| ee

Pa Oe Wie SRA OAK CAC oD OR OMe OT ow N A U K

MONOGRAFIE MATEMATYCZNE

TOM 34

WACLAW SIERPINSKI

CARDINAL AND ORDINAL

NUMBERS

WARSZAWA 1958
PANSTWOWE WYDAWNICTWO NAUKOWE

Pe
CARDINAL AND ORDINAL NUMBERS
POUL
6S) Ri Aa A RecA
ag tee Nie eee
MONOGRAFIE MATEMATYCZNE

KOMITET REDAKCYJNY

KAZIMIERZ KURATOWSKI REDAKTOR

KAROL BORSUK, BRONISLAW KNASTER, STANISLAW MAZUR
WACELAW SIERPINSKI, HUGO STEINHAUS, WLADYSLAW SLEBODZINSKI
ANTONI ZYGMUND

TOM 34

Pi AUN'S.T)-W-.0,W.EO W Y D°AIWN
IGG DW O72 NGA.
Gl ko Oawee
 WACLAW SIERPINSKI

CARDINAL AND ORDINAL

| NUMBERS

WARSZAWA 1958

HAFNER PUBLISHING COMPANY

NEW YORK
 COPYRIGHT, 1958, by
PANSTWOWE WYDAWNICTWO NAUKOWE
WARSZAWA (Poland) Miodowa 10

All Rights Reserved

No part of this book may be translated or reproduced
in any form, by mimeograph or any other means,
without permission in writing from the publishers.

PRINTED IN POLAND

Drukarnia Uniwersytetu Jagiellonskiego w Krakowie

TO MY GRANDDAUGHTERS
ANITA AND KASIA SIERPINSKI
 Digitized by the Internet Archive
in 2022 with funding from
Kahle/Austin Foundation —

https://archive.org/details/cardinalordinalnO00Ounse
 FOREWORD

This book is the product of long years of work. In the year 1909
appeared a lithographic edition of my lectures on the Theory of Sets
delivered at the University of Lw6w, and in 1912 the first edition of my
»An Outline of the Set Theory“ was published in Warsaw. The second
edition of that book appeared in 1923 and the third (divided into two
parts, having VITI+ 237 and VII+4 232 pages respectively) in 1928.
I lectured on the Theory of Sets repeatedly at the University of Warsaw
in the years 1920-1939. This work, however, does not comprise the whole
of the Theory of Sets or even the General Theory of Sets, but only deals
with certain important branches of it, namely the cardinal and the ordinal
numbers. I finished the manuscript of this book in Polish at the end
-of 1952. The process of translating it into English and the preparations
for printing occupied the next few years. During that period only very
slight changes and additions could be made. That is why it is only occasio-
nally that I was able to introduce results that were published later than
1952; in particular, I was unable to make use of those which are contained
in the work of Heinz Bachmann, Transfinite Zahlen, 1955.

W. Sierpinski

Warsaw, November 1957

 i ;cayae
eA

Pe
Pte. cr rT ar
f My ; Ake sage pene
3haa 4Deeks ie) Loa
Sy Peal
haan Sots eee
Os
¢ ' -

od.
7? fr, ere
tet nh”
 CHAPTER I

SETS AND ELEMENTARY SET OPERATIONS

1. Sets. From given objects we can form sets. For example we can
form, sets of letters {a, b,c¢,}, fe; 4,7, 9,h}, (0, f}-
From natural numbers (7. e. positive integers) we can form, for
example, the set of all natural numbers smaller than 10, the set of all
odd numbers, the set of all natural numbers between 100 and 1000, the
set of all squares, ete.
Here are some other examples of sets: the set of all the words used
in this book, the set of all rational numbers, the set of all points on a
straight line, the set of all spheres in the space.

2. Elements of a set. The objects constituting a given set are called

its elements. Thus, for instance, the elements of the set of letters {b, f}
are the letters b and f (and only these). The same object may be an ele-
ment of different sets: for example number 1 is an element of the set of
all natural numbers, but also an element of the set of all rational numbers
and an element of the set of all positive real numbers. Thus different sets
can have common elements.
The set of elements denoted by the letters a,b,...,1 is denoted
etd. Oe hesed
3. Symbols « and ¢. To denote that the object p is an element of the
set A we write
peA,

and to denote that the object p is not an element of the set A we write

pea.
Thus, for instance, if A = {a, b, c}, then we have acA, DeA, ceA,
but d¢A, if d is none of the objects a, b,c.
4, Set consisting of one element. According to our notation {a} de-
notes the set consisting of a single element, namely the object a. Indeed,
it could be maintained that no set exists where there is only one element,
6 I. Sets and elementary set operations

but actually that would be merely a question of terminology. Demanding

every set to contain at least two elements would cause certain incon-
veniences. Postulating this, we should be unable to speak in general of
the set of all the roots of a given equation, not knowing beforehand
whether it has more than one root. Similarly, we could not speak of the
set of all even prime numbers.
By forming a set of given objects we create a new object, distinct
from any of the original objects. Thus no set is its own element. In parti-
cular, for every object a we have :

{a} € {a},

and since, of course, for every object a we have

ae {a},

these formulae prove that {a}! Aa. Thus, the set containing only the single
element a must be distinguished from that element itself.
Therefore, if A,¢A,, then A,~A,. More generally, let us assume
that if, for a natural n >1,A,¢«A,¢ A,...€A,, then the objects A,,
A,,..., A, are all distinct from one another (Zermelo [5], p. 31).

5. The empty set. We often consider a set A of all objects p satisfying

a given condition C, not knowing beforehand whether there exists a single
object satisfying that condition (e. g. the set of all the roots of a given
equation). If no object satisfying the condition C exists,.we say that
the set A is empty and write A = 0.
In case of any doubt, it should be settled whether 0 denotes the
empty set or the number 0. In order to avoid this ambiguity some authors
denote the empty set by 9 or by A.
Thus, for instance, the set of all rational numbers « satisfying the
equation «? = 2 is empty.

6. Equality of sets. Two sets are regarded as equal if each element

of one set is an element of the other set and vice versa. To denote that
the sets A and B are equal we write A =B. Thus we have for instance

{a,b,c}= {e,
b, a} = {a,
¢, b} = {a, b,c, a}.

Here the same set is denoted by four different symbols.

From the definition of the equality of sets it follows that if the sets
A and B are empty, then A = B'). Therefore we can say that there exists
only one empty set.

1) A closer explanation of this fact will be found in the proposition (8.3).

 § 7. Sets of sets 7

7. Sets of sets. We can also form sets of objects which themselves

are sets. For example from the objects a,b,c,d,e, we can form two
sets, P = {a, b} and Q = {c, d, e}, and taking these two sets as elements
we can form the set
Z4= {P,Q} = {{a, b}, {c, d, e}}.

The set Z has therefore two elements: P and Y. It must be distin-

guished from the set 7 = {a,b,c¢,d,e}, obtained by joining into one
set all the elements of the set P and all the elements of the set Q. The
set 7 has five elements: a,b,c¢,d,e; none of them is an element of the
set Z. Thus, an element of an element of a given set is not necessarily an
element of that set. Therefore the relation « is not transitive: the formulae
aeP and Pe Z do not imply the formula ae Z.
In particular cases, however, an element of an element of a given
set may be an element of that set. H. g. a is an element of the set {a, }
and also of the set {a, {a, b}}, of which {a,b} is an element.
Sets of sets are sometimes called families of sets.
EXERCISES. 1. For every natural number n give an example of a set A of n
elements, such that each of two elements of the set A one is an element of the other !).
Answer. The set A = {a,, ao, ...,a,} Where a, 18 any object, and a, = {a, a,

> & 3} for k= 2,3,...,n. Thus we have a, = {a}, d= (a, {a,3} a fa, {ay},

{a {a,}}} ete.

2. Prove that if a, b, c and d are any objects (different from one another or not),
then in order to have

(7.1) {fa}, {a, b}} = { {}; fe, at,

is necessary and sufficient to have simultaneously

(7.2) ipa CONG Weebl.

Proof. It follows from (7.1) that {c} « {{a}, {a, b3\; therefore {c} = {a} or {e}=
= {a, b}. The latter equality can hold only when a = b, because if a 4 b, the set {a, b}
would have two different elements, while the set {c} has only one element. But if a= b,
we have {a, b} = {a, a} = {a}. Thus in any case we have the equality {c} = {a}, which
gives c= a.
If a ¥ 6, formula (7.1) gives {a, b} = {e, d} (since fa, bse {{c}, {e, d3\), the equality
{a, b} = {c} being impossible in view of a 4 b. Since, as we have already proved, a = ¢,
we have {a, b} = {a, d}, whence, in view of a 4 b, we find that b= d.
If a=b, formula (7.1) gives {{a3} = {{c}, {e, d}\, whence {c, d} = {a}, therefore
c=d=—a=b, whence again b = d.
Thus in any case formula (7.1) implies equality (7.2). Conversely, it is obvious
that equality (7.2) imples formula (7.1). The equivalence of formulae (7.1) and (7.2)
has therefore been proved.

1) Such sets play a part in certain reasonings; see e.g. Neumann [1].
8 I. Sets and elementary set operations

8. Subset of a set. If each element of a set A is also an element of

a set B, then we say that the set A is a part or subset of the set B, or that
the set A is contained in the set B and write A CB. The same is expres-
sed by BDA, which one reads: the set B contains the set A. We also say
then that the set B is a superset of the set A. (The relation C is called the
inclusion).
For every set A we have, of course, ACA.
If 4 CB and A+B, then the set A is called a proper subset of the
set B.
According to the definition of the relation C the formula 4 CB
expresses the same as the proposition

(8.1) For every, p, uf p <A, then p ¢B.

If the set A is empty, A = 0, then proposition (8.1) will be true for

every set B, because the antecedent of this conditional sentence will be
false for every object p, and every conditional sentence whose antecedent
is false is regarded as true. We have therefore 0CB for every set B.
Therefore: the empty set is a subset of every set. Of course, the empty set
has only one subset (namely the empty set). Every non-empty set has
at least one proper subset.
From the definition of the relation C it easily follows that this rela-
tion is transitive, 1. e. for any sets A, B and C the formulae ACB and
BCC imply the formula 4 CC.
The family of all subsets of the set A is denoted by U(A).
It is to be observed that a subset of a given set may also be an ele-
ment of that set, e. g. the subset B= {a} of the set A = {a, {a}. We
have here also BC A, therefore Be U(A) and Be A.
Similarly, for the sets P= |{{a}}} and Q= {a, P} we have Pe
and Pe UUU(Q), but P¢ U(Q), and likewise P ¢ UU(Q).

EXERCISES. 1. Prove that the equality A = B holds for the sets A and B if
and only if the two formulae Ac B and Bc A hold.
2. Prove that the formula p «A holds for an object p anda set A if and only
if {p}.c A.
3. Prove that for the sets 4, B and OC the formula 4c BC Oc A is equivalent to
the formula A = B=0O.
4, Let » denote a natural number >1. Prove that for the sets A,, A.,..., A, the
formula A,c A,c...c A, cA, is equivalent to the formula A, = A,=...=
5. Find all the subsets of the set {a,b} containing two different elements.
Answer: 0, {a}, {b}, {a, b} (since the empty set is a subset of every set).
6. Find all the subsets of the set {a,b,c} containing three different elements.
Answer. 0, {a}, {b}, {ce}, {a, b}, {a, C}, {b, Ch, ta, b, Ch.
 § 8. Subset of a set i)

7. Let k and n be two natural numbers, k < », How many different subsets of k
‘elements does a set of n elements contain?
Answer.
TON Mask n!}
k} kY(n—k)!’
since to find them it is sufficient to take into consideration all combinations of 7 ele-
ments of the set, taken k at a time.
8. 'dow many different subsets does a set of n elements contain (n being a natural
number)?
Answer. 2” (we include here the empty subset, of course).
9. Let Z(a), for real « > 0, denote the set of all natural numbers n = 2”?~'(2q—1),
where p and q are natural numbers such that p < qx. Prove that for 0 <a < y the
set Z(x) is always a proper subset of the set Z (y).
Proof. Let ~ and y be two real numbers >0, such that w < y. If ne Z(a), then
n = 2”-*(2q—1), where p and q are natural numbers such that p < ga and, since x < y,
we have p < qy, whence it follows that n « Z(y). Each element of the set 7 (x) belongs
therefore to the set Z(y), whence Z(x)c Z(y). Since 0 << a4 < y, there exists, as we
know, a rational positive number w such that « < w < y. The rational positive num-
ber w can be represented in the form w = r/s, where r and s are natural numbers. Thus
we shall have « < 1r/s < y, therefore r > sx and r < sy. Let n = 2"—*(2s—1). Since
y < sy, we shall have ne Z(y).
Suppose that » « Z(x). According to the definition of the set Z(x) we should then
have n = 2”~*(2q—1), where » and q are natural numbers such that p < qa. There-
fore we should have Rog i} = ea es 1), where p, q, 7 and s are natural numbers,
whence, as we know, it follows that p =r and q = s. This, however, is contrary to the
inequalities 7 > sw and p < qx. Thus n € Z(x), and since n « Z(y), we have Z(x) 4 Z(y).
Thus the set Z(x) is a proper subset of the set Z(y), q. e. d.
10. Give an example of a non-empty set whose all elements are subsets of that set.
Answer. The set Z = {A}, where dA denotes the empty set. Similarly, the set
9h |A, {A}} and the sets

9. Sum of sets. If / is a given family of sets, then by the swum of all

sets of the family # we understand the set S formed of all those (and
only those) elements which belong to at least one set of the family 7’.
We then write ,
s= A.
A€F

The sets of the family F are called cept nen: of the sum SN. If @ is
a family of families of sets and
Yi MF,
FeD
10 I. Sets and elementary set operations

then, as can easily be seen, we have the formula

Se oe
Aev FeD A€F

which expresses the generalized propertyof associativity of the sum of sets.

As can easily be observed, every set A is the sum of all its subsets;
this can be written in the form

i CaO aYe,
yeas
XEU(A)

or in the form

ASX.
XCA

Every set A is also the sum of all its one-element subsets (but gene-
rally it is not the sum of all its elements); this can be written in the form

A= >, {p}
peA

From the definition of the sum of sets and the relation C it follows
immediately that every swm of sets contains each of its components.
If n is a natural number and family F' consists of n sets A,, As, ..., An;
then the sum of all sets of the family F is denoted by A,+4,.+...+A,.
For every set A we have, as can easily be verified, the formulae
A+0=0+A=A and A+A=A. It follows from the last of these
formulae that in writing down a sum of sets there is no need to repeat
identical components.
It is also easy to see that if A and B are sets such that BCA, then
A+B= A. We can express this by saying that any component of a sum
of sets absorbs any other of its components contained in it. Conversely,
if d+B= A, then BCA.
It follows from the definition of the sum of sets that for any sets A
and B we have A+B= B+ A, and for any sets 4, B and CO we have
(A+B)+C=A+(B+0C).
The sum of sets, therefore, depends neither on the order of the
components nor on the way of joining them.

10. Difference of sets. The set of all those elements of the set A
which do not belong to the set B, is denoted by A—B and called the
difference of the sets A and B. ;
‘As can easily be seen, 4—B= 0 if and only if we have ACB,
and A—B= B—A if, and only if, A = B.
 § 10. Difference of sets Tt

Not all the properties of the algebraic sum of numbers are valid for
ets. H. g. if the sets A and C have an element in common, then

A+(B—C€)#(A+B)—C,
and if C—A #0, then
(A—B)+CA4A—(B—C).

On the other hand, for any sets A, B and C we have A—(B+C)=

= (A—B)—C and (A+8B)—C= (A=C)4-(B— 0).
EXERCISE. Prove that for any sets A, B and OC the following formulae hold:

(A—B)—O = (A—O)—(B_-0Q),
(BOB AeA] OG(A=B)-3(B <0)
11. Product of sets. If / is a given family of sets, then by the product
of all the sets of the family Ff we understand the set P containing all
those and only those elements which are contained in each set of the
family F. We then write
p= [[a.
A€F

The sets of the family P are called factors of the product P.

If ® is a family of families of sets and

y= »'P,
he,

then we have the formula zi

[l4=[][T-4,
AEev FeD A€F

which expresses the generalized property of associativity of the product

of sets.
It follows from the definition of the product of sets, that a product
of sets is contained in each of its factors.
If is a natural number and the family F consists of n sets A,, Ao, ...,
A,, then the product of all sets of the family F is denoted by A,-A,-...- A,
OE AatA sss Ay. :
From the definition of the product of sets it follows that the product
does not depend on the order of factors.
For every set A we have, as can easily be verified, the formulae:
A-0 = 0-4 =0 and A-A = A. From the last of these formulae it follows
that in writing down a product of sets there is no need to repeat identical
factors.
It is also easy to see that if A and B are such sets that A C B, then
AB = A. Conversely, if AB= A, then ACB. From the properties of
12 I. Sets and elementary set operations \

the product of sets it follows that for every two sets A and B we have
the formula 4B — BA, and for every three sets A, B and C we have the
formula (AB)C = A(BC).
It can easily be seen that if # is an arbitrary family of sets and B
an arbitrary set, then
BY A=) BA.
AeF A€F

This formula proves the distributivity of the multiplication of sets

with respect to addition. The converse is also true: the addition of sets
is distributive with respect to multiplication. We have the formula

(11.1) B+[[4=[] (B+).

AeF A€F

Here is the proof of this formula. The formula states an equality

of two sets. In order to prove it, we must demonstrate, according to the
definition of the equality of sets, that every element of the set given
by the left-hand side of the formula is an element of the set given by the
right-hand side of the formula, and vice versa.
Suppose, therefore, that p is an element of the set on the left-hand
side of (11.1). This set is a sum of two sets. Since p is an element of this
sum, p is an element of at least one of its components. Thus we have
peB or pe]]A or both of these relations.
If pe B, then, for every set A, we have pe(B+A), and p is the
element of each factor of the product given by the right-hand side of
formula (11.1), and is thus an element of that product. On the other
hand, if

pep a,
A€F

then, according to the definition of the product of sets, p is an element

of every factor of the product ][ A, i.e. of each set A of the family F and,
AeF
moreover, p is an element of every set B+A, where B is an arbitrary set
and A is a set of the family #. Hence we conclude again that p is an ele-
ment of each factor of the product given by the right-hand side of for-
mula (11.1), and therefore p is an element of that product.
Thus we have proved that each element of the set given by the left-
-hand side of formula (11.1) is an element of the set given by the right-
-hand side of this formula.
Suppose now that p is an element of the set given by the right-
-hand side of formula (11.1). It follows from the definition of the pro-
duct of sets that p « B-+A for each set A of the family I’. If p « B, then,
 § 11. Product of sets ° 13

obviously, p is an element of the set given by the left-hand side of formula

(11.1) (since p is an element of its first component). On the other hand,
it p ¢ B, then, in view of pe B+A for Ae F, we have p¢A for AeF
(since an element of a sum of two sets must belong to at least one of
its components, and if it does not belong to the first component, then
it must belong to the second one). The element p belongs to each set .
of the family F, therefore to each factor of the product [] A and thus is
A€F
an element of that product. Therefore, p is an element of the second
component of the sum constituting the left-hand side of formula (11.1)
and thus. p is an element of that sum.
Thus we have proved that each element of the set given by the right-
hand side of formula (11.1) is an element of the set given by the left-hand
side of that formula, and since the converse has been shown before, this
completes the proof of (11.1).
In particular, for any sets. A, B and C we have the formula B+
+40 = (B+A)(B+0).
12. Disjoint sums. A family of sets of which no two different sets
contain common elements is called a family of disjoint sets, and the sum
of the sets of such a family is called a disjoint sum. Hach element of such
a sum belongs to one, and only one, of its components.
‘It is easy to verify that, for any sets A and B, we have the following
decompositions of the sums A+8, or A+B+C, into sums of two,
or three disjoint sets: A+B= A+(B—A) A+B+C=A+(B-A)+
tO (Ae=B)I.
In general, if is a natural number, and we have the sets A,, As, ...,
A,, then putting S,= A,+A,.+...+A, for k=1,2,...,n, we have the
following decomposition of the sum A,+A,+...+A, into n disjoint sets:

Nb BRS Wat eT Te Mw, a Se” A Raeleg

Let © denote any family of disjoint sets whose sum is a given set A.
We say that two elements, a and J, of the set A are congruent modulo O
if they belong to the same set of the family 0, 7. e. to the same component
of the sum
5 A= »'E.
Ee

We then write a = b(mod@). In this way the most general concept of

congruence is reduced to the concept of set.
As a particular case we have the congruences a = b(modm), discussed
in the theory of numbers, where for a given natural number m (or, more
generally, for a complex number 40) the module is the decomposition
 \

14 » I. Sets and elementary set operations

of the set of all integers (or, more generally, of all complex numbers)
into disjoint components, the two numbers a and 6 being regarded as
belonging to the same component if and only if there exists an integer
k such that a—b = km.

EXERCISES. 1. Prove that for any two sets A and B we have the following
decomposition of the set A into two disjoint components (which can be empty sets):
A =(A—B)+AB.
2. Prove that for any sets A, B,O we have the following decomposition of their
sum into four disjoint components:

ALL BA Cea BY45( B30) se(Oma eA

3. Prove that for any sets A, B, C, D we have the formula

A+B+ C+D = (A—B)+ (B—C)+(C—D)+ (D—A)+ABCD,

and give an example of sets A, B, C, D for which the components of the right-hand
side are not disjoint.
Proof. The right-hand side of our formula is, of course, contained in the left-hand
side. In order to prove the formula it suffices to show that its left-hand side is contained
in the right-hand side. Therefore let p<« 4+B+O+D. Thus the element p belongs
to at least one of the sets A, B, VU, D. Obviously, if we perform a cyclic permutation
of the letters A, B, CO, D, our formula remains unchanged; therefore, we can suppose,
without loss of generality, that p « A. If we had p « ABCD, p would belong to the last
component of the right-hand side of our formula. Suppose further that p ¢ ABOD.
The element p, therefore, does not belong to at least one of the sets B, U and D (since
it belongs to A). If the first of the sets B, OC, D to which p does not belong is B, then
p¢«A—B; if Cis the set in question, then p< B—(; finally, if Dis that set, then peC0—D.
Thus, in any case, p belongs to at least one of the five components of the right- |
hand side, therefore p is an element of the right-hand side of our formula, q. e. d.
An example of the sets A, B, 0, D for which the components of the right-hand
side of our formula are not disjoint -is given by the sets A = 0 = {p}, B= D= 0. The
element p belongs here to the first and to the third component of the right-hand
side of our formula.
4. Prove that for any sets A and B we have A= A(A+B).
5. Prove that if A, B, O, D are sets such that 4+B=0O+D, then there exist
sets P,Q,
&, S such that A= P+Q, B=R+8, C=P+IR, D=Q+8.
Proof. As can easily be verified, we can assume that P = AO, Q = AD, R= BO,
S = BD, since we have, for instance, P+Q = A(O0+D) = A(A+B)=A.
6. Prove that if A, B, 0, D are sets such that A+ B= C+D and AB=CD=0O,
then there exists only one system of sets P, Y, Rk, S such that A =9P+Q, B= R+S8,
C=P+k, D=@Q-+S8, and moreover, we must have PY = RS = 0.
Proof. Let A, B,C, D be sets such that A+B=CO+D and AB=OD=0.
According to exercise 5 there. exist sets P,Q, R and S such that 4 = P+Q, B= R+8,
C= P+kh, D=Q+8. Now let P,Q, Rk, S be any sets satisfying these four equalities.
Hence we have Yc D, kc OC, therefore QR c CD = 0 and thus QR = 0, whence ACO =
= (P+Q)(P+8) = P+P(Q+8)+Qk =P. Therefore AO =P. Similarly, we shall
find AD=Q, BC = Kk, BD=S8. Thus the sets P, Q, RB, S are determined by the sets
 §$ 12. Disjoint sums 15

A, B,C, D, and, since Pc 0, Qc D, Rc C, Sc D and OD= 0, we find PQ = FS = 0.

This completes the proof.
7. Prove that if n is a natural number S 1, then for arbitrary sets A,, Az, ..., A,,
the following formula holds:

A,+A,+...+4, = (A,—A,)+ (A,—As;)+...4 (A,_, —A,)+ (4, —A1) + Are... A, .

13. Complement of a set. We often have to deal with subsets of the

same set W. If AC W, then the set W-+-A is called the complement of
the set A to the set W and denoted by CA.
It is easy to verify that for every set AC W we have

Ae (OA =a svandiae 40 (CA =A.

and that for the sets 4C BC W we have CA CB.

If F is a family of subsets of the set W, then we have the formulae

CiAcs. CAG) andy, OS)

Aas [ cae
Ae€F A€F A€F A€F

These are called De Morgan’s formulae for sets. The first of them
states that the complement of a product of sets is the sum of their comple-
ments, the second — that the complement of a sum of sets is the product
of their complements. x
We shall prove the first of these formulae. Let us suppose that
peO]] A. Therefore p ¢ ITA, and there exists a set A, of the family F
A€F €

such that p ¢ A,, whence, in view of the fact that the sets of the family 7
are subsets of the set W, it follows that pe W—A,=(CA,, and,
since A,«F, therefore p« >) CA. Thus we have
A€F

CA Mee iene:
A€F AeF

On the other hand, let us suppose that p « » OA. Thus there exists
A€F

a set A,¢«F such that peCA, therefore p¢A, and naturally p¢// A,
A€F

and since pe CA = W—A, whence’ pe W, therefore pe W—]|] A, i. «.

Aer
p«C][] A. Thus we have ) CACC/] A.
AcF AeF AeF
The two inclusions obtained give the equality

C[[a=) c4,
A€F AeF

qié5 de
16 I. Sets and elementary set operations

In particular, it follows from De Morgan’s formulae that if A and B

are subsets of the set W, with respect to which we take the complements,
then CAB=CA+CB and C(A+ 8B) =CA-OB.

EXERCISE. Prove that if the sets A and B are subsets of the set W, with respect
to which the complements are taken, then

BR ACB OA Bye Cee

A=B
= CCA BB). AB = C(O4-AOB)=
14. Ordered pairs. Let a and b be two different objects. Write

(a, b) ae {{a}, {a, by} :

According to this notation we shall have

and, since ab, a b) ~(b, a), for in case of (a, b) = (b, a) we should
have {{a}, {a, b}}= {{b}, (0, a}}, whence, as we have proved in section 7
(exercise 2), it would follow that a = b, contrary to our assumption that
Geer.
Thus, unlike the set {a,b}, the arrangement (a,b) depends on the
order in which the elements a and b occur in it. We call it an ordered pair.
In the ordered pair (a, b), a is called the first, b — the second element of
the pair.
By means of ordered pais we can define ordered triplets as the
arrangements ((a, Ay), Ms) , ordered quadruplets as the arrangements
(a1, Ml), M3), 44) and, generally, ordered sequences of n terms, n being an
arbitrary natural number.

15. Correspondence. Function. If we have an ordered pair (a, b),

we say that the object b is associated with (or corresponds to) the object a.
In this way the concept of correspondence is reduced to the concept
of set: :
Suppose now that we have a given set Z of ordered pairs. Let A
denote the set of all first elements of these pairs, B — the set of all their
second elements. For each element ae A let us denote by f(a) the set
of all those elements } for which (a, b) « Z. (Thus we obtain the function f,
associating with each element a of the set A a certain subset f(a) of the
set B). In the particular case where the set f(a) has only one element for
each element a of the set A, we say that we oee a single-valued function
(defined for the elements of the set A), whose values belong to the set B. -
In this way the concept of function is reduced to the concept of set.
 § 15. Correspondence. Function 17

If the set f(a) contains only one element dD, 7. e. if f(a) = {b}, then
we write b = y(a). In this way with each element a of the set A is as-
sociated an element g(a) of the set B. However, the formulae a,« A,
a,€ A, a, ~ 4, do not necessarily imply the formula y(a,) + y(a,). If to
each two different elements a, and a, of the set A always correspond two
different elements y(a,) and y(a,), then the function » is called invertible
in the set A.
We might also consider functions f(a) for which the values of variable «
do not form a set, e. g. functions whose variable can be any object. (As we
know, the concept of the set of all objects leads to an antinomy (Sier-
pinski [58], p. 50 and 113). For instance, f(7) and g(a#) will be such func-
tions if f(#) = w for every object «, while g(x) = 2? if 2 is a real number,
and g(a) = 0 if the object x is not a real number. U(X) will also be such
a function, where if X is a set, U(X) is the set of all subsets of the set X,
and if X is an object, not a set, U(X) is the empty set.

16. One-to-one correspondence. Suppose that the function is in-

vertible in the set A. Therefore to each element a of the set A there cor-
responds an element y(a) (not necessarily belonging to A), while to dif-
ferent elements a, and a, of the set A there always correspond different
elements g(a,) and gy(a,). Let us denote by B the set of all elements y(a),
corresponding to the elements a of the set A. Therefore, to each element
of the set A there corresponds an element of the set B, and to different
elements of the set A there correspond different elements of the set B.
From the definition of the set B it follows that each element b of the set B
corresponds to one, and only one, element of the set A, since otherwise
the function g would not be invertible. Therefore we say that the func-
tion g establishes a one-to-one (or (1-1)) correspondence between the
elements of the set A and the elements of the set B.
Now let b denote any element of the set B. It follows from the defini-
tion of the set B that there exists at least one element a of the set A such
that y(a) = b, and from the assumption that the function » is invertible
in the set A it follows that there is only one such element a of the set A.
Therefore, for each element b of the set B there exists one and only one
element a of the set A such that g(a) = b. Let us denote this element a
by (b). The pairs (b, y(b)), for all b < B, form the set y; moreover, from
b, = b, follows p(b,) = y(b,). The set y, therefore, is a function. It can
easily be seen that the function y establishes a (1-1) correspondence
between the elements of the set B and the elements of the set A. It is
also clear that for every element D of the set B we have y (p(d)) == 5, since
it follows from the definition of the function y that for b « B the equality
g(a)= b implies the equality a= y/(b). If, on the other hand, for an
Cardinal and ordinal numbers 2
18 I. Sets and elementary set operations

element a of the set A we take y(a) = b, then we shall have b « B, and,

according to the definition of the function y, y(b) = a, whence p(p(a)) 16
for each element a of the set A. Thus the operators y and yw placed succes-
sively, reduce each other. The function y is called inverse with respect
to y and is denoted by y-!. As can easily be seen, the function ¢ is in-
verse with respect to the function y.
EXAMPLES. 1. A (1-1) correspondence between the elements of the sets A =
= {x,y,z} and B= {z,t, u}, where 2, y, z, t and uw are different objects, is established
by the function » defined by the conditions p(x) =z, p(y) =t, y(z) = wu. Its inverse
function will be the function g, defined by the conditions: gy '(z)=2, p(t) = y,
y (uw) = z. But the function g,, defined by the conditions ¢,(x) = t, y,(y) = u, gi(z) = 2,
also establishes a (1-1) correspondence between the elements of the sets A and B. Its
inverse function will be the function g,1, defined by the conditions g;1(z) = 2, g(t) =
= 2, gy(u) = y. It is easy to verify that there exist four other functions establishing
a (1-1) correspondence between the elements of the sets A and B.
Similarly, it is easy to establish that if the sets A and B have n elements each,
n being a natural number, then there exist m! different functions establishing a (1-1)
correspondence between the elements of those sets.
Thus, for instance, a (1-1) correspondence between the set of all odd numbers
not exceeding 100 and the set of all even numbers not exceeding 100 can be established.
in 50! = 1-2...50 > 10: 11-12...50 > 10* different ways.
2. There exists a (1-1) correspondence between the set A = {1, 2,3,...} of all
natural numbers and the set B= {2,3,5,7,...} of all prime numbers.
Such a correspondence is established e. g. by the function relating to each natural
number 7 the n-th successive prime number (taken in order of increasing magnitude).
Thus we have y(1) = 2, (2) = 3, y(3) = 5, y(10) = 29. The calculation of the number
y(n) for large values of » presents considerable difficulties.
3. Let A denote the set of all straight lines on a plane which do not pass through
a given point O on that plane, and B — the set of all points on the plane different
from O. Clearly, a (1-1) correspondence between the elements of the sets A and B will
be established by relating to each straight line p of the set A the point y(p) on that line
nearest to the point O. It can easily be seen that for each point q of the set B, y(q) will
be the line passing through the point q and perpendicular to the segment Oq.
On the other hand, if we denote by A’ the set of all straight lines on a plane, and
by B’ the set of all points of the plane, and define for each straight line p « A’ the point
y(p) as the point on the line p nearest to a given point O on the plane, then, though we
should obtain a function g(p), determined in the set A’, whose values for p « A’ cons-
titute the set B’, that function would not establish a (1-1) correspondence between
the sets A’ and B’, since it would not be invertible-in the set A’: for each straight line p
passing through the point O we shall have y(p) = 0. It does not follow, however, that
between the elements of the sets A’ and B’ a (1-1) correspondence does not exist.
It could be proved that such a correspondence exists, but the definition of the function
which establishes it would be rather complicated (see Chapter II, section 6, Exercise 2).
4. If A is a set containing only one element and B a set containing more than
one element, then there exists no (1-1) correspondence between the elements of the
sets A and B, since, for every single-valued function defined in the set A, the set of its
values has only one element and thus is different from the set B.
Thus there are sets between whose elements no (1-1) correspondence exists.
 § 16. One-to-one correspondence ‘ 19

5. Let A denote a given set, 7’ — the set of all subsets of the set A, P — the set
of all functions f(«) defined for 7 « A and assuming only the values 0 and 1. We can
establish a (1-1) correspondence between the elements of the sets 7 and P, relating
to each element X of the set 7, 7. e. to each subset X of the set A, a function f(x) de-
fined for «<A by the conditions f(z)=1 for we X and f(x) = 0 for we A—X. This
function is the so-called characteristic function of the subset X of the set A. If we de-
note it by fy, then a (1-1) correspondence between the elements of the sets 7 and P
can be established by the formula y(X) = fy for X«T (i. e. for Xc A). Clearly, the
functions g71, inverse to the function @ will be defined for the elements f of the set P
by the following condition: g 4(f) is the set of all elements w of the set A for which
Gp) ake

17. Cartesian product of sets. By a Cartesian (or combinatorical)

product of the sets A and B we understand the set of ordered pairs (a, y)
where we A and ye B. Such a set is denoted by A xB.
A Cartesian product is generally neither commutative nor associative.
On the other hand, it is easy to establish a (1-1) correspondence between.
the elements of the sets A xB and Bx A assigning to each element
(x,y) of the set A xB an element (y, x) of the set BxA, where (a, y)
is an ordered pair in which we A and yeB. Similarly, it is easy
to establish a (1-1) correspondence between the elements of the sets
(A xB)xC and Ax(BxC). It suffices to assign to each of those
elements of the first set which have the form ((x,y),z), where we« A,
y«B, ze C, an element (x, (y,2)) of the second set.
EXERCISES. 1. Prove that the Cartesian multiplication of sets is distributive
with respect to the addition of sets, but that the addition of sets generally is not distri-
butive with respect to their Cartesian multiplication.
Proof. It is easy to verify that for arbitrary sets A, B, C we have the formulae
(A+B)x0=(AxO)+(BxC) and Ax(B+C)=(AxB)+(4xC). On the other
hand, ¢. g., for A = {a}, B=0, O= {c}, a#c, we have AxB=0, whence (4x B)+
+0 = {c}, and A+ C0 = {a,c}, B+ C = {c}. Therefore, (A+ C) x (B+ C) = {(a, ¢), (e, ¢)}
and thus (A x B)+ 0 ¢ (A+ C)x (B+ CQ). B,
2. Prove that for any sets A, B, C, D the following formula holds:

(AB) x (CD) = (Ax 0)(BxD).

3. Prove that if A,c A and B,c B, then (A, x B,)c (A xB).

4. Determine all the elements of the Cartesian product A xB, where A = {a, b, c}
and B= {e, d}.
Answer. The product A xB has six elements: (a,c), (a, d), (b,¢), (b, d), (c, ¢),
(c, d).
5. Let A denote a set having m elements and B a set having » elements, m and n
being natural numbers. How many elements has the set A xB?
Answer. mn.

6. Prove that if A and B are sets such that Ax d= BxB, then A = B.

20 : J. Sets and elementary set operations

Proof. If ae A, then (a,a)«e AXA = BxB, whence ac B. Thus every element

of the set A is an element of the set B, therefore A c B. We prove in a similar way that
BcA. Thus we have A = B, q.e. d.
_ 7. Prove that if A, B, C are sets such that A 4 0, B40, and (Ax#B)+(BxA)=
— Oe On bhenmeAs=— 10),
Proof. If ce O, then, in view of CxO = (Ax B)+ (BxA), we have (c,c)«AxB
or (c,c)«BxA. In the first case we have ce A and ce B, in the second ce B and
ce A. Thus, in any case, ifceO, then ce A and ce B. Therefore we have Oc A
and Oc B.
Now let us denote any element of the set A by a. Since B ~ 0, there exists an
element b of the set B. Therefore we shall have (a,b) «AxB, whence (a, b) «0x0,
and a«C. Thus we have AcC.
Finally let 6 denote any element of the set B. Since A + 0, there exists an element a
of the set A. Therefore we shall have (a, b) « A x B, and thus (a, b) « Ox 0, whence b « O.
Thus we shall have Bc 0.
The formulae Cc A, Cc B, AcO, BcO give A=O and B=(, therefore A =
== IB =a (Oh Gla Gy Cle
Let us observe that in the case of A = 0 we can assert only that CO = 0, while the
set B may be arbitrary.
8. Prove thatif A, B, C, D are sets such that 4 4~0,B4~0 and(AxB)+(BxA)=
=(0xD)+(Dx0O), then either 4 = C and B=D or d=D and B=(C.
9. Let A and B be any sets, not necessarily different. Let us denote by P the set
of all elements {{a}, {a, 63} where ae A, be B. Establish a (1-1) correspondence
between the elements of the sets A xB and P.
Solution. It suffices to assign an element {{a}, {a, b3} of the set P to each ele-
ment (a,b) of the set A xB. In order to prove that this is a (1-1) correspondence it is
sufficient to refer to I.7, Exercise 2.
10. Prove that if A, B, C, D are non-empty sets such that A ~B, O 4D and
{A,B} ~ {0, D}, then Ax (B—A)+Bx (A—B) 4 Ox (D—C)+D~x (C—D).
Proof. Clearly, at least one of sets A and B must be different from either of the
sets C and D. Otherwise each of the sets A and B would be equal to one and only one
(since C #~ D) of the sets O and D; moreover, since A ¢ B, the sets A and B could
not be equal to the same set, C or D. Hence it follows that {4, B} = {O0, D}, contrary
to our assumption. Therefore at least one of the sets A and B, e. g. the set A, is different
from either of the sets C and D. Similarly we prove that at least one of the sets O and D,
for example the set C, is different from either of the sets A and B. Thus we have A + O,
PAN-= pang eba-—Or
Suppose now that, contrary to our assertion, we have

(19.1) AX (Bad B (4B) = 00D OD (Oe Die

If we had B—A =O and A-~B=0, then AcBc4A, therefore A = B, contrary
to our assumption. Thus at least one of the sets B—A and A—O is non-empty. Similarly
we prove that at least one of the sets D—C and O—D is non-empty.
Suppose that B—A ~0: thus there exists an element b« B—A. Let x denote
any element of the set A. The ordered pair (x, b) is therefore an element of the
set Ax(B—A). In view of (19.1), it is an element of the right-hand side of
formula (19.1).
 § 17. Cartesian product of sets Zk

If we had 6 «(D—O), then b€ 0, thus b& O—D and we should have (a, b)«
« Ox (D—C), whence « « C. Therefore we should have w« 0 for w« A, and thus AcC.
But, in view of be D—O and b« B—A it could be proved in exactly the same way
that Cc A. Thus we should have A = OC, which is impossible. Therefore b € D—C and
thus 6 « C—D since if a « A, then the element (x, b) belongs to the left-hand side of
formula (19.1) and thus to the right-hand side of this formula. But then, just as before,
we come to the formula A = D, which is impossible. Thus the assumption that B—A + 0
leads to a contradiction. Therefore B—A = 0. Similarly we prove that D—C= 0.
Therefore we have A—B +4 0 and C-D + 0. Thus formula (19.1) gives Bx (A—B)=
= Dx(C—D), and since the sets B, D, A—B, O—D are non-empty, we must have
B=D and A—B = C-—D, and, since Bc A-and DcC, we have A = B+ (A—B)=
= D+(C—D) = OC, whence A = O, which is impossible. Therefore formula (19.1) does
not hold, q.e.d.

18. Exponentiation of sets. The set of all single-valued functions,

defined for the elements of the set A, whose values are elements of the
set B is denoted by B4.
Thus, for instance, for A = {a,, a, a3}, B = {b,, b,} the set B4 con-
tains eight functions (i =1,2,...,8), defined, respectively, by the
conditions: 9 (@,) = 91(42) = 9i(@3) = 013 Po( G1) = 92, Po(Ae) = Pols) = 43
P3(A1) = Ps(A3) = 01, Po(G2) = B23 alr) = Pale) = D1, Palas) = 925 Ps(Gi) =
= by, P5(42) = Ps(43) = 923 Pe(G1) = PolAs) = 92, Pols) = 01; Pal) = P(A) =
= be, Gy(43) = by; pa(b1) = %a(2) = Ye(43) = O2-
On the other hand, for the same sets A and B the set A¥® contains
nine functions y,(i1—1,2,...,9), defined, respectively, by the condi-
tions: y,(d1) = yr(d2) = 43 Yo(O1) = Gry P2(P2) = M25 Ys(Dy) = 1, Ya(O2) = 435
Pa(d,) = Mey YalD2) = G15 Ys (D1) = Y5(D2) = 23 Yo(D1) = 2, Yo(D2) = 33 Yr(O1) =
== As, Yr(D2) = O13 Ya(D1) = 3, Ya(Bo) = Ao3 Yo(O1) = Yo(O2) = as.
It is easy to prove that if the set A has n elements, and the set B
has m elements, m and » being natural numbers, then the set B4 is com-
posed of m” different functions.
In particular, if X denotes the set of all real numbers, then X*% is
the set of all real functions of a real variable, and if NV denotes the set of
all natural numbers, then X% is the set of all real functions of a natural
variable, or, which is the same, the set. of all infinite sequences of real
numbers 4).

1) The subject of this chapter belongs to the so-called Algebra of Sets. The reader
who wishes to study the subject at a greater length is referred to Sierpinski [58], espe-
cially Chapters II and III. The reader will find there a large number of exercises.
 CHAPTER II

EQUIVALENT SETS

1. Equivalent sets. Relation ~. Two sets A and B are said to be

equivalent (or to be of the same power) and written A~B, if between the
elements of these sets there exists:a (1-1) correspondence.
Examples of equivalent sets are given in 1.16 (Examples 1, 2, 3).
Here are some other examples:
1. The set of all natural numbers is equivalent to the set B of all
integers. The one-to-one correspondence between the sets A and B is
established by a function defined for natural numbers in the following
way: y(1) = 0, and for natural k, y(2k) =k, and p(2k+1) = —k.
2. From the axioms of geometry (e. g. from the axioms of Hilbert,
see Hilbert [3]) it follows that there exists a (1-1) correspondence between
all points of a straight line and all real numbers. If we fix two different
points O and Q on a straight line, then we are able to establish the above-
mentioned correspondence (to the point O we relate the number 0,
to every point P lying on the same side of the point O as the point Q —
a positive integer equal to the ratio of the lengths of the segments OP
and OQ, and, finally, to every point P not lying on the same side of O as
the point Q —a negative integer equal to —OP:0Q). Thus the set of
all points of a straight line is equivalent to the set of all real numbers.
This fact is usually regarded as obvious in analytic geometry. Cantor
(in 1872) was the first to point out that here we deal with a new axiom.
R. Dedekind drew attention to the close relation of this question to our
idea of the continuity of a straight line. Obviously, those problems could
not have arisen earlier than in the second half of the 19-th century;
there had been no exact definition of irrational numbers before that time.

2. Finite and infinite sets. If n is a natural number, and A is a set

equivalent to the set of all natural numbers < n, then there exists a (1-1)
correspondence between the elements of the set A and the numbers
1,2,...,”. In other words, the elements of the set A can be assigned,
successively, to the numbers 1,2,...,. If the element of the set A
bearing the number k is denoted by a,, then all the elements of the set
 § 2. Finite and infinite sets 23

A can be arranged as a sequence d,, a,...,a,. Therefore we shall have

A = {0y, Mg, ..-) Qn}. We say then that the set A has n elements.
Every set of n elements, where n is a natural number, can be arranged
as a sequence of n terms. It is easy to see that there exist n! different .
ways of arranging all the elements of a set of n elements as a sequence
of n terms, since there are n! permutations of the numbers 1, 2,..., ”.
For every given natural n every two sets of n elements are equi-
valent.
Every set which is empty or which has n elements, n being a natural
number, will be called a finite set). Any other set will be called infinite.
The set of all natural numbers, the set of all points on a straight line, the
set of all straight lines on a plane are examples of infinite sets.
There are examples of sets regarding which we are unable to decide
(at the present state of science) whether they are finite or not. Such is,
for instance, the set of all prime numbers of the form 2”+1 where n is
a natural number, and also the set of all prime numbers p for which p + 2
is also a prime number.
An example of a theorem whose proof is known but difficult is pro-
vided by the following theorem of F. P. Ramsey:
If, for a given natural number n, all n-element subsets of the set of
all natural numbers have been divided into a finite number of classes, then
there exists an infinite set of natural numbers such that each n-element
subset of that set belongs to the same class (Ramsey [1], p. 264-286, and
Erdés and Rado [1], p. 249-285).
The sum of two finite sets is a finite set. Indeed, if the set A has k
elements, and the set B has J elements, k and | being natural numbers
or = 0, then, clearly, the set A+B has at most k+1 elements (exactly
k+1 elements if the sets A and B are disjoint). Since the sum of two na-
tural numbers is, as we know from arithmetic, a natural number, and
k+0=k and 0+l=1, the set A+B is finite.
Hence it follows that if a set S is infinite and S= A+B, then at
least one of the sets A and B is infinite (since if both of the sets A and B
were finite, their sum S would also be finite, contrary to our assumption).
Very often, however, even in decomposing into two parts an infinite
set as simple as the set V of all natural numbers, we are not able to decide
which part is infinite. That is the case, for instance, if A denotes the
set of all natural numbers n for which the number 22"+1 is a prime,
and B= N—A. We know that at least one of the sets A and B must be
infinite, but we do not know which of them; and neither do we know whe-
ther both of them are infinite.

1) As regards other definitions of finite sets see III. 9.

‘24 II. Equivalent sets

Every set equivalent to a set of » elements, where n is a natural

number, is also a set of m elements. Hence it follows that every set equi-
valent to a finite set is a finite set. Thus also every set equivalent to an infinite
set is an infinite set.
We shall prove that if a set A is infinite, then for every natural num-
ber n the set A contains a subset B, of n elements.
The theorem is true for n = 1, since the set A, being infinite, is not
empty; thus there exists an object a which is an element of the set A,
and the set B= {a} is a ove-element subset of the set A.
Now let » denote a given natural number and suppose that the
set A contains a subset B of n elements, B, = {a,, ag. ..., dn}. If we had
A—B,=0, then (since B,C A) A= B, and the set A would be finite
(of nm elements), contrary to our assumption. Thus we have A—B, ~0
and there exists an element a,11 of the set A—B,. The set Bry=
= Bart {Gnai} = {01 Ags s+) Ony Mngit 18 thus a subset of the set A con-
taining n+1 elements.
Our theorem has been proved by induction.
Thus we have proved the existence, in every infinite set, of a subset
having any finite number of elements. We have not proved, however,
that in every infinite set we can indicate a subset having a given finite
number of elements, or even that in every infinite set we can indicate
one of its elements.
On the other hand, it is obvious that if the set A contains, for every
natural n, a subset of n elements, then it is not a finite set, and con-
sequently it is an infinite set. Thus we have the following theorem:
A set is infinite if, and only if, for every natural number n it contains
a subset of n elements.

3. Fundamental properties of the relation ~. The relation ~ has the

following fundamental properties.
I. Reflexivity. For every set A we have AW~A.
A (1-1) correspondence between the elements of the set A is determined
by an identical mapping of the set A upon itself, 7. e. by the function g,
defined by the condition p(x) = « for each element « of the set A. Its
inverse function is equal to the function itself, 7. e. we have g(a) =
= (2) for re A.
Let us observe that a function inverse to a given function may be
equal to that function not only when giving an identical mapping. L. g.
for the function y, defined for positive numbers x by the formula g(a) =
= 1/x, we have, for x >0, ¢-(%) = g(x), in spite of the fact that, for
example, o(2) 4 2.
II. Symmetry. If A and B are sets such that A~B, then B~A.
 § 3. Fundamental properties of the relation ~ 25

This follows from the fact that, as we have proved in I.16, if the
function gy establishes a (1-1) correspondence between the elements of
the sets A and B, then its inverse function g establishes a (1-1)
correspondence between the elements of the sets B and A.
Ill. Transitivity. If A, B, C are sets such that A~B and B~C,
then AW~C.
If the function gy, establishes a (1-1) correspondence between the
elements of the sets A and B, and the function g, establishes a (1-1)
correspondence between the elements of the sets B and O, then, obviously,
the function ¢, defined by the condition p(x) = ¢,(y,(x)) establishes a (1-1)
correspondence between the elements of the sets A and C. The function
p (a) — gt (v2 (2)) for we C will be inverse to the function 9.

4. Effectively equivalent sets. From the definition of the equivalence

of sets it follows that in order to prove that two given sets A and B are
equivalent it is necessary and sufficient to prove that there exists a (1-1)
correspondence between the elements of the set A and the elements of
the set B. Let us observe, however, that to prove the existence of
such correspondence it is not necessary to establish it, or to give an
example of it. Its existence may be deduced from the axioms assumed
and the theorems proved previously, and an indirect proof is also
conceivable. :
On. the other hand, in order to prove that two sets are not equivalent
it is necessary and sufficient to prove that there exists no (1-1) cor-
respondence between the elements of both sets. The proof is usually
obtained by assuming that there exists a (1-1) correspondence between
the elements of the two sets in question and showing that this assump-
tion leads to a contradiction.
We are often unable to decide whether two sets are equivalent or
not. For instance in the present state of science we cannot deter-
mine whether or not the set of all prime numbers of the form 2”+1,
where » is a natural number, is equivalent to the set of all natural
numbers.
If we can establish a (1-1) correspondence (at least one) between
the elements of two given sets A and B, then we say that the sets are
effectively equivalent, and write Aef~B').
It is easy to see that the relation ef~, like the relation ~, is ref-
lexive, symmetric and transitive.

1) Effectivity is a logical concept belonging to metamathematics. See Sierpinski

[7], p. 112, Knaster et Kuratowski [1], p. 251, Kuratowski [1], p. 142, Lindenbaum [3],
p. 118, Sierpinski [58], p. 41, Sierpinski [16], p. 280-287.
26 II. Equivalent sets

It should be noticed that the relation ef ~, unlike the relation ~,

cannot be defined by means of logical concepts and the relation « alone.
The formula Aef~B implies of course the formula A~B. Later
on we shall see examples of sets A and B about which we know (in virtue
of the axioms assumed) that A~B but we cannot (at the present state
of science) prove that Aef~B.
Let » denote the function establishing a (1-1) correspondence between
the elements of given sets A and B. For each subset P of the set A denote
by 0#(P) the set of all the elements y(p) where p « P. Obviously, the func-
tion # establishes a (1-1) correspondence between all subsets of the
set A on one hand and all subsets of the set B on the other. (For QC B,
o-(Q) will be the set of all the elements g—(q) where qe«Q). Hence it
follows that if the sets A and B are effectively equivalent, then the set
U(A) of all subsets of the set A is effectively equivalent to the set U(B)
of all subsets of the set B. In other words:

ifAef~B, then U(A)ef~U(B).

5. Various theorems on the equivalence of sets. We prove
THHOREM: 1; 1f “A,B, A,, Bb; ate sets such thal ABZ, A, Bie,
Aef~A, and Bef~B,, then A+Bef~A,+B,.
Proof. Since A ef~A, and B ef~B,, we are able to define a func-
tion g, establishing a (1-1) correspondence between the elements of
the sets A and A, and a function g, establishing a (1-1) correspondence
between the elements of the sets B and B,.
Now define the function g(x) for ce A+B as follows: f we A+B,
then we have we A or we B, only one of these formulae being valid since
AB=0. If we A, take y(x%) = 9,(x), and if we B, take g(x) = 9,(x). AS
can. easily be seen, the function g(a) establishes a (1-1) correspondence
between the elements of the sets A+B and A,+B,. Thus we have
A -B ef ~A, +B,, q. ©. d.
It is easy to find out what modifications should be made in the
proof of Theorem 1 in order to prove
THEOREM la. If A, B, A, and B, are sets such that AB = 0, A,B, = 0,
Aw~A,, B~B,, then A+B~A,+B,.
And here is a generalization of Theorem 1:
THEOREM 2. Let us suppose that in a non-empty set Z three functions
have been defined: functions A(&) and B(&) where A(&) and B(E) are
sets (for €« Z), and a function pg, where, for E€Z, y= ge(p) is a fune-
tion defined for the elements p of the set A(é) and establishing a (1-1) cor-
respondence between the elements of the sets A(é) and B(é). Moreover, assume
 § 5. Various theorems on the equivalence of sets 27

that A(é)A(n) = 0 and B(é)B(n) = 0 for Ee Z, ne Z, EA. Let S denote

the sum of all sets A(&) where &€« Z, i. e.

s= YAS,
é€Z
and similarly
APB(eys
&€Z
Then Sei ~ T.
Proof. Let p denote any element of the set S. Since S is a disjoint
sum of the sets A(&é), where « Z, there exists for the element p one,
and only one, element & of the set Z, such that p« A(é,).
Let (p)=¢2,(p). We have of course ¢;,(p)<« B(é,), and since
B(é) C 1, O(p) = vz,(p) « T. In the set S we have thus defined the func-
tion 3(p), whose values are elements of the set 7. The function @ is in-
vertible in the set S. Indeed, if pe S, ge S and pq, then, in case of
& = &,, we have z,(p) A @z,(¢), since the function g;, is invertible in
the set A(é,), and, in view of (p) = 9,(p) and 9(¢)= ¢z,(¢) ~ %,(P),
we have 0(p) # #(q); and in case of & ~ &, we have 0(p) = ¢z,(p) « B(&),
3(q) = ¢2,(q) ¢ B(Eq), and since, in view of & ~ &,, we have B(é,) B(é,) = 0,
therefore we must have #(p) 4 0(q). Finally, suppose that te 7. Thus
there exists an element of the set Z such that t« B(&,). Since the fune-
tion gq, establishes a (1-1) correspondence between the elements of the
sets A(é,) and B(é,), we have u = gz, (t) « A(&,)C S, and it follows from |
the definition of the function 0 that d(w)=t. Thus for such element t
of the set 7 there exists an element uw of the set S such that #(u) = t.
From the already proved properties of the function @ it follows that
this function establishes a (1-1) correspondence between the elements
of the sets S and 7. Thus we have proved Theorem 2.
It is easy to see how to weaken the assumptions of Theorem 2 in
order to get the statement S~T instead of the statement S ef~T.
The following theorems can easily be proved.
THEOREM 3. If A, B, A,, B, are seis such that A ef~A, and Bef~B,,
then A xBef~A, xB,.
THEOREM 4. If A,B, A,, B, are sets such that A~A, and B~B,,
then A xXB~A, xB,.
EXERCISES. Prove that if A -B~B—A, then A~B, but not necessarily vice versa.
Proot. We have, as we know, the decompositions 4 = (A—B)+AB and B=
= (B—A)+AB, and the equalities (A—B)-AB=0 and (B—A)-AB=0. According
to Theorem 1 (in which A and B should be replaced by A—B and AB respectively,
and A, and B, by B—A and AB respectively), and in view of 4—B~B—A and AB~AB,
we obtain (A—B)+AB~(B—A)+AB, and therefore A~B, q. e. d.
28 Il. Equivalent sets

On the other hand, let A denote the set of all integers, and B — the set of all
natural numbers. As already proved, A4~ B, but the relation A—B~B—A does not
hold since the set B—A is empty while the set 4 —B is not empty (having as elements 0
and all negative integers).
2. Prove that if d~A, and B~B,, then AB~ A,B, does not necessarily follow.
Proof. Let 4 = B= A, = {1}, B, = {2}. Then we have A~A,, B~B,, AB= {1},
A,B, =0, whence it follows that AB~A,B, does not hold.

THEOREM 5. If A, B, A,, B,, are sets such that A ef~A, and B ef~B,,
then A® ef~Aj?. :
Proof. It follows from the assumption of Aef~A, and Bef~B,
that we are able to establish a (1-1) correspondence gy between the ele-
ments of the sets A and A, and a (1-1) correspondence wy between the
elements of the sets B and Be
Let f « A’. According to the definition of the set A® (1.18) a function
f(wv) is defined for we B and such that f(a) « A for we B. To the function f
let us relate the function g = 0(f) defined for xe B in the following way:
§(e)= elt (y-())) for we B. Obviously, we have ge Af? and f(y) =
= 9 (9 (p( y))) for y e B, whence it follows that the function # establishes
a (1-1) correspondence between the elements of the sets A® and Ape
Thus we have A’ ef~A?, q.e. d.
Sunilarly we prove
THEOREM 6. If A,B, A,,B, are sets such that A~A, and B~B,,
then A?~A??,
Next let us prove
THEOREM 7. If A, B, C are sets such that BC = 0, then A8+©
ef ~A® Xx AS.
Proof. Let fe A?+¢. Therefore a function f(x) is defined for
ce B+C and such that f(x)« A for re B+C. By g=g(f) denote the
function g(x) defined for ce B by the equality g(#)=f(#), and by
h=~yp(f) denote the function h(w#) defined for «eC by the equality
h(x) = f(z) Obviously, we shall have g« A?, he A°, therefore
(g,h)« A®?x AC. Thus to each element f of the set A2+° corresponds an
element #(f) = (p(f), v(f)) of the set A?x A‘. We shall show that the
function ? establishes a (1-1) correspondence between the elements of
the sets A?+° and A?x AS. Indeed, let f « A?+°,f,<¢ A?*°,fAf,. Thus
there exists an element x7, « B+C (at least one) such that f(a) < f(a).
Since 7,« B+C, we have #,<B or w «CC. If a, ¢ B, then for g= g(f),
gi = plfi) we shall have g(%) = f(#o), 91(@o) = fi(@o); therefore g, 4 g and
thus, setting h = y(f), hi = y(f,), we shall have (g, h) 4 (g,, h,), therefore
O(f) AO(f,). Thus the function # is invertible in the set A®+C.
It remains to show that for each element 7 of the set A? AC there
exists an element f of the set A+© such that r= #(f).
 § 5. Various theorems on the equivalence of sets 29

Therefore, let r<¢ A? x AC. Thus r = (g, h) where g« A?, he AT. Now
let « denote an element of the set B+C. With regard to BC = 0 one,
and only one, of the two formulae #w¢« B and we C holds. If veB, let
f(x) = g(a); if we C, let f(v~) = h(a). The function f(x) will thus be defined
for «« B+C, while obviously f(v) « A. Thus we shall have f « A?+°. It
follows at once from the definition of the functions » and y that

g=¢(f), h=vlf),
therefore r= (g,h) = B(f), q. e. d.
It is easy to see that an immediate corollary of Theorem 7 is
THEOREM 7a. If A, B, C are sets such that BO = 0, then A®+*°C~A®X AS.
THEOREM 8. For any sets A, B,C we have (Ax B)° ef ~ACX BS.
Proof. Let fe(AxB)°. Thus f is a function f(x), defined for we C,
whose values are elements of the set A xB. For each x « C we thus have
f(x) = (u,v) where we A and ve B. Since the elements uw and v depend
on #, we can write u=g(x),v=h(z), while of course ge A°, he B°,
and thus (g,h)« ACx BS. Thus to each element / of the set (A x B)¢
have related an element (g, h) = #(f) of the set Acx BC.
Now we shall show that the function &? establishes a (1-1) corres-
pondence between the elements of the sets (Ax B)° and ACx BS.
Let fe(AxB)°,f,¢(AxB)°,f~Af,. Thus there exists an element
#,« C_such that 7(2%) © J,(@)- ihe OGY aA, 0, 8 (Fe (Gartlig) LO ere iC.
we shall then have f(x) = (g(x), h(a yey (onlay, h,(a)), therefore
(g (0); h (ao) i (91(%o) 5 hy(&»)) and thus either
g(#9) 7 g1(o) or h(%) A hy(Ho),
or both. If g(@,)#9,(v) then g+~9,, and if h(a.) ~ h(a), then h ~ hy.
Thus in any case (g, h)~(g,, h,), therefore #(f) 4 O(f,). Thus the fune-
tion # is invertible in the set (Ax B)¢.
It remains to show that for each element r of the set A° x BC there
exists at least one element f of the set (A xB)° such that r= #(f). Let
re A°x BS, and so r= (g,h) where ge A©, he BC. Thus g is a function
g(x) defined for x « O such that g(#) « A for we C, and h is a function h(x)
defined for «« C and such that h(w7)« B for we C. Let f(x) = (g(x), h(a))
for «eC; we shall have f(x)eAXxB for we C, therefore f<«(AxB)¢,
while +r = (g, h) = &(f) follows from the definition of the function #.
Thus we have proved Theorem 8.
It is easy to see that an immediate corollary of Theorem 8 is
THEOREM 8a. For any sets A, B, C we have (AX B)°~ACX
BS,
THEOREM 9. For any sets A, B, C we have (A®)° ef ~ A®XS,
Proof. Let f«(A)°. Thus j is a function f(x), defined for we C,
such that f(x) « A¥, v. e. f(x) is a function g, defined for y « B, such that
gxly) « A. ad let ge BXC; therefore = (y;#), where ye B, we C.
30 II. Equivalent sets

Let h(z) = gx(y): we shall have h(z)<« A for ze BXC, therefore h « A?*¢.
Thus to each element f of the set (A?)° we have related an element h=
= b(f) of the set A?*°. We shall show that the function ? establishes
a (1-1) correspondence between the elements of sets (A?)° and A®*C¢.
Let f « (A), f’ « (A%)°,f #f’. Thus there exists an element 2) « C such
that f(a) A f(x). But f(a) is a function g, where g,« A and, similarly,
f'(w) is a function gi, where g; « A. Therefore, since f(x.) 4 f(x), we have
Jxo F Jxo, and there exists an element y, of the set B, such that g..(Yo)~
F JxiYo). Let 2 = (Yo; Bo), h= AS), ho —O(f'). From the detinition of.
the function % it follows that h(Z) = Gx (Yo) and h'(2o) = Jxo(Yo), therefore:
h(Z) A h'(29), which proves that h ~ h’. Thus the function @ is invertible:
in the set (A)°. It remains to show that for each element A of the set.
A%xC there exists at least one element f of the set (A”)° such that the equal-
ity h=9#(f) holds.
Let h« A®*¢; thus fh is a function h(z), defined for z« BxC, such
that h(z)« A for ze BXC. For we C, let us denote by f(x) a function
gx(y) of the variable y, defined for y « B by the formula g,(y) = h(z) where
z= (y,«x). We shall have f(x) « A® for we C, therefore f « (A?)°, and it
follows at once from the definition of the function # that 0(f) = h.
Thus we have proved Theorem 9.
It is easy to see that an immediate corollary of Theorem 9 is
THEOREM 9a. For any sets A, B, C we have (A®)°~A®*XC.
As an example of a theorem which is of the same kind as Theorems 7,
8 and 9 but is difficult to prove we give without proof the following
theorem:
If A, B, C, D are sets such that AB= CD=0, Aei ~B, Cef ~D
and A+Bef ~C+D, then A ef ~C.
Proof. See Sierpinski [8], p. 1-6.

6. The Cantor-Bernstein Theorem. We shall prove

THEOREM 1. If A, B, A,, B, are sets such that A,C A, B,CB,
Aef ~B,, Bef ~A,, then A ef ~B.
Proof. Since A ef ~B, and B ef ~ A,, we are able to define a func-
tion gy establishing a (1-1) correspondence between the elements of the
sets A and B, and a function yp establishing a (1-1) correspondence between
the elements of the sets B and A,.
Let x, denote any element of the set A. If w « A,, then it follows
from the definition of the function y that there exists one, and only
one, element y of the set B such that y(y) = a. We denote this element
by y,. If, however, 7, ¢ A,, then we shall say that the element y, does
not exist.
 § 6. The Cantor-Bernstein theorem 31

If the element y, exists and y,¢ B, then it follows from the defini-
tion of the function g that there exists one and only one element « of
the set A such that y(x#)= y,. We shall denote this element by 7,. If,
however, y, ¢ B,, then we shall say that the element x, does not exist.
Generally, let us suppose that for a given natural number k there
exists an element 7,. If x, « A,, then by yxz41 we shall denote that single
element of the set B for which w(y,z41) = %, and if #,¢ A, we shall
say that the element y;,; does not exist. Similarly, if there exists an ele-
ment y, and y,« B,, then by a, we shall denote that single element of
the set A for which y(a,) = y;,, and if y,¢ B,, we shall say that the ele-
ment 2, does not exist.
Clearly, for each element x, of the set A one and only one of the
following three cases occurs:
1° For every natural number & there exist elements a, and y;;
2° For a certain natural number k there exist elements y,, 7%, y2,
Lay «+5 Ye-15 He-1, but the element y, does not exist;
3° For a certain natural number k there exist elements ¥,, 2, Y2,
May +009 Ye-1) Uk-1y Yk, Dut the element x, does not exist.
If the first or the second case occurs, we shall take 3(x) = p(X),
and if the third case occurs, we take #(a) = y, 7. e. B(a%>) = yp(Xp).
Thus we have defined the function 0(#) defined for we A. We shall
show that it establishes a (1-1) correspondence between the elements
of the sets A and B.
Let us suppose that for the elements x, and ao of the set A we have
P(X) = B(x). Case 1° or case 2° cannot occur for a, and case 3° cannot
occur for 2%, since we should then have #(a) = (4%), B(%o).= po),
whence (2) = y~(#), Which would give a = p (yp(xo) and therefore
y(%) = ¥, and x2 = a. Still it is clearly always the same of the three
cases 1°, 2°, 3° that arises both for x and #,. Similarly we prove that
case 1° or case 2° cannot occur for x), and case 3° cannot occur for «7.
Thus we have either 3(xo) = 9(%), (x5) = (2), 4. & (2) = v(x),
which, in view of the function being invertible in the set A, gives % = a,
or else 3(%) = pao), O(x,) = p'(a,), 4. & pHa) = p(x), which gives
Ly = x. Thus we have proved that the function # is invertible in the set A.
It follows at once from the definition of the function # that its values
are the elements of the set B.
Now let y denote any element of the set B. Let 2) = p(Yo); # will
be an element of the set A, and, according to our notation, y, will be the
element y, corresponding to a, 1. €& Yo = %-
If there exists no element x, corresponding to x, then case 3° occurs,
and we have #(a%)) = y,. From y,= y, we know that for y, there exists
an element 2 = %, of the set A such that (x) = yp.
32 II. Equivalent sets

Suppose now that for x, the element «, exists. If for x, and thus
also for 2,, case 1° or case 2° occurs, then we have #(x%,) = 9(%,). But,
according to the definition of the element 7,, we have (#,) = y,. Thus
O(xv,) = y,, and, since y, = yy, we conclude that there exists an element
a” = #, of the set A such that 3(x) = yy. Finally, if for a case 3° occurs,
then we have #(x%)=y,, and, since y,= Y), we conclude that there
exists an element © =v, of the set A, such that #(7) = yy.
Thus we have proved that for each element y, of the set B there
exists an element x of the set A such that #(x) = y,. Since, as we have
proved above, the function # defined in the set A is invertible in this
set and its values are the elements of the set B, # establishes a (1-1) cor-
respondence between the elements of the sets A and B, q.e. d.
Thus Theorem 1 has been proved.
Now if we replace, in Theorem 1, the formulae Aef~B, and
Bef ~A, by the formulae A~B, and B~A, and in the first sentence of
the proof of that theorem the words “we are able to define the func-
tion gm and the function »“ by the words “there exists a function m and
a function y“, then, as can easily be seen, we should obtain the proof of
THEOREM 2 (of Cantor-Bernstein). If A, B, A,, B, are sets such that
A,C A, B,C B, A~B,, B~A,, then A~B.
The proof of the Cantor-Bernstein theorem given above is based on
an idea of J. Konig. Two more proofs of this theorem will be found in
my book (Sierpiiski [58], p. 145-148).
The theorem of Cantor and Bernstein can also be formulated in the
following way:
Two sets of which each is equivalent to a subset of the other are equi-
valent.
Now let us apply Theorem 1 to the case of B= A. Replacing B by C,
and taking into account the symmetry and transitivity of the relation
ef ~ we obtain
THEOREM 3. If A, B, C are sets such that AD BIC and Aef ~C,
then Aef ~B and Cef ~B.
Similarly, from Theorem 2 we obtain
THEOREM 4. If A, B,C are sets such that AD BOC and A~O,
then A~BW~C.
In other words: if the set A is equivalent to its subset OC, then each set
intermediate between C and A (1. e. containing C and contained in A) is
equivalent both to the set A and to the set C.
EXERCISES. 1. Prove that if X5 Y and Yef ~Y+Z, then X ef ~X-+ Z.
Proof. We have Yc ¥+(Z—-X)c Y+ Z, whence, since Yef ~Y+Z and in
virtue of Theorem 3, we find Y ef~Y+(Z—X).
 § 6. The Cantor-Bernstein theorem 33

But X+Z=(X—YV)+[¥+(Z—-X)], the components X—Y and Y+(Z—-X)

and similarly the components X—Y and Y being disjoint. Thus in virtue of I1.5, Th. 1,
we have X = (X—Y)+YV ef ~(X—Y)+Y+(Z-X) = X+ Z, therefore X ef~X+ Z,
qaead.
2. Let us suppose that we are given, on a plane, a rectangular system of co-
ordinates with the origin O and a unit of length. Prove that the set A of all straight
lines on the plane is effectively equivalent to the set B of all points of the plane.
Proof. Let p denote a given straight line lying on our plane. If the line p does
not pass through the point O, denote by Q the point of the line p nearest to O, and by P
the point lying on the extension of the segment OQ at unit distance from the point Q.
If the line p passes through the point O but is not the axis of abscissae, denote by P
the point with a positive ordinate at which the line p intersects the circle with unit
radius and centre O. Finally, if the line p is the axis of abscissae, denote by P the point
on the axis of abscissae with the abscissa 1. In each case we relate to the straight line p
the point P = 9(p).
It is easy to prove that the function y is invertible in the set A, and its elements
are the values of the set B. Thus denoting by B, the set of all the points p(p) for p<« A
we shall have A ef ~B,.
On the other hand, for every point P of the plane different from O, let us de-
note by y(P) the straight line passing through the point P and perpendicular to the
segment OP, and take w(0) as the axis of abscissae. Itis easy to prove that the function
y(P) is defined for P « B and invertible in the set B. Thus, denoting by A, the set of
all the straight lines y(P) for Pe B, we shall have Bef ~A,.
The formulae A ef ~B, and B ef ~A, give, according to Theorem 1, A ef ~B,
q. 6. d.
3. Prove that if A ef ~B,c Bef ~0,c Oef ~A,c A, then A ef ~C and A ef ~B.
Proof. Since Bef ~O(, we are able to define the function % establishing a (1-1)
correspondence between the elements of sets B and O,. Since B,c B, the function # is
invertible in the set B and also in the set B,. Let us denote by OC, the set of all the ele-
ments #(p) for pe B,. Obviously, the function @ establishes a (1-1) correspondence
between the elements of the sets B, and C,. Thus B, ef ~O,, and since A ef ~B,, we
infer, in view of the transitivity of the relation ef ~, that A ef ~C,. But from the
definition of the set C, and the function # it follows that O,c OC. Thus A ef ~0,c 0
and, according to our assumption, O ef A,~A, whence, in virtue of I1.6, Th. 1, we
obtain A ef ~C. Similarly, in view of the cyclic character of the assumptions of our
exercise, we prove that Bef ~A, whence we infer, in virtue of the symmetry of the
relation ef ~, that A ef ~B.
4, Prove that if A, B, C, D are sets such that 42> C0, Bc D and 0+D~Q0O, then
A+BWA.
Proof. Since 450, we have 4+ B= (A—O)+[C+ (B—A)], while (A—O)[(C+
+(B—A)]=0. But Cc 0+ (B—A)c C+D, because Bc D. From C+D~CO, we infer,
in virtue of II.6, Th. 4, that C+ (B—A)~O, and since the set A —O, as we have proved,
is disjoint with the left-hand side of this formula, as well as with its right-hand side,
we have, in virtue of IJ.5, Th. la, (A—0)+ [0+ (B—A)]~(A—O)4+0,i.¢. A+ BW~A,
16 Ol.

Cardinal and ordinal numbers eS)

 CHAPTER III

DENUMERABLE AND NON-DENUMERABLE SETS

1. Denumerable and effectively denumerable sets. A set which is

equivalent to the set of all natural numbers is called denumerable or
countable, and a set which is effectively equivalent to the set of all natural
numbers is called effectively denwmerable.
If a given set A is effectively denumerable, then we are able to esta-
blish a (1-1) correspondence between the elements of the set A and the
natural numbers. In other words, we can assign indices to the elements
of the set A in such a way that each element of the set A will have its.
own index, different elements of the set A alvays having different indices
and every natural number appearing as the index of a certain element
of the set A. If the index of a given element of the set A is n, then we can
denote this element by a,. We shall then have

Ce A = {yy gy Ug, .-}5

in other words, all the elements of the set A will be arranged in an infinite
sequence @,, d,... Of different terms.
Conversely, if we are given an infinite sequence of different terms
(i. e. a function defined in the set of all natural numbers and invertible
in that set), then the set consisting of all terms of this sequence is effectively
denumerable.
Thus we can say that a given set is effectively denumerable if and
only if we are able to arrange all its elements in an infinite sequence.
If we know that a given set is denumerable, we can say that there’
exists an infinite sequence composed of all the elements of the set in
question, and vice versa.
If the set A is effectively denumerable, we are able to arrange its
elements in an infinite sequence in many ways. For example, we could
arrange all the elements of the set A in a sequence different from that de-
termined by (1.1), e. g. replacing the element a, in this formula by ele-
ment a,, and vice versa, or, more generally, replacing a, by a, and vice
versa, or substituting for every even index an index smaller by 1 and
 § 1. Denumerable and effectively denumerable sets 35

for every odd index an index greater by 1, which will give the infinite
sequence:
Az, Ay, Ag, Az, Ag, Ug, +++ Mak, Aok—1y Aek4+2) Aak+1; +
Similarly, for every denumerable set there exist many infinite se-
quences composed of all its elements.
THEOREM 1. Hach subset of an effectively denumerable set is a finite
set or an effectively denumerable set.
Proof. Let A denote an effectively denumerable set. We are able
to arrange all the elements of the set in an infinite sequence
(1.2) Ors es ign s-:
Now let B denote a subset of A which is not a finite set. Let us re-
move from the sequence (1.2) all those terms which are not elements of
the set B (without changing the order of the remaining terms of the se-
quence). In this way what remains of the sequence (1.2) will be a well-
defined infinite sequence, which of course will be composed of all the
elements of the set B. Thus we are able to arrange the elements of the
set B in an infinite sequence, and therefore the set B is effectively denu-
merable, q.e. d.
In particular, every infinite set of natural numbers is effectively denu-
merable (as an infinite subset of the effectively denumerable set of all
natural numbers).
The concept of effective denumerability is understood differently
by E. Borel. In his book (Borel [3]) on p. 229-232, in § 66, entitled Les
ensembles dénombrables et les ensembles effectivement énumérables, he as-
serts that an infinite subset of an effectively denumerable set need not
be an effectively denumerable set, and gives several examples of this,
among them one which is somewhat like the following. Let H denote
the set of all natural numbers which appear infinitely many times as
denominators in the development of the number z into a continued arith-
metical fraction. If the set H is infinite, then, according to Borel, it
is not effectively denumerable (although it is a subset of the effectively
denumerable set of all natural numbers), because we do not know the
smallest number belonging to it. Of course, we are not able to calculate
this number, but we can define it in a unique manner as the smallest
natural number belonging to H. Thus, what Borel calls effective de-
numerability is the caleulability of every term of the sequence’). I have
1) In his note (Borel [2], p. 164) E. Borel calls effectively denumerable every set
given in the form u,, u,,..., where every term wu, is known if its index nm is given and
conversely (cf. also Sierpinski [57], p. 42-43). The question is what should be meant
by a “known term. We understand by it an object which is effectively defined; for
Borel “a known number“ means a calculable number.
3%
36 Il]. Denumerable and non-denumerable sets

devoted more space to the concept of calculability, as well as to the dif-

ference between the effective definition of a number and its calculability
in my book (Sierpinski [58], p. 41-42) and the reader is referred to it in
this matter.
It can easily be seen what modifications should be made in the proof
of Theorem 1 in order to prove
THEOREM la. Hach subset of a denumerable set is a finite set or
a denumerable set.

2. Effective denumerability of the set of all rational numbers. We shall

prove
THEOREM 1. The set of all rational numbers is effectively denumerable.
Proof. As we know, there is only one way of representing a rational
number w in the form of an irreducible fraction p/q with a natural denomi-
nator and an integer numerator. (The fraction 0/1 should be regarded as
such a representation of number 0).
Now let us divide all rational numbers into classes assigning to the
k-th class all those irreducible fractions p/q for which |p|+q = k (where
{p| denotes the absolute value of the number p). Thus every rational
number will belong to a certain class. In every class we shall have a finite
number of rational numbers, which can be arranged in order of their
magnitude, since all the elements of the k-th class will clearly be obtained
from the sequence

Seal) =oa 2) DSi Ss 2 k—-2 k—1

Ry AMOI re eds sa bY ey ec omy ae eo era
by removing from it those fractions which can be reduced.
Now let us arrange in succession the elements of the first class, then
the elements of the second, the third etc., placing the elements of each
class in order of increasing magnitude. In this way we shall obtain the
well-defined infinite sequence

} i) x99 4) #4; ay

in which we shall find every rational number and find it only once. This
proves that the set of all rational numbers is effectively denumerable,
Quiesd.
If we wanted to determine the place occupied in our sequence by
a given rational number (e.g. 22/7) it might require long calculations;
this, however, is of no importance, since our aim has been only to establish
a (1-1) correspondence between rational numbers and natural numbers,
and not to apply this correspondence in practice.
 §2. Effective denumerability of the set of all rational numbers 34
da

From Theorem 1 and from the proof of III.1, Th. 1, we get some-
thing more than the information that every infinite set of rational numbers
is effectively denumerable. Namely, there obviously follows
THEOREM 2. In the set Z of all infinite sets of rational numbers we
are able to define a function y such that, for Pe Z, p(P) is an infinite se-
quence composed of all different elements of the set P.
In Theorem 2 we can of course replace the set Z by the set of all
infinite subsets of an arbitrary effectively denumerable set.

3. Effective denumerability of the infinite set of non-overlapping intervals.

Let a and b> a be two given real numbers. The set of all real num-
bers x satisfying the inequality a <a <b is called the interval [a, 6];
the numbers 4 and b are called the end-points of the interval [a, b], a — the
left-hand end-point, b — the right-hand end-point.
A given number @, is said to belong to the interval [a, b] or to be a point
of that interval if a <a <b; it is said to lie inside the interval [a, b] or
to be an interior point of that interval if a < 7 <b. The number b—a is
called the length of the interval [a, b].
The interval [c,d] is said to le inside the interval [a, b], if the end-
-points c and d le inside the interval [a,b], 7.e. if a<e<d<b.
We say that two given intervals [a, b] and [¢, d] do not overlap if they
possess no interior points in common. It is easy to show that then either
b<cord<a.
EXERCISE. Prove the following theorem of A. Denjoy: If three given imtervals
possess an interior point in common, then at least one of those intervals is such that
each of its interior points is also an interior point of at least one of the remaining two.
intervals. .

THEOREM 1. Every infinite set of non-overlapping intervals is effeetively

denumerable.
Proof. Let Z denote a given infinite set of non-overlapping inter-
vals. Let 6 denote a given element of the set Z. Thus 6 is an interval
[a,b] where a < b. Let m denote the least natural number greater than
1:(b—a), and let k = Ema (where Et denotes the greatest integer < t);
thus we shall have k+1> ma, therefore (k+1):m > a, and, since k < ma
and 1/m< b—a, we find (k+1):m <a+(b—a)=b. Thus the rational
number (k+1):m lies inside the interval [a, dD].
Let y(6) = (k+1):m. In this way we have defined, for 6 « Z, a func-
tion having rational values, y(6) being, for 6« Z, a number lying inside
the interval 6. Hence it follows immediately that the function (6) is.
invertible in the set Z, since if 6, ¢« Z, 6,¢ Z, 6, #6,, then the intervals:
6, and 6, do not overlap and thus have no interior points in common;
38 III. Denumerable and non-denumerable sets

therefore y(6,) = y(d,) is impossible since g(6,) lies inside the interval 6,
while y(é,) lies inside the interval 6,.
Denote by T the set of all numbers y(6) where 6 « Z. Since the func-
tion @ is invertible in the set Z, the set Z is effectively equivalent to the
set 7. Thus the set 7 is infinite since the set Z is infinite. But the set T
is a subset of the set of all rational numbers, which, according to III.1,
Th. 1, is effectively denumerable. In view of the transitivity of the rela-
tion ef ~, it follows that the set Z is effectively denumerable. Thus we
have proved Theorem 1.
We might also prove Theorem 1 in another way, observing that if
we are given an interval 6, then each set of non-overlapping intervals,
lying in the interval 6, and whose lengths are not less than a given positive
number, is always finite. Therefore let us first take the intervals of the
set Z having lengths >1 and lying in the interval [—1, 1] (if there are
such intervals in the set Z), arranging them, for instance, from left to
right; similarly, the intervals of the set 4 having lengths >1/2, and
lying in the interval [—2, 2], which have not yet been taken into account;
next, similarly, the intervals having lengths > 1/3, lying in the interval
{[—3,3], etc. In this way we obtain a well-defined infinite sequence,
made up of all the intervals of the set Z.
EXERCISES. 1. Prove that if D is a denumerable set of points on a plane, then D
is the sum of two sets, of which one is finite on every straight line parallel to the axis
of abscissae, and the other is finite on every straight line parallel to the axis of ordinates.
Proof. Let D be a denumerable set of points on a plane. Let D, denote the set
of all the abscissae of the points of the set D, and D, — the set of all the ordinates of
the points of the set D. The sets D, and D, are of coursé finite or denumerable: thus
there exist finite or infinite sequences ,, 7, ..., and Y,, Y2,..., composed, respectively,
of all the elements of those sets. It can easily be seen that if the first of those sets is
finite, then each straight line parallel to the axis of abscissae contains a finite number
of points of the set D and, setting D = D+ 0, we shall have the desired decomposi-
tion of the set D. Similarly, if the sequence y,, yo, ... is finite, then D = 0+D will be
the desidered decomposition of the set D. Therefore let us assume that the sequences
24, Ge, --- aNd Yi, Ya, --- are infinite.
The set D is of course contained in the set of all points of the plane (x,, y,), where k
and J are natural numbers.
Let A denote the set of all the points (x,, y,) of the set D for which k <1 and
let B= D—A. We shall have, of course, D= A+B,
Now let y = b.be a given straight line parallel to the axis of abscissae. If b is none
of the terms of the sequence y,, yz, ..., then, ebviously, the straight line y = b contains
no point of the set B, and, therefore, no point of the set A. If, on the other hand, 0 is
one of the terms of the sequence ¥,, Yo, ..., for example b = y,, then, as follows at once
from the definition of the set A, the straight line y = b contains only the points (x,, b)
of the set A where k <1, and thus at most J points of the set A.
Similarly, let « = a denote a straight line parallel to the axis of ordinates. If a is
none of the terms of the sequence a,, #,..., then the straight line =a contains no
point of the set D, and, therefore, no point of the set B. If, on the other hand, a = z,,
 § 3. The infinite set of non-overlapping intervals 39

then, in virtue of the definition of the set B, the straight line x = a contains only the
points (a, y,) of the set B, where / < k, and thus less than & points of the set B.
Thus the desired proof is completed.
2. Prove that if f(a”) is a function having real values, defined in the set A composed
of real numbers, and increasing in the set A (i. ec. such that f(x,) < f(a.) for a <A,
w,¢ A, %, < #2), then the set of all points of the set A at which the function f(a) is not
continuous in the set A is finite or effectively denumerable.
Proof. We shall prove first that the set A, of all points of the set A at which the
function f(x) is not continuous from the right in the set A, is finite or effectively de-
numerable. By III.2, Th. 1, we are able to define the infinite sequence 7,, 72, ..., formed
of all rational numbers. Now let a « A. We shall show that there exists a rational num-
ber 7 such that f(a) < r and f(x) > r for xe A, x > a. Indeed, if there were no such
rational number z, then for each rational number r > f(a) there would exist an element
b<« A such that b > a and f(b) <r. Since the function f(x) is increasing in the set 4,
we should have f(a) < f(x) < f(b)<yr, for each element xe A such that a<a<b,
therefore f(a) < f(x) <r, which, in view of the arbitrariness of the rational number
7 >a (on which the number b > a depends) proves that the function f is continuous
from the right at the point a, in the set A, contrary to our assumption. Thus there exists
the first term 7, of the sequence 7,,7,,..., such that f(a) < 7, and f(x) > 7, for ve A,
x >a;letr(a)=7,. Now let us suppose that ae A,, a’ « A,, a 4 a’, for instance a < a’.
Thus we have f(a’) < r(a’) and, since a’>a, f(a’) > r(@), we have r(a) < r(a’). Thus
the function r(a) is determined for @ « A, and relates different terms of the sequence
1,.%2,... to different elements of the set A. Hence it immediately follows that the set A,
is finite or effectively denumerable.
Similarly we prove that the set A, of all points of the set A at which the function f
is not continuous from the left in the set A is finite or effectively denumerable. And
since the set of all points of the set A at which the function f/ is not continuous in the
set A is the sum of the sets A, and A,, it follows from [1.5, Th. 1, that this set (4,+ A.)
is finite or effectively denumerable, q. e. d.

4. Effective denumerability of the set of all finite sequences of rational

numbers. We shall prove
THEOREM 1. The set of all finite sequences of rational numbers is
effectively denumerable.
Proof. Let Z denote the set of all finite sequences of rational numbers
(not necessarily different numbers). Let s« Z. Thus s is a finite sequence
of rational numbers, s = (7,, 725 .--, 7m), Where m is a natural number
and 71, 2) +») 1m are. rational numbers. For k = 1, 2,..., m, let r, = pzfde,
where p;/q, denotes an irreducible fraction with an integer numerator
and a natural denominator. Let

(4.1) iy = ||| Po]. r |Pm| + 41+ fet ee Om-5

n will be a natural number, well-defined by sequence s. We shall assign
this sequence to the n-th class.
In this way each sequence belonging to the set Z will belong to
a certain class with a natural index number n.
40 III. Denumerable and non-denumerable sets

We shall show that in every class there will be a finite number of

sequences.
Let n denote a given natural number, s — a sequence of the n-th
class. Since q,(k=1,2,...,m) are natural numbers and |p,|(k=
=1,2,...,m) are positive integers or zero, formula (4.1) implies the
inequalities n >m, n>qx, n> |px| for K=1, 2,..., m.
The first of these inequalities shows that the sequences of the n-th
class contain no more than » terms each; the remaining two inequalities
show that the numerator and the denominator of each of those terms
can assume only a finite number of different values. Therefore the number
of all the sequences of the n-th class must also be finite and, as can easily
be found, less than (2n?7)".
We can write all the sequences of the n-th class in a certain
order, e. g. always putting down first that one of two n-th class sequences
which has fewer terms, and if the number of terms is equal, applying the
so-called principle of first differences, 1. e. writing first that sequence
in which we first come across a term which is smaller than the term oc-
cupying the same place in the other sequence.
Now let us write down, in the order thus fixed, all the sequences
of the first class, then all the sequences of the 2-nd class, next those of
the third class, etc. Thus we shall obtain a well-defined infinite sequence,
whose terms will all be finite sequences of rational numbers 1+). Thus
we have proved Theorem 1.
As can. easily be seen Theorem 1 of III.2 is an immediate consequence
of Theorem 1 and Theorem 1 of III.1., since the set of all rational numbers
is of course equivalent to the set of all one-term sequences of rational
numbers.
Another immediate consequence of Theorem 1 and Theorem 1 of
III.4 is the theorem stating that the set of all ordered pairs of rational
numbers is effectively denumerable. Hence it follows that the set of all points
of a plane which have rational coordinates is effectively denuwmerable.
Similarly we deduce from Theorem 1 of III.1 and Theorem 1 of ITI.4
that the set of all three-term sequences of rational numbers is effectively
denumerable, whence it follows that the set of all points of a space which
have rational coordinates is effectively denumerable, and likewise that the
set of all so-called rational circles lying in a plane, 7. e. circles whose cen-
tres have rational coordinates and whose radii have rational lengths,
is effectively denumerable.
1) It is easy to see that the first terms of that sequence will be the sequences
(given in brackets):
(0), (—1), (1), (0, 0), (—2), (— 2), (4), (2), (-1, 0),
(0, —1), (0, We (Be 0), (0, 0, 0), (—3), (3) OOK
 § 4. The set of all finite sequences of rational numbers 41

Similarly we prove that the set of all rational spheres in a space is

effectively denumerable.
As another consequence of Theorem 1 of III.1 and Theorem 1 of
III.4 we shall prove
THEOREM 2. The set of all finite subsets of an effectively denumerable
set is effectively denumerable.
Proof. Let A denote a given effectively denumerable set, U(A) —
the set of all subsets of the set A, N — the set of all natural numbers.
Thus A ef~WN, whence, as proved at the end of I1.4, it follows that
U(A)ef ~U(N). The (1-1) correspondence between all subsets of the
set A and all subsets of the set N obviously also establishes a (1-1) cor-
respondence between the set P of all finite subsets of the set A and the
set Q of all finite subsets of the set NV. Thus, in view of the transitivity
of the relation ef ~, in order to prove Theorem 2 it suffices to show that
the set of all finite sets of natural numbers is effectively denumerable.
But this is an immediate consequence of Theorem 1 and Theorem 1 of
III.1 since every non-empty finite set of natural numbers determines
a certain finite sequence of natural numbers, constituting the elements
of the set in question arranged in order of magnitude. Thus we are able
to arrange the set of all non-empty finite sets of natural numbers in an
infinite sequence. If, moreover, before the first term of that sequence
we place the empty set, then we shall obtain the infinite sequence of all
finite sets of natural numbers. Thus we have proved Theorem 2.
It will be observed that a (1-1) correspondence between all finite
sequences of natural terms and all natural numbers can be established
by assigning to the sequence (7, %2,...,x) the natural number

Oe = Data —1
SRS eg a ata
Teas
-1

EXERCISES. 1. Prove the following theorem of A. Lindenbaum 1):

Every infinite sequence {u,} of rational numbers is a sum of three sequences,
Un, = Utui't+u” (n= 1, 2,...), each of which is composed of all different rational
numbers.
Proof. Let r,,7,,... denote a given infinite sequence, composed of all different
rational numbers. Let uj = 7,, uj’= u,, ui’’= —7,. Now let m denote a given natural
number greater than 1 and suppose that we have already defined all the terms uf?
where 1 = 1, 2,3, and k <n.
If n = 31—2 where I is a natural number, then let u/ denote the first term of the
sequence 7,,72,..., different from any of the numbers uz, where k <n; let i
denote the first term of the sequence 7,, 72, ..., different from any of the numbers u}/
and, also different from any of the numbers u,—u/—uj,’ for k <n, and let wi’=
= u,—u,—uj'. Hence it follows that u//’4~ uz’ for k <n.

1) Given without proof in Lindenbaum [1], p. 130.

42 III. Denumerable and non-denumerable sets

If n = 31—1, we shall replace everywhere in the above definition the upper in-
dices ’, ’’ and ’’” by ”’, ’” and ’, respectively, and if » = 31, we shall replace them respec-
tively by °”, %, and”.
It is easy to prove that the infinite sequences {u/}, {u/’}, {u/’’}, defined thus by
induction, satisfy the required conditions.
2. Find an infinite sequence {u,} of rational numbers which is not a sum of two
sequences, each of them composed of all different rational numbers.
Solution. Such is for instance the sequence {u,} where u,= 1, and u, = 0 for
SOs Be ep
Indeed, if we had u, = u,+ uj’ for n = 1, 2,..., where each of the sequences {u;,}
and, {u}’} is composed of all different rational numbers, then we should have u/-+ u}’ = 0
for n = 2,3,..., therefore u/’ = — us for n = 2, 3,... But the sequence ui, uj, ..., and
therefore also the sequence —u/, —uf,... are made up of all different rational numbers.
Thus the sequence w,’, u;’,... contains all rational numbers except the number —w{,
whence, in view of the property of the sequence {u/’}, we conclude that uj’/= —u/,
therefore u, = uj+ ui’ = 0, which is impossible since u, = 1.

5. Effective denumerability of the set of all algebraic numbers. Let P

denote a polynomial of degree >0 in one variable, with integer coef-
ficients, a)+a,7+a,v7-+...+4,0" where a, ~0. Denote by g(P) the
sequence of nm+1 terms (do, @, G:,..-,@,). Obviously the function
establishes a (1-1) correspondence between polynomials of degree >0 in
one variable with integer coefficients and finite sequences of integers
having more than one term, the last term +4 0.
Thus from Theorem 1 of III.1 and Theorem 1 of [11.4 it follows
that the set of all polynomials of degree > 0 in one variable with integer coef-
frcients is effectively denumerable. Thus we are able to arrange all such
polynomials in an infinite sequence

(5.1) fi(@), fo(@), fa(), ---

An algebraic number is, a8 we know, a root of a polynomial of degree

> 0 in one variable with integer coefficients. Thus if & is an algebraic
number, then there exists (at least one) polynomial f(#)= a,+a,e+
+...+4,0" of degree > 0 with integer coefficients such that f(&)= 0,
and therefore at least one term f, of sequence (5.1) such that f,(é) = 0.
THEOREM 1. The set of all algebraic numbers is effectively denumerable.
Proof. Let f,(#)(n = 1, 2,...) be the terms of the infinite sequence
of polynomials (5.1) defined above. It is proved in Algebra (in quite
an elementary way) that a polynomial of degree n in one variable cannot
have more than n different roots. Thus every polynomial in one variable
of degree > 0 has a finite number of different roots.
Now let us write down all the roots of the polynomial /,(#) (arrang-
ing them, for instance, in order of the increasing magnitude of their real
parts, and, in case of their real parts being equal, in order of the increasing
 § 5. The set of all algebraic numbers 43

magnitude of their imaginary parts), then (in the same way) all the roots
of the polynomial f,(7), next those of the polynomial f,(x), ete.
In the infinite sequence of algebraic numbers obtained in this way
let us retain only those terms which are different from any of the terms
preceding them. We shall obtain the well-defined infinite sequence

(5.2) pe er ane
composed of all and, also, of different algebraic numbers. That the
sequence is infinite follows from the remark that, in any case, this se-
quence will contain all rational numbers as roots of polynomials of the
first degree with integer coefficients.
Thus we are able to arrange all algebraic numbers in an infinite
sequence, which proves Theorem 1].
EXERCISES. 1. Prove that the set of all finite sequences whose terms are al-
gebraic numbers is effectively denumerable.
Proof. It is a consequence of Theorem 1 of III.5 and Theorem 2 of III.4.
2. Prove that the set of all straight lines in a plane each of which passes through
(at least) two different points with rational coordinates is effectively denumerable.
3. Prove that if & denotes the set of all points of a plane that have rational co-
ordinates, and D any straight line lying in the plane, then for the set D& only the fol-
lowing three cases are possible: 1° The set DR is empty, 2° The set DR consists of only
one point, 3° The set DR is denumerable.
4. Prove that the same applies to the case where #& denotes the set of all points
of a plane that have integer coordinates, and to the case where & denotes the set of
all points of a plane whose coordinates are algebraic numbers, but not to the case of &
denoting the set of all points of a plane that have natural coordinates (because then
the set DR may also have an arbitrary finite number of points).

6. Non-denumerable sets. Infinite sets with which we have dealt so

far have been denumerable.
An infinite set which is not denumerable is called non-denumerable.
We shall prove that such sets exist. For this purpose let us first prove
THEOREM 1. We are able to define a function (81, 8,...) relating
to every infinite sequence 81, 82, ... of infinite sequences of natural numbers
a certain infinite sequence s of natural numbers such that s~ 8, for n=
== ROM pete ‘sp
Proof. Let s,, 8, 83,... denote a given infinite sequence of infinite
sequences of natural numbers. Thus we have, for every natural n

(6.1) 9 MOE Ae elle a.)

where a”, af, af, ... is an infinite sequence (corresponding to the na-
tural number ») of natural numbers.
44 III. Denumerable and non-denumerable sets

Let us denote by 9 (81; S,,..-) the infinite sequence

P (S14 S25 ++) = (ay +1, ay +1, as +1, ver) y

i. e. the sequence (b,, b.,...) where b, = a” +1 for n= 1, 2,...

It is easy to see that the function » satisfies the conditions of our
theorem. Indeed, for every given natural number n, we have s= (8,
Sy, ..-) ~S,, Since the n-th term of the sequence s is a” +1, and the n-th
term of the sequence s, is ae Thus the sequences s and s, differ in their
n-th terms, 7. e. they are different. Thus we have proved Theorem l.
THEOREM 2. The set of all infinite sequences of natural numbers is
non-denumerable.
Proof. Suppose that the set Z of all infinite sequences of natural
numbers is denumerable. Therefore there exists an infinite sequence
$1, 8, ... composed of all the elements of the set Z. Let s = o(8;, 8, ...),
where is a function satisfying Theorem 1. Thus s 4s, for n= 1, 2,...,
therefore s « Z, which is impossible since s, being an infinite sequence of
natural numbers, is an element of the set Z.
Thus the set Z is not denumerable. But Z is an infinite set, since Z
contains every set whose terms are all equal to the same natural number n,
and there are infinitely many such sets. Thus the set Z is non-denumerable,
Ge...
A set maybe called effectively non-denumerable if we are able to re-
late to every infinite sequence of the elements of that set an element of
the set different from any of the elements of the sequence in question.
(Novikoff [1], p. 35-39). Obviously (in view of Theorem 1 of III.6), the set.
of all infinite sequences of natural numbers is effectively non-denumerable..
EXERCISES. 1. Define a function p(s,, s,, ...) relating to every infinite sequence:
S,, S,,... of infinite sequences of natural numbers an infinite sequence of natural num--
bers s such that for every natural number n there exist only a finite number of terms.
of the sequence s, that are equal to the corresponding terms (i. e. those occupying the
same place) of the sequence s.
Solution. If the sequences s, (n= 1, 2,...) are defined by the formula (6.1),
thenwlotwy (Sisesaenes = CreeCoeee) ML OLO a CE a) + a®) + weet a) + sto? Pees 5 coe
Let n denote a given natural number. Hence we shall have ¢, > a® for k>n;
therefore if ¢, = a, then k <n. Thus there are less than n terms of sequence (6.1)
that are equal to the corresponding terms of the sequence ¢,, ¢,, ... Therefore the funec-
tion y defined above satisfies the required conditions.
2. Let q denote a given natural number. Prove that there exists no function
&(s8,, 8, ...) relating to every infinite sequence s,, s,,... of infinite sequences of natural
numbers an infinite sequence of natural numbers s such that for every natural num-
ber n there exist at most g terms of the set s, equal to the corresponding terms
of the set s.
 § 6. Non-denumerable sets 45

Proof. As we proved in III.4 we are able to define an infinite sequence 0;, oy, ...
of all finite sequences of natural numbers. Now denote by s,, for natural n, the infinite
sequence arising from the finite sequence o, if, after the last term of that sequence, we
add the infinite sequence of numbers 1.
Now let us suppose that the function # mentioned in our exercise exists. For the
infinite sequence of the sequences s,, 8, ... just defined, let s = #(s,, 82, ...). For every
natural number nm there exist at most q terms of the sequence s,, equal to the correspond-
ing terms of the sequence s.
Let s = (k,, k,, ...). The finite sequence k,, k,, ...,-k q+ is one of the terms of the
infinite sequence o;, o2,...; thus there exists a natural number n such that (k,, ke,..., k, 41) =
= o,. According to the definition of the set s,, we shall then have s, = (k,, ky, ..., ky.4;
1,1,1,...). Thus each of the first g+1 terms of the sequence s, is equal to the cor-
responding term of the sequence s, contrary to the above-mentioned property of the
set s,. Thus the assumption of the existence of the function # mentioned in Exercise 2
deads to a contradiction. This completes the proof.

7. Properties of sets containing denumerable subsets. We shall prove

THEOREM 1. If CC A, AB= 0, and if the sets B and C are effectively
denumerable, then A+Bef ~A.
Proof. If the sets B and C are effectively denumerable then we are
able to arrange their elements in infinite sequences, B= {p,, po, ...}-
and O = (G1) qa) +++)
Let us denote by a function defined in the set A in the following
way. For pe A—C let y(p) =p. If pe C, then for one (and only one)
natural » we have p=q,. If n is an odd number, n = 2k—1, then let
9 (p) = px; if n is an even number, n = 2k, then let p(p) = q,. There is
no difficulty in proving that the function » establishes a (1-1) corres-
pondence between the elements of the sets A and A+B.
It is easy to see what modifications should be made in the proof
of Theorem 1 in order to prove
THEoREM 2. If CC A, AB = 0, and tf the sets B and C are denumerable,
then A+BWA.
It is also easy to modify the proof of Theorem 1 so as to prove
THEOREM 3. If OC A, AB = 0, and if the set B is finite and the set CO
denumerable, then A+BW~A.
Proof. The set B being finite, there exists a finite sequence composed
of its elements, e. g. B= {p1, Po, --») Pm}. And the set C being denumerable,
there exists an infinite sequence composed of its elements, e.g. C=
=a {his Yo» wee
Let us define function y in the set A in the following way. If
p« A—O, then let y(p) = p. Ii pe C, p= q,, and n <™m, then let y(p) =
= pn; and if n > m, let p(p) = dr—m- It is easy to verify that the function
establishes a (1-1) correspondence between the elements of the sets A
and A+B.
 46 III. Denumerable and non-denumerable sets

Theorems 2 and 3 imply immediately

TuroREM 4. A set containing a denumerable subset is equivalent to
the set obtained by adding to it a finite or a denumerable set of elements.
Now let us suppose that the sets B and C are effectively denumerable,
ADB and A—B2OC and let us replace in Theorem 1 the set A by the
set A—B. In view of (4—B)+B= A, we shall obtain in this way
TuEorEM 5. If ADB, A—BODO, the sets B and C being effectively
denumerable, then A ef ~A—B.
We can express this Theorem as follows: If from an arbitrary set
we remove an effectively denumerable set of elements and if the set
obtained in this way contains an effectively denumerable subset, then
this set will be effectively equivalent to the original set.
Obviously we also have
THEOREM 6. If ADB, A—BOC, the set C being denumerable and
the set B finite or denumerable, then A—B ~A.
The following two corollaries result immediately from Theorem 4:
COROLLARY 1. The sum of. two denwmerable sets is a denwmerable set.
COROLLARY 2. The sum of a denumerable set and a finite set is a denume-
rable set.
Moreover, we shall prove
COROLLARY 3. If the set A is non-denumerable, and the set B is finite
or denumerable, then the set A—B is non-denumerable.
Proof. Let 4—B = C; we shall have A = C+AB. Since the set B
is finite or denumerable, it follows from Theorem 1 of ITI.1 that the set
AB is finite or denumerable. Thus if the set C were finite or denumerable,
then, according to Corollaries 1 and 2, the set C+AB, 7. e. the set A,
would also be finite or denumerable, contrary to our assumption that
the set A is non-denumerable. Therefore the set C is neither finite nor
denumerable, 7. €. it is non-denumerable, q. e. d.
Thus: a non-denumerable set will not cease to be non-denumerable ‘if
we remove from ut an arbitrary finite or denumerable set of elements.
From Theorem 1a of III.1 it also follows immediately that a non-
denumerable set will not cease to be non-denumerable if we add to it an ar-
bitrary set.
It is easy to prove that the swm of two effectively denwmerable
sets
is an effectively denumerable set.
Indeed, let A and B be two effectively denumerable sets. Thus we
are able to arrange their elements as infinite sequences

A= {P1) Pes Ee
 § 7. Sets containing denumerable subsets 47

and B= {q%, q,...}. Now let us remove from the infinite sequence p,, q,
Dey 2) Ps) Is, --- every term that is equal to one of the terms preceding it.
As can easily be seen, we shall obtain in this way an infinite sequence
composed of all different elements of the set A+B. Thus the set 4+B
is effectively denumerable.
Hence we obtain by induction
THEOREM 7. The sum of a finite number of effectively denumerabie
sets is an effectively denumerable set.
Obviously we also have
THEOREM 8. The swum of a fimte number of denumerable sets is a de-
numerable set.
THEOREM 9. Every set in which we are able to indicate an effectively
denumerable subset is effectively equivalent to a certain proper subset of
itself (which we are able to indicate).
Proof. Let us suppose that in the set A we are able to indicate an
effectively denumerable subset D= {p,, p2,...} of that set. Let B=
== {o5 Day Dey ---} Oud C= {p,, Ps, Ps, .>-}; these will be.effectively. de-
numerable sets, and we shall have A DB, A—B DC, whence,’according
to Theorem 5, A ef ~A—B. But, in view of BCA and B #0, we have
A ~A-—B; therefore the set A—B is a proper subset of the set A. Thus
the set A is effectively equivalent to its proper subset A—B, which we
are able to define (since we know the effectively denumerable set D).
Thus we have proved Theorem 9.
We also have, as can easily be seen,
THEOREM 10. A set containing a denumerable subset is equivalent to
a certain proper subset of itself.
Theorem 10 can be reversed: namely we have
THEOREM 11. A set which is equivalent to a certain proper subset of
itself contains a denumerable subset.
Proof. Let us suppose that a set A is equivalent to a certain proper
subset of itself B. Thus 4 DB, A—B ~0 and there exists a function
establishing a (1-1) correspondence between the elements of the sets A
and B. Since A—B #0, there exists an element p,<«4—B. We shall
have p,= 9(p,)¢«B, therefore p, & p,- Let us determine by induction
the infinite sequence p,, pz,..., setting pp, = p(pr-1) for n= 2,3,... We
easily prove by induction that p, « B for n = 2,3,... We shall show that
all terms of the infinite sequence p,, p.,... are different from one another.
Let us assume that this is not true, and let p denote the first of the terms
which are equal to one of the preceding terms, thus p, = p, where 7 > q;
therefore p, « B, and since p,¢ B, the equality g = 1 is impossible. Thus
48 III. Denumerable and non-denumerable sets

we have r >q> 1, therefore p, = 9(Dr-1), Pq = P(Pq-1), Whence ~(p,-1) =

= y(Pq-1). Since the function p is invertible in the set A, the last equality
gives p,_1—= P91, contrary to our assumption that p, denotes the first
term of the infinite sequence p,, p.,... that is equal to one of the preced-
ing terms. Hence the sequence p,, p,,... is an infinite sequence composed
of the elements of the set A that are all different from one another. Thus
the set A contains a denumerable subset {p,, Po, ...}, q. e. d.
As regards the inverse of Theorem 9, we shall prove that the fol-
lowing propositions P and Q are equivalent:
P. If a set is effectively equivalent to a certain proper subset of itself,
which we can indicate, then we are able to indicate an effectively denumer-
able subset in that set.
Q. In every non-empty set we are able to indicate a certain element of tt
Let us suppose that proposition P is true; let A denote an arbitrary
non-empty set, and N — the set of all natural numbers. Let B= Ax WN.
Let g denote a function defined for qe B in the following way: If q=
= (p,n), where pe A, ne N, then o(q) = (p, +1).
Let € denote the set of all elements y(q) where qe B. The set C is
the required proper subset of the set B and, clearly, the function p defined
above establishes a (1-1) correspondence between the elements of the
sets B and C. Hence the set B is effectively equivalent to a certain prop-
er subset of itself, C, which we are able to indicate. Thus from the truth
of proposition P it follows that in the set B we are able to point out
a denumerable subset D. Let q, denote the first element of that subset.
We shall have q, = (p,, 4), where p, is an element defined by q, of the
set A. Therefore in the non-empty set A we are able to indicate a certain
element p, of that set. Thus we have proved that if proposition P is true,
then proposition Q is also true.
Now let us suppose that proposition Q is true, and let A denote
a set which is equivalent to a certain proper subset of itself, B, which
we are able to point out. Thus we are able to define in the set A a func-
tion g establishing a (1-1) correspondence between the elements of the
sets A and B. Since A—B +0, from the assumption that proposition Q
is true it follows that we are able to indicate an element p, of the set
A—B. Let us determine by induction an infinite sequence of elements
of the set A, p,, po,..., taking p, = 9(Pn_1) for n= 2,3,... As in the proof
of Theorem 11, we show that the terms of this sequence are all different
from one another. Therefore the set C= {p,, po,...} is an effectively
denumerable subset of the set A, which we are able to point out.
Thus we have proved that if proposition Q is true, then proposition P
is also true.
 § 7. Sets containing denumerable subsets 49

The equivalence of propositions P and Q is thus proved.

Now we shall show that at the present state of science proposition Q
is not true.
Let F denote the set of all real functions f(x) of areal variable « which
are not of the form ax (where a is constant) and which satisfy the func-
tional equation f(w+y)= f(#)+ f(y) for all real numbers x, y. Without
specifying whether the set F' is ae or not, let us take A= Fif F4~0,
and denote by A the set consisting of only one function f(v) = wif F = 0.
The set A is of course non-empty, but at the present state of science we
are not able to indicate any element of the set A (which does not exclude
the possibility of our doing so in the future) *).
Hence it follows that proposition Q is not true to-day ”).
8. Sets infinite in the sense of Dedekind. R. Dedekind [2] calls infinite
every set which is equivalent to a certain proper subset of itself. From
Theorem 10 of JII.7 and Theorem 11 we immediately obtain
THEOREM 1. A set is infinite in the sense of Dedekind if and only if
at contains a denumerable subset.
It is easy to see that a set equivalent to a set which is infinite in
the sense of Dedekind is also infinite in that sense.
Clearly, a set which is infinite in the sense of Dedekind cannot be
finite (since no finite set can contain an infinite sequence of different
elements). However, we are unable to prove without the use of the axiom
of choice (see IV.7) that every set infinite in the sense defined in IT.2
is infinite in the sense of Dedekind; neither can we prove that every non-
denumerable set is infinite in the sense of Dedekind.
Since without the aid of the axiom of choice we are unable to prove
that there exist no sets which are not finite (in the usual meaning) and
not infinite in the sense of Dedekind, the existence of such sets has been
conjectured *) and their properties investigated (Wrinch [1], p. 87-92).
EXERCISES. 1. Prove that the set A is infinite in the sense of Dedekind if and
only if itis equivalent to the set which will be obtained by joining to the set A an element
not belonging to A.
Proof. If the set A is infinite in the sense of Dedekind, then, according to Theo-
rem 1, it contains a denumerable subset; therefore, according to Theorem 4, the set A

1) If anyone wished to assert that the function f(x) = x is an element of the set A,
he would have to prove first that the set # is empty, which we are not able to do (and
from the so-called axiom of- choice it follows that IF + 0).
2) The truth of the proposition stating that we are able to indicate an element
of the set A depends on the time-and on the person who makes it. Logic knows
various propositions of this kind, e.g. the propositions: “I am 75 years old“, “I am
in Paris“, “It is Friday to-day“.
3) According to A. A. Fraenkel (L’axiome du choix, Revue Philosophique de Teena sin
50 (1952), p. 444), this conjecture was formulated as early as 1904 by H. Lebesgue.
Cardinal and ordinal numbers 4
50 III. Denumerable and non-denumerable sets

is equivalent to the set obtained by joining one element to the set A. Thus the condition
stated in the exercise is necessary.
On the other hand, let us suppose that the set A is equivalent to the set B = A+ {p}
where p ¢ A. Thus the set B is equivalent to its proper subset A, and therefore, accord-
ing to the definition of sets infinite in the sense of Dedekind, it is infinite in the sense
of Dedekind. Such is also the set A, being equivalent to the set B. Thus the condition
stated in the exercise is sufficient.
2. Prove that the set B is infinite in the sense of Dedekind if and only if it is
equivalent to the set obtained by removing from the set B one of its elements.
Proof. If pe B, let A = B— {p}, whence B= A+ {p}.
If the set B is infinite in the sense of Dedekind, then, according to Theorem 1
of III.1 B contains a denumerable subset {p,, p., ...}. The set {p,, pz, -..}—{p} will of
course be a denumerable subset of the set A; according to Theorem 1, the latter is in-
finite in the sense of Dedekind and, according to Exercise 1, we have A+ {p}~4dA,
Therefore B~ B— {p}, which proves that the condition stated in Exercise 2 is necessary.
On the other hand, if for a certain p«B we have B—{p}~B, then the set B is
equivalent to its proper subset B— {p}, therefore it is infinite in the sense of Dedekind.
Thus the condition stated in Exercise 2 is sufficient.
3. Prove that a set infinite in the sense of Dedekind is equivalent to the set ob-
tained by adding to it or removing from it an arbitrary finite number of elements.
Proof. The proof is easily obtained from Exercises 1 and 2, by induction.

9. Various definitions of finite sets. Every set which is not infinite

in the sense of Dedekind, 7. e. every set which is not equivalent to a prop-
er subset of itself, may be called finite in the sense of Dedekind. Such
a definition of finite sets would have the advantage of not being based
on the concept of natural numbers. But then we should not be able to
prove without the aid of the axiom of choice that this definition of finite
sets is equivalent to that given in II.2. Poincaré ridicules those who would
like to define the concept of a finite set in this way 4).
Besides the two definitions of finite sets discussed so far (the usual
one and that of Dedekind), other definitions are known. E. Zermelo,
for instance, defines a finite set using the concept of ordered set and
calls finite every set which can be ordered in such a way that each of
its subsets has a first and a least element (Zermelo [4], p. 188, and Stickel
[1], p. 425). Several authors give other different definitions of finite sets.

1) ,,De nombreux mathématiciens se sont lancés sur ses traces et se sont posé
une série de questions de méme genre. IIs se sont tellement familiarisés avec les nombres
transfinis qu’ils en sont arrivés 4 faire dépendre la théorie des nombres finis de celle
des nombres cardinaux de Cantor. A leurs yeux, pour enseigner l’arithmétique d’une
fagon vraiment logique, on devrait commencer par établir les propriétés générales des
nombres cardinaux transfinis, puis distinguer parmi eux une toute petite classe, celle
des nombres entiers ordinaires. Grace & ce détour on pourrait arriver 4 démontrer toutes
les propositions relatives 4 cette petite classe (c’est-4-dire toute notre arithmétique et
notre algébre) sans se servir d’aucun principe étranger 4 la logique“ (Poincaré [1]).
 § 9. Various definitions of finite sets pet

Russel calls a set inductive if it belongs to every family F of sets that

contains the empty set and satisfies the following condition: if X «
and if Y is the set obtained from the set X by joining to it one element,
then Y « F (Russell [2], p. 30). Weber [1], Sierpiiski [5] (p. 106), Kura-
towski [1] (p. 130), and Zermelo [3] have also considered definitions of
finite sets. An extensive paper on finite sets was published in 1924 by
A. Tarski [2]. Its object was the development of the theory of finite sets
as a part of the general theory of sets, without the aid of concepts or
theorems of the arithmetic of natural numbers. On page 49 of that paper
we find the following definition of finite sets:
The set Z is finite if in every non-empty set 7 of its subsets there
exists (at least one) subset Ye 7 such that no proper subset of Y be-
longs to T.

THEOREM 1. Tarski’s definition of finite sets is equivalent to the usual

definition (given in II.2).
Proof. Since the empty set is of course finite according to both
definitions, it suffices to consider only non-empty sets in order to prove
their equivalence.
Let Z denote a non-empty set, finite by the usual definition, 7 — an
arbitrary non-empty set of subsets of the set Z. Every subset of a finite
set (in the usual sense) is a finite set and thus the number of its elements
is a non-negative integer, and, as we know from Arithmetic, in every
set of non-negative integers there exists a least number, therefore in the
set of all the numbers of elements of the sets belonging to 7 there exists
a least number k. It follows from the definition of the number k that
there exists a set K « T with k elements, and that no set belonging to 7
can have less than k elements and therefore no set belonging to T is
a proper subset of AK. Thus the set Z satisfies Tarski’s condition and there-
fore it is finite in his sense.
Now let Z denote a set which is not finite by the usual definition.
Let 7 denote the set of all those subsets of the set Z which are not finite
(in the usual sense). The set 7 is not empty since it contains of course
the set Z. If A « 7, then, according to the definition of the set 7’, the set A
is not finite (in the usual sense), therefore it is not empty and there exists
in element pe A. The set B= A-— {p} is not finite (in the usual sense)
since if it had k elements, the set A would have k+1 elements, thus
being finite, which is impossible. Since, in view of AC Z, we have also
BC Z, it follows that Be T, and B is a proper subset of A belonging to Tf.
Thus every set belonging to 7 contains a proper subset belonging to 7.
Hence it follows that the set Z does not satisfy Tarski’s condition, and
thus it is not finite in his sense.
4*
52 III. Denumerable and non-denumerable sets

Thus the equivalence of the two definition of finite sets, 7. e. The-

orem 1, has been proved. It is left to the reader to prove that also the
following definition of finite sets given by Tarski is equivalent to the
usual definition:
The set Z is finite if in every non-empty set 7 of its subsets there
exists a subset which is not a proper subset of any element of 7.
Let us observe, moreover, that we can prove the equivalence of the
usual definition of finite sets and the following one, called the second
definition of Dedekind:
The set Z is finite if there exists a single-valued function, transform-
ing the set Z into a subset of itself, which does not transform any non-
-empty proper subset of the set Z into a subset of itself (Tarski [2], p. 92,
Theorem 52, Cavaillés [1], p. 142).
We can also prove that the set Z is finite or denumerable if and
only if there exists a (1-1) mapping of the set Z on itself which does
not transform into itself any non-empty proper subset of the set Z
(Tarski [2], p. 85, footnote 1).
Note, moreover, that A. Mostowski has published an extensive
paper (54 pages) On the independence of the definition of finiteness in the
system of logic (Mostowski [1]) *).
An algebraic characteristic of finite sets, rather complicated by the
way, has been given by Kiyoshi Iséki [2].!
10. Denumerability of the Cartesian product of two denumerable sets.
We shall prove
THEOREM 1. The Cartesian product of two denumerable sets is a denu-
merable set.
Proof. In view of II.5, Theorem 4, it is sufficient for the proof of
our theorem to suppose that one factor of our Cartesian product is the
set A of all terms of the infinite sequence 2”-* where n= 1, 2,..., and
the other factor is the set B of all odd natural numbers. Thus if p « Ax B,
then p is an ordered pair (2""*, 2k —1) where n and k are natural numbers.
Let p(p) = 2” *(2k—1). It is easy to verify that the function »
establishes a (1-1) correspondence between the elements of the set 4xB
and the natural numbers. Hence it follows that the set AxB is equi-
valent to the set of all natural numbers, therefore it is denumerable, q. e. d.

1) Cf. a note of Mostowski [2], p. 13-20, containing a discussion of the relations.

of implication between some definitions of finite sets. ;
 CHAPTER IV

SETS OF THE POWER OF THE CONTINUUM

1. Sets of the power of the continuum and sets effectively of the power of
the continuum. A set which is equivalent to the set of all real numbers
is said to be of the power of the continuum and the set which is effectively
equivalent to the set of all real numbers is said to be effectively of the
power of the continuum.
According to Brouwer the set of all real numbers is not a finished
creation: its elements cannot be said to exist but only to be coming into
existence in the process of our defining them. Thus the set of all real
numbers is presented as a set in which we are able to indicate every time
only a denumerable subset, but for every subset of this kind it is always
possible to define the real numbers which do not belong to it (Fraenkel
[3], p. 47; see Theorem 1 of IV.2).
Lusin also questions the concept of the set of all real numbers since
we do not know the general law of defining every real number (Lusin [2],
p. 242). Kronecker and Brouwer go still further, rejecting the concept
of the set of all natural numbers (Bernays [2]).
A. Fraenkel, however, is of the opinion that a theory of sets which
did not ensure the existence of sets of the power of the continuum (so
essential for both Analysis and Geometry) would be a mere shadow and
would mean a suicidal renunciation of the teachings of Cantor for the
sake of intuitionism (Fraenkel [3]).

2. Non-denumerability of the set of real numbers. We shall prove

THEOREM 1. We can define a function (a1, %,...) which relates to
every infinite sequence 1, %2,... of real numbers a certain real number
P= Ol, zn) SUCH tha aa, for Vial, 25.
Proof. Let 7, 2, ... denote a given infinite sequence of real numbers.
As we know, every real number has one, and only one, normal
decimal representation !), which we can always regard as infinite, re-

1) A finite decimal expansion or an infinite one in which infinitely many digits

are different from 9 is called normal. ;
54 IV. Sets of the power of the continuum

placing the missing digits by zeros. Let

(2.1) ; == Hn, 0, teCee

denote the normal decimal expansion of the number 7.

Now, for natural n, -let

2.2, Cn ==
Out Oe a0.
er Date Seti:
The infinite sequence ¢,, ¢,,... will be well-defined by these condi-
tions and the numbers ¢,(n = 1,2,...) will be digits of the decimal system
other than the digit 9. Let
(2.3) (Vig Was noo) =, Oy OC
oGyan

Formula (2.3) gives the normal decimal expansion of the real num-
DOL LL == eae ae)
Now let » denote a given natural number. Formulae (2.2) and (2.3)
prove that the n-th digit of the expansion (2.3) is different from the n-th
digit of the expansion (2.1). Since every real number has only one normal
decimal expansion, it follows that «4A x,. Thus we have «~ 42, for n=
= 1,2,... Therefore the function y(a#,, 7, ...) satisfies Theorem 1, which
has thus been proved.
As the set of all real numbers is infinite (since all natural numbers
belong to it), therefore it follows from Theorem 1 (in the same way as
Theorem 2 of III.6 follows from Theorem 1 of III.6) that the set of all
real numbers is non-denumerable.
In view of Theorem 1 we likewise conclude that the set of all real
numbers may be called effectively non-denumerable *).
From Theorem 1 it follows that we can define a function /(#) which
relates to every effectively denumerable set H of real numbers a real
number f(#) such that f(#)¢ H. Let us observe that we cannot define
a function g(D) relating to every denumerable set D of real numbers
a real number g(D) such that g(D)¢ D. Neither can we define a func-
tion h(D) relating to every denumerable set D of real numbers a real
number h(D) such that h(D) « D.

3. Removing a denumerable set from a set of the power of the continuum.

We shall prove
THEOREM 1. If X is the set of all real numbers, B— an effectively
denumerable set such that BC X, then we can indicate an effectively denu-
merable set C such that X—BO OC.

1) Cf. the notion of effective non-denumerability on p. 44.

 § 3. Removing a denumerable set 55

Proof. If B is an effectively denumerable set of real numbers, then

we are able to arrange all the elements of the set B in an infinite se-
quence: B= {b,, b,,...}. Let y denote the function satisfying Theorem 1
of IV.2 and define by induction an infinite sequence of real numbers
€,) Cz). by the conditions:

€, = p(b,, Be, .:.) , Cn4t-== Y(C1, Cay +++). ny O14 Og, «--)

ATOR Ci 1 Sei.)
It follows immediately from these formulae and from the properties
of the function y that the terms of the infinite sequence ¢,, ¢,... are all
different from one another, and that the set O = {¢,, G&,...} is disjoint
with respect to the set B. Therefore we have X—BOC and the set C is
an effectively denumerable set, which we are able to define. Thus we
have proved Theorem 1. (
From Theorem 1 and Theorem 5 of III.7 we immediately have
THEOREM 2. If we remove from the set of all real numbers an effectively
denumerable subset, we shall obtain a set effectively of the power of the con-
tinuum.
From this theorem, in view of Theorem 1 of III.2 and Theorem 1
of III.5 we immediately infer the following two corollaries:
COROLLARY 1. The set of all. irrational numbers is effectively of the
power of the continuum.
COROLLARY 2. The set of all real transcendental +) numbers is effectively
of the power of the continuum.
No example of a transcendental number had been known up to the
middle of the past century. However, from IV.2, Theorem 1, and from
IIL.5, Theorem 1, it immediately follows that we are able to define
a transcendental number.
A transcendental number defined in this way would be very com-
plicated and hardly interesting. The proof of the existence of transcen-
dental numbers, based on the denumerability of the set of all algebraical
numbers, comes from G. Cantor?). In the same year, 1873, Hermite
proved that the number e (the base of natural logarithms) is transcendental,
thus giving a simple example of such numbers. About 10 years later
Lindemann proved that the number z is also transcendental, and, con-
sequently, that the squaring of a circle cannot be performed by means
of compasses and ruler (Lindemann [1]). In 1934 it was proved (by Gel-
fond and Schneider) that the numbers 2V? and e* are transcendental.

1) A number which is not algebraical is termed transcendental.

*) It was published in Cantor [2], one of the first works on the Theory of Sets.
56 IV. Sets of the power of the continuum

An entirely different principle underlies Liouville’s proof of ‘the

existence of transcendental numbers (Liouville [1]), which is earlier than
Cantor’s. Liouville proves that every algebraical number of the n-th
degree has a certain property P,, and then constructs numbers which
do not have the property P, for any natural n, 7. e. transcendental num-

bers. One of them, for instance, is the number 10-”,

n=1

It is easy to see what modifications should be introduced into the

proof of Theorem 2 in order to obtain
THEOREM 3. If we remove from a set of the power of the continuum a fi-
nite or denumerable subset, we shall obtain a set of the power of the continuum.
From III.7, Theorem 4, and from the observation that the set of
all real numbers contains the denumerable set of all natural numbers,
1. e. that every set of the power of the continuum contains a denumerable
subset, immediately follows
THEOREM 4. The set obtained by adding to a set of the power of the
continuum a finite or denumerable set 1s of the power of the continuum.
We shall also prove
THEOREM 5. If we add one element to a set which is effectively of the
power of the continuum, we shall obtain a set effectively of the power of the
continuum.
Proof. If the set A is effectively of the power of the continuum,
then we are able to define a function y establishing a (1-1) correspondence
between all real numbers and all elements of the set A. Now let b denote
an element which does not belong to the set A. For real a, let p(x) = (ax)
if 7 is not a natural number, y(1) = db, and y(n+1) = y(n) for natural n.
It is easy to verify that the function w establishes a (1-1) correspondence
between all real numbers and all elements of the set A-+ {b}. Thus the
latter is effectively of the power of the continuum, q. e. d.
By induction we obtain from Theorem 5 the following
COROLLARY. If we add a finite number of elements to a set which is
effectively of the power of the continuum, we shall obtain a set effectively
of the power of the continuum.

4. Set of real numbers of an arbitrary interval. It can easily be seen

that the set X of all positive real numbers is effectively equivalent to
the set Y of all numbers lying inside the interval (0,1). In order to
establish a (1-1) correspondence between the elements of the sets X
and Y it suffices to associate with every positive real number « the number
y = «:(1+.), belonging of course to the set Y. The proof that the. cor-
respondence is (1-1) presents no difficulties.
 § 4. Set of real numbers ‘ave

Similarly, the set of all negative real numbers is effectively equi-

valent to the set of all numbers lying inside the interval (—1, 0). Hence
it immediately follows that the set of all real numbers is effectively equi-
valent to the set of all numbers lying inside the interval (—1s3 1). Thus
the latter set is effectively of the power of the continuum.
Further, it is easy to establish a (1-1) correspondence between all
numbers lying inside the interval (—1,1) and all numbers lying inside
any finite interval (a,b) where a and b are given real numbers, a <b.
For that purpose it suffices to assign to every real number w such that
—1<a#<1 the number

y=,ot
lying inside the interval (a, b); the proof that the correspondence is (1-1)
presents no difficulties.
Thus we have
THEOREM 1. The set of all real numbers lying inside the finite interval
(a,b) (where a and b are given real numbers, a <b), is effectively of the
power of the continuum.
By IV.3, Theorem 5, and the Corollary to it, Theorem 1 implies
THEOREM la. If a and b are real numbers, a <b, then both the set
of all real numbers x such that a< « <b and the set of all real numbers «x
such that a <<u“<b ave effectively of the power of the continuum.
EXERCISE. Give an example of the continuum of different increasing functions
mapping in a (1-1) way the set of all real numbers of the interval 0 < «<1 on the
set of all real numbers of the interval 0< y < 2.
Answer. The function
se err =e

where c¢ is an arbitrary constant equal to a real number > 0.

The question arises whether we are able to define a set which can
be proved to be of the power of the continuum but cannot be proved
(at least now) to be effectively of the power of the continuum.
Here is an example of such a set.
Let us divide all real numbers into classes, including two numbers
in the same class if and only if their difference is a rational number.
Let A denote the set of all classes obtained in this way. Now we shail
define the set B as follows. If the set A is of the power of the continuum,
let B= A; otherwise let us denote by B the set of all real numbers.
Clearly, the set B defined in this way is of the power of the con-
tinuum but at the present state of science we are unable to prove that
it is effectively of the power of the continuum.
58 IV. Sets of the power of the continuum

5. Sum of two sets of the power of the continuum. We shall prove

THEOREM 1. The sum of two sets which are both effectively of the power
of the continuum is a set effectively of the power of the continuum.
Proof. First, let A and B be two disjoint sets, each of which is
effectively of the power the continuum. Thus, by Theorem la of the
preceding section, the set A is effectively equivalent to the set of all real
numbers «w such that 0 <#<1 and the set B is effectively equivalent
to the set of all real numbers w such that 1 <a# < 2: Hence it immedia-
tely follows (since AB = 0) that the set 4+B is effectively equivalent
to the set of all real numbers a such that 0 <a< 2. By IV.4, Theorem 2a,
however, the latter is effectively of the power of the continuum. There-
fore the set A+B is also effectively of the power of the continuum.
Now let A and B be any two sets either of which is effectively of
the power of the continuum. Let us denote by A, the set of all pairs (p, 1)
where pe« A, and by B, the set of all pairs (p, 2) where p « B,. We shall
have of course A,ef ~A and B,ef ~B; therefore the sets A, and B,
will be effectively of the power of the continuum, and thus, as we have
proved, the set A,+B, will be effectively of the power of the continuum.
Since Bef ~B,, we are able to define the set Bi,C B, such that B—A
ef ~f;. By II.5, Theorem 1, and in view of A(B—A)=0 and A,Bi,C
CA,B,=0 we find 4+B=A+(B—A)ef ~A,+ Bi. But A,C A,+
+B; C A,+B, and the sets A, and A,+B, are effectively of the power
of the continuum, therefore A, ef ~A,+8B,. In virtue of I1.6, Theorem 4,
we conclude that A,+Bjef ~A, and thus that 4+Bef ~A,, 2. e. that
the set A+B is effectively of the power of the continuum. Thus we
have proved Theorem 1.
From Theorem 1 we immediately infer by induction the following
CoROLLARY. The sum of a finite number of sets each of which is effectively
of the power of the continuum is a set effectively of the power of the con-
tinwum.
Clearly, we also have
THEOREM la. The sum of a finite number of sets each of which is of
the power of the continuum is a set of the power of the continuum.
As has been proved above, the set of all negative real numbers is
effectively of the power of the continuum. We also know that for a na-
tural » the set X,41 of all real numbers w such that n—1 <a < n is ef-
fectively of the power of the continuum. But the set of all real numbers
is of course the disjoint sum of all sets XY, for n= 1,2,... Hence follows
THEOREM 2. We are able to decompose every set effectively of the power
of the continuum into a disjoint sum of an infinite sequence of sets each
of which is effectively of the power of the continuum.
 § 5. Sum of two sets 59

It is also easy to prove that we are able to decompose every set

effectively of the power of the continuum into a sum of two disjoint
sets each of which is effectively of the power of the continuum, and simi-
larly that every set of the power of the continuum is the sum of two
disjoint sets of the power of the continuum.
On the other hand, it is not easy (although it is possible) to prove
(without resorting to the axiom of choice) that if a set of the power
of the continuum is decomposed into two disjoint subsets equivalent
to each other, then each of them will be of the power of the continuum,
and if the set of real numbers is decomposed into disjoint pairs, then
the set of those pairs will be of the power of the continuum (see Sier-
pinski [8] and [33)).

6. Cartesian product of a denumerable set and a set of the power of the

continuum. We shall prove
THEOREM 1. The Cartesian product of two sets of which the first is
denumerable and the second of the power of the continuum is of the power
of the continuum.
Proof. In view of I1.5, Theorem 4, it*suffices for our theorem to
suppose that one of the factors of the Cartesian product in question is
the set A of all integers and the other, B, the set of all real numbers «x
such that 0 <a#< 1.
Thus if pe AB, then p is an ordered pair (k, x) where k is an in-
teger and x a real number >0 and <1. Let y(p)=k+za. It is easy to
verify that the function @ establishes a (1-1) correspondence between
the elements of the set A xB and all real numbers. Hence it follows that
the set AxB is of the power of. the continuum, q. e. d.

7. Set of all infinite sequences of natural numbers. We shall prove

THEOREM 1. The set. of all infinite sequences of natural numbers is
effectively of the power of the continuum.
Proof. Let S denote the set of all infinite sequences of natural num-
bers.. Phis it ese8... then s = {7,5°7,,-..)) where 97, %5 2.-18 an infinite
sequence of natural numbers. Let

Gls) See? “INO a Me ks

this sum gives, as we know, a certain real number > 0 and <1.
Basing ourselves on the theorem known from Analysis stating that
every real number x such that 0 <a# <1 can be represented uniquely
in the form
HE es irc Tat yeeis Ae
60 IV. Sets of the power of the continuum

where ,,%,... is an infinite sequence of natural numbers (7. e. in the

form of a strictly infinite dyadic fraction), we can easily prove that
the function g establishes a (1-1) correspondence between the elements:
of the set S and the real numbers w such that 0<a<1. But the
latter set, by IV.4, Theorem 1, and IV.4, Theorem 2, is effectively of the
power of the continuum. Therefore the set S is effectively of the power
of the continuum, q. e. d.
The set of all infinite sequences of natural numbers is identical with
the set of all functions of a variable whose values are natural numbers,
i. e. with the set N, where N% is the set of all natural numbers. Therefore
from Theorem 1 we immediately draw the following
COROLLARY 1. If N denotes the set of all natural numbers, then the
set NN is effectively of the power of the continuum.
EXERCISES. 1. Prove that the set of all increasing infinite sequences of natural
numbers is effectively of the power of the continuum.
Proof. In view of Theorem 1 it suffices to show that the set of all increasing in-
finite sequences of natural numbers is effectively equivalent to the set of all infinite
sequences of natural numbers. Now, a (1-1) correspondence between the elements of
those two sets is established by the function » relating to every increasing infinite se-
quence s,, S,... the sequence of natural numbers s,,8,—8,, 83— 82, $4— 83, «.
2. Prove that the set of all infinite sequences of rational numbers is effectively
of the power of the continuum.
Hint. Use Theorem 1 and Theorem 1 of III.2.
3. Prove that the set of all increasing infinite sequences of rational numbers is:
effectively of the power of the continuum.
Proof. We obtain the proof by referring to Exercise 2 and then proceeding as.
in Exercise 1.
4. Prove that the set of all infinite sequences of algebraic numbers is effectively
of the power of the continuum.
Hint. Use Theorem 1 and Theorem 1 of III.5.
5. Prove that the set of all sets of natural numbers is effectively of the power of
the continuum.
Proof. According to Theorem 1 of III.4 the set of all finite sets of natural num-
bers is effectively denumerable. Since we are able to arrange every infinite set of
natural numbers in an increasing infinite sequence, we conclude, in virtue of Exercise 1,
that the set of all infinite sets of natural numbers is effectively of the power of the conti-
nuum. Thus the set of all sets of natural numbers is the sum of two disjoint sets of
which one is effectively of the power of the continuum, and therefore contains an
effectively denumerable subset, and the other effectively denumerable. Thus in virtue
of Theorem | of III.7, we conclude that the sum mentioned above is a set effectively
of the power of the continuum.
Thus we have also proved that the set of all subsets of the set of all natural numbers
is effectively of the power of the continwum.
6. Give an example of a set H of the power of the continuum consisting of points
in the three-dimensional. space such that no three points of the set H lie on the same
 § 7. Set of all infinite sequences 61

straight line, and if any two points of the set H are connected by line segments, then
no two segments have a common point which is not an end point.
Solution. A set of that kind is the set # of all points in the three-dimensional
space with the coordinates (¢, ¢, t?) where ¢t is an arbitrary real number. No three points
of the set # lie on the same straight line because if they did, there would exist four
different points of the set # lying in the same plane, which is impossible since the equa-
tion Ai?+ B+ Ct+D = 0, provided we do not have simultaneously A = B= O= D= 0,
has no more than th ee different roots. And if two segments whose end points are
points of the set H had a common point other than their common end point, then they
would lie in the same plane, containing four different points of the set #, which is im-
possible. Thus the set H has the required properties.
7. Prove that the set of all infinite sets of natural numbers is effectively of
the power of the continuum.

8. Cartesian product of two sets of the power of the continuum. We

shall prove
Theorem 1. The Cartesian product of two sets which are both effectively
-of the power of the continuum is effectively of the power of the continuum,
Proof. By Theorem 2 of III.2 and Theorem 1 of III.7 it suffices,
for the proof of our theorem to suppose that each factor of our Cartesian
‘product is the set S of all infinite sequences of natural numbers. Thus
if pe SxS, then p is an ordered pair, p = (s,¢), where s and ¢ are infinite
sequences of natural numbers, s = (m,, ms, ...), t= (1, Mo, ...). Let |

P(P) = (My, Nyy My, Nay M34 Ngy ++) 5

4. é., let us denote by y(p) the sequence obtained by inserting between

‘the successive terms of the set s the successive terms of the set t. We
shall have of course y(p)«S for pe SXS.
It is easy to verify that the function » establishes a (1-1) corres-
‘pondence between the elements of the sets SxS and S. Indeed, if
pes OS, pe SX Sp £p 5 thenwwe: have: p =.(s51),-p' =: (s:, 4) where
either s ~ s’ or tt’ (or both), whence, in view of the definition of the
function g, it follows that in any case y(p)~ y(p’). Thus the function ~
is invertible in the set Sx S. On the other hand, if re S and therefore
Vie Wan Nayerels OMENS CAKING 28s 7(107 .<ay 5 srep )y Ui Mos Bags Nes a-s)s
p = (s,t), we shall have pe SxS and, in virtue of the definition of the
function 9, 9(p)=q. Therefore the function m establishes a (1-1) cor-
respondence between the elements of the sets S xS and S. Thus we have
proved Theorem 1.
It is easy to find what modifications should be introduced into the
-proof of Theorem 1 in order to prove
THEOREM la. The Cartesian product of two sets which are. both of
.the power of the continuum is of the power of the continuum.
62 IV. Sets of the power of the continuum

Analysing the proof of Theorem 1 we can define a function ® such

that if g, and g, are the mappings establishing the equivalences S,~X,
S,~X (where X is the set of all real numbers), then ®(¢y,, y,) establishes
the equivalence (S,xS,)~X.
The theorem stating that the function g (defined in the proof of
Theorem 1) has the above-mentioned property includes Theorems 1
and la as particular cases.
In the same way other pairs of similar theorems stating equivalence
or effective equivalence can be replaced by one theorem.
Now let X denote the set of all real numbers. From Theorem 1 it
immediately follows that the set XxX is effectively of the power of
the continuum. But the set X x X is the set of all pairs (wv, y) of two real
numbers. To every such pair we can relate a point of a plane with the
abscissa « and the ordinatey, and, as we know, this correspondence
s (1-1). Thus the set of all points on a plane is effectively of the power
of the continuum, 7. e. it. is effectively equivalent to the set of all
points on a straight line (cf. I1.1, Example 2). Thus we have
THEOREM 2. The set of all points on a plane is TY equivalent
to the set of all points on a straight line.
We could also say that the set of all complex numbers is effectively
equivalent to the set of all real numbers.
EXERCISE. Let P denote a set of the power of the continuum. The set P, c P
is called a smail subset of the set P if there exists a decomposition of the set P into disjoint
subsets each of which is of the power of the continuum, the set P, having at most one
common element with each oe those subsets. Prove that the set Pis the sum of two of
its small subsets.
Proof. It is easy to prove that in every (1-1) mapping of a set of the power of the
continuum each of its small subsets will be mapped on a small subset of its image.
Therefore it suffices to prove our theorem for any set of the power of the continuum,
e. g. for the set P of all points on a plane (which, according to Theorem 2, is of the power
of the continuum). Let P, denote the set of all points of the axis of abscissae. Let us
decompose the set P into a sum of straight lines parallel to the axis of ordinates. Each
of those lines is a subset of the power of the continuum of the set P and contains only
one point of the set P,, the components of our sum being disjoint. Thus the set P, is.
a small part of the set P. But the set P, is of the power of the continuum, and so is the
set P —P, (since the latter contains a set of the power of the continuum, e. g. the straight
line y = | is contained in the set P of the power of the poe Thus there exists.
a (1- a mapping f of the set P, in the set P—P,. Let g(x) = f(x) for we Py, and g(x) =
= f(x) for « «eP—P,; obviously the function g gives a Bra of the set.P on itself,
and he1) = P,—P,. Since the set P, is a small subset of the set P, it follows that the
set P—P, is also a small subset of the set P. Since P = P,+ (P—P,), P is the sum of
two of its small subsets, q.e. d.

It follows easily from Theorem 2 that we are able to decompose

effectively the set of all real numbers into a sum of two small subsets of it.
 § 8. Cartesian product 63

According to Theorem 2 we are able to define a function f(a, 7)

establishing a (1-1) correspondence between points (#, y) on a plane and
real numbers. Given a real number w we denote by A(x) the set of all
real numbers f(x, y) where y is an arbitrary real ce the set A(x)
is of the power of the continuum since f(x, y)¢ f(x, y’) for y#y’.
Denoting by X the set of all real numbers we Oe obviously have

3 X= SA),
xeX

‘while A(xv):A(ax’)=0 for we A, w € A, x Ax’, the function f being in-

vertible in the plane. Hence
COROLLARY 1. We are able to decompose the set of all real numbers
into the sum of a continuum of disjoint sets each of which is equivalent to
the continuum.
Hence we also immediately obtain
COROLLARY 2. Every set equivalent to the continuum is the sum of
a continuum of disjoint sets each of which is of the power of the continuum.
If we replace in the proof of Theorem 2 the set X of all real numbers
by the set of all points in the interval 0 <a <b, then, in view of Theo-
rem 2a of IV.4, we obtain
THEOREM 2a. The set of all points of a square is effectively ean
to the set of all points of tts side.
By proving Theorem 2 we have proved that if X denotes the set
of all real numbers, then X x X ef ~X. Hence, by Theorem 3 of I1.5, we
immediately obtain the formula:

OLDER Oe
But the set (X x X)xZX is, as we know, effectively equivalent to the set
of all triplets (v, y,z) where wv, y,z2 are real numbers, and also to the
set of all points in the three-dimensional space. Thus we have
THEOREM 2b. The set of all points in the three-dimensional space is
effectively equivalent to the set of all points on a straight line.
Now, in the (1-1) correspondence between the points on a plane
and the points on a straight line, which we are able to establish accord-
ing to Theorem 2, let f,(7,y) denote the point on the straight line cor-
responding to ee point (#, y) on the plane, and let (p(t), p(t )) denote
the point on the plane to which corresponds the point t on the line. We
have of course for any real w and y

(8.1) P(fo(@, y)) ae) v(fo(@, ¥)) We

64 IV. Sets of the power of the continuum

Further, let f(2,y) denote an arbitrary real function of two real

variables. For real ¢ let

(8.2) F(t) = f(p(t), p(t) .

This will be a real function (defined by the function f) of one real
variable. According to (8.2) and (8.1), we find for any real # and any y

F(folv, 9)) = fle(tolw, ¥)), vlfo(@, y))) = f(a, y)-

Therefore
For every real function f(a, y) of two real variables there exists a real-
function of one variable F(t) such that for any real ~ and y we have

(8.3) HosY) Et Vo)

We can express this result by saying that we are able to define a real
function of two real variables f(x, y) such that every real function f(x, y) of
two real variables is a function of one variable of the function f,, and more-
over, we are able to define that function of one variable if we know the fune-
tion f(x, y) (Bieberbach [1], p. 92, Lindenbaum [4], p. 26, Sierpinski [21],
Poli):
The function f/, appearing in this theorem can be defined in the form
g(x)+h(y) where g and h are functions of one variable. Considering only
the intervals 0 <<a <1, 0<y<1 we could assume
co

q(«) = >) 2-*"(H2"e— 282"2) ,

Z—/

h(y) = 39(y)
(Lindenbaum [4], p. 27, Sierpinski [21], p. 133, footnote).
We shall further observe that by using the so-called Continuum
Hypothesis (see V.2) we can prove that there exists a real function of
two real variables f,(~2, y) such that for every real function f(#, y) of two
real variables there exists a real function of one real variable #(¢) in-
vertible and such that for any real « and y we have f(a, y) = f,(9 (x), 3(y))
(Sierpinski [25], p. 8, Sierpinski [24], p. 9).

EXERCISES. 1. Prove that if the function f,(~, y) maps a plane on a straight

line, then the function f,(x, fo/y, 2)) maps the three-dimensional space on a straight
line, and each of the functions f)‘f.(v,
y), fo(z,t)) and fol, fol ys f(z, t))) maps the
four-dimensional space on a straight line.

2. Prove that if the function p(t) = (p(t), p(t)) is a (1-1) mapping of a straight
line on the whole plane, then the function q(é) = (p(t), (v(t), v(y@))) is a (1-1)
mapping of a straight line on the whole three-dimensional space.
 § 8. Cartesian product 65

Proof. Since the function p(t) is invertible in the set of all real numbers, we can
easily prove that the function q(t) is also invertible in the set of all real numbers. Thus
it remains to show that for every point (x, y, z) of the three-dimensional space there
exists a real number ¢ such that q(t) = (x, y, 2).
The properties of the function p(t) imply the existence of a real number uw such
that y = p(u), 2 =y(u) and of a real number ¢ such that « = y(t), wu= p(t). Hence we
have x = g(t), y= o(p(t)), 2 = y(y(t)), therefore g(t) = (x, y, 2), q.e. d.
3. Prove that we are able to define a function of three real variables f,(x, y, 2)
such that every real function f(x, y, 2) of three real variables is a function of one varia-
ble of the function f,, and that we are able to define that function of one variable if
we know the function f(x, y, 2).
Proof. In the (1-1) correspondence between the points of three-dimensional
space and the points on a straight line, which we are able to establish by Theorem 2b,
let f,(~, y, 2) denote the point of the line corresponding to the point (x,y,z) of the
space, and let p(t), w(t), (t) denote the point of the space to an Ae ae the
point ¢ of the straight line. It is easy to prove that, taking F(t) = f(y( t), ®(t)) for
real t, we shall have f(z, y,2) =F (f(x, y,2)) for any real a, y, 2.
4. Prove that we are able to define a real function f,(x, y) of two real variables
such that for every function f(x, y, 2) of three real variables we are able to define
a function g(x, y) of two real variables for which we have, with any real z, y, 2,
{8.4) f(a, Ys 2) = G(X fl¥x2))

Proof. Let f), y and y denote the same functions as in the proof of formula (8.3).
Let g(x,y) = f(x, p(y), p(y)). In view of (8.1) it is easy to verify that we shall have
identity (8.4).
This proves that functions of three real variables can be expressed as a composite
function of functions of two variables. Still, as has been proved by Bieberbach ({1],
p. 90), it is not every continuous function of three variables that can be expressed as
a composite function of a finite number of continuous functions of two real variables.
5. Prove that if the functions f,, y, y, have the same meaning as in the proof of
formula (8.3), then, for any real function of four real variables f(x, y, 2, t), and h(x, y) =
=f (v(x), p(x), p(y), ply)), we shall have identically

f(z, Y,%, t) = h(fola, Y), folz, t)) ’

and taking k(x, y) = f(a, p(y), p(yly)). v(y(y))). we shall have identically

f(x,y, 2, t) =k(a, foly, folz, t)))-

6. Prove that if the functions f,,g,y have the same meaning as in the proof of
formula (8.3), then for any real function of five real variables f(x, y,z,t,u), and

Uw,y)= 1(2, py), o(vly)), (vy). v(v(vy))))

we shall have identically
Uw, y, 2,t, 1) =1(2, folys fol2s folts u)))) -
Thus we can see that functions of a larger number of variables can also be expressed
as a composite functions of functions of two variables.
7. Prove that the set of all straight lines lying on a plane is effectively of the power
of continuum.
Cardinal and ordinal numbers : 5
66 IV. Sets of the power of the continuum

Proof. The proof follows immediately from Theorem 2 and Exercise 2 of II.6.
8. Prove that the set Z of all two-element subsets of the set X of all real numbers
is effectively of the power of the continuum.
Proof. To each two-element subset of the set X we can assign in a (1-1) manner
an ordered pair composed of two real numbers, its first element being the smaller of
the numbers constituting the elements of the given two-element subset of the set X
and its second element being the larger of those numbers. Thus the set Z is effectively
equivalent to a subset of the set X x X (consisting of those of its elements (x, y) where
a < y), and therefore, by Theorem I, it is effectively equivalent to a subset of the set X.
On the other hand, the set X is effectively equivalent to a certain subset of the set Z,
namely to the subset composed of all the pairs {~, 7+ 1} where wx is a real number.
Hence, by Theorem 1 of III.7, we conclude that the sets Z and X are effectively equi-
valent, q.e. d.
9. Prove that for every natural number n the set of n-element subsets of the set X
of all real numbers is effectively of the power of the continuum.
Proof. The proof is analogous to that applied in Exercise 8.
10. Prove that the set Q of all finite subsets of the set X of all real numbers ig
effectively of the power of the continuum.
Proof. For n= 1, 2,3,... let us denote by Q, the set of all n-element subsets
of the set X. We shall have of course 9 = Q,+9Q,+..., and from Exercises 8 and 9 it
easily follows that we can map the set Q on the set of all positive real numbers so that
the image of the set Q,, for n= 1, 2,..., will be the set of all real numbers X such
that n—1 <a<n. Thus the set Q is effectively equivalent to a subset of the set X.
Conversely, the set X is of course effectively equivalent to the set Q,, 7. e. to a subset
of the set Y. Hence, in virtue of Theorem 1 of II.6, we conclude that the sets Q and X
are effectively equivalent, q. e. d.

9. Impossibility of a continuous (1-1) mapping of a plane on a straight

line. The Theorem of Cantor, which we have proved in section 8, stating
that a plane can be mapped on a straight line (or a square on a segment)
apparently undermines our notion of dimensions. This is, however, only
a surface impression. What the theorem actually does is to overthrow
our false assumption, based merely on intuition, that sets of points in
spaces of different dimensions are not equivalent. The fundamental
difference between spaces of different dimensions should be sought
elsewhere; it results from the fact that spaces of different dimensions can-
not be mapped continuously in a (1-1) manner one on another. The proof
of this theorem in its full extent requires a certain knowledge of the theory
of point sets. Here we shall only prove analytically that a two-dimen-
sional space cannot be mapped continuously in a (1-1) manner on a one
dimensional space. For this purpose we shall prove
THEOREM 1. There exists no continuous function f(x,y) of two real
variables x and y (even continuous only with respect to each variable se-
parately) which for different pairs of real numbers <x,y> would always
assume different real values.
 § 9. Impossibility of a continuous (1-1) mapping 67

Proof. Let us assume that such a function f(x, y) exists. Let

(9.1) p(%) = f(#, 9),

1. €. y(v) will be a continuous function of the variable ~ Let us write:
y(0)= 0, p(1) = b. From the assumption that to different pairs (x,y)
correspond different values of the function f(«v,y), it follows by (9.1)
that a~b, e.g. a<b. The function y(#), being continuous in the inter-
val (0, 1), assumes in this interval every value contained between y(0) = a
and y(1) = b; therefore in the interval (0,1) there exists a value %, (at
least one) such that p(a#) = 4(a+5).
Further, let y(y) = f(x,y), 2. e. w(y) will be a continuous function
of the variable y, and we shall have

(0) = flat, 0) = gla) = 2°,

therefore, since a <b, a<y(0) <b, and since the function p(y) is conti-
nuous for y= 0, we conclude that for sufficiently small y we must
also have a<ywp(y) <b, t.¢. a<f(%,y) <b. This inequality, however,
is impossible for y’4 0, since the function gy(«#)= f(#,0) assumes for
0 <a#< 1 all values of the interval (a,b) and the function f(a, y) gives
different values for different pairs of variables (a, y).
Thus we have proved Theorem 1.
From Theorem 1 we can immediately infer the following
COROLLARY 1. A plane cannot be continuously (1-1)-mapped on a
straight line.
Restricting Theorem 1 and its proof to functions f(x, y) defined for
0<a<1 and 0 <y <1, we obtain
COROLLARY 2. A square cannot be mapped continuously in a (1-1)
manner on a segment of a straight line.
It follows hence immediately that also
COROLLARY 3. A segment of a straight line cannot be mapped continu-
ously in a@ (1-1) manner on a square.
Indeed, if such a mapping existed, then, as we know from Analysis,
its inverse mapping would also be continuous, 7. e. it would be a map-
ping of a square on a segment of a straight line, which we have proved
~ above to be impossible.
On the other hand, there exists a (1-1) mapping of a segment
on to a square such that the abscissa of a point on the square is
a continuous function of the abscissa of a point on the segment
(Hahn [1], p. 36).
5*
68 IV. Sets of the power of the continuum

10. Continuous curve filling up a square. Of course a square can be

mapped continuously on a segment of a straight line; the projection of
a square on its side is an example of such a transformation.
It should be noted that, likewise, a segment of a straight line can be
mapped continuously on a square, in other words: that there exist continu-
ous curves filling up a square, which has been proved by G. Peano. Here
we shall give I. J. Schoenberg’s proof of the existence of such a curve
(Schonberg [1], p. 519).
Let f(t) denote a periodic continuous function with the period 2
defined by the conditions:
0 for: Sth
=\; for 2
Bi

f(t) is a linear function for + <t<#, and f(—t)=/f(t) for 0 <t <1.
The curve in question is defined by the equations

(10.1) £= Oo), Vy =v) © fore. 0 t1,

where

p(t)=Eft) + 18%)+35f(3%)+
(10.2)
y(t) = =f(3t)+59(3%) + 3yACB)+
From the inequality 0 </f(t) <1 for real ¢ it follows, by (10.2),
that 0 <gy(t) <1, 0 <yp(t) <1 for 0 <t <1, and that the two series
(10.2) are uniformly convergent in the interval (0,1), 7. e. that the func-
tions g(t) and y(t) are continuous. It remains to show that the curve
(10.1) meets every point of the square 0 <x <1,0<y <1. Let (a, yp)
denote a given point of this square. The dyadic expansion of the numbers
# and y, can be given in the form

Stata 92 | 93 veng

(10.3) ot
Yn =
Yo 9 t sat 93 Se 008 )

where a; (¢=1,2,...) are the numbers 0 or 1.

* Let
20g ed ae Dt ye 2OE
(10.4) tye gg an Pe a ee

If a, = 0, then we have
Digi, 1
Oe a a ae
 § 10. Continuous curve filling up a square 69

therefore f(t,) = 0; and if a, =1, then we have

cat cues ae
ee Pasay eet
therefore f(t,)=1. Thus in every case we have f(to)= a. Similarly
20x 20k41
3*{, = an even number + ——+ 2 =
ae
(2)

whence it follows that

(10.5) Otel Ge 2 FOL = k= 0. U 22

Formulae (10.2), (10.3) and (10.5) give p(t.) = 2%, p(to) = Yo, q. e. d.
Continuous curves filling up a square have been dealt with by a large
number of authors. The first example of such a curve was given by
G. Peano in 1890 (Peano [1], p. 157). The equations for the Peano curve
were also given by E. Cesaro [1], p. 257, while the geometrical interpre-
tation of the Peano curve is due to A. Schoenflies and E. H. Moore.
Another continuous curve filling up a square was given by D. Hilbert [1],
p. 459. See also my paper in Polish (Sierpinski [2]) and my note in French
(Sierpinski [1], p. 462-478).
As regards a (1-1) mapping of the segment 0 <t <1 on the square
0<a<1, 0<y <1, we observe that the functions g(t) and y(t), for
which the formulae « = p(t), y = p(t) give the mapping in question, can
be defined in such a way that for 0 <¢t <1 they will be continuous every-
where from the left (Sierpifski [14], p. 193-197). Thus F. Klein was
wrong when he said once that every (1-1) correspondence between the
points of a segment.and the points of a square is aS non-continuous as
can be imagined, e. g. as if we put all points of a square into a sack
and then shook them so that they all got well mixed together.
And here is what L. Couturat says on the subject: ,,Par conséquent,
ce qui constitue proprement et essentielment les continus a plusieurs
dimensions (comme le continu linéaire), ce n’est pas un ensemble de
points, mais un ensemble de relations. Ce fait a une portée philosophique
manifeste; il signifie, en somme, que l’espace n’est pas une simple ,,multi-
plicité“, mais bien une multiplicité ordonnée; et il justifie la conception
de Leibniz, qui voyait dans espace, avant tout, un ordre“ (Couturat [1],
p. 134).

11. Set of all infinite sequences of real numbers. We shall prove

THEOREM 1. The set of all infinite sequences of real numbers is effec-
tively of the power of the continuum.
Proof. By Theorem 1 of IV.7 we can define a function p(x) estab-
lishing a (1-1) correspondence between all real numbers and all infinite
70 IV. Sets of the power of the continuum

sequences of natural numbers. Now we shall define, in the set S of all

infinite sequences of real numbers, a function # in the following way.
For every natural number k there exist, as we know, natural num-
bers p, and q,, well-defined by the number k, such that & = 2?*~*(2q,—1)
(pe—1 being the index of the highest power of the number 2 dividing
the number k, and 2q,—1 the greatest odd divisor of the number &).
Let s denote a given infinite sequence of real numbers. Thus we have
$ == (#1, @a,...) where a (= 1, 2,...) are real numbers. For every na-
tural number 7, w(#;) is an infinite sequence of natural numbers y(a;) =
= (ni?, n?,...) defined by the element s of the set S and by the
number 7. Let

(idea O(8) = png, n&, ...) .

We shall show that the function # establishes a (1-1) correspondence
between the elements of the set S and the real numbers.
From the definition of the function y it follows that, for every in-
finite sequence 7,,%”2,... of natural numbers, y(n, %2,...) 18 a real
number. Thus the values of the function @ are real numbers.
Now let # denote a given real number. We shall try to determine all
elements s of the set S such that #(s) = wv. Let us assume that s is a given
element of the set S and w a given real number. Let s = (a, %,...) and
p(x) = (n?, n®, ...) for i= 1,2,..., and let p(x) = (m,, ms, ...) a8 above.
Therefore we shall have (11.1), whence

y(3(s)) = (n@, ner , ae

Thus in order that ?(s) = #, which, in view of the properties of the

function y, is equivalent to the formula

p(9(s)) = y(L) = (mM, Mg, ».-),

it is necessary and sufficient to have

(12) ne? —m for k=1,2,..

Let 7 and h be arbitrary natural numbers. In view of the defini-
tions of the numbers p, and gq, we have, for k = 2/-l(2h—1), p,=7 and
9% = h, and (11.2) gives

(11.3) ny = Ms—10,1, for natural 7 and h.

Therefore, if, for a given real wv, 0(s) = 4, then every infinite se-
quence n,n, ..., i.e. also the numbers | pa y(n, no, - 3 OR
 § 11. Set of all infinite sequences 71

j =1,2,..., are well-defined (by the number a). This proves that (for
@ given x) there exists at most one element s of the set S such that #(s) = a.
On the other hand, given a real number « for which y(a) =
= (M,, Me, ...), We Write 7; = y'(Mp 41.) Mgi—1.5) Mgi—1,)---) ANA $ = (%, Ly,...).
Introducing the notation (11.3), we shall have y(a;) = (n®, n®, Jie AOD t=
=1,2,..., and thus, according to the definition of the function 0, we
get (11.1). But (11.3) for 7 = p, and h = q, gives (11.2). Thus (11.1) gives
WAS) = NM Ms 2.0) = (since we have assumed that w(x) = (m,, Mme, 32)) ,
Thus for every real number x there exists an element s of the set S
such that 0(s) = a.
Therefore we have proved that the function ? establishes a (1-1)
correspondence between the elements of the set S and all real numbers;
we have thus proved Theorem 1.
EXERCISES. 1. Prove that the set of all infinite sequences of complex numbers
is effectively of the power-of the continuum.
Proof. By Theorem 1 we are able to define a function # establishing a (1-1) cor-
respondence between infinite sequences of real numbers and real numbers. Now let
S = (2, 2,...) denote an infinite sequence of complex numbers. Thus, for natural n,
we have z, =a,+0,% where a, and b, are real numbers. Let t(s) = (a, b,, de, bz, a3,
bs, ...). The proof that the function 7 establishes a (1-1) correspondence between all
infinite sequences of complex numbers and all real numbers presents no difficulties.
2. Prove that the set of all power series with complex coefficients

P(z)i= ao a,2-++ a,27-1 ...

is effectively of the power of the continuum.

Proof. Let t denote the function defined in the proof of Exercise 1. Clearly the
function y(P) = t(do, a, Ge, ...) establishes a (1-1) correspondence between all power
series with complex coefficients and all real numbers.
3. Prove that the set of all convergent infinite sequences of real numbers is
effectively of the power of the continuum.
Proof. Let S denote the set of all infinite sequences of real numbers, A — the
set of all convergent infinite sequences of real numbers, B — the set of all real numbers.
By Theorem 1 we are able to define a function # establishing a (1-1) correspondence
between the elements of the sets S and B. At the same time the function & establishes
a (1-1) correspondence between the elements of the set A (which is a subset of the set 8)
and the elements of the subset B, = 3(A) of the set B. Thus we have A ef ~B,c B.
Further, let A, denote the set of all infinite sequences having all terms equal to the same
real number. Clearly, Bef ~A,c A. Thus,the sets A, B, A,, B, satisfy the conditions
of Theorem 1 of JI.6, whence, by that theorem, we find that A ef ~B, q.e.d.
4. Prove that the set of all convergent infinite sequences of complex numbers
is effectively of the power of the continuum.
5. Prove that the set of alli finite sequences of real numbers is effectively of the
power of the continuum.
Proof. We can easily carry out the proof basing ourselves on the fact thatthe
set A of all finite sequences of real numbers is effectively equivalent to a certain sub-,
72 IV. Sets of the power of the continuum

set B, of the set B of all infinite sequences of real numbers. In order to obtain the func-
tion t establishing a (1-1) correspondence between the elements of the sets A and B it
suffices to assign to every sequence (@,,4,,...,d,) belonging to A an infinite sequence
COA Gp Wn coon Gan Un Or cox)
belonging to B.
Note. We shall not obtain an invertible function +t defined in the set A, if
we write
T(r sag teeetoLey =" Class Og totes ouanak Or Ose Ogre)

for every sequence (a,, d.,...,4,) belonging to A, since we then have e.g. t(1, 0) =
= 1(1,0, 0) = (1, 0, 0, 0,...), the sequences (1,0) and (1,0,0) being different ele-
ments of the set A.
Every set of non-empty disjoint sets whose sum is the set A is called a partition
of the set A. .
6. Prove that if NV is the set of all natural numbers, then the set Z of all partitions
of the set N is effectively of the power of the continuum.
Proof. Let S denote the set of all increasing infinite sequences of natural numbers.
From IV.7, Exercise 1, it follows that X ef ~S where X is the set of all real numbers.
Now let s«S, therefore s = (n,, n,...) where 1, n,,... are natural numbers and n,<
<<... Let P, = {my +1,n,+1, ...}, 9, = N—P,. The set {P,,Q,} is a partition of
the set N (since 1 «Q,): we denote it by f(s). Obviously for s « S, s’« S, s #8’ we shall
have f(s) ~ f(s’). Thus the function f is a (1-1) mapping of the set 8 on a certain sub-
set Z, of the set Z, whence S ef ~Z, c Z, and since X ef ~S, we have also X ef ~Z, c Z.
Now let P denote any partition of the set V; thus N = » 2B, the set Hmaking up
EeP
the partition P being disjoint and non-empty. For each set # « P let us denote by #(£)
the least natural number belonging to H. Since the sets H « P are disjoint, we shall
have &(H) 4 O(H’) for He P, H’ « P, HEA LH’. Thus the sets He P can be arranged
in a finite or an infinite sequence H,, H,,... according to the magnitude of the natural
numbers #(H#) corresponding to them, and every set H,, as a non-empty set of natural
numbers, can be arranged in a finite or infinite sequence according to the magnitude
of the numbers belonging to it: H, = n®, ne, ...}. In this way we obtain a certain finite
or infinite sequence of finite or infinite sequences of natural numbers

ni, no), n®, ces

2) 2 3)
nf . nf i ni Et ae

n®), n), n&, eee

Creare sh) eye y aay ows

Completing, if necessary, this double sequence with zeros, we shall obtain an

infinite sequence of infinite sequences of integers >0; and increasing in this double
sequence each term by 1, we shall obtain the double infinite sequence of natural numbers

nO 4 al not Mm nO dence

n+ Lis nO 4 li nO) + 1S 8S

n®) +.1Be, nO + Is n®) + Gs) ae

which we are able to arrange, for instance by the diagonal method, in a single se-
quence (m,, M2, ...). Let g(P) = (m,, mr, ...).
 § 11. Set of all infinite sequences 73

In this way we have assigned to each partition P « Z an infinite sequence of na-

tural numbers, and, as can easily be proved, to different partitions of the set N cor-
respond different sequences of this kind. But, according to Theorem 1 of IV.7, the set 7
of all infinite sequences of natural numbers is effectively of the power of the continuum.
Thus Zef ~T,c T. But T ef ~X, and, as proved above, X ef ~Z,c Z. Thus we have
Tef~Z,cZ and Zef ~7T,c T, whence, by Theorem 1 of II.6, we conclude that
Z ef ~T; therefore, in view of T ef ~X, we have Z ef ~X, q.e. d.
7. Prove that the set of all partitions of an a) denumerable set is effectively
of the power of the continuum.
Hint. It is an easy conclusion from Exercise 6.
8. Determine all the partitions of the set {a, b, c}.
Answer. There are five of them, namely the sets {{a, b, c}} ; {{a}, {b, c}! 5
{0}, fa, ch}, {Xe}, fa, d}}, {{a}, (b}, {3}.
9. How many partitions has a four-element set?
The set of all infinite sequences of real numbers is identical with the set of all
real-valued functions of a natural variable, i. e. with the set X where NV denotes the
set of all natural numbers and YX the set of all real numbers. Thus, from Theorem 1 we
can immediately infer the following
CoroLLaRy 1. If N denotes the set of all natural numbers, and X the set of all real
numbers, then the set XN is effectively of the power of the continuum.
Combining this corollary with Corollary 1 of IV.7 we obtain the formula X% ef~N%,
which is remarkable in view of the fact that N is a denumerable set while X is a non-
denumerable set.
We also have
COROLLARY 2. The set of all infinite sequences of real numbers >0 and <1 is effec-
tively of the power of the continwum.
In order to prove this corollary we need only use Theorem 1 and Theorem 2a
of IV.4.

12. Continuous curve filling up a denumerably dimensional cube. The

Cartesian product of an infinite sequence of sets A,, Ag,... is what
we call the set of all infinite sequences (a,, a,,...) where a, « A, for n =
=1,2,... We denote this set by A,xA,xA,;x... The Cartesian pro-
duct of an infinite sequence of sets generally depends on the order of
factors, still it can easily be proved that by changing the order of factors
in it we shall obtain a set which is equivalent to it. Indeed, if ,, mg, ...
is an infinite sequence, differing from the sequence 1, 2,3,... at most
in the order of its terms, then in order to establish a (1- 1) bGaliappmulones
between the elements of the Cartesian products A,x A, A;x... and
An X An, X An, X ... it suffices to associate with each element (a), a2, dg, ..-)
of the former an element (j,, Ang, Gng, ---) Of the latter.
If N denotes the set of all natural numbers, then we shall obviously
have for every set A:

AN ef WAX AXAX on,

74 IV. Sets of the power of the continuum

thus, in particular for the set X of all real numbers, we have X” ef ~X x

xXx... and therefore, in view of Corollary 1 of IV.11:
The set XX XXXX... is effectively of the power of the continuum.
Let us denote by X, the set of all real numbers # such that 0 <@ <1.
The set X¥xXxXx... is called a denumerably dimensional cube. There-
fore, in view of Corollary 2 of IV.11, we can say that
A denumerably dimensional cube is effectively of the power of the con-
tinuum.
It should be noted that the theorem asserting the existence of a con-
tinuous curve filling up a square can be generalized to a denumerably
dimensional cube. Namely we have the following
THEOREM 1. We are able to define an infinite sequence of continuous
functions of a real variable f,(t), fo(t), ..., such that 0 <f,(t) <1 for n=
= 1,2,..., and such that for every infinite sequence of real numbers a,,
Gg, ..., Where 0 <a, <1 for n=1,2,... there exists,a real number t (at
least one) such that 0 <t <1 and j,(t)= a, for n=1, 2, ...
Proof. The proof given here was found a few years ago by a Japan-
ese mathematician Kiyoshi Iséki [1].
Let f(¢) denote the same continuous function as in Schoenberg’s
proof (given above) of the existence of a continuous curve filling up
a square. For natural » and real ¢, let

(12.1) flee k=1

ee
aaaeaat
The functions /,(¢) are continuous for real ¢, being sums of uniformly
convergent series of continuous functions, and, in view of 0 <f(t) <1
for real t, we shall have 0 <f,(t) <1 for real t.
Now let a,, a,,... be an arbitrary infinite sequence of real numbers,
such that 0 <a, <1 for n=1,2,... Let

(12.2) Gpaeh yen ae

k=l
denote the dyadic expansion of the number a,. Thus, for natural n and k,
we have a= 0 or a=1.
Now we shall define an infinite sequence of the numbers ¢,, fa, ...
in the following way.
If m = 2”"*(2k—1), then let t,, = a. Thus, in view of (12.2) we shall
have

(12.3) On = 2 Bhi, dias

=1
 § 12. Continuous curve 15

Let

(12.4) tog yyBET tae

m=1

In view of (12.4), we have, for p=0,1, 2, ...

2 2
(12.5) 3 = 2lp + 3 bot + 33byte ty

where 1, is an integer equal to 0 for p = 0 and equal to 3’~*t,

+3°~#,+
+...+4, for p> 1.
Jit ,44.= 0, then, -since. ¢, = 0) or tg= 1-for k= 1, 2, :.., we have-in
view of (12.5) ©
Diao:
0<3%—2,<4+ayte = Go|

which, according to the definition of the function f/f, gives

f(3°t) =f (’t—21,) = 0 = tar.

Now, if 44140, then we have t,,:=—1, and formula (12.5) gives
2 < 3’t— 21, <1, whence, according to the definition of the function f,
we find /(37t) = f(3"t —2l,) =1=441.
Thus we have proved that f(3’t)=t,4, for p=0,1, 2, ...
In view of (12.1), we have, for a natural n:

Tilt) = 2 2 tan—t00n 1) ’
=r
wanehe by <(12.2) Cives wai dn, (Oren. 1, 2...
Thus Theorem 1 has been proved.
It will be noted that the function f,(t) (n=1,2,...), satisfying
Theorem 1, can be obtained from every continuous curve (10.1) filling
up a square by writing /,(¢) = oy"st(t) for »— 1,,2,..., 0 <t< 14).

13. Set of all continuous functions. We shall prove

THEOREM 1. The set of all continuous real functions of one real va-
riable is effectively of the power of the continuum.
Proof. Let A denote the set of all continuous real functions of one
real variable, and let fe A. Thus f,is a function having real values, and
continuous for every real value of the variable z.
By Theorem 1 of ITI.2 we can arrange the set of all rational numbers
in an infinite sequence
(13.1) Pett osthiogh ce
1) See my note Remarque sur la courbe péanienne, Wiadomosci Matemat. 42 (1936),
p. 1-3.
76 IV. Sets of the power of the continuum

Now let us define a function ? in the set A by writing

O(f) = (f(r), £(7%2)5 ++) -

Thus the function % associates infinite sequences of real numbers
with the elements of the set A. We assert that the function # is inver-
tible in the set A.
To prove this we assume that feA, ge A and d#(f) = d(g). There-
fore the corresponding terms of the infinite sequences Gaede es)
and (g(7r1), 9(72), +.) must be equal; thus we have f(%)= g(tn) for n=
=1,2,..., and since the sequence (13.1) is made up of all rational num-
bers, we have

(13.2) f(r) =qg(r) for every rational number 7.

Let r denote an arbitrary real number. Thus, by (13.2), we shall have

Ene Ena
(13.3) (| = (=) for n= Li 2 sere

(where Et denotes the greatest integer <7). But, as we know,

. Ene
lim aero

which, in view of the continuity of the functions f and g (since they

belong to the set A) gives

lim #(="") —f(«) and lim g (=) = g(a) .

Since by (13.3) the left-hand sides of these equalities are equal,
their right-hand sides must also be equal, 7. e. f(%) = g(a).
Thus for every real number w we have f(x) = g(«), 7. e. the functions f
and g are equal. Therefore we have proved that if fe A, ge A and #(f) =
= #(g), then f = g. This proves that the function @ is invertible in the
set A. This result can also be expressed in the following way: every
continuous function of a real variable is defined by the values which it as-
sumes for rational values of the variable.
Let us denote by S the set of all infinite sequences of real numbers,
and by S, — the set of all sequences #(f) where f « A; we shall have 8, C 8.
The function #, being invertible in the set A, establishes a (1-1) cor-
respondence between the elements A and S,. Thus A ef ~8,.
Let B denote the set of all real numbers; according to Theorem 1
of IV.11 we have S ef ~B. Thus we are able to establish a (1-1) cor-
respondence between the elements of the sets S and B. Hence the subset B,
 § 13. Set of all continuous functions ae

of the set B corresponds to the subset S, of the set S and we have S, ef ~B,.
Owing to the transitivity of the relation ef ~ formulae A ef ~S, and
S,ef ~B, give Aef ~B,.
Now let us denote by A, the set of all functions (continuous of course)
such that f(#) =a for xe B where a is a given real number. We have
of course Bef ~A,. Formulae A ef ~B,C B and Bef ~A, CA give,
by Theorem 1 of I1.6, the formula A ef ~B, which proves Theorem 1.
EXERCISES. 1. Prove that the set of all continuous complex functions of one
complex variable is effectively of the power of the continuum.
2. Prove that the set of all continuous (at least with respect to each variable se-
parately) real functions of two real variables is effectively of the power of the continuum.

14. Decomposition of a set of natural numbers into a continuum of almost

disjoint sets. We say that two denumerable sets are almost disjoint if they
have only a finite (> 0) number of common elements.
THEOREM 1. We are able to define a set effectively of the power of the
continuum of infinite sets of natural numbers each two of which are almost
disjoint.
Proof. For every real number «# inside the interval (0,1) let us de-
note by A(«) the set of all (obviously different natural) numbers

2"(2Ene-+-1), where n=—1,2, ...

The sets A(a#),-for 0 < # <1, will be infinite sets of natural numbers.
We shall show that, for « ~y, the set A(x)-A(y) is finite.
Indeed, for « 4 y we have |x—y| > 0. Now if, for certain natural p
and q,
2?(2Epe +1) = 2%(aKgqx +1),
then obviously p= q and Epx = Eqy, whence Epx = Epy, which gives
|pa—py| <1, 4. :
=
which proves that less than 1: |v—y| elements of the set A(x) belong
to the set A(y). Hence it follows that the sets A(x) and A(y) have a finite
(> 0) number of common elements.
Now let us denote by Z the set of all sets A(x) for 0<a4<1. For
0<#<1, 0<y<1, «fy, the sets A(x) and A(y) have been proved
to be almost disjoint, and therefore different. Thus the function 3, defined
in the set Z by the formula 0(A (a)) = w, establishes a (1-1) correspond-
ence between the elements of the set Z and the real numbers lying in-
side the interval (0,1). But the latter set, by Theorem 2 of IV.4, is
effectively of the power of the continuum. Thus the set Z is also effect-
ively of the power of the continuum.
 \

78 IV. Sets of the power of the continuum

Therefore the set Z satisfies the conditions of Theorem 1, which

has thus been proved.
In connection with Theorem 1 it will be observed that the following
theorem holds:
If m is a given natural number, then every infinite set Z of infinite
sets of natural numbers each two of which have at most m common elements
is effectively denumerable.
Proof. Let m be a given natural number and let Z denote a set
satisfying the conditions of our theorem. If A « Z, then A is an infinite
set of natural numbers which can be ordered according to their magni-
tude. Therefore let A = {n,, m.,...} where n, <7”, <... Let us write

i AD oF ee ok
f(A) will be a natural number well-defined by the set A. Thus f(A) is
a function having natural values, defined for A « Z. We shall show that
the function f is invertible in the set Z.
Let us assume that A4<¢Z, A’e«Z, AAA’, f(A)=—f(A’). Tf A=
== {14 Noy =f Where, <n, <9... and ACs int, ne, 1.3) Where ny << Ne < 2.2,
then f(A) = 2%+2%4 ...4.9%+1, f(A’) = gn 4 ora eb gtme therefore,
in view of f(A) =f(A’),
OO p Orme OM ont poets
Since 7,, Me, .--5%m41 and ni, N3,'..., M441 are natural numbers satisfy-
ing the inequalities n,<,.<...< m1 and ni < Nz <<... << Mii, this
gives, as we know, n;= nj for i=1, 2,...,.m+1.
Thus sets A and A’ have at least m+1 common elements, contrary
to our assumption that Ae« Z, A’eZ and A~ A’. Thus the function f
is invertible in the set Z and maps this set on a certain subset of the set
of all natural numbers. Hence it immediately follows that the set Z is
effectively denumerable, q. e. d.
EXERCISES. 1. Define a set Z composed of the continuum of different infinite
sets of natural numbers such that of each two different sets belonging to Z one is always
a proper subset of the other.
Solution. By Theorem 1 of III.2 we are able to arrange all different rational
numbers in an infinite sequence 7,,7,,... Now, for a given real number z, let us de-
note by A(a) the set of all natural numbers » such that r, < x; it will of course be an
infinite set. Clearly, for real x and y where « < y, we shall have A(x) c A(y), because
ifn « A(x), then we have r, < x and, in view of x < y, we have r, < y, whence n « A(y).
On the other hand, if w < y, then, as we know, there exists a rational number r such
that «<r<_y. In view of the properties of the sequence 7, 7,,..., the number r is
one of its terms; thus there exists a natural number m such that r = 7,,. Thus we have
x<1, <Y, Whence it follows that m¢€ A(x) and me A(y). Therefore A(x) #4 A(y),
and since A(«#)c A(y), the set A(x) is a proper subset of the set A(y).
 § 14. Decomposition of a set of natural numbers 79

Now let us denote by Z the set of all sets A(x) where x is a real number. Obviously
the set Z satisfies the required conditions.
2. According to S. Hartman two infinite sequences of natural numbers ay, dp, ...
and b,, b,... are called distant if for every natural number ¢ the inequality |a,,—b,| <¢
holds only for a finite number of pairs (m,n) of natural numbers. Give an example of
a continuum of infinite sequences of natural numbers each two of which are distant.
Solution. For real « > 1, let us denote by S(#) an infinite sequence w,, Ue, ...,
where u, = 27"(2Ena-+ 1)? for n = 1, 2,... We shall show that for «> l,y>1,«##y,
the sequences S(x) and S(y) are distant.
Let & denote an arbitrary natural number. We need only show that if at least.
one of the natural numbers mand nis > k-+ 1/|a—y], then |2°"(2Ema-+ 1)?—2°*"(2Eny+
+ 1)?| > k. Suppose that at least one of the numbers m and n is >k-+1/|x—y|. If we
had 2”"(2Emaz+ 1) = 2”(2Eny+1), we should have m=n and Emax = Emy, there-
fore |na—ny| <1, whence m= n < 1/|x—y|, contrary to our assumption that at
least one of the numbers m and n is >k+1/|~—y|. Thus we have 2"(2Emz+1) #4
#~2"(aEny+1). But
|2°"(2Ema-+ 1)?—2*"(2E
ny + 1)>|
= |2"(2E ma + 1)—2°(2Eny + 1)|-|2"(2Ema+ 1)4 2"(Env+ 1)|.

The first factor is >1 (since we have proved it to be 40) and the second is of
course >2”+ 2" > m+n, therefore >k+1/|~—y| >k. The product of these two
factors, and thus also the left-hand side of the last equality, is >k, q. e. d.
 CHAPTER V

COMPARING THE POWER OF SETS

1. Sets of less and greater power. Let A and B be two given sets.
If the set A is equivalent to (7. e. of the same power as) a certain subset
of the set B, but the set B is not equivalent to any subset of the
set A, then we shall say that the set A is of less power than the set B and
write

(1.1) i ASB.
Obviously for finite sets A and B this formula holds if and only if
the set A has fewer elements than the set B.
7 If formula (1.1) holds, we shall also say that the set B is of greater
power than the set A and write

BSA:
From the definition of formula (1.1) it immediately follows that
if A < B, then neither A~B nor B < A.
THEOREM 1. If A, B, A,, B, are sets such that Aes B, A~A, and
B~B,, then A, < B,.
Proof. From A < B it follows that the set A is equivalent to a cer-
tain subset B’ of the set B. Thus A~B’, and since A~A,, we have
A,~B’. In view of B~B,, the subset B’ of the set B is equivalent to
a certain subset Bj of the set B,. Thus B’~Bj, whence, in view of A,~B’,
we have A,~B;. Thus the set A, is equivalent to a certain subset of the
set B,.
If the set B, were equivalent to a certain subset A; of the set A,,
then we should have B,~Aj. In view of A,~A, the subset Aj of the set A,
would be equivalent to a certain subset A’ of the set A. Thus we should
have B,~Aj, Ai~A’, and since B~B,; this would imply (in view of
the transitivity of the relation ~) B~A’, and the set B would be
equivalent to a certain subset of the set A, contrary to our assumption
that 4 < B.
 § 1. Sets of less and greater power 81

Thus the set B, is not equivalent to any subset of the set A,, and
since we have proved above that the set A, is equivalent to a certain
subset of the set B,, we have Ay =a Bis q-ecd.
We shall write A = B instead of A~B and A < B instead of A < B
or A = B. Obviously if A < B, then the set A is equivalent to a certain
subset of the set B. We shall prove that the inverse is also true, 7. e. that
we have
THEOREM 2. Formula A <B holds for the sets A and B if and only
af the set A is equivalent to a certain subset of the set B.
Proof. In view of the remark made above it is sufficient to prove
that if the set A is equivalent to a certain subset B, of the set B, then
A <B. For this purpose we shall distinguish two cases:
1° The set B is equivalent to a certain subset A, of the set <A.
Thus we have A~B,C B and B~A,C A, whence, in virtue of the
Cantor-Bernstein theorem (Theorem 2 in II.6), 4~B, therefore A = B.
2° The set B is not equivalent to any subset of the set A. Since the
set A is equivalent to a certain subset of the set B, we have A < B.
Thus we have either A = B or A < B, therefore in any case A < B,
Gee. d:
Thus we have proved Theorem 2.
It will be noted that not only is it easy to deduce Theorem 2 from
the Cantor-Bernstein theorem, but also conversely the latter theorem can
easily be deduced from Theorem 2.
Indeed, let us assume that Theorem 2 is true and let A, B, A,, B,
be sets such that A~B,C B and B~A,C A. From Theorem 2 it fol-
lows that A <B and B <A. If we did not have 4 = B, we should have
A<Band B< A, which, as we know, is impossible. Therefore we must
have A= B, i. e. A~B, which shows the truth of the Cantor-Bernstein
theorem.
THEOREM 3. If A, B,C are sets such that A < Band B< C, then
A= C.
Proof. Since A < zB, the set A is equivalent to a certain subset B,
of the set B, and since B < C, the set B is equivalent to a certain subset Co
of the set C. Thus the subset B, of the set B is equivalent to a certain
subset OC; of the set C. Therefore we have A~C;C C.
If the set C were equivalent to a certain subset A, of the set A, then
the set B, being equivalent to the subset C, of the set C, would be equi-
valent to a certain subset A; of the set A, contrary to our assumption
Cardinal and ordinal numbers 6
82 V. Comparing the power of sets

that A < B. Thus the set C is not equivalent to any subset of the set 4
and since A~ OC; C Od, a 2 Ole que. a:
Similarly we can prove
THEOREM 3a. If A, B, C are sets such that A <B and B < O,.then
A <0:
It is likewise easy to prove that the following theorem holds:
THEOREM 3b. If A, B, CO are sets such that A <B and B <0, then
Teg 6h
2. Sets of greater power than finite sets and denumerable sets. Hypothesis:
of the continuum. We shall prove
THEOREM 1. Any infinite set is of greater power than any finite set.
Proof. Let A denote an infinite set, B a finite set. If the set B is
empty, then it is equivalent to the empty subset of the set A. If the set B
is not empty, then the number of its elements is a natural number n.
In II.2 we have proved that if the set A is infinite, then for every natural
number n it contains a subset of n elements. Thus the set B is any case
equivalent to a certain subset of the set A, but the set A, being infinite,
is not equivalent to any subset of the finite set B. Therefore we have
B< A. Thus we have proved Theorem 1.
The inverse theorem is of course also true: every set which is of
greater power than every finite set is an infinite set.
THEOREM 2. A set of the power of the continuum is of greater power
than any denumerable set.
Proof. Let A denote a set of the power of the continuum, 7. e.
A~X where X is the set of all real numbers, and let B denote a denume-
rable set. Thus the set B is equivalent to the set N of all natural numbers.
Therefore B~N C X, but the set X is not equivalent to any subset of
the set B, since, by Theorem 1a of III.1, every subset of a denumerable
set is finite or denumerable, and the set X, as we have proved in IV.2,
is non-denumerable, 7. ¢. not equivalent to any finite or denumerable
set. Therefore B < X, and since X~A, by Theorem 1 of V.1, we have
B< A. Thus we have proved Theorem 2.
Obviously, a set of greater power than a denumerable set is non-
denumerable. It will be noted, however, that without the aid of the axiom
of choice we are not able to prove that any non-denumerable set is
greater than any denumerable set. The proof of this theorem, based on
the axiom of choice, will be given in VI.7 (see Corollary 1, p. 114).
We do not know whether there exists a set which is of greater power
than a denumerable set and of less power than a set of the power of the
 § 2. Hypothesis of the continuum 83

continuum. The assumption that there are no such sets is called the
Continuum Hypothesis. So far the Hypothesis has neither been proved
nor disproved. K. Goddel has proved that the continuum hypothesis is
consistent with the generally accepted axioms of the Theory of Sets,
provided the latter are consistent with one another (Go6del [2] and [3]).
In the paper The present state of research into the foundations of mathe-
matics, written by A. Mostowski in collaboration with A. Grzegorezyk,
A. Jaskowski, J. Los, S. Mazur, H. Rasiowa and R. Sikorski, we find
the following statement:
“In 1939 Godel published a proof of the consistency of the Continuum
Hypothesis with the axiom of choice. That proof was closely bound with
the constructive trend. Namely Gédel defined a certain simple method
of constructing new sets from sets already known, and showed that if
the method was iterated an arbitrary transfinite number of times, we
should reach a class of sets (the so-called constructible sets) in which all
axioms of the Set Theory as well as the axiom of choice, the Conti-
nuum Hypothesis and some other hypotheses of the Set Theory are
satisfied’’.
It should be noted, however, that the same K. Gédel writes in his
paper [1] as follows: ,,Therefore one may on good reason suspect that
the role of the continuum problem in set theory will be this, that it will
finally lead to the discovery of new axioms which will make it possible
to disprove Cantor’s conjecture“. By ,,Cantor’s conjecture“ the author
means the continuum hypothesis.

3. Cantor’s theorem on the set of all subsets of a given set. We shall

prove
THEOREM 1 (of Cantor). The set of all subsets of any given set A is
of greater power than the set A.
Proof. Let A denote a given set, B the set of all subsets of the set A
(including the empty set). Let us denote by B, the set of all sets {a}
where a< A. Since {a}C A for ae A, the set B, is the set of certain sub-
sets of the set A (namely the set of all one-element subsets of the set A),
and since B is the set of all subsets of the set A, we have B,C B. The
function f, defined by the formula ,f(a) = {a} for ae A, obviously estab-
lishes a (1-1) correspondence between the elements of the set A and B,.
Therefore 4~B, C B. In order to prove that A < B it remains to show
that the set B is not equivalent to any subset of A.
Let us assume that, contrary to what we wish to prove, B~A,C A.
Thus there exists a function m establishing a (1-1) correspondence between
the elements of the sets B and A,. For ae A, we shall have g(a) « B,
4. €. g(a) will be a subset of the set A.
6*
84 V. Comparing the power of sets

Let us denote by Z the set of all those elements a of the set A, for
which a¢qg-(a) (leaving open the question whether or not Z = 0); thus
we shall have ZC A,, and also ZC A, therefore Z « B. Hence 9(Z) € A;.
Let a) = y(Z) and let us consider two cases:
1° a,eZ. From the definition of the set Z it follows that a) ¢ g(a)).
But, since a) = y(Z) « A,, we have g(a.) = Z. Therefore a, ¢ Z, contrary
to our assumption. .
2° a,¢ Z. Since, in view of a,—(Z), we have Z = g—(a,), it fol-
lows that a,¢gy—(a,), and, in view of a)= y(Z)« A,, and according to
the definition of the set Z, we have a, « Z, again contrary to our assump-
tion. Thus the assumption that B~A,C A leads in any case to a con-
tradiction. Therefore the set B is not equivalent to any subset of the
Seb:A, "q3,6. G.
Thus we have proved Theorem 1.
From Theorem 1 we can immediately infer the following
CoROLLARY. The set of all subsets of real numbers is of greater power
than the continuum.
It will be noted that some authors, .¢e. g. N. Lusin, do not consider
the set of all subsets of real numbers as adequately defined 1), since we
do not know any regular procedure of obtaining all sets of real numbers.
When we speak of it — says Lusin — we actually only imagine various
methods which can be conceived to define very complicated sets, but
we have no complete definition which would allow us to comprise them
all. But as regards man’s capacity for creating methods of defining we
enter the field of subjectivity. Thus the set of all linear sets is different
for different mathematicians and therefore it cannot be regarded as
a mathematical concept to be introduced in reasonings.
Similarly, R. Baire thinks it does not follow that being given a cer-
tain infinite set we are eo ipso given the set of all its subsets (Menger [1],
p., o14):
E. Kamke, on the other hand, points out that although from the
logical point of view nothing can be said against the rejection of the set
of all subsets of a set, still it is mathematics we want to deal with and
our object is not to take the most limited view but to give the widest
possible range to mathematical considerations. Because of the extra-
ordinary fertility of the theory of sets of points, dealing in particular
with the continuum and its parts, those sets have become so familiar
to us that any decree forbidding us to consider them in future is not to

1) Lusin calls it “totalité illégitime“ (Lusin [1] and [2]).

 § 3. Cantor’s theorem 85

be thought of. Therefore the set of all subsets of a given set should be
regarded as well-defined *).
The concept of definable linear sets has been dealt with in an exten-
sive paper by A. Tarski [6], p. 210-233.

4. Generalized Continuum Hypothesis. The assumption that, no matter

what the infinite set A is like, there is no set which would be of greater
power than A and of less power than the set of all subsets of the set A
is called the generalized Continuum Hypothesis. K. Godel has proved (I. ¢.)
that the generalized Continuum Hypothesis is consistent with the gener-
ally accepted axioms of the theory of sets, provided they are themselves
consistent.
It can easily be seen that the Continuum Hypothesis is a particular
case of the generalized Continuum Hypothesis, namely the case where A
is the set of all natural numbers, since, as we have proved in IV.7, Exer-
cise 5, the set of all subsets of the set of all natural numbers is of the
power of the continuum.

5. Forming sets of ever greater powers. Theorem 1 of section 3 permits

us to construct an infinite sequence of infinite sets of ever greater powers.
Indeed, let us denote in general by U(A) the set of all subsets of the
set A, and let N denote the set of all natural numbers. By Theorem 1
(Section 3) every term of the infinite sequence of sets

N, U(N), U(U(N)), U[U(U(N)),...

will be a set of less power than the next term.
We can likewise construct a set of greater power than any term of
a given infinite sequence of sets; for we have
THEOREM 1. If A,, A,,... 1s an infinite sequence of sets such that
A, < A, <..., then putting S = A,+A,+... we have 8S> A, forn =1,2,...
Proof. Let us assume that the sets A,, A,,... and S satisfy the
conditions of our theorem. Let n denote a given natural number. Thus
we have A,C WS and, in order to prove that S > A, it suffices to show
that the set S is not equivalent’ to any subset of the set A,.
Let us assume that the set S is equivalent to a certain subset of
the set A,. Then the set A,41, being a subset of S, would also be equi-
valent to a certain subset of the set A,, contrary to our assumption that
A, Hogep Thus we have proved Theorem 1.

1) The article of Kamke [1], § 13, footnote 88, gives further bibliography on this
problem (Borel, Brouwer, Holder, Weyl).
86 V. Comparing the power of sets

Theorem 1 can be generalized as follows:

THEOREM 2. If Z is a set of sets of different power none of which is
of greater power than any other set belonging to Z, then the set

s= DA
AEeZ

is of greater power than any set belonging to Z.

Proof. Since

s= 4,
A€Z

we have of course S> A for Ac Z. Thus, in order to prove Theorem 2

it suffices to show that S= A for no set A « Z.
Let us assume that for a certain set A,« Z we have c= i Now
let Ae Z, AA A,. From the assumption of Theorem 2 it follows that
A# A,, therefore Az S, and since S > A, we must have S > A, i.e.
AS > A. Thus we should have A > A for A« Z, A#~ A, contrary to our
assumption that no set belonging to Z is of greater power than any other
set belonging to Z. Thus we have proved Theorem 2.

EXERCISE. Prove that for every set A there exists a set F(A) of sets, such that
1° A « F(A), 2° if Xe F(A), then U(X) « F(A) and 3° if @ is an arbitrary set such that
Ae«@® and formula X «® always implies formula U(X) «@, then F(A)c@.
Proof. It is easy to verify that

Ff (A) = {4p+ U(A)+UU(A)+UUU(A)+...

is such a set F(A). e

THEOREM 3. For every given infinite sequence of sets we are able to

define a set of greater power than any term of the sequence.
Proof. Let A,, A,,... denote a given infinite sequence of sets. Let
S= A,+A,+... and 7 = U(S). We know from the proof of Theorem 1
of section 3 that the set S is equivalent to a certain subset of the set 7
(namely to the set of all one-element subsets of the set S). Hence, for
every natural number n, the set A, (being C S) is equivalent to a certain
subset of the set 7. On the other hand, if the set 7 were equivalent to
a certain subset of the set A,, then, in view of A, CS, it would be equi-
valent to a certain subset of the set S, which is impossible since 7 = U(8)
and therefore, by Theorem 1 of section 3, we have T>8S. Thus the
set T is not equivalent to any subset of the set A,, and since the
set A, is equivalent to a certain subset of the set 7, we have eS Wb
for n=1,2,... Thus we have proved Theorem 3.
 § 5. Forming sets of ever greater powers 87

From Theorem 3 it follows immediately that there exists no infinite

sequence of sets such that every set is equivalent to one of its terms.
It is easy to find what modifications should be made in the proof
of Theorem 3 in order to prove
THEOREM 4. Given any set Z of sets we can define a set of greater
power than any set belonging to Z.
It follows from Theorem 4 that whatever set Z of sets we should
define, we can always define a set not belonging to the set Z.
As we know, sets are constructed gradually: first we form certain
sets of objects which themselves are not sets, then sets of objects which
are not sets and of sets already defined, etc. The process of forming more
and more new sets in this way will never come to an end. We shall never
get the set of all sets. The concept of the set of all sets, not only all
those which have already been defined but also those which are being
defined or will be defined at any future time, is a notion whose meaning
is not well-defined. Such a set would be its own element, therefore it
would have to be given before being defined, since we cannot define a set
until all its elements are determined.
The concept of the set of all sets leads to contradictions which are’
called antinomies. There is no reason for anxiety on this account
since the concept of the set of all sets is contradictory in itself, and by
the method, which we have adopted, of constructing sets gradually the
set of all sets will never be obtained.
 CHAPTER VI

AXIOM OF CHOICE

1. The axiom of choice. Controversy about it. In 1904 E. Zermelo

stated an axiom (Zermelo [1], p. 514, and Zermelo [2], p. 261) which
later gave rise to a lively exchange of opinions among mathemati-
cians!). To-day we have an extensive bibliography on the subject of
this axiom and its applications. It reads as follows:
AXIOM OF CHOICE. For every set Z whose elements are sets A, non-
-empty, having no common elements, there exists at least one set B having
one and only one element from each of the sets A belonging to Z.
Another, equivalent wording of the axiom of choice is the following:
If a non-empty set S is the sum of disjoint non-empty sets, then there
exists at least one subset of S which has one and only one common element
with each of those sets (Fraenkel [3], p. 80).
With respect to an axiom which does not contradict intuition or
other, already accepted axioms two attitudes are possible: we can either
accept the axiom or reject it). As regards in particular the axiom of
choice, it should be taken into account, in any case, that 1° a large number
of particular cases of this axiom are true (which has been proved inde-
pendently of it); 2° from the axiom of choice a great; many conclusions
have been drawn of which none so far has led to a contradiction; 3° the
axiom of choice simplifies considerably various parts of the Theory
of Sets and of the Calculus and is indispensable for the proof of many
important theorems of those theories.
Finally let it be mentioned that in 1938 K. Gédel proved that the
axiom of choice is consistent with other generally accepted axioms of
the Theory of Sets, provided they are consistent with one another (Gédel
[2, 3]). Therefore, if any conclusion from the axiom of choice and other

1) ,,L’axiome du choix peut étre considéré comme le plus intéressant et — bien

quil ait moins de cinquante d’age — comme l’axiome le plus discuté de la mathématique
aprés Vaxiome des paralléles d’Euclide“ — writes A. A. Fraenkel [7], p. 431.
*) The rejection or non-acceptance of a given axiom should be distinguished from
its negation, 7. e. the acceptance of an axiom which is the negation of the given one.
 § 1. The axiom of choice 89

generally accepted axioms of the Theory of Sets resulted in a contra-

diction, it would mean that the latter axioms are not consistent. It is
conceivable, of course, that the axioms of the Theory of Sets might be
replaced by others, which might prove inconsistent with the axiom of
choice.
B. Russel [2] (p. 32-33) writes about the axiom of choice: “Numerous
mathematicians, like Zermelo himself, assert that this axiom is as ob-
vious as the other axioms and that it can be accepted without hesita-
tion. Others say that there is no reason to believe that the axiom is true.
Peano [2] (p. 145-148), having proved the independence of the axiom,
devotes to the consideration of its truth only the following remark: “Are
we to believe now that the proposition (la proposition) is true or that
it is false? Our attitude is neutral (indifférente)’’. Peano maintains in the
same paper that the question of obviousness is a psychological problem,
not concerning logic’’. For Russell the axiom of choice ceases to be ob-
vious once its meaning is understood.
“It is possible — he continues — that later on some one will find
the reductio ad absurdum which will show that the axiom is false. But
at present I think it merely doubtful. It may be true but it lacks obvious-
-ness, and the conclusions drawn from it are astonishing. In these circum-
stances I think it would be well to abstain from using it, excepting those
premises which suggest the possibility of coming across an absurdity and
thus solving in the negative the question of the truth of the axiom’’.
An entirely opposite view as regards the axiom of choice is that
held by A. Fraenkel [1], who writes: “... wer aber dieses Prinzip ver-
wirrt, nicht etwa weil es zu logischen Widerspruchen ftihrte — denn
soleche haben sich in der axiomatisch begrindeten Mengenlehre nicht
gezeicht — sondern weil es bisher in der Mathematik nicht anerkannt
oder nicht benutzt worden sei, der kann grundsatzlich mit demselben
Recht die gesamte Mengenlehre, ja iiberhaupt jede Wissenschaft und im
besondern auch die ganze Mathematik ablehnen, die ja ebenfalls letzten
Endes auf unbeweisenen, mehr oder minder einleuchtende Voraussetzun-
gen (Axiome) sich griindet. Das Auswahlprinzip als weniger einleuchtend
zu betrachten als alle anderen Axiome der Mathematik, dazu dirfte ein
ernstlicher Grund nicht bestehen”’.
The same author points out elsewhere (Fraenkel [3], p. 96-97) that
the axiom of choice has been introduced in the same way as the other
axioms of mathematics, which have been obtained by the analysis of
already known processes of reasoning. On these lines Greek mathematics
was led to the axiom of the parallels, which was accepted although there
was not, and, as we know to-day, there could not be any proof of it. And
even now, when we know that the axiom of the parallels could be rejected,
90 VI. Axiom of choice

nobody thinks of rejecting it or of abandoning the further development

of Euclidean geometry, based on this axiom. Similarly, it would not be
justifiable to reject those branches of mathematics which are based on the
axiom of choice unless we wanted to restrict the Set Theory radically by
deleting several very important parts of it. According to Hilbert the
axiom of choice is based on a general logical principle, necessary and indis-
pensable for the very foundations of mathematical deduction (Hilbert [3],
p. 89, and Hilbert [2], p. 152).
As regards the controversy about the axiom of choice the following
opinion expressed by H. Lebesgue in 1938 is worth quoting:
,»A aucune époque les mathématiciens n’ont été entierement d’ac-
cord sur ensemble de leur science que l’on dit étre celle des verités
évidentes, absolues, indiscutables et définitives; ils ont toujours été en
controverse sur les parties en formation des mathématiques, toujours
ils ont estimé que leur époque était une période de crise“. (Lebesgue [6],
Diet 22). }
E. Borel [4] (p. 22) writes: ,,... Une branche nouvelle de la science
se trouve ainsi créée, a savoir Vensemble des théoremes qui peuvent
étre obtenus en partant de l’axiome de Zermelo. J’ai proposé (C. R. Acad.
Se. t. 230 (1950), p. 1989) de donner le nom de mathématiques euclidien-
nes aux mathématiques dans lesquelles on n’admet pas Paxiome de Zer-
melo; les relations entre les mathématiques de Zermelo et les mathé-
matiques euclidiennes seront intéressantes a étudier et, & mon avis, il
serait particuliérement intéressant d’examiner si les mathématiques de
Zermelo peuvent conduire a des résultats intéressant les mathémati-
ques euclidiennes, mais difficiles a démontrer directement, comme cela
a été le cas pour les imaginaires. C’est, 4 mon avis, ce qui décidera dans
Vavenir de la véritable valeur mathématique de Vaxiome de Zermelo“.
Stil, apart from our being personally inclined to accept the axiom
of choice, we must take into consideration, in any case, its role in the
Set Theory and in the Calculus. On the other hand, since the axiom of choice
has been questioned by some mathematicians‘), it is important to know
which theorems are proved with its aid and to realize the exact point at
which the proof has been based on the axiom of choice; for it has frequently
happened that various authors have made use of the axiom of choice
in their proofs without being aware of it. And after all, even if no one
questioned the axiom of choice, it would not be without interest to in-
vestigate which proofs are based on it and which theorems can be proved.
without its aid — this, as we know, is also done with regard to other axioms.
1) According to J. Konig the axiom of choice is not an axiom in the usual
meaning of the word: this is seen from the very fact that we can argue whether it
is true or not.
 § 1. The axiom of choice oy

One of the main causes of the difference of opinions as regards the

axiom of choice was undoubtedly the fact that it was understood in diffe-
rent ways. Therefore, if we want to discuss the axiom of choice, we must
first be agreed as to what we want to understand by it. To make our
task easier we shall begin with the simplest cases of this axiom.
The simplest case of the axiom of choice is that in which the set Z:
consists of a single set A, Z = {A}. Then of course the axiom of choice
is reduced to the statement that if the set A is not empty, then there
- exists at least one object forming an element of the set A. This state-
ment, however, is true, since the propositions “the set A is not empty“
and “there exists at least one object forming an element of the set A“
are equivalent. Making this statement we do not at all want to assert
that we can indicate one element in every non-empty set, or that we can
choose a certain element from every such set: we merely assert the existence
of such an element. But what does it mean: “to exist‘?
It is the great and ancient problem of existence!) that underlies the
whole controversy about the axiom of choice.
We might hold the view that in mathematics to assert that there
exists an object having a given property is to negate the proposition that
no object has the property in question. (This view is not shared by the
so-called intuitionists).
According to Fraenkel [4], p. 35-36, on one hand, disregarding the
existential character of our principle (of choice), it has caused numerous
errors and misunderstanding even on the part of such a profound scholar,
accepting the classical theory of sets, as Julius Konig; on the other hand
it is exactly this existential and non-constructive moment which is the
reason why the majority of intuitionists of a radical and conservative type
regard our principle as unimportant or inadmissible, since the identifica-
tion of mathematical existence with the constructibility is the fundamental
principle of intuitionism.
It should also be noted that Poincaré, whose tendencies were largely
of an intuitionist character, was not opposed to the axiom of choice;
it was not our principle but non-predicative definitions that his criticism
of the proof of the theorem on good ordering concerned.
If we asked ourselves whether the axiom of choice, in its full extent
or narrowed in any way, is true or not, we should encounter one more
difficulty, namely the necessity of defining first what is truth. A great
many volumes have been written about this problem. It has been
devoted a paper of over one hundred pazes by A. Tarski [7], p. 23-25,
and [9], p. 261-405.
1) A. Fraenkel devoted a special lecture to the problem of existence in mathematics
(see Fraenkel [4], p. 18-32).
92 VI. Axiom of choice

As we pointed out above, if all that we know about a given set is

that it is not empty, then we cannot pretend to be able to define one of
its elements (so that we might be sure of two persons thinking of the
same object), or to be able to choose one element of that set. The axiom
of Zermelo is called the axiom of choice (Auswahlpostulat, axiome du
choix) and it is undoubtedly this rather unfortunate name that has been
one of the main reasons why some mathematicians have rejected the
axiom. Actually, accepting Zermelo’s axiom we say nothing of the
possibility of choosing one element from every set (belonging to the given °
set of sets).
“Man kann das Axiom auch so ausdriicken, dass man sagt — writes
Zermelo in [6], p. 266 — es sei immer moglich aus jedem Elemente M, NV,
R,... von T einzelne Elemente m,n,7,... auszuwablen und alle diese
Elemente zu einer Menge S, zu vereinigen“.
Dieser letzte Satz — writes Fraenkel [6], p. 232 — hat vielfach die
Meinung hergerufen der Kern des Axioms liege in der Forderung der
Moglichkeit der “Auswahl eines ausgezeichneten Elementes“ aus jeder
Menge M,WN,... oder in der Forderung der Moglichkeit der “gleich-
zeitigen Auswahl“ aus ihnen allen. Beide Auffassungen durften unzu-
trettend sein. Denn fir jede einzelne Menge WM oder, anders ausgedrickt,
fir den Fall, dass 7 nur ein einziges Element M besitzt, ist die in dem
Axiom geforderte Auswahl beweisbar; ist néihmlich a irgendein Element
der (voraussetzungsgemass) von 0 verschiedenen Menge WM, so existiert
die Menge {a} nach Axiom II!) und sie besitzt die gewunschte Higen-
schaft. Ist so die Auswahl fur jede einzelne Menge moglich, so ist es na-
tirlich auch gleichzeitig fir alle Mengen von 7’, da die Auswahl wie jede
mathematische Operation als etwas Zeitloses anzusehen ist.
The above mentioned proposition explaining the axiom of choice
should for the present, according to Zermelo’s kindly acknowledgement
by letter, “be understood merely as a remark, not concerning theory;
this proposition, as well as the name “axiom of choice’’, concerns only
the psychological method of presentation, while the axiom, as its wor-
ding, by the way, makes sufficiently clear, should be regarded as a pure
axiom of existence (Existenzaxiom)”’.

2. The axiom of choice for a finite set of sets. From the truth of the
axiom of choice for the case where the set Z consists of only one set A
it follows immediately by induction that it is true for every finite set
of sets. (H.g. let us suppose that A, and A, are two non-empty sets
without common elements: thus there exists an object p, which is an

1) It is an axiom stating that for every object we can form a set consisting of that
object only (the author’s note).
 § 2. The axiom of choice for a finite set of sets 93

element of the set A,, and an object p, which is an element of the set A,;
the set {p,, p.} will of course contain one and only one element from each
of the sets A, and A,). We shall not dwell upon this case any longer, since
even. those authors who reject the axiom of choice in its general form,
do not question it in the case of a finite number of sets.

2a. The axiom of choice for an infinite sequence of sets. The next
simplest case is the one where the set Z consists of an infinite sequence
of sets
(2a.1) Tae Wap! a po
Are we able to choose one element from each of these sets? It is
impossible to take this question too literally: actually in order to choose
one element from each of infinitely many sets we should have to perform
infinitely many operations, for which human life is too short. But the
problem which is presented here can be understood as a search for a law
according to which to each of our sets would correspond a certain element
of that set'). In other words, we should define a function f(n) of a na-
tural variable such that f(n)« A, for every natural number n. Obviously
this question is closely connected with the following one: “Can we estab-
lish a law according to which to every given non-empty set A would
correspond a certain element of that set?”
We are unable to establish such a law, not only in a general way
for every non-empty set, but also for sets forming certain given sets
of sets, e. g. for all non-empty sets of real numbers.

3. Hilbert’s axiom. However, we can accept the following logical axiom:

There exists a function ¢, associating with every property P for which
there exists at least one object having the property P an object e(P) having
the property P (Hilbert [2], Cippola [1]).
Let us accept this axiom and for every set.A let us denote by Py(p)
the following property of the object p: the object p belongs to the set A,
4. e. for every set A and for every object p let Py(p)=(pe« A). If the
set A is not empty, then there exists an object p such that p « A, 2. e. an
object having the property Py,; therefore it follows from the accepted
axiom that the object «(P4) has the property P 4, 7. e. belongs to the
set A. Thus the function 1(A) = e(P4) associates with every non-empty
set A a certain element of that set.
Thus from the logical axiom which we have accepted it follows that
there exists a function associating with every non-empty set a certain

1) Jt will be observed that as early as 1890 G. Peano [1], p. 210, wrote: ,,Mais
on ne peut pas appliquer une infinité de fois une loi arbitraire avec laquelle & une
classe on a fait correspondre un individu de cette classe“.
94 VI. Axiom of choice

element of that set. Accepting the existence of such a function we do not.

pretend at all to being able to define it.
In particular, from the accepted logical axiom we can immediately
deduce the following
THEoREM T. For every set there exists a correspondence according to
which to every non-empty subset of that set corresponds a certain element
of that subset +).
We shall prove that theorem T is equivalent to the axiom of choice.
Therefore let us assume that the axiom of choice is true and let M
denote an arbitrary set. For every non-empty subset N of the set WM
let us denote by Ay the set of all ordered pairs (p, V) where p « NV, and
by Z — the set of all the sets Ay where 04 NC M. The sets Ay forming
the set Z are of course non-empty and have no common elements. There-
fore, by the axiom of choice, there exists a set B having one and only one
element from each of the sets belonging to Z. Thus for every set N
such that 0A NC WM the set AyB consists of only one element, which
we shall denote by (py, VY). For 0ANC M, let +(N)= py. Clearly,
the function t associates with each non-empty subset N of the set M
a certain element t(N) of that subset. Therefore theorem T is true.
Thus we have proved that theorem T follows from the axiom of choice.
In order to prove that the axiom of choice follows from theorem T
let us assume that theorem T is true, and let Z denote any set whose
elements are non-empty sets A having no common elements. Let W
denote the sum of all the sets A forming the set Z. By theorem T
there exists a function + such that for 0~A NC M we have t(N) «WN.
Since for A « Z we have 0~ AC M, we shall have 1t(A)e A for Ae Z.
Let us denote by B the set of all elements t(4) where A « Z. Thus we
shall have t(A)« AB for A « Z, and, since the sets A forming the set Z
have no common elements, t(A) is the only element of the set AB. Thus
the set B has one and only one element from each of the sets A belonging
to Z. Therefore the axiom of choice is true.
Thus we have proved the equivalence of the axiom of choice and
theorem T.
4. General principle of choice. It is easy to prove the equivalence of
the axiom of choice and the so-called general principle of choice (Allge-
meines Auswahlprinzip ).
G. For every set Z of non-empty sets A (which may have common
elements) there exists a correspondence according to which to every set A
belonging to Z corresponds a certain element of that set tA.
1) A. Church calls Theorem T the axiom of choice, and to the axiom in section I’
gives the term multiplicative axiom (proposed by Russell [1]; see Church [2], p. 114).
 § 4. General principle of choice 95

It suffices of course to prove that theorem G follows from the

axiom of choice. Therefore let Z denote a given set of non-empty sets
and let M denote the sum of all sets belonging to Z.
We have proved that from the axiom of choice follows Theorem T,
according to which there exists a function 7 associating a certain element
of the set M with every non-empty subset of that set, 7. e. also with
every set Ac Z. This function satisfies Theorem G. Thus the axiom of
choice implies Theorem G, q. e. d. .
Some mathematicians, in applying the axiom of choice, distinguish
the case where Z is a denumerable set of sets A and the case where Z is
a non-denumerable set of sets A. H. g. E. Borel believes that, in principle,
we should distinguish the law of a denumerably many successive and
arbitrary choices (which, by the way, he thinks to be rather doubtful)
from the law of a non-denumerably many choices (successive or
simultaneous). “The last concept seems to me — he writes in [1]1) — to
be entirely devoid of sense. As regards a denumerable infinity of choices,
they cannot, of course, all be performed, but we can at least indicate
such a procedure that, if we establish it beforehand, we may be sure
that each choice will be made within a finite period of time; therefore
if two given systems of choice are different, we are sure to notice this after
a finite number of operations. When an infinite number of choices is
not denumerable, it is impossible to imagine a way of defining it, 7. e.
distinguishing it from an analogous infinite number of choices; thus it
is impossible to regard it as a mathematical creation which can be intro-
duced in arguments’’.
H. Lebesgue writes (Borel [1], p. 156): “I entirely agree with Hada-
mard when he declares that when we speak of an infinite number of
choices without giving their law the difficulty is equally great whether
we deal with a denumerable infinity or not’’.
When we apply the axiom of choice to a denumerable infinity of
sets, we say that we apply the restricted axiom of choice.
N. Lusin regards the role of the axiom of choice as follows: “For me
the proof of a theorem by means of Zermelo’s axiom is valuable only as
an indication that it is useless to waste time on an exact proof of the
falsity of the theorem in question?’ ?). This opinion had been expressed
by N. Lusin many years beforé K. Gédel proved the consistency of the
axiom of choice with other generally accepted axioms of the Theory of
Sets (provided they are consistent themselves).

1) This criticism, by the way, concerns only the possibility of choices (a priori)
and not the axiom of choice itself.
2) Cf. also Fraenkel [3], p. 86, who points out that such an attitude is in agreement
with the views of Hilbert.
96 VI. Axiom of choice

It will be observed further that the axiom of choice is also valu-

able as a heuristic tool: it makes it possible to discover theorems (or to
construct non-effective examples) for which we can then seek proofs
(or effective examples) on other lines. Every proof by means of the axiom
of choice represents at the same time a certain mathematical fact, namely
that from certain assumptions (assuming the axiom of choice to be true
in a particular case) there follow certain conclusions. In any case, as
we pointed out above, it is most desirable to distinguish between theo-
rems which can be proved without the aid of the axiom of choice and
those which we are not able to prove without the aid of this axiom ?).
Analysing proofs based on the axiom of choice we can 1° ascertain
that the proof in question makes use of a certain particular case of the
axiom of choice (indicating the exact point in the proof where that case
is used), or even state that all known proofs of the given theorem
refer to the axiom of choice;
2° determine the particular case of the axiom of choice which is
sufficient for the proof of the theorem in question, and the case which
is necessary for the proof. (To say that the given particular case of the
axiom of choice is necessary for the proof of the given theorem is to as-
sert that the given theorem implies the truth of that particular case of
the axiom of choice); _
3° determine that particular case of the axiom of choice which is
both necessary and sufficient for the proof of the theorem in question.
It will further be observed that the share of the axiom of choice in
proofs of the existence of sets, functions, etc., satisfying given condi-
tions, may be varied. The axiom of choice may be necessary for the
very construction of the set or function giving the desired example; or,
the set (or function) having been constructed without the aid of the
axiom of choice, only the proof that it has the desired properties may
resort to that axiom. Corresponding examples will be given later.

EXERCISES. 1. Prove with the aid of the axiom of choice that in every infinite
set A of real numbers there exists a denumerable subset # dense in the set A, 4%. e. every
interval containing inside at least one element of the set A contains also at least one
element of the set H.
Proof. According to the general principle of choice there exists a function 7 as-
sociating with each non-empty subset X of the set «4 a certain element t(X) of the set A.
Thus, in particular, for every interval d with rational end-points such that dd #0
t(dA) will be an element of the set A. The set H of all the elements t(dA) where d is

1) H. Lebesgue [4] writes: ,,J’ai dit ailleurs (II, 259) que je ne comprends pas ces
questions tout 4 fait comme M. Sierpinski, mais je suis plainement d’accord avec lui
sur le fait qu’il faut attacher une bien autre importance aux exemples qui n’utilisent
pas Vaxiome de Zermelo qu’a ceux qui lutilisent“.
 § 4. General principle of choice 97

an interval with rational end-points such that dA 40 will obviously satisfy the re-
quired, conditions.
2. Prove with the aid of the axiom of choice that the set F of all real functions,
defined in a given non-empty set A of real numbers increasing in the set A is of the
power of the continuum.
Proof. It suffices of course to assume that the set A is infinite. Let f(x) denote
a function belonging to the set F. It follows from III.3, Exercise 2, that the set A of
all the points of the set A at which the function f(x) is not continuous in the set A is
finite or effectively denumerable. Therefore we are able to, arrange all elements of the set As
in an infinite sequence a,, a2, d3,... (repeating, if necessary, the same term infinitely
many times if the set A, is finite). From Exercise 2 it follows (with the aid of the axiom
of choice) that there exists an infinite sequence b,, b,,..., dense in the set A. Now let
us associate with each function f belonging to the set # the infinite sequence

Oy, f(A), f(D1), Ge, f(Ge), f(Be), «+

In this way different sequences will be associated with different functions belong-
ing to the set F’. Indeed, if the same sequence corresponded to the function f and to the
function g belonging to F’, then the function g would be non-continuous in the set A
at the same points as the function f and would assume at those points the same values
as the function f. We should also have g(b,) = f(0,) for n = 1, 2,... and the function g
would be equal to the function f at each point of a certain set dense in the set A, whence
it would be equal to the function f at all those points of the A at which the function f
(and hence also the function g) is continuous. Therefore, we should have f(x) = g(x)
for xe A.
Since there is a continuum of infinite sequences of real numbers (see IV.11), it
follows hence that the set & is of power <c (by ¢ we shall denote the power of the
continuum). On the other hand, the set F is of power >c since to this set belongs every
function f(x) defined for x « A by the formula f(z) = «+ ¢ where ¢ is an arbitrary real
number independent of x. Thus we have F =, qeaeside ;
Observe further that it would be easy to prove that the set of all increasing real
functions of a variable is effectively of the power of the continuum. For this purpose
we need only take the effectively denumerable sequence formed of all rational numbers
as the sequence ),, do, ...

5. Axiom of choice for finite sets. For a given natural number n let
us denote, after A. Mostowski, by [n] the following particular case of the
general principle of choice G:
[n] For every set Z of n-element sets A there exists a correspondence
such that to each set A belonging to Z corresponds a certain element of it, t(A).
Just as easily as we have proved above the equivalence of Theorem G
and the axiom of choice we can prove the equivalence of Theorem [n]
and the following particular case of the axiom of choice:
For every set Z of n-element disjoint sets there exists a set having one
element from each of the sets belonging to Z.
However, it is not for every set Z of n-element disjoint sets that
we are able to indicate a set having one element from each of the sets
Cardinal and ordinal numbers 7
98 VI. Axiom of choice

belonging to Z. For let 7 denote the set of all real functions defined
for 0 <«% <1 and assuming only the values 0 or 1. Two functions f(x)
and g(x) of the set 7 will be assigned to the same class if there exists
an integer k such that

g(x“) = j(v+2—B (0+ =} for some n>1.

It can be proved that if we could choose one and only one element from
each of the considered classes it would be possible to prove the existence
of linear sets non-measurable in the sense of Lebesgue without the aid of
the axiom of choice (Sierpinski [50]). Thus we cannot choose one and
only one element from each of the two-element sets forming a certain
set of sets. Still, it will be observed that we cannot give an effective
example of any pair of two functions of a real variable from which we
should not be able to choose one of those two functions.
(I have been asked whether such a pair of functions could not be
defined in the following way: the pair {sinw, cosx} if the last theorem of
Fermat is true, and the pair {tga, ctga} if the last theorem of Fermat is
false. From such a pair we are able to choose one function, eé. g. sing if
the theorem of Fermat is true, and tga if it is false).
THEOREM 1. [2]—[4]?).
Proof. Let Z denote any set of four-element sets, A = {a,, a, 3, 4}
denote any set belonging to Z, and A* — the set of all two-element
(not ordered) subsets of the set A:

A* = {{a1, Az}, {Ay, Ms} {M,, Hg}, {2,3}, 1d2, Aas, (45, a,}} :

As we know, [2] implies the existence of the function g associating

with each two-element set {x,y}, where re A, ye A, Ae Z, one of the
elements 7, y.
Dhus pe. 9))e.0,9) tor geA, ye Ave y.)Ave2. Kor Awl. 2,
3,4, let m denote the number of those pairs {x,y} belonging to A* for
which y({v, y}) = a;. The numbers n;(¢ = 1, 2,3, 4) are of course posi-
tive integers and zero, and n,+7.+7,+%” = 6, whence it follows that
they are not all equal to one another. Let us suppose that n, <n, <
<mz <n, and let B denote the set of all elements a; for which n; = ,.
Obviously the set B has at least one and at most three elements. If the
set B has one element, let us denote it by y(B). If the set B has three
elements, then the set A—B has one element, which we shall denote
by p(B). Finally if the set B has two elements, then let y(B) = (B).
Therefore we shall always have (A) « A.
1) See Mostowski [3], p. 138. [2]—>[4] signifies here that [4] can be deduced from [2]
and other generally accepted axioms other than the axiom of choice.
 ‘

§ 5. Axiom of choice for finite sets 99

Thus the function y associates with every set A « Z a certain element

of that set. This proves the truth of [4].
Thus we have proved Theorem 1. A. Mostowski attributes the proof
given here to A. Tarski (Szmielew [1], p. 79, Mostowski [3], p. 164,
Lemma 13).
THEOREM 2. For natural k and n we have [kn]+[n].
Proof. Let us suppose that the theorem [kn] is true, and let Z denote
a given set of n-element sets. For X « Z let us denote by X* the set of
all ordered pairs (v%,7) where we X and 7 is a natural number <k, and
let Z* denote the set of all sets X* where X « Z. Thus Z* will be a set
of kn-element sets, and the assumption that the theorem [kn] is true
implies the existence of a function » such that y(X*) « X* for X« Z.
If for a given set X « Z we have o(X*) = (a,7), then let p(X) = x. Ob-
viously we shall have y(X) « X for X « Z, which proves the truth of theo-
rem [n]. Thus we have proved Theorem 2.
In particular, from Theorem 2 it follows that [4]—[2], which, in
virtue of Theorem 1, gives the following
CoROLLARY. [4] = [2].
Thus the axiom of choice for a set of four-element sets is equivalent
to the axiom of choice for a set of two-element sets.
It should be noted, however, that neither the implication [2]~[3]
nor the implication [3]—[4] holds (Mostowski [3], p. 164, Theorem VII,
and p. 138). A Mostowski has proved that the implication [2]—[n] holds
only for n = 2 and n= 4. According to Mostowski the proposition “the
implication [m]—[n] does not hold’? means that the proposition [n] is
independent of the axioms of the Theory of Sets and of the proposition [m]
(Mostowski [3], p. 151).
Lemma. If p ts a prime factor of a natural number n and tf the proposi-
tion p is true, then for every set Z of n-element sets there exists a function g®
associating with every set A belonging to Z a certain proper non-empty subset
of it, g@(A) (Mostowski [3], p. 164, Lemma 15 (of Tarski).
Proof. Let p denote a prime factor of the number n and let us as-
sume the proposition [p] to be true. Let Z denote a set of n-element
sets. Let us denote by T the set of all p-element subsets of the sets belong-
ing to Z; it follows from [p] that there exists a function t such that
TEX rex ior X42.
Let A denote a set belonging to Z, 7. e. having n elements. The
set A has

(i
p
7%
100 VI. Axiom of choice

different p-element subsets. For each element of the set A let us denote
by q(a) the number of all those different p-element subsets X of the set A
for which t(X) =a. We shall have of course

and as the number

in n(n—1)...(n—p+1)
| Ledeen p

is not divisible by (the number (n—1)...(~—p—1) not being divisible

by the prime factor p of the number n), the numbers q(a) cannot all be
equal for a¢« A (since then the sum

DS: q(4) ,
acA

having » components, would be divisible by 7).

Thus the set A is decomposed into two non-empty disjoint sets A =
= A’+A" where A’ denotes the set of all those elements a of the set A
for which q(a) = maxgq(a«). The set A’ is well-defined by means of the
xeA

function +. Let g?(A) = A’ for Ae Z. The function go satisfies our

Lemma, which has thus been proved.
This lemma easily implies that for every natural number n the fol-
lowing theorem, 7’,, is true: if the proposition [p] is true for every prime
number p <n, then proposition [k] is true for every natural number
k<n. f
Indeed, theorem 7, is true. Now let n denote a natural number >1
and suppose that the theorem 7; is true for every natural number k < n
and that the proposition [p] is true for every prime number p <n. If n
were a prime number, then, in virtue of our assumption, the proposi-
tion [n] would be true; also, by our assumption, the proposition 7',_; is
true. Hence it immediately follows that theorem 7’, would be true.
Further let us assume that n is a compound number and let p denote
its smallest prime divisor: thus we shall have p<n, by our ‘lemma
a function gen will exist. Therefore, if A is a set having n elements,
then g(A) is a non-empty proper subset of the set A. Thus the set gS (A)
has at least one and at most ~—1 elements. But, since in virtue of our
assumption proposition [p] is true for every prime number p <n, and
since theorem T7,,_, is true, the proposition [k] is true for every natural
number k <n—1. Thus there exists a function gy, associating with each
non-empty proper subset of the n-element set A a certain element of
 § 5. Axiom of choice for finite sets 101

that subset. Thus we shall have 1(A) = 9(g$(A)) «A, which implies the
truth of the proposition [n], and therefore, in view of the truth of proposi-
tion [k] for k <n—1, also the truth of theorem 7’.
We have thus proved by induction the truth of theorem 17, for every
natural n.
Hence it immediately follows that if proposition [p] is true for every
prime number, then proposition [n] is true for every natural number.
As a further conclusion we shall prove
THEOREM 3. If n is a natural number and if proposition [p] is true
for every prime number p <n, then proposition [k] ts true for every compound
number k < 2n+1.

Proof. If proposition [p] is true for every prime number <n, then,
as we have proved, theorem IT, holds, 7. e. proposition [k] is true for
every natural number k <n. Now let k denote a compound number
<2n+1 and p — the smallest prime divisor of the number k. Thus we
have k= Up where | >2, therefore 2p <lp=k <2n+1, whence 2p <
<2n-+1, therefore 2p < 2n and so p <n; thus, in virtue of our assump-
tion, proposition [p] is true. According to the lemma, there exists a function.
Te associating with every k-element set A (belonging to a given family Z
of sets) a certain non-empty proper subset of that set, Q@ = OCA), There-
fore that subset has 1,2,... or k—1 elements. Since proposition [1] is
true for every natural number / <n, there exists for such J a function fp
associating with each l-element subset of the set A a certain element.
of that subset. If Q= L<n, then let +(A) = f,(Q), and if Ve nm, then,,
in view of @ <k—1l and
A = k < 2n+1, we shall have 1 < A_Q =1 <n,.
and we shall write +(A) = f,(A—Q). Thus we shall always have 1(A) « A..
Therefore proposition [k] is true for compound k < 2n+1, q.e.d.
Thus in particular, for n = 2, we obtain [2]—[4]. For n = 4 Theo-
rem 3 gives ([2]|[3])—([6][8][9]). But from Theorem 2 it follows that.
[6]—([2]-[8]). Therefore we have
THEOREM 4. ([2]+[3]) = [6].
In view of ([2]-[3])—(({6][8][9]) we also have [6]—[8][6]—[9].
For n= 6 it follows from Theorem 3 that [2][3][5] imply [k] for-
every compound k <12, and for n= 10 — that [2][8][5][7] imply [4],
for all compound k < 21.
We shall now prove
THEOREM 5. ([{2]-[5])—[8].
Proof. Let us assume that propositions [2] and [5] are true; in
virtue of Theorem 1 proposition [4] is also true. In view of [2], [5],. [4]
102 VI. Axiom of choice

there exist: a function m which associates with every two-element subset X

of a set belonging to a given family Z of eight-element sets the element
y(X) « X, a function y which associates with every five-element subset Y
of a set belonging to Z the element w(Y)« Y and the function ? which
associates with every four-element subset U of a set belonging to Z the
element 3(U) « U.
Now, in view of [2] and of our lemma, there exists a function Gs
Let A denote an eight-element set belonging to the family Z. Let Q = g$)(A):
as a non-empty proper subset of an eight-element set the set Q has 1, 2,
3, 4, 5,6 or 7 elements. In each of these cases let us take respectively:
1) t(A)= 4(Q@), 2) t(A)=(Q), 3) t(A)= p(A—Q), 4) t(A) = 0(Q),
5) t(A)= y(Q), 6) t(A) = o(A—Q), 7) t(A) = 1(A—Q@), where the sym-
bol «(X) denotes the unique element of the set X if X contains only one
element. We shall have +(A)e A for Ae Z, which proves the truth of
proposition [8]. Thus we have proved Theorem 5.
It will be useful to compare Theorem 5 with formula ([2]-[3])—[8]
obtained above (from Theorem 3, for n = 8).
In view of Theorem 2 we have [10]—([2]-[5]): thus from Theorem 5
we infer the following
COROLLARY. [10]—[8].
We shall also prove that

([2]-[3]-[5]) [15] .
In order to prove this, it suffices to observe that if the set Q@ = g$(A)
had 7 elements, then the set A—Q would have eight of them, and, as
we know, ([{2][3])—[8]. °
We leave it to the reader to prove that

([2][3][5]) [16], ([2][s][11]) [16]

and
([2][5][7][13]) +[16].
EXERCISES. 1. Prove that ({2][5][7])>[10], ([2] [3] [5] [7])
>((25] [27)),
((2] [3] [7]) > (14), ((2] [8] (5) (7) (11)
= ([28] [83] [85]), ((2] [8] [5] (7] [11] [13])
+ [44],
(({3] [5] [7] [11] [17]) > [40].
2. Prove a theorem of W. Szmielew, stating that

((2] [3] [5] [7] [11] [13] [17])

+[63].
(in other words, that if [p] holds for all prime numbers p < 17, then we have [63]).
Proof. From Theorem 3 for n = 18 and from our assumption it follows that [k]
holds for every compound number <36. Thus the assumption that [p] holds for prime
numbers p < 17 implies that the function 7, assigning to each l-element subset of the
set A (of 63 elements) a certain element of that subset exists for all natural 1< 36
except 1 = 19, 23, 29 and 31. From the last two formulae of Exercise 1 it follows more-
 § 5. Axiom of choice for finite sets 103

over that f, exists for |] = 44 andl = 40. If we take Q = gS)(A) where Q etl it (Al) —
= f,(Q) for 1= 40, 1 = 44 and for 1 < 36, except the numbers / = 19, 23, 29, 31, and
for those numbers respectively t(A) = f,,_,(A —Q) (we shall then have 63—1 = 44, 40,
34, 32), and for 36 < 1< 62 where 1 # 44 andI1#4 also 1(A) = f,,_,(A—Q), we shall .
always have t(A) «A, which proves that [p] for prime numbers p< 17 implies for-
mula [63].
3. Prove that ([2][3][5][17] [13]) [32] and ([2] [3] [5] [7] [11] [13] [17] [23])
>[64].
We shall also prove the following formula of A. Mostowski?):
/
([3]-[7]) >[9] -
Proof. The only difficulty arises in the case where the set 7 = g)(A) has four
elements. In view of [3] there exists a function y(X) associating with each three-element
subset X of the set A a certain element of that subset, and in view of [7] there exists
a function y(X) associating with each seven-element subset of the set A a certain ele-
ment of that set.
Let P denote the set of all two-element subsets of the set 7. Thus P = oo For
X «P let f(X) =y(A—X) and let § = {f(X)}y-p. If S = 1, let z(A)= 4(8); if &
let t(A) = y(A—S); if S = 3, let t(A) = o(S). If S = 4, then the set # of all the ele-
ments a«S for which there exist more than one set X «P such that f(X)=a ob-
viously has one or two elements (since f is then a function defined in a set of six elements
and assuming’four different values). If R = 1, let r(A) =«(R), and if R = 2, let r(A) =
=y(A—f).
If S = 5, then there exists one and only one element ae¢«S for which there are
more than one element X <P such that f(X) =a. Then let t(A) =a.
If S = 6, let t(A) = g(A—S).
Thus we have avoided the difficulty of the case 7 = 4. In case of T = 5 the set
A~—T would have four elements and we should deal with it just as we have done with
the set 7. In case of 7 = 6, 7 or 8, we should take respectively (A) = p(A—T), p(T)
or ((A—T).
Thus we have proved that [3] [7]—[9].
However, we do not know whether [3] [5][13]~[15] (Mostowski [3], p. 168).
Moreover let us observe the following. Let n denote an arbitrary positive integer.
From the assumption that Theorem [n] is true we can deduce, without the aid of the
axiom of choice, the existence of linear sets non-measurable in the sense of Lebesgue ”).

In connection with the axiom of choice for finite sets let it be ob-
served that A. Fraenkel has dealt with two problems: 1° whether the
axiom of choice for sets of two-element sets is more restricted than the
general principle of choice, and 2° whether the axiom of choice for finite
sets is more restricted than the general principle of choice. According to
A. Mostowski both works of Fraenkel quoted here have gaps, but their
general idea is correct. Mostowski makes use of it giving the correct solu-
tion of problem 2° in his paper [6] (p. 201-252), where he investigates

1) See Mostowski [3], p. 164, Theorem IX, and Sierpinski [67], p. 98.
2) See Sierpinski [50], p. 36. For m = 2 see also Sierpinski [13], p. 177.
104 VI. Axiom of choice

the relation between the principle of ordering and the axiom of choice,
the solution of problem 1° being a corollary to paper [6] of Mostowski.
THEOREM 6. Let n denote a given natural number.
Theorem T,,, stating that there exists a function f,(Z) associating with
every set Z of n elements a certain ordered set f,(Z) of the same elements,
is equivalent to the theorem stating that proposition [k] is true for every na-
tural number k <n.
Proof. Suppose that Theorem 7, is true; thus there exists a func-
tion f,. Let k denote a natural number <n. Let us associate with every
set Z of k elements a set g,(Z) of n elements (obviously different) which
are the elements of the set Z and ordered pairs (Z,7) where j is a natural
number <n—k. Thus the set f,(g,(Z)) is an ordered set whose elements
are all the elements of the set g,(Z). The set Z- fulgu( Z)) will be of course
an ordered set of the same elements as the set Z. Denoting in a general
way the first element of the ordered (finite) set A by p(A) and writing
t(Z)= p[Z-fa(gn(Z))|, we shall have a function associating with every
k-element set a certain element of that set. Thus we have Broyes that
T,—>(k] for k <n.
Now suppose that proposition [k] is true for natural k <n. Thus
for k <n there exists a function h,(A), associating with every k-element
set A a certain element h,(A) of that set. Now let Z denote an arbitrary
set of n elements. Let us denote by /,(Z) the sequence of n terms

PZ) otaZ th Z))) ) hgnalZ— hi, Z) pee Ze

he(Z —{hn(Z) Rn—s(Z), «+5 Ng(Z)}) ,¢(Z— {hn(Z), «++ Ro(Z)})
— it will be a sequence composed of all elements of the set Z. Thus the
function f,(Z) assigns to every set of n elements a certain ordered set
of the same elements, which proves that Theorem 7, is true.
Thus our theorem on the equivalence 7, =[1][2]...[m] has been
proved.
Therefore in particular 7, =[2], and, in view of [2][3] = [6], we
have T, = [6] and it can be proved that [3] does not imply 7,. Similarly,
in view of [4] =[2], we have 7, = [6], therefore T, = T,.
In view of [2] = T, it is easy to prove that proposition [2] implies
that the sum of a denumerably infinity of two-element sets is a denumerable
set, and more generally, in view of [1][2]...[n] = T, (for n=1,2,...),
it follows from [1][2]...[n] that the sum of a denumerable infinity of n-ele-
ment sets 1s a denumerable set.
It is also easy to prove that [k], for natural k, implies that the sum
of a denumerable infinity of- finite sets is a finite or denumerable set.
‘
 § 5. Axiom of choice for finite sets © 105

Let [s,] denote the theorem stating that there exists a function
assigning to every denumerable set a certain element of that set. [s,] imp-
lies that there exists a function f assigning to every denumerable set
a well-ordered set (XIII.1) composed of the same elements as the given
denumerable set; the proof, however, is not easy — it is a particular
case of Zermelo’s theorem for denumerable sets. We are not able, howe-
ver, to deduce from [s,] that there exists a function assigning to every
denumerable set an infinite sequence composed of all elements of that
set. We can deduce the existence of such a function from [c], where [c]
is the theorem stating that there exists a correspondence associating
with every set of the power of the continuum a certain element belong-
ing to that set.
Indeed, let us assign to every denumerable set Z the set s(Z) of all
infinite sequences composed of all elements of the set Z. Thus if
Z = {A,, 4, ...}, then the set s(Z) is the set of all infinite sequences
An.) Angy «+, Where 1, %,... 18 any sequence composed of all natural
numbers.
Therefore the set s(Z) will be of the power of the continuum. From
[No] it follows that there exists a function g assigning to every set of the
power of the continuum a certain element of that set. Thus the function
1A) = g(s(Z)) assigns to every denumerable set Z a certain infinite
sequence, composed of all elements of that set.
Hence we conclude that [c] implies that the sum of a denumerable
infinity of denumerable sets is a denumerable set. We are not able to deduce
this last theorem from [s,]. We even cannot deduce from [s,] the theorem
stating that the sum of a denumerable infinity of denumerable well-
ordered sets is a denumerable set.

6. Examples of cases where we are able or not able to make an effective

choice. Let Z denote a set of non-empty disjoint sets A. Let us now
consider the determination of a set B such that for every set A « Z the
set AB has one element.
1° Let us divide into classes all infinite sequences of polynomials
in one real variable, convergent for 0 <a# <1, including in the same
class all sequences converging to the same limit function. In this case
the definition of the set B containing one and only one sequence from
each of those classes is difficult but possible!). The problem would not
be easier if we confined ourselves to sequences of polynomials with ra-
tional coefficients. On the other hand, it would be much easier if we

1) In order to define the set B we should refer to the investigations of R. Baire

concerning non-continuous functions (see Kuratowski [2], p. 101).
 106 VI. Axiom of choice

considered only uniformly convergent sequences of polynomials (Sier-

pinski [4], p. 177).
Now, if we divided into classes all convergent sequences of real
numbers including in the same Class all sequences converging to the same
limit, then the definition of the set B having one sequence of each class
of that kind would present no difficulty (e.g. we would take as B the
set of all infinite sequences

Ena
a (nae 2h)

for real #).

2° Let us divide into classes all double sequences of polynomials
convergent in the interval (0,1) (¢. e. such that there exists a limit

lim [Tim frnn(2)]) 5

including in the same class all sequences converging to the same limit
function. Here we are not able either to define a set B containing one
and only one double sequence from each class or to prove without the aid
of the axiom of choice the existence of such a set B.
3° Let us divide all infinite sequences of real numbers into classes
in such a way that each class will contain all sequences that differ only
in the order of their terms. We are not able to define a set B containing
. one and only one sequence from each class of that kind.
We should be able, however, to define an analogous set B if our
division concerned only infinite sets of rational numbers.
4° Let Z denote the set of all two-element sets A = {P,Q} where P
and Q are sets such that if XY is the set of all real numbers, then we have
P+Q=X and PQ = 0. We are able to define a function f associating
with each set A belonging to Z a certain element of it, f(A). For this
purpose it suffices, for A = {P,Q}, to denote by f(A) that one of the
sets P and Q which has number 0 as element.
But if Z denotes the set of all two-element sets A = {P,Q} where P
and @ are denumerable sets such that P+-QC X and PQ = 0, then we
are not able to define a function f which associates with every set A belong-
ing to Z a certain element of it.
EXERCISES. 1. Define the correspondence assigning to every set of real numbers
containing more than one element a certain non-empty proper subset of that set.
Solution. From Theorem 1. of III.2 it follows that we are able to define the in-
finite sequence 7,, 7,, ... of all rational numbers. If X is a set of real numbers containing
more than one element, then there exist rational numbers r such that the set of all num-
bers of the set X that are smaller than r is a non-empty proper subset of the set X.
 § 6. Examples 107

¥or if the set X has more than one element, then there exist real numbers a and b such
that ae X, b« X, a < b, and it suffices to take as r a rational number lying between a
and. b. Now let 7, denote the first term of the sequence 7,, 72, ..., such that the set f(X)
of all numbers of the set A that are smaller than r, is a non-empty proper subset of the
set X. The function f/(X) satisfies the required conditions.
Note. We are not able to define the correspondence assigning to every set of
functions of a real variable containing more than one function a certain non-empty proper
subset of that set.
2. Establish the law according to which to every infinite set of real numbers X
corresponds a certain decomposition X — X,+X,1+... of the set X into an infinite
series of disjoint non-empty sets.
Solution. Let f denote the function satisfying the conditions of exercise 1. If X
is an infinite set of real numbers, then the sets {(X) and X —f(X) are both non-empty,
and at least one of them is infinite. If the set f(X) is infinite, let g(X) = f(X), other-
wise let g(X) = X—f(X). Thus, for every infinite set X of real numbers the set g(X)
is an infinite proper subset of the set X. Denoting by g” the n-th iteration of the func-
tion, 9g, we put X,— 9g(X)—g¢(X) for w= 2,3,..., and X,=.%. =(X,7 4,4...)
‘obviously we shall have X—g(X)c X,, the sets XY, (n =1,2,...) will be disjoint
and X = X,+X,+... will be the required decomposition of the set X into an infinite
series, defined by X, of disjoint non-empty sets.
It will be observed that we are not able to prove without the aid of the axiom of
choice that every infinite set (of arbitrary elements) is the sum of an infinite series of
‘disjoint non-empty sets (Corollary 2 of VI.7).
3. Define the function assigning to every infinite series A,+A,+... of different
non-empty sets an infinite series B,+B,+... of disjoint non-empty sets such that
A,+A,+...= B,+ B,+...1).
Solution. Let A,, A,,... denote a given infinite sequence of different non-empty
sets. Let us denote by C the set of all real numbers x giving the triadic expansion

co

t= 2 oe Bae

where x, = 0 or 1 forn=1, 2,... (GC, as we know, will be the so-called nowhere dense
‘perfect set of Cantor). For a given number we C denote by P(x) the set of all those
elements » of the set A = A,+A,+... for which with any natural n we have pe A,
Woo eandep (Ait a 1102).
Obviously we shall have P(x)P(a#’)=0 for we 0, w «CC, x Au’. Let Z denote
the set of all non-empty sets P(x), where x « 0. Each of the sets A, is the sum of certain
sets belonging to Z, namely the sum of all those sets P(x) for which we have x, = 1 in
the triadic expansion of the number z.
If the set Z had only a finite number of different sets then the sets A,, A,, ... could
not all be different from one another, since from a finite number of sets we can form only
a finite number of different sums. Thus the set Z is infinite. Therefore the set X of

1) This problem has been solved by K. Kuratowski (see Tarski [2], p. 94— 95).
‘The solution given here is different from that found by Kuratowski (cf. also Sierpiiski
[43], p. 60-61).
2) Cf. the definition of the characteristic function of sequences of sets given by
Marczewski (see Szpilrajn [2], p. 211).
108 VI. Axiom of choice

those numbers we 0 for which P(x) ~ 0 is also infinite. According to Exercise 2 we

are able to decompose the set X into an infinite series of disjoint non-empty sets X =.
= X,+X,+... Let
Boy Pa) for n=1,2,...;
xEeXy

the sets B, (n= 1, 2,...) will be non-empty, disjoint, and we shall have

foo}

Uva SsRie B
xzeX n=1

In connection with Exercises 3 it will be observed that a set which is the sum of
a non-denumerable infinity of different non-empty sets need not be the sum of a non-
denumerable infinity of disjoint non-empty sets, e.g. the set of all natural numbers
which is the sum of a non-denumerable infinity of the power of the continuum of its
different non-empty subsets.
4. Prove that the axiom of choice is equivalent to the following
THEOREM T. If Z is an arbitrary set and f(p) an arbitrary function (of arbitrary
values), defined for p « Z, then there exists a set Z,c Z such that the function f is invertible
am the set Z, and f(Z,) =f (Z).
Proof. Let us suppose that the axiom of choice is true and let f denote a function
defined in the set Z. The sets f(q) for q«f(Z) form afamily F of disjoint non-empty
sets contained in Z. The axiom of choice implies the existence of a set Z,, containing
one and only one element from each set of the family #, and thus one and only one
element y(q) from each set {~1(q) for q « f(Z). Hence Z, is the set of all the elements 9(q)
where q«f(Z). Now if p,« Z,, p.<« Z,, and p, ~ pz, then it follows from the definition
of the set Z, that there exist elements q,<«f(Z) and q,<f(Z) such that p, = y(q,),
P2 = 9(qz2), Whence in virtue of p, = p,, we find q, ~ q,; according to the definition of
the function » it follows that o(q) «f—*(q:) and ~(q.) = f(q), whence g, = f(p(qm)) =
= f(p;) and g. = f(~(q2)) = f(p2), which in virtue q, # q, gives f(pi) # f(p2). There-.
fore we have f(p,) ~ f(p.) for pi « Z,, Po « Zz, Pi # P2; thus the function f is invertible
in the set Z,. From the definition of the set Z, it follows that Z, c Z (since y(q) « f-"q) c Z
for q«f(Z)). In order to prove that theorem T is true it remains to show that {(Z,) =
= f(Z), i. e. that for every element q<«f(Z) there exists an element p« Z, such that
f(p) = q. Obviously p = ¢(q) is such an element. Thus we have proved that the axiom
of choice implies theorem T.
Now let us suppose that theorem T is true and let # denote an arbitrary family
of disjoint non-empty sets. Let us denote by Z the sum of all sets of the family F. Since
the sets of the family F#' are disjoint, therefore there exists for every element p of the
set Z one and only one set of the family F that contains p; let us denote it by f(p). Thus
the function f(p) (whose values are sets e /) will be defined in the set Z. In virtue of
theorem T there exists a set Z,c Z such that the function f is invertible in the set Z,
and f(Z,) = f(Z). We shall show that the set Z, contains one and only one element
from each,set of the family 7. Let H denote any set of the family 7; thus we shall have
H « Z. Since the sets of the family F are non-empty, there exists an element p « H, and
it follows from the definition of the function f that f(p) = H. Thus, in view of pe Hc Z
we have H «f(Z) = f(Z,) and there exists an element q of the set Z, such that H = f(q),
which, in virtue of the definition of the function f, proves that q« #. Thus the set Z,
contains elements from each set of the family Ff. Finally, if the set Z, contained two
 § 6. Examples 109

different elements g, and qg, from a certain set EH of the family F, then we should have
q ¢€# and q,«H, whence, in virtue of the definition of the function f and in view of Ec F,
we find f(q:) = # and f(q.) = H, therefore f(q,) = f(q.), and m¢Z,, g@¢4,, G# oq,
contrary to the fact that the function f is invertible in the set Z,. Therefore the set Z,
contains one and only one element from each set of the family F. The axiom of choice
is true for the family / of sets. Thus we have proved that theorem T implies the truth
of the axiom of choice.
Therefore the axiom of choice and theorem T are equivalent.

7. Applications of the axiom of choice. Let us now pass to the applica-

tions of the axiom of choice. As early as 1902, 7. e. a few years before
Zermelo announced the axiom of choice, Beppo Levi [1] pointed out
that in the general case we are not able to prove that the sum S of
disjoint non-empty sets forming a set Z of sets is of greater power than,
or of the same power as, the set Z and that the proof can be given in
all those cases in which we are able to distinguish one element in each
of the sets forming the set Z.
Referring to the axiom of choice we can prove the general case as
follows. In virtue of the axiom of choice there exists a set B having one
and only one element from each of the sets A forming the set Z. Since
‘S= D'A,
AceZ

we have BC S. On the other hand, we have Z~B; in order to obtain

a (1-1) correspondence between the elements of the sets Z and B it. suf-
fices to associate with every set A « Z the only element of the set AB.
Therefore the set Z is equivalent to a certain subset B of the set S, whence,
in virtue of Theorem 2 of V.1, we find Z < S, q. e. d. Thus we have proved
with the aid of the axiom of choice
THEOREM 1. If we decompose any set A into disjoint non-empty sub-
sets, then the set of all those subsets is of power < than the power of the set A.
It will be observed that without the aid of the axiom of choice we
are unable to prove even that if we decompose a set into disjoint non-
-empty subsets the set of those subsets cannot be of greater power than
the decomposed set, however improbable this might seem.
In particular, let us decompose into disjoint subsets the set of all
real numbers, including two real numbers in the same subset if and
only if their difference is a rational number. In this way the set of all
real numbers will be decomposed into a non-denumerable infinity of
disjoint subsets congruent by translation. Now, without the aid of the
axiom of choice we are unable to prove that the set of those subsets will
not be of greater power than the continuum.
On the other hand, we are able to prove without the aid of the axiom
of choice, although such proof is difficult, that if we decompose in any
110 VI. Axiom of choice

way the set of all real numbers into disjoint pairs, then the set of those
pairs is of the power of the continuum (Sierpinski [53], p. 31).
The proof that this set is of power < the continuum presents no
difficulties. We are also able to prove without the aid of the axiom of
choice that if we decompose the set of all real numbers into disjoint,
non-empty finite subsets, then the set of those subsets is of power < the
continuum, but we cannot prove without the aid of the axiom of choice
that it is of the power of the continuum. .
EXERCISES. 1. Prove without the aid of the axiom of choice that if we de-
compose an arbitrary non-empty set A into disjoint non-empty subsets, then the set
of those subsets is of less power than the set U(A) of all subsets of the set A.
Proof!). Let us suppose that a non-empty set A has been decomposed into
disjoint non-empty sets; let Z denote the set of those subsets. Let us associate with
each subset T of the set Z the sum f(7) of all those subsets of the set A which are ele-
ments of the set 7. Since the (non-empty) sets forming the elements of the set Z are
disjoint, we obviously assign to different subsets of the set A different subsets of the
set Z, whence it immediately follows that U(Z) < U(A), and since, by Theorem 1
of V, we have Z < U(Z), it follows that Z < U(A), q. e. d.
2. Prove without the aid of the axiom of choice that if we decompose the set X
of all real numbers into disjoint non-empty finite subsets, then the set of those subsets.
is of power =e
Proof. The proof follows immediately from the observation that we are able to
define a function associating with every non-empty finite subset of the set X a certain
element of that subset, namely the least number belonging to it. This function maps
the set of all the components into which we have decomposed the set X upon a certain
subset of the set X. Therefore the set of those components is of power <X.
It will be observed, however, that we are unable to prove without the aid of the
axiom of choice that in the case in question the set of all the components is of the power
of the continuum (or that it is of power >X).
3. Prove without the aid of the axiom of choice that if we decompose the set of
all real numbers into a denumerable infinity of components, then at least one of them
must be an infinite set.
Proof. Let us suppose that the set X of all real numbers is the sum of an infinite
series of components X = X,+4X,+... and that each of the sets X, (k = 1, 2,...) is
finite. Therefore the real numbers forming the set X,, arranged according to magnitude,
give a finite sequence a, a, 5006 an (well defined by the number k). Thus we
should have
X= ie5) ) As,
@ il
), ae al, af2 ) Rane Ged,
(2
af3 i eet

and the set X would be denumerable, which, as we know from IV.2, is not true. Thus.
each of the sets X,, X,,... cannot be finite, therefore at least one of them is infinite,
(anerads
Thus we are able to prove without the aid of the axiom of choice that the set of all real
numbers is not the sum of a denumerable infinity of finite sets.

1) We owe this proof to A. Tarski.

 § 7. Applications of the axiom of choice £1

‘Beside the relations < and < for the power of sets, already intro-
duced by Cantor, A. Tarski has introduced the relations <+* and <x
presenting certain analogies to the relations < and < and defined as
follows (Tarski [5], p. 301).
For given sets A and B we write A <+ B if A is the empty set, or
if the set B can be decomposed into disjoint non-empty sets in such
a way that the set of those sets is equivalent to the set A.
With the aid of the axiom of choice it is easy to prove that the rela-
‘tions <* and < are identical. Without the aid of the axiom of choice
we can prove that if for two sets A and B we have A < B, then A <x B,
and that if A < « B, then U(A)< U(B) (which, as can easily be seen,
has been deduced in the proof of the last exercise 1). Cf. also VIII.4.
From Theorem 1 we can easily infer the following
COROLLARY 1. If f(a”) is a single-valued function defined in the set A
then f(A)y<aon.
Proof. For each element y of the set f(A) there exists at least one
element x of the set A such that f(x)= y. For y« f(A), let us denote
by g(y) the set of all elements wx of the set A such that f(x) = y. The sets
g(y), for yef(A), will be non-empty, and obviously we shall have
g(y)g(y’) = 0 for ye f(A), y’ e f(A), y #y’. It is also clear that we shall
have .
A= Sy 9(Y);5
ye f(A)
thus we have a decomposition of the set A into disjoint non-empty
subsets, the set of those subsets 7h equivalent to the set f(A). Thus
in virtue of Theorem 1 we have f(A) = qg. e. d.
Thus we have proved that the a of all values of a single-valued
function is not of greater power than the set of all its arguments.
It should be noted that without the aid of the axiom of choice we
are unable to prove, for every set A, that if f(x) is a single-valued func-
tion defined for w« A, then the set f(A) is not of greater power than
the set A (Sierpinski [31], p. 157).
Further, Theorem 1 has the following
COROLLARY 2. The set of all denumerable sets of real numbers is of
the power of the continuum.
Proof. From Theorem 1 of IV.11 it follows that the set S of all
infinite sequences of real numbers is of the power of the continuum.
The set 7 of all infinite sequences whose terms are different real numbers,
being a subset of the set S, is of power < the continuum. Now let us
 112 VI. Axiom of choice

divide all infinite sequences forming the set 7 into classes, including
two sequences in the same class if and only if they differ only in the
order of their terms. Let Z denote the set of the classes obtained in this
way. By Theorem 1 we shall have Z < 7, therefore the set Z is of power <
the continuum.
Now let M denote the set of all denumerable sets of real numbers.
To each set A belonging to M corresponds a certain class of sequences
of real numbers belonging to Z, namely the class of those sequences s
for which A is the set of all the real numbers which are terms of the se-
quence s. Clearly, to different sets belonging to M there will correspond
different classes belonging to 7. Hence the conclusion that M <Z,
therefore the set M is of power < the continuum.
On the other hand, we know from Exercise 1 of IV.8 that the set NV
of all increasing infinite sequences of natural numbers is of the
power of the continuum. Now let us denote by P the set of all de-
numerable sets of natural numbers. Obviously P~WN since every in-
creasing infinite sequence of natural numbers determines a certain denu-
merable set of natural numbers (which are the terms of the sequence in
question), and, conversely, every denumerable set of natural numbers
can be arranged in an increasing infinite sequence of natural numbers (by
ordering the elements of this set according to their increasing magnitudes).
Therefore the set P is of the power of the continuum. But, of course,
PC M, whence it follows that P < M, 7%. e. that the set M is of power >
the continuum. Since we have already proved that the set M is of power
< the continuum, we conclude that the set M is of the power of the con-
tinuum, q. e. d.
However, it will be observed that we are unable to prove that the
set M is effectively of the power of the continuum. Thus we have here
an example of a set which can be proved with the aid of the axiom of
choice to be of the power of the continuum but cannot be proved to be
effectively of the power of the continuum.
Thus with the aid of the axiom of choice we can prove that the
set of all denumerable subsets of a set of the power of the continuum is
of the power of the continuum (but we cannot prove that it is effectively
of the power of the continuum). But, without referring to the axiom
of choice we are able to prove that the set of all finite subsets of a set
‘ of the power of the continuum is of the power of the continuum.
We can even prove that the set of all finite sets of real numbers is
effectively of the power of the continuum.
Indeed, it is easy to establish a (1-1) correspondence between finite
sets of real numbers and increasing finite sequences of real numbers (by
arranging the numbers of the finite set in a sequence according to their
 § 7. Applications of the axiom of choice 113

increasing magnitudes). Thus it suffices to prove that the set S of all

increasing finite sequences of real numbers is effectively of the power
of the continuum. Now, in virtue of VI.11, Exercise 5, the set T of all
finite sequences of real numbers is effectively of the power of the conti-
nuum. Thus the set SC T is effectively equivalent to a subset of the set X
of all real numbers. But of course X is effectively equivalent to a subset
of the set S, namely to the subset composed of all one-term sequences.
In virtue of Theorem 1 of II.6, we conclude that S ef ~ X, q.e. d.
COROLLARY 3. Every set of points is of power > than the set of the
points of its projection.
Proof. In order to prove this it suffices to decompose the set in
question into subsets, including in one subset all those points whose
projections coincide, and then to apply Theorem 1.
As an application of Corollary 3 we shall prove
THEOREM 2. If we decompose a set of the power of the continuum
into two subsets, then at least one of them will be of the power of the
continuum.
Proof. Let P denote the set of all points in a plane. In virtue of
Theorem 2 of IV.8 the set P is of the power of the continuum. In order
to prove our theorem it obviously suffices to prove that there exists no
decomposition P = 4+B where A and B are sets of less power than the
continuum. Therefore let us suppose that such a decomposition exists.
The projection of the set A on the axis of abscissae, in virtue of Corol-
lary 3, is of power < the power of the set A, and therefore of less power
than the continuum. This projection does not fill the whole axis of abscis-
sae. Therefore there exists an abscissa x, such that the straight line x= a,
does not contain any point of the set A. Similarly we conclude that there
exists an ordinate y, such that the straight line y = y, does not contain
any point of the set B. Thus the point (a, y,) belongs neither to A nor
to B, contrary to the assumption ‘that P = A+B.
Thus we have proved Theorem 2.
We are unable to prove Theorem 2 without the aid of the axiom
of choice. But we are able to prove without the aid of that axiom that
if we decompose a set of the power of the continuum into two equivalent
disjoint sets (or into a finite number of equivalent disjoint sets), then
each of them is of the power of the continuum; the proof, however, is
difficult (Sierpinski [7], p. 1, and Sierpinski [33], p. 32).
Much more difficult than the proof of Theorem 2 would be the proof,
with the aid of the axiom of choice, of the theorem stating that the Carte-
sian product of two sets neither of which is of the power of the continuum
cannot be of the power of the continuum (XVI.2).
Cardinal and ordinal numbers : 8
114 VI. Axiom of choice

EXERCISE. Prove with the aid of the axiom of choice that if we [decompose
a set of the power of the continuum into a finite number of subsets, then at least one
of them will be of the power of the continuum.

Now we shall prove with the aid of the axiom of choice

THEOREM 3. Every infinite set contains a denumerable subset.
Proof. Let A denote an infinite set. As we have proved in II.2,
for every natural number n the set A contains a subset B, of n elements;
thus for every natural number m there exists also at least one sequence
of n different terms, which are elements of the set A. Thus from the
axiom of choice follows the existence of an infinite sequence of sequences
815 8) 83, +.» Where, for n=1,2,..., s, = (a{”, as”, ..., a”) is a sequence
of terms which are different elements of the set A.
Consider the infinite sequence af, a, a, a®, a®, aP, a®, ... ob-
tained by taking successively the terms of the sequences 38, 82, 83, ..-
In the sequence obtained we omit those terms which are equal to any
of the terms preceding them. In this way we shall obtain a new se-
quence b,, b,, b;, ... containing only different elements of the set A. This
sequence cannot be finite since it contains of course all terms of each
of the sequences s, (n= 1, 2,...), and the sequence s, contains different
terms. Therefore if the set b,, b,, ... contained only a finite number m of
terms, then we should have m > vn for n= 1, 2,..., which is impossible.
Therefore the set b,, b,,... is infinite.
Thus we have proved (with the aid of the axiom of choice) that if
the set A is infinite, then there exists an infinite sequence containing
only different elements of the set A. Thus we have proved Theorem 3.
From Theorem 3 it follows that every non-denumerable set con-
tains a denumerable subset. And since no denumerable set contains
a non-denumerable subset, we can immediately infer from Theorem 3
COROLLARY 1. Any non-denumerable set is of greater power than any
denumerable set.
From Theorem 3 we can also immediately infer the following
COROLLARY 2. In order that an infinite set be denumerable it is neces- =

sary and sufficient that it be equivalent to each of its infinite subsets.

The necessity of the condition of Corollary 2 can be proved without
the aid of the axiom of choice, but the sufficiency cannot.
Theorem 3 implies also another corollary, namely
COROLLARY 3. Every infinite set is the sum of an infinite series of
non-empty disjoint sets.
This Corollary we are also unable to prove without the aid of the
axiom of choice.
 § 7. Applications of the axiom of choice 115

Theorem 3 can of course be reversed: every set containing a denumer-

able subset is infinite (the proof of which can be given without referring
to the axiom of choice). Thus from the axiom of choice it follows that
a set is infinite if and only if it contains a denumerable subset. Therefore,
in view of III.8, Theorem 1, we can say that with the aid of the axiom
of choice we have proved
THEOREM 4. The two definitions of an infinite set, Dedekind’s and
the usual one, are equivalent.
From Theorem 3 and from Theorem 10 of III.7 it follows that every
infinite set is equivalent to a certain proper subset of itself, whence it
follows that a set which is not equivalent to any of its proper subsets is
finite. Even this conclusion we are not able to prove without the aid
of the axiom of choice. But without referring to this axiom we are able
to prove.
THEOREM 5. In order that the set A be finite it is necessary and suf-
ficient that the set U(U(A)) be equivalent to no proper subset of itself
(Tarski [2], p. 74).
Proof. The proof of the necessityof the condition of Theorem 5
follows from the observation that the finiteness of the set A implies the
finiteness of the sets U(A) and U(U(A)).
Now let us suppose that the set A is not finite. As we know from IT.2,
for every natural number m there exists in this case at least one n-ele-
ment subset of the set A. Therefore, if we denote by Z, the set of all n-ele-
ment subsets of the set A, then obviously Z,, Z,, Z;,... will be an ef-
fectively defined infinite sequence of different elements of the set U(U(A)),
. whence according to Theorem 9 of III.7 we conclude that the set U(U(A))
is effectively equivalent to a certain proper subset of itself. This proves
that the condition of Theorem 5 is sufficient.
On the other hand, we cannot prove without the aid of the axiom
of choice that every infinite set is the sum of two disjoint infinite sets
(Lindenbaum and Mostowski [1], p. 28).
The easy proof of this theorem with the aid of the axiom of choice
is left to the reader.
In connection with Theorem 3 it will be observed that without the
aid of the axiom of choice we are not able to prove that every infinite
set of real numbers contains a denumerable subset but, as follows from
Exercise 2 of VI.6, we are able to prove that every infinite set of real
numbers is the sum of an infinite series of disjoint non-empty sets.
Theorem 3 and Theorem 4 of III.7 immediately imply
THEOREM 6. An infinite set is equivalent to a set which will be obtained
by joining to it a finite or a denumerable set of elements.
8*
116 VI. Axiom of choice

Now let us suppose that the set A is non-denumerable and that the
set B is finite or denumerable. According to Corollary 3 of III.7, the
set A—B will be non-denumerable, and therefore, according to Theo-
rem 3, it will contain a denumerable subset. By Theorem 6 of III.7 we
shall have 4—B~A. Thus with the aid of the axiom of choice we are
able to prove
THEOREM 7. If the set A is non-denumerable and the set B is finite
or denumerable, then A—B~A.
We are not able to prove Theorems 6 and 7 without the aid of the
axiom of choice.
Now we shall prove with the aid of the axiom of choice
THEOREM 8. If
(7.1) Aig Ag nw Ot as Bnet.
are two infinite sequences of sets such that no two sets belonging to the same
sequence have common elements, and
(7.2) Ay ewey | Ol eas Lad. aces
then putting
(7.3) S=A,+A4,... -and T=B,+B,++...,
we shall have
(7.4) S~T.
In other words: If the corresponding components of two infinite series
of disjoint sets are equivalent, then the swms of those series are also equivalent.
Proof. Let n denote a given natural number. According to the as-
sumptions of our theorem we have A,~B,; thus there exists a (1-1)
correspondence between the elements of the sets A, and B,. Let us denote
by @, the set of all (1-1) correspondences between the elements of the
sets A, and B,; therefore the set ©, is non-empty. Moreover, it is clear
that for natural m and n, where m #1, the sets ©, and ®, have no ele-
ments in common.
In virtue of the axiom of choice for the infinite sequence of sets
@,,,,... there exists an infinite sequence

(7.5) is Pay
such’ that ¢, ¢O,; for w= 1,2) te euch thatvior v=) 2, -.. 48
a (1-1) correspondence between the elements of the sets A, and B,.
Now let s denote a given element of the set S. In view of (7.3) and
the observation that the sets A,, A,,... have no common elements, there
exists one and only one natural number n such that s is an element of
the set A,. Let y(s) = 9,(s); we shall have p(s) « B, and, in view of (7.3),
 § 7. Applications of the axiom of choice G17

gy(s)« T. It is easy to prove that the function g establishes a (1-1) cor-

respondence between the elements of the sets S and 7. Thus we have
(73); "Qe e.d.
In the same way as we have proved Theorem 8 we can prove the
more general .
THEOREM 9. If E is an arbitrary non-empty set and, for é« H, A(é)
and B(é) are sets such that A(é)A(y)= B(&)B(n)=0 for eH, ne #,
E#~n, and A(é)~B(E) for 颫 H, while

Se GE), ti Bl),
&€E nee
then S~T.
Now we shall prove
THEOREM 10. The sum of an infinite series of disjoint non-empty
sets, finite or denumerable, is a denumerable set.
Proof. Let A,(m= 1, 2,...) denote an infinite sequence of disjoint
non-empty sets, finite or denumerable, and let S= A,+A,+... Let n
denote a given natural number. The set A,, being finite or denumerable,
is equivalent to a certain subset B, of the set
1 ee aie rate i Gna (oll nae
In virtue of Theorem 8 the set S= A,+A,+... will be equivalent
to the set 7 = B,+B,+..., which is a certain set of natural numbers.
Hence it follows that the set S is finite or denumerable. But the set S =
= (,+C,+..., being the sum of an infinite series of disjoint non-empty
sets, is an infinite set. Thus it is a denumerable set, q. e. d.
It would also be easy to prove
THEOREM 10a. The sum of an infinite series of denumerable sets is
a denumerable set.
We are not able to prove Theorems 10 and 10a without the aid of
the axiom of choice. But here is what H. Lebesgue has written on the
subject (Lebesgue [3], p. 260):
»ll est vrai qu’en apparence on fait parfois des choix sans loi. Mais
cest, ou bien parce qu’il importe peu qu’on ait fait tel choix ou tel autre,
pourvu qu’on en ait fait un, et qu’il est évident qu’on pourrait définir
logiquement un choix, “par un nombre fini de mots“, c’est, par exemple,
le cas du texte; ou bien c’est parceque le choix est imposé. Supposons,
par exemple, qwil s’agisse de prouver que la somme d@’une infinité dénom-
brable d’ensembles dénombrables est un ensemble dénombrable. Soit
ai, a;,... le premier ensemble F,, soit a3, a3, ... le second E,,... On peut
ranger tous les éléments dans la suite dénombrable
1 2 il 2 1
Qi, A15 Az, Ag, Mg, «..
118 VI. Axiom of choice

Mais, objecte-t-on, vous avez ,énuméré“ H#, d’une fagon particuliére

et il y a la un choix que vous avez fait sans loi. Et comme vous avez
opéré de méme sur chaque #,, voici une infinité de choix faits sans loi.
En aucune facon. On m’a donné les ensembles H,, H,,... non pas un
& un, mais par une loi. De cette loi j’ai du tirer la preuve que chacun
deux était dénombrable, ce que je n’ai pu faire séparament pour chaque
ensemble, mais grace & un raisonnement qui me permis d’effectuer le
classement des éléments de chaque ensemble H; dans une suite Ui 5 Ui nae
bien déterminée. Done des suites doivent étre considérées commes don-
nés, elle ne résultent pas de choix.
Je repondrais de la méme maniére aux remarques faites a l’occasion
de la démonstration de cette propriété: la somme d’une infinité dénom-
brable d’ensembles mesurables est mesurable *).
Il me semble que ceux qui voient dans ces démonstrations V’emploi
de Vaxiome de Zermelo, donnent aux énoncés un sens ,,idéaliste“, alors
que je ne concois que le sense ,,empiriste“*. Je montrerais mieux la dif-
férence entre ces interpretations en disant que je ne comprends pas ce
que Von veut dire quand on parle d’une ensemble dénombrable non ef-
fectivement énumérable. Il y a deux mentalités, en présence; on per-
drait son temps en essayant de prouver que Vune d’elles est la bonne et
que autre n’est pas cohérente. Les recherches faites en portant de ’axiome
de Zermelo sont autrement importantes: si ces travaux conduisaient
a quelque découverte notable, la cause de V’axiome serait bien prés d’étre
gagnée“.
And several years later, at a conference in Zurich on the foundations
and methods of mathematical sciences, in the discussion which followed
the paper I had delivered on the subject of the axiom of choice and the
Continuum Hypothesis, H. Lebesgue said:
»Depuis les travaux de M. Sierpiriski et de Ecole polonaise une
révolution s’est produite. Un ceratin nombre de mathématiciens se sont
servi fructueusement de Vaxiome du choix; les choses n’en sont plus au
méme point“ ?).
Without the aid of the axiom of choice we are not even able to prove
that the sum of an infinite series of two-element disjoint sets is a denume-
rable set. B. Russell illustrates this by an anecdoticaljexample of a certain
millionaire who possessed a denumerable set of pairs of shoes: the shoes
for the left and the right foot being exactly alike, it was impossible
to prove that the set of the shoes in question was denumerable (Russell [2]).

1) H. Lebesgue alludes here to the remarks which I made in a note (Sierpinski

[3], p. 688) and in my paper [5], p. 127.
2) See Lebesgue [6], p. 139, also the enunciations of H. Lebesgue on the subject
of the axiom of choice in his paper [6], p. 116-119.
 § 7. Applications of the axiom of choice 119

We shall show that we are able to prove without referring to the axiom
of choice that for every denumerable set A of infinite sequences (of arbitrary,
not necessarily different terms) there exists an infinite sequence containing
each term of each of the sequences forming the set A.
Indeed, since the set A of infinite sequences is denumerable, there
exists an infinite sequence A,, A,,..., containing all the sequences form-
ing the set A. Thus, for every natural number n, A, is a certain infinite
sequence A, = (a, aS, af, ...). For natural m= 2"-'(2k—1), let am =
= a. Obviously the infinite sequence (a,, a2, ...) will contain each term
of each sequence of the set A.
Further, it will be observed that although we have proved without
the aid of the axiom of choice the existence, for every denumerable set 4
of infinite sequences, of a sequence containing each term of each of those
sequences, we are not able to define the function associating with each
denumerable set A of infinite sequences a certain infinite sequence (defined
by the set A), containing each term of each sequence of the set 4. The
sequence (a, a, ...), whose existence we have proved, depends on the
way of arranging all the sequences forming the denumerable set A of
sequences in an infinite sequence A,, Ag, ...
After A. Tarski, every set which is the sum of a denumerable infinity
of denumerable sets will be called the set D,, and every set which is the
sum of a denumerable infinity of sets D, — the set D,. Without the aid
of the axiom of choice we are not able to prove that the set of all real
numbers is not a set D,. But we are able to prove without the aid of the
axiom of choice (although it is by no means easy) that the set of all sets ©
of real numbers is not a set D, (see XV.2). A. Tarski has remarked that-
without the aid of the axiom of choice we are not able to prove the exist-
ence of even one infinite set which is not a set D,. In this connection
A. Tarski asks whether the axiom stating that the set of all real numbers
is a set D,, or the axiom stating that every infinite set is a set D,, would
be proved to be consistent with the usually accepted axioms of the Theory
of Sets (excluding the axiom of choice) (Tarski [2]).
THEOREM 11. The sum of an infinite series of sets of the power of
the continuum tis a set equivalent to the continuum.
Proof. Let C, (n=1, 2,...) denote an infinite sequence of sets of
the power of the continuum, and let S = C,+0C,+... Writing A, = (C,,
and for n= 2,3,..., A, = C,—(C,+0C,+...+C,-1) we shall have S=
= A,+A,+..., the sets A, (n=1,2,...) being disjoint, and A,C C,
foray abe
Let » denote a given natural number. According to Theorem 2a
of IV.4, the set C,, being of the power of the continuum, is equivalent
120 VI. Axiom of choice

to the set D, of all real numbers w such that n—1 <a <n. Thus, since
A, C C,, the set A, will be equivalent to a certain subset B, of the set D,.
According to Theorem 8 the set S= A,+A,+... will be equivalent to
the set 7’= B,+B,+.... But, since B,C D, for n= 1,2,..., we have
TCD,+D,+..., and the set D,+D,+... is obviously the set of all positive
real numbers and zero. Therefore, since S~Z, the set S is equivalent
to a certain set of real numbers, whence we conclude that the set S is
of power <the continuum. On the other hand, in view of S0C,, and
observing that the set C, is of the power of the continuum, we conclude
that the set S is of power > the continuum. Therefore the set S is of the
power of the continuum, which proves the truth of the theorem.
EXERCISE. Prove the following
THEOREM lla. If the set Z is of the power of the continuum and the sets O(&) for
E<Z are of the power of the continuum, then the set

S => Ole)
&€Z
as of the power of the continuum
Proof. The proof of this theorem is a modification of the proof of Theorem 9.
We must base ourselves on Theorem 1 of IV.6 and on the fact that, for € « Z, the set
C(é), being of the power of the continuum, is equivalent to the set of all pairs (é, x)
where x is a real number.

Some theorems on infinite series of sets are difficult to prove even

with the aid of the axiom of choice, e. g. the following two theorems
of A. Tarski:
1. If A,, Ag, ... is an infinite sequence of sets, while B and D are sets
such that B~DC A,+A,+..., then etther B~A,+A,+..., or there exist
a natural number n and a set C such that B~CC A, +A,4+...+A,.
2. If A,, Az,... 1s an infinite sequence of sets, and B a set such that
for each natural number n there exists a set C, such that A,+A.+...+
+A,~C,CB, then there exists a set D such that A,+A,+...~DCB.
A. Tarski has proved that the first of the above theorems is equi-
valent to the axiom of choice, and that the second follows from the re-
stricted axiom of choice (Tarski [11], p. 87, Theorem 1, and p. 84, Theo-
rem 2).
In IV.2 we have given the term Cartesian product of infinite se-
quences of sets A,, A,,... to the set of all infinite sequences ‘a,, do, ..
where a,¢ A, for n=1,2,... We denote this set by
A,X Asx AX 2}
Without the aid of the axiom of choice we are not able to prove
that the Cartesian product of an infinite sequence of non-empty sets is
@ non-empty set.
 § 7. Applications of the axiom of choice 11

THEOREM 12. If
(7.6) AccAaeusy ONO. obiissDo ose:
are two infinite sequences of sets such that
(7.7) Ae ige, OR. wis ly,
De veers
then
(7.8) Aa CA ASE. DCsBa oe

Proof. From (7.7) and from the axiom of choice it follows, as in the
proof of Theorem 8, that there exists an infinite sequence 9, go, ...,
where y, denotes a function establishing a (1-1) correspondence between
the elements of the sets A, and B,.
Now let a denote an element of the set A,x A,X ... Therefore we have
== Adio. Oop.) WHC. On.< Ap LOr ay = 1),.2.59.... Let

g(a) = (p(41), P22) )+++) Pn(An) »Pas

obviously g(a) will be an element of the set B,xB,~x...
The proof that the function » establishes a (1-1) correspondence
between the elements of the sets A,x A,X... and B,xB,x... presents
no difficulties. Thus with the aid of the axiom of choice we have proved
Theorem 12.
THEOREM 13 (of J. Konig). Jf

(7.9) Aap As, and, Bi, Be ee:

are two infinite sequences of sets, dnd ;

(7.10) MON Be for 1 = Ie 28

then
(7.11) gee Ase sa Der cee
Proof. In view of (7.10), for every natural number n there exists
a proper subset of the set B,, equivalent to the set A,. Thus it follows
from the axiom of choice that there exists an infinite sequence of sets
C,, C., ... such that
(7.12) Agi, GB Aa, ALO on = IDs.
Let
(7.13) Re
By Ca tor (ate Rae
the sets FR, will be non-empty, and, in virtue of the axiom od choice,
there exists an infinite sequence 7,,7,,... such that

(7.14) ieee LOR Me ites BD | ty

122 VI. Axiom of choice

For a given natural n, let us denote by fF, the set of all infinite se-
quences obtained from the sequence 7,, 72, ... by replacing its n-th term
by an arbitrary element of the set C,. In view of (7.12) we shall have of
course
(7.15) Di AG OU ai) eee.

the sets-7,, 7,,.... being. disjomt, and -7,,C BOB. xX ..4 100. 7,15 2
Let Aj= A, and A, = A,—(A,4+4,4+...+An_1) for n= 2,3,...:
the sets A, (n= 1,2,...) will be disjoint and A; CA, for = 1, 2,...,
and also A,+A,+...= Aj+A3+... On the other hand, in view of (7.15)
and A;,C A,, for every natural number n there exists a subset of the
set T,, equivalent to the set A;, whence, in virtue of the axiom of choice,
we conclude that there exists an infinite sequence of sets T7, (n = 1, 2, ...)
such that 7, 07T,~A, (for n=1,2,...). Since both the sets Aj, Ag, ...,
and the sets Tj, 73,... are disjoint, we have according to Theorem 8
Aj+A2+...~+7{+73+..., hence, in view of 4,+A,+...= Aj+Az+... and
eT ics, GE, Tae BOB, 2, wey concludes thate4so5 Asse
= Bios, cus
Thus in order to prove formula (7.11) it suffices to show that the
sets S= A,+A,+... and P= B,xB,x... cannot be equivalent.
Assume that there exists a (1-1) correspondence gy between the
elements of the sets S and P. In view of y(S) = P and S§ = A,+A,+...
we have P = (8S) = 9(A,)+9(A,)+... and, taking P, = 9(A,) for n=
=. 25.05 We: Shall navel As ~P asor Gnesi Oe ane

(7.16) Pees ep oe
Let us consider all those sequences of the set P which belong to P,,
and denote by W,, the set of all the (different) n-th terms of those se-
quences. ae
Let us divide all the sequences belonging to P, into classes of se-
quences having the same n-th term; the set of those classes will of course
be equivalent to the set W,, whence, in virtue of Theorem 1 we conclude
that W, <P,, which in view of P,~A, and (7 10) gives

a7) W,<B, or r= 125s

On the other hand, the set W, is a subset of the set B, since the n-th
term of each sequence belonging to P is an element of the set B,. Thus
the inequality (7.17) proves that W,, is a proper subset of the set B, and
therefore the sets

(7.18) U, = BeaWe a ee
 § 7. Applications of the axiom of choice 123

are non-empty. Thus it follows from the axiom of choice that there exists
an infinite sequence
(7.19) Uys Was ess

where w,« U, for ~=1,2,... In view of (7.18) we thus have u,¢ W,

for n=1,2,... In view of the definition of the set W, we conclude
that the sequence w,, u,,... differs in its n-th term from each sequence
of the set P,. Since this holds for » = 1,2,..., the sequence %,, Uo, ...
in view of (7.16), does not belong to P, which is impossible because, in view
Oitvae 0 ole? =, Zeenat, (ito). Wer lave’ U;, €.b, 10M, We 1, 2 iy
and therefore (w,, u.,...)« B,XB,X...= P. Thus with the aid. of the
axiom of choice we have proved Theorem 13.
In connection with Theorem 13 it will be observed that if A,, Ag, ..
and B,, B,,... are two infinite sequences of non-empty sets such that
Aa B Morn 19 2 then twelmay have As A, eb SeB xB
E.-g. it is so in the case where each of.the sets A, and B, (for n = 1, 2, ...)
consists of only one element and where the sets A, (n= 1, 2,...) are
disjoint, since then the set A,+A,+... is infinite and the set B,x B,x ...
consists of only one element. On the other hand, if we slightly modify
the first part of the proof of Theorem 13, we can easily prove with the
aid of the axiom of choice
THEOREM 13a. If A,, A,.,... and B,, B,,... are two infinite sequences
of sets such that A, < B, forn =1,2,..., and if the sets B, (n= 1, 2, ...)
are infinite, then A,+A,+... < B,XB,X...
Indeed, if we analyse the first part of the proof of Theorem 13, we
see that in order to prove the inequality A,+A,+...<B,xB,x... the
inequalities (7.10) are not necessary: it is sufficient to refer to the
fact that for every natural number v the set A, is equivalent to a certain
proper subset of the set B,, which follows immediately from the assump-
tion that A, < B, and that the set B, is infinite, and therefore, in virtue
of Theorem 4, equivalent to a certain proper subset of itself.
With the aid of the axiom of choice we can also prove a stronger
theorem than Theorem 13a, obtained from Theorem 18a by replacing
in it the condition that the sets B, (n = 1, 2, ...) are infinite by the condi-
tion that each of the sets B, (n= 1, 2,...), has more than one element.
Indeed, from this assumption and from the axiom of choice it follows
that there exist infinite sequences p,, p., and qg,, q2.,-.. Such that p, « B,,
Qn€ B, and p, ~ q, for n=1,2,... Now for natural n let us denote by
P, {the set formed of all infinite sequences (4%, %%,...), where 2 = px
for kn, %n<¢« By, In F Pn, and of the sequence (7, #,...) where 7% = qx
for k AN, %,= py. Clearly we shall have B,~P, for n= 1, 2,... It is
124 V1. Axiom of choice

also obvious that the sets P,, P.,... are disjoint and contained in the
set BX Bix... Let Ay='A,, and A,= Aj,= Aig for’ 7". Le
sets A, (n= 1, 2,...) will be disjoint and we shall have A,+A,+...=
= Aj+Aj+... Further, for natural n, A; C A, and in view of a < B, =
= Jee there exists a set P/ C P, such that Aj;~P/,. In virtue of Theorem 8
we have A,+A,.+...=eAitAst.. ~Pit+P34+..CP,+P,4+...C Bx
x B,X ..., Whence ae BCBG desl.
EXERCISE. Prove that if A,, A,, ... is an infinite sequence of sets of ever greater
powers and A, ~ 0, then A,+A,+...< A,XA,X...
Proof. Let B, = A,,, for n = 1, 2,... In view of the assumption that Je < Ans
for n = 1, 2, ..., we shall have the itequalities (7.10), whence, in virtue of Theorem 13
Ae eee a a SNS ae
But the set A, not pene empty, the set A,x A; x... is obviously equivalent to
a certain subset of the set A, x A, xA,X ... Therefore we have A, x A,X... < A,X AgX «-
and thus A,+A,+...<A,x A,X ..., q.e.-d.

It will be observed that by generalizing the proof of Theorem 13

we could prove with the aid of the axiom of choice the much more
general
THEOREM 14. If Z is a non- empty set and to each element &« Z are
assigned two sets Ag and Bz such that Ay = Bs for Ee Z, then, writing

s=
>) A,
&€Z

and denoting by P the set of all functions f(&) defined for &« Z and such
that f(&) « Bg for EZ, we have 8S <P (Lermelo [2], p. 277) 2).
It will be observed that Theorem 14 immediately implies Theorem 1
of V.5. In order to show this we assume in Theorem 14 that A; = {&}
and B.= {0,1}. We shall then have S = Z, and P will be the set of all
functions defined for x « Z and assuming only the values 0 or 1. The set P
is equivalent to the set T of all subsets of the set Z, since in order to
establish a (1-1) correspondence between the elements of sets 7 and P it
suffices to associate with each subset of the set Z its characteristic func-
tion (1.16, Example 5).
Thus, in virtue of Theorem 14, we have Z < T, which gives Theo-
PELL OL INEO:
Now we shall prove with the aid of the axiom of choice

1) From Theorem 14 (for finite sets Z) it follows that Theorem 13 is also true for
finite sequences of sets. But also in this case we are not able to prove it without the
aid of the axiom of choice.
 § 7. Applications of the axiom of choice 125

THEOREM 15. If we decompose a set of the power of the continuum into

an infinite series of subsets, then at least one of them will be of the power of
the continuum.
Proof. It is of course sufficient to prove our theorem for the set X
of all real numbers. Let us suppose that X = X,+X,+... where none
of the sets X, (n=1,2,...) is of the power of the continuum. Since
eX tory = 1,0. we have Y= x tor. 1 = 1,2,..., whence in
virtue of Theorem 13
(7.20) DED
ie ee Ce
But X,4+X,+...= X, and, as we have proved in IV.12, the set
AXXX... is effectively of the power of the continuum, therefore
XxX XX...= X. Thus the inequality (7.20) gives X <_X, which is impos-
sible. Therefore at least one of the sets X, (n= 1, 2,...) must be of the
power of the continuum, which proves the truth of Theorem 20.
CoROLLARY. If we decompose a set of the power of the continuum into
a finite number of subsets, then at least one of them will be of the power of
the continuum.
Proof. Let us suppose that X = X,+X,+...+X,. Thus we also
have X = X,+X,4+...4+X,+2X,+42X,+... and, in virtue of Theorem 15,
we conclude that at least one of the components of the right side of
this formula, 7. e. at. least one of the sets X,, X,,..., X,, is of the power
of the continuum. Thus we have proved our corollary. Its particular
case is Theorem 2, proved on different lines.
EXERCISES. 1. Prove the following theorem of P. Hrdés:
If A is a plane set such that every parallel to the axis of abscissae intersects the set A
at a finite number of points, then there exists a parallel to the axis of ordinates which inter-
sects the set CA (complement of the set A to the plane) in ¢ points.
Proof. Let us suppose that each of the straight lines « = n (where n = 1, 2,...)
intersects the set CA in less than c points. Let us denote by Q the set of all real num-
bers y such that (nm, y) «CA. Therefore we should have Q, < ¢ for n=1,2,... Thus
it follows from Theorem 15 that the set X of all real numbers cannot be the sum of the
sets Q, (n= 1, 2,...). Thus we have X—(Q,+@Q,+...) 4 0 and there exists a number
be X—(Q,+Q.+...). Hence it follows that b<Q, for n= 1, 2,..., therefore (n, b)<« A
for n = 1, 2,..., which is impossible, since. in virtue of the property of the set A the
straight line y = b intersects the set A at a finite number of points. Thus there exists
a natural number n such that the straight line « = n intersects the set CA at c points,
which proves the truth of Erdés’ theorem.
2. Prove without the aid of the axiom of choice that Theorem 15 is equivalent
to the following
THEorEM T. If f(x) is a function of a real variable whose values are natural numbers,
then there exists a set A of real numbers, of the power of the continwwm such that f(a,) =
== (a3) for m.<« A, 7, <A.
126 . VI. Axiom of choice

Proof. Theorem T follows from Theorem 15. For if f(x) is a function of a real
variable having natural values and if we denote by X the set of all real numbers and
by X,, for n = 1, 2,..., the set of all those real numbers a for which f(x) = n, then we
shall have X = X,+ X,+... Thus according to Theorem 15 there exists such a natural
number m that the set X,, is of the power of the continuum. But, according to the de-
finition of the set X,,, we have f(x) = m for « « X,,, therefore f(x,) = f(a.) for x, « X,,,
m?

Lz X,,. Therefore Theorem T is true.

Now we shall show that Theorem 15 follows from Theorem T.
Indeed, let us suppose that Theorem T is true and let X = X,+X,+... denote
a decomposition of the set X of real numbers into an infinite series of disjomt subsets.
For n = 1,2,..., let f(x) =n for x « X,. In virtue of Theorem T there exists a set A
of the power of the continuum such that f(x) = f(x.) for v,<«A, x, «A. Let « denote
any element of the set A. From the definition of the function / it follows that f(a) is.
a natural number; we denote it by m. Thus f(x,) = m, whence in view of the definition
of the function f we find «, « X,,. But if « « A, then in view of x, « A and the properties.
of the set A, we have f(x) =f(a,), 2. e. f(%)=m, whence x«X,,. Thus we have
xe X,, for x <A, therefore Ac X,,c X, and since the sets A and X are of the power
of the continuum, the set X,, is also of the power of the continuum. Hence follows.
Theorem 15.

8. The m-to-n correspondence. We say that a (m-n) correspondence

exists between the elements of the sets A and B (where m and n are
natural numbers), if to each element of the set A corresponds a certain
set K (a) formed of m elements of the set B in such a way that for each
element 6 « B there exist exactly n different elements a of the set A for
which be K(a).
If an (m-n) correspondence (where n is a natural number) exists, then
it is easy to prove that the sets A and B have the same number of ele-
ments. Indeed, to each element a of the set A corresponds the set K (a)
formed of n elements of the set B, and for each element b « B there exist
exactly n different elements a of the set A such that be K(a). Let us
denote by S the set of all pairs (a, b), such that ae A and be K(a). If
the set A has k elements, and the set B has 1 elements, then the set S
can be divided (according to the elements a) into k disjoint sets each of
which has n elements, and also (according to the elements 6) into J disjoint
sets each of which has n elements. Thus we have kn = lm, whence k = l,.
which proves that A~B.
On the other hand, we are not able to prove without the aid of the
axiom of choice the proposition that if there exists ta two-two cor-
respondence between the elements of two arbitrary sets A and B, then.
A~B.
We can even give an example of two sets A and B between the
elements of which we are able to establish a (2-2) correspondence, but.
for which we cannot prove without the aid of the axiom of choice that:
A-~B (Sierpinski [40], p. 39-40).
 § 8. The m-to-n correspondence 127

With the aid of the axiom of choice D. Konig (the son of J. Kénig)
has proved the following theorem (Konig [4], p. 114):
If there exists between the sets A and B a (m-n) correspondence f (where n
is a natural number >2), then there exists between the elements of the sets A
and Ba(1-1) correspondence g such that, for each element of the set A, g(a) ws
one of the n elements associated with the element a in the correspondence f.
The proof of this theorem, even for n = 2, is not easy; for n = 3 it
is difficult, and even for finite sets it is not trivial.
D. Konig’s theorem is obviously equivalent to the following theorem
on the Cartesian product of sets:
If AXB ts the Cartesian product of two sets and P is a set, contained
in tt, which is. intersected by every parallel to the axis of abscissae or to the
axis of ordinates at n points (where n is a given natural number >1),
then there exists a subset P, of the set P such that every parallel to the
axis of abscissae or to the axis of ordinates intersects the set P, exactly at
one point.
Using the axiom of choice we shall prove this theorem for n = 2.
Let p = (a, b)« P. It follows from the properties of the set P that
the parallel x = a (or y = b) meets the set P exactly at two points, p and q;
let us denote the point q by g(p) (or y(p)). Therefore we shall have
gp) = yw(p) = p for pe P. Every point of the set P determines a se-
quence, infinite in both directions, of points which also belong to P but
are not necessarily different:

+39 YOY (DP), PY(P), V(P)s Ps P(P), VO(P)s PEP (P)s v5

We denote this set by Q(p)+). Obviously such sequences, corresponding
to two different points of the set P, have no common points or are a trans-
lation of one another, or finally are a translation with reversed order of terms..
In any case if p« P and qg« P, then either Q(p) = Q(q), or Q(p)Q(q) = 0.
Thus we can divide all elements p of the set P into disjoint classes, as-
signing two elements p and q of the set P to the same class if and only if
Q(p) = Q(q). In virtue of the axiom of choice there exists a set N contain-
ing one and only one element from each of those classes.
Let
P= N+y9(N)
+9y(N) +yeye (NY) + pygy(N) +5
1. €.

P,= N+ > (ypy"(N)+ > (py).

n=1 n=1

1) The set Q(p) might be defined as the product of all sets Q to which the element p
belongs and such that if qg«@, then also y(q) «<Q and y(q) <Q.
 128 VI. Axiom of choice

It can easily be proved that the set P satisfies the required conditions.
Thus, using the axiom of choice, we have proved the theorem of
D. Konig for n = 2. It will be observed that in the case of n = 2 K6nig’s
theorem could be proved by using proposition [2] instead of the axiom
of choice, but the proof would be somewhat longer than the one given
here (Sierpinski [40], p. 42-44).
As a corollary to D. K6énig’s theorem, proved (with the aid of the
axiom of choice) for » = 2, we shall now prove
THEOREM 1. If M,N, P,Q are sets such that MN = PQ=0, M~N,
P~Q, and M+N~P+Q, then M~P.
Proof. In view of M~WN there exists a (1-1) correspondence
between the elements of the sets M and N, and in view of P~Q there
exists a (1-1) correspondence py between the elements of the-sets P and Q;
finally, in view of M+N~P+49Q there exists a (1-1) correspondence 0
between the elements of the sets M+N and P+4Q.
; For ae M we shall distinguish four cases:
If d(a)e P and dp(a)« P, let K(a) = {8(a), dy(a)}.
If #(a)« P and Yp(a) <Q, let K(a) = {8(a), yp 0p (a)}.
If &(a)«Q and bp(a)e P, let K(a) = {y Ha), B—(a)}.
If B(a)eQ and dy(a)eQ, let K(a) = ty 8(a), p“(a)}.
It is easy to verify that in each of these four cases the set K(a) is
formed of two different elements of the set P, and that for each element D
of the set P there exist exactly two different elements of the set M such
that b« K(a). Thus between the elements of the sets M and P there
exists a (2-2) correspondence. In virtue of D. Koénig’s theorem for n = 2,
we conclude that M~P, q.e.d. Thus we have proved Theorem 1.
It will be observed that Theorem 1 can be proved without resorting
to the axiom of choice, but such the proof is much longer and more dif-
ficult than the one given here. It is even possible to prove that if we are
given the functions y, y, 3, establishing a (1-1) correspondence between
the sets M and V, P and@d, M+WN and P+Q, then we are able to define
a (1-1) correspondence between the sets M and P (Sierpinski [8], p. 1).
Theorem 1 is a particular case of a more general theorem of F. Bern-
stein (see Bernstein [1], p. 122, Hobson [1], p. 159-162, Konig [2], p. 462):
If n is a natural number and M; and P; («=1, 2,...,) are sets
such that, for 1<i<k<n, we have M;M,;=P,;P,=0, M;~M;,,
Pi~P,, while M,+M,+...4+M,~P,4+P2,+...+P,, then My~P,.
This theorem can easily be deduced from D. K6nig’s theorem (see
Konig [31]) but its first proof without the aid of the axiom of choice
was given by A. Tarski [12] (p. 77-92) as late as 1949. The proof in ques-
tion is long and difficult.
 § 8. The m-to-n correspondence 129

Further it will be observed that the following theorem can also be

proved, without the aid of the axiom of choice:
If M,N,P,Q are four disjoint sets, and we are given the mapping
of the set M on N, the mapping y of the set P on Q and the mapping 8 of
the set M+N on a certain subset of the set P+Q, then we are able to define
the mapping f of the set M on a certain subset of the set P (Sierpinski [31],
p. 148).
We can also prove without the aid of the axiom of choice that:
If n is.a natural number, and M; and P; ((=1, 2,...,) are sets such
that for l<i<k<n we have M;M,=P;P,=0, Mi~My,, Pi~Px,
and M,+M,+...4+0, <P, +P,4+...4P,, then M, <P.
According to a note of A. Lindenbaum and A. Tarski [1], p. 308
(published in 1926) A. Tarski proved this theorem for n = 2 without
the aid of the axiom of choice as early as 1924, and A. Lindenbaum did
the same in 1926 for arbitrary natural n. Those proofs have not appeared
in print. Another proof was published by A. Tarski in 1949 (Tarski [12],
~- 17,8 -neorem If):
9. Dependent choices. Suppose that in order to prove that the in-
finite set A contains an infinite sequence of different elements we reason
as follows. Since the set A is not finite, there exists an element a, of that
set. Since the set A is infinite, there. exists an element a, of that set,
different from a,. Similarly we conclude that there exists an element a,
of the set A, different from a, and from a, etc. In this argument we make
infinitely many choices, each choice depending on those previously made,
since element a, is not chosen arbitrarily from the set A but depends
on the already made choice of element a,, from which element a, must
be different. Similarly the choice of element a, depends on the already
made choices of elements a, and a,, etc. Now, in cases of this kind we
speak of making dependent choices.
As regards the example given here, we have already seen above
how to conduct the proof of the corresponding theorem, namely The-
orem 3, of VI.7 using the axiom of choice. Arguments of this kind can
often be replaced by others, based on the axiom of choice; that, however,
is not always an easy matter, even if we deal only with a denumerable
infinity of dependent choices. Suppose, for instance, that we want to
prove the following theorem.
If Z is a non-empty set of sets such that for every set A « Z there
exists a set Be Z such that A ~ BC A, then there exists an infinite se-
quence A,, A,,... such that A,<«Z and A, #4 Aj4i:C A, for n=1, 2,...
Making use of dependent choices we could reason as follows: since
the set Z of sets is non-empty, we can choose from it a set A, « Z. In
Cardinal and ordinal numbers 9
130 VI. Axiom of choice

virtue of the assumption and in view of A, « Z, there exist sets belonging

to Z and forming proper subsets of the set A,: let us choose one of them,
A,. Now let us select a set A, ¢« Z such that A, 4 A,C A,, etc. A denumer-
able infinity of dependent choices brings us to the infinite sequence
A,, A,,..., satisfying the required conditions.
This theorem can be deduced from the axiom of choice in the follow-
ing way.
For A « Z, let us denote by F(A) the set of all sets belonging to Z
and forming proper subsets of the set A. In virtue of our assumption,
the sets #(A) are non-empty for A« Z. From the general principle of
choice it follows that there exists a function f(A), defined for A « Z and
such that f(A) « F(A), 2. e. that A 4 f(A) C A for A « Z. Now let A denote
an arbitrary set belonging to Z; such a set exists because we have assumed
that the set Z of sets is non-empty.
Clearly the infinite sequence of sets

A, f(A), ff(A), fff(A), -.

satisfies the required conditions.
In our proof the general principle of choice has been applied with
respect to the set of sets #(A) where A « Z, that is with respect to a set
of sets which is equivalent to the set Z. Therefore, if Z is a non-denumer-
able set, we shall have to apply the general principle of choice to a non-
denumerable infinity of sets, although the infinity of choices is only denu-
merable. We are not able to prove our theorem (in the general case) if we
apply the axiom of choice only to a denumerable infinity of sets. (In
particular cases of our theorem we can sometimes even do without the
axiom of choice, e. g. where Z is the set of the power of continuum of
all infinite sets of natural numbers; the sequence in question will be
a sequence of sets A, = {n,n+1,n+2,...} for n= 1, 2,...).
We have an analogous situation as regards the proof of O. Tetch-
miiller’s principle, which reads:
If with every element a of a non-empty set A and with every natural
number v is associated a certain non-empty subset B,(a) of the set A,
then there exists an infinite sequence a,, a, .... such that a,41 € B,(a,)
for n=1, 2,... (Teichmuller [1], p. 568).
Making use of dependent choices we can choose an element a, from
the set A, next a, from the set B,(a,), then a, from B,(a,), etc., generally —
Qn41 from B,(a,) for n=1,2,... Thus we have here a denumerable in-
finity of dependent choices.
Teichmiller’s principle can be deduced from the general principle
of choice in the following way. Since the sets B,(a) are non-empty for
ae A and n=1,2,..., there exists, according to the general principle
 § 9. Dependent choices it

of choice, a function f(a,n), defined for ae A and n=1,2,..., such

that f(a,”)e«B,(a) for ae A and n=1,2,... Since the set A is non-
empty, there exists an element a, of this set. Now let us define the infinite
sequence 4@,, @,.... by induction, taking a,,, = f(a,,") for n= 1, 2,...
Obviously this sequence will satisfy the required conditions.
A. Tarski has formulated the principle of dependent choices in the
following way:
If A is a non-empty set and S a relation such that for every element
ae A there is an element y« A with «Sy, then there exists an infinite se-
quence of elements’ x,, 0, ...) in... n A such that’'c,S8¢,+1 forn= 0,1, 2,...
(Tarski [11], p. 96).
As Tarski remarks further, it can easily be shown that this principle
implies the axiom of choice for a denumerable set of sets, “we see, howe-
ver, no way of establishing the implication in the opposite direction.
On the other hand, A. Mostowski has shown that the principle in question
does not imply the general axiom of choice”? (Mostowski [5], p. 127-130).

g*
 CHAPTER VII

CARDINAL NUMBERS AND OPERATIONS ON THEM

1. Cardinal numbers. Let us divide the sets we have been consider-

ing into classes, assigning two sets to the same class if and only if they
are equivalent. Since the relation of equivalence (equality of power)
is, a8 we know from II.3, reflexive, symmetrical and transitive, the sets
we have been considering are divided into disjoint classes. The symbols
serving to denote those classes are called cardinal numbers 1). Cardinal
numbers relating to the same class are regarded as equal. In systems of
mathematical logic (Russell and Whitehead [1], p. 4 and foll.) cardinal
numbers and, in particular, natural numbers are usually regarded as classes
of all sets equivalent to (of the same power as) a certain given set; thus
for instance number 1 is defined as the class of all those sets which have
exactly one element 2). The concept of the class of all sets having a given
property (e. g. consisting of only one element) is attacked by some authors
(just as is the set of all objects). In order to avoid this difficulty, some
authors by the cardinal number or the power of the set A understand
the set A and every set equivalent to it.
Another way of avoiding the difficulty in question is to consider
not all possible sets, those which have been defined and those which will
be defined at any future time, but only a certain family F of sets,
and to divide into classes of equivalent sets (classes of the same power)
only the sets forming the family F (Mostowski [4], p. 171) °). In that case,
however, the concept of cardinal number becomes dependent on a fixed
family # of sets.
Note, however, that in one of his paper A. Tarski introduces cardinal
numbers by means of the following two axioms:

1) Lebesgue considers as meaningless the usual injunction that a number should

not be confused with the symbol which denotes it. See Lebesgue [5], p. 7.
) See Tarski [8], p. 84, footnote 76. In his book [13] Tarski writes on p. 237: “By
the cardinal number or the power of a set A we understand the class of all sets X with
A~X”’.
3) As a family F we can take the family of all subsets of a given set W (cf. Tarski
{11], p. 84).
 § 1. Cardinal numbers 133

1. To every set corresponds an object which is its cardinal number.

2. In order that any two sets be equivalent (of the same power)
it is necessary and sufficient that the same cardinal number
corresponds to them (Tarski [1], p. 147).
If M is a given set and m the cardinal number serving to denote
the class of equivalent sets to which M belongs, then we shall say that
to the set M corresponds the cardinal number m, or that the set M is of po-
wer m, and we shall write M—m and m= card M.
Thus equivalent sets (sets of the same power) have the same cardinal
number, and vice versa.
Since for finite sets the concept of equivalent sets is synonymous
with the concept of sets with equal number of elements, it is simplest
and most convenient to adopt natural numbers as the respective symbols
of cardinal numbers, and thus for a finite set of n elements — the num-
ber n itself. In this way natural numbers will be a particular case of cardinal
numbers. As the cardinal number of the empty set we adopt the number 0,
The cardinal number corresponding to denumerable sets is denoted,
after G. Cantor, by the symbol s, (aleph-zero), and the cardinal number
corresponding to sets of the power of the continuum is denoted by the
Gothic letter c.

2. Sum of cardinal numbers. Let m and n be two given cardinal num-

bers. In virtue of the definition of cardinal numbers (VII.1), m is a symbol
corresponding to a certain class of equivalent sets. Let M denote any
set of that class; thus we shall have M =m. Similarly, there exists a set V
such that VY =n. We can assume here that the sets M and N have no
common elements. For if they had, we could take instead of M the equi-
valent set M’ of all ordered pairs (m,1) where me M, and instead of
the set N — the equivalent set of all ordered pairs (n, 2) where ne NV.
Therefore, if we are given two cardinal numbers m and n, then there
exist two disjoint sets, MZ and WN, such that

(1.1) ar = isle seats

Let
(1.2) S == MAN.

Now let M, and N, be any two disjoint sets such that

(138) M,=m, N,=n,

and let

(1.4) S, = My+N 45
134 VII. Cardinal numbers

In view of (1.1) and (1.3) we have M~M,, N~N,, hence, in view

of (1.2) and (1.4) and observing that MN = M,N,=0, we conclude
by Theorem 1a of II.V that S~S,.
Thus to the sets S and S, corresponds the same cardinal number:
let us denote it by s.
Therefore we can say that to every two cardinal numbers m and n
corresponds a well-defined cardinal number $ such that if M and N
are two arbitrary disjoint sets satisfying the conditions M=mand N =n,
then M+N =s,
Obviously, if the cardinal numbers m and n are finite, then the
number s is their sum, Thus, for every two cardinal numbers m and n,
it will be natural to call the cardinal number s, obtained in the above
way, the sum of those two numbers, writing
$=mn.

Thus every two cardinal numbers m and n have a well-defined sum

(which is a cardinal number dependent only on m and n). Respectively
equal cardinal numbers have of course equal sums.
Since the sum of two sets does not depend on the order of the compo-
nents, it immediately follows that the sum of two cardinal numbers is
commutative:
m+n=n+m.
From the definition of the sum of two cardinal numbers it immediately
follows that if M and MN are any two disjoint sets, then
M+N= M+N.
From the associativity of the sum of sets immediately follows the
formula
(m-+1n)+p = m-+(n-+-p)
for any cardinal numbers m, n, p. Thus the sum of cardinal numbers is
associative.
The concept of the sum of cardinal numbers can be generalized by
induction to an arbitrary finite number of components; obviously such
a isum depends neither on the order of the components nor on the
manner of associating them.
EXERCISES. 1. Prove that if § is a set such that S = m-+n where m and n are
two cardinal numbers, then there exist sets M and N such that S= M+N, MN =0,
= ,N=n.
me Since m and n are given cardinal numbers, there exist as we know, sets
M, and N, such that M,N, = 0, MU, =m, Ne n.
Let S, = M,+N,; we shall have Bi, =m-+n — § therefore S,~S. In the (1-1)
correspondence between the elements of the sets S, and S, to the disjoint subsets M,
 § 2. Sum of cardinal numbers 135

‘and N,, whose sum is the-set S,, will correspond the disjoint subsets M and WN, whose
sum is the set S. Thus we shall have M~M,, N~N,, whence. M= M,== ih N= V,= n,
and S= M+N, MN = 0, Geresd's
We shall now give various examples of sums of cardinal numbers.
From the definition of the sum of cardinal numbers and from Theorem 13 of III.7
immediately follow the formulae
Notn=ys, £4xfor natural n
and
Not No = No ?

while from Theorem 4 of IV. 3 follow the formulae

ctn=c for natural n ,

and
Cc+No =;
finally, from Theorem la of IV.5 follows the formula
cte=c.
From the Corollary to Theorem 15 of VI.7 (proved with the aid of the axiom of
choice) and from Exercise 1 it easily follows that if ¢ is the sum of a finite number of
eardinal numbers,
C= Mm, Me--<-.-- mM, ,

then at least one of the cardinal numbers m,, m,, ..., m, is equal to c. We are not able
to prove this theorem without the aid of the axiom of choice (even for n = 2).
2. Prove without the aid of the axiom of choice that if m is a cardinal number
such that m+, =c, then m=c.
Proof. In view of m+, = c¢ and in virtue of Exercise 1, if X denotes the set of
all real numbers, then there exist sets M and WV such that ees m, Ne n,' UN =0
and X = M+ N, whence M = X—N. Since the set NV is denumerable, it follows from
the formula obtained and from Theorem 3 of IV.3 that Mu = Giese ti Cogn Onl
3. Prove that if m,n,p,q are cardinal numbers such that m+n=p-+q, then
there exist cardinal numbers ),, p.,4,,4., such that p=p,t+p.,, q=GQatq, m=
=Pitha, N= Pet q.-
Proof. Let M and N be two disjoint sets such that M =m, N =n; therefore
we shall have M+N —m-+n. In view of m+n=p-+q and in virtue of Exercise 1,
there exist two disjoint sets P and QY such that P=p, Q =g and M+N=P+9Q.
Hence we have M= MP+MQ, N= NP+NQ, P= MP+NP, Q= MQ+NQ, these
decompositions being disjoint. Taking MP = Pi> NP = p2, WQ =a, NQ =a, We
obviously obtain the cardinal numbers satisfying the required conditions.
3. Product of two cardinal numbers. Let m and n be two cardinal
numbers. Thus there exist sets, MZ and N such that M = m, N=n.
Let P= MXN and p= P. The cardinal number p does not depend on
the choice of the sets M and WN provided only that M=m and V=nt.
Indeed, let us suppose that also the sets M’ and N’ satisfy the condi-
tions I’ =m, N’=n. Therefore we shall have M’~M and N’~N.
But in virtue of Theorem 4 of II.5 these formulae imply the formula
M'xN'’~MxXN. Taking p’= M’XxWN’, we shall have p’=p, q.e. d.
 136 VII. Cardinal numbers

In the case of finite cardinal numbers m and n the number p will

obviously be their product. Number p will be called the product of the
cardinal numbers m and n also for arbitrary m and n, and we shall write
Dis Witt
In this way every two cardinal numbers give a product, well-defined
by them, which is a cardinal number.
Respectively equal factors give of course equal products. Thus, in
particular, for cardinal numbers the equality m= n implies the equality
mim = nn; however, we cannot prove without the help of the axiom of
’ choice that the inverse is also true.
From the definition of the product of cardinal numbers it immediately
follows that we have m-1 = m for every cardinal number tm.
As we know (see I[.17), if M and WN are arbitrary sets, then.
MxN~N x M; this immediately implies the commutativity of the product
of two cardinal numbers, 7. e. the formula

mt = NM

for any cardinal numbers m and n.

We also know (see ibid.) that if A, B, OC are arbitrary sets, then
(AxB)xXC~Ax(BXC); this immediately implies the associativity of
the product of cardinal numbers, 7. e. the formula

(1m, M,) M3 = m,(1M, M1)

for any cardinal numbers m,, m, and m3.

Now we shall prove that the law of distributivity of multiplication
with respect to addition, 7. e. the formula

(3.1) U(nt+n) = Im+In

holds for any cardinal numbers I, m and n.

Let L, M, N be sets such that L=1, M=m, N=n and MV=0.
In virtue of the definition of the sum of cardinal numbers, we shall have
M+N = m+n; therefore, according to the definition of the product of
cardinal numbers, I(m-+n)= (M+).
2X But (see 1.17, Exercise 1) we
have LX (M+WN) = (Lx M)+(Lx
WN); according to the definition of the
product of cardinal numbers, Lx M=Im, and Lx N=In, whence
(owing to the fact that in view of MN =0 we have (Lx M)(LxXN)=
= 0, and according to the definition of the sum of cardinal numbers)

Now we shall show that if a cardinal number m is finite, then for

every cardinal number n the product mn is equal to the sum of m com-
 § 3. Product of two cardinal numbers 137

ponents each of them being equal to the number n:

1 2 _m

(3.2) Mt NE ee ei -
Let M = {1,2,...,m}, let N denote a set such that N= n, and
let P= MXN. Thus we shall have mn= P. But of course P= P,+
+P.+...+Pm, where P, denotes for any given number k= 1, 2,...,m
the set of all pairs (k,) where ne N. We shall have of course P,= 1
for k= 1, 2,..., m, and the sets P, (k=1, 2,...) will be disjoint. Thus
the formula for P gives P= P, ay Fs Ee hey iB =nt+n-+...+n, and since
mn = P, we have formula (3.2), q. e. d.
The concept of the product of cardinal numbers can be generalized
by induction to an arbitrary finite number of factors, and such a product
obviously depends neither on the order of factors nor on the manner of
associating them.
Further it will be observed that if » denotes an arbitrary natural
number, and M,, M,,..., M, are sets such that M, = m, for k = 2 yore sfDs
then, taking P = M,x M,x...x M,, we shall have P= mm... tp
We shall now give various examples of the product of cardinal
numbers.
Formula (3.2) and Theorem 8 of III.7 immediately imply

SORES TROVE eyran is as ee

Theorem 1 of III.10 immediately implies the formula

NoNo = No.

Now, Theorem 1a of IV.5 and (3.2) immediately imply

tC OTe ial ee

Theorem 1 of IV.6 implies

(Ro ='C;
and Theorem 1a of IV.8 implies »
c=—C.

EXERCISES. 1. Prove that for any sets A and B we have A+B=A+B+AB.

Proof. For any sets A and B we have the following decompositions into disjoint
components: A=(A—B)+AB; B=(B—A)4+AB, A+B= (A—B)+(B—A)+AB,
whence A—A—B+AB, B=B—A+AB, A+B= A—B+B—A+AB, therefore
A-eP A Aeba A = (2B +B A+ AB)+ AB = A+B AB, gq: 6. a.
138 VII. Cardinal numbers

2. Prove that for any sets A, B,C

ALBLO= A-B4E-O-- Oa AO.

3. Prove that for any sets A, B, C

A+B+O0+ABO = A+B+O+AB+A0+B0.

Proof. In virtue of Exercise 1 we have for any sets A and 8: 4+ 8 = A+8+4AS8,

whence, taking S = B+0, we obtain A+B+C = A+B+0+AB+A0. Since (accord-
ing to Exercise 1) B+O0+BO=B+0
and AB+AO+ABC = AB+A0, we find
A+B+06+AB0 = A+ B+0+ 80+ 480 = A+B+O0+AB+A0+
B0+AB0 =
= A+B+C0+AB+A0+B0.
4. Prove that for any sets A, B,C, D we have

A+B+0+D+AB0+
A0D+ ABD+ BCD =
= Ae pO Dp AB ACT ADOBC- BD+ OD-_ABROD .

(As regards the generalization of this formula, see my note [49], p. 18-22).
5. Prove that if m and n are two cardinal numbers, then every set P of power mn
is a disjoint sum of sets, the set of its components being of power m and each component
being a set of power n.
Proof. Let P denote a set such that P = mn, and M and N — sets such that
M =m, N =n. From the definition of the product of cardinal numbers it follows that
MxWN=r1mn, therefore MxN~P; thus there exists a function » which associates
in a (1-1) manner to each element (m,n) of the set MxWN a certain element y(m, n)
of the set P. For each element m of the set M we denote by 4A,, the set of all elements
y(m,n) where n « N; obviously we shall have

Pw ete
meM

where the sets A,, for me M, are disjoint and of power n. This decomposition of the
set P satisfies the required conditions.
It should be noted that we are not able to prove the inverse theorem (stating
that if P is a disjoint sum of sets, the set of its components being of power m and each
component being a set of power n, then the set P is of power mn) without resorting to
the axiom of choice (even in the case where m = Ny), n = 2). The proof of that theorem
for arbitrary cardinal numbers m and n, based on the axiom of choice, will be given in
Chapter XVII. For m = &, and finite n, and for m = n= &,, the proof has been given
in VI.7 (Theorem 10).

4. Exponentiation of cardinal numbers. Let m and n be two given

cardinal numbers, and M and N — two sets such that M=n, N=n.
Let P = MN (see 1.18). If M, and WN, were any sets such that M, =m,
N= n, we should have M,~M, N,~N, whence, by Theorem 6 of II.5,
MM ~M™.
 § 4. Exponentiation 139

Thus the cardinal number p = MN depends only on the cardinal

numbers m and nt, and not on the choice of the sets M and NV, provided
only that M =m and V=n.
In the case of finite cardinal numbers m and n the cardinal number p,
a8 follows from the definition of the set IM” (see 1.18), will be equal to the
number m". We shall write p = m" also in the case of arbitrary cardinal
numbers m and n, and the number p will be called the power with basis m
and exponent n. Thus, for any two cardinal numbers m and n the power m"
is a well-defined cardinal number. It can easily be shown that, in parti-
cular, for any cardinal numbers m and n we have m!=m and 1"=— 1.
Let m, n,, n, be given cardinal numbers, and M, N, and N, — sets
such that M = ™m, i= tt, N,= n,, and N,N,= 0. In virtue of the
definitions of the sum, product and power, the cardinal number m+"2
will be the power of the set M1+%2, and the cardinal number m™m"™ will
be the power of the set Mix M2. But in virtue of Theorem 7a of II.5
we have M™+Ne—~M™x M2. Hence follows the equality

matt — mm":

for any cardinal numbers ™, 11,, 1.

It follows by induction that for any finite sequence of cardinal num-
bers 11,,1,,... 1%, and for any cardinal number m we have the formula

VAG Naki rer ah oe

notte
thie i (1 b

In particular, i 1,=t,—...=n,—1, it follows that -for every

natural number & and every cardinal number m the power m* is the
product of k factors, each of them being equal to the cardinal number m:
1 2 k

Now let m,, m,, t be given cardinal numbers, and M,, M,, N — sets
such that Ny, tlt ye Mn, = iit, N =n. From the definitions of the product
and the power of cardinal numbers it follows that m;m; is the power
of the set My x M2’, and (m,m,)" is the power of the set (M,xM,)%. But
in virtue of Theorem 8a of II.5 we have (M,x M,).~MY~x MY. Hence
follows the equality
(m,m,)" = mpm
for any cardinal numbers m,, m,, 1.
Now let m,un,p be given cardinal numbers, and M, NV, P — sets
~guch that M = m, N= Tt, P= p. From the definitions of the product
and the power of cardinal numbers it follows that (m")? is the power
140 VII. Cardinal numbers

of the set (M¥%)?, and m™ is the power of the set M%*?. But, according to
Theorem 9a of II.5, (M%)P~MN*x?, Hence we have the equality
(ins )® —— Tie.

for any cardinal numbers m, n and p.

5. Power of the set of all subsets of a given set. Let 1 denote a given
set of power m, U — the set of all subsets of the set WM. Let P = {0, 1}.
Obviously we shall have U~P™; a (1-1) correspondence between the
elements of the sets U and P™” can be established by associating with
each subset of the set M its characteristic function. Since, according to
the definition of exponentiation, the power of the set P” is 2™ (as P= 2),
the set U is of power 2™. Hence we have
THEOREM 1. The power of the set of all subsets of a set of power m is 2™
Thus, in particular, the power of the set of all subsets of the set
of natural numbers is 2°°. But, as we have proved in IV.8, Exercise 5,
the set of all subsets of the set of all natural numbers is of the power of
the continuum. Hence we have the formula

(5.1) C= 2",
In IV.7 (see the Corollary to Theorem 1) we have proved that if NV
denotes the set of all natural numbers, then the set N% is of the power
of the continuum. Hence, in view of N = s, and in virtue of the defini-
tion of exponentiation, we have the formula
iN
Noo
== Cs

therefore, in view of the previously obtained formula for the cardinal

number c: .
s s
yo = OP

Thus the exponentiation of cardinal numbers with different bases |

and the same exponents can give equal cardinal numbers. Exponentia-
tion with the same bases and different exponents may also give equal
results, since ¢. g. 89 = No. (From the formula s,%) = 8, it follows by induc-
fron hatin; fy fOr. == 1) Oe),
Now we do not know whether, for cardinal numbers m and n, the
equality 2™ = 2" always implies the equality m=, 4. e. whether two
sets for which the sets of all their subsets are equivalent must be equi-
valent themselves. This theorem follows from the so-called generalized °
continuum hypothesis (see Chapter XVI).
 § 5. Power of the set of all subsets 141

From the formulae ¢ = 2", s 98%) =), and from ,the formula proved
above for the exponentiation of a power it follows that ¢ °= (2%) =
= 2foro gf,
Thus we have the formula:

Coe Bee
EXERCISES. 1. Prove that ¢* = 2°.
Proof. In view of ¢ = 2°° and y,c =¢ we have c° = (2%9)* = 9% — 9°,
2. Prove that xf = 2°.

Proof. In view of ¢ =Ny)¢ and x5? =c we have N& = Nho* = (wX0)6 = cf = 2°.
3. Prove that if M is a set of the power m, then the set of all infinite sequences
whose terms are elements of the set M is of the power m*°.
Proof. The proof follows from the definition of exponentiation and from the
observation that infinite sequences of elements of the set M are in a (1-1) correse
pondence with functions of a natural variable whose values are elements of the set M,
4. Prove (without resorting to the axiom of choice) that a cardinal number m is
finite if and only if
ge 1 oe,

Proof. The necessity of our condition is obvious. It remains to prove its sufficiency.
Assume that m is not a finite cardinal number and let M denote a set of power m; thus M
will be an infinite set. Therefore for every natural number n there exists a subset of
the set M having n different elements. Denote by T,, the set of all subsets of the set M
that are composed of n elements. Thus we shall have 7, 4 0 for n= 1, 2,..., and
T, ~ T,, for different natural numbers k and n. Denoting generally by U(Z) the set
of all subsets of the set Z, we shall have of course 7, c U(M); thus T, « UU(M) for
nm =1,2,..., which proves that the set UU(M) contains a denumerable subset, there-
fore its power will remain unaltered (in virtue of Theorem 4 of III.7) if we join one
element to it. But according to Theorem 1 the power of the set UU (MM) is 22", and the
power of this set enlarged by one element is 22"-+4 1; thus we have 22"-+ 1 = 22", contrary
to our assumption. Therefore the cardinal number m must be finite, which proves the
sufficiency of our condition. Thus the required proof is completed.
Now we are not able to prove without the aid of the axiom of choice that the
cardinal number m is finite if and only if 2"11 4 2™. Neither can we prove without
the aid of the axiom of choice that if the cardinal number m is such that 22"t* 4 22™,
then m is a finite number.
5. Prove without the aid of the axiom of choice that if m is a cardinal number
m

which is not finite and n= al , then, n? =n.

Proof. If m is a cardinal number which is not finite, then, in virtue of Exercise 4,
we have gam 1 = 2". Hence

2 p20" = 92 ata 2 on ip 5
q. e. d.
However, we are not able to prove without resorting to the axiom of choice that
if the cardinal number m is not finite and p = g2 then p? = p, but we can prove without
using the axiom of choice that in that case 2p =p. :
142 _VII. Cardinal numbers

6. Prove that the set F of all real functions of a-real variable is of the power 2°,
’ Proof. From the definition of the exponentiation of sets it follows that / = DG
where X is the set of all real numbers. Thus f = F = c*. But, according to Exercise 1,
* — 2°; thus we have f = 2°, q.e. d.
It follows hence that the set of all real functions of a real variable has the same
power as the set of all functions of a real variable, assuming only two different values,
e.g. 0 and I.
7. Prove that a set of power c has 2° different subsets of power c.
Proof. It suffices of course to prove that the set X of all real numbers has ve
different subsets of power c.
Let us denote by 7 the set of all subsets of the set X, and by F the set of all those
subsets of the set X which are of power c. Let X, denote the set of all positive real num-
bers; X, is, as we know, a set of power c; therefore the set F’, of all its different subsets
is of power 2°. Thus the set F, of all sets of the form (X—X,)-+A where A«F, will
be of the same power. But the set X—X,, as the set of all real numbers <0, is of
power c, whence we easily conclude that every set belonging to Ff, is of power c. Thus
we have F,cF'c 7. But F, = 2° and, as we know, f= 2°; therefore F,~7, whence,
according to Theorem 4 of II.6, we conclude that F~7, and finally that TD q.e. d.
8. Prove without the aid of the axiom of choice that 1+ 22%° = 22%
Proof. In virtue of Theorem 4 of III. 7 it suffices to prove that a set of the power 27°
contains a denumerable subset. The set of all sets of real numbers is a set of power 27°
and its denumerable subset is the set of the sets {{1}, {2* 43%, ae
9. Prove without the aid of the axiom of choice that the total number of two-
-element subsets of a set of power 27° is 29,
Proof. As a set of power 22° we take the set U(X) of all subsets of the set X
of all real numbers. Let 7 denote the set of all two-element subsets of the set U(X),
a. e. the set of all sets {4, B} where A and B are different sets of real numbers. The
number of sets {A4, B} where one (and only one) of the sets A and B is empty is the
same as the number of non-empty subsets of the set X, i. e. 2%. Since 2?°°+ 27% —
= 20, 2 — o2%0+1 — 98%, the proof that the set T is of power 22°° is obtained by prov-
ing that the set 7, of all sets {4, B}, where A and B are different non-empty sets of
real numbers, is of power 2. Let f({4, B}) = Ax (B—A)+Bx (A—B). FromI.17,
Exercise 10, it follows that the function f maps the set 7, on a subset of the set U(X x X);
therefore there exists a function h mapping the set 7, on a subset of the set U(X),
since, according to Theorem la of IV.8, we have X x X~X, whence U(X x X)~U(X).
On the other hand, we have X~X — {1}, whence U(X)~U(X— {1}), and the function
g(A) = {1}, A} maps the set U(X — {1}) on a subset of the set 7,. In virtue of the
Cantor-Bernstein Theorem we thus have 7T,~U(X), whence Te alas Ce Ib
10. Let us divide all real functions of a real variable that are not everywhere = 0
into pairs, assigning two functions to the same pair if and only if they differ only
in their signs. Let M denote the set of the pairs obtained in this way. Prove without
the aid of the axiom of choice that M = 2°.
Proof. Let X denote the set of all real numbers. With every set Ac X let us as-
sociate a function f, such that f,(~) = 1 for «¢ A, and f(x) = 2 for we A. Obviously,
for AcX, Bc X,A +B the pairs {f,(a), —f,(x)} and {f,(x), —f,(x)} will be different
from eachother. Hence it follows that the set U(X) is equivalent to a certain
subset of the set MZ. On the other hand, the set M is of course a subset of the set of all
 § 5. Power of the set of all subsets 143

two-element subsets of the set of all real functions of a real variable, therefore, in virtue
of Exercise 6, it is a subset of a set whose power is 22°. Therefore the set M is equivalent
to a certain subset of the set U(X). Thus it follows from the Cantor-Bernstein theorem
that M~U(X), and finally M = 2°, g.e. d. ;
11. Prove that there are 2? different ways of decomposing the set of all points
on a straight line into two disjoint sets congruent by translation.
Proof. Let A denote an arbitrary set of points of the segment J = (0< 2 < l].
Let B = J—A and denote by E(a) the set congruent to the set EH by translation by the
length a along the straight line. Let
+00 +00
9 = » Aen >» B2k-1).
=—0C0O K=— —OO

Obviously, the sets 7, and 7 ,(1) are disjoint and congruent by translation, while
their sum is the whole straight line, different sets T., corresponding to different subsets A
of the interval J. Thus there are 22°° such sets.
 CHAPTER VIII

INEQUALITIES FOR CARDINAL NUMBERS

1. Definition of an inequality between two cardinal numbers. Let m

and n be two given cardinal numbers, M and N —sets such that M =m
and NV = n, and assume that M<WN (see V.1). If the cardinal numbers m
and nm were finite, we should have m <n. Now, if M< N , then also in
the case where the cardinal numbers m and n are not both finite we shall
write m <n and say that the cardinal number m is less than the cardinal
number n. We shall also write the same in the form n>™m™ saying that
the cardinal number n is greater than the cardinal number m.
Theorem 1 of V.1 shows that the formula m <n is independent of
the choice of the sets M and WN satisfying the condition M =m, N =n.
Thus, in certain cases, the formula m<n holds for the cardinal
numbers m and n. This formula is equivalent to the assertion that every
set of the power m is of less power than every set of the power n.
As we know from V.1, if the formula M < N holds for two sets M
and N, then neither MZ= N nor M > N. Hence it follows that for cardinal
numbers the formula m<n excludes the formulae m=n and m>n.
However, we cannot yet assert that every two cardinal numbers m and nt
can be joined by one (and only one) of the three signs <, =, > (¢. e. that
we have trichotomy for cardinal numbers); this theorem will be proved
later with the aid of the axiom of choice (see XVI.1). Thus we cannot
yet maintain that if for the cardinal numbers m and n the inequality
m<mn does not hold, then we must have either m=n or mM>nNt (Z. €.
m >n). In particular, the assertion that every cardinal number is either
<N) OF >, iS Obviously equivalent to Theorem 3 of VI.7 (stating that
every infinite set contains a denumerable subset), which has been proved
with the aid of the axiom of choice.
Theorem 1. Jf m and n are cardinal numbers such that m <n, then
there exists a cardinal number p (>0) such that n= m-+p.
Proof. Let M and N be sets such that M = ™m, N=n. In view
of m<n we have M < N, whence it follows (see V.1) that the set
 Pr
§ 1. Definition of an inequality 145

is equivalent to a certain subset NV, of the set V. Thus we shall have Y, =

= M=m. Let N,= N—N,; we shall have N = N,+N, and V,N,= 0,
whence VN = N,+N,. Let N,=p. In view of N,N,=0 we shall have
n= m-+p. The set N, is non-empty since we should then have V = N,,
i. €. n= Mm, contrary to the assumption that m < n. Therefore the cardinal
number p is positive. Thus Theorem 1 is proved.
It will be observed that Theorem 1 cannot be reversed, because we
have for instance s) = %)+1.
Now the question arises whether for any two cardinal numbers m
and n such that m<n there exists only one cardinal number p such
that n= m-+p. The theorem stating this proves to be equivalent to the
axiom of choice, as will be shown in XVI.2, Theorem 3.

Examples of inequalities for cardinal numbers

From Theorem 1 of V.2 it immediately follows that if n is a natural

number and m a cardinal number which is not finite, then nn < m.
Thus, ams particular, 2s, for m= 1.25...
It is easily seen that, conversely, if a cardinal number is < sy, then
it is a finite number.
From Theorem 5 of V.5 it immediately follows that

(1.1) - nae
From Theorem 6 of V.5 and Theorem 1 of VI.5 immediately follows
the inequality
(1.2) Sait

for any cardinal number m.

In view of formula (5.1) of VII, formula (1.2) immediately gives
(for m= y,) formula (1.1). In view of (1.2) we find further that
" Xo
Saeed ae er Oe ul
EXERCISES. 1. Prove that if M is a set of the power m, P — a set of power p,
such that Pc M, and n — a cardinal number such that m =n >p, then there exists
a set N of power n such that MONOP.
Proof. The proof can be based on the following lemma:
If P and Q are two arbitrary sets and if y(#) is an invertible function defined
in the set Q, and y(x) is an invertible function defined in the set P such that
y(P)cQ, then there exists a set H such that (99 H5Q—y(4#), y(L)y1(Q—-Z) =0,
y(£)+y1(Q—E) 2 P (where y() denotes the function, defined for x «y(P), inverse to
the function y(a)).
In order to prove this lemma it suffices to denote by H the product of all sets H
satisfying the conditions 9D H2Q—y(P) and »w[Pp(X)]c H for Xc H.
Cardinal and ordinal numbers 10
 146 VIII. Inequalities

In order to deduce the required theorem from this lemma it suffices to denote
by Q an arbitrary set of power n, and by gy — the mapping of the set P on a subset
of the set Q (such a mapping exists in view of n> p), and to take f(x) = y(#) for re H
and f(x) =y (x) for « «Q—E.
The function f maps the set Y on the set V = y(F)+y (Q—B). It follows from
our lemma that MZ5>N2>P and N-~Q, therefore NV = n}),

. Prove that if m is a cardinal number + 0, then there exists a cardinal number n

such Te Q™ _ on.
Proof. If m = 1, then 2" — 2-1, andifm 4 0 and m ¥ 1, then we obviously have
m > 1, and, by Theorem 1, there exists a cardinal number p such that m= 1+p,
Whence s? seo"= 250"
nt

3. Prove without the aid of the axiom of choice that if MW is an infinite set of
power m and » a natural number, then the set M,, of all m-element subsets of the set W
is of power > 1m.
Proof. If M is an infinite set and » a natural number, then we are able to prove
without the aid of the axiom of choice that there exists a sequence p,, P2, +++. Pa» Pa+1>
composed of n+ 1 different elements of the set M. Let

f (Pe) = (Pris Par +> Puno Pasat— {Pes for k=1,2,...,n—1,

and :
iGO) Ug iOen ecso iO sin Oe BOR foe NEO Fy ocon JOs_auhh o

It can easily be verified that the function f is invertible in the set VM and its values
are elements of the set M,5 thus it maps in a (1-1) manner the set M on a subset of
the set M,, n? whence JM < M,.; therefore. in view of M =m, M, Ss Gis Cs Ol
Note 1. Obviously if » is a natural number >1, then the above theorem holds
also for finite sets MW, having more than n elements.
Note 2. It can be proved with the aid of the axiom of choice that if M is an in-
finite set of power m, then the sets MZ, are of power m for n= 1, 2,...

4. Prove that if m is a cardinal number >, then the cardinal number 2™ is si-
multaneously even and odd, i. e. there exist cardinal numbers n and p such that 2™ =
== 21 = 2p-e 1.

Proof. If m>w8,, then, as we know, tm =m-+2, whence 2

ee Fon ee oC Ot ee Oe which sivese) == O40 So eee eee
Note. With the aid of the axiom of choice it can be proved that every cardinal
number which is not finite is simultaneously even and odd (see XVI.2); now without
the aid of the axiom of choice we are unable to prove that every cardinal number is
either even or odd (7. e. that it is at least of one of the forms 2n and 2n+1). Neither
can we prove without resorting to the axiom of choice that the power of an arbitrary
set which is the sum of two-element disjoint sets is an even cardinal number. (In order
to prove this theorem the axiom of choice for two-element sets, i. e. axiom [2] in VI.5,
would be sufficient).
5. Prove without the aid of the axiom of choice that if m is a cardinal number
>,, then m+, =m.

1) For a detailed proof see Sierpinski [39], p. 72-74; L. Vietoris, Jahresbericht

der Deutsch. Math. Ver., Bd 60, Heft 1, p. 2-3 (Losung der Aufgabe 361).
if
 § 1. Definition of an inequality 147

Proof. Let m denote a cardinal number =8,. In virtue of Theorem 1 there exists
a cardinal number p such that m = p-+-§\,. (In case of m = 8, we can assume that p = 0).
Hence, in view of the formula 8)+N8)=8), wé obtain m+, = (P+S,) +8) = pt
+ (No No) =P+N, =m, q.e. d.
It is also obvious that the converse is true: if, for a certain cardinal number m,
we have m+'S, =m, then m>k,.
6. Prove without the aid of the axiom of choice that if m is a cardinal number >c,
then: tt ¢ = mm.
7. Prove without the aid of the axiom of choice that if m and n are cardinal
numbers such that m>c>n, then m+n =m.
8. Prove without the aid of the axiom of choice that if m is a cardinal number
which is not finite, then for any natural n we have 2™ > nm.
Proof. Let m denote a cardinal number which is not finite, and n — a natural
number. Therefore we have m>vn?+n and there exists a cardinal number n> n?2
such that m—n-+». Since, for natural n, we have 2” >n+1 and also oS nt, we
obtain 2™ = 2"*" — 2".2" >n(n+ 1). Thus, in view of n>=n?, we have 2™>nn+
tn > nn+n?= n(n+n) = nm, 7. e. 2° > ni, Gs a GE
However, we are unable to prove without the aid of the axiom of choice that if m
is a cardinal number which is not finite, then 2™ > ,m, or even that 2" >y,.

2. Transitivity of the relation of inequality. Addition of inequalities.

From Theorems 3, 3a and 3b of V.1 and from the definition of the rela-
tion < for cardinal numbers immediately follows
THEOREM 1. Jf m, n and p are cardinal numbers, then

m<nand n<p implies m<p,

m<nandn<p implies mM<p,
m<n and n<p implies m <p.
THEOREM 2. Jf m,n, m,, n, are cardinal numbers such that

a | fete and. ie 1 = it 1)
then
(2.2) sis, SM ity

Proof. Let m,n, m,, 1, be cardinal numbers satisfying conditions

(2.1), MW, N, M,, N, — sets whose powers are the respective cardinal
.
numbers. We can assume that the sets V and N, are disjoint and, in view
of (2.1), that the set M is a subset of the set N and the set M, — a subset
of the set N,. For, in view of (2.1) ,we have M<N , whence it follows
that the set M is equivalent to a certain subset of the set N and we can
replace the set M by that subset, since what we want is only that the
set M be of power m. The same applies to the set M,. Since MM, = 0
and NN,=0, we have M+M,—m+m,, and NLN,=n+n,, and
since MC N and M,CN,, whence M+ M,C N+N,, we have inequality
(2.2), q.e. d. Thus we have proved Theorem 2.
10*
148 VIII. Inequalities

For finite cardinal numbers the inequalities

(2.3) m<n. and “mM, <n,

imply, as we know, inequality m+m'<1n+n,. For arbitrary cardinal

numbers it need not be so, because for instance sy) <s) and 1 < 2,
but x) +1—s,+2 (both sides being, as we know from VII.2, equal to w,).
Later on we shall prove with the aid of the axiom of choice that for
cardinal numbers the inequalities m<n and m,<nt, imply the ine-
quality m-+-m, <n-+n,. We shall even prove, after A. Tarski, that this
theorem is equivalent to the axiom of choice (see XVI.2, Theorem 10).
THEOREM 3. In order that, for the cardinal numbers m and n, we have
m <n, it ws necessary and sufficient that there exists a cardinal number p
such that n= m+p.
Proof. If m#~n, then it immediately follows from Theorem 1 of
VITI.1 that the condition of Theorem 3 is necessary. It is necessary also
in case of m=n since then we can assume that p= 0.
On the other hand, for every cardinal number p we have of course
p > 0 and, in virtue of Theorem 2, for any cardinal numbers m and p .
we have m+p >m+0=—m; therefore if n—m-+p, then n>m. This
proves that the condition of Theorem 3 is sufficient.
Thus we have proved Theorem 3. It follows from this theorem that
the definition of the relation < for cardinal numbers can be reduced to
the definition of addition for cardinal numbers.
EXERCISES. 1. Prove (without the aid of the axiom of choice) that for the cardinal
numbers m and n the inequality m < n+ 1 implies the inequality m <n (but not vice
versa).
Proof. Let m and n be two given cardinal numbers such that m < n+ 1. There-
fore there exists a set Q of power n+ 1 and its subset M of power m, M being a proper
subset of the set Q since in case of M =Q we should have m = n+ 1, contrary to our
assumption. Thus there exists an element g «Q—M. On the other hand, in view of Q =
= n-+ I, there exists a set N of power n and a set P, consisting of only one element p,
such that Q = N+P (see VII.2, Exercise 1). If p ¢ M, then, in view of McQ = N+P,
we have Mc N, whence M< N, i.e. m<n, and if p « M, then in view of q<«Q—M,
we have q # p; the set M = (M— {p})+ {p} is equivalent to the set (M— {p})+ {q}c N,
whence again m<n.
2. Prove without the aid of the axiom of choice that if m is a cardinal number
which is not finite and if m < 2%, then 2° < 2™}),
Proof. Let m denote a cardinal number, not finite and such that m < 2°°. Since 2°°
is of the power of the set of all real numbers (by (5.1) of VII), and m < 2°, there
exists a set X of real numbers of power m. According to Exercise 2 of VI.6 there exists

1) A. Tarski [10] (p. 309) stated without giving a proof the following theorem:
it can be proved without the use of the axiom of choice that if m< 2“ and m is not
a finite cardinal number, then 2° < 2™
 § 2. Addition of inequalities 149

a decomposition of the set X into an infinite series of non-empty disjoint sets, 4 =

= X,+X,+... Now associate with every set Z of natural numbers the set

=a) as
neZ

The sets X, (n= 1, 2,...) being disjoint and non-empty, to different sets Z of
natural numbers correspond different subsets 7'(Z) of the set X. Thus, denoting by WV
the set of all natural DUETS, we shall have a set U(N), equivalent to a certain subset
of the set U(X), whence U(N
(N) < U(X).3
But in view of V = No, X =m, and by Theo-
rem 1 of VII.5 we have U(N) = 2°, U(X) = 2™. Therefore we have 2°< 2™, q.e. d.
(Sierpinski [43], p. 60).
3. Prove without the aid of the axiom of choice that if, for a cardinal number m,
we have 2™ < 2%, then m < y, (in other words, that a set whose number of subsets 18
less than the continuum must be finite).
Proof. Let us assume that the cardinal number m is not finite. Since, as we know,
m < 2™ and 2™ < 2 by assumption, we obtain m < 2“ and, in virtue of Exercise 2,
we are able to prove without the aid of the axiom of choice that 2°° < 2™, contrary to
our assumption. Thus the cardinal number m must be finite, therefore m < &p, q. e. d.
Now, we cannot prove without the aid of the axiom of choice that if, for a cardinal
number m, we have 2™ > 2%, then m > &,, while without the aid of the continuum-
hypothesis we cannot prove that if 2™ = 2%, then m=X,.
4, Prove {without ;the aid of the axiom of choice that for a cardinal number m
not to be finite it is necessary and sufficient that 8,< 27”.
Proof. The sufficiency of the condition is obvious, {and the proof of its necessity
follows from the observation that in VII.5, Exercise 4, we have proved without the
aid of the axiom of choice that if M is an infinite set, then the set UU(M) contains
a denumerable subset.
Now we are not able to prove without the aid of the axiom of choice that if m is
a cardinal number which is not finite, then §,< 2™.
5. Prove without the aid of the axiom of choice that if, for a cardinal number m,
we have m? >>, then m > No.
Proof. Let M denote a set of power m. In view of m? > &,, the set Mx M (which
is of power m?) contains a denumerable subset. Therefore there exists an infinite sequence
(p,> %)(Kk= 1, 2,...) of different elements of the set MxM. Thus we have p,« M and
q, « M for k = 1, 2,... If each of the infinite sequences p,, po, -.. and q, Qs, ... contained
only a finite number of different terms, then, all terms of the sequence (p,, q:), (Po, Ge), ---
could not be different from one another. Therefore at least one of the sequences ,, po, .-.
and q, q, -.. contains infinitely many different terms, the set of which forms a denumer-
able subset of the set M. Thus we have m= M>k8,. If m=), we should have m? —
=? =p, contrary to our assumption that m? > 8). Therefore m >No, q.e. d.
Now we cannot prove without the aid of the axiom of choice that, for cardinal
numbers, the inequality m? > n® implies the inequality m > n, or that the inequality
m >n implies the inequality m? > n’.
6. Prove without the aid of the axiom of choice that there is no cardinal number m
such that §)< 2 < 2“ (Tarski [5], Theorem 66, and Sierpinski [43]).
Proof. Let us assume that m is a cardinal number such that 8) <2™ < 2%. In
virtue of Exercise 3 we should have m < XN, 7. e. the number m, and therefore also the
number 2™, would be finite, contrary to the assumption that %)< 2™.
150 VIII. Inequalities

7. Prove without the aid of the axiom of choice that there is no cardinal num-
ber m such that 22" — 2%,
e 3 ot y
Proof. Let us assume that m is a cardinal number such that 22> = 2°. Thus the
: 5 ct m mt ¥ 9°
number m is not finite, and we have m < 2” < 22) = 2%°, therefore m < 2°°, whence,
° . * y nt A ° . . . x Tm rit
in virtue of Exercise 2, 2°°< 2", which is impossible since 2°° = 22° > 2”.
8. Prove without the aid of the axicm of choice that in order that a cardinal num-
ber m be not finite it is necessary and sufficient that we have 2°° < 2" (Tarski [5],
Theorem 69).
Proof. The sufficiency of the condition is obvious. In order to prove its necessity
let us assume that m is a cardinal number which is not finite and let M denote a set
of power m. For natural n let us denote by TZ’, the set of all n-element subsets of the
set M and let us associate with every set Z of natural numbers a set

Dien ler,
nNEZ

we shall have of course f(Z)« UU(M), and to different sets Z of natural numbers
would obviously correspond different elements of the set UU(M). Hence we obtain
the formula 2°° < 22", and since according to Exercise 7 the relation of equality can-
Mot-hold, 2°95 < 92". q. end.
9. Prove without the aid of the axiom of choice that if m is a cardinal number
such that 2" >, then 2™ > 2° (i. e. prove that a set which contains at least a denumer-
able aggregate of subsets has at least a continuum of subsets). (Tarski [5], Theorem 68),
Proof. Let m denote a cardinal number such that 2" >, and let MW be a set of
power m. In view of U(M) = 2™ >, there exists an infinite sequence of different non-
empty subsets of the set M: M,, M,,... Let S= M,+M,+...: we shall have M=
= (M—S)4+ M,+ M,+...; all the components of this infinite series of sets will be dif-
ferent and, except perhaps the first, non-empty. Thus the set M is the sum of an infinite
series of different non-empty sets, and therefore, as we have proved without the aid
of the axiom of choice in VI.6, Exercise 3, the set M is also the sum of an infinite series
of disjomt non-empty sets, whence, as in the proof of Exercise 2, we conclude that
os Oe eqne.cd:
10. Prove without the aid of the axiom of choice that if m, and m, are cardinal
numbers >1, then m,-+m, < m,m,.
Proof. If m, > 1 and m, > 1, then, in virtue of Theorem 1 of VIII.1 there exist
cardinal numbers n, and n,, both >0, such that m, = n,+ 1, m, = n,+ 1, whence m,m, —
= (n,+ 1)(n,4+ 1) =n, n,+n,+n,+ 1, and since, in view of n, > 0 and n, > 0, we have
nin, > 1, therefore m,m, > 1+n,+n,+1=m,+m,; thus m,+m,<m,m,, q.e. d.
Note. If one of the numbers m, and m, are =—1, then the theorem proved above
may be false, since for instance 1+ 3 > 1-3. But 1+8,=1-.-
11. Prove without the aid of the axiom of choice that if k is a natural number
and m,,m,,...,m, are cardinal numbers >1, then m,+m,+...+m,<m,m,... m,.
Proof. According to Exercise 9, the theorem which we want to prove is true for
k = 2 and for k = 1 it is obvious. Now let & denote a natural number >2 and let us
suppose that the theorem is true for k—1 cardinal numbers; let m,m,,...,m, be k
given cardinal numbers >1. Thus we shall have m = m,+m,-+...- m,_, < mm,...m,_,
and (since the theorem is true for two cardinal numbers and since m >m,_, > 1)
m-+m,<mm, and m,+m,+...+m,_,+m,<m,m,...m,_;m*%, which proves the
 § 2. Addition of inequalities 151

theorem for k cardinal numbers. Hence, by induction, follows its truth for an arbit-
rary finite number of cardinal numbers.
Note. For finite cardinal numbers m, and m,, both >1, even the inequality
m,+m, < m,m, can easily be proved to hold. This inequality, however, is not neces-
sarily true for cardinal numbers m, and m, which are both > 2 but not both finite, since
for instance 3+ 8, = 3-,. It can even be proved with the aid of the axiom of choice
that if m, and m, are cardinal numbers <0 of which at least one is not finite, then
m,+-m, = m,m,, but the proof is difficult. It can also be proved that this theorem is
equivalent to the axiom of choice (see XVI.2, Exercise 1).
12. Prove without the aid of the axiom of choice that Theorem T,, stating that
for every cardinal number n > &, there exists only one cardinal number p such that
n=,+P, is equivalent to Theorem T,, stating that every cardinal number is either
SNL Ole N oe
Proof. Suppose Theorem T, to be true and let m denote a cardinal number for
which the formula m < 8, does not hold. From Theorem 2 it follows that in any case
M-+ S$) >, and m+, >m; thus if m+, =), we should have S) >, contrary to
our assumption. Therefore n = m+, > 8, and, according to Theorem T,, there exists
only one cardinal number p such that n = 8,+p, and since n= m+, =N,+m, we
must have p =m. But we also have &)+ (S>+m) = (S)>+8))-+m = +m =n, which,
in view of Theorem T,, gives $,.+m=m, whence m>8,. Thus we have T,—T,.
Now suppose that Theorem T, is true, and let n denote a cardinal number such
that N > 8. In virtue of Theorem 1 there exists a cardinal number p such that n =
=N,)+p. Ifp < Ny, we should have n = &)+P <Ny+N, = Ny, Whence n < Np, contrary
to our assumption. Therefore it follows from Theorem T, that p > Np.
Hence it follows that a set of power p contains a denumerable subset, and there-
fore, by Theorem 4 of III.7, it will not change its power if we join with it a denumer-
able set. Thus ps, = p, therefore p = n, which proves that number p is unique. Thus
we have T,—T,.
The equivalence of Theorems T, and T, is thus established.
13. Prove that if the cardinal numbers m,n, p are such that m+n =n andn<p,
then m+p=p.,
Proof. In view of t <p and in virtue of Theorem 3, there exists a cardinal num-
ber g such that.p =n-+q. Hence, in view of m+n=n

m+p=m-+ (n+ q) =(M+n)+q =N+q=P,

therefore m+p =p, q.e.d.
14. Prove that if m<n and n+p=p, then m+p=p (Tarski [13], p. 10,
Theorem 1.30).
Proof. Ifm <n andn-+p =p, then, by Theorem 2, m+p>p, andn+p >m-+p,
whence m+p >p=n+p>m-p, therefore m+p =p, q.e.d.

THEOREM 4. For cardinal numbers inequalities (2.1) imply the ine-

quality
(2.4) mm, <n, .

Proof. Let M, M,, N, N, be sets whose powers are respectively

m, m™m,, 1, 1,. In the same way as in the proof of Theorem 2 we can as-
sume that MC N and M,C N,. Therefore we shall have Mx M,C NxWN,,
152 VIII. Inequalities

whence Mx M, <NXN,; since Mx M, = mm,, and NXN,=nn,, we

have inequality (2.4), q. e. d.
For finite cardinal numbers inequalities (2.3) imply the inequality
mm, <nn,. For arbitrary cardinal numbers it need not be so because,
for instance, 8) = 8%), 1< 2, and s)-l1=%)-2. We also have s,<c but
Not = CC.
Later on we shall prove with the aid of the axiom of choice that
for cardinal numbers the inequalities m <n and m, <n, imply the ine-
quality mm, <1n,, and even (after Tarski) that this theorem is equi-
valent to the axiom of choice (see XVI.2, Theorem 2).
From Theorem 4, for m= n, m, = 1, n, = p > 1, we can immediately
infer the following
CoROLLARY. For any cardinal numbers m and p >1 we have mp >

3. Exponentiation of inequalities for cardinal numbers. We shail prove

THEOREM 1. For cardinal numbers inequalities (2.1) wmply the ine-
quality
(3.1) i Was SSS ta

Proof. Let m,n, m,, nm, be cardinal numbers satisfying (2.1), and
M,N, M,, N, — sets whose vowers are respectively equal to those
cardinal numbers. In view of (2.1) we can assume that MC NW and M,C N,.
According to the definition of exponentiation we shall have m™ = uM",
n™ = NV, But from the definition of power of sets (1.18) it immediately
follows, in view of MC N, that M@1C N™%, which gives m™ <n™. In case
of m, =n, we have formula (3.1). Now let us assume that m, ~1n,: in
view of (2.1) we have n, > m,, and, by Theorem 1 of VIII.1, there exists
a cardinal number p, > 0 such that n,= m,+),, whence, in virtue of
the properties of the exponentiation of cardinal numbers (VII.4), mn =
=n™n't, whence, by the Corollary to Theorem 3 of VIII.2, n>n™.
The inequalities m™ <1" and n= > n™ give again (in view of Theorem 1
of VIII.2) inequality (3.1).
Thus Theorem 1 is proved.
It will be observed that for the cardinal numbers m, n, p the ine-
quality m<n does not imply the inequality m <n” because, for
example, 2<c but 2°°—c™ (see VII.5). Similarly, for the cardinal
numbers m, p,q, the inequality p< q does not imply the inequality
m?® < m® since, for instance, 1 < 2 but xp = 8}.
Yet with the aid of the axiom of choice we can prove that, for car-
dinal numbers, the inequality m® < n” implies the SNE IEN m<n, and
the inequality m?’ < m* implies the inequality p < q.
 §$ 3. Exponentiation of inequalities 153

It should be noted that, as has been proved by A. Tarski [3] (p. 10),
the inequalities m<n and m, <n, for cardinal numbers do not imply
the inequality m™ < n"..,
Here is an elementary proof by A. Tarski (different from the one
given in his work quoted above).
Let N, denote an arbitrary infinite set. We define the sets NV, (n=
= 1,2,:.) Dy induction putting NV,;= U(N,-1)-for n= 1,2,... (where.
U(A) denotes the set of all subsets of the set <A).
Let
(3.2) M = Nyt Net e+. .
In view of NV,= U(N,_1), we have, for natural n, N,1¢« N, and,
by, (5:2), Vey ¢ 1 for = 1.2, . Since, in view of NV,= U (Nn 1) and
by Theorem 1 of V.5, we have ¥, Swi 1, this gives N, = N, oN, =.
whence it follows that the sets Ny, N,, N2,... are all different from ote
another, and since they are elements of the set M, the set M contains
a denumerable subset. For m= M we thus have

(3.3) M>N.

Now let Q denote any subset of the set M; by (3.2) we shall have

(3.4) Q = QN,+9N,+9N2 +...

For natural n we have QN,_1C N,_1, therefore QN,_1¢€ U(Nyz_1) =
= N,, and by (3.2) QN,_1« M. Thus to each subset QY of the set M cor-
responds an infinite sequence YN,,QN,,QNo, ... of elements of the set J,
and it follows from formula (3.4) that to different subsets of the set M
correspond different sequences. Therefore the set 7 = U(M) is (effectiv-
ely) equivalent to a certain subset of the set S of all infinite sequences
whose terms are elements of the set M. But in view of 7 = U(M) and
M= m, we have 1 = 2", while it follows from the definition of the set S8
that S =m (see VII.5, Exercise 3). Since 7~S,CS, we must have
T <8, and thus

(3.5) 2 me.

Now let P denote the set of all sets {{a}, {a, bs}, where ae M and
be M. As we know from 1.17, Exercise 9, we have P~M x M, whence
P =m. On the other hand, if ae M and be M, then by (3.2) we have
with certain natural k and l, ae Nx_1, be Nj_1, whence, in view of the
definition of the sets N,: {a}e Ny, {a,b} « Neyi+, and {{a},
{a, b}he
e Nga, M. Thus we have PC M, whence P<m, and since we
154 VIII. Inequalities

have previously found that P = m’, this gives m? <im, and since, by the
Corollary to Theorem 3 of VIII.2, m? > m, we have

(3.6) NV

Formulae (1.2), (3.3), (3.5) and (3.6) give at once by Theorem 4

of VITI2:

(3.7) Pie ag|8 COR 5 a Gaal eae ay

whence

(3.8) om mo,
Further, in view of (3.7) and (3.3), we have

2 Ny it = 2

whence 2™ = sf, and therefore, in view of (3.8)

(3.9) Sh alte.

It follows from the definition of the set M that MD U(N,), whence

u> UW); and since MW = m, N, == ty, we have m 2 2" > n,. Simee Wy
is an arbitrary infinite set, there exist cardinal numbers m™, satisfying
equality (3.9), greater than an arbitrary cardinal number. Thus there
exist infinitely many cardinal numbers m satisfying equation (3.9), and
therefore there exist also infinitely many pairs of different cardinal
numbers m and p satisfying the equation m? = p™. We owe this result
to S. Banach. (Among natural numbers there exists, as we know, only
one pair of different numbers m and p satisfying the last mentioned
equation, namely 24 = 4?).
In view of m > ny = N, and the observation that the set NV, is infinite,
we obtain m-~x,, and formula (3.3) gives mM > Xp.
Let
(3.10) {S= Sire ity —— Naa ttt

we shall have 1 = x5 2 2™ =i, 1, = Mm = 8) == my), therefore

(3.11) Nie itp ad) Ween,

while, in view of (3.10), (3.9) and (3.6): 1% = (s@™)™ = st = ys" — m= m™;

therefore
(3.12) LU pees |e be

Formulae (3.11) and (3.12) prove that inequalities for cardinal

numbers cannot be exponentiated by sides, q. e. d.
 § 3. Exponentiation of inequalities 15d

Further, it will be observed that without the aid of the so-called

generalized continuum hypothesis we are not able to settle the question
whether, for cardinal numbers, the inequality m <1 always implies the
inequality 2™ < 2" (ef. VII.5), and without the aid of the axiom of choice
we are not able to prove (even for m= ys,) that the inequality 2" < 2,
implies the inequality m<n. (The proof by means of the axiom of
choice will be given in XVI.1, Exercise 3).

4. Relation m <xn. A. Tarski [5] (p. 301) has introduced a rela-

tion <x between two cardinal numbers, which is defined as follows:
m <n denotes either that m= 0 or that every set of power n is the
sum of m non-empty disjoint sets.
It is easy to prove without the aid of the axiom of choice that if m <n,
then m<-xn. Indeed, suppose that m <n, and let N denote a set of
power n. In view of m < n there exists a subset M of power m of the set V.
According to the definition of the relation <x, if we had m= 0, we should
have m <n. Therefore suppose that m+ 0. Then the set M is non-empty
and there exists an element p, of the set M. Let us write A(p)= {p}
for pe M, p~py, and A(pyo) = {po} +(N—M). Clearly we shall have

Ne peM
Any
therefore N will be the sum of m non-empty disjoint sets. We have
proved without the aid of the axiom of choice that the formula m <n
implies the formula m <*xn. With the help of the axiom of choice it is
easy to prove that the converse is also true. This immediately follows
from Theorem 1 of VI.7.
Now we can prove without the aid of the axiom of choice that if
m<xn, then 2" < 2". Indeed, from the definition of the relation <x it
follows that if m <n and if NV is a set of power n and M a set of power my
then. there exists a decomposition

v=
>’ A(p)
peM

where A(p)(p « M) are non-empty: disjoint sets. Now for every subset 7
of the set WM let
ee
(C93 ae peT

It is easy to verify that the function f associates with different

subsets of the set M different subsets of the set N. Hence it follows
that _U(M) =< UY), 4. é. that 2" <2", q:e.d.
156 VIII. Inequalities

EXERCISES. 1. Prove without the aid of the axiom of choice that if m is a car-
dinal number such that m<-*,, then M< Wp.
2. Prove without the aid of the axiom of choice that if m,n, p are cardinal num-
bers such that m<x*«n and p 40, then p™ <p" (Tarski [5], Theorem 4).
Proof. Suppose that m<«*n; we can assume of course that m ~ 0. Therefore
if N denotes a set of power n, and M a set of power m, then there exists a decomposition

N=» A(p)
peM

where A(p), for p « M, are non-empty disjoint sets. Let P denote a set of power p and
let f denote an element of the set P™, i. e. a function f(x), defined for « « M, having
values which are elements of the set P. Now denote by (x) a function defined for
a « N as follows. If x « N, then, in view of

N=
)' Alp)
peM

and the disjointness of the sets 4(p) for p « M, there exists one and only one element
pe M such that x « A(p). Let y,(x) = f(p). Clearly we shall have 9; « P%, and to dif-
ferent functions f « P” will correspond different functions g,, 1. e. to different elements
of the set P™ will correspond different elements of the set P™ . Hence it follows that the
power of the set P is < the power of the set PY, i. e. that p™ <p", q.e. d.
3. Prove without the aid of the axiom of choice a theorem of Tarski [5] (p. 303,
Theorem 58) stating that if pq<m-+n, then either p< m or g<*«n.
Proof. In view of pq <m-+n there obviously exist sets P,Q, M,N such that
P=p,Q=q,M=m, N=n, MN =0 and PxQc M+WN. Let us suppose that p< m
is not the case, and let q denote any given element of the set Q. In view of PxQc M+WN,
we have, for p< P, (p,q) « M+N. We denote by A (q) the set of all ordered pairs (p, q)
such that (p,q) « NV.
If we had A(q) = 0, we should have (p,q) « M for p« P and jthe set M would
contain a subset of power p composed of all the elements (p,q) where p« P and we
should have p<m, contrary to our assumption. Thus ‘we have A(q) #0 for qeQ.
But of course the sets A(q) where q«Q are disjoint, and since A(q) c N for q«Q, the
set NV contains the sum
YAW
aeQ

composed of q non-empty disjoint sets, whence, in view of VN =n, it immediately fol-

lows that q<*n.
Thus we have proved that if we do not have p< m, then we have q <«n,q. e. d.

4, Prove without the aid of the axiom of choice a theorem of A. Tarski [5] (Theo-
rem 60) stating that if pq<* (m+n), then p<xm org<xn.
Proof. In view of pq <«(m-+n), there exist sets M, N, P and Q such that M =
=m, N=n, MN=0, P=p, Q=q, and, for peP, geQ, non-empty disjoint sets
A(p,q) such that :
M+N= »' Alp, q).
peP 7€Q
 $4. Relation m= *«n 157

Let us suppose that p<*m is not the case, and let q denote a given element of
the set Q. Let
Big) =D) N*A(p.0).
peP

If we had B(q) = 0, we should have N-A(p,q)=0 for p ¢ P and, in view of the

formula for M+ N, we should have A(p,q)c M for peP.
The set M would contain p non-empty disjoint sets A(p,q) where p « P, whence
it immediately follows that p<*m, contrary to our assumption. Therefore B(q) 4 0
for q <Q. But obviously the sets B(q) for q <Q are disjoint and contained in NV. Thus
the set N contains q non-empty disjoint sets, whence it follows that q<»*n.
5. Prove without the aid of the axiom of choice a theorem of Tarski [5] (p. 309,
Theorem 65) stating that no set of power m is the sum of 2™ non-empty disjoint sets.
Proof. Let M denote a set of power m and suppose that

MSTA
A€F

where F is a family of 2" non-empty disjoint sets. The family F has 2?" subsets (sub-
families). For Fc F, let
La) Als
A€Fy

we shall then have 7'(f,)c M for Ff,c F. Since the sets A, forming the family F, are
disjoint and non-empty, we shall obviously have 7(F,) 4 T(f,) for F, cH, F,cF,
i, # F,. Thus the sets 7(F,) where F,cF form a family of 2" different subsets of
the set M. The set M of power m would thus have > 2?" > 2™ different subsets, which
is impossible. ;
It will be observed that in above-mentioned work A. Tarski stated (without
proof) several further theorems concerning the relation <*, pointing out that they
could be proved without the aid of the axiom of choice (e. g. Tarski [5], Theorems 24-28).
Of the theorems which are difficult to prove without the aid of the axiom of choice we
shall mention, as an example, the following one (Tarski [5], Theorem 27):
If m+p
<x 2m, thenp <*m.
 CHAPTER IX

DIFFERENCE OF CARDINAL NUMBERS

1. Theorem of A. Tarski and F. Bernstein. For the cardinal numbers

m and n we say that the difference m—n exists if there exists one and only
one cardinal number p such that m=n+p. We denote this number p
by m—n.
In XVI.2 we shall prove with the aid of the axiom of choice that if m
is a cardinal number which is not finite, then for the existence of the dif-
ference m—n it is necessary and sufficient that n be a cardinal number <m
(and that then m—n=m). We shall even prove that this theorem is
equivalent to the axiom of choice. In this chapter we shall give a few
theorems on the differences of cardinal! numbers which can be proved
without the aid of the axiom of choice.
The essential properties of the difference of two cardinal numbers
are simple, but their proofs without the aid of the axioms of choice are
difficult. Those properties, above all, are the following: if, for given
cardinal numbers m and n, the difference m—n exists, then there exists
also the difference m,—n for every cardinal number m, > m, as well as
the difference m—n, for every cardinal number n, <n. We shall base
the proofs of these theorems upon the following theorem of A. Tarski '):
THEOREM 1. We are able to prove without the aid of the axiom of choice
that of A and B are sets such that A~B, then there exist sets C,, C,, D,, Dz
such that
Dab O80) 2 Bad =e) ee C= De Or
a
(1.1) O,tAB~AB~D,+AB.

Proof. Let A and B be two given sets, and 4~B. Thus there exists
a (1-1) mapping ¢ of the set A on the set B. Therefore for each element xr
of the set A, m(x) is a defined element of the set B.
If, moreover, g(x)«A, then y(@) = (p(x) is a defined element
of the set B. Generally speaking, if, for a given natural number n, ~(«)

1) It was given by him without proof in a note (written in collaboration with

A. Lindenbaum); see Tarski [5], p. 301.
 § 1. Theorem of A. Tarski and IF. Bernstein 159

is an element of the set A, then g"*'(x) = p(y"(x)) is an element of the

set B. Similarly, for each element y of the set B, y—(y) is a defined element
of the set A, and if, for a certain natural number n, g-"(y) is an element
of the set B, then y-"-'(x) = y7(p-"(a)) is an element of the set A.
Let us denote by C, the set of all those elements «x (of the set A)
for which
(1.2) ee A—B and oa):e:A “ior n= 1, 2)%..;

and by D, the set of those elements y (of the set B) for which

(1.3) Ue b=*A Seas © Oe wey ise Or ae 112), ae

and let
(1.4) OC, = (A—B)-C,, D,=(B-—A)—-D,.

We shall show that the sets C,, C,, D,, D, satisfy conditions (1.1).
For this purpose it is of course sufficient to show that

(1.5) C2), and — GeeAB 4p) ea

(since from the definition of the sets C, and D, it follows that C,C A—B,
D,C B—A, whence formulae (1.4) give A—B= 0C,+0,, B—A = D,+D,,
Cie e022 6D), Ds):
For natural n, let us denote by P, the set of all elements x (of the
set A) such that

(WO), 66 AB. ok(@).6 A for b=), 20.691, ‘and, o@).<« BA,

and by Q, the set of all elements y (of the set B) such that

(ety news = A or aie for hae 1, Doss Mok. ANd o="(4y) 6 AaB.
The sets*P,,P,, .. and Q;, Q,--. are obviously disjoint.
From formulae (1.4) and the definition of the sets C,, D,, P, and Q,
it immediately follows that

(1.8) Oe Pes es and D,=4:+92+.--

From the definition of the set P, it follows that for every natural
number 7 the function (x7) is determined for x « P,. We shall show that

(i: 9.) GPCRS) eho TOL Wie 1, 2,

Indeed, let « denote an element of the set P,; according to the defini-
tion of the set P, we have formulae (1.6)..Let y = g(x); in view of (1.6)
we have yeB—A and g-"(y)=9—"(p(x)) = xe A—B. But gy) =
= p-(p(a)) = vw) e B for k=1,2,...,n—1, since if for a certain
natural p there exists an element g(x), then this element belongs to the
160 IX. Difference of cardinal numbers

set B. Thus, in virtue of the definition of the set Y, we have y « Q,. There-
fore, if we P,, then p(x) «Q,; the function p(x) is defined for # « P, and
we have g"(P,)C Qn-
On the other hand, let y denote an element of the set Q,; according
to the definition of the set Q, we have formulae (1.7). Let 7= -"(y);
in view of (1.7) we shall have w« A—B and 9"(x) = g"(p-"(y)) = ye B—A.
But g(a) = o'(pa"(y)) = oy) <A, for k= 1;2,:..,n—1, since it
for a certain natural q there exists an element y—4(x) which in virtue of
(1.7) occurs for g=1,2,...,n—1, then y-4%(a”)e« A. According to the
definition of the set P,, we thus have w« P,. Therefore, for each element y
of the set Q, there exists an element x of the set P, such that 7 = g-"(y),
whence v(x) = y; in view of we P, we thus have ye g"(P,) for y «Qn,
whence Q, C g"(P,). But, above, we have established the inverse inclusion;
therefore we have formula (1.9), q. e. d.
The function g"(x#) is of course invertible in every set in which "(z)
is a defined element (since the function g(x#) is invertible in the set A);
therefore it is invertible in the set P,.
Let us now define a function f(#) in the set C,, putting
(1.10) TC) GAO) POT) «re PRN i ke ene
Since the sets P,, P;,... are disjoint, the function f(x) will be defined
by formulae (1.10) in the whole set O,, and, in view of (1.9), we shal] have
(1.11) P20, ors Moat ee,
and since, in view of (1.10), the function f(a) is invertible in each of the
sets P, (n=1,2,...), and the sets Q,,Q.,... are disjoint, we conclude,
in view of (1.11) and (1.8) that the function f(x) is invertible in the set C,.
In view of (1.8) and (1.11) we also obtain f(C,) = D,. Therefore the func-
tion f maps in a (1-1) manner the set C, on the set D,, and we have the
formula C,~D,.
According to the definition of the set C, we have C,C A—B, there-
fore C,-AB = 0. If xe C,, then, for natural n, v(x) « A, and since also
y"(x) « B (y"(x) being a defined element), we have g”(#) « AB. Thus, for
natural n,g"(C,) is a defined set CAB. Let

(1.12) AB—
)' 9"(Q,) =R.
n=1

In view of o(C,) C AB, for n= 1, 2,..., we shall have

(1.13) AB = R+—(Cz)
+ (C5) + 9(C2) +...
whence

(1.14) C,+AB=
R+C,+ (Cs) +9°(C2) +...
 § 1. Theorem of A. Tarski and F. Bernstein 161

Now, for # « C,+ AB, let g(v)=«@ if we R, and g(x) = ¢(a) if x¢ KR.
Since, in view of (1.12), RC AB, and C,C A—B, we have RC, = 0
and, in view of (1.12), Re(C,) = 0 for n= 1, 2,... Thus, in virtue of the
definition of the function g and in view of (1.14), we have g(x) = y(2)
for r= 0,+ 9(C,)+¢(C,)+..., in view of (1.13) and observing that g(R)=
= RK while g(C,)=9(,) and g(¢(C,)) = ¢*(C,) for n= 1,2,..., for-
mula (1.14) gives
(1.15) g(C,+AB) = B+ (Cz) +9°(C2) + P( C2) +... = AB.
Since the function y is invertible in the set A and in the set C,+
+ y(C,)+¢7(C,)+...CA, we conclude, in view of g(k)=Rk, RC,=0
and Ry"(C,) = 0 for n= 1, 2,,.. and in virtue of (1.14), that the func-
tion g(x) is invertible in the set C,+AB. Thus formula (1.15) proves that
C,+AB~AB.
The proof of the formula D,+AB~AB is analogous.
Formulae (1.5) have thus been proved. Therefore the sets Cj, C,,
D,, D, satisfy conditions (1.1), which proves Theorem 1.
A. Tarski points out 1) that Theorem 1, expressed in the language
of the theory of cardinal numbers, takes the following form:
THEOREM 2. We are able to prove without the aid of the axiom of
choice that 1f m,p, q are cardinal numbers such that m+p = m-+ q, then
there exist cardinal numbers n, p,, 9, such that
p= n+7,, g= n+, m+Pp= M= M+q,.

Proof. Suppose that m+p=m+q and let M, P,Q be ee

sets such that M = m, Joey Os q. Let A= M+P, B= MO; i
view of MP= MQ= PQ=0, we shall have AB= UM, peas
B—A=Q, and, in view of m+p= m-+q and the sets M, P,Q being
disjoint, we shall have M+P~M-+Q, i.e. A~B. Let C,, C,, D, and D,
be sets satisfying Theorem 1, and let = = D, =n, O,=P, iD Zlib
=
since P= A—-B=C,+C,, Q=B-A=D,+D,, ¢,C,=D,D,=0, we
have p=n+p,, q=n+q,; and since 0,4+M= 0,+AB~AB= M~
~D,+AB= D,+M, whence 0,+M~D,+M, and 0,M=D,M=0
(because C,C P, D,CQ, MP = MQ = 0), we have p,+m=m=aq,+m
Thus Theorem 2 is proved.
EXERCISES. 1. Prove without the aid of the axiom of choice that in order that
a set A be finite in the sense of Dedekind 2) it is necessary and sufficient that for every
set B such that A~B, we have A—B~B—A.

1) In the above-mentioned note (Tarski [5], p. 301, Theorem 6) given without

proof. The proof is given by A. Tarski [11], p. 85, and also in his book [13], p. 18, The-
orem 2.6.
*) 4. e. that the set A be not equivalent to any proper subset of itself (see III.8).
Cardinal and ordinal numbers
162 IX. Difference of cardinal numbers

Proof. Let us suppose that the set A is finite in the sense of Dedekind, and let B
denote a set such that A~B. In virtue of Theorem 1 there exist sets C,, O,, D,, Dz
satisfying conditions (1.1). We shall show that C, = 0. Suppose that C, 4 0. In view
of (1.1) we have 0,+ AB ~AB, and since O,:AB = 0, we obtain AB = A,+ A, where
A,~AB, A,~O,. Hence A = (A—B)+AB= (A—B)+A,+A,, where the sets 4—B,
A,, and, A, are disjoint. But, in view of A,~AB, we have A; = (A—B)+4,~(A—B)+
+AB= A. Therefore A = A,+ A, and A~A,. If we had A,=0, the set A would
be equivalent to its proper subset A;, in contradiction to the assumption that it is finite
in the sense of Dedekind. Therefore A, = 0 and, in view of A,~C,, also OC, = 0, q.e.d.
It can be proved in a similar way that D, = 0. Thus, in view of (1.1), dA—B= (0,,
B—A =D,, therefore A—B~B—A. Thus we have proved that the condition in our
exercise is necessary.
Now let us suppose that the set A is not finite in the sense of Dedekind, 7. e. that
it is infinite in the sense of Dedekind. It follows then (without the aid of the axiom
of choice) from III.8, Exercise 1, that the set A is equivalent to the set B= A- {p},
obtained by joining to the set A an element p not belonging to A. We shall have of course
A-—B=0, B—A = {p}, therefore 4—B non ~B-—A. Therefore if we have A—B ~
~B—A for every set B such that A~B, then the set A must be finite in the sense of
Dedekind. Thus we have proved that the condition of our exercise is sufficient.
2. Prove without the aid of the axiom of choice a theorem of A. Tarski stating
that in order that a cardinal number n satisfy the condition n 4 n+ 1 it is necessary
and sufficient that for any cardinal numbers p and q the formula n+p =—n-+q be
equivalent to the formula p = q (Tarski [12] p. 82).
Proof. Let us suppose that n is a cardinal number such that n 4 n+ 1, and that
for certain cardinal numbers p and q we have n+p = n+q. Let A denote a set of pow-
er n. If the set A were not finite in the sense of Dedekind, then it would be infinite
in the sense of Dedekind, and it follows from Exercise 1 in III.8 that the set A would
be equivalent to the set A, = A+ {p}, obtained by joining to the set A one element p
not belonging to A. But, in view of A =n, we have A, = n+ 1, and, in view of AW~A,,
we should have n =n+1, contrary to our assumption. Thus the set A must be finite
in the sense of Dedekind.
Now let C denote a set of power p, disjoint with respect to the set A. Thus we shall
have A+C =n+p=n-q, therefore A+C—=B+D where Pane’ D = gq, BD 0.
The set A is finite in the sense of Dedekind, therefore, according to Exercise 1, and in
view of A~B, we have 4—B ~ B—A. But it can easily be verified that, in view of
AC=BD=0, we have C=COD+(B—A), D=OD+(A—B), the sets OD, A—B,
B—A being disjoint. Hence, in view of A—B~ B—A, we obtain O~D, therefore
p =q. Thus we have proved that the formula n+ p = n+q implies the formula p = q.
Clearly the converse is also true. Therefore the formulae n+p =—n-+q and p =q are
equivalent.
On the other hand, if for any cardinal numbers p and q the formulae n+ p = n+q
and p = q are equivalent, then we cannot have n = n-+ 1, since it would follow hence
that n4+1—n+ 2, thus 1= 2, which is impossible.
Thus we have proved A. 'Tarski’s Theorem (without the aid of the axiom of choice).

As an easy conclusion from Theorem 2 we obtain the following theo-

rem of F. Bernstein:
THEOREM 3. We are able to prove without the aid of the axiom of choice
that if m and q are cardinal numbers such that m+m = m+ q, then m> q.
 § 1. Theorem of A. Tarski and F. Bernstein 163

Proof. In virtue of Theorem 2 (for p= m) there exist cardinal num-

bers n, py, q, Such that

Hw=—=— NED og= n+ ay i+ p.m

= m-kay,
whence
mM=M+q=—=n+pPit+qa=—qtquzr q,
eG. acl.

2. Theorem on increasing the diminuend. We shall prove

THEOREM 1. We are able to prove without the aid of the axiom of choice
that if for the cardinal numbers n and p the difference n—p exists, then for
any cardinal m the difference (m+n)—p exists and is equal to mM+(n—p)
(Sierpinski [34], p. 120).
Proof. Let n and p be two given cardinal numbers and suppose
that the difference n—p exists. Thus we have n> p, therefore m+n >p
and, according to Theorem 1 of VIII.1, there exists a cardinal number gq
such that m+n=p+q'). From this formula it follows that there exist
disjoint sets M and N and disjoint sets P and @ such that M =m, N= n,
Pa, Qe= Gand ee P+Q. In view of n>p, there exists a set
P,CN such that [ves =p. Thus we have P~P,, and, in virtue of Theo-
rem 1, there exist sets C,, C,, D, and D, such that P—P, = 0,+0¢,,
Pee), De, OC, == 0 == D Ds, BCi~D,, (Ch RP PPD RP.
Let f denote a (1-1) mapping of the set C, on D, (which exists in
view of 0,~D,), and g a (1-1) mapping of the set 0,+PP, upon PP,
(which exists in view of C,+PP,~PP,). From P,C N, we have PN =
= PP,+(P—P;)= PP} +C,N+C,N. But, in view of PP,+C,~PP,
and PP,C PP,+0C,N C PP,+0,, we have PP,+C,N ~PP,. Since (PP,+
CN) CN = 0 because: .P,C, = 0 and (C,0, = 0), we-have PN~PP, +
+C.N.
Now we shall prove that the set S= PN+/(C,M) is of power p.
Indeed, in view of {(0C,M)CD,C P,—P, the sets PN and /f(C,M)
are disjoint. But /(C,M)~C,M =0,—C,N (because C,C PC P+Q =
= M+N and MN=0O), and (C,+C,N)(PP,+0C,N) = 0. Ineview of PP, ~
~PP,+D,, PP,(C,+D,) = 0, (PP, +D,)D, = 0, we have
S = PN +}(0,M)~PP,+¢,N +(€,—0,N)
SEPP CE SP PAE DP Pet DEED 2 -PP. a Pyar) =P...
Thus S~P, and, in view of P, =p, we have Se p. Now we shall
prove that also the set 7 = g(PP,)+D,+D, is of power p.

1 In case of m+n=p we should assume g = 0.

LES
164 IX. Difference of cardinal numbers

Indeed, in view of g(PP,)C PP,, PP,(D,+D,),= 0, D,D, = 0, the

components of the sum 7 are disjoint and, in view of g(PP,)~PP, (be-
cause the function Lg igs invertible in 0,+.PP,, and thus in PP,), we ob-
tain T = g(PP,) +D,+D, = PP,+D,+D,=— PP,+D,+D,= P, =p.
SIMeer Ss GANee GANS S2T= p and in view of the assumption that
there exists the ainerencs t—p, the two sets R, = N—T and k, = N—S
are of power n—p and there exists a (1-1) mapping h of the set R, on R,.
Since C,PP, = 0, and g being a (1-1) function invertible in the set C,+-
+PP,, we find g(C,)g(PP,) = 0. In view of g(C,) C PP, and PP,(D,+
+D,) = 0, we have g(C,)-(D,+D,) = 0. Therefore g(C,)-T = 0, whence
in view of g(C,)C PP,CN, we obtain g(C,)C R, and naturally
g(C,M)
C Ry; thus there exists a set h(g(C,M)) C R.
Let. P= PN +f(0,M) + h(g(C,N)) = S+h(g(C,M)). Since. Sip= 0,
whence Sh(g(C,M)) = 0, and h(g(C,M))~C,M (the function g being
invertible in the set C,MC O,+PP, and the function h being invertible
in the set g(C,M)C R,), and since SC,M=0 (in view of SCN and
MN = 0), we have P, = §+-h(g(C,M))~S+0C,M.
But S= PN +f(C,M) I
PP, (since P,CN); thus S+0,M = (S—PP,)+[PP,+(¢.M —S8)],
and since or course PP,C PP,+(0,M—S8)CPP,+0, and 0,+PP,~
~PP,, we have according to Theorem 3 of II.6, PP,+(0C,M—S)~
~PP,. Adding on both sides the set S—PP,, disjoint with the sets form- .
ing those sides, we obtain S+C,M~S, and since we have proved that
P,~8+0,M, we conclude that P,~S, and thus P,=p. But in view
of P,C N, we have the decomposition into disjoint components N =
= P,+(N—P,) whence, in view of N =n and P,=p, n= po Ne bee
and since the difference n—p exists, we have N—P,—=—n—p. In view
of P,C.N and MN=0, we have MP, =0, therefore (I+N)—P,=
= M+(N—P,)
= m+(n—p).
It is easy to verify that we have the following decompositions into
disjoint components: P= PN+0O,M+C,M and P,= PN+/(C,M
+h(g(C,M)).
The set#®O,M+C,M and f(C,M)+h(g(0,M)) are disjoint because
the first is CM and the second C N, while MN = 0. Thus we have

P,—P = f(C,M)+h(g(C,M))~0,M+0,M
= P—P,,

therefore P,—P~P—P,. Adding to both sides of this formula the set

(M+N)—(P+P,), disjoint with respect to each of these sides, in view
of the equality (M+N)—P= [(M+N)—(P+P,)]+(P.—P), and (M+
+N)—P, = ((M+N)—(P+P,)|+(P—P,) (since P+P,C M+), we find
 § 2. Theorem on increasing the diminuend 165

(M+N)—P~(M+N)—P,; the set Q=(M+MN)—P is therefore of

power m+(n—p), whence g = m-+(n—p).
Thus we have proved that there exists only one cardinal number q
such that m+n= p+q, namely q = m+(n—p). This proves the existence
of the difference (m1+1n)—p being equal to the cardinal number m+
+(n—p). Therefore we have proved Theorem 1.

3. Theorem on increasing the subtrahend. We shall prove

THEOREM 1. We are able to prove without the aid of the axiom of choice
that if for the cardinal numbers m,n, p the difference m—(n+p) exists,
then the difference m—n exists also and is equal to the number [m—(n+p)]+p
(Sierpinski [37], p. 121).
Proof. Suppose that m,n,p are cardinal numbers for which the
difference m—(n+p) exists. Thus we have m>n+p and m>n.
Let M denote a set of power m, NV — its arbitrary subset of power n,
and 7 — an arbitrary subset of the set M of power n+p. Thus we have
T= N,+P, where V,P=0, N, =n, P=p. Therefore N~N, and, in
virtue of Theorem 1 of IX.1, there exist sets C,, C,, D,, D, such that
N—N,= (+ ©,, N,-—N = D,+D,, C,C, = 0 = DD, O~D,, C,+ NN ~
~NN,~D,+NN,. Let f denote a (1-1) mapping of the set C, on D,,
and g a (1-1) mapping of the set C,+NN, on NN,. Let

(3.1) S, = (,+0,+P+f(PC,)+9(NN)) .

We shall show that S, = n+p. Indeed, in view of (1.15) we obviously

have the decomposition of the set 8, into five disjoint components

(3.2) 8, =P+(O,—P)+(0,—P)+f(PC,)
+g(NN,) -
' We have NN,C PO,+NN,C 0,4+NN,, and 0,+NN,~WN,, there-
fore PC,+NN,~NN,; since g(NN,)~NN,, we have g(NN,)~PC,+
+NWN,. On the other hand, we have f(PC,)~PC,, and since both the
left-hand and the right-hand sides of the last two formulae are disjoint,
being also disjoint with respect to the set (C,—P)+(C,—P), we have

(C,—P) + (C,—P) + f(PC,)

+ 9(NN1)~
~(0,—P) +(0,—P)+P0,+P0,4+NN, =
= (4+0,+NN, =(N—-N,)+NN,=N.
Thus formula (3.2) proves that the set S, is the sum of two disjoint
sets one of which is P and the other ~N. Hence it follows that S,=
=P+N=p-+n, q.e.d.
Now let
(3.3) S.= N+P+f(PQ).
166 IX. Difference of cardinal numbers

In. view of (3.3) and of the equality N—N, = C,+C,, we have

8, = NN, +(N—W,)+P+f(PG,) = NN, 40,40,4P4+f(PG),

and, in view of C,+C,= N—N,, PN, = 0, f(PC,) CD, C N,—N, we have

(3.4) NV,(C,+0,P
+ f(P0,)) =0.
In view of NN,~g(NN,)C
NN, and (3.1) we thus have

By — NN
Cy C3 PCE Cyr (NN Cae Cai ee at (aCe a
whence S,~S,. The sets S, and S, are both of power n+p and, in view
of S,C M, S.C M and the existence of the difference m—(n+p), we have
M —S,~M—S,. Let h denote a (1-1) mapping of the set M—S, on
M—S8S,. We have C,-NN,=0 and, since the function g is invertible
in the set NN,+0,, we find that g(NN,)g(C.)=0 and naturally
g(NN,)g(PC,) = 0. On: the other hand, in view of g(PC,)C NN, and
(3.4), we have g(PC,)[C,+C,+P+f(PC,)] = 0. In view of (3.2) we have
g(PC,):S,= 0, whence g(PC,)C M—8, which proves the existence of
a set, h(g(PC,)) C M—S,. Let

(3.5) P, = (P=N)+f(PC,) +h(g(PC,)) .

Since in view of (3.4) we have P—N CS,, f(PC,)C 82, and since
h(g(PC,)) C M—S8, and f(PC,)CN, formula (3.5) gives a decomposition
of the set P, into three disjoint components. But /(PC,)~PC,, h(g(PC,))~
~PO,, the sets P—N, PC,, PC, being disjoint. Thus

PUP N Ure PC, = (P= Ny ep Ne Ps

therefore P, = p. But since f(PC,) C D,C N,—WN and h (g(PC,)) C= Sac
C M—N we have, in view of (3.5), P.N=0. Thus N4+P,=N+P=
=n+p, and since V+P,C M, M=, the existence of the difference
m—(n+p) being assumed, we have M—(N+P,)=m—(n+p); hence,
in view of M—N=[M—(N+P,)|+P,, M—N =[m—(n-+p)]4+p, the
subset NV of power nof the set M being arbitrary, this proves the existence
of the difference m—n and equal to the number [m—(n+p)]+p. Thus
we have proved Theorem 1.
COROLLARY 1. If one of the numbers m—(n+p) and (m—n)—p
exists, then the other number also exists and the two numbers are equal
(Tarski [5], p. 307, Theorem 49, without proof.)
Proof. According to Theorem 1, if the number m—(n+p) exists,
then the number m—n also exists, and m—n=[m—(n+p)]+p>p.
 § 3. Theorem on increasing the subtrahend 167

Let us assume that m—n=p-+q; therefore m=n-+p+q and, in

view of the existence of the difference m—(n+p), we shall have q=
=m-—(n+p). This proves the existence of the difference m— (1t—p)
and its being equal to m—(n+p). On the other hand, if the number
(m—n)—p = q exists, then m—n=yp+q and m=n+p+q. And if
m=n+n-+a,, then, in view of the existence of the difference m—n,
we have p+ q, = m—n, whence, in view of the existence of the difference
(m—n)—p, we obtain q, = (m—n)—yp, which proves the existence of the
difference m—(n+p) and its being equal to (m—n)—pP.
In this way Corollary 1 is proved (without the aid of the axiom of
choice).
COROLLARY 2. If the numbers m—n and n—p exist, then the number
m—(n—p) also exists and is equal to (tt—n)+p (Tarski [5], Theorem 50).
Proof. Let us suppose that the numbers m—n and n—p exist. Since
the difference n—p exists, we have n= (1—p)+p.
Now replace, in Theorem 1, n by n—p. The condition of Theorem 1
will then be satisfied since m—[(n—p)+p]= m—n and we have assumed
the existence of the difference m—n. Thus in virtue of Theorem 1 (re-
placed in it n by n—p) there exists a number m—(n—p) equal to
(m—[(n—p)+p]) +p = (m—mn)+p.
Corollary 2 has thus been proved (without the aid of the axiom ef
choice).

4. Difference in which the subtrahend is a natural number. We shall prove

THEOREM 1. We are able to prove without the aid of the axiom of choice
that if n is a natural number and m a cardinal number >n, then there exists
the difference m—n.
Proof. Let » denote a natural number and m a cardinal number >.
Thus there exists a cardinal number p, such that m= n+p. Now sup-
pose that q is a cardinal number such that m=n-+q. Let M denote
a set of power m. In view of m= n+p = n+ q, there exist sets N and N,
of n elements each, and sets P and Q such that NP = 0 = N,Q, P=»,
C= q and M= N+P= N,+@Q. In order to prove Theorem 1 it suffices
to show that q=p, ¢. ¢. that Q~P.
In view of M= N+P=N,+9Q we have the following decomposi-
tions ofe th sets P and Q into two disjoint components:

P= M—-N=[(M—(N+N)]+(%—Y),
Q= M—N, =(M—(N+N,)]+(¥—™,).
Therefore, to prove that P~Q it is sufficient to show that V,—N~
~N—N,. Now we have Vj—-N=W,—-NN,, N—-N, => N—NN,. The
 168 IX. Difference of cardinal numbers

sets N and N, have n elements each; let k denote the number of elements
of the set NN,: it will be a natural number or 0. Clearly the sets V,—NN,
and NV,—NN, will have n—k elements each, 7. e. they will be equivalent,
whence VN,—N~N—N,, q.e. d.
Thus we have proved Theorem 1.
CoroLLARY. If, for the cardinal numbers m and n, we have, with a cer-
tain natural n, M+n=n-+n, then M=nN.
Proof. Let p= m+n; we shall have also p= n-+vn. If m is a finite
cardinal number >0, then, p= m+n > n, which is of course true also in
the case of m not being a finite number (since » is a natural number).
Thus we always have p > n, and, in virtue of Theorem 1, the difference
p—n exists. But in view of p=m+n=n-+n we have p—n=m and
p—n=n, whence m—=—n, q.e. d.
Thus we have proved our Corollary (without the aid of the axiom
of choice).
It will be observed that in our Corollary the number n cannot be
replaced by s, since, for instance, 1+) = 2-++,), but we are able to
prove without the aid of the axiom of choice that if m and n are cardinal
numbers >s, and if m+%,—=n+s,, then m=n (since as we have
proved in Exercise 5, VIII.1, if m >x,, then m+, =m). On the other
hand, we are not able to prove without the aid of the axiom of choice
that if m and n are cardinal numbers such that m >, and n is not a finite
number, and if m+s,=n-+s,, then m=n. For it would easily follow
from such a theorem, without the aid of the axiom of choice, that every
cardinal number n which is not finite is >s,. (Indeed, taking n+», = m,
we should have m+s,=m->s, and n+s,=tm-+s8,, whence if our
theorem were true, then since n is not a finite number and m>x,, we
should have n= m= M+, > >).
EXERCISE. Prove without the aid of the axiom of choice that if m and n are
cardinal numbers such that n An-+1 and m>nNn, then the difference m—n exists.
Proof. In view of m > n there exists a cardinal number p ¥ 0, such that m = n+p.
Now let q denote any cardinal number such that m=n-+q. According to Exercise 2
(section 1) it follows hence without the aid of the axiom of choice that p = q. Thus
there exists one and only one cardinal number p such that m= n-+p. This proves the
existence of the difference m—n = p.

5. Proof of the formula 2™—m = 2™ for m >, without the aid of the
axiom of choice. We shall prove
THEOREM 1. We are able to prove without the aid of the axiom of choice
that if m is a cardinal number >, then
2" —m = 2™
(Tarski [5], p. 307, Theorem 56, without proof).
 § 5. Proof of the formula 2% —m = 2™ 169

Proof. (Sierpinski [32], p. 113). To begin with, we shall prove

without the aid of the axiom of choice the following
LemMMA. Jf m,p,$ are cardinal numbers such that 2" = m-+p and
m—=—m-+s, then p > 2°.
Proof of the lemma. Let M denote a set of power m. Since 2" =
= m-+py, the set 7 of all subsets of the set M is the sum of two disjoint
sets, M, and P, such that M, =m and P =p. On the other hand, since
m= m-+s, the set M is the sum of two disjoint sets M, and S such that
M,—=m and S=s. Thus we have M~M,~WM, and therefore there
exist a (1-1) mapping f of the set M on M, and a (1-1) mapping g of the
set M, on M. Now for X CS let

(5.1) n(X)
= X+ Ei[we My, x ¢ f(g(X))]
(F, [W (a)] denotes the set of all elements x satisfying the condition W(« Me
If, for a certain set X C 8, we had h(X) « M,, then, in view 7 Me
= f(M), there exists an element ¢ of the set M such that h(X) = f(t).
But, in view of te M and M = g(M,), there exists an eee ce M,
such that t = g(x). Therefore we should have h(X) = f(g (@)) where we M,.
In view of «e M,, Xe S and M,S=0, we have «¢X and if we h(X)
formula (5.1) gives « ¢ f(g(#)) = h(X), which gives a contradiction; now,
if x¢ h(X), then in view of we M, formula (5.1) gives x « f(g(x)) = h(X),
which again gives a contradiction. Thus the assumption that for a certain
X C S we have h(X) « VM, leads in any case to a contradiction. This proves
that h(X)¢ M for X C 8S, and since, in view of (5.1), we have h(X)« T=
= M,+P, h(X)« P for XC 8. Now since in view of M,8 = 0, formula
(5.1) implies immediately that h(X,) ~ h(X,) for X,C 8, X,C 8, X, ~ X,,
the function h(X) is invertible in the set U(S) (of all subsets of the set
and its values are elements of the set P. Hence it follows that U(8) < wee
1. e. that p > 2°, q. e. d. Thus we have proved our lemma without the
aid of the axiom of choice.
Proof of Theorem 1. Let m denote a cardinal number >ys,. As
we know, 2™ > m; thus there exists a cardinal number p such that 2" =
= m-+p. In view of m>ys, we have, as we know, m+1—m and thus
2 he ee em, and since 2™-m2> 2", we obtam: 2" =—
=2™1m. Thus we have m+p = m-+2™. In virtue of Theorem 2 of IX.1,
there exist cardinal numbers n, p,, q, such that

Peerhis, ft aes a Maas

SM > My
In view of 2™"=m+p and m=m-+q,, we have from our lemma
p > 2% > q,. Therefore p > q, and since p= n+p, >n, we have p+p >
170 IX. Difference of cardinal numbers

>n+q = 2 = 2™4, whence p+p > 2"41 = 2"

4 Om — Om Lm tp SQM 4
+p>2"1p>ptp, ie p=p+2™. Hence we conclude, in virtue of
Theorem 3 of IX.1, that p>2™, and since 2" = m-+p2>p, we have
Din
Thus we have proved that there exists only one cardinal number p
sueh that 2"—m-+p, namely p= 2™. This proves the existence of
the difference 2"—m and its being equal to 2™, which proves The-
orem i;
It will be observed that we are not able to prove without the aid
of the axiom of choice the theorem which would be obtained from Theo-
rem 1 by replacing the condition m > ys, by the condition that m be a car-
dinal number which is not finite.

6. Quotient of cardinal numbers. Similarly to our definition, in § 1, of

the difference of two cardinal numbers we can say that the quotient of the
cardinal numbers m:n exists if there exists one and only one cardinal
1umber p such that m= np. In XVI.2 we shall prove with the aid of
the axiom of choice that if m is a cardinal number which is not finite,
then in order that the quotient m:n exist it is necessary and sufficient
that n be a cardinal number <™m (and that then m:n=m). We shall
even prove that this theorem is equivalent to the axiom of choice.
However, in certain particular cases we are able to prove the existence
of the quotient of the cardinal numbers m: n without the aid of the axiom
of choice. £. g. it is easy to prove without resorting to the axiom of
choice that if n is a natural number, then there exists the quotient x): =
=). (For if p is a cardinal number such that s,=n-p, then x) >p,
and we cannot have x, > p since then n-p would be the product of two
natural numbers, 7. e. a natural number, contrary to our assumption
that s,= ”-p. Therefore p = s,, q. e. d.). But the proof of the existence
of the quotient ¢: 2, although it can be carried out without the aid of the
axiom of choice, is difficult.
We are able to prove without the axiom of choice a more general
theorem stating that for every cardinal number m and every natural
number & exists the quotient (km): k =m. Obviously this theorem is
equivalent to the following one:
Tf, for a natural number k and cardinal numbers m and n, we have
Fert Tet, eh tt = 11.
This theorem was proved without the aid of the axiom of choice
for k = 2 in 1905 by F. Bernstein, who also outlined a proof for natural k.
(Bernstein [1], p. 122; also Sierpinski [8], p. 1). A detailed proof for na-
tural & (without the aid of the axiom of choice) was published by A. Tar-
ski [12] (p. 77), who proved also (without the aid of the axiom of choice)
 § 6. Quotient of cardinal numbers : el:

that if for a natural number k and cardinal numbers m and 1 we have

km < kn, then m <n (for k = 2 see also Sierpinski [31], p. 148).
In connection with the existence of the quotient (km): k for natural k
and cardinal m it will be observed that by means of the axiom of choice
we can prove that the quotient (Am):m does not exist for any natural
number k and any cardinal number m which is not finite. Without the aid
of the axiom of choice we are able to prove that the quotient (ks): No
does not exist for any natural number k, since for any natural k and l
we have ks, = Is). Similarly we can prove without resorting to the axiom
of choice that the quotient (kc): ¢ does not exist for any natural number k.
Let us also mention the following theorem, which we are able to prove
without the aid of the axiom of choice:
If k and | ave relatively prime natural numbers and m and n cardinal
numbers such that km = In, then there exists a cardinal number p such that
m=lp and n= kp.
This theorem was published without proof by A. Lindenbaum in
1926; the proof was published by A. Tarski [12] (p. 90, Theorem 13)
in 1949. In particular, we deduce from this theorem (for k= 2,1 = 3)
the following corollary:
If a cardinal number is divisible by 2 and by 3, then it is also divisible
by 6.
The proof of this corollary without the aid of the axiom of choice
is rather difficult.
 CHAPTER X

INFINITE SERIES AND INFINITE PRODUCTS

OF CARDINAL NUMBERS

1. Sum of an infinite series of cardinal numbers. Let

(ae LULPAGe LLLey pce

denote a given infinite sequence of cardinal numbers. For every given

natural n there exists a set (at least one) of power m. As far as I know,
Kuratowski and Mostowski have been the first to point out that it is
not easy to prove with the aid of the axiom of choice the existence of
an infinite sequence of sets
(1.2) Ma ee Vigne
such that

(1.3) U.=m, for n=1,2,3,..

(see Kuratowski and Mostowski [1], p. 125 and p. 227).
For we cannot assert here that for every natural number n there
exists the set of all sets of power m,, since such an assumption (and even
the assumption of the existence of the set of all one-element sets) leads
to antinomy (see V.4).
This difficulty may be avoided by assuming temporarily that the
infinite sequence of cardinal numbers (1.1) satisfies the condition W
according to which there exists an infinite sequence of sets (1.2) satisfying
conditions (1.3), and proving later, with the aid of the so-called theorem
of Zermelo (following from the axiom of choice), that every infinite se-
quence of cardinal numbers (1.1) satisfies condition W. Another way of
avoiding the above-mentioned difficulty consists in resorting not to the
axiom of choice but to the logical axiom of Hilbert, discussed in VI.3.
Namely, for every cardinal number m let us denote by P», the property
of an object consisting in its being a set of power m. If m is a cardinal
number, then there exists a set M of power m; thus for every cardinal
number m there exists at least one object with the property P,,. In virtue
of Hilbert’s axiom there exists a function « such that e(P,) is a set with
 § 1. Sum of an infinite series 173

the property P,,. Thus, in particular, if (1.1) is a given infinite sequence

of cardinal numbers, then M, = ¢(Pym,) (n= 1, 2,...) will be an infinite
sequence (1.2) of sets satisfying conditions (1.3).
Further, we can suppose that the sets (1.2) are disjoint; if they were
not, we could replace each set M, by the set of all ordered pairs (p, 7)
where pe M,.
Let
S= M,+M,+...
and
(1.4) S=s.
Further, let Mz, M3, ... denote any infinite sequence of disjoint sets
such that
(1.5) M Silly ytOl geet acd yD ce

By (1.3) and (1.5) we shall have

om OT. aan Le inci

Let

S’= Myj+Mo+...
According to Theorem 8 of VI.7, we shall thus have S’~S which,
by (1.4), gives S’ =s.
Thus we have proved that to every infinite sequence (1.1) of cardinal
numbers corresponds a well-defined cardinal number $ such that if (1.2)
is an arbitrary infinite sequence of disjoint sets satisfying conditions (1.3),
then $ is the power of the sum of all sets of the sequence (1.2). Such a car-
dinal number ¢ is called the sum of an infinite series of cardinal numbers
m,t+m,+... and written

$=MmM+m+... or > m,=s.

n=1

From the fact that the sum of an infinite series of sets is independent
of the order of the components and of the manner of joining them im-
mediately follows the commutativity and associativity of the sum of an
infinite series of cardinal numbers. Thus we can say that
Hvery infinite series of cardinal numbers has a well-defined sum, which
is a cardinal number, and this sum does not depend on the order of the com-
ponents or on the manner of joining them.
From the definition of the sum of an infinite series of cardinal num-
bers it immediately follows that if M,, M,.,... is an infinite sequence of
disjointed sets, then writing
S= M,4+WM,+4+...,
 \

174 X. Infinite series and infinite products

we shall have
6S
> 307 -.
EXERCISE. Prove that if 8 is a set such that g= m,+m,+..., then there exists
an infinite sequence M,, M,,... of disjoint sets Such that u, SS tt WOM We Nh P45 cons
and S = M,+ M,+ ...
Proof. Suppose that S is a set such that Ts m,+m,-+... For an infinite sequence
of cardinal numbers m,, m,, ... there exists, as we know, an infinite sequence of disjoint
sets M’? (n=1, 2,...) such that Mw Sith, AONE Wy) Se Pa, coe) Aner SS UE US
we shall have §’= m,+m,-+-..., and therefore S’ = S, which proves that
Thus 8’~S.
there exists a (1-1) correspondence gm between the elements of the sets S’ and 8. Writ-
ing 9(M/) = M, for n= 1, 2,..., we shall obviously obtain sets satisfying the condi-
tions of our exercise.

2. Properties of an infinite series of cardinal numbers. It is easy to prove

that infinite series of cardinal numbers whose corresponding components
are equal give identical sums. It is also easy to prove that the sum of an
infinite series of cardinal numbers is > any component of the series and
that if we have two infinite series of cardinal numbers, m,+m,+... and
it + Me--.., SUCh that im, <1... for #— 1,2, :.., then

Tg fit ret
de oe
For let NV, (n= 1, 2, ...) denote an infinite sequence of disjoint sets
suchethat: N,)=sst, tore 4542... .Sinee myo; dorms by 2.8) there
exists for every natural number » a set VM, such that M, C NV, and M, —
=m,. Let S= M,+WM,+..., S’= N,+N.+... Of course we shall have
SCS) whence S =< 8/,-and since 8 =m, m,+...) 8’ =n, fn, +..., we
have m,+m,.+... <<m,+1,+..., q.e. d.
We can also easily prove that for any cardinal numbers the ine-
quality
(2 L) ie ite, Migs ae

implies the inequalities

(2.2) Nee ie ON crete Ten oy eea ds Le,

It is much more difficult to prove the converse: that inequalities

(2.2) imply inequality (2.1) (7. e., that the sum of an infinite series
of cardinal numbers is the least cardinal number that is not less than
any partial sum of a series) (Tarski [11], p. 79); it is difficult to prove
even that inequalities (2.1) exclude the inequality m<m,+m,-+... It
will be observed that A. Tarski has proved that the latter inequality
implies the existence of a natural number n such that m < m,+m,+...+
+m, and that this theorem is equivalent to the axiom of choice.
 § 2. Properties of an infinite series bi Ot

As regards other properties of infinite series of cardinal numbers,

we have the formulae

(2.3) m(m, +m,+...) = mm,+mm,+...

and
co CO Cc co co

prey Ns > My = Oe Wine Wepet eee ela),

k=1 [=1 @=1 ket n=1

their easy proof is left to the reader. Thus infinite series of cardinal num-
bers behave in the same wayias absolutely convergent series of real numbers.

3. Examples of infinite series of cardinal numbers. As an example of an

infinite series of cardinal numbers let us consider the series

M+Ng+...,

where n, (k = 1, 2,...) is an infinite series of natural numbers. Let s,= 0,

Si Ny Mo... for k= 1525... and Ny = {Sp
4 +1,:Sp-1-2, 5
Sx_1+m} for k=1,2,... Obviously the sets N,, N., ... will be disjoint.
N,~= for k=1,2,..., and N= N,+N,+... will be the set of all
natural numbers, 7. e. a set of power 8,. Hence

Syed = WN Na et et + on as
therefore

(3.1) ; So=%44+%7)4+.

for every infinite sequence of natural numbers 2, ”,,... In particular,

from formula (3.1) it follows that

(3.2) sett ER eat Ste oe

but also

as well as

These formulae prove that if we have two infinite series of cardinal

numbers such that any component of the first series is smaller than
the corresponding component of the second series, then the sum of the
first series is not necessarily smaller than the sum of the second series.
As a particular case of formula (2.3), where m,=1 for k= 1, 2,...
we obtain, by (3.2), for every cardinal number m, the formula

(3.3) Sym = m+m+m-+... ;

176 X. Infinite series and infinite products

in particular for m= sp, in view of the formula x8) = %>, formula (3.3)
gives
No a No +No + No +... 9

and for m= c, in view of the formula s,c =c, formula (3.3) gives

C= Usa
Cop coo ¢

From Theorem 15 of VI.7 it immediately follows that if

Cait, “Santa,

then there exists at least one natural number » such that m,=c. In
particular, for m, = m, n= 1, 2,..., we obtain in view of formula (3.2),
the conclusion that if m is a cardinal number such that c=s,m, then
m= c. We have obtained this conclusion with the aid of the axiom of
choice, since the definition of the sum of an infinite series of cardinal
numbers makes use of that axiom and, besides, we have used theorem 15
of VI.7, which was proved with the aid of the axiom of choice. Without
the aid of that axiom we cannot prove that if c=s,m then m=c. On
the other hand, we can prove without the aid of the axiom of choice
(although it is difficult) that if » is a natural number and m a cardinal
number such that c= nm, then m= cc (Sierpinski [35], p. 32 and Sier-
pinski [31], p. 154).
From Theorem 2 of V.5 it immediately follows that if m,, mb), ...
is an infinite sequence of increasing cardinal numbers, 7. e. if m, < mM, <...,
then. 1,-- Wt, = 1, for m= 1,2, .

EXERCISES. 1. Prove that if, for cardinal numbers, we have p+gq =m,+m,+
+m,-+..., then there exist infinite sequences of cardinal numbers Pj, p,, ... and q,, Jo,-- »
such that p=p,+P.4+..., G=G+4+.., and m,=—p,+q, for n= 1, 2,...1).
Proof. Since p+q =m,+m,-+..., it follows from the axiom of choice that there
exists an infinite sequence of disjoint sets M,, M,, ... such that M, =m, forn = 1, 2,...
and that there exist disjoint sets P and Q such that P=p, Q =q and P+Q= M,+
+ M+...
Let P,=PM,, Q, = QM, for n=1,2,... Obviously we shall have P= P,+
Pot, O' = O10... =, EO, tor, n= 1,2, .3, the sete Pp kse endl
Qz,-.. being disjoint. Let p, = P,, q, = 2, for n = 1, 2,... It can easily be proved that
the sequences p,,P,,... and q,,q4,,... will satisfy the required conditions.
2. Prove that if m,,m,,... and ,,n,,... are two infinite sequences of cardinal
numbers such that m,=n,+m,,, for n=1,2,..., then my >n+n,+mj+...%).

1) This is the refinement postulate of A. Tarski. See his book [13], p. 4.

*) See Tarski [11], p. 88, Lemma B. Cf. also Remainder of infinite chain postulate
of Tarski [13], p. 4.
 § 3. Examples of infinite series of cardinal numbers ser

Proof. Let m,,m,,... and n,,1,,... be two infinite sequences of cardinal numbers
such that m, = n,+- M44 for n= 1, 2,... As we know (see IX.1), in that case there
exist infinite sequences of sets M,, M,,...,.N,, .N,,... and M’, M!,... such that NM, =
iy Mita NM 0, Me NM fort = 1,2, .,. Since My a=
=™..= M! for n= 1, 2,..., we have M,1y~M, for n= 1, 2,..., and it follows from
the axiom of choice that there ex-sts an infinite sequence ¢, (n = 1, 2,...) where g,, is
a mapping of the set M,,, on the set ¢,(M,,,) = M,. Since M, = N,+ M,, we have
M,, = N+ On( M41) for m= 1, 2,..., and thus Mf, ——N7-49¢,(M,), M, = No--@, (M5), -::
Let us write py, = 9 and Y,11 = YnPn4, for n= 1, 2,... Thus we shall have

M, = N,+ 9,(M,) = N+ (Ne)

+ po( Ms) = -.. = Nit vile)
+ ya(Ns) ++

+ Yn—a(Na)+ Yn(M1) for n=1,2,...

But y, 1s obviously a mapping of the set W,,, (for n= 1, 2,...) and it is easy to
prove by induction that the sets V1, y,(N2), Po(N3), «+> Ya_y(Na)> y, (M41) are disjoint.
het Ny wy, CN, ) for nm —=.2,3,.-.; we shall have Ni == 11, 1OLN) leu 2,...5) Les SOLS
forming the infinite sequence N,, Vi, N/,... being disjoint and contained in M,. Hence
Moa NENT
EN + and m, = Mf, > 1,4 np t..., q.e. d.
3. Prove that in order that a cardinal number s be the sum of an infinite series
_ of cardinal numbers m,+m,+-... it is necessary and sufficient that the following two
conditions be satisfied:
1) there exists an infinite sequence of cardinal numbers n,, n,, ..., such that n, = s
and nn, =™,+T, 1) TONE Pr aes Ul Bis, goons :
2) if py, P., ... is an infinite sequence of cardinal numbers such that p, = m,-+p, 41
for 715 2)..., then s<p, (Tarski [13]; p. 11, Theorem 1-36).
Proof. Suppose that s=m,+m,+... and let n,=m,+m,,,+M,,.+.. It is
easy to verify that the sequence nj, n,, ... satisfies Condition 1 (since m,+ Ta Mot
+..=m,+(m,.,+ M,, a+ ---) for n= 1, 2,...). It also satisfies Condition 2. Indeed,
if p,,P.,... IS an infinite sequence of cardinal numbers such that p,=m,+p,_, for
m=1,2,..., then, according to Exercise 2, we have p, >m,+m,+...=s, whence
$< p,. Conditions 1 and 2 are thus necessary in order that s = m,+m,+...
On the other hand, suppose that Conditions 1 and 2 are satisfied. By 1 and ac-
cording to Exercise 2, we have s =n, >m,+m,+... Now let p, = ™,,-- 11,4) + +.» for
m=1,2,...; it is easy to verify that we shall have p, = m,+ Pat LOTR eee
therefore, by 2, we shall have s < p,, 7. e. 8 << m,+m,-+... Thus s = m,+m,-+ ... Condi-
tions 1 and 2 are therefore necessary in order that s = m,+m,+ ...
From Exercise 3 it follows by Theorem 3 of VIII.2 that the concept of the sum
of an infinite series of cardinal numbers can be reduced (with the help of the concept
of the infinite sequence) to the concept of the sum of two cardinal numbers.

4. Sum of an arbitrary set of cardinal numbers. The concept of the

sum of cardinal numbers has so far been defined only for a finite and for
a denumerable set of conrponents. We shall generalize it to an arbitrary
set of components.
Let # denote an arbitrary set and suppose that with every element
of the set H is associated a certain cardinal number m;. In XVI.1 we
shall deduce from the axiom of choice that there exists a function (é),
Cardinal and ordinal numbers 12
178 X. Infinite series and infinite products

defined for é« # and such that, for « H, p(é) = Me where M; is a set

of power m,, and M;M,=0 for eH, ne H, EAN"). Let

s= > M,;
é€E

and s = S. From Theorem 9 of VI.7 it easily follows that the cardinal

number s does not depend on the choice of the function gy, provided it
is such that y(é)= M; where M;= mg, for &¢HE, and M;M,=0 for
Ee H, »e¢ H, &An. The cardinal number ¢ is called the sum of cardinal
numbers me, extended to all elements & of the set HE, and written as

(4.1) s= )' me.

g€E

It is also obvious that if formula (4.1) holds and if S is a set of power 5,

then
s= >) Me,
é€E

where M;= wm, for ée H and M,M,=0 for é« BH, ne BH, EF.
Now let m and n be two arbitrary cardinal numbers, MW a set of
power m, # a set of power n, and suppose that

s= >’ Mm,
é€E

where M;=m for 颫 H#, and M:M,=0 for eH, ne H, EAN. Now,
for é« H, let us denote by M; the set of all ordered pairs (p, €) where
pe M. We shall have Mj=m= M,, whence M:;~M, for é« E, and
of course M:M; = 0 for feH, nie H, & = 7.
Let
i= ) u:;
é€E

in virtue of Theorem 9 of VI.7 we shall have S’~S, whence S’ 16:

But obviously 8S’ = Mx# whence S’ = mn. Thus s = mn.
It follows that if we have a family of power n of disjoint sets each
of which is of power m, then the sum of all sets of that family is of
power mn.
In view of the commutativity of the product of cardinal numbers,
it follows that the sum of a family of power n of disjoint sets each of
which 1s of power m is equivalent to the sum of a family of power m of
disjoint sets each of which is of power n. We cannot prove this theorem
without the aid of the axiom of choice even for m= 2, n=).

1) Cf. the beginning of § 1 of this chapter.

 §$ 5. Infinite product of cardinal numbers Lf,

5. Infinite product of cardinal numbers. Let (1.1) denote a given

infinite sequence of cardinal numbers. In § 1, Chapter XVI, we shall
deduce from the axiom of choice that there exists an infinite sequence
(1.2) of sets satisfying conditions (1.3)1). Let

Pe OV x
and
p=.
Further, let My, Mg, ... denote any infinite sequence of sets satisfy-
ing conditions (1.5). Thus we shall have M,~M, for n=1,2,... Let

iP eae SG Ma Cae

In virtue of Theorem 12 of VI.7 we shall have P’~P, i. e. P’ =p,

Thus we have proved that to every infinite sequence (1.1) of cardinal
numbers corresponds a well-defined cardinal number p such that if (1.2)
is an arbitrary infinite sequence of sets satisfying conditions (1.3), then p
is the power of the set M,x M,x M;x... Such a cardinal number p is
called the value of the infinite product of the cardinal numbers m,, Mg, ...
and written as

Die lynne sO if m1, =D).

Thus every infinite product of cardinal numbers has a well-defined

value, which is a cardinal number. Since, as we know (see 1V.12), by
changing the order of factors in a Cartesian product of sets we obtain
an equivalent set, the infinite product of cardinal numbers does not
depend on the order of factors. Neither does it depend on the manner
of joining them, as can easily be seen.
Obviously, if in the infinite product of cardinal numbers m,m,™m,..
we have m, = 1 for k > n, then the value of this infinite product is the
finite product m,m,... m,.
More generally, i: is obvious that in the infinite product of cardinal
numbers we can always omit factors equal to unity.

6. Properties of infinite products: of cardinal numbers. Examples. Sup-

pose that in the infinite product m,m,m,... all factors are equal to the
same cardinal number m. Let M denote a set of power m. From the defini-
tion of the infinite product of cardinal numbers it follows that as the
value p of our product we can regard the power of the set P of all in-
finite sequences whose terms are elements of the set M. But according

1) Cf. the beginning of §1 of this chapter.

12*
180 X. Infinite series and infinite products

to the definition of the exponentiation of cardinal numbers, the set P is

of power mo (see VII.5, Exercise 3). Hence follows the formula

(6.1) mio= 1-1: TI...

for every cardinal number m. In particular, for m= 2, in view of 2° = ¢

(VII.5, formula 1), we obtain

(6.2) (Bes AS,

This formula proves that for cardinal numbers the inequalities

(6.3) {lian IPRINeA ret TKI) Oo Vi Jumma le WAR Ee

do not always imply the inequality

(6.4) Ttec> ML Uta iit, --.

because, for instance, every partial product of the infinite product (6.2)
is finite, 7. e. <<), while the value of the infinite product (6.2) is >np.
The value of an infinite product of cardinal numbers need not necessarily
be the least cardinal number that is > each partial product of that
product.
On the other hand, however, it can easily be proved that for m, 4 0
(n=1,2,...) inequality (6.4) always implies inequality (6.3).
Assuming in formula (6.1) m=), or m= c, we obtain, in view of
the formulas s9°= c° = c, the formulae

(6.5) (=a NpNohoreely ead. iice=icetes.

These formulae prove that if the factors of an infinite product of

cardinal numbers are smaller than the corresponding factors of another
product, then the value of the first product is not necessarily less than
the value of the second. Still, it can easily be proved (by means of the
axiom of choice) that if we have two infinite products of cardinal numbers
M,m,m,... and 11,M,1,...., such that

(6.6) ity <1, tor) en a= ee,

then
My MgiMg... < My MeN...

For it follows from the axiom of choice, by (6.6), that there exist
infinite sequences of sets M,, M.,... and N,, N.,... such that i ie
N,=nt, and M,C N, forn—1,2,..., whence it follows that M,< Mox
x Msx...C NX N.XN,x... and thus the power m,m,m, of the set
M,x M,x... is < the power nnzgn3, ... of the set N,x N.x ...
 § 6. Properties of infinite products of cardinal numbers 181

As an application of the property proved above we shall calculate

the product of all successive natural numbers. Since

eae Ry LOR) WT es digod ye enss

we find that
bo So)
TA I ee << Gio Ae ae < NoNoNo--- 5

whence, by (6.2) and (6.5), we obtain at once

De 203a= *:
As regards other properties of infinite products of cardinal numbers,
here are the formulae
Cie
Aer ne) LIL Stay eng
and
WP tP2+ Pst.» — yPimr2em?s... ;

the proof of which is left to the reader.

7. Theorem of J. Kénig. From theorem 18, Chapter VI, follows at

once (in view of the definition of the sum and of the infinite product of
cardinal numbers)
’ THEOREM 1 (of J. Konig). Jf m,, m,,... and m,, th, ... are two infinite
sequences of cardinal numbers such that

C741) Ae ljee OTm ee Wie Ne arias,

then
(7.2) m, + m, + M54... < 1, M,N...

Suppose, in particular, that my <m,<... and assume nl, = M,41 for

n =1,2,...; we shall have inequalities (7.1), and, in virtue of Theorem 1,
inequality (7.2), 2. e. in this case the inequality

m, +m, +m,+...< MyMsMy... ,

whence, in case of m, ~ 0, also

my, + mt. + ms +... << mM, My mMs..-

Thus: the sum of an infinite series of increasing cardinal numbers

different from zero is always less than their product.
This, however, may be false for finite series, since for instance
ge a ed aan
EXERCISE. Let & denote a natural number >4. Prove that if the cardinal num-
bers m, (¢ = 1, 2,..., &) are finite 40 and form an increasing sequence, then m,-+ m,-}-
+m,+...+m, <™m,m,m,...m,; give an example of cardinal numbers 40 for which
the above does not hold.
182 X. Infinite series and infinite products

Solution. If the cardinal numbers m, (i = 1, 2, ..., k) are finite 4 0, 7. e. natural,

and if they form an increasing sequence, then we have m,_, > k—1, m,_,>k—2>2
and, in view of 2(k—1) = k+ (k—2) > k (since k > 4), we find that

m,M,...M,_,M,_,m,
> m,_,m,_,m,
> 2(k—1)m, > km, > m+mM,+...+M, ,

whence m,m,...m, > m,+m,+...+m,.

On the other hand, we have 14+ 24+...+(k—1)+8, = 1-2°3... (k—1)8, because
both sides are equal to Np.

From Theorem 13b of VI.7 immediately follows

THEOREM 2. If m,, m,,... and tl, Me, ... are two infinite RES Z
cardinal numbers, tn > 1 for n= 1,2, ...5 and if-m, <n, for n=1, 25%
then
Tt, 4 Mts + Mts >. S My Molt...

In the case of there being infinitely many numbers equal to unity

among the numbers t,, 1t,, .... Theorem 2 may be false, e. g. if m, = 1, = 1
fOrai == PO, wc. since hp le LA
It can be proved that the condition of Theorem 2 stating that n, > 1
for n= 1,2,... may be replaced by the condition stating that the num-
bers nt, (n = 1, 2,...) are all positive and infinitely many of them are >1,
or by the condition stating that the numbers n, (n = 1, 2,...) are all >0
and at least one of them is not finite.

8. Product of an arbitrary set of ‘cardinal numbers. The concept of

the product has been defined so far only for a finite and for a denume-
rable set of factors. Now we shall generalize it to an arbitrary set “ot
factors.
Let # denote an arbitrary set and suppose that with every element &
of the set H is associated a certain cardinal number m;. In XVI.1 we shall
deduce from the axiom of choice the existence of a function g(&), defined
for € « H and such that, for « HE, y(é) = Mz, where Mz is a set of power tm,.
Now let P denote the set of all functions f/(é) defined for é« EH and such
that f(£) « M; for &« E, and let p = P. We shall show (with the aid of
the axiom of choice) that the cardinal number p does not depend on the
choice of the} function y, provided it is such that, for é« H, y(&)= Meg
where M = We.
Therefore let us suppose that y(é) is also a function defined for € « H
and such that w(é) = Mz where M= me for €« H. Thus we have Me~Miz
for €« EH and it follows from the axiom of choice that there exists a func-
tion #(£) such that, for & « EH, # = %(p) is a (1-1) mapping of the set M;
upon the set M;. er us denote by P’ the set of all functions g(&), defined
for é« H and such that g(é) « Mz for « H. Obviously it suffices to show
 § 8. Product of an arbitrary set of cardinal numbers 183

that P~P’. For that purpose let us define for f « P a function y(f) in the
following way.
Let f denote a given element of the set P, and & a given element of
the set H. Thus we shall have f(&)« M;z, and therefore (f (€)) « M:. Let
g(&) = 8(f(€)). The function g(é) will thus be defined (by the function f)
for €« H and
we shall have g(&) « Mz for « H, whence ge P’. Let g=
= y(f). It is
easy to verify that the function y maps in a (1-1) manner
the set P onthe set P’. Thus we have P~P’, g.e. d.
We have
proved that the cardinal number p defined above depends
only on the set # and on the function h(é), associating with every element
«EH a certain cardinal number A(é) = m;. This number p we call the
product of the cardinal numbers m¢ extended to all elements & of the set EH
and write

Dp = | |We¢ 4
f€E

From Theorem 14 of VI.7 follows

THEOREM 1 (of KE. Zermelo [2], p. 227). If H is a non-empty set and

with every element & of the set E are associated two cardinal numbers, Me and Ne,
such that me< te for E« EH, then

(8.1) Sn ote:

From Theorem 1 for finite sets H it follows that Theorem 1 of X.7

is true also for finite sequences of cardinal numbers. Thus, in particular,
we have the following theorem T:
If m,, m,,1,, 1, are cardinal numbers such that m, <n, and m, < Nh,
then mM, +m, < 1yN,.
But we are not able to prove even this theorem without the aid
of the axiom of choice. As an easy conclusion from theorem T we shall
prove the following theorem T;:
Tf c=m+m,, then at least one of the numbers m, and m, ts
equal to c.
For suppose that c= m,+m, where m,<c and m,<c. According to
theorem T we thus have item = cc and therefore, since cc=c, m,+
+m,<c, contrary to the assumption.
Now, we cannot prove even theorem T, without the aid 01 the axiom
of choice. Theorem T, also follows immediately from the Corollary, to
Theorem 15 of VI.7, which implies also, more generally, that 7f a cardinal
number c is the sum of a finite number of cardinal numbers, then at least
one of them is equal to c.
184 X. Infinite series and infinite products

EXERCISE. Prove that if m is a cardinal number such that m- 2°09 = 22°, then
m = 22%,
Proof. Suppose that m- 2° — 2, Thus we have m- 2°°>m, whence m< 2”,
Therefore it suffices to show that m < 2° is impossible. Suppose that this inequality
holds. Let H denote a set of power 2° and let m,= m,n, = 2% for € « H. Therefore
m, <n, for € « # which, in virtue of Theorem 1, implies formula (8.1). But obviously
from 2 = 2° we get
Dm, =m-2% and =ff n, = (2%) = 2%
écE §<E

(because (2°°)? = 2°°). Hence, in view of (8.1), we should have m- 2% < 2%°, contrary
to our assumption. Thus the assumption that m < 2? leads to a contradiction, q. e. d.
 CHAPTER XI

ORDERED SETS

1. Ordered sets. Let A denote a given set, and Q — a given part of

its combinatorical square, 7. e. a set such that QC Ax A. We say that
the set Q determines a certain relation R between the elements of the set A,
namely the relation which holds between the elements a and bd (diffe-
rent or not) of the set A if and only if (a,b) «Q. In that case we write
a Rb, and if (a,b) éQ, we say that the relation R does not hold between
the elements a and b of the set A and write anon Rb or non(a BD).
For the set A there exist as many different relations between its
elements as there are different parts of the combinatorical square of the
set A, 7. e. for a set of power m there exist 2™ different relations between
its elements. We include here of course the relation which does not hold
between any two elements (different or not) of the set A, 7. e. when Q = 0,
as well as the relation which holds between every two élements of the
seund, 7.6. when-@ = Ax A,
A relation R, defined for the elements of the set A, is called con-
nected if for ae A, be A where a+b we have either a RD or bRa.
A relation R, defined for the elements of the set A, is called transitive
if for ae A, be A, ce A, from a RB and b Re always follows a Re.
If for the elements of the set A a relation R is defined, connected,
transitive and such that anon Ra for ae A’), then we say that the
set A is ordered by the relation R. This relation is then denoted by 2.
If the set A is ordered by the relation < and if ae A, be A and
a <b, then b Sa is impossible since, in view of the transitivity of the
relation <, we should then have a <a, which is impossible.
If for a relation R, defined in the set A the formula aRb always
implies the formula b non Ra, we say that the relation R is asymmetrical.
If the set A is ordered by the relation <, then, the relation < being
connected and asymmetrical, for every two different elements a and b of
the set A one and only one of the formulae a <b and b <a holds.

1) If anon Ra for ae A, then the relation R is called irreflexive in the set A.

186 XI. Ordered sets

Obviously the relation of order could also be defined as a connected,

transitive and asymmetrical relation (since the asymmetry implies
anon Ra for ae A).
A subset of an ordered set is obviously also an ordered set if the
order relations between the elements of the subset are the same as in
the whole set.
Examples of ordered sets

The set of all rational numbers will be ordered if, for the rational
numbers a and b, formula a <b means that a < b. However, it is pos-
sible to order that set differently, e. g. if a <b means that a > b. Another
ordering of the set of all rational numbers will be obtained in the fol-
lowing way. In III.4 we have defined a method of arranging all rational
numbers in an infinite sequence. If a<b means that in the sequence
number a precedes number b, we shall obtain a new ordering of the set
of all rational numbers.
The set of all natural pumbers, besides being ordered according to
increasing magnitudes or according to decreasing magnitudes, can be
ordered by assuming that a<b if and only if number a has smaller
natural divisors than number 0b, and in the case of an equal number of
divisors — if a<b. The reader will easily verify that the relation <
established in this way really orders our set (7. e. that it is connected,
transitive and asymmetrical).
We shall have here, for instance,

eeee Oo 08
The set of all complex numbers may be ordered by assuming that
a+bi<c+di if and only if a<e, or if a=ec and b < d. Thus we shall
have, for instance,

9423) 234.2) 31515

16125.
The set of all infinite sequences of real numbers may be ordered
according to the so-called principle of first differences, namely by
putting first that one of two different sequences in which we first
meet a smaller number with the same index. Here also, of course, it
should be proved that the relation between sequences defined in this
way satisfies the above-mentioned three conditions. If we apply this order
arrangement to infinite sequences composed of the successive denomina-
tors of the expansion of irrational numbers from the interval (0,1) into
continued arithmetical fractions, then we shall obtain an arrangement
of those irrational numbers different from the arrangement according to
' their magnitudes.
 § 1. Ordered sets 187

If we have a family F of sets such that one of every two sets belonging
to # is always a subset of the other, then we can order family / putting
first that one of two different sets belonging to it which is a proper
subset of the other.
It is remarkable that every ordered set A easily permits the construc-
tion of such a family F of sets: it suffices to take the family F' of all
sets A(a), for ae A, where A(a) denotes the set composed of the ele-
ment @ and of all elements x of the set A such that 7 <a.
We are not able to order every set (7. e to define in every set a con-
nected, transitive and asymmetrical relation); e. g. we are not able to order
the set of all sets of real numbers, or the set of all real functions of a real
variable. Also, it is difficult to give an effective example of an ordered
set of greater power than the continuum (Corollary to Theorem 3 of XVI.1).

Inverse order arrangement

If a set A is ordered by a relation o and if we define a relation o0*

for the elements of that set by the condition that ao*b holds for the
elements a and b of the set A if and only if boa, then the set A will be
ordered by the relation o*. Indeed, if the set A is ordered by the relation 0,
this relation must be connected, transitive and asymmetrical, and from
the definition of the relation o* it immediately follows that o* is also
connected, transitive and assymmetrical; thus the relation o* orders
the set A. The order arrangement of the set A by the relation o* (pro-
vided A has more than one element) is different from the arrangement
by the relation 0; we call that order arrangement inverse with respect
to that defined by the relation 0.

2. Partially ordered sets. If, for the elements of the set A, a relation R
is defined, transitive and such that a non R a for ae« A (or, which is the
same, if the relation R is transitive and asymmetrical), then we say that
the relation BR orders the set A partially, or that the set A is partially ordered
by the relation R.

Examples of partially ordered sets

The set of all real functions of a real variable will be partially ordered
by the relation /Rg denoting that for every real number x we have
f(«) <g(#).
The set of all natural numbers will be partially ordered by the rela-
tion aRb denoting that number a is a divisor of number b, smaller
than b.
188 XI. Ordered sets

If R, and R, are relations ordering (fully or even partially) the sets A,

and A,, then the set A = A,X A, will be partially ordered by the rela-
tion R defined as follows: for (a,, 0,)€¢ A, (@,, 0.) « A we have (a,, 0,)
R(a2, 52)
if and only if a,R,a, and b,R,d,.
If A is a set ordered (or even only partially ordered) by a relation 0,
then the set B of all subsets of the set A will be partially ordered by a rela-
tion 9, provided, for X¥ CA, YC A, we take Xo,Y if and only if, for 7 « X,
y « Y, we always have xey. The set B will also be partially ordered by rela-
tion e, provided, for XC A, YC A, we assume Xo,Y if and only if there
exists an element x of the set X such that for each element y of the
set Y we have zxoy.
The set of all positive functions of a real variable will be partially
ordered by the relation fRg denoting that

If the relation R partially orders the set A, we are able to associate

with every element a of the set A a certain subset f(a) of the set A in such
a way that, for ae A, be A, formula aRb will hold if and only if the set
f(a) is a proper subset of the set f(b). Obviously it suffices for this pur-
pose to denote by f(a), for ae A, the set composed of the element a and
of all those elements xv of the set A for which xRa.
On the other hand, every set of subsets of a given set is obviously
partially ordered by means of the relation ARB denoting that the
set A is a proper subset of the set B. The examination of partially ordered
sets is thus equivalent to the examination of families of sets with respect
to the relation of inclusion (C).
It can be proved that if the set A is partially ordered by the rela-
tion R, then the set A can be ordered by means of the relation S in such
a way that if, for any elements a and Db of the set A, the formula aRb
holds, then also a <b. The proof is difficult and (for non-denumerable
sets) based on the axiom of choice (Szpilrajn [1], p. 386).
Now let A denote a given set and F' a given set of its different order
arrangements. Let us assume that a <b, for ae A and be A, if and only
if the element a precedes b in every order arrangement of the set A belong-
ing to F.
It is easy to prove that in this way the set A will be partially or-
dered. It would be much more difficult to prove (with the aid of the
axiom of choice) that every partial order arrangement of the set A may
be obtained in this way (with the suitable choice of the set F of order
arrangements of the set A) (Dushnik and Miller [2], p. 600-610).
 § 2. Partially ordered sets 189

EXERCISES. 1. Prove (without the aid of the axiom of choice) that if an effectively
denumerable set A is partially ordered by a relation e, then we are able to define an order
arrangement @, of the set A such that for ae A, b « A formula agb always implies ag,b.
Proof. Let A = {a,, a, ...} denote a given effectively denumerable set, partially
ordered by a relation g. It is sufficient to prove that we are able to define a function f
which maps the set A on such a subset of the set of all rational numbers that for a « A,
be A the formula agb always implies f(a) < f(b).
Let 7,, 72, ... denote the infinite sequence composed, of all rational numbers (which
we are able to construct). Let f(a,) = 7,. Now let » denote a given natural number >1
and suppose that we have already defined numbers f(a,) for i < and that fork <n
and J < » the formula a, ga, always implies f(a,) < f(a,). We shall distinguish three cases:
1) In the sequence a,, a,,...,d,_, there are no such terms a, that a,ea,. In that
case we denote by f(a,) the first term of the sequence 7,,7,,... that is smaller than
any of the numbers f(a), f(d2),..-,f/(@,_,). (From the properties of the set of all ra-
tional numbers it follows that such a term exists).
2) Case 1) does not hold, and in the sequence ay, a2, ...,@,_, there are no such
terms a, that a, oa,. In that case we denote by f(a,) the first term of the sequence 7,
%,..., that is greater than any of the numbers f(a,), f(a), .--, f(Q,_)-
3) Cases 1) and 2) do not hold. We denote by p, the greatest of the numbers f(a,)
corresponding to elements a, such that k < n and a,ea,, and by q, the least of the num-
bers f(a,), corresponding to elements a, such that | < n and a,ea,; by f(a) we denote
the first term of the sequence 7,,7,,... that is different from f(a,), f(a), ---, f(@,_,),
greater than p, and smaller than q,.
In this way the infinite sequence of rational numbers f(a,) (n = 1, 2, ...) is defined
by induction. It remains to prove that if, for natural k and 1, we have a,ea,, then
f(a) < f(a).
Let us suppose, that it is not so. Thus there would exist an arrangement of
the natural numbers & and J such that a,ea, and f(a,) >f(a,). In view of the trre-
flexivity of the relation @ we should have k 41, and from the definition of the infi-
nite sequence f(a,) (n = 1, 2,...) it follows that all its terms are different from one
another. Thus we should have f(a,) # f(a,), and therefore f(a,) > f(a).
Among all the different arrangements of natural numbers k and | where a, oq,
and f(a,) > f(a,) there would exist an arrangement in which the number max(k, 1)
would be least. Let (p,q) denote such an arrangement of natural numbers. Thus we
have a, ea,,f(a,) > f(a,) and, if max(k,l) <= max(p,q), te. if k<n andl<n,
and if a,oa,, then f(a,) < f(a,).
If p > q, then n = p and, since a, ega,, we have a, ga,- From the definition of num-
ber f(a,) it follows that f(a,) < f(a,), which, in view of n=, contradicts the in-
equality f(a,) > f(a,).
If p <q, then qg=™ and, in view of a,ea,, we have a,ea,. From the definition
of number f(a,) it follows that f(a,) > f(a,), which, in view of n= q, again contra-
dicts the inequality f(a,) > f(a,). ’
Thus we have proved that for natural k and / the formula a, ea, implies the formula
f(a) < f(a), q.e. a.
2. Prove the following theorem T: If A is a partially ordered infinite set such that
each of its infinite subsets contains an ordered parr of elements, then there exists an ordered
infinite subset of the set A +).

1) The question whether theorem T is true has been put forward by a young Hun-
garian mathematician, A. Hajnal.
190 XI. Ordered sets

Proof. Let A be an infinite set, partially ordered by a relation o. We shall prove

that there exists an element a «A for which the set

H, = xrEeA
F (weal+ eAF [aex)+ {a}
is infinite. Suppose that for a «A each set H, is finite. We shall define by induction
an infinite sequence a,, da, ... of elements of the set A as follows. Let us suppose that we
have already defined the elements a,, dz, ..., a, of the set A (where n is a given natural
number). The set Ho,-+Ha,-+ ...+Ha, is finite and therefore the set A— (Ha,+ ...+Ha,)
is non-empty; let a, 4a denote any of its elements. The elements a,, a,.,..., defined in
this way by induction (with the aid of the axiom of choice, since we have not indicated
the method of choosing them), are of course all different from one another, since from
the definition of the sets H, it follows that a, « H, for k = 1, 2,... In virtue of the as-
sumption concerning the set A there exists among those elements an ordered pair,
2. e. there exist natural numbers k and 1 > k such that either a, ea, or a,oa,. Since k < I,
we have
£ a i teh
Gihs del. = EF. (wea,|+- F (a, or)-+ {a,},
Ke eA
whence it follows that we can have neither a,ea, nor a,ea,, which gives a contradiction.
Thus we have proved that there exists an element b, of the set A such that the set

H,, = z€A
EF (»eb.]+ x€eA
F tb, er]+ 1}
is infinite. It follows that at least one of the sets

F [webs]
xeA
and ff [b, ea]
xéeA

is infinite. Let us denote by A, the first of these sets that is infinite. As we have proved,
then there exists an element b, « A, such that the set

FE, (xob.)+ zeFEAy (b.02]

Ze Ay

is infinite. Let us denote by A, the first of these two components that is infinite and
let b, denote an element of the set A such that the set

EF (xebsl+ xeFEAg (bs oz]

zéeAg
is infinite, ete.
We shall prove that the set {b,, b,,...} will be ordered
by the relation 9. Indeed,
let k and J be two natural numbers such that k < 1. Obviously, we have 4 5 A, A,2...,
therefore A, ,c A,, and

A, = EK [web,] or A,= I; [b,00] .

eA, 4 weAy 4

Since b, « A,_, CcA, Cc A,_,, we have 6,0), or b,0b,, which, the natural numbers k and
1 > k being arbitrary, proves that the infinite set {b,, b,, ...} forms an infinite ordered
subset of the set A. Theorem T is thus proved.
Remark. We are not able to settle the question whether every non-denumerable
set satisfying the conditions of theorem T contains a non-denumerable ordered subset.
Prof. B. Knaster has put forward the question whether if A is a non-denumerable
 § 2. Partially ordered sets oH

partially ordered set such that each of its non-denumerable subsets contains an orde-
red infinite subset, there exists an ordered non-denumerable subset of the set A.
If the set A is finite, then every relation @ between its elements may be defined
by means of a table having as many rows and as many columns as there are elements
in the set A; in the table both the rows and the columns are supplied with headings
in which the elements of the set A are written out, while at the intersection of the row
with the heading a and the column with the heading b we write number | if agb and
number 0 if aeb does not hold.
Thus for instance the table

anes)
S|
Sh so
oro|!lH|
OD
|
Si
ma
—

defines a certain order arrangement of the set {a, b, c}, namely that in which a <b ~e.

2. Prove that the table

a) |) eld

Gis WO AO Th ye al
Sea Ma OPEDaL
er rono 0/0
d}olojijo

defines an order arrangement of the set {a, b,c, d} such that bSaxdwJe.

3. Prove that the table

aloe |-d
Cm COM set

a a i
e |0lo0lo0!o
d|0|o}o]o

defines a certain partial order arrangement of the set {a, b, ¢, d}.

4, Prove that the relation 9 defined by the table

is not a partial order arrangement of the set {a, b, c}.

192 XI. Ordered sets

Proof. It follows from the table of the relation 9 that we have agb and bee, but -
not ac; thus the relation g is not transitive, and therefore does not partially order
the set {a, b, c}. ‘
5. Calculate the number of partial order arrangements of a two-element set and
write out their tables.
Answer. There are three such partial order arrangements, and their tables are:

[e{e] fae ||3]

a | oh || Oi)I) a 0| |

a ) ieuinonos Teale oO

6. Calculate the number of partial order arrangements of a three-element

set
and write out their tables.
Answer. There are 19 such partial order arrangements and their tables are respec-
tively:

a\bje an a\biec a\b|e |a| ble rb|¢

a|0|0|0 a|O|1 a|0|0j)1 a|0/0!0 a}0|0{0 a|0|0|0
b{0/0}0 b|0/0 0/0|0 b/1/0/0 b/0}0/1 b/0/0/0
¢|0/0/0 ¢|0|0 ¢|0/010 ¢|0}6|0 c|0)0] 0) -|1fo|o|

|a|b]e |a|b)e |a] b|¢ a|b|e |a| ble [a] o] | a|b|e

a/0}0}0) aloliji a|0|0/0 a|0|0|0 a|0|0/0) a]o/1]o a|0|O|1
0|0 0/0} BI1;O]1} B/O;O]O; BIi1J/O}O] BJolojo| Bjojoj1
e/0/1/0] ej0jojo| elojojo| ef1jajo] ejijojo| folio ¢/0/0/0

|a]b|e a|b\e} a|b) «| |a|o|e| j}a|ble| a|b|e

a|o|1|1 a|O|1|1 a|0|0|1 a|0|0] 0) a|o|1|0 a|0|0|0
b}0/0]1 0/0/0 b}1/0/1 b/1/ol4 “ac0 b/1/0/0
c/0/0/0 c}0/1]0 e|0/0|0 ef1jojo| {1 |1jo| eli [1}o|
Remark. The numerical function f(n), expressing how many different partial
order arrangements of a set of n elements exist for a natural number , has been little
investigated and we do not know any easy method of calculating it even for natural
numbers <10. We have f(2) = 3, f(3) = 19, f(4) = 219 (G. Birkhoff).
7. Calculate the number of different partial order arrangements of a denumer-
able set.
Solution. Their number is the continuum. Indeed, on one hand, we know (XI.1)
that there are 2° — 2% different relations between the elements of a denumerable set,
whence it follows that the set Z of all the different partial order arrangements of a de-
numerable set is of power <2*°. On the other hand, there are of course no fewer different
order arrangements of a denumerable set than there are different permutations of the
 § 2. Partially ordered sets 193

set N of all natural numbers; the number of those is >2*°, since by associating with
every real number #, such that 0 < «<1, whose expansion into a proper infinite
dyadie fraction is
co

Te ele Gan
n=1

an infinite sequence of natural numbers

1+ a,, 2—a,, 3+ a), 4—a,, ..., 2n—1+4,, 2n—a4,, ...,

we shall obviously obtain 2“° different permutations of the set V. Therefore the num-
ber of all the different order arrangements of a denumerable set and, naturally, that of
all its partial order arrangements are >2”. The inequalities obtained, Z < 2% and Z > 2%
give Z = 2", q.e. d.
8. Calculate the number of different order arrangements (or of different partial
order arrangements) of a set equivalent to the continuum.
Solution. Their number is 27°. Indeed, on one hand, in view of the formula
(22%0)2 — 22%, we conclude that their number is <2”°. On the other hand, let X denote
the set of all real numbers, and 7’ — any set of positive real numbers; the set F of all
such sets is known to be of power 22°. For every set 7’ «I, let us denote by U(T) the
set X ordered in such a way that each number of the set T is regarded as preceding each
number of the set X—/F, and the numbers of each of the sets T and X—F are ordered
according to their magnitude. It is easy to verify that in this way to different sets T «Ff
would correspond, different order arrangements of the set X. The set of all the different
order arrangements and, naturally, the set of all the different partial order arrange-
ments of the set X are thus of power >2?° and since we proved before that it is of power
<2°, therefore it is of power 22%, q.e. d.
Remark. With the aid of the axiom of choice it can be proved that the set of all
the different order arrangements of an infinite set of power m is of power 2™.
9. Prove without the aid of the axiom of choice that the set of all the different
No, 2 28
order arrangements of a set of power 2° is of power 2?°”.

3. Lattices. Let A denote a given partially ordered set. If for two

given elements a and b of the set A there exists an element ¢ of the set A
such that ac (which means that either a=c or a~e) and b3e,
and that for each element w of the set A such that a <a fe b= a we
have ¢ = 2, then the element c will be called the swm of the elementsa and b
and denoted by a+b. Of course it is not for every two given elements
of a partially ordered set that such a sum exists; but if it exists,
then there is only one, since if an element c’ of the set A satisfied
the same conditions as ec, we should obviously have ¢s ¢’ and
c' 3 ¢, whence, in view of the asymmetry of the relation =i ke! =o Lt
is also obvious that if the sum of the elements a and b exists, it does not
depend on the order of the components. According to the definition of
the sum which we have accepted, we must also accept the existence of
the sum a+a=a for each element a of a partially ordered set.
Cardinal and ordinal numbers 1 3
194 XI. Ordered sets

An example of a partially ordered set each two elements of which

have a sum is given by the set of all subsets of a given set Z. The
sum of two elements of that set will obviously be the sum of the sets
constituting those elements.
If for two given elements, a and b, of a partially ordered set A there
exists an element c of the set A such that ¢ =a and ¢ = b, and for every
element « of the set A such that «3a and «3b we have xc, then
the element ¢ will be called the product of the elements a and b and de-
noted by ab.
It is not for every two elements of a partially ordered set that such
a product exists, but if it exists then obviously there is only one.
Clearly, if the product ab exists, then also the product ba = ab exists.
According to the definition of the product, we must also accept the
existence of the product aa=a for every element a of a partially or-
dered set.
As an example of a partially ordered set in which for each two of
its elements there exists a product we may again quote the set of all
subsets of a given set Z. The product of two elements of that set will
obviously be the product of the sets constituting those elements.
Another example of a partially ordered set in. which each two ele-
ments have a sum and a product is the set of all natural numbers, partially
ordered by the relation R where aRb means that number a is a divisor
of number b different from b. It is easy to prove that the ,,sum“ of two
elements of our set will be their smallest common multiple, and their
»product“ will be their largest common divisor.
A partially ordered set every two elements of which have a sum and
a product is called a lattice. Lattices represent a very important particular
case of partially ordered sets. Numerous papers have been devoted to
their examination, and G. Birkhoff has written a whole book on the
subject (Birkhoff [1}).
From what we have said above about the sum and the product of
two elements of a partially ordered set it follows immediately that if
such a set A is a lattice, then two operations are defined and performable
in it: addition and multiplication; these operations have the following
two properties.
1) ee =o anda aw Tora <A,
2) cy=ye and «+y=y+2a@ for we A, ye A.
It is easy to prove that they also have the following two properties:
3) a(yz) = (ay)e and «+(y+z2)=(a#+y)+e2 for 7, y,2¢«A;
4) a(e-+y) = wand @+ey =a for geA, y< At
Indeed, according to the definition of the product, yz is an element
of the set A such that ye sy and yz Sz, and if s3y and s Sz, then
 § 3. Lattices 195

ss yz, while x(yz) is an element such that w(yz) sa and «(yz) 3 yz;
therefore, since yz sy and yz 32, we have a(yz) 3 yand «(yz) 3, and
if te and t sy, then t 3a/(yz). Thus if usa, uy, ue, then U32,
whence also Us x (yz). On the other hand, ay igan element such that
vy sx and xy 3y and if ssa and s sy, then s say, while (vy)e is an
element such that (« y)23 vy and (ay)z sz, and if ¢ is an element such
that tS ay and t Sz, then iS ey) 2. Since (ay)z xy and ay 3a and
xy 3y,we have (#y)ze3a, (ay)e3y, (ay)e32, and if wa, w3y
and w sz, then ws xy, whence also w = (wy)z. Thus we have proved
that a(yz) 3a, x a(yz) 3 y, (yz) 2, and that if ua, u sy and u 3%,
then vu 3 a(yz), and algo that (ay)z-Su, (wy)ze SY, (ay)e3z and if w = Le
w3sy, wsz2, then w 3S (ay)z.
Therefore we coun pecan that wu= (w#y)z and w = «#(yz), which gives
(xy)2 3 xv(yz) and x(yz)S (wy)z, whence (ay)z= «(yz), which proves the
truth of the first formula in 3).
The truth of the second formula in 3) might be proved in an ana-
logous way, or it could be observed that between the two formulas there
is a kind of duality. For if a set A partially ordered by a relation R is
a lattice, then the same set ordered by a relation R’ such that for a« A,
be A we have aR’d if and only if b Ra, is obviously also a lattice, and it
(for certain elements a, b, ¢ of the set A) we have c= a-+b in the first
lattice, then in the second we shall have c= a-b, and vice versa, and if
we have ¢ = ab in the first, then we shall have ¢ = a+b and vice versa.
In order to prove the first formula in 4) let us observe that if the
elements « and y belong to a set A constituting a lattice, then, according
to the definition of the product, we have x(#+y) 3 a, and since, in view
of the properties of the sum, we have «3 7+y and1 of course sw, there-
fore, in virtue of the definition of the product, we have a = “(e+y).
Thus «3 a(%+y) 3a, whence w(v7+y)= a, g.e.d. We can prove the
second formula in 4) in an analogous way, or by resorting to the principle
of duality mentioned above.
Thus we have proved that if a partially ordered set A is a lattice Ai
according to the definition given above — let us call it a lattice of the
1-st kind — then by defining in it in the above way the addition and
the multiplication of two elements weobtain a set in which both operations
are uniquely performable and the properties 1) to 4) hold. Let us denote
the set obtained in this way by K, and let K, = f(K,).
Now we call a lattice of the 2nd kind every set A (not necessarily par-
tially ordered) in which the addition and the multiplication of elements
of the set A are defined, uniquely performable in that set and satisfying
conditions 1) to 4). From what we have proved it follows that if K, is
a lattice of the 1-st kind, then K, = f(K,) is a lattice of the 2nd kind.
13*
196 XI. Ordered sets

Now we shall prove that the correspondence f between lattices of

the 1-st kind and lattices of the 2-nd kind is (1-1).
For this purpose it suffices to show that for every lattice K, of the
2nd kind there exists one and only one lattice A, of the first kind such
that HK, = 7 (K,).
Since K, is a lattice of the 2-nd kind, then K, is a set A in which
the addition and the multiplication of elements are defined and uniquely
performable and the conditions 1)-4) hold. Now, for the elements a
and b of the set A let a<b if and only if a~b and ab = a. Thus if a<b
and b<c, then we have a4b, ab=a, b~c, be=b; therefore, by 3),
ac = (ab)c = a(bc) = ab =a, whence ac= a. Ti we had’a=—<, then, m
view of be = b, we should have ba = b; therefore, by 2), a = ab = ba = b,
whence a = b, which is impossible. Consequently, ac = a and a # ¢, which
proves that a <c. Relation < is thus transitive. It is also asymmetrical
because in case of a<b and b <a, in view of the already proved transi-
tivity of the relation S, we should have a<a; therefore, according to
the definition of the relation <, aa, which is impossible. Thus the
relation < partially orders the set A.
Now we shall prove that ab is an element of the set A such that
ab 3a, ab 3b, and if « Sa and « 3b, then x Sab. From the definition
of the pelation < it immediately follows by 1), that the formula a <b
is equivalent to the formula ab = a. Therefore if ~sa and #3), then
ta=e and «b= a, which, by 1), 2) and 3), gives «= a7 = (@ aN) =
= (wx)(ab) = x(ab), and this proves that a = ab.
On the other hand, by 1), 2) and 3), we have (ab)a= aab= ab,
whence ab Sa, and similarly we obtain ab sb. Thus we have proved
that ab is an element of the set A such that ab Sa, ab 3b and if asa
and « 3b, then # ab. fr iy ©
By 4) we have a(a+b) =a, which, aS we know, gives asa+b.
Similarly, by 4) and 3), we have ene = b, whence b sat.
Now let # denote an element such that a3 and bw. Thus we
have ax = a and bax= b. Hence, by 4) and 2), we have eta=e2tan=
=ae+x2a—= 4, and similarly «+b= 4; therefore, by 1), 2) and 3), +
+(a+b) = (#+a)+(#+b)= «4+a= x, whence, by 4) and 2), x(a+b)=
=[#%+(a+b)]|(a+b)=a+b, which gives a+b 3S @.
We have proved that a+b is an element of the set A such that
a sa+b, b3a-+b, and that if asa and b 3a, then a+b 32.
Thus the set A is a lattice dG, of the 1-st kind such that {(K 1) us
since, aS immediately follows from the already proved properties of the
elements ab and a+b, the sum and the product in the lattice K, are,
respectively, the same as in the lattice K,.
Now suppose that there exists another lattice Kz of the 1-st kind,
 $ 3. Lattices 197

different from A, and such that f(A7) = K,. The lattice A; would thus also
be a set A, but, being different from the lattice K,, it would be partially
ordered in a different way than A,. Thus there would exist two different
elements a and b of the set A such that a<b in the lattice A, and non
(a<b) in the lattice Aj, or non(a-—b) in the lattice K, and ab in
the lattice Ky;. Suppose that a<b in the lattice K, and non (a <b) in
the lattice Ay. In view of f(K,) = f(Ki) the same element of the set A
would be the product of the elements a and 6 both in the lattice K, and
in the lattice K;. But in the lattice K, the product of the elements a
and b is the element a, since, in view of a <b) in K,, we haveasa,ax3b
and, if sa and #3), then «3a. Now, in the lattice AK; the elementa
is not the product of the elements a and b, since, in view of ab and
non (a<b) in Ky, we cannot have asb in K;. We prove in a similar
way that in case of non (a<b) in K, and a<b in Kj the element a is
not the product of the elements a and b in K,, but it is their product in Kj.
Thus we have proved that the correspondence f between lattices
of the 1-st kind and lattices of the 2-nd kind is (1-1). The examination
of lattices of the 1-st kind is thus reduced to the examination of lattices
of the 2-nd kind. Since lattices of the 2-nd kind are defined in a purely al-
gebraic manner, the examination of partially ordered sets, which we
have called lattices (and then lattices of the 1-st kind), may be algebrized.
As regards conditions 1)-4), we shall observe that they are not
independent of one another. Namely, as already Dedekind (Birkhoff [1],
p. 18) observed, from 4) follows 1). For if, in the second formula of 4),
we replace y by «+ y, then, in virtue of the first formula in 4), we shall
obtain «= a#+ae(a+y)=a+a, and if, in the first formula 4), we re-
place y by xy, then, in virtue of the 2-nd formula in 4), we shall obtain
= Ue cy) = vw.
EXERCISE. Prove that conditions 1), 2), 3) and 4) are equivalent to conditions
1), 2), 3), 5) where 5) denotes the following condition:
5) For we A, ye A the formula cy = a is equivalent to the formula 7+ y = y.
Proof. Suppose that conditions 1), 2), 3) and 4) are satisfied and that wy = «;
by 2) and 4), we then have x+y = vy+y = y+cy = y. And if x+ y = y, then, by 4),
we have sy = «(a#+y) =a. Formulae cy = « and «+y=y are thus equivalent, 7. e.
condition 5) is satisfied.
On the other hand, if condition 5), is satisfied, then the formula w(x%+y) = aw is
equivalent to the formula #+ (#+ y) = «+ y, which is true by 3) and 1), and the for-
mula «+ xy = is equivalent, by 2), to the formula xy+ a= 4, which, by 5), is equi-
valent to the formula (xy)« = xy, which is true by 1) and 2). Thus conditions 1), 2), 3)
and 5) imply condition 4).

4, Similarity of sets. A set A ordered by a relation o will be denoted

by A(@), (@) being omitted only if there is no doubt as to what relation o
orders the set in question. Dealing with two ordered sets, A(e) and B(0’),
198 XI. Ordered sets

we shall say that the set A(o) is similar to the set B(oe’) and write A(e)~
~ B(o’) if there exists a function f, (1-1) mapping the set A on the set B
and such that if a, and a, are any two elements of the set A for which
4,043, then f(a,)0’f(d.). We say that the function f establishes similarity
between the sets A(o) and B(o’).
The relation of similarity between ordered sets is symmetrical, 7. e. if
A(o)~ B(o’), then B(oe’)~ A(e@). Indeed, suppose that A(e) ~ B(o’)
and let f denote a function establishing similarity between the sets A (@)
and B(o’), and f-! — the function inverse with respect to the function f.
Let b, and b, be any two elements of the set B such that b,0’b, and let
a, = f(b), a. = f(b2); we shall have a, « A and a, « A. If we had a,= a,,
we should have b, = f(a) =f(a.)=6,, which is impossible since }, 0’),
and the relation 0’ orders the set B. Therefore a, and a, are two different
elements of the set A, and since the relation @ orders the set A, we must
have either a, oa, or a,0a,. But if we had a,ea,, then, since the function /
establishes similarity between the set A(e@) and the set B(o’), we should
have f(a.) o’'f(a,), 2. €. b,0'b,, Which is impossible in view of b,0’b, and
of the asymmetry of the relation 0’ as the relation ordering the set B.
Therefore we must have a, oa,, 7. e. f(b) o'f(b). Thus we have proved
that if f-! is a (1-1) mapping of the set B on the set A such that if b, and b,
are any two elements of the set B for which b,0'b., then f—*(b,) of—(b.).
Thus the function f— establishes similarity between the set B(o’) and the
set A(e) and therefore we have B(o’)~ A(@), q.e. d.
It would also be easy to prove that the relation of similarity between
ordered sets is transitive, 2. e. that the formulae A(o)~+ B(e’) and B(o’) ~
~ C(0"’) always imply the formula A(e)~ C(e’’). Namely, it could easily
be proved that if the function f establishes similarity between the sets
A(o) and B(o’) while the function g establishes similarity between the
sets B(e’) and C(o’’), then the function g(f) establishes similarity between
the sets A(o) and CO(0’’).
For every ordered set A we have of course A(o0) ~ A(o); as the func-
tion f establishing similarity we can take here the identical mapping of
the set A on itself, 7. e. we can take f(a) = a for ae A. Thus we can say
that the relation of similarity between sets is reflexive.
If @ and o’ are two different relations ordering the same set A, then
A(e)~A(o’) may either hold or not. For example, if A = {1, 2,3,...}
and if, for the elements of the set A, zey means that 2 < y and the rela-
tion we’y means that the sum of the natural divisors of number @ is
smaller than the analogous sum for number y or that, if the sums of the
divisors of numbers # and y are equal, we have w < y, then the set A is ordered
by the relation o in a different way than by the relation 0’, but A(e) ~ A(0’)
will hold. We shall have 102030405o..., and 10’20'30'50'40'70'60'90’...
 § 4. Similarity of sets 199

If, however, xo’’y means that either number x has fewer natural divi-
sors than number y or,in caseof numbers # and y having the same
number of divisors, that » < y, then the sets A(e) and A(o’’) are not
similar.
The similarity of the sets A(o) and A(o’) here considered cannot
be established by means of an identical mapping.
However, it may happen that a non-identical mapping f of a given
ordered set A(e) upon itself has the property of making the formula xoy
equivalent to the formula f(x)oef(y); for example, in the set A of all inte-
gers, voy denotes that « < y and f(#) = «+5. But, if we wrote g(x) = —2,
then the formula vey would not imply the formula g(x) eg(y), since, for
instance, 1 <2 but g(1) = —1 > —2= g(2).
EXERCISES. 1. Prove that the set of all real numbers w such that —1 <a <1
ordered according to their magnitude is similar to the set of all real numbers ordered
according to their magnitude, and that the function

establishes the similarity of those sets. Give another function establishing the similarity
of those sets.
2. Prove that if a and b are real numbers such that a < b, then the set of all real
numbers a such that a < « < b ordered according to their magnitude is similar to
the set of all real numbers ordered according to their magnitude.
3. Prove that every function f(x) mapping the set of all real numbers ordered
according to their magnitude on itself in a similar way must be a continuous function
of a real variable.
Proof. Suppose that the function f(z) maps the set x of all real numbers ordered
according to their magnitude on itself in a similar way. Therefore the inverse function,
f(a), has the same property. Let 2, denote a given real number, and e — an arbitrary
positive number; f(x,)+« and f(x,)—e will be certain real numbers, and f(x#)—e <
< f(a) < f(%)+¢, whence a= f(f(x9)—e) < w% < f“f(x.)+¢) =, since the func-
tion f1 establishes the similarity of the set of all real numbers, ordered by means of
the relation <, to the same set, also ordered by the relation <. Now if « is a real num-
ber such that f"(f(a)—«) < # < f-(f()+ ¢), then, in view of the fact that the func-
tion f has the same property as the function f, to which we referred just now, and
in view of the fact that f(f(y)) = y for real y, we obtain

f(%)—e< f(x) <f(m)+«.

Thus we have proved that for every real number x, and every positive number «
there exist real numbers a and 6b such that a < a < b and f(%)—e« < f(x) < f(m)+¢
for a < x < b, which proves the continuity of the function f(z).
4. Calculate the power of the set / of all functions establishing the similarity of
the set of all real numbers ordered according to their magnitude to itself.
Solution. That power is the continuum. For, on one hand, according to Exercise 3,
every function belonging to the set / must be continuous and the set of all continuous
real functions of a real variable is equivalent to the continuum (see IV.13) whence
 200 XI. Ordered sets

F < 2%°, and on the other hand, every function f(z) = w+ a, where a is a real number,
obviously belongs to F', whence F > 2°. Therefore F = 2%°, q. e. d.
5. Calculate the power F of all functions establishing the similarity of the set
of all rational numbers ordered according to their Magnitude to itself.
Solution. That power is the continuum. For, on one hand, the number of all
functions defined in the set of rational numbers and assuming rational values is equal
to the continuum (since such is the number of infinite sequences of rational numbers),
whence < 2%. On the other hand, let x denote a real number such that 0<a< 1
and let

be its expansion into a dyadic fraction whichis really infinite. Let us define a function
f,(v) of a rational variable r as follows: f,(r) = r for r < 0. If k is an integer >0, then,
in case of a, = 0, let £,(r) =r for k<r<k-+1, and in case of a, = 1 let £,(r) = k+
+4(r—k) for kx r<k+# and f(r) =k+3(r—k)—3} for k+4<r<k+1.
Itis easy to verify that fe F forO<a<1 andf,~f, forO<#<1,0<y<l,
x #y, whence F > 2’, Thus F = 2%, q..e.d.

5. First and last element of an ordered set. Cuts. Jumps. Density and
continuity of a set. If, in an ordered set A(o), there exists an element a
such that for no element « of the set A does wea hold, then a is called the
first element of the set A(o). If, in the set A, there exists an element Dd
such that for no element x of the set A does box hold, then b is called
the last element of the set A.
In the set of all natural numbers ordered according to their magni-
tude there exists a first element, and that is number 1, but no last
element. In the set of all negative integers ordered according to their
magnitude, there is no first element but there is a last element —
number — 1. In the set of all integers ordered according to their magni-
_ tude there are no first and no last element. In the set of all real numbers wv
such that 0 <a# <1, ordered according to their magnitude, there exist
both a first element, 0, and a last element, 1.
Obviously, if the set A(@) is similar to the set B(0’) and if the set A(o)
has a first element, then the set B(o') also has a first element. If a is the first
element of the set A(e) and f is a function establishing the similarity
of the sets A(e) and B(o’), then b = f(a) is the first element of the set
B(o'). For if there existed an element ¢ « B such that ¢o’b, then we should
have a’= f“(c) ef(b) = a, whence a’« A and a’oa, which is impos-
sible since a is the first element of the set A(o).
We likewise prove that 7f the set A(o) is similar to the set B(o') and
has a last element, then the set B(o’) also has a last element.
As an easy conclusion, we shall prove that two ordered sets, A(o) and
B(o'), which are similar to subsets of one another are not necessarily similar.
Indeed, let A(g) denote the set of all rational numbers 7 such that
 § 5. First and last element of an ordered set 201

0<r <1 ordered according to the magnitude of those numbers and let
B(e) = A(e)— {1}. A similar mapping of the set A(o) on a subset of
the set B(e) is established by the function f(#) = #/2, and a similar map-~
ping of the set B(oe) on a subset of the set A(e) — by the function g(x) =
Still, the sets A(o) and B(o) are not similar since the set A(o) has a last
element while the set B(o) has no last element.
Now, from a theorem of Banach [2] (p. 239, Theorem 2) it follows
that if A(e) and B(o) are ordered sets which are similar to subsets of
one another, then there exist sets A,C A and B,C B such that A,(0)~
~ Bye’) and A(e)—A,(e)
+ B(e’) —B,(e’).
A decomposition of an ordered set A(o) into two non-empty com-
ponents, A= A,+A,, such that for any two elements a, and a, such
that a, « A,,a,<« A, we have a,oea,, is called a cut of that set. A cut is
denoted by the symbol [|A,, A].
If in a cut [A,, A,] the set A, has a last element and the set A, has
a first element, then we say that the cut gives a jump. If, however, the
set A, has no last element and the set A, has no first element, we say
that the cut [A,, A,] gives a gap.
An ordered set which has no jumps is called dense, if it has nenber
jump nor gaps it is called continuous.

EXERCISES. 1. Prove that if [4,, A,] is a cut of an ordered set A(o) and if a « A,,
x«A,, and xoa, then xe A,.

Proof. Let [4,, A,] be a cut of an ordered set A(g@) and let ae Ay, we A and xea.
If we had we A,, then, in view of a« A, and the properties of a cut, we should have
aex, which is impossible in view of wea and of the asymmetry of the relation 9. There-
fore x¢ A,, and thus, since A = A,+ A, and we A, we obtain we A,, q.e. d.

2. Prove that if [4,, A,] and [B,, B,] are cuts of the same ordered set, then
either A,c B, (and A,> B,) or B,c A, (and B,5 A,).
Proof. Suppose that a = [A,, A,] and 6 = [B,, B,] are two cuts of an ordered
set A(o). Thus we have A = A,+4,=— B+ B,, A,4,— BB, = 0. Hence A, = A—A,,
B, = A—B,. Therefore if we had A, c B,, we should have A—A,5 A—B,, 7. e. A, 5 B-
Let us suppose that A,cB, does not hold. Therefore A,—B, + 0 and there exists
an element a« A,—B,. Thus we have ae A, and, naturally, ae A (since A,c A) and
a¢ B,; therefore a « A—B,, i. e. ae B,. Thus the element a belongs in the cut f to the
class B,. Therefore if x « B,, then wea. But if ae A, and wea, then, according to Exer-
cise 1, we have we A,. We have proved that if # « B,, then x « A,, which proves that
B,c A,, whence it follows also that A—B,5 A—Aj,, 7. e. that B,> A,. Thus we have
proved our theorem.
If A(g) is an ordered set, and a and b are two elements of that set such that ab,
then the set composed of the elements a and b and of all elements # of the set A such
that aexeb is called a closed interval (a, b).
We say that an ordered set A(e) satisfies the axiom of Ascoli if, for every infinite
sequence of closed intervals each of which contains the next, there exists at least one
element common to all intervals of the sequence in question.
202 XI. Ordered sets

3. Prove that every ordered set that has no gaps satisfies the axiom of Ascoli.
Proof. Let (a,,,) (n= 1, 2,...) be an infinite sequence of closed intervals of
an ordered set A(o) such that each interval contains the next.
It follows that for every natural number n we have a,ea,,, OF a, = G4,- If
there exists such an index k that a,=a, for n>k, then a is a common element
of all our intervals. If there is no such index k, then obviously for each term a,, of the
sequence @,, a2, ... there exists in that sequence a term a, such that m < n and a, oa,.
In that case let us form a cut [A,, A,] of the set A(o) assigning to the class A, each
element a of the set A such that a,oa for k = 1, 2,..., and writing A, = A—A,. Thus
if ac A,, then there exists such an index k that the formula a, oa does not hold; there-
fore, in view of the connexity of the relation @ in the set A, we have either aga, or a = a,,
and then, as we know, for a certain natural n we have a, ea, and therefore aga,. Thus
in class A, there is no last element, and, in view of the assumption that the set A has
no gaps, there must exist a first element 6 in the set A,. From the fact that the interval
(Qn41> %n41) is contained in the interval (a,,, b,) we easily conclude that a,b, for all na-
tural k and n. This proves that b, « A, for n = 1, 2,..., and since b is the first element
of the set A,, we have either b, = b or beb, for every natural number n, whence also
a, ebeb, since a, <«A,, b«A,. Thus in any case b is an element of the closed interval
(a,,6,) for n= 1, 2,..., 7. e. a common element of all intervals of our sequence. The
set A(g) therefore satisfies the axiom of Ascoli, q. e. d.
It will be observed that an ordered set may satisfy the axiom of
Ascoli and have gaps’).
THEOREM 1. In order that an ordered set A(o) be dense it is necessary
and sufficient that for each two elements, a and b, of it such that aob there
exist at least one element xe A such that aox and xob.
Proof. The condition of Theorem 1 is necessary. For let us suppose
that it is not satisfied, 7. e. that there exist elements a and b of the set A(o)
such that aeb but there exists no element x of the set A such that aowx
and xob. Let us denote by A, the set of all elements x of the set A such
that xeb and let A, = A—A,. In view of agb we have ae A,, whence
A, #0, and since beb does not hold (in view of the irreflexivity of the
relation @), we have b¢ A, whence b « A, and A, #0. The sets A, and A,
are thus non-empty. Now if # denotes such an element of the set A, that
aex, then, according to our assumption, veb does not hold; therefore,
according to the definition of the set A,, we have x¢ A,, which gives
a contradiction. Thus in the set A, there is no such element x that aoa,
which proves that a is the last element of the set A,. Now, b is the first
element of the set A, since, according to the definition of the set A,,
if the element «x of the set] A satisfied the formula web, then we
should have #e« A,; consequently, since A, = A—A,, we could not have
vce A,. Finally, if 2, ¢A,, %,¢« A,, then we have 2,0b but not 2,ob. In
view of the connexity of the relation o we thus have either x7, = b or boas.

1) An example of such a set will be given by a set of type (144) (Q4+*)+1

(cf. Chapter XV, § 1).
 § 5. First and last element of an ordered set 203

If v, = b, then the formula 2, 0b gives x, o0x,, and if bex,, then, in view

of x, eb and the transitivity of the relation 0, we again obtain x, ex,. Thus
if 7, « A,, % ¢ A., then in any case x, 0x,. The decomposition A = A,+A,
gives the cut [A,, A,], in which the set A, has a last element, a, and the
set A, has a first element, D. Thus the cut[A,, A,] of the set A gives a jump
and the set A is not dense.
Now we shall prove that the condition of Theorem 1 is sufficient.
Suppose that for each two elements, a and bd, of the set A(o) such that
aob there exists at least one element we A such that aex and xob. If the
cut [A,, A,] of the set A(o0) gave a jump, then in the set A, there would
exist a last element, a, and in the set A, — a first element, b, and, since
ae A, and be A,, we should have aob. According to our assumption
there would exist an element a of the set A such that aex and xed. In
view of aox and of a being the last element of the set A,, 7 « A, cannot
hold, and in view of xeb and of b being the first element of the set A,,
xe A, cannot hold. Therefore x ¢ A, and «¢ A,, which is impossible since
xe A and A= A,+A, ([A,, A,] being a cut of the set A). Therefore no
cut of the set A gives a jump and thus the set A is dense.
Theorem 1 is thus proved.
An. element x of an ordered set A(e) such that aex and xeb or box
and woa is said to lie between the elements a and b. If a and 6 are two dif-
ferent elements of the set A(o), then, in view of the connexity of the
relation 0, we have either aeb or bea. Hence it follows that Theorem 1
may also be expressed in the following way:
In order that an ordered set be dense tt is necessary and sufficient that
between each two different elements of that set there lie at least one ele-
ment of it.
From Theorem 1 we shall now deduce the following
CoROLLARY. Between each two different elements of a dense ordered
set there lie infinitely many different elements of that set.
Proof. Let A(g) denote a given dense ordered set, and a and b —
two elements of that set such that agb. From Theorem 1 it follows that
there exists an element «, of the set A such that aex, and a,eb. Since x, ob
it follows from Theorem 1 that there exists an element x, of the set A
such that wv, ex, and x7, eb. Continuing in the same way we conclude that
for every natural number n there exists a sequence %%, 22, ..., ® of
elements of the set A(g) such that aox7,, 2, 0%2, 12 0%s, ...) Ln—10%n, Ln od.
In view of the transitivity of the relation eo, we conclude that aex, and
x,ob for k=1,2,...,, whence it follows that between the elements a
and b there lie at least n different elements of the set A(o). (Elements
1, Ua, ...,% are different since, in view of the transitivity of the rela-
204 XI. Ordered sets

tion 9, we have a, ea for 1 <k <1 <n, whence, in view of the irreflexivity
of the relation 0, we have a, ~ 2). The set of all elements of the set A(o)
lying between a and b cannot be finite. Our corollary is thus proved.
Remark. The above proof of the Corollary does not use the axiom
of choice since we have made only a finite number of choices (of the
elements %,, %,..-,%). We should not be able, however, to prove
without the aid of the axiom of choice that, in a dense set A(o), for each
two elements of it, a and b, such that aob, there exists an infinite sequence
of different elements of that set lying between a and Db.
It is easy to prove that a set which is similar to a dense set is dense
itself.
Tf a and b are elements of an ordered set A(e) such that aeb and if
there is no element x of that set for which aex and web, then the element b
is called the consequent of the element a and the element a — the ante-
cedent of the element b. Obviously each element of an ordered set has at
most one consequent and at most one antecedent.
EXERCISE. Prove that in order that an ordered set be dense it is necessary and
sufficient that no element of it have a consequent.

6. Finite ordered sets. We shall prove

THEOREM 1. Two finite ordered sets with the same number of elements
are always similar.
Proof. In view of the transitivity of the relation of similarity of
sets it suffices to prove that for every natural-number n an ordered set
of n elements A(go) is similar to the set {1, 2,...,} ordered according
to the magnitude of the numbers that compose it.
To begin with, we shall prove that the (non-empty) finite set A(o)
has a last element. For this purpose let us suppose that the set A(o)
has no last element. Leta, denote any element of the set A(g). Since
it is not the last element of the set A(o), there exists an element a, of
the set A(o) such that a,oa,. Similarly, we conclude that there exists
an element a, of the set A(e) such that a,oa,, etc., and finally an ele-
ment a,41 of the set A(e) such that a,ea,41. Thus we have a, oa, 0d 0...
0A, 04,41, Whence, in view of the transitivity of the relation 0, we conclude
that a,ea for 1<k<l<n-+1, which proves that the elements a,,
ay ++) Ans On41 Of the set A(o) are all different from one another. But
this is impossible because the set A(oe) has only n elements.
We have proved that every finite ordered set has a last element,
In order to prove Theorem 1 we shall apply the method of induction.
For sets consisting of one element Theorem 1 is obviously true. Now
let m denote a natural number >1 and suppose that Theorem 1 is true
for every ordered set of n—1 elements. Let A(o) denote an ordered set
 § 6. Finite ordered sets 205

of n elements. We have proved that A(o) has a last element, b. If «x is

an element of the set A(g) that is different from 6, then box cannot hold
and, in view of the connexity of the relation o, we must have xob. Let
A, = A—{b}; A, will be a set of n—1 elements which is also ordered hy
the relation 9 (as a subset of the set A(o)), and for each element x of
the set A, we shall have xoeb. In virtue of the assumption that Theorem 1
is true for every ordered set: of n —1 elements, the set A,(o) is similar to the
set (1, 2,...,2—1}, ordered according to the magnitude of the numbers
which compose it. But A = A,-+ {b}, while {1, 2,...,n} = (1, 2,..., n—1}+4+
+ {n}, and for z<« A, we have xob while for k « {1, 2,...,n—1} we have
k <n. Hence we easily conclude that if we associate the number n with
the element b of the set A, then the set A(o) will be similar to the set
Rg ore tcos's Ly.
Therefore Theorem 1 is true for sets of n elements. Hence, by induc-
tion, follows the truth of Theorem 1.
Since the ordered set A(o) of n elements is, in virtue of Theorem 1,
similar to the set {1,2,...,}, ordered according to the magnitude of
the numbers which compose it, there exists a (1-1) function /, associating
with each natural number k <n a certain element f(k) of the setA (@).
Lew 7(s)== a,. Therefore A(o) = {4;, a, ... ,}, ANd 4, 0d, 0d... oa,. Thus
the elements of every finite ordered set can be arranged in a finite se-
quence and numbered by means of successive natural numbers (not larger
-than the number of elements of the set) in such a way that aeb will hold
if and only if the index number of element a is smaller than the index
number of element b.
Let n denote a given natural number. Obviously different permuta-
tions of the set {1,2,...,} determine different arrangements of that
set provided we assume aob in a permutation if and only if a, as a term
of the permutation, precedes b. Since, as we know, there exist n! permuta-
tions of n elements, it follows from Theorem 1 that for a finite set of n
elements we have n! different order arrangements.
From Theorem 1 it immediately follows that two finite ordered
sets which are equivalent (of the same power) are always similar. This
does not apply to infinite sets. The set of all natural numbers and the
set of all positive numbers are equivalent (since they are both denumer-
able), but, if each is ordered according to the magnitude of the numbers
which belong to it, they are not similar since the former has a first element,
and the latter has not.

7. Sets of type w. An ordered set which is similar to the set of all

natural numbers ordered according to their magnitude is said to be of
type w. Thus for instance the set of all natural numbers ordered according
206 XI. Ordered sets

to their magnitude is of type w. Also the set of the terms of every infinite
sequence of different terms (a,, a, ...) ordered according to the magnitude
of the indices of those terms is of type wm. But the set of all positive num-
bers ordered according to their magnitude is not of type o.
Every ordered set of type wm is denumerable but not vice versa.
Every ordered set which is similar to a set of type w is obviously
(in view of the transitivity of the relation of similarity) itself of type o.
Each subset A, of an ordered set A(o) such that if ae A,, re A
and xoa (or aox), then we A,, is called a segment (or a remainder) of the
set A(o). A segment (or remainder) A, of a set A(o) is called a proper
segment (or remainder), if A, ~0 and A, 4 A. Obviously a segment of
a segment of a set A(o) is a segment of that set.
If a is a given element of an ordered set A(o), then the set of all
elements x of the set A(e) such that xea obviously forms a segment of
the set A(o) (which can be the empty set); we call it a segment corresponding
to the element a and denote it by A,(0).
Obviously, in a mapping of a set A(e) upon a set B(o’) which esta-
blishes their similarity, every segment of the set A(e) is transformed
into a segment of the set B(o’). The same applies to proper segments
and to remainders, as well as to proper remainders.

THEOREM 1. In order that an ordered set be of type w it is necessary

and sufficient that at be infinite and that every proper segment of that set
be finite.
Proof. If a given ordered set is infinite and each proper segment
of that set is finite, then obviously the same properties are possessed by
every ordered set that is similar to the given one. In order to prove the
necessity of the conditions of our theorem it suffices to prove that at
least one ordered set of type w satisfies them, e. g. the set N of all natural
numbers ordered according to their magnitude.
Let Q denote a proper segment of the set V. Thus VN—Q +0; let m
denote the smallest natural number belonging to N—Q. If we Q, then
“%<m, since if « >m, then in view of the fact that weQ, me WN and
that @ is a segment of the set NV, we should have meQ, contrary to
me« N—Q. Thus each element of the set Q is a natural number <m, whence
it follows that the set Q is finite. Therefore we have proved that every
proper segment of the set N is finite, and since the set WN is infinite, it
satisfies the conditions of our theorem. Thus we have proved that those
conditions are necessary.
Now suppose that an ordered set A(e) satisfies the conditions of
our theorem. Let a denote a given element of the set A(e), and A,(@)
a segment corresponding to it. If A,(o) = 0, then of course a is the first
 § 7. Sets of type w 207

element of the set A(o); let f(a) = 1. If A,(e) #90, then (according to
the definition of a segment corresponding to an element a) we also
have aéA,(e), whence A,(e) # A(o) and the segment A,(o) of the
set A(o) is a proper segment. According to the conditions of our theorem
it is thus a finite non-empty set; therefore its power, 4,(o), is a natural
number. Let f(a) = A,(o)+1; f(a) will be a natural number +1. In this
way a certain natural number f(a) has been associated with each element a
of the set A(g). Now suppose that a and b are two elements of the set
A(e) such that aeb. Thus the element a belongs to the segment A;/(0),
whence it easily follows that the segment A,(o) (since it does not contain
element a) is a proper subset of-the segment A,(0). Since the sets A,(0)
and A;(e) are finite, we have A,(0) < 4;(0), and thus f(a) < f(b). Thus
the function f establishes the similarity of the infinite set A(o) to a certain
subset (also infinite of course) of the set of all natural numbers ordered
according to the magnitude of the numbers which belong to it. But ob-
viously every infinite set of natural numbers ordered according to their
magnitude is of type w since we can arrange the numbers of that set
in an infinite sequence according to their magnitude. It follows that
the set A(oe) is of type w. Thus the conditions of our theorem are suffi-
cient and Theorem 1 is proved.
We also have the following
THEOREM la. In order that an ordered set be of type w it is necessary
and sufficient that it have no last element and that each proper segment of
that set be finite.
The proof of Theorem la is an easy modification of the proof of
Theorem 1.
EXERCISES. 1. Prove that one of two segments of the same ordered set is al-
ways a subset of the other.
Proof. Suppose that A, and A, are segments of an ordered set A(g) and suppose
that neither A,c A, nor A,c A,. Therefore we have A,—A, ~ 0 and A,—A,40 and
there exist elements a and b such thata « A,—A,, b « A,—A,. Since A,c A and A,c A,
we have ae A and be A, where a # b since a« A, and b € A,. Thus in view of the con-
nexity of the relation @ we have either agb or bea. If aeb, then, since 6 is an element
of the segment A,, we have, in view of the properties of a segment, a « A,, which is impos-
sible. If b = a, then, since ais an element of the segment A,, we should have 6 « A,, which
is impossible. Thus the assumption that neither A, c A, nor A, c A, leads to a contradic-
tion. Our theorem is thus proved.
2. Prove that if a non-empty set 4, 4 A is a segment of an ordered set A(g) and
if we put A, = A—A,, then [4,, A,] will be a cut of the set A(o).
Proof. Suppose that 4,,0 #4 A, # A, is a segment of the set A(oe); A, = A—A,,
a,<«A,, a,¢ A,. Thus a, «A, a,¢ A, and a, a, since a, ¢ A,, a, ¢ A,. In view of the
connexity of the relation @ we thus have either a, ga, or a,ea,. But if we had a,ea,,
then, since @ belongs to A,, we should have a, « A,, which is impossible. Thus a, ea,.
208 XI. Ordered sets

We have proved that if a,<«A,, a, ¢-A,, then a,ga,, and since A = A,+A,, it follows
that [A,, A.] is a cut of the set A(Q), q.e. d.
From Exercise 1] in § 5 and from the definition of a segment it immediately follows
that if [4,, A,] is a cut of the set A(@), then A, is a segment of the set A non-empty
and, different from A. Thus we see that in order that the decomposition A = A,+ Ag,
of an ordered set A(g) into two disjoint components determine the cut [A,, A,]
of the set A (g) it is necessary and sufficient that the set A, be a segment of the set A (g)
SUCH unatn Oka eA — eA
3. Prove the following theorem:
In order that an ordered set be of type w it is necessary and sufficient that it have the
following properties: 1) it should contain a first element, 2) it should contain no last element,
and 3) each cut of it should give a jump.
Proof. It is easy to prove that the set of all natural numbers ordered, according
to their magnitude has the, properties 1), 2) and 3), whence it follows that the condi-
tions of our theorem are necessary.
Now suppose that an ordered set A(o) satisfies conditions 1), 2) and-3). According
to 1) the set A(g) has a first element: let us denote it by a,. Now let us define by induc-
tion a certain infinite sequence of elements of the set A(@), a,, d2,..., in the following
way. Suppose that, for a given natural number n, we have defined the terms a, dg, ..., A,
(which is true for n = 1). Let A, = Aa,(e)+ {a,} and A, = A—A,. In view of a, « A,,
we shall have A, ~ 0, and also A, # A since the set A, has a last element, a,, and,
according to 2), the set A has no last element. Thus A, = A—A, ~ 0. It is easy to
verify that, in view of the definition of the sets A, and A,, for x « A,, ye A,, we have
xoy. Thus the sets A, and A, determine a certain cut [4,, A,] of the set A(g). In virtue
of property 3) that section must give a jump; according to the definition of a jump (§ 5),
the set A, must have a first element; let us denote it by a,,,. We shall have of course
a, 0A, 41 Since a, « A,, a,,, € A,, and between a, and a, ,, there is no element of the setA
since such an element « would satisfy the formulae a, ga and xea,,, and it would belong
neither to the set A, nor to the set A,(a, 4, being the first element of that set).
Thus we have defined by induction an infinite sequence a,, a,,... of elements of
the set A(g); we have a,ea,ea3ea,..., and between a, and a, ,, there is no element of
the set A(g). It easily follows by induction that between a, and a,,, lie only the ele-
ments dz, dg, .--, M1, Of the set A(e) (for n= 1, 2,...). Further, it follows that if
there existed an element a of the set A(@) different from all the terms of the sequence
@&, G,,..., then we should have either aga, for n=1,2,..., or a,oa for n=11, 2,...
The first case cannot hold since a, is the first element of the set A(g). And if a, ea for
nv —1,2,..., then let B,— {a,, a,,...), B= A—B,. Since we have proved that for
y « A—B, we must have a, oy for n = 1, 2,..., therefore for x « B,, y « B, we have xoy.
Of course B, 4 0: if we also had B, 4 0, the sets B, and B, would determine a cut
[B,, B,] of the set A(e), and, in virtue of property 3), the set B, would have a last ele-
ment, which is impossible in view of B, = {a,, a@,,...} and a,ea,o0a;... Thus we have
B, = 0, whence A(o) = B, = {a,, dz, ...}, Which, since a,ea,Qa3..., proves that the set
A(@) is of type w. Thus the conditions of our theorem are sufficient.
4, Prove the following theorem:
In order that an ordered set A(g) be of type w it is necessary and sufficient that it
have the following three properties:
1° it should have a first element, a,,
2° each element of the set A(@) should have its consequent,
3° if a,« XC A(e) and the set X contains the consequent of each of its elements,
then X = A (see Kuratowski and Mostowski [1], p. 151, Theorem 1).
 § 7, Sets of type w 209

It will be observed that an ordered set A(g) which satisfies only conditions 1°
and 2° is not necessarily of type w. It is even easy to give an example of a set of rational
numbers, not being of type w, ordered according to the magnitudes, having a first ele-
ment and such that every element has the consequent, and every element, except the
first one, has the antecedent.
As can easily be verified, such is the set composed of the numbers 1—1/n, 1+ 1/n
and 3—1/(n+ 1) where n= 1, 2,... We have here

Oy he a Le a Se 8 Sh
There exist also non-denumerable ordered sets A(g), having the properties 1°
and 2° and such that each of its elements that is different from the first element has
an antecedent.
5. Prove the following theorem:
In order that an ordered set be similar to the set of all integers ordered according to
their magnitude it is necessary and sufficient that it have neither a first nor a last element
and that each cut of it give a jump.
6. Prove that if an ordered set A(@) is of type w, then by removing an arbitrary
/
element of that set we shall obtain a set B(@) which is also of type ow.

8. Sets of type 7. We say that an ordered set which is similar to the

set of all rational numbers ordered according to their magnitude is of
type n. Thus for instance the set of all rational numbers ordered accord-
ing to their magnitude is of type 7. It is also easy to prove that the set
of all rational numbers r such that —1 <*r<1 ordered according to
their magnitude is of type 7 (cf. XI.4, Exercise 1).
THEOREM 1. In order that an ordered set be of type 7 it is necessary
and sufficient that it have the following three properties: 1) it should have
neither a first nor a last element, 2) tt should be dense and 3) it should be
denumerable.
Proof. As can easily be seen, the set of all rational numbers ordered
according to their magnitude, which is of type 7, has the properties 1),
2) and 3). Since those properties are retained in mappings establishing
similarity of sets, it follows that the conditions of Theorem 1 are nhe-
cessary. In order to prove that they are also sufficient it is enough
to show that if each of two ordered sets A(o) and B(o’) satisfies condi-
tions 1), 2) and 3), then those sets are similar.
Let A(o) and B(o’) be two ordered sets having the properties 1), 2)
and 3). In virtue of property 3) our sets are denumerable and there exists
an infinite sequence
(8.1) | Ons Gay Tay tees
formed of all the different elements of the set A(o), and an infinite se-
quence
(8.2) DiO sg lige es
formed of all the different elements of the set B(o’).
Cardinal and ordinal numbers 14
210 XI. Ordered sets

Now we shall define by induction two infinite sequences,

(8.3) Pir Po» Psy +

and
(8.4) G19 day W39 v0+5

the former composed of the terms of sequence (8.1) and the latter of the
terms of sequence (8.2), and a function f of the terms of sequence (8.3)
as follows.
Let. p,= 4,, ¢,= 01, f(p1) = g@. Now let » denote a given natural
number >1 and suppose that we have already defined all terms p,, q@,
Diss Qa, +2) Potts In=1) ond (thatthe: function: 7(p;)—4¢;, Lor 4— 1,2, ee,
nm—1, establishes the similarity of the sets {p,, po, -.-, Pn-rt(e) and
NG eGoaicoes Une (On) aWLC D piss ULUe 4Or a7) Ly,
The ordered set {p,, Ps, .--5 Pn—1}(e), being finite, it is, in virtue of
Theorem 1 of X1.6, similar to the set {1, 2, ..., ~—1}(<), and if y(t) = pz,
(¢=1,2,...,n—1) is a function establishing similarity between the
set {1,2,..., 7—1}(<) and the set {p;, Do, :-»5 Pri} (0), then
(8.5) Pky 0 Pky @ Pkg +++ Pkn—» Pkn—1

and since the function f(p;) = qi (¢ = 1, 2,..., »—1) establishes similarity

between the sets {p1, Po, --, Pn—1}(@) and {91, Qs, --- Gn—1}(@'), also

(8.6) Yi, © TkgO Wkg@ +++ Ung Uke_a

Next we shall distinguish two cases:
1) n is an even number. Let us denote by p, the first term of sequence
(8.1) that is different from p,, Po, ..-, Pn—1- If props, then let us denote
by q, the first term of sequence (8.2) for which q,o0’g; with 7 <n, and
if Px,_,@Pn, then let us denote by q, the first term of sequence (8.2) for
which 40d, with 7 <7; in both cases such a term q, exists in view of
property 1) of the set B(e’). In both cases let f(pn)= qn.
Finally, if neither p,opx, nor pz,_,@Pn, then, in view of the fact that
elements ,, Po, +», Pn are different from one another and belong to the
ordered set A(e), and that the relation @ is connected, we have px, OPn OP kn_4
and, by (8.5), we conclude that there exists a natural number 7 < »—1,
such that px,0Pn0Px,,,- Let us denote. by q, the first term of sequence
(8.2) for which gx,0'Gn0'Gx,4,3 Such a term exists in view of property (8.2)
of the set B(o’). Thus, by (8.5) and (8.6), we have:

Pk,
CP koQ +++ OPKk,CPnOPk,
410 +++ OPkn_y
and
Gk O'UhgO +++ O'Gk, O'InO'Vkpy
yO +++ OFOkn_ 9
 § 8. Sets of type 7 Zi

which proves that if we take f(p,)=q,, then the function f(p;) = q:

for «<n establishes the similarity of the sets {p,, p.,..-, Pnt(e) and
{91> Yay +9 Ini (O’). This will hold both in the case of p,op,, and in the
case Of Px, ODn-
2) n is an odd number. Here we must repeat the reasoning of case
(8.1) replacing everywhere the letter p by the letter g and vice versa,
o by e’ and vice versa, the set A(e) by the set B(o’) and vice versa, se-
quence (8.1) by sequence (8.2) and vice versa, leaving only the formula
1 (Pn) = Gr unchanged.
In this way we shall define by induction sequences (8.3) and (8.4)
and a function f such that f(p:) = q; for i= 1, 2,...; that function will
establish similarity between the sets {p,, po, ...}(0) and {q, go, ..-}(0’).
If we also proved that sequence (8.3) differs from sequence (8.1), and
sequence (8.4) from sequence (8.2) at most in the order of their terms,
Theorem 1 would be proved.
Since the terms of sequence (8.3), as follows from its definition,
are different terms of sequence (8.1), and the terms of sequence (8.4)
are different terms of sequence (8.2), therefore it suffices to show that
each term of sequence (8.1) is a term of sequence (8.3), while each term
of sequence (8.2) is a term of sequence (8.4).
Now, from the definition of sequences (8.3) and (8.4) it éasily follows
that, for n= 1, 2,..., the sequence pi, po, ..-» Pong Contains all the terms
of the sequence a,, a,,...,@,, and the sequence 4q,, go, .--» Gan—1 contains
all the terms of the sequence b,, 0.,..., 0, (since p,= 4, 9,= 0,, and,
for natural k, as po, has been taken the first term of sequence (8.1) that
is different from p,, pz, .--, Pex—1, ANd AS ox41 — the first term of sequence
(8.2) that is different from 4q,, 2, ---, ax).
Thus we can regard Theorem 1 as proved.
As an easy conclusion from Theorem 1 we find that:
The set of all rational numbers, the set of all positive rational num-
bers, the set of all rational numbers lying inside an arbitrary interval
(a,b) where a <b, the set of all finite decimal fractions and the set of all
real algebraical numbers, each of those sets ordered according to the
magnitude of the numbers belonging to it, are similar.
From Theorem 1 it also easily follows that if we remove from an
ordered set of type 7 an arbitrary finite number of any elements of that
set, then the remaining set will be of type 7.
An easy modification of the proof of Theorem 1 enables us to preve
THEOREM 2. Hvery denumerable ordered set is similar to a set of ra-
tional numbers ordered according to their magnitude. :
Proof. Let A(e@) denote a given denumerable ordered set, (1) an
infinite sequence composed of all different elements of the set A(o),
14*
212 XI. Ordered sets

(2) the infinite sequence of all different rational numbers, and as 0’

let us take the relation <. Let us repeat the proof of Theorem 1 with the
following difference: in case 2) (of an odd m) let us define elements p,
and qg, in the same way as in case 1). The function f(p;) = q; for1=1,
2,... obviously establishes similarity between the set A(o) and a certain
subset of the set of all rational numbers ordered according to their
magnitude. Thus we may regard Theorem 2 as proved.
From the proof of Theorem 2 and from the effective denumerability
of the set of all rational numbers immediately follows
THEOREM 2a. For every effectively denumerable ordered set we can
define a function establishing the similarity of that set to a certain set of
rational numbers ordered according to their magnitude.
Theorem 2 shows that the examination of denumerable ordered
sets can be reduced to the examination of sets of rational numbers
ordered according to their magnitude.
It can be proved that every denumerable ordered set is similar to
a certain proper subset of itself (see XII.3, Exercise 25). With the aid
. of the axiom of choice, however, it can be proved that there exists a set
of real numbers, equivalent to the continuum, which, when ordered
according to the magnitude of the numbers which compose it, is not
similar to any proper subset of itself (Dushnik and Miiller [1], p. 322-326).
If A(o) is an ordered set and B(o) its subset, we say that the set
B(o) is dense in the set A(o) if for each two elements a and b of the set
A(o) such that agob there exists at least one element x of the set B(o)
such that agxreb. A set which is dense in itself is dense, and vice versa.
It is also easy to prove that if A(e) 0 B(e) D C(e@) and if the set C(o) is
dense in the set B(o) while the set B(e) is dense in the set A(o), then
the set C(o@) is dense in the set A(o). In other words: a set which is dense
in a dense subset of an ordered set is dense in that set.

EXERCISES. 1. Prove the following theorem (Skolem [1)):

If A = A(e) and B = B(0’) are dense denwmerable ordered sets having no first and
no last elements, and if A = A’+A” and B= B’+B” where A’A”’ = B’/B” =0 and
where the sets A’ and A” are dense in the set A and the sets B’ and B” are dense in the set B
then there exists a function f establishing similarity between the setsA (e@) and B(0’) and such
that f(A) =B" and. f(A) = B.
Proof. The sets A’, A’’, B’, B’” being dense in dense denumerable sets having
no first and no last elements, are obviously themselves denumerable, dense and without
first or last elements. Let
(8.7) Gis» Oey cong
(8.8) GMO tens
(8.9) OO es,
(8.10) bY, BY, ...,
 § 8. Sets of type 7 213

be infinite sequences formed, respectively, of all! different elements of the sets 1’,
Mie basanda Bia.
Now let us define by induction four infinite sequences

(8.11) FOS (Dao atop

(8.12) Dis Do's os

(8.13) iin OR Seocoo

(8.14 ) OHO Rene

whose terms belong, respectively, to the sets A’, A’, B’, B’, and a function f of the
terms of sequences (8.11) and (8.12) as follows.
(Let p= 4, G =O, pi = ay Gi! = bY, F(pi) = a Flpy') = a"
Now let » be a natural number >1 and suppose that we have already defined the
terms Pi»4;, p;’ and gi’ for i < n, and a function f establishing similarity between the set
<Pie Pi 1 Pas Paces Deis Pag (@eana. whe set 4%, Q4. 9a dy es Gy dy ys (@) and such
that f(p;) = q and f (pi) = gy’ for 1< n. We shall distinguish two cases:
1) nm is an even ey As p} we shall take the first term of sequence (8.7) dif-
ferent from p;,p3,..-,p,_, and as p}’ the first term of sequence (8.8) different from
Dr's Po’, «+» pi_,- In view of the properties of the sets B’ and B”, we shall find (as in the
proof of Theorem 1) the first term of sequence (8.9), q/ such that if we take f(p/) = qi,
then the function f will establish similarity between the sets {p/, pi’, ..-. Ph_1> Pi_1> p}(@)
and {Q1, 91's-++> Gr_1> Mr_1> 9,}(0’); next we shall find the first term of sequence (8..0),
gq, for which if we take f(p/’) = q/’, then the function f will establish similarity between
the sets {p1, Pi’, -+> Daa» Pn_1> Pn» Pn3(Q@) amd {G55 G45 +> Una Ina» Ine In 3 (0’)-
2) n is an odd number. Here we repeat the reasoning of case 1), replacing every-
where the letter » by the letter q and vice versa, B’ and B” by A’ and A” and vice
versa, @ by @’ and vice versa, sequences (8.7), (8.8), (8.9) and (8.10), by (8.9), (8.10),
(8.7) and (8.8) respectively, and leaving the formulae f(p/,) = q/ and f(pi’) = qf’ un-
changed.
As in Theorem 1, we prove next that sequences (8.11), (8.12), (8.13) and (8.14)
differ from sequences (8.7), (8.8), (8.9) and (8.10) respectively at most in the order of
their terms, and that the function f establishes similarity between the sets A(g) and
B(o’), and we have f(A’) = B’ and f(A”) = B”, q.e. d.
2. Give a decomposition of the set & of all rational numbers ordered according
to their magnitude into an infinite series of disjoint sets, dense in LR.
Answer. Let p, denote the n-th successive prime number?!) and, for » > 1,
let A, denote the set of all numbers of the form 1/py where k is a natural nuntber and /
an integer, prime with respect to p n >
let A, = R—-(A,4+A,+...). R= A,+4,4+... will
be the required decomposition.
3. Give the decomposition of the set & of all rational numbers ordered according
to their magnitude into 2*° almost-disjoint sets *), dense in R.
Answer. Let p, denote the n-th successive prime number and @(n) — the set of all
numbers of the form l/p, where J is an integer prime with respect to p, and such

1) Since we do not know any simple formula expressing the n-th successive prime
number as function of n, the infinite sequence p, (n= 1,2,...) may be defined here
in a different way, e.g. by taking p, = 2°41 for n=1, 2,...
2) As regards the definition of almost-disjomt sets (see [V.14).
214 XI. Ordered sets

that || < pz. For real # > 0 let

co

= Ol Q"(2Hna+ 1)),
n=1

and
=R-
> Ala),
z2>0

where > denotes the sum extended to all positive real numbers.
z>0

R=
y Ate)
r>0

is the desired decomposition.

4, Prove that if A = A(o) and B= B(o’) are dense denumerable ordered sets,
having no first and no last element, and if

co co

Ap==y)
= gl@) aad
: oe
ez (72)

where A 4 — B©B® — 0 for k <1 and where, for n = 1, 2,..., the sets A and B™
are dense in the sets A and B respectively, then there exists a function f establishing
similarity between the sets A(o) and B(g), such that f(A™) = B® for n= 1, 2,...
5. Prove that every dense ordered set contains a subset of type 7.
6. Prove that if A is a set of type 7, B its segment, and C the remainder of the
set A corresponding to that segment, then at least one of the sets B and C is of type 7.
Proof. Obviously if the set B contains no last element and is not empty, then B
is of type 7; otherwise C is of type 7.
7. Let & denote the set of all rational numbers ordered according to their magni-
tude. Give an example of such a set Bc R that neither of the sets B and k—B is of
type 7.
Answer. An example of such a set B is the set formed of all rational numbers
lying inside the interval (0,1) and of the numbers 3 and 4.
8. Prove that if A is a set of type 7 and B — any subset of the set A, then at least
one of the sets B and A—B contains a subset of type 7.
Proof. If the set B is dense in the set A, then obviously it is of type 7. Otherwise
there exist in the set A two elements a and b, a 4 b, between which there is no element
of the set B. The set of all elements of the seb A lying between a and b will be a subset
of type 7 of the set A—B.

9. Dense ordered sets as subsets of continuous sets. We shall prove

THEOREM 1. Hvery dense ordered set is similar to a subset of a continuous
ordered set, dense in that set.
Proof. Let A(e) denote a given dense ordered set. Let us denote
by b the set composed of the empty set and of all segments of the set A (oe)
that have no last element. According to Exercise 1 (XI.7) the set B will
be ordered by the relation 0’ where, for b, « B, b, « B, b,0’b, means that
6; Gb, and b).210e
 \

§ 9. Dense ordered sets as subsets of continuous sets 215

With each element a of the set A let us associate a segment A,(0)

of the set A(e) composed of all elements « of the set A such that woa (in
particular the set A(e) may be empty). From the assumption that the
set A(e) is dense it follows that the set A,(o) has no last element. There-
fore A,(e)« B for ae A. Let us denote by B’ the set of all sets A,(0)
where ae A. Thus B’C B. We shall prove that A(o)~ B’(0’).
Let us suppose that a,« A, a,« A, a,0a,. Obviously we shall have
Ag(e)C Aa(e) and Ag(e) # Ago), whence A,,(0)0’da(Q). Since the
function f(a) = A,(@) is invertible in the set A and f(A) = B’, the formula
f(a,) o’f(a.) for a,e« A, aye A, aoa, proves that the function f establishes
similarity between the ordered sets A(o) and B’(0’).
Now we shall prove that the set B’ is dense in the set B(o’). Let b,
and b, be any two elements of the set B(o’) such that 6,0’b,. From the
definition of the set B(o’) it follows that 6, is either the empty set or
a segment of the set A(o) having no last element, and b, is a segment
of the set A(o) having no last element, where b, C b, and b, 4 b,. In both
cases there exists an element a of the set b,—b, and, since the segment b,
of the set A(e) has no last element, there exists an element a, of the
set b, such that aea,. Let b= A,,(e): we shall have b« B’, while ob-
viously b,c bCb, and b,4Ab+~b, since ac b—b,, a,€¢b,—b. Therefore
b, 0b, bo'b,, and b « B’. Thus we have proved that for any two elements
b, and b, of the set B(o’) such that b, 0’b, there exists an element b of the
set. B'C B such that b, 0’b and bo’b,. This proves that the set B’ is dense
in the set, B(o’), q.e. d.
For the proof of Theorem 1 it remains to show that the set B(o’)
is continuous.
Since the set B(o’) contains the subset B’, which is dense in it, it
is itself dense. In order to prove the continuity of the set B(o’) it suffices
to show that it has no gaps. Therefore let us suppose that the cut
[B,, B,| of the set B(e’) determines a gap.
Every element of the set B, C B(g) is either the empty set or a segment
of the set A(o) having no last element, the set 6, having more than one
element since it is non-empty and has no last element. Let us denote
by s the sum of all sets constituting elements of the set B,: it will ob-
viously be a segment of the set A(e); it will have no last element since
each element a of the set s is an élement of at least one of the segments b
of the set. A(o) belonging to B,, 7. e. of segments having no last element:
thus there exists an element a’ « b such that aoa’, and, in view of a’ «be B,
we have a’ «s. Therefore s « B, and it follows from the definition of the
set s that b Cs for b « B,, whence do's for be B,, b A 8. If we had sc B,,
then s would be the last element of the set B,, which is impossible. In
view of s « B, we thus have s « B,. Since the set 6,(0’) has no first element,
216 XI. Ordered sets

there exists an element b, « B, such that b,0’s, whence ),C s and b, 48.
But, in view of b,¢ B,, we have bo’b, for each element be B,, whence
bCb,; thus it follows from the definition of the sets s that sC b,, which
is impossible since b,C s and b, #s8.
Thus the assumption that there exists a cut [B,, B,| of the set B(o)
determining a gap leads to a contradiction. We have proved Theorem 1.
Remark. If the dense ordered set A(ge) had no first element, then,
in the proof of Theorem 1, when defining the set B we should not have
to include the empty set among its elements; the set B(o’) would also
have no first element. It is likewise obvious that if the set A(o) has no
last element, then the set B(o’) also has no last element.
Theorem 1 might also be expressed as follows:
Every dense ordered set is a dense subset of a continuous ordered set.
This should be understood in the following way: if A(o) is a dense
ordered set then there exists a continuous ordered set B(o’) containing
the set A, the relation 0’ being equivalent in the set A to the relation 0,
and the set A(o’) is dense in the set B(o’).
This theorem is usually proved by joining to the set A(o), as new
elements, all its cuts [A,, A,] giving gaps, and defining the relation o
as follows: for ae A, ae[A,, A] holds if and only if ae A,, and[A,, A,]oa
holds if and only if ae A,; finally if [A,, A,] and [Aj, Az] are two diffe-
rent cuts of the set, A giving gaps, then we write [A,, A,]o[Ai, Az] if
and only if A,C Aj. Then we prove that the new set obtained in this
way will be continuous and that the set A(o) will be dense in it.
In particular, if we applied a proof of this kind to the set of all rational
numbers ordered according to their magnitude, we should obtain Dede-
kind’s theory of irrational numbers, the irrational numbers being the
gap-giving cuts of the set of all rational numbers (see Dedekind [1]).
THEOREM 2. Two continuous ordered sets having no first and no last
elements and containing, respectively, similar dense ordered subsets are
similar.
Proof. Let A(e) and B(e’) be two continuous ordered sets having
no first and no last elements, and let A’(o) be a dense subset of the set
A(o) and B’(o’) a dense subset of the set B(o’), A’(e) + B'(o’). Thus there
exists a function f establishing the similarity of the sets A’(o) and B’(0’).
Now let «« A—A’. Let us denote by O the set of all elements a « A’ such
that aoxv. Since the set A(o) has no first element and the set A’(o) is dense
in A(e), we easily conclude that the set C is not empty and has no last
element. Similarly, since the set B(o@) has no last element, we conclude
that the set @,= A’—C, is non-empty and has no first element. Thus
[C,, C.] is a cut of the set A’(o) determining a gap.
 § 9. Dense ordered sets as subsets of continuous sets 217

Since the function f establishes similarity between the sets A’(@)

and B’(o’), it obviously follows that [f(C,), f(C.)] will be a cut of the set B’
determining a gap. Let us denote by HF, the set of all elements b of the
set B’(o’) such that bo’b, for b, « f(C,) and let H, = B—EH,. We have of
course {(C,) C E,, f(C,) C EF, and thus sets H, and FH, are not empty. It is
also easy to see that [#,, #,] is a cut of the set B(o’). _
In the set H,(0’) there is no first element. Indeed, suppose that
ye H,. Thus we have y¢ H, and there exists an element Dd, « f(C,) such
that yo'b, does not hold, whence, in view of the connexity of the relation 0’,
in the set B we have either y = Bb, or b,0’y. If y= b,, then, since there
is no first element in the set f(C,), there exists an element bz « f(C,) such
that b20’y. In any case there exists an element y’ « f(C,) such that y’o’y.
But, in view of y’«f(C,) and of the definition of the set H,, we have
y’ ¢ E,, whence y’ « H,. Thus in the set EH, there is no first element. Since
the set B(o’) is continuous, the section [H,, #,] cannot give a gap, and — as
we have proved — the set H, has no first element; therefore the set £,
must have a last element — let us denote it by f(z). In this way we have
defined the function f(#) for x« A—A’. The proof that the function /
establishes the similarity of the sets A and B presents no difficulties.
Thus we can regard Theorem 2 as proved.

10. Sets of type 4. An ordered set which is similar to the set of all
real numbers ordered according to their magnitude is said to be of type 2.
Thus, for instance, the set of all real numbers ordered according to their
magnitude is of type 4. The set of all real numbers lying inside an arbi-
trary interval (a,b) and ordered according to their magnitude is also
of type A. (See Exercise 2 in XI.4).
THEOREM 1. Jn order that an ordered set be of type A tt is necessary
and sufficient that it have the following three properties: 1) it should have
neither a first nor a last element, 2) it should have a denumerable subset
dense in tt, 3) it should be continuous.
Proof. As can easily be seen, the set of all real numbers ordered
according to their magnitude, which is of type A, has the properties 1),
2) and 3). (A denumerable subset dense in it is formed, for instance, by
the set of all rational numbers, while the continuity of the set of all
real numbers is proved in Analysis,’ in Dedekind’s theory of irrational
numbers (cf. § 9). Since properties 1), 2) and 3) are obviously valid in
transformations establishing the similarity of sets, it follows that the
conditions of Theorem 1 are necessary. In order to prove that they are
also sufficient it is enough to show that if each of two ordered sets,
A(o) and B(o’), satisfies conditions 1), 2) and 3), then those sets are
similar.
218 XI. Ordered sets

Let A(o) and B(o’) be two ordered sets having properties 1), 2)
and 3). According to property 2) the set A(o) has a denumerable subset
A,(o) dense in it and the set B(o’) has a denumerable subset B,(o') dense
in it. The set A,(0), aS a dense subset of the set A(o) which has property 1),
must obviously itself possess that property. Similarly, the set B,(e’) has
property 1). Further, the set 4,(o) being dense in a set that is continuous,
i. e. dense, is dense itself. Similarly the set B,(o0’) is dense. Thus the sets
A,(o) and B,(0’) satisfy the conditions of Theorem 1 of XIJ.8 and are
both of type 7, 7. e. similar. The continuous sets A(e) and B(oe’) thus
have neither a first nor a last element and contain, respectively, the
subsets 4,(e) and B,(o’) dense in them, those subsets being similar. Accord-
ing to Theorem 2 of XI.9 the sets A(e) and B(o’) are thus similar. We
have proved Theorem 1.
Tt will be observed that there exist ordered sets which satisfy condi-
tions 1) and 3) but are not of type A. Indeed, let Y denote a square com-
posed of points (~, y) of a plane, where 0 <a%<1, 0<y <1, and let
us order its points by means of a relation o where (a, b)o(c, d) means
that either a <¢ or a=ec and b <d. We leave it to the reader to prove
that, after removing from a set QY ordered in this way the points (0, 0)
and (1,1), we shall obtain an ordered set U satisfying conditions 1)
and 3). However, the set U does not satisfy condition 2), since, in view
of (7, 0)o(x, 1) for every real number x such that 0<a#< 1 each dense
subset of the set U contains at least one element (x, y,) where 0 < y, <1
for 0 <a <1, 2. e. itis obviously of the power of the continuum. It follows
that the set U is not similar to any subset of a set of type 4. Thus it is
not every ordered set of the power of the continuum that is similar to
a certain set of real numbers ordered according to their magnitude.
It can easily be seen that every set of type / is such that if we remove
from it an arbitrary element, the ordered set obtained in this way will
no longer be of type A (since it will have a gap).
If A(e) is an ordered set and a and b are two of its elements such
that aoeb, then the set composed of the elements a and b and all those
elements w of the set A(o) for which aov and «ob is called a (closed)
interval (a,b) (X1.5, Exercise 2). Two intervals, (a,b) and (c, d), where
aoc, are said to be non-overlapping if bec or b = c. It is easy to prove that
every ordered set A(g@) containing a dense denumerable subset is such that
4) every set of non-overlapping intervals in it is finite or denumerable.
Indeed, suppose that P= {p,, p.,...} is a denumerable subset of
the set A(o) dense in it, and let Z denote the set of non-overlapping
intervals of the set A(o). Let us associate with each interval belonging
to Z the first term of the sequence p,, p., ... that belongs to that interval
and is different from its end-points. (Such an element exists since the
 § 10. Sets of type 4 219

set P is dense in A(o)). In this way to different intervals of the set Z

obviously correspond different elements of the denumerable set P (since
the intervals of the set Z do not overlap). It follows that the set 7% is fi-
nite or denumerable. (Cf. Theorem 1 in III.3).
The question, still unsolved, whether every ordered set that satisfies
conditions 1° and 3° of Theorem 1 and condition 4° is of type 4 is called
the problem of Suslin (see Suslin [1}) *).

EXERCISES. 1. Prove that the set composed of number 0 and of all real num-
bers # such that |v%| > 1 ordered according to the magnitude of the numbers that
compose it is of type A.
2. Prove that if we remove from the set X of real numbers ordered according
to their magnitude the sum of an arbitrary set of non-overlapping intervals each of
which contains only one of its end-points, then the remaining set will be of type A.
3. Find how many different subsets of type /4 are contained in the set X of all real
numbers ordered according to their magnitude.
Solution. The number of such subsets is equal to the continuum. Suppose that
Ac X is a subset of the set A that is of type 4. Let x denote an element of the set X —A,
A, the set of all those numbers of the set A that are less than «, A, the set of all those
numbers of the set A that are greater than x. Let us denote by 6, the upper bound of
the numbers of the set A,, taking b, = — co if A, = 0, and by 6, the lower bound of
the numbers of the set A,, taking b, = + oco if A, = 0; let us associate with number «
the interval 6(x) = (b,, b,). We have here of course b, < «4 < b,, b, = db, being impos-
sible since then the cut [A,, A,] of the set A would give a gap, which is impossible in
view of the set A being of type 4, i. e. continuous. From the continuity of the ordered
set A it easily follows that one and only one of the numbers b, and b, belongs to the
set A. It also follows that if « and w’ are two different elements of the set X —A, then
either 6(”%) = 6(a’) or the intervals d(x”) and d(x’) do not overlap. Hence we draw the
conclusion that the set A will be obtained by removing from the set X the sum, finite
or infinite, of non-overlapping intervals, each of them containing only one of its end-
points; let us denote that sum by (A); we shall have A = X—S(A). But, in virtue
of Theorem 1 of III.3, a set of non-overlapping intervals is finite or effectively denumer-
able. Thus in order to determine the sum (A) it suffices to give a finite or infinite se-
quence of left-hand and right-hand end-points of intervals whose sum is (A), and another
sequence of those end-points which belong to the intervals in question. The set p(A)
can thus be determined by two infinite sequences of real numbers and since, as we know,
there are 2“° pairs of such sequences, there are at most 2*° sums S(A), whence it fol-
lows that the set X has at most 2*° subsets of type 4.
On the other hand, every interval without end-points is a type / subset of the
set X, whence we conclude that there are > 2*° such subsets.
Thus we have proved that the set X has 2°° different subsets of type A, q. e. d.
Remark. It is possible to construct a family, of the power of the continuum, of
disjoint subsets of the set X each of which is of type 4. Indeed, let A denote the set

1) Several works have been devoted to this problem, e.g. Kurepa [1], p. 2-4,
Maraham [1], Kurepa [3], p. 187, Kurepa [2], p. 1051, Kurepa [4], p. 327 and p. 1084,
Marezewski [1], p. 303, Inagaki [1], p. 25-46, Inagaki [2], p. 145-162 and 191-201,
Miller [1], p. 673-678, Sierpinski [46].
 220 XI. Ordered sets

of all real numbers lying inside the interval (0,1) and being of the form

va
co
my —2n

n=]

where a, (n= 1, 2,...) is an infinite sequence, formed of the numbers 0 and 1 and
containing infinitely many terms <0. It is easy to prove that the set A, ordered accord-
ing to the magnitude of the numbers which belong to it, is similar to the set X of all
real numbers lying inside the interval (0,1), ordered according to their magnitude,
i.e. it is of type 2. The function establishing the similarity of the sets 4 and X is
foe} co
l= on 7 —n
A> a, 2 ae
n=1 n=1

Now let us denote by 4,, for ¢¢ A, the set of all numbers «+ ¢/2 where x « A. The
sets A, where t « A are of course similar to the set A, i. e. they are of type /, and ob-
viously A,A, = 0 forte A, we A, t #u. The sets A,, where t « A, form a family, of the
power of the continuum, of disjoint subsets of the, set X each of which is of type 2.
4. Prove that if Q denotes the set of all irrational numbers ordered according
to their magnitude, and if we remove from it an arbitrary finite or denumerable set,
then we shall obtain a subset of the set Y which is similar to Q.
Proof. Let X denote the setwof all real numbers, & the set of all rational numbers,
D an arbitrary finite or denumerable set of irrational numbers, all of them ordered, ac-
cording to the magnitude of the numbers which belong to them. From theorem 4 it
follows that the sets R and K+D are similar: thus there exists an increasing function
f(z), defined in the set & and mapping that set upon the set R+D. As in the proof of
Theorem 2 of XI.9, we prove that there exists a (unique) increasing function of a real
variable, y(#), such that p(x) = f(a) for « « R. The function g(a) obviously establishes
similarity between the sets 9 = X—R and Q—D = (X—R)—D.
5. Prove that every ordered set containing a denumerable subset, dense in it
is similar to a certain set of real numbers ordered according to their magnitude.
ci] Proof. The proof should be based on theorems 6 and 8.
6. Prove that every continuous ordered, set contains a subset of type 4.
7. Construct a set of disjoint intervals contained in the interval (0, 1) which,
‘when ordered according to the magnitude of the abscissae of the left-hand end-points
of the intervals belonging to it, is of type 7.
Solution. Let us divide the interval (0,1) into three equal intervals 6), 0,, d2
and let us assign to a set Z the middle interval, i. e. 6, = (4, 3). Each of the remaining
two intervals 6, and 6, let us again divide into three equal intervals 699, do, Oo, and
Seo, Oo1, O22, and let us assign to the set Z the middle interval of each triplet, 7. e. the
intervals 69, = (4,4) and 6,, = (4, 8). Let us repeat the same procedure with regard
to each of the remaining four intervals 699, do2, 5205 022, dividing it into three equal
intervals, assigning the middle one to the set Z, etc. In this way we shall obtain an in-
finite sequence of intervals

(10.1) Or; Oo1> Oer> Ooo» Ooe1> Oo01> Ooa1 > Oooo1> Ooo2 ? O2001> eens,

composed of the interval 6 and of intervals of the form Once opt? where k is a natural
number and a,, a,..., a, are numbers 0 or 2. Those intervals, ordered according to the
magnitude of their left-hand end-points, form a set of type 7, since they obviously
satisfy the conditions of Theorem 1 of XI.8.
 § 10. Sets of type A 221

It is easy to prove by induction that the left-hand end-point of the interval

ayap...0,1 18 the number

‘and its right-hand end-point is the number

ay Ay , Oy, “a)
DP [isa ere a at pee
SS Dia Roa

It follows that every number lying inside the interval Onesg++ Rl

or constituting
its right-hand end-point gives an expansion into a triadic fraction, with infinitely many
digits different from 0, whose (k-+ 1)th digit is 1. -
Conversely, if # is a number from the interval (0,1) whose expansion into a triadic
fraction with infinitely many digits different from 0 has at least one digit equal to 1,
then that number lies inside one of the] intervals of the set Z (or is its right-hand
end-point). Indeed, let 6, be the first digit of the triadic expansion of number «x (of
infinitely many digits different from 0) that is equal to 1. We shall have

% a, eT 1 Bpa Bure
u +—4.,..-+
3 | 92 - gh gk | 3h ta y gk +2 Foss

where a; =0 or 2 for 1=1,2,...,k—1, whence we immediately conclude that

Z€O G1 Ap..-
An _}
1 Le
It follows that the numbers lying inside the interval (0,1) and neither lying
inside any of the intervals of the set Z nor being right-hand end-points of those inter-
vals are >0 and <1 and such that their expansions into a triadic fraction with infinitely
many digits different from 0 contain no digit equal to 1. Let us denote by A the set
of all such numbers, and by D the (denumerable) set composed of numbers 0 and 1
and, of all right-hand end-points of the intervals of the set Z. For each number «« A
we thus have an expansion into a triadic fraction with infinitely many digits different
from 0 whose all digits are even, 7. e. the, expansion

. €
o=2(S4 2+.)

where ¢, = 0 or 1 fori = 1, 2,... and where c, # 0 for infinitely many natural numbers 7.
It can easily be seen that if we associate with number x of the set A the number

we shall obtain a similar mapping of the set A upon the set of all real numbers lying
inside the interval (0, 1). The set A ordered according to the magnitude of the numbers
that belong to it, is thus of type 4, i. e. of the power of the continuum. The set
O=A-+D, which is called the nowhere dense perfect set of Cantor is therefore also
of the power of the continuum.
It is easy to prove that if we have two different numbers of the set C, then either
they are the end-points of a certain interval of the set Z or infinitely many different in-
tervals of the set Z lie between them.
 CHAPTER XII

ORDER TYPES AND OPERATIONS ON THEM

1. Order types. Just as the concept of equivalence (equality of power)

leads to the concept of cardinal numbers, so does the concept of similarity
between ordered sets lead to the concept of order types. We divide ordered
sets into classes, assigning two ordered sets to the same class if and only
if they are similar. Since the relation of similarity between ordered sets
is reflexive, symmetrical and transitive, ordered sets are divided into
disjoint classes. The symbols denoting those classes (or, if preferred,
the classes themselves) are called order types. Every ordered set A thus
determines a certain order type, namely the type determined by the
class of all ordered sets similar to the set A. The order type of the set A
is denoted, after Cantor, by A. G. Cantor regarded the order type of a set
as that property of the set which remains when we disregard the quality
of the elements of that set but not their order, which is marked by one
horizontal stroke over the symbol of the set; now, if we disregard both
the quality and the order of the elements of a set, we come, according
to Cantor, to the concept of the power of a set, which is denoted by two
horizontal strokes over the symbol of the set.
Thus, for two ordered sets A and B, we have A = B if and only if
those sets are similar. In that case we also have A = B but not vice versa
because equivalent ordered sets are not necessarily similar (e. g. the set
of all natural numbers and the set of all integers ordered according to
their magnitude). ee aie
However, if the ordered sets A and B are finite and if A = B, then,
according to Theorem 1 of XI.6, we have A= 8B. Thus if an ordered
set is finite, then in order to know its order type it suffices to know the
number of its elements. Therefore the order type of finite ordered sets
of n elements may be denoted by n. In this way natural numbers (and
number 0) appear here in a different role from that of cardinal numbers,
namely as order types.
The simplest type corresponding to infinite sets is the type to which
belongs the set of all natural numbers ordered according to magnitude;
we denote it by w (cf. XI.7).
 § 1. Order types 223

The set of all negative integers ordered according to their relative

magnitudes
? vy] ,

determines another type, ordered inversely with respect to type w: we

denote it by the symbol a%*.
In general, if a denotes a given order type, then the type ordered
inversely (see XI.1) will be denoted by a*. It may occur. that a= o*;
this is obviously the case for every finite type and for types 7 and A,
defined in XI.8 and XI.10.
We have of course (a*)* = a for every order type a.
EXERCISE. Prove with the aid of the axiom of choice that every infinite ordered
set contains either a subset of type w or a subset of type w*.
Proof. Suppose that each non-empty subset 7 of an infinite ordered set A (o)
has a first element, f(7). Let us define by induction an infinite sequence a,, a2, ... of
elements of the set A as follows: a, = f(A), and we take a, = f(A— {a, ae, ..-, G_})
forn > 1. (The set A—{a,, ao, ..., d_,}i8 non-empty since we have assumed the set A
to be infinite). It can easily be proved that a,_,ea, for n = 2,3,..., whence the in-
finite sequence a,, a2, ... will be a subset of type w of the set A(o).
Now suppose that a certain non-empty subset 7 of the set A(@) has no first element.
Thus for each element a of the set 7 there exists at least one element a’ « T such that
a’oa. From the axiom of choice it follows that there exists a function » defined in the
set 7’ and such that g(a) « T and g(a) ea for ace T. Let a, denote any element of the
(non-empty) set 7’. The set of the elements a, 9(4o), PP (a), PPY (Ay)... Will obviously be
a subset of type w* of the set A(g).
Thus the required proof has been given.

2. Sum of two order types. Let a and f be two given order types.
Thus there exist ordered sets A(o,) and B(o,) such that 4 =a, B= fp.
We can assume here that the sets A and B are disjoint. Otherwise it would
suffice to replace the set A by the set A, of all ordered pairs (a, 1) where
ae A, ordering the set A, by means of the relation oj where (a,, 1) o1(a2, 1)
holds in the set A, if and only if we have a,o,a, in the set A. Similarly
we should replace the set B by the set B, of all pairs (b, 2) where be B,
ordering the set 6, by means of the relation 03 where (b,, 2) 03(b,, 2) holds
in the set B, if and only if we have b, 0,6, in the set B. We shall have of
course A,(e1)~A(a), Bye) ~ B(@), whence A,=o, B,= 6 and
A,B, =.0.
Therefore let us suppose that A= A(o,), B= B(o,), d=a, B= B,
AB = 0. Let S = A+B and let us order the set S by the relation 0 defined
by the conditions: if s, « S, s,« 8, then the formula s, es, holds if and only
lime hesAr as, erAront S109, )00us; € Bs 28,icB “and <3, 08,4 oresre Ay 1s.) © B:
It is easy to prove that the relation @ defined in this way will order the
set S. In other words, we order the sum S of the sets A and B in such
a manner that for each two elements of that sum both belonging to A or
224 XII. Order types and operations on them

both belonging to B we retain the ordering in which they appeared in

those sets, and of two elements of the set S belonging one to A and the
other to B we regard the one belonging to A as preceding the other.
Now let A’ = A(oj) and B’ = B(e@z) be any two ordered sets such
that A’—a, B’ =p, A’B’=0. Let 8’ = A’+B’ and let us order the
set S’ by the relation 0’, such that for s, « S’ and s,« 8S’ we have s,0's if
and. Ollyais.s, 604 S,:c-Al, Sip $3; OF ae De oa etls. aS 075s Ol ala ys
s,« A’, s,¢B’. It is easy to prove that S(o0) ~ S8’(o’).
Indeed, since A = a and A’=—a, we have A(o,) ~ A’(ei) and there
exists a mapping f, of the set A on the set A’ establishing the similarity
of those sets. Similarly, in view of B= 6 and B’= fB we have B(o,) ~
~ B(os) and there exists a function f, establishing the similarity of the
sets B and B’. In order to obtain the function f establishing the similarity
of the sets S(o) and S(o’) it obviously suffices to put f(s) = f,(s) for se A
and j/(s) = /j,(s) for s¢ B.
Now let S = o; the order type o depends thus. merely on the order
types a and / and not on the choice of ordered sets A and B such that
A=a, B=, AB= 0. We call it the sum of the order types a and fp and
write
Or a8...

EXAMPLES. Denoting by A the set of all negative integers ordered according

to their relative magnitudes and by B the set of all natural numbers ordered according
to magnitude, we shall have (see § 1): A = w*, B=o. The set p= A+B, ordered as
in the definition of the sum of two types, will thus be the set of all integers ordered
according to their relative magnitudes. The type of that set will thus be w*+ w.
If we now denoted by A’ the set of all numbers of the form 1/n, where n is a natural
number, ordered according to their magnitude, and by B’ the set of all numbers of the
form —1/n, where n is a natural number, ordered according to their relative magnitudes,
then we should have A’= w*, B’ = w. Writing gy’= B’+ A’ and ordering the set ¢,
according to the relative magnitudes of the numbers which compose it, we shall obtain
a set of type w+ m*, formed of the numbers

—-t< -4<-4<..<57<4<1.

Types m*-+-@ and w+ * are different from one another: the former has neither
a first nor a last element, while the latter has both. The former has no gaps, while the
latter has one.
This example proves that a swum of] order types may depend on the order of the
components. Thus the components of a sum of two types play a different part each:
therefore they have been given different names: the first is called angendus and the
second addendus.
Let us take another example: the sums 1-+m and w+ 1. According to the defini-
tion of the sum of types, in order to obtain a set of type 1+ w we must take a set con-
taining one element only, e. g. a), and a set of type @, e. g. the infinite sequence a,, do, ...,
ordered according to the successive indices; having formed the sum of the two sets,
regard the element a) as preceding each term of the sequence a, (k = 1, 2, ...), in which
 § 2. Sum of order types 22 Ol

we retain the former ordering. In this way we shall obtain the ordered set

git,
<A ses 5
which is obviously also of type wo.
Now, for the type »+1 we obtain the ordered set

pA ean ne
whose type is different from m, since it has a last element while type wm has not. Thus
we have 1+o0 =o, o+1+ wo and therefore 1+o4~o+1.
Similarly we easily obtain 1+ 0* #4 o*+1= o%.
In certain cases, however, sum of two order types may be independent of the
order of the components, e. g. when both types are finite or when the components are
equal. Here is another example. Let £ = w+ *: obviously 1+&=é&+1 (here both
sums are equal to &).

The concept of the sum of order types can immediately be generalized

by induction to the sum of an arbitrary finite number of types and it
is easy to prove that this sum always has the property of associativity. EF. q.

(o+l1l)+o=—o+(1l+o)=o+0.

From the definition of the sum of ordered types it follows that if A

is an ordered set and [A,, A,] a cut of the set A, then A= A,+4A,.
Thus, for instance, if A denotes the set of all non-negative. rational num-
bers ordered according to magnitude, and if we define in that} set a cut
[A,,A,], taking A,= {0}, A,= A-—A,, then we shall of course
have A, = 1, A, = 7, whence A = 1+. Thus the set of all non-negative
rational numbers ordered according to their magnitude is of type 1+7.
If we now form a cut of all rational numbers FR ordered according
to magnitude, |f,, R,|, where R, is the set of all negative rational numbers
and R, = R—R,, then we shall have Rk, = 7, R,—1+7; since R=,
we have the formula 7 = 74+(1+
7). Thus

yt+1l+n=y7 .

We prove in an analogous way that

At+14+A=i.

If we formed a cut of the set R,[Ri, Ry], where Rj denotes the set
of all rational numbers < 2 and R3j—=— R—R{, we should have Rj=
= Ri = ny, whence, in view of R= hkj+ Rj, we should obtain the formula

U fs bred Bega Be
But :
2 -— A SH i, 9

since the type 4+ has a gap.

Cardinal and ordinal numbers 15
226 , XII. Order types and operations on them

EXERCISES. 1. Prove that for any order types a and 6 we have

(2.1) (a+ B)* = p*+a*.

2. Prove that if € denotes the order type of the set of all irrational numbers
ordered according to magnitude, then

f+f=& and E+14f=6.

3. Prove that, for an order type a, we have a = a* if and only if there exists an
order type & such that a = &+ &* or such that a = €+1-+ &*}).
Proof. Let A(o@) denote an ordered set of type a. If a= a*, then there exists
a mapping f of the set A on itself, such that, for xe A, ye A, the formula xey is equi-
valent to the formula f(y) ef(x). Let

Ai = (wef(@)1, A, =é.
We shall further distinguish two cases:
1) a Af(a) for ve A. Let A, = A—A,. If w¢ A,, then, according to the defini-
tion of the set A,, xef(x) does not hold, and since, by 1), x ~€ f(a), in view of the con-
nexity of the relation @ in the set A, we must have f(x) ox. Hence we easily conclude
that
A,= Ef (@) ee].
Now if a, <« A,, a,¢ A,, then we have a, ~ az, a ef (a1), f(a) 0a,. If we had a, ea,,
we should have hence f (@2) ea@2 0a, of (a,), and thus f(d,) ef (a), which, as we know, gives
4, 0d; this, in view of a,ea,, is impossible. Thus if a, « A,, a, ¢ Az, then a, ~ a, and
4, 0a, does not hold; therefore, in view of the connexity of the relation o@ in the set A,
we must have a,ea,. Thus, from the definition of the sum of order types it follows that
A = A,+A,. Now we shall prove that the function f maps the set A, on the set A,.
Let t¢ A,; we thus have tof(t) =u, whence f(u) of (t), i. e. f(f(t)) ef(t), which proves
that f(t) « A,. On the other hand, if we A,, then f(w)ew and, writing ¢ = f-1(v), we ob-
tain f(t)= wu, whence f(u)of(t), which gives tow, t.¢. tof(t), and thus te Aj, i. e.
f-(u) « A,. The function f maps the set A, on the set A,, and, since it inverts the order
of elements, it follows that A, = A* = &*, Thus a= A= 4,44, = €+&.
2) There exists an element a « A such that a = f(a). To begin with, we shall prove
that there may be only one such element. Indeed, if we had also b = f(b) for b 4a,
then we should have either agb or bea; suppose that aeb, whence f(b) oef(a), 7%. e. boa.
which gives a contradiction. Similarly we find that A = A,+ {a}+A, where

A, a [wof(x)], 4e= Ff [f(a)ex).

Now if a, <¢A,, then a, ef(a,) and, in view of a= f(a), we must have a, 4a. If
we had aea,, we should have f(a,)ef(a) = a, whence, in view of a, ef(a,), a, ea, which
is impossible. In view of a, 4 a and the connexity of the relation @ in the set A, we
thus have a, ea. Therefore, if a, « A,, then a,ea. We prove similarly that if a, « A,, then
aea,. In view of A = A,+ {a}4+-A, and the definition of the sum of order types, we

1) This theorem was given without proof by A. Lindenbaum (see Lindenbaum

et Tarski [1], p. 321, Théoréme 8).
 § 2. Sum of order types 227

have A = A,+1+4,. But, as in case 1), we obtain f(A,) = A,, whence A, = A*, and
if we take A, = &, we shall have a = + 1+ &*.
Thus we have proved that if a = a*, then there exists an order type € such that
a= &+ &* or such that a = €+ 1+ &* (it is possible for those cases to hold simultane-
‘ously, ¢. g. 7 = 7* =y+7* =7+1+7*). On the other hand, it is easy to verify (by
formula (2.1) and a** =a) that, for every order type &, we have (€+ é*)* = ¢+ &*
and (€+ 14+ é*)*= €+ 1+ &*. The required proof is thus completed.

Remark. If f is a function which maps an ordered set H upon itself, inverting

the order of its elements, then an element # « H such that f(f(x)) = # does not neces-
sarily exist. H.g. if # is the set of all Gey ee numbers of the interval (0, 1),
ordered according to their magnitude, and if f(#) = 1—2?, then the function f maps
the set # on itself, inverting the order of its fuhates but there is no element « of the
set H for which f(f(x))=, since then we should obviously have 2a?— «1 = a, an equa-
tion that is satisfied by no transcendental number z.
4. Prove that, for an order type a we have 1+a=a if and only if a=o+€
where € is an arbitrary order type (which can be equal to 0).
Proof. Let A(@) denote an ordered set of type a. If a= 1-+a, then, according
to the definition of the sum of order types, there exists a decomposition A = A,)+ A,
such that A, is an ordered subset of the set A that is of type 1, 7. e. the set A, contains
one element only, A) = {a}, and A, is a subset of the set A that is of type a, whence
Aw A,; thus there exists a function f establishing a similar mapping of the set A on
the set A,, and moreover the only element of the set A, precedes any element of the
set A,, whence it follows that a, is the first element of the set A. Since f(a.) e A, =
= A—A,, we have f(a.) ~ ao, and thus a, ef (a). Since the function f establishes a simi-
lar mapping of the set A(@) on its subset A,, if ce A, ye A and woy, then f(x) ef (y)
Since ay ef (ao), we have f(a)
ef(f(ao)) and, in general, writing f°(a)) = a), and f'*? (a) =
=f(f'(a)) for n=1,2,..., we obtain f"(a)of"**(a,) for n=0,1,2,... Thus
Ao Of (Ao) Of ?(Ao) Of8(ao)
@... Let B= {ao, f(ao), f?(Go), .--}3 16 will be a subset of type w of
the set A. Taking C = A—B, we obtain A = B+C, without excluding the possibility
of the set C being empty.
It can easily be seen that f(a,) is the first element of the set A,, 2. e. in the set A
there is no such element « that a) exof(a.). Next we prove by induction that for a na-
tural there is no such an element y in the set A that f"(ao) eyof’* (a>) because then
we should have y ~ ay, and since the function f~! establishes the a of the sets A,
and A, we should have also f~*(f"(ao)) of “(y) of (f"** (ao) 4. & f° (ao) of (y) ef" (ao)-
Now suppose that ce O. We shall prove that f"(a)) ec for n = ‘.Ws Pe eeeeee
since ce 0 = A—B, we have ¢c $ f*(a,) for n = 0, 1, 2, ... Therefore, if we did not have
f"(a) ec for n = 0, 1, 2, ..., then, in view of the connexity of the relation g in the set A,
there would exist such natural k& that cef*(a)). Assuming k to denote the least of such
natural numbers, we should have f*~*(a,) ecef*(ao), which, as we have proved, is im-
possible.
Thus we have proved that A = B+0, each element of the set B preceding each
element of the set O. Let C = & (we can have £ = 0 if the set O is empty). Since A = a,
B = , according to the definition of the sum of order types, we shall have a = w+ &.
On the other hand, in view of the formula 1+ » = w and of the Sees of
the addition of order types, we have for every order type €: 1+ (w+ &) = (l+)+&=
=o-+ &.
Thus the required proof is completed.
fomOr
228 XII. Order types and operations on them

5. Prove that, for an order type a, we have a+ 1=a if and only if a= &+o%,
where € is an arbitrary order type (which may be equal to 0).
Proof. Suppose that a is an order type such that a+ 1= a. According to Exer-
cise 1, we have 1+ a* = (a+ 1)* = a*, and, according to Exercise 4, there exists an
order type ¢, such that a* m+, whence a = (a*)* = (w+C)* =€*+ o*. Taking
* — & we shall have a = + w*: thus the condition is necessary.
On the other hand, in view of w*+ 1 = w* and the associativity of the sum of
types, we have for every order type &: (€+ m*)+1= &+(w*+1) = + o*; thus the
condition is sufficient.
6. Prove that, for an order type a, which is not finite, we have 1+ a = a-+ l, if
and only if there exists an order type € (which may be equal to 0) such that a = w+ &+
+ m* (and then we have also 1+a=a).
7. Determine the order type of the set of all infinite sequences composed of num-
bers 0 and 1 and having only a finite >0 number of terms <0, a) ordered according
to the principle of first differences‘), and b) ordered according to the principle of last
differences.
Answer. In case a) the required type is 1+ 7, in case b) it is w. In case a) it suffices
to prove that the set has a first element, has no last element and is denumerable and
dense (and to refer to Theorem 1 of XI.8). In case b) a mapping of the set in question
on the set of all natural numbers is obtained by associating with each sequence 4@,, a2,
a,,... @ natural number 1+a,+ 2a,-+ 2?a,+ ...
8. In Exercise 7 replace the words ,numbers 0 and 1“ by ,,integers‘. What is
the answer?
Answer. In case a) the type is 7, in case b) it is 7+ w*+o+ 7.
9. Prove that there exist only four different dense denumerable order types, na-
mely the types 7, 1+7,7+1 and 1+7+1. 3
Proof. Let a denote a dense denumerable order type, A(g) a set of type a. Thus
the set A satisfies conditions 2) and 3) of. Theorem 1 of XI.8. If, moreover, the set A
satisfies condition 1) of that theorem (i. e. has neither a first nor a last element), then,
in virtue of that theorem, it is of type 7. If the set A does not satisfy condition 1), then
of course only the following three cases are possible:
1’) The set A has a first element — a, but has no last element.
1’) The set A has a last element — 6b, but has no first element.
1’’) The set A has a first element — a, and a last element — b.
From properties 2) and 3) of the set A it easily follows that in case 1’) the set
A, = A— {a}, in case 1’) the set A, = A — {b}, and in case 1’) the set A; = A—{a, b}
will satisfy conditions 1), 2) and 3), 7. e. will be of type 7. In these three cases we shall
have, respectively, A = {a}+A,, A = A,+ {b}, A = {a}+A;+4 {6}, and since a is the
first element of the set A in cases 1’) and 1’), and b is the last element of the set A in
cases 1’’) and 1’’’), we shall find that the types of the set A in cases 1’), 1’’) and 1’”)
are 1+7,7+1,1+7+1 respectively. The proof that these types are denumerable,
dense, different from one another and different from 7 presents no difficulties.
10. Prove that if a, 6, a, and £, are order types such that a+ 6=a,+/,, then
there exists an order type y (which can be equal to 0), such that either a, =a+y, B=
=yt+f,, or a=a+y, Bp=y+8.

1) 4. e. so that (a,, dg, ...) = (b,, bg, ...) if and only if, denoting by k the least na-
tural number for which a, 4 b,, we have a, < 6,.
 § 2. Sum of order types 229

Proof. Let # denote an ordered set of type a+ 8. From the definition of the sum
of order types it follows that, in view of a+ $f=a,+ ,, there exist ordered sets A,
B, A, and B, such that A=a, B=, A, =a, B,=f,, F = A+ B= A,+B,, [A, B]
and [A,, B,] being cuts of the set F. From Exercise 2 in XI.5 it follows that either A c A,
and B,c B, or A,c A and Bc B,. In the first case the set A will obviously be a segment
of the set A, and the set B, — the remainder of the set B, and we shall have A, = A+
-+(A,—A), B= (A,—A)+B,, in the second case the set A, will be a segment of the
set A and the set B — the remainder of the set B,, and we shall have 4d= A,+(A—A,),
B, = (A—A,)+8. Taking y = A,—A in the first case and y = A—A,, in the second
case, we shall have either a, =a+ty, B=y+,, ora=a+t+y, fj} =y+f, q.e.d.
11. Give an example of order types a, B,y, aq, f,,y, such that af~Aa,, P-Af,,
ea+p=aq-p, a4=aty, pP=y+tBp, ¢=aty.- pi = i+B-
AMEWOR, C= 7, (i = IS Q, = i = We Uy hi = 7 DA = He
Remark. It can be proved that there are no such order types a, a, and y that
AAG, h=a+y, a=aty.

3. Product of two order types. Let a and / be two given order types;
there exist ordered sets A(o,) and B(o,) of types a and f respectively. Let
P= AXB and let us order the set P according to the principle of last
differences, 7. ¢., for (a, b)e« P, (c,d)eP, let (a, b)oe(e, d) if and only if
bod, or b= d and aoe. It is easy to verify that the relation o orders the
set P.
If we take, instead of the set A(o,), an ordered set A’(o;) also of
type a, and, instead of the set B(o,) — the set B’(o2) of type f, and if
we write P’ = A’xB’, ordering the set P’ by the relation 0’ according
to the principle of last differences, we have, as can easily be proved,
P(o)~ P(o’). It follows that the order type of the set P(e) does
not depend on the choice of the ordered sets A(o,) and B(@,), provided
the former is of type a and the latter of type 6. Thus the order type y
of the set P(e) depends only on the order types a and f (and on their
order). We call it the product of the order types a and #6 and write y =
ap.
From the definition of the product of order types it follows im-
mediately that a-l1=1-a=a for every order type a.

EXAMPLES. 1. Let us calculate the product 2-m. As the set 4(0,) we take the set
{1, 2}, and as the set B(o,) — the set of all natural numbers ordered according to their
magnitude. The set P will be the set of all ordered pairs (a, b) where a = 1 or 2, and b
is any natural number; ordering the set P by the relation @ according to the principle
of last differences, we shall obviously obtain

(1, 1l)e(2, Ie(1, 2)e(2, 2)e(1, 3)e(2, 3)eU, 4a...

whence it follows that the set P(g) is of type w. Thus we have
2 Oi Oe

Similarly we have n-w =o for every natural number n.

 230 XII. Order types and operations on them

2. Let us now calculate the product wm-2. As the set A we can take the set of all
natural numbers, as the set B — the set {1, 2}, both ordered according to the magnitude
of the numbers composing them. Ordering the set P = A xB by the relation @ accord-
ing to the principle of last differences, we obviously obtain:

(i, 1) @(2, 1) e(3, 1)@ coe e(l, 2) @(2, 2) (3, 2)0 OO ty

whence it follows that the set P(g) is of type a+. Thus we have
o-2=o10,

and therefore w: 2 4 2-w since the set P(g) is not of type w, its proper segment formed
by the element (1, 2) not being finite, in contradiction to Theorem 1 of XI.7 on the
type w. Therefore
The product of order types does not possess the property of commutativity.
The concept of the product of order types is generalized by induc-
tion to the product of an arbitrary finite number of types, and, as can
easily be seen, that product always has the property of associativity. E. gq.
(m-2)-oa=@o-(2-0)=o-o.
As regards the law of distributivity of the multiplication of order types
with respect to their addition, it holds only in one of its forms, namely
when the second factor (multiplier) is a sum. It can easily be proved .
that, for all order types a,f and y, we have

(3.1) a(B+y)= aB+ay,

but
GQ ho=2-o=ao -o-1--o-1.

In order to prove formula (3.1) we must take the ordered sets A, B

and OC of types a, 6 and y respectively, where BC = 0, and then use
the equality Ax(B+C)= AxCO+BxO and the method of obtaining
the sum and the product of order types.
By induction, we easily generalize formula (3.1) to an arbitrary
finite number of types — components of the second factor of the left-
hand side, and obtain the formula

(8, + Bot...
+ Bn) = af, + a8.+...+
a8, ,

for every finite sequence of types a, f,, Bo, ...) Bn.

In particular, for 6, = §, = ...= B, = 1, we obtain hence the formula
See Z
(3.2) an=a-+a-t...ta

for every order type a and every natural number n.

In view of the formula 7+ 7 = 7 (see § 2), formula (3.2) gives 7-2=
=n+n=yn. But 2-7A~yn, since the type 2-7 obviously has jumps.
Indeed, in order to obtain type 2-7, we can take the set A eae ale 2}
and the set B of all rational numbers, ordering each of those sets accord-
 § 3. Product of two order types 231

ing to the magnitude of the numbers composing them; then we order

the set A xB according to the principle of last differences. If r « B, then,
with this ordering of the set A x B, its element (2,7) sueceeds the element
{1,7); thus the set Ax B has jumps and cannot be of type 7.
Therefore 2-7 ~A7- 2.
We similarly prove that 2-4 -~4/-2.
Let us now calculate the product 7-7. It will be observed that if
each of two order types, a and f, satisfies any of the conditions 1), 2), 3)
of Theorem 1 of XI.8, then their product a-f also satisfies that condi-
tion. The type 7-7 therefore satisfies conditions 1), 2), 3) of that theorem
and thus is of type 7.
Therefore we have the formula

(3.3) Fe ee
For a natural n, the product of » order types each of which is of
type a is denoted by a”. From formula (3.3) it immediately follows by
IMOUCHION Lhab qe —= 7) LOT MWe 1,

EXERCISES. 1. Prove that for any order types a and fS we have the formula

(aB)* = a*B*.
2. Prove that if an order type a is dense and has no first or last element, then
for every order type 6 the type af is dense, but the type fa is not necessarily dense.
3. Prove that if a set of type a-f has a first element, then both a set of type a
and a set of type f have a first element. Prove that the word ,,first“ can be replaced
here (twice) by the word ,,last‘.
4. Prove that aw* ~< w*o.
Proof. We can prove the above by showing that the type w*m has segments of
type w*, while the type ww* has no such segments.
5. Prove that if a is any denumerable type, then ya = 7.
Proof. Let a denote a given denumerable order type. From |Exercise 3 and from
the observation that the type 7 has no first element it follows that type na has no first
element. We similarly prove that type ya has no last element. From Exercise 2 and
from the density of type 7 it follows that type ya is dense. Finally, from the denumer-
ability of types 7 and a it follows that type ja is denumerable. Thus type ya satisfies
the conditions of Theorem 1 of XI.8 and therefore is equal to 7, q.e. d.
6. Prove that if a is any denumerable type having no last element, then
(nt+lha=y.
Proof. Since the order type 7+ 1 is dense it follows from Exercise 2 that type
(n+ 1)a is dense. Since type 7+ 1 has no first element, and type a, according to the
assumption, has no last element, it follows from Exercise 3 that type (7+ 1)a has neither
@ first nor a last element. Finally, from the denumerability of types 7+ 1 and a follows
the denumerability of type (n+ 1)a. Thus type (7+ 1)a satisfies the conditions of
Theorem 1 of XI.8, and therefore it is equal to 7, q. e. d.
7. Prove that if ais any denumerable type having no first element, then (1-+ 7)a = 7.
232 XII. Order types and operations in them

8. Prove that wn-+o 4 wn.

Proof. The proof follows from the observation that the type wy+o has a re-
mainder w, while the type w7 has no such remainder.
Remark. As regards type wn, it is easy to prove that it is the type of the set of
all natural numbers ordered by the relation g, defined as follows: If m = 2?~*(2q—1),
n = 2'~*(2s—1) where p,q, 7 and s are natural numbers, then men if and only if p/q <
< r/s, of if p/q=7/s and m < n.
9. Prove that for a= on, B = w(y+1) we have a= f? but a+ B.
Proof. According to Exercise 5, we have nw = 7, whence a? = aa = (wy) (wn) =
= 0(nw)n = wnn = wn, Since yn=7 by (3.3). According to Exercise 7 we have
(7+1)@ =, whence B? = w(n+1)o(n+1) = wn (y+1) = on, since, according to Exer-
cise 5, 7(n+1) = 7. Thus a? = f?. But a = wy and, in view of the distributivity law,
B = o(n4+ 1) = on +o-1= n+; in virtue of Exercise 8, we thus have a + B.
Therefore: the equality of the squares of order types does not imply the equality of
those types.
The first example of two different denumerable order types whose squares are
equal has been given by Davis [1] (p. 382) and Davis and Sierpinski [1] (p. 850). We do
not know so far any example of two types m and y, such that g? = y? but gy? ¥ y’, or
types y and 6 such that y? 4 6? but y? = 6%. Neither do we know any type a such that
GP sim Oe == Oh

10. Give an example of three different order types whose squares are equal.
Answer. Such are, for instance, the following types: (w*+@)7, (w*+ @)(y+1)
and (w*+ w) (1+).
Remark. Anne C. Davis has given an example of infinitely many different order
types whose squares are equal. They are the types & = (@*+ )(w+on+k) where
k= 0,1, 2,... (see XII.4, Exercise 4). She has also given an example of a denumer-
able order type a for which the number of different order types & satisfying the equa-
tion £2 =a, and, more generally, the equation é” = a, where n = 2, 3,..., is equal to
the continuum. She has likewise proved that for each cardinal number m = 0, 1, 2,
suey No, 2° there exist 2“° denumerable types a such that for every natural number n > 2
the equation &" = a has exactly m different solutions (in order types &). In particular,
for given natural m and n, there exist exactly m-+ 2 order types satisfying the equa-
tion &” = (w*+ @)myn: they are &, = (w*+@)myn and Ga ok = (w*+ ow) (k+ mn)+ (m—k)
LODE One, Serceaite

1l. Give an example of three order types a, 6 and y such that for every natural »
we have a Ap’ #y" 4a’ but
gy ae p" = y" ; .

Solution. We can take here a=o, B=w-2, y=a@-3. For, in view of kw =

= w for natural k, we have (wk)? = wkwk = w(kw)k = wk, and if for a certain natural n
we have (wk)* =o'k, then (wk)"** = (wk)wk = o kok = o"(kw)k = wok = wk,
whence, by induction, we conclude that (wk)” = w"k for natural k and n. Thus, in view
of the distributivity law,

a’ + p” = w+ (0-2)" =o" +o" -2=a' (14 2) =o"-3 =(@-3)" =’.

The proof that the types w, w-2 and w-3 are different from one another presents
no difficulties.
Therefore: the great theorem of Fermat is false for order types.
 , §3. Product of two order types 233

12. Prove that (1+ 7) (n+ 1) = (n+1)(1+n) = 7, and that for natural n we have
(1-+.n)" = 1+ and (y+1)"=n+1.
Proof. In view of the distributivity law we have (1+ )(y+1) = (l+7)n+1+7.
But, in virtue of Exercise 7, (1+ 7)7 = 7, and since, as we know, 7+ 1+ = , we have
(1+ 7)(n+1)= 7. We similarly obtain (7+ 1)(1+7) = (7+ 1)+ (n+ ))n=9+14+7=
= yn. Further, we have (1+ 7)(1+ 7) = (l+7)+ (l+7)yn = 1+7+7=1+7, whence
by induction (1+ 7)"=1-+y for natural n. Finally (y+ 1)(y4+1) = (n4+1)n+n4+1,
and since, in virtue of Exercise 6, (7+ 1)7 =%, it follows that (y+ 1)? =7+y7+1=
= +1, whence by induction (y+1)° =y+1 for n=1, 2,...
13. Give an example of infinitely many different order types satisfying the equa-
tion €-2=—&.
Answer. Types 1+ 7+ (n—1), where » is a natural number, are obviously all
different from one another and we have [1+ 7+ (n—1)]-2=1+77+ (n—-1)+1+
+ n+ (n—1) = 14 nj + n+ m+ (n—1) = 14+ n(n+ 14+ )+ (n—-1) = 14+ 27+ (n—]).
14. Give an example of an order type a such that a-n=a for n= 1, 2,..., but
aw ~ a.
Answer. a=7-+1 since, if for a natural n we have (n+1)n=7+1, then
(n+ 1) (m+ 1) = (n+ 1) nt (9+: 1) = (n+ 14+ (9+ 1) = (H4+-14+9)+1= 741, whence, by
induction, (y+ 1)n = 7+ 1 for n= 1, 2,..., but, in virtue of Exercise 6, (7+1l)a=7
and n+1 7.
15. Prove that the types A-n(n= 1, 2,...) are all different from one another.
Proof. The proof follows from the observation that the type A-n has exactly » gaps.
16. Prove that A? ~ 2.
Proof. The proof may proceed as follows. We have 4+ 1+.A4= A, whence #2? =
=A(A+1+4+A) = P+A+ 72? = + (A+ 27). Hence it follows that type #2 is the sum of
two order types /? and A+ 22 of which the first has no last element and the second has
no first element. Type 2? thus has a gap, whence /? 4 /. -
17. Prove that types nA (n= 1, 2,...) are all different from one another.
Proof. The proof proceeds as follows. If n and k are natural numbers such that
n > k, then a set of type nd contains elements a and b #0, between which there lie
nm—2>k—1 different elements of that set, and a set of type kA contains no such
elements.
18. Give an example of an order type m for which there exist order types a and fp
such that mp = 2a+ 1 = 26 (i. e. of type g which is both odd and even) and of an order
type y for which there exists neither a type a such that y= 2a+ 1 nor a type f such
that y = 26 (i. e. of type y which is neither odd nor even).
Answer. We have w* = 2w* = 2w*+ 1, and type 7 is neither of the form 2é
nor of the form 2+ 1 since both forms give types which are not dense.
19. An infinite sequence of order types a, (n= 1, 2,...) for which there exists
an order type g such that a,,,=a,+ 0 forn = 1, 2,..., is called, an infinite arithmetical
progression. Give an example. of an infinite arthmetical progression composed of dif-
ferent terms each of which is the square of an order type.
Answer. For example a, = (w-n)? for n= 1, 2,..., since we have here O41 =
= [a(n+ 1) P= w(n+4+ 1) = w'n4+ w? = (w-n)?+ ow = a,+ w? for n = 1, 2,...
Remark. According to a certain theorem of Fermat (of 1640) there are no four
different finite order types forming an arithmetical progression and being each
a square. We can have three such types, e.g. 1, 5°, 7°. ,
234 XII. Order types and operations on them

20. Give an example of three order types a, f and y such that each of the six per-
mutations of those types gives a different sum.
Answer. Por example a—1, B=o, y= 7.
21. For every natural number k = 1, 2,3,4,5 give an example of three order
types a,, B,. 7, Such that in any permutations of those types we obtain exactly hk dif-
ferent sums.
ADS WOR: C=O, pr 0-2, Vy == fe) Opel ps On 7, ny Ae ge
=o ily Wy — Os Be Gn = Thy fin Oy SHS OP Ch = @, a= Wel vee Oe We Wane
here, for instance, a,+/;+y,; = @+(@+1)+o? = o+0+4+ (1+?) = 040+? =
=o(1t+1+o)=o0-o=o, B+a+y; = (@+1)to+0? = 0+0+0?= ow, Bs+yst
+a,=(o+1)+0?+o= 079+ 0, a4,+7,+f; = 0+ w+ (o+ 1)=0?+0+1, 75+ 6s+ a=
=w?+(o+1)+o=o?+o-2, ystast+f; = o+ot+(o+1)—e@?+q@-2+1, and the
types w?, w+ 0, w+ o+1, w+ 0-2, o?+o-2+1 are all different from one another
(see Sierpinski [48], p. 252).
22. For every natural number n give an example of m order types a,, dy, ..., a,
such that each of n! permutations of those types gives a different sum.
PACING WiCT Cee /Go17) a OTGe=— sl ge ae ae
23. Find all order types a 40 such that a4+y=y+a.
Answer. They are all order types of the form 7+ &+ 7 where € is an arbitrary
order type. Indeed, if a is an order type 40 such that a+ 7 = 7+ a, then ais a segment,
different from 0, of type 7+ a, and since every non-empty segment of type 7-+a is
of the form 7+ é, we have a=7+é, whence, since 7+ =, we obtain a= 7+ €&=
=y+yn+E=yn+a=a+n=7+E+7. Thus a=7+&+7. On the other hand, if, for
an arbitrary type €, we take a = 7+é+7, then obviously a+7=y+a=a.
24. Find all order types a #0 such that a+/A=A-+a.
Auswer. These are all order types of the form A-n, where n is a natural number,
and all the types of the form wj+ + w*/, where & is an arbitrary order type (which
can. be equal to 0).
25. Prove the theorem of B. Dushnik and E. W. Miller stating that every denu-
merable ordered set is similar to a proper subset of itself.
Proof. Let A(g) denote a denumerable ordered set. Let us decompose the set
A(e) into components, assigning two elements of the set A(g) to the same component
if and only if only a finite number (+0) of the elements of the set A(@) lies between
them. Obviously the set A(@) decomposes into disjomt components, each component
being either a finite set or a set of one of the following types: m, w* or w*+. If A,
and A, are two different components of the set 4 (g), then every element of one component
precedes every element of the other; thus the components of the set A(g) form an or-
dered -set.
If the set A(g) contains an infinite component Y, then obviously the ordered set
Q(oe) (as a set of type w, m* or m*+) is similar to a certain proper subset of itself;
denoting by P the set of all elements of the set A(e) which precede all elements of the
set Y (or, respectively, by & the set of all elements of the set A(o) which follow all
elements of the set Q), we shall have A = P+Q+8. Mapping the sets P and R identic-
ally each on itself, and the set Q similarly on its proper subset, we shall obviously ob-
tain a similar mapping of the set A(g@) upon its proper subset.
If the set A(g) contains no infinite component, then, obviously, the set A (@),
ordered by the relation @, where, for two components A, and A,, formula A,0,A, means
 § 3. Product of two order types 235

that a,ea, for a, « A,, a, « A,, will be a dense ordered set, since two finite components
cannot lie side by side (which immediately follows from the definition of components).
Taking from each (finite) component its first element, we shall obtain of course
a dense subset D of the set A(e@), denumerable (since the set A(g) is denumerable).
Thus, according to Exercise 9 in XII.2, the set A(g) is of type 7, 1+ 7, n+1 or 1+
+n-+ 1, and therefore contains a subset of type 7. Hence it follows that the set A (@)
contains a subset O of type 7. In virtue of Theorem 2 of XI.8, and in view of the equa-
lity 7 = n+ 7, the set A(g) is similar to a certain proper subset of the set CO, 7. e. also
to a certain proper subset of itself.
Thus the theorem of B. Dushnik and E. W. Miller is proved.
26. Give an example of a non-denumerable ordered set having a first element
and such that each element of it has a consequent and each element except the first
has an antecedent.
Answer. A set of type w+ (w*+ o)d.
27. Prove that an infinite ordered set has a first element, and is such that each
of its elements has a consequent and each element except the first has an antecedent
if and only if that set is of type w+ (w*+ w)& where & is an arbitrary order type (which
can be equal to 0).
28. Give an example of infinitely many different denumerable order types & for
which & = é.
Answer. Types €= ny where n= 1, 2, ...
Remark. It would be more difficult to give an example of 2°° different denumer-
able order types & for which &? = &, as well as an example of a non-denumerable type &
such that = €.
29. Prove that if y and # are arbitrary order types, and y = yw+0-+yo*, then
p+y = y+ (Aronszajn [1)).
30. For every type y, at most denumerable, give an example of a type y, of the
power of the continuum, such that p+y=y-+¢.
Solution. As follows from Exercise 29, such a type will be yw = yw+ A+ ga*.
31. Prove that [@(1+ 7)!’ = (1+ y) and [(@+1)(n+ 1] = (®+1)(n+1) for
Tf, ss MW Dan oe
32. Prove that (o+oy7+n) = o+on+n for n= 1, 2,...
33. Prove that the order types w+1, w+2 and w+3 give a different product
in each of their six permutations.
Proof. For natural n,, n., n3 we easily find: (w+ ,)(@+ N2)(@+ Ns) = w+ w'ng3+
+ on,+ n,, and w?+ wks-+ wk,+ ky= w3+ w?n3+ won,+ nif and only if k, = n, fori =1,2,3.
Hence it follows that with each change of the order of the factors w+1, w+ 2, w+ 3
we obtain a different product. ;
34. Find how many different values are given by the product of the order types w,
o+1, w+ 2 with all possible changes of the order of the factors.
Answer. There are five different values since w(w+1)(w+ 2) = w+ w?-2+ 0,
w(o+ 2)(@+ 1) = w?+ w?4+ w-2, (w+ 1)@(@+ 2) = w3+ w?- 2, (w+ 2)@(@+ 1) = w+ w?,
(+ 1)(o+2)o
= (0+ 2)(o+ 1)o
= a. |
35. For every natural number n give an example of n order types whose product
has a different value in each of their n! permutations.
Answer. Types a+ 1,@+2,...,@+N.
2356 XII. Order types and operations on them

36. The set of all infinite sequences whose terms are 0 or 1, the number of those
terms which are different from 0 being finite (> 0), is ordered according to the principle
of last differences. Prove that the order type of that set is m and determine a function f
establishing similarity between that set and the set of all natural numbers.
Answer. f(a, d, ...) = 1+ a,+ 2, a .+ 2?a3+...4 oo ae ... (Cf. XII.2. Exercise 7).

4. Sum of an infinite series of order types. Let a,, a, ... denote a given
infinite sequence of order types. From the axiom of Hilbert (VI.3) it
follows that there exists an infinite sequence of disjoint ordered sets
An(On) (n=1,2,...) such that A,(o,) is a set of type a,1). Let S=
= A,+A,+... and let us order the set S by the relation o defined as
follows. If the elements a and b of the set S belong to the same com-
ponent A,, then we shall write aob if and only if ao,b. And if ae Ax,
be A, where k ~#n, then we shall write aob if and only if k < n. It is easy
to verify that the set S will be ordered by the relation o. Let S = co. We
shall prove that the order type o depends only upon the infinite sequence
of order types a, a,..., and not upon the choice of ordered sets An( en)
disjoint and such that A,(0,) =a, for n=1,2,.
Let A,(o,) (1 = 1, 2,...) denote another infinite sequence of disjoint
ordered sets such that A,(o,) = a, for n= 1, 2,... Thus we have A,(0,)~
~ A,(o,) and it follows from the axiom of choice that there exists an in-
finite sequence of functions /7,,f,,..., such that, for n=1,2,..., the
function f, establishes a similar mapping of the set A,(o,) on the set
A,(o,). Let us define the function f in the set S as follows: if ae A, then
let f(a) = f,(a). Obviously, the function f establishes similarity between
the set S(o’) and the set S’ = Aj+A3+..., ordered by the relation 0’
defined by the conditions that a’ob’ if and only if we have either a’ « A,,
b’<« A, and a’o,b’ for a certain natural n or a’ e«Aj, b’e A, and k <n.
Thus 8’ = S =o.
We have proved that to every infinite sequence of order types a,
d,,... corresponds an order type o, well defined by that sequence, such
that if A,, A,,... is an infinite sequence of disjoint ordered sets where
A, = a forn=1,2,... and if we order the set S = A,+A,+... retaining

1) Here also the existence of such an infinite sequence of sets A,(0,) (vw = 1, 2, ...)
could be deduced from the axiom of choice (instead of the axiom of Hilbert), for we
shall prove in XVI.1 that from the axiom of choice follows the existence of a function f,
associating with each cardinal number m a set f(m) of power m. For n = 1, 2,..., let
A, =f(a,) and denote by ®, the family of all the order arrangements of the set A,
according to type a,. Since /(m) =m for cardinal m, we have A, = a,, and therefore
there exists at least one ordering of the set A, belonging to ®,. Thus it follows from the
axiom of choice that there exists an infinite sequence of order arrangements Q,, Q2, ..-;
such that 9, ¢«®, for n= 1, 2,..., 4. ¢. that A,(o,) =a, for n = 1, 2, ... The substitu-
tion of disjoint sets for the sets A,, A,,... presents no difficulty.
 § 4. Sum of an infinite series of order types 237

for the elements of the same component that order arrangement which
they had in that component, and — if the elements belong to different
components of the sum S — regarding that element as coming first which
belongs to a component that comes first, then we shall have S = o. The
order type o is called the swm of an infinite series of order types a,+a,+..
and we write
oO = Oy += Ag Og...

From the accepted definition of the sum of an infinite series of

order types it easily follows that

o=1414+1-3.,
but also
w= 242424..=142434...=
2427+ 23+...

It is easy to prove that for sums of infinite series of order types

the law of associativity holds; thus, for example,

Oy + Oy + Og ... = (4, + Oy) + (Gg + 4) + (Q5+ Gg) +...

= Oy + (Gy + Gg) + (Gq + As + Gg) + (G7 + Ag+ Ag+ Gyo) + ---
=(a, Az 4 4 Gn (Op Onto Tyas.) :

The law of distributivity of the multiplication of types with respect

to the sum of an infinite series also holds, but only in the form
a(a,+a,+a,+...) = aa,+aa,+aa,+...

(for any order types a, aj, ae, ...)

Thus we have for instance

ht+yntyt+..=y-ltyn-lty-14+...=n(14+1+...) = no
= 7;

whence n»+y+7+...=Y7
We also find in general that
am =a(14+1+...)=a-l4ta-1+...=atea+...;

therefore aw = a+a+a-++... for every order type a. Thus, in particular,

w=ototot...,

But, like the sum of two order types, the sum of an infinite series
of order types has no commutativity law. H.g. it can be proved that
the sum of the infinite series of order types

n+2n+3y-+...+nyn+...

changes its value with every change of the order of its components (7. €.
assumes 2” different values).
238 XII. Order types and operations on them

But if we form a new series from the infinite series of types

142434...

by changing arbitrarily the order of its components, we shall always ob-

tain a series whose sum is o.
It will be observed that two infinite order type series whose partial
sums of all orders are respectively equal can give different sums.
HE. g. for the infinite series a,+a,+... = (w*)??+1+14... and 6,+
18,2... == (@* PE ot ot +...) In view “ot (o*)?=- l= (oo (a).
we obtain a,+a,+...to,= §,+/.+...+ 6, = (w*) for n=1,2,..., but
a+ a,+...=(o*?+o ~6,+6.4+.. = (w*)? + o0*o.
EXERCISES. 1. Prove that 2=4A+1+44+1+4+4+1+...
Proof. The set X of all positive real numbers
ordered according to their magni-
tudes is, as we know, of type 4. Let us decompose the set X into an infinite series of
sets X = X,+ X,++..., where X, denotes the set of all real numbers w such that n—1 <
< «<n. The set X, will be a subset of type 1+ 1 of the set X. Hence, and in virtue
of the definition of the sum of an infinite series of order types, we easily obtain
A= (A+1)4+ ~A+1)4+ (44+ 1I4+...=4A+144A4+14+A4 1+...
2. Prove that (w+) =? forn=1,2,.
Proof. We have (@+n)o = (@+n)4+(m+n)+...= 0+ (n+ @)+ (n+ o)+..=
=ototot... =o’.
3. Prove that for natural k and J we have (w+k)(@+1) = w+ ol+k.
4. Prove that for §& = (w*+@)(w+o7+k) where k=0,1, 2, we have
& =(o*+ @) on for n = 2, 3,...
Proof. We have (w+ oa7+k)(w*+o)=...+o+@7+k+o Lk+o+ yo
+k+..=..+o+on+o+on+o+on+...=o(..1+7+ 14 7- cea ny + soe
= WN;
therefore & = (w*+ 0)(@+ o7n+hk)(w*+ @)(o+on+k) = (w* + o)on(@+ on + k) =
= (w*+q@)on, whence, more generally, & = (w*-+ w)an fork = 0,1,2,.., n= 2,3,...
On the other hand, it can easily be seen that types & (k = 0, 1, 2,...) are all dif-
ferent from one another.
5. Prove that for = w+ (w*+o)yn+o* we have &=& (L. Gillman).
Proof. ® = o+(o*+o)n+ o*+ w+ (@*+ o)yn+ w*+... = wot+ (w*4
+n+1-+...) = w+ (w*+o)7. It is easy to calculate that €* = €, whence w* =
I (w*+ @)n+ w*. Therefore: €(w*+ ow) = éw*+ Ew = (w*+ @) n+ w*+ w+ ((60)
= (w*+ w)(n+ 14 7) =(o*+ o)7n and thus & = o+ €(w*+ w) y+ éo* = w4+(
+ (w*+ 0) +(0*+ @)n
+ oF = © + (o*
+ ©)(n+. +7) + O* = 0 + (o*+ oN be
= 18; Gb Gh Gh

5. Power of the set of all denumerable order types. Let

(5.1) Ay, Az, Ag, ».
denote any infinite sequence formed of numbers 0 and 1. Let » = o* +a
and let us associate with sequence (5.1) the order type

(5.2) pt1i+a+gy+1+a+p+1i+a,+...
 § 5. Power of the set of all denumerable order types 239

We shall prove that to different sequences (5.1) will correspond

different types (5.2). Let A(o) denote an ordered set of type (5.2). From
the definition of the sum of an infinite series of order types and from
the observation that the set A is of type (5.2) it follows that there exist
disjoint sets F,, A,, Fz, As, Ps, A;,... such that

(5.3) Ae Dy Aye be Aso...

and F,,=q, A,=1+a, for n=1,2,..., while of two elements of the

set A belonging to different components of the sum (5.3) that one comes
first which belongs to the component that comes first.
Now let us group the elements of the set A(e) into components,
as In Exercise 25 of XII.3, and let us write down the successive finite
components: we shall obviously obtain an infinite sequence A,, A,,...,
with A, as the n-th successive finite component of the set A(o).
Now let
(5.4) Bebe
be any infinite sequence composed of numbers 0 and 1, different from
sequence (5.1) and let B(e,) be an ordered set of type

(5.5) DOME Dh 003 SEE ie NS I= eo

Therefore
B=G@,+B8,+G6,+B,4+...,

where G,= 9, B,=1+5), for n=1,2,... As before, we conclude that

the n-th successive finite component of the set B(o,) will be B,. If types
(5.1) and (5.5) were equal, there would exist a function f establishing
the similarity of the sets A(o) and B(o,), and the function f would ob-
viously map the n-th successive finite component of the set A(o) on
the n-th successive finite component of the set B(o,). Thus we should
have f(A,) = B, whence A,~B,, therefore 1+a,=1-+5),, which gives
a, = b, for n= 1, 2,..., in contradiction to the assumption that infinite
Sequences (5.1) and (5.4) are different.
_ Thus to every infinite sequence (5.1), composed of numbers 0 and 1,
corresponds a denumerable order type (5.2), different types (5.2) corres-
ponding to different sequences (5.1). Since, as we know, there are 2°
sequences (5.1) we obtain in this way a set composed of 2°° different denu-
merable order types. The set 7 of all the different denumerable order
types is thus of power > 2”.
Our proof of the formula 7 > 2° resorted to the axiom of choice
since we introduced infinite series of order types, and made use of the
axiom of choice in defining the sum of an infinite series of order types,
240 XII. Order types and operations on them

in XII.4. However, it is possible to prove that 7 > 2‘ without the aid

of the axiom of choice. For this purpose we should observe that we are
able to define a function f(a,, a, ...), associating with every infinite
sequence (5.1) of terms 0 or 1 a certain set of rational numbers (5.3)
ordered according to their magnitude, such that F, =, An=1+ 4
for n= 1, 2, and that of each two numbers of the set A belonging to
different components of the sum (5.3) that one is smaller which belongs
to that component which comes first. Then we should prove, as before,
that two sets (5.3) that correspond to different sequences (5.3) are never
similar. In this way we should obtain a set, of the power of the continuum,
of denumerable sets, of which no two different sets are similar. They
determine 2°° different denumerable order types, whence the formula
TS oe
We are not able, however, to prove without the aid of the axiom
of choice that T <2‘. With the aid of that axiom the proof of this for-
mula can proceed thus: From Theorem 2 of XIJ.8 it follows that every
denumerable order type is the type of a certain set of rational numbers
ordered according to their magnitude. The total number of such sets,
constituting denumerable subsets of the denumerable set of all rational
numbers, is, as we know, 2°°. Now let us divide those sets into classes,
assigning two sets of the same order type to the same class. According
to Theorem 1 of VI.7 the set of those classes will be of no greater power
than the power of all sets that we have been decomposing into classes,
i. e. of power <2". Now, our set of classes is of course of the same power
as the set 7. Hence the formula 7 < 2%?).
Formulae 7 > 2 and T < 2“ immediately give the equality T= 2%.
Hence
THEOREM 1. From the axiom of choice it follows that the set of all
denumerable order types is of the power of the continuum.

6. Power of the set of all order types of the power of the continuum. We
shall prove
THEOREM 1. We are able to define a set of power 2” ‘of sets of real
numbers, of the power of the continuum, of which no two different sets ordered
according to the magnitude of their elements are similar (see Sierpinski '[53],
p. 305-307).
Proof. In XI.10, Exercise 10, we have defined the perfect nowhere-
dense set C of Cantor, which is a complement to the closed interval (0, 1)

1) It will be observed that, analogously, we could prove in general (with the aid
of the axiom of choice) that the set of all order types of power m is of power <2™,
 § 6. Power of the set of all order types 241

of the sum of an infinite sequence of open intervals (10.1) (Chapter XI)

forming an everywhere-dense set in the interval (0,1) (intervals called
tangent to the set C). For brevity, let us denote those intervals by d,,
d,, ... We have also proved that the set C is of the power of the continuum.
For n=1,2,..., let a, and 0b, denote respectively the left-hand
and the right-hand end-point of the interval d,. We divide the interval d,
into n equal intervals; let c/, ci’, ...,c’~-» be the successive points of that
interval. Let P denote the set of all points

(6.1) Gat Gir Cr nan meee

WiOLOn = ole 2, a5
Let C, = C—P. Since the set C is of the power of the continuum, and
the set P is denumerable, the set C, is of the power of the continuum. The
set F of all subsets of the power of the continuum of the set C, will thus be
of power 27°‘
Now let us denote by @ the set of all sets 4+ P, where A « F; the
set ® will thus be of power 22°. Now we shall prove that two different
sets belonging to the set ®, ordered according to the magnitude of the
numbers that compose them, are not similar.
Let A and A’ be two sets belonging to F. Since the set of all intervals
d, (n= 1,2,...) is known to be dense and dense in the interval (0, 1),
therefore it is easy to see that the only component (in the sense of Exer-
cise 25 in XII.3) of the set A+ P (ordered according to the magnitude
of the numbers composing it) that consists of n+1 elements is the set
(6.1), which we shall denote by P,. It immediately follows that
if the sets 4+P and A’+P are similar and f denotes the function esta-
blishing the similarity of those sets, then f(P,)=P, for n=1, 2,...,
and since the set P,, being finite, can be mapped only identically upon
itself if the order of its elements is to be unchanged, we have f(a,) = a,
for n = 1, 2,... We also infer that /(P) = P, whence, in view of f(A +P) =
= A’+P and the invertibility of the function f in the set A+P, it fol-
lows that f(A) = A’.
Now suppose that A ~ A’; thus at least one of the sets A and A’
contains an element that is not contained in the other set, e.g. te A
and t¢ A’. Since te A, we have f(t) « f(A) = A’, and f(t) « A’. Since Ae F,
A’ <«F, we have AC C,, and A’C 0,, and 0, = C—P, whence it follows
that ¢ and f(t) are points of G. Cantor’s set which are not end-points of
any intervals tangent to C. It follows from the properties of Cantor’s
set (see XI.10) that between two points of the set C which are not end-
points of any interval tangent to the set C there lie infinitely many inter-
vals tangent to C. Thus if we had t< f(t), there would exist a natural
number such that t <a, < f(t); in view of t <a, and of the fact that
Cardinal and ordinal numbers 16
242 XII. Order types and operations on them

the function f retains the order arrangement of numbers according

to their magnitude, we find that /f(t)<f(a,), and since f(a,) = a,, we
have f(t)<a,, contrary to a, < f(t). If, however, we had f(t) <t, there
would exist a natural number n such that f(t) <a <t, whence f(a) <
<f(t), which, in view of f(a.) = a, again gives a contradiction. Thus we
must have f(t) = ¢ and since t « A, it follows that t « f(t) = f(A) = A’, con-
trary to t¢é A’.
Thus we have proved that if d«eFf, Be F and A ~ A’, then the
sets A4+P and A’+P (ordered according to the magnitude of the num-
bers belonging to them) cannot be similar.
Therefore the set ® is the set of power 22° of sets of real numbers,
of the power of the continuum, of which no two different sets ordered
according to the magnitude of their elements are similar. Theorem 1
is thus proved.
From Theorem 1 we immediately obtain the following
COROLLARY. The set of all order types of the power of the continuum is
of power > 2",
We have proved the above corollary without resorting to the axiom of
choice. We are not able, however, to prove without the aid of that axiom
that the set 7 of all order types of the power of the continuum is of power
< 22°. With the aid of the axiom of choice we can prove this in the fol-
lowing way.
As we know from XI.1, for a set A, of the power of the continuum
there exist 2@? — 22 different relations between its elements. Since
every order arrangement of the set A is a certain relation between its
elements, there exist at most 22° different order arrangements of the
set A. Thus if we divide the order arrangements of the set A into classes,
assigning two order arrangements of the set A to the same class if and
only if they are of the same type, then, in virtue of Theorem 1 of VI.7
(proved with the aid of the axiom of choice), the set of those classes will
be of power < 22°. Thus we have Wie 22, and since we have found be-
fore that T > 22°, it follows that T = 2", We have proved
THEOREM 2. From the axiom of choice it follows that the set of all order
types of the power of the continuum is of power 2?"°.
EXERCISES. 1. Define a set of the power of the continuum of denumerable sets
of rational numbers, of which no two sets, ordered according to the magnitude of their
elements, are similar.
Solution. Let 0, denote the set of all numbers of the form 1/3"-4, where n =
= 1, 2,... It can easily be proved that O, c OC, where O, denotes the set defined in the
proof of Theorem 1. Examining the proof of that theorem we conclude that the set
of all sets of the form A+ P, where A is a denumerable subset of the set O,, and P is
a set defined in the proof of Theorem 1, satisfies the conditions of our exercise.
 § 6. Power of the set of all order types 243

2. Let X denote the set of all real numbers ordered according to their magnitude.
Prove that for every non-empty set AcX there exist exactly 2° different sets cX
similar to the set A.
Proof. If AcX, Bc X andtif the sets A and B are jsimilar, then there exists a real
function f establishing their similarity, 7. e¢. a function defined for xe A and increasing
in the set A. But, as follows from Exercise{2 of VI.4, we can prove with the aid of the axiom
of choice that the set of all such functions is of power 2°. Hence the set of all differ-
ent sets cX similar to the set A is of power <2". Thus it remains to show that the
above set is of power > 2%. If the set A is bounded (even from one side only), then it
is easy to prove that all translations of the set A along a straight line are different from
one another and their number is 2*°. On the other hand every set of real numbers is
similar to the set of all real numbers lying inside the interval (0, 1) (since both sets
are of type A). Thus for every non-empty set Ac X there exist at least 2“ different
sets cX similar to the set A, q.e. d.
Now let m denote the power of the set of all different order types of the subsets
of the set X of all real numbers ordered according to their magnitude. From the theorem
proved in exercise 2 it immediately follows that all subsets of the set X can be divided
into m disjoint classes, each of which contains 2“° different sets. Hence the equality
m.-2%0 — 28°, which, as follows from Exercise of X.8,p.181,impliesm = 22°°. Thus we have
proved that the set of all different order types of the subsets of the set of all real num-
bers ordered according to their magnitude is of power 2?°, which immediately follows
from Theorem 1. :

7. Sum of an arbitrary ordered set of order types. Let ® be a given

ordered set whose elements are order types. From Hilbert’s axiom (VI.3)
follows the existence of a function f associating with each element » of
the set ® an ordered set A, of type y, the disjoint sets A, corresponding
to different elements g of the set ®+) Let

A= »'A,
ped

and let us order the set A in such a way that for two elements of the set A
belonging to the same component we shall retain the same order rela-
tion which existed between them in the set A,, and of two elements of
the set A belonging to two different components of the set A we shall
regard as coming first that element which belongs to the component
that comes first (2. e. the component whose index ¢ in the set ® comes first).
Obviously in this way the set A will be ordered and we shall easily prove
with the aid of the axiom od choice (as we did in XII.5 for the sum of
an infinite series of types) that the type o of the set A will depend: only
upon the ordered set ® of types gy, and not upon the choice of sets A,,
disjoint and such that A, = y for ye @®. The type o obtained in this way

1) We could resort here to the axiom of choice and not to Hilbert’s axiom Cf.
an analogical remark concerning the sum of an infinite series of types, § 4.
IGS
244 XII. Order types and operations on them

is called the sum of an ordered set ® of order types and is written

nays
Thus every ordered set of types gives a well-defined sum; that sum
depends in general upon the order arrangement of the components, but
it has the property of associativity. Particular cases of such a sum of
types are obviously finite and infinite series of types.
In particular, if the set ® is of type y and if each of its elements
constitutes the same type y, then it is easy to prove that

c= Me=ory.
gpeD

Thus the product of order types can be regarded as a DanC case of

the sum of an ordered set of order types.
If ® is an ordered set of type w* of order types, @ = "{...; a3,°a,, a},
then the type >) y is denoted by ....0,;+a,+
a.
ped
Thus for instance w* = ...--1-- 1-1)

8. Infinite products of order types. As follows from the definition of

the product of order types (XII.3), in order to form the product of a finite
number of order types a,, d2,..-, G1, we can proceed as follows. We take
ordered sets A; ({=1,2,...) such that A;= oa, for i=1,2,...,n, and
then we form their combinatorical product A,x A, ...x A, according
to the principle of last differences: a set ordered thus is of type
Cy Ag... An- |
Now suppose that we are given an infinite sequence of order types
(4 0) a, a,... If we took ordered sets A; (t= 1, 2,...) such that A;= a;
fori = 1, 2,..., then it would not always be possible to order the infinite
Cartesian product A,x A,x A,X... according to the principle of last
differences: it is impossible even in such a simple case as that of A;=
= {0,1} for 1=1, 2, since, for instance, for the sequences 0,1,0,1, 0,
1,... and 1,0,1,0,1,0,... there are no last different terms with the
same index. That is why the definition of the infinite product a,, az, as, ...
of order types cannot be introduced on these lines.
‘Now, all infinite sequences of type w*

++ Ag, Ag, Ay y

(v7. é. ordered according to decreasing indices) where a; « A; for? = 1, 2,...,

can always be ordered according to the principle of last differences since
for two different sequences of that kind, ..., a3, d@., a, and ..., ds, be, b,,
 § 8. Infinite products of order types 245

there always exists the least natural number ék such that a, ~ b,. Those
sequences form the infinite Cartesian product of type a*

ie eat Ae A Xe AG

The order type of the set P ordered according to the principle of

last differences is called the, infinite product of type w* of order types
@,, G,-.- and denoted by ...a,a,a,.
With the aid of the axiom of choice we can easily prove that if
A,, A, ...and B,, B,,... are ordered sets such that’ A;~B; fori = 1, 2,...,
phen, 22xGAaxX Antes. <abs eb, <b,. Hence tor every intimite se-
quence a,,a@,... of order types ~0 the infinite product (of type w*)
.. 30,0, 1S a well-defined order type, dependent only on types a,, a, ..-
and on their order (and not on the choice of ordered sets A; (¢ = 1, 2, ...)
such that A; = a; for i= 1, 2, ...).
It can easily be seen that if a,, a.,... is an infinite sequence of order
types such that a;= 1 for 71> (where vn is a given natural number),
then.
see Og Ag Oy — On On—1 eee Os Oy .

EXAMPLES 1. Let us calculate the infinite product of type w* of order types

... @@@ Which could also be denoted by w®”. It will be the type of the set of all infinite
sequences of type w*
wooy Ng5 No, Ny ;
composed of natural numbers, ordered according to the principle of last differences, which
is obviously similar to the set of all infinite sequences of type w

Ny, No, Ng5 oe »

composed of natural numbers ordered according to the principle of first differences.

If we associate with each sequence of this kind the number
2 —n 1192 —m—n
it ey) —Ny1 —Ny2 — 2.Seen
~

then we shall obviously establish similarity between our set of sequences and the set A
of all real numbers « such that 0 < x < 1, ordered according to the decreasing magni-
tudes of its elements. If we reversed the orderof the numbers of the latter set, we
should obtain, as we know, a set of type 4+ 1. It follows that the set A is of type
(A+ 1)* = 1*+4* =1+4. Therefore w® = 1-+4.
2. Let us calculate the infinite product of type w*

Soa AD PAD Mia

i. e. the power 2®'. It will obviously be the type of the set of all infinite sequences ay,
a, ..., composed. of numbers 0 and 1, ordered according to the principle of first differences.

1) In general, given a set F of type » of sets ordered by the relation @, we can define
the Cartesian product of type g of those sets as the set, of all sets {a,}4-, where a, «A
for A «Ff, ordered in such a way that for A, «Ff, A, «Ff, A, 4 A, we have Ag, 344, if
and only if 4,0A, in the set F.
246 XII. Order types and operations on them

With each sequence of this kind let us associate the number

a, a
(Sh 24...).
2(24
et-]
It can easily be seen that in this way we shall establish similarity between our
set of sequences and the set of all numbers forming the perfect nowhere-dense set
of Cantor (see XI.10), ordered according to the magnitude of the numbers which be-
ong to it.
Thus type 2® is the order type of the set of all numbers composing the perfect
nowhere-dense set of Cantor, ordered according to their magnitude.
3. Let us calculate the infinite product of type w*

. OF ww* ow*|o,
i. e. the power (w*w)”. It will obviously be the type of the set of all infinite sequences
Ny, —Ny,N3,—Ny,..., Where n,,%,,... are natural numbers ordered according to the
principle of first differences, 7. e. so that if m,, —m,, m3, ... and 21, —Ne, Nz, ... are two
different sequences of this kind and if k denotes the least natural number for which
M, F= Ny» then the first of those sequences is regarded as preceding or following the
second according to whether (—1)*t*m, < (—1)'t'n, or (—1)*t?*m, > (—1)'t',. Now if
we associate with each sequence n,, —N2, M3, —N4, ... an irrational number whose deve-
lopment into an infinite continued fraction is

4} cory
NM, |Ns

then we shall obviously establish similarity between the set of our sequences n,, —Nez»
Ng, +... ordered according to the principle of first differences, and the set of all irrational
numbers of the interval (0, 1) ordered according to their magnitude. Hence the conclu-
sion that (w*w)® is the type of the set of all irrational numbers ordered according to
their magnitude (Hausdorff [2], p. 151).
The infinite product of type w* of order types obviously has the property of as-
sociativity of its factors but not of their commutativity. It will also be observed that
two infinite products of type w* of order types may have all their partial products
respectively equal and still be different from one another. Thus for instance in each of
the infinite products 5
Spee oihogy. 6 ehiysl | sae SRA Oso,

all the partial products are equal to wm, and the products themselves are different from
each other since the first is obviously equal to wm and the second can easily be shown
to be of the type of the set C— {1} where C is the perfect nowhere-dense set of Cantor _
ordered according to the magnitude of the numbers belonging to it.
EXERCISES. 1. Prove that for every order type a0 there ‘exists an order
type €#~0 such that éa=&
Proof. It suffices to assume that =... aaa =a”.
_ 2. Prove that for every order type a 4 0 there exists an [order [type & ~ 0 such
that a& = é,
Proof. Since a # 0 we can of course take a= 8+1+y where # and y are order
types which can be equal to 0. Let us denote by é the sum of the infinite series of type
o*+q of order types
E=..+ aB+ oP B+ aB+ at ay+ ay+ ay+...
 § 8. Infinite products of order types 247

In view of the distributivity law of the multiplication of types with respect to

their addition we shall have hence

af =... a*B+ a®8+ a®8-+ a?+ a’y-+ a®y+ aty+ ...

whence, considering that a? = a(6+1+y) = aB+a+ay, we immediately obtain af = &.
3. Prove that for every order type a # 0 there exists an order type 40 such
that a = af = &,
Proof. According to Exercise 1 there exists an order type &, ~ 0 such thaté,a = &,,
and according to Exercise 2 there exists an order type & ~ 0 such that a& = &,. Let
&=&,&; we shall have

a = (&§,)a = §,(&a) = && = & and af = a(&,6,) = (a&)& = £26, = §

therefore a = a&é = &.

4. Prove that for every infinite sequence of order types 40, ay, a2, ..., there exists
an order type a ~ 0 and an infinite sequence of order types 6,, 6,,..., such that a =
SEACH Chyercs CO, WIE 1) 9 We Pts cos
Proof. Since a, # 0 for n= 1, 2,..., there exist, for n = 1, 2,..., order types £,,
and y, such that a, = 8,+1+y,. Let us denote by a the order type which is the sum
of the infinite series of type w*+ of order types
A = ooo + Qy Ay Ag By + Q, A, 83+ a, Po+ b+ 1+ Wt O12 Oy Ag 3+ Cy Ay Og Yq + eee

It is easy to prove by induction that the jsum of successive 2n+1 components

of this series, starting from the component a,a, ... a, __, 6, and ending with the component
OyAy ++» Gp_1 Yn» gives the type a,a,...a,, and hence, in view of the distributivity law,
we obtain a= aa,...a,6, where 6, = + On11%,19Basgt On41Beya+ Barat I+ Megat
+ Ons Papa Gn41%Mm+2 Bn4s+ 08

9. Segments and remainders of order types. Let a denote a given

order type. If o and og are order types such that a = o+ @ then the type o
is called a segment and the type e — a remainder of the order type a.
In particular, type 0, 2. e. the type of the empty set, is both a segment
and .a remainder of every order type. Order type 1 is both a segment
and a remainder of every finite order type as well as of types w+1,
1+ o0*, o©+o*, but it is neither a segment nor a remainder of type w*+
+q. Every order type is both its own segment and its own remainder.
Two different order types may be segments of one another, e. g. types 7
and 7+1, since 7 = (y7+1)+7. Similarly two different order types can
be remainders of one another, e. g. types 7 and 1+ 7 since y= 7+(1+7).

EXERCISES. 1. Give an example of infinitely many different order types each

two of which are segments of each other.
Solution. Such are for example the types ww*+n where n= 0,1, 2,... For
if m=0,1,2,..., and n= 0,1, 2,..., then in view of w*+1=—o*, n+o=o0, we
have mw*-+ m= w(wo*+ 1)+m = ow*+ 0+ mM = wo*+ (n+ o)+m = (@w*+n)+o+m,
which proves that the type ww*-+n is a segment of the type ww*+m. The proof that
our types are different follows from the observation that n is the greatest finite remain-
der of the type ww*+ n.
248 XII. Order types and operations on them

2. Give an example of infinitely many different order types any two of which
are remainders of each other.
Solution: Types n+ w*w where n= 0,1, 2,...
3. Prove that types /?, 2?-+- A, #?+A-+ 1 are different and each of them is a segment
of each of the other two.
Proof. The proof should be based on the equality 4= A+ 1+ A, whence
BHA=(REAL +A, BPSAAt1+A) = (PLATE = (PEAT D)+AT®).
THEOREM 1. Of two segments (remainders) of the same order type
at least one is a segment (remainder) of the other.
Proof. Let a=o+o@=o,+0,. Let A denote an ordered set of
type a. From the definition of the sum of order types it follows that
there exist decompositions of the set A, A= S+R=8,+R,, such that
[S,R] and [8,,R,] are cuts of the set A. But, as we have proved in
exercise 2, § 5, Chapter XI, we have either SCS, and ROR, or SOS,
and RC R,, whence it follows that either the set S is a segment of the
set S, and the set R, a remainder of the set F or the set S, is a segment
of the set S and the set R a remainder of the set #,. In the first case type o
will be a segment of type o,, and type 0, a remainder of type o, while in
the second case type o, will be a segment of type o and type o a remainder
of type 0.
THEOREM 2 (of A. Lindenbaum). If an order type a is a segment of an
order type B, and the order type B a remainder of the order type a, then a= f.
In other words, for order types the equalities

(Dek) a=o+p and ~p=ate

imply the equality a= f.
Proof. From formulae (9.1) it follows that B=o+f+o. Let B
denote an ordered set of type §. From the last formula for 6 and from the
definition of the sum of order types follows the existence of a decomposi-
tion B= 8,+B,+R, where S,=o, B, =f, R,=o and where the sum
of disjoint subsets S,, B, and FR, of the set B is ordered in such a way
that of two of its elements belonging to different components that one
comes first which belongs to the component that comes first. Since B = B
and B,=f, we have B~B, and there exists a function f establishing
a similar mapping of the set B on the set B,. Denoting the n-th itera-
tion of the function f by /”, we obtain, in view of B= S8,+B,+R, and
By= (Bb),

(9.2) P(B) = PS) + PB)

+ IR) for n=0,1,2,...
By (9.2), we obtain for natural n

(9.3) B= 8 +7(8i:)+P(8)
+... +78) + 7"4(B)
+ ("(B) +... + f(R) +h «
 § 9. Segments and remainders of order types 249

It follows from formula (9.2) that f"(S,) C f*(B) and /”+*(8,) C fet+*(B);
in view of S,f(B) = 8,B,= 0 and of the invertibility of the function /
we obtain /"(S,)/*+1(B) = 0, and since f(B)C B, whence /{*+*(B) C fe+1(B)
for natural k, we have /*+*(8,)C f-+1(B) and thus /*(8,)/"+*(S,)= 0 for
n= 0,1, 2,... and for natural k. Therefore from formula (9.3) it follows
that BD 8,+/(S,)+f(S,)+...; the components of this sum are disjoint
and of two of its elements belonging to different components always
that one comes first which belongs to the component that comes first, every
element of this sum preceding every element of the set B that does not
belong to this sum. Hence follows the existence of a set 7 such that

B= Si-F 7 (8) (7S) EL

and
B= S,4+f(S)+P(8) +..+7'.

But {*(8,) = o for n= 0,1, 2,...; therefore if we take 7 = 1, then

we shall have B= o+o+o04...+1, te. B= ow+t, whence a=ot+f=
=o-+00+T = o(1+o)+t= ow+7= £P, therefore a= B, q. e. d.
EXERCISES. 1. Determine all non-empty segments and remainders of type 7.
Answer. The only non-empty segments of type 7 are 7 and 7+ 1, and the only
non-empty remainders of type 7 are 7 and 1+ 7.
2. Determine all non-empty segments and remainders of type A.
Answer. The only such segments are 1 and 4+-1 and the only remainders 4 and
1+A.
3. Give an example of an order type different from 0 and from 1 whose segments
different from 0 are all equal.
Answer. Type w*.
4. Give an example of an infinite order type 4m whose remainders different
from 0 are all equal.
Answer. Type o?.
5. Prove that every order type different from 0, 1 and 2is a sum of two different
order types 40, but it is not every type of this kind that is a sum of two types dif-
ferent from that type. :
Proof. Let a denote an order type different from 0,1 and 2 and let A denote
an ordered set of type a. If the set A has a first element, then a = 1+ 6 where B 4 0
and 6 # 1 since a 4 1 and a # 2. If the set A has no first element, then let a denote
any element of the set A, A, the set of all elements of the set A that precede a, and
A, = A—A,. Thus we shall have A = A,+A, and, denoting by a, and a, the types
of the sets A, and A,, we shall have a=a,+a,, and a, 4 0, a, £0, a, $ ay,since the
set A, has no first element while the set A, has a first element, namely a. Thus in any
ease the type a is a sum of two different order types 40.
On the other hand it is easy to prove that the type 7 is not a sum of two order
types different from it. Types w, w* and 4 have a similar property.
6. Give an example of a denumerable order type having non-denumerably many
different segments (remainders).
 250 XII. Order types and operations on them

Solution. Let & denote the set of all rational numbers ordered according to
magnitude. In virtue of Theorem 1 of III.2, we can arrange the numbers of the set &
in an infinite sequence R = {7,,7,...}. For n=1, 2,... let us denote by ZN the set
of m systems (7,, 1), (7,5 2). ---, (7,, 7). Let

and let us order the set A by a relation 0 defined as follows: (7,,,k)o(r,,4) if 1, <p
or if 7, = 7, and k < l. We say that the set A(g) has 2° different segments and as many
different remainders.
Indeed, let » and y be two real numbers, x < y, and let us denote by A(x) the
segment of the set A composed of those of its elements (r,, &) for which 7, < x. Since
a2 <y, then there exists a natural number p such that « <r, < y. Hence it obviously
follows that the segment A (y) of the set A contains a component (in the sense of Exer-
cise 25 of XIT.3) 4, , consisting of p elements, but the segment A(x) of the set A con-
tains no such component. Therefore A(x) 4 A(y) for « ~y. Thus the order type A
has 2° different segments. We prove in a similar way that the type A has 2‘ different
remainders.
7. Prove that if a and f are order types such that a-2 = 8-2 then a= f.
Proof. Since a-2= 6-2, a+a= 6+. Types a and f are segments of the same
type. Thus it follows from the proof of theorem 1 that either a is a segment of type 6,
and $ a remainder of type a, or f is a segment of type a, and a remainder of type f. In
both cases, in virtue of theorem 2, we shall have a= f#.
8. Prove that if y is a finite order type, then for any order types a and f the for-
mula a+ y = 6+ y implies the formula a = f, and the formula y+ a= y+ 6 also implies
the formula a = 8, while if y is an infinite order type, then the following cases may
hold: 1) for any order types a and 6, at+y=f+y implies a= f, and y+a=y+ 6
implies a = f, 2) for any order types a and 6, a+ y= f+y implies a= 6 but there
exist order types a, and f, such that y+ a,= y+ f, but a, ~ f,, 3) for any order types a
and B, y+a=y+ implies a = 6 but there exist order types a, and 8, ~ a, such that
a+y=f,+y, 4) there exist both order types a, and 6, € a, such that qa+y=f,+y
and order types a, and B, ~ a such that y+a,= y+ py.
Proof. The proof of the first part of the exercise presents no difficulty. For the
second part we have the following easily provable examples: 1) y=-w*+o, 2) y=
Or, (ne Ih hie Dl Wes Oy Ch sil) fsa, 2b) Mew Caea th Sse lh, CRE
Bo=1+7.
9. Find two order types o and 6 £a such that a+7y=f+y and ynta=n7+8
Answer. a=1,f=1+4+ 7+1.

10. Divisors of order types. Let a denote a given order type. If »

and wy are order types such that a = yy, then is called a left-hand and wy
a right-hand divisor of the type a. In particular, the order type 1 is both
a left-hand and a right-hand divisor of every order type (since a= 1-a=
= a-1). Every order type is its own left-hand and its own right-hand
divisor. Two different order types can be right-hand divisors of each
- other, e. g. types 7 and 27, which are different since one of them is dense
while the other is not, and 7 = 7-27. It is more difficult to give an example
 § 10. Divisors of order types 251

of order types which are left-hand divisors of one another: we shall give
one in Chapter XIV.12. We do not know whether there exist two different
denumerable order types which are left-hand divisors of each other.
Neither do we know whether there exist two different order types which
are both left-hand and right-hand divisors of each other.
It is possible to give an example of a denumerable order type hav-
ing 2” divisors which are both left-hand and right-hand divisors, as
well as an example of a set of power 2 of denumerable order types such
that any type belonging to that set is a right-hand divisor of any other
type belonging to that set (Sierpinski [51]). It can also be proved that
there exist 2°° denumerable order types for which type 7 is a right-hand
divisor.
If a= gy, then the right-hand divisor y of type a is called comple-
mentary to the left-hand divisor » and conversely the divisor is called
complementary to the divisor y. Thus each (left-hand or right-hand)
divisor of a given order type has at least one complementary divisor.
For natural numbers, as we know, to every divisor corresponds a well-
defined complementary divisor, but a divisor of an order type may have
more than one complementary divisor. H. g. to the right-hand divisor
of the type w correspond infinitely many different complementary divisors,
since w = n-o for n= 1, 2,... (but to the left-hand divisor w of the type
® = ow-1 corresponds only one complementary divisor, 1). To the left-
hand divisor 7 of the type » correspond 2” different complementary
divisors as follows from Exercise 5 of XII.3. To the right-hand divisor
of the type w* correspond infinitely many different complementary divisors
since w? = (w-n)-w = (w+n)-o for n=1,2,... To the left-hand divisor
y+1 of the type 7 correspond infinitely many different complementary
divisors since for instance 7 = (y+1)-(@-+n) for n=1,2,..
EXERCISES. 1. Determine all the left-hand and all the right-hand divisors of
each of the types w, w*,7,7+1,1+7.
Solution. The left-hand divisors of type w are obviously the types 1, 2,3,...
and w (since w= nw for n=1,2,...) and those types only, the right-hand divisors
are only the types 1 and w. The left-hand divisors of the type w* are the types 1, 2, 8, ...
(since w*=nw* for. n=1,2,...), and the right-hand divisors are ,the types 1,
and w*. The left-hand divisors of the type 7 are the types 1,7, 1+7,7+1 and 1+7+1
(since: ny = (1+ )n = (1+ 7) = (1+7+1)y) and those types{only since leach {left-hand
divisor of a dense type is either equal to 1 or dense; thus type7 has’ exactly
five different left-hand divisors. The right-hand divisor of type 7 is every finite order
type 40 or denumerable type (cf. Exercise 5, XII.3); thus type 7 has 2°° different
right-hand! divisors. The left-hand divisors of type 7+1 are only the types 1, 7+ 1
and 1+7+1 (since 7+1=(1+7+1)(n+1) and eachi left-hand divisor of type
n+ 1 must be dense), and the right-hand divisors of the type 7+ 1 are all finite types
+0 and all denumerable types having a last element, and only those types. The left-
hand divisors of type 1+ 7 are [only the types 1, 1+ 7 and 1+ 7+1 (since 1+7 =
252 XII. Order types and operations on them

= (1+7+1)(1+7) and the right-hand divisors of the type 1+ 7 are all finite types 40
and all denumerable types having a first element, and only those types.
2. Prove that if 6 is a left-hand divisor of.an order type a ¢ 0, then there exist
order types 6 and y (which may be equal to 0) such that a = 6+6+y.
Proof. If 6 is a left-hand divisor of an order type a ~ 0, then there exists an order
type € £ 0 such that a = 6&. Since & # 0, there exist order types w and vy (which may
be equal to 0) such that €=yw+1+». Hence a= 6&= d(u+1+4+7) = du+6+6y and
taking du = B, dv=y, we have a= f+ 6+), q.e. d.
The order type 6 is called a portion of the type B+ 6+ y. The theorem proved
above can therefore be expressed as follows: Wach left-hand divisor of an order type
different from 0 is a portion of that type.
3. On the basis of Exercise 2 prove that the only left-hand divisors of type / are
the types 1, 4,4+ 1 and 1+4.

4. Prove that the types @, w*, w*+ @w*+o-2, w*+ w?, (w?)*+q@ are right-
hand divisors of type 4, and the type w+ w* is not a right-hand divisor of type 4.
Proof. The fact that the types w, w*, w*+ 0, o*+@-2, w*+o*, (w?)*+o are
right-hand, divisors of type 4 follows from the formulae 2= (A+ 1)w = (1+ A)wo* =
= (14-4) (@*+ @) = (1+ 4) (@*+ @- 2) = (1+ 4) (@*+ @?) = (A+ 1) ((@*)*+ 0), which can
easily be obtained from formulae (1+ A)m = 1+/, (a+ B)* = B*¥+a*, (aB)* = a*B* and
from the distributivity law.
It would be more difficult to determine all right-hand divisors of type. 4, because
they form a non-denumerable set.
In order to show that type w-+ w* is not a right-hand divisor of type 4 we should
use Exercise 3, which implies that if type w+ w* is a right-hand divisor of type A, then
at least one of the formulae 2=1-(w+o*), A=A(w+o*), A= (A+ 1)(@+ 0*), A=
= (1+ A)(@+ w*) holds; that, however, is not the case since 2 4 w+ w* and A(@+ w*) =
=A+AAZA, (A+-1)(@4+ w*) =A+A414/~, (14 A) (+o*) =14A4A Fd.
5. Prove that none of the types w, n+1, 1+7 and 1+7+1 is a product of
two order types different from it, but each of the types 7 and A is a product of two
order types different from it.
Proof. For type w the above immediately follows from the observation that this
type has only two right-hand divisors: 1 and w. In order to prove that type 7+ 1 has
the required property we shall observe that, as follows from Exercise 1, the only left-
hand divisors of type 7+ 1 are the types 1, 7+ 1 and 14+7-+1; thus if type 7+ 1 were
a product of two types different from it, then we should have 7+ 1= (1+7+1)f.
If the type 6 were not dense, then there would exist types y and 6 (which could be equal
to 0) such that 6 = y+ 2+ 6, whence 7+ 1 = (1+y+1)y+14+7+2+7y+4+14 (14+ 7+1)6
and the type 7+ 1 would not be dense, which is impossible. Thus type f is dense
and also it must have a last’ element since otherwise the type (1-++7+1)6 = 7+ 1 would
not have a last element, which is impossible. Moreover, the type f cannot have a first
element since otherwise the type n+ 1= (1+ 7+ 1) would have a first element. The
type B is at most denumerable as a divisor of a denumerable type. Thus it is a denumer-
able dense type having a last element and no first element. Therefore we must have
Bp =7-+1. Thus we have proved that type y+1 is not a product of two types dif-
ferent from it.
If we had 1+7= af where a41+7 and B#~1+y, we should have 7+1=
= (1+ 7)* = a*f*, where a* ¢ (1+7)* =n+1 and 6* #~7+1, which is known to
be impossible. Thus type 1+7 has the required property.
 § 10. Divisors of order types 253

As regards type 1+7+1, basing ourselves on Exercise 2 we easily prove that

the only left-hand divisors of this type are the types 1 and 1+7+ 1, whence it im-
mediately follows that type 1+7+ 1 is not a product of two types different from it.
(As regards the right-hand divisors of type 1+ 7+ 1, it is easy to prove that only the
types 1 and 1+7-+1 are such divisors, since each right-hand divisor of the type 1+
+n-+1, different from type 1, must be denumerable, dense and must have a first and
a last element).
But 7 = (n+ 1), A= (A+ 1); thus each of the types 7 and A is a product of
two types different from that type.
6. Give an example of an order type which is not a product of two order types
different from 1.
Answer. The type w+ 1. Indeed, let us suppose that w+1= af, where a4 1
and 6 ~ 1. It follows from Exercise 2 that a must be a portion of the type w+, 7. e
a finite type, w orw+1. If a were a finite type 41,a=n, where n is a natural num-
ber >1, then, on account of the fact that type 6 must have a last element (since the
* product af has such an element) and that 6 = y+ 1, we should have w+ 1 = n(y+1)=
=ny+n, which in view of n > 1 is impossible since type n > 1 is not a remainder of
type w+ 1. If we had a= ow, B £1, type B would have to possess a first and a last
element (since af has such an element) and we should have 8 = 1+ + 1, whence w+ 1 =
= o(1+ + 1) = w+ w&+ o, which is impossible. In a similar way we prove the impos-
sibility of a+ 1 = (w+ 1)6 where 6 ¥ 1. Thus type w+ 1is not a product of two types
different from 1, q.e.d.
It will be observed that type 1+7+1, although its only left-hand divisors are
the types 1 and 1+7-+1, the same types being its only right-hand divisors (see Exer-
cise 5), is a product of two types different from 1 since 14+ 7+1= (14+ 7+1)(1+
et):
7. Give an example of an order type for which every natural number is both
a left-hand and a right-hand divisor but neither m nor w* is a left-hand or right-
-hand divisor.
Answer. Type (@*+ )(1-+ 27+ 1).
8. Give an example of two order types which are common right-hand divisors
of a certain order type, but are not common left-hand divisors of any order type. ,

Answer. Such are for example the types w and 7 since nw = yn, but for every
order type € 4 0 type wé is not dense while type 7& is dense. Such are also the types w
and w* since 7a = nw*, and it can easily be proved that for any order types 4 0
and , 40 we have wi 4 w*é,.
It would be more difficult to give an example of two order types which are com-
mon left-hand divisors of a certain order type but are not common right-hand divisors
of any order type, or an example of two order types which are neither common left-
hand divisors nor common right-hand divisors of any order type (see Chapter XVI.12).
9. Give an example of two order types such that among their common lIcft-hand
divisors there is no divisor for which each common left-hand divisor of those types
would be a left-hand divisor.
Answer. The types w and w*, since their only common left-hand divisors are
natural numbers.
10. Prove that if a and f are two order types, then: a) a common left-hand divisor
of the types a and f is always a left-hand divisor of type a+ f, b) a common left-hand
divisor of the types a and a+ £ is not necessarily a left-hand divisor of type f, ¢) a com-
 254 XII. Order types and operations on them

mon left-hand divisor of the types 6 and a+f is not necessarily a left-hand divisor
of type a.
Proof. a) If a= da, B = 6f,, then a+ Bf= d(a,+ f,).
b) We have 7+ (1+) =7 and type 1+ 7, having a first element, cannot have
the left-hand divisor 7.
c) We have 1+w:1=q@-1 and for type 1 type o is not a left-hand divisor.
11. Prove that if a and B are two order types, then a) a common right-hand divisor
of the types a and f is not necessarily a right-hand divisor of type a+ f, b) a common
right-hand divisor of the types a and a+ is not necessarily a right-hand divisor of
type 8, c) a common right-hand divisor of the types 8 and a+ f is not necessarily a right-
hand divisor of type a.
Proof. a) Type 2 is a common right-hand divisor of the types n = 7-2 and 2,
but is not a right-hand divisor of type 7-++ 2 since if we had 7+ 2 = y- 2, we should have
n+2=y+y, whence y ~ 1 and thus type 2 would be a remainder of type y; there-
fore y=6+2, whence »=6+2+6, which is impossible {since 6+2+6 is not
a dense type: ;
b) We have w-2+1= (w+ 1)-2 and type 2 is not a right-hand divisor of type 1.
c) We have 1+1-m=1-q@ and type w is not a right-hand divisor of type 1.
12. Prove that type 1+7-+1 is a left-hand divisor of type 7+ 2+ 7.
Proof. The proof follows from the equality 7+ 2+7 = (1+n7+1)(n+2+7).
The proof of some theorems concerning divisors of order types is difficult, e.g.
the proof that every order type whose right-hand, divisors are the types 2 and 3 has
the right-hand divisor 6. A. Lindenbaum has given without proof a more general theo-
rem stating that if & and J are relatively prime numbers, and a and # are order types
such that ak = fl, then there exists a type € such that a = &l, 6B= ék (Lindenbaum
et Tarski [1], p. 321, Th. 14).
It can be proved that if n is a natural number and a and £ are order types such
that na = nf, then a = B. We could also replace the number n by the types w, w*, w*+
+q@,x7-+2, but we could not replace it either by n or by +1 since 7-2 = 7-1, while
(n+ 1)-2 = (y+1)-1. It can also be proved that if n is a natural number and a and 8
are order types such that an = fn, then a= f. But aw = fw does not imply a= 8
‘because 1-w = 2-q@; still, it can be proved that a(w+1)=£8(w+1) always implies
a =f and also a(1+o*) = B(1+o*) implies a = 6B (Sierpinski [45], p. 1-2).

11. Comparison of order types. Let a and / be two given order types,
A and B — ordered sets such that 4 = a, B=. One and only one of
the following cases holds:
1) The set A is similar to a certain subset of the set B but the set B
is not similar to any subset of the set A.
2) The set B is similar to a certain subset of the set A but the set A
is not similar to any subset of the set B.
3) The set A is similar to a certain subset of the set B and the set B
is similar to a certain subset of the set <A.
4) The set A is not similar to any subset of the set B, and the set B
is not similar to any subset of the set A.
It is easy to prove that if one of the above cases holds for the ordered
sets A and B and if A, and B, are ordered sets such that A,~A and
 § 11. Comparison of order types 255

B,~B then the same case holds for the sets A, and B,. Hence it follows
that the question which of the cases k (where k=1,2,3,4) holds de-
pends only on the order types a and 6 and not on the choice of ordered
sets A and B such that A=—a, B= f.
If it is case 1) that holds for the order types a and #6 we shall say,
after R. Fraissé [1] (p. 1330) (cf. also N. Cuesta [2], p. 132), that type
a is smaller than type f and type f greater than type a, and we shall
write a < 6 and 6 > a. Thus if it is case 2) that holds for the order types a
and f, then 6 <a and a>f.
If it is case 3) that holds for the types a and f, we shall say that
the types a and f are equivalent and we shall write a~f. Obviously in
that case we also have B~a.
If it is case 4) that holds for the types a and f, we shall say that
the types a and f are incomparable and write all.
It is easy to prove that if we have, for the order types a, 6 and y,
a—p and 6<y, then a4<y. If a<, then neither ~ < a. nor_a~f").
Tfia<f and f~y, thena<y;if a< 6 and a~y, then a < f. Thus order
types are partially ordered by the relation <. The relation of equivalence
of order types is obviously transitive.
EXAMPLES. It can easily be seen that if n is a natural number and a denotes
an infinite order type, then n < a. From Theorem 2 of XI1.8 it follows that for every
denumerable order type a we have a < yn ora 7. If a, B and y are order types such
that a = B+ y, then obviously 6 < aor Braand y <a orywxa. Since n = (n+ 1)+7,
we obtain 7+1# 7; thus two different types can be equivalent.
Each subset of an ordered set of type w is known to be either finite or of type w.
Each subset of an ordered set of type w* is either finite or of type w*. Since w 4 w*,
we conclude that the types w and w* are incomparable: || w*.
EXERCISES. 1. Prove that for every infinite type a one of the following four
formulae holds: a>w,axo,a>o*,avra*.

2. Prove that waxy <A<A-2<4A-3.

3. Prove that w+ 0*||wo*.
4. For every natural number n> 1 give an example of nm denumerable order
types each two of which are incomparable.
Answer. The types k+ w0*+ 0+ (n—1—k) for k= 0,1, 2,...,n—1.
Remark. We do not know whether there exists an infinite sequence of denumer-
able types each two of which are incomparable. Neither do we know whether there
exists an infinite sequence of decreasing denumerable order types (R. Fraissé [1] be-
lieves that there is no such sequence).
5. Give an example of an infinite series of order types a,+a,+... such that for
a certain order type 6 we have a,+a,+...+-a, < 6 forn=1, 2,..., but a,+a,+... ||B.
Answer. The series a,+a,+... where a,= 1 for n=1,2,..., and B= w*.

1) Comparing the power of sets V.1.

256 XII. Order types and operations on them

6. Give an example of an infinite series of order types a,+ a,+... with the sum o
and of a type f# such that

Q+at..ta,<Pp<o for ppm Wn eee

Answer. a, = (w*)?, a,—=o* for n=2,..., B= (w*)?+ o.. For we have here
a+ a,+...+ a, = (w*)?+ w*(n—1) = w*(w*+4+ n—1) = (w*)? for n= 1, 2,..., while o=
= 4+ ag+... = (w*)?+ H*w, and (w*)? < (w*)?+ a < (w*)?+ w*o.
7. Give an example of an infinite series of order types a,-+-a,+... with the sum o
and of types 6 and y such that

Che p Gh onapn i eo BO NN Mo oa

Answer. a, = (w*)*, a, = (w*)? for n=1,2,.., B= (w*+o, y = (o*)+ o*o.

We have here o = (w*)’+ (w*)?o@, a,ta,+...+ a, = (w*)® for n= 1, 2,...
8. Give an example of two ordered sets A and B, differing only in two elements
and such that the order types A and B are incomparable.
Solution. Let us denote by Q the set formed of the numbers 1—1/n and of the
numbers —1+ 1/n where n = 1, 2,..., and let A = {-1}+Q, B=Q-4 {1}. The sets A
and B, ordered according to the magnitude of the numbers that belong to them, satisfy
the required conditions.
9. Give an example of two incomparable types which are left-hand divisors of
the same order type.
Solution. The types w+o* and w*+q are incomparable and they are also
left-hand divisors of type (w*+ )? since (w+ w*)(w*+ @) = (w*+ @).
10. Give an example of two incomparable types which are both left-hand and
right-hand divisors of the same order type.
Solution. The types w+ w* and w*-+ w are both left-hand and right-hand divisors
of type (w*+ w)?7.

It will be observed that we know no example of a non-denumerable

type a such that a < 4. But with the aid of the axiom of choice we can
prove that there exists an order type a, of the power of the continuum,
such that a <A and also that for every order type, of the power of the
continuum, such that a <4 there exists-an order type, of the power of
the continuum, f, such that f < a (Sierpitski 53], p. 253, Théoréme 1).
With the aid of the continuum-hypothesis we immediately conclude that
every non-denumerable set of real numbers ordered according to their
magnitude contains a non-denumerable subset whose order type is smal-
ler than the order type of that set. We are unable to prove this theorem
without the aid of the continuum-hypothesis. Still, as will be proved
in Chapter XVII, it is not every non-denumerable ordered set that con-
tains a non-denumerable subset whose order type is smaller than the
order type of that set.
With the aid of the axiom of choice it can also be proved that for
every order type a < A there exists an order type f such that a< B <A,
and that if a, a,,... is an infinite sequence of order types <A, then
 § 11. Comparison of order types 257

a +a,+... <A (Sierpinski [53], p. 260, Th. 7, and p. 263, Th. 8). Now,
with the aid of the axiom of choice it can be proved that an ordered set
of type A is the sum of two subsets of itself, each of which is of type < 4.
Moreover, with the aid of the axiom of choice we can prove the fol-
lowing theorems.
There exist two order types a and f, of the power of the continuum,
such that a</<A and such that there is no order type & such that
o<€< 6 (Sierpinski [53];p-261, Th. 6).
There exist two order types g <A and y <A, of the power of the
‘continuum, for which there is no order type é, of the power of the con-
tinuum, such that <q and <y (Sierpinski [53], p. 261. Th. 10).
There exists a set, of the power of the continuum, of order types
<A each two of which are incomparable (Sierpinski [53], p. 258, Th. 4).
Furthermore: there exists a set, of the power of the continuum, of subsets
of a set of type 4 each two of which differ only in two elements and are
incomparable (Sierpinski [53], p. 258, Th. 5).
It can be proved that for every infinite order type there exists an
order type which is incomparable with it.

Cardinal and ordinal numbers 17

 CHAPTER XIII

WELL-ORDERED SETS

1. Well-ordered sets. An ordered set is said to be well-ordered if each

non-empty subset of that set has a first element; obviously we could
also say: if each proper subset of that set has a first element.
In order to prove that a given set A is well-ordered by the relation 0
it is not necessary to prove first that the set A is ordered by that rela-
tion: it suffices to show that for no elements a and b of the set A do aob
and boa hold simultaneously, and that in each non-empty subset A, of
the set A there exists an element a, such that a,ob for each element )b,
different from a,, of the set A,. (The connexity and transitivity of the
relation 0 follow hence easily).
We leave it to the reader to prove that in order that an ordered set
be well-ordered it is necessary and sufficient that it have a first element
and that in each proper cut of that set the upper class have a first
element. It can easily be seen that a set which is similar to a well-ordered
set is itself well-ordered.

Examples of well-ordered sets

Every finite ordered set is of course well-ordered. Every non-empty

subset of a finite ordered set has not only a first but also a last element.
This property is a characteristic of finite sets: every ordered set whose
non-empty subsets all have a first and a last element is finite. Let it be
observed that ordered sets whose non-empty subsets all have either
a first or a last element have also been investigated. It can be proved
that every set of that kind is the (ordered) sum of two components of
which the first is a well-ordered (or empty) set and the second (if it is
not empty) has an order arrangement which is inverse to well-ordering
(see Steckel [1]).
The sets of types o,. o+1, w+, w-m are obviously well-ordered.
The sets of types w*, 7, A are not well-ordered.
In contrast to well-ordered sets, there exist infinite ordered sets
no infinite subset of which has a first element; it is easy to prove that
such are ordered sets of type w* and only those sets.
 §$ 1. Well-ordered sets 259

Let A(o) denote a well-ordered set, A, — any non-empty subset of

that set. The set A,(e) is obviously well-ordered. To prove this it suffices
to observe that each non-empty subset of the set A, is at the same time
a non-empty subset of the set A, 7. e. it must have a first element. Thus
the well-ordering of a set is a hereditary property.
Hence it immediately follows that a well-ordered set contains no
subset of type m*. It is easy to prove with the aid of the axiom of choice
that, conversely, if an ordered set A(o) contains no subset of type o*,
then it is well-ordered.
Indeed, suppose that an ordered set A(o) is not well-ordered. The
set A(o) contains a subset A,, non-empty and having no first element.
Thus, for each element a of the set 4A,, the set S, of all elements x of the
set A, such that woa is non-empty. It follows from the axiom of choice
that there exists a function f(a), defined for a « A, and such that f(a) « S,
for a« A,, 7. e. f(a)oa for ae A,. Let a, denote any element of the (non-
-empty) set A,; the set {a,, f(a,), f(a), f(a,), ...} i8 a subset of type w*
of the set A(o), since ... of?(a,) of?(a,) of(a,) ea,. Therefore we have
THEOREM 1. Jt follows from the axiom of choice that the necessary and
sufficient condition for an ordered set to be well-ordered set is that it should
contain no subset of type w*.
It will be observed that we are not able to prove the sufficiency of
the condition of Theorem 1 without the aid of the axiom of choice.
It will also be observed that we are able to define an infinite ordered
set A(o) such that for each two elements of that set, a and b #£ a, we can
decide whether aob or bea, but we cannot decide whether the set A(o)
is or is not well-ordered. Such is for instance the set composed of all negative
integers and of all prime numbers of the form 2”+1, where n is a natural
number, ordered according to the decreasing magnitudes of the numbers
that belong to it.

2. The principle of transfinite induction. Let A(o) denote a given

well-ordered set, B — any set satisfying the following two conditions:
1° The first element of the set A belongs to B.
2° If a is an element of the set A such that each element «x of the
set A that precedes a belongs to 6, then the element a also belongs to B.
We shall prove that from assumptions 1° and 2° it follows that AC B.
Let us suppose that it is not so, 7. e. that the set C = A—B is not empty.
The set C, as a non-empty subset of the well-ordered set A, has a first
element, a. We shall distinguish two cases:
1° a is the first element of the set A. Then property 1° implies that a
belongs to B, which is impossible since ae C= A—B.
17*
260 XIII. Well-ordered sets

2° a is not the first element of the set A. Thus there exist elements
of the set A which precede a; let a’ denote any of them. The element a’,
since it precedes a, cannot belong to OC, a being the first element of the
set O. Thus we have a’ none C= A—B, whence, in view of a’ « A, it
follows that a’ « B. Therefore each element of the set A that precedes a
belongs to B, whence, according to property 2), we have ae Bb, which is
impossible since ae C= A—B.
Thus the assumption that A CB does not hold leads to a contradic-
tion. We have proved that AC B.
Let P denote the property of a well-ordered set A according to
which A has a first element and every set B satisfying conditions 1° and 2°
contains the set A. We have proved that every well-ordered set A has
the property P. Let us now prove that every ordered set having the pro--
perty P is well-ordered. Let A denote an ordered set with the property P
and let A, denote any non-empty subset of the set A and let us suppose
that the set A, has no first element. It follows that the first element
of the set A (which exists in virtue of the assumption that the set A has
the property P) precedes each element of the subset A,. Let us denote
by B the set of all those elements of the set A which precede each element
of its subset A,; thus the set B will satisfy condition 1°. We shall prove
that it will also satisfy condition 2°. Indeed, let a denote such an element
of the set A that each element of the set A preceding the element a be-
longs to B. If we had aé¢ B, then the element a would not precede each
element of the set A,, and since, according to our assumption, the set A,
has no first element, there would exist in the set A, an element a, preced-
ing a; in view of the definition of the element a, we should have a, « B,
which is impossible since A,B = 0. Therefore ae B and from the defini-
tion of the element a it follows that the set B satisfies condition 2°. Thus
the set B satisfies conditions 1° and 2° and from the assumption that
the set A has the property P it follows that AC B, which is impossible
since Ay G Ay AT 2 0 and i406 0:
Thus we have proved that each non-empty subset A, of the set A
has a first element. Therefore the set A is well-ordered.
We have proved that if an ordered set A has the property P, then
it is well-ordered, and since we proved before that also the converse
holds, the property P is a characteristic of well-ordered sets.
Now let A denote a well-ordered set having a first element, and T —
any theorem satisfying the following two conditions:
1° Theorem T is true for the first element of the set A,
2° If a is such an element of the set A that theorem T is true for
each element of the set A that precedes a, then theorem T is true for the
element a.
 § 2. Principle of transfinite induction 261

Let P, denote the property of an ordered set A according to which

the set A has a first element.and every theorem T satisfying conditions 1°
and 2° is true for each element of the set A. We shall prove that pro-
perty P, is equivalent to property P.
Suppose that an ordered set A has the property P and let T denote
any theorem satisfying conditions 1° and 2°. Let us denote by B the set
of all those elements of the set A for which theorem T is true. From
conditions 1° and 2° and from the definition of the set B it follows that
the set B satisfies conditions 1° and 2°; hence and from the assumption
that the set A has the property P it follows that A C B, 7. e. that theorem T
is true for each element of the set A. Thus we have proved that property P
implies property P,.
Suppose now that an ordered set A has the property P,, and let B
be an arbitrary set satisfying conditions 1) and 2). Let us denote by T
a theorem concerning an element of the set A, stating that the element
in question belongs to the set B. From the assumption that the set B
satisfies conditions 1) and 2) it immediately follows that theorem T
satisfies conditions 1° and 2°. Hence, in view of the assumption that the
set A has the property P,, it follows that theorem T is true for each ele-
ment of the set A, 7. e. that A C B. Thus we have shown that property P,
implies property P. The equivalence of the properties P and P, is proved.
We have proved above that property P is a characteristic of well-
ordered sets. In view of the equivalence, proved just now, of the properties
P and P,, also the property P, is a characteristic of well-ordered sets.
Property P, may be regarded as a generalization of the mathematical
induction principle; it is termed the principle of transfinite induction.
From what we have proved about property P, immediately follows
THEOREM 2. In order that the transfinite induction principle be applic-
able to an ordered set it is necessary and sufficient that the set be well-ordered.
The above theorem shows what an important role is played by
well-ordered sets.
3. Induction for ordered sets. We shall prove
THEOREM 1 (Blumberg [1], p. 818). Jf A is an ordered set and B its
subset such that
1° there exists a non-empty segment of the set A contained in B, and
2° for each non-empty segment A, of the set A such that A,C B and
A, 4A there exists a segment A, of the set A such that A,C A,C B
and A, ~ A,,
then A= B.
Proof. Let us denote by S the sum of all segments of the set A con-
tained in B; it will obviously be a segment of the set A contained in B,
262 XIII. Well-ordered sets

and, by 1°, we shall have S £0. If S # A then, by 2°, there exists a se-
ement 7 of the set A such that SC 7C Band T 4S. But, in view of
TC B and the definition of the set S, 7’ is one of the components of the
sum S; therefore we have TCS, which contradicts SC TAS. Thus
we must have S= A and, since SC B and BCA, we obtain A= B.
Theorem 1 is proved.

EXERCISE. Let P= AxXxB denote the Cartesian product of ordered sets A

and B. Each set A, xB, where A, is a segment of the set A and B, a segment of: the
setB let us call a segment of the set P. Prove that if C is a subset of the set P= AxB
such that
1) there exists a non-empty segment of the set P contained in C and
2) for each segment Q of the set P contained in C and different from P, there exists
a segment FR of the set P such that OcRcC and RFQ,
then C= A xB (Blumberg. [1], p. 819).

Induction for the set of all real numbers

From Theorem 1 we shall deduce the following theorem of A. Khint-

chine which may be regarded as the principle of induction for the set
of all real numbers (Khintchine [1], p. 165).
THEOREM 2. If theorem T (on a real number x) satisfies the fol-
lowing two conditions:
1) there exists a real number a such that theorem T is true for every
real number «<a,
2) af theorem T is true for every real number x < b, then there exists
a real number ¢ > b such that theorem T is true for every real num-
ber @& < ¢@,
then theorem T is true for every real number.
Proof. Let A denote the set of all real numbers ordered according
to their magnitude, and let T denote a theorem (on a real number 2)
satisfying conditions 1) and 2). Let us denote by B the set of all those
real numbers for which theorem T is true. By 1), condition 1° of Theorem 1
is satisfied. Now let A, denote a non-empty segment of the set A such
that A,C B and A, £4 A. Let Aj = A—A,; since A, # A we shall have
A; #0 and the sets A, and A; form a proper cut [A,, Ai] of the set A.
We shall now distinguish two cases.
a) In the set A, there exists a greatest number, a,. In view of A, C B
and of the fact that A, is a segment of the set A, theorem T is true for
every real number w < a, and, by 2), there exists a real number aj > a,
such that theorem T is true for every real number x < aj. The set of all
real numbers 7 < a; forms a segment A, of the set A such that A,C A,C B
andi. Gamay.
 ¥

§ 3. Induction for ordered sets 263

6) In the set A, there is no greatest number. From the continuity

of the set of all real numbers it follows that in the set Aj there exists
a least number, a;, and that the set of all real numbers w < az, is a segment
of A,; therefore, in view of A, C B, theorem T is true for every real num-
ber x < aj. By 2 there exists a Bo number aj’ > a; such that theorem T
is true for every real number « < ay’. The set of all real numbers «2 < a”
forms a segment A, of the set A such that A,C A,C B and A, # A,.
We see that in both cases, a and f, there exists a segment A, of the
set A such that A,C A,C Band A, #4 A,. We have proved that the set A
satisfies condition 2° of Theorem 1.
Thus the set A satisfies conditions 1° and 2° of Theorem 1; in virtue
of that theorem we have A = B, whence, in view of the definition of the
‘set B, we conclude that theorem T is true for every real number «.
Thus Theorem 2 is proved.

EXERCISE. Prove that among the ordered sets A(@) sets having no gaps, and
only those, have the property P according to which any theorem T (on an element of
the set A) is true for each element of the set A if that theorem satisfies the following
two conditions:
1. there exists an element a« A such that theorem T is true for each element «x of the
set A such that xea,
2. if be A and if theorem T is true for each element x « A such that xeb, then there
exists an element ¢ « A such that bec and that theorem 'T is true for every element x of the
set A such that xee.
Proof. The fact that ordered sets having no gaps have the property P follows
from the proof of Theorem 2, in which we use only the property of the set of all real
numbers (ordered according to their magnitude) consisting in their having no gaps.
Suppose now that an ordered set A (go) has a gap; let [A,, A,] be a cut of the set A (@)
giving a gap. The sets A, and A, are thus non-empty, the set A, having no last element
and the set A, having no first element. Now let T denote a theorem on an element x
of the set A stating that x « A,. Theorem T satisfies condition 1) since the set A, is
non-empty and is a segment of the set A; thus if a denotes any element of the set A,,
then theorem T is true for each element x of the set A such that xoa.
Suppose now that 6b « A and that theorem T is true for every element x « A such
that web. We cannot have b « A,, since in that case, the set A, having no first element,
there would exist an element b, « A, such that b, eb; in view of b,eb and of our assump-
tion concerning the element b, theorem T would be true for the element b,, which is impos-
sible because in view of b, € A, we have b, € A,. Therefore b « A,, and since the set A,
has no last element, there exists an element ¢ « A, such that bec. In view of ¢ e A, and
of the fact that the set A, is a segment of the set A we conclude that x « A, for xoc,
i.e. that theorem T is true for each element x of the set A such that xec. Therefore
theorem T satisfies condition 2).
Thus theorem T satisfies conditions 1) and 2) and yet it is not true for each element
of the set 4 since it is not true for any element of the non-empty subset A, of the set A.
Therefore the set A does not have the property P. We have proved that if an ordered
set A(ge) has a gap, then it does not possess the property P.
‘This completes the required proof.
*
 264 XIII. Well-ordered sets

4. Similar mapping of a well-ordered set on its subset. We shall prove

THEOREM 1. If f(x) 1s a function, defined for the elements of a well-
ordered set A(o), such that f(x)« A for xe A and f(x)of(y) for xe A, ye A,
xoy, then f(a)oea can hold for no element a of the set A. ;
Proof. Suppose that, for a certain element a of the set A(o), we
have the formula f(a) ea. Let us denote by B the set of all those elements «
of the set A for which f(x) ox. We have a« B and the set B, as a non-
empty subset of a well-ordered set A, has a first element, a,. In view
of a,« B and of the definition of the set B, we have f(a,)ea,. Let a, =
= f(a); a, will be an element of the set A and we shall have a, oa,, which,
in view of the properties of the function /,, implies the formula f(a.) of (a),
a. é. the formula f(a.) ea, which proves that a,« B. But this is impos-
sible since a, oa, and a, is the first element of the set B. Thus the assump-
' tion that the formula f(a)ea holds for a certain element a of the set A
leads to a contradiction. This proves the truth of Theorem 1.
If the function f establishes a similar mapping of the well-ordered
set A(g) on its subset, then we have of course f(x) « A for xe A and f(x) of(y)
for re A, ye A, voy. Thus from Theorem 1 immediately follows
COROLLARY 1. There exists no similar mapping of a well-ordered set
upon its subset in which the wmage of a certain elementa is an element
preceding a.
Suppose that a well-ordered set A(e) is similarly mapped on itself;
let f denote a function establishing that mapping, f-! — its inverse fune-
tion. Of course the function f- also establishes a similar mapping of
the set A(o) on itself. In virtue of Theorem 1 neither f(a)oa nor f—(a)oa
can. hold for any element a of the set A; hence it follows also that we
cannot have aef(a), for it would imply that {~(a) ef(f(a)), 7. e. f(a)ea
(since the function f establishes a similar mapping of the set A(o) on
itself), which is impossible. Thus, for each element a of the set A both
the formula f(a)oa and the formula aof(a) are false, which proves that
we must have f(a) = a for ac A. Therefore we have
COROLLARY 2. A well-ordered set can be mapped on itself only iden-
tically.
Suppose now that f and g are two different similar mappings of
a well-ordered set A on a set B. The function g(f(a)) obviously establishes
a similar mapping of the set A on itself and therefore it follows from
Corollary 2 that g7(f(a)) =a for aeA, whence f(a)= g(a) for ae A,
contrary to the assumption that the mappings f and g are different from
each other. Thus we have
COROLLARY 3. Two similar well-ordered sets can be similarly mapped
on each other in one way only.
 § 5. Properties of segments of well-ordered sets 268

5. Properties of segments of well-ordered sets. Principal theorem on

well-ordered sets. Let A(o) denote a given well-ordered set, a — a given
element of the set A. Let us denote by A, the set of all elements of the
set A that precede a; the set A, will be a segment of the set A different
from A (since it does not contain the element a of the set A), which may
even be empty (when a denotes the first element of the set A). Thus to
each element a of the set A corresponds a certain segment A, of the set A,
different from A.
Conversely, let A’ denote any segment of the set A that is different
from A. The set A—A’ is non-empty and, as a subset of the well-ordered
set A, it has a first element, a. Obviously A’ = A,. Thus the set of all
segments of the well-ordered set A that are different from A (including
the empty segment) is the set of all sets A, for ae A where A, denotes
the set of all elements of the set A which precede a.
From Theorem 1 of XIJI.4 immediately follows
COROLLARY 1. A well-ordered set is not similar to any subset of any
of tts segments that is different from the set itself.
Indeed, in a similar mapping of a well-ordered set A(o) upon a sub-
set T of a segment A, of the set A, to the element a of the set A would
correspond the element f(a) of the set A,, 7. e. an element preceding a,
contrary to Corollary 1 of XIIT.4.
In particular, we have hence
COROLLARY 2. A well-ordered set cannot be similar to any segment of
itself different from the set crtself.
As regards Corollary 1, it will also be observed that, as can easily
be proved, a well-ordered set is not similar to any segment of any of its
subsets that is different from the subset itself.
Of two different segments of a well-ordered set A(o) one is always
a segment of the other, namely if ae A, be A, aob, then the segment A,
is a segment of the segment A,. Thus from Corollary 2 it follows that
two different segments of a well-ordered set cannot be similar.
Of two different segments A, and A, of a well-ordered set A(o)
that one is said to be smaller which is a segment of the other. Thus
A, < A, if and only if aob. Hence the set of all segments of a well-ordered
set A that are different from A, ordered according to their magnitude, is
similar to the set A.
Therefore if two well-ordered sets are similar, then the sets of their
segments, different from the sets themselves, ordered according to their
magnitude, are similar, and vice versa. Since the set of all segments of
a given well-ordered set that are different from that set is well-ordered
according to their magnitude (being similar to the set in question), in
266 XIII. Well-ordered sets \

each non-empty set of segments of a given well-ordered set there exists a smal-
lest segment.
If two well-ordered sets are similar, then to each segment of one
set corresponds a similar segment (only one) of the other set. Now we
shall prove the inverse theorem, namely
THEOREM 1. Jf, for each segment of a well-ordered set A(e), different
from A, there exists a similar segment of a well-ordered set A’(o’), different
from A’, and vice versa, then the sets A and A’ are similar.
Proof. Let a denote an arbitrary element of the set A; according
to our assumption, for the segment A, of the set A there exists a similar
segment Sj of the set A’, and, as we know, only one such segment. The
element a’ of the set A’ is thus well-defined by the element a of the set 4;
let f(a) = a’. The function f establishes a similar mapping of the set A
on the set A’. Indeed, let a’ denote an arbitrary element of the set A’;
to the segment Aj, of the set A’ corresponds, in virtue of our assumption;
a similar segment A, of the set A and it follows from the definition of
function f/ that f(a)= a’. Thus each element of the set A’ is an image
of a certain element of the set A. Finally if a, and a, are elements of the
set A such that a,oa,, then, as we know, A,, < A,, and, for aj = f(a),
az = f(s), Ag < Ag (since Ay,~A, and Ag,~A,,), therefore aj gaz,
whence it follows that each element of the set A’ is an image of only
one element of the set A and the function f establishes the similarity of
the sets A and A’. Theorem 1 is proved:
It will be observed that Theorem 1 is generally not true for ordered
sets; e. g. it is false for a set A of type 7 and a set A’ of type 7+1.
Now let A(oe) and A’(o0’) be two given well-ordered sets and sup-
pose that in the set A(o) there exists a segment, different from A, that
is not similar to any segment of the set A’(o) different from A’. Let S,
denote the smallest of such segments of the set A. We shall prove that
each segment of the set A’ that is different from A’ is similar to a segment
of the set A, different from Aj.
Indeed, if there exist segments of the set A’, different from A’, not
similar to any segment of the set A, different from A,, then let Sj} de-
note the smallest of them. Thus each segment Aj; of the set Aj, different
from Aj, being a segment of the set Aj that is smaller than Aj, is similar
to a segment A, of the set A, different from A,, and we have A, < A,.
Therefore each segment of the set Aj different from Aj/ is similar to
a segment of the set A, different from A,. But also each segment A,,
of the set A, different from A is, according to the definition of the se-
ement A,, similar to a segment Ag of the set A’ different from A’, and
it follows from the definition of the segment Aj that Ay < Aj. Thus
 § 5. Properties of segments of well-ordered sets 267

each segment of the set A, different from A, is similar to a segment of

the set Aj, different from 4A/,. From Theorem 1 it follows that the sets 4,
and Az are similar, which contradicts the definition of the set A,.
Thus we have proved that each segment of the set A’ different from A’
is Similar to a segment of the set A, different from A,. But from the defini-
tion of the segment A, it follows that also each segment of the set A,
different from A, (being smaller than A,) is similar to a segment of the
set A’, different from 4A’. Thus from Theorem 1 it follows that the sets A’
and A, are similar, whence the set A’ is similar to a certain segment of
the set A different from A.
We have proved that if A(e) and A’(e’) are two well-ordered sets
and if in the set A there exists a segment, different from A, that is not
similar to any segment of the set A’ different from A’, then the set A’
is Similar to a certain segment of the set A different from A.
It follows hence that if in the set A’ there exists a segment, different
from A’, not similar to any segment of the set A different from A, then
the set A is similar to a certain segment of the set A’ different from A’.
Comparing these results with Theorem 1 we immediately obtain
THEOREM 2. Two well-ordered sets are either similar or one of them
as similar to a certain segment of the other set, different from that set.
If A and A’ are two well-ordered sets, then A~ A’ or the set A is
similar to a certain segment A; of the set A’ different from A’, and then,
by Corollary 2, we cannot have A~A’, or finally the set A’ is similar
to a certain segment A, of the set A different from A and then obviously
neither of the first two cases can hold.
We shall now deduce from Theorem 2 the following corollaries.
COROLLARY 1. Hach subset of a well-ordered set is similar either to the
set itself or to a certain segment of that set (different from the set itself).
Indeed, if a subset A’ of a well-ordered set A is not similar either
to the set A or to any of its segments, then, by Theorem 2, the set A
must be similar to a certain segment A; of the subset A’ different from A’.
In a mapping establishing that similarity the element of the set A which
forms the segment A; of the subset A’ would be transformed into an
element of that segment, 7. e. into an element preceding it, contrary to
Corollary 1 of XITI.4.
COROLLARY 2. Two well-ordered sets each of which is similar to a sub-
set of the other are similar.
Proof. Suppose that A and A’ are two well-ordered sets such that
AwA{C A’ and A’~A,C A and that the sets A and A’ are not similar.
Hence neither the sets Aj and A’ not the sets A, and A are similar, and
it follows from Corollary 1 that the set Az, and therefore also the similar
268 XIII. Well-ordered sets

set A, are similar to a certain segment of the set A’ different from A’,
and that the set Ai, and therefore also the similar set A’ are similar to
a certain segment of the set A different from A. Hence it follows that
the set A is similar to a certain segment of a segment of itself, different
from A, contrary to Corollary 4 of XII.4. Thus we have proved Corol-
lary 2.
Corollary 2 is not always true for ordered sets, e.g. it is not true
for sets of which one is of type 7 and the other of type 7+1.
Theorem 2 and Corollary 2 immediately imply
COROLLARY 3. If A and A’ are two well-ordered sets, then either they
are similar or the set A is similar to a certain segment of the set A’ different
from A’ and the set A’ is not similar to any subset of the set A, or finally
the set A’ is similar to a certain segment of the set A different from A, and
the set A is not similar to any subset of the set A’.
Now, if we recall the definition of inequality for order types, given
in XII.11, from Theorem 1 of XIII.3 we immediately deduce
COROLLARY 4. If A and A’ are two well-ordered sets, then one (and
only one) of the formulae
AT We A aA Vee Wea
holds.
In particular, the inequality A < A’ holds for well-ordered sets A
and A’ if and only if the set A is similar to a certain segment of the set A’
different from A’. From Theorem 2 also immediately follows
COROLLARY 5. If A and A’ are two well-ordered -sets, then always one
(and only one) of the formulae

Ae ba Lowry
holds.
Thus for the powers of two well-ordered sets we have trichotomy.
 CHAPTER XIV

ORDINAL NUMBERS

1. Ordinal numbers. Ordinal numbers as indices of the elements of

well-ordered sets. Order types of well-ordered sets are called ordinal
numbers. From Corollary 4 of XIII.4 it follows that each two ordinal
numbers can be joined by one and only one of the signs >, =, <. We
also know that each of these relations has the property of transitivity.
Number 0 (order type of the empty set) is also included among ordinal
numbers; it should be regarded as smaller than any other ordinal number.
Let A denote a given non-empty well-ordered set of type a. Further,
let. a denote an arbitrary element of the set A, A, — a segment of the
set A formed by the element a (7. e. the set of all elements of the set A
preceding a), y(a) — the order type of the segment A,, where y(a) = 0
if a is the first element of the set A. We shall have of course y(a) < a
for ae A and p(a,) < p(dg) if a, and a, are elements of the set A, a, prece-
ing a. In this way to each element a of the set A corresponds a certain
ordinal number & < a, a greater number always corresponding to a later
element. .On the other hand, every ordinal number § <a corresponds to
a certain element of the set A; indeed, if <a, then the set B of type &
is similar to a certain segment A’ of the set A different from A, and if
a denotes the first element of the set A—A’, then we shall obviosuly have
A’ = A,, whence é=y(a). Thus we have
THEOREM 1. A well-ordered set of type a € 0 is similar to the set of all
ordinal numbers >0 and <a, ordered according to their magnitude.
Thus the elements of a well-ordered set of type a may be denoted
‘by means of symbols a: where the indices = y(a) are ordinal numfers
<a including the number 0, as the index of the first element, a), of the
set in question. Thus for instance the elements of a finite set consisting
of n elements will be denoted by

yy Ay, Wz, +++ 9 An—-15

the elements of a set ‘of type ow by
Ags Ay5 Ae, see
 270 XIV. Ordinal numbers

the elements of a set of type w+ n where n is a natural number by

Dy Ay, Az +++ 5Woy Uo41) +++) Motn—-13

the elements of a set of type w-2+1 by
he lis ay Gey Gaia ialGniey er Caley
and so forth. In general, the elements of a well-ordered set A of type a
ean be arranged in a transfinite sequence of type a:
Cg py) Oa ee een iO aun ieee me (Sa Oe
which we shall also express by. writing A = {dz}s<q.
2. Sets of ordinal numbers. Now let. Z denote any non-empty set of
different ordinal numbers. Let m be any number of the set Z. The set
of all ordinal numbers (>0) smaller than ¢ is, in virtue of theorem 1,
well-ordered according to their magnitude and of type gy. Hence it im-
mediately follows that the set of all ordinal numbers >0 and <@ is also
well-ordered according to their magnitude (of type +1). Thus in that
set there exists a smallest number, belonging to Z (since m belongs to Z):
let us denote it by a. No ordinal number <a belongs to Z, since such
a number would be <q, which is contrary to the definition of number a.
Hence a is the smallest number of the set Z. Thus we have proved
THEOREM 1. In every non-empty set of ordinal numbers there exists
a smallest number.
From Theorem 1 we immediatelly deduce the following
JOROLLARY. Every set of ordinal numbers is well-ordered according
to their magnitude.
Like the concept of the set of all sets (see V.4), the assumption of
the existence of the set of all ordinal numbers leads to a contradiction.
Indeed, such a set W would, in virtue of the corollary proved above,
be well-ordered according to the magnitude of the numbers belonging to
it, 7. e. its order type, W =, would be a certain ordinal number, and
thus a certain element of the set W. According to Theorem 1 of XIV.1
the set W would thus be similar to the set of all ordinal numbers >0
smaller than qg, 7. e. to a certain segment of itself different from W, which
is contrary to Corollary 5 of XIII.4.
This is the famous antinomy of Burali-Forti, the oldest of all antino-
mies of the theory of sets, found as early as 1897.
The explanation of this antinomy is simple. In order to define an
“ordinal number we must first define a well-ordered set whose order type
is that number, and we are not able to construct the set of all ordinal
numbers necessary for the construction of the set W (it. being required
that sets be constructed gradually from previously defined elements).
 § 2. Sets of ordinal numbers 21
e

EXERCISE. Prove that if a,, aj, a3, ... is an infinite sequence of ordinal numbers,
then there exists an increasing infinite sequence of natural numbers k, < ky, < ks < ....
such that a, Ken <a, Kn sy for n= 1, 2,... (in other words that from an arbitrary: infinite
sequence of ordinal numbers we can pick out a non-decreasing infinite sequence).
Proof. Suppose that there exists an infinite sequence picked out from the sequence
Ay, Az, ++. &. g. the sequence a,, a, ... where I, <I, <..., in which there is no term
that is not smaller than each term of the given sequence. Then for the term a, there
exist terms a, such that a, <a,. Let a, denote the first of them; we shall have a, <
s s 81 a

<4, for 1 < s,, whence, in view of aq, <a , we shall have a, >a, fori<s,. If
: sy 8 G

we had a >a, for i> s,, then the term a would not be smaller than each term
sy v &

of our sequence, in contradiction to the assumption regarding that sequence. Thus

there exists a least index s, > s, such that a, <a, , whence we conclude that a <
a . . . . “il 82 . . . . i : a
<a, for i < s,. Continuing to reason in this way we obtain an infinite sequence of
89 =

Increasing indices s, < s, < s;,..., Such that a. <a, <..., therefore there exists an
increasing infinite sequence picked out from the sequence a,, a. ...
. . . . . . ‘1 52

Thus if it is impossible to pick out from the sequence a,, a, ... an increasing in-
finite sequence, then every increasing infinite sequence picked out from the sequence
@,, M,... contains a term that is not smaller than any term of the extracted sequence.
In particular, in the sequence a,, a, ... itself there exists a term a,,, > a, fori = 1, 2,...
Removing from that sequence the terms q,, a, ..., An» We shall obtain an infinite se-
quence G45 Oyo ss picked out from the sequence a,, a, ... in which there exists a term
Gn, = 4; for t > m,. Similarly in the set a, 1,4, 19+ there exists a term a, > a; for
«> m, ete. In this manner we finally obtain an infinite sequence a, > 4, > On, = +
and m, < m, < ... In this infinite sequence the sign > cannot appear infinitely many
times, since then we should obtain a decreasing infinite sequence of ordinal numbers,
which is impossible. Hence it follows that from some place in the sequence Ons = oe Shh
the sign = must appear. Thus there exists a non-decreasing infinite sequence (of equal
terms) picked out from the sequence a,,4:,... °
The required proof is thus complete.

3. Sum of ordinal numbers. Let pw denote a given ordinal number

and suppose that with each ordinal number é < @ a certain ordinal num-
ber az to be associated. We say then that we have a defined (finite or
transfinite) sequence Z of type g of ordinal numbers apo, a), ..., Go) Gotis «+s
az, -.- (<q). AS we know (XII.7), the sum of all ordinal numbers for-
ming a well-ordered sequence Z, is a certain order type o. We shall
prove that o is an ordinal number.
The ordinal numbers a: forming the sequence Z are ordered in this
sequence according to the magnitude of the indices ¢ corresponding to
them (and not necessarily to their:own magnitudes).
For 0 <é<p@ let As denote a set of type as; we may assume that
A,;Ay= 0 for & 4 é’1). Let S denote the sum of all sets A; for 0 <E<g@.
We order the set S so that for each two elements of the set S belonging

1) For example, we can assume B, to be the set of all ordinal numbers <a, and
take A, = B, x {§}. ;
272 XIV. Ordinal numbers

to the same component A, of the sum S that order relation which existed
between them in the set Az is retained, and of two elements belonging
to different components of the sum S we shall regard that one which
belongs to the component with the smaller index as preceding the other.
From the definition of the sum of types, given in XII.7, it follows that
the type of the set S ordered in this way will be o. Thus in order to prove
that o is an ordinal number, it remains to show that the set S is well-
ordered.
Let 7 denote any non-empty subset of the set S. Since SD T # 0 and

S= »' Ap,
0<é<

there exist indices € <q such that A;T ~0, and among them a smallest
index a (since the set of all ordinal numbers & such that 0 <é<@ is
well-ordered according to their magnitude). Thus we have A,T <0.
The set A,7, as a non-empty subset of the well-ordered set A,, contains
a first element, a. Obviously a is also the first element of the subset 7’
of the set S. Therefore each non-empty subset of the set S has a first
element, which proves that the set S is well-ordered, q.e.d. Thus we
have proved
THEOREM 1. The sum of every well-ordered sequence of ordinal numbers
is an ordinal number.
From Corollary 1 of XIII.5 immediately follows
COROLLARY 1. The sum of a well-ordered sequence of ordinal numbers
is not smaller than any of its components.
Thus if we add number 1 to such a sum, then in view of o+1>6
(since o is the order type of a segment of a set of type o+1 different
from that set), we obtain
COROLLARY 2. For every set of ordinal numbers there exists an ordinal
number greater than any number of that set.
The sum of two or more ordinal numbers depends in general on the
order of the components. The function /(n), expressing, for natural n,
the greatest number of different values that can be assumed by the
sum of » ordinal numbers in all the n! permutations of its components
has been investigated (Wakulicz [1], p. 254, Wakulicz [3], Erdos and
Rado [1], p. 127). It has been calculated that f(1) = 1, f(2) = 2, f(3)=5,
F(4)= 13, f(5) = 33, f(6) = 81, f(7) = 193, f(8) = 449, f(9) = 337, f(10) =
= 33°81, f(Cil) = 81%) (12) = SES103.- 4 (3) = 19528 =) (14) aoe sae
J(LD) = 33-18 17,5f (16) S09 C7 re Od 9S.) ULE) ae Mle ee
= 193%, #(20) = 33-813 and |
jin) =S8li(m—5), for m2 21.
 § 3. Sum of ordinal numbers 273

Thus we know the function f(7) for all natural numbers n. We have
f(n) <n! for n>3 and
limf(n)/nt = 0.

For each natural number k <5 there exist three ordinal numbers
whose sum, in all their permutations, gives k different values (Sierpin-
ski [48], p. 252). Similarly, for each natural number k < f(4) = 13 there
exist four ordinal numbers whose sum, in all their permutations, gives
k different values (Wakulicz [2], p. 23). But, as has been proved by Wa-
kulicz [1] (p. 261), there are no five ordinal numbers whose sum assumes 30
different values, although there exist five ordinal numbers such that
their sum, in all their permutations, assumes 31 different values. (Such
are for instance the numbers w?,@+1, w-2+2, w-3+3, w-4+4).

EXERCISES. 1. Prove that numbers w+ 2, w-2+1, w-4, w? in all their 24 per-

mutations give 13 different sums.
2. Prove that numbers 1, 2,4, 5, qm give 13 different sums in all their 120 per-
mutations.

3. Find how many different values are assumed by the sum of numbers 1, 2, 3,
4,q@ in all their permutations.
Answer. 11.

4. Prove that numbers w+ 1, w+ 2, wo-34 3, w-5+4, w, in all their permuta-

tions, give 29 different sums.
5. Prove that numbers w+1, w-2+2, w-4+3, w-8+4, mw’, in all their permuta-
tions, give 33 different sums.
6. Find all the different values that are assumed by the sum of the numbers w+ 1,
@m-2+2, wo-3+3, w-5+4, w*? in all their permutations.
Answer. There are 31 values; they are: w?,

wtoatl, w+w-242, #+o-34+1, w4+o-34+2, w+o-3-+3,

w+to-:44+1, o#+o-443, o+o-54+2, w+qo-54+ 3, w+o0-54+4,
o+o-64+1, o+0-642, w#+o-64 3, w+o0-6+4, w?+@-7-+2,
w+toa-7+4, o+o-8+1, w#+o-842, w+o-84+3, w+o-8+4,
oto-94+1, #+o-94+3, o+o-94+4, w?+w-104+2, w+o-10+3,
o+o-10+4, o+o-1l1l4+1, #@+o-114 2, w#+o-114+3, w#+o-1144

7. Determine a permutation of numbers w, w-2+1, w:3, w-5 and w? in which

their sum will be w?+ w-10+ 1 and a permutation in which their sum will be w?+
+q@-ll+1.
Answer. o+o?+o-3+@-54+ (@-2+1)=o?+o-10+1,

otota:-3+o-5+(0-2+1)=e?+o-ll+1.

8. Prove that the sum of n ordinal numbers w-+ 1+ 2+ 22+...+ 2”~*, in all the
possible permutations of the components, assumes 2”~* different values (namely it may
assume each of the values w+ k where k = 0, 1, 2,3, ..., 2"-*—1 and only those values).
Cardinal and ordinal numbers 18
274. XIV. Ordinal numbers

4. Properties of the sum of ordinal numbers. Numbers of the 1-st and

of the 2-nd kind. We shall prove
THEOREM 1. Jf a and B are ordinal numbers and B > 0, then a+fB> a.
Proof. From the definition of the sum of order types it follows that
if S is an ordered set of type a+, then S= A+B where A= a, B= f,
the set A being a segment of the ordered set S different from S since
£8 #0 and B+ 0. Hence, as we know (see XIII.5), S > A, and therefore
at+tp>a, q.e.d. i
Thus the sum of two ordinal numbers different from 0 is always
greater than the first component. But it is not necessarily greater than
the second component, since for instance 1+ w = w. Now, from corollary 1
of theorem 3 it follows that a+ 6 > #for any ordinal numbers a and f.
From Theorem 1 it follows in particular that for every ordinal num-
ber a we have
at+l>a.

It is also easy to prove that between the ordinal numbers a and

a+1 there is no intermediate number. jIndeed, let A denote a well-
-ordered set of type a. Number a+ 1 is thus the order type of the set A’,
obtained by joining to the set A an element a’, not belonging to it, which
should be regarded as succeeding each element of the set A. The set A
will thus be a segment of the set A’ formed by the element a’.
Now let » denote any ordinal number <a+1. Thus the set F of type
is similar to a certain segment of the set A’ different from A’, e. g. to the
segment formed by the element a of that set. Since a’ is the last element
of the set A’, we have either a = a’ or the element a precedes a’ (in the
set A’). In the first case we shall of course have g = a, in the second
case gp <a. We have proved that the inequality gy <a+1 implies the
inequality gm < a; thus there is no ordinal number é such that a << €<a+l,
quicxcs
Therefore every ordinal number has its next number: namely num-
ber a has a+1 as its next number.
However, it is not every ordinal number that is next to another
ordinal number, 7. é€. it is not every ordinal number that has its predeces-
sor, €. g. number w has none. Ordinal numbers having their predecessors
are called numbers of the 1-st kind, those having no predecessors are
called numbers of the 2-nd kind. Thus for instance 3,@+1, #:2+5 —
are numbers of the 1-st kind, wm, w-2, w? are numbers of the 2-nd kind.
Tf a is a number of the 1-st kind, then the number preceding a is
obviously the only solution & of the equation €+1—=a for numbers of
the 2-nd kind this equation has of course no solutions in ordinal num-
bers €.
 § 4. Properties of the sum 275

THEOREM 2. If a and 6 are ordinal numbers and a> p, then there

exists one and only one ordinal number y > 0 such that a= B+y; we de-
note that number, after Cantor, by a—f.
Proof. Let A denote a set of type a. In view of 6 <a, f is the type
of a certain segment B of the set A different from A. Let C= A—B;
we shall have C 40, A = B+O and, if we denote the type of the set C
by y,a=6+y.
Suppose now that we have simultaneously

(4.1) a=Pt+ty and a=f+y,

where y #,, ¢. g. y > yi. There exists, aS we proved just now, at least
one ordinal number 6 > 0 such that y = y,+6, whence, in view of (4.1)
and the associativity of the addition of order types:

a=B+y=6+(y146)=(6+y)+é6=a+6,
which is impossible since, by theorem 1, we have a+6> a.
Thus Theorem 2 is proved. As an immediate conclusion we obtain
COROLLARY 1. If, for ordinal numbers B, y and y,, the formula B+y =
= 6+, holds, then we must have y = ¥,.
It would also be easy to prove that if 6 is an ordinal number, and y
and y, are order types such that 8+ y= 6+ y,, then y = y,. Now, w*+1=
== SM

EXERCISES. 1. Prove that if a, 8, y and 6 are ordinal numbers such that a> B
and y > 6, then a+y>+6, and if a>f and y > 6, then a+y > B+.
2. Give an example of ordinal numbers a, f, y and 6 such that a > 8, y > 6 holds
but the inequality a+ y > B+6 does not hold.
Answer. a= 2, B= 1, y= 6=o. We have here 2+ 0= 1+ o.
3. Prove that if a and f# are ordinal numbers such that a > B, then we have a+
+n>B+n for n=1,2,...
Proof. As we know, between the numbers f and 8+ 1 there is no intermediate
number; therefore if a > B then we cannot have a < 6+1, t.¢. a>f+1, whence
atn>f6+1+n. But, since 1+n > n, we have 6+1+n > B+n. Thus a+n> B+n,
Qa e.2d.
4, Prove that for an ordinal number a the equality 1+.a = a holds if and only
ifalSoa.
Proof. On one hand, if a < w, then number a is finite and we have l+a> a.
On the other hand, if a > mw, then, by theorem 2, there exists an ordinal number y > 0
such that a=w-+y whence l+a=1+(0+y)=(l+o)+y=o+y=a.
5. Give an example of ordinal numbers a and B > a 1) such that a+ 8 < B-+4a,
2) such that a+ $f > B+a.
Answer. 1) a> 1 and 1+a#<o+1, 2) o+2>o+1 and (#+1)4(@+2)>
> (@+ 2)+ (@+ 1).
18*
276 XIV. Ordinal numbers

6. Prove that for every ordinal number a we have a+ 1+a= 1+a+a.

Proof. If a is a finite number, then the above equality is obvious, and if a is not
a finite number, then a > w and, according to Exercise 4, we have 1+a=a; there-
fore a+ l+ta=a+(l+a)=a+a=(l+a)+a.
From Theorem 2 of XIV.4 follows that for any ordinal numbers a and 6 <a the
difference a—f is the defined ordinal number, such that a—f = 0 and for a > 6 we
have a—f > 0. Moreover, it follows from Exercise 5 that for a > w we have a—1= a.
Hence it follows that if an ordinal number a > m has the antecedent, then this antecedent
is <a—l. From Theorem 2 of XIV.4 it follows immediately that for any numbers a and 6
we have (a+ 8)—a=/f. However, for a=o, B =w+1 we have (a+$)—f ¢a (he-
cause w-2+1= (wa+1)+(m+4+ 1), whence [w+ (+ 1)]—(@+1)=0+1F o).
7. Prove that if a,a, and £ are ordinal numbers such that a2>a,>f, then
(a—B)—(a,—B) =a—aq,. '
8. Find ordinal numbers a and £ such that (a—f)+f6 4a.
Answer. a=ot+1, B=o.
9. Prove that if a > a, > Bf, then a—f > a,—8.
10. Prove that if a > 6B > f,, then a—fS <a—f, and give an example showing
that we can have here a—f =a—f,.
11. Prove that (m?+ 1)(@+ 1)— (w+ 1)(w?+-1) = w?+1.
12. Prove that if a > B, then d(a—f) = da—Odfp.
Proof. If a> 6, then taking a—6=y we shall have a=f+y, whence da =
= 0(6+y) = 68+ dy, and therefore da—6B = édy = 6(a—f), q.e. d.
Thus for ordinal numbers the law of distributivity of multiplication with respect
to subtraction holds, but only if the difference is the second factor, since for instance

(2-lhoa=o#~2-o—-1-wo=0,

(2—1)(@+
1) =@4+14 2(@4+ 1)—(@4+1)=1.

13. Prove that if y is an order type whichis not an ordinal number, then there
exists one and only one decomposition y= a+y where a is an ordinal number >0
and y is an order type having no first element.
Proof. Let F denote an ordered set of type y. Let us assign to class A each element
ae such that the set of all elements of the set F that precede a is well-ordered.
Let B= f—A; if b<« B, then the set of all elements of the set F that precede 6 is
not well-ordered (since in that case we should have b« A, which is impossible) and
therefore we obviously cannot have b < a or b= a for any a« A. Thus a 3 b for ae A,
b « B. We have here a cut [A, B] of the set F (class A may be empty).
The set A is well-ordered, for if it contained a subset a,> a,\a,;© ... of type w*,
then the set of all elements of the set # preceding a,, as containing the subset a,,
Gz, +.., of type w*, would not be well-ordered, contrary to the fact that a, « A. The set B
cannot be empty since in that case we should have Ff = A, in contradiction to the as-
sumption that the type @ of the set # is not an ordinal number. The set B has no first
element since for such an element 0b the set of all elements of the set F that precede b
would be identical with the set A, 2. e. it would be well-ordered and we should have
b ¢ A, which is impossible. Let A = a, B = y; thus p = a+y where a is an ordinal num-
ber y — an order type having no first element. Thus the required decomposition exists.
Suppose now that we have two different decompositions: p=a+ty=at+y
where a and, a, are ordinal numbers and py and y, are order types having no first ele-
 § 4. Properties of the sum 277

ment. From XII.2, Exercise 10, it follows that there exists an order type y such that
either a, =a+y and p=y+y, or a=at+y,w=yty.
If we had y = 0, then we should have, in either case, a, = a and y = y,, which
is impossible because the decompositions a+ y and a,+y, are different from each other.
Therefore y ~¢ 0. In the first case the formula y = y+y, implies that the type y has
no first element, which is impossible since a, = a+ y and a, is an ordinal number. In
a similar way we obtain a contradiction in the second case. Thus there is only one
decomposition of type g having the required properties. The desired proof is complete.
14. Prove the following theorem of Z. Ohajoth: In every infinite ordered set we
can change the place of one element so that the order type of the set will be changed (Cha-
joth [1], p. 133).
Proof. If the set in question is well-ordered of type a> wo, then 1+a—=a and
a—l=a, therefore (a—1)+1—a+4+1>a; thus moving the first element of the set
to the last place we shall change the type of the set. And if the ordered set in question
is not well-ordered, then, in virtue of Exercise 13, it is of type a+y where a is an
ordinal number and y an order type having no first element. Thus there exist order
types y, and yw, such that y= y,+1+y,., type y, having no first element, whence it
follows that a+1+y,ty,4a+y,+1+y4,. By changing the place of one element
of our ordered set we can obtain an ordered set of a different type.
15. Prove that every ordered set is a subset of an ordered set of a different type,
having only one element more than the given set.

5. Remainders of ordinal numbers. According to the definition of the

remainders of order types (see XII.9), the remainder of an ordinal num-
ber a is what we call every ordinal number o for which there exists an
ordinal number € > 0 such that

(5.1) a=étQ;
the number & satisfying equation (5.1) is called a segment of number a
corresponding to the remainder o. For given a and o there may exist
more than one ordinal number é satisfying equation (5.1); e.g. for a=
= w-2, 0 =o formula (5.1) holds for =o and for é=w-+n where
n=1,2,...; thus a segment is not uniquely determined by a number
and its remainder, but among the segments of a given number cor-
responding to the given remainder of it there always exists a smallest
segment. H.g. the smallest segment corresponding to the remainder w
of number w-2 is number ow.
In virtue of Corollary 1 of XIV.3, formula (5.1) gives o <a; thus the
remainders of an ordinal number are not greater than the number itself.
THEOREM 1. Every ordinal number has a finite number of different
remainders.
Proof. Suppose that o and 0, are two different remainders of num-
ber a and that, for instance, o > o,. Thus there exist ordinal numbers &
and é, such that a=é+o and a= é,+0,, whence §+9=—£¢,+0,. We
shall show that & < é,.
278 XIV. Ordinal numbers

Indeed, if we had é > &,, then, by Theorem 2 of XIV.4, there would

exist an ordinal number 6 > 0 such that = &6,+6 and in view of + 0=
&,+0,, we should have é,+6+o0= &+0,, which, according to Corol-
lary 1 of XIV.4, gives 6+ 0= 0,, whence, according to Corollary 1 of
XIV.3, 0, > e, contrary to our assumption.
Thus we have proved that to a greater remainder of an ordinal number
always corresponds a smaller segment.
In virtue of the corollary to theorem 1, XIV.2, the set of all different
remainders of a given ordinal number a is well-ordered according to their
magnitudes. Thus if that set were infinite, there would exist an in-
creasing infinite sequence of different remainders of number a: 0, <
<Q. < @3<..., to which would correspond a decreasing infinite sequence
of segments corresponding to those remainders, ¢, > é, > &,>... (which
could be obtained for instance by associating with a remainder of number a
the smallest segment corresponding to it), contrary to the corollary to
Theorem 1, XIV.2, which implies that there exists no infinite sequence
of decreasing ordinal numbers. Thus Theorem 1 is proved.
THEOREM 2. The smallest positive remainder of every positive ordinal
number is not a sum of two ordinal numbers smaller than that remainder.
Proof. Let o denote the smallest positive remainder of an ordinal
number a> 0 and suppose that e@ = w+yv where uw and » are ordinal
numbers such that uw < @ and y < @. Since o is a remainder of number a,
there exists an ordinal number £ for which formula (5.1) holds, whence
a= &+(utyv)= (€+y)+7; thus number y is a remainder of number @
and »~0 because in case of »= 0 the formula 90 = u+yv would give
0 =m, contrary to our assumption uw < oe. The remainder v of number a
would thus be positive and, since » < 0, smaller than the smallest positive
remainder 0 of number a, which is impossible. Theorem 2 is therefore
proved.
THEOREM 3. If 0 and 0, > 0 are two remainders of the same ordinal
number a, then o is a remainder of 0,.
Proof. Denoting by é and , the segments of number a corresponding
to the remainders &0 and 0,, we shall have

(5.2) a=€+e=¢,+6

and > é, since (see the proof of theorem 1) to the greater remainder
corresponds the smaller segment. Thus, by theorem 5, there exists an
ordinal number 6 > 0 such that € = &,+ 6 and formula (5.2) gives 6,+6+
+o= &+0,, whence, according to Corollary 1 of XIV.4, we have 6+ 0=
= 0,, which proves that o is a remainder of number @,, q. e. d.
 § 5. Remainders 279

In particular from theorem 3 it follows that the smallest positive

remainder of every ordinal number is a remainder of every other positive
remainder of that number.
6. Prime components. Every ‘ordinal number >0 that is not a sum
of two ordinal numbers smaller than that number (or, which is obviously
the same thing: different from that number) is called a prime component
or a principal number of addition. Thus if an ordinal number a is a prime
component, then there exists no decomposition a= /+y where 6 <a
and y <a.
Among finite ordinal numbers only number 1 is a prime component.
Number @ is a prime component since every ordinal number <w and
thus also every sum of two such numbers is finite. The numbers w +1,
o-+q@ are not prime components of course.
From Theorem 2 of XIV.5 it immediately follows that the smallest
positive remainder of every ordinal number is:a prime component. (This
theorem is analogous to the theorem of arithmetic stating that the small-
est. divisor, greater than 1, of every natural number is a prime factor).
Jf a is a prime component and a= f+y, then 0<y<a cannot
hold, since, by theorem 4, we should then have also 6 < a, in contradic-
tion to the properties of number a; since, on the other hand, in view
of a= f6+y and in virtue of Corollary 1 of XIV.3, we have a> y, we
must have y= 0 or y= a. Hence a prime component has no positive re-
mainder different from tt. It is also easy to prove that among the positive
remainders of a given ordinal number only the smallest is a prime component.
This follows from Theorem 3 of XTV.5 and from the property of prime
components proved above.
THEOREM 1. In order that an ordinal number 0 > 0 be a prime com-
ponent it is necessary and sufficient that for every ordinal number E < 0
we have ;
(6.1) €+0=0.
Proof. Suppose that the ordinal number o is a prime component
and let é denote an ordinal number such that € < 0. According to Theo-
rem 2 of XIV.4 there exists a number y > 0 such that
(6.2) . Oe).
If we had y < g, then, by (5.1), number @ would be the sum of two
ordinal numbers smaller than 0, contrary to the assumption that it is
a prime component. On the other hand, by (5.1) and in virtue of Corol-
lary 1 of XIV.3, we have y < o. Therefore we must have y = o and for-
mula (6.2) gives formula (6.1). Thus we have proved that the condition
of Theorem 1 is necessary.
280 XIV. Ordinal numbers

Suppose now that 0 is a positive ordinal number such that for every
ordinal number & < @ formula (6.1) holds. If number o were not a prime
component, there would exist ordinal numbers mw and y such that @=
= pty, uw <o,»<o. By formula (6.1) for € < @ we should have »+ o= @
and w+o= oe, therefore
OO We?) Oe (0 8) = OO,
which is impossible since, in view of 9 > 0 and according to theorem 4,
we have 90+0 > o. Thus number 0 is a prime component. We have proved
that the condition of Theorem 1 is sufficient.
Theorem 1 is thus proved. It implies that a prime component absorbs
every component that goes immediately before it and is smaller than 1.
EXERCISE. Prove the following theorem:
If @ is a prime component and a any ordinal number, then

(6.3) Et+atoe=ate. for E<@g.

Proof. If @ is a prime component, then, according to theorem 1, we have for-
mula (6.1) for é< oe. If f+a<a, then of course we have +a=a and formula (6.3)
is true. If +a > a then we cannot have a> &m, since then, by Theorem 2 of XIV.4,
we should have a = w+7 where t>0, whence +a= &+é@m+7= (1l+o)47T=
=ém+t1t=a, therefore é+a=a, contrary to the assumption. Thus a < w = +
+é&+é+..., and a is a segment of number + é&+&+..., different from it, whence i
easily follows that there exists a natural number » such that a < én, which gives @<
<at+ex<é+a+e<E+m+e=0 (since +0=0 for &<@). Hence ateg=é+
+a+to, i.e. formula (6.3).

THEOREM 2. Hvery ordinal number > 0 is the sum of a finite number

of non-increasing prime components, and there is only one such decomposi-
tion for every ordinal number > 0.
Proof. Let a denote a given ordinal humber > 0; let us denote by @,
its smallest positive remainder and by a’ the smallest segment of number a
corresponding to that remainder; suppose that a’ > 0. Further, let us
denote by o’’ the smallest positive remainder of number a’ and by
the smallest segment of number a’ corresponding to that remainder. Thus

(6.4) a=a' +o’,

(6.5) O16, 0.
whence
(6.6) a=a’+o’+o'.

If we had 0’ < 0’, then, in view of the fact that o’ as the smallest
positive remainder of number a is, by Theorem 2 of XIV.5, a Pane
component and it follows from Theorem 1 that we should have 0+ 0’ = 0’,
therefore, by (6.6),
(6.7) (6, == a’ +o
 § 6. Prime components 281

From formula (6.5) it follows, in view of o’’ > 0 and of Theorem 1

of XIV.4, that a’ > a’; number a’’ would thus be, by (6.7), a segment
of number a, smaller than a’, corresponding to the remainder 0’ of num- /

ber a, contradicting the definition of a’. Thus we have proved that 0’ > 0’
must hold.
If we had a’ > 0, then, denoting by 0’ the smallest positive re- Q

mainder of number a’ and by a tt

the smallest segment of number a”
ttt

corresponding to that remainder, we could write

(6.8) a” = e+ 6",

and we should again have a’”’ , <a’ iad

and 0’ > 0”.
The sequence of inequalities obtained in this way cannot be infinite
since the numbers a’, a’, a’’’ constantly decrease and, as we know, there
is no decreasing infinite sequence of ordinal numbers. Thus we shall
finally obtain the equality
(6.9) aD = a + 9
where a= 0. From the successive equalities (6.4), (6.5), (6.9) we im-
mediately obtain
—— oo + o@—D + weet oO. -L 0.

Numbers 0’, 0”,...,0™ are prime components, as follows from

their definition and fork Theorem 2 of XIV.5, and we have 0’ <9” <...
Thus we have proved that every ordinal number > 9 is the sum of a finite
number of non-increasing prime components.
We shall now prove that there is only one such decomposition for
every ordinal number.
Let
(6.10) Q@ = 0:+ Cot... + On

be a decomposition of an ordinal number a> 0 into a finite series of

non-increasing prime components, 7. e. such that 0, > 0, >... > on. To
begin with, we shall prove that the first component 0, of the expansion
(6.10) is the greatest prime component <a.
Let us assume that there exists a prime component 0 <a greater
than o0,. In view of 0, < So and in virtue of Theorem 1, we easily obtain
1 nA if
Y Y VY VY Y VY

C= y+t0Hatate=.H—atat-- tate > atat---+O;

whence, in view of (6.10) and of the observation that 0, > 0. >... > 6,
we should obtain 9 > a, contrary to the assumption that o@ <a. Thus
we have proved the following
282 XIV. Ordinal numbers

LemMa. If an ordinal number a is the sum of a finite number of non-

-increasing prime components, a= 0,+ 0.+...+0n, then o, ws the greatest
prume component <a.
In virtue of our lemma, number 0, in the decomposition (6.10) is
well-defined by number a (as the greatest prime component <a). If the
series (6.10) contains more than one component, then we have a> 0,
and, by Theorem 2 of XIV.4, there exists one and only one ordinal num-
ber a, such that a= 0,+a,, and since, by (6:10), that number is e,+
+ o03,+...+0,, we have

(6.11) Oj == 09-F 054 .. <4 On

From formula (6.11), which gives a decomposition of number a

into a finite series of non-increasing prime components, we similarly
conclude that number o, is the greatest prime component <a; thus
number 0, is also well-defined by number a. If n > 2, then writing further
(ly = 02+a, we Shall obtain a, = o,+...+0, and number oe, will be the
greatest prime component < a,, ete.
In this way we conclude that all the components of series (6.10)
are well-defined by number a. Thus there is only one decomposition
of the ordinal number a into a sum of a finite series of non-increasing
prime components.
Thus Theorem 2 is proved.
Moreover, it will be observed that Theorem 2 and our lemma im-
mediately imply that in the set of all prime components not greater
than an arbitrary ordinal number a there always exists a greatest prime
component. Hence it immediately follows
THEOREM 3. If in a given set P of prime components there is no great-
est number, then the smallest ordinal nwmber a that is greater than any
number of the set P is a prime component.
Indeed, let us denote by P, the set of all prime components < a.
As has been proved, there exists in the set P, a greatest prime compo-
nent, o. Since it follows from the definitions of number a and of the set P,
that PC P, and since there is no greatest number in the set P, we con-
clude that o is greater than any number of the set P, whence 0 > a. But,
on the other hand, it follows from the definition of the set P,, in view
of oe P,, that @ <a. Thus a = 9, therefore a is a prime component, q. e. d.
And here is another proof of Theorem 3. If number a were not a prime
number, we should have a= uw+yv where u<a and »<a. From the
definition of number a it follows, in view of uw < a, that in the set P there
exist a number o, > uw, and a number oe, > 7; since there is no greatest
number in the set P, there exists in it a number oe such that 0 > 0, and
 § 6. Prime components 283

0 > @,. Thus @ > wu and @ >», and since 0, as a number of the set P, is
a prime component, we have, according to Theorem 1, a+o0= (uw+y)+
+o0=p+(y+e)= u+e=e, whence a < og, contrary to the definition
of number a. Thus Theorem 3 is true.
THEOREM 4. If a= 0,+0.+...4+ 0, is a decomposition of an ordinal
number a> into a finite series of non-increasing prime components,
then all the different positive remainders of number a, ordered according
to their decreasing magnitudes, are the numbers og+ Og41 +... +0, where
| TP etn
LEMMA. If a= e+06 where o@ is a@ prime component and if t is a re-
mainder of number a smaller than a, then t <6.
Proof of the lemma. If we had 1+ > 6, then, since (as has been
shown. in the proof of theorem 1 of XIV.5) to a greater remainder of
the same ordinal number always corresponds a smaller segment, the
segment 0, corresponding to the remainder 6 of number a, would be greater
than the segment € of number a, corresponding to its remainder t; thus
we should have 9 = +y where y > 0, whence (in view of the fact that
a prime component has no positive remainder different from itself), we
should have y= o. Therefore 0 = +0, which, in view of the formula
a=é4+17, gives €+0+6=¢+t1, whence, in virtue of Corollary 1 of
XIII.4: a= o+6=1T, in contradiction to the assumption that 1 < a.
Thus our lemma is proved.
Proof of Theorem 4. From the lemma it immediately follows
that if a= o+6 where o is a prime component and 6> 0, then there
is no intermediate remainder of number a between 6 and a; if, moreover,
6< a, then 6 is the greatest positive remainder of number a that is
smaller than a. In particular, let us consider decomposition (6.10) of
the number a into non-increasing prime components. Suppose that n> 1
and let
(6.12) 02 + O03 ++ + On = O43
we shall have, by (6.10):

(6.13) = 01+ 4;
which, as we know, gives a > a. If we had a= a,, formula (6.13) would
give a= o,+a, whence ’
1 2 n
gf
— i { ie lead .
O04 Oy oes Oy OS
10) TO ae e+ 1 01 5
therefore, in view of 0, > 0, >... > @, and (6.10), a < a, which is impos-
sible. Thus a, <a and therefore, according to the conclusion from the
lemma, formula (6.13) proves that a,, 7. e. number (6.12), is the greatest
positive remainder of number a that is smaller than a.
234 XIV. Ordinal numbers

Each remainder of number a that is smaller than the remainder a,

must, by Theorem 3 of XIII.5, be a remainder of number a,. Thus if
n>2, then writing 0,+ 0,+...+0,= 4 and reasoning as above, we
shall obtain a, < a, and prove that a, is the smallest remainder of num-
ber a, that is smaller than a,; therefore a, is also the greatest remainder
of number a that is smaller than a,. Reasoning thus further, we obtain
the proof of Theorem 4. .
7. Transfinite sequences of ordinal numbers and their limits. Let ¢
denote a given transfinite ordinal number and suppose that we are given
a transfinite sequence of type yw of ordinal numbers
(ib) Wis iy ecg Cas Cee yc ass Cente, ec ae

We denote the sequence (7.1) shortly by {ag}ecy.

If for E<f<a we always have ag < a, then the sequence (7.1)
is called increasing. In virtue of Corollary 2 of XIII.3, there exist ordinal
numbers greater than any term of sequence (7.1) and among them there
exists the smallest term (in virtue of Theorem 1 of XIII.2); denote it by A.
Suppose next that the type m of the sequence (7.1) is a number of the
2-nd kind (7. e. that the sequence (7.1) has no last term), and let ~ denote
any ordinal number </. From the definition of the number 4 (as the
smallest ordinal number that is larger than any number of sequence
(7.1)) it follows that the number w cannot be greater than each number
of (7.1); thus there exists an index », being an ordinal number <q de-
pendent on uw, such that uw <a,; since sequence (7.1) is increasing, we
shall have
(T2) [DOE DV LOL SON.
In other words: for every ordinal number uw <A all the terms of
sequence (7.1) starting from a certain place are contained between wu
and ». Thus the number 4 has a property analogous to the property of
the limit of an increasing sequence of real numbers. It will be natural
to call this number 4 the limit of the transfinite sequence (7.1), and to
write
(7.3) A= limes
Exo
Number / will obviously be of the 2-nd kind (for, in case of 4 =
= v-+1, formula (7.2) could not hold for u = y).
Therefore we can state
THEOREM 1. Hvery ancreasing transfinite sequence of ordinal numbers
whose type is an ordinal number of the 2-nd kind has a well-defined limit,
which is a number of the 2-nd kind (and also the smallest ordinal number
greater than any term of the sequence in question ).
 § 7. Transfinite sequences 285

Thus we have, for instance,

o = lmn = limn? = lim2”.

n<ow n<ow n<o

In general, for every ordinal number 4 of the 2-nd kind, we have

Av lines,
E<A

7. e. every ordinal number of the 2-nd kind is the limit of a transfinite

sequence formed of all ordinal numbers smaller than that number (ordered
according to their magnitude). In view of this fact (and of Theorem 1),
numbers of the 2-nd kind are also called limit numbers.
In spite of the above-mentioned analogy between the limit of a se-
quence of real numbers and the limit of a sequence of ordinal numbers,
there is a certain difference between them, noticed by Hoborski [1]
(0.193),
If a real number / is the limit of an infinite sequence of real num-
bers u, (n= 1, 2,...), then, writing /= u,+7,, we shall have

im 7,==.0..

But if an ordinal number / is the limit of an increasing transfinite se-

quence of ordinal numbers (7.1) and if we write 2 = a¢+ oz, then, starting
from a certain place, all the terms g¢ will be equal to a certain transfinite
ordinal number, namely to the smallest positive remainder 0 of number A.
Indeed, let
(7.4) A= O¢+ 0: for SES or

since 24 > a; for € < gy, numbers g; are all > 0, and since they are remainders
of number 4 (by (7.4)), we have oe; > @ for §& < gy. Number 4, as we know,
is of the 2-nd kind, therefore 0 is a transfinite number. All numbers o¢,
where <q, are thus not smaller than a certain transfinite number;
this fact alone proves that the limits of transfinite sequences behave in
a different way from the limits of infinite sequences of real numbers.
Since 0 is a remainder of number 4, there exists an ordinal number
ff > O90 such that
5
(7.5) A=p+e.
In view of (7.5) and 0 > 0, we have A > yw; thus, by (7.3), there exists
a number y for which formula (7.2) holds. By (7.2) and in wirtue of Theo-
rem 2 of XIII.4 there exists, for y <& <q, a number t; > 0 such that

(7.6) Cpe ie a) LOT 9 <b om.

286 XIV. Ordinal numbers

Formulae (7.4) and (7.6) give A= w+te+o, for » <E<g, whence,

by (7.5) and according to Corollary 1 of XIII.4, 97= t+ 0; for »<&<g,
therefore 9 > o; for » <&<g@, and since we proved before that 9; > 0
for <q, we have o:-=o for »<é<gq, q.e.d.
There are other differences between the properties of the lmits
of transfinite sequences of ordinal numbers and those of the limits of
infinite sequences of real numbers. Thus for instance the limit of the
sum of two transfinite sequences of ordinal numbers need not be equal
to the sum of the limits of those sequences. H. g.

lim(n+n)=o<o+o=lmn+lmn.
na<@w n<ow n<w

But we have
THEOREM 2. The formula
lim ag = A
é<p

implies, for every ordinal number y, the formula

(7.7) lim
E<q
(y+ag)= y +A.
Proof. Since
A —— hm ag
E<@p

we have A> a, for <q, whence y+A>y-+a, for é <q. On the other
hand, suppose that uw is an ordinal number such that w > y+ a; for <9.
Therefore we have w > y and, according to Theorem 2 of XIII.4, there
exists an ordinal number 6 > 0 such that uw = y+ 6, whence y+6>y+a¢
for <q, which gives 6 > az for <q; since 4 is the smallest ordinal
number greater than any of the numbers a for <q we must have
6>A, whence w= y+d>y+4. Hence y+A is the smallest ordinal
number that is > y+ ag for < y, which proves the truth of formula (7.7).
Thus Theorem 2 is proved.
But the formula
limag = A
E<—p
does not imply the formula
lim(ag-+y) = 4+y
E<o
for every ordinal number y, since for instance, in view of o+1> a0,
we have
lim(n +1) < limn +1.
1<@ n<o

Obviously, if {ag}ec, and {Pe}ecg are increasing transfinite sequences

of type y of ordinal numbers where » is a number of the 2-nd kind and
 § 7. Transfinite sequences hooe A

if ag > Bz then
lim de < lim Be ‘
E<p é<

Hence it immediately follows that if, moreover, {felec, 18 an increasing

transfinite sequence of ordinal numbers such that a: < ye < fz and if

lim ag= om Be,

<p
then
lim ye = limag .
é<p é<o

The concept of limit of a transfinite sequence of ordinal numbers

has been introduced only for increasing sequences. However, we may
generalize it to any transfinite sequences of ordinal numbers by saying
that the ordinal number / is the limit of a transfinite sequence {ag}z<g,
where gy is a number of the 2-nd kind, if and only if for every ordinal
number uw < A there exists an ordinal number » < y such that uw <a <A
for »v< é <@ (Sierpinski [60], p. 204).

8. Infinite series of ordinal numbers and their sums. Let {az}:<, denote
any transfinite sequence of type qm of ordinal numbers (not necessarily
increasing), where m is any ordinal number (not necessarily a number of
the 2-nd kind), and let o denote the sum of all numbers a: where & < 9g,
ordered according to the magnitude of their indices €. In virtue of Theo-
rem 1 of XIV.3, o will be an ordinal number. We shall write

C= Ata taet...+ Ant Gori t...tagt... (E< gq)

or

(8.1) oe
Let
(8.2) = dog for v<g.

The numbers o, are partial sums of the series (8.1). Thus for instance
we shall have o, = a, +a,+...+a,-1 for natural n, og = a) +a,+a,+...,
Cota—— Og O75 Cas O~.9 = Ag Oy Gy 4- +--+} Ao + Anti Cmte + «.-
We shall prove that if g is an ordinal number of the 2-nd kind and
ag > 0 for <q, then
(8.3) oe = limo; ,
v<@p

2. €. the sum of a transfinite series of positive ordinal numbers is the limit

of the transfinite sequence of its partial sums (as is the case with convergent
infinite seriesof real numbers).
288 XIV. Ordinal numbers

Indeed, if the components ag of series (8.1) are positive, then for

y<t<g we have t>v+1; therefore o, > 0,11: = 0,+4, >0,, whence
we conclude that the transfinite sequence {o,},<, is increasing. Since 1s
a number of the 2-nd kind, then for »y< y we also have »+ 1 < @ (because
it follows from » < g that v+1< q, and y = v+1 is impossible). By (8.1)
and (8.2) we have o > o,,; > o, and in order to prove that formula (8.3)
holds, it suffices to show that no ordinal number t < a is > 9, for all »y < 9.
Therefore suppose that t < o.
Let S denote a well-ordered set of type o; by (8.1) we have

S= )' Ag,
E<p

where A; is a set of type az, for é <q, each element of the set A; preced-
ing each element of the set A; for €<¢<y. Since t<a, t is the type
of a certain segment 7’ of the set S different from S. Let a denote the first
element of the (non-empty) set S—T; T will be the set of all the elements
of the set S that precede a. Since

aeS= > Ag,

E<@

there exists an index y <q such that ae A,, whence it follows that t is
the type of a certain segment of the set

pre
é<y

which is of type o,,1 and, in view of y <q and of number @ being of the
2-nd kind, we have y+1<p@. Thus t <o,41, which proves that t > 0,
for » <q is impossible. We have proved that formula (8.3) holds.
It will be observed that the sum of an infinite series of positive
ordinal numbers (unlike the sum of an infinite series of positive real
numbers) can change its value with the change of the order of its com-
ponents. For example, since w+? = w(1+o) = w?, we have

o+o?+14141+..=0?+a,
but
w+towti+i4+14+..=e?+o-2>ae%+o.

It can be proved that by changing the order of the components of an

infinite series of type of ordinal numbers we can obtain only a finite
number of different values of its sum (Sierpinski [48], p. 248). It can also
be proved that an infinite series of type w of increasing ordinal numbers
does not change its sum with any permutation of the components (Sier-
pinski [48], p. 250).
 § 8. Infinite series 289

EXERCISES. 1. Prove that the infinite series (of type w) w+ 1+2+3-+... does
not change its sum with any permutation of the components.
2. Prove that the sum of the transfinite series of type w+ 1:

(8.4) 14+ 24+ 34+ ..+ 0

assumes infinitely many different values with the change of the order of its components.
Proof. Let m-denote any given natural number. Interchanging the components n
and w in.the series (8.4) we shall obviously obtain the series (of type w+ 1) 14+ 2+...
t(m—1)+@+ (n+ 1)4+ (n+ 2)+...+”=@-2+n, and for n = 1, 2,... all these sums are
different from one another.
3. Prove that the sum of the transfinite series of type w-2

o-l+o@-2+o@-34+..+142434.

assumes infinitely many different values with the change of the order of the components.
Proof. Interchanging the components w-n and 1, we shall obviously obtain from
our series a series (of type w-2) whose sum is w?+ (n+ 1), and these sums are all
different from one another for n = 1, 2,...
4. For every natural number n give an example of an nfinite series (of type w)
of positive ordinal numbers whose sum, in all the permutations of the components,
assumes exactly n different values.
Answer. The infinite series a)+a,+a,+... where a =o*,a,=@ for i=l,
2,...,n—1, and a, = 1 fori > n. Indeed, if after a given permutation of the components
of this series. the number of all the components equal to wm and preceding the component w?
is k (where 0<k<n-—1), then the sum of the series is obviously w?+ w(n—k) and
these numbers are al) different from one another for k = 0,1, 2,...,n—1

5. Ordinal numbers of the form

Dee
E<a

where a is a given ordinal number are called triangular. Determine all the successive
transfinite triangular numbers < ?.
Answer. 0, :2, o-3+1, w:-44+2,..., w-n+(n—2),..., w?, w?+@-2, w?+
+o:-4+1, w?+o-6+2,..., w+q@-2n+ (n—1),..., w?-2, w?-21+0-3, w?-2+@-6+
+1...,, w?-2+@-3n+ (n—1),..., w?-3,..., w?-k, wo? -k+ow(k+ 1), w?-kt+o-2(k+1)+
+1,..., wo -kto-n(k+1)+ (n—1), ..., w%.
Remark. The number
on Ds E<w2

is the smallest transfinite triangular ordinal number which is the cube of an ordinal
number. In virtue of a theorem of Fermat, there exist no finite triangular numbers >1
that are cubes.

6. Prove that, for natural n, we have

au —
» ES "=",
E<w”

Cardinal and ordinal numbers 19

290 XIV. Ordinal numbers

9. Product of ordinal numbers. We shall prove

THEOREM 1. The product of two ordinal numbers is an ordinal number.
Proof. The product of two ordinal numbers is equal to 0 if and
only if at least one factor is equal to 0. Thus, for the proof of our theorem
we can assume that a and # are ordinal numbers >0. Let A and B be two
well-ordered sets such that A—a, B=f and lett P=AxXB. Let us
order the set P according to the principle of last differences; in virtue
of the definition of the product of types we shall have P = af and in
order to prove our theorem it suffices to show that the set P will be well-
-ordered.
Let C denote any non-empty subset of the set P. Thus there exist
elements b of the set B such that, for a certain ae A, (a,b) « C; let b, de-
note the first of those elements (such an element exists because the set B
is well-ordered). Thus there exist elements a of the set A such that
(a, b,) « O; let a, denote the first of those elements (such an element exists
because the set A is well-ordered). It can easily be seen that (a,, b,) is
the first element of the set C.
Indeed, from the definition of the elements b, and a,, we have
(a,,5,)«C. If there were an element (a,b) of the set C preceding
(a,, 5,) in that set, then, according to the definition of the ordering of
the set P, either b would precede b, in the set B, contrary to the defini-
tion of the element b,, or b would be equal to b, and a would precede a,
in the set A, contrary to the definition of the element a.
We have proved that each non-empty subset of the set P has a first
element, 7. e. the set P is well-ordered, q. e. d. Thus Theorem 1 is proved.
From Theorem 1 it immediately follows by induction that the product
of an arbitrary finite number of ordinal numbers is an ordinal number, and
from the properties of order types it follows that the product of ordinal
numbers has the property of associativity, (af)y = a(fy), and the property
of distributivity in the case where the sum is the second factor, a(f+ y) =
= af+ay (but the property of commutativity and the property of distri-
butivity in its second form do not hold because, for instance, w-2 ~2-a,
(1+1)-@ Al-w+1-@).
EXERCISES. 1. Prove that (@?+ o)5 = (w>+ 3).
2. Determine all the successive transfinite ordinal numbers <w? which are
squares of ordinal numbers.
Answer. w?, #+@+1, w+ 0-24 2, #+-3+3,..., e@to-ntn,..., w?-2+
@-2+1, w+o0-412, w+o-64+3,..., wto-Wntn,.., wo-3+@-3+1, w?+
o-6+ 2, w+o0-9+3, ..., wo-kt+o-k+1, w-kt+@-2k+2,:.., wo kt+@-kn+n,.:.
3. Determine the product of the positive ordinal numbers a and f if it is known
that a+ f=o.
Answer. af =o.
 § 9. Product of ordinal numbers 291

4. What values can be assumed by the product of two positive ordinal numbers a
and # for which a+fB=o-+1.
Answer. w+k where k= 0,1, 2,...

5. Represent the number w+ 1 as a difference of the squares of two ordinal numbers.

Answer. o+ 1= (+ 1)?—@?*.
6. Represent the number ow” **, where n is a natural number, as the difference of
the squares of two ordinal numbers.
Answer. wo"? = (w"+ @)?— (w")®.
7. Prove that number w is not a difference of the squares of any two ordinal
numbers.
Proof. Suppose that wo = a?— f?. Thus a? = 62+, whence a? > f? and a > &,
therefore a=>f6+1, whence P+o=a?> (8+1)?> £(6+1) = £?+ 8, which gives
o>f, therefore @2?= #+o<@?+o < wo+o+1=(w+1)?, whence a < (o+ iy
and a < w+1, therefore a<w and since o? = §?+ o>, whence a>o, we have
a= aq, and thus w? = B?+ @ whence it follows that 6 > w (because for B < w we have
f?-+-o =o), and thus w? > w?+ 0 > w?, which is impossible.
8. Represent the number w? in infinitely many ways as a difference of the squares
of two ordinal numbers.
Answer. w? = [w(n+ 1)}/?—(on)? for n= 1, 2,...
9. Represent number w® as a difference of the cubes of two ordinal numbers.
Answer. w® = (w?+ w)3— (w?)3.
‘10. Prove that for every natural number p there exists an infinite arithmetical
progression each term of which is the p-th power of an ordinal number.
Proof. Such a progression is (wn)? (n= 1,2,...) since [w(n-+ 1)])’— (wn)? =
=o’(n+1)—o'n =o” for n= 1, 2, ...
11. Give an example of ordinal numbers a, f and y such that B—a = y—f, but
aty#~B-2.
Answer.a=o, B=o+1,y=o-+ 2. We have here B—a = y—6 = landa+y=
=o:-21+2>8-2=0-2+1.
12. Give an example of two ordinal numbers a and f such that none of the num-
bers af” and fa” is the n-th power of an ordinal number, for natural n>1.
Solution. a= 2, B =w+1. Indeed, it is easy to prove by induction that, for
natural n, we have (w+ 1)"= "+ "—'+...4 + Land (w+ 2)"=o"+ 0". 2+ 0" 72+
= .-@ +24 9) and since 2(@-- 1) =o +o, i+..2 0-2 , for n=1

(ory =< (6-2 = (o2)",

whence it follows that the number 2”(w+ 1)” lies between two successive n-th powers
of ordinal numbers, 7%. e. it cannot itself be the n-th power of an ordinal number.
On the other hand (w+1)"- 2” = o"-2"+"-'+o"74+...4@+1, and (w-2")” =
=@"-2"< (w+ 1)"-2" <(@- 2°+ 1)"= wo” 2°+ w"—-2°4+ ...4 w- 2"+ 1, therefore for n>1

(On2.)) SS (O-P Neat elas Fe T)*

whence it follows that the number (w+ 1)”-2” lies between two successive n-th powers
n

of ordinal numbers, 7. e. it cannot itself be the m-th power of an ordinal number.

19*
292 XIV. Ordinal numbers

13. Give an example of two ordinal numbers a and # such that for every natural
number > 1 the number af” is, and the number f”a” is not, the n-th power of an
ordinal number.
Solution. a= +1, B = o. For we have here (w+ 1)°w” = (?)", but @"(@+ 1)” =
= oot nore cody ha!
(o@ + o) = Or RG ee eo (a+ 1)" < OG as
+o+1= (o?+o-+ 1)", whence it follows that the number w"(w+ 1)” lies between the
n-th powers of two successive ordinal numbers and therefore it cannot itself be the
n-th power of an ordinal number.
14. Give an example of two ordinal numbers a and 6 such that a+6=f+a
but a?+ 62+ 64 a.
Answer. a=o+l1, B=a@-2+1. For we have here a+ f=f6+a=—o:-3+1,
a+B? = w7?-3+ 0-24 1'> @-3+4+ 0+ 1 = f+ a.
15. Prove that there exist infinitely many pairs of ordinal numbers a and f such
that a? = 6? and that we do not have a = 73, 6 = 7, where t is any ordinal number.
Proof. a=o”"+o"", p=o"+o, where n=1,2,.., or a=o'+o”-24
bo 29 p=—o
+o -24-2 for n= 1,2.

10. Properties of the product of ordinal numbers. We shall prove

THEOREM 10. If a, B, 8, are ordinal numbers, a>0, B> p,, then
ap > ap,.
Proof. In view of 6 >, and Theorem 2 of XIII.4 we can write
B= p,+y where y > 0. Since a> 0 and y > 0 we have ay > 0, whence,
in view of the law of distributivity and by Theorem 1 of XIV.4, we obtain
af = o(6, +7) = a6, +ay > af;, i. @ a6 > af,, q. ed.
However, it will be observed that if a, a,, 6 are ordinal numbers such
that a >a, and 6 > 0, then we can have af = a,f because 2-0 =1-a.
THEOREM 2. If a, a,, 6, B, are ordinal numbers such that a > a, and
b> Py, then af > a,f,.
Proof. From the definition of the product of order types it im-
mediately follows, in view of a > a, and f > f,, that a,f, is the type of
a subset 7’ of the set Z which is of type af, and since, according to Corol-
lary 1 of XII.5, the set T is similar either to the set Z or to a certain
segment of the set Z that is different from Z, we have in the first case
af, = af, and in the second case a,f, < af, which proves the truth of
Theorem 2. ;
From Theorems 1 and 2 we obtain, in particular, for 6, = 1, or for
G = 1, py — £, thestollowme
COROLLARY 1. The product of two positive ordinal numbers is not small-
er than each factor, and if the second factor is >1 then the product is greater
than the first factor.
From Theorem 2 we also easily obtain
COROLLARY 2. If a > a, and if B is an ordinal number of the 1-st kind,
then ab > a,f.
 § 10. Properties of the product 293

Proof. Corollary 2 is obvious if 6 = 1. If 6 is a number of the 1-st

kind >1, then 8 = y+1, where y > 0. In view of a > a, and by theorem 2
we have ay > a,y, whence, according to exercise 1 in § 4, ay+a>ay+
+a > aqy+a,, and since af = a(y+1) = ay+a, and ay+a,= a(y+1)=
== ap, we Nave, ap > a,f, dq. e..d.
Corollary 2 immediately implies the following two corollaries:
COROLLARY 3. If a, a, and 6 are ordinal numbers such that afb = a, B
and B is a number of the 1-st kind, then a= a,.
COROLLARY 4. If a, a, and B are ordinal numbers such that aB = a,B
and a ~a,, then 6 is a number of the 2-nd kind.
From Theorem 2 easily follows
COROLLARY 5. For ordinal numbers the inequality ap > a,b, implies
p> py.
Proof. If we had 6 < f,, then, according to Theorem 2, we should
have af < af, in contradiction to the assumption. Thus 6 > f,, q.e. d.
In an analogous way we prove
COROLLARY 6. For ordinal numbers the inequality ab > ap implies
a> ada.

From Theorem 1 easily follows

COROLLARY 7. If a, 6 and 6, are ordinal numbers such that a > 0 and
GD. =O), then B= p,.
Proof. If we had 6 > f,, then, in view of a> 0 and by Theorem 1,
we should have af >af,, contrary to our assumption. We prove in
a Similar way that 6 < f, is impossible. Thus 6 = f,, q. e. d.
From Corollary 7 it immediately follows that for each left-hand
divisor of a positive ordinal number there exists a well-defined comple-
mentary right-hand divisor of that number. That is not the case as re-
gards right-hand divisors of an ordinal number; for a right-hand divisor
of an ordinal number there may exist infinitely many different comple-
mentary left-hand divisors of that number, since for instance (m-”)-o@=
= for n=1,2,..., and the numbers w-n are all different from one
Anocner TOV i== 1 Zier...
It has been observed by A. Lindenbaum that Corollary 7 remains
true if a is an ordinal number while / and /, are order types (Linden-
baum et Tarski [1], p. 322, Théoréme 15). But if 6 and f, are ordinal
numbers and a is an order type, then Corollary 7 is not necessarily
true, since for instance 7-1=7-2°and similarly (y+1)-1=(7+1)-2.
But from (y+2)8 = (7+2)6,, where 6 and #, are order types, follows
B = py.
294. XIV. Ordinal numbers

EXERCISES. 1. Prove that if, for the ordinal numbers a and 6 we have a-2=
== /HoD, WN -G == [Xe
Proof. The proof immediately follows from Corollary 3.
2. Prove that for ordinal numbers a and f the formulae af = fa and a?f? = fa?
are equivalent.
Proof. If af = Ba, then af? = a(aB)B = a(Ba)B = (af) (aB) = (Ba) (Ba) = B(aB)a =
= B(Ba)a = Ba’.
On the other hand, if aB > fa, then a?6? = a(aB)6 > a(Ba)6 = (aB) (af) > (Ba) (Pa)=
= fp(ap)a> B(fa)a= fa", therefore a?B? > Ba?.
Similarly, if Ba > af, then fa? > a26?. Thus a?6? = fa? implies af = fa.
3. Prove that for a natural number » and ordinal numbers a and f the formulae
ap = Ba and a’p” = "a" are equivalent 2).

11. Theorem on the division of ordinal numbers. We shall prove

THEOREM 1. If a,f,y are ordinal numbers such that y < af, then
we can represent number y uniquely in the form y = af,+a,, where a, < a,
By < p.
Proof. Let A and B be two well-ordered sets such that A= a,
B= f, and let us order the set P= AXB according to the principle of
last differences; thus we shall have P = af and, since y < af, y will be
the type of a certain segment Q of the well-ordered set P, different from P,
7. €. corresponding to a certain element (a,, b,) of P. Thus the set Q is
the set of all systems (a, b) where a ¢ A, b « B, (a, b) S (a, b,) in the set P.
Let us denote by M the set of all systems (a, b) of the set P for which
b <b, in the set Band by N — the set of all systems (a, b) of the set P for
which a<a, in the set A and b= dy.
Further, let us denote by a, the type of the segment of the set A
formed by the element a, (7. e. the set of all elements of the set A
that precede a,), and by 6, — the type of the segment of the set B formed
by the element b,; thus we shall have a, < a and pf, < f.
From the definition of the product of order types it immediately
follows that the type of the subset M of the set P will be af,, and since
the type of the subset N of the set P is of course a,, and Y= M+N,
where each element of the set M precedes in the set P each element of
the set N, therefore, according to the definition of the sum of types, we
obtain y = af,+a,. Thus we have proved that number y can be repre-
sented as y = af,+a, where a, <a, pf, <a.
Suppose now that number y can also be represented in the form
y = ap,+a,, where a,*< a, B,< 6. Therefore we should have
(Gein s) ap, +a, = af.+a,.

1) More generally, m and n being natural numbers, and a and f ordinal numbers,
the formulae af = Ba and a™p” = f"a™ are equivalent.
 § 11. Theorem on the division 295

If we had $6, = £., then, by Corollary 1 of XIV.4, we should have

a, = a, and our decompositions would be identical. Therefore let us as-
sume that 6, ~ f,, e.g. that 6, > B,. By Theorem 4 of XIV.4 we have
B, = Bo+t where t > 0, whence af, = a(f,+17) = af,+art; therefore, by
(28): aB.+at+a, = af,+a,, which, by Corollary 1 to theorem 5, gives

(11.2) at +a, = a.

But +21; by Theorem 5 of XIV.10 we thus have ar >a-l=a,

whence by (11.2), we obtain a, > a, which is contrary to the assumption
regarding number a,. Therefore we must have f, = 6, and a, = a,. Thus
we have proved Theorem 1.
It will be observed that Theorem 1 is not necessarily true for order
types if the inequality for them is defined after Fraissé (see XI.11)
Indeed, we obviously have 7 < 2-7, but if we had 7 = 28,+a, where
a, < 2, bj <7 we could not have a, = 1 because the type 7 has no last
element; we should have a, = 0 and therefore 7 = 28,, which is impos-
sible because the type 7 is dense while the type 2/, is not.
EXERCISE. Prove that number w? is a prime component.
Proof. Let & denote an ordinal number <w?. From Theorem 1 it follows that
~— = wk-+1 where k and J are finite integers >0. Hence + w? = wk+ 14 w? = wk+ ao? =
= w(k+ o) = wo. Thus + w?= w* for — < w*, whence, according to Theorem 1 of
XIV.6, we conclude that number mw? is a prime component.

THEOREM 2. Jf a and £ are ordinal numbers and a> 0, then number p

can be represented only in one way as

(11.3) P= Takao Meowheres | 10 fora).

Proof. Since a> 0, i.e. a>1, we have, by Theorem 2 of XIV.10

ap >1-fP= 6. If aB =f we can assume, in formula (11.3), that = 8,
a= 0:
If af > B we can apply Theorem 1, which gives 6 = af,+a, where
a <a, 6, < B; thus it suffices to write ¢= 6,, o = a, in order to obtain
formula (11.3).
The proof that numbers € and o satisfying formula (11.3), are well
defined by numbers a and f is similar to the proof of unicity of de-
composition in Theorem 1.
Thus Theorem 2 may be regarded as proved.
In particular, for a = 2, it follows from Theorem 2 that every ordinal
number can be represented either as 2é or as 2é+1 (7. e. it is either even
or odd). This is not the case with order types, where for instance the
type 7 is neither even nor odd, while the type m* is both even and
odd since o* = 2@* = 2o*+1.
296 XIV. Ordinal numbers

It is remarkable that the number (w+1)-2 is odd [=2(m-2)+1].

But not every ordinal number has the form é-2 or the form
&-2+1, e.g. w is of neither of these forms. For if we had w= &-2
or w= ¢é-2+1, we should have >, whence §-2>o@-2>o and
&-2+1>o. More generally, none of the numbers w+k where 0 <k<o
is of the form &-2 or of the form ¢-2+1 because w+k<o@-2: But
each number w-2-+k where 0 < k < w is both of the form €-2 and of the
form £-2+1 because o-2+k=(w+k)-2= ((o+k)—1)-2+41.
For a=o it follows from Theorem 2 that every ordinal number
can be represented as 6 = w+ o where oe is a finite number (> 0); if
o > 0, then f is of course a number of the 1-st kind (6 = wé+(e—1)+1);
it follows that every ordinal number of the 2-nd kind is of the
form wé. Obviously the converse is also true, because if an ordinal num-
ber f is of the 1-st kind, then 6 = a+1, and since a, as an ordinal number,
must be of the form w&+o where o0< a, B= w&+(e+1) where o+1
is a finite number. Thus if we had 6 = wé, at the same time, it would
be contrary to Theorem 2, from which it follows, in particular, that every
ordinal number can be represented uniquely as wé+ 0, where 90 <o.
Therefore a number of the 1-st kind cannot be of the form wé; every
number of this form is of the 2-nd kind. Thus we obtain
COROLLARY 1. In order that an ordinal number be of the 2-nd kind i
is necessary and sufficient that it have w as a left-hand divisor.

EXERCISES. 1. Prove that in order that an ordinal number f be of the

2-nd kind, it is necessary and sufficient that we have nf = for every natural
number n.
Proof. The necessity of the condition follows from corollary 1 and the observa-
tion that, in view of nw = w, we have, for every ordinal number &, n(wé) = (nw)é =
= w&. On the other hand, if B =a+1, then 268 = 2(a+1) = 24+2>5>a+2>a4+1=
= B, therefore 26 > 6. Thus if 6 is a number of the 1-st kind, then 28 > B, whence it
follows that the condition is sufficient.
2. Determine all natural left-hand divisors of the ordinal number of the 1-st
kind 6 = w&+k where k is a natural number.

Solution. The natural left-hand divisors of the number of the 1-st kind B=
= w&+k are divisors of number k and only these. Indeed,} if I|k, k =1k,, then B=
= wE+k = I(wE+k,) and 1 is a left-hand divisor of number f#. On the other hand, if
the natural number J is a left-hand divisor of number $ then 6 = w&+ k = I(wé,+ k,) =
= wé,+ lk, and it follows from Theorem 2 (for a = w) that = & and k = lk,, which
proves that J is a divisor of number k.
3. Prove that if a is an ordinal number >0, then for ordinal numbers f of the
1-st kind we have (a+ 1)6 > af and for ordinal numbers f of the 2-nd kind we have
(a+ 1)B = af.
Proof. If 6 is a number of the 1-st kind, 6 = y+ 1, then (a+ 1)6 = (a+ 1)(y+1)=
=(a+|l)ytat+le>ay+a+1=al(y+1)+1=a8+1> af,
 § 11. Theorem on the division 29%

If 6 is a number of the 2-nd kind, then we have 6B = wé. But since 2w = w and
a > 0 we have aw< (a+ 1)0< (a+ a)o = (a: 2)m = a: (2m) = aw, whence (a+ 1)o=
=ao and (a+ 1)6 = (a+ 1) wé = awE = af.

4. Prove that the product of two ordinal numbers >0 is a number of the 1-st
kind if and only if both factors are of the 1-st kind.
Proof. On one hand we have (a+ 1)(6+ 1) = (a+1)6+ a+ 1; on the other hand,
if aisa number of the 2-nd kind, then a = wé, whence, for B > 0, a8 = (w&)B = w(&f) is
a number of the 2-nd kind, and if 6 is a number of the 2-nd kind, then it follows from
the definition of the product of order types that (for a > 0 and f > 0) the type af
has no last element.
5. Prove that if a > 0 is a number of the 2-nd kind and 6 a number of the 1-st
kind, then (a+ 1)6 =af+1.
Proof. It 6 is a number of the 1-st kind, then we have 8 = wé+n where n is
a natural number, and according to exercise 3 we have (a+ 1)w& =awé. If a > 0 is
a number of the 2-nd kind, then we have a > and according to exercise 4 in § 4 we
have 1+a=a, whence (a+ 1)n = (a+1)4+ (a+ 1)+...+ (a+ 1)+ (a+ 1) = atoat..+
t+ta+(a+1)=a-n+1. Thus we have (a+ 1)6 = (a+ 1) (m+n) = (a+ 1) w€4+ (a4 1l)n=.
=awét+a-n+1=—a(woé+n)+1=af+1.
It will also be observed that if 6 is a number of the 2-nd kind, then it follows from
Exercise 4 that the equality (a+ 1)8 = aB+1 cannot hold for any ordinal number a,
while if a and £8 are numbers of the 1-st kind, then the above equality may be either
true or false, because, for instance, it is true for a= 6 = 1, as well as for B=ow+1
with an arbitrary a (since, in virtue of exercise 3, (a+ 1)@ = aw, whence (a+ 1)(@+ 1) =
= (a+ 1l)o+a+1=ao+a+1=a(w+1)+1), but it is false fora=—1, B = 2.
6. Prove that if 4 is an ordinal number of the 2-nd kind and & and n are natural
numbers, then (An)* =n.
Proof. If 4 is an ordinal number of the 2-nd kind, it follows from Corollary 1
that A= wé&, where é is an ordinal number >0. Thus, if for natural &k and n we
have (An)* = 2'n (which is true for k = 1), then (An)**? = (An)*An = A*ndn = A*nwén =
= win = i'*'n (since nw = w). Thus the proof follows by induction.

Basing ourselves upon Theorem 2 we shall now prove

THEOREM 3. The formula
lima: —/
E<o

implies, for every ordinal number y > 0, the formula

(11.4) lim(yas) = yA.

E<o

Brook, at
A= lim Qe ,
é<p

then we have 41> oa; for <q, whence for y>0; yA> ya; for <q.
On the other hand, suppose that wu > yo; for E<g. By Theorem 2 we
have u = y6-+e where 0 < y, whence y(6¢+1)= y6ty>yte=u> ya.
298 XIV. Ordinal numbers

for <q; therefore y(€+1) > yag for E<g, which gives ¢+1 > a; for
— <q. In view of
lima=i,
E<o

€+1 > A; since A, as the limit of a transfinite sequence of ordinal numbers,

is a number of the 2-nd kind, we have €¢+1->4A, which gives ¢>4,
whence w= y6+o> yA+o> yd. Thus yd is the least ordinal number
>yag for <q, which proves the truth of formula (11.4), q. e. d.
In connection with Theorem 3, however, it is to be observed that,
in view of w? >, we have
lim(rnw) < (umn): @.
n<w n<o

12. Divisors of ordinal numbers. We shall prove

THEOREM 1. Every ordinal number >0 has a finite number of different
right-hand divisors.
Proof. Suppose that a certain ordinal number a> 0 has infinitely
many different right-hand divisors. The set of those divisors would be
well-ordered according to their magnitudes, and there would exist an
increasing infinite sequence 0, < 6, < 0; <..., each term of which would
be a right-hand divisor of number a.
For every 6, let us denote by y, the smallest of the left-hand divisors
of number a that correspond to the right-hand divisor 6,; thus a = y,6,.
Now let k and I be two natural numbers, k <l. We have 6; < 6;. If we
had yx <y,, then by Theorem 2 of XIV.10, we should have y;,6, < y6;,
and in view of 6, < 6; and by Theorem 1 of XIV.10: yd, < y6,, there-
fore a = y,0x < y:0; = a, which is impossible. Thus », > for k <l,
whence y, > y. > y3 >..., Which is impossible since there exists no de-
creasing infinite sequence of ordinal numbers. The ordinal number a > 0
cannot have infinitely many different right-hand divisors, which proves
the truth of Theorem 1.
THEOREM 2. Ordinal numbers of the 1-st kind and they alone (among
ordinal numbers) have a finite number of left-hand divisors.
Proof. Suppose that a certain number of the 1-st kind a= 6+1,
has infinitely many left-hand divisors. As in the proof of Theorem 1,
we conclude that there exists an increasing infinite sequence y, < y. <...,
each term of which is a left-hand divisor of the number a. Therefore for
every natural number n there exists an ordinal number 6, (as follows
from Corollary 7 of XIV.10, Theorem 2, only one) such that a= y,6,.
Now let k and J be two natural numbers, k < 1. Therefore we have », < y;.
If we had 6, < 6;, then the inequality y,; < y, and Theorem 2 of XIV.10
give y,0x <y;0, and Theorem 2 of XIV.10 gives y)6, < y;6;, whence a=
 \ § 12. Divisors 299

= yx0x~ < y16,, Which is impossible. Therefore 6, > 6,, and from Exercise 4
of XIV.4 it follows that numbers 6, and 6, must be of the 1-st kind (the
number a being of the first kind). Therefore 6,=—¢,+1, 6.=¢4+1.
Hence ypdg = yelCxt+1) = vebetye and y6;= wility. Since a= yd, =
= 10, we thus have y,xl,+y,= yidity. If we had ype, <j), then,
since yz< yw, we should have y,7;+y,% < yi; +71, which is impossible.
Therefore. we have y,¢, > y¢;; if we had ¢,<¢,, then, since »,< ,
we should have y,¢, <+,¢;, which is impossible. Therefore we have
én C7, whence op C. +1 > ¢,+1 = 0), v. €. 0, >0;. Therefore 6, > 0,
for k<l, whence 6, >06,>..., which is impossible. Thus an ordinal
number a of the first kind cannot have infinitely many different left-
hand divisors. But from Exercise 1 of XIV.10 it follows that every ordinal
number of the 2-nd kind has infinitely many different left-hand divisors.
Theorem 2 is thus proved.
Here is another proof of the theorem stating that every ordinal
number of the first kind has a finite number of left-hand divisors; this
proof is based on Theorem 1.
Let a denote a number of the first kind. As we know, to each left-
-hand divisor of the number a corresponds a well-defined complementary
right-hand divisor. If the same complementary right-hand divisor f
corresponded to two different left-hand divisors, y and y, > y of the num-
ber a, then we should have a= yd = y,6, and, according to Exercise 4
of XIV.10, 6 would be a number of the first kind; in view of y < y,, and
by Corollary 2 of Theorem 2 of XIV.10, we should have yd < y,6, which
is impossible. Thus to different left-hand divisors of the number a cor-
respond different right-hand divisors of that number, and since the
number of the latter, by Theorem 1, is finite, the number of left-hand
divisors of the number a is also finite, q. e. d.
EXERCISES. 1. Prove that every number of the first kind gives a finite number
of different decompositions into a product of two factors, and also into a product of
three factors.
Proof. The fact that every number of the first kind gives a finite number of dif-
ferent decompositions into a product of two factors follows from the observation that,
in every such decomposition, a = yd, the number y is a left-hand divisor and the num-
ber 6 a right-hand divisor of the number a, and by Theorems | and 2 a number of the
first kind has a finite number both of left-hand and of right-hand divisors. In every
decomposition a = fy6d of the number a into a product of three factors, the number £
is a left-hand divisor of the number a, the number 6 — a right-hand divisor of the num-
ber a, and the number y — a left-hand divisor of the right-hand divisor yd of the num-
ber a, which, according to Exercise 4 of XIV.10, is also a number of the first kind.
Hence, and by Theorems 1 and 2, it follows that the number of all decompositions of
@ number of the first kind into a product of three factors is finite.
2. Prove that, for every natural number m, every ordinal number of the first
kind gives a finite number of different decompositions into a product of m factors.
300 : XIV. Ordinal numbers

3. Give an example showing that a right-hand divisor of each of two given ordinal
numbers need not be a right-hand divisor of their sum.
Solution. The number w-w+o-o = @*-2 is not divisible on the right by o,
since if w?-2 = éw, we should have < w-w (because w?-2 < w?-w) and, by Theo-
rem 1 of XIV.11 we should have € = wk-+l, where k < wm and 1 < w, whence w =
= (wok+1)o = w < w?-2, which is impossible. But a left-hand divisor of each of two
given ordinal numbers is a left-hand divisor of their sum, since 6a+ 66 = 6(a+ ).

Common divisors of ordinal numbers

From Theorem 1 it immediately follows that there is a finite number

of common right-hand divisors of two given ordinal numbers: thus
among them there always exists the greatest. But there are always in-
finitely many common left-hand divisors of two given ordinal numbers
(of the 2-nd kind), since at any rate all natural numbers are such divisors;
still there always exists among them the greatest, which (just as with
finite numbers) is divisible on the left by any other common divisor of
the numbers in question. The proof of this property, just as with finite
numbers, is based on the principle of successive divisions, resulting from
Theorem 2 of XIV.11. For if a and £ are ordinal numbers, p > a> 0,
then, by Theorem 2 of XIV.11, we have B= a&+o where 0 <0 <a,
therefore 90 = 6—aé. From the formula 6= aé+e it follows that each
common left-hand divisor of numbers a and o is a left-hand divisor of
number f/, and from formula 90 = 6/—aé and Exercise 9, of XIV.4, it
follows that each common left-hand divisor of numbers a and f is a left-
hand divisor of number oe. Thus the pair of numbers f and a has the same
common left-hand divisors as the pairs of numbers a and o. If o=0,
then 6 = ag and a is the greatest common left-hand divisor of numbers a
and 6, but even then the pair 6 and a may be said to have the same com-
mon left-hand divisors as the pair a and oe. Let us denote, generally, by
Z(6,a) the set of all common left-hand divisors of numbers 6 and a.
Thus we have Z(f, a) = Z(a, @). If o> 0, then we have a= 0é,+0,, where
0<o,< oe and we again obtain Z(a, 0) = Z(e@, o,). If o, #0, then we have
© = 0162+ @ where 0 < ge, < 0, and we obtain Z(0, 0,) = Z(@,, e:). Proceeding
in this way we get to the equality @,-1= Onén41+ On41, Where 0n41 = 0, since
we have a>e>o,>0,>..., and a decreasing sequence of ordinal numbers
cannot be infinite. But then 0,41 = @né,41 Where §&41 40 (since @n41> On),
whence it easily follows that the set Z(0,_1, 0,) is the set of all left-hand
divisors of number.o,. Since 7(8,¢)— 7 (a, ov= Zi0, 0))— ZiOy. 02) —
= Z(0n-1, Qn), We conclude that the set Z7(f, a) is the set of all left-hand
divisors of number e,. It immediately follows hence that the greatest
common left-hand divisor of numbers f and a is oe, and that @, is divisible
on the left by each common left-hand divisor of numbers £ and a, q. e. d.
 § 12. Divisors 301

Common multiples of ordinal numbers

For two given ordinal numbers #0 an ordinal number 40 divisible

on the right by each of the two numbers does not necessarily exist.
E. g. There is no ordinal number 0 divisible on the right by both 2
and mw. Indeed, if such an ordinal number y existed, we should have
y=a-2=6-o, and a<a-2= Bw; by Theorem 1 of XIV.11 we should
have a= $k+8, where k<q@ and a,<#, whence a-2—a+a= pk+
+ 6,+ 6kK+6, < bk +6+ pk+6 = B(2k+2) < Bo, which is impossible (see
also Exercise 1). There exist, however, ordinal types divisible on the
right by both 2 and , since we have, for instance, 7-2 = y-@, (1+7)-2=
= (1+ 7): o, (14+4)-2=(1+A)-o. It would be more difficult to prove
(see XV.4, Exercise 5) that there exist ordinal numbers a and f which
are not common right-hand divisors of any order type.
Now, for every two positive ordinal numbers a and fp there exists an
ordinal number y ~ 0 for which a and 6 are left-hand divisors. Indeed,
let 0<a<f. By Theorem 1 of XIV.11 we have 6 = a&+ a, where a, < a.
Hence aé <8 < a€+a= a(é+1) and am < fw <a(Eé+1)m. But, accord-
ing to Exercise 3, XIV.10, éw = (é+1),. whence aém = a(E+1)w
and fo= ass, which proves that numbers a and £ are left-hand
divisors of number fw. We have also proved here that for every ordinal
number f > 0 every positive ordinal number </ is a left-hand divisor
of the number fo.
We can also prove, more generally, that for every set.Z of positive
ordinal numbers there exists an ordinal number y > 0 such that each
number of the set Z is a left-hand divisor of y.
Indeed, let Z denote a given set of ordinal numbers. From Corol-
lary 2 of XIV.3 it follows that there exists an ordinal number / greater
than any number of the set Z. The number fw has been proved to be
divisible on the left by each positive ordinal number </, 7. e. by each
number of the set Z, q.e. d.
If w denotes any ordinal number 0 divisible on the left by the
numbers a and f, and v denotes the least of those numbers — let us de-
note it by [a, B] — then w= vg. Indeed, by Theorem 2 of XIV.11 we
have «= vé+y7,, where »,< y», whence »,= w—vé and, by Theorem 1
of XIV.6, we conclude that a and 7 are left-hand divisors of the number ?,.
In viwe of the definition of the number y, we conclude that », = 0, there-
fore = VE,.G.e.-d..
It immediately follows that if an ordinal number y is divisible on
the left by the ordinal numbers a and £, then it is also divisible on the
left by the number [a, 8]. Hence, in particular, an ordinal number that
is divisible on the left by 2 and 3 is divisible on the left by 6.
302 XIV. Ordinal numbers

EXERCISES. 1. Prove that an ordinal number >0, divisible on the right by

w cannot be divisible on the right by any number of the first kind >1.
Proof. Suppose that 0 < a= Bw = yd where 6 is a number of the first kind >1.
Since 6 is a number of the first kind, we have 6 = w&+n where n is a natural number.
Hence yd = y(wé+n) = yw&+ yn, therefore ywE < yi = fw and, by Theorem 1 of
XIV.11, we have ywé = Bk,+ B,, where k, < w and f, < f. Similarly, in view of 6 > 1,
we have y < yO= fw, therefore y = 6k,+ 8, where k, < w and fp, < B. Thus yd =
= poé+ yn = Bk,+ Bi + (Bho+ B.)n < Blk,+ 1+ (k,.+ 1)n] < Bo, which is impossible.
2. Prove that w+ 6 is the least ordinal number >0 divisible on the right by w + 2
and by w+ 3.
Proof. The fact that number w+ 6 is divisible on the right both by w+ 2 and
by +3 follows from the equality +6 = 3(m+ 2) = 2(@+ 3). Since every ordinal
number >0, divisible on the right by +2 and by w+ 3, must be >w+3, therefore
it remains to prove that none of the numbers w+.3, o+4 and w+ 5 is divisible on the
right both by w+ 2 and w+ 3. For this purpose it suffices to show that the number
@+3 is not divisible on the right by w+ 2, while numbers w+ 4 and w-+5 are not
divisible on the right by w+ 3. Now, if we had w+ 3 = a(w+ 2), we should have w+ 3 =
=aw+a-2, and the number a would have to be finite, whence aw = w and #+ 3 =
= m+a-2, which gives 3=a-2, which is impossible. Similarly, if we had o+4=
= a(wm-+ 3), we should have w+ 4= aw+a-3 and we should obtain, as above, 4 = a-3,
which is impossible. We prove in a similar way that we cannot have w+ 5 = a(w-+ 3).
3. Find the least ordinal number >0, divisible on the left by w+2 and by w+ 3.
Solution. The sought number is w?. For, on one hand, we have w? = (w+ 2) =
= (m+ 3)qo. On the other hand, every ordinal number <q? is of the form wk--1 where
k and J are finite numbers >0. If we had wk+1= (m+ 2)a = (w+ 3), the numbers a
and 6 would obviously have to be finite and we should have (m+ 2)a = wa+2 and
(w+ 3)8 = wf +3 therefore ok +1 = wa+2 = wf+3, which is impossible.
4. Prove that an ordinal number, divisible on the left by w+ 2 and by +3 is
divisible on the left by o?.
Proof. The proof follows at once from Exercise 3 since we have proved above
that an ordinal number divisible on the left by the number a and f is divisible on the
left by the least ordinal number that is divisible on the left by both a and f.
5. Prove that if, for ordinal numbers a and 6 we have a-2=— 8-3, then a=
= (a—B)-3, B = (a—8)-2.
Proof. The theorem is of course true if a = 0 (since then also 6 = 0. Therefore
let us assume that numbers a and 6 are #0. Since a-2 = 8-3, we have a-2 > p-2
whence, by Theorem 2 of XIV.9, a > f, therefore y= a—fB > 0 anda=f-+y. Hence,
in view of a+a=— f+ 8+ 8, we obtain B+ y+6+y = 84+ 6+ 6 which, according to
Corollary 1 of XIV.4, gives y+ 6+y= 6+. For B<y, we should have y+ fBty<
<y+ , whence B+ y < B, which, in view of y > 0, is impossible. Thus 6 > y, there-
fore B = y+ 6 where 6 > 0, therefore, in view of y+ B+ y = B+ B, we have y+ y+ 6+ i
=y+o+y-+0, whence y+6+y=6+y+6.
For 6#y, ¢.g.6 < y, we should have 6+y< y+y, therefore y+ 6+ y = 6+ y+
+d6<y+y+0, which gives y < 6, contrary to the assumption. We prove in a similar
way that 6 > y is impossible. Thus 6 = y, therefore B = y+ d= y-2,a=fp+y=y:3.
where y=a—f, q.e. d.
6. Prove that an ordinal number divisible on the right by 2 and by 3 is divisible
on the right by 6.
 § 12. Divisors 303

Proof. If y=a-2= 6-3, then by Exercise 5, we have y = (a—f)-6, whence it

follows that number y is divisible on the right by 6.
7. Prove that if for ordinal numbers a and 6 we have a+ f+a=f+a-+ , then
C08
Proof. If a+fta=f+a+f, let p=a+f+a, p=a+f. We have 9-2=
=(a+6+a)+ (fb+a+f) = (a+ f)-3=y-3 and in view of Exercise 5 there exists
an ordinal number @ such that p= %-3, y=%-2, therefore a+ f+a=%-3, a+ B=
= 0-2, whence }-2+a=—0-24+4%, thus a=#@, whence $+8=%+9%, thus fh=%=a
and a=, q.e.d.
8. Find all pairs of ordinal numbers a and f# such that 2a = 3f.
Answer. a= 3¢, 6B = 2¢, where € is an arbitrary ordinal number >0. Indeed we
have 2a = 2-3¢ = 3-2¢ = 38. On the other hand if a and f are ordinal numbers, we
have a= wé+7, B = wyn+s, where € and 7 are ordinal numbers >0 and r and s are
finite ordinal numbers > 0. If 2a = 38, we have 2(w&+ 7) = 3(@yn+ 8), whence wé+
+ 2r = wn+ 3s; thus = 7 and 2r = 3s and there exists a finite ordinal number k such
that r = 3k, s = 2k, whence for € = w&+k we have a= w&+ 3k = 30, B = w&4+ 2k=
SHE5 Olas. Ole
Remark. Later on we shall prove, more generally (see XIV.21), that if an ordinal
number is divisible on the right by natural numbers k and J, then it is also divisible
on the right by their smallest common multiple. I do not known if it is true for ordinal
types. On the other hand it may be proved that if an ordinal type is divisible on the
left by natural numbers k and l, then it is also divisible on the left by their smallest
multiple.
9. Give an example of two ordinal numbers a and f such that if 6 denotes their
greatest common left divisor and y» the smallest ordinal number >0 divisible on the
left by a and by £, then af + ov, aB £76, Ba $ dv, Ba 4 vO.
Answer. a=o+2, 6=a@+3. We have here d=1, v=o, aB=@?+o-3+2,
pa = w+ w-2+3.

13. Prime factors of ordinal numbers. Every ordinal number >1 that
is not the product of two ordinal numbers smaller than that number
is called a prime factor or a number indecomposable with respect to multi-
plication.
If a> 1, then there exist right-hand divisors of the number a that
are greater than 1 (e.g. the number a itself), and among them there
exists the smallest, z,. It is easy to prove that z, is a prime factor. Other-
wise, in view of the definition of prime factors and of x, > 1, the number z,
would be the product of two ordinal numbers smaller than 2,: 2, = py
where uw <a, and »<2,, and since u<a,, we should have y>1 and
the number » would be a right-hand divisor of the number a, greater
than 1 and smaller than z,, which is contrary to the definition of num-
ber. z,.
Since z, is a right-hand divisor of number a, there exists an ordinal
number a, such that a = a,. If a4,=1, then we have a= 2, and the
number a is a prime factor. And if a, > 1, then, in view of z, > 1 and by
Theorem 1 of XIV.10, we obtain a > a,. Number a, can be dealt with in the
304 XIV. Ordinal numbers

same way as number a, which gives a, = ag7,, where z, is a prime factor,

and a, >a,. If a, >1, then we shall obtain next a, = a,7, where zz 1s
a prime factor and a, > a;. Since a decreasing sequence of ordinal num-
bers a > a, >a, >... cannot be infinite, we obtain in this way an expan-
sion of the number a into a finite product of prime factors. Thus we have
THEOREM 1. Every ordinal number >1 is the product of a finite number
of prime factors.
A decomposition of ordinal numbers into prime factors, however,
is not unique+) in general; we have for instance (o+1)o=@-o, and
if can easily be proved that numbers w and w+1 are prime factors.
Finite prime factors coincide of course with prime numbers. The
smallest transfinite prime factor is obviously number @ (which follows
from the observation that the product of two numbers smaller than w
is finite). Next comes number w+1. For if number w+ 1 were not a prime
factor, we should have w+1= wy, where uw<w+1 and y< w+1, there-
fore uw<q@ and »y <o. If wu< , and therefore wu= k where k is a natural
number, then, in view of »<w, we should have w= ky <ko= a,
therefore o+1 = wy <@, which is impossible. Thus we must have uw = o
If »=1, then we should have o+1= uwy=o-1, which is impossible;
if »>1, and thus vy > 2, we should have o+1= wv >o-2, which is
also impossible.
Thus there exist two prime factors which are successive transfinite
ordinal numbers (just as 2 and 3 are successive finite: prime numbers);
we shall prove later on (in XIV.22) that there are infinitely many pairs
of successive ordinal numbers that are prime factors. Numbers w+k
for k = 2,3,... are not prime factors because w+k = k(w+1); number
m-2 is likewise not a prime factor, nor are the numbers o-k+1= (o+l1)k
for b==2',3,...3- = 0, 1,2)... Lhe: numbero*--1 is the: next.) prime
factor after w+1.
In order to prove that w?+ 1 is a prime factor we shall prove, more
generally, that 7f @ 7s a prueme component then o +1 is a prime factor. Indeed,
if we had 0+1= my, where u<oe+1 and »<o0+1, then since 90+1
is a number of the 1-st kind, numbers uw and » would also be of the 1-st
kind; therefore w= &,+1, v= +1 where §,+1<0+1 and &4+1<
<o+i, thus é,<@ and é,<g, and also e+1=(&41)(&41) = (&+
+1)é,+& +1, whence 0 = (6,+1)6+¢,, which gives &, = 0 since &, < 0
and @ is a prime component. We should thus have o = &, which is impos-
sible because & < 0. Since m? is a prime component (see XIV.9, Exercise 15),
om? +1 is a prime factor.

1) A. Schoenflies in his book [1], p. 161, gives a theorem on the unicity of de-
composition, which is probably an ovérlooked mistake. Cf. F. Sieezka [1], p. 172.
 Sis Prime factors 305

We shall concern ourselves with the determination of all prime

factors in § 22. A prime factor (only a transfinite one of course) may
have (but does not always have) a left-hand divisor different from 1
and from that factor itself, e.g. number w= 2 (but not w+1). But
in order that an ordinal number >1 be a prime factor it is necessary and
sufficient that it have no right-hand divisor different from 1 and from
that number itself. Indeed, if a= uy >1 where » ~1 and » £a, i.e,
l<y<a, then a=ypr>w and a, as a product of two numbers
smaller than a, is not a prime factor; thus the condition is necessary.
On the other hand, if a>1 is not a prime factor, then a= ur where
fa<a and » <a, t.e. also y #1 since if y= 1‘then a= yp, contrary to
“a@<a. Thus number a has a right-hand divisor different from 1 and
from a itself, which proves that our condition is sufficient.

14. Certain properties of prime components. We shall prove

THEOREM 10. Jf 0 is a prime component >1, then, for every ordinal
number a>0, ao is a prime component.
Proof. Suppose that ao is not a prime component, 7. e. that ao =
= fB+y where f <ag and y<ag. By Theorem 1 of XIV.11 we have
p= 08, Cue y= e641 where 6, < 01, 45<.0,°j65 @ and » =a. Thus
Bty<aé,+a+a,+a= a(6,+1+6 +1). But in view of §,< 0, &<e
and by Theorem 1 of XIV.6 we obtain successively 1+o=o, 6,4. 1+
-o=¢,+0='0, 1+6,+lto—1I+e=o0 4+14+6+1+¢e=—£+0=
=o. If we had 6,4+1+44+1+2 0, we should have §é,+1+4+1l+o0>
>o+o> oe, which is impossible. Thus ¢,+1+é+1< o and therefore,
by ‘Theorem 1, of XIV.10, 6-+y <a(é,+1+8 +1) < ae, contrary to
ago=f-+y. Thus Theorem 1 is proved.
COROLLARY. If a is an ordinal number +0, then the smallest prime
component >a is the number aw.
Proof. Number w is known to be a prime component (XIV.6):
thus if a > 0, then, by Theorem 1, number aw will be a prime component.
In order to prove our corollary it remains only to show that between a
and aw there is no prime component.
Suppose that 6 is a number between a and aw, v. e. that a< 6 < aw.
By Theorem 1 of XIV.11 we have 8 = au-+y, where w<o, vy <a. If
y > 0, then the formula for 6 proves that 6 is the sum of two ordinal
numbers smaller than /, and thus it is not a prime component. If » = 0,
then owing to the fact that, in view of u < w, wis a finite number and 40
(since 6B>a>0), we can write ~= a(u—1)+a, whence (in view of
0 <a < #) it follows that number /, as the sum of two ordinal numbers
smaller than £, is not a prime component.
Cardinal and ordinal numbers 20
306 XIV. Ordinal numbers

Thus our corollary is proved. In particular, it follows from it that

if 9 is a given prime component, then the next prime component after
it is ew. Together with Theorem 3. of XIV.6 this gives us a means of
determining successive prime components. Thus the first w non finite
prime components will be the numbers w, aw, www, ..., v. e. the numbers
wo, @, o,...3 the least ordinal number that is greater than any term of
that sequence will be the next successive prime component; that number
multiplied on the right by w will come next, ete.
THEOREM 2. Hvery prime component is divisible on the left by every
positive ordinal number smaller than that component, the quotient being
again a prime component.
Proof. Let @ denote a given prime component and let a denote an
ordinal number such that 0 <a< oe. By Theorem 1 of XIV.11 we have
= a+ 0,, where 0 < 0, <a, therefore oe, < oe. But, as we know from
XIV.6, a prime component has no positive remainder different from
itself. Thus we must have 0, = 0, whence @ = af, which proves that a
is a left-hand divisor of number o.
Now we shall prove that the quotient § is a prime component. If
we had
(14.1) c= py Mwhere- Wire and? “pees,

then, by Theorem 1, of XIV.10, we should haveau < a& and ay < aé;
therefore, in view of 90= ag, this would give au<o and ay <o, and
since, by (14.1), oe= a&=a(u+yv)=au+ay, the prime component o
would be the sum of two ordinal numbers smaller than 0, which is impos-
sible. Thus number & is not the sum of two ordinal numbers smaller than &,
z. €. it is @ prime component.
Thus Theorem 2 is proved.

15. Exponentiation of ordinal numbers. Let « and a be two given

ordinal numbers >0. Let us denote by Z(u, a) the set of all sequences
(finite or transfinite) of type a whose terms are ordinal numbers >0
and <p, the number of those terms which are different from 0 being
finite (>0).
Let a= {dg}ze<q and b= {bg}ecq be two different sequences belonging
to the set Z(u, a). The number of indices € for which 0 <&é<aanda; 40,
or be 40, is thus finite. Hence it immediately follows that the number
of those ordinal numbers § (0 <<a) for which a; ¢ Dz is also finite
(such indices exist, since we assume the sequences a and b to be
different). Thus among them there exists the greatest, ». Let a<b if
a,<b,, and b<a if a,>b,. We shall prove that the relation < or-
ders the set Z(u, a).
 § 15. Exponentiation 307

The relation < defined in the set Z(u, a) is of course asymmetrical

and connected; therefore it suffices to show that it is transitive. Let
a= {dghe<ay ) = {beheca, and ¢ = {Ce}z<q be the given sequences, belonging
to Z(u, a), such that a<b and b<c. Let » denote the greatest index
such that ag 4b, and 2 — the greatest index such that b, ~c¢,. In view
of a<b and of our agreement as to joining our sequences by the sign X,
we have a, < b,; similarly, in view of b<e we have b, < ¢,. Further we
distinguish three cases:
1) A= ». In this case we have a, < b, and b, < ¢,, which gives a, < ¢,.
On the other hand, in view of the definition of numbers » and A, we have
az = b; for v.<é< a and b= ¢ for A<é< a; therefore, since A=,
we have de = cg for y << €< a. It follows that » is the greatest index such
that a, ~¢,, and since a,<c¢,, therefore, in view of the definition of
the relation <, we have a<e.
2) A<y. Thus we have (in view of the definition of number A) b, = ¢,,
and since a, < b,, we have a,<c,. On the other hand, in view of the
definition of numbers » and 4 and in view of 4 <», we have a; = b; and
be = cg for » << € < a, whence a; = cg for v< é < a. Thus we have ase.
3) A> v. Thus we have (in view of the definition of number v) a, = b,,
and since b,<«, we have a,<«. On the other hand, in view of the
definition of numbers » and 4 and in view of »< A, we have az= b: = ¢¢
for 4< & <a, whence it again follows that a<e.
Thus we have proved that the relation < orders the set Z(u, a) of
sequences (according to the principle of last differences).

EXAMPLES. For u = a= 2 the set Z(u, a) obviously consists of four two-term

sequences which are ordered thus:

(0,0) (1,0) = (0,1) (1, 1).

For «= 2, a=3 we obtain eight three-term sequences: .

(0,
0, 0) 3 (1; 0, 0) S (0, 1, 0) S (1, 1, 0) = (0,0, 1) = (1,0, 1) $(0, 1, 1)
S(1, 1, 1).
For 4 = 3, a= 2 we obtain nine two-term sequences:

(0, 0) S (1, 0) S (2, 0) S (0, 1) S (1, 1) S (2, 1) S (0, 2) S (1, 2) S (2, 2).

It can easily be seen that if the numbers uw and a are finite, then
the ordered set Z(u,a) will be of type wm.
For any positive ordinal numbers uw and a, let us agree to denote
the type of the set Z(u, a) (ordered by the relation <) by the symbol p..
Thus, for any positive ordinal numbers wu and a, u* will be a well-defined.
(by ~ and a) order type.
It can easily be seen that w'= mw for every ordinal number yu > 0.
Now we shall prove that for any positive ordinal numbers yu, a and f we
20*
308 XIV. Ordinal numbers

have the formula

(15.1) FD ei lay 1

Let c= {Cg}sca+g denote any sequence belonging to the set Z(u, a+ f).
The term c, of the sequence ¢ splits this sequence into two sequences:
the segment a= {dag}ecq and the remainder b = {Cg}acgca+g- Clearly the
sequence a belongs to the set Z(u,a), and the sequence b (regardless
of the symbols used as indices) — to the set Z(u, B). Conversely, if we
take an arbitrary sequence b, belonging to Z(u, f), followed by an arbi-
trary sequence a, belonging to Z(u, a), then we should obtain the se-
quence ¢=a+b, belonging to Z(u, a+).
Furthermore, it is obvious that if e=a+b and c’=a’'+bD’ are
decompositions of the sequences ¢ and c’ belonging to Z(u,a+), then
we shall have c<ce’ in the set Z(w, a+) if and only if b <b’ in the set
Z(u, B), or if we have simultaneously b) = b’ and a<a’ in the set Z(u, a).
Hence, and in view of the fact that the sets Z(u, a), Z(u, 6) and Z(u, a+6)
are of types wt, uw’ and yt+? respectively, it follows that (according to the
definition of the product of two order types) formula (15.1) holds, q. e. d.
In particular, for 6 = 1, formula (15.1) gives

(15.2) | path = uu
for any positiwe ordinal numbers yw and a.
Now we shall prove that for any positive ordinal numbers u and a
the type wt is an ordinal number.
Suppose that for a certain positive ordinal number w there exist
positive ordinal numbers a such that uw is not an ordinal number. Thus
among such numbers there exists the smallest, which will be denoted by ».
If number » were of the. 1-st kind, »y=a+1, then by (15.2), we
should have w’= utu. From the definition of number »v it follows, in
view of a<y, that the type y* is an ordinal number; thus the type wp’
would be the product of two ordinal numbers, 7. e. an ordinal number,
contrary to the definition of number v. Thus number » is of the 2-nd kind.
Since yw’ is not an ordinal number, the set Z(u, v) (which is of type py’)
is not well-ordered. There exists in it a non-empty subset Z,, having
no first element. Let a= {ag}cc, denote a sequence belonging to Z,.
Since only a finite (>0) number of terms a; of this sequence is different
from 0, there exists the greatest ordinal number 4 < y such that a, £0.
(If we always had ag= 0 for 0 <&<», then the sequence a would of
course be the first element of the set Z(u,v), 2. e. also of its subset Z,,
in contradiction to the assumption that Z, has no first element; thus
number / exists). From the manner of ordering the set Z(u, v) it follows
 § 15. Exponentiation 309

that for every sequence b= {bz\z-, belonging to Z, and preceding a

(provided such sets exist) we must have b:=0 for A<&<». But the
set of all sequences belonging to Z(u,v) in which all terms with indices
§ >A are equal to zero is obviously of type u’+1, ¢. e. it is well-ordered
in view of the definition of number vy and the observation that 2+1 < »
(since 2 < »y and v is a number of the 2-nd kind). Therefore the set of all
sequences belonging to Z, and preceding a would naturally also be well-
ordered (or empty). Hence the conclusion that the subset Z, contains
a first element, which contradicts the assumption.
Thus the assumption that there exist positive ordinal numbers a for
which y* is not an ordinal number leads to a contradiction. We have proved
that for any positive ordinal numbers u and a the type utis an ordinal number.
For «= 1, with any ordinal number a > 0, the set Z(u, a) consists
of course of only one sequence, whose terms are all equal to zero; thus
we have 1*= 1 for every positive ordinal number a.
Suppose now that w~>1 and let a and f be two positive ordinal
numbers, a < ~. The set Z(u, a) is obviously similar to the segment of
the set Z(u, 6) formed by a sequence whose term with the index a is equal
to 1 and whose other terms are all equal to zero. Hence it follows at once that
(15.3) gee ett Ie and arr ale 0
This immediately implies that if mw is an ordinal number greater
than 1 and if, for ordinal numbers a and /, the equality ut = yw? holds,
then a= fp.
Now let wu > 1 and let a denote an ordinal number of the 2-nd kind.
By (15.3) we shall have yw* > ui for <a. We shall prove that 7 is the
smallest ordinal number >yé for & < a.
Indeed, let y denote an ordinal number <y*. Thus number y is the
type of a certain segment of the set Z(u, a), e. g. of the segment formed
of all the elements of the set Z(u, a) that precede the element a = {dghecg
of this set. Since {ag}<<, 18 a sequence belonging to the set Z(w, a), there
exists the greatest ordinal number 4 <a such that a, ~0, and we shall
also have 4+1 <a because a is a number of the 2-nd kind. Thus a; = 0
for A+1<é&< a. Therefore, naturally, in all sequences of the set Z(u, a)
that precede the sequence a all the terms with indices >A4+1 are equal
to zero. Since the set of all the sequences of the set Z(w, a) in which the
terms with indices >A+1 are equal to zero is of type p’*1, we have
y <p’*+1, Thus no number y < yu* satisfies the inequality y > wi for § < a.
We have proved that if uw is an ordinal number >1 and if a is an
ordinal number of the 2-nd kind, then

ee WE:
E<a
310 XIV. Ordinal numbers

Now, for a given ordinal number uw > 1 and for every ordinal num-
ber a > 0, let f(a) = mw. Thus f(a) will be a function of an ordinal number,
satisfying the following three conditions: .
1) f)=4,
2) f(a+1) = f(a)-w for every ordinal number a,
3) If a is a number of the 2-nd kind, then

f(a) = limf(é) .
E<a
Now we shall prove that for every ordinal number y« > 1 there exists
only one function f(a) of an ordinal variable a satisfying conditions 1),
2) and 3).
Let us suppose that for a given ordinal number w > 1 there exist
two different functions, f,(a) and f.(a) of an ordinal number a satisfying
conditions 1), 2), 3). Since the functions /, and f, are different from each
other, there exist ordinal numbers é such that f,(é) #/.(é) and among
them there exists the smallest, a. Thus we have f,(a) ¢ f,(a), but

(15.4) f(é)=f.(€) for <a.

By 1) we must have a> 1. If a were a number of the 1-st kind,

we could take a= +1 where <a and, in view of property 2) of the
functions f, and f,, we should have

Ala)=AlE+D=AE)-e, hla) =f(F+1) = fl) -u,

whence, by (15.4), f,(a) = f,(a), which contradicts the definition of num-
ber a. Now, if a were a number of the 2-nd kind, then from property 3).
of the functions f, and f, and from formula (15.4) it would follow that
f(a) = f,(a), again in contradiction to the definition of number a. Thus
the assumption that f/, 4/7, leads to a contradiction.
Thus we have proved
THEOREM 1. For every ordinal number u > 1 there ewists one and only
one function f(a), defined for every ordinal number a>0 and_ satisfying
conditions 1), 2) and 3).
Let us denote function f(a) satisfying Theorem 1 by yu*. Moreover,
if we assume that 1° = 1 for every ordinal number a and p° = 1 for every
ordinal number » > 0, then the power u* will be defined for any ordinal
numbers u>0O and a>0, and for any ordinal numbers n»>0, a>0
and 6 >0 we shall have formula (15.1).
The power of an ordinal number greater than 0 with an ordinal
exponent has thus been defined constructively, by forming an ordered
set Z(u, a) whose order type is p*. ;
 § 15. Exponentiation 311

An apparently shorter method of introducing the power of an ordinal

number consists in applying a definition by transfinite induction. In
order to prove that for a given ordinal number u > 1 there exists a func-
tion f of an ordinal variable, satisfying conditions 1), 2) and 3), we reason
as follows:
Let us write f(1) = uw. Now let a denote a given ordinal ‘number >1
and suppose that we have already defined a function f(é) for every positive
ordinal number é < a. If a is of the 1-st kind, a= £+1, then we write
f(a) = f(&)u, and if a is a number of the 2-nd kind, then we write

f(a) = lim} (é)

é<a
In this way the function f(a) will be defined by transfinite induction
for every ordinal number a > 0 and it will obviously satisfy conditions 1),
2) and 3).
As regards the above argument, there might be certain objections
in connection with the ,,assumption that we have already defined someth-
ing“. The explanation of defining by transfinite induction will be given
in the next paragraph.’
16. Definitions by transfinite induction. A general (but not the most
general, as will be seen in XIV.17) form of definitions by transfinite
induction is the following.
Let h denote a function associating with every finite or transfinite
sequence {d@g}s<4, including the ,empty“ sequence A= {de}eco (or with
every sequence of type a, such that a <@ for a given ordinal number 9g)
a certain element h({ag}<<q).
A function f(a) of a variable a, which is an ordinal number, is defined
by transfinite induction as follows. We write /(0)=h(A). Now let a
denote an ordinal number a > 0 (or an ordinal number such that 0 < a <9)
and suppose that we have already defined the values of f(é) for <a.
We then write f(a)= h({f(E)}e<a). We say that, in this way, we have
defined the function f(a) for every ordinal number a (or for every ordinal
number a <q) by ee induction.
Thus we ie for instance, f(1)=h ({#(0)3), f(2)= h(£f( 1)}),
f(3)= h(EF(0), F(1), £(2)3),
--» F(@)= A({F(O),
(1), F(2), --.}), terete
= h({f(0), f(1), F(2), «+5 f(@)})-
The Fee of defining by transfinite induction finds its justifica-
tion in the following
THEOREM 1 (on defining by transfinite induction). Let h denote
a function which associates with every finite or transfinite sequence {astsca,
including the empty sequence A = {ag}s<o(or with every sequence, of type a,
312 XIV. Ordinal numbers

such that for a given ordinal number p we have a <q) a certain element
h({ag\e<a). Then there exists one and only one function f(a), defined for
every ordinal number a (or for every ordinal number a <q), such that

eune Pe f(a) = h(Ef(2}ece)

for every ordinal number a (or for every ordinal number a <q) (Kuratowski
and Mostowski [1], p. 198, Theorem 5).
Proof. To begin with, we shall prove that there exists at most
one function f satisfying Theorem 1. Suppose that there exist two dif-
ferent functions, f, and f,, satisfying Theorem 1. Since the functions f,
and f, are different from each other, there exist ordinal numbers a (or
a<g) such that /,(a) ~ f.(a), and among them there exists the smallest,
which we shall still denote by a. We must have a > 0, since if the func-
tions f, and f, satisfy the conditions of Theorem 1, then 7,(0) = h( A)
and 7,(0) = h( A), whence /,(0) = 7,(0).
From the definition of number a it follows further that 7,(¢) = f2(é)
for <a, which, by formula (16.1) for f= f, or f = f., gives f,(a) = f,(a),
contrary to the definition of number a. Thus the assumption of the exist-
ence of two different functions, /, and f,; satisfying Theorem 1 leads to
a contradiction.
Now we shall prove that for every ordinal number py > 0 there exists
a function f = f,, defined for every ordinal number a <q and satisfying
formula (16.1) for a <q.
Suppose that for a certain ordinal number ¢y there exists no such
function f,. Among such numbers @ there exists the smallest, y. There-
fore for »<y each of the functions f, exists (and, as has been
proved, there is only one such function f, for every ordinal number @ < y),
but the function f, does not exist.
Let g denote an ordinal number <y. From the definition of the func-
tion f, it follows that the formula

(16.2) fola) = h ({fol§) }e<a) for a<—.

is satisfied.
Similarly, if 6<q, and thus of course 6<y, then the function fo
satisfies the formula

fe(a) = h({fo(€)}s<a) for a<6,

and since, as has been proved above, there is at most one such function fe,
by (16.2) and in view of 0<q, we obtain

(16.3) fala) =fp(a) for «a<0<o9<y.

 § 16. Definitions by transfinite induction 315

Now let us write

(16.4) f(a)=f.{a) ‘for a<yp and f(p~)=h({f(é)}ecy)-

By (16.4) and (16.2) we have

(16.5) f(a) = fala) = h({falE)}ece) for a<y,

and by (16.4) and (16.3) we obtain

(16.6) f(E)= fle) =felé) for E<a<y.

Formulae (16.5) and (16.6) give at once formula (16.1) for a<y
and the second of the two formulae (16.4) proves that formula (16.1)
is also true for a= y. Thus the function f(a), defined for a <y would
be the function /,, in contradiction to our assumption that such a func-
tion does not exist.
We have proved that for every ordinal number g > 0 there exists
a function f = f,, defined for every ordinal number a <@ and satisfying
formula (16.1) for a <9.
Now, if we write f(a) = f,(a) for every ordinal number a we could
easily prove that the function f satisfies formula (16.1) for every ordinal
number a, q. e- d.
In particular, if we wanted to apply Theorem 1 to the definition
by transfinite induction of the power uw, where wu and a are ordinal num-
bers, « > 1 and a> 0, we should take h as a function defined in the fol-
lowing way: 1) h(A) = 1, 2) if a is an ordinal number of the 1-st kind,
a= f6+1, and {dz}cc, iS a Sequence (of type a) of ordinal numbers, then
Witeice = d,-m, 3) if a is a number of the 2-nd kind and {ag}s<, 18 an
increasing sequence of ordinal numbers, then

h({ag}eca) = limag ,
é<a

4) in all other cases h({d@e}eca) = 1.

Thus the definition of the exponentiation of ordinal numbers by
means of transfinite induction, as well as the justification of defining
by transfinite induction, is more complicated than the constructive defini-
tion of the power uw given in XIV.15.
17. Transfinite products of ordinal numbers. Let {a:}:<, denote a trans-
finite sequence of type gy of positive ordinal numbers. By transfinite
induction we can easily define the product

(171) [] 4:
<p
as follows.
314 XIV. Ordinal numbers

J] a; is the smallest ordinal number > |[ ag- a9 for 0 <q (Haus-

EXP E<6
dort (2), 02199):
To justify this definition, however, we should need a more general
theorem than Theorem 1 of XIV.16. Namely we should modify theorem 1
of XIV.16 assuming that the function h depends not only on the sequence
{Ag}e<q but also on the ordinal number a, 7. e. we should replace the func-
tion A({ag}eca) by the function h({ag\eca, a); moreover, formula (16.1)
should be replaced by the formula

f(a) = (Ef (E)}e<ay a)«

The proof of the theorem thus modified differs very slightly from
the proof of theorem 1 of XIV.16 (Kuratowski and Mostowski [1], p. 198).
In order to justify the above-metioned definition of the transfinite
product |] a: of ordinal numbers, given a sequence {dghecp Of ordinal
é<
numbers, we should define the function h({az}z-,, a) for the sequences
{de}ecq Of ordinal numbers as the smallest ordinal number > aga, for § < a.
In particular, if a,, a,, ... is an infinite sequence (of type mw) of positive
ordinal numbers, then the infinite product a,a,a,... is the smallest ordinal
number > a,a,... a, for n=1,2,... and if, for infinitely many different
indices n, we have a, >1 then

Oy Oy One LEN tents

n<ow

If ag = a for € < » where oa and @ are given ordinal numbers, a > 0,

g > o, then

H. Le TIS
E<p

the power of an ordinal number with an ordinal number as an exponent

is thus a particular case of the finite or transfinite product of ordinal
numbers.
In order to prove this formula it suffices to compare the definition
of the product (17.1) with the following, easily provable property of the
power of ordinal numbers: :
a? is the smallest ordinal number >oa®-a for 0<> .
It can easily be proved that the product of a transfinite sequence of
ordinal numbers is associative but, in general, it is not commutative. In
particular, it can be proved that by changing the order of factors in a infi-'
nite product of type w of ordinal numbers we always obtain only a finite
number of different values, and in the case where the factors of such
a product form an increasing infinite sequence of ordinal numbers the
 § 17. Transfinite products 315

value of the product does not depend on the order of the factors (Sier-
pinski [56], p. 21 and 24).
Now, a transfinite product of type +1 of increasing ordinal num-
bers may give infinitely many different values if we change the order
of its factors. H.g. it is easy to find that the transfinite product
(of type w+1) 1-2-3...@, gives the value wn if we interchange the
factors n and. w, these numbers being different for n= 1, 2,...
For ordinal numbers a let us write

raay=[fé and at=r(a+1).

E<a

It is easy to find that ['(w) = w,w!= w*. We can also calculate

(atelio 2 a=? (i. 2) t= Mette? Ia?) wm =a (ot) (ol)! = (o*)h—
See std C—O» 1 (0?) =a 2,

EXERCISE. For a natural number n give an example of a infinite product (of

type w) of ordinal numbers which gives n different values if we change the order of
its factors.
Solution. Such is for instance the infinite product a,a,a;... where a,= 2 for
4=1,2,...,n—1, while a, = » and a,=1 fori >n-+1. It can be proved that if we
change the order of the factors of this product, its values are numbers w-2*—? —1 where
k =1,2,...,n and those numbers only.

18. Properties of the powers of ordinal numbers. An increasing func-

tion of a variable which is an ordinal number whose values are ordinal
numbers satisfying condition 3) of XIV.15, p. 310, is called continuous.
Thus from Theorem 1 of XIV.15 and formula (15.1) immediately follows
THEOREM 1. If uw denotes a given ordinal number >1, then the func-
tion f(a) = ut is the only continuous function of a variable which is an
ordinal number satisfying the conditions f(1) = mu and f(a+6) = f(a)f(p)
for any ordinal numbers a and fp.
As we know, an analogical theorem holds for real functions of a real
variable with real yw > 0.
It will be observed that the power yw*, as a continuous function
of the exponent a, is not a continuous function of the base mw, since for
instance
limn = w
n<o

but
limn? = wm < @?.
n<o

Also
o® ~ limn?
na<o@
 316 XIV. Ordinal numbers

because n®= w for n=1,2,..., stnce we have

== noer,
k<o
whence
lim k f
Wye 8 ee Ie =
k<o

From formula (15.3) it follows that the power m* is an increasing

function of the exponent. Still there are exponents a for which the power p*
is not an increasing function of the base w since for instance 2° = 3° = a.
However, with a given exponent a, the power yu? is a non-decreasing fune-
tion of the base since obviously, for uw, < ps, the set Z(u,, a) is a subset
of the set Z(s, a).
Now we shall prove that for any positive ordinal numbers wv, a and £
we have the formula
(18.1) (u2}@ = y%
Let w and a be given positive ordinal numbers. Formula (18.1) nat-
urally holds for 6 = 1 since (w*)! = pt = pw 1. Now let y denote an ordinal
number >1 and suppose that formula (18.1) is true for every ordinal
number f < y. We shall distinguish two cases.
1) Number y is of the 1-st kind, therefore y= 6+1 where 6 < y.
Thus: formulas(48.1) "is true... But.= by “(251), \Co%)7 (6 eo nt eeee
bys(18:1) and. (15.1) we, thus) have. (ut)! ==) ue ys pee ye) a
-which proves that formula (18.1) is true for B = y.
2) Number y is of the 2-nd kind. As we know from XIV.6, we then
have the formula
y == Mme
<y
From property 3) of the power it thus follows that

(u2)” = lim (u)é.

é<y

But in view of € < y formula (18.1) holds for 6 = ¢; therefore (u2)* = py.
Thus we have
(U2)? ite
é<y
But, according to Theorem 3 of XIV.11, in view of
IDR as See
é<y
we have
JONES Ye rea pie
é<y
 § 18. Properties of the powers of ordinal numbers 317

whence, in view of property 3) of the power:

Lama t* = uv.
S<y

Therefore (u*)” = uw” and formula (18.1) is true for B= y.

We have proved that if y is any ordinal number >1 and if formula
(18.1) is true for every ordinal number f< y, then it is also true for B = y.
Hence (and from the fact that formula (18.1) is true for 6 = 1) we infer
by transfinite induction that formula (18.1) is true for every positive
ordinal number f.
EXERCISES. 1. Prove that the sum and the product of two continuous functions
of a variable which is an ordinal number whose values are ordinal numbers need not
be a continuous function.
Proof. Let f,(é) = f.(é) = € for every ordinal number &. The functions f, and f,
are obviously continuous. The functions g(é) = f,(€)+ f.(€) = €-2 and h(&) = f,(&)-f.(€) =
= & however, are increasing but not continuous, since for example

liimn = w
n<w
while
lim g(n) = lim(n-2) = @ < g(w) = @-2
nN<w Row
and
liimh(n) =hlmn? = wo < h(w)=@?.
n<w N<w

oO (22)
2. Prove that n® =o for natural n> 1.
Proof. In view of 1+. a =o we have wm? =w'*® =w-q@”, and since, as we know,
@
for natural n > 1 we have n® = @, therefore, by (18.1), n°” = 27° = (n®)°” = w®
3. Give an example of a function f(a, f) associating with every ordered pair a, B
of positive ordinal numbers a positive ordinal number f(a, #) and increasing with respect
to a for every given f and with respect to f for every given a.
Solution. Such is the function (a+ 1/8 -_ a, since for a < a, we have (a+ 1)P ax
“<< (a,+ te a <(a,+1)'+a,. and for B < Bi: (a+ 1)? + a> (a+ 1) > (atift=
= (a+ 1)%(a+ 1) = (a+ 1)'a+ (at 1% S (at 18+ (a+ 1) > (at Uf +a.
4, Prove that for every ordinal number a we have
224+ 3% = B24 27,
Proof. This is obvious for finite a. If a= om. then a = wé+ k where €>1 andk
“is a finite number >0. Hence 2%+4.3% = 20F+*1 grbth _ (9@)F oF1 (35. gF 2 wy. oF
+ hs F = w®- (2° 3°) = w° (3+ 2") = 37+ 2°
5.; Prove that (@-+ 1)? = o®.
Proof. For natural n. we have: wo < (w+1)" < (w+) = (w-2)" = w'-2<
<0'oO= or whence, on account of the fact that

lim w” = lim w” tT
1
— @®,
nN<w n<o
‘we easily conclude that
lim (@-+ 1)” = w®
N<w

which gives (w+ 1) =, q.e.d.

318 XIV. Ordinal numbers

6. Prove that if f(€) is an increasing continuous function of an ordinal variable &

and if for a certain ordinal number a we have f(a) 4 a and write a, = a and a, = f(a,_,)
forn=1,2,..., then the infinite sequence ay, a,, a, ... will be increasing and, writing

Ave I coe
N<w!
we shall have {(A) =A.
Proof. As the function f(&) is increasing, we easily deduce by transfinite induc-
tion that f(€) => é for every ordinal number &. Since f(a) #~a, we have f(a) >a,
and thus a,= f(a) = f(a) > a=a,, whence a, > a. Now if we have a, >
> a,_, for a certain natural n, then, since the function f is increasing, we obtain hence
F (Qn) > f(Gn_i)> 4. @- G4, > a,. Hence it follows by induction that this inequality holds
for »= 0,1, 2,... Thus there exists a limit

A = lima, .
n<w

In view of the continuity of the function f we have

f(A) = limf(a,) = lima,,, = 4,
n<w r<o
therefore f(A) =A, q.e. a.
7. Prove that for transfinite ordinal numbers 4 of the second kind we have 14+
+ 2 — 34,
Proof. If 2 is a transfinite ordinal number of the second kind, we have 4 = wé,
where é is a positive ordinal number. Thus 14+ 27= 1+ 9% 9° and-since 2°= 3° = w,
we have 1’+ 24 = 325 = 34.
Remark. For Fermat’s last theorem on transfinite exponents see Sierpinski [55]
(p. 202-205).
The roots of the equation f(&)= & are called the critical numbers
of the function f. From the theorem which we have proved in Exercise 6
it follows that if f(é) is an increasing continuous function of an ordinal
number, then for every ordinal number a there exists a critical number
of that function >a.
Now we shall prove that for every ordinal number uw >1 and for
every ordinal number a
(18.2) we>a.
For a= 0 formula (18.2) is obvious, and for a> 0 it suffices to
observe that u* is the type of the set Z(u, a) (see § 14) and that the se-
quence of type a whose one term is equal to 1 and the remaining terms
are equal to 0 form a subset of type a of the set Z(u, a).
Now we shall prove
THEOREM 2. For every ordinal number u > 1 and every ordinal number
—& > 0 there exists one and only one ordinal number a such that

(18.3) Ue ei
Proof. From formula (18.2) it follows that (for given w>1 and
& > 0) there exist ordinal numbers y such that w > &, since for example
 § 18. Properties of the powers of ordinal numbers 319
9

pitt > ub > &; let » denote the smallest of them. If the number » were
of the 2-nd kind, then, as we know XIV.15) the number ~” would be
the smallest ordinal number >! for €< ». But also, if » were a number
of the 2-nd kind, then, for ¢<¥v we should have ¢+1<¥, which, in
view of the definition of the number » gives the inequality ptt! < ,
therefore uw? < & Thus the number é < w” would be greater than any of
the numbers uw for ¢ <», in contradiction to the properties of the num-
ber uw’. Therefore number »y is of the 1-st kind and we can write y= a+1,
whence, in view of the definition of number », we obtain formula (18.3).
We have proved the existence of an ordinal number a for which
inequalities (18.3) hold. It can easily be seen that there is only one such
number, because if, besides the inequalities (18.3), also the inequalities
wP<E< pot) held, we should have pt < pot? and pf < ptt, whence,
by (15.3), a< 6+1 and 6 < a+1, therefore a < 6 and 6 <a, whence f=a.
We have proved theorem 2.

19. The power w*. Normal expansions of ordinal numbers. Now we

shall deal in particular with the power *.
To begin with, we shall prove that for every ordinal number a the
number wt is a prime component. Suppose that it is not so; thus there
exists the smallest ordinal number a such that w* is not a prime com-
ponent. We have a > 1, since the numbers w° = 1 and w! = w are known
to be prime components.
If a were a number of the 1-st kind, then we could write a= +1
where 0 < <a. Thus the numbor w* would be a prime component and
also, by theorem 1 of XIV.14, the number w*-@ = w+! = w* would be
a prime component, in contradiction to the definition of number a.
Therefore number a is of the 2-nd kind. Thus, as we know from
XIV.15, w* is the smallest ordinal number greater“than any of the num-
bers w* for € <a, and the set P of all the numbers w* where € < a does
not contain a greatest number (since if <a, then, a being a number
of the 2-nd kind, we have also é+1<a, and w* < w'*1). And since all
the numbers of the set P, as follows from the definition of number a,
are prime components, we conclude according to theorem 3 of XIV.6
that the number w* (as the smallest ordinal number that is greater than
any number of the set P) is a prime component, again in contradiction
to the definition of number a.
Thus the assumption that it is not every number w*, where a is an
ordinal number, that is a prime component leads to a contradiction.
Now we shall prove that, conversely, every prime component is
a certain power of the number w (with an exponent which is an ordinal
number). Therefore let 9 denote a given prime component. By Theorem 2
320 XIV. Ordinal numbers

of XIV.18 there exists an ordinal number a such that

(19.1) Oe Pe OCs
According to the corollary to Theorem 1 of XIV.14 the smallest
prime component >o* is wt w= w%t!; the inequalities (19.1) prove
that 0 = w*. We have proved
THEOREM 1. Prime components are powers of number w (whose
exponents are ordinal numbers) and conversely.
From Theorem 1 and Theorem 2 of XIV.6 it immediately follows
that every ordinal number a> 0 may be represented uniquely in the form

a= +. we? ar ee on,

where m is a natural number and p,, Bo, .--, bm are ordinal numbers such
that By > pp >... > Bm > 0. Joining equal components in this decomposi-
tion of number a we immediately obtain
THEOREM 2. Hvery ordinal number a > 0 may be represented uniquely
in the form
(19.2) a= w74,+ w%a,+...+ wa,

where * and 4a,, a,,..., a are natural numbers while a,, a2, ..., ax 18 a de-
creasing sequence of ordinal numbers.
Formula (19.2) is called, after Cantor, the normal form of the ordinal
number a and the number a, is called its degree. By (19.2) and (18.2)
we have a>o7>a,; thus the degree of a number (in its normal form)
is not greater than the number itself.
EXERCISES. 1. Determine the normal expansion of the number
(m+ 1)2(m4+ 1)3(m+ 1)4.
Answer. 3-44 @?- ogi m:2+ 1.
2. Determine the normal expansion of the number 2% where a is a given ordinal
number.
Solution. As we know from XIV.11, every ordinal number a may be represented
(uniquely) in the form a = wé+k where & is an ordinal number >0 and k is an inte-
ger >0. Since 2° = w, we have 2% = 2°°+* — (2°)§.9* — w*.2*. Thus the normal ex-
pansion of number 2% is 2% = w*- 2". ;
3. Prove that prime components >1 are powers of number 2 whose exponents
are numbers of the 2-nd kind and vice versa.
Proof. If 9 is a prime component >1, then, by Theorem 1, there exists an ordinal
number & > 0 such that 9 = o*, and since w = 2°, we have 9 = 2°, which shows that @
is a power of number 2 with an‘exponent which is an ordinal number of the 2-nd kind.
On the other hand, if a is a number of the 2-nd kind, a=&, then 2% = 2° = (2%)§ = w®
and it follows from Theorem 1 that 2° is a prime component (>1 of course).
4. Knowing the normal expansion of positive ordinal numbers a and f determine
the normal expansion of the numbers a+ and af.
 § 19. The power wm" 321

Answer. Let a = w%a,+ o™”a,+ ...4 oa, and B= wtb, + wb, + ...4 ot, be the
normal expansions of numbers a and f. If a, < f,, then a+f=. Otherwise there
exist indices 1<k such that a, > f,; let 7 denote the greatest of them. If a, > fy,
then the normal expansion of the number a+ is w%a,+ oa,.+ ...+ o%a,4 wtb, 4
a7 el ee wlth, , and if a,=,, then the normal expansion is wla,;-+ w"a,+ ...4+-
+ w%—la, »+ cot (a, + b,)+ w2b,+ wF8b,+ ...+ wht,.
If number
f is finite (i.e. B =56,), then af = w%a,b,+ w%a,+ w8as+ ...F w%a,.
If number f is transfinite (i. e. B, > 0) and of the 1-st kind (i. e. 6, = 0), then af =
= mAthip+ matfapt 4+ oa +Piap, s+ ola,
b, + w%a,+ ...+ w%a,. Finally, if number f
is of the 2-nd kind (i.e. B, > 0), then af = w% 11d, + wt ap, + ... + wt th,.
5. Knowing the normal expansion of an ordinal number a > 0 determine the nor-
mal expansion of the number a”.
Solution. If a=1, then a® = 1. If 1 < a < a, then, as we know, the normal
expansion of the number a® is w. If a => and the normal expansion of number a is
(19.1), then we have a, > 1 and a+ watt — mt! therefore w% <a < w4*' whence
wt < g® < mt", But, according to exercise 3 of $10, we have (in view of a, > 1)
(a, +1) = aw. Thus a? = o%®. The normal expansion of the number a® (where
a>) is therefore a? = w%®.
6. Prove that if « is an ordinal number >1, k — a natural number, a,, a, ..., a, —
ordinal numbers such that a, > a, >... > a, > 0, and »,, »,...,% — ordinal numbers
greater than 0 and smaller than mw, then w%y,-+ u™r_+ ...4+ uy, < watt.
Proof. In view of », << pw and w > 1 we have wy, < ww = ons Thus the inequa-
lity which we are to prove is true for k = 1. Now let us suppose that it is true for a na-
tural number k& and let a, > a, > a, >... > a, be k+1 ordinal numbers not smaller
than 0 and 7), 7, ¥2,..-,%—k+ 1 ordinal numbers greater than 0 and smaller than wy.
. ; +)
We shall have uv,+ w%y,+ ...+ wy, < w%yyt wT,
But, since a) > a,, we have a,+1<a,. Therefore p“ovp+ ut? < wy, + jis
= p(y +1) < wn = wt, because, in view of ») < uw, we have »+1< yp. Thus
Oy + wat? < n%**, whence it follows that our inequality is true for k+ 1 components.
It has been proved by induction for an arbitrary finite number of components, q. e. d.

7. Prove that if 6 = tb,+-...4 0%), is the normal expansion of the number

6 = o, then, for y > 0, the first component of the normal expansion of the number p”,
is wb, or wt”, according to whether y is a number of the 1-st kind or of the 2-nd kind.
Proof. For finite y the proof presents no difficulties. If y is a number of the 1-st
kind >, then y= wé+-k where > 0 and & is a natural number, whence fp” = peet* =
= (6%) -p*, and since B® = wt? (Exercise 5), we have — epi. Be But the first com-
ponent of the normal expansion of the number f” is w"t"b,. Therefore the first component
of the expansion of the number £” is wt. @t"b, = oP S)p — wtb, for numbers
-y > o of the 1-st kind. Now, if y > w is a number of the 2-nd kind, then y = w& where
& > 0, whence fp” = (p%)* == (p15 = pile) yh? ,
The required proof is complete.
8. How many components has the expansion of the m-th power (where m= 1,
2,...) of an ordinal number a whose normal expansion has k components?

Answer. kif a is a number of the 2-nd kind, and mk—m-+ 1 if a is a number of

the 1-st kind.
Cardinal and ordinal numbers 21
322 XIV. Ordinal numbers

Let us turn again to the normal form of ordinal numbers. Formula

(19.1) is analogous to the decimal expansion of natural numbers, the
base 10 being replaced by o.
However, the following more general theorem holds:
THEOREM 3. For every ordinal number u > 1 every ordinal number
a>0 may be represented uniquely in the form

(19.2) a= wy, + wr +... ty.

where k is a natural number, a1, a2, ..-, a are decreasing ordinal numbers >0
aNd 1, Vo, ..-5 % are ordinal numbers satisfying the inequalities

(19.3) OFS eco LOR 4 =e eee

Proof. Let « denote a given ordinal number >1, a — an ordinal

number >0. By Theorem 3 of XIV.18, there exists an ordinal number a,
such that
(19.4) ie 0 =
In view of a< u2t*— uy, and by Theorem 1 of XIV.11 we can
represent number a in the form

(19.5) a= p 'y,+a’,

where 0 <a’ < w™ <a and, in view of (19.4) 0<»< yp. If a’ > 0, then
we can further write

(19.6) C= 7,4 ae

where 0 <0 2 ~<a < p+ whence a,.¢, ands 0.— 9 Jivo 0,

we can deal with the number a’’ in the same way as with a’, etc. Since,
moreover, we have a>a’>a’’>..., for a certain natural k we shall
obtain the formula
Buea) = uty, al Ae
where a = 0.
From formulae (19.5), (19,6) ete. we immediately obtain formula
(19.2).
In order to prove that there is only one decomposition (19.2) of
number a let us first observe that formula (19.2) and exercise 6 imply
1 ea oi eee wrt, thus it follows from Theorem 3 of XIV.18 that number a,
is well-defined (by number a), and, in view of a < w™* = uw. and by
Theorem 1 of XIV.11 there exists only one system of ordinal numbers
a’ < w™ and », < w such that a= w%y,+a’. Hence, by (19.2), we obtain
a’ = wy, + wv,+...+ u»,. Repeating our argument for numbers a,
we conclude that numbers a, and », are well-defined, and we have
a’= wy,+ a" and a’ = ur,+...+ u»,. Repeating this argument k times
 § 19. The power w¢ 323

we reach the conclusion that numbers a,, dz, ... a, aNd 11, 2, ..-) MR are
well defined by number a, which proves that there is only one decom-
position (19.2) (where numbers aj, ag,...,a, and ,%,...,% Satisfy
known. conditions).
Theorem 3 of XIV.19 is thus proved.
In particular, if ~ = 2, we obtain for every ordinal number a > 0
the expansion
Q 27 ee
where a,, a), ..., 4, 1S a decreasing finite sequence of ordinal numbers > 0,
there being only one such expansion for every ordinal number a > 0.
Thus every positive ordinal number is the sum of a finite number of
powers of number two (just as is the case with natural numbers).
It will be observed that it is not every ordinal number that can be
represented by numbers smaller than itself with the aid of expansion
(19.1), or even, more generally, with the aid of expansion (19.2). There
exist ordinal numbers a > 1 such that for any uw < a the first exponent a,
in formula (19.2) is equal to a; we shall get acquainted with such numbers
in the next paragraph.
It is almost self-evident that with certain mw for certain a the ex-
ponent a, in formula (19.2) is equal to a; e. g. we have w = 2° and, more
generally, .0 == n° 10n n= 1,72...

20. Epsilon numbers. Let

(20.1) Rie (10 ee

eeTe ae

The components of this infinite series, as powers of number w, are prime

components according to Theorem 1 of XIV.19; Therefore, by Theorem 1
of XIV.6, they absorb smaller components preceding them. On the other
hand, the components of series (20.1) increase, since denoting by ¢,
the n-th component we have of course 9, = wo > w= gq, and the ine-
quality g@ > Gn—1 implies the inequality g,41 = wo" > w™"-! = gy.
Thus each component of series (20.1) absorbs all components preced-
ing it and therefore each partial sum of series (20.1) is equal to the last
component of that sum, 7. &. 9, +¢.+...+ Gn = 9. for n= 1, 2,... Since,
as we know from XIV.6, the sum of an infinite series of ordinal numbers
is the limit of its partial sums, we have
é== mg,
nrA<@

Since a power is a continuous function of the exponent, it follows

that
wo = lima” = HIM Gye == 2
n<o@ n<w
O24 XIV. Ordinal numbers

Thus « satisfies the condition

(20.2) Ores
Cantor calls every ordinal number « satisfying equation (20.2) an
epsilon-number. Thus they are the critical numbers of the function f(é) =
= w*. We shall prove that number (20.1) is the smallest of such numbers.
Indeed, suppose that w*= a. We have hence a> 1, therefore a= w* >
>o=gy, whence a>g,. On the other hand, if for a certain natural
number » we have a>g,, then a= > w™= @ai1, Whence a > gait.
We infer by induction that a>, for n=1,2,..., and thus

a= limg, =e,
n<o

tones Ook Cs
For every ordinal number y there exists an epsilon-number > y.
Indeed, let »,=y+-1 and y=" for, 7=—2,3,... From Formula
(18.2) it follows that w’+t > y+1, and since the left-hand side is a num-
ber of the 2-nd kind and the right-hand side of the 1-st kind, we
have w?tt>y-+1, and thus y.>y,. Hence, by induction, we easily
conclude that y, < y, < y;... and then we prove, in the same way as we
have done above for number «, that the number

H(y) = limy,
n<o

is an epsilon-number and that, as can easily be proved, it is the smallest

epsilon-number >y. For if w= a> y, then, on account of a= w* being
a number of the 2-nd kind, we conclude that a>y+4+1, 4.e a>y,
whence a= o*>o"=y, and we infer by induction that a> y, for.
n—1,2,..., whence
a > limy, ,
n<ow

4%.€ a> EH(y).

Thus we shall have «, = H(0)= «; the next epsilon-number after «
will be ¢,= E(e), the next after that — EH, = H(e,) ete.
\ If a is a given epsilon-number, then the set of all epsilon-numbers
smaller than a is well-ordered (as a certain set of ordinal numbers). Leté
denote the type of that set: it will be a certain ordinal number >0..Let us
denote a by es. Since, as we have proved above, H(y) is the smallest
epsilon-number greater than y, for every ordinal number é we shall have
€e41 = LH (é,). | .
Now we shall. prove that 7f 2 ts an ordinal number of the 2-nd
kind, then
oy legis
E<A
 § 20. Epsilon numbers 32 Ct

Indeed, let
a= lim €e.
E<A
From the properties of the power of ordinal numbers it follows that

wo = limw®.
E<A

But w* = e for € <A, since each of the numbers ¢; is an epsilon-

-number. Thus we have
Otc MNS: Sue, 5
<A

therefore m* = a, which proves that a is an epsilon-number. Since

a=lme,
E<a

a is the smallest number that is greater than any of the numbers «; for
—& <A. It follows hence that a= «,. Thus

6, = lime: , ).g-e.7d.
E<A
In particular, the next epsilon-number after the numbers ¢é9, &, é, ..-
will be*
So es
n<o

Thus all epsilon-numbers may be obtained from number 0 by applying

two operations: H(y) and on

It is also easy to ane et to every ordinal number a there cor-

responds an epsilon-number «, well defined by number a.
It will also be observed that epsilon-numbers may also be defined
as ordinal numbers a >o satisfying the equation 2*°— a. For, on one
hand, if a is an epsilon-number, then a>qm and we have l+a=a
whence a= 0° =o **=w-o"= a, therefore a= wa and 2° = 2% =
= (2°)*= w* =a. Thus every epsilon-number a satisfies the equation
2°— a (and of course is an ordinal number >, although the number
a=, which is not an epsilon-number, also satisfies this equation).
On the other hand, if the ordinal number a> o satisfies the equation
pe iene where 62 and ac 2 20 0° = Or
=n" > w*, whence w+a=a and therefore, in view of a=w+é, we
obtain € = a, and thus a = w* = w* = w*, whence w* = a, which ‘proves
that a is an epsilon-number.
We shall also prove the following theorem on_ epsilon-numbers:
326 XIV. Ordinal numbers

THEOREM 1. If a is an epsilon-number, then

(20.3) Eta=a for Fa,

(20.4) EV (OT oo,
(20.5) t= Oe Ol tae Sorc,

Proof. Formula (20.3) immediately follows from the definition of

epsilon-numbers and from Theorem 1 of XIV.19 and Theorem 1 of XIV.6
(since, by Theorem 1 of XIV.19, epsilon-numbers are prime components).
In view of (20.3), and w*= a, we have, further, for 1<&<a,a<éa<
<wfwot= wit*= @t=a, whence fa=a, i.e. formula (20.3). Finally,
if 2 <é <a, then (since, as we know, a, being an epsilon-number, satisfies
the equation 2°= a) we have, by (20.4): a = 2° < & <(o')*= 0"= w=4a,
whence &* =a, 7. e. formula (20.5). Theorem 1 is thus proved.
Transfinite ordinal numbers a satisfying condition (20.4) are termed
by G. Hessenberg 6-nwmbers. Thus every epsilon-number is a 6-number
but not necessarily vice versa. It is easy to prove that in order that an
ordinal number be a d-number it is necessary and sufficient that it be of
the form w®” where y is an ordinal number >0. Ordinal numbers of the
form af? ¥ have also been investigated and a more convenient notation
[a,, A, ..., G] has been introduced for them (Neumer [1], p. 419).
Obviously, apart from epsilon-numbers, only the number a=
satisfies property (20.5), since (20.5) implies that 2° = a.
If a is an epsilon-number while 6 and y are ordinal numbers <a,
then, by (20.5), for B >2 we have fp” < f*=a, whence f” <a, which
is also true for 6=0 and 6=1. Thus if a is an epsilon-number, then
Bb’ <a for B<a and y<a.
On the other hand if a is an ordinal number > ao and if fp” < a for
p<a and y<a, then a cannot be a number of the 1-st kind, since
with a = +1, in view of a > a, we should have é > w and é < a, whence
f&<qa=é&+1, and thus é§*<é, which is impossible for &>1. There-
fore a is a number of the 2-nd kind and thus

= nee
<a

In view of w <a and of the assumption concerning number a, we

have w* < a for & < a, whence

ot = limo’ <a,
E<a

and since, by (18.2), w* >a, we have w*= a, which proves that a is an

epsilon-number.
 § 20. Epsilon numbers 327

We have thus proved

THEOREM 2. In order that an ordinal number a be an epsilon-number
at is necessary and sufficient that a > w and

(20.6) Ban Kore ips os Sandie y <a.

Ordinal numbers a > @ satisfying condition (20.6) are termed prin-

cipal numbers of exponentiation. Thus in order that an ordinal number
be a principal number of exponentiation it is necessary and sufficient
that it be an epsilon-number.
With the aid of number ¢ it is easy to give an example of two dif-
ferent transfinite ordinal numbers a and f# such that

(20.7) of = pa;
such are for instance the numbers a= w and 6 = ew. Indeed, by (20.2)
we have ew = ww = w*t1, and since, by XIV.11, Exercise 3, we have
(2 Dw co, this) cives: b2==.(em)P = (w*t!)? = wo6tDo— o@.— of, Num-
ber « may of course be replaced by any epsilon-number. Thus equation
(20.7) has infinitely many solutions in different ordinal numbers a and f.
Among finite ordinal numbers a and />a satisfying equation (20.7)
there is, as we know, only one pair: a= 2, B= 4.
Equation (20.7) for ordinal numbers will be dealt with in greater
detail in XIV.27, where all its solutions will be determined.

EXERCISES. 1. Prove that if 6 > and 6” = y, then y is an epsilon-number.

Proof. If pf’ =y and B>a, then obviously y is a transfinite ordinal number.
Let 6 = w%tb,+... and y = w"!¢,+... be the normal expansions of numbers f and y».
According to XIV.10 Exercise 7 the first component of the normal expansion of num-
ber B” is w1’b, or w1”, whence, in view of 6” = y, by comparison with the normal ex-
pansion of number y, we obtain f,y = y,, thus y,; = f,y > y. Since, on the other hand,
y > w1> y, by formula (18.2), we have y = y, and thus y > w”; since w” > y by (18.2),
we obtain y=’, which proves that y is an epsilon-number.
2. Prove that if y is an epsilon-number, then for every ordinal number a we have
ae (a),
Proof. The proof follows from the observation that (a°)” = a®” and that for every
epsilon-number y we have y = w” > w, and thus, by Theorem 1, formula (20.4): wy =
ae
Note. We shall investigate all the solutions of the equation a iG. )’ in ordinal
numbers in XIV.27, Exercise 2.
3. Prove that for ordinal numbers a property (20.5) implies property (20.4), and
property (20.4) implies property (20.3).

21. Applications of the normal form. Right-hand natural divisors of

an ordinal number a>O0 whose normal expansion is (19.1) are identical
with the divisors of number ay.
328 XIV. Ordinal numbers

For if a = Bn, where the normal expansion of number f is 6 = ab, +

+ wb, +...+0%,, then a= pr= w%dn+o%),+...+0%,, whence, in
view of the uniqueness of the normal expansion, we obtain a, = f, and
a, = byn, and thus n|a,.
Conversely, if n]a,, then a,= bn, where b, is a natural number
and, by (19.1), we obtain a = (w%b,+ w"a,+...+@%a,)>n and number n
is a right-hand divisor of number a,.
Therefore, among the right-hand natural divisors of an ordinal num-
ber a> 0 there exists a largest, a,, which is divisible by each right-hand
natural divisor of number a. It also follows that if the ordinal num-
ber a is divisible on the right by the natural numbers m and n, then
it is also divisible on the right by their smallest common multiple [m, n]
(since if m|a, and n|a,, then also [m, n]|a,) 4).
As regards left-hand natural divisors of ordinal numbers of the
1-st kind, we have dealt with them in Exercise 2 of XIV.10, while accord-
ing to Exercise 1 of XIV.11 every natural number is a left-hand divisor
of every ordinal number of the 2-nd kind. For ordinal numbers of the
2-nd kind we have moreover
THEOREM 1. An ordinal number a of the 2-nd kind with the normal
expansion (19.1) has as its left-hand divisors all positive ordinal numbers
<w™, and besides only a finite number, not smaller than 0, of left-hand
divisors (Shermann [1], p. 111-116).
Proof. Let a> 0 be an ordinal number of the 2-nd kind with the
normal expansion (19.1). Thus we have a,>a,>...>a,>0 and, by
(19.1), we obtain a= w*(@" %a,+o7 “a,+...+@%-)
*a,_1+a,), whence
‘(follows that w™ is a left-hand divisor of number a. Since, as follows
from Theorems 1 of XIV.19 and 2 of XIV.14, every positive ordinal
Sumber £ < w® is a left-hand divisor of number w”, every such num-
per € is also a left-hand divisor of number a. By the way, it will be ob-
served that the formula obtained for a proves also that every ordinal
number a of the 2-nd kind may be represented in the form a= wf
where y=a,>0 and 6 is a number of the 1-st kind.
Now let 6 denote a left-hand divisor of number a such that B > w*
(which can occur only if a > w”). Thus there exists an ordinal number y
such that a= py. Let B = wd,+0%,4+...+a%, and Y == SC, 96,
+...+@”%c, be the normal expansions of numbers f and y. If we had
ys
> 0, then (cf. Exercise 4 of XIV.19) we should have py= oy =
= wo Me,4...+o%%e,, whence, in view of a= fy, (19.1) and of the
uniqueness of the normal expansion, 6,+y,;= a,, and thus, in view of

1) An analogous theorem was given by A. Lindenbaum for order types in Linden-

baum et Tarski [1] (p. 321, Theorem 14).
 § 21. Applications, of the normal form 329

ys > 0, By < ax; therefore 6 < m”, which is contrary to the assumption.
Thus y,;= 0 and number y is of the 1-st kind, whence (cf. Exercise 4
of XIV.19) we obtain a= py = w™*%¢,+ ote, +... + ot t"—I¢,_ 4 +
+ w'tb,¢, + wb, +... + we, whence, by comparison with the expansion
(Oe) ery ODUALL! Mie eG RN Opt teh Depa syeinpee Pp Sey U1Ce=
= Ok—pii, Jo = Op—ria; +. 0p = a. The natural number r may assume k
different values, and for a given r the numbers #,, f.,..., 8, and b.,
bs, ---, b, are well defined (by number a) while, in view of b,|a,_,41 num-
ber b, may assume only a finite number of values. Thus there exists only
a finite number (> 0) of left-hand divisors 6 of number a greater than w”.
Theorem 1 is thus proved.
THEOREM 2. For any ordinal numbers a, B and y the following ine-
quality holds (Shermann [1], p. 111):

(20.1) (a+ B)y <ay+ fy.

Proof. We shall distinguish two cases.

1) y is a number of the 2-nd kind, and consequently y = wé where
E>0. Let a= w%a,+...t 0%, and p= wd, +...+
0% be the normal
expansions of numbers a and f. Thus, we have aw = ww, whence ay=
=awé = ww = wy and similarly py = oy.
In view of Exercise 4 of XIV.19 we find that:
If a, > 6,, then the first component of the normal expansion of
number a+ f is wa,.
If a,= f,, then the first component of the normal expansion of
number a+ 8 is w™(a,+),).
If a, <p, then a+f= 6 and the first component of the normal
expansion of number a+ is w,.
We conclude that if a,>f,, then (a+ f)y= oy and if a < fi,
then (a+f)y= oy. Since both wo%y <ay+o%y and aity < w y+
+ oy, we always have formula (21.1).
2) y is a number of the 1-st kind, and consequently y= w£+k where &
is an ordinal number >0 and k a natural number. Thus we have ay =
awé+ak = w%w&+ak and similarly By = wwE+ Bk, whence ay + fy =
— wowé+ak+ wwe + bk.
On the other hand, we easily find that if a, > 6,, then (a+f)y=
= wwi+(a+f)k, and if a, <f,, then (a+f)y = w%wE+(a+f)k.
If a, > f,, then (a+ f)kK=a+f+a+f4+..t+a+8 = ok+8, there-
fore (a+ B)y = wtwE+ak+P = ay+6 <ay+ fy.
If a,=f,, then we have (a+f)k=a+f+a+/4+..4+44a4+f8=
=o" [ka, +(k—1)b,]+6, and if &>0, then ak+owé + Bk = wwé + Bk
and w™[ka,-+(k—1)b,] < w7wé. Since 8 < Bk, we obtain formula (21.1).
330 XIV. Ordinal numbers

If €=0, then ay+ By = ak+ pk = w"[ka,+(k—1)),]+ PB = (a+f6)k =

= (a+f)y, whence we again obtain formula (21.1).
If a, <f,, then (a+f)k = Bk, (a+f)y = ot wé+(at+f)k = otwé+
+ pk and ay+ py = wwé+ ak +o%wé + Bk = of wé + Bk; thus (a+f)y=
= ay+ Py, whence we again obtain formula (21.1).
Finally, for y = 0 formula (21.1) is of course true.
Theorem 2 is thus proved.
We shall give here one more application of the normal form of
ordinal numbers — to the proof of a certain theorem of Tarski [5] (p. 308-
309) of which we shall make use later on in XIV.28. Namely we shall
prove that we are able to define effectively a one-to-one correspondence
between ordinal numbers and ordered pairs of those numbers.
We shall define a function of two variables g(é, 7), satisfying the
following conditions:
(a) for any two ordinal numbers é and 7, ¢ = (&, 7) is a well-defined
ordinal number;
(b) for every ordinal number ¢ there exists one and only one ordered
pair of ordinal numbers & and 7 such that € = g(é, »);
(c) for every ordinal number a the inequality »(é, 7) < wt is equi-
valent to the system of inequalities € < w* and 7 < w*?).
Proof. Let € and 7 be two given ordinal numbers, & = w x,+
+0 ty +...+0%r, and n= woy,+o™y,+...+o"y, — their normal ex-
pansions. Let ¢, >¢, >... > ¢, denote a (finite) sequence formed of those
ordinal numbers which belong to one at least of the two sequences
E,, fo, ---) && and 7, Ho, ..-, m%. Numbers & and 7 may of course be written
in the form = o%m,+o%m,+...+ am, , = won, + wn, +... + on,
where m1, M2, ..-, M, and 1, Ng, ..., N, are well-defined (by & and 7) inte-
gers >0. Further, let f(m, ) denote a function establishing a one-to-one
correspondence between non-negative integers and ordered pairs of such
numbers, and let f(0, 0) = 0. (#. g. we can assume f(m, n) = 2"(2n +1)—1).
In order to obtain the function y(é,7) satisfying conditions (a), (b),
and (c) it suffices, as can easily be verified, to take

p(é, 7) = of (my, N) + wf (me, Ne) + oe oof (mr, Ny)

a closer investigation of which we leave to the reader 2).

1) A. Tarski, Comptes rendus de la Soc. des Se. et des Letires de Varsovie,

Cl. III, 19 (1926), p. 308-309.
*) If we wanted to obtain a function o(&, 7) satisfying only conditions (a) and (b)
we could define it more simply in the following way:

ae 25(2n+-1) if at least one of the numbers & and 7 is not finite,

PN\so oT)
2(2n+1)—-1 if E<o and n<o.
 § 21. Applications of the normal form 331

EXERCISES. 1. Prove that an ordinal number that is greater than 0 has a right-
hand divisor of the 1-st kind greater than 1 if and only if it is not a prime component.
Proof. Suppose that an ordinal number @ > 1 is a prime component. By The-
orem 1 of XIV.19 it is a number of the 2-nd kind. Suppose that number ¢ has a right-
hand divisor of the 1-st kind y+1> 1; therefore »y> 0 and there exists an ordinal
number f such that @ = B(y+1), and 6 > 1 since number o is not of the 1-st kind.
Hence @ = fy+ 6 > By > B (since y > 1), and thus 9 > fy and e > f, and the prime
component @ is the sum of two ordinal numbers smaller than itself, which is impossible.
Thus our condition is necessary.
On the other hand, if an ordinal number a > 0 is not a prime component, then,
by Theorem 1 of XIV.19, it is not of the form w> where € > 0, and then in the normal
expansion (19.2) we have either k = 1 and a, > 1, and number a = wa, has a finite
right-hand divisor a, > 1, or k > 1 and number a has a right-hand divisor of the 1-st
kind 17 %a,+ w%2”%*a,4+ ...4 @%-1~%q, _ ,+a,. Thus our condition is sufficient. The
required proof is complete.
2. Prove that if each of two given ordinal numbers is divisible on the right by
a natural number n, then their sum is also divisible on the right by n.
Proof. The proof follows easily from the theorem proved at the beginning of
XIV.21 and from the formula for the addition of normal forms of two ordinal numbers.
3. Prove that if an ordinal number y is >, then number y+y is not divisible
on the right by y.
Proof. If y is an ordinal number > q@ and if wc, is the first component of its nor-
mal expansion, then the first component of the normal expansion of number y+y is
@’12c,. If number y+ y were divisible on the right by y, we should have y+ y= &y
where € is an ordinal number > 0. But it is easy to find that the first component
of the normal expansion of number €y has the form w*c, where B > y,. In view of y+ y=
=&y and the uniqueness of the normal expansion of an ordinal number, we should thus
have 2c, = c,, which is impossible since ¢, is a natural number.
4, Give an example of two order types each of which is divisible on the right
by 2, but the sum of which is not divisible on the right by 2.
Answer. Such are the types 7 and 2.
5. Prove that if a and f are arbitrary ordinal numbers y >a and y=, then
a+B+y=B+at+y.
Proof. The proof is easily carried out with the aid of the normal expansions of
numbers a, 6 and y. ;
6. Prove that in order that an ordinal number & satisfy the equation wé = € it
is necessary and sufficient that number w® be its left-hand divisor.
7. Find all systems &, 7 of ordinal numbers such that &-.2 = 7.
Solution. Suppose that € and 7 are positive ordinal numbers such that &.2 = 7?,
and that & is of the first kind. Then 7 is also,of the first kind, and the numbers & and 7
cannot be finite (because V2 is irrational). Thus the normal expansions of &-2 and 7?
have at least three components and, since &:-2 = 7?, the number of their components
is equal. Comparing in these expansions the first, the middle and the last components,
we easily obtain a contradiction. Thus é is a number of the 2-nd kind and we have
&-2 = (&-2)2, 1. e. (E> 2)? = 7?, which gives &-2=7. Thus all positive ordinal num-
bers & and 7 such that &*- 2 = 7? are: 1 = €-2, where é is an arbitrary ordinal number
of the 2-nd kind.
332 XIV. Ordinal numbers —

8. Prove that there exist no transfinite ordinal numbers § and y such that =
=7>-+ 1.

Proof. See Sierpinski [71], p. 1-2.

ra
9. Prove that there exist an infinity of solutions of the equation é* = 1+7% for
ordinal numbers.
Proof. = 73, 7 = 7, where t is an arbitrary transfinite ordinal number, satisfy
our equation. There ,exist other solutions, e.g. § = w'+o?, 7 = w+, or = on+
+o"- 24 o"-2+2, n= o”+o"-24+ 2, where n= 1, 2,...

22. ‘Determination of all ordinal numbers that are prime factors.

Let «z denote a prime factor (see XIV.13) which is a transfinite number
of the 1-st kind. We have 7= wu+n where wu > 0 and n is a natural
‘number. Hence x= n(wu+1). If n>1, we should have z= wut+n>
>out+ 1 and the transfinite number z would be the product of two
numbers smaller than x, which is contrary to the assumption that it is
a prime factor. Thus 7 = 1, and therefore z= wu+1. But, as has been
shown in the proof of Theorem 1 of XIV.21, number wy, as a number
of the 2-nd kind (since uw > 0), is of the form amu = wf where y > 0 and 6
is a number of the 1-st kind. Thus we have z= w’S+1, whence, as we
know (Exercise 5 of XIV.11), it follows that ~=(w’+1)f (since £ is
a number of the 1-st kind). If 6 > 1, we should have a > ’+1. On the
other hand, «= o’f+1>o’f > B, whence a>. The number z=
= (w’+1)6 would thus be the product of two ordinal numbers smaller
than xz, which is contrary to the assumption of its being a prime factor.
Therefore 6 =1 and thus = w?+1.
We have proved that every prime factor that is a transfinite number
of the 1-st kind is of the form w?+1 where y is an ordinal number >0.
Now we shall prove that, conversely, every ordinal number of the form
w”’+1 where y > 0, is a prime factor (transfinite and of the 1-st kind
of course). Indeed, let @ = w’; by theorem 1 of XIV.19 it will be a prime.
component. But we. have proved at the end of XIV.13 that if @ is a prime ©
component then @+1 is a prime factor. Number w%+1 is thus a prime
factor, q.e. d. We have proved;
THEOREM 1. In order that a transfinite ordinal number of the 1-st
kind be a prime factor it is necessary and sufficient that it be of the form
w?+1 where y is an ordinal number >0.
Suppose now that an ordinal number a of the 2-nd kind is a prime
factor. As has been shown in the proof of Theorem 1 of XIV.21, if a is
a number of the 2-nd kind, then a = w’f where y > 0 and £ is an ordinal
number of the 1-st kind: If 6 > 1, we should have a> w’, but also a > B
since a= wf > B, the equality a = f being impossible if a is of the 2-nd
kind and £ of the 1-st kind. Number a = w’fS would thus be the product
/
 § 22. Determination of prime factors 333

of two ordinal numbers smaller than a, in contradiction to the assump-

tion of its being a prime factor. Thus 6 = 1 and therefore a= ow’. Num-
ber y must be a prime component since if y= ~+yv where uw < y and
y<y we should have a= ow’ = w/+*= w-@ where w4#< wY=a and
wo” < wo’ =a, and a would be the product of two ordinal numbers smaller
than a, in contradiction to the assumption of its being a prime factor.
Thus number y is a prime component, whence, by Theorem 1 of XIV.19.
it is of the form w* where & is an ordinal number >0. Hence a = ow” = ow”.
We have thus proved that every prime factor that is an ordinal
number of the 2-nd kind is of the form w®* where é is an ordinal number
>0. Now we shall prove that, conversely, every ordinal number of the
form w°’ where & is an ordinal number >0 is a prime factor (which is
of course a number of the 2-nd kind).
This is true for & = 0 because, as has been proved in § 12, number w® =
= w' = w is a prime factor. Therefore suppose that > 0: the number
A= of is thus of the 2-nd kind. ;
If number w®” = w’ were not a prime factor, we should have o*=
= py where uw < o and »< om, But if J is a number of the 2-nd kind,
then, as we know,
wo? = imw
o<A

and, in view of uw < w+, we cannot have uw > w for € < 4. Thus there exists
a number ¢, < A such that w < 1, and similarly we conclude that there
exists a number ¢, < 4 such that » < w. But, as we know from Theo-
rem 1 of XIV.19, number A = wé is a prime component: therefore if ¢, < A
and ¢, <A, then we must have ¢,+¢, <A; thus py < wm? = white < @),
in contradiction to the equality #4 = uy. Therefore number ow” is a prime
factor, q. e. d. We have proved
THEOREM .2. In order that an ordinal number of the 2-nd kind be
a prime factor it is necessary and sufficient that it be of the form wo” where E
is an ordinal number > 0.
Combining Theorems 1 and 2 we immediately obtain
THEOREM 3. All numbers of the form w’+1 where y is an ordinal
number +0 and all numbers of the form a” where € is an ordinal number > 0
are, besides finite prime numbers, the only ordinal numbers that are prime
factors.
This theorem immediately implies that among ordinal numbers
there exist infinitely many pairs of successive ordinal numbers which are
prime factors: they are the numbers w® and w® +1 where é is an arbitrary
ordinal number > 0.
304 XIV. Ordinal numbers

But if z is not a finite prime factor, 7+2 is not a prime factor, be-
cause if m= w’+1, we have 7+2= 3m and 3<a2+2, a<2+2, and
if «= w”, we have 7+2= 2(x+1) and 2<2”+2, a4+1<a2-+2.

EXERCISES. 1. Find the smallest ordinal number >1 that is not the sum of
a finite number of prime factors.
Answer. o?.

2. Find the smallest transfinite ordinal number a such that between a and a-2
there is no prime factor.
Answer a=q@-+1 since the next prime factor after w+ 1 1s w?+ 1 > (@+1)-2=
=@-2+1. This proves that the so-called postulate of Bertrand is not true for ordinal
numbers.
3. Prove that in order that an ordinal number be a prime factor it is necessary
and sufficient that it have two and only two different right-hand divisors.
Proof. Cf. the last section of XIV.13.

4. Prove that in order th t an ordinal number be a prime factor of the 1-st kind
it is necessary and sufficient that it have two and only two different left-hand divisors.
Proof. Suppose that an ordinal number z is a prime factor of the 1-st kind. Thus
it has at least two different left-hand divisors: 1 and z. If it also had a left-hand divisor
w~Al1 and w~2x, we should have 7 =~. If y= 2, we should have 1-7 = u-a, and
on account of the fact that z is a number of the 1-st kind and according to Corollary 3
of XIV.10 we should have 1 = uw, which is impossible. Therefore » 4 a and number a
is the product of two numbers u and vy smaller than 2, which is contrary to the assump-
tion that z is a prime number. Thus our condition is necessary.
On the other hand, if an ordinal number a has two and only two left-hand divisors,
then they are 1 and a (where a > 1). Then there is no decomposition a = wy where
“u <aand vy < a because in that case we should have «= 1, whence v = a. Number a
is thus a prime factor and it must be of the 1-st kind since otherwise it would have
infinitely many different left-hand divisors: 1, 2, 3,... Thus our condition is sufficient.

4. Find all ordinal numbers having three and only three different left-hand divisors.
Solution. As regards finite numbers, such are, as we know, only the squares of
prime numbers. Suppose that number a is transfinite and has only three different left-
hand divisors. Number a must be of the 1-st kind (since a number of the 2-nd kind
has infinitely many left-hand divisors, e. g. all natural numbers). Thus besides numbers 1
and a the number a has only one more left-hand, divisor, w. Therefore we have a = py
where 1 < uw < a. If number p were not a prime factor, we should have u = u,7, where
fy < wand », < mw; therefore | < uw, < mw and m, would be a left-hand divisor of num-
ber a such that 1 < uw, < u < a, which is impossible because a has only three different
left-hand divisors: 1, w and a. Thus number yw is a prime factor. If number y were not
a prime factor, then, in view of »y > 1 (since a = uv > pw), we should have v = u.¥, where
1 < py < v and, in view of a = wyyv,, number py would be a left-hand divisor of num-
ber a such that 1 < uw < wy, < a, which is impossible. We have proved that if a trans-
finite ordinal number a has only three different left-hand divisors, then it is of the 1-st
kind and is the product of two prime factors (of the 1-st kind of course).
Suppose now that a transfinite ordinal number a is the product of two prime
factors of the 1-st kind, a = 2,2,, at least one of the numbers z, and z, being transfinite
since such is number a. Number a has at least three different left-hand divisors: 1, z,,
 § 22. Determination of prime factors 335

and a. Suppose that number a has one more left-hand divisor mw, different from the other
three. Thus a = pr.
If x, is a finite number, i. e. a prime number p, then we have wy = pa, % = o?+1
where & > 0, therefore wy = ore p. Number w+ p, as we know from XIYV.21, does
not possess a natural right-hand divisor >1, and since vy > 1 in view of u +a, » can-
not be a natural number. 2-1 p being a number of the 1-st kind, vymust be a transfinite
number of the 1-st kind. If w were also a transfinite number (of the 1-st kind of course),
then the normal expansion of the product wy would have at least three components
which is impossible since this expansion is w°?-++ p. Therefore « must be a natural num-
ber. But number a+ p has (see Exercise 2 of XIV.11) only two natural left-hand
divisors, 1 and p, whence follows a contradiction since ~~ 1 and w~2,=p.
If a, is a finite prime number, z, = p, then wy= 1p, 1 = w+ 1 where €, > 0,
whence wv = wolp+ 1. Number wtp+ 1, as we know, does not possess a natural left-
-hand divisor >1, and since w 4 1, ~ must be a transfinite number (of the 1-st kind),
We conclude as above that number » cannot be transfinite because the normal ex-
pansion of the product uv = w*!p+1 has only two components. Thus » is a natural
number, and since number wp+ 1 has only two natural right-hand divisors, 1 and p,
therefore y= 1 of y= p. But if y= 1, a = wy would give «=a, which is impossible,
and if »y= p, we should have wp = 2,p, whence uw = xz, which is also impossible.
Finally let us suppose that numbers z, and a, are both transfinite, and since they
are prime factors of the 1-st kind, therefore 2, = wo +1, ii == w2+ 1, where &,>0
and & > 0; thus wy = w+ o t+ 1. This number has no left-hand or right-hand
natural divisors different from 1, and since « ~£ 1 and vy # I (in view of u ~ a), we con-
clude that numbers yu and v are both transfinite. Let u = w“!m,+ o?m,.+ ...4+ wo*—Im,_1 ++
+m, and vy = own,-+...+ 0"-In,_,+n, be their normal expansions. The normal ex-
pansion of number wy will thus be oT n,+ wMtT”n,-+ ...-+ Ml Tlg, + om, Ne+
+om,+-...+m,. By comparing both normal expansions of the product uv, we ob-
tain k+1—-1=3, i.e. k+l=4, and since k > 2 and 1>2 (numbers yw and » being
transfinite of the 1-st kind), therefore k= |= 2; further, the comparison of the normal
expansions gives w,+ 7, = 6+ &, w= &, m = Mn, =m, = 1. whence m, = m, = 4 =
= —= sls UNCLOLOT ONi — ot 1] = m,, Which is impossible.
We have thus proved that in each of the three cases investigated number a has
only three left-hand divisors. We have proved that in order that an ordinal number a
have three and only three different left-hand divisors it is necessary and sufficient that
it be either the square of a finite prime number or a transfinite number of the 1-st kind
which is the product of two prime factors (different or not), one of which may be finite.
Thus for instance the following numbers are ordinal numbers having three left-
hand divisors: o+ p, op+1, w?+p, w*p+1 where p is a finite prime number, as well
as (w+ 1)(@?+ 1) = w?+ 0+ 1 and (w?+1)(@+ 1) = w+ w+ 1.
5. For every natural number n give an example of an ordinal number a which
is the product ot two prime factors and has more than n different right-hand divisors.
Solution. Such is, for natural n, the number a = w”. Indeed, w” is the product
of two prime factors, w”~?+ 1 and », and, in view of w" = w"—*w" for k= 0,1, ..., 7,
numbers 1, w, w?,..., @", Which are all different from one another, are its right-hand
divisors.
6. Find all the ordinal numbers having three and only three different right-hand
_ divisors.

Solution. The necessary and sufficient condition for an ordinal number a to have
three and only three different right-hand divisors is rather complicated. We shall prove
336 XIV. Ordinal numbers

that in order that an ordinal number a have three and only three different right-hand
divisors it is necessary and sufficient that it be either the square of a finite prime number
or a transfinite number which is the product of two prime factors satisfying one of the
following two conditions: 1) the second factor is a number of the 1-st kind (finite or
transfinite), 2) both factors are of the 2-nd kind, the first being not smaller than the
second.
For the proof it is of course sufficient to investigate the case of number a being
transfinite. Therefore let us suppose that number a is transfinite and has only three
different right-hand divisor, i. e. besides the right-hand divisors 1 and a one more
divisor, v, different from the other two. Thus there exists a smallest ordinal number yp
such that a = wv. We have uw > 1 because vy £ a. If w were not a prime factor, we should
have w= .47,, where pw, < uw and », < mw, and a= ,,». If »»=», we should have
a = wy, which, in view of uw, < mw, is contrary to a definition of number uw. Thus »yv ¥ »,
therefore v,y > v, and on the other hand »y < a, since a= wy > vy, and the equa-
lity a = »» is impossible in view of », < w and the definition of number yw. Thus num-
ber a would have at least 4 different right-hand divisors 1, y, v,y, and a, in contradic-
tion to the assumption that it has only three right-hand divisors. Thus number yp is
a prime factor.
It is also easy to prove that number » must be a prime factor since otherwise it
would have a right-hand divisor 6 such that 1 < 6 < », and that divisor would also
be a right-hand divisor of number a such that 1 < 6 < » < a, which is impossible since a
has only three right-hand divisors: 1,» and a.
We have proved that if a transfinite ordinal number a has only three different
right-hand divisors, then it is the product of two prime factors. It is impossible simultane-
ously for the first of them to be a finite number and for the second to be a number of
the 2-nd kind since in that case number a would be equal to the second factor, i. e.
it would be itself a prime factor and therefore, according to Exercise 3, it would have
only two different right-hand divisors. Thus if the first factor is a finite number, then
the second is a transfinite number of the 1-st kind. Suppose now that the first factor
is a transfinite number of the 1-st kind, and the second a number of the 2-nd kind
not smaller than the first factor. By Theorem 1 and 2 we thus have a = 2,2, where
m%=ol+l, m= Boe &>0 &>0 and 2,<a2,; therefore w#!+1< on whence
wol< eae Wa Covey << w2, which, on account of the fact that number w*? — as a prime
factor — absorbs the preceding component &, smaller than itself, gives &+ wo?
= w®,
>) €2 é2 : = ‘ A
whence a = ™2, = (w+ 1) @°" = wt? — w?” = n,, which is impossible, since the
prime number z, has only two right-hand divisors. Thus if the first factor in the de-
composition a = 7,7, is a transfinite number of the 1-st kind, and the second a number
of the 2-nd kind, then we must have a, > a,. We shall also prove that 2, < 22. Indeed,
*s ay &2. : : €2.
suppose that 2,>73. Thus wot+t1> wo”, whence w! > wo” (since if w! < ow,
we should have w!-+-1< we?) which gives &, > w*?-2; therefore £,— w*?-2+-5 where
56> 0. Thus we should have a = pfitotee —- ofte+d+o% 514 number p= om tetomt
would be a right-hand divisor of number a such that 2, < » < a; number a would thus
possess at least four different right-hand divisors, 1, z,,», and a, which is impossible.
é é2. ¥
Therefore 2, < 2, < 22 and thus o°" <@!+1<o 2 whence Roe <@ml< oot 2
and ow? < &, < w-2. Therefore &,= w°!+7 where 0<1t < w. If €& = 0, we have
0<+t< 1, whence t=0 and t+ ow? = w? and if €& > 0, we also have t+ we? = wm?
in view of t < wo and by Theorems 1 of XIX.14 and 1 of XIV.6. Therefore &,+ o =
 § 22. Determination of prime factors Goospa

és a 7 5
= wo?+ 7+ wo = w?-2, and thus a= aa = who = mito? — we? — zz. Condi-
tion 2) is satisfied.
Finally let us suppose that both factors, a, and 2,, are of the 2-nd kind and that
é1 é2 Ey §2 :
Ty < %,. We have 2, =, 1 =o” , OX €, < &, and wo” < w””, whence wo! < w®
E ea ase E14 gySe €2 Fe Pe |
and w*!+ w= w*; therefore a = 12,= 0° w= ot?” = w®” = ay, whichis impos-
sible. Thus if both factors, a, and z,, are of the 2-nd kind, then we must have 2, > a,
2. e. condition 2) must be satisfied.
We have proved in this way that if a is an ordinal number having only three
different right-hand divisors, then it is the product of two prime factors satisfying one
of the conditions 1) and 2). Thus in order to prove our theorem it remains to show that
if a is a transfinite ordinal number which is the product of two prime factors satisfying
any of the conditions 1) and 2), then number a has three and only three different right-
hand divisors. Therefore let us suppose that a = 2,72, where 2, and a, are prime factors
satisfying condition 1). The second factor, a, is thus a number of the 1-st kind. If z,
is a finite number, then 2, = p where p is a prime number and, since a = 2,7, is a trans-
finite number, z, must be a transfinite number. If z,, is a number of the 1-st kind, then
we have xz, = w*!+ 1 where &, >-0; therefore a = a,p = op + 1. This number, as we —
know, has no natural right-hand divisors other than 1 and p. If number a had, besides
1, p and a, another right-hand divisor y, then » would have to be a transfinite number,
of course of the first kind (since such is number a). We should have a = wy. Let w=
=o m, + ..+o"m, and vy = w'n,+...+@"m, be the normal expansions of numbers yz
and »v. By comparing the normal expansions of numbers wv and a we obtain k+1J—1= 2.
But 11> 2 since y is a transfinite number of the 1-st kind. Hence k = 1, therefore uw =
= olm,. We cannot have yw, > 0 since a = wy, i. e. wis also a number of the 1-st kind.
Thus ~,=0, whence w= m, and p is a natural number. But number a= op+ 1
has no natural left-hand divisor >1. Therefore « = m, = 1, and thus, in view of a=
= wv,v =a, which is contrary to the assumption concerning number v. Now if x, is
a number of the 2-nd kind, then we have z, = ow? where & > 0, therefore a= p=

= oy, If number a had a right-hand divisor y different from 1, p and a, then we

should have a = uy and, the normal expansion of the number a having only one com-
ponent, from the normal expansion of the product of two ordinal numbers it follows
that the normal expansion of » has no more than one component. Thus v = m”!n,, and,
for w= wo'm,+...+0%m, we have wy = o”!T"m,, and by comparing the normal expan-
. . E. . . .
sions of numbers a and wr we obtain n, = p and u,+», = w*; since w*! is a prime com-
ponent, yp, = 0 or»,=0. In the first case we have »,= w*! and y= 0p,
therefore v = op =a, which is impossible. In the second case we have y, = 0, there-
fore y = p, which is also impossible. We have thus proved that if 2, is a finite number,
then number a has only three right-hand divisors: 1, 2, and a.
Therefore let us suppose that x, is a transfinite number (of the 1-st kind). Thus
Ro wet 1. If a, is a finite number, 2, = p, where p is a prime number, then a =
= 7,1, = w°?+-p. Suppose that v is a right-hand divisor of number a different from 1,
zt, and a. Thus we should have a = wr. By comparing the normal expansions of num-
bers a and yy we obtain k+1—1 = 2, and since /> 2, number a = w+ p having no
finite right-hand divisors >1,' we have k= 1 and uw must be a finite number since
otherwise it would be a number of the 2-nd kind, which is impossible as number a is
of the 1-st kind. Therefore 4 = m,. But number a = w+ p has no natural left-hand
divisors other than 1 and p. If w= 1, then a=», which is impossible, and if u = p
then a =p (w?+ 1) = py, whence » = o?+1= 2, which is also impossible.
Cardinal and ordinal numbers boins)
338 XIV. Ordinal numbers

We have proved that if 2, is a finite number and a, a transfinite number of the

1-st kind, then number a has only three right-hand divisors. Finally let us suppose
that 2, is a transfinite number and a, a transfinite number of the 1-st kind. If a, is
a number of the 1-st kind, then 2, = w°!-+1, a, = w*?+1 where é,> 0 and & > 0,
and if a = wv where » is different from 1, 2, and a, then by comparing the normal ex-
pansions of numbers a = 2,7, = wits 1 ot 1 > mz, and my we obtain k+1/—1= 3.
But number a, as can be seen from its normal expansion, has neither left-hand nor right-
hand natural divisors >1: thus numbers « and y must both be transfinite because if
y #1 and »y ~<a, then mw ~ 1; they must also be of the’ 1-st kind since a is of the 1-st
kind. Thus we have k > 2 and 1> 2 and
gives k=1= 2. the equality k+1]—1=3
We have pr = on, + m,n, ms, Whence, by comparison with the normal ex-
pansion of number a, w#,+%=6+&, m= &; therefore », = &, m,n, = 1, whence
n, = 1, and thus »y = w+ 1 = a,. Now, if 2, is a number of the 2-nd kind, then 2, =
1
= , %=ow?+1 and if a= pw, then, in view of a = wets +@®' we obtain k+
~W
Ww

--1—1 = 2. Since number a has neither a left-hand nor a right-hand natural divisor > 1,
therefore, if y A 1 and yA a i.e. wu£1, then numbers mw and y are transfinite. If we
had 1 — 1, number » would be a power of number @ with a positive exponent and the
number «wv would also be a power of number w, which in view of uw» = a is impossible.
Therefore | > 2 and from the equality k+1/—1= 2 it follows that k= 1 and l= 2.
Hence pr = wo!m,(@"'n, + ny) = 0! "In, + co! m,n, and by comparison with the normal
expansion of number a we obtain m,+ », = wi+té,, Mi, w'', therefore v, = &; further
m, = 1 and m,n. = 1, whence n, = 1, and thus »y = w*?-- 1 = z,. Thus in any case num-
ber a has only three right-hand divisors.
We have proved that if a is an ordinal number which is the product of two prime
factors satisfying condition 1), then a has only three different right-hand divisors.
Suppose now that a= 2,2, where a, and z, are prime factors satisfying condi-
: Ey &2 4 oe Me
tion 2). Therefore m= 0° ,7%,=@0° , m =a, and thus & > & > 0. If number a —
fe
mel +
E

=o “had one more right-hand divisor, besides the right-hand divisors 1. a and a,
E14 $2 :
then we should have m® °°” = wy, whence we easily conelude that » must be a power
of number w, vy = w1 and we obtain w®'*t ieee wt, whence w!-+ wo? — 4, +»,. Num-
ber », is thus the remainder of the number w*'-- w*? where &, > &,. But the latter, as
follows from Theorems 4, p. 283 and 1, p. 320, has only two different positive remainders:
wo? and w*!-- w?. Thus either », = o*? whence »y = w’! = 2,, which is impossible, or
vy, = w8'-+ w, whence » = w = a, which is also impossible. Thus we have proved that
if a is an ordinal number which is the product of two prime factors satisfying condition 2,
then a has only three different rnght-hand divisors.
Our theorem is thus proved.
7. Find all the ordinal numbers giving three and only three different decomposi-
tions into the product of two ordinal numbers.
Answer. Obviously they are all the ordinal numbers having three and only three
left-hand divisors (7. e. the numbers determined in Exercise 4).
8. Prove that if an ordinal number has three and only three left-hand divisors,
then it also has three and only three right-hand divisors but not vice versa.
Proof. The proof immediately follows from Exercises 4 and 6. 5
9. Give an example of a transfinite ordinal number having the same three and
only three left-hand and right-hand divisors.
Answer. Such is, for instance the number (m+ 1)?. In general, they are numbers
(w+ 1)? where & > 0 and only such numbers.
 § 22. Determination of prime factors 309

10. Find for what transfinite ordinal numbers a the numbers 2*+ 1 are prime.
Solution. For ordinal numbers a of the 2-nd kind and only for such transfinite
numbers a. Indeed, if a transfinite ordinal number a were of the 1-st kind, we should
have a = w&+k where & is an ordinal number greater than 0 and & a natural number.
Hence 2° = 2°+* — (2)°.9* — m>.2*, therefore 2°+1=o*- 9°41 1 = (w+ 1)-2",. and
thus the ordinal number 2%+ 1 is the product of two ordinal numbers smaller than
that number, i. ¢.it is not a prime number. But if a is a number of the 2-nd kind, then
ce £ ' F ; 5
a = wé where — > 0 and we have 2°+ 1 = 27% +1 = w* +1, which is a prime number-
Note. It is much more difficult to find for what finite ordinal numbers a the num-
bers 2*-- 1 are prime numbers. We know only five such numbers a, namely 0, 1, 2,
4,8 and 16, and we do not know whether there are more of them.
11, Ordinal numbers of the form F = 2%*+ 1 are called Fermat numbers. Prove
that every Fermat number with a transfinite index is a prime number.
Proof. Ifa is a transfinite number, then a = w+ & where = O and therefore the
number 2% = 2°°5 — 2°.2° — w- 2° is of the 2-nd kind. According to the preceding exer-
cise, number fF, = 2%*+ 1 is a prime number, q. e. d.
Note. As regards numbers F, for finite a we know that they are prime for a =
0,1,2,3,4 and composite for 4< a< 12. We do not know whether numbers F,,;
and F,, are prime or composite. Numbers /’,, and F’,, are composite.

Now we shall prove

THEOREM 4. Kvery ordinal number has at most one transfinite right-
-hand divisor which is a prime factor.
Proof. To begin with, we shall prove that an ordinal number having
a right-hand divisor which is a prime factor of the 2-nd kind cannot
have a right-hand divisor which is a transfinite prime factor of the 1-st
kind. Indeed, if a = fa = yx’ where x is a prime factor of the 2-nd kind
and z’ — a transfinite prime factor of the 1-st kind, then, by Theorem 3,
a= ow and a’ = w¥+1. Let p= wd, ate oe +!" b, and y= w¢4+...4+
+o"c, be the normal expansions of numbers f and y. We shall have
Br = wit and yr’ = (we, +... + we.) (@” +1) = w+ o%e, +... +0",
both expansions being normal. Thus in view of the uniqueness of the
normal expansion of an ordinal number, we cannot have Ba = yz’.
For the proof of Theorem 4 it suffices to show that an ordinal number
cannot have two different right-hand divisors which are prime factors
of the 2-nd kind, nor can it have two different right-hand divisors which
are transfinite prime factors of the 1-st kind.
Therefore let us suppose that a= pa = yx’ where x and z’ are prime
factors of the 2-nd kind, 7. e. = wo, 2’ = w®*”. Denoting as above the
normal expansions of numbers 6 and y, we shall have Ba = white, Vi =
=quto whence, in view of px= yn’ we obtain f,+of = »,+ 7.
Numbers o* and w" are thus remainders of the same number; by Theo-
rem 3 of XIV.5 one of them is thus a remainder of the other. But, by
Theorem 1 of XIV.19, numbers o£ and w” are prime components, 7. e.
22%
340 XIV. Ordinal numbers

they have no positive remainders different from those numbers them-

selves. Hence it follows that wf = o, 7. e. also 7= a7’.
Finally, let us suppose that a = Ba = ya’ where a and z’ are trans-
finite prime factors of the 1-st kind, therefore z= w*+1, a’ = w+,
and let us denote as above the normal expansion of numbers f and y.
We should have f(@m'+1)= y(@"+1), whence pof+ fp= ywt+ y; there-
fore wt + wth, + ob, +... + 0, = wo" +4 we,+...+0%c,, both sides
representing normal expansions. In view of the uniqueness of the normal
expansion of an ordinal number we conclude that 6,+é= y,+ 7 and
y= yas sWOMCEsS = nd enor ar.
Theorem 4 is thus proved.

23. Expanding ordinal numbers into prime factors. By theorem 1 of § 13

every ordinal number > 1 is the product of a finite number of prime
factors. We shall show now how to deduce such an expansion of
a g:ven ordinal number a > 1 directly from its normal expansion (19.1).
It’ is easy to verify that we have, for k>1, w%a,+7a,+...+
0%a, =
= (@ "Ao +O 0g-- 4. @ O,)(@ > 2-1 )yay since (@. 0-4 La, 0) “aa,
and! @ 205450 0.-- 9 Oy) Or mae Oe
Similarly, for k > 2 we have
a 3 ; a, : =O
@ "Ay 4+ Og o.. + @ "Oy = (@ 2A, +e waz) (oo? 1) a,
etc., and finally
—ak
Ol) hier aa way, == Or UA =. ate 1) G4

Hence follows the formula

(2381) 9G =i) O(a AD a or es re

In order to obtain from formula (23.1) the expansion of an ordinal

number a into prime factors, we should moreover expand into prime
components the natural numbers a; (1 = 1,2,...,k) (provided they are
greater than 1), arranging them for example according to non-increasing
magnitudes, and the number w* (by decomposing into prime compo-
nents the exponent a,). If a, = 0, then formula (23.1) immediately gives the
expansion of an ordinal number a (of the 1-st kind) into prime factors,
which are then all of the 1-st kind (finite or transfinite). If a, > 0 and
a, = om +o0my,+...+ 0°"m is the normal expansion of number a,;,
then in view of (23.1) we obtain

(23.2) (OS) (OPA A (OP

eta (er 2m ala eee cnet aed Nice

whence we immediately get the expansion of number a into prime factors.

We have “4, > fy > ... > fn. By Theorem 4 of XIV.22 we can easily prove
that an expansion into prime factors (in which prime factors of the 2-nd
 § 23. Expanding into prime factors 341

kind, if they exist, follow one another in order of non-increasing magnitude

and precede all prime factors of the first kind, and in which the successive
finite prime factors are arranged in order of non-increasing magnitude)
is unique (Jacobsthal [2], p. 184, and Aigner [1], p. 297-299) and identical
with the expansion obtained by determining for the number a the smal-
lest of its right-hand divisors greater than 1, which. we denote by z,
writing a = a’x where a’ is the smallest ordinal number & such that a = &z,
and then proceeding with the number a’ in the same way as with a, etc.
If, however, we make no restrictions as to the succession of prime
factors, then in general an expansion of an ordinal number into prime
factors is not unique (Sieczka [1], p. 176), since, for instance, w? = aw =
== 0) 00) == Go Bo Zo () = (= NOR2h GS? == arom = (==if Wena == (@2 ==1) oy

EXERCISES. 1. Decompose into prime factors the following ordinal numbers:

a) ¢, b) @ + @+4+ 1, ¢) ow +4 2, d) wo+o-2+], e) w+ w+ 1, f) wo +? 241, 2) ot
+. >: 2+ @?: 6.

Answer. a) Since « = w° we have ¢ = w® , whence it follows that number « (and

similarly every other epsilon-number) is a prime factor, b) m+ m+ 1 = (m+ 1)(m?+ I),
ec) a@+a+ 2 = 2(Ma+ 1)(w?+ 1), d) w+ o-24 1= (m+ 1): 2(@?+ 1) e) o+oa?4+1=
=(e~?+1)(o+1), f) w8+o?:24+1 = (m+ 1):-2:(@+1), 2) w+ @m?+24 w?-6 =
= ww: 3- 2(w+ 1): 2(w?+ 1).
2. Prove that every transfinite ordinal number of the 1-st kind has at most one
left-hand divisor which is a transfinite prime factor (of the 1-st kind of course) and
give an example of an ordinal number of the 2-nd kind having infinitely many left-
hand divisors which are transfinite prime factors of the 1-st kind.
Proof. Suppose that a and a’ are two different transfinite prime factors which
are left-hand divisors of an ordinal number a of the 1-st kind. We have a = af = a’y
and, a being a number of the 1-st kind, numbers a, 2’, f and y must also be of the 1-st
kind. Therefore we have B = w&+k, y = wé&’+k’ where > 0, & > 0. k and k’ being
natural numbers, and by Theorem 3 of XIV.22 we have m= m+ 1, 2’ = w’+1 where
“w>O0 and »>0. Thus 26 = (w*+ 1)(@€+ kh) = wo" Tet @“k+1 and similarly 2’y =
= ~"**y+ w'l4+- 1, whence we easily conclude that the last two components of the nor-
mal expansion of the number wf are mk and 1, and for the number z’y, they are o'l
and 1. From the equality au = a’y and the uniqueness of the normal expansion of an
ordinal number it follows that wk = w'l; therefore 4 =v, whence a= 2’, which is
contrary to the assumption that aA a’. x
An example of an ordinal number of the 2-nd kind having infinitely many left-
-hand divisors which are transfinite prime factors of the 1-st kind is given by the number
ow, because w® = (wo +1)H® for natural n, the numbers wm’+1 (n= 1,2,..) being
different transfinite prime factors of the 1-st kind.
3. Expand into prime factors according to formula (23.2) the number (w® + l)w””
(which itself is the product of two prime factors).
2 2, 2 2
ACA Siac (ls) Omi Or.

24. Roots of ordinal numbers. If for an ordinal number a and an

ordinal number £ there exists an ordinal number f > 1 such that & = a,
then number é is termed a root of number a. The necessary and sufficient
342 XIV. Ordinal numbers

conditions for an ordinal number & to be a root of a given ordinal num-

ber a and the properties of roots have been investigated by Ph. W. Car-
ruth [1], (p. 470-480).
In particular, if for given ordinal numbers a and f/ the ordinal num-
ber & satisfies the equation é*=— a, then it is termed a f-th root of
number a.
It is easy to prove that if a is an arbitrary ordinal number and p —
an ordinal number of the 1-st kind, then there exists at most one B-th root
of number a. For if B= y+1 and é<é, then = &= FE < HEX
< &&, = & += &, whence & < &. But if 6 is a number of the 2-nd
kind, then there may exist infinitely many /th roots of number a. F. g.
since n® = q@ for n= 2,3,..., every natural number >1 is an w-th root
of number w. In Chapter XV.5 we shall prove that there exist ordinal
numbers a having non-denumerably many m-th roots.

EXERCISES. 1. Prove that if 6 is an ordinal number of the 2-nd kind, 6 = wy,

then in order that there be at least one 6th root of an ordinal number a > 1 it is neces-
sary and sufficient that a = w” or that a be a power of number , with an exponent >0,
divisible on the right by f, and that then there exist infinitely many /th roots of
number a.

Proof. Let a denote an ordinal number > 1, 6 = wy. where y is an ordinal number
>0O and suppose that there exists an ordinal number € such that &P =a. In view of
a> 1, we must have > 1. If € is a natural number then, in view of € => 2, E°= o,
therefore g = &? = &°” = @’.. If € is not a natural number, then > w. Let w lx, denote
the first component of the normal expansion of number €; we shall have &, > 0. Hence
«8 <M) & 4 !& .
we obviously have &°e — w*!, therefore a = &eB gO” _. (@51)? — @?, Thus our condi-
tion is necessary.
On the other hand, if a = m’, then for every natural number n > 2 we have n° = o,
therefore n? = n °Y — ~” — a and every natural number n > 2 is a B-th root of number a.
If a=", where rt > 0, then for every natural number n we have (@ ny oS gp TnMOY see
== (1) as oY
— ~? — g, the numbers w™” (n = 1, 2, ...) being all different from one another,
and again there exist infinitely many $-th roots of number a. Thus our condition is
sufficient and the required proof is complete.
2. Determine all m-th roots of number ow”.

Answer. They are all ordinal numbers € such that wo < & < w”.
3. Find the smallest ordinal number >1 of which there exists an (w+ 1)-th root.
. =i
Answer. It-is the number 2°° = - 2.

4. Find the smallest transfinite ordinal number of which there exist no roots of
degree > 1.
Answer. w+. Indeed, on one hand, 2° = wm. On the other hand, if for a certain
ordinal number f > 1 we had e? — w+ 1 and it — were a finite number, then we should
have > 2 and f > @ (since n® = w < w+1 for n=2,3,...), whence & > 9¢+? —
=o-:2 >o+1, which gives a contradiction. Therefore we should have €>o, and
gh 10)" > w+ 1, whence we again obtain a contradiction. Thus there are no roots of
degree P > 1 of number m+ 1, q. e. d. 5
 § 24. Roots 343

5. Of what degree >1 are there roots of number e?

Answer. Only of degree ¢: according to theorem 1 of XIV.20 every ordinal num-
ber € such that 2< é < « is a root of degree ¢ of number e.

25. On ordinal numbers commutative with respect to addition. Two

ordinal numbers a and f are said to be commutative with respect to addi-
tion, or additively commutative if
(25.1) a+Pp=f+a.

THEOREM 1. Jn order that ordinal numbers a and 6 be additively com-

mutative it is necessary and sufficient that there exist an ordinal nwmber &
and non-negative integers k and | such that

(25.2) EK, (OP == El

Proof. We shall say that two ordinal numbers a and / satisfy condi-
tion C if there exist an ordinal number & and non-negative integers k
and J for which formulae (25.2) hold.
If ordinal numbers a and f satisfy condition C, then, in view of the
distributivity of multiplication of ordinal numbers with respect to addi-
tion we obtain, by (25.2):

a+ p= &k+ = (k+l) = E(l +h) = dl + &k = Ba,

which proves that the condition of our theorem is sufficient.
In order to prove that it is also necessary let us suppose that there
exist pairs of ordinal numbers a, /, satisfying formula (25.1) but not
satisfying condition C. Among such pairs there must exist at least one
pair, a, 6, for which the sum a+f= 6-+a is the smallest. On account
of the symmetry of conditions relating to numbers a and f$, we can sup-
pose, without loss of the generality of the proof, that a <#. Thus there
exists an ordinal number y > 0 such that 6=a-+y and, by (25.1), we
have a+(a+y)=(a+y)+a=a+(y+a), whence a+y=y-+a. If there
existed an ordinal number & and non-negative integers k, and 1, such
that a= &k,, y= él,, we should have f=a+y= &k,+ él, = €(4,+4),
in contradiction to the assumption that the pair a, does not satisfy
condition C. Therefore the pair a,y cannot satisfy condition C and it
follows from the definition of the pair a, f that a+y >a+f/. But a=
— 0 is impossible, because the pair a, 6 does not satisfy condition C. Thus
a>0,and B=a+y>a+f=f8+a> 8, which is impossible.
We have proved that the condition of our theorem is necessary. Thus
Theorem 1 is proved.
Knowing the normal expansion (19.1) of an ordinal number a > 0
we shall now determine the smallest ordinal number f > 0 additively
commutative with the number a. From Theorem 1 it immediately follows
o44 XIV. Ordinal numbers

that it will be the number $= & where & is the smallest ordinal number
such that for a certain natural k’ we have a = ék’. The number é will thus
be the smallest left-hand divisor of the number a for which the comple-
mentary right-hand divisor of the number a is a natural number. And
since, as we know from XIV.21, the greatest natural right-hand divisor
of the number (19.1) is a,, this number é will be determined from the
equation a= éa,, whence it is easy to find that &= o1+o7%a,+...+
ao Oe. <
Thus: Lf the normal expansion of an ordinal number a> 0 is (19.1),
then the smallest ordinal number a, > 0 additively commutative with num-
ber a 18 a) = @'+07%0,-.. + otae
Let us now determine all ordinal numbers 6 > 0 additively com-
mutative with a positive number (19.1). By Theorem 1 they will all be
numbers f= él where | is a natural number and é a left-hand divisor
of number a for which the complementary right-hand divisor is a na-
tural number, ¢. e. a divisor of number a,. Thus in order to determine
number é we must solve the equation a= &k’ where k’|\a. If a, = k’ky,
then we easily obtain §= w%ki+o"a,1+...+o%a,, therefore 6 = l=
=on+wa,t+...t@%*a, where n= kil is a natural number. On the
other hand it is easy to verify that for any natural » number $ = w%n +
+ wa,+...+ 0a, is additively commutative with number a. It follows
that im order that an ordinal number 6 >0 be additively commutative
with number a > 0 having the normal expansion (19.1) it is necessary and
sufficient that B = w n+ w®a,+...+ o%a, where n is a natural number.
Thus we have 6 = ayn and, conversely, for any natural n the number
fb = aon is additively commutative with the number a. Thus we have
THEOREM 2. If a is an ordinal number >0 and ay denotes the smallest
ordinal number > 0 additively commutative with number a, then every ordinal
number p additively commutative with a is of the form 6 = aps where s is
an integer >0, and vice versa.
From Theorem 2 immediately follows
COROLLARY 1. For every ordinal number a> 0 the set of all ordinal
numbers additively commutative with number a is denumerable.
From Theorem 2 we can also easily deduce
COROLLARY 2. Jf a, B and y are ordinal numbers such that a 40,
a+f=B+a and a+y=y-+a, then B+y=y+6.
Proof. If a, 6 and y are ordinal numbers such that a> 0, a+ p=
= f+a and a+y=—y-+a, then, by Theorem 2, there exist integers >0,
s and r, such that 6 = ays, y= ayr, whence B+ y = as+apr = a(s+r)=
‘= ao? +8) = apr + as = y+, whence B+y= y+, q.e. d.
From Corollaries 1 and 2 immediately follows
 § 25. On ordinal numbers commutative with respect to addition 345

COROLLARY 3. Positive ordinal numbers decompose into disjoint denumer-

able sets such that two positive ordinal numbers are additively commutative
if and only if they belong to the same set.
The additive commutativity of order types has been dealt with by
N. Aronszajn [1] (p. 65-96).
THEOREM 3. In order that positive ordinal numbers a and B be additively
commutative it is necessary and sufficient that there exist natural numbers m
and n such that
(25.3) am pn «

Proof. Suppose that a and f are positive ordinal numbers, additively

commutative. By Theorem 1 there exist an ordinal number € and non-
-negative integers k and / for whichformula (25.2) holds, numbers k and J,
in view of a>0 and fp > 0, being natural. In view of (25.2) we obtain
al = kl = lk = Bk, which proves the necessity of the condition of Theo-
rem 3.
Now let a and f be positive ordinal numbers and suppose that there
exist natural numbers m and » for which formula (25.3) holds. Number a
is known to be additively commutative with number am, and therefore,
in view of (25.3), also with number fn, while the latter is additively com-
“mutative with number #. According to Corollary 2 of Theorem 2 we thus
have a+ f= f+a. The condition of Theorem 3 is thus sufficient. Theo-
rem 3 is proved.
EXERCISES. 1. Find all ordinal numbers additively commutative with number lL.
Answer. Such are all finite ordinal numbers.
2. Find all ordinal numbers additively commutative with number o.
Answer. Such are all numbers of the form m/ where J is an integer > 0.
3. Find all positive ordinal numbers additively commutative with number w+ I.
Answer. Such are all numbers of the form w/+ 1 where J is a natural number.
4. Find all ordinal numbers additively commutative with number e.
Answer. Such are all numbers of the form el where | is an integer > 0.
5. Prove that in order that the sum of three positive ordinal numbers a,, a and ay,
be independent of the order of the components it is necessary and sufficient that there
exist natural numbers k,, hk, and k, such that a,k, = a,k, = as3k,.
Proof. Let j,, j,, 73 denote any permutation of numbers 1, 2, 3. If the sum of
numbers a,, a, and a; does not depend on their order, then a;,+ (a, + = a, (a, + d;,)s
which gives a, + a), = G+ a, hence each of the numbers a,, a, and a; is additively com-
mutative with each of the other two numbers. By Theorem 3 there exist numbers m, n, p
and g such that a,m = a,n and a,p = a3q, whence a,mp = a,np = ag,nq, which proves
the necessity of the condition.
On the other hand, if there exist natural numbers k,, ky and k, such that a,k,=
= ak, = a,k,, then it follows from Theorem 3 that each of the numbers a,,; a), a3 is
additively commutative with each of the other two numbers, whence it casily follows
346 XIV. Ordinal numbers

that the sum of numbers a,, a, and a, is independent of their order. Thus the condi-
tion is sufficient.
6. Let » denote a natural number >1. Prove that in order that the sum n of
positive ordinal numbers a,, @,..., a, be independent of the order of those numbers
it is necessary and sufficient that there exist natural numbers k,, k,.,...,k, such that
Thy == C1) = oo == Cla bic
7. Prove that Corollary 2 to theorem 2 is not true for order types.
Proof. If a=1,.8 = o+o*, »y=o+1+o*, then a+ $= f+a,a0-y—y+7a@
but B+yAy+ fp.
8. Prove that an ordinal number a > 0 is not additively commutative with any
ordinal number >0 smaller than a if and only if number a has no natural right-hand
divisor greater than 1.
9. Prove that if a is an ordinal number > 0, then all ordinal numbers > 0 additively
commutative with number a form an increasing transfinite sequence of type @ whose
limit is number aw.
Proof. It is easy to prove that if (19.1) is the normal expansion of number a > 0,
dg
then the sequence in question is the sequence of numbers w n+ w@d,+ w7dg-+ ...
l . . - = Es |
+ oa, where n = 1, 2,..., whose limit is the number m2" = aw.
10. For an ordinal number a > 0 find the smallest ordinal number f such that
number a is not additively commutative with any ordinal number => f.
Answer. It easily follows from Exercise 9 that this number is 6 = ao.
11. Prove J. Stupecki’s theorem stating that if m is an infinite order type, then
there exist order types w and » such that p= pwtvAv+u.
Proof. If m is a transfinite ordinal number, then, as we know, y = 1+ , and,
according to Exercise 1, we have 1+ 9 4 y+ 1; thus we can assume that uw = 1,vy= 9.
If y is not an ordinal number, then ~ is the order type of an ordered set /,, which is
not well-ordered, 7. e. it contains a transfinite sequence of type m*. Denoting by A the
set of all elements of the set / that precede every term of that sequence and writing
B = F—A, we shall obviously obtain a cut fF= A+B of the set F in which the set B
has no first element (and the set A may be empty). Let 4 =a, B = f. If the type @
has a first element, then of course gy= a+/ + B+ a, since then a has a first element
while £ has not. And if y has no first element, then let a denote any element of the set F,
let us denote by WV the set of all elements of the set / that precede a and let VN= F—M.
The set J/ will thus have no first element and the set N will have such an element,
viz. a. Writing M = yw, N =», we shall thus have g=putyvA~Ar+ypm. We have proved
J. Shupecki’s theorem.
12. Give an example of two ordinal numbers a and f such that a®+ p° = p®+ | a®
Bee ie eRe n Ce pee ¢
butcesepe- p +a for n=1,2,...
:
ADswer. a=, Bp = w*;2. for
¢ we have here a® = f° = w® anda +f =o 2 ° <o an,
+ i]
ae aK z= p Bt ee

13. Prove that if a and f# are ordinal numbers >1 such that a+ 6 = P+ a, then
a® = 6° but not vice versa.
Proof. If a and f are natural numbers >1, then we have a® = f° = o. If a and B
are transfinite ordinal numbers with the normal expansions a= w“a,+... and Bp=
8. os :
= w'b,+... and if a+ f= fB+a, then we must have a, = f,, therefore a? = 0% =
= wi — B°, Now, w® = (w)® but oto? < w+o.
 § 25. On ordinal numbers commutative with respect to addition oad.

14. Prove that if a and f are transfinite ordinal numbers of the 1-st kind and if
for a certain natural m > 1 we have a”+ fp” = p”+ a”, then a= f.
Broof, Wet a= w4a,4-...-- wk la, +a, and p= @ Lb ack i | ene b, be
the normal expansions of numbers a and $. Thus we have a, > 0, B;> 0 (since a and P
are transfinite). It is easy to calculate that for a natural m > 1 the sum of the first k
terms of the normal expansion of number a” will be
ay(m—1)-+ ay ru” =D) et the, ay("—1)
Oa eo a+... jo A,Uy,

and the sum of the first | terms of the normal expansion of number ~”ym will be
m B —1)+-£8 5. —1)+ 7. 34(m—1
wht b,+ oi—Y 2 ae lee ai) rPr_ip, Sain phi” dh, Ds

the expansion of number a” having mk—m-+ 1 components and the expansion of num-
ber 6”, ml—~m—1 components. If a” + p” = p”"+ a”, then the normal expansions of num-
rs a” andand p”
bers p° maymay differ
differ atat most
most in ththe coefficient
ff t of the
of the highest
highest power
er of of number
numbe w;
thus mk—m+1= ml—m-+1, whence k =I; further a,m = f,m. whence a, = f,,
and a,(m—1)+ a, = B(m—1)+, whence, in view of a, = a we obtain a, = f,, etc.,
IU Ch 5 [Gh a Gly = Oy, Che Wynccon Cheoy =b, ,, &a, = b,b,. But in the expansion
of a” the term free of @ is a, and in the expansion of f” it is b, = b,; thus we must have
a, = b,, whence, in view of a,a, = b,b,, we obtain a, = b,. The normal expansions of
numbers a,a and f, are thus identical, i.e. a= p, ¢
15. Prove that for positive ordinal numbers a and f the formula a+f = p+a
is equivalent to the formula aw* = po*.
Proof. If a+fP=f+a and numbers a and f are positive, then by Theorem 1
of XIV.25 there exist an ordinal number € and natural numbers m and ” such that
=a a= CN CCN OO” —= GIO) =O) |== CNG) —=
On the other hand, if a and f are positive ordinal numbers such that aw* = pw*,
then there exist a natural number m such that a is a remainder of number fim and
a natural number » such that 6 is the remainder of number an. Thus there exist ordinal
numbers y and 6 such that Pm =y+a, an=6+ f. If a=o™a,+...4 0%a, and B=
= wb, +...+ ob, are the normal expansions of numbers a and f, then we have
ob, m+ w2b,+ ...4 o'lb, =yto%a,+..+o0%a, and wa,n+ w%d.+ ...+ wa, = 0+
a 10 =... eke Sinie we easily find that fp, > a, > f,, therefore a, = f,; further
Cig == [855 Gaon =P olin = Oy coon Cs = Ors EHNG! oe € = w+ w%a,+ ...+ o%a,, We
obtain a Sia. = a5 == <b, theretore a 3 —= ba.
16. Give an example of ordinal numbers a and f for which aw = pw but a+ pF
# B+ a.
ANSwer a=, f)—@-— I.
Give an example of order types a and £ for which aw* = pw* but a+ Pp~ B+ a.
AISWiers @— 0, p— Teo”
18. Give an example of order types a and f for which a+ 6 = f+ a but am* ~ pa*.
Answer. a=1, B =w+o*

26. On ordinal numbers commutative with respect to multiplication.

We say that two ordinal numbers a and f are commutative with respect
to multiplication, or multiplicatively commutative if
(26.1) Cp == 0a..
348 XIV. Ordinal numbers

Formula (26.1) holds of course if at least one of the numbers a and

is equal to 0 or to 1. It also holds in the case where numbers a and / are
both finite. Numbers 0 and 1 are thus multiplicatively commutative
with every ordinal number and each two finite ordinal numbers are
multiplicatively commutative.
Now we shall show that no natural number >1 is multiplicatively
commutative with any transfinite ordinal number.
Let n denote a natural number >1 and f a transfinite ordinal number.
Thus, as we know, 8 = w&+k where é is an ordinal number >0 and k&
a natural number or 0. As we know,
np = n(wé+k) = wE+nk, pn = (wé+k)n=owén+k,

whence it follows that if n6 = pn we should have én = &, therefore, in

view of é>0, n=1, which is contrary to our assumption.
Thus every natural number >1 is multiplicatively commutative with
every finite number and only with every finite ordinal number.
In this way we have investigated the problem of multiplicative
commutativity for finite ordinal numbers. Subsequently we shall deal
only with transfinite ordinal numbers.
Every ordinal number is of course multiplicatively commutative
with respect to itself: thus we shall concern ourselves only with the
multiplicative commutativity of two different ordinal numbers.
The two smallest transfinite ordinal numbers, m and w+1, are not
multiplicatively commutative because w(@+1) = w+ and (o+l)a=
=o? <w?+o. A more general theorem holds, namely:
No transfinite ordinal number of the 1-st kind is multiplicatively com-
mutative with any transfinite ordinal number of the 2-nd kind.
Indeed, let a be a number of the 1-st kind and # — a number of
the 2-nd kind, and let
2. | B j Be f 3
a= 0 1d,+ 070, +... + w °1a;,_1+ a, , =O; wo! 2b, +... + wd;

be the normal expansions of these numbers. We have

:
ap = wp, + wth, +... + 2 + Ph, ;
, +a | Simeone Nn | it) ae | eA By B
Bo wit 1a, + wrt "ig +... + efit oe lip=4 OO Oo Dee

The first of these expansions has 1 components and the second

k+1—1, and since they are the normal expansions of numbers af and fa,
in the case of af = Pa they must be identical, whence 1 = k+1—1, there-
fore k = 1, which is impossible in view of a being a transfinite number.
Thus subsequently it will suffice to deal with the multiplicative
commutativity of transfinite numbers of the 1-st kind and of the 2-nd
kind separately.
 § 26. On ordinal numbers commutative with respect to multiplication 349

It is easy to prove that there exist infinitely many pairs of different

multiplicatively commutative transfinite numbers both of the 1-st and
of the 2-nd kind. For let é denote an arbitrary ordinal number and m
and » two different natural numbers. Numbers
(26.2) o) = (Ge and) ~pS=25"

are obviously multiplicatively commutative, and moreover they are of

the same kind as number é.
For numbers (26.2) the following formula of course holds:

(26.3) : an = pm,
The question arises whether for each two transfinite ordinal num-
bers, a and f, multiplicatively commutative there exist an ordinal num-
ber € and natural numbers m and » for which formulas (26.2) hold. We
find that this is the case only for transfinite ordinal numbers of the 1-st
kind (the proof of this theorem, which will be given later, is difficult);
for numbers of the 2-nd kind it need not be so. Indeed, let

(26.4) a=o+o, p= +o’.

It is easy to calculate that

(@? + w) (wm? + w?) = w® + wt = (wo? + w7)(w? +0),

therefore af = fa.
Suppose now that for a certain ordinal number & and natural num-
bers m and » we have formula (26.2). Number & would thus have to be
of the 2-nd kind. If we had m = 1, we should have é= a= w?+o > a’,
and, since é = @?+ w <@*+ w? = B, n > 2, whence p = &" > & > 4, which
is impossible as 8 = w*+ w? < w*. Thus m > 2, therefore w?+@= é" > ®,
whence @ < w?-2 = (w-2)*, which gives €< m-2. Since number ¢ is of
the 2-nd kind, we have = w, whence w*?+ wm? = &" = w”, which is impos-
sible.
Thus formulae (26.2) cannot hold for numbers (26.4) with any ordinal
numbers € and natural numbers m and n.
We find, however, as will be proved later (see Theorem 1), that in
order that transfinite ordinal numbers a and f be multiplicatively com-
mutative it is necessary and sufficient that there exist natural num-
bers m and n for which formula (26.3) holds.
EXERCISES. 1. For numbers (26.4) find natural numbers m and mn for which
formulae (26.2) hold.
Answer. m= 2, n= 3, because a? = f? = w*+ @?.
2. Find all transfinite ordinal numbers that are multiplicatively commutative
with number a= o®+ o.
350 XIV. Ordinal numbers

Solution, Let # be a transfinite ordinal number multiplicatively commutative

with number a = w?+ @. As we know, it will be a number of the 2-nd kind (since a is
a number of the 2-nd kind). Let 6 = wtb, + wb, + ...+ ob, be its normal expansion.
We shall have af = Pigs|ee See at Ph. and fa = wt 4 wit? Thus in order that
aB = Ba it is necessary and sufficient that k = 2, b,; = b,= 1, 2+6, = ~i4+2, 2+ 6p, =
— B,+ 1. Number /, is thus additively commutative with the finite number 2, whence
it follows, as we know (e. g. in view of Theorem 3 of XIV.25) that it is itself fmnite. Thus
the formula 2+ 6,= 6,+1 proves that number f, is also finite and that 6,= 6,+ 1,
But, as we snow Bb, > O (number f being of the 2-nd kind); thus itis a natural number,
B, =n, whence f, = n+ 1, therefore B = w''*+ a".
Thus every transfinite ordinal number which is multiplicatively commutative
with number w?+ @ has the form w"!*+ w", where n is a natural number, and, conversely,
every number of this form is multiplicatively commutative with number w?+ w. Since
number w?+-m, being transfinite, is not multiplicatively commutative with any finite
number >1, it follows hence that number ?+ @ is not multiplicatively commutative
with any ordinal number >1 smaller than itself.
3. Prove that numbers a= w3+@ and f= w'+ qm? are maultiplicatively com-
mutative, but that, for these numbers, there exist no ordinal number € and no
natural numbers m and n for which formulae (26.2) hold. Find for the above numbers a
and # natural numbers m and n for which formula (26.3) holds.
4. Prove that for a pa k the only transfinite ordinal numbers multiplicatively
commutative with number w+ k are the powers of that number with a natural exponent.
Proof. Since number a = w+ k is of the 1-st kind, if a transfinite ordinal number /
is multiplicatively commutative with it, then it must also be of the 1-st kind. Let 6 —
= wtb, + eee ae b, denote the normal expansion of number /; we shall have / > 1
since 6 is a transfinite number. By comparing the normal expansions of numbers af
and Pa :
af = w'*'1b, + wT 2b, +... + © HERD re cote lok

po = wit} ah wtb, ke 2b, + ae at—1h, +b,,

Weulnters in vilewrOteGa >:= San ulnciul Sym — Nome) Oy == Symite fl ==)9m =) eee rent oe
= p,, 1+ 6, = 6,+ 1, whence we find that 6, , = 1, ae Say cae,” Ba =I—1. By com-
paring the coefficients in the normal expansions of numbers af and fa we obtain b, = 1,
(Dy =o, Oy = yh soon We = Dg 1 lays, WN, CAE, Oh Se, Oy SD = cen = Dy se he MR
ei ern nae wo k+ ok + + wk+-k, which, as can easily be verified, is equal to
number (@+ k)'~". Thus ne
=a ~' where |—1 (in view of 1 > 1) is a natural number,
Ms Gs Ok
Prove that if a, f and y are ordinal numbers such that af = pa and ay = ya,
then also a(fy) = (by)a
Proof. In view of our assumptions we have a(fy) = (ap)y = (Ba)y = B(ay) =
= B(ya) = (By)a, therefore a(By)= (By)a, q.e. d.
6. Prove Jacobsthal’s theorem stating that if a, f,y +40 are ordinal numbers
such that ay = ya and a(yf) = (yf)a, then af = fa.
Proof. In view of our assumptions we have y(af) = (ya)B = (ay)B = a(yB) =
= (yB)a = y(Ba), therefore y(af) = (Ba), which, in view of y#0, gives af = fa,
Gio Cn Bh
 § 26. On ordinal numbers commutative with respect to multiplication 351

THEOREM 1. In order that transfinite ordinal numbers a and fp be

multiplicatively commutative it is necessary and sufficient that there exist
natural numbers m and n for which a’ = pr”.
We shall first prove Theorem 1 for numbers of the 2-nd kind.
Let a and f be ordinal numbers of the 2-nd kind, and

(26.5) a= O'a,+:.+o0%, and p=ab,+...+ ab,

their normal expansions. For numbers af and fa we obtain hence the

normal expansions
(26.6) ap = wtp, + wb, + 2. + wo, ,
(26.7) pe = att aq, aL ait 2q, +...+ art! ea,

lf af = fa, then by comparing the normal expansions (26.6) and

(26.7) we obtain k = 1 and, above all, the formula a,+/, = 6,+4a,, from
which it follows that the positive ordinal numbers a, and /, are additively
commutative, which, by Theorem 3, of XIV.25, implies the existence of
natural numbers m and » such that
(26.8) Gn = bm.
Further, from the comparison of the expansions (26.6) and (26.7)
we obtain
(26.9) Pret = Gp4- 6; 9 for 4 == 235 as, Be.

Adding equality (26.9) to equality (26.8) we obtain

(26.10) Qi bp OG Py N= Oy py «LOT? 4 32 Sy cs KA

Since numbers a, and f, are additively commutative,

an+6,=a4+f6,+a(n—1) and pym+a,—a,4+ 6,+/,(m—1)

in view of which formula (26.10) gives

a +f, +a,(n—1)+a,;=a,+6,+P\(m—-1)+6; for t=

whence
(26.11) ony -Ih) 4 oF == Bn) Bb; for i923,
2.2, Rh.

The comparison of coefficients in the expansions (26.6) and

(26.7) gives moreover
[eat eikcie ay Neapaed ioe
From formulae (26.5) it easily follows that
nn =Ij+ are
(26.13) Oe ou sere Game i fae
- . Sioa aay
(26.14) eee wt", 4b gi + Bap, HAY Ae fu) Pp, J
By XIV. Ordinal numbers

whence, in view of k = 1, and formulae (26.8), (26.11), and (26.12), we

obtain the formula a” = p”. Thus for numbers of the 2-nd kind the condi-
tion of Theorem 1 is necessary.
Suppose now that this condition is satisfied for numbers of the
2-nd kind (26.5). Thus for certain natural m and » we have the equality
a" = 6”. Since formulae (26.5) imply the normal expansions (26.6) and
(26.7), we obtain by comparing them k= / and formulae (26.8), (26.11)
and (26.12).
Formula (26.8) proves, according to Theorem 3 of XIV.25, that the
numbers a and f are additively commutative, 7. e. that

(26.15) 6, Bice Ons oe

Adding equality (26.11) to equality (26.15), we obtain, in view of
the additive commutativity of the numbers a, and f,, equality (26.10),
which, by (26.8), implies formula (26.9).
But formulae (26.5) imply formulae (26.6) and (26.7). In view of
>= 1 and of formulae (26.6), (26.7), (26.15), (26.9) and (26.12) we obtain
the equality af = fa. Thus for numbers of the 2-nd kind the condition
of Theorem 1 is sufficient.
Another proof that if for natural numbers m and » and ordinal num-
bers a and f we have a” = f”, then af = fa, is the following: if a = pf”,
then a”"f” = fra", thus, according to Exercise 3 of XIV.10, af = Ba, q. e. d.
Theorem 44 is thus proved for numbers of the 2-nd kind.
In order to prove it for transfinite ordinal numbers of the 1-st kind,
we Shall first prove the following theorem of E. Jacobsthal, which is in
itself interesting:
THEOREM 2. If a is a transfinite ordinal number of the 1-st kind,
and ay denotes the smallest ordinal number >1 such that aay = aya and B any
ordinal number >1 such that ap = Ba, then B is a power of number ag with
a natural exponent (Jacobsthal [1], p. 484).
In order to prove Theorem 2 we shall first prove two lemmata.
LEMMA 1. If a and 6 are transfinite ordinal numbers of the 1-st kind
such that a > 6 and af = Ba, then there exists an ordinal number B’ such
that a = Bp’= B’B and af’= p’a (Jacobsthal [1], p. 482, Theorem 8).
Proof of Lemma 1. Let

(26.16) a= wa, + @ 7a, +... + wo -ay_1+ aK,

B= wd, + wd, +... + 10,1 + by

(where a, > 0 and b; > 0) be the normal expansions of numbers a and #.
 § 26. On ordinal numbers commutative with respect to multiplication 353

Hence we have the normal expansions

4 +f +B Ei} :
(26.17) “ap = wo" Pip, + gt Pap, ae ee aS TFG) *dy Op @ "Be 4, Oey
F a y+ A ;
61S) 5 Boo Mao! 80, 4. a a a Dag + aD, ES -EOy.
Suppose that a > 6 and af = fa. We have b, = a,. If we had a, = /,,
then in view of af = Bu the components w”a,b; and wb,a, of the normal
expansions (26.17) and (26.18) would have to be equal: in view of the
equality of the expansions (26.17) and (26.18) the next components would
also have to be equal, ete., whence we should obtain

Go Poel Og == Pas rec gh ay Bn D5 hala Og yxy) Oj. = OF, nO 0.4 Opes

whence a, = b,; therefore in view of (26.16) we should have a = f, -which

is contrary to the assumption that a > 6. Thus a, # /,, therefore, in view
of a > f and a, = b,, a, > B,. In view of the equality of the normal ex-
pansions (26.17) and (26.18) the component a,b; of the expansion
(26.17) must be equal to a certain component of the expansion (26.18)
preceding the component wba, since a, > ~,. Hence it follows that the
component m2 a, of the expansion (26.17) is equal a to certain component
of the expansion (26.18) not later than the component w%b,a,. Thus
there exists an ordinal number. o (of the 2-nd kind or equal to 0) such
that
B ? .
(26.19) — wa, 4+ w% d+... +a; = w4(o + d,a%) 4 Ob, ie eee

Moreover, wo? < S w”, therefore

é me B.
wo < w%, whence _
o < ot,
Let

(26.20) ~ 6’ = wo a, +o + ax.
In view of (26.16) and (26.20) we obtain

(26.21) BB’ = w7a,+ B(o+ ax).

Since number o is of the 2-nd kind or equal to 0, by (26.16), we

obtain fo = wo but Ba, = od,a,+ ob, +... 4 b;, therefore, by (26.19):
B(o+4a%) = Bo+ Bay = wo(o+d,a,) + 8), +...+0, = WMG,
+ wy +... + ng
and therefore, by (26.21) and (26.16) pp’ = a
since “opi= pa, we have 6p) = 0p — ba vt. .e) PUpby =-8a; which
gives p p= o."Thus.a = pp = p's. Hence we have ap ='(6"B)p’ = p'(BB') =
= f'a, therefore af’ = f'a.
Lemma 1 is thus proved.
LEMMA 2. If a, 6 and 6 are transfinite numbers of the 1-st kind and
= 610. 0D == po, a0 0a, then

(26.22) aay,
Cardinal and ordinal numbers to[o)
354 XIV. Ordinal numbers

where

(26.23) pee
and ap, = pya.
Proof of Lemma 21). Suppose that a, 6 and 6 are transfinite
ordinal numbers of the 1-st kind and a>f> 6, af= fa, ad= da.
According to Lemma 1 there exists an ordinal number (’ such that a=
= Bf’ = B’B and af’ = f’a; the number f’, being greater than 1 (since
a>fP and a= fp’) and multiplicatively commutative with the trans-
finite number of the 1-st kind, a, must also be transfinite of the 1-st
kind. Let £, = 6’6: it will be a number of the 1-st kind greater than o.
In view of 6 < f we shall have f, = p’6 < f’B = a, therefore B, <a and
apy = af'6 = p’ad = f'da = f,a, and thus af) = f,a. By Lemma 1 (for a
and f,) there exists a number f, such that a= 6,f,= 6,f) and af,=
= fi.a “Wurther,. swe {have a ==!6,0; = poppy, aud. “smce..a%—7 85.
B = 6f,. If B = £,, we should have hence 6 = 6f, and since # is a number
of the 1-st kind, 6=w+1, we should have w+1= 6(u¥+1)= 6u¢+
+6>u+0, therefore w+1 > 4+6, which gives 6 <1, which is impos-
sible, 6 being a transfinite number. Thus 6 ~ 6,, therefore, in view of
6 = 68,,8 > B,. Lemma 2 is thus proved.
Proof of Theorem 2. Let a denote a transfinite ordinal number
of the 1-st kind. There exist transfinite numbers >1 of the 1-st kind
multiplicatively commutative with number a, e. g. number a itself. Let a,
denote the smallest number of the 1-st kind, greater than 1, multiplica-
tively commutative with number a; as we know, it will be a transfinite
number. We shall show that a is a power of number a, with a natural
exponent. This would be obvious in the case of a = a,. Therefore let us
assume that a ~ a); from the definition of number a, it follows that
a >a). By Lemma 1 there exists a number ag such that a = aoa = ajay
and adg = aga.
We shall have a> ag since a) > 1 and a= agay. Number a, being
wultiplicatively commutative with the transfinite number of the .1-st
kind, a, is also transfinite of the 1-st kind In view of aaj= aja and of
the definition of number a, we thus have a > ay. If aj = a), we should
have a= a0 = aa.
Therefore let us suppose further that a, = aj >a). By Lemma 1
there exists a number a, such that a, = apa, = aya) and aya. = Q20,; Since,
in view of a,=— ao, we have a,a= aa,, and since aa,= aja, we have
Ap Ms = O09A, = Ayaa,, whence a4 = aa,.

') The idea of the proof is the same as Jacobsthal’s [1] (p. 483), but his proof is
somewhat longer than ours.
 § 26. On ordinal numbers commutative with respect to multiplication 355

In view of the definition of number a, we thus have a, > a) and since

a, = Gd) and a, is a transfinite number, we have a, >a,. If a, = ay,
then a,= a5 and a=aa,=o9. If a> Qo, then we deal with the
humber a, in the same way as with the number a, above, etc. Since
a> a, >a, > ..., we Shall finally obtain a,_; = a, with a certain natural n,
ILC SIMCOE Ol yinOh, nly ye 8.845 On tg =" G4 Gg) WeHAVE a= Op:
We have thus proved that there exists a natural number » such that
a= ao. Hence it follows that we cannot have a,f = fa, for any ordinal
number 6 such that 1< $< ap, since in view of a= ao we should then
have also af = fa, and the number #, being greater than 1 and multi-
plicatively commutative with the transfinite number oa of the first kind,
would be transfinite of the first kind and smaller than a,, in contradic-
tion to the definition of the number a,. Therefore the number a, is not
multiplicatively commutative with any ordinal number smaller than a,
and greater than 1.
Now let 6 denote any ordinal number such that 1 < 6 < a and af =
== Ppa. The number / will thus be transfinite of the first kind and ff > ay.
If 6 > ay, then we put 6 = a, and apply Lemma 2. We shall have 6 = af,
where 6, < 6 and af, = f,a. If 6, > a), then applying Lemma 2 to the
numbers a, £,, a), we Shall obtain 6, = ay, where §, < f, and af, = fa.
If 6, >a), we deal with the number #, in the same way as with
number /, above, etc. Since the numbers /,, /,, ... decrease, we shall finally
obtain, with a certain natural k, a number f, <a, such that af, = pra,
whence it follows, as we know, that 6,=1 or B, = a). Since B = af,,
=) le ones Pha Og PE awe) ODLAIN . 6 eee and hence p= ag OF
ee
We have thus proved that if 6 is any ordinal number multiplicatively
commutative with a and such that 1< ~<a, then for a certain na-
tural m we have B= ao.
Now let y denote any ordinal number >a and multiplicatively com-
mutative with number a. Number y will be transfinite of the 1-st kind
(since such is number a). If y = a, then y = ao. If y > a, then by Lemma 1
there exists an ordinal number a such that y = aa, = a,a and ya, = my.
If a, >a, then similarly, by Lemma 1, there exists an ordinal number a,
Such that! a, — aa, — d,0-28nd “a,.0,4- 4,0,, and ih view Of @, == o,0, ‘and
a >1 we have a, > a,. If a, >a, then find similarly a number a,, etc.,
and, in view of a, > a, >... we Shall finally obtain, for a certain natural p,
a number =o, <a. ius we shall have y= aa;,0,— a0,,..., @,.1 =
= ag, — ap; therefore y= off where Pp <a and a,_1 = aa, = a,a, whence
af = Ba. If B=a, then y=a""", therefore y=of'™; if B=1, then
y= a’ — ap; finally, if 1 < #6 <a, then, in view of af = Ba (number 2B
being then transfinite), we have, as has been proved above, 8 = aq, there-
29%
23
356 XIV. Ordinal numbers

fore y = a’B = af"? . Thus in any case y is a power of number a, with

a natural exponent.
We have proved that if y > a and y is an ordinal number multiplica-
tively commutative with number a, then y is a power of number a, with
a natural exponent. Since we have already proved a similar property of
ordinal numbers / such that 1 < 6 <a, Theorem 2 is proved.
Now we shall prove
THEOREM 3. If a, B and y are arbitrary ordinal numbers such that
o> Ly. ob = Po-ond.ay.=— ya, inen”’ py = va.
Proof. If a is a finite number, then, as we know, in view of af= faQO

and ay = ya, the numbers 6 and y are also finite and py = yf.
If a is a transfinite number of the first kind, then, in view of af = fa
and of ay = ya, the numbers f and y will also be transfinite of the first
kind, and in that case, by Theorem 2, there exist natural numbers p
and gq such that 6 = af, y = 0, whence fy = yf.
Finally, if a is a number of the 2-nd kind, then, in view of af = pa
and ay = ya, the numbers 6 and y will also be of the 2-nd kind, and in
that case, by Theorem 1, which has already been proved for numbers
of the 2-nd kind, and in view of af = fa there exist natural numbers m
and » such that a" = p”, and in view of ay = ya there exist natural num-
bers p and q such that a? = y7. Hence we have fp = ar" = y%, and apply-
ing Theorem 1 to numbers of the 2-nd kind, 6 and y, we again infer
that By = yp.
Theorem 3 is thus proved. It will be observed that it is not necessarily
true for order types, for it can be shown for instance that for a= 2,
B= on, y= on we have af = pa and ay = ya but By A y6.
We shall now prove Theorem 1 for transfinite ordinal numbers of
the first kind. Suppose that a and f are transfinite ordinal numbers of
the first kind. By Theorem 2 there exist natural numbers m and n» such
that == 0550 = Bos whence a =p"
Conversely, suppose that a and / are transfinite ordinal numbers
of the first kind such that for certain natural numbers m and n» we
have a”?= Bp". We have of course 6f"= BB, therefore fa* = a”B, and
since also aa” = a"a, we have, by Theorem 3 (for the numbers a”, 6 and a),
ONO /6x0s
Theorem 1 is thus proved for transfinite ordinal numbers of the
1-st kind, and since it has already been proved for numbers of the 2-nd
kind, it is now proved in full.
From Theorem 1 follows
COROLLARY 1. For every ordinal number a > 1 the set of all ordinal
numbers multiplicatively commutative with number a is denumerable.
 § 26. On ordinal numbers commutative with respect to multiplication ODT

Proof. If an ordinal number a > 1 is finite, then it is known to be

multiplicatively commutative with finite numbers and only with finite
numbers. Thus for finite numbers a > 1 our corollary is true. If a > o,
then, on one hand, numbers a” (n= 1, 2,...) are all different from one
another and multiplicatively commutative with number a; on the other
hand, if af = pa, then, by Theorem 1, there exist natural numbers m
and » such that a”? = 6” and for a given ordinal number a and given na-
tural m and » there exists at most one ordinal number / for which a" = p™:
for if Ps > p,, we OMe have p:' > fi and we could not have simultane-
ously a” = fy and a” — fs. Since the set of all systems m,n of two na-
tural numbers m, » is denumerable, it follows that the set of all
ordinal numbers ( satisfying the condition af= fa is (for a given ordinal
number a > m) at most denumerable, and since, as has been proved before,
it is at most denumerable, it is denumerable. Corollary 1 is thus proved.
From Corollary 1 and Theorem 3 immediately follows
COROLLARY 2. Ordinal numbers >1 form denumerable disjoint sets
such that two ordinal numbers >1 are multiplicatively commutative if and
only if they belong to the same set.
EXERCISES. 1. For an ordinal number a > 1 find the smallest ordinal num-
ber y such that a is not multiplicatively commutative with any ordinal number >».
Solution. It is the number y ~ @°. Indeed, if number ais finite and greater than1,
then it is known to be multiplicatively commutative only with finite numbers, whence
it follows that y = w, which agrees with formula y = a® since for finite a > 1 we have
a° — wm. If number a is transfinite and if B is an ordinal number multiplicatively com-
mutative with a, then either 6 = I or number f is transfinite and in that case, by The-
orem 1, there exist natural numbers m and n such that a” = p". Let a= o4a,+...+
+ o%a, and B = w"b,--...+ ob, be the normal expansions of numbers a and f. The
first components of the normal expansions of numbers a” and 6” will be w%a, and aw” b,
respectively, whence, in view of a” = 6”, we obtain ayn = B,m. Thus p, < bpm = an <
<a,@, whence we easily obtain B < 4° = a”. Thus every ordinal number ea ae
eatively commutative with number a, is smaller than a®. On the other hand, if u< a®,
then, in view of
a”? =lima ,
N=

we have, for a certain natural n, « <a”, and number a is of course multipheatively

commutative with number a”. Therefore y= a® is the smallest ordinal number such
that a is not multiplicatively commutative with any ordinal number >y, q. e. d.
2. Find all ordinal numbers >1 multiplicatively commutative with number o.
Answer. They are all numbers of the form w” where n = 1, 2,3
3. Prove that if a is any of the numbers w+ 1, w”-2 (where n= 1, 2, ac CD
w°-+ a, o°-2, then numbers a”, where m = 1,.2,..., are the sole ordinal numbers > 1
that are multiplicatively commutative with number a.
4. Find all ordinal numbers >1 multiplicatively commutative: with number a=
2 : e x
= w°” + q@ and prove that for no ordinal number & they are of the form &” where n=
se 1 oe
358 XIV. Ordinal numbers

Solution. They are all numbers of the form w?o(?"-+ ) where k = 0,1, 2, ...
The smallest of them is a. If they were all represented by the formula &” (m = 1, 2, ...)
where é is an ordinal number (greater than 1 of course), then we should have =a
and there would exist no ordinal number f multiplicatively commutative with nuim-
ber a and such that a? < 6 < a’, while in fact B = w?*(w®*-+ @) is such a number be-
cause
a = m°7*(m°2 +) and a= w°*(w?* +o).

5. Prove that each two of the order types 7, 1+7, 4+1 and 14+ 4+ 1 are multi-
plicatively commutative.
6. Determine all order types, at most denumerable, that are multiplicatively
commutative with the type 7.
Solution. They are types 1, 7, 14-7. 4+1 and 1+ 7+ 1. Their multiplicative
commutativity with type 7 follows from Exercise 5. On the other hand, if a is an order
type at most denumerable and such that an = ya, then from Exercise 1 of XII.10 it
follows that ya = 7; therefore an = 7 and a is a left-hand divisor of type 7, 7. e., accord-
ing to Exercise 5 of XII.3, a must be one of the types 1, 7, 1+ 7, 7+1 or 1+ +1.
7. Give an example of a non-denumerable order type, multiplicatively commutative
with type 7.
Answer. Type 7/7.
8. Give an example of a non-denumerable order type € such that 2-€= €-2.
Answer. € = wAn.
9. Give an example of a non-denumerable order type &€ such that wé = éw.
Answer. €= An.
10. Give an example of denumerable order types f, y and 6 such that 2-6 = p- 2,
2-y=y-2, 2-6= 6-2, but Py ~ yB, BO ~ OB, yd H$ Oy.
Answer. B=on, y= ow*n, 6 = (w*+ ow).
11. Give an example of two different infinite order types both additively and
multiplicatively commutative.
Answer. Types w+@* and (w+ o*)?.
12. Prove that there exist no two different transfinite ordinal numbers both ad-
ditively and multiplicatively commutative.
Proof. Suppose that there exist two transfinite ordinal numbers a and f such
that a #4f, a+ f = B+ a, af = Ba. In view of af = Ba, numbers a and f are known
to be either both of the 1-st kind or both of the 2-nd kind.
Let us suppose that the transfinite ordinal numbers a and 8B ~£ a are both of the
1-st kind and assume, for instance, that a > 6. By Theorem 2 (in view of af = fa)
there exists an ordinal number a, (transfinite of course) and natural numbers m and n
such that a= /, f =a, and, in view of a > B, we have m > n, therefore m = n+ p
where p is a natural number. In view of a+ f = B+a we have a, +a, = a,+a" there-
fore aj(aj+ 1) = aj(a+ aj), which gives a}+1=1+ a> which, as we know, is impos-
sible since number | is not additively commutative with any transfinite number.
Suppose therefore that number a and 6 ~a are of the 2-nd kind. In view of
a+ Pp=f6-+a and by Theorem 1 of XIV.25 there exists an ordinal number é, of the
2-nd kind of course, such that = wé,, and natural numbers k and 1 such that a =
=ék, 6 = él. Thus we have af = ékw&,l= €0&,l, Ba = Elwé,k = Ewé,k, therefore, in
view of af = Ba, (fm&,)l = (wé,)k, which gives | = k, therefore a = B, which is contrary
to the assumption.
® § 26. On ordinal numbers commutative with respect to multiplication 359

The required proof is thus completed.

13. Give an example of two infinite order types a and # for which af = fa but
for which there exist no natural numbers m and n such that a” = p”.
Answer. a=ao, 8 =n.
14. Let n denote a natural number >1. Prove that in order that the product
of n transfinite ordinal numbers aj, a, ..., a, be independent of the order of those num-
bers it is necessary and sufficient that there exist natural numbers k,, kz, ..., k, such
that ail = aka =..=air,
15. Determine the power of the set of all different denumerable order types
multiplicatively commutative with number 2.
Angwer. Their number is continuum. The proof is based on Theorem 1 of XII.5
and on the observation that every type

(3 (AC)as V9 con) = |M48(@m,+ o*)|y

kI »

where 1, %.,... 18 an arbitrary infinite sequence of natural numbers satisfies the condi-
tion 2-€= &-2 and that &(m,, me, ...) = &(m, me, ...) only if m =n, for k= 1, 2,...
16. Give an example of 2° denumerable order types no two of which are multi-
plicatively commutative.
Answer. Types &(n,, 2, ...) from Exercise 15.
17. Find what ordinal numbers a may be represented in the form a = yy where
Ly F ve.
Answer. All transfinite ordinal numbers that are not of the form (wo + 1)”, where&
is an ordinal number, n =1, 2,..., and only such numbers.

27. On the equation of = 6" for ordinal numbers. Rejecting the

trivial case where a = f, we can assume further that a < f. For natural
numbers the equation a’ = 62, where a < f, is known to have only one
solution: a= 2, B= 4. In XIV.20 we have given the solution of our
equation in different transfinite ordinal numbers: a=, B= ew. Now
we shall determine all the solutions of this equation in ordinal numbers.
We shall prove the following theorem of E. Jacobsthal [1] (p. 488):
THEOREM 1. All solutions in ordinal numbers a and fb >a of the
equation
(27.1) af = pr,
besides the solution a= 2 and B = 4, will be obtained by taking for a an
arbitrary transfinite ordinal number of the 2-nd kind and writing B = ta
where t denotes an arbitrary epsilon-number >a.
Proof. To begin with we shall prove that equation (27,1) has no
solutions in which a is a natural number and f a transfinite ordinal number.
For this purpose let us suppose that a, is such a solution. |
If 6 is a number of the 1-st kind, 6 = w&+k, where k is a natural
number and > 0, and if a= n where n is a natural number, then af =
 360 XIV. Ordinal numbers

= nb = ye. nk = w§- nk; therefore a is a number of the 2-nd kind while

(6 Fre pr is a number of the 1-st kind since such is number 6. Equation
(27.1) is thus impossible.
If 6 is a number of the 2-nd kind, 6 = wé where £>0 and a=n
where » is a natural number, then it follows from (27.1) that 6” = n? =
ne =o. If 6 = ob, +... is the normal expansion of number f, then
HG normal expansion of number f” is p" = w"b,+..., whence in view
of B" = wf we obtain € = 6,n, therefore 8 = wé = wf,n and f" = (wp,n)"* =
= (wf,)"n, which proves that number f” has a right-hand natural divisor n.
But we have obtained above p" = ow, and the number m5 is known to be
a prime component. and as such it has no right-hand natural divisor >1.
Thus n= 1, therefore a= 1, and by (27.1) also 6 = 1, which is impos-
sible in view of 6 >a
Besides solutions in finite numbers, equation (27.1) can thus have
solutions only in ordinal numbers a and f > a which are both transfinite.
We shall prove that in that case both a and 6 must be of the 2-nd kind.
For this purpose we suppose that one of the numbers a and f, e. q-.
the number /, is of the first kind, and the other, a, of the 2-nd kind.
Let a= @74,+.:.+ 07%, and B= @ Dy 525 wb, me be ee normal
expansions of the numbers a and f; we shall have o, >1, pj > 1, l> 2,
b, > 0. Let 6 = 6’ +b,; thus fp’ will be a number of the 2 ae kind, =e,
therefore B = w&+b,. We have w72 <a< w'*, But, as we know, for
: : . is +1
a, 21 we have (a,+1)®—a,. Therefore w'” <a < tne Owe
whence a®° = w” and thus
ayo

wy g
= (wa?!b = wo” awe
19s b 21
a" = w ap’ b
B wE+ by by; iP Pt
a =a SAAT as

Since a = wt,a being a number of the 2-nd kind, we have f* = (f*)¥=

By = o
= (wt)? = ow" e
and thus, in the case of equality (27.1), we have
ag?! = wo", whence it follows that the number a” is a power of the
number w, and since, in view of the normal expansion of the number a
we have a Dik ed Aye, a= wo, whence a= oo and a= w™ 5
since we have Siren above p*= w'*, we have by (27.1) 48 = pia.
But p,a = p,o = w*, therefore a,B = wf, while from the normal expansion
of the number f, which has at least two components (since | > 2) it fol-
lows that the normal expansion of the number a,/ has also at least two
components, which gives a contradiction.
We have proved that if transfinite ordinal numbers a and / satisfy
- equation (27.1), then a cannot be a number of the 2-nd kind and p
simultaneously a number of the first kind. In the same way we prove
that a cannot be of the first kind and / simultaneously of the 2-nd kind.
Suppose now that numbers a and # are both of the first kind and
let a= o%a,4+...+@* gata, B= oD, ose w'—-1b)_4 +b, be their
 § 27. On the equation a8 = fa 361

normal expansions. Thus a, > 0, b, > 0 and we can write a= a’+ax,

Bp = p'+ 6, where a’ and #’ are numbers of the 2-nd kind. Further, we
< . : Ende}
easily obtain a = oa", p= wo 4 p%,Aap a= oaL
t..ta,, ay:
b=
a ole, Sl, Nery and by ;
(27.1) we obtain: owBf 4 *™%¢,4+...+0%
b Y a, =
Bya + é ; Z 7
OG 2 ei bp whence ° a/b eb; = plait 6, an) 0 e. a, 6 =
= f,a; by comparing the last components we have a,/p’ = f,a’ and a, = b7;
therefore a,6;= B,a,, which in view of 6b; = a,>0 gives a,;= p,, and
in view of the equality a,$ = ~,a obtained above this proves that 6 = a,
in contradiction to the assumption that a + fp.
We have thus proved that if a, 6 is the solution of equation (27.1)
in ordinal numbers, and if 6 > a and numbers a and f are not both finite,
then a and. # are transfinite numbers of the 2-nd kind. Let a= w“a,+
+...+0%a, B= wd, +...40%b, be their normal expansions. Hence
we have a° = w%, f*= w™, therefore, by (27.1), ajf = p,a. Let o, =
=o m+... and p, = w1n,+... be the normal expansions of the num-
bers a, and /,. We have yu, > 0 and », > 0. Since a and # are numbers
of the 2-nd kind, we have a,f' = wo" and f,a= @"a, therefore, in view
ofa, 6B= Ba, of = wta. In view of B>a-we obtain’ hence yu, <%,
therefore »,= 4, -+t where +> 0. Hence of = ao "a= oo", which
eee
gives B5% = w'a = wo + '"a,+...= wb,
B
+..., whence t+a, = f; = w+y
= By a sty7, theretore . 14 om, + ...= O + om... Qs
Since +> 0; and
ly < fy +7, the component w”*n,, which is the first component of the
normal Saronen of the right-hand side, must be equal to the first com-
ponent of the normal expansion of the number 1, whence w''n, <t
and m* <t, and since, in view of + >0 and by formula (18.2) we have
wt >t, we obtain w*= 7, which proves that t is an epsilon-number
(XIV.20). Thus we have 8 = wa = ta, and t > w"* > w** > a,, whence
t>a+1 and t= o > wo" >a, therefore t > a.
We have proved that if ordinal numbers a and f > a, which are not
both finite, satisfy equation (27.1), then 6 = ta where 7 is an epsilon-
number >a and a is a transfinite number of the 2-nd kind.
Now let a denote any transfinite ordinal number of the 2-nd kind
and t any epsilon-number >a and let 6 = ta. Let a= wa,+... denote
the normal expansion of the number a. In view of a < t= @* we have
a, < t; therefore, in view of a > 0 (the number a being transfinite) and
by Theorem 1 of XIV.20, formula (20.4), we have at = vt. In view of
p= ta= ota, the number f is of the 2-nd kind, therefore a = wo” =
= wo" = w™. On the other hand, Bp = ta= w'a= wo “a,+..., whence,
on account of a being a number of the 2-nd kind, 6* = w°*™*. But in
view of t>a>a,, and by Theorem 1 of XIV.20, formula (20.3), we
have a,tt=t, hence Ta <t+q+t=t+t=7-2; therefore tw <
<(t+a,)0 <t:-2w=tw, whence (t+a,)o=—tw, and since a= wé,a
362 XIV. Ordinal numbers

being a number of the 2-nd, kind, (1+ 4,)a= (t+ 4,)wé = tw§ = ta.
Thus £* = w™. Therefore we have af = fe, 7. e. formula (27.1).
Theorem 1 is thus proved.
From Theorem 1 it immediately follows that the solution of equa-
tion (27.1) in the smallest transfinite ordinal numbers a and f >a will
be obtained by setting a= m and 6 = ew where « is the smallest epsilon-
number, 7. e. the number (20.1). Another solution will be obtained by
setting a= w-2 and P= «w- 2.
From Theorem 1 we also immediately obtain the following
COROLLARY. All solutions in ordinal numbers & 4 w of the equation
a = & are obtained by setting — = tw where t denotes any epsilon-number.
This corollary can also easily be proved directly (Sierpinski [52],
p. 49). ;
In XIV.20 we have proved that for every ordinal number y there
exists an epsilon-number >y. Thus from Theorem 1 it immediately fol-
lows that for every ordinal number a of the 2-nd kind and for every
ordinal number y there exists an ordinal number f > y satisfying (27.1).
However, there are ordinal numbers a of the 2-nd kind for which there
exists no ordinal number 6 <a satisfying equation (27.1). H.g. there
is no such number f/ for any ordinal number a such that wo <a< «qa,
e. g. for a= e. It can be proved however that for every transfinite ordinal
number a (of the 2-nd kind) there exists at most one ordinal number B < a,
satisfying equation (27.1).
Indeed, let a be a transfinite ordinal number of the 2-nd kind,
a= ow a,+...+ oa, — its normal expansion, and suppose that 6 <a
is an ordinal number satisfying equation (27.1). In that case number f,
as we know, must also be of the 2-nd kind: let 6 = w%b,+...+ 0b; be
its normal expansion. By Theorem 1 in view of a>, we must have
a= tf where t is an epsilon-number >f. Thus a= 18 = 0B = w'"b,+
+o",
+ ...+.0°°%b,, whence, by comparison with the normal ex-
pansion of number a, we obtain /= k,b; = a; for i= 1, 2,...,1,7+fi= a,
LOrt.== 1, 2.2. 51. Whee: .B; —30)— 7 fOr a= 2s. 1. Number deiedt
exists) is thus well defined by number a.

EXERCISES. 1. Prove that if a, 6 and y are different ordinal numbers and a° =

= p* and p” = »%, then a’ 4 y*.
Proof. It suffices of course to show that there exists no system of three different
ordinal numbers a, f andy suchthat a® = p*, a” = y* and fp” = 76. Suppose therefore that
such a system does exist. In view of the symmetry of conditions for a, 6 and y we can
of course assume that a < f < y. From Theorem 1 it easily follows that numbers a, pb
and y must all be transfinite of the 2-nd kind, and that there exist epsilon-numbers
%, Tt, and t, such that t, > a, B = %14, t, > a, y= 7G,T; > B, y = 38, Whence ta =
=%T,a and t,a = 138. Since a < B, we have hence 7, > t; and in view of 7; > fp =
 § 27. On the equation tof = fe 365

aioe Le Cm AAMC aiam= Tee LNs rete ty ead t= Pb > Ga Anew. OL etsai—
=T3T,a we have 1t,at,= 7,7,aT,. But, since epsilon-numbers have the property of
absorbing positive factors smaller than they are and immediately preceding them (see
Theorem 1 of XIV.20, formula (20.4)), we have at, = t, and 73;7,aT, = 7T,; thus equality
Tet, —= TstyAt, LIVES 72 = %,, which is impossible since 7? = 7,7, > 7,-

2. Find all systems of positive ordinal numbers a, f and y such that a8” = (a*)’.

Solution. Ifa= 1, 6 and y may of course be arbitrary ordinal numbers. If a > 2,

the equality a8” = (a*)” = a’ gives Bp’ = By. In the case of y= 1, a and f may of course
be arbitrary ordinal numbers. In the case of y = 2 we obtain 6 = 2, while a may be an
arbitrary ordinal number. For y > 1 the equality B” = By gives p’—* = y and, if y is
a finite number >2, we obtain 6 > 1, therefore 6 > 2, and since by means of an easy
induction we prove that 2’? > y for y > 3, we should have y = B”—* > 2”? > y for
y => 3, which is impossible. Thus y cannot be a finite number > 2. In the case of y = a,
B may be an arbitrary finite number >1 since n® = nw for n = 2,3,... In the case
of B > w and y= o, in view of B’—* = y and of the fact that y—1 = y» for y >, we
have 6” = y, whence by Exercise 1 of § 20, we conclude that y is an epsilon-number,
and in view of fp” = y we must have 2< f < y. On the other hand, if y is an epsilon-
number and 2< 6 < y, then, by theorem 1, of XIV.20, formulae (20.5) and (20.4),
we have fp? = y = By.
Thus we finally reach the conclusion that all the systems of positive ordinal num-
bers a, 6 and y satisfying equation a8’ = (a*)” are: 1) a= 1, arbitrary B and y, 2) a> 2,
Veal AL DULLalere psa os Cee 2,0 Pa 2, — Aa ee Da le <1), 1D) Oe
y = an arbitrary epsilon-number, 1 < B < y,

3. Find out for what systems of positive ordinal numbers a, 6, y we have a8” < (a’)”.
Solution. The above inequality may hold only for a > 1, y > 1 and then it is
equivalent to the inequality p’~* < y. The latter inequality holds of course for
B=1 y>1. Hy>o, then we have y-l=y and, in view of y>1 and of
formula (18.2), we obtain p’—* = B” > y which proves that for 6B > 1, y > @ our inequa-
lity does not hold. Therefore let us suppose that 6 > 1, 2<y< @. By means of an
easy induction we prove that 2” > m+ 1 for natural n. Thus B’~* > 2’ > y for B>1
and 2<y< wo and in this case our inequality does not hold.
Thus finally we reach the conclusion that inequality a6” < (a°)” holds for positive
ordinal numbers a, 6 and y if and only ifa>1, B=1, y > 1.

4. Find all ordinal numbers a > 0 such that ae® = (a”)’.

Solution. This equality holds of course for a= 1. For a > | it is equivalent to
the equality a? = a’, which (for a > 1) is equivalent to the equality a = 2. Thus our
equality holds only for the positive ordinal numbers | and 2.

28. Natural sum apd natural product of ordinal numbers. Let € and 7
be two given ordinal numbers. As we know (cf. the proof of A. Tarski’s
theorem in § 21) they can be represented in the form ¢ = wm, + wm, +
+... om, , (pire om, + oN, + jee oN, where ¢,,¢5,...,¢, are decreas-
ing ordinal numbers, and} m,, m2, ...,m, and 1, %,...,”, are integers >
> 0. Let

C(+)y = w(m, +m) + omy + Ne) +... + O'(M, + N,) 5

364 XIEV. Ordinal numbers

obviously number &(+)7 is an ordinal number well defined by § and 7.

After G. Hessenberg it is termed the natural swm of the ordinal numbers &
and 7 (Hessenberg [1], p. 591-594).
The natural sum of (a finite number of) ordinal numbers is of course
commutative and associative.
In order to form the natural product &(-)n of the ordinal numbers &
and 7 we multiply their normal expansions in such a way as if they were
polynomials of the variable w; multiplying two powers of number w we
form the natural sum of the exponents and arrange the terms obtained
from the multiplication according to decreasing exponents. The natural
product is commutative and associative.

EXERCISES. 1. Prove that for any ordinal numbers a and f we have a+ P<
<a(+)f.
2. Determine the smallest ordinal number a such that a+a < a(+)f.
Answer. a=—a-+1l. We have here a+a=q@-1+1 and a(+)a=o-2-+ 2.
3. Determine the smallest ordinal number a such that a(-)a < a-a.
Answer. a=o-+1. We have here a-a=o?+o+1, a(-)a=o@*+@-2+1.

29. Exponentiation of order types. Let a denote an arbitrary order

type, 6 — an ordinal number, A — an ordered set of type a, B* — an
ordered set of type £*. Let P = A®*; it will be the set all functions, defined
in the set B*, whose values are elements of the set A. If f, «-P, f,e P
and f, ~f,, then the set of those elements Db of the set B* for which
f(b) A f.(b) is non-empty. On the other hand, it is a set which possesses
a last element, ¢ (as a subset of the set B*, ordered inversely to a well-
ordered set); thus f,(¢c) €f,(¢). But f,(c) and f,(c) are elements of the
ordered set A. If f,(¢) <f,(c) in the set A, then we shall assume that f, < f.
in the set P. Otherwise let us assume that f, </, in the set P. It can easily
be proved that the accepted agreement orders the set P (according to
the principle of last differences) and that the order type 0 of the set P
ordered in this way depends only on the types a and / and is independent
of the individual choice of the sets 4 and B*, provided only that 4A = a
and B* = p*. In particular, if 6 is a natural number, then we obviously
have (according to the definition of the product of types P = af. Haus-
dorff denotes the type of the ordered set P obtained in the above way
by a. Thus the power a* is a definite order type for every order type a
and every ordinal number ep
The particular cases of that power deserve attention. H.g. let us
assume that a= P= w. Thus as A and B we can take the set of al! na-
tural numbers ordered according to their magnitudes. The set P will be
the set of all infinite sequences of type o*, formed of natural numbers
(t. e. sequences of the form..., 23, 2, ,) and ordered according to the
 § 29. Exponentiation of order types 365

principle of last differences. Thus it will obviously be a set similar to the

set of all ordinary infinite sequences (7. e. of type m), formed of natural
numbers and ordered according to the principle of the first differences.
Associating with each sequence of that kind, ,, n2, 73,..., a real num-
ber 20722 2> 7-9-2) 3 3 st we shall-_obviously obtain: “a similar
mapping of the set P on the set of all real numbers x such that 0 < # <1,
ordered according to decreasing magnitudes. Hence it easily follows
that w®** = 1+/ (Hausdorff [2], p. 48 and 73). Type w** is thus of the
power of the continuum, while type w® is denumerable.
It is also easy to prove that, generally, a*” — (a&*)*, Thus we have
OAT:
~o™

A closer investigation of types 2”, 7°. 2°° and (w*w)°” is left to the
reader. The last type can be proved to be the type of the set of all irra-
tional number, ordered according to their magnitude.
We shall also mention here another definition of the power of order
types, constituing a generalization of the definition of the power of
ordinal numbers given in XIV.14. Let a and w be two given types, A and JW
two ordered sets such that 4 = a, M = yw, and assume that the set M
has a first element, m,. Let us consider the set Z of all functions f(x)
defined for the elements « of the set A, whose values are elements of the
set M, the inequality f(r) #4 my holding only for a finite number (> 0)
of the elements # of the set A. Thus if 7, and f/, are two different functions
belonging to the set 7, then among the elements « of the set A, for which
f(x) A f.(x), there exists (in the’set A) a last,-a. Let us assume that
hfe if f(a) <fxa) (in the ordered set M), and /.<j, if f.(a)<f,(a)-
It is easy to verify that in this way the set Z will be ordered, and its order
type will depend only on types a and w. This could be accepted as a defini-
tion of the power uv*. According to this definition we should have, for
instance, 2°Qo* = 14-7 (while according to the definition given at the be-
5 Q@ *
ginning of this section type 2” is of the power of the continuum).
 CHAPTER XV

NUMBER CLASSES AND ALEPHS

1. Numbers of the 1-st and of the 2-nd class. If a is a given ordinal

number, then all sets of type a (being similar) are equivalent (of the
same power): the cardinal number corresponding to them is denoted
by a and called the power of number a. Since, as we know (XIV.1), a is
at the same time the type of the set of all ordinal numbers & < a (includ-
ing 0), a is the power of the set of all ordinal numbers <a.
Transfinite ordinal numbers (7. e. >q@) are divided into classes, all
numbers of the same power being assigned to the same class. All finite
numbers form the first class of ordinal numbers. Ordinal numbers a
constituting types of denumerable sets (7. e. such that a = s,) are assigned
to the second class. The smallest ordinal number of the 2-nd class is of
course number o.
We shall now deal more closely with the properties of numbers of
the 2-nd class. :
From the definition of sum, product and power of ordinal numbers
and from the properties of denumerable sets it easily follows that if a
and 6 are numbers of the 1-st or of the 2-nd class, then a+, af and af
are also numbers of the 1-st or of the 2-nd class.
From the axiom of choice it follows, as we know (see X.3), that
the sum of a finite or denumerable series of cardinal numbers < &, is
a cardinal number <x,. It follows that the sum of a finite or transfinite
series of type a, where a is a number of the 2-nd class, of nwmbers of
the I-st or 2-nd class is always a number of the I-st or of the 2-nd class;
for if

where a <8, and g: <s, for é<a, then

Co => yes Ve < No °

E<a

Using this fact we shall now prove

 § 1. Numbers of the 1-st and of the 2-nd class 367

THEOREM 1. The limit of an increasing infinite sequence of type w of

numbers of the 2-nd class is a number of the 2-nd class
Proof. Let
(1.1) A = limg,
n<@

where gy, (n=1,2,...) are numbers of the 2-nd class and

(1.2) Cite =O LOU pee Peel 52 ance
In view of (1.2) we have
(1.3) Cee Carne LOL 6 Sa 2 acs 5
whence gni1 = Gn +a,, therefore a, < G41 = ie
In view of (1.3) we easily verify by induction that for natural » > 1
we have 9,=9,+a,+a,+...+a,-1, Whence it follows that q, is the
n-th partial sum of the infinite series

Py + Oy + Ay +...

(for n= 1, 2,...), which, by (1.1), proves as we know (see XIV.8) that

A=@+%+a.4+...,
whence
Dero Hag Ossetia

and since gy, = No, and a, <x, for n= 1, 2,..., we have A= Sis, Ue (On Avis
a number of the 2-nd class. Theorem 1 is thus proved. The proof is based
on the axiom of choice, since it makes use of the theorem (which we are
unable to prove without the aid of the axiom of choice) stating that the
sum of an infinite series of cardinal numbers a, <s, is a cardinal num-
ber =.
It will be observed that we are unable to prove Theorem 1 without
the aid of the axiom of choice.
As regards the limit of an increasing infinite sequence of type w
of numbers of the 1-st class, it is of course always number wo.
. From the theorem stating that the sum of a denumerable series of
numbers of the 2-nd class is always a number of the 2-nd class it im-
mediately follows that for every denumerable set of numbers of the 2-nd
class there exists a number of the 2-nd class greater than any number of
that set.
As an immediate conclusion we obtain
THEOREM 2. The set of all numbers of the 2-nd class is non-denumerable.
The proof of Theorem 2 given above has been based on the axiom
of choice. We are able, however, to prove Theorem 2 without the aid
ot the axiom of choice, namely in the following way.
368 XV. Number classes and alephs

Suppose that the set Z, of all numbers of the 2-nd class is denumer-
able. Therefore the set Z = Z,+Z, of all numbers of the 1-st and the 2-nd
classes would also be denumerable. The set 7, as a set of ordinal numbers,
is well ordered (according to the magnitude-of those numbers). Let us
denote its type by 2; it will be an ordinal number, and from our assump-
tion that Z =», it follows that Q=x,, 7. e. that Q is a number of the
2-nd class, which means that it is an element of the set Z. But every
ordinal number is known to be the type of the set of all ordinal numbers
smaller than that number; thus each number of the set Z is the type of
all numbers of the set Z which are smaller than that number, 7. e. the
type of the segment which it forms in the set Z Thus number 2, which
is the type of the well-ordered set Z, would likewise be the type of a cer-
tain proper segment of that set. Thus the well-ordered set Z would be
similar to a certain proper segment of itself, which is known to be impos-
sible (see XIV.5, Corollary 5). Therefore the set Z, cannot be denumer-
able. And being infinite, as it contains a denumerable subset composed
of numbers m+n where n= 0,1,2,..., it is (effectively) non-denumer-|
able, q.e: d.
From Theorem 2 it also immediately follows that for every denumer-
able set of ordinal numbers of the 2-nd class there exists a number of
the 2-nd class different from any number of that set. This conclusion
we are also able to prove without the aid of the axiom of choice. But we
cannot prove without the aid of that axiom that for every denumer-
able set of numbers of the 2-nd class there exists a number of the 2-nd
class greater than any number of that set.
For ae Z, let us denote by A(a) the set of all ordinal numbers <a;
it will be a set of type a, 7. e. a denumerable set. Let F denote the family
of all sets A(a) for ae Z,. For ae Z,, Be Z, a< 6 we shall have A(a)C
C A(f). The sum of all the sets of family F will of course be the set of
all ordinal numbers of the 1-st class and of the 2-nd class, 7. e. it is a non-
denumerable set. Thus there exists a family F/ of denumerable sets such
that for each two different sets of family F one is contained in the other
and the sum of all the sets of family F is a non-denumerable set.
It is also easy to give an example of a family with the same properties
formed of 2” different denumerable sets. Such is the family of all proper
segments of an ordered set of type 72 (where 2 denotes the order type
of the set of all ordinal numbers of classes one and two, ordered according
to their magnitude). From the formula 7a = 7 for every ordinal number > 0
of the first and of the second class (following from Exercise 5 of XI.3)
we easily conclude that each element of a set of type 72 determines
a segment of type 7 of that set. The type 7 is known to have similar pro-
perties. Thus from the fact that each element of a given ordered set,
 § 1. Numbers of the l-st and of the 2-nd class 369

having no last element, determines a segment similar to a segment

determined by an arbitrary element of another ordered set, also having
no last element, we cannot infer either that those sets are similar or that
they are of the same power (since a set of type

72= »'n,
a<Q

being the sum of a non-denumerable aggregate of non-empty sets, is

a non-denumerable set).
The type 4+ (1+) also has an interesting property. Hach element
of a set of this type determines a segment of type 2 just as each element
of a set of type A does. Still we have 1+(14+A4)2 44, since a set of type
A+(1+A)Q2 contains a non-denumerable non-overlapping intervals of
type 1+, while a set of type 4 has no such property.
Another question: what is the condition necessary and sufficient in
order that for a given order type y there exist a family F of type of
denumerable sets such that for each two different sets of family F one is
contained in the other? It can be proved that this condition consists in @
being the type of a subset of a set of type (1+ 4)Q (Sierpiiski [15], p. 242).
2. Cardinal number s,. The power of the set Z, of all numbers of
the 2-nd class is denoted by s, (read aleph-one). Since the set Z, contains
a denumerable subset (formed for example by the numbers w+ for
n=1,2,...), and, by Theorem 2 of XV.1, it is not denumerable, we have
(2.1) iS es Ray.
Now we shall prove without the aid of the axiom of choice
THEOREM 1. Between s, and 8, there is no cardinal number.
Proof. Suppose that there exists a cardinal number m satisfying
the inequality
(2.2) Ngee lien)
and let M denote a set of power m. In view of M =m, Z,=xs, and of
(2.2), the set M is of the same power as a certain subset 7 of the set Z,.
But the set 7, being a subset of the well-ordered set Z,, is similar either
to the set Z, or to a certain proper segment of that set. The first case
is impossible since we should then have m= M = Z,=y,, which is
contrary to (2.2). The second case is, by (2.2), also impossible, each proper
segment of the set Z, being of power <8, (since a segment formed by
an element a of the set Z, is a subset of the set of all ordinal numbers <a,
which is of type a, 7. e. of power 8, a being a number of the 2-nd class).
Thus there exists no cardinal number m satisfying the inequalities (2.2),
q.e. d.
Cardinal and ordinal numbers 24.
370 XV. Number classes and alephs

In view of Theorem 1 we may regard number s, aS the next cardinal

number following s,. But we have not proved yet that there is no cardinal
number n, different from s,, which could also be regarded as the next
after S$), 7. e. that there is no cardinal number n for which there is no
cardinal number m satisfying the inequalities s, < m <n. This property
of number s, will be proved later with the aid of the axiom of choice
(see XX.VLob).
From Theorem 1 it immediately follows that each non-denumerable
subset of a set of power s, is of power s,. A set of power s, is thus equi-
valent to each non-denumerable subset of itself. With the aid of the
axiom of choice we shall prove later (Chapter XVI, p. 412) that also con-
versely: a set that is equivalent to each non-denumerable subset of itself
is of power s,.
Now we shall prove with the aid of the axiom of choice
THEOREM 2. §, < 2".
Proof. By Theorem 1 of IIl.2 we are able to define an infinite
sequence
(2.3) Tae Gaetces

formed of all (different) rational numbers. Let us divide the set X of all
real numbers into s, disjoint subsets as follows.
For a given real number «x let us denote by
Wap nD i) aa)

its expansion (unique, aS we know) into an infinite dual fraction, contain-

ing infinitely many digits different from 0. Thus 7,(#), (a), ... will be
an increasing infinite sequence of natural numbers, well-defined by
number @. -
Now let us consider the infinite sequence
(2.4) Ping) 9 Vre(x) 9 sony

taken out from sequence (2.3). The set of all terms of sequence (2.3)
ordered according to their magnitude is a certain denumerable order
type g(x). If p(x) is not an ordinal number, then we shall assign number »
to the set X, and if g(x) is an ordinal number a (of the 2-nd class of course,
since the set of numbers (2.4) is denumerable), then we shall assign num-
ber x to the set X,. We shall have of course

(2.5) Kap), Xe 4 Wee ek ee

aéZ2,

where the sum is extended to all ordinal numbers a of class two. Thus
there are aS many components in sum (2.5) as there are ordinal numbers
 § 2. Cardinal number ys, 371

of the 2-nd class, 7. e. s,, the components of sum (2.5) obviously being
disjoint. Finally, in order to prove that they are non-empty we use
theorem 2 of VIII.8 according to which the subset of the set of all rational
numbers, ordered according to the magnitude of the numbers, can re-
present any denumerable order type. Thus for any ordinal number a
of the 2-nd class there exists an increasing infinite sequence of natural
numbers 71, .,... such that the set {7q,,1,,-..}, ordered according to
the magnitude of the numbers that belong to it, is of type a. Writing

Ee ED ES le
we shall obviously obtain a real number belonging to the set X,. Thus
we have X, £0 for ae Z,. Thus we have proved (without resorting to
the axiom of choice) the following
THEOREM 3. We ave able to give an effective decomposition of the
set of all real numbers into s, non-empty disjoint sets *).
Using the relation <-« introduced after A. Tarski in VIII.4, we im-
mediately obtain the following
COROLLARY. We are able to prove without the aid of the axiom of choice
that s; <2".
The decomposition (2.5) given by H. Lebesgue [1] (p. 213) has been
investigated in detail by N. Lusin [1] (p. 1-95) in his extensive work on
analytic sets. As to the nature of the components X, of the sum (2.5),
Lusin has proved that for a> oq they are B-measurable (7. e. they can
be obtained from intervals by means of a denumerable aggregate of addi-
tions and multiplications), while the set X, is not B-measurable but it
is an analytic set (7. e. a projection of a B-measurable plane set, or,
which is the same, a continuous image of the set of all irrational numbers).
It could also be proved that, with a suitable choice of the index a,
X, may be a set whose class (in the classification of B-measurable sets)
is higher than an arbitrary ordinal number of the 2-nd class.
In virtue of the axiom of choice there exists a set N containing one
and only one element from each of the sets forming the components of
series (2.5); the set N is of course equivalent to the set of all components
of series (2.5), 7. e. it is of power s,. Thus we have proved with the aid of
the axiom of choice that there exists a set of real numbers of power s,,
whence follows Theorem 2.
I have given another effective decomposition of the set of all real
numbers into s, non-empty disjoint components in Fund. Math. 29 (1937),
p. 2-3.

1) Tt can easily be shown that each of those sets (in the decomposition (2.5)) is
of the power of the continuum (Church [1], p. 183).
24%
372 XV. Number classes and alephs

We are unable to prove the inequality s, <2” without resorting

to the axiom of choice. Neither do we know an effective example of a set
of real numbers of power s,. But we are able to prove without the aid
of the axiom of choice that
(2.6) Rie
The sets X,, constituting the components of sum (2.5), are different
subsets of the set X of all real numbers, 7. e. the set of the components
of series (2.5) is a subset of the set of all subsets of the set X, which is
of power 2* = 22°, whence follows the inequality s, < 22°, the sign = being
impossible here since then we should have s, = 27° > 2° > s,, which is
contrary to Theorem 1, proved without the aid of the axiom of choice.
Thus we have inequality (2.6).
It is also easy to prove without the aid of the axiom of choice that

(2.7) Qs < 92% |

Indeed, for each subset 7 of the set Z, (which is known to be of
power s,) let us denote by (7) the sum

p(L)
= SX,
aeT

where X, are componenets of sum (2.5). Since the Components of sum

(2.5) are disjoint and non-empty, to different subsets 7 of the set Z,
will obviously correspond different sums S(T), 7. e. different sub-
sets of the set X, whence it immediately follows that U(Z,) < U(X),
which gives inequality (2.7).
Let us also observe, after A. Tarski, that if the inequality s, <2”
were not true, then we should be able to decompose effectively the set
of all real numbers into more than the continuum of non-empty disjoint
subsets. Indeed, formula (2.5) makes it possible to decompose effectively
the interval (0, 1) into s, of disjoint non-empty sets. On the other hand,
we are also able to decompose effectively the set of all real numbers not
belonging to the interval (0,1) into 2° disjoint non-empty sets. In this
way we obtain a decomposition of the set X of all real numbers into
m= ¥,+2" disjoint non-empty sets. But of course s,-+-2™ > 2", and we
cannot have x,+ 2° = 2” if the formula s, < 2“ igs not true. Thus if the
formula x, < 2° is not true, then we have m> 2”, i. e, a decomposition
of the set of all real numbers into more than the continuum of disjoint
non-empty sets.
THEOREM 4. The following two propositions, P, and P,, are equi-
valent:
 § 2. Cardinal number x, 373

P,. There exists a function f associating with every denumerable set D

of real numbers a real number f(D) ¢ D.
P,. There exists a set of real numbers of power s, (Sierpiiski [66],
p. 53-54).
Proof. P,—P,. Suppose that proposition P, is true and let D, =
= {1, 2,....}, f(D.) = %- Let a denote a given ordinal number, w <a,
aeZ, and suppose that we have already defined the real numbers 2;
foro <€< a. The set D,, composed of all natural numbers and of numbers
_ae Where € < a, é« Z,, is denumerable (since a « Z,): % = f(D.) will thus be
a defined (by the function f) real number. In this manner numbers x, are
defined, for ae Z,, by transfinite induction. For a<f, ae Z,, pe Z,
we have 7,= f(D.) «Dg and since, in view of the properties of func-
tion f, x ¢ Dg, we have «a, ~ av; thus the terms of the transfinite sequence’
{vq\yez, are all different from one another. The set of all numbers x,
where ae Z, is of the same power as the set of all ordinal numbers « Z,,
2. €. 18 Of power s,. We have proved that P,—P,.
P,—P,. Suppose that proposition P, is true; there exists a set A of
real numbers of power s,, 7. e. equivalent to the set of all ordinal numbers
ae Z,. Thus there exists a transfinite sequence {X}aez,, composed of
all numbers of the set A. Let D be an arbitrary denumerable set of real
numbers. We cannot have AC PD because the set A, being of power s,,
is non-denumerable. Therefore A—D~0, and thus in the sequence
{Gataez, there exist terms that do not belong to D. Let us denote by f(D)
the first of those terms. We shall have /f(D)é¢ D. Thus the function f
associates with every denumerable set D of real numbers a certain real
number that does not belong to D. We have proved that P,—P,.
The equivalence P, = P, is thus proved (without the aid of the axiom
of choice).
It will also be observed that, modifying slightly the above proof,
we could prove that
The following propositions Py and Pz are equivalent:
Pi. We are able to define a function f associating with every denume-
rable set D of real numbers a certain real number f(D) D.
Ps. We are able to define an ordered set of type 2 of real numbers.
Moreover, it will.be observed that, as has been proved in III.6, we
are able to define a function associating with every infinite sequence
of real numbers a certain real number which is not a term of that
sequence.
Basing ourselves on decomposition (2.5) we shall now prove that;
we are able to associate with every at most denumerable set A of sets of real
374 XV. Number classes and alephs

numbers a set of real numbers f(A) different from each of the sets contained
an the set A.
Indeed, let 4 denote a given at most denumerable set of sets of real
numbers. Since the components X, of sum (2.5) are disjoint, non-empty,
and their number is s,, while the set 4 is at most denumerable, there
exist among them components which do not belong to the set 4. Let
mw denote the smallest ordinal number (>) such that X,¢ A and let
f(A)= X,. Thus f(A) will be a set of real numbers not belonging to
the set A.
Hence it easily follows that we are able to associate with every na-
tural number n and with every at most denumerable set A of sets of real
numbers, contained in the interval J,=[n—l<a<n], a set of real
numbers f,(4;,), contained in the interval J, and different from each of
the sets belonging to the set A,.
Now let F denote the set of all sets of real numbers and let us sup-
pose that F is the sum of a denumerable aggregate of denumerable sets
of sets of real numbers. Thus there would exist an infinite sequence
Gi, Go,..., where G, (n=1,2,...) is @ denumerable set of sets of real
numbers, such that F=G,+G,+... For n=1,2,..., let us denote
by #H, the set of all sets-1,4 where 4 <G,. The sets A, (v= 1,27)
will of course be at most denumerable sets of sets of real numbers of the
interval J,. Now let
(2.8) Q = fi( Hy) + foe) +... :
we shall of course have Qe; therefore, in view of F = G,+G,-+...,
there exists a natural number » such that @ « G,, whence in view of the
definition of the set H,, we have J,Q « H?. But from the definition of
the function f, it follows that /,(H,)¢H, and f,(H,) «J, and, the inter-
vals I,, I,,... being disjoint, formula (2.8) gives J,Q = f,(H,). Thus we
should have J,Q ¢ H,, which is contrary to the formula J,Q « H, proved
above.
Thus we have proved without resorting to the axiom of choice A. Tar-
ski’s theorem stating that the set of all sets of real numbers is not the sum
of a denumerable aggregate of denumerable sets whose elements are sets of
real numbers.
In VI.7 sets constituting sums of a denumerable aggregate of de-
numerable sets have been termed D,-sets. Therefore the above theorem
can also be expressed as follows:
We are able to prove without the aid of the axiom of choice that the set
of ‘all sets of real numbers is not a D,-set.
Analogously, we could prove without the aid of the axiom of choice
that a set of power 2% is not a D,-set. For this purpose we should base
 § 2. Cardinal number 3, 375

ourselves on the observation that we are able to decompose effectively

the set of all ordinal numbers of the 2-nd class into an infinite sequence
of disjoint sets of power s,, assigning to the k-th component, for k=
= 0,1,2,..., all ordinal numbers of the form w&+k where é is an arbi-
trary ordinal number of the 1-st or of the 2-nd class >0. For the rest,
the proof would proceed analogously to the proof for sets of power 22°.
EXERCISES. 1. Prove (without the aid of the axiom of choice) that
NiNo = Si
Proof. The set A of all numbers of the form w&+ k where e Z, and k= 0, 1, 2,...
is obviously equivalent to the set of all systems (&,k) where é«Z,, k=0,1,2.,...,
therefore, in virtue of the definition of the product of cardinal numbers, it is of power
ZaNo =,N,- But the set A is of course a subset of the set Z,; thus A <,. Hence y,§, <
<,, and since, on the other hand, we have of course ¥, < 8,N,, therefore $,8) = Si,
Ts ils
Since Si < NHt+N <N +N, = ° 2<8,8, it follows from the equality
NiNo =, that ys, +8, = ,. But so+s, = TET. the set of all ordinal numbers of
the 1-st class and of the 2-nd class is thus of power y,; if Q denotes the order type of
that set ordered according to the magnitude of the numbers that belong to it, then Q=\,.
2. Prove without the aid of the axiom of choice that 2 > y*.
Proof. In virtue of Exercise 1 we have y, =8,N,. Hence, in view of 2 >8,,
we obtain 2°! — Q*1*0 — (2%1)*0 > yf, therefore 2°! > y%, g. ec. d.
3. Prove without the aid of the axiom of choice A. Tarski’s theorem stating that
a set of power s¥° is not a D,-set.
Proof. As a set of power s}° we can obviously take the set H of all infinite sequences
(a, dz, ...) Whose terms are ordinal numbers of the 1-st or of the 2-nd class. Suppose
that the set H# is a D,-set. Thus there exists an infinite sequence H,, #,, ... of denumer-
able sets, formed of the elements of the set #, such that H = H,4+ H,-+...; thus #, will
be a denumerable set of infinite sequences of ordinal numbers of the 1-st or of the 2-nd
class. As we have proved in VI.7, p. 119, we are able to prove without the aid of the
axiom of choice that there exists an infinite sequence containing each term of each of
the sequences forming the set H,. Hence it immediately follows that the set of all
different ordinal numbers constituting the terms of the sequences of the set #,, is finite
or denumerable and since the set of all ordinal numbers of the 1-st class and of the
2-nd class (as has been proved without the aid of the axiom of choice in XV.1, Theo-
rem 2) is non-denumerable. there exists a smallest ordinal number u,, of the 1-st
or of the 2-nd class, that is not a term of any sequence of the set #,. Let f(n) = ,-
In this way the function f will be defined for every natural number n (and dependent
only on the sequence #,, #,, ..., which can be proved to exist without the use of the
axiom of choice). The infinite sequence a = (,, fe, ...) belongs of course to the set H
and therefore, in view of the formula H = H#,+ H,--... and of the definition of the sets #,,,
there exists a natural number m such that a « H,,. Number y,, would thus be a term of
a certain sequence of the set H,,, in contradiction to the definition of numbers y,.
Thus the assumption that the set # is a D,-set results in a contradiction. The re-
quired proof is complete.
It will be observed that from formula (2.7) and exercises 2 and 3 it immediately
follows that we are able to prove without the aid of the axiom of choice that a set of
power 2“! is not a D,-set (which we have found before on different lines).
376 XV. Number classes and alephs

3. 8, = 2° hypothesis. In V.2, the term continuum hypothesis has

been given to the supposition that there is no set of power greater than
denumerable and smaller than the continuum, 7. e. the supposition that
there is no cardinal number m such that

(uh) Pat oa
Let. us denote this supposition by H. By Theorem 2 of XV.2 and
formula (2.1) it follows from H that we must have

(3.2) cee
since otherwise s, would be an intermediate cardinal number between Xo
and 2°
On the other hand, from formula (3.2) and Theorem 1 of XV.2 it
follows that there is no cardinal number 1m satisfying inequality (3.1),
which means the same as hypothesis H.
Hypothesis H is thus equivalent to equality (3.2). However, our
proof of this equivalence has been based on the axiom of choice since
we have used in it Theorem 2 of XV.2, proved with the aid of
the axiom of choice. Without the aid the axiom of choice we are unable
to prove that hypothesis H implies formula (3.2).
THEOREM 1. The hypothesis that 2°—vs, is equivalent to the theorem
stating that the set of all points in a plane is the swm of two sets of which
one is at most denumerable on every parallel to the axis of abscissae and
the other is at most denumerable on every parallel to the axis of ordinates
(Sierpinski [5] and [12], p. 179; Tietze [1], § 18, p. 307).
Proof. Let us assume that 2°°=—x,. The set X of all real numbers
is thus of power s, and therefore is equivalent to the set Z—= Z,+Z,
(of type 2) of all ordinal numbers of the 1-st class and the 2-nd class
(ordered according to their magnitude). Thus there exists a transfinite
sequence of type 2

(3.3) fhe ne cle bare tae) En eee

formed of all the different numbers of the set YX.
Let us denote by A the set of all points in the plane P with the
coordinates (f.,t%), where a<f6 <Q and let B= P—A.
Let b denote a given real number; therefore it is one of the terms
of sequence (3.3), e.g. b=t, where BeZ. Points of the set A whose
ordinate is b are the points (t¢, ts) where é < f, and since f « Z, i. e. B is
an ordinal nuntber of the 1-st or of the 2-nd class, the set of all ordinal
humbers </ is of type 6 and thus at most denumerable, whence it fol-
lows that the line y = b contains an at most denumerable aggregate of
 § 3. §, = 2%e hypothesis 377

points of the set A. On the other hand let a denote a given real number,
é. g. @=t, where ae Z. Points of the set B= P—A with the abscissa a
are obviously the points (f,, t2) where & < a; in view of ae Z we conclude
that the line « = a contains an at most denumerable aggregate of points
of the set B.
Thus from the assumption that 2°°—s, follows the existence of
a decomposition of the plane P= A+B satisfying the required condi-
tions. Now we shall prove that, conversely, the existence of such a de-
composition implies the formula 2°° = x,.
Let us assume that 2° ~s,. By Theorem 2 of XV.2 we should have
2" — sy, and there would exists a set Q formed of x, lines parallel to the
axis of abscissae. Let us denote by N the set of all points of the set A
lying on the lines forming the set Q. Since each of the lines belonging
to Y contains av at most denumerable aggregate of points of the set A
‘(according to the properties of the set A), and the set Q contains ys, lines,
we easily conclude with the aid of the axiom of choice that the set N is
of power <,Nq, 7. €. invirtue of the Exercise 1 in XV.2, of power <y,,
whence, by Corollary 3 of theorem 1, Chapter VI, we conelude that
also the projection of the set N on the axis of abscissae is of power <x.
Thus in view of 2° >, there exists a-point x, of the axis of abscissae
that is not a projection of any point of the set N on that axis. Hence
it follows (in view of the definition of the set N) that each point of
intersection of the line x= a, with the parallels to the axis of abscissae
belonging to Q belongs to the set P—A = B. The line x= x would thus
contain a non-denumerable aggregate of points of the set B, which is
in contradiction to the properties of that set. Therefore we must have
2° — 9,
‘Theorem 1 is thus proved (with the aid of the axiom of choice).
As has been observed by N. Lusin, this theorem can also be ex-
pressed in a different way. Every set of points in a plane that has one
and only one point in common. with every parallel to the axis of ordinates
is called a curve of the form y = f(x), and every set of points in a plane
that has one and only one common point with every parallel to the axis
of abscissae is called a curve of the form x = f(y). Theorem 1 may obviously
be expressed as follows:

The continuum hypothesis is equivalent to the theorem stating that

a plane may be covered with a denumerable aggregate of curves which
are either of the form y= f(a) or of the form «= f(y) each.

Now let us call a curve every set of points in a plane that is con-
gruent (by translation or rotation) to the graph of a one-valued function
of a real variable (7. e. to a curve of the form y = f(#)). It is by no means
378 XV. Number classes and alephs

easy to prove that a plane is not the sum of a finite number of curves:
it was proved in 1933 by S. Mazurkiewicz [2]. In the same year I
proved with the aid of the continuum hypothesis (using Theorem 1)
that a plane is the sum of a denumerable aggregate of curves congruent
to one another, and also that there exists a function of a real variable (x)
whose graph y = y(#) may be divided into a denumerable aggregate of
fragments (corresponding to the intervals k <#<k+1) with which by
suitable changes of position (by translation and rotation) we can cover
the whole plane (Sierpinski, Fund. Math. 21 (1933), p. 39-42).
It has not been resolved whether from the assumption that a plane
is the sum of a denumerable aggregate of curves follows the continuum
hypothesis.
Without the aid of the continuum hypothesis we are able to prove
that a plane is not the swm of two sets, one of them finite on every parallel
to the axis of abscissae and the other at most denwmerable on every parallel
to the axis of ordinates (Tietze [1], p. 291, and Sierpinski [59], p. 6-7).
We can also prove without the aid of transfinite numbers that the
continuum hypothesis is equivalent to the proposition that every plane
set that is not of the power of the continuum is the sum of two sets of which
one is finite on every parallel to the axis of abscissae and the other finite
on every parallel to the axis of ordinates (Sierpinski [63], p. 297).

THEOREM la. The continuum hypothesis H is equivalent to the the-

orem Q, stating that the three-dimensional Huclidean space E is the sum
of three sets H; (¢ = 1, 2, 3) such that if we denote the Cartesian coordinates
axis inthe ‘space’ EH by OX, (4 =F, 253), then, for v— 1, 2, 5, the
set H; is at most denumerable on every plane perpendicular to the axis OX;
(Sierpiiski [59], p. 11, Theorem 8).

Proof. H+Q. From hypothesis H it follows (cf. the begining of

the proof of Theorem 1) that there exists a transfinite sequence (3.3) of
type 2, formed of all different real numbers. Let H, denote the set
of all those points (¢,, ts, t,) of the space # for which a> 6 and a> y,
Ei, — the set of those points for which a<f and 6 >y and #, — the
set of those points for which a < y and 8 < y. We shall of course have
HK = #,+H,+8#,.
The set H, is at most denumerable on every plane perpendicular to
the axis OX, = OX. Indeed, let « = a be such a plane. Since the trans-
finite sequence (3.3) contains every real number, there exists an ordinal
number a <2 such that a=t,. Let p= (a,y,2) denote a point of. the
set H, lying in the plane «=a. Since all the terms of sequence (3.3)
(which contains every real number) are different from one another, there
exist two ordinal numbers 6 < Q and y <2, well-defined by the point p,
 § 3. §, = 280 hypothesis 379

such that y=%t%, z=1t,. In view of pe H, and of the definition of the

set E, we easily conclude that 6 <a and y <a. In view of a< the
set of all systems.of ordinal numbers f,y such that 6 <a and y <a,
as a set of power (a+1)?, is at most denumerable. It follows that
the set of all points (a, y, 2) of the set H, is at most denumerable. Thus
we have proved that the set H, is at most denumerable on every plane
perpendicular to the axis OX.
Analogously, we prove that the set FH, is at most denumerable on
every plane perpendicular to the axis OX,= OY, and that the set FH,
is at most denumerable on every plane perpendicular to the axis OX;=
= OZ. Thus the sets EH; (¢ = 1, 2,3) satisfy theorem Q. We have proved
that H+Q.
Q—H. Suppose that theorem Q is true and hypothesis H is false,
td. e. 2°° > s,. Let N denote a set of real numbers of power s,. For each
number a « N the set of all points (a, y, 0) of the set H, is at most denumer-
able. The set G of all points (a, y, 0) of the setH, where ae N is thus
of power <x pk8,, 7. €. in virtue of the Exercise 1 in XV.2, of power <x, < 2”,
and the projection 7 of the set G on the axis OY is (according to Corol-
lary 3 of VI.7) of power <2". Thus there exists a point (0,0, 0) of the
axis OY that does not belong to 7. Hence it follows that the plane y = b
does not contain any point of the set @. Therefore, for 7« N, we have
(7, b,0)¢ #,, and thus (z, b,0)« H,+#, for we N. Since the set of all
points (7, b,0) where we N is of power s,, and hence non-denumerable,
at least one of the sets #, and #, contains non-denumerably many points
(x, b, 0) each of those points lying of course both in the plane y = b and
in the plane z = 0, whence easily follows a contradiction of the assump-
tions of theorem Q with regard to the sets H, and H,. Thus the assump-
tion that theorem Q is true and hypothesis H is false results in a contradic-
tion. Hence it follows that Q—-H.
Theorem la is thus proved.
Later on (in XV.9) we shall give some more theorems equivalent to
the continuum hypothesis.
4. Properties of ordinal numbers of the 2-nd class. In XV.1 we have
proved that the limit of an increasing infinite sequence (of type w) of
numbers of the 1-st or of the 2-nd class is a number of the 2-nd class,
of the 2-nd kind of course (which follows directly from the definition
of the limit of an increasing transfinite sequence of ordinal numbers).
We shall now prove the inverse theorem, namely
THEOREM 1. Hvery number of the 2-nd class and of the 2-nd kind is
the limit of a certain increasing infinite sequence (of type wm) of numbers
of the L-st or of the 2-nd class.
380 XV. Number classes and alephs

Proof. Let a denote a given number of the 2-nd class and of the
2-nd kind. The set of all ordinal numbers é < a is thus denumerable and
does not contain a greatest number. Let

(4.1) Ex, S95 $39:++: 5

denote an infinite sequence formed of all those numbers. Let a = &,.
Now let » denote a given natural number and suppose that we have
already defined a number a, aS a certain term &, of the sequence (4.1).
Since the sequence (4.1) does not contain a greatest term, there exist
numbers >a, in it; let a,;, denote the first of them. In this way we shall
obtain an increasing infinite sequence of ordinal numbers
(4.2) Or lain Oe eae
obviously by taking a, = ¢,, for n= 1, 2,... we-shall have k, < k, < ...
and for any natural n the term &,, of the sequence (4.2) will be an ordinal
number greater than any preceding term of that sequence. Thus if &,
denotes any given term of the sequence (4.2), then for sufficiently large n
we shall have k,>p and thus a,= &, > &. Since the sequence (4.2)
contains all ordinal numbers <a, it follows that none of them is greater
than any term of the sequence (4.2); the number a (which is of course
greater than any number of the sequence (4.2)), taken out the of sequence
(4.1), is thus the least ordinal number that is greater than any number
of the sequence (4.2). In view of the definition of the limit of a transfinite
sequence of ordinal numbers, we have

(4.3) == iM ar
n<w

which proves the validity of Theorem 1.

Thus for every number of the 2-nd class and of the 2-nd kind there
exists an infinite sequence (of type w) of numbers smaller than the num-
ber in question, that number constituting its limit; we are unable, however,
to establish the law according to which to every number of the 2-nd class
and the 2-nd kind would correspond one such sequence, neither can
we establish the law according to which to every number of the 2-nd
class would correspond an infinite sequence (4.1) formed of all ordinal
numbers smaller than the number in question. We have
THEOREM 2. If we were able to define a function associating with
every ordinal number a of the 2-nd class an infinite sequence formed of all
ordinal numbers <a, then we should be able to define effectively a set of real
numbers of power s,.
Proof. Let a denote an ordinal number of the 2-nd class, and a,, a...
a sequence associated with if which is formed of all ordinal numbers < a.
Let us denote by N(a) the set of all natural numbers n = 24-1(21—1)
 § 4. Properties of ordinal numbers of the 2-nd class 331

for which o, < a. We shall have N(a) 4 N(f) for different numbers a
and f of the 2-nd class since if (,, /,,... is a sequence, associated with
number f, of all ordinal numbers </, then, in the case of N(a) = N (fp),
the formula oa, < q, for any natural k and /, would be equivalent to the
formula $6, < 6;, which is impossible because the set {a,, a, ...}, ordered
according to the magnitude of the numbers that compose it, is of type a
while the set {6,, B., ...} is of type $8 4 a. Now let

ie)
= 2 Eo n n
oa n
)

where ”,, %.,... are numbers of the set N(a) ordered according to their
magnitude. We shall have «(a) £ «(f) for different numbers a and f/ of
the 2-nd class. The set of all real numbers «x(a) corresponding to the
ordinal numbers a of the 2-nd class is thus of power ys, 1).
Let it also be observed that A. Church has deduced the formula
s, < 2 without the aid of the axiom of choice from the assumption that
there exists a function associating with every ordinal number a of the
2-nd class an infinite sequence with the upper bound a (Church [1], p. 191,
Hardy [1], p. 87-94, Kurepa [6], p. 175-176).
Finally, it will be observed that it is possible to prove that if we know
a function f(a) establishing a (1-1) correspondence between the set of all or-
dinal numbers of the first and of the 2-nd class and the set of all real num-
bers, then we should be able (with the aid of the function /) to define a func-
tion g(a), associating with every ordinal number a of the 2-nd class
a certain infinite sequence a,, a,,..., formed of all ordinal numbers < a
(Sierpinski [59], p. 4).
From Theorem 1 it easily follows that every number of the 2-nd
class and of the 2-nd kind is the sum of an infinite series (of type w) of
positive ordinal numbers smaller than the number in question. Indeed,
if a is a number of the 2-nd class and of the 2-nd kind, then we have
TORMIIIA (4,5), “were: dg, <0, < .... Thus, for 7 ='2, 3, 2250p = Gna +p,
where 6, >0 for »—2,3,..., and a,+/.+ f.+...+6,—= a, for n=
= 2,3,..., whence we conclude that
a= lima, = a,+$,+ 63+...
n<w

If a set whose elements are numbers of the first or of the 2-nd class,
ordered according to the magnitude of the numbers that compose it,
gives an increasing transfinite sequence of type 2, f{as}ecqg such that

him (a¢+4 = ae) =? ;

E<Q

then we shall call that set sparse.

1) Another proof of Theorem 2 was given by P. Finsler [1], p. 78.

382 XV. Number classes and alephs

It can be proved (with the aid of the axiom of choice) that the set Z
of all ordinal numbers of the 1-st and of the 2-nd class is the sum of
a denumerable aggregate of sparse sets. We are unable, however, to
decompose the set Z effectively into the sum of an infinite series (of
type w) of sparse sets; it can be proved that if we were able to do it, we
should also be able to define a function associating with every ordinal
number a of the 2-nd class a certain infinite sequence formed of all
ordinal numbers <a (see Sierpinski [33], p. 289-296).
EXERCISES. 1. Prove the following theorem:
For every transfinite sequence of type 2, (dg}ecg whose terms are ordinal numbers < 2,
and for every ordinal number uw < 2 there exists an ordinal number 4 of the 2-nd hind
such that w<24< 2 and
(4.4) dp < A for Ns

Proof. Let {a,}.-9 be an transfinite sequence of ordinal numbers <Q, and yu

a positive ordinal number <Q. Let

(4.5) Be = wc ey +t+yu+é for &é<Q.

(Be}ecq Will of course be an increasing transfinite sequence of ordinal numbers,

and we shall have l < Be < 9 for 1< & < 2. Moreover, let

(4.6) Y= Pr and Ys,=— fo, for n=1, 2,...

Since f, > 1 and the sequence {f,},_9 is increasing, y, = By, > Py = V1, t. & Y2 > V-
Now let » denote a given natural number and suppose that y,., > y,. Since sequence
(4.5) is increasing, therefore, by (4.6), we have y,.. = ee Si Vane this way
we have proved by induction that y, << y.< ...
By (4.6) and since numbers (4.5) are ordinal numbers <2 we conclude that the
ordinal numbers y, (n= 1, 2,...) are also <Q. Thus, by Theorem 1, of XV.1 there
exists a number
(4.7) A = limy, <a Q
Now
and

(4.8) Lb > Nps HOE @ Hi S2 ls Ba coc

Now let € denote an ordinal number <A. By (4.7) and (4.8) there exists a natural
number m such that € < y,, whence B, < Pap =Ymi1 <A and, by (4.5), we naturally
have a, < A. Formula (4.4) is thus proved.
In an analogous way we can prove the following theorem:
Tf f(&) is @ function associating with every ordinal number € an ordinal number f(é).
then for every ordinal number wu there exists an ordinal number 2 > yw such that

HEPA fore Seas

2. Prove the following theorem:
Lf f(&,) ts a function associating with every ordered pair &,n, formed of ordinal
numbers, an ordinal number f(&, 7). then for every ordinal number u there exists an ordinal
number 2 > w such that
(4.9) T(é,
Wr =7 /\ xA for Bee Win lir< The
 § 4. Properties of ordinal numbers of the 2-nd class 383

o
Hint. To prove this it is sufficient to write, for ordinal numbers 6,

set) B= dD f(E, n)tu+¢

EC
nxt

and to define an ordinal number 4 as in the solution of exercise 1. We shall have Be <A
for € < A, whence, by (4.10), we obtain formula (4.9).
3. Prove the following theorem:
There exists no function f(f) defined for numbers of the 2-nd kind € < Q and such that

(4.11) T(@E) OF for hE 2

and

(4.12) i Sa) Se QE) OP Gr SE = 2 ;

Proof. For an ordinal number a < @ let us denote by H, the set of all ordinal
numbers <a.
Suppose that there exists a function f(f), defined for numbers of the 2-nd kind
€ < Q2 and satisfying conditions (4.11) and (4.12). For the ordinal number a < Q let
us denote by g(a) the least number of the 2-nd kind greater than any number of the
set pa) where tf (.) denotes the set of all ordinal numbers of the 2-nd kind € < Q
for which {(¢) « H,. We shall have g(a) < Q since, in view of a < Q, the set H, is at
most denumerable, as well as the set i :) (which follows from (4.12)), and thus there
exists an ordinal number of the 2-nd kind <Q, greater than any number of the set
fei i.): We shall prove that

(4.13) <0) (CO) eh On

Indeed, if we had g(a) < a for a certain ordinal number a < Q, then (on account
of the fact that g(a) isa number of the 2-nd kind) we should have, by (4.11), f(g(a)) <
< g(a), therefore f(g(a)) <a, which gives f(y(a))«H,, whence g(a) «f '(H,), which
is contrary to the definition of the number g(a). Thus formula (4.13) holds. Thus the
infinite sequence (of ordinal numbers <Q)

(4.14) w, g(@), g?(m), g(@),

is increasing and the number

(4.15) A =limg"(@)
N<w

(i. e. the smallest ordinal number greater than any of the terms of the sequence (4.14)
is a number (4.15) of the 2-nd kind <Q. By (4.11) we have f(A) < 4, therefore, by (4.15),
for a certain natural number m we have f(A) < g”(w), which gives f(A) « H, aw)? and since
A=f (f(A), we have 4 « f'(Ayney)> which, in view of the definition of the function g,
gives g”**(w) = g(g"(w)) > 4, in contradiction to (4.15). Thus the assumption that there
exists a function f having properties (4.11) and (4.12) results in a contradiction,
q. e. d
From the above theorem we easily deduce the following paradoxical theorem of
J. Novak:
“We take away one element s, from the given infinite countable set 4,, we add
a new infinite countable set A, to the remainder, from the set
is
» 41,- Boies
A<2
384 XV. Number classes and alephs

we take away one element s,, add the new infinite countable set A, and we continue
in this way so that from the set
eu

d, A,— ‘8Wika
1

A<a

(unless it is empty) we take away one element s, and then we add a new infinite coun-
table set A,. Then there exists an ordinal number # <2 such that the set of all
given and added elements is the same as the set of elements taken away, 7. €.
.) 2°
ye A, = Shes
: A<B
(Nioweikes lal ep emaed):
For let us suppose that we can choose the elements s, in such a way that
mar 7 =
» A,#~{s},-5 for @<Q.
A<B

Since 4,(A < 2) are arbitrary denumerable disjoint sets, we can put A, =
Be
= {0., Garr> Gara> } Or evienne:
even simply A, a= {wA,@A+1,
l wA+2,‘ ...}.) Let f(wa) 2505
= *a for
a <Q. In view of
Suq \!
f€ yi A, De Sire
a
8)54
(oar
A<a

we shall have f(ma) < wa and f(wma) ~ f(md) for 2 < a < Q, which is contrary to the
theorem of Exercise 3. Thus we have proved Novak’s theorem.
4. Prove that an ordered set of type Q will not change its order type if we remove
from it an arbitrary finite or denumerable aggregate of elements.
5. Prove that there is no order type 40 for which the types m and 2 are right-
-hand divisors.
Proof. Suppose that such is the type gy € 0, i. e. that gp = aw = BQ. In view of
go #0, we havea 40 and § #0. Let A,B,C and D be ordered sets of types a, 8, w
and © respectively, e.g. C= {1, 2,3, ...}, D= {S}eeg-
Let us order the Cartesian products 4 xC and Bx D according to the principle
of last differences. In view of aw = BQ, the ordered sets 4 xO and BxD obtained in
this way, will be similar and let f denote a similar mapping of the former upon the
latter. Let a be any element of the set A: we shall have (a, n).«AxCO for n= 1, 2,...
In the set A xO there is no element that would succeed each of the elements (a, »)
where n = 1, 2,..., since each element of the set A xC is of the form (a, k) where «« A
and k = 1, 2,..., and in the set A xC we have (a, k) < (a, k+ 1). In view of the simi-
larity of the sets A xU and Bx D, also the set BxD has no element succeeding each
of the elements f((a,)) where n = 1, 2,... But, in view of f((a,”))«BxD for n=
= 1,2,.., and D= {&),-,. for every natural number n there exist one (and only one)
element 6, of the set B and a (unique) ordinal number &, < Q, such that f((a,)) =
='(0,,'6,). In view of €, <= O'forn = 1, 2; ..; numberase: (2 = 1, 25%..) are or the: esp
or of the 2-nd class, and since we have proved in XV.1 (with the aid of the axiom of
choice) that for every denumerable set of numbers of the 2-nd class there exists a num-
ber of the 2-nd class greater than any number of that set, there exists a number uw < Q
such that w > &, for n = 1, 2, ... Thus in the set B x D we shall have f((a, n)) = (b,é,) <
< (b,) and (b, u) e Bx D, which contradicts the fact that in the set BD there is no
element succeeding each of the elements f((a, )) = (b,, &,) for n = 1, 2, ... The assump-
tion that there exists an order type 40 whose right-hand divisors are types w and Q
leads to a contradiction, q. e. d.
 § 4. Properties of ordinal numbers of the 2-nd class 38: wt

It will be observed that our proof is based on the axiom of choice.

Using Exercise 5 it is easy to give an example of two different order types each
of which is a left-hand divisor of the other. Such are for instance the types a = Q°*(=
= ,..QQQ) and 6 = aw. Each of them is a left-hand divisor of the other since, as will
be proved in XV.8, number 2 is a power of number w, of course with a transfinite ex-
ponent, whence Q=w” and 1+ =; therefore #2 = w: 0’ = '*’ = wo” =Q and
thus BQ = (aw)2 = a(@2) = aQ =a, whence a= (2. Types a and f are different
from each other since the first has a right-hand divisor 2 while the second has a right-
hand divisor w; according to Exercise 5, w and 2 cannot be right-hand divisors of the
same order type.
6. Prove that if we were able to define a function associating with every ordinal
number a of the 2-nd class and of the 2-nd kind an increasing infinite sequence of or-
dinal numbers whose limit is a, then we should be able to define a function associating
with every ordinal number a of the 2-nd class an infinite sequence formed of all ordinal
numbers smaller than a.
Proof (Finsler [1], p. 78). For an ordinal number a of the 2-nd class and of the
2-nd kind let us denote by a,, a, a3, ... an infinite sequence associated with it, such that

Ges Ihnen, «
N<@

Let f(@) = (0, 1, 2,...). Now let a be an ordinal number of the 2-nd class >
and suppose that we have already defined a function f(&) fora <& <a. Ifa=f+1,
then let us denote by f(a) an infinite sequence whose first term is 6, and the (n+ 1)-th
term is the n-th term of the sequence f(f) (for n = 1, 2,...). If a is a number of the
2-nd kind
a = lima nm?
Now

then let us arrange the double sequence f(qa,), f(q),... into an ordinary sequence by
the diagonal method and remove from the sequence thus obtained all terms equal
to any of the preceding terms. Let us denote the remaining sequence by f(a). The func-
tion f(a) is thus defined by transfinite induction for all numbers a of the 2-nd class
and obviously satisfies the required condition.
7. Prove the following theorem of A. Denjoy [2], p. 1394: If f(&) ts a@ function as-
sociating with every number & of the 2-nd class a certain positive ordinal number f(&) < &
(not necessarily of the 2-nd class), then there exists an ordinal number a < 2 for which
there are non-denumerably many ordinal numbers —< 2 such that f(§) =a.
Proof. Let f(€) denote a function, defined for # < € < 2 whose values are ordinal
numbers, such that f(€) < € for om< € < 2 and suppose that Denjoys’ theorem is not
true, 7. e. that for every ordinal number a < 2 the set

BE,= KUO=a)
é=Q

is at most denumerable (or empty). For every given ordinal number uw < 2 let

it will be an at most denumerable set of ordinal numbers <2 and thus there exists
the least ordinal number g(mu) < 2, greater than any number of the set H,. Moreover,
Cardinal and ordinal numbers 25
386 XV. Number classes and alephs

we have ae cH, for up<<» < Q, therefore p(u) < p(v) for uw< » < Q. In view of we Hy 5
we have u < p(u) < 2 for wu<2. The infinite sequence

1, p(1), pp(1), ppp(1), --

will thus be an increasing sequence of type » of ordinal numbers <Q; let

A = lim”(1) ;
n<u
it will be an ordinal number of the 2-nd class. In view of the properties of function f
we shall have a = f(A) < A; thus there exists a natural number n such that gy” *(1) =
<a< q’(1). In view of f(A) = a we have Ae HE, and since H,c H,, A« H,, whence in
view of the definition of function y we obtain g(a) > 2. But in view of a < y"(1) and
v(t) < y(v) for uw < » < 2 we obtain p(a) <y"" (1) < 4, therefore g(a) < /, in contra-
diction to the inequality p(a) > A obtained above. Thus Denjoy’s theorem must be true.
Having proved this theorem (by a different method from ours), A. Denjoy [2],
(p. 1396) writes: “Thus we see how illusory, intangible, paradoxical is the nature of
the 2-nd class of ordinal numbers’.
8. Find all the ordinal numbers € for which &? = Q°.
Answer. These’are all the ordinal numbers é such that Q< € < Q®. For if 9 =
= 2°, then > (because for § < 2 we had &° < 2 < ®) and also & < 2® since
(G2E\P ss Ge, :
On the other hand, if Q< é < ®, then, in view of

Q® = lim",
n<mw

there exists a natural number » such that € < Q”, whence Q° < &® < Q” = Q” (since
nw =), therefore €° = Q®. Since there exist uncountably many numbers & for which
2<=& <Q” (such are, for instance, all the numbers 2+ & where 0 < & < Q), the equa-
tion €® = 2° has uncountably many solutions for the ordinal numbers &.
9. Prove that for each positive ordinal number a <2 we have ya=y7 and
(1+A)a = 1-+A (where 7 is the order type of all rational numbers and A the order
type of all real numbers ordered according to their magnitude).
10. Prove that the order type 72 is 47 but to each element a of any ordered set
type of type 72 corresponds a segment of the type 7.
11. Prove that the order type (1+4)2 is #41-+A, but to each element a of any
ordered set of type (1+4)Q2 corresponds a segment of the type 1+.
12. Prove that for € = (14+A)@ any ordered set A of the type + é* has a gap,
but the axiom of Ascoli is true in the set A(cf. XI.5).

5. Transfinite induction for numbers of the i-st class and of the 2-nd class.
Theorem 1 of XV.4 imples the following
THEOREM 1. Jf a given set A of ordinal numbers
1) contains number 0,
2) contains number a+1 if it contains number a
3) contains the limit of every increasing infinite sequence (of type wo)
whose terms are elements of the set A,
then the set A contains every ordinal number of the 1-st or of the 2-nd class.
 fa
§ 5. Transfinite induction 387

In other words, the set of all ordinal numbers of the 1-st or of the
2-nd class is the least set of ordinal numbers that satisfies conditions 1),
2) and 3).
Proof. Suppose that a given set jA, satisfying conditions 1), 2)
and 3), does not contain a certain ordinal number of the 1-st or the 2-nd
class; let ~ denote the least number of the 1-st or of the 2-nd class that
does not belong to A; by 1) we must have w > 0. If w were a number
of the 1-st kind, then we could write = a-+1 where a < yu: in virtue
of the definition of number uw, the number a, being smaller than uw, belongs
to A, whence, according to property 2) of the set A, we conclude that
number w= a+1 also belongs to A, in contradiction to its definition.
Thus w must be a number of the 2-nd kind. But then, by Theorem 1
of XV.4, we have
fineealinneyem
n<w

where oa, <u for n=1,2,..., and from the definition of number « it
follows that a,« A for n=1,2,..., whence, according to property 3) of
the set A, we conclude that we A, which is contrary to the definition of
number pu.
Theorem 1 is thus proved. Just as in XIIT.2 we immediately deduce
from it the principle of transfinite induction for numbers of the L-st class
and of the 2-nd class, which runs as follows: in order to prove that a the-
orem T is true for every number of the 1-st and of the 2-nd class it suffices
to prove that 1) theorem VT is true for number 0, 2) if theorem T ts true for
number a, then it must also be true for number a+1 and 3) if theorem T is
true for numbers a,, dg,..., forming an increasing infinite sequence (of
type wm), then it is also true for the limit of that sequence.
For the proof of a theorem T for every number of the 1-st class
finite induction, aS we know, is sufficient; it consists in proving that
theorem T satisfies conditions 1) and 2).

6. Convergence and limit of transfinite sequences of real numbers.

A transfinite sequence of type 2, {ae}s<q of real terms is called convergent
if for every real number «> 0 there exists an ordinal number uw < 2
such that |a;—a,|<« for €>y,y > wu. It can be shown that in that
case there exists an ordinal number v < 2, such that a; = a, for y< &<2
(Sierpinski [6], p. 132-141, Malchair [1], p. 47-50, and [2], p. 133-114).
Tf a is an ordinal number of the 2-nd kind <2 and {ag}<<, is a trans-
finite sequence of type a of real terms, then we write

a = lima:
£<a
t x
to
388 XV. Number classes and alephs :

if for every real number ¢ > 0 there exists an ordinal number w= ule) < a
such that |a;—a! < «for « <&<a. It can easily be proved that in order
that we have
lima: =a
E<a

where a < Q it is necessary and sufficient that for every infinite sequence
of ordinal numbers &, (n= 1, 2,...) such that

line eo
n=oo

we have
Hin Sn
G2. a
n=oco

In order that the limit

a= lima;
E<Q

exist it is obviously necessary and sufficient that the fransfinite sequence

{dg}e<q be convergent.

Iterations of functions of transfinite orders

For a given function f(z) of a real variable we define its iterations of

finite orders by induction in the following way. We take f,(a#)= f(a) ~
Bd TOP Miteaall 62 eee ( ahs t(fn—a(a)). Function /,(v) is termed the n-th
iteration of function f(a).
Iterations of a function f(#) of transfinite order a <2 are defined
by transfinite induction in the following way.
If a= 6+1 and if the iteration f(x) is already defined, then we
write f, = t(fe(«)).
If a is an ordinal number of the 2-nd kind <2 and if all functions
f(av) are defined for <a, then we say that there exists an iteration of
order a of the function f(a) if for every real number x exists the limit

lim f<(7)
E<a

in the sense defined in the first part of this section. That limit is then
the iteration f,(a).
It can be proved that if for a given function f(x) of a real variable
the iteration fo(x) exists, then we have f(fo(«)) = fo(x), therefore also f,(~) =
= fo(xv) for every ordinal number a > 2 (Sierpinski [68]). Thus there is
no need to consider iterations of orders higher than Q.
With the aid of the axiom of choice it can be proved that there exists
a function of a real variable f(x) for which all iterations f,(7) where a < Q
exist and are different from one another.
 § 7. Initial numbers, alephs and their notation 389

7. Initial numbers, alephs and their notation. In XV.1 we have divid-

ed ordinal numbers >a into classes, assigning two ordinal numbers
to the same class if and only if they are of the same power. In every class
(constituting a certain set of ordinal numbers) there exists a smallest
number; we call it the initial number of the class in question. Thus for
instance w is the initial number of the 2-nd class, 2 denotes the initial
number of the 3-rd class (composed of ordinal numbers of power s;).
Let Z denote a given class of ordinal numbers >, y — its initial
number, P — the set of all initia] (transfinite) numbers <y. The set P
(as a certain set of ordinal numbers) is well-ordered according to the
magnitude of the numbers composing it; thus its type will be a certain
ordinal number a. Let us agree to denote the initial number wy by a,.
Thus every initial number will have a certain index a, which is an ordinal
number, associated with it. According to our notation m= m, (since
the corresponding set P is empty, and 2 = w,). We have of course @,= Xo,
@®, =8,; let us write in general w, =s,: it will be the power of every
ordinal number belonging to the class whose initial number is w,. This
class will be denoted, after Hausdorff, by Z(x,): it is formed by ordinal
numbers of power s,. Thus Z(s,) denotes the 2-nd class of ordinal numbers,
Numbers xs, are called alephs. The power of every well-ordered set
is thus an aleph and conversely: every aleph is the power of a certain
well-ordered set (for the moment we do not raise the question whether
to every ordinal number a corresponds a certain aleph s,; this will
be proved below, Theorem 1).
Every cardinal number smaller than a given aleph is either a finite
number or an aléph, which immediately follows from the observation
that a subset of a well ordered set is a well-ordered set.
It is easy to show that if a < # and the alephs xs, and x, exist, then
Xa < Ng. Indeed, it follows from the definition of numbers w; that if a < f
then om, < mg (since a is the type of the set P, of all initial numbers < a,
and similarly 6 is the type of the set P, of all initial numbers < ,; if
®_ > wg, we Should of course have # <a, in contradiction to the assump-
tion that a<). Thus we have @, <@,; but we cannot have o,=o
since numbers w, and, wg are, in view of a < f#, different from one another,
and thus, being initial, they belong to different classes. Thus a, < @,,
therefore s,<,, q. e. d.
Hence it follows that a trichotomy holds for alephs.

THEOREM 1. For every ordinal number «a there exists a cardinal num-

ber ss,.

Proof. Suppose that there exist ordinal numbers £ for which the
cardinal number x; does not exist and let a denote the smallest of them.
390 XV. Number classes and alephs

Thus there exist all numbers x; for & < a, whence also all sets Ug = Z(x¢)
for <a. Let us denote by S the set composed of all numbers of the
1-st class and of the set

ues
E<a

Let us arrange the set S according to the magnitude of the ordinal

numbers which compose it and let S = o; thus it will be a certain ordinal
number and as such it will belong to a certain class Z(xq,).
Suppose now that y > » and y « S; thus number y belongs to a certain
component of the sum

and thus to a certain class Z% (Na,) where a, < a. Now if # is an ordinal

number <y then, if / is a number of the 1-st class, 6 belongs to S in view
of the definition of the set 8; if 6 > @ then it belongs to a certain class
Z(Sa,) and, in view of 6 < y, we have B</~ i.e. Nag Na, Whence, as
we have proved above, it follows that ag <a,, and in view of a, < a we
have also 6, < a, which proves that Z (Sag) is one of the components of
the sum S. In view of Be Z (Sa,) we thus have Be S. We have proved
that the set S is such that if a certain ordinal number y belongs to it,
then every ordinal number / < y also belongs to it.
Let uw denote the least ordinal number that does not belong to the
set S. If »vé 8, then we cannot have » < uw in view of the definition of
number uw. On the other hand, if ve 8, then we must have » < uw because
in the case of » > uw and in view of the property proved above, of the
set S, number uw would belong to the set S in contradiction to the defini-
tion of that number. The set S is thus the set of all ordinal numbers < yz,
then it is of type uw. Hence wp =a.
Now if € < a, then w,; « Z(sz) C S; therefore ; < o whence xs; < o = x,,.
On the other hand, we cannot have x; = &,, because in that case o = sz,
whence o« Z(s<) CS, which is impossible since number o= wu does not
belong to the set S. Thus s;<s,, for <a, whence <a, for <a,
which proves that a <a,. But if the cardinal number x,, exists, then there
exists also the initial ordinal number @,,. In view of a <a, the initial
number w, and thus also the cardinal number s, must exist, in contradic-
tion to the assumption regarding the ordinal number a. Theorem 1 is
thus proved.
If, for a given a, €« Z(s,.), then we have of course w, < & < wi1.
But also conversely, because if wm, <§& < mai and if € belongs to the class
Z(ss), then wg <& < wgii, Whence wy < wg+1 and wg < @ai1, Which gives
 § 7. Initial numbers, alephs and their notation 39L

a<fP~+1 and 6<a+1, whence f=a. Thus the class Z(s,) is the set
of all ordinal numbers E satisfying the inequalities wy <& < Wg41.
Let us now assume that a is a given ordinal number and m a trans-
finite cardinal number such that m<s,. Number m is thus the power
of a certain subset M of the set 7 of all ordinal numbers & < a, (which
is of type a, 7%. €. of power @ =,). But each subset of the set 7 (which
is well-ordered) is similar either to the set 7 itself or to a segment of
that set. Still the set M cannot be similar to the set Z because M =m
while 7’ is of power s, > m. Thus the subset / must be similar to a segment
of the set 7, e. g. to the one determined by number é of that set. Hence
iM = E, since the segment determined by number é in the set 7 is the
set of all ordinal numbers <é, therefore a set of type &. In view of é« 7
and the definition of the set 7 we have & < a; thus if we denote by Z(s;
the class to which number & belongs, then we shall have ¢€ < a. We have
proved that for a certain ¢< a we shall have m= M = b= Ne.
Thus we have proved that if for a given ordinal number a a cardinal
transfinite number mM is <s,, then there exists an ordinal number 6 < a
such that m = s;. Hence it immediately follows that there is no cardinal
number greater than each of the numbers s;, where ¢ < a, and at the same
time smaller than s,. This proves that for every ordinal number a number sq
is the next cardinal number after all numbers s- where 6 < a.
In particular if a is a number of the 1-st kind, a= £+1, then the
inequality €< a= 6+1 is equivalent to the inequality ¢ <<, and from
the theorem proved above it follows that the inequality m < s,+1 implies
the inequality m<s,. This proves that for every ordinal number f, the
cardinal number Sg41 18 neat after 83, 1. €. between Xz and Sp11 there is no
intermediate cardinal number. (We do not forejudge whether there exists,
besides 8,41, another cardinal number n, being not an aleph and such
that between s, and 1 there is no intermediate cardinal number. This
question will be resolved when we prove in Chapter XVI with the aid of
the axiom of choice that every transfinite cardinal numbers is an aleph).
In particular there is no intermediate cardinal number between s,
and s,, which has been directly proved (Theorem 1 of XV.2). The
cardinal number immediately following s, is s,, then comes s,, etc. The
cardinal number immediately following all numbers of the sequence s,
(7— es) IS Spy UNEN “COMES Soir, CtC.
In XV.2 (in the proof of Theorem 2) we have associated to every
real number w @ certain denumerable order type g(x). Now let us denote
by Q the set of all real numbers w such that y(#) is an ordinal number
satisfying the condition

(7.1) 2 = Rox «
392 XV. Number classes and alephs

If the set Q is not empty, let us write M = Q, and if Q = 0, let M de-

note the set of all real numbers.
The set V/ is a set with regard to which we are able to prove without the
aid of the axiom of choice that it is not empty, but in which we are not
able to point out any element !). However, we should be able to prove,
using the continuum hypothesis, that the number 1 is an element of
the set M, because that hypothesis obviously implies Q = 0. (The con-
tinuum hypothesis could be replaced here by the weaker hypothesis
that 2°° is an aleph with a finite index). But if we replace, in the defini-
tion of the set Q, the condition (7.1) by the condition 27° = s,¢), then
even by accepting the continuum hypothesis we should not be able to
point out any element of the set M (we could do it by assuming Cantor’s
hypothesis regarding alephs, implying 2?°° = x,).

8. Formula s; = s, and conclusions from it. We shall prove

THEOREM 1. For every ordinal number a the following formula holds:

se =a Te
(8.1)

Proof. Let us assume that for a certain number «a formula (8.1)

does not hold. We can also assume that a is the smallest of such numbers.
Thus we shall have a> 0 because for a = 0 formula (8.1) is known to
be true, and
(8.2) Race Nant ORI) ean,
Let us denote by P the set of all ordered pairs (u,v) where uw and y
are ordinal numbers <a,. Let us assign to the set P, (for 2<@,) all
those systems (u,v) for which 4+v= 4d. We shall prove that

(8.3) P=) Py.

Ag

If uwtv=A<a,; then we shall have ~<A and » <A, therefore

i < @, and »y < w, and thus (uw, v) « P. Thus in order to prove formula (8.3)
it suffices to show that if w<, and »<a@,, then wt»v<a,. Let us

1) See Sierpinski [17]. p. 191. The first example of such a set has been constructed
with the aid of the theory of analytic sets by N. Lusin [1] (p. 92) and also in his
book [3].
I have been asked why a non-empty set in which we are unable to point out any
element cannot be defined more simply as a set of numbers of the interval (0, 1) if the
continuum hypothesis is true. and as a set of numbers of the interval (2,3) if the conti-
nuum hypothesis is false. Now, in such a set M we are able to point out an element,
for instance the number 0 if the continuum hypothesis is true and the number 2 if it
is false.
 § 8. Formula 53 = &, 393

assume that wu < w, and v < w,. In view of u < a, we have uw < sq (since w,
is the least ordinal number of power s,), whence, as we know from § 7,
it follows that uw <s, or w= s~ where & < a. Similarly, in view of » < a,,
we shall have » < s, or » =x, where 7 < a. Let us denote ¢ = max(é, 7);
we shall have €< a ae bw <s: and » <s;. Hence, by (8.2), we obtain
uty <setee = 2np <8? =, which proves that for 4= w+» we have
A <s; where 6 <a, therefore 7<s, and A<a,, q.e.d. Thus we have
proved that formula (8.3) holds.
Now let us consider, for a given 2< a@,, the set P,, 4. e. the set of
all systems (u,v) where w+yv= A. For every given uw <A there exists,
as we know, one and only one ordinal number v such that w+yv= A.
The systems (u,v) forming the set P, can be arranged according to the
increasing magnitudes of numbers wu; the set obtained in this way will
of course be well ordered of type 4+ 1 (since as uw we can take only ordinal
numbers satisfying the inequalities 0 < wu <A).
Now let us arrange the set P, regarding that one of the two systems
(u,v) and (w’, »’) as preceding which belongs to a preceding component
P, (vt. e. gives a smaller sum uw+y); and if both systems belong to the
same component P, we shall retain for them the same order relation
as in the set P,, ordered as above. Clearly, the set P will be well ordered
in this way. We shall prove that its type will be P = o,.
Let us assume that P > w,. Thus there exists a proper segment of
the set P which is of type w,; suppose it is the segment A determined
by the element (u,, »,) of the set P. In view of (u,, »,) « P, we have uw, + 7, =
=A, <,. Further, it can easily be seen that for every element (p, v)
of the segment A, since it precedes (,,%,), we Shall have uw <A, and
y <4,. Since the set of all numbers uw <4, is of type 4’ = 1,+1, where
2’ <q in view of 2, < a, and thus of power 2’, therefore segment A is
of power A < Aji’.
But in view of 2’ <a,, we have 4’ < Sa, which, as we know from
XV.7, gives 2’ = x; where é < a; thus we have TE sz where é < a, there-
fore? by (8.2), a <<, and since xz < 8, for € < a, we have ee Sa, which
contradicts the assumption that A is a segment of type a,.
We have proved that P >, is impossible. On the other hand? we
have P > @, because the systems (uw, 0), where « <@,, form a subset
of type ,, of the set P. Thus we have P = w,, whence P = gq.
But from the definition of the set P and of the definition of the
product of two cardinal numbers it immediately follows that Beas
In view of the formula P = Na obtained just now we should have so Ra»
in contradiction to the definition of number a. Therefore formula (8.1)
must be true for every ordinal number a. Theorem 1 is thus proved.
394 XV. Number classes and alephs

Remark. In our proof we have decomposed the set of pairs (wu, v)

into the sets P,. Although they are in a certain way similar to the defini-
tion of diagonals (frequently applied in the proof of formula 80 = No))
are not really diagonals. They do contain at most one element from each
line of the double table (containing s, lines and s, columns numbered
by means of ordinal numbers <«,), but they may contain more than
one element from the same column of that table. H. g. the set P,, is formed
of infinitely many pairs of the w-th column:

(0, @), (1, @), (2, @), ...

and of one pair, (@, 0), belonging to the w-th line and to the 0-th column.
The set P,41 18 formed of the pairs

(0, o+1), (1, -o@+ 1), (2; m+. 1), ..., (, 1),(o -), 0),

the last of which belongs to the 0-th column, the last but one to the 1-st
column and all the remaining ones to the column of order m+ 1.
The set Py. is formed of the pairs

(0, we 2), (L.@>2)5, (2,000 2) sees Diy @) (ol GD), (0 2 i}) 205 gel = 2 Ne

belonging to three columns only: of the order 0, » and @- 2.

The set P,2 is formed of the pairs

COL?) GL Beo*) 2 ew") ae, (Os oe (TT eo?) ey (ee, Oa es CAO

belonging to two columns only: the 0-th and the o?-th.

We shall also give here another proof of Theorem 1, based on A. Tar-
ski’s theorem proved in XIV.21. It suffices of course to show that every
initial number is a power of number w, which Tarski proves in the follow-
ing way.
Let a, = oa, +o7%a.+...+.0%a, denote the normal expansion of
number w,. We have of course a, > 0 and we can write a,=1+/,. We
also~ have ow < o, <w(a,+1),- whence o” <@, <.o%(a,+1). But
(aq +1)o=o, whence (a,+1)0" = (a,+1)0:
0" = w:o" = ow, there-
fore (a4,+1)0% = wo”. From the definition of the product of order types
it follows that wy = yu for arbitrary types w and », whence, in particular,

@ (Oy ly (Oe Vl ae
Thus @7 <@, <a", whence @,—a@".ay Numbers o, and w™ thus
belong to the same class, whence, on account of the fact that wm, is an
initial number, we obtain w, < w%, and since wm! <@,, we have w,= w” ’
and thus @, is a power of number @, q. e. d.
 § 8. Formula s5:=Nz, 395

It could even be shown that o, = ow, i. e. that every initial number

is an epsilon-number. We shall prove it for number 2. As we know, 2 =
= w, = w’. If we had y <Q, then, as we know from XV.1, number w”
would be a number of the 2-nd class (as a power of a number of the 2-nd
class with an exponent which is a number of the 1-st or of the 2-nd class),
which, in view of 2 = w’ is impossible. Thus y > Q, and since Q= w’ > y,
We, havens, == Olt ica) oq. end:
As an easy corollary to Theorem 1 we obtain the formula

8.4 Nog| = 8 B Ol perigee

: :
since for a < 6 we have x, <sg, whence NaX8s <3; 2 = 8s and on the other
hand, of course SuNg > Nz-
Another corollary to Theorem 1 is given by the formula
iB) Sots—=,
(ee) for a<f

because for a < 6 we have sq <4, whence sa+ 8s < 8g-+8s = 28 < Ses =
= ,, and on the other hand we have of course s,+, > &,. It is equally
easy to prove that for every ordinal number a we have n+sx,=s, for
ell seen hs
From Theorem 1 we easily deduce by induction the formula

(8.6) SC ae Ne
for every ordinal number a and every natural number 7.
However, we have
Shorea, <
No = No,

since, as we know from VIII.5, 5° = 2°°. Still, there exist cardinal num-
bers m for which m”°— m; such is for instance the number m= 2”.
As regards number s,, we obtain, in view of the inequality 2 <s, < 2°°
(proved with the aid of the axiom of choice in XV.2, Theorem 2), 2°° <
<s7° < (2%)*° — 2 whence
(8.7) pie 0%
(we are unable to prove this formula without the aid of the axiom of
choice); it follows hence that the question whether x=, or si° >, is
equivalent to the question whether the continuum hypothesis 2 = s, is true
or false.
It will also be observed that we have a formula which is, in some
sense, reciprocal with respect to (8.7), namely

(8.8) gh en ON
396 XV. Number classes and alephs

It is the particular case of the more general formula

(8.9) ep a2 efor Seep
Indeed, for a < 8 we have 2 <x, < 2% < 28, whence by Theorem 1,
ONE <i << (Q"B)'F = oF — 28, which gives formula (8.9).
Let us also observe that the inequality
(8.10) Soret
can be proved to be equivalent to the continuum hypothesis. For, on
one hand, we have of course s3° > 8,, which, in case of the validity of the
continuum hypothesis, gives, by (8.7), inequality (8.10); on the other
hand, from inequality (8.10), by (8.7) and the observation that 2°¢=
= (2°°), follows the inequality s2° > (2). In Chapter XVI we shall
prove that for cardinal numbers we have the trichotomy; thus if s, > 2°°
did not hold, we should have s, < 2“°, whence 2° < (2°°)°, which is con-
trary to the inequality obtained just now. Therefore we must have s, > 2°;
since 2°° >, and since between the cardinal numbers s, and xs, there is
no intermediate one (see XV.7), it follows 2° = x,.
The continuum hypothesis can also be expressed with the aid of the
cardinal number s, in the following way.
We shall say that a cardinal number m has property P if for every
set MW of power m there exists an infinite sequence M,, M,, ... of subsets
of the set M such that for each two different elements p and q of the
set M there exist two disjoint sets, which are terms of that sequence, one
of which contains p and the other g. Now, it can be proved that the conti-
nuum hypothesis is equivalent to the theorem stating that the cardinal nwm-
ber s. does not have property P.
The proof is based on the lemma stating that in order that a cardinal
number m has property P it is necessary and sufficient that m < 2°
(see Sierpinski [22]).
It will also be observed that we are able to define effectively a de-
composition of the set F# of all functions of two real variables f(a, y)
assuming only the values 0 and 1 into x, disjoint non-empty sets (Sier-
pinski [28]).
Indeed, let a denote an ordinal number of the 3-rd class, 7. e. such
that Q <a < qo, and a=-s,. We shall assign a function feF to a set ®,
if and only if by taking Ry for é< Qand 7 < Q if and only if f(#, y)=1
for ve Xoiz, Ye Xoi, (where X, are the components of the sum (2.4),
defined in XV.2), we obtain the relation R, well-ordering the set of all
ordinal numbers <2 according to type a.
Taking
’,=F-— » ®,,
Qea<wg
 § 8. Formula x7 = x,y 397

we shall obtain an effective decomposition of the set F

ie Doce 2.
QK<aK<wg
into s, disjoint non-empty components, which can be proved without
resorting to the axiom of choice.
. Indeed, let a denote an ordinal number such that 2 <a < @,, thus
a —s,. In view of Q =x, we have a = Q, and since a is the order type
of the set #, of all ordinal numbers <a, and 2 the type of the set Ho of
all ordinal numbers <Q, there exists a (1-1) correspondence f between
the numbers of the sets Hg and H#,. Let R denote the relation defined in
the set Hg by the conditions that Ry for <2, 4 <2 if and only if,
T(€) < f(m). By the relation R the set He is obviously well-ordered accord-
ing to type a.
Now we shall define a Woouion f(x,y) of two real variables in the
following way. If # and y are given ae numbers, then, by (2.5), there
exist ordinal numbers § < 2 and 7 <2, well-defined by x and y, such
that we X,+<, ye Xo+,- If, moreover, we have Ry, then let f(x, y) = 1;
otherwise let f(#, y) = 0. From the definition of the set ®, it easily fol-
lows that fe @®,. Thus the sets ® are non-empty for Q <a<@y,, q.e. d.
It is easy to prove without the aid of the axiom of choice that the
set F is of power 22°. Thus we are able to prove without resorting to the
axiom of choice that a set of power 27° is the sum of s, disjoint non-
empty sets, 7. ¢. that we have the formula s, < *2?°. It easily follows
that 2°° < 22°, whence x, < 2”°. Let us also observe that it is pos-
sible to prove without the aid of the axiom of choice that s, < 2%! and,
after A. Tarski, that 2°¢+? < 27° and x,41 < *2"¢ for every ordinal number a.

9. A proposition of elementary geometry, equivalent to the continuum

hypothesis. We shall prove
THEOREM 1. The continuum hypothesis H is equivalent to the proposi-
tion P that the three-dimensional Euclidean space HE is the sum of three
sets EH; (0 = 1, 2,3) such that if we denote the axes of Cartesian coordinates
am the space HE! by OX; (1 = 1,2, 3), then, for 1= 1,2,3, the set E; is finite
on every parallel to the axes OX; aa [61], p. 1046; Sierpinski [64],
p. 1-6; Sierpiski [59], p. 1-4).
Proof. 1. HP. Let us assume that hypothesis H is true. Thus
there exist sets A; (¢=1, 2,3) satisfying Theorem la of XV.3. For
a real a and for 71 = 1, 2, 3, let us denote by P;(a) the plane x; = a. Thus,
for 7 = 1,2, 3,' we: have
K,= >) K,P\(a)
398 XV. Number classes and alephs

where the summation is extended to all real numbers a. According to the

properties of the sets A; expressed in XV.3, Theorem la, the sets I; P;(a)
are at most denumerable for i—1,2,3 and for real a. In virtue of
Exercise 1 of III.3, the set A,P,(a) is the sum of two sets A,P,(a) =
= A,(a)+A,(a), where A,(a) is a set finite on every parallel to the axis OY,
and A,(a) is a set finite on every parallel to the axis OZ. Similarly,
K,P.(a) = B,(a)+B,(a) where B,(a) is a set finite on every parallel to
the axis OX and B,(a) is a set finite on every parallel to the axis OZ.
Finally, A,P;(a) = C,(a)+C,(a) where O,(a) is a set finite on every
parallel to the axis OX and (C,(a) is a set finite on every parallel to the
axis OY. Let

B, = » [B(a)+G(a)], 2B,= S[A,(a)+6,(a)},

a a

H,= » [A,(a) +B,(a)] .

a
It is easy to verify that the sets H; (¢=1,2,3) satisfy the condi-
tions of the proposition P. Thus we have proved that H—P.
2. P+H. Let us suppose that the proposition P is true and that
hypothesis H is false, 7. e. that 2°° > s,. Let N denote a set of real num-
bers of power s,. The set 7 of all points (7, y) of the plane OY such that
xe N, ye N is thus of power xi, therefore, by formula (8.1), of power s,,
and in view of s, < 2° of power less than the continuum. According to
proposition P the set KE, is finite on every parallel to the axis OX, = OZ.
Since the sum of ys, finite sets is of power <3,Np, 2. €. by (8.4) of power
<s, < 2°, the set of all points (#, y, z) of the set H, for which we N and
ye WN is of power less than the continuum. The projection of that set
on the axis.OZ, in view of Corollary 3 of VI. 3 is thus also of power less
than the continuum, and there exists a real number ¢ such that the plane
z= ec contains no point (xz, y, 2) of the set #, such that (x, y) e T. There-
fore, in view of H = H,+H#,+#,, we have (7, y, c) « H,+H, for (a, y)« 7.
Now let D denote a denumerable subset of the set N. The set H, is
finite on every parallel to the axis OX,= OY, therefore the set of all
points (x,y, ¢) of the set B,, where ve D and y e N, is at most denumer-
able. Since the set NV is non-denumerable, it follows that there
exists a number } of the set N such that the line y = b does not contain
any point (x, b,c) of the set H, such that we D. But in view of DCN
and be N for ce D we have (x,b)« T and thus (#,b,c)« H,+#,, and
since (@,b,c)¢ H,, we obtain (w,b,c)e H, for we D. The straight line
y=b, 2=c, parallel to the axis OX = OX,, would thus contain (in
view of the denumerability of the set D) infinitely many points of the
set £,, in contradiction to the properties of the set #,. Thus the assump-
 § 9. A proposition equivalent to the continuum hypothesis 399

tion that proposition P is true and hypothesis H is false leads to a contra-

diction. We have proved that P—H.
Theorem 1 is thus proved.
Theorem 1 makes it possible to express the continuum hypothesis
by means of notions of elementary geometry without the use of the con-
cept of infinity.
In connection with Theorem 1 let us observe that it has not been
solved yet whether there exist in a plane three straight lines D; (¢ = 1, 2, 3),
such that the plane is the sum of three sets E; (i= 1,2,3) such that
for i= 1,2,3 the set H; is finite on every parallel to the line D.
. In connection with Theorem 1 of XV.3 and Theorem 1 it will also
be observed that the following theorems hold:
T,. The hypothesis 2° <s, is equivalent to the assertion that the three-
dimensional Euclidean space is the sum of three sets HE; (= 1, 2,3) such
that if we denote the axes of Cartesian coordinates in the space E by OX;
(¢=1, 2,3), then, for i= 1, 2,3 the set E; is denumerable on every parallel
to the axes OX, (Sierpinski [60], p. 6, Theorem 3) !).
T,. Lhe hypothesis 2° <y,, where n is a given natural number, is
equivalent to the assertion that the (n+-2)-dimensional Euclidean space
is the sum of n-+-2 sets H; (1 = 1,2, «.,%+2) such that, forv=—1,2 ...,
n+2, the set H; is finite on every parallel to the i-th axis of Cartesian co-
ordinates in that space.
EXERCISES. 1. Prove that if m denotes a natural number, then the three-di-
mensional Euclidean space # is not the sum of three sets H, (i = 1, 2, 3) such that,
for 1 = 1, 2, 3, the set H, has at most m points on every parallel to the axis OX, (Sier-
pinski [60], p. 9, Theorem 6, and p. 10, Theorem 7).
2. Prove that the three-dimensional Euclidean space # is not the sum of three
sets H, such that, for i= 1, 2, 3, the set H, is finite on every plane perpendicular to
the axis OX, (Sierpinski [60], p. 12, Theorem 9).
3. Prove the following two theorems:
Every plane set of power §, is the sum of two sets of which one is at most denumer-
able on every parallel to the axis of abscissae and the other is at most denumerable
on every parallel to the axis of ordinates.
The hypothesis 2“° < x, is equivalent to the assertion that the three-dimensional
space # is the sum of three sets H, («= 1, 2, 3) such that for 1= 1, 2,3 the set H, is
at most of power , on every plane perpendicular to the axis OX,.
From these theorems deduce theorem T, (just as we have deduced Theorem 1
from Theorem 3a of XV.3).
10. Difference of alephs. Sums and products of transfinite sequences of
successive alephs. Let a denote a given ordinal number, nt — a given car-
‘dinal number such that 1 <»s,. Thus there exists a cardinal number p
such that s, = n+p. If tt is a finite number, then of course p cannot be

1) See also Exercise 3 below.

400 XV. Number classes and alephs

a finite number and, in view of s, > p, the cardinal number p is an aleph,

é.g. p=%,. But then, as we know, in view of number n being finite,
we have n+, = Nz, therefore s, = 8s, whence p=x,=k,. If n is not
a finite number, then in view of n <8, humber 11 is an aleph, ¢. g. n= &,,
and in view of s, <8. we must have y < a. Thus the cardinal number p
cannot be finite since in that case we should have x,=s,+p=k,, in
contradiction to s, <s.. Thus number p is an aleph, p= sz. If Bb <y,
we should have, by (8.5), Sa=%8sgt8,=8,, Which is impossible. Thus
b> -y, whence s, = 8,+N,s = %,, therefore p= s,. We have proved that
in any case p = »,. Thus for n < s, there exists only one cardinal number p
such that s, = n+p, namely p= s,. From the definition of the difference
of cardinal numbers (IX.1) it follows that if s, is a given aleph, then
for every cardinal number n<»s, exists the difference s,—n=k,.
At the same time we have proved that 7f we subtract from a set of power xa
a set of power less than s, then the remaining set will be of power s,. Moreover
we have the formula

(10.1) RosaheeaiN, LO ayoon

We can also say that the power of an infinite well-ordered set does not
change if we subtract from it a set of less power.
As an application of formula (10.1) we shall calculate the power
of the set of all ordinal numbers forming the class Z(s,). In § 7 we have
proved that the class-Z(s,) is the set of all ordinal numbers & satisfying
the inequalities wm, <& < wyii1. Since that set will be obtained by sub-
tracting from the set of all ordinal numbers <a@,+1, which is of type @g41,
i.e. Of power @y+1—= Nai1, the set of all ordinal numbers <a, which is
of power @, = &., the power of the set Z(s,) 1s, by (10.1), So41—Sa = Not1>
Hence we obtain
THEOREM 1. The class Z(s.) has the power Sq+1-
The set S of all ordinal numbers € < m, is the sum of disjoint sets

(10.2) S= 8+) 4s),E<a

where S, denotes the set of all numbers of the 1-st class (7. e. of the num-
bers €é < m). For if €< @,, then é<@,=8, and in that case either ¢ is
a finite number, and thus £« 8, or & is an aleph and thus = 3 <»,,
whence 6 <a and therefore ¢« Z(s,) where fp < a.
Since, by Theorem 1, the set Z(xz) is of power sei and S,= x),
S=s,., we obtain by (10.2),

(10.3)
eee
Rees pAb NE ai
B}

E<a
 o

§ 10. Difference of alephs 401

In particular, for a= , formula (10.3) gives

(10.4) No = Noth: +8.+...

In virtue of Theorem 1 of VI.8 (proved with the aid of the axiom

of choice), if we decompose a set of the same power as the continuum
into a denumerable aggregate of subsets, then at least one of them must
be equivalent to the continuum. It follows hence that number 2” cannot
be the sum of an infinite sequence (of type w) of increasing cardinal num-
bers. Since, by (10.4), number x, is such a sum, we have (Kénig’s theorem )
(10:5) 2% Zy,.

By Theorem 1 of X.7 we have

(10.6) Sot
8, +o t+... << NyNoNs-.-5

Pde since s-< ky form = 152, 1, S88... << Ne, 1ormula (10:6) gives,
by (10.4),
(10.7) peas.ce
It is easy to prove that

(10.8) RoNiNees ="

On one hand we have soXys2--- <s., and on the other hand, since
the product of an infinite sequence of cardinal numbers does not depend
on the order and grouping of the factors, we have

NpNoNg--. = (NiNoNs) (NokgN19 +++) (NahroNo---)«++

and since for every increasing infinite sequence of natural numbers

Ky, hy, ... WE ave RN Nkg--- S NiNeN3--- > No, Ib follows that s,N.83... >
2 ScSeR pss! Nos
ae s

Formula (10.8) is & particular case of the formula

a
| | Ne = Ra
E<a

for numbers of the 2-nd kind, proved in 1925 by Tarski [3] (p. 11), who
also proved that for numbers a of the first kind we have
4 a
[]s:=s%.
§<a

Kor sums it would be easy to deduce from formula (10.3) the follow-
ing formulae:

f=, for any ordinal numbers a,

E<a

Cardinal and ordinal numbers 26

4()2 XV. Number classes and alephs

and
. Se=, for a of the 2-nd kind.

Finally, we shall give here without proof the formula established

by Hausdorff [1] (p. 570) (the so-called recurrent formula):
de
X
ae RS
s

ata Dye aa

and its easy generalization:

88 SB a ¢
shoatn SYSa-atn EOP ieeesd tes
tee i

Putting in the latter formula a= 0, we obtain, by (8.9), F. Bern-

stein’s formula (Bernstein [1], p. 150):

(10.9)
( NP
Di) =SS 2x
ONales for
On =n
Pe Oe 5) Re:

Generalizations of these formulae and of some other formulae for

alephs have been given by Tarski [3], p. 1-14.
EXERCISES. 1. Prove F. Bernstein’s formula (10.9) for n= 1.
Proof. In virtue of Theorem 1 of XV.2 we have 2°° < 2%%y, < 9*09*o — 9%0, whence
2%oy, = 2%, and since, by (8.7), we have 5° = 2°, it follows that {9 = 2%°y,, which
proves the validity of formula (10.9) for n=1, B=0. If 8 > 0, then y, < Np < 28
and thus, by (8.5) and (8.9), 2° <8, < 2-98 — 20"
“8— 9°8, whence sy°s,=2%,
and since, by (8.9), wi? = 2° and so? = 2%, we have sf? 26 — yoy, = 2's, which
proves the validity of formula (10.9) for n= 1, 6 > 0. Thus we have proved that
s,/ = 2 Fy, for every ordinal number f > 0, q.e. d.
2. Give an example of a cardinal number m such that m*° > 2°m.
Solution (Tarski [3], p. 10). Let m = y,+ 2°°-+ 2704 ... It is easy to prove that
: R ; : x ‘ Xoe2ke N ‘
98m =m. Hence m*° = mmm... >> 2%o- 928092?" _. gNot 200+ A" +... _ om __ 9280Ont __ 9Xony
therefore m*® > 2°%m, q. e. d.

11. Limit of transfinite sequences of initial numbers. Regular and singular

initial numbers. Inaccessible alephs. Let a denote a given ordinal number
of the 2-nd kind, and {ag}.c, — a given increasing transfinite sequence (of
type gy) of ordinal numbers tending to a, 7. e. '

(Ge esp) Nin

ae Sher.
<9
We shall prove that
(11.2) Gg == Das.
sap 7
In order to prove formula (11.2) let us observe, to begin with, that
we have the inequality

Oq > Wa, fOr ay

 § 11. Regular and singular initial numbers 403

since, by (11.1), a> a; for é<q. Number @, is thus greater than any
term of the sequence {a,,}s<g- On the other hand, it is easy to prove
that @, is the least ordinal number greater than any term of the sequence
under consideration. For if we have y <a, then p<, since a, is the
least ordinal number of power s,. In view of y<s,, however, we have
Y=: (see XV.7) for a certain € <a. By (11.1), for the ordinal number
¢<a there exists an ordinal number u«<g such that a, >¢, whence
Na, > Xo = wy; therefore a, > Ys which proves that number y cannot be
greater than each term of the sequence {wa,}:<,. In view of the definition
of the limit of a transfinite sequence of ordinal numbers we conclude
that equality (11.2) holds.
Therefore the function of an ordinal variable f(a) = w, is increasing and
continuous. .
Thus we have also proved that the limit of every increasing transfinite
sequence of initial numbers is an initial number.
In particular, we have for instance

Og = nos
n<o

which proves that number w, is the limit of a denumerable sequence

(of type w) of numbers smaller than w,. But number w, = 2 is not the
limit of any sequence of type w of numbers smaller than w, because such
numbers would belong to the 1-st or to the 2-nd class, and since num-
ber 2 is not of the 2-nd class, this would contradict Theorem 1 of XV.1.
Initial numbers o, which are the limits of transfinite sequences of
type <a, are called singular; all other initial numbers (7. e. such
numbers w, that are not the limits of any transfinite sequences of type
<@,) are called regular.
Thus for instance number w, = 2 is regular, number @, is singular
(as the limit of a sequence of type wo <,). Similarly, number ,,. is
singular as the limit of a sequence of type w <@,. since, in view of

o = lino”,
n<o
we have
Oo(2) = LUMO,» -
n<=o

Number @,, = @g is also singular because we have

og=limw and Q2=a,< ag.

E<Q

We say that an ordinal number 4 is confinal with a number w of

the 2-nd kind if 4 is the limit of an increasing sequence of type wu of ordinal
numbers.\ Clearly, in order that an initial number @, be singular it is
26*
404 XV. Number classes and alephs

necessary and sufficient that it be confinal with a certain number of

the 2-nd kind smaller than a,.
It is easy to prove that if a is a number of the 1-st kind, then ,
is a regular number. But we do not know any regular number w, whose
index a is a number of the 2-nd kind, although on the other hand we
do not know whether every initial number whose index is a number of
the 2-nd kind must be singular. We know, however, that every regular
number o, with an index a.of the 2-nd kind would have to satisfy the
condition
(11.3) (Ota e

For we obviously have wo, >a for every ordinal number a since
the- set of all ordinal numbers § < m, contains a subset {m}s<, of type a;
thus if a is a number of the 2-nd kind and ow, > a, then in view of

Og = ima:
E<a

we conclude that number @, is singular.

It is remarkable that there exist initial numbers o, satisfying condi-
tion (11.3), 7. e. equal to their indices. Indeed, let

(11.4) Cy OF OC MPCp tt ee te Len eer eae

We have hence a,= ao, > w= a,, therefore a, > a,. Suppose that
for a certain natural n >1 we have a, >a,_1. We should have hence
Day > g,_,) therefore a,11 > a,. Therefore we infer by induction that
the sequence a, (n= 1, 2,...) is increasing. Let

a= litota,3
now

by (11.4) we have hence

(ide ==) 0) ML Gy,vega Le et,

n<@ n<@ n<@

which gives formula (11.3). The number @,, however, is not regular,
being the limit of an infinite sequence of type m < a,. It can easily be
proved that the number w, which we have defined above is the least
initial number satisfying equation (11.3). For if wm; = 6, then we have
Op > ®@ = a,, and from the assumption that wg > a, it follows that Wop >
> Day yt. €. Op > Ont1, Whence ws > a, for n= 1,2,;...; and thus w, >@,:
Neither can we solve the question whether every initial number,
with an index of the 2-nd kind, of power <2” is singular. The assump-
tion that if is so is of course weaker than the continuum hypothesis;
it can be proved that it follows already from the assumption that 2° < Sor,
(Sierpinski [23], p. 153). With its aid it is possible to prove some important
 § 11. Regular and singular initial numbers 405

theorems which previously could only be proved with the aid of the
continuum hypothesis (Sierpiiski [23, p. 153-161; 19, p. 214; 20, p. 1],
Ulam [1, p. 140; 2, p. 22)).
K. Gédel has proved that the assumption that there exist no regular
initial numbers with an index of the 2-nd kind is consistent with the
axioms of the theory of sets (if these are consistent); it is not known if
the same apples to the assumption that such numbers exist.
Regular alephs s, whose indices are numbers of the 2-nd kind have
been termed inaccessible by K. Kuratowski't). With the aid of Cantor’s
hypothesis regarding alephs (according to which for every ordinal num-
ber a we have 2°*= s,11) it can be proved that the above definition is
equivalent to the following definition of inaccessible transfinite cardinal
numbers:
A cardinal number m > s, is termed inaccessible if for every trans-
finite sequence {mMg}:<, of type uw, where 0 << m, formed of cardinal
numbers m; < m for é< mu) we have

[ [ me<m
E<u

(Sierpinski and Tarski [1], p. 292).

It will also be observed that with the aid of the axiom of choice we
can prove that in order that a cardinal number m be greater than any
sum of fewer than m cardinal numbers smaller than im it is necessary and
sufficient that 1 =s, and that m, be a regular initial number (Tarski
(Seep. 69);
In his extensive work (Tarski [18], p. 71, Theorem 9) on inacces-
sible numbers Tarski calls a cardinal number m inaccessible in the wider
sense if m #0 and if the following two conditions are satisfied:
1. if X is a set of power <m and if with every element x « X is as-
sociated a cardinal number n, < m, then

|
\) Tanita
i’ Z
xeX

2. ifm < m, then there exists a cardinal number p such that n < p< m.
The same author calls a cardinal number m inaccessible in the nar-
rower sense if m~0 and if m satisfies condition 1 and the following
condition:
1) Without the aid of the continuum hypothesis we are not able to prove that
there are no inaccessible numbers among the cardinal numbers <2“. In their book [1]
(p. 233) Kuratowski and Mostowski write that the addition to the set theory axioms
of the assumption that all cardinal numbers are accessible does not result in a con-
tradiction, and that the proof thereof is outlined by K. Kuratowski [3] (p. 146) and by
Firestone and Rosser [1] (p. 79).
406 XV. Number classes and alephs

3. if t<m and p<m, then n° <m-.

With the aid of Cantor’s hypothesis concerning alephs it can be
proved that for transfinite cardinal numbers both definitions are equi-
valent (Tarski [18], p. 71, Theorem 9). .
In another paper of his (Tarski [10], p. 154-155) A. Tarski introduces
the concept of a cardinal number accessible through another cardinal
number (weakly or strongly).
EXERCISES. 1. Prove that every number of the 2-nd class and of the 2-nd kind
is confinal with the number ow.
Proof. It is an immediate conclusion from Theorem 1 of XV.4.
2. Prove that if an ordinal number 4 is confinal with a number of the 2-nd kind
and the number y is confinal with a number of the 2-nd kind y, then the number A/ is
confinal with the number y.
Proof. The proof may be based on the easily provable proposition that if

A=lima, and yw=lim§,,

é<u E<y
then
A= lim ag F
E<v §
 CHAPTER XVI

-ZERMELO’S THEOREM AND OTHER THEOREMS EQUIVALENT

TO THE AXIOM OF CHOICE

1. Equivalence of the axiom of choice, of Zermelo’s theorem and of the

problem of trichotomy. We shall prove
THEOREM 1. We are able to define a function s(m) which associates
with every cardinal number m a certain aleph s(m), which is not <m
(Hartogs [1], p. 436-443; Sierpinski [7], p. 118). f
Proof. If m is a finite cardinal number, then we can of course
write s(m) =s,. We can further assume that the cardinal number m is
not finite. We denote by Z the set of all ordinal numbers a such that
a<wm. The set Z is infinite because every natural number belongs to it;
for if the cardinal number im is not finite, then we have n < ™ for n=
=1,2,... The set Z is well-ordered according. to the magnitude of the
ordinal numbers belonging to it. Let € denote its order type; it will be
a transfinite ordinal number (since the set Z is infinite). The cardinal
number é (the power of the set Z) will thus be a certain aleph, s(m).
If s(m) <_m, we should have € <m and, in virtue of the definition of the
set Z, €« Z, whence naturally ae Z for a < é, since for a< & we have
a<&<im. But the set of all ordinal numbers a < é is known to be of
type &, and on the other hand, in view of é« Z, it is a segment of the
set Z formed by a certain element of the set Z, 7. e. it is a proper segment
of the set Z. Thus the well-ordered set Z would be similar to a certain
proper segment of itself, which, as we know, is impossible. Therefore we
cannot have s(m) <™m.
Theorem 1 is thus proved (without the aid of the axiom of choice).
In the above proof of Theorem 1 we have made use of ordinal num-
bers. We can also prove this theorem without the aid of ordinal numbers,
using merely the properties of well-ordered sets. In this way even
a stronger theorem can be proved, namely |
THEOREM}2. We are able to define a function f(M) which associates
with every set M a certain well-ordered set f(M)C UUU(M), whose power
is neither equal to nor less than the power of the set M (Sierpinski [36], p. 1,
Lemma 1).
408 XVI. Zermelo’s theorem

Proof. Let M denote a given set. According to the notation which

we have adopted, U(M) denotes the set of all subsets of the set M, and
UU(M) the set of all sets whose elements are subsets of the set M. Let F
denote the set of all those sets of the subsets of M which are well-ordered
according to the relation C (between the subsets of the set M which are
their elements). We shall have / C UU(M), therefore F «eVUUU(M). (Thus,
for instance, if M+ {1,2,3,...3, then {{3}}«F, {41,2}, 1,2, djleF,
{1,3}, 1,2, 33}¢F but {1}, 2,3}}¢F). Now let us divide the
sets belonging to the set F into classes, assigning two sets to the same
class if and only if they are similarly well ordered. Let ® denote the
set of the classes obtained in this way. Our classes are of course sub-
sets of the set F, whence ‘t follows that the set ® is contained in the set
U(F), i.e. ©C U(F), and since FC UU(M), we have U(F)C UUU(M)
and thus ®C UUU(M). If K, and K, are two different classes belonging
to ®, A,« K,, A,e« K,, then each of the sets A, and A, is well ordered
but they are not similar since they belong to different classes, K, and K,.
One of these sets is similar to a proper segment of the other. Let K,< K,
if the set A, is similar to a proper segment of A,. In that case if instead
of sets A, and A, we took sets Aj and Aj such that A; « K,, Aj« K,, then
we should have A,= A{, A,=Aj and the set A{ would be similar to
a proper segment of the set As. Obviously the set @ is well-ordered by
the relation < which we have established. We shall prove that ® < W
is Impossible.
Suppose that ® <M. Thus there would exist a subset M, of the
set M such that 6~M, and a mapping of the set © on M, by which the
set M, is well-ordered according to the same type as the set ®. Let us
denote by A the set of all proper segments of the set M,; it will be a cer-
tain set of subsets of the set M, well-ordered by the relation C of the
same type as the set ®. Thus we shall have A « # and therefore the set A
will belong to a certain class K «®. Similarly, each proper segment of
the set M, will obviously determine a certain class K,«@®, such that
K,<K, and since the set of all proper segments of the well-ordered set M,,
when ordered according to their magnitude, is of the same type as the
set M,, the segment of the well-ordered set ® formed of the element K
of that set would contain a well-ordered subset of the same type |as the
set ©, which is known to be impossible. Thus we cannot have ® <M.
Now let f(J1)= ®; clearly the function f(I/) satisfies the conditions
of Theorem 2, which is thus proved.
Now let m denote a non-finite cardinal number and let M be a set
of power m. The power of the set (Jf) will obviously be an aleph, inde-
pendent of the choice of the set M of power m; let us denote it by s(m).
 \

§ 1. Equivalences 409

In view of f(M)C UUU(M) and M = m, we shall have s(n) < pee

Thus. Theorem 2 has the following
COROLLARY. For every non-finite cardinal number w there exisise an
aleph s(m) such that
nt
(1.1) s (im) <2” and s(m) is not <m.

Among the alephs that are not <m there exists of course a smallest:
and (in view of the corollary proved above) that aleph — which we shall
continue to denote by s(m) — will obviously satisfy conditions (1.1).
It is also clear that between the cardinal numbers m and s(nt) there is
no intermediate cardinal number (cf. Tarski [14], p. 26-32).
We have m<m-+s(m) since obviously m<m+s(m) and m=
= M+(M) give s(m) <m, which is impossible.
THEOREM 3. For every given ordered set M we are able to define an
ordered set N of greater power than the set M (Sierpinski [41], p. 137).
Proof. It suffices of course to prove Theorem 3 for infinite sets M.
Let M denote a given infinite ordered set and f — a function satisfying
Theorem 2. Thus we shall have f(J7)C UUU(M)=T. As we know,
there exists an ordered set M’ similar to the set WM and disjoint with
the set 7. Therefore M’-f(M)—0 and M’= M. Let N= M’+f(M)
and let us order the set N, retaining for the components M’ and f(J/)
their previous order arrangement (f(J/) is, by Theorem 2, a well-ordered
set), and regarding the elements of the component M’ as preceding the
elements of the component f(M/). Let M =m; thus f(M) = 8 (Nt), and,
by Theorem 2, s(t) <m does not hold. In view, of M’-f(M)= 0, we
shall have n= NVN=m+s(m). Thus n>m. If we had n=M, 7. e.
m-+s(m) =m, we should have s(m) <™m, which is impossible. There-
fore n ~m, and thus n> m. The ordered set N is thus of greater power
than the set M. We have proved Theorem 3 (without the aid of the axiom
of choice).
In particular, if W/ is the set of all real numbers ordered according
to their magnitude, then from Theorem 3 we immediately obtain the
following
CoroLLARY. We are able to define an ordered set N and to prove without
the aid of the axiom of choice that the set N ws of greater power than the
continuum.
Similarly, in virtue of Theorem 3, we are able to define an ordered
set NV, of greater power than the set N, then an ordered set JN, of
greater power than the set N,, etc. But we are not able to define an
410 XVI. Zermelo’s theorem

ordered set with regard to which we could prove without the aid of
Cantor’s hypothesis on alephs that it is of power > 27°. From Cantor’s
hypothesis it follows that the set N is of power 22°, and so is the well-
ordered set composed of all ordinal numpers & < ay.

THEOREM 4. Knowing a function f associating with each non-empty

subset X of a given set M a certain element of that subset, f(X)« X, we are
able to define the well-ordering of the set M.
Proof. Let M denote a given set of power m, Z — the set of all
ordinal numbers a such that a <m, ¢— the order type of the set Z
ordered according to the magnitude of the numbers belonging to it.
As we have shown in the proof of Theorem 1, ¢ <m does not hold.
Let a) = f(M). Now let o denote an ordinal number >0 and suppose
that we have already defined all elements a; where § < a as certain ele-
ments of the set M; let M, denote their set. If we had M, ~ M, then
the set M—M, would be non-empty and we could set a, = f(M—M,).
The elements a, are thus defined by transfinite induction for every ordinal
number a > 0 for which the set M, is different from the set M. We shall
prove that this cannot hold for every ordinal number a < ¢. Indeed, in
that case we could define by transfinite induction all elements a, for
€<(¢. Those elements are all different from one another because a, =
= f(M—M,)« M—M,, therefore a,¢ M,= {ag}, and thus 4,44
for <a. The set M,;= {ag}:<¢ would thus be of power ¢ and therefore
it would not be of power <™m, whence follows a contradiction since all
its elements belong to the set M of power m. We have proved that there
exists an ordinal number, obviously only one, a <¢ such that M,= M.
Then M = {ag}s<, and the set M is well ordered of type a. Theorem 4 is
thus proved.
Relation o, well-ordering the set M, has been defined in the proof
of Theorem 4 by means of function f with the aid of ordinal numbers
and of transfinite induction. However, it could be defined without the
use of transfinite ordinal numbers in the following way.
As in the proof of Theorem 4, let f denote a function associating
with every non-empty subset X of the set M a certain element of that
subset, f(X)« X. Let us denote by ® the smallest family of sets such
that: 19 Me@, 2° if X«@ then X—f(X)«@ and 3° the product of an
arbitrary aggregate of sets belonging to ® belongs to ®. It is easy to
prove that such a family ® of sets exists. Now let pe M, qe M, pq.
Let us denote by T= T(p,q) the product of all sets X of the family ®
such that pe X and ge X. Let poq if and only if p=—f(T). It could be
proved that the set M is well-ordered by the relation o thus defined and
it is the same ordering as has been given by the proof of Theorem 4.
 § 1. Equivalences 411

A proof thereof, especially if we wished to conduct it without the aid

of transfinite numbers, would not be easy 4).
From Theorem 4 immediately follows
THEOREM 5. It follows from the axiom of choice that for every set there
exists a relation o well-ordering that set.
Proof. In VI.3 we have proved that from the axiom of choice fol-
lows theorem T stating that for every set M there exists a function f
associating with each non-empty subset X of the set M a certain element
of that subset. Hence and from Theorem 4 immediately follows the
validity of Theorem 5.
The first proof of Theorem 5 was given by E. Zermelo in 1904 in
the paper [1] (p. 514-516). That proof was based, similarly to ours, on
‘the theory of ordinal numbers. A few years later Zermelo gave another
proof of his theorem, not using ordinal numbers. (Germelo [2], p. 111-128;
Hausdorf [2], p. 136-138, and p. 56-58). The idea of that proof has been
utilized by C. Kuratowski to avoid the use of ordinal numbers in the
proofs of theorems in whose wording they do not appear (Kuratowski [2]).
The sentence appearing in the title of Zermelo’s first proof — every
set can be well-ordered — must of course be understood as referring to
“possibility”? in the ideal sense, for it is not every set that we are able
to well-order. For instance we are not able to well-order the set of
_ all real numbers. There are many sets with which we have to deal in
mathematics that we are unable not only to well-order but even to order
at all, e. g. the set of all real functions of a real variable or the set of all
denumerable sets of real numbers.
The assertion that for every set there exists a relation well-ordering
that set is called the well-ordering theorem of Zermelo (Wohlordnungs-
satz). In virtue of Theorem 5, Zermelo’s theorem results from the axiom
of choice. But it is easy to prove that, conversely, the axiom of choice
results from Zermelo’s theorem.
Indeed, let Z denote a set whose elements are non-empty sets having
no common elements. Let S be the sum of all the sets A forming the set Z
of sets. In virtue of Zermelo’s theorem, there exists a relation o well-
ordering the set S. Each set A belonging to 7% is of course a non-empty
subset of the well-ordered set S; let f(A) denote the first of the elements
of set 8 belonging to A. The set B of all elements of f(A), where A « Z,
will of course contain one and only one element of each of the sets A
belonging to Z. The axiom of choice is thus true.
yy Cf. Zermelo’s so-called second, proof of his theorem, dating from 1908, which
we cite below. It will also be observed that Kuratowski [2] (p. 76-108) proves that
theorems of a certain general type can be proved without the aid of transfinite numbers.
Cf. also Milgram [1], p. 18-30, and Vaughan [1], p. 407.
412 XVI. Zermelo’s theorem

It can easily be seen that Zermelo’s theorem is equivalent to the

assertion that every cardinal number that is not finite is an aleph. Hor if
a cardinal number m is not finite and M is a set power m, then the set WZ
is infinite, and since there exists a relation well-ordering it, therefore m,
as the power of an infinite well-ordered set, is an aleph. On the other
hand, if every non-finite cardinal number is an aleph and if M is an in-
finite set, then its power is an aleph, and therefore there exists a rela-
tion well-ordering the set M, whence follows Zermelo’s theorem for in-
finite sets (it being unnecessary to prove it for finite sets).
Thus, in particular, it follows from the axiom of choice that the
cardinal number 2°° is an aleph, but we do not know which one and we
do not know any ordinal number f for which 2°° < xz.
Since for ordinal numbers a we have s, = @,, the cardinal number s,
is the power of the set of all ordinal numbers <a. From the theorem
stating that every non-finite cardinal number is an aleph it follows that
there exists a function /(m) associating with every cardinal number m
a set of power m. Namely, if m is a finite cardinal number, then we can
set f(m) = {1, 2,..., mt}, and if m is not a finite number, then there exists
only one ordinal number a such that m=s, and we can set f(m) =
= {Se<o,- Thus, in particular, it follows from the axiom of choice that
every infinite sequence of cardinal numbers satisfies condition W of X.1.
By the theorem on trichotomy we understand the proposition that
every two cardinal numbers can be joined by one of the three signs:
<, =, >. The theorem on trichotomy implies Zermelo’s theorem. Indeed,
let M denote a given set, m — its power. In virtue of Theorem 1 (proved
without the aid of the axiom of choice) there exists an aleph s(im) which
isnot <m. Thus from the theorem on trichotomy it follows that m < s(m),
which proves that the set M is equivalent to a subset of a well-ordered
set of power s(11), whence immediately follows the existence of a rela-
tion well-ordering the set M.
Now Zermelo’s theorem also implies the theorem on _ trichotomy
since, aS we know, trichotomy holds for alephs. Thus we have proved
THEOREM 6. The axiom of choice, Zermelo’s theorem on well-ordering
and the theorem on trichotomy are equivalent ').
According to Fraenkel [4], p. 37, it would be interesting to investigate
how far we could go in the Calculus and in the set theory if we accepted the
axiom of the possibility of ordering every set instead of the axiom of
choice *). As has been observed by Kuratowski (see Tarski [2], p. 82)
1) It was proved by Hartogs [1] (p. 442).
*) This axiom is essentially weaker than the axiom of choice. See A. M ostowski,
Uber die Unabhdngigkeit des Wohlordnungssatzes von Ordnungsprinzip, Fund. Math.
32(1939), p. 201-232.
 as As Equivalences 413

from the assumption that every set can be ordered follows the axiom of
choice for an arbitrary aggregate of finite sets’). Indeed, let Z denote
a set of finite non-empty disjoint sets, S — the sum of all sets belonging
to Z. The order arrangement of the set S determines the order arrangement
of each of the sets belonging to Z. But every finite ordered set has a first
element. The set N of all the first elements of the sets forming the set Z
will be a set containing one and only one element from each of those sets.
It is also easy to prove without resorting to the axiom of choice
that the proposition T, that every set can be ordered is equivalent to the
proposition T, that every set in which every ordering is a well-ordering
is finite. The proof that T,->T, presents no difficulties. Suppose now
that proposition T, is false for a set A. Thus in the set A there exists no
order arrangement, 7. ¢.gbhat set satisfies the assumption of proposition T,
and therefore, in virtue of that proposition, it is finite, 7. e. it can be
ordered, which gives a contradiction. Therefore T,—>T,. (Tarski [2], p. 82).
W. Kinna has proved that if for a given set A we know the function /
associating with every subset B, composed of more than one element
of the set A, a certain proper subset f(B) of the set B, then we are able
to order the set A (see XVI.4, p. 4382, Exercise).

EXERCISES. 1. Prove without resorting to the axiom of choice that if m is a

cardinal number and y is an aleph such that y<m, then m+y =m.

Proof. In view of y < m, there exists a cardinal number m, such that m= m,+ 9,
and, since we are able to prove without the aid of the axiom of choice that y+" = y,
we have m+s=(m,+8)+N=m,+ (N+) = m+ =m, whence m+N=—m, q. e. d.

2. Prove without the aid of the axiom of choice A. Tarski’s proposition that if
a cardinal number m is not finite and y(m) denotes the least aleph that is not <m, then
between the number m and m+y(m) there is no intermediate cardinal number.
Proof (after A. Tarski). Suppose that there exists a cardinal number n such that

(a) m<aon<mm+y(m).

In view of (a) we can write

(b) n= m’+ 9’
where

(c) m’<m and yw’ <N(m)

(since if Pc M+N, then P= M,+N,, where M,= MPcM, and N,= NPC N).
Further, we shall distinguish two cases:
1) s’=s(m). By (b) we have 8’ <n, whence, in virtue of Exercise 1, n+’ =n,
and since, by (a), n > m, we have n=n+y’ >m-+W’ and thus, by 1), n>m+x(m),
which contradicts (a).

1) But not vice versa (see Fraenkel [5], p. 1-25).

414 XVI. Zermelo’s theorem

2) ys’ £N(m). In view of (c) and the observation that for alephs trichotomy holds
(which is proved without the aid of the axiom of choice), we have x’ < s(m). Since
s(m) is the least aleph that is not <m and wy’ < ¥(m), we must have y’ < m, whence,
according to exercise 1, m-+¥’=m, which, by (c) and (b), gives m>m’+y’=n
contradicting (a).
Thus the assumption that there exists a cardinal number n satisfying inequalities (a)
leads in any case to a contradiction.
3. Prove without the aid of the axiom of choice that for cardinal numbers m
and n the inequality 2™ < 2" implies the inequality m < n.
Proof. Suppose that for cardinal numbers m and m the inequality m <n does
not hold. As we know, from the axiom of choice follows the theorem on trichotomy,
and, in view of (m < n) not holding, it follows that m > n, which, as we know, implies
the inequality 2 > 2". Thus if we have 2™ < 2" then m <n, q.e. d.

2. Various theorems on cardinal numbers equivalent to the axiom of

choice. AS has been proved in XVI.1, the axiowt of choice is equivalent
to the proposition that every non-finite cardinal number is an aleph.
From the theorems deduced for alephs in XV.8 and XV.10 it follows
that the axiom of choice implies for any cardinal numbers m and 1 the
formulae (where the cardinal number n is not finite):

(2.1) Tee
Ot ea Or One
ai rh,
(2.2) Naat LOT meray ae eae
(2.3) itt it for” iti an;
(2 4) Wie as TORAt) de
iit =o dh
l
From formula (2.1) it immediately follows that no.cardinal number
that is not finite can be the sum of two cardinal numbers smaller than that
number. This theorem thus follows from the axiom of choice. On the other
hand, as has been observed by S. Lesniewski, this theorem implies the
theorem on trichotomy. Indeed, if the cardinal numbers m and n are
not finite, then in virtue of this theorem their sum m+n cannot be greater
than each of those numbers, and since of course m+n >m and m+
+n >n, therefore either m+n—n, whence n <i or m+n= n, whence
m <n. And since, in virtue of Theorem 6 of XVI.1, the theorem on tri-
chotomy is equivalent to the axiom of choice, we have
THEOREM 1. Yhe proposition that a non-finite cardinal number is
not the swm.of two cardinal numbers smaller than that number is equivalent
to the axiom of choice (cf. Kiyoshi Iseki [1], p. 109).
Now let a cardinal :uinber m be termed prime if it is not the product
of two cardinal numbers :maller than that number.
THEOREM 2. The axicin of choice is equivalent to the proposition that
every non-fnite card na’ number is a prime number (Sierpinski [41],
ps 11742).
 § 2. Theorems equivalent to the axiom of choice 415

Proof. As we know, from the axiom of choice follows formula (2.1).

Suppose now that a non-finite cardinal number p is not a prime number,
i.e. that p= mn where p>m and p>n. As we know, the axiom of
choice implies the theorem on trichotomy: thus we may suppose that
m <n. The cardinal number n cannot be finite since in that case numbers
m<n and p= mn would be finite, in contradiction to the assumption
concerning number p. Thus we have formula (2.1), therefore p = n, con-
tradicting the assumption. Thus number p must be prime.
On the other hand, let us suppose that every non-finite cardinal
number p is a prime number and let m and it be any two cardinal num-
bers that are not finite. We have of course mi >m and mn>n, and the
cardinal number mn cannot be finite. Thus it is a prime number and
therefore we cannot have at the same time mn > im and mn>n. Thus
either mn = m, which gives m >n, or mn= Nn, which gives 11> im. Thus
the proposition that every Nodetinite cardinal number is a prime number
implies the trichotomy, 7. e. also, by Theorem 6 of XVI.1, the axiom of
choice.
Theorem 2 is thus proved.

EXERCISES. 1. Give an example of a cardinal number which is the product

of an infinite sequence of cardinal numbers smaller than that number.
Solution. 2° = 2-2-2...
2. Give an example of an aleph which is the sum of an infivite series of alephs
smaller than that aleph.
¢

Solution. §, =Si+ %+%s+..

3. Prove with the aid of the axiom of choice that every infinite set is the sum of
disjoint denumerable sets.
Proof. The axiomof choice is known to imply trichotomy: thus if m is a non-
finite cardinal number, then m >, and. by (2.1), we have m = ,m, whence the re-
quired proof easily follows.
4. Prove with the aid of the axiom of choice D. K6nig’s proposition that for
every set having more than one element there exists a function (1-1) mapping that
set on itself, such that no element of the set is mapped on itself (Konig [1], p. 336, and
Schoenflies [1], p. 42).
Proof. For finite sets the proof presents no difficulties, and for sets that are not
finite it easily follows from the formula = n+n tor every cardinal number nt that is
not finite (that formula being a particular case of formula (2.1), for m=).
5. Prove (with the aid of the axiom of choice) the formula of J. Mycielski:

(x) (Sim)? > TF |] m,,

teT teT

where 7 igs am arbitrary (non-empty) set and m, (for te 7’) are positive cardinal
numbers.
416 XVI. Zermelo’s theorém

Proof. For T finite the proof is easy. For JT not finite we have T = 2-7 and

(ex) (Sm)? =(YmJF-(Y) mJ.

We have

If m is a cardinal number, we have

m? — [J m,
: ueT
therefore
(ey rn a (Sita eae ame pati,
teT ueT teT ueT eT

Thus formula («*) gives formula (*) q. e. d.

From formula (2.1) it follows that 7f a cardinal number n is not finite,
then 2n =n. In this connection it will be observed that without resorting
not only to the axiom of choice but also to Zermelo’s theorem we are
not able to prove that every cardinal number n may be represented in
at least one of the following two forms:
We OG gt——ezitt
a It

(7. e. that every set, either whole or with one element removed from it,
can be decomposed into disjoint pairs, which intuitively seems almost
self-evident).
From formula (2.1) it follows that if a cardinal number n is not finite,
then n= 2n= 2n+1. It immediately follows that a set which is not
the disjoint sum of two sets of the same power as the set itself is finite.
This theorem we are also unable to prove without the aid of the axiom
of choice‘ (Tarski [2], p. 93, Def. V).
THEOREM 3. The axiom of choice is equivalent to the proposition that
the difference mw—n ewxists for every cardinal number m and every cardinal
number 1 <— m+).
Proof. It follows from the axiom of choice that the difference m—n
exists for every cardinal number wt and every cardinal number n < m.
Indeed, if m is a finite number and n < m, then the existence of the dif-
ference m—11 is obvious. Therefore let 1m be a non-finite cardinal number.
In order to prove the existence of the difference m—n it is necessary
and sufficient to prove, in accordance with the definition of the difference
of two cardinal numbers (see [X.1), that there exists one and only one
cardinal number p such that m=n+p. That such a number p exists

1) This theorem was given without proof by A. Tarski [5], p. 312, th. 82, A. I gave
the proof in [37], p. 125, Theorem 5,
 § 2. Theorems equivalent to the axiom of choice 417

follows from the fact that, by formula (2.1) (which follows from the
axiom of choice) and on account of the fact that number m is not finite,
we have m= n+m for n <m. Thus it remains to show that if for a cer-
tain cardinal number p we have m=n+p and n<m, then p=m. In
view of m=n-+p we have m>p; if, we had m>p, then number m
would be the sum of two cardinal numbers smaller than itself, in contra-
diction to Theorem 1. Therefore p= 1m, q. e. d.
Suppose now that the proposition that the difference m—n exists
for every cardinal number m and every cardinal number n < ™ is true.
Let m denote any non-finite cardinal number. In XVI.1 we have proved
without the aid of the axiom of choice that there exists an aleph s(m)
such that s(m) <m does not hold (Theorem 1 of XVI.1). We have of
course m+s(m) >s(m). If we had m+s(m) >s(m), then from our
assumption of the existence of the difference of cardinal numbers would
follow the existence of the difference [m+s(m)]—s(m), and thus the
existence of only one cardinal number p such that m+s(m) = p+s(m).
But the latter formula is of course true for p=m, as well as for p =
= m-+wx(m) since, as we know (see XV, (8.5)), s(m)+ (mM) = s(m).
Thus we should have m = m-+s(m), whence s(m) <m, which contradicts
the property of number s(m). Thus the assumption that m+s(m) >
>s(m) leads to a contradiction. Therefore we must have m+s(m) =
=x(m), whence m<s(m), whence it follows that the cardinal num-
ber m (which is not finite) is an aleph.
Thus from the assumption of the -existence of the difference of
cardinal numbers m—n (for 1<™m) it follows without the aid the of
axiom of choice that every non-finite cardinal number is an aleph. And
this, as we know, implies Zermelo’s theorem and the axiom of choice
GxcV fell).
Theorem 3 is thus proved.
THEOREM 4. The axiom of choice is: equivalent to the theorem on the
addition of inequalities for cardinal numbers, 1. e. to the proposition that
for any cardinal numbers m,n, m, and n, the inequalities

(2.5) Meat, Ona. iit =< thy

imply the inequality

(2.6) m+m, <nt+n, +).

Proof. Suppose that the axiom of choice is true and let us assume
that for cardinal numbers m, 1, m, and n, inequalities (2.5) hold. If the
cardinal numbers m and m, are both finite, then the number m+ 1, is
also finite, and inequalities (2.5) (both in the case where numbers 1 and 1,
1) A. Tarski, Fund. Math. 5(1924), p. 147-154.
Cardinal and ordinal numbers bo “1
418 XVI. Zermelo’s theorem

are finite and in the case where one or both of them are not finite) imply
inequality (2.6). Suppose therefore that at least one of the numbers m
and m, is not finite. The axiom of choice, in virtue of Theorem 6 of XVI.1,
implies trichotomy; thus we have either m<m, or m, <m. Suppose
that m << m,: number m, is thus not finite and, by formula (2.1) (resulting
from the axiom of choice), we have m+ ™m, = m,. Hence, by (2.5), we
obtain m+m, = m, <n, <n+n,, and thus formula (2.6). Therefore the
axiom of choice implies the theorem on the addition of inequalities for
cardinal numbers.
Suppose now that the theorem on the addition of inequalities for
cardinal numbers is true and let m denote any cardinal number that is
not finite. Let
(2.7) Teale 3
whence we immediately obtain
(2.8) 2 2, — t(2Ko) — Is, — It -

By (2.7), the cardinal number n is not finite. Hence it follows, as

has been proved in XVI.1 without the aid of the axiom of choice, that
there exists an aleph x(t) such that s(n) <n does not hold. We have
of course s(n) <n+nx(n). If we had x(n) <n+s(n), then, in view of
n<1N+x(n) (since of course n <n+x(n) and from n= n+x(n) it would
follow that s(1) <n in contradiction to the property of number s(n))
and in virtue of the theorem on the addition of inequalities, we should
obtain
n+s(1) <[n+s(n)]+[1+s(1)] = 2n-+4 2s(1).

From (2.8) and from the formula for alephs 2s = s proved in XV.8,
we should have the formula
n+x(t)<n+s(n),

which is impossible. The inequality s(n)<n+s(n) cannot hold and

therefore we must have x(n) = n+ s(n), whence n <s(n) and, by (2.7),
naturally m <s(n); it follows that the cardinal number m (which is not
finite) is an aleph.
Thus from the theorem on the addition of inequalities it follows
without the aid of the axiom of choice that every cardinal number that
is not finite is an aleph, whence, as has been proved in XVI.1, follows
the axiom of choice.
Theorem 4 is thus proved.
The theorem on the addition of two inequalities for cardinal num-
bers immediately implies, by induction, the theorem on the addition of
an arbitrary finite number of inequalities for cardinal numbers. However,
 § 2. Theorems equivalent to the axiom of choice 419

it will be observed that the theorem on the addition of an infinite sequence

of inequalities does not hold for cardinal numbers, since we have for
instance n <s, for n=1,2,... but 14+24+3+4+...=8,+8,+... because
both sides are equal to sp.
Let us also note that A. Tarski has proved the equivalence of the
axiom of choice to the proposition that if m,, m,, ... is an infinite sequence
of cardinal numbers such that for a certain cardinal number n we have
fore}

1 eS ey: mM, 9

ne

then there exists a natural number p such that

Pp

a eeuNeTte

(Tarski [11],. p.. 79).

Similarly to Theorem 4 we can prove
THEOREM 5. The axiom of choice is equivalent to the theorem on the
multiplication of inequalities for cardinal numbers, 7. e. to the proposition
that the mequalities (2.5) imply the inequality

(2.9) mm, <1,

(Tarski [1], p. 153-154).

Proof. The proof that the axiom of choice implies the theorem on
the multiplication of inequalities for cardinal numbers is analogical to
the first part of the proof of Theorem 4. In order to prove that the con-
verse is also true we take n= m*, instead of (2.7), whence it follows
that n? = mo = me =—n. Now n< s(n) (which is proved in exactly
the same way as the inequality n <n+s(n) was proved above). If we
had simultaneously the inequalities n < ns(n) and s(n) < ms(n), the first
of them being proved in a similar manner as the inequality n < n+x(n),
then from the equalities n? = n, [s(n)]? = x(n) (the theorem of Chapter XV
being valid for alephs) and from the assumption that inequalities for
cardinal numbers can be multiplied by sides, we should obtain ms(n) <
< [xs (11)}? = 1x(1), which is impossible. Thus in view of n < nx(n), we
must have s(n) = ns(n), which gives n <x(n) and m < me =n < x(n)
and proves that a cardinal number m which is not finite is an aleph.
THEOREM 6. The axiom of choice is equivalent to the proposition that
for every cardinal number m that is not finite the following formula holds:
(2.10) eal
(Tarski [1], p. 147-151).
423.0 XVI. Zermelo’s theorem

Proof. Formula (2.10) for cardinal numbers m that are not finite
immediately follows from formula (2.1); thus formula (2.10) follows
from the axiom of choice.
Suppose now that formula (2.10) is true for every cardinal number m
that is not finite and let m denote any such number. Let m= s(m). In
virtue of our assumption, we shall have:
(2.11) SA eh) le NW, (m+ n)?=m-+n
and since (m+n)? = n?+ 2mn+1, by (2.11), we have
m+n=m+2mn4+n,

whence mn < 2mn < m+ 2mn+n = m+N, @. €.

(2.12) Mit <—tt 1

On. the other hand, we may of course set m= m,+1, n= n, +1 where m,

and n, are cardinal numbers >0, whence mn = (m,+1)(1, +1) = m,n, +
+m,+n,+1>14+m,+n,+1=m-+n; thus inequality (2.12) gives the
formula
(2.13) mni=m-+n.
Let M and NV be sets of power m and n respectively: as we. know,
it can be assumed that the sets WM and N have no common elements,
and in view of n= s(m) we can assume that the set N is well-ordered.
Further, let P= Mx WN. Thus P= m1, therefore, by (2.13),

P=m-+4+n and P= M,+N,

where
(2.14) UNO MN en
Two cases are then possible: 1) there exists an element m, of the
set M such that for every element n of the set N the element (m,, 7)
belongs to M,, or 2) for every element m of the set M there exists at least
one element » of the set N such that the system (m,n) belongs to N,.
In the first case let us denote by N, the set of all systems (m,, 2)
where n « N; of course we shall have N, C M,, N,= N therefore M, > N
and thus, in view of V =n and (2.14),
(2.15) TU e
In the second case, for me M, let us denote by f(m) the first element
of the well-ordered set N such that the system (m, 7) belongs to N,.
Let M, denote the set of all systems (m,f(m)) where me M. We shall
have M,C N,, M,= M, whence M < iN which, in view of M—=m and
(2.14), gives
(2.16) its 1.
 § 2. Theorems equivalent to the axiom of choice 421

Thus we have either inequality (2.15) or inequality (2.16). But,

in view of n= s(m), inequality (2.15) cannot hold. Therefore inequality
(2.16) must hold, 2. e. m <s(m), whence it follows that a cardinal num-
ber m that is not finite is an aleph.
We have proved (without resorting to the axiom of choice) that if
for every cardinal number m that is not finite formula (2.10) holds, then
every cardinal number that is not finite is an aleph, and hence, as we
know (see XVI.1) follows the axiom of choice.
Theorem 6 is thus proved. What makes it remarkable is that formula
(2.10) (for cardinal numbers m that are not finite) might seem to be
a very particular conclusion from the axiom of choice, while actually,
in virtue of Theorem 6, it is equivalent to the axiom of choice.
It will be observed, however, that we are unable to solve the ques-
tion whether the formula 2m = m for cardinal numbers that are not
finite is equivalent to the axiom of choice.
A Tarski has observed moreover that the proposition (for cardinal
numbers) stating that

(2.17) If 2m<am-+n, then m<n

is equivalent to the axiom of choice, while the proposition:

(2.18) If 2m >m-+n, then m>n

can be proved without the aid of the axiom of choice.

The proof that the axiom of choice implies formula (6.17) presents
no difficulties. For the axiom of choice, by Theorem 1 of XVI.1, implies
trichotomy; therefore if we did not have m<n, then we should have
m>n, whence 2m = m+m > m+n, which is contrary to the assump-
tion in (2.17). In order to prove that (2.17) implies the axiom of choice
let us suppose that n is a cardinal number that is not finite and let m=
= x(n). Thus 2m=m<m-+n. If we had 2m<m-+n, then, by (2.17),
we should have m<n, 2. ¢e. s(t) <n, which contradicts the property
of number s(1). Thus we have 2m = m= m+n, therefore n < m= x(n),
whence it follows that number n is an aleph.
As regards the proof of (2.18) without the aid of the axiom of choice,
it will be observed, to begin with, that the inequality 2m > m-+11 implies
the existence of a cardinal number.p such that 2m = m+n+p, whence,
by Theorem 3 of I[X.1 (of F. Bernstein) (proved without the aid of the
axiom of choice), we obtain m>n-+p, therefore m>n. If m=n, we
should have 2m = m+n, which contradicts the assumption. Thus m > n,
seams Pam
As regards Theorem 6, it will also be observed that we can express
it without the aid of cardinal numbers in the following way:
422 XVI. Zermelo’s theorem

The axiom of choice is equivalent to the proposition that every non-

finite set M is of the same power as the set MxM.
However, it is not for every set which is not finite that we can
establish a (1-1) correspondence between the sets M and Mx M.
THEOREM 7. The axiom of choice is equivalent to the proposition that
for cardinal numbers m and n the equality m? = 1 implies the equality
ni" (Dalrskiz 2], Spr ilp ied 52):
Proof. If m and n are finite cardinal numbers such that m? = n
then we have of course m = n. If m and nt are non-finite cardinal numbers,
-then, aS we know, the axiom of choice implies formula (2.2), therefore
nm? = m and n? =n and thus in the case of m? = rn? we also have m= n.
We have proved that from the axiom of choice it follows that for cardinal
numbers the equality m? = 1? implies the equality m= n.
Suppose now the proposition that for cardinal numbers m and n
the equality nt? = n? implies the equality m = n to be true. Let m denote
an arbitrary non-finite cardinal number and let

(2.19) n = me,
(2.20) p=n+rx(t),
(ON) q = n-s(n).
By (2.19) we obtain
(2.22) n2 = (m0)? — mo? — me =n,
whence, by (2.20)

(ee) p? = [1-48 (1)? = 12 + 2ns (0) + [9 (1) =

= n+: 28(n)+ x(n) = [N+x(1)]4+1-s(n).

In view of n>1 and x(n) >1 we have n+ x(n) < nKx(n), whence
Ms (11) < [11+ 8 (11)]+ 118 (11) < THs (11) + 18 (11) = 1128 (11) ] = Ts (11)
whence
(2.24) Ms (11) = [11 + 8s(1) ] + 1s (11)
Formulae (2.23) and (2.24) give the equality
(2.25) p? = its (11).
By (2.21), (2.22) and (2.25), we obtain

g? = [ns (1) P= ref (1) P= mx (11) =p?

which in view of the assumed proposition implies the equality p= aq,
1. @., by (2.20) and (2.21), the equality
(2.26) n+
8(1) = 18 (N).
 § 2. Theorems equivalent to the axiom of choice 423

But in the proof of Theorem 6 we have shown that formula (2.13),

where n = (11), implies (without the aid of the axiom of choice) that
a cardinal number m, provided it is not finite, is an aleph. Thus formula
(2.26) implies that the cardinal number n (which, in view of (2.19), is
not finite) is an aleph, and since, by (2.19), 1 > m, the number m, which
is not finite, is an aleph.
Thus the assumed proposition implies that every non-finite cardinal
number is an aleph, whence, as we know from XVI.1, follows the axiom
of choice.
Theorem 7 is thus proved.

EXERCISES. 1. Prove A. Tarski’s [1] (p. 150) theorem stating that the axiom
of choice is equivalent to the proposition that for non-finite cardinal numbers m and n
we have mn = m-—n.
Proof. The axiom of choice, by Tbeorem 6 of XVI.1, implies trichotomy: thus
if m and n are non-finite cardinal numbers, then m<n or Mn, e.g. m<n. But,
as we know, the axiom of choice implies also formula (2.1): thus mn = m+n,
Suppose now the proposition that for non-finite cardinal numbers mt and n we
have mn = m-+-n to be true and let m denote an arbitrary non-finite cardinal number.
Let n = §$(m); we shall have formula (2.13), which, as we have shown in the proof of
Theorem 6, implies that the cardinal number m is an aleph. Further proof is obvious.
2. Prove A. Tarski’s ([1], p. 154) theorem stating that the axiom of choice is equi-
valent to the proposition that for any cardinal numbers m,n and p the inequality m-+-
+ <n-+p implies the inequality m <n.
Proof. As we know, the axiom of choice implies trichotomy; thus if m < n does
not hold, then m =n, whence m+ p >n-+p. Thus it follows from the axiom of choice
that if m+ p <n-+p, then we must have m <n.
Suppose now that for cardinal numbers m,n and p the inequality m+p <n-+p
always implies the inequality m <n. Let n denote an arbitrary non-finite cardinal
number. We have of course $(n) << n+wN(n). If S(n) < N+N(N) then, in view of $(m)-+
+8(n) = S(n) we should have §$(n)+ (nN) < n+8(n), whence, in view of the assumed
proposition (form = $(n), p = 8$(n)), S$(N) <n, which contradicts the property of num-
ber N(n). Therefore we must have S$(n) = n+8(n), whence n <N(n) and number n is
an aleph. Further proof is obvious.
3. Prove A. Tarski’s theorem stating that the axiom of choice is equivalent to
the proposition that for any cardinal numbers m, n and p the inequality mp < np implies
the inequality m <n.
Proof. The proof is analogical to the proof of Exercise2, the sums of cardinal
numbers being replaced by their products.
4. Prove A. Tarski’s') theorem stating that the axiom of choice is equivalent to
the proposition that for any cardinal numbers m, p and g the equality m+ p = im+q
implies that either p=g or p<m and g<m.
Proof. Suppose that for cardinal numbers m, p and q we have the equality m+ p =
= m-+q andp =q does not hold. In virtue of Theorem 6 of XVI.1, the axiom of choice
implies trichotomy: thus we have either p > q org > p, e. g.p > q. If we had alsop > m

1) This theorem was given without proof in Tarski [5], p. 312.

424 XVI. Zermelo’s theorem

and if the number p were finite, then the numbers m and q would also be finite and
it would follow from the equality m+p =m-+q that p=4q, which is contrary to the
assumption. Thus if p > m, then the number p is not finite and, by formula (2.1) result-
ing from the axiom of choice, we obtain m-+ p =p, therefore p =m-+q where p > m
and p > q, while, by Theorem 1, it follows from the axiom of choice that a non finite
number p may not be the sum of two cardinal numbers smaller than itself. Thus p > m
is impossible, therefore we have p< m, and, in view of p > q. q <m. Of course if
gq >p, we should similarly obtain q<m and p < m.
Thus we have proved that from the axiom of choice follows the proposition that
for any cardinal numbers m, p and q the equality m+ p= m-4q implies that either
De qorep— Nl andad site
Suppose now that the latter proposition is true and let n denote any unon-finite
cardinal number. As we know, n <n+y(n) and y(n)+n = y(n)+[n+(n)]; thus the
assumed proposition implies (for m = y(n), p =n, gq=n+y(n)) that n < y(n), whence
it follows that the cardinal number n is an aleph. Further proof is obvious.

THEOREM 8 (of A. Tarski). The axiom of choice A is equivalent to the

following proposition. Q:
Q. For every cardinal number m there exists a cardinal number
such that
Lh) aie
and
2) m<p implies n<p
(Tarski [14], p. 26-32) (in other words: for every cardinal number among
all cardinal numbers greater than that number there exists a least).
In order to prove this equivalence we shall first prove without the
aid of the axiom of choice the following
LemMA. If mis any cardinal number and s an aleph such that s < 1m’,
then & <™m.
Proof. Let M denote a set of power m. If s < n¥, then there exists
a well-ordered subset P of the set Mx M which is of power s. Let us
denote by A the set of all elements x of the set M tor which there exists
at least one element y « M such that (7, y)« P, and by B — the set of
all elements y of the set M for which there exists at least one element x
of the set M such that (x7, y) « P. Since the set P is well-ordered, there-
fore for each element ve A the (non-empty) set of all elements y of the
set M such that (7, y) « P.is well-ordered; let y, denote the first element
of the set. In this manner with each element xe A there is associated
a certain element (x7, y,) of the set P, and of course with different elements
of A different slerrenis of the set P are associated. Hence it follows
thou P, therefore A <x. Similarly we prove that Reo ee te
Slee
TB; we shall have n <A eyes
=S+8 =k, whence nt <x, therefore
1? <ss =x. But obviously PC Ax B, whence s= P= AXB<in-n< Ss,
 § 2. Theorems equivalent. to the axiom of choice 425

which gives n?=y. It follows that n is not a finite number and

thus, in view of n <x, 1 is an aleph, therefore n? = n. Thus we have n= sx.
But in view of AC M, BC M we have A+BC_M and thusn= 4+B<m,
which gives s <™m, q.e. d. Our lemma is thus proved (without the aid
of the axiom of choice). .
Proof of Theorem 8. From the axiom of choice it follows, as we
know, that every non-finite cardinal number is an aleph, and for every
aleph there exists a next aleph, which is the least of all alephs greater
than the given aleph. Thus it follows from the axiom of choice that pro-
position Q is true for every non-finite cardinal number im; for finite num-
bers m proposition Q is obviously true (and then n= m+1). Thus we
have AQ.
Suppose now that theorem Q is true. Let m denote a given non-
finite cardinal number, and m—.a cardinal number | satisfying condi-
tions 1) and 2). Since, as we know, m < m+s(m) (see XVI.1. therefore,
by 2), m <n <m-+ns(m), and since, as we know (see Exercise 2 of XVI.1),
there is no intermediate cardinal number between m and m-+ (mt), there-
fore we must have n = m+s(m). Thus from property 2) of number 1 it
follows that if m<p then m+s(m) <p.
Now suppose that m< m’*. In view of property 2) of number n=
= m-+sx(m) we should have m+s(m) <1? and therefore s(m) < 1m’,
which, in virtue of our lemma, gives s(m) <m, which contradicts the
definition of number s(m). Thus in view of m < 1? we have m= nv? for
every non-finite cardinal number, which according to Theorem 6 implies
the axiom of choice. Thus we have proved that Q—A.
In view of A+Q and Q—A, we obtain A =Q and Theorem 8 is
thus proved.
THEOREM 9 (of G. Kurepa [7]). The axiom of choice is equivalent to
the proposition that the following two sentences are true:
K,. There exists an (ordinary) order arrangement of every set.
K,. In every partially ordered set there exists a subset that is saturated
with respect to the following property: each two of its elements are incomparable.

Proof. Since propositions A, and A, easily follows from the axiom

of choice, it suffices to deduce the axiom of choice from those proposi-
tions. Therefore suppose that propositions K, and Wy, are true, and let I
denote a family of non-empty disjoint sets, and S — the sum of all
sets of family F. In virtue of A, there exists an order arrangement o of
the set S. Let us partially order the set S by the relation 0, defined as
follows: for ae S,beS the relation ae,b holds if and only if a and b are
elements of the same set of family F and aob. In virtue of A, there exists
426 : XVI. Zermelo’s theorem

in the set S a subset 7 saturated with respect to the following property:

the relation 9 does not hold between any two different elements of it.
Now let Z denote any set belonging to family F. If TZ = 0, then,
the set Z being non-empty since it belongs to family F, there exists
an element pe Z, and from the definition of the relation o, and of the
set T it follows that the relation 0, does not hold between any two
different elements of the set 7+ {p}. But in view of pe Z and TZ= 0
we have péT; this contradicts the above-mentioned property of the
subset 7 of the set S, namely its being saturated with respect to the follo-
wing property: the relation 0, holds between no two different elements of it.
Thus TZ + 0 for every set Z of family F, and since the set 7 cannot have
more than one element in common with any set Z of family F, the rela-
tion o,, holding between each’ two different elements of the set Z,
the set 7 has one and only one element from each set Z of family F.
In view of the arbitrariness of the family F of non-empty disjoint sets,
this proves the truth of the axiom of choice.

3. A. Lindenbaum’s theorem. We shall prove .

THEOREM 1 (of A. Lindenbaum)!). The axiom of choice is equivalent
to the proposition that for any cardinal numbers m and n we have either
11 Se TOP Seite
Proof. In VIII.4 we have denoted by m < «n, after A. Tarski, the
relation between cardinal numbers m and n according to which either
m = 0 or every set of power nt is the sum of m disjoint non-empty sets.
By Theorem 6 of XVI.1, the axiom of choice implies trichotomy; thus
if m and nm are given cardinal numbers, then either m <n or m>n. If
m<n and m0, then every set of power n is of course the sum of m
disjoint non-empty sets, for instance such that all of them except one
are one-element sets. If m >n and n+ 0, then every set of power m is
the sum of n disjoint non-empty sets. Thus the axiom of choice implies
that either m <*n or n <*™.
To continue the proof of Theorem 1 we must now prove without
the aid of the axiom of choice the following
LemMMA. For every cardinal number m there exists an aleph s‘(m)
such that s'(m) <*m does not hold.
Proof of the lemma. If the cardinal number m were finite, we
could of course set (1m) = 8); therefore we shall assume further that
the cardinal number m is not finite. Let us denote by Z the set of all
ordinal numbers a such that a <*m. From the assumption that the

1) This theorem was given without proof (see Lindenbaum et Tarski [1]). I gave
the proof in my paper [47].
 § 3. A. Lindenbaum’s theorem 427

cardinal number m is not finite it follows that the set Z is infinite since
of course n« Z for n= 1, 2,... It is easy to verify that the set Z is such
that if ae Z and €<a then fe Z. The set Z is well-ordered according to
the magnitude of the ordinal numbers that belong to it; let ¢ denote its
order type. If we had ¢ « Z, then the well-ordered set Z would be similar
to its proper segment formed by the element ¢ of the set Z, which is
impossible. Therefore ¢¢ Z and if we set s/(m) =, then we shall not
Se

have s/(m) <*«m™m.

To remove a susp:c'on of the set Z being the set of all ordinal
numbers, which would lead to the antinomy of Burali-Forti, it can be
proved that every number a of the set Z satisfies the inequality a < 2™.
For if n <*m, then there exists a decomposition of the set m of power m
into n disjoint non-empty subsets. Thus there exists a set of power n,
composed of different subsets of the set MM, and since the set of all
different subsets of the set M is of power 2™, we have n < 2™. Thus each
number a of the set Z satisfies the inequality a < 2", q. e. d.
Let us return to the proof of Theorem 1. Suppose that the proposi-
tion that for any cardinal numbers m and n either m < «nt or 1<*m
is true. Let im denote an arbitrary cardinal number 40 and s’(m) — an
aleph satisfying our lemma. Thus for 1 = s/(m) we do not have n <*™;
therefore, in virtue of the assumed proposition, we must have m < «*n,
a.€. m <x«y(m). This means that a well-ordered set A of power s’/(m)
is the sum of m disjoint non-empty sets. The first elements of each of
those sets form of course a subset of power m of the set A. Thus the car-
dinal number m is the power of a certain well-ordered subset of the set A,
2. é. it is either a finite number or an aleph. Thus it follows from the as-
sumed proposition that every cardinal number is either a finite number
or an aleph, which, as we know from XVI.1, implies the axiom of choice.
Theorem 1 is thus proved.
4. Equivalence of the axiom of choice to the theorems of Zorn and Teich-
miiller. Let us call a chain every family of sets such that if A and B are
two sets of that family, then either ACB or BCA.
Let us call closed every family F of sets such that if | is a chain
contained in #, then the sum of all sets belonging to /, belongs to F.
Zorn’s theorem can be formulated in the following way (Zorn [1],
Ds 92): ~apalice
Z. In every closed family of sets there is at least one set not contained
im any other set of that family.
THEOREM 1. Theorem Z is equivalent to the axiom of choice (Bernays
[1], p. 98, Remark; Birkhoff [1], p. 44, Theorem 10).
Proof. 1. From theorem Z follows the axiom of choice.
428 XVI. Zermelo’s theorem

Let A denote a given set, U = U(A) — the set of all subsets of the
set A. Let us denote by J’ the set of all functions f7;(X) defined for X « T
where 7C U and such that fr(X)e¢«X for XY « 7. In particular, for in-
stance, if 7, denotes the set of all subsets of the set A consisting of one
element only and if we set f;,({v}) = x for we A, then we shall have fr, € 1’.
Let fr,<9r, if frel, gr,e DT, TyC Ts, Ty ATs, and fr(X)= gr,(X)
fore West...
The set /’ contains subsets ordered according to the relation x,
e.g. the subset 4) = {f7,} (A one-element set CJ is regarded as ordered.
according to the relation <).
Let @ denote the family of all subsets of the set /’, ordered accord-
ing to the relation S. Thus J, « G. We shall prove that the family @ is
closed.
Let G, denote a chain contained in G and let

Sea —
ahi:
HeG,

In view of the definition of G, we have G, C @; every component of the

sum S thus belongs to G, ¢. e. is a subset of the set J’, whence it follows
that also S is a subset of the set J’, 7. e. it is the set of certain functions fr.
Now if fr,e gy and gr,<«g where fr, ~ gr,, then, in view of

s= >a,
HeG,

there exist sets H, « G, and H, « G, such that fr, « H,, Gr, « Hz, and since G,
is a chain we have either H,CH, or H,C H,, e.g. H,C H,. But then
fr,« H, and Gr, «H,. Since H,¢G, it follows from the definition of fa-
mily G that the functions forming the set H, are ordered according to
the relation <; in view of fr, 4 gr, we have either fr, Sgr, or gr, < fr,-
The set (of functions) S is thus ordered according to the relation < and,
in view of the definition of family G, it follows that S « G.
We have proved that family @ is closed. In virtue of theorem Z
there exists a set V eG, not contained in any other set of family @. In
view of V eG, V is the set of certain functions f;, ordered according to
the relation <.
We shall prove now that if XC A and X 40, then there exists
a function fre V such that XY « 7. Indeed, suppose that for a certain non-
empty set X,C A.and for every function freV we have X,¢ T. Let
x,e X, (such an element exists because VY, 4 0) and let us define a set O
and a function fe as follows. Let 0 = YT where the summation is extended
over all sets 7’C U for which there exists a function fr such that fre V.
Now if X «0, then, in view of the definition of the set 0, for a certain
 § 4. Theorems of Zorn and Teichmiiller 429

component 7’ of the sum @ we have X « 7 and there exists a function

freV. Let fo(X) = fr(X). Obviously, fo(X) will not depend on 7 since
if X « T and if there exists a function g7,<« V, then, in view of fre V and
of the definition of the set V, we shall have f7 = gr, or fr S gr, OF Gr, <r.
If for instance fr-Sgr,, then (in view of the definition of the relation <)
we have TC T, and fr(X) = gr,(X) (since X « T).
Thus the function fe(X) is defined for every set X ¢«@ and we ob-
viously have gr~<fe for every function gre« V. .
Now let 0, = 0+ {X,} and for X <«@: feX) = fe(X), and fe( X,) = 2%.
We shall have foe<fe, for fre V.
Let V,=V-+ {fe}: we shall obviously have V,«G and VCY,,
VAY, because fo¢V (since X,<«O, and X,¢ T for every T such that
there exists a function fre V; thus if fo,« V, then we should have X, «<0,,
which gives a contradiction). Thus the set V is a subset of the set V, « G,
different from V, which contradicts the property of the set V.
We have proved that if X C A, then there exists a function fe V
such that X « 7. In view of the definition of the set 0 we thus have X ¢@
for X C A. The function fe(X) associates with each subset X of the set A
a certain element of that subset. Thus theorem Z implies the existence
of such a function for every set A, which, as we have proved in VI.3,
implies the axiom of choice.
2. From the axiom of choice follows theorem Z').
Let F denote a given closed family of sets. By Theorem 6 of XVI.1,
the axiom of choice implies Zermelo’s theorem: thus there exists a trans-
finite sequence {Azg}:-,, formed of all the sets of family PF.
Suppose that theorem Z is not true for family F. Each set of the -
sequence {As}ec, 18S thus contained in another set of that sequence.
Now for every ordinal number a we shall define by transfinite induc-
tion an ordinal number 4, <q as follows. Let 2, = 1. Now let a denote
a given ordinal number >1 and suppose that we have already defined
numbers Az: < m for é <a and that A;, @ Aj, for 6 <2 7 = as Let

= Ly Aig’
E<a
it will be a set of family # (in view of its being closed) and therefore it
is contained in a certain set, different from it, of family 7. Thus there
exists a least ordinal number 4,<g such that 8,C A, and SA Ap
The ordinal numbers A, are thus defined by transfinite induction for
every ordinal number a and /, < 9.

1) Cf. Hausdorff [2], p. 140, where the author proves the existence in every partially
ordered set of a maximal ordered subset.
430 XVI. Zermelo’s theorem

Now let a and f be two ordinal numbers such that a < p. Hence,
in view of
S:= >) Ar,
é<p
we obtain. A; C 8s, and since Sp C Aas, we have Ajo An. If we had
Aq = 4g, then, in view of A,,C SpC Ag, we should have S; = Aig, which
is impossible. Thus A, 4 4g. On the other hand, in view of S8,C 4,C A a5
we obtain S,C A;,, and since 4, is the least ordinal number <q such
that S,C As, A, <dAg. Thus A, < dg. Therefore A, is an increasing func-
tion of the ordinal variable a, whence, as we know, it follows by trans-
finite induction that 2, > a for every ordinal number a and, in particular,
Ay > v, Which is impossible, because for every ordinal number a we have
Aa < gy. Therefore theorem Z must be true.
Theorem 1 is thus proved ?).
We shall say that a given family F of sets has property T if the condi-
tion Ae F is equivalent to the condition that each finite subset of the
set A belongs to F.
Property T of a family of sets obviously implies its being closed
(but not vice versa). Indeed, if F,C F, if family F, is ordered according
to the relation C, if
s= A
A€F,

and if. P denotes a finite subset of the set S, e.g. P = {pry Do, --»5 Pn}
then there exist sets A,, A,,..., A, of the family F, such |that p; «A;
for +=1,2,...,”. But the family F, is ordered according to the rela-
tion C; thus among the sets A,, A,,..., A, there exists such a set A
that A;C A, for += 1,2,...,. Thus the set P is a finite subset of the
set A, of family F; if family F has property T, then Pe F. Thus each
finite subset of the set S belongs to #, whence, again in view of the pro-
perty T of family F, we conclude that S« #. Family F is thus closed,
ire. .e:,
Feichmiller’s theorem runs as follows:
T. In every family of sets that has property T there exists at least one
set not contained in any other set of that family (Tarski [16]; Bernays [1],
j es) a)
THEOREM 2. Theorem T is equivalent to the axiom of choice.

1) Various authors deduce Zorn’s theorem directly from the axiom of choice
without using Zermelo’s theorem, e.g. Kneser [1], p. 110-134; Szele [1], p. 254-256;
With [1], p. 434-438.
*) G. Birkhoff [1], p. 42, theorem (AC 3), attributes this theorem to J. W. Tukey.
 § 4. Theorems of Zorn and Teichmiiller 431

Proof. In view of Theorem 1 it suffices to prove that theorem T

is equivalent to theorem Z. Since, as has been proved above, property T
of a family of sets implies its being closed, we have ZT. It remains to
show that TZ.
For this purpose we shall first prove that theorem T implies that in
every family F of sets there exists a chain which is not a subset of any
other chain contained in family F#’. To begin with we shall observe that
a family of sets is a chain if and only if each finite subset of that family
is a chain. Thus the family F’ of all chains contained in an arbitrary
family # of sets has property T. Therefore it follows from theorem T —
that every family F' of sets contains at least one chain C that is not con-
tained in any other chain contained in F. Now if family / is closed, then

S= AcF.
AEC

We shall prove that the set S is not contained in any other set of
family Ff. Suppose that SC A,eF and SA#A,. Let C, = C+{S+A,)}.
Since ACS for A « 0, we have ACS+A, for Ae C,; from the formula
for C, and the observation that OC is a chain it follows that the family C,
of sets is a chain. But CCC, and C4~C, because S+A,¢C (for if
S+A,« C, we should have S+A,C 8S, which is contrary to Aj #4SC A,).
Chain C would thus be contained in another chain contained in F, which
is contrary to the property of chain C.
We have proved that the set S (belonging to family F’) is not con-
tained in any other set of family F.
We have proved that T—Z and thus the proof of Theorem 2 is
complete.
It will be observed, moreover, that we can prove (without resorting
to the axiom of choice) that theorem Z is equivalent to a certain theo-
rem Z,, concerning partially ordered sets.
Let A denote a partially ordered set, A, — an ordered subset of
the set A. If there exist in the set A elements succeeding each element
of the set A, and if among those elements there exists one that precedes
all the others, then it is termed the wpper bound of the set A,. A partially
ordered set A whose each non-empty ordered subset has an wpper bound
is called inductive. Each element Of a partially ordered set A that does
not precede any other element of the set A is termed a maximal element
of the set A.
Theorem Z is equivalent to the following theorem Z,:
Z,. In every inductive set there exists at least one maximal element
(Bourbaki [1], p. 37; Wallace [1], p. 278).
432 XVI. Zermelo’s theorem

EXERCISE. Prove the following theorem of W. Kinna!):

If for a given set A of power >1 there is defined a function f associating with
each subset X of the set A, having more than one element, a certain proper subset f(X)
of the set X, then we are able to define in the set A a relation o ordering the set A.
Hint. Let @ denote the least family of subsets of the set A that has the following
three properties: 1) A «@, 2) If X «@® and X > 1, then f(X) «@ and X —f(X) «@, and 3)
The product of an arbitrary aggregate of sets belonging to ® belongs to @.
It is easy to prove that the least family of sets satisfying conditions 1), 2) and 3)
does exist: it is the product of all families of subsets of the set A satisfying condi-
tions 1), 2) and 3).
Now let p and qg be two given elements of the set A, p ~ q. Let us denote by 7
the product of all sets of the family ® containing the elements p and q. If p « f(T), let
poq: m the contrary case let gep. It can be proved (although it is by no means easy)
that the relation 9 defined in this way orders the set A.
Frony our exercise it follows that in order to prove that every set can
be ordered it is sufficient to use, instead of the axiom stating the existence
of a function associating with every non-empty set a certain element of
that set, a weaker axiom, which states that there exists a function associat-
ing with every set consisting of more than one element a certain proper
subset of that set. By a reasoning similar to that used by A. Mostowski
in his proof that the theorem on the existence of an order arrangement of
every set is weaker than the theorem on the existence of a well-ordering
arrangement of every set it is possible to prove that this axiom is weaker
than the axiom of choice (see the paper quoted in?) on p. 412).
THEOREM 3. Theorem T stating that in every family F of non-empty
sets there exists a subfamily F,, maxvmal monotonic (i.e. such that of each
two sets of family F one is contained in the other and that family F, is not
contained in any other subfamily of family F with similar properties) is
equivalent to the axiom of choice.
Proof. It is easy to prove that Zermelo’s theorem implies theorem T.
Therefore we shall confine ourselves to proving that theorem T implies
the axiom of choice.
Let A denote a family of disjoint non-empty sets. Let # be the fa-
mily of all sets X contained in the sum of all sets of family A and
having at most one element in common with each set of family A. It
follows from theorem T that there exists a maximal monotonic sub-
family F, of family F’. Let Z denote the sum of all sets of family F,.
we shall prove that the set Z contains one and only one element from
each set of family A.
Suppose that the set Z contains at least two different elements,
a, and a,, from a certain set X, of family A. Since Z is the sum of all

1) Given (without proof) by W. Kinna in a letter to the present author, dating

from 7-th July, 1952.
 § 4. Theorems of Zorn and Teichmiller 433

sets of family F',, there exist sets Y, and Y, of family F, such that
a,¢« Y, and a,« Y,, and since family F, is monotonic, we have either
Y,C Y,or Y,C Y,. Thus in any case one of the sets Y, and Y, of family 7, CF
contains two different elements of a certain set X, of family A, contrary
to the definition of family F. Hence the set Z contains at most one element
from each of the sets of family A.
Suppose now that for a certain set X, of family A we have X,Z = 0.
The set X, is not empty since it belongs to family A; accordingly there
exists an element ae X,. Thus we have aé Z and it follows from the
definition of the set Z that a is not an element of any set of family F,.
Denote by F, the family composed of the sets of family F, and of the set

Q= {a+ DY.
YeFy

Obviously we shall have Qe F and family F, will be a monotonic sub-

family of family F, family F, being a proper subset of family #,, which
contradicts the fact that family F, is the maximal monotonic subfamily
of family F Thus the set Z contains one and only one element from each
set of family A. The set A of non-empty disjoint sets being arbitrary,
we have proved that theorem T implies the axiom of choice; q. e. d.
THEOREM 4 (OF R. VAUGHT). Theorem T stating that in every family F
of non-empty sets there exists a maximal subfamily F, of disjoint sets (4. e.
sets not contained in any other subfamily of family F composed of disjoint
sets) is equivalent to the axiom of choice (Vaught [1], p. 66).
Proof. It is easy to prove that Zermelo’s theorem implies theorem T.
We shall prove that theorem T implies the axiom of choice.
Let A denote a family of disjoint non-empty sets and F’ the family
of all sets {X, {w}} where xe X « A. (If [X, {a}} e F and if the set X has
only one element, then in view of we X we have X = {x} and the set
{X, {v}} has only one element. If the set XY has more than one element,
then Y # {w} and the set {X,, {v}} has two elements). It follows from theo-
rem T that there exists a maximal subfamily F, of disjoint sets from
family #. Let Z denote the set of all x for which, with a certain X « A,
we have {xX : {x} e F,. We shall prove that the set Z has one and only
one element from each set of family A.
Suppose that the set Z contains two different elements, 7, and x,
from a certain set X, of family A. Since w,« Z and «,« Z, there exist
sets X, and XY, of family A such that {X,, {a,}}«F, and {X, {w,}} « F,,
and in view of the definition of family F’, we have a, « X,, 7, « X,. Thus
X,X, #0 and X,X, #0, whence, the sets of family A being disjoint,
it follows that X,= X, and X,= X,, and therefore {X,, {w,}|¢F, and
Cardinal and ordinal numbers 28
434 XVI. Zermelo’s theorem

{Xo, {vo}|} « F,, which is impossible iin view x, ~ #, and the fact that the
sets of family F', are disjoint.
Hence the set Z contains at most one element from each set of
family A.
Suppose that for a certain set X, of family A we have X,.4 = 0.
Since the sets of family A are non-empty, there exists an element % « Xy-
Let {X, {v}} denote any set of family F,. It follows from the definition
of the set Z that w« Z, and thus XZ ~0, which, in view of X,Z = 0,
proves that XY 4 X,, whence, the sets of family A being disjoint, it fol-
lows that YX, = 0. Accordingly the set {X,, {x }} belonging to family F
(since w),¢« X,¢ A), is disjoint with each set of family #,, contrary to
the assumption that F’, is the maximal subfamily of family / composed
of disjoint sets.
Thus the set Z has at least one element in common with each set
of family A.
The set A of non-empty disjoint sets being arbitrary, we have proved
that theorem T implies the axiom of choice, q. e. d.
It will be observed that the sets {X, {v}} in our proof could not
ie
be by the sets {X,x} since, for instance, in the case of
= {1}, {{1}} we should have F = (a, fe Hay, atl, and the
maximal subfamily of disjoint sets of Aes ow seen consist of one set
only, whence it would follow that the set Z has only one element, while
in fact it is the set fi, {1}} (only) that contains one element from each
set of family A.

5. Inference of the axiom of choice from the generalized continuum

hypothesis. In § 3, Chapter 5, the term generalized continuum hypothesis
has been given to the supposition that for any infinite set A there is no
set of power greater than the set A and less than the set U(A). With
the aid of cardinal numbers this hypothesis can be expressed as follows:
G. If m is a non-finite cardinal number, then there exists no cardinal
number n such that
(5.1) Abell yk
Now Cantor's hypothesis on alephs is the term given to the following
hypothesis:
C. For every ordinal number a we have the formula
Fone 9s
(5.2) ass Sat1-

A. Lindenbaum and A. Tarski have given without proof the following

THEOREM 1. Theorem G is equivalent to the logical product of theorem C
and the axiom of choice (Lindenbaum et Tarski [1], p. 314, Theorem 95).
 § 5. Generalized continuum hypothesis 435

In other words theorem G implies the axiom of choice and theorem C,

and if theorem C and the axiom of choice are true, then theorem G is
also true.
Proof (Sierpitski [36], p. 1-5). Suppose that theorem G is true
and let n denote an arbitrary non-finite cardinal number. Let m= n+;
we shall have
iG O22 ¢ ~
(5.3) Le ae 2 ee Ni

But if p>s,, then 2° < 2°+p < 2°+2° — 2°t+1— 2°, whence 2?+
Se ae
In view of (5.3) we thus have
1 nt
(5.4) eet =O 2g Oe Oe OF ge

In virtue of Theorem 1 of IX.5 and of (5.3) we are also able to prove

without the aid of the axiom of choice that

(5.5) Qe == 2", 22 OM be
it nt
Pe K . ; ogm .
———
‘ gm
>
2
a aw
o2 —
22 .

In XVI.1 we have proved without resorting to the axiom of choice

that for the cardinal number m, which (as we know) is not finite, there
exists an aleph s(m) such that
mt
(5.6) s (m1) < 22 and s(m) is not <m
(see Corollary to Theorem 2 of XVI.1).
By (5.5) and (5.4) we obtain

(5.7)
~ SCD 24 eee oe
m 2"
If'we had s(m)+2? = 2° , then, by (5.5), we should have s(m)=
mt . . . as *
= 2" +m>n and the cardinal number n (which is not finite) would
be an aleph.
; mt
Theorefore suppose that s(m)+
22" = 2 does not hold. By (5.7)
we thus have
; Z POA iC oe
2? wi RG)
Se 22 a2
: ; ott
whence, by theorem G applied to number 27, we have

ss(tt) + 22" — 22"

whence-s(m) < 2?" and, by (5.4),

(5.8) s(t) 4-2" < 2m 22")

If we had s(m)+2™ = 22", then, by (5.5), we should have $$)(Ctt)i=—
m
= 2? >+>m2>n and the cardinal number n would be an aleph.
ppGoSO XVI. Zermelo’s theorem

: F m : Pee gS) , 5
Therefore suppose that s(m)+2™ 4~ 2?°, whence, by (5.8), we obtain

2" < (mm) +2" <2",

whence, in virtue of theorem G applied to number 2", it follows that we
must have

whence x(m) < 2™ and, by (5.5),

(3,9) s(t) =m < 2%--m = 2™.

If we had s(m)+m= 2", then, by (5.5), we should have s(m)=

= 2"™>m>n and the cardinal number n would be an aleph.
Therefore suppose that s(m)+m 4~ 2™, whence, by (5.9),

m<s(m)+m < 2",

which, in view of theorem G, gives s(m)+m= m, and thus s(m) < 1m,
contradicting (5.6).
We have proved that the cardinal number n must be an aleph, and
since n is an arbitrary non-finiie cardinal number, this, as we know
from XVI.1, imphes the axiom of choice.
We have proved that theorem G implies the axiom of choice.
Now we shall prove that theorem G implies theorem C.
Let a be any ordinal number. We have 2° = 9. (see XIII) -and
Na <Nai1, and we know that that there is no cardinal number between x,
and 8,41. AS we know, theorem G implies the axiom of choice, and thus,
in virtue of Theorem 6 of XVI.1, it implies trichotomy. Therefore either
QO 81g OF 2° > key. Th we had 2% <oxecy,, then in view Of 02“2-+5,;
the cardinal number 2™¢ would be intermediate between s, and +1,
which we know to be impossible. Thus 2°*>x,41. If 2%>,11, then
we should have xg < %ai1 < 2, which contradicts theorem G (for m=
—N8,, = Na41). Therefore we must have 2°*=s,13, 7. e. formula (Ono
Thus we have proved that G+C.
Suppose now that theorem C and the axiom of choice are true and
let m be an arbitrary non-finite cardinal number. From the axiom of
choice it follows (as we know from XVI.1) that m is a certain aleph;
thus there exists an ordinal number a such that m—=s,. Now, theorem C
implies formula (5.2). Thus if there existed a cardinal number n for which
inequalities (5.1) would hold, we should have s¢ < 1 < S41, which we
know to be impossible. Thus there is no cardinal number 1 satisfying
inequalities (5.1). Therefore theorem G is true. We have proved that
the validity of theorem © and the axiom of choice implies the validity
of theorem G.
 § 5. Generalized continuum hypothesis 437

Theorem 1 is thus proved.

For a given cardinal number m let us denote by G(m) the proposi-
tion that there exists no cardinal number n satisfying inequalities (5.1).
A. Lindenbaum and A. Tarski have formulated without proof the pro-
position that from theorems G(m), G(2™) and @(22") it follows that cardinal
numbers m,2™ and 22" are alephs (Lindenbaum et Tarski [1], p. 314,
Theorem 89).
We shall prove (without the aid of the axiom of choice) that from
theorems G(m), G(2™) and G(22") it follows that the cardinal number m is
an aleph.
For this purpose we shall prove (without resorting to the axiom of
choice) that for m > 1 7 follows from theorem G(m) that m > sy.
Let m be a cardinal number >1. [If m > 4, we can set m=n+1
where n2>3, whence 2™= 2%1— 2°.22n-2=nine>n+3=m-+2,
therefore 2" > m-+2, which is also true for m= 2 and for m= 3. We
have m+2 >m-+1. From Theorem 1 of III.8 it follows that we are
able to prove without the aid of the axiom of choice that if p is a cardinal
number such that p+1=p, then p>x,. Thus if m+2=m+1, then
we should have m+12>8,, whence m>x,. If m+2>m-+1, then, in
view of 2" > m+2, we should have 2% >m-+1>m and in view of the
validity of theorem G(m) we could not have m+ 1 > m. Thus we should
have m+1= m, which, as we know, gives m>s,. We have proved
that for m > 1 theorem G(m) implies that m > xp. |
By analyzing the first part of the proof of Theorem 1 we find at
once that we have proved without the aid of the axiom of choice that
for m > x, theorem G(m), G(2™) and G@(22") imply that the cardinal num-
ber m is an aleph. Thus for m > 1 it follows from theorems G(m), G(2™)
and G(2?") that m > s, and that the cardinal number m is an aleph, q. e. d.
Now we shall prove with the aid of the axiom of choice

THEOREM 2. Proposition OC, that (for a given ordinal number a)

2*¢*— s,41 ts equivalent to the proposition T, that every set of power 2°* is
the sum of sets of power s, forming a chain.
Proof. 1. Suppose that for a given ordinal number a proposition C,
is true. Thus 2°* = x,41 and a set of power 2“ is well-ordered and of type
+1. Therefore it is the sum of all its segments of type >a,, t.e. of
power s,, those segments of course forming a chain. Thus we have C,-—T,.
2. Suppose that proposition T, is true. Let A denote a set of power 2"¢.
In view of 2“* > x, it follows from the axiom of choice that 2°* > s,41(2°* <
<Ngii being impossible since no cardinal number lies between s, and
Na41). Thus there exists a subset A, of power ,41 of the set A. From
proposition T, it follows that the set A is the sum of certain subsets of
438 XVI. Zermelo’s theorem

power x, forming a chain; let # denote the family of those sets. For each
element a of the set A there exists a set B(a) « F such that a« B(a). Let

= » Bla)
aéAy

We have of course A, C S, whence S> A, =N,+1- On the other hand,

the set S, as the sum of Sat+1 sets of power x, is Of power <Sy41Na= Nati)
nes =a Sai1- Therefore s= Nai1- Now let a, denote an arbitrary element
of the set A. Since the set B(a,), as belonging to family F, is of power x,
and the set A, is of power 8,41, there exists an element a, of the set A,
such that a,¢ B(a,). But a,« B(a,), and since B(a,)«F, B(a,)eF and
family F is a chain, we must have either B(a,) C B(a)) or B(a) C B(a,).
But, in view of a,« B(a,)—B(a,), B(a,)C B(a,) cannot hold; thus
B(a,)C B(a,), and in view of a,e A, we have a,« B(a,)C B(a,)CS.
Therefore we have a,<« S for a,« A, whence it follows that ACS, and,
in view of SC A, we have A = S, waa 2% — 4 — § = x,41. We have
proved that T, > O,.
Theorem 2 is thus proved.
In the particular case of a = 0, we obtain from theorem2 the following
CoROLLARY. The continuum hypothesis is equivalent to the proposi-
tion that the set of all real numbers is the sum of denumerable sets forming
a chain (Sierpinski [10], p. 112, and [12], p. 180.) |
For m>1 theorem G(m) imples that m is not a finite number,
and the axiom of choice implies that there exists an ordinal number a
such that m = x,. In view of the axiom of choice we then have G(m)-->C,,
and it easily follows from Theorem 2 that we are able to prove with the
aid of the axiom of choice
THEOREM 3. For every cardinal number m > 1 theorem G(m) is equi-
valent to the proposition that every set of power 2™ is the sum of sets of power m
forming a chain.
For a cardinal number m let us denote by 7 (m) the proposition that
if FE is a set such that H = 2", then the set EXE is the sum of two sets,
H x H= A-+B, such that the set A is at most of power m on every parallel
to the axis of abscissae and the set B is at most of power m on every
parallel to the axis of ordinates. With the aid of the axiom of choice it
can be proved that the following theorem holds:
For every non- tee cardinal number m theorems G(m) and T(m) are
equivalent.
For m= x, this gives Theorem 1 of XV.3.
It is easy to prove that the theorems T (1) and T (2) are true and
that theorems T(m) are false for natural m > 3.
 § 5. Generalized continuum hypothesis 439

We shall also prove

THEOREM 4. It follows from hypothesis G that for any cardinal num-
bers m and n the equalities m= n and 2™ = 2" are equivalent.
Proof. It suffices of course to show that formula 2™— 2" implies
m=n. In the case where one at least of the numbers m and n is
finite this is obvious. Suppose therefore that neither of the numbers m
and n is finite, and that theorem G is true. In view of Theorem 1 also
the axiom of choice is true; as we know, it implies trichotomy. Thus if
we had 2" = 2" and m 4 n, then either m< nor m>n,eéeg. m<n. But
then, in view of 2™= 2", we should have m<n < 2"™= 2", which contra-
dicts theorem G.
Theorem 3 is thus proved.
Suppose now that a cardinal number m is not finite. If n is a cardinal
number >m, then, by formula (2.1) we have mn =n, whence 2" <m" <
—o2. = 2, theretore mt" — 2".
If n is a cardinal number such that 0<n<m, then, by (2.1), we
have mn=m, whence m <m" < (2™)"= 2™— 2™ and it follows from
hypothesis G that either m*= m or m" = 2™. Therefore:
It follows from hypothesis G that apart from numbers of the form 2%
no non-finite cardinal number p is of the form p = m" where m <p.
 CHAPTER. XVII

APPLICATIONS OF ZERMELO’S THEOREM

1. Hamel’s basis. There are a great many applications of Zermelo’s

theorem in various branches of mathematics. Here we shall present some
of them by way of example.
Zermelo’s theorem implies the existence of a well-ordered set

(UL) Lg ce gilng eet thes Aomecae aa eee

formed of all different real numbers +0.

Now let us form a set B of real numbers, assigning to it the number #p
and every term a, of the sequence (1.1) for which there exists no finite
sequence a, d,..., d, of ordinal numbers <a and no sequence of rational
numbers 7,, 72, ---;%n, Such that

(1.2) La= 11 Ve, + 12boq + ++» + Tn Vay

From the definition of the set B it follows that for no finite sequence
b,, by, ..., 5, of different numbers belonging to B the formula r,b,+
+7.b,+...+7nbn = 0, where n is a natural number and 7,, 72, ..., 7, are
rational numbers ~0, is true. Indeed, if we assumed that this formula
holds and that the numbers b; (¢= 1, 2,...,) are written in the same
order as they appear in the sequence (1.1), we should have
Dn = 8,0, + 89bg+...+ Sp_10n-1 ,

where s;= —7;/r7, for +=1,2,...,2—1, which is impossible since b,

belongs to B and the numbers J,, by, ..., b,-1 precede b, in the sequence
(1.1), the coefficients s; («= 1,2,...,%—1) being rational and <0.
Hence it immediately follows that every real number xv 4 u can be
represented at most in one manner in the form
(1.3) = 17,6, +72b,+...+%m0m,
where m is a natural number, 5; («= 1,2,...,m) are different numbers
of the set B and 7; ((=1,2,...,m) are rational numbers 40 (if we do
not consider as different two sums (1.3) differing only in the order of
their components).
 § 1. Hamel’s basis 44]

Now we shall prove that every real number 40 can be represented

in. the form (1.3).
Suppose that this is not true and let x, denote the first number of
sequence (1.1) that is different from 0 and cannot be represented in the
form (1.3). Number x, cannot belong to B since each number Db of the
set B can be represented in the form b = 1-3), 7. e. in the form (1.3) (for
m=1,7r,= 1). On the other hand, from the definition of the set B it
follows that if number #, does not belong to B, then there exists a finite
sequence a,, a, ..., 4, Of ordinal numbers <a and a sequence of rational
numbers 7,,72,.--, 7, for which formula (1.2) holds. But from the defini-
tion of number «, it follows that each of the numbers 7%, (¢= 1, 2,..., ”),
as preceding number x, in sequence (1.1), can be represented in the form
(1.3); thus. for 7 = 1, 2,...,” we have

(1.4) Gi = Oy ETS Dy ho Tm
where m (2=1,2,...,”) are natural numbers, pi? tea RA a
1=1,2,...,) are numbers of the set B and rp Cee Le roaeg HIS 0 a=
=1,2,...,”) are rational numbers 40. Introducing formulae (1.4) into
formula (1.2) we obtain for number #,, by reduction, an expression of
the form (1.3), which contradicts the assumption that x», cannot be re-
presented in that form.
Thus we have proved (with the aid of the axiom of choice) the
existence of such a set B of real numbers that every real number x
different from 0 can be represented in one manner only in the form (1.3)
where 0; («=1, 2,...,m) are different numbers of the set B and 7;
(¢=1,2,...,m) are rational numbers 40. A set B with such proper-
ties is called the basis of Hamel (of the set of all real numbers) (Hamel
[1], p. 459-462). :
So far we do not know any effective example of Hamel’s basis, and
neither can we prove its existence without the aid of the axiom of choice,
Now let b, denote any given number belonging to the basis B. Let
us define a function f(#) of a real variable, setting f(#)= 0 for r=0
and for every real number x for which in the expansion (1.3) number b,
does not appear (7. e. for which in formula (1.3) we have b; ~ b, for +=
=1,2,...,m). If in the expansion (1.3) number b, does appear with
the (rational) coefficient r, then we shall set f(x) =r. It is easy to verify
that the function f(x) of a real variable defined in this way satisfies the
functional equation

(1.5) f(at+y) = f(a) +F(y)

for any real numbers «x and y.
442 XVII. Applications of Zermelo’s theorem

However, our function f(x) is not of the form Aw+B where A and B
are real numbers independent of x, since /(b,) = 1 and f(b) = 0 for every
number b ¢ b, belonging to the basis B, there being obviously infinitely
many such numbers b (since the set of all real numbers is non-denumer-
able while there is a denumerable aggregate of numbers of the form (1.3)
where 0,, b,,..-, bm belong to a finite sequence of numbers). Now, the
function Av+B, provided it is not constantly 0, has at most one root.
The existence of Hamel’s basis implies thus the existence of non-
linear solutions of the functional equation (1.5). However, we do not know
any effective example of a non-linear function satisfying the functional
equation (1.5), and neither can we prove the existence of such a function
without the aid of the axiom of choice.
It can be proved without the aid of the axiom of choice that every
non-linear function satisfying the functional equation (1.5) is disconti-
nuous (which was already proved by Cauchy), and even non-measurable
in the sense of Lebesgue *).
Numerous works have been devoted to functions satisfying equa-
tion (1.5) (Ostrowski [1]; Maceaferri [1], p. 92-101; Alexiewiez and Or-
licz [1], p. 314; Haeperin [1], p. 221-224).
However, we can easily give examples of discontinuous functions
f(x) of a real variable satisfying the inequality

f(a+y) > f(a) +f(y)

for any real x and y; such is for instance the function Ex (Cooper [1],
p. 430).
From the fact that the functional equation (1.5) has discontinuous
solutions it easily follows that there exist discontinuous functions g(a)
satisfying the functional equation

g(ay) = g(@)gly)
for any real « and y; in order to obtain such a function it suffices to set
g(0) = 0 and, for real x 4 0, g(x) = ef(slr),
But it can easily be proved that, besides the function f(a) constantly
equal to 0 and the function f(a) = a, there is no real function of a real
variable satisfying the functional equation (1.5) and the functional
equation

(1.6) f(xy) = f(a) f(y)

*) It was proved with the aid of the axiom of choice by M. Fréchet [1] (p. 390-392).
I proved it without the aid of the axiom of choice in [6] (p. 116); see also Banach [1],
p. 123, and Kae [1], p. 170.
 § 1. Hamel’s basis 443

for any real x and y (see below, Exercise 1). However, it can be proved
with the aid of Zermelo’s theorem that there exist complex functions
f(z) of a complex variable, not constantly equal to 0 and different from
the functions f(z) = 2 and f(z) = 2’ (where 2’ denotes a complex number
conjugate with number z), which satisfy the functional equations

(1.7) (4 +22) =f(%i)+F(%) and = f (222) = f(%)f (2)

for any complex numbers 2, and 2, (Lebesgue [2], p. 532, and [7], p. 296-
300; Schoenflies [1], p. 181-184).
It has also been proved that there exist non-trivial functions satisfy-
ing (for any complex z, and 2,) the functional equations (1.7) and as-
suming every complex value once and only once (i. e. establishing the
automorphism of the set of all complex numbers with respect to addi-
tion and multiplication (Segre [1], p. 419, and Kestelman [1], p. 1).

EXERCISE. Prove without the aid of the axiom of choice that besides the func-
tion f(x) constantly equal to 0 and the function f(z) = » there is no real function of
a real variable satisfying the functional equations (1.5) and (1.6).
Proof. By (1.6), we have f(x?)= [f(x)]* for real «, whence f(t)> 0 for t> 0. By
(1.5), we have f(x) = f(~—y)+ f(y) for real « and y, therefore f(x) > f( +ffoe o> y.
If we had f(1) = 0, then, on ee of the fact that, by (1.6). f(x)= f(a)f(1), we should
have f(x) = 0 for real x. In the contrary case we have f(1) 4 0 and, by (1. zewe obtain
f(1) = 1 for x = y = 1, By (1.5) we easily obtain f(ma) = mf(x) for natural m, whence
for «= l1/m, in view of f(1) = 1,.we have f(1/m) = 1/m, and since jf(lx) = If(a) for
natural 1, we have
1 1 i
i(-.) Su) (5) mM

Thus, for positive rational r, we have f(r) = 7, and since, by (1.5), for « = y = 0,
we obtain {(0) = 0, whence for y = —~x, f(—«”) = —f(x), we conclude that the formula
f(r) =r is true for any rational r.
Now ne that there exists a real number w such that f(x . Thus either
iD) SS 8 Oe (Ca) <a Reex) x «x, then there exists a rational ea r a that f(x) >
SPS oO weave f(r) > f(a); therefore, in view of f(r)= 7, r > f(x”), which is impos-
sible. If f(x) < z, Ateae ci a rational number 7 such that f(x) < r < x, whence
f(r) <f (x); therefore r < f(x), which is impossible.
Thus we must have f(”%) =« for every real number x and the ae es, proof is
complete.

With the aid of Hamel’s basis it is easy to prove that there exists
a set of points P on a straight line such that the family of all the different
sets lying on that straight line and congruent to the set P is denumer-
able +).

1) Without the aid of the axiom of choice we are unable to prove the existence of
such a set.
444 XVII. Applications of Zermelo’s theorem

Indeed, let B denote Hamel’s basis and b any given element of it.
The set P, formed of number 0 and all real numbers x for which we have
b; £6 for i=1,2,...,m in the expansion (1.3), obviously has the re-
quired property. If we denote by P(a) the translation of the set P along
the straight line for a distance of a, then it is easy to prove that for ra-
tional 7; and 7, where 7, #7, the sets P(r,b) and P(r.b) are disjoint
and that for every real number a there exists a rational number 7 such
that P(a) = P(rb) (Sierpinski [44], p. 161).
It will be observed, however, that I. Kapuano asserts that there
exists no plane set Q such that the family of all different plane sets
congruent with QY (by translation or by rotation) is denumerable.
With the aid of Hamel’s basis it is also easy to prove E. Cech’s
proposition that there exists a non-empty set of real numbers A different
from the set of all real numbers and such that for every real number x
in the infinite sequence A, A(x), A(2ax), A(3a@),... there is only a finite
number of different sets (W. Sierpinski, Fund. Math. 35(1948), p. 160).
Indeed, let A denote a set formed of number 0 and of all real num-
bers x for which in the expansion (1.3) numbers 7; (2 = 1, 2,..., m) are
all integers. The set A is not the set of all real numbers because, for in-
stance, if b is an element of Hamel’s basis, then the number b/2 does not
belong to A. Now let x denote any real number, n — the common de-
nominator of the rational numbers 7,, 72, ...,7%m in the expansion (1.3) of
number v. Clearly nxe A for n=1,2,... and A(nw) = A. Hence it im-
mediately follows that each term of the infinite sequence of sets A, A(z),
A(2x), A(3a), ... is equal to one of the sets A, A(x), A(2z), ..., A((n—1)a).
Gech’s theorem is thus proved.
In connection with translations of linear sets let us also observe
that the following proposition is not trivial: for every non-empty set A
of real numbers that is not the set of all real numbers there exist infinitely
many different sets of real numbers congruent with the set A by trans-
lation (W. Sierpinski, Fund. Math. 35(1948), p. 159, Th. 1).
We say that a given set X of real numbers is formed of rationally
independent numbers if for every finite sequence «,, 2, ...,%m of dif-
ferent numbers of the set X and for every sequence 7,, 72, ...,%m Of ra-
tional numbers the equality 7,7, +7,0,+...+7m@%m—= 0 implies the equa-
it Nea Op fOnei ea ed a. 00.
We say that a real number z is rationally independent of the numbers
of the set X if for every finite sequence 7,, %,..., %m of different numbers
of the set X and for every sequence 7,, 72, ..., 7m of rational numbers 40
we have the formula + #7,0,+7,0,+...+1m@m-
It can easily be seen that every basis of Hamel is formed of rationally
independent numbers. It is also easy to prove that if A is a set of rationally
 § 1. Hamel’s basis 445

independent numbers, then there exists a basis of Hamel B such that

AC B. Indeed, let B denote the set formed of all numbers of the set A
and of all terms x, of sequence (1.1) such that number «x, is independent
of the numbers of the set A+ {X¢}-<,. It is easy to prove that the set B
is a basis of Hamel.
From the continuum hypothesis it follows that the set of all real num-
bers 40 is the sum of a denumerable aggregate of sets each of which is a basis
of Hamel.
Indeed, let us suppose that the continuum hypothesis is true. Ob-
viously every basis of Hamel is a non-denumerable set since there is
a denumerable aggregate of numbers of the form (1.3), where ),, bo, ..., Dm
is a finite set of numbers of any given denumerable set and numbers
W145 725 -++57%m are rational. Thus it follows from the continuum hypothesis
that every basis of Hamel is a set of power s,. Let B denote a given basis
of Hamel. There exists a transfinite sequence of type 2, {be}s<e, formed
of all different numbers of basis B. For every ordinal number a < 2
and every rational number r 4 0 let us denote by A? the set of all num-
bers of the form 17,b;,+7,0;,+...+1,-10;,_,+7ba, where is a natural
cP Ctra teen = hy oa cd ANI 9, 7, eo te ee raulOonal NUInDers.
In view of a< the sets A’ are obviously at most denumerable. Thus for
every ordinal number a < 2 and every rational number r * 0 there exists
an infinite sequence af”, af”, af®,... formed of all (not necessarily dif-
ferent) elements of the set A’. For natural. and rational 7 = 0 let

Q(r,n)
= DY)fay.
(r,a)
a<Q

Obviously each of the sets Y(7, 2) consists of rationally independent

numbers, since for every number a <2 there exists in the set Q(r, n)
only one number from the set A’. Thus the numbers forming each finite
subset of the set Q(r,) can be arranged in a sequence a’, a?®,...,
av™ where a, <a) <...< dm, and the component (rb,,,) containing b,,,
appears only in the expansion (of the form 7, bg, +7202, +... + Tn-19¢,_, +71ba)
of number a”, whence we conclude that the numbers of the set in ques-
tion are rationally independent.
Since the set Y(7, n) is made up of rationally independent numbers,
there exists for it, as we know, a basis of Hamel, B(7,n), such that
OO”, nn) EG Bir, n). The set

s= » dB,n), hoard
446 XVII. Applications of Zermelo’s theorem

. 2f . .
where the summation > extends over all rational numbers r different
r

from 0, is of course the sum of a denumerable aggregate of Hamel’s bases.

On the other hand,

and the set

NOONae
pila,
PED At
<a a

is obviously the set of all real numbers different from 0, whence we easily
conclude that also S is the set of all real numbers ~0. Thus the latter
set is the sum of a denumerable aggregate of Hamel’s bases, q. e. d.
It can be proved that also conversely: from the assumption that
the set of all real numbers 40 is the sum of a denumerable aggregate
of Hamel’s bases follows the continuum hypothesis. Therefore the conti-
nuum hypothesis is equivalent to the proposition that the set of all real num-
bers different from 0 is the sum of a denumerable aggregate of Hamel’s bases
(Erd6s and Kakutani [1], p. 459, Theorem 2).

2. Plane set having exactly two points in common with every straight
line. We shall prove
THEOREM 1 (of S. Mazurkiewicz). There exists a set of points in
a plane that has two and only two points in common with every straight
line lying in the plane.
Proof (Mazurkiewiez [1], p. 382-383). Since the set of all points
in a plane is of the same power as the continuum and the set of all straight
lines lying in a plane is of the same power as the continuum (in virtue
of Theorem 2 of IV.8 and Exercise 2 of II.6) and from Zermelo’s theorem
follows the existence of the least ordinal number @ of the power of the
continuum, there exists a transfinite sequence of type @, {pels<,, formed
of all different points in a plane and a transfinite sequence of type
@, (Sshe<p, formed of all different straight lines lying in a plane.
_ Now let us define by transfinite induction a transfinite sequence of
type @, {dehe<y Of points in a plane as follows.
Let us denote by q, the first point of the sequence {pe}s<, lying on
the straight line S,. Now let a denote an ordinal number such that 1 <
<a<g and suppose that we have already defined all points gq: where
E<a. Let us denote by Q, the set of all points gq: where & < a; in view
of a <q we have Y, < 2”, therefore also the set Q.XQ, is of power <2,
and so is the set 7, of all straight lines having at least two points in com-
mon with the set, Q,. Since the number of all straight lines lying in
a plane is 2", there exist lines that do not belong to the set 7',; thus there
 § 2, Plane set 447

exists the first term S, of the sequence {S¢}:<, which is such a line. Let &
denote the set of all intersection points of the line S,, with the lines of
the set 7,; since each of those lines intersects the line Sz at one point
only and the set 7, is of less power than the continuum, the set Rk will
also be of less power ‘than the continuum, whence it follows that the
linee Ss, contains points belonging neither to Q, nor to R (since Oe Rie
ZO. +R < 2°, for it follows from the axiom of choice that the sum of
iv sets of power less than the continuum is of power less than the conti-
nuum). Let us denote by q, the first term of the sequence {pets<,, Which
is exactly such a point.
In this way the points q (a <q) have been defined by transfinite
induction. Let us denote their set by Q. We shall prove that the set Q
has two and only two points in common with every straight line lying
in a plane. Suppose that a certain straight line S contains more than
two of the set Q; let qa,, 4a, and q,, be those three of them which have
the least indices; thus a, < a,< a3, therefore q,¢Q., and q.,¢€Q,, and
thus S« T,,, while from the definition of point q,, it follows that it does
not lie on any straight line belonging to 7,,, which gives a contradiction.
Thus every straight line lying in a plane contains at most two points
of the set QY. Now suppose that there exist straight lines containing less
than two points of the set Q and let S, denote the first term of the se-
quence {Sz}s<g with this property. ;
From the definition of the lines S, it easily follows that fa, < Pa,
for a, <a,<g. But q,¢ Sg,, and, as we already know, no straight line
contains more than two points of the set @. Hence it follows that 6,, =
= Ba, = Pa, 18 impossible for a, < a, < a, <q, whence we conclude that
Ba < Pate for a <<, whence it easily follows that there exists an ordinal
number a<q such that p, > y. Since S, is the first term of the se-
quence {Sz}s<, that is a straight line not belonging to 7, in view of
y <P, we must have S,« T,, whence it follows that the line 8, has at
least two points in common with the set Q, and thus in view of Q, CQ
with the set Q, which contradicts the definition of the line S,. We have
proved that every straight line lying in a plane contains two points of
the set Q.
Theorem 1 is thus proved. We are unable to prove it without the
aid of the axiom of choice. Neither can we define effectively a plane set
which has at least one and at most two points in common with every
straight line lying in a plane. But it is easy to give an example of a plane
set which every straight line intersects at a denumerable aggregate of
points: such a set is for instance the sum of the circumferences of all
circles «2+ y? = n? where n = 1, 2,...
448 XVII. Applications of Zermelo’s theorem

The theorem of Mazurkiewicz has been generalized by IF. Bage-

mihl [1] (p. 34). We can prove (with the aid of Zermelo’s theorem) the
following theorem:
Suppose that with every straight line S lying im a plane a certain cardinal
number Ws such that 2 <img <2" is associated. Then there exists a plane
set Q such that QS = msg for every straight line S lying in a plane ?).
3. Decomposition of an infinite set of power 1 into more than m almost
disjoint sets. Two infinite sets are said to be almost disjoint if the set of
their common elements is of less power than either of those sets. We
shall prove with the aid of the axiom of choice
THEOREM 1. Every infinite set of power m is the sum of more than ™
almost disjoint sets each of which is of power m (Sierpinski [70], p. 15, [26],
p. 118; Tarski, Fund. Math. 12(1928), p. 188-205, and 14(1929), p. 205-215).
Proof. Let M denote an infinite set of power m and let P= Mx M.
Since it follows from the axiom of choice that m= m (see Theorem 6
of XVI.2), we have P = M. Thus in order to prove our theorem it suf-
fices to show that the set P is the sum of more than m almost disjoint
sets of power m.
Let the term curve y = f(x) be given to the set of all elements (x, y)
of the set P such that y = f(x) where f(x) is a given function, defined
for «« M, whose values belong to M. It suffices of course to prove that
the set P is the sum of more than m almost disjoint curves.
LemMaA. If F is a family of power <m of curves, then there exists
a curve almost disjoint with every curve of the family F.
Proof of the lemma. From Zermelo’s theorem it follows that
there exists an ordinal number a such that m = x, and, in view of M= m,
there exists a transfinite sequence of type @a, {Ws}e<w,, formed of all
different elements of the set M, and in view of F < mthere exists a trans-
finite sequence of type @,, {Cz}s<w, formed of all curves of the family F
that can be repeated (if # < m). For € < a, let y = f(a) denote a curve Cs.
Now we shall define a curve y=.g(#) by transfinite induction as
follows, Let g(x,) = 7,. Now let A denote a given ordinal number, 1 <
<A<@, and suppose that we have already defined all values g(a-)
for <A. Let Q, denote the set of all elements f,(v,) for <4. Since
hh ids Ari ean daa m, and thus in the sequence {Le}e<w, there exist
elements that do not belong to Q,. Let xe, denote the first of those elements:
let g(a) = we,. In this manner the funetion g(x) has been defined by

1) W. Sierpinski, Une généralisation des théorémes de 8S. Mazurkiewicz etF.Bage-

mihl, Fund. Math. 40(1953), p. 1-2.
 § 3. Decomposition of an infinite set 449

transfinite induction for all terms of the sequence {wg}s<o,, 7. e. for all
elements of the set M, and
(3.1) g(a) Afeas) for E<A<ay.
Now let y = f(x) be any curve belonging to family F. Thus it is one
of the terms of the sequence {C;}:<, and therefore there exists an ordinal
number uw < w, such that f(a”) = f,(x) for 7 « M. By (3.1) we have g(a) ¢
AF ful.) for w<A<o,; thus the equality g(a) = f,(a,) may hold at
most only for ordinal numbers 4 < yw whose set is of power <p<m
(since u < @,). Elements (x, y) of the set P such that y = g(#) and y =
= f,(v) form a set of power <m, which proves that the curves C, and
y = g(x) are almost disjoint. The curve y = g(a) is almost disjoint with
every curve of family #. Our lemma is thus proved.
Let us return to the proof of Theorem 1. We shall define a trans-
finite sequence of type +1 of curves {Ke}ecw,,, a8 follows. Let fe(v) = xe
for €< a, ce M. The curves Ky = f,(x)], where ¢< @,, are of course
disjoint and their family is of power m. Now let 4 denote an ordinal num-
ber such that w, <A < mg41 and suppose that we have already defined
all curves K,; for & <A; their set F, is of power A4=s,=™ since a, <
<A < @g41. Therefore we can apply our lemma, which gives a curve K,;,
almost disjoint with every curve Ke for §<A. The curves K; where
€ < Wg+1 are thus defined by transfinite induction and it can easily be
seen that each two of them are almost disjoint; their sum (even the sum
») Kz) gives the set P. Theorem 1 is thus proved.
$<
It will be observed that the proof of Theorem 1 for m= xs, may
be carried out without the use of the axiom of choice, as has been done
in Theorem 18 of IV.14. .
Given two denumerable sets, A and B, we say that the set B is
almost contained in the set A if the set B—A is finite. The sets A and B
are said to be essentially different if the set (A—B)+(B—A) is infinite.
With the aid of the axiom of choice we can prove the following
theorem:
There exists a transfinite sequence {Ngsecq of type 2 of infinite sets
of natural numbers such that, for a<B <Q, the set Ng is almost: con-
tained in the set N, and essentially different from N, (Sierpinski [30], p. 9).
With the aid of the continuum hypothesis we can prove the follow-
ing theorem:
There exists a transfinite sequence {Nelsco of type 2 of infinite sets
of natural numbers such that the sets N,—N: are finite for <7 <2 and
that there exists no infinite set of natural numbers A such that the sets A—N;
are finite for € <Q (Sierpinski [46], p. 148, [62], p. 99).
Cardinal and ordinal numbers 99
450 XVII. Applications of Zermelo’s theorem

Finally it will be observed that F. Rothberger has proved the fol-

lowing theorem:
The continuum hypothesis is equivalent to the proposition T that there
exists a transfinite sequence {Nelseo of type 2 of infinite sets of natural
numbers such that for every infinite set A of natural numbers there exists
an index a <Q such that N,—A <s, (Rothberger [1], p. 33).
It is obvious that the continuum hypothesis implies proposition T
since the set of all infinite sets of natural numbers is of power 2”, 7. ¢., in
virtue of the continuum hypothesis, NV is of power s,, and therefore there
exists a transfinite sequence of type 2, {Nz}:-o formed of all infinite
sets of natural numbers.
Now suppose that proposition T is true and let {N<}:-,g denote a trans-
finite sequence of type 2 of infinite sets of natural numbers satisfying
theorem T. In virtue of Theorem 1 of IV.14 the set N of all sets of natural
numbers is the sum of 2° almost disjoint infinite sets. Let ® denote the
family of those sets; therefore ® = 2“. Let A and B be any two different
sets belonging to family ©; they are almost disjoint, whence AB < 2h
In virtue of proposition ft hers exist ordinal numbers ay < 2 and ag < &
such that N,Need So and iWones ap < No. If we had a,= ag, then N,,
= N,, and, in view of the Formula INCN A) tN age BDepee, he
set N,, would be contained in the sum of three finite sets, 7. e. it would
be finite, which is impossible. Therefore a, 4 ag for Ae ®, Be ®@, AA BR,
and since @ = 2° and a4 <Q for Ae ®, we conclude that > 2? Sate:
that s, > 2°°, which, as we know, gives s, = 2. Thus proposition T implies
the continuum hypothesis.
We have proved Rothberger’s theorem.
4. Some theorems on families of subsets of sets of given power. We
shall prove
THEOREM 1. For every infinite set M of power m there exists a family F
composed of 2™ subsets of the set M none of which is a subset of any other.
In other words if Ae F, Be F and A ~B, then A—B £0 (and of
course also B—A 40).
Proof. Let M denote an infinite set of power m. As has been proved
in XVI.2 (e. g. see (2.1)), from the axiom of choice follows the formula
m= 2m = m-+m, whence the set M is the sum of two disjoint sets, M—
= M,+M, each of which is of power m. Thus we have ,~M, and there
exists a function g (1-1) mapping the set M, on the set M,. With each
subset N of the set M,, let us associate the set y(NV) = N+[M,—q(N)];
it will of course be a subset of the set MW. Let F denote the family of all
sets p(N) where NC M,. If Ace Ff, Be F and A ~B, then there exist
 § 4. Some theorems on families of subsets 451

sets V,C M, and N,C M, such that A= y(N,), B= y(N,), therefore,

in view of A ~ B, we have N, ~ N,. In view of the definition of the
function y we have A = N,+[M,— 9(N,)], B= N,+[M.— ¢(N,.)]. If we
had ACB, then N,+[M,—9(N,)]|C N.+[M,— (N,)], whence, on ac-
count of the fact that V,M, = 0 (since N,C M, and M,M, = 0) and since
similarly V,M, = 0, we should obtain VN, C N, and M,—(N,) C M,—9q(N,),
whence y(N,)C y(N,), which gives N,C N, since @ is a (1-1) mapping
of the set M, upon the set M, while N, C M, and N,« M,. Thus we should
have N,C N, and N,C N,, whence NV, = N,, which is impossible. There-
fore A non CB, i.e. A—B £0. Henceit follows that if V,C M,, N,C M,
and N, ~ N., then y(N,) #y(N,). The family F of all sets y(N) where
N C M, is thus of the same power as the set of all different subsets
of the set M,, i. e. in view of M,=m, of power 2".
Family F satisfies the conditions of our theorem, which is thus proved.
We have deduced Theorem 1 from the formula 2m = m for non-
finite cardinal numbers m without the aid of the axiom of choice, but
the proof of that formula has been based on Zermelo’s theorem. So far
we do not know any proof of Theorem 1 that is not based on Zermelo’s
theorem. But with regard to those cardinal numbers m for which the
formula 2m = m can be proved without the aid of the axiom of choice
(e. gy. for m= Ny, 2°°, 22°) we can also prove Theorem 1 without resorting
to the axiom of choice.
Theorem 1 may be regarded as a particular case of a more general
theorem proved by F. Hausdorff ([3], 18-19, and Tarski [17], p. 61)
(as a generalization of a certain theorem of G. Fichtenholz and L. Kan-
torovitch '); it runs as follows:
For every infinite set M of power m there exists a family F', formed
of 2™ subsets of the set M, such that for every finite set A,, A.,..., Ap,
Ay, Az, ..., Aj of different subsets of this family we have
(4.1) Au Ag A (Ube
AL Meal) CM A) <2:00
A family of sets satisfying this condition (where M denotes the sum
of all sets of family F) is said to be a family of independent sets in the
sense of the general set theory (Marezewski [2], p. 14)
EXERCISE. Prove that if # is an infinite family of infinite sets of natural num-
bers, then there exists an infinite sequence A,, A,, ... of different sets of family F such
that
(4.2) TAT ASA me tetAm MetOLIAE Aine Dhue:

Proof. We shall distinguish two cases:

1) Proved by them only for m = ¥, and m = 2” (see Fichtenholz et Kantorovitch

{1], p. 80 and 82).
29+
452 XVII. Applications of Zermelo’s theorem

a) Family F is denumerable, therefore Ff = {A,, A,, ...}. Then we of course have

formula (4.2).
b) Family F is non-denumerable. Let P denote the product of all sets of family /
and let NV = {1, 2,...}. The set NW—P cannot be finite because in that case each set of
family F would contain all natural numbers except a finite number of them and the
set of all different sets of family # would have to be at most denumerable, which
is contrary to our assumption. Thus the set N—P is denumerable, therefore V—P =
== {N,, Np, ...} Where n, < n, <... For every natural number k we have n, ¢ P and from
the definition of the set P it follows that there exists a set A, of family F' such that
n, € A,. It can easily be seen that A,A,A,...c P, therefore, in view of the definition
of the set P, we have A,A,A,...c A for A «Ff.
Thus the proof is complete; we have made use of the axiom of choice in it because
in case b) we have not given the rule of determining the set A, for every natural num-
ber k. Without the aid of the axiom of choice we are unable to carry out the proof in
question.

A family F of sets such that if X « fFandY « X then YC F is called

hereditary. From Theorem 1 we shall now deduce the following
COROLLARY 1. The total number of different hereditary families which
can be composed from the subsets of a given infinite set M of power m is 2?"
(A. Tarski, Sur les classes densembles closes par rapport a certaines
opérations élémentaires, Fund. Math. 16(1930), p. 231, Theorem II).
Proof. Let M denote an infinite set of power m, and F — a family
of its subsets satisfying Theorem 1. Family F, being of power 2™, has 22”
different subfamilies. For every subfamily G@ of family F (7. e. for every
family G@ of sets such that @CF) let

G)= S'U(A
AeG

(where U(A) Bae the family of all subsets of the set A). It is easy to
verify that f(G@) is (for GCF) a hereditary family formed of subsets of
the set M.
Now suppose that G,CF, G,CF, G,~G,. We shall prove that
f(G,) #7 (G,). Since G, #~G,, we have either G,—G, 40 or G,—G, 4 0.
Suppose for instance that G,—G, #40; then there exists a set A, such
that A, « G,—G,. In view of A, « G, we have A, « f(@). If we had A,« f(G
there would exist a set A « G, such that A,« U(A), t.e. A,yCA, and in
view of Aye G,—G,, AeG we have A, #4A. The sets A, and A, both
belonging to family F (since G,C F and G@, CF), would thus be two dif-
ferent subsets of the set M, the first of them being a subset of the second,
which er sa the property of family F. Therefore A, ¢f(@,) and
thus A, «/f(G,)—f(G,). We have proved that if G,CF, G,C F- and
G,—G, 4 0, hr f(G,)—f(G,) #0. Hence it follows that to different
subfamilies G of family # correspond different hereditary families f(G@)
made up of subsets of the set M. Since there are 2?" different subfamilies
 § 4. Some theorems on families of subsets 453

of family #, the number of different hereditary families formed of sub-

sets of the set M is >2?". But the set of all families composed of sub-
sets of the set M is the set UU(M), i. e. it is of power 22", whence it fol-
lows that the total number of different hereditary families formed of
subsets of the set M is <2". In view of the result obtained previously
we conclude that their number is exactly 2°". Corollary 1 is thus proved.
EXERCISE. Give an example of a hereditary family of sets which does not
constitute the family of all subsets of any given set.
Answer. The family of all finite subsets of an infinite set; the family of all almost
denumerable subsets of a non-denumerable set; the family composed of the empty set
and the sets {1} and {2}.

A family of sets such that the product of an arbitrary aggregate of

sets of that family also belongs to it is termed absolutely multiplicative. Every
hereditary family of sets is of course absolutely multiplicative. The family
of complements, to a given set M, of sets of a hereditary family contained
in M is (as follows from the generalized theorem of De Morgan) absolutely
additive, 7. e. the sum of an arbitrary aggregate of sets of that family
also belongs to it. Thus from Corollary 1 follows
COROLLARY 2. The total number of different absolutely additive (ab-
solutely multiplicative) families which can be formed of the subsets of a given
infinite set of power m is 2™
However, it will be observed that, as has been proved by A. Tar-
ski (l. ¢., p. 242, Theorem VI), the total number of different families at
the same time absolutely additive and absolutely multiplicative which
can be formed of the subsets of a given infinite set of power m is
only 2™. ;
EXERCISE 1. From Theorem 1 deduce (with the aid of the axiom of choice) the
following corollary:
For every infinite set M of power m there exists a family ®, formed of 2™ subsets of
the set M, such that for X «®, Ye@®, X# Y¥, we have X—Y =m.
Proof. Let M denote an infinite set of power m. It follows from the axiom of
choice that m? = m (see Chapter XVI, formula (2.2)). Thus the set M is the sum of m
disjoint components each of which is of power m. Thus there exists a function f associat-
ing with each element x of the set M a set f(x) c M such that f(z) =m for a « M and
f(x): fly) =0 for re M, ye M, x#y. Let F denote a family satisfying Theorem 3
and let ® denote the family of all sets

where X «Ff. We shall have of course p(A)c M for Ace the family @ is formed of
subsets of the set M.
If A ¢«F,BeF, A ~ B, then, in view of the property of familyF,we have A -BS 0;
thus there exists an element a¢« A—B such that ae A and a€ B, therefore a ¢ b for
454 XVII. Applications of Zermelo’s theorem

b « B whence, in virtue of the properties of the function f, we obtain f (a)-f(b) = 0

for b « B, which gives
flay) fa) =20,,
xeA

i. €. f(a): p(B) = 0, and since, in view of ae A, we have f(a) cp(A) and p(A) ¥ (Bb).
Thus p(A) ~ g(B) for A «Ff, Be F, A ~ B, and since family F is of power 2™, the family
of all sets p(A) for A « F, i. e. family ® is of power 2™. Suppose now that X «®, ¥ «9,
X = JY. From the definition of family @ it follows that there exist sets A ef and Bel’
such that X = ~(A), Y = y(B). In view of X 4 Y we have A + B, therefore 4—B A 0
and there exists an element ae«A—B, whence, as above, we obtain f(a) Ccp(A) = X
and f(a): g(B) = 0, i.e. f(a)¥ = 0, whence f(a) c X—Y, and since f(a) =m (because
ae M), we have X¥—Y >m. On the other hand, in view of X—Y c M we have X-—Y<
<m. Therefore XY = miforex «®, Ye@®, X~AY. Thus family © satisfies the re-
quired conditions.
2. Let us divide all subsets of an infinite set J of power m into classes, assigning
to the same class two subsets 4 and B of the set M if and only if dA—B<m and B—A <
<m. How many such classes are there?
Answer. There are 2™ of them. On one hand, our classes are disjoint; they
are subsets of the set of all subsets of the set M of power m, the number of which is 2™;
if a set of power 2™ is decomposed into disjoint subsets, then, by Theorem 6 of VI.5
the number of those subsets is <2™. Thus the number of our classes is <2™. On the other
hand, let ® denote a family of subsets of the set M, satisfying the conditions of Exer-
cise 1. Clearly, different subsets of the set M belonging to family ® belong to different
classes. Therefore the number of our classes is >® = 2™. Thus the number of our clas-
ses. is 2") +g. end,
Remark. Without the aid of the axiom of choice we are unable not only to solve
Ixercise 2 for infinite sets of arbitrary power but also to prove that if we define, for
sets of real numbers, a relation Aeb by the condition that for two sets of real num-
bers A and B we have AgB if and only if each of the sets 4—B and B—A is of less
power than the continuum, then the relation @ will be transitive. But we can prove
without resorting to the axiom of choice that the set of all real numbers can be de-
composed into disjoint classes in such a manner that two sets, 4 and B, belong to the
same class if and only if the sets 4 —B and A—A are finite, and we can prove that there
are 2? such classes. We are also able without the aid of the axiom of choice to solve
Exercise 2 for m = &,.
3. Prove that the total number of different famihes of disjoint sets which can
be formed of the subsets of a given infinite set of power m is 2™.
Proof. Every family of disjoint sets which can be formed from the subsets of
a given set of power m is obviously of power < mand it is easy to prove with the aid of the
axiom of choice that the total number of families of power <m formed of the subsets
of a given infinite set of power m is (2")™ — Qm’ _ 9m. Thus the total number of different
families of disjoint sets which can be formed of the subsets of a given infinite set of
power m is <2™. That their number is >2™ follows from the fact that each subset of
a set of power m (there are 2™ of them) determines a family of one-element disjoint
sets. different subsets determining different families.
Cf. also Exercise 6 of IV.11.
4. Let X denote!an arbitrary infinite set of real numbers and D the straight line
y =. Prove with the aid of Zermelo’s theorem that the set 7 = (X x X)—D contains
 § 4. Some theorems on families of subsets 455

a subset symmetrical with respect to the line D and such that for every number a ¢« X
the line x = a intersects the set Z exactly at one point.
5. Prove with the aid of Zermelo’s theorem the proposition of.P. Erdos that if
all straight lines lying in a plane P are divided into two disjoint classes L, and L,, then
there exists a decomposition of the plane into two sets P = S,+8, such that, for 1 =
= 1, 2, each line of class Z, has fewer than 2*° points in common with the set S,.
Proof. Let my denote the least ordinal number of power 2*° and {de}ec, a trans-
finite sequence of type y formed of all straight lines of the plane P. For i= 1, 2 and
a<~y let us denote by 7, the sum of all lines d. where € < a belonging to the class L,.
For a < let HE, = d,(T?—T!) if d, «Ll, and #, = d(T. —T:) if d,«L,. For i=1, 2,
let H denote the sum of all sets H, where a<g and d,¢«L,. Let 8,= E+ (P—T’).
We easily prove that the sets S, and S, satisfy the required bonaiacn

THEOREM 2. If M is an infinite set of power m and F — a family

of power m of subsets of power m of the set M, then there exist m disjoint
sets each of which has m elements in common with each set of family F.
Proof. Let M denote an infinite set of power m. Thus the cardinal
number nt is not finite and therefore (XVI.1) it is an aleph, m= x,
Thus there exists a transfinite sequence of type o,, Wshle<oys formed of
all different elements of the set J.
Further, let / denote a family of power m of subsets of power m of
the set M. In view of Theorem 5 of XV.8 we have x), = 8, and thus there
exists a transfinite sequence {Aghecw, formed of all sets of family / and
containing each of them m times. Now we shall define by means of trans-
finite induction for 6 <a< , a double sequence {qs} of elements of the
set M in the following way.
Let q denote the first term of the sequence Pehe<w, belonging to the
set A. For 1 <1 < a, the set Q, of all terms g; where 6 < a < / is of course
of power <™m (since A < w,), while, in view of A, « F, the set A, is of power
m. By formula (2.3) of Chapter XVI there exist m different elements of
the sequence {Pe}e<w, belonging to the set A,—Q,; they form a sub-
sequence of the sequence {ps}e<w,, Which is also of type ,. Let (eves
be the first 2 terms of this subsequence. The double sequence {qj} where
p<a<o, has thus been defined by transfinite induction.
Now, for 6 < w,, let us denote by R, the set. of all elements q3 where
pB <a<o, and consider any two elements qj ¢«Rz, and qs, « Rz, where
p= 0, = 0,, po < Oo =< &, and py —<.Py’
If a, = a, then the elements qs! and gq? = qs. are (in view of f, < f,)
different terms of the sequence {qé'}:<,, Whence qj}A qs, since this se-
quence has been taken out of a sequence of different terms. In the case
of a, < a, we have (in virtue of the definition of the sets Q,) qi € Qa,+1 © Qos
and (in virtue of the definition of the terms qj) qj « A.,—Q.,, therefore
ds, AF Ye. The same inequality is deduced analogically in the case of a, < a,.
456 XVII. Applications of Zermelo’s theorem

In this way the inequality 6, < 6B, (where B, <a, < my, Be < & < w,)
always implies the inequality qj 4 q, which proves that the sets Rg,
and Rs, are disjoint for 6, < py < ay,.
Since there are m sets Rs where 6 < w,, in order to prove Theorem 2
it remains to show that each of them has m elements in common with
each of the sets A of family F.
Therefore let A denote a given set of family /. In virtue of the de-
finition of the sequence tAghe<o, there exist m different ordinal numbers
a<o, such that A= A,; thus also for every given ordinal number
6 <o, there exist m different ordinal numbers a such that B <a<ao,
and A = A,. For each of those numbers a the element qj is a term of the
sequence {gzte<, and therefore, in virtue of the definition of that se-
quence, it belongs to the set A,; on the other hand qj belongs to the set Rg
(in virtue of the definition of that set). Thus we have qj « A,Rs = ARg
for m different ordinal numbers a satisfying the condition 6 <a < a,.
But from the definitions of the sets Q, and the terms qg it follows that
fOr hay G),68-< 0; <0, 0b as0, We. Nave gs «Oo uaheds te 4s, Gs
Qa,+1C Q., therefore q;' ~ q;. The set AR, contains m different elements
and of course no more (since 4R;C M and M =m).
Theorem 2 is thus proved. We easily deduce from it the following
COROLLARY. If {Xs}e<w, is a transfinite sequence of type ow, of sets of
power S$,, then there exists a transfinite sequence {Yehe<w, of disjoint sets
of power s, such that Y;C Xe for E<a,.
Without the use of transfinite numbers we could express this corol-
lary in the following way:
If we have m > sy sets of power m, then with each of them we can as-
sociate a subset of it of power m such that all those subsets are disjoint.
We shall also mention, without proof, the following two theorems,
which can be proved with the aid of Zermelo’s theorem:
If M is an infinite set of power m and F a family of power m of fune-
tions defined for the elements of the set M (whose values do not necessarily
belong to M), then there exists a family © of power 2™ of different subsets
of the set M such that f(H) AH for Hed, He®, HAH, fe F').
If M is an infinite set of power m and F a family of power m of func-
tions defined for the elements of the set M and assuming each value from
the set M at most once (assuming besides, for the elements of the set M,
values not belonging to M), then there exists a family ® formed of 2™ dif-
ferent subsets of the set M such that f(H)—H 40 for He®@, He®,
EH, f«F (Sierpinski, Fund. Math. 34(1947), p. 33).
1) This theorem was given without proof by A. Lindenbaum [1], p. 185; I gave
the proof in Fund. Math. 34(1947), p. 29-33.
 § 5. Set of all order types of a given power 457
ry

5. The power of the set of all order types of a given power. We shall prove
THEOREM 1. Jf m ts a non-finite cardinal number, then the set of all
order types of power m is of power 2™.
Proof. Let m denote a non-finite cardinal number and / the set
of all order types of power m. As has been proved in XI.1, for a set M of
power m there exist 2™ different relations between its elements, then not
more than 2™ different order arrangements of the set M. Thus if we
divide the order arrangements of the set M into classes, assigning two
order arrangements of the set M to the same class if and only if they
determine the same order type, then, by Theorem 6 of VI.5, the set of
those classes will be of power <2™’, hence by formula (2.2) of Chapter XVI,
of power <2". Thus T < 2™.
On the other hand, if a cardinal number is not finite, then there
exists an ordinal number uw such that m=x,. Let s= {As}e<w, denote
an arbitrary transfinite sequence of type w, with the terms 0 or 1. Let

pS) a= ss (o*+o-+1+ 4).

E<o,

It is easy to see how to generalize the argument of XII.5 (carried

out for the case of «= 0) in order to prdéve that r(s) A 7(s’) for s #8’.
Since the number of different sequences s is 2", we have J > 2",
which, in view of the formula 7 <2" obtained previously, gives 7 = 2",
which proves the truth of theorem 1.
The particular cases of Theorem 1 for mi = s, and m= 2° have been
proved with the aid of the axiom of choice but without the use of trans-
finite numbers in XII.5 and XIJ.6 (see Theorem of XII.5 and Theorem 2
of XIT.6).
EXERCISES. 1. Prove with the aid of Zermelo’s theorem that every ordered
set is a subset of an ordered set no two different segments of which are similar.
Proof. Let g be the type of a given ordered set #’. This set is of course similar to
a subset of a dense set H of type np. Let {a,}._,, be a transfinite sequence formed of all
the elements of the set H; such a sequence exists in virtue of Zermelo’s theorem. Let 7
denote the set of all ordered pairs (a,,¢) where €< uw, €<& and let us order the
set 7 according to the principle of first differences, 7. e. let us assume that (ae, a=
< (az, ; ¢,) if and only if we have Ag< Mg, in, the set A or if €=&, and ¢ < ¢,. The set H
is obviously similar to the subset of the set 7 formed of the pairs (a,,0) for § < p,
whence it follows that also the set / is similar to a subset of the set 7. On the other
hand, it can easily be seen that no two different segments of the set 7 are similar. The
required proof is thus complete.
Remark. A. Denjoy [1] (p. 121 and foll.) gives the term ensembles rangés to ordered
sets no two segments of which are similar and devotes to their investigation an extensive
passage in his book.
458 XVII. Applications of Zermelo’s theorem

2. Give an example of an ordered set whose all non-empty segments are equa-
but all remainders different.
Answer. A set of type w*. More generally, a set of type o* where g is an ordinal
number >1 and a prime component.
3. Prove that if m is a non-finite cardinal number, then the set Z of all order
types gm of power m such that an ordered set of type y will not change its power if we
delete from it an arbitrary finite set of elements is of power 2™.
Proof (Chajoth [1], p. 132). If m is an infinite type of power m, then the type wp
is of power $)Mm, 7. e., by formula(2.1) of Chapter XVI, of power m, and moreover, as
can easily be seen, a set of type wy will not change its type if we remove from it an ar-
bitrary finite set of elements. Further, it is easy to prove that if y, Ag. then wg, ~
A wy, (Chajoth [1], p. 132). Hence it follows that the set Z is of no less power than the
set T of all order types of power m, whence Z > 7, and since of course Zc T, whence
HM, SKE T; therefore Z — 7, and thus in virtues of Theorem 1, Z = 2", qu Cad:

6. Applications of Zermelo’s theorem to the theory of ordered sets.

We say that an ordered set M is confinal with its subset N if there is
no element of the set M that succeeds every element of the set N. If
the set M has a last element, a, then it is of course confinal with its one-
-element subset {a}. The set of all real numbers ordered according to
their magnitude is confinal with the set of all rational numbers and also
with the set of all natural numbers as well as with the set of all prime
numbers.
THEOREM 1. Every ordered set that has no last element is confinal
with a certain subset of itself whose type is a regular ordinal number.
Proof. Let M be a set, ordered by a relation 0, having no last ele-
ment; thus it is an infinite set and it follows from Zermelo’s theorem
that its power is an aleph, ANE = s,- Thus there exists a transfinite se-
quence of type o,, {As}e<e, formed of all different elements of the
set M. Let us denote by N the set composed of the element a, and of
all terms a, of our transfinite sequence such that az:ea, for <a. Now
if a,e N, age N and a < f, then from the definition of the set N it fol-
lows, in view of ag,« N that a,oa,; thus the ordering of the subset N of
the set M by relation o is identical with the ordering of its elements
according to the magnitude of their indices. Hence it follows that the
set N is well-ordered by relation @. Now we shall prove that the set M is
confinal with the set NV. Suppose that it is not so. Thus there would
exist an element a, of the set M with the least index y such that a, ea,
for each element a, of the set V. From the definition of the index y it
follows that if €<y, then there exists an element a, of the set N such
that a,ea¢ does not hold; therefore we have either a, = az, whence a¢oa,,
OY dg 0d, Whence, in view of a,oa, again azea,. We should thus have
aoa, for <y and, according to the definition of the set N, a,« N,
which is impossible because a, ea, for each element of the set YN.
 § 6. Applications of Zermelo’s theorem 459

We have proved that the set N, well-ordered by relation 0, is con-

final with the set M. Thus we have proved that every ordered set having
no last element is confinal with a certain well-ordered subset of itself.
If M is an ordered set of power x, having no last element, then it
is confinal with a certain subset of itself whose order type is an ordinal
number <w,. Among such ordinal numbers 4 (for which the set M is
confinal with a certain subset of itself of type A) there exists a least, t.
Number rt is not confinal with any ordinal number / smaller than t since
in that case the set M would obviously be confinal with a certain subset
of itself of type 4< 7, which contradicts the definition of number t.
Therefore number t is regular. Theorem 1 is thus proved.
It could be proved, moreover, that a regular transfinite ordinal
number must be an initial number and that for every ordered set M
having no last element there exists only one regular initial number t such
that the set M is confinal with a certain subset of itself of type + (Haus-
dorif [2], jp. 132).
We say that an ordered set M is coinitial with its subset N if there
is no element of the set M that precedes every element of the set N.
Analogically to the proof of Theorem 1 we prove that every ordered set
is oinitial with a certain subset of itself whose order type is t* where t is
either number 1 or a regular ordinal number.
Let M denote an ordered set, a — a given element of that set, A — the
set of all elements of the set M that precede a, B — the set of all elements
of the set MW that succeed a. Let a denote the least ordinal number such
that the set A is confinal with a certain subset of itself of type a (in virtue
of Theorem 1 such an ordinal number a does exist; in the case of A = 0
we shall have a= 0), and let / denote the least ordinal number such
that the set B is coinitial with a certain subset of itself of type p*. Type a
is termed the left-side character of the element a and type f* — the
right-side character of the element a, while the system (a, /*) is termed,
after Hausdorff, the character of the element a in the ordered set WM.
Thus for instance in the set of all real numbers ordered according
to their magnitude each element has the character (@, m*), in the set
of all natural numbers number 1 has the character (0,1) and every na-
tural number >1 has the character (1,1). In the set of all ordinal num-
bers of the 1-st and the 2-nd class every positive number of the 1-st
kind has the character (1,1), and every number of the 2-nd kind has
the character (@, 1).
We can also ascribe characters to the gaps of a given ordered
set M. If the cut [A, B] of the set MZ determines a gap, then by the cha-
‘acter (a, B*) of that gap we understand the character of the element
which would fill up that gap (7. e. of an element, added to the set M,
460 XVII. Applications of Zermelo’s theorem

which would have to be regarded as succeeding each element of the set A

and preceding each element of the set B). Thus for instance in the set
of all rational numbers ordered according to their magnitude every gap
has the character (w, m*). In a set of type Q+* the only gap has the
character (Q, 2*)
THEOREM 2. Hvery ordered set of power s, (where wis a given ordinal
number) is similar to a certain set of transfinite sequences of type w, com-
posed of numbers 0 and 1 and ordered according to the principle of first
differences (Sierpinski [29], p. 207-208; Cuesta [1], p. 130-131).
LEMMA. Let 0 be a given transfinite ordinal number (not necessarily
of the 2-nd kind) and let Ss denote the set of all transfinite sequences of type 0,
formed of numbers 0 and 1 and ordered VE AUS to the principle of first
differences. The set Ss has no gaps.
Proof of the lemma. Let [A,B] denote a given cut of the ordered
set Sy. For €<@ let us define numbers Cs by transfinite induction as
follows.
Let ¢) = 1 if there exist in the set A sequences whose first term is 1;
in the contrary case let ¢,= 0. Let a denote a given ordinal number,
0<a<%#, and suppose that we have already defined all numbers ¢;
for <a. Let ¢, = 1 if there exists in the set A a sequence {%z}:e9 such
that 2, = cz for <a and a = 1, and let « = 0 if there is no such se-
quence in the set A. The sequence c = {#e}s<g 18 thus defined by transfinite
induction.
Now let 7 = {we}s-9 be any sequence belonging to the set A. If we
had ¢<~ in the set Ss, then there would exist an index a < @% such that
Cg= % for <a and q=0, %=1, which obviously contradicts the
definition of number c,. Thus we cannot have ¢~<w and therefore we
have either + = c¢ or « Sc. Thus either ¢ is the last element of the set A
or ce B. We shall prove that in the latter case ¢ is the first element of
the set B. Suppose that it is not so, 7. e. that there exists an element
yeé«B such that y<e. Let y = {ye}seg. In view of y<e there exists an
ordinal number a <? such that

(6.1) Ysa Ce AOR eo” and wl yo =F O Pandas Cea lee

In view of ¢,=1 and of the definition of number ¢, there exists

a sequence xe A such that w= cs for €< a and #,=1 which by (6.1)
gives wv = yg for § < a, and 4% = 1, ¥,= 0, whence y <2, which is impos-
sible because ve A and ye B. We have proved that if the sequence ¢
belongs to B,.then it is the first element of the set B.
Thus we have proved that the cut [A, B] gives no gap, 7. e. we have
proved our lemma.
 § 6. Applications of Zermelo’s theorem 461

Remark. The set Ss, satisfying the conditions of our lemma, has
a first and a last element but it is not continuous because it has jumps.
H.g. the sequences a = {dag}:<9 where a,=0 and a=1 for 0<&<%
and the sequence b= {be}scg where 6, = 1 and bs: = 0 for 0< €<@% can
easily be shown to be successive elements of the set Sy (7. e. there is no
intermediate element between them).
In the case where ? is a number of the 2-nd kind we can obtain
a continuous set from the set Ss by removing from it all those sequences
for which there exists such an ordinal number yw that all their terms with
indices > are equal to 0.
Proof of Theorem 2. Let M denote an ordered set of power x,
and let (Ug}e<o, Genote a transfinite sequence of type w,, formed of all
different elements of the set M. We shall define by transfinite induc-
tion a double transfinite sequence {a5}, where & < o,, 1 < , formed of
numbers 0 and 1, as follows.
Let aj = a,= 0. Let a denote a given ordinal number such that
0 <a<o, and suppose that we have already defined all numbers Can
and. Bsn for é<aand y <a. If there is no element wu: of the set M such
that €< a and uz: S u,, then let az, = a3,+1 for 7 <a. If such elements uz
exist in the set M, then let A,, denote the set of all sequences ony ees
where € is an ordinal number such that é<a and u:<u,. In virtue
of our lemma in the set S,, exists the first sequence that does not precede
any sequence of the set A»a,; let {a,},<2q be that sequence. (If in the set Seq
there were no sequences succeeding each sequence of the set Aa, we
should have of course a, = 1 for 7 < 2a). Further, for <a let On=
= Oi1= 0 if ug Ua, ANd Ag= G5a1i1 if ue—<us. Finally let ag,=0
and 43,11= 1. Numbers as, and Bint are thus defined for <a and
H< a.
Thus we have defined numbers A, for <q, and 7 <a, by trans-
finite induction.
Let a® = {On ln<e, for §<, and T'= {a*}e<, and let us order the
set 7 according to the principle of first differences. We shall prove that
the sets M and 7 will be ordered similarly. For this purpose it is of course
sufficient to prove that

(6.2) Ti aes 8 oy, eer endwe << ig, then: | 10% ar:
(6.3) ff oxi =o, ieand “w—=2w,, then “a? ae?
Suppose that a< 6 < ,, U.< Ug. According to the definition of
the double sequence {a;}, the sequence {an in<ee in the set Sg, does not
precede the sequence {a;},<2,: thus we have either at*<a’ or
(6.4) a,—a for <28.
462a XVII. Applications of Zermelo’s theorem

But, in view of a< 6 and u,<ug, according to the definition of

numbers dzg and dyg;1, We have dz, = ds341— 0, and since ab; = 0 and
Op = 1, we obtain, by (14), at < a’ and formula (6.2) is thus proved.
Suppose now that formula (6.3) is not true. Then there obviously
exist two least ordinal numbers a and 6 such that a < 6 <, and
(6.5) gS Ua
and
(6.6) at 3 a8,
whence
(Gio at ao 0 eG and Up e,, Nneh) Or Oe,
Suppose now that €< #6 and u<u,. H a<, then, by (6.5) and
(6.7), we obtain af < a*. If € <a, then, in view of u;< ug, (6.5) and (6.2),
we have a'<a*. Finally if {=a then a =a‘. Thus we have the
following formula:

(6.8) .TfC<P and u<ug, then a sar.

From the definition of the double sequence {az} it follows that either
we have A, == Ose = 0 for 7 < 6, and in that case we of course have
(in the set So,)

(6.9) {0p 239 (Gn peasy

or the sequence {a5},1, is the first sequence in the set Ss; that does not
precede any sequence of the set Ass. In view of the definition of the
set Aog, if ¢< 6 and u;-<u~,, then ery, e Ag, and vice versa. Therefore
{a},<9» i8 the first element of the set Sop such that for ¢< 6 and uz
<us,
we have lan loeas BE) lay eae
But, by (6.8), for ¢<f and u<ug we have {aj}ncop 3 {azhn<op
(since if {a%},<2~< {d;\n<99 in the set Ss,, we should have a*—aé, in
contradiction of (6.8). In view of a < f, {a;},<2, is one of the elements
of the set So, such that for ¢< 6 and u;-<ug we have {an in<26 3 {Ae} ncops
in view of the properties of the element ees we again obtain for-
mula (6.9).
In view of (6.5), a < 6 and of the definition of numbers a, we obtain
dog = 1 and Gee: thus, by (6.9), we have a’<a*, which contradicts (6.6).
The assumption that formula (6.3) is not true leads to a contradiction.
We have proved formulae (6.2) and (6.3) and thus also Theorem 2.
THEOREM 3. For every ordinal number a there exists an ordered set
of power 2° (of elements of the set So.) which for every ordered set of power
Sa+1 contains a subset which is similar to that set (Sierpinski [47], p. 62).
 $ 6. Appheations of Zermelo’s theorem 463

Proof. Let 4 = a+1 and let us denote by H, the set of all sequences
{ds}e<oy belonging to the set S,,, for which there exists an ordinal number
(dependent on the set in question) 2 < w, such that

(6.10) Gn and te $ = 0 for Year eA

Let us denote by Pi and Ps respectively the following two properties

of an ordered set EH:
Pr. If Y is a subset of power <x, of the set H, then there exist ele-
ments a and b of the set E such that (in the set £)

(6.11) mem 0 LOR es,

mo iy and 0, are subsets of power <x, of the set H such that

(6.12) GOW PION Vee rb € Oy.

then there exists an element ¢ of the set H such that

(6.13) UC mrtotes hie.O), 06 Oe

(¢. e. according to property Pf the set H is neither coinitial nor confinal

with any subset of itself of power <s,, and according to property P?
the set H# contains no two successive subsets of power <x,).
Hausdorff gives the term y7,-set to every ordered set having pro-
perties Pf and P3.
We shall prove that the set H, is an 7,-set.
Indeed, let @ be a subset of the set H, which is of power <y,,, hence,
in view of w= a+1, of power <x,. In view of @C H,CS,,, we can set
Q = {@}:<, where y < ow, and “= {43}<<o,- In view of a « H, for 6 <q,
there exists for every number ¢ <q an ordinal number A; < w, such
that aj, =1 and a;=—0 for 4 <é <a,.
The set of all numbers Az where €<@ is of power 9 <a, =8, and
thus there exists an ordinal number 4 < o,41 = , such that A; < A for
£ <q (for we can set

jes NAS
pAb)
cep

in view of A,< w, we have A; <0, —=%8, =Nai1; therefore A; <s, for

€ <q, whence, in view of » <x, we obtain

aie aN ne
AS eames, ai,
i<@
which proves that 2 < wg11 = o,).
 \
464 XVII. Applications of Zermelo’s theorem

Let
dge= On LOY Sen} aya

On=acl™ LOR ORE <A 0 0 Oe LOL SA ace ace

Cr (As he<oys b= Dehe<o, :

We shall of course have ae H,, be H, and formula (6.11) will be

true. Thus the set H, has property Pr.
Now let Q, and Q, be subsets of the set H, which are of power <k,,
thus of power <x,, and for which formula (6.12) holds. In virtue of the
lemma, the set S,, has no gaps. Hence it easily follows that there exists
a sequence p = {Ps}e<m, Which is the first sequence of the set 8, such that

(6.14) Css Pape tOr a mae Or

By (6.12) we shall thus have

(6.15) p36 for beQ,.

If we had peQ,, we should have, in view of Q,C H,, pe H, and

there would exist an index y < w, such that p,=1 and p-= 0 for y<
NY
Let? pe—= p, lor € =< y,.p,= 0 and pe— efor wy 6 =<G),.
Setting p’ = {Pehe<wy we shall of course have p’ Sp in the set So,,
and in the set S,, there is no sequence w such that p’ <u <p; in view
of (6.14) and p«Q, we should thus have a 3 p’ for aeQ,, which is con-
trary to the definition of the sequence p. ad
Thus we have péQ,, therefore, by (6.15), we obtain

(6.16) Dic OpmAOter GU ets

Since the sets Q, and Q, are of power <x,, we can set Q, = {a5}r<9,,
where 9, <a, Qo = {D*}c<q, Where g, <%,. Since Q,C H, and 9,C H,,
we shall obtain, as above, an ordinal number 4 < , such that

(6.17) OO) for, Sg A eee,

and
(6.18) Bb=0 for E<q,A<E<
ay.
Let
(6.19) Cp pe te OTS ees
@=1 and c=0 for A<E<o,, C= {Cg}e<a,-
 § 6. Applications of Zermelo’s theorem 465

We shall of course have ce H,. Let a= {dg}e<w, and b= Wehe<w,

be sequences such that aeQ,, b<«Q.; by (6.14) and (6.16) we shall have
a3p~<, which, by (6.18), as can easily be seen, gives (in the set S,)

(6.20) {Ggheca 3 (Deheca S (Deheca.

But by (6.17) we have a,= 0 and by (6.19) and (6.20) we obtain

{Aghe<a S {Ce}ex, in the set Sjy1, thus a Re.
Since p< and, by (6.18), there exists an ordinal number x < 4,
such that p,— 0; for é< x, p, = 0, b, = 1, we.have by (6:19) c.= b, for
Ck — pu 0,0, = 1,and thus (in the set -S.41) {:tec.—< {Bsheuy}
which gives c<b. Thus we have a<c<b and the set H, has the pro-
perty Pe.
Therefore we have proved that the set AH, is an 7,-set.
F. Hausdorff [2] (p. 181-182) has proved that every 7,-set contains,
for every ordered set of power s,, a subset. which is similar to that set.
Indeed, let M denote an ordered set of power x, and let (Usheca,
denote a transfinite sequence of type m, formed of all different ele-
ments of the set M. Further, let H denote an ,-set and {he}s<, — a trans-
finite sequence of type g formed of all different elements of the set H.
Let f(u,) = h,. Let y denote an ordinal number such that 1 < y < o,
and suppose that we have already defined all elements f(w;) for < y
and we have f(ug) < f(uz) if Ex y, €<y and ug<u;. In view of y < o,
we have y < x, and the sets of all elements f(u;) where & < y is of power
y <x,. We define f(u,) as the first term h of the sequence {hele<, which,
for every index é < y, has with respect to f(uz) the same order relation
in the set H as wu, has with respect to uw; in the set M. Since the set H,
as an 7,-Set, has properties Pt and P2, such a term h of the set {he}scy
always exists. In this manner the elements /(u;) for é< @, are defined
by transfinite induction and it can easily be seen that their set is similar
to the set M.
We shall now calculate the power of the set H,. From the definition
of the set H, it follows that if, for ) < w,, we denote by 7, the set of all
sequences {dg}:+o, of the set S,, where a;= 0 for << ,, then

(6.21) HO eT
o<o,, |

But in. view of 0 < w, = @ai1, we-have 3 <x,, and j= Sati < 2°;
thus we have 75 = 2 < 2% and formula (6.21) gives
Sa =’ N 8
UE EO a icy ali WR
Cardinal and ordinal numbers 30
466 XVII. Applications of Zermelo’s theorem

On the other hand we have of course H, > 2™ since every sequence

{ds}e<o, of the set So, where Me, = 1 and a; = 0 for a, < é< o, belongs
to H,. Thus we have H,= 2™.
The set H,, satisfies the conditions of Theorem 3, which is thus proved.
At the same time we have proved

THEOREM 4. For every ordinal number a there exists an q+1-8et of

power oe

It will be observed that we can prove that for every ordinal num-
ber a the least power of the y,.1-sets is 2°*, whence it follows that for
a given ordinal number a an 4,41-Set of power s,+41 exists if and only if
s,
Zi ee Na+1-
For a= 0, we obtain from Theorem 4 and from the properties of
Hy- Sets

COROLLARY 1. There exists an ordered set of the power of the continuum

which for every ordered set of power <s, contains a subset similar to
that set*).
Hence immediately follows

COROLLARY 2. From the continuum hypothesis follows the existence of-

an ordered set of the power of the continuum which for every ordered set
of power not greater than the continwum contains a subset similar to that set.
We do not know whether the existence of such a set is equivalent
to the continuum hypothesis ”).

EXERCISE. Prove with the aid of Zermelo’s theorem that every real function
of a real variable is the sum of two (1-1)-mapping functions*).
Proof. Let m denote the least ordinal number of the power of the continuum.
From Zermelo’s theorem it follows that there exists a transfinite sequence of type ,
Wehccg: formed of all different real numbers. Let f(x) denote a given real function
of a real variable. Let f,(”,) = f(x), f2(z,) = 0. Let a denote an ordinal number such
that 1 << a< gq and suppose that we have already defined all values fil@s) and f(%z)
for § < a. We shall define number /,(7,) as the first term of the sequence {x,} E<p that is
different from each of the numbers

(x) fi(é) and f(x,)—f(wg)+filz) for E<a.

Of numbers (*) there are not more than a+ a, 7. e. fewer than 2“ since if a+ a>
=> 2*0 — 2%0 + 20 we should have @ > 2*° = », which is impossible because m denotes

1) I proved this theorem on different lines in [18], p. 281.

*) For further investigations of 7,-sets see Gillman [2] and [1], p. 12-13.
3) This theorem was given without proof by A. Lindenkaum LO 90% ae
 4

§ 6. Applications of Zermelo’s theorem 467

the least ordinal number of power 2°. Thus there exist terms of the sequence {w,}.— é
that are different from each of the numbers (*).
Finally let 7.(a,) = f(~,)—fi(w,). In this manner the functions f,(#) and /,(x) are
defined for real x» by transfinite induction and we always have f (az) = flv) + fo(ae)
for § <p; it follows from the definition of numbers f,(~,) that the function f,(x) is
(1-1)-mapping and that f,(~,) A f(x) — f(x) + fil@e) for §<a<p@, and thus f(x,)—
— fila) A f(%s)—fi(wg), 0. e. falXq) F fale) for §<a<q. whence it follows that the
function f,(#) is (1-1) mapping.
Thus we have proved that for real # the function f(x) is the sum of two (1-1) mapp-
ing functions f.(”)+ f(z), q.e. d.

30*
 APPENDIX

In a letter dated January 3rd, 1958, Mr. Elliot Mendelson gave

the following simple proof of theorem 2 (p. 460):
THEOREM. Every ordered set of power s, (where uw is a given ordinal
number) is similar to a certain set of transfinite sequences of type w, of
numbers 0 and 1 and ordered according to the principle of first diffe-
rences.
Proof. Let M denote an ordered set of power s, and let (Ug e<e,
denote a transfinite sequence of type w,, made up of all the different
elements of the set M. For Eé<a,, €<, define

a =1 if us Sue and az = 0 otherwise ,

where = is the order in M. This defines a mapping f(u;) = {A}s<w,-

Now assume uw <u, in M. If a = 1, we have ug 3 uz S u,, thus a? = 1.
Thus a: < az for §<w,. Also, since usu,, u=0<l=u,. Thus
(uz) sf (UM) in Sa. Hence f is a 1-1 order-preserving mapping and
the theorem is proved.
 BIBLIOGRAPHY

Aigner, A.
|1] Der multiplikative Aufbau beliebiger unendlicher Ordnungszahlen, Monats-
hefte f. Math. 55 (1951), p. 297-300.
Alexiewicz, A. and Orlicz, W.
[1] Remarque sur Véquation fonctionnelle f(x+y) = f(w)+ fly), Fund. Math. 33
(1945), p. 314-315.
Aronszajn, N.
[1] Characterization of types of order satisfying aj+ a, = a,+ a), Fund. Math. 39
(1952), p. 65-96.
Bagemihl, F.
{l] ‘A theorem on intersections of prescribed cardinality, Annals of Math. 55 (1952),
p. 34.
Banach, Ss.
[1] Sur Péquation fonctionnelle f (a+ y) = f(a)+ f(y), Fund. Math. 1 (1920), p. 123.
[2] Un théoréme sur les transformations biunivoques, ibidem 6 (1924). p. 236-239.
Bernays, P.
[1] A system of axiomatic set theory, Journal of Symbolic Logie 8 (1934), p. 89-106.
[2] Sur le Platonisme dans les Mathématiques, Enseignement Math. 34 (1935),
p. 52-69.
Bernstein, F.
[1] Untersuchungen aus der Mengenlehre, Math. Annalen 61 (1905), p. 117-155.
[2] Inaugural Dissertation, Halle 1909.

Bieberbach, L.
[1] Bemerkung zum dreizehnten Hilbertschen Problem, Journal f. reine u. angew.
Math. 165 (1931), p. 89-92.
Birkhoff, G.
[1] Lattice theory, New York 1948.
Blumberg, H.
[1] Methods in point sets and theory of real numbers, Bull. Amer. Math. Soe. 36
(1930), p. 809.
Borel, E.
[1] Legons sur la théorie des fonctions, Paris 1914.
[2] Sur les ensembles effectivement énumérables et sur les définitions effectives,
Rend. Acad. dei Lincei 5. 28, sem. II (1919), p. 163-165.
470 Bibliography

[3] Bléments de la théorie des ensembles, Paris 1929.

[4] Les nombres inaccessibles, Paris 1952.

Bourbaki, M.
[1] Bléments de Mathématique, Partie I-Livre I (Théorie des Hnsembles, fascicule
de résultats), Actualités Scientifiques et Ind. 843 (1939), p. 37.

Cantor, G.
[1] Gesammelte Abhandlungen, Berlin 1932.
[2] Uber eine Higenschaft des Inbegriffes aller reellen algebraischen Zahlen, Journal tf.
reine u. angew. Math. 77 (1874), p. 258-262.

Carruth, Ph. W.
[1] Roots and factors of ordinals, Proc. Amer. Math. Soc. 1 (1950), p. 470-480.

Cavaillés, I.
[1] Sur la dewxiéme definition des ensembles finis donnée par Dedekind, Fund.
Math. 19 (1932), p. 143-184.
Chajoth, Z.
[1] Beitrdge zur Theorie der geordneten Mengen, Bull. Se. Math. 2 (1897), p. 257.
[2] Fund. Math. 16 (1930), p. 132-136.

Church, A.
[1] Alternatives to Zermelo’s assumption, Trans. Amer. Math. Soc. 29 (1927),
p. 178-208.
[2] Introduction to mathematical logic, Ann. of. Math. Studies 13, Princeton 1944.

Cippola, M.
[1] Sur postulate di Zermelo e la teoria dei limiti delle funzioni, Atti Acad. Gioenia
Se. Nat. in Catania 6. V (1913). i
[2] Sut fondamenti logict delle’ Matematica, Annali di Matematica IV. 1 (1924),
Deer 9): ge :

Cooper, R.
[1] The converses of the Cauchy -Hoélder inequality and the solutions of the inequality
g(a+y) <g(x)+ gy), Proc. London Math. Soc. 2.26 (1927), p. 415-432.

Couturat, L.
1) Les principes des Mathématiques, Paris 1905.

Cuesta, N.
[1] Notas sobre unos trabajos de Sierpinski, Revista Mat. Hisp.-Amer. 4 (1947),
p. 130-131.
[2] Ibidem 9 (1949), p. 132.
Dayis, A. C.
[1] Cancellation theorems for products of order types, Bull. Amer. Math. Soc: 58
(1952), p. 63. f
Davis, A. C. and Sierpinski, W.
[1] Sur les types Wordre distincts dont les carrés sont éqaux, C. R. Acad. Se.
Parise Zoom (L952) p.ss00:
 Bibliography 471

Dedekind, R.
[1] Stetigkett und irrationale Zahlen, Braunschweig 1872.
[2] Was sind und was sollen die Zahlen, Braunschweig 1888.

Denjoy, A.
[1] L’énumération transfinie (Livre I, La notion du rang), Paris 1946.
[2] Lordination des ensembles, C. R. Acad. Sc. Paris 236 (1953), p. 1393-1396.

Dushnik, B. and Miiller, E. W.

[1] Concerning similarity transformations of linearly ordered sets, Bull. Amer.
Math. Soc. 46 (1940), p. 322.
[2] Partially ordered sets, Amer. J. Math. 63 (1941), p. 600-611.
Erdos, P.
[1] Some remarks on set theory, Proc. Amer. Math. Soc. 1 (1950), p. 127-141.

Erd6és, P. and Kakutani, S.

[1] On non-denumerable graphs, Bull. Amer. Math. Soc. 49 (1943), p. 457-461.

Erdos P. and Rado, T.

[1] A combinatorical theorem, Journal of the London Math. Soe. 25 (1950),
p. 249-255. sf
Fichtenholz, G. et Kantoroyitch, L.
[1] Sur les opérations linéaires dans espace des fonctions bornées, Studia Math. 5
(1934), p. 69-98. i
Finsler, P.
[1] Hine transfinite Folge arithmetischer Operationen, Comm. Math. Helvetici 25
(1951), p. 75-90.
Firestone, P. and Rosser, J. B.
[1] J. Symb. Logie 14 (1949), p. 19.

Fraenkel, A.
[1] Hinleitung in die Mengenlehre, Berlin 1919.
[2] Uber den Begriff ,,definit “und die Unabhdngigkeit des Auswahlaxioms, Sitzungs-
ber. Preuss. Akad. Wiss. Phis.-Math. (1922), p. 253-257.
[3] Zehn Vorlesungen uber die Grundlagen der Mengenlehre, Leipzig und Berlin 1927.
[4] Sur la notion dexistence dans les mathématiques, Enseignement Math, 34 (1935),
De LS=a2"
[5] Uber eine abgeschwichte Form des Auswahlaxioms, J. Symbolic Logie 2 (1937),
p-. 1-25.
[6] Zu den Grundlagen der Cantor-Zermeloschen Mengenlehre, Math. Ann. 86
(1922), p. 230-237. .
[7] Laxiome du choix, Revue Phil. de Louvain 50 (1950), p. 431.

Fraissé, R.
[1] Sur la comparaison des types de relations, C. R. Acad. Sc. Paris 226 (1948),
Ds GSE
Fréchet, M.
[1] Pri la funkeia ekvacio, Enseignement Mathématique 15 (1923), p. 390-393.
472 Bibliography

Gillman, L.
[1] Remarque sur les ensembles 7,,.C. R. Acad. Sc. Paris 241 (1955), p. 12-13.
[2] Some remarks on 1,-sets, Fund. Math. 43 (1956), p. 77-82.
Godel, K.
[1] What is Cantor continuum problem, Amer. Math. Monthly 54 (1947), p. 515-525.
[2] Proc. Nat. Acad. Se. 24 (1938), p. 556-557.
[3] The consistency of the axiom of choice and the generalized continuum hypothesis,
Ann. Math. Stud. 3 (1940).

Hahn, H.
[1] Uber die Abbildung einer Strecke auf ein Quadrat, Ann. di Matematica
3.21 (1913), p. 33-35. ;
Hamel, G.
[1] Hine Basis aller Zahlen und die unstetige Losungen der Funktionalgleichung
f(@+y) = f(x)+ fly), Math. Ann. 60 (1905), p. 459-462.

Haperin, [.
[1] Non-measurable sets and the equation f (a+ y) = f(#)+ f(y), Proc. Amer. Math.
Soc. 2 (1951), p. 221-224.
Hardy, G. H.
[1] A theorem concerning the infinite cardinal numbers,’ Journal of pure and appl.
Math. 35 (1903), p. 87-94.
‘ Hartogs, F.
[1] Uber das Problem der Wohlordnung, Math. Ann. 76 (1914), p. 438-443.
Hausdorff, F.
[1] Der Potenzbegriff in der Mengenlehre, Jahresber. der deutschen Math. Ver. 13
(1904), p. 569-571.
[2] Grundziige der Mengenlehre, Leipzig 1914.
[3] Uber zwei Sdtze von G. Fichtenholtz wnd L. Kantoroviteh, Studia Math. 6 (1936),
p. 18-19.
Hessenberg, G.
[1] Grundbegriffe der Mengenlehre, Gottingen 1906.
Hilbert, D.
[1] Uber die stetige Abbildung einer Linie auf ein Flichenstiick, Math. Ann. 38
(1891), p. 459-460.
[2] Die logischen Grundlagen der Mathematik, Math. Ann. 88 (1923), p. 151-165.
[3] Grundlagen der Geometrie, re 1923.
Hoborski,A
[1] Une remarque sur la limite des nombres ordinaux, Fund. Math. 2 (1921), p. 193-198.
Hobson, E. W.
[1] Theory of functions of real variable, Cambridge 1907.
Inagaki, T.
[1] Le probleme de Souslin dans les espaces SONS Journal Fac. Se. Hokkaido
Univ. 1. 8 (1939-1940), p. 25-46.
[2] Ibidem 1. 9 (1940), p. 145-201.
 Bibliography 473

Jacobsthal, E.
[1] Vertauschbarkeit transfiniter Ordnungszahlen, Math. Ann. 64 (1907), p. 475-488.
[2] Uber den Aufbau der transfiniter Arithmetik, ibidem 66 (1909), p. 145-194.
Kac, M.
[1] Une remarque sur les équations fonetionnelles, Comm. Math. Helv. 9 (1936/37),
p. 170.

Kamke, E.
[1] Allgemeine Mengenlehre, Encyklop. der Math. Wiss., § 13.
Kestelman, H.
[1] Automorphisms of the field of complex numbers, Proc. London Math. Soe. 2, 53
(1951), p. 1-12.
Khintchine, A.
[1] Sur les suites de fonctions analytiques bornées dans leur ensemble, Fund.
Math. 4 (1924), p. 72-75.

Kiyoshi Iséki
[1] Journ. Osaka Inst. Se. Technol. 2 (1950), p. 109.
[2] Sur les ensembles finis, C. R. Acad. Sc. Paris 231 (1950), p. 1396-1397.

Knaster, B. et Kuratowski, C.
[1] Sur les ensembles connexes, Fund. Math. 2 (1921), p. 206-259.
Kneser, H.
{1] Hine direkte Ableitung des Zornschen Lemmas aus dem Auswahlaxiom, Math.
Zeistschrift 53 (1950), p. 110-122.

K6nig, D.
[1] Zur
Theorie der Médchtigkeiten, Rend. Palermo 26 (1908), p. 339-342.
[2] Uber Graphen wnd ihre Anwendung auf Determinantentheorie und Mengenlehre,
Math. Ann. 77 (1916), p. 453-465.
[3] Sur les correspondances multivoques des ensembles, Fund. Math. 8 (1926),
p. 114-134.

Kuratowski, K.
[1] Sur la notion Wensemble fini, Fund. Math. 1 (1920), p. 129-131.
[2] Une méthode @ elimination des nombres transfinis des raisonnements mathéma-
tiques, ibidem 3 (1922), p. 77-108.
[3] Sur Vétat actuel de Vaxiomatique de la théorie des ensembles, Ann. Soe. Pol.
Math. 3 (1924), p. 146-147.
[4] Topologie I, Warszawa-Wroclaw 1948.

Kuratowski, K. and Mostowski, A.

[1] Teoria mnogosci, Warszawa-Wroclaw 1952.

Kurepa, G.
[1] Ensembles ordonnés et ramifiés, Publ. Math. Univ. Belgrade 4 (1935), p. 1-138.
[2] Le probléme de Souslin et les espaces abstraits, C. R. Acad. Se. Paris 203 (1936),
p. 1049-1051.
[3] L’hypothése de ramification, ibidem 202 (1936), p. 185-187.
ATA Bibliography

[4] Lhypothése du continu et les ensembles partiellement ordonnés, ibidem 204

(1937). p. 1196-1198.
[5| Transformations monotones des ensembles partiellement ordonnés, ibidem 204
(1937), p. 1033-1035.
[6] Deux conséquences équivalentes, relatives aux nombres ordinaux, de la bonne
ordination du continu linéaire, ibidem 234 (1952), p. 175-176.
[7] Math. Ann. 126 (1953), p. 381-382.

Lebesgue, H.
[1] Sur les fonctions représentables analytiquement, Journal de Math. 1 (1905),
p. 213.
[2] Atti Acad. Se. di Torino 42 (1907), p. 532.
[3] Sur les correspondances entre les points de deux espaces, Fund. Math. 2 (1921),
p. 256-286.
[4] Bull. Sc. Math. 2. 46 (1922).
[5] Sur la mesure des grandeurs, Enseign. Math. 32 (1933), p. 23-51.
[6] Les controverses de la théorie des ensembles et la question des fondaments, Les
entretiens de Zurich sur les fondaments et les méthodes des sciences mathématiques, 8-9,
Décembre 1938 — Exposés et discussions publiés par le président des débats F. Gonseth,
Zurich ,1941, p. 109-122.
[7] Legons sur les constructions géométriques, Paris 1950.
Levi, B.
[1] Intorno alla teoria degli aggregati, Rend. R. Sct. Lomb. di Se. e Lett. Il. 35
(1902), p. 863-868.
Lindemann, F.
[1] Uber die Zahl x, Ann. Math. 20 (1882), p. 213-225.
Lindenbaum, A.
[1] Sur le nombre des invariants des familles de transformations arbitraires, Ann.
Soc. Pol. Math. 15.(1936), p. 185.
[2] Sur quelques propriétés des fonctions de variable réelle, ibidem 6 (1927),
p. 129-130.
[3] Sur les constructions non-effectives dans lV Arithmetique élémentaire, ibidem 10
(1931), p. 118-119.
[4] Sur les ensembles dans lesquels towtes les équations dune famille donnée on
un nombre de solutions fixé @avance, Fund. Math. 20 (1933), p. 1-29.
Lindenbaum, A. und Mostowski, A.
[1] Uber die Unabhéngigkeit des Auswahlarioms wnd einiger seiner Folgerungen,
C. R. Soc. Sc. et Lett. de Varsovie, Cl. III. 31 (1938), p. 27-32.

Lindenbaum, A. et Tarski, A.
[1] Communication sur les recherches de la théorie des ensembles, C. R. Soc. Se.
et Lett. de Varsovie, Cl. III. 19 (1926), p. 299-330.
Liouville, J.
[1] Sur des classes trés entendues de quantités dont la valeur west ni algébrique,
ni méme reductive a des irrationnelles algébriques, Jour. Math. 16 (1851).
Lusin, N.
[1] Sur les ensembles analytiques, Fund. Math. 10 (1926), p. 1-95.
 Bibliography AT

[2] Sur les ensembles analytiques, Recueil Math. Moscou 33 (1926), p. 237-289.
[3] Legons sur les ensembles analytiques, Paris 1930.
Maccaferri, E.
[1] Per. di Matem. IV (1939), p. 92-101.
Malchair, H.
[1] Bull. Soc. Se. Liége 1.
[2] Ibidem 3, p. 133-134.
Maraham, D.
[1] Set functions and Souslin’s Hypothesis, Bull. Amer. Math. Soe. 54 (1948),
p. 587-590.
Marczewski, E.
[1] Sur deux propriétés des classes Wensembles, Fund. Math. 33 (1945), p. 303-307.
[2] Ensembles indépendants et leurs applications a la théorie de la mesure, ibidem 35
(1948), p. 13-28.
Mazurkiewicz, S.
[1] C. R. Soc. Sc. et Lettres de Varsovie 7 (1914), p. 322-383.
[2] Sur la décomposition du plan en courbes, Fund. Math. 21 (1933), p. 43-45.
Menger, K.
[1] Der Intuitionismus, Blatter f. deutsche Phil. 4 (1930), p. 314.

Milgram, A. P .
[1] Rep. Math. Coll. II. 1 (1939), p. 18-30.

Mostowski, A.
[1] Bericht aus einem Satz von Gédel viber die Widerspruchsfreiheit des Auswahl-
axioms, Ann. Soc. Pol. Math. 17 (1938), p. 115.
[2] On the concept of a finite set, Sprawozd. z posiedzen Warsz. Tow. Mat. 31
(1938), p. 13-20.
[3] Axiom of choice for finite sets, Fund. Math. 33 (1945), p. 137-168.
[4] Logika Matematycena, Warszawa-Wroclaw 1948.
[5] On the principle of dependent choices, Fund. Math. 35 (1948), p. 127-130.
[6] Uber die Unabhdngigkeit des Wohlordnungssatzes von Ordnungsprinzip, ibidem
32 (1939), p. 201-232.

Miller, E. W.
[1] A note of Souslin’s problem, Amer. J. Math. 65 (1943), p. 673-678.

Miiller, E. W. and Dushnik, B.

[1] Concerning similarity transformations of linearly ordered sets, Bull. Amer.
Math. Soc. 46 (1940), p. 322. i
[2] Partially ordered sets, Amer. J. Math. 63 (1941), p. 600.
(

y. Neumann, J.
[1] Zur Hinfihrung der transfiniten Zahlen, Acta Litt. Acad. Sc. Szeged X. 1 (1923),
p. 199-208.

Neumer, W.
[1] Uber den Aufbaw der Ordnwngszahlen, Math. Zeitschr. 53 (1951), p. 59-69.
 476 Bibliography

Novak, J.
[1] A paradoxical theorem, Fund. Math. 37 (1950), p. 77-83.
_ Novikoff, P.
[1] Bull. Acad. Se. URSS 1939, p. 35-39.

Ostrowski, A.
[1] Uber die Funktionalgleichung der Haponentialfunktion, Jahresber. d. deutschen
Math. Ver. 38 (1929).
Peano, G.
[1] Sur une courbe qui remplit toute une aire plane, Math. Ann. 36 (1890), p. 157-160.
[2] Additione, Revista di Mat. III. 8 (1906), p. 143-155.
Poincaré, H.
[1] Science et méthode, Paris 1912.
Ramsey, F. P.
[1] On a problem in formal logic, Proc. London Math. Soc. 2. 30 (1930), p. 264-286.
Rothberger, F.
[1] On some problems of Hausdorff and of Sierpinski, Fund. Math. 35 (1948),
p-. 29-46. ;
Russell, B.
[1] On some difficulties in the theory of transfinite numbers and order types, Proc.
London Math. Soc. 4 (1906), p. 29-53.
[2] Sur les axiomes de Vinfini et du transfini, C. R. Soc. Math. de France, Séance
du 22 Mars 1911, p. 22-35.

Russell, B. and Whitehead, A. N.

[1] Principia Mathematica, vol. 2, Cambridge 1927.

Schonberg, I. J.
_ [1] On the Peano curve of Lebesgue, Bull. Amer. Math. Soc. 44 (1938), p. 519.
Schonflies, A.
[1] Entwickelung der Mengenlehre, Leipzig und Berlin 1913.
Segre, B.
fll Atti Acad. Naz. Lincei (1947), p. 414.

Sehrmann, S.
[1] Some new properties of transfinite ordinals, Bull. Amer. Math. Soc. 47 (1941),
p. 111-116.
Sieczka, F.
[1] Sur Punicité de la decomposition de nombres ordinaux en facteurs irréductibles,
Fund. Math. 5 (1924), p. 172-176.

Sierpinski, W.
[1] Sur Vensemble des points angulaires dune courbe y = f(x), Bull. Acad. Se. de
Cracovie 1912, p. 850-855.
[2] O kreywych wypetniajacych kwadrat, Prace Mat.-Fiz. 23 (1912), p. 193-219.
[3] C. R. Acad. Se. de Paris 163 (1916), p. 688.
 Bibliography AT 7

[4] Dowéd ogdlnego wzorw interpolacyjnego S. Bernsteina, Wiad. Mat. 22 (1918),

p. 177-180.
[5] L’axiome de M. Zermelo et son réle dans la Théorie des Ensembles et VvAnalyse,
Bull. Acad. Se. Cracovie (1918), p. 97-152.
[6] Sur Véquation fonctionnelle f(a+y)=f(~)+fly), Fund. Math. 1 (1920),
p. 116-122.
[7| Les exemples effectives et Vaxiome du choix, ibidem 2 (1921), p. 112-118.
[8] Sur Pégalité 2m = 2n pour les nombres cardinaux, ibidem 3 (1922), p. 1-16.
[9] Ibidem 3 (1922), p. 180.
[10] Sur un probléme concernant les sous-ensembles croissants du continu, ibidem
3 (1922), p. 109-112.
[11] Sur wne propriété des fonctions de M. Hamel, ibidem 5 (1924), p. 334.
[12] Sur Phypothése du continu, ibidem 5 (1924), p. 117-187.
[13] Sur wn probleme condwisant a wn ensemble non meswrable, ibidem 10 (1927),
Dep Laie lis.Oe ;
[14] Sobre la correspondencia entre los puntos de un seqmento y los de wn cuadrado,
Rev. Mat. Hisp.-Amer.-7 (1927), p. 193-197.
[15] Sur les familles croissantes de sous-ensembles d’un ensemble dénombrable, Enseign.
Math. 30 (1931), p. 240-242.
[16] Sur les ensembles de points qu'on sait définir effectivement, Verh. Intern. Math.
Kongres, Ziirich 1932, p. 280-287.
[17] Sur les ensembles de la méme puissance qui ne sont pas effectivement de la méme
puissance, Fund. Math. 18 (1932). p. 189-192.
[18] Généralisation @un théoréme de Cantor concernant les ensembles ordonnés deé-
nombrables, ibidem 18 (1932), p. 280-284.
[19] Sur un théoreéme de recouvrement dans la théorie générale des ensembles, ibidem
20 (1933), p. 214-220.
[20] Sur les ensembles partout de deuxiéme categorie, ibidem 22 (1934), p. 1-3.
[21] Remarques sur les fonctions de plusieurs variables réelles, Prace Mat.-Fiz. 41
(1934), p. 171-175.
[22] C. R. Soc. Se. Lett. Varsovie, Cl. III. 27 (1934), p. 128-129.
-[23] Hypothéese du continu, Warszawa-Lwow 1934.
[241 Sur les suites transfinies multiples wniverselles, Fund. Math. 27 (1936), p. 1-9.
[25] Un théoréme sur les fonctions définies dans les ensembles infinis queleonques,
Bull. Intern. Acad. Pol. (1936), p. 433-435.
[26] Sur une décomposition effective @ensembles, Fund. Math. 29 (1937), p. 1-4.
[27] Sur wne décomposition effective Vensembles, ibidem 29 (1937), p. 1-4.
[28] Sur le rapport de la propriété (C) a la théorie des ensembles, ibidem 29 (1957), p. 118.
[29] Sur une propriété des ensembles ordonnés, Pontificia Acad. Se. 4 (1940),
p. 207-208.
[30] Sur une suite transfinie Vensemble de nombres naturels, Fund. Math. 33 (1945),
Oo Mall,
[31] Sur Vimplication (2m < 2n)>(m <n) pour les nombres cardinauc, Fund.
Math. 34 (1947), p. 148-154.
[32] Démonstration de Végalité 2™—m = 2™ nour les nombres cardinaux transfinis,
ibidem 34 (1947), p. 113-118.
[33] Remarque sur les ensembles des nombres ordinawx de classes I et II, Revista
de Ciencias XLI, N° 429, p. 289-296.
[34] Sur la difference de deux nombres cardinawxz, ibidem 34 (1947), p. 119-129.
[35] Sur la décomposition das ensembles en pairs, Proc. Benares Math. Soc. 8 (1946), p.31.
478 : Bibliography

[36] Dhypothése généralisée du continu et Vaxiome du choix, Fund. Math. 34 (1947),

15 WSs
[37] Sur la différence de deux nombres cardinaws, ibidem 34 (1947), p. 119-126.
[38] Deux théorémes sur les familles de transformations, ibidem 34 (1947), p. 30-33.
[39] Un théoréme sur les puissances des ensembles, ibidem 34 (1947), p. 72-74.
[40] Les correspondances multivoques et Vaxriome du choix, ibidem 34 (1947),
p. 39-44.
[41] Sur wne proposition équivalente a Vaxiome du choix. Actas Acad. Nat. Cien.
de Lima 11 (1946), p. 111-112.
[42] C. R. Soc. Se. Lett. Varsovie (1947), p. 1-3.
[43] Sur quelques propriétés du nombre 2*°, Mathematica 23 (1947/48), p. 60-64.
[44] Sur les translations des ensembles linéaires, Fund. Math. 35 (1948), p. 159-164.
[45] Sur la division des types ordinaux, ibidem 35 (1948), p. 1-12.
[46] Sur les ensembles presque contenus les wns dans les autres, ibidem 35 (1948),
p. 141-150.
[47] Sur wne propriété des ensembles ordonnés, ibidem 36 (1949), p. 56-67.
[48] Sur les séries infinies de nombres ordinawr, ibidem 36 (1949), p. 248-253.
[49] C. R. Soc. Se. Lett. Varsovie 42 (1949), p. 312.
[50] Sur quelques propositions qui entrainent existence des ensembles non mesurables,
ibidem 42 (1949), p. 36-40.
[51] Sur les diviseurs de types ordinaux, Az Elz6 Magyar Mat. Kongr., Budapest
1952, p. 397-399.
[52] Solution de Véquation w2 = &° pour les nombres ordinaur, Acta. Se. Math. 12B,
Szeged 1950, p. 49-50.
[53] Sur les types W@ordre des ensembles linéaires, Fund. Math. 37 (1950), p. 253
- 264.
[54] Sur wn type ordinal dénombrable qui a une infinite indénombrable de diviseurs
gauches, ibidem 37 (1950), p. 206-208.
[55] Le dernier théeoréme de Fermat pour les nombres ordinaux, ibidem 37 (1950),
p. 201-205.
156] Sur les produits infinis de nombres ordinaux, C. R. Soc. Se. ILL (1950), p. 20-24.
[57] Legons sur les nombres transfinis, Paris 1928 (2-nd ed. 1950).
[58] Algébre des ensembles, Warszawa-Wroclaw 1951.
[59] Sur quelques propositions concernant la puissance du continu, Fund. Math.
38 (1951), p. 1-13.
[60] Sur les fonctions continues @une variable ordinale, ibidem 38 (1951), p. 204-208.
[61] C. R. Acad, Sc. Paris 232 (1951).
[62] Sur le probleme de M. J. Novak, Czechoslovak Math. J. 1. 76 (1951), p. 97-101.
[63] Sur wne propriété des ensembles plans équivalente a Vhypothése du continu,
Bull. Soc. R. Se. Liege 1951.
[64] Derniéres recherches et problémes de la Théorie des Hnsembles, Rend. Math,
et delle sue appl. V. 10; Fase. 3-4 (11951), p. 1-11.
[65] Sur une propriété paradowale de Vespace a trois dimensions équivalente a Vhypo-
these dw continu, Rend. Cire. Mat. Palermo II. 1 (1952), p. 7-10.
[66] Sur une proposition équivalente a Venistence @un ensemble de nombres réels
de puissance aleph 1, Bull. Acad. Pol. Sc. 2 (1954), p. 53-54.
[67] L’axiome du choix pour les ensembles finis, Le Matem. 10 (1955), p. 98.
[68] Bull. de la Section Scient. Acad. Roum., 18-iéme année, no 1-2.
[69] Sur les types Wordre de puissance du continu, Revista Cien. 48, p. 305-307.
[70] Sur la décomposition des ensembles en sows-ensembles presque disjoints, Mathe-
matica 14 (1938), p. 15.
 Bibliography 479

[71] Sur Véquation 2 = 7?+1 pour les nombres ordinaus transfinis, Fund. Math.
43 (1956), p. 1-2.
Sierpinski, W. et Davis, A. C.
[1] Sur les types dordre distinets dont les carrés sont égaux, C. R. Acad. Sc. Paris 235
(1952), p. 850.

Sierpinski, W. et Tarski, A.
[1] Sur une propriété caractéristique des nombres inaccessibles, Fund. Math. 15
(1930), p. 292.
Skolem, S.
[1] Logisch-kombinatorische Untersuchung, Skrifer utgit av Videnskap. Kristiania
I KI. (1920), p. 4.
Stackel, P.
[1] Zu H. Webers elementarer Mengenlehre, Jahresber. deutsch. Math. Ver. 16
(1907), p. 425-428.
Steckel, S.
[1] Remarques sur classe Wensembles ordonnés, Fund. Math. 11 (1928), p. 285-287.
Souslin, M.
[1] Fund. Math. 1 (1920), p. 223, Probléme 3.
Szele, T.
[1] On Zorn’s Lemma, Publ. Math. 1 (1950), p. 254-256.
Szmielew, W.
[1] On choices from finite sets, Fund. Math. 34 (1946), ps 75-80.

Szpilrajn, E.
[1] Sur
Vextension de Vordre partiel, Fund. Math. 16 (1930), p. 386-390.
[2] The characteristic function of a sequence of sets and some of tts applications,
ibidem 31 (1938), p. 207-223.

Tarski, A.
[1] Sur quelques théoremes qui équivalent a Vaxiome de choiz, Fund. Math. 5
p. 147-154.
[2] Sur les ensembles finis, ibidem 6 (1924), p. 45-95.
[3] Quelques théoremes sur les alephs, ibidem 7 (1925), p. 1-13.
[4] Remarques concernant Varithmeéetique des nombres cardinaux, Ann. Soc. Pol.
Math. 5 (1926), p. 106-107.
[5] C. R. Soc. Se. et Lett. Varsovie III. 19 (1926), p. 308-309.
[6] Sur les ensembles définissables des nombres réels I, Fund. Math. 17 (1931),
p- 210-240.
[7] Der Wahrhettsbegriff in den Sprachen der deductiven Disziplinen, Anzeiger Akad.
Wien 69 (1932), p. 23-25.
8] Pojecie prawdy w jezykach nauk dedukeyjnych, Prace Tow. Nauk. Warsz.
Ii. 34 (1933).
[9] Der Wahrheitsbegriff in formalisierten Sprachen, Stud. Phil. 1 (1935), p. 261-406,
‘ 110] C. R. Soc. Se. et Lett. Varsovie 30 (1937).
[11] Axiomatic and algebraic aspects of two theorems on sums of cardinals, Fund.
Math. 35 (1948), p. 79-104.
480 Bibliography as.

[12] Cancellation laws in the arithmetic of cardinals, ibidem 36 (1949), p. 77-92.

[13] Cardinal algebras, New York 1949.
[14] Theorems on the existence of successors of cardinals and the axiom of choice,
Indag. Math. 16 (1954), p. 26-32,
[15] Sur la décomposition des ensembles en sous-ensembles presque disjoints, Fund.
Math. 12 (1928), p. 186-205.
[16] Sur les classes d’ensembles closes par rapport a certaines opérations élémentaires,
ibidem 16 (1930), p. 181-304.
[17] Ideale in vollstandigen Mengenkorpern I, ibidem 32 (1939), p. 45-63.
[18] Uber wnerreichbare Kardinalzahlen, ibidem 30 (1938), p. 68-89.
Teichmiiller, O.
[1] Braucht der Algebraiker das Auswahlaxiom?, Deutsche Math. 4 (1939),
p. 567-577.
Tietze, H.
[1] Beittrdge zur allgemeinen Topologie I, Math. Ann. 88 (1923), p. 290-312.
Ulam, S.
[1] Zur Masstheorie in der allgemeinen Mengenlehre, Fund. Math. 16 (1930),
p. 140-150.
[2] Uber gewisse Zerlegungen von Mengen, ibidem 20 (1933), p. 221-223.
Vaughan, H. E.
[1] Well-ordered subsets and maximal members of ordered sets, Pacific J. Math.
2 (1952), p. 407.
Vaugt, R. L.
[1] On the equivalence of the axiom of choice and a maximal principle, Bull. Amer.
Math. Soc. 58 (1952), p. 66.
Wakiulicz, A.
[1] Sur la somme @un nombre fini de nombres ordinaux, Fund. Math. 36 (1949),
p. 254-266. :
[2] C. R. Soc. Sc. et Lett. Varsovie III. 13 (1949), p. 23.
[3] Correction au travail ,,Sur les sommes d’un nombre fini de nombres ordinaua
de A. Wakulicz, Fund. Math. 38 (1951), p. 239.
Wallace, A. D.
[1] A substitute for the axiom of choice, Bull. Amer. Math. Soc. 50 (1944), p. 278.
Weber, H.
[1] Hlementare Mengenlehre, Jahresber. deutsch. Math. Ver. 16 (1907).
Witt, E.
[1] Beweisstudien zum Satz von M. Zorn, Math. Nachr. 4 (1951), p. 434-438.
Wrinch, D. H.
[1] On medial cardinals, Amer. J. Math. 45 (1923), p. 87-92.
Zermelo, E.
[1] Beweis dass jede Menge wohlgeordnet werden kann, Math. Ann. 59 (1904),
p. 514-516.
[2] Untersuchungen tiber die Grundlagen der Mengenlehre, ibidem 65 (1908),
p. 261-281.
 Bibliography 481

[3] Uber die Grundlagen der Arithmetik, Atti de! TV Congr. Intern. dei Math. 2. J,
Roma 1909.
[4] Sur les ensembles finis et le principe de Vinduction complete, Acta Math. 32
(1909), p. 185-193.
[5] Uber Grenzzahlen und Mengenbereiche, Fund. Math. 16 (1930), p. 29-47.
[6] Untersuchungen tiber die Grundlagen der Mengenlehre I, Math. Ann. 65 (1907),
p. 261-281.
Zorn, M.
{1] A remark on method in transfinite algebra, Bull. Amer. Math. Soc. 41 (1935),
p- 667-670.

Cardinal and ordinal numbers | 31

 INDEX

Alephs, 389. Normal form, 320.

Algebraic number, 42. Number, Indecomposable, 303; Ratio-
Axiom of choice, 88; Restricted, 95. nally independent, 444.
Basis of Hamel, 441. Ordered sequences, 16.
Cantor-Bernstein theorem, 32. Postulate of Bertrand, 334.
Cardinal numbers, 132; Accessible, 406;
Power, of sets, 80; of an ordinal number,
Inaceessible 405.
366.
Cartesian (combinatorical) product, 19, 73.
Prime factor (indecomposable number), 303.
Character, of an element, 459; Left-side,
Principle of first differences, 40.
Right-side, 459.
Product, Cartesian, 19; of sets, 11.
Commutative ordinal numbers, 343.
Complement, 15. Root of an ordinal number, 342.
Components, 9.
Second class of ordinal numbers, 366.
Continuum Hypothesis, 83; Generalized, 85.
Sets, 5; Coinitial, 459; Confinal, 458;
Correspondence, 16; m-to-n, 126; One-to-
Denumerable, countable, 34; Disjoint,
-one, 17.
13; Effectively denumerable, 34;
Critical numbers of a function, 318.
Effectively equivalent, 25; LEffecti-
De Morgan’s formulae, 15. vely non-denumerable, 44; Empty, 6;
. . ¢ w a
Difference, of cardinal numbers, 158; of Equivalent, 22; Equal, 6; Essentially
sets, 10. different, 449; Ny» 463; Finite, 23, 51;
Disjoint, sets, 13; Sum, 13. Finite in the sense of Dedekind, 50;
Inductive, 51, 431; Infinite, 23; Infi-
Epsilon number, 324. nite in the sense of Dedekind, 49;
Families of sets, 7. Non-denumerable, 43; of the power of
First class of ordinal numbers, 366. the continuum, 53.
Function, 16; Characteristic, 19; Inverse, Single-valued function, 16.
18; Invertible, 17. Subset (part), of a set, 8; Proper, 8.
Sum, of an infinite series, 173; of sets, 9.
General principle of choice, 94. Superset, 8.
Inclusion, 8.
Teichmiiller’s principle, 130.
Inductive sets, 431.
Theorem, of Cantor-Bernstein, 32; of
Initial nwmbers, 389.
Denjoy, 37; of J. Kénig, 121; of
Interior point, 37.
Lindenbaum, 41, 426; of P. Erddés,
Interval, 37.
125; of R. Vaught, 433; on tricho-
Length of an interval, 37. tomy, 412; Well-ordering, of Zermelo,
Maximal element, 431. 411.

Natural product and natural sum_ of Upper bound, 431.

ordinal numbers, 364. Well-ordering theorem of Zermelo, 411.
 CONTENTS
Page
Preface .

Chapter I. Sets and elementary set operations

Sets. .
. Elements of a ae
. Symbols « and ¢ . Or
Ol
ON

Set consisting of one ererien Or

. The empty set. o>)

. Equality of sets
. Sets of sets
. Subset of a set.
. Sum of sets .
. Difference of sets. .
. Product of sets.
. Disjoint sums
. Complement of a set .
. Ordered pairs :
. Correspondence. Tinneuion :
. One-to-one correspondence. .
. Cartesian product of sets
. Exponentiation of sets.

Chapter II. Equivalent sets

. Equivalent sets. Relation ~. .

. Finite and infinite sets . wo bo
be

. Fundamental properties of the rel abion ~.

. Effectively equivalent sets. >OU

. Various theorems on the equivalence of Be OD

[SNe
NwWNNWW
& . The Cantor-Bernstein
aor
wr Theorem . w =)

Chapter IfI. Denumerable and non-denumerable sets

. Denumerable and effectively denumerable sets. : 34

. Effective denumerability of the set of all rational aa bees 36
. Effective denumerability of the infinite set of non-overlapping atonal ‘ 37
. Effective denumerability of the set of all finite sequences of rational numbers 39
. Effective denumerability of the set of all algebraic numbers . 42
He
Dor
WW . Non-denumerable sets . 43
31*
484 Contents

Page

7. Properties of sets containing denumerable subsets. ........... «45

8, Setsuniinitesiny the sense of Dedekimdl arenes er caret niente wren tor a
9. Various definitions of finite sets... . Ss SRLS OU
10. Denumerability of the Cartesian product of te donumerania ee SSAA ae ele

Chapter IV. Sets of the power of the continuum

1. Sets of the power of the continuum and sets effectively of the power of the
CONMULMUTLUTE sane ne ; PL NE Rae CHe NAL Ot Pei ec 53
2. Non-denumerability is ae ee of Sal Stan ee eths eines 53
3. Removing a denumerable set from a set of the power of he Rone Ah geen
4. Set of real numbers of an arbitrary interval ........4:~... 5. 56
5. Sum of two sets of the power of the continuum .. . kos
6. Cartesian product of a denumerable set and a set of the power of tie convene 59
7. Set of all infinite sequences of natural numbers... . te 5 ee oO
°8. Cartesian product of two sets of the power of the count Bae Ar 61
9. Impossibility of a continuous (1-1) mapping of a plane on a Siralewe a 66
LOSContinuoustcurye: fillies up, aesquares.: s-y -nle) -nie olen ce) ten en SS
11. Set of all infinite sequences of real numbers ... . oats "a SE deen aeRO
12. Continuous curve filling up a denumerably iaenone) ene bE) de aw Re oe eT
13. Set of all continuous functions .. . 75
14. Decomposition of a set of natural Panbers tc a; coatnnimn of mirtioet disjoueh
SOLS u.hye: rove deelia RSe Mennee ease eeVga Lo) My, Twn eh roa ericfetes Sue ea 0 rr

Chapter V. Comparing the power of sets

1. Sets of less and greater power... . 80

2. Sets of greater power than finite sets avid SSSA Nts eae Hypocneas ee
the continuum. . . oi lg roe in he RES
3. Cantor’s theorem on aes si ‘of all Bibects Ae a vee eet gat Ae See oS
4.) Generalized) Continuum’) Hiypothesisi. ee. sre) connie eect eenemEEnECS Ey
Oy Horming sets ol ever prea ter powers: .ae ine) encircle eeeene yr ene an 85

Chapter VI. Axiom of choice

1. The axiom of choice. Controversy about itt... 2 2) ee eS

2 Lhe axiom Of Choice tors aatinibe Seb, OL SOLSH sepa e mein ntcme etn atCnnen t n an anno?
2a. The axiom of choice for an infinite sequence of sets. ........: ~ «298
3. Hilbert’s axiom... . Tee e OR LEA ats Oe a ohio) ces ga OR
4, General principle of bcicel ee ae er ert, hm Ee Gly sh yo We
5. Axiom of choice for finite sets .... : ! 97
6. Examples of cases where we are able or AGE able td Praleeean woffedtire enor 105
i. Applications of the axiom oLschoieo; wes os. eee one ra nen e n nn een TC)
8, The m-to-n correspondence: 2 yen. ol hy ce wal ren eee ne ae To?
9, Dependent choices -. 2) 2)! disease Nap oe clta Ace Ah Ree TE
 Contents

Chapter VII. Cardinal numbers and operations on them

. Cardinal numbers
. Sum of cardinal numbers
. Product of two cardinal numbers
Rm
=wh. Exponentiation of cardinal numbers ‘
. Power of the set of all subsets of a given Neate

Chapter VIII. Inequalities for cardinal numbers

. Definition of an inequality between two cardinal numbers

. Transitivity of the relation of inequality Addition of inequalities .
. Exponentiation of inequalities for cardinal numbers .
. Relation
Bmwhdo m< «xn.

Chapter IX. Difference of cardinal numbers

. Theorem of -A. Tarski and F. Bernstein

. Theorem on increasing the diminuend
. Theorem on increasing the subtrahend ,
. Difference in which the subtrahend is a natural Huber ;
. Proof of formula 2%™—m = 2" for m>k, pone the aid of fhou axiom “BF
choice .
. Quotient of Rater namnbore

Chapter X. Infinite series and infinite products of cardinal numbers

. Sum of an infinite series of cardinal numbers. .

i Properties of an infinite series of cardinal numbers
. Examples .of infinite series of cardinal numbers .
. Sum of an arbitrary set of cardinal numbers
Infinite product of cardinal numbers .
. Properties of infinite products of cardinal etn Bere. Honiples
. Theorem of J. Konig . :
. Product of an arbitrary set of eordigal “hin

Chapter XI. Ordered sets

. Ordered sets . as
. Partially ordered sets . pao
. Lattices . . . bau o kite bsp. M lade oa uth
. Similarity of oaks : :
oO
Rm
WD . First and last element of an brdered aot Guise i;umps. Denes nad continuity
of a set eo
Finite ordered sets
. Sets of type w .
. Sets of type 7
. Dense
oor ordered sets as ibsets of Peenupad baie
10. Séts of type A
186 Contents

Page

Chapter XII. Order types and operations on them

. Order types
ed. Sum of two praered Cane
. Product of two order types .
mw. Sum of an infinite series of order oes : Ww
ve
wwSOW
by

. Power of the set of all denumerable order types mal bodW

nelee
bv
ww

- Power of the set of all order types of the power of the pour : 240
. Sum of an arbitrary ordered set of order types . tonN w
. Infinite products of order types : 244
oN. Segments and remainders of order types
coro 247
10. Divisors of order types 250
ee Comparison of order types. . 254

Chapter XIII. Well-ordered sets

. Well-ordered sets . 258

The principle of transfinite idonon ; 259
. Induction for ordered sets . . : 261
. Similar mapping of a well-ordered St on ie Prince : 264
as . Properties
om
i
Ww of segments of well-ordered sets 265

Chapter XIV. Ordinal numbers

. Ordinal numbers. Ordinal numbers as indices of the elements of well-ordered

sets . 269
. Sets of otdinal aurbor : 270,
. Sum of ordinal numbers. . 271
. Properties of the sum of ordinal ernie eNamabore of Hite ieae ia of che
2-nd kind .
. Remainders of ordinal Camnern
. Prime components
. Transfinite sequences of erteinal Rambor and there fits BO
7

S
>»
I
a
Owat
. Infinite series of ordinal numbers and their sums
. Product of ordinal numbers .
. Properties of the product of ordinal ber ; ©
bo

. Theorem on the division of ordinal numbers Ww

ww
www
mow
bv

. Divisors of ordinal numbers .

. Prime factors of ordinal numbers
. Certain properties of prime components.
. Exponentiation of ordinal numbers. .
. Definitions by transfinite induction.
. Transfinite products of ordinal numbers
. Properties of the powers of ordinal numbers :
. The power w®. Normal expansions of ordinal arbors :
. Epsilon Bans ; :
. Applications of the HORE ou ; :
. Determination of all cardinal numbers nee are prime Sinetors
. Expanding ordinal numbers into prime factors
 Jontents

24. Roots of ordinal numbers . ;

25. On ordinal numbers commutative catia eet to fidinon :
26. On ordinal numbers commutative with respect to multiplication
27. On the equation a? = 6% for ordinal numbers
28. Natural sum and natural product of ordinal numbers
29. Exponentiation of order types

Chapter XV. Number classes and alephs

Numbers of the 1-st and of the 2-nd class.
Cardinal number s,
, = 2*° hypothesis. :
Properties of ordinal Arner of rie 2.sl cee :
Transfinite induction for numbers of the 1-st class and of ae 2. Hd lass
Convergence and limit of transfinite sequences of real numbers .
aes
Pi Initial numbers, alephs and their notation
Formula 8eee =, and conclusions from it ; : sets
A proposition of elementary geometry, equivalent to thie éottiniune ie
thesis le Ea es Be
— S2%. Difference of alephs,eave ma Produ of Weanatinite sequences of successive
alephs . : inn ei erRes Pt can
= — . Limit of ectnite sequences of ral eee Regular and singular initial
numbers. Inaccessible alephs . 4()2

Chapter XVI. Zermelo’s theorem and other theorems equivalent to the axiom of chcice

1. Equivalence of the axiom od choice, of Zermelo’s theorem and of the problem

of trichotomy 5 b oe ; Sa 407
2. Various theorems on te dinal members Maeralent to ihe axiom of SRG 414
3. A. Lindenbaum’s theorem . : 426
4. Equivalence of the axiom of eieice Re abe ahunoae A ise a) Teichmuiller 427
5. Inference of the axiom of choice from the generalized continuum hypothesis 434

Chapter XVII. Applications of Zermelo’s theorem

Hamel’s basis eens Pu

Be : 440
Plane set having Satie te 0 Points in common
vat every Reais ine 446
Decomposition of an infinite set of power m into more than m almost disjoint
were

sets . Pe gs F 448
4. Some AN sece,on Pratuies of acces. of sats re given power . 450
5. The power of the set of all order types of a given power 457
6. Applications of Zermelo’s theorem to the theory of ordered sets 458

Appendix 468
Bibliography 469
Index 482
 iF . 1 iene
nie. bike f ni ed
; A a iy} | ‘
' ra) wih te
hive. th 3
1 tt erh

eg ney |

int Saw ae

Sie Ue ey, ant

Den veal Fa i
1
“yy
4 Ni :
; CE
a
SS
vm
c-'ee
7eve
~~
a
nn
.peel
=.=‘
 PUBLICATIONS DE L’INSTITUT MATHEMATIQUE
DE L’ACADEMIE POLONAISE DES SCIENCES

JOURNAUX
FUNDAMENTA MATHEMATICAE I-XLV.1.
STUDIA MATHEMATICA I-XVI.
COLLOQUIUM MATHEMATICUM I-V.1.
ZASTOSOWANIA MATEMATYKI I-III.
ROZPRAWY MATEMATYCZNE I-XIV.
ANNALES POLONICI MATHEMATICI I-IV.1.

MONOGRAFIE MATEMATYCZNE

Mostowski, Logika matematyczna, 1948, p. VIII+388.

. Sierpinski, Teoria liczb, 3-6me éd., 1950, p. VIII+544.
Kuratowski, Topologie I, 3-éme éd., 1957, p, XI+450.
Kuratowski, Topologie II, 2-éme éd., 1952, p. VIII+444.
F2C3 . Sierpinski, Algébre des Ensembles, 1951, p. 202.
Banach, Mechanics, 1951, p. 1V+546.
aie
-Kuratowski i A. Mostowski, Teoria mnogoégci, 1952,
p. 1X+311.
S. Saks and A. Zygmund, Analytic Functions, 1953, p. VIII+452.
J. Mikusinski, Rachunek operatoréw, 2-éme éd., 1957, p. I1+374
W. Slebodzinski, Formes extérieures et leurs applications, I, 1954,
p. VI+156.
8S. Mazurkiewicz, Podstawy rachunku prawdopodobienstwa, 1956,
p. 270.
A. Walfisz, Gitterpunkte in mehrdimensionalen Kugeln, 1957, p. 471.
W. Sierpinski, Cardinal and ordinal numbers, 1958, p. [V+ 480.
35. R. Sikorski, Funkcje rzeczywiste, I, 1958, p. 536.
K. Maurin, Metody przestrzeni Hilberta (sous presse).

Stownik statystyczny rosyjsko-polski i angielsko-polski, 1952, p. 20.