THE BANACH TARSKI PARADOX

By
Naba Kumar Bhattacharya
and
Pratit Goswami
under the supervision of
Prof. Rabiul Islam
 What is a paradox?
 MATHEMATICAL PARADOXES ARE STATEMENTS THAT RUN COUNTER TO ONE’S
INTUITION, SOMETIMES IN SIMPLE PLAYFUL WAYS AND SOMETIMES IN EXTREMELY
ESOTERIC AND PROFOUND WAYS.
 THE WORD “PARADOX” ORIGINATED FROM THE GREEK WORD “PARADOXON”
WHICH REFERS TO A STATEMENT CONTRARY TO AN ACCEPTED OPINION.
 FROM ITS DICTIONARY MEANING, THE WORD “PARADOX” REFERS TO HIGHLY
COUNTER INTUITIVE STATEMENTS THAT ORIGINATES FROM THE EXSITING SET OF
AXIOMS IN A PARTICUAR TOPIC.
 PARADOXES ARE USEFUL SINCE SOMETIMES THEY REFLECT SOME VOID IN A
PARTICULAR SET OF AXIOMS TAKEN AS A BASIC FOR A PARTICULAR SUBJECT
WHEREAS SOMETIMES THEY REFER SHOW US SOME TRUE RESULTS ARISING FROM
THE GIVEN SET OF AXIOMS WHICH ARE SEEMINGLY IMPOSSIBLE FROM A LAYMAN’S
POINT OF VIEW.
A mother has
two sons, but she
has only one
burger.
How can she get
another one
without buying?
Can she do some
magic?
 The Magic Begins!
In Banach-Tarsky paradox we can partition a solid
spherical ball in IR3 and reassemble them to form two
solid balls identical to the original one. This process
seems to be interesting as well as unbelievable in the
first look as if it was really possible in our real world;
then we could make our burgers as spherical shaped
and would have made it double whenever we wish!!!
But sadly the sense of equality in mathematics does
not coincide with our common sense!
 Let’s set out the journey to the world of apparent impossibilities!!!

Definition:- Let, G be a group acting on a set X and

suppose E be a subset of X. E is
said to be G-paradoxical if for some m,n
there exist g1,g2,..,gm and h1,h2,..,hn in G
and pairwise disjoint A1,A2,..,Am and
B1,B2,...,Bn subsets of E such that E
m ሺ ሻ n
=!!!!!!!‫ڂ‬
i=1 ig A i = j=1 hj (Bj ).
!!!!!!!‫ڂ‬
Mathematically, the Banach-Tarski Paradox tells that any ball in IR3 can
be paradoxically decomposed by rotations and translations.
Before starting formal discussions, we must take note of some more
definitions and results.

Definition:- Let G act on a set X, and let A,B be subsets of X. A

and B are said to be G-equidecomposable if A and B
can be partitioned into A1,A2,...,An and
B1,B2,....,Bn respectively such that each Ai is congruent
to Bi (i.e., there is a gi in G such that
gi(Ai)=Bi).
We write A ~G B, or A ~ B, when the ambient group is
clear.
This means that A can be partitioned into a finite
number of subsets and reassembled to form B. This
reassembling depends on the group G and its action on
X.
Properties:
1.The relation on P(X), the set containing all the subsets of X, defined by "A is -related to B iff A~G B", where
A,B are subsets of X is an equivalence relation on P(X).

2. If E~G F and E is paradoxical then F is also paradoxical.

It is necessary to introduce one more definition in this regard:

Definition : For subsets E and F of X, to say a bijection f of E onto F is a puzzle map, or a G-puzzle map
means that there is a partition {Ei:i=1(1)n} of E and g1,...,gn in G such that f(Ei)=gi(Ei) for all i.

A characterisation of G-equidecompasble subsets of X is as follows:

Subsets E and F of X are G-equidecomposable iff there is a G-puzzle map taking E onto F.

Again A characterisation of G-paradoxical subsets of X can be given by:

A subset E of X is said to be paradoxical if and only if there are disjoint subsets A and B of E so that A~E~B.
Theorem 1: S1 \{pt} is equidecomposable to S1. By omission of the group we understand it to be the isometry
group of R2, and we identify S1 with .
Proof: Consider R2 identified with C, the complex plane. Let the pt be 1 = ei0.
Let A = {ein : n ∈ {1, 2, 3, ...}}, and let B = (S1 \ {pt}) \ A be everything else. The points ein are unique, since 2π
is irrational. Then, leaving B fixed, rotate A by 1 radian— i.e., use the isometry ρ(z) = e -iz. This rotation sends each
point “back” one radian, by which we obtain a complete circle.

We made an infinite set namely, A and shift them to the predecessors in terms of the mapping preimage in N and
make the void filled up again to get the new circle. Only idea used here is that A is infinite, the rotation of 1 radian
clockwise sends every element of A say ein to ei(n-1) which sends ei to 1 and the void is filled up. And in the more
general case we rotate the circle to align it in a way that the missing point coincides with the point 1 itself and then
the same proof as before comes into picture.
The above construction gives us an interesting idea for the construction of paradoxes. Here we considered the
image of a point (namely, 1) under the subsemigroup generated by f -1 (that is, only positive rotations by 1 radian).
Because the semigroup was free, we could use the inverse rotation (obtained from the group containing this
semigroup) to "shift" everything back.
 
Paradoxical Groups:
 We note that the previous examples of paradoxes involved finding one in a group or a subsemigroup and then in
some manner "lifting" the paradox to set on which it acts. Thus it is natural to study paradoxical groups, where
the group acts on itself by left multiplication.

Definition : Let, G be a group acting on a set X. An action of a group G on a set X is paradoxical if X is G-

paradoxical under that action. The group itself is called paradoxical if the action of the group on itself by left
multiplication is paradoxical.

The primary example of a paradoxical group is the free group on two generators (sometimes called a free group
of rank 2). Recall that a free group F on two generators a, b is the group of all finite words in a ±1,b±1 with no
adjacent pairs of inverse letters, and the operation is juxtaposition with removal of possibly any adjacent pairs of
inverse letters.
Theorem 2: The free group F on two generators is F-paradoxical.
 
Proof: Let B() = set of all words beginning on the left with x, where x may be .
Then F = , where all sets are pair-wise disjoint.
But F = and F = , which shows that F is indeed paradoxical. Hence. Proved.
 
The underlying mathematical beauty here is we form all the words in F by concatenating finite terms made of
written in a series and reduced to a compact form by cancelling adjacent inverse terms (if any). So, The words in F
starting with x-1 (here x is is just the words we get by putting an x-1 in front of the words starting with every other
symbols except x. So, if we apply x again in front of those words by reducing them we get all those finite strings
which starts with every symbols excluding x (including 1). So, this set of words taken union with the words
starting with x in particular forms the set of all words in F which justifies the composition.
To ensure that the paradox "lifts" to disjoint well-defined sets, we need a condition on G's action, namely that no
element but the identity fixes any points of the set.

Formally, by a free action of G on a set X we mean that, the equation gx=x with g in G and x in X means g=1
under the action of G on X. Sometimes we refer it as an action of G on X without nontrivial fixed points.

In this regard we note some more results, which are indeed an important one in establishing Banach-Tarsky
paradox, is based upon the free actions of a group G on a set X and it associate paradoxical groups with sets on
which it acts.

1. If a group a G is paradoxical then any free action of G on a set X is paradoxical. more precisely, if
a paradoxical group G acts on a set X without any nontrivial fixed points. Then X is G-paradoxical.

2. If a group G contains H, a paradoxical subgroup then G is paradoxical.

3. Any group G with a free subgroup of rank 2 is paradoxical.

4. A stronger converse of result 1 is also true: If X is G-paradoxical then G is also paradoxical.

THE HAUSDORFF PARADOX :
In 1914, F. Hausdorff constructed the following paradox:
Except for a countable number of points, S2 may be partitioned into three subsets A,B,C such that A B C (BC),
where denotes congruence (not to be confused with ~ which denotes equidecomposability)[Here congruence
means that one part can be formed from another part by using an isometry or a sequence of finitely many
isometries on it].

Since we are primarily interested in proving the Banach-Tarski Paradox , it will suffice to find two rotations
which generate a free subgroup of rank 2 of SO3. This will enable us to obtain a paradoxical decomposition by
lifting to S2, the technology used here is to find a free subgroup of rank 2 and except for a countably many points
(negligible in terms of measure) this free subgroup freely acts on the remaining points on S 2 to give the required
paradoxical decomposition.

However we shall not go into the proof of the statement written above in the name of Hausdorff paradox but we
will prove the existence of the free group and find out the countable subset, say, D as it will be required to prove
Banach-Tarsky paradox.
 Theorem 3: There exist two independent rotations which fix the origin in R3. Hence, SO3 contains a free
subgroup of rank 2.

As it turns out, "most" pairs of rotations are independent, and in the following proof arccos(3/5) can be replaced
by arccos(r) where r is any rational other than, 0,-1,1,-1/2,1/2.

Proof: We shall let be counter-clockwise rotations about the z axis and x axis (z axis is the axis passing through
(0,0,1) and x axis is the axis passing through (1,0,0) in R3, respectively, each through the angle of arccos(3/5).
It can be shown that no nontrivial reduced word in and in the identity.
For if there is such a word then we may conjugate (if necessary) to obtain a word w ending in which equals the
identity.
Then, we can establish that, w(1,0,0) is of the form (a,b,c)/5 k where, a, b, c are integers and b is not divisible by
5.Which shows that, w(1,0,0) cannot be (1,0,0) and hence gives a contradiction to the assumption that w is the
identity.
Thus, there is no nontrivial relationships between the generators and of the subgroup R=<,of SO 3. So, by a
homomorphism f defined on F2, the free group generated by two symbols, to R which sends the generators of F 2
to and respectively. The detailed proof of the last statement guarantees the injectivity of this homomorphism and
hence, F2 can be embedded into SO3, which proves the theorem.
Now, we prepare the final step in the staircase to enter the floor of BANACH-TARSKY Paradox

We see in the last proof that the subgroup R is isomorphic to F2, the free subgroup of rank 2 and the results
mentioned before ensures it to be paradoxical.
Now, every g in R, our free subgroup fixed two points on the sphere, viz, the intersections points of the axis of
rotation to the sphere. Let, D be the set of all such points on S2 fixed by some rotation in R. Since R is finitely
generated group, hence it is countable, and each point fixes two points on S2. So, D is also countable. And note
that R acts freely on S2\D and R is a paradoxical group.
This motivates the following:

Theorem (Hausdorff Paradox) 4: There is a countable set D such that S2\D is SO3- paradoxical.

Proof: D is the countable set described before. Since R acts on S2\D freely and R is a paradoxical group, so, S2\D
is R-paradoxical. And R< SO3 , so, from the definition of G-paradoxical we can say, S2\D is SO3-paradoxical.
Hence proved.
THE BANACH TARSKI PARADOX (in it’s weak form):
We shall soon show, however, this countable set D does not matter much.

The classical Banach-Tarski paradox amounts to a decomposition of a unit ball into finitely many pieces, rearranging this pieces into unit
balls. Since the group of rotations preserves the origin, it would not be sufficient to obtain Banach-Tarski paradox. For this purpose we add
translations into the picture.

Theorem 5: S2 and S2\D are SO3-equidecomposable.

Proof: Choose an axis of rotation which doesn't fix any points of D. Then consider the set of all rotations around this axis for which some
integer multiple of it sends a point of D to another point of D. This is a countable set. But there are uncountably many possible angles; pick
one, say .

Let ρ be a rotation by θ around L, then no multiple of ρ will send a point of D to another point of D and ρn(D) ∩ ρk(D) = ∅ for all
nonnegative integers k n.
Let, A=
So, S2= A(S2\A) ~ (S2\A)= (A\D) (S2\A)=S2\D.
So, S2 is SO3-equidecomposable to S2\D, hence proved.
 
Corollary : S2 is SO3- paradoxical.
Proof: By Hausdorff paradox, S2\D is SO3-paradoxical, and by last theorem, S2\D ~ S2 so, S2 is SO3-paradoxical, which proves the
statement.
Now we are ready to prove Banach-Tarski paradox.
Theorem (Banach-Tarski Paradox) (Weak form) 6: Every ball in R3 can be paradoxical decomposed by
rotations and translations. More precisely, B3, the solid ball in R3, is G3-paradoxical.
Proof: Let B3 be a unit ball around the origin.
Since, S2 is SO3-paradoxical, we can obtain a paradox for any “thickened” shell by producing a paradox for one
sphere in this shell, and then putting points along a radius in the same piece in a decomposition. In particular, we
see
B3\{0} is paradoxical.
We will show that it is equidecomposable to B3.
Take a small circle which passes through 0 and is contained in B3 . Then let ρ be a 1 radian rotation of the circle.
The it follows that 0, ρ(0), ρ2(0), ρ3(0), . . . are all distinct. If ρ is then applied to this set of points again, the set
remains the same except 0 is now gone. This creates a equidecomposition of B3 with B3\{0}. One piece is 0, ρ(0),
ρ2(0), ρ3(0), . . . going to ρ(0), ρ2(0), ρ3(0), . . . under rotation ρ, the other is the remaining points in the ball going
to itself.
So choose a broken circle C' B3\{0} with the origin as its broken point.
We derived earlier that C' is equidecomposable with a completed circle C. Thus,
B3\{0}=B3\({0} C') C' ~ B3\({0} C') (C'{0})=B3.
Since, B3\{0} is paradoxical and is equidecomposable to B3 so, B3 is also paradoxical. Hence, proved.

 
Corollary : R3 is paradoxical

Proof: We constructed a paradoxical decomposition of B3\{0} by drawing radial lines from every point in S2 to
the origin, and identifying the lines with the points . The paradoxicity of S 2 was then shown to be equivalent to
that of B3\{0}. If, rather than terminating the lines at the surface of the unit ball, we extended them infinitely, then
we obtain the set R3\{0}, and see that this set is also SO3 paradoxical. We can then reproduce substituting B3\{0}
with R3\{0}in the proof of the theorem 9, to show that R3\{0} is equidecomposable with R3 . The result follows.
Hence the weak form is strongly established.
However, the children
are not satisfied with the
size of the burger; they
want a much bigger
burger now! So, the
previous magic trick of
the mother doesn’t work
anymore!

Can she now come up

with a new magic?
 Preparing for the magic
The equidecomposability relation is a useful tool, as we have seen in our proofs, and so it
deserves further investigation. It is an equivalence relation, so we can define another
relation between ∼-equivalence classes of sets: A ≤ B if and only if A is equidecomposable
with a subset of B.
This relation ≤ is both reflexive and transitive, by definition. Further- more, it is also anti-
symmetric: if A ≤ B and B ≤ A, then A ∼GB. This fact is proved by a generalization of the
well-known Schr¨oder-Bernstein Theorem of set theory— Banach was able to generalize it
for equivalence relations satisfying these two properties (which ∼G satisfies):
(i) if A ≤ B, then there exists a bijection g : A → B such that if C ⊆ A, then C ∼ g(C),
(ii) if A1 ∩ A2 = B1 ∩ B2 = ∅, and if A1 ∼ B1 and A2 ∼ B2, then A1 ∪ A2 ∼ B1 ∪ B2
 Banach-Schr¨oder-Bernstein Theorem:
Let G act on X and A, B ⊆ X. Hey! I am
Then A ≤ B and B ≤ A implies that smaller
A∼GB. than you!
Proof:-
By Condition (i), we have bijections f : A → B1
and g : B → A1; A1 ⊆ A; B1 ⊆ B.
Let us define C0 = A\A1 and Cn+1 = g(f(Cn))
with C= ∪ Cn ,n =1 to ∞
Then x ∈ f(C) ⇔ g(x) ∈ C.
We can see that A\C ⊆ A1 = g(B). So A\C = g(B\
f(C)) But, I am
smaller
But, thanks to Condition (i), this shows that A\C than you
is equidecomposable with B\f(C). Also, C∼f(C). too!
But does
Then, by Condition (ii): A = A\C ∪ C ∼ B\f(C) ∪ this mean
f(C) = B that we
are equal?
The mother has now got “stronger” and
is ready for her new magic!!
The Banach-Tarski Paradox (Strong Form)

If A and B are bounded subsets of R3 with nonempty interior, then A ∼ B.

Proof: We want to show A ≤ B, for then B ≤ A is similarly proved, and we

obtain by previous theorem that A ∼ B.
So choose balls K and L such that K contains A and L is contained by B.
Then for some n large enough, K ⊆ n copies of L. Then we have that A
⊆ K ⊆ (n copies of L) ≤ L ⊆ B, where the ≤ arises from repeated Banach-
Tarski Paradox duplications of L.
Hence A ≤ B.
Similarly, we obtain B ≤ A, and so A ∼ B.
The next thing that comes up into our mind is
that can the mother cut short her efforts
during her first magic while breaking the
burger into a finite number of pieces?

In other words, what are the minimum

number of pieces do we require for the
duplication of the ball?
MINIMIZING THE NUMBER OF PIECES
Before we proceed further, we need the following theorems.

Theorem7:- Let F be the free group generated by σ and τ . Then we can actually partition F into four sets A1, ..., A4 such that
σ(A2) = A2 ∪ A3 ∪ A4 and τ (A4) = A1 ∪ A2 ∪ A4.
Moreover, for any fixed w ∈ F, the partition can be chosen such that w is
in the same piece as the identity of F

Proof:- From Theorem 2, we get B(x) = set of all words beginning on the left with x, where x may be σ,τ,σ -1,τ -1.
Then F = {1}∪B(σ)∪B(σ -1 )∪B(τ)∪B(τ -1), where all sets are pair-wise disjoint.
But F = B(σ)∪σB(σ -1)and F = B(τ)∪τB(τ -1), which shows that F is indeed paradoxical.
However, we may choose to partition F so that all four pieces in the paradoxical decomposition.
We fix some w in F and ρ denote the leftmost letter in te word w.
Now suppose, for instance, that ρ = τ -1. (The other cases are similar.)
Then we may choose our partition as follows:
A1 = B(σ)
A2 = B(σ -1)
A3 = B(τ ) \ {τn : n ∈ N+}
A4 = B(τ -1) ∪ {τn : n ∈ N+} ∪ {1}
These sets have the property that 1, w are in the same piece, as well as satisfy the relations above: σ(A 2) = A2∪ A3∪ A4 and τ
(A4) = A1 ∪ A2 ∪ A4.
Theorem 8:- Let F be a free group of rank 2 acting on X without nontrivial fixed points.
Then X is F -paradoxical using four pieces.

Theorem 9:- Let F be a free subgroup of rank 2 whose action on X is locally

commutative.
Then X is paradoxical using four pieces.
(Definition of locally commutative group actions:-
Let G act on X. This action will be called locally commutative if for every point x ∈ X
the stabilizer of x is commutative, i.e., if the subgroup of G which fixes x is
commutative.)

Definition:- Suppose G acts on X and E, A, B ⊆ X. Then E is G- paradoxical using r

pieces if A ∩ B = ∅ with A ∪B = E such that
A ∼mE ∼nB where m + n = r.
Theorem 10:- S2 is SO3-paradoxical using four pieces, and the four cannot
be improved.

Proof:- The action of rotations on the sphere is locally commutative

and SO3 have a subgroup isomorphic to F2.
To demonstrate that a composition in four pieces cannot be improved,
assume the contrary:
that S2 contains two disjoint non-empty subsets A, B such that
A ∼ mS2 ∼nB where m + n < 4.
At least one of m or n must equal 1.
If m = 1, then S2 = g(A) for some g ∈ SO3, whence A = g-1 (S2 )=S2
and B = ∅, a contradiction.
‘For n=1,m=2 a similar type of contradiction will hold.’
Using the previous results, it can be shown that
“B3, the solid ball in R3, is paradoxical using five pieces, and this is best possible.”
 CONCLUSION

We have derived a completely unintuitive, somewhat disturbing result. What are we to conclude from
this? At first glance, the Banach-Tarski Paradox appears to be an actual contradiction with measure
theory, for if we measure all the pieces used in the construction, they must add up both to the volume
of the ball and to twice the volume of the ball, and these are different positive real numbers. The
resolution of this paradox is that the pieces must not be Lebesgue measurable—or more generally,
any function that satisfies the properties of measure listed above cannot be defined on these pieces.
So we have examined the Banach-Tarski Paradox inside out. Even after all this analysis, we have
only scratched the surface of the literature on paradoxes.
Since, the closed bounded balls in Euclidean spaces are compact sets as well as Borel sets, so one
may think whether compact metric space can be paradoxical using Borel pieces? But, this remains
an open problem till now!
Clearly such questions will entertain mathematicians for many years to come!
Thus The Magic Show Ends Here!