Appendix C

The Exponential Function

Various calculus books used at ASU have included in the title Calculus: Early
Trancendentals. This is because exponential functions are introduced early in the
text. By delaying their use until after integration, exponentials could be rigorously
introduced by defining logarithms using integrals and exponentials as the inverse
of logarithms. Here we show a way to define exponential functions using the com-
pleteness axiom - thus justifying the approach in the freshman calculus books. Here
is the main result.
Theorem C.1. Let a, b > 0, x, y R. There exists a real-valued continuous
function on the real numbers, f (x) = ax , with the properties

(i) ax ay = ax+y ; ax
(v) = axy ;
(ii) (ab)x = ax bx ; ay
(vi) For a > 1, x < y iff ax < ay ;
(iii) (ax )y = axy ;
(vii) x > 0 and ax = 1/ax ;
 a x ax
(iv) = x; (viii) For x > 0, a < b iff ax < bx .
b b

Exponentials with Natural Numbers

To prove the theorem we first define an for n a natural number. Then define
it for rational numbers and finally for irrational numbers. Suppose a > 0 is a real
number. For n N, we define an recursively. Set
a0 = 1;

(C.1) n N, an+1 = a an .
So a2 = a a, a3 = a a2 , a4 = a a3 ...
Lemma C.2. For any real numbers a, b > 0 and for all n, m N

(i) an am = an+m ;
 a n an
(iv) = n;
b b
(ii) (ab)n = an bn ;
an
(iii) (an )m = anm ; (v) For n m, m = anm ;
a
103
 (vi) For a > 1, n < m iff an < am ; (viii) a < b iff an < bn .
(vii) an > 0;

Proof. Let a > 0 be any real number and m be any natural number. We use
mathematical induction on n. By definition, (C.1), aam = am+1 . Suppose an am =
an+m . Then
a(n+1)+m = a(n+m)+1
= a an+m
= a an am
= an+1 am .
This proves (i).
Similarly, (ii) holds for n = 1. Suppose (ab)n = an bn . Then,
(ab)n+1 = ab (ab)n
= ab an bn
= an+1 bn+1 .

The third assertion clearly holds for n = 1. Suppose (an )m = anm . Then
(an+1 )m = (a an )m
= am anm
= am+nm
= a(n+1)m .

Part (iv) obviously is true for n = 1. Suppose (a/b)n = an /bn , then (a/b)n+1 =
(a/b)(a/b)n = (a/b)(an /bn ) = an+1 /bn+1 .
an
By definition part (v) is true for n = m. Suppose m = anm . Then
a
an+1 a an
=
am am
= a anm
= an+1m
= an+1m ,
and (v) follows.
To prove (vi) suppose a > 1 and n < m. If an1 > 1, then multiplying by a
and using Part (i), an > a > 1. By induction an > 1 for all n N. If m > n,
m n N. Hence, amn > 1. Part (v) then shows that an < am .
Conversely, suppose a > 1 and an < am . If n m, then Part (v) shows
nm
a 1. This contradicts an > 1 for all n. Thus n < m.
Part (vii) follows from a simple induction argument. To prove (viii) suppose
a < b and an < bn for n N. Then an+1 = aan < abn < bbn = bn+1 , and an < bn
for all natural numbers n. Conversely, suppose an < bn for all n. Part (iv) shows
(a/b)n < 1. As observed in the proof of Part (vi), c 1, implies cn 1. Thus
a/b < 1, and part (viii) follows.

Exponentials with Integers
Next we extend the results for integers.

Definition 1.1 Suppose n N. Then

an = 1/an .
Lemma C.3. Properties (i) (viii) in Lemma C.2 are valid for all integers.

Proof. We first prove (v) with no restrictions on m and n. Indeed, we reprove (v)
using induction on m. Suppose n is a given natural number and an /am = 1/amn
for m > n. Then
an an 1 1
= = mn = m+1n
am+1 aam aa a
and so an /am = 1/amn holds for all m n. Moreover, since 1/amn = anm , we
find (v) in Lemma C.2 is true for all n, m N.
To prove (i), suppose m > 0. Then an am = an /am = anm . Similarly,
1 1 1
an am = n m
= n+m = anm .
a a a
Parts (i) and (v) follows for all integers. To prove (ii) let n > 0 be a natural
number. Then
1 1
(ab)n = n
= n n = an bn .
(ab) a b
To prove (vi) we again assume a > 1. Suppose z1 < z2 with z1 , z2 integers.
Then z2 z1 N, and az2 z1 > 1. That is, az2 az1 > 1. Using the properties
proven above, this implies az1 < az2 . Since the argument can be reversed, the
converse holds as well. The remaining properties follow in a similar way. 

Exponentials with Rational Numbers.

Next we define ar for rational numbers. To do so, we need to show that positive
numbers have nth roots.
Lemma C.4. Let a > 0 and n N. The equation z n = a has a unique positive
solution.

Proof. Consider the set

E = {x R | 0 xn a}.
The set is non empty since 0 E. If a > 1, then a is an upper bound for E. Indeed,
for any y a, y n an > a, and y is not in E. If a 1, then one is an upper
bound for E since for any number y > 1, y n > 1 and therefore not in E. By the
completeness axiom, sup(E) exists.
 Set z = sup(E). We need to show z n = a and use trichotomy to do so. Some
estimates are needed to proceed. Using induction, one can prove the binomial
expansion
n  
n
X n k nk
(x + y) = x y ,
k
k=0
where  
n n!
= .
k k!(n k)!
Suppose 0 < < 1. Apply the binomial expansion with x = and y = 1. We find
there is a constant, K so that
n   n  
n
X n k X n k
(1 + ) = =1+
k k
k=0 k=1
n  
X n
< 1+
k
k=1
(C.2) = 1 + K.
Suppose z n < a. Then there exists an > 0 such that z n (1+K) < a. Equation
(C.2) shows
(z (1 + ))n = z n (1 + )n < z n (1 + K) < a.
That is, z (1 + ) is in E and larger than z. This contradicts z being an upper
bound.
Similarly, if z n > a, then > 0 exists so that z n /(1 + K) > a. Equation (C.2)
shows n
zn zn

z
a< < = .
1 + K (1 + )n 1+
That is, z/(1 + ) is and upper bound for E. This contradicts z being the least
upper bound.
To show uniqueness suppose positive z1 and z2 exist with zin = a, i = 1, 2.
Property (viii) in Lemma C.2 implies z1 = z2 . 

Definition C.5. Let r = m/n Q be positive, and z be the unique solution of

z n = am . We define ar = z. For negative rationals, we define ar = 1/ar , r > 0.

We need to check that this definition is well defined. Specifically, the represen-
tation for r is not unique (it may not be in reduced form). We need to check that
ar is independent of the form of r. The following lemma will help.
Lemma C.6. Let a, b > 0 be real and suppose ap = bp for some natural number p.
Then a = b.

Proof. By the Property (iv) in Lemma C.2 we find (a/b)p = 1. Let z be the unique
pth root of one. That is, z p = 1. Since one solves this equation and the pth root is
unique a/b = 1. That is a = b. 

Lemma C.7. Suppose r = m/n = m /n = r for n, m N. Then ar = ar . In
particular, the definition of ar is well defined.

Proof. Let z and z be the unique solution of z n = am and z n = am . Using the

properties in Lemma C.2, z nn = amn and z nn = am n . Since mn = m n, we find

z nn = z nn , and the previous lemma shows z = z . 
Lemma C.8. For any real numbers a, b > 0 and for all r, s Q

(i) ar as = ar+s ; ar
(v) = ars ;
(ii) (ab)r = ar br ; as
(vi) For a > 1, r < s, iff ar < as ;
(iii) (ar )s = ars ;
(vii) ar > 0;
 a r ar
(iv) = r; (viii) For r > 0, a < b, iff ar < br .
b b

Proof. To Prove (i) suppose r = m/n and s = k/l. Let zr and zs be the unique
solutions to zrn = am and zsl = ak respectively. Using the properties in Lemma
C.2, zrnl = aml and zsnl = ank . Then (zr zs )nl = aml+nk . Definition C.5 implies
zr zs = a(ml+nk)/nl . This is (i) for positive rationals. If s = k/l < 0, then zs solves
zsl = 1/ak and zsnl = 1/ank . By the properties in Lemma C.3 (zr zs )nl = amlnk ,
which again gives (i).
To prove (vi), suppose 0 < x < y and that x = m/n, y = l/k. Then zx and
zy exists so that zxn = am and zyk = al . That is, zxkn = amk and zykn = aln . Since
mk < ln, Part (vi) in Lemma C.2 shows zxkn < zykn . Part (vi) in Lemma C.2 again
applies and zx < zy which is (vi). Similar arguments hold for negative rationals.
Moreover, the other properties follow in the same manner. 

Exponentials with Minor Restrictions.

We define ax for any real x and for a > 1. We will remove all restriction in the
next section. Consider the set
(C.3) Ex = {ar : r Q, r < x}.
The set Ex is non empty since ar Ex for any rational number less than x. Since
ar is increasing with increasing r, Ex is bounded above by ar1 for any rational
number r1 > x.
Definition C.9. For any x R we set
ax = sup Ex .

We should check that the new definition agrees with our definition for ax when
x is rational.
Lemma C.10. Suppose a > 1. If x = r1 Q, then sup Er1 = ar1 . In particular,
for rational x Definition C.9 agrees with Definition C.5.

Proof. Property (vi) in Lemma C.8 shows ar1 is an upper bound of Er1 . Let > 0.
We show ar1 is not an upper bound of Ex . In the appendix we prove that
lim a1/n = 1. Let N N be chosen so that |a1/n 1| < /ar1 for all n N .
n
Property (vi) in Lemma C.8 implies |ar 1| < /ar1 for rational 0 r < 1/N . Let
s be any rational number with r1 1/N < s < r1 . Then 0 r1 s < 1/N and
as
0 < ar1 as = as (ar1 s 1) < r1 < .
a
Since s < r1 , as Er1 and ar1 is the least upper bound. 

Lemma C.11. (Theorem C.1 (i)) Suppose a > 1 and x, y R. Then ax ay = ax+y .

Proof. We have to show

sup{ar |r < x, r Q} sup{as |s < y, s Q} = sup{at |t < x + y, t Q}.
To do so consider the sets A = {r + s : r < x, s < y, r, s Q}, and B = {t Q :
t < x + y}. We claim A = B.

Remark. The sets are not

the same if equality is included in the sets. For example,
set x = 2 and y = 2 2. Then 2 B but not in A. This is one reason equality
was not included in (C.3).

To prove the claim set t = r + s A. Then t < x + y and hence is in B.

Conversely, suppose t B. By the density of rationals, r Q exists so that
t y < r < x. Set s = t r. Then r < x, s < y, and t = r + s A. Hence, B A.
Thus
ax+y = sup{et : t < x + y, t Q} = sup{ar+s : r < x, s < y, r, s Q}
= sup{ar as : r < x, s < y, r, s Q} = ax ay .
(We leave it as an exercise to show sup(AB) = sup(A) sup(B) when A and B are
subsets of the positive reals. See the homework below.) 

Lemma C.12. (Theorem C.1 (ii)). Suppose a, b > 1, x R. Then (ab)x = ax bx .

Proof. We have to show

sup{(ab)r |r < x, r Q} = sup{ar |r < x, r Q} sup{bs |s < x, s Q}.
To do so, we first claim that
sup{ar br : r < x, r Q} = sup{ar bs : r < x, s < x, r, s Q}.
Indeed, set E1 = {ar br : r < x} and E2 = {ar bs : r < x, s < x}. Part (vi)
in Lemma C.8 implies that any element in E2 is bounded by an element in E1 .
Specifically, if ar1 bs1 E2 , then ar1 bs1 ar br E1 , where r = max{r1 , s2 }.
Hence, sup E2 sup E1 . Since E1 E2 , sup E1 sup E2 .
As in the proof of Part (i),
(ab)x = sup{(ab)r : r < x, r Q} = sup{ar br : r < x, r Q}
= sup{ar bs : r < x, s < x, r, s Q}
= sup{ar : r < x, r Q} sup{br : r < x, r Q}
= ax b x .

Lemma C.13. (Theorem C.1 (iii)) Suppose a > 1 and x, y are positive numbers.
Then (ax )y = axy .

Remark. We have only defined ax when a > 1. Thus we need ax > 1 for (ax )y to
be defined. The restriction x > 0 is therefore necessary at this juncture. Indeed,
one can check using our definitions of ax and ar for r Q that ax > ar > 1 for
0 < r < x.

Proof. Set E1 = {(ax )s |0 < s < y, s Q} and E2 = {(ar )s |0 < r < x, 0 < s <
y, r, s Q}. By our definitions, (ax )y = sup E1 . Since ar < ax for all 0 < r < x,
Property (viii) in Lemma C.8 shows (ar )s < (ax )s , and thus (ax )y is an upper
bound for E2 , and sup E2 sup E1 . We show it is the least upper bound of E2 by
showing every element in E1 is bounded by an element in E2 (i.e. sup E1 sup E2 ).
Indeed, let (ax )s1 E1 with 0 < s1 < y. Pick s2 Q such that 0 < s1 < s2 < y
and r1 mathbbQ such that xs1 /s2 < r1 < x. It follows that ar1 s2 E2 and

ar1 s2 = (ar1 s2 /s1 )s1 > (ax )s1 .

Therefore, sup E1 = sup E2 .

We need a claim similar to the one above. Set A = {rs : 0 < r < x, 0 < s <
y, r, s Q} and B = {t Q : 0 < t < xy}. We claim A = B. Indeed, if r < x and
s < y, rs < xy. That is A B. Conversely, suppose t < xy. Choose s Q so that
t/x < s < y, and set r = t/s < x. Then t = rs < xy, and t A.
This shows

(ax )y = sup{(ar )s |r < x, s < y, r, s Q} = sup{(ars |r < x, s < y, r, s Q}

= sup{(at |t < xy, t Q}
:= axy .


Before proceeding to Property (iv) we need another lemma. The lemma uses
the notion of limits. With more work the use of limits could be avoided.

Lemma C.14. Suppose rn Q is any monotone increasing sequence with rn < x

for all n N and that rn x as n . Then ax = lim arn .
n

Proof. By Property (vi) in Lemma C.8, the sequence {arn } is monotone increasing.
Moreover, arn < ax by the definition of ax . By the monotone convergence theorem
the limit exits. If the limit is less than ax , the supremum of (C.3), then there
is an element in Ex between the limit and ax . That is, as exists in Ex with
arn as < ax . By Property (vi) in Lemma C.8, rn s < x for all n N. This
provides a contradiction to lim rn = x. 
n
 a x ax
Lemma C.15. (Theorem C.1 (iv)) Suppose a > b > 1. Then = .
b bx

Remark. We need the restriction a > b since we have only defined ax for a > 1.
Proof. We need to show

sup{ar |r < x, r Q}
(C.4) sup{(a/b)r | r < x, r Q} = .
sup{br |r < x, r Q}

Let {rn } be as in Lemma C.14. Then the algebra of limits and Property (iv)
in Lemma C.8 implies (C.4). Indeed,
 a  rn ar n sup{ar |r < x, r Q}
sup{(a/b)r | r < x, r Q} = lim = lim r
= .
n b n b n sup{br |r < x, r Q}
 a x ax
That is, = . 
b bx
ax
Lemma C.16. (Theorem C.1 (v)) Suppose a > 1, x > y. Then = axy .
zy

Remark. There should be no restrictions on x and y since we have defined ax for

all reals in Definition C.9. The constraint x > y can easily be removed after we
prove Property (vii).

Proof. Let rn = x 1/(2n) and sn = y 1/n be two monotone increasing sequences

converging to x and y respectively. Then Property (v) in Lemma C.8 applies for
all n N, and
ar n
= arn sn .
asn
Set tn = rn sn . Then {tn } is monotone increasing with limit x y. Lemma C.14
applies and the stated result in the lemma follows. 

Lemma C.17. (Theorem C.1 (vi)). Suppose a > 1. Then x < y if and only if
a x < ay .

Proof. Suppose x < y are real numbers, and consider the sets Ex = {ar : r
Q, r < x} and Ey = {as : s Q, s < y}. Since x < y, Ex Ey . Hence
sup Ex sup Ey . The inequality is strict for the following reason. A rational
number exists so that x < r1 < y. Then ar1 Ey , and Part (vi) in Lemma C.8
shows that ar1 < sup Ey . Moreover, since ar < ar1 for all r x, sup Ex ar1 .
That is, ax = sup Ex ar1 < sup Ey = ay .
Conversely, suppose ax < ay . That is, sup Ex < sup Ey . Suppose y x. As
we noted above y x implies Ey Ex , and hence sup Ey sup Ex , contrary to
our assumption. 

1
Lemma C.18. (Theorem C.1 (vii)) Suppose a > 1. For all x, ax = ax .

Remark. We cannot take this as a definition here since we have already defined
ax for all real numbers.
Proof. For x > 0, the properties in Lemma C.8 imply
ax = sup{ar | r < x, r Q} = sup{ar | r > x, r Q}
1 1
= sup{ r | r > x, r Q} = inf{ r | r < x, r Q}
a a
1 1
= = x.
sup{ar | r < x, r Q} a
The case x < 0 follows from the above. Suppose x = 0. Since zero is rational,
Definition C.9 agrees with the rational definition - which implies a0 = 1. The
statement is trivial in this case. 
Lemma C.19. (Theorem C.1 (viii)) Suppose a, b > 1 and x > 0. Then a < b if
and only if ax < bx .

Proof. Lemma C.17 shows that ar < ax for all 0 < r < x and a > 1. We claim
ar > 1 for r > 0. Indeed, if r = m/n, then z n = am where z = ar . Since a > 1,
am > 1. Thus z n > 1 and z > 1. Consequently ax > 1 for such a and x.
Thus a < b and Lemma C.15 implies 1 < (b/a)x = bx /ax , or ax < bx . Con-
versely, suppose ax < bx . If b a, the same argument shows bx ax contrary to
our assumption. 

Exponentials with No Restrictions.

For 0 < a < 1 we define ax = (1/a)x . The constraints in the previous section
are easily removed. Except that the inequality in (vi) in Lemma C.2, Lemma
C.8, and Theorem C.1 is reversed. We do not provide all of the details since the
calculations follow in a straightforward manner. Some representative cases are
considered here.
Lemma C.20. (Theorem C.1 (i)) Suppose a > 0 and x, y R. Then ax ay = ax+y .

Proof. The proof for a > 1 was completed in the previous section. So suppose
0 < a < 1. Using the lemmas in the previous section and taking care to notice the
restrictions, we have
 (x+y)  x  y
1 1 1
ax+y = = = ax ay .
a a a

Lemma C.21. (Theorem C.1 (ii)) Suppose a, b > 0, x R. Then (ab)x = ax bx .

Proof. We proved this lemma for a, b > 1 in the previous section. There are several
cases to check. We examine only one. Suppose 0 < a < 1 and ab > 1. Lemmas
C.15 and C.18 apply (with restrictions)
 x  x
1 bx b
ax b x = bx = 1
x = 1 = (ab)x .
a a a
We assume the other cases follow as easily. 
Lemma C.22. (Theorem C.1 (iii)) Suppose a > 0 and x, y R. Then (ax )y = axy .

Proof. Suppose a > 1. We first drop the restriction that x, y be positive.

Suppose x < 0 and y > 0. Then ax = 1/ax < 1 as we proved in the beginning of
Lemma C.19. Thus
 y
x y 1 1 1 1
(a ) = x
= 1
y = x y = xy = axy ,
a ax
(a ) a
where we have used Lemma C.13 and Lemma C.18. We again assume the other
cases work in a similar manner. 
 a x ax
Lemma C.23. (Theorem C.1 (iv)) Suppose a, b > 0. Then = .
b bx

Proof. We only examine the case 1 < a < b. We have

 a x  1 x  b x bx ax
= a = = x = x
b b a a b

The others follow in a similar way. As a final lemma, we show the continuity
of f (x) = ax .

Lemma C.24. Suppose a > 0. Then ax is continuous on R.

Proof. Let > 0, x0 R and {xn } be any sequence converging to x0 . Then

axn ax0 = ax0 (axn x0 1).
According to the result in the appendix, an N1 N exists so that |a1/n 1| < /ax0
for all n N1 . There exists an N N such that |xn x0 | < 1/N1 for all n N .
Then
|axn ax0 | = ax0 |axn x0 1| <
for n max{N1 , N } and ax is continuous for any x0 R. 

Appendix.

Here we establish the limit lim a1/n = 1.

n

Lemma C.25. Suppose a > 0. Then lim a1/n = 1.

n

Proof. We show the sequence {a1/n } is bounded and monotone. First suppose
0 < a < 1. If 1 < a1/n , Lemma C.8 shows that 1 < a. Hence, a1/n < 1 for all
n N. To see the sequence is monotone note
1
a1/n a1/(n1) = a1/n (1 a n(n1) ) > 0
for all n N. The monotone convergence theorem applies and the limit exists
which we call a.
 To find the
limitwe recall that xn x, n , with xn > 0 for all n N
implies lim xn = x. Moreover, since all subsequences of a1/n converge to a,
n
p
lim a1/n = lim a1/(2n) = a
n n

while p
lim a1/n = a.
n

That is, a = a, and a = 1 or a = 0. Since a1/n is increasing, the limit must be
one.
If a > 1, then 1/a < 1 and
1 1
lim a1/n = lim = = 1.
n n (1/a)1/n lim (1/a)1/n
n


Homework
1. Suppose A and B are subsets of positive numbers. Show
sup(AB) = sup(A) sup(B).

Solution. Let A = sup(A) and B = sup(B). Since A, B R+ , A B is an

upper bound of AB. We show it is the least upper bound by showing A B is
not an upper bound of AB for all > 0.
Consider the function f (x) = x(A + B ) x2 . One checks that f (0) = 0 =
f (A + B ) and that f (x) > 0 on (0, A + B ). Thus, given any > 0, there
exists an with > (A + B ) 2 and 0 < < (A + B )/2. This is the same
statement as
A B < (A )(B ) < A B .
Since A and B are not upper bounds of A and B respectively, there exists
a A, b B such that A < a A and B < b B . Therefore,
A B < ab A B .
Since ab AB, A B is not an upper bound of AB, and A B is the least
upper bound of AB.
Note that A and B must be positive. Indeed, if A = {1, 2, 3} and B =
{1, 2, 3}, then sup(AB) = 9, while sup(A) sup(B) = 1.

2. Let g : R R and not identically zero. Moreover, suppose g satisfies for all
x, y R
g(x + y) = g(x)g(y). (1)
The purpose of this problem is to show that g must be an exponential function.
(a) Show that g is continuous at every point of R if and only if it is continuous at
x = 0.
(b) Show that g > 0 for all x R.
(c) Prove that g(0) = 1. If a = g(0), then a > 0 and g(r) = ar for all r Q.
(d) Suppose g is continuous, Show g is strictly increasing, is constant, or is strictly
decreasing if g(1) > 1, g(1) = 1, or g(1) < 1 respectively.
(e) If g(x) > 1 for some x (0, ), > 0, then f is strictly increasing and
continuous.
(f ) If a > 0, then there exists at most one continuous function g satisfying (1).
(g) Suppose a > 1. Show there exists a unique continuous function satisfying (1)
with g(1) = a.