Skip to content

Chapter 4 — Exponential and Logarithmic Functions

4.1 - Exponential Functions

Having rational exponents was, until now, arithmetic disguised as algebra. What follows is the step in which that arithmetic becomes, without losing one iota of precision, analysis.


Chapter 1 left constructed, with all the rigor this book demands, the power with rational exponent: for a∈Ra\in\mathbb{R} and r=m/n∈Qr=m/n\in\mathbb{Q}, the symbol ara^r was defined via radicals, and it was proved, not assumed, that this value does not depend on which representation of the fraction m/nm/n is used. This section does not repeat that construction: it uses it as a starting point and asks, precisely, about what ara^r cannot yet do, namely accept any real exponent, and not only the rational ones, in order then to resolve it.

§1. What has already been constructed, and the question it leaves open

Section titled “§1. What has already been constructed, and the question it leaves open”

Recall, without proving it again, for economy and because it is already proved, that for a∈Ra\in\mathbb{R}, n∈Nn\in\mathbb{N} and m∈Zm\in\mathbb{Z}, the symbol am/na^{m/n} was defined in §4 of section 1.2 as (a1/n)m\left(a^{1/n}\right)^m, and that this definition requires, when nn is even, the restriction a≥0a\ge 0, since otherwise a1/na^{1/n} does not exist in R\mathbb{R}. It was also proved, in that same section, that the value of am/na^{m/n} does not depend on which fraction equivalent to m/nm/n is chosen to compute it.

That second property, independence of the representation, is more delicate than it seems when viewed with this chapter’s goal in mind. The proof of §1.2 assumes that both representations, m/nm/n and p/qp/q, give rise to well-defined powers; it does not say what happens if one of them does not exist in R\mathbb{R}. And here the problem appears: one and the same rational number admits representations with even denominator and also with odd denominator (1/21/2 and also, for example, 3/63/6), so that if the base aa is negative, some representations of the same rational will produce a well-defined real number (those with odd index) and others none (those with even index). The symbol ara^r, for a<0a<0, is not badly defined: it is defined halfway, on a subset of Q\mathbb{Q} that depends, in an unnatural way, on how the exponent has been written.

For a function f(x)=axf(x)=a^x that aspires to have domain R\mathbb{R}, or even Q\mathbb{Q}, as a first step, that situation is inadmissible. We therefore need a condition on aa guaranteeing that ara^r exist for every r∈Qr\in\mathbb{Q}, without exception and without depending on the way the exponent is written.

§2. The condition a > 0 as a necessary and sufficient restriction

Section titled “§2. The condition a > 0 as a necessary and sufficient restriction”

Proposition. Let a∈Ra\in\mathbb{R}. The value ara^r is defined in R\mathbb{R} for all r∈Qr\in\mathbb{Q} if and only if a>0a>0; in that case, moreover, ar>0a^r>0 for all r∈Qr\in\mathbb{Q}.

Proof. Sufficiency and necessity are proved separately.

(Sufficiency: a>0  ⟹  ara>0 \implies a^r defined and positive for all r∈Qr\in\mathbb{Q}.) Let r=m/n∈Qr=m/n\in\mathbb{Q}, with n∈Nn\in\mathbb{N}, in any representation. Since a>0a>0, in particular a≥0a\ge 0, so the restriction of §1.2 §3, necessary only when nn is even, is satisfied without further ado, whether nn is even or odd. Hence a1/na^{1/n} exists in R\mathbb{R}, and with it, am/n=(a1/n)ma^{m/n}=(a^{1/n})^m. Since this holds for any representation of rr, and §1.2 §4 already guaranteed that the result does not depend on which one is used, ara^r is well defined, without ambiguity, for all r∈Qr\in\mathbb{Q}. As for the sign: a1/na^{1/n} is, by the definition of principal root, nonnegative, and since a>0a>0 it cannot be a1/n=0a^{1/n}=0, for if it were, a=(a1/n)n=0a=(a^{1/n})^n=0, a contradiction, so that a1/n>0a^{1/n}>0. A positive base raised to any integer exponent mm, positive, negative or zero, remains positive, by the laws of exponents already established; hence ar=(a1/n)m>0a^r=(a^{1/n})^m>0.

(Necessity: if a≤0a\le 0, there exists r∈Qr\in\mathbb{Q} with ara^r undefined.) If a=0a=0, r=−1r=-1 already fails: a−1=1/0a^{-1}=1/0 is not defined in R\mathbb{R}. If a<0a<0, take r=1/2r=1/2, with the only possible reduced representation n=2n=2 even. The restriction of §1.2 §3 requires, for nn even, that a≥0a\ge 0; since a<0a<0, there is no real b≥0b\ge 0 with b2=ab^2=a, so a1/2a^{1/2} is not defined in R\mathbb{R}. In either case, a≤0a\le 0 produces at least one rational for which ara^r does not exist. ■\blacksquare

§3. The convention of the exponential function: a > 0, a ≠ 1

Section titled “§3. The convention of the exponential function: a > 0, a ≠ 1”

The previous Proposition fixes the first hypothesis on the base: a>0a>0. A second one remains, of a different nature, not algebraic but of purpose. If a=1a=1, then 1r=11^r=1 for all r∈Qr\in\mathbb{Q}, and, as will be seen, for every real once the extension is completed, so that f(x)=1xf(x)=1^x is simply the constant function f≡1f\equiv 1, already studied in the catalog of prototypes of §2.4. There is nothing wrong with that function, but there is also nothing exponential in it: it lacks the strict monotonicity and the injectivity that are, as will be seen in §9, the central content of this chapter. It is excluded, then, not because it is badly defined but because its inclusion would make the following theorems vacuous.

With this, the base of every exponential function, whose formal definition will be completed only in §6–§7, once it is extended to R\mathbb{R}, is restricted to

a∈(0,1)∪(1,∞)a\in(0,1)\cup(1,\infty)

that is, a>0a>0 and a≠1a\neq 1. This is the standard hypothesis that will be maintained, without repeating it in each statement, for the rest of the chapter.


With the base now restricted to a>0, a≠1a>0,\ a\neq 1, and ara^r well defined and positive for every rational rr, the central problem of this section is posed: to extend ara^r to an arbitrary real exponent xx, in a unique way and without losing any of its algebraic properties. The tool that makes this possible is the supremum axiom, fixed in §9 of section 1.1; but before invoking it an auxiliary fact is needed that this axiom has as an immediate consequence and that, until now, the book never needed to make explicit.

Lemma (Archimedean property). The set N\mathbb{N} is not bounded above in R\mathbb{R}.

Proof. We proceed by contradiction. If N\mathbb{N} were bounded above, the supremum axiom would guarantee the existence of s=sup⁡N∈Rs=\sup\mathbb{N}\in\mathbb{R}. Since s−1<ss-1<s, the number s−1s-1 cannot be an upper bound of N\mathbb{N}, for if it were, ss would not be the least of the upper bounds, so there exists some n∈Nn\in\mathbb{N} with n>s−1n>s-1, that is, n+1>sn+1>s. But n+1∈Nn+1\in\mathbb{N} as well, so ss does not bound n+1n+1: a contradiction, since ss was supposed to be an upper bound of all of N\mathbb{N}. Hence N\mathbb{N} is not bounded above. ■\blacksquare

In particular, for every x∈Rx\in\mathbb{R} there exists some natural number nn with n>xn>x; if there were none, xx would be an upper bound of N\mathbb{N}, contradicting the Lemma. With this we prove what really matters here: that between two distinct reals, however close they are, there is always a rational.

Proposition (density of Q\mathbb{Q} in R\mathbb{R}). For every pair x,y∈Rx,y\in\mathbb{R} with x<yx<y, there exists r∈Qr\in\mathbb{Q} such that x<r<yx<r<y.

Proof. Since y−x>0y-x>0, the Archimedean Lemma gives a natural number nn with n>1y−xn>\dfrac{1}{y-x}, whence 1n<y−x\dfrac{1}{n}<y-x, that is, nx+1<nynx+1<ny. Consider the set of integers strictly greater than nxnx; it is nonempty, again by the Lemma, applied this time to nxnx, and it is bounded below by nxnx, so it has a least element mm, by the well-ordering property of Z\mathbb{Z}. The minimality of mm forces m−1≤nxm-1\le nx, that is, m≤nx+1m\le nx+1; and by construction, m>nxm>nx. Combining both inequalities with the one above,

nx<m≤nx+1<nynx<m\le nx+1<ny

and dividing by n>0n>0 throughout the chain, x<mn<yx<\dfrac{m}{n}<y. The rational r=m/nr=m/n is the one sought. ■\blacksquare

Before defining the extension we need to know that, over Q\mathbb{Q}, the power ara^r already behaves as expected of an increasing function when a>1a>1; otherwise, the supremum to be used in §6 would have no increasing sense to extend.

Lemma. Let a>1a>1. If r,s∈Qr,s\in\mathbb{Q} and r<sr<s, then ar<asa^r<a^s.

Proof. Write s−r=p/qs-r=p/q, with p,q∈Np,q\in\mathbb{N}, in reduced form; since s>rs>r, p/q>0p/q>0. By the laws of exponents already established in §1.2 §2, valid also for rational exponents, according to §4 of that same section,

asar=as−r=ap/q=(a1/q)p\frac{a^s}{a^r}=a^{s-r}=a^{p/q}=\left(a^{1/q}\right)^p

We claim that a1/q>1a^{1/q}>1. Indeed, if a1/q≤1a^{1/q}\le 1, raising both sides to the integer power qq, an operation that preserves order among nonnegatives, we would obtain a=(a1/q)q≤1a=\left(a^{1/q}\right)^q\le 1, contradicting the hypothesis a>1a>1. Hence a1/q>1a^{1/q}>1, and a base greater than 11 raised to a positive integer pp remains greater than 11: (a1/q)p>1\left(a^{1/q}\right)^p>1. Thus as/ar>1a^s/a^r>1, and since ar>0a^r>0 by the Proposition of §2, we conclude as>ara^s>a^r. ■\blacksquare

§6. Existence and uniqueness of the continuous extension for a > 1

Section titled “§6. Existence and uniqueness of the continuous extension for a > 1”

With the density of Q\mathbb{Q} in R\mathbb{R} and rational monotonicity now available, the central result of this section can be formulated and proved.

Proposition (continuous extension, case a>1a>1). Let a>1a>1. There exists a unique function f:R→R+f:\mathbb{R}\to\mathbb{R}^+ that coincides with r↦arr\mapsto a^r on Q\mathbb{Q} and that is strictly increasing on R\mathbb{R}. It is given, for each x∈Rx\in\mathbb{R}, by

f(x)=sup⁡{ar∣r∈Q, r≤x}f(x)=\sup\{a^r \mid r\in\mathbb{Q},\ r\le x\}

Proof. It is organized in three parts: that the supremum exists (good definition), that ff coincides with the already known rational power when x∈Qx\in\mathbb{Q} (consistency), and that no other function with the required properties can differ from ff (uniqueness).

Good definition. Fix x∈Rx\in\mathbb{R}, and let Sx={ar∣r∈Q, r≤x}S_x=\{a^r\mid r\in\mathbb{Q},\,r\le x\}. The set Sx⊆RS_x\subseteq\mathbb{R} is not empty: by the Archimedean Lemma there exists a natural number greater than −x-x, and from it an integer k≤xk\le x, take for example the negative of that natural number, which in particular is rational, so that ak∈Sxa^k\in S_x. Nor is SxS_x unbounded: by the density Proposition there exists a rational t>xt>x, and by the Lemma of §5, every rational r≤x<tr\le x<t satisfies ar<ata^r<a^t; hence ata^t is an upper bound of SxS_x. Nonempty and bounded above, the supremum axiom guarantees that sup⁡Sx\sup S_x exists in R\mathbb{R}, and being the supremum of a set of positive numbers, sup⁡Sx>0\sup S_x>0. This defines f(x)=sup⁡Sxf(x)=\sup S_x for each x∈Rx\in\mathbb{R}, without ambiguity.

Consistency with the rational case. Let x=r0∈Qx=r_0\in\mathbb{Q}. Since r0∈Sr0r_0\in S_{r_0}, taking r=r0r=r_0 in the definition of the set, we have ar0≤sup⁡Sr0=f(r0)a^{r_0}\le\sup S_{r_0}=f(r_0). On the other hand, every r∈Sr0r\in S_{r_0} satisfies r≤r0r\le r_0, and by the Lemma of §5, or equality if r=r0r=r_0, ar≤ar0a^r\le a^{r_0}; hence ar0a^{r_0} is an upper bound of Sr0S_{r_0}, and since f(r0)f(r_0) is the least of the upper bounds, f(r0)≤ar0f(r_0)\le a^{r_0}. Both inequalities together give f(r0)=ar0f(r_0)=a^{r_0}: the extension contradicts, at no rational point, what was already constructed in Chapter 1.

Uniqueness. Let g:R→R+g:\mathbb{R}\to\mathbb{R}^+ be another strictly increasing function that coincides with ara^r on Q\mathbb{Q}. Define, alongside SxS_x, the set Tx={at∣t∈Q, t>x}T_x=\{a^t\mid t\in\mathbb{Q},\,t>x\}. By the same density and monotonicity argument of §5, TxT_x is nonempty and bounded below, by any ara^r with r≤xr\le x, so that inf⁡Tx\inf T_x exists. Since every element of SxS_x is smaller than every element of TxT_x (if r≤x<tr\le x<t, the Lemma of §5 gives ar<ata^r<a^t), it follows immediately that

f(x)=sup⁡Sx ≤ inf⁡Txf(x)=\sup S_x\ \le\ \inf T_x

Sublemma (approximation of unity). For a>1a>1 and every rational ε>0\varepsilon>0, there exists n∈Nn\in\mathbb{N} with a1/n<1+εa^{1/n}<1+\varepsilon.

Proof of the sublemma. We first use Bernoulli’s inequality: for n∈Nn\in\mathbb{N} and real ε>0\varepsilon>0, (1+ε)n≥1+nε(1+\varepsilon)^n\ge 1+n\varepsilon, a fact proved by induction on nn, without difficulty, from (1+ε)n+1=(1+ε)n(1+ε)≥(1+nε)(1+ε)=1+(n+1)ε+nε2≥1+(n+1)ε(1+\varepsilon)^{n+1}=(1+\varepsilon)^n(1+\varepsilon)\ge(1+n\varepsilon)(1+\varepsilon)=1+(n+1)\varepsilon+n\varepsilon^2\ge 1+(n+1)\varepsilon. By the Archimedean Lemma of §4, there exists n∈Nn\in\mathbb{N} with n>(a−1)/εn>(a-1)/\varepsilon, that is, 1+nε>a1+n\varepsilon>a. Combining with Bernoulli, (1+ε)n≥1+nε>a(1+\varepsilon)^n\ge 1+n\varepsilon>a. Both sides of (1+ε)n>a(1+\varepsilon)^n>a are positive, and the nnth root, of which it was already proved, in §1.2 §3, that it is a well-defined operation and that it preserves order among nonnegative reals, gives 1+ε>a1/n1+\varepsilon>a^{1/n}. ■\blacksquare

With this sublemma the missing equality is closed. Suppose, for contradiction, δ:=inf⁡Tx−sup⁡Sx>0\delta:=\inf T_x-\sup S_x>0. Let m∈Zm\in\mathbb{Z} be an integer with m≥xm\ge x, which exists by the Archimedean Lemma of §4, applied to −x-x, and let M=am>0M=a^m>0, a fixed bound that does not depend on what follows. Apply the sublemma with ε=δ2M>0\varepsilon=\dfrac{\delta}{2M}>0: there exists n∈Nn\in\mathbb{N} with a1/n<1+εa^{1/n}<1+\varepsilon. By the density Proposition, there exists a rational r∈(x−1/n, x]r\in(x-1/n,\,x]; take t=r+1/n>xt=r+1/n>x, also rational. Then r∈Sxr\in S_x and t∈Txt\in T_x. Since r≤x≤mr\le x\le m, the Lemma of §5 gives ar≤am=Ma^r\le a^m=M, so that

at−ar=ar(a1/n−1)<ar ε ≤ Mε=M⋅δ2M=δ2<δa^t-a^r=a^r\left(a^{1/n}-1\right)<a^r\,\varepsilon\ \le\ M\varepsilon=M\cdot\frac{\delta}{2M}=\frac{\delta}{2}<\delta

We have thus obtained a pair r∈Sx, t∈Txr\in S_x,\,t\in T_x with at−ar<δa^t-a^r<\delta, whence inf⁡Tx−sup⁡Sx≤at−ar<δ\inf T_x-\sup S_x\le a^t-a^r<\delta, contradicting the definition of δ\delta. Hence δ=0\delta=0, that is,

f(x)=sup⁡Sx=inf⁡Txf(x)=\sup S_x=\inf T_x

With this equality, uniqueness is immediate: for every rational r≤xr\le x, g(r)=ar≤g(x)g(r)=a^r\le g(x) by monotonicity of gg, so that g(x)g(x) is an upper bound of SxS_x and f(x)≤g(x)f(x)\le g(x); symmetrically, for every rational t>xt>x, g(x)≤g(t)=atg(x)\le g(t)=a^t, so that g(x)g(x) is a lower bound of TxT_x and g(x)≤inf⁡Tx=f(x)g(x)\le\inf T_x=f(x). Both inequalities give g(x)=f(x)g(x)=f(x) for all x∈Rx\in\mathbb{R}, and therefore g=fg=f. ■\blacksquare


The extension for a>1a>1 is thus resolved. The case 0<a<10<a<1 does not follow from this one by simple algebraic symmetry (the base changes monotonicity regime, not only the sign of an exponent) and requires its own proof, dual by means of infima instead of suprema: it is the content of §7.

§7. Existence and uniqueness of the continuous extension for 0 < a < 1

Section titled “§7. Existence and uniqueness of the continuous extension for 0 < a < 1”

The case 0<a<10<a<1 is not a mere notational mirror: the power ara^r now decreases as rr increases, so that the relevant set is bounded not above but below, and it is the infimum, not the supremum, that produces the extension. Each step is proved autonomously, without appealing to the relation a=1/ba=1/b with b>1b>1.

Lemma (rational monotonicity, case 0<a<10<a<1). Let 0<a<10<a<1. If r,s∈Qr,s\in\mathbb{Q} and r<sr<s, then ar>asa^r>a^s.

Proof. As in §5, write s−r=p/qs-r=p/q with p,q∈Np,q\in\mathbb{N}. We first claim that a1/q<1a^{1/q}<1: if it were a1/q≥1a^{1/q}\ge 1, raising to the power qq would give a=(a1/q)q≥1a=(a^{1/q})^q\ge 1, contrary to the hypothesis a<1a<1. Since also a1/q>0a^{1/q}>0, being the principal root of a positive real, we have 0<a1/q<10<a^{1/q}<1. A base in that range raised to a positive integer pp remains in that range: indeed, if 0<c<10<c<1 then cp+1=cp⋅c<cp⋅1=cpc^{p+1}=c^p\cdot c<c^p\cdot 1=c^p, so that, by induction on pp starting from c1=c<1c^1=c<1, it follows that 0<cp<10<c^p<1 for all p∈Np\in\mathbb{N}. Applied to c=a1/qc=a^{1/q}: 0<(a1/q)p<10<(a^{1/q})^p<1, that is, as−r<1a^{s-r}<1. Since ar>0a^r>0, multiplying by ara^r preserves the inequality: as=ar⋅as−r<ara^s=a^r\cdot a^{s-r}<a^r. ■\blacksquare

Lemma (existence of the infimum). Every nonempty subset of R\mathbb{R} bounded below has an infimum in R\mathbb{R}.

Proof. Let S⊆RS\subseteq\mathbb{R} be nonempty and bounded below by mm. The set −S={−x∣x∈S}-S=\{-x\mid x\in S\} is nonempty and bounded above by −m-m, since x≥mx\ge m for all x∈Sx\in S is equivalent to −x≤−m-x\le -m. By the supremum axiom, sup⁡(−S)∈R\sup(-S)\in\mathbb{R} exists. It is verified directly that −sup⁡(−S)-\sup(-S) satisfies the two conditions that define the infimum of SS: it is a lower bound of SS, because sup⁡(−S)\sup(-S) is an upper bound of −S-S, and no greater lower bound exceeds it, because sup⁡(−S)\sup(-S) is the least upper bound of −S-S. Hence inf⁡S=−sup⁡(−S)\inf S=-\sup(-S) exists. ■\blacksquare

With both lemmas available, let Sx′={ar∣r∈Q, r≤x}S'_x=\{a^r\mid r\in\mathbb{Q},\,r\le x\}, for fixed a∈(0,1)a\in(0,1) and x∈Rx\in\mathbb{R}. It is nonempty by the same Archimedean argument of §6, and it is bounded below: by density there exists a rational t>xt>x, and since r≤x<tr\le x<t for every rr in the set, the dual monotonicity Lemma gives ar>ata^r>a^t, so that ata^t is a lower bound. By the Lemma on the existence of the infimum, f(x):=inf⁡Sx′f(x):=\inf S'_x exists and is positive, being the infimum of a set of positive reals, bounded below by one of them.

Sublemma (approximation of unity, case 0<a<10<a<1). For every rational ε∈(0,1)\varepsilon\in(0,1), there exists n∈Nn\in\mathbb{N} such that a1/n>1−εa^{1/n}>1-\varepsilon.

Proof. Let c=1−ε∈(0,1)c=1-\varepsilon\in(0,1), and write 1/c=1+h1/c=1+h with h>0h>0, which exists because c<1c<1 implies 1/c>11/c>1. By Bernoulli’s inequality, already proved in §6, (1+h)n≥1+nh(1+h)^n\ge 1+nh for all n∈Nn\in\mathbb{N}. By the Archimedean Lemma of §4, there exists n∈Nn\in\mathbb{N} with n>1/a−1hn>\dfrac{1/a-1}{h}, that is, 1+nh>1/a1+nh>1/a. Combining, (1/c)n=(1+h)n≥1+nh>1/a(1/c)^n=(1+h)^n\ge 1+nh>1/a, whence cn<ac^n<a. Taking the nnth root, order being preserved among positives, according to §1.2 §3, we obtain c<a1/nc<a^{1/n}, that is, a1/n>1−εa^{1/n}>1-\varepsilon. ■\blacksquare

Proposition 2 (continuous extension, case 0<a<10<a<1). Let 0<a<10<a<1. There exists a unique function f:R→R+f:\mathbb{R}\to\mathbb{R}^+ that coincides with r↦arr\mapsto a^r on Q\mathbb{Q} and that is strictly decreasing on R\mathbb{R}. It is given, for each x∈Rx\in\mathbb{R}, by f(x)=inf⁡Sx′=inf⁡{ar∣r∈Q, r≤x}f(x)=\inf S'_x=\inf\{a^r\mid r\in\mathbb{Q},\,r\le x\}.

Proof. The good definition was already established above. For consistency, let x=r0∈Qx=r_0\in\mathbb{Q}: since r0∈Sr0′r_0\in S'_{r_0}, inf⁡Sr0′≤ar0\inf S'_{r_0}\le a^{r_0}; and since every r∈Sr0′r\in S'_{r_0} satisfies r≤r0r\le r_0, the dual monotonicity Lemma gives ar≥ar0a^r\ge a^{r_0}, a lower bound, so that ar0≤inf⁡Sr0′a^{r_0}\le\inf S'_{r_0}. Both inequalities give f(r0)=ar0f(r_0)=a^{r_0}.

For uniqueness, let Tx′={at∣t∈Q, t>x}T'_x=\{a^t\mid t\in\mathbb{Q},\,t>x\}, nonempty and bounded above by any ara^r with r≤xr\le x, via the same dual monotonicity Lemma, so that sup⁡Tx′\sup T'_x exists. Every element of Tx′T'_x is smaller than every element of Sx′S'_x (if r≤x<tr\le x<t, ar>ata^r>a^t), hence sup⁡Tx′≤inf⁡Sx′=f(x)\sup T'_x\le\inf S'_x=f(x). For the reverse inequality, let δ:=f(x)−sup⁡Tx′\delta:=f(x)-\sup T'_x and suppose δ>0\delta>0. Let m∈Zm\in\mathbb{Z} be the unique integer with m≤x−1<m+1m\le x-1<m+1, which exists by the same well-ordering argument used in the density Proposition of §4, and let M=amM=a^{m}, a fixed bound. Apply the sublemma with ε=δ/(2M)\varepsilon=\delta/(2M): there exists n∈Nn\in\mathbb{N} with a1/n>1−εa^{1/n}>1-\varepsilon. By density, there exists a rational r∈(x−1/n, x]r\in(x-1/n,\,x]; let t=r+1/n>xt=r+1/n>x. Since 1/n≤11/n\le 1, we have r>x−1/n≥x−1>mr>x-1/n\ge x-1>m, so that the dual monotonicity Lemma gives ar<am=Ma^r<a^m=M. Hence

ar−at=ar(1−a1/n)<arε<Mε=δ2a^r-a^t=a^r\left(1-a^{1/n}\right)<a^r\varepsilon<M\varepsilon=\frac{\delta}{2}

Thus, f(x)−sup⁡Tx′≤ar−at<δf(x)-\sup T'_x\le a^r-a^t<\delta, contradicting the definition of δ\delta. Hence δ=0\delta=0 and f(x)=sup⁡Tx′=inf⁡Sx′f(x)=\sup T'_x=\inf S'_x.

From this equality uniqueness follows exactly as in §6: if gg is decreasing and coincides with ara^r on Q\mathbb{Q}, then for r≤xr\le x it is g(x)≤g(r)=arg(x)\le g(r)=a^r, an upper bound of g(x)g(x) over Sx′S'_x inverted, so that g(x)≤inf⁡Sx′=f(x)g(x)\le\inf S'_x=f(x); and for t>xt>x, g(x)≥g(t)=atg(x)\ge g(t)=a^t, so that g(x)≥sup⁡Tx′=f(x)g(x)\ge\sup T'_x=f(x). Both give g(x)=f(x)g(x)=f(x). ■\blacksquare


With both Propositions, §6 for a>1a>1 and §7 for 0<a<10<a<1, the function f(x)=axf(x)=a^x is defined and uniquely determined on all of R\mathbb{R}, for any admissible base. What remains is what gives this construction its true operational value: to prove that ff keeps the homomorphism property, f(x+y)=f(x)f(y)f(x+y)=f(x)f(y), beyond Q\mathbb{Q}. That is the content of §8.

§8. The fundamental homomorphism: f(x+y) = f(x)f(y) on ℝ

Section titled “§8. The fundamental homomorphism: f(x+y) = f(x)f(y) on ℝ”

Over Q\mathbb{Q}, the identity ar+s=arasa^{r+s}=a^r a^s is, essentially, the product law for powers of §1.2 §2, extended to rational exponents in §1.2 §4. Extending it to R\mathbb{R} is not automatic: f(x+y)f(x+y) and f(x)f(y)f(x)f(y) are, a priori, two numbers constructed by different procedures, a supremum (or infimum) taken over rationals close to x+yx+y, against a product of two suprema (or infima) taken separately, and an argument is needed to know that they coincide. The tool that makes this possible is a general fact about how the supremum and the infimum interact with the product.

Lemma (supremum of a product). Let A,B⊆(0,∞)A,B\subseteq(0,\infty) be nonempty and bounded above, with α=sup⁡A\alpha=\sup A, β=sup⁡B\beta=\sup B. Then AB={ab∣a∈A, b∈B}AB=\{ab\mid a\in A,\,b\in B\} is bounded above and sup⁡(AB)=αβ\sup(AB)=\alpha\beta.

Proof. For any a∈A, b∈Ba\in A,\,b\in B, 0<a≤α0<a\le\alpha and 0<b≤β0<b\le\beta, and multiplying two inequalities between positives preserves the direction: ab≤αβab\le\alpha\beta. Hence αβ\alpha\beta is an upper bound of ABAB, and sup⁡(AB)≤αβ\sup(AB)\le\alpha\beta.

For the reverse inequality, fix ε>0\varepsilon>0 and take ε′=εα+β>0\varepsilon'=\dfrac{\varepsilon}{\alpha+\beta}>0. By the characterization of the supremum of §1.1 §9, for every ε′>0\varepsilon'>0 there exists an element of the set exceeding sup⁡−ε′\sup-\varepsilon', there exist a∈Aa\in A with a>α−ε′a>\alpha-\varepsilon' and b∈Bb\in B with b>β−ε′b>\beta-\varepsilon'. Then

ab>(α−ε′)(β−ε′)=αβ−ε′(α+β)+(ε′)2=αβ−ε+(ε′)2>αβ−εab>(\alpha-\varepsilon')(\beta-\varepsilon')=\alpha\beta-\varepsilon'(\alpha+\beta)+(\varepsilon')^2=\alpha\beta-\varepsilon+(\varepsilon')^2>\alpha\beta-\varepsilon

so that sup⁡(AB)≥ab>αβ−ε\sup(AB)\ge ab>\alpha\beta-\varepsilon. Since this holds for all ε>0\varepsilon>0, it follows that sup⁡(AB)≥αβ\sup(AB)\ge\alpha\beta. Both inequalities give sup⁡(AB)=αβ\sup(AB)=\alpha\beta. ■\blacksquare

Lemma (infimum of a product). Let A,B⊆(0,∞)A,B\subseteq(0,\infty) be nonempty and bounded below by positive bounds, with α=inf⁡A\alpha=\inf A, β=inf⁡B\beta=\inf B. Then inf⁡(AB)=αβ\inf(AB)=\alpha\beta.

Proof. The inequality inf⁡(AB)≥αβ\inf(AB)\ge\alpha\beta is immediate: a≥α>0a\ge\alpha>0, b≥β>0b\ge\beta>0 give ab≥αβab\ge\alpha\beta for all a∈A,b∈Ba\in A,b\in B, so αβ\alpha\beta is a lower bound of ABAB. For the reverse, fix ε>0\varepsilon>0 and take ε′=min⁡ ⁣(1, ε2(α+β+1))\varepsilon'=\min\!\left(1,\ \dfrac{\varepsilon}{2(\alpha+\beta+1)}\right). By the dual characterization of the infimum, for every ε′>0\varepsilon'>0 there exists an element of the set smaller than inf⁡+ε′\inf+\varepsilon', there exist a∈Aa\in A with a<α+ε′a<\alpha+\varepsilon' and b∈Bb\in B with b<β+ε′b<\beta+\varepsilon'. Since ε′≤1\varepsilon'\le 1, (ε′)2≤ε′(\varepsilon')^2\le\varepsilon', and

ab<(α+ε′)(β+ε′)=αβ+ε′(α+β)+(ε′)2≤αβ+ε′(α+β+1)≤αβ+ε2<αβ+εab<(\alpha+\varepsilon')(\beta+\varepsilon')=\alpha\beta+\varepsilon'(\alpha+\beta)+(\varepsilon')^2\le\alpha\beta+\varepsilon'(\alpha+\beta+1)\le\alpha\beta+\frac{\varepsilon}{2}<\alpha\beta+\varepsilon

so that inf⁡(AB)≤ab<αβ+ε\inf(AB)\le ab<\alpha\beta+\varepsilon. Since this holds for all ε>0\varepsilon>0, it follows that inf⁡(AB)≤αβ\inf(AB)\le\alpha\beta. Both inequalities give inf⁡(AB)=αβ\inf(AB)=\alpha\beta. ■\blacksquare

With these two lemmas, the proof of the homomorphism reduces to a question of set inclusion.

Proposition (fundamental homomorphism). Let a>0a>0, a≠1a\neq 1, and f:R→R+f:\mathbb{R}\to\mathbb{R}^+ the extension constructed in §6 (if a>1a>1) or in §7 (if 0<a<10<a<1). Then, for all x,y∈Rx,y\in\mathbb{R},

f(x+y)=f(x) f(y)f(x+y)=f(x)\,f(y)

Proof. We treat the case a>1a>1; the case 0<a<10<a<1 is analogous, exchanging the roles of SS and TT, as detailed at the end.

Case a>1a>1. Recall, from §6, the sets Sx={ar∣r∈Q,r≤x}S_x=\{a^r\mid r\in\mathbb{Q},r\le x\} and Tx={at∣t∈Q,t>x}T_x=\{a^t\mid t\in\mathbb{Q},t>x\}, with f(x)=sup⁡Sx=inf⁡Txf(x)=\sup S_x=\inf T_x.

If r∈Sxr\in S_x and s∈Sys\in S_y, that is, r≤xr\le x, s≤ys\le y, both rational, then r+s≤x+yr+s\le x+y is rational, so that aras=ar+s∈Sx+ya^{r}a^{s}=a^{r+s}\in S_{x+y}. This shows SxSy⊆Sx+yS_xS_y\subseteq S_{x+y}, and therefore sup⁡(SxSy)≤sup⁡Sx+y=f(x+y)\sup(S_xS_y)\le\sup S_{x+y}=f(x+y). By the Lemma of the supremum of a product, sup⁡(SxSy)=sup⁡Sx⋅sup⁡Sy=f(x)f(y)\sup(S_xS_y)=\sup S_x\cdot\sup S_y=f(x)f(y). Hence

f(x)f(y)≤f(x+y)f(x)f(y)\le f(x+y)

Symmetrically, if t∈Txt\in T_x and u∈Tyu\in T_y, that is, t>xt>x, u>yu>y, then t+u>x+yt+u>x+y is rational, so that atau=at+u∈Tx+ya^t a^u=a^{t+u}\in T_{x+y}. This shows TxTy⊆Tx+yT_xT_y\subseteq T_{x+y}, and therefore inf⁡(TxTy)≥inf⁡Tx+y=f(x+y)\inf(T_xT_y)\ge\inf T_{x+y}=f(x+y). By the Lemma of the infimum of a product, inf⁡(TxTy)=inf⁡Tx⋅inf⁡Ty=f(x)f(y)\inf(T_xT_y)=\inf T_x\cdot\inf T_y=f(x)f(y). Hence

f(x)f(y)≥f(x+y)f(x)f(y)\ge f(x+y)

Both inequalities together give f(x+y)=f(x)f(y)f(x+y)=f(x)f(y).

Case 0<a<10<a<1. With Sx′={ar∣r∈Q,r≤x}S'_x=\{a^r\mid r\in\mathbb{Q},r\le x\} and Tx′={at∣t∈Q,t>x}T'_x=\{a^t\mid t\in\mathbb{Q},t>x\} of §7, now f(x)=inf⁡Sx′=sup⁡Tx′f(x)=\inf S'_x=\sup T'_x. The same inclusion Sx′Sy′⊆Sx+y′S'_xS'_y\subseteq S'_{x+y}, via r+s≤x+yr+s\le x+y, gives, by the Lemma of the infimum of a product, inf⁡(Sx′Sy′)=f(x)f(y)≥inf⁡Sx+y′=f(x+y)\inf(S'_xS'_y)=f(x)f(y)\ge\inf S'_{x+y}=f(x+y), that is, f(x)f(y)≥f(x+y)f(x)f(y)\ge f(x+y). And the inclusion Tx′Ty′⊆Tx+y′T'_xT'_y\subseteq T'_{x+y}, via t+u>x+yt+u>x+y, gives, by the Lemma of the supremum of a product, sup⁡(Tx′Ty′)=f(x)f(y)≤sup⁡Tx+y′=f(x+y)\sup(T'_xT'_y)=f(x)f(y)\le\sup T'_{x+y}=f(x+y), that is, f(x)f(y)≤f(x+y)f(x)f(y)\le f(x+y). Both inequalities together give, again, f(x+y)=f(x)f(y)f(x+y)=f(x)f(y). ■\blacksquare

As an immediate corollary, and without need of a new proof, being a direct algebraic consequence of the homomorphism already proved, we obtain the general laws of exponents in R\mathbb{R}: ax−y=ax/aya^{x-y}=a^x/a^y (taking y′=−yy'=-y and using ax=a(x−y)+y=ax−yaya^{x}=a^{(x-y)+y}=a^{x-y}a^y), (ax)y=axy(a^x)^y=a^{xy}, and (ab)x=axbx(ab)^x=a^xb^x for a,b>0a,b>0. The verification of each is left as an exercise in §12.


With the homomorphism established, it remains to characterize how ff behaves as a function: its strict monotonicity on all of R\mathbb{R}, not only on Q\mathbb{Q} as until now, its injectivity, and the exact sets it occupies as domain and range. That is the content of §9.

§9. Strict monotonicity and injectivity on ℝ

Section titled “§9. Strict monotonicity and injectivity on ℝ”

The Lemmas of §5 and §7 proved the monotonicity of ara^r only for rational rr. It remains to extend it to every pair of reals, which is not automatic: that a function increase on every pair of rationals is not enough, by itself, to guarantee that it increase on every pair of reals, unless it is proved.

Proposition (strict monotonicity). Let ff be the extension constructed in §6 or in §7. If a>1a>1, then ff is strictly increasing on R\mathbb{R}: x<y  ⟹  f(x)<f(y)x<y\implies f(x)<f(y). If 0<a<10<a<1, then ff is strictly decreasing: x<y  ⟹  f(x)>f(y)x<y\implies f(x)>f(y).

Proof. Let x,y∈Rx,y\in\mathbb{R} with x<yx<y. By the density Proposition of §4, there exists a rational r1∈(x,y)r_1\in(x,y); applied again to the pair (r1,y)(r_1,y), there exists a rational r2∈(r1,y)r_2\in(r_1,y). In particular, x<r1<r2<yx<r_1<r_2<y, with r1,r2∈Qr_1,r_2\in\mathbb{Q}.

Case a>1a>1. Since r1>xr_1>x, r1∈Txr_1\in T_x and therefore f(x)=inf⁡Tx≤ar1f(x)=\inf T_x\le a^{r_1}. Since r2<yr_2<y, r2∈Syr_2\in S_y and therefore ar2≤sup⁡Sy=f(y)a^{r_2}\le\sup S_y=f(y). Since r1<r2r_1<r_2, the rational monotonicity Lemma of §5 gives ar1<ar2a^{r_1}<a^{r_2}. Chaining the three relations,

f(x) ≤ ar1 < ar2 ≤ f(y)f(x)\ \le\ a^{r_1}\ <\ a^{r_2}\ \le\ f(y)

whence f(x)<f(y)f(x)<f(y).

Case 0<a<10<a<1. Since r1>xr_1>x, r1∈Tx′r_1\in T'_x and therefore ar1≤sup⁡Tx′=f(x)a^{r_1}\le\sup T'_x=f(x). Since r2<yr_2<y, r2∈Sy′r_2\in S'_y and therefore f(y)=inf⁡Sy′≤ar2f(y)=\inf S'_y\le a^{r_2}. Since r1<r2r_1<r_2, the dual monotonicity Lemma of §7 gives ar1>ar2a^{r_1}>a^{r_2}. Chaining,

f(y) ≤ ar2 < ar1 ≤ f(x)f(y)\ \le\ a^{r_2}\ <\ a^{r_1}\ \le\ f(x)

whence f(y)<f(x)f(y)<f(x), that is, f(x)>f(y)f(x)>f(y). ■\blacksquare

Corollary (injectivity). ff is injective on R\mathbb{R}: if x≠yx\neq y, then f(x)≠f(y)f(x)\neq f(y).

Proof. Without loss of generality, x<yx<y. By the previous Proposition, f(x)<f(y)f(x)<f(y) or f(x)>f(y)f(x)>f(y) according as a>1a>1 or 0<a<10<a<1; in either case, f(x)≠f(y)f(x)\neq f(y). ■\blacksquare


With strict monotonicity and injectivity proved, what remains is to determine the exact domain and range of ff, the latter, in particular, requiring a genuinely new argument, and the behavior of ff at the ends of the domain. That is the content of §10.

Proposition (domain). Dom⁡(f)=R\operatorname{Dom}(f)=\mathbb{R}.

Proof. Immediate: the good-definition part of the Propositions of §6 and §7 showed that f(x)f(x) exists, as a supremum or as an infimum, according to the case, for each x∈Rx\in\mathbb{R}, without exception. ■\blacksquare

The range is the nontrivial part. Before tackling it, a lemma is needed on the growth of integer powers, which supports both surjectivity and, in §11, the asymptote.

Lemma (unboundedness of integer powers). Let b>1b>1. The set {bn∣n∈N}\{b^n\mid n\in\mathbb{N}\} is not bounded above in R\mathbb{R}.

Proof. Let h=b−1>0h=b-1>0. By Bernoulli’s inequality, already proved in §6, bn=(1+h)n≥1+nhb^n=(1+h)^n\ge 1+nh for all n∈Nn\in\mathbb{N}. Given any M∈RM\in\mathbb{R}, the Archimedean Lemma of §4 gives a natural number n>(M−1)/hn>(M-1)/h, whence 1+nh>M1+nh>M, and therefore bn≥1+nh>Mb^n\ge 1+nh>M. Since MM was arbitrary, the set has no upper bound. ■\blacksquare

This lemma, applied to b=ab=a when a>1a>1, and to b=1/ab=1/a when 0<a<10<a<1, a case in which 1/a>11/a>1, covers both regimes of the base without need of a second proof: it is a purely arithmetic fact about numbers greater than 11, independent of which exponential function is being constructed with it.

Proposition (range). Rec⁡(f)=(0,∞)\operatorname{Rec}(f)=(0,\infty).

Proof. The inclusion Rec⁡(f)⊆(0,∞)\operatorname{Rec}(f)\subseteq(0,\infty) was already proved: both the supremum and the infimum of a set of positive reals are positive, as established in the good definition of §6 and §7. The reverse inclusion remains: given any y>0y>0, exhibit x∈Rx\in\mathbb{R} with f(x)=yf(x)=y.

Case a>1a>1. Let E={x∈R∣f(x)<y}E=\{x\in\mathbb{R}\mid f(x)<y\}. It is nonempty: by the previous Lemma there exists n∈Nn\in\mathbb{N} with an>1/ya^n>1/y, whence f(−n)=a−n=1/an<yf(-n)=a^{-n}=1/a^n<y, using the rational consistency of §6, that is, −n∈E-n\in E. It is bounded above: there exists n∈Nn\in\mathbb{N} with an>ya^n>y, and by the strict monotonicity of §9, every x>nx>n satisfies f(x)>f(n)=an>yf(x)>f(n)=a^n>y, so that no such xx belongs to EE; hence nn is an upper bound. By the supremum axiom, x0=sup⁡Ex_0=\sup E exists in R\mathbb{R}.

We claim f(x0)=yf(x_0)=y. Suppose first f(x0)<yf(x_0)<y. By the homomorphism of §8 and rational consistency, for any n∈Nn\in\mathbb{N},

f ⁣(x0+1n)=f(x0) f ⁣(1n)=f(x0)⋅a1/nf\!\left(x_0+\tfrac{1}{n}\right)=f(x_0)\,f\!\left(\tfrac{1}{n}\right)=f(x_0)\cdot a^{1/n}

Since y/f(x0)>1y/f(x_0)>1, because f(x0)<yf(x_0)<y and f(x0)>0f(x_0)>0, the approximation of unity sublemma of §6 gives an nn with a1/n<y/f(x0)a^{1/n}<y/f(x_0), whence f(x0+1/n)<f(x0)⋅(y/f(x0))=yf(x_0+1/n)<f(x_0)\cdot(y/f(x_0))=y. Thus, x0+1/n∈Ex_0+1/n\in E and x0+1/n>x0x_0+1/n>x_0, contradicting that x0x_0 is an upper bound of EE.

Suppose, instead, f(x0)>yf(x_0)>y. By the same homomorphism,

f ⁣(x0−1n)=f(x0)⋅a−1/n=f(x0)a1/nf\!\left(x_0-\tfrac{1}{n}\right)=f(x_0)\cdot a^{-1/n}=\frac{f(x_0)}{a^{1/n}}

Since f(x0)/y>1f(x_0)/y>1, the approximation of unity sublemma gives an nn with a1/n<f(x0)/ya^{1/n}<f(x_0)/y, whence f(x0−1/n)>f(x0)/(f(x0)/y)=yf(x_0-1/n)>f(x_0)/(f(x_0)/y)=y. If there existed x∈Ex\in E with x>x0−1/nx>x_0-1/n, the strict monotonicity of §9 would give f(x)>f(x0−1/n)>yf(x)>f(x_0-1/n)>y, contradicting x∈Ex\in E (which requires f(x)<yf(x)<y). Hence no element of EE exceeds x0−1/nx_0-1/n: it is an upper bound of EE, and x0−1/n<x0x_0-1/n<x_0, contradicting that x0x_0 is the least upper bound.

Both strict inequalities having been ruled out, f(x0)=yf(x_0)=y.

Case 0<a<10<a<1. Let E′={x∈R∣f(x)>y}E'=\{x\in\mathbb{R}\mid f(x)>y\}. By an argument symmetric to the previous one, using the Lemma applied to b=1/ab=1/a to obtain a−na^{-n} arbitrarily large and ana^n arbitrarily small, E′E' is nonempty and bounded above, so that x0=sup⁡E′x_0=\sup E' exists.

If f(x0)>yf(x_0)>y: by the homomorphism, f(x0+1/n)=f(x0)⋅a1/nf(x_0+1/n)=f(x_0)\cdot a^{1/n}. By the approximation of unity sublemma of §7, there exists nn with a1/n>1−εa^{1/n}>1-\varepsilon for ε=(f(x0)−y)/f(x0)>0\varepsilon=(f(x_0)-y)/f(x_0)>0, whence f(x0+1/n)>f(x0)(1−ε)=yf(x_0+1/n)>f(x_0)(1-\varepsilon)=y. Thus x0+1/n∈E′x_0+1/n\in E', exceeding x0x_0: a contradiction.

If f(x0)<yf(x_0)<y: by the homomorphism, f(x0−1/n)=f(x0)⋅a−1/nf(x_0-1/n)=f(x_0)\cdot a^{-1/n}. Applying the approximation of unity sublemma of §6, this time to b=1/a>1b=1/a>1, which gives b1/n=a−1/nb^{1/n}=a^{-1/n} arbitrarily close to 11 from above, there exists nn with a−1/n<1+ε′a^{-1/n}<1+\varepsilon' for ε′=(y−f(x0))/f(x0)>0\varepsilon'=(y-f(x_0))/f(x_0)>0, whence f(x0−1/n)<f(x0)(1+ε′)=yf(x_0-1/n)<f(x_0)(1+\varepsilon')=y. By the strict decreasing monotonicity of §9, every x>x0−1/nx>x_0-1/n satisfies f(x)<f(x0−1/n)<yf(x)<f(x_0-1/n)<y, so that no such xx is in E′E': x0−1/nx_0-1/n is an upper bound of E′E', smaller than x0x_0: a contradiction.

In both cases, f(x0)=yf(x_0)=y. ■\blacksquare

With domain and range established, and f:R→(0,∞)f:\mathbb{R}\to(0,\infty) being bijective by §9 and this Proposition, the piece that §4.3 will need in order to define the logarithm as an inverse function is fixed.


It remains to describe how ff behaves at the ends of its domain, behavior that, unlike the range, requires no new tool: the unboundedness Lemma of §10 suffices.

Proposition (horizontal asymptote). Let a>1a>1. For every ε>0\varepsilon>0 there exists N∈NN\in\mathbb{N} such that x<−N  ⟹  f(x)<εx<-N\implies f(x)<\varepsilon. Moreover, ff is not bounded above: for every M∈RM\in\mathbb{R} there exists N′∈NN'\in\mathbb{N} such that x>N′  ⟹  f(x)>Mx>N'\implies f(x)>M.

Proof. Given ε>0\varepsilon>0, the Lemma of §10 gives N∈NN\in\mathbb{N} with aN>1/εa^N>1/\varepsilon. If x<−Nx<-N, the strict monotonicity of §9 gives f(x)<f(−N)=a−N=1/aN<εf(x)<f(-N)=a^{-N}=1/a^N<\varepsilon; and since f(x)>0f(x)>0 always, ∣f(x)−0∣=f(x)<ε|f(x)-0|=f(x)<\varepsilon. Given M∈RM\in\mathbb{R}, the same Lemma gives N′∈NN'\in\mathbb{N} with aN′>Ma^{N'}>M; if x>N′x>N', strict monotonicity gives f(x)>f(N′)=aN′>Mf(x)>f(N')=a^{N'}>M. ■\blacksquare

The case 0<a<10<a<1 is exactly the mirror: ff decreases, so that the unbounded behavior occurs on the left and the approach to zero on the right.

Proposition (horizontal asymptote, case 0<a<10<a<1). Let 0<a<10<a<1. For every ε>0\varepsilon>0 there exists N∈NN\in\mathbb{N} such that x>N  ⟹  f(x)<εx>N\implies f(x)<\varepsilon. Moreover, for every M∈RM\in\mathbb{R} there exists N′∈NN'\in\mathbb{N} such that x<−N′  ⟹  f(x)>Mx<-N'\implies f(x)>M.

Proof. Apply the Lemma of §10 to b=1/a>1b=1/a>1: given ε>0\varepsilon>0, there exists NN with (1/a)N>1/ε(1/a)^N>1/\varepsilon, that is, aN<εa^N<\varepsilon. If x>Nx>N, the strict decreasing monotonicity of §9 gives f(x)<f(N)=aN<εf(x)<f(N)=a^N<\varepsilon. Given MM, there exists N′N' with (1/a)N′>M(1/a)^{N'}>M, that is, a−N′>Ma^{-N'}>M; if x<−N′x<-N', strict monotonicity gives f(x)>f(−N′)=a−N′>Mf(x)>f(-N')=a^{-N'}>M. ■\blacksquare


With this the analytic characterization of the exponential function is complete: existence and uniqueness of its extension to R\mathbb{R} (§6–§7), its homomorphism (§8), its strict monotonicity and injectivity (§9), its domain and range (§10), and its asymptotic behavior (§11). What remains is the block of worked exercises and proofs, and the closing reflection on the passage from discrete algebra to continuous analysis, the content of §12.

Part A — Worked exercises and complete proofs

Section titled “Part A — Worked exercises and complete proofs”

Exercise 1. Prove that ax−y=ax/aya^{x-y}=a^x/a^y for all x,y∈Rx,y\in\mathbb{R}.

Solution. By the homomorphism of §8, applied to u=x−yu=x-y and v=yv=y,

ax=a(x−y)+y=ax−y⋅aya^x=a^{(x-y)+y}=a^{x-y}\cdot a^y

Since ay>0a^y>0, the range of ff, §10, both sides can be divided by aya^y without altering the equality: ax−y=ax/aya^{x-y}=a^x/a^y. ■\blacksquare

The other two deferred corollaries, (ax)y=axy(a^x)^y=a^{xy} and (ab)x=axbx(ab)^x=a^xb^x, do not follow from a single algebraic step: they ask us to compare two functions constructed by different procedures (one a composition of extensions, the other a product of extensions), and a general criterion is needed to know when two such constructions coincide. That criterion is developed before using it.

Lemma (continuity at 00 of every exponential extension). Let c>0c>0. For every ε>0\varepsilon>0 there exists δ>0\delta>0 such that, for every h∈Rh\in\mathbb{R} with ∣h∣<δ|h|<\delta, ∣ch−1∣<ε|c^h-1|<\varepsilon.

Proof. If c=1c=1, the claim is trivial: ch=1c^h=1 for all hh. Suppose c≠1c\neq 1; without loss of generality, c>1c>1; if 0<c<10<c<1, the argument is the mirror, using the sublemma of §7 instead of that of §6. By the approximation of unity sublemma of §6, given ε>0\varepsilon>0, there exists n∈Nn\in\mathbb{N} with c1/n<1+εc^{1/n}<1+\varepsilon; for the same reason, applied to 1/c>11/c>1, or directly by rational monotonicity of §5, which gives c−1/n<1c^{-1/n}<1 and, together with the approximation, c−1/n>1−εc^{-1/n}>1-\varepsilon for nn sufficiently large, by an identical argument, also c−1/n>1−εc^{-1/n}>1-\varepsilon by choosing nn larger if necessary. Let δ=1/n\delta=1/n. For ∣h∣<δ|h|<\delta, −1/n<h<1/n-1/n<h<1/n, and by the strict monotonicity of §9, c−1/n<ch<c1/nc^{-1/n}<c^h<c^{1/n}. Combining, 1−ε<ch<1+ε1-\varepsilon<c^h<1+\varepsilon, that is, ∣ch−1∣<ε|c^h-1|<\varepsilon. ■\blacksquare

Lemma (uniqueness of homomorphisms continuous at 00). Let φ,ψ:R→(0,∞)\varphi,\psi:\mathbb{R}\to(0,\infty) be such that φ(u+v)=φ(u)φ(v)\varphi(u+v)=\varphi(u)\varphi(v) and ψ(u+v)=ψ(u)ψ(v)\psi(u+v)=\psi(u)\psi(v) for all u,v∈Ru,v\in\mathbb{R}, that coincide on Q\mathbb{Q}, and that are both continuous at 00 in the sense of the previous Lemma. Then φ=ψ\varphi=\psi on R\mathbb{R}.

Proof. Every homomorphism φ:R→(0,∞)\varphi:\mathbb{R}\to(0,\infty) satisfies φ(0)=1\varphi(0)=1, since φ(0)=φ(0+0)=φ(0)2\varphi(0)=\varphi(0+0)=\varphi(0)^2, and as φ(0)>0\varphi(0)>0, the only solution of t=t2t=t^2 with t>0t>0 is t=1t=1, and φ(x)φ(−x)=φ(0)=1\varphi(x)\varphi(-x)=\varphi(0)=1, so that φ(x)>0\varphi(x)>0 never vanishes, consistent with the hypothesis. Define ρ(x)=φ(x)/ψ(x)\rho(x)=\varphi(x)/\psi(x), well defined since ψ(x)>0\psi(x)>0. It is a homomorphism: ρ(u+v)=φ(u)φ(v)ψ(u)ψ(v)=ρ(u)ρ(v)\rho(u+v)=\dfrac{\varphi(u)\varphi(v)}{\psi(u)\psi(v)}=\rho(u)\rho(v). It equals 11 on Q\mathbb{Q}: if r∈Qr\in\mathbb{Q}, φ(r)=ψ(r)\varphi(r)=\psi(r) by hypothesis, hence ρ(r)=1\rho(r)=1. And it is continuous at 00 with value 11: given ε>0\varepsilon>0, take ε′=min⁡(ε/3,1/2)\varepsilon'=\min(\varepsilon/3,1/2); by the continuity at 00 of φ\varphi and ψ\psi, there exists δ>0\delta>0 such that ∣h∣<δ|h|<\delta gives φ(h),ψ(h)∈(1−ε′,1+ε′)\varphi(h),\psi(h)\in(1-\varepsilon',1+\varepsilon'), whence ρ(h)=φ(h)/ψ(h)∈(1−ε′1+ε′,1+ε′1−ε′)\rho(h)=\varphi(h)/\psi(h)\in\left(\dfrac{1-\varepsilon'}{1+\varepsilon'},\dfrac{1+\varepsilon'}{1-\varepsilon'}\right), an interval contained in (1−ε,1+ε)(1-\varepsilon,1+\varepsilon) for ε′≤ε/3\varepsilon'\le\varepsilon/3, a direct algebraic verification.

Now fix arbitrary x∈Rx\in\mathbb{R} and ε>0\varepsilon>0. By the continuity at 00 of ρ\rho, there exists δ>0\delta>0 with ∣h∣<δ  ⟹  ∣ρ(h)−1∣<ε|h|<\delta\implies|\rho(h)-1|<\varepsilon. By the density of §4, there exists r∈Qr\in\mathbb{Q} with ∣x−r∣<δ|x-r|<\delta. Then

ρ(x)=ρ(r+(x−r))=ρ(r)⋅ρ(x−r)=1⋅ρ(x−r)=ρ(x−r)\rho(x)=\rho\big(r+(x-r)\big)=\rho(r)\cdot\rho(x-r)=1\cdot\rho(x-r)=\rho(x-r)

using that ρ(r)=1\rho(r)=1. Since ∣x−r∣<δ|x-r|<\delta, it follows that ∣ρ(x)−1∣=∣ρ(x−r)−1∣<ε|\rho(x)-1|=|\rho(x-r)-1|<\varepsilon. Since ε>0\varepsilon>0 was arbitrary, ρ(x)=1\rho(x)=1, that is, φ(x)=ψ(x)\varphi(x)=\psi(x). Since xx was arbitrary, φ=ψ\varphi=\psi. ■\blacksquare

Exercise 2. Prove that (ax)y=axy(a^x)^y=a^{xy} for all x,y∈Rx,y\in\mathbb{R}, understanding (ax)y(a^x)^y as the real extension (§6–§7) applied to the base c=ax>0c=a^x>0.

Solution. Fix x∈Rx\in\mathbb{R}. Define φ(y)=(ax)y\varphi(y)=(a^x)^y and ψ(y)=axy\psi(y)=a^{xy}, both functions of yy. The three hypotheses of the Uniqueness Lemma are verified.

Homomorphism. φ(y1+y2)=(ax)y1+y2=(ax)y1(ax)y2=φ(y1)φ(y2)\varphi(y_1+y_2)=(a^x)^{y_1+y_2}=(a^x)^{y_1}(a^x)^{y_2}=\varphi(y_1)\varphi(y_2), by §8 applied to the base axa^x. ψ(y1+y2)=ax(y1+y2)=axy1+xy2=axy1axy2=ψ(y1)ψ(y2)\psi(y_1+y_2)=a^{x(y_1+y_2)}=a^{xy_1+xy_2}=a^{xy_1}a^{xy_2}=\psi(y_1)\psi(y_2), by §8 applied to the base aa.

Coincidence on Q\mathbb{Q}. Let r=m/n∈Qr=m/n\in\mathbb{Q}. By iterated homomorphism, applying §8 nn times, (ax/n)n=ax/n+⋯+x/n=ax\left(a^{x/n}\right)^n=a^{x/n+\cdots+x/n}=a^x. Thus, ax/na^{x/n} is a positive real whose nnth power is axa^x; by the uniqueness of the principal nnth root (§1.2 §3), ax/n=(ax)1/na^{x/n}=(a^x)^{1/n}. Hence, using again iterated homomorphism and Exercise 1 for negative exponents,

φ(r)=(ax)m/n=((ax)1/n)m=(ax/n)m=amx/n=axr=ψ(r)\varphi(r)=(a^x)^{m/n}=\left((a^x)^{1/n}\right)^m=\left(a^{x/n}\right)^m=a^{mx/n}=a^{xr}=\psi(r)

Continuity at 00. φ\varphi is continuous at 00 by the previous Lemma, applied to c=axc=a^x. For ψ\psi: given ε>0\varepsilon>0, if x=0x=0 the claim (a0)y=1y=1=a0(a^0)^y=1^y=1=a^0 is trivial for all yy; if x≠0x\neq 0, by the continuity at 00 of faf_a there exists δ′\delta' with ∣z∣<δ′  ⟹  ∣az−1∣<ε|z|<\delta'\implies|a^z-1|<\varepsilon, and taking δ=δ′/∣x∣\delta=\delta'/|x|, ∣h∣<δ  ⟹  ∣xh∣<δ′  ⟹  ∣ψ(h)−1∣=∣axh−1∣<ε|h|<\delta\implies|xh|<\delta'\implies|\psi(h)-1|=|a^{xh}-1|<\varepsilon.

By the Uniqueness Lemma, φ=ψ\varphi=\psi, that is, (ax)y=axy(a^x)^y=a^{xy} for all y∈Ry\in\mathbb{R}, and since xx was arbitrary, for all x,y∈Rx,y\in\mathbb{R}. ■\blacksquare

Exercise 3. Prove that (ab)x=axbx(ab)^x=a^xb^x for a,b>0a,b>0 and all x∈Rx\in\mathbb{R}.

Solution. Define φ(x)=(ab)x\varphi(x)=(ab)^x and ψ(x)=axbx\psi(x)=a^xb^x. Homomorphism: φ\varphi is one by construction (§8, base abab, or trivially if ab=1ab=1); ψ(x1+x2)=ax1+x2bx1+x2=(ax1bx1)(ax2bx2)=ψ(x1)ψ(x2)\psi(x_1+x_2)=a^{x_1+x_2}b^{x_1+x_2}=(a^{x_1}b^{x_1})(a^{x_2}b^{x_2})=\psi(x_1)\psi(x_2), using §8 for aa and for bb separately. Coincidence on Q\mathbb{Q}: (ab)r=arbr(ab)^r=a^rb^r for r∈Qr\in\mathbb{Q} is the law of rational exponents already extended in §1.2 §4. Continuity at 00: φ\varphi, by the Lemma, applied to c=abc=ab; ψ(h)=ahbh\psi(h)=a^hb^h, a product of two functions each continuous at 00 with value 11, whence ψ(h)→1\psi(h)\to 1 by a bounding argument analogous to the one used inside the Uniqueness Lemma. By the Uniqueness Lemma, φ=ψ\varphi=\psi: (ab)x=axbx(ab)^x=a^xb^x. ■\blacksquare


Exercise 4 (exponential equation, same base). Solve 23x−1=2x+52^{3x-1}=2^{x+5}.

Solution. By the injectivity of §9, a corollary of strict monotonicity, 23x−1=2x+52^{3x-1}=2^{x+5} if and only if the exponents coincide: 3x−1=x+53x-1=x+5. Hence 2x=62x=6, that is, x=3x=3. Injectivity is what justifies the step, not an informal manipulation: without it, two distinct exponents could, in principle, produce the same value.

Exercise 5 (exponential equation, different reducible bases). Solve 9x=27x−19^x=27^{x-1}.

Solution. Rewrite both bases as powers of 33: 9=329=3^2, 27=3327=3^3. By Exercise 2, (32)x=32x(3^2)^x=3^{2x} and (33)x−1=33(x−1)=33x−3(3^3)^{x-1}=3^{3(x-1)}=3^{3x-3}. The equation reduces to 32x=33x−33^{2x}=3^{3x-3}, and by injectivity (§9), 2x=3x−32x=3x-3, whence x=3x=3.

Exercise 6 (exponential inequality, base between 00 and 11). Solve (12)x>(12)3\left(\tfrac12\right)^x>\left(\tfrac12\right)^3.

Solution. Since 0<12<10<\tfrac12<1, f(x)=(1/2)xf(x)=(1/2)^x is strictly decreasing (§9). For a strictly decreasing function, f(x)>f(3)f(x)>f(3) forces x<3x<3: if it were x=3x=3, the values would coincide by direct substitution; if it were x>3x>3, decreasing monotonicity would give f(x)<f(3)f(x)<f(3), contradicting the hypothesis. Hence the solution is x<3x<3.

Part B — Philosophical close and transition

Section titled “Part B — Philosophical close and transition”

Before closing this section, it is worth taking stock of what really made possible the passage from Q\mathbb{Q} to R\mathbb{R} in exponentiation. It was not a new set of numbers, unlike what was required, in Chapter 3, by the passage from R\mathbb{R} to C\mathbb{C}, nor a new operation: it was a single property of R\mathbb{R}, completeness, already fixed as an axiom in §1.1, long before this chapter needed it. All the machinery constructed throughout these twelve sections (supremum, infimum, density, Bernoulli’s inequality, the approximation of unity) is, seen as a whole, a single idea applied over and over from different angles: that a pattern defined on a discrete and perforated set such as Q\mathbb{Q} can be forced, without ambiguity or exception, to occupy every hole of the continuum, provided one has a bound that sustains it from both sides. The discrete algebra of Chapter 1 asked “what is aa raised to this exact fraction worth?”; the analysis of this chapter asks, instead, “what value is obliged, not chosen but obliged, to occupy the place of aa raised to a number that is not a fraction of anything?” The answer, as the uniqueness Propositions of §6 and §7 proved, depends on no arbitrary decision: it is completely determined the moment the base aa is fixed and continuity with what was already constructed on Q\mathbb{Q} is demanded.

It is worth noting that this rigor was late in arriving, and that its absence did not prevent, for more than a century, the exponential function from being used with complete confidence. Euler, in his Introductio in analysin infinitorum of 1748, wrote expressions such as axa^x for irrational xx without stopping to ask what it meant exactly to raise a number to a power that admits no representation as a quotient of integers; he relied, implicitly, on the same intuition of “approximating by rationals” that this chapter formalized step by step, but without the completeness axiom, which would not be stated precisely until Dedekind and Cantor, more than a century later, to sustain that intuition with a proof. Euler’s practical result was correct; what was missing was not the calculation, but the guarantee that that calculation had, indeed, a single possible result. Closing that gap, between the fruitful use of an object and the certainty that this object exists and is unique, was, to a great extent, the program of real analysis in the nineteenth century, and this chapter, in miniature, repeated that program: it started from a function that was already being used with ease on Q\mathbb{Q} and demanded of it, before moving on, that it give an account of its own existence on R\mathbb{R}.

That demand, to give an account before operating, is not exclusive to exponentiation, and it will not be the last time this book applies it. Whenever an operation defined on a discrete set must be extended to a continuum, the general strategy is the same: bound from both sides with elements of the discrete set, invoke completeness to guarantee a meeting point, and verify that this meeting point does not depend on how it was approximated. The logarithm of §4.3, being defined as the inverse of a function already continuous on its entire domain, will inherit for free a good part of this work; but the methodological lesson, that the continuum is not reached by decree, but by a specific proof in each case, is worth carrying, explicitly, to the rest of the chapter.

One question remains, however, that this chapter deliberately left open, because it was not its place to close it. Among all possible bases, a∈(0,1)∪(1,∞)a\in(0,1)\cup(1,\infty), each gives rise to a genuinely different exponential function, with its own rate of growth or decay; §12 already showed how to compare two of them by means of (ax)y=axy(a^x)^y=a^{xy}, but it said nothing about whether some base occupies a privileged place among all. The question worth formulating, and leaving resonating until §4.2, is this: does there exist, among that infinity of bases, one that is distinguished not by typographic convention or computational convenience, as choosing a=10a=10 might seem, by affinity with the decimal system, but by a genuine structural property, internal to the function itself? Concretely: is there a base for which the speed with which the function grows at each point coincides, exactly, with the value the function has at that point? Formulated thus, the question does not yet have rigorous meaning, since it must first be made precise what “speed of growth at a point” means for a function that is neither linear nor polynomial, a notion that this book has not yet constructed, but its mere formulation already suggests that, if such a base exists, it will not be chosen: it will be discovered, in the same way that 2\sqrt{2} was not chosen to solve x2=2x^2=2, but was proved, in §1.1, to have to exist.


With §12 section 4.1 is closed. §4.2 takes up the question just posed in order to construct, from it, the natural exponential function and the number ee.