Skip to content

Chapter 4 — Exponential and Logarithmic Functions

4.2 - The Natural Exponential Function

Among all possible bases, is there one that arithmetic itself singles out as distinguished, not by computational convenience, but because a concrete construction forces it to exist?


§4.1 closed with a question that, formulated precisely, still made no sense in this book: whether there exists a base aa for which the speed of growth of f(x)=axf(x)=a^x coincides, at each point, with the value of the function itself. That question requires, in order to be answered, the concept of derivative, and this book does not yet have it. But there is a second question, more modest and entirely verifiable with what has already been constructed, that points to the same number without needing that machinery: what happens to a quantity that grows by interest, when the interest is compounded ever more frequently?

§1. From the open question to a verifiable construction

Section titled “§1. From the open question to a verifiable construction”

Suppose a capital of one unit, invested for one year at an annual rate of one hundred percent, figures deliberately simple, chosen to isolate the mathematical phenomenon from any accessory financial detail. If the interest is compounded only once, at the end of the year we have 1+1=21+1=2 units. If it is compounded twice, half the rate, twice in the year, we have (1+12)2=2.25\left(1+\tfrac12\right)^2=2.25. If it is compounded nn times, each compounding applies a fraction 1/n1/n of the annual rate on the capital already accumulated, and the result at the end of the year is

an=(1+1n)na_n=\left(1+\frac{1}{n}\right)^n

The financial question, which is better, compounding little or much?, is, strictly, a question about this sequence of rational numbers indexed by n∈Nn\in\mathbb{N}: whether ana_n grows without bound as nn grows, so that compounding more frequently would always be strictly better, without limit, or whether, instead, ana_n approaches a fixed value, however many times the year is subdivided.

This is, incidentally, a question of the same type as those that the catalog of §2.4 already resolved: a sequence of values, and the matter of whether it grows without bound or stabilizes. The difference with that catalog is that there the variable ran over a continuum (x→+∞x\to+\infty over the reals) and here it runs only over the naturals, a sequence, not a function of a real variable; but the conceptual apparatus, as that same section made precise in its definitive form, is the same: in both cases it is a limit, and in the catalog of §2.4 it was already made explicit that the name changes nothing of the technique. Here, the technique that corresponds is not ϵ\epsilon-NN applied to a function already defined on R\mathbb{R}, but something closer to completeness itself: to show that (an)(a_n) is bounded and monotone, and to appeal to the supremum axiom of §1.1 §9, the same tool that constructed, from scratch, the whole extension of axa^x in §4.1.


Before deciding whether (an)(a_n) is bounded, it is advisable to show that it at least grows, or, more precisely, that it does not decrease, and to find a second sequence that bounds it from above without needing to guess the limit value in advance. That is the content of §2.

Bernoulli’s inequality of §6 was proved for ε>0\varepsilon>0, with non-strict inequality. What follows needs a slightly stronger version, strict, and valid also for negative exponents between −1-1 and 00, which is obtained with the same inductive argument, without a new tool.

Sublemma (Bernoulli, strict version). For x∈(−1,∞)x\in(-1,\infty), x≠0x\neq 0, and n∈Nn\in\mathbb{N}, n≥2n\ge 2, we have (1+x)n>1+nx(1+x)^n>1+nx.

Proof. Base case n=2n=2: (1+x)2=1+2x+x2>1+2x(1+x)^2=1+2x+x^2>1+2x, since x2>0x^2>0 because x≠0x\neq 0. Inductive step: suppose (1+x)n>1+nx(1+x)^n>1+nx for some n≥2n\ge 2. Since x>−1x>-1, 1+x>01+x>0, and multiplying the inductive hypothesis by 1+x>01+x>0 preserves the direction of the inequality:

(1+x)n+1=(1+x)n(1+x)>(1+nx)(1+x)=1+(n+1)x+nx2>1+(n+1)x(1+x)^{n+1}=(1+x)^n(1+x)>(1+nx)(1+x)=1+(n+1)x+nx^2>1+(n+1)x

using once again nx2>0nx^2>0. ■\blacksquare

Proposition (ana_n is strictly increasing). an=(1+1n)na_n=\left(1+\dfrac1n\right)^n satisfies an<an+1a_n<a_{n+1} for all n∈Nn\in\mathbb{N}.

Proof. Apply the Sublemma with x=−1(n+1)2∈(−1,0)x=-\dfrac{1}{(n+1)^2}\in(-1,0) and exponent n+1≥2n+1\ge 2:

(1−1(n+1)2)n+1>1−n+1(n+1)2=1−1n+1=nn+1\left(1-\frac{1}{(n+1)^2}\right)^{n+1}>1-\frac{n+1}{(n+1)^2}=1-\frac{1}{n+1}=\frac{n}{n+1}

The base factors without remainder: 1−1(n+1)2=(n+1)2−1(n+1)2=n(n+2)(n+1)2=nn+1⋅n+2n+11-\dfrac{1}{(n+1)^2}=\dfrac{(n+1)^2-1}{(n+1)^2}=\dfrac{n(n+2)}{(n+1)^2}=\dfrac{n}{n+1}\cdot\dfrac{n+2}{n+1}. Substituting,

(nn+1)n+1(n+2n+1)n+1>nn+1\left(\frac{n}{n+1}\right)^{n+1}\left(\frac{n+2}{n+1}\right)^{n+1}>\frac{n}{n+1}

Dividing both sides by (nn+1)n+1>0\left(\dfrac{n}{n+1}\right)^{n+1}>0,

(n+2n+1)n+1>(nn+1)1−(n+1)=(nn+1)−n=(n+1n)n\left(\frac{n+2}{n+1}\right)^{n+1}>\left(\frac{n}{n+1}\right)^{1-(n+1)}=\left(\frac{n}{n+1}\right)^{-n}=\left(\frac{n+1}{n}\right)^n

The left side is exactly an+1=(1+1n+1)n+1a_{n+1}=\left(1+\dfrac{1}{n+1}\right)^{n+1}, and the right, ana_n. Hence an+1>ana_{n+1}>a_n. ■\blacksquare

Proposition (bnb_n is strictly decreasing). bn=(1+1n)n+1b_n=\left(1+\dfrac1n\right)^{n+1} satisfies bn>bn+1b_n>b_{n+1} for all n∈Nn\in\mathbb{N}.

Proof. Apply the Sublemma with x=1n(n+2)>0x=\dfrac{1}{n(n+2)}>0 and exponent n+2≥2n+2\ge 2, valid for n≥1n\ge 1, since then n+2≥3n+2\ge3:

(1+1n(n+2))n+2>1+n+2n(n+2)=1+1n=n+1n\left(1+\frac{1}{n(n+2)}\right)^{n+2}>1+\frac{n+2}{n(n+2)}=1+\frac{1}{n}=\frac{n+1}{n}

The base also factors without remainder: 1+1n(n+2)=n(n+2)+1n(n+2)=(n+1)2n(n+2)1+\dfrac{1}{n(n+2)}=\dfrac{n(n+2)+1}{n(n+2)}=\dfrac{(n+1)^2}{n(n+2)}. Substituting,

(n+1)2(n+2)[n(n+2)]n+2>n+1n\frac{(n+1)^{2(n+2)}}{\big[n(n+2)\big]^{n+2}}>\frac{n+1}{n}

Multiplying both sides by n[n(n+2)]n+2>0n\big[n(n+2)\big]^{n+2}>0 and solving,

n (n+1)2n+4>(n+1) nn+2(n+2)n+2n\,(n+1)^{2n+4}>(n+1)\,n^{n+2}(n+2)^{n+2}

Dividing both sides by n(n+1)>0n(n+1)>0,

(n+1)2n+3>nn+1(n+2)n+2(n+1)^{2n+3}>n^{n+1}(n+2)^{n+2}

This is, precisely, the inequality bn>bn+1b_n>b_{n+1} rewritten: indeed, bnbn+1=(n+1)n+1/nn+1(n+2)n+2/(n+1)n+2=(n+1)2n+3nn+1(n+2)n+2\dfrac{b_n}{b_{n+1}}=\dfrac{(n+1)^{n+1}/n^{n+1}}{(n+2)^{n+2}/(n+1)^{n+2}}=\dfrac{(n+1)^{2n+3}}{n^{n+1}(n+2)^{n+2}}, and the inequality just obtained says exactly that this quotient is greater than 11. Hence bn>bn+1b_n>b_{n+1}. ■\blacksquare


It remains to show that both sequences not only move in opposite directions, but squeeze against each other: that no term of (an)(a_n) ever exceeds any term of (bn)(b_n), whatever the indices. That crossed bound is what makes it possible, in §4, to appeal to the completeness axiom. It is the content of §3.

§3. The crossed bound: aₙ < bₘ for every pair of indices

Section titled “§3. The crossed bound: aₙ < bₘ for every pair of indices”

Proposition. For all n,m∈Nn,m\in\mathbb{N}, an<bma_n<b_m.

Proof. The equal-index case is proved first, and then extended with the monotonicity already established.

Same index. bn=an⋅(1+1n)>an⋅1=anb_n=a_n\cdot\left(1+\dfrac1n\right)>a_n\cdot 1=a_n, since 1+1/n>11+1/n>1 and an>0a_n>0.

Different indices. Let k=max⁡(n,m)k=\max(n,m). Since (aj)(a_j) is strictly increasing (§2), n≤kn\le k gives an≤aka_n\le a_k, with equality only if n=kn=k. By the equal-index case, ak<bka_k<b_k. Since (bj)(b_j) is strictly decreasing (§2), m≤km\le k gives bk≤bmb_k\le b_m, with equality only if m=km=k. Chaining,

an ≤ ak < bk ≤ bma_n\ \le\ a_k\ <\ b_k\ \le\ b_m

whence an<bma_n<b_m. ■\blacksquare


With (an)(a_n) bounded above by any bmb_m, the completeness axiom guarantees that its supremum exists. It remains to check that this supremum coincides, moreover, with the infimum of (bn)(b_n), without which the construction would leave a margin of ambiguity. That is the content of §4, and with it, the definition of the number ee.

By the Proposition of §3, the set {an∣n∈N}\{a_n\mid n\in\mathbb{N}\} is nonempty and bounded above, for example by b1=4b_1=4. By the supremum axiom (§1.1 §9), there exists

L:=sup⁡{an∣n∈N}∈RL:=\sup\{a_n\mid n\in\mathbb{N}\}\in\mathbb{R}

By the same Proposition, {bn∣n∈N}\{b_n\mid n\in\mathbb{N}\} is bounded below by any ana_n, so that, by the Lemma on the existence of the infimum (section 4.1, §7), there exists ℓ:=inf⁡{bn∣n∈N}\ell:=\inf\{b_n\mid n\in\mathbb{N}\}. Since each ana_n is a lower bound of (bn)(b_n), and LL is the least upper bound of (an)(a_n) while each bmb_m is an upper bound of (an)(a_n), by §3, we have L≤bmL\le b_m for all mm, whence L≤ℓL\le\ell. The reverse inequality remains.

Proposition. L=ℓL=\ell.

Proof. The distance between ana_n and bnb_n is bounded:

bn−an=an(1+1n)−an=annb_n-a_n=a_n\left(1+\frac1n\right)-a_n=\frac{a_n}{n}

Since an<b1=4a_n<b_1=4 for all nn, a particular case of §3, it follows that bn−an<4nb_n-a_n<\dfrac{4}{n}. Given ε>0\varepsilon>0, by the Archimedean Lemma (section 4.1, §4) there exists N∈NN\in\mathbb{N} with N>4/εN>4/\varepsilon, so that for all n≥Nn\ge N,

bn−an<4n≤4N<εb_n-a_n<\frac4n\le\frac4N<\varepsilon

Since an≤La_n\le L, by definition of supremum, and L≤ℓ≤bnL\le\ell\le b_n, already proved, plus the definition of infimum, we have, for that same n≥Nn\ge N,

0≤ℓ−L≤bn−an<ε0\le\ell-L\le b_n-a_n<\varepsilon

Since ε>0\varepsilon>0 was arbitrary, ℓ−L=0\ell-L=0, that is, L=ℓL=\ell. ■\blacksquare

Definition. The number ee is defined as the common value

e:=sup⁡{an∣n∈N}=inf⁡{bn∣n∈N}e:=\sup\{a_n\mid n\in\mathbb{N}\}=\inf\{b_n\mid n\in\mathbb{N}\}

With ee defined without ambiguity, what remains is what gives this construction its first contact with numerical intuition: to bound it between two concrete numbers, to confirm that the abstract construction corresponds to the familiar value. That is the content of §5.

Proposition. 2<e<32<e<3.

Proof. For the lower bound: a2=(1+12)2=94=2.25a_2=\left(1+\dfrac12\right)^2=\dfrac94=2.25, and since e=sup⁡{an}≥a2e=\sup\{a_n\}\ge a_2, by the definition of supremum, any element of the set is less than or equal to it, it follows that e≥2.25>2e\ge 2.25>2.

For the upper bound: since e=inf⁡{bn}≤b5e=\inf\{b_n\}\le b_5, by the definition of infimum, it suffices to bound b5b_5. By definition, b5=(1+15)6=(65)6=6656b_5=\left(1+\dfrac15\right)^6=\left(\dfrac65\right)^6=\dfrac{6^6}{5^6}. Computing exactly, 66=466566^6=46656 and 56=156255^6=15625, so that

b5=4665615625b_5=\frac{46656}{15625}

Since 46656<46875=3×1562546656<46875=3\times 15625, it follows that b5<3b_5<3. Hence e≤b5<3e\le b_5<3. ■\blacksquare


With ee defined and bounded, and knowing that e>1e>1, all the machinery of §4.1, homomorphism, strict monotonicity, injectivity, domain, range, asymptote, applies without need of a single new proof. That is the content of §6.

§6. The natural exponential function f(x) = eˣ

Section titled “§6. The natural exponential function f(x) = eˣ”

Since e>1e>1, §5, ee satisfies exactly the hypothesis a>1a>1 under which all the Propositions of section 4.1 were proved. It is not necessary, then, to repeat or adapt a single proof: it suffices to instantiate a=ea=e in each of them.

Definition. The natural exponential function is f(x)=exf(x)=e^x, defined as the continuous extension of ere^r (r∈Qr\in\mathbb{Q}) constructed in §6 of section 4.1, specialized to the base ee.

By that construction and the Propositions that follow it, all applied directly, without modification:

  • Dom⁡(f)=R\operatorname{Dom}(f)=\mathbb{R} and Rec⁡(f)=(0,∞)\operatorname{Rec}(f)=(0,\infty) (§10, section 4.1).
  • ff is strictly increasing and injective on R\mathbb{R} (§9, section 4.1), since e>1e>1.
  • ex+y=exeye^{x+y}=e^x e^y for all x,y∈Rx,y\in\mathbb{R} (§8, section 4.1).
  • (ex)y=exy(e^x)^y=e^{xy} and, for c>0c>0, (ce)x=cxex(ce)^x=c^xe^x (§12, section 4.1, Exercises 2 and 3).
  • y=0y=0 is a horizontal asymptote as x→−∞x\to-\infty, and ff is not bounded above as x→+∞x\to+\infty (§11, section 4.1).

With the natural exponential function completely characterized by direct inheritance, what remains is to exercise its concrete handling before the philosophical close of the section. That is the content of §7.

Exercise 1. Prove that e2<9e^2<9.

Solution. By §6, e2=e⋅ee^2=e\cdot e, and by §12 of section 4.1 (Exercise 2, with x=1,y=2x=1,y=2), e2e^2 coincides with the usual power. Since e<3e<3 (§5), and squaring preserves order among positives, e2<32=9e^2<3^2=9. ■\blacksquare

Exercise 2. Solve e2x−1=ex+3e^{2x-1}=e^{x+3}.

Solution. By the injectivity of f(x)=exf(x)=e^x (§6, inherited from §9 of section 4.1), the equality of the powers is equivalent to the equality of the exponents: 2x−1=x+32x-1=x+3, whence x=4x=4.

Exercise 3. Solve the inequality ex−1≥e2−xe^{x-1}\ge e^{2-x}.

Solution. Since e>1e>1, f(x)=exf(x)=e^x is strictly increasing (§6). For a strictly increasing function, f(u)≥f(v)  ⟺  u≥vf(u)\ge f(v)\iff u\ge v, a direct consequence of the strict monotonicity Proposition of §9, section 4.1. Applied here, x−1≥2−xx-1\ge 2-x, whence 2x≥32x\ge 3, that is, x≥32x\ge\dfrac32.

Exercise 4. Using the bounds of §5, bound e3/2e^{3/2} between two rational numbers.

Solution. By §12 of section 4.1 (Exercise 2), e3/2=(e3)1/2e^{3/2}=(e^3)^{1/2}, the positive square root of e3e^3. Since 2<e<32<e<3 (§5), and cubing preserves order among positives, 8<e3<278<e^3<27. Since the square root also preserves order among positives, §1.2 §3, 8<e3/2<27\sqrt8<e^{3/2}<\sqrt{27}, that is, 22<e3/2<332\sqrt2<e^{3/2}<3\sqrt3.


What remains, to close the section, is to situate philosophically what ee is and what ee cannot yet be within this book. That is the content of §8.

ee was not chosen. Unlike the base 1010, convenient by affinity with the decimal system, or any other base that a text might prefer by custom, ee was determined by a construction that left no margin of decision at any step: two sequences, a monotonicity condition proved with Bernoulli, a crossed bound, and the completeness axiom doing the rest. If some mathematics book had decided never to study compound interest, ee would have appeared anyway, under another disguise, in another problem, because it is not an artifact of the financial question that motivated it, but of the structure of R\mathbb{R} itself, in the same way that 2\sqrt2 does not depend on someone asking about the diagonal of a square.

And yet, this chapter left pending exactly the question that gives ee its central place in analysis: whether there exists a base for which the speed of change of the function coincides, at each point, with the value of the function. §4.1 posed it; this paragraph resolved it, but only halfway: it constructed the candidate number, not the property that would distinguish it among all positive reals. Answering it requires the derivative, and this book, deliberately, does not yet have it. It is not an omission: it is a frontier drawn on purpose, of the same kind as the one that separated, in the close of §2.4, what this book can name from what it can only construct in a Calculus I course.

It is worth noting, before closing, that this frontier has an almost exact historical precedent, although for reasons inverse to ours. Newton, in the Principia Mathematica of 1687, needed to reason about quantities that approached one another, arcs and chords, areas and their polygonal approximations, and he did so by means of what he called first and last ratios: the relation between two geometric quantities at the very instant when their difference became smaller than any given magnitude. He had neither the word “limit” nor an algebraic definition like the ϵ\epsilon-NN that this book already named in §2.4, and yet he arrived, by a geometric and intuitive route, at results that the analysis of the nineteenth century, Cauchy, Weierstrass, would take more than a hundred and fifty years to ground with the rigor demanded today. The difference between that delay and this book’s is one of nature, not of degree: Newton avoided “limit” because the concept did not yet exist; this book avoids it, in the terrain of the derivative, because it decided to construct real analysis from its foundations, one floor at a time, and that floor has not yet arrived. The question that opened §4.1 remains, then, formulated with full precision, with its candidate already constructed and bounded between 22 and 33, and with the complete answer waiting, unhurried, for the chapter that corresponds to it.


With this, section 4.2 is closed. §4.3 takes up the injectivity of f(x)=exf(x)=e^x, and of axa^x in general, already proved in section 4.1, to construct its inverse function: the logarithm.