Skip to content

Chapter 4 — Exponential and Logarithmic Functions

4.7 - Close: Epistemology, History and Scientific Applications


Throughout this chapter three names appeared, one per section and each time as an isolated note: Euler in the close of §4.1, Newton in the close of §4.2, Napier in the close of §4.4. Taken separately, each illustrated a specific point: Euler wrote axa^x for irrational xx without grounding it; Newton reasoned about approximations without the word “limit”; Napier constructed tables of logarithms with neither of the two. Taken together, the three names trace a single line, and it is worth traveling it from end to end before closing the chapter.

Napier published his tables in 1614. He does not start from any exponential function, since the concept of function, in the set-theoretic sense of §2.1, would take more than two centuries to be formalized, but from two progressions, one arithmetic and one geometric, compared term by term; his motivation is entirely instrumental: to reduce the multiplication of large numbers to an addition. Seventy-three years later, Newton, in the Principia of 1687, needs to reason about geometric magnitudes that approach one another, areas, arcs, tangents, and he does so with first and last ratios, a geometric intuition of the limit without the algebraic machinery to express it. And sixty-one years after that, Euler, in the Introductio in analysin infinitorum of 1748, the same book that contains, among many other things, the formula that will be taken up in §3, already operates with exe^x, with natural logarithms and with infinite series, in a treatise that systematizes the analysis of the eighteenth century with a boldness of calculation that its own century could not yet fully justify.

Between Napier and Euler one hundred and thirty-four years pass; between Euler and the moment when mathematics finally has the instrument that this chapter used from its first page, the completeness axiom of R\mathbb{R}, formulated precisely by Dedekind and by Cantor only in the 1870s, more than a further century passes. Cauchy, in the 1820s, is the first to demand definitions of limit and of continuity with the ε\varepsilon-δ\delta rigor that §2.4 of this book already named; Weierstrass, in the following decades, completes that demand and makes it standard. By the time Dedekind publishes his construction of the real numbers by means of cuts, in 1872, almost exactly contemporaneous with Cantor’s by means of sequences, the exponential already has two hundred and fifty years of fruitful use, without anyone having been able to prove, with the rigor demanded today, that axa^x for irrational xx was a well-defined and unique number.

This chapter compressed those two and a half centuries into twelve sections. The construction of §6 and §7 of section 4.1, supremum, infimum, Bernoulli, density, is, in essence, the answer that Dedekind and Cantor made possible; that this work occupies a few pages instead of two centuries does not speak of a lesser difficulty, but of the fact that the right question, what does Q\mathbb{Q} lack to be R\mathbb{R}?, already formulated in §1.1, was, this time, resolved in advance.


That the exponential was used with ease for two and a half centuries before being grounded is not an accident of this particular function. §3.6 of this same book, in closing the construction of C\mathbb{C}, found exactly the same pattern: Cardano and Bombelli manipulated square roots of negative numbers since the sixteenth century, with results that worked in calculations even when nobody could say, with precision, what those objects were, until Hamilton, in the nineteenth century, gave the axiomatic construction that §3.3 of this book repeated. The exponential and the complex numbers are, in this sense, the same story told twice, with different protagonists and about completely different mathematical objects.

It is worth asking why the pattern repeats with such regularity, instead of treating it as a coincidence between two chapters of the same book. One possible answer, not the only one, but the most economical, is that operational intuition and rigorous proof solve different problems, and not always in the same order. Operational intuition answers the question “does this object behave consistently in the calculations I already know how to do?”, and that question can often be answered by analogy, pattern, or simply trial and error, without needing to know what the object is in an ultimate sense. Rigorous proof answers a different and more demanding question: “why must this object exist, and why must it be unique?”, the question that this chapter spent twelve sections answering for axa^x. There is no logical necessity for both questions to be answered in the same century, and the history of mathematics, time and again, shows that they are not.

This does not reduce the value of operational intuition to a mere provisional shortcut. Euler did not calculate worse for lacking ε\varepsilon-δ\delta; his results, in their immense majority, proved correct when the following century subjected them to rigorous scrutiny. What the foundation contributes is not, then, mainly correctness, since the object already worked, but something different: the guarantee that this correctness was not chance, and solid ground on which to safely build the next generation of results, which intuition alone can no longer sustain without ambiguity.


§3. The exponential and the logarithm in contemporary science

Section titled “§3. The exponential and the logarithm in contemporary science”

§4.6 already showed three applications, compound interest, decay, measurement scales, in which the exponential and the logarithm solve a concrete and bounded problem. There are, beyond those three, two places in the science of the nineteenth and twentieth centuries where the logarithm is not a convenient calculation tool, but the way in which a fundamental physical magnitude is defined.

Boltzmann’s entropy. In 1877, Ludwig Boltzmann proposes a relation between the entropy SS of a thermodynamic system and the number WW of microstates, possible microscopic configurations, compatible with its observed macroscopic state:

S=kln⁡WS=k\ln W

where kk is the constant that today bears his name. The choice of the logarithm is not aesthetic: if two independent systems are combined, the number of microstates of the combined system is the product W1⋅W2W_1\cdot W_2, each configuration of one can be combined with each configuration of the other, while entropy, a physical magnitude expected to be additive when combining independent systems, must satisfy S=S1+S2S=S_1+S_2. The only way for a function of WW to translate a product into a sum is, exactly, the product law of §4.4 §1: ln⁡(W1W2)=ln⁡W1+ln⁡W2\ln(W_1W_2)=\ln W_1+\ln W_2. The logarithm was not chosen for convenience of calculation: it is the only function, up to a scale factor, compatible with the physical requirement that entropy be additive.

Shannon’s entropy. Seventy-one years later, in 1948, Claude Shannon publishes A Mathematical Theory of Communication and defines, for an information source with possible outcomes of probabilities p1,…,pnp_1,\ldots,p_n, the quantity

H=−∑ipilog⁡2piH=-\sum_i p_i\log_2 p_i

as a measure of the uncertainty, or, in the terminology that Shannon would end up popularizing, of the information, associated with that source. The base 22 is not arbitrary: it measures information in bits, the minimal unit of a choice between two equally probable alternatives, a direct consequence of log⁡22=1\log_2 2=1, according to §5 of section 4.3. And the same structural reason that led Boltzmann to the logarithm leads Shannon here: the probability of two independent events multiplies, while the information they contribute must add, receiving two independent messages should give as much information as the sum of what each gives separately, never more or less. That Shannon’s formula and Boltzmann’s share the same form, a logarithm of a probability, summed with a certain weight, is not a historical coincidence: it is the same requirement of additivity, resolved by the same tool, in two branches of science that in 1948 barely spoke to each other.

A bounded mention: Euler’s identity. The same book of Euler that this chapter cited in the close of §4.1, the Introductio of 1748, also contains the formula that relates the exponential to the trigonometric functions by means of an imaginary exponent: eiθ=cos⁡θ+isin⁡θe^{i\theta}=\cos\theta+i\sin\theta, from which, with θ=π\theta=\pi, the identity follows

eiπ+1=0e^{i\pi}+1=0

which is often cited as the most beautiful in mathematics, for gathering in a single equation five fundamental constants, ee, ii, π\pi, 11 and 00, and the three basic operations, addition, product, exponentiation. This book cannot prove it, and it is important to say precisely why not: eiθe^{i\theta} requires extending the exponential function to complex exponents, and this chapter, from beginning to end, constructed axa^x only for x∈Rx\in\mathbb{R}; the extension to x∈Cx\in\mathbb{C} is a different object, which would require taking up again the axiomatic construction of C\mathbb{C} of §3.3 and combining it with this one, something no chapter of this book has undertaken. It remains, then, as this chapter left the question of the derivative in §4.2: named with precision, situated in the exact place where this book stops, and without the concession of a proof that does not yet correspond to it.


§4. Close: the question that remains open

Section titled “§4. Close: the question that remains open”

§4.1 opened this chapter with a question that was never completely resolved: whether there exists a base for which the rate of change of the exponential function coincides, at each point, with the value of the function itself. §4.2 constructed the candidate, ee, determined without ambiguity by two squeezing sequences, and left the question formulated with precision, but unresolved. None of the following sections, logarithm, laws, equations, applications, and this very close, resolved it either, because none had the instrument to do so: the derivative.

It is not an omission for which this chapter must apologize. It is, rather, the most honest mark that a book can leave of its own reach: to construct everything that can be constructed with the available foundations, and to point out with precision the exact point where those foundations run out, instead of feigning a continuity that does not exist or appealing, even for a line, to a tool that the reader does not yet have. The question remains, then, formulated with the greatest precision that precalculus allows: ee exists, is unique, is bounded between 22 and 33, and satisfies ex+y=exeye^{x+y}=e^xe^y for every pair of reals, all of this proved; the only thing left to say is in what precise sense its rate of change is equal to itself, and that answer, with all reason, belongs to the first chapter of Calculus I.