Chapter IX · The logarithmic, exponential and circular functions of a real variable

The exponential function

§§212–213 35 min

eye^y defined for every real yy as the inverse of log⁡\log; it equals its own derivative, satisfies ey+z=eyeze^{y+z} = e^y e^z, and outgrows every power of yy.

Why this matters

We know eye^y only for rational yy. Rather than extend the definition of powers piece by piece, Hardy turns the logarithm round: eye^y is the number whose logarithm is yy. Every property of the exponential then follows from a property of the logarithm: the derivative, the addition rule, the speed of growth. This is the function at the heart of differential equations, probability and complex analysis, and §212 and §213 lay its foundation.

The idea

The logarithm is a one-to-one correspondence between the positive numbers xx and all real numbers yy: it increases strictly, and it runs from −∞-\infty (as x→0x \to 0) to +∞+\infty (as x→∞x \to \infty) without jumps. So every real yy is the logarithm of exactly one positive xx. Call that xx "eye^y". Its graph is the graph of log⁡\log reflected in the diagonal y=xy = x.

Explore The exponential as a mirror image

Try this. Drag PP along the logarithm. Watch its mirror image QQ on the exponential curve and compare the slopes of the two tangents. Then put PP at x=1x = 1 and at x=e≈2.72x = e \approx 2.72.

  • y = log x
  • x = e^y, drawn as the mirror image
  • mirror line
  • P (drag) and the two tangents
P=(2, 0.693)P = (2,\ 0.693) on the logarithm, slope 1/x=0.51/x = 0.5. Its mirror image Q=(0.693, 2)Q = (0.693,\ 2) on the exponential has slope 1/(1/x)=x=21/(1/x) = x = 2, which is exactly the height of QQ: ddyey=ey\tfrac{d}{dy}e^{y} = e^{y} at y=0.693y = 0.693.
What to notice

Reflection swaps horizontal and vertical, so the slope at QQ is the reciprocal of the slope at PP: 1/(1/x)=x1/(1/x) = x. But xx is precisely the height of QQ. So at every point the exponential's slope equals its height, ddyey=ey\tfrac{d}{dy}e^y = e^y. At x=1x = 1, Q=(0,1)Q = (0, 1) with slope 11; at x=ex = e, Q=(1,e)Q = (1, e).

The speed of the exponential is the logarithm’s slowness seen in the mirror. The previous lessons showed that log⁡x\log x loses to every power of xx; substituting x=eyx = e^y turns ”(log⁡x)α(\log x)^\alpha is negligible compared with xx” into ”yαy^\alpha is negligible compared with eye^y”: the exponential beats every power of yy.

Explore Race a large power against exe^x

Try this. Set α=20\alpha = 20 and find where exe^x finally overtakes x20x^{20}. Then try α=100\alpha = 100 and α=200\alpha = 200 (widen the view).

x20againstexx^{20} \quad\text{against}\quad e^x
xlog₁₀ (value)
  • x20x^{20}
  • exe^x
  • last crossing
  • read-out point
At x=40x = 40: x20/ex≈1014.67x^{20} \big/ e^x \approx 10^{14.67} (the slower function is still ahead). The last crossing is at x≈90x \approx 90; beyond it the ratio only falls, towards 0. The ratio (slower ÷ faster) is largest at x=α=20x = \alpha = 20.
What to notice

On these semi-logarithmic axes exe^x is a straight line and xαx^\alpha a flattening curve: the same picture as the race between log⁡x\log x and xαx^\alpha, one level up. x100x^{100} leads until x≈647x \approx 647, but the line always wins. The ratio xα/exx^\alpha/e^x is largest at x=αx = \alpha.

Precise statements

Definition The exponential function Hardy: §212

For every real yy, eye^y (also written exp⁡y\exp y) is the unique positive number xx such that log⁡x=y\log x = y. Thus

x=ey  ⟺  y=log⁡x(x>0, y real).x = e^{y} \iff y = \log x \qquad (x > 0,\ y \text{ real}).
Theorem Basic properties Hardy: §212

eye^y is positive, continuous and strictly increasing on the whole real line; ey→0e^y \to 0 as y→−∞y \to -\infty and ey→∞e^y \to \infty as y→∞y \to \infty; e0=1e^0 = 1; and for rational yy, eye^y is the positive yy-th power of the number ee.

Theorem The exponential is its own derivative Hardy: §213 (1)
ddyey=ey,and more generallyddyeay=a eay.\frac{d}{dy}e^{y} = e^{y}, \qquad\text{and more generally}\qquad \frac{d}{dy}e^{ay} = a\,e^{ay}.
Theorem The addition formula Hardy: §213 (2)

For all real yy and zz, ey+z=eyeze^{y+z} = e^{y}e^{z}. In particular e−y=1/eye^{-y} = 1/e^{y}.

Theorem The exponential beats every power Hardy: §213 (3)

For every α\alpha, however large, y−αey→∞y^{-\alpha}e^{y} \to \infty as y→∞y \to \infty. More generally eγye^{\gamma y} tends to ∞\infty if γ>0\gamma > 0 and to 00 if γ<0\gamma < 0, in each case faster than any power of yy.

This adds a new end to the scale of infinity: ⋯≺xn≺⋯≺ex≺e2x≺⋯≺ex2≺⋯≺eex≺…\dots \prec x^{n} \prec \dots \prec e^{x} \prec e^{2x} \prec \dots \prec e^{x^2} \prec \dots \prec e^{e^x} \prec \dots, where ex2e^{x^2} means e(x2)e^{(x^2)} and eexe^{e^x} means e(ex)e^{(e^x)}.

Proofs

Proof The exponential is well defined, continuous and increasing
Strategy

Show that log⁡\log maps (0,∞)(0, \infty) one-to-one onto the real line; then the inverse exists, is increasing, and inherits continuity. Hardy gives a direct proof of the continuity, which we follow.

Onto

Given yy, the limits log⁡x→−∞\log x \to -\infty (x→0x \to 0) and log⁡x→∞\log x \to \infty (x→∞x \to \infty) give a<ba < b with log⁡a<y<log⁡b\log a < y < \log b. By the intermediate value theorem log⁡x=y\log x = y for some xx in (a,b)(a, b).

One-to-one, and the inverse increases

log⁡\log is strictly increasing, so the xx is unique; and if y1<y2y_1 < y_2 then ey1<ey2e^{y_1} < e^{y_2}, since otherwise log⁡\log would give y1≥y2y_1 \ge y_2.

Continuity: write x=eyx = e^y, x+ξ=ey+ηx + \xi = e^{y + \eta}

Then η=log⁡(x+ξ)−log⁡x=log⁡(1+ξ/x)\eta = \log(x + \xi) - \log x = \log(1 + \xi/x). We must show that ξ→0\xi \to 0 when η→0\eta \to 0.

Bound ξ\xi by η\eta

Use h1+h<log⁡(1+h)<h\dfrac{h}{1+h} < \log(1 + h) < h with h=ξ/xh = \xi/x. If ξ>0\xi > 0: η>ξx+ξ\eta > \dfrac{\xi}{x + \xi}, which rearranges to ξ<xη1−η\xi < \dfrac{x\eta}{1 - \eta} when η<1\eta < 1. If ξ<0\xi < 0: η<ξ/x\eta < \xi/x, so ∣ξ∣<x∣η∣|\xi| < x|\eta|. Either way ∣ξ∣≤x∣η∣1−∣η∣|\xi| \le \dfrac{x|\eta|}{1 - |\eta|} for ∣η∣<1|\eta| < 1, which tends to 00 with η\eta.

Values

e0=1e^0 = 1 because log⁡1=0\log 1 = 0. For rational yy, the previous lesson showed log⁡(positive y-th power of e)=y\log(\text{positive } y\text{-th power of } e) = y, so by uniqueness that power is eye^y. Since eye^y is increasing and takes every positive value, it is unbounded above, so ey→∞e^y \to \infty as y→∞y \to \infty; and it takes values arbitrarily close to 00, so ey→0e^y \to 0 as y→−∞y \to -\infty.

Proof ddyey=ey\tfrac{d}{dy}e^y = e^y
Strategy

Differentiate y=log⁡xy = \log x with respect to xx and invert: the derivative of an inverse function is the reciprocal of the derivative.

Difference quotients are reciprocals

With x=eyx = e^y and x+ξ=ey+ηx + \xi = e^{y+\eta} as above (η≠0\eta \ne 0, so ξ≠0\xi \ne 0): ξη=1/ηξ=1/log⁡(x+ξ)−log⁡xξ\dfrac{\xi}{\eta} = 1 \Big/ \dfrac{\eta}{\xi} = 1 \Big/ \dfrac{\log(x + \xi) - \log x}{\xi}.

Let η→0\eta \to 0

By continuity, ξ→0\xi \to 0, so the denominator tends to (log⁡)′(x)=1/x(\log)'(x) = 1/x. Hence ξ/η→x=ey\xi/\eta \to x = e^y.

eaye^{ay}

By the chain rule, ddyeay=eay⋅a\frac{d}{dy}e^{ay} = e^{ay}\cdot a.

Proof ey+z=eyeze^{y+z} = e^y e^z
Strategy

Translate into logarithms, where the statement becomes log⁡xy=log⁡x+log⁡y\log xy = \log x + \log y.

Take logarithms of the right-hand side

log⁡(eyez)=log⁡ey+log⁡ez=y+z\log(e^y e^z) = \log e^y + \log e^z = y + z, by the functional equation of the logarithm.

Invert

So eyeze^y e^z is the positive number whose logarithm is y+zy + z, which by definition is ey+ze^{y+z}. With z=−yz = -y: eye−y=e0=1e^y e^{-y} = e^0 = 1.

Hardy also gives a second proof: the rule is true for rational yy and zz by the laws of indices, and both sides are continuous, so it holds in the limit along rational sequences yn→yy_n \to y, zn→zz_n \to z. This “prove it for rationals, extend by continuity” argument is used again for general powers in the next lesson.

Proof ey/yα→∞e^y/y^\alpha \to \infty
Strategy

Substitute x=eyx = e^y in the theorem (log⁡x)α/x→0(\log x)^\alpha / x \to 0 of the third lesson.

The known result

For every α>0\alpha > 0, (log⁡x)α/x→0(\log x)^\alpha/x \to 0 as x→∞x \to \infty (the corollary with k=αk = \alpha and power 11).

Put x=eyx = e^y

As y→∞y \to \infty, x=ey→∞x = e^y \to \infty, and (log⁡x)α/x=yα/ey(\log x)^\alpha/x = y^\alpha/e^y. So yα/ey→0y^\alpha/e^y \to 0, that is, ey/yα→∞e^y/y^\alpha \to \infty (the ratio is positive). For α≤0\alpha \le 0 the claim is trivial.

eγye^{\gamma y}

If γ>0\gamma > 0, eγy/yα=γα eu/uαe^{\gamma y}/y^\alpha = \gamma^\alpha\, e^{u}/u^{\alpha} with u=γy→∞u = \gamma y \to \infty. If γ<0\gamma < 0, yαeγy=1/(e∣γ∣y/yα)→0y^\alpha e^{\gamma y} = 1\big/\big(e^{|\gamma|y}/y^{\alpha}\big) \to 0.

Worked examples and exercises

Exercise ●○○ Hardy: Examples LXXXVI, 1

Show that if dxdy=ax\dfrac{dx}{dy} = ax for all yy, where aa is a constant, then x=Keayx = Ke^{ay} for a constant KK.

Hint
Differentiate x e−ayx\,e^{-ay}.

Solution

ddy(xe−ay)=dxdye−ay−axe−ay=(ax−ax)e−ay=0\dfrac{d}{dy}\left(x e^{-ay}\right) = \dfrac{dx}{dy}e^{-ay} - a x e^{-ay} = (ax - ax)e^{-ay} = 0. A function with zero derivative everywhere is constant (mean value theorem), so xe−ay=Kx e^{-ay} = K and x=Keayx = Ke^{ay}.

Exercise ●●○ Hardy: Examples LXXXVI, 2

Show that a differentiable solution of f(y+z)=f(y)f(z)f(y + z) = f(y)f(z) is either identically 00 or of the form eaye^{ay}.

Hint
Differentiate in yy and in zz: f′(y+z)=f′(y)f(z)=f(y)f′(z)f'(y+z) = f'(y)f(z) = f(y)f'(z).
Hint
First deal with f(0)f(0): put y=z=0y = z = 0.

Solution

Putting y=z=0y = z = 0 gives f(0)=f(0)2f(0) = f(0)^2, so f(0)=0f(0) = 0 or 11. If f(0)=0f(0) = 0 then f(y)=f(y+0)=f(y)f(0)=0f(y) = f(y + 0) = f(y)f(0) = 0 for all yy.

Suppose f(0)=1f(0) = 1. Differentiating in zz and then setting z=0z = 0: f′(y)=f(y)f′(0)f'(y) = f(y)f'(0). Put a=f′(0)a = f'(0). By the previous exercise f(y)=Keayf(y) = Ke^{ay}, and f(0)=1f(0) = 1 gives K=1K = 1. (Hardy writes KeayKe^{ay}; substituting into the equation shows K=K2K = K^2, so the only possibilities are again K=0K = 0 and K=1K = 1.)

Exercise ●○○ Hardy: Examples LXXXVI, 3

Show that eay−1y→a\dfrac{e^{ay} - 1}{y} \to a as y→0y \to 0.

Hint
Either recognise a derivative at 00, or use the mean value theorem as Hardy suggests.

Solution

By the mean value theorem, eay−1=eay−e0=y⋅aeaηe^{ay} - 1 = e^{ay} - e^{0} = y \cdot a e^{a\eta} for some η\eta between 00 and yy. So eay−1y=aeaη\dfrac{e^{ay} - 1}{y} = a e^{a\eta}. As y→0y \to 0, η→0\eta \to 0 and, by continuity, eaη→1e^{a\eta} \to 1. The limit is aa. (Equivalently: it is the derivative of eaye^{ay} at y=0y = 0.)

Exercise ●●○ Hardy: Examples LXXXVI, 4

Show that ex−1−xe^x - 1 - x,  e−x−1+x\ e^{-x} - 1 + x and  1−12x2+13x3−(1+x)e−x\ 1 - \tfrac12 x^2 + \tfrac13 x^3 - (1 + x)e^{-x} are positive and strictly increasing for x>0x > 0.

Hint
Each vanishes at x=0x = 0. Show its derivative is positive for x>0x > 0.
Hint
For the third, the derivative factorises as x(x−1+e−x)x(x - 1 + e^{-x}).

Solution

First. g(x)=ex−1−xg(x) = e^x - 1 - x: g(0)=0g(0) = 0 and g′(x)=ex−1>0g'(x) = e^x - 1 > 0 for x>0x > 0, since exe^x increases and e0=1e^0 = 1. So gg increases strictly on [0,∞)[0, \infty) and is positive for x>0x > 0.

Second. h(x)=e−x−1+xh(x) = e^{-x} - 1 + x: h(0)=0h(0) = 0, h′(x)=1−e−x>0h'(x) = 1 - e^{-x} > 0 for x>0x > 0.

Third. k(x)=1−12x2+13x3−(1+x)e−xk(x) = 1 - \tfrac12x^2 + \tfrac13x^3 - (1+x)e^{-x}: k(0)=1−1=0k(0) = 1 - 1 = 0, and

k′(x)=−x+x2−e−x+(1+x)e−x=−x+x2+xe−x=x (x−1+e−x)=x h(x),k'(x) = -x + x^2 - e^{-x} + (1 + x)e^{-x} = -x + x^2 + xe^{-x} = x\,(x - 1 + e^{-x}) = x\,h(x),

which is positive for x>0x > 0 by the second part.

Exercise Polynomials against exponentials ●○○ Hardy: §213 (3)

Show that P(y)e−y→0P(y)e^{-y} \to 0 as y→∞y \to \infty for every polynomial PP, and that ∫0∞yne−y dy\displaystyle\int_0^\infty y^{n}e^{-y}\,dy converges for every n≥0n \ge 0.

Hint
For the integral, compare yne−yy^n e^{-y} with y−2y^{-2} for large yy.

Solution

Each term ckyke−y→0c_k y^k e^{-y} \to 0 by the theorem, so the finite sum does too.

For the integral: yn+2e−y→0y^{n+2}e^{-y} \to 0, so yn+2e−y<1y^{n+2}e^{-y} < 1 for y≥Yy \ge Y, i.e. yne−y<y−2y^n e^{-y} < y^{-2} there. Since ∫Y∞y−2 dy\int_Y^\infty y^{-2}\,dy converges and the integrand is positive and continuous on [0,Y][0, Y], the integral converges by comparison (§184). (Its value is n!n!, as integration by parts shows.)

Check yourself

Check yourself Choose one

What is the slope of the graph of eye^y at the point where y=1y = 1?

Why

The slope of eye^y equals its value, e1=ee^1 = e. The reciprocal 1/e1/e is the slope of the logarithm at the mirror point (e,1)(e, 1).

Check yourself True or false?

ey>0e^y > 0 for every real yy, including large negative yy.

Why

True. By definition eye^y is a number xx in the domain of log⁡\log, which consists of positive numbers. Large negative yy correspond to xx close to 00, never 00 or negative.

Check yourself Enter a number

Beyond which xx (approximately) is ex>x100e^x > x^{100} for good? Use the race above, or solve x=100log⁡xx = 100\log x.

Why

Taking logs, ex>x100e^x > x^{100} means x>100log⁡xx > 100\log x. The function x−100log⁡xx - 100\log x is smallest at x=100x = 100 and has its last zero near x=647.3x = 647.3. The same number appeared in the third lesson, in the race between log⁡u\log u and u0.01u^{0.01}: putting x=log⁡ux = \log u turns one race into the other.

Check yourself Choose one

Which of these tends to infinity fastest as x→∞x \to \infty?

Why

Compare logarithms: 1000log⁡x1000\log x, xx, x1.01x^{1.01} and xlog⁡1000≈6.9xx\log 1000 \approx 6.9x. The last two are the largest; and x1.01/(6.9x)=x0.01/6.9→∞x^{1.01}/(6.9x) = x^{0.01}/6.9 \to \infty, so ex1.01e^{x^{1.01}} wins (eventually: only beyond x≈6.9100x \approx 6.9^{100}). Comparing logarithms is the standard way to place functions on the exponential end of the scale.

Hardy’s language

Summary

Summary
  • eye^y is the positive number whose logarithm is yy; its graph is the mirror image of the logarithm’s in y=xy = x.
  • It is positive, continuous and strictly increasing from 00 to ∞\infty, and agrees with ep/qe^{p/q} for rational exponents.
  • ddyey=ey\frac{d}{dy}e^y = e^y (reciprocal slopes under reflection) and ey+z=eyeze^{y+z} = e^y e^z (from log⁡xy=log⁡x+log⁡y\log xy = \log x + \log y).
  • eye^y beats every power of yy: substitute x=eyx = e^y in ”log⁡x\log x loses to every power of xx”. This gives the exponential end of the scale of infinity.