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

The general power axa^x

§214 30 min

How axa^x is defined for irrational xx as exlog⁡ae^{x\log a}, why that is the only sensible choice, and the laws and derivatives that follow.

Why this matters

What is 222^{\sqrt 2}? School algebra defines ap/qa^{p/q} as a root of a power and stops there, yet calculus freely uses xπx^\pi, 10x10^x for all real xx, and the rule ddxxm=mxm−1\frac{d}{dx}x^m = mx^{m-1} for irrational mm. In §214 Hardy fills the gap with one line, ax=exlog⁡aa^x = e^{x\log a}, and shows that all the familiar rules survive. This also removes the restriction ”xx rational” from many results of earlier chapters.

The idea

For a rational exponent we already know two things: ap/qa^{p/q} is the positive qq-th root of apa^p, and its logarithm is pqlog⁡a\frac{p}{q}\log a. So for rational rr,

ar=e rlog⁡a.a^{r} = e^{\,r\log a}.

The right-hand side makes sense for every real rr, because eye^y is now defined for every real yy. So we take it as the definition when rr is irrational. Is that the “right” value? The test is continuity: if r1,r2,…r_1, r_2, \dots are rationals tending to xx, the elementary powers arna^{r_n} should tend to the new axa^x.

Explore Close in on 222^{\sqrt2} with rational exponents

Try this. With a=2a = 2 and exponent 2\sqrt2, increase the number of decimal places from 00 to 88. Each row uses only roots and powers: 21.41=21411002^{1.41} = \sqrt[100]{2^{141}}. Then try a<1a < 1 and the exponent π\pi.

xaˣ
  • ax=exlog⁡aa^x = e^{x\log a}
  • rational powers ap/qa^{p/q}
  • a2a^{\sqrt2}
rational exponent p/qap/q=apqa^{p/q} = \sqrt[q]{a^p}distance from exlog⁡ae^{x\log a}
12.0000000000.665
1.42.6390158220.0261
1.412.6573716280.00777
1.4142.6647496500.000394
Definition: a2=e2 log⁡a=e0.980258=2.665144143a^{\sqrt2} = e^{\sqrt2\,\log a} = e^{0.980258} = 2.665144143. Each extra decimal place in the exponent brings ap/qa^{p/q} roughly ten times closer, because exlog⁡ae^{x\log a} is continuous (indeed differentiable) in xx.
What to notice

The elementary values 1.41,1.414,1.4142,…1.41, 1.414, 1.4142, \dots give 21.41≈2.65742^{1.41} \approx 2.6574, 21.414≈2.66472^{1.414} \approx 2.6647, … and settle on 2.665144…=e2log⁡22.665144\ldots = e^{\sqrt2\log 2}, each extra digit gaining about one more correct digit. Every rational point lies on the one smooth curve exlog⁡ae^{x\log a}: no other continuous way of filling in the irrational exponents exists. For a<1a < 1 the curve falls instead of rising.

Precise statements

Definition The general power Hardy: §214

For a>0a > 0 and every real xx,

ax=e xlog⁡a.a^{x} = e^{\,x\log a}.

For rational xx this is the positive value of the elementary power. Consequently ax>0a^x > 0 and log⁡ax=xlog⁡a\log a^x = x\log a for all real xx. The power axa^x with irrational xx is defined only for positive aa.

Theorem Properties of axa^x Hardy: §214 (1)–(5)

Let a>0a > 0 and b>0b > 0, and let xx, yy be real.

  1. Laws of indices: axay=ax+ya^{x}a^{y} = a^{x+y} and (ax)y=axy(a^{x})^{y} = a^{xy}; also (ab)x=axbx(ab)^x = a^x b^x.
  2. Growth: if a>1a > 1, ax→∞a^x \to \infty as x→∞x \to \infty, faster than any power of xx; if 0<a<10 < a < 1, ax→0a^x \to 0 as x→∞x \to \infty, faster than any power of 1/x1/x.
  3. Derivative in xx: ddxax=axlog⁡a\dfrac{d}{dx}a^{x} = a^{x}\log a.
  4. Derivative in aa: ddaax=x ax−1\dfrac{d}{da}a^{x} = x\,a^{x-1}. In other words ddttm=m tm−1\dfrac{d}{dt}t^m = m\,t^{m-1} for t>0t > 0 and every real mm.
  5. A limit: ax−1x→log⁡a\dfrac{a^{x} - 1}{x} \to \log a as x→0x \to 0.

Proofs

Proof The laws of indices
Strategy

Everything reduces, through the definition, to eu+v=eueve^{u+v} = e^u e^v and log⁡eu=u\log e^u = u.

axay=ax+ya^x a^y = a^{x+y}

axay=exlog⁡aeylog⁡a=e(x+y)log⁡a=ax+ya^x a^y = e^{x\log a}e^{y\log a} = e^{(x+y)\log a} = a^{x+y}.

(ax)y=axy(a^x)^y = a^{xy}

By definition (ax)y=eylog⁡(ax)(a^x)^y = e^{y\log(a^x)}, and log⁡ax=log⁡exlog⁡a=xlog⁡a\log a^x = \log e^{x\log a} = x\log a. So (ax)y=exylog⁡a=axy(a^x)^y = e^{xy\log a} = a^{xy}.

(ab)x=axbx(ab)^x = a^x b^x

(ab)x=exlog⁡ab=exlog⁡a+xlog⁡b=axbx(ab)^x = e^{x\log ab} = e^{x\log a + x\log b} = a^x b^x.

Proof Growth
Strategy

Write axa^x as an exponential eγxe^{\gamma x} with γ=log⁡a\gamma = \log a and use the previous lesson.

a>1a > 1

Then γ=log⁡a>0\gamma = \log a > 0 and ax=eγxa^x = e^{\gamma x}, which tends to ∞\infty faster than any power of xx.

0<a<10 < a < 1

Then γ=log⁡a<0\gamma = \log a < 0, and eγx⋅xk→0e^{\gamma x} \cdot x^{k} \to 0 for every kk: axa^x tends to 00 faster than every power x−kx^{-k}. The graph is that of (1/a)x(1/a)^x reflected left to right, since ax=(1/a)−xa^x = (1/a)^{-x}.

Proof The derivatives and the limit
Strategy

Differentiate exlog⁡ae^{x\log a} by the chain rule, once treating xx as the variable and once aa.

In xx

ddxexlog⁡a=exlog⁡a⋅log⁡a=axlog⁡a\dfrac{d}{dx}e^{x\log a} = e^{x\log a}\cdot\log a = a^x\log a.

In aa

ddaexlog⁡a=exlog⁡a⋅xa=x axa−1=x ax−1\dfrac{d}{da}e^{x\log a} = e^{x\log a}\cdot\dfrac{x}{a} = x\,a^{x}a^{-1} = x\,a^{x-1}, using the law of indices.

The limit

ax−1x=ax−a0x−0\dfrac{a^x - 1}{x} = \dfrac{a^x - a^0}{x - 0} is a difference quotient of axa^x at 00, so it tends to the derivative there, a0log⁡a=log⁡aa^0\log a = \log a.

Part 4 extends the power rule, which Chapter VI proved for rational exponents, to every real exponent. Part 5 is Examples LXXXVI, 3 in a new notation (there (eαy−1)/y→α(e^{\alpha y} - 1)/y \to \alpha; take α=log⁡a\alpha = \log a, so that eαy=aye^{\alpha y} = a^y), as Hardy remarks, and its sequence form n(a1/n−1)→log⁡an(a^{1/n} - 1) \to \log a is the subject of §216.

Why no other definition would do. Suppose F(x)F(x) is any function that agrees with the elementary axa^x at rational xx and is continuous. Given irrational xx, choose rationals rn→xr_n \to x. Then F(x)=lim⁡F(rn)=lim⁡ernlog⁡a=exlog⁡aF(x) = \lim F(r_n) = \lim e^{r_n\log a} = e^{x\log a}, because eye^{y} is continuous. So continuity alone forces ax=exlog⁡aa^x = e^{x\log a}, which is what the widget shows numerically.

Worked examples and exercises

Exercise ●●○ Hardy: Examples LXXXVIII, 31

Show that x1/x→1x^{1/x} \to 1 as x→∞x \to \infty, that x1/xx^{1/x} has its greatest value at x=ex = e, and sketch its graph for x>0x > 0.

Hint
x1/x=e(log⁡x)/xx^{1/x} = e^{(\log x)/x}. What does (log⁡x)/x(\log x)/x do?
Hint
Differentiate g(x)=(log⁡x)/xg(x) = (\log x)/x: g′(x)=(1−log⁡x)/x2g'(x) = (1 - \log x)/x^2.

Solution

x1/x=eg(x)x^{1/x} = e^{g(x)} with g(x)=(log⁡x)/xg(x) = (\log x)/x. By §208, g(x)→0g(x) \to 0 as x→∞x \to \infty; since eye^y is continuous, x1/x→e0=1x^{1/x} \to e^0 = 1.

g′(x)=(1−log⁡x)/x2g'(x) = (1 - \log x)/x^2 is positive for x<ex < e and negative for x>ex > e, so gg, and with it ege^{g}, is greatest at x=ex = e: the maximum value is e1/e≈1.4447e^{1/e} \approx 1.4447.

As x→+0x \to +0, g(x)=(log⁡x)/x→−∞g(x) = (\log x)/x \to -\infty (a large negative number divided by a small positive one), so x1/x→0x^{1/x} \to 0. The graph rises from 00, crosses 11 at x=1x = 1, peaks at (e,e1/e)(e, e^{1/e}), and falls slowly back towards 11.

Exercise eπe^\pi or πe\pi^e? ●●○ Hardy: Examples LXXXVIII, 31

Without a calculator, decide which is larger, eπe^{\pi} or πe\pi^{e}.

Hint
Raise both to the power 1/(eπ)1/(e\pi) and use the previous exercise.

Solution

eπ>πee^\pi > \pi^e if and only if e1/e>π1/πe^{1/e} > \pi^{1/\pi} (raise both sides to the positive power 1/(eπ)1/(e\pi); t↦t1/(eπ)t \mapsto t^{1/(e\pi)} is increasing). The function x1/xx^{1/x} has its strict maximum at x=ex = e, and π≠e\pi \ne e, so π1/π<e1/e\pi^{1/\pi} < e^{1/e}. Hence eπ>πee^\pi > \pi^e. (Numerically 23.14>22.4623.14 > 22.46.)

Exercise ●○○ Hardy: Examples LXXXVIII, 32

Show that xx→1x^{x} \to 1 as x→+0x \to +0. Where does xxx^x (for x>0x > 0) take its least value?

Hint
xx=exlog⁡xx^x = e^{x\log x}, and xlog⁡x→0x\log x \to 0 as x→+0x \to +0 (§209).

Solution

xx=exlog⁡xx^x = e^{x\log x} and xlog⁡x→0x\log x \to 0, so xx→e0=1x^x \to e^0 = 1 by continuity of the exponential.

ddxxx=ddxexlog⁡x=xx(log⁡x+1)\dfrac{d}{dx}x^x = \dfrac{d}{dx}e^{x\log x} = x^x(\log x + 1), which is negative for x<1/ex < 1/e and positive for x>1/ex > 1/e. So the least value is at x=1/ex = 1/e, where xx=e−1/e≈0.6922x^x = e^{-1/e} \approx 0.6922. Note that neither rule "ddxxm=mxm−1\frac{d}{dx}x^m = mx^{m-1}" (exponent fixed) nor "ddxax=axlog⁡a\frac{d}{dx}a^x = a^x\log a" (base fixed) applies alone; the derivative is the sum of the two.

Exercise The power rule for every real exponent ●○○ Hardy: §214 (4)

Show that ddxx2=2 x2−1\dfrac{d}{dx}x^{\sqrt2} = \sqrt2\, x^{\sqrt2 - 1} for x>0x > 0, and find ∫12xπ dx\displaystyle\int_1^2 x^{\pi}\,dx.

Hint
Part 4 of the theorem, with aa renamed xx. Then reverse it.

Solution

By part 4 with m=2m = \sqrt2: ddxx2=2 x2−1\frac{d}{dx}x^{\sqrt 2} = \sqrt2\,x^{\sqrt2-1}. In the same way ddxxπ+1π+1=xπ\frac{d}{dx}\frac{x^{\pi+1}}{\pi+1} = x^\pi, so ∫12xπdx=2π+1−1π+1≈16.654.142≈4.02\int_1^2 x^\pi dx = \dfrac{2^{\pi+1} - 1}{\pi + 1} \approx \dfrac{16.65}{4.142} \approx 4.02.

Check yourself

Check yourself Enter a number

Give 222^{\sqrt2} to four decimal places.

Why

22=e2log⁡2=e1.41421×0.69315=e0.98026=2.66514…2^{\sqrt2} = e^{\sqrt2\log2} = e^{1.41421 \times 0.69315} = e^{0.98026} = 2.66514\ldots. The widget’s rational approximations 21.41422^{1.4142} and 21.414212^{1.41421} agree with this to four places.

Check yourself True or false?

(−8)2(-8)^{\sqrt2} is defined by Hardy’s definition, since (−8)1/3=−2(-8)^{1/3} = -2 makes sense.

Why

False. The definition ax=exlog⁡aa^x = e^{x\log a} needs log⁡a\log a, so a>0a > 0. Some rational powers of negative numbers make sense ((−8)1/3=−2(-8)^{1/3} = -2), but not all ((−8)1/2(-8)^{1/2}), and there is no continuous way to fill in the irrational exponents. Complex powers of negative numbers are the subject of Chapter X.

Check yourself Choose one

What is ddx2x\dfrac{d}{dx}2^{x}?

Why

2x=exlog⁡22^x = e^{x\log 2}, so the derivative is exlog⁡2⋅log⁡2e^{x\log2}\cdot\log 2. The first option is the derivative with respect to the base (part 4), the wrong variable here. Only base ee gives a function equal to its own derivative, since log⁡e=1\log e = 1.

Check yourself Choose one

For 0<a<10 < a < 1, as x→∞x \to \infty, axa^x

Why

ax=e−βxa^x = e^{-\beta x} with β=−log⁡a>0\beta = -\log a > 0, and xke−βx→0x^k e^{-\beta x} \to 0 for every kk. The exponent tends to −∞-\infty, but the power itself is always positive.

Hardy’s language

Summary

Summary
  • For a>0a > 0 and any real xx: ax=exlog⁡aa^x = e^{x\log a}. It agrees with roots of powers for rational xx, and it is the only continuous function that does.
  • log⁡ax=xlog⁡a\log a^x = x\log a; the laws of indices axay=ax+ya^xa^y = a^{x+y}, (ax)y=axy(a^x)^y = a^{xy} hold for all real exponents.
  • ddxax=axlog⁡a\frac{d}{dx}a^x = a^x\log a, and ddttm=mtm−1\frac{d}{dt}t^m = mt^{m-1} for every real mm.
  • ax→∞a^x \to \infty (if a>1a > 1) or →0\to 0 (if a<1a < 1) faster than any power; (ax−1)/x→log⁡a(a^x - 1)/x \to \log a as x→0x \to 0.