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

How slowly log⁡x\log x tends to infinity

§§207–209 30 min

log⁡x→∞\log x \to \infty, but more slowly than every positive power of xx; and log⁡x→−∞\log x \to -\infty as x→0x \to 0, more slowly than every power of 1/x1/x.

Why this matters

In §98 Hardy measured how fast a function tends to infinity by comparing it with powers: ff is “of order α\alpha” if f(x)/xαf(x)/x^\alpha tends to a non-zero limit. The powers xαx^\alpha, with α\alpha as small or as large as we please, look like a complete measuring stick. The logarithm shows that they are not: log⁡x\log x tends to infinity, yet it is beaten by x0.0000001x^{0.0000001}. This single fact, proved in §208 in three lines, is used in nearly every estimate in analysis.

The idea

Compare log⁡x\log x with a small power such as x0.1x^{0.1}. For moderate xx the logarithm is far ahead: at x=1010x = 10^{10}, log⁡x≈23\log x \approx 23 while x0.1=10x^{0.1} = 10. It looks as if the logarithm wins. But plot both on log–log axes (horizontally log⁡10x\log_{10} x, vertically log⁡10\log_{10} of the value) and the picture changes. A power xαx^\alpha becomes a straight line of slope α\alpha, because log⁡10xα=αlog⁡10x\log_{10} x^\alpha = \alpha \log_{10} x. The logarithm becomes log⁡10(log⁡x)\log_{10}(\log x), a curve that keeps bending over. A straight line rising at any positive slope must overtake a curve that flattens out.

Explore Race a power of log⁡x\log x against a small power of xx

Try this. Start with k=3k = 3 and α=0.25\alpha = 0.25 and find the last crossing. Then make α\alpha smaller (try 0.050.05) or kk larger, and widen the view until the crossing reappears.

(log⁡x)3againstx0.25(\log x)^{3} \quad\text{against}\quad x^{0.25}
log₁₀ xlog₁₀ (value)
  • (log⁡x)3(\log x)^{3}
  • x0.25x^{0.25}
  • last crossing
  • read-out point
At x=1010x = 10^{10}: (log⁡x)3/x0.25≈101.587(\log x)^{3} \big/ x^{0.25} \approx 10^{1.587} (the slower function is still ahead). The last crossing is at x≈1019.94x \approx 10^{19.94}; beyond it the ratio only falls, towards 0. The ratio (slower ÷ faster) is largest at x=105.21x = 10^{5.21}, i.e. log⁡x=k/α\log x = k/\alpha.
What to notice

However you handicap the power, it wins in the end: the crossing moves out to astronomical xx (for log⁡x\log x against x0.1x^{0.1} it is at about x=3.4×1015x = 3.4 \times 10^{15}), but it is always there, and after it the ratio falls without limit. The ratio peaks where log⁡x=k/α\log x = k/\alpha, then declines for ever.

So the picture suggests two things: the race is always won by the power, and the win may come very late. The theorem below captures the first; the second is a warning about trusting tables of values.

Precise statements

Theorem The logarithm is slower than every power Hardy: §207

log⁡x→∞\log x \to \infty as x→∞x \to \infty, but for every positive α\alpha

log⁡xxα→0(x→∞).\frac{\log x}{x^{\alpha}} \to 0 \qquad (x \to \infty).

Hardy states this for rational α\alpha, since xαx^\alpha has not yet been defined for irrational α\alpha; once it has (§214), the same proof works for every real α>0\alpha > 0.

Corollary Any power of the logarithm

For every k>0k > 0 and α>0\alpha > 0, (log⁡x)kxα→0\dfrac{(\log x)^k}{x^\alpha} \to 0 as x→∞x \to \infty.

Theorem Behaviour as x→+0x \to +0 Hardy: §209

For every positive α\alpha, xαlog⁡x→0x^\alpha \log x \to 0 as x→+0x \to +0. Thus log⁡(1/x)→∞\log(1/x) \to \infty as x→+0x \to +0, but more slowly than any positive power of 1/x1/x.

Proofs

The whole difficulty is that log⁡\log is defined by an integral we cannot evaluate. Hardy’s trick is to compare the integrand 1/t1/t with an integrand we can integrate, tβ−1t^{\beta - 1}, which is only a little larger.

Proof x−αlog⁡x→0x^{-\alpha}\log x \to 0
Strategy

Bound log⁡x\log x above by a slightly smaller power xβ/βx^\beta/\beta with 0<β<α0 < \beta < \alpha. Then x−αlog⁡xx^{-\alpha}\log x is at most a constant times xβ−αx^{\beta - \alpha}, and a negative power tends to 00.

For t>1t > 1 and β>0\beta > 0: 1/t<tβ−11/t < t^{\beta - 1}

Because tβ−1=tβ/tt^{\beta - 1} = t^\beta / t and tβ>1t^\beta > 1 when t>1t > 1.

Integrate from 11 to xx (for x>1x > 1)

log⁡x=∫1xdtt<∫1xtβ−1 dt=xβ−1β<xββ.\log x = \int_1^x \frac{dt}{t} < \int_1^x t^{\beta - 1}\,dt = \frac{x^\beta - 1}{\beta} < \frac{x^\beta}{\beta}.

Choose β\beta between 00 and α\alpha

For example β=α/2\beta = \alpha/2. Then for x>1x > 1,

0<x−αlog⁡x<xβ−αβ=2α x−α/2.0 < x^{-\alpha}\log x < \frac{x^{\beta - \alpha}}{\beta} = \frac{2}{\alpha}\, x^{-\alpha/2}.

Squeeze

Since α/2>0\alpha/2 > 0, x−α/2→0x^{-\alpha/2} \to 0 as x→∞x \to \infty, so the right-hand side tends to 00, and x−αlog⁡xx^{-\alpha}\log x, trapped between 00 and it, tends to 00 as well.

The bound log⁡x<xβ/β\log x < x^\beta/\beta is worth remembering by itself. With β=12\beta = \tfrac12 it says log⁡x<2x\log x < 2\sqrt{x} for all x>1x > 1; with β=1\beta = 1, log⁡x<x\log x < x.

Proof The corollary: (log⁡x)k/xα→0(\log x)^k / x^\alpha \to 0
Strategy

Take the kk-th root to reduce to the theorem.

Rewrite

For x>1x > 1, (log⁡x)kxα=(log⁡xxα/k)k\dfrac{(\log x)^k}{x^\alpha} = \left(\dfrac{\log x}{x^{\alpha/k}}\right)^k.

Apply the theorem with exponent α/k\alpha/k

The bracket is positive and tends to 00, so its kk-th power tends to 00 (for rational kk by the rules for limits; for any k>0k > 0 because u↦uku \mapsto u^k is continuous at 00 with value 00).

Proof xαlog⁡x→0x^\alpha\log x \to 0 as x→+0x \to +0
Strategy

Turn x→+0x \to +0 into y→∞y \to \infty by putting x=1/yx = 1/y, and use log⁡(1/y)=−log⁡y\log(1/y) = -\log y.

Substitute x=1/yx = 1/y

As x→+0x \to +0, y→+∞y \to +\infty. And xαlog⁡x=y−αlog⁡(1/y)=− y−αlog⁡yx^\alpha \log x = y^{-\alpha}\log(1/y) = -\,y^{-\alpha}\log y.

Apply the theorem

y−αlog⁡y→0y^{-\alpha}\log y \to 0 as y→∞y \to \infty, so xαlog⁡x→0x^\alpha\log x \to 0. Equivalently log⁡(1/x)/(1/x)α→0\log(1/x) \big/ (1/x)^\alpha \to 0: the logarithm of 1/x1/x is beaten by every power of 1/x1/x.

Explore xlog⁡xx \log x near 00

Try this. Move the trace point towards x=0x = 0 from the right and read off xlog⁡xx\log x.

y=xlog⁡xy = x\log xHardy: Examples LXXXVIII, 16
xy
  • x log x
  • trace point (drag)
At x=0.3x = 0.3: y=−0.36119y = -0.36119
What to notice

Although log⁡x→−∞\log x \to -\infty, the factor x→0x \to 0 wins: the product creeps back up to 00. Its lowest point is at x=1/e≈0.37x = 1/e \approx 0.37, where the value is −1/e-1/e. The same happens with x0.1log⁡xx^{0.1}\log x, only the dip is much deeper and the recovery much later.

Worked examples and exercises

Exercise An explicit bound ●○○ Hardy: §208

Show that log⁡x<2x\log x < 2\sqrt{x} for every x>0x > 0, and that log⁡x<10 x0.1\log x < 10\, x^{0.1} for every x>0x > 0.

Hint
Use log⁡x<(xβ−1)/β\log x < (x^\beta - 1)/\beta for x>1x > 1. What about 0<x≤10 < x \le 1?

Solution

For x>1x > 1 the proof above gives log⁡x<(xβ−1)/β<xβ/β\log x < (x^\beta - 1)/\beta < x^\beta/\beta. With β=12\beta = \frac12 this is log⁡x<2x\log x < 2\sqrt x; with β=0.1\beta = 0.1 it is log⁡x<10x0.1\log x < 10x^{0.1}. For 0<x≤10 < x \le 1 we have log⁡x≤0\log x \le 0, while 2x2\sqrt x and 10x0.110x^{0.1} are positive. So both inequalities hold for all x>0x > 0.

Exercise How late is “eventually”? ●●○ Hardy: §207

Show that log⁡x>x0.01\log x > x^{0.01} at x=10100x = 10^{100}, although log⁡x/x0.01→0\log x / x^{0.01} \to 0. Estimate where log⁡x\log x finally drops below x0.01x^{0.01} for good.

Hint
Write u=log⁡xu = \log x. The inequality log⁡x<x0.01\log x < x^{0.01} becomes u<e0.01uu < e^{0.01u}, or log⁡u<0.01 u\log u < 0.01\,u, that is, u>100log⁡uu > 100\log u.
Hint
log⁡10≈2.3026\log 10 \approx 2.3026, so log⁡10100≈230\log 10^{100} \approx 230.

Solution

At x=10100x = 10^{100}: log⁡x=100log⁡10≈230.3\log x = 100 \log 10 \approx 230.3, while x0.01=101=10x^{0.01} = 10^{1} = 10. So log⁡x\log x is far ahead.

With u=log⁡xu = \log x, we have x0.01=e0.01ux^{0.01} = e^{0.01u}, and log⁡x<x0.01\log x < x^{0.01} means log⁡u<0.01u\log u < 0.01u, i.e. u>100log⁡uu > 100\log u. The function u−100log⁡uu - 100\log u decreases until u=100u = 100 and then increases, so there is a last crossing u1>100u_1 > 100 with u1=100log⁡u1u_1 = 100\log u_1. Trying values: at u=600u = 600, 100log⁡600≈639.7>600100\log 600 \approx 639.7 > 600; at u=650u = 650, 100log⁡650≈647.7<650100\log 650 \approx 647.7 < 650. Bisection gives u1≈647.3u_1 \approx 647.3. So log⁡x<x0.01\log x < x^{0.01} for good only once x>e647.3≈10281x > e^{647.3} \approx 10^{281}. The theorem is about the limit, and says nothing about where the limit “starts to show”.

Exercise ●○○ Hardy: §209

Show that xα{log⁡(1/x)}k→0x^\alpha \{\log(1/x)\}^k \to 0 as x→+0x \to +0, for every α>0\alpha > 0 and k>0k > 0.

Hint
Put x=1/yx = 1/y and use the corollary.

Solution

With x=1/yx = 1/y, y→∞y \to \infty and xα{log⁡(1/x)}k=(log⁡y)k/yα→0x^\alpha\{\log(1/x)\}^k = (\log y)^k / y^\alpha \to 0 by the corollary.

Exercise Sequences ●○○ Hardy: §208

Show that (log⁡n)10n100→0\dfrac{(\log n)^{10}}{\sqrt[100]{n}} \to 0 as n→∞n \to \infty, and that ∑n≥2(log⁡n)10n2\sum_{n \ge 2} \dfrac{(\log n)^{10}}{n^{2}} converges.

Hint
A limit of a function as x→∞x \to \infty gives the same limit along x=nx = n. For the series, compare with ∑n−3/2\sum n^{-3/2}.

Solution

The corollary with k=10k = 10, α=0.01\alpha = 0.01 gives (log⁡x)10/x0.01→0(\log x)^{10}/x^{0.01} \to 0 as x→∞x \to \infty, hence also along the integers.

For the series: by the corollary with α=12\alpha = \tfrac12, (log⁡n)10/n1/2→0(\log n)^{10}/n^{1/2} \to 0, so it is less than 11 for n≥n0n \ge n_0. Then (log⁡n)10/n2<n1/2/n2=n−3/2(\log n)^{10}/n^2 < n^{1/2}/n^2 = n^{-3/2} for n≥n0n \ge n_0, and ∑n−3/2\sum n^{-3/2} converges (§181). By comparison the series converges. Logarithmic factors never change the convergence of ∑n−s\sum n^{-s} when s≠1s \ne 1; the case s=1s = 1 is the subject of §218.

Check yourself

Check yourself Choose one

Which of these tends to 00 as x→∞x \to \infty?

Why

Any power of log⁡x\log x over any positive power of xx tends to 00, so the second tends to 00 and the first (its reciprocal) to ∞\infty. The third is the constant log⁡x/(0.01log⁡x)=100\log x / (0.01\log x) = 100. The large exponent 100100 only delays the moment when x0.01x^{0.01} takes over, to around x=e116 700≈1050 670x = e^{116\,700} \approx 10^{50\,670} (the ratio peaks where log⁡x=100/0.01=104\log x = 100/0.01 = 10^4).

Check yourself True or false?

At x=1010x = 10^{10} we have log⁡x≈23>10=x0.1\log x \approx 23 > 10 = x^{0.1}. This shows that log⁡x/x0.1\log x / x^{0.1} does not tend to 00.

Why

False. A limit as x→∞x \to \infty is about all sufficiently large xx, and no single value decides it. The ratio rises until log⁡x=10\log x = 10 (that is x=e10≈22 000x = e^{10} \approx 22\,000) and then falls for ever; it drops below 11 near x≈3.4×1015x \approx 3.4 \times 10^{15} and tends to 00.

Check yourself Enter a number

The inequality log⁡x<xβ/β\log x < x^\beta/\beta with β=12\beta = \tfrac12 gives an upper bound for log⁡106\log 10^6. What is it?

Why

x1/2/(1/2)=2106=2000x^{1/2}/(1/2) = 2\sqrt{10^6} = 2000. The true value is log⁡106≈13.8\log 10^6 \approx 13.8, so the bound is very crude, but crude bounds are all a limit proof needs.

Check yourself Choose one

As x→+0x \to +0, x log⁡x\sqrt{x}\,\log x

Why

Putting x=1/yx = 1/y gives −log⁡y/y→0-\log y/\sqrt y \to 0. Any positive power of xx beats the logarithm at 00, just as at infinity.

Hardy’s language

Summary

Summary
  • log⁡x→∞\log x \to \infty, but log⁡x/xα→0\log x / x^\alpha \to 0 for every α>0\alpha > 0; also (log⁡x)k/xα→0(\log x)^k/x^\alpha \to 0 for every kk.
  • Proof: 1/t<tβ−11/t < t^{\beta - 1} for t>1t > 1, so log⁡x<xβ/β\log x < x^\beta/\beta; choose 0<β<α0 < \beta < \alpha.
  • At 00: xαlog⁡x→0x^\alpha \log x \to 0 as x→+0x \to +0 (put x=1/yx = 1/y).
  • “Eventually” can be very late: log⁡x\log x stays ahead of x0.01x^{0.01} until about x=10281x = 10^{281}.