nx tends to 1 and n tends to ∞, so n(nx−1) is another tug of war, a "0×∞". §75 shows that it has a limit for every x>0, using only the monotone theorem and the inequalities of the previous lesson. The limit is a new function f(x) with the defining property of a logarithm, f(xy)=f(x)+f(y). Hardy cannot yet call it logx, because the logarithm has not been defined; Chapter IX makes the identification. This is the first time in the book that a genuinely new function is created as a limit.
The idea
Take x=2. Then n(21/n−1) is 1 for n=1, 2(2−1)≈0.828 for n=2, 3(32−1)≈0.780 for n=3, and so on: decreasing, and (as you will see) never below 21. A decreasing sequence bounded below has a limit.
Why should it decrease? Write n(x1/n−1)=1/nx1/n−1. This is the quotient g(r)=(xr−1)/r of the last lesson at r=1/n. That quotient increases with r, and r=1/n decreases as n grows. So the sequence decreases: the previous lesson did all the work.
ExploreA decreasing sequence with a positive floor
Try this.With x=2, find n0 for δ=0.01 (show more terms if needed). Then move x to 10 and to 0.2.
ϕ(n)=n(x1/n−1) , candidate limit l=0.6931
nφ(n)
band l ± δ
l
n₀ (drag)
n ≥ n₀, good
n ≥ n₀, bad
5 terms with n ≥ n₀ fall outside the band: n = 1, 2, 3, 4, 5.
What to notice
For x>1 the terms decrease towards a positive limit; for x<1 they are negative but also decrease, towards a negative limit (as 1/n decreases, g(1/n) decreases for every x=1). The candidate limit shown, 0.693… for x=2, is log2: the widget already knows what Chapter IX will prove.
Looking at all x at once shows the new function being born.
ExploreCurves settling onto one curve
Try this.Double n step by step from 1 to 65536 and watch the thick curve. Then choose x and y and compare fn(x)+fn(y) with fn(xy) in the read-out.
fn(x)=n(x1/n−1),n=8
xy
fₙ for n = 8
f₁, f₂, f₄ (earlier n)
1 − 1/x, a lower bound
f8(2)+f8(3)=1.90168 and f8(6)=2.00827: they differ by 0.107. As n grows the difference tends to 0, so the limit satisfies f(xy)=f(x)+f(y).
What to notice
The curves fn(x)=n(x1/n−1) come down for x>1 (and for x<1) and settle onto one limiting curve, always above the dashed lower bound 1−1/x. The difference fn(xy)−fn(x)−fn(y) shrinks towards 0 as n grows: in the limit, products become sums.
Precise statements
TheoremA function defined by a limitHardy: §75
For every x>0 the limit
f(x)=n→∞limn(nx−1)
exists. For x>1 the sequence decreases and f(x)≥1−1/x>0; f(1)=0; and f(1/x)=−f(x), so f(x)<0 for 0<x<1.
TheoremThe functional equationHardy: §75, Example
For all x,y>0: f(xy)=f(x)+f(y).
Proofs
ProofThe limit exists, and f(x)>0 for x>1
Strategy
Show that ϕ(n)=n(α1/n−1) decreases (by §74 (3)) and is bounded below by a positive number (by §74 (7)); then apply the monotone theorem.
Decreasing
Let α>1 and n≥2. Apply §74 (3), rαr−1>sαs−1, with r=n−11>s=n1. Since dividing by 1/(n−1) is multiplying by n−1, this says (n−1)(α1/(n−1)−1)>n(α1/n−1), that is, ϕ(n−1)>ϕ(n).
Bounded below
§74 (7) with s=1/n<1 gives α1/n−1>n1α1/n−1(α−1). Multiply by n: ϕ(n)>α1/n(1−α1)>1−α1, because α1/n>1.
Converges
A decreasing sequence bounded below tends to a limit (the decreasing form of the theorem of §69), and the limit is at least the lower bound: f(α)≥1−1/α>0.
∎
Proofx≤1: reduce to x>1
Strategy
For β<1 write β=1/α with α>1, and express the sequence for β through the one for α.
x=1
n(n1−1)=0 for every n, so f(1)=0.
An identity
With β=1/α: n(β1/n−1)=n(α1/n1−1)=−α1/nn(α1/n−1).
Take limits
The numerator tends to f(α) (just proved) and the denominator α1/n→1 (The limit of xn, Examples XXVII, 10). By Theorem IV, n(β1/n−1)→−f(α). So f(β) exists and f(1/α)=−f(α)<0.
∎
Prooff(xy)=f(x)+f(y)
Strategy
Split (xy)1/n−1 into a part involving x and a part involving y, so that each part is one of the sequences whose limits we know.
On the right, n(x1/n−1)→f(x), y1/n→1 and n(y1/n−1)→f(y). By Theorems I and II the right side tends to f(x)⋅1+f(y). The left side tends to f(xy). So f(xy)=f(x)+f(y).
∎
A function from positive numbers to reals that turns products into sums, is 0 at 1, and is positive for x>1, is behaving exactly like a logarithm. Chapter IX (§204) defines logx as an area, ∫1xdt/t, and then shows that this f is the same function.
Worked examples and exercises
ExerciseBounds for f(2)●○○Hardy: §75
Using the theorem and the first terms of the sequence, show that 21≤f(2)<0.83.
Solution
The lower bound f(α)≥1−1/α gives f(2)≥21. The sequence decreases strictly, so its limit is below each term; ϕ(2)=2(2−1)=0.8284…, so f(2)<0.83. (In fact f(2)=log2=0.6931…)
ExerciseAn upper bound for f●○○Hardy: §75
Show that f(x)≤x−1 for every x>0, and hence 1−1/x≤f(x)≤x−1.
Hint
For x>1, compare f(x) with the first term of a decreasing sequence. For x<1, use f(x)=−f(1/x) and the lower bound.
Solution
For x>1 the sequence decreases from ϕ(1)=x−1, so f(x)≤x−1. For x=1 both sides are 0. For 0<x<1, 1/x>1, so f(1/x)≥1−x, and f(x)=−f(1/x)≤x−1. For the lower bound when x<1: f(1/x)≤1/x−1 by the first part, so f(x)=−f(1/x)≥1−1/x. So 1−1/x≤f(x)≤x−1 for every x>0, the familiar bounds for logx.
ExercisePowers and quotients●●○Hardy: §75
Show that f(xk)=kf(x) for every positive integer k, that f(x/y)=f(x)−f(y), and that f is strictly increasing.
Solution
Powers. By the functional equation and induction, f(xk)=f(xk−1)+f(x)=⋯=kf(x).
Quotients.f(x)=f((x/y)y)=f(x/y)+f(y), so f(x/y)=f(x)−f(y).
Increasing. If x>y>0 then x/y>1, so f(x)−f(y)=f(x/y)≥1−y/x>0.
ExerciseAnother limit for f●●○Hardy: §75
Show that n(1−x−1/n)→f(x) as well, and that n(1−x−1/n)≤f(x)≤n(x1/n−1) for every n when x>1.
Solution
n(1−x−1/n)=−n((1/x)1/n−1)→−f(1/x)=f(x). For x>1, the sequence n(x1/n−1) decreases to f(x), so it is at least f(x). And n(1−x−1/n)=n(x1/n−1)x−1/n; the sequence ψ(n)=−n((1/x)1/n−1) is −1 times a sequence that decreases (by the same §74 argument applied to 1/x<1, where g(1/n) decreases as n grows), so ψ increases to f(x) and is at most f(x). This traps f(x) between two computable sequences: for x=2 and n=1000, 0.69290<f(2)<0.69339.
Check yourself
Check yourselfChoose one
Why does n(nα−1) decrease as n grows (for α>1)?
Why
The factor n grows while nα−1 shrinks, so neither factor alone decides the matter. The monotonicity of (αr−1)/r from §74 does. Knowing the limit would not show monotonicity anyway.
Check yourselfTrue or false?
f(x)<0 for 0<x<1.
Why
True: f(x)=−f(1/x) and f(1/x)≥1−x>0.
Check yourselfEnter a number
If f(a)=1.2, what is f(a)+f(1/a)?
Why
f(a)+f(1/a)=f(a⋅1/a)=f(1)=0. So f(1/a)=−1.2.
Hardy’s language
Later on we shall be able to identify this function with the Napierian logarithm of x.
— G. H. Hardy, §75
Summary
Summary
n(nx−1) is the quotient (xr−1)/r at r=1/n; since that quotient increases with r, the sequence decreases as n grows.
For x>1 it is bounded below by 1−1/x>0, so it converges to f(x)>0; f(1)=0 and f(1/x)=−f(x).
Splitting (xy)1/n−1 shows f(xy)=f(x)+f(y); also 1−1/x≤f(x)≤x−1.
This f is the natural logarithm, created as a limit before logarithms are defined.