The series 1 + 2^(−s) + 3^(−s) + … converges exactly when s > 1, and it becomes the standard yardstick for every series whose terms behave like a power of n.
The geometric series decides series whose terms shrink geometrically. Most series that arise from algebra shrink like a power of n: n3+53n+1 behaves like 3/n2, n2+11 like 1/n. To decide those we need to know exactly which powers give convergent series. §181 answers this with the integral test, and the answer is the single most useful fact in the chapter: ∑n−s converges if s>1 and diverges if s≤1. Combined with the comparison theorem D, it decides at a glance every series whose terms are asymptotic to a power of n.
The idea
For s>0 the terms n−s decrease, so the integral test applies with ϕ(x)=x−s. The question becomes: does the area under y=x−s from 1 to ξ stay bounded as ξ→∞? The integral is easy to compute, and the answer turns on whether ξ1−s tends to 0 or to ∞, that is, on whether s>1 or s<1. The borderline s=1 is the harmonic series, which diverges.
But knowing that a series converges is not the same as being able to sum it in practice. Near the borderline the convergence is extremely slow.
Explore∑n−s for every s
Try this.Start at s=1.5, then move s slowly down to 1 and below. Then set s=1.1 and compare the partial sum at N=105 with the true sum. Finally try s=3.
∑n=1Nn1.51 for N up to 105
N (log scale)S_N
partial sums S_N
1 + ∫₁ᴺ x⁻ˢ dx, always above S_N
Hardy’s bound s/(s − 1)
the sum ζ(s)
S100000=2.60605. The series converges, to 2.61238, which is below s/(s−1)=3. The part still missing after N terms is about N1−s/(s−1), so an error below 10−3 needs roughly N≈106.6 terms.
What to notice
For s>1 the partial sums level off under the ochre curve, below the dashed bound s/(s−1). For s≤1 they keep rising. At s=1.1 the picture looks almost the same as at s=1: after 105 terms the partial sum is still far below the sum ≈10.58, and an accuracy of 0.001 would need about 1040 terms. Convergence and practical summability are different things.
Precise statements
At this point in the book xs is defined only for rational s (irrational powers come in Ch. IX), so Hardy states the result for rational s. The proof is word for word the same for every real s once powers are available.
TheoremThe series ∑n−sHardy: §181
The series
1−s+2−s+3−s+⋯+n−s+⋯
converges if s>1 and diverges if s≤1. When s>1 its sum is less than s/(s−1).
Together with Theorem D (§173) this gives a test that covers almost every series with “algebraic” terms.
CorollaryComparison with ∑n−s
If vn>0 and nsvn tends to a limit l=0 (that is, vn∼ln−s), then ∑vn converges if s>1 and diverges if s≤1.
For example, n3+53n+1∼3n−2, so ∑n3+53n+1 converges; n2+11∼n−1, so ∑n2+11 diverges. More generally (Examples LXXI, 3), ∑ns/(nt+a) with a,t>0 converges exactly when t>1+s.
Proofs
ProofThe series ∑n−s
Strategy
Dispose of s≤0 (the terms do not tend to 0). For s>0 apply the integral test to ϕ(x)=x−s, which is positive, continuous and decreasing. Compute Φ(ξ)=∫1ξx−sdx and see whether it has a limit. The case s=1 needs a separate argument, since its integral is a logarithm.
s≤0: divergence
Then n−s≥1 for every n, so the terms do not tend to 0 and the series diverges.
The integral for s>0, s=1
An antiderivative of x−s is x1−s/(1−s), so
Φ(ξ)=∫1ξxsdx=1−sξ1−s−1.
s>1: convergence, with the bound
Now 1−s<0, so ξ1−s→0 and Φ(ξ)→s−11=l. By the integral test the series converges, and its sum is less than ϕ(1)+l=1+s−11=s−1s.
0<s<1: divergence
Now 1−s>0, so ξ1−s→∞ and Φ(ξ)→∞. The series diverges. (So, more simply, does comparison with ∑1/n, since n−s≥n−1.)
s=1 without logarithms
Split ∫12ndx/x at 2,4,…,2n−1. The substitution x=2ru turns ∫2r2r+1dx/x into ∫12du/u, the same positive number c for every r. So Φ(2n)=nc→∞; since Φ increases, Φ(ξ)→∞, and the harmonic series diverges. (This is the integral version of the grouping argument of §77.)
∎
ProofComparison with ∑n−s
Strategy
Theorem D with un=n−s.
Apply Theorem D
vn/un=nsvn→l=0, so ∑vn and ∑n−s converge or diverge together.
Read off the answer
By the theorem, ∑n−s converges exactly when s>1.
∎
The picture of the integral test shows where the bound s/(s−1) comes from, and how good it is.
ExploreWhere s/(s−1) comes from
Try this.Set s=2 and slide the slivers left. Then try s=1.2 and s=3.
ϕ(x)=x−2: rectangles of height ϕ(ν) against the area under the curve
xy
rectangles: total area ϕ(1)+⋯+ϕ(n); shaded: ∫1n+1ϕ
the excess slivers
Slivers
ϕ(1)+⋯+ϕ(10)=1.5498, ∫111ϕ=0.90909; the slivers have total area 0.6407, which is less than ϕ(1)=1 however large n is. So the sums stay bounded exactly when the integrals do: here ∫1∞ϕ=1, the series converges (to 1.6449), and its sum lies between 1 and ϕ(1)+1=2.
What to notice
The rectangles exceed the area under the curve, ∫1∞x−sdx=1/(s−1), by the slivers, which fit in the column of height ϕ(1)=1. Hence the sum lies between 1/(s−1) and 1+1/(s−1)=s/(s−1). For s=2 these are 1 and 2, and the true sum is π2/6≈1.645.
Worked examples and exercises
Exercise●○○Hardy: Examples LXXI, 2
Show that ∑n−2, ∑n−3/2 and ∑n−11/10 converge, with sums at most 2, 3 and 11 respectively, and that ∑n−1/2 and ∑n−10/11 diverge.
Solution
The exponents 2, 23, 1011 exceed 1, so the series converge, with sums below s/(s−1): 2/1=2, 23/21=3, 1011/101=11. The exponents 21 and 1110 are at most 1, so those series diverge. (The bounds are crude near s=1: ∑n−11/10≈10.58.)
Exercise●○○Hardy: Examples LXXI, 3
Let a>0 and t>0. Show that ∑nt+ans converges if t>1+s and diverges if t≤1+s.
Hint
Divide the term by ns−t.
Solution
ns−tns/(nt+a)=nt+ant=1+an−t1→1, since n−t→0 for t>0. By the Corollary the series behaves like ∑n−(t−s), which converges exactly when t−s>1, that is t>1+s. (Hardy’s examples tacitly have t>0; if t<0 the term behaves like ns/a instead.)
Exercise●●○Hardy: Examples LXXI, 5
Show that, for every positive integer m,
m21+(m+1)21+(m+2)21+⋯<m2m+1.Hint
Apply the bound of the integral test to the series that starts at n=m.
Solution
The integral test, started at x=m instead of x=1, says that ∑n≥mϕ(n)<ϕ(m)+∫m∞ϕ. With ϕ(x)=x−2: ϕ(m)=1/m2 and ∫m∞x−2dx=1/m. So the tail is less than m21+m1=m2m+1. The same picture gives the lower bound ∫m∞x−2dx=1/m: the tail after m−1 terms is about 1/m, which is why ∑1/n2 needs about 107 terms for 7 correct decimals.
Exercise●●○Hardy: Examples LXXI, 6
Show that n=1∑∞n2+11<21+41π.
Hint
Use the integral test with ϕ(x)=1/(1+x2) and ∫dx/(1+x2)=arctanx.
Solution
ϕ(x)=1/(1+x2) is positive, continuous and decreasing for x≥1. ∫1ξϕ=arctanξ−41π→21π−41π=41π=l. By the integral test the series converges and its sum is less than ϕ(1)+l=21+41π≈1.285. (The sum is about 1.077; the lower bound from the same picture is 41π≈0.785.) Choose 1/(1+x2) in the widget above to see the picture.
Exercise●●○Hardy: Examples LXXI, 9
If ϕ(n)→l>1, show that ∑n−ϕ(n) converges; if ϕ(n)→l<1, show that it diverges.
Hint
Pick s strictly between l and 1.
Solution
If l>1, choose s with 1<s<l. For large n, ϕ(n)>s, so n−ϕ(n)<n−s (because n≥1), and ∑n−s converges. If l<1, choose s with l<s<1; for large n, ϕ(n)<s and n−ϕ(n)>n−s, with ∑n−s divergent. The case l=1 is genuinely undecided: ϕ(n)=1 diverges, ϕ(n)=1+2loglogn/logn converges (Ch. IX).
Exercise●●●Hardy: Examples LXXI, 1
Without integrating, show that ∫1ξx−sdx→∞ as ξ→∞ when s<1.
Hint
Imitate the case s=1: split at the powers of 2 and substitute x=2ru.
Solution
For s<1, substituting x=2ru gives
∫2r2r+1xsdx=2r(1−s)∫12usdu=2r(1−s)c,c>0.
Since 1−s>0, 2r(1−s)≥1, so each of these pieces is at least c. Therefore ∫12nx−sdx≥nc→∞, and since the integral increases with ξ it tends to ∞.
Check yourself
Check yourselfTrue or false?
∑1/n1.0001 converges.
Why
True: s=1.0001>1. The sum is below s/(s−1)=10001 (it is about 10000.58), and in practice the series is hopeless to sum by adding terms. Convergence says nothing about speed.
Check yourselfTrue or false?
∑n−s converges for every s>0, because then n−s→0.
Why
False. Terms tending to 0 are necessary, not sufficient. For 0<s≤1 the series diverges, for example ∑1/n. The dividing line is s=1, not s=0.
Check yourselfChoose one
Which of these series converges?
Why
Find the power of n that each term behaves like. 1/n(n+1)∼n−1: diverges. n/(n2−1)∼n−3/2: converges. n/(n2+1)∼n−1: diverges. 1/(n1/2+n1/3)∼n−1/2: diverges (the larger power in the denominator dominates).
Check yourselfEnter a number
Hardy’s bound for the sum of ∑n−3/2 is s/(s−1). What is it?
Why
23/21=3. The true sum is ζ(23)≈2.612.
Check yourselfEnter a number
The tail ∑n>N1/n2 is about 1/N. Roughly how many terms of ∑1/n2 are needed for an error below 0.001?
Why
1/N<0.001 requires N>1000. Each extra correct decimal costs ten times as many terms, which is Hardy’s footnote: some ten million terms for seven decimals.
Hardy’s language
Much the most important application of the integral test is to the series 1⁻ˢ + 2⁻ˢ + 3⁻ˢ + … + n⁻ˢ + …, where s is any rational number.
— G. H. Hardy, §181
Summary
Summary
∑n−s converges for s>1 (with sum below s/(s−1)) and diverges for s≤1.
Proof: integral test with ∫1ξx−sdx=(ξ1−s−1)/(1−s); the harmonic case s=1 by doubling intervals.
If vn∼ln−s with l=0, ∑vn behaves like ∑n−s: find the dominant power and compare it with 1.
Near s=1 convergence is extremely slow: the tail after N terms is about N1−s/(s−1).