Chapter IV · Limits of functions of a positive integral variable

Infinite series

§§76–77 35 min

An infinite series is nothing but the sequence of its partial sums; the general theorems that follow, and why terms tending to zero is not enough.

Why this matters

What could 1+12+14+18+⋯1 + \tfrac12 + \tfrac14 + \tfrac18 + \cdots mean? Nobody can add infinitely many numbers. In §76 Hardy gives the expression a meaning without any new machinery: an infinite series is a disguised sequence, the sequence of its partial sums, and its “sum” is that sequence’s limit. Everything proved about limits so far applies at once. The series that do not converge are just as instructive, and the most famous of them, 1+12+13+⋯1 + \tfrac12 + \tfrac13 + \cdots, shows that terms which tend to 00 are not enough.

The idea

Start with any sequence of terms u1,u2,u3,…u_1, u_2, u_3, \dots and form the running totals

s1=u1,s2=u1+u2,s3=u1+u2+u3,…,sn=u1+u2+⋯+un.s_1 = u_1, \quad s_2 = u_1 + u_2, \quad s_3 = u_1 + u_2 + u_3, \quad \dots, \quad s_n = u_1 + u_2 + \cdots + u_n.

Each sns_n is an ordinary finite sum. The question “what is u1+u2+u3+⋯u_1 + u_2 + u_3 + \cdots?” becomes “does the sequence sns_n tend to a limit?”, which is exactly the question of the earlier lessons. Two different sequences are involved, and keeping them apart is half the battle: the terms unu_n and the partial sums sns_n.

Explore Terms against partial sums

Try this. Switch the plot between partial sums and terms for ∑1/n\sum 1/n (blue) and ∑1/n2\sum 1/n^2 (ochre). Then show 1000 terms.

∑n=1∞1n\sum_{n=1}^{\infty} \frac1n compared with ∑n=1∞1n2\sum_{n=1}^{\infty} \frac{1}{n^2}
nsₙ
  • sₙ for 1n\frac1n
  • sₙ for 1n2\frac{1}{n^2}
  • the sum s
Plot
After 100 terms s100=5.187377518s_{100} = 5.187377518. The series diverges. The comparison series converges to 1.6449341.
What to notice

Both sequences of terms tend to 00 and look almost the same. The partial sums behave quite differently: those of ∑1/n2\sum 1/n^2 level off below 1.651.65, while those of ∑1/n\sum 1/n keep creeping upwards, by about 2.32.3 each time the number of terms is multiplied by ten. A picture cannot prove that they never level off; the argument below does.

The creeping upward of the harmonic sums is too slow to be convincing in a plot. Hardy’s argument makes it certain, by cutting the series into blocks that double in length.

Explore Why 1+12+13+⋯1 + \tfrac12 + \tfrac13 + \cdots diverges

Try this. Increase the number of blocks KK for ∑1/n\sum 1/n. Read the bound 1+K/21 + K/2 in the read-out. Then switch to ∑1/n2\sum 1/n^2 and do the same.

1+12+(13+14)+(15+⋯+18)+⋯1 + \tfrac12 + \big(\tfrac13 + \tfrac14\big) + \big(\tfrac15 + \dots + \tfrac18\big) + \cdots
block kblock sum
  • sum of the terms in block k (alternate blocks shaded differently)
  • every block sum is at least ½
Series
Block k≥1k \ge 1 holds the 2k−12^{k-1} terms with 2k−1<n≤2k2^{k-1} < n \le 2^k. Blocks 0 to 6 hold the first 26=642^{6} = 64 terms, and their sum is s64=4.74389s_{64} = 4.74389. Each term of block k is at least 1/2k1/2^k, so each block sums to at least ½ and s2K≥1+K/2=4s_{2^K} \ge 1 + K/2 = 4. Adding blocks adds at least ½ every time: the sums pass every bound, although the terms tend to 0.
What to notice

For 1/n1/n every block, however far out, sums to at least 12\tfrac12: the blocks get longer exactly as fast as their terms get smaller. So the partial sums pass every bound. For 1/n21/n^2 the terms shrink like the square of their position, block kk sums to less than 1/2k−11/2^{k-1}, a bound that halves from block to block, and the total stays below 33.

Precise statements

Definition Convergent series, sum Hardy: §76

Let u1,u2,…u_1, u_2, \dots be numbers and sn=u1+u2+⋯+uns_n = u_1 + u_2 + \cdots + u_n. If sn→ss_n \to s as n→∞n \to \infty, the series u1+u2+⋯u_1 + u_2 + \cdots is convergent, ss is its sum, and we write

∑ν=1∞uν=soru1+u2+u3+⋯=s.\sum_{\nu=1}^{\infty} u_\nu = s \qquad\text{or}\qquad u_1 + u_2 + u_3 + \cdots = s.

In terms of δ\delta: for every δ>0\delta > 0 there is n0(δ)n_0(\delta) with s−δ<sn<s+δs - \delta < s_n < s + \delta for all n≥n0(δ)n \ge n_0(\delta). If sn→+∞s_n \to +\infty or sn→−∞s_n \to -\infty the series diverges to +∞+\infty or −∞-\infty; if sns_n oscillates, finitely or infinitely, so does the series.

Conversely every sequence is the sequence of partial sums of some series: take u1=ϕ(1)u_1 = \phi(1) and un=ϕ(n)−ϕ(n−1)u_n = \phi(n) - \phi(n-1) for n≥2n \ge 2. Series and sequences are two ways of describing the same thing.

§77 lists the general theorems. The first six are bookkeeping; each follows from the theorems on sums and multiples of limits applied to sns_n.

Theorem Bookkeeping for series Hardy: §77 (1)–(6)

Suppose ∑un\sum u_n converges to ss and ∑vn\sum v_n converges to tt.

  • (1) Putting finitely many terms a,b,…,ka, b, \dots, k in front gives a convergent series with sum a+b+⋯+k+sa + b + \cdots + k + s.
  • (2) Deleting the first mm terms gives a convergent series um+1+um+2+⋯u_{m+1} + u_{m+2} + \cdots with sum s−u1−⋯−ums - u_1 - \cdots - u_m.
  • (3) If any of the series in (1) or (2) diverges or oscillates, so do the others.
  • (4) ∑kun\sum k u_n converges to ksks; and (5) if ∑un\sum u_n diverges or oscillates, so does ∑kun\sum k u_n, unless k=0k = 0.
  • (6) ∑(un+vn)\sum (u_n + v_n) converges to s+ts + t.

The next theorems are of a different kind: each says something about the terms or about rearranging the brackets.

Theorem The terms of a convergent series tend to zero Hardy: §77 (7)

If ∑un\sum u_n converges, then un→0u_n \to 0. The converse is false: un=1/nu_n = 1/n tends to 00, but ∑1/n\sum 1/n diverges to +∞+\infty.

Theorem Brackets may be inserted Hardy: §77 (8)

If ∑un\sum u_n converges to ss, then so does any series obtained by grouping consecutive terms in brackets and treating each bracket as one term. The converse is false: (1−1)+(1−1)+⋯=0+0+⋯(1 - 1) + (1 - 1) + \cdots = 0 + 0 + \cdots converges, but 1−1+1−1+⋯1 - 1 + 1 - 1 + \cdots oscillates.

Theorem Series of positive terms Hardy: §77 (9), (10)

If un≥0u_n \ge 0 for every nn, then ∑un\sum u_n either converges or diverges to +∞+\infty: it cannot oscillate. It converges if and only if there is a number KK with sn≤Ks_n \le K for every nn, and then its sum is at most KK.

Theorem Comparison Hardy: §77 (11)

Let un≥0u_n \ge 0 and vn≥0v_n \ge 0, and let un≤Kvnu_n \le K v_n for every nn, where KK is a constant. If ∑vn\sum v_n converges, so does ∑un\sum u_n, and its sum is at most KK times that of ∑vn\sum v_n. If ∑un\sum u_n diverges, so does ∑vn\sum v_n (when K>0K > 0).

Proofs

Proof un→0u_n \to 0 when ∑un\sum u_n converges
Strategy

A single term is the difference of two consecutive partial sums. If both partial sums are close to ss, their difference is close to 00.

un=sn−sn−1u_n = s_n - s_{n-1} for n≥2n \ge 2

The partial sum sns_n is sn−1s_{n-1} with unu_n added on.

Both sns_n and sn−1s_{n-1} tend to ss

The first by hypothesis; the second because sn−1s_{n-1} is the same sequence shifted by one place, and shifting changes no limit.

So un→s−s=0u_n \to s - s = 0

By the theorem on the limit of a difference (§63).

Proof 1+12+13+⋯1 + \tfrac12 + \tfrac13 + \cdots diverges to +∞+\infty
Strategy

The terms are positive, so the partial sums increase; it is enough to show that they are not bounded. Group the terms in blocks whose lengths double, and bound each block below by its length times its smallest term.

Block kk consists of the terms 1/n1/n with 2k−1<n≤2k2^{k-1} < n \le 2^k

For k=1,2,3,…k = 1, 2, 3, \dots this gives 12\tfrac12, then 13+14\tfrac13 + \tfrac14, then 15+⋯+18\tfrac15 + \cdots + \tfrac18, and so on. Block kk has 2k−12^{k-1} terms.

Each block sums to at least 12\tfrac12

Every term of block kk is at least 1/2k1/2^k, its last term. So the block sum is at least 2k−1⋅2−k=122^{k-1} \cdot 2^{-k} = \tfrac12.

s2K≥1+K/2s_{2^K} \ge 1 + K/2

The first 2K2^K terms are the term 11 followed by blocks 1,…,K1, \dots, K.

Conclude

Given any Δ\Delta, choose K>2ΔK > 2\Delta; then s2K>Δs_{2^K} > \Delta, and since sns_n increases, sn>Δs_n > \Delta for every n≥2Kn \ge 2^K. That is the definition of sn→+∞s_n \to +\infty.

Hardy leaves the bracket theorem to the reader. The point is that bracketing only thins out the sequence of partial sums.

Proof Brackets may be inserted
Strategy

The partial sums of the bracketed series are some of the partial sums of the original one, taken in order. A sequence that tends to ss still tends to ss when we keep only some of its terms.

Describe the bracketed series

Let the brackets end after the terms un1,un2,…u_{n_1}, u_{n_2}, \dots, where n1<n2<⋯n_1 < n_2 < \cdots. The jj-th partial sum of the bracketed series is Sj=snjS_j = s_{n_j}.

Use the n0n_0 for the original series

Let δ>0\delta > 0, and choose n0n_0 with ∣sn−s∣<δ|s_n - s| < \delta for n≥n0n \ge n_0. Since nj≥jn_j \ge j, we have nj≥n0n_j \ge n_0 whenever j≥n0j \ge n_0, so ∣Sj−s∣<δ|S_j - s| < \delta for all j≥n0j \ge n_0.

Why the converse fails

The thinned-out sequence can converge when the full one does not: sn=1,0,1,0,…s_n = 1, 0, 1, 0, \dots has the convergent subsequence s2,s4,⋯=0,0,…s_2, s_4, \dots = 0, 0, \dots.

Proof Series of positive terms, and comparison
Strategy

With un≥0u_n \ge 0 the partial sums form an increasing sequence, and §69 says that an increasing sequence either tends to a limit or tends to +∞+\infty, the first exactly when it is bounded.

The partial sums increase

sn+1=sn+un+1≥sns_{n+1} = s_n + u_{n+1} \ge s_n. By §69, sns_n either converges or tends to +∞+\infty; it cannot oscillate.

Bounded if and only if convergent

If sn≤Ks_n \le K for all nn, then sns_n cannot tend to +∞+\infty, so it converges, and its limit is at most KK (a limit of numbers ≤K\le K is ≤K\le K). Conversely a convergent sequence is bounded.

Comparison

If un≤Kvnu_n \le K v_n and ∑vn=t\sum v_n = t, then u1+⋯+un≤K(v1+⋯+vn)≤Ktu_1 + \cdots + u_n \le K(v_1 + \cdots + v_n) \le K t, because the partial sums of ∑vn\sum v_n increase to tt. So the partial sums of ∑un\sum u_n are bounded by KtKt, and the previous step applies. If ∑un\sum u_n diverged while ∑vn\sum v_n converged, this would be a contradiction.

Without positive terms, bounded partial sums prove nothing: the partial sums of 1−1+1−⋯1 - 1 + 1 - \cdots are 1,0,1,0,…1, 0, 1, 0, \dots, bounded by 11, and the series oscillates.

Worked examples and exercises

Exercise ●○○ Hardy: Examples XXX, 13

Show that 1+11+11⋅2+11⋅2⋅3+⋯1 + \dfrac{1}{1} + \dfrac{1}{1\cdot 2} + \dfrac{1}{1 \cdot 2 \cdot 3} + \cdots converges.

Hint
Compare 1/n!1/n! with a geometric quantity.
Hint
n!=1⋅2⋅3⋯n≥2n−1n! = 1 \cdot 2 \cdot 3 \cdots n \ge 2^{n-1} for n≥1n \ge 1.

Solution

The terms are positive. For n≥1n \ge 1, each of the factors 2,3,…,n2, 3, \dots, n of n!n! is at least 22, so n!≥2n−1n! \ge 2^{n-1} and 1/n!≤1/2n−11/n! \le 1/2^{n-1}. The series 1+12+14+⋯1 + \tfrac12 + \tfrac14 + \cdots converges (to 22; see the next lesson), so by comparison (§77 (11), with K=1K = 1) the series ∑n≥11/n!\sum_{n \ge 1} 1/n! converges with sum at most 22, and adding the initial term 11 keeps it convergent (§77 (1)). Its sum is at most 33. (It is e=2.71828…e = 2.71828\ldots; Chapter IX.)

Exercise ●●○ Hardy: Examples XXX, 15

Let a>0a > 0 and b>0b > 0. Show that 1a+1a+b+1a+2b+⋯\dfrac{1}{a} + \dfrac{1}{a+b} + \dfrac{1}{a + 2b} + \cdots diverges to +∞+\infty.

Hint
Compare with the harmonic series, but in the direction that proves divergence: you need a lower bound for each term.
Hint
For n≥1n \ge 1, a+nb≤n(a+b)a + nb \le n(a + b).

Solution

The general term is un=1/(a+nb)u_n = 1/(a + nb) for n=0,1,2,…n = 0, 1, 2, \dots. For n≥1n \ge 1 we have a+nb≤na+nb=n(a+b)a + nb \le na + nb = n(a + b), so

un≥1a+b⋅1n.u_n \ge \frac{1}{a + b} \cdot \frac{1}{n}.

If ∑n≥1un\sum_{n\ge1} u_n converged, then by comparison (with K=a+bK = a + b) so would ∑1/n\sum 1/n, which diverges. So ∑n≥1un\sum_{n \ge 1} u_n, a series of positive terms, diverges to +∞+\infty, and putting back the first term 1/a1/a changes nothing (§77 (3)).

Exercise Telescoping series ●○○ Hardy: Examples XXX, 16

Show that (u0−u1)+(u1−u2)+(u2−u3)+⋯(u_0 - u_1) + (u_1 - u_2) + (u_2 - u_3) + \cdots converges if and only if unu_n tends to a limit, and find its sum. Use this to sum ∑n≥11n(n+1)\sum_{n \ge 1} \dfrac{1}{n(n+1)}.

Hint
Write out the nn-th partial sum and watch the cancellation.

Solution

The nn-th partial sum is (u0−u1)+⋯+(un−1−un)=u0−un(u_0 - u_1) + \cdots + (u_{n-1} - u_n) = u_0 - u_n. This tends to a limit if and only if unu_n does, and if un→lu_n \to l the sum is u0−lu_0 - l.

Since 1n(n+1)=1n−1n+1\dfrac{1}{n(n+1)} = \dfrac1n - \dfrac{1}{n+1}, take un=1/(n+1)u_n = 1/(n+1) (so u0=1u_0 = 1): the series is telescoping with sum 1−0=11 - 0 = 1.

Exercise ●●○ Hardy: Examples XXX, 17

Show that if ∑un\sum u_n diverges (to +∞+\infty or −∞-\infty), then so does every series obtained from it by grouping terms in brackets. Does the same hold if “diverges” is replaced by “oscillates”?

Hint
As in the proof above, the partial sums of the bracketed series are a subsequence sn1,sn2,…s_{n_1}, s_{n_2}, \dots of the original partial sums.

Solution

Let sn→+∞s_n \to +\infty and let the brackets close after un1,un2,…u_{n_1}, u_{n_2}, \dots with n1<n2<⋯n_1 < n_2 < \cdots, so nj≥jn_j \ge j. Given Δ\Delta, choose n0n_0 with sn>Δs_n > \Delta for n≥n0n \ge n_0. For j≥n0j \ge n_0, nj≥n0n_j \ge n_0, so the bracketed partial sum Sj=snj>ΔS_j = s_{n_j} > \Delta. Hence Sj→+∞S_j \to +\infty. The case −∞-\infty is the same.

For oscillation the statement fails: 1−1+1−⋯1 - 1 + 1 - \cdots oscillates, and (1−1)+(1−1)+⋯(1 - 1) + (1 - 1) + \cdots converges. A subsequence of an oscillating sequence may converge.

Exercise ●○○ Hardy: Examples XXX, 18

Show that any series formed by selecting some of the terms (in their original order) of a convergent series of positive terms is itself convergent.

Hint
Use the bounded-partial-sums criterion, §77 (10).

Solution

Let ∑un=s\sum u_n = s with un≥0u_n \ge 0. Every partial sum of the selected series is a sum of finitely many distinct terms um1+⋯+umju_{m_1} + \cdots + u_{m_j} with mjm_j the largest index, so it is at most smj≤ss_{m_j} \le s. The selected series has positive terms and partial sums bounded by ss, so it converges, with sum at most ss. (Positivity is essential: from 1−1+12−12+13−13+⋯1 - 1 + \tfrac12 - \tfrac12 + \tfrac13 - \tfrac13 + \cdots, which converges to 00, select the positive terms and get the harmonic series.)

Check yourself

Check yourself True or false?

If un→0u_n \to 0, then ∑un\sum u_n converges.

Why

False. This is the converse of §77 (7), and the harmonic series refutes it. Terms tending to 00 are necessary for convergence, not sufficient. The useful form of the true theorem is the contrapositive: if unu_n does not tend to 00, the series cannot converge.

Check yourself Choose one

For the series 1+12+14+⋯1 + \tfrac12 + \tfrac14 + \cdots, which sequence has to tend to 22 for us to write 1+12+14+⋯=21 + \tfrac12 + \tfrac14 + \cdots = 2?

Why

The sum of a series is by definition the limit of its partial sums. The terms of a convergent series tend to 00, never to the sum (unless the sum happens to be 00 and that is a coincidence). There is no way to “add infinitely many numbers directly”: the limit is the definition.

Check yourself True or false?

Since (1−1)+(1−1)+⋯=0(1 - 1) + (1 - 1) + \cdots = 0, the series 1−1+1−1+⋯1 - 1 + 1 - 1 + \cdots has sum 00.

Why

False. Its partial sums are 1,0,1,0,…1, 0, 1, 0, \dots, which oscillate. Inserting brackets keeps only every second partial sum; that subsequence converges although the whole sequence does not. Brackets may be inserted in a convergent series, but not removed in general.

Check yourself Enter a number

The block argument gives s2K≥1+K/2s_{2^K} \ge 1 + K/2 for the harmonic series. What is the smallest KK for which this bound guarantees s2K≥5s_{2^K} \ge 5?

Why

1+K/2≥51 + K/2 \ge 5 means K≥8K \ge 8, so K=8K = 8: the first 28=2562^8 = 256 terms sum to at least 55. (In fact s256≈6.12s_{256} \approx 6.12; the bound is crude, but it is enough to prove divergence. The smallest nn with sn>5s_n > 5 is 8383.)

Check yourself Choose one

Which of these statements needs the hypothesis that every un≥0u_n \ge 0?

Why

Only the second. For positive terms the partial sums increase, and an increasing bounded sequence converges (§69). Without positivity, 1−1+1−⋯1 - 1 + 1 - \cdots has bounded partial sums but oscillates. The other three hold for all series, with any signs.

Hardy’s language

the consideration of such infinite series introduces no new ideas beyond those with which the early part of this chapter should already have made the reader familiar.

— G. H. Hardy, §76

Summary

Summary
  • A series ∑un\sum u_n converges to ss when its partial sums sn=u1+⋯+uns_n = u_1 + \cdots + u_n tend to ss. Series are sequences in disguise: every limit theorem applies to sns_n.
  • If ∑un\sum u_n converges then un→0u_n \to 0; the converse is false, as the harmonic series shows by its blocks of sum at least 12\tfrac12.
  • Brackets may be inserted in a convergent series without changing its sum, but not removed in general.
  • A series of terms ≥0\ge 0 cannot oscillate: it converges exactly when its partial sums are bounded. This gives the comparison test.