A.3 Chapter 3

Exercise 3.5

(i) Let aa, rr\in\mathbb{R} with r1r\neq 1 and, for nn\in\mathbb{N}, consider the partial sum sn:=k=1narks_{n}:=\sum_{k=1}^{n}ar^{k} of the series k=1ark\sum_{k=1}^{\infty}ar^{k}. It follows that

snrsn=k=1narkrk=1ark=k=1narkk=1nark+1.s_{n}-rs_{n}=\sum_{k=1}^{n}ar^{k}-r\sum_{k=1}^{\infty}ar^{k}=\sum_{k=1}^{n}ar^% {k}-\sum_{k=1}^{n}ar^{k+1}.

Relabelling the index of summation,

snrsn=k=1narkk=2n+1ark=ar+k=2narkk=2narkarn+1.s_{n}-rs_{n}=\sum_{k=1}^{n}ar^{k}-\sum_{k=2}^{n+1}ar^{k}=ar+\sum_{k=2}^{n}ar^{% k}-\sum_{k=2}^{n}ar^{k}-ar^{n+1}.

Since the middle two terms cancel, we have snrsn=ararn+1s_{n}-rs_{n}=ar-ar^{n+1}. If r1r\neq 1, then this rearranges to give

sn=ararn+11r=ar(1rn)1r,s_{n}=\frac{ar-ar^{n+1}}{1-r}=\frac{ar(1-r^{n})}{1-r},

as required.

(ii) Let aa, rr\in\mathbb{R} with r1r\neq 1. We shall show by induction on nn that

sn:=k=1narksatisfiessn=ar(1rn)1rfor all n.s_{n}:=\sum_{k=1}^{n}ar^{k}\qquad\text{satisfies}\qquad s_{n}=\frac{ar(1-r^{n}% )}{1-r}\qquad\text{for all $n\in\mathbb{N}$.}

Base case: For n=1n=1, it is immediate that

s1=ar=ar(1r)1r,s_{1}=ar=\frac{ar(1-r)}{1-r},

which establishes the base case.

Inductive step: Suppose, as an induction hypothesis, that the formula holds for some nn\in\mathbb{N}. Since sn+1=sn+arn+1s_{n+1}=s_{n}+ar^{n+1}, we may apply the induction hypothesis to deduce that

sn+1=ar(1rn)1r+arn+1=ar(1rn)1r+arn+1(1r)1r.s_{n+1}=\frac{ar(1-r^{n})}{1-r}+ar^{n+1}=\frac{ar(1-r^{n})}{1-r}+\frac{ar^{n+1% }(1-r)}{1-r}.

Thus,

sn+1=ararn+1+arn+1arn+21r=ar(1rn+1)1r.s_{n+1}=\frac{ar-ar^{n+1}+ar^{n+1}-ar^{n+2}}{1-r}=\frac{ar(1-r^{n+1})}{1-r}.

This shows that the formula is true for n+1n+1, which closes the induction and thus completes the proof.

Exercise 3.15

Observe that

k2+3k+2=(k+1)(k+2).k^{2}+3k+2=(k+1)(k+2).

We wish to form the partial fraction decomposition of 1k2+3k+2\frac{1}{k^{2}+3k+2}, which means writing

1k2+3k+2=1(k+1)(k+2)=Ak+1+Bk+2for some AB.\frac{1}{k^{2}+3k+2}=\frac{1}{(k+1)(k+2)}=\frac{A}{k+1}+\frac{B}{k+2}\qquad% \text{for some $A$, $B\in\mathbb{R}$.}

Multiplying through by the denominator,

1=(k+2)A+(k+1)B=k(A+B)+2A+B.1=(k+2)A+(k+1)B=k(A+B)+2A+B.

Since there is no dependence on kk on the left-hand side, we must have A+B=0A+B=0 and so B=AB=-A. Substituting this into the above, A=1A=1 and so A=1A=1 and B=1B=-1. Thus, we have

1k2+3k+2=1k+11k+2.\frac{1}{k^{2}+3k+2}=\frac{1}{k+1}-\frac{1}{k+2}.

From this, we see that k=11k2+3k+2\sum_{k=1}^{\infty}\frac{1}{k^{2}+3k+2} is a telescoping series. Using the cancellation, the nnth partial sum is given by

sn=k=1n(1k+11k+2)=121n+2for all n.s_{n}=\sum_{k=1}^{n}\left(\frac{1}{k+1}-\frac{1}{k+2}\right)=\frac{1}{2}-\frac% {1}{n+2}\quad\text{for all $n\in\mathbb{N}$.}

Since 1n+20\frac{1}{n+2}\to 0 as nn\to\infty, it follows that limnsn=12\lim_{n\to\infty}s_{n}=\frac{1}{2} and so, by definition, k=11k2+3k+2=12\sum_{k=1}^{\infty}\frac{1}{k^{2}+3k+2}=\frac{1}{2}.

Exercise 3.17

By writing

rk=1r1rk(r1)=1r1(rk+1rk),r^{k}=\frac{1}{r-1}\cdot r^{k}\cdot(r-1)=\frac{1}{r-1}\cdot(r^{k+1}-r^{k}),

we see that k=1rk\sum_{k=1}^{\infty}r^{k} is a telescoping series. Using the cancellation,

sn=1r1k=1n(rk+1rk)=rn+1rr1.s_{n}=\frac{1}{r-1}\cdot\sum_{k=1}^{n}(r^{k+1}-r^{k})=\frac{r^{n+1}-r}{r-1}.

Since |r|>1|r|>1, the sequence (rn+1)n(r^{n+1})_{n\in\mathbb{N}} is unbounded. It follows from the above formula that the sequence of partial sums (sn)n(s_{n})_{n\in\mathbb{N}} is unbounded and therefore diverges. Thus, by definition, the series diverges.

Exercise 3.20

(i) For all kk\in\mathbb{N}, by dividing the numerator and dominator by kk we have

kk+1=11+1/k.\frac{k}{k+1}=\frac{1}{1+1/k}.

Since 1/k01/k\to 0 as kk\to\infty, it therefore follows from the limit laws that kk+11\frac{k}{k+1}\to 1 as kk\to\infty. Consequently, by a result from Worksheet 4, the sequence ((1)kkk+1)k((-1)^{k}\frac{k}{k+1})_{k\in\mathbb{N}} diverges. Explicitly, the subsequence of odd terms tends to 1-1, whilst the subsequence of even terms tends to 11. It therefore follows from the subsequence test that ((1)kkk+1)k\big{(}(-1)^{k}\cdot\frac{k}{k+1}\big{)}_{k\in\mathbb{N}} is a divergent sequence. In particular, it does not converge to 0. Thus, by the kkth term test, the series k=1(1)kkk+1\sum_{k=1}^{\infty}(-1)^{k}\frac{k}{k+1} diverges.

(ii) For all kk\in\mathbb{N}, by dividing the numerator and dominator by k2k^{2} we have

k23k2k=131/k.\frac{k^{2}}{3k^{2}-k}=\frac{1}{3-1/k}.

Since 1/k01/k\to 0 as kk\to\infty, it therefore follows from the limit laws that k23k2k13\frac{k^{2}}{3k^{2}-k}\to\frac{1}{3} as kk\to\infty. Thus, by the kkth term test, the series k=1k23k2k\sum_{k=1}^{\infty}\frac{k^{2}}{3k^{2}-k} diverges.

(iii) Since (3k)k(3^{k})_{k\in\mathbb{N}} is unbounded, 3k↛03^{k}\not\to 0 as kk\to\infty. Thus, by the kkth term test, the series k=13k\sum_{k=1}^{\infty}3^{k} diverges.

Exercise 3.22

1. We write the nnth partial sum as

sn=k=1n(ak+bk)=k=1nak+k=1nbk=:An+Bnfor all n,s_{n}=\sum_{k=1}^{n}(a_{k}+b_{k})=\sum_{k=1}^{n}a_{k}+\sum_{k=1}^{n}b_{k}=:A_{% n}+B_{n}\qquad\text{for all $n\in\mathbb{N}$},

where we denote by AnA_{n} and BnB_{n} the nnth partial sum of k=1ak\sum_{k=1}^{\infty}a_{k} and k=1bk\sum_{k=1}^{\infty}b_{k}, respectively. By assumption, both k=1ak\sum_{k=1}^{\infty}a_{k} and k=1bk\sum_{k=1}^{\infty}b_{k} converge, and so both the sequences AnA_{n} and BnB_{n} converge, with limnAn=A\lim_{n\to\infty}A_{n}=A and limnBn=B\lim_{n\to\infty}B_{n}=B for A=k=1akA=\sum_{k=1}^{\infty}a_{k}, B=k=1bkB=\sum_{k=1}^{\infty}b_{k}\in\mathbb{R}. By the limit laws for sequences, the sequence An+BnA_{n}+B_{n} converges, with limnAn+Bn=A+B\lim_{n\to\infty}A_{n}+B_{n}=A+B. This implies that limnsn=A+B\lim_{n\to\infty}s_{n}=A+B which, by definition, means that the series k=1(ak+bk)\sum_{k=1}^{\infty}(a_{k}+b_{k}) converges and k=1(ak+bk)=k=1ak+k=1bk\sum_{k=1}^{\infty}(a_{k}+b_{k})=\sum_{k=1}^{\infty}a_{k}+\sum_{k=1}^{\infty}b% _{k}.

2. We write the nnth partial of k=1(λak)\sum_{k=1}^{\infty}(\lambda\cdot a_{k}) sum as

sn:=k=1n(λak)=λk=1nak=λAnfor all n,s_{n}:=\sum_{k=1}^{n}(\lambda\cdot a_{k})=\lambda\sum_{k=1}^{n}a_{k}=\lambda A% _{n}\qquad\text{for all $n\in\mathbb{N}$},

where AnA_{n} is as defined in the proof of part 1. By assumption, k=1ak\sum_{k=1}^{\infty}a_{k} converges and so the sequence AnA_{n} converges, with limnAn=A\lim_{n\to\infty}A_{n}=A for A=k=1akA=\sum_{k=1}^{\infty}a_{k}\in\mathbb{R}. By the limit laws for sequence, the sequence λAn\lambda A_{n} converges, with limnλAn=λA\lim_{n\to\infty}\lambda A_{n}=\lambda A. This implies that limnsn=λA\lim_{n\to\infty}s_{n}=\lambda A which, by definition, means that the series k=1(λak)\sum_{k=1}^{\infty}(\lambda\cdot a_{k}) converges and that k=1(λak)=λk=1ak\sum_{k=1}^{\infty}(\lambda\cdot a_{k})=\lambda\sum_{k=1}^{\infty}a_{k}.

Exercise 3.23

Let (ak)k(a_{k})_{k\in\mathbb{N}} be a sequence such that k=1ak\sum_{k=1}^{\infty}a_{k} converges. From our earlier observations, given mm\in\mathbb{N}, the tail k=m+1ak\sum_{k=m+1}^{\infty}a_{k} converges and

k=m+1ak=k=1akk=1mak.\sum_{k=m+1}^{\infty}a_{k}=\sum_{k=1}^{\infty}a_{k}-\sum_{k=1}^{m}a_{k}.

However, by definition limmk=1mak=k=1ak\lim_{m\to\infty}\sum_{k=1}^{m}a_{k}=\sum_{k=1}^{\infty}a_{k} and so, by the limit laws,

limmk=m+1ak=k=1aklimmk=1mak=k=1akk=1ak=0,\lim_{m\to\infty}\sum_{k=m+1}^{\infty}a_{k}=\sum_{k=1}^{\infty}a_{k}-\lim_{m% \to\infty}\sum_{k=1}^{m}a_{k}=\sum_{k=1}^{\infty}a_{k}-\sum_{k=1}^{\infty}a_{k% }=0,

as required.

Exercise 3.25

Let rr\in\mathbb{R} with |r|<1|r|<1. Let (tn)n(t_{n})_{n\in\mathbb{N}} denote the partial sums of the series k=1ark=ar1r\sum_{k=1}^{\infty}ar^{k}=\frac{ar}{1-r} and (sn)n0(s_{n})_{n\in\mathbb{N}_{0}} the partial sums of k=0ark\sum_{k=0}^{\infty}ar^{k}. It follows that sn=tn+as_{n}=t_{n}+a for all nn\in\mathbb{N}. Thus, by the limit laws and (3.2), we have

k=0ark=limnsn=limn(tn+a)=(limntn)+a=k=1ark+a=ar1r+a=a1r,\sum_{k=0}^{\infty}ar^{k}=\lim_{n\to\infty}s_{n}=\lim_{n\to\infty}(t_{n}+a)=(% \lim_{n\to\infty}t_{n})+a=\sum_{k=1}^{\infty}ar^{k}+a=\frac{ar}{1-r}+a=\frac{a% }{1-r},

as required.

Exercise 3.28

Let (ak)k(a_{k})_{k\in\mathbb{N}} and (bk)k(b_{k})_{k\in\mathbb{N}} satisfy 0akbk0\leq a_{k}\leq b_{k} for all kk\in\mathbb{N} and suppose k=1ak\sum_{k=1}^{\infty}a_{k} diverges.

Our goal is to show k=1bk\sum_{k=1}^{\infty}b_{k} diverges; that is, that the sequence of partial sums (k=1nbk)n\big{(}\sum_{k=1}^{n}b_{k}\big{)}_{n\in\mathbb{N}} is unbounded.

Let M>0M>0 be given. By the boundedness test from Lemma 3.26, since k=1ak\sum_{k=1}^{\infty}a_{k} diverges, the sequence of partial sums of (k=1nak)n\big{(}\sum_{k=1}^{n}a_{k}\big{)}_{n\in\mathbb{N}} is unbounded. In particular, there exists some nn\in\mathbb{N} such that

k=1nak>M.\sum_{k=1}^{n}a_{k}>M.

Since 0akbk0\leq a_{k}\leq b_{k} for all kk\in\mathbb{N}, we therefore see that

k=1nbkk=1nak>M.\sum_{k=1}^{n}b_{k}\geq\sum_{k=1}^{n}a_{k}>M.

Since M>0M>0 was chosen arbitrarily, it follows that the sequence of partial sums (k=1nbk)n\big{(}\sum_{k=1}^{n}b_{k}\big{)}_{n\in\mathbb{N}} is unbounded, as required.

Exercise 3.30

(i) For all kk\in\mathbb{N}, we have

03k+14k333k+k4k33k3=4kk3=4k2.0\leq\frac{3k+1}{4k^{3}-3}\leq\frac{3k+k}{4k^{3}-3k^{3}}=\frac{4k}{k^{3}}=% \frac{4}{k^{2}}.

Since we know k=14k2\sum_{k=1}^{\infty}\frac{4}{k^{2}} converges by Example 3.29 and the limit laws, it follows by the comparison test that k=13k+14k33\sum_{k=1}^{\infty}\frac{3k+1}{4k^{3}-3} converges.

(ii) If k3k\geq 3, then log2(k+1)log2(3+1)=log24=2\log_{2}(k+1)\geq\log_{2}(3+1)=\log_{2}4=2. Hence

01log2(k+1)k12kfor all k3.0\leq\frac{1}{\log_{2}(k+1)^{k}}\leq\frac{1}{2^{k}}\qquad\text{for all $k\geq 3% .$}

Since we know k=32k\sum_{k=3}^{\infty}2^{-k} is a convergent geometric series, it follows by the comparison test that the tail k=31log2(k+1)k\sum_{k=3}^{\infty}\frac{1}{\log_{2}(k+1)^{k}} converges. Hence, k=11log2(k+1)k\sum_{k=1}^{\infty}\frac{1}{\log_{2}(k+1)^{k}} also converges.

(iii) For k4k\geq 4, we have

k!=k(k1)(k2)(k3)(k4)!k(k1)(k2)(k3)(k3)4.k!=k(k-1)(k-2)(k-3)(k-4)!\geq k(k-1)(k-2)(k-3)\geq(k-3)^{4}.

Furthermore, if k6k\geq 6, then we have 3k/23\leq k/2 and so (k3)4(k/2)4=k4/16(k-3)^{4}\geq(k/2)^{4}=k^{4}/16. From this, we see that

0k2k!16k2k4=16k2for all k6.0\leq\frac{k^{2}}{k!}\leq\frac{16k^{2}}{k^{4}}=\frac{16}{k^{2}}\qquad\text{for% all $k\geq 6$.}

Since we know k=616k2\sum_{k=6}^{\infty}\frac{16}{k^{2}} converges by Example 3.29 and the limit laws, it follows by the comparison test that the tail k=6k2k!\sum_{k=6}^{\infty}\frac{k^{2}}{k!} converges. Thus, k=1k2k!\sum_{k=1}^{\infty}\frac{k^{2}}{k!} also converges.

Exercise 3.31

Let (dk)k(d_{k})_{k\in\mathbb{N}} be a sequence with dk{0,1,,9}d_{k}\in\{0,1,\dots,9\} for all kk\in\mathbb{N}. Then

0dk10k910kfor all k.0\leq d_{k}\cdot 10^{-k}\leq 9\cdot 10^{-k}\qquad\text{for all $k\in\mathbb{N}% $.}

We know from the geometric series formula that k=1910k=1\sum_{k=1}^{\infty}9\cdot 10^{-k}=1. Thus, by the comparison test, k=1dk10k\sum_{k=1}^{\infty}d_{k}\cdot 10^{-k} converges and

0x:=k=1dk10kk=1910k=1.0\leq x:=\sum_{k=1}^{\infty}d_{k}\cdot 10^{-k}\leq\sum_{k=1}^{\infty}9\cdot 10% ^{-k}=1.

It follows by definition of decimal expansion that x=0.d1d2d3x=0.d_{1}d_{2}d_{3}\ldots.

Exercise 3.35

(i) Let (ak)k(a_{k})_{k\in\mathbb{N}} be a positive sequence and suppose the sequence (rk)k(r_{k})_{k\in\mathbb{N}} given by rk:=ak+1/akr_{k}:=a_{k+1}/a_{k} for all kk\in\mathbb{N} converges with r:=limkrk>1r:=\lim_{k\to\infty}r_{k}>1.

Let ε:=r12>0\varepsilon:=\frac{r-1}{2}>0. By the ε\varepsilon-NN definition of a limit, there exists NN\in\mathbb{N} such that |rnr|<ε|r_{n}-r|<\varepsilon for all nNn\geq N. Thus, by the triangle inequality,

(A.5) (A.5) rn=|r+rnr|r|rnr|>rε=1+r2for all nN,r_{n}=|r+r_{n}-r|\geq r-|r_{n}-r|>r-\varepsilon=\frac{1+r}{2}\qquad\text{for % all $n\geq N$,}

For any kk\in\mathbb{N} with k>Nk>N, we may write

(A.6) (A.6) ak=aNaN+1aNak1ak2akak1=aNrNrk1.a_{k}=a_{N}\cdot\frac{a_{N+1}}{a_{N}}\cdot\dots\cdot\frac{a_{k-1}}{a_{k-2}}% \cdot\frac{a_{k}}{a_{k-1}}=a_{N}\cdot r_{N}\cdot\dots\cdot r_{k-1}.

Applying (A.5) to each of the factors rnr_{n} for Nnk1N\leq n\leq k-1, we deduce that

(A.7) (A.7) ak>aNskNwhere s:=1+r2 satisfies s>1a_{k}>a_{N}s^{k-N}\qquad\text{where $s:=\displaystyle\frac{1+r}{2}$ satisfies % $s>1$. }

Since s>1s>1, we know from Example 3.4 that k=N+1aNskN\sum_{k=N+1}^{\infty}a_{N}s^{k-N} is a divergent geometric series. In light of (A.6), the series k=N+1ak\sum_{k=N+1}^{\infty}a_{k} diverges by the comparison test. Finally, since we have shown a tail diverges, we conclude that the whole series k=1ak\sum_{k=1}^{\infty}a_{k} diverges.

(ii) Let (ak)k(a_{k})_{k\in\mathbb{N}} be a positive sequence and suppose the sequence (rk)k(r_{k})_{k\in\mathbb{N}} given by rk:=ak+1/akr_{k}:=a_{k+1}/a_{k} for all kk\in\mathbb{N} satisfies limkrk=\lim_{k\to\infty}r_{k}=\infty. Then there exists some NN\in\mathbb{N} such that rn>2r_{n}>2 for all nNn\geq N.

Given k>Nk>N, we may express aka_{k} as in (A.6) to deduce that ak>aN2kNa_{k}>a_{N}2^{k-N}. The proof now concludes as in part (i): we know from Example 3.4 that k=N+1aN2kN\sum_{k=N+1}^{\infty}a_{N}2^{k-N} diverges and so k=N+1ak\sum_{k=N+1}^{\infty}a_{k} diverges by the comparison test. Finally, since we have shown a tail diverges, we conclude that the whole series k=1ak\sum_{k=1}^{\infty}a_{k} diverges.

Exercise 3.36

(i) Let ak:=k322k3k+1a_{k}:=\frac{k^{3}2^{2k}}{3^{k}+1} for kk\in\mathbb{N}. Then

ak+1ak=(k+1)322k+23k+1+13k+1k322k=4(k+1)3k33k+13k+1+1=4(1+1k)33k+13k+1+1.\frac{a_{k+1}}{a_{k}}=\frac{(k+1)^{3}2^{2k+2}}{3^{k+1}+1}\cdot\frac{3^{k}+1}{k% ^{3}2^{2k}}=4\cdot\frac{(k+1)^{3}}{k^{3}}\cdot\frac{3^{k}+1}{3^{k+1}+1}=4\cdot% \Big{(}1+\frac{1}{k}\Big{)}^{3}\cdot\frac{3^{k}+1}{3^{k+1}+1}.

Since 1/k01/k\to 0 as kk\to\infty, it follows from the limit laws that (1+1k)31\big{(}1+\frac{1}{k}\big{)}^{3}\to 1 as kk\to\infty. Moreover, by multiplying the numerator and denominator by 3k3^{-k}, we see that

3k+13k+1+1=1+3k3+3k13as k,\frac{3^{k}+1}{3^{k+1}+1}=\frac{1+3^{-k}}{3+3^{-k}}\to\frac{1}{3}\qquad\text{% as $k\to\infty$,}

where we have again used the limit laws. Thus, by one final application of the limit laws,

ak+1ak43as k.\frac{a_{k+1}}{a_{k}}\to\frac{4}{3}\qquad\text{as $k\to\infty$.}

Since 4/3>14/3>1, by the ratio test, the series k=1k322k3k+1\sum_{k=1}^{\infty}\frac{k^{3}2^{2k}}{3^{k}+1} diverges.

(ii) Let bk:=5kk!b_{k}:=\frac{5^{k}}{k!} for kk\in\mathbb{N}. Then

bk+1bk=5k+1(k+1)!k!5k=5k+10as k.\frac{b_{k+1}}{b_{k}}=\frac{5^{k+1}}{(k+1)!}\cdot\frac{k!}{5^{k}}=\frac{5}{k+1% }\to 0\qquad\text{as $k\to\infty$.}

Since 0<10<1, by the ratio test, the series k=15kk!\sum_{k=1}^{\infty}\frac{5^{k}}{k!} converges.

(iii) Let ck:=k!5kc_{k}:=\frac{k!}{5^{k}} for kk\in\mathbb{N}. Then

ck+1ck=(k+1)!5k+15kk!=k+15as k.\frac{c_{k+1}}{c_{k}}=\frac{(k+1)!}{5^{k+1}}\cdot\frac{5^{k}}{k!}=\frac{k+1}{5% }\to\infty\qquad\text{as $k\to\infty$.}

Thus, by the ratio test, the series k=1k!5k\sum_{k=1}^{\infty}\frac{k!}{5^{k}} diverges.

(iv) Let dk:=k10(2k1)!d_{k}:=\frac{k^{10}}{(2k-1)!} for kk\in\mathbb{N}. Then

dk+1dk=(k+1)10(2k+1)!(2k1)!k10=(k+1)10k10(2k1)!(2k+1)!=(1+1k)1012k(2k+1).\frac{d_{k+1}}{d_{k}}=\frac{(k+1)^{10}}{(2k+1)!}\cdot\frac{(2k-1)!}{k^{10}}=% \frac{(k+1)^{10}}{k^{10}}\cdot\frac{(2k-1)!}{(2k+1)!}=\Big{(}1+\frac{1}{k}\Big% {)}^{10}\cdot\frac{1}{2k(2k+1)}.

Since 1/k01/k\to 0 as kk\to\infty, it follows from the limit laws that (1+1k)101\big{(}1+\frac{1}{k}\big{)}^{10}\to 1 as kk\to\infty. On the other hand, 012k(2k+1)1k0\leq\frac{1}{2k(2k+1)}\leq\frac{1}{k} for all kk\in\mathbb{N} and so, by the squeeze theorem, 12k(2k+1)0\frac{1}{2k(2k+1)}\to 0 as kk\to\infty. Thus, by another application of the limit laws,

dk+1dk0as k.\frac{d_{k+1}}{d_{k}}\to 0\qquad\text{as $k\to\infty$}.

Since 0<10<1, by the ratio test, the series k=1k10(2k1)!\sum_{k=1}^{\infty}\frac{k^{10}}{(2k-1)!} converges.

Exercise 3.40

Let

sn\displaystyle s_{n} :=k=1nak=a1+a2+a3+a4++an,\displaystyle:=\sum_{k=1}^{n}a_{k}=a_{1}+a_{2}+a_{3}+a_{4}+\cdots+a_{n},
tn\displaystyle t_{n} :=k=0n12ka2k=a1+2a2+4a4+8a8++2n1a2n1for all n,\displaystyle:=\sum_{k=0}^{n-1}2^{k}a_{2^{k}}=a_{1}+2a_{2}+4a_{4}+8a_{8}+% \cdots+2^{n-1}a_{2^{n-1}}\qquad\text{for all $n\in\mathbb{N}$,}

so that (sn)n(s_{n})_{n\in\mathbb{N}} and (tn)n(t_{n})_{n\in\mathbb{N}} are the sequences of partial sums of the series k=1ak\sum_{k=1}^{\infty}a_{k} and k=02ka2k\sum_{k=0}^{\infty}2^{k}a_{2^{k}}, respectively. Consider the subsequence (s2n)n(s_{2^{n}})_{n\in\mathbb{N}} and partition the terms of s2ns_{2^{n}} into ‘blocks’

s2n=(a1)1a1+(a2+a3)2a2+(a4+a6+a7+a8)4a4++(a2n1+a2n1+2++a2n1)2n1a2n1+a2n,s_{2^{n}}=\underbrace{(a_{1})}_{\displaystyle\leq 1\cdot a_{1}}+\underbrace{(a% _{2}+a_{3})}_{\displaystyle\leq 2\cdot a_{2}}\\ +\underbrace{(a_{4}+a_{6}+a_{7}+a_{8})}_{\displaystyle\leq 4\cdot a_{4}}+% \cdots+\underbrace{(a_{2^{n-1}}+a_{2^{n-1}+2}+\cdots+a_{2^{n}}-1)}_{% \displaystyle\leq 2^{n-1}\cdot a_{2^{n-1}}}+a_{2^{n}},

where we have used the hypothesis that (an)n(a_{n})_{n\in\mathbb{N}} is nonincreasing. Therefore

s2n\displaystyle s_{2^{n}} a1+2a2+4a4+8a8++2n1a2n1+a2n\displaystyle\leq a_{1}+2a_{2}+4a_{4}+8a_{8}+\cdots+2^{n-1}a_{2^{n-1}}+a_{2^{n}}
(A.8) (A.8) =tn+a2ntn+1.\displaystyle=t_{n}+a_{2^{n}}\leq t_{n+1}.

If the condensed series k=02ka2k\sum_{k=0}^{\infty}2^{k}a_{2^{k}} converges, then the sequence of partial sums (tn)n(t_{n})_{n\in\mathbb{N}} is bounded. By (A.8), the subsequence of partial sums (s2n)n(s_{2^{n}})_{n\in\mathbb{N}} is therefore also bounded, so there exists some M>0M>0 such that 0s2nM0\leq s_{2^{n}}\leq M for all nn\in\mathbb{N}. But then 0snM0\leq s_{n}\leq M for all nn\in\mathbb{N} and so the complete sequence of partial sums (sn)n(s_{n})_{n\in\mathbb{N}} is also bounded. Indeed, given nn\in\mathbb{N}, we have n2nn\leq 2^{n} and since the terms of the sequence (ak)k(a_{k})_{k\in\mathbb{N}} are non-negative, it follows that 0sns2nM0\leq s_{n}\leq s_{2^{n}}\leq M. Consequently, k=1ak\sum_{k=1}^{\infty}a_{k} converges by the boundedness test from Lemma 3.26.

Exercise 3.41

(i) Since the sequence of terms (1k(log2k)2)k=2\big{(}\frac{1}{k(\log_{2}k)^{2}}\big{)}_{k=2}^{\infty} is nonincreasing and non-negative, we can apply the condensation test. The condensed series is given by

k=12k2k(log22k)2=k=11k2,\sum_{k=1}^{\infty}\frac{2^{k}}{2^{k}(\log_{2}2^{k})^{2}}=\sum_{k=1}^{\infty}% \frac{1}{k^{2}},

which know converges. Thus, by the condensation test, k=21k(log2k)2\sum_{k=2}^{\infty}\frac{1}{k(\log_{2}k)^{2}} converges.

(ii) Since the sequence of terms (1(log2k)log2k)k=2\big{(}\frac{1}{(\log_{2}k)^{\log_{2}k}}\big{)}_{k=2}^{\infty} is nonincreasing and non-negative, we can apply the condensation test. The condensed series is given by

k=12k(log22k)log22k=k=12kkk.\sum_{k=1}^{\infty}\frac{2^{k}}{(\log_{2}2^{k})^{\log_{2}2^{k}}}=\sum_{k=1}^{% \infty}\frac{2^{k}}{k^{k}}.

We claim this series converges. Once we have proved the claim, it follows by the condensation test that k=21(log2k)log2k\sum_{k=2}^{\infty}\frac{1}{(\log_{2}k)^{\log_{2}k}} converges.

We can show k=12kkk\sum_{k=1}^{\infty}\frac{2^{k}}{k^{k}} converges using the ratio test. Indeed, let ak:=2kkka_{k}:=\frac{2^{k}}{k^{k}} for all kk\in\mathbb{N} so that

0ak+1ak=2k+1(k+1)k+1kk2k=2k+12k1k+1kk(k+1)k=2k+1(11k+1)k.0\leq\frac{a_{k+1}}{a_{k}}=\frac{2^{k+1}}{(k+1)^{k+1}}\cdot\frac{k^{k}}{2^{k}}% =\frac{2^{k+1}}{2^{k}}\cdot\frac{1}{k+1}\cdot\frac{k^{k}}{(k+1)^{k}}=\frac{2}{% k+1}\cdot\Big{(}1-\frac{1}{k+1}\Big{)}^{k}.

Observe that

0(11k+1)k1and so0ak+1ak2k+12kfor all k.0\leq\Big{(}1-\frac{1}{k+1}\Big{)}^{k}\leq 1\qquad\text{and so}\qquad 0\leq% \frac{a_{k+1}}{a_{k}}\leq\frac{2}{k+1}\leq\frac{2}{k}\qquad\text{for all $k\in% \mathbb{N}$.}

Thus, by the squeeze theorem, ak+1/ak0a_{k+1}/a_{k}\to 0 as kk\to\infty. Since 0<10<1, by the ratio test, the series k=12kkk\sum_{k=1}^{\infty}\frac{2^{k}}{k^{k}} converges, as claimed.

Exercise 3.44

(i) Observe that

0k1/3k2+10k1/3k2=1k5/3for all k.0\leq\frac{k^{1/3}}{k^{2}+10}\leq\frac{k^{1/3}}{k^{2}}=\frac{1}{k^{5/3}}\qquad% \text{for all $k\in\mathbb{N}$.}

Since 5/3>15/3>1, it follows by the pp-series test that k=11k5/3\sum_{k=1}^{\infty}\frac{1}{k^{5/3}} converges. Hence, by the comparison test, k=1k1/3k2+10\sum_{k=1}^{\infty}\frac{k^{1/3}}{k^{2}+10} converges.

(ii) For kk\in\mathbb{N}, observe that

0k2+1k=(k2+1k)(k2+1+k)k2+1+k=k2+1k2k2+1+k1k.0\leq\sqrt{k^{2}+1}-k=\frac{(\sqrt{k^{2}+1}-k)(\sqrt{k^{2}+1}+k)}{\sqrt{k^{2}+% 1}+k}=\frac{k^{2}+1-k^{2}}{\sqrt{k^{2}+1}+k}\leq\frac{1}{k}.

Thus,

0k2+1kk1/1001k1+1/100for all k.0\leq\frac{\sqrt{k^{2}+1}-k}{k^{1/100}}\leq\frac{1}{k^{1+1/100}}\qquad\text{% for all $k\in\mathbb{N}$.}

Since 1+1/100>11+1/100>1, it follows that k=11k1+1/100\sum_{k=1}^{\infty}\frac{1}{k^{1+1/100}} converges by the pp-series test. Hence, k=1k2+1kk1/100\sum_{k=1}^{\infty}\frac{\sqrt{k^{2}+1}-k}{k^{1/100}} converges by the comparison test.

Exercise 3.48

(i) Recall that our goal is to show

(A.9) (A.9) |k=m+1n(1)kak|am+1for all nm with n>m\Big{|}\sum_{k=m+1}^{n}(-1)^{k}a_{k}\Big{|}\leq a_{m+1}\qquad\text{for all $n$% , $m\in\mathbb{N}$ with $n>m$. }

Let mm, nn\in\mathbb{N} with n>mn>m and suppose mm and nn have the same parity: that is, either both are odd or both are even. Then

(1)m+1k=m+1n(1)kak=(am+1am+2)+(am+3am+4)++(an1an).(-1)^{m+1}\sum_{k=m+1}^{n}(-1)^{k}a_{k}=(a_{m+1}-a_{m+2})+(a_{m+3}-a_{m+4})+% \cdots+(a_{n-1}-a_{n}).

Since, by hypothesis, (ak)k(a_{k})_{k\in\mathbb{N}} is nonincreasing, ak1ak0a_{k-1}-a_{k}\geq 0 for all k2k\geq 2. In particular, the above shows that (1)m+1k=m+1n(1)kak(-1)^{m+1}\sum_{k=m+1}^{n}(-1)^{k}a_{k} is a sum of non-negative terms and therefore

(A.10) (A.10) |k=m+1n(1)kak|=(1)m+1k=m+1n(1)kak.\Big{|}\sum_{k=m+1}^{n}(-1)^{k}a_{k}\Big{|}=(-1)^{m+1}\sum_{k=m+1}^{n}(-1)^{k}% a_{k}.

Using (A.10), we now group the terms as

|k=m+1n(1)kak|=am+1+(am+3am+2)++(an1an2)an.\Big{|}\sum_{k=m+1}^{n}(-1)^{k}a_{k}\Big{|}=a_{m+1}+(a_{m+3}-a_{m+2})+\dots+(a% _{n-1}-a_{n-2})-a_{n}.

Arguing as before, ak+1ak0a_{k+1}-a_{k}\leq 0 for all kk\in\mathbb{N}. On the other hand, since (ak)k(a_{k})_{k\in\mathbb{N}} is non-negative, an0-a_{n}\leq 0. Combining these observations, we see that

|k=m+1n(1)kak|am+1+0++0+0=am+1,\Big{|}\sum_{k=m+1}^{n}(-1)^{k}a_{k}\Big{|}\leq a_{m+1}+0+\cdots+0+0=a_{m+1},

which verifies (A.9) in this case.

(ii) Now suppose mm and nn are have different parity. Then

(1)m+1k=m+1n(1)kak=(am+1am+2)+(am+3am+4)++(an2an1)+an(-1)^{m+1}\sum_{k=m+1}^{n}(-1)^{k}a_{k}=(a_{m+1}-a_{m+2})+(a_{m+3}-a_{m+4})+% \cdots+(a_{n-2}-a_{n-1})+a_{n}

is again a sum of non-negative terms. Thus, (A.10) continues to hold in this case. Using (A.10), we now group the terms as

|k=m+1n(1)kak|\displaystyle\Big{|}\sum_{k=m+1}^{n}(-1)^{k}a_{k}\Big{|} =am+1+(am+3am+2)++(anan1)\displaystyle=a_{m+1}+(a_{m+3}-a_{m+2})+\dots+(a_{n}-a_{n-1})
am+1+0++0\displaystyle\leq a_{m+1}+0+\cdots+0
=am+1,\displaystyle=a_{m+1},

where we have again used ak+1ak0a_{k+1}-a_{k}\leq 0 for all kk\in\mathbb{N}. This completes the proof.

Exercise 3.49

(i) Since the sequence (1/k3)k(1/k^{3})_{k\in\mathbb{N}} is nonincreasing, non-negative and 1/k301/k^{3}\to 0 as kk\to\infty, the alternating sign test implies that k=1(1)kk3\sum_{k=1}^{\infty}\frac{(-1)^{k}}{k^{3}} converges.

(ii) Since the sequence (1/log2(k+1))k(1/\log_{2}(k+1))_{k\in\mathbb{N}} is nonincreasing, non-negative and 1/log2(k+1)01/\log_{2}(k+1)\to 0 as kk\to\infty, the alternating sign test implies that k=1(1)klog2(k+1)\sum_{k=1}^{\infty}\frac{(-1)^{k}}{\log_{2}(k+1)} converges.

(iii) Recall that

k+1k=(k+1k)(k+1+k)k+1+k=1k+1+k\sqrt{k+1}-\sqrt{k}=\frac{(\sqrt{k+1}-\sqrt{k})(\sqrt{k+1}+\sqrt{k})}{\sqrt{k+% 1}+\sqrt{k}}=\frac{1}{\sqrt{k+1}+\sqrt{k}}

for all kk\in\mathbb{N}, so that (k+1k)k(\sqrt{k+1}-\sqrt{k})_{k\in\mathbb{N}} is nonincreasing, non-negative and k+1k0\sqrt{k+1}-\sqrt{k}\to 0 as kk\to\infty. Thus, the alternating sign test implies that k=1(1)k(k+1k)\sum_{k=1}^{\infty}(-1)^{k}(\sqrt{k+1}-\sqrt{k}) converges.

(iv) An easy application of the limit laws shows that 2k3k+123\frac{2k}{3k+1}\to\frac{2}{3} as kk\to\infty. It then follows from a result from Worksheet 4 that ((1)k2k3k+1)k\big{(}(-1)^{k}\cdot\frac{2k}{3k+1}\big{)}_{k\in\mathbb{N}}. Explicitly, the subsequence of odd terms tends to 2/3-2/3, whilst the subsequence of even terms tends to 2/32/3; it therefore follows from the subsequence test that ((1)k2k3k+1)k\big{(}(-1)^{k}\cdot\frac{2k}{3k+1}\big{)}_{k\in\mathbb{N}} diverges. Hence, by the kkth term test for convergent series, the series k=1(1)kkk+1\sum_{k=1}^{\infty}(-1)^{k}\cdot\frac{k}{k+1} diverges.

Exercise 3.51

(i) Taking absolute values,

|(1)klog2log2k|=1log2log2kfor k4.\Big{|}\frac{(-1)^{k}}{\log_{2}\log_{2}k}\Big{|}=\frac{1}{\log_{2}\log_{2}k}% \qquad\text{for $k\geq 4$.}

To determine whether the series k=41log2log2k\sum_{k=4}^{\infty}\frac{1}{\log_{2}\log_{2}k} converges, we apply the condensation test. The condensed series is given by

k=12klog2log22k=k=12klog2k.\sum_{k=1}^{\infty}\frac{2^{k}}{\log_{2}\log_{2}2^{k}}=\sum_{k=1}^{\infty}% \frac{2^{k}}{\log_{2}k}.

If we condense further, we arrive at

k=12k22klog22k=k=12k22kk.\sum_{k=1}^{\infty}\frac{2^{k}2^{2^{k}}}{\log_{2}2^{k}}=\sum_{k=1}^{\infty}% \frac{2^{k}2^{2^{k}}}{k}.

Since we know 2kk2^{k}\geq k for all kk\in\mathbb{N}, it is clear that 2k22kk\frac{2^{k}2^{2^{k}}}{k}\to\infty as kk\to\infty and so the series k=12k22kk\sum_{k=1}^{\infty}\frac{2^{k}2^{2^{k}}}{k} diverges by the kkth term test. Thus, the series k=41log2log2k\sum_{k=4}^{\infty}\frac{1}{\log_{2}\log_{2}k} diverges by a two-fold application of the condensation test. In particular, k=4(1)klog2log2k\sum_{k=4}^{\infty}\frac{(-1)^{k}}{\log_{2}\log_{2}k} does not converge absolutely.

On the other hand, the sequence (1/log2log2k)k(1/\log_{2}\log_{2}k)_{k\in\mathbb{N}} is nonincreasing, non-negative and 1/log2log2k01/\log_{2}\log_{2}k\to 0 as kk\to\infty. Thus, the alternating sign test implies that the series k=4(1)klog2log2k\sum_{k=4}^{\infty}\frac{(-1)^{k}}{\log_{2}\log_{2}k} converges.

Since k=4(1)klog2log2k\sum_{k=4}^{\infty}\frac{(-1)^{k}}{\log_{2}\log_{2}k} converges but does not converge absolutely, by definition it converges conditionally.

(ii) Since |1/k3|=1/k3|1/k^{3}|=1/k^{3} for all kk\in\mathbb{N}, by the pp-series test applied to p=3>1p=3>1, we see that k=11k3\sum_{k=1}^{\infty}\frac{1}{k^{3}} converges absolutely.

(iii) We have already see in Exercise 3.49 that k=1(1)k(k+1k)\sum_{k=1}^{\infty}(-1)^{k}(\sqrt{k+1}-\sqrt{k}) converges.

Taking absolute values we have

|(1)k(k+1k)|=k+1k=1k+1+k12k+114k0.|(-1)^{k}(\sqrt{k+1}-\sqrt{k})|=\sqrt{k+1}-\sqrt{k}=\frac{1}{\sqrt{k+1}+\sqrt{% k}}\geq\frac{1}{2\sqrt{k+1}}\geq\frac{1}{4\sqrt{k}}\geq 0.

For the last inequality, we used the fact that k+12k\sqrt{k+1}\leq 2\sqrt{k} for all kk\in\mathbb{N}, which follows since k+14kk+1\leq 4k for all kk\in\mathbb{N}. Since 1/2<11/2<1, by the pp-series test and the limit laws, we know k=114k\sum_{k=1}^{\infty}\frac{1}{4\sqrt{k}} diverges. By comparison, k=1|(1)k(k+1k)|\sum_{k=1}^{\infty}|(-1)^{k}(\sqrt{k+1}-\sqrt{k})| diverges. Thus, k=1(1)k(k+1k)\sum_{k=1}^{\infty}(-1)^{k}(\sqrt{k+1}-\sqrt{k}) does not converge absolutely.

Since k=1(1)k(k+1k)\sum_{k=1}^{\infty}(-1)^{k}(\sqrt{k+1}-\sqrt{k}) converges but does not converge absolutely, by definition it converges conditionally.

(iv) Since, by the limit laws,

2k2k+1=11+2k1as k,\frac{2^{k}}{2^{k}+1}=\frac{1}{1+2^{-k}}\to 1\qquad\text{as $k\to\infty$,}

it follows from Worksheet 4 that (1)k2k2k+1↛0(-1)^{k}\cdot\frac{2^{k}}{2^{k}+1}\not\to 0 as kk\to\infty. Explicitly, the subsequence of odd terms tends to 1-1, whilst the subsequence of even terms tends to 11; it therefore follows from the subsequence test that ((1)k2k2k+1)k\big{(}(-1)^{k}\cdot\frac{2^{k}}{2^{k}+1}\big{)}_{k\in\mathbb{N}} is a divergent sequence. Hence, by the kkth term test for convergent series, k=1(1)k2k2k+1\sum_{k=1}^{\infty}(-1)^{k}\cdot\frac{2^{k}}{2^{k}+1} diverges. It therefore does not converge conditionally or absolutely.

Exercise 3.54

(i) Since |sink|1|\sin k|\leq 1 for all kk\in\mathbb{N}, it follows that

0|sinkk(log2k)2|1klog22kfor all k2.0\leq\Big{|}\frac{\sin k}{k(\log_{2}k)^{2}}\Big{|}\leq\frac{1}{k\log_{2}^{2}k}% \qquad\text{for all $k\geq 2$.}

Furthermore, we know from Exercise 3.41 that k=21k(log2k)2\sum_{k=2}^{\infty}\frac{1}{k(\log_{2}k)^{2}} converges. Hence, by the comparison test, k=2|sinkk(log2k)2|\sum_{k=2}^{\infty}\Big{|}\frac{\sin k}{k(\log_{2}k)^{2}}\Big{|} converges. Thus, k=2sinkk(log2k)2\sum_{k=2}^{\infty}\frac{\sin k}{k(\log_{2}k)^{2}} converges absolutely and therefore converges by the absolute convergence test.

(ii) Observe that

|(5)kk!|=5kk!for all k.\Big{|}\frac{(-5)^{k}}{k!}\Big{|}=\frac{5^{k}}{k!}\qquad\text{for all $k\in% \mathbb{N}$.}

Furthermore, we know from Exercise 3.36 that k=15kk!\sum_{k=1}^{\infty}\frac{5^{k}}{k!} converges. Thus, k=1(5)kk!\sum_{k=1}^{\infty}\frac{(-5)^{k}}{k!} converges absolutely and therefore converges by the absolute convergence test.

(iii) Since |cosk|1|\cos k|\leq 1 for all kk\in\mathbb{N}, it follows that

0|cosk3k+14k33|3k+14k33for all k.0\leq\Big{|}\cos k\cdot\frac{3k+1}{4k^{3}-3}\Big{|}\leq\frac{3k+1}{4k^{3}-3}% \qquad\text{for all $k\in\mathbb{N}$.}

Furthermore, we know from Exercise 3.30 that k=13k+14k33\sum_{k=1}^{\infty}\frac{3k+1}{4k^{3}-3} converges. Hence, by the comparison test, k=1|cosk3k+14k33|\sum_{k=1}^{\infty}\Big{|}\cos k\cdot\frac{3k+1}{4k^{3}-3}\Big{|} converges. Thus, k=1cosk3k+14k33\sum_{k=1}^{\infty}\cos k\cdot\frac{3k+1}{4k^{3}-3} converges absolutely and therefore converges by the absolute convergence test.

Exercise 3.55

Let (ak)n(a_{k})_{n\in\mathbb{N}} be a sequence of real numbers and suppose the series k=1ak\sum_{k=1}^{\infty}a_{k} converges absolutely. Our goal is to show that k=1ak\sum_{k=1}^{\infty}a_{k} converges.

The key idea of this proof is to consider the positive and negative terms of (ak)n(a_{k})_{n\in\mathbb{N}} separately; that is, the non-negative sequences (bk)n(b_{k})_{n\in\mathbb{N}} and (ck)n(c_{k})_{n\in\mathbb{N}} defined by

bk=max(ak,0)andck=max(ak,0),b_{k}=\max(a_{k},0)\quad\text{and}\quad c_{k}=\max(-a_{k},0),

so that ak=bkcka_{k}=b_{k}-c_{k} and |ak|=bk+ck|a_{k}|=b_{k}+c_{k}.

Now, since both 0bk|ak|0\leq b_{k}\leq|a_{k}| and 0ck|ak|0\leq c_{k}\leq|a_{k}|, and by hypothesis the series k=1ak\sum_{k=1}^{\infty}a_{k} converges absolutely, we have

B=k=1bkandC=k=1ckB=\sum_{k=1}^{\infty}b_{k}\quad\text{and}\quad C=\sum_{k=1}^{\infty}c_{k}

are each convergent by comparison with k=1|ak|\sum_{k=1}^{\infty}|a_{k}| (Corollary 3.27). Thus, using the limit laws for series (Theorem 3.21),

BC=k=1(bkck)=k=1akB-C=\sum_{k=1}^{\infty}(b_{k}-c_{k})=\sum_{k=1}^{\infty}a_{k}

is convergent.

Note that this proof took more work than the proof that used the Cauchy criterion, since that is much better suited to working with signed series.

Exercise 3.58

We carry out the same recursive process as used in the example, but this time we keep raising the threshold of the limit.

1. We begin by adding together (positive) even terms of the series only until we pass the threshold 1010.

More precisely, since ene_{n}\to\infty as nn\to\infty, we can find some least value M1M_{1}\in\mathbb{N} such that

k=1M1a2k>10.\sum_{k=1}^{M_{1}}a_{2k}>10.

2. We now include some negative terms to ensure we obtain a rearrangement of the series (if we just included the positive terms, then the resulting σ:\sigma\colon\mathbb{N}\to\mathbb{N} would not be a bijection). We add together (negative) odd terms until we again drop below the threshold 1010. More precisely, since ono_{n}\to-\infty, there exists a least value N1N_{1}\in\mathbb{N} such that

(A.11) (A.11) k=1M1a2k+k=1N1a2k1<10.\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}}a_{2k-1}<10.

3. We again add together (positive) even terms until we again rise above the higher threshold 100100. In particular, we let M2M_{2}\in\mathbb{N} be the least value such that

k=1M1a2k+k=1N1a2k1+k=M1+1M2a2k>100.\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}}a_{2k-1}+\sum_{k=M_{1}+1}^{M_{2}}a_% {2k}>100.

4. Repeating this process, we recursively define increasing sequences of integers M1<M2<M_{1}<M_{2}<\cdots and N1<N2N_{1}<N_{2}\cdots such that

k=1M1a2k+k=1N1a2k1++k=MJ1+1MJa2k>10J\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}}a_{2k-1}+\cdots+\sum_{k=M_{J-1}+1}^% {M_{J}}a_{2k}>10^{J}

and

k=1M1a2k+k=1N1a2k1++k=MJ1+1MJa2k+k=NJ1+1NJa2k1<10J\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}}a_{2k-1}+\cdots+\sum_{k=M_{J-1}+1}^% {M_{J}}a_{2k}+\sum_{k=N_{J-1}+1}^{N_{J}}a_{2k-1}<10^{J}

for each JJ\in\mathbb{N}. This defines a rearrangement of the series. Indeed, it is clear that every term will appear in our reordering of the sequence, and each term appears at most once.

5. It remains to show the rearranged series satisfies k=1naσ(k)\sum_{k=1}^{n}a_{\sigma(k)}\to\infty and nn\to\infty. As in the example, this is quite intuitive and we only provide a sketch.

Our only concern is in step 2 and subsequent steps where we lower the value of the partial sums by adding successive negative terms. We need to see that the value of the sum does not drop down too low in these stages. We illustrate what happens in step 2 only. Recall that we chose N1N_{1} to be the least value such that (A.11) holds. This means that

k=1M1a2k+k=1N11a2k110.\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}-1}a_{2k-1}\geq 10.

Moreover, taking into account the signs of the terms,

10+a2N11k=1M1a2k+k=1N1a2k1k=1M1a2k+k=1na2k1k=1M1a2k10+a_{2N_{1}-1}\leq\sum_{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{N_{1}}a_{2k-1}\leq\sum% _{k=1}^{M_{1}}a_{2k}+\sum_{k=1}^{n}a_{2k-1}\leq\sum_{k=1}^{M_{1}}a_{2k}

for all 1nN11\leq n\leq N_{1}, where we have also used the bound from the previous step. Thus, the amount (in absolute value) by which we drop below the threshold 1010 in step 2 is bounded by a2N11-a_{2N_{1}-1}, which is small.