Week 7 Discussion Notes

Table of Contents

Midterm

I'm just gonna do two problems that I thought were relatively interesting.

Problem 2

Use induction to prove that 2n(n+1)!2^n \leq \p{n+1}! for all n0n \geq 0.

Solution.

Base case: (n=0n = 0)

When n=0n = 0, we get 20=1=1!2^0 = 1 = 1!, so the base case holds.

Inductive step:

Suppose 2n(n+1)!2^n \leq \p{n+1}!. We want to show that 2n+1(n+2)!2^{n+1} \leq \p{n+2}! also. First, notice that since n0n \geq 0, we get n+22n + 2 \geq 2. Thus,

2n+1=22n(n+2)(n+1)!=(n+2)!.2^{n+1} = 2 \cdot 2^n \leq \p{n+2} \cdot \p{n+1}! = \p{n+2}!.

By induction, the inequality holds for all n0n \geq 0.

Problem 4

Let {sn}n\set{s_n}_n be a recursive sequence such that s0=12s_0 = \frac{1}{2} and sn+1=sn3s_{n+1} = s_n^3 for all n0n \geq 0.

  1. Prove that sn[0,1]s_n \in \br{0, 1} for all n0n \geq 0.
  2. Prove that sns_n is decreasing.
  3. Prove that sns_n is convergent and limnsn=0\displaystyle\lim_{n\to\infty} s_n = 0.
Solution.
  1. We'll do this by induction.

    Base case: (n=0n = 0)

    By definition, s0=12[0,1]s_0 = \frac{1}{2} \in \br{0, 1} already, so the base case holds.

    Inductive step:

    Suppose that sn[0,1]s_n \in \br{0, 1}, or equivalently, that 0sn10 \leq s_n \leq 1. Then by cubing everything, we get

    03sn313    0sn+11,0^3 \leq s_n^3 \leq 1^3 \implies 0 \leq s_{n+1} \leq 1,

    since sn+1=sn3s_{n+1} = s_n^3 by definition. Thus, by induction, sn[0,1]s_n \in \br{0, 1} for all n0n \geq 0.

  2. First, notice that

    snsn+1=snsn3=sn(1sn2).s_n - s_{n+1} = s_n - s_n^3 = s_n\p{1 - s_n^2}.

    By (1), we know that sn0s_n \geq 0, and because sn[0,1]s_n \in \br{0, 1}, we also know that sn21s_n^2 \leq 1. Thus, 1sn201 - s_n^2 \geq 0, so snsn+10s_n - s_{n+1} \geq 0 also, i.e., snsn+1s_n \geq s_{n+1}, so sns_n is decreasing.

  3. By (1) and (2), we see that sns_n is a bounded monotone sequence, so it converges to some ss. Thus, by the limit laws, we can take limits on both sides of the recursive relation:

    sn+1=sn3    s=s3    s(1s2)=0.s_{n+1} = s_n^3 \implies s = s^3 \implies s\p{1 - s^2} = 0.

    So, there are three possibilities: s{1,0,1}s \in \set{-1, 0, 1}. Since sn0s_n \geq 0 for all n0n \geq 0, we get s0s \geq 0, so s1s \neq -1. Because sns_n is decreasing, we also know that sns0=12s_n \leq s_0 = \frac{1}{2}, so s12s \leq \frac{1}{2} also. Thus, s=0s = 0.

Subsequences

Definition

Let {sn}nm\set{s_n}_{n \geq m} be a sequence. We say another sequence {tk}k1\set{t_k}_{k \geq 1} is a subsequence of {sn}nm\set{s_n}_{n \geq m} if there exist integers

mn1<n2<m \leq n_1 < n_2 < \cdots

such that tk=snkt_k = s_{n_k}.

Example 1.

If sn=(1)ns_n = \p{-1}^n, then s2k=1s_{2k} = 1 and s2k+1=1s_{2k+1} = -1.

Subsequences are similar to subsets, but with one important distinction: you have to respect the ordering of the original sequence. For example, if sn=ns_n = n, then

1,3,2,4,5,1, 3, 2, 4, 5, \ldots

is not a subsequence of {sn}n\set{s_n}_n.

Theorem (Bolzano-Weierstrass)

Every bounded sequence has a convergent subsequence.

This is a very important idea in analysis, but it's won't be particularly useful for us while we're looking at regular sequences.

Theorem

A sequence {sn}n\set{s_n}_n converges if and only if every subsequence of {sn}n\set{s_n}_n converges to the same limit.

This theorem is useful for us because it gives us another way to prove that sequence diverges. Instead of using the ε\epsilon definition of a limit, we can just find two subsequences which converge to different numbers.

Example 2.

Let sn=(1)n(1+1n)s_n = \p{-1}^n\p{1 + \frac{1}{n}}. Notice that

s2k=1+12kk1ands2k+1=(1+12k+1)k1,s_{2k} = 1 + \frac{1}{2k} \xrightarrow{k\to\infty} 1 \quad\text{and}\quad s_{2k+1} = -\p{1 + \frac{1}{2k+1}} \xrightarrow{k\to\infty} -1,

so {sn}n\set{s_n}_n diverges.

Example 3.

Let sn=cosnπ3s_n = \abs{\cos{\frac{n\pi}{3}}}. Then

s3k=1ands3k+1=12,s_{3k} = 1 \quad\text{and}\quad s_{3k+1} = \frac{1}{2},

so {sn}n\set{s_n}_n must diverge.

Example 4.
(9.9)

This example will be helpful for one of the homework problems this week.

Suppose there exists N0N_0 such that sntns_n \leq t_n for all n>N0n > N_0.

  1. Prove that if limnsn=\displaystyle\lim_{n\to\infty} s_n = \infty, then limntn=\displaystyle\lim_{n\to\infty} t_n = \infty.
  2. Prove that if limnsn\displaystyle\lim_{n\to\infty} s_n and limntn\displaystyle\lim_{n\to\infty} t_n exist, then limnsnlimntn\displaystyle\lim_{n\to\infty} s_n \leq \displaystyle\lim_{n\to\infty} t_n.
Solution.
  1. Let M>0M > 0. Since limnsn=\displaystyle\lim_{n\to\infty} s_n = \infty, there exists N1RN_1 \in \R such that if n>N1n > N_1, then snMs_n \geq M. Thus, if n>max{N0,N1}n > \max\,\set{N_0, N_1}, we get

    Msntn.M \leq s_n \leq t_n.

    Since MM was arbitrary, this proves that limntn=\displaystyle\lim_{n\to\infty} t_n = \infty also.

  2. In Week 4, I proved a fact about non-negative sequences: if {an}n\set{a_n}_n is a sequence such that an0a_n \geq 0 for all n0n \geq 0 and that limnan\displaystyle\lim_{n\to\infty} a_n exists, then limnan0\displaystyle\lim_{n\to\infty} a_n \geq 0.

    If we set an=tnsna_n = t_n - s_n, then an0a_n \geq 0 (technically only for n>N0n > N_0, but this doesn't make a big difference since we're working with limits), then because all the limits in question exist, the proposition and the limit laws tell us

    limntnlimnsn=limn(tnsn)0    limnsnlimntn.\lim_{n\to\infty} t_n - \lim_{n\to\infty} s_n = \lim_{n\to\infty} \p{t_n - s_n} \geq 0 \implies \lim_{n\to\infty} s_n \leq \lim_{n\to\infty} t_n.

    Alternatively, if we didn't have this theorem, we can also prove it using the ε\epsilon definition directly. Let s=limnsns = \displaystyle\lim_{n\to\infty} s_n and t=limntnt = \displaystyle\lim_{n\to\infty} t_n, and suppose for the sake of contradiction that s>ts > t.

    The idea is this: eventually, sns_n should be very close to ss, and tnt_n should be very close to tt. If they're close enough, then we should eventually end up with sn>tns_n > t_n, which contradicts our assumption that sntns_n \leq t_n for n>N0n > N_0.

    Let ε=st2\epsilon = \frac{s - t}{2}, i.e., half the distance between ss and tt. Since s=limnsns = \displaystyle\lim_{n\to\infty} s_n, there exists N1RN_1 \in \R such that if n>N1n > N_1, then

    sns<ε    sns>ε    sn>sε=s(st2)=s+t2.\begin{aligned} \abs{s_n - s} < \epsilon &\implies s_n - s > - \epsilon \\ &\implies s_n > s - \epsilon = s - \p{\frac{s - t}{2}} = \frac{s + t}{2}. \end{aligned}

    Similarly, there exists N2RN_2 \in \R such that if n>N2n > N_2, then

    tnt<ε    tnt<ε    tn<t+ε=t+ts2=s+t2.\begin{aligned} \abs{t_n - t} < \epsilon &\implies t_n - t < \epsilon \\ &\implies t_n < t + \epsilon = t + \frac{t - s}{2} = \frac{s + t}{2}. \end{aligned}

    Thus, if n>max{N0,N1,N2}n > \max\,\set{N_0, N_1, N_2}, then

    tn<n>N2s+t2<n>N1snn>N0tnt_n \underbrace{<}_{n > N_2} \frac{s + t}{2} \underbrace{<}_{n > N_1} s_n \underbrace{\leq}_{n > N_0} t_n

    which is a contradiction.

Example 5.
(basically 11.11)

Let SS be a bounded set. Prove that there is a decreasing sequence {sn}n\set{s_n}_n of points in SS such that limnsn=infS\displaystyle\lim_{n\to\infty} s_n = \inf{S}.

First, if infSS\inf{S} \in S, then we can just set sn=infSs_n = \inf{S} for all n1n \geq 1. Thus, from now on, we will assume that infSS\inf{S} \notin S.

Our strategy will be to construct a sequence that satisfies two properties:

  1. sninfS+1ns_n \leq \inf{S} + \frac{1}{n}.
  2. sn+1sns_{n+1} \leq s_n for all n0n \geq 0.

As before, (1) is here to make sure {sn}n\set{s_n}_n actually converges to infS\inf{S}, and (2) is here to make sure the sequence is decreasing. We can do this by induction:

Base case: (n=1n = 1)

Since infS\inf{S} is the greatest lower bound, this means that infS+1\inf{S} + 1 cannot be a lower bound. Thus, there exists s1Ss_1 \in S such that sinfS+1s \leq \inf{S} + 1.

Inductive step:

Suppose we have s1,,snSs_1, \ldots, s_n \in S such that skinfS+1ks_k \leq \inf{S} + \frac{1}{k} for all 1kn1 \leq k \leq n and so that s1s2sns_1 \geq s_2 \geq \cdots \geq s_n. We need to find sn+1Ss_{n+1} \in S such that sn+1infS+1n+1s_{n+1} \leq \inf{S} + \frac{1}{n+1} and sn+1sns_{n+1} \leq s_n.

As before, we know that infS+1n+1\inf{S} + \frac{1}{n+1} cannot be a lower bound for SS. Similarly, because infSS\inf{S} \notin S, we get infS<sn\inf{S} < s_n, so sns_n is not a lower bound for SS either. Hence if we set ε=min{1n+1,sninfS}\epsilon = \min\,\set{\frac{1}{n+1}, s_n - \inf{S}}, then ε>0\epsilon > 0, so infS+ε\inf{S} + \epsilon is not a lower bound for SS. By definition, there exists sn+1Ss_{n+1} \in S such that sn+1infS+εs_{n+1} \leq \inf{S} + \epsilon. Thus,

ε1n+1    sn+1infS+1n+1εsninfS    sn+1infS+(sninfS)=sn.\begin{aligned} \epsilon \leq \frac{1}{n+1} &\implies s_{n+1} \leq \inf{S} + \frac{1}{n+1} \\ \epsilon \leq s_n - \inf{S} &\implies s_{n+1} \leq \inf{S} + \p{s_n - \inf{S}} = s_n. \end{aligned}

By induction, we have a decreasing sequence {sn}n\set{s_n}_n of elements in SS such that sninfS+1ns_n \leq \inf{S} + \frac{1}{n}. Since infS\inf{S} is a lower bound for SS, we have

infSsninfS+1n\inf{S} \leq s_n \leq \inf{S} + \frac{1}{n}

for all n1n \geq 1. By the squeeze theorem, we get limnsn=infS\displaystyle\lim_{n\to\infty} s_n = \inf{S}.