Skip to Content
Math4111Introduction to Real Analysis (Lecture 22)

Lecture 22

Review

Let (an)(a_n) be a sequence, and let bn=af(n)b_n = a_{f(n)} be a rearrangement, where ff is given by the following:

nn123456789\dots
f(n)f(n)12436851012\dots

(The pattern is “odd,even,even,”) Defined the partial sums sn=k=1naks_n = \sum_{k=1}^n a_k and tn=k=1nbkt_n = \sum_{k=1}^n b_k.

  1. In terms of a1,a2,a_1,a_2,\ldots, determine sntns_n-t_n for n=1,2,3,4,5,6,7n=1,2,3,4,5,6,7. (For example, s3t3=a3a4s_3-t_3 = a_3-a_4.)
    s1t1=a1a1=0s_1 - t_1 = a_1 - a_1 = 0
    s2t2=a2a2=0s_2 - t_2 = a_2 - a_2 = 0
    s3t3=a3a4s_3 - t_3 = a_3 - a_4
    s4t4=a4a4=0s_4 - t_4 = a_4 - a_4 = 0
    s5t5=a5a6s_5 - t_5 = a_5 - a_6
    s6t6=a6a8s_6 - t_6 = a_6 - a_8
    s7t7=a7a8s_7 - t_7 = a_7 - a_8

  2. What is the smallest nn so that sntns_n - t_n does not contain any of the terms a1,,a5a_1,\dots, a_5?
    n=7n=7

  3. What is the smallest nn so that sntns_n - t_n does not contain any of the terms a1,,a10a_1,\dots, a_{10}?
    n=13n=13

New Material

Absolute Convergence

Theorem 3.55

Let (an)(a_n) be a sequence in C\mathbb{C} such that n=1an\sum_{n=1}^\infty |a_n| converges. If (bn)(b_n) is a rearrangement of (an)(a_n), then n=1an=n=1bn\sum_{n=1}^\infty a_n = \sum_{n=1}^\infty b_n.

Proof:

Let f:NNf:\mathbb{N}\to\mathbb{N} be a bijection and let bn=af(n)b_n = a_{f(n)}.

Let I(n)={1,2,,n}I(n)=\{1,2,\dots,n\}, J(n)={f(1),f(2),,f(n)}J(n)=\{f(1),f(2),\dots,f(n)\}.

Then sn=kI(n)aks_n = \sum_{k\in I(n)} a_k and tn=kJ(n)akt_n = \sum_{k\in J(n)} a_k.

sntn=k=1nakk=1nbk=kI(n)akkJ(n)af(k)=kI(n)\J(n)akkJ(n)\I(n)af(k)sntni(I(n)\J(n))(J(n)\I(n))ai\begin{aligned} s_n - t_n &= \sum_{k=1}^n a_k - \sum_{k=1}^n b_k \\ &= \sum_{k\in I(n)} a_k - \sum_{k\in J(n)} a_{f(k)} \\ &= \sum_{k\in I(n)\backslash J(n)} a_k - \sum_{k\in J(n)\backslash I(n)} a_{f(k)} \\ |s_n - t_n|&\leq \sum_{i\in (I(n)\backslash J(n))\cup(J(n)\backslash I(n))} |a_i| \end{aligned}

We will show that limnsntn=0\lim_{n\to\infty} |s_n - t_n| = 0.

Let ϵ>0\epsilon > 0.

By the Cauchy criterion applied to n=1an\sum_{n=1}^\infty |a_n|, there exists NN such that if m,nNm,n\geq N, then k=m+1nak<ϵ\sum_{k=m+1}^n |a_k| < \epsilon.

Then we choose pNp\in\mathbb{N} such that I(n)J(p)I(n)\subset J(p) ({1,2,,N}{f(1),f(2),,f(p)}\{1,2,\dots,N\}\subset\{f(1),f(2),\dots,f(p)\}). p=max{f1(1),f1(2),,f1(N)}p=\max\{f^{-1}(1),f^{-1}(2),\dots,f^{-1}(N)\}.

Note: This implies that pp is at least NN.

If npn\geq p, then I(n)J(p)I(n)J(n)I(n)\subset J(p)\subset I(n)\cap J(n) so sn=tns_n = t_n.

So,

sntni=N+1maxJ(n)ai<ϵ|s_n - t_n| \leq \sum_{i=N+1}^{\max J(n)} |a_i| < \epsilon

Back to the example of the review question.

I(9)={1,2,,9}I(9)=\{1,2,\dots,9\}, J(9)={1,2,4,3,6,8,5,10,12}J(9)=\{1,2,4,3,6,8,5,10,12\}, I(9)\J(9)={7,9}I(9)\backslash J(9)=\{7,9\}, J(9)\I(9)={10,12}J(9)\backslash I(9)=\{10,12\}.

s9t9a7+a9+a10+a12k=712ak<ϵ|s_9 - t_9| \leq |a_7|+|a_9|+|a_{10}|+|a_{12}| \leq \sum_{k=7}^{12} |a_k| < \epsilon

This proves that limnsntn=0\lim_{n\to\infty} |s_n - t_n| = 0.

Since limnsn\lim_{n\to\infty} s_n exists, limnsn=limntn\lim_{n\to\infty} s_n = \lim_{n\to\infty} t_n.

QED

Theorem 3.54

Special case of Riemann rearrangement theorem

Suppose anRa_n\in \mathbb{R}, n=1an\sum_{n=1}^\infty a_n converges conditionally. (i.e. n=1an\sum_{n=1}^\infty a_n converges but n=1an\sum_{n=1}^\infty |a_n| diverges.) Then αR\forall \alpha\in\mathbb{R}, there exists a rearrangement (bn)(b_n) of (an)(a_n) such that n=1bn=α\sum_{n=1}^\infty b_n = \alpha.

Chapter 4 Continuity

Limits of Functions

Definition 4.1

Let (X,dX)(X,d_X) and (Y,dY)(Y,d_Y) be metric spaces and EXE\subset X, pEp\in E', qYq\in Y. Let f:EYf:E\to Y. We say that limxpf(x)=q\lim_{x\to p} f(x) = q if ϵ>0\forall \epsilon > 0, δ>0\exists \delta > 0 such that xE\forall x\in E, if 0<dX(x,p)<δ0 < d_X(x,p) < \delta, then dY(f(x),q)<ϵd_Y(f(x),q) < \epsilon.

This is same as:

If limxpf(x)=q\lim_{x\to p} f(x)=q, then

ϵ>0,δ>0,xE,0<dX(x,p)<δ    dY(f(x),q)<ϵ\forall \epsilon > 0, \exists \delta > 0, \forall x\in E, 0 < d_X(x,p) < \delta \implies d_Y(f(x),q) < \epsilon

In many definitions, E=XE=X

Theorem 4.2

limxpf(x)=q    \lim_{x\to p} f(x) = q \iff forall sequence (pn)(p_n) in E\{p}E\backslash\{p\} with pnpp_n\to p, f(pn)qf(p_n)\to q.

Proof:

    \implies

Suppose limxpf(x)=q\lim_{x\to p} f(x) = q.

Let (pn)(p_n) be a sequence in E\{p}E\backslash\{p\} with pnpp_n\to p.

Let ϵ>0\epsilon > 0.

By definition of limit of function, δ>0\exists \delta > 0 such that if 0<dX(x,p)<δ0 < d_X(x,p) < \delta, then dY(f(x),q)<ϵd_Y(f(x),q) < \epsilon.

Since pnpp_n\to p, N\exists N such that if nNn\geq N, then 0<dX(pn,p)<δ0 < d_X(p_n,p) < \delta. So f(pn)Bϵ(q)f(p_n)\in B_\epsilon(q).

Thus, we showed ϵ>0\forall \epsilon > 0, N\exists N such that if nNn\geq N, then f(pn)Bϵ(q)f(p_n)\in B_\epsilon(q).

    \impliedby

We proceed by contrapositive.

Suppose limxpf(x)q\lim_{x\to p} f(x) \neq q, then \exists sequence (pn)(p_n) in E\{p}E\backslash\{p\} with pnpp_n\to p such that f(pn)qf(p_n)\cancel{\to} q.

Suppose limnf(pn)q\lim_{n\to\infty} f(p_n) \neq q, then ϵ>0\exists \epsilon > 0 such that for all δ>0\delta > 0, there exists xEBδ(p)\{p}x\in E\cap B_\delta(p)\backslash\{p\} such that f(x)Bϵ(q)f(x)\notin B_\epsilon(q).

For nNn\in\mathbb{N}, if we apply this with δ=1n\delta = \frac{1}{n}, we get pnEB1n(p)\{p}p_n\in E\cap B_{\frac{1}{n}}(p)\backslash\{p\} such that f(pn)Bϵ(q)f(p_n)\notin B_\epsilon(q).

Then: (pn)(p_n) is a sequence in E\{p}E\backslash\{p\} with dX(pn,p)=1n0d_X(p_n,p) = \frac{1}{n}\to 0 so that as nn\to\infty, f(pn)Bϵ(q)f(p_n)\notin B_\epsilon(q).

So limnf(pn)q\lim_{n\to\infty} f(p_n) \neq q.

QED

With this theorem, we can use the properties of limit of sequences to study limits of functions.

To prove things about limits of functions, we can use sequences.

  • If limxpf(x)=q\lim_{x\to p} f(x) = q, and limxpf(x)=q\lim_{x\to p} f(x)=q', then q=qq=q'.
  • If limxpf(x)=A\lim_{x\to p} f(x) = A and limxpg(x)=B\lim_{x\to p} g(x) = B, then limxpf(x)+g(x)=A+B\lim_{x\to p} f(x) + g(x) = A+B.

Continuous functions

Definition 4.5

Let (X,dX)(X,d_X) and (Y,dY)(Y,d_Y) be metric spaces and EXE\subset X, pEp\in E. Let f:EYf:E\to Y. We say that ff is continuous at pp if ϵ>0\forall \epsilon > 0, δ>0\exists \delta > 0 such that f(EBδ(p))Bϵ(f(p))f(E\cap B_\delta(p))\subset B_\epsilon(f(p)).

  • We say that ff is continuous on EE if ff is continuous at every pEp\in E.

Remark: For pEp\in E, there are two possibilities.

  • pp is an isolated point of EE.
  • pp is a limit point of EE.

In the first case, ff is automatically continuous at pp. (EBδ(p)={p}E\cap B_\delta(p)=\{p\} for all δ>0\delta > 0.)

In the second case, ff is continuous at pp if and only if limxpf(x)=f(p)\lim_{x\to p} f(x) = f(p).

Theorem 4.8

A function f:EYf:E\to Y is continuous at pEp\in E if the pre-image of every open set is open.

Two consequences if f:EYf:E\to Y is continuous:

  • Image of compact set is compact. (Implies Extreme Value Theorem)
  • Image of connected set is connected. (Implies Intermediate Value Theorem)
Last updated on