2025-09-15

Math Analysis(Class Note)

Math analysis

Class 1 Some Inequality and Def of limit

Triangle Inequality

a,bR    aba+ba+ba,b\in R \implies {\left \vert {\left \vert a \right \vert} -{\left \vert b \right \vert} \right \vert} \le {\left \vert a+b \right \vert} \leq {\left \vert a \right \vert} + {\left \vert b \right \vert}
ababab    a22ab+a+b2a2+2ab+b2a2+2ab+a+b2    (ab)2(a+b)2(a+b)2\begin{gathered} -{\left \vert a \right \vert} {\left \vert b \right \vert} \le ab\le {\left \vert a \right \vert} {\left \vert b \right \vert} \implies \\ {\left \vert a \right \vert} ^2-2 {\left \vert a \right \vert} {\left \vert b \right \vert} + {\left \vert a+b \right \vert} ^2 \le a^2+2ab+b^2 \le {\left \vert a \right \vert} ^2+2 {\left \vert a \right \vert} {\left \vert b \right \vert} + {\left \vert a+b \right \vert} ^2 \implies \\ ({\left \vert a \right \vert} -{\left \vert b \right \vert} )^2\le (a+b)^2\le ({\left \vert a \right \vert} +{\left \vert b \right \vert} )^2 \end{gathered}

向量的模长也满足 也可以用这个做法.

Cauchy-Schwarz Inequality

0<xiR    ixinixinni1xi\begin{gathered} 0<x_i \in R \implies \dfrac{\sum_i x_i}{n} \ge \sqrt[n]{\prod_i x_i} \ge \dfrac{n}{\sum_i \dfrac{1}{x_i} } \end{gathered}

显然 取xi=1xix_i=\dfrac{1}{x_i}能用左边不等式推右边 只证左边

归纳,22的幂次显然吧

对非22的幂次nn,用X=ixinX=\sqrt[n]{\prod_i x_i}补完,直接套

Array's Limit

limnan=A    ϵ>0N s.t. nN    anA<ϵ\begin{gathered} lim_{n\to \infty} a_n = A \iff \\ \forall \epsilon >0 \exists N \ s.t.\ n\ge N \implies \left \vert a_n-A \right \vert < \epsilon \end{gathered}

注意

  • N=N(ϵ)N=N(\epsilon)
  • NN 不唯一
  • ϵ\epsilon 可限制在任意(0,a),a>0(0,a),a>0

Eg 1

limnn1n=1\lim_{n \to \infty} n^{\frac{1}{n} }=1

n1n1=(nn)1n1=(i1nn)1n1n2+2nn12n\begin{gathered} \vert n^{\frac{1}{n} }-1 \vert \\ = \vert (\sqrt n \sqrt n)^{\frac{1}{n} } -1 \vert \\ ={\left \vert (\prod_i 1 \cdot \sqrt n\sqrt n)^{\frac{1}{n} } -1 \right \vert} \\ \le {\left \vert \dfrac{n-2+2\sqrt n}{n} -1 \right \vert} \\ \le \dfrac{2}{\sqrt n} \end{gathered}

Or try this:

n1n1=n1n1<ϵ    n1n<(ϵ+1)    n<(1+ϵ)n    n<1+ϵn+ϵ2n(n1)2\begin{gathered} {\left \vert n^{\frac{1}{n} }-1 \right \vert} =n^{\frac{1}{n} }-1<\epsilon \\ \impliedby n^{\frac{1}{n}}<(\epsilon+1) \\ \impliedby n<(1+\epsilon)^n \\ \impliedby n<1+\epsilon n+\frac{\epsilon^2n(n-1)}{2} \end{gathered}

Solve the equation, it's a parabola with upward opening so the solution exists.

[think] At this stage, we cannot say limg(f(n))=g(limf(n))\lim g(f(n))=g(\lim f(n)) which depends on continuity of function. But you sometimes can rewrite the proof with inequality.

Class 2

About QQ

nZQC\begin{gathered} \sqrt n \in Z \cup Q^C \end{gathered}

p2=q2np^2=q^2n Integer factorization

More natural than textbook but it doesn't depend on Integer factorization.

pq\dfrac{p}{q} is finite decimal or repeating decimal

Simulate how the division is done and consider the remainder will be repeated.

About Cardinality

A Example

Find a bijection of [0,1][0,1] and (0,1)(0,1)

f(x):(0,1)[0,1]={0,x=121,x=131n2,x=1nx,otherwise\begin{gathered} f(x):(0,1) \to [0,1]=\begin{cases} 0,x=\dfrac{1}{2} \\ 1,x=\dfrac{1}{3} \\ \dfrac{1}{n-2} , x=\dfrac{1}{n} \\ x, \text{otherwise} \end{cases} \end{gathered}

Wow!

Cardinality is comparable

A,B, f(x):ABs.t.f is injective or surjective\begin{gathered} \forall A,B, {\ } \exists f(x):A\to B \\ s.t.\\ \text{f is injective or surjective} \end{gathered}

(承认选择公理)

首先选择公理出良序定理,找到A,BA,B的良序(A,<),(B,<)(A,<),(B,<).

然后进行超限归纳,把AA,BB按照良序(保证了每个元素存在序里的后继),最小元素匹配,然后剩下的次小元素匹配,这么做下去.

然后这个看起来进行的是普通的自然数归纳显然是错的. 你需要用超限归纳也就是在序上做归纳.

然后问题来到序是什么.

序的结构是这样的 首先是自然数 自然数定义是00定义为空集开始 然后定义succ(S)=S{S}\operatorname{succ}(S)=S \cup \{S\}

然后定义完所有自然数后 定义ω=ii\omega=\bigcup_i i,就是把所有的自然数的集合并起来(显然它包含所有的自然数). 然后我们可以接下来用后继的定于去定义ω+1,ω+2\omega+1,\omega+2\ldots,并且你又可以把它们并起来得到ω×2\omega\times 2,不断走后继,ω,2ω,3ω\omega,2\omega,3\omega\ldots可以变成ω2\omega^2,又有ω3ωω\omega^3\ldots \omega^\omega等等

基本上是每个层次的运算完了之后进下一个层次的序数构造 总之它看起来包含了各种各样的无穷,可以应付所有大小的集合.

然后我们要在序数的结构上做归纳法,就要证明:xsucc(x)x\to \operatorname{succ}(x),还要证明极限这一把也对,就是a1aniaia_1\ldots a_n \to \cup_i a_i这个操作(也就是对某个极限序数,如果所有它以前的序数推到它自己)合法,就满足了你可以不断到下一极限.这样推出对全体元素合法. 这就是超限归纳法.

那你对照一下我们的归纳就是一一对应啊,所以是合法的. 就结束了.

然后我们刚才是说明了可以借良序去给两个集合配对,那么你就一定能找到一个对另一个的单射,所以一定可比.

Cantor-Bernstein-Schröder Theorem

Card(A)Card(B),Card(B)Card(A)    Card(A)=Card(B)\operatorname{Card}(A)\le \operatorname{Card}(B),\operatorname{Card}(B)\le \operatorname{Card}(A) \implies \operatorname{Card}(A)=\operatorname{Card}(B)

或者表达为

{A,B,g:AB,f:BAf,g is injective    h:AB is bijective\begin{gathered} \begin{cases} \forall A,B,g:A\to B,f:B\to A\\ f,g \text{ is injective} \end{cases} \implies \exists h:A \leftrightarrow B \text{ is bijective} \end{gathered}

考虑从任意元素uAu\in A可以引出一条链:uf(u)g(f(u)f(g(f(u)))u\to f(u)\to g(f(u)\to f(g(f(u)))\to \ldots. 同理vBv \in B开始的链vg(v)f(g(v))g(f(g(v)))v\to g(v)\to f(g(v))\to g(f(g(v)))\to \ldots.

注意到所有元素一定都在某条链上(映射).每个点度数一定至多一进一出(单射).

于是每条链上构造一个双射(显然的)拼起来即可.

Q.E.D\text{Q.E.D}

其实这些应该算集合论还是什么?

Class 3 Properties of Limit

  • Uniqueness:Obviously
  • Local boundedness: 取ϵ=1\epsilon=1NN,NN后面显然,前面有限项也显然.
  • About subsequence:
    • about any subsequence
    • about some subsequence whose union is the sequence

Limitaion's Calculation

limnan=A,limnbn=B\lim_{n \to \infty} a_n=A,\lim_{n \to \infty} b_n=B
limnanbn=AB\lim_{n \to \infty} a_nb_n=AB
anbnAB=anbnanB+anBABaNbnB+BanA\begin{gathered} {\left \vert a_nb_n-AB \right \vert} = \\ {\left \vert a_nb_n-a_nB+a_nB-AB \right \vert} \\ \\ \le {\left \vert a_N \right \vert} {\left \vert b_n-B \right \vert} +{\left \vert B \right \vert} {\left \vert a_n-A \right \vert} \\ \end{gathered}

显然an,bna_n,b_n有界,令MM是他俩共同的界,取ϵ=1114514M+114514\epsilon'=\dfrac{1}{114514M+114514}用到an,bna_n,b_n上即证.

limnanbn=AB\lim_{n \to \infty} \frac{a_n}{b_n} =\frac{A}{B}

先上一个保号性干掉分母上出现00的事.

证取倒数:

1bn1B<ϵ    Bbn<ϵbnB    ϵ=ϵBM+114514(M>bn)Q.E.D\begin{gathered} {\left \vert \frac{1}{b_n}-\frac{1}{B} \right \vert}<\epsilon \\ \impliedby {\left \vert B-b_n \right \vert} <\epsilon b_nB \\ \impliedby \epsilon'=\frac{\epsilon}{BM+114514} (\vert M\vert>b_n) \end{gathered} \\ \text{Q.E.D}

那转化到乘法是显然的.

无穷小

基本就是收敛到00的数列啊.

Eg

{limnan=alimnbn=b    limniaibnin=ab\begin{cases} \lim_{n \to \infty} a_n=a \\ \lim_{n \to \infty} b_n=b \end{cases} \\ \implies \lim_{n \to \infty} \frac{\sum_i a_ib_{n-i}}{n}=ab

Solution 1

b=0b=0时,考虑a有界Ma有界M.

limniaibninlimnMbnin=0\begin{gathered} \lim_{n \to \infty} \dfrac{\sum_i a_ib_{n-i}}{n} \\ \le \lim_{n \to \infty} M\dfrac{b_{n-i}}{n} =0 \end{gathered}

b0b\ne 0,bi=bibb'_i=b_i-b,有

limniaibnin=limniaibnin+iainb=0+ab\begin{gathered} \lim_{n \to \infty} \frac{\sum_i a_ib_{n-i}}{n} =\lim_{n \to \infty} \frac{\sum_i a_ib'_{n-i}}{n} +\frac{\sum_i a_i}{n} b=0+ab \end{gathered} \begin{gathered} \end{gathered}

Solution 2

不妨设a,b>0a,b>0,则有界性可以得到后面是an,bn>0a_n,b_n>0

考虑影响值的肯定是中间的项,所以直接拆,取ϵ\epsilon,对a,ba,b可以得到N1N_1

limniaibnin==limni=1N1aibni+i=1N1biani+i=N1+1nN11aibnin=limni=1N1aibnin+limni=1N1bianin+limni=N1+1nN11aibnin(limnn2N11n(aϵ)(bϵ),limnn2N11n(a+ϵ)(b+ϵ))=((aϵ)(bϵ),(a+ϵ)(b+ϵ))\begin{gathered} \lim_{n \to \infty} \dfrac{\sum_i a_ib_{n-i}}{n}= \\ =\lim_{n \to \infty} \dfrac{\sum _{i = 1} ^{N_1} a_ib_{n-i}+\sum _{i = 1} ^{N_1} b_ia_{n-i}+\sum _{i = N_1+1} ^{n-N_1-1} a_ib_{n-i} }{n} \\ =\lim_{n \to \infty} \dfrac{\sum _{i = 1} ^{N_1} a_ib_{n-i}}{n} \\ + \lim_{n \to \infty} \dfrac{\sum _{i = 1} ^{N_1} b_ia_{n-i}}{n} \\ +\lim_{n \to \infty} \dfrac{\sum _{i = N_1+1} ^{n-N_1-1} a_ib_{n-i}}{n} \\ \in (\lim_{n \to \infty} \dfrac{n-2N_1-1}{n} (a-\epsilon)(b-\epsilon) \\ ,\lim_{n \to \infty} \dfrac{n-2N_1-1}{n} (a+\epsilon)(b+\epsilon)) \\ = ((a-\epsilon)(b-\epsilon),(a+\epsilon)(b+\epsilon)) \end{gathered}

后面显然.`

Class 4

eg1

a>1,kN,limnnkan=0\begin{gathered} a>1,k\in N^*,\lim_{n \to \infty} \frac{n^k}{a^n} =0 \end{gathered}

an=(1+b)na^n=(1+b)^n展开会出现nn的任意次方,然后显然.

Stolz

{yn,limnyn=limnxnxn1ynyn1=a[,+]    limnxnyn=a\begin{gathered} \begin{cases} y_n \uparrow,\lim_{n \to \infty} y_n=\infty \\ \lim_{n \to \infty} \dfrac{x_n-x_{n-1}}{y_n-y_{n-1}} =a\in[-\infty,+\infty] \end{cases} \\ \implies \lim_{n \to \infty} \dfrac{x_n}{y_n} =a \end{gathered}

先简化,不妨设a0,ΔxnΔyn>0a\ge 0,\frac{\Delta x_n}{\Delta y_n}>0

然后可以证a=0a=0时,你取ϵ1\epsilon_1N1N_1,变成 ΔxnΔyn<ϵ\dfrac{\Delta x_n}{\Delta y_n} <\epsilon.于是

xnyn=xN1+i=N1+1nΔxiyN1+i=N1+1nΔyixN1+ϵi=N1+1nΔyiyN1+i=N1+1nΔyixN1+ϵi=N1+1nΔyii=N1+1nΔyi\begin{gathered} \dfrac{x_n}{y_n} \\ =\dfrac{x_{N_1}+\sum _{i = N_1+1} ^{n} \Delta x_i}{y_{N_1}+\sum _{i = N_1+1} ^{n} \Delta y_i} \\ \le \dfrac{x_{N_1}+\epsilon\sum _{i = N_1+1} ^{n} \Delta y_i}{y_{N_1}+\sum _{i = N_1+1} ^{n} \Delta y_i} \\ \le \dfrac{x_{N_1}+\epsilon\sum _{i = N_1+1} ^{n} \Delta y_i}{\sum _{i = N_1+1} ^{n} \Delta y_i} \end{gathered}

因为你 Y=i=N1+1nΔyi+Y=\sum _{i = N_1+1} ^{n} \Delta y_i \to +\infty,所以一定可以有ϵY>xN1\epsilon Y>x_{N_1}.就能证<2ϵ<2\epsilon,你再取一下ϵ\epsilon就得证了.

然后我说0<aR0<a\in R时你直接x=xayx'=x-ay就用结论,a=+a=+\infty的时候你取倒数用结论,就做完了.

Class 5

单调有界数列有极限

{ai>ai1ai<M    limnai exists\begin{gathered} \begin{cases} a_i>a_{i-1} \\ a_i<M \end{cases} \implies \lim_{n \to \infty} a_i \text{ exists} \end{gathered}

{ai}\{a_i\}上确界M0M_0,则任意ϵ\epsilon,取M=M0ϵM'=M_0-\epsilon,由确界知i s.t. ai(M,M0]\exists i \ s.t.\ a_i\in (M',M_0],则n>i,anM<ϵ\forall n>i,\vert a_n-M\vert<\epsilon.

然后书上因为没教你上确界,试图用无限小数说明.那其实就是你一位一位考虑,就先这一位增加到最大,然后进入下一位,容易发现最后区间长度趋近于00

其实无限小数就是区间套啊.

e

an:=limn(1+1n)n=bn:=limni=0n1i!\begin{gathered} a_n:=\lim_{n \to \infty} (1+\dfrac{1}{n} )^n = b_n:=\lim_{n \to \infty} \sum _{i = 0} ^{n} \dfrac{1}{i!} \end{gathered}

右边那个单调有界都是好证的.证相等:

an=(1+1n)n=i=0n1ni(ni)=i=0n1i!j=0i1(1jn)<bn\begin{gathered} a_n=(1+\dfrac{1}{n} )^n=\sum _{i = 0} ^{n} \dfrac{1}{n^i} \binom{n}{i} \\ =\sum _{i = 0} ^{n} \dfrac{1}{i!} \prod_{j=0}^{i-1} (1-\dfrac{j}{n} ) <b_n \\ \end{gathered}

又有你考虑级数aa展开的某个部分和i=0k1i!j=0i1(1jn)\sum _{i = 0} ^{k} \dfrac{1}{i!} \prod_{j=0}^{i-1} (1-\dfrac{j}{n} ),当nn趋近于无穷大的时候这个部分和趋近于bkb_k,于是会大于bk1b_{k-1},也就是说你对每个bkb_k都能找打比他大的aa,就解决了.

重点是把一次到极限拆成两个取极限.但这个为什么对呢.

是好证的,其实就是

[think]

limalimbf(a,b)=X    limnf(n,n)=X\begin{gathered} \lim_{a \to \infty} \lim_{b \to \infty} f(a,b) = X \implies \lim_{n \to \infty} f(n,n) = X \end{gathered}

Fixed:这是不一定的啊,当我们学了多元函数,需要一致收敛之类的条件啊,但其实这里也用不到这个内容啊,我们实际上是用到的是对函数中的无关变量取极限就没有影响.

0<ei=0n1i!<1nn!\begin{gathered} 0<e-\sum _{i = 0} ^{n} \dfrac{1}{i!} < \dfrac{1}{n\cdot n!} \end{gathered}
bn:=i=0n1i!X=bn+mbn=i=n+1n+m1i!<1(n+1)!i=0m11(n+1)i<1n!n    ebn=limmX<1n!n\begin{gathered} b_n:=\sum _{i = 0} ^{n} \dfrac{1}{i!} \\ X=b_{n+m}-b_n=\sum _{i = n+1} ^{n+m} \dfrac{1}{i!} \\ <\dfrac{1}{(n+1)!} \sum _{i = 0} ^{m-1} \dfrac{1}{(n+1)^i}<\dfrac{1}{n!\cdot n} \\ \implies e-b_n= \lim_{m \to \infty} X<\dfrac{1}{n!\cdot n} \end{gathered}

然后说为什么取极限小于号还是小于号呢?

X=bn+mbn<bn+m+kbn<1n!nlimkebn<1n!n\begin{gathered} X=b_{n+m}-b_n<b_{n+m+k}-b_n<\dfrac{1}{n!\cdot n} \\ \stackrel{\lim_{k \to \infty} }{\Longrightarrow} e-b_n<\dfrac{1}{n!\cdot n} \end{gathered}

[think] 对单调数列让一个独立变量趋近无穷说明严格不等号.

有个魔怔法,你先去下一条证明ee是无理数再回来说取不了等.

e∉Qe\not \in Q

假设e=pqe=\dfrac{p}{q}

pqbn<1nn!n=qp(q1)!bqq!<1q\begin{gathered} \dfrac{p}{q} -b_n<\dfrac{1}{n\cdot n!} \stackrel{n=q}{\Longrightarrow} p(q-1)!-b_q q!<\dfrac{1}{q} \end{gathered}

左边是不为00的整数(为00bb增矛盾),右边是分数

[think] 一个无理数等价于存在一列分母到无穷的有理数始终有更高阶(相比序列中分母)的余项.

(1+1n)n(1+1n)n+1\begin{gathered} (1+\dfrac{1}{n} )^n \uparrow \\ (1+\dfrac{1}{n} )^{n+1} \downarrow \end{gathered}
(1+1n)n1<(n(1+1n+1)n+1)n+1=(1+1n+1)n+1\begin{gathered} (1+\dfrac{1}{n} )^n\cdot 1< (\dfrac{n\cdot (1+\dfrac{1}{n}+1 )}{n+1} )^{n+1}=(1+\dfrac{1}{n+1} )^{n+1} \end{gathered}

第二个取倒数同理.

从这个出发可以说明ln\ln的切线放缩系列不等式.

有界数列必有单调子列

有界数列有收敛子列

可以考虑后缀max,取出一个单调子列.

也可以区间套,每次进有无穷多项的区间.

Class 6

Cauchy Convergance Theorem

Cauchy Convergance Theorem

ϵ,Ns.t.n,m>N,anam<ϵ    limnan exists\begin{gathered} \forall \epsilon, \exists N \\ s.t.\\ \forall n,m>N, \vert a_n-a_m\vert<\epsilon \iff \lim_{n \to \infty} a_n \text{ exists} \end{gathered}

右推左是显然的.

首先容易得到有界.于是它有收敛子列.

然后你其他的项到你的收敛子列的距离拿柯西的条件放缩一下就证完了.

或者也可以闭区间套.

确界原理

有上界的数集一定有上确界.

来闭区间套,二等分,如果上面的(包含边界)有就取上面,否则取下面,框出一个数.

然后来看,比他小的不是上界是好说的(取个区间即可).怎么说明它是上界呢?

考虑任何一个数,递归后一定有某一次它在下半区间(包含中点),那就证完了.

可能甚至不如无穷小数简洁 反正本质相同.

有限覆盖定理

S={(x,y)x<y},TS,ITI[a,b]    AT,IAI[a,b],AN(not infinity)\begin{gathered} S=\{ (x,y) \vert x<y \},T\subset S , \cup_{I\in T} I \supset [a,b] \\ \implies \exists A \in T,\cup_{I\in A} I \supset [a,b],\vert A\vert\in N(\text{not infinity} ) \end{gathered}

反证,设[A,B][A,B]不能被有限覆盖.

闭区间套 二等分 则一定有一半区间也是不能被有限覆盖的.递归到不能被有限覆盖的区间,最后弄出一个数.

但包含这个数的极小区间显然可以被有限覆盖.矛盾,得证.

很棒的啊,它完全不关心你无穷覆盖的结构而是到数的结构去了.

Class 7

Funciton's Limits

limxx0f(x)=A    ϵ,δ,s.t.x,xx0(0,δ),f(x)A<ϵ\begin{gathered} \lim_{x \to x_0} f(x) = A \\ \iff \forall \epsilon, \exists \delta,\\ s.t.\\ \forall x, \vert x-x_0 \vert \in (0,\delta), \vert f(x)-A \vert <\epsilon \end{gathered}

Heine Theorem

limxx0f(x)=A    {xn},limnxn=x0limnf(xn)=A\begin{gathered} \lim_{x \to x_0} f(x)=A \\ \iff \forall \{ x_n \} ,\lim_{n \to \infty} x_n = x_0 \\ \lim_{n \to \infty} f(x_n) = A \end{gathered}

正向是显然的.

反向你就反证,然后极限不存在就翻译成

ϵ,δ,x,xx0f(x)Aϵ\exists \epsilon, \forall \delta, \exists x,\vert x-x_0 \vert \\ \vert f(x)-A \vert \ge \epsilon

于是你取一个极限是00δ\delta,得到一列收敛到x0x_0xx,然后用这个数列就矛盾了.

然后你可以用这种方法,把函数极限的各种性质转化到数列极限,四则运算,夹逼等.

柯西收敛

limxx0 exists    ϵ,δ,x1,x2N(x0,δ),f(x1)f(x2)<ϵ\begin{gathered} \lim_{x \to x_0} \text{ exists} \iff \forall \epsilon, \exists \delta,\forall x_1,x_2 \in N^*(x_0,\delta),\vert f(x_1)-f(x_2) \vert < \epsilon \end{gathered}

正向是显然的.

反向的话你可以用上面Heine转化成数列,则你只需要证所有这样的数列极限相等.

然后你发现你直接取任意两个数列,然后插(奇数偶数项分别放两个数列的元素)就可以直接证明这两个数列极限相等.于是结束.

{limxx0f(x)=Alimtt0g(t)=x0η>0,tN(t0,η)    g(t)0    limtt0f(g(t))=A\begin{gathered} \begin{cases} \lim_{x \to x_0} f(x) = A \\ \lim_{t \to t_0} g(t) = x_0 \\ \exists \eta>0, t\in N(t_0,\eta) \implies g(t)\ne 0 \end{cases} \\ \implies \lim_{t \to t_0} f(g(t)) = A \end{gathered}

直接翻译成ϵδ\epsilon-\delta是显然的.

lim0sin(x)x=1\begin{gathered} \lim_{ \to _0} \dfrac{\sin(x)}{x} = 1 \end{gathered}

通过sin(x)xtan(x)\sin(x)\le x \le \tan(x)同时除以xx得到:

{sin(x)x1sin(x)xcos(x)Squeeze Theoremlimx0sin(x)x=1\begin{gathered} \begin{cases} \dfrac{\sin(x)}{x} \le 1 \\ \dfrac{\sin(x)}{x} \ge \cos(x) \end{cases} \\ \stackrel{\text{Squeeze Theorem}}{\Longrightarrow} \\ \lim_{x \to 0} \dfrac{\sin(x)}{x} =1 \end{gathered}
R(x)={1,x=01q,x=pq0,xQ    limxx0R(x)=0\begin{gathered} R(x)=\begin{cases} 1,x=0 \\ \dfrac{1}{q}, x=\dfrac{p}{q} \\ 0,x\notin Q \end{cases} \implies \lim_{x \to x_0} R(x)=0 \end{gathered}

首先显然是周期函数可以只看[0,1][0,1].

其次对于任意x0x_0,对ϵ\epsilon,对所有 n1ϵn\le \dfrac{1}{\epsilon},令 δ=12minjn,ij{ijx0}\delta=\dfrac{1}{2} \min_{j\le n,i\le j} \{ \vert \dfrac{i}{j} -x_0 \vert \}即可保证,该区间[xδ,x+δ][x-\delta,x+\delta]内没有分母比nn小的有理数,于是就证明了.

Ex Class 1

定义指数函数

承认确界原理,实数的四则运算,<<关系,等式性质等.

定义指数函数ax(a>0),xR下a^x(a>0),x\in R

首先良好的定义xNx\in N的情况.

然后尝试定义 x=1k,kNx=\dfrac{1}{k},k\in N的情况,那么要证明

x>0,nN,!y s.t. yn=x\begin{gathered} \forall x>0,n\in N^*,\exists! y {\ } s.t. {\ } y^n=x \end{gathered}

那么构造 E={ttn<x}E=\{ t \vert t^n<x \},容易证明:

  • 非空:x1+xE\dfrac{x}{1+x}\in E
  • 有上界:(x+1)n>x(x+1)^n>x

于是确界存在,设上确界为yy,尝试证明yn=xy^n=x.

假设yn<xy^n<x,设h(0,min(1,xynn(y+1)n1))h\in (0,\min(1,\dfrac{x-y^n}{n(y+1)^{n-1}})).

于是发现(y+h)nyn<hn(y+h)n1<xyn(y+h)^n-y^n<hn(y+h)^{n-1}<x-y^n

于是(y+h)n<x,y+hE(y+h)^n<x,y+h\in E,与yy是上确界矛盾.

再假设yn>xy^n>x,设h=ynxnyn1h=\dfrac{y^n-x}{ny^{n-1}}.

yn(yh)n<hnyn1=ynxy^n-(y-h)^n<hny^{n-1}=y^n-x,即(yh)n>x(y-h)^n>x,故yhy-h也是EE的上界,与上确界矛盾.

于是就证明出yn=xy^n=x.

那么现在我们能定义 ax,xQa^x,x\in Q. 了吗?

不行.接下来你要证明有理数约不约分结果是一样的.即:

r=mn=pq    a=(xm)1n=(xp)1q=b\begin{gathered} r=\dfrac{m}{n} =\dfrac{p}{q} \implies a=(x^m)^{\frac1n}=(x^p)^{\frac1q}=b \end{gathered}
anq=xnrq,bnq=xnrq    anq=bnq    a=b\begin{gathered} a^{nq}=x^{nrq},b^{nq}=x^{nrq} \\ \implies a^{nq}=b^{nq} \\ \implies a=b \end{gathered}

下一个目标是实数!

定义ax,xRa^x,x\in R是所有aq,qQ,qxa^q,q\in Q,q\le x的上确界就行了吧!

那么你是不是得说说实数这个满足和刚才一样的运算律,也就是:

axay=ax+y(ax)y=axy\begin{gathered} a^xa^y=a^{x+y} \\ (a^x)^y=a^{xy} \end{gathered}

然后他下课不讲了

todo

Class 8

等价无穷小替换

重点就是你只能替换形如F(x)P(x)F(x)Q(x)(P(x)Q(x))F(x)P(x)\to F(x)Q(x)(P(x)\sim Q(x)).

你不能替换F(x,P(x))F(x,Q(x))F(x,P(x))\to F(x,Q(x))这种.

这样正确性就保证了.

然后对F(x)(P(x)+H(x))F(x)(Q(x)+H(x))F(x)(P(x)+H(x))\to F(x)(Q(x)+H(x))是合法的,当且仅当你低次项没消掉.

然后通用就是你换的时候带上配亚诺余项就不会错了.

反函数连续性

连续函数的反函数是连续的.

f(x) is continuous,ff1=x    f1is continuous\begin{gathered} f(x) \text{ is continuous} ,f\circ f^{-1}=x \\ \implies f^{-1} \text{is continuous} \end{gathered}

Sol1

limyy0f1(y)=f1(y0)=x0    {yn},limnyn=y0 s.t. limnf1(y)=x0Contrapose! Assumeϵ,δ,y1N(y0,δ),f1(y1)x0>ϵ.{xn},xn=y1(δ).xnis bounded    {in},xinis convergent,limnxin=Xx0limnf(xin)=f(X)limnf(xn)=limnyn=y0f(X)=y0=f(x0),Xx0Contradiction to injection\begin{gathered} \lim_{y \to y_0} f^{-1}(y)=f^{-1}(y_0)=x_0 \\ \iff \forall \{ y_n \} ,\lim_{n \to \infty} y_n=y_0 \ s.t.\ \lim_{n \to \infty} f^{-1}(y)=x_0 \\ \text{Contrapose! Assume} \exists \epsilon,\forall \delta,\exists y_1\in N^*(y_0,\delta),\vert f^{-1}(y_1)-x_0 \vert > \epsilon. \\ \exists \{ x_n \} ,x_n=y_1(\delta). \\ x_n \text{is bounded} \implies \exists \{ i_n \} , \\ x_{i_n} \text{is convergent} ,\lim_{n \to \infty} x_{i_n}=X\ne x_0 \\ \lim_{n \to \infty} f(x_{i_n})=f(X) \\ \because \lim_{n \to \infty} f(x_n) =\lim_{n \to \infty} y_n=y_0 \\ \therefore f(X)=y_0=f(x_0),X\ne x_0 \\ \text{Contradiction to injection} \end{gathered}

[think] 这里为什么数列这么好用呢?感觉因为函数极限的定义不是对称的,但 limnxn=x0,limnyn=y0\lim_{n \to \infty} x_n=x_0,\lim_{n \to \infty} y_n=y_0 这个关系是对称的.

Sol2

容易证明连续函数是单射必须严格单调.

x0x_0的任意邻域NN,f(N)f(N)也是y0y_0的邻域,于是 ϵ,xN(x0,ϵ)    yf(N(x0,ϵ))\forall \epsilon,x\in N(x_0,\epsilon) \implies y\in f(N(x_0,\epsilon)),把这个翻译成ϵδ\epsilon-\delta.

[think] 考虑的是连续函数把邻域变成邻域(或者说连续函数把闭集映到闭集,于是在一边有极限在另一边也有)吧.

所以证连续其实是用不着单调的. 不过我们知道连续函数单射一定是单调的.

初等函数都是连续函数

初等函数都是连续函数

xa=ealnxx^a=e^{a\ln x},于是只要证:

  • 指数函数和三角函数是连续的
  • 连续函数的反函数是连续的

指数:

limxx0ex=ex0limxx0exx0=ex0limx0ex    ex is continuous    ex is continuous at 0limx0ex=0    xn,limnexn=0,limnxn=0limnn1n=1    limnexn=0\begin{gathered} \lim_{x \to x_0} e^x = e^{x_0} \lim_{x \to x_0} e^{x-x_0} = e^{x_0} \lim{x\to 0} e^x \\ \implies e^x \text{ is continuous} \iff e^x \text{ is continuous at } 0 \\ \lim_{x \to 0} e^x=0 \\ \iff \forall x_n,\lim_{n \to \infty} e^{x_n} = 0,\lim_{n \to \infty} x_n=0 \\ \lim_{n \to \infty} n^{\frac{1}{n}}=1 \implies \lim_{n \to \infty} e^{x_n}=0 \end{gathered}

三角:

limΔx0sin(x+Δx)=sinxcosΔx+cosxsinΔx=sinx\begin{gathered} \lim_{\Delta x \to 0} \sin(x+\Delta x)=\sin x\cos \Delta x+\cos x\sin \Delta x \\ =\sin x \end{gathered}

反函数用前面的方法.

Class 9

Intermediate value theorem

{x[a,b],f is continuousy[f(a),f(b)]    x0,f(x0)=y\begin{cases} x\in [a,b], f\text{ is continuous} \\ y\in [f(a),f(b)] \end{cases} \implies \exists x_0, f(x_0)=y

Sol1

不妨设f(a)yf(b)f(a)\le y \le f(b),==情况显然,只考虑f(a)<y<f(b)f(a)<y<f(b).

A={xf(x)<y}A=\{ x\vert f(x)<y \} 则它有上确界x1x_1.

那么要证明f(x1)=yf(x_1)=y,考虑:

f(x1)<yf(x_1)<y,则 limxx1f(x)=y    ϵ=yf(x1),x(x1δ,x1+δ)    f(x)(f(x1)ϵ,f(x1)+ϵ)<y\lim_{x \to x_1} f(x)=y \implies \epsilon=y-f(x_1),\forall x\in (x_1-\delta,x_1+\delta)\implies f(x)\in (f(x_1)-\epsilon,f(x_1)+\epsilon)<y,于是x2>x1,f(x2)<y\exists x_2>x_1,f(x_2)<y,与x1x_1上确界矛盾.故f(x1)yf(x_1)\le y.

同理f(x1)yf(x_1)\ge y,于是f(x1)=yf(x_1)=y.

那么一定存在一个收敛到x1x_1的数列 {zn}\{ z_n \} 你就直接发现 limnf(zn)=f(limnzn)\lim_{n \to \infty} f(z_n) = f(\lim_{n \to \infty} z_n)

Q.E.D\text{Q.E.D}

[think] 核心在 x0=sup{xf(x)<y}x_0=\sup \{ x \vert f(x)<y \},即先看到构造x0x_0的方式.

Sol2

另一个做法是闭区间套,把区间二等分,那么把平凡情况讨论掉后,一定有(f(a)y)(f(b)y)<0(f(a)-y)(f(b)-y)<0,那么现在区间中点mm处,f(m)=yf(m)=y直接结束,否则递归到值域区间包含yy的一边.

如果过程没有在中间停止,那么闭区间套定理,!ξ\exists !\xi在所有的区间中.考虑区间序列[an,bn][a_n,b_n],那么显然f(an)pf(a_n)-pf(bn)pf(b_n)-p符号始终不变且相异,于是对a,ba,b分别取极限可以证f(ξ)f(\xi)收敛到yy.

[think] 讲题顺序很迷惑(这个证明是Class 11讲零点存在弄出来的). 不过你注意到这个闭区间套证法是可以证明其他几条连续函数性质的(闭区间上一致连续,有界,极值定理都是可以的).

间断点分类.

when limxx0f(x)=f(x0) not satisfied\begin{gathered} \text{when } \lim_{x \to x_0} f(x)=f(x_0) \text{ not satisfied} \end{gathered}

分类:

  • f(x0+)f(x0)f(x_0^+)\ne f(x_0^-):跳跃间断点
  • f(x0+)=f(x0)f(x0)f(x_0^+)=f(x_0^-)\ne f(x_0):可去间断点
  • f(x0+) or f(x0) not existsf(x_0^+) \text{ or } f(x_0^-) \text{ not exists}:无穷间断点

[a,b][a,b]上有定义的单调函数的间断点必然是跳跃间断点.

那么先证明左右极限存在:若x0x_0是间断点,这里和介值定理的证明是一样的:

let S={f(x)x[a,x0)}A=supSif A<f(x0),by limit’s local sign-preserving propertyx1N(x0),x1<x0,f(x1)>A,Contradiction!if A>f(x0),xn,f(xn)>Aϵ>f(x0)Contradiction!A=f(x0)same for B=f(x0+)x1<x0,x2>x0,f(x1)<f(x2)f(x1)<f(x0)limx1x0Af(x0)f(x2)>f(x0)limx2x0Bf(x0)A=B    A=f(x0)=BAB\begin{gathered} \text{let } S=\{ f(x)\vert x\in[a,x_0) \} \\ A=\sup S \\ \text{if }A<f(x_0^-),\text{by limit's local sign-preserving property} \\ \exists x_1\in N^-(x_0),x_1<x_0,f(x_1)>A,\text{Contradiction!} \\ \text{if } A>f(x_0^-),\exists x_n,f(x_n)>A-\epsilon>f(x_0^-)\text{Contradiction!} \\ \therefore A=f(x_0^-) \\ \text{same for } B=f(x_0^+) \\ \forall x_1<x_0,x_2>x_0,f(x_1)<f(x_2) \\ f(x_1)<f(x_0) \stackrel{\lim_{x_1 \to x_0} }{\Longrightarrow}A\le f(x_0) \\ f(x_2)>f(x_0)\stackrel{\lim_{x_2\to x_0}}{\Longrightarrow}B\ge f(x_0) \\ \therefore A=B \implies A=f(x_0)=B \\ \therefore A\ne B \end{gathered}

[think] 这个证明和介值定理证明那里都走了:构造集合,确界存在,确界等于某值,构造集合里极限为确界的收敛数列,连续性 的流程, 似乎是推出集合边界极限的套路.

[a,b][a,b]上有定义的单调函数的间断点必然至多可数个.

x1,x2x_1,x_2两个跳跃间断点:

f(x1)<f(x1+)f(x2)<f(x2+)\begin{gathered} f(x_1^-)<f(x_1^+)\le f(x_2^-)<f(x_2^+) \end{gathered}

于是每个间断点xix_i对应一个(f(xi),f(xi+))(f(x_i^-),f(x_i^+)),且不同xix_i对应区间互不相交.

于是每个区间可以对应一个不同的有理数,有理数至多可数.

[think] 于是任何一个区间的不交划分只有可数个.(禁止[a,a][a,a]的情况)

Ex Class 2

其实这节课在Class 10后面 但讲间断点的话和这里更近.

f(x),x(a,b),f has at most countable point of discontinuity of the first kind\begin{gathered} \forall f(x), x\in (a,b),f \text{ has at most countable point of discontinuity of the first kind} \end{gathered}

首先我们考虑右极限大于左极限的间断点,证明它们是可数的.

对于一个间断点x0x_0,limxx0f(x)=A,limxx0+f(x)=B\lim_{x \to x_0^-} f(x)=A,\lim_{x \to x_0^+} f(x)=B,显然我们可以找到pQ,p(A,B)p\in Q,p\in (A,B).

而由于极限保号性,你也容易找到q,rQq,r\in Q使得x(q,x0),f(x)<p,x(x0,r),f(x)>p\forall x\in (q,x_0),f(x)<p,\forall x\in (x_0,r),f(x)>p.

考虑是否可能有两个间断点x1<x2x_1<x_2对应相同的p,r,qp,r,q,那么显然q<x1<x2<rq<x_1<x_2<r,但是(x1,r)(x_1,r)中的值小于pp,(q,x2)(q,x_2)中的值大于pp,而这两个区间有交,所以不可能有两个间断点对应相同的p,r,qp,r,q,那么这些间断点可以单射到Q3Q^3上是可数的.

而对于左极限小于右极限的间断点同理可得.

接下来考虑左右极限相等的间断点,让p(A,f(x0))p\in (A,f(x_0)),而q,rq,r满足(q,x0),(x0,r)(q,x_0),(x_0,r)f(x)<pf(x)<p,那么考虑有两个相同的这样的p,q,rp,q,rx1,x2x_1,x_2,你会发现f(x1)f(x_1)f(x2)f(x_2)不能都小于pp你就爆炸了.所以也是不同的间断点对应不同的数对.

于是都是可数的.总和也是可数的.

Class 10

Some Limits' Calculation

f1,glimfg=elimglnf=elimg(f1)\begin{gathered} \\ f\to 1,g\to \infty \\ \lim f^g=e^{\lim g\ln f}=e^{\lim g(f-1)} \end{gathered}
limx(1+1x)x=e\begin{gathered} \lim_{x \to \infty} (1+\dfrac{1}{x} )^x=e \end{gathered}
limx+(1+1x)x((1+1[x]+1)[x],(1+1[x])[x]+1)let en=limn(1+1n)nlimx+(1+1[x]+1)[x]=limne[x]+1(1+1[x]+1)1=elimx+(1+1[x])[x]+1=limne[x](1+1[x])=e Squeeze Theorem limx+(1+1x)x=elimx(1+1x)x=limx+(1+1x)x=limx+(xx1)x=limx+(1+1x1)x=e\begin{gathered} \lim_{x \to +\infty} (1+\dfrac{1}{x} )^x\in ((1+\dfrac{1}{[x]+1} )^{[x]},(1+\dfrac{1}{[x]} )^{[x]+1}) \\ \text{let } e_n=\lim_{n \to \infty} (1+\dfrac{1}{n} )^n \\ \lim_{x \to +\infty} (1+\dfrac{1}{[x]+1} )^{[x]} \\ =\lim_{n \to \infty} e_{[x]+1}(1+\dfrac{1}{[x]+1})^{-1} \\ =e \\ \lim_{x \to +\infty} (1+\dfrac{1}{[x]} )^{[x]+1} \\ =\lim_{n \to \infty} e_{[x]}(1+\dfrac{1}{[x]} ) \\ =e \\ \stackrel{\text{ Squeeze Theorem }}{\Longrightarrow} \lim_{x \to +\infty} (1+\dfrac{1}{x} )^x=e \\ \lim_{x \to -\infty} (1+\dfrac{1}{x} )^x \\ =\lim_{x \to +\infty}(1+\dfrac{1}{-x} )^{-x} \\ =\lim_{x \to +\infty}(\dfrac{x}{x-1} )^x \\ =\lim_{x \to +\infty}(1+\dfrac{1}{x-1} )^x \\ =e \end{gathered}

真麻烦.一句话就是取整夹你.

limx1mxm1nxn1\begin{gathered} \lim_{x \to 1} \dfrac{m}{x^m-1} -\dfrac{n}{x^n-1} \end{gathered}
limx1mxm1nxn1=limx1m(xn1)n(xm1)(xm1)(xn1)=limx1m(n(x1)+n(n1)2(x1)2)mn(x1)2n(m(x1)+m(m1)2(x1)2)mn(x1)2=limx1mn2nm22nm=nm2\begin{gathered} \lim_{x \to 1} \dfrac{m}{x^m-1} -\dfrac{n}{x^n-1} =\lim_{x \to 1} \dfrac{m(x^n-1)-n(x^m-1)}{(x^m-1)(x^n-1)} \\ =\lim_{x \to 1} \dfrac{m(n(x-1)+\dfrac{n(n-1)}{2} (x-1)^2)}{mn(x-1)^2} \\ -\dfrac{-n(m(x-1)+\dfrac{m(m-1)}{2} (x-1)^2)}{mn(x-1)^2} \\ =\lim_{x \to 1} \dfrac{mn^2-nm^2}{2nm} \\ =\dfrac{n-m}{2} \end{gathered}

Uniform Continuity

Uniform Continuity

f(x) is uniformly continuous in [a,b]    ϵ>0,δ,x1,x2[a,b],x1x2<δ    f(x1)f(x2)<ϵ\begin{gathered} f(x) \text{ is uniformly continuous in } [a,b] \\ \iff \forall \epsilon>0,\exists \delta,\forall x_1,x_2\in [a,b], \\ \vert x_1-x_2 \vert <\delta \implies \vert f(x_1)-f(x_2) \vert <\epsilon \end{gathered}

Not Uniform Continuity

ϵ,sn,tn,sntn<1n,f(sn)f(tn)>ϵ\begin{gathered} \exists \epsilon,s_n,t_n, \\ \vert s_n-t_n \vert < \dfrac{1}{n} ,\vert f(s_n)-f(t_n) \vert >\epsilon \end{gathered}
f(x)=x is uniformly continuous in [0,+)\begin{gathered} f(x)=\sqrt{ x } \text{ is uniformly continuous in } [0,+\infty) \end{gathered}
x1>x2    x1x2=x1x2x1+x2<x1x2x1x2=x1x2<δ\begin{gathered} x_1>x_2 \implies \\ \vert \sqrt{x_1}-\sqrt{ x_2 } \vert ={\left \vert \dfrac{x_1-x_2}{\sqrt{ x_1 } +\sqrt{ x_2 } } \right \vert} < {\left \vert \dfrac{x_1-x_2}{\sqrt{x_1-x_2}} \right \vert} =\sqrt{ x_1-x_2 }<\sqrt \delta \end{gathered}
f(x)=1x is not uniformly continuous in (0,1)\begin{gathered} f(x)=\dfrac{1}{x} \text{ is not uniformly continuous in } (0,1) \end{gathered}
1n12n<1n,y(1n)y(12n)=n1\begin{gathered} \dfrac{1}{n} -\dfrac{1}{2n} <\dfrac{1}{n} , \\ y(\dfrac{1}{n} )-y(\dfrac{1}{2n} )=n\ge 1 \end{gathered}

闭区间连续函数性质

闭区间上的连续函数一致连续

反证

ϵ,sn,tn,sntn<1n,f(sn)f(tn)>ϵ\begin{gathered} \exists \epsilon,s_n,t_n,\vert s_n-t_n \vert <\dfrac{1}{n} ,\vert f(s_n)-f(t_n) \vert >\epsilon \\ \end{gathered}

sn,tns_n,t_n的收敛子列sn,tns'_n,t'_n:

letx0=limnsn=limnsn=limntn=limntn[a,b]limnf(sn)=limnf(tn)=f(x0)    limnf(sn)f(tn)=0\begin{gathered} \text{let} x_0=\lim_{n \to \infty} s'_n=\lim_{n \to \infty} s_n=\lim_{n \to \infty} t'_n=\lim_{n \to \infty} t_n \in [a,b] \\ \lim_{n \to \infty} f(s'_n)=\lim_{n \to \infty} f(t_n')=f(x_0) \\ \implies \lim_{n \to \infty} \vert f(s'_n)-f(t'_n) \vert =0 \end{gathered}

闭区间上连续函数一定有界.

Sol1

反证,设无界,xn\exists x_n满足f(xn)>nf(x_n)>n,取xnx_n的收敛子列,那么 f(limnxn)=f(\lim_{n \to \infty} x_n)=\infty矛盾.

Sol2

用上面一致连续,那么区间的值域跨度不超过 baδϵ\dfrac{b-a}{\delta} \epsilon.

闭区间上连续函数一定能取到最大值最小值.

{M=sup{f(x)x[a,b]}f is continuous    x[a,b],f(x)=M\begin{cases} M=\sup \{ f(x) \vert x\in [a,b] \} \\ \text{f is continuous} \end{cases} \implies \exists x\in [a,b], f(x)=M

考虑取任意M1<MM_1<M,可以得到一个f(x1)(M1,M)f(x_1)\in (M_1,M),取Mn>f(xn1)M_n>f(x_{n-1})可得xn(Mn,M)x_n \in (M_n,M),显然xnx_n有界可以取收敛子列yny_n,则limnf(yn)=f(limnyn)\lim_{n \to \infty} f(y_n)=f(\lim_{n \to \infty} y_n),则于是得证

Class 11

Continuous Function's Periodicty

{f(x) is a periodic function that isn’t constantf(x) is continuous    f(x) has min positive period\begin{cases} f(x)\text{ is a periodic function that isn't constant} \\ f(x)\text{ is continuous} \end{cases} \\ \implies f(x)\text{ has min positive period}
T:={tt>0,f(x+t)=f(x)}t0:=infT\begin{gathered} T:= \{ t \vert t>0,f(x+t)=f(x) \} \\ t_0:= \inf T \\ \end{gathered}

t00t_0\ne 0,取 {tn},tiT,limntn=t0\{ t_n \},t_i\in T,\lim_{n \to \infty} t_n=t_0,则 f(x)=f(x+tn)=limnf(x+tn)=f(x+limntn)=f(x+t0)f(x)=f(x+t_n)=\lim_{n \to \infty} f(x+t_n)=f(x+\lim_{n \to \infty} t_n)=f(x+t_0)

t0=0t_0=0,因为不是常函数, 则x0 s.t. f(x0)f(0)\exists x_0 {\ } s.t. {\ } f(x_0)\ne f(0),因为连续,和极限有保号性你可以说xO(x0,δ),f(x)f(0)\forall x\in O(x_0,\delta),f(x)\ne f(0),但由t0=0t_0=0,你可以找到t<δt<\delta,那就可以再找到NN使得Nt(x0δ,x0+δ)Nt\in (x_0-\delta,x_0+\delta),NtNt一定是周期,但f(0+Nt)f(0)f(0+Nt)\ne f(0),矛盾.

于是得证.

Directive

导数的定义,初等函数求导,导数的四则运算.

Class 12

极限里换元

u(x) is continuous,limuu0G(u)=A    limxx0G(u(x))=A\begin{gathered} u(x) \text{ is continuous},\lim_{u \to u_0} G(u)=A \\ \implies \lim_{x \to x_0} G(u(x)) = A \end{gathered}
ϵ,δ,u(x)N(u(x),δ1)    G(u(x))N(A,ϵ)    ϵ,δ,xN(x0,δ)    u(x)N(u(x),δ1)\begin{gathered} \forall \epsilon,\exists \delta, u(x)\in N^*(u(x),\delta_1) \implies G(u(x))\in N(A,\epsilon) \\ \implies \forall \epsilon,\exists \delta,x\in N^*(x_0,\delta) \implies u(x)\in N(u(x),\delta_1) \\ \end{gathered}

这里有一点小小的问题,不过你发现只要定义G1(u(x0))=G1(u0)=AG_1(u(x_0))=G_1(u_0)=A即可避免,因为条件里这个点反正是任意的.

于是结束.

复合函数求导(链式法则)

(f(g(x)))=f(g(x))g(x)\begin{gathered} (f(g(x)))'=f'(g(x))g'(x) \end{gathered}

Sol

用上面的极限换元一换就出来啦

My Sol

f(x)=f(x0)+f(x0)(xx0)+o(xx0)g(x)=g(x0)+g(x0)(xx0)+o(xx0)limxx0f(g(x))f(g(x0))xx0=limxx0f(g(x0)+g(x0)(xx0)+o(xx0))f(g(x0))xx0=limxx0f(g(x0))(g(x0)(xx0)+o(xx0))xx0+o(g(x0)(xx0)+o(xx0))xx0=limxx0f(g(x0))g(x0)+o(g(x0)(xx0)+o(xx0))xx0\begin{gathered} f(x)=f(x_0)+f'(x_0)(x-x_0)+o(x-x_0) \\ g(x)=g(x_0)+g'(x_0)(x-x_0)+o(x-x_0) \\ \lim_{x \to x_0} \dfrac{f(g(x))-f(g(x_0))}{x-x_0} \\ =\lim_{x \to x_0} \dfrac{f(g(x_0)+g'(x_0)(x-x_0)+o(x-x_0))-f(g(x_0))}{x-x_0} \\ =\lim_{x \to x_0} \dfrac{f'(g(x_0))(g'(x_0)(x-x_0)+o(x-x_0))}{x-x_0} \\ +\dfrac{o(g'(x_0)(x-x_0)+o(x-x_0))}{x-x_0} \\ =\lim_{x \to x_0} f'(g(x_0))g'(x_0)+\dfrac{o(g'(x_0)(x-x_0)+o(x-x_0))}{x-x_0} \\ \end{gathered}

而满足o(xx0)ϵ(xx0)o(x-x_0)\le\epsilon(x-x_0)xxx0x_0的一个邻域,于是

limxx0o(g(x0)(xx0)+o(xx0))xx0limxx0o((g(x0)+ϵ)(xx0))xx0=0\begin{gathered} \lim_{x \to x_0} \dfrac{o(g'(x_0)(x-x_0)+o(x-x_0))}{x-x_0} \\ \le \lim_{x \to x_0} \dfrac{ o((g'(x_0)+\epsilon)(x-x_0))}{x-x_0} \\ =0 \end{gathered}

得证.

只能说很暴力.

莱布尼茨求导公式

(uv)(n)=i=10(ni)u(i)v(ni)\begin{gathered} (uv)^{(n)}=\sum _{i = 1} ^{0} \binom{n}{i} u^{(i)}v^{(n-i)} \end{gathered}

归纳

据说一般用在有一个高阶导数是00的情况

arctan(50)(0)=0\begin{gathered} \arctan^{(50)}(0)=0 \end{gathered}

Sol 1

arctan(x)=11+x2    (1+x2)arctan(x)=10=((1+x2)arctan(x))(n)=(1+x2)arctan(n+1)(x)+(n1)2xarctan(n)(x)+(n2)2arctan(n1)(x)x=0arctan(n+1)(0)+n(n1)arctan(n1)(0)=0\begin{gathered} \arctan'(x)=\dfrac{1}{1+x^2} \\ \implies (1+x^2)\arctan'(x)=1 \\ 0=((1+x^2)\arctan'(x))^{(n)} \\ =(1+x^2)\arctan^{(n+1)}(x)+\binom{n}{1}2x\arctan^{(n)}(x)+\binom{n}{2}2\arctan^{(n-1)}(x) \\ \stackrel{ x=0 }{\Longrightarrow} \arctan^{(n+1)}(0)+n(n-1) \arctan^{(n-1)}(0)=0 \end{gathered}

于是有递推,得到是00.

Sol 2

注意到一阶导是偶函数,又求了奇数次变成奇函数,所以说00.

Sol 3

11+x2=1(xi)(x+i)=1(1ix)(1+ix)=12i(1xi1x+i)\begin{gathered} \dfrac{1}{1+x^2}=\dfrac{1}{(x-i)(x+i)} =\dfrac{1}{(1-ix)(1+ix) } \\ =\dfrac{1}{2i} (\dfrac{1}{x-i}-\dfrac{1}{x+i}) \end{gathered}

于是可以对两个分数分别求导.复变函数从实数轴上逼近的导数当然就是原函数的导数.

Class 13

中值定理

Theorems

极值点

f(x0)f(x_0)是极大值当且仅当存在δ,xN(x,δ),f(x)<f(x0)\delta,\forall x\in N^*(x,\delta),f(x)<f(x_0)

极值和最值既不充分也不必要.

最值是极值当且仅当最值在区间内部,端点不行.

Fermat Theorem

f(x0)f(x_0)是极值点,f(x0)f'(x_0)存在则f(x0)=0f'(x_0)=0

不妨设是极大值.

f(x0)=limxx0f(x)f(x0)xx00f(x0+)=limxx0f(x)f(x0)xx00    f(x0)=f(x0)=f(x0+)=0\begin{gathered} f'(x_0^-)=\lim_{x \to x_0^-} \dfrac{f(x)-f(x_0)}{x-x_0} \ge 0 \\ f'(x_0^+)=\lim_{x \to x_0} \dfrac{f(x)-f(x_0)}{x-x_0} \le 0 \\ \implies f'(x_0)=f'(x_0^-)=f'(x_0^+)=0 \end{gathered}

Rolle's Theorem

{f(x)C[a,b]x(a,b)    f(x)f(a)=f(b)    ξ(a,b),f(ξ)=0\begin{gathered} \begin{cases} f(x) \in C[a,b] \\ x\in (a,b) \implies \exists f'(x) \\ f(a)=f(b) \end{cases} \\ \implies \exists \xi\in (a,b),f'(\xi)=0 \end{gathered}

用连续函数最值定理,并且在ff不是常函数由f(a)=f(b)f(a)=f(b)显然有至少一个最值在中间取.这个最值是极值.极值处费马定理,得证.

Lagrange Mean Value Theorem

{f(x)C[a,b]x(a,b)    f(x)    ξ(a,b),f(ξ)=f(a)f(b)ab\begin{gathered} \begin{cases} f(x) \in C[a,b] \\ x\in (a,b) \implies \exists f'(x) \end{cases} \\ \implies \exists \xi\in (a,b),f'(\xi)=\dfrac{f(a)-f(b)}{a-b} \end{gathered}
let F(x)=f(x)f(a)f(b)(ab)(xa)\begin{gathered} \text{let } F(x)=f(x)-\dfrac{f(a)-f(b)}{(a-b)} (x-a) \end{gathered}

然后直接 Rolle's Theorem.

Cauchy Mean Value Theorem

{f(x)C[a,b]x(a,b)    f(x)g(x)C[a,b]x(a,b)    g(x)(a,b),x(a,b),g(x)≢0ξ(a,b),f(ξ)g(ξ)=f(a)f(b)g(a)g(b)\begin{gathered} \begin{cases} f(x) \in C[a,b] \\ x\in (a,b) \implies \exists f'(x) \\ g(x)\in C[a,b] \\ x\in (a,b) \implies \exists g'(x) \\ \forall (a',b'),x\in (a',b'),g'(x)\not \equiv 0 \end{cases} \\ \exists \xi \in (a,b),\dfrac{f'(\xi)}{g'(\xi)} =\dfrac{f(a)-f(b)}{g(a)-g(b)} \end{gathered}
let F(x)=f(x)f(a)f(b)g(a)g(b)(g(x)g(a))\begin{gathered} \text{let } F(x)=f(x)-\dfrac{f(a)-f(b)}{g(a)-g(b)} (g(x)-g(a)) \end{gathered}

Rolle's Theorem 启动

Usage

他说你应该在研究导函数零点的时候用Rolle,研究函数和导函数关系用Lagrange,研究两个函数的时候用Cauchy

limx+f(x)=0    limx+f(x)x=0\begin{gathered} \lim_{x \to +\infty} f'(x)=0 \implies \lim_{x \to +\infty} \dfrac{f(x)}{x} =0 \end{gathered}
x>A>X    f(x)f(A)<ϵ1(xA)    f(x)x<f(A)ϵ1Ax+ϵ1\begin{gathered} x>A>X \\ \implies \vert f(x)-f(A) \vert <\epsilon_1(x-A) \\ \implies \vert \dfrac{f(x)}{x} \vert <\dfrac{f(A)-\epsilon_1A}{x} +\epsilon_1 \end{gathered}

其实从这就能看出来了,同时取极限则 limx+f(x)x\lim_{x \to +\infty} \vert \dfrac{f(x)}{x} \vert小于任意正数,于是是00.

a>b,ξ(a,b),aebbea=(1ξ)eξ(ab)\begin{gathered} \forall a>b,\exists \xi \in (a,b),ae^b-be^a=(1-\xi)e^\xi(a-b) \end{gathered}
    ebbeaa1b1a=(1ξ)eξ\begin{gathered} \implies \dfrac{\dfrac{e^b}{b} -\dfrac{e^a}{a} }{\dfrac{1}{b} -\dfrac{1}{a}} =(1-\xi)e^{\xi} \end{gathered}

然后柯西中值结束.

ξC[a,b],f(x)    ξ(a,b),f(b)+f(a)2f(a+b2)=(ba2)2f(ξ)\begin{gathered} \exists \xi\in C[a,b],\exists f''(x) \\ \implies \exists \xi \in (a,b), \\ f(b)+f(a)-2f(\dfrac{a+b}{2})=(\dfrac{b-a}{2})^2f''(\xi) \end{gathered}
let F(x)=f(x)f(xba2)F(b)F(a+b2)=f(b)+f(a)2f(a+b2)=ba2(F(ξ1)F(ξ1ba2))=(ba2)2F(ξ)\begin{gathered} \text{let } F(x)=f(x)-f(x-\dfrac{b-a}{2}) \\ F(b)-F(\dfrac{a+b}{2} )=f(b)+f(a)-2f(\dfrac{a+b}{2}) \\ =\dfrac{b-a}{2} (F'(\xi_1)-F'(\xi_1-\dfrac{b-a}{2} )) \\ =(\dfrac{b-a}{2} )^2F''(\xi) \end{gathered}

我们注意到如果你上来不构造函数对f(b)f(mid),f(mid)f(a)f(b)-f(mid),f(mid)-f(a)分别用拉格朗日中值是做不出来的.这么干限制强(要求值对应相等而非差相等)显然就不如分开.

f(x)C[1,+),f(x)exf(x) is bounded in [1,+)    exf(x) is bounded in (1,+)\begin{gathered} f(x)\in C[1,+\infty),\exists f'(x) \\ e^{-x}f'(x) \text{ is bounded in } [1,+\infty) \\ \implies e^{-x}f(x) \text{ is bounded in } (1,+\infty) \end{gathered}
f(x1)f(x2)ex1ex2=f(ξ)eξMexf(x)f(x)f(1)ex+f(1)f(x)f(1)exe1+f(1)=f(ξ)eξ+f(1)M+f(1)\begin{gathered} \dfrac{f(x_1)-f(x_2)}{e^{x_1}-e^{x_2}} =\dfrac{f'(\xi)}{e^{\xi}}\le M \\ e^{-x}f(x)\le \dfrac{f(x)-f(1)}{e^x} +f(1) \\ \le \dfrac{f(x)-f(1)}{e^x-e^1} +f(1) \\ = \dfrac{f'(\xi)}{e^\xi}+f(1) \\ \le M+f(1) \end{gathered}
f(x)C(0,1],f(x)limx0+xf(x)    f(x)UC(0,1]\begin{gathered} f(x)\in C(0,1],\exists f'(x) \\ \exists\lim_{x \to 0^+} \sqrt xf'(x) \\ \implies f(x) \in UC(0,1] \end{gathered}

todo

Class 14

Darbox Theorem

x[a,b],f(x)    {v[f(a),f(b)],ξ,f(ξ)=vf(x) has no discontinuity of first kind\begin{gathered} \forall x\in [a,b], \\ \exists f'(x) \\ \implies \begin{cases} \forall v\in [f'(a),f'(b)],\exists \xi,f'(\xi)=v \\ f'(x) \text{ has no discontinuity of first kind} \end{cases} \end{gathered}

(1)

先不妨设f(a)<0<f(b)f'(a)<0<f'(b)

由定义容易说明a,ba,b不是极小值.

于是存在最值定理,(a,b)(a,b)中存在最值xx,于是存在f(x)=0f'(x)=0

其他情况显然可以规约过来.

(2)

考虑对一个间断点x0x_0

第一类间断点所以有左右极限L,RL,R,那么ϵδ,x(x0δ,x0)    f(x0)N(L,ϵ)\forall \epsilon\exists \delta, x \in (x_0-\delta,x_0) \implies f'(x_0)\in N(L,\epsilon).同理有x(x0,x0+δ)    f(x)N(R,ϵ)\forall x\in(x_0,x_0+\delta) \implies f'(x)\in N(R,\epsilon).

于是可以取ϵ\epsilon使得两个邻域不交,则这个小区间上至少越过了一个值.对[x0δ2,x0+δ2][x_0-\dfrac{\delta}{2},x_0+\dfrac{\delta}{2}]用(1)

单调性

f(x)C[a,b],x[a,b]D,f(x)>0D is finite set    f(x) is strictly increasing at [a,b]\begin{gathered} f(x)\in C[a,b],\forall x\in [a,b]-D,f'(x)>0 \\ D \text{ is finite set} \\ \implies f(x) \text{ is strictly increasing at }[a,b] \end{gathered}

x1<x2\forall x_1<x_2,把DD(x1,x2)(x_1,x_2)中的点排序得到d1dkd_1\ldots d_k,令d0=x1,dk=x2d_0=x_1,d_k=x_2,显然有

f(x2)f(x1)=i=1k+1f(di)f(di1)=i=1k+1f(ξi)(didi1)>0\begin{gathered} f(x_2)-f(x_1)=\sum _{i = 1} ^{k+1} f(d_i)-f(d_{i-1}) \\ =\sum _{i = 1} ^{k+1} f'(\xi_i) (d_i-d_{i-1}) \\ >0 \end{gathered}

每个区间都有闭区间连续开区间可导推导闭区间上中值定理.

p,q>1,a,b>0,1p+1q=1    app+bqqab\begin{gathered} p,q>1,a,b>0,\dfrac{1}{p} + \dfrac{1}{q} =1 \\ \implies \dfrac{a^p}{p} +\dfrac{b^q}{q} \ge ab \end{gathered}

也是求偏导啊.

然后下节课讲了凸函数之后还有个做法说的是

f(x)=lnx,x1=ap,x2=aq\begin{gathered} f(x)=\ln x,x_1=a^p,x_2=a^q \end{gathered}

用琴声.

abϵalna+ϵeebϵ\begin{gathered} ab\le \epsilon a\ln a+\dfrac{\epsilon}{e} e^{\frac{b}{\epsilon} } \end{gathered}

求偏导,代入结束.

f(b)=ϵalna+ϵeebϵabf(b)=ebϵ1aa0=eb0ϵ1f(b0)=(bϵ)ebϵ1+ϵeebϵbebϵ1=ϵeebϵϵebe1=0\begin{gathered} f(b)=\epsilon a\ln a+\dfrac{\epsilon}{e} e^{\frac{b}{\epsilon}}-ab \\ f'(b)=e^{\frac{b}{\epsilon} -1}-a \\ a_0=e^{\frac{b_0}{\epsilon}-1} \\ f(b_0)=(b-\epsilon)e^{\frac{b}{\epsilon}-1 }+\dfrac{\epsilon}{e} e^{\frac{b}{\epsilon}}-be^{\frac{b}{\epsilon}-1} \\ =\dfrac{\epsilon}{e} e^{\frac{b}{\epsilon}}-\epsilon e^{\frac{b}{e}-1 } \\ =0 \end{gathered}

这节课其实是讲单调性和导数关系的,只是看起来比较高中能记的不多.

Class 15

凹凸性

定义成了

Convex/Concave function

f(x) is convex function    x,yD,λ(0,1)f(λx+(1λ)y)λf(x)+(1λ)f(y)\begin{gathered} f(x) \text{ is convex function} \iff \\ \forall x,y\in D,\lambda \in (0,1) \\ f(\lambda x+(1-\lambda)y)\le \lambda f(x)+(1-\lambda)f(y) \end{gathered}
f(x) is convex    {xn},{λn},f(i=1nλixii=1nλi)i=1nλif(xi)i=1nλi\begin{gathered} f(x) \text{ is convex} \iff \\ \forall \{ x_n \} ,\{ \lambda_n \}, f(\dfrac{\sum _{i = 1} ^{n} \lambda_ix_i}{\sum _{i = 1} ^{n} \lambda_i} )\le \dfrac{\sum _{i = 1} ^{n} \lambda_if(x_i)}{\sum _{i = 1} ^{n} \lambda_i} \end{gathered}

归纳显然.

f(x) is convex,f(x)    f(x) is increasing\begin{gathered} f(x) \text{ is convex} ,\exists f'(x) \\ \iff f'(x) \text{ is increasing} \end{gathered}

显然凸性等价于

x1<x2<x3    f(x1)f(x2)x1x2<f(x2)f(x3)x2x3\begin{gathered} x_1<x_2<x_3 \implies \\ \dfrac{f(x_1)-f(x_2)}{x_1-x_2} <\dfrac{f(x_2)-f(x_3)}{x_2-x_3} \end{gathered}

这个定理反向用拉格朗日中值是显然的.对正向:

x1<x2f(x1)=limxx1f(x1)f(x)x1xf(x1)f(x2)x1x2same forf(x2)f(x1)f(x2)x1x2\begin{gathered} \forall x_1<x_2 \\ f'(x_1)=\lim_{x \to x_1^-} \dfrac{f(x_1)-f(x)}{x_1-x} \le \dfrac{f(x_1)-f(x_2)}{x_1-x_2} \\ \text{same for} f'(x_2)\ge \dfrac{f(x_1)-f(x_2)}{x_1-x_2} \end{gathered}

于是结束.

i=1naibi(i=1nap)1p(i=1nbq)1q\begin{gathered} \sum _{i = 1} ^{n} a_ib_i\le (\sum _{i = 1} ^{n} a^p )^{\frac1p}(\sum _{i = 1} ^{n} b^q)^{\frac1q} \end{gathered}
(ai(i=1nap)1p)(bi(i=1nbq)1q)1paipi=1nap+1qbiqi=1nbq\begin{gathered} (\dfrac{a_i}{(\sum _{i = 1} ^{n} a^p)^{\frac{1}{p}}} )(\dfrac{b_i}{(\sum _{i = 1} ^{n} b^q)^{\frac{1}{q}}} ) \le \dfrac{1}{p} \dfrac{a_i^p}{\sum _{i = 1} ^{n} a^p}+\dfrac{1}{q} \dfrac{b_i^q}{\sum _{i = 1} ^{n} b^q} \end{gathered}

同时累加即证.

Ex Class 3

对这样的常微分方程:

y=Φ(x,y)y(a)=0\begin{gathered} y'=\Phi(x,y) \\ y(a)=0 \end{gathered}

A,y1,y2,Φ(x,y1)Φ(x,y2)Ay1y2\begin{gathered} \exists A,\forall y_1,y_2, \\ \vert \Phi(x,y_1)-\Phi(x,y_2) \vert \le A \vert y_1-y_2 \vert \end{gathered}

则微分方程的解唯一.

假设存在两个解y1(x),y2(x)y_1(x),y_2(x),令g(x)=y2(x)y1(x)g(x)=y_2(x)-y_1(x)

g(x)=Φ(x,y2)Φ(x,y1)Ay2y1=Ag(x)\begin{gathered} g'(x)=\vert \Phi(x,y_2)-\Phi(x,y_1) \vert \le A \vert y_2-y_1 \vert =A g(x) \end{gathered}

于是由作业Math Analysis Homework Week5-Class1-T6易证g(x)=0g(x)=0

压缩映射

f(x) is contraction     A(0,1),x1,x2f(x1)f(x2)Ax1x2\begin{gathered} f(x) \text{ is contraction } \iff \\ \exists A\in (0,1),\forall x_1,x_2 \\ \vert f(x_1)-f(x_2) \vert \le A\vert x_1-x_2 \vert \end{gathered}

Banach不动点定理

f(x) is contraction,{xn} s.t. xn+1=f(xn)    limnxn=XX is the unique fixed point of f\begin{gathered} f(x) \text{ is contraction},\{ x_n \} \ s.t.\ x_{n+1}=f(x_n) \\ \implies \lim_{n \to \infty} x_n=X\land X \text{ is the unique fixed point of }f \end{gathered}

考虑任意数列 {xn}\{ x_n \},则

n<mxmxn=i=nm1xi+1xii=1m1x1x2An+i1x1x2An1A0\begin{gathered} \forall n<m \\ \vert x_m-x_n \vert =\sum _{i = n} ^{m-1} \vert x_{i+1}-x_i \vert \\ \le \sum _{i = 1} ^{m-1} \vert x_1-x_2 \vert A^{n+i-1} \\ \le \vert x_1-x_2 \vert \dfrac{A^n}{1-A} \\ \to 0 \end{gathered}

于是xnx_n收敛,且同时取极限能看出极限是不动点.

考虑如果由两个不动点X1,X2X_1,X_2,那么 X1X2Af(X1)f(X2)=AX1X2\vert X_1-X_2 \vert \le A\vert f(X_1)-f(X_2) \vert=A\vert X_1-X_2 \vert,可以得到X1=X2X_1=X_2,得证.

{fC2[a,b],f(x)δ>0,f(x)(0,M),f(X)=0x1(X,b),xn+1=xnf(xn)f(xn)    {limnxn=XN,n>N    (xnX)(xn1X)2\begin{gathered} \begin{cases} f\in C^2[a,b],f'(x)\ge \delta>0,f''(x)\in (0,M),f(X)=0 \\ x_1\in (X,b),x_{n+1}=x_n-\dfrac{f(x_n)}{f'(x_n)} \end{cases} \\ \implies \begin{cases} \lim_{n \to \infty} x_n=X \\ \exists N,n>N \implies (x_n-X)\sim (x_{n-1}-X)^2 \end{cases} \end{gathered}

就是牛顿迭代

显然f(x)>0f'(x)>0得到ff增.

尝试证明 xn>Xx_n>X,即f(xn)>0f(x_n)>0.

L' Hopital

{f,gC1(a,b),g(x)0limxa+f(x)=limxa+g(x)=0    limxa+f(x)g(x)=limxa+f(x)g(x)\begin{gathered} \begin{cases} f,g \in C^1(a,b),g'(x)\ne 0 \\ \lim_{x \to a^+} f(x)=\lim_{x \to a^+} g(x)=0 \end{cases} \\ \implies \lim_{x \to a^+} \dfrac{f(x)}{g(x)} =\lim_{x \to a^+} \dfrac{f'(x)}{g'(x)} \end{gathered}

不妨设f(a)=g(a)=0f(a)=g(a)=0

然后你可以对一个[a,a+b2][a,\dfrac{a+b}{2}]用柯西中值,有

ξ(a,x)s.t.f(x)f(a)g(x)g(a)=f(ξ)g(ξ)limxa+f(x)g(x)=limxa+f(ξ)g(ξ)=limξa+f(ξ)g(ξ)\begin{gathered} \exists \xi\in (a,x) \\ s.t.\\ \dfrac{f(x)-f(a)}{g(x)-g(a)}=\dfrac{f'(\xi)}{g'(\xi)} \\ \therefore \lim_{x \to a^+} \dfrac{f(x)}{g(x)} =\lim_{x \to a^+} \dfrac{f'(\xi)}{g'(\xi)}=\lim_{\xi \to a^+} \dfrac{f'(\xi)}{g'(\xi)} \end{gathered}

[think] 为什么 limxa+f(x)=limξa+f(ξ)\lim_{x \to a^+} f(x)=\lim_{\xi \to a^+} f(\xi)ξ(a,x)\xi\in (a,x).因为你转换到数列上,就是对任意xx趋近到aa的序列你可以找到一个ξ\xi趋近到aa的序列.

你还可以对无穷区间用:对(a,+)(a,+\infty),考虑复合 1x\dfrac1x

limx+f(x)g(x)=limx0f(1x)g(1x)=limx01x2f(1x)1x2g(1x)=limx+f(x)g(x)\begin{gathered} \lim_{x \to +\infty} \dfrac{f(x)}{g(x)}=\lim_{x \to 0} \dfrac{f(\dfrac{1}{x} )}{g(\dfrac{1}{x})} \\ =\lim_{x \to 0} \dfrac{-\dfrac{1}{x^2} f'(\dfrac{1}{x})}{-\dfrac{1}{x^2} g'(\dfrac{1}{x} )} \\ =\lim_{x \to +\infty} \dfrac{f'(x)}{g'(x)} \end{gathered}

另外,考虑若g,fg,f\to \infty,可以

gf=1f1g=ff2gg2=g2ff2g    fg=fg\begin{gathered} \dfrac{g}{f} =\dfrac{\dfrac{1}{f} }{\dfrac{1}{g} } = \dfrac{\dfrac{f'}{f^2} }{\dfrac{g'}{g^2} }=\dfrac{g^2f'}{f^2g'} \\ \implies \dfrac{f'}{g'} =\dfrac{f}{g} \end{gathered}

于是也可以对 \dfrac{\infty}{\infty}用?但问题在于这样要求预先知道 fg\dfrac{f}{g}极限存在,所以考虑下面的结论.

{f,gC1(a,b),g(x)0limxa+g(x)=+    limxa+f(x)g(x)=limxa+f(x)g(x)\begin{gathered} \begin{cases} f,g \in C^1(a,b),g'(x)\ne 0 \\ \lim_{x \to a^+} g(x)=+\infty \end{cases} \\ \implies \lim_{x \to a^+} \dfrac{f(x)}{g(x)} =\lim_{x \to a^+} \dfrac{f'(x)}{g'(x)} \end{gathered}

Solution 1

你先转化成对于任意趋近到a+a^+且满足g(x)g(x)单调的序列xx,然后有

limnf(xn)g(xn)=Stolzlimnf(xn)f(xn1)g(xn)g(xn1)=limn(xnxn1)f(ξn)(xnxn1)g(ξn),ξn(xn1,xn)=limnf(ξn)g(ξn)=limxa+f(x)g(x)\begin{gathered} \lim_{n \to \infty} \dfrac{f(x_n)}{g(x_n)} \\ \xlongequal{Stolz}\lim_{n \to \infty} \dfrac{f(x_n)-f(x_{n-1})}{g(x_n)-g(x_{n-1})} \\ =\lim_{n \to \infty} \dfrac{(x_n-x_{n-1})f'(\xi_n)}{(x_n-x_{n-1})g'(\xi_n)},\xi_n\in (x_{n-1},x_n) \\ =\lim_{n \to \infty} \dfrac{f'(\xi_n)}{g'(\xi_n)} \\ =\lim_{x \to a^+} \dfrac{f'(x)}{g'(x)} \end{gathered}

但是海涅要求对任意的序列xx怎么办呢?

考虑一个数列的一个性质:若数列的任何一个子列都有一个子列趋近到AA,则原数列也趋近到AA,这个反证是容易的.

那么而对于你这个任意数列的任意子列,显然都能再取一个子列使得g(x)g(x)单调,于是得证.

Solution 2

我们现在的问题是f(a)f(a)没有任何信息,所以想办法取c(a,a+δ)c\in (a,a+\delta),然后对x(a,c)x\in(a,c)[x,c][x,c] 用柯西证明

你有

f(x)f(c)g(x)g(c)=f(ξ)g(ξ)    f(x)g(x)=f(ξ)g(ξ)g(c)g(x)f(ξ)g(ξ)+f(c)g(x)\begin{gathered} \dfrac{f(x)-f(c)}{g(x)-g(c)} =\dfrac{f'(\xi)}{g'(\xi)} \\ \implies \dfrac{f(x)}{g(x)} =\dfrac{f'(\xi)}{g'(\xi)} -\dfrac{g(c)}{g(x)} \dfrac{f'(\xi)}{g'(\xi)} +\dfrac{f(c)}{g(x)} \end{gathered}

然后第二项,第三项显然趋近于00.

然后你同时趋向于aa就得证了.

[think] 但是我还是喜欢第一种. 因为它一直保持着fg\dfrac{f}{g}的结构()

{fC1(a,+),α>0limx+(αf(x)+xf(x))=β    limx+f(x)=αβ\begin{gathered} \begin{cases} f\in C^1(a,+\infty),\alpha>0 \\ \lim_{x \to +\infty} (\alpha f(x)+xf'(x))=\beta \\ \end{cases} \\ \implies \lim_{x \to +\infty} f(x)=\dfrac{\alpha}{\beta} \end{gathered}
limx+f(x)=limx+f(x)xαxα=limx+f(x)αxα1+f(x)xααxα1=αβ\begin{gathered} \lim_{x \to +\infty}f(x) \\ =\lim_{x \to +\infty}\dfrac{f(x)x^\alpha}{x^\alpha} \\ =\lim_{x \to +\infty}\dfrac{f(x)\alpha x^{\alpha-1}+f'(x)x^\alpha}{\alpha x^{\alpha-1}} \\ =\dfrac{\alpha}{\beta} \end{gathered}

如果你好奇如何注意到,那么考虑解微分方程用的那个积分因子,对P(x)f(x)+f(x)P(x)f(x)+f'(x)应该使用eP(x)dxe^{\int P(x)dx}

Class 16

Taylor's Formula With Peano Reminder

f(x)=i=0nf(i)(x0)i!(xx0)i+o((xx0)n)\begin{gathered} f(x)=\sum _{i = 0} ^{n} \dfrac{f^{(i)}(x_0)}{i!} (x-x_0)^i+o((x-x_0)^n) \end{gathered}

减过来变成

f(x)i=0nf(i)i!(xx0)i(xx0)n\begin{gathered} \dfrac{f(x)-\sum _{i = 0} ^{n} \dfrac{f^{(i)}}{i!}(x-x_0)^i }{(x-x_0)^{n}} \end{gathered}

然后洛n1n-1次.

f(k)(x0)0f^{(k)}(x_0)\ne 0kk为满足条件的最小正整数,则若kk为奇数x0x_0不是极值点.若kk为偶数则x0x_0是极值点.

limxx0f(x)f(x0)(xx0)k=f(k)(x0)k!0\begin{gathered} \lim_{x \to x_0} \dfrac{f(x)-f(x_0)}{(x-x_0)^k} =\dfrac{f^{(k)}(x_0)}{k!} \ne 0 \end{gathered}

于是邻域内 f(x)f(x0)(xx0)k\dfrac{f(x)-f(x_0)}{(x-x_0)^k}f(k)(x0)f^{(k)}(x_0)同号.

Taylor's Formula With Lagrange/Cauchy Reminder

f(x)=i=0nf(i)i!(xx0)i+f(n+1)(ξ)(n+1)!(xx0)n+1\begin{gathered} f(x)=\sum _{i = 0} ^{n} \dfrac{f^{(i)}}{i!} (x-x_0)^i + \dfrac{f^{(n+1)}(\xi)}{(n+1)!} (x-x_0)^{n+1} \end{gathered} f(x)=i=0nf(i)i!(xx0)i+f(n+1)(ξ)n!(xx0)(xξ)n\begin{gathered} f(x)=\sum _{i = 0} ^{n} \dfrac{f^{(i)}}{i!} (x-x_0)^i + \dfrac{f^{(n+1)}(\xi)}{n!} (x-x_0)(x-\xi)^n \end{gathered}
let F(t,x)=i=0nf(i)(t)i!(xt)iδFδt=f(t)+i=1nf(i+1)(t)i!(xt)if(i)(t)(i1)!(xt)i1=f(n+1)(t)n!(xt)nf(x)Tn(x)=F(x,x)F(x0,x)=(xx0)δFδt(ξ)=f(n+1)n!(xx0)(xξ)nf(x)Tn(x)(xx0)n+1=F(x,x)F(x0,x)(xx0)n+1(x0x0)n+1=δFδt(ξ)(n+1)(ξx0)n    f(x)Tn(x)=f(n+1)(ξ)(n+1)!(xx0)n+1\begin{gathered} \text{let } F(t,x)=\sum _{i = 0} ^{n} \dfrac{f^{(i)}(t)}{i!} (x-t)^i \\ \dfrac{\delta F}{\delta t}=f'(t)+\sum _{i = 1} ^{n} \dfrac{f^{(i+1)}(t)}{i!} (x-t)^i-\dfrac{f^{(i)}(t)}{(i-1)!} (x-t)^{i-1} \\ =\dfrac{f^{(n+1)(t)}}{n!} (x-t)^n \\ f(x)-T_n(x) \\ =F(x,x)-F(x_0,x) \\ =(x-x_0)\dfrac{\delta F}{\delta t}(\xi) \\ =\dfrac{f^{(n+1)}}{n!} (x-x_0)(x-\xi)^n \\ \dfrac{f(x)-T_n(x)}{(x-x_0)^{n+1}} \\ =\dfrac{F(x,x)-F(x_0,x)}{(x-x_0)^{n+1}-(x_0-x_0)^{n+1}} \\ =\dfrac{\dfrac{\delta F}{\delta t} (\xi)}{(n+1)(\xi-x_0)^n} \\ \implies f(x)-T_n(x)= \dfrac{f^{(n+1)}(\xi)}{(n+1)!} (x-x_0)^{n+1} \end{gathered}
fC2[0,1],f(x)1,f(x)2    f(x)3\begin{gathered} f\in C^2[0,1],\vert f(x) \vert \le 1,\vert f''(x) \vert \le 2 \\ \implies \vert f'(x) \vert \le 3 \end{gathered}
f(a)=f(x)+f(x)(ax)+f(ξ)(ax)22\begin{gathered} f(a)=f(x)+f'(x)(a-x)+\dfrac{f''(\xi)(a-x)^2}{2} \end{gathered}

带入a=1a=100然后解方程,把范围带进去.

{fC3[a,+)limx+f(x)=A,limx+f(x)=B    limx+f(x)=limx+f(x)=limx+f(x)=0\begin{gathered} \begin{cases} f\in C^3[a,+\infty) \\ \exists \lim_{x \to +\infty} f(x)=A,\lim_{x \to +\infty} f'''(x)=B \\ \end{cases} \implies \\ \lim_{x \to +\infty} f'(x)= \lim_{x \to +\infty} f''(x)=\lim_{x \to +\infty} f'''(x)=0 \end{gathered}

考虑展开

f(x+d)=f(x)+f(x)d+f(x)d22+f(x+θd)d36\begin{gathered} f(x+d)=f(x)+f'(x)d+\dfrac{f''(x)d^2}{2} +\dfrac{f'''(x+\theta d)d^3}{6} \end{gathered}

然后随便取几个固定d=1,1,2d=1,-1,2,让xx到无穷,解方程就能解出来.

因为你导数条件在无穷所以你要先趋近无穷.

[think] 当我们试图拿几个数的多项式表示函数,那函数相关条件自然都变成这几个数的多项式的方程.

Class 17

不定积分

找反导数

F(x)=f(x)dx    F(x)=f(x)\begin{gathered} F(x)=\int f(x)dx \iff F'(x)=f(x) \end{gathered}
secdx\begin{gathered} \int \sec dx \end{gathered}
=cosxcos2xdx=dsinx(1+sinx)(1sinx)=12(11sinx+11+sinxdsinx)=12ln1+sinx1sinx+C=lnsecx+tanx+C\begin{gathered} =\int \dfrac{\cos x}{\cos^2 x}dx \\ =\int \dfrac{d\sin x}{(1+\sin x)(1-\sin x)} \\ =\int \dfrac{1}{2} (\dfrac{1}{1-\sin x} +\dfrac{1}{1+\sin x} d\sin x ) \\ =\dfrac{1}{2} \ln \vert \dfrac{1+\sin x}{1-\sin x} \vert +C \\ =\ln \vert \sec x+\tan x \vert +C \end{gathered}

分部积分

udv=uvvdu\begin{gathered} \int udv=uv-\int vdu \end{gathered}

求导的乘法法则,同时积分一下.

(1x1x2)exdx\begin{gathered} \int (\dfrac{1}{x} -\dfrac{1}{x^2} )e^xdx \end{gathered}
=exxdx+d(1x)ex=exxdx+exxexx+C=exx+C\begin{gathered} =\int \dfrac{e^x}{x} dx+\int d(\dfrac{1}{x} )e^x \\ =\int \dfrac{e^x}{x} dx+\dfrac{e^x}{x} -\int \dfrac{e^x}{x}+C \\ =\dfrac{e^x}{x} +C \end{gathered}
sec3xdx\begin{gathered} \int \sec^3 x dx \end{gathered}
sec3xdx=secxdtanx=secxtanxtan2xsecx=secxtanxsec3xdx+secxdx    sec3xdx=12(secxtanx+secx)=12(secxtanx+ln(secx+tanx))+C\begin{gathered} \int \sec^3 xdx=\int \sec x d\tan x \\ =\sec x\tan x-\int \tan^2 x\sec x \\ =\sec x\tan x-\int \sec^3 xdx+\int \sec xdx \\ \implies \int \sec^3 xdx=\dfrac{1}{2} (\sec x\tan x+\int \sec x) \\=\dfrac{1}{2} (\sec x\tan x+\ln (\sec x+\tan x))+C \end{gathered}

换元积分

换就不好了......

3sinx+cosxsinx+2cosxdx\begin{gathered} \int \dfrac{3\sin x+\cos x}{\sin x+2\cos x} dx \\ \end{gathered}
3sinx+cosx=(sinx+2cosx)(cosx2sinx)3sinx+cosxsinx+2cosxdx=xd(sinx+2cosx)sinx+2cosx=xlnsinx+2cosx\begin{gathered} 3\sin x+\cos x=(\sin x+2\cos x)-(\cos x-2\sin x) \\ \int \dfrac{3\sin x+\cos x}{\sin x+2\cos x} dx \\ =x-\int \dfrac{d(\sin x+2\cos x)}{\sin x+2\cos x} \\ =x-\ln \vert \sin x+2\cos x \vert \end{gathered}
xarctanx(1+x2)2dx\begin{gathered} \int \dfrac{x\arctan x}{(1+x^2)^2} dx \end{gathered}
=12arctanx(1+x2)2d(x2+1)=12arctanxd(11+x2)=arctanx2(1+x2)12dx(1+x2)2let x=tant=arctanx2(1+x2)12sec2xsec4xdx=arctanx2(1+x2)12sec2xsec4xdx=arctanx2(1+x2)122sin2t4\begin{gathered} =\dfrac{1}{2} \int \dfrac{\arctan x}{(1+x^2)^2} d(x^2+1) \\ =\dfrac{1}{2} \int \arctan x d(\dfrac{1}{1+x^2}) \\ =\dfrac{\arctan x}{2(1+x^2)}-\dfrac{1}{2} \int \dfrac{dx}{(1+x^2)^2} \\ \text{let } x=\tan t \\ =\dfrac{\arctan x}{2(1+x^2)}-\dfrac{1}{2} \int \dfrac{\sec^2 x}{\sec^4 x}dx \\ =\dfrac{\arctan x}{2(1+x^2)}-\dfrac{1}{2} \int \dfrac{\sec^2 x}{\sec^4 x}dx \\ =\dfrac{\arctan x}{2(1+x^2)}-\dfrac{1}{2} \dfrac{2-\sin2t}{4} \end{gathered}

然后换成xx就行了 懒得写了.

dx(x+1)2(x+2)3\begin{gathered} \int \dfrac{dx}{(x+1)^2(x+2)^3} \end{gathered}
let t=x+1x+2ans=(1t)5t2(1t)2dt\begin{gathered} \text{let } t=\dfrac{x+1}{x+2} \\ ans=\int \dfrac{(1-t)^5}{t^2(1-t)^2}dt \end{gathered}

然后拆了做就行了.

就是分母是两个一次多项式的幂的积你换元它们的比.

Class 18

有理分式积分

P(x),degP(x)<deg(iQi(x)),degQi(x)2Ri(x),degRi(x)<degQi(x)s.t.P(x)i=1kQi(x)=i=1nRi(x)Qqi(x)ki\begin{gathered} \forall P(x),\deg P(x)<\deg (\prod_i Q_i(x)),\deg Q_i(x)\le 2 \\ \exists R_i(x),\deg R_i(x)<\deg Q_i(x) \\ s.t.\\ \dfrac{P(x)}{\prod_{i=1}^k Q_i(x)} =\sum _{i = 1} ^{n} \dfrac{R_i(x)}{Q_{q_i}(x)^{k_i}} \end{gathered}

待定系数,启动!

左边有degP\deg P个方程.

右边有degQ\deg Q个未知数.

结束.

注意分解的时候如果有若干个相同的QiQ_i,则拆成Qik(x)Q_i^k(x),如出现

ax2+bx+c(x1)3\begin{gathered} \dfrac{ax^2+bx+c}{(x-1)^3} \end{gathered}

就可以拆成r1x1,r2(x1)2,r3(x1)3\dfrac{r_1}{x-1} ,\dfrac{r_2}{(x-1)^2},\dfrac{r_3}{(x-1)^3}去表示,riRr_i\in R.

所以有理分式你先分解,分解完了分母一次的和r(xx0)k\dfrac{r}{(x-x_0)^k}的直接秒,现在问题来到分母是二次的函数.

5x+6x2+x+1dx\begin{gathered} \int \dfrac{5x+6}{x^2+x+1} dx \end{gathered}

考虑我们会积1x2+a\dfrac1{x^2+a}的形式,于是你要把分子干成一次,分母干成没有一次,先换元分母t=x2+x+1t=x^2+x+1,dt=2x+1dt=2x+1.

=52dtt+72(x2+x+1)dx\begin{gathered} =\int \dfrac{\dfrac{5}{2} dt }{t} +\int\dfrac{7}{2(x^2+x+1)}dx \end{gathered}

然后右边配方即可.

那如果分母是二次项的幂呢?可以用一样的办法消掉分子上的一次项,然后还是三角换元,这样你变成积三角函数的幂了.

根式积分

去分母根号:

(xa)(xb)t=xaxbx2+bx+cx2+bx+c=t+x,x=t2cb2tax2+bx+1ax2+bx+c=xt+1\begin{gathered} \sqrt{ (x-a)(x-b) } \to t=\sqrt{ \dfrac{x-a}{x-b} } \\ \sqrt{ x^2+bx+c } \to \sqrt{x^2+bx+c}=t+x,x=\dfrac{t^2-c}{b-2t} \\ \sqrt{ ax^2+bx+1 } \to \sqrt{ ax^2+bx+c } = xt+1 \end{gathered}

Class 19

定积分

abf(x)=limT0i=1nf(ξi)ΔxiwhereT={t1tn},ti<ti+1,t0=a,tn=b,Δxi=titi1,ξ(ti1,ti)T=maxiΔxi\begin{gathered} \int_a^b f(x)=\lim_{\vert\vert T \vert\vert \to 0} \sum _{i = 1} ^{n} f(\xi_i)\Delta x_i \\ \text{where} \\ T=\{ t_1\ldots t_n \} ,t_i<t_{i+1},t_0=a,t_n=b, \\ \Delta x_i=t_i-t_{i-1},\xi\in (t_{i-1},t_i) \\ \vert\vert T \vert\vert =\max_i \Delta x_i \end{gathered}

极限存在则称为黎曼可积.

牛顿莱布尼茨公式

{f(x) is integrable on [a,b]F(x)C[a,b],D(a,b)f(x)=F(x)    abf(x)=F(b)F(a)\begin{gathered} \begin{cases} f(x) \text{ is integrable on }[a,b] \\ F(x)\in C[a,b],D(a,b) \\ f(x)=F'(x) \end{cases} \\ \implies \int_a^b f(x)=F(b)-F(a) \end{gathered}
F(b)F(a)=i=1nF(ti)F(ti1)=i=1n(titi1)f(ξi),ξi(ti1ti)=SlimT0F(b)F(a)=limT0S=abf(x)\begin{gathered} F(b)-F(a) \\ =\sum _{i = 1} ^{n} F(t_i)-F(t_{i-1}) \\ =\sum _{i = 1} ^{n} (t_i-t_{i-1})f(\xi_i),\xi_i\in (t_{i-1}-t_i) \\ =S \\ \lim_{\vert\vert T \vert\vert \to 0} F(b)-F(a)=\lim_{\vert\vert T \vert\vert \to 0} S=\int_a^b f(x) \end{gathered}

Class 20

积分练习

f(x)C[a,b],f(x)>0    1baabln(f(x))dxln(1baabf(x)dx)\begin{gathered} f(x)\in C[a,b],f(x)>0 \\ \implies \dfrac{1}{b-a} \int_a^b \ln(f(x))dx\le \ln(\dfrac{1}{b-a} \int_a^b f(x)dx) \end{gathered}

几何均值小于代数均值

我们两边任取相同的[a,b][a,b]的分割和取点,于是就是ln\ln的琴生,即

LHS=1bailn(f(ξi))(titi1)RHS=ln(1baif(ξi)(titi1))\begin{gathered} LHS=\dfrac{1}{b-a} \sum_i \ln(f(\xi_i)) (t_i-t_{i-1})\\ RHS=\ln(\dfrac{1}{b-a} \sum_i f(\xi_i)(t_i-t_{i-1})) \end{gathered}

Holder

{p,q>1,1p+1qf,g is integrable    abf(x)g(x)dx(abf(x)pdx)1p(abg(x)qdx)1q\begin{gathered} \begin{cases} p,q>1,\dfrac{1}{p} +\dfrac{1}{q} \\ f,g \text{ is integrable} \end{cases} \\ \implies \int_a^b f(x)g(x)dx\le (\int_a^b f(x)^pdx)^{\frac{1}{p} }(\int_a^b g(x)^qdx)^{\frac{1}{q} } \end{gathered}

感觉你直接弄到离散上也行啊.

要么就是抄带权柯西的步骤

f(x)g(x)(abf(x)pdx)1pabg(x)qdx)1q1pfp(x)(abf(x)pdx)+1qfq(x)(abg(x)qdx)\begin{gathered} \dfrac{f(x)g(x)}{(\int_a^b f(x)^pdx)^{\frac1p}\int_a^b g(x)^qdx)^{\frac1q}}\le \dfrac{1}{p} \dfrac{f^p(x)}{(\int_a^b f(x)^pdx)}+\dfrac{1}{q} \dfrac{f^q(x)}{(\int_a^b g(x)^qdx)} \end{gathered}

然后同时积分,结束.

f(x)D2[a,b],f(x)>0,f(x)0    f(x)2baabf(x)dx\begin{gathered} f(x)\in D^2[a,b],f''(x)>0,f(x)\le 0 \\ \implies f(x)\ge \dfrac{2}{b-a} \int_a^b f(x)dx \end{gathered}
f(x)=f(t)+f(t)(xt)+f(ξ)2(xt)2f(t)+f(t)(xt)(ba)f(x)abf(t)dt+abf(t)(xt)dt\begin{gathered} f(x)=f(t)+f'(t)(x-t)+\dfrac{f''(\xi)}{2} (x-t)^2 \\ \ge f(t)+f'(t)(x-t) \\ (b-a)f(x)\ge \int_a^bf(t)dt+\int_a^bf'(t)(x-t)dt \end{gathered}

只需证

abf(t)(xt)dtabf(t)dt\begin{gathered} \int_a^b f'(t)(x-t)dt\ge \int_a^b f(t)dt \end{gathered}

考虑

abf(t)(xt)dt=x(f(b)f(a))(bf(b)af(a))+abf(t)dt=(bx)f(b)(xa)f(a)+abf(t)dt\begin{gathered} \int_a^b f'(t)(x-t)dt \\ =x(f(b)-f(a))-(bf(b)-af(a))+\int_a^b f(t)dt \\ =-(b-x)f(b)-(x-a)f(a)+\int_a^b f(t)dt \end{gathered}

得证.

唉傻了,你考虑这个的意思其实就是,一个上凸的图形的面积面积比直接连起来的大.所以你琴声才是正解.

f(x)C[0,1]    limn01f(xn)dx=f(1)\begin{gathered} f(x)\in C[0,1] \\ \implies \lim_{n \to \infty} \int_0^1 f(\sqrt[ n ]{ x } )dx =f(1) \end{gathered}

你大概体会一下,nn很大的时候基本上xn\sqrt[n]{x}都很接近11,而在一个00的小邻域内离11比较远,所以能猜到答案是f(1)f(1),能交换.

01f(xn)=01nf(xn)dx+1n1f(xn)dx\begin{gathered} \int_0^1 f(\sqrt[ n ]{ x } )=\int_0^{\frac1n} f(\sqrt[ n ]{ x } )dx+\int_{\frac1n}^1f(\sqrt[ n ]{ x } )dx \\ \end{gathered}

因为ff有界,所以左边是00.右边用中值得到

=(11n)f(ξn)\begin{gathered} =(1-\dfrac{1}{n})f(\sqrt[n]{\xi}) \end{gathered}

因为你1ξn1nn=11\ge \sqrt[n]{\xi}\ge \sqrt[n]{\dfrac1n}=1,于是结束了.

原函数存在定理

f(x) is integrable on [a,b]    F(x)=axf(x)dxC[a,b]\begin{gathered} f(x) \text{ is integrable on } [a,b] \\ \implies F(x)=\int_a^x f(x)dx \in C[a,b] \end{gathered}

因为可积,设fM\vert f\vert \le M.

F(x0+h)F(x0)=x0x0+hf(x)dxhM\begin{gathered} \vert F(x_0+h)-F(x_0) \vert =\vert \int_{x_0}^{x_0+h} f(x)dx \vert \le hM \end{gathered}

显然连续.

fC[a,b]    F(x)=axf(x)dx s.t. F(x)=f(x)\begin{gathered} f\in C[a,b] \implies F(x)=\int_a^x f(x)dx \ s.t.\ F'(x)=f(x) \end{gathered}
F(x0)=limh0F(x0+h)F(x0)h=limh0x0x0+hf(x)dxh=limh0hf(ξ)h,ξ(x0,x0+h)=limh0f(ξ)=f(x0)\begin{gathered} F'(x_0)=\lim_{h \to 0} \dfrac{F(x_0+h)-F(x_0)}{h} \\ =\lim_{h \to 0} \dfrac{\int_{x_0}^{x_0+h}f(x)dx}{h} \\ =\lim_{h \to 0} \dfrac{hf(\xi)}{h},\xi\in (x_0,x_0+h) \\ =\lim_{h \to 0} f(\xi) \\ =f(x_0) \end{gathered}
g(x)DF(x)=g(0)g(x)f(x)dx    F(x)=g(x)f(x)\begin{gathered} g(x)\in D \\ F(x)=\int_{g(0)}^{g(x)}f(x)dx \\ \implies F'(x)=g'(x)f(x) \end{gathered}

换元,x=g(t)x=g(t),看成关于tt的复合函数求导.

fC1[a,b],f(a)=0    abf2(x)dx12(ba)2ab(f(x))2dx\begin{gathered} f\in C^1[a,b],f(a)=0 \\ \implies \int_a^b f^2(x)dx\le \dfrac{1}{2} (b-a)^2 \int_a^b (f'(x))^2 dx \end{gathered}
f(x)=axf(t)dt=ax1f(t)dt(ax12dt)12(axf2(t)dt)12(xa)12(abf2(t)dt)12\begin{gathered} f(x)=\int_a^x f'(t)dt \\ =\int_a^x 1\cdot f'(t)dt \\ \le (\int_a^x 1^2dt)^\frac12(\int_a^x f'^2(t)dt)^\frac12 \\ \le (x-a)^{\frac12}(\int_a^b f'^2(t)dt)^\frac12 \end{gathered}

同时平方再积分,结束.

另一个神秘做法是

ax(f(x)+c)20\begin{gathered} \int_a^x (f'(x)+c)^2\ge 0 \end{gathered}

展开后,让左边的式子关于cc最小得到一个不等式去做,但同样难以想到啊.

Class 21

一道例题

{f(x)C2[0,1]f(0)=f(1)=0x(0,1),f(x)0    01f(x)f(x)dx4\begin{gathered} \begin{cases} f(x)\in C^2[0,1] \\ f(0)=f(1)=0 \\ \forall x\in(0,1),f(x)\ne 0 \end{cases} \\ \implies \int_0^1 \vert \dfrac{f''(x)}{f(x)} \vert dx\ge 4 \end{gathered}

这个ff\dfrac{f''}{f}太鬼了,于是你直接把ff放掉.不妨设f(x)>0f(x)>0,f(x0)f(x_0)ff最大值,则

f(x)=f(x0)+(xx0)f(ξ)    {0=f(0)=f(x0)+x0f(ξ1)0=f(1)=f(x0)+(1x0)f(ξ2)\begin{gathered} f(x)=f(x_0)+(x-x_0)f'(\xi) \\ \implies \begin{cases} 0=f(0)=f(x_0)+x_0f'(\xi_1) \\ 0=f(1)=f(x_0)+(1-x_0)f'(\xi_2) \end{cases} \end{gathered}

于是

01f(x)f(x)dx1f(x0)01f(x)dx1f(x0)ξ1ξ2f(x)dx=1f(x0)f(ξ1)f(ξ2)=1f(x0)f(x0)x0f(x0)1x014\begin{gathered} \int_0^1 \vert \dfrac{f''(x)}{f(x)} \vert dx \\ \ge \dfrac{1}{f(x_0)} \vert \int_0^1 f''(x) \vert dx \\ \ge \dfrac{1}{f(x_0)} \vert \int_{\xi_1}^{\xi_2}f''(x) \vert dx \\ =\dfrac{1}{f(x_0)} \vert f'(\xi_1)-f'(\xi_2) \vert \\ =\dfrac{1}{f(x_0)} \vert \dfrac{f(x_0)}{x_0}-\dfrac{f(x_0)}{1-x_0} \vert \\ \ge \dfrac{1}{4} \end{gathered}

[think] todo

Im=0π2cosmxdx\begin{gathered} I_m=\int_0^\frac\pi2 \cos^m xdx \end{gathered}
Im=0π2cosxcosm1xdx=sinxcosm10π20π2sinx(m1)cosm2x(sinx)dx=(m1)0π2(1cos2x)cosm2xdx=(m1)Im2(m1)Im    Im=m1mIm2\begin{gathered} I_m=\int_0^\frac\pi2 \cos x\cos^{m-1}x dx \\ =\sin x \cos^{m-1} \vert^{\frac\pi2}_0- \int_0^\frac\pi2 \sin x(m-1) \cos^{m-2} x(-\sin x)dx \\ =(m-1)\int_0^\frac\pi2 (1-cos^2 x)\cos^{m-2}xdx \\ =(m-1)I_{m-2}-(m-1)I_m \\ \implies I_m=\dfrac{m-1}{m} I_{m-2} \end{gathered}

于是递推即可.

积分型泰勒

fCn+1    f(x)=Tn(f,x0,x)+x0xf(n+1)(t)(xt)ndt\begin{gathered} f\in C^{n+1} \\ \implies f(x)=T_n(f,x_0,x)+\int_{x_0}^x f^{(n+1)}(t)(x-t)^ndt \end{gathered}

注意到

x0xf(n)(t)(tx0)n1(n1)!dt=f(n)(t)(tx0)nn!x0xf(n+1)(t)(tx0)nn!dt\begin{gathered} \int_{x_0}^x f^{(n)}(t)\dfrac{(t-x_0)^{n-1}}{(n-1)!}dt \\ =f^{(n)}(t)\dfrac{(t-x_0)^n}{n!}-\int_{x_0}^x f^{(n+1)}(t)\dfrac{(t-x_0)^n}{n!} dt \end{gathered}

于是你直接递归的做一下:

f(x)f(x0)=i=1nf(i)(x)i!(1)i1(xx0)i+(1)nx0xf(n+1)(t)(tx0)nn!dt\begin{gathered} f(x)-f(x_0)=\sum _{i = 1} ^{n} \dfrac{f^{(i)}(x)}{i!} (-1)^{i-1}(x-x_0)^i + (-1)^n \int_{x_0}^x f^{(n+1)}(t)\dfrac{(t-x_0)^n}{n!} dt \end{gathered}

发现两个问题:求导在xx上,以及交错,所以你直接把x0x_0xx交换一下:

f(x0)f(x)=i=1nf(i)(x0)i!(xx0)ix0xf(n+1)(t)(tx0)nn!dt    f(x)=i=0nf(i)(x0)i!(xx0)i+x0xf(n+1)(t)(tx0)nn!dt\begin{gathered} f(x_0)-f(x)=-\sum _{i = 1} ^{n} \dfrac{f^{(i)}(x_0)}{i!} (x-x_0)^i -\int_{x_0}^x f^{(n+1)}(t) \dfrac{(t-x_0)^n}{n!} dt \\ \iff \\ f(x)=\sum _{i = 0} ^{n} \dfrac{f^{(i)}(x_0)}{i!} (x-x_0)^i+\int_{x_0}^x f^{(n+1)}(t) \dfrac{(t-x_0)^n}{n!} dt \end{gathered}

最后的积分部分还涉及一个对称的换元.

所以其实就是你一开始对x0x_0积分就对了.

从这里我们可以注意到,对最后一个式子用积分中值是柯西中值.

要得到拉格朗日余项,也对最后一个式子用积分中值f(x)g(x)dx\int f(x)g(x)dx的版本即可.

换元为什么对

也就是为什么可以说:

abf(x)dx=cdf(g(t))g(t)dt\begin{gathered} \int_a^b f(x)dx=\int_c^d f(g(t))g'(t)dt \end{gathered}

gg可逆,g(c)=a,g(d)=bg(c)=a,g(d)=b

如果条件是fC,gC1f\in C,g\in C^1,那么直接对两边的变上限积分求导即可.

如果条件是黎曼可积,你需要回归定义+拉格朗日中值得到gg'去证明.

Class 22

一道例题

01f(x)dx=1,01xf(x)dx=0    supx[0,1]f(x)2+1\begin{gathered} \int_0^1 f(x)dx=1,\int_0^1 xf(x)dx=0 \\ \implies \sup_{x\in [0,1]} \vert f(x) \vert \ge \sqrt 2+1 \end{gathered}

考虑反证,则 f(x)<2+1\vert f(x) \vert <\sqrt 2+1

1=01f(x)(2x1)dx1=01f(x)(2x1)dx<(2+1)012x1dx=(2+1)(21)=1\begin{gathered} -1=\int_0^1 f(x)(\sqrt 2x-1)dx \\ 1=\vert \int_0^1 f(x)(\sqrt 2x-1)dx \vert \\ <(\sqrt2+1)\int_0^1 \vert \sqrt2 x-1 \vert dx \\ =(\sqrt 2+1)(\sqrt 2-1) \\ =1 \end{gathered}

得证.

[think] 首先,01(ax+b)f(x)=b\int_0^1 (ax+b)f(x)=b包含了所有信息是你应该想到的.其次f(x)f(x)的最优形态看起来是最大值最小值的绝对值相等的二值的函数,所以这个放缩是有的.

可积性理论

按照这个路径:

  • 定义上和,下和
  • 固定分割,任意和介于上和,下和之间
  • 于是显然可积等价于上和极限与下和极限相等
  • 细分之后,上和不增下和不降
  • 于是我们发现对于一列确定的上和下和的分割TnT_n,如果它们收敛,那么对任意另一列SnS_n,我们把它们的分割并起来拼一个新的SnTkS_n\cup T_k一定收敛,而拼起来的过程中的值的变化是不超过 K=Tk2MSnK=\vert T_k \vert 2M\Vert S_n \Vert,MM为被积函数的界.于是你让SS去和满足K<ϵK<\epsilonTkT_k去拼得到一个新的SnTkS_n\cup T_k,显然这个也收敛.且这个和我们原来的SS每项差都小于ϵ\epsilon,所以原SS也收敛.
  • 于是你说明任意一列分割的上下和极限相等就所有的都相等.
  • 然后还有一个可积性的理论是任意值vvϵ\epsilon可以取TT,使得TT中最大值减最小值超过vv的区间长度和小于ϵ\epsilon:由这个推可积只要容易转化到wiΔx\sum w_i \Delta x的可积条件,而如果这个不成立,则存在v,ϵv,\epsilon,那你能找到两列差vv的就爆炸了.

第二积分中值

{g is decreasing on [a,b]g(x)0fR[a,b]    abf(x)g(x)dx=g(a)acf(x)dx\begin{gathered} \begin{cases} g \text{ is decreasing on } [a,b] \\ g(x)\ge 0 \\ f\in R[a,b] \end{cases}\\ \implies \int_a^b f(x)g(x)dx = g(a)\int_a^c f(x)dx \end{gathered}

首先我们定义F(x)=axf(t)dtF(x)=\int_a^x f(t)dt,我们就是要证明g(a)F(x)g(a)F(x)的最大值比左侧大,最小值比左侧小.

其实最大值是显然的.

考虑

abf(x)g(x)dx=i=1nxixi+1f(x)g(x)dx=i=1nxixi+1f(x)(g(x)g(xi))dx+i=1nxixi+1f(x)g(xi)dx\begin{gathered} \int_a^b f(x)g(x)dx \\ =\sum _{i = 1} ^{n} \int_{x_i}^{x_{i+1}}f(x)g(x)dx \\ =\sum _{i = 1} ^{n} \int_{x_i}^{x_{i+1}}f(x)(g(x)-g(x_i))dx \\ +\sum_{i=1}^n \int_{x_i}^{x_{i+1}}f(x)g(x_i)dx \end{gathered}

其中对第一项,考虑gg可积所以我们可以取分割是的g(x)g(xi)g(x)-g(x_i)的积分是任意小,然后你把f(x)f(x)放成他的界就说明这块最后是00.

对第二项转化成

todo

gg是增的情况是相同的,而结合可以证明:

{g is monotonic on [a,b]f(x)R[a,b]    c,abf(x)g(x)dx=g(a)acf(x)dx+g(b)cbf(x)dx\begin{gathered} \begin{cases} g \text{ is monotonic on } [a,b] \\ f(x)\in R[a,b] \end{cases} \\ \implies \exists c,\int_a^b f(x)g(x)dx=g(a)\int_a^cf(x)dx+g(b)\int_c^b f(x)dx \end{gathered}

g0(x)=g(x)g(a)g_0(x)=g(x)-g(a)(或者gg递减,取g0(x)=g(x)g(b)g_0(x)=g(x)-g(b))用上面的形式.

Class 23

f is decreasing on [0,2π]    nZ,02πf(x)sin(nx)dx0\begin{gathered} f \text{ is decreasing on } [0,2\pi] \\ \implies \forall n\in Z,\int_0^{2\pi} f(x)\sin(nx)dx\ge 0 \end{gathered}

直接用中值就是

f(0)(cosnt+cos0)+f(2π)(cosntcos2π)0\begin{gathered} f(0)(-\cos nt+\cos 0) \\ +f(2\pi)(\cos nt-\cos 2\pi) \\ \ge 0 \end{gathered}
x>0    c,xx+csin(t2)dt1x\begin{gathered} x>0 \implies \forall c,\vert \int_x^{x+c} \sin(t^2)dt \vert \le \dfrac{1}{x} \end{gathered}

你考虑t2t^2肯定要处理:

=xx+c1t2sin(t2)d(t2)=x2(x+c)212ssinsds\begin{gathered} =\int_x^{x+c} \dfrac{1}{\sqrt{t^2}} \sin(t^2)d(t^2) \\ =\int_{x^2}^{(x+c)^2}\dfrac{1}{2\sqrt s}\sin sds \\ \end{gathered}

然后用第二中值

=12xx2ξsinsds12x\begin{gathered} =\vert \dfrac{1}{2x} \int_{x^2}^\xi \sin sds \vert \\ \le \dfrac{1}{2x} \end{gathered}

Lebeisge Theorem

勒贝格测度下,RR上零测集等价于可以被总长为任意小的可数个区间覆盖.

因为勒贝格测度是完备的.任意一个区间覆盖的测度是有定义 的,空集是有定义的,于是完备化定义了中间的是00.

ff可积当且仅当ff有界且不连续点的集合测度为00.

Sol 1

定义函数一点的振幅为

w(x)=limd0supy1,y2[xd,x+d]f(y1)f(y2)\begin{gathered} w(x)=\lim_{d \to 0} \sup_{y_1,y_2\in[x-d,x+d]}\vert f(y_1)-f(y_2) \vert \end{gathered}

容易看出连续等价于w(x)=0w(x)=0.

(实际上是上极限减下极限,等于0就是只有一个极限).

于是考虑零测集的一组开覆盖I=I1,I2,I=\bigcup I_1,I_2,\ldots满足长度和小于任意ϵ1\epsilon_1.

那么对区间[a,b][a,b]的每个点,考虑

  • 若当前点在II中,则你取一个它的小邻域也在II中(II是开集).
  • 否则,当前点是连续点,则连续点满足w(x)=0w(x)=0是一个极限,所以存在一个dd使得N(x,d)N(x,d)上的振幅小于任意ϵ2\epsilon_2.我们取这样一个邻域N(x,d)N(x,d).

于是所有点对应的区间构成对[a,b][a,b]的无限开覆盖,于是它有有限子覆盖,于是你可以拿出有限个区间.

把这些区间的端点组成划分TT,考虑TT中的每个区间:

如果它被一个来自II的点的区间包含,那么这样的区间总长度小于II的总长度.而剩下的区间被第二种区间包含,每个振幅小于ϵ2\epsilon_2.

这样你的wΔx\sum w\Delta x就是第一类的加第二类的,第一类的是2Mϵ12M\epsilon_1,第二类是(ba)ϵ2(b-a)\epsilon_2,都是乘常数,所以可以任意小得证.这是充分.

对于必要性,那么如果ff真的可积,你考虑等价条件之一是对任意ϵ,δ\epsilon,\delta,可以找到分割TT使得振幅大于ϵ\epsilon的区间长度小于δ\delta.

于是你取ϵ=1n\epsilon=\dfrac1n,取δ=12n\delta=\dfrac1{2^n}再把这些区间并起来结束.

Sol 2

或者我们看到振幅是上极限减下极限(包含自身),所以是上半连续的,所以振幅大于等于ϵ\epsilon的集合是一个闭集,所以覆盖零测集的一组集合可以取有限覆盖来覆盖大于等于ϵ\epsilon的.后面的仍然一样走.

Sol 3

老师上课讲的:

首先我们需要一个引理:若D={xw(x)0}D=\{x \vert w(x)\ne 0\}有一个可数开区间覆盖且是零测集,则

ϵ,δ,x[a,b]D,y[a,b],xy<δ    f(x)f(y)<ϵ\begin{gathered} \forall \epsilon,\exists \delta,\forall x\in [a,b]-D,y\in [a,b],\vert x-y \vert <\delta \\ \implies \vert f(x)-f(y) \vert <\epsilon \end{gathered}

证明考虑反证,则存在

cn[a,b]D,dn[a,b]limncndn=0limnf(cn)f(dn)>ϵ\begin{gathered} c_n\in [a,b]-D,d_n\in [a,b] \\ \lim_{n \to \infty} \vert c_n-d_n \vert =0 \\ \lim_{n \to \infty} \vert f(c_n)-f(d_n) \vert >\epsilon \end{gathered}

那我们取aa的一个收敛子列,则bb对应子列也收敛到相同点,而你用闭集减去若干开集得到的是闭集,所以这个共同的点落在[a,b]D[a,b]-D中,对这个连续点用连续性定义即推出矛盾.

有了引理之后,对任意分割TT满足T<δ\Vert T\Vert<\delta,考虑TT的每个小区间把和K=[ab]DK=[a-b]-D的交是否为空集分类.为空的集合完全包含于覆盖长度很小,而不为空集的有该区间内振幅一定小于2ϵ2\epsilon,于是随便取一下ϵ\epsilon把结果加一下就得证.