微分中值定理

定理

定理0(费马引理)

\[\begin{align*} &f:U(x_0)\to\mathbb R,\ f'(x_0)\text{ 存在},\ x_0\text{ 为局部极值点}\\ \implies &f'(x_0)=0 \end{align*} \]

定理1(Rolle)

\[\begin{align*} &f\in C[a,b],\ f\in D(a,b),\ f(a)=f(b)\\ \implies &\exists\xi\in(a,b),\ f'(\xi)=0 \end{align*} \]

定理2(Lagrange)

\[\begin{align*} &f\in C[a,b],\ f\in D(a,b)\\ \implies &\exists\xi\in(a,b),\ f(b)-f(a)=f'(\xi)(b-a) \end{align*} \]

\[\xi=a+\theta(b-a),\quad \theta\in(0,1) \]

端点不必可导:定理只要求 \(f\) 在开区间 \((a,b)\) 可导,\(f'(a)\)、\(f'(b)\) 可以不存在。
例:\(f(x) = 5\sqrt{x} - x\sqrt{x}\) 于 \([0,4]\),
\(f'(x) = \dfrac{5}{2\sqrt{x}} - \dfrac{3}{2}\sqrt{x}\) 在 \(x = 0\) 无定义,
但 MVT (\(Mean\ Value\ Theorem\))照样适用。

定理 3(柯西 Cauchy)

\[f,g\in C[a,b],\ f,g\in D(a,b),\ g'(x)\neq0,\forall x\in(a,b) \implies \exists\xi\in(a,b),\ \dfrac{f(b)-f(a)}{g(b)-g(a)}=\dfrac{f'(\xi)}{g'(\xi)} \]

定理 4(泰勒 Taylor)

\[f\in C^{n}(x_0),\ f\in C^{n+1}(I) \implies f(x)=\sum_{k=0}^{n}\dfrac{f^{(k)}(x_0)}{k!}(x-x_0)^k+R_n \]

其中

\[R_n=\dfrac{f^{(n+1)}(\xi)}{(n+1)!}(x-x_0)^{n+1} \quad(\text{Lagrange}) \]

\[R_n=\dfrac{f^{(n+1)}(\xi)}{n!}(x-x_0)(x-\xi)^n \quad(\text{Cauchy}) \]

\[R_n=o\left((x-x_0)^n\right) \quad(\text{Peano}) \]

\[R_n = \frac{f^{(n+1)}(\xi)}{n!\,p}(x - x_0)^p(x - \xi)^{n+1-p},\quad p>0 (Schlömilch–Roche) \]

\(p=n+1 \implies \text{Lagrange},\quad p=1 \implies \text{Cauchy}\)

定理 5(Darboux) 导函数具有介值性

\[f\in D(I),\ x_1,x_2\in I,\ \min\{f'(x_1),f'(x_2)\}<\lambda<\max\{f'(x_1),f'(x_2)\} \implies \exists \eta \text{ 介于 }x_1,x_2,\ f'(\eta)=\lambda \]

定理6(多元标量)

\[[x,y]\subset D\subset\mathbb R^n,\ f\in C^1(D) \implies \exists\theta\in(0,1),\ f(y)-f(x)=\nabla f(x+\theta(y-x))\cdot(y-x) \]

推论6.1

\[|f(y)-f(x)|\le \sup_{z\in[x,y]}\|\nabla f(z)\|\|y-x\| \]

\[D\text{ 凸},\ \|\nabla f\|\le M \implies |f(y)-f(x)|\le M\|y-x\| \]

定理7(向量值)

\[F:D\to\mathbb R^m,\ F\in \mathcal{D}(D) \implies \|F(y)-F(x)\|\le \sup_{z\in[x,y]}\|DF(z)\|_{\text{op}}\|y-x\| \]

推论7.1

\[D\text{ 连通},\ \nabla f\equiv 0 \implies \exists c\in\mathbb R,\forall x\in D,\ f(x)=c \]

\[D\text{ 连通},\ DF\equiv 0 \implies \exists \boldsymbol{c}\in\mathbb R^m,\forall x\in D,\ F(x)=\boldsymbol{c} \]

定理8(多元Taylor)

\[f(x+h)=\sum_{|\alpha|\le k}\frac{D^\alpha f(x)}{\alpha!}h^\alpha+\sum_{|\alpha|=k+1}\frac{D^\alpha f(x+\theta h)}{\alpha!}h^\alpha,\quad \theta\in(0,1) \]

二阶:

\[f(x+h)=f(x)+\nabla f(x)^T h+\frac12 h^T\nabla^2 f(x+\theta h)h,\quad \theta\in(0,1) \]

定理9(积分形式)

\[F(y)-F(x)=\int_{0}^{1}DF\big(x+t(y-x)\big)(y-x)\,\mathrm dt \]

证明

费马引理

不妨 \(x_0\) 为极大值点:\(\exists\delta>0,\ |h|<\delta \implies f(x_0+h)-f(x_0)\le 0\)

\[f'_+(x_0)=\lim_{h\to0^+}\frac{f(x_0+h)-f(x_0)}{h}\le 0,\quad f'_-(x_0)=\lim_{h\to0^-}\frac{f(x_0+h)-f(x_0)}{h}\ge 0 \]

\(f\) 可导 \(\implies f'_+(x_0)=f'_-(x_0)=f'(x_0)\),故 \(f'(x_0)=0\).■

罗尔定理

证法一(费马引理):

case1:

\[\max_{[a,b]}f=\min_{[a,b]}f \implies f\equiv\text{常量} \]

case2:

\[\max_{[a,b]}f\neq\min_{[a,b]}f,\ f(a)=f(b) \implies \max f \text{ 或 } \min f \text{ 在 }(a,b)\text{ 内取到} \]

由费马引理,极值点可导 \(\implies f'(\xi)=0\)

证法二:

设 \(\xi\in(a,b)\) 为最大值点:\(f(\xi+\Delta x)\le f(\xi)\)

\[\Delta x>0 \implies \frac{f(\xi+\Delta x)-f(\xi)}{\Delta x}\le 0 \]

\[\Delta x<0 \implies \frac{f(\xi+\Delta x)-f(\xi)}{\Delta x}\ge 0 \]

\(f'(\xi)=\lim\limits_{\Delta x\to0}\dfrac{f(\xi+\Delta x)-f(\xi)}{\Delta x}\) 存在

\[\implies f'(\xi)\le 0,\ f'(\xi)\ge 0 \implies f'(\xi)=0 \]

若最大值仅在端点,则最小值必在内部,同理。

证法三:(Darboux 反证)

\(f\in C[a,b],\ f\in \mathcal{D}(a,b),\ f(a)=f(b)\)
\(\text{假设 } \forall x\in(a,b),\,f'(x)\neq 0\stackrel{\text{Darboux}}{\implies} (\forall x\in(a,b),f'(x)>0)\lor(\forall x\in(a,b),f'(x)<0)\)
\(f'>0\implies f(b)>f(a),\quad f'<0\implies f(b)<f(a)\)
\(\text{均与 }f(a)=f(b)\text{ 矛盾}\implies \exists\xi\in(a,b),\,f'(\xi)=0\)

证法四:(区间套定理)

引理:设 \(h\in C[\alpha,\beta]\),\(\alpha<\beta\),\(h(\alpha)=h(\beta)\)。令 \(\ell=\dfrac{\beta-\alpha}{2}\),在 \([\alpha,\alpha+\ell]\) 上定义

\[\Phi(x):=h(x+\ell)-h(x). \]

则 \(\Phi\in C[\alpha,\alpha+\ell]\),且(注意 \(\alpha+2\ell=\beta\))

\[\Phi(\alpha+\ell)=h(\beta)-h(\alpha+\ell)=h(\alpha)-h(\alpha+\ell)=-\Phi(\alpha). \]

  • 若 \(\Phi(\alpha)=0\):取 \([\alpha',\beta']=[\alpha,\alpha+\ell]\);
  • 若 \(\Phi(\alpha)\neq0\):则 \(\Phi(\alpha)\) 与 \(\Phi(\alpha+\ell)=-\Phi(\alpha)\) 异号,由介值定理 \(\exists x_0\in(\alpha,\alpha+\ell)\) 使 \(\Phi(x_0)=0\),取 \([\alpha',\beta']=[x_0,x_0+\ell]\)。

两种情形都有:\([\alpha',\beta']\subseteq[\alpha,\beta]\),\(\alpha'<\beta'\),\(h(\alpha')=h(\beta')\),且

\[\beta'-\alpha'=\ell=\frac{\beta-\alpha}{2}\quad(\text{严格对半}). \]

下证罗尔定理:
若 \(f\) 在 \([a,b]\) 上恒等于 \(f(a)\),则任取 \(\xi\in(a,b)\) 有 \(f'(\xi)=0\),证毕。
否则 \(\exists p\in(a,b)\) 使 \(f(p)\neq f(a)\)(非恒常,而两端点值相等,故差异点必在内部)。不妨设 \(f(p)>f(a)\),否则对 \(-f\) 证明(\((-f)(a)=(-f)(b)\),且 \((-f)'(\xi)=0\iff f'(\xi)=0\))。取

\[v:=\frac{f(a)+f(p)}{2}\in\big(f(a),\,f(p)\big). \]

由介值定理:\(f(a)<v<f(p)\Rightarrow\exists a_1\in(a,p)\),\(f(a_1)=v\);\(f(p)>v>f(b)\Rightarrow\exists b_1\in(p,b)\),\(f(b_1)=v\)。于是

\[a<a_1<p<b_1<b,\qquad f(a_1)=f(b_1),\qquad [a_1,b_1]\subset(a,b). \]

对 \(n\ge1\),把引理用于 \(h=f|_{[a_n,b_n]}\)(其中\(f|_{[a_n,b_n]}:[a_n,b_n]\to\mathbb{R},\quad x\mapsto f(x)\))(有 \(f(a_n)=f(b_n)\)),得 \([a_{n+1},b_{n+1}]\)。归纳地:

\[[a_1,b_1]\supset[a_2,b_2]\supset\cdots,\qquad b_n-a_n=\frac{b_1-a_1}{2^{\,n-1}}\xrightarrow[n\to\infty]{}0,\qquad f(a_n)=f(b_n), \]

且所有区间都含于 \([a_1,b_1]\subset(a,b)\),故一切端点 \(a_n,b_n\in(a,b)\)。

由闭区间套定理:

\[\exists!\,\xi\in\bigcap_{n\ge1}[a_n,b_n]. \]

因 \(\xi\in[a_1,b_1]\subset(a,b)\),故 \(f\) 在 \(\xi\) 可导;又 \(a_n\le\xi\le b_n\) 且 \(b_n-a_n\to0\),得

\[|a_n-\xi|\le b_n-a_n\to0,\qquad |b_n-\xi|\le b_n-a_n\to0 . \]

在 \([a,b]\) 上定义

\[\rho(x):=\begin{cases}\dfrac{f(x)-f(\xi)-f'(\xi)(x-\xi)}{x-\xi},&x\neq\xi,\\[2mm]0,&x=\xi.\end{cases} \]

由 \(f\) 在 \(\xi\) 可导:\(\rho(x)\to0\ (x\to\xi)\);故 \(\rho(a_n)\to0,\ \rho(b_n)\to0\)。且对一切 \(x\) 成立恒等式

\[f(x)=f(\xi)+f'(\xi)(x-\xi)+\rho(x)(x-\xi). \]

两式相减(\(x=b_n\) 减 \(x=a_n\)),利用 \(f(b_n)=f(a_n)\):

\[0=f'(\xi)(b_n-a_n)+\rho(b_n)(b_n-\xi)-\rho(a_n)(a_n-\xi). \]

除以 \(b_n-a_n>0\),并注意 \(0\le\dfrac{b_n-\xi}{b_n-a_n}\le1,\ 0\le\dfrac{\xi-a_n}{b_n-a_n}\le1\):

\[\big|f'(\xi)\big|\le\big|\rho(b_n)\big|\cdot\frac{b_n-\xi}{b_n-a_n}+\big|\rho(a_n)\big|\cdot\frac{\xi-a_n}{b_n-a_n}\le|\rho(b_n)|+|\rho(a_n)|\xrightarrow[n\to\infty]{}0 . \]

故 \(f'(\xi)=0\)。\(\blacksquare\)

拉格朗日中值定理

证法一:(辅助函数)

令\(k=\dfrac{f(b)-f(a)}{b-a}\),设\(\ \varphi(x)=f(x)-f(a)-k(x-a)\)
\(\varphi(a)=\varphi(b)=0 \stackrel{\text{Rolle}}{\implies} \varphi'(\xi)=0 \implies f'(\xi)=k\)

证法二:(行列式)

\(\displaystyle\Phi(x)=\begin{vmatrix}1&1&1\\a&b&x\\f(a)&f(b)&f(x)\end{vmatrix}\)
\(\Phi(a)=\Phi(b)=0 \stackrel{\text{Rolle}}{\implies} (b-a)f'(\xi)-\big(f(b)-f(a)\big)=0\)

证法三:

令\(\psi(x)=f(x)-kx,\ k=\dfrac{f(b)-f(a)}{b-a}\)
\(\psi(a)=\psi(b) \stackrel{\text{Rolle}}{\implies} \psi'(\xi)=0 \implies f'(\xi)=k\)

证法四:(反证)

\(k:=\dfrac{f(b)-f(a)}{b-a},\ g(x):=f(x)-kx\)
\(g(b)-g(a)=0\implies g(a)=g(b)\)

反设
\(\forall x\in(a,b),\ g'(x)\neq 0\)

\(\exists\xi_1<\xi_2,\ g'(\xi_1)g'(\xi_2)<0\stackrel{\text{Darboux}}{\implies}\exists\eta\in(\xi_1,\xi_2),\ g'(\eta)=0 \Rightarrow\bot\)
\(\implies \operatorname{sgn}(g')\text{ 于 }(a,b)\text{ 恒定}\)
\({Without\ Loss\ Of\ Generality}\) \(g'(x)>0,\ \forall x\in(a,b)\)

\(lemma:\)\(g\in C[u,v],\ \forall t\in[u,v),g'(t)>0 \implies g(v)\ge g(u)\)

\(\forall\varepsilon>0,\ A_\varepsilon:=\{t\in[u,v]\mid g(t)-g(u)+\varepsilon(t-u)\ge0\}\)
\(A_\varepsilon\text{ 非空闭集}\ (u\in A_\varepsilon)\implies \exists c=\max A_\varepsilon\)
\(c<v \implies \exists\delta>0,\forall t\in(c,c+\delta],\ g(t)>g(c)\)
\(g(t)-g(u)+\varepsilon(t-u)=\big(g(t)-g(c)\big)+\big(g(c)-g(u)+\varepsilon(c-u)\big)+\varepsilon(t-c)>0\)
\(\implies t\in A_\varepsilon,\ t>c\),矛盾
\(\implies c=v,\ g(v)\ge g(u)-\varepsilon(v-u)\)
\(\varepsilon\to0^+\implies g(v)\ge g(u)\)

\(\forall a<a_0<x<b:\)
\([a_0,x]:\ g(x)\ge g(a_0),\quad [x,b]:\ g(b)\ge g(x)\)
\(a_0\to a^+,\ g\in C[a,b]\implies g(a)\le g(x)\le g(b),\ \forall x\in[a,b]\)
\(g(a)=g(b)\implies g\equiv {C}\implies g'\equiv0 \Rightarrow\ 矛盾\)
\(\implies\exists\xi\in(a,b),\ g'(\xi)=0\implies f'(\xi)-k=0\implies f'(\xi)=\dfrac{f(b)-f(a)}{b-a}\quad\blacksquare\)

柯西中值定理

证法一:(辅助函数)

\(k:=\dfrac{f(b)-f(a)}{g(b)-g(a)},\psi(x):=f(x)-f(a)-k(g(x)-g(a))\)
\(\psi(a)=\psi(b)=0\stackrel{\text{Rolle}}{\implies}\psi'(\xi)=0\implies f'(\xi)=k g'(\xi)\)
■

证法二(行列式)

\(\displaystyle\Psi(x)=\begin{vmatrix}1&1&1\\g(a)&g(b)&g(x)\\f(a)&f(b)&f(x)\end{vmatrix}\)
\(\Psi(a)=\Psi(b)=0\stackrel{\text{Rolle}}{\implies}(g(b)-g(a))f'(\xi)=(f(b)-f(a))g'(\xi)\)
■

泰勒定理

证法1:Cauchy

条件:\(f\in C^n[a,x],\ f^{(n+1)}(t)\text{ 在 }(a,x)\text{ 存在}\)
目标:\(\exists\xi\in(a,x),\ R_n(x)=\dfrac{f^{(n+1)}(\xi)}{(n+1)!}(x-a)^{n+1}\)

\(\varphi(t):=R_n(t),\ \psi(t):=(t-a)^{n+1}\)

\[\varphi^{(k)}(a)=0,\ \psi^{(k)}(a)=0,\quad 0\le k\le n \]

\(\varphi,\psi\in C[a,x]\cap D(a,x),\ \psi'(t)=(n+1)(t-a)^n\neq0,\ t\in(a,x]\)

\[\frac{\varphi(x)}{\psi(x)} =\frac{\varphi'(\xi_1)}{\psi'(\xi_1)} =\frac{\varphi''(\xi_2)}{\psi''(\xi_2)} =\dots =\frac{\varphi^{(n+1)}(\xi)}{\psi^{(n+1)}(\xi)},\xi\in(a,x) \quad({n+1} 次Cauchy) \]

\(\varphi^{(n+1)}(t)=f^{(n+1)}(t),\ \psi^{(n+1)}(t)=(n+1)!\)

\[\frac{R_n(x)}{(x-a)^{n+1}}=\frac{f^{(n+1)}(\xi)}{(n+1)!} \implies R_n(x)=\frac{f^{(n+1)}(\xi)}{(n+1)!}(x-a)^{n+1} \]

■

证法2:\(f\in C^{n+1}[a,x]\)

基例 \(n=0\):

\[R_0(x)=f(x)-f(a)=\int_{a}^{x} f'(t) dt \]

\[\int_{a}^{x}\frac{f^{(1)}(t)}{0!}(x-t)^0 dt=\int_{a}^{x}f'(t)dt \]

\(\implies n=0\) 成立。

归纳假设:\(n=m\) 时,

\[R_m(x)=\int_{a}^{x}\frac{f^{(m+1)}(t)}{m!}(x-t)^m dt \]

\(n=m+1\):

\[\begin{aligned} R_{m+1}(x) &=f(x)-\sum_{k=0}^{m+1}\frac{f^{(k)}(a)}{k!}(x-a)^k \\ &=\left(f(x)-\sum_{k=0}^{m}\frac{f^{(k)}(a)}{k!}(x-a)^k\right)-\frac{f^{(m+1)}(a)}{(m+1)!}(x-a)^{m+1}\\ &=R_m(x)-\frac{f^{(m+1)}(a)}{(m+1)!}(x-a)^{m+1} \end{aligned} \]

对 \(\displaystyle R_m(x)=\int_{a}^{x}\frac{f^{(m+1)}(t)}{m!}(x-t)^m dt\) 分部积分:

\[u:=f^{(m+1)}(t),\quad dv:=\frac{1}{m!}(x-t)^m dt \]

\[du:=f^{(m+2)}(t)dt,\quad v:=-\frac{(x-t)^{m+1}}{(m+1)!} \]

\[\begin{aligned} \int_{a}^{x} u\,dv &=uv\Big|_{a}^{x}-\int_{a}^{x} v\,du \\ &= \left.-f^{(m+1)}(t)\frac{(x-t)^{m+1}}{(m+1)!}\right|_{a}^{x} +\int_{a}^{x}\frac{f^{(m+2)}(t)}{(m+1)!}(x-t)^{m+1}dt \\ &= \frac{f^{(m+1)}(a)}{(m+1)!}(x-a)^{m+1} +\int_{a}^{x}\frac{f^{(m+2)}(t)}{(m+1)!}(x-t)^{m+1}dt \end{aligned} \]

\[\implies \int_{a}^{x}\frac{f^{(m+2)}(t)}{(m+1)!}(x-t)^{m+1}dt =R_m(x)-\frac{f^{(m+1)}(a)}{(m+1)!}(x-a)^{m+1}=R_{m+1}(x) \]

由数学归纳法,\(\forall\,n\in\mathbb N\):

\[R_n(x)=\int_a^x \frac{f^{(n+1)}(t)}{n!}(x-t)^n dt \]

积分第一中值定理,\((x-t)^n\ge0\):

\[\int_a^x f^{(n+1)}(t)(x-t)^n dt =f^{(n+1)}(\xi)\int_a^x (x-t)^n dt,\quad \xi\in(a,x) \]

\[\int_a^x (x-t)^n dt=\frac{(x-a)^{n+1}}{n+1} \]

\[R_n(x)=\frac{f^{(n+1)}(\xi)}{n!}\cdot\frac{(x-a)^{n+1}}{n+1} =\frac{f^{(n+1)}(\xi)}{(n+1)!}(x-a)^{n+1} \]

■

证法3:Peano余项

\(f^{(n)}(a)\) 存在
目标:\(R_n(x)=o\left((x-a)^n\right),\ x\to a\)

\[R_n(x)=f(x)-\sum_{k=0}^n\frac{f^{(k)}(a)}{k!}(x-a)^k \]

\[L=\lim_{x\to a}\frac{R_n(x)}{(x-a)^n} =\lim_{x\to a}\frac{f(x)-\displaystyle\sum_{k=0}^n\frac{f^{(k)}(a)}{k!}(x-a)^k}{(x-a)^n} \]

\(\dfrac{0}{0}\) 型,连续 \(n-1\) 次 \(L'Hospital\):

\[L=\lim_{x\to a}\frac{f^{(n-1)}(x)-f^{(n-1)}(a)-f^{(n)}(a)(x-a)}{n!\,(x-a)} \]

导数定义:\(\displaystyle\lim_{x\to a}\frac{f^{(n-1)}(x)-f^{(n-1)}(a)}{x-a}=f^{(n)}(a)\)

\[L=\frac{1}{n!}\big(f^{(n)}(a)-f^{(n)}(a)\big)=0 \]

\[\implies R_n(x)=o\big((x-a)^n\big)\quad(x\to a) \]

■

证法4:

\(f\in C^n[a,x],\ f^{(n+1)}(t)\) 在 \((a,x)\) 存在
固定 \(x\),定义常数 \(K\):

\[f(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^k+K\big(\psi(x)-\psi(a)\big) \]

构造辅助函数:

\[\Phi(t)=f(x)-\sum_{k=0}^{n}\frac{f^{(k)}(t)}{k!}(x-t)^k-K\big(\psi(x)-\psi(t)\big) \]

验证端点:

\[\Phi(a)=0,\quad \Phi(x)=0 \]

对 \(t\) 求导,

\[\frac{d}{dt}\left[\sum_{k=0}^{n}\frac{f^{(k)}(t)}{k!}(x-t)^k\right] =\sum_{k=0}^{n}\frac{d}{dt}\left[\frac{f^{(k)}(t)}{k!}(x-t)^k\right] \]

\[=\sum_{k=0}^{n}\left[\frac{f^{(k+1)}(t)}{k!}(x-t)^k-\frac{f^{(k)}(t)}{(k-1)!}(x-t)^{k-1}\right] =\frac{f^{(n+1)}(t)}{n!}(x-t)^n \]

于是

\[\Phi'(t)=-\frac{f^{(n+1)}(t)}{n!}(x-t)^n+K\psi'(t) \]

由 \(Rolle\) 定理,\(\exists\xi\in(a,x),\ \Phi'(\xi)=0\):

\[-\frac{f^{(n+1)}(\xi)}{n!}(x-\xi)^n+K\psi'(\xi)=0 \]

\[K=\frac{f^{(n+1)}(\xi)}{n!\,\psi'(\xi)}(x-\xi)^n \]

取 \({\psi(t)=(x-t)^p}\):

\[\psi(x)=0,\quad \psi(a)=(x-a)^p,\quad \psi'(\xi)=-p(x-\xi)^{p-1} \]

代入得到一般余项,即

\[K=\frac{f^{(n+1)}(\xi)}{n!\cdot\big(-p(x-\xi)^{p-1}\big)}(x-\xi)^n =-\frac{f^{(n+1)}(\xi)}{n!\,p}(x-\xi)^{n-p+1} \]

回到

\[f(x)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}(x-a)^k + K\big(\psi(x)-\psi(a)\big) \]

\(\psi(x)=0,\psi(a)=(x-a)^p\),于是

\[\begin{aligned} R_n(x)=K\big(\psi(x)-\psi(a)\big) &=K\big(0-(x-a)^p\big) \\ &=\left(-\frac{f^{(n+1)}(\xi)}{n!\,p}(x-\xi)^{n-p+1}\right)\cdot\big(-(x-a)^p\big) \end{aligned} \]

一般余项(Schlömilch-Roche余项)

\[{R_n(x)=\frac{f^{(n+1)}(\xi)}{n!\,p}(x-a)^p(x-\xi)^{n-p+1},\quad \xi\in(a,x)} \]

  • \(p=n+1\):Lagrange余项
  • \(p=1\):Cauchy余项

高维微分中值

记号:

  • \(\boldsymbol{x},\boldsymbol{y}\in\mathbb R^n\);
  • \(\nabla f\):梯度;
  • \(DF\):Frechet导数(雅可比矩阵);
  • \(\|\cdot\|_{\text{op}}\):算子范数(诱导范数)

定理6

证明:\(f:\Omega\subset\mathbb R^n\to\mathbb R\),线段 \([\boldsymbol{x},\boldsymbol{y}]\subset\Omega\),\(f\) 在 \([\boldsymbol{x},\boldsymbol{y}]\) 连续,在 \((\boldsymbol{x},\boldsymbol{y})\) 内可微。

\[g(t):=f\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big),\quad t\in[0,1] \]

复合求导:

\[g'(t)=\nabla f\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big)\cdot (\boldsymbol{y}-\boldsymbol{x}) \]

\(g:[0,1]\to\mathbb R\) 满足一元Lagrange中值条件,\(\exists\theta\in(0,1)\),记 \(\boldsymbol{\xi}=\boldsymbol{x}+\theta(\boldsymbol{y}-\boldsymbol{x})\in(\boldsymbol{x},\boldsymbol{y})\):

\[g(1)-g(0)=g'(\theta) \]

\[{f(\boldsymbol{y})-f(\boldsymbol{x})=\nabla f(\boldsymbol{\xi})\cdot (\boldsymbol{y}-\boldsymbol{x})} \]

定理7

证明:设 \(F:\Omega\subset\mathbb R^n\to\mathbb R^m\),\([\boldsymbol{x},\boldsymbol{y}]\subset\Omega\),\(F\) 在 \([\boldsymbol{x},\boldsymbol{y}]\) 连续、在内部Frechet可微。
若 \(F(\boldsymbol{y})=F(\boldsymbol{x})\),不等式平凡成立。
若 \(F(\boldsymbol{y})\neq F(\boldsymbol{x})\),取单位向量(投影方向)

\[\boldsymbol{u}:=\frac{F(\boldsymbol{y})-F(\boldsymbol{x})}{\|F(\boldsymbol{y})-F(\boldsymbol{x})\|}\in\mathbb R^m,\quad \|\boldsymbol{u}\|=1 \]

构造标量一元函数(投影)

\[\varphi(t):=\boldsymbol{u}\cdot F\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big),\quad t\in[0,1] \]

\(\varphi:[0,1]\to\mathbb R\) 是一元可微函数,对 \(\varphi\) 使用一维Lagrange中值定理:\(\exists\theta\in(0,1),\ \boldsymbol{z}=\boldsymbol{x}+\theta(\boldsymbol{y}-\boldsymbol{x})\)

\[\varphi(1)-\varphi(0)=\varphi'(\theta) \]

\[\boldsymbol{u}\cdot\big(F(\boldsymbol{y})-F(\boldsymbol{x})\big)=\boldsymbol{u}\cdot DF(\boldsymbol{z})(\boldsymbol{y}-\boldsymbol{x}) \]

左边:

\[\boldsymbol{u}\cdot\big(F(\boldsymbol{y})-F(\boldsymbol{x})\big) =\frac{F(\boldsymbol{y})-F(\boldsymbol{x})}{\|F(\boldsymbol{y})-F(\boldsymbol{x})\|}\cdot\big(F(\boldsymbol{y})-F(\boldsymbol{x})\big) =\|F(\boldsymbol{y})-F(\boldsymbol{x})\| \]

右边由算子范数定义:

\[\big|\boldsymbol{u}\cdot DF(\boldsymbol{z})(\boldsymbol{y}-\boldsymbol{x})\big|\le \|\boldsymbol{u}\|\,\|DF(\boldsymbol{z})(\boldsymbol{y}-\boldsymbol{x})\| \le \|DF(\boldsymbol{z})\|_{\text{op}}\|\boldsymbol{y}-\boldsymbol{x}\| \]

合并:

\[{\|F(\boldsymbol{y})-F(\boldsymbol{x})\|\le \|DF(\boldsymbol{z})\|_{\text{op}}\|\boldsymbol{y}-\boldsymbol{x}\|} \]

定理9

证明:同样 \(F:\Omega\subset\mathbb R^n\to\mathbb R^m\),线段 \([\boldsymbol{x},\boldsymbol{y}]\subset\Omega\),\(DF(\cdot)\) 在 \([\boldsymbol{x},\boldsymbol{y}]\) 连续。
对任意单位向量 \(\boldsymbol{u}\in\mathbb R^m\),令路径 \(\boldsymbol{\gamma}(t)=\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\),

\[\varphi(t)=\boldsymbol{u}\cdot F(\boldsymbol{\gamma}(t)) \]

\(\varphi'(t)=\boldsymbol{u}\cdot DF(\boldsymbol{\gamma}(t))(\boldsymbol{y}-\boldsymbol{x})\),由一元牛顿–莱布尼茨公式:

\[\varphi(1)-\varphi(0)=\int_0^1 \varphi'(t)\,dt \]

\[\boldsymbol{u}\cdot\big(F(\boldsymbol{y})-F(\boldsymbol{x})\big)=\int_0^1 \boldsymbol{u}\cdot DF\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big)(\boldsymbol{y}-\boldsymbol{x})\,dt \]

对等式两端取绝对值,结合积分三角不等式与算子范数估计:

\[\big|\boldsymbol{u}\cdot\big(F(\boldsymbol{y})-F(\boldsymbol{x})\big)\big| \le \|\boldsymbol{y}-\boldsymbol{x}\|\int_0^1 \big\|DF\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big)\big\|_{\text{op}}\,dt \]

左边对所有单位向量 \(\boldsymbol{u}\) 取上确界:

\[\sup_{\|\boldsymbol{u}\|=1}\big|\boldsymbol{u}\cdot(F(\boldsymbol{y})-F(\boldsymbol{x}))\big|=\|F(\boldsymbol{y})-F(\boldsymbol{x})\| \]

最终:

\[{\|F(\boldsymbol{y})-F(\boldsymbol{x})\| \le \|\boldsymbol{y}-\boldsymbol{x}\|\int_0^1 \big\|DF\big(\boldsymbol{x}+t(\boldsymbol{y}-\boldsymbol{x})\big)\big\|_{\text{op}}\,dt} \]

■

达布定理(Darboux)

证明:
Step 1
\(f'(x_1)<0<f'(x_2)\Rightarrow\exists\xi\in(x_1,x_2):f'(\xi)=0\)

设 \(x_1<x_2\)。

\[f'(x_1)=\lim_{t\to x_1^+}\frac{f(t)-f(x_1)}{t-x_1}<0 \xRightarrow{\ \text{保号}\ } \exists x_1'\in(x_1,x_2):\frac{f(x_1')-f(x_1)}{x_1'-x_1}<0 \xRightarrow{\ x_1'-x_1>0\ } f(x_1')<f(x_1) \]

\[f'(x_2)=\lim_{t\to x_2^-}\frac{f(t)-f(x_2)}{t-x_2}>0 \xRightarrow{\ \text{保号}\ } \exists x_2'\in(x_1,x_2):\frac{f(x_2')-f(x_2)}{x_2'-x_2}>0 \xRightarrow{\ x_2'-x_2<0\ } f(x_2')<f(x_2) \]

\[f\in D[x_1,x_2]\Rightarrow f\in C[x_1,x_2] \xRightarrow{\ \text{Weierstrass}\ } \exists\xi\in[x_1,x_2]:\ f(\xi)=\min_{[x_1,x_2]}f=:m \]

\[f(x_1')<f(x_1)\Rightarrow m<f(x_1);\qquad f(x_2')<f(x_2)\Rightarrow m<f(x_2) \ \Longrightarrow\ \xi\notin\{x_1,x_2\}\ \Longrightarrow\ \xi\in(x_1,x_2) \]

\[\xi\ \text{为极小点},\ f\in D(\xi)\xRightarrow{\ \text{Fermat}\ } f'(\xi)=0\qquad\blacksquare \]

Step 2

\({Without\ Loss\ Of\ Generality}\) \(f'(x_1)<f'(x_2)\);\(\mu\in\{f'(x_1),f'(x_2)\}\) 取端点,平凡。下设 \(f'(x_1)<\mu<f'(x_2)\)。

\[F:=f-\mu\cdot\mathrm{id}\ \Longrightarrow\ F\in D(I),\quad F'=f'-\mu \]

\(\mathrm{id}(x)=x\) \((identity\ function)\)

\[F'(x_1)=f'(x_1)-\mu<0<f'(x_2)-\mu=F'(x_2)\]

\[\xRightarrow{\ \text{Step 1}\ } \exists\xi\in(x_1,x_2):\ F'(\xi)=0 \ \Longrightarrow\ f'(\xi)=\mu\qquad\blacksquare \]

推论:\(f'\) 无第一类间断点

反设 \(x_0\) 为 \(f'\) 的第一类间断点:\(\exists A=\lim\limits_{x\to x_0^-}f'(x),\ B=\lim\limits_{x\to x_0^+}f'(x)\in\mathbb R\),且 \(\bigl(A,B\bigr)\ne\bigl(f'(x_0),f'(x_0)\bigr)\)。

\(\exists L\in\{A,B\}:L\ne f'(x_0)\); \({Without\ Loss\ Of\ Generality}\) \(L=A\)(右侧同理)。取 \(\mu\) 严格介于 \(A\) 与 \(f'(x_0)\) 之间,则由极限保号性

\[\exists\delta>0:\ \forall x\in(x_0-\delta,x_0),\ \mu\ \text{严格介于}\ f'(x)\ \text{与}\ f'(x_0)\ \text{之间} \]

取定 \(x\in(x_0-\delta,x_0)\),由 Darboux:\(\exists\xi\in(x,x_0):f'(\xi)=\mu\) —— 与上句 \(\forall x\in(x_0-\delta,x_0)\) 的量词矛盾。\(\blacksquare\)

故 \(f'\) 的间断点只能是第二类(振荡型),如 \(f(x)=x^2\sin\frac1x\) 在 \(x=0\) 处。

应用

不等式

经典
1.\(\displaystyle \frac{x}{1+x}<\ln(1+x)<x,\quad x>0\)
取 \(f(t)=\ln(1+t)\),在 \([0,x]\) 上用Lagrange中值定理:

\[\exists\,\xi\in(0,x),\quad \ln(1+x)=\frac{x}{1+\xi} \]

2.\(e^{x}>1+x,\quad x\neq 0\)
分别讨论 \(x>0\) 与 \(x<0\),对 \(f(t)=e^t\) 在 \([0,x]\)(或 \([x,0]\))用Lagrange中值定理定号。
命题:\(\displaystyle e^{x}>1+x,\quad \forall\,x\neq 0\)

证明
令 \(f(t)=e^t\),则 \(f'(t)=e^t\),\(f(t)\) 在 \(\mathbb R\) 上可导。

  • 情形1:\({x>0}\)
    \(f(t)\) 在闭区间 \([0,x]\) 由 Lagrange 中值定理

\[\exists\,\xi\in(0,x),\quad f(x)-f(0)=f'(\xi)(x-0) \]

代入:

\[e^{x}-e^{0}=e^{\xi}\cdot x \implies e^{x}-1 = x e^{\xi} \]

由 \(0<\xi<x\),得 \(e^{\xi}>e^{0}=1\),又 \(x>0\),故

\[e^{x}-1 = x e^{\xi} > x\cdot 1 = x \]

即

\[e^{x}>1+x \]

情形2:\({x<0}\)
\(f(t)\) 在闭区间 \([x,0]\) 由 Lagrange 中值定理

\[\exists\,\xi\in(x,0),\quad f(0)-f(x)=f'(\xi)(0-x) \]

代入:

\[e^{0}-e^{x}=e^{\xi}(-x) \implies 1-e^{x} = -x e^{\xi} \]

由 \(x<0\),有 \(-x>0\);\(\xi<0 \implies 0<e^{\xi}<1\)。

\[1-e^{x}=(-x)e^{\xi} < (-x)\cdot 1 = -x \]

\[1-e^{x} < -x \]

\[1+x < e^{x} \]

综上,\(\forall\,x\neq 0\),有

\[{e^{x}>1+x} \]

3.\(|\sin a-\sin b|\le |a-b|,\quad |\arctan a-\arctan b|\le |a-b|,\ \forall a,b\in\mathbb R\)

4.0.记号

\[\Delta_h^2f:=f(x+h)+f(x-h)-2f(x),\qquad D_h:=\frac{\Delta_h^2f}{h^2} \]

P

\[f''>0\ \text{于}\ I,\ [x-h,x+h]\subset I\ \Longrightarrow\ \forall x,\ \forall h>0:\ \Delta_h^2f>0 \]

证明
对 \(f\) 分别于 \([x,x+h],\,[x-h,x]\) 应用 Lagrange 中值定理:

\[\begin{align*} f(x+h)-f(x)&=f'(c_1)h,\quad \exists\,c_1\in(x,x+h)\\ f(x)-f(x-h)&=f'(c_2)h,\quad \exists\,c_2\in(x-h,x) \end{align*} \]

两式相减:

\[f(x+h)+f(x-h)-2f(x)=\big[f'(c_1)-f'(c_2)\big]h \]

对 \(f'\) 在 \([c_2,c_1]\) 应用 Lagrange 中值定理:

\[\exists\,c\in(c_2,c_1),\quad f'(c_1)-f'(c_2)=f''(c)(c_1-c_2) \]

代入得

\[f(x+h)+f(x-h)-2f(x)=f''(c)(c_1-c_2)h \]

由 \(f''(c)>0,\ c_1>c_2,\ h>0\),故

\[f(x+h)+f(x-h)-2f(x)>0 \]

■

Q

\[f''(x)\ \text{存在}\ \Longrightarrow\ \lim_{h\to0}D_h=f''(x) \]

证明

\[\begin{aligned} &A:=f''(x)=\lim_{t\to0}\frac{f'(x+t)-f'(x)}{t}\\ &\rho(t):=f'(x+t)-f'(x)-At\ \Longrightarrow\ \frac{\rho(t)}{t}\to0\ (t\to0)\ \Longrightarrow\ \rho(t)=o(t)\\ &\varphi(t):=f(x+t)-f(x)-f'(x)t-\tfrac{A}{2}t^2\ \Longrightarrow\ \varphi(0)=0,\ \varphi'(t)=\rho(t)=o(t) \end{aligned} \]

\[\text{MVT}\ \Longrightarrow\ \varphi(h)=\varphi'(\theta)h,\quad |\theta|<|h| \]

\[|\varphi(h)|\le\underbrace{\frac{|\rho(\theta)|}{|\theta|}}_{\to0}\cdot|\theta|\cdot|h|\le\varepsilon|h|^2(|h|<\delta)\ \Longrightarrow\ \varphi(h)=o(h^2) \]

\[\begin{aligned} f(x+h)&=f(x)+f'(x)h+\tfrac{A}{2}h^2+o(h^2)\\ f(x-h)&=f(x)-f'(x)h+\tfrac{A}{2}h^2+o(h^2)\\ \Longrightarrow\ \Delta_h^2f&=Ah^2+o(h^2)\\ \Longrightarrow\ D_h&=A+\frac{o(h^2)}{h^2}\xrightarrow[h\to0]{}A=f''(x) \end{aligned} \]

■

P''

\[f''\ \text{在}\ (x-h,x+h)\ \text{存在}\ \Longrightarrow\ \exists\xi\in(x-h,x+h):\ D_h=f''(\xi) \]

证明

\[\begin{aligned} &A:=D_h,\quad \psi(t):=f(x+t)+f(x-t)-2f(x)-At^2,\quad t\in[0,h]\\[2pt] &\psi(0)=f(x)+f(x)-2f(x)=0,\quad \psi(h)=f(x+h)+f(x-h)-2f(x)-Ah^2=\Delta_h^2f-\tfrac{\Delta_h^2f}{h^2}h^2=0\\[2pt] &f''\ \text{存在}\Rightarrow f\ \text{可导}\Rightarrow f\in C\ \Longrightarrow\ \psi\in C[0,h]\cap D(0,h)\xRightarrow{\ \text{Rolle}\ }\exists\eta\in(0,h):\psi'(\eta)=0\\[2pt] &\psi'(t)=f'(x+t)-f'(x-t)-2At\ \Longrightarrow\ f'(x+\eta)-f'(x-\eta)=2A\eta\\[2pt] &\eta\in(0,h)\Rightarrow[x-\eta,x+\eta]\subset(x-h,x+h)\xRightarrow{\ \text{MVT on}\ f'\ }\exists\xi\in(x-\eta,x+\eta):\\[2pt] &\qquad f''(\xi)=\frac{f'(x+\eta)-f'(x-\eta)}{(x+\eta)-(x-\eta)}=\frac{2A\eta}{2\eta}=A=D_h,\quad x-h<x-\eta<\xi<x+\eta<x+h\\[2pt] &\Longrightarrow\ \frac{f(x+h)+f(x-h)-2f(x)}{h^2}=f''(\xi),\quad \xi\in(x-h,x+h)\qquad\blacksquare \end{aligned} \]

P'' ⇒ P

证明

\[\begin{aligned} f''>0\ \text{于}\ I &\xRightarrow{[x-h,x+h]\subset I}\ \forall t\in(x-h,x+h):f''(t)>0\\ &\xRightarrow{\ P''\ }\ \exists\xi\in(x-h,x+h):\ D_h=f''(\xi)\\ &\xRightarrow{\ f''(\xi)>0,\ h>0\ }\ \Delta_h^2f=D_hh^2=f''(\xi)h^2>0 \end{aligned} \]

■

P'' ∧ \(f''∈C(x)\) ⇒ Q

证明

\[\begin{aligned} &\big[f''\ \text{在}\ (x-h,x+h)\ \text{存在}\big]\wedge\big[f''\in C(x)\big]\\ &\xRightarrow{\ P''\ }\ D_h=f''(\xi_h),\quad \xi_h\in(x-h,x+h)\ \Longrightarrow\ |\xi_h-x|<h\\ &\xRightarrow{\ h\to0\ }\ \xi_h\to x\\ &\xRightarrow{\ f''\in C(x)\ }\ f''(\xi_h)\to f''(x)\\ &\Longrightarrow\ \lim_{h\to0}D_h=\lim_{h\to0}f''(\xi_h)=f''(x) \end{aligned} \]

■

Landau 不等式

命题:设 \(f\in C^2(\mathbb R)\),

\[M_k=\sup_{x\in\mathbb R}|f^{(k)}(x)|<\infty,\quad k=0,1,2 \]

则

\[M_1^2 \le 2M_0M_2 \]

证明
\(\forall\,x\in\mathbb R,\forall\,h>0\),由带Lagrange余项Taylor公式:

\[f(x+h)=f(x)+f'(x)h+\frac{f''(\xi_1)}{2}h^2,\quad \exists\,\xi_1\in(x,x+h) \]

\[f(x-h)=f(x)-f'(x)h+\frac{f''(\xi_2)}{2}h^2,\quad \exists\,\xi_2\in(x-h,x) \]

两式相减:

\[f(x+h)-f(x-h)=2hf'(x)+\frac{h^2}{2}\big(f''(\xi_1)-f''(\xi_2)\big) \]

\[2hf'(x)=f(x+h)-f(x-h)-\frac{h^2}{2}\big(f''(\xi_1)-f''(\xi_2)\big) \]

\[\begin{aligned} 2h|f'(x)| &\le |f(x+h)|+|f(x-h)|+\frac{h^2}{2}\big(|f''(\xi_1)|+|f''(\xi_2)|\big) \\ &\le 2M_0 + h^2 M_2 \end{aligned} \]

令 \(u=|f'(x)|\):

\[M_2 h^2 - 2u h + 2M_0 \ge 0,\quad \forall\,h>0 \]

若 \(M_2=0\),则 \(f''\equiv 0\),\(M_0=\sup_{x\in\mathbb R}|f(x)|<\infty\) (\(f\) 有界) \(\implies f\) 为常函数\(\implies\) \(f'\equiv 0\)
\(\implies\)\(M_1=0\),不等式成立。

若 \(M_2>0\) \(\implies \Delta\le 0\):

\[\Delta = (-2u)^2 - 4\cdot M_2\cdot 2M_0 \le 0 \]

\[4u^2 - 8M_0M_2 \le 0 \implies u^2 \le 2M_0M_2 \]

\[|f'(x)|^2 \le 2M_0M_2,\quad \forall\,x\in\mathbb R \]

取上确界:

\[M_1^2 \le 2M_0M_2 \]

■

恒等式

例1

\[{\arcsin x + \arccos x \equiv \frac{\pi}{2},\quad x\in[-1,1]} \]

证明
令 \(f(x)=\arcsin x+\arccos x,\ x\in[-1,1]\)。

\[f'(x)=\frac{1}{\sqrt{1-x^2}}-\frac{1}{\sqrt{1-x^2}}=0,\quad x\in(-1,1) \]

故 \(f(x)\) 在 \((-1,1)\) 上恒为常数。
代入特殊值 \(x=0\):

\[f(0)=\arcsin 0+\arccos 0=0+\frac{\pi}{2}=\frac{\pi}{2} \]

于是

\[\arcsin x+\arccos x=\frac{\pi}{2},\ x\in(-1,1) \]

由连续性,端点 \(x=\pm1\) 也成立。

\[\arcsin x + \arccos x \equiv \frac{\pi}{2},\quad x\in[-1,1] \]

■

例2

\[{\arctan x + \arctan \frac1x \equiv \frac{\pi}{2},\quad x>0} \]

证明
令 \(\displaystyle g(x)=\arctan x+\arctan\frac1x,\ x>0\)。

\[g'(x)=\frac{1}{1+x^2}+\frac{1}{1+\left(\frac1x\right)^2}\cdot\left(-\frac{1}{x^2}\right) =\frac{1}{1+x^2}-\frac{1}{1+x^2}=0 \]

\(g(x)\) 在 \((0,+\infty)\) 上恒为常数。
代入特殊值 \(x=1\):

\[g(1)=\arctan 1+\arctan 1=\frac{\pi}{4}+\frac{\pi}{4}=\frac{\pi}{2} \]

因此

\[\arctan x + \arctan\frac1x \equiv \frac{\pi}{2},\quad x>0 \]

■
补充:\(x<0\) 时,\(\displaystyle \arctan x+\arctan\frac1x = -\frac{\pi}{2}\)

例3

\[{\arctan 3x-\arctan x-\arctan\frac{2x}{1+3x^2}\equiv 0,\quad |x|<\frac{1}{\sqrt{3}}} \]

证明
令

\[f(x)=\arctan 3x-\arctan x-\arctan\frac{2x}{1+3x^2},\quad x\in\left(-\frac{1}{\sqrt{3}},\frac{1}{\sqrt{3}}\right) \]

\[(\arctan 3x)'=\frac{3}{1+9x^2},\quad (\arctan x)'=\frac{1}{1+x^2} \]

记 \(\displaystyle u=\frac{2x}{1+3x^2}\),

\[u'=\frac{2(1+3x^2)-2x\cdot 6x}{(1+3x^2)^2}=\frac{2-6x^2}{(1+3x^2)^2} \]

\[\big(\arctan u\big)'=\frac{u'}{1+u^2}= \frac{\dfrac{2-6x^2}{(1+3x^2)^2}}{1+\dfrac{4x^2}{(1+3x^2)^2}} \]

\[=\frac{2-6x^2}{(1+3x^2)^2+4x^2} =\frac{2-6x^2}{1+10x^2+9x^4} =\frac{2(1-3x^2)}{(1+9x^2)(1+x^2)} \]

\[\begin{aligned} f'(x)&=\frac{3}{1+9x^2}-\frac{1}{1+x^2}-\frac{2(1-3x^2)}{(1+9x^2)(1+x^2)}\\ &=\frac{3(1+x^2)-(1+9x^2)-2(1-3x^2)}{(1+9x^2)(1+x^2)}\\ &=\frac{3+3x^2-1-9x^2-2+6x^2}{(1+9x^2)(1+x^2)}\\ &=0 \end{aligned} \]

所以 \(f(x)\) 在区间 \(\left(-\dfrac{1}{\sqrt{3}},\dfrac{1}{\sqrt{3}}\right)\) 上恒为常数。

取 \(x=0\):

\[f(0)=\arctan0-\arctan0-\arctan0=0 \]

因此

\[\arctan 3x-\arctan x-\arctan\frac{2x}{1+3x^2}\equiv 0,\quad |x|<\frac{1}{\sqrt{3}} \]

■

中值点的渐近位置(统一常数 · Abian)

命题
设 \(f\) 在 \(a\) 处 \(n+2\) 阶可导,且 \(f^{(n+2)}(a)\neq 0\)。记 \(h=b-a\),带Lagrange余项的Taylor展开:

\[f(b)=\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}h^{k}+\frac{f^{(n+1)}(\xi)}{(n+1)!}h^{n+1},\quad \xi=a+\theta h,\ \theta\in(0,1) \]

则

\[\lim_{b\to a}\frac{\xi-a}{b-a}=\lim_{h\to 0}\theta=\frac{1}{n+2} \]

证明
记余项

\[R(h)=f(a+h)-\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}h^{k} \]

一方面,Lagrange形式:

\[R(h)=\frac{f^{(n+1)}(\xi)}{(n+1)!}h^{n+1}=\frac{f^{(n+1)}(a+\theta h)}{(n+1)!}h^{n+1} \]

将 \(f^{(n+1)}(a+\theta h)\) 在 \(a\) 处做一阶Peano展开:

\[f^{(n+1)}(a+\theta h)=f^{(n+1)}(a)+f^{(n+2)}(a)\cdot \theta h+o(h) \]

代入余项:

\[\begin{aligned} R(h)&=\frac{h^{n+1}}{(n+1)!}\Big[f^{(n+1)}(a)+\theta f^{(n+2)}(a)h+o(h)\Big]\\ &=\frac{f^{(n+1)}(a)}{(n+1)!}h^{n+1}+\frac{f^{(n+2)}(a)\cdot \theta}{(n+1)!}h^{n+2}+o(h^{n+2}) \end{aligned} \]

另一方面,直接对 \(f(a+h)\) 展开至 \(h^{n+2}\) (Peano余项):

\[f(a+h)=\sum_{k=0}^{n+2}\frac{f^{(k)}(a)}{k!}h^{k}+o(h^{n+2}) \]

分离余项:

\[R(h)=\frac{f^{(n+1)}(a)}{(n+1)!}h^{n+1}+\frac{f^{(n+2)}(a)}{(n+2)!}h^{n+2}+o(h^{n+2}) \]

两个表达式是同一个 \(R(h)\),\(h^{n+1}\) 项自动相等;对比 \(h^{n+2}\) 的系数:

\[\frac{f^{(n+2)}(a)\cdot \theta}{(n+1)!}=\frac{f^{(n+2)}(a)}{(n+2)!} \]

由条件 \(f^{(n+2)}(a)\neq 0\),两边约去:

\[\frac{\theta}{(n+1)!}=\frac{1}{(n+2)(n+1)!} \]

\[\theta=\frac{1}{n+2}+o(1),\quad h\to 0 \]

于是

\[\lim_{h\to0}\theta=\lim_{b\to a}\frac{\xi-a}{b-a}=\frac{1}{n+2} \]

特例

  • \(n=0\)(Lagrange中值定理):\(\displaystyle \lim_{b\to a}\theta=\frac12\)
  • \(n=1\)(二阶Taylor Lagrange余项):\(\displaystyle \lim_{b\to a}\theta=\frac13\)
  • \(n=2\)(三阶Taylor Lagrange余项):\(\displaystyle \lim_{b\to a}\theta=\frac14\)

推广:广义 Cauchy–Taylor 中值点渐近位置(Abian)
命题

\[\begin{aligned} &f,g\ \text{在}\ a\ \text{处}\ n+2\ \text{阶可导},\quad h:=b-a\\ &g^{(k)}(a)\neq0\ (k=0,1,\dots,n+1),\quad g^{(n+1)}(x)\neq0\ (x\in(a,b))\\ &\Delta(a):=f^{(n+2)}(a)g^{(n+1)}(a)-f^{(n+1)}(a)g^{(n+2)}(a)\neq0 \end{aligned} \]

\[\frac{f(b)-\displaystyle\sum_{k=0}^{n}\frac{f^{(k)}(a)}{k!}h^{k}}{g(b)-\displaystyle\sum_{k=0}^{n}\frac{g^{(k)}(a)}{k!}h^{k}}=\frac{f^{(n+1)}(\xi)}{g^{(n+1)}(\xi)},\quad \xi=a+\theta h,\ \theta\in(0,1) \]

\[\Longrightarrow\ \lim_{b\to a}\frac{\xi-a}{b-a}=\lim_{h\to0}\theta=\frac{1}{n+2} \]

证明:

\[\begin{aligned} &N(h):=f(a+h)-\sum_{k=0}^{n}\tfrac{f^{(k)}(a)}{k!}h^{k},\qquad M(h):=g(a+h)-\sum_{k=0}^{n}\tfrac{g^{(k)}(a)}{k!}h^{k}\\ &P:=f^{(n+1)}(a),\quad P':=f^{(n+2)}(a),\qquad Q:=g^{(n+1)}(a)\neq0,\quad Q':=g^{(n+2)}(a) \end{aligned} \]

\[\begin{aligned} N(h)&=\tfrac{P}{(n+1)!}h^{n+1}+\tfrac{P'}{(n+2)!}h^{n+2}+o(h^{n+2})\\ M(h)&=\tfrac{Q}{(n+1)!}h^{n+1}+\tfrac{Q'}{(n+2)!}h^{n+2}+o(h^{n+2}) \end{aligned} \]

\[\xRightarrow{\ Q\neq0\ }\ \frac{N(h)}{M(h)}=\frac{P+\tfrac{P'}{n+2}h+o(h)}{Q+\tfrac{Q'}{n+2}h+o(h)}=\frac{P}{Q}+\frac{\Delta(a)}{(n+2)Q^{2}}h+o(h)\tag{1} \]

\[\frac{N(h)}{M(h)}=\frac{f^{(n+1)}(a+\theta h)}{g^{(n+1)}(a+\theta h)}=\frac{P+P'\theta h+o(h)}{Q+Q'\theta h+o(h)}=\frac{P}{Q}+\theta\,\frac{\Delta(a)}{Q^{2}}h+o(h)\tag{2} \]

\[(1)=(2)\xRightarrow{\ \text{比对}\ h\ }\ \frac{\Delta(a)}{(n+2)Q^{2}}=\theta\,\frac{\Delta(a)}{Q^{2}}+o(1) \]

\[\xRightarrow{\ \Delta(a)\neq0\ }\ \theta=\frac{1}{n+2}+o(1)\ \xrightarrow{\ h\to0\ }\ \lim_{h\to0}\theta=\lim_{b\to a}\frac{\xi-a}{b-a}=\frac{1}{n+2} \]

■

根的存在性

例:\(f(x)=x^3-3x+a\) 在区间 \([-1,1]\)
求导:

\[f'(x)=3x^2-3=3(x^2-1) \]

当 \(x\in(-1,1)\),\(f'(x)<0\),\(f'(x)\) 在 \((-1,1)\) 内没有零点。

反证:
若 \(f(x)\) 在 \((-1,1)\) 内有两个相异零点 \(x_1<x_2\),则 \(\exists\xi\in(x_1,x_2)\subset(-1,1)\),使得 \(f'(\xi)=0\),矛盾。
\(\implies f(x)\) 在 \([-1,1]\) 至多 1 个零点。

存在性(介值定理)

\[\begin{cases} f(-1)=(-1)^3-3(-1)+a = 2+a\\ f(1)=1^3-3\cdot 1+a = -2+a \end{cases} \]

当 \(|a|<2\):
\(a>-2 \implies f(-1)=a+2>0\);
\(a<2 \implies f(1)=a-2<0\)。
\(f(-1)f(1)<0\),由介值定理,\(\exists c\in(-1,1),\ f(c)=0\)。

结合至多一根:\({|a|<2}\) 时,\(f(x)\) 在 \((-1,1)\) 存在唯一实根。

根数上界定理

命题:设 \(f\) 在 \(\mathbb R\) 上 \(n\) 阶可导。
若 \(f\) 有 \(n+1\) 个两两不同的零点:\(x_1<x_2<\dots<x_{n+1}\),
则:
\(f'\) 至少含有 \(n\) 个不同零点,
\(f''\) 至少含有 \(n-1\) 个不同零点,
\(\dots\)
\(f^{(n)}\) 至少含有 \(1\) 个零点。

逆否形式(核心结论,做题常用):

\[f^{(n)}(x) \neq 0,\ \forall x \in I \implies f(x)\text{ 在区间 }I\text{ 至多有 }n\text{ 个零点} \]

证明:
对 \(k\) 用数学归纳法,\(k=1,2,\dots,n\)。

基例 \(k=1\):
\(f(x_1)=f(x_2)=\dots=f(x_{n+1})=0\),共 \(n+1\) 个零点。
对每个区间 \([x_i,x_{i+1}],\ i=1,2,\dots,n\),由罗尔定理,\(\exists \xi_i\in(x_i,x_{i+1})\),使得 \(f'(\xi_i)=0\)。
\(\xi_1<\xi_2<\dots<\xi_n\) 互不相同,故 \(f'\) 至少有 \(n\) 个零点。

归纳假设:
设对 \(k=m\;(1\le m<n)\),\(f^{(m)}\) 至少有 \(n+1-m\) 个两两不同零点:

\[\eta_1<\eta_2<\dots<\eta_{n+1-m}. \]

归纳步骤:
对相邻零点区间 \([\eta_i,\eta_{i+1}]\) 再用罗尔定理,\(\exists \zeta_i\in(\eta_i,\eta_{i+1})\),满足

\[f^{(m+1)}(\zeta_i)=0,\quad i=1,2,\dots,(n+1-m)-1=n-m. \]

于是 \(f^{(m+1)}\) 至少有 \(n-m\) 个不同零点。

取 \(m=n-1\),得到 \(f^{(n)}\) 至少有 \(1\) 个零点。归纳完成。■

逆否命题

\[\big(\forall\,x\in I,\; f^{(n)}(x)\neq 0\big) \implies \big(f\text{ 在 }I\text{ 上至多 }n\text{ 个零点}\big) \]

实例:方程 \(e^x = ax^2+bx+c\)
令

\[f(x)=e^x - ax^2 - bx - c \]

求三阶导数:

\[f'(x)=e^x-2ax-b,\quad f''(x)=e^x-2a,\quad {f'''(x)=e^x} \]

\(\forall x\in\mathbb R,\ e^x>0 \implies f'''(x)\) 无零点。
套用上面定理:\(f\) 至多 \(3\) 个零点。
反证:假设存在4个相异实根,则反复罗尔推出 \(f'''\) 至少1个零点,与 \(e^x>0\) 矛盾。
\(\implies\) 方程 \(e^x=ax^2+bx+c\) 的实根个数不超过 3。

导数极限定理

\[f\in C\bigl(U(x_0)\bigr),\quad f\in D\bigl(\mathring U(x_0)\bigr),\quad \lim_{x\to x_0}f'(x)=A\in\mathbb R \ \Longrightarrow\ \exists f'(x_0)=A \]

证明

\[\begin{aligned} &\forall x>x_0:\ f\in C[x_0,x]\cap D(x_0,x)\xRightarrow{\ \text{Lagrange}\ }\exists\xi_x\in(x_0,x):\ \frac{f(x)-f(x_0)}{x-x_0}=f'(\xi_x)\\ &x\to x_0^+\Rightarrow\xi_x\to x_0^+,\ \xi_x\ne x_0\\ &\Longrightarrow\ f'_+(x_0)=\lim_{x\to x_0^+}\frac{f(x)-f(x_0)}{x-x_0}=\lim_{\xi_x\to x_0^+}f'(\xi_x)=A \end{aligned} \]

\[\begin{aligned} &\forall x<x_0:\ f\in C[x,x_0]\cap D(x,x_0)\xRightarrow{\ \text{Lagrange}\ }\exists\eta_x\in(x,x_0):\ \frac{f(x)-f(x_0)}{x-x_0}=f'(\eta_x)\\ &x\to x_0^-\Rightarrow\eta_x\to x_0^-,\ \eta_x\ne x_0\\ &\Longrightarrow\ f'_-(x_0)=\lim_{x\to x_0^-}\frac{f(x)-f(x_0)}{x-x_0}=\lim_{\eta_x\to x_0^-}f'(\eta_x)=A \end{aligned} \]

\[f'_+(x_0)=f'_-(x_0)=A\ \Longrightarrow\ \exists f'(x_0)=A\qquad\blacksquare \]

\(inference\):

\(\ \S1\)

\[\boxed{\ f'\ \text{的间断点必为第二类}\ }\quad\bigl(\text{第一类}\Rightarrow\text{极限存在}\xRightarrow{\ \text{导数极限定理}\ }f'\in C\Rightarrow\text{矛盾}\bigr) \]

\[\boxed{\ f'\ \text{单调}\Rightarrow f'\in C\ }\quad\bigl(\text{单调函数仅有第一类间断点}\bigr)\wedge\bigl(f'\ \text{无第一类间断点}\bigr) \]

\(\ \S2\)单点 \(f'(x_0)>0\) 不足以推出局部单调

\[f(x)=\begin{cases}\dfrac{x}{2}+x^{2}\sin\dfrac1x,&x\ne0\\[8pt]0,&x=0\end{cases} \]

\[f'(0)=\lim_{h\to0}\frac{f(h)-f(0)}{h}=\lim_{h\to0}\Bigl(\frac12+h\sin\frac1h\Bigr)=\frac12>0 \]

\[x\ne0:\quad f'(x)=\frac12+2x\sin\frac1x-\cos\frac1x \]

\[x'_n:=\frac{1}{2n\pi+\pi}\Rightarrow\sin\tfrac1{x'_n}=0,\ \cos\tfrac1{x'_n}=-1\Rightarrow f'(x'_n)=\frac12-(-1)=\frac32>0 \]

\[x''_n:=\frac{1}{2n\pi}\Rightarrow\sin\tfrac1{x''_n}=0,\ \cos\tfrac1{x''_n}=1\Rightarrow f'(x''_n)=\frac12-1=-\frac12<0 \]

\[x'_n,x''_n\to0\ \Longrightarrow\ \forall\delta>0,\ \exists x'_n,x''_n\in(-\delta,\delta):f'(x'_n)>0>f'(x''_n) \]

\[\Longrightarrow\ f'\ \text{在}\ U(0)\ \text{内变号}\xRightarrow{\ \text{Fermat/Darboux}\ }f\ \text{在}\ U(0)\ \text{内不单调}\qquad\blacksquare \]

\(\ \S3\)两个充分条件

\[\begin{aligned} &\text{(a)}\ f'(x_0)>0\ \wedge\ f'\in C(x_0)\xRightarrow{\ \text{保号性}\ }\exists\delta>0:\ f'>0\ \text{于}\ U(x_0,\delta)\xRightarrow{\ \text{MVT}\ }f\ \text{严格增于}\ U(x_0,\delta)\\[4pt] &\text{(b)}\ f'(x_0)>0\ \wedge\ f'\ \text{单调}\xRightarrow{\ \S1\ }\ f'\in C(x_0)\xRightarrow{\ \text{(a)}\ }\ f\ \text{严格增于}\ U(x_0,\delta) \end{aligned} \]

\[\begin{array}{ccc} f'(x_0)>0&\xrightarrow{\ \not\Rightarrow\ }&f\ \text{局部单调}\quad(\S2\ \text{反例})\\ f'(x_0)>0\ \wedge\ f'\in C(x_0)&\xrightarrow{\ \text{保号+MVT}\ }&f\ \text{局部严格增}\\ f'(x_0)>0\ \wedge\ f'\ \text{单调}&\xrightarrow{\ f'\ \text{单调}\Rightarrow f'\in C\ }&f\ \text{局部严格增} \end{array} \]

数列极限的存在性(Euler 常数)

\[\frac{h}{1+h}<\ln(1+h)<h\quad(h>0)\ \xrightarrow{\ h=\frac1n\ }\ \frac{1}{n+1}<\underbrace{\ln\left(1+\frac1n\right)}_{=\ln(n+1)-\ln n}<\frac1n\qquad(*) \]

\[H_n:=\sum_{k=1}^{n}\frac1k,\qquad x_n:=H_n-\ln n \]

\[\begin{aligned} &x_{n+1}-x_n=\frac{1}{n+1}-\bigl[\ln(n+1)-\ln n\bigr]\overset{(*)}{<}0\ \Longrightarrow\ \{x_n\}\searrow\\[4pt] &\ln n=\sum_{k=1}^{n-1}\ln\left(1+\frac1k\right)\overset{(*)}{<}\sum_{k=1}^{n-1}\frac1k=H_n-\frac1n\ \Longrightarrow\ x_n=H_n-\ln n>\frac1n>0\\[4pt]\Longrightarrow &\{x_n\}\ 收敛 \end{aligned} \]

Rolle定理条件

记号

\[\begin{aligned} &A:\ f\in C[a,b],\quad B:\ f\in D(a,b),\quad C:\ f(a)=f(b),\quad D:\ \exists\xi\in(a,b):\ f'(\xi)=0\\ &\textbf{Rolle}:\quad A\wedge B\wedge C\ \Longrightarrow\ D \end{aligned} \]

必要性:

\[\begin{aligned} &\text{缺}\ C:\ A\wedge B\ \nRightarrow\ D,\quad f(x)=x\ \text{于}\ [0,1]:\ f'=1\ne0,\ f(0)=0\ne1=f(1)\\[4pt] &\text{缺}\ B:\ A\wedge C\ \nRightarrow\ D,\quad f(x)=|1-2x|\ \text{于}\ [0,1]:\ f\notin D\left(\tfrac12\right),\ f'=\pm2\ne0\\[4pt] &\text{缺}\ A:\ B\wedge C\ \nRightarrow\ D,\quad f(x)=x\ (0\le x<1),\ f(1)=0:\ f\notin C(1),\ f'=1\ne0 \end{aligned} \]

充分性不成立

\[\begin{aligned} &D\ \nRightarrow\ A:\quad f(x)=\left(x-\tfrac12\right)^2\ (x\in(0,1)),\ f(0)=f(1)=0:\ f'\left(\tfrac12\right)=0,\ f\notin C(\{0,1\})\\[4pt] &D\ \nRightarrow\ C:\quad f(x)=-x(x-1)\ \text{于}\ \left[\tfrac14,1\right]:\ f'(x)=1-2x,\ f'\left(\tfrac12\right)=0,\ f\left(\tfrac14\right)=\tfrac{3}{16}\ne0=f(1) \end{aligned} \]

\(A\) 不可弱化为开区间连续

\[f(x)=x\ (x\in(0,1)),\ f(0)=f(1)=-1:\quad f\in C(0,1)\cap D(0,1),\ f(0)=f(1),\ f'=1\ne0 \]

\[\Longrightarrow\ \bigl[f\in C(a,b)\bigr]\wedge B\wedge C\ \nRightarrow\ D\qquad\blacksquare \]

单调性、凹凸性与端点

一阶

\[\begin{aligned} &f\in C[a,b]\cap D(a,b),\quad \forall x\in(a,b):\ f'(x)\ge0\ \Longrightarrow\ f\ \text{于}\ [a,b]\ \text{单调不减}\\ &\forall x_1<x_2:\ f(x_2)-f(x_1)\overset{\text{Lagrange}}{=}f'(\xi)(x_2-x_1)\ge0,\quad \xi\in(x_1,x_2) \end{aligned} \]

严格单调充要条件

\[\begin{aligned} &f\in D(a,b)\ \Longrightarrow\ \Bigl[f\ \text{严格增}\Bigr]\iff \begin{cases} \text{(i)}\ \forall x\in(a,b):\ f'(x)\ge0\\[2pt] \text{(ii)}\ f'\ \text{不在任何子区间上恒为}\ 0 \end{cases} \end{aligned} \]

\[\begin{aligned} &\text{反例说明 (ii) 不可去}:\quad f(x)=x^3:\ f'(0)=0,\ f\ \text{严格增}\\ &f(x)=x+\sin x:\ f'(x)=1+\cos x=0\ \text{于}\ x=(2k+1)\pi\ (\text{孤立零点}),\ f\ \text{严格增} \end{aligned} \]

端点引理

\[\begin{aligned} &f\ \text{于}\ (a,b)\ \text{严格增}\ \wedge\ f\in C(a^+)\ \Longrightarrow\ f\ \text{于}\ [a,b)\ \text{严格增}\\ &\forall x_0\in(a,b),\ \forall x_1\in(a,x_0),\ \forall x\in(a,x_1):\\ &\qquad a<x<x_1<x_0\ \Longrightarrow\ f(x)<f(x_1)<f(x_0)\\ &\qquad \xRightarrow{\ x\to a^+\ }\ f(a)=\lim_{x\to a^+}f(x)\le f(x_1)<f(x_0)\ \Longrightarrow\ f(a)<f(x_0) \end{aligned} \]

\[\text{用途:}\ e^x>1+x\ (x\ne0);\qquad \sin x>x-\frac{x^3}{6}\ (x>0) \]

凸性(一阶条件)

\[f''\ge0\ \text{于区间}\ \Longrightarrow\ f\ \text{凸}\ \Longrightarrow\ \forall x,y:\quad f(y)\ge f(x)+\nabla f(x)^{\top}(y-x)=f(x)+\sum_{i=1}^{n}\frac{\partial f}{\partial x_i}(x)(y_i-x_i) \]

二阶Taylor判极值

\[\begin{aligned} &\nabla f(x_0)=0\ \wedge\ \nabla^2 f(x_0)\succ0\ \Longrightarrow\ x_0\ \text{为严格极小点}\\ &f(x_0+h)-f(x_0)=\tfrac12h^{\top}\nabla^2f(x_0)h+o(\|h\|^2)>0\quad(0<\|h\|<\delta) \end{aligned} \]

posted @ 2026-10-26 19:06  星灯社小吊扇  阅读(19)  评论(0)    收藏  举报