多元函数积分中值定理 Fubini定理

第三周讲完了积分中值定理(也就是积分性质应该讲完了),积分中值定理多用于估计积分值,可以利用一个函数值来估计整个积分的值,并学了如何使用Fubini定理去计算多元函数积分值。

多元函数积分中值定理

定义1(有界集的“体积”,积分平均值,加权积分平均值)

ARnA\subset \mathbb R^n 有界,m(A)=0m^*(\partial A)=0,则称 V(A):=A1dxV(A):=\int_A1\,dxAA 的“体积”。

V(A)>0,f:ARV(A) > 0, f:A\rightarrow \mathbb R 可积,则称 1V(A)Af\frac{1}{V(A)}\int_AfffAA 上得积分平均值

φ:AR\varphi:A\rightarrow \mathbb Rφ\varphi 可积,φ0\varphi\geqslant 0Aφ=1\int_A\varphi = 1,则称 φ\varphiAA 上的权函数

Afφ\int_A f\varphiff加权积分平均值

定理2(积分中值定理)

KRnK\subset\mathbb R^n连通的紧集m(K)=0,V(K)>0m^*(\partial K) = 0, V(K) > 0,设 fC(K)f\in C(K)

ξK\exists\xi\in K,使 1V(K)Kf=f(ξ)\frac{1}{V(K)}\int_Kf=f(\xi)


证明: (取最大和最小值,再利用连通紧集上连续函数的介值定理)

由于 fC(K)f\in C(K),则 ffKK 上存在最值,记 m=minKf,M=maxKfm=\min\limits_K f, M=\max\limits_K f,又 m(K)=0m^*(\partial K) = 0ff 连续有界,故 ff 可积,则有

1V(K)Km1V(K)Kf1V(K)KMm1V(K)KfM\begin{aligned} \frac{1}{V(K)}\int_Km&\leqslant\frac{1}{V(K)}\int_K f\leqslant\frac{1}{V(K)}\int_K M\\ \Rightarrow m&\leqslant\frac{1}{V(K)}\int_Kf\leqslant M \end{aligned}

介值定理知,ξK\exists\xi\in K,使得 f(ξ)=1V(K)Kff(\xi)=\frac{1}{V(K)}\int_Kf

定理3(加权积分中值定理)

KRnK\subset \mathbb R^n连通的紧集m(K)=0m^*(K) = 0,设 φ:KR\varphi:K\rightarrow \mathbb R 可积,φ0,Kφ=1\varphi\geqslant 0, \int_K\varphi = 1,设 fC(K)f\in C(K),则 ξK\exists \xi\in K,使得 Kfφ=f(ξ)\int_K f\varphi=f(\xi)


证明: (思路和定理2的证明类似)

m=minKf,M=maxKfm=\min\limits_K f, M=\max\limits_K f,由于 D(fφ)D(φ)m(D(fφ))m(D(φ))=0D(f\varphi)\subset D(\varphi)\Rightarrow m^*(D(f\varphi))\leqslant m^*(D(\varphi)) = 0

则有

m=KmφKfφKMφ=Mm = \int_Km\varphi\leqslant\int_Kf\varphi\leqslant\int_KM\varphi = M

介值定理知,ξK\exists \xi \in K,使得 f(ξ)=Afφf(\xi) = \int_Af\varphi

Fubini 定理

定理1(Fubini定理)

PRn,QRmP\subset \mathbb R^n, Q\subset \mathbb R^m 均为闭方体,f:P×QRf:P\times Q\rightarrow \mathbb R 可积,f=f(x,y),xP,yQf=f(x, y),x\in P, y\in Q,则下列函数分别关于 x,yx, y 均可积:(这里把 f(x,y)f(x,y) 看做“二元函数”,但其实 xxnn 维的,yymm 维的)

  • xP\forall x\in P,关于 xx 的函数:Qf(x,y)dy, Qf(x,y)dy\underline{\int}_Qf(x, y)\,dy,\ \overline{\int}_Qf(x, y)\,dy。(固定 xxyy 进行积分)

  • yQ\forall y\in Q,关于 yy 的函数:Pf(x,y)dx, Pf(x,y)dx\underline{\int}_Pf(x, y)\,dx,\ \overline{\int}_Pf(x, y)\,dx。(固定 yyxx 进行积分)

且有:

P×Qf=P{Qf(x,y)dy}dx=P{Qf(x,y)dy}dx=Q{Pf(x,y)dx}dy=Q{Pf(x,y)dx}dy\begin{aligned} \int_{P\times Q}f & = \int_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx = \int_P\left\{\overline{\int}_Qf(x, y)\,dy\right\}\,dx\\ & = \int_Q\left\{\underline{\int}_Pf(x, y)\,dx\right\}\,dy = \int_Q\left\{\overline{\int}_Pf(x, y)\,dx\right\}\,dy \end{aligned}


注: 定理1比较复杂其原因是 Qf(x,y)dy, Pf(x,y)dx\int_Qf(x, y)dy,\ \int_Pf(x, y)dx 不一定存在,而它们的Darboux上下积分却可以继续积分,这说明它们的Darboux上下积分的间断点构成零测集,而且它们四个Darboux上下积分继续积分出来的结果都等于 P×Qf\int_{P\times Q} f。下面的推论2就没这么复杂了。

思路: 构造 P×QP\times Q 上的一个划分,将这个划分分别分成 P,QP, Q 上的两个划分,再利用Darboux上下积分定义,对它们进行估计,具体来说,是用 ffP×QP\times Q 上的Darboux上下积分进行夹逼,由于 ff 是可积的,利用 ff Darboux上下积分收敛的性质,得出结论。

证明: 只证明其中一个,其他同理可证。

下面证明:

P×Qf=P{Qf(x,y)dy}dx\int_{P\times Q}f=\int_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx

只需证:

P×Qf=P{Qf(x,y)dy}dx=P{Qf(x,y)dy}dx\begin{aligned} \int_{P\times Q}f=\underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx=\overline{\int}_P\left\{\underline{\int}_Qf(x,y)\,dy\right\}\,dx \end{aligned}

π\piP×QP\times Q 的分划 π\pi,则 P\exists P 的分划 π1\pi_1QQ 的分划 π2\pi_2,使得 π={p×q:pπ1,qπ2}\pi = \left\{p\times q: p\in \pi_1, q\in\pi_2\right\}

  • 我们先对 P{Qf(x,y)dy}dx\underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx 进行估计。

P{Qf(x,y)dy}dxpπ1(infxpQf(x,y)dy)V(p)\underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx \geqslant \sum_{p\in\pi_1}\left(\inf_{x\in p}\underline{\int}_Qf(x, y)\,dy\right)V(p)

xpx\in p,则

Qf(x,y)dyqπ2(infyqf(x,y))V(q)qπ2mp×qV(q)infxpQf(x,y)dyqπ2mp×qV(q)\begin{aligned} &\underline{\int}_Qf(x, y)\,dy\geqslant \sum_{q\in\pi_2}\left(\inf_{y\in q}f(x, y)\right)V(q)\geqslant\sum_{q\in\pi_2}m_{p\times q}V(q)\\ \Rightarrow \inf_{x\in p}&\underline{\int}_Qf(x, y)dy \geqslant \sum_{q\in \pi_2}m_{p\times q}V(q) \end{aligned}

其中 mp×q=infxp×qf(x)m_{p\times q} = \inf\limits_{x \in p\times q} f(x)

P{Qf(x,y)dy}dxpπ1qπ2mp×qV(p)V(q)p×qπmp×qV(p×q)P×Qf\begin{aligned} \underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx &\geqslant \sum_{p\in\pi_1}\sum_{q\in\pi_2}m_{p\times q}V(p)V(q)\\ &\geqslant \sum_{p\times q\in \pi}m_{p\times q} V(p\times q)\\ &\geqslant \underline{\int}_{P\times Q} f \end{aligned}

Δπ0\Delta\pi\rightarrow 0,得

P{Qf(x,y)dy}dxP×Qf=P×Qf\underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx\geqslant\underline{\int}_{P\times Q} f = \int_{P\times Q}f

  • 我们再对 P{Qf(x,y)dy}dx\overline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx 进行估计。(原理相同,只需把 \geqslant 换成 \leqslantinf\inf 换成 sup\sup

P{Qf(x,y)dy}dxpπ1(supxpQf(x,y)dy)V(p)\overline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx \leqslant \sum_{p\in\pi_1}\left(\sup_{x\in p}\underline{\int}_Qf(x, y)\,dy\right)V(p)

xpx\in p,则(这里多了一个估计,Darboux下积分 \leqslant Darboux上积分,为了保持 \leqslant

Qf(x,y)dyQf(x,y)dyqπ2(supyqf(x,y))V(q)qπ2Mp×qV(q)supxpQf(x,y)dyqπ2Mp×qV(q)\begin{aligned} &\underline{\int}_Qf(x, y)\,dy\leqslant \overline{\int}_Qf(x, y)\,dy\leqslant \sum_{q\in\pi_2}\left(\sup_{y\in q}f(x, y)\right)V(q)\leqslant\sum_{q\in\pi_2}M_{p\times q}V(q)\\ \Rightarrow \sup_{x\in p}&\underline{\int}_Qf(x, y)dy \leqslant \sum_{q\in \pi_2}M_{p\times q}V(q) \end{aligned}

其中 Mp×q=supxp×qf(x)M_{p\times q} = \sup\limits_{x \in p\times q} f(x)

P{Qf(x,y)dy}dxpπ1qπ2Mp×qV(p)V(q)p×qπMp×qV(p×q)P×Qf\begin{aligned} \overline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx &\leqslant \sum_{p\in\pi_1}\sum_{q\in\pi_2}M_{p\times q}V(p)V(q)\\ &\leqslant \sum_{p\times q\in \pi}M_{p\times q} V(p\times q)\\ &\leqslant \overline{\int}_{P\times Q} f \end{aligned}

Δπ0\Delta\pi\rightarrow 0,得

P{Qf(x,y)dy}dxP×Qf=P×Qf\overline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx\leqslant\overline{\int}_{P\times Q} f = \int_{P\times Q}f

  • 综上,有

P×Qf=P×QfP{Qf(x,y)dy}dxP{Qf(x,y)dy}dxP×Qf=P×Qf\begin{aligned} \int_{P\times Q}f = \underline{\int}_{P\times Q} f\leqslant \underline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx\leqslant\overline{\int}_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx\leqslant\overline{\int}_{P\times Q}f = \int_{P\times Q}f \end{aligned}

则,Qf(x,y)dy\underline{\int}_Qf(x, y)\,dy 可积,且

P×Qf=P{Qf(x,y)dy}dx\int_{P\times Q} f = \int_P\left\{\underline{\int}_Qf(x, y)\,dy\right\}\,dx

推论2(积分函数连续)

PRn,QRmP\subset\mathbb R^n, Q\subset\mathbb R^m 均为闭方体,fC(P×Q)f\in C(P\times Q),则

P×Qf=P{Qf(x,y)dy}dx=Q{Pf(x,y)dx}dy\begin{aligned} \int_{P\times Q} f = \int_P\left\{\int_Qf(x, y)\,dy\right\}\,dx = \int_Q\left\{\int_Pf(x, y)\,dx\right\}\,dy \end{aligned}


证明: 由于 fC(P×Q)f\in C(P\times Q),所以 ffPP 上连续,也在 QQ 上连续,且 P,QP,Q 均为闭方体,则 ffP,QP, Q 上可积,于是由 Darboux定理 知:

Qf(x,y)dy=Qf(x,y)dy=Qf(x,y)dyPf(x,y)dx=Pf(x,y)dx=Qf(x,y)dx\begin{aligned} \underline{\int}_Qf(x,y)\,dy = \overline{\int}_Qf(x, y)\,dy = \int_Qf(x, y)\, dy\\ \underline{\int}_Pf(x,y)\,dx = \overline{\int}_Pf(x, y)\,dx = \int_Qf(x, y)\, dx\\ \end{aligned}

再通过 定理1 得证。

推论3(求曲面柱的“体积”)

ΩRn1\Omega\subset\mathbb R^{n-1} 为有界开集,m(Ω)=0m^*(\partial\Omega) = 0,设 φ,ψC(Ω)\varphi, \psi\in C(\overline{\Omega}),且 xΩ, φ(x)<ψ(x)\forall x\in \Omega,\ \varphi(x) < \psi(x)

D={(x,y):xΩ,φ(x)<y<ψ(x)}D = \{(x, y): x\in \Omega, \varphi(x) < y < \psi(x)\}。(注:这里 xxn1n-1 维的,yy11 维的)

fC(Dˉ)f\in C(\bar{D}),则

Dˉf=Ω{φ(x)ψ(x)f(x,y)dy}dx\begin{aligned} \int_{\bar{D}}f = \int_{\overline{\Omega}}\left\{\int_{\varphi(x)}^{\psi(x)}f(x, y)\,dy\right\}\,dx \end{aligned}


思路: 先证明内侧积分有意义,再用闭方体对 Dˉ\bar{D} 进行一个覆盖,通过有界集积分的定义写出来,最后将恒为零的部分删去即可。

证明:

由于 D(Ω×R)(graph φ)(graph ψ)\partial D\subset(\partial\Omega\times\mathbb R)\cup(\text{graph }\varphi)\cup(\text{graph }\psi)

因为 m(Ω)=0m^*(\partial\Omega) = 0,则 m(Ω×R)=0m^*(\partial\Omega\times\mathbb R)=0(通过定义证明),又由 note1 - 命题10 知,m(graph φ)=m(graph ψ)=0m^*(\text{graph }\varphi)=m^*(\text{graph }\psi)=0

m(D)=0m^*(\partial D)=0,又因为 fC(Dˉ)f\in C(\bar{D}) 有界,所以 ff 可积。

PRn1P\subset \mathbb R^{n-1} 为闭方体,使得 ΩP\overline{\Omega}\subset P^\circ。设 m+1φ(x)<ψ(x)M1m+1\leqslant\varphi(x) < \psi(x)\leqslant M-1

Q=P×[m,M]Q = P\times [m,M],则 DˉQ\bar{D}\subset Q

f~\tilde{f}ffQQ 上的零延拓,则

Dˉf=Qf~=P{mMf~(x,y)dy}dx\begin{aligned} \int_{\bar{D}}f=\int_Q\tilde{f}=\int_P\left\{\int_m^M\tilde{f}(x, y)\,dy\right\}\,dx \end{aligned}

I(x)=mMf~(x,y)dyI(x) = \int_m^M\tilde{f}(x, y)\,dy

xPΩx\in P-\overline{\Omega}f~(x,y)=0I(x)=0\tilde{f}(x, y)=0\Rightarrow I(x) = 0

xΩx\in\overline{\Omega} 时,

I(x)=mMf~(x,y)dy=φ(x)ψ(x)f~(x,y)dy=φ(x)ψ(x)f(x,y)dy\begin{aligned} I(x)=\int_m^M\tilde{f}(x, y)dy = \int_{\varphi(x)}^{\psi(x)}\tilde{f}(x, y)dy = \int_{\varphi(x)}^{\psi(x)}f(x, y)dy \end{aligned}

Dˉf=FubiniP{mMf~(x,y)dy}dx=Ω{mMf~(x,y)dy}dx=Ω{φ(x)ψ(x)f(x,y)dy}dx\begin{aligned} \int_{\bar{D}}f \xlongequal{\text{Fubini}}& \int_P\left\{\int_m^M\tilde{f}(x, y)\,dy\right\}\,dx\\ =& \int_{\overline{\Omega}}\left\{\int_m^M\tilde{f}(x, y)\,dy\right\}\,dx\\ =& \int_{\overline{\Omega}}\left\{\int_{\varphi(x)}^{\psi(x)}f(x, y)\,dy\right\}\,dx\\ \end{aligned}

QED

由于边界积分为零: 故可以把边界去掉,即

Df=Ω{φ(x)ψ(x)f(x,y)dy}dx\begin{aligned} \int_Df=\int_\Omega\left\{\int_{\varphi(x)}^{\psi(x)}f(x, y)\,dy\right\}\,dx \end{aligned}

求曲面柱体积:f=1f=1,即

V(D)=D1=Ω(ψ(x)φ(x))dx\begin{aligned} V(D)=\int_D1=\int_{\Omega}(\psi(x)-\varphi(x))\,dx \end{aligned}


多元函数积分中值定理 Fubini定理
https://wty-yy.github.io/posts/54113/
作者
wty
发布于
2021年10月3日
许可协议