好的,我们来深入探讨一下调和函数的相关证明。我将尝试用一种更像人类的叙述方式,层层递进地引导您理解其中的逻辑和技巧。
问题的背景:调和函数与拉普拉斯方程
首先,我们需要明确我们谈论的是什么——调和函数。一个在某个区域 $D$ 上足够光滑(通常是二次连续可微)的函数 $u(x, y)$,如果它满足 拉普拉斯方程,即:
$Delta u = frac{partial^2 u}{partial x^2} + frac{partial^2 u}{partial y^2} = 0$
那么我们就称 $u$ 为区域 $D$ 上的调和函数。拉普拉斯方程是偏微分方程领域中一个非常核心且重要的方程,它出现在物理学的许多领域,比如稳态热传导、电势、流体动力学等。
我们想证明什么?
虽然您没有直接给出具体要证明的命题,但我可以假设您想深入理解调和函数的一些基本性质的证明,或者您可能有一个关于调和函数的具体定理需要证明。为了让讨论更具普适性,我们不妨从一些最基础但又至关重要的性质入手,例如 调和函数的平均值性质。
平均值性质:一个非常直观且重要的性质
调和函数的一个核心性质是它的“平均值性质”。简单来说,就是调和函数在一个圆盘上的平均值等于它在圆盘中心的函数值。听起来可能有点抽象,我们把它写成数学公式:
定理(调和函数的平均值性质):
设 $u(x, y)$ 是区域 $D$ 上的调和函数,$D$ 是一个包含闭圆盘 $ar{B}(x_0, r)$ 的开集。那么:
$u(x_0, y_0) = frac{1}{2pi r} oint_{partial B(x_0, r)} u(x, y) ds$
其中,$(x_0, y_0)$ 是圆盘的圆心,$r$ 是圆盘的半径,$partial B(x_0, r)$ 是圆盘的边界(一个圆周),$ds$ 是弧长微分。
更进一步,对于一个半径为 $r$ 的圆盘,其函数值在圆盘内的平均值也等于圆心处的函数值:
$u(x_0, y_0) = frac{1}{pi r^2} iint_{B(x_0, r)} u(x, y) dx dy$
我们先来证明这个积分形式的平均值性质,因为它通常是证明圆周平均值性质的基础。
证明思路:Green公式大显身手
要证明这个积分性质,我们很自然地会想到利用 Green公式。Green公式在二维平面上,它将一个区域上的面积分与该区域边界上的线积分联系起来。它本身就是从高维的散度定理发展而来的。
回忆一下 Green公式:对于一个在区域 $D$ 上足够光滑的向量场 $mathbf{F} = (P, Q)$,我们有:
$iint_D left( frac{partial Q}{partial x} frac{partial P}{partial y}
ight) dx dy = oint_{partial D} (P dx + Q dy)$
现在,我们如何将它与拉普拉斯方程 $Delta u = 0$ 联系起来呢?我们需要找到一个合适的向量场,使得 $frac{partial Q}{partial x} frac{partial P}{partial y}$ 能够与 $Delta u$ 相关联。
让我们考虑一个巧妙的选择。我们知道 $Delta u = frac{partial^2 u}{partial x^2} + frac{partial^2 u}{partial y^2}$。如果我们选择一个向量场,它的散度恰好是 $Delta u$,那就太好了!
考虑向量场 $mathbf{F} =
abla u = left( frac{partial u}{partial x}, frac{partial u}{partial y}
ight)$。这个向量场的散度是:
$
abla cdot mathbf{F} = frac{partial}{partial x} left( frac{partial u}{partial x}
ight) + frac{partial}{partial y} left( frac{partial u}{partial y}
ight) = frac{partial^2 u}{partial x^2} + frac{partial^2 u}{partial y^2} = Delta u$
所以,如果我们将 $mathbf{F} =
abla u$ 代入 Green公式的变种(即散度定理在二维下的形式),我们得到:
$iint_D (
abla cdot
abla u) dx dy = oint_{partial D} (
abla u cdot mathbf{n}) ds$
其中 $mathbf{n}$ 是边界 $partial D$ 的外法线单位向量,$ds$ 是弧长。
现在,我们考虑我们想要证明的平均值性质中的积分区域:一个半径为 $r$、圆心在 $(x_0, y_0)$ 的圆盘 $B(x_0, r)$。拉普拉斯方程告诉我们 $Delta u = 0$ 在这个圆盘内部成立。
所以,根据上面的散度定理形式:
$iint_{B(x_0, r)} Delta u , dx dy = oint_{partial B(x_0, r)}
abla u cdot mathbf{n} , ds$
因为 $u$ 是调和函数,所以 $Delta u = 0$ 在 $B(x_0, r)$ 上成立。因此,左边的积分是零:
$iint_{B(x_0, r)} 0 , dx dy = 0$
这意味着右边的边界积分也必须是零:
$oint_{partial B(x_0, r)}
abla u cdot mathbf{n} , ds = 0$
这看起来很有用,但它并不是我们想要的那个平均值性质。我们需要一种方法来引入圆盘内的积分。这里就需要一些技巧了。
更巧妙的构造:引入一个特殊的向量场
我们不能直接从 $Delta u = 0$ 推导出圆盘内的平均值性质。我们需要对被积函数做一些“处理”。让我们考虑另一个向量场,它能帮助我们“隔离”圆心处的函数值。
考虑向量场 $mathbf{F}(x, y) = u(x, y)
abla phi(x, y)$,其中 $phi$ 是一个我们选择的函数。我们想让这个向量场的散度,在利用 Green公式后,能够产生我们需要的积分形式。
我们知道散度的计算公式:$
abla cdot (f mathbf{v}) = (
abla f) cdot mathbf{v} + f (
abla cdot mathbf{v})$。
所以,对于 $mathbf{F} = u
abla phi$,我们有:
$
abla cdot mathbf{F} =
abla cdot (u
abla phi) = (
abla u) cdot (
abla phi) + u (
abla cdot
abla phi) = (
abla u) cdot (
abla phi) + u Delta phi$
将这个代入散度定理:
$iint_D [(
abla u) cdot (
abla phi) + u Delta phi] , dx dy = oint_{partial D} (u
abla phi) cdot mathbf{n} , ds$
现在,我们回过头来看我们想要证明的平均值性质:
$u(x_0, y_0) = frac{1}{pi r^2} iint_{B(x_0, r)} u(x, y) dx dy$
或者说,
$pi r^2 u(x_0, y_0) = iint_{B(x_0, r)} u(x, y) dx dy$
我们希望左边的积分项最终能简化为 $u(x_0, y_0)$,而右边的边界积分能够以某种方式被处理掉或者与左边抵消。
这里的关键在于如何选择 $phi$。我们想要的是当我们在圆盘内积分时,当 $Delta u = 0$ 时,我们能够得到 $iint_D u , dx dy$ 的形式,并且它能被 $u(x_0, y_0)$ 表示。
一个“杀手锏”的函数 $phi$
让我们考虑一个在原点(如果圆心在原点的话)具有奇点的函数。如果我们希望 $Delta phi$ 在圆盘内部是零,但又能在边界上和 $
abla u cdot mathbf{n}$ 发生作用,我们可能会想到像 $ln|mathbf{x}|$ 这样的函数(在二维上)。
让我们假设圆心在原点 $(0,0)$,所以 $x_0=0, y_0=0$。我们考虑圆盘 $B(0, r)$。
选择 $phi(x, y) = lnleft(sqrt{x^2 + y^2}
ight) = frac{1}{2} ln(x^2 + y^2)$。
先计算 $Delta phi$:
$frac{partial phi}{partial x} = frac{1}{2} frac{2x}{x^2 + y^2} = frac{x}{x^2 + y^2}$
$frac{partial^2 phi}{partial x^2} = frac{(x^2+y^2) x(2x)}{(x^2+y^2)^2} = frac{y^2 x^2}{(x^2+y^2)^2}$
同理,
$frac{partial^2 phi}{partial y^2} = frac{x^2 y^2}{(x^2+y^2)^2}$
所以,$Delta phi = frac{partial^2 phi}{partial x^2} + frac{partial^2 phi}{partial y^2} = frac{y^2 x^2 + x^2 y^2}{(x^2+y^2)^2} = 0$。
太棒了!$phi(x, y) = frac{1}{2} ln(x^2 + y^2)$ 在除了原点以外的地方是调和的。
现在我们有了 $Delta phi = 0$(在我们的圆盘内部,只要圆心不在圆盘上,它就是调和的)。
将 $phi$ 和 $Delta phi = 0$ 代入散度定理的形式:
$iint_D [(
abla u) cdot (
abla phi) + u cdot 0] , dx dy = oint_{partial D} (u
abla phi) cdot mathbf{n} , ds$
$iint_D (
abla u) cdot (
abla phi) , dx dy = oint_{partial D} u (
abla phi cdot mathbf{n}) , ds$
我们知道 $
abla phi = left( frac{x}{x^2+y^2}, frac{y}{x^2+y^2}
ight)$。
在圆盘的边界 $partial B(0, r)$ 上,径向单位向量是 $mathbf{r} = frac{(x, y)}{sqrt{x^2+y^2}} = frac{(x, y)}{r}$。
外法线单位向量 $mathbf{n}$ 和径向单位向量 $mathbf{r}$ 是同一个方向。
所以,$
abla phi cdot mathbf{n} =
abla phi cdot mathbf{r} = left( frac{x}{x^2+y^2}, frac{y}{x^2+y^2}
ight) cdot left( frac{x}{r}, frac{y}{r}
ight)$
$= frac{x^2+y^2}{r(x^2+y^2)} = frac{1}{r}$。
因此,右边的边界积分变成:
$oint_{partial B(0, r)} u left( frac{1}{r}
ight) ds = frac{1}{r} oint_{partial B(0, r)} u , ds$。
到目前为止我们得到:
$iint_{B(0, r)} (
abla u) cdot (
abla phi) , dx dy = frac{1}{r} oint_{partial B(0, r)} u , ds$
我们离目标 $u(0,0) = frac{1}{pi r^2} iint_{B(0, r)} u , dx dy$ 还有一步。我们需要将左边的积分项处理成 $iint u , dx dy$ 的形式。
这里需要引入另一个非常关键的工具:高维调和函数的一种特殊形式的积分表示。
我们回到更一般的散度定理:
$iint_D
abla cdot mathbf{F} , dx dy = oint_{partial D} mathbf{F} cdot mathbf{n} , ds$
我们可以这样重新安排 Green 公式:
$iint_D u Delta v , dx dy = oint_{partial D} u (
abla v cdot mathbf{n}) , ds oint_{partial D} v (
abla u cdot mathbf{n}) , ds$
这个公式被称为 Green 第二公式。
让我们在其中令 $v = phi$。由于 $Delta phi = 0$ 在圆盘内部成立,并且 $u$ 是调和函数 ($Delta u = 0$),我们不能直接应用 Green 第二公式将 $Delta v$ 变成零。
我们需要换个角度。我们可以直接处理与拉普拉斯算子相关的形式。
考虑一个更通用的形式:
$iint_D u Delta v , dx dy = iint_D left( frac{partial}{partial x} (u frac{partial v}{partial x}) frac{partial}{partial x} (frac{partial u}{partial x} v) + frac{partial}{partial y} (u frac{partial v}{partial y}) frac{partial}{partial y} (frac{partial u}{partial y} v)
ight) dx dy$
这个展开是基于 $(f g_x)_x = f_x g_x + f g_{xx}$ 等的恒等式。
Let's try a simpler approach again.
We want to show $frac{1}{pi r^2} iint_{B(0, r)} u(x, y) , dx dy = u(0, 0)$.
This means we want to show $iint_{B(0, r)} u(x, y) , dx dy = pi r^2 u(0, 0)$.
Consider the function $g(x, y) = u(x, y) u(0, 0)$. This function is also harmonic because $Delta g = Delta u Delta u(0, 0) = 0 0 = 0$.
If we can show that for a harmonic function $g$ with $g(0,0) = 0$, then $iint_{B(0, r)} g , dx dy = 0$, then we are done.
$iint_{B(0, r)} u , dx dy = iint_{B(0, r)} (g(x, y) + u(0, 0)) , dx dy = iint_{B(0, r)} g(x, y) , dx dy + iint_{B(0, r)} u(0, 0) , dx dy$
$= 0 + u(0, 0) cdot ( ext{Area of } B(0, r)) = u(0, 0) cdot pi r^2$.
So the problem boils down to showing that if $g$ is harmonic and $g(0,0)=0$, then its integral over any disk centered at the origin is zero.
This is where the original $phi = ln r$ comes into play for a slightly different purpose, or perhaps we need a different approach.
Let's return to the $phi = ln r$ idea and see where it leads us for the circle average.
We had $iint_{B(0, r)}
abla u cdot
abla (ln
ho) , dx dy = frac{1}{r} oint_{partial B(0, r)} u , ds$, where $
ho = sqrt{x^2+y^2}$.
Now consider another identity using Green's second formula:
$iint_D (v Delta u u Delta v) , dx dy = oint_{partial D} (v
abla u u
abla v) cdot mathbf{n} , ds$.
Let $u$ be our harmonic function, and let $v$ be the auxiliary harmonic function $phi = ln
ho$.
$Delta u = 0$ and $Delta v = 0$ in $B(0, r)$.
So the left side of Green's second formula is 0:
$iint_D (v cdot 0 u cdot 0) , dx dy = 0$.
This implies that the boundary integral is also 0:
$oint_{partial D} (v
abla u u
abla v) cdot mathbf{n} , ds = 0$.
$oint_{partial B(0, r)} (ln
ho
abla u u
abla (ln
ho)) cdot mathbf{n} , ds = 0$.
We know $
abla (ln
ho) cdot mathbf{n} = frac{1}{r}$ on $partial B(0, r)$.
And $
abla u cdot mathbf{n} = frac{partial u}{partial
ho}$ in the radial direction.
So the integral becomes:
$oint_{partial B(0, r)} (ln r frac{partial u}{partial
ho} u frac{1}{r}) , ds = 0$.
$oint_{partial B(0, r)} frac{partial u}{partial
ho} ln r , ds frac{1}{r} oint_{partial B(0, r)} u , ds = 0$.
$frac{1}{r} oint_{partial B(0, r)} u , ds = ln r oint_{partial B(0, r)} frac{partial u}{partial
ho} , ds$.
This is still not exactly what we want. The issue here is that $ln
ho$ has a singularity at the origin. When we apply Green's theorem on a disk containing the singularity, we need to be careful. The standard Green's theorem applies to functions that are smooth everywhere in the region and on its boundary.
The Correct Approach Using a Modified Green Identity
The average value property is usually proven using a different version of Green's identity or by constructing a special kernel. A common technique involves using the Poisson kernel. However, let's try to build it up from Green's theorem in a way that addresses the singularity.
Let's consider the integral $iint_{B(x_0, r)} u(x, y) , dx dy$.
We can use polar coordinates centered at $(x_0, y_0)$. Let $x = x_0 +
ho cos heta$ and $y = y_0 +
ho sin heta$.
Then $dx dy =
ho , d
ho , d heta$.
The integral becomes $int_0^{2pi} int_0^r u(x_0 +
ho cos heta, y_0 +
ho sin heta)
ho , d
ho , d heta$.
We want to show this equals $pi r^2 u(x_0, y_0)$.
Let's consider the radial derivative of the integral of $u$ on a circle of radius $
ho$:
$I(
ho) = int_0^{2pi} u(x_0 +
ho cos heta, y_0 +
ho sin heta) , d heta$.
Then the average value is $frac{1}{pi r^2} int_0^r I(
ho)
ho , d
ho$.
We need to show that $frac{dI}{d
ho} = int_0^{2pi}
abla u cdot frac{partial mathbf{x}}{partial
ho} , d heta$, where $mathbf{x}(
ho, heta) = (x_0 +
ho cos heta, y_0 +
ho sin heta)$.
$frac{partial mathbf{x}}{partial
ho} = (cos heta, sin heta)$.
So, $frac{dI}{d
ho} = int_0^{2pi}
abla u cdot (cos heta, sin heta) , d heta$.
$
abla u cdot (cos heta, sin heta) = frac{partial u}{partial x} cos heta + frac{partial u}{partial y} sin heta$. This is the directional derivative of $u$ in the direction $(cos heta, sin heta)$.
Now, we use the harmonic property $Delta u = 0$.
Let's consider $int_0^r frac{dI}{d
ho} , d
ho = I(r) I(0)$.
$I(0) = int_0^{2pi} u(x_0, y_0) , d heta = 2pi u(x_0, y_0)$.
So, $int_0^r frac{dI}{d
ho} , d
ho = int_0^r left( int_0^{2pi} left( frac{partial u}{partial x} cos heta + frac{partial u}{partial y} sin heta
ight) , d heta
ight) , d
ho$.
This still looks complicated. Let's go back to Green's theorem and choose the auxiliary function wisely.
Consider the identity:
$iint_{B(x_0, r)} Delta u , dx dy = oint_{partial B(x_0, r)}
abla u cdot mathbf{n} , ds$.
Since $Delta u = 0$, we have $oint_{partial B(x_0, r)}
abla u cdot mathbf{n} , ds = 0$.
This means the average of the normal derivative on the boundary is zero.
The core idea for the average value theorem is to introduce a function whose Laplacian is related to $u$ itself in a way that cancels out.
Let's focus on the integral $iint_{B(x_0, r)} u , dx dy$.
Consider the function $v(x, y)$ such that $Delta v = u$. Such a function exists and is related to the fundamental solution of the Laplacian.
However, a more direct method uses the property that $Delta (
ho^2 u)$ for harmonic $u$ is $2
ho frac{partial u}{partial
ho} + 4u$ (this is wrong calculation)
Let's use the fact that for a harmonic function $u$, the integral over a disk is related to the value at the center.
A More Rigorous Approach Using Divergence Theorem and Parameter Differentiation
Let $u(x,y)$ be harmonic in a disk $D_R = {(x,y) : x^2+y^2 < R^2}$.
We want to show $u(0,0) = frac{1}{pi r^2} iint_{B(0,r)} u(x,y) dx dy$ for $r < R$.
Let's consider the integral:
$I(r) = iint_{B(0,r)} u(x,y) dx dy$.
We want to show $frac{dI}{dr} = int_0^{2pi} u(
ho cos heta,
ho sin heta)
ho , d heta$ evaluated at $
ho=r$? No, this is incorrect.
$frac{dI}{dr} = frac{d}{dr} int_0^r int_0^{2pi} u(
ho cos heta,
ho sin heta)
ho , d heta , d
ho$.
Using Leibniz integral rule:
$frac{dI}{dr} = int_0^{2pi} u(r cos heta, r sin heta) r , d heta$.
Now we need to relate this to $u(0,0)$.
Let's differentiate $I(r)$ with respect to $r$ again.
$frac{d^2 I}{dr^2} = frac{d}{dr} left( r int_0^{2pi} u(r cos heta, r sin heta) , d heta
ight)$.
Let $J(r) = int_0^{2pi} u(r cos heta, r sin heta) , d heta$.
Then $frac{dI}{dr} = r J(r)$.
$frac{d^2 I}{dr^2} = J(r) + r frac{dJ}{dr}$.
$frac{dJ}{dr} = int_0^{2pi}
abla u(r cos heta, r sin heta) cdot (cos heta, sin heta) , d heta$.
This is the integral of the radial derivative of $u$ on the circle of radius $r$.
Now we use Green's theorem again. Let $mathbf{F} = u
abla phi$ where $phi = ln r$. We saw $Delta phi = 0$.
$iint_{B(0,r)}
abla cdot (u
abla phi) , dx dy = oint_{partial B(0,r)} u
abla phi cdot mathbf{n} , ds$.
$iint_{B(0,r)} [(
abla u) cdot (
abla phi) + u Delta phi] , dx dy = oint_{partial B(0,r)} u (
abla phi cdot mathbf{n}) , ds$.
Since $Delta phi = 0$:
$iint_{B(0,r)} (
abla u) cdot (
abla (ln
ho)) , dx dy = oint_{partial B(0,r)} u frac{1}{
ho} , ds$.
With $
ho=r$ on the boundary:
$iint_{B(0,r)}
abla u cdot frac{mathbf{x}}{
ho^2} , dx dy = frac{1}{r} oint_{partial B(0,r)} u , ds$.
Consider the function $G(x,y) = u(x,y) u(0,0)$. $G$ is harmonic and $G(0,0)=0$.
We want to show $iint_{B(0,r)} G(x,y) dx dy = 0$.
Let's use the identity:
$iint_{B(0,r)} Delta u , dx dy = 0$.
Now consider the identity for a harmonic function $u$:
$iint_{B(0,r)} u , dx dy = pi r^2 u(0,0)$.
This identity is often proven using the Poisson integral formula for the disk. The Poisson integral formula is derived using Fourier series or complex analysis, and it explicitly shows the average value property.
If the goal is to prove the average value property from basic principles (like Green's theorem), the derivation can be quite involved.
Let's summarize the approach using Green's theorem with an auxiliary function that is harmonic except at the center.
The function $phi(x, y) = ln(sqrt{x^2+y^2})$ is harmonic for $(x,y)
eq (0,0)$.
We consider the identity derived from Green's second theorem, applied to a region that excludes a small disk around the origin:
Let $D_epsilon = B(0, r) setminus B(0, epsilon)$ for small $epsilon > 0$.
$iint_{D_epsilon} (v Delta u u Delta v) , dx dy = oint_{partial D_epsilon} (v
abla u u
abla v) cdot mathbf{n} , ds$.
Here $u$ is harmonic, so $Delta u = 0$. Let $v = ln
ho$. $Delta v = 0$ in $D_epsilon$.
So the left side is zero.
$oint_{partial D_epsilon} (ln
ho
abla u u
abla (ln
ho)) cdot mathbf{n} , ds = 0$.
The boundary $partial D_epsilon$ consists of the outer circle $partial B(0, r)$ and the inner circle $partial B(0, epsilon)$.
Let $mathbf{n}_r$ be the outward normal for $partial B(0, r)$, and $mathbf{n}_epsilon$ be the outward normal for $partial B(0, epsilon)$ (which points inward towards the origin).
$oint_{partial B(0, r)} (ln r
abla u u
abla (ln r)) cdot mathbf{n}_r , ds + oint_{partial B(0, epsilon)} (ln epsilon
abla u u
abla (ln epsilon)) cdot mathbf{n}_epsilon , ds = 0$.
On $partial B(0, r)$: $mathbf{n}_r = frac{mathbf{x}}{r}$, $
abla ln
ho cdot mathbf{n}_r = frac{1}{r}$.
$oint_{partial B(0, r)} (ln r frac{partial u}{partial
ho} u frac{1}{r}) , ds = 0$.
$ln r oint_{partial B(0, r)} frac{partial u}{partial
ho} , ds frac{1}{r} oint_{partial B(0, r)} u , ds = 0$.
On $partial B(0, epsilon)$: $mathbf{n}_epsilon = frac{mathbf{x}}{epsilon}$, $
abla ln
ho cdot mathbf{n}_epsilon = frac{1}{epsilon}$.
$oint_{partial B(0, epsilon)} (ln epsilon frac{partial u}{partial
ho} u (frac{1}{epsilon})) , ds = 0$.
$oint_{partial B(0, epsilon)} (ln epsilon frac{partial u}{partial
ho} + frac{u}{epsilon}) , ds = 0$.
As $epsilon o 0$:
$frac{partial u}{partial
ho}$ is bounded on $partial B(0, epsilon)$ because $u$ is smooth. So $ln epsilon frac{partial u}{partial
ho}$ term goes to $infty$ if $frac{partial u}{partial
ho}$ is not zero. This indicates a problem with $ln
ho$.
The correct auxiliary function is often related to the fundamental solution of the Laplacian, which is $frac{1}{2pi} ln |mathbf{x}mathbf{y}|$. For the average value property, the Poisson kernel is more direct.
If the question is to prove a specific property of harmonic functions, please provide the property. The average value property is a good example that requires careful application of Green's theorem and handling of singularities, or using more advanced tools like the Poisson kernel.
Let me know if you had a specific statement in mind, and I can tailor the detailed proof to that statement. The journey to prove properties of harmonic functions often involves navigating the subtle applications of integral theorems and understanding the behavior of solutions near singularities.