Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

This question stems from a step of a proof in this paper:

The question itself I didn't feel to be "research level" as it is only an asymptotic estimate, so I'm posting this here.

The step occurs near the bottom of page 8:

"Since $\displaystyle-\frac{1}{4\pi}\frac{t}{(|x-x^\prime |^2+t^2)^{3/2}}$ is the Poisson kernel for the half space and $f(\cdot,h\zeta^\prime)$ has compact support in $\Omega$,

$$-\frac{1}{4\pi}\int_\Omega \frac{h^2 (\zeta-\zeta^\prime)}{(|x-x^\prime|^2 + h^2|\zeta-\zeta^\prime |^2 )^{3/2}} f(x^\prime , h\zeta^\prime) \,dx^\prime =-\frac{h}{2} f(x,0) \frac{\zeta - \zeta^\prime}{ |\zeta-\zeta^\prime |} +O(h^2)\text{"}$$

$x,x^\prime \in \Omega\subset\mathbb{R}^2$ where $\zeta,\zeta^\prime \in \mathbb{R}$. So as a big picture, we are looking at functions of $(x,h\zeta)$ and $(x^\prime , h\zeta^\prime)$ which are in $\mathbb R^3$. As $h$ goes to zero, the third component of both go to the boundary (i.e. 0). The Poisson theorem about $x$ going to the boundary that I mention below requires the third component of both to equal 0.

In Evans' PDE book chapter 2.2 theorem 14, we have that if $\zeta^\prime=0$ and (ignoring one of the $h$ in front) then we would have

$$-\frac{1}{4\pi}\int_\Omega \frac{h (\zeta-\zeta^\prime)}{(|x-x^\prime|^2 + h^2|\zeta-\zeta^\prime |^2 )^{3/2}} f(x^\prime , h\zeta^\prime) \,dx^\prime =-f(x,0)$$ as $h\to 0$. I'm confused on (a) how they get the $O$ estimate and (b) how the fraction $\frac{\zeta - \zeta^\prime}{ |\zeta-\zeta^\prime |}$ pops out. I think it has to do with $\zeta^\prime$ possibly not being on $\partial\mathbb{R}^{n}_+$ which is what is required in Evan's theorem, but I'm stuck there.

For (a): Let $\displaystyle u(x)= \frac{2 x_n}{n\alpha(n)}\int_{\partial\mathbb{R}^n_+}\frac{g(y)}{|x-y|^n}\, dy$ with $x\in\mathbb{R}^n_+$ and $y\in\partial\mathbb{R}^n_+$. For convenience we write: $u(x)=\int_{\partial\mathbb{R}^n_+}K(x,y)g(y)\, dy$ with $K(x,y)$ defined accordingly.

Following the steps in Evan's proof: Let $\epsilon>0$ be given and $\delta>0$ such that $|g(y)-g(x_0)|<\epsilon$ if $|y-x_0|<\delta$ and $y\in\partial\mathbb{R}^n_+$. we can find that for $x_0$ on the boundary, if $|x-x_0|<\delta/2$ So as $x\to x_0\in\partial\mathbb{R}^n_+$, $$|u(x)-g(x_0)|=\left |\int_{\partial\mathbb{R}^n_+}K(x,y)[g(y)-g(x_0)]\, dy \right| \leq \int_{\partial\mathbb{R}^n_+ \cap B(x_0,\delta)}K(x,y)|g(y)-g(x_0)|\, dy +\int_{\partial\mathbb{R}^n_+ \setminus B(x_0,\delta)}K(x,y)|g(y)-g(x_0)|\, dy$$ $$=I+J$$

In the book it is shown $I\leq\epsilon$ and that $J=O(x_n)$. So what I'm wondering is this: does this mean I can write:

$$u(x) = g(x_0) + O(x_n)$$ as $x_n \to 0$?

share|cite|improve this question
up vote 1 down vote accepted

I admire the book by Evans, but his notation for half-space Poisson kernel does not align with the way I think of it. Loosely following Wikipedia, I write the Poisson kernel for half-space $\mathbb R^3_+$ as $$ P_t(x,x')=\frac{t}{2\pi}\frac{1}{(|x-x'|^2+t^2)^{3/2}},\qquad x,x'\in\mathbb R, \ t>0 \tag{I} $$ This kernel has the property that $\int_{\mathbb R^2} P_t(x-x')g(x')\,dx'$ is the value at $(x,t)\in\mathbb R^3_+$ of the harmonic extension of $g$ to the half-space. The harmonic extension of a continuous functions is continuous up to the boundary, which means we get the original function $g$ back as $t\to 0$.

Comparing (I) with the integral (with one less $h$, as in your post): $$-\frac{1}{4\pi}\int_\Omega \frac{h (\zeta-\zeta^\prime)}{(|x-x^\prime|^2 + h^2|\zeta-\zeta^\prime |^2 )^{3/2}} f(x^\prime , h\zeta^\prime) \,dx^\prime \tag{II}$$ we decide to let $t=h|\zeta-\zeta^\prime |$. We become aware at this point that the numerator in (II) is $\pm t$, depending on the sign of $\zeta-\zeta'$. So, rewrite the integral as $$-\frac{1}{4\pi}\frac{\zeta-\zeta^\prime}{|\zeta-\zeta^\prime|}\int_\Omega \frac{h |\zeta-\zeta^\prime|}{(|x-x^\prime|^2 + h^2|\zeta-\zeta^\prime |^2 )^{3/2}} f(x^\prime , h\zeta^\prime) \,dx^\prime \\ = -\frac{1}{2}\frac{\zeta-\zeta^\prime}{|\zeta-\zeta^\prime|}\int_\Omega P_t(x,x') f(x^\prime , h\zeta^\prime) \,dx^\prime $$ Some people, including myself, would rather write $\mathrm{sign}\,(\zeta-\zeta')$ instead of the fraction.

As $h\to 0$, two things happen. First, the function whose harmonic extension we consider changes: $f(\cdot, h\zeta')\to f(\cdot, 0)$ in some way (I hope uniformly, but did not check). Second, we approach the boundary where the Poisson integral recovers the original function. This, $$-\frac12\frac{\zeta-\zeta^\prime}{|\zeta-\zeta^\prime|} f(x,0)\tag{III}$$ is exactly what we should expect in the limit.

The authors claim that the difference between (II) and (III) is bounded by $h\|f\|_{C^1}$. I don't quite believe them, because harmonic extension of a $C^1$ function is not necessarily Lipschitz. Anything a little better than $C^1$, such as $C^{1,\alpha}$, would be enough.

share|cite|improve this answer
Thanks for your response! I have a better understanding now. I'm going to fully work through it tomorrow to make sure I get everything. Do you have any comments about the O-estimate? – toypajme Aug 5 '12 at 19:24
@toypajme The last paragraph of my answer is about the O-estimate. As I said, I don't think the authors got it right. If $f$ has two continuous derivatives, the O-estimate holds, but one derivative may not be enough. – user31373 Aug 5 '12 at 19:58
Oh, right. I figured out where my confusion was. – toypajme Aug 6 '12 at 16:01
I appreciate your help. I'm missing how, even with $C^{1,\alpha}$ we get the bound $h\|f\|_{C^{1}}$. I see when subtracting II from III in absolute value and substituting in the poisson kernel $f(x,0)$ as a "1" in III, we can bound to get $|f(x^\prime,h\zeta)-f(x,0)|$ inside the integral, but I don't know how to continue from here to get $h\|f\|_{C^{1}}$. – toypajme Aug 8 '12 at 14:11
I'm guessing it has to do with a metric inequality and using the norm for the Lipschitz "C," but I can't write it down satisfactory. Is it just that for a fixed $h$, we may find a $C$ such that $|(x^\prime,h\zeta^\prime)-(x,0)|\leq C |h\zeta^\prime| = Ch$ ? – toypajme Aug 8 '12 at 14:17

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.