2
$\begingroup$

I am currently trying to find the Laplacian of $u(x)=\frac{1}{|\vec x|^2}$ in $3$ Dimensions in the sense of distributions. Now, let us have a test function $\phi(x)$. We consider two cases, when $0\notin \text{supp}\>\phi$, and when $0\in \text{supp }\phi$. Now in first case, it is obvious that the Laplacian in the sense of distributions is just the pointwise Laplacian. Now I considered when $0\in\text{supp }\phi$. Then, let $B(0,\epsilon)$ be some ball centered at $0$ with radius $\epsilon$, and consider $\Omega$ the support of $\phi$. Then, consider $\Omega_\epsilon=\Omega\setminus B(0,\epsilon)$. Integrating over this using Green's Second Identity yielded:$$\iiint_{\Omega_{\epsilon}}\frac{1}{|x|^2}\Delta\phi dx = \iiint_{\Omega_{\epsilon}}\Delta\left(\frac{1}{|x|^2}\right)\phi dx+\iint_{\partial \Omega_{\epsilon}}\frac{1}{|x|^2} \frac{\partial\phi}{\partial\vec n} dS_{x}-\iint_{\partial \Omega_{\epsilon}}\frac{\partial}{\partial\vec n}\left(\frac{1}{|x|^2}\right)\phi dS_x\>\>\>\>(1)$$ Now, I focused on the last integral on the right hand side above: $$\iint_{\partial \Omega_{\epsilon}}\frac{\partial}{\partial\vec n}\left(\frac{1}{|x|^2}\right)\phi dS_x=8\pi\epsilon\left(\frac{1}{4\pi\epsilon^2}\iint_{\partial B(0,\epsilon)}\phi dS_x\right)$$ By the Mean Value Property, the value in the brackets above gave $\phi(0)$ and if we take $\epsilon\rightarrow 0$, we get the above tends to $0$. Now I looked at the second integral on the right hand side of $(1)$. $$\iint_{\partial B(0,\epsilon)}\frac{1}{|x|^2}\frac{\partial\phi}{\partial \vec n} dS_x = \frac{1}{\epsilon^2}\iint_{\partial B(0,\epsilon)}\frac{\partial\phi}{\partial\vec n}dS_x$$ Now since $\phi$ is bounded as it is compact, its derivatives hold the same characteristics, and thus: $$\frac{1}{\epsilon^2}\iint_{\partial B(0,\epsilon)}\frac{\partial\phi}{\partial\vec n}dS_x\leq C\cdot \frac{1}{\epsilon^2}\cdot 4\pi\epsilon^2$$ The above gives a constant. Thus, am I supposed to conclude that the distributional Laplacian when the support of $\phi$ includes $0$ is simply the pointwise Laplacian plus some arbitrary constant?

$\endgroup$
  • $\begingroup$ In three dimensions? $\endgroup$ – md2perpe Dec 17 '17 at 21:27
  • $\begingroup$ @md2perpe ah yes, forgot to mention the dimension. $\endgroup$ – Felicio Grande Dec 17 '17 at 21:30
1
$\begingroup$

No need for "two cases": the second case is the general case, since we are never going to use the assumption $0\in \operatorname{supp}\phi$.

Incorrect

By the Mean Value Property, the value in the brackets above gave $ϕ(0)$

The function $\phi$ is not harmonic, so it does not satisfy the Mean Value Property. But it is true that $$\frac{1}{4\pi\epsilon^2}\iint_{\partial B(0,\epsilon)}\phi dS_x \to \phi(0)$$ because $\phi$ is continuous. However, there is more to say about this term...

Incorrect

$$\iint_{\partial \Omega_{\epsilon}}\frac{\partial}{\partial\vec n}\left(\frac{1}{|x|^2}\right)\phi dS_x=8\pi\epsilon\left(\frac{1}{4\pi\epsilon^2}\iint_{\partial B(0,\epsilon)}\phi dS_x\right)$$ The normal derivative of $1/|x|^2$ on the sphere of radius $\epsilon$ is $2\epsilon^{-3}$, not $2\epsilon^{-1}$. As a result, this term tends to infinity, which is troublesome (but see below).

Correct but incomplete

$$\frac{1}{\epsilon^2}\iint_{\partial B(0,\epsilon)}\frac{\partial\phi}{\partial\vec n}dS_x\leq C\cdot \frac{1}{\epsilon^2}\cdot 4\pi\epsilon^2$$ This term actually tends to $0$, because the divergence theorem implies $$ \iint_{\partial B(0,\epsilon)}\frac{\partial\phi}{\partial\vec n}dS_x = \iiint_{B(0, \epsilon)} \Delta \phi(x)\,dx \le C \epsilon^3 $$

Conclusion

We need the limit as $\epsilon\to 0$ of $$ \iiint_{\Omega_{\epsilon}}\Delta\left(\frac{1}{|x|^2}\right)\phi dx -2\epsilon^{-3}\iint_{\partial \Omega_{\epsilon}} \phi dS_x \tag{A}$$ The pointwise Laplacian should not be casually dismissed: it is not locally integrable, hence does not define a distribution. The formula (A) represents its regularization; the subtracted term tempers the singularity at $0$.

Using the divergence formula, one can compute $$ \iiint_{\Omega_{\epsilon}}\Delta\left(\frac{1}{|x|^2}\right) dx = 2\epsilon^{-3} (4\pi \epsilon^2)$$ which allows us to rewrite (A) as $$ \iiint_{\Omega_{\epsilon}}\Delta\left(\frac{1}{|x|^2}\right)(\phi-\phi(0))\,dx -2\epsilon^{-3}\iint_{\partial \Omega_{\epsilon}}( \phi-\phi(0)) dS_x \tag{B}$$ Here the second term tends to zero because $\phi(x) - \phi(0 ) = \nabla \phi(0) x + O(|x|^2)$, and the integral of $x$ over a sphere is zero. Final answer: the Laplacian is $$ \phi\mapsto \lim_{\epsilon\to 0} \iiint_{\Omega_{\epsilon}}\Delta\left(\frac{1}{|x|^2}\right)(\phi-\phi(0))\,dx $$ where you can put the formula for pointwise $\Delta\left(\frac{1}{|x|^2}\right)$ (some constant times $|x|^{-4}$) to make it more explicit.

This is not "pointwise Laplacian plus a constant", this is pointwise Laplacian regularized.

$\endgroup$
  • 1
    $\begingroup$ I cannot see that you show $\Delta \left( \frac{1}{|x|^2} \right) = \delta$ in the sense of distributions as you are not using derivatives of $\phi$. $\endgroup$ – md2perpe Dec 17 '17 at 21:23
  • $\begingroup$ (a) I use $\Delta \phi$ and $\nabla \phi$; (b) the Laplacian of this function is not the Dirac delta. You may be thinking of $1/|x|$. $\endgroup$ – user357151 Dec 17 '17 at 21:27
  • $\begingroup$ Yes, I was too fast when I wrote that formula. $\endgroup$ – md2perpe Dec 17 '17 at 21:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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