My exercise is to proof the divergence theorem from Gauß for a radial invariant vector field on a Ball centered a $0$ and radius $R$ through actually calculating both integrals.

So let $f:A\rightarrow \mathbb{R}^n, f(x)=g(\lVert x\rVert)x$ for some $g:\mathbb{R}^n\rightarrow\mathbb{R}$ and $A=\{x\in\mathbb{R}: \lVert x\rVert \leq R\}$. Then for all $x\in A$ is the standardized normal vector $n=\frac{x}{\lVert x\rVert}$. I already calculated $$\int_{\partial A} f(x) \cdot n\ S(x) \, dS =\int_{\partial A}g(\lVert x\rVert)x \cdot \frac{x}{\lVert x\rVert} S(x) \, dS =\int_{\partial A}g(R)R S(x) \, dS =g(R)RS(A),$$ where $S(A)$ denotes the volume of the surface of A (which is $2 R^{n-1} \frac{\pi^{n/2}}{\Gamma(n/2)}$).

However, I am having problems calculating $\int_{A} \operatorname{div}(f(x)) \, d^nx$. First I tried to calculate the divergence of $f(x)=g(\lVert x\rVert)x$: $$\frac{\partial}{\partial x_i}f_i(x)=\frac{\partial}{\partial x_i}g(\lVert x\rVert)x_i=\frac{g'(\lVert x\rVert)x_i^2}{\lVert x\rVert}+g(\lVert x\rVert).$$ Here I am already not sure whether this is right or wrong. I used the product and the chain rule to calculate the derivation. Then I get: $$\int_{A} \operatorname{div}(f(x)) d^nx=\int_{A}g'(\lVert x\rVert)\lVert x\rVert+ng(\lVert x\rVert) \, d^nx.$$

Now I don't know how to proceed, so could you please help me?

  • $\begingroup$ Have you tried using a Dirac-Delta method instead? $\endgroup$
    – Johnq
    Jun 24 '17 at 0:26
  • $\begingroup$ Unfortunately I've never heard of the Dirac-Delta method $\endgroup$
    – MightyGuy
    Jun 24 '17 at 1:05
  • $\begingroup$ The Dirac-Delta distribution is a distribution that's defined always in terms of the integral $\int^{\infty}_{-\infty}f(x)dx = f(0)$. I usually see it in situations where you have to do something like a volume integral around a singularity. You may be able to throw that in to simplify some things. For more info: mathworld.wolfram.com/DeltaFunction.html $\endgroup$
    – Johnq
    Jun 24 '17 at 1:13
  • $\begingroup$ @Johnq That's really not going to help here. $\endgroup$
    – Chappers
    Jun 24 '17 at 1:49
  • $\begingroup$ @Chappers : Out of curiosity, why not? I mean, your approach seems pretty clean. I'm just curious as to why you wouldn't go about it like that here? $\endgroup$
    – Johnq
    Jun 24 '17 at 1:52

Write $\lVert x \rVert = r$ for brevity. Then $\partial_i r^2 = 2x_i$, so $\partial_i r = \partial_i \sqrt{r^2} = \frac{1}{2\sqrt{r^2}}2x_i = x_i/r $. Now, I think you actually want the field to be $g(r) \hat{x} = xg(r)/r $ (so that the radial direction given to it is as a unit vector) and so $$ \frac{\partial}{\partial x_i} \left(g(r)\frac{x_i}{r}\right) = \frac{g(r)}{r}\operatorname{div}{x}+\left( \frac{g'(r)}{r} -\frac{g(r)}{r^2} \right)\frac{x_ix_i}{r} = g'(r)+\frac{n-1}{r}g(r). $$ Now, it is useful to note that this is $$ \frac{1}{r^{n-1}}(r^{n-1}g(r))'. $$

The radial integration measure in $n$ dimensions is found by taking the volume of a spherical annulus as the width goes to zero, which is $S(r) \, dr= \frac{2\pi^{n/2}}{\Gamma(n/2)} r^{n-1}\, dr$. Hence the divergence theorem gives $$ S(1) R^{n-1}g(R) = S(1)\int_0^R \frac{1}{r^{n-1}}(r^{n-1}g(r))' r^{n-1} \, dr, $$ which is clearly the case using the Fundamental Theorem of Calculus.


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.