Problem statement

I took an exam, where I had the following task: Determine the electrostatic potential for the charge distribution $$\varrho(\mathbf{r}) = \sigma \cos\left(\frac{2 \pi}{L} x\right) \, \delta(y)$$

Approach 1

Solvong the Coulomb integral is futile, as $$ \phi(\mathbf{r}) = \int_{\mathbb{R}^3} \frac{\varrho(\mathbf{r}')}{|\mathbf{r} - \mathbf{r}'|} \, \mathrm{d}^3 r' = \sigma \int\limits_{-\infty}^{\infty} \mathrm{d}x \int\limits_{-\infty}^{\infty} \mathrm{d}y \int\limits_{-\infty}^{\infty} \mathrm{d}z \, \frac{\cos\left(\frac{2 \pi}{L} x\right) \, \delta(y)}{\sqrt{(x-x')^2 + (y-y')^2 + (z-z')^2}} $$ is not integrable.

Approach 2

Solving Poisson's equation is the only other possibility, that came to my mind. In electrical and magnetic units $$ \nabla^2 \phi(\mathbf{r}) = - 4 \pi \varrho(\mathbf{r}) $$ Now I started off Fourier-transforming the equation, where $\mathbf{k}^2 = k_x^2 + k_y^2 + k_z^2$ $$ \begin{aligned} \frac{1}{\sqrt{(2 \pi)^3}} \int_{\mathbb{R}^3} \nabla^2 \phi(\mathbf{r}) \, \mathrm{e}^{-\mathrm{i} \mathbf{k} \cdot \mathbf{r}} \, \mathrm{d}^3 r' &= - \frac{4 \pi }{\sqrt{(2 \pi)^3}} \int_{\mathbb{R}^3} \varrho(\mathbf{r}) \, \mathrm{e}^{-\mathrm{i} \mathbf{k} \cdot \mathbf{r}} \, \mathrm{d}^3 r' \\ \mathbf{k}^2 \, \hat{\phi}(\mathbf{k}) &= - 4 \pi \sigma \int_{\mathbb{R}^3} \cos\left(\frac{2 \pi}{L} x\right) \, \delta(y) \, \mathrm{e}^{-\mathrm{i} \mathbf{k} \cdot \mathbf{r}} \, \mathrm{d}^3 r' \\ \mathbf{k}^2 \, \hat{\phi}(\mathbf{k}) &= - 4 \pi \sigma \int\limits_{-\infty}^{\infty} \cos\left(\frac{2 \pi}{L} x\right) \, \mathrm{e}^{-\mathrm{i} k_x x} \, \mathrm{d}x \int\limits_{-\infty}^{\infty} \delta(y) \, \mathrm{e}^{-\mathrm{i} k_y y} \, \mathrm{d}y \int\limits_{-\infty}^{\infty} \mathrm{e}^{-\mathrm{i} k_z z} \, \mathrm{d}z \\ \mathbf{k}^2 \, \hat{\phi}(\mathbf{k}) &= - 4 \pi \sigma \cdot \frac{1}{2} \left( \delta\left(k_x - \frac{2 \pi}{L}\right) + \delta\left(k_x + \frac{2 \pi}{L}\right) \right) \cdot 1 \cdot \int\limits_{-\infty}^{\infty} \mathrm{e}^{-\mathrm{i} k_z z} \, \mathrm{d}z \end{aligned} $$ And here I'm stuck, because the leftover integral $$ \int\limits_{-\infty}^{\infty} \mathrm{e}^{-\mathrm{i} k_z z} \, \mathrm{d}z $$ does not converge.

Résumé: What am I doing wrong? Is my approach wrong, or did I miscalculate something?

  • 5
    $\begingroup$ @HenriMenke Please excuse the poor attitude of the user above. Your question seems well thought out, and you deserve a real answer, not discouragement, so I have put a small bounty on this question. $\endgroup$ – Alexander Gruber Jul 28 '13 at 21:48
  • $\begingroup$ Thank you very much @AlexanderGruber! The bounty brought an excellent answer. $\endgroup$ – Henri Menke Jul 29 '13 at 17:08

Note that in the unitary definition of the Fourier transform: $$\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} 1\mathrm{e}^{-\mathrm{i} k_z z} \, \mathrm{d}z=\sqrt{2\pi}\delta(k_z).$$ To see why, inverse Fourier transform both sides from momentum space back to the spatial domain: $$\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty}\left(\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} 1\mathrm{e}^{-\mathrm{i} k_z z} \,{dz}\right)\mathrm{e}^{\mathrm{i} k_z z} {dk_z}=\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty}\sqrt{2\pi}\delta(k_z)\mathrm{e}^{\mathrm{i} k_z z} {dk_z}=1,$$ so this is a Fourier transform pair of functions. More generally: $$\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \text{exp}\left[{\mathrm{i} k_y y_0}\right]\text{exp}\left[-{\mathrm{i} k_y y}\right] \,{dk_y}=\sqrt{2\pi}\delta(y-y_0)$$ and the previous identity follows from letting $y_0=0$.

Part 1
Lets see if your Dirac Delta function answer then checks out, define (note the corrected sign): $$\mathbf{k}^2 \, \hat{\phi}(\mathbf{k}) = 4\pi \sigma \hat{I_1}(k_x)\hat{I_2}(k_y) \hat{I_3}(k_z) , \tag{1}$$ $$\hat{I_1}(k_x)=\sqrt{2\pi}\frac{1}{2} \left( \delta\left(k_x - \frac{2 \pi}{L}\right) + \delta\left(k_x + \frac{2 \pi}{L}\right) \right),$$ $$\hat{I_2}(k_y)=\frac{1}{\sqrt{2\pi}},$$ $$\hat{I_3}(k_z)=\sqrt{2\pi}\delta(k_z),$$

then Fourier transform into the spatial domain:

$$\frac{1}{(2\pi)^{3/2}}\int\limits_{\mathbb{R}^3}\mathbf{k}^2 \, \hat{\phi}(\mathbf{k})\,e^{i\mathbf{k}\cdot\mathbf{r}}{d\mathbf{k}}=$$ $$\frac{4\pi\sigma}{(2\pi)^{3/2}}\int\limits_{-\infty}^{\infty} \hat{I_1}(k_x)\,e^{\mathrm{i} k_x x}{dk_x}\int\limits_{-\infty}^{\infty} \hat{I_2}(k_y)\,e^{\mathrm{i} k_y y}{dk_y}\int\limits_{-\infty}^{\infty} \hat{I_3}(k_z)\,e^{\mathrm{i} k_z z}{dk_z},$$

$$\frac{-1}{(2\pi)^{3/2}}\int\limits_{\mathbb{R}^3}\mathbf{k}^2 \, \hat{\phi}(\mathbf{k})\,e^{i\mathbf{k}\cdot\mathbf{r}}{d\mathbf{k}}=\nabla^2 \phi(\mathbf{r}),$$

\begin{aligned}\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \hat{I_1}(k_x)\,e^{\mathrm{i} k_x x}{dk_x}&=\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \frac{\sqrt{2\pi}}{2} \left[ \delta\left(k_x - \frac{2 \pi}{L}\right) + \delta\left(k_x + \frac{2 \pi}{L}\right) \right]\,e^{\mathrm{i} k_x x} {dk_x}\\&=\cos\left(\frac{2\pi x}{L}\right),\end{aligned}

$$\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \hat{I_2}(k_y)\,e^{\mathrm{i} k_y y}{dk_y}=\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \frac{1}{\sqrt{2\pi}} \,e^{\mathrm{i} k_y y} {dk_y}=\delta(y),$$

$$\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \hat{I_3}(k_z)\,e^{\mathrm{i} k_z z}{dk_z}=\frac{1}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \sqrt{2\pi}\delta(k_z) \,e^{\mathrm{i} k_z z} {dk_z}=1,$$

and therefore: $$\nabla^2 \phi(\mathbf{r}) = - 4 \pi \sigma\cos\left(\frac{2\pi x}{L}\right)\delta(y),\tag{2}$$ $$\nabla^2 \phi(\mathbf{r}) = - 4 \pi\varrho(\mathbf{r}) .$$

So it seems that works if we introduce the bit about the Fourier transform of $1$ being a delta function at zero.

Part 2

From $(1)$ we then have:

$$\hat{\phi}(\mathbf{k}) = \dfrac{4\pi \sigma \sqrt{2\pi}\frac{1}{2} \left[ \delta\left(k_x - \frac{2 \pi}{L}\right) + \delta\left(k_x + \frac{2 \pi}{L}\right) \right]\delta(k_z)}{k_x^2+k_y^2+k_z^2},$$

$$\begin{aligned}[b] \phi(\mathbf{r})&=\frac{1}{(2\pi)^{3/2}}\int\limits_{\mathbb{R}^3} \, \hat{\phi}(\mathbf{k})\,e^{i\mathbf{k}\cdot\mathbf{r}}{d\mathbf{k}},\\&=\frac{\sigma\cos\left(\dfrac{2\pi x}{L}\right)}{\sqrt{2\pi}}\int\limits_{-\infty}^{\infty} \dfrac{4\pi}{\left(\dfrac{2\pi}{L}\right)^2+k_y^2}\,e^{ik_y y}{dk_y},\\&=L\,\sigma\cos\left(\dfrac{2\pi x}{L}\right)\text{exp}\left[{-\dfrac{2\pi}{L}|y|}\right]\,,L\ge0.\end{aligned}\tag{3}$$ At $y\ne0$, $(3)$ solves the homogeneous equation: $$\nabla^2 \phi(\mathbf{r}) =0,$$ as required and at $y=0$ the derivative of $(3)$ w.r.t $y$ is not defined in the conventional sense as the limit of the derivative is not the same from both sides, however as @O.L points out in the comments, calculating derivatives in the distributional sense leads to the $\delta$ function at $y=0$ (see below).

  • 2
    $\begingroup$ The derivatives in the end should be calculated in the distributional sense. And the distributional derivative $\partial_{yy}$ of $\exp\left\{-\frac{2\pi|y|}{L}\right\}$ will produce the necessary delta function: $\partial_y$ produces a function with jump at $y=0$, and the next differentiation will give $\delta(y)$. $\endgroup$ – Start wearing purple Jul 29 '13 at 13:34
  • $\begingroup$ OK, cool cheers! $\endgroup$ – Graham Hesketh Jul 29 '13 at 13:49

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.