It is often quoted in physics textbooks for finding the electric potential using Green's function that

$$\nabla ^2 \left(\frac{1}{r}\right)=-4\pi\delta^3({\bf r}),$$

or more generally

$$\nabla ^2 \left(\frac{1}{|| \vec x - \vec x'||}\right)=-4\pi\delta^3(\vec x - \vec x'),$$

where $\delta^3$ is the 3-dimensional Dirac delta distribution. However I don't understand how/where this comes from. Would anyone mind explaining?


The gradient of $\frac1r$ (noting that $r=\sqrt{x^2+y^2+z^2}$) is

$$ \nabla \frac1r = -\frac{\mathbf{r}}{r^3} $$ when $r\neq 0$, where $\mathbf{r}=x\mathbf{i}+y\mathbf{j}+z\mathbf{k}$. Now, the divergence of this is

$$ \nabla\cdot \left(-\frac{\mathbf{r}}{r^3}\right) = 0 $$ when $r\neq 0$. Therefore, for all points for which $r\neq 0$,

$$ \nabla^2\frac1r = 0 $$ However, if we integrate this function over a sphere, $S$, of radius $a$, then, applying Gauss's Theorem, we get

$$ \iiint_S \nabla^2\frac1rdV = \iint_{\Delta S} -\frac{\mathbf{r}}{r^3}.d\mathbf{S} $$ where $\Delta S$ is the surface of the sphere, and is outward-facing. Now, $d\mathbf{S}=\mathbf{\hat r}dA$, where $dA=r^2\sin\theta d\phi d\theta$. Therefore, we may write our surface integral as $$\begin{align} \iint_{\Delta S} -\frac{\mathbf{r}}{r^3}.d\mathbf{S}&=-\int_0^\pi\int_0^{2\pi}\frac{r}{r^3}r^2\sin\theta d\phi d\theta\\ &=-\int_0^\pi\sin\theta d\theta\int_0^{2\pi}d\phi\\ &= -2\cdot 2\pi = -4\pi \end{align}$$ Therefore, the value of the laplacian is zero everywhere except zero, and the integral over any volume containing the origin is equal to $-4\pi$. Therefore, the laplacian is equal to $-4\pi \delta(\mathbf{r})$.

EDIT: The general case is then obtained by replacing $r=|\mathbf{r}|$ with $s=|\mathbf{r}-\mathbf{r_0}|$, in which case the function shifts to $-4\pi \delta(\mathbf{r}-\mathbf{r_0})$

  • 2
    $\begingroup$ Thank you very much, Glen! $\endgroup$ – Jenna Apr 21 '13 at 14:59
  • 7
    $\begingroup$ Comment to the answer (v2): Playing the devil's advocate: The usual divergence theorem assumes that the vector field ${\bf F}$ is $C^1$. But in this case, the vector field ${\bf F}=-\frac{\bf r}{r^3}$ is singular at $r=0$. So how can the use of the divergence theorem be justified? $\endgroup$ – Qmechanic May 22 '15 at 14:36
  • 1
    $\begingroup$ @Qmechanic - for the same reason as why we use the Dirac Delta function... which isn't technically a function, but a distribution. Note that the usual definition of integration doesn't apply to the dirac delta function in one dimension, because it requires that the function be real-valued (or complex-valued, as appropriate). $\endgroup$ – Glen O May 22 '15 at 17:20
  • 2
    $\begingroup$ A distribution is usually evaluated on an arbitrary test function (within a certain class of functions). The answer (v2) makes no mentioning of test functions. $\endgroup$ – Qmechanic May 23 '15 at 14:38
  • 2
    $\begingroup$ @Cookie - there is undoubtedly some sort of generalisation, but I don't believe other cases are as simple as $n=3$. I believe the useful observation is that $\nabla\cdot (\mathbf{r}/r^n) = 0$ for $r\neq 0$ and $n$ dimensions. $\endgroup$ – Glen O Jun 4 '16 at 8:28

I'm new around here (so suggestions about posting are welcome!) and want to give my contribution to this question, even though a bit old. I feel I need to because using the divergence theorem in this context is not quite rigorous. Strictly speaking $1/r$ is not even differentiable at the origin. So here's a proof using limits of distributions.

Let $\mathbf{x}\in\mathbb{R}^{3}$ and $r=|\mathbf{x}|=\sqrt{x^2+y^2+z^2}$. It is evident from direct calculation that $\nabla^{2}\left(\frac{1}{r}\right)=0$ everywhere except in $\mathbf{x}=0$, where it is in fact not defined. Thus, the integral of $\nabla^{2}\left(\frac{1}{r}\right)$ over any volume non containing the origin is zero.

So let $\eta>0$ and $r_{\eta}=\sqrt{x^2+y^2+z^2+\eta^2}$. Obviously $\lim\limits_{\eta\rightarrow 0}r_{\eta}=r$. Direct calculation brings \begin{equation} \nabla^{2}\left(\frac{1}{r_{\eta}}\right) = \frac{-3\eta^{2}}{r_{\eta}^5} \end{equation} Now let us consider the distribution represented by $\nabla^{2}\left(\frac{1}{r_{\eta}}\right)$ and let $\rho$ be a test function (for example in the Schwartz space). I use Dirac's bra-ket notation to express the action of a distribution over a test function. Let $S^{2}$ be the unit sphere. Thus we calculate \begin{align} \lim\limits_{\eta\rightarrow 0}\left.\left\langle \nabla^{2}\left(\frac{1}{r_{\eta}}\right)\right|\rho\right\rangle &= \lim\limits_{\eta\rightarrow 0}\iiint\limits_{\mathbb{R}^3}\mathrm{d}^{3}x\, \nabla^{2}\left(\frac{1}{r_{\eta}}\right)\rho(\mathbf{x})\\ &=\lim\limits_{\eta\rightarrow 0}\left\{\iiint\limits_{\mathbb{R}^3\setminus S^{2}}\mathrm{d}^{3}x\, \frac{-3\eta^{2}}{r_{\eta}^5}\rho(\mathbf{x})+ \iiint\limits_{S^{2}}\mathrm{d}^{3}x\, \frac{-3\eta^{2}}{r_{\eta}^5}\rho(\mathbf{x})\right\}\\ &=\lim\limits_{\eta\rightarrow 0}\iiint\limits_{S^{2}}\mathrm{d}^{3}x\, \frac{-3\eta^{2}}{r_{\eta}^5}\rho(\mathbf{x}) \end{align} Where the limit of the first of the integrals in the curly braces is zero (easy to show, referring to the laplacian of $1/r$, no need for $\eta$ in sets not containing the origin). Now Taylor expand $\rho$ at $\mathbf{x}=0$ and integrate using spherical polar coordinates: \begin{equation} \lim\limits_{\eta\rightarrow 0}\int_{0}^{\pi}\mathrm{d}\theta\,\sin\theta\int_{0}^{2\pi}\mathrm{d}\varphi\,\int_{0}^{1}\mathrm{d}t\,\frac{-3\eta^{2}t^{2}}{(t^{2}+\eta^{2})^{5/2}}(\rho(0)+O(t^{2})) \end{equation} Integrating you will get that all the terms contained in $O(t^2)$ vanish as $\eta\rightarrow 0$, while the term with $\rho(0)$ remains. In fact you get \begin{align} \lim\limits_{\eta\rightarrow 0}\frac{-4\pi\rho(0)}{\sqrt{1+\eta^{2}}}&=-4\pi\rho(0)\\ &=\langle -4\pi\delta_{0}^{(3)}|\rho\rangle \end{align}

From this argument one defines the limit distribution \begin{equation} \nabla^{2}\left(\frac{1}{r}\right):=\lim\limits_{\eta\rightarrow 0}\nabla^{2}\left(\frac{1}{r_{\eta}}\right)=-4\pi\delta_{0}^{(3)} \end{equation} The generalization to $r=|\mathbf{x}-\mathbf{x}_{0}|$ is obvious.

  • 1
    $\begingroup$ Great answer! However, could you clarify on the expansion in Taylor series? It seems to me you are supposing that $\rho(\mathbf{x})$ is only radial. $\endgroup$ – rmk236 Aug 18 '15 at 20:17
  • 2
    $\begingroup$ @rmk236 It doesn't look that way to me. Taylor expanding inside the integral you get the $\rho(0)$ term and then all the others of course. But all those other terms (which account for the possibility of $\rho$ to be not radial) are included in the $O(t^{2})$. If you need the actual terms or you still have doubts I could carry out the calculation again and post more details on that part. $\endgroup$ – Daniele Oriani Aug 19 '15 at 22:25
  • $\begingroup$ This approach is the most popular in electrodynamics (see Chap. 1 of Classical electrodynamics, Jackson). Neverteless, the first one is more intuitive. $\endgroup$ – nunodsousa Oct 23 '19 at 8:06

Here is a back-alley derivation using Fourier transform properties. Take the Fourier transform of $\frac{1}{4 \pi r}$ to get $\frac{1}{k^2}$. Therefore, $$ \frac{1}{4\pi r} = \int \frac{d^3k}{(2\pi)^3} \frac{e^{-ik\cdot r}}{k^2} $$ Now, apply $-\nabla^2$ to get $$ -\nabla^2 \frac{1}{4\pi r} = \int \frac{d^3k}{(2\pi)^3} e^{-ik\cdot r} $$ This is the delta function, but we can explicitly put in our test function: \begin{align*} \int d^3r \left(-\nabla^2 \frac{1}{4 \pi r}\right) f(r) &= \int \frac{d^3k}{(2\pi)^3} \int d^3r e^{-ik\cdot r} f(r) \\ &= \int \frac{d^3k}{(2\pi)^3} \tilde{f}(k) \\ & = f(r=0). \end{align*}


We can use the simplest method to display the results, as shown below : -

$$ \nabla ^2 \left(\frac{1}{r}\right) = \nabla \cdot \nabla \left( \frac 1 r \right) = \nabla \cdot \frac {-1 \mathbf {e_r}} {r^2} $$

Suppose there is a sphere centered on the origin, then the total flux on the surface of the sphere is : -

$$ \text {Total flux} = 4 \pi r^2 \frac {-1} {r^2} = -4 \pi $$

Suppose the volume of the sphere be $ \mathbf {v(r)}$, so by the definition, the divergence is : -

$$ \lim_{\text {volume} \to zero} \frac {\text {Total Flux}} {\text {Volume}} = \lim_{\text {v(r)} \to 0} \left(\frac {-4 \pi} {v(r)}\right) $$

So obviously,

$$ \lim_{\text {r} \to 0} \left[ \nabla ^2 \left(\frac{1}{r}\right) \right]= \lim_{\text {r} \to 0} \left[ \nabla \cdot \nabla \left( \frac 1 r \right) \right]= \lim_{\text {r} \to 0} \left(\frac {-4 \pi} {v(r)} \right) = \text {infinite} $$

$$ \lim_{\text r\to 0} \int \nabla ^2 \left( \frac 1 r \right) dv(r) = \lim_{\text r\to 0}\int \frac {-4 \pi} {v(r)} dv(r) = -4\pi$$

Since the laplacian is zero everywhere except $r \to 0$, it is true and real that :-

$$ \nabla ^2 \left(\frac{1}{r}\right)=-4\pi\delta^3({\bf r}) $$


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.