# Heat equation with dirac delta as source term

I have the following equation in $\mathbb{R}^2$,

$\partial_t u(x,t) - \Delta u(x,t) = \delta_0,$

with $u(x,0) = 0.$

I received this equation and I instantly got suspicious. I have tried to prove that anything that satisfies the PDE will have a singularity in 0 at all times, implying that actually there isn't a solution for this initial data.

My intuition comes from the fact that if a Dirac delta arises from differentiating, it should arise in at least one of the derivatives. If it there is one on the time derivative, it would mean that the solution is discontinuous in time, and this seems highly unlikely. In the case of the space derivatives, I think it should look like the Green function of the Laplacian (the steady state actually IS), which is singular at 0.

I have tried reducing to dimension 1 and obtaining an explicit solution by means of Fourier transform, and the solution gets so unmanageable that I abandoned that path.

Do you have any clues on this?

• Do you mean $\delta_0(x)$ or $\delta_0(x) \delta(t)$? – Ian Sep 1 '17 at 15:27

A solution exists indeed. Let us define the operator $L = -\Delta$ and rewrite the heat equation with source $f$ as $$\nabla_t u + L[u] = f$$ with initial conditions $u(x,0)=0$
A formal solution can be written in terms of exponential operator as $$u = \int_0 ^t e^{-(t-s)L } [f(s)] \mathrm{d}s$$ where the operator $e^{-tL}$ is given by an integral operator having the Green's function as kernel $$G[f] = \int_{R^n} G(x, \eta) f(\eta) \mathrm{d} \eta$$
Now we specialise the $f$ to a Dirac delta and using its "chief"property under integration, find the solution $$u(x,t) \int_0 ^t G(x, t-s) \mathrm{d}s$$ and further, as $$G = \frac{1}{(4 \pi t)^{\frac{n}{2}}} e^{-\frac{x^2}{4t}}$$ the solution $$u(x,t) =\frac{1}{(4 \pi t)^{\frac{n}{2}}} \int_0 ^t \frac{e^{-\frac{x^2}{4t}}}{s^{\frac{n}{2}}} \mathrm{d}s$$ is recovered, which I believe might be further simplified by using the Gamma function, if one needs to do thermal calculations. The main point is that by all means, a solution (of relevant practical importance, for what matters) exists.
• Thank you for your reply ! Now I got it. There are several things to correct in this answer (if anyone in the future comes here). 1) The operator $e^{-t\Delta}$ is the green function of the Heat Equation (when writting it just afterwards, there is a missing time dependence that confuses a bit). 2) "=" sign missing in the following equation. 3) The solution I believe is wrongfully written, it should be $u(x,t) = \int_0^t e^{-x^2/4s}/(4\pi s)^{n/2} ds$ – pancho Sep 2 '17 at 10:53