MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Dear Math Overflowers,

I'm looking for references on the parabolic Harnack inequality for distributional solutions of the heat equation on the whole space $$ \partial_t u=\Delta u\quad\text{and}\quad u(0)=u_0\quad\Leftrightarrow\quad u(t,x)=\int\limits_{R^d}\Gamma_t(x-y)u_0(y)dy. $$ Here $\Gamma_t(z)=\frac{e^{-\frac{|z|^2}{4t}}}{(4\pi t)^{d/2}}$ is the usual heat kernel and my initial data $u_0\geq 0$ is only in $L^1\cap L^p(R^d)$ for some $p>1$.

More precisely, I'm interested in the case when the initial datum $u_0$ is supported on a positive finite measure set $E=supp(u_0)$ and $u_0|_E\geq M>0$ for some constant $M$. I would like to control $u(t,x)\geq (\ldots)$ for points $x\in supp(u_0)$ and small times $t>0$, which really looks like Harnack inequality to me. The thing is that I have no information at all on the initial support (except that it has finite measure so it may not even be bounded) and I only have mere $L^1\cap L^p$ regularity for $u_0$. I strongly doubt that I can get pointwise estimates, but maybe $L^p$ estimates for $|u(t)|_{L^p(E)}\geq (\ldots)$?

Thank you in advance!

share|cite|improve this question
up vote 4 down vote accepted

Without knowing how the initial data is "spread," we can't get either pointwise or $L^p$ estimates independent of the measure of the initial support. Take for example the building block solution $v$ with initial data $\chi_{B_{\epsilon}}$. It follows from the representation formula that at time $t = \epsilon$, $v \leq C\epsilon^{n/2}$ on $B_{\epsilon}$. If we add together $\epsilon^{-n}$ translations of $v$ sufficiently "spread apart" (depending on $\epsilon$) to get $u$, it is easy to see (by exponential spatial decay) that the same bound will hold at all the points where $u$ was initially supported, which has measure like $1$.

No pointwise estimate follows immediately. To see no $L^p$, if we compute at $t = \epsilon$ we get $$\int_{E} u^p \leq C\epsilon^{-n} \int_{B_{\epsilon}} \epsilon^{np/2} \leq C\epsilon^{np/2}.$$

share|cite|improve this answer
great answer @Connor Mooney, thanks a lot! But darn, I whish you were wrong ;-) – leo monsaingeon Oct 24 '13 at 16:22

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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