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

On pg. 133 of Heat Kernels and Spectral Theory, Davies is studying the heat kernel $K(x,y,t)$ of the operator $H = -\Delta + |x|^{\alpha}$ for $\alpha > 0$. He wishes to prove a lower bound, and writes, "If $H_{B}$ is the operator obtained from $H$ by imposing Dirichlet boundary conditions on the surface of the ball with center $x$ and radius $1$, then $$K(x,x,t) \geq K_{B}(x,x,t)$$ for all $t > 0$." (This I see follows from the parabolic maximum principle and nonnegativity of $K$.) He then continues, "Moreover, $$|y|^{\alpha} \leq (|x| + 1)^{\alpha}$$ for all $y \in B$ so $$K_{B}(x,x,t) \geq K_{0}(x,x,t)\exp\left\{-(|x| + 1)^{\alpha}t\right\}$$ for all $t > 0$ where $K_{0}$ is the heat kernel of $-\Delta$ on $B$ with Dirichlet boundary conditions."

This is where he loses me. I'm aware that for, say, $V_{1}, V_{2} \in L^{\infty}(\mathbb{R}^{n})$ with $V_{1} \leq V_{2}$, the Feynman-Kac formula implies that the operator $H_{1} = -\Delta + V_{1}$ has heat kernel greater than or equal to that of $H_{2} = -\Delta + V_{2}$.

But I'm not sure if something analogous is the justification in Davies' argument, because I can't find a version of the Feynman-Kac formula that would apply in this setting.

Does anyone happen to know the proper explanation? Thank you.

share|cite|improve this question
Perhaps one could reverse the order of the two reductions? Instead of going from $-\Delta+|y|^\alpha$ on $R^n$ to $-\Delta+|y|^\alpha$ on $B$ with Dirichlet conditions to $-\Delta+(|x|+1)^\alpha$ on $B$ with Dirichlet conditions, one could use $\Delta + \max(|x|+1,|y|)^\alpha$ on $R^n$ as the intermediate operator and then the Feynman-Kac formula on $R^n$ could be used. – Terry Tao Aug 26 '13 at 11:06
Regarding the question of whether there is a Feynman-Kac formula assuming Dirichlet conditions, it seems that the formula would be signed rather than unsigned (basically one picks up a minus sign whenever one bounces off the boundary) which looks not particularly useful for proving comparison theorems. On the other hand, the heat kernel is still non-negative, so there may still be some probabilistic interpretation, perhaps after performing some additional cancellation... – Terry Tao Aug 26 '13 at 11:10
@TerryTao: thanks for the reply! What would you consider the best reference for the Feynman-Kac formula on $\mathbb{R}^{n}$? (I have seen Broderix, K. & Hundertmark, D. & Leschke, H.: Continuity properties of Schrödinger semigroups with magnetic fields mentioned.) – Michael Tinker Aug 26 '13 at 13:26
I guess Davies may have just used a second application of the maximum principle; namely, $u = K_{B} - K_{0}\exp\left\{(|x| + 1)^{\alpha}\right\}$ is a supersolution of $-\Delta + (|x|+1)^{\alpha}$ which vanishes on $\partial B$, so $u \geq 0$ inside $B$. – Michael Tinker Aug 26 '13 at 13:44
up vote 3 down vote accepted

Yes, if $H = -\Delta + V$ on a smooth domain $D$ with Dirichlet boundary conditions, then the solution to $\partial_t u = - H u$ with initial condition $u_0$ is given by $$ u(t,x) = E_x \Bigl[ \exp\Bigl(-\int_0^t V(X_s)\,ds\Bigr) u_0(X_s) 1_{t < \tau}\Bigr]\;, $$ where $X$ is a Brownian motion started at $x$ and $\tau$ is the first time when $X$ hits the boundary of $D$. I don't have a reference off the top of my head, but you can see this quite easily by arguing that $H$ is the limit (in resolvent sense) of $H_n$ where $H_n$ is defined on all of $\mathbf{R}^d$ and has potential $V_n$ which is equal to $V$ inside $D$ and $n$ outside $D$. (The intuition is that the Dirichlet boundary condition really corresponds to having a potential that is infinite outside of $D$.) For Neumann boundary conditions, you get a similar formula by using a reflected Brownian motion instead of a killed Brownian motion.

share|cite|improve this answer
Thanks Martin! Would it suffice for $V$ to be in $L^{q}(D)$ for some $q > 1$, or is something stronger required? Also, is $1_{\tau < t}$ supposed to read $1_{t < \tau}$? – Michael Tinker Aug 26 '13 at 22:50
So I see that we can take $u_{0}(x) = \delta(y - x)$ for fixed $y \in D$, and the appropriate inequality on kernels would you think this is what Davies had in mind? – Michael Tinker Aug 26 '13 at 22:52
You are right, it should be $t < \tau$, I fixed it. I suppose that Davies had something like this in mind, but of course I cannot speak for him ;-) Not sure what the minimal conditions on $V$ are, probably something like $L^q$ for $q > d/2$, see the paper "Schrödinger semigroups" by Barry Simon. – Martin Hairer Aug 27 '13 at 8:36

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.