Let $\Omega$ be a domain in $\mathbb{R}^n$ with smooth boundary, and let $L$ be a negative definite second order elliptic differential operator defined with $\mathcal{D}(L) \subset H^2(\Omega)$, given by either Dirichlet or Neumann boundary conditions. My question is, for $f \in L^1(\Omega)$, do we have the following for small time $t$: $$ \Vert e^{tL}f \Vert_{L^1(\Omega)} \leq C\Vert f\Vert_{L^1(\Omega)}? $$

The reason why I expect this to hold is, I expect something like $e^{tL}f \to f$ pointwise almost everywhere. Is this true? In particular, does $\Omega$ need to be bounded for this? Thanks in advance.


I can give an answer for divergence-type operators. Let $$ Q(u,v)=\int_\Omega a\nabla u\cdot\nabla v\,dx $$ for $a=(a_{jk})$ with $a_{jk}\in L^\infty_{\mathrm{loc}}(\Omega)$ and $\sum_{j,k}a_{jk}(x)\xi_j\xi_k\geq c|\xi|^2$ for some $c>0$ and all $x\in\Omega$, $\xi\in\mathbb{R}^n$ (indeed, it suffices to assume the strong ellipticity condition only locally). Then $Q$ is a closed form on $D(Q)=\{u\in L^2(\Omega)\mid \int a\nabla u\cdot\nabla u<\infty\}$. If $a_{jk}\in L^\infty(\Omega)$, then $D(Q)=H^1(\Omega)$.

Furthermore, the generator of $Q$ is the divergence-type operator $$ Lu=\mathrm{div}(a\nabla u) $$ with Neumann boundary conditions whenever you can make sense of that expression (say, if $a_{jk}\in C^\infty$).

If one restricts $Q$ to $D(Q_0)=\overline{C_c^\infty(\Omega)}^{\lVert\cdot\rVert_Q}$, then the generator is the corresponding divergence-type operator with Dirichlet boundary conditions.

The important property of $Q$ is that it is a Dirichlet form (see [MR92], Sectiion II.2), that is, if $u\in D(Q)$, then $(u\vee 0)\wedge 1\in D(Q)$ and $Q((u\vee 0)\wedge 1)\leq Q(u)$. The same is true for $Q_0$. Essentially, one only needs a sufficiently strong chain rule to see that $$ Q((u\vee 0)\wedge 1)=\int_\Omega 1_{0\leq u\leq 1} a\nabla u\cdot \nabla u\,dx\leq Q(u). $$ By abstract theory, this property of $Q$ is equivalent to $0\leq e^{tL}u\leq 1$ for all $0\leq u\leq 1$ (such semigroups are called sub-Markovian). This equivalence was probably first stated in [BD58], but without proof. A proof can be deduced from the more general main theorem in [Ouh96] (of course, there are also direct proofs, I just don't have a reference at hand).

Since $(e^{tL})$ is sub-Markovian, it extends to a strongly continuous contraction semigroup on $L^\infty(\Omega)$, and by taking adjoints one also gets a strongly continuous contraction semigroup on $L^1(\Omega)$. Since $^{tL}$ is symmetric on $L^2$, this construction yields the original semigroup on $L^1\cap L^2$.

So, to answer your question:

Yes, the inequality holds, you can choose $C=1$, you do not even have to restrict to small times, $e^{tL}f$ converges to $f$ in $L^1$-norm as $t\to 0$, and it does not matter whether $\Omega$ is bounded or not.

[MR92] Ma, Röckner. Introduction to the theory of (non-symmetric) Dirichlet forms.
[BD58] Beurling, Deny. Espaces de Dirichlet. I. Le cas élémentaire.
[Ouh96] Ouhabaz. Invariance of closed convex sets and domination criteria for semigroups.

  • $\begingroup$ Very nice answer, thanks! One question, if I may: if we were on a compact setting, by the Hille-Yosida theorem, we would have $e^{tL} f \to f$ in $L^2$-norm, which would imply, $e^{tL}f \to f$ in $L^1$ norm. In that case, we would not need divergence form, correct? $\endgroup$ – user95820 Aug 6 '16 at 9:02
  • $\begingroup$ @user95820 Yes, as long as you can assure that $L$ generates a strongly continuous semigroup. I should add that divergence form is not really necessary for all I wrote above, you get a Dirichlet form under much more general conditions. Take a look at "Feller semigroups, Lp-sub-Markovian semigroups, and applications to pseudo-differential operators with negative definite symbols" by Farkas, Jacob and Schilling. Their main interest lies in pseudo-differential operators, but they also have an theorem dealing explicitely with quite general elliptic differential operators. $\endgroup$ – MaoWao Aug 6 '16 at 14:30
  • $\begingroup$ @MaoWao Does the same hold true (with $C=1$ and in $L^\infty$ norm) also for other general boundary conditions (for example, Robin)? Also, do you have a reference that states and proves such maximum principle? $\endgroup$ – user60665 Mar 6 at 18:08

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.