X is an n-dim Riemannian manifold with the Dirichlet form $$ \varepsilon (u,v) =-\int_X \langle \nabla u,\nabla v \rangle $$

for $u,v \in W^{1,2}(X)$. Let $P_t$ and $p_t(x,y)$ be the associate semigroup and heat kernel respectively.

If there exists $f \in L^2(x)$ such that $$ \varepsilon (u,v) =-\int_X \langle \nabla u,\nabla v \rangle= \int_X f\phi dvol $$ Then we say $\Delta u=f$. Denote $$D(\Delta)=\{u:u \in W^{1,2}(X),\Delta u \in L^2(X)\}$$(Obviously, the definition of $\Delta$ and $D(\Delta)$ coincide with the generator and the domain of the generator of the Dirichlet form).

Suppose $Ric(X) \ge -(n-1)$, then for any $h \in L^2(X)$, can we get $P_t h \in D(\Delta) \cap L^\infty$?

Fix $x=x_0$, then $p_t(x_0,y) \in D(\Delta) \cap L^\infty$ and $\Delta p_t(x_0,y) \in D(\Delta) \cap L^\infty$?



It follows from the spectral theorem that for any $h \in L^2$, we have $P_t h \in D(\Delta)$.

I think this is easiest to see from the multiplication operator form of the spectral theorem; i.e., there is a measure space $(\hat{X}, \mu)$, a measurable function $g : \hat{X} \to \mathbb{R}$, and a unitary operator $U : L^2(X) \to L^2(\hat{X})$ such that $\Delta = U^{-1} M_g U$. Here $M_g$ is the multiplication operator on $L^2(\hat{X})$ defined by $M_g v = gv$ with domain $D(M_g) = \{ v \in L^2(\hat{X}) : gv \in L^2(\hat{X})\}$. Since $\Delta$ is a negative definite operator, $g \le 0$ almost everywhere.

Now it is easy to check that $P_t = U^{-1} M_{e^{tg}} U$. But notice that $g e^{tg}$ is a bounded function. Thus for any $v \in L^2(\hat{X})$, we have $M_{e^{tg}} v \in D(M_g)$. Using $U$ to move back to $L^2(X)$, we see that for any $h \in L^2(X)$, $P_t h \in D(\Delta)$.

A similar argument shows that $P_t h \in D(\Delta^\infty)$.

To show $P_t h \in L^\infty$, one could use the heat kernel bounds shown by Li and Yau:

Li, Peter; Yau, Shing-Tung. On the parabolic kernel of the Schrödinger operator. Acta Math. 156 (1986), no. 3-4, 153–201.

They show, among other things, that in a complete Riemannian manifold with Ricci curvature bounded from below, we have Gaussian upper estimates for the heat kernel. In particular, for each $t$, $p_t$ is bounded. From this it is obvious that $P_t$ maps $L^1$ into $L^\infty$.

Now $P_t$ also maps $L^\infty$ into $L^\infty$, so it maps $L^1 + L^\infty$ into $L^\infty$; in particular it maps $L^2$ into $L^\infty$.

Since $P_t$ maps $L^\infty$ into $L^\infty$ and is symmetric, it also maps $L^1$ into $L^1$. So it maps $L^1$ into $L^1 \cap L^\infty$; in particular it maps $L^1$ into $L^2$. In particular, since $p_t(x_0, \cdot) = P_{t/2} [p_{t/2}(x_0, \cdot)]$, where $p_{t/2}(x_0, \cdot) \in L^1$, we have $p_t(x_0, \cdot) \in L^2$. Then by using $p_t(x_0, \cdot) = P_{t/2} [p_{t/2}(x_0, \cdot)]$ again, we get $p_t(x_0, \cdot) \in D(\Delta)$.

Finally, we note that for $h \in D(\Delta)$, we have $\Delta P_t h = P_t \Delta h$. So we have $$\Delta [p_t(x_0, \cdot)] = \Delta P_{t/2} [p_{t/2}(x_0, \cdot)] = P_{t/2} \Delta [p_{t/2}(x_0, \cdot)].$$ Since $p_{t/2}(x_0, \cdot) \in D(\Delta)$ as previously argued, we have $\Delta p_{t/2}(x_0, \cdot) \in L^2$. Since $P_{t/2}$ maps $L^2$ into $L^\infty \cap D(\Delta)$, we have $\Delta [p_t(x_0, \cdot)] \in L^\infty \cap D(\Delta)$ as desired.

| cite | improve this answer | |
  • $\begingroup$ :Your answer is quite helpful. 1 why "sine $P_t$ maps $L^\infty$ to $L^\infty$ and is symmetric, it also maps $L^1$ to $L^1$"? 2 $P_t$ maps $L^\infty$ to Lipschitz function? $\endgroup$ – wang mu Nov 1 '13 at 13:46
  • $\begingroup$ For every Dirichlet form, suppose $f \in L^2(X)$ but $\int_X |\nabla f|^2=\infty$. You still can get $P_t f \in W^{1,2}(X) $because $\int_X |\nabla P_tf|^2=-\int_X \langle \Delta P_t f,f \rangle<\infty$? $\endgroup$ – wang mu Nov 1 '13 at 14:09
  • $\begingroup$ @wangmu: I edited to explain why $P_t$ maps $L^1$ to $L^1$. I am not sure in general whether $P_t$ maps $L^\infty$ to Lipschitz functions. And yes, your argument is essentially the reason why $P_t$ maps $L^2$ into $W^{1,2}$. $\endgroup$ – Nate Eldredge Nov 1 '13 at 17:42
  • $\begingroup$ @wangmu: Oops, I didn't save the edit. But the idea is that the $L^1$ norm of $P_t h$ for $h \ge 0$ is the sup of $\int (P_t h) f$ over all $0 \le f \le 1$. (Restrict both $h$ and $f$ to also lie in $L^2$ if necessary.) Since $\int (P_t h)f = \int h P_t f$ and $0 \le P_t f \le 1$ we get the result. $\endgroup$ – Nate Eldredge Nov 1 '13 at 17:44
  • 1
    $\begingroup$ @NateEldredge $Ricci \ge 0$ alone is not enough to get a uniform bound $p_t(x,y) \le C(t)$ on the heat kernel. The non-collapsing condition $\inf_{x \in \mathbb{M}} \mu (B(x,1)) >0$ is also needed. $\endgroup$ – Fabrice Baudoin Jul 1 '14 at 19:10

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.