10
$\begingroup$

Let $A(t)$ be a smooth family of positive definite operators on a Hilbert space $H$. Consider the operator $$D:= \frac{d}{dt}+A(t)$$ and let $U(t):H\to H$ be the evolution operator, i.e., $U(0)=I$ and $U'(t)+A(t)U(t)=0$.

Question: What condition on $A(t)$ guarantees that $U(t)$ is of trace class for all $t>0$?

More specifically, I am interested in situation when the following situation: Let $M$ be a compact manifold, $H=L^2(M)$, and let $A(t)$ be a smooth family of positive definite elliptic operators on $M$. If $A(t)$ is independent of $t$, then $U(t)=e^{-tA}$ is of trace class. So I hope that if $A'(t)$ is not very large, then $U(t)$ is still of trace class. What is the precise condition on $A'(t)$ which guarantees this?

Improve this question
$\endgroup$
3
  • $\begingroup$ Don't you want $U(t)$ is identity plus trace class? $\endgroup$
    – Peter Michor
    May 29, 2015 at 14:53
  • 2
    $\begingroup$ I don't think so. If $A$ is independent of $t$, then $U(t)$ is the heat operator and is of trace class. $\endgroup$
    – Maxim Braverman
    May 29, 2015 at 19:18
  • 1
    $\begingroup$ In the case $A$ is independent of $t$ and is elliptic, you need more to ensure that $e^{-tA}$ is of trace class. This is false if $Af=af$ for some $a\in(0,+\infty)$, but it is true if $A$ has order $m>0$. Presumably, you had in mind the case where $A$ is differential of second order. $\endgroup$
    – Denis Serre
    Oct 26, 2015 at 20:08

2 Answers 2

Reset to default
2
$\begingroup$

You may want to take a look at this Ph.D. thesis, especially at Thm. 5.5.2, which as far as I know is the sharpest currently available result concerning regularity of nonautonomous problems. The theorem states that the solution to your problem is at most in $V$ for each $t$, under suitable assumptions. Now, if you are thinking of a time-dependent diffusion equation, $V=H^1(\Omega)$, which is never trace-class-embedded in $L^2(\Omega)$, even in 1-dimension. In the autonomous case this is not a problem, since the semigroup law still allows us to conclude that $e^{-tA}$ is of trace class for all $t>0$ iff $V\hookrightarrow H$ is of $p$-Schatten class for some $p<\infty$.

But in the autonomous case this smoothing property seems to fail, so in general you won't get any more than $p$-Schatten class for all $p>1$ in the case of a nonautonomoum problem associated with a differential operator of order 1.

As suggested by @DenisSerre, things change if you consider higher order operators: if $A(t)$ is for all $t$ a 2k-th order differential operator with domain $H^k(\Omega)$, then you do have trace class evolution operators as soon as $k>d$.

Improve this answer
$\endgroup$
0
$\begingroup$

Maybe, the following paper or references therein can help:

  • Josef Teichmann: Another approach to some rough and stochastic partial differential equations, arXiv/0908.2814, Stochastics and Dynamics 11 (2011), no. 2-3, 535-550.
Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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