3
$\begingroup$

Let $\Omega \subset \mathbb R^2$ be a bounded domain with a smooth boundary. Let $$\bar{\partial}= \frac{1}{2} ( \partial_{x^1} + i \,\partial_{x^2}),$$ and let $G: L^2(\Omega)\to H^2(\Omega)$ be the Dirichlet Green's function on $\Omega$ that is defined by $GF:=u$ where $u\in H^2(\Omega)$ is the unique solution to $-\Delta u =F$ on $\Omega$ with $u|_{\partial \Omega}=0$.

Let us define the compact operator $T:L^2(\Omega)\to L^2(\Omega)$ via $$ T F = \bar{\partial}GF.$$

My question is to try to understand the spectrum of the operator $T$ and whether for instance it admits a discrete spectrum that provides a basis for $L^2(\Omega)$.

$\endgroup$
3
  • $\begingroup$ The only thing I see is compactness: $G$ is bounded form $L^2$ to $H^2$ and then $T$ is bounded form $L^2$ to $H^1$, hence compact from $L^2$ into itself. $\endgroup$ Apr 29 at 14:17
  • $\begingroup$ It actually turns out that the spectrum contains only zero and nothing else. So I guess the best next question to ask is what is the best resolvent estimates for T near zero. $\endgroup$
    – Ali
    Apr 29 at 23:03
  • $\begingroup$ Nice. How can you prove this? In vol 1 of Dunford Schwartz there is a result saying that the linear span of the generalized eigenfunctions is dense if the resolvent decays like $1/|\lambda|$ on certain arcs. Then this is not your case. $\endgroup$ Apr 30 at 8:11

1 Answer 1

1
$\begingroup$

I claim that the spectrum contains only zero, which amounts to proving that $T-\lambda$ has a bounded inverse for $\lambda$ away from zero. To see this, we need to construct a bounded inverse which I will denote by $A$. Let us first define the linear operator $A: C^{\infty}_c(\Omega)\to L^2(\Omega)$ by $u:=AF$ where $u$ is the unique solution to $$ -e^{\frac{1}{4\lambda}z} \Delta(e^{-\frac{1}{4\lambda}z}u) =\frac{1}{\lambda}\bar \partial u - \Delta u = \frac{1}{\lambda} \Delta F,\quad \text{on $\Omega$},$$ subject to $u|_{\partial \Omega}=0$. Note that $$ \|AF\|_{L^2(\Omega)} \leq C_\lambda \|F\|_{L^2(\Omega)}.$$ Subsequently, we can define $A:L^2(\Omega)\to L^2(\Omega)$ by continuous extension and density of $C^{\infty}_c(\Omega)$ in $L^2(\Omega)$.

$\endgroup$
1
  • $\begingroup$ I see. Writing $Tf-\lambda f=F$ and setting $u=Gf$, then $u=-\frac 1 \lambda e^{z/4 \lambda} G(F e^{-z/4\lambda})$ which shows that the resolvent satisfies an exponential estimate near zero (in 1d this is in fact the Volterra operator). Since 0 is not an eigenvalue ($Tf=0$ gives $f=\partial Tf=0$), 0 is not a pole of the resolvent, otherwise would be an eigenvalue. $\endgroup$ May 1 at 14:36

Your Answer

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

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