[Paper Review] Resolvent estimates for non-self-adjoint operators via semi-groups

Johannes Sjoestrand|ArXiv.org|May 30, 2009

Spectral Theory in Mathematical Physics14 references3 citations

TL;DR

This paper establishes improved resolvent estimates for non-self-adjoint $h$-pseudodifferential operators in the semi-classical limit by introducing a novel semi-group-based method. Under a subellipticity condition, it proves that the resolvent extends into the numerical range up to distance $\mathcal{O}((h\ln\frac{1}{h})^{k/(k+1)})$ from certain boundary points, refining prior results by Dencker, Zworski, and Sjöstrand and matching a recent result by Bordeaux Montrieux for $k=2$. The method avoids classical calculus and instead uses microlocal semi-group estimates via Bargmann-FBI transforms.

ABSTRACT

We consider a non-self-adjoint $h$-pseudodifferential operator $P$ in the semi-classical limit ($h o 0$). If $p$ is the leading symbol, then under suitable assumptions about the behaviour of $p$ at infinity, we know that the resolvent $(z-P)^{-1}$ is uniformly bounded for $z$ in any compact set not intersecting the closure of the range of $p$. Under a subellipticity condition, we show that the resolvent extends locally inside the range up to a distance ${\cal O}(1)((h\ln \frac{1}{h})^{k/(k+1)})$ from certain boundary points, where $k\in \{2,4,...\}$. This is a slight improvement of a result by Dencker, Zworski and the author, and it has recently been obtained by W. Bordeaux Montrieux in a model situation where $k=2$. The method of proof is different from the one of Dencker et al, and is based on estimates of an associated semi-group.

Motivation & Objective

• To improve the known resolvent estimates for non-self-adjoint $h$-pseudodifferential operators near the boundary of the numerical range in the semi-classical limit.
• To provide a new proof technique for subellipticity-based resolvent extensions, differing from the Weyl-Hörmander calculus used in prior work.
• To extend the range of resolvent analyticity to $\mathcal{O}((h\ln\frac{1}{h})^{k/(k+1)})$ from boundary points under a subellipticity condition, improving upon the $\mathcal{O}(h^{k/(k+1)})$ bound.
• To unify and generalize results on spectral localization and pseudospectral behavior for non-self-adjoint operators using semi-group and microlocal analysis.

Proposed method

• The method employs microlocal semi-group estimates associated with the operator $P$, treating the semi-group as a Fourier integral operator with complex phase.
• It uses the Bargmann-FBI transform to convert the problem into a more tractable setting, enabling microlocal analysis of the semi-group kernel.
• The analysis relies on estimating the growth of the semi-group in the complex domain, particularly near points where the principal symbol $p$ satisfies a subellipticity condition of order $k \geq 2$.
• The proof avoids classical pseudodifferential calculus and instead uses a modified FBI representation to control the resolvent near the boundary of the numerical range.
• Key estimates are derived from the evolution of the transformed operator $\widehat{P}$, with bounds transferred back to the original resolvent via duality and regularity propagation.
• The method is adapted to handle both compact manifolds and $\mathbb{R}^n$, with careful treatment of symbol decay and behavior at infinity.

Experimental results

Research questions

• RQ1 Can the resolvent of a non-self-adjoint $h$-pseudodifferential operator be extended further into the numerical range than previously known under subellipticity?
• RQ2 Is it possible to improve the $\mathcal{O}(h^{k/(k+1)})$ resolvent extension bound to $\mathcal{O}((h\ln\frac{1}{h})^{k/(k+1)})$ using a new analytical method?
• RQ3 How does the semi-group approach compare to classical Weyl-Hörmander calculus in proving subellipticity-based resolvent estimates?
• RQ4 Can the method be extended to operators with non-elliptic behavior at infinity, such as the Kramers–Fokker–Planck operator?

Key findings

• The resolvent $(z - P)^{-1}$ extends analytically into the numerical range up to distance $\mathcal{O}((h\ln\frac{1}{h})^{k/(k+1)})$ from certain boundary points under a subellipticity condition of order $k \geq 2$.
• This improves upon the previous bound of $\mathcal{O}(h^{k/(k+1)})$ established by Dencker, Zworski, and Sjöstrand, matching a recent result by Bordeaux Montrieux for $k=2$.
• The method is fundamentally different from prior approaches, relying on semi-group estimates and microlocal analysis via the Bargmann-FBI transform rather than classical pseudodifferential calculus.
• The result applies to both $\mathbb{R}^n$ and compact manifolds, provided the symbol $p$ satisfies appropriate decay and growth conditions at infinity.
• For the non-self-adjoint harmonic oscillator $P = -h^2\frac{d^2}{dx^2} + ix^2$, the resolvent is bounded up to $\mathcal{O}((h\ln\frac{1}{h})^{2/3})$ from the boundary, matching the improved bound.
• The method suggests that the spectrum of the Kramers–Fokker–Planck operator may be localized more tightly, potentially improving the confinement from $\mathcal{O}(h^{2/3})$ to $\mathcal{O}(h\ln\frac{1}{h})^{2/3}$.