Abstract:It has been known since the work of Avakumovíc, Hörmander and Levitan that, on any compact smooth Riemannian manifold, if $-\Delta_g \psi_\lambda = \lambda \psi_\lambda$, then $\|\psi_\lambda\|_{L^\infty} \leq C \lambda^{\frac{d-1}{4}} \|\psi_\lambda\|_{L^2}$. It is believed that, on manifolds of negative curvature, such a bound can be largely improved; however, only logarithmic improvements in $\lambda$ have been obtained so far. In the present paper, we obtain polynomial improvements over the previous bound in a generic setting, by adding a small random pseudodifferential perturbation to the Laplace-Beltrami operator.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to improve the \(L^\infty\) bound estimates of eigenfunctions on manifolds of negative curvature by adding small random pseudodifferential perturbations. Specifically, it is known that on any compact smooth Riemannian manifold, if \(-\Delta_g \psi_\lambda=\lambda \psi_\lambda\), then:
\[ \|\psi_\lambda\|_{L^\infty} \leq C \lambda^{\frac{d - 1}{4}} \|\psi_\lambda\|_{L^2}. \]
On manifolds of negative curvature, it is believed that this bound can be greatly improved. However, so far, there has only been a logarithmic improvement for \(\lambda\). The goal of this paper is to obtain a polynomial improvement in the general setting by adding a small random pseudodifferential perturbation to the Laplace - Beltrami operator.
### Main Results
The main results of the article are as follows:
Let \((X, g)\) be a compact smooth Riemannian manifold of negative sectional curvature, and \(0 < \mu_1 < \mu_2\). Consider an operator \(\tilde{P}_\delta^h\) of the form:
\[ \tilde{P}_\delta^h=e^{-it/h} P_\delta^h(-h^2 \Delta_g) e^{it/h P_\delta^h}, \]
where \(P_\delta^h =-\frac{h^2}{2} \Delta_g+\delta Q_\omega\), and \(Q_\omega\) is a self - adjoint pseudodifferential operator whose symbol \(q_\omega\) oscillates rapidly on a mesoscopic scale.
For a normalized eigenfunction \(\psi_h\) satisfying \(\tilde{P}_\delta^h \psi_h = E_h \psi_h\), where \(E_h\in(\mu_1, \mu_2)\), there exists a set \(\Omega_h\subset\Omega\) such that \(P(\Omega_h) = 1 - O(h^\infty)\), and for any \(\omega\in\Omega_h\), we have:
\[ \|\psi_h\|_{L^\infty} \leq C h^{1-\frac{d}{2}+\Gamma'-\epsilon} \|\psi_h\|_{L^2}, \]
where \(\Gamma'\) is defined as:
\[ \Gamma'=\min\left(\Gamma, \frac{(d - 1)\beta}{2}\right), \]
and \(\Gamma\) satisfies:
\[ \delta h^{-2\beta} \leq h^\Gamma\quad\text{and}\quad h^{1-\Gamma} \leq \delta h^{\beta/2}. \]
### Remarks
1. When \(d = 2\), we can choose \(\alpha=\frac{5}{7}\) and \(\beta=\frac{2}{7}-\epsilon\), thus obtaining \(\Gamma'=\frac{1}{7}-\frac{\epsilon}{2}\). This gives:
\[ \|\psi_h\|_{L^\infty} \leq C h^{-\frac{5}{14}-\epsilon} \|\psi_h\|_{L^2}. \]
2. When \(d = 3\), the optimal value of \(\Gamma'\) is \(\Gamma'=\frac{2}{9}-\epsilon\).
3. Estimates for other \(L^p\) norms can be obtained by interpolation, but these estimates are not better than Sogge's estimates.