Spectral projections of an anharmonic oscillator with complex polynomial potential
Boris Mityagin, Petr Siegl
TL;DR
We analyze the non-selfadjoint Schrödinger-type operator \(L = -\dfrac{d^2}{dx^2} + V(x)\) with the complex polynomial potential \(V(x) = x^{2a} + i x^b\) and establish that for the regime \(a-1 < b < 2a\) the spectral projections \(\{P_n\}\) fail to form a (Riesz) basis in \(L^2(\mathbb{R})\). A novel resolvent identity combined with a gauge function \(F\) built from an infinite product enables a regularized resolvent expansion and a partial fraction decomposition of \(1/F\), linking bounds on \|P_n\| to bounds on the resolvent \|(z-L)^{-1}\|. The key finding is that \limsup_{n\to\infty} \|P_n\|/\exp(\gamma n^{\sigma}) = \infty\ for small \gamma>0\) with \(\sigma = \frac{b-(a-1)}{1+a} \in (0,1)\), proving absence of a basis. The framework further yields resolvent-growth bounds and basis-implications for related models, including imaginary even/odd oscillators and conjugated real oscillators, illustrating how exponential-type projection growth drives non-basis behavior in a broad class of non-selfadjoint perturbations.
Abstract
For a broad class of polynomial potentials $V$, with an important and instructive representative being $V(x) = x^{2a} + i x^b$, $x \in \mathbb R$, $a, b \in \mathbb N$, we show that the system of spectral projections $\{P_n\}_n$ of an anharmonic operator $L = - (\mathrm{d}/ \mathrm{d}x)^2 + V(x)$ does not generate a (Riesz) basis in $L^2(\mathbb R)$ if $a - 1 < b < 2a$. Moreover, for $σ= [b - (a - 1)]/(1 + a)$ and $γ> 0$ small enough, $\limsup_n \|P_n\|/ \exp(γn^σ) = \infty$. Proofs are based on two groups of results which are of great interest on their own: (a) relationship between behavior (growth) of the norms of projections $\|P_n\|$ and of the resolvent $\|(z - L)^{-1}\|$ outside of the spectrum $σ(L)$; (b) partial fraction decompositions of special meromorphic functions $1/F$ where $F(w) = \prod_{k=1}^\infty \left( 1 + \frac{w}{a_k} \right)$, $a_{k+1} \geq a_k>0$, $k \in \mathbb N$, and the generalization of the first resolvent identity.
