A general framework for universal bounds on the analytic functional calculus

Abstract

We establish a general principle for bounding the matrix-valued analytic functional calculus $F\mapsto F(T)$ of a Hilbert space operator $T$, independently of the matrix size. It rests on two properties of an auxiliary Hilbert space of holomorphic functions: the accretivity of its moment map at $T$ and a uniform bound for its multipliers. The mechanism of OpenAI's direct proof of the complete Crouzeix conjecture is the instance in which these properties come from the positivity of the double-layer potential and from norm estimates for Faber series of a convex domain. The main advantage of our method is that the auxiliary space can be chosen freely, to fit the class of operators, and a well-chosen space can give sharp constants, also on domains that are not convex, simply connected or bounded. On every intersection of two disks of the Riemann sphere, for operators with both closed disks as spectral sets, we obtain a complete bound independent of the two disks, which is sharp for the quantum annulus and the strip. For operators, bounded or m-sectorial, with numerical range in a sector, our bounds improve on the complete Crouzeix inequality for every opening; they are optimal for the right angle, answering a question of Crouzeix, and nearly optimal for acute angles. On Jordan domains with regular real-analytic boundary, a space built from the exterior conformal map and the Faber polynomials of the domain gives complete bounds for operators whose double-layer potential is bounded below. They are optimal for the disk, recover the complete Crouzeix inequality on convex domains, and improve on earlier bounds, including bounds of Crouzeix and Greenbaum, also on nonconvex domains.