Spectral Convergence Proofs for Multidimensional Random Feature Methods

This work proves spectral convergence for the random feature method (RFM) across Sobolev, Gevrey, and ultra-analytic function classes. It establishes high-probability approximation estimates where a single coefficient vector can achieve spectral accuracy for targets in a prescribed source ball.