Research questionWhen do second-order elliptic PDE solutions gain two orders of spectral Barron regularity for dimension-independent neural-network approximation?For elliptic PDEs on \(\mathbb{R}^d\), approximation depends on whether solution regularity can be strengthened in a dimension-robust way. The key issue is how spectral Barron regularity relates to the width required by neural networks.