Dirac resonances as non-self-adjoint eigenvalues

What the math gives to ML

The paper gives a constructive complex-coordinate distortion that turns resonances, defined through meromorphically continued resolvents, into discrete eigenvalues of a non-self-adjoint distorted operator while preserving localized resolvent information. The transferable asset is the combination of an identity deformation near a compact interaction region, controlled deformation outside it, and compactness-based spectral discretization. This suggests a complex-stretched boundary module for neural operators that model wave or scattering dynamics, allowing outgoing radiation to be damped without corrupting the physical interior. A first experiment should test whether this improves resonance estimation and reduces boundary-reflection errors compared with ordinary finite-domain boundary conditions.