Two-dimensional Toda--Arnoldi correspondence: Holomorphic Krylov geometry and counterdiabatic transport
arXiv:2608.04850
2026
Architecture
2 ideas extracted · analyzed Aug 31, 2026
What the math gives to ML
The paper exposes a finite-dimensional Krylov geometry in which Arnoldi's non-Hermitian upper-Hessenberg reduction has a closed local sector: Gram determinants generate Toda tau functions, while their ratios are the squared subdiagonal Arnoldi coefficients. These coefficients measure how much genuinely new direction each application of a learned operator contributes, and they vanish at rank loss or Arnoldi breakdown. This suggests neural modules with adaptive Krylov depth and geometry-derived gates instead of treating all polynomial powers equally. The Arnoldi-frame connection also suggests a counterdiabatic correction that transports hidden representations smoothly while a learned operator changes during training.
Ideas from this paper
△ Mechanism confirmed, baseline not beaten
2026
Replace a fixed-order polynomial or recurrent state-space block by an Arnoldi basis built from a learned operator and the current input, and use subdiagonal coefficients as geometry-aware gates over Krylov depth. The gates quantify how much genuinely new direction each operator application contributes, allowing the layer to stop early near Krylov breakdown and suppress redundant or unstable directions.
Useful7/10
Difficulty5/10
Novelty7/10
Unverified
2026
When a learned operator changes during training, add a frame-connection correction that transports its current Arnoldi representation instead of allowing hidden states to jump between evolving spectral directions. This is a geometry-aware residual or optimizer correction intended to reduce representation drift during aggressive learning-rate schedules, fine-tuning, and continual learning.
Useful6/10
Difficulty7/10
Novelty8/10