Dirac–Frenkel dynamics with inertia for nonlinearly parametrized solutions of evolution problems

\name Matteo Raviola, Benjamin Peherstorfer · 2026

Read Dirac–Frenkel dynamics with inertia for nonlinearly parametrized solutions of evolution problems by \name Matteo Raviola, Benjamin Peherstorfer (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-24769.txt.

Description

Matteo Raviola and Benjamin Peherstorfer add inertia to Dirac Frenkel dynamics, a method for evolving solutions written in nonlinear parameters. Such parametrisations, seen in neural networks and mixture models, can be redundant. The plain dynamics may then give parameter motion that is not unique or is badly conditioned. Inertia lets useful velocity information persist from the recent past in directions the current data barely inform. Well informed directions keep following the standard dynamics as before.

The inertial version provably yields well posed parameter dynamics, the authors show. They also derive error bounds that can be checked after the fact. After time discretisation, each step solves the same kind of regularised least squares problem as the standard method. The previous velocity simply enters as an anchor. Numerical experiments then demonstrate the payoff. Robustness improves clearly when inertia is switched on. The method targets evolution problems with learned parametrisations.

Catalog record

CREATOR\name Matteo Raviola, Benjamin Peherstorfer
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE74 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.24769

Browse the full library · About the Tower of Babel