Moduli space of optimization algorithms

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pasechnyuk-Vilensky, Dmitry, Takáč, Martin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909917512728576
author Pasechnyuk-Vilensky, Dmitry
Takáč, Martin
author_facet Pasechnyuk-Vilensky, Dmitry
Takáč, Martin
contents We introduce a geometric and operator-theoretic formalism viewing optimization algorithms as discrete connections on a space of update operators. Each iterative method is encoded by two coupled channels-drift and diffusion-whose algebraic curvature measures the deviation from ideal reversibility and determines the attainable order of accuracy. Flat connections correspond to methods whose updates commute up to higher order and thus achieve minimal numerical dissipation while preserving stability. The formalism recovers classical gradient, proximal, and momentum schemes as first-order flat cases and extends naturally to stochastic, preconditioned, and adaptive algorithms through perturbations controlled by curvature order. Explicit gauge corrections yield higher-order variants with guaranteed energy monotonicity and noise stability. The associated determinantal and isomonodromic formulations yield exact nonasymptotic bounds and describe acceleration effects via Painlevé-type invariants and Stokes corrections.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18004
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Moduli space of optimization algorithms
Pasechnyuk-Vilensky, Dmitry
Takáč, Martin
Optimization and Control
Differential Geometry
65K05, 65L07, 49M27, 53C05
G.1.0; F.2.1; J.2
We introduce a geometric and operator-theoretic formalism viewing optimization algorithms as discrete connections on a space of update operators. Each iterative method is encoded by two coupled channels-drift and diffusion-whose algebraic curvature measures the deviation from ideal reversibility and determines the attainable order of accuracy. Flat connections correspond to methods whose updates commute up to higher order and thus achieve minimal numerical dissipation while preserving stability. The formalism recovers classical gradient, proximal, and momentum schemes as first-order flat cases and extends naturally to stochastic, preconditioned, and adaptive algorithms through perturbations controlled by curvature order. Explicit gauge corrections yield higher-order variants with guaranteed energy monotonicity and noise stability. The associated determinantal and isomonodromic formulations yield exact nonasymptotic bounds and describe acceleration effects via Painlevé-type invariants and Stokes corrections.
title Moduli space of optimization algorithms
topic Optimization and Control
Differential Geometry
65K05, 65L07, 49M27, 53C05
G.1.0; F.2.1; J.2
url https://arxiv.org/abs/2511.18004