Randomized Gradient Descents on Riemannian Manifolds: Almost Sure Convergence to Global Minima in and beyond Quantum Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Malvetti, Emanuel, Arenz, Christian, Dirr, Gunther, Schulte-Herbrüggen, Thomas
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909675168989184
author Malvetti, Emanuel
Arenz, Christian
Dirr, Gunther
Schulte-Herbrüggen, Thomas
author_facet Malvetti, Emanuel
Arenz, Christian
Dirr, Gunther
Schulte-Herbrüggen, Thomas
contents We analyze convergence of gradient-descent methods on Riemannian manifolds. In particular, we study randomization of Riemannian gradient algorithms for minimizing smooth cost functions (of Morse-Bott type). We prove that randomized gradient descent methods, where the Riemannian gradient is replaced by a random projection of it, converge to a single local optimum almost surely despite the existence of saddle points. We consider both uniformly distributed and discrete random projections. We also discuss the time required to pass a saddle point. As a major application, we consider ground-state preparation through quantum optimization over the unitary group. In mathematical terms our randomized algorithm applied to the trace function $U \to \operatorname{tr}(AUρU^*)$ almost surely converges to its global minimum. The minimum corresponds to the smallest eigenvalue (ground state) of the selfadjoint operator $A$ (Hamiltonian) if $ρ$ is a rank-one projector (pure state). In this setting, one can efficiently replace the uniform random projections by implementing so-called discrete unitary 2-designs.
format Preprint
id arxiv_https___arxiv_org_abs_2405_12039
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Randomized Gradient Descents on Riemannian Manifolds: Almost Sure Convergence to Global Minima in and beyond Quantum Optimization
Malvetti, Emanuel
Arenz, Christian
Dirr, Gunther
Schulte-Herbrüggen, Thomas
Optimization and Control
Quantum Physics
65K10 (Primary) 53B21, 37H10 (Secondary)
We analyze convergence of gradient-descent methods on Riemannian manifolds. In particular, we study randomization of Riemannian gradient algorithms for minimizing smooth cost functions (of Morse-Bott type). We prove that randomized gradient descent methods, where the Riemannian gradient is replaced by a random projection of it, converge to a single local optimum almost surely despite the existence of saddle points. We consider both uniformly distributed and discrete random projections. We also discuss the time required to pass a saddle point. As a major application, we consider ground-state preparation through quantum optimization over the unitary group. In mathematical terms our randomized algorithm applied to the trace function $U \to \operatorname{tr}(AUρU^*)$ almost surely converges to its global minimum. The minimum corresponds to the smallest eigenvalue (ground state) of the selfadjoint operator $A$ (Hamiltonian) if $ρ$ is a rank-one projector (pure state). In this setting, one can efficiently replace the uniform random projections by implementing so-called discrete unitary 2-designs.
title Randomized Gradient Descents on Riemannian Manifolds: Almost Sure Convergence to Global Minima in and beyond Quantum Optimization
topic Optimization and Control
Quantum Physics
65K10 (Primary) 53B21, 37H10 (Secondary)
url https://arxiv.org/abs/2405.12039