Alternating Minimal Energy Methods for Linear Systems in Higher Dimensions

Методы попеременной минимальной энергии для линейных систем в пространствах высокой размерности
Dmitry Savostyanov, Sergey Dolgov
2014-01-01

Alternating minimal energy methodsChemical master equationDensity matrix renormalization groupLow-rank tensor formatSymmetric positive definite linear systems
We propose algorithms for the solution of high-dimensional symmetrical positive definite (SPD) linear systems with the matrix and the right-hand side given and the solution sought in a low-rank format. Similarly to density matrix renormalization group (DMRG) algorithms, our methods optimize the components of the tensor product format subsequently. To improve the convergence, we expand the search space by an inexact gradient direction. We prove the geometrical convergence and estimate the convergence rate of the proposed methods utilizing the analysis of the steepest descent algorithm. The complexity of the presented algorithms is linear in the mode size and dimension, and the demonstrated convergence is comparable to or even better than the one of the DMRG algorithm. In the numerical experiment we show that the proposed methods are also efficient for non-SPD systems, for example, those arising from the chemical master equation describing the gene regulatory model at the mesoscopic scale.
1
Geometric convergence is proved, and convergence rates are estimated using analysis of the steepest descent method.
2
Numerical experiments demonstrate efficiency for non-SPD systems, including chemical master equations modeling gene regulation at the mesoscopic scale.
3
The algorithms have complexity linear in both mode size and dimension, with convergence comparable to or better than DMRG.
4
The methods optimize tensor-product components sequentially, similarly to density matrix renormalization group algorithms, while enlarging the search space with an inexact gradient direction.
5
The paper introduces low-rank algorithms for solving high-dimensional symmetric positive definite linear systems with known matrix and right-hand side.

High-dimensional symmetric positive definite linear systems with solutions represented in low-rank tensor-product format

Convergence, computational complexity, and efficiency of alternating minimal energy methods with inexact gradient-based search-space expansion for solving the systems

Publication Details
Publication Date
2014-01-01
Journal
Publisher
ISSN
Access Type
Author Information
Authors
Dmitry Savostyanov
Sergey Dolgov
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat
Make a presentation
100%