English

Solving Dense Linear Systems Faster Than via Preconditioning

Data Structures and Algorithms 2024-06-10 v2 Machine Learning Numerical Analysis Numerical Analysis Optimization and Control

Abstract

We give a stochastic optimization algorithm that solves a dense n×nn\times n real-valued linear system Ax=bAx=b, returning x~\tilde x such that Ax~bϵb\|A\tilde x-b\|\leq \epsilon\|b\| in time: O~((n2+nkω1)log1/ϵ),\tilde O((n^2+nk^{\omega-1})\log1/\epsilon), where kk is the number of singular values of AA larger than O(1)O(1) times its smallest positive singular value, ω<2.372\omega < 2.372 is the matrix multiplication exponent, and O~\tilde O hides a poly-logarithmic in nn factor. When k=O(n1θ)k=O(n^{1-\theta}) (namely, AA has a flat-tailed spectrum, e.g., due to noisy data or regularization), this improves on both the cost of solving the system directly, as well as on the cost of preconditioning an iterative method such as conjugate gradient. In particular, our algorithm has an O~(n2)\tilde O(n^2) runtime when k=O(n0.729)k=O(n^{0.729}). We further adapt this result to sparse positive semidefinite matrices and least squares regression. Our main algorithm can be viewed as a randomized block coordinate descent method, where the key challenge is simultaneously ensuring good convergence and fast per-iteration time. In our analysis, we use theory of majorization for elementary symmetric polynomials to establish a sharp convergence guarantee when coordinate blocks are sampled using a determinantal point process. We then use a Markov chain coupling argument to show that similar convergence can be attained with a cheaper sampling scheme, and accelerate the block coordinate descent update via matrix sketching.

Keywords

Cite

@article{arxiv.2312.08893,
  title  = {Solving Dense Linear Systems Faster Than via Preconditioning},
  author = {Michał Dereziński and Jiaming Yang},
  journal= {arXiv preprint arXiv:2312.08893},
  year   = {2024}
}

Comments

STOC 2024

R2 v1 2026-06-28T13:50:51.945Z