English

On Solving Linear Systems in Sublinear Time

Data Structures and Algorithms 2026-02-23 v1

Abstract

We study \emph{sublinear} algorithms that solve linear systems locally. In the classical version of this problem the input is a matrix SRn×nS\in \mathbb{R}^{n\times n} and a vector bRnb\in\mathbb{R}^n in the range of SS, and the goal is to output xRnx\in \mathbb{R}^n satisfying Sx=bSx=b. For the case when the matrix SS is symmetric diagonally dominant (SDD), the breakthrough algorithm of Spielman and Teng [STOC 2004] approximately solves this problem in near-linear time (in the input size which is the number of non-zeros in SS), and subsequent papers have further simplified, improved, and generalized the algorithms for this setting. Here we focus on computing one (or a few) coordinates of xx, which potentially allows for sublinear algorithms. Formally, given an index u[n]u\in [n] together with SS and bb as above, the goal is to output an approximation x^u\hat{x}_u for xux^*_u, where xx^* is a fixed solution to Sx=bSx=b. Our results show that there is a qualitative gap between SDD matrices and the more general class of positive semidefinite (PSD) matrices. For SDD matrices, we develop an algorithm that approximates a single coordinate xux_{u} in time that is polylogarithmic in nn, provided that SS is sparse and has a small condition number (e.g., Laplacian of an expander graph). The approximation guarantee is additive x^uxuϵx| \hat{x}_u-x^*_u | \le \epsilon \| x^* \|_\infty for accuracy parameter ϵ>0\epsilon>0. We further prove that the condition-number assumption is necessary and tight. In contrast to the SDD matrices, we prove that for certain PSD matrices SS, the running time must be at least polynomial in nn. This holds even when one wants to obtain the same additive approximation, and SS has bounded sparsity and condition number.

Keywords

Cite

@article{arxiv.1809.02995,
  title  = {On Solving Linear Systems in Sublinear Time},
  author = {Alexandr Andoni and Robert Krauthgamer and Yosef Pogrow},
  journal= {arXiv preprint arXiv:1809.02995},
  year   = {2026}
}
R2 v1 2026-06-23T03:59:25.111Z