Solving Sparse Linear Systems Faster than Matrix Multiplication
Solving Sparse Linear Systems Faster than Matrix Multiplication
复制标题
比矩阵乘法更快地求解稀疏线性系统
DOI:
10.1137/1.9781611976465.31
复制
发表时间:
2021
期刊:
影响因子:
--
通讯作者:
Vempala, Santosh S.
中科院分区:
文献类型:
--
作者:
Peng, Richard;Vempala, Santosh S.
Can linear systems be solved faster than matrix multiplication? While there has been remarkable progress for the special cases of graph structured linear systems, in the general setting, the bit complexity of solving ann×nlinear systemAx=bisÕ(nω), whereω< 2.372864 is the matrix multiplication exponent. Improving on this has been an open problem even for sparse linear systems with poly(n) condition number.In this paper, we present an algorithm that solves linear systems in sparse matrices asymptotically faster than matrix multiplication for anyω> 2. This speedup holds for any input matrixAwitho(nω–1/log(κ(A))) non-zeros, whereκ(A) is the condition number ofA. For poly(n)-conditioned matrices withÕ(n) nonzeros, and the current value ofω, the bit complexity of our algorithm to solve to within any 1/poly(n) error isO(n2.331645).Our algorithm can be viewed as an efficient, randomized implementation of the block Krylov method via recursive low displacement rank factorizations. It is inspired by the algorithm of [Eberly et al. ISSAC ‘06 ‘07] for inverting matrices over finite fields. In our analysis of numerical stability, we develop matrix anti-concentration techniques to bound the smallest eigenvalue and the smallest gap in eigenvalues of semi-random matrices.
登录
查看更多内容
DOI:
--
发表时间:
2007
期刊:
影响因子:
--
作者:
M. G. Bruin
通讯作者:
M. G. Bruin
影响因子:
2
作者:
H. Nguyen;T. Tao;V. Vu
通讯作者:
V. Vu
DOI:
10.1214/19-aihp1032
发表时间:
2020
期刊:
Probabilités et Statistiques
影响因子:
--
作者:
Luh, Kyle;Vu, Van
通讯作者:
Vu, Van
DOI:
--
发表时间:
2009
期刊:
影响因子:
--
作者:
D. Spielman;N. Srivastava
通讯作者:
N. Srivastava
DOI:
--
发表时间:
2011
期刊:
Symposium on the Theory of Computing
影响因子:
--
作者:
L. Orecchia;Sushant Sachdeva;Nisheeth K. Vishnoi
通讯作者:
Nisheeth K. Vishnoi