Optimal Deterministic Fully Sparse Matrix Multiplication

Abstract

We give the first deterministic algorithm for fully sparse matrix multiplication that attains the optimal running-time exponent. This result matches the best previously known randomized algorithm running-time exponent. Given compatible matrices $A$ and $B$ over an arbitrary associative ring with identity, with $\operatorname{nnz}(A),\operatorname{nnz}(B)=O(n^{δ_{\mathrm{in}}})$ and $\operatorname{nnz}(AB)=O(n^{δ_{\mathrm{out}}})$, our algorithm finds the support of $AB$ and computes the product exactly in $$O\!\left(n^{β_R(δ_{\mathrm{in}},\min\{δ_{\mathrm{out}},2δ_{\mathrm{in}}\})+\varepsilon}\right)$$ operations, where $β_R(δ_{\mathrm{in}},δ)$ denotes the maximum of $δ_{\mathrm{in}}$ and $ω_{δ_{\mathrm{in}},R}(a,1,b)$ over all $a,b\in[0,1]$ satisfying $a+b=δ$. For dense inputs over a commutative ring, this bound simplifies to $O(n^{ω_R((δ_{\mathrm{out}}-1)_+,1,1)+\varepsilon})$. With the current rectangular matrix multiplication bounds, this is nearly quadratic, namely $O(n^{2+\varepsilon})$, for every $δ_\mathrm{out}\le1.321334$, improving the previous deterministic range of $δ_{\mathrm{out}}\le 0.642668$. To prove this result, we develop a general deterministic recovery technique that finds and fixes sparse parts of an unknown matrix while keeping temporary errors in denser parts under control.

Abstract: arXiv descriptive metadata (CC0). Thank you to arXiv for use of its open access interoperability. This service was not reviewed or approved by, nor does it necessarily express or reflect the policies or opinions of, arXiv.