Interlacing Polynomial Method for Matrix Approximation via Generalized Column and Row Selection
Jian-Feng Cai,Zhiqiang Xu,Zili Xu
2023-12-04
Abstract:This paper delves into the spectral norm aspect of the Generalized Column and Row Subset Selection (GCRSS) problem. Given a target matrix $\mathbf{A}$, the objective of GCRSS is to select a column submatrix $\mathbf{B}_{:,S}$ from the source matrix $\mathbf{B}$ and a row submatrix $\mathbf{C}_{R,:}$ from the source maitrx $\mathbf{C}$, with the aim of minimizing the spectral norm of the residual matrix $(\mathbf{I}_n-\mathbf{B}_{:,S}\mathbf{B}_{:,S}^{\dagger})\mathbf{A}(\mathbf{I}_d-\mathbf{C}_{R,:}^{\dagger} \mathbf{C}_{R,:})$. By employing the interlacing polynomials method, we show that the largest root of the expected characteristic polynomial of the residual matrix serves as an upper bound on the smallest spectral norm of the residual matrix. We estimate this root for two specific GCRSS scenarios, one where $r=0$, simplifying the problem to the Generalized Column Subset Selection (GCSS) problem, and the other where $\mathbf{B}=\mathbf{C}=\mathbf{I}_d$, reducing the problem to the submatrix selection problem. In the GCSS scenario, we connect the expected characteristic polynomials to the convolution of multi-affine polynomials, leading to the derivation of the first provable reconstruction bound on the spectral norm of the residual matrix for the GCSS problem. In the submatrix selection scenario, we show that for any sufficiently small $\varepsilon>0$ and any square matrix $\mathbf{A}\in\mathbb{R}^{d\times d}$, there exist two subsets $S\subset [d]$ and $R\subset [d]$ of sizes $O(d\cdot \varepsilon^2)$ such that $\Vert\mathbf{A}_{S,R}\Vert_2\leq \varepsilon\cdot \Vert\mathbf{A}\Vert_2$. Unlike previous studies that have produced comparable results for very special cases where the matrix is either a zero-diagonal or a positive semidefinite matrix, our results apply universally to any matrix $\mathbf{A}$.
Functional Analysis,Combinatorics,Operator Algebras