A New Framework for Matrix Discrepancy: Partial Coloring Bounds via Mirror Descent
PROCEEDINGS OF THE 54TH ANNUAL ACM SIGACT SYMPOSIUM ON THEORY OF COMPUTING (STOC '22)(2022)
Abstract
Motivated by the Matrix Spencer conjecture, we study the problem of finding signed sums of matrices with a small matrix norm. A well-known strategy to obtain these signs is to prove, given matrices A(1),...,A(n) is an element of R-mxm, a Gaussian measure lower bound of 2(-O (n)) for a scaling of the discrepancy body {x is an element of R-n : parallel to Sigma(n)(i=1) x(i)A(i) parallel to <= 1}. We show this is equivalent to covering its polar with 2(O (n)) translates of the cube 1/n B-infinity(n), and construct such a cover via mirror descent. As applications of our framework, we show: Matrix Spencer for Low-Rank Matrices. If the matrices satisfy parallel to A(i)parallel to(op) <= 1 and rank (A(i)) <= r, we can efficiently find a coloring x is an element of{+/- 1}(n) with discrepancy parallel to Sigma(n)(i=1) x(i)Lambda(i)parallel to(op) less than or similar to root nlog(min(rm/n,r)). This improves upon the naive O (root n log r) bound for random coloring and proves the matrix Spencer conjecture when rm <= n. Matrix Spencer for Block Diagonal Matrices. For block diagonal matrices with parallel to A(i)parallel to(op) <= 1 and block size h, we can efficiently find a coloring x is an element of{+/- 1}(n) with. parallel to Sigma(n)(i=1) x(i)A(i) parallel to(op) less than or similar to root nlog(hm/n). This bound was previously shown in [Levy, Ramadas and Rothvoss, IPCO 2017] under the assumption h <= root n, which we remove. Using our proof, we reduce the matrix Spencer conjecture to the existence of a O ( log(m/n)) quantum relative entropy net on the spectraplex. Matrix Discrepancy for Schatten Norms. We generalize our discrepancy bound for matrix Spencer to Schatten norms 2 <= p <= q. Given parallel to A(i)parallel to(Sp) less than or similar to root nmin(p, log(rk)) center dot k(1/p-1/q), = 1 where k := min(1, m/n). Our partial coloring bound is tight when m = Theta(root n). We also provide tight lower bounds of Omega(root n) for rank-1 matrix Spencer when m = n, and Omega(root min(m, n)) for S-2 -> S-infinity discrepancy, precluding a matrix version of the Komlos conjecture.
MoreTranslated text
Key words
matrix,discrepancy,mirror descent,Gaussian measure,covering number,operator norm
AI Read Science
Must-Reading Tree
Example
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined