Stochastic Systems Group  

James Saunderson
SSG
Suppose an n x n matrix X is the sum of a diagonal and a positive semifedinite lowrank matrix. Decomposing X into these unknown constituents has applications in statistics, signal processing, and elsewhere. We give a simple condition on the column space of the low rank matrix that ensures a tractible convex program can correctly decompose X. Our analysis highlights connections between this problem and the structure of the set of correlation matrices.
Problems with this site should be emailed to jonesb@mit.edu