On the Generic Low-Rank Matrix Completion

This paper investigates the low-rank matrix completion (LRMC) problem from a generic vantage point. Unlike most existing work that has focused on recovering a low-rank matrix from a subset of the entries with specified values, the only information available here is just the pattern (i.e., positions) of observed entries. To be precise, given is an $n\times m$ pattern matrix ($n\le m$) whose entries take fixed zero, unknown generic values, and missing values that are free to be chosen from the complex field, the question of interest is whether there is a matrix completion with rank no more than $n-k$ ($k\ge 1$) for almost all values of the unknown generic entries, which is called the generic low-rank matrix completion (GLRMC) associated with $({\cal M},k)$, and abbreviated as GLRMC$({\cal M},k)$. Leveraging an elementary tool from the theory on systems of polynomial equations, we give a simple proof for genericity of this problem, which was first observed by $Kir\'aly et al.$, that is, depending on the pattern of the observed entries, the answer is either yes or no, for almost all realizations of the unknown generic entries. We provide necessary and sufficient conditions ensuring feasibility of GLRMC$({\cal M},1)$. Aso, we provide a sufficient condition and a necessary condition (which we conjecture to be sufficient too) for the general case (i.e., $k\ge 1$). In the following, two randomized algorithms are presented to determine upper and lower bounds for the generic minimum completion rank. Our approaches are based on the algebraic geometry theory, and a novel basis preservation principle discovered herein. Finally, numerical simulations are given to corroborate the theoretical findings and the effectiveness of the proposed algorithms.