We consider the problems of completing a low-rank positive semidefinite square matrix M or a low-rank rectangular matrix N from a given subset of their entries. We study the local and global uniqueness of such completions by analyzing the structure of the graphs determined by the positions of the known entries of M or N. We show that, in the generic setting, the unique completability testing of rectangular matrices is a special case of the unique completability testing of positive semidefinite matrices. We prove that a generic partially filled semidefinite n×n matrix is globally uniquely rank d completable if any principal minor of size n−1 is locally uniquely rank d completable. These results are based on new geometric observations that extend similar results of the theory of rigid frameworks. We also give an example showing that global completability is not a generic property in R2. We provide sufficient conditions for local and global unique completability of a partially filled matrix in terms of either the minimum number of known entries per row or the total number of known entries.
- Positive semidefinite matrices
- Rigidity of frameworks
- Unique low rank matrix completion
ASJC Scopus subject areas
- Theoretical Computer Science
- Discrete Mathematics and Combinatorics
- Computational Theory and Mathematics