We present two algorithms for large-scale low-rank Euclidean distance matrix completion problems, based on semidefinite optimization. Our first method works by relating cliques in the graph of the known distances to faces of the positive semidefinite cone, yielding a combinatorial procedure that is provably robust and parallelizable. Our second algorithm is a first order method for maximizing the trace-a popular low-rank inducing regularizer-in the formulation of the problem with a constrained misfit. Both of the methods output a point configuration that can serve as a high-quality initialization for local optimization techniques. Numerical experiments on large-scale sensor localization problems illustrate the two approaches.
AMS subject classifications. 90C22, 90C25 , 52A991 Introduction. A pervasive task in distance geometry is the inverse problem: given only local pairwise Euclidean distances among a set of points, recover their locations in space. More precisely, given a weighted undirected graph G = (V, E, d) on a vertex set {1, . . . , n} and an integer r, find (if possible) a set of points x 1 , . . . , x n in R r satisfyingwhere · denotes the usual Euclidean norm on R r . In most applications, the given squared distances d ij are inexact, and one then seeks points x 1 , . . . , x n satisfying the distance constraints only approximately. This problem appears under numerous names in the literature, such as Euclidean Distance Matrix (EDM) completion and graph realization [2,12,26], and is broadly applicable for example in wireless networks, statistics, robotics, protein reconstruction, and dimensionality reduction in data analysis; the recent survey [27] has an extensive list of relevant references. Fixing notation, we will refer to this problem as EDM completion, throughout. The EDM completion problem can be modeled as the nonconvex feasibility problem: find a symmetric n × n matrix X satisfying (1.1)for all ij ∈ E, Xe = 0, rank X ≤ r, X 0,where e stands for the vector of all ones. Indeed, if X = P P T is a maximal rank factorization of such a matrix X, then the rows of P yield a solution to the EDM