arXiv subjects
Publications and source records attributed to Henry Wolkowicz.
We consider partial symmetric Toeplitz matrices where a positive definite completion exists. We characterize those patterns where the maximum determinant completion is itself Toeplitz. We then extend these results with positive definite replaced by positive semidefinite, and maximum determinant replaced by maximum rank. These results are used to determine the singularity degree of a family of semidefinite optimization problems.
A spectrahedron is the feasible set of a semidefinite program, SDP, i.e., the intersection of an affine set with the positive semidefinite cone. While strict feasibility is a generic property for random problems, there are many classes of problems where strict feasibility fails and this means that strong duality can fail as well. If the minimal face containing the spectrahedron is known, the SDPcan easily be transformed into an equivalent problem where strict feasibility holds and thus strong duality follows as well. The minimal face is fully characterized by the range or nullspace of any of the matrices in its relative interior. Obtaining such a matrix may require many facial reduction steps and is currently not known to be a tractable problem for spectrahedra with singularity degree greater than one. We propose a single parametric optimization problem with a resulting type of central path and prove that the optimal solution is unique and in the relative interior of the spectrahedron. Numerical tests illustrate the efficacy of our approach and its usefulness in regularizing SDPs.
Slater's condition -- existence of a "strictly feasible solution" -- is a common assumption in conic optimization. Without strict feasibility, first-order optimality conditions may be meaningless, the dual problem may yield little information about the primal, and small changes in the data may render the problem infeasible. Hence, failure of strict feasibility can negatively impact off-the-shelf numerical methods, such as primal-dual interior point methods, in particular. New optimization modelling techniques and convex relaxations for hard nonconvex problems have shown that the loss of strict feasibility is a more pronounced phenomenon than has previously been realized. In this text, we describe various reasons for the loss of strict feasibility, whether due to poor modelling choices or (more interestingly) rich underlying structure, and discuss ways to cope with it and, in many pronounced cases, how to use it as an advantage. In large part, we emphasize the facial reduction preprocessing technique due to its mathematical elegance, geometric transparency, and computational potential.
Recent breakthroughs have been made in the use of semidefinite programming and its application to real polynomial solving. For example, the real radical of a zero dimensional ideal, can be determined by such approaches as shown by Lasserre and collaborators. Some progress has been made on the determination of the real radical in positive dimension by Ma, Wang and Zhi. Such work involves the determination of maximal rank semidefinite moment matrices. Existing methods are computationally expensive and have poorer accuracy on larger examples. This paper is motivated by problems in the numerical computation of the real radical ideal in the general positive case. In this paper we give a method to compute the generators of the real radical for any given degree $d$. We combine the use of moment matrices and techniques from SDP optimization: facial reduction first developed by Borwein and Wolkowicz. In use of the semidefinite moment matrices to compute the real radical, the maximum rank property is very key, and with facial reduction, it can be guaranteed with very high accuracy. Our algorithm can be used to test the real radical membership of a given polynomial. In a special situation, we can determine the real radical ideal in the positive dimensional case.
We study the closure of the projection of the (nonconvex) cone of rank restricted positive semidefinite matrices onto subsets of the matrix entries. This defines the feasible sets for semidefinite completion problems with restrictions on the ranks. Applications include conditions for low-rank completions using the nuclear norm heuristic.
Minimization of the nuclear norm is often used as a surrogate, convex relaxation, for finding the minimum rank completion (recovery) of a partial matrix. The minimum nuclear norm problem can be solved as a trace minimization semidefinite programming problem, (SDP). The SDP and its dual are regular in the sense that they both satisfy strict feasibility. Interior point algorithms are the current methods of choice for these problems. This means that it is difficult to solve large scale problems and difficult to get high accuracy solutions. In this paper we take advantage of the structure at optimality for the minimum nuclear norm problem. We show that even though strict feasibility holds, the facial reduction framework can be successfully applied to obtain a proper face that contains the optimal set, and thus can dramatically reduce the size of the final nuclear norm problem while guaranteeing a low-rank solution. We include numerical tests for both exact and noisy cases. In all cases we assume that knowledge of a target rank is available.
We study large scale extended trust region subproblems (eTRS) i.e., the minimization of a general quadratic function subject to a norm constraint, known as the trust region subproblem (TRS) but with an additional linear inequality constraint. It is well known that strong duality holds for the TRS and that there are efficient algorithms for solving large scale TRS problems. It is also known that there can exist at most one local non-global minimizer (LNGM) for TRS. We combine this with known characterizations for strong duality for eTRS and, in particular, connect this with the so-called hard case for TRS. We begin with a recent characterization of the minimum for the TRS via a generalized eigenvalue problem and extend this result to the LNGM. We then use this to derive an efficient algorithm that finds the global minimum for eTRS by solving at most three generalized eigenvalue problems.
The semidefinite programming (SDP) relaxation has proven to be extremely strong for many hard discrete optimization problems. This is in particular true for the quadratic assignment problem (QAP), arguably one of the hardest NP-hard discrete optimization problems. There are several difficulties that arise in efficiently solving the SDP relaxation, e.g.,~increased dimension; inefficiency of the current primal-dual interior point solvers in terms of both time and accuracy; and difficulty and high expense in adding cutting plane constraints. We propose using the alternating direction method of multipliers (ADMM) to solve the SDP relaxation. This first order approach allows for inexpensive iterations, a method of cheaply obtaining low rank solutions, as well a trivial way of adding cutting plane inequalities. When compared to current approaches and current best available bounds we obtain remarkable robustness, efficiency and improved bounds.
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.
The real radical ideal of a system of polynomials with finitely many complex roots is generated by a system of real polynomials having only real roots and free of multiplicities. It is a central object in computational real algebraic geometry and important as a preconditioner for numerical solvers. Lasserre and co-workers have shown that the real radical ideal of real polynomial systems with finitely many real solutions can be determined by a combination of semi-definite programming (SDP) and geometric involution techniques. A conjectured extension of such methods to positive dimensional polynomial systems has been given recently by Ma, Wang and Zhi. We show that regularity in the form of the Slater constraint qualification (strict feasibility) fails for the resulting SDP feasibility problems. Facial reduction is then a popular technique whereby SDP problems that fail strict feasibility can be regularized by projecting onto a face of the convex cone of semi-definite problems. In this paper we introduce a framework for combining facial reduction with such SDP methods for analyzing $0$ and positive dimensional real ideals of real polynomial systems. The SDP methods are implemented in MATLAB and our geometric involutive form is implemented in Maple. We use two approaches to find a feasible moment matrix. We use an interior point method within the CVX package for MATLAB and also the Douglas-Rachford (DR) projection-reflection method. Illustrative examples show the advantages of the DR approach for some problems over standard interior point methods. We also see the advantage of facial reduction both in regularizing the problem and also in reducing the dimension of the moment matrices. Problems requiring more than one facial reduction are also presented.
We consider the problem of partitioning the node set of a graph into $k$ sets of given sizes in order to \emph{minimize the cut} obtained using (removing) the $k$-th set. If the resulting cut has value $0$, then we have obtained a vertex separator. This problem is closely related to the graph partitioning problem. In fact, the model we use is the same as that for the graph partitioning problem except for a different \emph{quadratic} objective function. We look at known and new bounds obtained from various relaxations for this NP-hard problem. This includes: the standard eigenvalue bound, projected eigenvalue bounds using both the adjacency matrix and the Laplacian, quadratic programming (QP) bounds based on recent successful QP bounds for the quadratic assignment problems, and semidefinite programming bounds. We include numerical tests for large and \emph{huge} problems that illustrate the efficiency of the bounds in terms of strength and time.
We consider the problem of constructing quantum operations or channels, if they exist, that transform a given set of quantum states $\{ρ_1, \dots, ρ_k\}$ to another such set $\{\hatρ_1, \dots, \hatρ_k\}$. In other words, we must find a {\em completely positive linear map}, if it exists, that maps a given set of density matrices to another given set of density matrices. This problem, in turn, is an instance of a positive semi-definite feasibility problem, but with highly structured constraints. The nature of the constraints makes projection based algorithms very appealing when the number of variables is huge and standard interior point-methods for semi-definite programming are not applicable. We provide emperical evidence to this effect. We moreover present heuristics for finding both high rank and low rank solutions. Our experiments are based on the \emph{method of alternating projections} and the \emph{Douglas-Rachford} reflection method.
The sensor network localization, SNL, problem in embedding dimension r, consists of locating the positions of wireless sensors, given only the distances between sensors that are within radio range and the positions of a subset of the sensors (called anchors). Current solution techniques relax this problem to a weighted, nearest, (positive) semidefinite programming, SDP, completion problem, by using the linear mapping between Euclidean distance matrices, EDM, and semidefinite matrices. The resulting SDP is solved using primal-dual interior point solvers, yielding an expensive and inexact solution. This relaxation is highly degenerate in the sense that the feasible set is restricted to a low dimensional face of the SDP cone, implying that the Slater constraint qualification fails. Cliques in the graph of the SNL problem give rise to this degeneracy in the SDP relaxation. In this paper, we take advantage of the absence of the Slater constraint qualification and derive a technique for the SNL problem, with exact data, that explicitly solves the corresponding rank restricted SDP problem. No SDP solvers are used. For randomly generated instances, we are able to efficiently solve many huge instances of this NP-hard problem to high accuracy, by finding a representation of the minimal face of the SDP cone that contains the SDP matrix representation of the EDM. The main work of our algorithm consists in repeatedly finding the intersection of subspaces that represent the faces of the SDP cone that correspond to cliques of the SNL problem.
We study Semidefinite Programming, \SDPc relaxations for Sensor Network Localization, \SNLc with anchors and with noisy distance information. The main point of the paper is to view \SNL as a (nearest) Euclidean Distance Matrix, \EDM, completion problem and to show the advantages for using this latter, well studied model. We first show that the current popular \SDP relaxation is equivalent to known relaxations in the literature for \EDM completions. The existence of anchors in the problem is {\em not} special. The set of anchors simply corresponds to a given fixed clique for the graph of the \EDM problem. We next propose a method of projection when a large clique or a dense subgraph is identified in the underlying graph. This projection reduces the size, and improves the stability, of the relaxation. In addition, viewing the problem as an \EDM completion problem yields better low rank approximations for the low dimensional realizations. And, the projection/reduction procedure can be repeated for other given cliques of sensors or for sets of sensors, where many distances are known. Thus, further size reduction can be obtained. Optimality/duality conditions and a primal-dual interior-exterior path following algorithm are derived for the \SDP relaxations We discuss the relative stability and strength of two formulations and the corresponding algorithms that are used. In particular, we show that the quadratic formulation arising from the \SDP relaxation is better conditioned than the linearized form, that is used in the literature and that arises from applying a Schur complement.