Tim Hoheisel

Theoretical and numerical comparison of relaxation methods for mathematical programs with complementarity constraints

Mathematical programs with equilibrium constraints (MPECs) are difficult optimization problems whose feasible sets do not satisfy most of the standard constraint qualifications. Hence MPECs cause difficulties both from a theoretical and a numerical point of view. As a consequence, a number of MPEC-tailored solution methods have been suggested during the last decade which are known to converge under suitable assumptions. Among these MPEC-tailored solution schemes, the relaxation methods are certainly one of the most prominent class of solution methods. Several different relaxation schemes are available in the meantime, and the aim of this paper is to provide a theoretical and numerical comparison of these schemes. More precisely, in the theoretical part, we improve the convergence theorems of several existing relaxation methods. There, we also take a closer look at the properties of the feasible sets of the relaxed problems and show which standard constraint qualifications are satisfied for these relaxed problems. Finally, the numerical comparison is based on the MacMPEC test problem collection.

On the Abadie and Guignard constraint qualifications for Mathematical Programmes with Vanishing Constraints

We consider a special class of optimization problems that we call a Mathematical Programme with Vanishing Constraints. It has a number of important applications in structural and topology optimization, but typically does not satisfy standard constraint qualifications like the linear independence and the Mangasarian–Fromovitz constraint qualification. We therefore investigate the Abadie and Guignard constraint qualifications in more detail. In particular, it follows from our results that also the Abadie constraint qualification is typically not satisfied, whereas the Guignard constraint qualification holds under fairly mild assumptions for our particular class of optimization problems.

Convergence of a local regularization approach for mathematical programmes with complementarity or vanishing constraints

Mathematical programmes with equilibrium or vanishing constraints (MPECs or MPVCs) are both known to be difficult optimization problems which typically violate all standard constraint qualifications. A number of methods try to exploit the particular structure of MPECs and MPVCs in order to overcome these difficulties. In a recent paper by Steffensen and Ulbrich (S. Steffensen and M. Ulbrich, A new regularization scheme for mathematical programs with equilibrium constraints, SIAM J. Optim. 2010.), this was done for MPECs by a local regularization idea that may be viewed as a modification of the popular global regularization technique by Scholtes (S. Scholtes, Convergence properties of a regularization scheme for mathematical programs with complementarity constraints, SIAM J. Optim. 11 (2001), pp. 918–936.). The aim of this paper is twofold. First, we improve the convergence theory from (S. Steffensen and M. Ulbrich, A new regularization scheme for mathematical programs with equilibrium constraints, SIAM J. Optim. 2010.) in the MPEC setting, and second we translate this local regularization idea to MPVCs and obtain a new solution method for this class of optimization problems for which several convergence results are given.

A smoothing-regularization approach to mathematical programs with vanishing constraints

We consider a numerical approach for the solution of a difficult class of optimization problems called mathematical programs with vanishing constraints. The basic idea is to reformulate the characteristic constraints of the program via a nonsmooth function and to eventually smooth it and regularize the feasible set with the aid of a certain smoothing- and regularization parameter t>0 such that the resulting problem is more tractable and coincides with the original program for t=0. We investigate the convergence behavior of a sequence of stationary points of the smooth and regularized problems by letting t tend to zero. Numerical results illustrating the performance of the approach are given. In particular, a large-scale example from topology optimization of mechanical structures with local stress constraints is investigated.

Epi-convergent Smoothing with Applications to Convex Composite Functions

Smoothing methods have become part of the standard tool set for the study and solution of nondifferentiable and constrained optimization problems as well as a range of other variational and equilibrium problems. In this note we synthesize and extend recent results due to Beck and Teboulle on infimal convolution smoothing for convex functions with those of X. Chen on gradient consistency for nonconvex functions. We use epi-convergence techniques to define a notion of epi-smoothing that allows us to tap into the rich variational structure of the subdifferential calculus for nonsmooth, nonconvex, and nonfinite-valued functions. As an illustration of the versatility and range of epi-smoothing techniques, the results are applied to the general constrained optimization for which nonlinear programming is a special case.

On a Smooth Dual Gap Function for a Class of Quasi-Variational Inequalities

A well-known technique for the solution of quasi-variational inequalities (QVIs) consists in the reformulation of this problem as a constrained or unconstrained optimization problem by means of so-called gap functions. In contrast to standard variational inequalities, however, these gap functions turn out to be nonsmooth in general. Here, it is shown that one can obtain an unconstrained optimization reformulation of a class of QVIs with affine operator by using a continuously differentiable dual gap function. This extends an idea from Dietrich (J. Math. Anal. Appl. 235:380–393 [24]). Some numerical results illustrate the practical behavior of this dual gap function approach.

MATRIX SUPPORT FUNCTIONALS FOR INVERSE PROBLEMS, REGULARIZATION, AND LEARNING ∗

A new class of matrix support functionals is presented which establish a connection between optimal value functions for quadratic optimization problems, the matrix-fractional function, the pseudo-matrix-fractional function, the nuclear norm, and multitask learning. The support function is based on the graph of the product of a matrix with its transpose. Closed form expressions for the support functional and its subdifferential are derived. In particular, the support functional is shown to be continuously differentiable on the interior of its domain, and a formula for the derivative is given when it exists.

On a Smooth Dual Gap Function for a Class of Player Convex Generalized Nash Equilibrium Problems

We consider a class of generalized Nash equilibrium problems, where both objective functions and constraints are allowed to depend on the decision variables of the other players. It is well known that this problem can be reformulated as a constrained optimization problem via the (regularized) Nikaido–Isoda-function, but this reformulation is usually nonsmooth. Here we observe that, under suitable conditions, this reformulation turns out to be the difference of two convex functions. This allows the application of the Toland-Singer duality theory in order to obtain a dual formulation, which provides an unconstrained and continuously differentiable optimization reformulation of the generalized Nash equilibrium problem. Moreover, based on a result from parametric optimization, the gradient of the unconstrained objective function is shown to be piecewise smooth. Some numerical results are presented to illustrate the theory.

Mathematical programs with vanishing constraints: a new regularization approach with strong convergence properties

Motivated by a recent method introduced by Kanzow and Schwartz [C. Kanzow and A. Schwartz, A new regularization method for mathematical programs with complementarity constraints with strong convergence properties, Preprint 296, Institute of Mathematics, University of Würzburg, Würzburg, 2010] for mathematical programs with complementarity constraints (MPCCs), we present a related regularization scheme for the solution of mathematical programs with vanishing constraints (MPVCs). This new regularization method has stronger convergence properties than the existing ones. In particular, it is shown that every limit point is at least M-stationary under a linear independence-type constraint qualification. If, in addition, an asymptotic weak nondegeneracy assumption holds, the limit point is shown to be S-stationary. Second-order conditions are not needed to obtain these results. Furthermore, some results are given which state that the regularized subproblems satisfy suitable standard constraint qualifications such that the existing software can be applied to these regularized problems.

Convex Geometry of the Generalized Matrix-Fractional Function

Generalized matrix-fractional (GMF) functions are a class of matrix support functions introduced by Burke and Hoheisel as a tool for unifying a range of seemingly divergent matrix optimization prob...