Search

Several manifold optimization schemes are presented and analyzed for solving a specialized inverse structured symmetric matrix problem with prescribed spectrum. Some entries in the desired matrix are assigned in advance and cannot be altered. The rest of the entries are free, some of them preferably away from zero. The reconstructed matrix must satisfy these requirements and its eigenvalues must be the given ones. This inverse eigenvalue problem is related to the problem of determining the graph, with weights on the undirected edges, of the matrix associated with its sparse pattern. Our optimization schemes are based on considering the eigenvector matrix as the only unknown and iteratively moving on the manifold of orthogonal matrices, forcing the additional structural requirements through a change of variables and a convenient differentiable objective function in the space of square matrices. We propose Riemannian gradient-type methods combined with two different well-known retractions, and with two well-known constrained optimization strategies: penalization and augmented Lagrangian. We also present a block alternating technique that takes advantage of a proper separation of variables. Convergence properties of the penalty alternating approach are established. Finally, we present initial numerical results to demonstrate the effectiveness of our proposals. [ABSTRACT FROM AUTHOR]

In this two-part study, we develop a general theory of the so-called exact augmented Lagrangians for constrained optimization problems in Hilbert spaces. In contrast to traditional nonsmooth exact penalty functions, these augmented Lagrangians are continuously differentiable for smooth problems and do not suffer from the Maratos effect, which makes them especially appealing for applications in numerical optimization. Our aim is to present a detailed study of various theoretical properties of exact augmented Lagrangians and discuss several applications of these functions to constrained variational problems, problems with PDE constraints, and optimal control problems. The first paper is devoted to a theoretical analysis of an exact augmented Lagrangian for optimization problems in Hilbert spaces. We obtain several useful estimates of this augmented Lagrangian and its gradient, and present several types of sufficient conditions for KKT-points of a constrained problem corresponding to locally/globally optimal solutions to be local/global minimizers of the exact augmented Lagrangian. [ABSTRACT FROM AUTHOR]

This paper provides a theoretical and numerical investigation of a penalty decomposition scheme for the solution of optimization problems with geometric constraints. In particular, we consider some situations where parts of the constraints are nonconvex and complicated, like cardinality constraints, disjunctive programs, or matrix problems involving rank constraints. By a variable duplication and decomposition strategy, the method presented here explicitly handles these difficult constraints, thus generating iterates which are feasible with respect to them, while the remaining (standard and supposingly simple) constraints are tackled by sequential penalization. Inexact optimization steps are proven sufficient for the resulting algorithm to work, so that it is employable even with difficult objective functions. The current work is therefore a significant generalization of existing papers on penalty decomposition methods. On the other hand, it is related to some recent publications which use an augmented Lagrangian idea to solve optimization problems with geometric constraints. Compared to these methods, the decomposition idea is shown to be numerically superior since it allows much more freedom in the choice of the subproblem solver, and since the number of certain (possibly expensive) projection steps is significantly less. Extensive numerical results on several highly complicated classes of optimization problems in vector and matrix spaces indicate that the current method is indeed very efficient to solve these problems. [ABSTRACT FROM AUTHOR]

We consider solving nonlinear optimization problems with a stochastic objective and deterministic equality constraints. We assume for the objective that its evaluation, gradient, and Hessian are inaccessible, while one can compute their stochastic estimates by, for example, subsampling. We propose a stochastic algorithm based on sequential quadratic programming (SQP) that uses a differentiable exact augmented Lagrangian as the merit function. To motivate our algorithm design, we first revisit and simplify an old SQP method Lucidi (J. Optim. Theory Appl. 67(2): 227–245, 1990) developed for solving deterministic problems, which serves as the skeleton of our stochastic algorithm. Based on the simplified deterministic algorithm, we then propose a non-adaptive SQP for dealing with stochastic objective, where the gradient and Hessian are replaced by stochastic estimates but the stepsizes are deterministic and prespecified. Finally, we incorporate a recent stochastic line search procedure Paquette and Scheinberg (SIAM J. Optim. 30(1): 349–376 2020) into the non-adaptive stochastic SQP to adaptively select the random stepsizes, which leads to an adaptive stochastic SQP. The global "almost sure" convergence for both non-adaptive and adaptive SQP methods is established. Numerical experiments on nonlinear problems in CUTEst test set demonstrate the superiority of the adaptive algorithm. [ABSTRACT FROM AUTHOR]

We introduce a perturbed augmented Lagrangian method framework, which is a convenient tool for local analyses of convergence and rates of convergence of some modifications of the classical augmented Lagrangian algorithm. One example to which our development applies is the proximal augmented Lagrangian method. Previous results for this version required twice differentiability of the problem data, the linear independence constraint qualification, strict complementarity, and second-order sufficiency; or the linear independence constraint qualification and strong second-order sufficiency. We obtain a set of convergence properties under significantly weaker assumptions: once (not twice) differentiability of the problem data, uniqueness of the Lagrange multiplier, and second-order sufficiency (no linear independence constraint qualification and no strict complementarity); or even second-order sufficiency only. Another version to which the general framework applies is the smoothed augmented Lagrangian method, where the plus-function associated with penalization of inequality constraints is approximated by a family of smooth functions (so that the subproblems are twice differentiable if the problem data are). Furthermore, for all the modifications, inexact solution of subproblems is handled naturally. The presented framework also subsumes the basic augmented Lagrangian method, both exact and inexact. [ABSTRACT FROM AUTHOR]

It is well-known that the primal quadratic growth condition of the classical augmented Lagrangian around a local minimizer can be obtained under the second-order sufficient optimality condition. In this paper, we show that those conditions are indeed equivalent. Moreover, we prove that the primal quadratic growth condition of the sharp augmented Lagrangian around a local minimizer is in fact equivalent to the weak second-order sufficient optimality condition. In addition, we present some secondary results involving the sharp augmented Lagrangian. [ABSTRACT FROM AUTHOR]

The alternating direction method of multipliers (ADMM) is a classical effective method for solving two-block convex optimization subject to linear constraints. However, its convergence may not be guaranteed for multiple-block case without additional assumptions. One remedy might be the block-wise ADMM (BADMM), in which the variables are regrouped into two groups firstly and then the augmented Lagrangian function is minimized w.r.t. each block variable by the following scheme: using a Gauss–Seidel fashion to update the variables between each group, while using a Jacobi fashion to update the variables within each group. In order to derive its convergence property, a special proximal term is added to each subproblem. In this paper, we propose a new partial PPA block-wise ADMM where we only need to add proximal terms to the subproblems in the first group. At the end of each iteration, an extension step on all variables is performed with a fixed step size. As the subproblems in the second group are unmodified, the resulting sequence might yield better quality as well as potentially faster convergence speed. Preliminary experimental results show that the new algorithm is empirically effective on solving both synthetic and real problems when it is compared with several very efficient ADMM-based algorithms. [ABSTRACT FROM AUTHOR]

We introduce a filter mechanism to enforce convergence for augmented Lagrangian methods for nonlinear programming. In contrast to traditional augmented Lagrangian methods, our approach does not require the use of forcing sequences that drive the first-order error to zero. Instead, we employ a filter to drive the optimality measures to zero. Our algorithm is flexible in the sense that it allows for equality-constrained quadratic programming steps to accelerate local convergence. We also include a feasibility restoration phase that allows fast detection of infeasible problems. We provide a convergence proof that shows that our algorithm converges to first-order stationary points. We provide preliminary numerical results that demonstrate the effectiveness of our proposed method. [ABSTRACT FROM AUTHOR]

At each iteration of the augmented Lagrangian algorithm, a nonlinear subproblem is being solved. The number of inner iterations (of some/any method) needed to obtain a solution of the subproblem, or even a suitable approximate stationary point, is in principle unknown. In this paper we show that to compute an approximate stationary point sufficient to guarantee local superlinear convergence of the augmented Lagrangian iterations, it is enough to solve two quadratic programming problems (or two linear systems in the equality-constrained case). In other words, two inner Newtonian iterations are sufficient. To the best of our knowledge, such results are not available even under the strongest assumptions (of second-order sufficiency, strict complementarity, and the linear independence constraint qualification). Our analysis is performed under second-order sufficiency only, which is the weakest assumption for obtaining local convergence and rate of convergence of outer iterations of the augmented Lagrangian algorithm. The structure of the quadratic problems in question is related to the stabilized sequential quadratic programming and to second-order corrections. [ABSTRACT FROM AUTHOR]

Interior point methods have attracted most of the attention in the recent decades for solving large scale convex quadratic programming problems. In this paper we take a different route as we present an augmented Lagrangian method for convex quadratic programming based on recent developments for nonlinear programming. In our approach, box constraints are penalized while equality constraints are kept within the subproblems. The motivation for this approach is that Newton's method can be efficient for minimizing a piecewise quadratic function. Moreover, since augmented Lagrangian methods do not rely on proximity to the central path, some of the inherent difficulties in interior point methods can be avoided. In addition, a good starting point can be easily exploited, which can be relevant for solving subproblems arising from sequential quadratic programming, in sensitivity analysis and in branch and bound techniques. We prove well-definedness and finite convergence of the method proposed. Numerical experiments on separable strictly convex quadratic problems formulated from the Netlib collection show that our method can be competitive with interior point methods, in particular when a good initial point is available and a second-order Lagrange multiplier update is used. [ABSTRACT FROM AUTHOR]

We consider the problem of minimizing a smooth nonconvex function over a structured convex feasible set, that is, defined by two sets of constraints that are easy to treat when considered separately. In order to exploit the structure of the problem, we define an equivalent formulation by duplicating the variables and we consider the augmented Lagrangian of this latter formulation. Following the idea of the Alternating Direction Method of Multipliers (ADMM), we propose an algorithm where a two-blocks decomposition method is embedded within an augmented Lagrangian framework. The peculiarities of the proposed algorithm are the following: (1) the computation of the exact solution of a possibly nonconvex subproblem is not required; (2) the penalty parameter is iteratively updated once an approximated stationary point of the augmented Lagrangian is determined. Global convergence results are stated under mild assumptions and without requiring convexity of the objective function. Although the primary aim of the paper is theoretical, we perform numerical experiments on a nonconvex problem arising in machine learning, and the obtained results show the practical advantages of the proposed approach with respect to classical ADMM. [ABSTRACT FROM AUTHOR]

In this work, we present an Augmented Lagrangian algorithm for nonlinear semidefinite problems (NLSDPs), which is a natural extension of its consolidated counterpart in nonlinear programming. This method works with two levels of constraints; one that is penalized and other that is kept within the subproblems. This is done to allow exploiting the subproblem structure while solving it. The global convergence theory is based on recent results regarding approximate Karush–Kuhn–Tucker optimality conditions for NLSDPs, which are stronger than the usually employed Fritz John optimality conditions. Additionally, we approach the problem of covering a given object with a fixed number of balls with a minimum radius, where we employ some convex algebraic geometry tools, such as Stengle's Positivstellensatz and its variations, which allows for a much more general model. Preliminary numerical experiments are presented. [ABSTRACT FROM AUTHOR]

Interior-point methods have been shown to be very efficient for large-scale nonlinear programming. The combination with penalty methods increases their robustness due to the regularization of the constraints caused by the penalty term. In this paper a primal-dual penalty-interior-point algorithm is proposed, that is based on an augmented Lagrangian approach with an ℓ2-exact penalty function. Global convergence is maintained by a combination of a merit function and a filter approach. Unlike the majority of filter methods, no separate feasibility restoration phase is required. The algorithm has been implemented within the solver WORHP to study different penalty and line search options and to compare its numerical performance to two other state-of-the-art nonlinear programming algorithms, the interior-point method IPOPT and the sequential quadratic programming method of WORHP. [ABSTRACT FROM AUTHOR]

In this article we present a general theory of augmented Lagrangian functions for cone constrained optimization problems that allows one to study almost all known augmented Lagrangians for these problems within a unified framework. We develop a new general method for proving the existence of global saddle points of augmented Lagrangian functions, called the localization principle. The localization principle unifies, generalizes and sharpens most of the known results on the existence of global saddle points, and, in essence, reduces the problem of the existence of global saddle points to a local analysis of optimality conditions. With the use of the localization principle we obtain first necessary and sufficient conditions for the existence of a global saddle point of an augmented Lagrangian for cone constrained minimax problems via both second and first order optimality conditions. In the second part of the paper, we present a general approach to the construction of globally exact augmented Lagrangian functions. The general approach developed in this paper allowed us not only to sharpen most of the existing results on globally exact augmented Lagrangians, but also to construct first globally exact augmented Lagrangian functions for equality constrained optimization problems, for nonlinear second order cone programs and for nonlinear semidefinite programs. These globally exact augmented Lagrangians can be utilized in order to design new superlinearly (or even quadratically) convergent optimization methods for cone constrained optimization problems. [ABSTRACT FROM AUTHOR]

In this article, we present new general results on existence of augmented Lagrange multipliers. We define a penalty function associated with an augmented Lagrangian, and prove that, under a certain growth assumption on the augmenting function, an augmented Lagrange multiplier exists if and only if this penalty function is exact. We also develop a new general approach to the study of augmented Lagrange multipliers called the localization principle. The localization principle allows one to study the local behaviour of the augmented Lagrangian near global optimal solutions of the initial optimization problem in order to prove the existence of augmented Lagrange multipliers. [ABSTRACT FROM AUTHOR]

We present a primal-dual algorithm for solving a constrained optimization problem. This method is based on a Newtonian method applied to a sequence of perturbed KKT systems. These systems follow from a reformulation of the initial problem under the form of a sequence of penalized problems, by introducing an augmented Lagrangian for handling the equality constraints and a log-barrier penalty for the inequalities. We detail the updating rules for monitoring the different parameters (Lagrange multiplier estimate, quadratic penalty and log-barrier parameter), in order to get strong global convergence properties. We show that one advantage of this approach is that it introduces a natural regularization of the linear system to solve at each iteration, for the solution of a problem with a rank deficient Jacobian of constraints. The numerical experiments show the good practical performances of the proposed method especially for degenerate problems. [ABSTRACT FROM AUTHOR]

Stabilized sequential quadratic programming (sSQP) methods for nonlinear optimization generate a sequence of iterates with fast local convergence regardless of whether or not the active-constraint gradients are linearly dependent. This paper concerns the local convergence analysis of an sSQP method that uses a line search with a primal-dual augmented Lagrangian merit function to enforce global convergence. The method is provably well-defined and is based on solving a strictly convex quadratic programming subproblem at each iteration. It is shown that the method has superlinear local convergence under assumptions that are no stronger than those required by conventional stabilized SQP methods. The fast local convergence is obtained by allowing a small relaxation of the optimality conditions for the quadratic programming subproblem in the neighborhood of a solution. In the limit, the line search selects the unit step length, which implies that the method does not suffer from the Maratos effect. The analysis indicates that the method has the same strong first- and second-order global convergence properties that have been established for augmented Lagrangian methods, yet is able to transition seamlessly to sSQP with fast local convergence in the neighborhood of a solution. Numerical results on some degenerate problems are reported. [ABSTRACT FROM AUTHOR]

The multiplier-penalty approach is one of the classical methods for the solution of constrained optimization problems. This method was generalized to the solution of quasi-variational inequalities by Pang and Fukushima (Comput Manag Sci 2:21-56, 2005). Based on the recent improvements achieved for the multiplier-penalty approach for optimization, we generalize the method by Pang and Fukushima for quasi-variational inequalities in several respects: (a) We allow to compute inexact KKT-points of the resulting subproblems; (b) We improve the existing convergence theory; (c) We investigate some special classes of quasi-variational inequalities where the resulting subproblems turn out to be easy to solve. Some numerical results indicate that the corresponding method works quite reliable in practice. [ABSTRACT FROM AUTHOR]

Augmented Lagrangian function is one of the most important tools used in solving some constrained optimization problems. In this article, we study an augmented Lagrangian objective penalty function and a modified augmented Lagrangian objective penalty function for inequality constrained optimization problems. First, we prove the dual properties of the augmented Lagrangian objective penalty function, which are at least as good as the traditional Lagrangian function's. Under some conditions, the saddle point of the augmented Lagrangian objective penalty function satisfies the first-order Karush-Kuhn-Tucker condition. This is especially so when the Karush-Kuhn-Tucker condition holds for convex programming of its saddle point existence. Second, we prove the dual properties of the modified augmented Lagrangian objective penalty function. For a global optimal solution, when the exactness of the modified augmented Lagrangian objective penalty function holds, its saddle point exists. The sufficient and necessary stability conditions used to determine whether the modified augmented Lagrangian objective penalty function is exact for a global solution is proved. Based on the modified augmented Lagrangian objective penalty function, an algorithm is developed to find a global solution to an inequality constrained optimization problem, and its global convergence is also proved under some conditions. Furthermore, the sufficient and necessary calmness condition on the exactness of the modified augmented Lagrangian objective penalty function is proved for a local solution. An algorithm is presented in finding a local solution, with its convergence proved under some conditions. [ABSTRACT FROM PUBLISHER]

In this paper, we propose an augmented Lagrangian affine scaling (ALAS) algorithm for general nonlinear programming, for which a quadratic approximation to the augmented Lagrangian is minimized at each iteration. Different from the classical sequential quadratic programming (SQP), the linearization of nonlinear constraints is put into the penalty term of this quadratic approximation, which results in smooth objective of the subproblem and avoids possible inconsistency among the linearized constraints and trust region constraint. By applying affine scaling techniques to handle the strict bound constraints, an active set type affine scaling trust region subproblem is proposed. Through special strategies of updating Lagrange multipliers and adjusting penalty parameters, this subproblem is able to reduce the objective function value and feasibility errors in an adaptive well-balanced way. Global convergence of the ALAS algorithm is established under mild assumptions. Furthermore, boundedness of the penalty parameter is analyzed under certain conditions. Preliminary numerical experiments of ALAS are performed for solving a set of general constrained nonlinear optimization problems in CUTEr problem library. [ABSTRACT FROM AUTHOR]

In this paper, we introduce two primal-dual active-set methods for solving large-scale constrained optimization problems. The first method minimizes a sequence of primal-dual augmented Lagrangian functions subject to bounds on the primal variables and artificial bounds on the dual variables. The basic structure is similar to the well-known optimization package Lancelot (Conn, et al. in SIAM J Numer Anal 28:545-572, ), which uses the traditional primal augmented Lagrangian function. Like Lancelot, our algorithm may use gradient projection-based methods enhanced by subspace acceleration techniques to solve each subproblem and therefore may be implemented matrix-free. The artificial bounds on the dual variables are a unique feature of our method and serve as a form of dual regularization. Our second algorithm is a two-phase method. The first phase computes iterates using our primal-dual augmented Lagrangian algorithm, which benefits from using cheap gradient projections and matrix-free linear CG calculations. The final iterate produced during this phase is then used as input for phase two, which is a stabilized sequential quadratic programming method (Gill and Robinson in SIAM J Opt 1-45, ). Obtaining superlinear local convergence under weak assumptions is an important benefit of the transition to a stabilized sequential quadratic programming algorithm. Interestingly, the bound-constrained subproblem used in phase one is equivalent to the stabilized subproblem used in phase two under certain assumptions. This fact makes our choice of algorithms a natural one. [ABSTRACT FROM AUTHOR]

We discuss the question of which features and/or properties make a method for solving a given problem belong to the 'Newtonian class.' Is it the strategy of linearization (or perhaps, second-order approximation) of the problem data (maybe only part of the problem data)? Or is it fast local convergence of the method under natural assumptions and at a reasonable computational cost of its iteration? We consider both points of view, and also how they relate to each other. In particular, we discuss abstract Newtonian frameworks for generalized equations, and how a number of important algorithms for constrained optimization can be related to them by introducing structured perturbations to the basic Newton iteration. This gives useful tools for local convergence and rate-of-convergence analysis of various algorithms from unified perspectives, often yielding sharper results than provided by other approaches. Specific constrained optimization algorithms, which can be conveniently analyzed within perturbed Newtonian frameworks, include the sequential quadratic programming method and its various modifications (truncated, augmented Lagrangian, composite step, stabilized, and equipped with second-order corrections), the linearly constrained Lagrangian methods, inexact restoration, sequential quadratically constrained quadratic programming, and certain interior feasible directions methods. We recall most of those algorithms as examples to illustrate the underlying viewpoint. We also discuss how the main ideas of this approach go beyond clearly Newton-related methods and are applicable, for example, to the augmented Lagrangian algorithm (also known as the method of multipliers), which is in principle not of Newton type since its iterations do not approximate any part of the problem data. [ABSTRACT FROM AUTHOR]

In this work we introduce a relaxed version of the constant positive linear dependence constraint qualification (CPLD) that we call RCPLD. This development is inspired by a recent generalization of the constant rank constraint qualification by Minchenko and Stakhovski that was called RCRCQ. We show that RCPLD is enough to ensure the convergence of an augmented Lagrangian algorithm and that it asserts the validity of an error bound. We also provide proofs and counter-examples that show the relations of RCRCQ and RCPLD with other known constraint qualifications. In particular, RCPLD is strictly weaker than CPLD and RCRCQ, while still stronger than Abadie's constraint qualification. We also verify that the second order necessary optimality condition holds under RCRCQ. [ABSTRACT FROM AUTHOR]

In this paper, an approximate augmented Lagrangian function for nonlinear semidefinite programs is introduced. Some basic properties of the approximate augmented Lagrange function such as monotonicity and convexity are discussed. Necessary and sufficient conditions for approximate strong duality results are derived. Conditions for an approximate exact penalty representation in the framework of augmented Lagrangian are given. Under certain conditions, it is shown that any limit point of a sequence of stationary points of approximate augmented Lagrangian problems is a KKT point of the original semidefinite program and that a sequence of optimal solutions to augmented Lagrangian problems converges to a solution of the original semidefinite program. [ABSTRACT FROM AUTHOR]

We show how to exploit the structure inherent in the linear algebra for constrained nonlinear optimization problems when inequality constraints have been converted to equations by adding slack variables and the problem is solved using an augmented Lagrangian method. [ABSTRACT FROM AUTHOR]