AN INTERIOR TRUST REGION ALGORITHM FOR NONLINEAR MINIMIZATION WITH LINEAR CONSTRAINTS

AbstractAn interior trust-region-based algorithm for linearly constrained minimization problems is proposed and analyzed. This algorithm is similar to trust region algorithms for unconstrained minimization: a trust region subproblem on a subspace is solved in each iteration. We establish that the proposed algorithm has convergence properties analogous to those of the trust region algorithms for unconstrained minimization. Namely, every limit point of the generated sequence satisfies the Krush-Kuhn-Tucker (KKT) conditions and at least one limit point satisfies second order necessary optimality conditions. In addition, if one limit point is a strong local minimizer and the Hessian is Lipschitz continuous in a neighborhood of that point, then the generated sequence converges globally to that point in the rate of at least 2-step quadratic. We are mainly concerned with the theoretical properties of the algorithm in this paper. Implementation issues and adaptation to large-scale problems will be addressed in a     

Modular solvers for image restoration problems using the discrepancy principle

Many problems in image restoration can be formulated as either an unconstrained non‐linear minimization problem, usually with a Tikhonov‐like regularization, where the regularization parameter has to be determined; or as a fully constrained problem, where an estimate of the noise level, either the variance or the signal‐to‐noise ratio, is available. The formulations are mathematically equivalent. However, in practice, it is much easier to develop algorithms for the unconstrained problem, and not always obvious how to adapt such methods to solve the corresponding constrained problem. In this paper, we present a new method which can make use of any existing convergent method for the unconstrained problem to solve the constrained one. The new method is based on a Newton iteration applied to an extended system of non‐linear equations, which couples the constraint and the regularized problem, but it does not require knowledge of the Jacobian of the irregularity functional. The existing solver is only used as a black box solver, which for a fixed regularization parameter returns an improved solution to the unconstrained minimization problem given an initial guess. The new modular solver enables us to easily solve the constrained image restoration problem; the solver automatically identifies the regularization parameter, during the iterative solution process. We present some numerical results. The results indicate that even in the worst case the constrained solver requires only about twice as much work as the unconstrained one, and in some instances the constrained solver can be even faster. Copyright © 2002 John Wiley & Sons, Ltd.     

MM algorithms for geometric and signomial programming

This paper derives new algorithms for signomial programming, a generalization of geometric programming. The algorithms are based on a generic principle for optimization called the MM algorithm. In this setting, one can apply the geometric-arithmetic mean inequality and a supporting hyperplane inequality to create a surrogate function with parameters separated. Thus, unconstrained signomial programming reduces to a sequence of one-dimensional minimization problems. Simple examples demonstrate that the MM algorithm derived can converge to a boundary point or to one point of a continuum of minimum points. Conditions under which the minimum point is unique or occurs in the interior of parameter space are proved for geometric programming. Convergence to an interior point occurs at a linear rate. Finally, the MM framework easily accommodates equality and inequality constraints of signomial type. For the most important special case, constrained quadratic programming, the MM algorithm involves very simple updates.     

A new augmented Lagrangian function for inequality constraints in nonlinear programming problems

In this paper, a new augmented Lagrangian function is introduced for solving nonlinear programming problems with inequality constraints. The relevant feature of the proposed approach is that, under suitable assumptions, it enables one to obtain the solution of the constrained problem by a single unconstrained minimization of a continuously differentiable function, so that standard unconstrained minimization techniques can be employed. Numerical examples are reported.     

On the role of continuously differentiable exact penalty functions in constrained global optimization

The aim of this paper is to show that the new continuously differentiable exact penalty functions recently proposed in literature can play an important role in the field of constrained global optimization. In fact they allow us to transfer ideas and results proposed in unconstrained global optimization to the constrained case.First, by drawing our inspiration from the unconstrained case and by using the strong exactness properties of a particular continuously differentiable penalty function, we propose a sufficient condition for a local constrained minimum point to be global.Then we show that every constrained local minimum point satisfying the second order sufficient conditions is an attraction point for a particular implementable minimization algorithm based on the considered penalty function. This result can be used to define new classes of global algorithms for the solution of general constrained global minimization problems. As an example, in this paper we describe a simulated annealing algorithm which produces a sequence of points converging in probability to a global minimum of the original constrained problem.     

An Ideal Penalty Function for Constrained Optimization

A well known approach to constrained optimization is via a sequenceof unconstrained minimization calculations applied to a penaltyfunction. This paper shown how it is posiible to generalizePowell's penelty function to solve constrained problems withboth equality and inequality constraints. The resulting methodsare equivalent to the Hestenes' method of multipliers, and ageneralization of this to inequality constraints suggested byRockafellar. Local duality results (not all of which have appearedbefore) for these methods are reviewed, with particular emphasison those of practical importance. It is shown that various strategiesfor varying control parameters are possible, all of which canbe viewed as Newton or Newton-like iterations applied to thedual problem. Practical strategies for guaranteeing convergenceare also discussed. A wide selection of numerical evidence isreported, and the algorithms are compared both amongst themselvesand with other penalty function methods. The new penalty functionis well conditioned, without singularities, and it is not necessaryfor the control parameters to tend to infinity in order to forceconvergence. The rate of convergence is rapid and high accuracyis achieved in few unconstrained minimizations.; furthermorethe computational effort for successive minimizations goes downrapidly. The methods are very easy to program efficiently, usingan established quasi-Newton subroutine for unconstrained minimization.     

Elimination of bounds in optimization problems by transforming variables

The problem of minimizing a nonlinear objective function ofn variables, with continuous first and second partial derivatives, subject to nonnegativity constraints or upper and lower bounds on the variables is studied. The advisability of solving such a constrained optimization problem by making a suitable transformation of its variables in order to change the problem into one of unconstrained minimization is considered. A set of conditions which guarantees that every local minimum of the new unconstrained problem also satisfies the first-order necessary (Kuhn—Tucker) conditions for a local minimum of the original constrained problem is developed. It is shown that there are certain conditions under which the transformed objective function will maintain the convexity of the original objective function in a neighborhood of the solution. A modification of the method of transformations which moves away from extraneous stationary points is introduced and conditions under which the method generates a sequence of points which converges to the solution at a superlinear rate are given.     

A sufficient optimality theoriem for protoconvex programs

We establish two first order sufficient optimality theorems; one for unconstrained nonlinear minimization problem, and the other for constrained nonlinear minimization problems, both with non-differentiable protoconvex or quasiconvex data functions that are not necessarily locally Lipschitz     

Quadratic Termination Properties of Minimization Algorithms I. Statement and Discussion of Results

There is a family of gradient algorithms (Broyden, 1970) thatincludes many useful methods for calculating the least valueof a function F(x), and some of these algorithms have been extendedto solve linearly constrained problems (Fletcher, 1971). Somenew and fundamental properties of these algorithms are given,in the case that F(x) is a positive definite quadratic function.In particular these properties are relevant to the case whenonly some of the iterations of an algorithm make a completelinear search. They suggest that Goldfarb's (1969) algorithmfor linearly constrained problems has excellent quadratic terminationproperties, and it is proved that these properties are betterthan has been stated in previously published papers. Also anew technique is identified for unconstrained minimization withoutlinear searches.     

Alternating Minimization as Sequential Unconstrained Minimization: A Survey

Sequential unconstrained minimization is a general iterative method for minimizing a function over a given set. At each step of the iteration we minimize the sum of the objective function and an auxiliary function. The aim is to select the auxiliary functions so that, at least, we get convergence in function value to the constrained minimum. The SUMMA is a broad class of these methods for which such convergence holds. Included in the SUMMA class are the barrier-function methods, entropic and other proximal minimization algorithms, the simultaneous multiplicative algebraic reconstruction technique, and, after some reformulation, penalty-function methods. The alternating minimization method of Csiszár and Tusnády also falls within the SUMMA class, whenever their five-point property holds. Therefore, the expectation maximization maximum likelihood algorithm for the Poisson case is also in the SUMMA class.     

Dynamic Programming Algorithms for Generating Optimal Strip Layouts

This paper presents dynamic programming algorithms for generating optimal strip layouts of equal blanks processed by shearing and punching. The shearing and punching process includes two stages. The sheet is cut into strips using orthogonal guillotine cuts at the first stage. The blanks are punched from the strips at the second stage. The algorithms are applicable in solving the unconstrained problem where the blank demand is unconstrained, the constrained problem where the demand is exact, the unconstrained problem with blade length constraint, and the constrained problem with blade length constraint. When the sheet length is longer than the blade length of the guillotine shear used, the dynamic programming algorithm is applied to generate optimal layouts on segments of lengths not longer than the blade length, and the knapsack algorithm is employed to find the optimal layout of the segments on the sheet. Experimental computations show that the algorithms are efficient.     

Integral global minimization: Algorithms,implementations and numerical tests

The theoretical foundation of integral global optimization has become widely known and well accepted [4],[24],[25]. However, more effort is needed to demonstrate the effectiveness of the integral global optimization algorithms. In this work we detail the implementation of the integral global minimization algorithms. We describe how the integral global optimization method handles nonconvex unconstrained or box constrained, constrained or discrete minimization problems. We illustrate the flexibility and the efficiency of integral global optimization method by presenting the performance of algorithms on a collection of well known test problems in global optimization literature. We provide the software which solves these test problems and other minimization problems. The performance of the computations demonstrates that the integral global algorithms are not only extremely flexible and reliable but also very efficient.Research supported partially by NSERC grant and Mount St Vincent University research grant.     

Another version of the proximal point algorithm in a Banach space

In this paper, we study some non-traditional schemes of proximal point algorithm for nonsmooth convex functionals in a Banach space. The proximal approximations to their minimal points and/or their minimal values are considered separately for unconstrained and constrained minimization problems on convex closed sets. For the latter we use proximal point algorithms with the metric projection operators and first establish the estimates of the convergence rate with respect to functionals. We also investigate the perturbed projection proximal point algorithms and prove their stability. Some results concerning the classical proximal point method for minimization problems in a Banach space is also presented in this paper.     

A Single Component Mutation Evolutionary Programming

In this paper, a novel evolutionary programming is proposed for solving the upper and lower bound optimization problems as well as the linear constrained optimization problems. There are two characteristics of the algorithm: first, only one component of the current solution is mutated in each iteration; second, it can solve the linear constrained optimization problems directly without converting it into unconstrained problems. By solving two kinds of the optimization problems, the algorithm can not only effectively find optimal or close to optimal solutions but also reduce the number of function evolutions compared with the other heuristic algorithms.     

Nonlinear rescaling and proximal-like methods in convex optimization

The nonlinear rescaling principle (NRP) consists of transforming the objective function and/or the constraints of a given constrained optimization problem into another problem which is equivalent to the original one in the sense that their optimal set of solutions coincides. A nonlinear transformation parameterized by a positive scalar parameter and based on a smooth sealing function is used to transform the constraints. The methods based on NRP consist of sequential unconstrained minimization of the classical Lagrangian for the equivalent problem, followed by an explicit formula updating the Lagrange multipliers. We first show that the NRP leads naturally to proximal methods with an entropy-like kernel, which is defined by the conjugate of the scaling function, and establish that the two methods are dually equivalent for convex constrained minimization problems. We then study the convergence properties of the nonlinear rescaling algorithm and the corresponding entropy-like proximal methods for convex constrained optimization problems. Special cases of the nonlinear rescaling algorithm are presented. In particular a new class of exponential penalty-modified barrier functions methods is introduced. Partially supported by the National Science Foundation, under Grants DMS-9201297, and DMS-9401871. Partially supported by NASA Grant NAG3-1397 and NSF Grant DMS-9403218.     

The Equality Constrained Indefinite Least Squares Problem: Theory and Algorithms

We present theory and algorithms for the equality constrained indefinite least squares problem, which requires minimization of an indefinite quadratic form subject to a linear equality constraint. A generalized hyperbolic QR factorization is introduced and used in the derivation of perturbation bounds and to construct a numerical method. An alternative method is obtained by employing a generalized QR factorization in combination with a Cholesky factorization. Rounding error analysis is given to show that both methods have satisfactory numerical stability properties and numerical experiments are given for illustration. This work builds on recent work on the unconstrained indefinite least squares problem by Chandrasekaran, Gu, and Sayed and by the present authors.     

Solution to Global Minimization of Polynomials by Backward Differential Flow

This paper presents a study on solutions to the global minimization of polynomials. The backward differential flow by the K–T equation with respect to the optimization problem is introduced to deal with a ball-constrained optimization problem. The unconstrained optimization is reduced to a constrained optimization problem which can be solved by a backward differential flow. Some examples are illustrated with an algorithm for computing the backward flow.     

对等式约束非线性规划问题的Hestenes-Powell增广拉格朗日函数的进一步研究

本文对用无约束极小化方法求解等式约束非线性规划问题的Hestenes-Powell 增广拉格朗日函数作了进一步研究．在适当的条件下,我们建立了Hestenes-Powell增广拉格朗日函数在原问题变量空间上的无约束极小与原约束问题的解之间的关系,并且也给出了Hestenes-Powell增广拉格朗日函数在原问题变量和乘子变量的积空间上的无约束极小与原约束问题的解之间的一个关系．因此,从理论的观点来看,原约束问题的解和对应的拉格朗日乘子值不仅可以用众所周知的乘子法求得,而且可以通过对Hestenes-Powell 增广拉格朗日函数在原问题变量和乘子变量的积空间上执行一个单一的无约束极小化来获得．     

A sufficient condition for exact penalty functions

In this paper, we use the penalty approach for constrained minimization problems in infinite dimensional Banach spaces. A penalty function is said to have the exact penalty property if there is a penalty coefficient for which a solution of an unconstrained penalized problem is a solution of the corresponding constrained problem. We establish a simple sufficient condition for exact penalty property for two large classes of constrained minimization problems.     

A Learning Framework for Neural Networks Using Constrained Optimization Methods

Conventional supervised learning in neural networks is carried out by performing unconstrained minimization of a suitably defined cost function. This approach has certain drawbacks, which can be overcome by incorporating additional knowledge in the training formalism. In this paper, two types of such additional knowledge are examined: Network specific knowledge (associated with the neural network irrespectively of the problem whose solution is sought) or problem specific knowledge (which helps to solve a specific learning task). A constrained optimization framework is introduced for incorporating these types of knowledge into the learning formalism. We present three examples of improvement in the learning behaviour of neural networks using additional knowledge in the context of our constrained optimization framework. The two network specific examples are designed to improve convergence and learning speed in the broad class of feedforward networks, while the third problem specific example is related to the efficient factorization of 2-D polynomials using suitably constructed sigma-pi networks.