Aggregation bounds in stochastic linear programming

Stochastic linear programs become extremely large and complex as additional uncertainties and possible future outcomes are included in their formulation. Row and column aggregation can significantly reduce this complexity, but the solutions of the aggregated problem only provide an approximation of the true solution. In this paper, error bounds on the value of the optimal solution of the original problem are obtained from the solution of the aggregated problem. These bounds apply for aggregation of both random variables and time periods.     

Aggregation and discretization in multistage stochastic programming

Multistage stochastic programs have applications in many areas and support policy makers in finding rational decisions that hedge against unforeseen negative events. In order to ensure computational tractability, continuous-state stochastic programs are usually discretized; and frequently, the curse of dimensionality dictates that decision stages must be aggregated. In this article we construct two discrete, stage-aggregated stochastic programs which provide upper and lower bounds on the optimal value of the original problem. The approximate problems involve finitely many decisions and constraints, thus principally allowing for numerical solution.      

Test problems in stochastic multistage programming

This paper provides a set of stochastic multistage programs where the evolvement of uncertain factors is given by stochastic processes. We treat a practical problem statement within the field of managing fixed-income securities. Detailed information on the used parameter values in various interest rate models is given. Barycentric approximation is applied to obtain computational results; different measures of the achieved goodness of approximation are indicated     

Implicit representation of generalized variable upper bounds in linear programming

In certain linear programs, especially those derived from integer programs, large numbers of constraints may have very simple form. Examples are:x _ij 1 (simple upper bounds [SUB]), _i x _ij = 1 (generalized upper bounds [GUB]) andx _ij y _i (variable upper bounds [VUB]). A class of constraints called generalized VUB [GVUB] is introduced which includes GUB and VUB as special cases. Also introduced is a method for representing GVUB constraints implicitly within the mechanics of the simplex method.Research supported in part by the Mobil Oil Corporation.     

An implementation of the simplex method for linear programming problems with variable upper bounds

Special methods for dealing with constraints of the formx _j x _k , called variable upper bounds, were introduced by Schrage. Here we describe a method that circumvents the massive degeneracy inherent in these constraints and show how it can be implemented using triangular basis factorizations.This research was partially supported by National Science Foundation Grant ECS-7921279 and by a Guggenheim fellowship.     

The stochastic bottleneck linear programming problem

In this paper we consider some stochastic bottleneck linear programming problems. We overview the solution methods in the literature. In the case when the coefficients of the objective functions are simple randomized, the minimum-risk approach will be used for solving these problems. We prove that, under some positivity conditions, these stochastic problems are reduced to certain deterministic bottleneck linear problems. An application of these problems to bottleneck spanning tree problems is given. Two simple numerical examples are presented. This paper was written when I.M. Stancu-Minasian was visiting the Instituto Complutense de Análisis Económico, in the Universidad Complutensen de Madrid, from October 1, 1997 to November 15, 1997 and from October 24, 1998 to November, 9, 1998, as invited researcher. He is grateful to the Institution.     

On the Glivenko-Cantelli problem in stochastic programming: Mixed-integer linear recourse

Expected recourse functions in linear two-stage stochastic programs with mixed-integer second stage are approximated by estimating the underlying probability distribution via empirical measures. Under mild conditions, almost sure uniform convergence of the empirical means to the original expected recourse function is established.     

Analysis of stochastic dual dynamic programming method

In this paper we discuss statistical properties and convergence of the Stochastic Dual Dynamic Programming (SDDP) method applied to multistage linear stochastic programming problems. We assume that the underline data process is stagewise independent and consider the framework where at first a random sample from the original (true) distribution is generated and consequently the SDDP algorithm is applied to the constructed Sample Average Approximation (SAA) problem. Then we proceed to analysis of the SDDP solutions of the SAA problem and their relations to solutions of the “true” problem. Finally we discuss an extension of the SDDP method to a risk averse formulation of multistage stochastic programs. We argue that the computational complexity of the corresponding SDDP algorithm is almost the same as in the risk neutral case.     

Computational aspects of two-segment separable programming

Recursive separable programming algorithms based on local, two-segment approximations are described for the solution of separable convex programs. Details are also given for the computation of lower bounds on the optimal value by both a primal and a dual approach, and these approaches are compared. Computational comparisons of the methods are provided for a variety of test problems, including a water supply application (with more than 600 constraints and more than 900 variables) and an econometric modelling problem (with more than 200 variables). Research supported by National Science Foundation Grants MCS74-20584 A02 and MCS-7901066.     

The value of the stochastic solution in stochastic linear programs with fixed recourse

Stochastic linear programs have been rarely used in practical situations largely because of their complexity. In evaluating these problems without finding the exact solution, a common method has been to find bounds on the expected value of perfect information. In this paper, we consider a different method. We present bounds on the value of the stochastic solution, that is, the potential benefit from solving the stochastic program over solving a deterministic program in which expected values have replaced random parameters. These bounds are calculated by solving smaller programs related to the stochastic recourse problem.This paper is an extension of part of the author's dissertation in the Department of Operations Research, Stanford University, Stanford, California. The research was supported at Stanford by the Department of Energy under Contract DE-AC03-76SF00326, PA#DE-AT03-76ER72018, Office of Naval Research under Contract N00014-75-C-0267 and the National Science Foundation under Grants MCS76-81259, MCS-7926009 and ECS-8012974 (formerly ENG77-06761).     

Optimality in transient markov chains and linear programming

It is shown how a discrete Markov programming problem can be transformed, using a linear program, into an equivalent problem from which the optimal decision rule can be trivially deduced. This transformation is applied to problems which have either transient probabilities or discounted costs.This research was supported by the National Research Council of Canada, Grant A7751.     

Dynamic programming in stochastic control of systems with delay

We consider optimal control problems for systems described by stochastic differential equations with delay (SDDE). We prove a version of Bellman's principle of optimality (the dynamic programming principle) for a general class of such problems. That the class in general means that both the dynamics and the cost depends on the past in a general way. As an application, we study systems where the value function depends on the past only through some weighted average. For such systems we obtain a Hamilton-Jacobi-Bellman partial differential equation that the value function must solve if it is smooth enough. The weak uniqueness of the SDDEs we consider is our main tool in proving the result. Notions of strong and weak uniqueness for SDDEs are introduced, and we prove that strong uniqueness implies weak uniqueness, just as for ordinary stochastic differential equations.     

Multistage stochastic programming: Error analysis for the convex case

We consider convex stochastic multistage problems and present an approximation technique which allows to analyse the error with respect to time. The technique is based on barycentric approximation of conditional and marginal probability spaces and requiresstrict nonanticipativity for the constraint multifunction and thesaddle property for the value functions.Part of this work was carried out at the Institute of Operations Research of the University of Zurich.     

Nonlinear programming and stationary strategies in stochastic games

We show that an undiscounted stochastic game possesses optimal stationary strategies if and only if a global minimum with objective value zero can be found to an appropriate nonlinear program with linear constraints. This nonlinear program arises as a method for solving a certain bilinear system, satisfaction of which is also equivalent to finding a stationary optimal solution for the game. The objective function of the program is a nonnegatively valued quadric polynomial.This research was supported in part by the National Science Foundation under the grant #ECS-8503440. We wish to thank the referee for many helpful comments and in streamlining the presentation.     

On the computational utility of posynomial geometric programming solution methods

Ten codes or code variants were used to solve the five equivalent posynomial GP problem formulations. Four of these codes were general NLP codes; six were specialized GP codes. A total of forty-two test problems was solved with up to twenty randomly generated starting points per problem. The convex primal formulation is shown to be intrinsically easiest to solve. The general purpose GRG code called OPT appears to be the most efficient code for GP problem solution. The reputed superiority of the specialized GP codes GGP and GPKTC appears to be largely due to the fact that these codes solve the convex primal formulation. The dual approaches are only likely to be competitive for small degree of difficulty, tightly constrained problems.     

Improving constants of strong convexity in linear stochastic programming

We derive formulas for constants of strong convexity (CSCs) of expectation functions encountered in two-stage stochastic programs with linear recourse. One of them yields a CSC as the optimal value of a certain quadratically constrained quadratic program, another one in terms of the thickness of the feasibility polytope of the dual problem associated to the recourse problem. CSCs appear in Hoelder-type estimates relating the distance of optimal solution sets of stochastic programs to a suitable distance of underlying probability distributions.     

Implementing bounds-based approximations in convex-concave two-stage stochastic programming

This paper is concerned with implementational issues and computational testing of bounds-based approximations for solving two-stage stochastic programs with fixed recourse. The implemented bounds are those derived by the authors previously, using first and cross moment information of the random parameters and a convex-concave saddle property of the recourse function. The paper first examines these bounds with regard to their tightness, monotonic behavior, convergence properties, and computationally exploitable decomposition structures. Subsequently, the bounds are implemented under various partitioning/refining strategies for the successive approximation. The detailed numerical experiments demonstrate the effectiveness in solving large scenario-based two-stage stochastic optimization problems throughsuccessive scenario clusters induced by refining the approximations.     

Inference of statistical bounds for multistage stochastic programming problems

We discuss in this paper statistical inference of sample average approximations of multistage stochastic programming problems. We show that any random sampling scheme provides a valid statistical lower bound for the optimal (minimum) value of the true problem. However, in order for such lower bound to be consistent one needs to employ the conditional sampling procedure. We also indicate that fixing a feasible first-stage solution and then solving the sampling approximation of the corresponding (T–1)-stage problem, does not give a valid statistical upper bound for the optimal value of the true problem.Supported, in part, by the National Science Foundation under grant DMS-0073770.     

Relaxations of linear programming problems with first order stochastic dominance constraints

Linear stochastic programming problems with first order stochastic dominance (FSD) constraints are non-convex. For their mixed 0-1 linear programming formulation we present two convex relaxations based on second order stochastic dominance (SSD). We develop necessary and sufficient conditions for FSD, used to obtain a disjunctive programming formulation and to strengthen one of the SSD-based relaxations.     

Nonlinear programming methods in the presence of noise

The problem of minimizing a nonlinear function with nonlinear constraints when the values of the objective, the constraints and their gradients have errors, is studied. This noise may be due to the stochastic nature of the problem or to numerical error.Various previously proposed methods are reviewed. Generally, the minimization algorithms involve methods of subgradient optimization, with the constraints introduced through penalty, Lagrange, or extended Lagrange functions. Probabilistic convergence theorems are obtained. Finally, an algorithm to solve the general convex (nondifferentiable) programming problem with noise is proposed.Originally written for presentation at the 1976 Budapest Symposium on Mathematical Programming.