Mathematical programming models with noisy, erroneous, or incomplete data are common in operations research applications. Difficulties with such data are typically dealt with reactively-through sensitivity analysis-or proactively-through stochastic programming formulations. In this paper, we characterize the desirable properties of a solution to models, when the problem data are described by a set of scenarios for their value, instead of using point estimates. A solution to an optimization model is defined as: solution robust if it remains "close" to optimal for all scenarios of the input data, and model robust if it remains "almost" feasible for all data scenarios. We then develop a general model formulation, called robust optimization RO, that explicitly incorporates the conflicting objectives of solution and model robustness. Robust optimization is compared with the traditional approaches of sensitivity analysis and stochastic linear programming. The classical diet problem illustrates the issues. Robust optimization models are then developed for several real-world applications: power capacity expansion; matrix balancing and image reconstruction; air-force airline scheduling; scenario immunization for financial planning; and minimum weight structural design. We also comment on the suitability of parallel and distributed computer architectures for the solution of robust optimization models.

Robust Optimization of Large-Scale Systems

http://www.cs.hunter.cuny.edu/~felisav/StochasticProcesses/Course_Material_files/BBG_jedc96.pdf

Monte Carlo methods for security pricing

The Benders decomposition algorithm: A literature review

We consider the design of multiproduct, multi-echelon supply chain networks. The networks comprise a number of manufacturing sites at fixed locations, a number of warehouses and distribution centers of unknown locations (to be selected from a set of potential locations), and finally a number of customer zones at fixed locations. The system is modeled mathematically as a mixed-integer linear programming optimization problem. The decisions to be determined include the number, location, and capacity of warehouses and distribution centers to be set up, the transportation links that need to be established in the network, and the flows and production rates of materials. The objective is the minimization of the total annualized cost of the network, taking into account both infrastructure and operating costs. A case study illustrates the applicability of such an integrated approach for production and distribution systems with or without product demand uncertainty.

Design of Multi-echelon Supply Chain Networks under Demand Uncertainty

A major issue in any application of multistage stochastic programming is the representation of the underlying random data process. We discuss the case when enough data paths can be generated according to an accepted parametric or nonparametric stochastic model. No assumptions on convexity with respect to the random parameters are required. We emphasize the notion of representative scenarios (or a representative scenario tree) relative to the problem being modeled.

Scenarios for Multistage Stochastic Programs

We develop an algorithm for solving nonlinear, two-stage stochastic problems with network recourse. The algorithm is based on the framework of row-action methods. The problem is formulated by replicating the first-stage variables and then adding nonanticipativity side constraints. A series of independent deterministic network problems are solved at each step of the algorithm, followed by an iterative step over the nonanticipativity constraints. The solution point of the iterates over the nonanticipativity constraints is obtained analytically. The row-action nature of the algorithm makes it suitable for parallel implementations. A data representation of the problem is developed that permits the massively parallel solution of all the scenario subproblems concurrently. The algorithm is implemented on a Connection Machine CM-2 with up to 32K processing elements and achieves computing rates of 276 MFLOPS. Very large problems-8,192 scenarios with a deterministic equivalent nonlinear program with 868,367 constraints and 2,474,017 variables-are solved within a few minutes. We report extensive numerical results regarding the effects of stochasticity on the efficiency of the algorithm.

/pdf/a-massively-parallel-algorithm-for-nonlinear-stochastic-2yhzdsys4a.pdf

A massively parallel algorithm for nonlinear stochastic network problems

We develop a scalable parallel implementation of the classical Benders decomposition algorithm for two-stage stochastic linear programs. Using a primal-dual, path-following algorithm for solving the scenario subproblems we develop a parallel implementation that alleviates the difficulties of load balancing. Furthermore, the dual and primal step calculations can be implemented using a data-parallel programming paradigm. With this approach the code effectively utilizes both the multiple, independent processors and the vector units of the target architecture, the Connection Machine CM-5. The, usually limiting, master program is solved very efficiently using the interior point code LoQo on the front-end workstation. The implementation scales almost perfectly with problem and machine size. Extensive computational testing is reported with several large problems with up to 2 million constraints and 13.8 million variables.

Scalable parallel Benders decomposition for stochastic linear programming

We develop four iterative algorithms for the solution of separable, convex nonlinear optimization problems with a single linear constraint and bounded variables. The design of the algorithms makes them suitable for implementation on massively parallel computers of the SIMD (i.e., Single Instruction, Multiple Data) class. The algorithms are specialized for the solution of network problems whereby the linear constraint reflects conservation of flow. Details of implementations on a Connection Machine CM-2 are reported. The numerical results indicate that all algorithms are very effective, and can solve very large problems. Three of the algorithms are also very efficient when implemented on the massively parallel system. Interestingly, the most effective algorithm (in number of steps required to solve the test problems) is the least efficient (in solution time) when implemented in parallel. INFORMS Journal on Computing, ISSN 1091-9856, was published as ORSA Journal on Computing from 1989 to 1995 under ISSN 08...

Massively Parallel Algorithms for Singly Constrained Convex Programs

Single Premium Deferred Annuities (SPDAs) are investment vehicles, offered to investors by insurance companies as a means of providing income past their retirement age. They are mirror images of insurance policies. However, the propensity of individuals to shift part, or all, of their investment into different annuities creates substantial uncertainties for the insurance company. In this paper we develop amultiperiod, dynamic stochastic program that deals with the problem of funding SPDA liabilities. The model recognizes explicitly the uncertainties inherent in this problem due to both interest rate volatility and the behavior of individual investors. Empirical results are presented with the use of the model for the funding of an SPDA liability stream using government bonds, mortgage-backed securities and derivative products.

A stochastic programming model for funding single premium deferred annuities

Multi-stage stochastic programs are typically extremely large, and can be prohibitively expensive to solve on the computer. In this paper we develop an algorithm for multistage programs that integrates the primal-dual row-action framework with proximal minimization. The algorithm exploits the structure of stochastic programs with network recourse, using a suitable problem formulation based on split variables, to decompose the solution into a large number of simple operations. It is therefore possible to use massively parallel computers to solve large instances of these problems. The algorithm is implemented on a Connection Machine CM-2 with up to 32K processors. We solve stochastic programs from an application from the insurance industry, as well as random problems, with up to 9 stages, and with up to 16392 scenarios, where the deterministic equivalent programs have a half million constraints and 1.3 million variables.

Soren S. Nielsen

Papers

A massively parallel algorithm for nonlinear stochastic network problems

Scalable parallel Benders decomposition for stochastic linear programming

Massively Parallel Algorithms for Singly Constrained Convex Programs

A stochastic programming model for funding single premium deferred annuities

Solving multistage stochastic network programs on massively parallel computers