Bøger af Darinka Dentcheva

This book offers a comprehensive presentation of the theory and methods of risk-averse optimization and control. Problems of this type arise in finance, energy production and distribution, supply chain management, medicine, and many other areas, where not only the average performance of a stochastic system is essential, but also high-impact and low-probability events must be taken into account. The book is a self-contained presentation of the utility theory, the theory of measures of risk, including systemic and dynamic measures of risk, and their use in optimization and control models. It also covers stochastic dominance relations and their application as constraints in optimization models. Optimality conditions for problems with nondifferentiable and nonconvex functions and operators involving risk measures and stochastic dominance relations are discussed. Much attention is paid to multi-stage risk-averse optimization problems and to risk-averse Markov decision problems. Specialized algorithms for solving risk-averse optimization and control problems are presented and analyzed: stochastic subgradient methods for risk optimization, decomposition methods for dynamic problems, event cut and dual methods for stochastic dominance constraints, and policy iteration methods for control problems. The target audience is researchers and graduate students in the areas of mathematics, business analytics, insurance and finance, engineering, and computer science. The theoretical considerations are illustrated with examples, which make the book useful material for advanced courses in the area.

The methodology presented in this work yields existence of selections and Castaing representations of these sets that enjoy stability properties. Particularly, work on differential stability of solution sets of two-stage stochastic optimization problems is presented. We identify conditions such that the optimal value function has first- and second-order directional derivatives and the solution-set mapping is directionally differentiable into admissible directions. Moreover, the form of the semi-derivative is identified, and we have given a formula for it. The sensitivity analysis is carried out by exploiting structural properties of the optimization model. We obtain differentiability properties of solution sets and extend earlier results on directional differentiability of optimal values considerably.

A comprehensive treatment of optimization problems involving uncertain parameters for which stochastic models are available.