Kurt Marti – författare

This book examines application and methods to incorporating stochastic parameter variations into the optimization process to decrease expense in corrective measures. Basic types of deterministic substitute problems occurring mostly in practice involve i) minimization of the expected primary costs subject to expected recourse cost constraints (reliability constraints) and remaining deterministic constraints, e.g. box constraints, as well as ii) minimization of the expected total costs (costs of construction, design, recourse costs, etc.) subject to the remaining deterministic constraints.

Basic results on the convergence and convergence rates of random search methods are presented. Moreover, for the improvement of the – sometimes very low – convergence rate of RSM, search methods based on optimal stochastic decision processes are presented. In order to improve the convergence behavior of RSM, the random search procedure is embedded into a stochastic decision process for an optimal control ofthe probability distributions of the search variates (mutation random variables).

This book examines optimization problems that in practice involve random model parameters. It outlines the computation of robust optimal solutions, i.e., optimal solutions that are insensitive to random parameter variations, where appropriate deterministic substitute problems are needed. Based on the probability distribution of the random data and using decision theoretical concepts, optimization problems under stochastic uncertainty are converted into corresponding deterministic problems.

Due to the probabilities and expectations involved, the book also shows how to apply approximative solution techniques. Several deterministic and stochastic approximation methods are provided: Taylor expansion methods, regression and response surface methods (RSM), probability inequalities, multiple linearization of survival/failure domains, discretization methods, convex approximation/deterministic descent directions/efficient points, stochastic approximation and gradient procedures, and differentiation formulas for probabilities and expectations.

The fourth edition of this classic text has been carefully and thoroughly revised. It includes new chapters on the solution of stochastic linear programs by discretization of the underlying probability distribution, and on solving deterministic optimization problems by means of controlled random search methods and multiple random search procedures. It also presents a new application of stochastic optimization methods to machine learning problems with different loss functions. For the computation of optimal feedback controls under stochastic uncertainty, besides the open-loop feedback procedures, a new method based on Taylor expansions with respect to the gain parameters is presented. 

The book is intended for researchers and graduate students who are interested in stochastics, stochastic optimization, and control. It will also benefit professionals and practitioners whose work involves technical, economicand/or operations research problems under stochastic uncertainty.

Optimization problems arising in practice involve random parameters. For the computation of robust optimal solutions, i.e., optimal solutions being insensitive with respect to random parameter variations, deterministic substitute problems are needed. Based on the distribution of the random data, and using decision theoretical concepts, optimization problems under stochastic uncertainty are converted into deterministic substitute problems. Due to the occurring probabilities and expectations, approximative solution techniques must be applied. Deterministic and stochastic approximation methods and their analytical properties are provided: Taylor expansion, regression and response surface methods, probability inequalities, First Order Reliability Methods, convex approximation/deterministic descent directions/efficient points, stochastic approximation methods, differentiation of probability and mean value functions. Convergence results of the resulting iterative solution procedures are given.

Das Buch gibt eine Einführung in das neue Gebiet der Analyse und Optimierung von Tragwerken unter stochastischer Unsicherheit. Es werden die Grundlagen ausführlich dargestellt und zum Teil von unterschiedlichen Standpunkten aus beleuchtet.

In Teil I wird die lineare Theorie der Stabtragwerke als Grundlage für die FEM entwickelt. Vorausgesetzt werden dabei nur wenige Kenntnisse aus  der Technischen Mechanik und der Ingenieurmathematik, insbesondere eine gewisse Vertrautheit mit der Matrizenrechnung.

In Teil II wird dargestellt, wie sich die Wahrscheinlichkeitsverteilungen der Verschiebungen und Spannungen in den Knoten aus denen der stochastischen Stabparameter und äußeren Lasten – zumindest approximativ – berechnen lassen.

In Teil III schließlich wird die Optimierung von Tragwerken mit stochastischen Parametern behandelt. Dazu wird ein geeignetes deterministisches Ersatzproblem des Ausgangsproblems mit stochastischen Modellparametern formuliert.

Eine kurze Beschreibung einiger Optimierungsverfahren findet man im letzten Abschnitt. Besondere Mühe wurde auf die zahlreichen und eingehend behandelten Beispiele verwandt.

Das Buch ist geschrieben für Studierende, praktisch tätige Ingenieure und Mathematiker.

Ongoing global changes bring fundamentally new scientific problems requiring new concepts and tools. A key issue concerns a vast variety of practically irreducible uncertainties, which challenge our traditional models and require new concepts and analytical tools. The uncertainty critically dominantes, e.g., the climate change debates. In short, the dilemma is concerned with enormous costs vs. massive uncertainties of potential extreme impacts. Traditional scientific approaches usually rely on real observations and experiments. Yet no sufficient observations exist for new problems, and "pure" experiments and learning by doing may be very expensive, dangerous, or simply impossible. In addition, available historical observations are contaminated by actions, policies. The complexity of new problems does not allow to achieve enough certainty by increasing the resolution of models or by bringing in more links. Hence, new tools for modeling and management of uncertainty are needed, as given in this book.

Optimization problems arising in practice involve random model parameters. For the computation of robust optimal solutions, i.e., optimal solutions being insenistive with respect to random parameter variations, appropriate deterministic substitute problems are needed. Based on the probability distribution of the random data, and using decision theoretical concepts, optimization problems under stochastic uncertainty are converted into appropriate deterministic substitute problems. Due to the occurring probabilities and expectations, approximative solution techniques must be applied. Several deterministic and stochastic approximation methods are provided: Taylor expansion methods, regression and response surface methods (RSM), probability inequalities, multiple linearization of survival/failure domains, discretization methods, convex approximation/deterministic descent directions/efficient points, stochastic approximation and gradient procedures, differentiation formulas for probabilities and expectations.

Support for addressing the on-going global changes needs solutions for new scientific problems which in turn require new concepts and tools. A key issue concerns a vast variety of irreducible uncertainties, including extreme events of high multidimensional consequences, e.g., the climate change. The dilemma is concerned with enormous costs versus massive uncertainties of extreme impacts. Traditional scientific approaches rely on real observations and experiments. Yet no sufficient observations exist for new problems, and "pure" experiments, and learning by doing may be expensive, dangerous, or impossible. In addition, the available historical observations are often contaminated by past actions, and policies. Thus, tools are presented for the explicit treatment of uncertainties using "synthetic" information composed of available "hard" data from historical observations, the results of possible experiments, and scientific facts, as well as "soft" data from experts'' opinions, and scenarios.

Managing safety of diverse systems requires decision-making under uncertainties and risks. Such systems are typically characterized by spatio-temporal heterogeneities, inter-dependencies, externalities, endogenous risks, discontinuities, irreversibility, practically irreducible uncertainties, and rare events with catastrophic consequences. Traditional scientific approaches rely on data from real observations and experiments; yet no sufficient observations exist for new problems, and experiments are usually impossible. Therefore, science-based support for addressing such new class of problems needs to replace the traditional “deterministic predictions” analysis by new methods and tools for designing decisions that are robust against the involved uncertainties and risks. The new methods treat uncertainties explicitly by using “synthetic” information derived by integration of “hard” elements, including available data, results of possible experiments, and formal representations of scientific facts, with “soft” elements based on diverse representations of scenarios and opinions of public, stakeholders, and experts. The volume presents such effective new methods, and illustrates their applications in different problem areas, including engineering, economy, finance, agriculture, environment, and policy making.