Even convexity and optimization. Handling strict inequalities (Q2215593)

This book is a valuable contribution to optimization, especially generalized versions of convex analysis and convex optimization, its theory, methods and applications, on the interface between (i) algebra and linear algebra, calculus and analysis, functional analysis and numerical analysis, with an emphasis on both existence and computational aspects, and rapidly growing fields of (ii) operational research, mathematical modeling, game theory and further applied areas, herewith paving the way to future advances in (iii) data science, statistical and machine learning, deep learning and, ultimately, mathematically supported artificial intelligence. The new monograph is timely and needed, theoretically and methodologically rigorous, and promising for practice. A main reason for the expected promises of this book lies in the fact that so many of our present and upcoming highly complex problems are nonlinear and tackled by locally approximating them with the help of linear ones, such as deeply investigated in this book. With this book's help in upcoming times, the gifted youth of mathematics, OR and other applied disciplines can be educated, guided and trained better. Many further scientific and practical inquiries can be conducted and real-world applications may be conducted, in the vast and quickly expanding world of modern research in the quantum and the cosmic worlds, for our emerging industries, our environment, for health, prosperity, justice and freedom of our humankind and whole creation. This book is the first one dedicated to linear systems with both general or weak lower inequalities (with ``less or equal'' relation) and strict lower inequalities (with ``less than'' relation), in total maybe infinitely many constraints, in the $n$-dimensional Euclidean space with standard inner product, with the solution set or feasible set $\sigma$ (called evenly convex by Fenchel in 1952) implicitly defined by these linear systems, and to those extended real-valued functions whose epigraphs (or, at least, their lower level sets) are evenly convex. To the authors the necessity of this book arose from the almost vanishing number of available monographs so far mind about existence theorems for linear systems containing strict inequalities. On the one hand, these were the monographs of Schrijver (on linear programming) and Mangasarian (on nonlinear optimization), Owen (on game theory), all referring to the overall index set of inequalities, $T$, being finite. On the other hand, Goberna and López (on semi-infinite optimization) address an arbitrary entity $T$. The situation is even harder regarding evenly convex sets and related functions, which have just been mentioned by now in Soltan's book, its 2nd edition, on convex sets (devoting 2 notes to evenly convex sets and evenly convex hulls) and the PhD thesis of Maggis (where evenly convex and evenly quasiconvex functions are used in finance and economics). This basically complete absence of linear systems containing strict inequalities, evenly convex sets and related functions in monographs and textbooks, has caused decades of delay and stagnation of interesting research and, over the years, rediscoveries of already known facts on mathematical subjects under different names. The authors call $\sigma$ as ordinary if there are no strict inequality constraints, in symbols: $S=\emptyset$ (i.e., the solution set is exclusively set up by weak linear inequalities). Then the solution set is closed and convex (a well-known type of evenly convex set). The authors say that the above linear system $\sigma$ is finite if $T$ is finite. Otherwise they call it as semi-infinte (a name coming from their finitely many variables or dimensions, and their infinitely many inequalities). Finite ordinary linear systems were firstly considered by Fourier in 1826 and secondly by Farkas around 1900, characterizing equilibrium points in mechanics. For their crucial role in linear programming, they have been well analyzed, because this widely used optimization model is computationally equivalent to feasibility problem for finite ordinary linear systems. This comes out of duality theorem and the Fourier-Motzkin elimination method. Semi-infinite ordinary linear systems were first studied by Haar in 1824, in a paper unnoticed until the publication of a free translation by Charnes, Cooper and Kortanek in 1963, in their seminal paper on linear semi-infinite programming. Here the authors merely address ordinary linear systems with comparative purposes as they are treated in the abovementioned monographs, whereas their non-ordinary counterparts have been systematically neglected by now. Finite non-ordinary linear systems were first regarded by Gordan in 1873. They were used a couple of times in the 20th century, e.g., by Kuhn in 1956, Walkup and Wets in 1969, and Kannan in 1992, while they have become intensively employed in the 21st century in OR (such as in optimization and games), computational sciences, etc. Their feasible sets, which the authors name as evenly convex polyhedra, were rediscovered and studied repeatedly under names such as wholefaced polyhedra, copolyhedra, not necessarily closed convex polyhedra, G-polyhedra, and semiclosed polyhedra. As much as functions are concerned, right in same way as quasiconvex and convex functions are defined by their lower level sets or epigraphs being convex, respectively, evenly quasiconvex and evenly convex functions are defined by their lower level sets or epigraphs being evenly convex, respectively. Evenly quasiconvex functions, applied in economics, were introduced by Martínez-Legaz under the name of normal quasiconvex. Independently they were presented by Passy and Prisman in the early 1980s. Evenly convex functions became introduced by Rodríguez and Vicente-Pérez in 2011. The authors' team enjoy a remarkable reputation, especially, for their strong research achievements over many decades, for example, in semi-infinite optimization and related subjects, along with their coauthors from Alicante, from other Spanish cities -- premium centers as well, and from all over the world. Therefore, the present book stands another academic success of the authors and their team. It is based on vast experience and strong foundation in mathematics and OR. This innovative and truly unique work is offered to us together with several special and additional benefits and promises, namely: (i) It contains numerous ``classical'' OR and mathematical applications, so that the reader is well motivated to get involved into the scientific terminology of the book and trust in its real-world meaning and impact. Those applications could also be used by the reader in education and other lectures and presentations, (ii) it contains many classical and, especially, emerging conditions and their characterizations which enrich the mathematics of operational research and intellectually train and trim the readers both from theory and from practice, (iii) it is highly flexible in order to be applied in future on other ``linear problems'', to be generalized in theory for ``nonlinear problems'' and applied on them, and on infinite-dimensional and stochastic problems, both linear and nonlinear ones. Here we may think of infinite programming, infinite kernel learning, calculus of variations, theories of optimal control and dynamical games, and of their stochastic counterparts. The four chapters and all the other parts of this book are these. After the Preface, Acknowledgments, etc., there are: Chapter 1: Evenly convex sets: linear systems containing strict inequalities, Chapter 2: Evenly convex polyhedra: finite linear systems containing strict inequalities, Chapter 3: Evenly quasiconvex functions, Chapter 4: Evenly convex functions, and the Appendix: Extensions to infinite-dimensional spaces. On Chapter 1: Here the authors are concerned with linear systems of a possibly infinite number of weak or strict inequalities and with their solution sets, which they name as evenly convex sets. Both sets and functions are ``equivalent'' to each other. In different ways they first characterize evenly convex sets which fulfill most of the well-known properties of closed convex sets. Any set has an evenly convex hull. Then they turn to the operations with evenly convex hulls and how they relate to other hulls. Next on the agenda are diverse separation theorems with evenly convex sets, followed by existence theorems for linear systems with strict inequalities and characterizations of linear inequalities. This permits them to tackle set containment problems with convex sets. Next, the authors turn to evenly linear semi-infinite programming problems, before they make applications to the following subjects: polarity that once inspired the authors to their evenly convex set, semi-infinite games, approximate reasoning, optimality conditions, and strict separation of set families. On Chapter 2: Here the authors deal with linear systems containing finitely many weak or strict inequalities, whose solution of feasible sets (if nonempty) are called as evenly (or: e-) convex polyhedral. All corresponding results in Chapter 1 remain valid, but the finiteness of linear representations of e-polyhedra allows obtaining specific results and methods. Many convex-set families have both Internal and external representations. The outstanding advantage of the polyhedra against the other two families of convex sets is given by the double description methods permitting an external representation from the internal one, and vice versa. In this chapter, the authors extend this method from polyhedra to e-polyhedra. They first describe the Fourier-Motzkin elimination method to project a given e-polyhedron on the coordinate planes. Iteratively applied this allows the finding of solutions for finite linear systems containing strict inequalities. Then they associate each finite non-ordinary system with its so-called ``representative cone'', containing all relevant information about the systems. By this, existence theorems and characterizations of the consequent inequalities can be simplified for an arbitrary number of constraints. Then they present the aimed double description method for e-polyhedra, before they investigate the minimization of linear functions under weak and strict linear inequalities, named as evenly linear programming problems. On Chapter 3: After all the careful preparations made in the previous chapters, now the authors investigate deeply on evenly quasiconvex functions, evenly quasiconvex hulls, conjugacy and subdifferentiability, duality in quasiconvex optimization, and an application to consumer theory, and bibliographic notes. In this chapter, again the mathematics is done with rigor and completely. Many small examples facilitate an easier understanding and help the newcomer in the field, or any reader who comes from fields rather different from mathematics, to endure, enjoy and memorized key conditions, configurations and results. On Chapter 4: Now the authors introduce evenly convex functions as such whose epigraphs are evenly convex sets. They unfold a duality theory for nonlinear optimization involving evenly convex optimization. First they present the main properties of this problem class of convex functions which includes the important subclass of lower semicontinuous convex functions, whose relevance in convex analysis comes from the fact that Fenchel conjugacy is an ``involution'' on most of them. Then the authors offer the evenly convex hull of a function and schemes of conjugation for evenly convex functions, before they unfold the c-conjugate duality theory along with regularity conditions of type ``closedness''. The latter ones will be expressed by even convexity of the functions involved, for both strong and stable strong duality of convex problems. Appendix A: At least a dozen of results provided by this book also possess an infinite-dimensional counterpart, in which the finite-dimensional Euclidean space can be substituted and in fact generalized by some Banach spaces or even a locally convex separated, i.e., Hausdorff, topological vector space. These cases and places of the book are carefully listed in a table. This excellent book is clearly and well structured, analytically deep, well exemplified, beautifully illustrated, and written with care and taste. In the future, refinements and extensions in analytic, theoretical foundations and algorithmic techniques may be considered by the authors and within the academic family, prepared, supported and inspired by this book. These could be made in the form of articles and monographs, and in terms of, e.g., singularity and Morse theory, of non-smooth, discontinuous or robust optimization, optimal control, stochastic optimal control, and of discrete-combinatorial elements such as thresholds, regime switching and hybrid systems, collaborative game theory and stochastic sames. Just to mention one most recent research area where those particular mixtures between weak and strict inequality constraints naturally occur and can help in modeling, we mention the new studies by the author of this report, G.-W. Weber, on boundaries, surfaces and membranes, on interior and exterior, in dimensionally generalized space-time, and in all states of matter. Those future advancements could nurture and support successes in management, economics and finance, in the natural sciences and high-technology, in bio-, neuro- science and medical sciences, in environmental, geo-, earth- and space-sciences, and in societal and developmental sciences. PS: Some more information about the book and its underlying series are noteworthy. We quote them here. The EURO Advanced Tutorials on Operational Research are a series of short books devoted to an advanced topic -- a topic that is not treated in depth in available textbooks. The series covers comprehensively all aspects of operations research. The scope of a Tutorial is to provide an understanding of an advanced topic to young researchers, such as Ph.D. students or Post-docs, but also to senior researchers and practitioners. Tutorials may be used as textbooks in graduate courses. More information about this series can be found at \url{http://www.springer.com/series/13840}. The editors of EURO Advanced Tutorials on Operational Research are Prof. Dr. M. Grazia Speranza (University of Brescia, Italy) and Prof. Dr. José Fernando Oliveira (University of Porto, Portugal).