The concepts of continuity for set-valued mappings were introduced by G. Bouligand and K. Kuratowski. There are two ways defining differentiability of set-valued mapping. One is defined by classical differentiability theorem and another is defined by normal cone which was introduced by B.S. Mordukhovich. In this thesis, we survey various definitions of continuity and differentiability for set-valued mapping.

Cette thèse est consacrée à l'analyse variationnelle des problèmes d'équilibre avec contraintes d'équilibre (problème d'optimisation bi-niveaux). Ce travail tire sa motivation de modèles de marchés de l'électricité issus de la théorie des jeux non coopératifs. Dans de tels marchés, un régulateur, appelé ISO (Independant System Operator), gère le clearing et les flux d'électricité entre zones d'enchère. Nous avons développé cette analyse variationnelle selon différents axes. Tout d'abord et sur un modèle spécifique, nous avons analysé de façon exhaustive la notion de meilleure réponse d'un producteur, grâce à la détermination d'une formule explicite pour l'unique solution du problème de bas niveau de l'ISO. Puis, pour un modèle plus général de marché, la stabilité des points M(ordukhovitch)-stationnaires a été étudiée via la notion de codérivée limiting de second ordre des opérateurs multivoques. En fin, le concept d'opérateur normal limiting a été introduit et des règles de calcul ont été obtenues, fournissant ainsi un nouvel outil performant pour l'analyse quasiconvexe. L'idée de base a été l'utilisation, pour des cônes normaux à des sous-niveaux d'une fonction, de la construction limiting classiqueen analyse variationnelle moderne. Cette approche est motivée par l'hypothèse de la quasiconvexité des fonctions de coût généralement faite dans de nombreux jeux non-coopératifs. / This thesis is focused on nonsmooth variational analysis of equilibrium problems with equilibrium constraints. Such an eff ort is directly motivated by a model of electricity markets encountered in non-cooperative game theory. In such a marketthere is the so-called Independent System Operator (ISO), a regulator entity that manages the market clearing and the electricity dispatch. This market structure makes the problem of electricity markets challenging from the mathematicalpoint of view. In this area, we discovered several possibilities for further development. First, the best responses of producers in a speci fic variant of a model are fully analysed. This progress was due to an analytical formula for a uniquesolution to the lower level ISO problem. Then, for a more general model of the market, stability of the so-called M(ordukhovich)-stationarity points is provided based on the concept of coderivatives. To this end, the respective second order limiting coderivative was computed. Finally, the concept of limiting normal operator is proposed, a new tool for quasiconvex analysis exhibiting workable calculus rules. The basic idea is to employ the same limiting construction that is used in modern variational analysis in connection with normal cones to sets. This topic is motivated by the classical assumption in many non-cooperative games where the loss function of players is often assumed to be quasiconvex.

Die vorliegende Arbeit beschaeftigt sich mit Gleichgewichtsproblemen unter Gleichgewichtsrestriktionen, sogenannten EPECs (Englisch: Equilibrium Problems with Equilibrium Constraints). Konkret handelt es sich um gekoppelte Zwei-Ebenen-Optimierungsprobleme, bei denen Nash- Gleichgewichte fuer die Entscheidungen der oberen Ebene gesucht sind. Ein Ziel der Arbeit besteht in der Formulierung dualer Stationaritaetsbedingungen zu solchen Problemen. Als Anwendung wird ein oligopolistisches Wettbewerbsmodell fuer Strommaerkte betrachtet. Zur Gewinnung qualitativer Hypothesen ueber die Struktur der betrachteten Modelle (z.B. Inaktivitaet bestimmter Marktteilnehmer) aber auch fuer moegliche numerische Zugaenge ist es wesentlich, EPEC-Loesungen explizit bezueglich der Eingangsdaten des Problems zu formulieren. Der Weg dorthin erfordert eine Strukturanalyse der involvierten Optimierungsprobleme (constraint qualifications, Regularitaet), die Herleitung von Stabilitaetsresultaten bestimmter mengenwertiger Abbildungen und die Nutzung von Transformationsformeln fuer die sogenannte Ko-Ableitung. Weitere Schwerpunkte befassen sich mit der Beziehung zwischen verschiedenen dualen Stationaritaetstypen (S- und M-Stationaritaet) sowie mit stochastischen Erweiterungen der betrachteten Problemklasse, sogenannten SEPECs. / This thesis is concerned with equilibrium problems with equilibrium constraints or EPECs. Concretely, we consider models composed by coupling together two-level optimization problems, the upper-level solutions to which are non-cooperative (Nash-Cournot) equilibria. One of the main goals of the thesis involves the formulation of dual stationarity conditions to EPECs. A model of oligopolistic competition for electricity markets is considered as an application. In order to profit from qualitative hypotheses concerning the structure of the considered models, e.g., inactivity of certain market participants at equilibrium, as well as to provide conditions useful for numerical procedures, the ablilty to formulate EPEC solutions in relation to the input data of the problem is of considerable importance. The way to do this requires a structural analysis of the involved optimization problems, e.g., constraints qualifications, regularity; the derivation of stability results for certain multivalued mappings, and the usage of transformation formulae for so-called coderivatives. Further important topics address the relationship between various dual stationarity types, e.g., S- and M-stationarity, as well as the extension of the considered problem classes to a stochastic setting, i.e., stochastic EPECs or SEPECs.

Title: Hierarchical Problems with Evolutionary Equilibrium Constraints Author: Lukáš Adam Supervisor: Prof. Jiří Outrata Abstract: In the presented thesis, we are interested in hierarchical models with evolutionary equilibrium constraints. Such models arise naturally when a time-dependent problem is to be controlled or if parameters in such a model are to be identified. We intend to discretize the problem and solve it on the basis of the so-called implicit programming approach. This technique requires knowledge of a generalized derivative of the solution mapping which assigns the state variable to the control variable/parameter. The computation of this generalized derivative amounts equivalently to the computation of (limiting) normal cone to the graph of the solution mapping. In the first part we summarize known techniques for computation of the normal cone to the set which can be represented as a finite union of convex polyhedra. Then we propose a new approach based on the so-called normally admissible stratification and simplify the obtained formulas for the case of time-dependent problems. The theoretical results are then applied first to deriving a criterion for the sensitivity analysis of the solution mapping and then to the solution of two practically motivated problems. The first one concerns optimal...