GNEP in Banach Space Equilibrium Analysis

• GNEP in Banach space is defined as an equilibrium system where each agent’s strategies and constraints depend on other agents in infinite-dimensional settings.
• Variational reformulations using quasi-variational inequalities and KKM methods enable existence proofs under conditions like lower semicontinuity and graph convexity.
• Applications in optimal control and PDE-constrained optimization illustrate the practical impact of these techniques on multi-agent system analysis.

A generalized Nash equilibrium problem (GNEP) in Banach space arises when several agents (players) compete in a system where both the objective functions and the feasible sets for each agent depend on the strategies chosen by all players. The infinite-dimensional setting notably arises in optimal control, PDE-constrained optimization, and game-theoretic models in functional spaces. Banach-space GNEPs generalize classical finite-dimensional GNEPs by incorporating strategies and constraints in arbitrarily complex, infinite-dimensional vector spaces, and require specialized analytic and variational tools for study.

1. Formal Definition and Variational Formulation

Given a finite set of players $A = \{1,2,\ldots,N\}$, each player $v$ is equipped with a real Banach space $X_v$ and a closed, convex, nonempty private constraint set $C_v \subset X_v$. The product space $X := X_1 \times \cdots \times X_N$ and admissible set $C := C_1 \times \cdots \times C_N$ structure the joint feasible profile $x = (x_1, \ldots, x_N) \in C$. Each player's available strategies are further restricted by a set-valued constraint function $K_v: C \to 2^{C_v}$, potentially depending on all rivals' choices, and preferences are encoded either

• by cost functionals $f_v: X \to \mathbb{R}$,
• or by preference correspondences $P_v: C \to 2^{C_v}$ specifying strict-preference sets.

A profile $x^* \in C$ solves the GNEP if for each $v \in A$,

• $x^*_v \in K_v(x^*_{-v})$,
• $x^*_v$ is at least as preferred as any $z_v \in P_v(x^*) \cap K_v(x^*_{-v})$.

Convex GNEPs require that the feasible sets $K_v(x_{-v})$ are convex and cost functionals $f_v(\cdot,\, x_{-v})$ are convex for each fixed $x_{-v}$.

Variational analysis recasts the equilibrium search as a quasi-variational inequality (QVI) problem: find $x^*\in K(x^*)$ and $x^* \in F(x^*)$ such that

$$\langle x^\ast, y - x^\ast \rangle \ge 0, \quad \forall y \in K(x^*)$$

for a constructed principal operator $F$ capturing the normal cone structure of the preferences or gradients (Sultana et al., 2024).

2. Constraint Structures and Regularity Conditions

The analytic properties of the constraint maps $K_v$ are fundamentally important for existence and uniqueness results.

• Lower semicontinuity (LSC): A set-valued map $F: U \to 2^Y$ is lower semicontinuous at $u_0$ if for every $y_0 \in F(u_0)$ and neighborhood $V$ of $y_0$, there is a neighborhood $U_0$ of $u_0$ such that $F(u) \cap V \neq \emptyset$ for $u \in U_0$. LSC of $K_v$ underpins upper semicontinuity of best-response maps, and is standard in Kakutani-type fixed-point proofs. However, LSC can be difficult to verify in infinite-dimensional function spaces (Bongarti et al., 14 Dec 2025).
• Graph convexity: $K_v$ is graph-convex if the set $\{(x_{-v},x_v) : x_v \in K_v(x_{-v})\}$ is convex in $X_{-v} \times X_v$. Graph convexity is purely geometric and easier to check in situations such as PDE state-constrained games.
• KKM property: $K: X \to 2^X$ is a KKM-map if for any finite set $\{u_1,\ldots,u_m\} \subset X$, $$\mathrm{co}\{u_1, \ldots, u_m\} \subset \bigcup_{i=1}^m K(u_i)$$. The KKM property enables the application of the Knaster-Kuratowski-Mazurkiewicz lemma, leading to intersection results for constraint maps without relying on semicontinuity.

The following table summarizes these notions:

Property	Definition	Utility in GNEP Existence
Lower semicontinuity	For all $y_0\in F(u_0)$, open $V$ of $y_0$, $\exists$ neighborhood $U_0$ s.t. $F(u)\cap V\neq \emptyset$ for $u\in U_0$	Enables fixed-point existence via Kakutani
Graph-convexity	Graph of $F$ is convex in $X\times Y$	Suits PDE/open convex games
KKM property	Convex hull of any finite subset is covered by values	Fixed-point alternative to LSC

3. Existence Theorems: Analytic and Geometric Criteria

Existence of GNEs in Banach spaces can be established under several sets of hypotheses:

• Classical (LSC): Compactness, convexity, LSC of $K_v$, and continuity plus convexity of $f_v$ yield existence of GNEs via an upper semicontinuous best-response correspondence and Kakutani’s theorem (Bongarti et al., 14 Dec 2025).
• Graph-convexity: If $K_v$ are convex-valued, have closed graph, and are graph-convex, existence follows from arguments based on the Nikaido–Isoda–Fan function, convexity of feasible profiles, and fixed-point methods.
• KKM-based: If $K$ is KKM with closed convex values and each $X_i^{ad}$ is convex, compact, then a GNE exists via the KKM lemma.

For games whose equilibrium structure is captured by preference correspondences, similar existence results are established if $K_v$ are LSC and $P_v$ have open graph, convex values, and appropriate irreflexivity/closedness properties (Bongarti et al., 14 Dec 2025).

4. Variational Analytic Characterizations

GNEPs are reformulated as QVIs via the construction of a principal operator:

• For each player $v$, the normal-cone operator to preferences $N_{P_v}(x)$ is defined by

$$N_{P_v}(x) = \{ x_v^* \in X_v^* : \langle x_v^*, y_v - x_v \rangle \leq 0, \ \forall y_v \in P_v(x) \}$$

and the total operator $F(x) := F_1(x) \times \cdots \times F_N(x)$ is built from selections with norm-weak* upper semicontinuity, convexity, and weak* compactness (Sultana et al., 2024).

• The equilibrium profile $x^*$ solves the QVI if it belongs to the product constraint $K(x^*)$ and, for some $x^* \in F(x^*)$, satisfies the variational inequality against all feasible $y$.
• For convex $f_v$ with numerical representation, these reduce to the KKT-based VI system underpinning classical GNEP theory (Facchinei–Kanzow).

5. Uniqueness and Diagonal Strict Monotonicity

Uniqueness of variational equilibria is ensured under diagonal strict monotonicity conditions, extending Rosen’s finite-dimensional theory:

• For given positive weights $r = (r_1, \ldots, r_N)$ and Gâteaux-differentiable $f_i$, the pseudogradient operator $d(x,r)$ is constructed as

$$d(x,r)(h) = \sum_{i=1}^N r_i \langle \partial_i f_i(x), h_i \rangle_{X_i^*, X_i}$$

Diagonal strict monotonicity requires

$$d(x,r)(y-x) + d(y, r)(x-y) < 0 \quad \forall x\neq y \in C$$

This yields uniqueness of the variational equilibrium in the shared constraint case, and allows controlled selection of equilibria via manipulation of the multipliers $r$ (termed "multiplier bias") (Bongarti et al., 14 Dec 2025).

6. Geometric Preference-Based Approach

The use of strict-preference correspondences $P_i: X \to 2^{X_i}$—rather than cost functionals—permits a geometric, order-theoretic framing of equilibrium theory:

• Each $P_i$ satisfies open-graph, convexity, and topological closedness properties.
• The geometric generalized Nash equilibrium is $x^* \in K(x^*)$ such that $P_i(x^*) \cap K_i(x^*_{-i}) = \emptyset$ for all $i$.
• Existence results parallel the analytic case, provided $K_i$ is LSC and $X_i^{ad}$ is compact. Variational and fixed-point arguments adapt to the purely combinatorial structure afforded by the KKM property or convexity of the graph (Bongarti et al., 14 Dec 2025).

7. Special Cases and Applications

• In finite dimensions, selections for the principal operator $F_v(x)$ recover classical VI operators.
• If preference correspondences admit a utility representation and are complete and transitive, QVI formulations coincide with the classical variational inequality characterization (Sultana et al., 2024).
• Graph-convexity and KKM approaches are particularly advantageous in PDE-constrained or function-space games, where LSC is typically unavailable but convexity of feasible sets is immediate.
• The structural results unify earlier finite-dimensional existence theorems and clarify the analytic role of regularity versus geometry of set-valued maps in infinite-dimensional equilibrium analysis (Bongarti et al., 14 Dec 2025).

These frameworks collectively enable the rigorous study of multi-agent optimal control, PDE-constraint interaction, and other infinite-dimensional competition models under broad and weak regularity conditions.