CoopetitiveV: Theory and Applications

• CoopetitiveV is a framework that quantifies both cooperative and competitive interactions in multi-agent, economic, and game-theoretic environments.
• It employs indices such as Banzhaf, Uniform–Shapley, and Shapley–Owen to measure coalition dynamics and incentive structures within varying game settings.
• The approach extends to practical applications including multi-agent learning, online auctions, network economies, and dynamical systems to drive actionable insights.

CoopetitiveV encompasses the theoretical, algorithmic, and empirical study of systems and indices that quantify, facilitate, or exploit the dual presence of cooperation and competition—termed "coopetition"—within multi-agent, economic, game-theoretic, and networked environments. This concept has seen rigorous mathematical formalization, particularly through indices designed to measure the cooperative/competitive tendencies of coalitions, the incentive structures in strategic interactions, and the dynamical properties of coopetitive systems.

1. Formal Definition and Index Construction

At the core of CoopetitiveV lies the coopetition index, now formulated for general monotone transferable utility (TU) games as well as non-transferable simple games. For a TU-game $(N, v)$ and any nonempty coalition $S \subseteq N$, define two sets of probabilities: $p_S$ over non-trivial two-partitions of $S$, and $q_S$ over subsets $T \subseteq N \setminus S$ (external coalitions). The coopetition index is then

$\C^v_{p,q}(S) = \sum_{T \subseteq N \setminus S} q_S(T) \, \A^v_p(S, T)$

where the attitude function $\A^v_p(S, T)$ is: $\A^v_p(S, T) = \begin{cases} v(S \cup T) - v(T) & \text{if } |S| = 1, \ v(S \cup T) - v(T) - \sum_{\{S_1, S_2\} \in \Pi_2(S)} p_S(\{S_1, S_2\}) [v(S_1 \cup T) + v(S_2 \cup T) - 2v(T)] & \text{if } |S| \geq 2. \end{cases}$ The absolute coopetition index $\overline\C^v_{p,q}(S)$ normalizes $\C^v_{p,q}(S)$ via the generalized semivalue $\Phi^v_q(S)$, yielding a universal range in $[-1, 1]$: $\overline\C^v_{p,q}(S) = \begin{cases} \frac{\C^v_{p,q}(S)}{\Phi^v_q(S)}, & \Phi^v_q(S) \neq 0, \ 0, & \Phi^v_q(S) = 0. \end{cases}$ where

$$\Phi^v_q(S) = \sum_{T \subseteq N \setminus S} q_S(T)[v(S \cup T) - v(T)].$$

This construction subsumes and generalizes the relative and absolute indices for simple games, providing a precise, normed measure of group-level competition and cooperation in both binary (simple) and scalar (TU) settings (Aleandri et al., 19 Nov 2025, Aleandri et al., 2024).

2. Instantiations: Banzhaf, Uniform Shapley, and Shapley–Owen Indices

Three canonical indices arise from specific choices of $(p, q)$:

• Banzhaf coopetition: uniform internal ($p^u_S$) and external ($q^u_S$) weights; coincides with the Banzhaf value on singletons.
• Uniform–Shapley coopetition: uniform internal, Shapley-style permutation external distribution; reduces to the classical Shapley value on singletons and satisfies a group version of the Shapley axioms.
• Shapley–Owen coopetition: permutation-based weights for both internal and external splits; recovers Owen value on singletons.

Each version possesses explicit computational formulas and normalization properties, with the absolute index admitting clear interpretation and comparison across coalitions, including singletons. The endpoints $+1$/$-1$ correspond to maximal internal cooperation/competition, while $0$ can represent either powerlessness or perfect balance of antagonistic substructures (Aleandri et al., 19 Nov 2025, Aleandri et al., 2024).

3. Axiomatic Characterizations

Uniform–Shapley and Shapley–Owen coopetition indices are each the unique group indices satisfying:

• Linearity: Additive in the game-form.
• Symmetry on unanimity games: For $v(S)=1$ iff $S=N$.
• External null-player neutrality: Removing a null player external to $S$ leaves the index unchanged.
• Internal contraction: Adding an internal null player scales the index by a contraction factor specific to $(p, q)$.

For Uniform–Shapley, the contraction uses the uniform partitioning, for Shapley–Owen the permutation-based distribution. These axiomatic results unify coopetitive indices with classical group value theory in cooperative game theory (Aleandri et al., 19 Nov 2025).

4. Dynamical, Algorithmic, and Game-Theoretic Contexts

CoopetitiveV is not restricted to static coalition games. Its analytical machinery and conceptual framework extend to several domains:

• Dynamic Systems: In nonlinear dynamics, "eventually cooperative" or "eventually competitive" flows are those preserving order after a finite transient. The non-oscillation principle and three-dimensional Poincaré–Bendixson theorems guarantee that for such systems, complex limit behavior (e.g., chaotic attractors) is ruled out and only simple structures (equilibria or cycles) are possible. This underpins the theoretical tractability of broad "coopetitive" dynamical models arising in biology, economics, and networked control (Niu et al., 2018).
• Multi-agent Learning: In Markov games with mixed-team objectives, CoopetitiveV is operationalized via centralized-decentralized value estimation, as in QMIX, MAVEN, and QVMix. Population-based training against multiple evolving strategies consistently yields more robust (higher Elo) coopetitive teams than single-opponent or pure self-play approaches (Leroy et al., 2022).
• Online Decision-making: Multi-player bandit problems with a linear cooperation parameter $\lambda \in [-1,1]$ establish a spectrum from full competition to joint cooperation, affecting both exploration thresholds and aggregate rewards (Brânzei et al., 2019).
• Network Economy Models: In business networks (e.g., blockchain-enabled collaborations), CoopetitiveV is instantiated as a joint optimization (total network value maximization) combined with competitive bargaining over surplus division, formalizing both joint-creation and competitive division layers in a unified equilibrium (Wasserkrug et al., 2020).

5. Applications and Model Examples

The CoopetitiveV formalism admits immediate application:

• Voting and Apex Games: In simple games, e.g., the apex game, the coopetitive index can distinguish between powerlessness ($V=0$) and true balance of antagonistic forces by comparing to the decisiveness index ($\mathcal{D}$), which quantifies the magnitude (not sign) of internal tensions (Aleandri et al., 2024).
• Ad Auctions: When advertising benefits are distributed across overlapping coalitions, first-price auctions with cooperative envy-freeness yield strong welfare and revenue guarantees, outperforming benefit-aware VCG mechanisms prone to zero-revenue collusion—a direct operationalization of coopetitive mechanism design (Hoy et al., 2012).
• Polymatrix Games: Algorithmically, CoopetitiveV underpins manipulation policies in polymatrix settings, with provably small costs: agents can enforce desired coalition structures or dominance via low-cost LP-based modifications to payoff matrices, facilitating practical "coopetitive" control without explicit negotiation (Mahesh et al., 2021).
• Data Sharing under Competition: In data-driven markets with dominant external competitors (e.g., "Amazon"), optimal mediator thresholds for full vs. partial data sharing can be exactly or approximately characterized using CoopetitiveV-style analysis. Outside competition typically lowers the sharing threshold, but regime shifts may reverse this effect, depending on joint identifiability and information structure (Gradwohl et al., 2020).

6. Interpretive Insights and Comparative Analysis

Key properties and implications across settings include:

• The coopetitive index $\overline\C^v_{p,q}(S)$ quantifies the net pull toward cohesion vs. antagonism among coalition members, normalized for external marginal power and internal split. This enables principled, objective comparison of cohesion across coalitions, game structures, and application domains (Aleandri et al., 19 Nov 2025).
• The indices uniquely satisfy monotonicity under minimal-winning splits, invariance under relabeling, and reduce to classical (singleton) value measures in degenerate cases.
• In dynamical and algorithmic environments, the presence of explicit or latent coopetition shapes learning dynamics, equilibrium selection, and welfare outcomes in ways not captured by purely cooperative or competitive frameworks.

7. Role in Broader Multi-agent and Game-theoretic Research

CoopetitiveV provides a rigorous, generalizable language for articulating and analyzing the pervasive duality of cooperation and competition at both micro (coalition, strategy profile) and macro (dynamical, equilibrium, or population-level) scales. It encompasses static indices, learning frameworks, policy design, and structural stability theory, positioning coopetitive analysis as a central tool in the theoretical and applied study of complex multi-agent interactions (Aleandri et al., 19 Nov 2025, Aleandri et al., 2024, Niu et al., 2018, Wasserkrug et al., 2020, Brânzei et al., 2019, Leroy et al., 2022).