Binary Coordination Games

• Binary coordination games are strategic interactions on networks where agents select one of two actions, aiming for payoff gains by matching neighbors' choices.
• They exhibit a tractable equilibrium structure under supermodular conditions, enabling convergence to consensus profiles and efficient welfare or potential maximization in pure coordination settings.
• Research highlights sharp complexity contrasts with anti-coordination games and investigates rich dynamical phenomena through various learning dynamics and network-based extensions.

A binary coordination game is a strategic interaction, typically defined on a network, in which each agent chooses one of two available actions and derives utility from matching the choices of neighbors or co-players. Such games model social conventions, technology adoption, task allocation, and agreement formation, and are a canonical instance of games with strategic complements. Binary coordination games are notable for their tractable equilibrium structure, rich dynamical phenomena, and sharp complexity-theoretic dichotomies versus anti-coordination settings.

1. Formal Model and Equilibrium Structure

A binary coordination game is defined by a graph $G = (V, E)$, where each node (agent) $i \in V$ selects $a_i \in \{0, 1\}$. The payoff to agent $i$ is typically a function increasing in the agreement between $a_i$ and the actions of its neighbors. In the simplest symmetric case, the utility can be written as

$$U_i(a_i, a_{-i}) = \sum_{j \in N(i)} w_{ij} \cdot \mathbf{1}\{ a_i = a_j \} + b_i a_i$$

with nonnegative weights $w_{ij}$ encoding the strength of coordination, and $b_i$ an individual bias term (Vanelli et al., 14 May 2025). The profile $a = (a_1, \dots, a_n)$ is a Nash equilibrium if no agent can increase their utility by unilaterally deviating. When $w_{ij} \geq 0$ for all $i, j$, the game is supermodular and possesses a lattice of equilibria; best-response dynamics converge to extremal consensus profiles (Vanelli et al., 14 May 2025).

Generalizations allow for signed weights and mixed supermodular/submodular structures, where existence and selection of consensus (all-0 or all-1) or polarization equilibria depends on core cohesiveness and structural balance properties of the underlying graph (Vanelli et al., 14 May 2025).

2. Solution Concepts, Social Welfare, and Complexity

Analysis of binary coordination games frequently centers on three optimization objectives for a given instance (network, payoffs):

• Welfare maximization: Find a profile maximizing total sum of payoffs $W(a) = \sum_i U_i(a)$.
• Potential maximization: Find $a$ maximizing the game's potential function $\Phi(a)$.
• Best Nash equilibrium: Find a welfare-maximizing $a$ that is also a Nash equilibrium.

In pure coordination games, welfare and potential maximization reduce to submodular or graph-cut-type problems and are solvable in polynomial time via s-t min-cut algorithms (Deligkas et al., 2023, Deligkas et al., 2023). For anti-coordination games, these problems become NP-hard due to their equivalence to Max-Cut (Deligkas et al., 2023). However, even in pure coordination, finding the optimal Nash equilibrium is NP-hard, via reductions from hypergraph transversal (Deligkas et al., 2023, Deligkas et al., 2023):

Objective	Pure Coordination	Anti-Coordination
Welfare maximization	P (min-cut)	NP-hard
Potential maximization	P (min-cut)	NP-hard / PLS-cpt
Best Nash equilibrium (max-welfare)	NP-hard	NP-hard

This reveals a dichotomy: environments of strategic complements—coordination—are algorithmically friendlier than those of substitutes.

3. Equilibrium Selection, Learning Dynamics, and Metastability

Equilibrium selection in binary coordination games depends on the update rule, noise, and network structure. Various processes have been studied:

• Best response / replicator dynamics: These favor risk-dominant equilibria independent of local effects (Raducha et al., 2021, Auletta et al., 2011).
• Unconditional imitation: Coordination on the payoff-dominant equilibrium occurs only below a critical network connectivity—a phenomenon absent on complete graphs (Raducha et al., 2021).
• Aspiration learning: With fading-memory aspiration levels and occasional trembles, the stationary distribution concentrates arbitrarily on efficient coordination profiles, achieving both efficiency and fairness in symmetric games (Chasparis et al., 2011).
• Logit (Glauber) dynamics: Stationary distributions concentrate on risk-dominant equilibria as noise vanishes. For large networks and moderate noise, long-lived metastable states emerge, with timescales exponential in system size for cliques (Auletta et al., 2011). The behavioral regime alternates between rapid coalescence into one "phase" and exponentially slow mixing across them.

On bipartite graphs, there is a formal isomorphism: binary coordination and anti-coordination games have identical mixing and hitting times under symmetric stochastic decision rules (Jones et al., 2021), so neither problem is generically easier.

4. Network Structure, Heterogeneity, and Equilibrium Fuzziness

Network topology plays a critical role. On large, finely connected graphs, the set of possible equilibrium averages is sharply constrained. "Fuzzy conventions" arise when agents are subject to persistent idiosyncratic random utility shocks; equilibria generically feature positive fractions of agents playing both actions (Pęski, 2021). The stable equilibrium outcome set in the mean field (complete graph) is the interval of stable fixed points of the best-response map, but on lattices, convergence is to a single "RU-dominant" fraction maximizing a suitable robustness integral—generalizing classical risk-dominance to random-utility settings (Pęski, 2021). Complete-graph copies (cliques) realize the largest possible set of equilibrium averages, while lattices select the minimal, unique convention.

Effects of network connectivity interact with agent rationality (modeled as inverse temperature $\beta$ in log-linear learning). Higher connectivity allows for lower rationality while still ensuring high-probability convergence to the global potential maximizer—a "wisdom of crowds" effect (Zhang et al., 2023).

5. Extensions: Preference Heterogeneity, Multilayer Games, and Complex Constraints

Extensions of binary coordination games capture heterogeneous preference and multilayer effects:

• Discrete preference games: Players balance individual strategy preferences and neighbor agreement. The social optimum is efficiently computable by min-cut for two strategies. The price of stability (PoS) is 1 for $\alpha \leq 1/2$ (weight on intrinsic preference) or $\alpha=2/3$, but can approach 2 as $\alpha \to 1$ outside these regimes (Chierichetti et al., 2013).
• Multilayer networks: Agents embedded in two interacting layers, each with opposite preferences, can experience synchronization transitions. In such systems, imitative dynamics (replicator, unconditional imitation) robustly select the Pareto-optimal equilibrium over a wide parameter range, in contrast to best-response, which preserves risk-dominance symmetry (Raducha et al., 2022).
• Signed network coordination games: When network edges are allowed positive (coordination) or negative (anti-coordination) weights, robust consensus or polarization equilibria arise if a large enough supermodular "core" is present (Vanelli et al., 14 May 2025).

6. Binary Constraint System Games and Quantum Extensions

Binary coordination is central to the theory of binary constraint system (BCS) games, which are nonlocal games with binary variables subject to linear (e.g., parity) constraints. A solution consists of coordinated local assignments respecting consistency across overlapping variables. Classical satisfiability reduces to (un)coordinated solution; quantum strategies (entangled assignments) admit sought-after "quantum satisfying assignments" obeying specific operator product and commutation relations. The existence of such assignments is equivalent to perfect quantum strategies, but their decidability is not known to be in NP and is open in general. Strict upper bounds on quantum success probability for some BCS games are established via commutation algebra arguments (Cleve et al., 2012).

7. Implications and Applications

The mathematical tractability and sharp structural characterizations of binary coordination games underpin a wide range of applications, including distributed task allocation, opinion dynamics, technology adoption, and the design of efficient protocols for decentralized decision-making. The differential complexity profile vis-à-vis anti-coordination and the network-theoretic wisdom-of-crowds effect underscore the importance of interaction topology and agent updating mechanisms. These findings have direct implications for mechanism design, network formation, algorithmic game theory, and the study of social conventions in large-scale systems (Deligkas et al., 2023, Pęski, 2021, Raducha et al., 2022, Zhang et al., 2023, Vanelli et al., 14 May 2025).