Multiplayer & Asymmetric Games
Beyond pairwise interactions — public goods, multiplayer dynamics, and bimatrix games
A Brief History
Beyond Pairwise Interactions
The two-player games of Evolutionary Game Theory provide powerful intuition, but they omit a central feature of biological reality: many interactions involve multiple participants. A bacterium producing antibiotics affects an entire community. A predator pack hunts cooperatively with many individuals. Immune cells coordinate through cytokine signalling. These many-player or multiplayer games fundamentally reshape the dynamics of evolution.
Why do multiplayer games matter? First, the payoff landscape becomes richer. In a two-player game, a strategy's success depends on one opponent's strategy. In an $n$-player game, success can depend nonlinearly on the number of cooperators—introducing threshold effects, saturation, and synergies absent in pairwise games. Second, a single strategy can have different fitness depending on the composition of the rest of the group—creating subtle frequency-dependent selection. Third, multiple stable equilibria become possible: where two-player games typically admit at most one interior equilibrium, multiplayer games can exhibit dozens.
Gokhale and Traulsen (2014) provided a foundational review of this landscape in Evolutionary Multiplayer Games (Dynamic Games and Applications, 4, 468–488). They showed how public goods games, production dilemmas, and other multiplayer setups challenge the simplicity of Hawks and Doves. The replicator equation generalises, but the algebra becomes intricate: we must sum over all possible group compositions, weight by binomial coefficients, and track payoff differences across different alliance sizes.
The Public Goods Game
The simplest and most widely studied multiplayer game is the Public Goods Game (PGG). Imagine a group of $n$ individuals, each with an initial endowment. Each individual can contribute (cooperate) an amount $c$ to a common pool, or keep their endowment (defect). The pool is then multiplied by a factor $r$ (the multiplication factor) and divided equally among all $n$ members, regardless of their contribution.
Consider a group of $n$ individuals. Let $j$ be the number of cooperators. Each cooperator contributes $c$ to the pool; each defector contributes $0$. The pool total is $j \cdot c$, multiplied by factor $r$, yielding $r \cdot c \cdot j$. This is then split equally among all $n$ members:
Cooperator's payoff: $\pi_C = \frac{r \cdot c \cdot j}{n} - c$
Defector's payoff: $\pi_D = \frac{r \cdot c \cdot j}{n}$
A cooperator loses the cost $c$ but gains a share of the multiplied pool. A defector loses nothing and still gains from others' contributions.
The public goods game exhibits a profound tragedy of the commons. A defector always earns more than a cooperator (since $\pi_D > \pi_C$ for any group composition with $j < n$). Yet if everyone defects, the pool is empty and all earn $0$. If everyone cooperates, each earns $\frac{r \cdot c \cdot n}{n} - c = c(r - 1) > 0$ (assuming $r > 1$). Thus, mutual cooperation is collectively optimal, but individually irrational.
A crucial feature of the PGG is the role of the multiplication factor $r$. If $r$ is close to $1$, cooperation is barely worth it. If $r$ is large, cooperation yields enormous returns—but the incentive to free-ride remains unchanged. When can cooperation resist invasion? In a fully cooperative group of $n$, each cooperator earns $c(r - 1)$. A lone defector in that group earns $r c (n-1)/n$, free-riding on the remaining $n-1$ cooperators. For cooperation to resist invasion we need:
$$ c(r - 1) > \frac{r\,c\,(n-1)}{n} \implies \frac{r}{n} > 1 \implies r > n $$The multiplication factor must exceed the group size for cooperation to be stable against rare defectors. In a two-player game ($n = 2$) the threshold is $r > 2$; in a group of five it is $r > 5$. Larger groups therefore make cooperation harder to sustain—a quantitative expression of the tragedy of the commons. Below this threshold defection dominates, and the population spirals toward universal defection despite the collective loss.
Nonlinear Multiplayer Games
The public goods game assumes linear payoff scaling: each cooperator contributes equally, and the benefit is split equally. But real biological systems often show nonlinearity. A predator pack's hunting success may rise superlinearly with group size (synergistic effect). A biofilm's antibiotic tolerance may saturate with the number of producers (diminishing returns). These nonlinearities fundamentally alter the game's structure.
Consider a game where each individual plays against $d-1$ other individuals (so the group has size $d$). Let $k$ be the number of co-players who cooperate (ranging from $0$ to $d-1$). If the focal player cooperates, their payoff depends on $k$:
$\pi_C(k) = a_k$
If the focal player defects:
$\pi_D(k) = b_k$
The payoff matrix is specified by the sequences $\{a_0, a_1, \ldots, a_{d-1}\}$ and $\{b_0, b_1, \ldots, b_{d-1}\}$.
The key insight of Gokhale and Traulsen (2010), published in Evolutionary games in the multiverse (PNAS, 107, 5500–5504), is that nonlinear payoffs allow for multiple internal equilibria. In a two-player game with two strategies, the replicator equation is one-dimensional, and an interior equilibrium $x^*$ (if it exists) is generically unique. But in a $d$-player game, the replicator equation becomes higher-dimensional, and the fitness landscape can have many plateaus and valleys.
Specifically, in a $d$-player game with two strategies, the replicator equation can exhibit up to $\mathbf{d - 1}$ internal equilibria, compared to at most $1$ for two-player games. This multiplicity arises because fitness differences $(a_k - b_k)$ can change sign multiple times as $k$ varies from $0$ to $d-1$. Such changes create "waves" in the fitness landscape, each crossing giving rise to a potential equilibrium.
Biologically, this richness translates to intricate coexistence patterns. Imagine a cooperative bacterium and a cheat. At low frequencies of cooperators, the benefit from cooperation is weak, and cheats thrive. At high frequencies, cooperation yields strong returns, but then rare cheats invade. This can lead to cycling, chaotic dynamics, or multiple stable equilibria—depending on the precise payoff matrix.
Replicator Dynamics for d-Player Games
To understand how strategy frequencies evolve in a multiplayer setting, we must generalise the replicator equation. Consider a well-mixed population with two strategies. The focal individual interacts with $d-1$ other individuals drawn randomly from the population. The probability that exactly $k$ of these $d-1$ others are cooperators is binomial:
$$ P(k) = \binom{d-1}{k} x^k (1-x)^{d-1-k} $$The expected fitness of a cooperator is:
$$ f_C = \sum_{k=0}^{d-1} \binom{d-1}{k} x^k (1-x)^{d-1-k} a_k $$Similarly, the expected fitness of a defector is:
$$ f_D = \sum_{k=0}^{d-1} \binom{d-1}{k} x^k (1-x)^{d-1-k} b_k $$The replicator equation generalises to:
$$\dot{x} = x(1-x)\sum_{k=0}^{d-1}\binom{d-1}{k}x^k(1-x)^{d-1-k}(a_k - b_k)$$
The sum is a weighted difference in payoffs, where the weights are Bernstein polynomials, the basis functions for Bézier curves and other smooth approximations. Each term represents a possible group composition, weighted by its probability.
This equation reveals a beautiful mathematical structure. Writing $\dot{x} = x(1-x)\,P_{d-1}(x)$, the factor $P_{d-1}$ is a polynomial of degree $d-1$ in $x$ (a linear combination of Bernstein basis polynomials). For two-player games ($d=2$), $P_1$ is linear, giving at most one interior root. For larger groups the degree increases, allowing for richer dynamics.
The equilibria $x=0$ and $x=1$ are always present (enforced by the $x(1-x)$ factor). Interior equilibria are roots of $P_{d-1}$ in $(0,1)$: a polynomial of degree $d-1$ can have up to $d-1$ such roots—hence the potential for many coexisting equilibria mentioned earlier.
Reference: Gokhale, C. S. & Traulsen, A. (2010). Evolutionary games in the multiverse. Proc. Natl. Acad. Sci. USA, 107, 5500–5504.
Explore: Nonlinear d-Player Games
Asymmetric (Bimatrix) Games
Many biological conflicts are inherently asymmetric: the players occupy different roles. Think of a property owner defending a resource against an intruder, males versus females with different reproductive strategies, or an infected cell versus an immune cell. In these cases a single payoff matrix is insufficient—we need a bimatrix game, where each role has its own payoff matrix.
The defining feature of a bimatrix game is that the two roles are described by two separate matrices. Suppose each role has two strategies. We write:
| $\mathbf{A}$ — Payoffs to Role I | ||
|---|---|---|
| Role II: $S_1$ | Role II: $S_2$ | |
| Role I: $S_1$ | $a_{11}$ | $a_{12}$ |
| Role I: $S_2$ | $a_{21}$ | $a_{22}$ |
| $\mathbf{B}$ — Payoffs to Role II | ||
|---|---|---|
| Role II: $S_1$ | Role II: $S_2$ | |
| Role I: $S_1$ | $b_{11}$ | $b_{12}$ |
| Role I: $S_2$ | $b_{21}$ | $b_{22}$ |
When Role I plays $S_i$ and Role II plays $S_j$, Role I receives $a_{ij}$ and Role II receives $b_{ij}$. If $\mathbf{A} = \mathbf{B}^{\!\top}$ the game is symmetric and a single matrix suffices; when $\mathbf{A} \neq \mathbf{B}^{\!\top}$ the game is genuinely asymmetric and the two roles face different selective pressures.
The dynamics become two-dimensional. Let $x$ be the frequency of $S_1$ among Role I individuals and $y$ the frequency of $S_1$ among Role II individuals. The expected payoffs to each strategy in each role are:
$$ f_I(S_1) = a_{11}\,y + a_{12}(1-y), \quad f_I(S_2) = a_{21}\,y + a_{22}(1-y) $$ $$ f_{II}(S_1) = b_{11}\,x + b_{21}(1-x), \quad f_{II}(S_2) = b_{12}\,x + b_{22}(1-x) $$and the two coupled replicator equations are:
$$ \dot{x} = x(1-x)\bigl[f_I(S_1) - f_I(S_2)\bigr] $$ $$ \dot{y} = y(1-y)\bigl[f_{II}(S_1) - f_{II}(S_2)\bigr] $$Crucially, the fitness of each role depends on the other role's composition, coupling the two equations. The Nash equilibrium concept generalises: we seek a pair $(x^*,y^*)$ such that neither role benefits from unilateral deviation. Interior equilibria, where both roles mix, require both fitness differences to vanish simultaneously—a condition that is generically harder to satisfy than in the symmetric case.
An owner occupies a resource of value $V$; an intruder arrives. Both can Fight or Retreat. If both fight, each pays cost $C$ and wins $V$ with probability $\frac{1}{2}$. The bimatrix is:
$\mathbf{A}_{\text{owner}} = \begin{pmatrix} \frac{V-C}{2} & V \\ 0 & 0 \end{pmatrix}, \qquad \mathbf{B}_{\text{intruder}} = \begin{pmatrix} \frac{V-C}{2} & 0 \\ V & 0 \end{pmatrix}$
When $C > V$ the unique Nash equilibrium in pure strategies is owner fights, intruder retreats—the Bourgeois strategy. This relies on roles being distinguishable; if identities were unknown the game would reduce to the symmetric Hawk–Dove.
Reference: Selten, R. (1980). A Note on Evolutionarily Stable Strategies in Asymmetric Animal Conflicts. Journal of Theoretical Biology, 84, 93–101.
Mutualism and the Red King Effect
Mutualism—cooperation between individuals of different species—is an inherently asymmetric game: the two partners come from separate populations with different evolutionary rates, generation times, and effective population sizes. This makes it a natural application of bimatrix game theory. Bergstrom and Lachmann (2003) asked a deceptively simple question: when two mutualist species coevolve, which partner captures the larger share of the surplus?
Their model frames mutualism as a Nash demand game between two species. Each species evolves strategies that determine how much of the mutualistic surplus it demands. The two populations follow coupled replicator dynamics, but with different evolutionary rates reflecting differences in mutation rate, generation time, or population size. The key finding is counterintuitive: the slower-evolving species captures the larger share of the surplus.
The mechanism is not about passivity being rewarded, but about commitment power. A slowly evolving species cannot quickly adjust its demands—its evolutionary inertia acts as a credible commitment to its current bargaining position. The faster-evolving partner, able to adapt rapidly, converges toward strategies that accommodate the slower species' demands. In the language of bargaining theory, slow evolution "ties one's hands," and a player who cannot easily back down extracts more from the negotiation. This is the same logic that makes it advantageous, in a game of chicken, to visibly throw away the steering wheel.
Bergstrom and Lachmann named this the Red King effect, after the Red King in Lewis Carroll's Through the Looking-Glass. The Red King sleeps motionless throughout the entire story while all other pieces move around him—yet Tweedledee claims the whole world is the Red King's dream. The analogy is precise: the slow-evolving species sits still while its partner scrambles to adapt, yet the immobile species ends up with the better deal. This contrasts with the Red Queen effect from the same book, where the Red Queen must run constantly just to stay in place—the standard metaphor for antagonistic coevolution (host–parasite arms races).
Does the Red King effect hold in multiplayer mutualistic interactions? Gokhale and Traulsen (2012) revisited this question in Mutualism and evolutionary multiplayer games: revisiting the Red King (Proc. R. Soc. B, 279, 4611–4616). They showed that when mutualism involves more than two species, the commitment advantage of slow evolution erodes. In a pairwise interaction the faster species has no choice but to accommodate its single partner; in a multiplayer setting the faster species can redirect its adaptive effort toward other partners, weakening the slower species' bargaining leverage. The Red King effect can reverse entirely: the faster-evolving partner may capture more of the surplus.
Pairwise mutualism (2 species): The slower-evolving partner's evolutionary inertia acts as a credible commitment, forcing the faster partner to concede a larger share of the surplus. Slow evolution is strategically advantageous.
Multiplayer mutualism (3+ species): The commitment advantage of slowness diminishes because the faster species can redirect adaptation toward alternative partners. In the multiplayer regime the Red King becomes a Red Queen: all partners must keep evolving or risk being outcompeted.
This transition from Red King to Red Queen dynamics illustrates a profound lesson: the structure of interactions—whether pairwise or multiplayer—is not a technical detail. It fundamentally reshapes the coevolutionary landscape and can reverse long-held biological principles.
Finite Populations: d-Player Moran Process
Just as Evolutionary Game Theory extended two-player games to finite populations via the Moran process, we can extend multiplayer games. In a finite population of size $N$, we model $d$-player games by randomly selecting $d$ individuals at each "interaction step" and computing their payoffs. Individuals then reproduce according to fitness: high-fitness individuals are more likely to have offspring, which replace random members of the population.
The key modifications are:
- Group sampling: A focal individual's payoff is computed by averaging over random groups of $d-1$ co-players drawn from the remaining $N-1$ individuals. In a finite population this follows a hypergeometric distribution (not binomial), adding combinatorial complexity.
- Fitness heterogeneity: In a finite population, different individuals can have very different numbers of interactions, creating sampling variance.
- Fixation probability: The probability that a rare mutant takes over is no longer the simple formula $\phi = (1 - 1/w) / (1 - 1/w^N)$ from two-player games. In multiplayer settings, fixation probabilities can differ dramatically depending on group size and payoff structure.
Wu, Traulsen and Gokhale (2013) provided a comprehensive treatment in Dynamic Properties of Evolutionary Multi-player Games in Finite Populations (Games, 4, 182–199). They showed that:
- For "synergistic" games where group size $d$ is large, small populations face strong drift, and the fixation threshold is correspondingly relaxed.
- For games with strong internal equilibria (multiple stable states), finite populations can be trapped in low-fitness equilibria due to drift.
- The relationship between infinite-population (deterministic) equilibria and finite-population fixation probabilities is non-trivial: an equilibrium with high relative fitness in the infinite-size limit may have low fixation probability in small groups.
These results underscore that finitude is not a minor perturbation—it reshapes the selective landscape and can reverse predictions from deterministic theory.
Exercises
Conceptual Questions
- What is the key difference between 2-player and $n$-player games? Why is the payoff per player often nonlinear in $n$ (e.g., public goods games)?
- In an asymmetric game, what is the difference between a strategy that is an ESS in the symmetric game and one that is an ESS when players have different roles?
- Explain the "Red King Hypothesis" or "Red King Effect" in evolutionary dynamics. Why might the slowest runner sometimes win in a coevolutionary race?
- In the $n$-player Prisoner's Dilemma (public goods game), why is defection favoured at the individual level even though cooperation maximises group payoff? What mechanisms can align individual and group interest?
- How does population finitude (Moran process) change the outcome of evolutionary games compared to infinite-population predictions? Why can drift allow dominated strategies to invade?
Computer Problems
- Public Goods Game with Variable Group Size. Implement the public goods game where $n$ individuals each receive endowment $e$, can contribute $0 \leq c_i \leq e$, and receive equal share of $r\sum c_i$. Compute payoff $\pi_i = e - c_i + r\sum c_i / n$. For different group sizes ($n = 2, 5, 10$), show how the optimal group contribution changes and why individual incentive leads to free-riding.
- Asymmetric Hawk-Dove with Roles. Implement an asymmetric game where an "owner" (defender of resource) and "intruder" (challenger) have different payoffs. Derive the ESS where owner plays Hawk with frequency $p$ and intruder with frequency $q$. Show how role asymmetry stabilises different equilibria than the symmetric game.
- Voluntary Contributions Mechanism ($n$-player Snowdrift). Model $n$ players with continuous strategies $s_i \in [0,1]$ contributing to a common pool. Payoff is $\pi_i = b(1 - \prod(1-s_j)) - c s_i$ (at least one must contribute for benefit). Use replicator dynamics to find ESS. Show how $n$ affects equilibrium contribution level.
- Finite Population Moran Process with Asymmetry. Implement a finite Moran process where individuals play an asymmetric public goods game (e.g., some players are "helpers," others are "free-riders"). Include mutation. Show how small population size allows temporary invasion of dominated strategies and how drift can reverse infinite-population predictions.
- Reward and Punishment in Multiplayer Games. Extend the public goods game to include optional punishment ($p_i$) and reward ($r_i$) of others. Compute equilibrium punishment levels and show how they depend on cost of punishment, group size, and baseline parameters. Demonstrate the "paradox of punishment" where it sustains cooperation even if costly.
References
- Gokhale, C. S. & Traulsen, A. (2014). Evolutionary Multiplayer Games: An Introduction. Dynamic Games and Applications, 4, 468–488.
- Gokhale, C. S. & Traulsen, A. (2010). Evolutionary games in the multiverse. Proc. Natl. Acad. Sci. USA, 107, 5500–5504.
- Hardin, G. (1968). The Tragedy of the Commons. Science, 162, 1243–1248.
- Perc, M., Jordan, J. J., Rand, D. G., Wang, Z., Boccaletti, S. & Szolnoki, A. (2017). Statistical physics of human cooperation. Physics Reports, 687, 1–51.
- Selten, R. (1980). A Note on Evolutionarily Stable Strategies in Asymmetric Animal Conflicts. Journal of Theoretical Biology, 84, 93–101.
- Bergstrom, C. T. & Lachmann, M. (2003). The Red King effect: when the slowest runner wins the evolutionary race. Proc. Natl. Acad. Sci. USA, 100, 593–598.
- Gokhale, C. S. & Traulsen, A. (2012). Mutualism and evolutionary multiplayer games: revisiting the Red King. Proc. R. Soc. B, 279, 4611–4616.
- Wu, B., Traulsen, A. & Gokhale, C. S. (2013). Dynamic Properties of Evolutionary Multi-player Games in Finite Populations. Games, 4, 182–199.
- Szabó, G. & Tőke, C. (1998). Evolutionary prisoner's dilemma game on a square lattice. Phys. Rev. E, 58, 69.
- Nowak, M. A. (2006). Evolutionary Dynamics: Exploring the Equations of Life. Harvard University Press.
- Santos, F. C. & Pacheco, J. M. (2005). Scale-free networks provide a unifying framework for the emergence of cooperation. Phys. Rev. Lett., 95, 098104.
- Hofbauer, J. & Sigmund, K. (1998). Evolutionary Games and Population Dynamics. Cambridge University Press.