Statistical physics of human cooperation

Matjaz Perc; Jillian J. Jordan; David G. Rand; Zhen Wang; Stefano Boccaletti; Attila Szolnoki

Source-linked AI summary

Statistical physics of human cooperation

Matjaz Perc, Jillian J. Jordan, David G. Rand, Zhen Wang, Stefano Boccaletti, Attila Szolnoki

arXiv:1705.07161v1May 19, 2017physics.soc-phcond-mat.stat-mechcs.SInlin.AOq-bio.PE

TL;DR

Human cooperation is extensive and consequential but remains difficult to explain, particularly when interactions involve groups and multiple possible states. This review synthesizes experiments and theoretical models using statistical-physics approaches to study cooperation, incentives, spatial organization, and dynamics. It reports mechanisms and patterns that can promote prosocial outcomes, while noting that an integrated framework combining the models remains lacking.

Problem
Statistical physics has produced insights into cooperation, but an integrative framework combining its diverse mathematical models remains lacking.
Method
The review synthesizes human experiments and mathematical models of public-goods cooperation, incentives, reciprocity, tolerance, and structured populations using statistical-physics methods.
Results
The review identifies strategies, mechanisms, and external factors associated with prosocial outcomes, including cyclic dominance, self-organized incentives, and structured-population dynamics.
Takeaways & Limitations
Statistical physics has led to a better understanding of which cooperation-promoting mechanisms operate in competitive settings and why.

Abstract

from arXiv · show

Extensive cooperation among unrelated individuals is unique to humans, who often sacrifice personal benefits for the common good and work together to achieve what they are unable to execute alone. The evolutionary success of our species is indeed due, to a large degree, to our unparalleled other-regarding abilities. Yet, a comprehensive understanding of human cooperation remains a formidable challenge. Recent research in social science indicates that it is important to focus on the collective behavior that emerges as the result of the interactions among individuals, groups, and even societies. Non-equilibrium statistical physics, in particular Monte Carlo methods and the theory of collective behavior of interacting particles near phase transition points, has proven to be very valuable for understanding counterintuitive evolutionary outcomes. By studying models of human cooperation as classical spin models, a physicist can draw on familiar settings from statistical physics. However, unlike pairwise interactions among particles that typically govern solid-state physics systems, interactions among humans often involve group interactions, and they also involve a larger number of possible states even for the most simplified description of reality. The complexity of solutions therefore often surpasses that observed in physical systems. Here we review experimental and theoretical research that advances our understanding of human cooperation, focusing on spatial pattern formation, on the spatiotemporal dynamics of observed solutions, and on self-organization that may either promote or hinder socially favorable states.

I. INTRODUCTION

Human cooperation is evolutionarily consequential yet remains difficult to understand, especially because it involves interactions among individuals and groups. This review applies non-equilibrium statistical-physics methods to models of cooperation, emphasizing spatial patterns, spatiotemporal dynamics, and self-organization.

Human cooperation: Extensive cooperation among unrelated individuals became central to human evolutionary success, although the evidence for its specific origins remains scarce and circumstantial.Proposed drivers include alloparental care, provisioning, and conflicts between groups.
Human cooperation: Modern societies combine large-scale cooperation and technological progress with severe failures to meet basic needs for millions of people.The review frames this contrast as motivating better understanding of cooperation and its broader consequences.
The role of statistical physics: Statistical physics offers tools for studying cooperation as collective behavior in structured populations, including network reciprocity, pattern formation, and phase transitions.Cooperator clusters can protect interior members from exploitation by defectors.
The role of statistical physics: Human cooperation can be supported or undermined by positive incentives, negative incentives, tolerance, and social norms, while incentive provision itself creates second-order freeriding.Rewards benefit prosocial behavior, whereas punishment targets free-riding; both impose costs on those providing them.
The role of statistical physics: The review combines human experiments with mathematical models, using public-goods extensions and non-equilibrium methods such as Monte Carlo simulations and phase-transition theory.The models address punishment, rewarding, reciprocity, tolerant players, and structured-population stability.

A. Goals of human experiments

Human-cooperation experiments use incentivized economic games to measure prosocial behavior and test how interaction structure, repetition, observability, and culture affect cooperation. Across games, cooperation is sensitive to strategic incentives and often declines over repeated public-goods play.

Goals of human experiments: Experiments test what empirically occurs, complementing theoretical models that describe what is theoretically possible.Researchers use human subjects and economic games to test predictions about cooperation.
Goals of human experiments: Dictator games measure anonymous generosity, while ultimatum games add responder rejection power and thereby alter the strategic structure of giving.In the ultimatum game, rejection causes both players to earn nothing.
Goals of human experiments: Repeated prisoner’s-dilemma play supports more cooperation than one-shot play, with cooperation increasing as the continuation probability rises.Repeated interaction can make cooperation self-interested when the probability of another round is sufficiently high.
Cooperation in groups: Public-goods games model positive-sum group cooperation while making contribution individually costly, so contributing nothing is payoff-maximizing regardless of others’ contributions.Contributions are multiplied and divided equally among group members.
Cooperation in groups: Around 50% is the typical average contribution in one-shot public-goods games, but repeated play commonly erodes cooperation as conditional cooperators switch toward defection.Responses are often bimodal, with many participants contributing nothing or everything.
Cooperation in groups: Across 16 developed societies, first-period contributions ranged from approximately 70% in Copenhagen to approximately 40% in Athens, while repeated-game decline was consistent across cultures.The result motivates experiments across diverse cultures rather than relying primarily on WEIRD populations.

2. Cooperation in dyads

Experiments use dyadic games to measure prosociality, trust, and punishment, revealing substantial cooperation alongside culturally variable generosity and costly responses to unfairness.

The dictator game measures generosity toward strangers by asking anonymous participants to share an endowment in a one-shot interaction.
Dictator-game sharing varies across cultures, including a single mode at 10 percent among Hadza hunter-gatherers rather than the canonical bimodal pattern.
Trust-game participants show substantial trust and trustworthiness, with larger returns when trustors send more and greater trust toward previously cooperative trustees.
Punishment opportunities increase public-goods contributions from the first round, suggesting deterrence operates before defectors personally experience punishment.
Punishment experiments distinguish retaliation from norm enforcement through ultimatum and third-party punishment games.
Third-party punishment occurs across cultures, declines as offers rise toward half the stake, and is almost absent for hyper-generous offers.

4. Rewarding

Experiments and models examine rewarding, network structure, and the gap between laboratory incentives and real-world cooperation. Their findings show that social mechanisms can sustain cooperation, but experimental success does not by itself establish real-world operation.

4. Rewarding: People preferentially reward contributors, and repeated rewarding can sustain cooperation within groups.
Network effects: Fixed networks stabilize cooperation only when cooperative neighbors satisfy the conditions required by the relevant theoretical mechanism.
Network effects: Dynamic networks promote cooperation by allowing individuals to form or maintain connections with cooperators and avoid or break connections with defectors.
Experiment-model divide: Laboratory tests that recreate model incentives can show that a mechanism promotes cooperation when implemented without showing that it operates outside the laboratory.
Experiment-model divide: Evidence less dependent on laboratory payoff maximization includes reciprocal behavior without financial incentive and intuitive responses undermined by deliberation.
Experiment-model divide: Third-party punishment is associated with greater trust and costly returns, supporting its interpretation as a signal of trustworthiness.

A. Public goods game as the null model

The public goods game models cooperation and defection in groups, where individual incentives conflict with collective benefits. Spatial structure changes cooperation thresholds relative to well-mixed populations and motivates extensions involving punishment and other mechanisms.

Cooperators contribute c = 1 to a common pool, defectors contribute nothing, and the multiplied total is divided among group members.The model generalizes the pairwise prisoner’s dilemma to group interactions through the synergy factor r > 1.
For r < G, defectors earn more than cooperators, creating a social dilemma in which cooperation benefits the group but defection maximizes short-term individual payoff.If nobody cooperates, public goods vanish, producing the tragedy of the commons.
At r = G in a well-mixed population, the system crosses between universal defection and universal cooperation.Under payoff-based imitation, r < G leads to all defection, whereas r > G leads to population-wide cooperation.
On a square lattice, players interact in overlapping groups of size G = 5 with their four nearest neighbors under periodic boundary conditions.Each player’s overall payoff sums payoffs from all groups in which they participate.
Cooperators survive on the square lattice only if r > 3.74 and eliminate defectors completely for r > 5.49, with both phase transitions continuous.The lattice benchmark shows how network reciprocity supports cooperation below the well-mixed r = G limit.
The framework extends the null model by adding peer punishers, who cooperate while fining defectors and paying punishment costs.Punishment parameters are normalized by the number of other players in each group to facilitate comparisons across networks and group sizes.

2. Pool punishment

Pool punishment models institutionalized sanctioning by adding cooperators who fund a common punishment pool. The review contrasts this mechanism with peer punishment and with rewarding-based institutions.

2. Pool punishment: Pool punishment is institutionalized sanctioning in which punishers contribute to a common pool that covers punishment institutions independently of their necessity or efficiency.This differs from peer punishment, where costs are paid only when defectors are identified and sanctioned.
2. Pool punishment: Pool punishers O contribute c = 1 like traditional cooperators C, while defectors contribute nothing to the public good.The multiplied contributions are distributed equally among group members before punishment is applied.
2. Pool punishment: The review treats rewarding as positive reciprocity that costs the provider while conferring a benefit on another player.Peer rewarding targets cooperators individually, whereas institutionalized rewarding uses a common pool.
2. Pool punishment: Pool rewarding is described as much rarer than pool punishment and mainly associated with particular subcultures rather than official law-enforcing institutions.The model therefore includes both prosocial and antisocial rewarding schemes.
2. Pool punishment: Rewarding cooperators contribute c = 1 to a prosocial pool, whose total is multiplied by r2 > 1 and distributed among rewarding cooperators.Rewarding defectors operate an analogous antisocial pool under the same synergy factor.
2. Pool punishment: The pool-rewarding model uses equal contributions c = 1 and the same multiplication factor r2 for prosocial and antisocial pools.This strategy-neutral parametrization avoids assigning either form of rewarding an obvious advantage.

3. Self-organized rewarding

Self-organized rewarding allows cooperation-supporting incentives to adapt over time rather than remain fixed. The model specifies how rewards, costs, and activity levels enter payoffs while framing reciprocity as a broader evolutionary problem.

3. Self-organized rewarding: Self-organized rewarding adds an activity parameter µx that lets rewarding efforts adapt according to the success of cooperation.This extends models in which reward costs and intensities remain fixed after initialization.
3. Self-organized rewarding: Each cooperator or rewarding cooperator receives µi∆/(G −1) from every rewarding cooperator in the same group, excluding self-rewarding.Reward providers pay µiα∆/(G −1) for every cooperator rewarded.
3. Self-organized rewarding: The parameter α is the ratio between the cost of rewarding and the reward allotted to cooperators.Together with ∆, it controls the cost and incremental step of rewarding activity.
3. Self-organized rewarding: Economic experiments and everyday observations challenge strong reciprocity by suggesting that people often reward cooperation or punish defection, but seldom do both.Statistical-physics methods are presented as tools for studying this “stick versus carrot” dilemma.
3. Self-organized rewarding: The correlated-reciprocity model compares defectors with cooperators who punish, reward, or both, while all cooperative strategies contribute c = 1.The payoff equations assign sanctions to defectors and rewards to cooperative strategies according to their neighborhood composition.

E. Tolerance

Tolerance is modeled by players who cooperate when defections remain below an individual threshold and abstain otherwise. The broader simulation framework uses spatial interactions, payoff-based imitation, and prepared states to assess stationary solutions.

E. Tolerance: Tolerant players Mi join cooperation or abstain as loners depending on the number of defectors in their group.The tolerance level i determines how many defectors they accept before refusing cooperation.
E. Tolerance: All cooperative strategies contribute c = 1, while defectors and loners contribute nothing; public goods are shared among non-loners.The synergy factor r applies only when at least two contributions enter the common pool, and loners receive σ = 1.
E. Tolerance: The tolerance model assigns tolerant players a payoff that combines the cooperator payoff or secure loner payoff with a punishment cost.The competing payoffs are represented by ΠC = ΠD −1, ΠL = σ, and ΠMi = δiΠC + (1 −δi)σ −γ.
E. Tolerance: A simplified related model uses one tolerant-player type at a time, with the tolerated-defector threshold denoted H.The full model allows all tolerance levels i = 0, …, G −1.
E. Tolerance: Monte Carlo simulations update strategies randomly and sequentially, enabling comparisons with generalized mean-field approximations and determination of phase transitions.The procedure is described for a square lattice but can be adapted to other interaction networks.
E. Tolerance: Each update selects a player and a random nearest neighbor, computes both payoffs across overlapping groups, and lets the neighbor imitate the player via a Fermi probability.The uncertainty parameter K controls how strongly payoff differences influence adoption.
E. Tolerance: Unlike imitation, the logit rule evaluates the payoff of changing strategy directly and can generate new strategies spontaneously.The review focuses primarily on imitation dynamics because logit updating has been less studied for public goods games.
E. Tolerance: Random initial states may fail to relax to the most stable stationary solution, so specially prepared initial conditions are often needed to verify stability.In the pool-punishment example, the D + C + O phase is the correct stationary state rather than the O phase reached from a random state.

B. Random initial conditions

Random initial conditions give strategies comparable starting representation but cannot ensure that every stable multi-strategy subsystem solution emerges. Large systems and subsequent prepared-state analyses are therefore needed to assess stability and phase behavior reliably.

B. Random initial conditions: Stable solutions may consist of two-, three-, or four-strategy alliances rather than single strategies.Some strategies survive only within these subsystem solutions, which must form before competing with other solutions.
B. Random initial conditions: Random initial conditions are useful because the stable subsystem solutions are initially unknown and may require a sufficiently large population to emerge.The recommended approach is to use large systems first, then identify subsystem solutions for further analysis.
B. Random initial conditions: Prepared initial states are more efficient and fair for comparing identified subsystem solutions than continuing with random initial conditions.They provide equal survival chances when the relevant subsystem solutions are already known.
B. Random initial conditions: Random initial conditions are unnatural for many social processes, which often begin locally among like-minded individuals, but remain useful for discovering unknown solutions.This creates a methodological tension between discovery and realism.
B. Random initial conditions: Continuous transitions in cooperation models can be characterized by spreading-process critical exponents, with directed percolation reported as the predominant universality class.The cited exponent is ζ = 0.584(4).
B. Random initial conditions: Discontinuous transitions can arise from indirect territorial competition or spontaneous cyclic dominance, including cases where average strategy fractions remain far from zero.The cyclic-dominance phase can terminate discontinuously because fluctuations diverge near the critical control value.

D. Stability of subsystem solutions

With three or more competing strategies, many subsystem solutions can coexist, so stationary states and phase boundaries require systematic stability analysis. Reliable results depend on adequate system size, prepared interfaces, and sufficient relaxation.

D. Stability of subsystem solutions: Monte Carlo studies with three or more strategies require sufficiently large systems and long relaxation times to avoid misleading stationary outcomes.Finite-size and relaxation effects can prevent relevant strategies or phases from emerging.
D. Stability of subsystem solutions: Improper simulation conditions can produce one- or two-strategy states that are unstable to invasion by mutant groups.The review also notes that quenched heterogeneities can generate long-lived patches and obscure the true stationary state.
D. Stability of subsystem solutions: All subsystem solutions of a multi-strategy game are also solutions of the whole system, making stability analysis necessary for accurately locating phase transitions.The analysis must account for the large number of possible subsystem states.
D. Stability of subsystem solutions: The winner between two subsystem solutions is determined by the average invasion-front velocity after each solution has developed its proper composition and spatiotemporal structure.Interfaces should be opened only after the competing solutions reach their stationary dynamics.
D. Stability of subsystem solutions: In the tolerant-player example, the D + C + L and D + C + M1 + M2 phases are individually stable but separated by a discontinuous transition.The dominant phase changes as the multiplication factor r increases slightly.
D. Stability of subsystem solutions: Finite-size effects can suppress necessary strategies or destabilize cyclic phases before their characteristic stationary patterns emerge.Even prepared-state analyses require sufficiently large lattice regions for each subsystem solution.

A. Peer punishment and indirect territorial competition

Peer punishment and rewarding can promote cooperation through spatial organization, but they generate distinct competitive dynamics. Punishment can produce territorial replacement and discontinuous transitions, whereas rewarding can create cyclic coexistence and nonmonotonic outcomes.

A. Peer punishment and indirect territorial competition: Peer punishers can survive despite lower payoffs against defectors because cooperators and punishers form compact clusters that compete separately against defectors.Punishers ultimately win this indirect territorial competition in the illustrated dynamics.
A. Peer punishment and indirect territorial competition: A sufficiently cheap punishment cost favors peer punishers over cooperators, while excessive cost favors cooperators, producing a discontinuous switch between C + D and C + P phases.The phase diagram confirms this cost-to-fine dependence.
A. Peer punishment and indirect territorial competition: The transition from C + D is discontinuous, whereas the D + P → P transition is continuous and agrees with the directed-percolation universality class conjecture.The order parameter for the latter transition is the defector fraction ρD.
A. Peer punishment and indirect territorial competition: Peer punishment can produce very slow delayed replacement of cooperators, hysteresis, and opposite effects on defectors depending on whether the system lies below or above a critical point.Punishment may stabilize cooperation below the critical point but stabilize defectors above it.
B. Peer rewarding and the emergence of cyclic dominance: Rewarding produces a sequence from D to D + R, R, D + C + R, and finally C + R as reward increases at moderate cost.The three-strategy phase is governed by cyclic dominance, while C and R can coexist stably because they remain nonequivalent without defectors.
B. Peer rewarding and the emergence of cyclic dominance: In the D + C + R phase, defectors invade traditional cooperators, rewarding cooperators defeat defectors, and traditional cooperators spread within rewarding-cooperator regions.These interactions form a closed loop of dominance that sustains the three-strategy dynamics.
B. Peer rewarding and the emergence of cyclic dominance: Rewarding is viable but produces counterintuitive outcomes, including cases where moderate rewards promote cooperation more effectively than high rewards.The review concludes that these results do not settle the stick-versus-carrot dilemma.

C. Correlated strategies and an exotic first-order phase transition

Correlated positive and negative reciprocity produces complex phase behavior dominated by discontinuous transitions and cyclic dominance. The correlated strategy survives only in narrow, unrealistic parameter regions.

C. Correlated strategies and an exotic first-order phase transition: The correlated reciprocity model reveals phase behavior dominated by discontinuous transitions caused by spontaneous cyclic dominance among D, P, and B.The cycle is D outperforming P, P outperforming B, and B outperforming D.
C. Correlated strategies and an exotic first-order phase transition: Around β = 0.37, increasing γ causes P and B fractions to decay until P vanishes, interrupting the dominance cycle and producing the D(B) phase.
C. Correlated strategies and an exotic first-order phase transition: At β = 0.55, the D + P + B phase terminates when oscillation amplitudes increase with γ and drive the system into a uniform absorbing phase.This mechanism persists independently of system size rather than arising as a finite-size effect.
C. Correlated strategies and an exotic first-order phase transition: χ becomes size-independent and diverges as γ approaches γc = 0.1242(6), forcing a discontinuous transition from D + P + B to D(B).Within D(B), either pure D or pure B can result depending on which strategy disappears first.
C. Correlated strategies and an exotic first-order phase transition: Despite indirect territorial competition and cyclic dominance, the correlated strategy survives only in very narrow and unrealistic parameter regions.
C. Correlated strategies and an exotic first-order phase transition: The reviewed institutionalized-punishment dynamics include complex phase transitions, subsystem stability, and sustainability constraints involving second-order freeriders.Pool punishment remains sustainable only when second-order freeriders are sanctioned strongly enough to prevent their prevalence.

B. The non-existent institutionalized rewarding

Institutionalized rewarding is less extensively studied than punishment, but its effectiveness depends on interaction heterogeneity, synergy, and adaptive sanctioning dynamics. Adaptive punishment can restore smooth interfaces and enhance cooperation through network reciprocity.

B. The non-existent institutionalized rewarding: Pool rewarding is less common in human societies than pool punishment, and dedicated statistical-physics research on it remains limited.
B. The non-existent institutionalized rewarding: When group synergies are weak, equal rewards maximize cooperation, whereas large multiplication factors favor rewarding high-degree nodes more than low-degree nodes.
B. The non-existent institutionalized rewarding: Adaptive sanctioning lets players adjust punishment according to neighboring defectors’ success, explaining spontaneous punishment and deterrence with globally negligible investment.
B. The non-existent institutionalized rewarding: Adaptive punishment promotes cooperation by enhancing network reciprocity, preventing cyclic dominance, or providing advantages to those sanctioning antisocial behavior.
B. The non-existent institutionalized rewarding: In phase diagrams, cheap punishment can restore coexistence or full cooperation, while increasing punishment costs shifts cooperative survival and dominance toward larger synergy factors.
B. The non-existent institutionalized rewarding: Adaptive punishers spontaneously restore smooth interfaces after defector invasions, disabling further invasion and eventually allowing punishment to cease in the pure cooperative phase.

ADAPTIVE PUNISHMENT

Adaptive and probabilistic punishment reshape spatial interfaces and distribute sanctioning effort. Cooperation is promoted when punishment is localized or shared, but extreme punishment propensities favor defection.

ADAPTIVE PUNISHMENT: Adaptive punishment preserves smooth cooperator–defector interfaces, strengthening network reciprocity against invasion.
ADAPTIVE PUNISHMENT: Probabilistic sanctioning assigns cooperators a probability p of being selected as peer punishers, distributing responsibility rather than relying on permanent punishers.
ADAPTIVE PUNISHMENT: Sharing sanctioning responsibility can make punishment successful even when sanctioning costs substantially exceed the fines imposed on defectors.
ADAPTIVE PUNISHMENT: Cooperation increases with the multiplication factor r, while the punishment fine α and punishment probability p require careful adjustment.
ADAPTIVE PUNISHMENT: Intermediate punishment propensity maintains compact, smooth cooperative clusters, whereas p = 0 and p = 1 produce more frequent defector invasions and widespread defection.
ADAPTIVE PUNISHMENT: Cooperators defeat defectors only in the presence of punishers, indicating that the counterintuitive outcome requires multi-point interaction.
ADAPTIVE PUNISHMENT: The combined effect is summarized as two weaker strategies forming a stronger one, paralleling Parrondo’s paradox.

C. Evolutionary advantages of adaptive rewarding

Adaptive rewarding creates several evolutionary advantages but has a narrower route to defector-free cooperation than adaptive punishment. Its aggressive invasion can undermine the compact spatial structure needed for network reciprocity.

C. Evolutionary advantages of adaptive rewarding: Adaptive rewarding promotes collaboration through self-organized rewarding, indirect territorial competition, and spontaneous cyclic dominance.
C. Evolutionary advantages of adaptive rewarding: At r = 3.5, discontinuous phase transitions are absent, while sufficiently costly rewarding leaves defectors as the only surviving strategy.
C. Evolutionary advantages of adaptive rewarding: Adaptive rewarding can produce a defector-free R(C) state when rewarding is efficient enough, with logarithmically slow voter-model-like coarsening.
C. Evolutionary advantages of adaptive rewarding: Second-order freeriders survive only in D + C + R, requiring high ∆, intermediate α, and sufficiently high r.
C. Evolutionary advantages of adaptive rewarding: At small synergy factors, second-order freeriding requires a fine balance of the remaining parameters.
C. Evolutionary advantages of adaptive rewarding: Adaptive punishment eliminates defectors at lower required ∆ than adaptive rewarding because punishers maintain compact phases while rewarding cooperators advance faster but fail to create defector-free states.
C. Evolutionary advantages of adaptive rewarding: Allowing antisocial punishment reduces cooperation to approximately the level observed without punishment, whereas traditional punishment alone yields more cooperative outcomes.

B. Lack of adverse effects with antisocial rewarding

Antisocial rewarding does not undermine cooperation when prosocial rewarding is also available. Increasing pool-reward effectiveness strengthens cooperative aggregation and can produce cooperative dominance despite equally strong antisocial rewards.

Antisocial rewarding does not deter public cooperation as long as cooperators can also distribute rewards.
At r2 = 1, cooperators survive only above the critical multiplication factor r1c > 3.74, below the well-mixed threshold G = 5 because of network reciprocity.
Increasing r2 steadily lowers the r1 threshold for rewarding cooperators to survive and for rewarding defectors to dominate.
At r2 = 1.3, larger cooperative clusters and more stable interfaces emerge because pool rewarding reinforces aggregation despite equally strong antisocial and prosocial rewards.
At sufficiently high r2, the cooperative-defector interface becomes impenetrable, allowing cooperators to spread until a pure RC phase remains.

B. Diverse tolerance levels

Tolerance and tolerance diversity reshape public-goods outcomes through phase transitions, cyclic dominance, and competition among subsystem solutions. Intermediate tolerance and heterogeneous tolerant strategies can support cooperation, but finite populations may conceal the strongest outcomes.

With diverse tolerance levels, loners dominate at very low r, while higher r supports cyclic D+C+L dynamics before tolerant strategies become more competitive.
The diverse-tolerance phase diagram contains discontinuous transitions whose locations require stability analysis of competing subsystem solutions.
Starting from all eight strategies, the M1 + M2 subsystem can eliminate defectors and achieve maximal cooperation, outperforming the D + M2 outcome.
Finite-size effects can hide the M1 + M2 phase: its fixation probability is 1 at 6000 × 6000 but can remain invisible at 500 × 500.
These dynamics show that diversity and tolerance matter through spatiotemporal competition among subsystem solutions in structured populations.

X. SUMMARY

The review synthesizes statistical-physics approaches to human cooperation, spanning experiments, spatial public-goods models, Monte Carlo methods, phase transitions, and self-organized incentives. It reports broad mechanistic insights while identifying missing integration across models and social-science perspectives.

The review systematically covers Monte Carlo methods and phase-transition theory for pattern formation, spatiotemporal dynamics, and self-organization in cooperation.
Experiments show prosocial behavior, willingness to punish selfishness and reward cooperation, and sensitivity to whether interaction networks are fixed or changing.
The spatial public-goods game serves as a null model extended with punishment, rewards, correlated reciprocity, and tolerance.
The reviewed methodology includes random sequential updating, analysis of random initial conditions, phase-transition concepts, and stability analysis of subsystem solutions.
Across model families, the review highlights mechanisms including adaptive punishment, probabilistic sharing, peer-strategy dynamics, institutionalized incentives, antisocial strategies, and tolerance.
An integrative framework merging the separate mathematical models is still lacking, and anthropology, psychology, and sociology remain insufficiently integrated.
The framework does not yet account for how differences in individuals’ goals, status, and potential losses affect cooperation or defection.
Antisocial strategies remain understudied systematically, especially antisocial punishment and stability of subsystem solutions in spatial public-goods models.