Long-range social pressure and the evolution of cooperation in multiplex networks

arXiv:2607.06056 · physics.soc-ph · Submitted 2026-07-07 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "Long-range social pressure and the evolution of cooperation in multiplex networks".

Mira: Here's the gist of the paper. Long-range social pressure and the evolution of cooperation in multiplex networks shows that extending social vigilance beyond direct neighbors substantially promotes cooperation.

Kai: This work introduces distance-resolved, multi-hop social pressure into an evolutionary-game framework, bridging the literature on structured and multilayer cooperation with studies of multi-step social influence and complex contagion.

Mira: Model Setup The model considers a population of N = 1000 agents arranged on a two-layer multiplex network with a shared node set.

Kai: Each agent carries two binary variables: a strategy Ci ∈

0, 1: (cooperator/defector) and a vigilance state Vi ∈

0, 1: (vigilant/non-vigilant).

Paper summary: Kai: So, this paper, "Long-range social pressure and the evolution of cooperation in multiplex networks," it’s looking at how extending social vigilance beyond just direct neighbors helps cooperation. They set up a population where everyone has two variables: their strategy, which is either a cooperator or a defector, and their vigilance state.

Mira: And they use this structure on a two-layer multiplex network with shared nodes. The main idea is that the social pressure you get from people further away than your immediate neighbors actually helps cooperation grow in the system.

Kai: They’re bridging the gap between studies on structured cooperation and complex contagion, specifically looking at how multi-step social influence works in an evolutionary game framework. It connects a few different areas of literature together that hadn't really been connected before.

Lev: From an engineering standpoint, they take this concept of monitoring and propagate it continuously across L circles of influence, with the weight decaying geometrically with network distance. That’s a distance-resolved, multi-hop form of social pressure that they haven't combined with an evolutionary game before.

Mira: They introduce this idea by coupling a Prisoner’s Dilemma on one layer of the network to a vigilance cascade on the other layer, where influence decays geometrically with network distance. This is how they introduce distance-resolved, multi-hop social pressure into an evolutionary game framework >

Conclusion: Kai: So, thinking about this paper, "Long-range social pressure and the evolution of cooperation in multiplex networks," it really focuses on how extending that vigilance beyond direct neighbors substantially promotes cooperation. The authors are looking at a two-layer network setup where agents have strategy and vigilance states.

Mira: What they claim is that this extension of social pressure, moving to more than one circle of influence, has a significant effect on cooperation levels across all the network topologies they study, both synthetic and real ones.

Kai: The authors show that even extending vigilance to just the second circle of influence already recovers most of the cooperative gain in sparse networks. This is interesting because it suggests the benefit isn't always massive or immediate for every network structure.

Lev: I wonder how quickly that gain accrues, though they mention it’s topology dependent. If you look at sparse Erdős–Rényi networks, extending to just the second circle of influence recovers most of what you get from one step alone. But in sparse Barabási–Albert networks, they needed a third circle before the gain felt complete >

Mira: And they also looked at how inter-layer correlation matters. They found that for sparse Erdős–Rényi networks, decorrelating those layers actually suppresses the benefit of extending that influence further out. The shared hub structure in any two realizations of the same degree distribution helps preserve some of that correlation even where it's usually absent >

Kai: So, what does this mean practically? It suggests that for social influence to work well in complex systems, you need to consider these higher-order interactions, not just the immediate neighbors. It points toward a more nuanced understanding of how social pressure builds up over time and distance >

Mará Pereda, Ãbriel Muller

Grupo de Investigaci࠹n Ingenier࠹a de Organizaci࠹n y Logistica (IOL) · Departamento de Ingenier࠹a de Organizaci࠹n, Administraci࠹n de Empresas y Estadࡁstica, Escuela Técnica Superior de Ingenieros Industriales, Universidad Politˬnica de Madrid · Grupo Interdisciplinar de Sistemas Complejos (GISC)

physics.soc-ph

Submitted: 2026-07-07

Updated: 2026-09-07

Journal ref: Phys. Rev. E 114, 054305, 2026

DOI: 10.1103/yvh4-8yxx

Code: https://github.com/mpereda/long-range-social-pressure

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 83/100

The gist: Long-range social pressure and the evolution of cooperation in multiplex networks shows that extending social vigilance beyond direct neighbors substantially promotes cooperation.

Key concepts

Multiplex Network
A network structure consisting of multiple interacting layers or groups. In this model, agents belong to a two-layer system where they have connections within one layer and shared connections with another layer, creating a richer social environment than simple single-layer networks.
Vigilance State ($V_i$)
A binary state for each agent indicating whether they are vigilant or not. This state determines how much influence an agent receives from others. Agents update their vigilance based on the level of vigilance they perceive in their environment, following a threshold rule.
Social Pressure Range ($L$ circles)
The distance or number of layers (circles) through which an agent's vigilance influence is calculated. Extending this range ($L$) beyond direct neighbors tests how far social monitoring can reach to encourage cooperative behavior.

Terminology

Summary

Long-range social pressure and the evolution of cooperation in multiplex networks shows that extending social vigilance beyond direct neighbors substantially promotes cooperation. p. 0

Where the paper says it

Long-range social pressure and the evolution of cooperation in multiplex networks Mar´ıa Pereda∗ Grupo de Investigaci´on Ingenier´ıa de Organizaci´on y Log´ıstica (IOL), Departamento de Ingenier´ıa de Organizaci´on, Administraci´on de Empresas y Estad´ıstica, Escuela T´ecnica Superior de Ingenieros Industriales, Universidad Polit´ecnica de Madrid, 28006 Madrid, Spain and Grupo Interdisciplinar de Sistemas Complejos (GISC), 28911 Legan´es, Madrid, Spain Gabrielle Muller Universidad Polit´ecnica de Madrid, 28006 Madrid, Spain (Dated: September 9, 2026) Social pressure—the awareness of being observed by others—is a fundamental driver of prosocial behavior in human societies. Yet it is typically assumed that only direct neighbors exert vigilance pressure on an individual, despite empirical evidence that social influence persists to at least three degrees of separation.

This work introduces distance-resolved, multi-hop social pressure into an evolutionary-game framework, bridging the literature on structured and multilayer cooperation with studies of multi-step social influence and complex contagion. p. 1
Where the paper says it

Our vigilance mechanism instead propagates continuously across L circles of influence, with weight decaying geometrically with network distance—a distance-resolved, multi-hop form of social pressure that, to our knowledge, has not been previously combined with an evolutionary game. Beyond extending the vigilance mechanism of [12] to multiple circles of influence, the conceptual step this model takes is to introduce distance-resolved, multi-hop social pressure into an evolutionary-game framework, bridging the literature on cooperation in structured and multilayer populations with empirical and theoretical work on multi-step social influence and complex contagion—two bodies of work that have so far developed largely independently of one another.

Model Setup

The model considers a population of N = 1000 agents arranged on a two-layer multiplex network with a shared node set. Each agent carries two binary variables: a strategy Ci ∈ [0, 1] (cooperator/defector) and a vigilance state Vi ∈ [0, 1] (vigilant/non-vigilant). p. 2

Where the paper says it

graph model. Each agent i carries two binary variables: a strategy Ci ∈ {0, 1} (cooperator/defector) and a vigilance state Vi ∈ {0, 1} (vigilant/non-vigilant).

Agents play a weak Prisoner’s Dilemma on Ggame. p. 2
Where the paper says it

Each agent i carries two binary variables: a strategy Ci ∈ {0, 1} (cooperator/defector) and a vigilance state Vi ∈ {0, 1} (vigilant/non-vigilant). Agents play a weak Prisoner’s Dilemma on Ggame.

The total payoff of agent i is πi = Pj∈∂i [CiCj ·R+Ci(1−Cj)·S + (1−Ci)Cj ·Ti + (1 − Ci)(1 − Cj)· P], where ∂i denotes the neighborhood of i in Ggame. p. 2
Where the paper says it

Payoffs are R = 1 (cooperator–cooperator), P = 0 (defector–defector), S = 0 (cooperator against defector), and Ti (defector against cooperator i), where Ti depends on the local vigilance environment of i (see below). The total payoff of agent i is πi = P j∈∂i [CiCj ·R+Ci(1−Cj )·S + (1−Ci)Cj ·Ti + (1 − Ci)(1 − Cj )· P], where ∂i denotes the neighborhood of i in Ggame.

Vigilance Mechanism

The vigilance influence received by agent i at generation t, across L circles, is Ii(t) = min 1, X L d=1 λ d−1 md i(t) k d i!, (1). The individual temptation of defecting against cooperator i at generation t is Ti(t) = 1 + (b − 1)(1 − Ii(t)), (2). p. 2

Where the paper says it

004 overall), before adopting the unnormalized kernel used throughout this paper, which instead lets range and magnitude increase together, consistent with the additive weighting used in [23]. The individual temptation of defecting against cooperator i at generation t is Ti(t) = 1 + (b − 1)(1 − Ii(t)), (2) with b ∈ [1, 2] the baseline temptation parameter.

Vigilance follows a Watts threshold rule where Vi(t + 1) = Ci(t + 1) · 1[Ii(t) ≥ θ]. p. 3
Where the paper says it

Second, λ is a single continuous parameter that smoothly interpolates between purely local social pressure (λ = 0, equivalent to L = 1) and uniform weighting across all circles (λ → 1), allowing us to study the effect of the decay rate independently of the range L (Sec. Vigilance dynamics Vigilance follows a Watts threshold rule [14], reevaluated every generation: agent i is vigilant at generation t + 1 if and only if it is a cooperator at t + 1 and the influence it received at the previous generation met or exceeded a threshold θ, Vi(t + 1) = Ci(t + 1) · 1[Ii(t) ≥ θ].

Key Findings on Range Extension

The central result is that extending vigilance beyond direct neighbors substantially promotes cooperation, though how quickly the gain accrues is itself topology dependent. In sparse Erd˝os–R´enyi networks, the L = 1 → 2 step alone already recovers most of it, while sparse Barab´asi–Albert networks need a third circle before the gain is similarly complete. p. 6

Where the paper says it

S7): in both Erd˝os–R´enyi densities, the L = 1 → 2 step alone already captures the large majority of the total gain (93–96%), and the further step to L = 3 adds only 4–8% as much again; in both Barab´asi–Albert densities the cumulative gain instead builds much more evenly across the first two additional circles (52–65% already reached at L = 2, a further increment 59–87% as large still to come at L = 3). Sparse Erd˝os–R´enyi topologies thus front-load nearly the entire benefit into the first additional circle, while Barab´asi– Albert topologies spread it across the first two, converging by L = 3 to essentially the same outcome across all four topologies (96–103% of the total gain captured).

The effect is strongest precisely where the original model is weakest—sparse networks, where direct neighbor vigilance alone struggles to sustain cooperation at high temptation. p. 0
Where the paper says it

Both shifts are topology dependent: the L = 1 → 2 step alone already accounts for most of the gain in these sparse networks, whereas the pattern is more gradual in denser, hub-dominated Barab´asi–Albert networks, consistent with the decay coefficients of social influence reported in controlled experiments. The effect is strongest in sparse networks, requires that the vigilance and game layers be aligned, and reproduces directly on a real social network of physicians; in dense, hub-dominated networks the gain instead depends sharply on how fast influence decays with distance, switching between weak and strong cooperation as the decay rate crosses a threshold.

Role of Correlation and Decay

Inter-layer correlation matters, and its effect grows with L, most prominently in sparse Erd˝os–R´enyi networks. In ER z = 4, decorrelating the layers increasingly suppresses the benefit of extending L. p. 7

Where the paper says it

5 shows, inter-layer correlation matters, and its effect grows with L, most prominently in sparse Erd˝os– R´enyi networks. In ER z = 4, decorrelating the layers increasingly suppresses the benefit of extending L: the correlated critical temptation at θ = 0.

The shared hub structure of any two realizations of the same degree distribution partially preserves the correlation that Erd˝os–R´enyi networks lack entirely. p. 7
Where the paper says it

86 for z = 16 in the nominally uncorrelated BA multiplex, compared with rk ≈ 0 for Erd˝os–R´enyi networks, whose nodes are exchangeable and carry no such arrival-order signal. The high-degree hubs that dominate vigilance propagation in BA networks thus do tend to occupy structurally similar roles in Gvig and Ggame even when the two layers share no edges—but this alignment follows substantially from the shared node-arrival order of the network generator, not from a property intrinsic to the vigilance dynamics itself, and the residual sensitivity BA does show to decorrelation, reported above, is consistent with rk falling short of 1 rather than reaching it.

Robustness and Real Networks

Testing the model’s robustness on the CKM physician network [27, 28] showed that this mechanism is not an artifact of idealized random-graph topologies. The qualitative pattern—cooperation gains concentrated at low threshold, vanishing at high threshold—reproduces on a real, sparse, historically documented social network at an order of magnitude smaller N than our synthetic networks. p. 0

Where the paper says it

Both shifts are topology dependent: the L = 1 → 2 step alone already accounts for most of the gain in these sparse networks, whereas the pattern is more gradual in denser, hub-dominated Barab´asi–Albert networks, consistent with the decay coefficients of social influence reported in controlled experiments. The effect is strongest in sparse networks, requires that the vigilance and game layers be aligned, and reproduces directly on a real social network of physicians; in dense, hub-dominated networks the gain instead depends sharply on how fast influence decays with distance, switching between weak and strong cooperation as the decay rate crosses a threshold.

The sensitivity analysis further revealed that the decay parameter λ does not act uniformly across topologies. p. 1
Where the paper says it

We adopt the Fermi update rule [3, 24]— new for this model—and validate it against the replicator dynamics of [12] at L = 1 before turning to L > 1. We then study how the vigilance range affects stationary cooperation in both correlated and uncorrelated two-layer multiplex networks built on Barab´asi–Albert and Erd˝os– R´enyi topologies (mean degree z = 4 and z = 16), assess the sensitivity of the results to the decay parameter λ, check their robustness to the network realization, population size, Fermi noise, and update scheme, and test the robustness of the main result on a real social network.

Conclusion

The extension to L circles of influence bears a conceptual connection to the emerging literature on higher-order interactions in complex systems. The results suggest that this misalignment, not merely the existence of a wide monitoring network, can blunt its cooperative benefit. p. 0

Where the paper says it

Vigilance range and alignment act jointly: extending vigilance range promotes cooperation only where the monitoring and game structures are aligned, and neither factor alone is sufficient. Our results strongly suggest that even modest expansions of social awareness—such as those enabled by online social platforms—can substantially reshape the landscape of cooperative behavior in human populations.

The model retains the original’s simplifying assumptions: vigilance and strategy are binary rather than graded, and monitoring carries no direct cost to the monitor. p. 2
Where the paper says it

graph model. Each agent i carries two binary variables: a strategy Ci ∈ {0, 1} (cooperator/defector) and a vigilance state Vi ∈ {0, 1} (vigilant/non-vigilant).

The extension to L circles of influence bears a conceptual connection to the emerging literature on higher-order interactions in complex systems [7–9]. p. 1
Where the paper says it

Our vigilance mechanism instead propagates continuously across L circles of influence, with weight decaying geometrically with network distance—a distance-resolved, multi-hop form of social pressure that, to our knowledge, has not been previously combined with an evolutionary game. Beyond extending the vigilance mechanism of [12] to multiple circles of influence, the conceptual step this model takes is to introduce distance-resolved, multi-hop social pressure into an evolutionary-game framework, bridging the literature on cooperation in structured and multilayer populations with empirical and theoretical work on multi-step social influence and complex contagion—two bodies of work that have so far developed largely independently of one another.

The paper is organized as follows. p. 1
Where the paper says it

We then study how the vigilance range affects stationary cooperation in both correlated and uncorrelated two-layer multiplex networks built on Barab´asi–Albert and Erd˝os– R´enyi topologies (mean degree z = 4 and z = 16), assess the sensitivity of the results to the decay parameter λ, check their robustness to the network realization, population size, Fermi noise, and update scheme, and test the robustness of the main result on a real social network. The paper is organized as follows.

The main finding is that extending vigilance beyond direct neighbors—even to just the second circle of influence—substantially promotes cooperation across all network topologies studied, synthetic and real alike. p. 1
Where the paper says it

Section IV summarizes the findings and discusses their implications for the spread of cooperative vigilance in real social networks. Our main finding is that extending vigilance beyond direct neighbors—even to just the second circle of influence—substantially promotes cooperation across all network topologies studied, synthetic and real alike, consistent with the empirical evidence that social influence reaches well beyond direct contacts [18, 23].

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive vigilance range selection: The system can dynamically adjust its monitoring range L based on real-time network conditions to maximize cooperation, shifting from L=1 to L=4 only when necessary for a given temptation level.

  2. Topology-dependent gain prediction: The AI can predict the speed at which cooperation gain accrues based on network structure, specifically distinguishing between sparse Erd˝os–R´enyi networks where the L = 1 → 2 step recovers most of the gain and dense Barab´asi–Albert networks requiring a third circle.

  3. Threshold sensitivity analysis: The system can precisely map how the critical temptation for defection shifts across different vigilance decay parameters lambda, identifying that dense, hub-dominated networks switch abruptly between low- and high-cooperation regimes when lambda crosses 0.5 to 0.75.

  4. Correlation detection in monitoring: The AI can quantify the impact of inter-layer correlation on cooperation gains, noting that this effect grows with L most prominently in sparse Erd˝os–R´enyi networks, indicating that extending vigilance amplifies alignment between game and monitoring layers.

Abstract

Social pressure---the awareness of being observed by others---is a fundamental driver of prosocial behavior in human societies. Yet it is typically assumed that only direct neighbors exert vigilance pressure on an individual, despite empirical evidence that social influence persists to at least three degrees of separation. Here we show that extending the reach of social vigilance beyond direct neighbors substantially promotes cooperation. We couple a Prisoner's Dilemma on one layer of a multiplex network to a vigilance cascade on the other, with influence decaying geometrically with network distance. Extending vigilance to just the second circle of influence (L=2) shifts the critical temptation for defection by roughly 36% in sparse Erdős--Rényi networks. Extending to four circles (L=4) raises this threshold by over 50% in the same networks. Both shifts are topology dependent: the L=1 to2 step alone already accounts for most of the gain in these sparse networks, whereas the pattern is more gradual in denser, hub-dominated Barabási--Albert networks, consistent with the decay coefficients of social influence reported in controlled experiments. The effect is strongest in sparse networks, requires that the vigilance and game layers be aligned, and reproduces directly on a real social network of physicians; in dense, hub-dominated networks the gain instead depends sharply on how fast influence decays with distance, switching between weak and strong cooperation as the decay rate crosses a threshold. Our results strongly suggest that even modest expansions of social awareness---such as those enabled by online social platforms---can substantially reshape the landscape of cooperative behavior in human populations.

Related papers