Behavioral patterns and mean-field games in epidemiological models

arXiv:2512.20547 · q-bio.PE, math.OC · Submitted 2025-12-23 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: I'm Ines.

Marcus: And I'm Marcus, here with Yuki.

Ines: We're looking at the paper: "Behavioral patterns and mean-field games in epidemiological models".

Marcus: Here's the gist of the paper. We introduce a new type of Mean Field Game epidemiological models in which subpopulations have different behavioral patterns: some are viewed as “highly rational” (choosing Nashequilibrium long-term strategies) while others follow pre-specified “non-rational” patterns (e.g., either sticking to their usual habits or trying to mimic those around them) <ref:2512.20547#pg2> The paper aims to build a bridge between two distinct communities of epidemiological modelers and to promote the consideration of behavioral patterns in broader Mean Field Games literature <ref:2512.20547#pg4>.

Ines: Model Framework and Assumptions The model extends standard Mean Field Games by incorporating subpopulations with distinct behavioral patterns: “highly rational” Optimizers, “non-rational” Followers, and “non-rational” Reckless individuals <ref:2512.20547#pg2> The framework allows for occasional behavioral switches among individuals which rational agents also take into account when formulating their Nash-equilibrium strategies <ref:2512.20547#pg5>.

Marcus: The baseline epidemiological model is a Susceptible-InfectedRecovered-Susceptible-Dead (SIRSD) epidemic model based on a frequency-based force of infection and a mortality rate that rapidly increases once the proportion of infected individuals exceeds some prescribed threshold <ref:2512.20547#pg7>.

Ines: Rational Optimization Dynamics The core of the rational agents, termed “Optimizers,” strive to maximize their personal cumulative reward by integrating over the time-horizon their instantaneous utility function that changes based on their current status and chosen contact rate <ref:2512.20547#pg8>.

Paper summary: Ines: So this paper is about extending Mean Field Game models to look at how different groups in a population behave when dealing with an epidemic, specifically introducing subpopulations that aren't all perfectly rational.

Marcus: Right, it claims that some people are highly rational, they're choosing long-term strategies based on Nash equilibrium, while others follow simpler patterns like sticking to old habits or just mimicking what everyone else is doing.

Ines: And the key idea here is allowing for these behavioral switches even among the highly rational agents when they formulate their strategies > This approach tries to bridge the gap between two different groups of epidemiologists who usually assume total rationality and those who include more human-like behaviors like inertia or conformity in their models.

Yuki: From a population genetics standpoint, I wonder how much weight we should give to these different behavioral patterns when we try to model disease spread across a whole species > Does this heterogeneity really matter, or is it just adding complexity?

Ines: The paper focuses on applying this framework to a specific SIRSD epidemic model based on frequency-based force of infection and a mortality rate that jumps up when the infected proportion passes some threshold, which creates strain on medical resources > They use this as the baseline for their tests.

Marcus: And what they're really looking at is how these non-rational patterns—the reckless ones, the followers—increase the impact of infectious diseases compared to just having rational optimizers in the population > The numerical tests show that adding these other behaviors makes a difference.

Yuki: So if we look at the results, it seems like even when you only have the highly rational Optimizers, they still manage to flatten the curve of the epidemic and reduce both the peak number of infected people and total deaths compared to a fully reckless population > That's an interesting finding for understanding how group behavior influences public health outcomes.

Ines: Exactly, but there's a caveat they mention; behavioral switching makes it harder for those rational Optimizers to lower the peak of infection > So while rationality helps, the presence of unpredictable switches can actually hurt their ability to control the outbreak.

Marcus: And what they explore is how these different subpopulations interact with each other regarding their contact rates—the way people choose how much they interact with others > It sets up a system where reckless individuals ignore the epidemic completely, followers try to match the average behavior, and optimizers make choices based on maximizing their personal reward over time.

Yuki: I'm curious about what that means for real-world modeling; are we suggesting that we need to explicitly model these kinds of behavioral biases when we look at how diseases move through human populations > Or is this just a theoretical extension of Mean Field Games?

Ines: The authors want to build a bridge between the community focused on pure rational optimization and the community that includes more complex, non-rational patterns like inertia or mimicry > It's about making the broader literature on Mean Field Games more inclusive of these kinds of human factors.

Marcus: It really matters because standard Mean Field Games often assume everyone is fully rational, but this model shows how relaxing that assumption—by having different types of agents—can lead to different epidemiological outcomes.

Yuki: So, in simple terms for someone listening who just wants the gist, the paper shows that mixing different decision-making styles within a population changes how severe an epidemic turns out > It moves beyond just assuming everyone is perfectly calculating their best move.

Ines: And they also point out some limitations; they mention that the model is mostly qualitative and suggests future work focusing on incorporating fatigue-induced behavioral switches or uncertainty for those rational agents > They are flagging that we still don't fully know how these real-world cognitive biases affect contact rate choices.

Marcus: And I think that part is important because it reminds us that some human decisions aren't based purely on game theory; cognitive biases and just sheer inertia could be driving the contact rates in ways the current math doesn't capture > It’s a reminder to consider those non-game-theoretic aspects.

Yuki: So, to wrap up this summary of "Behavioral patterns and mean-field games in epidemiological models", the main point is that introducing different behavioral rules for different groups fundamentally alters the epidemic trajectory compared to models where everyone acts identically > It’s about acknowledging that not all agents are making decisions based on pure game theory.

Ines: And for the conclusion of this paper, it's about showing how these specific behavioral patterns—highly rational versus non-rational followers or reckless individuals—can be incorporated into epidemiological models to get a richer understanding of disease spread > It's about moving beyond the simplest, most uniform agent assumptions in Mean Field Games.

Marcus: It’s important because it gives epidemiologists and mathematicians a new toolset to look at population heterogeneity in disease modeling, moving beyond just the standard SIRSD setup > The authors are really pushing for a broader consideration of these behavioral patterns in Mean Field Games literature.

Conclusion: Marcus: So we're talking about these new behavioral patterns in epidemiological models, specifically how some people are highly rational and others aren't following any clear rules at all.

Ines: That’s right. The authors are looking at how incorporating things like stubbornness or just people trying to copy what everyone else is doing changes the whole picture of an epidemic over time.

Yuki: It sounds like they're trying to build a bridge between two groups of modelers who usually assume perfect rationality and those who want to include more human-like stuff in their math.

Marcus: Exactly. They’re showing how those different behaviors affect the numbers we get when we look at disease spread, which is really important for understanding real-world public health situations.

Ines: It’s about taking that basic Susceptible Infected Recovered Susceptible Dead model and adding these layers of behavioral logic to see what happens.

Yuki: And the results they show are pretty interesting because even when you only have the most rational agents, they can still manage to flatten the curve of an epidemic and reduce both how many people get sick at once and the total number of deaths compared to a fully reckless population.

Marcus: That’s a big difference. But they also point out that those behavioral switches themselves make it harder for those rational optimizers to lower that peak infection number.

Ines: So, the paper is really suggesting that while pure rationality helps control the disease, those unpredictable human behaviors can actually complicate things and make it harder to get results as good as we might expect from a purely mathematical optimization standpoint.

Yuki: And they’re not just stopping there though. They mention that this model is pretty qualitative, meaning it gives us a good feel for the concepts but doesn't give us all the hard data yet.

Marcus: Right, and they suggest future work should focus on things like how fatigue causes people to switch behaviors or how those rational agents make choices when they don't have perfect information about what everyone else is doing.

Ines: So, this paper sets up a framework for thinking about population heterogeneity in disease spread using Mean Field Games and points us toward where the next set of questions should go.

Cornell University

q-bio.PE, math.OC

Submitted: 2025-12-23

Updated: 2026-10-30

Comments: 45 pages; 13 figures; 2 appendices

Code: https://github.com/eikonal-equation/Behavioral-MFGs-in-epidemiology

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 61/100

The gist: We introduce a new type of Mean Field Game epidemiological models in which subpopulations have different behavioral patterns…

Key concepts

Mean Field Game (MFG)
A mathematical tool used to model large groups of interacting agents where each agent's decision depends on the average behavior of the entire population. It helps analyze complex systems like how many people get infected based on collective choices.
Optimizers
Highly rational individuals who choose contact rates strategically to maximize their personal long-term reward, considering how their choices interact with the strategies of others in a Nash equilibrium setting.
Behavioral Switching
The possibility that individuals change their behavioral patterns over time. Rational agents account for this uncertainty when planning their optimal contact rate strategies, making the model more realistic.

Terminology

Summary

We introduce a new type of Mean Field Game epidemiological models in which subpopulations have different behavioral patterns: some are viewed as “highly rational” (choosing Nashequilibrium long-term strategies) while others follow pre-specified “non-rational” patterns (e. p. 0g., either sticking to their usual habits or trying to mimic those around them) <ref:2512. p. 0 p. 020547#pg2> The paper aims to build a bridge between two distinct communities of epidemiological modelers and to promote the consideration of behavioral patterns in broader Mean Field Games literature <ref:2512. p. 0 p. 020547#pg4>.

Model Framework and Assumptions

The model extends standard Mean Field Games by incorporating subpopulations with distinct behavioral patterns: “highly rational” Optimizers, “non-rational” Followers, and “non-rational” Reckless individuals <ref:2512.20547#pg2> The framework allows for occasional behavioral switches among individuals which rational agents also take into account when formulating their Nash-equilibrium strategies <ref:2512. p. 020547#pg5>. The baseline epidemiological model is a Susceptible-InfectedRecovered-Susceptible-Dead (SIRSD) epidemic model based on a frequency-based force of infection and a mortality rate that rapidly increases once the proportion of infected individuals exceeds some prescribed threshold <ref:2512. p. 020547#pg7>.

Rational Optimization Dynamics

The core of the rational agents, termed “Optimizers,” strive to maximize their personal cumulative reward by integrating over the time-horizon their instantaneous utility function that changes based on their current status and chosen contact rate <ref:2512.20547#pg8>. When a lone focal individual chooses a different strategy, they will not influence the course of the epidemic but will change their personal chances of getting ill and their personal lifestyle <ref:2512. p. 920547#pg10>. The value functions are defined to quantify the result of optimizing the contact rates, leading to optimality equations that define the Nash equilibrium in a Mean Field Game setting <ref:2512. p. 920547#pg9>.

Behavioral Pattern Interactions

The model incorporates three subpopulations: Reckless/stubborn individuals who largely ignore the epidemiological situation, Followers/consensus-seekers who try to adopt the “population-averaged” behavior, and rational Optimizers <ref:2512.20547#pg5>. Susceptible Reckless individuals select a contact rate that myopically maximizes their utility function, while Susceptibles Followers prefer to adopt the “consensus” modeled by averaging the contact rate used by all Susceptibles <ref:2512. p. 320547#pg9>. The rational Susceptible Optimizers select a Nash-type contact rate taking into account the behavior of other subpopulations and the possibility of occasional behavioral switches <ref:2512. p. 320547#pg10>.

Epidemiological Outcomes and Sensitivity

Numerical tests illustrate that additional/non-rational behavioral patterns and behavioral switching increase the impact of infectious diseases <ref:2512.20547#pg4>. Comparing scenarios, the Optimizers-only population still manages to flatten the curve of the epidemic and reduce the peak number of Infected, achieving a reduction in total deaths compared to a fully Reckless population <ref:2512.20547#pg9>. The impact of behavioral switching is shown to make it harder for Optimizers to lower the peak of infection <ref:2512. p. 020547#pg10>.

Future Directions and Limitations

The paper notes that the model remains primarily qualitative in nature and suggests extensions such as other epidemiological compartment models, more sub-population types with different behavioral logic, and effects of vaccinations <ref:2512.20547#pg10>. Future work should focus on refining a model of fatigue-induced behavioral switches and incorporating uncertainty or partial information for Optimizers’ contact-rate choices <ref:2512. p. 320547#pg10>. The authors emphasize that some humans are not making decisions based on game-theoretic considerations, suggesting that cognitive biases and inertia should be considered in contact rate modeling <ref:2512.20547#pg10>.

Acknowledgments

This research was partially funded by the National Science Foundation (award DMS-2111522) as well as the Air Force Office of Scientific Research (award FA9550-22-1-0528) and the second author’s Royal Society Wolfson Visiting Fellowship. The authors also thank A. Gumel for pointing out many relevant papers and S. p. 0 Ellner for useful suggestions on improving this manuscript. p. 2

Data and Code Availability

All data and code used to generate the figures in this manuscript will be available online at https://github.com/eikonal-equation/Behavioral-MFGs-in-epidemiology. p. 0 The paper is 2512. p. 020547.

--- Page 1 ---

We introduce a new type of Mean Field Game epidemiological models, in which subpopulations have different behavioral patterns: some are viewed as “highly rational” (choosing Nashequilibrium long-term strategies) while others follow pre-specified “non-rational” patterns (e.g., either sticking to their usual habits or trying to mimic those around them) <ref:2512.20547#pg2>. Our model also allows for occasional behavioral switches, which rational individuals also take into account when formulating their Nash-equilibrium strategies <ref:2512. p. 020547#pg5>. While this modeling approach is general, here we develop it for individuals choosing their “contact rates” within a particular Susceptible-InfectedRecovered-Susceptible-Dead (SIRSD) epidemics model <ref:2512. p. 020547#pg7>. The latter is based on a frequency-based force of infection and the mortality rate that rapidly increases once the proportion of infected individuals exceeds some prescribed threshold, resulting in a strain on medical resources <ref:2512. p. 020547#pg7>. Numerical tests illustrate the properties of our model and highlight the ways in which additional/non-rational behavioral patterns and behavioral switching increase the impact of infectious diseases <ref:2512. p. 020547#pg4>. The paper aims to build a bridge between two distinct communities of epidemiological modelers and to promote the consideration of behavioral patterns in broader Mean Field Games literature <ref:2512.20547#pg4>.

--- Page 2 ---

Mathematical epidemiology is a classical area of applied mathematics, which saw a significant increase in research activities since the onset of COVID pandemic Many mathematicians seized the opportunity to bring new tools and ideas to these problems of obvious and immediate societal importance. Among them, the experts in Mean Field Games (MFGs) have proposed a number of models to explain individuals’ choices on how much to follow official rules and recommendations (on vaccinations, social distancing, mask wearing, etc. p. 1) [1, 2, 5, 6]. In MFG literature, these choices are viewed as fully rational in a selfish or decentralized manner: “optimal” from the point of view of individual decision makers. p. 1 But computing and analyzing Nash equilibria is generally harder in dynamic games involving a large number of players. p. 2

--- Page 3 ---

Mean Field Games (MFGs) were introduced to address the latter challenge [22, 26, 27] and have by now found much success, particularly in economics and engineering applications. The key idea in MFGs is to consider a game with infinitely many “small” players, each of whom interacts with an evolving density of players rather than with specific individual players. p. 2 Computationally, this leads to a coupled system of differential equations describing the evolution of players’ density and players’ optimal (Nash equilibrium) behavior. p. 2

--- Page 4 ---

The rational optimization perspective allows for comparing the epidemiological and economic impact of Nash equilibrium solutions with what could be achieved with “socially optimal” solutions [10, 40]. Other models go beyond this comparison and formulate the problem as a “Stackelberg game” between the authorities, who choose a public policy under assumption that people will respond to them in a rational but selfish manner. p. 2 While the above are usually formulated with the assumption that the logic of all participants is homogeneous, there is also a broader literature on multipopulation MFGs [4, 11, 13], which allows each subpopulation to have different optimization goals and dynamics. p. 2

--- Page 5 ---

But the full rationality and consistency assumptions remain fundamental even in such standard MFG multipopulation models, with all individuals assumed to have the full knowledge of others’ priorities and risks, ability to compute a personal (time-dependent) Nash equilibrium strategy. In this paper, we suggest an approach for relaxing several of these assumptions by recognizing that only some of the people are highly rational, while others frequently adopt behavioral patterns that are not grounded in game theory <ref:2512. p. 320547#pg10>. More specifically, we consider three subpopulations: Reckless/stubborn individuals, Followers/consensus-seekers, and rational Optimizers <ref:2512. p. 320547#pg10>.

--- Page 6 ---

Since the goal of the paper is largely to foster the dialogue between two distinct research communities, this also influences our exposition: we primarily focus on fundamental modeling assumptions and their consequences illustrated through numerical experiments, leaving other important aspects (theoretical, numerical, and data-centric) for the future <ref:2512.20547#pg4>. To make the discussion concrete, we introduce a specific simple SIRSD ODE model in § 1. p. 31 <ref:2512. p. 020547#pg7>. We then develop its generalized MFG version in § 2: starting with the optimization task faced by a single rational individual when others’ strategies are already known, deriving a Nash description when all such individuals make decisions simultaneously, then adding other behavioral subpopulations and the possibility of switching between them <ref:2512. p. 320547#pg7>.

--- Page 7 ---

We assume that at any given time each individual can be either Susceptible (S), Infected (I), Recovered/Immune (R) or Dead (D), with the corresponding letters also used to encode the fraction of individuals in each state. For the sake of simplicity, we also ignore the births and disease-unrelated deaths. p. 4 We further assume that Susceptible, Infected, and Recovered individuals each on average contact cS, cI, and cR other individuals (respectively) per unit time. p. 4 For now, these will be treated as constants, but starting in the next section we will interpret these as individual’s time-dependent control variables. p. 4

--- Page 8 ---

An epidemic trajectory resulting from this model is illustrated in Figure 1 using the baseline contact rates (cS = 5, cI = 3, cR = 5), which will be motivated in the next section <ref:2512.20547#pg7>.

--- Page 9 ---

Since the goal of the paper is largely to foster the dialogue between two distinct research communities, we describe our approach in greater detail than it is usually done in MFG literature, focusing on modeling assumptions as well as on their mathematical and practical implications <ref:2512.20547#pg4>. Our focus will be a mixed population of agents, all of whom change their personal contact rate independently with the full knowledge of current epidemiological situation <ref:2512. p. 620547#pg9>. Some of these agents (referred to as “Optimizers”) are assumed to be highly rational in a sense that they strive to maximize their personal cumulative reward, which is computed by integrating over the time-horizon [0, T] their instantaneous utility function u that changes based on their current status (S, I, or R) and the chosen contact rate <ref:2512. p. 620547#pg9>.

--- Page 10 ---

Individuals need to make contact with others for economic gains and for their personal well being. Thus, we assume that an individual’s instantaneous utility is a function of their contact rate with others. p. 7 While our general model does not require a specific type of utility functions, all of the numerical experiments in § 3 are based on the utility functions previously formulated in [10] <ref:2512. p. 720547#pg9>.

--- Page 11 ---

For each status z ∈ (S,I,R) and contact rate c we use uz(c) = bzc − c 2/g − az. This specifies the same instantaneous utility for all healthy individuals (status S or R) and a different, generally lower, instantaneous utility for Infected individuals <ref:2512. p. 720547#pg9>.

--- Page 12 ---

The maximal instantaneous utility values u¯S = uR and u¯I are achieved with contact rates c = bz/2. These are also the contact rates chosen in the previous section to produce Figure 1 <ref:2512. p. 820547#pg7>.

--- Page 13 ---

An individual in the population accumulates reward according to the utility function of their current state, which is subject to random transitions. To begin, suppose that the entire population uses some specific and known contact rate strategy c(t) = (cS(t), cI(t), cR(t)) <ref:2512. p. 820547#pg7>. If a lone focal individual chooses some different strategy ˆc = (ˆcS(t), cˆI(t), cˆR(t)), they will not influence the course of the epidemic (i. p. 8e., y(t) will remain the same) but they will change their personal chances of getting ill and their personal lifestyle <ref:2512. p. 920547#pg9>.

--- Page 14 ---

Mathematically, for a focal individual starting from a state ξ ∈ (S, I, R, D) at the time t < T and using contact rate ˆc(·) until the time T, their expected cumulative reward will be Jξ, t,ˆc(·); y(t), c(t) = E "Z min(T,T D) t u z(r) cˆz(r)(r) dr + V D 1T D≤T <ref:2512.20547#pg9>.

--- Page 15 ---

Here, D is the absorbing state, T D = min(r ≥ t z(t) = D) is a random time of death, and V D is a large negative terminal “reward” received if this sad outcome occurs within the planning horizon <ref:2512.20547#pg9>. Some of these transition rates are constant; e. p. 9g., for the recovery from infection or the loss of immunity to happen within a short time interval τ, P z(r + τ) = R z(r) = I = µτ + o(τ) and P z(r + τ) = S z(r) = R = γτ + o(τ) <ref:2512. p. 920547#pg9>. Other rates are y(r)-dependent or even y(t), c(r),ˆc(r)-dependent; e. p. 9g., for a death to occur within a short time interval τ, P z(r + τ) = D z(r) = I = δ I (r) τ + oτ) and for a new infection, P z(r + τ) = I z(r) = S = ˆcS (r) p y(t), c (r) τ + oτ), with p determined by the current state of the epidemic and the current contact rates used by all others <ref:2512. p. 920547#pg7>.

--- Page 16 ---

Assuming that the focal individual strives to maximize their expected cumulative reward, one can define value functions to quantify the result of optimizing the contact rates: Vξ(t) = sup ˆc(·) Jξ, t,ˆc(·); y(t), c(t) <ref:2512.20547#pg9>.

--- Page 17 ---

When z(r) = I or z(r) = R, the focal individual’s choices do not influence their transition rates, and thus they will always choose the contact rate that maximizes their current utility function <ref:2512.20547#pg9>. On the other hand, that individual has a non-trivial choice to make when z(r) = S: a contact rate lower than ¯cS is often optimal since it reduces the immediate chances of getting infected <ref:2512. p. 920547#pg9>.

--- Page 18 ---

Using the above expressions for transition rates, the optimality equations satisfied by the value functions can be written as V S(t) = max cˆS(·) (9) where Z t+τ t uS cˆS(r) dr! + p y(t), cˆS VI(t + τ) + 1 − p cˆS(t)τ VS (t + τ) <ref:2512. p. 020547#pg9>.

--- Page 19 ---

A standard control-theoretic argument based on a Taylor-series expansion of (9-11) yields a system of ODEs for the value functions V′S(t) = −uS(cˆS) + p y(t), cˆS S (VS(t) − VI(t)) <ref:2512.20547#pg9>.

--- Page 20 ---

In MFG literature, such systems are often treated numerically by forward-backward iterations, which, for stability reasons, might also be combined with an under relaxation [28] or with the so called “fictitious play” [7] <ref:2512.20547#pg9>. Since our models in this paper are over a finite state space, we instead employ the standard collocation methods for TPBVPs (e. p. 2g., the built-in Matlab function bvp5c [25] <ref:2512. p. 020547#pg9>.

--- Page 21 ---

The resulting model can be now formally described by a system of 19 coupled ODEs using indices j ∈ (r, f, o) and k ∈ (r, f, o) excluding j: S′j = −p cˆS(cˆS) S jSj + γR j + X k!=j λk,jSk − λj,kSk <ref:2512. p. 020547#pg10>.

--- Page …

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. System can simulate epidemic spread under heterogeneous rational behavior by distinguishing between Optimizers, Followers, and Reckless subpopulations to predict how different behavioral patterns alter the epidemic trajectory compared to standard models.

  2. The AI can determine the Nash equilibrium contact rate for Susceptible Optimizers by solving a specific maximization problem that balances instantaneous utility against the predicted future epidemiological trajectory influenced by all other agents' strategies.

  3. The system can model dynamic behavioral switching in rational agents, where Optimizers occasionally switch to Reckless behavior based on factors like fatigue or frustration, leading to non-constant contact rates over time.

  4. The AI can quantify the impact of behavioral patterns by comparing epidemic outcomes between a fully rational population of Optimizers and a purely reckless population, showing how strategic choices reduce the peak infection count and deaths.

  5. The system can analyze how the presence of Followers affects an Optimizer's behavior, learning that Susceptible Followers tend to be heavily influenced by Optimizers near the epidemic peak when Reckless individuals are prevalent.

Sources

Related papers