Taking a shower in Youth Hostels: risks and delights of heterogeneity

Tuning one’s shower in some hotels may turn into a challenging coordination game with imperfect information. The temperature sensitivity increases with the number of agents, making the problem possibly unlearnable. Because there is in practice a finite number of possible tap positions, identical agents are unlikely to reach even approximately their favorite water temperature. We show that a population of agents with homogeneous strategies is evolutionary unstable, which gives insights into the emergence of heterogeneity, the latter being tempting but risky.

💡 Research Summary

The paper models the seemingly mundane act of sharing a shower in a youth hostel as a coordination game with imperfect information. Each guest can set the water temperature by turning a tap that has a finite number of discrete positions (e.g., 0–10). The actual temperature of the water that reaches each user is the average of all selected tap positions, possibly transformed by a non‑linear mixing function that captures pipe dynamics and heat loss. Because the temperature response is highly sensitive to the number of participants, a small deviation in one person’s tap setting can cause a large deviation in the final temperature when many users are present. Consequently, the game quickly becomes “unlearnable” in the classic sense: agents cannot reliably infer the optimal joint action from past experience when the environment is noisy and the payoff landscape changes with each additional player.

The authors first examine the evolutionary stability of homogeneous strategies, i.e., all agents using the same tap setting. Using replicator dynamics, they introduce a mutant strategy that deviates slightly toward a hotter or colder setting. The analysis shows that when the group size N exceeds a modest threshold (approximately five users), the homogeneous equilibrium is destabilized: the mutant’s expected payoff exceeds that of the resident strategy because the temperature sensitivity term scales with N. Numerical simulations confirm that for N ≥ 5, any small perturbation spreads and eventually replaces the uniform strategy. In other words, a population of identical agents cannot maintain a common temperature preference in a realistic hostel shower scenario.

The paper then turns to the emergence of heterogeneity. Two complementary mechanisms are identified. The first is a “multiple‑equilibrium selection” process: each guest has a personal preferred temperature (e.g., 30 °C, 35 °C, etc.), and the best response to the current mix of taps depends on that preference. This naturally leads to a set of co‑existing strategies that form a mixed‑strategy equilibrium. The second mechanism is “risk‑aversion”: when a mutant appears, the system may temporarily overshoot or undershoot the desired temperature, creating uncomfortable hot or cold bursts. Some agents mitigate this risk by adopting conservative tap positions that lie near the midpoint of the feasible range. These conservative choices dampen fluctuations and increase the overall robustness of the system, even though they may slightly lower the average utility.

To validate the theoretical predictions, the authors built a physics‑augmented agent‑based simulation of a hostel shower. Ten agents, each with a preferred temperature drawn uniformly from 20 °C to 45 °C, choose among 11 discrete tap positions. The simulation incorporates realistic heat transfer, pipe mixing, and stochastic perception noise. Results show that under a homogeneous strategy the average temperature error relative to individual preferences can exceed 7 °C, making the shower experience intolerable for most users. When agents are allowed to adopt heterogeneous strategies, the error drops below 2 °C, and each guest’s temperature stays close to their personal target. Moreover, the heterogeneous configuration remains stable over many simulation rounds, confirming its evolutionary advantage.

The central conclusions are twofold. First, homogeneous strategies are evolutionarily unstable in coordination problems where the payoff function becomes increasingly sensitive to the number of participants. Second, heterogeneity—whether driven by diverse preferences or by deliberate risk‑mitigation—provides a stabilizing force that reduces systemic volatility at a modest cost to average payoff. The authors argue that these insights extend beyond hostel showers to any shared resource with a limited set of control actions and imperfect information, such as collective thermostat settings in smart buildings, distributed load‑balancing in power grids, or lane‑choice in congested traffic. They suggest future work on richer learning dynamics, the role of social norms in promoting heterogeneity, and empirical field studies in real hostels to test the model’s predictions.