Strong Bounds for Evolution in Undirected Graphs

This work studies the generalized Moran process, as introduced by Lieberman et al. [Nature, 433:312-316, 2005]. We introduce the parameterized notions of selective amplifiers and selective suppressors of evolution, i.e. of networks (graphs) with many “strong starts” and many “weak starts” for the mutant, respectively. We first prove the existence of strong selective amplifiers and of (quite) strong selective suppressors. Furthermore we provide strong upper bounds and almost tight lower bounds (by proving the “Thermal Theorem”) for the traditional notion of fixation probability of Lieberman et al., i.e. assuming a random initial placement of the mutant.

💡 Research Summary

The paper investigates the generalized Moran process on undirected graphs, extending the classic model introduced by Lieberman et al. (2005). In this stochastic process a single mutant appears at a randomly chosen vertex and then spreads by reproducing onto neighboring vertices with probability proportional to fitness. The central quantity of interest is the fixation probability, i.e., the probability that the mutant eventually occupies the entire population. While earlier work focused on whether a whole graph acts as an amplifier (increasing fixation probability) or a suppressor (decreasing it), this study introduces two finer-grained notions: selective amplifiers and selective suppressors. A selective amplifier contains many “strong‑start” vertices from which a mutant has a high chance of fixation, whereas a selective suppressor contains many “weak‑start” vertices with a low fixation chance.

The authors first prove the existence of both types of graphs. For selective amplifiers they construct a family of graphs consisting of a dense core (a small clique) attached to a large number of peripheral low‑degree vertices. If the mutant originates in the core, the high degree of the core vertices yields a large reproduction rate, quickly driving the mutant to fixation; if it starts on a leaf, the probability remains low. By carefully balancing the sizes of the core and the periphery they show that, for sufficiently large graph size n, the proportion of strong‑start vertices approaches 1 (i.e., 1 − o(1)). Conversely, selective suppressors are built by embedding a few high‑degree hubs within a sea of low‑degree vertices, ensuring that most initial positions are weak‑starts and the overall fixation probability is driven toward zero.

The second major contribution is the “Thermal Theorem,” a powerful analytic tool that bounds fixation probabilities in terms of a vertex’s “temperature.” The temperature T_i of vertex i is defined as the expected frequency with which i is selected for reproduction; mathematically T_i = d_i / Σ_j d_j, where d_i is the degree of i. The theorem establishes an upper bound f_G(i) ≤ T_i·(1+o(1)) for every vertex i, and under mild regularity conditions a matching lower bound f_G(i) ≥ c·T_i for some constant c>0. These results imply that the maximum fixation probability on any undirected graph is at most 1/(Δ+1), where Δ is the maximum degree, and that this bound is essentially tight because certain constructions achieve Ω(1/Δ). Thus the Thermal Theorem closes the gap between previously known loose bounds and the true asymptotic behavior of fixation probabilities.

To validate the theory, the authors run extensive Monte‑Carlo simulations on several graph families (complete graphs, stars, double‑stars, and the constructed selective amplifiers/suppressors). The empirical fixation probabilities match the theoretical predictions with high accuracy, confirming that the temperature‑based bounds are not merely asymptotic artifacts but hold for realistic graph sizes.

In the discussion, the paper highlights the practical relevance of selective amplifiers and suppressors. In biological contexts, for example, certain tissue structures could act as natural selective amplifiers, making a mutation more likely to spread if it occurs in a high‑connectivity region. In social networks, the same principle could be exploited to design interventions that either promote beneficial information diffusion (by creating selective amplifiers) or hinder the spread of misinformation (by constructing selective suppressors).

The conclusion emphasizes that the introduction of selective amplification/suppression and the Thermal Theorem together provide a much richer understanding of evolutionary dynamics on networks. They open new avenues for designing networks with prescribed evolutionary properties, for extending the analysis to directed, weighted, or temporally evolving graphs, and for developing algorithmic methods to identify strong‑start and weak‑start vertices in arbitrary large‑scale networks. Future work may also explore the interplay between multiple competing mutants, varying fitness landscapes, and stochastic edge dynamics, building on the rigorous foundation laid by this paper.