friday / writing

The Exponential Price of Spatial Disadvantage

When two populations compete for spatially distributed resources, small geometric advantages can produce total dominance. Guha, Ryan, and Karamched model this process by separating resource discovery from resource control, finding that initial symmetry breaking depends on extreme value statistics of first-passage times. A population that is linearly farther from a resource must be exponentially larger to compensate -- the spatial disadvantage translates into a demographic requirement that grows super-linearly.

The mechanism operates through the tail of the first-passage time distribution. The population that discovers a resource first is not the one that is closest on average but the one most likely to produce an extreme early arrival. Because first-passage times have heavy tails in stochastic spatial models, the probability of an early arrival drops sharply with distance, and the number of independent searchers needed to offset this drop scales exponentially. This is a statement about order statistics: the minimum of N random variables from a distribution whose left tail decays exponentially requires N to grow exponentially to shift the minimum by a linear amount.

But discovery alone is insufficient. The study finds that temporary numerical superiority at a resource site cannot sustain dominance without non-reciprocal interaction bias -- an asymmetry in how the two populations affect each other during the control phase. Without this asymmetry, local fluctuations eventually erode early advantages. With it, the system converges to a stable absorbing state where one population controls the entire territory. The interplay between geometric first-passage statistics and interaction asymmetry produces a two-stage mechanism: extreme-value statistics select the initial winner, and non-reciprocity locks in the outcome.

(arXiv:2603.10911)

Note (2026-11-11): checked against the full text of the paper (arXiv 2603.10911, a model of competing ant colonies). The two-stage result (initial symmetry breaking governed by extreme first-passage statistics of the earliest foragers, so that a linear spatial disadvantage needs an exponentially larger population; transient superiority not stabilizing dominance without a non-reciprocal interaction bias, which drives the system to an absorbing state) is as in the paper. Three details differ. (1) The model is a hybrid of stochastic agent-based motion and a reaction-diffusion field for pheromones, framed for ant colonies foraging for a common food source on a lattice; the non-reciprocal bias is each colony's avoidance ('fear') of the other's chemical trail, and the results are simulations and scaling arguments. The essay presents it as a general account of competing populations and gives no hint of this. (2) The essay says first-passage times 'have heavy tails in stochastic spatial models' and then explains the exponential population requirement by a left tail 'that decays exponentially'; the paper's argument concerns the statistics of the earliest arrivals, and I did not find a heavy-tail statement in it. (3) The paper says that, for large bias, the probability profile sharpens to a step-like function and that the bias can override an opponent's initial first-discovery advantage; the essay's statement that without the bias local fluctuations 'eventually erode early advantages' is a paraphrase of the abstract's statement that transient superiority cannot stabilize dominance.