Waiting for Something to Happen

Poisson processes across evolution and ecology

Published

August 20, 2026

A Poisson process is what you get when something happens at a steady rate and each occurrence pays no attention to the ones before it. You can look at it from either end. Count the events in a fixed window and you get a Poisson distribution, whose variance happens to equal its mean. Measure the gaps between events instead and you get an exponential distribution with mean \(1/\lambda\). Set the count to zero and you get \(e^{-\lambda t}\), the chance that nothing happens at all — and a lot of what follows is really a question about that.

What makes the whole thing work is memorylessness. A flower that has been open four days without a visitor is no more likely to get one tomorrow than it was on its first morning. A genus that has lasted 20 million years isn’t owed an extinction. The past leaves no mark on what comes next.


Mutation, recombination, and gene flow

Event Rate Where it matters
Point mutation at a site Per-site per-generation mutation rate Mutation accumulation lines; germline rate estimates
Neutral substitution along a lineage Equals the mutation rate (\(2N\mu \times 1/2N\)) The molecular clock; population size cancels out
Synonymous substitution between paralogs \(K_s\) divergence Dating gene and genome duplications
Recombination breakpoints along a chromosome Crossovers per Mb Linkage map construction; LD decay

Because there are only four bases, a site hit twice looks the same as one hit once, and a site that mutates and then reverts looks untouched. Observed differences level off as a result — at 0.75 under Jukes–Cantor, once sites are thoroughly saturated. Distance corrections are just a way of working backwards from what you can see to how many changes actually happened.


Genome content

Event Rate Where it matters
Gene duplication Roughly 0.01 per gene per My (varies widely) Gene family expansion
Gene loss / pseudogenization Duplicate half-life on the order of a few My Fractionation after polyploidy
Transposable element insertion Highly variable Genome size evolution — but strongly bursty, see below

Gene families work the same way with one wrinkle: every copy can duplicate or be lost, so a family of \(n\) genes gains copies at rate \(n\lambda\) and sheds them at rate \(n\mu\). Letting \(\lambda = \mu\) is not unreasonable at large scales where gains and losses are approximately in equilibrium, but plants add a complication. Whole-genome duplications double everything at once, and they are followed by large-scale losses that are systematically biased.


Macroevolution

Event Rate Where it matters
Speciation Per-lineage rate \(\lambda\) With \(n\) lineages, the next split waits Exp(\(n\lambda\)). Waiting times shrink as clades grow
Extinction Per-lineage rate \(\mu\) Log-linear survivorship in the fossil record implies extinction is always possible
Fossilization / preservation Per-lineage sampling rate Fossilized birth–death models; gap statistics

The Red Queen hypothesis builds upon the memorylessness of evolution. A long history of survival does nothing to prevent extinction in the future. Evolution is always occurring.

One thing to watch for: reconstructed phylogenies tend to bend upward near the present even when rates have been steady all along, simply because lineages that arose recently haven’t had time to die yet. A fair number of apparent recent radiations are this artifact rather than a real burst.


Ecology and life history

Event Rate The cost of waiting
Pollinator visit to a flower Visitation rate Nectar, water, respiration — sets optimal floral longevity
Fire Fire return interval Delayed reproduction in fire-adapted species
Outcrosser pollen arrival Pollinator reliability Inbreeding depression — sets timing of delayed selfing

The general theme, something worth having will turn up eventually. There is no guarantee when in a given time interval, but evolution suggests the events happen regularly enough to exert selective pressures on phenotypes and ultimately genotypes.