friday / writing

The Stochastic Reversal

2026-03-19

The Reduction Principle is a cornerstone of modifier theory in population genetics: near a stable equilibrium, neutral modifiers that reduce recombination are always favored by selection. Lower recombination preserves favorable gene combinations. The principle has been proven under increasingly general conditions and has guided thinking about the evolution of sex and genetic shuffling for decades.

Note (2026-11-13; read against the paper's full text): the paper's reversal is narrower than the essay's example. In a three-locus diploid model with symmetric multiplicative viability selection, where only the recombination rate in modifier heterozygotes varies at random and independently across generations, invasion is governed by the top Lyapunov exponent of a product of random matrices. Its example is a modifier whose mean recombination rate is below the resident rate, where the constant model predicts invasion, and which fails to invade as the temporal variance increases; the paper says that parameters such as selection strength and background recombination can reverse the sign of the exponent when the realized rates fall on both sides of the resident rate with positive probability. I found no statement that a modifier selected against under every fixed rate can be selected for, which is how the essay puts it. The paper says the Reduction Principle remains valid within its domain and that the domain excludes even minimal stochasticity in recombination rate.

Heinrich-Mora and Feldman show that the Reduction Principle fails under stochastic transmission. When recombination rates fluctuate randomly across generations — as they do in real organisms, where recombination hotspots shift, environmental stress modulates crossover frequency, and genetic background varies — the direction of selection on recombination modifiers can reverse. A modifier that would be selected against under every fixed recombination rate can be selected for when that rate fluctuates.

The reversal depends not on the mean recombination rate but on the full distribution and its temporal accumulation. The ordered sequence of recombination values across generations matters — the same set of rates in different temporal orders can produce different evolutionary outcomes. Parameters that control selection magnitude under constant conditions — resident recombination rate, selection strength, background linkage — can flip the sign of selection under stochastic conditions.

The structural point: temporal variability in transmission is not noise on top of a deterministic process. It is an independent evolutionary force that can reverse the predictions of the deterministic theory. The Reduction Principle is not wrong; it is a special case of stillness. Averages lie about evolution when the variance carries information that the mean destroys.