Copy a genome badly enough, forever, and it can only get worse. In a finite population that never recombines, the best genotype left alive is one unlucky generation from gone — and once it's gone, nothing can build it back. So the whole population walks, one irreversible step at a time, toward more mutations. Here is the machine that does it, the number that sets its speed, and the one thing that stops it.
Here is a fact that sounds like it can't be true. Take a population of living things that reproduce asexually — each parent making near-copies of itself, no sex, no mixing. Let mutations trickle in the way they really do: mostly harmful, each one a small tax on survival. Now make the population finite — any real size, a million, a billion, it doesn't matter. Over time, the average number of harmful mutations per individual does not settle. It climbs, in discrete clicks, and it never comes back down. The population decays, by a mechanism with no reverse gear.
This is Muller's ratchet — argued verbally by Hermann Muller in 1964, named by Joe Felsenstein in 1974, and turned into exact mathematics by John Haigh in 1978. The name is precise: a ratchet is a wheel with a pawl that lets it turn one way only. The reason it turns one way is almost embarrassingly simple, and the page will let you watch it happen.
Give every individual a fitness that falls off multiplicatively with the number of harmful mutations it carries: an individual with k of them reproduces in proportion to (1−s)ᵏ, where s is the cost of one mutation. Two dials do the rest: U, the number of new harmful mutations a genome picks up per generation, and N, the population size. Each generation, individuals have offspring in proportion to fitness, and each offspring picks up Poisson(U) new mutations. That's the whole model. Watch one population run.
Run it. The distribution marches right and never retreats — because to move left, some individual would have to lose a mutation, and asexual copying with no back-mutation can only add. When the green class is thin, it blinks out by pure chance, and the front jumps forward a notch. That notch can never be undone. Set s small or U large and watch the ratchet race.
Why does the front class blink out — and why does that decide everything? Because the fittest genotypes are always the rarest. At mutation–selection balance the distribution of mutation counts is exactly a Poisson with mean U/s (Haigh proved this — selection on a Poisson returns a Poisson, mutation shifts it back, and the fixed point is U/s). The mutation-free class is the thin left tail of that Poisson: a fraction e−U/s of the population. So its expected head-count is
n₀ = N · e−U/s
and Haigh called it the most important parameter of the whole system. It is a small number sitting on the edge of extinction. When n₀ is a few dozen, the front class is a robust little population that drifts but rarely dies — the ratchet is effectively frozen. When n₀ drops toward 1, the front class is one or two individuals, chance wipes it out constantly, and the ratchet races. Dial n₀ yourself and watch the front class fight for its life.
This is Haigh's engine laid bare. The ratchet's whole speed rides on one number, n₀. Push n₀ above ~25 (big N, or a small U/s) and the clicks nearly stop. Pull it below ~1 and they come in a flood. The exact click rate for the in-between — n₀ of a handful — has no closed form to this day; it's the honest open edge of a fifty-year-old model (Haigh 1978; Gordo & Charlesworth 2000).
Before the ratchet even starts to turn, the mutations already cost the population something: at mutation–selection balance the mean fitness sits at a reduced level called the mutational load. You'd guess a nastier mutation (bigger s) costs more. It doesn't. The mean fitness at balance is
w̄ = e−U
which depends only on U, the rate mutations arrive — not at all on s, how bad each one is. Make each mutation twice as deadly and selection removes it twice as efficiently; the two effects cancel exactly. This is the Haldane–Muller principle (Haldane 1937; Muller 1950): a population pays for its mutations at the rate they occur, whatever their severity. In Instrument I, slide s at fixed U and watch the mean-mutation count balloon while the mean fitness barely stirs — the load is holding at e−U, exactly as the check confirms.
So an asexual finite population is doomed to decay. Real asexual lineages exist and persist — so something must be reaching in and turning the ratchet back. Muller's answer, and Felsenstein's, was sex. Recombination takes two genomes that each lost the front class in different ways — one carrying mutation A, the other mutation B — and reshuffles them, and among the offspring is one carrying neither. The mutation-free class, extinct in both parents' lineages, is rebuilt from parts. Put the two side by side, same population, same mutations, same luck — only the shuffling differs.
Both start identical — every individual mutation-free — and see the same mutations at the same moments. The only difference is that the right-hand population reshuffles its genes each generation. That is enough. The clones ratchet downhill: their kmin is a staircase that only ever climbs, and the mean load climbs with it, without bound. The recombiners rebuild their best genotype from parts, so their kmin wanders up and back down — not a ratchet at all — and the mean load holds near the balance value U/s. This is the oldest surviving argument for why sex is worth its enormous costs — though, honestly, not the only one.
Every bar, staircase, and number above is computed live in your browser by the same arithmetic the verifier proves offline: fitness-proportional Wright–Fisher resampling, Poisson(U) mutation, and — in Instrument III — free recombination over a long chromosome, all from a seed so a screenshot is reproducible. The predicted n₀ = N·e−U/s and load e−U are Haigh's exact analytic results, and the verifier confirms the simulation lands on them.
The named frontier. Four honest edges. First, the model gives every deleterious mutation the same cost s and assumes no interaction between them (multiplicative fitness); real mutations have a whole distribution of effects, and epistasis between them can speed the ratchet or slow it — a deliberate idealization, Haigh's and ours. Second, the click rate for intermediate n₀ (a handful of individuals in the front class) still has no closed form; the two ends — n₀ ≲ 1 fast, n₀ ≫ 1 nearly frozen — are well understood (Gordo & Charlesworth 2000), the middle is fit by simulation. Third, the ratchet can be halted or reversed in the real world by back-mutation, compensatory mutations, or the occasional beneficial one; Haigh showed back-mutation is usually far too rare to matter, but the others are real exits. Fourth, Muller's ratchet is one hypothesis for the evolution of sex, not the verdict — it sits beside the Fisher–Muller effect (sex combines separate beneficial mutations), Kondrashov's deterministic-mutation hypothesis (which needs synergistic epistasis and works even in infinite populations), and the Red Queen (sex to outrun coevolving parasites). This page shows that the ratchet is real and does turn; it does not claim it is the whole story of sex.
Where the ratchet is thought to actually bite: in things that are small and don't recombine. RNA viruses under repeated bottlenecks lose fitness in measurable steps (Chao's 1990 experiment on phage φ6). The tiny, clonal genomes of insect endosymbiotic bacteria show accelerated decay consistent with it (Moran 1996, on Buchnera). And the non-recombining Y chromosome, and the mitochondrion, are the ratchet's suspected fingerprints inside sexual organisms like us — the parts of our own genome that don't get to shuffle.
research/mullers-ratchet/verify.mjs,
20/20. Recomputed, not asserted:
(1) the mutation–selection balance is exactly Poisson(U/s)
— a machine-precision fixed point of selection∘mutation at three parameter sets;
(2) selection on Poisson(λ) returns Poisson(λ(1−s)) exactly;
(3) mean fitness at balance = e−U, independent of s (Haldane–Muller);
(4) the simulated least-loaded class tracks n₀ = N·e−U/s, mean load ≈ U/s, mean fitness ≈ e−U;
(5) kmin is a non-decreasing staircase — the ratchet is irreversible;
(6) smaller n₀ turns it faster, larger n₀ freezes it;
(7) the loci model matches the count model at balance;
(8) free recombination clicks far less than clonal copying — sex halts the ratchet.
Run it: node research/mullers-ratchet/verify.mjs.