Chapter 11 The mathematics of evolution

12 minute read

Illustration for Chapter 11, The mathematics of evolution

In November 1859, Charles Darwin and Alfred Russel Wallace presented the world with a simple, powerful mechanism for the origin of species. Organisms produce more offspring than scarce resources can support, individuals vary in heritable traits, and those bearing advantageous variations leave more descendants. Yet Darwin spent the rest of his life haunted by a gap at the heart of his theory. He had no working model of heredity.

Throughout the 19th century, most naturalists assumed what was called blending inheritance. Offspring were thought to be a smooth, continuous mixture of their parents, just as mixing red paint with white paint yields pink paint. In 1867, a Scottish professor of engineering named Fleeming Jenkin pointed out that blending inheritance would strangle natural selection in its cradle. If a single mutant individual arises with a beneficial trait in a large population, mating with normal individuals cuts the trait's deviation in half every generation. Statistically, blending inheritance cuts the genetic variance in half in every reproductive cycle. Within 10 generations, less than one-thousandth of the original variation remains, diluting any favorable novelty into the background average long before differential survival can amplify it.

Ironically, the solution was already sitting unread on the shelves of Darwin's library. In 1866, an Augustinian friar named Gregor Mendel published the results of eight years of breeding 28,000 pea plants in a monastery garden in Brno. Mendel discovered that heredity is not a continuous fluid that blends away upon mixing. Instead, traits pass from parent to child as discrete, indivisible particles—what we now call genes—that remain intact even when hidden in a hybrid generation. Between 1918 and 1932, three brilliant thinkers—Ronald Fisher and J.B.S. Haldane in Britain, and Sewall Wright in the United States—fused Mendel's particulate genetics with Darwin's natural selection, founding the mathematical science of population genetics.

11.1Particulate inheritance and the adaptive landscape

Consider a large, randomly mating diploid population in which every individual carries two copies of each chromosome, one inherited from each parent. Focus on a single genetic locus that occurs in two alternative forms, or alleles. Suppose the first allele makes up a fraction pp of the gene pool, while the second allele makes up the remaining complement.

In 1908, the English mathematician G.H. Hardy and the German physician Wilhelm Weinberg independently asked what happens to this gene pool in the absence of evolutionary forces. Without selection, mutation, migration, or finite-sample drift, random mating is equivalent to drawing two alleles independently from the pool. Expanding the binomial square locks the three diploid genotypes into invariant Hardy-Weinberg proportions after a single generation. Homozygotes carrying two copies of the first allele occur at the square of its frequency. Meanwhile, heterozygotes occur at twice the product of the two allele frequencies, and homozygotes for the second allele occur at the square of the second frequency.

Because genes are discrete DNA sequences rather than miscible liquids, Mendelian segregation preserves these allele frequencies from one generation to the next. Homozygotes transmit the first allele every time and heterozygotes transmit it half the time, so the allele frequency in the next generation remains completely unchanged. Heritable variation never decays by blending, giving natural selection a permanent reservoir of diversity on which to act.

Now let natural selection enter the picture. Suppose the three diploid genotypes survive and reproduce at different rates. After one generation of selection acting on a population in Hardy-Weinberg proportions, the average reproductive output across the entire population is its mean fitness Wˉ\bar{W}. In the next generation, each allele changes in frequency in proportion to how much its marginal fitness exceeds the population mean. Differentiating the mean population fitness with respect to the allele frequency yields Sewall Wright's adaptive landscape equation for the single-generation change in allele frequency:

Δp=p(1−p)2WˉdWˉdp\Delta p = \frac{p(1 - p)}{2\bar{W}} \frac{d\bar{W}}{dp}

This compact equation reveals the geometric nature of adaptation. Because the genetic variance factor p(1−p)p(1 - p) and the mean fitness are never negative, the change in allele frequency always shares the exact sign of the fitness slope. Natural selection operates as a gradient-ascent algorithm that pushes allele frequencies steadily uphill on the surface of mean population fitness until the population reaches a local peak.

Depending on how the three genotype fitnesses compare, Wright's equation predicts two classic evolutionary regimes. In directional selection, one allele is consistently favored over the other in both heterozygotes and homozygotes. In continuous time, the allele frequency obeys a logistic differential equation whose S-shaped curve traces a selective sweep from a rare mutation all the way to complete fixation.

In balancing selection, by contrast, the heterozygote is fitter than either homozygote. Here the slope of mean fitness vanishes at an interior peak where both alleles coexist permanently. Both homozygotes suffer a fitness penalty relative to the heterozygote. Consequently, selection holds both alleles at a stable interior equilibrium determined by the ratio of their homozygote fitness costs.

In the early 1950s, the geneticist Anthony Allison showed that this equilibrium explains one of the most puzzling diseases in human medicine. Across equatorial Africa, the mutant sickle-cell hemoglobin allele remains at frequencies of 10% to 15%. It stays common even though children who inherit two copies of the sickle allele suffer severe anemia with an 80% fitness cost. The allele persists because individuals with two normal hemoglobin alleles suffer roughly a 10% fitness loss from lethal Plasmodium falciparum malaria. Meanwhile, heterozygous carriers enjoy strong resistance to malaria without developing anemia. Balancing selection holds the gene pool at the interior peak of one-ninth, or about 11%, trading a tragic genetic load in homozygotes for the survival of the wider population.

11.2Fisher's theorem, the Price equation, and the puzzle of altruism

In 1930, Ronald Fisher sought a law of evolution as universal as the second law of thermodynamics. Whereas entropy measures the inexorable decay of closed physical systems toward disorder, Fisher asked what quantity measures the upward progress of adaptive organization. In his book 'The Genetical Theory of Natural Selection', he proved the Fundamental Theorem of Natural Selection.

In a simple clonal population of asexual types, the single-generation increase in mean fitness equals the total genetic variance in fitness divided by mean fitness. In a sexual diploid population, however, Mendelian segregation and recombination break apart non-additive dominance and epistatic gene combinations every generation. As Fisher proved and George Price later clarified, natural selection increases mean fitness at a rate equal to the additive genetic variance in fitness, VarA(w)\text{Var}_A(w), divided by mean fitness. Here the additive genetic variance is the variance of transmissible breeding values across individuals. Because variance can never be negative, natural selection always drives the additive component of mean fitness upward at a pace set by standing heritable variation. When a population runs out of additive genetic variance in fitness, directional selection halts.

In 1970, an American physical chemist named George Price walked into University College London with no formal training in evolutionary biology. Working alone in a rented flat, he derived an equation that generalized Fisher's theorem to every evolving system. Price asked what mathematical relation must hold whenever any parent collection of entities gives rise to a descendant collection.

Assign each parent individual a numerical trait value, a reproductive fitness, and an average offspring trait shift. Using the definitions of statistical covariance and expectation, Price proved that the single-generation change in the average trait obeys an exact identity. From that covariance identity, William Donald Hamilton's rule for the evolution of altruism follows directly:

wˉ Δzˉ=Cov(wi,zi)+E[wi Δzi]⟹rB>C\bar{w} \, \Delta \bar{z} = \text{Cov}(w_i, z_i) + \mathbb{E}\left[ w_i \, \Delta z_i \right] \quad \Longrightarrow \quad r B > C

The Price equation cleanly splits every evolutionary change into two parts. The first term, the covariance between fitness and trait value, measures pure selection: how strongly the trait covaries with reproductive output. Choose the trait to be additive breeding value for fitness and hold transmission faithful. Then the covariance of breeding value with fitness recovers Fisher's Fundamental Theorem in a single line. The second term, the expected fitness-weighted trait shift, captures transmission bias from mutation, recombination, meiotic drive, or cultural learning.

Even more remarkably, Price realized that his equation unlocked the deepest riddle Darwin had ever faced. In 'On the Origin of Species', Darwin confessed that sterile worker castes in ants and bees initially struck him as fatal to his entire theory. These sterile insects never reproduce, yet they labor and die to defend the colony. How can natural selection ever favor a gene that lowers the personal reproductive fitness of its bearer to help another individual?

In 1964, Hamilton had proposed the concept of inclusive fitness, and Price's covariance algebra revealed its exact structure. Suppose an altruistic gene causes an actor to pay a personal reproductive cost in order to confer a fitness benefit on a social partner. Under unbiased Mendelian transmission, the Price equation shows when the altruistic gene increases in frequency. It spreads whenever the statistical regression slope of the partner's gene on the actor's gene exceeds the ratio of personal cost to partner benefit. That regression slope is the coefficient of genetic relatedness rr, yielding Hamilton's rule.

An allele for self-sacrifice spreads whenever the benefit to the recipient, weighted by genetic relatedness, exceeds the personal cost to the actor. Full siblings share half their genes by descent, while first cousins share one-eighth. J.B.S. Haldane once joked in a London pub that he would not jump into a river to save one drowning brother. Yet he would gladly lay down his life for two brothers or eight cousins. In ants, bees, and wasps, males develop from unfertilized haploid eggs and females develop from fertilized diploid eggs. When a queen mates with a single male, this haplodiploid sex-determination system causes full sisters to share identical paternal genes and half of their maternal genes. Full sisters therefore share an average relatedness of three-quarters. A female worker is more closely related to her full sisters than she would be to her own offspring, which share only half of her genes.

For George Price, the realization that even selfless kindness obeys a cold covariance algebra proved personally devastating. Determined to refute his own equation through radical charity, he gave away all his possessions to homeless strangers in London before taking his own life in 1975. The next year, Richard Dawkins wove Fisher, Hamilton, and Price together in 'The Selfish Gene', showing that individual organisms are transient vehicles built by immortal genetic replicators.

11.3Genetic drift, neutral evolution, and the molecular clock

All our equations so far have treated populations as infinite, so that allele frequencies follow deterministic trajectories. Real populations, however, are finite. A diploid population of effective size NeN_e contains twice that many gene copies, and each new generation is a finite random sample of gametes drawn from the previous one. Just as flipping a fair coin 10 times rarely yields five heads, random sampling introduces chance fluctuations into allele frequencies every generation, a phenomenon Sewall Wright called genetic drift.

Under the Wright-Fisher model, random binomial sampling across the gene pool leaves the expected allele frequency unchanged, but injects a single-generation sampling variance inversely proportional to the effective population size. In small populations, this random sampling noise can easily overpower weak natural selection.

In the 1950s and 1960s, the Japanese mathematical geneticist Motoo Kimura took the continuous-time limit of this discrete sampling process to derive the Fokker-Planck diffusion equation of population genetics. Kimura solved the backward diffusion equation with absorbing boundaries at extinction and complete takeover. His solution gives the ultimate probability of fixation for an allele starting at an initial frequency, along with the long-term rate of neutral molecular substitution:

u(p0)=1−exp⁡(−4Nesp0)1−exp⁡(−4Nes),K=(2Neμ0)(12Ne)=μ0u(p_0) = \frac{1 - \exp(-4 N_e s p_0)}{1 - \exp(-4 N_e s)}, \qquad K = \left( 2 N_e \mu_0 \right) \left( \frac{1}{2 N_e} \right) = \mu_0

Consider a brand-new mutation that arises in a single copy within a diploid population, so its initial frequency is one divided by twice the effective population size. Kimura's formula reveals two fundamental regimes. Everything depends on the dimensionless product of four times the effective population size and the selection coefficient, which measures selection relative to random drift.

When selection is much stronger than drift and the mutation is weakly beneficial, Kimura's fixation probability simplifies to twice the selection coefficient, recovering Haldane's rule. Even a beneficial mutation with a 1% selective advantage faces a 98% chance of vanishing by random drift in the first few generations. Only the remaining 2% survive the early lottery to sweep to fixation.

When selection is much weaker than drift, the mutation is effectively neutral. Taking the limit as the selection coefficient approaches zero shows that a newly arising neutral allele fixes with a probability equal to its initial share of the gene pool. In a diploid population, that initial share is simply one divided by twice the effective population size.

In 1968, Kimura multiplied that fixation probability by the total number of new neutral mutations arising across the population each generation. Suppose new neutral mutations occur at a constant rate per gene copy per generation. Multiplying that rate by twice the effective population size gives the total number of new neutral mutations entering the diploid gene pool each generation. Because each one has a probability of one divided by twice the effective population size of eventually fixing, the population size cancels out completely.

Because population size cancels out, in a vast population of bacteria, millions of neutral mutations arise every generation, yet each faces astronomical odds against fixing. In a small population of whales, very few mutations arise, but each enjoys a substantial chance of drifting to fixation. Across both extremes, the rate at which neutral DNA substitutions accumulate in an evolutionary lineage equals the raw neutral mutation rate per generation. This exact cancellation forms the bedrock of Kimura's neutral theory of molecular evolution. It also explains the molecular clock discovered by Émile Zuckerkandl and Linus Pauling, letting geneticists date the divergence of any two species directly from their DNA sequences.

11.4Evolutionary games and ritualized conflict

Until the 1970s, models usually assumed that a genotype's fitness is fixed by the external environment. In nature, however, the payoff of a behavior often depends on what everyone else is doing. A tree wins sunlight only relative to neighboring trees, and an aggressive animal risks injury only if its rival also chooses to fight.

In 1973, John Maynard Smith and George Price brought the mathematical game theory of John von Neumann and John Nash into biology. Instead of rational human players calculating optimal moves, evolutionary game theory lets natural selection adjust the frequencies of inherited behavioral strategies across generations. In the replicator equation of Peter Taylor and Leo Jonker, each strategy's per-capita growth rate is proportional to how much its expected payoff exceeds the population average.

Maynard Smith and Price defined a resident strategy to be an evolutionarily stable strategy if no rare mutant strategy can invade the population. Stability requires two conditions. First, the resident strategy must be a Nash equilibrium, earning at least as high a payoff against itself as any mutant earns against it. Second, whenever a mutant ties the resident against the resident, the resident must earn a strictly higher payoff against the mutant than the mutant earns against itself.

To see the power of this idea, Maynard Smith and Price analyzed the Hawk-Dove game of animal conflict over a territory or mate. Winning the resource brings a fitness gain, whereas escalating to an all-out physical fight inflicts a much larger injury cost on the loser. A Hawk always escalates until injured or victorious, earning a negative average payoff against another Hawk and the full prize against a Dove. A Dove displays peacefully, sharing half the prize when meeting another Dove and retreating unharmed with zero payoff the moment a Hawk attacks.

Because injury costs more than the prize, the Hawk-on-Hawk payoff is negative. A population of pure Hawks is unstable because a rare Dove mutant avoids brutal injuries and earns zero rather than a negative payoff. Yet a population of pure Doves is equally unstable, because a rare Hawk mutant steals the entire prize uncontested. Instead, selection drives the population to a stable interior balance where Hawks and Doves earn equal expected fitness. At that equilibrium, Hawks make up a fraction equal to the resource value divided by the injury cost. The deadlier the weaponry—antlers in stags, horns in bighorn sheep, or venom fangs in rattlesnakes—the larger the injury cost and the lower the equilibrium frequency of unrestrained violence. Restraint in animal combat evolved not for the good of the species, but because ritualized displays maximize the survival of each contender's own genes. Yet before stags or snakes could clash over territory, single-celled microbes first had to breach a steep bioenergetic barrier to build macroscopic multicellular bodies.