Adaptive dynamics

Adaptive dynamics is a mathematical framework for modeling evolution as a dynamical process driven by the repeated substitution of mutant traits into a resident population. Developed by Odo Diekmann, Johan Metz, and collaborators in the 1990s, it bridges the gap between population genetics and evolutionary game theory by treating phenotypic traits as continuous variables and evolution as a trajectory through trait space.

The Core Idea

Adaptive dynamics begins with the assumption that a population is monomorphic — all residents share the same trait value x. The population is at an ecological equilibrium determined by x. A rare mutant with a slightly different trait y appears. The mutant's growth rate when rare, called the invasion fitness f(y,x), determines whether it can spread. If f(y,x) > 0, the mutant invades; if f(y,x) < 0, it is eliminated.

The key insight is that invasion fitness is a function of both the mutant trait y and the resident trait x. This creates a feedback loop: as successful mutants replace residents, the resident trait changes, which changes the selective pressure on future mutants. Evolution is not climbing a fixed fitness landscape; it is co-evolving with the landscape itself.

The Canonical Equation

When mutations are small and rare, the long-term dynamics of the resident trait can be approximated by a differential equation called the canonical equation of adaptive dynamics:

dx/dt = k(x) · ∂f(y,x)/∂y|_{y=x}

where k(x) is a mutation rate scaled by the variance of mutational effects. This equation says that the population evolves in the direction of the local fitness gradient, at a speed proportional to the mutational input.

The canonical equation is not a law of evolution; it is a scaling limit. It holds when mutations are sufficiently small that the population remains effectively monomorphic between invasion events — a regime called trait substitution sequences. When mutational steps are large or the population is polymorphic, the approximation breaks down and stochastic individual-based models become necessary.

Evolutionary Singularities

Points where the fitness gradient vanishes — where ∂f/∂y = 0 — are called evolutionary singularities. These are the candidate endpoints of evolution, but their classification is subtle. A singularity may be:

• Convergence stable — traits near it evolve toward it
• Evolutionarily stable — no nearby mutant can invade once the population is at the singularity
• Branching points — convergence stable but not evolutionarily stable, leading to disruptive selection and the emergence of polymorphism

The last case is particularly significant. A branching point predicts evolutionary divergence — the splitting of a single lineage into two distinct strategies. This provides a mechanistic foundation for speciation and phenotypic diversification without requiring geographic isolation.

Connections and Limitations

Adaptive dynamics makes several simplifying assumptions that limit its applicability. It assumes asexual reproduction, small mutational steps, and rare mutations. Extensions to sexual populations, frequency-dependent selection, and spatial structure exist but complicate the analysis. The framework also struggles with Evolvability — the capacity to generate viable mutations — which it treats as a parameter rather than an evolving property.

The deeper connection is to agent-based models: adaptive dynamics is the macroscopic approximation of microscopic evolutionary processes, just as the Navier-Stokes equations approximate molecular motion. When the approximations fail, the full individual-based model must be simulated. This two-scale structure — microsimulation and macroequation — is a recurring pattern in systems theory.

Adaptive dynamics reveals that evolution is not optimization on a fixed landscape but a dynamical system in which the landscape itself is a function of the current state. The implication is severe: any claim that evolution "optimizes" fitness is either false or true only in the trivial sense that the current state defines what "fitness" means. Evolution has no target; it has only a local gradient that disappears the moment you reach it.