ut J

for i — 1,... ,N (e. g., van Kampen 1981). After inserting r(x , x) as derived above, this yields d dt

Xi Hi(Xi)bi(xi,x)rii(x) j si(xi:x)(Xi )MMi(xi,xi)dxi .

By expanding si(x'i, x) — max(0 , fi(xi, x))/bi(x'i, x) around xi to first order in xi, we obtain si(x'i,x) — max(0, (xi — xi)gi(x))/bi(xi,x) with gi,(x) —

dXfi(x'i,x) ; notice here that fi(xi,x) — 0. This means that in the xi-integral above only half of the total xi-range contributes, while for the other half the integrand is 0. If mutation distributions Mi are symmetric -

Mi(xi + Axi, xi) — Mi(xi — Axi, xi) for all i, xi,, and Axi - we obtain d^xi — ^Pi(xi)ni(x) j (xi — xi)T(xi — xi)Mi(x'i,xi)dx' gi(x) .

The integral is the variance-covariance matrix of the mutation distribution Mi around trait value xi, denoted by o~2(xi,). Hence we recover the canonical equation of adaptive dynamics (Dieckmann 1994; Dieckmann and Law 1996), d^xi — 1 Pi(xi )ni (x)af(xi )gi(x) for i — 1,... ,N. When mutational steps xi ^ xi are not small, higher-order correction terms can be derived: these improve the accuracy of the canonical equation and also cover non-symmetric mutation distributions (Dieckmann 1994; Dieckmann and Law 1996).

Polymorphic Deterministic Model. When mutation probabilities are high, evolution is no longer mutation-limited, so that the two classes of models introduced above - both being derived from the analysis of invasions into essentially monomorphic populations - cannot offer quantitatively accurate approximations of the underlying individual-based birth-death-mutation processes. Provided that population sizes are sufficiently large, it instead becomes appropriate to investigate the average distibution-valued dynamics of many realizations of the birth-death-mutation process, p(x) = j [p'(x) — p(x)]r(p',p)dp' .

Inserting the transition rates r(p' ,p) specified above for the individual-based evolutionary model, we can infer (by collapsing the integrals using the sifting properties of the Dirac delta function and of the generalized delta function)

-dtpi(x) = r+ (xi ,p) — ri (xi,p) for i = l,... ,N. Inserting r+ (xi,p) and r-(xi,p) from above, this gives

+ j pi(x'i)bi(x'i,p)pi(x'i)Mi(x'i,xi)dx'i — di(xi,p)pi(xi) .

Further analysis is simplified by assuming that the mutation distributions Mi are not only symmetric but also homogeneous - Mi(x'i + Axi,xi + Axj) = Mi(x'i,xi) for all i, xi, xi, and Axi. Expanding pi(x'i)bi(x'i,p)pi(x'i) up to second order in x' around xi, pi(x'i)bi(x'i ,p)pi(x'i) = pi(xi)bi(xi,p)pi(xi) + (xi — xi) Pi(xi)bi(xi,p)pi(x{) + 2 (xi — xi)T[ dx Pi(xi )bi(xi,p)pi(xi)](xi — xi), then yields d l d2

dtpi(x) = fi(xi,p)pi(xi) + 2a2(xi) * dx2 Pi(xi)bi(xi,p)pi(xi) , with fi(xi,p) = bi(xi,p) — di(xi,p), a2(xH) = ¡(xi — xH)T(xi — xi)Mi(xi,xi) dxi, and with * denoting the elementwise multiplication of two matrices followed by summation over all resultant matrix elements. This result also provides a good approximation when mutation distributions are heterogeneous, as long as a2(xi), rather than being strictly independent of xi, varies very slowly with xi on the scale given by its elements.

Was this article helpful?

0 0

Post a comment