Appendix 3—figure 26. Deviations from Lande’s approximation as a function of time after the shift.

In the Lande case, i.e., when $C\ll 1$ , the mean distance from the new optimum is well approximated by $D_L(t)$ . Deviations increase with shift size and extent of polygenicity (as seen, e.g., in the red curve), primarily due to a moderate increase in phenotypic variance during the rapid phase. The increase in the 3^rd phenotypic moment is minimal, which is why, in this case, there are no substantial long-term deviations. In the non-Lande case, the distance from the optimum decays faster than predicted by Lande’s approximation ( $D(t)-D_L(t)<0$ ) during the rapid phase, because of the increase in phenotypic variance. The increase in the 3^rd central moment during the rapid phase leads to a slower decay than predicted by Lande’s approximation during the equilibration phase. Even in the non-Lande case, however, we always find the distance from the optimum during equilibration to be smaller than $\delta$ . The simulation results were generated using the all allele simulation, as described in Section 2.1.