Appendix 3—figure 19. The non-Lande case: As predicted by the linear approximation, the long-term contribution, per unit mutational input, to the change in mean phenotype coming from standing variation and new mutations scales with $(1+C)\cdot \mathrm{\Lambda }/V_A(0)$ (Equation A3-83).

Both the linear and nonlinear approximations perform fairly well. However, when the polygenicity is low and the shift is large, the nonlinear approximation overestimates the proportion of the change in mean due to large effect standing variation (Appendix 3—figure 18) and therefore overestimates the proportion contribution coming from all large effect alleles (see the red box in A). The simulation results were generated using the all allele simulation, as described in Section 2.1. We calculate the relative contribution of alleles in each effect size squared bin $(a_1^2,a_2^2]$ (between the gray gridlines) by dividing the contribution of fixations in the bin by the total mutational input per generation times the size of the bin, $2NU{\int }_{a_1}^{a_2}g(a)da$ . The analytic predictions were based on the linear non-Lande (Equation A3-83 or Equation 28 in the main text) and nonlinear non-Lande (Section 5.4.3) approximations with the corresponding parameters.