Figure 5.

Behavior of G over time. (A) Variance of trait one (${\sigma }_1^2,$ dashed line), its covariance with trait two (σ_1,2, dotted line), and its contribution to adaptation (${\mathit{\upsilon }}_1,$ solid line). Variances of ${\sigma }_1^2$ (1.307) and σ_1,2 (1.02) are around four times larger than the variance of ${\mathit{\upsilon }}_1$ (0.27) [units of $\mathit{\upsilon }{\left(U,N,s\right)}^2$]. (B) The magnitude of the angle between the second eigenvector of G and direction of selection [vector (1,1)] is on average 7.2° for the time period shown. Its grand mean is 0 and deviations from this mean orientation are usually small. (C) Normalized eigenvalues of G, ${\lambda }_2$ and ${\lambda }_1,$ measure genetic variation along the direction of selection and the perpendicular “neutral direction,” respectively. Normalization disguises the fact that $\overline{{\lambda }_1}\sim 5\times \overline{{\lambda }_2};$ as a result, it is fluctuations in ${\lambda }_1$ that drive those in variances, covariance, and angle. Fluctuations in ${\lambda }_2$ have a different cause and are uncorrelated. Spikes in ${\lambda }_1$ are due to enlargement of the high-fitness front, followed by its collapse (colored lines mark time points at which the 2D distribution is shown in Figure 7). Simulation parameters: s = 0.02, U = 10⁻⁵, and N = 10⁹.