Evolution in a rugged 2d fitness landscape. A: The fitness landscape, obtained by sampling a Gaussian process with unit standard deviation and unit correlation length; the global fitness maximum is indicated by the green dot. It is a priori difficult to predict the path taken by a population evolving in this landscape. B: The quasispecies distributions \(Q\) for two different values of the mutation rate \(\mu\), localized at the fitness peak (low \(\mu\)) or in some lower but flatter region (high \(\mu\)). C: The effective potential \(U=-2\log Q\) for \(\mu\ =\ 0.05\) is much smoother than the fitness landscape, with few local minima which act as local attractors for an evolving population (black dots). Note how the population conspicuously moves away from the global fitness maximum.