Complementary Variety: When Can Cooperation in Uncertain Environments Outperform Competitive Selection?

Figure 5

Space-time fitness matrixes of Japanese exports 1974–2010. Fitness values are calculated as geometric means based on annual averages. Horizontal types are based on Figure 4 (see for full names of types). Vertical environmental states are identified by superior relative fitness of a type within a squared split (see text). Fitness optimization suggests placing resources only at the shaded types, while nothing should be bet on the nonshaded ones. Optimal population shares were calculated with constrained nonlinear optimization algorithms in R (NLoptr). The values we calculated with the Augmented Lagrangian Algorithm, using the Broyden-Fletcher-Goldfarb-Shanno method (auglag-lbfsg), and confirmed with the Sequential Quadratic Programming method (auglag-slsqp), as well as with Sequential Quadratic Programming (slsqp).