Designs

The equations of beamtracer.optimization.design, declared on the code that implements them and checked by the tests.

rank

functionsource

Return the sort key of the searches: lower is better.

rank

$$\text{rank} = \begin{cases} (0, f) & \text{feasible} \\ (1, v) & \text{ok but infeasible} \\ (2, 0) & \text{failed or unconverged} \end{cases}, \qquad v = \sum_i \max(g_i, 0)$$

Assumes

• ok: converged, no failure, finite objectives; feasible: ok and every constraint $g_i \le 0$; a NaN constraint counts as $v = \infty$
• sorted lexicographically, lower is better: a penalised evaluation never ranks before a feasible one