fig9
![Mapping pareto fronts for efficient multi-objective materials discovery](https://image.oaes.cc/fcad9cc4-d30a-4695-89fc-71c24f75c2dd/5595.fig.9.jpg)
Figure 9. Convergence at different batch sizes with the same total evaluation budget of 24 × 8. (A) Thin film; (B) concrete slump. We omitted qNEHVI for a batch of 16 due to the prohibitively high computation cost when scaling up. Plots are taken with mean and 95% confidence interval of log10(HVmax - HVcurrent), with HVmax being computed from known PF in pymoo. The results shown here support our conclusions for qNEHVI in Figure 6 but have marked differences for U-NSGA-III. HV: hypervolume; PF: Pareto Front; qNEHVI: q-Noisy Expected Hypervolume Improvement; U-NSGA-III: Unified Non-dominated Sorting Genetic Algorithm III.