fig5

From: Robust global optimization of atomic structures via a learning loss-informed on-the-fly firefly algorithm

Figure 5. (A) Schematic illustration showing robustness to error in the target potential. In E_PBE, Structure B represents the global minimum, but with vdW corrections, Structure A becomes more stable. Conventional GO algorithms always tend toward Structure B, the minimum of the chosen target potential, whereas the FA explores both minima, capturing both A and B (shaded region); (B) Agents from the final generation from a single NARA active learning run are shown on the PES. Top-left structure: non-planar global minimum (capped trigonal prism). Bottom-left structure: planar global minimum (edge-capped elongated hexagon). vdW: Van der Waals; GO: global optimization; FA: firefly algorithm; NARA: nature-inspired algorithm for robust atomic structure search; PES: potential energy surface; PBE: Perdew, Burke, and Ernzerhof.