Testing Lennard-Jones clusters for optimality

• Published in: The Journal of chemical physics 2023-07, Vol.159 (1)
• Main Author: Kiessling, Michael K.-H.
• Format: Article
• Language: English
• Subjects: Clusters; Data points; Monomers; Optimization; Search algorithms
• ISSN: 0021-9606; 1089-7690
• DOI: 10.1063/5.0158931

This note advertises a simple necessary condition for optimality that any list N ↦ vx(N) of computer-generated putative lowest average pair energies vx(N) of clusters that consist of N monomers has to satisfy whenever the monomers interact with each other through pair forces satisfying Newton’s “action equals re-action.” These can be quite complicated, as, for instance, in the TIP5P model with five-site potential for a rigid tetrahedral-shaped H2O monomer of water, or as simple as the Lennard-Jones single-site potential for the center of an atomic monomer (which is also used for one site of the H2O monomer in the TIP5P model, which in addition has four peripheral sites with Coulomb potentials). The empirical usefulness of the necessary condition is demonstrated by testing a list of publicly available Lennard-Jones cluster data that have been pooled from 17 sources, covering the interval 2 ≤ N ≤ 1610 without gaps. The data point for N = 447 failed this test, meaning the listed 447-particle Lennard-Jones cluster energy was not optimal. To implement this test for optimality in search algorithms for putatively optimal configurations is an easy task. Publishing only the data that pass the test would increase the odds that these are actually optimal, without guaranteeing it, though.