`A Genetic Algorithm for molecular structure optimization`

Nature article

Ames Laboratory

Fastest Real Application award at Supercomputing '95

Atomistic models of materials provide accurate total energies. Practical applications, however, often require extremely long time scale simulations. Structural optimization of an atomic cluster requires a simulated annealing run whose length scales exponentially with the number of atoms in the cluster. Here, the 60 atom carbon buckyball is formed from random coordinates in our computer simulation.

How we do it

The genetic algorithm. Inspired by the Darwinian evolution process, a population of structures is maintained. New generations are produced by `mating' clusters and selecting the lowest energy relaxed children.
Carbon clusters in a tight-binding model. We found fullerene cages starting from random coordinates, including the 60 atom buckyball illustrated above.
The mating procedure. We use a cut and mating operation. The parent clusters are split in half and the child is created from one half of each parent. Mutations may also be applied.

At left, we show some of the structures our algorithm generates for twenty carbon atoms. At right, the lowest energy structure is found only when mutations are added.

We have applied the physical cut-and- mating genetic algorithm to other systems, including point charges on a sphere (Thomson problem), lattice spin glass models in 2D and 3D, and Lennard-Jones clusters.

deaven@deaven.net