UCLA

Global optimization techniques for molecules, solids, and clusters are numerous and can be algorithmically elegant. Yet many of them are time-consuming and prone to getting trapped in local minima. Among the available methods, Coalescence Kick (CK) is attractive: it combines a nearly insulting simplicity with thoroughness. A new version of CK is reported here, called Adaptive Force-Field-Assisted Coalescence Kick (AFFCK). The generation of stationary points on the potential energy surface is tremendously accelerated as compared to that of the earlier, pure ab initio CK, through the introduction of an intermediate step where structures are optimized using a classical force field (FF). The FF itself is system-specific, developed on-the-fly within the algorithm. The pre-computed energies resulting from the FF step are found to be surprisingly indicative of energies in subsequent Density Functional Theory optimization, which enables AFFCK to effectively screen thousands of initial CK-generated structures for favorable starting geometries. Additionally, AFFCK incorporates the use of symmetry operations in order to enhance the diversity in the search space, increase the chance for highly symmetric structures to appear, and speed up convergence of optimizations. A structure-recognition routine ensures diversity in the search space by preventing multiple copies of the same starting geometry from being generated and run. The tests show that AFFCK is much faster than traditional ab initio-only CK. We applied AFFCK to the search for global and low-energy local minima of gas-phase clusters of boron and platinum. For Pt8 a new global minimum structure is found, which is significantly lower in energy than previously reported Pt8 minima. Although AFFCK confirms the global minima of B5(-), B8, and B9(-), it proves to be less efficient for systems with nontrivial bonding.