A comparison of genetic algorithms and simulated annealing in maximizing the thermal conductance of harmonic lattices

Functions of chemical composition are complex and discrete in nature making it impossible to optimize them with gradient methods. Genetic algorithms, which do not use derivative information, are used to maximize the thermal conductance of one-dimensional classical harmonic oscillators constructed from a fixed library of randomly generated molecular units. The ability for the genetic algorithm to build structures with components having no physical increment is important in optimizing molecules from a library of unrelated polymer units. The performance of genetic algorithms in this problem is compared with simulated annealing. Hyper-parameters for these routines are selected from a grid search in order to optimize their expected solution strength. The solutions found via the genetic algorithm consistently outperform those of simulated annealing at the cost of longer computer time. Together, these algorithms are able to find thermal conductor candidates that mirror solutions in continuous media.