What to do with statistical mechanics when the figure of merit cannot be calculated: library design for high-throughput materials development

摘要

By analogy with Monte Carlo algorithms, we discuss new strategies for design and redesign of libraries in high-throughput experimentation, or combinatorial chemistry. Several Monte Carlo methods are examined, including Metropolis, several types of biased schemes, and composite moves that include swapping or parallel tempering. Among them, the biased Monte Carlo schemes exhibit particularly high efficiency in locating optimal compounds. The Monte Carlo strategies are compared to a genetic algorithm approach. Although the best compounds identified by the genetic algorithm are comparable to those from the better Monte Carlo schemes, the diversity of favorable compounds identified is reduced. Applications to materials discovery, small molecule discovery, and templated materials synthesis are discussed.