Wednesday April 12: On-The-Fly Heuristic Reordering Approach to Deterministic Optimization For Qualitative Chemical Property Prediction

Speaker

B. Christopher Rinderspacher
Army Research Laboratory

Title

On-The-Fly Heuristic Reordering Approach to Deterministic Optimization For Qualitative Chemical Property Prediction

Location

Wednesday, April 12th
1-2pm
307 Eberly Hall

Abstract

Chemical optimization and design affords current and future researchers the ability to harness the potential of chemical space towards the fast and efficient discovery of novel materials. In the present work, a multi-constraint deterministic optimization technique based on the on-the-fly heuristic reordering of chemical subspace has been developed and used towards materials discovery in several systems of interest. The competitive advantage of the deterministic optimization method results from the combination of fast computational techniques and innovative design algorithms which allow for intelligent screening of a large number of chemical compounds within a reasonable computational time.  A family of search algorithms has been used to approach the problem of navigating chemical subspace, including general base line search and general base gradient local search techniques. Because ideal ordering of the chemical subspace is not known, measures must be taken to ensure that the space is being sampled properly. There are several strategies utilized in this work to ensure this requirement is being satisfied. First, optimization algorithms based on heuristic reordering of the chemical subspace are developed to assist and direct the optimization procedure. The heuristic reordering algorithms play an essential role in optimization efficiency and each heuristic scheme has been uniquely developed as an effort to further enhance the subspace sampling. Each of these algorithms have been benchmarked and tested for their performance with respect to candidate structure discovery. In addition, to further combat the potential subspace sampling partiality, a binary entropic, enhanced sampling approach has been employed. This technique allows for generation and searching of structures which are chemically maximally different from the local best candidate structure. This is advantageous because a larger breadth of space is able to be sampled in an equally efficient manner. This method has been applied to several systems of interest including high-hyperpolarizability materials, energetic materials and optically switchable materials. Detailed analysis was performed for each of these systems and qualitative structure property relationships were determined.