(620w) Quantum-Level Descriptors In Computational Molecular Design | AIChE

(620w) Quantum-Level Descriptors In Computational Molecular Design

Authors 

Chen, Q. - Presenter, The University of Kansas
Camarda, K. D. - Presenter, Washburn University
Bishop, K. - Presenter, The University of Kansas

    A fundamental need in chemical engineering is the ability to predict the properties of small molecules a priori, purely from 2-D structure. While group contributions are effective for some properties, other properties require knowledge of the topology of the molecule, as well as a quantum-level description of the bonding environments around each atom. However, current quantum-level simulations require a great deal of time to arrive at acceptable accuracies. By developing links between quantum-level simulation information and linearized property prediction models, quantum-level molecular information can be incorporated into optimization-based design algorithms for novel molecules. In this project, parallel quantum simulations were performed for a representative set of small molecules, which span the combinatorial space of interest for a solvent design problem. The results of the quantum computations were then used compute numerical descriptors, which were next correlated with sets of experimentally measured property values for properties such as solubility, toxicity, viscosity, and melting point.  The correlations were then combined with a set of structural constraints to form an optimization framework for molecular design. This optimization problem, an MINLP with some black-box constraints, can be solved to near-optimality by stochastic methods, including the Tabu search algorithm applied in this work. Examples are shown which present successful designs for both solvent and ionic liquid design tasks.