A relatively well-known field in the literature, Computer-Aided Molecular Design (CAMD) defines a class of problems aimed at designing a single, optimal molecular structure to best fit some purpose. While CAMD1 has seen some success in the literature, its mixture analogue, Computer-Aided Mixture Design (CAMxD), still has some significant limitations. CAMxD is, unsurprisingly, much more complicated. This class of problems now involves the design of a blend of compounds – both every compound’s molecular structure and mole fraction in the composition – and this greatly increases the number of variables in the problem and adds many more nonlinear constraints in the form phase equilibria equations, miscibility checks, and safeguards against including two molecular species which might react.
The traditional approach to CAMxD in the literature has two main shortcomings: (1) many approaches rely on enumerating every possible structure for every possible component and then checking every combination for mixture feasibility; and (2) the search space is inherently curtailed because these approaches often solve the mixture part of the problem with UNIFAC2 method, which can only work when a regressed interaction parameter is known for every pair of molecular substructures in the blend. These issues have resulted in approaches to the CAMxD problem that can only investigate a very limited area of the molecular search space.
Our approach to the CAMxD problem circumvents these two issues by: (1) using a derivative-free optimization scheme to design for structures and test their mixture properties on-the-fly; and (2) deriving a group-contribution method to estimate COSMO-RS3 and COSMO-SAC4 sigma profiles, which can be used to calculate mixture properties without the need for binary interaction parameters. These two steps enable a much more efficient optimization of a much larger chemical search space. In this presentation, we will outline our approach in more detail and discuss how the COSMO group contribution methods were made. We will also demonstrate our algorithm as applied to a solvent design problem and a reaction rates optimization problem.
 Group-contribution based estimation of pure component properties. Jorge Marrero and Rafiqul Gani.Fluid Phase Equilibria. 2001. 183, 183-208.
 Group-contribution estimation of activity coefficients in nonideal liquid mixtures. Aage Fredenslund, Russel L. Jones, and John M. Prausnitz. AIChE Journal. 1975. 21(6), 1086-99.
 Fast solvent screening via quantum chemistry: COSMO‐RS approach. Frank Eckert and Andreas Klamt. " AIChE Journal. 2002. 48(2), 369-385.
 A priori phase equilibrium prediction from a segment contribution solvation model. Shiang-Tai Lin and Stanley I. Sandler. " Industrial & engineering chemistry research. 2002. 41(5), 899-913.