In modern chemistry, computational methods can replace costly experimentation and measurements to predict molecular structure properties or to check the results of experimental work. Integrated molecular and process design aims at finding the best molecule for applications in chemical or energy engineering.

A major challenge is the representation of molecules in a machine-readable way so that integrated design can be formulated as a mathematical optimization problem. One has to find a trade-off between accurate representations and reasonable computation time. Simplified representations lead to loss of information both in the optimization process and the interpretation of the results.

The authors introduce a new molecular representation where the full structural information of the molecule is available during the optimization. Coupled with advanced property prediction methods, they find the optimal working fluid for a high-temperature organic Rankine cycle. The discrete nature of molecules implies integer variables. Combined with the process model described by continuous variables and nonlinear equations, this results in a mixed-integer nonlinear programming (MINLP) problem.

Artelys Knitro was used to solve the resulting problem with its implementation of the branch and bound algorithm and the sequential quadratic programming (SQP) algorithm. In addition, the multistart feature has been crucial to find the best solutions to this nonconvex model. The researchers’ approach coupled with Artelys Knitro enables high-fidelity property predictions and can be used in process design problems with complex non-linear target functions and constraints.