431785 Molecular Modeling to Design and Control (M2DC): A First Principles Approach to Polymorph Prediction and Crystallization Unit Design

Friday, November 13, 2015: 9:20 AM
Ballroom D (Salt Palace Convention Center)
Conor Parks1, Andy Koswara2, Ramon Pena3, Doraiswami Ramkrishna3, Zoltan K. Nagy3, Hsien-Hsin Tung4 and Shailendra Bordawekar5, (1)ChE, Purdue, West Lafeyette, IN, (2)Chemical Engineering, Purdue University, West Lafayette, IN, (3)School of Chemical Engineering, Purdue University, West Lafayette, IN, (4)Process Research & Development, AbbVie Inc., North Chicago, IL, (5)Process Research & Development, AbbVie, North Chicago, IL


Crystallization remains one of the most important industrial separation processes to date.  Despite its wide spread use, a fully fundamental understanding of the nucleation process is lacking. The classical nucleation theory (CNT) rate expression, formulated by Gibbs, Weber, and Volmer, at the beginning of the 20thcentury, remains the main framework against which experimental evidence is compared, despite being overly simplistic, and failing to capture data well in a range of systems. Skepticism towards the degree to which thermodynamic bulk properties used in the CNT rate expression can be even applied to nano scale clusters is warranted. Rather, with the advent of modern supercomputers, the possibility to relax the suit of CNT assumptions in favor of fully atomistic calculations provided via molecular dynamics is becoming a more and more appealing alternative. From a solid-state physics perspective, polymorphism, the ability of a molecule to exist in multiple crystalline phases, remains a thorn in the side of both industrial crystallization experts, and molecular modelers alike. Landscape energy minimization techniques, although elegant in their own right, fail to include standard design variables such as solvent, temperature, or pressure. Molecular dynamics, which encapsulates the aforementioned design variables naturally, has suffered from time scale challenges associated with activated rate processes, such as nucleation. However, with the advent of modern supercomputers, and HPC hardware such as xeon phi coprocessors, these types of simulations are becoming more and more feasible. The crystallization community would more than like molecular simulations to become primetime, because in the absence of in silico polymorph specific nucleation rates, expensive solvent screening experiments must be performed. Furthermore, the question of “have we truly found all possible forms” is never truly answered by such experiments.

In this work, we investigate the potential of atomistic information provided by hybrid monte carlo molecular dynamics (MCMD) to be used to not only predict polymorphism as a function of operation conditions, but also the degree to which polymorph specific solubility, and metastable zone curves can be calculated. These calculations will serve as inputs into control models, allowing a fully atomistic design of a crystallization unit. As microfluidic devices provide a means to measure homogeneous nucleation rates using minimal reagents and in minimal time, they will serve as a benchmark for determining the degree of accuracy of our homogeneous nucleation rate calculations prior to further scale-up for industrial applications.

Extended Abstract: File Not Uploaded