(357h) Modeling Functional Nanoporous and Soft Colloidal Materials Using Molecular Simulations and Machine Learning
AIChE Annual Meeting
Meet the Industry Candidates Poster Session: Pharmaceutical Discovery, Development and Manufacturing Forum
Tuesday, November 15, 2022 - 1:00pm to 3:00pm
I am a computational scientist with expertise in multi-scale modeling of functional materials using advanced molecular modeling and simulation techniques. As a postdoctoral fellow, I have extensively worked with nanoporous materials including zeolites, amorphous materials, and metal-organic frameworks. I have experience with applying advanced computational algorithms to characterize pore geometries and architectures in nanoporous materials. I have utilized force-field-based Monte Carlo and Molecular Dynamics simulations to predict gas adsorption isotherms and diffusion coefficients.
My Ph.D. thesis focused on modeling the directed assembly of small clusters of soft colloidal particles for meta-material applications. I constructed thermodynamic process models for facilitating the assembly of colloidal clusters using advanced Monte Carlo simulations and machine learning algorithms. The work involved dimensionality reduction of large data sets containing particle trajectories into coarse-grained free-energy landscapes.
Molecular Modeling of Adsorption and Diffusion in Nanoporous Rigid Amorphous Materials: The development of publicly available materials databases for materials including zeolites (IZA-SC), metal organic frameworks (CoRE-MOF), and inorganic materials (The Materials Project) has been a key enabler of high-throughput computational screening and data-driven discovery of materials for potential use in new technologies. We recently introduced a similar database for structures of porous rigid amorphous materials1, an important class of materials for which no such resource was available. The database includes atomically detailed structures of disordered materials like amorphous carbons, kerogens, hyper-cross-linked polymers etc. generated using a wide range of simulation techniques by multiple research groups. We present extensive computational analyses for material characterization by calculating a series of scalar (e.g., accessible surface area) and vector (e.g., pore size distribution) descriptors. A variety of gas adsorption isotherms for both single component and binary mixtures are predicted for each structure. We also discuss the agreement between binary adsorption data and predictions from the Ideal Adsorbed Solution Theory.
In addition to adsorption isotherms, we have computed the diffusion coefficients of CH4 and CO2 in several of these structures.2 We present a diverse collection of molecular diffusivities in amorphous materials and examine their concentration dependence by comparing data from adsorption and diffusion simulations. We also discuss the correlations of computed self-diffusivities with scalar descriptors and Knudsen diffusivities.
Thermodynamic and Kinetic Models for Directed Assembly of Small Clusters of Colloidal Particles: Self and directed assembly of colloidal particles into crystalline objects is an emerging area of scientific interest that finds applications in manufacturing of photonic crystals and other meta-materials. Quantitatively accurate models of the thermodynamics and dynamics of these systems are essential to producing defect-free crystals. Robust methods for controlling the assembly of these crystals would require reduced dimension process-models that link the particle-level dynamics of the colloids to the actuator states. In this work, we describe the building of such models for two systems comprising 10 â100 micron sized silica particles in aqueous solution that employ either a temperature-tunable depletion interaction potential3 or externally applied electric field4 as a mechanism to promote the assembly process. We model the assembly process using coarse-grained representations, based on the Fokker-Planck equation, which can capture both the dynamics and the equilibrium properties of these small clusters. We use diffusion maps (DMaps), a machine learning technique to identify the slow, low-dimensional manifolds in these systems. The DMap coordinates are correlated against a set of candidate order parameters (OPs) to identify a suitable choice of observables. The DMap technique is sensitive to the nature of defects observed in these two systems and this is manifest in the correlations with OPs. We construct free energy and diffusivity landscapes in the chosen OPs that serve as reduced order models for process control policy maps providing an optimal route to defect-free crystals.
- Thyagarajan, R.; Sholl, D. S. A Database of Porous Rigid Amorphous Materials. Chem. Mater. 2020, 32, 8020-8033.
- Thyagarajan, R.; Sholl, D. S. Molecular Simulations of CH4 and CO2 Diffusion in Rigid Nanoporous Amorphous Materials. J. Phys. Chem. C 2022, 126, 8530-8538.
- Bevan, M. A.; Ford, D. M.; Grover, M. A.; Shapiro, B.; Maroudas, D.; Yang, Y.; Thyagarajan, R.; Tang, X.; Sehgal, R. M. Controlling assembly of colloidal particles into structured objects: Basic strategy and a case study. J. Process Control 2015, 27, 64-75.
- Yang, Y.; Thyagarajan, R.; Ford, D. M.; Bevan, M. A. Dynamic colloidal assembly pathways via low dimensional models. J Chem Phys 2016, 144, 204904.