Honors and Awards

• AIChE’s Alpha Chi Sigma Award, 2024
• BASF Distinguished Lecturer, Wayne State U., 2021
• SEAS Distinguished Teacher Award, Princeton U., 2020
• Robert L. Pigford Memorial Lecturer, U. of Delaware, 2018
• Keith E. Gubbins Inaugural Lecturer, N. Carolina State U., 2016
• Fellow, American Institute of Chemical Engineers, 2014
• Chemical Engineering Distinguished Lecturer, Texas A&M at Qatar, 2013
• American Academy of Arts and Sciences, 2012
• National Academy of Engineering, 2004
• J.M. Prausnitz Award in Applied Chemical Thermodynamics, 1998
• Allan P. Colburn Award, American Institute of Chemical Engineers, 1995
• Teacher-Scholar Award, Camille and Henry Dreyfus Foundation, 1992
• Presidential Young Investigator, National Science Foundation, 1989

Affiliations

• Associated Faculty, Princeton Institute for Computational Science and Engineering
• Associated Faculty, Princeton Materials Institute

Research Interests

Research in our group focuses on development and application of theoretical and computer simulation techniques for the study of properties of fluids and materials. Emphasis is on molecular-based models that explicitly represent the main interactions among microscopic constituents of a system. These models can be used to predict the behavior of materials at conditions inaccessible to experiment and to gain a fundamental understanding of the microscopic basis for the observed macroscopic properties. Our work usually requires large-scale numerical calculations involving a number of powerful molecular simulation methodologies. An example of such a methodology is Gibbs ensemble Monte Carlo, which provides a direct way to obtain coexistence properties of fluids from a single simulation.

Ab initio derived machine-learning models. The group use the “Deep Potential” molecular dynamics (DPMD) approach of Car, E, and coworkers [Phys. Rev. Lett. 120:143001 (2018)] to develop force fields for fluids. The approach involves selecting a number (hundreds to thousands) of configurations for a system at thermodynamic state points representative of the conditions of interest. Kohn-Sham density functional theory is then used to obtain the “reference” energies of the corresponding configurations. For every configuration in the training set, a local coordinate frame is set up for every atom and its neighbors inside a smooth cutoff radius to allow for translational and rotational symmetries to be preserved. Permutational invariance is maintained by summing over all possible permutation of atoms of the same type. Then, these “descriptor” representations serve as inputs to a deep neural network to undergo linear and nonlinear transformations and output the energy for every atom. A molecular dynamics simulation is then run using the resulting potential energy surface, generating new configurations; their reference energies are evaluated. If differences between reference energies and the neural network model exceed specified thresholds, the network parameters are adjusted appropriately and the procedure is repeated.

A fundamental shortcoming of many machine learning models is that most such models only take into account interactions within a localized atom-centered spherical representation with a pre-defined radial cutoff. In a recent development, we have been developing methods for incorporation of long-range interactions into such models, as shown in the figure below, showing the training error in system energies for a finite-range (blue) and long-range (red) version of a model.

Electrolyte systems play an important role in chemical engineering separations, and also in geochemical environments and for biophysics. The mean ionic activity coefficients quantify the deviation of salt chemical potential from ideal solution behavior; experimental measurements are available for many salts over broad ranges of concentration and temperature, but there have been practically no prior simulation studies of these quantities, because if sampling difficulties for explicit-solvent electrolyte solutions. We are developing new methods for determination of properties of aqueous electrolytes and using them to improve the models for water and ions, by incorporating polarizability in the intermolecular potential models. For example, we have recently used forward-flux-sampling and metadynamics methods to obtain insights on the nucleation rates and pathways for salt crystallization from supersaturated aqueous solutions and to compare different salt and water models with respect to their ability to describe experimental measurements, as shown in the figure. We are also exploring the properties of molten carbonate electrolytes, which have applications to high-temperature fuel cells that can be used to separate CO₂ for carbon sequestration.

Liquid-liquid phase transitions in disordered proteins. Phase transitions within the intracellular environment help organize and regulate biochemical processes necessary for biological function. In collaboration with the Brangwynne group, we utilize grand canonical Monte Carlo and molecular dynamics simulations for simple coarse-grained models of disordered proteins to systematically investigate how sequence distribution, hydrophobic fraction and chain length influence phase behavior and regulate the formation of finite-size aggregates preempting macroscopic phase separation for some sequences. We show that a normalized sequence charge decoration parameter establishes a "soft" criterion for predicting the underlying phase transition of a model protein. We find that the overall phase separation propensity of sequences increases with the chain length and the hydrophobic fraction. At sufficiently long chain lengths, a vast majority of sequences are expected to phase separate. We hypothesize that this effect might contribute toward the ubiquity of phase separation and consequent, the relative rarity of aggregation behavior within cells.