EDUCATION
University of Wisconsin-Madison Ph.D. candidate in Chemistry, 2015 – current Sogang University, South Korea M.S. in Chemistry, 2010
Sogang University, South Korea B.S. in Chemistry and B.E. in Integrated Biotechnology, 2008
RESEARCH DESCRIPTION
Supervised machine learning for prediction of phase boundary of 3D complex systems
Supervised machine learning (ML) has recently shown its great potential for identifying the phase boundaries or phase diagram of complex physical systems. I propose a new scheme of convolutional neural network (CNN), that is, a way to utilize grid-interpolated coordinates of molecules as input data of ML and optimize some hyperparameters of CNN model. We test our method for obtaining phase separation boundaries of off-lattice toy models: the Widom-Rowlinson model and a symmetric freely jointed polymer blend. The ML result shows excellent agreement with phase boundaries obtained from previous simulation studies. We believe that our CNN approach could be widely used to study phase behavior of condensed matters which may not be feasible for molecular simulations studies.
Figure 1. Pictorial scheme for predicting phase boundary of 3D off-lattice models
Critical point prediction of polymer solution using hybrid of feed forward neural network and molecular simulation method
I study methylation effects on the phase separation behavior (or LCST) of polymer in ionic liquids; poly(ethylene oxide) (or PEO) and anion [BF4] with various imidazolium cations, [CnMIM], and [CnMMIM], that is, methylated [CnMIM] cation to hinder the C2 hydrogen bonds between PEO and IL cation. For fast and accurate simulation data, I develop the symmetry-adapted perturbation theory united atom (SAPT-UA) force field and a neural network model trained by mixture properties. As a result, the predicted phase boundaries of PEO/ILs shows C2 methylation lowers critical temperature, which is consistent with the experimental result. Also, [CnMMIM] can drive that PEO chains are likely to not only wrap single cation and but also form crown ether shape near cation, which represents entropic driven LCST behavior of PEO/[CnMMIM] [BF4].
Figure 2. Phase diagrams of PEO/ILs from our ML work and from an experimental study of Lee, et al. in Macromolecules 2012, 45, 8, 3627-3633
SELF-INTRODUCTION
I am interested in employing modern data-driven techniques such as machine learning to solve tough questions on polymer science in such a creative way; currently, I am looking for my next career to enjoy machine learning applications with chemistry knowledge. I like to travel a new place for a long day to learn history such as Route 66 and Chicago skyscraper architectures and put unseen scenes in my eyes.