Current research

I am advised by Prof. Simon Billinge at Columbia University. My research topic, titled “ActionGraph,” attempts to build an open-source database infrastructure for time-series materials synthesis data, intended for integration with self-driving labs. I am expected to graduate in May 2025 with the research goal of using graph neural networks to address the inverse problem of predicting simulated synthesis recipes, constrained by inputs (ingredients) and outputs (physical properties). More updates to come.

Reseach Area 1. First principles & computational materials science

1. Raman mode quasiparticle frequencies across the H-bond symmetrization in δ-AlOOH

During the first year of my MS program advised Dr. Renata Wentzovitch and with guidance from Chenxing Luo, I validated the DeePMD-trained force field by generating phonon frequencies for AlOOH using PhonoLAMMPS, Dynaphopy, and Phonopy. I also automated a series of necessary computations using Snakemake and Bash on HPC.

Abstract: δ-AlOOH is a prototypical deep mantle hydrous phase that undergoes a pressure-induced H-bond disorder-symmetrization transition from P21nm to Pnnm at 300 K and approximately 9.0 GPa. This second-order (or near second-order) transition is characterized by significant anharmonicity. The precise transition pressure remains somewhat uncertain in static compression studies, but Raman frequency measurements provide valuable insights, revealing mode softening that cannot be fully captured by harmonic phonon calculations. In this study, we combine molecular dynamics and phonon calculations using DeePMD to obtain phonon quasiparticle properties, enabling us to examine the softening of Raman mode frequencies across the H-bond disorder-symmetrization transition.

Keywords: phonons, anharmonicty, deep potential, LAMMPS, Snakemake, HPC


2. Monte Carlo geometry optimization and first-principles nanoparticles study

From January 2022 to May 2023, I conducted three semesters of independent computational chemistry research with Dr. Robert Topper at the Chemistry Department at The Cooper Union. I was involved in the group’s focus on using theoretical and computational tools to study the formation behavior of salt nanoparticles. I focused on the formation behavior of ammonium chloride solvated in two to eight water molecules.

For geometry optimization, the group developed an open-source geometry optimizer called “TransRot”. I validated the accuracy and reliability of the open-source software by implementing water models such as TIP4P and TIP4P/2005. I used the TransRot software and conducted ab initio calculations in Psi4, ORCA, and SPARTAN to study the formation behavior with ammonium by determining the binding energy as one water molecule is added to the complex.

Book chapter | Poster | GitHub

Keywords: optimizer, Monte Carlo, ab initio, TIP4P, Psi4, ORCA


Research Area 2. Data-driven materials discovery

1. High-throughput geometric crystal featurizer for .cif files

Crystal data extraction schematic from raw CIF files and crystal databases by Dr. Anton Oliynyk

In Summer 2023, in collaboration with Dr. Anton Oliynyk, we developed an open-source Python tool crafted for the high-throughput extraction of crystal structure features. Traditional ML models in solid-state materials have predominantly relied on composition data to predict properties. The lack of structural information in the model often leads to incomplete property mapping and predictive noise

This geometric featurizer systematically processes raw Crystallographic Information Files (CIF) to construct supercells and meticulously extracts a comprehensive suite of descriptors, such as coordination numbers, interatomic distances, and atomic environments. These descriptors are not mere extrapolations from generic values but are tailored to the material’s specific structure, reducing noise and enhancing the relevance of ML applications for binary and ternary compounds. The package has demonstrated its robustness, having processed more than 10,000 binary and ternary CIF files. Leveraging structure-property correlations, we believe this crystal structure featurizer can be a promising tool for increasing predictive capabilities for material property optimization.

GitHub | Journal of Alloys and Compounds

Keywords: CIF, geometric descriptors, crystal structure, feature engineering, solid-state


2. ML descriptors for materials chemistry with multiple experimentally validated studies

Crystal prediction with compositional data

Materials informatics employs data-driven approaches for analysis and discovery of materials. Features also referred to as descriptors are essential in generating reliable and accurate machine-learning models. While general data can be obtained through public and commercial sources, features must be tailored to specific applications.

Common featurizers suitable for generic chemical problems may not be effective in features-property mapping in solid-state materials with ML models. Here, we have assembled the Oliynyk property list for compositional feature generation, which performs well on limited datasets (50 to 1,000 training data points) in the solid-state materials domain. The dataset contains 98 elemental features for atomic numbers from 1 to 92, including thermodynamic properties, electronic structure data, size, electronegativity, and bulk properties such as melting point, density, and conductivity. The dataset has been utilized peer-reviewed publications in predicting material hardness, classification, discovery of novel Heusler compounds, band gap prediction, and determining the site preference of atoms using machine learning models including support vector machines, random forests for classification, and support vector regression for regression problems. We have compiled the dataset by parsing data from publicly available databases and literature and further supplementing it by interpolating values with Gaussian process regression (GPR).

Data in Brief | Download dataset

Keywords: materials infomatics, feature engineering, GPR, elemental database, ML