Hello.
I am Xiang Zhang.
My research is in ab initio materials simulations and ML.
Here's my research at MIT, Oxford, and more.
Pathological behavior during the geometry relaxation of bare PbS quantum dots
Off-stoichiometric ligand-less wulff-structure polyhedrons, the popular computational toy model for QDs, aren't necessarily theoretically solid.
In accordance with their unphysicalness, we've noticed pathological behaviors, such as multiple local energy minima and phase-change-like behaviors for off-stoichiometric QDs.
We tried to characterize and quantify such behavior through "XRD"s and anomaly detection algorithms to understand and insure against the unphysicalness of the computational model.
Neural network potential (NNPs) for PbS quantum dots
Quantum dot DFT calculations are expensive; NNPs can help. We attempted both the proven Behler-Parrinello approach and the more recent Graph Neural Network approach in the search for NNPs for PbS QDs, unfortunately without much success.
A much less popular Cluster Expansion approach, and to a lesser degree recurrent neural networks, are able to make meaningful predictions on both ab-initio forces and geometry optimizations.
A more fundamental understanding of the bias and variances of summation layers is still needed. [Project summary 1][2]
A toolkit, framework and application for automating DFT calculations
After 6+ years of VASP experience, I envisioned my own workflow management system, one that doesn't fall clumsily behind nimble, explorative, data-centric computational research.
Up from a few population analysis shell scripts, to a Python library that represents unit cells, handles INCAR logic, prepares files, and talks to clusters, paying special attention to where existing solutions like ASE fall short.
And beyond, to a framework that autonomously executes pre-planned static computation graphs of VASP runs, and a production-ready Flask-backended web GUI.
Currently, I'm refactoring towards a more modern, lightweight, object-oriented version 2.0.0. [Design Doc]
Novel oxidation states in planar La2MnRhO6-x and Sr2FeIrO4
In a jointed experimental and theoretical study, we sought to understand the novel oxidation states observed in synthesized La2MnRhO6-x and Sr2FeIrO6-x. The unit cell geometry, charge distribution and magnetic ground state were found to depend strongly on the stoichiometry. After I left, the project culminated in a series of papers. [La2MnRhO6-x] [Sr2FeIrO6-x]
