Several postdoctoral positions in the Materials Intelligence Research group at Harvard University are open to develop and apply first principles and machine learning methods for computational materials physics and chemistry. Applications include investigation and design of catalysts, soft materials, energy conversion and storage materials, power electronics and thermoelectrics.
The desired technical qualifications are experience with DFT or quantum chemistry calculations, method development and implementation of high-performance scientific software, with GPU capability, machine learning methods and automated computing workflows. Projects include:
Machine learning methods for large scale molecular dynamics. We are developing equivariant neural network models NequIP and Allegro for interatomic potentials, that advance the state of the art in accuracy and data efficiency. Efforts are aimed at learning computationally lean and geometrically rich representations, and designing methods for quantifying uncertainty of predictions. We are also developing Bayesian Force Fields in the FLARE framework that combines rigorous uncertainty in Gaussian process regression with active learning. Resulting models are implemented in LAMMPS and are used to perform reactive dynamics simulations of billions of atoms.
Electrical and thermal transport from first principles. We are developing methods for predicting electrical, thermal, and magneto-transport coefficients in semiconductors within the Boltzmann transport and Wigner transport formalism. Applications include thermoelectric materials and 2D systems. We implement these methods in the Phoebe software framework which relies on Wannier/Fourier interpolation of first-principles carrier spectra and couplings.
Machine learning of exchange-correlation functionals.. Current work in the group is focused on improvements and performance optimizations for the recently developed CIDER formalism for designing non-local XC functionals, with an eye toward applying the resulting functionals to currently intractable problems in catalysis and energy storage materials. Effort is aimed at generating training sets with high-order quantum calculations and designing combined models for the exchange and correlation energy as an explicit non-local functional of the electron density.
Simulations of catalysis reactions and ionic transport. The aim is to investigate large-scale and long-time evolution and discover mechanisms of reconstruction of surfaces and nanoparticles in reactive and solvated environments, ionic diffusion and phase transformations complex ceramic and polymer electrolytes. Methods will combine first principles, molecular dynamics, enhanced sampling and Monte Carlo simulations, with state of the art machine learning force field models.
Preferred start date as soon as possible but flexible.