Machine learning for hidden order detection and large-scale material simulation
The quantum materials research landscape has entered a frontier enabled by exponentially growing data streams reflecting their complexity. Substantial progress within this complexity frontier has remained largely intractable to siloed researchers employing conventional materials physics tools. Instead, transdisciplinary team science is necessary, involving cutting-edge experiments and high-throughput theoretical modeling of quantum materials, combined with advanced machine learning (ML). Supervised and unsupervised learning techniques enable the predictive design, and provide understandable descriptions of quantum materials phenomena.
We aim to apply ML to quantum materials via two themes: Learning Models from Data, and Foundational AI for Materials. We will:
- Develop and realize ML-based model for experimental and theoretical spectroscopy, including circular (linear) dichroism, angle-resolved photoemission, raman spectroscopy, and neutron scattering, that will accelerate the detection of hidden order in quantum materials.
- Implement novel multi-scale computational modeling methods that maintain quantum accuracy through machine learning. Density functional theory (DFT) electronic structure calculations will be performed to obtain structural, electronic, and magnetic properties of prototype materials. First-principles-based effective Hamiltonians will be constructed by energy mapping analysis, linear response theory, and finally by machine learning. With the machine-learning energy model, atomic-scale modeling using the Molecular dynamics, Monte Carlo or Landau-Lifshitz equation will be employed to simulate equilibrium/non-equilibrium collective phenomena and evaluate thermodynamic properties.
