Research

Computation, data science and AI for energy materials.

We use computation, data science and AI to understand and design materials for energy and sustainability. Our current work follows three directions: solid-state ionics for batteries, porous materials for carbon capture and gas separation, and AI-driven multiscale modelling. The publication graph shows how our papers connect these themes.

Solid-state ionics and battery materials

Rendering of electrode and electrolyte particles in a solid-state battery

All-solid-state batteries replace the flammable liquid electrolyte with a solid fast-ion conductor, which can improve the safety and energy density of lithium- and sodium-based cells. We study what controls ion transport and stability in these materials, from the bulk to grain boundaries, coatings and interfaces. Recent work covers the aqueous stability of lithium halide and oxyhalide electrolytes, the effects of grain boundaries and surfaces on the electronic and mechanical properties of solid electrolytes, and sodium-ion transport in NaSICON. We combine density functional theory with cluster expansion, Monte Carlo and kinetic Monte Carlo sampling, machine-learning interatomic potentials, and phase-field modelling of dendrite growth.

Selected papers:

People: Shoutong Jin, Yan Li, Naibing Wu and Cheng Peng (team). See all battery papers.


Porous materials for carbon capture and gas separation

Selective CO2 capture in aluminium formate A mixed gas of CO2, N2 and CH4 from flue gas or biogas flows into the ReO3-type aluminium formate framework. CO2 molecules are held in the pore cavities while N2 and CH4 pass through. gas flow CO₂ N₂ CH₄ Flue gas or biogas N₂ and CH₄ pass CO₂ held in the pore cavities ALF, Al(HCOO)₃ · ReO₃-type framework
Schematic of selective CO2 capture by aluminium formate, after Das et al., Small (2026).

Aluminium formate, Al(HCOO)3 (ALF), is a low-cost porous metal formate with a ReO3-type framework. With experimental collaborators, we study how its hydrogen-confined pore cavities recognise CO2 over hydrocarbons, how temperature-regulated gating enables gas separations, and how hydrophobic composites and related metal formates improve CO2 capture. ALF has also been evaluated for noncryogenic air separation and hydrogen storage. Our calculations address adsorption sites, binding energies and adsorption kinetics.

Selected papers:

People: Fuzhi Li (Jason) (team). See all ALF papers.


AI and multiscale modelling

Modelling methods across time and length scales A staircase of five scales on time and length axes: quantum mechanics (DFT), molecular dynamics (machine-learning interatomic potentials, luMOD), mesoscale kinetics (Monte Carlo, kMCpy), continuum models (phase field, finite elements) and device level (digital twins). Arrows show energies and forces, barriers and rates, transport coefficients, and cell state passed from each scale to the next, organised by AI. fs ps ns µs s Å nm µm mm m Time Length AI-organised workflow energies, forces barriers, rates transport cell state Quantum DFT Atomistic MLIPs · luMOD Mesoscale Monte Carlo · kMCpy Continuum Phase field · FEM Device Digital twins
Methods across time and length scales, adapted from Jin et al., Small Struct. (2026).

The performance of a battery or a sorbent emerges across length and time scales, from electrons and atoms to devices. We develop methods that carry information between these scales: machine-learning interatomic potentials that reach the size and time scales of disordered materials, descriptors of local atomic environments such as luMOD, machine-learning models of thermodynamic stability, and AI-assisted workflows that link first-principles calculations to mesoscale and continuum models.

Selected papers:

People: Tzu-Chen Liu, Yanhao Deng and Jingyuan Lin (team). See all machine learning papers.

Software

kMCpy overview: model, structure and kinetic Monte Carlo results
  • kMCpy is an open-source Python package for computing ion transport in solids with kinetic Monte Carlo (paper, tutorial).
  • atombatch batches energy and force evaluations from many independent ASE tasks into shared model calls.
  • Wendao turns course notes into an interactive knowledge graph with a course-grounded AI tutor. It also powers our publication graph.

More code is on our GitHub.


Hybrid halide perovskites and optical materials

Crystal structures of hybrid halide perovskites

Our earlier work on hybrid halide perovskites focused on the discovery of lead-free compositions through the double perovskite approach. With experimental collaborators, we continue to study low-dimensional hybrid halides and germanium perovskite ferroelectrics for light emission and nonlinear optics.

Selected papers:

See all perovskite papers.