Research Digest — 2026-09-19¶
Solid Polymer Electrolytes & Dendrite Growth¶
1. Phase-field simulation and machine learning-assisted prediction of dendrite growth and crack damage in solid polymer electrolytes¶
Source: Journal of Energy Storage · 📅 2026-09-01 · ↗ Open paper
Develops a multiphysics-coupled phase-field model describing coupled dendrite growth and crack damage evolution in solid polymer electrolytes, with machine learning used to assist property prediction. Coupled temperature-pressure loading reduces maximum dendrite height by 62.2% and maximum lithium deposition by ~28%.
Relevance to DENG.Group
Highly relevant to both Naibing Wu (polymer electrolyte simulation) and Shoutong Jin (phase-field dendrite growth): combines the group's two active threads and shows how ML surrogates can accelerate phase-field parameter exploration under coupled thermal-mechanical loading.
Machine Learning Interatomic Potentials¶
2. Truncated automatic sparse differentiation for machine learning interatomic potentials¶
Source: arXiv · 📅 2026-09-17 · ↗ Open paper
Introduces a sparse, truncated automatic differentiation scheme to compute Hessians and higher-order derivatives directly from MLIPs at low cost, enabling routine vibrational analysis (phonons) from learned potentials without dense second-derivative evaluations.
Relevance to DENG.Group
Two direct hooks: Yanhao Deng can extend his MLIP workflow for solid electrolytes to phonon/lattice-dynamics observables, and Mengke Li can use cheap Hessians from MLIPs for the thermal-property component of his halide electrolyte work.
3. ALIGNN 2.0: A Unified Line-Graph Neural Network Framework for Materials Screening, Force Fields, Inverse Design, Spectroscopy, and Microscopy¶
Source: arXiv · 📅 2026-09-16 · ↗ Open paper
Rewrites the ALIGNN line-graph neural network as a dependency-free pure-PyTorch framework unifying property prediction, force fields, inverse design, spectroscopy, and microscopy in one codebase, removing the split between property and potential graph pipelines.
Relevance to DENG.Group
Useful reference for Yanhao Deng's MLIP development: a single framework spanning screening-to-potentials reduces the tooling overhead of training group-specific models for solid electrolytes and interfaces.
4. A budget-dependent crossover between coverage- and response-based training-set selection for machine-learned interatomic potentials¶
Source: arXiv · 📅 2026-09-05 · ↗ Open paper
Compares structural-diversity (coverage-based) against model-disagreement (response-based) training-set selection for MLIPs and shows the optimal strategy depends on the data budget: coverage wins at small retained sets, response-based selection takes over at larger budgets.
Relevance to DENG.Group
Practical guidance for Yanhao Deng's MLIP training on solid electrolytes: when building compact DFT training sets under a fixed computational budget, this defines when to prioritize diverse structures vs. configurations where candidate models disagree.
Potentials for Electrochemical Interfaces¶
5. Voltage-embedded equivariant machine learning potential for open system simulations¶
Source: arXiv · 📅 2026-09-04 · ↗ Open paper
Embeds electrode voltage explicitly into an equivariant ML potential, enabling open-system atomistic simulations of electrochemical interfaces under realistic non-equilibrium bias conditions rather than closed-cell approximations with homogeneous fields.
Relevance to DENG.Group
Directly applicable to Umang Agarwal's electrode/electrolyte heterogeneous interface problems: voltage-aware potentials would let the group simulate interfaces at realistic operating potentials instead of fixed-charge approximations.
Halides & Thermal Transport with ML Potentials¶
6. Strong Impact of Halide Ordering on Structural Phase Transitions in Mixed Perovskites¶
Source: arXiv · 📅 2026-09-09 · ↗ Open paper
Trains machine-learned interatomic potentials on DFT data to resolve how local halide ordering controls structural phase transitions in mixed halide perovskites, showing that cation/anion ordering dramatically shifts transition temperatures and mechanisms.
Relevance to DENG.Group
Order-disorder physics in halide frameworks speaks directly to Yan Li and Mengke Li's halide electrolyte work (halide sublattice ordering affects transport and degradation), while demonstrating the MLIP methodology Yanhao Deng uses for chemically disordered phases.
7. Origins of Pressure-Enhanced Thermal Transport in Organic Semiconductors¶
Source: arXiv · 📅 2026-09-13 · ↗ Open paper
Combines machine-learned potentials with the Wigner transport equation to compute pressure-dependent thermal conductivity of crystalline naphthalene, showing pressure stiffens the lattice while suppressing anharmonic scattering, yielding enhanced thermal transport.
Relevance to DENG.Group
Methodological blueprint for Mengke Li's thermal-property studies of halide electrolytes: the MLIP + Wigner transport toolbox is directly transferable to pressure-dependent thermal transport in solid electrolytes, complementing the stack-pressure theme of this cycle's interface papers.