Skip to content

Research Digest — 2026-05-19

ML Interatomic Potentials & Free Energy

1. Reweighting Free Energy Profiles Between Universal Machine Learning Interatomic Potentials for Fast Consensus Building

Source: arXiv:2605.15630 · 📅 2026-05-15 · ↗ Open paper

Presents a systematic framework for reweighting potential-of-mean-force (PMF) profiles, initially sampled with a single source MLIP, across multiple target MLIPs. Uses robust analytical corrections (mean energy-gap approximation) to bypass statistical collapse when phase-space overlap between potentials is critically low. Demonstrated on a 601-atom Li⁺ transport system in a nanoconfined electrolyte, recovering high-fidelity thermodynamics across PBE+D3, PBE-sol, r²SCAN, and r²SCAN-D4 reference levels at a fraction of the cost of full simulations.

Relevance to DENG.Group

Highly relevant to Yanhao Deng's MLIP development. The reweighting framework enables affordable cross-model consensus on ion-transport free energy barriers without redundant DFT simulations. The Li⁺ nanoconfined-transport application directly parallels the group's work on Li⁺ migration in grain boundaries and interfaces of solid electrolytes. The finding that MLIPs partition into distinct clusters based on training data has implications for how the group selects and validates potentials for battery materials.

Source: arXiv:2605.14154 · 📅 2026-05-13 · ↗ Open paper

Proposes TSAgent, an agentic workflow that automates transition-state search at DFT-level accuracy through a persistent plan-execute-analyze-replan loop. Evaluated on 100 examples from the OC20NEB heterogeneous catalysis benchmark, TSAgent achieves 83% success rate. In direct comparison against expert DFT practitioners on 10 held-out examples, TSAgent achieves 70% success versus human-expert average of 73±12%. Independently reproduces Brønsted-Evans-Polanyi scaling relationships for NH₃ dissociation on metal and single-atom alloy surfaces.

Relevance to DENG.Group

Directly relevant to Yanhao Deng's computational workflow. Transition-state searches are critical for calculating Li⁺ migration barriers, interfacial reaction pathways, and degradation mechanisms in solid-state batteries. TSAgent could automate the tedious NEB/CI-NEB workflow that is currently a major bottleneck for studying electrode-electrolyte interface reactions. The agentic plan-replan loop approach complements the group's existing DFT calculations for halide and sulfide electrolyte stability.

Solid Electrolytes & Interfaces

3. Bulk-to-Interface Fluorination for Stable and Low-Pressure All-Solid-State Lithium Metal Batteries

Source: Nature Communications (10.1038/s41467-026-73012-4) · 📅 2026-05-14 · ↗ Open paper

Designs a core-shell structured sulfide electrolyte (Li₅.₄PS₄.₄Cl₁.₄F₀.₂-0.2LiF) with a 50 nm LiF nanoshell and F-enriched bulk via fast thermodynamic diffusion of fluorine atoms. During cycling, F atoms diffuse into the NCM cathode lattice, enhancing structural robustness and mitigating mechanochemical failure, while the LiF nanoshell stabilizes both Li metal and cathode interfaces. Pouch cells achieve stable cycling over 350 cycles at 1 C under only 2.5 MPa stack pressure, with >400 Wh/kg specific energy.

Relevance to DENG.Group

Highly relevant to the entire Deng group's solid-state battery research. The bulk-to-interface fluorination strategy provides a new design principle for achieving low-pressure operation, which is critical for practical ASSBs. The LiF nanoshell concept could be explored computationally to understand interfacial thermodynamics and F-diffusion kinetics. Cheng Peng could apply the group's computational tools to model F diffusion at grain boundaries and interfaces in halide electrolytes.

4. Amorphous Sulfo-Halide Solid Electrolytes With Enhanced Anion Dynamics for Highly Stable All-Solid-State Sodium Batteries

Source: Advanced Energy Materials (10.1002/aenm.71091) · 📅 2026-05-18 · ↗ Open paper

Develops amorphous sulfo-halide solid electrolytes with enhanced anion dynamics for all-solid-state sodium batteries. The amorphous structure provides superior deformability, stable high-voltage cycling, and excellent interfacial contact under moderate stack pressures. The mixed-anion design strategy leverages the coexistence of sulfide and halide anions to achieve high Na⁺ conductivity while maintaining electrochemical stability against both Na metal and high-voltage cathodes.

Relevance to DENG.Group

Directly relevant to the group's interest in halide solid electrolytes and sodium-ion systems. The sulfo-halide mixed-anion approach extends the design space beyond pure chloride or bromide halides. The amorphous structure strategy addresses the grain-boundary resistance problem that Cheng Peng studies. The sodium-ion focus opens computational opportunities for the group to explore Na⁺ migration in amorphous vs. crystalline sulfo-halide environments using MLIP-driven MD simulations.

Cross-Scale Modeling & Reviews

5. AI-Enabled Cross-Scale Modeling and Parameter Transfer for Lithium-Ion Batteries

Source: Energy Storage Materials (10.1016/j.ensm.2026.scaff) · 📅 2026-05-15 · ↗ Open paper

Proposes the AI-enabled Cross-Scale Parameter Chain (ACPC) as a carrier-centered framework for cross-scale LIB modeling. The framework traces how physically meaningful quantities are generated at atomic scales, transformed across model interfaces, compressed into deployable descriptors, and updated through operational feedback. Reviews AI's role at each interface: parameter inversion, structure reconstruction, surrogate acceleration, uncertainty-aware compression, and online recalibration across atom-to-module scales.

Relevance to DENG.Group

A valuable reference for the Deng group's computational strategy. The ACPC framework formalizes how atomistic calculations (DFT, MLIP-MD) feed into electrode- and cell-level models, which is exactly the challenge the group faces in connecting MLIP predictions to battery-level performance. The discussion of uncertainty-aware compression and parameter transfer is particularly relevant for Yanhao Deng's MLIP development, ensuring that atomistic accuracy propagates meaningfully to engineering-scale predictions.