Research Digest — 2026-10-05¶
Solid Electrolytes & Ion Transport¶
1. Topo-Spectral Percolation Descriptors for Mechanistic Ion Transport Pathways from Static Crystal Structures¶
Source: arXiv (2610.01070) · 📅 2026-10-01 · ↗ Open paper
Introduces topology- and spectral-analysis-based descriptors that identify diffusion pathways and bottleneck atoms directly from static crystal structures, using percolation theory on the ionic sublattice. The approach recovers mechanistic transport information that normally requires costly AIMD, enabling fast pre-screening of solid electrolytes, electrodes, and ion-selective membranes.
Relevance to DENG.Group
Mengke Li (halide ion transport) and Yanhao Deng (ML-potential MD) can use these descriptors as a cheap first-pass filter before running expensive AIMD/MLIP-MD on halide or sulfide candidates — a natural fit for the group's multi-stage screening workflow.
Battery Interfaces & Characterization¶
2. Fluctuation-induced ionic friction at solid-electrolyte interfaces¶
Source: arXiv (2609.40142) · 📅 2026-09-30 · ↗ Open paper
Combines MD simulations with analytical theory to show that interfacial ionic friction at solid-electrolyte contacts is dominated by collective solid and electrolyte fluctuations rather than individual ion drag. Ionic contributions to the fluctuation-induced interfacial friction arise from many-body effects, and high-frequency molecular modes turn out to be ion-specific.
Relevance to DENG.Group
Umang Agarwal (electrolyte/electrode heterogeneous interfaces) gains a theoretical lens for interpreting interfacial ion friction and contact resistance in his simulations; Naibing Wu's polymer interfacial transport work could test whether collective-mode dissipation carries over to soft interfaces.
ML Interatomic Potentials¶
3. BranchIP: Learning Adaptive Equivariant Computation for Interatomic Potentials¶
Source: arXiv (2610.02013) · 📅 2026-10-01 · ↗ Open paper
Presents a single-model framework for learned adaptive tensor-product computation in equivariant MLIPs, trained with a novel distillation loss that allocates expensive operations only to environments that need them. Achieves near-full accuracy at a fraction of the inference cost on a heterogeneous catalysis system and a proton-conducting solid acid electrolyte.
Relevance to DENG.Group
Yanhao Deng (ML interatomic potentials for solid electrolytes) can borrow the adaptive-compute strategy to accelerate foundation-potential-based MD of solid electrolytes and electrodes, extending accessible trajectory length- and time-scales.
4. From Energy-Force Weighting to Primal-Dual Optimization of Machine-Learned Interatomic Potentials¶
Source: arXiv (2610.00876) · 📅 2026-10-01 · ↗ Open paper
Reinterprets energy-force balancing in MLIP fitting as a protocol-dependent model-selection problem and replaces weighted-sum scalarization with primal-dual optimization: a regularized energy objective under exact force constraints. Studies on molten LiCl, liquid water, and Si reveal dominated weighting regimes and nonmonotone out-of-distribution behavior.
Relevance to DENG.Group
Yanhao Deng trains custom potentials for solid electrolytes and electrodes — this constraint-based fitting scheme offers a principled alternative to hand-tuned energy/force loss weights in the group's DeepMD/ACE training protocols.
5. Reference-free certification of machine-learning interatomic potentials¶
Source: arXiv (2610.00585) · 📅 2026-09-30 · ↗ Open paper
Shows that MLIPs can be graded without any reference calculation by testing properties the exact Born-Oppenheimer surface must satisfy: symmetry and invariance, self-consistency and integrability, and statistical-mechanical equilibrium. Organizes these into fourteen inexpensive probes that return absolute pass/fail certifications independent of chemistry and reference method.
Relevance to DENG.Group
A cheap quality-control gate for every potential the group trains — Yanhao Deng and Umang Agarwal can run these probes before production MD to catch failure modes that held-out error metrics miss.
6. A latent-space extrapolation grade built into graph atomic cluster expansion foundation potentials¶
Source: arXiv (2609.40060) · 📅 2026-09-30 · ↗ Open paper
Introduces the calibrated Mahalanobis (CALM) extrapolation grade for GRACE/ACE foundation potentials: a piecewise-differentiable, per-atom uncertainty measure computed in the same pass as energies and forces. Controlled tests show it reliably flags extrapolative atomic environments during simulations, enabling detection of unreliable predictions on the fly.
Relevance to DENG.Group
Directly useful for Yanhao Deng's active-learning loops: per-atom extrapolation grades can trigger targeted DFT labeling near interfaces and grain boundaries where the group's potentials are most likely to fail.
Halide & Oxychloride Electrolytes¶
7. Crystalline–amorphous heterophase oxychloride solid electrolyte for achieving fast Li-ion conduction and enhanced moisture tolerance¶
Source: Journal of Power Sources (10.1016/j.jpowsour.2026.241596) · 📅 2026-09-25 · ↗ Open paper
Engineers a crystalline-amorphous heterophase structure in oxychloride solid electrolytes, where amorphous domains provide moisture tolerance while crystalline regions preserve fast Li-ion conduction. The heterophase design relaxes the usual trade-off between ionic conductivity and air stability in halide-family electrolytes.
Relevance to DENG.Group
Yan Li and Mengke Li (halide electrolyte degradation and transport) can use this as a model system for simulating how amorphous/crystalline phase mixtures alter degradation pathways and ion transport — directly relevant to their air-stability and conductivity mechanism studies.
Dendrite Growth & Phase-Field Methods¶
8. Peridynamic modeling of lithium dendrite growth¶
Source: Journal of Energy Storage (10.1016/j.est.2026.123162) · 📅 2026-10-01 · ↗ Open paper
First peridynamics framework for lithium electrodeposition, coupling the Poisson-Nernst-Planck ion transport system with a nonlocal solid mechanics description. The nonlocal formulation naturally handles crack-like dendrite propagation and interfacial evolution without explicit interface tracking.
Relevance to DENG.Group
Shoutong Jin (phase-field simulation of dendrite growth) gains a nonlocal alternative to compare against phase-field dendrite models — useful for cross-validating morphology predictions and understanding where nonlocality matters during Li metal electrodeposition.