Research Digest — 2026-09-11¶
ML Potentials & Simulation Tools¶
1. MLIP Detective: Active Failure Mode Discovery Beyond Benchmark Scores for Machine-Learning Interatomic Potentials¶
Source: arXiv · 📅 2026-09-08 · ↗ Open paper
Introduces an agentic framework that goes beyond static benchmarks to actively discover hidden failure modes in universal ML interatomic potentials. Physics-informed failure hypotheses are generated from benchmark evidence, screened with inexpensive simulations, and only the most suspicious cases are escalated to human experts with proposed verification protocols.
Relevance to DENG.Group
Directly useful for Yanhao Deng's ML interatomic potential work on solid electrolytes and electrodes: as u-MLIPs (MACE, etc.) become the group's workhorse, a systematic workflow for finding where they break — before production MD — addresses the validation gap in our own fine-tuning pipelines.
2. HiPoly: a hierarchical polymer-native AI framework for property prediction and generative design¶
Source: arXiv · 📅 2026-09-02 · ↗ Open paper
A polymer-native AI framework that processes complete polymer descriptions through a three-level hierarchical graph architecture built on the G2RINS representation, encoding stochastic inter-monomer connectivity and composition. Targets property prediction and generative design for polymeric materials across energy applications.
Relevance to DENG.Group
Naibing Wu's solid polymer electrolyte and composite simulations could borrow both the hierarchical representation and the pretraining strategy: a learned polymer prior calibrated against MD could pre-screen PEO-composite chemistries before committing to expensive atomistic runs.
Interfaces & SEI¶
3. Insights into interfacial reactions between Li-argyrodite solid electrolyte and lithium metal via machine learning-based molecular dynamics¶
Source: Journal of Energy Storage · 📅 2026-08-01 · ↗ Open paper
Uses machine-learning-based molecular dynamics to resolve atomistic interfacial interactions, charge transfer, and amorphous structural evolution at the Li/Li6PS5Cl/Li interface. Connects the chemistry of interphase formation to the electronic structure across the electrolyte-metal boundary.
Relevance to DENG.Group
A template case for Yan Li and Mengke Li's halide electrolyte degradation work — same Li-metal contact problem, different anion chemistry — and methodologically close to Umang Agarwal's electrolyte/electrode interface simulations: ML-MD now reaches the tens-of-nanoseconds scale needed for spontaneous interphase growth.
4. A locally ab initio computational framework for arbitrary incommensurate materials interfaces¶
Source: arXiv · 📅 2026-09-04 · ↗ Open paper
Introduces a scalable framework for constructing locally ab initio electronic Hamiltonians for arbitrary incommensurate interfaces, exploiting the nearsightedness of Wannier Hamiltonian matrix elements to extrapolate and interpolate across interlayer registries. Retains first-principles accuracy while avoiding prohibitively large commensurate supercells.
Relevance to DENG.Group
Real electrolyte/electrode contacts (Umang Agarwal's heterogeneous interfaces, Cheng Peng's GB-electrode boundaries) are generically incommensurate — this removes the lattice-matching constraint that usually forces strained or oversized interface models, enabling DFT-quality electronic structure at realistic solid-solid contacts.
Electrolyte Solvation & Ion Transport¶
5. Differentiable Solvation Shell Model for Rational Electrolyte Design¶
Source: arXiv · 📅 2026-09-04 · ↗ Open paper
Presents a mean-field Ising-type model that predicts Li+ solvation shell composition from donor number, acceptor number, molar ratio, and molecular size. The framework is end-to-end differentiable and parameterized directly from MD, enabling gradient-based optimization of electrolyte formulations toward target solvation structures.
Relevance to DENG.Group
Complements Naibing Wu's electrolyte simulation work with a cheap, differentiable surrogate sitting between molecular descriptors and solvation structure; the MD-parameterization-to-optimization loop is the same pattern Yan Li and Mengke Li could apply to halide precursor speciation in synthesis-relevant solutions.
6. Revealing the origin of ionic conduction in silver-iodide-doped silver phosphate glass¶
Source: arXiv · 📅 2026-09-06 · ↗ Open paper
Uses time-domain terahertz spectroscopy across a series of (AgI)x(AgPO3)(1-x) glassy electrolytes to probe the mesoscopic frequency regime where ionic transport emerges. Identifies how AgI concentration controls the crossover from bound polarization to mobile-ion response, clarifying the microscopic origin of fast-ion conduction in glass.
Relevance to DENG.Group
Connects to the group's glassy solid electrolyte theme from late August (structure-transport relations in glass): Mengke Li's transport-mechanism studies and any halide-glass work gain a spectroscopy benchmark for separating localized hopping from percolating transport — a distinction that also matters for interpreting conductivity plateaus in MD.
Methods & Theory¶
7. Modeling electrolytes in nanopores by Monte Carlo simulations and the Bazant-Storey-Kornyshev model¶
Source: arXiv · 📅 2026-09-09 · ↗ Open paper
Benchmarks the Bazant-Storey-Kornyshev (BSK) mean-field theory, augmented with a Stern layer and a mechanical-equilibrium boundary condition for the outer Helmholtz plane, against Monte Carlo simulations of electrolyte-filled cylindrical nanopores. The BSKB-Stern model outperforms classical Poisson-Boltzmann-Stern for multivalent electrolytes where ion-size correlations dominate.
Relevance to DENG.Group
The EDL theory Underpinning interfacial ion distributions matters for Umang Agarwal's interface charge-transfer models and for how the group parameterizes boundary conditions in coarse-grained or continuum descriptions of electrode-electrolyte contacts; multivalency (Li+, trivalent halide precursors) is exactly where PB fails and correlated theories are needed.