AQVolt26: Accelerating Solid-State Battery Discovery with High-Fidelity Data and AI

From electric vehicles to grid storage and defense, demand for safer, higher-energy density solid-state batteries is growing faster than traditional R&D can keep up. Yet solid-state electrolytes are notoriously difficult to model: standard DFT workflows are too slow for large-scale dynamics, and existing foundation datasets struggle in the high-temperature, highly distorted regimes where ion transport actually happens in working cells.

In this webinar, SandboxAQ’s battery AI team will introduce AQVolt26 — a 322,656-frame r²SCAN Li-halide dataset and family of universal machine-learning force fields designed to close this high-temperature “blind spot.” You’ll see how AQVolt26 complements datasets like MatPES and MP-ALOE, delivers near-zero failure rates under extreme lattice strain, and enables robust, high-throughput screening of halide electrolytes for next-generation solid-state batteries — moving from prediction → simulation → materials discovery with Large Quantitative Models (LQMs).

Key Learning Objectives

1. Understand the modeling gap for solid-state electrolytes – why elevated-temperature, far-from-equilibrium configurations break many existing ML potentials, and how AQVolt26 was designed to systematically cover this regime with r²SCAN-level fidelity.
2. See how AQVolt26-powered universal ML force fields behave in practice – including potential energy surface benchmarks, dynamic stability under ±20% lattice deformations, and ionic conductivity predictions that track demanding solid-state electrolyte benchmarks.
3. Learn how to incorporate AQVolt26 models into your own workflows – from accessing checkpoints on Hugging Face and interpreting licensing terms, to integrating the potentials into existing MD and materials screening pipelines for battery R&D.
4. Explore the roadmap for LQM-driven battery innovation – how AQVolt26 extends SandboxAQ’s earlier work in battery lifetime prediction and lays the foundation for future datasets and models spanning sulfides, cathodes, interfaces, and full solid-state cell design.

AQVolt26 represents one of the largest high-fidelity datasets for off-equilibrium lithium halide materials and enables more accurate modeling of ionic conductivity and structural stability in next-generation battery systems.

Access the dataset and models here: https://huggingface.co/SandboxAQ

Read the paper on arXiv here: https://arxiv.org/html/2604.02524v1

Learn more about what the AQChemSim team is up to here: https://www.sandboxaq.com

Questions? Contact the team here: materials@sandboxaq.com

Chapters
00:00 - Introduction
00:42 - Large Quantitative Models vs Large Language Models
01:18 - Why batteries are critical infrastructure technology
02:37 - DFT vs classical simulation methods
03:52 - Machine learning interatomic potentials explained
04:40 - Catalyst modeling acceleration example
06:00 - Importance of spin polarization in datasets
07:24 - Configuration coverage challenges
08:55 - Expanding sampling beyond equilibrium structures
10:20 - Moving toward higher-fidelity training data
10:44 - Modeling lithium halide electrolytes
11:46 - Generating 200M configurations with ML molecular dynamics
12:27 - Building the AQVolt26 dataset
14:09 - Training equivariant neural network potentials
15:57 - Strain stability benchmarking results
16:36 - High-temperature simulation robustness
17:13 - Impact on next-generation battery discovery
17:43 - Dataset availability and collaboration opportunities

#sandboxaq #aqchemsim #lqms

Видео AQVolt26: Accelerating Solid-State Battery Discovery with High-Fidelity Data and AI канала SandboxAQ