Researchers from Chinese supercomputing centres, universities and ETH Zurich have calculated the electronic structure of a solid-state battery interface containing 11.3 million atoms. Their XLSDFT software also completed silicon calculations containing up to 200 million atoms on China's LineShine exascale computer. The September 11 preprint describes the battery model as about 1,000 times larger than earlier studies of comparable interfaces.
Density functional theory, or DFT, calculates how electrons are distributed through a material's atomic structure. It can reveal charge transfer and energy changes that are difficult to measure directly. Its computing cost rises quickly as atoms are added, so conventional models often cover only hundreds or thousands of atoms. Battery reaction layers can extend for tens of nanometres and require a much larger view.
The team combined machine learning with a divide-and-compute method. A force model trained on small quantum simulations first prepared the 11.3-million-atom lithium-LGPS interface at operating temperature. XLSDFT then split that structure into overlapping local regions. Each region was solved separately before the results were joined into a global picture of the electrons. The final electronic signatures were compared with X-ray photoelectron spectroscopy measurements from the same material system.
The scaling tests ran silicon models and the battery interface on as many as 20,480 LineShine compute nodes. The battery model measured roughly 57 by 44 by 90 nanometres, large enough to span the reaction region between lithium metal and the solid electrolyte. The researchers measured the complete calculation time as well as the performance of the main numerical stage.
XLSDFT kept 96.6% of its ideal scaling efficiency when the silicon model reached 100 million atoms. The 11.3-million-atom battery calculation finished in just under two hours. It placed altered forms of germanium and phosphorus most heavily near the lithium boundary, with their concentration falling deeper into the electrolyte. The laboratory depth profile showed the same direction of change.
The calculation addresses one interface chemistry and begins from a structure produced by a learned force model. It also relies on a specialised exascale machine and stated electronic approximations. Studies of other materials must show that the method can predict defects and reaction products that later experiments confirm. Those comparisons will determine whether unprecedented scale also brings more reliable battery design decisions.
Related coverage
Sources
- Qimen Xu and colleagues, XLSDFT, arXiv, September 11, 2026. Abstract, authorship, submission date and headline results.
- Full paper. Workflow, system configuration, scaling measurements, battery-interface results and limitations.
- XLSDFT source code. Public implementation linked by the authors.
- Lead image: Oak Ridge National Laboratory, Wikimedia Commons, CC BY 2.0. The photograph is illustrative and does not depict the XLSDFT study or its lithium-LGPS interface.
.jpg){kind=link}