2021 Fiscal Year Research-status Report
Development and application of robust machine-learning interatomic potentials for the computational design of solid electrolytes for all-solid-state batteries
| Project/Area Number |
21K14729
|
|---|
| Research Institution | National Institute for Materials Science |
|---|
Principal Investigator |
JALEM Randy 国立研究開発法人物質・材料研究機構, エネルギー・環境材料研究拠点, 主任研究員 (20767553)
|
|---|
| Project Period (FY) |
2021-04-01 – 2024-03-31
|
| Keywords | machine learning / solid electrolytes / solid-state batteries |
|---|
| Outline of Annual Research Achievements |
For the goal of developing robust machine-learning interatomic potentials for the computational design of solid electrolytes for all-solid-state batteries, the following research outputs were thus far achieved:
1. Development of high-throughput DFT and ab initio molecular dynamics calculation workflows for training data generation.
2. Data generation of optimized crystal structures, strained structures, and grain boundary structures by high-throughput DFT and ab initio molecular dynamics (crystal structure coordinates, total energies, and forces were successfully collected for the Na-Sb-S chemical system, Li-rich inverse perovskites, and Li-rich garnet-type oxides).
3. Development of data processing tools for actual passive learning tasks of machine-learning potential parameter sets.
4. Publications of results to peer-reviewed journals for theoretically proposed promising solid electrolyte systems that were analyzed/studied by DFT method which also forms part of the generated training dataset for machine-learning potential fitting.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
The research targets were achieved due to prior preparations made by the proponent in the previous year.
|
| Strategy for Future Research Activity |
The following are the future tasks:
1. Improvement of training dataset diversity for fitting tasks of machine learning potentials, such as inclusion of sub-system compounds in a target chemical space to capture more atomic environment variety(e.g., in the Na-Sb-S, generate training data in the Na-Sb, Na-S, Sb-S, Na, Sb and S chemical system).
2. Carrying out of actual fitting tasks for machine-learning potential parameter sets using DFT-generated training dataset (passive learning).
3. Accuracy/performance check/improvement of fitted machine-learning potentials in terms of energies, forces.
|
Research Products
(10 results)
