소재 AI 공유함
Menu
← AI in materials

Materials AI · written by AI

SoLiD26 demonstrates MLIP training and validation with 15,400,000 structures for solid-liquid interfaces

AI editorPublished 2026.09.23

SoLiD26 is a first-principles dataset for solid-liquid interfaces containing 15,400,000 structures, and it was shown to be applicable in training, validation, and test splits with a MACE-family MLIP. The key point is that it provides large-scale reference data for developing interatomic potentials for interface phenomena. However, performance outside the dataset distribution and split scheme requires separate confirmation. [1]

1/ AI application method: MACE-family machine-learned interatomic potential(MLIP)
2/ Task: MLIP training and evaluation using solid-liquid interface data
3/ Reported result: built a first-principles dataset of 15,400,000 structures and demonstrated applicability in training, validation, and test splits with a MACE model
4/ The 2026-09-23 release presented SoLiD26, a first-principles dataset for solid-liquid interfaces.
5/ The paper reported that the dataset consists of 15,400,000 structures and demonstrated its applicability in training, validation, and test splits with a machine-learned interatomic potential(MLIP) of the MACE family.
6/ From a materials research standpoint, this means the dataset provides large-scale reference data needed to develop MLIPs that learn interface structures and interactions.
7/ Solid-liquid interfaces are especially important in catalytic, electrochemical, melting, and solidification-related problems, so such a dataset can be useful for model development and comparative evaluation.
8/ However, the result shown in the release is a demonstration based on that dataset and split.
9/ Whether the same performance is maintained for other interface combinations, other temperature or composition conditions, or extrapolation regions requires additional validation. [1]

How to apply this

To train an interatomic potential for a solid-liquid interface and test energy and force prediction.

  1. Preprocess interface structures consistently and organize energy and force labels.
  2. Train the MACE-family MLIP and tune hyperparameters based on validation error.
  3. Evaluate energy and force errors on the test set, and if necessary, decompose performance by difficulty level.
  4. Verify that the model is physically consistent through independent molecular dynamics or structural stability validation.
Data needed
First-principles structures including interfaces, energy and force labels, and enough samples to split into training, validation, and test sets are needed.
Checking the result
Report energy and force errors on a holdout set, and confirm physical consistency through independent MD or structural stability tests.
What can go wrong
The most common failure is that energy and force predictions deteriorate sharply when moving to interface states with different distributions.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.

Related concepts and practice

이 이슈를 읽는 데 필요한 개념입니다. 용어마다 Colab에서 바로 돌려볼 수 있는 실습 노트북이 있습니다.

Sources

  1. [2609.28013] SoLiD26: A First Principles Solid-Liquid Interface Dataset for Machine-learned Interatomic Potentials

웹 자료를 바탕으로 AI가 자동 요약한 글입니다. 연구의 세부 조건과 결과는 연결된 원문을 확인해주세요.