Building a Machine Learning Potential for MOF-303 Hydrogen Adsorption with Hybrid Sampling

AI editorPublished 2026.10.05Posted 2026.10.06 06:01 KST

An accepted paper in APS reports that a machine learning potential for hydrogen adsorption in MOF-303 was created using hybrid sampling and active-learning refinement, and applied to GCMC adsorption simulations. It states that energies and forces were matched accurately, and that temperature-dependent isotherms, heats of adsorption, diffusion coefficients, and identification of 8 adsorption sites were possible. [1]

1/ AI utilization method: Construction of a machine learning potential using hybrid sampling and active-learning refinement
2/ Application task: Potential training for GCMC simulations of hydrogen adsorption in MOF-303
3/ Reported results: Accurate evaluation of energies and forces, enabling calculation of temperature-dependent isotherms, heats of adsorption, diffusion coefficients, and identification of 8 adsorption sites
4/ This paper combined hybrid sampling and active-learning refinement to construct a machine learning interatomic potential for handling hydrogen adsorption in MOF-303.
5/ Using this potential, hydrogen adsorption GCMC simulations were performed, and it was reported that energies and forces could be evaluated accurately.
6/ It also stated that calculation of temperature-dependent adsorption isotherms, heats of adsorption, and diffusion coefficients was possible, and that 4 main adsorption sites and 4 secondary adsorption sites around N/O/C/–OH were identified.
7/ In terms of materials research, this case demonstrates the practical value of an approach that accelerates expensive adsorption behavior analysis by training an MLP for adsorbable porous materials.
8/ However, the reported performance and the number of adsorption sites are results limited to MOF-303 and hydrogen, so they cannot be directly generalized to other frameworks or adsorbates. [1]

How to apply this

For porous materials, one can train an MLP with hybrid sampling and active-learning refinement to predict adsorption behavior.

1. Build an initial training set including representative framework structures, adsorbate placements, and defects and ground states.
2. Generate labels for energies and forces from high-precision calculations, and actively add structures with large errors.
3. Run GCMC or MD with the trained MLP to calculate isotherms, heats of adsorption, and diffusion coefficients.
4. Compare with independent data under different conditions to verify whether adsorption sites and properties are consistent.

Data needed: Representative framework structures, adsorbate placements, defect and ground-state configurations, initial high-precision energy and force calculations, and adsorption simulation conditions matched to the temperature and pressure range are needed.
Checking the result: Check energy and force errors on holdout structures and adsorbate placements, and compare isotherms, heats of adsorption, and diffusion coefficients against independent GCMC or MD results to assess reproducibility.
What can go wrong: If the initial sample is biased toward specific adsorption sites or local environments, the MLP may become vulnerable to extrapolation, causing the number of adsorption sites or thermodynamic quantities to appear overly optimistic.

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

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

Building a Machine Learning Potential for MOF-303 Hydrogen Adsorption with Hybrid Sampling

How to apply this

Related concepts and practice

Sources