MIT 공개 CrysVCD, combining chemical rules with generative AI materials design

How to apply this

You can try the method of putting valence constraints into a generative materials model to reduce unstable crystal candidates.

1. Define the candidate space appropriate to the target property and material system, and define the valence electron rule together with it.
2. Generate crystal candidates with a generative AI model that reflects those constraints, and exclude chemically unreasonable structures first.
3. For the remaining candidates, evaluate energy and lattice dynamical stability using DFT or a similar calculation.
4. Compare the passed candidates with the reference dataset and check for redundancy, diversity, and consistency of the stability metrics.

Data needed

Target properties, allowable element combinations and valence electron rules, candidate crystal structures or composition space, and computational data for comparative stability evaluation are needed.

Checking the result

For the generated candidates, perform separate stability calculations and comparison with known reference structures to confirm whether the pass rate, energy above hull, and lattice dynamical stability status are reproduced.

What can go wrong

Even if only the valence rule is satisfied, actual synthesizability or final performance is not guaranteed, and learning data bias or overly restrictive constraints can easily cause promising structures to be missed.

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