Application of MLP to global structure search in high-pressure Bi binary systems discovers many unreported compounds

1/ AI utilization method: global structure search based on machine learning potential
2/ Application task: search for possible crystal structures and compounds in high-pressure Bi binary systems
3/ Reported result: found several unreported compounds across 11 binary systems at 0–20 GPa and reproduced previously known compounds
4/ The researchers carried out a global structure search using a machine learning potential (MLP) for 11 bismuth-based binary systems.
5/ The search pressure range was 0–20 GPa, and several unreported compounds were proposed in the process.
6/ It was also reported that previously experimentally known compounds were reproduced in the explored configuration space.
7/ From the perspective of materials informatics, this case shows an approach in which an MLP is used not simply as an energy predictor, but as a structure-search engine under high-pressure conditions.
8/ The key point is that when the search target is a complex composition-pressure space, a much broader set of structural candidates can be examined than with classical exhaustive search.
9/ However, such results depend on the selected compositions, pressure range, and the training scope of the potential.
10/ The proposal of unreported compounds indicates structural candidates only, and actual stability requires additional high-accuracy calculations and experimental confirmation.
11/ Reproducing known compounds is one basis for reliability, but it does not guarantee all pressure conditions and the entire composition range. [1][2]