Machine learning explores the metastable phase diagram of covalent carbon, which can be used to discover or design metastable materials in the future

Traditional phase diagram generation involves experiments to provide an initial estimate of a set of thermodynamic accessible phases and their boundaries , Then the phenomenological model is used to interpolate between the available experimental data points and extrapolate to the inaccessible region of the experiment .

This method is combined with high-throughput first principles computing and data mining techniques , This leads to a detailed thermodynamic database （ For example, CALPHAD Method compatible ）, Although the focus is on the reduced phase set observed at different thermodynamic equilibria .

by comparison , Materials in their synthesis 、 The thermodynamic equilibrium state may not be reached during operation or processing , But keep it local （ Metastable ） Of the minimum free energy , This may exhibit desirable characteristics .

Researchers from the nano materials center of Argonne National Laboratory introduced an automated workflow , It combines first principles physics and atomic simulation with machine learning (ML) And high-performance computing , To allow rapid exploration of metastable phases , So as to build... For materials far from equilibrium 「 Metastable 」 Phase diagram .

Using carbon as a prototype system , The researchers demonstrated automatic metastable phase diagram construction , To plot hundreds of metastable states , Range from near equilibrium to far from equilibrium （400 meV/ atom ）. The free energy calculation is combined with the state equation learning based on neural network , Thus, the metastable phase diagram can be constructed effectively . The team used metastable phase diagrams and determined the relative stability and composability domains of metastable materials . And through the experiment , Confirmed their metastable prediction .

The research 「 Machine learning the metastable phase diagram of covalently bonded carbon」 entitled , On 2022 year 6 month 6 Published on 《 Nature Communications》.

Thesis link ： https://www.nature.com/articles/s41467-022-30820-8