Stability of binary precipitates in Cu-Ni-Si-Cr alloys investigated through active learning

AD Carral and X Xu and S Gravelle and A YazdanYar and S Schmauder and M Fyta, MATERIALS CHEMISTRY AND PHYSICS, 306, 128053 (2023).

DOI: 10.1016/j.matchemphys.2023.128053

Binary complexes that can be found in copper alloys are investigated in this work through a combination of computer simulations and machine learning. Copper alloys are made of a copper matrix and a combination of single alloying elements in n-ary forms. Due to the coexistence of different types of phases in this matrix, complex regions exist for which information on their precise atomistic structure is missing. In order to uncover such information, we apply active learning and generate moment tensor potentials. This development is based on quantum- mechanical calculations. This approach allows the on-the-fly relaxation of many thousands of potentially novel candidates and check their stability. The ground-state energy of these structures is used to build active learning-generated convex hulls, which are in turn being compared to those from the simulations and the AFLOW database. This procedure provides an insight to additional new stable copper alloy relevant binary complexes. Here, in view of Cu-Ni-Si-Cr alloys, the binary complexes Cu-Si, Ni-Si, Cr-Si, Cr-Ni, Cu- Ni, and Cu-Cr have been investigated. Their stability and the identification of novel stable candidates are discussed based on energetic arguments and the analysis of the respective phonon dispersion. The pipeline followed in this work is able to successfully predict binary phases in Cu-Ni-Si-Cr alloys, specifically for the Cu-Si, Ni-Si, Cr-Ni and Cu-Ni complexes, and to extend the already reported structures in the AFLOW library. In the end, we show the applicability of a predicted Cu-Si stable phase and the developed machine learned potentials at the larger scale of atomistic simulations for the calculation of their mechanical properties and melting behavior. This work provides a computationally efficient framework for material structure prediction and calculation of their properties at a quantum-mechanical accuracy.

Return to Publications page