University of Kazan Develops Highly Accurate Machine-Learned Potential (MLP) for Aluminum Using n2p2 Methodology in MedeA MLPG

Structural model of liquid aluminum and phonon dispersion curves as calculated using the present NNP-Al in comparison to DFT results and experimental data.

Aluminum is an important material, with at least $150 billion worth produced and sold annually, and its properties and alloys form the foundation of an enormous range of products. Recently, Akhmerov and colleagues from the University of Kazan have developed a highly accurate machine-learned potential (MLP) for aluminum using the n2p2 methodology available in the MedeA MLPG, achieving excellent agreement with experimental data and DFT results and enabling the calculation of thermodynamic, elastic, vibrational, and generalized stacking fault energy properties. In addition, excellent agreement of liquid phase results at ambient and elevated pressures with measured data with measured data underline the wide range of applicability of this new potential. See the paper for the detailed procedure including optimal training set selection and configuration parameter configuration - it is a fascinating read.

The official reference is as follows:

R. F. Akhmerov, I. I. Piyanzina, O. V. Nedopekin, and V. Eyert, A neural-network potential for aluminum, Comput. Mater. Sci. 244, 113159 (2024). 

#modeling #simulation #materialsscience #materials #MLP #MLPG #DFT #Aluminum #alloys

---