| About this Abstract |
| Meeting |
MS&T23: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
Development of Machine Learning Interatomic Potentials to Model Materials Processing & Performance in Multicomponent Systems |
| Author(s) |
Ridwan Sakidja, Marium Mostafiz Mou, Nur Aziz Octoviawan, Tyler McGilvry-James, Gaige M Riggs |
| On-Site Speaker (Planned) |
Ridwan Sakidja |
| Abstract Scope |
Developing the classical interatomic potentials to model materials processing in multi-component systems has been “the holy grail” in the field of computational materials science. A variety of Machine Learning Interatomic Potentials (MLIP) have emerged as the necessary means to address this issue. The potential development typically starts with the generation of the critical database through the electronic structure calculations within the Density Functional Theory (DFT) approximation at the ground state and at elevated temperatures. The extracted critical data (energy, stress, and forces) are then fed to the neural networks or machine learning algorithms with various schemes of invariant/equivariant representations. Due to the linear scalability of the molecular dynamics (MD) simulations that utilize these interatomic potentials, many large-scale simulations can now be performed to help understand the key material processing and performance. Additionally, a Virtual Autonomous Materials Discovery (v-AMD) may be established to accelerate the materials development in multi-component systems. |