About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Materials Processing and Fundamental Understanding Based on Machine Learning and Data Informatics
|
| Presentation Title |
ML-Informed ReaxFF Development for Complex Metal Carbide, Oxide and Nitride Materials |
| Author(s) |
Adrianus C. Van Duin, Asma Ul Hosna, Mozhdeh Mirakhory, Zihan Wang, Wei Chen, Yun Kyung Shin |
| On-Site Speaker (Planned) |
Yun Kyung Shin |
| Abstract Scope |
The ReaxFF method [1] provides highly transferable simulation methods for atomistic scale simulations on chemical reactions at the nanosecond and nanometer scale. This method combines concepts of bond-order based potentials with a polarizable charge distribution. At this moment, over 1000 ReaxFF publications have appeared in open literature and the ReaxFF code has been distributed around the world.
Here we describe development and applications of the ReaxFF methods to complex, multi-element metal carbide/nitride/oxide materials, with applications to hypersonic environments. For this, we utilize Machine Learning (ML) methods to compare configurations encountered during high-temperature ReaxFF molecular dynamics simulations with the data from the ReaxFF training set, enabling us to identify the ReaxFF accuracy and allowing us to design an automatic, self-evaluating and self-improving framework for the design of reactive force fields for complex materials and their interfaces.
[1] van Duin et al. (2001) Journal of Physical Chemistry A 105, 9396-9409. |