About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithms Development in Materials Science and Engineering
|
Presentation Title |
Accelerating molecular dynamics simulations using mass rescaling and asynchronous time integrators |
Author(s) |
Amuthan Ramabathiran |
On-Site Speaker (Planned) |
Amuthan Ramabathiran |
Abstract Scope |
Molecular Dynamics (MD) simulations provide significant atomistic insights into material behavior, but suffer from well known length and time scale restrictions. We present a new computational scheme to accelerate MD simulations by systematically reformulating the governing equations in a manner that naturally introduces mass-rescaling, and discuss the role of interpretable machine learning models in this regard. We subsequently employ asynchronous time integrators to achieve the desired acceleration. The advantage of this formalism is that it permits the use of large time steps for most atoms, with small time steps employed only in regions where critical dynamics is expected, in the spirit of the quasicontinuum method. We present a variety of illustrations using both 2D and 3D simulations to highlight the utility of the proposed acceleration scheme. The talk will conclude with a broader discussion of the challenges that need to be overcome to achieve better acceleration of MD simulations. |
Proceedings Inclusion? |
Planned: |
Keywords |
Computational Materials Science & Engineering, Modeling and Simulation, Machine Learning |