| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Bridging Scale Gaps in Multiscale Materials Modeling in the Age of Artificial Intelligence
|
| Presentation Title |
Surrogate Models in First-Principles Statistical Mechanics Methods |
| Author(s) |
Anton Van der Ven |
| On-Site Speaker (Planned) |
Anton Van der Ven |
| Abstract Scope |
The ab initio calculation of thermodynamic and kinetic properties of materials requires the use of surrogate models to interpolate and generalize computationally expensive first-principles electronic structure calculations within Monte Carlo simulations. A wide variety of surrogate models can be formulated, including lattice model Hamiltonians (e.g. cluster expansions, lattice-dynamical Hamiltonians, magnetic Heisenberg models) to simulate excitations relative to a high symmetry reference crystal structure and off-lattice interatomic potentials to simulate disordered states (e.g. liquids, amorphous states and extended defects). In this talk I will describe mathematical methods to enable the rigorous formulation of surrogate models as well as Bayesian approaches to enable uncertainty quantification. A key requirement of surrogate models is that they reproduce the correct ground states to ensure that the finite temperature phase diagram is qualitatively correct. Bayesian approaches will be described that enforce the correct ground states in the calculation of phase diagrams of multi-component alloys. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Phase Transformations, Machine Learning, |