About this Abstract |
Meeting |
2025 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithms Development in Materials Science and Engineering
|
Presentation Title |
Enhancing the Performance of Constrained Minimization Algorithm |
Author(s) |
Sunyong Kwon, Benjamin Stump, Ying Yang, Alex J Plotkowski |
On-Site Speaker (Planned) |
Sunyong Kwon |
Abstract Scope |
Optimization problems involving equality and inequality constraints frequently arise in materials science, particularly in minimizing free energies. Many algorithms utilize the Lagrangian multiplier method (LMM) which requires either an inverse matrix solver for linear constraints or a nested optimizer for non-linear constraints at each step of gradient descent. This approach leads to a significant increase in computational cost as dimensionality grows. In this presentation, we introduce an algorithm that circumvents the need to solve constraint equations explicitly at every iteration by gradually penalizing the constraints with the number of iterations. To enhance the convergence rate, we augmented the Adam optimization algorithm, a gradient decent method with an adaptive estimator of lower-order moments. The performance of our algorithm will be compared with LMM in a case study of calculating multicomponent-multiphase phase equilibria by the CALPHAD approach. This work was sponsored by the DOE Vehicle Technology Office’s Lightweight Metal Core Program. |
Proceedings Inclusion? |
Planned: |
Keywords |
Computational Materials Science & Engineering, ICME, Modeling and Simulation |