Abstract Details
Abstracts
Author: Pablo Rodriguez-Fernandez
Requested Type: Poster
Submitted: 2024-04-11 09:25:49
Co-authors: N.T. Howard, C. Holland, J. Candy
Contact Info:
MIT Plasma Science and Fusion Center
77 Massachusetts Ave
Cambridge, Massachuse 02139
United States
Abstract Text:
In recent years, efforts to bring together advances in turbulent transport modeling and more efficient transport solvers are making high-fidelity predictions of kinetic profiles and device performance possible [1-4]. Here, we present the open-source PORTALS framework [1,2], which leverages surrogate modeling and optimization to find solutions to the multi-channel, steady-state transport problem with much fewer evaluations than standard numerical (finite differences Jacobian) methods. PORTALS has been successfully applied to simulate the temperature and density profiles of 40 different plasma conditions, by multi-channel flux-matching simulations of nonlinear turbulence with GPU-accelerated CGYRO [5]. In this work, challenges of surrogate-optimization will be presented, together with the plans to address them, to improve the robustness and automation of the framework, and to enable the coupling with gyrokinetic databases for building global surrogate models. This will become essential for the integration of the PORTALS solver as a central piece for the new SMARTS (Surrogate Models for Accurate and Rapid Transport Solutions) SciDAC partnership.
This work was funded by US DoE DE-SC0024399.
[1] P. Rodriguez-Fernandez et al Nucl. Fusion 62 076036 (2022)
[2] P. Rodriguez-Fernandez et al arXiv (2023)
[3] N.R. Mandell et al arXiv (2022)
[4] A. Di Siena et al Nucl. Fusion 62 106025 (2022)
[5] J. Candy et al Journal of Computational Physics 324, 73–93 (2016)
Comments:
Integrated modeling of fusion plasmas