Machine learning, harnessed to extreme computing, aids fusion energy development

One of the most demanding calculations in fusion science has just been completed by MIT researchers. Even the most advanced supercomputers can't solve this problem by brute force. The researchers used a machine learning methodology to reduce the time required for the solution while maintaining its accuracy.

There is fusion energy.

The same physical process that powers the sun and the stars can be used to power fusion. It requires heating the fuel to a temperature of 100 million degrees above the point where the electrons are stripped from their atoms. Researchers use strong magnetic fields to keep the hot plasma out of ordinary matter. The magnetic field's strength affects the quality of insulation.

Rodriguez-Fernandez and Howard have focused on predicting the performance expected in the SPARC device, a compact, high-magnetic-field fusion experiment, currently under construction by the MIT spin-out company Commonwealth Fusion Systems. The calculation required an extraordinary amount of computer time, over 8 million hours, but how little was used was remarkable.

Computational challenge of fusion energy.

The greatest problem remaining in classical physics is turbulence, which is the mechanism for most of the heat loss. The equations that govern fusion are well known, but analytic solutions are not possible in the regimes of interest, where nonlinearities are important and solutions encompass an enormous range of spatial and temporal scales. Scientists use computers to solve equations. fusion researchers have been at the forefront of computational physics for 50 years.

The only way to reliably predict the temperature and density of the plasma is to use the magnetic field configuration and input power. The external power and heat input from the fusion process are lost in confinement devices. The turbulence is caused by the difference in the temperatures of the plasma core and the edge. Predicting the performance of a self-heated fusion plasma requires a calculation of the power balance between the fusion power input and the losses due to turbulence.

The heat is transported locally by turbulence and is assumed to be at a particular location. A useful prediction requires a self-consistent calculation of the profiles across the entire plasma, which includes both the heat input and turbulent losses. Researchers have come up with an approach that stitches the profiles together from a series of demanding but tractable local calculations. The method works, but it can take a long time to converge since the heat and particles depend on many parameters.

Techniques emerging from the field of machine learning can be used to improve the calculation. Rodriguez-Fernandez and Howard fit a surrogate mathematical model which was used to explore and maximize the local calculations. The system was iterated to a desired level of accuracy after the results of the optimization were compared to the exact calculations at each optimum point. The technique reduced the number of runs by a factor of four.

Confidence in predictions increases with a new approach.

The highest fidelity calculation ever made of the core of a fusion plasma was described in a recent publication. It confirms predictions made with less demanding models. Professor Jonathan Citrin, leader of the fusion modeling group for DIFFER, the Dutch Institute for Fundamental Energy Research, commented: "The work significantly accelerates our capabilities in more routinely performing ultra-high-fidelity tokamak scenario prediction." The ultimate validation test of machine design can be provided with the help of this algorithm.

This technique provides a path to check and calibrate reduced physics models, which run with a small fraction of the computational power, in addition to increasing confidence in the fusion performance of the SPARC experiment. Predicting before each SPARC discharge will be provided by models cross-checked against turbulence simulations, helping to guide experimental campaigns and improve the scientific exploitation of the device. It can be used to improve even simple data-driven models, which run extremely quickly, allowing researchers to narrow down possible experiments or future machines.

More information: Pablo Rodriguez-Fernandez et al, Nonlinear gyrokinetic predictions of SPARC burning plasma profiles enabled by surrogate modeling, Nuclear Fusion (2022). DOI: 10.1088/1741-4326/ac64b2

The story was re-posted by MIT News, a popular site that covers news about MIT research, innovation and teaching.

Citation: Machine learning, harnessed to extreme computing, aids fusion energy development (2022, April 27) retrieved 27 April 2022 from https://phys.org/news/2022-04-machine-harnessed-extreme-aids-fusion.html This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.