Deep Learning Leader William Tang
To capture and control fusion in doughnut-shaped tokamaks, scientists must confront disruptions that can halt reactions and damage the facilities. Now an artificial intelligence system under development at PPPL and Princeton University to predict and tame disruptions has been selected as an Aurora Early Science project by the Argonne Leadership Computing Facility, a DOE Office of Science User Facility at Argonne National Laboratory.
The project is one of 10 Early Science Projects for the Aurora supercomputer, which is set to become the first U.S. exascale system in 2021. Aurora will be capable of performing a quintillion (1018) calculations per second — 50-to-100 times faster than the most powerful supercomputers today.
The project aims to develop a method for predicting and controlling disruptions in burning plasma — or nearly self-sustaining — fusion reactions in ITER, the international tokamak under construction in France. “Our research will utilize capabilities to accelerate progress that can only come from the deep learning form of artificial intelligence,” said William Tang, a principal research physicist at PPPL and a lecturer with the rank and title of professor in the Department of Astrophysical Sciences at Princeton University who heads the project.
Deep learning can be trained to solve with accuracy and speed highly complex problems.
The project has greatly benefited from access to the huge disruption-relevant data base of the Joint European Torus (JET) in the United Kingdom, the largest and most powerful tokamak in the world today, and the DIII-D National Fusion Facility that General Atomics operates for the DOE, the largest U.S. tokamak. For ITER, the overall goal will be predictions that are 95 percent accurate with less than 5 percent false alarms, made at least 30 milliseconds before disruptions occur. ☀︎
Physicist Vinícius Duarte, left, and advisor Nikolai Gorelenkov
Birds do it and so do doughnut-shaped tokamak fusion facilities. But tokamak chirping— a rapidly changing frequency wave that can be far above what the human ear can detect — is hardly welcome to researchers who seek to bring the fusion that powers the sun and stars to Earth. Such chirping signals a loss of heat that can slow fusion reactions, a loss that has long puzzled scientists.
Chirps have commonly occurred in the National Spherical Torus Experiment Upgrade (NSTX-U) at PPPL but have been rare in the DIII-D National Fusion Facility tokamak that General Atomics operates for the DOE. Understanding why some tokamaks chirp and some do not could allow researchers to predict and avoid chirping in the international ITER tokamak going up in France.
Scientists led by PPPL researchers have now modeled the conditions that give rise to chirping in plasma, the free electrons and atomic nuclei, or ions, that fuel fusion reactions. The model, developed by PPPL physicist Vinícius Duarte, describes how turbulence can affect the fast ions that are produced by fusion reactions and can act as a wind to cause the plasma chirping. Turbulence can scatter and weaken these ions, keeping them from creating the chirping and reducing plasma heat and fusion power.
The theory indicates why some plasmas chirp and some do not, since turbulence is much less effective in scattering fast ions in some devices than in others. For researchers, the next step will be to use this knowledge to design methods to prevent chirping in present experiments, and to include the methods in the design of future fusion reactors such as ITER. ☀︎
Physicist Isabel Krebs
Sawtooth swings — up-and-down ripples found in everything from stock prices on Wall Street to ocean waves — occur periodically in the temperature and density of the plasma that fuels fusion reactions in doughnut-shaped facilities called tokamaks. These swings can sometimes combine with other instabilities in the plasma to produce a perfect storm that halts the reactions. However, some plasmas are free of sawtooth gyrations thanks to a mechanism that has long puzzled physicists.
Researchers at the Princeton Plasma Physics Laboratory (PPPL) have recently produced complex simulations of the process that may show the physics behind this mechanism, which is called “magnetic flux pumping.” Unraveling the process could advance the development of fusion energy.
The mechanism limits the current in the core of the plasma that completes the twisting magnetic field that confines the hot, charged gas that produces fusion reactions. This development, found in some fusion plasmas, keeps the current from becoming strong enough to trigger the sawtooth instability.
Spearheading the research that uncovered the physics behind the process was physicist Isabel Krebs, lead author of a Physics of Plasmas paper describing the mechanism. Krebs used the PPPL-developed M3D-C1 code to simulate the process on the high-performance computer cluster at PPPL, working closely with physicists Stephen Jardin and Nate Ferraro, developers of the code. “The mechanism behind magnetic flux pumping had not been understood,” Jardin said. “Isabel’s paper describes the process.”
The simulations could lead to measures to avoid the troublesome swings. “This mechanism may be of considerable interest for future large-scale fusion experiments such as ITER,” Krebs said. For ITER, the major international fusion experiment under construction in France, creation of the scenario that the model depicts could produce flux pumping and deter sawtooth instabilities. ☀︎
The Cori supercomputer
Physicist Stephen Jardin
Scientists led by Stephen Jardin, principal research physicist at PPPL and Principal Investigator for the SciDAC Center for Tokamak Transient Simulations, have won 40 million core hours of supercomputer time to simulate plasma disruptions that can halt fusion reactions and damage fusion facilities. The PPPL team will apply its findings to ITER, the international tokamak under construction in France, and the results could help operators of ITER mitigate the large-scale disruptions the facility inevitably will face.
Receipt of the highly competitive 2018 ASCR Leadership Computing Challenge award entitles the physicists to simulate the disruption on Cori, the newest and most powerful supercomputer at the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory. NERSC, a U.S. Department of Energy Office of Science user facility, is a world leader in accelerating scientific discovery through computation.
“Our objective is to model development of the entire disruption from stability to instability to completion of the event,” said Jardin, who has led previous studies of fusion plasma off-normal events. “Our software can now simulate the full sequence of an ITER disruption, which could not be done before.”
The award on Cori, a supercomputer named for Nobel Prize-winning biochemist Gerty Cori that has hundreds of thousands of cores that act in parallel, will enable the physicists to complete in weeks what a single-core laptop computer would need thousands of years to accomplish. The high-performance computing machine will scale up simulations for ITER and perform other tasks that less powerful computers would be unable to complete.
On Cori the team will run the M3D-C1 code primarily developed by Jardin and PPPL physicist Nate Ferraro. The code, developed and upgraded over a decade, will evolve the disruption simulation forward in a realistic manner to produce quantitative results.
The simulations will cover strategies for the mitigation of ITER disruptions, which could develop from start to finish within roughly a tenth of a second. Such strategies require a firm understanding of the physics behind mitigations, which the PPPL team that includes physicist Cesar Clauser and Computational Scientist Jin Chen aims to create. ☀︎
Physicist Francesca Poli
A key goal for ITER, the international fusion device under construction in France, will be to produce 10 times more power than it takes to heat the hot, charged plasma that produces fusion reactions. Among the steps needed to reach that goal will be controlling instabilities called “neoclassical tearing modes” that can cause magnetic islands to grow in the plasma and shut down the reactions.
Outlining a direction for the control of tearing modes has been PPPL physicist Francesca Poli. She and coauthors have published a piece in the journal Nuclear Fusion that describes an approach that for the first time simultaneously simulates the plasma, the magnetic islands, and the feedback control from waves that provide so-called electron cyclotron heating and current drive.
These waves, which accelerate the electrons gyrating along magnetic field lines in tokamak plasmas, can deposit energy into the islands and should have the ability to suppress or stabilize tearing modes in the ITER device. “Basically, you are trying to shoot the beam of electron-cyclotron waves inside the island to replace current that has been depleted from the island,” Poli said. “If you shoot inside the island, that’s okay. If you shoot outside the island, that’s not okay. The simulations we performed can determine the maximum misalignment that can be tolerated and under which conditions experiments should be run.”
Poli conducted the research on PPPL computers running TRANSP, the PPPL-developed code that physicists around the world employ to analyze and predict fusion experiments. Results showed that simulations are generally most helpful when they model the plasma in an integrated manner. Rather than depict a magnetic island in isolation, simulations should take into account what happens to the surrounding plasma, both when the island grows and after the application of radio waves to produce electron-cyclotron heating. Only by using a more integrated model can scientists determine under which conditions the island stabilization will be successful. ☀︎