DOE Argonne National Laboratory’s fusion simulation reaches Gordon Bell Prize final

Researchers led by Elia Merzari of Pennsylvania State University and Argonne National Laboratory have completed the largest and most detailed simulations to date of a water-cooled lead-lithium (WCLL) blanket, a critical component for the ITER fusion experiment currently under construction in France. The team’s work, which simultaneously modeled six interacting processes including liquid-metal flow and heat transfer under strong magnetic fields, is a finalist for the 2026 Gordon Bell Prize. “This is the first time a blanket has been modeled at this fidelity, including all the physics,” Merzari said, detailing a simulation carried out using exascale supercomputers at Argonne and Oak Ridge National Laboratories.

WCLL Blanket Simulations Advance ITER Fusion Reactor Design

The simulations achieved a resolution previously unattainable for a full-scale WCLL blanket, revealing complex details of fluid behavior within the breeding zone. The team’s calculations used more than 2 billion spectral elements and 2 trillion grid points, making them the largest liquid-metal magnetohydrodynamics (MHD) simulations to date. Understanding how magnetic fields influence the flow is critical, as temperature differences within the liquid lead-lithium drive buoyancy, further altering the flow patterns, particularly near the plasma-facing wall.

A member of the team explained the necessity for detailed modeling to accurately capture these complex interactions. This high-fidelity modeling required a combined effort in fusion energy science and high-performance computing, using two of the Department of Energy’s exascale supercomputers.

The simulations went beyond simple fluid dynamics, incorporating heat transfer, buoyancy, tritium transport and water cooling alongside the magnetic field effects. “My role was to develop the solver that brings MHD into NekRS and validate it against similar problems in the literature to make sure everything was working correctly before we pushed it to larger scales,” said a researcher involved in the computational aspects of the project.

The resulting data provides a valuable complement to physical experiments, offering a cost-effective method for studying and predicting the behavior of fusion systems under extreme conditions. These simulations allow scientists to explore a wider range of designs and operating conditions than would be feasible with purely experimental approaches.

The team anticipates that the detailed data will also facilitate the development of faster, lower-fidelity models, which can then be used for more computationally intensive tasks like AI-driven design optimization and the creation of digital twins. “Now that we’ve shown this is doable, people can use the results or the code as a tool to validate a low-fidelity model,” said a researcher, emphasizing the broader impact of the work.

Beyond the award itself, the simulations represent a step toward realizing practical fusion energy. A team member noted the potential for continued advancements in the field, stating, “There is such interesting physics in simulations at this scale and so much we can learn as we continue working to develop better fusion energy technologies.” The ability to accurately model these complex systems will be essential for optimizing future fusion reactor designs and ultimately achieving fusion as a clean and sustainable energy source.

There is such interesting physics in simulations at this scale and so much we can learn as we continue working to develop better fusion energy technologies.

Elia Merzari, a professor of nuclear engineering and mechanical engineering at Pennsylvania State University with a joint appointment at DOE’s Argonne National Laboratory

Exascale Computing Powers Liquid-Metal MHD Simulations

This level of detail was made possible by access to the Aurora and Frontier exascale supercomputers at the Argonne and Oak Ridge Leadership Computing Facilities, both supported by the U.S. Department of Energy. “We’re tackling a very challenging multiphysics, multiscale problem involving fluid systems with very different properties,” said project lead Elia Merzari, a professor of nuclear engineering and mechanical engineering at Pennsylvania State University with a joint appointment at Argonne National Laboratory.

The team’s approach combined a high-resolution multiscale model with the computational power necessary to capture the detailed interplay of these phenomena. This allowed for examination of how these physical effects influence blanket behavior, revealing that magnetic fields suppress turbulence and reorganize liquid-metal flow, impacting pressure, temperature and heat transfer.

The team used NekRS, a spectral element code, and expanded its capabilities to model magnetohydrodynamic (MHD) effects. Yu-Hsiang Lan, a doctoral student at the University of Illinois Urbana-Champaign, led the effort to optimize the code for simulations of this scale. Complementing NekRS, the open-source OpenMC code calculated neutron transport, providing essential data on heating, tritium production and water activation as inputs to the NekRS simulations.

“It really helped us get this work done in a fraction of the time,” Merzari said. The scale of the simulations, over 2 billion spectral elements and 2 trillion grid points, necessitated the exascale computing power of Aurora and Frontier. “Modeling two trillion grid points is something you can only do with exascale computing,” Merzari said.

NekRS Code Development Enables 2 Trillion Grid Point Models

Achieving this scale demanded years of dedicated code development and validation, focusing on the NekRS computational fluid dynamics code originally built on the earlier Nek5000 framework. NekRS was specifically adapted and improved to use the processing power of GPUs, extending its capabilities for high-fidelity multiphysics simulations.

The simulations ran on more than 9,000 nodes on each system, nearly exhausting the capacity of both Aurora and Frontier, and demonstrated strong scaling, as increasing nodes continued to reduce simulation time. Frontier served as the primary platform for code development, testing and validation, while Aurora hosted the largest simulation, capitalizing on its substantial memory capacity to run the complete multiphysics model, including MHD, at a resolution of 2 trillion grid points.

Aurora and Frontier Validate High-Resolution Multiphysics Results

This technique allowed the team to progressively add detail to a lower-resolution model, significantly reducing computation time. The team’s approach to simulation involved more than 2 billion spectral elements, dividing the blanket model into smaller regions representing its geometry, and required access to exascale computing power from both Aurora Leadership Computing Facility and Frontier Leadership Computing Facility.

In-situ visualization using the open-source Ascent software further enhanced the analysis, enabling researchers to capture high-frequency simulation data and study transient flow behavior in detail. Paul Fischer of the Urbana-Champaign and, along with Misun Min of, mentored doctoral student Yu-Hsiang Lan, who focused on developing and validating the MHD solver within NekRS.

The team’s ability to run NekRS on different types of GPUs proved important, allowing them to perform and scale the simulations effectively across both supercomputing platforms. “In terms of workflow, this was a lifesaver,” Merzari said. Lan is already applying the data generated from these blanket simulations to a separate -led project focused on building simplified system-analysis tools for broader fusion research. The team’s success highlights the importance of sustained investment in both advanced computing resources and the development of sophisticated modeling tools.

My role was to develop the solver that brings MHD into NekRS and validate it against similar problems in the literature to make sure everything was working correctly before we pushed it to larger scales.

Yu-Hsiang Lan, a doctoral student at the University of Illinois Urbana-Champaign

Now that we’ve shown this is doable, people can use the results or the code as a tool to validate a low-fidelity model.

Yu-Hsiang Lan, a doctoral student at the University of Illinois Urbana-Champaign

Modeling two trillion grid points is something you can only do with exascale computing.

Elia Merzari, a professor of nuclear engineering and mechanical engineering at Pennsylvania State University with a joint appointment at DOE’s Argonne National Laboratory
Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Ivy Delaney

Ivy Delaney

Ivy Delaney has been working with neural networks and machine learning since the mid-nineties, back when a couple of hidden layers and a long afternoon of training counted as ambitious. She has watched the field go from academic curiosity to the thing quietly running underneath everything, and she brings that long view to quantum computing. For Quantum Zeitgeist she covers the ground where the two fields meet. That means quantum machine learning and the variational algorithms it leans on, and it also means the less glamorous but more interesting story of classical machine learning already doing real work inside quantum machines, decoding error-correcting codes, calibrating noisy hardware and learning the error models that simulators depend on. She writes about the hardware those algorithms have to run on too, and about the post-quantum cryptography scramble that the same hardware has set off. Her stories typically start with the paper, whether that is peer-reviewed work, conference proceedings or an arXiv preprint, with the source linked so you can hold a claim up against the research it came from. She is unimpressed by benchmarks that will not say what they beat, and by demonstrations that only work in the press release.

Latest Posts by Ivy Delaney: