User Summit 2026
Description: You're looking at where and how the planet mixes itself. The movie opens on a slice of the temperature field stirred by geophysical turbulence into thin, convoluted filaments. The view zooms in to expose the smallest scales – even at 4K resolution, each pixel contains roughly 8 grid points of simulation data so what you see is still a compressed view of the underlying data. The final sequence switches to the local mixing rate: bright filaments mark intense bursts where heat, salt, nutrients and gases are irreversibly scrambled. These rare, extreme events set the pace at which the climate system stirs itselves together. The simulation has 4 trillion grid points; one slice printed at viewing-distance quality would cover over 50 square meters, larger than most living-room ceilings.
Tools: Custom GPU spectral Direct Numerical Simulation code; Python rendering.
System: Frontier, Oak Ridge National Laboratory. Up to ~16 000 AMD MI250X GPUs across ~2 000 nodes, 87 TB per three-dimensional snapshot.
Novelty and impact: One of the first exascale datasets of geophysically realistic stratified turbulence, approaching the most extreme oceanic conditions, never before fully resolved. Pushing to larger domains and weaker diffusivity, the flow self-organises into bigger coherent structures with rarer, more extreme mixing events and ever more convoluted filaments – bursts that ocean and atmosphere field probes routinely under-sample. The dataset gives observers a fully resolved reference and modellers a long-missing benchmark for Earth-system turbulent closures crucial to climate research.
PRESENTER: Miles M. P. Couchman
AUTHORS: Adrien Lefauve, Miles M. P. Couchman, Stephen M. de Bruyn Kops
Laser-driven cryogenic hydrogen jet
PRESENTER
Klaus Steiniger
AUTHORS
Richard Pausch, Michael Bussmann, Julian Lenz, Felix Meyer, Martin Rehwald, Ulrich Schramm, Klaus Steiniger, Rene Widera, Karl Zeil
ABSTRACT
This video visualizes compact laser-driven proton acceleration from solid density hydrogen jets. Laser particle acceleration enables ultrahigh dose rate proton sources delivering 4-20 Gy per shot to millimeter-scale volumes on nanosecond timescales, equivalent to around 10^9 Gy/s. This technology provides a unique infrastructure for translational research with protons at ultrahigh dose rates. Experiments conducted at HZDR routinely produce proton bunches with energies greater than 60 MeV for radiobiological in vivo studies. Simulations of these special accelerators require exascale compute capabilities. They produce several hundred Terabyte per second of data, at twenty time steps and more per second. With the particle-in-cell code [PIConGPU](https://github.com/ComputationalRadiationPhysics/picongpu) and its tightly-coupled in-situ visualization plugin ISAAC real time interactive movies of these simulations could be produced on Frontier.
Leadership-Scale AI-Driven CFD on OLCF Frontier: Image-to-Geometry Synthesis, Adaptive Mesh Refinement, and Hypersonic Aerotherm
PRESENTER
Allan Grosvenor
AUTHORS
Allan Grosvenor
ABSTRACT
This visualization showcases MSBAI's end-to-end AI-driven CFD pipeline, developed under ALCC project LRN039 and executing autonomously on OLCF Frontier across aerial vehicle geometries spanning subsonic, transonic, supersonic, and hypersonic regimes. The demonstration traces multiple autonomously driven workflows start to finish: image-to-3D geometry synthesis, automated geometry validation, aerodynamic condition configuration, HPC job submission, multi-cycle solution-adaptive mesh refinement, and convergence of the hypersonic aerothermodynamics solution, rendered in ParaView. Featured cases illustrate flow-field development, residual convergence, surface heat-flux distributions, and adapted-mesh refinement near body and wake.
Tools: OpenFOAM and Pentagrow (grid generation), PyMeshFix and PyMeshLab (surface repair), SU2 (hypersonic aerothermodynamics solver), ParaView (containerized visualization), fine-tuned TRELLIS (image-to-3D synthesis), OpenVSP (parametric aircraft), Apptainer, MPI4Py, DeepSpeed.
System: OLCF Frontier under ALCC allocation LRN039, with companion runs on ALCF Aurora. Grid-generation and repair pipeline shows near-linear strong scaling to 1,000 nodes on Frontier. We have demonstrated utilization of this capability supporting large-scale design studies, or complete system operational characterization in ensembles that would consume a majority of the system at capability scale.
Novelty: First fully autonomous end-to-end demonstration closing the loop from input image to converged modern flight aerodynamics CFD solution at leadership-class scale, removing some of the most manual and arduous labor commonly necessary to generate computational grids in practical engineering, and procure their computational physics solutions.
Leadership-Scale AI-Driven CFD on OLCF Frontier: Image-to-Geometry Synthesis, Adaptive Mesh Refinement, and Hypersonic Aerotherm
PRESENTER
Allan Grosvenor
AUTHORS
Allan Grosvenor
ABSTRACT
This visualization showcases MSBAI's end-to-end AI-driven CFD pipeline, developed under ALCC project LRN039 and executing autonomously on OLCF Frontier across aerial vehicle geometries spanning subsonic, transonic, supersonic, and hypersonic regimes. The demonstration traces multiple autonomously driven workflows start to finish: image-to-3D geometry synthesis, automated geometry validation, aerodynamic condition configuration, HPC job submission, multi-cycle solution-adaptive mesh refinement, and convergence of the hypersonic aerothermodynamics solution, rendered in ParaView. Featured cases illustrate flow-field development, residual convergence, surface heat-flux distributions, and adapted-mesh refinement near body and wake.
Tools: OpenFOAM and Pentagrow (grid generation), PyMeshFix and PyMeshLab (surface repair), SU2 (hypersonic aerothermodynamics solver), ParaView (containerized visualization), fine-tuned TRELLIS (image-to-3D synthesis), OpenVSP (parametric aircraft), Apptainer, MPI4Py, DeepSpeed.
System: OLCF Frontier under ALCC allocation LRN039, with companion runs on ALCF Aurora. Grid-generation and repair pipeline shows near-linear strong scaling to 1,000 nodes on Frontier. We have demonstrated utilization of this capability supporting large-scale design studies, or complete system operational characterization in ensembles that would consume a majority of the system at capability scale.
Novelty: First fully autonomous end-to-end demonstration closing the loop from input image to converged modern flight aerodynamics CFD solution at leadership-class scale, removing some of the most manual and arduous labor commonly necessary to generate computational grids in practical engineering, and procure their computational physics solutions.
Leadership-Scale AI-Driven CFD on OLCF Frontier: Image-to-Geometry Synthesis, Adaptive Mesh Refinement, and Hypersonic Aerotherm
PRESENTER
Allan Grosvenor
AUTHORS
Allan Grosvenor
ABSTRACT
This visualization showcases MSBAI's end-to-end AI-driven CFD pipeline, developed under ALCC project LRN039 and executing autonomously on OLCF Frontier across aerial vehicle geometries spanning subsonic, transonic, supersonic, and hypersonic regimes. The demonstration traces multiple autonomously driven workflows start to finish: image-to-3D geometry synthesis, automated geometry validation, aerodynamic condition configuration, HPC job submission, multi-cycle solution-adaptive mesh refinement, and convergence of the hypersonic aerothermodynamics solution, rendered in ParaView. Featured cases illustrate flow-field development, residual convergence, surface heat-flux distributions, and adapted-mesh refinement near body and wake.
Tools: OpenFOAM and Pentagrow (grid generation), PyMeshFix and PyMeshLab (surface repair), SU2 (hypersonic aerothermodynamics solver), ParaView (containerized visualization), fine-tuned TRELLIS (image-to-3D synthesis), OpenVSP (parametric aircraft), Apptainer, MPI4Py, DeepSpeed.
System: OLCF Frontier under ALCC allocation LRN039, with companion runs on ALCF Aurora. Grid-generation and repair pipeline shows near-linear strong scaling to 1,000 nodes on Frontier. We have demonstrated utilization of this capability supporting large-scale design studies, or complete system operational characterization in ensembles that would consume a majority of the system at capability scale.
Novelty: First fully autonomous end-to-end demonstration closing the loop from input image to converged modern flight aerodynamics CFD solution at leadership-class scale, removing some of the most manual and arduous labor commonly necessary to generate computational grids in practical engineering, and procure their computational physics solutions.
Pressure Fluctuations and Vortical Structures in Turbulence at High Reynolds Number
PRESENTER
Daniel L. Dotson
AUTHORS
Daniel L. Dotson, Rohini Uma-Vaideswaran, P.K. Yeung
ABSTRACT
The pressure within an incompressible fluid satisfies a Poisson equation whose source term is the second invariant Q of the velocity gradient tensor, which is positive in rotation-dominated regions and negative where strain dominates. Within a turbulent flow, one might therefore expect pressure minima to shadow the small filamentary vortical structures known as 'worms' which emerge intermittently. However, the Poisson equation is inherently nonlocal, and so pressure fluctuations depend not just on the vorticity within individual worms, but also on their collective spatial organization. Worms are not uniformly distributed within a turbulent flow, and instead seem to organize into progressively larger tangles and braids, a multi-scaled hierarchy which can be revealed by spectrally sifting the vorticity field with a low-pass filter. Using a GPU-accelerated shear-warp volume rendering visualization technique, implemented in Fortran with OpenMP, we find that turbulence simulations at high Reynolds number exhibit a remarkable correspondence between moderately large negative pressure fluctuations and a hierarchy of vortical structures. This picture lends credence to the idea of 'vortices on all possible scales' which inspired classical turbulence theories. Ongoing work aims to identify the statistical correlations and conditioning that best capture these observations, in order to help explain the pronounced negative skewness in the pressure probability distribution and other signatures of turbulence intermittency.
The data visualized here, which was generated on Frontier, is accessible through the Johns Hopkins Turbulence Database. For further details on our most recent simulations, see Yeung et al. J. Fluid Mech. Vol. 1019 (2025).
Simulation of the Satellite Tobacco Mosaic Virus in Explicit Water
PRESENTER
David Pugmire
AUTHORS
David Pugmire, Dilip Asthagiri
ABSTRACT
Simulation of a satellite tobacco mosaic virus in explicit water. The system comprises slightly over 1 million atoms. The system was built at Urbana-Champaign around 2006 and represented the first major virus particle simulation. The original work required 256 nodes of the NCSA Altix supercomputer, but now we routinely run this for testing on a single node of Frontier.
This visualization summarizes a 1,000-year simulation of forest dynamics across the contiguous United States using GGap, a GPU-native, individual-tree, agent-based forest model designed for exascale computing and executed on Frontier at OLCF. The full simulation tracks 712 million individual tree agents across 1,424 spatially coupled sites and 235 tree species, with inter-site seed dispersal represented through a haversine-distance network. The central map displays NLDAS-2-derived forest fraction at 0.125° spatial resolution. Ten numbered markers identify representative sites spanning the major ecoregions of the contiguous United States, ranging from Pacific Northwest mixed conifer forests to Gulf Coast loblolly pine systems. For each site, two companion panels summarize long-term ecosystem development: a stacked area plot showing temporal shifts in species composition as a percentage of stand biomass carbon over the millennium of simulated succession, and a bar chart illustrating the tree diameter distribution at simulation year 1,000. The visualization highlights strong regional differentiation in species dominance, productivity, and forest structure emerging entirely from climate-driven competition resolved at the level of individual trees. The simulated communities—including Calocedrus and Pseudotsuga in the Pacific Northwest, Pinus banksiana in the Upper Midwest, and Pinus taeda along the Gulf Coast—closely reproduce forest associations documented by the USDA Forest Inventory and Analysis. These results demonstrate the potential of exascale, high-fidelity forest simulation to capture long-term U.S. forest dynamics under changing weather and disturbance regimes, while supporting next-generation Earth system modeling and national energy resilience strategies.
PRESENTER: Xi Zhang
AUTHORS: Xi Zhang, Bin Wang, Herman H. Shugart, Forrest M. Hoffman, Robert Patton
Trajectories of Wind in Urban Environments
PRESENTER
Kenneth Moreland
AUTHORS
Kenneth Moreland and Matthew Norman
ABSTRACT
Wind helps clear pollutants from urban environments, but only when airflow carries particles away from the city. These paired visualizations depict wind trajectories through a section of New York City using animated particles. The flow reveals how air is funneled between buildings, while also exposing recirculating zones where pollutants can become trapped and accumulate near street level.
Trajectories of Wind in Urban Environments
PRESENTER
Kenneth Moreland
AUTHORS
Kenneth Moreland and Matthew Norman
ABSTRACT
Wind helps clear pollutants from urban environments, but only when airflow carries particles away from the city. These paired visualizations depict wind trajectories through a section of New York City using animated particles. The flow reveals how air is funneled between buildings, while also exposing recirculating zones where pollutants can become trapped and accumulate near street level.
Visualization of AI-Driven Hyper-Resolution Climate Downscaling with ORBIT-2
PRESENTER
David Pugmire
AUTHORS
Xiao Wang, Jong-Youl Choi, Takuya Kurihaya, Isaac Lyngaas, Hong-Jun Yoon, Xi Xiao, David Pugmire, Ming Fan, Nasik Muhammad Nafi, Aristeidis Tsaris, Ashwin M. Aji, Maliha Hossain, Mohamed Wahib, Dali Wang, Peter Thornton, Prasanna Balaprakash, Moetasim Ashfaq, Dan Lu
ABSTRACT
Visualization of ORBIT-2 climate downscaling results generated using ParaView from data produced on the Frontier exascale supercomputer at Oak Ridge National Laboratory. The rendering highlights the model’s ability to produce high-resolution global climate fields from coarse input data, revealing fine-scale atmospheric and geographic structures that are difficult to capture with traditional downscaling approaches. The visualization demonstrates the scalability and fidelity of ORBIT-2 across large spatial domains, emphasizing both the detail preserved at hyper-resolution and the efficiency of processing massive climate datasets.
Visualization of Core-Collapse Supernova Simulation Data
PRESENTER
David Pugmire
AUTHORS
Sudarshan Neopane, William Hix, Bronson Messer, David Pugmire
ABSTRACT
Visualization of large-scale core-collapse supernova (CCSN) simulation data generated on the Frontier exascale supercomputer at Oak Ridge National Laboratory. Iso-surfaces of helium and nickel isotopes were generated using VisIt, and final rendering was performed in Blender. Core-collapse supernovae are explosions of massive stars that have exhausted the nuclear fuel in their cores, enriching the interstellar medium with elements ranging from oxygen to iron and driving galactic chemical evolution. The visualization highlights turbulent mixing, shock-driven instabilities, and the evolving spatial distribution of helium and nickel isotopes produced during the explosion.
