America’s next major exascale supercomputer already has its first scientific challenges lined up. The Discovery supercomputer is not expected to go online until 2028, but the US Department of Energy is preparing nine applications that will be ready to run when the machine becomes available.
The system will succeed Frontier at Oak Ridge National Laboratory and support the Genesis Mission’s goal of accelerating science and engineering through AI.
Here are the nine science projects selected to push Discovery’s power across aviation, fusion, quantum materials, astrophysics, and energy research.
Gas turbine engines are incredibly complex, especially when turbulence, secondary flows, and film cooling begin to affect performance and component temperatures.
That is where GlennHT comes in. Led by Kenji Miki of NASA, the project will help researchers run large eddy simulations for gas turbine engines.
These simulations are more accurate than standard Reynolds-averaged Navier–Stokes simulations, but they are also far more computationally expensive. Discovery’s AMD GPUs are expected to help NASA run routine turbomachinery simulations at unprecedented scale and complexity.
Dark matter remains one of the biggest unsolved questions in astrophysics, and understanding it requires simulations that can connect the very large with the very small.
GIZMO, led by Philip Hopkins of the California Institute of Technology, will use Discovery to model cosmological physics across a huge range of scales.
The project will study everything from the cosmic web to supermassive black holes. It will also test how smaller, faster effects, such as jets and accretion flows, connect with larger cosmic structures.
With the Discovery supercomputer, researchers hope to build bigger, more detailed simulations with richer physics.
Quantum materials could shape future technologies, but simulating how they behave is still limited by available computing power.
QMCPACK, led by Paul Kent of ORNL, focuses on quantum materials and their properties, including magnetism and conductivity.
Understanding these materials more accurately could support future technologies such as quantum computing and new material formulations with fewer critical elements.
The Discovery supercomputer is expected to let QMCPACK run larger, faster, and more accurate simulations, while also helping benchmark other methods.
To understand how protons and neutrons interact, scientists need to model quantum chromodynamics, or QCD.
QUDA_LAPH will support research into QCD, the theory that explains the strong force binding quarks into subatomic particles.
The project is led by Andre Walker-Loud of Lawrence Berkeley National Laboratory and aims to use Discovery to run larger simulations with fewer approximations.
That could help scientists calculate quantum masses more accurately and better understand interactions involving protons and neutrons.
Atomistic simulations depend on extremely precise calculations of total energy and atomic forces.
The Active Learning Framework, or ALF, will help scientists improve those simulations by focusing on machine learning interatomic potentials, or MLIPs. These sit between classical force fields and quantum mechanical methods.
Led by Richard Messerly of ORNL, the project could use Discovery to speed up the full ALF workflow, from selecting data and running quantum mechanical calculations to training models.
The goal is to build MLIPs faster using smaller, more manageable datasets.
Next-generation flex-fuel gas turbines could help address energy demand, but their nitrogen oxides pollution is not yet fully understood.
S3D-Regent, led by Jacqueline Chen of Sandia National Laboratories, will model turbulent flame mixing and complex chemistry inside such turbines.
Discovery will allow researchers to run direct numerical simulations that capture both large-scale turbulence and small-scale instabilities.
The work could support cleaner, safer, and more efficient turbine designs.
Open-fan aircraft engines could reduce fuel consumption by 20 percent, but simulating airflow around the engine and full aircraft is extremely demanding.
The GENESIS project from GE Aerospace Research will tackle that challenge using Discovery.
Led by Eduardo Jourdan de Araujo Jorge Filho, the project will combine machine learning and AI models with a state-of-the-art LES solver.
The simulations could track trillions of variables over billions of time steps, helping pave the way for future open-fan aircraft designs.
Fusion energy promises a potential source of nearly unlimited clean energy, but reactor design remains a major technical challenge.
PIConGPU will support research into laser-driven fusion, including the design of advanced nanostructured targets.
The project is led by Sunita Chandrasekaran of the University of Delaware and includes Helmholtz-Zentrum Dresden-Rossendorf in Germany.
Using Discovery’s computing power and AI optimization, researchers aim to run high-fidelity simulations that can identify more effective fusion target designs faster and with fewer simulations.
Turning methane into methanol has long been a goal in the energy industry, but the chemistry remains difficult to pin down.
SPARC, led by Phanish Suryanarayana of the Georgia Institute of Technology, will study that conversion process.
The project focuses on copper-zeolite catalysts, where scientists still do not fully understand the exact copper-oxygen structure responsible for carbon-hydrogen bond activation. Experimental results also do not align with density functional theory calculations.
SPARC will use many-body random phase approximation calculations, which are more accurate than typical DFT approaches, to produce more reliable benchmark reaction energies and activation barriers.
The nine projects show how Discovery is being prepared before it even comes online.
From aircraft engines and fusion energy to dark matter, quantum materials, and methane conversion, the system is being positioned as a launchpad for AI-driven science from day one.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。