Quantum Computing
Tomorrow’s quantum computing solutions hold the potential to solve mission-critical HEDS research priorities, offering unparalleled resolution, speed, and physics fidelity. With this aim in mind, LLNL scientists are engaged in efforts to harness the power of quantum technology to solve increasingly complex national security challenges.
LLNL’s two-stage approach for algorithm development focuses on mission-relevant applications. Today, we are in the first stage of these efforts, with our focus on noisy intermediate-scale quantum (NISQ) devices. We need to understand what can be accomplished using today’s noisy, error-prone platforms, as well as their limits. For example, on today’s platforms, we can use quantum hardware to emulate a quantum many-body system, enabling us to study challenges such as spin systems, nuclear physics in plasma environments, and material properties at extreme conditions.
At the same time, LLNL researchers want to be ready for the post-NISQ era. The hardware in the post-NISQ era will be characterized by large-scale, fault-tolerant, universal quantum computing, where the quantum computers will provide quadratic to exponential speedup of time-to-solution and numerical accuracy of HEDS simulations (~103 qubits). Today, LLNL scientists are developing quantum algorithms for both NISQ era and post-NISQ era machines, including algorithms that can tackle complex research challenges related to partial differential equations (radiation-hydrodynamic, plasmas), spin systems, nuclear physics, and material properties at extreme conditions.
Learn More
Contact
- Frank Graziani
- graziani1 [at] llnl.gov (graziani1[at]llnl[dot]gov)
Artificial Intelligence and Machine Learning
Artificial intelligence and machine learning (AI/ML) are rapidly becoming core infrastructure for scientific discovery, and high energy density science (HEDS) is an especially strong match. HED plasmas exhibit nonlinear, multiphysics behavior, with tight coupling among radiation transport, hydrodynamics, atomic physics, and kinetic effects. Experiments and simulations therefore produce high-dimensional parameter spaces and large, complex datasets, making it challenging to identify optimal designs and choose future research directions.
Researchers at LLNL use AI/ML to:
- Support scientifically defensible decision-making using predictive modeling with uncertainty quantification, spanning integrated experimental and simulation datasets
- Explore and optimize large design spaces efficiently
- Accelerate computational workflows while preserving key physics
Integration of AI/ML into predictive modeling enables probabilistic forecasts and explicit uncertainty bounds. For example, LLNL researchers have combined radiation-hydrodynamics simulations, National Ignition Facility experimental data, and Bayesian inference within a physics-informed deep-learning framework to estimate fusion ignition likelihood for proposed shots before they are executed. Rather than focusing on post-shot tuning, the method learns shot-to-shot variability from uncertain inputs (e.g., conditions and modeling assumptions) and propagates those uncertainties through fast neural-network surrogates. The result is risk-aware, pre-shot design and planning consistent with observed performance.
LLNL is also reshaping the computational experiment cycle around large ensembles rather than a small number of multiphysics “hero” simulations. Project ICECap (Inertial Confinement on El Capitan) leverages the Lab’s exascale supercomputer El Capitan to enable systematic model exploration, optimization, and design under uncertainty. The framework combines graphics-processing-unit‒accelerated multiphysics simulations, ML surrogates that speed up expensive sub-physics, and coordinated multi-fidelity studies across 1D, 2D, and 3D models. Simulation, workflow automation, and decision-making are integrated into a single engineered workflow.
As an embedded capability, AI/ML is also replacing costly sub-grid physics invoked by hydrodynamics codes at every time step and grid point. A representative HEDS example is deep-learning–accelerated, non-local thermodynamic equilibrium (NLTE) atomic modeling. NLTE calculations can be a dominant cost driver in radiation–hydrodynamics simulations of inertial confinement fusion and other HED systems. Early studies demonstrated that neural networks could replace inline NLTE calculations while preserving key radiative quantities and delivering substantial speedups. Current efforts focus on improving physical fidelity and robustness for end-to-end deployment in production simulations.
Looking ahead, AI/ML in HEDS will be guided by, and will contribute to, the Genesis Mission, a national initiative led by the U.S. Department of Energy, aimed at building an integrated, AI-enabled scientific discovery platform. LLNL research efforts will prioritize uncertainty-quantified predictive modeling, large-scale ensemble exploration, and tightly integrated experiment–simulation–ML workflows. These capabilities support design optimization, risk assessment, and campaign planning in regimes where experimental iteration is inherently limited, making AI/ML a foundational capability for HEDS design and discovery.
Learn More
- LLNL researchers employed an AI-driven model to predict fusion ignition shot
- High performance computing, AI, and cognitive simulation helped LLNL conquer fusion ignition
- Breaking the ICE: ICECap looks to use exascale fusion simulations to pioneer digital design
- Autonomous multiscale: Embedded machine learning for smart simulations
Contact
- Min Sang Cho
- cho28 [at] llnl.gov (cho28[at]llnl[dot]gov )
