OFFICE OF ADVANCED SIMULATION AND COMPUTING AND INSTITUTIONAL R&D PROGRAMS
Welcome to our summer 2026 issue of the ASC newsletter - published quarterly to socialize the impactful work performed by the National Nuclear Security Administration (NNSA) laboratories and our other partners. This edition begins with a special thanks to Lawrence Livermore National Laboratory (LLNL), Nevada National Security Site (NNSS), the Trilab planning team, and the speakers who presented during this year’s ASC Program Annual Meeting (APAM) in May 2026. Other featured highlights in this edition include:
- Use of ASC’s high-performance computer (HPC) resources in support of Aires Tide, a proof-of-concept flight test vehicle developed to accelerate system design to flight testing at lower cost. In the banner above, Sandia National Laboratories (SNL) researchers prepare the 1:2 scale Aires Tide flight vehicle for release from a balloon at Dugway Proving Ground, Utah (completed in May).
- An update on the technology planned for the “Mission” HPC system at Los Alamos National Laboratory (LANL).
- SNL’s Accelerated Model Development (AMD) team automating model creation and simulation workflows to enable a W80-5 subassembly model to be built and simulated in a single day.
Please join me in thanking the professionals who delivered the achievements highlighted in this newsletter and on an ongoing basis, all in support of our national security mission.
Dr. Stephen Rinehart
Assistant Deputy Administrator, ASC
LLNL hosted a successful 2026 ASC Program Annual Meeting (APAM) at the Remote Sensing Laboratory in Las Vegas.
The 2026 APAM, held May 19-21, 2026 at the Remote Sensing Laboratory on Nellis Air Force Base in Las Vegas and hosted by LLNL, brought together about 175 participants from across the NNSA ASC enterprise, including labs, plants, sites, and federal program managers. The meeting featured program updates from ASC HQ along with presentations on NNSS activities, advanced high explosives (HE) modeling, verification and validation, integrated codes, Artificial Intelligence for Nuclear Security (AI4NS) activities, digital engineering, quantum computing, and artificial intelligence (AI) coding agents.
Each day concluded with breakout sessions focused on key themes such as digital transformation, formal methods, system software strategy, future design and production simulation, metrics, workload characterization, the ASC computing strategy, and AI coding practices. The event was described as a valuable and highly informative opportunity for attendees to see the breadth and impact of the ASC mission, while also strengthening connections across the enterprise (LLNL-ABS-2022365).
ASC supported Aires Tide, a national security innovation developed using AI, HPC, and additive manufacturing under the Genesis Mission.
In June 2026 the NNSA announced Aires Tide, an NNSA-led proof-of-concept flight test vehicle developed using AI, HPC, and additive manufacturing to move from system design to flight testing on a sharply compressed timeline and at lower cost.
On November 24, 2025, President Trump issued an Executive Order launching the Genesis Mission, a historic effort led by the DOE to establish an interconnected web of national laboratory supercomputers empowered by AI. NNSA leveraged Genesis to develop, design, and demonstrate Aires Tide, marking the first tangible demonstration of the platform.
Aires Tide illustrates NNSA’s ability to apply advanced tools to rapidly design and deliver national security solutions, leading to a product developed 15 times cheaper and seven times faster than traditional manufacturing. NNSA’s National Laboratories – LANL, LLNL, and SNL – worked in close collaboration with NNSA’s Kansas City National Security Campus (KCNSC), to showcase the Nuclear Security Enterprise’s ability to move faster to meet urgent mission needs.
“Aires Tide is a remarkable early demonstration of how NNSA is putting the Genesis Mission into action,” said NNSA Administrator Brandon Williams. “President Trump has made it clear that America must lead the world in artificial intelligence and use emerging technologies to strengthen our national security. By combining AI, high-performance computing, and additive manufacturing, we are pioneering a faster, more efficient model to design and produce capabilities for national security while keeping human judgment firmly at the center.”
In May, Nuclear Security Enterprise scientists conducted two successful flight tests of Aires Tide, dropping the vehicle from 32,000 feet at the U.S. Army’s Dugway Proving Ground in Utah. Data from the flight tests will be used to optimize future systems developed using the same design and manufacturing model.
Two of NNSA's flagship supercomputers – Venado and El Capitan (funded by the ASC program) – were used to enable the design of Aires Tide. The project reflects a broader NNSA effort to use supercomputing platforms and cutting-edge additive manufacturing technologies to shorten development cycles and improve efficiency, strengthening the enterprise’s ability to respond to emerging national security challenges and keeping America safe (see the NNSA press release).
Improving time-to-insight for national security science with the Mission system at LANL.
The fifth Advanced Technology System (ATS-5) in the ASC program, called “Mission,” will be a critical resource supporting the NNSA stockpile management and modernization programs. Mission will support current and future simulation codes, future AI-augmented simulations, and AI-augmented workflows to support the design, assessment, and certification of the U.S. stockpile.
Mission will replace Crossroads (ATS-3) at LANL as a primary computing resource to support the stockpile modernization mission and current and planned life extension programs (LEPs). Delivery of the Mission system will begin in 2027 with a planned transition to classified production computing in 2028.
Codesigned by LANL for NNSA workloads, Mission will be built by Hewlett Packard Enterprise (HPE) and NVIDIA. As the first ATS resource to combine multiple architectures to accelerate the full end-to-end workflow, Mission will improve NNSA’s ability and agility to manage the stockpile with modeling and simulation (mod/sim) and AI.
Mission’s Three Integrated Partitions
Mission is the first ATS with diverse, purpose-built technologies to support NNSA’s most pressing workloads in support of the weapons program. Rather than a single, general-purpose architecture, Mission is composed of three purpose-built architectures (three specialized partitions), each optimized to support the most important components of the complex workloads of the future: Ranger, Sandstone, and Starlight.
- Ranger will continue LANL’s strong support for the highest-complexity, 3D multi-physics simulations. Ranger supports sparse optimized compute for large-scale, high-complexity modeling and simulation with large memory footprint and sparse memory access requirements. This Vera Vera central processing unit (CPU) partition delivers high-frequency, high-bandwidth cores with large aggregate and per core memory.
- Sandstone will support rapid turnaround of moderate-scale 3D simulation suites and provide an optimized environment for large-scale AI training. This partition supports dense-optimized compute for AI and machine learning (ML) training, including models with 500+ billion parameters, and moderate-scale 3D simulations with dense memory access and high floating point compute intensity. This Vera Rubin NVL4 graphics processing unit (GPU) partition provides dense-optimized compute and high-bandwidth memory.
- Starlight will provide the first scale-up architecture for frontier AI inference and strong scaling of modeling and simulation. This partition supports AI inference-optimized compute for advanced frontier models, agentic AI workflows, and strong-scaling modeling and simulation with large shared-memory and extreme-bandwidth scale-up network requirements. This Vera Rubin NVL72 partition provides the same Vera Rubin GPUs as Sandstone, but with 72 GPUs interconnected into a scale-up NVLINK fabric for a massive, shared memory footprint with an order of magnitude higher interconnect bandwidth.
These architectures will be integrated with a common software environment and shared high-performance storage optimized for modeling and simulation and AI. All flash high-performance storage is connected to each of these partitions with flexible configurations optimized for modeling and simulation and AI/ML needs. Underpinning these capabilities is Mission’s robust infrastructure supporting workflow management, Python notebooks, and planned compute lifecycles to ensure operational continuity and maximized mission impact.
Different from previous systems, Mission is driven by deep workload analysis and designed around NNSA’s most pressing application bottlenecks, such as memory-bound sparse accesses and branching found in complex, multi-physics codes. Mission is the first system designed to optimize the end-to-end performance of workflows that couple modeling and simulation, AI, and data analysis.
Data and Storage
The Mission storage system is an all-flash, high-bandwidth, high-input/output operations per second (IOPs) configuration with large capacity. Mission will provide two storage environments using this hardware. One storage environment will be based on the Lustre parallel file system, which has served as a high-performance scratch space for modeling and simulation on every ATS platform at LANL.
In addition to Lustre, an AI-optimized storage environment based on the Hammerspace parallel Network File System (pNFS) will be available for high-IOP workloads commonly found in AI. This dual-stack approach allows for sharing of resources (capacity and performance) for modeling and simulation and AI with flexible balancing of storage capacity and performance, which was not possible in prior systems.
The Goal: Improving Time-To-Insight
Mission will drive down time-to-solution for large-scale simulations significantly further than Crossroads while providing much higher memory capacity per core/node. Mission will enable dense compute-bound simulations and AI workloads to achieve multiple factors of speedup. AI training workloads will be accelerated through the most advanced CPU/GPU capability available and a robust set of best of breed, industry standard, software tools. Scale-up workloads (frontier AI models and strong-scaled simulations) will have a first-of-its-kind scale-up network on Mission. Taken together, Mission is more than the sum of its partitions. It provides a first-of-a-kind, capability-class system that will significantly reduce time-to-insight for NNSA’s most challenging problems.
LANL staff recently tested NVIDIA’s Vera CPU platform in preparation for deployment in upcoming platforms, and the early performance results are impressive. Vera’s memory bandwidth and core performance is on track to meet expectations. Similar testing will be conducted on the Rubin GPU followed by the deployment of a number of early testbeds prior to full-system deployment.
Early application readiness activities have already begun through Mission’s Center of Excellence. These activities will lay the early groundwork to ensure that both ASC codes and the Mission system are ready to meet the needs of the NNSA on day one. For more information on Mission, see the NNSA press release and the LANL press release (LA-UR-26-25071).
A W80-5 subassembly model was built and simulated in a single day at SNL.
Within the SNL Accelerated Model Development (AMD) Initiative, support was provided to the W80-5 systems engineering team in the development of a new subassembly redesign. Analysts outside the systems group reached out to SNL ASC AMD team members to inquire whether AMD tools could speed up modeling and simulation efforts to inform the technical basis for a redesign. By automating model creation and simulation workflows, the AMD team quickly produced results in one day compared to traditional modeling and simulation efforts that can take weeks to generate. Additionally, due to the successful utilization of AMD tools, the potential to use new AI-driven tools to further accelerate critical system analyses was highlighted. The success of this effort led the W80-5 structural dynamics analysis team to now leverage automated model-building tools to inform design decisions concurrently with design development that simply cannot be met with their current workflows (SAND2026-21731M).
LANL’s Discrete Diffusion Monte Carlo (DDMC) scheme accelerates and improves the fidelity of radiation hydrodynamics simulations.
The standard Implicit Monte Carlo (IMC) approach fails to capture the diffusion limit. Adhering to the diffusion limit is necessary to model thermal radiation correctly in optically thick materials. LANL’s thermal radiation transport (TRT) team, Jayenne, has developed a novel DDMC acceleration scheme for unstructured spatial meshes. The new DDMC method improves the performance and accuracy of radiation hydrodynamics calculations on unstructured meshes. Jayenne performance more than doubled on the classic crooked pipe numerical experiment while also demonstrating improved convergence of radiation propagation speeds in optically thick materials. The DDMC implementation will enable both faster and more accurate simulations in coupled radiation hydrodynamics simulations on unstructured meshes.
By design, the DDMC algorithm strictly enforces the diffusion limit. This is illustrated using the classic crooked pipe problem as an example. The rate that radiation travels through the optically thin material is driven by a method’s accuracy in accounting for radiation diffusing into the optically thick walls. Figure 9 shows how DDMC drastically reduces excessive radiation leakage into the optically thick walls relative to the standard IMC approach. In the right half of the figure, such leakage degrades the rate at which the radiation wave propagates through the optically thin material. The left half of Figure 9 highlights this effect, showing temperature as a function of time at two fiducial points. The DDMC algorithm also improves computational performance by replacing a long series of tightly coupled MC collisions with single discrete jumps between cells. This results in more than a double in speedup for the classic crooked pipe problem shown here.
The challenge of this work was developing an accurate diffusion discretization scheme that preserves accuracy on unstructured grids while strictly enforcing positivity in the resulting DDMC probability distributions. The LANL team adapted an existing second order finite volume diffusion discretization for unstructured meshes developed by Maire and Breil [1] to form suitable DDMC probability distributions. The new DDMC algorithm needed a modification to the original continuity condition to ensure positivity and a physical solution; this comes at the cost of slightly degrading the second order accuracy of the original discretization for highly skewed meshes. Future work will examine alternative approaches to preserve positivity and second order spatial accurate solutions in these extreme limits. Development of DDMC helps ensure LANL’s radiation transport capabilities are accurate across the broad range of optical thicknesses required by mission applications (LA-UR-26-23391).
[1] Maire, Pierre-Henri, and Jérôme Breil. "A nominally second-order accurate finite volume cell-centered scheme for anisotropic diffusion on two-dimensional unstructured grids." Journal of Computational Physics 231.5 (2012): 2259-2299.
LANL collaboration supports early fault-tolerant neutral atom quantum computers for nuclear dynamics.
Fault-tolerant quantum computing is essential for solving scientifically and nationally important problems that exceed the capabilities of classical HPC. A major barrier to reaching the “megaquop” regime—where error-corrected quantum computers can perform millions of reliable operations and deliver beyond-classical simulation capabilities—is the substantial resource overhead imposed by quantum error correction. Given advances like these, we have increasing optimism in our ability to address mission-relevant problems with quantum computing.
LANL collaborated with QuEra Computing and university partners to co-design and publish the transversal Space-Time Efficient Analog Rotation (STAR) architecture in PRX Quantum. The team developed a fault-tolerant quantum simulation framework (Figure 10) optimized for neutral-atom quantum hardware that eliminates key bottlenecks associated with magic-state synthesis (Figure 10a), lattice-surgery routing (Figure 10b), and modifications for high-rate codes. Through detailed circuit-level simulations using hardware-informed noise models, the researchers demonstrated a scalable approach that leverages reconfigurable connectivity and large-scale parallelism to significantly improve the efficiency of structured quantum simulations.
The transversal STAR architecture reduces quantum simulation resource requirements by orders of magnitude, enabling megaquop-scale simulations with approximately half the physical qubits required by conventional fault-tolerant approaches and up to 250 times faster execution. The architecture provides a credible near-term pathway to beyond-classical simulations in materials science, condensed matter physics, and non-equilibrium dynamics, accelerating progress toward practical fault-tolerant quantum computing and reinforcing LANL's leadership in quantum information science and co-designed computing architectures (LA-UR-26-25824).
Reference: Ismail, Refaat, I-Chi Chen, Chen Zhao, Ronen Weiss, Fangli Liu, Hengyun Zhou, Sheng-Tao Wang, Andrew Sornborger, and Milan Kornjača. "Transversal Architecture for Megaquop-Scale Quantum Simulation with Neutral Atoms." PRX Quantum 7, no. 2 (2026): 020343.
LLNL’s cross-code comparisons show tight agreement in National Ignition Facility capsule simulations.
Simulations of National Ignition Facility (NIF) capsules play a critical role in the pursuit of high-yield, high energy density (HED) experiments. Different computational codes employ different numerical discretizations and physics models, which can lead to varying results depending on the code used. A study has been carried out to quantify the spread in predicted capability when using various radiation hydrodynamics (rad-hydro) codes at LLNL to simulate igniting NIF capsules. These simulations account for a wide range of physics in the HED regime, including multimaterial hydrodynamics, thermal radiation, heat conduction, 3T physics, dense plasma models, thermonuclear burn, and more. Results obtained thus far indicate that quantities of interest predicted by the codes HYDRA, Ares, and MARBL, which use different approaches to model HED experiments, are in close agreement. The plots below show that the spread in predicted yield, radii of various layers, and mass-averaged temperatures is small (LLNL-ABS-2020853).
Digital thread CREO plugin accelerates thermal battery design at SNL.
Within the SNL ASC Accelerate Digital Engineering (ADE) Initiative, a custom CREO plugin is being developed to automate computer-aided design (CAD) model and drawing generation for thermal battery designs, enhancing the digital thread workflow. By enabling rapid, automated generation of 3D CAD models and 2D drawings, the digital thread reduces design cycle time and cost, directly supporting the modernization goals of the SNL Nuclear Deterrence (ND) program. The new plugin minimizes designer workload, scales to other component technologies, and ensures CAD models meet next assembly requirements, thereby increasing the ND program’s confidence in and utilization of ASC’s advanced design capabilities (SAND2026-21731M).
SNL supports the NNSA Trilabs in studying application hardware usage, exploring ways to leverage powerful GPU features.
At the recent El Capitan Center of Excellence Hackathon, SNL helped create a lightweight tool to profile specific double-precision, 64-bit floating point precision (FP64) and integer usage at scale. Teams from SNL and LANL then used the tool to analyze applications. HPC platforms have long relied on vector instructions, which apply the same operation across many data elements in parallel. GPUs extended this model with specialized matrix instructions, first appearing prominently in the Sierra supercomputer through NVIDIA’s Volta V100 tensor cores. These matrix instructions boost AI performance but have a more complex role in traditional simulations. By collecting data across the Tri-lab and working with AMD, HPE, and NVIDIA, the ASC program is exploring ways to better leverage these powerful GPU features (SAND2026-21731M).
Livermore Computing rapidly defeats cybersecurity vulnerability.
Livermore Computing (LC) responded quickly to the Linux “Copy Fail” vulnerability, a kernel privilege-escalation flaw, after being alerted on April 29th by a LLNL employee. John Allen, as LC Organizational Information System Security Officer (OISSO), quickly coordinated the response, informing the LC Tri-Lab Operating System Stack (TOSS) Lead Developer, Jim Foraker, and System Administrators, Py Watson and Rigo Moreno Delgado. The team confirmed LC systems were vulnerable, even though one proof-of-concept failed on some setups because of factors like Python version and whether “su” (substitute user) was setuid Root, while a second exploit worked with only minor adjustments. That helped them verify the urgency and move fast. Jim Foraker created and tested a live kernel patch, or “kpatch,” for both Trilab Operating System Stack-4 (TOSS-4) and TOSS-5, also checking that it would not conflict with an existing patch from a prior issue. Py and Rigo helped validate the exploit behavior and Py handled rapid deployment across systems using Ansible. LC had the mitigation fully deployed by midnight, and by the next morning John Allen shared the fix with the broader TOSS community, ahead of Red Hat’s official patch (LLNL-ABS-2022367).
Welcome Aboard...
LANL ASC program
Jeff Haack is the new ASC Integrated Codes DevOps Project Leader at LANL. Jeff is a scientist in the Computational Physics and Methods group with extensive experience in HPC across a wide range of applications. He first came to LANL as a summer student in 2008, working on asymptotic analysis of transport equations. He earned his PhD in Mathematics from the University of Wisconsin and conducted postdoctoral research at the University of Texas at Austin, developing spectral methods for the nonlinear Boltzmann equation and its application to neutral gases and kinetic plasmas. He joined LANL in 2014 working on theory and software for kinetic effects in ICF modeling and experiments. Jeff currently contributes to the infrastructure, GPU porting, and performance optimization for ASC radiation transport Capsaicin project, while maintaining active, collaborative relationships with other integrated code teams as well as external library developers and hardware vendors. In his free time, he enjoys playing outdoors with his cocker spaniel, baking, and listening to Chicago Cubs games on the radio.
SNL ASC program
John Ossorgin has been at SNL for a little over six years working in Global Security focusing on all aspects of software development and delivery for satellites and ground stations. Currently, he is the newest member of the ASC DevSecOps team, assisting on the upcoming SIERRA code build system transition to Spack. John received both his Bachelor's and Master's degrees in Computer Science from New Mexico State University where he focused on AI and HPC research. As a native New Mexican who requires a steady supply of chile (red over green), John enjoys spending time with family, friends, and most importantly, his dog, Phife. Most weekends, John can be found at a local coffee shop with his pup.
NNSA LDRD/SDRD Quarterly Highlights
LLNL’s project DarkStar is investigating AI and ML applications to scientific problems of complex hydrodynamics, shockwave physics, and energetic materials.
LLNL’s LDRD-funded Strategic Initiative Project DarkStar made advancements in controlling material deformation by investigating the scientific problems of complex hydrodynamics, shockwave physics, and energetic materials. The purpose of this project was to take the quantum leap from understanding dynamic material response to controlling it. The scientific and technological achievements attained will enable a completely new paradigm for weapons physics and design and advance the U.S. national security posture. Today, foundational methods in surrogate modeling, multimodal anomaly, detection, and explainable reasoning are reusable across mission spaces. Advancements in predicting and controlling instability formation were demonstrated on both HE experiments and ICF studies. Read more about LLNL’s Project DarkStar.
SDRD embraces AI and ML to advance the NNSS mission.
Within the SDRD AI and ML portfolio, one of the most exciting projects currently underway is “Background Subtraction and Noise Reduction via Machine Learning,” which is led by NNSS Principal Investigator (PI), Cliff Watkins. Cliff’s aim with this project is to explore methods for interpreting data coming into SDRD using ML. Specifically, his goal is to perform high-quality signal extractions by using a neural network to subtract background and noise from data, which would be applicable to radiography, interferometry, hyper-spectral data, and gamma spectroscopy.
Cliff sees more value for ML as a tool to meet certain requirements in the research process, not as a blanket solution for all scientific problems. He describes his approach as building an ML toolkit instead of replacing human workers. From the insights gained through this project, the SDRD program can focus on strategic deployment of quality ML approaches rather than inefficiently applying ML to large quantities of data. Read more about this work in the SDRD Highlight available online.
LANL LDRD: Learning robust parameter inference and density reconstruction in flyer plate impact experiments.
Researchers at LANL are using advanced ML techniques to improve how scientists estimate material properties from radiographic images collected during shock physics experiments. By combining low- and high-velocity impact data with generative AI models, the team developed a new approach for inferring equation-of-state and porosity parameters directly from radiographs, even in the presence of noise or previously unseen physics. The work could significantly enhance the ability to characterize materials and predict their behavior under extreme conditions (read more in LANL’s paper).
Questions? Comments? Contact Us.
ASC Assistant Deputy Administrator: stephen.rinehart [at] nnsa.doe.gov (Dr. Stephen Rinehart)
ASC Deputy Assistant Deputy Administrator: thuc.hoang [at] nnsa.doe.gov (Thuc Hoang)
Program Director for Computing: simon.hammond [at] nnsa.doe.gov (Dr. Si Hammond)
Program Director for Simulation: anthony.lewis [at] nnsa.doe.gov (Anthony Lewis)
- Integrated Codes: anthony.lewis [at] nnsa.doe.gov (Anthony Lewis) (Acting)
- Physics and Engineering Models: robert.spencer [at] nnsa.doe.gov (Robert Spencer)
- Verification and Validation/PSAAP/CSGF: david.etim [at] nnsa.doe.gov (Dr. David Etim)
- Capabilities for Nuclear Intelligence: anthony.lewis [at] nnsa.doe.gov (Anthony Lewis)
- Computational Systems and Software Environment: simon.hammond [at] nnsa.doe.gov (Dr. Si Hammond), sara.campbell [at] nnsa.doe.gov (Sara Campbell)
- Artificial Intelligence for Nuclear Security: cheri.hautala-bateman [at] nnsa.doe.gov (Dr. Cheri Hautala-Bateman)
- Facility Operations and User Support: michael.lang [at] nnsa.doe.gov (K. Mike Lang)
- LDRD/SDRD: anthony.lewis [at] nnsa.doe.gov (Anthony Lewis)
