The last 30 years have brought phenomenal changes in high-performance computing (HPC). Performance and capability have surged on all fronts: processor acceleration, platform portability, open-source software, artificial intelligence (AI) and machine learning (ML) models, data analysis and visualization, cloud integration, numerical algorithms, programming models and languages, and more. LLNL’s Center for Applied Scientific Computing (CASC) occupies an important role in this fast-moving landscape, connecting research and applied science while linking the Laboratory with collaborators.

Founded in 1996 to coalesce applied mathematics, computer science, and data science, CASC has grown into a dynamic organization of 160 scientists and a dedicated administrative team. Today, its scope includes HPC software tools, large-scale modeling and simulation, AI/ML, research software engineering, and numerical methods and mathematical algorithms.

According to CASC director Kathryn Mohror, “Our researchers strive to apply their work to real-world problems. CASC people are satisfied when they know their work has made a difference at the Laboratory and in the broader scientific community.”

“CASC is a tremendous resource for the Laboratory and the Department of Energy [DOE],” adds Jeffrey Hittinger, who led the Center from 2018 to 2025. “It’s a relatively unique organization with discipline knowledge across many areas within HPC. CASC creates new opportunities and encourages new discoveries.”

Mission Frontiers

The Laboratory’s national security mission evolves with federal priorities and geopolitical conditions, always under the direction of DOE and the National Nuclear Security Administration (NNSA). Across mission-driven programs, CASC’s impact is evident in high-fidelity simulations of complex physical phenomena and in high-dimensional visualization of massive datasets.

This influence extends into applications ranging from steel manufacturing and drug design to power grid monitoring and earthquake modeling. In HPC ecosystems, CASC expertise improves scientific workflows, data management, performance analysis, and resource optimization. Mohror states, “Domain scientists seek out CASC researchers to help overcome roadblocks and reach insights faster and with greater accuracy.”

Throughout decades of technological advancements and mission demands, CASC has stayed true to its name. “What hasn’t changed is the focus on being an applied scientific computing organization that aligns with programmatic needs. Our research consistently shines a light on the problems Livermore cares about,” explains CASC deputy director Tom Epperly.

As Hittinger notes, “From the beginning, CASC has conducted forward-leading research and maintained a strong connection between computing disciplines and mission programs.” These cross-cutting associations often lead to or arise from projects funded by the Laboratory Directed Research and Development program.

CASC also excels in the theoretical research that underpins applied work. Longstanding projects such as HYPRE (linear solvers), SUNDIALS (nonlinear solvers and time integrators), SAMRAI (adaptive mesh refinement), and MFEM (finite element methods) provide advanced mathematical methods that support large-scale scientific computing. Likewise, researchers make significant contributions to AI/ML interpretability and safety through novel frameworks for model steerability, failure detection, trustworthiness evaluation, and adversarial robustness.

Optimization is another area where CASC research spans theory and application in service of national security. For instance, one effort optimizes power grid resources to help prevent attacks and increase critical infrastructure resilience, while another project investigates Bayesian optimization using Gaussian process surrogate models to identify specific parameter values. Additional CASC work in this field focuses on simulating contact mechanics to optimize engineering designs. “Everyone in the Uncertainty Quantification [UQ] and Optimization group straddles the line between research and programmatic impact,” says group leader Kathleen Schmidt.

According to HYPRE lead Rob Falgout, “Being a bridge between fundamental research and simulation science has worked well for CASC. We’ve filled that gap, and this resonates with a lot of people.” Falgout credits CASC’s first director, Steven Ashby (1996–2001), with encouraging staff to join application teams, noting, “This approach ensures an explicit connection that’s really important for integrating foundational software, like HYPRE’s solvers, into simulation codes. We’re there to guide and assist.”

Ahead of the Curve

CASC does not merely react to changing tides. “In addition to solving today’s challenges, our researchers think ahead 5 or 10 years to develop solutions to anticipated challenges. They analyze trends and predict roadblocks the Laboratory might face, then work to make sure we are ready,” explains Mohror.

Many researchers are leaders in their fields, and the Center has consistently been at the forefront of new technologies and applications. For instance, access to a range of HPC systems—including NNSA’s first exascale supercomputer, El Capitan—drives innovation and experimentation with large-scale workflows. Other examples include virtual reality, quantum control, and HPC+cloud convergence.

Moreover, Epperly states, “CASC dove into AI and has been successful at attracting researchers in AI-related technologies.” This expertise helps advance major national initiatives. For example, computer scientist Jae-Seung Yeom leads an HPC workloads project for DOE’s AI-focused Genesis Mission, while several CASC staff represent LLNL on Genesis Mission projects led by other institutions.

CASC also participates in the U.S. Department of War’s Generative Unconstrained Intelligent Drug Engineering program, which leverages simulation, ML, and bioinformatics to accelerate discovery of biothreat countermeasures. The breadth of CASC’s AI/ML research is reflected across nearly all of its 12 groups, including those with deep expertise in Data Science and Analytics, Informatics, and Machine Intelligence. (Read more about groups in the companion article CASC at 30: Individual excellence, collective impact.)

Another emerging area is found in the Graph Science and Irregular Computing group, formed in 2025. Group leader Min Priest explains, “Irregular computing is the use of HPC in nontraditional ways. When people think of HPC at the Laboratory, they think of modeling and simulation for physics codes. My group is working on data science communication patterns that require an entirely different software stack.”

This group also explores the frontier of graph and network processing from trillions to tens or hundreds of trillions of edges and vertices—datasets so huge, Priest says, “they can’t be stored in traditional memory or communicated efficiently.” Managing such extreme-scale data will become more important for science and security applications, underscoring CASC’s foresight in this field.

Ready for the Revolution

CASC’s future may well echo its past and present, as the Center’s agility has always been crucial to its success. Hittinger states, “CASC is resilient and can pivot as new challenges arise. I’m confident it will help shape the next 30 years.” Current projects anticipate this future, including graph analysis supporting Livermore’s global security activities; ML algorithm innovation amid the rise of computer vision and large language models (LLMs); projects addressing efficiency and security of the nation’s power grid; analysis of threats to satellites and spacecraft; and advanced techniques to identify logic errors in HPC applications.

As technologies evolve, CASC’s mathematical research will remain fundamental to scientific and engineering applications. Even as one type of solution comes along, so too do new problems demanding adaptation and efficiency. For example, “HPC applications will need more specialized solvers and greater flexibility in arithmetic precision in order to achieve even better acceleration,” says Falgout, who was among CASC’s first dozen staff.

These mathematical foundations will support CASC researchers as they contend with varied computing configurations. “Computing platforms used to be homogeneous, and now we’re developing portable ways to write large-scale scientific applications for systems with varying compute architectures. We’ll need to keep codes maintainable and performant on both existing and hybrid solutions,” explains Epperly, who joined the Center in 2000.

Priest agrees, “Computing environments are trending toward greater heterogeneity and will include quantum technologies eventually. We’ll see more specialized hardware use cases and have to manage scalability with a smaller energy footprint.” Such scaling solutions will also need to accommodate increasingly massive datasets.

According to Schmidt, many UQ questions persist as AI gains traction and model trustworthiness comes under intense scrutiny. “Quantifying uncertainty for LLMs is a challenge when considering variability. A prompt phrased slightly differently can change a model’s output,” she points out. “Assessing uncertainty and correctness is important to national laboratories studying high-risk, high-stakes problems.”

In a time of transformation, Mohror is optimistic. “For most of a century, scientific computing was on an evolutionary road, but it’s now a revolution,” she says. “I am truly excited to see CASC’s contributions to ensuring the Laboratory is at the forefront of this new future. I know we’ll rise to meet the challenge.”

Next: Individual excellence, collective impact

The image at the top of this page shows deformity and displacement magnitude of two linear elastic bodies in contact. A collaboration of six CASC researchers and three colleagues from LLNL Engineering, the paper Algebraic Multigrid with Filtering: An Efficient Preconditioner for Interior Point Methods in Large-Scale Contact Mechanics Optimization was published in SIAM Journal on Scientific Computing this year.

—Holly Auten