Back to top
Artificial intelligence (AI) is fast becoming one of the most important tools for analysis of scientific data. Uniting AI with high-performance computing simulations yields entirely new capabilities that amplify researchers’ abilities to solve complex national security and science problems. This integration—called cognitive simulation (CogSim)—is the subject of this issue’s feature story.
The article, Cognitive Simulation Supercharges Scientific Research, describes how Lawrence Livermore uses machine learning—one of AI’s most important components—to analyze scientific data across a wide range of applications. Combined with high-performance computing simulations long central to Livermore’s scientific applications, researchers observe outcomes arguably greater than the sum of their theoretical and experimental parts.
CogSim can amplify computing capabilities by speeding up simulations. Replacing a time-consuming physics calculation with a fast machine-learning model can, in some cases, greatly accelerate the code while maintaining the quality of the simulation. As a result, CogSim improves the predictive power of models and guides optimization methods for unique designs and superior performance in multiple Laboratory applications.
CogSim strengthens the Laboratory’s approach to fundamental research and development as well. Advances in scientific discovery occur only when researchers challenge their own theories and previous understanding with experimental results. For complex questions, theory-based predictions require computational simulations with many variables and complex conditions. Comparing simulation predictions to experimental outcomes can validate understanding or, alternatively, provide clues to incorrect or missing parts of scientific theories. Until now, this comparison has taken place in a scientist’s mind, limiting the complexity of the evaluation, as multidimensional simulations and experimental results are often reduced to single quantities for comparison. As a result, information in data products such as time series, spectra, or images can be lost. Machine learning offers approaches to systematically unite simulation and experiment in a computational framework that enables scientists to retain more of the complex data for comparison.
Today at Lawrence Livermore, CogSim is used across many applications from U.S. nuclear stockpile stewardship to designing inertial confinement fusion experiments to developing new SARS-CoV-2 therapeutics—several of which are detailed in the feature article. Looking ahead, CogSim promises faster and higher quality answers to complex analysis and design problems that inform improvements in national security, health care, and the economy.
The research highlights in this issue touch on other Laboratory programs and research aimed at maintaining nuclear safety, preserving valuable scientific data, and improving experimental capabilities. The first story, Operational Resilience during the COVID-19 Pandemic, describes efforts by Livermore’s Global Material Security Program to continue critical work securing nuclear and other radioactive materials despite travel restrictions during the COVID-19 pandemic. Once solely a boots-on-the-ground program to lead nuclear forensics workshops, train international partners in developing their own capabilities, and network internationally with forensics experts, the program pivoted to establish virtual inspection, learning, and conference platforms plus new approaches to joint sample analysis and facility upgrade planning without in-person interactions. The second highlight, High Performance Storage System — Taking the Long View, presents a nearly 30-year collaboration across multiple national laboratories to ensure data storage and retrieval systems match the growing data output of supercomputers. The High Performance Storage System, designed to serve the world’s fastest supercomputers as well as protect sensitive data, has met its mission by evolving over the decades, reliably storing data long term for government, academic, and commercial organizations around the globe. The final article, Cinema in a Nanosecond , details a Livermore diagnostics system designed to capture images of instantaneous experimental effects. Improving on earlier technologies to capture images of material behavior across nanosecond timescales, the Bipolar Reset Experiment—called BiRX—acquires far more images from a single experiment than predecessors, eliminating the need for multiple experiments, reducing cost and potential variability.
Each highlighted achievement reflects how the Laboratory and its people grow and evolve ideas, just as CogSim has grown to advance scientific research and national security. Combined, these stories provide a snapshot of Lawrence Livermore’s ongoing commitment to scientific excellence and discovery.