Skip to main content

News

Berkeley Lab AI Learns Classic Physics to Scale Quantum Simulations

A multi-part technical diagram showing the architecture and function of the Sigma-Attention AI model. The left side shows a vertical flowchart titled "Encoder-only Transformer" with interconnected boxes labeled with AI processes like Multi-Head Attention and Feed Forward, leading to a final output labeled with a Sigma symbol. The top middle shows a chart titled "Dataset regime" that plots interaction strength against system size, shading three areas labeled MBPT for weak coupling, ED for intermediate coupling, and SCE for strong coupling. The top right features a line graph titled "Green's function prediction" displaying multiple U-shaped curves. Along the bottom, text reading "Generalizable self-energy learning across regime" sits next to a small, simple dot-and-line grid labeled "Small system." An arrow containing the AI model's Sigma symbol points from this small grid to a significantly larger grid labeled "Large system."

Excited-States Framework Could Illuminate New Optical Materials

Zenesis: A ‘ChatGPT’ for Scientific Image Analysis

A split graphic illustrating the Zenesis AI tool in action. The left panel shows a raw grayscale scientific image of a textured material with a prominent, jagged dark crack running through it. The middle panel displays a chat interface where a user message asks, "what is the length of the largest crack?" An AI chatbot responds, stating it found the requested structure, highlighted it in green, and provides an estimated length of 904.15 pixels. The right panel displays the original scientific image, but the main crack is now precisely outlined in bright green, matching the AI's response.

A Berkeley Lab Mentorship Comes Full Circle

Three people stand smiling for a photo outside the entrance of a modern, grey-paneled building labeled "Shyh Wang Hall". From left to right: a man wearing sunglasses and a pink polo shirt, a woman wearing sunglasses and a bright yellow patterned top, and a young man wearing a black denim jacket over a black shirt with blue jeans. They are standing on a wide concrete walkway on a bright, sunny day.

Pele Combustion Simulation Suite Named R&D 100 Award Winner

A high-resolution 3D computer simulation of a turbulent flame expanding from left to right against a light grey background. The simulation shows a textured, silver-grey plume of fuel igniting into a bright core of blue, red, orange, and yellow colors representing intense heat and complex chemical reactions. In the top left corner is a black and gold badge that reads R&D 100 WINNER.

Software Highlight: AstraAI Masters the Complex Rules of Scientific HPC Software

A flowchart outlining the system architecture for AstraAI. The process begins on the left with a natural language user prompt, which is passed to an LLM for user intent extraction utilizing run-time tools like Hugging Face, Llama.cpp, or Ollama. The workflow then splits into two parallel processes: Retrieval Augmented Generation to extract relevant code chunks, and Abstract Syntax Tree code structure extraction utilizing Clang tools like intercept_build and clang-query. The outputs from these two processes are merged into a final prompt that includes the original user prompt, the RAG chunks, and the structure information. This final prompt is sent to an LLM for code generation, which ultimately leads to AST-safe code editing featuring GitHub conflict-style markers.

Scientists Develop Predictive Roadmap to Boost Performance in Next-Gen Spintronics

Scanning electron microscope (SEM) image of a chiral 2D metal halide perovskite material synthesized with acetonitrile. Chiral perovskite films made with acetonitrile respond strongly to circularly polarized light.

New Hybrid Approach Bypasses Hardware Limits to Unlock Complex Quantum Simulations

A schematic diagram divided into two main sections illustrating a computational workflow, with Quantum measurements on the left and Classical postprocessing on the right. The left section displays a series of quantum circuit diagrams at increasing time steps, featuring green blocks labeled Hamiltonian simulation and blue blocks labeled Shallow shadows, which output histograms and color-coded data vectors. The right section shows the flow of classical data processing, starting at the top with multi-layered, colorful grids representing data matrices being used to compute a central system matrix. An arrow points down from this matrix to a diagram labeled Spectral decomposition, which shows data points plotted around the perimeter of a circle to represent eigenvalues. A final arrow points left to a line graph labeled Energy error, illustrating the error rates for both an excited state and a ground state steadily decreasing as the number of data points increases.

Mathematical Innovation Advances Complex Simulations for Science’s Toughest Problems

A two-part diagram illustrating a mathematical framework for simplifying complex models. The top diagram shows a V-shaped path of blue arrows enclosing a horizontal line that extends from a point labeled '0' towards an infinity symbol. Small circles along the line represent key numbers describing a system's dynamics. The blue path illustrates a calculation where orange and green segments at the far right can be ignored, allowing scientists to focus on the system's essential features for faster results. The bottom diagram shows a horizontal line with colored tick marks representing a random process visiting different states over time. Above the line, red brackets group larger sets of states into repeating loops, or cycles. Below the line, blue brackets highlight more specific cycles. This illustrates how analyzing these cycles can simplify and speed up complex simulations.

Berkeley Lab Researchers Evaluate Generative AI Models for Filling Scientific Imaging Gaps

A comparative infographic showcasing the results of different generative AI models in creating scientific images. The infographic is divided into three vertical sections, labeled 'Ceramics,' 'Plants,' and 'Rocks.' Each section contains a grid of images comparing a 'Raw' scientific source image to AI-generated versions created by models including DCGAN, StyleGAN, DALL-E2, and DALL-E3. The 'Ceramics' section shows microscopic views of packed particles, the 'Plants' section displays images of seedlings and root systems, and the 'Rocks' section features grayscale micro-scans of rock sediment, illustrating the varying success of different AI models in reproducing scientifically accurate images.

Unprecedented Perlmutter Simulation Details Quantum Chip

Computing Sciences logo

Connect

Our Organization