Berkeley Lab Computing Sciences

Ready, set, SLAM! 🕹️ Support our Computing Sciences postdocs as they join a fantastic lineup of 12 early-career scientists taking the stage at the 2026 Berkeley Lab Research SLAM.
Anupam Mitra will discuss "Tuning into a Quantum Magnet," and Durga Keerthi Mandarapu will explore "Are Your Fluffy Dogs as Unpredictable as my Computer Programs?"
📅 Join us on Wednesday, September 23, from 3:00 to 4:30 p.m. in the Building 50 Auditorium.
🏆 If you attend in person, you can cast your vote for the People’s Choice Award!
💻 Can't make it to the auditorium? You can still support our researchers virtually by tuning in at streaming.lbl.gov.
🎉 Keep the energy going at the reception immediately following the talks on the 3rd floor of B91 IGB.
It is the perfect opportunity to honor all of our SLAM participants and celebrate National Postdoc Appreciation Week together.

⚛️ Predicting how electrons behave in quantum materials is one of the greatest computational challenges in modern physics. To solve this, Berkeley Lab researchers have developed Σ-Attention (Sigma-Attention), a highly scalable new AI framework.
🧠 By adapting the same transformer architecture that powers modern LLMs, the team trained their AI on a combination of three traditional quantum simulation techniques. Instead of predicting the next word in a sentence, Σ-Attention learns to predict “self-energy,” a mathematical function capturing how strongly correlated electrons interact and influence one another.
⚡ Traditional computational methods force researchers to trade off between accuracy and system size. Σ-Attention breaks this barrier by accurately predicting collective electron behavior across various conditions while drastically reducing the required computing power. This breakthrough allows scientists to simulate much larger, more realistic quantum materials, paving the way for next-generation technologies like lossless power transmission, advanced quantum sensors, and ultra-efficient computing.
Learn more in comments ⬇️
(📸: Hardware of a current quantum computer, the exact kind of technology Σ-Attention will help advance.)

______________________________________

Did you know that Berkeley Lab is seeking early-career scientists for the prestigious Luis W. Alvarez and Admiral Grace M. Hopper Postdoctoral Fellowships?
As a fellow, you will push the boundaries of extreme-scale computing, pioneer next-generation AI, and design mathematical models to tackle global challenges in fields ranging from materials science to quantum computing.
You will have unparalleled access to world-class facilities—including NERSC supercomputers and the Energy Sciences Network (ESnet) high-speed network—empowering you to perform independent research that shapes the future of technology. If you are earning a Ph.D. in computer science, mathematics, data science, or a related computational discipline by October 1, 2027, a single application puts you in the running for both of these highly competitive roles.
Read the full fellowship details in the comments ⬇️, and submit your application by October 30, 2026.

cc: U.S. Department of Energy

New optical materials hold the key to the next generation of quantum computers, sensors, and solar panels, but understanding their behavior at the microscopic level requires massive computational power. 🔬💻
In a new paper, researchers from Berkeley Lab and the University of Southern California have successfully simulated complex optical materials at an unprecedented scale of 10,000 atoms to uncover exotic quantum phenomena from first principles. By using the BerkeleyGW software package—optimized with support from the NERSC Science Acceleration Program (NESAP)—the team created a powerful new computational framework to study phenomena such as Wigner-crystalline excitons. ⚛️
Berkeley Lab's Mauro Del Ben notes that this accomplishment allows researchers to explain these exotic quantum phenomena and pushes predictive computational capabilities for quantum technologies even further.
As the team looks ahead, they are already planning to leverage AI, mixed-precision algorithms, and NERSC’s upcoming flagship supercomputer, Doudna, to continue expanding this critical research.
Learn more at the links in the comments. ⬇️

___________________________________

Mainstream AI models struggle with complex scientific images, forcing researchers to spend countless hours manually annotating data. Enter Zenesis, a “ChatGPT-like” AI platform for scientific image analysis.
Zenesis operates directly on raw data with over 98% accuracy without needing any prior model training or hand-labeled data, turning tedious manual annotation into a process that takes just minutes.
Developed by Berkeley Lab researchers Shubhabrata Mukherjee and Daniela Ushizima, this interactive, no-code platform allows domain scientists to isolate intricate features—from compact crystals and pores to branching cracks—using simple, natural-language text prompts.
Learn more about Zenesis in the link in comments ⬇️
_________________________________
cc: U.S. Department of Energy

Did you know Berkeley Lab is seeking a visionary leader to serve as the next Director of NERSC? This is a rare opportunity to steer the mission HPC facility for the U.S. Department of Energy's Office of Science and directly accelerate the pace of global scientific discovery.
We are searching for an innovator with a distinguished background in HPC, AI, and scientific workflows to guide NERSC into its next era. As Director, you will be leading a world-class team to tackle the most complex challenges in exascale and post-exascale computing, quantum technologies, and AI for science. If you are passionate about cutting-edge supercomputing and fostering a collaborative culture of scientific excellence, this is your chance to make a tangible impact on the future of computational research.

🔗 Apply here: https://lbl.taleo.net/careersection/2/jobdetail.ftl?lang=en&job=107114

_______________________________________

Seven years after her own transformative internship, Meharry Medical College Assistant Professor Naw Safrin Sattar returned to Berkeley Lab this summer with her graduate students: David Reyes and Evan Browne. Together, they tackled an AI-driven materials science project alongside Khaled Ibrahim, Sattar’s former summer mentor and longtime collaborator.
“Because Dr. Sattar has been through it herself, she recognizes the challenges we face and has the patience to guide us through them. This summer has exceeded my expectations. I’ve had the chance to grow and learn so much from the people here at Berkeley Lab. My mentors have given me a lot of room to explore and contribute my own ideas, and they’ve always provided thoughtful feedback and direction,” said David Reyes.
Read more ⬇️ at the link in comments.

______________________________
Berkeley Lab A-Lift

Last week, current and former DOE CSGF fellows gathered at the home of Berkeley Lab Computing Sciences Senior Advisor David Brown to celebrate and welcome our latest cohort of fellows. This year, Berkeley Lab is hosting 18 CSGF fellows.
If you are passionate about advancing scientific discovery through HPC, we would love to see you in our next cohort. Learn more and apply to the CSGF program at the link below ⬇️

cc: Krell Institute

This week marks the end of the 2026 Berkeley Lab Computing Sciences Area summer student program! 📸 Swipe through to see highlights from our concluding Poster Session on August 11th, where 83 community college, undergraduate, and graduate students presented their projects to staff from across the area.
🧠 Over the past ten weeks, our incredible cohort of 145 students, affiliates, and guest faculty—including participants from the Berkeley Lab A-Lift—gained hands-on experience with cutting-edge tools across a variety of departments. Beyond their research, they honed vital professional skills like resume building and science communication, and even toured world-class facilities like NERSC, the Advanced Light Source, and the The Molecular Foundry.
Want to know what it’s really like to spend a summer at Berkeley Lab? Read our summer student testimonials at the link in the comments ⬇️

cc: Energy Sciences Network (ESnet)

🧠 Artificial intelligence holds immense potential for scientific discovery, but AI systems often act as opaque "black boxes" that struggle to accurately simulate the real world because they ignore fundamental physical laws. Now, an open-source framework from Berkeley Lab is changing this by guiding generative AI with physical constraints learned directly from real-world data—a major leap forward for scientific machine learning that could soon tackle computational bottlenecks in weather and fluid dynamics modeling, and medical image reconstruction.

🏙️ As a proof of concept, the researchers applied this approach to one of the most complex physical challenges: earthquake modeling. Predicting exact seismic impacts traditionally requires hours of expensive supercomputing time. But their new tool, Conditional Generative Modeling for Ground Motion (CGM-GM), can simulate thousands of highly realistic earthquake scenarios across the San Francisco Bay Area in just minutes on a standard GPU-enabled computer.

💻 Best of all, this incredible speed does not come at the cost of accuracy or real-world utility. The model generates broadband motions up to 15 Hertz, providing the crucial high-frequency resolution engineers need to determine exactly how different structures like tall buildings and underground pipelines will withstand a quake.

Learn more at the link in comments. ⬇️

___________________________________

cc: Earth & Environmental Sciences Area