USC AI Model Simulates 4 Billion Atoms to Advance Carbon-Neutral Concrete Design
Image Credit: Matthias Heyde | Splash
Researchers at the University of Southern California's Viterbi School of Engineering have developed an artificial intelligence model capable of simulating more than four billion atoms simultaneously, demonstrating its application in virtual simulations of concrete chemistry that involve carbon dioxide interactions. The breakthrough, published in The Journal of Physical Chemistry Letters, focuses on advancing computational efficiency to address challenges in materials design, including those related to cement production, which contributes about 8% of global carbon dioxide emissions.
Development of the AI Model
The AI model, named Allegro-FM, was created by a team led by Aiichiro Nakano, a professor of computer science, physics and astronomy, and quantitative and computational biology at USC Viterbi. Collaborators include Ken-Ichi Nomura, professor of chemical engineering and materials science practice; Priya Vashishta, professor of chemical engineering and materials science; and Rajiv Kalia, professor of physics and astronomy. The group, which has collaborated for over 20 years, tested the model on the Aurora supercomputer at Argonne National Laboratory in Illinois, achieving 97.5% efficiency in simulations.
Allegro-FM uses machine learning to predict atomic interactions, generating training data that avoids deriving quantum mechanics equations from first principles for each element. This method provides quantum-level accuracy with reduced computational demands compared to traditional approaches, which are generally limited to thousands or millions of atoms. The model encompasses 89 chemical elements, supporting simulations of complex materials involving multiple phases and interfaces.
Background and Motivation
The project stems from efforts to tackle climate change and environmental risks in California, such as wildfires affecting Los Angeles in January 2025. The researchers aimed to explore ways to reduce the carbon footprint of concrete, where production entails heating limestone, thereby releasing substantial CO2. Traditional molecular simulations have been constrained by scale, limiting the speed of testing new sustainable chemistries.
Allegro-FM scales simulations up to 1,000 times larger than previous capabilities, marking progress in AI-driven materials science through decades of computational development. It transitions from high-resource supercomputing to more efficient machine learning frameworks, potentially broadening access to detailed atomic modeling.
Applications in Concrete Design
Through Allegro-FM, the team conducted simulations of concrete formulations incorporating CO2 from cement manufacturing, forming a carbonate layer that researchers suggest could improve mechanical robustness. This approach indicates theoretical potential for making concrete a carbon sink rather than a major emitter, though full carbon neutrality would require system-level validation and life-cycle assessments.
The simulations explore properties that might extend durability, with researchers drawing comparisons to ancient Roman concrete's longevity of over 2,000 years against modern variants often designed for 50-100 years, though such outcomes remain undemonstrated. Virtual testing could decrease reliance on physical experiments, potentially lowering costs and development time for materials in bridges, buildings and pipelines.
Impact on Infrastructure and Environment
The work highlights potential for reducing global CO2 emissions by targeting a significant industrial contributor, while suggesting enhancements to infrastructure resilience in areas vulnerable to extreme weather. In materials science, it supports accelerated discovery via large-scale atomic predictions, with possible future applications in areas like battery technologies, carbon storage systems and biomedical devices.
Nomura highlighted the model's capacity to manage concrete's complexity, including varied elements and structures, overcoming prior simulation constraints. Nakano noted that directly embedding CO2 could theoretically strengthen the material, providing a route toward sustainability.
Future Trends in AI Materials Simulation
The USC team intends to extend simulations to more complex concrete geometries and surfaces, aiming to refine designs for practical use. Wider trends indicate AI may play a growing role in materials research, diminishing dependence on trial-and-error laboratory methods and promoting cross-disciplinary work between computer science and engineering.
As the technology is in its initial phases, its scalability suggests prospects for quicker advancements in eco-friendly materials, which could eventually shape international standards for sustainable construction.
TheDayAfterAI News
We are a leading AI-focused digital news platform, combining AI-generated reporting with human editorial oversight. By aggregating and synthesizing the latest developments in AI — spanning innovation, technology, ethics, policy and business — we deliver timely, accurate and thought-provoking content.