Researchers at the USC Viterbi School of Engineering have developed an artificial intelligence (AI) model capable of simulating the behaviour of billions of atoms at once, opening new avenues for material discovery and climate solutions.
The model, called Allegro-FM, demonstrated its potential in tackling one of the world’s most pressing challenges: reducing carbon emissions from concrete. Concrete production accounts for about 8% of global CO2 emissions, yet the team’s work showed that it may be possible to turn the material into a carbon sink.
Aiichiro Nakano, Professor of Computer Science, Physics and Astronomy, and Quantitative and Computational Biology at USC Viterbi, was moved to act after the January wildfires in Los Angeles. He reconnected with his long-term collaborator, Ken-Ichi Nomura, Professor of Chemical Engineering and Materials Science Practice, to explore new solutions.
Their discussions led to Allegro-FM, an AI-driven molecular simulation system that made a striking theoretical finding: carbon dioxide released during concrete production can be reabsorbed and embedded back into the same concrete.
Nakano explained: “You can just put the CO2 inside the concrete, and then that makes a carbon-neutral concrete.”
The project brought together Nakano, Nomura, Priya Vashishta, Professor of Chemical Engineering and Materials Science, and Rajiv Kalia, Professor of Physics and Astronomy. Their research centred on ‘CO2 sequestration’, the process of capturing and storing carbon dioxide.
By running simulations at atomic scale, Allegro-FM tested different concrete chemistries virtually before real-world trials. This approach promises faster and cheaper experimentation, with the aim of creating concrete that not only endures but also contributes to carbon reduction.
The model’s scalability distinguishes it from existing methods. While traditional molecular simulations handle only thousands or millions of atoms, Allegro-FM achieved 97.5% efficiency simulating more than four billion atoms on the Aurora supercomputer at Argonne National Laboratory – a system about 1,000 times larger than conventional approaches.
It also extended its reach beyond cement. Covering 89 chemical elements, Allegro-FM predicted molecular behaviour for diverse applications, including carbon storage and cement chemistry.
Nomura said:“Concrete is also a very complex material. It consists of many elements and different phases and interfaces. So, traditionally, we didn’t have a way to simulate phenomena involving concrete material. But now we can use this Allegro-FM to simulate mechanical properties [and] structural properties.”
Concrete’s durability and fire resistance make it a preferred material for cities like Los Angeles, especially in the wake of devastating wildfires. But its heavy carbon footprint remains problematic. Allegro-FM simulations suggest carbon-neutral production is within reach.
The work also hinted at longer lifespans for future concrete structures. Modern concrete typically lasts around 100 years, whereas Roman concrete survived more than 2,000 years. Incorporating CO2 into concrete may strengthen it in a similar way.
Nakano noted: “If you put in the CO2, the so-called ‘carbonate layer,’ it becomes more robust.”
The use of AI proved critical in accelerating this research. Nomura described how the process had shifted over the past two years: “Now, because of this machine-learning AI breakthrough, instead of deriving all these quantum mechanics from scratch, researchers are taking [the] approach of generating a training set and then letting the machine learning model run.”
This reduced the reliance on large-scale supercomputing, allowing the model to achieve quantum-level accuracy with fewer resources.
Nakano added: “[The AI can] achieve quantum mechanical accuracy with much, much smaller computing resources.”
Looking ahead, the team plan to refine its simulations. Nomura confirmed: “We will certainly continue this concrete study research, making more complex geometries and surfaces.”
