Imagine the concrete in our homes and bridges not only withstanding the ravages of time and natural disasters like the intense heat of wildfires, but actively self-healing or capturing carbon dioxide from the atmosphere.

New possibilities for materials design

Now, researchers at the USC Viterbi School of Engineering have developed a revolutionary AI model that can simulate the behaviour of billions of atoms simultaneously, opening new possibilities for materials design and discovery at unprecedented scales. 

The model also covers 89 chemical elements and can predict molecular behaviour for applications ranging from cement chemistry to carbon storage.

"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," says Nomura.

Concrete is a fire-resistant material, making it an ideal building choice in the wake of the January wildfires. But concrete production is also a huge emitter of carbon dioxide, a particularly concerning environmental problem in a city like Los Angeles. In their simulations, Allegro-FM has been shown to be carbon neutral, making it a better choice than other concrete.

This breakthrough doesn't only solve one problem. Modern concrete only lasts about 100 years on average, whereas ancient Roman concrete has lasted for more than 2,000 years. But the recapture of CO2 can help this as well.

"If you put in the CO2, the so-called 'carbonate layer', it becomes more robust," says Nakano.

In other words, Allegro-FM can simulate a carbon-neutral concrete that could also last much longer than the 100 years concrete typically lasts nowadays. Now it is just a matter of building it.

Behind the scenes

The professors led the development of Allegro-FM with an appreciation for how AI has been an accelerator of their complex work. Normally, to simulate the behaviour of atoms, the professors would need a precise series of mathematical formulas – or, as Nomura called them, "profound, deep quantum mechanics phenomena".

But the last two years have changed the way the two research.

"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," says Nomura. This makes the professors' process much faster as well as more efficient in its technology use. 

Allegro-FM can accurately predict 'interaction functions' between atoms – in other words, how atoms react and interact with each other. Normally, these interaction functions would require lots of individual simulations.

But this new model changes that. Originally, there were different equations for individual elements within the periodic table, with several unique functions for these elements. With the help of AI and machine learning, though, we can now potentially simulate these interaction functions with nearly the entire periodic table at the same time, without the requirement for separate formulas.

"The traditional approach is to simulate a certain set of materials. So, you can simulate, let's say, silica glass, but you cannot simulate [that] with, let's say, a drug molecule," says Nomura.

This new system is also a lot more efficient on the technology side, with AI models making lots of precise calculations that used to be done by a large supercomputer, simplifying tasks and freeing up that supercomputer's resources for more advanced research.

"[The AI can] achieve quantum mechanical accuracy with much, much smaller computing resources," says Nakano.

Nomura and Nakano say their work is far from over.

"We will certainly continue this concrete study research, making more complex geometries and surfaces," says Nomura.

This research was published recently in The Journal of Physical Chemistry Letters and was featured as the journal's cover image.