Firms working on the frontier of aerospace, energy, and computing are continuously on the lookout for recent materials to enhance performance. But with a view to understand how those materials will actually behave once they’re inside rockets or on computer chips, corporations first need to make the fabric after which test it. That’s because even probably the most powerful simulation techniques struggle to model the complex chemical arrangements in most of today’s solid materials. The issue adds costs and time to materials innovation.
Now a team of MIT researchers has created a method to accurately model the behavior of metals, whatever the complexity of their chemical arrangement. At the middle of the approach are machine-learning models that make simulations of materials faster and more accurate. The researchers improved those models by constructing training datasets that capture the variety of atomic environments in chemically disordered materials.
In a recent paper in Sciences Advances, the researchers showed their approach might be used to accurately predict material properties for a various group of metal alloys under a spread of conditions. Additionally they showed how the approach might be used to develop recent materials, especially in scenarios where experimentation is pricey.
“The main target of the paper is metallic alloys, which is the sphere I work in, but this might be adapted to other kinds of materials, like semiconductors,” says senior creator Rodrigo Freitas, MIT’s TDK Profession Development Professor in Materials Science and Engineering. “This will not be specific to anyone application — you would use this approach to create recent sustainable steels, recent materials for aerospace, and more. That’s what makes this exciting.”
Joining Freitas on the paper are first creator Killian Sheriff PhD ’26; MIT PhD students Daniel Xiao and Yifan Cao; and University of Sheffield Senior Lecturer Lewis R. Owen.
Modeling metals
Material properties are mostly determined by the interior arrangement of their chemical elements. Even when two materials have the identical mixture of chemical elements, different chemical arrangements could make the difference between a brittle material and one which deforms without breaking.
Capturing that distinction requires simulating materials atom by atom. To try this, researchers depend on models that describe how atoms interact with one another. During the last twenty years, machine learning has turn into probably the most accurate method to construct those models. Such models work well when the chemical arrangements inside materials follow highly ordered patterns, but that’s not the case with most solid materials, whose atomic chemical arrangements are disordered and vary from one region to a different.
“The actual challenge in our field is modelling these chemically disordered phases,” Freitas says. “Chemical disorder means there’s an enormous number of local chemical environments, which is difficult for the machine-learning model to learn. This can be a problem because each metal we use in practice is chemically disordered.”
The issue comes right down to an absence of representative training data for those atom-by-atom simulations. The present leading approach for creating such data works by brute force, often requiring greater than 100,000 hours of computation to create the training data for a single material. Even then, it doesn’t transfer well when researchers change the fabric’s composition.
In previous work, Freitas’ group had developed a method to measure the chemical complexity of solid materials by analyzing the frequency and spacing of tiny groups of atoms. For this study, the researchers used that capability to construct higher training datasets. They used a mathematical approach generally known as information theory to generate training datasets that capture a greater variety of local chemical environments inside disordered materials. The strategy works by swapping out atoms from samples to scale back repetition and expose the model to chemical environments it would otherwise miss.
“We kept optimizing the training set so it captured as many alternative local environments as possible,” Freitas says. “If the identical sort of environment showed up repeatedly, we replaced redundant examples with ones the model hadn’t seen before. That makes the training set rather more informative because each example adds something recent.”
When trained on the researchers’ datasets, the models predicted material properties more accurately than models trained using random sampling or one other popular sampling method.
“The start line for all these atom-by-atom simulations is: Are you capable of accurately describe the chemical bond between atoms?” Freitas explains. “If not, it may possibly still teach you about materials generally, but it surely doesn’t let you know what’s going to occur to specific materials in the true world. This approach makes the simulations high fidelity when it comes to their chemistry, to raised reflect what’s happening to materials.”
The researchers applied their technique to create machine-learning training datasets for a gaggle of chemically diverse metal alloys. Using a set of machine-learning models, they showed the models trained on their datasets are more accurate than much larger models created by corporations like Google and Microsoft.
“We got to some extent where we were convinced it worked without using these expensive brute-force methods,” Freitas says. “I told Killian, ‘That is a very good paper. But for those who can show that simulations with these models can now accurately predict useful materials properties, then it becomes a excellent paper.’ Killian took that to heart and tested this as widely as he could.”
Sheriff worked with Xiao and Cao to check the approach across different alloys and properties. The team also drew on Owen’s experimental data to match the simulations against real measurements of atomic ordering in alloys.
From the lab to industry
The strategy works, partly, by capturing hidden patterns within the sample data. The researchers describe the patterns within the paper as “subtle energetic biases toward certain local chemical configurations.”
Those small energetic differences matter because they determine which phases form in an alloy, how those phases change with temperature and composition, and ultimately which properties the fabric may have. As one test, Daniel Xiao led simulations showing that the team’s models could predict phase diagrams that closely matched experimental data. Phase diagrams map which phases are stable across different temperatures and chemical compositions, and so they are a central tool for designing and processing alloys.
“Phase diagrams are one among the primary ways people connect materials modeling to real processing decisions,” Freitas says. “If you happen to are welding, casting, or heat-treating an alloy, it’s good to know which phases are more likely to form under different conditions. Our goal is to make these sorts of predictions accurate enough, and accessible enough, that they turn into a part of how people design materials.”
The researchers are actually using the approach to review how changing an alloy’s composition affects mechanical properties and radiation tolerance, with the goal of designing materials that remain strong and damage-tolerant in harsh environments. Also they are working to make the tactic easier to make use of with the sorts of tools and workflows materials engineers already depend on.
“Industry isn’t going to alter the best way they do things if what you’re creating doesn’t fit into their existing operating procedures,” Freitas says. “The goal is to make these predictions useful within the places where materials decisions are literally made.”
The research was supported by the U.S. Air Force Office of Scientific Research.