For greater than 100 years, scientists have used X-ray crystallography to find out the structure of crystalline materials corresponding to metals, rocks and ceramics.
This technique works best when the crystal is undamaged, but in lots of cases scientists only have a powdered version of the fabric that accommodates random fragments of the crystal, making it harder to piece together the general structure.
Chemists at MIT have now developed a brand new generative AI model that makes it much easier to find out the structure of those powder crystals. The predictive model could help researchers characterize materials to be used in batteries, magnets, and lots of other applications.
“Structure is the very first thing it is advisable to learn about any material. It's essential for superconductivity, it's essential for magnets, it's essential for knowing what photovoltaics you're creating. It's essential for each conceivable application that involves materials,” says Danna Freedman, Frederick George Keyes Professor of Chemistry at MIT.
Freedman and Jure Leskovec, professor of computer science at Stanford University, are the lead authors of the brand new study, which appears today within the MIT graduate student Eric Riesel and Yale University graduate student Tsach Mackey are the lead authors of the paper.
Striking patterns
Crystalline materials, which include metals and most other inorganic solids, consist of lattices made up of many similar, repeating units. These units may be regarded as “boxes” of a selected shape and size wherein the atoms are precisely arranged.
When X-rays are directed at these grids, they’re diffracted by atoms at different angles and intensities, revealing the positions of the atoms and the bonds between them. Since the early twentieth century, this method has been used to investigate materials, including biological molecules with a crystalline structure corresponding to DNA and a few proteins.
For materials that exist only as crystal powder, deciphering these structures becomes far more difficult since the fragments don’t exhibit the whole 3D structure of the unique crystal.
“The exact lattice still exists because what we call powder is definitely a set of microcrystals. So they’ve the identical lattice as a big crystal, however the orientation is totally random,” says Freedman.
X-ray diffraction patterns exist for hundreds of those materials, but they’ve not yet been deciphered. To decipher the structures of those materials, Freedman and her colleagues trained a machine learning model using data from a database called the Materials Project, which accommodates greater than 150,000 materials. First, they fed tens of hundreds of those materials into an existing model that may simulate what the X-ray diffraction patterns would seem like. They then used these patterns to coach their AI model, which they call Crystalyze, to predict structures based on the X-ray patterns.
The model breaks the technique of structure prediction into several subtasks. First, it determines the scale and shape of the grid box and which atoms fit into it. Then it predicts the arrangement of the atoms inside the box. For each diffraction pattern, the model generates several possible structures that may be tested by feeding the structures right into a model that determines diffraction patterns for a given structure.
“Our model is a generative AI, which implies it generates something it has never seen before, and that enables us to make several different guesses,” says Riesel. “We could make 100 guesses after which predict what the powder pattern should seem like for our guesses. And if the input then looks exactly just like the output, we all know we were right.”
Solve unknown structures
The researchers tested the model on several thousand simulated diffraction patterns from the Materials Project. They also tested it on greater than 100 experimental diffraction patterns from the RRUFF database, which accommodates powdered X-ray diffraction data for nearly 14,000 natural crystalline minerals that that they had kept out of the training data. On this data, the model was accurate about 67 percent of the time. They then began testing the model on diffraction patterns that had not been solved before. This data got here from the Powder Diffraction File, which accommodates diffraction data for greater than 400,000 dissolved and undissolved materials.
Using their model, the researchers were capable of find structures for greater than 100 of those previously unsolved patterns. They also used their model to find structures for 3 materials that Freedman's lab created by forcing elements that don’t react at atmospheric pressure to form bonds under high pressure. This approach can create recent materials which have radically different crystal structures and physical properties despite having the identical chemical composition.
Examples of such materials are graphite and diamond – each fabricated from pure carbon. The materials developed by Freedman, each containing bismuth and one other element, might be useful in the event of recent materials for everlasting magnets.
“We have found many recent materials in existing data and, most significantly, deciphered three unknown structures from our laboratory that represent the primary recent binary phases of those element mixtures,” says Freedman.
The ability to find out the structures of powdered crystalline materials could help researchers in just about all materials-related fields, in accordance with the MIT team, which has developed an internet interface for the model at www.crystallyze.org.
The research was funded by the U.S. Department of Energy and the National Science Foundation.
