Accelerating the search for energy-efficient materials through machine learning


PhD candidate Nina Andrejevic combines spectroscopy and machine learning techniques to identify new and valuable properties in matter.

Born into a family of architects, Nina Andrejevic enjoyed creating drawings of her house and other buildings as a child in Serbia. She and her twin sister shared this passion, as well as an appetite for math and science. Over time, those interests have converged into a scientific path that shares some attributes with the family profession, according to Andrejevic, a doctoral candidate in materials science and engineering at MIT.

“Architecture is both a creative and a technical field, where you try to optimize the desired characteristics for certain types of functionality, like the size of a building or the layout of different rooms in a house,” says- she. Andrejevic’s work in machine learning resembles that of architects, she believes: “We start with an empty site – a mathematical model that has random parameters – and our goal is to train this model, called neural network, so that it has the functionality that we desire.”

Andrejevic is a doctoral student with Mingda Li, Assistant Professor in the Department of Nuclear Science and Engineering. As a research assistant in Li’s quantum measurement group, she trains her machine learning models to search for useful new traits in materials. His work with the lab has landed in major journals such as Nature Communication, Advanced sciences, physical examination letters, and Nano-letters.

Nina and Jovana Andrejevic

MIT doctoral candidate Nina Andrejević (right) developed with her twin sister Jovana (left), a doctoral student at Harvard University, a method for testing material samples to predict the presence of topological features that is faster and more versatile than other methods. Credits: Gretchen Ertl

A particular area of ​​interest for his group is that of topological materials. “These materials are an exotic phase of matter that can carry electrons to the surface without loss of energy,” she says. “That makes them very interesting for creating more energy-efficient technologies.”

Together with his sister Jovana, a PhD student in applied physics at Harvard University, Andrejevic developed a method of testing material samples to predict the presence of topological features that is faster and more versatile than other methods.

If the ultimate goal is to “produce more efficient and energy-efficient technologies,” she says, “we first need to know which materials are good candidates for these applications, and that’s something our research can help confirm.

Teaming up

The seeds of this research were sown over a year ago. “My sister and I always said it would be cool to do a project together, and when Mingda suggested this study of topological materials, it occurred to me that we could make it a formal collaboration,” says Andrejevic . The sisters are more alike than most twins, she notes, sharing many academic interests. “Being a twin is a big part of my life and we work well together, helping each other in areas we don’t understand.”

Andrejevic’s thesis work, which encompasses several projects, uses specialized spectroscopic techniques and data analysis, enhanced by machine learning, which can find patterns in large amounts of data more efficiently than even computers. more efficient.

Nina Andrejevic

When she graduates this winter, Nina Andrejević will travel to Argonne National Laboratory, where she plans to focus on designing physics-informed neural networks. Credits: Gretchen Ertl

“The common thread running through all my projects is this idea of ​​trying to speed up or improve our understanding when applying these characterization tools, and thus get more useful information than we can with more traditional or approximate models,” she says. An example of this is the twins’ research into topological materials.

In order to unravel new and potentially useful properties of materials, researchers must interrogate them at the atomic and quantum scale. Neutron and photon spectroscopy techniques can help capture previously unidentified structures and dynamics and determine how heat, electric or magnetic fields, and mechanical stress affect materials at the Lilliputian level. The laws governing this domain, where materials do not behave as they could at the macro scale, are those of quantum mechanics.

Current experimental approaches to identify topological materials are technically difficult and inaccurate, potentially excluding viable candidates. The sisters believed they could avoid these pitfalls by using a widely applied imaging technique called X-ray absorption spectroscopy (XAS) combined with a trained neural network. XAS sends beams of focused X-rays into matter to help map its geometry and electronic structure. The radiation data it provides provides a unique signature to the sampled material.

“We wanted to develop a neural network that could identify topology from a material’s XAS signature, a much more accessible metric than other approaches,” says Andrejevic. “This would hopefully allow us to filter out a much broader category of potential topological materials.”

Over the months, the researchers fed their neural network information from two databases: one contained materials theoretically predicted to be topological, and the other contained X-ray absorption data for a wide range of materials. “When properly trained, the model should serve as a tool to read new XAS signatures that it has never seen before and indicate whether you know whether the material that produced the spectrum is topological,” says Andrejevic.

The research duo’s technique has demonstrated promising results, which they have already published in a preprint, “Machine Learning Topology Spectral Indicators.” “For me, the fun of these machine learning projects is seeing underlying patterns and being able to understand them in terms of physical quantities,” says Andrejevic.

Towards materials studies

It was during his freshman year at Cornell University that Andrejevic first experienced the pleasure of scrutinizing matter on an intimate level. After a course in nanosciences and nanoengineering, she joined a research group in materials imaging at the atomic scale. “I feel like I’m a very visual person, and this idea of ​​being able to see things that until then were just equations or concepts, that was really exciting,” she says. “This experience brought me closer to the field of materials science.”

Machine learning, central to Andrejevic’s doctoral work, will be central to his life after MIT. When she graduates this winter, she’s headed straight for Argonne National Laboratory, where she’s won a prestigious Maria Goeppert Mayer Fellowship, awarded “internationally to outstanding doctoral scientists and engineers who are at the beginning promising careers. “We will try to design physics-informed neural networks, with a focus on quantum materials,” she says.

It will mean saying goodbye to her sister, from whom she has never been separated for a long time. “It will be very different,” says Andrejevic. But, she adds, “I hope Jovana and I will collaborate more in the future, no matter how far apart!”

Sherry J. Basler