Twin sisters from MIT and Harvard employ AI and spectroscopy in a novel method that accelerates the search for energy-efficient topological materials.
In a significant breakthrough spearheaded by sibling scientists from two esteemed universities, the search for energy-efficient materials has been expedited. Nina Andrejević, an MIT doctoral candidate, and her twin sister Jovana, a PhD candidate at Harvard University, have harnessed the capabilities of artificial intelligence (AI) and spectroscopy to streamline the identification of innovative and valuable traits in topological materials. The research has been reported by the World Economic Forum.
Topological materials are an extraordinary phase of matter with intriguing properties. These materials have the ability to transport electrons on the surface without any energy loss. For this reason alone it makes them exceptionally interesting for fabricating more energy-efficient technologies. The recent approach introduced by the Andrejević sisters significantly enhances the pace and versatility of testing for these characteristics.
Raised in Serbia, the two sisters grew up fascinated by the fields of maths, science, and architecture. This academic interest naturally evolved into their respective scholarly paths. Nina’s work in machine learning resonates with architectural endeavors. In architecture, just like in her AI models, one begins with a blank canvas, subsequently optimizing various features to achieve the desired functionality.
Nina, under the mentorship of Mingda Li, an assistant professor at MIT’s Department of Nuclear Science and Engineering, has used AI to seek out new and beneficial attributes in materials. Her research has contributed to leading publications like Nature Communications, Advanced Science, Physical Review Letters, and Nano Letters.
Working in tandem, the Andrejević sisters have innovated a method to anticipate topological characteristics in material samples. Their technique may eventually contribute to the development of better-performing, energy-conserving technologies. Their collaborative research, which utilizes machine learning and specialized spectroscopic techniques for data analysis, can detect patterns in large data sets more efficiently than conventional high-throughput computers.
Their AI environment, equipped with X-ray absorption spectroscopy (XAS), works in conjunction with a trained neural network. This approach sends concentrated X-ray beams into matter to generate a unique signature, mapping the geometry and electron structure of the sampled material. The sisters have effectively created a neural network that can identify topological traits from a material’s XAS signature, expanding the screening potential for a wider category of prospective topological materials.
Over time, the researchers have fed their neural network data from two separate databases: one containing theoretically predicted topological materials and another with X-ray absorption data for a broad spectrum of materials. The well-trained model can subsequently read new XAS signatures and discern if the corresponding material is topological.
Their innovative approach has yielded encouraging results, published in a preprint entitled “Machine learning spectral indicators of topology”. Their method could potentially hasten the progress in the field of materials science.
Nina’s life post MIT will be focused on machine learning. She will be proceeding to Argonne National Laboratory after her graduation, having secured the prestigious Maria Goeppert Mayer Fellowship. Her work will concentrate on designing physics-informed neural networks, with a particular focus on quantum materials.
As the two sisters part ways for their individual careers, they remain hopeful about future collaborations. Their pioneering efforts in creating an efficient AI environment for topological materials research is likely to have far-reaching implications in the development of energy-efficient technologies, a step closer towards a more sustainable future.