Aidan Duncan ’23

Harvard University, School of Engineering and Applied Sciences, Spaepen Group, Cambridge, MA

This summer I worked with a materials science group at the School of Engineering and Applied Science (SEAS) at Harvard University under Professor Frans Spaepen. I also worked closely with a graduate student and post doc working in Professor Michael Brenner’s applied mathematics group also at SEAS.

Over the past couple months, I worked on identifying structural flow defects in colloidal glasses. A colloid is a system of particles, on the scale of 1-10 microns, suspended in a liquid, and can be used to simulate a system of atoms. While atoms are too small to image conventionally, colloidal particles are big enough to track as they move through the sample. Our colloidal sample is a glass as opposed to a crystal. In a crystal, the atoms are ordered into a regular structure whereas in a glass, there is very little structure. Once we have our colloidal glass sample and have tracked all of the particles, which was done in 2014 by Professor Jensen for her Ph.D. thesis, we can look for flow defects. In a crystal, a defect is easy to find because most particles are in a regular lattice. In a glass, it is not obvious where these defects are located, because there is no regular structure. Both of my projects this summer have revolved around developing a method to systematically find these flow defects, also called shear transformation zones (STZs).

Starting in June, I worked with two of Professor Brenner’s students on using machine learning to try to find these STZs. Machine learning algorithms, on the most basic level, find patterns in data. Our hope was that the algorithm would be able to notice patterns in the structure around certain parts of the glass, and label them as STZs. While I got to learn about many different machine-learning algorithms, including support vector machines, convolutional neural networks, recurrent neural networks, and graph neural networks, our attempts were ultimately fruitless. We decided that we were not giving enough information, or the right information, to the networks. In July and August, I used a more traditional data analysis tool, cross correlation, in combination with ideas from physics and materials science, to attempt to systematically find the STZs. While we have not ultimately succeeded, we have made significant progress, and I will continue to work with Professor Spaepen’s group this semester.

I had so much fun this summer, even if I was only able to meet with others through Zoom. Working with Professor Spaepen’s and Professor Brenner’s students, I gained a better appreciation for what it might be like to be a graduate student in physics. It has been my dream to become a physics professor and obtain a Ph.D. after I graduate and this summer has given me valuable experience in this regard.

I would like to extend my thanks to the Kraft Family and to the ’68 Center for Career Exploration for this opportunity.