about arthur

Dr. Arthur McCray is a postdoctoral researcher in the materials science department at Stanford University. He studied physics at Carleton College and received his PhD in Applied Physics from Northwestern University in 2023. Arthur grew up near Seattle and attended Lakeside for middle and high school.  He loved sports and reading--especially science fiction--as a kid. He knew he wanted to become a scientist when he read, “A Brief History of Time” by Stephen Hawking because it took the scale and epic grandeur of science fiction concepts and turned them into something he could study. He decided to become a physicist in high school because physics helps us to answer the “why” questions we ask about the world, e.g. why is the sky blue? In his free time, he loves getting outdoors. Biking, hiking, and climbing are his main hobbies and he also loves to cook!

What does Arthur study?

Arthur works in a lab that focuses on developing methods for very high resolution imaging. They use transmission electron microscopes (TEMs, like the one further down the page) that use electrons to image a sample rather than photons (which are light particles, and what typical microscopes use). This lets a TEM capture images with enough resolution to see individual atoms!

One topic that is particularly exciting to Arthur is how we can use computational tools to solve inverse problems that we encounter when using a TEM (transmission electron microscope). One example is that of reconstructing a 3D structure from 2D projected images. This can’t be done from a single image (the problem is “under-constrained”, meaning there are many solutions that could work), but it can be done by recording lots of images with the sample at different orientations. These images can be used to reconstruct the 3D sample in a process known as tomography. 

In all cases, inverse problems occur because the images that we collect don’t readily contain the most important information about a sample. This occurs frequently when using a TEM, and it can cause problems both when imaging the atomic structure of a sample and when imaging other things like magnetic fields. 

Much of Arthur’s work focuses on developing computation tools, i.e. writing computer programs, that help us solve inverse problems (determining 3D structure from 2D information) and better understand images that are recorded with a TEM. Machine learning is a particularly important tool in his work, as it has greatly expanded the types of tasks that computers can perform. With machine learning, we can train artificial neural networks, which are a type of computer program loosely based on the cell networks that make up our brains. Neural networks can be trained to do lots of useful and otherwise tedious tasks, for example they can be trained to identify specific features in an image more accurately and much, much faster than can be done by hand. 

We can also use machine learning to help us solve inverse problems directly. If we can develop a mathematical forward model for a particular problem (eg recognizing the 2D projection of a 3D object), then we can apply a machine learning technique called “backpropagation” to help us directly solve the problem, meaning the program could be trained to determine 3D structure from 2D images. This is not a magic bullet for solving every hard problem, but we have shown it to be a very powerful technique that can be applied to many situations. In many cases we can combine this technique with neural networks and other physics-based computational techniques, and this provides a powerful basis for much of Arthur’s work.