Jack Wiskemann ’29 spent his summer at Kenyon, developing programs in a cozy room in Hayes Hall labeled “LIGO Lab” bedecked with humming computers and chalk-covered blackboards bearing notes and equations.
A Summer Science Scholar, he’s part of the College’s LIGO (Laser Interferometer Gravitational Wave Observatory) Lab that is working on a variety of research projects using the pair of U.S.-based facilities in Washington state and Louisiana. These instruments were the first to directly detect evidence of gravitational waves in 2015 from two colliding black holes, strengthening evidence for Einstein’s theory of general relativity.
All objects with mass distort spacetime, creating the effect we call gravity. When massive bodies like black holes and neutron stars orbit and collide, their distortions of spacetime interact, creating “ripples” that are called gravitational waves. These propagate outwards like ripples on a pond, distorting space imperceptibly, such that only incredibly precise instruments like LIGO can detect their presence.
At Kenyon, Wiskemann is working with Associate Professor of Physics Madeline Wade on a project that is using machine learning to try to improve our ability to understand the raw data generated by LIGO. Instrumental factors like thermalization, which happens when the startup of the lasers heats the lenses enough to slightly change their properties, as well as the environment around the facility, can create complexities in the data that must be accounted for before we can understand what the data is telling us.
To help, Wiskemann is building Physics-Informed Neural Networks (PINNs) to interpret the data from LIGO’s four-kilometer lasers. These networks are programs designed to learn how to do something while incorporating an understanding of the fundamental laws of physics.
“There is the detector readout that we get, that we want to convert back into what was actually, physically happening,” Wiskemann says.
In order to go from the raw data from LIGO’s lasers to information about the gravitational waves themselves, physicists must create a mathematical model of the detector called a “response function.” Researchers have a good grasp of the value of the response function for certain frequencies of gravitational waves, but not all. Wiskemann is building specialized neural networks to predict the values of the response function for all frequencies, based on the frequencies for which the value is already known.
This work is important not just in helping to interpret data now but also in keeping models updated as equipment becomes increasingly sensitive. “LIGO’s always working to upgrade the instruments,” Wiskemann explains, and with greater sensitivity comes more sources of data noise, like thermalization. “We don’t want that to be something that holds us back.”
In addition, Wiskemann is interested in the potential of PINNs to not only help based on the current model of physics, but to also advance our understanding of the physics that it is based on.
“The physics we know gives us a starting point for the neural networks, and then hopefully the neural networks will give us a starting point for the rest of the physics,” he said.
The most unexpected thing about Wiskemann’s work this summer, though, had nothing to do with neural networks or the physics of gravitational waves, but rather the amount of freedom and agency he has over his work at Kenyon.
“I think doing research — especially research at Kenyon, where you really get your own project — it really becomes your own,” he says. “It’s awesome.”
This article was written by Cole Szostak ’27 as part of the Hoskins Frame Summer Science Writing Scholars program.