Gravitational-Wave Physics @ Kenyon

We are members of the LIGO Scientific Collaboration and run a LIGO group at Kenyon.  LIGO stands for Laser Interferometer Gravitational-wave Observatory.  There are two LIGO observatories in the United States.  These instruments were the first to directly detect the existence of gravitational waves on September 14, 2015!  This momentous discovery of the gravitational waves from two colliding black holes received a wide array of press and attention, including the co-founders of the LIGO experiment being awarded the 2017 Nobel Prize in Physics.  Since 2015, LIGO has discovered over 100 more binary black hole mergers through their gravitational wave signatures, several binary neutron star mergers, and several neutron star - black hole mergers.  Most notably, on August 17, 2017, LIGO detected gravitational waves from the collision of two neutron stars.  This event also emitted radiation across the electromagnetic spectrum and was observed by hundreds of telescopes around the world.  This was the first multi-messenger astronomy event that involved gravitational waves, and for reference of the importance of this discovery and the science it has enabled, the detection paper for this event has been cited over 6,600 times since its publication.  You can read more about LIGO and the exciting discoveries we have been involved in as members of the collaboration at www.ligo.org.

Current activities of the LIGO research group at Kenyon College involve the maintenance and further development of the low-latency calibration pipeline for Advanced LIGO, which is the critical first step in low-latency gravitational-wave science, the development of new tools and methods for improving the monitoring and accuracy of low-latency calibration with LIGO, the development and investigation of methods to use gravitational- wave detections to probe the elusive equation of state of neutron star matter, the development of machine learning algorithms used to predictively veto glitches from LIGO data, and a search for a newly-hypothesized type of gravitational wave signal known as a gravitational-wave glint.

The gravitational-wave physics group at Kenyon is also involved in the NANOGrav collaboration through the NANOStars undergraduate research group.  The majority of this work is described below in the ROAR section, but a significant effort has also been put into developing a machine learning algorithm that ranks potential pulsar candidates in radio telescope data.

Projects

Calibration

Calibration is the necessary bridge between the instrument and data analysis. Data analysts search the interferometer strain, or fractional change in length of the interferometer arms, for gravitational-wave signals. However, the output of the interferometer does not directly represent the interferometer strain. The calibration effort is the process of reconstructing the strain from the interferometer output.  This process involves modeling the physics of the interferometer, digitizing these models, computing the strain, and thoroughly understanding all of the systematic errors that impact this process.

Prof. Maddie Wade was co-chair of the calibration working group within LIGO from 2015-2021.  She did a lot of the original development work on the official LIGO calibration pipeline that computes the strain data and continues to help maintain this software, and she is now helping to lead an effort to develop the next generation of calibration software.  Prof. Maddie Wade, along with several students and postdoctoral scholars, has also developed a new set of tools for monitoring the accuracy of the real-time calibrated strain data and better informing estimations of the uncertainty of the calibrated strain data.  Additionally, Prof. Maddie Wade and students have studied the impact of calibration errors on astrophysical results, such as the estimation of the parameters of a discovered gravitational wave source and the detection of compact binary events like merging black holes and neutron stars.  Prof. Maddie Wade and postdoctoral scholar Dripta Bhattacharjee are actively working to develop a new tool that will provide real-time estimates of the calibration uncertainty using machine learning algorithms and ongoing measurements of the interferometer response.

Neutron star equation of state parameter estimation

Neutron stars are some of the most interesting astrophysical bodies known.  They contain the mass of roughly one and a half of our Sun squished into the size of a city spinning faster than a kitchen blender.  Just a single teaspoon of neutron star matter weighs 10 million pounds.  The physics of matter at this density is not well understood, and the mathematical relationship that described the inner workings of these stars is called the neutron star equation of state.

On August 17, 2017, LIGO picked up the gravitational ripples from two merging neutron stars.  Prof. Leslie Wade and two of his research students had been developing software to measure the neutron star equation of state from binary neutron star mergers like this.  Using this software, the LIGO Scientific Collaboration reported on the first gravitational-wave constraint on the neutron star equation of state.  However, several more observations will need to be made in order to get a precise measurement.  Prof. Leslie Wade and students continue developing robust models for equation of state inference.  Additionally, Prof. Leslie Wade and Prof. Maddie Wade have worked with several research students to explore the use of machine learning algorithms in efficiently mapping between the gravitational-wave tidal deformability parameters and parameterized equation of state models for neutron stars. 

Machine learning and LIGO data quality

Gravitational wave signals from merging compact objects are signals that quickly enter and exit the LIGO detectors.  These types of signals are known as transient signals.  Unfortunately, terrestrial noise sources also show up as transient signals in ground-based interferometers like LIGO.  One of the major challenges in the analysis of LIGO data is to tease signals out of the noisy background of transient terrestrial signals.  

Our group has developed several hierarchical machine learning algorithms that predict the probability of a glitch being present in LIGO data based on input from auxiliary sensors at the LIGO detectors, such as thermometers, seismometers, and photometers.  We have developed an algorithm that is showing promising results where we have a series of neural networks that are trained to be experts in specific auxiliary systems of the LIGO detectors.  Predictions from each of these "experts" are then combined to form an overarching prediction about the presence of a glitch in LIGO data.  We are also working to develop a similarly-structured algorithm that will contain a series of experts in different types of glitches that are common to LIGO data.

Searching for gravitational-wave glints

Profs. Les and Maddie Wade's collaborators, Dr. Copi and Dr. Starkman at CWRU, predicted observable so-called gravitational-wave glints caused by the scattering of gravitational waves off of dense perturbers.  Profs. Les and Maddie Wade have been working with students at Kenyon College to search archival gravitational-wave data from the LIGO and Virgo detectors for evidence of gravitational-wave glints.  In the absence of a detection, we have also been developing methods for setting upper limits on the amplitude of gravitational-wave glints that could be present in the LIGO and Virgo data.

Pulsar candidate ranking using machine learning algorithms

Pulsars are rapidly spinning, highly dense remnants of massive stars.  They emit radiation, most observable in the radio frequencies, from their poles.  This emission, combined with their rapid rotation, leads to a lighthouse effect in how the radio waves are observed on earth.  Radio telescopes are the main tool used to search for new pulsars in our galaxy.  However, it takes significant effort to mine through radio telescope data in search of potential new pulsars.  Profs. Les and Maddie Wade have worked with two students over the years to develop a machine learning algorithm that will identify the likelihood of a particular pulsar candidate being a pulsar or not a pulsar.  This algorithm has performed well on test data and is currently being used to search archival radio telescope data to hopefully find elusive new pulsars.

Current Students

Kuba Kopczuk '26 - Kuba has been developing and refining methods for setting upper limits on the amplitude of gravitational-wave glints in LIGO and Virgo data.

Josephine Smith '26 - Josephine has continued the development of a machine learning algorithm for mapping from tidal parameters to neutron star equation of state parameters with several different equation of state models.

David Chintala '26 - David has worked to implement a class structure and unit tests into our machine learning algorithm for glitch identification.

Teddy Masters '25 - Teddy has continued development, implementation and testing of a machine learning algorithm to identify pulsars in radio telescope data.

Josh Temple '25 - Josh has worked to build a new data set for specific glitch types for our machine learning algorithm for glitch identification.

Luke Wilson '25 - Luke has performed hyperparameter tuning and feature studies for our machine learning algorithm for glitch identification.

Emmanuel Makelele '25 - Emmanuel has developed diagnostic tools for LIGO's photon calibrator system.

Past Students

Isa Braun '26 - Isa began development of a machine learning algorithm for mapping from tidal parameters to neutron star equation of state parameters with several different equation of state models.

Michael Van Keuren '24 - Michael continued with the development of a hierarchical machine learning algorithm for predicting the presence of glitches in LIGO strain data and specifically began developing an algorithm that would be trained to become an expert in predicting specific types of common glitches.  Michael also assisted in the search for gravitational-wave glints in LIGO and Virgo data.

James Hart '24 - James developed software for neutron star equation of state parameter estimation in bilby.

Kalista Wayt '23 - Kali developed a hierarchical machine learning algorithm where different neural networks were trained to become experts in specific auxiliary subsystems and then these networks were combined to form one final prediction.

Olivia Wilks '23 - Olivia began development of equation of state models with phase transitions for equation of state inference.

Sophie Schmitz, '24 - Sophie developed a hyperparameter tuning method for machine learning algorithms used to predict the presence of glitches in LIGO data.

Phillip Diamond, '24 - Phillip investigated the effectiveness of different features in a machine learning algorithm used to predict the presence of glitches in LIGO data.

Jeremy Baier, '22 - Jeremy upgraded equation of state software to be robust to equation of states with sharp phase transitions and implemented this new software in the bilby inference package.

Quinn Curren, '22 - Quinn developed an effective machine learning to identify pulsar candidates in data from radio telescope surveys.

Ezra Moguel, '21 - Ezra worked on improving the latency of LIGO calibration through new and innovative filter design techniques.  He has also worked to develop analysis and visualization techniques for the machine learning algorithms being developed to identify noise in LIGO data.

Paul Neubauer, '21 - Paul optimized the artificial neural network used for determining the quality of LIGO data and continuing development of a recurrent neural network for use in determining LIGO data quality.

Joe Lucaccioni, '21 - Joe developed and implemented a generalized polytropic model for measuring the neutron star equation of state from gravitational wave detections of binary neutron star mergers into the official LIGO software infrastructure.

Devon Nothard, '20 - Devon worked on implementing a neutron star equation of state model that is everywhere causal.

Maddie Stover, '20 - Maddie studied the impact of LIGO calibration errors on searches and parameter estimation for coalescing binary systems.

Chase Frederick, '20 - Chase worked on improving techniques for determining the accuracy of filters used in LIGO calibration.

Georgia Stolle-McAllister, '20 - Georgia performed a search for intermediate mass black hole binary systems on LIGO's second observing run data.

Burke Irwin, '19 - Burke developed methods for visualizing the results of an equation of state parameter estimation.

Kyle Rose, '19 - Kyle explored  improvements to the inputs of machine learning algorithms used to classify loud, transient noise events in LIGO data.

Campbell Fee, '18 - Campbell studied the effects of calibration error on gravitational-wave searches.

Matthew Carney, '18 - Matt developed LIGO parameter estimation software's ability to measure the neutron star equation of state.

Donald Moffa, '18 - Doni explored machine learning algorithms for the use of classifying loud, transient noise events in LIGO data.

Theresa Chmiel, '17 - Tracy worked on modeling Advanced LIGO calibration error and its effects on parameter estimation.

John Zellweger, '17 - Jack worked on optimizing algorithms to search for intermediate mass black hole binaries.

Radio and Optical Astronomy Research (ROAR) group

We run the ROAR group at Kenyon with the help of our colleague (and real astronomer) Paula Turner.  Our group is an integrated research and education group focusing on radio and optical astronomy.   Radio astronomy activities include taking observations with radio telescopes, participating in the NANOStars group (the undergraduate research group that is part of the NANOGrav collaboration), and building our very own radio telescope.   In particular, we have historically been trained to remotely control one of the world's largest radio telescopes, the Green Bank Telescope in West Virginia, to search for rapidly rotating neutron stars called pulsars.  The group helps search for new pulsars in radio telescope data through a process known as candidate ranking.  In addition, the group also studies variable stars using the 20-inch optical telescope in the Miller Observatory at Kenyon.  The group is open to anyone with an interest in astronomy.   We generally have 15-20 active student members.   

We use a team structure in the ROAR group.  Students are broken up into teams of 3-6 people with a team leader for each group.   The ROAR group meets as a large group once a week and each team must meet outside of the large group on a weekly or bi-weekly basis.  Teams study specific astronomy topics in more depth and develop learning tools and presentations on the topics they study.

Our motto in the ROAR group is that we  do astronomy,  learn astronomy, and promote astronomy.

Doing astronomy

We are active in two different astronomy wavelengths in the ROAR group - optical and radio astronomy.  Both optical and radio observing sessions are performed using the team structure in ROAR.  Data analysis activities can be performed either in teams or as a large group depending on the activity. 

Optical astronomy

We use our telescopes housed at Kenyon's own Miller Observatory to perform optical astronomy observations.  Prof. Paula Turner runs observations focused on variable stars in our galaxy.  Students are then able to analyze the data taken during their observations and draw conclusions from these analyses.

Radio astronomy

At Kenyon , we historically performed observations using the GBT.  Students both ran remote observing sessions that survey the sky looking for new pulsars and perform data analysis activities on data collected through this survey.  Students rank pulsar candidates in terms of their likelihood of being an astrophysical pulsar signal versus noise or RFI through various candidate ranking tools.  By participating in survey observations and pulsar candidate ranking, students are contributing to the effort to add pulsars to the pulsar timing arrays used by collaborations such as NANOGrav.  We are also actively building three of our own radio telescopes that can be used for observations of 21 cm emission in the Milky Way.

Learning astronomy

Students in ROAR engage in discussions about modern astronomy topics and encounter astronomy material in interactive lessons both in the large group meetings and in team meetings.  Our large group meetings often involve active discussions about recently published astronomy articles, fun activities to teach students more about a specific topic in astronomy, or discussions about videos or articles that were targeted at teaching students some introductory astronomy material.

Promoting astronomy

We have partnered with the local Mount Vernon High School (MVHS) to help start an astronomy club for students.  The Kenyon College ROAR group members help design activities for the MVHS astronomy club and help coordinate and run the MVHS astronomy club meetings.  Members of the MVHS astronomy club engage in hands-on activities learning about astronomy and participate in current data analysis activities through the Pulsar Search Collaboratory.  We also organize activities and events that involve both the MVHS astronomy club members and Kenyon ROAR group members working together collaboratively or just having fun up at the Miller Observatory.