Here I give a quick blurb about some projects I'm working on and some old projects from the past.
UNDER CONSTRUCTION
A Pulsar Timing Array (PTA) use many radio pulsars to detect gravitational waves (GW). The pulsars give off periodic radio pulses (hence their name). Pulsars are some of the most stable clocks in the universe, meaning we can predict the time of arrival of the pulses from a pulsar to very high accuracy.
By looking for differences between the expected and actual time of arrival of the radio pulses we can detect gravitational waves that pass between the pulsar and the Earth. Because the Earth is moving with respect to the Sun, we actually use the center of mass of the Solar System as our local reference point.
LIGO monitors the distance between mirrors on Earth, looking for changes due to passing gravitational waves.
A PTA looks for a changing distance between the solar system and radio pulsars.
I am one of the development leads for enterprise, NANOGrav's flagship GW data analysis software. I have also done a lot of work on enterprise_extensions, a collection of "recipes" to implement common PTA data analysis techniques in enterprise.
NANOGrav's Pulsar Signal Simulator, PsrSigSim, is an end to end simulation of pulsar data. It simulates the emission from the pulsar, the pulse's propagation through the interstellar medium, and all the way to detection by a radio telescope on Earth. It is being used by NANOGrav to study sources of noise in our galactic GW detector.
I did some early work on the BayesWave algorithm, which is a method for detecting unmodeled bursts of GWs. Instead of knowing a priori what GWs to look for, you look for any excess of power coherently across multiple detectors. That way if your model of what you expect is wrong, you can still find something. That's also a good way to discover things that you don't know are even out there!
BayesWave fits GWs as a collection of wavelets. Originally we used a discrete wavelet decomposition, but now BayesWave uses continuous Morlet-Gabor wavelets. BayesWave figures out how many wavelets to use and what the wavelets' frequencies and durations are. It uses a reversible jump Markov chain Monte Carlo (RJMCMC) to determine the best fit wavelets and the uncertainty of the fit.
BayesWave can also be sent to work to characterize the detector noise. LIGO suffers from non-stationary, non-Gaussian noise. BayesWave is able to fit the non-Gaussian parts of the noise (glitches) and can output a betting odds that an event is a GW or a glitch. BayesWave's glitch fitting capabilities have now been used to clean LIGO data by fitting and removing a glitch that occurred during the GW170817 binary neutron star detection!
You can see what was done in figure 2 of the detection paper (reproduced left). A brief overview also appears in the science summary.
When the orbits of a binary black hole system are highly eccentric, the GWs produced will look like a series of repeated bursts. The bursts occur at periastron, when the black holes are moving the fastest and producing the most GWs.
I am working with a collaborator at WVU to develop a method based on BayesWave that associates the repeated bursts as a single GW signal. This allows us to detect highly eccentric binary black holes without fancy waveform modeling.
BayesWave is very computationally expensive and operates as a follow-up to candidate events from other low-latency searches. BayesWave doesn't always agree with those other searches as to what is interesting, so it spends some time following up events that are not very likely to be GWs. I'm currently working to develop an new low-latency burst search to generate candidate events (triggers) tuned specifically for BayesWave. We are calling this method FastBurst.
FastBurst uses a technique called wavelet denoising, which is computationally cheap enough to run in real time as data is collected. We have demonstrated that it works and are currently optimizing code and preparing for the first large scale tests!
4 seconds of simulated data for two LIGO-like detectors. Data contains an injected GW signal that appears in both detectors, and an injected non-Gaussian noise glitch that appears in detector 1 only.
FastBurst coherent reconstruction of the gravitational wave signal. Note the good agreement with the injected waveform.
Way back in graduate school I worked on the search for black hole ringdowns during initial LIGO. The ringdown is the final phase of a binary blackhole coalescence: inspiral, merger, ringdown. The two black holes have merged into one, resulting in a black hole that's event horizon is perturbed from the stationary Kerr black hole. The event horizon vibrates like a ringing bell with harmonics determined by the mass and angular momentum of the black hole. The vibration stimulates gravitational wave (GW) emission that carries away energy until the black hole reaches the stationary Kerr state.
The ringdown GWs occur at higher frequency than the insprial and merger. A ringdown only search like ours looks for black holes with higher masses than inspiral searches. Our target black holes would inspiral and merge outside of LIGO's frequency band, but the ringdown would be in band.
We applied a machine learning algorithm, random forest of bagged trees, to better separate candidate events from noise glitches (the first use of machine learning in GW astronomy!). Our non-detection placed the most stringent limits on the rate of intermediate mass black hole mergers at the time.
The image at the top of this page was the Astronomy Picture of the Day on 11 February 2016 the day of the announcement of the first direct detection of gravitational waves. The image was made by Aurore Simonnet of Sonoma State University for LIGO.