Waveform Systematics for Black Hole Binary Mergers Models
Submission information
Submission Number: 178
Submission ID: 3972
Submission UUID: 3a2470e7-58fc-44a7-a6b8-3f0996458c7b
Submission URI: /form/project
Created: Thu, 08/24/2023 - 11:52
Completed: Thu, 08/24/2023 - 11:59
Changed: Thu, 06/13/2024 - 08:23
Remote IP address: 131.109.33.100
Submitted by: Gaurav Khanna
Language: English
Is draft: No
Webform: Project
Waveform Systematics for Black Hole Binary Mergers Models
Complete
Project Leader
Project Personnel
Project Information
Since the breakthrough in 2015, gravitational waves (GW) from about 90 black hole binaries
have already been observed. As GW detectors, such as LIGO, become ever more sensitive,
imperfections in the theoretical models of the GWs emitted from merging black hole binaries are
expected lead to significant biases in the estimated parameters (e.g. masses and spins) of
particularly loud GW events. This project will perform a study of such systematic effects by
leveraging the ML code "Dingo" to rapidly obtain posterior distributions for a number of relevant
waveform models. The main goal of this study is to create a visual map of measures of
discrepancies between the posteriors obtained for different waveform families for the same set
of signals.
The student will focus on learning computational tools to generate waveform datasets, train
neural networks, perform Bayesian inference with the Python-based Dingo code, compare the
resulting posterior distributions, and visualize their discrepancies on URI’s UNITY cluster.
have already been observed. As GW detectors, such as LIGO, become ever more sensitive,
imperfections in the theoretical models of the GWs emitted from merging black hole binaries are
expected lead to significant biases in the estimated parameters (e.g. masses and spins) of
particularly loud GW events. This project will perform a study of such systematic effects by
leveraging the ML code "Dingo" to rapidly obtain posterior distributions for a number of relevant
waveform models. The main goal of this study is to create a visual map of measures of
discrepancies between the posteriors obtained for different waveform families for the same set
of signals.
The student will focus on learning computational tools to generate waveform datasets, train
neural networks, perform Bayesian inference with the Python-based Dingo code, compare the
resulting posterior distributions, and visualize their discrepancies on URI’s UNITY cluster.
Project Information Subsection
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
University of Rhode Island -- Center for Computational Research
{Empty}
CR-University of Rhode Island
{Empty}
No
Already behind3Start date is flexible
6
Samuel_Clyne_Careers_project_launch.pptx
(1.87 MB)
{Empty}
Samuel_Clyne_6month_wrap.pptx.pdf
(1.77 MB)
{Empty}
- Milestone Title: Milestone #1
Milestone Description: The student will setup a Python environment and install the Dingo
code https://github.com/dingo-gw/dingo on Unity. He will familiarize himself with running the Dingo code for waveform generation and training of neural networks. The student will write Slurm scripts to submit such jobs to Unity. Launch presentation.
Completion Date Goal: 2023-10-01 - Milestone Title: Milestone #2
Milestone Description: The student will progress to train simple networks for binaries
with spins aligned with the orbital angular momentum vector. He will learn to make efficient use of CPU nodes for parallel waveform generation and NVIDIA A100 GPUs for training. He will then perform inference on a set of synthetic signals which vary over masses and effective spin.
Completion Date Goal: 2023-11-01 - Milestone Title: Milestone #3
Milestone Description: He will analyze and compare samples from posterior distributions obtained in Milestone #2 in jupyter notebooks using Unity's OpenOnDemand frontend, and compute different measures of discrepancy.
Completion Date Goal: 2023-12-01 - Milestone Title: Milestone #4
Milestone Description: The student will train networks for waveforms with higher order modes. He will perform inference on an extended signal parameter space which also varies the inclination angle of the source. Posteriors will be analyzed and compared in jupyter notebooks.
Completion Date Goal: 2024-01-01 - Milestone Title: Milestone #5
Milestone Description: The student will learn to perform importance sampling of Dingo
posteriors, making use of CPU cores for parallel computation of likelihoods. He will compute
Bayes factors between models.
Completion Date Goal: 2024-02-01 - Milestone Title: Milestone #6
Milestone Description: The student will create visualizations of the computed discrepancy
measures over the signal parameter space using Python notebooks. Wrap presentation.
Completion Date Goal: 2024-03-01
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
Final Report
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}
{Empty}