English

# Item

ITEM ACTIONSEXPORT
Statistical Gravitational Waveform Models: What to Simulate Next?

Doctor, Z., Farr, B., Holz, D. E., & Pürrer, M. (2017). Statistical Gravitational Waveform Models: What to Simulate Next? Physical Review D, 96: 123011. doi:10.1103/PhysRevD.96.123011.

Item is

### Basic

show hide
Genre: Journal Article

### Files

show Files
hide Files
:
1706.05408.pdf (Preprint), 2MB
Name:
1706.05408.pdf
Description:
Visibility:
Public
MIME-Type / Checksum:
application/pdf / [MD5]
-
-
:
PRD.96.123011.pdf (Publisher version), 2MB

-
Name:
PRD.96.123011.pdf
Description:
-
Visibility:
Restricted (Max Planck Institute for Gravitational Physics (Albert Einstein Institute), Potsdam-Golm; )
MIME-Type / Checksum:
application/pdf
-
-
-

show

### Creators

show
hide
Creators:
Doctor, Zoheyr, Author
Farr, Ben, Author
Holz, Daniel E., Author
Pürrer, Michael1, Author
Affiliations:
1Astrophysical and Cosmological Relativity, AEI-Golm, MPI for Gravitational Physics, Max Planck Society, ou_1933290

### Content

show
hide
Free keywords: Astrophysics, High Energy Astrophysical Phenomena, astro-ph.HE,General Relativity and Quantum Cosmology, gr-qc
Abstract: Models of gravitational waveforms play a critical role in detecting and characterizing the gravitational waves (GWs) from compact binary coalescences. Waveforms from numerical relativity (NR), while highly accurate, are too computationally expensive to produce to be directly used with Bayesian parameter estimation tools like Markov-chain-Monte-Carlo and nested sampling. We propose a Gaussian process regression (GPR) method to generate accurate reduced-order-model waveforms based only on existing accurate (e.g. NR) simulations. Using a training set of simulated waveforms, our GPR approach produces interpolated waveforms along with uncertainties across the parameter space. As a proof of concept, we use a training set of IMRPhenomD waveforms to build a GPR model in the 2-d parameter space of mass ratio $q$ and equal-and-aligned spin $\chi_1=\chi_2$. Using a regular, equally-spaced grid of 120 IMRPhenomD training waveforms in $q\in[1,3]$ and $\chi_1 \in [-0.5,0.5]$, the GPR mean approximates IMRPhenomD in this space to mismatches below $4.3\times 10^{-5}$. Our approach can alternatively use training waveforms directly from numerical relativity. Beyond interpolation of waveforms, we also present a greedy algorithm that utilizes the errors provided by our GPR model to optimize the placement of future simulations. In a fiducial test case we find that using the greedy algorithm to iteratively add simulations achieves GPR errors that are $\sim 1$ order of magnitude lower than the errors from using Latin-hypercube or square training grids.

### Details

show
hide
Language(s):
Dates: 2017-06-162017
Publication Status: Published in print
Pages: -
Publishing info: -
Rev. Method: -
Identifiers: arXiv: 1706.05408
URI: http://arxiv.org/abs/1706.05408
DOI: 10.1103/PhysRevD.96.123011
Degree: -

show

show

show

### Source 1

show
hide
Title: Physical Review D
Other : Phys. Rev. D.
Source Genre: Journal
Creator(s):
Affiliations:
Publ. Info: Lancaster, Pa. : American Physical Society
Pages: - Volume / Issue: 96 Sequence Number: 123011 Start / End Page: - Identifier: ISSN: 0556-2821
CoNE: /journals/resource/111088197762258