Laurence Datrier

Introduction
The detection of the binary neutron star merger (BNS) GW170817 and its electromagnetic (EM) counterparts marked the first joint gravitational wave-electromagnetic observations. The use of gravitational wave triggers is the most promising strategy for detecting more kilonovae (KNe), the faint optical transient associated with binary neutron star and neutron star-black hole (NSBH) mergers. However, new generations of optical telescopes like the LSST are expected to make serendipitous observations of kilonovae. We use KN models interpolated through Gaussian process regression to constrain kilonova parameters from incomplete light curves. We focus on recovering the merger time of the BNS, and consider the prospects for EM-triggered gravitational wave searches.

GW170817
Gravitational waves from two neutron stars coalescing were detected for the first time on 17th August 2017. These observations were followed
by a kilonova (AT 2017gfo) and a short GRB.

GAUSSIAN PROCESSES
A Gaussian process is defined as a collection of random variables, any finite number of which have a joint Gaussian distribution. Gaussian processes are used to infer directly in function space, by describing a distribution over functions.

KILONOVAE
Kilonovae are the faint, mostly isotropically emitting, long-lived optical and infrared transients associated with the merger of two neutron stars (BNS) or of a neutron star with a black hole (NSBH). They are promising candidates for the electromagnetic follow-up of gravitational wave observations of BNS and NSBH. New generations of telescopes will unveil populations of kilonovae both with and without GW triggers. The ejected neutron-rich matter in kilonovae undergoes rapid neutron capture (r-process) nucleosynthesis. This process enriches the universe with heavy elements such as gold and platinum.

MODELS
Models of kilonova light curves are obtained from time-resolved spectra by Kasen(2017). Each kilonova is made up of a tidal and a dynamical component. These models are the result of computationally expensive radiative transfer simulations, and are therefore available for only a discrete set of ejecta parameters. We use Gaussian Process regression to extend the models.

EM triggered GW searches
Up to 200 kilonovae detected by LSST could be generated by BNS mergers associated with subthreshold gravitational wave signals. While strategies are in place for the electromagnetic follow-up of gravitational wave triggers, optimising searches for the serendipitous discovery of kilonovae could lead to more prospects for multi-messenger Astronomy.

RESULTS
We run a full parameter estimation on kilonova light curves, both simu-
lated and real. The kilonovae are made up of two components (tidal and
dynamical), each with three parameters:
• Lanthanide fraction X lan
• Mass of ejecta m ej
• Vvelocity of ejecta v k .
The merger time t 0 and luminosity distance d L are also allowed to vary.
We use wide, flat priors for all parameters.
Figure 1 and Figure 2 show the results of the full parameter estimation
on g, r, i observations of AT 2017 gfo. In Figure 1, observations start
3 days after the merger, while in Figure 2, observations start 1.2 days
after the merger. Parameters are recovered, with some bias in distance.
Figure 3 shows results for a simulated kilonova light curve, with a fixed
luminosity distance d L , for nightly g, r, i observations starting 2 days after
the merger.

DISCUSSION
Using incomplete kilonova light curves, we can accurately recover the merger time of BNS to within one to three days, depending on when observations of the transient start. All ejecta parameters are also recovered, with the largest source of degeneracy coming from the luminosity distance. This however can be fixed with host galaxy identification. The recovery of the merger time from kilonova light curves only is a promising prospect for confirming subthreshold or single detector gravitational wave events