GPCCPaper

❗Following the recent updated GPCC.jl, the instructions below are no longer up to date❗ An update will follow soon.

ℹ What is this?

This repository contains scripts for reproducing all the results presented in

A Gaussian process cross-correlation approach to time delay estimation for reverberation mapping of active galactic nuclei (https://doi.org/10.1051/0004-6361/202345932)

using the implementation at GPCC.jl and the data available at GPCCData.jl.

Please clone this repository in order to use it.

To reproduce the results you will need to add the AINJuliaRegistry.

▶ Synthetic observations drawn from GP with OU kernel

We generate synthetic data that conform to the model with increasing noise levels.We note how various other peaks arise as the noise increases. In the presence of low noise, the true peak at 2 days is the incontestable winner. Beyond that, other peaks start to appear as potential candidates.

Let us look closer at case σ=1.0. We see that there is the true peak at two days and large peaks towards the right end of the posterior plot. We align the light curves according to these two candidate peaks. We note that they also seem to lead to plausible alignments.

These synthetic experiments can be reproduced following instructions here and the code located in the corresponding folder.

▶ Real observations

#object	v	ev	mass	emass	delay	edelay	z	luminosity
Mrk335	1293	64	4.6e6	0.5e6	14.0	0.9	0.0258	5.01e43
Mrk1501	3321	107	33.4e6	4.9e6	13.8	5.4	0.0893	2.09e44
3C120	1514	65	12.2e6	1.2e6	25.6	2.4	0.0330	9.12e43
Mrk6	3714	68	24.8e6	2.3e6	10.2	1.2	0.0188	5.62e43
PG2130099	1825	65	8.3e6	0.7e6	9.7	1.3	0.0630	1.41e44

The data can be conveniently accessed here.

We compute the following posteriors:

The above results can be reproduced by following the instructions here.

▶ Three lightcurve experiment, posterior for Mgc0811

We compute the following posteriors:

The above results can be reproduced by following the instructions here.

▶ Kernel selection via cross-validation

The proposed GPCC model delivers out-of-sample predictions. It can be subjected to model selection in a cross-validation framework. This allows us to compare GPCC to other models. It also allows us to select the kernel function.

The above results can be reproduced by following the instructions here.

▶ Mass distribution for 3C120 and Mrk1501

The above results can be reproduced by following the instructions here.