Constraining the Origins of Binary Black Holes using Multiple Formation Pathways

Dataset for Zevin et al. 2021. The codebase AMAZE used in this work can be found on Github. 

• "models.tar.gz" contains and hdf5 file with all the astrophysical models, organized in the hierarchical structure used in the code AMAZE (*channel*/*param1*/*param2*/...). This is used in the main analysis for inference on branching fractions and physical prescriptions.
  
  The .hdf5 file contains systems from each of the 5 formation channels considered: Common Envelope (CE), Chemically Homogeneous Evolution (CHE), Globular Clusters (GC), Nuclear Star Clusters (NSC), and Stable Mass Transfer (SMT). Each formation channel has 5 variations on the natal spin of black holes: natal spins of 0.0 (chi00), 0.1 (chi01), 0.2 (chi02), and 0.5 (chi05). For the common envelope channel, there are also 5 variations of the common envelope efficiency: 0.2 (alpha02), 0.5 (alpha05), 1.0 (alpha10), 2.0 (alpha20), and 5.0 (alpha50). 
  
  After downloading the file, one can use the following code to investigate, for example, the CE channel with natal spins of 0.2 and common envelope efficiency of 1.0, as well as the GC channel with natal spins of 0.0.
  First, unzip the tarball using the command line:
  tar -xzvf models.tar.gz
  Next, in python read in the models as a Pandas dataframe
  import pandas as pd
  CE_chi02_alpha10 = pd.read_hdf('models_reduced.hdf5', key='CE/chi02/alpha10')
  GC_chi00 = pd.read_hdf('models_reduced.hdf5', key='GC/chi00')
  
  These dataframes hold information to the systems in each population model. The keys in the dataframe are: 
  • 'mchirp', 'q', 'chieff', 'z' : the chirp mass, mass ratio (secondary mass divided by primary mass), effective spin, and redshift of merger. These are the 4 parameters used in the analysis of Zevin et al. 2021. 
  • 'm1', 'm2', 's1x', 's1y', 's1z', 's2x', 's2y', 's2z' : the component masses and component dimensionless spin vectors. These construct the parameters above, though are not explicitly used in the analysis of Zevin et al. 2021. 
  • 'weight' : the relative weight of each system in the population model. This includes the astrophysical weight (i.e., system 1's probability of being formed in a specific channel compared to system 2) as well as the cosmological weight of this system being detected by GW detectors (accounting for the volume of the universe at a particular merger redshift and time dilation). These weights do not hold meaningful units and are not comparable across different population models; they are used to construct a weighted distribution for each channel to provide the properties of systems we would see for that channel if we had infinitely-sensitive GW detectors. 
  • 'pdet_midhighlatelow_network', 'snropt_midhighlatelow_network', 'pdet_midhighlatelow', 'snropt_midhighlatelow' : the detection probabilities and optimal SNRs of each system. The values with the '_network' tag assume a 3-detector network consisting of LIGO-Hanford, LIGO-Livingston, and Virgo operating at midhighlatelow sensitivity with a SNR detection threshold of 10, whereas the values without this tag assume a single LIGO detector operating at midhighlatelow sensitivity with a SNR detection threshold of 8. These are used, for example, to construct detection-weighted distributions for each channel. 
     
• "gw_events.tar.gz" contains the data for all binary black hole events from GWTC1 and GWTC2, processed to have the parameters used in this project. These use the publicly-available samples from the previous links, using the 'Combined' and 'PublicationSamples' posterior samples for GWTC-1 and GWTC-2, respectively. The Jupyter notebook that processes the various LVK data releases to get the GW event posteriors in the proper format can be found on the public git repository. In addition to the event-level parameter posteriors, prior weights for each posterior sample are also included (labeled `p_theta` in the files), since these are needed to get draws from the likelihood that is needed in hierarchical inference. Note that this notebook has been updated to also process the data from GWTC-2.1 and GWTC-3, though these catalogs were not used in Zevin et al. 2021. 
   
• "beta_prior.tar.gz" contains prior samples for the branching fractions in the 2-channel and 5-channel case. This was used to create the prior distribution in Figure 5 of Zevin et al. 2021.
   
• "trails.tar.gz" contains all the inference output used in the study. Both underlying samples (key='model_selection/samples') and detectable samples (key='model_selection/detectable_samples') are stored as pandas dataframes in each hdf5 file. The trailing number in each hdf5 file is the trial number, each with a different random seed. These output files were used to create Figure 2-8 of Zevin et al. 2021. 

IF USING THE ASTROPHYSICAL MODELS IN `models.tar.gz`, please cite the relevant work: 

CE/SMT models: Bavera et al. 2021

CHE models: du Buisson et al. 2020

GC models: Rodriguez et al. 2019

NSC models: Antonini et al. 2019

In addition, one should cite Zevin et al. 2021 (this work) with any use of this dataset.