Galadirith / lalinference-injections

Examples of doing model and NR injections and parameter estimation with lalinference

Injections with lalinference

Summary

1. Selecting a noise curve

   If you are not provided with a noise file derived from actual data, then you can use the analytically calculated PSDs/ASDs. A useful reference for where to find ASCII files containing noise curves is bilbys docs. You can also find many noise curves already installed on clusters, although their names can be quite cryptic. A typical ASD noise curve used for Livingston, Hanford and Virgo are

   /cvmfs/ligo-containers.opensciencegrid.org/lscsoft/conda/latest/envs/ligo-py36/share/lalsimulation/LIGO-T0900288-v3-ZERO_DET_high_P.txt
   /cvmfs/ligo-containers.opensciencegrid.org/lscsoft/conda/latest/envs/ligo-py36/share/lalsimulation/LIGO-P1200087-v18-AdV_DESIGN.txt
   

   However there are many sources of noise files (LIGO-T1800044-v5, LIGO-P1200087-v47) so if you are unsure please ask those currently working with parameter estimation.

   While it is possible to pass a noise model (ie specify aLIGOZeroDetHighPower rather than the path to a PSD file) to some parts of the parameter estimation pipeline, models are not consistently names, and it is also not always possible. As such the best thing to do is use a noise file. Noise files are supported at every stage and you know exactly what you are providing the PE pipeline.

2. Generate gravitational wave detector responses

   For each detector that is in your network, we create a *txt file for each detector. This is done using the program pycbc_generate_hwinj and has all the options of the normal waveform interface in lalsimulation. You also specify the PSDs or ASDs for each detector which are used to determine the correct distance based on an SNR that you provide. Also optionally to also generate noise however it is typical for most injection runs to be performed with no noise.

3. Prepare responses as frame files for each detector

   The *txt detector responses have to be converted to *gwf frame files. They are a mostly difficult to understand binary format. The easiest way to manipulate them is using gwpy.

4. Perform parameter estimation on frame files

   You can now configure your parameter estimation job in mostly the same way as you normally would. There are a few modifications that need to be made to make it consistent with your frame files. In particular you need to make sure that the data.channels match your frame files and that you use the same PSDs/ASDs that you used to generate your frame detector responses.

Example steps (S190814bv)

The following steps work when running on the LIGO hawk head node. These instructions set up IMRPhenomHM injection using the estimated parameters for S190814bv. We then perform parameter estimation using IMRPhenomHM.

1. Activate compute environment

   If you are not already in the (ligo-py36) environment on the hawk head node then you can activate it using the quickstart guide on docs.ligo.org which are summarised here

   source /cvmfs/ligo-containers.opensciencegrid.org/lscsoft/conda/latest/etc/profile.d/conda.sh
   conda activate ligo-py36

2. Generate gravitational wave detector responses

   You now need to generate ASCII injection files for each interferometer in your network. In this example we will generate injection files for a HLV network. The program that does this is pycbc_generate_hwinj. The file ./generate_hwinj_HM.sh has already been populated with all the necessary arguments.

   ./generate_hwinj_HM.sh

3. Prepare responses as frame files for each detector

   The ASCII injection files now need to be converted to *gwf frame files. This uses the program pycbc_insert_frame_hwinj. One *gwf if generated for each interferometer. The file ./insert_frame_hwinj.sh has been populated with all the necessary arguments.

   ./insert_frame_hwinj.sh

4. Perform parameter estimation on frame files

   Finally we generate the resources for parameter estimation and start the parameter estimation run. This uses the program lalinference_pipe. In order for the parameter estimation programs to find the *gwf files that were generated in the previous step, they need to be placed in lalinferencemcmc and lalinferencenest subdirectories of the working directory for the lalinference job. The file ./pipe_HM.sh has been populated with all the necessary features to achieve this. The *gwf are system linked into the appropriate directories.

   ./pipe_HM.sh

License

lalinference-injections is released under the MIT license.