TFPV

DESI Tully Fisher Peculiar Velocity

Instructions

Data

Data Environmental variables

Location of input and output data files are communicated through environmental variables. For example on my local computer I do export DESI_SGA_DIR=/Users/akim/Projects/DESI_SGA export OUTPUT_DIR=/Users/akim/Projects/TFPV/output export DATA_DIR=/Users/akim/Projects/TFPV/data

Input data

• Download DESI_SGA repository https://github.com/DESI-UR/DESI_SGA. These have data required for the analysis:

  • Set environmental variable DESI_SGA_DIR to the directory
  • Tully's table of cluster information $DESI_SGA_DIR/TF/Tully15-Table3.fits
  • Clusters and which galaxies are in them $DESI_SGA_DIR/TF/output_??????.txt

• Get other required private data

  • Set environmental variable DATA_DIR to the directory where these files will be located
  • On NERSC DATA_DIR=/global/cfs/cdirs/desi/science/td/pv/tfgalaxies/ otherwise wherever you
  • Other data on NERSC /global/cfs/cdirs/desi/science/td/pv/tfgalaxies/.
    • The cluster catalog "./Y1/DESI-DR1_TF_pv_cat_v3.fits"
    • Coma "SV/SGA-2020_fuji_Vrot.fits".

Ouput data

• Set environmental variable OUTPUT_DIR to where created data go

Docker image

• alexgkim/tfpv:dev
• docker run -v $DATA_DIR:/data -v $OUTPUT_DIR:/output -v $DESI_SGA_DIR:/DESI_SGA alexgkim/tfpv:dev python /opt/TFPV-docker/fitstojson.py; /opt/TFPV-docker/cluster sample algorithm=hmc engine=nuts max_depth=17 adapt delta=0.999 num_warmup=3000 num_samples=1000 num_chains=4 init=/output/iron_cluster_init.json data file=/output/iron_cluster.json output file=/output/cluster.csv

Local install

• Install cmdstan
• Download TFPV repository https://github.com/AlexGKim/TFPV. Personally I cloned the repository to live inside the cmdstan installation cmdstan/TFPV given that conformed to the STAN examples.
• I don't have a "requirements.txt" file with all the python packages needed to run the software. Just create a new environment, run, crash, install, repeat...
• Create the data and initial condition files input to STAN. This is done with the command "python fitstojson.py". If you look at the file filtstojson.py, you will find several functions that can be called through main.
  • iron_cluster_json() creates the data for IRON, coma_json() creates the data for EDR.
  • coma_json() has a keyword for cuts, which in my analysis is set to True.
• The STAN code to fit IRON is cluster.stan, the code to fit Coma is coma.stan.
• Each code can be run with different configurations. The one to care about is controled by the variable "dispersion_case". "dispersion_case=4" is the perpendicular fit, "dispersion_case=3" is the ITF fit despite what is written in the comments.
• Use standard cmdstan to compile the STAN code. From cmdstan/ "make TFPV/cluster" for example. You will find a new executable TFPV/cluster. Personally I immediately rename it to include the dispersion case. So I have files "cluster311", "cluster411", "coma411", "coma311"
• At the top of cluster.stan, coma.stan are commented out command line commands I used to run the fit. Run them and hope for the best. Note it puts the output into a directory "output" you have to create.
• STAN provides two useful tools to check the result. The first checks for the convergence. from TFPV "../bin/diagnose output/cluster_311_?.csv" for example. The second shows statistics on the posterior "../bin/stansummary output/cluster_311_?.csv"
• Plots for both cluster and coma runs are made using methods in plot_iron.py