act-axion-analysis

Overview

This repo contains code to study time-dependent polarization rotations that could be induced in CMB polarization measurements made by the Atacama Cosmology Telescope by axion-like particles.

The main analysis functions live in act_axion_analysis itself. The scripts folder contains the main scripts for running the code - get_depth1_angle_parallel.py and get_depth1_angle_serial.py. As the names suggests, these scripts obtain a rotation angle for an ACT depth-1 (read "depth one") map, which is a map in which every pixel is visited only once in a given observational period (usually on the order of hours), relative to a reference map (usually a season-long or longer coadd map for each array) in either parallel (using mpi4py) or serial fashion using a power spectrum based estimator. They can also calculate a calibration factor of the TT spectrum relative to two coadd maps. All of the various settings and paths to the needed maps are set in the configuration files, dr6_depth1_ps_config.yaml or dr6_depth1_ps_serial_config.yaml. The only difference between the two is a few more plotting options in the serial version. See the configuration file section below for more information.

For running on Perlmutter at NERSC, there is also a shell script, run_alp_analysis.sh, that sets up the working environment, generates a slurm shell script to distribute the code across many nodes and processes, and then runs the slurm shell script with sbatch. Before using, make sure to change the RUN_TAG environment variable, the maximum walltime, the number of nodes, the number of processes per node, and any paths.

Standard procedure for running the parallel script by itself is:

$ python3 get_depth1_angle_parallel.py dr6_depth1_ps_config.yaml <tag>

where the tag is some identifying string that gets added to the output directory name and the output files to distinguish them as belonging to this run. In the MPI-enabled version, an output npy file will be saved for each depth-1 map that contains a dictionary with all the output products generated by the script. The YAML config file will also be saved to the output directory and includes a list of all the maps processed during the run. Various plots can also be generated by the script if the correct configuration flags are set.

There are also helper scripts for the parallel version on NERSC. To ensure that all paths are correct and maps are correct in the config file, run validate_config_paths.py with the path to the config file to check as an argument. To gather the npy files from each depth-1 map into a single npy file that contains a dictionary whose top-level keys are the names of the depth-1 maps, the script collect_npy_files.py can be used. It is automatically called at the end of run_alp_analysis.sh.

To run the serial script, use:

$ python3 get_depth1_angle_serial.py dr6_depth1_ps_serial_config.yaml

Note that there is no tag here - the default is to use the start time of the run to generate a unique output file. As mentioned above, the serial config file is identical to the parallel one except for a handful of extra plotting options to generate summary plots after all of the maps have finished running and one additional boolean The serial version has been made more like the parallel version, so it also saves an output npy file for each map in addition to saving one large npy file containing results for all maps.

The resources folder contains a handful of files that can be used for the CAMB theory spectra and the ACT DR4 bins, if desired.

There is also a branch called "serial" that contains only an old version of the serial code and a notebooks folder with various Jupyter notebooks that were used for testing and debugging aspects of the code. The notebooks themselves contain some documentation, but these provide a sort of history of this project including notebooks with simulations that test parts of the code for accuracy. Many of the notebooks, as a result, use older or developmental parts of the code, and they may be difficult to interpret without additional notes or careful study.

Dependencies

The following packages are required to use this code. I usually install them from conda-forge in a conda environment, and the version numbers in parentheses represent the versions against which the code has been tested most extensively (as of February 2025).

• Python (3.12.8)
• pixell (0.27.2)
• numpy (1.26.4 - has not been tested yet for numpy 2.0+!)
• matplotlib (3.10.0)
• scipy (1.15.1)
• pyyaml (6.0.2)
• tqdm (4.67.1)
• mpi4py (4.0.2)

Other versions may work (especially slightly older versions than these), but no guarantees.

Installing all of these from conda-forge does have the downside that the pixell v0.27.2 wheel does not allow the Intel MKL backend for numpy to be installed (nor numpy 2.0+). The ease of installation and ensuring that everything works together is enough that I have not done more to get around these constraints.

Installation

Currently must download from GitHub directly via `git clone`. To avoid downloading the old and rather large testing notebooks, I would recommend using:

$ git clone https://github.com/zhuber21/act-axion-analysis --single-branch --depth 5

This pulls down only the main branch and just the last five commits. Once downloaded, it can be installed with:

$ pip install . --user

Configuration File

As of February 2025, here are the available settings in the YAML config file and information about best practices for setting them. The paths, in particular, assume that I am running the code for ACT DR6 data, but they could be placed with appropriately similar maps, beams, etc. for future analyses with ACT/SO/etc.

• freq - the frequency of maps being run. Options are 'f090', 'f150', and 'f220' (though production runs on NERSC only occured for f090 and f150)
• Filtering parameters
  • kx_cut - cutoff in x Fourier modes (default 90)
  • ky_cut - cutoff in x Fourier modes (default 50)
  • unpixwin - boolean about whether to remove pixel window (default True)
• Apodization parameter
  • filter_radius - the apodization radius in degrees (default 0.5, though this is applied twice if using ivar weighting)
• Likelihood fitting settings
  • angle_min_deg - the minimum angle for the likelihood fitting (default -50.0)
  • angle_max_deg - the maximum angle for the likelihood fitting (default 50.0)
  • num_pts - the number of points between angle_min_deg and angle_max_deg at which the likelihood is evaluated (default 50000) * If using curvefit, 50,000 points is almost certanily enough to guarantee results match many more points out to O(0.001) precision. If not using curve_fit, may want to use more points to guarantee good precision.
  • fit_method - which method to use to fit angle likelihoods (default 'fwhm' - options are 'fwhm', 'curvefit', and 'moment')
• Calibration factor likelihood fitting settings
  • y_min - the minimum calibration factor for the likelihood fitting (default -1.0 - allows us to catch low values with large errorbars)
  • y_max - the maximum calibration factor for the likelihood fitting (default 2.0)
  • cal_num_pts - the number of points between y_min and y_max at which the likelihood is evaluated (default 50000)
  • cal_fit_method - which method to use to fit cal likelihoods (default 'fwhm' - options are 'fwhm', 'curvefit', and 'moment')
• Calibration factor binning settings
  • cal_bin_size - bin width (in ell) of the bins for the TT calibration (usually 200)
  • cal_lmin - minimum ell for the TT calibration (usually 1000)
  • cal_lmax - maximum ell for the TT calibration (usually 2001 - this ensures that we get the bin ending at 2000)
• Angle estimator binning settings
  • bin_settings - options are "regular" and "DR4" to use even bins of "bin_size" or to use the ACT DR4 bins, respectively (default "regular" - the DR4 option is a legacy test)
  • bin_size - (used with "regular") bin width (in ell) of the bins for the angle estimation (usually 400)
  • lmin - (used with "regular") minimum ell for the angle estimation (usually 1000)
  • lmax - (used with "regular") maximum ell for the angle estimation (usually 3001)
  • start_index: 11 # Used with "DR4" - refers to index in DR4 bin file
  • stop_index: 47 # Used with "DR4" - refers to index in DR4 bin file
• Power spectra analysis settings
  • use_ivar_weight - boolean setting whether or not to use inverse variance weighting for calculating spectra (default True)
  • cross_calibrate - boolean setting whether or not to do TT calibration (usually True, but will not affect angle estimation and will speed things up to set to False)
• Output options - all the plotting booleans are generally False on NERSC, but were very helpful for debugging during local testing. There are additional options for the serial branch: plot_all_spectra, plot_summary_spectra, and plot_angle_hist.
  • output_dir_root - path to the directory to which output files are saved (npy files with results, config YAML, any plots)
  • plot_maps - boolean for whether to save plots of trimmed maps and masks in analysis
  • plot_likelihood - boolean for whether to save plots of angle estimation likelihood
  • plot_beam - boolean for whether to save plots of binned beams
  • plot_tfunc - boolean for whether to save plot of binned filtering transfer function
• Paths
  • theory_curves_path - the path to a CAMB .dat file containing the best-fit LCDM cosmology spectra
  • ref_pa4_path - path to the pa4 reference map for the angle estimation (usually a night-time pa4 ACT DR6 coadd)
  • ref_pa4_ivar_path - path to the pa4 reference map inverse variance (ivar) map for the angle estimation
  • ref_pa4_beam_path - path to the beam for the pa4 reference map
  • ref_pa5_path - path to the pa5 reference map for the angle estimation (usually a night-time pa5 ACT DR6 coadd)
  • ref_pa5_ivar_path - path to the pa5 reference map inverse variance (ivar) map for the angle estimation
  • ref_pa5_beam_path - path to the beam for the pa5 reference map
  • ref_pa6_path - path to the pa6 reference map for the angle estimation (usually a night-time pa6 ACT DR6 coadd)
  • ref_pa6_ivar_path - path to the pa6 reference map inverse variance (ivar) map for the angle estimation
  • ref_pa6_beam_path - path to the beam for the pa6 reference map
  • pa4_beam_path - path to beam tform file for ACT DR6 pa4 (e.g. coadd_pa4_f150_night_beam_tform_jitter_cmb.txt")
  • pa5_beam_path - path to beam tform file for ACT DR6 pa5 (e.g. coadd_pa5_f150_night_beam_tform_jitter_cmb.txt")
  • pa6_beam_path - path to beam tform file for ACT DR6 pa6 (e.g. coadd_pa6_f150_night_beam_tform_jitter_cmb.txt")
  • galaxy_mask_path - path to the galaxy mask (usually using the ACT 70% galaxy mask)
  • cal_map1_path - path to the map for the first calibration coadd
  • cal_ivar1_path - path to the ivar map for the first calibration coadd
  • cal_beam1_path - path to the beam for the first calibration coadd
  • cal_map2_path - path to the map for the second calibration coadd
  • cal_ivar2_path - path to the ivar map for the second calibration coadd
  • cal_beam2_path - path to the beam for the second calibration coadd
  • obs_list - a .txt file containing the names of all of the maps to run
  • obs_path_stem - the path to the directory containing all of the depth-1 maps