ACT DR6 + SPT-3G + Planck NPIPE Lensing Likelihood

This repository contains likelihood software for the ACT + SPT CMB lensing analysis. If you use this software and/or the associated data, please cite the following papers:

• ACT and SPT Collaborations (2025), arXiv:2504.20038
• F. Ge et al (SPT-3G Collaboration) 2024, arXiv: 2411.06000
• Mathew S. Madhavacheril et al (ACT Collaboration) 2024 ApJ 962, arXiv:2304.05203
• Frank J. Qu et al (ACT Collaboration) 2024 ApJ 962 112, arXiv:2304.05202
• Niall MacCrann et al (ACT Collaboration) 2024 ApJ 966 138, arXiv: 2304.05196

In addition, if you use any variant involving Planck lensing from the likelihood, please also cite:

• Julien Carron et al JCAP09(2022)039, arXiv:2206.07773

Step 1: Install

First clone this repository, then install:

pip install .

Tests can be run using

python -m unittest discover -s act_dr6_spt_lenslike/tests

Step 2: download and unpack data

ACT

Download the likelihood data tarball for ACT DR6 lensing from NASA's LAMBDA archive.

Extract the tarball into the act_dr6_spt_lenslike/data/ directory in the cloned repository such that the directory v1.2 is directly inside it. Note that the bandpowers and covariances are already available in the above directory. Only the likelihood correction matrices are required in the installation. Only then should you proceed with the next steps.

SPT

The SPT bandpowers are found already in the data directory above.

Step 3: use in Python codes

Generic Python likelihood

import act_dr6_spt_lenslike as apslike

variant = 'act_baseline'
lens_only = False # use True if not combining with any primary CMB data
like_corrections = True # should be False if lens_only is True

# Do this once
data_dict = apslike.load_data(variant,lens_only=lens_only,like_corrections=like_corrections)
# This dict will now have entries like `data_binned_clkk` (binned data vector), `cov`
# (covariance matrix) and `binmat_act` (binning matrix to be applied to a theory
# curve starting at ell=0).

# Get cl_kk, cl_tt, cl_ee, cl_te, cl_bb predictions from your Boltzmann code.
# These are the CMB lensing convergence spectra (not potential or deflection)
# as well as the TT, EE, TE, BB CMB spectra (needed for likelihood corrections)
# in uK^2 units. All of these are C_ell (not D_ell), no ell or 2pi factors.
# Then call
lnlike=apslike.generic_lnlike(data_dict,ell_kk,cl_kk,ell_cmb,cl_tt,cl_ee,cl_te,cl_bb)

Cobaya likelihood

Your Cobaya YAML or dictionary should have an entry of this form

likelihood:
  act_dr6_spt_lenslike.ACTDR6LensLike:
    lens_only: False
    stop_at_error: True
    lmax: 4000
    variant: act_baseline

No other parameters need to be set. (e.g. do not manually set like_corrections or no_like_corrections here). An example is provided in XXX.yaml. If, however, you are combining with the ACT DR4 CMB 2-point power spectrum likelihood, you should also set no_actlike_cmb_corrections: True (in addition to lens_only: True as described below). You do not need to do this if you are combining with Planck CMB 2-point power spectrum likelihoods. Similarly, SPT data do not require likelihood corrections either. For more details on likelihood corrections, see Appendix B in Qu et al 2024.

Important parameters

• variant should be
  • act_baseline for the ACT-only lensing power spectrum with the baseline multipole range
  • act_extended for the ACT-only lensing power spectrum with the extended multipole range (L<1250)
  • actplanck_baseline for the ACT+Planck lensing power spectrum with the baseline multipole range
  • actplanck_extended for the ACT+Planck lensing power spectrum with the extended multipole range (L<1250)
  • spt_3g for the SPT-only lensing power spectrum
  • actspt3g_baseline for the ACT+SPT lensing power spectrum
  • actspt3g_extended for the ACT+SPT lensing power spectrum with the extended multipole range (L<1250) for ACT
  • actplanckspt3g_baseline for the ACT+SPT+Planck lensing power spectrum with the baseline multipole range
  • actplanckspt3g_extended for the ACT+SPT+Planck lensing power spectrum with the extended multipole range (L<1250) for ACT
• lens_only should be
  • False when combining with any primary CMB measurement
  • True when not combining with any primary CMB measurement

Recommended theory accuracy

For CAMB calls, we recommend the following (or higher accuracy):

lmax: 4000 lens_margin:1250 lens_potential_accuracy: 4 AccuracyBoost:1 lSampleBoost:1 lAccuracyBoost:1 halofit_version:mead2016

If you use class-sz, the following accuracy parameters are needed for the theory code:

theory:
  classy_szfast.classy_sz.classy_sz:
    use_class_sz_fast_mode: 1
    lensing_lkl: ACT
    path: null
    stop_at_error: true
    extra_args:
      output: tCl,lCl,pCl,mPk
      skip_chi: 0
      skip_hubble: 0
      skip_sigma8_and_der: 0
      cosmo_model: 6
      N_ur: 0.00441
      N_ncdm: 1
      deg_ncdm: 3
      class_sz_verbose: 0