[rreischke]

Summary

Halo-model ingredients for the covariance between angular power spectra $C_\ell$ and the number counts of selected halos (e.g., clusters), based on Takada & Bridle (2007).

New API

• halos.HaloProfileSelection(mass_def, selection, f_off=0, r_off=0): a point-like profile of selected halos, where selection(M, a) is the selection probability. Mass–observable scatter and redshift bins enter through selection. Optional miscentering: 2D Gaussian offset with dispersion r_off·R_Δ.
• halos.Profile2ptSelection(joint_selection=None): the 1-halo two-point moment of selected halos. It returns $S$ for the same selection, and a user-supplied joint selection $S_{12}$ for different selections (e.g. 0 for disjoint bins).
• HMCalculator.I_0_3: analogous to I_1_3.
• halos.halomod_bispectrum_squeezed and halos.halomod_Pk2D_bispectrum_squeezed: the 1h + 2h squeezed bispectrum.
• halos.halomod_bispectrum_squeezed_SSC and halos.halomod_Pk2D_bispectrum_squeezed_SSC: the super-sample term, $I^1_1(N),\partial P_{uv}/\partial\delta_L$. It uses the same response as halomod_Tk3D_SSC, including the number-count counter-terms.
• angular_cl_cov_counts and angular_cl_cov_counts_SSC: the projected covariances.
  • The local part is angular_cl(B) / (4π f_sky).
  • The SSC part is a Limber integral.
  • Both take fsky and sigma2_B in the same way as angular_cl_cov_SSC.

Validation (pyccl/tests/test_pk3pt.py)

• Reproduces the plots in Takada & Bridle by eye
• Consistency: with $S = M/\bar\rho_m$, the SSC term equals the halomod_Tk3D_SSC response, both for NFW and for HOD with counter-terms.
• Output shapes, both Pk2D wrappers, $\sigma^2_B$ passed as arrays, and the error paths are also tested.
• Tested against OneCovariance

[coveralls]

Coverage Report for CI Build 40332004325

Coverage decreased (-0.04%) to 97.431%

Details

• Coverage decreased (-0.04%) from the base build.
• Patch coverage: 8 uncovered changes across 2 files (200 of 208 lines covered, 96.15%).
• No coverage regressions found.

Uncovered Changes

File	Changed	Covered	%
pyccl/halos/pk_3pt.py	115	108	93.91%
pyccl/halos/halo_model.py	10	9	90.0%
Total (7 files)	208	200	96.15%

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	7006
Covered Lines:	6826
Line Coverage:	97.43%
Coverage Strength:	4.87 hits per line

---

💛 - Coveralls

[carlosggarcia]

Question: is the test against OneCovariance in the PR? I couldn't find it. Before merging, we should add a benchmark against a external code.