pysanepic

GLS map-making with correlated (1/f) noise for CMB experiments, in Python.

pysanepic is a Python port of SANEPIC (Patanchon et al. 2008, ApJ 681, 708). It solves the generalized least-squares problem (PᵀN⁻¹P) m = PᵀN⁻¹d with preconditioned conjugate gradients, applying N⁻¹ in Fourier space on chunks of data, and returns I, Q, U HEALPix maps.

• Designed for experiments without HWP, where 1/f noise affects both intensity and polarization. An ideal HWP is supported too.
• Pure Python with numba and ducc0; runs with MPI (mpi4py) and threads.
• Supports polarization efficiency and any coordinate system.
• Noise correlated between detectors (common modes of a wafer), with a two-level preconditioner for the large scales (GLSParameters.coarse_nside; by default on the I monopole alone, the slowest mode).
• Noise spectra, couplings and common modes estimated from the data (estimate_noise).
• Gain calibration on the dipole, fitted with the maps (CalibrationParameters), with masks for the Galactic plane.
• Sky model as alm interpolated at the pointing (no pixelization error) and bandpass templates fitted per detector (make_maps(sky_model=..., bandpass_templates=...)).
• Independent of any simulation pipeline: data are passed as plain arrays.
• Validated against the original C++ code, which is kept in cpp/.

Documentation: https://pysanepic.readthedocs.io (sources in docs/).

Installation

pip install pysanepic
pip install "pysanepic[mpi]"                # with MPI
pip install git+https://github.com/litebird/pysanepic   # development version

Usage

With litebird_sim (on master, not yet in a release):

from pysanepic import GLSParameters

result = sim.make_sanepic_gls_map(nside=512, params=GLSParameters(chunk_s=86400.0, tol=1e-8))
result.maps       # (3, npix) I, Q, U; healpy.UNSEEN where not solved

With any other pipeline, one DetectorData per detector and stretch of time:

from pysanepic import DetectorData, OneOverFNoise, Pointing, make_maps

# angles in any coordinate system, or Pointing(pix=..., nside=..., psi=..., coordinates=...)
pointing = Pointing(theta=theta, phi=phi, psi=psi, coordinates="E")
noise = OneOverFNoise(net_ukrts=40.0, fknee_hz=0.02, alpha=1.0, fmin_hz=1e-5)  # or TabulatedNoise(freqs_hz=..., psd=...)
data = [DetectorData(tod=tod, pointing=pointing, pol_angle_rad=0.0, sampling_rate_hz=19.0, noise=noise)]
result = make_maps(data, nside=64, coordinates="G")

See the documentation for the conventions, the parameters and how to choose them, MPI, memory and benchmarks, and the notebooks.

Tests

pip install -e ".[test]"
make -C cpp ARCH=linux        # or ARCH=mac: builds the C++ reference
pytest                        # includes the comparison with C++ SANEPIC
mpirun -n 4 python tests/test_mpi.py

Documentation

pip install -e ".[docs]"
sphinx-build -b html docs docs/_build/html

License

Distributed under the GPL3 License.