Skip to content

MAP Estimation

map_joint(data_set, num_steps = 30, constant_step = False) (cmb_lensing/map_joint.py) finds the joint maximum a posteriori (f, phi) given the data, following Millea et al. 2017. The objective is

\[ -2 \log P(f, \phi \mid d) = \frac{(d - M B L(\phi) f)^2}{C_n} + \frac{f^2}{C_f} + \frac{\phi^2}{C_\phi} + \text{const.} \]

Each step alternates

  1. a Wiener filter for f at the current phi, and
  2. a BFGS line search on phi along the preconditioned mixed gradient (mixed_grad_phi_logpdf), with Hessian approximation pinv(Cphi^-1 + Nphi^-1).

The G mixing matrix is set to the identity (the MAP is invariant to it).

from cmb_lensing.simulate import load_sim
from cmb_lensing.map_joint import map_joint

data_set = load_sim(nside = 128, theta_pix = 2.5, pol = "I", master_seed = 67, uk_arcmin_t = 5)
field_predict, phi_predict = map_joint(data_set, num_steps = 30)

A full worked example, with plots of the lensing, inverse lensing, logpdf and the cross spectra of the reconstruction against the truth, can be found in the map_joint tutorial notebook.