Maximum Likelihood Maps Of
Complete Single Frequency Planck Data
Christopher Cantalupo, Julian Borrill & Radek Stompor,
Computational Research Division, Lawrence Berkeley National Laboratory
& Space Sciences Laboratory, UC Berkeley
This work is part of an ongoing collaboration with colleagues in the US Planck Data Center and the Planck CTP Working Group. The input data were simulated using the reference sky model being developed by Planck Working Group 2 and software provided by the Planck Level S team. The output data were post-processed using the HEALPix package.
This project in particular was made possible by the exceptional support offered by colleagues in the NERSC Center's Systems Support and User Services Groups, and the Berkeley Lab Computational Research Division's Scientific Computing & Visualization Groups.
The Planck satellite will gather an unprecedented volume of Cosmic Microwave Background (CMB) temperature and polarization data whose analysis will present a major computational challenge. Temporal correlations in the detector noise mean that maximum likelihood (generalized least squares) methods must be used to obtain the highest fidelity CMB sky maps. Here we have used MADmap - a massively parallel implementation of the preconditioned conjugate gradient solution to maximum likelihood map-making - to make the first optimal, full-resolution, I, Q & U maps from 1 year of simulated data from all of the detectors at one of Planck's most CMB-sensitive frequencies. This calculation (mapping 75 billion observations to 150 million pixels) used 6000 processors of NERSC's Seaborg supercomputer for 2 hours, demonstrating the practicality of processing such data volumes by these methods. Scaling to this concurrency required breaking significant MPI and I/O bottlenecks, but these results show that continued access to state-of-the-art supercomputers, and the development of codes that can take advantage of their full capabilities, will be critical for maximizing the scientific return of the Planck mission.
Full Planck Resolution Single Frequency 1 Year CMB Intensity Map
The CMB signals for these simulations were generated by matching spherical harmonics up to l = 70 to the best WMAP map, and up to l = 3000 to the concordance model spectrum. These images compare the Planck CMB intensity map to its WMAP equivalent, showing both the WMAP full sky at the same number of map pixels per unit image area, and a common 100 square degree low-foreground patch of the sky from each map at the same number of sky square degrees per unit image area.
This analysis also generated polarization maps, and all three Stokes parameter maps can be compared with their corresponding input. To increase the signal to noise in these maps their resolution has been reduced by a factor of 8; even at such reduced resolution the Q and U maps are still noise-dominated.
Planck is a mission of the European Space Agency and the US National Aeronautics & Space Administration. This work is supported by NASA Planck funding under JPL Subcontract No. 1255574, and by the NASA Advanced Information Systems Research Program under Award No. S-92548-F.