Big Bang from Big Data (of the cosmic microwave background) – B3DCMB

This project is in the area of data analysis of cosmological data sets as collected by contemporary and forthcoming observatories. This is one of the most dynamic areas of modern cosmology. Our specific target are data sets of Cosmic Microwave Background (CMB) anisotropies, measurements of which have been one of the most fruitful of cosmological probes. CMB photons are remnants of the very early evolution of the Universe and carry information about its physical state at the time when the Universe was much younger, hotter and denser, and simpler to model mathematically. The CMB has been, and continue to be, a unique source of information for modern cosmology and fundamental physics. In particular, polarization of the CMB photons is expected to carry a unique signature from the time when the Universe, as we know it, was born. A detection of this signature, called primordial B-mode polarization, would be truly a ground-breaking achievement for cosmology and fundamental physics.

CMB B-mode polarization is the current frontier in cosmology. Consequently, there is a number of high profile experimental efforts aiming at its discovery, which follow on over 20 years of technological developments. Ultimately, the information will have to be extracted from huge and complex data sets those experiments are, or will be, collecting. This will require advanced numerical and statistical techniques from far-beyond the current standard toolbox of CMB data analysts. Indeed, it is well-recognized that a true leap in the efficiency, computational proficiency, and statistical robustness of the CMB data analysis methods is necessary in order to ensure that this next generation of the data sets is properly exploited and the cosmological objectives posed for the CMB field reached.

The main objective of this project is to empower the CMB data analysis with novel high performance tools and algorithms superior to those available today and which are capable of overcoming the existing performance gap. The CMB data analysis is a complex process. While the CMB data sets with projects volumes in excess of many Petabytes may not be the largest even by the cosmological standards, the signals of interest are tiny and distributed over all measurements which are dominated by instrumental noise and, often, non-cosmological signals, all possessing strong correlations best characterized in different domains. This precludes 'divide and conquer' approaches, instead, sophisticated statistical techniques, capable of harnessing efficiently the computational power of the modern, massively parallel supercomputers, need to be employed to allow processing an entire data set at once. CMB data analysis is a genuinely, multi-disciplinary endeavor involving expertise in high performance scientific supercomputing, statistical methods and physical insights.

This project is based on a few key observations. First, in spite of its apparent complexity and diversity of physical contexts, the CMB data analysis comes down to a relatively limited number of algebraic problems. The two most important ones are structured linear systems of equation of so-called map-making or Wiener-filter types. Second, these linear problems solvers need to be optimized and devised for a specific statistical method used in the analysis. Likewise, the most suitable statistical method may be the one for which the solvers can be best optimized. In this proposal, we will adapt an integrated outlook and considered together efficient statistical methods and appropriate numerical solvers. Third, the solvers need to be parallelizable and their implementations scalable up to hundreds of thousands of cores and adaptable to the hierarchical and heterogeneous architectures of modern supercomputers. These insights define the main challenges which need to be addressed to meet this project's objectives and which we propose to overcome with help of an interdisciplinary, closely working team of experts.