The JASMINE astrometric flexible block iterative solver (JAXBIS): concept and verification with full-mission sized mock data

Abstract

The upcoming Japan Astrometry Satellite Mission for Infrared Exploration (JASMINE) in the only specifically designed to survey the Nuclear Region of the Milky Way and obtain high precision astrometry (positions, proper motions and parallaxes) of its stars with which to study its origin and formation mechanisms. The best way to guarantee that its excellent observations become scientifically meaningful measurements (kinematics and distances) is to use the observations themselves to self-calibrate the telescope, which requires: i) a flexible calibration model with millions of parameters to describe the distortions induced by the optics and detectors, and ii) a solver capable of fitting said model by processing billions of observations within a few hours. To that end, we have developed JAXBIS, an iterative Least Squares solver that splits the >1000 Petabytes Design Matrix into millions of smaller, independent problems that can be solved in parallel. Its main distinctive feature is the use of automatic differentiation which allows us to freely modify our models, quickly test new improvements and ways to calibrate the instrument, and thus vastly speeds up the search for the optimal calibration model. The implementation of JAXBIS relies on a scalable software architecture with distributed memory and massive multiprocessing tuned for large-memory workstations and CPU clusters. We tested its performance on a series of simulated datasets and proved that it can process the whole JASMINE mission in ∼15 hours. The flexibility of JAXBIS allows us to easily transfer it to other astrometric missions beyond JASMINE and thus has the potential of becoming an industry standard for high-precision astrometric surveys.