Optimisation of the COMPAS rapid binary population synthesis code

CI: – Dr. I. Mandel

We are requesting ADACS software support to enable the optimisation of the binary evolution component of the COMPAS code [about 30,000 lines of C++ code] for high-performance computing.

COMPAS (Compact Object Mergers: Population Astrophysics & Statistics) is a software toolbox for exploring the evolution of binary stars and confronting model predictions with observations. COMPAS was originally developed to enable the interpretation of gravitational-wave observations, and the bulk of the 15 papers based on COMPAS written during the past 3 years deal with this topic; however, we have also incorporated other observational constraints that can illuminate massive binary evolution, including luminous red novae, X-ray binaries, and radio pulsars (see http://compas.science for a full list).

The COMPAS code is public: https://github.com/TeamCOMPAS/COMPAS. It is being used and developed internationally: in addition to a core Australian group at Monash and Swinburne, users and contributing developers are based at Harvard (USA), AEI (Germany), U of Amsterdam (Netherlands), Nils Bohr Institute (Denmark), etc. We are ready to deploy it on the rapidly growing gravitational-wave data set. We showed (Barrett et al., 2018) that such data will enable us to accurately constrain wind mass loss rates, supernova natal kicks, and mass transfer physics. However, because of the high dimensionality of this parameter space, as well as the need to integrate over a broad range of star-formation metallicities and the relatively low yield of binaries of interest (low fraction of merging double compact objects among all massive binaries), we estimate that we may need to simulate 10 to 100 billion systems. Even if each binary requires 1 second to simulate, this translates to a total computational cost of order 10 million CPU hours. Therefore, a modest investment in code optimisation is both necessary and economical.

While our team has broad expertise in theoretical astrophysics, and some experience in software development, we lack the specific knowledge about high-performance computing optimisation. The optimisation expertise of the ADACS team, which we do not have available in-house, would enable us to provide a truly efficient code that will allow both ourselves and the broader community to capitalise on the growing data sets of gravitational-wave sources and other transients associated with massive binaries.