Keywords: Computer simulator; emulation; model calibration
Computer simulators are often used to study complex physical systems. We aim to investigate the formation of binary black holes by means of a computer model, with inputs that consist of both initial conditions and physical parameters to be inferred. Forming a statistical surrogate, or emulator, for the simulator of the "chirp mass" is particularly challenging in this case because the response is zero when a binary black hole fails to form. In order to find regions in the input space that lead to non-zero outputs, a local GP classifier is employed The classifier is then combined with a local Gaussian process model to construct an emulator of the physics model and subsequently used for computer model calibration.