Abstract:
|
Computer experiments are often used to simulate complex physical systems to gain insights of their behavior. In our work, we aim to investigate the formation of binary black holes by means of computer models and explore the initial conditions and physical parameters governing the models. Knowledge of the physical parameters can greatly improve the efficiency of the simulator (the success rate of producing binary black holes, characterized by a "chirp mass", is quite low). A first stage of our study is to construct an emulator of the computer model. A unique feature of this setting is that, under certain initial conditions, binary black holes do not form. Thus, no chirp mass is observed. In this a talk, methodology for emulating computer models where outputs are limited to only a subset of the input space is presented. Our approach combines a Gaussian process classifier and local Gaussian process models to emulate the simulator and provide estimated of uncertainty.
|