Tempered Multifidelity Importance Sampling for Gravitational Wave Parameter Estimation

In an effort to accelerate gravitational wave parameter estimation, inexpensive but lower-fidelity waveform models may be used to reduce the cost of likelihood evaluations when sampling the posterior. Importance sampling can then be used to reweight these samples in order to represent the true target posterior from a higher-fidelity waveform model. However, importance sampling can be inefficient, resulting in a drastic reduction in the effective number of samples. We propose a new method of tempering the low-fidelity likelihood, allowing for potentially better overlap between the low- and high-fidelity distributions and thus better efficiency. This method is generally applicable to multifidelity importance sampling. We further propose a low-cost strategy for approximating the optimal temperature for our method. I will discuss our methods, motivate this approach by applying it to Gaussian examples, and present applications to gravitational wave parameter estimation.