Predicting Match to Optimize Numerical Relativity Template Placement

As gravitational wave detectors continue to observe an ever increasing number of merging binaries, it's crucial to have template banks which sufficiently span the parameter space. Unfortunately, numerical relativity simulations, which provide the basis for these template banks, are time consuming and computationally expensive. To mitigate this, we can fill in the parameter space more efficiently by identifying new simulations which will provide the most new information. One approach is to perform simulations which will be as different from existing simulations as possible, within the desired parameter space. A useful measure of this is the match, the noise-weighted inner product between the two waveforms. With the goal of performing new simulations in such a way as to minimize their match with existing waveforms, we train a model to predict the match that any two waveforms will have based on their initial parameters.