Fast and efficient evaluation of gravitational waveforms via reduced-order spline interpolation

Numerical simulations of merging black hole binaries produce the most accurate gravitational waveforms. The availability of hundreds of these numerical relativity (NR) waveforms, often containing many higher spherical harmonic modes, allows one to study many aspects of gravitational waves. Amongst these are the response of data analysis pipelines, the calibration of semi-analytical models, the building of reduced-order surrogates, the estimation of the parameters of detected gravitational waves, and the composition of public catalogs of NR waveform data. The large number of generated NR waveforms consequently requires efficient data storage and handling, especially since many more waveforms will be generated at an increased rate in the forthcoming years. In addition, gravitational wave data analyses often require the NR waveforms to be interpolated and uniformly resampled at high sampling rates. Previously, this resulted in very large data files (up to $\sim$ several GB) in memory-intensive operations, which is unfeasible when confronted with hundreds of multi-modal NR waveforms. To handle these challenges, we present a simple and efficient method to significantly \emph{compress} the original waveform data sets while accurately reproducing the original data via spline interpolation. The method is generically applicable to relatively smooth, one-dimensional datasets and uses a greedy algorithm to determine the most relevant subset of the full data such that a spline interpolant of a specified polynomial degree will represent the original data to within a requested point-wise tolerance. We find significant compression of the original NR data sets presented here. These compressed data sets can then be evaluated fast and efficiently and resampled as desired.