Eccentric surrogate waveform model for non-spinning unequal mass binary black holes

Recent interest in modeling gravitational waves (GWs) from eccentric binary black hole (BBH) systems arises from formation channels that predict detectable binaries with measurable eccentricity entering the LIGO sensitivity band. While numerical relativity (NR) accurately evolves BBH systems to generate emitted GW signals, its high computational cost limits direct use for template waveform generation. Surrogate modeling offers a fast and accurate data-driven method to replicate NR waveforms. We present the first 3-dimensional NR surrogate model for waveforms from eccentric BBH systems that retain eccentricity to merger, with mass ratios from 1 to 4 eccentricity less than 0.4 as measured at a reference time about 20 orbits before merger. Waveform amplitudes and phases from eccentric BBHs are typically oscillatory on the time-scale of the orbital period, which significantly complicates the construction of highly-accurate surrogates. We compare modeling efficacy for different data pieces derived from eccentric waveforms that imprint orbital timescale oscillations in the wave's amplitude and phase evolution, separately for the inspiral and merger-ringdown parts of the waveform. We also explore physically motivated ideas like using eccentricities and mean anomalies as functions of time to improve the surrogates's accuracy. Trained directly against 104 NR simulations, the model can reproduce the simulations nearly as accurately as the simulations themselves.