A comparison between best-fit eccentricity definitions and the standardized definition of eccentricity

Abstract: In the absence of a unique, gauge-independent definition of eccentricity in General Relativity, there have been efforts to standardize the definition for Gravitational-Wave astronomy. Recently, Shaikh et al. proposed a model-independent measurement of eccentricity $e_{\mathrm{gw}}$ from the phase evolution of the dominant mode. Many works use loss functions (LFs) to assign eccentricity to a reference waveform, for instance by fitting a Post-Newtonian expression to assign eccentricity to Numerical Relativity (NR) simulations. Therefore, we ask whether minimizing common LFs on gauge-dependent model parameters, such as the mismatch $\mathcal{M}$ or the $L_2$-norm of the dominant mode $h_{22}$ residuals, for non-precessing binaries, ensures a sufficient $e_{\mathrm{gw}}$ agreement. We use $10$ eccentric NR simulations and the eccentric waveform TEOBResumS-Dali as the parametric model to fit on eccentricity $e_0$ and reference frequency $f_0$. We first show that a minimized mismatch, the $\mathcal{M} \sim 10^{-3}- 10^{-2}$ results in better $e_{\mathrm{gw}}$ fractional differences ($\sim 1\%$) than with the minimized $h_{22}$ residuals. Nonetheless, for small eccentricity NR simulations $(e_{\mathrm{gw}} \lesssim 10^{-2}$), the mismatch can favor quasi-circular ($e_0=0$) best-fit models. Thus, with sufficiently long NR simulations, we can include $e_{\mathrm{gw}}$ in the LF. We explain why solely fitting with $e_{\mathrm{gw}}$ constitutes a degenerate problem. To circumvent these limitations, we propose to minimize a convex sum of $\mathcal{M}$ and the $e_{\mathrm{gw}}$ difference to both assign non-zero eccentric values to NR strains and to control the mismatch threshold.