Linear vs. nonlinear modeling of black hole ringdowns

The ringdown (RD) phase of gravitational waves is of prime interest for testing general relativity (GR). The modelling of the linear quasi-normal modes (QNMs) within the Kerr spectrum -- or with agnostic parameterized deviations to that GR spectrum -- has become ordinary; however, specific attention has recently emerged to calibrate the effects of nonlinear perturbations for the predominant quadrupolar $l=2$, $m=2$ mode. In this paper, we test the performance of a few nonlinear toy models and of the nonlinear inspiral-merger-ringdown (IMR) model IMRPhenomD for faithfully representing the RD regime and we compare them with the results obtained using linear solutions as sums of QNM tones. Using several quasi-circular, non-precessing numerical waveforms, we fit the dominant $l=2$, $m=2$ mode of the strain, and we assess the results in terms of both the Bayes factor and the inferred posterior distributions for the mass and spin of the final black hole (BH). We find that the nonlinear models can be comparable or preferred over the linear QNM-only solutions when the analysis is performed from the peak of the strain, especially at high signal-to-noise ratios consistent with third-generation observatories. Since the calibration of the tones' relative amplitudes and phases in high-overtone models to the progenitor parameters is still missing, or even not achievable, we consider the use of non-linear models to be more pertinent for performing confident tests of general relativity based on the RD regime starting from early times.