Phenomenological gravitational waveform model of binary black holes incorporating horizon fluxes

Subjected to the tidal field of its companion, each component of a coalescing binary suffers a slow change in its mass (tidal heating) and spin (tidal torquing) during the inspiral and merger. For black holes, these changes are associated with their absorption of energy and angular momentum fluxes. This effect modifies the inspiral rate of the binary, and consequently, the phase and amplitude of its gravitational waveform. Numerical relativity (NR) waveforms contain these effects inherently, whereas analytical approximants for the early inspiral phase have to include them manually in the energy balance equation. In this work, we construct IMRPhenomD_Horizon, a frequency-domain gravitational waveform model that incorporates the effects of tidal heating of black holes. This is achieved by recalibrating the inspiral phase of the waveform model IMRPhenomD to incorporate the phase corrections for tidal heating. We also include corrections to the amplitude, but add them directly to the inspiral amplitude model of IMRPhenomD. First we demonstrate that the inclusion of the corrections, especially in the phase, confers an overall improvement in the phase agreement between the analytical inspiral model (uncalibrated SEOBNRv2) and NR data. The model presented here is faithful, with less than 1% mismatches against a set of hybrid waveforms (except for one outlier that barely breaches this limit). The recalibrated model shows mismatches of up to $\ensuremath{\sim}14%$ with IMRPhenomD for high mass ratios and spins. Amplitude corrections become less significant for higher mass ratios, whereas the phase corrections leave more impact---suggesting that the former is practically irrelevant for gravitational wave data analysis in Advanced LIGO (aLIGO), Virgo and KAGRA. Comparing with a set of 219 numerical relativity waveforms, we find that the median of mismatches decreases by $\ensuremath{\sim}4%$ in aLIGO zero-detuned high power noise curve, and by $\ensuremath{\sim}1.5%$ with a flat noise curve. This implies a modest but notable improvement in waveform accuracy.