We study the impact of gas accretion on the orbital evolution of black-hole binaries initially at large separation in the band of the planned Laser Interferometer Space Antenna (LISA). We focus on two sources: (i)~stellar-origin black-hole binaries~(SOBHBs) that can migrate from the LISA band to the band of ground-based gravitational-wave observatories within weeks/months; and (ii) intermediate-mass black-hole binaries~(IMBHBs) in the LISA band only. Because of the large number of observable gravitational-wave cycles, the phase evolution of these systems needs to be modeled to great accuracy to avoid biasing the estimation of the source parameters. Accretion affects the gravitational-wave phase at negative ($-4$) post-Newtonian order, and is therefore dominant for binaries at large separations. If accretion takes place at the Eddington or at super-Eddington rate, it will leave a detectable imprint on the dynamics of SOBHBs. In optimistic astrophysical scenarios, a multiwavelength strategy with LISA and a ground-based interferometer can detect about $10$ (a few) SOBHB events for which the accretion rate can be measured at $50\%$ ($10\%$) level. In all cases the sky position can be identified within much less than $0.4\,{\rm deg}^2$ uncertainty. Likewise, accretion at $\gtrsim 10\%$ ($\gtrsim 100\%$) of the Eddington rate can be measured in IMBHBs up to redshift $z\approx 0.1$ ($z\approx 0.5$), and the position of these sources can be identified within less than $0.01\,{\rm deg}^2$ uncertainty. Altogether, a detection of SOBHBs or IMBHBs would allow for targeted searches of electromagnetic counterparts to black-hole mergers in gas-rich environments with future X-ray detectors (such as Athena) and radio observatories (such as SKA).

We consider the observation of stellar-mass black holes binaries with the Laser Interferometer Space Antenna (LISA). Preliminary results based on Fisher information matrix analyses have suggested that gravitational waves from those sources could be very sensitive to possible deviations from the theory of general relativity and from the strong equivalence principle during the low-frequency binary inspiral. We perform a full Markov Chain Monte Carlo Bayesian analysis to quantify the sensitivity of these signals to two phenomenological modifications of general relativity, namely a putative gravitational dipole emission and a non-zero mass for the graviton, properly accounting for the detector's response. Moreover, we consider a scenario where those sources could be observed also with Earth-based detectors, which should measure the coalescence time with precision better than $1 \ {\rm ms}$. This constraint on the coalescence time further improves the bounds that we can set on those phenomenological deviations from general relativity. We show that tests of dipole radiation and the graviton's mass should improve respectively by seven and half an order(s) of magnitude over current bounds. Finally, we discuss under which conditions one may claim the detection of a modification to general relativity.