Title: Unveiling early black hole growth with multi-frequency gravitational wave observations

Abstract: Third Generation ground based Gravitational Wave Interferometers, like the Einstein Telescope (ET), Cosmic Explorer (CE), and the Laser Interferometer Space Antenna (LISA) will detectcoalescing binary black holes over a wide mass spectrum and across all cosmic epochs. We track the cosmological growth of the earliest light and heavy seeds that swiftly transit into the supermassive domain using a semi analytical model for the formation of quasars at $z=6.4$, 2 and $0.2$, in which we follow black hole coalescences driven by triple interactions. We find that light seed binaries of several $10^2$ M$_\odot$ are accessible to ET with a signal-to-noise ratio ($S/N$) of $10-20$ at $6<z<15$. They then enter the LISA domain with larger $S/N$ as they grow toa few $10^4$ M$_\odot$. Detecting their gravitational signal would provide first time evidence that light seeds form, grow and dynamically pair during galaxy mergers. The electromagnetic emission of accreting black holes of similar mass and redshift is too faint to be detected even for the deepest future facilities. ET will be our only chance to discover light seeds forming at cosmicdawn. At $2<z<8$, we predict a population of "starved binaries", long-lived marginally-growing light seed pairs, to be loud sources in the ET bandwidth ($S/N>20$). Mergers involving heavy seeds ($\sim 10^5 M_\odot - 10^6 M_\odot$) would be within reach up to $z=20$ in the LISA frequency domain. The lower-z model predicts $11.25(18.7)$ ET(LISA) events per year, overall.