A 1.9 solar-mass neutron star candidate in a 2-year orbit

We report discovery and characterization of a main-sequence G star orbiting a dark object with mass $1.90\pm 0.04\,M_{\odot}$. The system was discovered via Gaia astrometry and has an orbital period of 731 days. We obtained multi-epoch RV follow-up over a period of 639 days, allowing us to refine the Gaia orbital solution and precisely constrain the masses of both components. The luminous star is a $\gtrsim 12$\,Gyr-old, low-metallicity halo star near the main-sequence turnoff ($T_{\rm eff} \approx 6000$\,K; $\log\left(g/\left[{\rm cm\,s^{-2}}\right]\right)\approx 4.0$; $\rm [Fe/H]\approx-1.25$; $M\approx0.79\,M_{\odot}$) with a highly enhanced lithium abundance. The RV mass function sets a minimum companion mass for an edge-on orbit of $M_2 > 1.67\,M_{\odot}$, well above the Chandrasekhar limit. The Gaia inclination constraint, $i=68.7\pm 1.4$\,deg, then implies a companion mass of $M_2= 1.90\pm 0.04\,M_{\odot}$. The companion is most likely a massive neutron star: the only viable alternative is two massive white dwarfs in a close binary, but this scenario is disfavored on evolutionary grounds. The system's low eccentricity ($e=0.122\pm 0.002$) disfavors dynamical formation channels and implies that the neutron star likely formed with little mass loss ($\lesssim 1\,M_{\odot}$) and with a weak natal kick ($v_{\rm kick}\lesssim 20\,\rm km\,s^{-1}$). Stronger kicks with more mass loss are not fully ruled out but would imply that a larger population of similar systems with higher eccentricities should exist. The current orbit is too small to have accommodated the neutron star progenitor as a red supergiant or super-AGB star. The simplest formation scenario -- isolated binary evolution -- requires the system to have survived unstable mass transfer and common envelope evolution with a donor-to-accretor mass ratio $>10$. The system, which we call Gaia NS1, is likely a progenitor of symbiotic X-ray binaries and long-period millisecond pulsars. Its discovery challenges binary evolution models and bodes well for Gaia's census of compact objects in wide binaries.