TDCOSMO XIX: Measuring stellar velocity dispersion with sub-percent accuracy for cosmography

Abstract: Stellar velocity dispersion ($\sigma$) of massive elliptical galaxies is a key ingredient to breaking the mass-sheet degeneracy and obtaining precise and accurate cosmography from gravitational time delays. The relative uncertainty on the Hubble constant H$_0$ is double the relative error on $\sigma$. Therefore, time-delay cosmography imposes much more demanding requirements on the precision and accuracy of $\sigma$ than galaxy studies. While precision can be achieved with an adequate signal-to-noise ratio (SNR), accuracy depends critically on factors such as elemental abundances and temperature of stellar templates, flux calibration, and wavelength ranges. We carry out a detailed study of the problem using multiple sets of galaxy spectra of massive elliptical galaxies with SNR$\sim$30-160 \AA$^{-1}$, and state-of-the-art empirical and semi-empirical stellar libraries and stellar population synthesis templates. We show that the choice of stellar library is generally the dominant source of residual systematic error. We propose a general recipe to mitigate and account for residual uncertainties. We show that sub-percent accuracy can be achieved on individual spectra with our data quality. Covariance between velocity dispersions measured for a sample of spectra can also be reduced to sub-percent levels. We recommend this recipe for all applications that require high precision and accurate stellar kinematics, and make all software publicly available to facilitate its implementation. This recipe will be used in future TDCOSMO collaboration papers.