Title: Consequences of the lack of azimuthal freedom in the modeling of lensing galaxies

Abstract: Massive elliptical galaxies can display structures that deviate from a pure elliptical shape, such as a twist of the principal axis or variations in the axis ratio with galactocentric distance. Although satisfactory lens modeling is generally achieved without accounting for these azimuthal structures, the question about their impact on inferred lens parameters remains, in particular, on time delays as they are used in time-delay cosmography. This paper aims at characterizing these effects and quantifying their impact considering realistic amplitudes of the variations. We achieved this goal by creating mock lensing galaxies with morphologies based on two data sets: observational data of local elliptical galaxies, and hydrodynamical simulations of elliptical galaxies at a typical lens redshift. We then simulated images of the lensing systems with space-based data quality and modeled them in a standard way to assess the impact of a lack of azimuthal freedom in the lens model. We find that twists in lensing galaxies are easily absorbed in homoeidal lens models by a change in orientation of the lens up to 10{\deg} with respect to the reference orientation at the Einstein radius, and of the shear by up to 20{\deg} with respect to the input shear orientation. The ellipticity gradients, on the other hand, can introduce a substantial amount of shear that may impact the radial mass model and consequently bias $H_0$, up to 10 km/s/Mpc. However, we find that light is a good tracer of azimuthal structures, meaning that direct imaging should be capable of diagnosing their presence. This in turn implies that such a large bias is unlikely to be unaccounted for in standard modeling practices. Furthermore, the overall impact of twists and ellipticity gradients averages out at a population level. For the galaxy populations we considered, the cosmological inference remains unbiased.