Solar velocity references from 3D HD photospheric models

Context. The measurement of Doppler velocities in spectroscopic solar observations requires a reference for the local frame of rest. The rotational and radial velocities of the Earth and the rotation of the Sun introduce velocity offsets in the observations. Normally, good references for velocities are missing (e.g. telluric lines), especially in filter-based spectropolarimetric observations. Aims. We determine an absolute reference for line-of-sight velocities measured from solar observations for any heliocentric angle, calibrating the convective line shift of spatially-averaged profiles on quiet sun from a 3D hydrodynamical simulation. This method works whenever there is quiet sun in the field-of-view, and it has the advantage of being relatively insensitive to uncertainties in the atomic data. Methods. We carry out radiative transfer computations in LTE for selected C I and Fe I lines, whereas the Ca II infrared lines are synthesized in non-LTE. Radiative transfer calculations are done with a modified version of Multi, using the snapshots of a nonmagnetic 3D hydrodynamical simulation of the photosphere. Results. The resulting synthetic profiles show the expected C-shaped bisector at disk center. The degree of asymmetry and the line shifts, however, show a clear dependence on the heliocentric angle and the properties of the lines. The profiles at mu = 1 are compared with observed profiles to prove their reliability, and they are tested against errors induced by the LTE calculations, inaccuracies in the atomic data and the 3D simulation. Conclusions. Theoretical quiet-sun profiles of lines commonly used by solar observers are provided to the community. Those can be used as absolute references for line-of-sight velocities. The limb effect is produced by the projection of the 3D atmosphere along the line of sight. Non-LTE effects on Fe I lines are found to have a small impact on the convective shifts of the lines, reinforcing the usability of the LTE approximation in this case. We estimate the precision of the disk-center line shifts to be approximately 50 m s(-1), but the off-center profiles remain to be tested against observations.