Three-dimensional surface convection simulations of metal-poor stars -- The effect of scattering on the photospheric temperature stratification
Three-dimensional (3D) radiative hydrodynamic model atmospheres of metal-poor late-type stars are characterized by cooler upper photospheric layers than their 1D counterparts. This property of 3D models can dramatically affect elemental abundances derived from temperature-sensitive spectral lines. We investigate whether the cool surface temperatures predicted by metal-poor 3D models can be ascribed to the approximated treatment of scattering in the radiative transfer. We use the Bifrost code to test three different ways to handle scattering in 3D model atmospheres of metal-poor stars. First, we solve self-consistently the radiative transfer equation for a source function with a coherent scattering term. Second, we solve the radiative transfer equation for a Planckian source function, neglecting the contribution of continuum scattering to extinction in the optically thin layers; this has been the default mode in previous models of ours. Third, we treat scattering as pure absorption everywhere, which is the standard case in CO5BOLD models. We find that the second approach produces temperature structures with cool upper photospheric layers very similar to the correct coherent scattering solution. In contrast, treating scattering as pure absorption leads to significantly hotter and shallower temperature stratifications. The main differences in temperature structure between our published models and those generated with the CO5BOLD code can be traced to the different treatments of scattering. Neglecting the contribution of continuum scattering to extinction in optically thin layers provides a good approximation to the full radiative transfer solution for metal-poor stars. Our results demonstrate that the cool temperature stratifications predicted for metal-poor late-type stellar atmospheres by previous models of ours are not an artifact of the approximated treatment of scattering.