Comparison of the observational data and the empirical orbit superposed model: the stellar density distribution. The top panels from left to right are the 2D density distribution ρ_star in R_gc versus z_gc constructed by data, our best-fitting model (q_DM = 0.92, p_DM = 0.80), a model with a flatter DM halo, and a model with a rounder DM halo. The bottom panels show the relative error of the data, and the corresponding model residuals. The inset panel in the right shows the flattening of the stellar halo, q_star ≡ z_gc/R_gc, as a function of the radius, R_gc, extracted from the iso-density contours of the 2D density distribution. The star symbols with the error bar are data. The solid red, light blue, and black curves indicate the three models shown on the left. The dashed red curve indicates a model allowing q_DM and p_DM to vary as a function of radius. The best-fitting model constructed stellar density distribution that matches the observational data well, while models with flatter or rounder DM halos constructed stellar halos that are either too flat or too round. The density distribution of the halo stars, especially the flattening q_star, has strong constraints on the shape of the underlying DM halo.