The local ray tracing algorithm should be as simple as possible because it has to be run many times. Ideally, we would like to have the analytical expressions for the coordinates of ray trajectories, which would allow us to find ray-centered coordinates efficiently. However, this would be difficult if we use bilinear interpolation within grid cells. Thus, we need to make approximations. The approximation with straight rays is not appropriate for slowness models interpolated bilinearly. I expect that approximating ray trajectories with quadratic functions should meet the needs.