I included the changes i made to the tutorial during the espresso summer school to make it momentum conserving. The comparison with the reference data is now done after 1000 timesteps for the two-component version and also only done up to an error of 1e-4. The reason is that there is a difference in the velocity-calculation using the Guo-force correction, the reference implementation always uses the current force whereas this notebook uses the old force value. I might be wrong here but the version from this notebook seems more physical to me.