I have a dataset in which some exposure groups have extreme Cs values, including negative ones. I’m doing a global CTF refinement now, resetting all the values to their defaults (and Cs to 2.7mm), to see if running multiple refinements (via non-uniform) allowed the values to drift too far.
Still, it seems like the refined Cs should be constrained to be positive?
(PS resolution of this data is ~2.7Å)
Update - resetting parameters and running just a global CTF from the final refinement also gives the negative and extreme Cs values.
Thanks for this observation. We have noticed that for many datasets even at quite high resolution, estimating the fourth order aberrations can be difficult. For now, we’ll make a note to throw a warning if the spherical aberration refines to an unphysical value. Our recommendations remain to check each aberration independently to see if it benefits the resolution (most easily done via reconstruct only). In the long term, we’d like to better characterize how well these physical parameters can be estimated from the data.