Convergence problems with big box size (720) in RELION-3-beta-2

[Ahmed-Noor A. Agip]

Hi,

So I tried to run a 3D autorefinement job with a big box size (720720720) using RELION-3-beta-2, initially using GPU until it crashes and then switching to CPU and using the --force_converge command. The resulting map is very noisy, even with post-processing (7.4-angstrom resolution). Repeating the same refinement job with RELION-3-beta-2 using GPU (again), and then switching to RELION-2.1 for the convergence step gave me a decent map and an even better post-processed map (4.1-angstrom resolution).

The noisy map from RELION3 I refer to is the class001.mrc map, not the unfiltered half-maps, those look even worse.

I remember this was an issue in RELION-2.1-b1. My data was collected at 0.66-angstrom per pixel. Any help would be gladly appreciated.

Best wishes,

Noor

[Takanori Nakane]

@bforsbe Could this be related to the issue we have seen before with a huge box?