Non uniform improvements in local refinement?


The new alterations in non uniform refinement in 3.0 are fantastic, particularly for smaller molecules. I’ve already seen significant improvements on test cases.

It seems like these alterations would really help for local refinement, where one is often dealing with very small masks. In addition, the improved memory efficiency that allows for 1024px boxes on a 12GB card would be very helpful when performing local refinement on smaller regions of larger complexes.

Is there any chance of bringing these improvements to the local refinement job in the near future?


Hi @olibclarke,

This is certainly a planned addition, and is actively being worked on. Experimentation thus far has shown us improved results on local refinements using the new features such as adaptive marginalization and the new method non-uniform regularization, but there is some tuning needed. There are also other additional features that have been requested and that we hope to bring alongside this, such as enforced and relaxed symmetry.


1 Like

Sounds good, if you need any alpha/beta testers let me know! :joy:


So all these improvements are now available (and very useful!), except for symmetry relaxation… is symmetry relaxation for either NU-refine or local refinement on the roadmap? would be great to have for pseudosymmetric structures!

Hey @olibclarke,

Yes I’d really like to add this :slight_smile: and it also seems like it would be a reasonably straightforward addition. I think something like the symmetry relaxation feature described here is best?

A feature like symmetry relaxation could definitely be useful with pseudosymmetric structures, I get the sense that adding classification w/o alignment would also help in this workflow too, which we also hope to add soon.


1 Like

Yes exactly - and classification without alignment would be helpful in pretty much any workflow, symmetry or no