I noticed this kind of thing myself a long time ago, and wondered what
refmac was doing to make things "worse", so I let it keep going. And
going and going. I was delighted to "discover" that although R and/or
Rfree could rise over up to hundreds of cycles, it almost invariably
turns around again, usually getting much better than that "blip" I saw
after a measly 2-3 cycles. Leaving me to wonder why I was so greedy in
the first place. Then I got even more greedy and let it keep running.
There can be several hills and valleys when you do this, usually
corresponding to larger-scale "wiggle" motions of backbone and the like,
or even side chains like lysine flipping rotamers. It is perhaps
understandable that these don't happen right away. After all, you are
performing a search in a ~10,000-dimensional space. But eventually you
get to the point where refmac no longer moves any of the atoms and the
input and output files are essentially identical. (rmsd < 0.002 A or
so). This is what I call "convergence", and this seems to be one of
those rare occasions where Ian and I are in total agreement!

After all, if you think about it, if you stop refinement just before R
and Rfree go on a "significant" upswing, how are you supposed to
interpret a "significant" upswing in your next round of refinement? Is
it due to the rebuild you just did? Or is it something far away from the
atoms you moved that was about to "wiggle"? The only way to really know
is to let the whole thing settle down first. Then you can be sure that
the subsequent change in R is due to something you just did, and not
something you did a hundred cycles ago.

Oh, and BTW, as far as I know, it is still a good idea to let refmac
itself run for only ~5 cycles and just keep re-cycling the input file.
In the old days, you wanted to do this because after the atoms moved a
bit you had to re-run "PROTIN" to re-determine the restraints. If you
didn't, you could get what I called "water cannons" where a
poorly-modeled water would fly unrestrained through the unit cell,
knocking side chains out of their density. Admittedly, this was a long
time ago, and I'm sure it has been fixed by now. Garib can correct me if
I'm wrong about that.

Anyway, to answer the OP's question, I saw let refmac, phenix.refine, 
shelx or whatever refinement program you are using do it's thing. You 
wouldn't inject your protein until the column had a stable baseline 
would you?

-James Holton
MAD Scientist


On Aug 24, 2011, at 5:24 PM, Tim Gruene <[log in to unmask]> wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Dear all,
>
> especially at the beginning of model building and/or at low resolution
> both Rfree and "-LL free" as reported in the refmac logfile show a
> minimum at a some cycle before rising again.
>
> I am certainly not the only one tempted to first run refmac with a large
> number of refinement cycles, determine that minimum and rerun refmac
> with ncyc set to that minimum.
>
> Of course I want the resulting model and phases/map to be as close to
> the what's in the crystal as possible in order to facilitate model building.
>
> Is it therefore good practice to interrupt refmac wherever it finds a
> minimum (if so, the minimum w.r.t. which number reported in the log-file)?
>
> Thanks for everyone's opinion and experience,
>
> Tim
>
> - --
> Dr Tim Gruene
> Institut fuer anorganische Chemie
> Tammannstr. 4
> D-37077 Goettingen
>
> GPG Key ID = A46BEE1A
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.11 (GNU/Linux)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
>
> iD8DBQFOVReTUxlJ7aRr7hoRAqyzAKCZpMJPVSQJTDEoWGxZEymwvqfFcACeMNLL
> rvIDPlXiL5HQmoNm7yrTt6k=
> =UnKT
> -----END PGP SIGNATURE-----