[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[ccp4bb]: Re: Bug in numerical support library for refmac

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***

Dear Ethan,

> I have found a bug that causes refmac, both versions 4 and 5, to crash
> on a divide-by-zero error under some circumstances.  This bug is only
> triggered during refinement of individual anisotropic parameters Uij,
> and the specific input values that trigger it probably vary with the
> computer architecture and compiler options.  Nevertheless it is a bug,
> and I was only able to proceed with my current aniso refinement after
> tracking it down and squashing it.

Thank you. I have added this correction although I have not seen it to
cause problem

> I recommend applying the fix below to all versions of refmac, and also
> The deeper cause of the problem is that the threshold value for
> numerical convergence, THR, is becoming much too small.
> But fixing that would require more extensive code changes and I didn't
> want to fiddle with what looks to be ancient IBM library code.
> If improved performance and general architecture independence is an
> issue here, it would probably be better to replace all these support
> routines with calls to the equivalent routines in the lapack library,
> which is actively maintained and is tweaked during installation to
> match the host architecture and compiler.

I have already replaced this and some other routines with corresponding
in a version I am working on.


Garib N. Murshudov, Chemistry Department, University of York, U.K.
Tel: Home +44 (1904) 42 62 87, work:  +44 (1904) 43 25 65