Hello,
On Fri, Jul 29, 2011 at 9:58 AM, Gordon Freeman <gfreeman.wisc.edu> wrote:
> Quick question,
>
> After waging war against gcc4.1.2 (thanks to the listserve archive I
> discovered that there have been problems similar to mine reported in
> the past about 4.1.2) I compiled Amber10 with gcc version 4.4.0 and
> had considerably more luck. The tests that "possibly" failed are
> below. Previous advice on the listserve was to not worry about
> rounding errors in the very small decimal places and to copy errors of
> greater magnitude here for comments.
>
> If anyone can take a quick look and let me know if the tests reported
> in TEST_FAILURES.diff are meaningful, I would be very appreciative!
> FYI I will be using Amber10 for MD simulations of protein with both
> implicit and explicit solvent in conjunction with PLUMED (hence using
> Amber10 rather than Amber11).
>
> -The first failure (AMBERHOME/test/tip5p) seems innocuous as the test
> output file is identical to .save for the fact that rms fluctuations
> were not calculated for EKCMT, VIRIAL, VOLUME, and Density in the test
> output file (not sure why but this problem does seem to have come up
> in any other test cases).
>
This bug exists for some compilers, but not for others. I never really
looked into it, since it's not terribly important to me (and like you said,
all other values are identical).
> -The failure occuring for AMBERHOME/test/jar seems maybe a bit more
> nefarious (markedly different output in the line that differed)? But I
> do not know much about it so any comments would be helpful.
>
That error is actually completely innocent, IIRC. The only thing that
happened was that the input was changed (the endpoint of a restraint was
changed). However, this never actually affected the results. A quick
explanation about why: the restraint is a flat-well potential; that is we
have a continuous potential with a continuous first derivative (for forces,
very important), which is linear past R1, quadratic between R1 and R2, zero
between R2 and R3, quadratic between R3 and R4, and linear past R4. Thus,
you can move R1 around wherever you want (or R4 for that matter), and the
results will be unchanged *if* the system never migrates past the smallest
value that you set for R1 and R4 (since the effective restraint potential is
quadratic through the whole explored space regardless of what R1 and R4 are
set to).
In any case, if you're using PLUMED, I'm assuming you'd use PLUMED advanced
sampling options (i.e. jarzynski SMD via PLUMED rather than sander's
built-in options (i.e. ABMD, SMD, umbrella sampling, etc.)
-Otherwise all failures occur in the last decimal place which leads me
> to believe that I should not worry but assumptions like that can lead
> to trouble...
>
These are fine. What you should really watch out for are substantial errors
for features you actually want to use.
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jul 30 2011 - 14:00:03 PDT
