Hi Dave,
This is a particularly test, I think, and a lot works together here to make
these appear to fail.
First, this is a fake test system. I just wanted a test case that would
make sure that MMPBSA.py would run through the Nmode test without failing.
Any numerical differences are going to result from system-dependent
differences in the nab functions we use (mme, xmin, and nmode). The system
calculates the "binding free energy" of trialanine with diglycine; quite
small and fictitious.
Furthermore, the starting (and intermediate) coordinate files are mdcrd,
which have severely truncated precision (which in turn are fed in as PDBs
with the same precision limitations) in order to do the nmode calculations.
Then, xmin is called which minimizes until a specific criteria is met (0.001
kcal/mol), which is fairly large, especially for such a small system. Nmode
is called after all of this is done.
So my explanation for why this test is so system-dependent is that the
minimizer detects "convergence" in a different place for each system, so the
actual structures that get run through the nmode routine differs on each
system. For instance, on my Linux box, the last minimization step looks
like this:
LS: step= 1 it= 1
MIN: It= 43 nfunc= 677 E= -72.22597 ( 5.47e-04)
----------------------------------------------------------------
FIN: :-) E= -72.23 ( 0.001)
Whereas on my mac it looks like this:
LS: step= 1 it= 1
MIN: It= 43 nfunc= 577 E= -72.22596 ( 7.66e-04)
----------------------------------------------------------------
FIN: :-) E= -72.23 ( 0.001)
When nmode is run on these, my linux box starts like this:
freq. E Cv S
cm**-1 kcal/mol cal/mol-K cal/mol-K
Total: 203.508 103.054 277.615
translational: 0.888 2.979 43.535
rotational: 0.888 2.979 39.844
vibrational: 273.958 97.097 194.237
ff energy: -72.226
1 -0.380
2 -0.152
3 -0.086
4 -0.000
5 -0.000
6 -0.000
7 0.089 0.592 1.986 17.391
8 0.123 0.592 1.986 16.744
9 0.218 0.592 1.986 15.605
Whereas my mac starts like this:
freq. E Cv S
cm**-1 kcal/mol cal/mol-K cal/mol-K
Total: 202.916 101.069 262.163
translational: 0.888 2.979 43.535
rotational: 0.888 2.979 39.844
vibrational: 273.366 95.111 178.784
ff energy: -72.226
1 -0.615
2 -0.244
3 -0.133
4 -0.086
5 -0.000
6 -0.000
7 0.000
8 0.051 0.592 1.986 18.497
9 0.224 0.592 1.986 15.548
10 0.282 0.592 1.986 15.088
So with many modes the error adds up, but IMO they're benign. In any case,
I don't think you should trust nmode results much beyond that level of
precision, anyway.
The reason (I think) that you're seeing 30+ kcal/mol differences instead of
the 5 or so that it *should* be (and even smaller for DELTAs) is that I made
a mistake in one of my commits that I have since fixed. However, the
differences that you were seeing before today will continue to be there, I
just told dacdif to be more lenient.
Another difference between systems is the math library used to diagonalize
the Hessian, which is another possible source of discrepancy.
All the best,
Jason
On Mon, Mar 28, 2011 at 12:47 PM, David A Case <case.biomaps.rutgers.edu>wrote:
> On Mon, Mar 28, 2011, Jason Swails wrote:
>
> > > mmpbsa_py/06_NAB_Nmode/FINAL_RESULTS_MMPBSA.dat
> > This one is fine. I'm actually changing the dacdif call to be less
> > stringent when it compares the files, so it'll pass next time around.
>
> Here are the diffs:
>
> 9c9
> < Vibrational: 58.9 6.1
> > Vibrational: 42.2 2.0
> 10c10
> < Total: 84.4 6.1
> > Total: 67.6 2.2
> 16c16
> < Vibrational: 20.2 0.6
> > Vibrational: 14.3 0.5
> 17c17
> < Total: 42.7 0.6
> > Total: 36.7 0.5
> 23c23
> < Vibrational: 9.0 0.9
> > Vibrational: 6.4 0.4
> 24c24
> < Total: 29.9 0.9
> > Total: 27.2 0.4
> 25c25
> < DELTA S total= 11.7 +/- 6.3
> > DELTA S total= 3.5 +/- 2.8
>
>
> This have always seems big to me...can you explain what is going on? (That
> is, why is 16 kcal/mol(?) not a diff that suggests something is wrong?
>
> ...thx...dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Mar 28 2011 - 14:30:02 PDT
