AMBER Archive (2005)

Subject: Re: AMBER: test suite possible failures

From: David A. Case (case_at_scripps.edu)
Date: Mon Jan 24 2005 - 16:38:21 CST


On Mon, Jan 24, 2005, Stern, Julie wrote:

> I compiles the serial linux version and got a list of
> possible failures. The programs producing those possible
> failures are listed. But, in each of the *.dif files the results
> are only off in the 3rd decimal place. However, since this
> is still much greater than machine precision we are concerned.
>
>
> ./pheTI/out.p1.dif
> ./antechamber/tp/prmtop.mol2.dif
> ./antechamber/tp/prmtop.dif
> ./antechamber/ash/ash.prepin.dif
> ./antechamber/ash/prmtop.dif
> ./antechamber/guanine/prmtop.dif
> ./antechamber/sustiva/sustiva.prepin.dif
> ./antechamber/sustiva/prmtop.dif
> ./antechamber/fluorescein/fluorescein.prepin.dif
> ./antechamber/fluorescein/prmtop.dif
> ./antechamber/guanine_amber/DGN.prepin.dif

Antechamber differences between machines are not uncommon, since slight
differences between computers/compilers/OSes can affect the self-consistent
iterations, and hence lead to charges that are slightly different (0.01
electron, even more sometimes). But you should examine the outputs
carefully: it's not always possible to discriminate important vs. un-important
diffs just from a list of files.

The out.p1.dif file is often a dif where one machine will give an rms
flucutation of 10**-14 or so, vs. zero. I've never tracked this down,
since it look innocuous; but I also don't know if this is what you are seeing
or not.

....dac

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu