" Anyway, it would be much nicer if round-off errors do not affect the elapsed time."
I believe Prof. Case and the rest of the team have stated that "code modifications for improvement are always welcome".
Jim Kress
-----Original Message-----
From: Song-Ho Chong <songho.chong.gmail.com>
Sent: Friday, September 28, 2018 3:15 AM
To: AMBER Mailing List <amber.ambermd.org>
Subject: Re: [AMBER] annoying elapsed time in mdout of amber18
After reading the reply from Prof. Case and checking my simulation results again, I realized that this has nothing to do with the amber versions, but probably with the fact that, in my previous simulations I have been using "formatted" restart/coordinates files, whereas in my newer simulations, I'm using binary ones.
(I still sometimes use formatted output files since some of in-house codes directly handles restart/coordinate files and they are not updated to cope with the binary formats.)
Anyway, it would be much nicer if round-off errors do not affect the elapsed time.
Song-Ho
2018年9月28日(金) 5:36 David Case <david.case.rutgers.edu>:
> On Thu, Sep 27, 2018, Song-Ho Chong wrote:
>
> > But, for some reason, right after 336 ns production run, I see
> >
> > NSTEP = 500000 TIME(PS) = 336219.999 TEMP(K) = 309.97 PRESS =
> 0.0
> >
> > instead of TIME(PS) = 336220.000 that I expected.
>
> The printed time is obtained by repeatedly adding dt to starting time,
> and what you observe is an expected feature of floating point
> roundoff. It's not clear why you didn't see this in earlier versions
> of Amber. Scripts that process such files will need to be made aware
> of what the output might look like.
>
> It's possible that doing something like multiplying the step number by
> dt, then adding it to the starting time, would yield more
> pleasing-looking results. Suggested code revisions are welcome.
>
> .....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 28 2018 - 09:30:03 PDT
