Hello,
On Wed, Apr 28, 2010 at 11:04 AM, g t <sketchfoot.googlemail.com> wrote:
> Hello all,
>
> I was hoping someone could help me with analyzing the setup in my
> ti_calculations. I am limited by my cluster to running jobs of 48 hours
> only. Therefore, I am using a script that splits a large job of nanoseconds
> into multiple segments of around 40 picoseconds. The script waits for a
> unique restart file to be written and submits the next segment. As a
> result,
> I get multiple .out and .crd files:
>
> e.g.
>
> hexanol_complex_prod_v0_l7_no1.out (50-90 picoseconds)
> NSTEP = 20000
> hexanol_complex_prod_v0_l7_no2.out (90-130 picoseconds)
> NSTEP = 20000
> hexanol_complex_prod_v0_l7_no3.out (130-170 picoseconds)
> NSTEP = 20000
> hexanol_complex_prod_v0_l7_no4.out (170-210 picoseconds)
> NSTEP = 20000
> hexanol_complex_prod_v0_l7_no5.out (210-250 picoseconds)
> NSTEP = 20000
>
> Rejoining the .crd files with ptraj is not a problem. However getting the
> data from the out file has been troublesome. The problem is that upon
> restarting, all the recorded numbers in the outfile are averaged for the
> number of steps of that segment only. i.e. I don't get an average that
> includes all the steps that preceded it
>
> e.g. for hexanol_complex_prod_v0_l7_no2.out:
>
>
>
> A V E R A G E S O V E R 20000 S T E P S
>
>
> NSTEP = 20000 TIME(PS) = 130.000 TEMP(K) = 299.91 PRESS =
> 5.5
> Etot = -65097.5659 EKtot = 15993.4386 EPtot =
> -81091.0045
> BOND = 480.0218 ANGLE = 1275.2825 DIHED =
> 1629.2670
> 1-4 NB = 565.9616 1-4 EEL = 7122.1569 VDWAALS =
> 9865.2479
> EELEC = -102028.9421 EHBOND = 0.0000 RESTRAINT =
> 0.0000
> DV/DL = -0.2145
> EKCMT = 7087.7220 VIRIAL = 7056.5130 VOLUME =
> 261867.6828
> Density =
> 1.0197
> Ewald error estimate: 0.7137E-04
>
>
> These numbers don't take into account the preceding 50 picoseconds. Only
> the
> time seems to have continued on from the previous segment.
>
> I have set ntpr to write every 1000 steps and have wriitten a script to
> average the dv/dl values that i am interested in. However in a test
> situation the averages calculated with the script do not match that in the
> 'A V E R A G E S O V E R 20000 S T E P S' section. Would it be accurate
>
I'm pretty sure these do not match because the AVERAGE values printed in the
mdout file is averaged over every frame in the simulation opposed to over
every ntpr frames which is all your script can do. This can be easily
verified by changing ntpr to 1 for ~10 frames and comparing your script and
the output file, then changing ntpr to 2 and checking that they differ.
> to just average all the dv/dl values in the 'A V E R A G E S O V E R
>
This will give you the proper average, but only as long as each segment that
you're adding has the same number of steps. The problem with doing this for
files with differing numbers of steps is that a file with more frames will
contribute more to the average, clearly. You can verify this pretty easily
by writing on the equation for the average in each file (Ave =
sum(data)/number points), and then extending it to each file. The total
number of points will simply end up being, in your case, 20 000 * number
files, so the final average is the sum of all the averages
(sum(data)/20,000) divided by the number of files: namely the average of the
averages (again, this only works because 20 000 is common to each
simulation).
20000 S T E P S' section? It is impossible to check manually because of the
>
random seed issue.
>
Not quite sure what you mean here... I'd say it's impossible to check
because not every frame used to compute the averages is printed to the
output file... In any case, it doesn't have any bearing on what I stated
above.
>
> Any advice on recalculation of segmented outfiles would be greatly
> apprecaited.
>
> Kind regards,
> GT
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
Good luck!
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Apr 28 2010 - 09:00:03 PDT
