Project

General

Profile

Bug #246

Saving data with uneven step size in pullf.xvg file.

Added by Yong Nam almost 11 years ago. Updated almost 11 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Erik Lindahl
Category:
mdrun
Target version:
Affected version - extra info:
Affected version:
Difficulty:
uncategorized
Close

Description

I performed MD simulation with pull code.

1. First, a input .tpr file was made from .gro , .top, .mdp and .ndx files.
(grompp -f H1T3.mdp -p H1T3.top -c cin*.gro -n index.ndx -o H1T3.tpr)
(I attached these files blow.)

2. And then, MD simulation was performed for 100ns with 0.025ps of step size.
(mdrun -v -s H1T3.tpr -o H1T3.trr -x H1T3.xtc)

I saved the force in pullf.xvg file every single time step.
("pull_nstfout = 1" in H1T3.mdp file.)

However, the force was saved with uneven step size. The saved step size is bigger or smaller than 0.025ps and the err increased as simulation going forward. ( In other words, the err was relatively small in the first part of simulation, but it became pretty large in the last part of simulation.)

I have tried to change the "pull_nstfout" value from 1 to 20 and realize that if the saved step size is 5, 10, 15 or 20, the force is saved with even step size in pullf.xvg file. But, for the case of any other value, it is saved with uneven step size.

H1T3.mdp (5.51 KB) H1T3.mdp mdp file Yong Nam, 11/02/2008 08:19 PM
H1T3.top (102 Bytes) H1T3.top top file Yong Nam, 11/02/2008 08:20 PM
cin_15.21050.000_2.241.gro (445 KB) cin_15.21050.000_2.241.gro gro file Yong Nam, 11/02/2008 08:21 PM
ff_v1.4.itp (37.9 KB) ff_v1.4.itp itp file Yong Nam, 11/02/2008 08:24 PM
index.ndx (101 KB) index.ndx index file Yong Nam, 11/02/2008 08:24 PM
pullf_tail.xvg (523 Bytes) pullf_tail.xvg The last part of the force file Yong Nam, 11/05/2008 10:53 PM

History

#1 Updated by Yong Nam almost 11 years ago

Created an attachment (id=313)
mdp file

#2 Updated by Yong Nam almost 11 years ago

Created an attachment (id=314)
top file

#3 Updated by Yong Nam almost 11 years ago

Created an attachment (id=315)
gro file

#4 Updated by Yong Nam almost 11 years ago

Created an attachment (id=316)
itp file

#5 Updated by Yong Nam almost 11 years ago

Created an attachment (id=317)
index file

#6 Updated by Yong Nam almost 11 years ago

I performed MD simulation for 200ns, not 100ns.

#7 Updated by Berk Hess almost 11 years ago

When does the step size start deviating, and by how much?

I ran the simulation for a minute and everything looks fine.
Could it be that you are seeing numerical rounding errors,
because the time is stored as a float, which has quite limited
precision for long simulation times?

Berk

#8 Updated by Yong Nam almost 11 years ago

The step size starts deviating around 8ps. (Since I set the single step size as 0.025ps, 8ps corresponds with 320 simulation steps.)

The degree of deviation is pretty small at its starting point, so it might be treated as a numerical rounding errors.

However, the values of saved step size deviation becomes over 0.007ps after passing approximately 130ns(5200000 simulation steps).
I cannot treat this deviation as rounding errors because it is relatively large comparing with the single step size(0.025ps).

#9 Updated by Berk Hess almost 11 years ago

These are indeed numerical rounding errors due to the float precision.
The same problem is present in the trajectory and energy files.

For the pull output files I can fix this by determining the time
with more accuracy.
But I guess that in most cases you are not really interested
in the exact times, so you can just ignore the deviations.

Berk

#10 Updated by Berk Hess almost 11 years ago

I have looked into this issue, but it can not be solved easily,
because of lack in accuracy of the time step (delta_t),
which is stored as a float.

We will probably change all time variables to double
precision for Gromacs 4.1.

Berk

#11 Updated by Yong Nam almost 11 years ago

I am not sure if the variations of the output time-step can be explained by the
numerical error. Consider for example, the attached file 'pullf_tail.xvg' file
containing information for the last 10 time-steps of my simulations. Although at the beginning of the simulation the time step is fixed at 0.025, towards the end it oscillated between dt1 = 0.03125 and dt2 = 0.015625. The large fluctuation magnitude within one step size and the fact that dt1 = dt1*2 suggest that something other than the round-off error is involved.

I am computing autocorrelation function of the force and I would like to make sure
that the output time-steps are consistent.

#12 Updated by Yong Nam almost 11 years ago

Created an attachment (id=321)
The last part of the force file

#13 Updated by Berk Hess almost 11 years ago

That is exactly the effect of rouding.

Just ignore the times.
You get the pull output every pull_nstfout steps,
there is no doubt abut that.

Berk

Also available in: Atom PDF