Re: Pressure Discrepancy

From: Sterling Paramore (
Date: Fri Apr 20 2007 - 11:18:39 CDT

First of all, you need to expand the autocorrelation graph so that you
can see more precicely how fast it goes to zero (I can't tell from the
graph). Try looking at the range [0:100ps]. Second, according to my
analytical chemistry book (Harris), the student's t for the 95% CI when
you have more than 200 measurements is 1.96, not 2.96. If your mean Pzz
is 5 and the 95% CI really is 10, then it would suggest that the
measured Pzz is statistically indistinguishible from 1 bar and
everything is ok with your simulations. Third, the barostat used in
NAMD works by changing the simulation cell dimensions in a way that the
average pressure hits the target. You have constrained the area to be
constant, so the barostatting mechanism can only work by changing the
simulation cell along the z dimension. If the system was isotropic,
then that would be enough to give the same pressure in every direction.
But since it's a membrane, it is anisotropic and you shouldn't expect
the pressure in the x and y directions to be the same as the z. If you
really want the pressure in the x and y directions to be 1 bar, then you
need to let the area fluctuate as well. However, that all depends on
what you want to do with these simulations.


Morad Alawneh wrote:

> *Dear Sterling,
> I reported the SE = SD/sqrt(N), so to get 95% CI, we need to multiply
> the SE by 2.96. By using 17 as the 95% CI, the 1 bar value for Pzz is
> within the range, which is I think what you are looking for. Regarding
> the autocorrelation function, see the attachment, it looks the same
> as the one for the total pressure. It shows how fast the no
> correlation for Pzz. Just to let you know I have printed out the data
> every 1 ps.
> Within the statistical error Pzz is fine, but the total pressure is
> not, even I set the pressure to be 1 atm, and that what makes me
> confused. You said it is not necessary to expect the total pressure be
> the same as the target value, will you explain that to me?
> **Unless the target value meant to be for Pzz, then every thing make
> sense.*
> *
> Thanks
> Morad Alawneh
> *
> Sterling Paramore wrote:
>> What did the decorrelation time for Pzz end up being in your system?
>> And what confidence interval are you reporting the standard error at?
>> Also, you shouldn't necessarily expect the average total pressure to
>> hit 1 bar exactly (only the zz component) since you used constant area.
>> -Sterling
>> On Apr 19, 2007, at 3:05 PM, Morad Alawneh wrote:
>>> *Dear all,
>>> I have here new results to show the pressure problem:
>>> <Pzz> (bar) = 10.93 (+/- 5.75 SE) (+/- 575.24 SD)
>>> <P> (bar) = -33.18 (+/- 3.61 SE) (+/- 361.23 SD)
>>> <V> (A^-3) = 119800.89 (+/- 4.73 SE) (+/- 473.10 SD)
>>> The results show stable average values but away from the target
>>> pressure, which I conclude something is not right about pressure and
>>> any quantity related to pressure should not be trusted until this
>>> problem be figured out.
>>> *

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:44:35 CST