# Re: Pressure Discrepancy

From: Sterling Paramore (paramore_at_hec.utah.edu)
Date: Wed Mar 28 2007 - 22:19:36 CDT

But Morad's running in NPT, right? The algorithm is designed so that
the volume is automatically adjusted to reproduce the desired
pressure, on average. It could very well be the piston decay time
giving rise to a very large autocorrelation time (so getting the
average to be 1atm would require an enormously long simulation
time). But before you go and run another long simulation, I really
suggest that you do a quick calculation to determine the pressure
autocorrelation function.... that'll tell you a lot.

-Sterling

On Mar 28, 2007, at 5:23 PM, Richard Wood wrote:

> Hi all,
>
> I still say his box is too big, which is why his presure is too
> large (and negative!).
>
> Here are some numbers from some MD simulations I did in which I
> varied the sides of a cube of water (512 water molecules) and
> calculated the average pressure, the goal of which was to find a
> box length that gave me an average pressure of 1 atm. If I plot
> the average pressure vs. box length, I get a straight line y =
> -2171.8x + 53499, so an average pressure of 1 atm would correspond
> to my water box being 24.633 A on a side.
>
>
> volume, Å3 box length, Å Average pressure, atm.
> 14706.125 24.5 286.0323
> 14886.936 24.6 63.11659
> 15069.223 24.7 -155.146
> 15160.92188 24.75 -229.526
> 15286.41305 24.8181 -401.745
> 15366.93394 24.8616 -487.346
>
> He needs to decrease his box size, since his average pressure is
> negative (see my data).
>
> Richard
>
> Richard L. Wood, Ph. D.
> Computational Chemist
> Cockeysville, MD 21030
> rwoodphd_at_yahoo.com
>
>
> ----- Original Message ----
> From: Sterling Paramore <paramore_at_hec.utah.edu>
> Cc: namd-l <namd-l_at_ks.uiuc.edu>
> Sent: Wednesday, March 28, 2007 5:12:06 PM
> Subject: Re: namd-l: Pressure Discrepancy
>
> Try calculating a pressure autocorrelation time. From this, you
> should
> be able to estimate how many independent pressure observations you
> over the course of your 10ns simulation. Using that number, you can
> calculate the standard deviation of the mean pressure. If the
> standard
> deviation of the mean is larger than the average, you don't have
> enough
> sampling; if not, then you've got another problem.
>
> -Sterling
>
>
> > *The simulation was for 10 ns with the following SD:
> >
> > Pxx Pyy
> > Pzz Pt P <P>
> > average -105.7232 -119.0253 -16.4359 -112.3742
> > -80.3948 -54.0773
> > SD 516.3578 520.2953 578.8550 395.4912
> > 362.1872 106.7834
> >
> > Thanks.
> >