Bug #137

g_dielectric segmentation fault

Added by George Patargias over 13 years ago. Updated over 13 years ago.

analysis tools
Target version:
Affected version - extra info:
Affected version:


I calculated ACF of the total dipole moment of a box of SPC water molecules
(PME, 300K) by issuing:
g_dipoles -s prod_in.tpr -f prod_out.xtc -corr total

Then I used the resulting dipcorr.xvg file as an input to g_dielectric which I
run as follows:
g_dielectric -f dipcorr.xvg -epsRF 0 -ffn aexp

The meessage I got was the following:

Read data set containing 2 colums and 1501 rows
Assuming (from data) that timestep is 1, nxtail = 500
Creating standard deviation numbers ...
nbegin = 5, x[nbegin] = 5, tbegin = 5
Segmentation fault

Any feedback will be much appreciated

Best wishes
George Patargias

dipcorr.xvg (34.1 KB) dipcorr.xvg Autocorrelation function of the total dipole moment George Patargias, 03/20/2007 03:51 PM


#1 Updated by David van der Spoel over 13 years ago

Could you please upload your dipcorr.xvg file?

#2 Updated by George Patargias over 13 years ago

Created an attachment (id=109)
Autocorrelation function of the total dipole moment

#3 Updated by David van der Spoel over 13 years ago

I can not reproduce the problem. I get an answer as it should, have also tried
exp_exp for the fit, and that gives more meaningful results probably.

What compiler and OS did you use for this?
It could be another of the gcc-4.1 problems...

#4 Updated by David van der Spoel over 13 years ago

Hi David

Sorry if you got this twice. I just reply to your previous bugzilla mail.

I compiled gmx on Fedora Core 6 x86_64 architecture with gcc-4.1.1-30.

Have you tried the rpms that are provided on the gmx website?
Or upgrading/downgrading the compiler? Or using e.g. the Intel compiler?
Alternatively if you compile with -g and run it in the debugger to see
where it crashes.

It would be very usefull if you could help figure out the steps that


makes to calculate the frequency dependent dielectric constant. I will try to
repeat them "manually" (e.g. in Mathematica).

So it first fits the ACF to, say, an exp, then performs the numerical

derivation and

then Laplace-Fourier transform? Is this the right order? Can we do the

derivation and

the transform without prior fitting. This is if we simulate a polymer or a

protein rather

than a simple polar Debye liquid like water.

It is described in JCP paper.
The largest problem is the smooth transition from ACF to fitted curve.
This was definitely done prior to taking the derivative.

Also available in: Atom PDF