calculated potential energy goes wrong in different versions of LAMMPS

Dear all*,*
I find that there is something wrong about the “calculated potential energy (Pe)” in defferent version of LAMMPS in Windows.

Details:
The MD potential consisting of Bond-style Morse and Angle-style cosine/squared is implemented in my calculation. In Ver. 21 May 2008, everything seems right, and of course, the Pe is less than 0, but with the later version such as Ver. 17 Jul 2010, Ver. 4 May 2011 and Ver.12 Apr 2013, the Pe becomes greater than 0, which is quite unreasonable. And the results shows that ONLY Pe (and related total energy) get wrong, which means the kinetic energy, virial stress and elastic modulus are nearly the same in different versions. Moreover, it seems that there is only a difference of the initial value of Pe in Ver 2008 and laters, and the trends of Pe-step curves during relaxation and loading are consistent.

I wonder what happened on the Pe in defferent version of LAMMPS. Will someone please give me some advice?
PS: The atoms in the interaction of specific bond or angle are defined before calculation. So the neighbor list , which is required more strictly in later versions, seems meaningless in my work. Does it matter with this problem?

Many thanks!
Huang

Dear all,
I find that there is something wrong about the "calculated potential energy
(Pe)" in defferent version of LAMMPS in Windows.

Details:
The MD potential consisting of Bond-style Morse and Angle-style
cosine/squared is implemented in my calculation. In Ver. 21 May 2008,
everything seems right, and of course, the Pe is less than 0, but with the
later version such as Ver. 17 Jul 2010, Ver. 4 May 2011 and Ver.12 Apr
2013, the Pe becomes greater than 0, which is quite unreasonable. And the

i disagree with that assessment. for bonded interactions the general
convention is that the potential energy is 0.0 at the minimum of the
potential. that being said, the absolute potential energy in any
simulation is just a diagnostic output and has no specific meaning
since lots of contributions to the total energy are integrated out in
classical (and ab initio) models and thus not included.

i doubt that what you observe is specific to the windows version and
you can monitor all changes to the sources in the SVN/Git
repositories.

axel.

Hello,

in general, no one is likely to look into 1,5-6,5 years old code. So, at first please check if there are any problems with 9 Dec 2014 version.

If you are sure that some problem was introduced before 2008 and 2010 and then went unnoticed (that is unlikely), you should provide proofs.

Regards,
Oleg

16:58, 10 декабря 2014 г., “黄贲” <[email protected]…223…>: