Bug #1061

MTTK conserved energy is drifting

Added by Shun Sakuraba over 4 years ago. Updated about 4 years ago.

Target version:
Affected version - extra info:
Affected version:


When I used MTTK integrator, the conserved energy (not total energy) of MTTK is drifting during the simulation.
It is drifting with a constant ratio of ca. -185 kJ/mol/ps.
The drift is caused even when I use the double precision version, and switching potential, and 0.5 fs time-step.
Although the fluctuation in the conserved energy is possible by nature of MTTK equation, constant drift during the full simulation time is not what I expect.
The problem is reproduced with either gcc 4.6.1 or icc 11.1.

Reproduce procedure:
grompp_d -f mpt-mttk05.mdp -c conf.gro -p -o mttk
mdrun_d -deffnm mttk

npt-mttk05.mdp - Simulation condition (558 Bytes) Shun Sakuraba, 12/06/2012 05:29 AM

conf.gro - Initial configuration (prepared with Berendsen barostat) (202 KB) Shun Sakuraba, 12/06/2012 05:29 AM - Topology (102 Bytes) Shun Sakuraba, 12/06/2012 05:29 AM

mttk05.xvg - Plot of energy drifting (779 KB) Shun Sakuraba, 12/06/2012 05:29 AM

Associated revisions

Revision b18b7833 (diff)
Added by Michael Shirts over 4 years ago

Fix for the pressure in MTTK with constraints + dispersion

Fix for pressure in MTTK - broken at some point in the
merge process. The MTTK+constraints code definitely needs to
be simplified for 5.0, it's too easy to break on code rearrangements
right now!

Testing with simple systems over long time shows that the pressure is
unbiased, while the PR pressure is off by a factor of 1/N (which is totally
fine for most realistic systems).

Fixes redmine bug #1061
Change-Id: Ie82cf8d6bdeb3e47db044ae34fc42c077ad73c75

Revision 66cab279 (diff)
Added by Mark Abraham over 4 years ago

New reference files for freeenergy/relative

The code version includes Berk's LINCS fix, and Michael's MTTK fix

Refs #1061

Change-Id: I5791bf2fb5f9e02b41d4a3543b9f0a6489f546f1


#1 Updated by Michael Shirts over 4 years ago

  • % Done changed from 0 to 100

OK, the commit submitted to 4.6 appears to fix this:

You can check it out -- in a run out to 16 ps. the conserved quantity has stayed within 10 kJ/mol from the total energy, as it should, and is not drifting up -- in the .xvg above, it had dropped 2000 kJ/mol by this point in the code.

When 4.6 is out, I'll go back and figure out the proper changes to backport to 4.5-patches.

#2 Updated by Erik Lindahl about 4 years ago

  • Status changed from New to Closed

Fixed in git by Michael's revision b18b7833.

Also available in: Atom PDF