Bug #175

Fwd and rev transformation dgdl values are not complementary without soft-coring during TI with PME

Added by Bharat V. almost 13 years ago. Updated almost 13 years ago.

Erik Lindahl
Target version:
Affected version - extra info:
Affected version:


Created an attachment (id=253)
All input and output files for Ala<->Gly and Ethane<->Methane transformations

I was trying do the charge transformations keeping LJ parameters intact for Ace-Ala-Nac -> Ace-Gly-Nac system in forward (ala->gly) and reverse (gly->ala) directions. Soft-core potentials are not requied for these transformations. I am using PME for coulomb calculations. Fifteen intermediate state are assumed during transformations. The following equality should be checked for the assumption that the there is no hysteresis during forward and reverse simulations:
dG_fwd/dlambda (lambda_i) = -dG_rev/dlambda (1-lambda_i)

For the above mentioned system and for Ethane->Methane also, when PME is used for the coulomb-calculations, i do not get the curves of dG/dlambda vs lambda overlapping when reverse transformation values are plotted onto that of fwd, with both axes reversed and negative of dG/dlambda values (y-axis).
At the same time, I tried the calculations for 20 ps equlibration step, not with PME, but with cut-off for coulomb calculations. I approximately see the equality mentioned above.
This leads to the suspicion that PME implementation for TI might have bug.

Please see the attachments for the run-input and output files.

Following are my machine details and installation specifications:
kernel: 2.6.11-1.14_FC3smp
uname -mpio : i686 i686 i386 GNU/Linux
gcc 3.4
gromacs version 3.3.1 installed with --disable-double and FFTW-3.0.1

for_bugzilla.tgz (87.2 KB) for_bugzilla.tgz All input and output files for Ala<->Gly and Ethane<->Methane transformations Bharat V., 11/13/2007 10:50 AM


#1 Updated by Berk Hess almost 13 years ago

There was indeed a bug in PME.
For charges that are zero in the A-state, but non-zero
in the B-state the splines were not calculated.
This could lead to NAN's or incorrect results.

Any other combination, such as non-zero A-state charges
and zero B-state charges was correct.

Note that often the effect of the bug can be negligible,
since usually the free-energies are dominated by the real-space
part and the mesh contribution is negligible.

I have fixed this bug for the upcoming 3.3.3 release and 4.0.


Also available in: Atom PDF