Hi Giacomo,

Sorry for writing back before your reply, but I think I found the bug.

In the attached input/output, you will observe the following:

1_md_control:

Normal MD, no extended system, no bias.

2_extended_mass_langevin:

Identical to #1, except with the following extended system:

extendedFluctuation 1

extendedTimeConstant 100,000

extendedTemperature 5,032

extendedLangevinDamping 100

3_extended_mass_langevin_harmonic_hack:

Identical to #2, except added the following harmonic block **(with zero force constant)**:

harmonic {

name tamd_colvar_1_harmonic

colvars tamd_colvar_1

forceConstant 0

centers 45

}

Here are the values of the CV at timestep 400,000, for each of these simulations:

Sim# Timestep Dihedral angle CV Extended CV

#1 400000 4.75343480219982e+01 N/A

#2 400000 4.75343480219982e+01 -1.51452462251092e+02

#3 400000 -1.55088564365035e+02 -1.51021479619022e+02

To me, #2 and #3 should be identical since the force constant is zero. #1 and #2 should not be identical. I think that forces are not being fed back to the atomic dof unless I call “harmonic”. I realize that my parameters for the extended system may not be optimal… however, when after 400,000 steps #1 and #2 are identical to 15 significant digits (and so are the energies), I think this is probably a bug. #3 exhibits, what I consider, expected behavior… tight coupling of the atomic and extended CV, and variation from the system with no extended system (#1).

Thanks again for the assistance!

lammps_small_example_ALA_peptide_harmonic_hack.tar.gz (1010 KB)