Thanks Axel,
It seems like I will have to decrease the shake tolerance as the efield strength increases. I'll play around with that and see if it helps.
On a similar note, I did a shake tolerance test with this same case (~600 waters, Efield ~7 volts/Angstrom, LJ walls, etc.). I tested tolerances of 1E-2, 1E-3, 1E-4, 1E-5, and 1E-7. None of them conserved energy, but I found that 1E-5 did better than all the rest, including the smallest tolerance 1E-7 (which increased in energy at about twice the rate of the 1E-5 case). Granted, I did not change the number of iterations for most of these, so it could have been hitting the max iterations, but is it possible that the tolerance can be too small?
Jonathan
Thanks Axel,
It seems like I will have to decrease the shake tolerance as the efield strength increases. I'll play around with that and see if it helps.
On a similar note, I did a shake tolerance test with this same case (~600 waters, Efield ~7 volts/Angstrom, LJ walls, etc.). I tested tolerances of 1E-2, 1E-3, 1E-4, 1E-5, and 1E-7. None of them conserved energy, but I found that 1E-5 did better than all the rest, including the smallest tolerance 1E-7 (which increased in energy at about twice the rate of the 1E-5 case). Granted, I did not change the number of iterations for most of these, so it could have been hitting the max iterations, but is it possible that the tolerance can be too small?
i suspect that it is a combination of both, having the right order
of the fixes and the smaller convergence in the shake iterations
that make the difference. due to the water dipole, you'll have a
systematic force tearing the molecule apart, so going for a tighter
convergence could be needed, but since shake does operate
based on the "current" force, adding the electrostatic field after
the applying the constraint may result in an incorrect application
of the constraint. so you could have a cancellation of errors up
to a point and then - with even tighter shake convergence -
see only the error due to ordering of the fixes.
mind you, i am only guessing here based on
what i saw looking at what fix shake does.
axel.