Hello,
I am writing my masters thesis about quantum free energy differences using Jarzynski’s equality. We are studying the double proton transfer in the Formic Acid Dimer (FAD). In order to do that I combined the i-pi server with DFTB+ for electronic structure calculations and with LAMMPS as a client including the Colvars module to perform steered md (SMD) in the stiff-spring limit. First of all we are just interested in the classical case to understand the basic mechanisms. Before I studied conformational changes in Alanine Dipeptide (ADP) to understand SMD. That worked actually quite well.
But now my problem: as I started studying the FAD I ran into problems and I am not really sure from which part they arise. When I change the friction coefficient of the Langevin thermostat I use, the barrier height obtained with SMD changes significantly. And I can’t find a range of values for the coefficient where the free energy profile doesn’t change. I actually hoped to find such a plateau.
Searching about the reasons for this behaviour I ran into this post in the namd mailing list: http://www.ks.uiuc.edu/Research/namd/mailing_list/namd-l.2006-2007/2035.html
There someone states that
"2 - Algorithms used to control temperature usually are not designed to handle systems in which external forces induce motions in a preferred direction.
The Langevin thermostat for instance, will apply a net (average) zero force to atoms that do not move on a preferred direction, but likely will introduce an artificial viscous drag to atoms that are being pulled. Thus, your force peaks will be larger.
3 - Some temperature control methods induced center of mass motion, which is not good when you have fixed reference points like in SMD."
Do you have experience with this? So are there problems using a Langevin thermostat due to the directional force?
And are there maybe ways to prevent that from happening?
Maybe you can make a great Christmas present since most parts of my thesis worked well until now but now I am stuck and it is not so much time left.
Thanks and kind regards,
Nicolas
Hello,
I am writing my masters thesis about quantum free energy differences using
Jarzynski's equality. We are studying the double proton transfer in the
Formic Acid Dimer (FAD). In order to do that I combined the i-pi server
with DFTB+ for electronic structure calculations and with LAMMPS as a
client including the Colvars module to perform steered md (SMD) in the
stiff-spring limit. First of all we are just interested in the classical
case to understand the basic mechanisms. Before I studied conformational
changes in Alanine Dipeptide (ADP) to understand SMD. That worked actually
quite well.
But now my problem: as I started studying the FAD I ran into problems and
I am not really sure from which part they arise. When I change the friction
coefficient of the Langevin thermostat I use, the barrier height obtained
with SMD changes significantly. And I can't find a range of values for the
coefficient where the free energy profile doesn't change. I actually hoped
to find such a plateau.
Searching about the reasons for this behaviour I ran into this post in the
namd mailing list: http://www.ks.uiuc.edu/Research/namd/mailing_list/
namd-l.2006-2007/2035.html
There someone states that
"*2 - Algorithms used to control temperature usually are not designed to **handle
**systems in which external forces induce motions in a preferred
direction. *
*The **Langevin thermostat for instance, will apply a net (average) zero
force **to **atoms that do not move on a preferred direction, but likely
will **introduce **an **artificial viscous drag to atoms that are being
pulled. Thus, your force **peaks **will be larger. *
*3 - Some temperature control methods induced center of mass motion, which
**is *
*not good when you have fixed reference points like in SMD." *Do you have
experience with this? So are there problems using a Langevin thermostat due
to the directional force?
And are there maybe ways to prevent that from happening?
have you tried a different thermostat? LAMMPS has a whole zoo of them.
while fix langevin is very convenient during equilibration (as it quickly
achieves equipartitioning of energy, a prerequisite for equilibration), it
does manipulate the dynamics. so for sensitive simulation methods, the time
constant - after equilibration - should be so large, that the frictional
damping has no significant impact. it would be simpler, though, to use a
nose-hoover thermostat (after initial equilibration with fix langevin) and
avoid direct manipulation of the velocity altogether.
axel.
Hi Nicolas, I second Axel’s suggestion to try different thermostats. The hypothesis that the work applied equals the change in free energy can only be used if the dissipated work is nearly zero. When you use a stiff spring, you are effectively applying very large forces, and the resulting excess kinetic energy will be eaten up by the thermostat. For this reason, Jarzynski’s equality is not the most used free-energy estimator, because of its slow convergence.
You could consider, for example, using the WHAM or MBAR estimators (used in conjunction with umbrella sampling) or thermodynamic integration. To run an umbrella sampling protocol, simply run multiple SMD simulations with different final points, and the part of the colvars’ trajectory after targetNumSteps are done into one of the available implementations of WHAM.
Alternatively, if you are using a recent version of LAMMPS, the new Colvars option writeTIPMF should be available for the moving harmonic restraint that you are using.
Giacomo