From: Jeff Comer (jeffcomer_at_gmail.com)
Date: Thu Nov 13 2014 - 14:37:13 CST
Dear Ms Sumbul,
The difference between a Langevin thermostat and Langevin dynamics is
the strength of the coupling. The algorithm is identical. When used as
a thermostat in an explicit solvent simulation, the damping constant
is typically small and only slightly perturbs the dynamics, although
it includes both dissipation and a random force.
Let's say you have a simulation of a single Na+ ion with explicit
water molecules and no thermostat — just Newton's second law. If you
look at the motion on a long enough timescale (roughly >1 ps), you'll
see the Na+ ion acts like it is Brownian particle with a diffusion
coefficient of D=133 Å^2/ns (assuming your force field agrees with
experiment). This is how diffusive motion emerges in real life —
although the underlying motion is Newtonian, the Na+ ion bounces off a
bunch of chaotically moving water molecules and diffusive motion
naturally emerges.
Let's calculate what langevinDamping coefficient this diffusive motion
would correspond to. langevinDamping = kT/(mD), where kT is the
thermal energy, m is the mass, and D is the diffusion coefficient.
NAMD takes langevinDamping in units of ps^-1. Using the program GNU
units, I calculate
units 'k (300K)/(22.9898u 133*angstroms^2/ns)' 'ps^-1'
81.577159
This damping coefficient is much much bigger than people typically use
as a thermostat (~1 ps^-1). Hence, in an explicit water simulation,
the natural dissipation and random buffeting is 80 times more than is
typically used for maintaining the temperature. A 1 ps^-1 damping
constant has little effect on the dynamics of the ion, but applied to
the whole system, is sufficient to maintain the temperature. On the
other hand, a 20 ps^-1 damping constant on the whole system will
measurably reduce the diffusion coefficient of the ion. I've done this
for water in:
http://dx.doi.org/10.1021/ct300867e
Getting back to the implicit solvent simulations, what we just
calculated is the langevinDamping coefficient we'd need to emulate the
effect of explicit water molecules on the ion. I just ran NAMD for 200
ps for a system consisting of a single Na+ ion. Here are the
thermostat parameters I used:
langevin on
langevinTemp 300
langevinHydrogen on
langevinDamping 81.577159
>From the trajectory, I estimated the diffusion coefficient by
<Δx^2>/(2Δt) and that gives 136.266 Å^2/ns, which is pretty close to
the diffusion coefficient that we wanted. So to run Langevin dynamics
on an implicit solvent model in NAMD, you set langevinDamping to
kT/(mD). If you have atoms or particles with different diffusion
coefficients, you can use langevinFile to set the damping constants
individually.
I hope this is clear!
Jeff
–––––––––––––––––––––––––––––––––––———————
Jeffrey Comer, PhD
Assistant Professor
Institute of Computational Comparative Medicine
Nanotechnology Innovation Center of Kansas State
Kansas State University
Office: P-213 Mosier Hall
Phone: 785-532-6311
On Thu, Nov 13, 2014 at 10:41 AM, Fidan Sumbul <fidansumbul_at_gmail.com> wrote:
> Thank you for your response Dr. Comer.
>
> I'd like to perform Langevin dynamics simulation (stochastic dynamis), but
> not overdamped as Brownian dynamics.
> If Langevin thermostat and Langevin dynamics are the same, why we call
> Molecular dynamics that NAMD MD runs do with langevin thermostat?
>
> Could you please help me to understand what is the real distinction between
> MD simulations with Langevin thermostat and Langevin dynamics simulations?
> Does MD with Langevin thermostat also have random force?
>
> Regards,
>
>
> On Thu, Nov 13, 2014 at 12:37 AM, Jeff Comer <jeffcomer_at_gmail.com> wrote:
>>
>> Hi,
>>
>> The Langevin thermostat and Langevin dynamics are really one in the
>> same, assuming you aren't talking about overdamped Langevin dynamics.
>> You can use generalized Born implicit solvent
>> (http://www.ks.uiuc.edu/Research/namd/2.10b1/ug/node31.html) coupled
>> with "langevin on"
>> (http://www.ks.uiuc.edu/Research/namd/2.10b1/ug/node36.html).
>>
>> I've used the "langevinFile" option to set different friction
>> coefficients for different particles. This means that you can set the
>> diffusion coefficient (in the long time limit) for your particles (or
>> molecules).
>>
>> Best wishes,
>> Jeff
>>
>> –––––––––––––––––––––––––––––––––––———————
>> Jeffrey Comer, PhD
>> Assistant Professor
>> Institute of Computational Comparative Medicine
>> Nanotechnology Innovation Center of Kansas State
>> Kansas State University
>> Office: P-213 Mosier Hall
>> Phone: 785-532-6311
>>
>>
>> On Wed, Nov 12, 2014 at 2:48 PM, Fidan Sumbul <fidansumbul_at_gmail.com>
>> wrote:
>> > Dear NAMD users,
>> >
>> > I'd like to perform an implicit solvent Langevin Dynamics simulation. I
>> > can
>> > do it in Amber, but I wonder if I can use NAMD for this job because I am
>> > familiar with NAMD. Could you provide me a tutorial or link to a
>> > tutorial if
>> > NAMD does Langevin dynamics simulations. I know NAMD uses Langevin
>> > dynamics
>> > in normal MD simulations to keep temperature constant.
>> >
>> > Thanks for your help.
>> >
>> > --
>> > Fidan Sumbul
>> > PhD Student and T.A.
>> > Polymer Research Center
>> > Department of Chemical Engineering
>> > Bogazici University
>
>
>
>
> --
> Fidan Sumbul
> PhD Student and T.A.
> Polymer Research Center
> Department of Chemical Engineering
> Bogazici University
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:21:22 CST