From: Sterling Paramore (paramore_at_hec.utah.edu)
Date: Wed May 17 2006 - 20:50:10 CDT
True, you do not need to know the numerical value of the energy in order
to run dynamics. My point is, what do the dynamics mean if the energy
is ill defined? In order to interpret MD in terms of meaningful
physical quantities (e.g., entropy, free energy, material properties,
etc.), you always have to take an ensemble average. If the energy is
ill defined, the distribution function is also ill defined and so are
any ensemble averages of phase functions.
Mark Abraham wrote:
> Sterling Paramore wrote:
>> My 2 cents:
>> I've never heard of this neutralizing plasma.
> It's a physical interpretation of what happens in the "trick" where
> the conditionally-convergent Ewald sum is partitioned into two
> functions that converge rapidly in either real or reciprocal space.
> See (among others) Deserno & Holm, J Chem Phys 109:7678. The four
> early PME papers I checked just now require neutrality for their
> derivations, but none state that neutrality is prerequisite for
> application of the methods.
>> However, if your system is charged and the Ewald sum does not
>> converge then you do not have well defined energy. Without a well
>> defined energy, you cannot have a well defined thermodynamic state or
>> phase space distribution function. Since the whole point of
>> molecular dynamics is to sample the distribution function, even if
>> the dynamics are "right," they are without meaning if the energy
> I disagree. If the energy as evaluated is not being used in
> determining the dynamics step - as is normal in MD where the
> integration of the forces is what happens - then the value might as
> well be zero. You are sampling the correct phase space through the
> forces being correct. The dynamics being "right" is normally all you
> need, although an accurate energy evaluation is nice as a check that
> things are going as planned. Obviously an MC simulation would be a
> different affair.
> You can take a single particle in a harmonic oscillator, give it a
> position and velocity and integrate the forces to watch the time
> evolution. You never need to compute the energy to generate the
> dynamics. The fact that the total energy is a simple function of the
> position and momentum is irrelevant until one comes to want to check
> that the integration is good enough.
> Certainly for many systems you won't want a highly charged state
> because it is unphysical.
> Does anyone know whether astrophysicists use the Ewald sum for their
> gravitational simulations - if every particle has positive mass than
> it is equivalent to an all-positively-charged electrostatic
> simulation? They certainly use multipole methods...
>>> Dear NAMD users,
>>> I have a general question on the charged system. Although this is
>>> not directly associated with the NAMD , I hope you will let me ask
>>> Some people claim that the Ewald sum is meaningful only for a
>>> neutral system. They say that in a case of a charged system, we
>>> should neutralize the system to avoid the divergence in the Ewald
>>> sum. And in order to neutralize the system, people use the "uniform
>>> background neutralizing plasma" method with Ewald sum or particle
>>> mesh Ewald (PME) for charged systems. But, when I look at the
>>> equation for the uniform neutralizing plasma, it is just a constant
>>> (I mean, independent of the positions of particles) added to the
>>> Ewald sum equation.
> Can you provide a reference for this please?
> In fact, since what we need in the simulation, is not the
>>> energy, but force, I don't think that the uniform neutralizing
>>> plasma DOES NOT affect the dynamics AT ALL. Consequently, unless we
>>> are interested in the properties associated with the energy of the
>>> system, I don't think we have to include the neutralizing plasma
>>> term in the Ewald sum. I mean that we don't have to neutralize the
>>> system if we are interested in properties such as the structure of
>>> the system.
>>> As a result, if there is something wrong with my argument, please
>>> correct me.
>>> Thanks a lot.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:42:04 CST