Re: [ESPResSo-users] Calibration of friction/mass

[Owen Hickey]

Hmmm... While Green-Kubo is elegant in theory, in practice it actually
takes a rather long time (I think it's weeks compared to hours using a
quiescent flow). Joost actually looked into the best way of measuring
mobilities/friction/diffusion in http://arxiv.org/abs/1503.02671. That
may go into more detail than you'd like though, but I think he also
noticed a difference between the friction in a thermalized LB and a
quiescent LB (I am not 100% sure though). I actually though of a
simpler way of getting a half-decent solution on the way home. If one
simply shifts the time so that t_prime = t_0 -dt/2, the curve should
look much better. Essentially the particles lose the first half of the
first integration step, so shifting the time should fix the problem to
first order... I hope.

Owen

On Tue, May 19, 2015 at 6:06 PM, Marcello Sega
<address@hidden> wrote:
> Hi,
>
> to obtain the friction, it should be enough to compute the force
> autocorrelation at equilibrium [e.g. Kubo, Rep. Prog. Phys. 29 255
> (1966) doi:10.1088/0034-4885/29/1/306  eq 8.15 ].
>
> This way you don't need to perform out-of-equilibrium simulations of
> any kind, neither fitting exponential decays -- which are anyway not
> exponential when the hydrodynamic interaction is involved, nor
> terminal velocity calculations.
>
> This should bypass the 1st step problem.
>
> Hope this helps,
>
> M.
>
>
> On Tue, May 19, 2015 at 5:51 PM, Owen Hickey <address@hidden> wrote:
>> Hey Markus,
>>
>> The Verlet algorithm normally works as follows:
>>
>> v+=0.5*F*dt
>> x+=v*dt
>> *recalculate forces, including the LB coupling force, LB gets whole force 
>> here*
>> v+=0.5*F*dt //note here only half of the force is applied to the particles
>>
>> The issue is that the Velocity verlet works as follows on the first step:
>>
>> *recalculate the forces ignoring the LB coupling force if something
>> has change in the simulation setup*
>> v+=0.5*F*dt
>> x+=v*dt
>> *recalculate forces, including the LB coupling force, LB gets whole force 
>> here*
>> v+=0.5*F*dt //note here only half of the force is applied to the particles
>>
>> One can avoid the first recalculation of the forces by using the
>> reuse_forces flag when calling integrate from Tcl. The issue is of
>> course that when one redefines potentials, or in your case changes the
>> velocity of the particles, the appropriate forces also change. In a
>> sense you are only really missing the first half of the first time
>> step, since in the second half of the Verlet step the forces will be
>> recalculated with the coupling force. The issue is that there is no
>> obvious way of applying only half of the force to the LB fluid without
>> calculating the coupling force twice, which would obviously be
>> inefficient. One could certainly due something where recalc_forces
>> flag is used and the LB force coupling is calculated twice at every
>> integration step (the bare friction would then also be off by a factor
>> of two).
>>
>> To summarize, it is doable, the question is if it is worth the effort
>> of hacking in a solution...
>>
>> Hope that helps,
>> Owen
>>
>> On Tue, May 19, 2015 at 5:41 PM, Wink, Markus
>> <address@hidden> wrote:
>>> Hello everybody,
>>>
>>>
>>>
>>> I exhibit the same behavior. To be more precisely, it happens in the first
>>> integration step.
>>>
>>> I used the OIF Implementation and wanted to perform the mass/friction
>>> calibration concerning doi:10.1016/j.camwa.2012.01.062.
>>>
>>> The object was put into a fluid at rest, all surface nodes were set to an
>>> initial velocity and the exponential decay of the velocity with time was
>>> investigated.
>>>
>>>
>>>
>>> The problem seems to be that ,when performing the first integration step,
>>> the LB forces are not included in the particle force (concerning the warning
>>> message). Having a look in integrate.cpp shows, that “transfer_momentum = 0”
>>> in line 245. Thus we don’t have a fluid/particle-coupling there since no
>>> momentum transfer between fluid and particle occurs (?). Why is that the
>>> case?
>>>
>>>
>>>
>>> I tried to perform an integration step before setting the surface nodes to
>>> the initial velocity (I thought this might help). Nevertheless, the warning
>>> always occurs in the first integration step after setting the velocity. Is
>>> there any way to avoid that to get a smooth exponential decay of the
>>> velocity? Does one really need to bother about the particle not being
>>> coupled to the fluid in the first integration step?
>>>
>>>
>>>
>>> Greetings and thanks for your answer,
>>>
>>>
>>>
>>> Markus
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> Von: address@hidden
>>> [mailto:address@hidden
>>> Im Auftrag von Holst, Lena
>>> Gesendet: Montag, 18. Mai 2015 17:29
>>> An: 'address@hidden'
>>> Betreff: [ESPResSo-users] Calibration of friction/mass
>>>
>>>
>>>
>>> Hello,
>>>
>>>
>>>
>>> I like to simulate a round object in a fluid.
>>>
>>> Now, I try to calibrate the relation mass/friction.
>>>
>>> I have a fixed mass of 15.004 and change the friction.
>>>
>>> The Reynoldsnumber has a value of 0,0579 and the timesteps are 0,05us.
>>>
>>> The problem is, that the velocity does not shows an exponent trend at the
>>> second timestep, the value is too big. For better understanding there is a
>>> picture in the appendix.
>>>
>>> The blue line is the computed trend, the red and green simulated trends.
>>>
>>> I have try the same at a mass of 1 and a friction of 11, there I have the
>>> same problem.
>>>
>>> Could you help me?
>>>
>>> Thank you.
>>>
>>>
>>>
>>> Kind regards,
>>>
>>>
>>>
>>> Lena Holst
>>>
>>> _________________________________________________________
>>>
>>> Fraunhofer ICT-IMM
>>>
>>> Gruppe Modellierung / Modeling Group
>>>
>>> Abteilung Zukunftstechnologien / Future Technologies Department
>>>
>>> Carl-Zeiss-Straße 18-20, 55129 Mainz, Germany
>>>
>>> Phone +49 6131 990-396 / Fax +49 6131 990-205 / Mobile +49 1577-5327027
>>>
>>> address@hidden / www.imm.fraunhofer.de
>>>
>>>
>>
>
>
>
> --
> University of Vienna, Institute of Computational Physics