Re: [ESPResSo-users] No conservation of momentum/mass in LBM ??

[Ulf Schiller]

Hi Markus,

On 04/05/2016 04:44 AM, Wink, Markus wrote:
> Hi Ulf,
> 
> Some mails ago you wrote: 
>> You could check that the convergence is O(Ma^2).
> 
> Why should the convergence be O(Ma^2)? Could you be so kind and explain it to 
> me?  It is not clear for me.

This is simply because the equilibrium distribution includes terms up to
O(Ma^2), so the truncation errors should actually be O(Ma^3) - sorry for
the confusion. In practice there may be subtle cancellations of errors,
or other errors may dominate, but if there are compressibility
artifacts, they should at least be smaller for smaller velocity.

Hope this helps,
Ulf

> -----Ursprüngliche Nachricht-----
> Von: Ulf Schiller [mailto:address@hidden 
> Gesendet: Montag, 21. März 2016 12:17
> An: Wink, Markus; address@hidden
> Cc: 'Georg Rempfer'; Joost de Graaf
> Betreff: Re: AW: [ESPResSo-users] No conservation of momentum/mass in LBM ??
> 
> Hi,
> 
> On 03/21/2016 12:05 PM, Wink, Markus wrote:
>> Hello everybody,
>>
>>  
>>
>> by „reduce the force density” I guess you mean the velocity of the 
>> rhomboids. I did that, and indeed, the extents of in- and outlet 
>> effects are almost equal. So I guess, that I have some compressibility 
>> artefacts in my system, which gets more pronounced if the velocity of 
>> the rhomboids increase. So I was thinking, that it might be related to 
>> the mach number. But looking at the mach number in the script I was 
>> attaching some mails ago, it is of the order of  1E-3 (if I am not 
>> mistaken). So compressibility effects shouldn’t play any role.
>>
>>  
>>
>> Still puzzling…
> 
> You could check that the convergence is O(Ma^2).
> 
> Cheers,
> Ulf
> 
>> *Von:address@hidden [mailto:address@hidden *Im 
>> Auftrag von *Georg Rempfer
>> *Gesendet:* Freitag, 18. März 2016 10:19
>> *An:* Wink, Markus
>> *Cc:* Joost de Graaf; address@hidden; Ulf Schiller
>> *Betreff:* Re: [ESPResSo-users] No conservation of momentum/mass in LBM ??
>>
>>  
>>
>> I think it's possible that the asymmetry stems from the fact that at 
>> one boundary, there is a density discontinuity and on the other one, 
>> there is not.
>>
>>  
>>
>> Can you reduce the force density and rerun? I think that will reduce 
>> the density gradient and therefore the discontinuity at one end.
>>
>>  
>>
>> I might be completely wrong about this.
>>
>>  
>>
>> On Fri, Mar 18, 2016 at 9:11 AM, Wink, Markus 
>> <address@hidden <mailto:address@hidden>>
>> wrote:
>>
>> Hello everybody,
>>
>>  
>>
>> the mass flux is constant in the system. Makes sense ... Thanks a lot!
>>
>>  
>>
>> The only question that is left for me now is, why the entry and outlet 
>> are not symmetric. Actually I am not sure anymore, if they should be 
>> symmetric. The system does not exhibit a mirror symmetry, so my 
>> assumption of having symmetric in- and outlet effects might not be 
>> valid (but I am not sure about it).
>>
>>  
>>
>> Any ideas about it? Notice, that although the graph implies the in- 
>> and outlet have different velocities, I set the velocities of the 
>> rhomboids at both sides to same velocities.
>>
>>  
>>
>> Greetings
>>
>>  
>>
>> Markus
>>
>>  
>>
>>  
>>
>>  
>>
>> *Von:*Joost de Graaf [mailto:address@hidden 
>> <mailto:address@hidden>]
>> *Gesendet:* Donnerstag, 17. März 2016 16:58
>> *An:* Georg Rempfer
>> *Cc:* Wink, Markus; address@hidden 
>> <mailto:address@hidden>; Ulf Schiller
>>
>>
>> *Betreff:* Re: [ESPResSo-users] No conservation of momentum/mass in LBM ??
>>
>>  
>>
>> Georg is right. Done that once for a sphere, it worked out perfectly 
>> back then.
>>
>>  
>>
>> On 17 March 2016 at 15:55, Georg Rempfer <address@hidden 
>> <mailto:address@hidden>> wrote:
>>
>> If the density goes up along the channel, the velocity has to go down 
>> to fulfill mass conservation.
>>
>>  
>>
>> Since there are density variations in the system, you should check the 
>> actual mass flux instead of the averaged velocity. That means you 
>> should integrate density*velocity over every slice and look at the 
>> component normal to the slice.
>>
>>  
>>
>> On Thu, Mar 17, 2016 at 10:33 AM, Wink, Markus 
>> <address@hidden <mailto:address@hidden>>
>> wrote:
>>
>> Hello everybody,
>>
>> I checked the mass flux. It is constant over the length of the channel 
>> (some oscillations at the outlet, but I am not concerned about it). I 
>> revised the script and now I get the maximum velocity right (less than 
>> 1% deviation to the theoretical one). But I am still puzzling with 
>> some aspects. First of all I checked whether the density is constant. 
>> Within one plane perpendicular to the direction of flow, that is the case.
>> Along the direction of transport, I notice a drop of the density. This 
>> makes sense to me, since the density is proportional to the pressure 
>> and I expect a linear pressure profile along the channel.
>>
>> Nevertheless, there are some questions left.
>>
>> 1) I have noticed, that in both cases, whether if I apply a body force 
>> to the fluid or an constant velocity inlet boundary condition, the 
>> maximum velocity of the profile is a bit lower than the expected one 
>> (although it is quite good with less than 1% deviation). I am just 
>> wondering, since I checked two "methods", whether this deviation lies 
>> in the nature of the LBM algorithm?
>> 3) With a constant velocity inlet and constant velocity outlet I have 
>> inlet and outlet effects of cause a certain length (until the profile 
>> develops). I was expecting that the two lengths should be equal, since 
>> I have equivalent boundary conditions. In the appendix you will see, 
>> that this is not the case (it shows the velocity as a function of the 
>> x-position, while y- and z- are set to half the channel width/height).
>> The entry effect seems to be much more pronounced, but I am not sure 
>> why. Does anyone have an idea?
>> 4) As an outlet condition it would be neat to have a constant pressure 
>> boundary condition (with that, one would eliminate the outlet effect). 
>> I was thinking to put the outlet-nodes to a constant pressure via 
>> lbnode set. Is there a command for setting the pressure of a node to a 
>> given value (investigating the source code it seems, that there is 
>> only "lbnode x y z print pi" but not set).
>> 5) What is the proper way to get the mean velocity out of the mass flux?
>> If I sum up the velocities and divide by the cross section, I get a 
>> slight increase of v_mean along the channel (I have no idea why).
>> 6) Have anyone ever checked the second order accuracy of the LBM in 
>> ESPResSo?
>>
>> Greetings and thanks a lot for your help!
>>
>> Markus
>>
>>
>>
>>
>> -----Ursprüngliche Nachricht-----
>> Von: address@hidden
>> <mailto:address@hidden>
>> [mailto:espressomd-users-bounces+markus.wink
>> <mailto:espressomd-users-bounces%2Bmarkus.wink>address@hidden
>> nu.org <mailto:address@hidden>] Im Auftrag von Ulf 
>> Schiller
>> Gesendet: Mittwoch, 16. März 2016 13:49
>> An: address@hidden <mailto:address@hidden>
>> Betreff: Re: [ESPResSo-users] No conservation of momentum/mass in LBM ??
>>
>>
>> Hi Markus,
>>
>> sorry for the confusion. In my earlier email I should have said mass 
>> flux of course, i.e., the Q in the formulas Kai provided. You can 
>> calculate that from the simulated velocity profile for each plane 
>> along the channel, and it should be constant.
>>
>> Best,
>> Ulf
>>
>> On 03/16/2016 08:21 AM, Georg Rempfer wrote:
>>> Assuming the fluid is not compressed (you could check that, but it's 
>>> likely true), the mass flux is proportional to the velocity. You 
>>> claim the average velocity in the channel direction is too low half 
>>> way between inlet and outlet. This implies that mass gets lost 
>>> between the inlet/outlet and the middle (or that the lb fluid is 
>>> denser in the middle). Can you check that?
>>>
>>> On Wed, Mar 16, 2016 at 1:02 PM, Wink, Markus 
>>> <address@hidden <mailto:address@hidden>
>> <mailto:address@hidden
>> <mailto:address@hidden>>>
>>> wrote:
>>>
>>>     Hello everybody,____
>>>
>>>     __ __
>>>
>>>     thank you for your answers. I did not get it. Which quantity is of
>>>     interest? Mass flux or momentum flux? I am not sure about it,
>>>     although to check whether mass conservation is fulfilled, both
>>>     should work, am I right?____
>>>
>>>     __ __
>>>
>>>     __ __
>>>
>>>     Greetings____
>>>
>>>     __ __
>>>
>>>     Markus____
>>>
>>>     ____
>>>
>>>     __ __
>>>
>>>     __ __
>>>
>>>   
>>  
>> *Von:address@hidden
>> g
>> <mailto:address@hidden>
>>>     <mailto:address@hidden
>> <mailto:address@hidden>>
>>>     [mailto:espressomd-users-bounces+markus.wink
>> <mailto:espressomd-users-bounces%2Bmarkus.wink>
>>>     <mailto:espressomd-users-bounces%2Bmarkus.wink
>> <mailto:espressomd-users-bounces%252Bmarkus.wink>>address@hidden
>> ongnu.org
>> <mailto:address@hidden>
>>>     <mailto:address@hidden
>> <mailto:address@hidden>>] *Im Auftrag von *Georg Rempfer
>>>     *Gesendet:* Mittwoch, 16. März 2016 11:26
>>>     *An:* Kai Szuttor
>>>     *Cc:* address@hidden
>> <mailto:address@hidden> 
>> <mailto:address@hidden
>> <mailto:address@hidden>>
>>>     *Betreff:* Re: [ESPResSo-users] No conservation of momentum/mass in
>>>     LBM ??____
>>>
>>>     __ __
>>>
>>>     I agree with you argument, Markus. Mass conservation dictates that
>>>     the normal flow through every surface along the channel should be
>>>     the same (assuming the flow is incompressible). Together with the
>>>     fixed shape of the fully developed flow profile, this uniquely
>>>     determines the flow in regions far away from the inlet/outlet. So if
>>>     this does not come out correctly, mass conservation should be broken
>>>     somewhere. I don't think this is possible in the LB. Can you
>>>     calculate this flux through the surfaces along the channel and show
>>>     us where exactly it differs from the inlet/outlet?____
>>>
>>>     __ __
>>>
>>>     On Tue, Mar 15, 2016 at 5:09 PM, Kai Szuttor
>>>     <address@hidden <mailto:address@hidden>
>> <mailto:address@hidden <mailto:address@hidden>>>
>>> wrote:____
>>>
>>>     Now with attachment :)
>>>
>>>     Am 15/03/16 um 14:07 schrieb Ulf D Schiller:
>>>     > Did you check the flow rates directly, i.e., the momentum flux per
>>>     plane? Your argument seems correct, so I can only guess that there's
>>>     some
>>>     > flaw in the calculation of the mean velocity. I think there's an
>>>     expression for the flux in rectangular channels that one could use.
>>>     >
>>>     > Best,
>>>     > Ulf
>>>     >
>>>     > Sent from a mobile device.
>>>     >
>>>     >
>>>     > -------- Original message --------
>>>     > From: "Wink, Markus" <address@hidden
>> <mailto:address@hidden>
>>>     <mailto:address@hidden
>> <mailto:address@hidden>>>
>>>     > Date: 3/15/2016 8:47 AM (GMT-05:00)
>>>     > To: 'Ivan Cimrak' <address@hidden <mailto:address@hidden>
>> <mailto:address@hidden <mailto:address@hidden>>>,
>>>     address@hidden <mailto:address@hidden>
>> <mailto:address@hidden 
>> <mailto:address@hidden>>
>>>     > Subject: Re: [ESPResSo-users] No conservation of momentum/mass in
>>>     LBM ??
>>>     >
>>>     > Hi Ivan, Hi Florian,
>>>     >
>>>     >
>>>     >
>>>     >>/How did you compute the expected maximum velocity? As far as I
>>>     know, the poisseuille flow has an exact expression for the velocity
>>>     in the case
>>>     > of channel with circular cross section, and you have a rectangular
>>>     one.///
>>>     >
>>>     > / /
>>>     >
>>>     > I know the velocity of the rhomboid. Thus I know the mean velocity
>>>     of the fluid (assuming it is incompressible). I took that for
>>>     calculating the
>>>     > Reynoldsnumber, pressure gradient and theoretical velocity profile
>>>     (using the expression in the book  “Viscous Fluid Flow” of Frank M.
>>>     White).
>>>     >
>>>     >
>>>     >
>>>     > /> //The boundaries are momentum sinks. (Florian)/
>>>     >
>>>     > /> Now I read the comment of Florian -//does that mean that amount
>>>     of fluid is decreasing when no-slip is prescribed?/
>>>     >
>>>     > I still don’t get it. That the boundaries are momentum sinks, I
>>>     agree. Due to the present of the walls and the “friction” of the
>>>     fluid there, I
>>>     > achieve the poiseuille profile. But I still hold the opinion, that
>>>     the mean velocity of the fluid should be the same.
>>>     > Imagine the following physical experiment: you have a syringe pump
>>>     set up with a constant flow rate Q0 connected to a rectangular
>>>     channel having
>>>     > a cross section A=w*h. The fluid in the channel then has a mean
>>>     velocity of v_mean=Q/A. Assuming an incompressible medium, this
>>>     means the
>>>     > velocity should be the same at every slice normal the direction of
>>>     transport.
>>>     > In my simulation, the mean velocity should be velocity v0 of the
>>>     rhomboid.
>>>     >
>>>     > So I still don’t get the deviation to the theoretical value…
>>>     >
>>>     > Greetings Markus
>>>     >
>>>     >
>>>     >
>>>     >
>>>     >
>>>     >
>>>     >
>>>     >
>>>   
>>  
>> *Von:address@hidden
>> g
>> <mailto:address@hidden>
>>>     <mailto:address@hidden
>> <mailto:address@hidden>>
>>>     > [mailto:espressomd-users-bounces+markus.wink
>> <mailto:espressomd-users-bounces%2Bmarkus.wink>
>>>     <mailto:espressomd-users-bounces%2Bmarkus.wink
>> <mailto:espressomd-users-bounces%252Bmarkus.wink>>address@hidden
>> ongnu.org
>> <mailto:address@hidden>
>>>     <mailto:address@hidden
>> <mailto:address@hidden>>] *Im Auftrag von *Ivan Cimrak
>>>     > *Gesendet:* Dienstag, 15. März 2016 13:22
>>>     > *An:* address@hidden
>> <mailto:address@hidden> 
>> <mailto:address@hidden
>> <mailto:address@hidden>>
>>>     > *Betreff:* Re: [ESPResSo-users] No conservation of momentum/mass
>>>     in LBM ??____
>>>
>>>     >
>>>     >
>>>     >
>>>     > Hi Markus,
>>>     >
>>>     >
>>>     >
>>>     >     Hello Everybody,
>>>     >
>>>     >
>>>     >
>>>     >     so far, in the LBM scheme only the body force is implemented
>>>     and no velocity/pressure boundary condition. So I was thinking on a
>>>     way of
>>>     >     mimicking a “velocity boundary” condition without changing the
>>>     source code. I am aware that one should, as a proper approach, using
>>>     Zou/He
>>>     >     boundary conditions and adjusting the distribution functions
>>>     according to the boundary conditions.
>>>     >
>>>     >
>>>     >
>>>     >     For that I have constructed a channel with rectangular cross
>>>     section and put the fluid inside. Furthermore, two rhomboids where
>>>     put inside,
>>>     >     one at the inlet of the channel, one at the outlet. The cross
>>>     section of the two rhomboids is equal to the cross section of the
>>>     channel,
>>>     >     furthermore they have a constant velocity v0.
>>>     >
>>>     >     My idea was, that, since the no-slip boundary condition is
>>>     implemented, I force the fluid nodes adjacent to the rhomboids to
>>>     have a constant
>>>     >     velocity, thus achieving constant velocity inlet/outlet
>> condition.
>>>     >
>>>     >
>>>     >
>>>     >     As a result I achieve a poiseuille profile in the middle of
>>>     the channel (fully developed flow after inlet/outlet effects). The
>>>     qualitative
>>>     >     pressure gradient looks proper, too.
>>>     >
>>>     >     Nevertheless, the maximum velocity is not the same as I
>>>     expected (factor 3 to the expected one).
>>>     >
>>>     > How did you compute the expected maximum velocity? As far as I
>>>     know, the poisseuille flow has an exact expression for the velocity
>>>     in the case
>>>     > of channel with circular cross section, and you have a rectangular
>>>     one.
>>>     >
>>>     >
>>>     > I have checked the mean velocity. I would expect, that the mean
>>>     velocity of the fluid should be the velocity v0 of the rhomboid
>> (due to
>>>     > mass/momentum conservation), I get less (10 %).
>>>     >
>>>     > This is strange. The amount of fluid at the inlet (integral of
>>>     velocity over the inlet surface, in this case is the velocity
>>>     constant over the
>>>     > inlet surface) should be the same as integral over the middle
>>>     cross section, as well as integral over the outlet surface....
>>>     Supposing you
>>>     > computed the average velocity as sum of velocities over the LB
>>>     nodes at middle cross section divided by number of these nodes, you
>>>     should have
>>>     > obtained the velocity at the inlet...
>>>     >
>>>     > Now I read the comment of Florian - does that mean that amount of
>>>     fluid is decreasing when no-slip is prescribed?
>>>     >
>>>     > Ivan
>>>     >
>>>     >
>>>     >
>>>     > What is wrong with my idea stated here? Obviously, something is
>>>     not correct, but I have no idea, what the reason for that is. Where
>>>     does the
>>>     > momentum vanish?
>>>     >
>>>     >
>>>     >
>>>     > Does anybody have an idea?
>>>     >
>>>     >
>>>     >
>>>     > Sincerely,
>>>     >
>>>     >
>>>     >
>>>     > Markus
>>>     >
>>>     >
>>>     >
>>>     >
>>>     >____
>>>
>>>     __ __
>>>
>>>
>>
>>
>> --
>> Dr. Ulf D. Schiller
>> Assistant Professor
>> Department of Materials Science and Engineering Clemson University
>> 161 Sirrine Hall
>> Clemson, SC 29634
>>
>> Office: 299c Sirrine Hall
>> Phone: 1-864-656-2669 <tel:1-864-656-2669>
>> Fax: 1-864-656-5973 <tel:1-864-656-5973>
>>
>>  
>>
>>  
>>
>>  
>>
> 
> 
> --
> Dr. Ulf D. Schiller
> Assistant Professor
> Department of Materials Science and Engineering Clemson University
> 161 Sirrine Hall
> Clemson, SC 29634
> 
> Office: 299c Sirrine Hall
> Phone: 1-864-656-2669
> Fax: 1-864-656-5973
> 


-- 
Dr. Ulf D. Schiller
Assistant Professor
Department of Materials Science and Engineering
Clemson University
161 Sirrine Hall
Clemson, SC 29634

Office: 299c Sirrine Hall
Phone: 1-864-656-2669
Fax: 1-864-656-5973