Is my weak form right?

What do you mean by 'wrong'? Does it fail to converge or do you get an unexpected result?

Thank you for your help.

The PDE was derived using conservation of mass, but the results from Fenics showed the mass actually kept growing with time. Because my finite difference code doesn't grow, I consider my finite element code got some problems somewhere.

Interestingly, if I get rid of the (du/dx)^2 term in both Fenics and my finite difference code, the results match each other. Because I haven't seen any examples on manipulating the power of derivatives, I think it might be the cause. I don't think I can take integration by parts for v*(du/dx)^2*dx, I don't know what to do with it, except keep it in this way.

I also tried the convergence study, it's getting a little better if I improve the mash refinement and the polynomial degree. But it doesn't change much, the mass still grows.

Pay special attention to the boundary conditions, especially the expression that you're implicitly setting equal to zero when deriving your weak from. Make sure that that term is the mass flux, otherwise you may have mass coming in across the boundary.

On Thursday April 29 2010 08:57:52 cutejeff wrote:
> Question #107819 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/107819
>
> Status: Answered => Open
>
> cutejeff is still having a problem:
> Thank you for your help.
>
> The PDE was derived using conservation of mass, but the results from
> Fenics showed the mass actually kept growing with time. Because my
> finite difference code doesn't grow, I consider my finite element code
> got some problems somewhere.
>
> Interestingly, if I get rid of the (du/dx)^2 term in both Fenics and my
> finite difference code, the results match each other. Because I haven't
> seen any examples on manipulating the power of derivatives, I think it
> might be the cause. I don't think I can take integration by parts for
> v*(du/dx)^2*dx, I don't know what to do with it, except keep it in this
> way.

Are you having (du/dx)^2 or d2u/dx2 as you state in your original question?

If it is the latter we do integrate that by part in several demos, Poisson for
example. But you have u^3*d2u/dx2 which makes the integration by parts non
trivial.

You need to get that correct in addition to Garth's suggestions, I think.

Johan

Thank you for both of your suggestions.
Dr. Wells: you meant the concomitant term from integration by parts might not be zero, but I implicitly set to zero. Do I understand it correctly?

 What I did for the B.C. was that I made the domain big enough and set 0 for B.C. at both ends. I understand u^3*d2u/dx2 need to be taken integration by parts, but according to my BC setting, the concomitant term is indeed zero.

Thank you again for your time to read my problem. Let me describe it in details.

I want to use this evolution equation to simulate the thin film flow on an incline. The PDE is derived using 2D Navier-Stokes equation, coupled with continuity equation and kinematic condition at free surface. See Acheson's book for detailed derivations.

Acheson, D.J., Elementary Fluid Dynamics. Oxford University Press, 1990.

In this 1D PDE,

du/dt+3*lmbda*u^2*du/dx*(sina-cosa*du/dx)-lmbda*u^3*cosa*d2u/dx2=0

u is the height of the free surface. I use a parabola “cap” as the initial condition
u=0.3[1-〖(x-2)〗^2] for1<x<3
u=0 for the other domain

To make the domain wide enough that in a short time period, the flow doesn't cross the boundary. I set the domain mesh = Interval(1000,0,16), at least in 12 sec, the flow won't touch the boundaries. Hence, u=0 at both ends all the time, which is also my BC.

According to the derivations, if I integrate u over the domain, I should get the mass of the fluids, it shouldn't change with time due to the conservation of mass. But the fact is that it grows for my Fenics code. I couldn't find any problem in my model so far.

I just start to learn the finite element processes, not very good at it. But I really have lots interests on this Fenics package. Any helps will be appreicated.

On 30/04/10 17:58, cutejeff wrote:
> Question #107819 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/107819
>
> Status: Answered => Open
>
> cutejeff is still having a problem:
> Thank you for both of your suggestions.
> Dr. Wells: you meant the concomitant term from integration by parts might not be zero, but I implicitly set to zero. Do I understand it correctly?
>

Yes.

Garth

> What I did for the B.C. was that I made the domain big enough and set 0
> for B.C. at both ends. I understand u^3*d2u/dx2 need to be taken
> integration by parts, but according to my BC setting, the concomitant
> term is indeed zero.
>
> Thank you again for your time to read my problem. Let me describe it in
> details.
>
> I want to use this evolution equation to simulate the thin film flow on
> an incline. The PDE is derived using 2D Navier-Stokes equation, coupled
> with continuity equation and kinematic condition at free surface. See
> Acheson's book for detailed derivations.
>
> Acheson, D.J., Elementary Fluid Dynamics. Oxford University Press, 1990.
>
> In this 1D PDE,
>
> du/dt+3*lmbda*u^2*du/dx*(sina-cosa*du/dx)-lmbda*u^3*cosa*d2u/dx2=0
>
> u is the height of the free surface. I use a parabola “cap” as the initial condition
> u=0.3[1-〖(x-2)〗^2] for1<x<3
> u=0 for the other domain
>
> To make the domain wide enough that in a short time period, the flow
> doesn't cross the boundary. I set the domain mesh = Interval(1000,0,16),
> at least in 12 sec, the flow won't touch the boundaries. Hence, u=0 at
> both ends all the time, which is also my BC.
>
> According to the derivations, if I integrate u over the domain, I should
> get the mass of the fluids, it shouldn't change with time due to the
> conservation of mass. But the fact is that it grows for my Fenics code.
> I couldn't find any problem in my model so far.
>
> I just start to learn the finite element processes, not very good at it.
> But I really have lots interests on this Fenics package. Any helps will
> be appreicated.
>

Thank for the helps.

As I said, the boundary conditions setting forced the the concomitant term from integration by parts be zero, so there shouldn't be any problem on B.Cs.

I still couldn't find where the problem is. I appreciate the suggestions. Any other thoughts about the mistake I could make?

Thank you.

Have you figured out the integration by part?

Johan

On Monday May 3 2010 07:28:06 cutejeff wrote:
> Question #107819 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/107819
>
> cutejeff posted a new comment:
> Thank for the helps.
>
> As I said, the boundary conditions setting forced the the concomitant
> term from integration by parts be zero, so there shouldn't be any
> problem on B.Cs.
>
> I still couldn't find where the problem is. I appreciate the
> suggestions. Any other thoughts about the mistake I could make?
>
> Thank you.

Johan:
Thank you for your reply.
For you last question, I have both (du/dx)^2 and d2u/dx2 terms in my PDE. I don't do integration by part for (du/dx)^2, I just make it multiply test function v. For d2u/dx2 term, I take integration by part as shown below,

d2u/dx2*v*dx=-du/dx*dv/dx*dx+du/dx*v@BC

Since I set u=0 at the boundaries, the concomitant term du/dx*v@BC is zero.

I think the FE processes on the power of the derivative may have some problem. I will open a question and make an easier example to describe it.

I close this one. Please refer to my new question. I appreciate the helps.