3
$\begingroup$

This question is a follow-up of this one.

The weak form of Navier Stokes equation is (assuming $v,q$ test functions for the velocity and the pressure, respectively)

$$(\frac{du}{dt},v)_{\Omega} + (\nabla u, \nabla v) _{\Omega}- (\operatorname{div}(v),p)_{\Omega} - (q,\operatorname{div}(u))_{\Omega} + c(u,u,v)_{\Omega}= (v,f)_{\Omega}$$

Now I've been told to advance in time using a time integrator (so I really need to use some available routine like the ones provided by SUNDIALS, instead of doing the time-stepping on my own). The system above becomes the following one:

$$M \dot{u} = f- Au -B^tp - C(u(t))u(t)$$ $$B u(t)=0$$

but I really do not see the usual form of an ODE. (I was thinking to a DAE, but my professor explicitely told us that we don't have to solve a DAE). I'm using deal.II, so I need to use the SUNDIALS wrappers to advance in time. My biggest problem is that I can't "see" that system as a system of ODEs. What am I missing?

Improve this question
$\endgroup$
5
  • $\begingroup$ If you don't want to solve this as ODE+DAE, then the only other option is to use fractional-step (or projection) methods. See this tutorial from deal.ii. dealii.org/current/doxygen/deal.II/step_35.html $\endgroup$
    – Chenna K
    Jul 5, 2021 at 10:12
  • $\begingroup$ If you make that run with SUNDIALS, it would be very nice to eventually get that program into the deal.II code gallery! $\endgroup$
    – Wolfgang Bangerth
    Jul 5, 2021 at 16:57
  • $\begingroup$ @WolfgangBangerth I'd be happy to share it, of course. The only thing I need to clarify is how I can use arkode interface. I mean, $$M \dot{u} = f- Au -B^tp - C(u(t))u(t)$$ is certainly in the right form, but with the incompressibility constraint I don't have an ODE anymore. $\endgroup$
    – Vefhug
    Jul 5, 2021 at 18:36
  • $\begingroup$ The N-S equations are definitely a DAE, but I can't say whether one would solve it as a DAE when using SUNDIALS. The methods I'm familiar with are of projection type, and they generally are built using hand-written time integration loops. But that may not be the most efficient way. If you have the time, I would just give it a try with SUNDIALS' IDA integrator and if that doesn't work, try a different approach. $\endgroup$
    – Wolfgang Bangerth
    Jul 5, 2021 at 22:19
  • 1
    $\begingroup$ Also, ask your professor why he thinks that you shouldn't see the equation as a DAE -- teaching is one of the jobs of professors :-) $\endgroup$
    – Wolfgang Bangerth
    Jul 5, 2021 at 22:19

1 Answer 1

Reset to default
1
$\begingroup$

I don't see what you're not seeing. You've written out a perfectly good non-linear weak form of the PDE in your last two equations. If you invert $M$ and apply it, then you have a non-linear equation for $\dot{u}(t)$ that you can try to solve. It's a continuous in space and time PDE that you would need to discretize in space to then have only an ODE, with $M$, $C$, $A$, and $B$ being space/time operators that need spatial discretization in space by introducing some finite element grid/mesh functions (e.g., $u_h(t)$) to generate a method that is only an ODE in time. Also, you need some sort of initial conditions (ICs) for $u_(t_0)$. You also need boundary conditions on $\Omega$ for $u_h(t)$. You also need to linearize $c$ so that you end up with a solvable system. My notation is a little loose here, but I think all the ideas are there.

Improve this answer
$\endgroup$
6
  • $\begingroup$ The fact is that I need to advance in time using SUNDIALS wrappers (dealii.org/current/doxygen/deal.II/classSUNDIALS_1_1ARKode.html). Yes, the first equation is an ODE, while the second contraint $Bu=0$ makes the system a DAE. (here with you $u$ I actually means $u_h$, I already applied a FE discretization in the last equations). Therefore I cannot see how I can use that interface. $\endgroup$
    – Vefhug
    Jul 5, 2021 at 17:49
  • $\begingroup$ Is this a homework assignment? $\endgroup$
    – Bill Barth
    Jul 5, 2021 at 17:51
  • $\begingroup$ Honestly, it's a bonus point for a project. But I'm doing this mainly for my future research. @BillBarth $\endgroup$
    – Vefhug
    Jul 5, 2021 at 18:17
  • $\begingroup$ @Vefhug, it looks like to me everything you need is in the deal.ii tutorials and lectures. Professor Bangerth's videos about this are on YouTube as well. In your case $M$ is $M$ (i.e. the mass matrix) and you're going to have to figure out your Jacobian, but the pieces are basically there. $\endgroup$
    – Bill Barth
    Jul 5, 2021 at 18:46
  • $\begingroup$ So, are you going to solve only $$M \dot{u} = f- Au -B^tp - C(u(t))u(t)$$ ? I'm really sorry to keep asking, but I looked up all the deal.ii tutorials in the past days, and they never use an ODE solver to solve the time dependent NS equation. $\endgroup$
    – Vefhug
    Jul 5, 2021 at 19:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.