search

mathLab / pi-DoMUS

Parallel Deal.II MUltiphysics Solver
http://mathlab.github.io/pi-DoMUS/
GNU Lesser General Public License v2.1
16 stars 16 forks source link

New interface is broken #86

Closed luca-heltai closed 8 years ago

luca-heltai commented 8 years ago

The new interface requires the function Interface::Init() to be called, but pi-DoMUS does not call it anywhere... I think the content of the Init function should end up directly in the constructor.

Time dependency of the IMEXTimeStepper seems to be broken in pi-DoMUS

See branch:

https://github.com/luca-heltai/pi-DoMUS/tree/scalar-reaction-diffusion

If you put the time dependence in the equation (we tried this with @gpitton ), nans appear in the matrix, and everything fails...

It suffices to put in the residual ut*v instead of u*v to see this

asartori86 commented 8 years ago

The new interface requires the function Interface::Init() to be called, but pi-DoMUS does not call it anywhere... I think the content of the Init function should end up directly in the constructor.

which constructor? we may just call that function within the constructor of pi-DoMUS.

Time dependency of the IMEXTimeStepper seems to be broken in pi-DoMUS

is this an issue of the IMEXTimeStepper itself (i.e., we need to address it in deal2lkit) or pi-DoMUS?

luca-heltai commented 8 years ago

Why do we need a separate init() function? I'm in favour of putting its content in the constructor of BaseInterface

luca-heltai commented 8 years ago

I don't know if the issue is with IMEXTimeStepper, or with pi-DoMUS...

asartori86 commented 8 years ago

Why do we need a separate init() function? I'm in favour of putting its content in the constructor of BaseInterface

It calls virtual functions so it cannot be inside the constructor of the base class

asartori86 commented 8 years ago

I don't know if the issue is with IMEXTimeStepper, or with pi-DoMUS...

what happens with ida?

asartori86 commented 8 years ago

@luca-heltai @gpitton, I tried a simple poisson equation time-dep with both ida and euler and it seems to work.

ESeNonFossiIo commented 8 years ago

I have problems with both euler and ida. Old tests do not work with the new interface. Sometimes it seems to work and it breaks after a refinement.

asartori86 commented 8 years ago

Sometimes it seems to work and it breaks after a refinement.

this is because of #72

asartori86 commented 8 years ago

I have problems with both euler and ida. Old tests do not work with the new interface.

what kind of problems?

ESeNonFossiIo commented 8 years ago

I am not using adaptivity and old_solutions.

On 07 Dec 2015, at 16:29, Alberto Sartori notifications@github.com wrote:

I have problems with both euler and ida. Old tests do not work with the new interface.

what kind of problems?

— Reply to this email directly or view it on GitHub https://github.com/mathLab/pi-DoMUS/issues/86#issuecomment-162557688.

ESeNonFossiIo commented 8 years ago
  1. No convergence in tests I wrote for the old interface.
  2. Sometime it requires a lot of time (10 minutes) for one step of a system (187Dof)

On 07 Dec 2015, at 16:29, Alberto Sartori notifications@github.com wrote:

I have problems with both euler and ida. Old tests do not work with the new interface.

what kind of problems?

— Reply to this email directly or view it on GitHub https://github.com/mathLab/pi-DoMUS/issues/86#issuecomment-162557688.

ESeNonFossiIo commented 8 years ago

Now, my tests are ok. Some parameters needed to be tuned...

ESeNonFossiIo commented 8 years ago

I have pushed tests for stokes and dynamic stokes. These tests looks good and works. I noticed some strange behaviour with direct method with non time depending problems: if you use the right solution as initial solution, the method will pass. Otherwise it does not. Am I doing something wrong? Or did I not consider something? Euler and Ida seem ok.