[aja1345]

Hi dear friends,

I am simulating a turbine. When I start Simulation the message error appears "Error: Divergence detected in AMG solver: pressure correction
Error: Divergence detected in AMG solver: pressure correction
Error Object: #f" and I´m not able to solve it.

Boundary conditions:
- inlet: velocity-inlet (direction y)
- outlet: pressure-outlet=0

mesh is structure and good. I selected inlet velocity for Initial conditions.
the fluid is incompressible(constant density) and the flow is steady and The working fluid was air.

Thanks,

Best,

Aja

[arjun]

The error message was certainly very interesting

Fluent by default uses correction scaling, so my question is are you or are you not running simulation in parallel. I guess it was single processor solve.
(I am just curious here).

About the divergence, the clue is the turbulent viscosity ration which is limited to 1e5 or so. Since it diverged at very first iteration, i would suggest use better initial guess for turbulence quantities so that initial turbulent viscosity is good to go.

[aja1345]

Originally Posted by arjun

The error message was certainly very interesting

Fluent by default uses correction scaling, so my question is are you or are you not running simulation in parallel. I guess it was single processor solve.
(I am just curious here).

About the divergence, the clue is the turbulent viscosity ration which is limited to 1e5 or so. Since it diverged at very first iteration, i would suggest use better initial guess for turbulence quantities so that initial turbulent viscosity is good to go.

Thanks your answer,

What is the correction scaling? can you explain more?

Yes, I am running simulation in parallel.

I am grateful that guide me.

[arjun]

Correction scaling is when during the prolongation phase of AMG corrections are multiplied by some scaling factor (typically between 1 to 2) to speed up convergence. It is kind of like an over-correction scheme.

StarCCM by default had them switched off (early versions of starccm had them on by default if i am right).
Current version of starccm seems to put it to auto and this is possibly off when in parallel. The justification is that calculating this factor requires two all to all exchange every time prolongation is made and so this cost adds up as you add more processors.

So i was thinking may be fluent also does the same.

Second thing is that in your error message fluent writes that when amg diverged , it switched off these corrections. This is because sometimes this makes AMG unstable so there is no perfect way to apply these scalings.

Stueben in his book mentions that because of this reason scaling shall be used when AMG is used as preconditioner so that even when amg fails to converge, the main linear system is not diverged. It gets more chances to recover.

Originally Posted by aja1345

Thanks your answer,

What is the correction scaling? can you explain more?

Yes, I am running simulation in parallel.

I am grateful that guide me.