Linear Solvers: Poisson problems with piecewise constant coefficients and interface jumps

[gdeskos]

Dear all,

I am an amex user/developer working within the (AMR-Wind project ). For the last six monthsI have been developing the multiphase capabilities of AMR-WIND in order to enable water waves and free-surface turbulence simulations. One of the main problems I am facing at the moment is dealing with linear solvers, particularly in the vicinity of the two-fluid interface. Such problems occur both during a MAC projection and a Nodal projection (poisson solve). MLMG really struggles with solving these problems particularly when the density ratio between the two fluids is rho1/rho2>100.

A possible solution would be to use AMG (using for example Hypre) but this is probably going to really slow down the simulations. I was therefore wondering if there is an easy hack within the native linear solvers. The problem really boils down to \beta (the non-constant coefficient - see here) being a piecewise constant coefficient (1/\rho1 in fluid 1 and 1/\rho2 in fluid 2) and the source term containing a jump at the interface.

Please let me know if someone has already worked on this, or/and have already found a solution. If not I think would be a very useful addition to the existing suite of linear solvers.

Best wishes,
Georgios (Yorgos) Deskos
Postdoctoral Research
National Renewable Energy Laboratory

[sayerhs]

@gdeskos Recently there were a few updates to AMReX hypre interface (see #1437 and #1439) that allow you to use hypre solvers and preconditioners from AMReX applications. If you have compiled AMR-Wind with HYPRE support -DAMR_WIND_ENABLE_HYPRE=ON, then you can try out the different hypre options. See the example input file, also see Exawind/amr-wind#221. It would be a good idea to check if even hypre can solve this issue or not (even if it is slow) or if some thinking of the algorithm.

[asalmgren]

@gdeskos --I believe this issue can be closed -- can you confirm?

[gdeskos]

yep, we can close this issue !