asalmgren
commented
4 years ago There are several strategies to try -- some in place and some we are actively developing. One of the things that may be affecting your convergence is whether the EB constrains how many multigrid levels you are using.
But I am somewhat confused by your description so let me address it in parts.
First -- you say the projection makes vel and gradp 0 -- is that just that it starts at rest therefore doesn't do anything, or are you saying it starts with a non-divergent field but zeroes the output? The former makes sense, the latter doesn't.
To diagnose convergence issues, it would be helpful to turn on more verbosity for the nodal solve -- could you re-run with nodal_proj.verbose = 4 and nodal_proj.set_verbose = 4?
On Mon, Jun 22, 2020 at 9:35 AM dcoveney notifications@github.com wrote:
Hi,
I encounter errors when I try and include EB geometry into a problem with large density ratios (816:1).
The issue seems to be with the nodal projection - it makes the velocity and gradp fields 0, which leads to an inability to obtain a time step estimate:
NavierStokes::advance(): before nodal projection max(abs(u/v)) = 0.2998059427 0.2482103605 max(abs(gpx/gpy/p)) = 13270.61685 16865.77065 1158.284095 ... Projection::level_project() at level 0 Projection::level_project(): lev: 0, time: 0.142975462 NavierStokes::advance(): after velocity update max(abs(u/v)) = 0 0 max(abs(gpx/gpy/p)) = 0 0 0
...
NavierStokesBase::estTimeStep() failed to provide a good timestep (probably because initial velocity field is zero with no external forcing). Use ns.init_dt to provide a reasonable timestep on coarsest level. Note that ns.init_shrink will be applied to init_dt. amrex::Abort::0:: !!!
If I change parameters to do with the projection, e.g. the bottom solver or tolerances, the projection fails outright:
Projection::initialVelocityProject(): iteration 0 After nodal projection: lev 0: max(abs(u,v)) = 0 0 Projection::initialVelocityProject(): time: 0.002017629 done calling initialVelocityProject calling initialPressureProject Projection::initialPressureProject(): levels = 0 0 amrex::Abort::0::MLMG failed !!! SIGABRT
The projection issues seem to occur when the more dense region of fluid interacts with the EB. I do not get this issue when I run with a lower density ratio (I've done it with 10:1 but haven't found the cut-off where it stops working).
I have tried tweaking things such as num_pre_smooth, which works for challenging problems without EB but not when I include geometry.
If there are any suggestions for things to try they would be most welcome. Is this a known limitation of IAMR with EB? I haven't found any examples of EB cases that use larger density ratios.
Thanks!
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/AMReX-Codes/IAMR/issues/46, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACRE6YWQWWQ55ZH3YLBN7DLRX6B4DANCNFSM4OE3BKNA .
-- Ann Almgren Senior Scientist; CCSE Group Lead
dcoveney
Hi,
I encounter errors when I try and include EB geometry into a problem with large density ratios (816:1).
The issue seems to be with the nodal projection - it makes the velocity and gradp fields 0, which leads to an inability to obtain a time step estimate:
If I change parameters to do with the projection, e.g. the bottom solver or tolerances, the projection fails outright:
The projection issues seem to occur when the more dense region of fluid interacts with the EB. I do not get this issue when I run with a lower density ratio (I've done it with 10:1 but haven't found the cut-off where it stops working).
I have tried tweaking things such as
num_pre_smooth, which works for challenging problems without EB but not when I include geometry.If there are any suggestions for things to try they would be most welcome. Is this a known limitation of IAMR with EB? I haven't found any examples of EB cases that use larger density ratios.
Thanks!