[FLASH-USERS] Magnetized Inflow: Torus Problem

[Ryan Farber]

Dear Flash users,

Sorry for the spam. Hopefully this will help someone in the future.

Setting BCs to outflow rather than periodic (with xl_boundary_type =
"user") appears to have fixed the problem (initialization and one evolution
step completes successfully).

Also, the reason I was seeing minval(unk(DENS_VAR,:,:,:, :)) = 0 even at
the very start of mpi_amr_guardcell is because the last index of unk is
blocks which includes values for all MAXBLOCKS, with unused block values
set to zero.

I remedied that by checking minval(unk(DENS_VAR, :,:,:, 1)) and found the
issue I was having with xl_boundary_type = "user" and other BCs set to
periodic is because unk1 is set to zeros in mpi_amr_1blk_guardcell whereas
that doesn't happen if the other BCs are set to outflow.

I should also mention that my write statements at the very start of
Grid_bcApplyToRegionSpecialized were executed for mostly outflow BCs
whereas they were not for mostly periodic BCs.

So, if someone does need to mix periodic with user BCs I suspect the issue
is ultimately in mpi_amr_1blk_guardcell.

However, I don't plan on investigating further since outflow BCs are
preferable for my problem to periodic.

Best,
--------
Ryan


On Sat, Feb 2, 2019 at 3:44 PM Ryan Farber <rjfarber at umich.edu> wrote:

> Dear all,
>
> I have not fixed the issue of zero density values within Eos yet but would
> like to share the following update and would be very grateful for help in
> understanding why unk (in the physicaldata module) would have all zero
> values despite solnData receiving values in Simulation_initBlock.
>
> First, I should mention that the error occurs in eos_idealGamma which is
> called by Eos -> Eos_wrapped -> Eos_guardCells -> Grid_fillGuardCells ->
> Grid_markRefineDerefine -> gr_expandDomain -> Grid_initDomain ->
> Driver_initFlash
>
> At first, I found that the minimum value of density was zero at the end of
> Simulation_initBlock because the loop setting density did not include guard
> cells. So, I modified that (replaced blkLimits with blkLimitsGC for the
> loop start and stop values) and found that the minimum density was fine
> until after the call to amr_guardcells in Grid_fillGuardCells.
>
> Within Grid_fillGuardCells, just before the call to amr_guardcells I found
> that minval(solnData(DENS_VAR,:,:,:) was the value I set for all (5)
> blocks. However, the very start of amr_guardcells (within
> mpi_amr_guardcells.F90) has minval(unk(DENS_VAR,:,:,:,:) = 0.0 (and also
> all zeros for all other variables).
>
>
> Additionally, the end result of amr_guardcells is to set the density =
> -1.0 for the parent block while setting the density to zero for the leaf
> blocks.
>
> This is undesirable yet I am not sure how to proceed. I'd be very grateful
> for any suggestions or ideas people may have.
>
> Best
> --------
> Ryan Farber
> PhD Candidate, University of Michigan
>
>
> On Sat, Feb 2, 2019 at 1:12 PM Ryan Farber <rjfarber at umich.edu> wrote:
>
>> Dear FLASH users,
>>
>> TL;DR: Setting xl_boundary_type = "user", cells are filled with zero
>> density although Grid_bcApplyToRegionSpecialized was not yet called. There
>> is no such problem with all periodic BCs. If someone familiar with user BCs
>> could mention which additional routines are executed that I might want to
>> look into to debug this problem I'd very much appreciate it.
>>
>> =========================================================
>>
>> I am trying to modify the Torus problem included in FLASH4.5 to include a
>> magnetized inflow from the lower x-boundary.
>>
>> So, I modified flash.par to turn off gravity, comment out gr_prmp*
>> variables, set geometry to cartesian, lrefine_min = 3, and set all BCs to
>> periodic except xl_boundary_type = "user".
>>
>> Additionally, I modified Simulation_initBlock so the domain is initially
>> at a uniform density and temperature with zero velocity and zero magnetic
>> field.
>>
>> When I run the simulation with all BCs set to periodic, I receive many
>> warning messages about the density being the value I set (modulo roundoff
>> error) with a lot of output (possibly printing the entire density array?);
>> however, it runs to completion fine and plots show the domain remained
>> uniform (as to be expected).
>>
>> However, when I set xl_boundary_type = "user" then the warning messages
>> say minimum density is -1.0 and it crashes with:
>>
>> ERROR After calling Eos, eosData(EOS_EINT) or eosData(EOS_DENS) are zero
>>
>> Compiling with -debug I receive a "floating invalid" error with traceback
>> to eos_idealGamma.F90 with a line in which there is a division by density
>> (which is zero, triggering the floating invalid error). However, I have
>> write statements at the start of Grid_bcApplyToRegionSpecialized that are
>> never executed suggesting that routine is not called yet.
>>
>> I am seeking to eliminate this zero density issue; I'd be very grateful
>> for any pointers those familiar with user BCs could point out.
>>
>> Best,
>> --------
>> Ryan Farber
>> PhD Candidate, University of Michigan
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20190202/505af425/attachment-0001.htm>