Hello,
thanks for you reply. I was not aware about the Llama code. I will have a look at this and also at the overlap zones, if it will help.
Best regards, Petra
-----Původní zpráva----- From: Philipp Moesta Sent: Wednesday, March 9, 2016 6:46 PM To: users@einsteintoolkit.org Cc: psukova@cft.edu.pl Subject: Re: [Users] Problem with accretion of magnetised gas and mesh refinement
Hi Petra,
to add to this. The problem you are describing sounds very familiar to me. I had faced similar issues in simulating magnetized core-collapse with AMR.
One thing that helped me a lot was utilizing additional buffer zones (called 'overlap' zones). Using these seem to effectively decouple the divB violations from the bulk of the grid when using constrained transport.
The relevant parameter is
Carpet::use_overlap_zones = "yes"
All that said, you may want to follow Erik's suggestion and use multi-block for simulating gas accretion onto a black hole.
Best, Philipp
On 3/9/16 7:23 AM, Erik Schnetter wrote:
Petra
Without going into the details of the issues you are encountering:
Accretion onto a black hole has a near-spherical or axisymmetric symmetry. In my experience, this is better modelled using a multi-block structure than mesh refinement. Are you aware of http://llamacode.bitbucket.org?
-erik
On Wed, Mar 9, 2016 at 10:10 AM, Petra Suková psukova@cft.edu.pl wrote:
Hello everyone,
I am a postdoc at CFT PAN Warsaw and a relatively new user of ET. I am trying to use ET for simulating accretion of low angular momentum magnetised gas onto Kerr black hole.
I use GRHydro and GRHydro_Init thorns on stationary Kerr spacetime. Using mesh refinement with nonzero magnetic field always leads to crash of the computation, after few hundreds or thousands of iterations. It looks like the cause of the crash is violation of divB=0 constrain in the buffer zones on the edge of refined level, which grows and later starts to propagate into the grid.
More details: I have adapted GRHydro_BondiM.c and run some tests with low angular momentum accretion, B=0, spin=0 and several refinement levels, which after some modifications worked quite well. Now I am trying to include magnetic field. I have set initially vertical magnetic field. After first iteration, divB in the buffer zones increases to values around 10e-7 and it is growing during the evolution. After some time (depending on the Bfield strength, grid size etc, typically tens or hundreds of M), the values of divB in the buffer zones exceed 0.1 and start to spread inside the grid zones and spoil the simulation. I thought, that maybe some of my changes caused this, so I tried to run an example par file MBondi_B5.70_Mdot12.57_Npts100_Sch.par with clean ET release 12, but that crashed immediately in the first iteration. So I tried to lower bondi_bmag from 5.7 to 0.1 and this worked and was stationary, but it runs only on unigrid without refinement. When I added the mesh refinement and fixed GRHydro_BondiM_Boundary to not act on the refinement boundary, the same problem with buffer zones appears. I am attaching the par file of this computation together with few snapshots of divB, rho and Bvec. My guess is that during the prolongation and restriction of magnetic field vector the DivB=0 condition is violated and this error is continuously increasing when the Bfield and matter is accreted through the refinement boundary. Has anyone faced similar problem? Or does anyone have some working parfile with similar computation with magnetised gas and at least two refinement levels, which he can send me? I could test that and hopefully discover my problem.
Thanks a lot, Petra
Tato zpráva byla zkontrolována na viry programem Avast Antivirus. https://www.avast.com/antivirus
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users