Meeting update

MHD Shear Flows

  • Need to decide on field orientation for the 'perpendicular' case. The Beta=1 effects for a uniform field in y or z are too strong (see below). Talking with Jonathan on this, it seems a highly unrealistic initial field orientation to begin with — how do you have this large scale perpendicular field and a tilted interface to begin with? If there was some large scale perpendicular field, then by the time the flows collided, they would have no tilted interface. A more realistic perp. case might be a helical field inside of the colliding flows only (no ambient field)? This would simulate a perpendicular field, but without concerning the broken symmetry played by the tilted interface. For now though, am running 3D low res sims of the y and z fields with a weaker field.
  • The data set for these runs as they are now, is extremely large. I am at ~ 120 GB now, ¼ of the way through the production Shear15, Beta=1, x-field run. This is for ~ 15003 cell resolution (48 + 5 levels). I estimate each run will be ~ 1TB in size. Where to store all this data?
  • Stuff still need to finalize for production runs: 1) Potentially increasing the box size to prevent back-flow into the outflow-boundary. This most likely will be a problem in the Shear 30 and 60 cases (see my page). Can do this either by adding code to enable a restart with a bigger box once the back-flow hits the wall, or just run sim with a longer box in x to start. 2) Decide on parameters for the flow, i.e. the mach, ramp down times, etc. for production.

3D High Res Colliding Flows

I ran a Beta = 1, Shear = 15, field = parallel case on BS, 8000 cores. It got ~¼ of the way through. It ran into some bad cfl numbers (at first, finite, then cfl=infinity) and src errors, requested restarts, and then made a few more frames until it died with the error:

"hyperbolic/sweep/stencil_control.cpp.f90, line 353: 1525-108 Error encountered while attempting to allocate a data object. The program will stop."

I'm attaching log files from machine to this blog post.

The global info allocations are at 300 GB at frame 57. The Info allocations didn't reach that size until frame 160 in the Hydro run on BS.

I'm curious how the src errors are mitigated

I also am curious about this - I saw that on BH (~64 cores), the info allocations in the standard out was 10x larger than the size of the chombo. This we talked about before as being related to not counting ghost zones. When I moved to BS, the difference was a factor of 100. Is this likely ghost zones again, this time the larger factor being due to the many more smaller patches being distributed over 8,000 cores?

Here are density and mesh images over time:

Everything looks good..

3D Low Res Perpendicular Field Cases

I ran both the y-field and z-field cases. This is a uniform field that runs through the ambient object and the CF object, perpendicular to the CFs. Here is a movie of the y-field case. This is density with the B vector field. Note the flow is colliding in x. Beta = 1.

We see that the oncoming flows cause the field to begin bending strongly, especially near the x faces of the box. The incoming flow is then directed along these field lines, squashing the material into a pancake in the xz plane. The field is pushed together by the flow, and piles up strongly in the collision region, where the magnetic pressure becomes so strong it vacates/resists gas from collecting there. Hence, the vacated cavity in the collision region, traced out by billowing shell of material traveling in y, away from the colliding flows.

The field is so strong in this case, that no interesting features of the flow are coming through. The NTSI for example is completely wiped out.

movie​

I also made column density maps of this case and the 3D low res z-field case and posted them on my page here.

I am running these same simulations now with Beta = 10 and with extrapolating/outflow only boundary conditions on all faces of the box, since the problem with the Bfields we saw before only happens when the Bfields are normal to the outflow boundary. The cases above had reflecting, Bnormal boundaries where the field was coming from, but in trying to avoid the bounce off of the boundary, changing BC's now too. Directories:

/bamboodata/erica/CollidingFlows/CollidingFlows/MHD_3D/Beta10/Shear15/

They are ~¼ way through on bamboo and grass and take ~ 20 hours to complete.

Attachments (12)

Comments

No comments.