[FLASH-USERS] MHD staggered mesh (Flash4a)

[dongwook at flash.uchicago.edu]

Hi Seyit,

> Dear Flash developers,
>
> I am testing the USM solver a bit in Flash4-alpha and noticed some
> curious features. I use my standard setup, which I normally use with
> split PPM, and do +usm in the setup command. Everything looks and runs
> fine, however, whenever I increase the number of processors, dt_hydro
> seems to drop by an equal amount. I do not use super-time-stepping,
> which I don't understand yet.

We haven't seen this issue with large number of processor runs with the
USM solver. I'd like to check what kind of problem you're trying to setup
and seeing this problem. One quick note is that, in case you have your own
MHD simulation, you NEED to put that under

source/Simulation/SimulationMain/magnetoHD/YourMHDsimulation

in order to preserve divB=0 condition on a staggered grid.

It is best if you start from those MHD simulations provided in /magnetoHD/
directory,
copy and save it to another name, and start modifying
Simulation_initBlock.F90, etc.

It sounds like you probably ran a simulation that is in
Simulation/SimulationMain/
and setup up once with PPM and then with USM by simply adding +usm.

This won't work, especially for USM, and will need to setup B-field on a
staggered
grid correctly to start with divB=0 condition.

> Another question I have is that if it is normal that USM MHD requires so
> much more (ram) memory than PPM. It is about 3-4 times more memory
> intensive. I noticed that there are much more (about 3 times more)
> variables than in PPM. These are mainly the scratch and the flux
> variables. So, is it normal that USM requires this much memory?

Yes, it is true that the memory requirement in USM is higher that any split
solvers. The expense comes from the fact that USM being an unsplit solver,
plus there are additional variables for MHD including B-fields & electric
fields in the
USM MHD formulation.

Of course, these variables are not needed in gas dynamics
solver. And the PPM solver is a split solver, therefore, it does not require
transverse fluxes in each directional sweep (i.e., no need y,z fluxes in
x-sweep).
This is not true for all directionally unsplit solvers.

Because of this fact, the USM solver needs to store all fluxes in x,y, and
z directions,
requiring more memory.


> Lastly, I am new to magnetic field studies and thus not so familiar how
> to implement them. The way I implement magnetic fields is as follow: I
> use my standard setup +usm and in Simulation_initBlock, I give
> reasonable values to center values of MAGX/Y/Z and also do the same for
> the face values of MAG (facex, facey, facez), similar to how it is done
> in the supplied test runs. My run differs from these test runs in the
> fact that I have gravity (and particles) included. I'm curious how the
> magnetic fields will be amplified/weakened over time. I'm also wondering
> why it is not necessary for the code to know the ion/electron abundance.
> Is this assuming some flux freezing state?

As I mentioned in the above, you should be very careful in initializing
B-fields
for the USM solver. You need to setup the fields that satisfy divB=0 on a
staggered
grid. If you use constant B-fields then it is going to be straightforward,
otherwise,
you will also need to initialize facex, facey, facez (face-centered)
differently from
MAGX/Y/Z (cell-centered) because face centers are dx/2 away from cell
centers.

Please have a look at the initialization of the fields in CurrentSheet
problem, if your
B-fields are complicated. Or if you know your analytic forms of B-fields,
then please
have a look at Simulation_initBlock in OrszagTang to help you increase
your understanding.

Magnetic energy will eventually decrease if you don't have any explicit
source terms such
as B-field injection as a boundary condition, or biermann battery term for
dynamo effect.
In ideal MHD, magnetic energy will decrease because of numerical
diffusivity on a grid
scale, and in resistive MHD, magnetic diffusivity (resistivity) will
diffuse out the strength
of the fields.

Best,
Dongwook

>
> Kind regards,
> Seyit
>