[yt-users] integrated quantities in cylindrical

[Matthew Turk]

Hi Roman,

On Tue, Jun 18, 2013 at 6:25 PM, Roman Yurchak <rth at crans.org> wrote:
> Hi Suoqing,
>
> On 18/06/13 19:00, Suoqing Ji wrote:
>> I assume the quantity 'CylindricalCellMass' is a derived field defined by yourself? I usually define the 2d cylindrical cell volume as followings:
>> [...]
> yes, that's pretty much the expression I was using.
>
>> And for my data set, the output given by YT (use dd.quantities['TotalQuantity']('CylCellMass')[0]) and FLASH subroutine (IO_writeIntegralQuantities) match down to the machine accuracy.
>
> ahh, thank you for your response! It made me realize two things:
>
> (a) I made a stupid mistake while sorting different time steps, and for
> a normal 2d cylindrical FLASH simulation I do have convergence to
> machine precision between the two methods.

Ah!  Good, that is great to hear.

>
> (b) I was post-processing a simulation that has a few coarse blocks
> removed and used as new boundary conditions in order to get a
> non-rectangular domain (see
> source/Simulation/SimulationMain/WindTunnel/Simulation_defineDomain.F90
> in FLASH source, or the illustration here:
>   http://perso.crans.org/yurchak/i/illustations_yt.pdf
>  ). This seems to be the origin of the issue I've been experiencing (and
> that's still there).

I think I see what's happening here.  There are a few diagnostics to
run to see if we can understand and then correct yt's understanding of
the mesh; basically, yt will construct a full mesh based on what it
sees in the file, and we need to make sure that the mesh matches what
is being evolved in the simulation.  For instance, it's not clear to
me completely whether or not the block that has been removed exists in
the mesh described in the FLASH output file, and if so, if we can
simply remove it.  We may be able to do this directly via the FLASH
frontend.  I do not think so, but there may be assumptions in the
FLASH frontend that the system resembles the typical refinement
pattern for FLASH, which this may not match.

In the past we've been able to do funny things to the hierarchy via
the Stream frontend -- interposing IO and whatnot, mocking up new
hierarchies -- which I would be happy to make work with this output if
we can't make the FLASH frontend work exactly as desired.

>
> I looks like yt, treats this configuration partly ok as for instance
> none of leaf blocks has for parent a block that was removed. On the
> other side though:
>   * there this issue with d.quantities['TotalQuantity']
>   * a slice shows that there is some plasma propagating through the
> region that's supposed to be treated as outside of the simulation (with
> removed blocks) and there is nothing there in VisIt.
>
> So, anyway, thanks for narrowing it down.
>
> Would you have any suggestions how to deal with this issue?

So let's start out by taking a look at a few things.  These are some
common diagnostics to figure out whether or not a hierarchy is
"well-posed" in yt.  One thing I am wondering is if the non-regular
cutout is causing irregular problems -- specifically, I am thinking
that it may not be "masked" in the parent grids with which it
overlaps.  So those operations that utilize "masking" information will
not know about the missing block, but operations that *don't* will
not.

I wrote up some simple diagnostics that I *think* will be sufficient
to detect problems like this.  They work properly for the yt sample
FLASH dataset, and I think they should draw out problems like the ones
you describe.  Here is a script:

http://paste.yt-project.org/show/3619/

Can you run this on your dataset and see if it works?  As a note,
there have been issues recently with hierarchies getting serialized;
I'd recommend removing and .yt files scattered about and also setting
this in ~/.yt/config :

[yt]
serialize = False

Just to be on the safe side.

-Matt

>
> --
> Roman
> _______________________________________________
> yt-users mailing list
> yt-users at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org