On Dec 21, 2012 8:16 PM, "Nathan Goldbaum" <nathan12343@gmail.com> wrote:

This is a discussion that began on yt-users but since I'm going to speculate about how to change the code, I'm replying over here on yt-dev.

I see now that it's difficult to check for periodic boundary conditions in a code-independent way in the middle of a field definition.  There are a number of places where yt assumes periodic boundary conditions (mostly in plotting routines) but I don't see any places that programatically check the boundary conditions that the simulation was run with.  Am I missing something somewhere?

I feel like this functionality should be there but I'm not sure what the best solution is.

Just spitballing: perhaps there be something like pf.periodic which is True or False?  This would have to be defined in the individual code frontends by looking at runtime parameters.  It's not obvious to me that this will be easy to do for codes that don't have self-describing data formats.  Another issue is that a binary Periodic attribute won't be sufficient for simulations run with mixed boundary conditions.

Does anyone else have any thoughts on this?  I don't want to cause a huge disruption but I'm strongly -1 on the way the Radius field is currently written since it quietly produces incorrect results on a large number of test problem-like datasets.  It won't be a problem in cosmological simulations so I worry that inexperienced users will tend to get bit by it.

Cheers,

Nathan

On 12/21/12 4:26 PM, Geoffrey So wrote:

I've seen and used something similar in my ellipsoids, where the domain width (DW) is used to calculate the shortest distance between two points due to periodic boundary condition.  So if you do not have a periodic box then that line (involving DW) should be changed.  This is part of the reason why my ellipsoid container is limited to periodic box only.

I'm hoping to put in a flag to turn it off and on later, but if a flag is put in to turn it off maybe I'll imitate it for the ellipsoids, too.

From
G.S.

On Fri, Dec 21, 2012 at 4:18 PM, Nathan Goldbaum <nathan12343@gmail.com <mailto:nathan12343@gmail.com>> wrote:

    Hi all,

    I encountered a bit of odd code deep in the field definition code
    and I need some help to understand why it is the way it is.

    The field I'm curious about is the Radius field (defined on line
    796 of universal_fields.py):

    def _Radius(field, data):
        center = data.get_field_parameter("center")
        DW = data.pf.domain_right_edge - data.pf.domain_left_edge
        radius = np.zeros(data["x"].shape, dtype='float64')
        for i, ax in enumerate('xyz'):
            r = np.abs(data[ax] - center[i])
            radius += np.minimum(r, np.abs(DW[i]-r))**2.0
        np.sqrt(radius, radius)
        return radius

    In particular, I don't understand this line:

    radius += np.minimum(r, np.abs(DW[i]-r))**2.0

    Is this supposed to correct for some effect in periodic simulations?

    The reason I bring this up is because it caused some very weird
    behavior in a 1D FLASH simulation.  In this simulation, the domain
    went from x = -.01 pc to x = +500 pc.  Since the middle of the
    domain wasn't lined up with the 'center' of the simulation at x =
    0, the offending line in the radius calculation led to an
    incorrect calculation of the radius beyond x = 250 pc.  In that
    case, I was able to fix it by changing the offending line to

    radius += r**2.0

    Before I issue a patch, I want to see why it's defined the way it
    is currently and see if we can come up with a workaround.

    Cheers,

    Nathan
    _______________________________________________
    yt-users mailing list
    yt-users@lists.spacepope.org <mailto:yt-users@lists.spacepope.org>
    http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org




_______________________________________________
yt-users mailing list
yt-users@lists.spacepope.org
http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org

_______________________________________________
yt-dev mailing list
yt-dev@lists.spacepope.org
http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org