[MOAB-dev] Edges at the ends of a periodic mesh?

[Iulian Grindeanu]

----- Original Message -----
| On May 23, 2012, at 11:20 PM, Timothy J. Tautges wrote:
| >
| >
| > ----- Original Message -----
| >> From: "Robert Jacob" <jacob at mcs.anl.gov>
| >> To: "Timothy J. Tautges" <tautges at mcs.anl.gov>
| >> Cc: "Iulian Grindeanu" <iulian at mcs.anl.gov>, moab-dev at mcs.anl.gov
| >> Sent: Wednesday, May 23, 2012 11:17:06 PM
| >> Subject: Re: [MOAB-dev] Edges at the ends of a periodic mesh?
| >> This was motivated by a more practical question. If there are N
| >> cells
| >> in the j-direction, there will be N+1 edges but the last is the
| >> same
| >> as
| >> the first. So in the call to box create, do we specify N vertices
| >> in
| >> the j-direction or N+1? I'm still not sure.
| >>
| >
| > In parallel, N+1, with is_periodic false. In serial, N, with
| > is_periodic true.
| >
| Do you assume the partition is over j direction? What if the partition
| is over other directions, which ends up with the mesh on each process
| is really a periodic mesh?
| Xiabing
So, first you decide if the global mesh is periodic. Let's say it is periodic in i direction. When you SCDInterface::compute_partition, you pass gdims [3]-gdims[0] = N+1, where N is the number of cells, even if it is periodic in i. If (ldims[3]-ldims[0] == gdims[3]-gdims[0]) it means that your partition was not broken in i direction, so when you create the box, you have to do periodic in i, so it will create only N nodes in i direction. It is a bad case (small number of processors, or number of data in j direction is very large compared to i). In "most" of the cases, you should have a break in both directions, then your periodicity will be broken. But you have to account for these "worse case scenarios" too. Iulian | > - tim
| >
| >> Rob
| >>
| >>
| >> On 5/23/12 10:26 PM, Timothy J. Tautges wrote:
| >>> Yes. But, in parallel, each piece of mesh looks like a serial
| >>> mesh,
| >>> so locally it won't appear periodic (the right hand edge will be
| >>> distinct from the left hand edge). Logically across processors,
| >>> though, the right edge on the rightmost processor will be the same
| >>> as the left edge on the leftmost processor. If you write the mesh
| >>> to
| >>> a single file, there will be a common edge between the rightmost
| >>> and
| >>> leftmost quad.
| >>>
| >>> - tim
| >>>
| >>> ----- Original Message -----
| >>> From: "Robert Jacob"<jacob at mcs.anl.gov>
| >>> To: "Timothy J. Tautges"<tautges at mcs.anl.gov>
| >>> Cc: "Iulian Grindeanu"<iulian at mcs.anl.gov>, moab-dev at mcs.anl.gov
| >>> Sent: Wednesday, May 23, 2012 9:13:20 PM
| >>> Subject: Re: [MOAB-dev] Edges at the ends of a periodic mesh?
| >>>
| >>>
| >>>
| >>> On 5/23/12 9:01 PM, Timothy J. Tautges wrote:
| >>>> Be careful about the parallel case though, where on each proc
| >>>> sharing the edge the edge is represented.
| >>>
| >>> So in the parallel case the edge WOULD be represented on both
| >>> sides
| >>> if
| >>> there are multiple processors in the j direction?
| >>>
| >>> Rob
| >>>
| >>>>
| >>>> Also, if you're just trying to create all the edges in one big
| >>>> batch, just put all the quads in a range, then call
| >>>> get_adjacencies, with the last argument Interface::UNION, and
| >>>> that
| >>>> should create/return all of them.
| >>>>
| >>>> - tim
| >>>>
| >>>> ----- Original Message -----
| >>>> From: "Iulian Grindeanu"<iulian at mcs.anl.gov>
| >>>> To: "Robert Jacob"<jacob at mcs.anl.gov>
| >>>> Cc: moab-dev at mcs.anl.gov
| >>>> Sent: Wednesday, May 23, 2012 12:34:12 PM
| >>>> Subject: Re: [MOAB-dev] Edges at the ends of a periodic mesh?
| >>>>
| >>>>
| >>>>
| >>>>
| >>>> ----- Original Message -----
| >>>>
| >>>>
| >>>>
| >>>> Hi,
| >>>>
| >>>> If I have a 2D rectangular mesh which is periodic in one
| >>>> direction,
| >>>> lets
| >>>> say the j direction, then the leftmost and rightmost edge of the
| >>>> mesh
| >>>> are really the same edge. Do they both need to be declared in
| >>>> MOAB?
| >>>> Or
| >>>> does the periodic property mean you only need to describe one of
| >>>> the
| >>>> edges and the other end will know its adjacent to it?
| >>>>
| >>>> Thanks,
| >>>> Rob
| >>>> Hi,
| >>>> I assume that you use ScdInterface;
| >>>> If the box is periodic in j direction, the nodes in that
| >>>> direction
| >>>> will be "periodic".
| >>>> Nodes are not duplicated, the edges that you get on the leftmost
| >>>> and rightmost sides are really pointing to the same nodes, they
| >>>> will not be duplicated, either.
| >>>>
| >>>> How do you get edges after you create quads?
| >>>> You can use "get_adjacency" methods, with "create_if_missing"
| >>>> flag
| >>>> true. Only one edge would be created.
| >>>> The problem is that the order in which these are created is not
| >>>> very "structured"
| >>>> Or do you create them with new scd boxes, of dimension 1, in a
| >>>> loop
| >>>> over i (and then j?)? In that case, you have complete control
| >>>> over
| >>>> "order" they are created in.
| >>>> I hope this helps.
| >>>> Iulian
| >>>>
| >>>>
| >>>> Iulian
| >>>>
| >>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/moab-dev/attachments/20120525/70885dd1/attachment.html>