[MOAB-dev] coupler test can not work well

[Vijay S. Mahadevan]

Huayi, we are in the process of redesigning several aspects of the
coupler to include more robust projection, normalization and
conservation algorithms. There is currently global and subset based
normalization, which as the name specifies only aims to rescale the
solution data globally or patch-wise. If you have intricate
requirements on conservation of data, this might not satisfy your
criteria. Or if you want to do local mass conservation on elements,
this might still not be sufficient without writing extra code.

We are always happy to take in contributions if you want to add
conservation techniques to the solution transfer capability.

Vijay

On Thu, Jul 9, 2015 at 5:44 PM, Grindeanu, Iulian R. <iulian at mcs.anl.gov> wrote:
> so you will probably want to look over normalization method (implemented in coupler too)
> In order to have conservation, you need to calculate integrals, on both target and source
>
> "normalization" will compute a factor (with which you multiply the target "computed" field)  such as the target integral is equal to the source integral
>
> Some of this work can be done with the coupler, if you understand what it does, and if you like the formula used for integration.
>
> If you do not like them, or if your "conservation" is defined differently, you may need to implement your own.
>
> Iulian
>
> ________________________________________
> From: 魏华祎 [huayiwei1984 at gmail.com]
> Sent: Thursday, July 09, 2015 5:31 PM
> To: Grindeanu, Iulian R.
> Cc: Vijay S. Mahadevan; MOAB Dev
> Subject: Re: [MOAB-dev] coupler test can not work well
>
> Hi, Lulian,
>
> Thanks very much for your help. It works well now.
>
> I have another question, does mb coupler consider the conservation
> law? For example, mass conservation.
>
> In every quad of my mesh,  there is a constant mass density. When I
> tansfer it from old mesh to new mesh, I need to keep the mass
> conservation as much as possible.
>
> In the above test, it is just a points interpolation. So I need to do
> more work to keep the mass conservation, right?
>
> Thanks again.
>
> Best
>
> Huayi
>
> 2015-07-09 22:47 GMT+08:00 Grindeanu, Iulian R. <iulian at mcs.anl.gov>:
>> you used a different scenario
>> parallel comm needs to know what is the source set
>>
>> If you load from file, that gets populated automatically; in your case, you create mesh in memory
>>
>> so one issue:
>> pc0->partition_sets().insert(src_mesh);
>>
>> also, you delete pc0 before mb, also mbc needs to be new too, otherwise it crashes (basically, mbc will be deleted when it goes out of scope, by that time mb and pc0 are gone)
>>
>> see attached t1.cpp, it works as expected
>>
>>
>> ________________________________________
>> From: moab-dev-bounces at mcs.anl.gov [moab-dev-bounces at mcs.anl.gov] on behalf of 魏华祎 [huayiwei1984 at gmail.com]
>> Sent: Thursday, July 09, 2015 8:44 AM
>> To: Vijay S. Mahadevan
>> Cc: MOAB Dev
>> Subject: [MOAB-dev] coupler test can not work well
>>
>> Hi, MOAB Developer,
>>
>> I  code a very simple test , please see the attachment.  The source
>> mesh is  following:
>>
>> (-1, 1) ------------(0, 1)--------------(1, 1)
>>     |                     |                    |
>>     |                     |                    |
>> (-1, 0)-------------(0, 0)-------------(1, 0)
>>     |                     |                   |
>>     |                     |                   |
>> (-1, -1)------------(0, -1)------------(1, -1)
>>
>> In every quad, there is a constant. For example, [1.0, 2.0, 2.0, 3.0]
>> ( from left to right and bottom to up) .
>>
>> I just change point (0, 0)  as (0.3, 0.4) to get the target mesh.
>>
>> I can compile it, but when I run it, it throw a segment error.
>>
>> This is my first time to use mb coupler tool, and not sure what
>> mistake I have made, so I need your help.
>>
>> Thanks very much.
>>
>> Best
>>
>> Huayi
>>
>> 2015-07-09 9:56 GMT+08:00 魏华祎 <huayiwei1984 at gmail.com>:
>>> My application mesh is a block structured mesh, which vertices will
>>> move and hence the mesh quality will become bad,  so I need to
>>> optimize the location of mesh vertices. Then I need to tansfer data
>>> from old mesh to new mesh. That's my problem.  My program is parallel
>>> and there is a structured quad mesh on each processor.  By the way,
>>> my program is not based on Moab. But I want to use
>>> the coupler tool in moab. So fristly I need transfer my mesh into Moab.
>>>
>>> At present, I explicityly transfer my every block mesh to the
>>> repesentation of vertices and coonecicity, just as your suggection. I
>>> am writing a test first. Dose you have better idea about it?
>>>
>>> 2015-07-09 9:36 GMT+08:00 Vijay S. Mahadevan <vijay.m at gmail.com>:
>>>> If you are asking about the ScdBox structure, I'm unsure. I don't
>>>> think we have a test for this. However, if you explicitly represent
>>>> the vertices and connectivity, usage of a structured grid is
>>>> transparent to MOAB interfaces and everything should work as is.
>>>>
>>>> What type of problems are you solving ? Are you using a logically
>>>> structured or performing stencil computations ?
>>>>
>>>> Vijay
>>>>
>>>> On Wed, Jul 8, 2015 at 8:33 PM, 魏华祎 <huayiwei1984 at gmail.com> wrote:
>>>>> Hi, Vijay,
>>>>>
>>>>> Thanks for your reply.  So it also work on Structured Mesh
>>>>> Representation of Moab?
>>>>>
>>>>> Best
>>>>>
>>>>> Huayi
>>>>>
>>>>> 2015-07-09 6:50 GMT+08:00 Vijay S. Mahadevan <vijay.m at gmail.com>:
>>>>>> No it should work for any dimension and is not restricted to 3-D. Have
>>>>>> you tried using the coupler for your problem ? If you have issues,
>>>>>> send us a test case and the meshes. You can also use mbcoupler_test to
>>>>>> verify whether the solution defined on your source mesh is being
>>>>>> transferred correctly onto the target mesh.
>>>>>>
>>>>>> Vijay
>>>>>>
>>>>>> On Wed, Jul 8, 2015 at 5:33 PM, 魏华祎 <huayiwei1984 at gmail.com> wrote:
>>>>>>> Hi, There,
>>>>>>>
>>>>>>> I am trying to use Moab Coupler transfer data between 2d meshes, but I
>>>>>>> am not sure if  it works on 2d mesh.
>>>>>>>
>>>>>>> Please give me some suggestion, thanks very much.
>>>>>>>
>>>>>>> Best
>>>>>>>
>>>>>>> Huayi