[petsc-users] Modeling a Network with DMPlex?

[Florian Meier]

I will build an example and send it to you, but it might take some time.

One short question: Should I preprocess my graph so that vertices that
are close in the graph are close in the data (i.e. get indices that have
small distances) or will this be done automatically?

On 02/10/2014 07:33 PM, Abhyankar, Shrirang G. wrote:
> 
> On 2/10/14 11:45 AM, "Florian Meier" <florian.meier at koalo.de> wrote:
> 
>> Hi Shri, Hi Jed,
>> thank you very much, that helped a lot!
>> DMCircuit is perfect for my application!
>>
>> I have finally managed to get a very simple example running after I
>> realized that comment lines inside of arrays are not allowed for pflow
>> input format ;-)
> 
> Cool! Note that the input format is specific to the application and is not
> a part of DMCircuit.
> 
>>
>> I am going to implement a more complex example and will probably come
>> across something for what I need your help again.
> 
> Please feel free to ask any questions you have. As DMCircuit is a recent
> development, your questions/comments/suggestions will help
> us to improve this feature.
> 
> It would be great if you could contribute your simple example as a
> tutorial example for DMCircuit.
> 
> Thanks,
> Shri
> 
>>
>> Thanks again,
>> Florian
>>
>> On 02/10/2014 08:33 AM, Abhyankar, Shrirang G. wrote:
>>> Hi Florian,
>>>  You can use DMPlex directly as Jed suggests or you can use a new DM
>>> that
>>> has been
>>> added recently to PETSc specifically for circuit/network/graph
>>> applications. This DM, named DMCircuit, is built on top of
>>> DMPlex. It is currently in the next branch of PETSc and there is an
>>> example from a power grid
>>> application in src/snes/examples/tutorials/pflow.
>>>
>>> Take a look at DMCircuit and let us know if you want to use it. I'll
>>> give
>>> you more information on its usage if you are
>>> interested.
>>>
>>> Shri
>>>
>>> On 2/10/14 12:35 AM, "Jed Brown" <jed at jedbrown.org> wrote:
>>>
>>>> Florian Meier <florian.meier at koalo.de> writes:
>>>>
>>>>> Hi,
>>>>> I would like to solve a non-linear equation system over a computer
>>>>> network. This is somewhat similar to FEM. There is a bunch of
>>>>> equations for each link that includes variables of adjacent links. I
>>>>> assume there is some support in PETSc for mapping the network to
>>>>> memory, distribute it and iterate over the adjacent links, but all
>>>>> examples I could found for DMPlex are so tightly coupled to FEM that I
>>>>> am not able to separate the relevant parts from the parts that are
>>>>> only needed for FEM (e.g. continuous elements, discretization).
>>>>>
>>>>> Is DMPlex actually the right place to look?
>>>>
>>>> You can describe a network having only edges and vertices.  This sounds
>>>> similar to what Shri does for power networks.  You wouldn't use the FEM
>>>> interfaces, but the DMPlex primitives still apply.
>>>
>>> You can We've recently added a DM for problems arising from circuit
>>> applications that may be
>>> of use to you. The interface is called DMCircuit and is currently in the
>>> next branch of PETSc.
>>> There is currently one DMCircuit example coming from a power grid
>>> application 
>>> in src/snes/examples/tutorials/circuit/pflow.
>>>
>>>
>>>>
>>>
>