[sreehahaha]

Dear Foamers,

I'm new to use OF and I'm developing my own solver for my problem. I have a simple cubic geometry. In it, I have to define a 'surfaceScalarField payal' and I want to give pre-calculated values to it at every time step.

It is easy to give values to the internal faces, so I'll skip that here. I give values to the boundary patches with:

// to loop over all boundary faces
forAll(payal.boundaryFieldRef(), patchID)
{
	forAll(payal.boundaryFieldRef()[patchID],facei) 
   {
   	payal.boundaryFieldRef()[patchID][facei]=<aPre-calculatedValue>;
   }
}

But I use cyclicBC, so I have to do something like below for two neighbourPatches with patchIDs 0,1:

// set values on patchID=1 same as patchID=0, because cyclicBC between patches 0,1
forAll(payal.boundaryFieldRef()[1],facei)
{
	payal.boundaryFieldRef()[1][facei]=payal.boundaryFieldRef()[0][facei];
}

and likewise for other pairs of neighbourPatches. This works fine in a serial run. But, I imagine this would be problematic in a parallel run, as both patches may be dealt with by different processes. Can somebody please suggest an elegant way which works for both serial and parallel runs? It would be very helpful.

Many thanks and have a great day/weekend!

[SmileMax]

you can use decomposePar -no-sets(.com) or decomposePar -noSets(.org) to skip decomposing specified cellSets, faceSets, pointSets.

[sreehahaha]

Thank you for the reply, SmileMax, and sorry for my late reply.

The <aPre-calculatedValue> that I have written in my original question contains a random number. So, I think, not decomposing the surfaceScalarField Payal will give its copies to all processors. And, since there is a random number involved, each processor will end up having different versions of Payal. Please correct me if I'm wrong, I'm new to OF programming.

If I'm correct above, can you please think of another way to solve the issue.

Many thanks and have a nice day,
S

[SmileMax]

If you use OpenFOAM's built-in random number class, you can use the functions prefixed with global to ensure that the random numbers generated each time are kept in sync on all MPI nodes，such as Random::globalSample01(), Random::globalGaussNormal().

see src/OpenFOAM/primitives/random/Random/Random.C for these source codes:

Foam::scalar Foam::Random::globalSample01()
{
    scalar value(-GREAT);
 
    if (Pstream::master())
    {
        value = scalar01(); // generate random number on master MPI node.
    }
    Pstream::scatter(value); // broadcast this value to all MPI nodes.
    return value;
}

But in fact, the random numbers obtained by the built-in functions of OpenFOAM are always the same on different MPI nodes and even in different running stages if you give a fixed random seed. You can use Pout to output the value of each node to observe this phenomenon:

Random rndGen(<a fixed random seed>);

scalar a = rndGen.sample01<scalar>();
Pout << "random number: "<< a << endl;

scalar b = rndGen.globalSample01<scalar>();
Pout << "random number: "<< b << endl;

// a is always equal to b here, and each MPI node have the same value of a and b.

If you are using an external random number class, such as std:rand, You can refer to the implementation method of Random::globalSample01 above to manually complete random number synchronization between MPI nodes.