[t.teschner]

I am trying to add a source term to the momentum equation which is depending on the velocity vector, i.e. ). Let's assume I have the momentum equation in the following form (no pressure gradient):

fvVectorMatrix UEqn
(
    fvm::ddt(U)
  + fvm::div(phi, U)
  - fvm::laplacian(nu, U)
  + turbulence->divDevReff(U)
);
solve(UEqn == - source.evaluate());

Here, source is an object which has access to the velocity vector (specifically a const reference, i.e. it is defined as const volVectorField &U;) and appropriately initialised in the constructor.

I want source.evaluate() to be executed for each iteration in the linear solver, i.e. if I have 5 iterations using a GaussSeidel solver, I want this function to be executed 5 times (and since U is changing during each iteration, the velocity vector dependend sourceterm should change for each iteration as well).

Testing the above, it seems that source.evaluate() is only executed once at the beginning of the iteration but never called for subsequent iterations. It doesn't matter if it is directly included in the UEqn or in the solve() method.

The source term is explicit and does not contribute to the coefficient matrix.

Are there any suggestions how to achieve that?

[t.teschner]

I managed to fix my problem, in case someone stumbles upon this in the future, the trick is to add the contribution of the source (depending on the variable that is being solved for) as an implicit source term (either fvm::Sp() or fvm::SuSp). The book of Moukalled et al. (The finite volume Method in Computational Fluid Dynamics, Chapter 14.1 and 14.5.2) is particularly helpful.

[dlahaye]

Hmm, are you sure about change the right-hand side vector b while solving the linear system A u = b ?

[t.teschner]

Originally Posted by dlahaye

Hmm, are you sure about change the right-hand side vector b while solving the linear system A u = b ?

Yes, as explained in the opening post, the source term is depending on the variable that is being solved for (here the velocity vector U) so it needs to be added implicitly. The contribution will be added to the coefficient matrix A using fvm::Sp(). Actually the full story is a bit more complicated in my case (there are some explicit contributions as well, which are just added to the right hand side vector b) but with this change I got it to work (validated against analytic data)