Implicit treatment of coupled boundary conditions

Merged Sergio Ferraris requested to merge feature-implicitCyclics-merge into develop


The alternative addressing insert new internal faces on selected coupled patches. Extending the matrix addressing allows the implicit treatment of some boundary conditions. At the moment, the BCs which can be made implicit are: cyclic, AMI, ACMI and mapped.

The entry useImplicit true is needed in the field BC. Only scalar fields can be made implicit (p, p_rgh, k, epsilon, etc). The U field cannot be implicit.

The top solvers are unchanged except for the cht solvers where a single he matrix is built for the multi-region case.

In the case that more than one patch is present, the user can choose which ones are treated implicitly, i.e p can be implicit in AMI1 and explicit in AMI2 and k the other way around.

Currently, there exists a limitation for parallel cases where AMI patch-pairs need to be on the same processor. For cyclics and mapped patches, the number of faces on each processor MUST be the same. Therefore a constraint decomposition needs to be used.

Details of new models

The model adds a new numbering of internal/boundary coefficients and patch ID's. This is constructed at the call of the matrix.solve() - it is reset to the original addressing after solving.

The original internal/boundary coeffs are cached and used (in some cases with some extra information) to fill the coefficients on the new internal faces and/or source term on the newly formed fvMatrix.


  • tutorials/heatTransfer/chtMultiRegionSimpleFoam/cpuCabinet/
  • tutorials/heatTransfer/chtMultiRegionFoam/snappyMultiRegionHeaterImplicit/
  • tutorials/basic/laplacianFoam/implicitAMI/
  • tutorials/basic/simpleFoam/implicitAMI
  • tutorials/basic/chtMultiRegionFoam/2DImplicitCyclic/


Addressing of cells on the patches are changed, this needs to be passed to all the lduInterfaces. Use alternative lduAddressing in lduMatrix.

The coupled BCs (cyclics, AMI, mapped) which made implicit don't exist at the moment of solving the matrix. Thus, any fvOption or FO referring to this BC will Fail

In general terms when using the implicit approach the lduaddressing changes and therefore any matrix solve will be affected in all top level solvers and all the linear solvers.

Edited by Andrew Heather

Merge request reports