overset: normalisation of new matrix entries for interpolated/hole cells does not include remote contributions
Summary
Division by zero (producing sigfpe) in the linear solver for p in some decompositions.
Steps to reproduce
basic/overPotentialFoam/cylinder/cylinderAndBackground and use e.g.
//- The total number of domains (mandatory)
numberOfSubdomains 30;
//- The decomposition method (mandatory)
method random;
The idea is that it creates cells where all the neighbours are on other processors. This leads to a zero normalisation in the matrix rewrite.
Environment information
- OpenFOAM version : v1906, develop
Possible fixes
When inserting the new matrix 'rows' for interpolated and/or hole cells make sure to use a normalisation that includes the remote contributions.