Severe bug in HiParSimulator

[klreuther]

Symptoms (I) The same input cell configuration yields different results depending on its positioning inside the domain as well as the domain size. (II) In all cases the results are wrong in several cells. (III) The results are reproducible.

All of these symptoms point to problems in the ghost-zoning process.

LGD Parameters

public APITraits::HasTorusTopology<3>,
public APITraits::HasStencil<Stencils::Moore<3,5>>,

(Torus topology, 11x11x11 square neighborhood for calculations)

Partitioner is RecursiveBisectionPartition<3>

Example Exemplary portions of results for three different test cases are appended. The initial cell configuration is the same in all cases. Parallelization with 48 processes.

1. 10x10x10 domain, initial cell configuration centered on (5 5 5)
2. 20x20x20 domain, initial cell configuraiton centered on (5 5 5)
3. 20x20x20 domain, initial cell configuration centered on (10 10 10)

(data expectations: the results shown should be horizontally and vertically symmetric with respect to their absolute values. All values to the left of the center should be negative, all values to the right should be positive)