Restart of lagrangian particles fails for reactingParcelFoam dependent on domain decomposition
Case particles impining on cube: Particles are impinging on a cube and form a liquid film with the solver reactingParcelFoam (V22/12). The duration of the spray is 1000 s. The simulation (on single processor) shows the impingment and the formation of a film and after approx. 1.5 the re-entrainment of particles back in the flow. The simulation is splitted in two simulations 0-1s and 1-2s which is a restart of the results for 1s (starting from latestTime). The results show (see massfilm.jpg) different results based on single / parallel simulation:
- single processor: Mass of film increases after the restart (which was expected)
- 4 proc. (simple 4/1/1) which is a distribution where the particle positons for the injection in reactingCloudProperties are on one processor shows the same results, like on a single processor. The mass of film increases after the restart
- 4 proc. (simple 1/2/2) which is a distribution where the particle injection positions are not on one procesor fails. After the restart no particles are injected and the mass drops.
This is a reproducible behaviour for the steps
- Use endtime = 1 s in controlDict. Run reactingParcelFoam > log1
- Change endtime = 2 s in controlDict. Run reactingParcelFoam > log2
- Evaluate result: cat log1 log2 > log
- ./auswert_particles and look at the mass in the system
This can be run for three different cases, single, parallel (1/2/2) and parallel (4/1/1) distribution in simple decomposition.
Summary: I.e. dependent on domain decomposition the restart works or works not.
Testcase
Environment information
OpenFOAM version : v2212 Operating system : ubuntu Hardware info : different systems Compiler : gcc
Possible fixes
If one removes uniform directory in processor?/1 in all directories and sets start of injection SOI in reactingCloudProperties equal 1 the particles seem to be calculated again independent on decomposition