View Issue Details

IDProjectCategoryView StatusLast Update
0001916OpenFOAMBugpublic2016-02-03 21:45
Reporterjherb Assigned Tohenry  
PrioritynormalSeverityminorReproducibilityalways
Status closedResolutionsuspended 
PlatformGNU/LinuxOSOpenSuSEOS Version13.2
Summary0001916: epsilonm seems not to converge in $FOAM_TUTORIAL/multiphase/twoPhaseEulerFoam/RAS/bubbleColumn
DescriptionIn the tutorial $FOAM_TUTORIAL/multiphase/twoPhaseEulerFoam/RAS/bubbleColumn for the solver twoPhaseEulerFoam the residuals for epsilonm stay somewhere between 0.01 and 0.1 even though all other equations seem to converge.

Is this an expected behavior?
TagsNo tags attached.

Relationships

related to 0001857 closedhenry possible bug in mixtureKEpsilon turbulence model 

Activities

jherb

2015-11-18 12:17

reporter   ~0005626

The same happens with OpenFOAM 3.0.0

henry

2015-11-18 12:31

manager   ~0005627

Yes this is the current behavior. Feel free to investigate.

wyldckat

2015-11-29 18:48

updater  

residuals_epsilon.png (8,546 bytes)   
residuals_epsilon.png (8,546 bytes)   

wyldckat

2015-11-29 18:49

updater  

residuals_p_k.png (7,119 bytes)   
residuals_p_k.png (7,119 bytes)   

wyldckat

2015-11-29 18:50

updater  

bubbleColumn_1916.tar.gz (20,097 bytes)

wyldckat

2015-11-29 19:13

updater   ~0005684

I was curious about this one and decided to at least give it a shot at trying to diagnose at least part of the issue.

Attached are the following files:

  - bubbleColumn_1916.tar.gz - tutorial case "multiphase/twoPhaseEulerFoam/RAS/bubbleColumn" from OpenFOAM 3.0.0, modified to print out some stats regarding the residuals and normalization factors, along with plots of the respective values. Simply unpack and run: ./Allrun

  - residuals_epsilon.png - plot that shows the residuals and normalization factors used for the first "e.air" and e.water iteration residuals, along with the pair for "epsilonm".

  - residuals_p_k.png - plot that shows the residuals and normalization factors for the "p_rgh" (first iteration) and "km" fields/equations.

The idea of the plots is to get a perception of the scales at play here.


My diagnosis of this is as follows:

 1. Not clearly visible in these plots, but the initial residual for the first iteration of "p_rgh" is roughly on the same scale as "epsilonm".

 2. The normalization factor for "km" is pretty high (~1e+9) and the respective residual is fairly low (~1e-12).

 3. The residuals for "e.air" and "e.water" are on different scales: ~1e-4 and ~1e-5 respectively.

 4. Comparatively the normalization factors for "e.air" and "e.water" are also on completely different scales: ~1e+5 and ~1e+7 (to ~1e+8) respectively. A pattern rises from these values, where the normalization factors seem to be scaled by roughly 1000, the relation between rho of air versus water's rho.

 5. The other pattern is that since "epsilonm" is the mixture of both phases, we are mixing two different scales of epsilon values, even if scaled by rho. A factor of 1000 can introduce a fair amount of numerical error, which would explain the pattern that the addition/mixture of the two normalization factors and residuals, that seem to lead to the consequential increase of both values for "epsilonm": residual of ~1e-2 and normalization factor of 1e+8 to 1e+9.

 6. The idea I get from this is that the initial residual for "epsilonm" can be disregarded and that its final residual is what's important, because the other indicators are more important:

   - the initial residual for "km" is very small;
   - the initial residuals for "e.air" and "e.water" are also very small.

 7. Nonetheless, it might be worthwhile looking into whether "epsilonm" is a proper mix of "e.air" and "e.water", since it does seem like the issue could also arise from a missing or an excess rho factor. On the other hand, it might simply be due to the numerical collision of two scales of turbulence, which can make it more difficult to have a good initial solution.


As for "numerical collision" that I'm referring to is along the lines of what happens when we add "1000" with "0.0001" = "1000.0001", where the "0.0001" in "1000.0001" can easily get lost in the equations, due to numerical precision.

wyldckat

2015-11-29 19:28

updater   ~0005685

In addition, this bug report might be related: http://www.openfoam.org/mantisbt/view.php?id=1857

@jherb: Can you take a look into #1857, as it seems to be related?

henry

2016-02-03 21:44

manager   ~0005910

Orphaned report.

Issue History

Date Modified Username Field Change
2015-11-18 12:15 jherb New Issue
2015-11-18 12:17 jherb Note Added: 0005626
2015-11-18 12:31 henry Note Added: 0005627
2015-11-29 18:48 wyldckat File Added: residuals_epsilon.png
2015-11-29 18:49 wyldckat File Added: residuals_p_k.png
2015-11-29 18:50 wyldckat File Added: bubbleColumn_1916.tar.gz
2015-11-29 19:13 wyldckat Note Added: 0005684
2015-11-29 19:28 wyldckat Note Added: 0005685
2015-11-29 19:29 wyldckat Relationship added related to 0001857
2016-02-03 21:44 henry Note Added: 0005910
2016-02-03 21:44 henry Status new => closed
2016-02-03 21:45 henry Assigned To => henry
2016-02-03 21:45 henry Resolution open => suspended