Pruning with pressure-dependence about rmg-java

[stijnvranckx]

This is still not working.

On Josh's branch at commit
125a79b
http://github.com/jwallen/RMG-Java/tree/125a79
we still run out of memory

Settings:

DynamicSimulator: DASSL

TimeStep: AUTOPRUNE

TerminationTolerance: 1E30

PruningTolerance: 1.0E-18

MinSpeciesForPruning: 1000

MaxEdgeSpeciesAfterPruning: 30000

As you can see, it is not pruning down to 30k species:

bfm_stijn]$ grep pruning RMG.log
The number of species in the model's edge, after pruning: 521
The number of species in the model's edge, before pruning: 3033
The number of species in the model's edge, after pruning: 3010
The number of species in the model's edge, before pruning: 3563
The number of species in the model's edge, after pruning: 3540
The number of species in the model's edge, before pruning: 3721
The number of species in the model's edge, after pruning: 3698
The number of species in the model's edge, before pruning: 4849
The number of species in the model's edge, after pruning: 4826
The number of species in the model's edge, before pruning: 5782
The number of species in the model's edge, after pruning: 5759
The number of species in the model's edge, before pruning: 6286
The number of species in the model's edge, after pruning: 6266
The number of species in the model's edge, before pruning: 6901
The number of species in the model's edge, after pruning: 6881
The number of species in the model's edge, before pruning: 7657
The number of species in the model's edge, after pruning: 7635
The number of species in the model's edge, before pruning: 8043
The number of species in the model's edge, after pruning: 8017
The number of species in the model's edge, before pruning: 10365
The number of species in the model's edge, after pruning: 10339
The number of species in the model's edge, before pruning: 11681
The number of species in the model's edge, after pruning: 11659
The number of species in the model's edge, before pruning: 13441
The number of species in the model's edge, after pruning: 13418
The number of species in the model's edge, before pruning: 14973
The number of species in the model's edge, after pruning: 14952
The number of species in the model's edge, before pruning: 16096
The number of species in the model's edge, after pruning: 16077
The number of species in the model's edge, before pruning: 17307
The number of species in the model's edge, after pruning: 17240
The number of species in the model's edge, before pruning: 18391
The number of species in the model's edge, after pruning: 18376
The number of species in the model's edge, before pruning: 19994
The number of species in the model's edge, after pruning: 19979
The number of species in the model's edge, before pruning: 21575
The number of species in the model's edge, after pruning: 21553
The number of species in the model's edge, before pruning: 22672
The number of species in the model's edge, after pruning: 22655
The number of species in the model's edge, before pruning: 23528
The number of species in the model's edge, after pruning: 23505
The number of species in the model's edge, before pruning: 24823
The number of species in the model's edge, after pruning: 24805
The number of species in the model's edge, before pruning: 27016
The number of species in the model's edge, after pruning: 26959
The number of species in the model's edge, before pruning: 29355
The number of species in the model's edge, after pruning: 29319
The number of species in the model's edge, before pruning: 32774
The number of species in the model's edge, after pruning: 31669
The number of species in the model's edge, before pruning: 31980
The number of species in the model's edge, after pruning: 31738
The number of species in the model's edge, before pruning: 34223
The number of species in the model's edge, after pruning: 34004
The number of species in the model's edge, before pruning: 35491
The number of species in the model's edge, after pruning: 35237
The number of species in the model's edge, before pruning: 37562
The number of species in the model's edge, after pruning: 37329
The number of species in the model's edge, before pruning: 38735
The number of species in the model's edge, after pruning: 38473
The number of species in the model's edge, before pruning: 41671
The number of species in the model's edge, after pruning: 41464
The number of species in the model's edge, before pruning: 44865
The number of species in the model's edge, after pruning: 44656
The number of species in the model's edge, before pruning: 47356
The number of species in the model's edge, after pruning: 47106
The number of species in the model's edge, before pruning: 50701
The number of species in the model's edge, after pruning: 50417
The number of species in the model's edge, before pruning: 51693
The number of species in the model's edge, after pruning: 51496

logs currently on monch in ~/RMGruns/bfm_stijn

[gmagoon]

It looks like at most 922 species are elligible for pruning as of the last iteration based on the number of edge species being tracked.

[gmagoon]

OK, I think I've narrowed down the issue. If you look at the lines with DEBUG at the beginning (the reactions where RMG thinks the reactants are not in the core; I estimate there are about 6700 such reactions at the the last iteration before running out of memory), they all seem to involve a small molecule (probably a seed species). Also, I noticed that Stijn is using "GenerateReactions: no" . For some reason, it seems that RMG does not recognize these species as being in the core, even though in at least some cases, they should be recognized as such. Is there some reason to expect that these seed mechanism species would not be part of the core?

(Also I should clarify that although I may have stated earlier that the reaction set has both forward and reverse reactions; this is not correct, it only has one of them, but it seems to be the direction taking core to edge (cf. appendUnreactedSpeciesStatus in ReactionSystem.java))

[gmagoon]

Another (probably unrelated) issue: Stijn's output contains many of the following type lines:

Warning: Attempted to read unreacted species flux for C10H14O5(55713), but there are only 54444 fluxes.

This is produced by RateBasedPDepRME and appears to reflect a mismatch between unreactedSpeciesFlux.length and the maximum species ID number.

[gmagoon]

The issue with the DEBUG lines turns out to be a non-issue, I think: RMG was iterating through all Pdep reactions (core and edge) and was identifying reactions that had no edge species. The unrelated issue of mismatch between unreactedSpeciesFlux.length and the maximum species ID number still remains, as far as I know. JWA and I think we have pinpointed the source of the bug that Stijn is observing and JWA has developed and implemented a fix which he will soon push for testing.

[gmagoon]

Post-mortem analysis: I failed to incorporate the changes in eee9ce6 (in particular the change to getLeakFlux) into the code for writing auto edge reaction info. This is also likely the source of Issue #114 (http://github.com/GreenGroup/RMG-Java/issues#issue/114)

[stijnvranckx]

the pruning seems to be working fine! It is keeping the amount of species in the model edge at 30000 and pruning some 1000 species on average every time it prunes.

[gmagoon]

This should be fixed now.

[gmagoon]

Clarification...this is fixed on PDepRMEMods branch...I've reopened and we can close once we bring it to master

[gmagoon]

Merge branch 'PDepRMEMods' per discussion between grm, jwa, rhw

Conflicts:
source/RMG/jing/rxnSys/JDAS.java

closed by 61872e8 (temporarily)