-
Notifications
You must be signed in to change notification settings - Fork 141
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Halogen training reactions #515
base: main
Are you sure you want to change the base?
Conversation
a10dea8
to
604712c
Compare
This is the only test that is failing
This test seems to bounce back and forth between 6 and 7. When we added CO to primaryThermoLib, we changed it from 6 to 7, now it want to be 6 again. Can we change this test to |
dd66f46
to
e0d02b7
Compare
053cd4e
to
6fa5908
Compare
6fa5908
to
8f7ba91
Compare
b27d3b6
to
1cada62
Compare
Cross validation for R_Addition_MultipleBond tree |
caec39b
to
cb4a63c
Compare
looks like the tests dies here |
f4aeaee
to
80a0f4f
Compare
80a0f4f
to
5838493
Compare
5838493
to
908eecb
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This looks good to me. I looked over the files and ran a bunch of examples including 2-BTP, ethane pyrolysis (minimal), cpox (surface), Nitrogen, and Sulfur, and none of them crashed. I picked the Nitrogen regression test as the one to look at for a more thorough review. I was slightly concerned that the halogens version finds fewer species and reactions than the main branch because my lizard brain is convinced that more=better. But most of the "missing" reactions are from Disproportionation, which was regenerated for all those new training reactions, so that makes sense that it would be different. I also ran a few simulations in Cantera for the Nitrogen example to see how the results changed, and you get nearly identical results at high temperatures. The halogens model is marginally slower than the main branch's version, especially at lower temperatures. I don't know much about the chemistry, but it seems reasonable to me.
Well done David! This PR must have taken a lot of work. I'm excited for it to be merged in.
Thanks for the awesome review @sevyharris, and thanks for getting the rmg regressions tests working! I will also take a look at the regression tests to see if there is anything that catches my eye. |
I'm thinking about dropping this commit 4a17fe7 which makes |
Sounds like a reasonable idea. Would leave H migration unperturbed; speed up generation of models with no halogens; and avoid the very different rates from influencing each other. |
908eecb
to
fed41cb
Compare
The new |
fed41cb
to
814180d
Compare
Cross validation for H Abstraction tree |
Disprop tree tree_cross_val.pdf This is outlier
ln(kest/ktrain)=10.2 |
added reverseMap, reactant and product Num for H Abs
ran a script to relabel spcs in H Abs training
…ction H2-2 + O2 <=> HO2-3 + H-2 violates the collision limit: reverse ratio=120558544715676.7 condition=500.0 K, 1.0 bar reverse ratio=55.46194345811393 condition=1500.0 K, 1.0 bar
Imported reactions from literature reactions are changed to rank 10 (since we don't always know the source) and AutoTST calculated rates are changed to rank 7 since most are DFT level without rotor scans
there is no rate rule for this group
0142778
to
e635e1d
Compare
changing to C2s prevents matching of C2sc groups which raise atomtype error when template is applied: Error: Could not update atomtypes for this molecule: multiplicity -187 1 F u0 p3 c0 {4,S} 2 O u0 p1 c+1 {3,D} {4,S} 3 O u0 p2 c0 {2,D} 4 *3 C u0 p0 c-1 {1,S} {2,S} {5,S} {6,S} {8,S} 5 H u0 p0 c0 {4,S} 6 *2 O u0 p2 c0 {4,S} {8,S} 7 O u0 p2 c0 {8,D} 8 *1 C u0 p0 c0 {4,S} {6,S} {7,D} Error: Problem family: 1+2_Cycloaddition Error: Problem reactants: (Molecule(smiles="O=C=O"), Molecule(smiles="O=[O+][C-]F")) Error: <Molecule "O=C=O"> 1 *2 O u0 p2 c0 {3,D} 2 O u0 p2 c0 {3,D} 3 *1 C u0 p0 c0 {1,D} {2,D} Error: <Molecule "O=[O+][C-]F"> 1 F u0 p3 c0 {4,S} 2 O u0 p1 c+1 {3,D} {4,S} 3 O u0 p2 c0 {2,D} 4 *3 C u0 p1 c-1 {1,S} {2,S} {5,S} 5 H u0 p0 c0 {4,S}
added training reactions from halogen kinetic models include NIST HFC and 2-BTP/CF3Br mech and Westmoreland YF mech. Also corrected CH3 in disproportionation training
twin with ReactionMechanismGenerator/RMG-Py#2135