Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) #1935

dranasinghe · 2020-04-27T03:10:35Z

Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T)/def2-tzvp normalPNO and wb97xd/def2tzvp were added.

Motivation or Problem

DLPNO- CCSD(T) method is use full to calculate thermochemistry parameters of large molecules. The BAC and Atom Corrections are important for accurate thermochemical estimation. BAC correction for DLPO-CCSD(T)/def2-tzvp NormalPNO and wb97xd/deftzvp calculated using https://github.com/cgrambow/bac
algorithm. I only considered neutral molecules for the fitting. I notice cation produced large deviations. So cations were left out form the fitting. Further, rotors were not considered
model chemistry BAC can be used for
DLPNO-CCSD(T)/def2-tzvp//wb97xd/deftzvp NormalPNO
wb97xd/deftzvp//wb97xd/deftzvp

Description of Changes

Atom corrections and Petersson type bond correction were added.

Testing

Reran the test set using ARC.

Reviewer Tips

easy to check H298 by running H2, N2, O2, F2, and Cl2

…-tzvp normalPNO and wb97xd/def2tzvp were added.

amarkpayne · 2020-04-27T04:16:15Z

We should also update the ReferenceSpecies database with these calculations. @dranasinghe do you have the Arkane YAML files from the calculations?

codecov · 2020-04-27T04:21:02Z

Codecov Report

Merging #1935 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1935   +/-   ##
=======================================
  Coverage   45.05%   45.05%           
=======================================
  Files          86       86           
  Lines       22213    22213           
  Branches     5788     5788           
=======================================
  Hits        10008    10008           
  Misses      11099    11099           
  Partials     1106     1106

Impacted Files	Coverage Δ
arkane/encorr/data.py	`100.00% <100.00%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 35d114a...630fa34. Read the comment docs.

alongd

Thanks!!
Could you also address Mark's comment?
Also, the commit message is too long, could you make a short header and add the remaining details to the commit message body?

alongd · 2020-04-27T16:39:00Z

arkane/encorr/data.py

-SOC = {'H': 0.0, 'N': 0.0, 'O': -0.000355, 'C': -0.000135, 'S': -0.000893, 'P': 0.0, 'I': -0.011547226}
+# Spin orbit correction for F, Si, Cl, Br and B taken form https://cccbdb.nist.gov/elecspin.asp
+SOC = {'H': 0.0, 'N': 0.0, 'O': -0.000355, 'C': -0.000135, 'S': -0.000893, 'P': 0.0, 'I': -0.011547226,
+        'F': -0.000614, 'Si': -0.000682 , 'Cl': -0.001338, 'Br': -0.005597 , 'B': -0.000046 } 


please correct white spaces: there's an etra one before 'F', and an extra one after 'B': -0.000046.

alongd · 2020-04-27T16:41:19Z

arkane/encorr/data.py

+    'O-O': 0.09663, 'C=S': 0.80888, 'H-S': 1.23648, 'C=O': 0.88066, 'C=N': 0.03998,
+    'H-N': -0.49564, 'H-O': -0.41183, 'H-H': 0.6263, 'N#N': 3.71325, 'N-N': 0.74915, 
+    'N-O': -0.62156, 'C=C': -0.63901, 'O=S': -1.39626, 'O-S': -1.37002, 'S-S': 0.1515,
+    'F-S': -0.68693, 'F-O': 0.09202, 'F-H': -1.68214, 'F-F': 0.95483, 'O=O': -2.64949


please add a , at the end of the line (and also for the block below)

These corrections are obtained from calculations that use dlpno-ccsd(t)/def2-tzvp for both energy calculations and geometry optimizations? If the geometry optimization was done with a different level of theory, it should be included in the model chemistry.

I am not sure how to include it current structure. So I left a comment.

alongd · 2020-04-27T16:43:07Z

arkane/encorr/data.py

+    'C=O': 1.00501, 'C=N': -0.68305, 'H-N': -0.52232, 'H-O': -1.18129, 'H-H': -1.76862, 'N#N': -3.84259,
+    'N-N': 2.66325, 'N-O': 1.69619, 'C=C': -0.11192, 'O=S': -1.33397, 'O-S': -1.71863, 'S-S': 0.5224,
+    'F-S': -1.28933, 'F-O': -0.03756, 'F-H': -3.71018, 'F-F': -1.71494, 'O=O': -6.70857
+    },


Please add two lines to the Model Chemistry table in the documentation (https://github.com/ReactionMechanismGenerator/RMG-Py/blob/master/documentation/source/users/arkane/input.rst).

dranasinghe · 2020-04-28T04:46:45Z

@aelong and @amarkpayne Thank you for the comments. I made the changes. I have all the YAML files. I will update the reference database.

alongd

Thanks! I added a couple of minor comments to the docs commit.

alongd · 2020-04-28T17:06:37Z

documentation/source/users/arkane/input.rst

@@ -119,12 +119,16 @@ Model Chemistry                                  AEC   BC   SOC  Freq Scale Supp
 ``'B3LYP/6-31G(d,p)'``                            v    v         v (0.961)  H, C, O, S
 ``'MRCI+Davidson/aug-cc-pV(T+d)Z'``               v         v               H, C, N, O, S
 ``'wb97x-d/aug-cc-pvtz'``                         v         v               H, C, N, O
+``DLPNO-CCSD(T)/def2-tzvp``                       v    v    v               H, C, N, O, S, F, Cl


Do you mind placing DLPNO after B-CCSD(T)-F12/aug-cc-pVnZ so all CCSD and all wb97 methods are adjacent?

alongd · 2020-04-28T17:06:39Z

documentation/source/users/arkane/input.rst

 ================================================ ===== ==== ==== ========== ====================

 Notes:

 - The ``'CBS-QB3-Paraskevas'`` model chemistry is identical to ``'CBS-QB3'`` except for BCs for C/H/O bonds, which are from Paraskevas et al. (DOI: 10.1002/chem.201301381) instead of Petersson et al. (DOI: 10.1063/1.477794). Beware, combining BCs from different sources may lead to unforeseen results.
 - In ``'M08SO/MG3S*'`` the grid size used in the [QChem] electronic structure calculation utilizes 75 radial points and 434 angular points.
+- In ``DLPNO-CCSD(T)/def2-tzvp`` run using Orca 4.2.1 with NormalPNO
+- In ``wb97xd/def2tzvp`` run using Gaussian16. Do n't use the AEC and BC for wb97xd calculated with Qchem.


These are excellent tips!
In Arkane, which wb97xd were calculated by QChem? Is it 'wb97x-d/aug-cc-pvtz'? If so, Arkane won't use it due to the basis set (and also the dash in the functional name). If I misunderstood, could you elaborate?

I'm not sure we should be using the notation without hyphens. We could consider using wb97x-d/def2-tzvp to be consistent with notation in the literature rather than notation Gaussian uses. I doubt any wb97x-d calculations were done using Q-Chem because you would usually use wb97x-d3 in Q-Chem.

amarkpayne

Couple of additional comments but overall it looks good, thanks for running these calcs!

When you update this branch next, drop the merge from master commit and rebase.

amarkpayne · 2020-04-28T20:51:02Z

arkane/encorr/data.py

+    # Calculated atomic energies fitted orca 4.2.1  dlpno-ccsd(t)/def2-tzvp NormalPNO
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 7.99 and 5.96 kJ/mol respectively.
+    'dlpno-ccsd(t)/def2-tzvp': {
+        'H': -0.49641082, 'C': -37.77274125, 'N': -54.50388932, 'O': -74.96760414,
+        'F': -99.62044819, 'S': -397.63480236, 'Cl': -459.65784960
+    },
+    # wb97xd/def2tzvp conducted using G16. 
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 9.16 and 6.66 kJ/mol respectively.
+    'wb97xd/def2tzvp': {
+        'H': -0.50224721, 'C': -37.84235323, 'N': -54.58757237, 'O': -75.06982154,
+        'F': -99.74180094, 'S': -398.10765642, 'Cl': -460.14650064
+    },


Instead of manually adding in the SOC terms, write the energy and explicitly write in the SOC correction. For example, the correction for oxygen for dlpno-ccsd(t)/def2-tzvp would be 'O': -74.96724914 + SOC['O'] (which will get evaluated to -74.96760414). Do this for all atoms, even if the SOC is zero. This is a very minor thing, but it is better just in case we ever need to update the SOC values, and it makes it very clear that the correct SOC values were used.

Figured I would chime in here. So these were fitted and not just calculated?

amarkpayne · 2020-04-28T20:55:33Z

documentation/source/users/arkane/input.rst

 ================================================ ===== ==== ==== ========== ====================

 Notes:

 - The ``'CBS-QB3-Paraskevas'`` model chemistry is identical to ``'CBS-QB3'`` except for BCs for C/H/O bonds, which are from Paraskevas et al. (DOI: 10.1002/chem.201301381) instead of Petersson et al. (DOI: 10.1063/1.477794). Beware, combining BCs from different sources may lead to unforeseen results.
 - In ``'M08SO/MG3S*'`` the grid size used in the [QChem] electronic structure calculation utilizes 75 radial points and 434 angular points.
+- In ``DLPNO-CCSD(T)/def2-tzvp`` run using Orca 4.2.1 with NormalPNO


Did you use anything like TightSCF? It can be added here if so. I agree with @alongd it is a great idea to include this here

amarkpayne · 2020-04-28T20:56:14Z

documentation/source/users/arkane/input.rst

 ================================================ ===== ==== ==== ========== ====================

 Notes:

 - The ``'CBS-QB3-Paraskevas'`` model chemistry is identical to ``'CBS-QB3'`` except for BCs for C/H/O bonds, which are from Paraskevas et al. (DOI: 10.1002/chem.201301381) instead of Petersson et al. (DOI: 10.1063/1.477794). Beware, combining BCs from different sources may lead to unforeseen results.
 - In ``'M08SO/MG3S*'`` the grid size used in the [QChem] electronic structure calculation utilizes 75 radial points and 434 angular points.
+- In ``DLPNO-CCSD(T)/def2-tzvp`` run using Orca 4.2.1 with NormalPNO
+- In ``wb97xd/def2tzvp`` run using Gaussian16. Do n't use the AEC and BC for wb97xd calculated with Qchem.


Typo in don't

cgrambow

Just saw this PR and wanted to make some comments. In addition to the individual comments I made, I wanted to address some bigger issues.

What exactly is the issue with ionic/radical species? We have the full dataset including such species, so I think they should be included in the fit. I am able to obtain very good fits including ions and radicals using DFT methods, so I'd be surprised if it didn't work for DLPNO. Additionally, if we are providing Petersson-type corrections, we should also be providing Melius-type corrections.

Were the corrections fit with or without substructure weighting?

Note that a recent RMG-database PR corrected some species in the reference database and ReactionMechanismGenerator/RMG-database#404 corrects some more. The corrections in this PR should be refit after the database PR has been merged.

cgrambow · 2020-04-30T19:37:56Z

arkane/encorr/data.py

+    # Calculated atomic energies fitted orca 4.2.1  dlpno-ccsd(t)/def2-tzvp NormalPNO
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 7.99 and 5.96 kJ/mol respectively.
+    'dlpno-ccsd(t)/def2-tzvp': {
+        'H': -0.49641082, 'C': -37.77274125, 'N': -54.50388932, 'O': -74.96760414,
+        'F': -99.62044819, 'S': -397.63480236, 'Cl': -459.65784960
+    },
+    # wb97xd/def2tzvp conducted using G16. 
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 9.16 and 6.66 kJ/mol respectively.
+    'wb97xd/def2tzvp': {
+        'H': -0.50224721, 'C': -37.84235323, 'N': -54.58757237, 'O': -75.06982154,
+        'F': -99.74180094, 'S': -398.10765642, 'Cl': -460.14650064
+    },


Figured I would chime in here. So these were fitted and not just calculated?

cgrambow · 2020-04-30T19:39:59Z

arkane/encorr/data.py

+    'O-O': 0.09663, 'C=S': 0.80888, 'H-S': 1.23648, 'C=O': 0.88066, 'C=N': 0.03998,
+    'H-N': -0.49564, 'H-O': -0.41183, 'H-H': 0.6263, 'N#N': 3.71325, 'N-N': 0.74915, 
+    'N-O': -0.62156, 'C=C': -0.63901, 'O=S': -1.39626, 'O-S': -1.37002, 'S-S': 0.1515,
+    'F-S': -0.68693, 'F-O': 0.09202, 'F-H': -1.68214, 'F-F': 0.95483, 'O=O': -2.64949


These corrections are obtained from calculations that use dlpno-ccsd(t)/def2-tzvp for both energy calculations and geometry optimizations? If the geometry optimization was done with a different level of theory, it should be included in the model chemistry.

cgrambow · 2020-04-30T19:42:54Z

documentation/source/users/arkane/input.rst

 ================================================ ===== ==== ==== ========== ====================

 Notes:

 - The ``'CBS-QB3-Paraskevas'`` model chemistry is identical to ``'CBS-QB3'`` except for BCs for C/H/O bonds, which are from Paraskevas et al. (DOI: 10.1002/chem.201301381) instead of Petersson et al. (DOI: 10.1063/1.477794). Beware, combining BCs from different sources may lead to unforeseen results.
 - In ``'M08SO/MG3S*'`` the grid size used in the [QChem] electronic structure calculation utilizes 75 radial points and 434 angular points.
+- In ``DLPNO-CCSD(T)/def2-tzvp`` run using Orca 4.2.1 with NormalPNO
+- In ``wb97xd/def2tzvp`` run using Gaussian16. Do n't use the AEC and BC for wb97xd calculated with Qchem.


I'm not sure we should be using the notation without hyphens. We could consider using wb97x-d/def2-tzvp to be consistent with notation in the literature rather than notation Gaussian uses. I doubt any wb97x-d calculations were done using Q-Chem because you would usually use wb97x-d3 in Q-Chem.

amarkpayne · 2020-04-30T21:24:40Z

I'll let Duminda respond to the comments as well, but my understanding for the lack of hyphens in certain place is to distinguish that this model chemistry is run using Gaussian, which does not use hypens for this in the input files.

This should probably be part of a separate discussion on the future of model chemistry. In ARC, they have separate variables for "level of theory" and "model chemistry". I am not certain on what they envisioned the difference between these two to be, but I think that we need to make a distinction between the verbatim text that must be supplied to QM software to specify the level of theory and associated items like grid size and tolerances, and on the other hand have a string/dictionary/something that uniquely identifies the model chemistry to the user in a standardized way, that ideally can be used as keys in the reference species database.

@alongd , @oscarwumit , @dranasinghe , @cgrambow it might be worth having a discussion over Zoom soon to discuss how to standardize model chemistry/level of theory between ARC, Arkane, and the reference species database. Thoughts?

…_def2tzvp_BAC

dranasinghe

Sorry for the late reply. I made the changes you requested. I added Melius-type bond additivity correction parameters and push the changes.

dranasinghe · 2020-05-08T22:55:16Z

arkane/encorr/data.py

+    # Calculated atomic energies fitted orca 4.2.1  dlpno-ccsd(t)/def2-tzvp NormalPNO
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 7.99 and 5.96 kJ/mol respectively.
+    'dlpno-ccsd(t)/def2-tzvp': {
+        'H': -0.49641082, 'C': -37.77274125, 'N': -54.50388932, 'O': -74.96760414,
+        'F': -99.62044819, 'S': -397.63480236, 'Cl': -459.65784960
+    },
+    # wb97xd/def2tzvp conducted using G16. 
+    # fitted using Colin's BAC algorithm and SOC are included in the correction
+    # AEs are fitted to neutral molecules with RMSE and MAE 9.16 and 6.66 kJ/mol respectively.
+    'wb97xd/def2tzvp': {
+        'H': -0.50224721, 'C': -37.84235323, 'N': -54.58757237, 'O': -75.06982154,
+        'F': -99.74180094, 'S': -398.10765642, 'Cl': -460.14650064
+    },


dranasinghe · 2020-05-08T22:59:36Z

arkane/encorr/data.py

+    'O-O': 0.09663, 'C=S': 0.80888, 'H-S': 1.23648, 'C=O': 0.88066, 'C=N': 0.03998,
+    'H-N': -0.49564, 'H-O': -0.41183, 'H-H': 0.6263, 'N#N': 3.71325, 'N-N': 0.74915, 
+    'N-O': -0.62156, 'C=C': -0.63901, 'O=S': -1.39626, 'O-S': -1.37002, 'S-S': 0.1515,
+    'F-S': -0.68693, 'F-O': 0.09202, 'F-H': -1.68214, 'F-F': 0.95483, 'O=O': -2.64949


I am not sure how to include it current structure. So I left a comment.

dranasinghe · 2020-05-09T23:00:22Z

Sorry I realized bac has moved to RMGdatabase. I will make a new PR to RMG database

amarkpayne · 2020-05-12T19:05:22Z

Everything in this PR with the exception of the documentation update has been moved to ReactionMechanismGenerator/RMG-database#404. I'll close out this PR, but we'll add the documentation update in a later PR along with the other new model chemistries once the level of theory standardization PR #1940 has been merged in.

Atom Corrections ad Bond additivity Correction for DLPNO-CCSD(T)/def2…

65578e7

…-tzvp normalPNO and wb97xd/def2tzvp were added.

dranasinghe requested review from alongd, amarkpayne and oscarwumit April 27, 2020 03:10

alongd changed the title ~~Atom Corrections ad Bond additivity Correction for DLPNO-CCSD(T)/def2…~~ Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) Apr 27, 2020

alongd reviewed Apr 27, 2020

View reviewed changes

dranasinghe added 2 commits April 28, 2020 00:43

add model chemistry to documentation

6d49741

fix-up formatting.

9d647c2

Merge branch 'master' into DLPNO_def2tzvp_BAC

92e5d8e

alongd reviewed Apr 28, 2020

View reviewed changes

amarkpayne reviewed Apr 28, 2020

View reviewed changes

This was referenced Apr 28, 2020

Move Arkane Data to RMG-database #1870

Merged

Add data for AEC, BAC, and frequency scale factors ReactionMechanismGenerator/RMG-database#403

Merged

cgrambow requested changes Apr 30, 2020

View reviewed changes

cgrambow mentioned this pull request Apr 30, 2020

Add in DLPNO-CCSD(T)/def2-tzvp//wb97xd/def2tzvp model chemistry into the reference database ReactionMechanismGenerator/RMG-database#404

Draft

dranasinghe added 2 commits May 9, 2020 18:48

add Melius-type bond additivity correction parameters

18562a2

Merge remote-tracking branch 'official/DLPNO_def2tzvp_BAC' into DLPNO…

630fa34

…_def2tzvp_BAC

dranasinghe commented May 9, 2020

View reviewed changes

amarkpayne closed this May 12, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) #1935

Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) #1935

dranasinghe commented Apr 27, 2020 •

edited by alongd

Loading

amarkpayne commented Apr 27, 2020

codecov bot commented Apr 27, 2020 •

edited

Loading

alongd left a comment •

edited

Loading

alongd Apr 27, 2020

dranasinghe Apr 28, 2020

alongd Apr 27, 2020

cgrambow Apr 30, 2020

dranasinghe May 8, 2020

alongd Apr 27, 2020

dranasinghe commented Apr 28, 2020

alongd left a comment

alongd Apr 28, 2020

alongd Apr 28, 2020

cgrambow Apr 30, 2020

amarkpayne left a comment

amarkpayne Apr 28, 2020

cgrambow Apr 30, 2020

dranasinghe May 8, 2020

amarkpayne Apr 28, 2020

amarkpayne Apr 28, 2020

cgrambow left a comment

cgrambow Apr 30, 2020

cgrambow Apr 30, 2020

cgrambow Apr 30, 2020

amarkpayne commented Apr 30, 2020

dranasinghe left a comment

dranasinghe May 8, 2020

dranasinghe May 8, 2020

dranasinghe commented May 9, 2020

amarkpayne commented May 12, 2020

Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) #1935

Atom Corrections and Bond additivity Correction for DLPNO-CCSD(T) #1935

Conversation

dranasinghe commented Apr 27, 2020 • edited by alongd Loading

Motivation or Problem

Description of Changes

Testing

Reviewer Tips

amarkpayne commented Apr 27, 2020

codecov bot commented Apr 27, 2020 • edited Loading

Codecov Report

alongd left a comment • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dranasinghe commented Apr 28, 2020

alongd left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

amarkpayne left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cgrambow left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

amarkpayne commented Apr 30, 2020

dranasinghe left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dranasinghe commented May 9, 2020

amarkpayne commented May 12, 2020

dranasinghe commented Apr 27, 2020 •

edited by alongd

Loading

codecov bot commented Apr 27, 2020 •

edited

Loading

alongd left a comment •

edited

Loading