Why functionals better B3LYP, failing to explain the experimental...

maximos

Hi
I performed a computational study of an organic reaction in order to explain the observed regioselectivity. I chosen the wB97XD/6-311G(d,p) level for the study, but unfortunately the obtained results predict a contradectory regioselectivity of the experimental (Transition states energies), I tested other functionals, M06-2X and B3LYP-D3(BJ), and I get the same reversed regioselectivity, I came back to the old functional B3LYP, the results deal with the experimental. I get the same problem with other reactions, when B3LYP succed to explain the experimental results other functionals better than B3LYP can't.
I didn't understand this, is there a reasonable explanation for this issu ?

sobereva

If you're not a native Chinese speaker, please post your question in English, then the content will be much easier for us to smoothly understand... Posting in English is highly welcomed in this forum.

maximos

sobereva 发表于 2023-2-2 16:36
If you're not a native Chinese speaker, please post your question in English, then the content will  ...

I'm so sorry for the inconvenience, I modified the post.

wzkchem5

Please note that wB97XD and M06-2X are not necessarily better than B3LYP. Indeed, the former two have higher amounts of HF exchange, which make them more suitable in cases where self-interaction error is important (e.g. radical ions, charge transfer). However, this also makes them less capable of describing non-dynamic correlation, and fail for systems where static correlation is important (B3LYP is also unreliable under these cases, but usually less so than functionals with higher amounts of HF exchange).
I would recommend you use a method that is fundamentally better than the methods that you are using. If your system is indeed manifestly multi-referential, then NEVPT2 may be a good idea. Otherwise, DLPNO-CCSD(T) can generally be recommended. If DLPNO-CCSD(T) is not affordable, then double hybrids like PWPB95-D3 may also be considered.

maximos

wzkchem5 发表于 2023-2-2 17:45
Please note that wB97XD and M06-2X are not necessarily better than B3LYP. Indeed, the former two hav ...

Dear wzkchem5,
Thank you very much for your reply,but, all the alternatives you mentioned is out of my reach, due to i only can use Gaussian 09 and  the modest computational resources.
I work with molecules contain at the most 50 atoms, Is there another alternative to overcome this issue ?

wzkchem5

maximos 发表于 2023-2-2 11:12
Dear wzkchem5,
Thank you very much for your reply,but, all the alternatives you mentioned is out  ...

You can use ORCA for free as long as you are an academic user, and ORCA is easy to learn, particularly if you only use a few features of ORCA. This is unlike Gaussian, which comes at a price.
Within Gaussian you may be able to do a brute-force B2PLYP calculation, for example. But that only works when your system is single-referential. If your system is multireferential, then sooner or later you'll have to switch to ORCA or Molpro etc.

maximos

wzkchem5 发表于 2023-2-2 18:47
You can use ORCA for free as long as you are an academic user, and ORCA is easy to learn, particul ...

I uploaded a transition state of the system as gjf file to see if my chosen level is suitable.
TS.gjf (3.49 KB, 下载次数 Times of downloads: 14)

2023-2-3 00:33 上传 Uploaded

点击下载
Click to download

sobereva

The quality of 6-311G(d,p) is insufficient. I suggest using def2-TZVP for evaluating single point energies based on your optimized geometries.

Contradiction with experimental observation is not always caused by the choice of theoretical method, you should also pay attention to other factors, especially thermodynamical correction (Gibbs free energy should be used instead of electronic energy) and solvent effect (solvation model should be employed).

LiHuaYu

What I can see is the TS was located in gas phase. Is it the condition that experiment was carried out? I think you should strongly consider factors that @sobereva mentioned: solvent effects and thermodynamics. Your system is not that big, so I suggest that electronic energies need to be improved based on your final TS geometry. Def2TZVPD or a bigger basis set is highly recommended here. You also might need to pay attention to standard condition (1M) if you would like to identified your TS in a solvent.

Here is my favorite solvent model: scrf(smd, solvent = your_solvent).

七尺贱

LiHuaYu 发表于 2023-2-5 13:18
What I can see is the TS was located in gas phase. Is it the condition that experiment was carried o ...

游客，本帖隐藏的内容需要积分高于 25 才可浏览，您当前积分为 0

wjc404

maximos 发表于 2023-2-3 00:33
I uploaded a transition state of the system as gjf file to see if my chosen level is suitable.

...

The static correlation effect of the TS structure seems moderate.

(Some groups far from the reaction site are replaced by H atoms to make the calculations affordable. )  

(1) The TS structure was optimized at B3LYP-D3/def-TZVP level with IEEPCM solvation model (in dichloromethane). The optimized geometry is given below.

1. C                 -0.08111300   -0.55990300    0.67244800
   
2. C                 -1.17341100   -1.32475300    0.38458300
   
3. C                  0.98650900   -0.27212900   -0.33182000
   
4. C                 -2.74038400    0.08636000   -0.09259700
   
5. H                 -1.75568100   -1.76210300    1.18578600
   
6. H                 -1.25242100   -1.80361900   -0.58522200
   
7. O                 -1.07705300    1.60051300    0.39567400
   
8. H                  0.13735200   -0.29301000    1.69948700
   
9. C                  1.71205000    1.07309100   -0.14580300
   
10. O                  2.06608800   -1.24717800   -0.21153800
    
11. H                  0.58604400   -0.37118800   -1.34228500
    
12. C                 -4.01310700   -0.53412200   -0.47423800
    
13. N                 -2.18034900    1.13538100    0.06949600
    
14. O                  3.08668700    0.76255300   -0.39456500
    
15. H                  1.39143000    1.83331200   -0.85064600
    
16. H                  1.58380900    1.44324100    0.87615700
    
17. C                  3.23530900   -0.52745900    0.14965000
    
18. H                 -4.75336900    0.23659000   -0.69168700
    
19. H                 -3.88186000   -1.15774400   -1.35975400
    
20. H                 -4.38659000   -1.16438800    0.33418400
    
21. H                  3.32381700   -0.47381600    1.24447100
    
22. H                  4.10901200   -1.00856500   -0.28687200

复制代码

(2) The stable UHF wave function of the above structure was obtained (S2=0.847). It was used as the initial guess for a Brueckner-CCD calculation. The latter yielded a reference wave function with <S**2> lower than 0.001, indicating the spin polarization has been removed. The maximum double excitation amplitude from that reference is 0.079.

(3) The natural orbitals at Brueckner-CCD/def-TZVP level were then generated. The result can be downloaded from https://pan.baidu.com/s/1nG6kRcrTkE75dW5Xoqk8Ew with extraction code bkxa. From the occupation numbers (each ≤0.10 or ≥1.90) we can see that this structure is not strongly correlated.

Since the multireferential character is not significant, DLPNO-CCSD(T) should be OK here for single-point energy correction of TS.

sobereva

七尺贱 发表于 2023-2-5 20:39
**** 本内容被作者隐藏 ****

The current topic is quantum chemistry rather than wavefunction analysis or Multiwfn. Discussing in this forum is fully appropriate.

maximos

wjc404 发表于 2023-2-5 21:40
The static correlation effect of the TS structure seems moderate.

(Some groups far from the rea ...

Dear wjc404,
Thank you very much for your valuable effort.

maximos

sobereva 发表于 2023-2-3 02:09
The quality of 6-311G(d,p) is insufficient. I suggest using def2-TZVP for evaluating single point en ...

Dear Pr. Sobereva,
I would like to add that I performed a Mulliken atomic spin density analysis (Parr indices) in order to explore the reactive sites, and the most favorable interactions, and a perfect agreement with the experimental was found. (See attached images).