Davidson Diagonalization fails in Orca

andrzejek · 2025-01-23 07:57:54

Hi,
I study a system contained of C, H and O in its neutral and anions forms in Orca 6. Therefore, I decided to use ma-def2-TZVP basis set. While optimization goes fine, generation of the absorption spectra fails at Davidson Diagonalization procedure. I tried using AutoAux and def2J but without a success. This is an aromatic system so I cannot limit diffused basis set just to oxygens undergoing deprotonation. What else could I do? My main route part is:

! LibXC(TPSSh) ma-def2-TZVP AutoAux TightSCF D4 SMD(H2O)

Thanks

------------------------
DAVIDSON-DIAGONALIZATION
------------------------

Dimension of the eigenvalue problem            ...  34500
Number of roots to be determined               ...     40
Maximum size of the expansion space            ...    400
Maximum number of iterations                   ...    100
Convergence tolerance for the residual         ...    2.500e-07
Convergence tolerance for the energies         ...    2.500e-07
Orthogonality tolerance                        ...    1.000e-14
Level Shift                                    ...    0.000e+00
Constructing the preconditioner                ... o.k.
Building the initial guess                     ... o.k.
Number of trial vectors determined             ...    400


                       ****Iteration    0****
   Time for iteration     :  TOTAL=69.2 TRAFO=1.2 RIJ=6.8 COSX=27.6 XC=21.4
   Size of expansion space: 120
   Lowest Energy          : -94740.194830773849
   Maximum Energy change  : 94740.194830773849 (vector 0)
   Maximum residual norm  : 1023356330656.517578125000

Last edited by andrzejek (2025-01-23 07:59:05)

sobereva · 2025-01-24 01:57:15

Using "ma-def2-TZVP AutoAux" usually is not a good choice, strong linear dependency problem often occurs.
Consider:
(1) Do not use RIJCOSX and hence auxiliary basis set is not needed.
(2) Do not add diffuse functions
(3) Using def2/J as auxiliary basis set. Generally, it is found that def2/J still works reasonably for ma-def2-TZVP.
(4) Using aug-cc-pVDZ (or TZ), and use aug-cc-pVDZ/JK (or TZ) as auxiliary basis set.

Best,

Tian

andrzejek · 2025-01-24 08:35:34

Dear Tian,
Thanks for your suggestions. I might indeed check with (1) and (3). On the other hand, (2) doesn't seem legit if I study anion and later will compute electron affinities, does it? Option (4) also sound reasonable, my system is not so big, but aren't Dunning basis set quite demanding and less efficient in DFT and SCF procedure? Have read somewhere about it.

In the meantime, I checked some simpler functional - O3LYP instead of TPSSh and as for now looks fine. Given the absorption spectrum are coherent with the experimental one, I suppose the swap is validated?

Best,

sobereva · 2025-01-25 09:50:49

To calculate electron affinity, diffuse functions are absolutely needed.

aug-cc-pVDZ is not expensive. Dunning's basis sets are not preferential choice for DFT calculation of thermodynamic data, but for electron excitation studies, it is usable.

O3LYP is a rarely used functional, I do not recommend it. Generally, for calculating electronic absorption spectrum, DFT functionals with HF composition lower than 20% are not recommended, because excitation energies are usually notably underestimated. O3LYP only contains 11.61% HF composition.

andrzejek · 2025-01-26 08:17:48

Dear Tian,

To calculate electron affinity, diffuse functions are absolutely needed.

That's why I switched from the regular def2-TZVP to ma-def2-TZVP. Apparently, on another forum, people are arguing that Karlsruhe's basis sets are developed in a way that they don’t require diffuse functions as much as, for example, Pople’s basis sets. As many options exist, there are as many opinions... However, to avoid receiving negative comments from reviewers, I decided to stick with an augmented basis set.

aug-cc-pVDZ is not expensive. Dunning's basis sets are not preferential choice for DFT calculation of thermodynamic data, but for electron excitation studies, it is usable.

Okay, but shouldn’t I actually use aug-cc-pVTZ? I read somewhere that "For DFT calculations, the aug-cc-pVnZ basis set family may not be the best choice (non-ideal contraction for DFT, overly large and often poor SCF energies)". My system is also relatively small, with no more than 30 heavy atoms.

O3LYP is a rarely used functional, I do not recommend it. Generally, for calculating electronic absorption spectrum, DFT functionals with HF composition lower than 20% are not recommended, because excitation energies are usually notably underestimated. O3LYP only contains 11.61% HF composition.

I am aware that O3LYP is uncommon, but only it and TPSSh yield reasonable excitation energies compared to other functionals—no more than 10 nm greater than experimental results. I tested a bunch of them.

Last edited by andrzejek (2025-01-26 10:31:13)

sobereva · 2025-01-26 16:23:11

1 Although def2-TZVP is more diffuse than 6-311G series of basis set, for the cases that diffuse functions are absolutely needed (e.g. electron affinity), def2-TZVP is still not diffuse enough.

2 If you can afford, certainly aug-cc-pVTZ is much better choice than aug-cc-VDZ. aug-cc-pVDZ is just the lowest acceptable basis set for this purpose.

3 Degree of agreement with the experiment should never be simply taken as the sole criterion for selecting the functional. One should consider all possible sources of discrepancy with experimental observation (e.g. conformation, solvation effect, protonation state, vibronic effect, double-excitation, and so on).

andrzejek · 2025-01-26 19:34:59

3 Degree of agreement with the experiment should never be simply taken as the sole criterion for selecting the functional. One should consider all possible sources of discrepancy with experimental observation (e.g. conformation, solvation effect, protonation state, vibronic effect, double-excitation, and so on)

Yes, I am aware of that but right now referring solely to the excitation which is probably the most relevant at the moment of discussing whether O3LYP is fine?

sobereva · 2025-01-27 06:34:31

If you believe the source of error exclusively comes from functional, compared to using O3LYP, I would rather to recommend considering more advanced and robust methods, for example:
(1) w-tuned functional (e.g. LC-wPBE*)
(2) TDDFT with double hybrid functional (e.g. TD-DSD-PBEP86, TD-wB2GP-PLYP)

andrzejek · 2025-01-27 08:51:02

I use Orca. LC-wPBE is not available. On the other hand, double hybrids are too expensive, I believe. Would love to sth like TPSSh but cannot due to the linear dependence as I told. And, as I said earlier, only O3LYP and TPSSh yielded reasonable vertical excitation energies so least their use is validated. I tested b3lyp, b3lyp/g, b3pw91, b97, bhandhlyp, cam-b3lyp, lc-blyp, lc-pbe, m06, m062x, m06l, o3lyp, pbe0, revTPSS, tpss, tpssh,wb97, wb97x, x3lyp...

Multiwfn forum

#1 2025-01-23 07:57:54

Davidson Diagonalization fails in Orca

#2 2025-01-24 01:57:15

Re: Davidson Diagonalization fails in Orca

#3 2025-01-24 08:35:34

Re: Davidson Diagonalization fails in Orca

#4 2025-01-25 09:50:49

Re: Davidson Diagonalization fails in Orca

#5 2025-01-26 08:17:48

Re: Davidson Diagonalization fails in Orca

#6 2025-01-26 16:23:11

Re: Davidson Diagonalization fails in Orca

#7 2025-01-26 19:34:59

Re: Davidson Diagonalization fails in Orca

#8 2025-01-27 06:34:31

Re: Davidson Diagonalization fails in Orca

#9 2025-01-27 08:51:02

Re: Davidson Diagonalization fails in Orca

Board footer