Some tchnical questions (.INTFLG, xyz)

[Balogh Szabolcs]

Dear Dirac users!

Slowly getting familiar with the program, sadly i have just a few hours per day but it really helps a lot recalling what i did previous day with the automatic renaming of the same outputs.

Also i really like the easy combination of inputs and structures. Very convenient to do one project in one folder.

Just made the next test isopropyliodide 13C NMR shift calculation gave 20ppm for the CH and 34ppm for the Me while the measured value is 22ppm and 31ppm respectively. While non-relativistic but high quality DFT gives 40-50ppm for the CH and the Me is the same (two baonds from I weaker rel. effect i guess).

And this one was made still on a non-rel DFT optimized structure and without solvent.

I can see in the output that even if i set .INTFLG to (1 1 1) under **HAMILTONIAN only the LL and LS integrals (Default) are involved. In my understanding the SS could be significant for shileding calculation (and i am not sure about importance of the GT integrals), at least i would like to see their effect. The calculation goes well and the results are satisfying but still. Is the syntax wrong?

Lines 951-960

I am using xyz for the structure input and i am asking for the calculation of 3  atomic centers namely 1, 2 and 7 (lines 285-300) but the program changes the order (527-542) but still calculates the atoms 1, 2 and 7 and the latter is changing to H and it is calculated instead of the C.

What is the logic in the atom order here since it is not the atomic number? How should i arrange the atom order in the xyz? With big complexes it will be a mess. Should i use .mol files instead? Or i can start the calculation, stop it and replace the contents of the xyz with the rearranged coordinate list. 

Thank you for your time if you check on these questions!

Szabolcs

[Balogh Szabolcs]

Sorry for the many questions :) Now i need to ask about structure optimization.

First i tried to run them with xyz inputs. It looked fine but it seems that the program is just running in circles. The structure is not changing but it is working for a long time on just a simple molecule like TMS or iPr-iodide.

Then based on the Troubleshooting section i tried it with mol files (after struggling a lot but finally learning from an old post that atomic charge needs to be put in columns 11-15 XD) but the result seems like the same. The computer is working a lot but the initially good structure (optimized with DFT in other software) is messed up, the equal Si-C distances getting different and i dont see convergence.

The test optimization jobs finishing fine.

I dont know what could be wrong with the input maybe?

Could you please take a look at the attached example?

The input is a modification of the dft/lda/opt test example. Just replaced LDA with PBE0

The file is to big i guess

https://drive.google.com/file/d/1h2-xLcxDzpJaw9Vw6O6rtFHF8SDQwXix/view?usp=drive_link

[Miroslav Iliaš]

Hello,

switch your  xyz input for mol input for the geometry optimization.

M

--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dirac-users/a8f36517-6d38-4b2c-a225-090bdd31c3cdn%40googlegroups.com.

[Balogh Szabolcs]

Dear Miroslav, 

Thank you for the swift response! 

I did! But the optimization calculation seems to be running in circles. This is why i am asking.

Best regards,

Szabolcs

[Miroslav Iliaš]

Hello,

In such a case try to perform pre-optimization with cheaper methods - try Avogadro with some force fields, the MOPAC or xtb to see if there is some circling of the relaxing geometry.

After you may try nwchem with scalar relativistic DFT structure relaxation.

At the end, apply DIRAC x2c (with spin-orbit on) geometry relaxation of such "pre-processed" structure. 

Best,

M

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/a64aa3bb-45e9-44dc-a8a7-4fd906fe646en%40googlegroups.com.

[Kenneth Dyall]

If you have methyl groups in your structure, you might simply be getting rotation of these groups, which have only a small barrier to rotation and can easily move a lot in a geometry optimization. 

Ken.

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/CA%2BGrSJxQr3nHp8SaQ_Vmu-Lfo1yo6H4cJW-yjRKfSAF3USt_uA%40mail.gmail.com.

[Balogh Szabolcs]

Thanks both of you for the feedback. It was important that i am not making something totally wrong. I have experience with NWChem and Orca but Dirac by default behaves differently. I will play with it and maybe look around at the structure convergence criteria parameters.

I have still a lot of questions but also some ideas how to resolve them. I will ask if i am stuck :)

[Hans Jørgen Aagaard Jensen]

Dear Balogh,

 

I have looked at your output.

1. It is not running in circles, but it is calculating the molecular gradient numerically because analytical gradient is not implemented for your calculation. To calculate the numerical gradient with central differences one needs 2*(3*N) + 1 energy evaluations. For N=17 atoms, this is 103 energy evaluations! This is the reason for the many total energies, and that all 103 numbers are almost equal, some a little up, some a little down compared to the reference energy (the first calculation)
2. However, your calculation revealed a bug: Dirac tries to find the symmetry for each of the 103 close-lying geometries, and for that purpose it moves the molecule. This gave errors in the molecular gradient calculation. To do the calculation you want, you should change your .mol file to:

INTGRL

 

 

C   3    0

        6.   4

C       -1.4690315229      3.2081156044      0.3172840864

C        0.2660889149     -0.8428584389     -3.4301265473

C       -2.0234085498     -2.3941639827      1.6503044899

C        3.2264700405      0.0287710782      1.4622895286

LARGE BASIS dyall.2zp

        1.   12

H        4.4539446366      1.4083909718      0.5374366161

H        3.1346866728      0.5162888437      3.4670777957

H        4.1335086195     -1.8201946547      1.3078484030

H       -1.5885217507     -0.8758821376     -4.3381813643

H        1.4373351678      0.5210816423     -4.4464974722

H        1.1183573011     -2.7077461904     -3.6772287427

H       -2.2160920414     -1.9533800479      3.6587249993

H       -3.9209410343     -2.4569433067      0.8374984012

H       -1.2138736146     -4.2881240724      1.4989233598

H       -0.3308652258      4.6482129469     -0.6292264844

H       -3.3568086371      3.2517152868     -0.5191511739

H       -1.6496247177      3.7554915933      2.3004418884

LARGE BASIS dyall.2zp

       14.   1

Si      -0.0001298998      0.0001359658      0.0002731788

LARGE BASIS dyall.2zp

FINISH

 

The red 0 in column 10 is the change compared to your .mol file, it tells Dirac to use zero symmetry operators, which is the same as telling Dirac not to attempt to use symmetry. I quickly tested this change with your molecule, but STO-3G and LDA, and the geometry converged in six geometry iterations.

 

Two more comments:

I recommend to use the standard DFT grid, i.e. remove the *GRID section in the .inp file. Do not use the lousy dft grid from the test suite, it is there to test that setting your own grid works.

I can also see that the output becomes much too verbose with 103 SCF evaluations for each molecular gradient (1.3 million lines in my STO-3G, LDA test), I will reduce that in next version of DIRAC.

 

Best regards, Hans Jørgen.

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/865fbaed-c788-44d5-9dc5-5f6554ba1cc5n%40googlegroups.com.

[Balogh Szabolcs]

Dear Hans Jørgen

Thank you for your time looking into this and for the solution! Really appreciate it! I can try this tonight, cant wait!

Yes the grid is there becaus in my desperate attempts was using /build/test feils as a base.

Best regards, 

Szabolcs

[Balogh Szabolcs]

Dear Hans Jørgen,

Seemingly the optimization worked after adding the 0 to column 10 in the .mol file (see attached file geometry-converged.out).

Was this an analytical solution, since it converged in 4 steps? 

Then after making the pcm module work wanted to repeat the same optimization (starting from the optimized structure output from the calculation above) but with including pcm effects.

Basically i have just added the pcm block to the input in the working optimization input above but it again does not mek the structure optimization.

This block is only at the beginning of the output while it repeats in the working optimization job:

  Optimization information
                             ------------------------

 Iteration number               :       3
 End of optimization            :       T
      ****  Geometry optimization has converged  ****
 Energy at this geometry is     :  -28514.225958539719
 Energy change from last geom.  : -2.445668D-07 (the threshold is: 1.00D-06)
 Norm of gradient               :  2.830911D-05 (the threshold is: 1.00D-04)
 Norm of step                   :  5.745541D-04 (the threshold is: 1.00D-04)
 Updated trust radius           : 5.000000D-01
 Total Hessian index            :       0

I dont get what is going on. 

May i ask you to take a look into these two outputs please? 

h...@sdu.dk a következőt írta (2025. szeptember 2., kedd, 22:37:47 UTC+2):

[Hans Jørgen Aagaard Jensen]

Yes, Dirac calculated the molecular gradients analytically. This is NOT related to convergence in 4 geometry steps, the geometry convergence algorithm is the same no matter how the gradients were evaluated (as long as the molecular gradient is correct, of course).  Thus, it would have been the same convergence with numerical molecular gradients.

 

For the pcm case analytical molecular gradients is not implemented, and you stopped it before finishing the calculation of the first numerical gradient.

 

Regards, Hans Jørgen.

To view this discussion visit https://groups.google.com/d/msgid/dirac-users/411083b6-d3f8-43a9-a184-df4c1eb92278n%40googlegroups.com.