convergence issue

[mzmansour]

I want to determine the wavefunctions and energy levels of Argon configurations 1s1 2s2 2p6 3s2 3p6 6p1 1s1 2s2 2p6 3s2 3p6 7p1 1s1 2s2 2p6 3s2 3p6 8p1 1s1 2s2 2p6 3s2 3p6 9p1 1s2 2s2 2p6 3s1 3p5 6p2 1s2 2s2 2p6 3s1 3p5 7p2 1s2 2s2 2p6 3s1 3p5 8p2 1s2 2s2 2p6 3s1 3p5 9p2 my target configurations are the first four configurations (with J=1 and odd parity) and added the last 4 to make it easier to use HF . I have used different techniques such as 1- increasing the charge of the isodata file 2- changing the damping factor 3- mix the core with different configurations 3- using HF

but with no solution. I will be very thankfull if anyone could help to get the energy levels of the first four configurations. I am attaching the GRASP file for help 0-file.txt

[jongrumer]

Hi Mohammad,

Why are you including the 3p5 np CSF's in your list? Have a look at e.g. the book "Computational Atomic Structure: An MCHF Approach" for details on how to build a multireference and important substitutions.

Try to remove all 3p5 np CSF's and see what happens. Note that in general it can be hard to converge a single runt with multiple active p-shells, but try this first.

Jon

[mzmansour]

Thanks Jon, The reason for including 3p5 np is that these configurations have closed shells (1s2 2s2 2p6) so I could use in HF to get the initial estimation of wavefunction. when removing 3p5 np CSF's it is going more harder to solve.

I don't know what is the best guess for the multireference configurations to start !

Regards, Mohammad

[jongrumer]

Ah I didn't see you had 3s1 in those extra configurations - and that these are core-excited states with holes in 1s. I suggest you try to get convergence in rmcdhf on one 1s1 2s2 2p6 3s2 3p6 np1 group at the time, collect all wavefunction files, and then use rwfnestimate to combine them all in a final run with all target states where you optimize on all states in rmchdf - I'm not sure this is possible though, you might have to calculate each group separately. Bit give it a try. Unfortunately my experience with core-excited states is fairly limited.

For the MR - once you have a converged calculation in which you also have included some correlation (i.e. single and double substations to a new layer of orbitals), run jj2lsj and check the lsj.lbl files and include those states that contribute say more than +-0.01 or so for the mixing coefficients, maybe make this cut a bit higher.

[jongrumer]

About the initial estimates: maybe you can run on 1s1 2s2 2p6 3s2 3p6 without the outermost np electron first to get initial estimates for the core - no idea, but could be worth a try.

[cffischer]

Let me comment. If you want to compute the energies of 1s1 2s2 2p6 3s2 3p6 6p1 1s1 2s2 2p6 3s2 3p6 7p1 1s1 2s2 2p6 3s2 3p6 8p1 1s1 2s2 2p6 3s2 3p6 9p1 then you have an EAL calculation (not EOL) and the best code for finding these energies and the initial estimates is the spline program DBSR_HF. I think the command would be
dbsr_hf 'name' atom=Ar term=jj out_w=1 provided 'name'.c was the usual rcsf.inp file. It is the file that includes the CSFs for an EAL calculation.
I am not sure of the command -- you need to check this out. The out_w=1 gives you a 'name'.w file that could be used as input for GRASP using rwfnestimate.

[jongrumer]

@cffischer Fantastic Charlotte! @mzmansour EAL means "extended average level" - just ask if you need input on what it means. You'll find the code Charlotte refers to on CPC. Well shall soon put up a development version here on github.com/compas as well.

[cffischer]

I just noticed spmething strange -- I had 'name' in angular brackets and they did not display. I have editted the comment.

[jongrumer]

@cffischer Github does some formatting, i.e. enclosing a word in a pair of _ results in italic, enclosing in ** gives bold, a word within `` gives monospace code etx...maybe something like that happened 🤷 But it's great that one can edit of course!

You can also do full code blocks which is kinda of neat, and even specify the language for syntax highlighting, just do ```fortran followed by a new row and the code, and end with a new row, again with three apostrophes, e.g.

program hello
   implicit none
   
   print*, 'Hello Vancouver, how are you?'

end program hello

[mortenpi]

I had 'name' in angular brackets and they did not display.

Markdown supports (limited) HTML, so it most likely interpreted <name> as an HTML tag. Surrounding it with backticks (i.e. `<name>`) as Jon suggested would work around that.

[jongrumer]

I like how this thread now has turned into a markdown formating crash course 👌😆

[mzmansour]

Ok! Thanks very much, I am working on it and will give you the final results as soon as possible.

[jongrumer]

Great! It would be appreciated if you could summarize in detail how you used the method suggested by Charlotte. I.e what input commands you used in the codes etc.

[mzmansour]

Many thanks for all of you who let me get different directions of solutions. I made the multireference configurations as 1s1 2s2 2p6 3s2 3p6 6p1 1s1 2s2 2p6 3s2 3p6 7p1 1s1 2s2 2p6 3s2 3p6 8p1 1s1 2s2 2p6 3s2 3p6 9p1 1s2 2s2 2p6 3s1 3p6 6p1 1s2 2s2 2p6 3s1 3p6 7p1 1s2 2s2 2p6 3s1 3p6 8p1 1s2 2s2 2p6 3s1 3p6 9p1

and used HF and the results are going well.

The usage of the spline program DBSR_HF gave me an error

Regards, Mohammad

[jongrumer]

You have to provide some more info - what command did you run DBSR_HF with? As Charlotte said in her post, you need to double-check the command she suggested.

But you said that everything is running well, so maybe you don't need more input at this point.