New HF code convergence problem

[thierry....@gmail.com]

Dear all

we hit a serious convergence problem with the HF code in Molpro2025.

We tried to redo some old calculations which converged without any problem using the old Molpro2009 HF code. The systems studied are carbon nanotubes and nanocones.

Unfortunately, the only way to make these converge with the new HF is to remove random (it depends on the system) pairs of electrons to obtain starting orbitals.

We tried several option of the new code (SO-SCI, level shifts, algorithm,...) nothing helped.

I join the input and the corresponding Molpro2009 and Molpro2025 outputs.

Thank you for any hint you may bring.

Best wishes,

 Thierry

[tibo...@gmail.com]

Dear  Thierry ,

We have also been seeing strangely worse HF/KS convergence ever since the default HF program was replaced in version 2021.1.

But for us SO-SCI, or SO usually works, albeit slower than the old code.

My recommendation is to try using the old SCF code, accessible by the old option, eg. {hf,old}.

Best regards,

Tibor Győri

[hjwern...@gmail.com]

Dear Thierry & All,

this is due to a poor starting guess. By default, it uses a minimal basis for generating the guess, but for reasons that I do not yet fully understand this does not work well in the current case. You can use 

hf,guess_basis=vtz

then it converges without problems (no so-sci and no shifts are needed). I'll investigate the problem further and will try to fix it in the upcoming next release.

Best wishes

Joachim

[Thierry Leininger]

Thank you Joachim.

Best wishes,

  Thierry

-- 
You received this message because you are subscribed to the Google Groups "molpro-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to molpro-user...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/molpro-user/c89466df-0e5e-42d3-9bb6-b3e6df8b1e09n%40googlegroups.com.