Which solid_dmft tutorials are realistic on a local machine vs HPC?

[Curvokarol]

Hello,

I have a question about the practical use of "solid_dmft", and i would also like to ask for guidance from people who have already gone through this workflow.

I was able to install "TRIQS", "triqs_dft_tools", and "solid_dmft" locally, and i can read the provided 'svo.h5' tutorial file with 'SumkDFTTools', so the installation itself seems to work.

However, when I looked at the tutorial scripts more closely, I noticed examples using commands such as mpirun -n 72 solid_dmft and Monte Carlo settings like 'n_cycles_tot = 10e+6', which seem much more appropriate for HPC than for a local laptop/workstation.

So I would like to ask:

1. Which tutorials are realistic for a minimal local test run?
2. Which ones are mainly intended for cluster/HPC use?
3. Is there a recommended minimal or toy setup to test the full solid_dmft workflow on a personal machine?

I am also asking because i am trying to build a DFT+DMFT workflow for monolayer NbS2, and at this stage this is honestly both a technical question and a request for help. I already have the DFT side in progress, but I am still struggling with the practical DFT+DMFT implementation and would be very grateful for guidance from anyone who has experience with 'TRIQS', 'DFTTools', 'solid_dmft', or similar workflows.

Thank you.

[alberto-carta]

Ciao Curvokarol,

Alberto here.

For the tutorials we discussed this back in the day and we preferred being honest and giving a setup which gives reasonably converged DMFT calculations instead of giving toy examples.

Having said that, there has been incredible progress from the point of view of the solvers, both CTHYB and CTSEG have become incredibly efficient (factor of 10 or even more) in the past year thanks to the amazing work of the people at flatiron.

This means that you could take a subset of the calculations in the svo tutorial and probably run them overnight on a laptop.

Just for your information, we had a DFT school in Ghana last year and we managed to run the coarse phase diagram in two days on a workstation from 2019 using 10 cores in the mpirun call.

You can try and set

loc_n_max = 3
loc_n_min = 0

In the [solver] card in the toml, this kills the states with electron number more than 3 for the vanadium. Given this is V4+, so 1 electron in the t2g shell, it is very energetically unfavorable to have 4 electrons or more in it. This will remove highly unlikely states in the proposal and make the QMC in CTHYB even faster.

Please tell me if you need more help with the specifics of your material, if you do not feel comfortable with sharing your question on a public forum like github, feel free to contact me via email, I will be very happy to help you set up the workflow:

https://www.psi.ch/en/lms/people/alberto-carta

[Curvokarol]

Ciao Alberto,

Thank you so much for your message. I was truly very happy to receive your reply.

I have already tested your suggestion of setting

"loc_n_max = 3"
"loc_n_min = 0"

in the [solver] card of the .toml file, and it really helped. Your explanation was also very clear and made a lot of sense physically, especially considering the V$^{4+}$ configuration and the fact that states with 4 or more electrons in the (t_{2g}) shell are extremely unlikely.

I also want to say that I feel very grateful to have access to a community like TRIQS. It has been helping me a great deal, both in learning and in building my workflow. Beyond the technical side, it is genuinely encouraging to see an open-source scientific framework supported by such a generous and knowledgeable community.

I truly hope that one day I will be able to contribute actively to the growth of this framework and to give something back to the community that has already helped me so much.

Thank you again for your kindness, for your time, and for your willingness to help.

Warm regards,
Karollaine

[the-hampel]

Hi @Curvokarol ,

just to add to @alberto-carta comments. The tutorials here: https://github.com/TRIQS/tutorials/tree/3.3.x/AbinitioDMFT are all with solid_dmft and are specifically designed for running also on your laptop / workstation. We designed them for a TRIQS school and they are complementary to the tutorials on the solid_dmft documention page.

Best,
Alex

[Curvokarol]

Hi @the-hampel,

Thank you very much for your message and for adding this information!!!

I was very happy to learn that these tutorials were specifically designed to run on a laptop or workstation. That is extremely encouraging for someone like me, who is still setting up calculations and learning how to build a reliable workflow.

I also very much appreciate the care that went into preparing these materials for the TRIQS school. It is clear that they were created with accessibility and learning in mind, and that makes a real difference for people who are still growing in this field.

I have been learning a great deal from the TRIQS ecosystem, and I am genuinely grateful for how supportive and generous the community is. It is very motivating to see this kind of effort behind an open source scientific framework.

I truly hope that, with time, I will also be able to contribute actively to this community and to the development of this framework.

Thank you again for your kindness and for taking the time to clarify this.

Warm regards,
Karollaine