Singular Matrix Error in 03,04-dmet-nio-afm,fm Example

[chemsm]

Hello, I am a researcher trying to reproduce the 03-dmet-nio-afm, 04-dmet-nio-fm problem example provided in your library as I explore using DMET for my studies.

nio_afm_output.txt

nio_fm_output.txt

Reproducing the both code the example 03-dmet-nio-afm and 04-dmet-nio-fm example as provided leads to a UICCSD calculation error displaying the following message:

1) 03-nio-afm case (2 hours running)

******* WARN: diis singular, eigh(h) [-9.2521e-07 4.0011e-09 1.4570e-07 1.9689e-07 2.0508e+06 4.3446e+06 4.0183e+09]

---

This causes the computation to halt. The traceback is as follows:

Traceback (most recent call last):
File "/home/chemsm/programs/anaconda3/lib/python3.11/site-packages/numpy/linalg/linalg.py", line 112, in _raise_linalgerror_singular
raise LinAlgError("Singular matrix")
numpy.linalg.LinAlgError: Singular matrix

2) 04-nio-fm case (20 minutes running)

******* WARN: diis singular, eigh(h) [ -0.0001 -0. 0. 0. 202.2266 32164.7754 13988017.9382]

---

This causes the computation to halt. The traceback is as follows:

Traceback (most recent call last):
File "/home/chemsm/programs/anaconda3/lib/python3.11/site-packages/numpy/linalg/linalg.py", line 112, in _raise_linalgerror_singular
raise LinAlgError("Singular matrix")
numpy.linalg.LinAlgError: Singular matrix

[zhcui]

Hello chemsm,

Thanks for reporting this.

From the output, the UHF-in-UHF embedding is exact. However, the UCCSD based on the UHF looks weird. The t amplitudes becomes diverge after a few CC iterations.
This example has been tested before (with pyscf 2.0) and gives typical results. I will try to reproduce your error based on the latest version.

I will get back to you after I re-run it.

[chemsm]

Thank you for your reply.

Since I am unable to fix the code on my own, I will await further communication from the developer regarding updates or improvements that will render the code executable in the latest version of PySCF.

Thank you,

chemsm

[zhcui]

Hi chemsm,

I rerun the example scripts.

However, I can't reproduce your error when using the latest version of pyscf.
Here is my output:
03-dmet-nio-afm.txt

It looks like we can get the same periodic HF solution, and in the CC solver, HF energy is the same. However, MP2 energy is different. This is strange. Can you try to use the latest version of pyscf?

[chemsm]

Thank you for rerunning the example scripts. I also tried running my previous code(That is made a little modification where lines[877] = 'class CIAHMinimizer(ciah.CIAHOptimizerMixin):\n'.) again.
Additionally, I've been using pyscf version 2.5.0 all along.

I also tried running the code exactly as in your 03-dmet-nio-afm.txt output file, but I'm still getting the same singular matrix error.

Interestingly, when I download and run the nio_afm DMET code directly from its directory, I encounter an error that says "array must not contain infs or NaNs". I have attached the relevant files below.

In a nutshell,
I rerun

1. initial error code (a little modification ver) -> singular matrix error
2. your successful recent attached code -> singular matrix error
3. 03-afm original code & original path -> valueError: array must not contain infs or NaNs
4. 04-fm original code & original path -> singular matrix error

I attached 3, 4 output file.
nio_afm.txt
nio_fm.txt