SSCHA tutorial 2023: Tutorial 2: ADVANCED FREE ENERGY MINIMIZATION

[anyi-mxh]

Dear Developer
Hi, I am following this tutorial by following the video and pdf.
Following the tutorial, I performed 8 steps:

1. pw.x < pw.scf.in > pw.scf.out
2. ph.x < ph.in > ph.out
3. python generate_ensemble.py
4. bash generate_input_files.sh
5. mpirun -np NPROC pw.x -i input_file.pwi > output_file.pwo
6. bash get_energy_forces.sh
7. python minimize.py
8. python plotter.py
   I got bad results. The left part of the figure below is my result, and the right part is the result from the video.
   
   Kindly help me solving this problem.
   Here, I attached my input file.
   02test.zip
   In addition, I directly utilized the file from 02_manual_submission of SSCHA2023, and there are also some other problems.
   In 02_manual_submission of SSCHA2023, after executing bash generate_input_files.sh, the obtained structure_2.pwi is completely inconsistent with 02_manual_submission/data/input_files_population1/structure_2.pwi provided by SSCHA2023. This is very confusing. I did not change any parameters, but I got a different structure_2.pwi. The same problem exists for other structure_$i.pwi.
   
   What is the standard for setting the parameter "temperature" in generate_ensemble.py and minimize.py? In 02_manual_submission of SSCHA2023, the temperature of generate_ensemble.py is set to 300 K, while minimize.py is set to 0 K. Do the temperatures in these two files need to be consistent, or what is the purpose of the two temperature settings? If calculating the superconducting critical temperature, how to set the temperatures in these two files.
   

Please help me answer the above questions.

[mesonepigreco]

Hi,
The difference in results is likely due to the limited number of configurations being used—only 10 in this case (the same as we are using). This small sample size introduces significant stochastic error. While the seed should help ensure consistency, differences can still arise if you're using a different Python version or a different version of the NumPy library.

For a proper calculation, we would typically need between 500 and 1000 configurations to achieve a converged result.

Regarding the temperature, you are absolutely correct—the two temperatures must match. That discrepancy is indeed a mistake.

[anyi-mxh]

Dear Sir:
Thank you for your reply. In addition to the above questions, I have a new question.
I tried two solutions:
Set nq1=2, nq2=2, nq3=2 in the ph.in file, generate the harmonic_dyn file, and run python generate_ensemble.py and bash generate_input_files.sh in turn. generate_ensemble.py and bash generate_input_files.sh, and the resulting supercell is 2×2×2.
Set nq1=4, nq2=4, nq3=4 in the ph.in file, generate the harmonic_dyn file, and run the same commands in sequence. The resulting supercell is 4×4×4.

Is the size of the supercell generated according to the default settings of the program? Or does it need to be manually adjusted to meet specific needs? If we need to customize the size of the supercell, do we need to add specific parameters in generate_ensemble.py to achieve it?

Kindly help me solving this problem.