The calculations of phase transition TERMINATED due to the imaginary frequencies

[Sssss-jg]

Dear all,

I am currently developing my research topic on phase transitions based on the SSCHA 2026 school tutorials (namely Lecture2 and Hands-on-session2). But when I performed the CONFIGS convergence test, the job terminated with the following warning:

RuntimeWarning: invalid value encountered in sqrt a[:] = np.sqrt(1.0 / (2.0 * w))

I suspect that minimizer.run() generated imaginary phonon frequencies, which caused the calculation to fail.

To investigate this, I debugged the script and confirmed that imaginary frequencies were indeed present after minimizer.run() by adding
w_check, _ = minimizer.dyn.DiagonalizeSupercell() if np.any(w_check <= 0): print(f"[Warning] {nconf} configs has negative frequencies: {w_check[w_check<=0]}") # Save the refined auxiliary dynamical matrix refined_prefix = os.path.join(OUTPUT_DIR, f"refined_dyn_{nconf}_") minimizer.dyn.save_qe(refined_prefix)

I tried to explicitly force the use of minimizer.dyn.ForcePositiveDefinite() and minimizer.dyn.Symmetrize() after the minimizer.run() and run the script again, but nothing changed.

Why is this happening? I expected ForcePositiveDefinite() to eliminate the imaginary modes. Is there something I might be missing, or is there a recommended way to resolve this issue? Any suggestions would be greatly appreciated.

I have uploaded the script 6_hessian_converg_nconf.py for your reference.

I look forward to any comments or suggestions. Thank you in advance for your help again.

Jianguo Si
DIMST, CAS

[mesonepigreco]

Hi, it seems very strange.
I suspect the error is not due to imaginary frequencies, in this case it should be written very clearly in the output. The check you are doing on imaginary frequency can just spot a very slightly imaginary part on acoustic modes, that is completely normal and not an issue (even after imposing the ASR). Check the frequencies of the final dynamical matrix inspecting them: where are these imaginary modes? If the code crashes, there should be an error (not just a warning). What is the error? If it crashes without errors, the most likely cause is that you just run out of memory for computing the hessian. Does the terminal says that the process has been killed?

[Sssss-jg]

Thanks for your reply again.

I understand that the memory requirement correlates with the number of atoms and the q-point grid density.

Now, I have tested a case with a reduced q-mesh; however, no significant improvement was observed. For your reference, I have included the full running command, 6_hessian_converg_nconf.py, and log output below:
running command： python List/6_hessian_converg_nconf.py
log:

/share/home/jgsi/anaconda3/envs/sscha/lib/python3.10/site-packages/sscha/aiida_ensemble.py:18: UserWarning: aiida or aiida-quantumespresso are not installed
  warnings.warn('aiida or aiida-quantumespresso are not installed')

Running calculation with 20 configurations
Force computed shape: 20
Computing configuration 1 out of 20 (nat = 72)
/share/home/jgsi/anaconda3/envs/sscha/lib/python3.10/site-packages/cellconstructor/calculators.py:90: FutureWarning: Please use atoms.calc = calc
  atm.set_calculator(calculator)
Computing configuration 2 out of 20 (nat = 72)
Computing configuration 3 out of 20 (nat = 72)
Computing configuration 4 out of 20 (nat = 72)
Computing configuration 5 out of 20 (nat = 72)
Computing configuration 6 out of 20 (nat = 72)
Computing configuration 7 out of 20 (nat = 72)
Computing configuration 8 out of 20 (nat = 72)
Computing configuration 9 out of 20 (nat = 72)
Computing configuration 10 out of 20 (nat = 72)
Computing configuration 11 out of 20 (nat = 72)
Computing configuration 12 out of 20 (nat = 72)
Computing configuration 13 out of 20 (nat = 72)
Computing configuration 14 out of 20 (nat = 72)
Computing configuration 15 out of 20 (nat = 72)
Computing configuration 16 out of 20 (nat = 72)
Computing configuration 17 out of 20 (nat = 72)
Computing configuration 18 out of 20 (nat = 72)
Computing configuration 19 out of 20 (nat = 72)
Computing configuration 20 out of 20 (nat = 72)
Force computed shape: 20
Traceback (most recent call last):
  File "/share/home/jgsi/quip/workspace/Test_5_USED/SSCHA_222/List/6_hessian_converg_nconf.py", line 89, in <module>
    minimizer.init()
  File "/share/home/jgsi/anaconda3/envs/sscha/lib/python3.10/site-packages/sscha/SchaMinimizer.py", line 1108, in init
    qe_sym.SymmetrizeFCQ(fcq, self.dyn.q_stars, asr = "custom")
  File "/share/home/jgsi/anaconda3/envs/sscha/lib/python3.10/site-packages/cellconstructor/symmetries.py", line 1032, in SymmetrizeFCQ
    self.SymmetrizeDynQ(fcq[iq, :,:], q_points[iq,:])
  File "/share/home/jgsi/anaconda3/envs/sscha/lib/python3.10/site-packages/cellconstructor/symmetries.py", line 1573, in SymmetrizeDynQ
    dyn_matrix[3 * na : 3* na + 3, 3*nb: 3 * nb + 3] = Methods.convert_matrix_cart_cryst(fc, self.structure.unit_cell, True)
numpy.exceptions.ComplexWarning: Casting complex values to real discards the imaginary part

Besides, the process execution time was significantly shorter than the time limit I configured.

Thanks again.

Jianguo Si

[Sssss-jg]

Apologies for the delayed response (I had subscribed to this thread, but unfortunately, I didn't receive any email notifications.).

[DjordjeDangic]

Hey it could be just numpy issue. Can you add this at the top of your script and try again:

import warnings

try:
    ComplexWarning = np.exceptions.ComplexWarning
except AttributeError:
    ComplexWarning = np.ComplexWarning

warnings.filterwarnings("ignore", category=ComplexWarning)

[Sssss-jg]

Thank you for your valuable reply.

I’ve already added this code snippet at the beginning of the script as a global variable, but it does NOT seem to be taking effect.

Best Regards,
Jianguo SI

[mesonepigreco]

Hi @Sssss-jg , are you sure that the @DjordjeDangic solution does not work, or that the error has not changed? The Error trace from your log says explicitly that the problem is the numpy ComplexCasting from your output:

numpy.exceptions.ComplexWarning: Casting complex values to real discards the imaginary part

Which the Dorde solution should fix (as it is telling the code to ignore that error: this is a NumPy warning that is treated in newer versions as an error, but it is not a bug):

import warnings

try:
    ComplexWarning = np.exceptions.ComplexWarning
    except AttributeError:
    ComplexWarning = np.ComplexWarning

warnings.filterwarnings("ignore", category=ComplexWarning)

So it is possible that there is another error that is now blocking your code, but we need to see the log to understand

[Sssss-jg]

Apologies for the delayed response.

Thank you all for the helpful responses.

After extensive testing, I found that this was a crash caused by an OOM error!!

Jianguo SI
DIMST, CAS