Problems with structure generation, heredity, and wrong k-grid generation

[Roman Lucrezi]

Hey everybody!

I am currently running USPEX for fixed composition searches in CsPH8 at 50GPa and I have encountered a couple of errors/warnings and a weird behaviour regarding the k-grid generation.

Any help or explanation would be much appreciated!
(please find my input files attached)

1) initial structure generation
By running USPEX for CsPH8 at 50GPa without any additional information about volumes or ion distances, the code struggles to create structures and gets stuck eventually (log messages like [1]). So, I've set the volume by hand (% Latticevalues) to a larger value than the default value at that pressure, and reduced the minimum ion distances (% IonDistances). With these settings, I could generate my initial population, but I'm still getting a few messages like [1] and it takes quite some time, as the structure generator has to retry several times to find a suitable structure (message [2]). Should I increase/reduce the mentioned values even further?

2) generate second population / heredity
While 1) might just take a bit more time, but eventually generates and calculates a population, the actual problem appears at the generation of the second population. After all the structures from population 1 have been relaxed, the code prints "status = Local optimization finished" and proceeds to the next population, where it tries to generate structures also by heredity. Now, I'm getting warnings like [3], and the "change to random" might work for a few new structures, but eventually the code gets stuck after a warning like [3]. Even restarting several times from pickUpGen 1, results qualitatively in the same outcome.
Is there any additional parameter to tune the heredity generation? Any other suggestions?

A sidenote to this problem: only 34/50 structures from this first population are recognised as finished/relaxed by USPEX, even though the QE calculation finished properly. Are those perhaps too few structures in order to successfully generate new structures by heredity?

And actually, these 16 unfinished structures also raise a third issue:

3) wrong k-grid generation
In my current setup, I use a 5 step relaxation procedure with increasing accuracy. The following happens for these 16 structures that USPEX does not recognise as "relaxed": Steps 1-4 proceed normally, converge and print final enthalpy and structure. At the 5th step, where the k-grid spacing is supposed to be the smallest, USPEX creates the correct input file with the relaxed and symmetrized structure from step 4, but the k-grid is just the Gamma point (1 1 1 0 0 0).
I managed to correct the erroneous input files by checking them via the submitJob_remote.py script before actually submitting the job, and now the 5th relaxation step runs and finishes smoothly for all of the 50 structures. Nevertheless, USPEX does not recognise those structures that were created with a 111-grid as relaxed; in fact, their entry in the log file is just gone after the 5th-step job finished [4].
Any idea what's happening here?

Kind regards,
Roman Lucrezi

-------------------------------------------

Software setup:
- USPEX v10.5.0
- python 3.8.20 on Debian 12
- remote job submission to a cluster
- Quantum Espresso as DFT code, 5 relaxation steps with increasing accuracy from USPEX
--> this setup works well for other systems and even for CsPH8 at 0GPa without any additional input on volumes or ion distances

[1] Random : Can not randomly generate a structure after 5 minutes. The minimum distance threshold will be lowered by 0.1!
Random : Please check your IonDistances parameter.

[2] "not possible to place given number of atoms into unit cell of given space-group symmetry" in CalcFoldTemp/rc.out

[3] Heredity : Failed to do Heredity in 50 attempts, switch to Random!

[4] while the 5th step is still running, a line like this is present in the log:
Structure27 step5 at CalcFold27
JobID=11227528