ERROR in subspace rotation PDSYEVX/ PZHEEVX: I2,IFAIL= 1 117

[pablog._lustemberg1]

Hi!

I'm experiencing an error when running an Ab-initio MD (Langevin, MDALGO=2) at 600 K crashes early with the banner:
ERROR in subspace rotation PDSYEVX/ PZHEEVX: I2,IFAIL= 1 117
----> I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <----

This occurs during the electronic SCF of the first MD steps. The run is spin-polarized (ISPIN=2) without explicit MAGMOM in INCAR (so VASP uses default 1 μB). Dispersion is enabled (IVDW=11). The job uses Γ-only sampling (1×1×1). I’m trying to understand whether this is a diagonalization/subspace-rotation robustness issue under these MD settings, or a problem with my choices (ALGO=All, LREAL=Auto, NPAR/NCORE, etc.). After 216 iterations the temperature remains equal to 0....

Technical details

VASP version: vasp.6.5.1 10Mar25 (build Mar 26 2025)
Platform / toolchain: executed on LinuxIFC (as printed in OUTCAR)
Parallel setup (from OUTCAR):
running 224 mpi-ranks on 2 nodes
distrk: each k-point on 224 cores, 1 groups
distr: one band on NCORE=14 cores, 16 groups
K-points: Γ-only (ASE-generated KPOINTS: Monkhorst–Pack 1 1 1, shift 0 0 0)

Key INCAR tags (MD):
IBRION = 0
MDALGO = 2
TEBEG = 600
TEEND = 600
NSW = 1000
POTIM = 0.2
ISYM = 0
ENCUT = 525
SIGMA = 0.2
EDIFF = 1E-6
ALGO = All
IVDW = 11
ISPIN = 2
NELM = 500
NELMIN = 8
LCHARG = .FALSE.
LWAVE = .TRUE.
LREAL = Auto
ISIF = 0
ISMEAR = 0

NPAR = 16 ! 8 for 112, 16 for 224 cpu's
LORBIT = 11

Is this a known stability issue for ALGO=All during MD with dispersion (IVDW=11) and ISPIN=2?

Would you recommend switching to ALGO=Normal and/or LREAL=.FALSE. for MD to avoid subspace-rotation breakdowns?

Any guidance on parallel settings (e.g., avoiding legacy NPAR, choosing NCORE, Γ-only with many ranks) that could prevent this diagonalization failure?

Given the default MAGMOM warning, could the initial ferromagnetic guess be triggering instabilities in the subspace rotation for this system?

Thanks in advance for any pointers. Happy to test suggested flags and report back.

BR,

Pablo

[michael_wolloch]

Dear Pablo,

Thanks for the bug report.

You write that you want to use a Langevin thermostat, for your molecular dynamics run, but your MDALGO tag is 2, which selects a Nose-Hoover thermostat. The implementation of the Nose-Hoover that you selected with MDALGO=2 would also need an appropriate SMASS value, which you do not set. This explains why your temperature is zero, and could also explain the stability issue and the crash.
Please try again with MDALGO=3, the correct option for the Langevin thermostat, and select the appropriate LANGEVIN_GAMMA and, if you want to run an NpT and not an NVT ensemble, LANGEVIN_GAMMA_L.

Some additional thoughts:

• Not setting MAGMOM is usually not advisable. I would either investigate the magnetic ground state without MD beforehand and then initialize it accordingly for the MD, or run the MD with ISPIN=1. Especially since your system does not seem to contain ions that are usually magnetic.

• ALGO=All is very stable (especially with the improved line search, selected by setting ISEARCH=1, which you should do), but it is comparatively slow. I would recommend to try ALGO=Fast for the MD, with optimized parallelization settings.

• The ENCUT=525 choice seems a bit high for MD and your system (max ENMAX=400). Are you sure you need to increase the plane-wave cutoff so much?

Please report back once you have sorted out the thermostat issues,
all the best,
Michael

[pablog._lustemberg1]

Dear Michael,

Thank you so much, now it seems everything is working properly.

Thanks again for teaching me and the support.

Best,

Pablo