Skip to content

SCF convergence failure for NiBDC surface relaxation in LCAO/GPU mode, while the same structure converges in VASP #7320

@EnciHan

Description

@EnciHan

Describe the bug

Dear ABACUS developers,

I am trying to relax a NiBDC surface slab using ABACUS. The same initial structure can be successfully relaxed in VASP, but the ABACUS calculation fails to converge during the electronic SCF iterations at each ionic step.

The main issue is that the SCF residual keeps oscillating around ~1e-2 and does not decrease to the target threshold, even after testing several mixing, smearing, relaxation, dipole-correction, and DFT+U settings.

I would like to ask whether this behavior may be related to my input settings, the choice of pseudopotentials/numerical orbitals, the LCAO basis, magnetic initialization, or a possible SCF convergence issue in ABACUS for this type of Ni-based MOF surface system.

System

  • System: NiBDC surface slab
  • Elements: Ni, C, O, H
  • Initial magnetic moment: Ni atoms are initialized with mag = 2 in STRU
  • All atoms are allowed to relax

I have tested the following settings, but the SCF still fails to converge:

smearing_sigma: 0.005, 0.01, 0.015 Ry
mixing_beta: 0.05, 0.08, 0.10, 0.15
mixing_beta_mag: 0.05, 0.08, 0.10, 0.15
mixing_gg0: 1.0, 1.5
relax_method: cg and bfgs_trad
Dipole correction: tested both on and off
DFT+U: tested both on and off
Ni initial magnetic moment in STRU: mag = 2

In all cases, the electronic SCF residual oscillates around ~1e-2 at each ionic step and does not converge.

The same initial structure can be relaxed successfully using VASP. However, it takes a long time.

The VASP INCAR is:

SYSTEM = qvasp

PREC = Normal
ENCUT = 400 eV
NELMIN = 5
NELM = 300
LREAL = F

ALGO = Fast
EDIFF = 1E-5
ISMEAR = 0
SIGMA = 0.05

ISPIN = 2
MAGMOM = 960 660 720 242

NSIM = 64

ICHARG = 2
EDIFFG = -0.03
NSW = 300
IBRION = 2
POTIM = 0.5
ISIF = 2

LCHARG = T
LWAVE = T
ISYM = 0
IVDW = 11

The VASP calculation converges normally with the same initial structure.

Questions

Are there recommended SCF settings for Ni-based MOF/surface systems using ABACUS LCAO?
Could the convergence failure be related to the numerical atomic orbitals or pseudopotentials for Ni/O?
Is there any recommended strategy for magnetic transition-metal systems with slab geometries?
Since the SCF residual remains around ~1e-2, is this more likely caused by charge mixing, magnetic-density mixing, the initial magnetic configuration, or the LCAO basis?

Expected behavior

I expected ABACUS to complete the electronic SCF convergence at each ionic step and then proceed with the optimization until the force convergence criterion is reached.

To Reproduce

Use ABACUS version:

  • ABACUS 3.10.0 LTS

Run the calculation on one NVIDIA Tesla V100 32GB GPU.

Environment

No response

Additional Context

INPUT.txt
STRU.txt

Task list for Issue attackers (only for developers)

  • Verify the issue is not a duplicate.
  • Describe the bug.
  • Steps to reproduce.
  • Expected behavior.
  • Error message.
  • Environment details.
  • Additional context.
  • Assign a priority level (low, medium, high, urgent).
  • Assign the issue to a team member.
  • Label the issue with relevant tags.
  • Identify possible related issues.
  • Create a unit test or automated test to reproduce the bug (if applicable).
  • Fix the bug.
  • Test the fix.
  • Update documentation (if necessary).
  • Close the issue and inform the reporter (if applicable).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions