Hello,
You have supplied some input but no output where I could see the error appear. It for instance is important to see if it appears in the first electronic minimization or later on. Based on the input I can only speculate and perhaps this could be a starting point for you to investigate the issue.
The geometry of your system seems far from equilibrium. Your H atom (atom 39) is placed exactly above O (atom 37) — identical fractional x,y coordinates, offset only in z by exactly 0.05. This gives a linear H–O–O angle of 180°, which is unphysical for HO2 (the equilibrium H–O–O angle is 104°). Additionally, the O–O distance in your structure is 1.46 Å, which is significantly stretched compared to the gas-phase HO2 value of 1.33 Å. A poor initial geometry for the adsorbate can cause large forces and erratic SCF behavior during relaxation, eventually leading to atoms approaching each other too closely and triggering the ZHEGV error. Especially since EDIFF = 1E-5 is not a very tight criterion. I recommend starting from a properly optimized gas-phase HO2 geometry (O–O ≈ 1.33 Å, O–H ≈ 0.97 Å, H–O–O angle ≈ 104°) placed on the surface at a reasonable adsorption distance.
You have ISTART = 1, which tells VASP to read wave functions from a WAVECAR file. Did you try starting from scratch or did you encounter the problem restarting from an existing WAVECAR?
For magnetic systems we recommend using ALGO=Conjugate ISEARCH=1. Maybe you can try that in addition to ALGO=Normal.
You perhaps need to revisit the choice of your pseudopotential. For O you may need a hard one and for Ni you may need the sv version to get accurate results.
Generally I'd recommend splitting the tasks. E.g. first relax the molecule, then the surface, then place the molecule near the surface and scan for the right adsorbtion site and for the distance using multiple static calculations. Then place the molecule at the adsorbtion site at the appropriate distance and let the surface and the molecule relax alternating between one and the other. It is sometimes worth a try to do all at once, but if you encounter problems splitting in smaller tasks with fewer degrees of freedom will be a safer strategy.
Please let me know how it goes and feel free to ask a follow up question (preferably with exampel stdout, OSZICAR and OUTCAR as well as submission script).
Best regards,
Marie-Therese