I am using VASP to perform a structural relaxation of $\ce{CuNCN}$. I am using the PAW-PBE POTCAR files for the same and the KPOINTS have been generated using a script. The POSCAR file has been generated from the CIF file obtained via PXRD measurements. Even after multiple iterations (using the CONTCAR as POSCAR in subsequent calculations), the CONTCAR data is nowhere close to my POSCAR data, i.e. the structure is relaxing to an incorrect energy minimum. Note that I am using the DFT-D3 dispersion scheme for the optimization. In the literature, dispersion has been employed for $\ce{PbNCN}$, which has a similar layered structure.
My CONTCAR structure (After 5 runs):
My INCAR:
ISTART = 0
ICHARG = 2
PREC = Accurate
ISIF = 3
IBRION = 2
EDIFF = 0.00000001
EDIFFG = -0.0005
NSW = 200
NELM = 100
ENCUT = 400
PSTRESS = 0
IVDW = 12
ISMEAR = 0
SIGMA = 0.1
ISYM = 1
LMAXMIX = 4