Why is the calculated energy of hexagonal In2Se3 lattice not accurate ?

[Jeffreyfox]

I am using VASP to relax the hexagonal In2Se3 structure and calculate its energy. The lattice dimension is 4A＊4A＊28.8A, the number of In and Se atoms in a cell is 6 and 9. One cell contains 15 layers along c axis and each layer only contains one representing atom. The 15 layers is divided into 3 groups, each groups is stacked as SeInSeInSe, with a vacancy layer between each group.

INCAR is as below：
SYSTEM = In2Se3 Layered structure
ISPIN = 2
IBRION = 1
ISYM = 0
EDIFFG = -0.005
ENCUT = 160
POTIM = 0.1
NSW = 500
PREC = med
LREAL = Auto
ISIF = 3 #relax the lattice

KPOINTS is as below:
Automatic Mesh
0
Monkhorst-Pack
3 3 1
0. 0. 0.

By increasing KPOINTS and ENCUT and comparing the energy results, I found that the energy is fluctuating and cannot reach the desired accuracy. (the value is fluctuating by several tenths of eV). Can someone tell me what has caused this problem? Is it because of the parameters or the structure? Thanks!

[lcyin]

Why you use "ISPIN=2" for this calculation? Is this system a open-shell system? If this system is a polar one, maybe, you should take the Dipole correction into account.