nonmagentic system gives different energy when doing spin polarized calculations.

[jhouze]

Using Vasp 4.6.31 calculating a seven layer thick 1x1 Tungsten(100) surface.

Setting:
ISPIN = 2
VOSKOWN = 1

I get -82.65 eV for total energy and mag= 0.0000
For non spin polarized calculations I get -85.51 eV.

This should be the same total energy, because the system is not magnetic.

To further complicated matters I had previously done the calculation with vasp 4.5.7 and with and without the spin flags I got -85.51 eV. Using Vasp 4.5.8 with and without the spin flags I got -82.65 eV
<span class='smallblacktext'>[ Edited ]</span>

[admin]

please check whether it helps to decrease the compilation optimization level of xcgrad and xcspin. (some compilers need lowest optimization for these subroutines).
A very small difference (a few micro-eV) may still persist, even though the system is not magnetic

[jhouze]

Thanks. It indeed was a result of compiler optimization.
Removing "-tp p6" from the pg makefile corrected the issue.
-O2 is okay.