My Community

I am running vasp version 5 on a machine of type cluster with operating system Centos 5.4, fortran compiler ifort and math libraries intel mkl.

The purpose of my calculations in to get the magnetic property of Fe doped Zn. I generate the structure of FeZn35 by using the lattice parameter of pure Zn and making a supercell of 2x2x2, then substitutes a Zn atom with Fe atom. I use the exchange correlation of PBE-GGA and k-mesh 8x8x8.

After the scf calculation, it comes to that the system is nonmagnetic with 0.001 bohr magnetic moment on Fe atom.

However, I did a calculation using another DFT software Wien2K with the same lattice parameter, the same exchange correlation, except k-mesh. In Wien2K, the k-mesh is 11x11x11. But the result of Wien2K shows the system is magnetism and with 2.54149 bohr magnetic moment on Fe atom.

Although in the calculations I did before, the results of the two software have something different within 0.1 bohr magnetic moment. The huge difference of the above result puzzled me.

By the way, I find a paper published in Phys. Rev. B 57, 7004 (1998). The author did a calculation using RS-LMTO-ASA method of Fe impurity in Zn host and found a local moment of 2.6 muB for substitutional Fe in Zn.

Any suggestion will be greatly appreciated! Thanks in advanced.

How have you initialized the magnetic moments in the system? A reasonable staring point is to set the MAGMOM tag to 1.2 or 1.5 times the expected converged configuration. If the initial magnetic moments are too small the system may converge to another solution (a local minima) if it exists.

Also make sure that your electronic convergence was completed within NELM steps (default 60).

Cheers,
/Dan
<span class='smallblacktext'>[ Edited Fri May 21 2010, 12:31PM ]</span>

Thank you for reply!

I did not set the MAGMOM tag and the value is default since I didn't know the expected configuration. The system is nonmagnetic in expriment, but none of the calculations without relaxtion is nonmagnetic.
Actually, I intend to do a relaxtion using vasp. And I have no idea whether the result of vasp is correct or that of wien2k.
By the way, if I do a relaxtion using vasp, should I need to set the MAGMON?

Well, you claim to know at least two configurations, so see if you can converge to the same configuration as in Wien2K by taking that magnetic configuration as input to your VASP calculation (i.e. set MAGMOM accordingly and possible multiply with 1.2 or 1.5). Then compare if the nonmagnetic solution or the solution where Fe has an magnetic moment has the lowest energy.

Experiments are performed at a finite temperature, so it's possible that a stable Fe mag. moment at T = 0 K disappears due to thermal expansion or other thermal effects.

MAGMOM sets the initial guess for the magnetic moments. This tag can effect the electronic convergence in the first ionic step. If your calculation does not converge to the correct lowest mag. state it can therefore help to modify the MAGMOM tag from its default value. So yes, if you want to have a proper start configuration for the relaxation you will need to specify MAGMOM.

/Dan

Thank you for your suggestions!
I set the MAGMOM = 35*0 5 and then do a calculation without relaxation using vasp, the result still comes to nonmagnetic and the magnetic moment in Fe atom is 0.002.

Actually, I did a calculation without spin in wien2k, the nonmagnetic solution has the lower energy than magnetic solution.

Experiments shows the Fe impurity with PPM concentration is nonmagnetic, for the magnetic moment from the spin and from the orbital cancel out each other in the range of 20K to 300K.

ref. Phys. Rev. Lett. 61, 2129

Normally you would not expect to obtain the correct properties of these strongly correlated systems using DFT
since it is a broken-spin-symmetry and single-particle theory. In DFT the singlet (non magnetic) solution may be internally of different nature than the real (Kondo) singlet. I would like to see a comparison of the charge density given by the DFT singlet solution and that obtained with a multi-reference quantum chemical calculation.