"I_CONSTRAINED_M = 2" option?
Posted: Thu Jan 03, 2008 3:00 pm
A question regarding constraints on the directions for magnetic moments. As an example I have two atoms (in a bcc structure) under the contraint M_CONSTR = 0 0 1 0 0 -1. I wish to examine the antiferromagnetic-state (AFM) and carry out two calculations with different initial magnetic moments:
Case A: MAGMOM = 0 0 3 0 0 -1
Case B: MAGMOM = 0 0 3 0 0 -3
Case B relax to the antiferromagnetic solution while Case A relax to the ferromagnetic solution without any penalty energy. After looking at the source code I find this reasonable since only the perpendicular component enters into the penalty energy.
The issue is that I want to restrict the magnetic moment from changing sign along the z-axis, i.e. I want to take into account the parallell component, so that a vector pointing in the positive direction (with the 0 0 -1 constraint) will give a penalty energy.
In the source code there appears to be an "I_CONSTRAINED_M = 2" option where the norm of the difference between magnetic moment and the constraint vector enters the penalty term instead, which seems to actually solve the issue. Performing the case A again also give me the AFM-state, although the convergence is quite slow (I guess this will depend on lambda).
So finally the main questions:
In the cases where I can't for instance use NUPDOWN = 0 to set an AFM-state, like heterogenous interfaces where material A have a AFM-structure while the material B have a FM-structure, is there any possibility to restrict the AFM-phase by the use of the "I_CONSTRAINED_M = 2" option? Will this be reliable in terms of convergence?
Best regards,
/Dan
--- INCAR ---
...
LNONCOLLINEAR = .TRUE.
I_CONSTRAINED_M = 1
LAMBDA = 10
RWIGS = 1.302
M_CONSTR = 0 0 1 0 0 -1
%---------------------
<span class='smallblacktext'>[ Edited ]</span>
Case A: MAGMOM = 0 0 3 0 0 -1
Case B: MAGMOM = 0 0 3 0 0 -3
Case B relax to the antiferromagnetic solution while Case A relax to the ferromagnetic solution without any penalty energy. After looking at the source code I find this reasonable since only the perpendicular component enters into the penalty energy.
The issue is that I want to restrict the magnetic moment from changing sign along the z-axis, i.e. I want to take into account the parallell component, so that a vector pointing in the positive direction (with the 0 0 -1 constraint) will give a penalty energy.
In the source code there appears to be an "I_CONSTRAINED_M = 2" option where the norm of the difference between magnetic moment and the constraint vector enters the penalty term instead, which seems to actually solve the issue. Performing the case A again also give me the AFM-state, although the convergence is quite slow (I guess this will depend on lambda).
So finally the main questions:
In the cases where I can't for instance use NUPDOWN = 0 to set an AFM-state, like heterogenous interfaces where material A have a AFM-structure while the material B have a FM-structure, is there any possibility to restrict the AFM-phase by the use of the "I_CONSTRAINED_M = 2" option? Will this be reliable in terms of convergence?
Best regards,
/Dan
--- INCAR ---
...
LNONCOLLINEAR = .TRUE.
I_CONSTRAINED_M = 1
LAMBDA = 10
RWIGS = 1.302
M_CONSTR = 0 0 1 0 0 -1
%---------------------
<span class='smallblacktext'>[ Edited ]</span>