Antiferromagnetic (AFM) calculations (From the userguide)

Several considerations are necessary, when you want to perform an AFM calculation. Please have also a look into $WIENROOT/SRC_afminput/afminput_test.

You must construct a unit cell which allows for the desired AF ordering. For example for bcc Cr you must select a ``P'' lattice and specify both atoms, Cr1 at (0,0,0) and Cr2 at (.5,.5,.5), corresponding to a CsCl structure. Note, that it is important to label the two Cr atoms with ``Cr1'' and ``Cr2'', since only then the symmetry programs can detect that those atoms should be different (although they have the same Z). If sgroup has interchanged some axis, try to undo these changes, since afminput may not properly find the correct symmetry operations in such a case.

When you generate case.inst you must specify the correct magnetic order and flip the spin of the AF atoms (i.e. invert the spin up and dn occupation numbers). In addition you should set a zero moment (identical spin up and dn occupations) for all ``non-magnetic'' atoms. This can be done conveniently using instgen_lapw -ask or during ``initialization'' using w2web.

Now you can run either a ``normal'' spinpolarized initialization (without AFM option) and runsp_lapw or:

Create a struct file of the non-magnetic (or ferro-magnetic) supergroup (run init_lapw up to lstart). Name it case.struct_supergroup. (For example for bcc Cr, this would be a struct file with the ordinary cubic lattice parameters, ``B'' type lattice and just one Cr at (0,0,0).)

Run init_lapw. At the end AFMINPUT creates an input file for the program CLMCOPY. Depending on the presence of case.struct_supergroup and the specific symmetry it may/may not ask you to supply a symmetry operation/nonprimitive translation (see Sect. 9.5 .

Run runafm_lapw. This script calls LAPW1 and LAPW2 only for spin-up but the corresponding spin-dn density is created by CLMCOPY according to the rules defined during initialization. This reduces the required cpu time by a factor of 2 (and in addition the scf cycle is much more stable).

It is highly recommended that you save your work (save_lapw) and check the results by continuing with a regular runsp_lapw. If nothing changes (E-tot and other properties), then you are ok, otherwise make sure the scf calculation is well converged (-cc 0.0001 or better). Eventually the system may not want to be antiferromagnetic (but for instance it is ferrimagnetic!).

runafm_lapw saves you more than a factor of 2 in in computer time, since only spin-up is calculated and in addition the scf-convergence may be MUCH faster. It works also with LDA+U (case.dmatup/dn are also copied), but does NOT work with Hybrid-DFT nor spin-orbit coupling, since this requires the presence of both vector files in the LAPWSO step.