[QE-users] constrained magnetization with non-colin and spin-orbit

Mon Jan 18 11:42:01 CET 2021

Dear Matteo,
Non collinear calculations are usually very difficult to converge with QE (usually codes based on localized basis set are easier to converge since I guess there are less degrees of freedom..).
This is even more difficult in finite systems such as clusters or molecules where you can often have multiple magnetic states.
Did you try to perform non-collinear calculations without SOC? Just to check that collinear configurations converge and are independent of the magnetization angle?

Cyrille

========================
Cyrille Barreteau
CEA Saclay, IRAMIS, SPEC Bat. 771
91191 Gif sur Yvette Cedex, FRANCE
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+33 1 69 08 38 56 /+33  6 47 53 66 52  (mobile)
email:     cyrille.barreteau at cea.fr
Web:     http://iramis.cea.fr/Pisp/cyrille.barreteau/
========================
________________________________
De : users [users-bounces at lists.quantum-espresso.org] de la part de Matteo Cococcioni [matteo.cococcioni at unipv.it]
Envoyé : lundi 18 janvier 2021 11:05
À : Quantum ESPRESSO users Forum
Objet : Re: [QE-users] constrained magnetization with non-colin and spin-orbit

Dear Cyrille,

thanks for your reply and advice.

Il giorno lun 18 gen 2021 alle ore 10:25 BARRETEAU Cyrille <cyrille.barreteau at cea.fr<mailto:cyrille.barreteau at cea.fr>> ha scritto:
Dear Matteo

If I understand well you want to calculate the magnetic anisotropy of a magnetic molecule.

Yes, that's correct

I am not sure I have understood well the way you proceed when you say that you "start" from a collinear lsda.

Well, I tried to start a non collinear calculation of this system from scratch, but I could never achieve convergence. The convergence with lsda is much more robust and using tot_magnetization I could achieve both a ferromagnetic and an antiferromagnetic ground state. Then I tried to use the charge-density of these ground states to initialize the non-collinear calculation (of course after updating the Mo PP to its fully relativistic version) with spin-orbit. This way I manage to converge to the same AFM and FM ground state with the non-collinear calculation. Now I want to use these calculations as starting points to take the magnetic moments away from the z direction to which they are (anti)aligned.

I would say that you can try two strategies:

i) using the force theorem as implemented in QE (one scf lsda calculation and then nscf with SOC starting from different theta angles..)

ok, I hadn't thought of this. Maybe it's sufficient to estimate the MAE. If you initialize just the theta angle (angle1) will the modulus of the magnetic moment be preserved with respect to the lsda or collinear ground state it starts from? How does the code choose the angle phi (angle2) in this case? In presence of SOC this also might make a difference and contribute to the MAE (unless the system has a cylindrical symmetry around the easy axis, which is not my case).

ii) use the magnetic constraint with penalization parameter lambda and perform a scf calculation. It will be more delicate but maybe more precise (not sure since you might face convergence problems)

This is what I was trying to do (see above, also): I am trying to change angle1 with lambda. I wanted to do a scf calculation to leave the system free to chose angle2. But the convergence is very delicate. Actually there seems to be no finite contribution to the potential from the constraint, so the calculation goes on for a while without changing the angles from their starting value and then suddenly crashes (not sure why). It seems that the variable pointlist is never different from 0 in my case which causes the potential to be insensitive to lambda. But I haven't yet understood why.

thanks again.

Best regards,

Matteo

best
Cyrille

========================
Cyrille Barreteau
CEA Saclay, IRAMIS, SPEC Bat. 771
91191 Gif sur Yvette Cedex, FRANCE
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+33 1 69 08 38 56 /+33  6 47 53 66 52  (mobile)
email:     cyrille.barreteau at cea.fr<mailto:cyrille.barreteau at cea.fr>
Web:     http://iramis.cea.fr/Pisp/cyrille.barreteau/
========================
________________________________
De : users [users-bounces at lists.quantum-espresso.org<mailto:users-bounces at lists.quantum-espresso.org>] de la part de Matteo Cococcioni [matteo.cococcioni at unipv.it<mailto:matteo.cococcioni at unipv.it>]
Envoyé : dimanche 17 janvier 2021 11:09
À : Quantum ESPRESSO users Forum
Objet : [QE-users] constrained magnetization with non-colin and spin-orbit

Dear all,

I am trying to run some calculations on a molecule with two magnetic ions (Mo) and I want to use the non collinear spin with spin-orbit to calculate the energy needed to change their direction. Starting from a collinear (lsda) calculation I managed to converge the non-collinear one in a ferromagnetic configuration, with magnetic moments aligned along z. Now I am using this ground state as starting point for a calculation where one or both spins are somehow deviated from the z direction. Following the instructions in INPUT_PW I am using the following settings (in &system):

    noncolin = .true.
    lspinorb=.true.
    angle1(1) = 30.0
    angle1(2) = 30.0
    constrained_magnetization = 'atomic direction'
    lambda = 1.0

where species 1 and 2 correspond to the two Mo, angle1 is the angle I want to have between z and the final magnetization, lambda is the strength of the quadratic constraint.

If I start from the potential of the ground state with the magnetization along z, the code starts with no problem and even pretends to converge for a number of iterations. Then suddenly the energy explodes and the code crashes saying that there are too many not converged eigenvalues.
While it seems converging nothing relevant seems to happen to angles: the constraint energy remains almost the same and the magnetization of both atoms maintains its original direction (aside small fluctuations). The same behavior is observed independently from the value of lambda (which is very strange) and beta (the mixing parameter).

looking into the code (add_bfield.f90) the implementation of the constraint seems fine, as far as I can tell (except that the code tries to constraint all magnetic moments once a lambda is present, which I fixed), and the potential seems to get a term from the constraint on magnetization. So I would expect it to do something.

Does anyone have any experience with this type of calculations? What am I missing or not doing right? Thanks in advance for any help/advice.

Best regards,

Matteo

--
Matteo Cococcioni
Department of Physics
University of Pavia
Via Bassi 6, I-27100 Pavia, Italy
tel +39-0382-987485
e-mail matteo.cococcioni at unipv.it<mailto:lucio.andreani at unipv.it>
_______________________________________________
Quantum ESPRESSO is supported by MaX (www.max-centre.eu<http://www.max-centre.eu>)
users mailing list users at lists.quantum-espresso.org<mailto:users at lists.quantum-espresso.org>
https://lists.quantum-espresso.org/mailman/listinfo/users

--
Matteo Cococcioni
Department of Physics
University of Pavia
Via Bassi 6, I-27100 Pavia, Italy
tel +39-0382-987485
e-mail matteo.cococcioni at unipv.it<mailto:lucio.andreani at unipv.it>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20210118/81a62937/attachment.html>