Dear Sourav,
Thanks for the detailed description.
- In the Wannier90 input, use Cr instead of Cr1. For both spin
up and spin down, you'll need not only the occupied orbitals but
also the unoccupied ones. As in TB2J, the local spin field, which
is the difference between Hamiltonian for spin up and spin down
for each magnetic site are needed.
- In the --element option, use Cr instead of Cr1 Cr2.
Best wishes,
HeXu
--
You received this message because you are subscribed to the Google Groups "TB2J" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tb2j+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tb2j/bf17fe26-34a8-47bc-8ebf-29c444fbd49bn%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
also some clarifications needed. whether am i using spin = up for up spin wannier calculation as before in CrN_up.win file
and nscf calculation with nosym =.true. ?
To view this discussion on the web visit https://groups.google.com/d/msgid/tb2j/2b4cf70c-cc3c-4efe-ad56-695efaa3dc7dn%40googlegroups.com.
Hi,
- For the TB2J results, it looks quite nice although I don't know
the details. Looks like the ligand magnetization effect is not
quite strong, which is quite normal.
- For the Multibinit calculation: To make is less noisy, making
the supercell larger, and use a larger spin_ntime will be useful.
Adding a tiny SIA (uncomment the SIA lines) might also help. One
thing to note is that the Cv at 0K is meaningless, so it's better
to remove that point from the figure.
- The agreement with experiment seems quite good. It is usually
not expected that the result should completely agree, as the model
is based on approximations and both computation and experiments
are with errors.
Best wishes,
HeXu
To view this discussion on the web visit https://groups.google.com/d/msgid/tb2j/1a80d6af-8b97-454e-aab5-f245ae45d20dn%40googlegroups.com.
I need some more help also. It takes a lot of time to run spin dynamics for this calculation it has taken 30 hrs. so increasing spin_ntime will take a lot of time to execute. will spin_ntime 300000 and spin_ntime_pre 50000 better?
There are several options:
- Use mpi. Currently the scaling of cpu numbers is not quite good in the spin dynamics, but it could help.
- Split the calculation into different temperature ranges and run
them in parallel.
- Tune the time step to reduce the number of total time steps
required. You'll need to try with different values to gain some
intuition on what number is the best for the system. Perhaps use a
small supercell size. But be careful not to use a too large time
step which will make the dynamics diverge. There is always a
balance between efficiency and stability.
I have also taken spin_dynamics = 3 (Monte Carlo) on 28 28 28 ncell and 500000 spin_ntime but cv and x came to be negative for MC case.
This is weird. Could you send me the input files including the
TB2J results and the Multibinit input/output? Looks like a bug in
Multibinit for MC. In principle these numbers cannot be negative
from the equation.
In the time of J calculation I have not taken any SIA. So in Multibinit ,can i add small SIA? if i can then sia_k1amp =1e-06 and sia_kidir = 0.0 0.0 1.0 hold good?
Also i have found that in tutorial spin_ntime = 10* spin_ntime_pre. is it mandatory in all case?
My magnon dispersion has come with negative magnon energy. is this due to Pseudo potentialeffect? by changing Pseudo potential can i get positive magnon energy? As negative energy is not physically
Note that the current implementation of magnon band structure is
only for FERROMAGNETIC materials. The negative energy means the
ground state is not ferromagnetic. As from your result indeed it
is antiferromagnetic.
Best regards,
HeXu
To view this discussion on the web visit https://groups.google.com/d/msgid/tb2j/84c6571b-7062-438d-ae01-5c611b46caa7n%40googlegroups.com.
Hi,
-I guess the compuation time could be reduced if you use a smaller
number of CPUs. The parallelization in the spin dynamics is not
yet optimized and the scaling is quite bad...
From my experience, it is only helpful with a few CPUs, and it
actually slows down the computation when the number is further
increased.
- Indeed with the Monte carlo method, these quantities are broken. I have found where the bug is and it will be fixed in the next release of Multibinit.Thanks.
- Yes, I mean spin_dt. If you want to make the simulation faster,
try with tuning this parameter to get a balance between stability
and efficiency. Small spin_dt will be more stable but requires
more steps, whereas larger with large spin_dt there is the risk of
divergence. So this needs some careful test.
Best wishes,
HeXu
To view this discussion on the web visit https://groups.google.com/d/msgid/tb2j/8ff66138-afe4-49ea-9a9d-5fb686d4a245n%40googlegroups.com.