Spin Polarization and Magnetic Systems
This week we'll cover the topic of spin-polarized systems. For metals, there are a couple of complications which mean we have to treat them differently from systems with a non-zero band gap.
Spin Polarization
Up untill now we have been assuming that we always had some set of bands which could each fit two electrons. Essentially we have been ignoring the electron spin. If you want to examine, for example, a magnetic system then the spin of the electrons is important. It can also be important in modelling atomic or molecular systems. We'll cover different examples of this in this lab.
The Oxygen Molecule
If a system is not necessarily magnetic we might imagine that representing it with some set of fully occupied, doubly degenerate bands will work. However, in some cases including spin polarization can lead to important differences. One example of this is the O2 molecule.
In this case, we have a system with two interacting oxygen atoms. Each oxygen
has 8 electrons in total, with the configuration 1s2 2s2 2p4
(the 1s orbital
will be contained within the pseudopotential for the DFT calculations done
here, so you will have 6 electrons from each oxygen atom).
For a single oxygen, from Hund's rule the three p orbitals should be filled singly before being filled in pairs, so that one of the p-orbitals will have two electrons, and the other two should have one each. However, if we assume doubly occupied orbitals, we'll have the two p-orbitals with two electrons and one that is empty. This means a calculation where we assume a set of doubly occupied bands will have trouble converging to the ground state of the system. For the molecule the situation is similar, but the s and p orbitals from each atom combine to form bonding and anti-bonding \(\sigma\) and \(\pi\) orbitals.
The directory 02_O2
contains an input file to calculate the total energy of
the system at the measured bond length. Here the calculation has been set up
exactly as you've seen in the past (i.e., assuming doubly degenerate band
occupation without smearing or spin polarization:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 |
|
Task 2.1 - Assming Spin Degenerate Insulator
Try running the calculation in this directory. Does it converge?
Answer
While it's possible that the system may randomly meet the convergence criteria in the self-consistent cycle, this calculation will most likely not converge. If you look at the estimate accuracy at the end of each iteration in the output, it will likely vary from step to step, rather than steadily decreasing as in a well-behaved calculation.
The situation we have is similar to a metal: we have two bands and the ground state of the system should be when there is one electron in each of them.
To get around this, we can use a metallic occupation scheme with a small smearing width. This will allow the system to converge to the correct ground state. The relevant input variables are the ones highlighed below:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 |
|
Task 2.2 - Assuming Spin Degenerate Metal
Create a copy of the 01_O2
directory called 01_O2_metal
. Modify the
input file in it to use a metallic occupation scheme with a small smearing
width and run the calculation (as above).
- Does the calculation now converge?
Answer
Yes, the calculation should now converge.
- Take a look at the file
pwscf.xml
in the calculation directory, and try to find the occupations of each band at each k-point. Are these as expected?
Answer
The occupations should be fractional for the highest occupied valence band which is not physical for a molecule.
1 2 3 4 5 |
|
While treating this system as a metal may help converging the calculation, it
may not necessarily reach the ground state since the spin-degress of freedom is
constrained. Instead, we can do a spin polarized calculation by adding nspin
and tot_magnetization
variables to the input file (highlighted below):
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 |
|
nspin
: this is 1 by default so no spin polarization is taken into account. To perform a spin polarized calculation it should be set to 2.tot_magnetization
: this is difference between the number of spin-up and spin-down electrons in the cell. If we want a single spin up electron we can set this to1.0
.
Task 2.3 - Assuming Spin Polarized Metal
Create another copy of 02_O2
called 02_O2_spin
. Then, try to:
-
Only turn on spin polarization. Does the calculation run?
Answer
The calculation will not run.
1 2 3 4
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Error in routine iosys (1): fixed occupations and lsda need tot_magnetization %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
Setting the total magnetization to 0, which would be the case if we don't have any net magnetization in the molecule, as both spins point in opposite directions.
Answer
The calculation converges to an energy of -63.25520699 Ry.
-
Setting the total magnetization to 2.0, which corresponds to both spins pointing in the same direction. Is the energy lower? How do the orbital energies vary?
Answer
The calculation converges to an energy of -63.29338911 Ry. The energy becomes lower with this configuration and the orbital energies become different between spin channels.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
<ks_energies> <k_point weight="1.00000000000000">0.000000000000000E+000 0.000000000000000E+000 0.000000000000000E+000</k_point> <npw>26462</npw> <eigenvalues size="16"> -1.025922874232402E+000 -7.812538236389854E-001 -4.590520636167327E-001 -4.198711506107195E-001 -3.871387741292614E-001 -2.912685945326537E-001 -2.532555221576769E-001 -1.076727247566867E-001 -1.025922792354764E+000 -7.812537346931661E-001 -4.590520549643138E-001 -4.198710925526331E-001 -3.871386455493686E-001 -2.912685126157339E-001 -2.532553593136832E-001 -1.076726566570645E-001 </eigenvalues> <occupations size="16"> 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 0.000000000000000E+000 0.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 1.000000000000000E+000 0.000000000000000E+000 0.000000000000000E+000 </occupations> </ks_energies>
Finally, comparing the energy of the spin polarized calculation with the spin degnerate metal calculation, we can see that the spin polarized calculation gives a lower energy.
Fun facts
O2 in its singlet state can be dangerous (see e.g.
this paper
),
so treating the spin correctly is important!
Iron
Now that you've seen how including spin polarization can allow us a correctly describe the ground state of our system in your calculation, the next step is to use it to describe a magnetic system.
In a magnetic system there is a net spin polarization in the unit cell. This means that we'll probably have an odd number of electrons, and the energy of the system when we include a net spin polarization is lower than the energy when we don't.
One of the most common magnetic systems is iron, so we'll examine this. The
directory 03_Fe
contains an input file for iron. Note this is a BCC structure
(as set by ibrav = 3
in the input file), whereas most of the crystals
structures you have examined previously were FCC. The calculation has been set
up in the usual way for a metallic system.
Task 3.1 - fixed magnetization
-
Run this calculation and check everything worked as expected. What is the final energy?
Answer
The final energy should be -55.52528610 Ry.
-
Now make a copy of the calculation directory and in this, modify the calculation to turn on spin polarization. Try running the calculation with
tot_magnetization = 0.0
first, and compare your total energy to that obtained using doubly degenerate bands.Note
while in the case of the O2 above, we were able to get our calculations to at least converge by using a metallic occupation instead of using spin polarization, in the case of iron, it will still be a metal when you use spin polarization, so you should not remove the input variables associated with this.
Answer
The total energy becomes -55.52528589 Ry. Almost identical to the one obtained with the doubly degenerate bands. This is because these two calculations are essentially identical.
-
Now try setting the total magnetization to 1.0 and see how total energy changes: Which is the more energetically favourable configuration?
Answer
The total energy becomes -55.53839616 Ry. Lower than the spin degenerate case.
-
Try setting the total magnetization to 2.0. How does the final energy compare to the previous value?
Answer
The total energy becomes -55.56226730 Ry. Lower than all previous cases.
From this we could test many guesses for the total magnetization, and find
the value which gives the lowest overall total energy. However, we can instead
pass an option that tells quantum espresso to automatically find the best
value. This is done by setting the starting_magnetization
input variable.
Task 3.2 - Relaxed magnetization
-
Make another copy of the
02_Fe
directory, and this time setnspin = 2
, andstarting_magnetization = 1.0
(do not include thetot_magnetization
variable as this fixes a value). Run the calculation and see what the final total magnetization per cell is. See if you can find a measured value for iron to compare to.Answer
The total magnetization becomes larger than 2.0.
This is becuase we are allowing the spin to fully relax in the system.1
total magnetization = 2.21 Bohr mag/cell
-
See if you can use what we covered in previous labs to calculate and make a plot of the electronic band structure of BCC Fe.
- Plot the spin-up and spin-down bands in different colours.
- Indicate the Fermi energy on your plot in some sensible way.
- As the Brillouin zone is different to the ones you have calculated so far you'll need to select a few sensible high-symmetry points yourself to plot with .
Answer
You can find the relevant input file in the directory
03_Fe/extra_bandstructure
. The band structure should look like The following:
Summary
In this lab you have seen:
- How to do a DFT calculation including spin polarization.
- How some systems need to be done with spin polarization to converge to the correct ground state.
- How to use spin polarized calculations to find the correct magnetization of a magnetic system by letting the code find the total magnetization that produces the lowest overall total energy.