Sunday, May 15, 2022

Flat bands: Why you might care, and one way to get them

When physicists talk about the electronic properties of solids, we often talk about "band theory".  I've written a bit about this before here.  In classical mechanics, a free particle of mass \(m\) and momentum \(\mathbf{p}\) has a kinetic energy given by \(p^2/2m\).  In a crystalline solid, we can define a parameter, the crystal momentum, \(\hbar \mathbf{k}\), that acts a lot like momentum (accounting for the ability to transfer momentum to and from the whole lattice).  The energy near the top or bottom of a band is often described by an effective mass \(m_{*}\), so that \(E(\mathbf{k}) = E_{0} + (\hbar^2 k^2/2m_{*})\).  The whole energy band spans some range of energies called the bandwidth, \(\Delta\). If a band is "flat", that means that its energy is independent of \(\mathbf{k}\) and \(\Delta = 0\).  In the language above, that would imply an infinite effective mass; in a semiclassical picture, that implies zero velocity - the electrons are "localized", stuck around particular spatial locations.  

Why is this an interesting situation?  Well, the typical band picture basically ignores electron-electron interactions - the assumption is that the interaction energy scale is small compared to \(\Delta\).  If there is a flat band, then interactions can become the dominant physics, leading potentially to all kinds of interesting physics, like magnetism, superconductivity, etc.  There has been enormous excitement in the last few years about this because twisting adjacent layers of atomically thin materials like graphene by the right amount can lead to flat bands and does go along with a ton of cool phenomena.  

How else can you get a flat band?  Quantum interference is one way.  When worrying about quantum interference in electron motion, you have to add the complex amplitudes for different electronic trajectories.  This is what gives you the interference pattern in the two-slit experiment.   When trajectories to a certain position interfere destructively, the electron can't end up there.  

It turns out that destructive interference can come about from lattice symmetry. Shown in the figure is a panel adapted from this paper, a snapshot of part of a 2D kagome lattice.  For the labeled hexagon of atoms there, you can think of that rather like the carbon atoms in benzene, and it turns out that there are states such that the electrons tend to be localized to that hexagon.  Within a Wannier framework, the amplitudes for an electron to hop from the + and - labeled sites to the nearest (red) site are equal in magnitude but opposite in sign.  So, hopping out of the hexagon does not happen, due to destructive interference of the two trajectories (one from the + site, and one from the - site).  

Of course, if the flat band is empty, or if the flat band is buried deep down among the completely occupied electronic states, that's not likely to have readily observable consequences.  The situation is much more interesting if the flat band is near the Fermi level, the border between filled and empty electronic states.  Happily, this does seem to happen - one example is Ni3In, as discussed here showing "strange metal" response; another example is the (semiconducting?) system Nb3Cl8, described here.  These flat bands are one reason why there is a lot of interest these days in "kagome metals".

8 comments:

Anonymous said...

It sounds like anytime the electronic correlation is strong, then you should get for localized bands? In that case why are things like cuprate superconductors with dispersing bands considered to be more strongly correlated than twisted bilayer graphene?

Douglas Natelson said...

Anon, I’m not sure that I fully understand your question. For the cuprates, in the undoped limit, the interactions are sufficiently strong to localize the carriers on the Cu atoms even though the bandwidth isn’t infinitesimal, so that the parent compounds are Mott insulators. In the TBLG case, it also looks like there are correlated insulating states. I believe that one way to quantify correlation is to compare the integrated Drude spectral weight to what one would expect in the zero interaction case. Others may have better ideas of how to compare correlation strengths.

Anonymous said...

Correlations do not result in flat bands, instead flat bands may result in a situation where electronic interactions dominate the energy scales.

Douglas Natelson said...

Anon@7:46, I agree. That’s what I was trying to say.

Anonymous said...

If interactions localize carriers as in the Mott insulator, would you not expect a flat band in momentum space (localized in real space)? Why would correlations not result in flat bands?

Anonymous said...

I think it is a matter of ratios. The bandwidth/interaction in oxides is very small. For comparable effect in moire, bandwidth must become very small since interaction absolute value is smaller.

Just my opinion

Douglas Natelson said...

Anon@2:24, I need to be more careful with my words, and I think Anon@9:40 is basically correct. You can have an insulator without carriers being so localized that the bands lack dispersion; this is what happens in any band insulator. In the Mott case, U/t (on-site repulsion compared to the site-to-site hopping energy scale) is large; in the flat band case, t approaches zero.

Anonymous said...

Most Hubbard bands (i.e. the ones dictating the energy for electron addition or removal) are indeed not as flat as cartoons often show them.
Though they are away from the Fermi energy (or better, chemical potential).