Tuesday, July 31, 2018

What is Berry phase?

On the road to discussing the Modern Theory of Polarization (e.g., pdf), it's necessary to talk about Berry phase - here, unlike many uses of the word on this blog, "phase" actually refers to a phase angle, as in a complex number \(e^{i\phi}\).   The Berry phase, named for Michael Berry, is a so-called geometric phase, in that the value of the phase depends on the "space" itself and the trajectory the system takes.  (For reference, the original paper is here (pdf), a nice talk about this is here, and reviews on how this shows up in electronic properties are here and here.)

A similar-in-spirit angle shows up in the problem of "parallel transport" (jargony wiki) along curved surfaces.  Imagine taking a walk while holding an arrow, initially pointed east, say.  You walk, always keeping the arrow pointed in the local direction of east, in the closed path shown at right.  On a flat surface, when you get back to your starting point, the arrow is pointing in the same direction it did initially.   On a curved (say spherical) surface, though, something different has happened.  As shown, when you get back to your starting point, the arrow has rotated from its initial position, despite the fact that you always kept it pointed in the local east direction.  The angle of rotation is a geometric phase analogous to Berry phase.  The issue is that the local definition of "east" varies over the surface of the sphere.   In more mathematical language, the basis vectors (that point in the local cardinal directions) vary in space.  If you want to keep track of how the arrow vector changes along the path, you have to account for both the changing of the numerical components of the vector along each basis direction, and the change in the basis vectors themselves.  This kind of thing crops up in general relativity, where it is calculated using Christoffel symbols.

So what about the actual Berry phase?  To deal with this with a minimum of math, it's best to use some of the language that Feynman employed in his popular book QED.   The actual math is laid out here.  In Feynman's language, we can picture the quantum mechanical phase associated with some quantum state as the hand of a stopwatch, winding around.  For a state \(| \psi\rangle \) (an energy eigenstate, one of the "energy levels" of our system) with energy \(E\), we learn in quantum mechanics that the phase accumulates at a rate of \(E/\hbar\), so that the phase angle after some time \(t\) is given by \(\Delta \phi = Et/\hbar\).   Now suppose we were able to mess about with our system, so that energy levels varied as a function of some tuning parameter \(\lambda\).  For example, maybe we can dial around an externally applied electric field by applying a voltage to some capacitor plates.  If we do this slowly (adiabatically), then the system always stays in its instantaneous version of that state with instantaneous energy \(E(\lambda)\).  So, in the Feynman watch picture, sometimes the stopwatch is winding fast, sometimes it's winding slow, depending on the instantaneous value of \(E(\lambda)\).  You might think that the phase that would be racked up would just be found by adding up the little contributions, \(\Delta \phi = \int (E(\lambda(t))/\hbar) dt\).

However, this misses something!  In the parallel transport problem above, to get the right total answer about how the vector rotates globally we have to keep track of how the basis vectors vary along the path.  Here, it turns out that we have to keep track of how the state itself, \(| \psi \rangle\), varies locally with \(\lambda\).  To stretch the stopwatch analogy, imagine that the hand of the stopwatch can also gain or lose time along the way because the positioning of the numbers on the watch face (determined by \(| \psi \rangle \) ) is actually also varying along the path.

[Mathematically, that second contribution to the phase adds up to be \( \int \langle \psi(\lambda)| \partial_{\lambda}| \psi(\lambda) \rangle d \lambda\).  Generally \(\lambda\) could be a vectorial thing with multiple components, so that \(\partial_{\lambda}\) would be a gradient operator with respect to \(\lambda\), and the integral would be a line integral along some trajectory of \(\lambda\).  It turns out that if you want to, you can define the integrand to be an effective vector potential called the Berry connection.  The curl of that vector potential is some effective magnetic field, called the Berry curvature.  Then the line integral above, if it's around some closed path in \(\lambda\), is equal to the flux of that effective magnetic field through the closed path, and the accumulated Berry phase around that closed path is then analogous to the Aharonov-Bohm phase.]

Why is any of this of interest in condensed matter?

Well, one approach to worrying about the electronic properties of conducting (crystalline) materials is to think about starting off some electronic wavepacket, initially centered around some particular Bloch state at an initial (crystal) momentum \(\mathbf{p} = \hbar \mathbf{k}\).  Then we let that wavepacket propagate around, following the rules of "semiclassical dynamics" - the idea that there is some Fermi velocity \(\partial E(\mathbf{k})/\partial \mathbf{k}\) (related to how the wavepacket racks up phase as it propagates in space), and we basically write down \(\mathbf{F} = m\mathbf{a}\) using electric and magnetic fields.  Here, there is the usual phase that adds up from the wavepacket propagating in space (the Fermi velocity piece), but there can be an additional Berry phase which here comes from how the Bloch states actually vary throughout \(\mathbf{k}\)-space.  That can be written in terms of an "anomalous velocity" (anomalous because it's not from the usual Fermi velocity picture), and can lead to things like the anomalous Hall effect and a bunch of other measurable consequences, including topological fun.


Pizza Perusing Physicist said...

Fantastic post! Thanks for taking the time to do such a thorough job on such a difficult topic. I don’t know how you find the time!

Anonymous said...

I'm a bit confused as I get the impression that even on the sphere the arrow actually does *not* rotate; it wobbles up, but then comes down again pointing in the same direction of "12 o'clock"?
As long as the north pole is not contained within the closed path, I do not think the arrow rotates. It changes direction, but does not rotate (a full circle)?
Can you explain ?

And yes, thanks for taking the time to write illuminating posts like this!

Douglas Natelson said...

Anon, I think this is an issue of my word choice. What I am trying to say is that the final orientation of the arrow (after completing the closed path on the surface of the sphere, where it has been "parallel transported") is not identical to its initial orientation. "Rotate" here is meant to imply reorientation by some angle \(\theta\), not a complete rotation (\(\theta = 2 \pi\)).

Anonymous said...

This is something new, https://www.v3.co.uk/v3-uk/news/3036984/quantum-material-developed-that-could-become-a-building-block-of-the-quantum-age

Don't know what do they mean when they say "it has more than one electron pattern".

Anonymous said...

I enjoyed your writing. You should probably mention that the idea of geometric phase predates Michael Berry by almost thirty years. It was described and worked out by S. Pancharatnam in the context of polarized light. See for example, the resource letter on geometric phases in the American Journal of Physics 65, 180 (1997).


Anonymous said...

news: https://electroiq.com/2018/08/the-rebirth-of-the-semiconductor-industry/

Douglas Natelson said...

Anon@3:34, that refers to this paper: https://arxiv.org/abs/1711.07390. Looks at a quick glance like that material has a very complicated band structure, with various Dirac cones and lines depending on the chemical potential. Very pretty photoemission data with an exotic Fermi surface, but I don’t know how big a deal this is.

MisterBee, you are correct, of course. Pancharatnam looked at the rotation of polarization as light propagates through an optical fiber - another case of parallel/adiabatic transport affecting a phase angle. I suspect the fact that Pancharatnam is already deceased is one reason why my perennial Nobel suggestion of Berry and Aharanov is doomed....

Anon@6:00, interesting. We shall see. DARPA is also putting a ton of money into post CMOS ideas, but the economics of Si are a powerful force. I’d greatly like to see the electronics industry genuinely get excited about other materials, rather than their reflexive reaction that everything is a threat to CMOS yield.