Tuesday, April 10, 2007

The curious rotational memory of the Electron, Part 2

A spinning top keeps going because it has angular momentum, and the angular momentum arises because the top is made up of mass that is distributed around the axis of rotation. The angular momentum is represented by a vector (well, a pseudovector really) directed along the axis of spin (at least when it's spinning around the top's axis). Similarly electrons have angular momentum. It's tempting to think of electrons as like microscopic spinning tops but the angular momentum of electrons isn't due to an extended mass rotating about some axis. In fact, as far as we know, electrons have no extension. Instead their angular momentum is a peculiar kind of intrinsic angular momentum that has a lot in common with ordinary angular momentum - eg. it's conserved and it's interconvertible with ordinary angular momentum. Because of it's not being associated with a mass distribution it's given the special name of spin, but as long as you bear in mind these provisos it's probably fair to use at least some intuitions about spinning tops when thinking about electrons.

But it's a little more complex than this. We also have quantum mechanics to contend with. The spin of an electron is a vector. But we find that when we measure one of the components of this vector this value is quantised and can only take values +hbar/2 and -hbar/2, where hbar is Planck's constant. We choose units where h-bar is 1 so the z-component of the spin is always measured to be +1/2 or -1/2. If we write these two states as |+> and |-> then because we are dealing with quantum mechanics, the z-component of the spin can be represented by the linear combination a|+>+b|->. This corresponds to a state in which there is a probability |a|² of measuring +1/2 and a probability |b|² of measuring -1/2. This is what might have been written as a.*return (1/2)+b.*return (-1/2) in my earlier Haskell code. But that's just one component of the spin. What about the x- and y-components? Amazingly the state a|+>+b|-> tells us everything we can possibly know about the spin of an electron and we'll call it a spin state.

Suppose we have an electron in the state ψ = a|+>+b|->. What happens if we measure the y-component of its spin? One way to answer that question is to rotate the electron through π/2 so that its x-axis is rotated to align with the z-axis and then measure the z-component of its spin. In order to to that we need to know how to rotate spin states. The rule for rotation through θ about the x-axis is this (in a suitable coordinate frame):

|+> → cos(θ/2)|+>-sin(θ/2)|->
|-> → sin(θ/2)|+>+cos(θ/2)|->

Note how choosing θ=0 gives the identity, as expected. Note also that θ=π maps a|+>+b|-> to b|+>-a|-> so that the probabilities of measuring +1/2 and -1/2 are simply swapped, exactly what you'd expect for turning a state upside down. But there's something else that you should notice - there's an ambiguity. A rotation through 2π should give the same as a rotation through 0 and yet setting θ=2π in that transformation maps a state ψ to -ψ. Now |a|² = |-a|² so the probability of observing spin up or spin down is unaffected. But as I've been showing over previous posts, flipping a sign in a state can make a big difference as soon as you start performing interference experiments. The same goes for any angle: if I rotate through π should I use θ=π or θ = 3π? So can the transformation I've given make sense?

The transformation does make sense if you consider that in any physical process that rotates an electron the transformation will evolve continuously over time. Electrons don't just instantly rotate. In other words, if a rotation is applied to an electron then it will follow a path in SO(3), not just be an instantaneous application of an element of SO(3). And that allows us to resolve the ambiguity: the rotations of electrons are described by the double cover of SO(3) known as SU(2). So a rotation through 360 degrees doesn't return you to the identity although a 720 degree rotation does. The transformation I gave above is completely unambiguous if you continuously rotate an electron around the x-axis tracking a continuous value of θ, after all, the double cover is basically just the set of continuous paths from the identitiy in SO(3) (with homotopic paths considered equivalent).

And that's the bizarre fact: electron rotations aren't described by SO(3), they're described by SU(2). In particular, rotating an electron through 360 degrees does not return it to its original state, but a rotation through 720 degrees does! In a sense, like Dirac's belt, electrons can remember something about the path they took to get where they are, in particular they remember how many twists there were in the path.

What does this mean experimentally? the first thing to note is that this is true not just for electrons but any spin-1/2 fermion. This included protons and neutrons. The stuff I've been talking about manifests itself in a number of ways. In particular, the spin of a particle affects how a magnetic field acts on it. For example, spin-up and spin-down particles can be separated into distinct beams using Stern-Gerlach apparatus. Also, the spin of particles precesses in a magnetic field and this is used on a regular basis in NMR. These two facts allow us to easily manipulate and measure the spin of fermions. In other words, the fact that fermions remember how many twists there are in their rotations isn't just some esoteric nonsense, it's now engineering and the theory is tested repeatedly all over the world.

Every familiar object is invariant under rotations through 360 degrees. So the fact that electrons need to be rotated through 720 degrees to return them to their original state seems like one of the most bizarre facts about the universe I know of. And yet many books that introduce spin just slip in this fact in a routine way as if it were no different to any other.

The fact that the biggest connected cover of SO(3) is the double cover puts a big constraint on the kinds of weird effects like this can happen. We can have a 360 degree rotation multiply by -1, but not by i, because a 720 degree rotation absolutely has to return us to where we started from. But suppose the universe were 2-dimensional. If you remember what I said about SO(2) you may notice that no such constraints apply because SO(2) has an infinite cover. There is a group in which all of the rotations through 360n degrees are distinct for distinct n. This means that a physical system could have its state multiplied by any factor (of modulus 1) when rotated through 360 degrees. Particle that behave this way are called anyons. But we live in a 3D universe so we don't expect any fundamental particles to have this property. However, in quantum mechanics any kind of 'excitation' of a physical system is quantised and can be thought of as a type of particle. These are known as quasiparticles. For example, just as light is made of photons, sound is also quantised as phonons. In the right kind of solid state medium, especially those that arise from some kind of 2D lattice, it seems quite plausible that anyons might arise. This gives rise to the so called fractional quantum hall effect. Anyons might one day play an important role in quantum computing via topological quantum computation.


Rahul Narain said...

I just wanted to say that I have been really enjoying your recent articles. You have a great writing style, and the topics you cover are mind-stretchingly fascinating.

Not to mention I had great fun doing the Dirac belt trick and the Feynman plate trick. It's delightful how the great minds of our time invented little party tricks that have deep mathematical connections.

By the way, you may be interested in using Unicode characters U+2329 and U+232A for the bra-ket notation.

sigfpe said...

Thanks for the kind words!

Those unicode characters are great. But I still don't have a good feel for how many font and browser combinations can display them correctly.

sigfpe said...


I forgot to mention Feynman's Dirac memorial lecture which talks a bit about the half angle formula and relates it to the plate trick. I was lucky enough to attend that lecture and see the plate trick first hand! Sadly I was very hungover that day (exams finished the night before) and didn't fully appreciate it at the time!

William said...

Regarding notation, have you tried this for rendering latex? I gave it a try and it seems to work quite well.

sakthi said...

This is the best explanation for Sipn 1/2 particle behaviour


Blog Archive