Michael Weiss - Lie Groups and QM PDF

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Lie Groups and Quantum Mechanics

Michael Weiss

1 Introduction
These notes attempt to develop some intuition about Lie groups, Lie algebras,
spin in quantum mechanics, and a network of related ideas. The level is rather
elementary— linear algebra, a little topology, a little physics. I don’t see any
point in copying proofs or formal definitions that can be had from a shelf full
of standard texts. I focus on a couple of concrete examples, at the expense of
precision, generality, and elegance. See the first paragraph on Lie groups to get
the flavor of my “definitions”. I state many facts without proof. Verification
may involve anything from routine calculation to a deep theorem. Phrases like
“Fact:” or “it turns out that” give warning that an assertion is not meant to
be obvious.
A quote from the Russian mathematician V. I. Arnol’d:

It is almost impossible for me to read contemporary mathemati-


cians who, instead of saying “Petya washed his hands,” write simply:
“There is a t1 < 0 such that the image of t1 under the natural map-
ping t1 → Petya(t1 ) belongs to the set of dirty hands, and a t2 ,
t1 < t2 ≤ 0, such that the image of t2 under the above-mentioned
mapping belongs to the complement of the set defined in the pre-
ceding sentence.”

A taste of things to come: consider the spin of an electron. One would like to
visualize the electron as a little spinning ball. This is not right, yet not totally
wrong. A spinning ball spins about an axis, and the angular velocity vector
points along this axis. You can imagine changing the axis by rotating the space
containing the ball. 1 Analogously, the quantum spin state of an electron has
an associated axis, which can be changed by rotating the ambient space.
The classical concepts of rotation and angular velocity are associated with
SO(3), the group of rotations in 3-space. SO(3) is an example of a Lie group.
1 If you want to be really concrete, imagine a spinning gyroscope fitting snugly in a box.

Rotate the box.

1
Another Lie group, SU (2), plays a key role in the theory of electron spin. Now
SO(3) and SU (2) are not isomorphic, but they are “locally isomorphic”, mean-
ing that as long as we consider only small rotations, we can’t detect any differ-
ence. However, a rotation of 360◦ corresponds to a element of SU (2) that is not
the identity. Technically, SU (2) is a double cover of SO(3).
Associated with every Lie group is something called its Lie algebra. The Lie
algebra is a vector space, but it has additional structure: a binary operation
called the Lie bracket. For the rotation group, the elements of the corresponding
Lie algebra can be thought of as angular velocities. Indeed, angular velocities
are usually pictured as vectors in elementary physics (right hand rule of thumb).
The Lie bracket for this example turns out to be the familiar cross-product from
vector algebra. (Unfortunately, I won’t get round to discussing the Lie bracket.)
The Lie algebras of SO(3) and SU (2) are isomorphic. This is the chief technical
justification for the “electron = spinning ball” analogy. The non-isomorphism
of SU (2) and SO(3) has subtle consequences. I can’t resist mentioning them,
though these notes contain few further details. Electrons are fermions, a term in
quantum mechanics which implies (among other things) that the Pauli exclusion
principle applies to them. Photons on the other hand are bosons, and do not
obey the exclusion principle. This is intimately related to the difference between
the groups SU (2) and SO(3). Electrons have spin 12 , and photons have spin 1.
In general, particles with half-odd-integer spin are fermions, and particles with
integer spin are bosons.
The deeper study of the electron involves the Dirac equation, which arose out
of Dirac’s attempt to marry special relativity and quantum mechanics. The
relevant Lie group here is the group of all proper Lorentz transformations.

A Rough Road-map. The basic plan of attack: show how elements of SU (2)
correspond to rotations; then apply this to the spin of the electron.
I start with the most basic concepts of Lie group and Lie algebra theory. SO(3)
is the ideal illustrative example: readily pictured, yet complicated enough to be
interesting. The main goal is the double covering result. I do not take the most
direct path to this goal, attempting to make it appear “naturally”. Once we do
have it, the urge to explore some of the related topology is irresistible.
Next comes physics. Usually introductory quantum mechanics starts off with
things like wave/particle duality, the Heisenberg uncertainty principle, and so
forth. Technically these are associated with the Hilbert space of complex-valued
L2 functions on R3 — not the simplest Hilbert space to start with. If one ignores
these issues and concentrates solely on spin, the relevant Hilbert space is C2 .
(Feynman’s Lectures on Physics, volume III, was the first textbook to take this
approach in its first few chapters.) SU (2) makes its entrance as a symmetry
group on C2 . I conclude with a few hand-waves on some loose ends.

2
2 Lie Groups
A Lie matrix group is a continuous subgroup of the group of all non-singular
n × n matrices over a field K, where K is either Ror C. “Continuous” really is
a shorthand for saying that the Lie group is a manifold. The rough idea is that
the components of a matrix in the group can vary smoothly; thus, concepts like
“differentiable function f : R → G” should make sense. I’ll just say “Lie group”
for Lie matrix group, though many mathematicians would groan at this.
Example: O(n) is the group of all orthogonal n × n matrices, i.e. all matrices A
with real components such that At A = 1. This is just the group of all isometries
of Rn which leave the origin fixed. (Standard calculation: let x be a column
vector. Then (Ax)t (Ax) = xt x, i.e., the norm of x equals the norm of Ax.)
Note also that the equations At = A−1 and AAt = 1 follow from At A = 1.
If At A = 1, then we have immediately det(A)2 = 1, i.e., det(A) = ±1. SO(n)
is the subgroup of all matrices in O(n) with determinant 1. Fact: SO(n) is
connected, and is in fact the connectedness component of 1in O(n). I will focus
initially on SO(2) and SO(3). These are, colloquially, the groups of rotations
in 2-space and 3-space. O(n) is the group of reflections and rotations.
Digression: the well-known puzzle, “why do mirrors reverse left and right, but
not up and down?” is resolved mathematically by pointing out that a mirror
perpendicular to the y-axis performs the reflection:

(x, y, z) → (x, −y, z)

i.e., corresponds to the following matrix in O(3):


⎡ ⎤
1 0 0
⎣ 0 −1 0 ⎦
0 0 1

For some psychological reason, people tend to think of this as the composition of
a 180◦ rotation about the z-axis followed by a reflection in a plane perpendicular
to the x-axis: (x, y, z) → (−x, −y, z) → (x, −y, z). This makes it seem that the
mirror is treating the x and z axes differently (left/right vs. updown), though
it really isn’t. End of digression.
Example: SO(2), rotations in 2-space. Since det(A) = 1, it is easy to write
down the components of A−1 . Equating these to At , we see that A has the
form:  
a −b
b a
with the constraint that a2 + b2 = 1. We can set up a one-one correspondence
between this matrix and the complex number a + ib on the unit circle. This is

3
a Lie group isomorphism between SO(2) and the unit circle. We can of course
find an angle θ for which a = cos θ and b = sin θ.
Elements of SO(2) have real components, but it is enlightening to consider
SO(2) as a subgroup of the group of all non-singular complex 2 × 2 matrices.
Fact: any matrix in SO(2) is similar to a matrix of the form
   iθ 
a + ib 0 e 0
=
0 a − ib 0 e−iθ

Of course, the new basis vectors have complex components.


Example: SO(3), rotations in 3-space. Fact: any element of SO(3) leaves a
line through the origin fixed. This seems obvious to anyone who has rolled a
ball, but it is not totally trivial to prove. I will outline two arguments, both
instructive.
First, an approach Euclid might have followed. Fact: any isometry of 3-space
that leaves the origin fixed is the composition of at most three reflections. For
consider the initial and final positions of the x, y, and z axes. One reflection
moves the x-axis to its final position, pointing in the correct direction; a second
reflection takes care of the y-axis; a third may be needed to reverse the direction
of the z-axis. We see that an orientation-preserving isometry is the composition
of two reflections. The intersection of the planes of these two reflections gives a
line that is left fixed.
Next, a linear algebra approach. We want to show that A ∈ SO(3) leaves a vec-
tor fixed (Av = v), or in other words that 1 is an eigenvalue. Note first that the
characteristic polynomial for A is a cubic with real coefficients. Hence it has at
least one real root. Furthermore, At A = 1 implies that A∗ A = 1 trivially, since
all components of A are real. By the spectral theorem, A can be diagonalized
(working over C) and its eigenvalues are all on the unit circle. Since the char-
acteristic polynomial has real coefficients, non-real roots appear in conjugate
pairs. It follows that A is similar to a matrix of the form diag(±1, ±1, ±1) or
diag(±1, eiθ , e−iθ ). Since det(A) = 1, and the determinant is the product of the
roots, at least one eigenvalue must be 1.
The first argument can also be cast in linear algebra form. This leads to House-
holder transformations.

3 Lie Algebras
Let G be a Lie group. Let x(t) be a smooth curve in G passing through the
unit element 1of G, i.e., a smooth mapping from a neighborhood of 0 on the
real line into G with x(0) = 1. T (G), the tangent space of G at 1, consists of
dx(t) 
all matrices of the form dt  , or just x (0) in a less clumsy notation.
t=0

4
T (G) is the Lie algebra of G. I will show in a moment that T (G) is a vector
space over R, and I really should (but I won’t) define a binary operation [x, y]
(the Lie bracket) on T (G) × T (G).
Proof that T (G) is a vector space over R: if x(t) is a smooth curve and x(0) = 1,
then set y(t) = x(kt), k ∈ R. This is also a smooth curve and y  (0) = kx (0). So
T (G) is closed under multiplication by elements of R. (Note this argument fails
for complex k.) Similarly, differentiating z(t) = x(t)y(t) (with x(0) = y(0) = 1,
as usual) shows that T (G) is closed under addition.
Historically, the Lie algebra arose from considering elements of G “infinites-
imally close to the identity”. Suppose  ∈ R is very small, or (pardon the
expression), “infinitesimally small”. Then x (0) is approximately x()−x(0)
 , or
(remembering x(0) = 1)
x() ≈ 1 + x (0)
Historically, x() is a so-called infinitesimal generator of G.
Robinson has shown how this classical approach can be made rigorous, using
non-standard analysis. Even without this, the classical notions provide a lot of
insight. For example, let n be an “infinite” integer. Then if t ∈ R is an ordinary
real number (not “infinitesimal”), we can let  = t/n and so

n
tx (0) 
x()n ≈ 1 + ≈ etx (0)
n
Assume that the left hand side is an ordinary “finite” element of G. Write v for
x (0), an arbitrary element of the Lie algebra T (G). This suggests there should
be a map (t, v) → etv from R × T (G) into G.
In fact, the following is true: for any Lie group G with Lie algebra T (G), we have
a mapping exp from T (G) into G such that exp(0) = 1, and exp ((t1 + t2 )v) =
exp(t1 v) exp(t2 v), for any t1 , t2 ∈ R and v ∈ T (G).
It also turns out that the Lie algebra structure determines the Lie group struc-
ture “locally”: if the Lie algebras of two Lie groups are isomorphic, then the
Lie groups are locally isomorphic. Here, the Lie algebra structure includes the
bracket operation, and of course one has to define local isomorphism.
Now for our standard example, SO(n). Notation: the Lie algebra of SO(n) is
so(n). If you differentiate the condition At A = 1 and plug in A(0) = 1, you
will conclude that all elements of so(n) are anti-symmetric. Fact: the converse
is true.
Example: SO(2), rotations in 2-space. All elements of so(2) have the form
 
0 −c
c 0

An element of so(2) can be thought of as an angular speed.

5
In the earlier discussion of SO(2), I set up a one-one correspondence
 
a −b
↔ a + ib
b a

between matrices and complex numbers. (We had the restriction a2 + b2 = 1,


but we can drop this and still get an isomorphism between matrices of this
form and C.) Then the displayed element of so(2) corresponds to the purely
imaginary number ic, and we have a map (t, ic) → eict mapping R × so(2) onto
the unit circle, which in turn is isomorphic to SO(2).
SO(3), rotations in 3-space. Elements of so(3) have the form:
⎡ ⎤
0 −c b
⎣ c 0 −a ⎦ ↔ (a, b, c) ∈ R3
−b a 0

(We’ll see the reason for the peculiar choice of signs and arrangement of a, b,
and c shortly.)
Fact: the vector (a, b, c) is the angular velocity vector for the above element of
so(3). What does this mean? Well first, let v0 ∈ R3 be some arbitrary vector;
if A(t) is a curve in SO(3), and we set v(t) = A(t)v0 , then v(t) is a rotating
vector, whose tip traces out the trajectory of a moving point. The velocity of
this point at t = 0 is A (0)v0 . It turns out that A (0)v0 equals the cross-product
(a, b, c) × v0 , which characterizes the angular velocity vector. The next few
paragraphs demonstrate this equality less tediously than by direct calculation.
Let
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 0 0 0 0 1 0 −1 0
x̂ = ⎣ 0 0 −1 ⎦ , ŷ = ⎣ 0 0 0 ⎦ , ẑ = ⎣ 1 0 0 ⎦
0 1 0 −1 0 0 0 0 0

so the general element of so(3) can be written ax̂ + bŷ + cẑ. And x̂, ŷ, and
ẑare simply the elements of so(3) corresponding to unit speed uniform rotations
about the x, y, and z axes, respectively— as can be seen by considering their
effects on the standard orthonormal basis.
This verifies the equation A (0)v0 = (a, b, c) × v0 for the special cases of A (0) =
x̂, ŷ, and ẑ. The general case now follows by linearity.

The Adjoint Representation. Elements of SO(3) act on R3 , or equivalently


on orthonormal bases of R3 (frames).2 But SO(3) can also be regarded as a
set of transformations on the vector space so(3), as we will see in a moment.
Intuitively, the triple (x̂, ŷ, ẑ) takes the place of the standard frame for R3 .
2 In other words, a element of SO(3) determines a mapping of R3 to R3 . As it happens,

the action is faithful, i.e., the mapping determines the element of SO(3).

6
General fact: for any Lie group G, there is a homomorphism (also known as a
representation) of G into the group of non-singular linear transformations on
the vector space T (G), with kernel Z(G), the center of G.
Here’s how it goes. For any group G, we have the group of inner automorphism
Inn(G) and a homomorphism G → Inn(G) defined by g → ιg , where ιg (h) =
ghg −1 . The kernel is Z(G). The automorphism ιg is furthermore determined
completely by its effects on any set of generators for G.
Now take G to be a Lie group. Let’s consider the effect of ιg on an “infinitesimal”
generator 1 + h, where h ∈ T (G).

g(1 + h)g −1 = 1 + ghg −1

Or in terms of derivatives, if x(t) is our prototypical smooth curve through 1,


then the derivative of gx(t)g −1 at t = 0 is gx (0)g −1 . So the vector space T (G)
is closed under the map h → ghg −1 . (Remember that both G and T (G) are
sets of matrices.) Z(G) is clearly contained in the kernel, and in fact: Z(G) is
the kernel.
For SO(3), this is rather intuitive. Suppose h ∈ SO(3) is a rotation about the
axis determined by vector v ∈ R3 . Then ghg −1 is a rotation about the axis gv:
(ghg −1 )(gv) = ghv = gv. If we think of h as an infinitesimal rotation, then we
see that the action of SO(3) on so(3) given by ιg looks just like the action of
SO(3) on R3 .
Only in three dimensions do things work out so neatly. SO(2) is abelian, and the
adjoint representation for abelian Lie groups is boring— the trivial homomor-
phism. And the vector space so(4) has dimension 6, so the adjoint representation
gives an imbedding of SO(4) in the group of non-singular 6 × 6 matrices.

Unitary Matrices: SU (n). Now for a different example. U (n) is the group
of unitary n × n matrices, i.e., complex matrices satifying A∗ A = 1. An easy
computation shows that | det(A)| = 1. SU (n) is the subgroup for which the
determinant is 1 (unimodular matrices). Unlike the situation with O(n) and
SO(n), the dimensions of U (n) and SU (n) (as manifolds) differ by 1.
The Lie algebras of U (n) and SU (n) are denoted u(n) and su(n), respectively.
Differentiating A∗ A = 1 we conclude that u(n) consists of anti-Hermitian ma-
trices: B ∗ = −B. Note that B is anti-Hermitian if and only if iB is Hermitian.


Fact: if A(0) = 1, then d detdtA(t)  = tr A (0) (where tr = trace). (Expanding
t=0
by minors does the trick.) This makes one half of the following fact obvious: the
Lie algebra for the Lie group of unimodular matrices consists of all the traceless
matrices.
For the special case SU (2), things work out very nicely. Since det(A) = 1, one
can write down the components for A−1 easily, and equating them to A∗ , one

7
concludes that SU (2) consists of all matrices of the form:
         
a + id c + ib 1 0 0 i 0 1 i 0
=a +b +c +d
−c + ib a − id 0 1 i 0 −1 0 0 −i

= a1 + bi + cj + dk, a2 + b2 + c2 + d2 = 1
defining i,j,kas the given 2 × 2 matrices in SU (2). Exercise: these four elements
satisfy the multiplication table of the quaternions, so SU (2) is isomorphic to
the group of quaternions of norm 1. (The somewhat peculiar arrangement of
a, b, c, d in the displayed element of SU (2) is dictated by convention.)
Next, an arbitary anti-Hermitian matrix looks like:
         
i(a + d) c + ib i 0 0 i 0 1 i 0
=a +b +c +d
−c + ib i(a − d) 0 i i 0 −1 0 0 −i

= ai1 + bi + cj + dk
This is traceless if and only if a = 0. So we have a canonical 1–1 correspondence
between su(2) and R3 , and so also with so(3): bi + cj + dk ↔ (b, c, d) ↔
bx̂ + cŷ + dẑ.
It turns out that this correspondence is a Lie algebra isomorphism. SU (2) and
SO(3) are locally isomorphic, but not isomorphic— as we will see next.
SU (2) acts on su(2) via the adjoint representation. But we have a 1–1 corre-
spondence between su(2) and R3 , so we have a representation of SU (2) in the
group of real 3×3 matrices. Let A be an element of SU (2) and v = bi+cj+dk be
an element of su(2). Note that det v = b2 + c2 + d2 . Since the map v → AvA−1
preserves determinants, it preserves norms when considered as acting on R3 .
So the adjoint representation maps SU (2) into O(3). Fact: it maps SU (2) onto
SO(3).
Incidentally, you can see directly that v → AvA−1 preserves anti-Hermiticity
by writing it v → AvA∗ .
The kernel of the adjoint representation for SU (2) is its center, which clearly
contains ±1— and in fact, consists of just those two elements. So we have a
2–1 mapping SU (2) → SO(3). Our double cover! I’ll look at the topology of
this in a moment.
Physicists prefer to work with the Pauli spin matrices instead of the quaternions.
The Pauli matrices are just the Hermitian counterparts to i, j, and k:
i = iσx , j = iσy , k = iσz
They form a basis (with 1)for the vector space of Hermitian 2 × 2 matrices:
         
a + d b − ic 1 0 0 1 0 −i 1 0
=a +b +c +d
b + ic a − d 0 1 1 0 i 0 0 −1

8
= a1 + bσx + cσy + dσz
SU (2) acts on the space of traceless Hermitian 2 × 2 matrices in the same way
as on su(2): h → ghg −1 .

Picturing the correspondences. We have an embarrassment of riches: so


many correspondences that it is easy to get confused. This paragraph tries to
fit things into a coherent framework, to make them easier to remember.
We have two fundamental concepts that are easy to visualize:

acts on vectors in 3-space ≡


Rotations −→
angular velocities
The rotation group acts on the space of vectors. For any representation of the
rotation group and any representation of the vector space, we would like to have
an intuitive grasp of the action. (Since the action preserves distances, one can
also consider the action of the group just on the set of vectors of norm 1, i.e.,
on the sphere.)
Possible representations for the rotation group: SO(3), SU (2), quaternions of
norm 1. Possible representations for the vector space: R3 , so(3), su(2), and the
space of traceless Hermitian matrices. For SO(3) with R3 , the action is matrix
multiplication on the left: v → Av. For SO(3) with so(3), or SU (2) with su(2)
or the traceless Hermitian matrices, the action is conjugation: v → AvA−1 =
AvA∗ .
The actions with SO(3) are all faithful. The actions with SU (2) are all two-to-
one— A and −A determine the same action (i.e., rotation).
Let’s look at the SU (2) actions in more detail. The space of traceless Hermitian
matrices consists of all matrices of the form xσx + yσy + zσz , x, y, z ∈ R. This
is in one-one correspondence with su(2):

xσx + yσy + zσz ↔ i(xσx + yσy + zσz ) = xi + yj + zk

So if we understand one SU (2) action, we understand the other. I’ll use Pauli
matrices from now on.
An arbitary element A of SU (2) looks like
 
a + id c + ib
A= = a1 + biσx + ciσy + diσz , a2 + b2 + c2 + d2 = 1
−c + ib a − id

and we see that A∗ = a1 − biσx − ciσy − diσz . So the result of acting with
A on v can be computed simply by working out the product (a1 + biσx +
ciσy + diσz )(xσx + yσy + zσz )(a1 − biσx − ciσy − diσz ). For this we need the

9
multiplication table of the σ’s. This is simply:

σx2 = σy2 = σz2 = 1


σx σy = −σy σx = iσz
σy σz = −σz σy = iσx
σz σx = −σx σy = iσy

The easiest way to check this is to work out the action of the σ’s on (p, q) ∈ C2 :
   
p q
σx =
q p
   
p −iq
σy =
q ip
   
p p
σz =
q −q

We have here non-commuting Hermitian matrices whose product is not Hermitian—


in fact is anti-Hermitian. Now, one has an analogy between matrices and com-
plex numbers, under which “Hermitian” goes with “real”, “anti-Hermitian”
goes with “purely imaginary”, and “unitary” goes with “on the unit circle”.
The σ matrices provide a striking example of the analogy breaking down. Non-
commutativity is the culprit— for of course the product of commuting Hermitian
matrices is Hermitian.
Example: what rotation does iσz represent? Its action on the unit x vector, σx ,
is just σx → iσz σx (−i)σz = iσy σz = −σx . Similar calculations show that y goes
to −y and z stays put, so we have a 180◦ rotation about the z-axis. Similarly
for iσx and iσy .
The exponential map provides a mapping from su(2) into SU (2):

ibσx + icσy + idσz → exp(ibσx + icσy + idσz )

(Warning: the exponential of a sum is not in general the product of the expo-
nentials, because of non-commutativity.) For the rotation group (as we’ve seen)
this says simply that an angular velocity determines a rotation— e.g., by the
prescription “rotate at the given angular velocity for one time unit”. The basis
of “angular velocities” in su(2) is (iσz , iσy , iσz ). Let us consider rotations about
the z-axis.
 ib 
ibσz e 0
e = = cos b 1 + i sin b σz
0 e−ib

(since iσz acts separately on each coordinate.) Perhaps the clearest way to
exhibit the action of this rotation on v = xσx + yσy + zσz is to work entirely in

10
matrix form:
 ib     
e 0 z x − iy e−ib 0 z e2ib (x − iy)
=
0 e−ib x + iy −z 0 eib e −2ib
(x + iy) −z

i.e., a rotation about the z-axis of −2b radians. Exercise: the same sort of thing
holds for iσy and iσz .
So one can directly picture the action of SU (2) on vectors in 3-space. The
“double angles” 2b, etc., stem from the two multiplications in the action: v →
AvA∗ . And the double angles in turn are the reason the map from SU (2) to
SO(3) is two-to-one.
SU (2) acts on C2 via left multiplication: v → Av, where v is a column vector.
Can one picture v as some kind of geometric object in 3-space? Yes indeed!
An object known as a spin vector embodies v geometrically. But I won’t get to
them.

Topology. Fact: SU (n) is simply connected. So SU (2) is the universal cov-


ering space of SO(3). The covering map is 2–1. It follows from standard results
in topology that the fundamental group of SO(3) is Z2 .
Recall that elements of SU (2) all take the form a1 + bi + cj + dk with a2 + b2 +
c2 +d2 = 1. Therefore SU (2) is topologically S 3 , the 3-dimensional hypersphere.
(This is enough to show that SU (2) is simply connected.) Because the kernel is
{±1} (verify!), antipodal points are identified on mapping into SO(3), so SO(3)
is homeomorphic to real projective 3-space.
This can be seen another way. A rotation can be specified by a vector along
the axis of the rotation, with magnitude giving the angle of the rotation. This
serves to identify elements of SO(3) with points inside or on a ball of radius
π. However, antipodal points on the surface of the ball represent the same
rotation. The resulting space (a three-dimensional ball with antipodal points
on the surface identified) is well-known to be homeomorphic to real projective
3-space. (If you think about this argument for a bit, you should see an implicit
use of the exponential mapping from so(3) into SO(3).)
A loop in the topological space SO(3) can be visualized as a continous “trajec-
tory” of rotations: we take a rigid object and turn it around in some fashion,
finally restoring it to its original orientation. The following fact can be deduced
from this: if a solid object is connected by threads to a surrounding room, and
the object is turned through 720◦ , then the threads can be untangled without
turning the object any more. However, if the object is turned through 360◦,
then the threads cannot be untangled. (The two-thread version of this is known
as “Dirac’s string trick”.) In this sense, a continuous turn through 360◦ is not
the same as no turn at all (but a 720◦ turn is.)

11
4 Quantum Mechanics: Two-state Systems
The framework of quantum mechanics rests on three pillars: the Hilbert space
of quantum states; the Hermitian operators, also called observables; and the
unitary evolution operators. I start by trying to attach some pictures to these
abstractions.
The simplest classical system consists of a single point particle coasting along
in space (perhaps subject to a force field). To “quantize” this, you’ll need
the Hilbert space of complex-valued L2 functions on R3 , and you’ll encounter
unbounded operators on this space. So goes the tale of history: Heisenberg,
Schrödinger, Dirac and company cut their milk teeth on this problem.
I will take an ahistorical but mathematically gentler approach. The Hilbert
space for a two-state quantum system is C2 , and the operators can all be rep-
resented as 2 × 2 complex matrices. The spin of an electron provides a physical
example. That is, if we simply ignore position and momentum (“mod them
out”, so to speak), we have a physical picture that can be modelled by this
(relatively) simple framework. (As noted before, Feynman’s Lectures, volume
III, starts off like this.)

Quantum states. Generally, the quantum state of a physical system is spec-


ified by a non-zero vector in a Hilbert space over the complex numbers (call
it H), with the understanding that x and cx specify the same quantum state,
where x is in H and c is a non-zero complex number. In other words, the set
of quantum states is (H − {0})/ ∼, where ∼ is the equivalence relation defined
by x ∼ y if and only if x = cy for some c
= 0. Sometimes I will abuse language,
and say “the state vector of the system is v ∈ H” instead of saying “the state
of the system is specified by v ∈ H.”
The state of our spinning electron is therefore specified by giving a ratio a : b
of two complex numbers (since H = C2 ), not both zero. Mathematically, this
is just the complex projective line. The correspondence between points on the
complex projective line and points on the Riemann sphere is well-known; I will
make use of it in a moment. 3
Physically, we want to picture the electron as a little ball, spinning about some
axis. It has angular momentum. In more detail, the angular momentum vector
3 A quick review: write (a : b) for the equivalence class of (a, b). We will associate either a

complex number w or else ∞ with each class (a : b). If b = 0, then (a : b) = (w : 1), where
w = a/b. All pairs of the form (a, 0) belong to the class (1 : 0). So we associate a/b with
(a : b) if b = 0, and ∞ with (1 : 0). Mapping the complex plane plus ∞ to the Riemann
sphere via the usual stereographic projection completes the trick. Some sample points to bear
in mind: the north pole is (1 : 0); the south pole is (0 : 1); points on the equator have the
form (eiθ : 1). (For the purist, the special treatment of ∞ rankles. It is not singled out on
either the complex projective line or on the Riemann sphere. Later I will show how to set up
the correspondence without this blemish.)

12
is a vector pointing along the axis of rotation, with length proportional to the
speed of rotation. This defines the vector up to sign. The sign ambiguity is
resolved, conventionally, by the right hand rule of thumb: if you curl the fingers
of your right hand in the direction of rotation, the thumb points in the direction
of the angular momentum vector.
Classically, the electron could be spinning at any speed, so its angular momen-
tum could have any magnitude. In quantum mechanics, the angular momentum
is quantized: its magnitude (measured along the axis of rotation) must be h̄/2
for any spin 12 particle. (Note: h̄ = h/2π, where h of course is Planck’s con-
stant.)
The quantization of angular momentum, although inexplicable classically, is
easy enough to picture: we just stipulate that our spinning ball must be spinning
at a particular rate. So we should be able to specify the spin state of the electron
just by giving a direction in 3-space, or equivalently, by picking a point on the
sphere S 2 . But I just noted that the set of states is “isomorphic” to the Riemann
sphere. So everything fits.

Hermitian observables. So much for quantum states. Next I look at mea-


surement. Generally, a measurement of something in quantum mechanics (spin,
energy, momentum . . . ) is associated with a Hermitian operator, say A, on the
Hilbert space of states. In finite dimensional cases, possible results of the mea-
surement must be eigenvalues of A. If v is an eigenvector of A with eigenvalue
λ, then measuring a system with state vector v will always give you the result
λ.
What happens if the state vector of the system is not an eigenvector? Here we
skate perilously close to philosophy! To make things simple4 , suppose (v1 , ..., vn )
is an orthonormal basis for H consisting of eigenvectors for A, say Avi = λi vi .
Suppose the system has state vector w = a1 v1 + . . . + an vn , and assume w is
normalized (i.e., |w| = 1). The so-called “collapse” interpretation of quantum
mechanics says: (i) The measurement process forces the system to jump ran-
domly to a state with state vector v1 , or v2 , or . . . or vn . (ii) The probability
that the system jumps to a state with state vector vi is |ai |2 = | w, vi |2 . (iii) If
the system ends up with state vector vi , then the measurment yields result λi .
(Note that you’ll get real-valued measurement results because A is Hermitian.)
Nearly everyone agrees that the collapse interpretation will correctly predict the
results of experiments. Whether it is what’s “really going on” is the subject of
endless debates. What we have here are rules for calculating probabilities. At
least four different philosophies have been draped around the rules.
How does measurement look for the electron? Say we want to measure the
4 In the infinite dimensional case, you have to use the spectral decomposition of A instead

of an orthonormal basis of eigenvectors; another reason why spin is simpler than position.

13
component of spin along the z-axis. Since the electron is charged, it acts like a
little magnet, with north and south poles along the axis of rotation. (Circulating
charge causes a magnetic field. Think of an electromagnet— a coil of wire with
an electric current flowing around in it.) A physicist would say that the electron
has a magnetic moment.
We can use the magnetic moment to measure the spin. Stern and Gerlach got a
Nobel prize for doing just that. They sent a beam of electrically neutral silver
atoms through a magnetic field. It turns out that the magnetic moments of
the electrons in a silver atom cancel out in pairs except for one electron, so we
can pretend (so far as the spin is concerned) that we’re looking at a beam of
electrons passing through a magnetic field.
The magnetic field was designed to produce a force on the electrons. An electron
N
with spin pointing up would look like an S magnet, and would experience
S
an upward force; an electron with spin pointing down would look like an N
magnet, and would experience a downward force. Classically, you would expect
an electron with spin at angle α to the vertical to experience an upward force
proportional to cos α.5 So the electron beam should be spread out into a vertical
smear, according to classical mechanics.
In fact, the beam splits into two beams, one up, one down. In other words, if we
measure the component of the spin along the vertical axis, we always find that
the spin is entirely up or entirely down. This is the most basic sense in which
the “spinning ball” analogy is wrong. The same two-valued behavior holds for
any measurement axis.
Classically this is inexplicable. How can the electron have spin up and spin
sideways at the same time? Answer: it doesn’t. After you’ve measured the
spin along the z-axis, the electron has vertical spin (say spin up). If you take
your vertically spinning electron and measure its spin along the x-axis, you have
a 50–50 chance at getting spin left or spin right. If you now repeat the spin
measurement along the z-axis, you have a 50–50 chance of getting spin up or
spin down. The x-axis measurement has destroyed the information obtained
from the z-axis measurement.
Let A be the Hermitian operator corresponding to “measure the spin along the
z-axis”. The eigenvalues (i.e., possible results) will be 1 and −1, if we choose
our units right.
 Pick abasis of two eigenvectors; then the matrix for A in this
1 0
basis is just , i.e., the Pauli matrix σz . (Common notation for the
0 −1
eigenvectors is |up and |down , although |dead and |alive are popular for that
other famous two-state system, Schrödinger’s Cat.)
5 Why wouldn’t the electron simply snap into alignment with the magnetic field? Answer:

the spinning electron would act like a gyroscope, and precess in response to the torque exerted
by the field. Thus it would maintain its angle of inclination to the field.

14
If σz is here, can σx and σy be far behind? In fact, these are the matrices for
measuring the x-component (respectively y-component) of spin, provided we
continue to use the same basis |up and |down .
It turns out that (1 : 0) represents spin up along the z-axis, (1 : 1) represents
spin along the x-axis, and (i : 1) represents spin along the y-axis. In a different
notation, |up + |down and i|up + |down are the state vectors for these two
spin directions. The x-axis and y-axis state vectors are not eigenvectors of σz .
The rules for calculating probabilities (clothed in any philosophy you like) yield
the 50–50 chances mentioned earlier.
As an exercise, you may like to chew on these remarks: if two measurements
can be done simultaneously, then the associated Hermitian operators must have
the same set of eigenvectors, and so the operators must commute. But the σ
matrices don’t commute. This accounts mathematically for the non-intuitive (or
at least non-classical) results of the Stern-Gerlach experiment. The Heisenberg
uncertainty principle stems from the same sort of considerations.
Now a general comment. Any linear operator on a Hilbert space H induces
a mapping on the space of states, since if x and cx are two state vectors for
the same state, then Ax and Acx = cAx will represent the same state. Can I
dispense with the Hilbert space entirely and just work with the space of states
and the induced mappings? The answer is yes, but it would be inconvenient.
If A is an observable with eigenvector v, say Av = λv, then the eigenvalue λ
has physical significance. But when we look at the action of A on the space of
states, all we notice (at first) is that A leaves the state corresponding to v fixed.
Nonetheless, the results of measurement are encoded in the action of A on the
space of states. A and cA, c
= 0, induce the same mapping on the states, and
the converse holds for the cases of interest to us (if A and B induce the same
state mapping, then A = cB for some scalar c
= 0). This scalar c will be real
for Hermitian A and B. If c
= 1, then A and B really represent the same
measurement, but expressed in different units (e.g., foot-pounds vs. ergs.)
Example: suppose Av = λv and Aw = μw, and Bv = λ v, Bw = μw, with
λ
= λ . A and B do the same thing to the quantum states determined by v
and w— namely, the states are left fixed. However, A and B send the state
determined by v + w to different states.

Unitary evolution operators. Suppose I change the state of a quantum


system “smoothly”. For example, I could move the system through space, or
I could “move it through time” (i.e., just wait— hence the term, “evolution
operator”), or I could (surprise!) rotate it.
It happens quite generally in quantum mechanics that all such state changes are
induced by unitary operators. Proving this clearly would require some physical
assumptions, and I won’t go into this at all. For our spinning electron, it turns

15
out you can even assume a little more: the change of state caused by rotating
the electron is induced by an operator in SU (2). The operator in question is
called a rotation operator.
How should we visualize the action of a rotation operator R on a state vector
v? We saw how to picture R as a rotation in 3-space by looking at its effects on
traceless Hermitian matrices: A → RAR∗ , where A = xσx +yσy +zσz . How can
we hook up the action of R on state vectors with the action of R on traceless
Hermitian matrices? It seems we need a correspondence between states (say
(a : b)) and matrices of the form xσx + yσy + zσz . We won’t get quite this, but
we’ll get something just as good.
The trick is to set up a correspondence between states and yet another kind
of matrix: a projection matrix. You probably noticed that the matrix σz does
a pretty good job specifying the state “spin up along the z-axis”. As it turns
out, σz is not a projection matrix, but it corresponds in a natural fashion to
1 1
2 1 + 2 σz ), which is.
Here’s how it goes for an arbitrary state vector v = a|up + b|down . Suppose v
is normalized, so |a|2 + |b|2 = 1. The projection matrix for v is given by taking
the product of the column vector v with the row vector v ∗ :
   
∗ a ∗ ∗ aa∗ ab∗
vv = [a , b ] =
b a∗ b bb∗

(Standard physicists’ notation is |v for v and v| for v ∗ . The product is |v v|.


The norm is v|v . Mathematicians prefer to talk about a vector space and its
dual instead of column vectors and row vectors, but these notes prefer concrete-
ness to elegance.)
Now, vv ∗ is a Hermitian matrix with determinant 0 (check!). It must therefore
take the form:
 
t + z x − iy
= t1 + xσx + yσy + zσz , t2 − x2 − y 2 − z 2 = 0
x + iy t − z

(If the appearence of t2 − x2 − y 2 − z 2 makes you think “Special Relativity!”,


you’re on the right track, but I won’t get into that.) However, the trace is not 0,
but aa∗ + bb∗ = 2t. Since I took v to be normalized (|a|2 + |b|2 = 1), it follows
that t = 12 .
So we have a mapping from state vectors to Hermitian matrices of the form
1 2 2 2 1
2 1 + xσx + yσy + zσz with x + y + z = 4 . And the latter are in an obvious
one-one correspondence with points on a sphere of radius one-half.
The mapping v → vv ∗ (restricted to normalized vectors) actually establishes
a one-one correspondence between states and our special class of Hermitian
matrices. For let c be a complex number of norm 1; then (cv)(cv)∗ = vv ∗ .

16
Why do I call vv ∗ a projection matrix? Answer: by analogy with projections in
ordinary real vector spaces, say R3 . If v is a vector of norm 1, and w is an arbi-
trary vector, then the projection of w “along the vector v” (i.e., in the subspace
spanned by v) is (v · w)v. Analogous to this, we define projv (w) = v, w v,
using the notation v, w for the inner product. In the “row vector, column
vector” notation, this is (v ∗ w)v = v(v ∗ w) = vv ∗ w. In physicists’ notation, this
is |v v|w .
We have acquired a new way of picturing the action of SU (2) on 3-space. The
formula v → Avv ∗ A∗ captures it succinctly. The mapping v → vv ∗ sets up a
one-one correspondence between the states (a : b) (i.e., the complex projective
line) and points on a sphere in 3-space. In fact this is just the Riemann sphere
mapping!
So the quantum states for the spin of an electron can be pictured as points
on a sphere. Elements of SU (2) correspond to the change in state induced by
rotating the electron, and this action of SU (2) can be pictured as a rotation of
the sphere. The naive pictures match up with the SU (2) formalism flawlessly.
The element −1 of SU (2) induces the identity mapping on the space of states,
since v and −v represent the same quantum state.
A simple computation illustrates how everything meshes. The rotation operator
for a clockwise 90◦ rotation about the y-axis is √12 (1 + iσy ). Indeed, if you work
out (1 + iσy )σx (1 − iσy ), you get 2σz , and likewise (1 + iσy )σz (1 − iσy ) = −2σx .
The x-axis maps to the z-axis, and the z-axis maps to minus the x-axis.
The example of electron spin illustrates two features of quantum mechanics very
clearly.

• Probabalistic character: The quantum state does not uniquely deter-


mine the result of experiment. Hence Einstein’s famous complaint, “I shall
never believe that God plays dice with the universe.” (Perhaps he plays
cards with the physicists?)
• Correspondence with classical physics: Classical physics emerges
from quantum mechanics by taking averages.

Some more technical features, also embodied in this example, and typical of
quantum mechanics:

• Need for complex numbers: The neat correspondence with the classi-
cal spinning ball picture wouldn’t work if we did everything over R.
• Non-commuting observables: You cannot simultaneously measure the
x and z components of spin (for example), because σx and σz do not
commute.

17
• Symmetry groups and observables: The rotation symmetry group
gives rise indirectly to the σ matrices, and ultimately to the notion of
angular momentum. The mathematical basis is the Lie groupLie algebra
correspondence.

Had I started with the first historical example, the single spinless particle coast-
ing in space, I would be illustrating the same morals with different actors:

• Need for complex numbers: The appropriate Hilbert space is the space
of complex-valued L2 functions on 3-space.
• Non-commuting observables: The momentum and position operators
do not commute, and you cannot simultaneously measure position and
momentum.
• Symmetry groups and observables: The group of translations in 3-
space gives rise to Lie group acting on the L2 Hilbert space; the momentum
operator emerges from the corresponding Lie algebra.

5 Loose Ends
There are so many loose ends that it seems pointless to try to tie them all up. I
will finish off with a few observations, meant more to tantalize than enlighten.

The Lie bracket. Can one reconstruct the Lie group from the Lie algebra?
In general, the multiplication table of a group is determined if you know the
multiplication table for its generators; why not try this with the “infinitesi-
mal” generators? If you try this approach, you will find you need to know the
commutators of infinitesimal elements, like x()y()x()−1 y()−1 .
My “definition” of the Lie algebra involved approximating infinitesimal gener-
ators by Taylor expansions out to the first order. In other words, I used only
first order derivatives. But to the first order, the commutators are zero!
Say we approximate an “infinitesimal” element of the Lie group out to the
second order:
2
x() ≈ 1 + x (0) + x (0)
2
If you work out the commutator, you will find expressions of the form vw − wv
appearing, where v and w belong to the Lie algebra. And one can verify that
vw − wv belongs in fact to the Lie algebra, as I’ve defined it, although vw and
wv in general don’t.
Remarkably, knowledge of these second order terms completely specifies the
structure of the Lie group near the identity. That is, if the Lie algebras are

18
isomorphic, then the Lie groups are locally isomorphic. Third and higher-order
terms are not needed.

Special relativity, spinors, and the Dirac equation. The crucial Lie
group for special relativity is the Poincaré group: all transformations of Minkowski
4-space (spacetime) that preserve the Minkowski pseudo-metric. The Lorentz
group is the subgroup that leaves the origin fixed, and the proper Lorentz group
is the subgroup of orientation preserving Lorentz transformations. The proper
Lorentz group in turn contains the rotation group of 3-space, SO(3).
Just as SU (2) is the double cover of SO(3), so SL(2) is the double cover of the
proper Lorentz group, where SL(2) is the group of unimodular 2 × 2 complex
matrices.
Say A is in SL(2) and v is in C2 . It turns out to be important to pry the
mapping vv ∗ → Avv ∗ A∗ apart into v → Av and v ∗ → v ∗ A∗ . The vector v
can be pictured as a geometric object consisting of a vector in space (rooted at
the orgin) with an attached “flag”, i.e., a half-plane whose “edge” contains the
vector. Moreover, if the flag is rotated through 360◦ , v turns into −v! (Recall
the earlier remarks on untangling threads.) Such an object is called a spin
vector. And just as one can create tensors out of the raw material of vectors,
so one creates spinors out of spin vectors.
Dirac invented spinors in the course of inventing (or discovering) the Dirac
equation, the correct relativistic wave equation for the electron. As it happens,
the σ matrices are not enough to carry the load; Dirac had to go up to 4 × 4
matrices (called the Dirac matrices). The σ matrices are imbedded in the Dirac
matrices.
I won’t repeat the story of how Dirac discovered antiparticles. Nor the story of
how he rediscovered knitting and purling (see Gamow’s Thirty Years that Shook
Physics.)

Spin and Statistics. The spin-statistics theorem of quantum field theory


says that particles with half-odd-integer spin (like the electron) must be fermi-
ons, while particles with integer spin (like the photon) must be bosons. Fermi-
ons obey Fermi-Dirac statistics, and hence obey the Pauli exclusion principle.
Bosons obey Bose-Einstein statistics.
The difference in statistics stems from the properties of the exchange operator.
This is a unitary operator, say Pex , which represents the effect of exchanging
two identical fermions, or two identical bosons. For the fermion case, one has
at a critical point in the calculations

Pex v = −v

19
and for the boson case,
Pex v = v
The minus sign for fermions ultimately derives from the double covering of
SO(3) via SU (2). Spinors also get into the act. Since I don’t fully understand
the story myself, this seems like a good place to stop.

20

You might also like