You are on page 1of 12

Lecture 4: Angular momentum review:

algebra, ladders, spin, addition (9/29/2005)


We will look at the angular momentum. According to Emmy Noethers
theorem, the conservation of (a component of) angular momentum is unseparable from the symmetry of physical laws with respect to rotations (around
an axis). We will discuss rotations next Tuesday. Today, we will focus on
the definition of the operator of angular momentum in quantum mechanics,
its algebraic properties such as commutators, the different eigenvalues that
the components of the angular momentum (and its square) may have, spherical harmonics as examples of eigenstates, and maybe spin and addition of
angular momenta.
Additional reading if you wish: Griffiths ch. 4.3-4.4 (especially
to completely understand the differential equations in , ). And
various tables of special functions, spherical harmonics, Legendre
polynomials, Clebsch-Gordan coefficients, and so forth.

Defining the orbital angular momentum


Even in classical physics, the orbital angular momentum is defined as the
cross product of ~x = (x, y, z) and ~p = (px , py , pz ):
~ = ~r p~ (ypz zpy , zpx xpz , xpy ypx )
L
Note that only the other components of the vectors ~x, ~p appear in a given
~ the relative sign is minus, and all three coordinates of L
~ may
component of L,
be obtained from each other by cyclical permutations of x y z x.
Recall that the cross product is perpendicular both to ~x and p~ and its
length is |x| |p| sin x,p while the orientation is given by the right hand rule.
~ can be used in quantum mechanics, too, by adding hats.
The definition of L
Because p is composed of derivatives, we have
~ = i
L
h(yz zy , zx xz , xy yx )
where i
h from p is included in all components and y means /y, for
example. Note that in the products of x, y, z with the derivatives, only
different components appear in the same product, and therefore there is no
ambiguity whether we define it as xp or px (different components commute
with each other).
1

Commutators and uncertainty


We now want to calculate the commutators
[Li , Lj ] = ?
where i, j = 1, 2, 3 (or, using different conventions, x, y, z). Lets start with
[Lx , Ly ]. For the sake of clarity, lets act on a wavefunction (x, y, z) with
this combined operator. The more experienced among you will be able to
write down this equation without , as an operator equation in which we
divide by the overall factor (i
h )2 :
(i
h)2 [Lx , Ly ](x, y, z) =

+
=

(yz zx zx yz
yz xz + xz yz
zy zx + zx zy +
zy xz xz zy ) (x, y, z) =
(yx xy ) (x)

Remember that all operators acted on the whole function on the right from
them; the nonzero contributions came from the first and fourth line because
[z , z] = 1, using the operator language, or if you need to imagine
because (z)0 = z 0 + . The result for the commutator may be recognized
as
[Lx , Ly ] = i
hLz .
Thats nice because we may also write the other two nonzero commutators
using the cyclical permutations x y z x and unify them into
[Li , Lj ] = i
hijk Lk
where ijk is completely antisymmetric and 123 = 1. At any rate, different
components of the angular momentum dont commute, and therefore they
cant be measured simultaneously; they are not compatible. By the generalized uncertainty principle this implies:
h

|hLz i|
2
where you have the uncertainties on the LHS but the actual expectation
value of the third component on the other side. The uncertainties of Lx , Ly
can only be simultaneously zero if Lz = 0; it will in fact only occur if the
~ equals zero.
whole L
Lx Ly

Squared angular momentum and ladders


Fine. So if the different components dont commute with each other, does it
mean that we can only determine one number related to the angular momentum (one component)? That would be bad because the angular momentum
should describe the dependence on the angles and there are two of them.
Fortunately it is not true because we can also define
L2 L2x + L2y + L2z
~ for example Lz :
that commutes with every component of L,
[L2x + L2y + L2z , Lz ] = Lx [Lx , Lz ] + [Lx , Lz ]Lx + Ly [Ly , Lz ] + [Ly , Lz ]Ly
= i
h(Lx Ly Ly Lx + Ly Lx + Lx Ly ) = 0.
We will later understand the vanishing of this commutator simply from the
~ 2 is a scalar and it thus does not transform under rotations genfact that L
erated by Lz .
Nevertheless, its great because we can now use the pair of operators
L2

and Lz

as a set of compatible observables which becomes complete if we add the


radial part of the energy, for example, in the case of spherically symmetric
potentials (we will instantly clarify that l, m in n, l, m of the Hydrogen atom
are related to the eigenvalues of L2 and Lz ).
So what are the eigenvalues of these two operators? In order to find
them, it is useful not to throw away Lx , Ly but use them in the following
combinations:
L+ = Lx + iLy , L = Lx iLy .
The first is called the raising operator and the second is the lowering operator;
both of them are expected to help you to climb a ladder. Why? Because of
the commutators
[Lz , L ] =
hL
where all the signs are the same. Its derived simply from
[Lz , Lx iLy ] = i
h(Ly iLx )
3

and from a careful verification of the signs. You may still ask why the words
raising and lowering. Imagine that you have an eigenstate of Lz :
Lz = m
h
where we called the eigenvalue m
h and you act on with the commutator
[Lz , L ]. What will you get?
[Lz , L ] =
hL
If we expand the commutator, use the eigenvalue relation, and exchange two
terms, we get
Lz L L m
h = L

i.e. Lz L = m
hL L

But the last equation says that L is an eigenstate of Lz whose eigenvalue


is (m 1)
h. In other words, the number m determining the eigenvalue
either jumped by one or decreased by one in the case of L+ (raising) or L
(lowering).

Climbing the ladder


We can now create a lot of new eigenstates of Lz using the ladder technique
which is very analogous to the harmonic oscillator. We said that the Hilbert
space can be written in terms of the eigenstates of L2 and Lz . Start with
one of them and act with L+ many times to get (L+ )N . Obviously, if
N is too large, the eigenvalue of Lz will become so large that its square L2z
will exceed the pre-determined eigenvalue of L2 which cant happen because
L2x + L2y is a positive operator; note, for example, that the expectation value
of L2x is always positive because it is the squared norm of Lx |i.
The only loophole is that (L+ )N will be zero for all N N0 since vanishing vectors are (completely uninteresting) eigenvectors of anything with any
0 1
eigenvalue: 0 = 0. Take the last nonzero state in the sequence, = LN
.
+
It satisfies
L+ = 0.
Here, has the maximal allowed eigenvalue of Lz . It is now useful to write
L2 using L , Lz :
1
L2 = (L+ L + L L+ ) + L2z = L+ L + L2z h
Lz = L L+ + L2z + h
Lz .
2
4

~ (symThe first form would be naively true even for ordinary numbers L
metrized product of L+ and L often behaves classically). For operators,
the ordering matters. If you choose a particular ordering, new terms proportional to h
appear because of the nonzero commutators
[L+ , L ] = 2
hLz
which can be easily obtained from the definition of L and the known commutators of Lx , Ly . If you are annoyed by h
everywhere, you can set it
equal to one, much like the adult physicists. I cant do it because this is an
undergrad course.
At any rate, if you use the last form for L2 and act on
L2 = (L L+ + L2z + h
Lz ) = . . . ,
you easily see that the first term vanishes because L+ = 0 and you obviously
get
. . . = m(m + 1)
h2 .
In plain English, the eigenvalue of L2 is h
2 times m(m + 1) where m is
the maximal allowed eigenvalue of Lz that you can achieve by the raising
operators. We call this maximum allowed value l = mmax which means
L2 = l(l + 1)
h2 .
Note that when you now take , the highest eigenstate of Lz the top rung
of a ladder you may obtain other rungs by climbing down as
LK

Their eigenvalues of Lz will be (m K)


h and the eigenvalue of L2 will still
be l(l + 1)
h2 because
[L2 , L ] = 0
i.e. because the action of L does not change the eigenvalue of L2 . By
very similar arguments as above, the states LK
must become zero for large
enough K = (l m) because the eigenvalue of Lz cant be too negative
either. The last nonzero state, the bottom rung 0 whose Lz eigenvalue is
m0 , must be annihilated1 by L by the mirror of the arguments above. Using
1

A state is annihilated by Q if Q = 0.

the previous formula for L2 , its eigenvalue of L2 is m0 (m0 1)


h2 but we know
2
that it must still be l(l + 1)
h:
m0 (m0 1) = l(l + 1)
which means that either m0 = l + 1 (which is nonsense because we couldnt
get higher by lowering operators) or, more likely,
m0 = l.
This means that among all the states you can achieve from a given state
by the raising and lowering operator, the eigenvalue of Lz goes between
l
h and +l
h with spacing h
. The difference between the top rung and the
bottom rung is 2l
h which must be a multiple of the rung height h
which
implies that l is either integer or half-integer (integer plus one half). In
reality, only integer values of l can be obtained from orbital momentum, as
discussed under spherical harmonics, but the internal angular momentum
of particles may also be half-integer (e.g. the electron has j = 1/2 where j
is an internal version of l) as we will discuss in 2 pages.

Spherical harmonics
You should be impressed by the previous section in which the spectrum of the
angular momentum was obtained without solving any differential equations
just by pure commutators and algebra. (Incidentally, one can even solve the
spectrum of the Hydrogen atom, including the radial part, by such tricks.)
But sometimes we need to solve the differential equations anyway, especially if we want the exact eigenfunctions for spherically symmetric potentials.
In these cases, we often use the spherical coordinates (r, , ). The Laplacian
(in the kinetic energy) may be divided to the radial part and the angular
part as
1 2
1
2 =
r + 2 L2
2
r r
r
where the last term is known as the centrifugal potential whose numerator
L2 may be written in the spherical coordinates as
1

1 2
L =
sin
+
.
sin
sin2 2
2

How do you prove these two equations? First, the nabla operator may be
written in spherical coordinates as
~ = r + 1 + 1

r
r
r sin
now dont denote operators but unit vectors tangent to the
where r, ,
directions where only r, , increase. If you square this nabla operator, you
obtain the expression for the Laplacian with L2 substituted from the other
equation. An explicit way to prove the formula for L2 in spherical coordinates
is to first derive (for example by an explicit translation of ~r p~ to spherical
coordinates) that
!

~ = i
L
h
.

sin
Note that it does not contain any derivatives with respect to r (because
all components of angular momentum are differentiating with respect to directions induced by rotations around an axis that goes through the origin).
Squaring the previous formula again leads to the formula above for L2 .
When you solved the Hydrogen atom, you already used the eigenfunctions
of L2 and Lz whose eigenvalues are l(l + 1)
h2 and m
h, respectively. For any
l = 0, 1, 2, . . . and m = l, l + 1, . . . + l 1, l we will call the eigenfunction
Ylm a spherical harmonic. Because in the spherical coordinates
Lz = i
h

the dependence on must be purely exp(im) because the -derivative picks


the right factor. The number m must be an integer to make the exponential
single-valued in the real space (periodic with period 2); this also means that
l (maximal m) must be integer.
The dependence on was easy. On the other hand, the dependence on
is affected both by l and m: Ylm / exp(im) may really be written as the
so-called associated Legendre polynomial of cos() (times sinl to make the
differential equation well-behaved around the poles).
It is still true that Yl,l must be annihilated by L+ and so forth, which
simplifies the task to a first order ordinary differential equation. The other
functions may be obtained as Llm
Yl,l (times a constant) as long as you are
able to spend a few minutes with Griffiths (or alone) and derive that in
7

spherical coordinates,
i

L =
he

i cot

So what will we get from the requirement that L+ annihilates Yll ? Assuming
Lz = m, we obtain its full form (up to the normalization):
1
Yll =
2l l!

(2l + 1)! l il
sin e .
4

You should remember that the normalization of Ylm is not determined and
may depend on l, m. Almost everyone agrees that the normalization of its
norm must be such that the functions form an orthonormal set of functions
on the sphere
Z
d2 Yl0 m0 (, )Ylm (, ) = l,l0 m,m0
so
that theR norm of each Ylm is one, for example. (Remember that d2 =
R
2
0 sin d 0 d and the Kronecker delta is ij = 1 for i = j and zero
otherwise.)
How do these spherical harmonics look like? You should remember that
R

1
Y00 = .
4
With no angular momentum, the function cannot depend on angles. The
integral of its mod squared must equal one which fixes the normalization.
And the phase is a pure convention and every good convention makes Y00
positive and real. What about l = 1? First of all, you should remember that
on the unit sphere where we use Cartesian coordinates x, y, z with x2 + y 2 +
z 2 = 1, the spherical harmonics Y11 , Y10 , Y1,1 are proportional to x + iy, z,
x iy. In terms of angles,
Y1,1

3
sin ei ,
=
4

Y10 =

3
cos .
4

Remember from the discussion of the Hydrogen atom that the most typical
convention for the phases implies that
Yl,m = (1)m Ylm .
You are invited to study Ylm for l 2, too.
8

Internal spin
I hope that you liked the algebra the way how the ladder operators created new states from old states. These ladder operators may be written as
matrices with respect to the basis vectors which are eigenvalues of L2 and
Lz . Think about the states with a fixed L2 because theyre not mixed with
~ Well,
states with other values of l (because L2 commutes with L).
Lz = h
diag(+l, +l 1, . . . , l + 1, l)
Also, L+ has nonzero entries just above the diagonal and L has nonzero
entries just below the diagonal. In fact, the precise matrix elements for the
orthonormal basis may be derived from the algebra, too. These matrices will
still satisfy the relations like
[Lx , Ly ] = i
hLz
where the products are defined as matrix multiplication. Imagine that you
dont care about the origin of the angular momentum of a particle H (imagine
the Hydrogen atom). It may come from the orbital angular momentum of
two smaller particles inside H and we use the word spin and the letter J~
in all mutations for this internal angular momentum. You can treat H as a
single particle and describe its internal angular momentum using a label l, m
and the operators that affect these internal numbers as matrices.

Pauli matrices
A funny thing is that there also exist matrices where l, m may be half-integer
but they still satisfy the commutators above. In that case, we call them j, m
instead of l, m. The simplest and most important case describing electrons
~
and other elementary fermions is the case of J = 1/2. The matrices L
that we prefer to call J~ because they are internal spin are the so-called
Pauli matrices:
h

J~ = ~ ,
2

x =

0 +1
+1
0

y =

0 i
+i 0

z =

+1
0
0 1

You should definitely check that [Jx , Jy ] = i


hJz just like before and
J 2 = Jx2 + Jy2 + Jz2 = j(j + 1)
h2 = 0.5 1.5
h2 = 0.75
h2
9

times the identity matrix, in a full agreement with the rules we derived for
the orbital angular momentum but with a half-integer m. Note that this
half-integrality means that if we imagine a (non-existent) function of , it
would have to be antiperiodic with antiperiod 2 which means
( + 2) = ().
The electrons wavefunction changes the sign if you rotate the electron by
360 degrees. We have already used the existence of the spin of the electron
in the past. There are two linearly independent vectors,
up =

+1
0

down =

0
+1

and their Jz eigenvalues are clearly m = +1/2 and m = 1/2, going between
j and +j as always. The most general wavefunction (a, b)T determines the
probability for the spin to be up as |a|2 and the spin to be down as |b|2 and
it is usually normalized to give a unit total probability e.g.
|a|2 + |b|2 = 1.
Non-relativistic physics does not depend on the spin much. It only controls
the number of electrons that we can fit into some particular states. However,
if you include the magnetic fields, you should add the following term to the
Hamiltonian:
= ~ B
~
H
~ is called magnetic moment and
where the coefficient ~ that multiplies B
it is
~
~ = S
~ was chosen to label the old good J~ to make the text
where the letter S
even more refreshing. The universal constant is called the gyromagnetic
ratio. Its value for a naive charged object is = q/2m. But because of
relativity, it turns out that for electrons, it is twice as large, = q/m,
according to the Dirac equation that replaces the Schrodinger equation in
relativity. In fact, quantum electrodynamics (quantum mechanics combined
with Maxwells theory and relativity, the first quantum field theory that
anyone studied) gives extra corrections of order 0.1% to . The full may
be calculated theoretically and it agrees with experiments with the accuracy
of 13 decimal points. This anomalous magnetic moment of the electron
remains the most accurately verified prediction in the history of science.
10

Adding angular momentum


The angular momentum is additive. Imagine that you have two contributions
~ J~ = J~1 +J~2 . The two contributions may be
to the total angular momentum J:
the orbital angular momenta of two electrons, or the orbital momentum and
a spin, or two spins, and so forth. We will be talking about two particles
for the sake of clarity. Imagine that we know that the eigenvalue of J12 is
j1 (j1 + 1)
h2 . Also imagine that you allow me to omit h
that you can always
add by dimensional analysis. And the eigenvalue of J22 is j2 (j2 + 1). The spin
states of the two particles are labeled by
(j1 , m1 , j2 , m2 ).
One should also realize that if J~1 satisfies the usual commutators like
[J1x , J1y ] = iJ1z
and so on and J~2 satisfies analogous relations, then also
[Jx , Jy ] = iJz
and so forth. So the states of the two particles can also be labeled by the
eigenvalues of J 2 and Jz . What are the allowed values?
J 2 describes the squared length of the vector. Classically, the length of
the vector ~a + ~b can be anything between the sum of their lengths and the
absolute value of their difference, depending on the mutual orientation. An
analogous statement holds in quantum mechanics. The quantum number j
(the total spin) may be between
j = |j1 j2 |, |j1 j2 | + 1, . . . , j1 + j2
Note that it is not true that the total spin must be j1 + j2 ; it can be smaller.
Lets count the states. There are 2j1 + 1 states of the first particle and 2j2 + 1
states of the second particle. It turns out that the total number of states
(2j1 + 1) (2j2 + 1) =

j1X
+j2

(2j + 1)

j=|j1 j2 |

may also be counted from all states of the total system whose j, m belong to
the usual intervals. The sum above can be proved following Gauss; when he
was a kid, his teacher wanted to reduce Gauss activity in the class and asked
Gauss to sum up the numbers from 1 to 100. In a minute, Gauss announced
it was 5,050. How did he do it?
11

Clebsch-Gordan coefficients
At any rate, you see that there is the same number of independent states in
each description, and therefore there must exist a relation between the two
bases. The matrix elements translating these two bases into each other are
called the Clebsch-Gordan coefficients which are the inner products h. . .i at
the end of this completeness relation:
|j, m, j1 , j2 i =

m1 +m
X2 =m
m1 ,m2

|j1 , m1 , j2 , m2 ihj1 , m1 , j2 , m2 |j, m, j1 , j2 i

Note that if we know m, we only know m1 + m2 and not m1 , m2 separately.


And on the contrary, if we know m1 , m2 , we only know m = m1 + m2 and
not j. So there is some mixing going on. To derive the CG coefficients, start
with the doubly-maximal m1 = j1 , m2 = j2 two-particle state which must
satisfy
|j, ji = |j1 , j1 i |j2 , j2 i
and therefore j, m must also be maximal, j = j1 + j2 , m = j. We will omit
the tensor product sign because it may distract you. Apply J = J1 +J2
on this equation to get
q

j(j + 1) j(j 1)|j, j 1i =


+

j1 (j1 + 1) j1 (j1 1)|j1 , j1 1i|j2 , j2 i

j2 (j2 + 1) j2 (j2 1)|j1 , j1 i|j2 , j2 1i

When we apply J repeatedly, we may write all |j, mi for the maximal j =
l1 + l2 as combinations of the two-particle states. What about the states
with smaller values of the total j? We may first start with the submaximal
|j 1, j 1i that satisfies
J+ |j 1, j 1i = 0.
That gives us enough conditions to express it in terms of the two-particle
states. Alternatively, we may impose the orthogonality of |j 1, j 1i to
|j, j 1i which is also enough. In both cases, we may again apply J many
times. This procedure can be repeated to obtain all CG coefficients for a
given pair j1 , j2 , and these zoos of numbers are printed in many books.

12

You might also like