14.1
We are going to be adding angular momenta in a variety of ways. We may add the spin
angular momentum S of a particle to its orbital angular momentum L. Or we may want to
add the spin angular momentum S(1) of a particle to the spin angular momentum S(2) of
another particle. At first sight we may feel like we are trying to add apples to oranges! For
a given particle its spin angular momentum has nothing to do with spatial wavefunctions,
while its orbital angular momentum does. How could we ever add such things? Adding
the spins of two different particles also seems unusual if, for example, the particles are faraway from each other. Vectors that live at different places are seldom added: you dont
typically add the electric field at one point to the electric field at another point: the sum
has no obvious interpretation. This is even more severe in general relativity: you cannot
309
310
add vectors that live at different points of spacetime. To add them you need a procedure
to first bring them to a common point. Once they both live at that common point you can
add them.
I want to make clear, however, that at a basic algebraic level all angular momenta are
apples (granny smith, reddelicious, macintosh, fuji, etc.) and therefore they can be added
and it is natural to add them. We are not adding apples to oranges; we are adding apples to
apples! The physics requires it: we will see that in a class of Hamiltonians energy eigenstates
will be eigenstates of the sum of angular momenta. The mathematics allows it: the sum of
angular momenta is an angular momentum acting in the appropriate tensor product. As
we will see below, while each angular momentum lives on a different vector space, the sum
finds a home in the tensor product of the vector spaces.
What is an angular momentum? It is a triplet Ji of Hermitian linear operators on some
complex vector space V satisfying the commutation relations
[Ji , Jj ] = i~ ijk Jk .
(14.1.1)
As we have learned, this is a very powerful statement. When coupled with the requirement
that no negative normsquared states exist, it implies that the vector space V on which
these operators act can be decomposed into sums of finite dimensional subspaces that carry
irreducible representations of angular momentum.
Let us now assume we have two angular momenta:
(1)
on V1 satisfying [Ji
(2)
on V2 satisfying [Ji
Hermitian operators Ji
Hermitian operators Ji
(1)
(1)
(1)
, Jj ] = i~ ijk Jk ,
(2)
(2)
(2)
, Jj ] = i~ ijk Jk .
(14.1.2)
Our claim is that the sum of angular momenta is an angular momentum in the tensor
product:
(1)
(2)
Ji Ji 1 + 1 Ji satisfy, [Ji , Jj ] = i~ ijk Jk acting on V1 V2 .
(14.1.3)
(1)
(2)
(1)
(2)
Ji 1 + 1 Ji , Jj 1 + 1 Jj
(1)
(1)
(2)
(2)
= Ji 1 , Jj 1 + 1 Ji , 1 Jj
,
[Ji , Jj ] =
(14.1.4)
since the mixed terms, which represent commutators of the operators in the different spaces
vanish:
(1)
(2)
(2)
(1)
Ji 1 , 1 Jj ] = 0 ,
1 Ji , Jj 1 ] = 0 .
(14.1.5)
311
(If this is not obvious, think about it and do the oneline computation that will make it
obvious!). Writing out the commutators we see that (14.1.4) becomes
(1) (1)
(2) (2)
[Ji , Jj ] = Ji , Jj
1 + 1 Ji , Jj
.
(14.1.6)
(14.1.7)
= i~ ijk Jk ,
(14.1.8)
If either constant is zero we are not really summing. The commutator calculation above
this time yields
(2)
(1)
(14.1.9)
[Ji , Jj ] = i~ ijk 2 Jk 1 + 2 1 Jk .
14.2
To set up the notation recall that for a spin onehalf particle and spin operators S we write
S2 s, mi = ~2 s(s + 1)s, mi ,
Sz s, mi = ~m s, mi ,
with s =
1
2
, m = 21 . (14.2.1)
 21 , 12 i ,
(14.2.2)
states that we used to label as +i and i, respectively. The action of S2 on any of these
states gives 34 ~2 and the action of Sz /~ gives 21 on the first and 21 on the second.
312
We now consider the case in which our system features two spin onehalf particles. For
the first particle we have the triplet of spin operators S(1) acting on the vector space V1
spanned by
(14.2.3)
 12 , 12 i1 ,  12 , 21 i1 ,
For the second particle we have the triplet spin operators S(2) acting on the vector space
V2 spanned by
 12 , 12 i2 ,  12 , 21 i2 .
(14.2.4)
We now form the total spin
(1)
(2)
Si Si 1 + 1 Si ,
(14.2.5)
(1)
(2)
Si = Si + Si , for example, Sz = Sz(1) + Sz(2) ,
(14.2.6)
with the understanding that each operator on the righthand sides acts on the appropriate
factor in the tensor product. The state space for the dynamics of the two particles must
contain the tensor product V1 V2 (more spaces might be needed if the particles have orbital
angular momentum or they are moving). As we learned before, V1 V2 is a fourdimensional
complex vector space spanned by the products of states in (14.2.3) and (14.2.4):
 12 , 12 i1  21 , 12 i2 ,
 12 , 12 i1  12 , 21 i2 ,
 12 , 12 i1  21 , 12 i2 ,
 12 , 12 i1  21 , 12 i2 .
(14.2.7)
Since the total spin operators have a well defined action on these states and the vector
space they span, it must be possible to describe the vector space as a sum of subspaces
that carry irreducible representations of the total spin angular momentum. The irreducible
representations of an angular momentum J are those characterized by the number j that
appears in the eigenvalue of J2 . A representation is said to be reducible if it is the direct
sum of irreducible representations.
We have four basis states so the possibilities for multiplets of total spin s are
1. Four singlets (s = 0).
2. Two doublets (s = 12 ).
3. One doublet (s = 12 ) and two singlets (s = 0).
4. One triplet (s = 1) and one singlet (s = 0).
5. One s =
3
2
multiplet.
Only the last option is an irreducible representation. All others are reducible. It may
be instructive at this point if you pause to make sure no other option exists and then to
consider which option you think is the correct one! Guess it!
313
The main clue is that the states in the tensor product are eigenstates of Sz , the total
zcomponent of angular momentum. We see by inspection of (14.2.7) that the possible
values of Sz /~ and +1, 0, and 1. Since we have a state with m = 1 and no state with
higher m we must have a triplet s = 1. Thus the only option is the fourth one listed above:
a triplet and a singlet. This is written as
(s = 21 ) (s = 12 ) = (s = 1) (s = 0) .
(14.2.8)
Note that in the lefthand side we have the tensor product of the two state spaces, but in
the righthand side the direct sum of the representations of total spin angular momentum.
This is a fundamental result and is written more briefly as
1
2
1
2
= 1 0.
(14.2.9)
We use bold type for the numbers to make clear that these represent vector spaces.
Let us understand this very explicitly by organizing the basis states according to the eigenvalue m of Sz /~. We readily observe that
m =
1:
 21 , 12 i1  12 , 12 i2 ,
m =
0:
 12 , 12 i1  12 , 21 i2 ,
m = 1 :
 12 , 12 i1  12 , 21 i2 ,
(14.2.10)
 12 , 12 i1  12 , 21 i2 .
We see that we get two states with m = 0. This is as it should be. One linear combination
of these two states must correspond to the m = 0 state of the triplet and another linear
combination must correspond to the singlet s = m = 0. Those two states are in fact
entangled states. Denoting by s, mi the eigenstates of S2 and Sz (total spin) we must have
a triplet with states
1, 1i =  21 , 12 i1  12 , 12 i2 ,
1, 0i =  12 , 21 i1  12 , 12 i2 +  21 , 12 i1  12 , 12 i2 ,
(14.2.11)
1, 1i =  21 , 12 i1  12 , 21 i2 ,
for some constants and , as well as a singlet
0, 0i =  12 , 21 i1  12 , 12 i2 +  12 , 21 i1  12 , 12 i2 ,
(14.2.12)
for some constants and . We must determine these four constants. Let us begin with
the state in the triplet. Recalling the general formula
p
J j, mi = ~ j(j + 1) m(m 1) j, m 1i ,
(14.2.13)
314
J 1, 1i = ~ 2 1, 0i ,
q
J  12 , 21 i = ~ 12 32 21 ( 12 )  12 , 21 i = ~  12 , 12 i ,
q
J+  12 , 21 i = ~ 12 32 ( 21 ) ( 12 )  12 , 21 i = ~  12 , 12 i .
(14.2.14)
(1)
(2)
(2)
S 1, 1i = (S  12 , 12 i1 )  12 , 21 i2 +  12 , 12 i1 (S  12 , 21 i2 ) ,
Using the results in (14.2.14) we find
2~ 1, 0i = ~ 12 , 21 i1  12 , 12 i2 +  12 , 21 i1 ~  21 , 12 i2 .
(14.2.15)
(14.2.16)
Cancelling the common factors of ~ and switching the order of the terms we find that the
1, 0i state takes the form
(14.2.17)
1, 0i = 12  12 , 12 i1  21 , 12 i2 +  12 , 21 i1  12 , 12 i2 .
Having found the m = 0 state of the s = 1 multiplet there are a number of ways to find the
m = 0 of the s = 0 singlet. One way is orthogonality: the latter state must be orthogonal
to the m = 0 state above because these are two states with different eigenvalue s of the
Hermitian operator S2 . Since the overall sign or phase is irrelevant, we can simply take for
the singlet
(14.2.18)
0, 0i = 12  12 , 12 i1  21 , 12 i2  12 , 21 i1  12 , 12 i2 .
You probably remember that we found this state in a previous chapter by requiring that it is
annihilated by the sum of spin angular momentum operators. This is exactly the condition
for a singlet.
As an instructive calculation let us confirm that S2 is zero acting on 0, 0i. For this it
is useful to note that
S2 = (S(1) + S(2) )2 = (S(1) )2 + (S(2) )2 + 2S(1) S(2)
(1) (2)
(1) (2)
(14.2.19)
1
2
(1) (2)
(1) (2)
J+ J + J J+ + Jz(1) Jz(2) ,
(14.2.20)
valid for arbitrary angular momenta. Written in explicit tensor notation it reads
(1) J
(2)
J
3
X
i=1
(2)
(1)
Ji Ji =
1
2
(2)
(1)
(2)
(1)
J+ J + J J+ + Jz(1) Jz(2) .
(14.2.21)
315
Back to our calculation, all states have s1 = s2 = 21 , and therefore (S(1) )2 = (S(2) )2 = 34 ~2 .
We thus have
(1) (2)
(1) (2)
(14.2.22)
S2 0, 0i = 23 ~2 0, 0i + S+ S + S S+ + 2Sz(1) Sz(2) 0, 0i .
It is simple to see that
(14.2.23)
because the singlet is a superposition of tensor states where each has one state up and one
state down. Similarly recalling that
S  21 , 12 i = ~  12 , 12 i ,
we quickly find that
(1) (2)
(1) (2)
S+ S + S S+ 0, 0i = ~2 0, 0i ,
(1) (2)
(14.2.24)
(14.2.25)
(1) (2)
since each of the operators S+ S and S S+ kills one term in the singlet and acting on
the other term it gives ~2 times the killed one. Check it! Back in (14.2.26) we get
(14.2.26)
S2 0, 0i = 23 ~2 0, 0i + ~2 21 ~2 0, 0i = 0 ,
as we wanted to show.
Let us summarize our results. The triplet states and singlet states are given by
 12 , 12 i1  12 , 12 i2 ,
1, 0i = 12  12 , 12 i1  21 , 12 i2 +  12 , 21 i1  12 , 12 i2 ,
1, 1i =
1, 1i =
0, 0i =
 12 , 12 i1  12 , 12 i2 .
1
2
(14.2.27)
 12 , 12 i1  21 , 12 i2  12 , 21 i1  12 , 12 i2 .
 12 , 12 i  i ,
and  12 , 21 i  i .
(14.2.28)
We then have
1, 1i =  i1  i2 ,
1, 0i = 12  i1  i2 +  i1  i2 ,
1, 1i =  i1  i2 .
0, 0i =
1
2
 i1  i2  i1  i2 .
(14.2.29)
316
With the understanding that the first arrow refers to the first particle and the second arrow
to the second particle, we can finally write all of this quite briefly
1, 1i =
1, 0i =
1, 1i =
0, 0i =
 i ,
1
2
 i +  i ,
(14.2.30)
 i .
1
2
 i  i .
This decomposition of the tensor product of two spin onehalf state spaces is useful for the
computation of the hyperfine splitting in the hydrogen atom, where the relevant spins are
those of the proton and the electron.
14.3
FeynmanHellman lemma
(14.3.1)
This is the FeynmanHellman (FH) lemma. Note that by assumption n () is an eigenstate and is normalized for all values of . The FH lemma tells relates the instantaneous
change of energy of an eigenstate to the expectation value of the derivative of the Hamiltonian evaluated at the instantaneous eigenstate. The proof of (14.3.1) is straightforward:
Proof. The statement that n ()i is an energy eigenstate of H() with eigenvalue E()
means that
H()n ()i = En ()n ()i .
(14.3.2)
Forming the inner product with the bra hn () and recalling that the state is normalized
gives
En () = hn ()H()n ()i .
(14.3.3)
317
(14.3.4)
(14.3.5)
We thus have
d
dH()
dEn
() = En () hn ()n ()i + hn ()
n ()i .
d
d
d
(14.3.6)
Since the state is normalized for all the first term vanishes and we are left with the result
we wanted to prove. Note that the FH lemma is an exact result.
In many problems plays the role of a deformation parameter. The Hamiltonian H( =
0) may be simple enough that all its eigenstates and eigenvalues are know. The problem then
becomes to calculate the deformed eigenvectors n ()i and eigenvalues En () as 6= 0.
The FH lemma relates these quantities but does not give them to us. We will show now
how to use FH lemma to formulate an approximate result for En () for small, in terms
of known quantities about the = 0 Hamiltonian.
For this purpose consider a Hamiltonian where the dependence is linear:
H() = H(0) + H1 = H0 + H .
(14.3.7)
(14.3.8)
(14.3.9)
En () = En (0) +
dEn ()
+O(2 ) ,
d =0
(14.3.10)
318
which we write as
En () = En (0) + hn (0)Hn (0)i + O(H 2 ) .
(14.3.11)
We can see that appears on the lefthand side, but given our notation it does not appear
on the righthand side. So let us change notation a bit so that is appears nowhere. Let us
use H for H() so that the Hamiltonian is now
H = H0 + H .
(14.3.12)
Instead of using En () we will call it En , which is naturally identified as the energy eigenvalue for the full Hamiltonian H . Additionally, En (0), which is the energy for = 0, is
(0)
(0)
now written as En , the energy associated with H0 . Similarly we rewrite n (0)i as n i,
so that (14.3.11) becomes
En = En(0) + hn(0) Hn(0) i + O(H 2 ) .
(14.3.13)
(0)
The second term on the righthand side is the energy shift for the eigenstate n i of the
original Hamiltonian H0 as we add the perturbation H. This is the simplest and most
important result in perturbation theory. The energy shift of a state is simply given by the
expectation value of the change H of the Hamiltonian in the unperturbed state. You do
not need to know how the state changes due to the perturbation to compute how its energy
changes to first order!
Example 1: Let H0 and H be given by
1 0
H0 =
,
0 2
H =
(14.3.14)
where all parameters , , in H are small. We want to find the energies to first order in
the perturbation. Since the original Hamiltonian H0 is diagonal, the two energy eigenstates
and the two eigenvalues are written by inspection
1
(0)
(0)
1 = 1 ,
1 =
,
0
(14.3.15)
0
(0)
(0)
.
2 = 2 ,
2 =
1
Note the subscripts 1 and 2 for the first and second states, as well as the superscript (0) that
refers to the unperturbed Hamiltonian H0 . The results to first order in the perturbation
are
(0)
(0)
(0)
(0)
(0)
(0)
1 = 1 + h1 H1 i ,
2 = 2 + h2 H2 i .
(14.3.16)
319
= 1 + 1,0
= 2 + 0,1
1
= 1+,
0
0
= 2+.
(14.3.17)
Note that to this order the parameter does not enter in the answer. In this simple case
you can quickly calculate the exact eigenvalues, and you should check that to linear order
in the small parameters you reproduce the above results.
The above Hamiltonian H0 has a nondegenerate spectrum: the two eigenvalues are
different and the eigenstates are unique up to normalization. We used the FH result (14.3.13)
for each of the eigenstates; we had n = 1 and n = 2! The calculation of perturbed energies is
more subtle if the Hamiltonian H0 has a degenerate spectrum. Let us consider an example
to see what is the difficulty!
Example 2: Let H0 and H be given by
1 0
H0 =
,
0 1
H =
0
,
0
(14.3.18)
where the parameter in H is small. We want to find the energies to first order in
the perturbation. We could again simply diagonalize the matrix H0 + H, but that is
not the point. What we want to learn is how to use properly the FH lemma to compute
the perturbed energies. This time, the eigenvalues of H0 are both equal to one and the
eigenstates can be chosen to be
1
(0)
(0)
1 = 1 ,
1 =
,
0
(14.3.19)
0
(0)
(0)
2 = 2 ,
2 =
.
1
If we simply applied the rule as before we would get that the eigenvalues are not changed:
0
1
1 = 1 + 1 , 0
= 1,
0
0
(14.3.20)
0
0
2 = 1 + 0 , 1
= 2.
0
1
This is just plain wrong! The eigenvalues are changed. We could have anticipated the
problem. The choice of two eigenvectors above was totally arbitrary. Given the degeneracy,
any superposition of these two is also an eigenvector. We can in fact parameterize the
general normalized eigenvector using an angle as follows
!
cos
(0)
() =
.
(14.3.21)
sin
320
The choice of vector (or ) clearly affects the FH result. For a given the perturbed
eigenvalue would be
!
cos
0
= 1 + cos , sin
= 1 + sin 2 .
(14.3.22)
0
sin
This makes clear our problem: the energy shift depends on the value that parametrizes
the eigenstates. We must find a way to choose the two values of that give the two states
that we must use in the FH lemma to compute the energy shifts.
To solve this problem we note that in the derivation of the FH lemma a key assumption
was our ability to track the eigenstates continuously as the parameter was turned on.
In our present example the choice of eigenvectors for H0 , which the = 0 Hamiltonian, is
ambiguous. Figure 14.1 aims to represent the evolution of two eigenstates as grows from
zero to finite values. At = 0, because of the degeneracy, all unit vectors in that shaded
plane are normalized eigenvectors of H0 and there seem to be no special ones. As soon as
the perturbation is turned on, however, the degeneracy is broken and the eigenstates are
uniquely determined. It follows that to track continuously the eigenstates we must choose
(0)
(0)
for 1 and for 2 the limits of the nonzero eigenstates as goes to zero.
In our example we must identify the eigenstates of the perturbation H as 0. In
fact in this case the eigenstates are independent
1
1 1
1
0
(0)
(0)
,
2 =
.
(14.3.23)
eigenvectors: 1 =
1
1
0
2
2
These are the states that must be used for the FH lemma. They correspond to (0) () for
1 = 4 and 2 = 4 . We then use our earlier result (14.3.22) to find
1 = 1 + sin 21 = 1 + ,
and 2 = 1 + sin 22 = 1 .
(14.3.24)
14.4
Hyperfine splitting
As our first physical application we will consider hyperfine splitting. For this example the
construction of eigenstates of total angular momentum is simple and was considered before.
321
Figure 14.1: Evolution of the eigenvectors as grows from zero at the bottom of the figure. The
Hamiltonian H( = 0) defines a space of degenerate eigenvectors. In the figure this subspace is
represented by the twodimensional shaded disk. Any vector from the origin to the circle is an
eigenvector with the same eigenvalue. Nevertheless the perturbation of the Hamiltonian breaks the
degeneracy and for any 6= 0 the two eigenvectors 1 () and 2 () are uniquely determined. In
(0)
(0)
the limit as 0 they fix the two vectors 1 and 2 that must be used in the FH lemma.
We will discuss here some of the general features of addition of angular momentum, leaving
the study of complete sets of observables to our second example, the case of spinorbit
coupling.
The simple hydrogen atom Hamiltonian we have used so far receives a small correction
because both the proton and the electron have magnetic dipole moments. The proton
magnetic dipole creates a magnetic field. The contribution to the Hamiltonian is the energy
of the electron dipole in the magnetic field created by the proton. Recall that that for the
proton and the electron we have:
e
gp e
Sp , gp 5.59 ,
e = Se .
(14.4.1)
p =
2mp
me
The extra term H in the Hamiltonian is
H = e Bp =
e
Se Bp ,
me
(14.4.2)
where Bp is the magnetic field due to the proton dipole, at the position of the electron. The
magnetic field of a magnetic dipole contains two contributions: the familiar piece that falls
off like 1/r 3 and a deltafunction contribution at the position of the dipole:
20
1
p (r) .
(14.4.3)
Bp = 3 +
r
3
322
e
20
0 gp e2
Se
p (r) =
Se Sp (r) .
m
3
3mp me
(14.4.4)
In this situation the original Hamiltonian H0 is the hydrogen atom Hamiltonian that does
not depend on the spin degrees of freedom of the electron nor on those of the proton. Since
each of them can be in either of two states, up or down, we have four degenerate basis states
that share a common spatial wavefunction, the ground state 1,0,0 , but differ on the spin
degrees of freedom:
1,0,0 (r)  i ,
1,0,0 (r)  i ,
1,0,0 (r)  i ,
(14.4.5)
1,0,0 (r)  i .
Before we include the perturbation H these states are all degenerate at energy equal to
the ground state energy of the hydrogen atom. As discussed at the end of section 14.3 we
can get the perturbed energies if we find new basis states, linear combinations of the above
states for which the matrix H is diagonal. In computing the matrix elements of H let
us write the states as 1,0,0 (r)si i, where si i with i = 1, 2, 3, 4, are some yet undetermined
combinations of the spin states for the protonelectron pair. A matrix element Hij then
denotes
Z
Hij =
d3 x1,0,0 (r)2 hsi Hsj i ,
0 gp e2
hsi Se Sp sj i ,
3mp me
0 gp e2 ~2 1
=
hsi Se Sp sj i
3mp me a30 ~2
1
= E 2 hsi Se Sp sj i ,
~
= 1,0,0 (0)2
(14.4.6)
where we recalled that 1,0,0 (0)2 = 1/(a30 ) and introduced the energy scale E relevant
to the hyperfine splitting:
0 e2 gp ~2
E =
.
(14.4.7)
3mp me a30
Our earlier work adding two spin onehalf angular momenta is relevant now that we
need to understand the operator product Se Sp . More precisely, this product is a tensor
323
Sei Spi .
(14.4.8)
(14.4.9)
As we have seen, the Si operators satisfy the algebra of angular momentum. It is now a
simple computation to expand S2 :
X
X
S2 =
Si Si =
Sei 1 + 1 Spi Sei 1 + 1 Spi
i
X
i
X
i
S2e
Sei Sei 1 + 2
1+2
X
i
X
i
Sei Spi + 1
Sei Spi + 1
S2p .
(14.4.10)
Spi Spi
(14.4.11)
(14.4.12)
1
2
S2 S2e S2p .
(14.4.13)
This rewriting shows the relevance of the total spin angular momentum. We can effectively
trade Se Sp for S2 because the other operators S2e and S2p on the righthand side are simply
multiples of the identity. Since they are spin onehalf operators S2e = S2p = 34 ~2 . Given this,
the operator whose matrix elements we are considering in (14.4.6) is
1
Se Sp =
~2
1 1 2
2 ~2 S
3
2
1 2
S
~2
3
2
sj i .
As a result, we have
Hij =
1
2
E hsi 
(14.4.14)
(14.4.15)
This matrix is diagonal when the states si i are eigenstates of the total spin S. As we have
seen before the four states in the tensor product of two spin onehalf particles combine into
324
a triplet (s = 1) and a singlet (s = 0). Thus we can take si i with i = 1, 2, 3 to be the
triplet and s4 i to be the singlet
s1 i =  i ,
 i +  i ,
1
2
s2 i =
(14.4.16)
s3 i =  i .
s4 i =
 i  i .
1
2
3
2
=2
3
2
1
2
(14.4.17)
3
2
=0
3
2
= 32 .
0
0
0
0
.
1
0
4 E
0
34 E
(14.4.18)
(14.4.19)
The three states on the triplet are pushed up 14 E and the singlet is pushed down 34 E.
The total split between the two set of states is E, which is now identified as the magnitude
of the hyperfine splitting of the ground state of the hydrogen atom.
Let us determine the value of E. Since we are using SI units we have
a0 =
40 ~2
me e2
0 e2 =
4~2
,
c2 me a0
(14.4.20)
as well as
e2
1 ~
a0 =
.
40 ~c
me c
These help us rewrite E in a comprehensible way:
4gp
4gp ~ 4
( me c)4 ,
=
E =
2
2
3mp me c a0
3mp m2e c2
=
and finally
E =
4 me 4
gp
(me c2 ) .
3
mp
(14.4.21)
(14.4.22)
(14.4.23)
Bohrs energy is roughly 2 1/19, 000 times the rest energy of the electron. The socalled
spinorbit splitting (next section!) is another factor of 2 times the Bohr energy. Finally, the
325
me
hyperfine splitting is roughly m
times the spinorbit splitting, namely, about twothousand
p
times smaller! We quickly plug in the numbers:
E =
1 4
1
4
511, 000 eV 5.88 106 eV .
5.59
3
1836
137
(14.4.24)
c
=
c
E
h
2 197 MeV fm
2~c
=
21.1 cm.
E
5.88 106 eV
(14.4.25)
The corresponding frequency is = 1420MHz. This 21centimeter line was first observed
by Ewen and Purcell in 1951. The splitting is so small that even microwave background
photons (of typical wavelength 2 mm) can easily excite the ground state into the triplet.
The lifetime of the triplet state is extremely long, about 10 million years. As a result, this 21cm line is extraordinarily sharp and very useful for astronomical Doppler shift
measurements. The rotation curve of our galaxy has been calculated using this 21cm line.
14.5
SpinOrbit coupling
14.5.1
Spinorbit coupling refers to a correction to the hydrogen atom Hamiltonian when we take
into account that the dipole moment of the orbiting electron actually feels a magnetic field.
You may ask: Where does this magnetic field come from? There are two ways to think
about it. As seen by the electron in its own instantaneous rest frame, there is a proton
going around the electron, that rotating proton defines a current, and the current creates
a magnetic field. Another way is as follows. The proton creates an electric field, but any
particle moving with some velocity in an electric field E will observe a magnetic field B
given by
v
v
(14.5.1)
B E = E .
c
c
This is a relativistic effect: in the frame of the proton there is only its electric field (ignoring
the electron), but a Lorentz transformation to the moving frame of the electron gives an
extra magnetic field. The formula above is accurate for small velocities.
The interaction of the dipole moment of the electron with this magnetic field is
described by a correction H to the Hamiltonian:
H = B .
(14.5.2)
For the electron dipole we will use the result in Gaussian units, that has an extra c in the
denominator:
e
e
S H =
S B.
(14.5.3)
=
mc
mc
326
Let us now calculate the magnetic field. The electrostatic potential (r) created by the
proton is (r) = e/r and therefore the associated electric field is
E = =
e
r
r3
e
v
e
r
=
L.
3
r
c
mcr 3
(14.5.4)
The magnetic field B at the electron is in fact aligned with the orbital angular momentum
L of the electron. Back to the Hamiltonian correction:
H =
e2
S L.
m2 c2 r 3
(14.5.5)
Unfortunately this derivation is off by a factor of two. The electron rest frame is not an
inertial frame and there is a Thomas correction, a correction to precession rates for objects
that are themselves in a rotating frame. We can view H as a precession Hamiltonian, as
a B interaction causes dipoles to precess in a magnetic field. Thomas correction implies
that the precession rate is half of the one recorded above. The correct form of H is
H =
e2
S L.
2m2 c2 r 3
(14.5.6)
The estimation of the magnitude E of this splitting is easy because we are using Gaussian
units. We estimate S L ~2 and r a0 , to find
e2 ~2
.
m2 c2 a30
(14.5.7)
e8
mc2 = 4 (mc2 ) .
~4 c4
(14.5.8)
E =
Using the simple a0 =
~2
me2
we have
E =
Here 1/137 is the fine structure constant. Since the ground state energy is roughly
2 (mc2 ), the fine structure of the hydrogen atom is an effect roughly 2 1/19, 000 smaller
than the scale of the energy levels!
In the following we will use the standard SI units version of our H in (14.5.6):
H =
e2
1
L S.
2
80 m c2 r 3
(14.5.9)
We reversed the order of the L and S operators, but this has no effect whatsoever. The
operators commute since they act on different degrees of freedom. We order them in this
way because states are usually written with tensor products in which orbital states come
before spin states.
327
14.5.2
We will look into the simplest example. Since we want some nonvanishing angular momentum, so we will take = 1. The lowest energy states with = 1 have principal quantum
number n = 2. The = 1 multiplet has three states, with m = 1, and m = 0. But since
the spin of the electron can be either up or down, we have a total of six states. The states
can be written as
2,1 (r)  = 1, mi s = 21 , ms i ,
with
ms = 12 .
m = 1, 0,
(14.5.10)
h1H1i
0
0
h2H2i
0
(14.5.11)
(H)ij =
.
..
..
..
..
.
.
.
.
0
h6H6i
(14.5.12)
where there is no sum over i. The entries of the above diagonal matrix are the shifts. We
will see that the required basis states are eigenstates of the total angular momentum.1
Let us focus on the L S term in H. As we discussed in the previous section the term
should be thought as a tensor product of operators
X
L S L1 S 1 + L2 S 2 + L3 S 3 =
Li S i .
(14.5.13)
i
(14.5.14)
1
Curiously, the states ii are not eigenstates of H. In fact the subspace spanned by the states in
(14.5.10) is not invariant under the action of H, because this operator carries radial dependence and thus
upon action on states with a given radial dependence it gives states with a different radial dependence. Still
it is true that the matrix of H is diagonal in the subspace we are working in. These same remarks apply
to the case of the hyperfine splitting.
328
The computation of J2 is completely analogous to the computation we did before and gives:
J2 = L2 1 + 2 L S + 1 S2 ,
which leads to the useful
LS =
1
2
J2 L2 S2 .
(14.5.15)
(14.5.16)
1. The uncoupled states (14.5.10) are all eigenstates of L2 and S2 with eigenvalues ~2 (+
1) and ~2 s(s + 1), respectively. Of course for these states = 1 and s = 1/2.
2. Any linear combination of these states are still eigenstates of L2 and S2 . But only
peculiar linear combinations will be eigenstates of J2 .
3. Since we have
[L2 , Li ] = 0 ,
[S2 , Si ] = 0 ,
[Li , Sj ] = 0 ,
(14.5.17)
L S
r3
with =
1
e2
.
80 m2 c2
(14.5.18)
A chosen set of basis states can be written with the r dependence out in the open
 ii = 2,1 (r) i i ,
(14.5.19)
(14.5.20)
Here the subscripts 2, 1 on the expectation value indicate that one is using the n =
2, = 1 radial wavefunction. If the  i i are eigenstates of L S the matrix for H will
be diagonal and the diagonal elements are the energy shifts.
Let us now consider the complete set of commuting observables suitable for the spinorbit
supplemented hydrogen Hamiltonian. For the unperturbed Hydrogen atom Hamiltonian H0
the complete set of observables is
{H0 , L2 , Lz } ,
(14.5.21)
329
Recall that [H0 , Li ] = 0, so all the operators here commute. Suppose we now consider the
spin of electron but do not change the Hamiltonian. In this case the list would be enlarged to
{H0 , L2 , Lz , S2 , Sz } .
(14.5.22)
Note that H0 contains no spin operator, so trivially [H0 , Si ] = 0. Moreover since Li s and
Sj commute, all operators on the list commute. Let us now consider the case when the full
Hamiltonian H is
H = H0 + H , H = 3 L S .
(14.5.23)
r
The list of commuting observables now will have
{H , L2 , S2 , . . .} ,
(14.5.24)
where the dots indicate additional operators we may be able to add. Notice that so far all
is good, since both L2 and S2 commute with H and thus with H . This is because H
is built from Li and Si operators that commute with L2 and S2 . H also contains r but
that operator commutes with any Li and commutes trivially with any Si . Neither Li nor
Si , for any value of i, can be added to the list above because
[ Li , L S ] 6= 0 ,
[ Si , L S ] 6= 0 ,
(14.5.25)
1
2
J2 L 2 S 2 .
(14.5.26)
(14.5.27)
The last two lines imply that J2 commutes with H and therefore it commutes with H .
Yes, we can add J2 and the new list is
{H , L2 , S2 , J2 , . . .} ,
(14.5.28)
We can try now to add some Ji to the list. You can see that Ji commutes with H0 , L2 , S2 ,
and J2 , the last one by the Casimir property. Moreover
[Ji , L S] =
Ji , 12 J2 L2 S2
= 0.
(14.5.29)
330
Since [Ji , f (r)] = 0 by the same arguments as before, we can add one Ji to the list. By
convention, we add Jz to find our final list!
CSCO for SpinOrbit coupling:
{H , L2 , S2 , J2 , Jz } .
(14.5.30)
We will confirm that all eigenstates can be labelled uniquely by eigenvalues of the above
operators.
14.5.3
Computation of 1 12
We have our six states in the uncoupled basis. Ignoring the radial dependence which is not
relevant in this section, the states are:
, mi s, ms i ,
= 1,
s=
1
2
(14.5.31)
and exist for all combinations of m = 1, 0, 1 and ms = 12 Our goal in this section is to
determined the coupled states, the states that are eigenstates of J2 and Jz where J = L + S.
We note that even the uncoupled states are eigenstates of Jz = Lz + Sz
Jz , mi s, ms i = (Lz , mi) s, ms i + , mi Sz s, ms i
= ~m , mi s, ms i + ~ms , mi s, ms i
(14.5.32)
1
~
1
~
1
~
Jz
3
2
1
~
Jz
1
2
Jz = 21 :
Jz = 23 :
1, 1i  12 , 21 i ,
1, 0i  12 , 21 i ,
1, 1i  21 , 12 i ,
1, 0i  12 , 21 i , 1, 1i  12 , 12 i ,
(14.5.33)
1, 1i  21 , 12 i .
We want to discover the J2 eigenstates. A couple of states are quickly recognized. The top
state on this list has Jz = 32 ~ and since there are no states with higher Jz it must be the
top state of a j = 23 multiplet:
j = 23 , m = 32 i = 1, 1i  12 , 12 i .
(14.5.34)
331
Notice that J+ = L+ + S+ kills the state, as it should: L+ kills the first factor and S+ kills
the second factor. The bottom state of this j = 32 multiplet is also recognized; it is the
bottom state in the above table
j = 23 , m = 32 i = 1, 1i  12 , 21 i .
(14.5.35)
Since we must get complete multiplets, we must get the 4 states in j = 23 . The state with
Jz = 21 ~ must arise from a linear combination of states on the second line of the table and
the state with Jz = 12 ~ must arise from a linear combination of states on the third line of
the table. We are then left with two basis states, one with Jz = 21 ~ and one with Jz = 12 ~.
These must assemble into an additional j = 21 multiplet! Thus we will write
1
1
2
3
2
1
2
(14.5.36)
The lefthand side indicates that we have tensored the vector spaces of = 1 and s = 21 .
On the right hand side we have the direct sum of j = 32 and j = 12 . In terms of numbers of
basis states the above relation holds because
3 2 = 4 + 2.
(14.5.37)
This information is all we really need to find the splittings due to H, but let us finish the
computation. To find the j = 23 , m = 12 i state we will apply J to the equation
j = 23 , m = 32 i = 1, 1i  12 , 12 i .
Recalling that
J j, mi = ~
(14.5.38)
(14.5.39)
j = 32 , m = 12 i = ~ 3 j = 23 , m = 21 i . (14.5.40)
J 1, 1i  21 , 12 i = (L 1 + 1 S )1, 1i  12 , 12 i
= L 1, 1i  21 , 12 i + 1, 1i S  21 , 12 i
q
= ~ 1 2 1 01, 0i  12 , 12 i + ~ 12 32
= ~ 21, 0i  12 , 21 i + ~1, 1i  12 , 21 i .
1
2
( 21 )1, 1i  12 , 21 i
(14.5.41)
Equating the results of the two last calculations we find the desired state
j =
3
2, m
1
2i
2
3
1, 0i
 21 , 12 i
1
3
1, 1i  21 , 12 i .
(14.5.42)
332
The state came out normalized, as it should! As expected, this state is a linear combination
of the two states with Jz = 21 ~. The state j = 21 , m = 12 i must be an orthogonal linear
combination of the same two states (recall that eigenstates of hermitian operators with
different eigenvalues must be orthogonal). This allows us to easily write
j = 21 , m = 12 i =
1
3
1, 0i  21 , 12 i +
2
3
1, 1i  21 , 12 i .
(14.5.43)
The overall sign of this state is arbitrary and a matter of convention. The above state could
also have been calculated as follows.
Exercise. Consider the ansatz
j = 12 , m = 21 i = 1, 0i  12 , 21 i + 1, 1i  12 , 12 i ,
(14.5.44)
and determine the coefficients and , up to normalization, by the condition that the state
is annihilated by J+ .
We have identified by now four out of the six states (three states in the j = 32 multiplet
and one in the j = 21 multiplet). The remaining two are quickly obtained. Writing the
j, mi states without the explicit j = . . . and m = . . ., the full j = 32 and j = 12 multiplets
are
j=
3
2
 23 ,
3
2
i = 1, 1i  12 , 12 i
 32 ,
1
2
i =
 32 , 12 i =
2
3
1, 0i  21 , 12 i +
2
3
1, 0i  21 , 12 i +
 32 , 32 i = 1, 1i  12 , 12 i
j=
1
2
1
3
1, 1i  21 , 12 i
1
3
1, 1i  21 , 12 i
(14.5.45)
:
q
 ,
1
2i
1
3
=
1, 0i
+ 23 1, 1i  21 , 12 i
q
q
1
2
1
1
1
1
1 1
2, 2i =
3 1, 0i  2 , 2 i
3 1, 1i  2 , 2 i
1
2
 12 , 21 i
Let us comment on the states we did not derive. The state  32 , 12 i can be found (up to
an overall sign) by applying J to  32 , 12 i or by applying J+ to  32 , 32 i, both of which were
determined before. Similarly, the state  12 , 12 i can be found by applying J to  21 , 12 i or by
demanding orthogonality to  23 , 12 i. If you are new to this subject you should test your
knowledge by doing these short computations.
333
14.5.4
Spinorbit wrapup
It is now time to complete our determination of the spinorbit levelsplitting. All the hard
work has already been done, we just have to put the various pieces together. Begin with
the simplification of L S given that = 1 and s = 21 :
L S = 21 J2 L2 S2 ,
(14.5.46)
= 12 J2 ~2 1 2 ~2 12 23 ,
= 12 ~2 ~12 J2 11
4 .
hiHji =
1
r3
2,1
hiL Sj i.
(14.5.47)
Here the kets  i i represent the orbital and spin states, while the kets ii include the radial
state. It is clear now that if we choose the  i i to be the J2 eigenstates the above matrix of
H expectation values becomes diagonal. More explicitly writing the kets  i i as j, mi we
have
1
h2,1 , j , m H2,1 , j, mi =
hj , m  L S j, mi ,
r 3 2,1
1
1 2
11
1 2
hj
,
m

J
= 2~
2
(14.5.48)
4 j, mi ,
~
r 3 2,1
1
1 2
= 2~
j j m m .
j(j + 1) 11
4
3
r 2,1
The matrix is now manifestly diagonal, and the diagonal elements are the energy shifts.
There are four diagonal elements for j = 23 corresponding to the four possible values of the
associated m, and two diagonal elements for j = 12 corresponding to the two possible values
of the associated m. We can simplify our writing if we define an energy E0 by
1
1
e2
1
1
1 2
,
, with =
=
,
(14.5.49)
E0 2 ~
3
2
2
3
r 2,1
80 m c
r 2,1
24a30
where the last equality is the result of an explicit calculation using the radial wavefunction.
The energy shifts implied by (14.5.48) then take the form:
(14.5.50)
Ej,m = E0 j(j + 1) 11
4 ,
and are m independent. Thus the six states split into the four j = 23 states that shift up
together and the two j = 21 states shift down together :
= E0 23 52 11
= E0 ,
j = 32 : E 3
4
2 ,m
(14.5.51)
= E0 21 32 11
j = 12 : E 1
= 2E0 .
4
2 ,m
334
This is the final result for the spinorbit correction to the n = 2, = 1 levels of the hydrogen atom. As it turns out there is also a relativistic correction to these states that gives
comparable shifts. We will not consider them here.
14.6
We have examined the tensor states of two particles with spin one half, and the states of a
particle that has both orbital angular momentum and spin angular momentum and whose
state space is a tensor product of angular momentum states and spin states. In both cases
we found it useful to form a basis for the tensor product in which states were eigenstates of
the total angular momentum. Now we will consider the general situation and discuss some
important regularities.
1 and H
2 each of which may contain a number of multiplets
Consider two state spaces H
of angular momenta J1 and J2 , respectively,
M
M (j )
(j )
1 =
j1 , m1 i , J1 ,
H1 1
H1 1 =
H
2 =
H
j1
m1
j2
(j )
H2 2
(j )
H2 2 =
m2
(14.6.1)
j2 , m2 i ,
J2 .
i (with i = 1, 2) contains a set of multiplets with angular momenAs the above indicates H
tum ji , each of which contains the familiar 2ji + 1 states with different values of mi . The
direct sum symbol is appropriate as the various ji multiplet vector spaces are put together
1 and H
2 . The goal is to form the
as orthogonal subspaces inside the big total spaces H
1 H
2 and find a basis of states that are eigenstates of the
tensor product vector space H
total angular momentum, more precisely, eigenstantes of J2 and Jz , where
J = J1 1 + 1 J2 .
(14.6.2)
The full tensor product is given as a direct sum of the tensor product of the various multiplet
subspaces:
M (j )
(j )
1 H
2 =
H1 1 H2 2 .
(14.6.3)
H
j1 ,j2
It is therefore sufficient for us to understand the structure of the general summand Vj1 ,j2 in
the above relation
(j )
(j )
Vj1 ,j2 H1 1 H2 2 .
(14.6.4)
Note that
1
2,2
(14.6.5)
1
1, 2
when discussing
335
(j )
1. The uncoupled basis states for Vj1 ,j2 follow naturally by tensor product of H1 1
(j )
basis states and H2 2 basis states:
j1 , j2 , m1 , m2 i j1 , m1 i j2 , m2 i .
(14.6.6)
Here m1 runs over 2j1 + 1 values and m2 runs over 2j2 + 1 values. The complete set
of commuting observables CSCOu, with subscript u for uncoupled, is given by the
following list:
CSCOu : { J1 2 , J1z , J2 2 , J2z } .
(14.6.7)
The states j1 , j2 , m1 , m2 i are eigenstates of the operators in the list with eigenvalues
given by the list
{ ~2 j1 (j1 + 1), ~m1 , ~2 j2 (j2 + 1) , ~m2 } .
(14.6.8)
2. The coupled basis states for Vj1 ,j2 are eigenstates of the coupled set of commuting
observables CSCOc, with subscript c for coupled, that contains the following operators
CSCOu :
{ J1 2 , J2 2 , J2 , Jz } .
(14.6.9)
(14.6.10)
and the eigenvalues for the list of operators in CSCOc is given by the list
{ ~2 j1 (j1 + 1), ~2 j2 (j2 + 1) , ~2 j(j + 1), ~m } .
(14.6.11)
The states (14.6.10) come in full J multiplets Hj that contain all allowed m values
m = j, j 1, . . . , j:
M
Hj =
j1 , j2 ; j, mi .
(14.6.12)
m
If we are aiming to give a basis for Vj1 ,j2 , which has dimension (2j1 + 1)(2j2 + 1), we
will need a collection of J multiplets Hj with various values of j.
M
Vj1 ,j2 =
Hj .
(14.6.13)
jS(j1 ,j2 )
Here S(j1 , j2 ) is a list of values that depend on j1 and j2 . In combining two spin
onehalf particles we learned that for j1 = j2 = 12 that list is S( 21 , 12 ) = {1, 0}. In
doing spin orbit coupling we learned that for j1 = 1, j2 = 12 , we have S(1, 12 ) = { 23 , 12 }.
Both the coupled and uncoupled bases are orthonormal bases. One natural question is
how to relate the two bases. We did this explicitly for the case of 12 21 and for 1 12 , in
both cases figuring out how to write the coupled states in terms of the uncoupled ones. The
question of relating the two sets of basis vectors is the problem of calculating the socalled
336
ClebschGordan coefficients. To write this out clearly consider the completeness relation
for the uncoupled basis of the space Vj1 ,j2 . With the sums over m1 and m2 running over
the usual values,
X
j1 j2 ; m1 m2 ihj1 j2 ; m1 m2  ,
acting on Vj1 ,j2 .
(14.6.14)
1 =
m1 ,m2
Apply both sides of this equation to the coupled basis state j1 , j2 ; j, mi to find
X
j1 j2 ; m1 m2 i hj1 j2 ; m1 m2  j1 , j2 ; j, mi .
j1 , j2 ; j, mi =
{z
}

m1 ,m2
(14.6.15)
ClebschGordan coefficients
The overlaps selected above are those numbers called ClebschGordan (CG) coefficients.
If known, we know the relations between the two sets of basis vectors. Those numbers
appearing on the righthand sides of equations (14.5.45) are ClebschGordan coefficients.
Schematically, a CG coefficient is an overlap
hui cj i ,
(14.6.16)
where ui denotes uncoupled basis element and cj a coupled basis element. Since the two
bases are orthonormal, the transformation from one to the other is produced by a unitary
operator
X
U=
ck ihuk  ,
(14.6.17)
k
that manifestly maps up i to cp i. The matrix elements of this operator
X
X
hui U uj i =
hui ck ihuk uj i =
hui ck ikj = hui cj i ,
k
(14.6.18)
are precisely the ClebschGordan coefficients (recall that hui U uj i = hci U cj i because U is
a basischanging operator.)
There are selection rules for CG coefficients that must be explained:
1. We claim that
hj1 j2 ; m1 m2  j1 , j2 ; j, mi = 0 ,
unless m = m1 + m2 .
(14.6.19)
This is easily proven by inserting the Jz operator in between the overlap and setting
equal the two possible ways of evaluating the matrix element:
hj1 j2 ; m1 m2 Jz j1 , j2 ; j, mi = hj1 j2 ; m1 m2 (J1z + J2z )j1 , j2 ; j, mi .
(14.6.20)
On the lefthand side we let the operator act on the ket and on the righthand side
we let the operator act on the bra (by using its Hermiticity):
~m hj1 j2 ; m1 m2 j1 , j2 ; j, mi = (~m1 + ~m2 )hj1 j2 ; m1 m2 j1 , j2 ; j, mi .
(14.6.21)
337
(14.6.22)
This shows that the CG coefficient must vanish when m 6= m1 +m2 , which is equivalent
to our claim. Note that when we computed coupled states in terms of uncoupled ones
we used this rule: a coupled state with Jz = ~m could only be related to uncoupled
states in which the J1z and J2z eigenvalues added to ~m ; see for example the states
in (14.5.45).
2. The second selection rule specifies the values of j for the coupled state that can have
a nonvanishing overlap with uncoupled states defined by j1 and j2 . This really goes
back to the question posed when writing (14.6.13)
M
(j )
(j )
Vj1 ,j2 = H1 1 H2 2 =
Hj .
(14.6.23)
jS(j1 ,j2 )
We must figure out what values of j appear in the list S(j1 , j2 ). Only for those values
of j the CG coefficient hj1 j2 ; m1 m2  j1 , j2 ; j, mi need not vanish. There is also the
constraint of dimensionality. The sum of the dimensions of the Hj s in the equation
above must equal the dimensionality of Vj1 ,j2 .
We can quickly deduce the largest value of j in the list S(j1 , j2 ). The largest value
(j )
(j )
of Jz in H1 1 H2 2 is J1z + J2z , for J1z and J2z the largest possible. The largest
value of J1z is ~j1 and the largest value of J2z is ~j2 . Thus the largest value of Jz is
~(j1 + j2 ), implying that the largest value of j in the list is j1 + j2 . As it turns out,
the smallest value of j in the list is j1 j2  and the full list also includes all values of
j in the range separated by intervals of one unit:
S(j1 , j2 ) = {j1 + j2 , j1 + j2 1, . . . , j1 j2  } .
(14.6.24)
Note that each value of j appears only once. The customary form of stating (14.6.23)
now that the list of j values is known omits the Hs and simply states that
j1 j2 = (j1 + j2 ) (j1 + j2 1) ... j1 j2  .
(14.6.25)
This must be viewed as an equality between vector spaces. Or perhaps, more explicitly,
as a way to write the vector space on the lefthand as a direct sum of useful invariant
subspaces. The vector space on the lefthand side is the tensor product of the vector
space associated with the multiplet j1 of J1 and the vector space associated with
the multiplet j2 of J2 . On the righthand side we have a collection of multiplets
of the angular moment sum J = J1 + J2 . Thus each term on the righthand side
is an invariant subspace of the total angular momentum operators J. Since various
invariant subspaces appear, the total state space forms a reducible representation of
the total angular momentum.
338
There is one basic test of (14.6.25). The dimensionality of j1 j2 is (2j1 + 1)(2j2 + 1).
The dimensionality of the righthand side must be the same. Since the j1 and j2 labels can
be exchanged assume, without loss of generality, that j1 j2 . The dimensionality of the
right hand side is the sum of the dimensions of each of the representations. The sum is
readily done and the expected value is obtained:
jX
1 +j2
(2j + 1) = 2
1 +j2
jX
j=0
j=j1 j2
j1 j
2 1
X
j=0
j + j1 + j2 (j1 j2 ) + 1
(14.6.26)
j1 , j1 i
j1 , j1 1i
j1 , j1 2i
..
.
j1 , 2j2 j1 i
j2 , j2 i
j2 , j2 1i
..
.
[2]
..
.
j1 , j1 i
j2 , j2 i
[2j2 + 1]
j1 , 2j2 j1 1i
..
.
[1]
(14.6.27)
The data is aligned at the bottom: the lowest m states are on the same line. Since j1 j2
the higher m states of the j1 multiplet need not have sameline counterparts in the j2
multiplet. We have also added number labels to the states. They appear in brackets to the
left of the j1 multiplet and to the right of the j2 multiplet. Those labels allow us to refer
quickly to some particular tensor product. Thus, for example,
j1 , j1 2i j2 , j2 1i [3] [2]
(14.6.28)
We now walk through the construction of Figure 14.2 which shows why the selection
rule for j holds. The heavy dots are states and dots on a horizontal line are states with the
same value of Jz = ~m, the total zcomponent of angular momentum.
The first row consists of the single state [1] [1] with m = j1 + j2 .
For the second line we get two states [1] [2] and [2] [1] both with m = j1 + j2 1.
339
Figure 14.2: Diagram illustrating how j1 j2 with j1 j2 contains total angular momentum
representations with j = j1 + j2 all the way down to j = j1 j2 . Each heavy dot in the figure is a
state. The columns are full j multiplets, with j value indicated on the topmost state.
For the third line we get three states [1] [3] and [2] [2] and [3] [1] all with with
m = j1 + j2 2.
The number of states on each line will keep growing until we get a maximum of 2j2 + 1
states
[1] [2j2 + 1] , . . . , [2j2 + 1] [1] ,
(14.6.29)
all with m = j1 j2 . At this point we are using all of the states of the j2 multiplet.
The number of states on each line now remains the same until we get the set of states
[2j1 2j2 + 1] [2j2 + 1] , . . . , [2j1 + 1] [1] ,
(14.6.30)
340
We should note that the above argument shows that there are the right number of states
at all values of m that they may be assembled into the expected multiplets of total angular
momentum. If we think literally about the columns in the figure forming the multiplets,
then each dot represents some linear combination of the uncoupled basis states with the
right value of m. Since the tensor space carries a representation of total angular momentum,
the states must fall into a set of irreducible representations of total angular momentum.
The above argument shows that the claimed set of representations is the only possible one,
and therefore it must be realized.
14.7
Let us begin by reviewing the basics of the hydrogen atom spectrum which we discussed in
section 13.6. This is the spectrum we wish to derive by algebraic methods. We have energy
levels with energies En , where
En =
e2 1
,
2a0 n2
n = 1, 2, . . .
(14.7.1)
For each value n we have the following multiplets of orbital angular momentum L:
= 0, 1, . . . , n 1 .
(14.7.2)
A simple computation shows that this gives a total of n2 states. So at each n have a
degenerate n2 dimensional space Hn of energy eigenstates that can be written as the direct
sum of angular momentum multiplets
Hn = ( = n 1) ( = n 2) ( = 0) .
(14.7.3)
341
p2
e2
,
2m
r
(14.7.4)
L L = i~L .
(14.7.5)
1
r
(p L L p) .
2me2
r
(14.7.6)
As written this operator is Hermitian. It is also conserved, as it too commutes with the
Hamiltonian
[H, R ] = 0 .
(14.7.7)
Two rewritings of R are possible using the identity p L = L p + 2i~p. These are
R =
r
r
1
1
p
i~p
=
L
p
+
i~p
.
me2
r
me2
r
(14.7.8)
If R and L are conserved, all of their scalar products are conserved too since the Hamiltonian
would commute with them using the derivation property of commutators. We are already
i . Let us consider now R2 ,
familiar with the Casimir L2 that in fact commutes with all L
which must be a conserved scalar. Its evaluation was left as an exercise and gives
R2 = 1 +
2H
L2 + ~2 .
4
me
(14.7.9)
Notice that H, which appears in the above righthand side can be moved, if desired, to the
right of the parenthesis, as it commutes with L. The righthand side is indeed a conserved
scalar. We now look into RL, which must also be conserved. Classically RL vanishes since
R lies on the plane of the orbit (along the major axis of the ellipse) while L is orthogonal
to the plane of the orbit. The dot product of the quantum operators also vanishes, as we
now confirm. First recall that
r L = 0,
p L = 0.
(14.7.10)
1
(p L) L .
me2
(14.7.11)
(14.7.12)
342
(14.7.13)
Recall now (12.2.18) that states that for any vector u under rotations
L u + u L = 2i~ u .
(14.7.14)
The RungeLenz vector R is a vector under rotations. This we know without any computation, since it is built using cross products from p and L, which are both vectors under
rotations. Therefore we must have
L R + R L = 2i~ R .
(14.7.15)
(14.7.16)
This gives
[S1 , S2 ] , H
[S1 , H ] , S2
[S2 , H ] , S1 .
(14.7.17)
(14.7.18)
(14.7.19)
Each term on the righthand side vanishes: the first by conservation of S1 and the second
by conservation of S2 . It follows that
[S1 , S2 ] , H = 0 ,
(14.7.20)
which establishes the conservation of the commutator [S1 , S2 ]. This result is useful, for
it tells us that the commutator [Ri , Rj ] must be some conserved object. We can focus,
equivalently, on the cross product of two Rs which encodes the commutator. We must
have
R R = ( ) conserved vector ,
(14.7.21)
where the dots represent some conserved scalar. Since L and R are conserved vectors, the
possible vectors on the righthand side are L, R, and L R. To narrow down the options
we examine the behavior of various vectors under the parity transformation
r r .
(14.7.22)
343
L L,
R R .
(14.7.23)
The first follows from the preservation of the commutators between r and p. The second
from L = r p, and the third from the expression for R in terms of r, p, L. Since the
lefthand side of (14.7.21) does not change sign under the parity transformation, nor should
the righthand side. From our choices L, R, and L R, we can only have L. We must
therefore have
R R = ( )L ,
(14.7.24)
where, again, the expression in parenthesis must be a conserved scalar. A calculation (that
we dont show here) gives
2H
(14.7.25)
R R = i~ 4 L .
me
This completes the determination of all commutators relevant to L and R.
Now we come to the main point. We will derive algebraically the full characterization of
the subspaces of degenerate energy eigenstates. For this we will assume we have one such
subspace H at some energy E , where is a parameter to be specified below. We will look
at our operators in that subspace. In our operator relations (14.7.9) and (14.7.25) we are
actually allowed to replace H by the energy E , given that H, which commutes with L, can
be brought to the right, to be directly in front of the states. We can therefore set H equal
to the energy, that we write in the form
me4 1
, R.
(14.7.26)
2~2 2
Here is a unitfree, real constant, to be determined. The rest of the factors, except for
a convenient factor of two, provide the right units. Of course, we know that the correct
answer for these energies emerges if is a positive integer. This, however, is something we
will be able to derive. It follows from the above equation that we can set
H = E =
1
2H
= 2 2.
4
me
~
We can use this expression to simplify our key relations (14.7.25) and (14.7.9):
RR =
2
i~
1
~2 2
(14.7.27)
L,
1
= 1 2 2 (L2 + ~2 ) .
~
(14.7.28)
2 2
L +~ R
= ~ ( 1) .
(14.7.29)
344
These are clear and simple algebraic relations between our operators. The first one shows
that ~R has the units of angular momentum and kind of behaves like one, except that the
operator to the right is not ~R but rather L.
Our next step is to show that with the help of L and R we can construct two independent,
commuting algebras of angular momentum. Of course, it is clear that L is an algebra of
angular momentum. But, using suitable linear combinations of L and R we will obtain two
such algebras. Indeed, define
J1
J2
1
2 (L
1
2 (L
+ ~R) ,
(14.7.30)
~R) .
(14.7.31)
~R = J1 J2 .
It is important to realize that L is nothing but the sum of J1 and J2 . It is the total angular
momentum. We now claim that the operators J1 and J2 commute with each other. This is
quickly confirmed by direct computation:
[J1i , J2j ] = 41 Li + ~Ri , Lj ~Rj
(14.7.32)
= 14 i~ ijk Lk ~[Li , Rj ] ~[Lj , Ri ] i~ ijk Lk = 0 ,
where we noted that the first and last terms on the righthand side cancel each other out,
and the second and third terms also cancel each other out using (14.7.16). Now we want to
show that J1 and J2 are indeed angular momenta. We check both operators at the same
time using the notation J1/2 with the subscripts 1 and 2 going with the top and bottom
signs, respectively, on the righthand side:
J1/2 J1/2 = 14 (L ~R) (L ~R)
=
=
=
1
4 i~L + i~L (L ~R
1
4 (2i~L 2i~~R)
i~ 12 (L ~R) = J1/2 .
+ ~R L)
(14.7.33)
In the first step we used the first equation in (14.7.29) and in the second step (14.7.15). In
summary, we have confirmed that J1 and J2 are indeed two commuting angular momentum
operators:
J1 J1 = i~ J1 ,
J2 J2 = i~ J2 ,
[ J1 , J2 ] = 0 .
(14.7.34)
345
J21 = J22 .
(14.7.35)
Both angular momenta have the same magnitude on the subspace H of degenerate energy
eigenstates. Let us look at J21 . Again, using R L = 0 and the second of (14.7.29), we find
(14.7.36)
J21 = 14 L2 + ~2 2 R2 = 41 ~2 ( 2 1) .
Note that the energy parameter determines the magnitude of J21 . This is our eureka
moment: the quantization of angular momentum is going to imply the quantization of the
energy!
Since both J1 and J2 commute with the Hamiltonian, the degenerate subspace H
must furnish a simultaneous representation of both of these angular momenta! All states
in the subspace carry the same value of J21 and carry the same value of J22 . So all states
are simultaneously in some (irreducible) representation j1 of J1 and in some (irreducible)
representation j2 of J2 . The equality J21 = J22 , however, implies j1 = j2 j. We thus have
J21 = J22 =
1 2 2
~ ( 1) = ~2 j(j + 1) .
4
(14.7.37)
= 2j + 1 .
(14.7.38)
Note that as anticipated, the energy is determined by the value of j. This shows that in
fact each subspace H of degenerate energy eigenstates cannot carry more than one value
of j. As j runs over all possible values takes all positive integer values and thus can be
indentified with the principal quantum number n:
j = 0,
1
2,
1,
3
2, . . .
n = 2j + 1 = 1, 2, 3, 4, . . . .
(14.7.39)
We have recovered the quantization of the energy levels in the hydrogen atom!
What is the structure of the degenerate subspace Hn ? We already know that each state
must be an eigenstate of J1 with eigenvalue ~2 j(j + 1) and at the same time an eigenstate
of J2 with the same eigenvalue. Since the two angular momenta are independent, the states
can be described as a tensor product of a space that carries the representation j of J1 with
a space that carries the representation j of J2 . Therefore Hn , with n = 2j + 1, must be
j j where the first j is the value of J1 and the second j the value of J2 . More explicitly,
the degenerate subspace has the structure
Hn=2j+1 = j j,
Basis states
j; m1 i j; m2 i ,
j m1 , m2 j .
(14.7.40)
346
Since m1 and m2 each take 2j +1 values, the dimension of Hn=2j+1 is (2j +1)2 = n2 . This is
indeed, the expected number of states that we have at the energy level. As we are familiar,
the tensor product breaks into a sum of representations of the sum of angular momenta.
But the sum here is simply the conventional angular momentum L = J1 + J2 . Since we
know that
j j = 2j 2j 1 ... 0 ,
(14.7.41)
the representations on the righthand side are the multiplets that arise. Thus the degenerate subspace is a direct sum of the values
( = 2j) ( = 2j 1) ... 0 .
(14.7.42)
(14.7.43)
This is exactly the familiar set of multiplets at the degenerate subspace labeled by the
principal quantum number n. Done!
We should emphasize that the above algebraic analysis characterizes for us the possible
subspaces Hn of degenerate energy eigenstates. These subspaces are labeled by the values
of j, with j 0 and 2j Z and n = 2j + 1. This gives an infinite list {0, 21 , 1, 32 , 2, 52 , . . .}
of possible values of j. The algebraic analysis alone cannot tell us which values of j in the
above list are used by the hydrogen atom. In physics, however, it is often the case that
whatever is possible is in fact compulsory! Thus, it is not all that surprising that all possible
values of j actually appear in the hydrogen atom spectrum.
As the simplest nontrivial example of the structure of subspaces of degenerate energies
consider j = 12 which gives us n = 2. We then have H2 = 12 21 = 10, where the righthand
side are the triplet = 1 and the singlet = 0. The uncoupled basis states are of the form
 12 , m1 i  21 , m2 i and thus the four of them can be written briefly as  i,  i,  , i,  i.
The singlet and triplet are therefore
1, 1i =  i ,
= 1 : 1, 0i = 12  i +  i , = 0 : 0, 0i = 12  i  i . (14.7.44)
1, 1i =  i
These are the four states of the n = 2 energy level. Note that we have built them out of
spinonehalf states. These are mathematical entities, not related in any way to elementary
particle spin. In the hydrogen atom Hamiltonian that we used the electron and the proton
are treated as spinless. This is reminiscent of our analysis of the spectrum of the isotropic
twodimensional oscillator. There also, the energy levels of the oscillator formed representations of a hidden angular momentum, and included representations with fractional angular
momentum.