221B Lecture Notes

Relativistic Quantum Mechanics

1 Need for Relativistic Quantum Mechanics

We discussed the interaction of matter and radiation field based on the Hami-
tonian
~ 2 Ze2 Z
(~p − ec A) 1 ~2 ~2
H= − + d~x (E + B ). (1)
2m r 8π
(Coulomb potential is there only if there is another static charged particle.)
The Hamiltonian of the radiation field is Lorentz-covariant. In fact, the
Lorentz covariance of the Maxwell equations is what led Einstein to propose
his special theory of relativity. The problem here is that the matter Hami-
tonian which describes the time evolution of the matter wave function is not
covariant. A natural question is: can we find a new matter Hamiltonian
consistent with relativity?
The answer turned out to be yes and no. In the end, a fully consistent for-
mulation was not obtained by modifying the single-particle Schrödinger wave
equation, but obtained only by going to quantum field theory. We briefly re-
view the failed attempts to promote Schrödinger equation to a relativistically
covariant one.

2 Klein–Gordon Equation
The Schrödinger equation is based on the non-relativisitc expression of the
kinetic energy
p~2
E= . (2)
2m
By the standard replacement
∂ ~
E → ih̄ , p~ → −ih̄∇, (3)
∂t
we obtain the Schrödinger equation for a free particle

∂ h̄2 ∆
ih̄ ψ=− ψ. (4)
∂t 2m

1
A natural attempt is to use the relativistic version of Eq. (2), namely
2
E

= p~2 + m2 c2 . (5)
c
Then using the same replacements Eq. (3), we obtain a wave equation
!2
h̄ ∂
φ = (h̄2 ∆ − m2 c2 )φ. (6)
c ∂t

It is often written as
m2 c2
!
2+ 2 φ = 0, (7)
h̄
where 2 = ( 1c ∂t )2 − ∆ is called D’Alambertian and is Lorentz-invariant. This
equation is called Klein–Gordon equation.
You can find plane-wave solutions to the Klein–Gordon equation easily.
Taking φ = ei(~p·~x−Et)/h̄ , Eq. (6) reduces to Eq. (5). Therefore, as long as
energy and momentum follows the Einstein’s relation Eq. (5), the plane wave
is a solution to the Klein–Gordon equation. So far so good!
The problem arises when you try to rely on the standard probability
interpretation of Schrödinger wave function. If a wave function ψ satisfies
Schrödinger equation Eq. (4), the total probability is normalized to unity
Z
d~xψ ∗ (~x, t)ψ(~x, t) = 1. (8)

Because the probability has to be conserved (unless you are interested in

seeing 5 times more particles scattered than what you have put in), this
normalization must be independent of time. In other words,
d Z
d~xψ ∗ (~x, t)ψ(~x, t) = 0. (9)
dt
It is easy to see that Schrödinger equation Eq. (4) makes this requirement
satisfied automatically thanks to Hermiticity of the Hamiltonian.
On ther other hand, the probability defined the same way is not conserved
for Klein–Gordon equation. The point is that the Klein–Gordon equation is
second order in time derivative, similarly to the Newton’s equation of motion
in mechanics. The initial condtions to solve the Newton’s equation of motion
are the initial positions and initial velocities. Similarly, you have to give both

2
initial configuration φ(~x) and its time derivative φ̇(~x) as the initial conditions
at time t. The time derivative of the “total probability” is
d Z Z
d~xφ (~x, t)φ(~x, t) = d~x(φ̇∗ (~x, t)φ(~x, t) + φ∗ (~x, t)φ̇(~x, t)),
∗
(10)
dt
and φ and φ̇ are independent initial conditions, it in general does not vanish,
and hence the “total probability” is not conserved. In other words, this is
an unacceptable definition for the probability, and the standard probability
interpretation does not work with Klein–Gordon equation.
One may then ask, if there is a conserved quantity we can possibly call
“probability.” It is easy to see that the following quantity is conserved:
Z
d~x(iφ∗ φ̇ − iφ̇∗ φ) (11)

using the Klein–Gordon equation. However, this quantity cannot be called

probability either because it is not positive definite.
Overall, the Klein–Gordon equation appears to be a good relativistic
replacement for the non-relativistic Schrödinger equation at the first sight,
but it completely fails to give the conventional probability interpretation of a
single-particle wave function. In other words, the Klein–Gordon equation, if
useful at all, does not describe the probability wave, which the Schrödinger
equation does, but describes something else. Because of this reason, the
Klein–Gordon equation was abandoned for a while. We will come back to
the question what it actually describes later on.

3 Dirac Equation
3.1 Heuristic Derivation
Dirac was the first to realize the problem with the probability interpretation
for equations with second-order time derivatives. He insisted on finding an
equation with only first-order time derivatives. Because the relativity re-
quires to treat time and space on equal footing, it means that the equation
has to be only first-order in spatial derivatives, too. Given the replacements
Eq. (3), the Hamiltonian must be linear in the momentum. Then the only
equation you can write down is of this form:
∂
ih̄ α · p~ + mc2 β]ψ.
ψ = Hψ = [c~ (12)
∂t
3
At this point, we don’t know what α ~ and β are. The Dirac further required
that this equation gives Einstein’s dispersion relation E 2 = p~2 c2 + m2 c4 like
the Klein–Gordon equation. Because the energy E is the eigenvalue of the
Hamitonian, we act H again on the Dirac wave function and find

H 2 ψ = [c2 αi αj pi pj + mc3 (αi β + βαi )pi + m2 c4 β 2 ]ψ. (13)

In order for the r.h.s. to give just p~2 c2 + m2 c4 , we need

αi αj + αj αi = 2δ ij , β 2 = 1, αi β + βαi = 0. (14)

These equations can be satisfied if αi , β are matrices! Setting the issue

aside why the hell we have to have matrices in the wave equation, let us
find solutions to the above equations. There are of course infinite number of
solutions related by unitary rotations, but the canonical choice Dirac made
was ! !
i 0 σi 1 0
α = , β= . (15)
σi 0 0 −1
They are four-by-four matrices, and σ i are the conventional Pauli matrices.
You can easily check the relations Eq. (14) using the matrices in Eq. (15).
Correspondingly, the wave function ψ must be a four-component column
vector. We will come back to the meaning of the multi-component-ness
later. But the first point to check is that this equation does allow a conserved
probability
d Z Z
ih̄ d~xψ † ψ = d~x[ψ † (Hψ) − (Hψ)† ψ] = 0, (16)
dt
simply because of the hermiticity of the Hamiltonian (note that α ~ , β matri-
ces are hermitean). This way, Dirac found a wave equation which satisfies
the relativistic dispersion relation E 2 = p~2 c2 + m2 c4 while admitting the
probability interpretation of the wave function.

3.2 Solutions to the Dirac Equation

Let us solve the Dirac equation Eq. (12) together with the matrices Eq. (15).
For a plane-wave solution ψ = u(p)ei(~p·~x−Et)/h̄ , the equation becomes
!
mc2 c~σ · p~
u(p) = Eu(p). (17)
c~σ · p~ −mc2

4
This matrix equation is fairly easy to solve. The first point to note is that the
matrix ~σ ·~p has eigenvalues ±|~p| because (~σ ·~p)2 = σ i σ j pi pj = 12 {σ i , σ j }pi pj =
δ ij pi pj = p~2 . Using polar coordinates p~ = |~p|(sin θ cos φ, sin θ sin φ, cos θ), we
find
! !
pz px − ipy cos θ sin θe−iφ
~σ · p~ = = |~p| , (18)
px + ipy −pz sin θeiφ − cos θ
and their eigenvectors
!
cos 2θ
~σ · p~χ+ (~p) = ~σ · p~ = +|~p|χ+ (~p), (19)
sin 2θ eiφ
!
− sin 2θ e−iφ
~σ · p~χ− (~p) = ~σ · p~ = −|~p|χ− (~p). (20)
cos 2θ
Once ~σ · p~ is replaced by eigenvalues ±|~p|, the rest of the job is to diagonalize
the matrix !
mc2 ±|~p|c
. (21)
±|~p|c −mc2
q
This is easily done using the fact that E = |~p|2 c2 + m2 c4 . In the end we
find two eingenvectors
 q   q 
E+mc2 E+mc2
χ (~p)
q 2mc2 +
χ (~p)
q 2mc2 −
u+ (p) =  E−mc2
, u− (p) =  E−mc2
. (22)
2mc2
χ+ (~p) − 2mc2 χ− (~p)
In the non-relativistic limit E → mc2 , the upper two components remain
O(1) while the lower two components vanish. Because of this reason, the up-
per two components are called “large components” while the lower two “small
components.” This point will play an important role when we systematically
expand from the non-relativistic limit.
An amazing thing is that there are two solutions with the same momen-
tum and energy, and they seem to correspond to two spin states. Then the
wave equation describes a particle of spin 1/2! In order to make this point
clearer, we look at the conservation of angular momentum. The commutator
[H, Li ] = [c~
α · p~ + mc2 β, ijk xj pk ] = −ih̄cijk αj pk 6= 0 (23)
does not vanish, and hence the orbital angular momentum is not conserved.
On the other hand, the matrix
!
~ = ~σ 0
Σ , (24)
0 ~σ

5
has the commutator

[H, Σi ] = [c~
α · p~ + mc2 β, Σi ] = cpj [αj , Σi ] = −2iijk cpj αk . (25)

Therefore, the sum

~ + h̄ Σ
J~ = L ~ (26)
2
commutes with the Hamiltonian and hence is conserved. Clearly, the matrix
h̄ ~
2
Σ has eigenvalues ± h̄2 and hence corresponds to spin 1/2 particle. The
eigenvectors u± (p) we obtained above are also eigenvectors of Σ ~ · p~ = ±|~p|
by construction, and hence J~ · p~ = h̄2 Σ ~ · p~ = ± h̄ |~p|. In other words, they
2
are helicity eigenstates (J~ · p~)/|~p| = ± h̄2 . Helicity is the angular momentum
projected along the direction of the momentum, where the orbital angular
momentum trivially drops out because of the projection. And hence the
helicity is purely spin. This analysis demonstrates that the Dirac equation
indeed describes a particle of spin 1/2 as guessed above.
This line of reasoning is fascinating. It is as if the consevation of prob-
ability requires spin 1/2. Maybe that is why all matter particles (quarks,
leptons) we see in Nature have spin 1/2!
But the equation starts showing a problem here. The Dirac wave function
ψ has four components, while we have obtained so far only two solutions.
There must be two more independent vectors orthogonal to the ones obtained
above. What are they? It turns out, they correspond to negative energy
solutions. Writing ψ = v(p)e−i(~p·~x−Et)/h̄ , the vectors v(p) must satisfy the
following matrix equation similar to Eq. (17) but with the opposite sign for
the mass term !
−mc2 c~σ · p~
v(p) = Ev(p). (27)
c~σ · p~ mc2
Therefore the solutions are obtained in the same manner but the upper two
and lower two components interchanged
 q   q 
E−mc2 2
χ (~p)
q 2mc2 +
− E−mc
2mc2
χ− (~p) 
v+ (p) =  E+mc2
, v− (p) =  q E+mc 2
. (28)
2mc2
χ+ (~p) 2mc2
χ− (~p)

Note that the definition ψ = v(p)e−i(~p·~x−Et)/h̄ has the energy and momentum
in the plane wave
q with the opposite sign from the normal one, and hence
positive E = |~p|2 c2 + m2 c4 means negative energy solution. There is no

6
reason to prefer positive energy solutions over negative energy ones as far as
the Dirac equation itself is concerned.
What is wrong with having negative energy solutions? For example,
suppose you have a hydrogen atom in the 1s ground state. Normally, it is
the ground state and it is absolutely stable because there is no lower energy
state it can decay into. But with the Dirac equation, the story is different.
There are infinite number of negative energy solutions. Then the 1s state can
emit a photon and drop into one of the negative energy states, and it happens
very fast (it is of the same order of magnitude as the 2p to 1s transition and
hence happens within 10−8 sec for a single negative energy state. If you sum
over all final negative-energy states, the decay rate is infinite and hence the
lifetime is zero)! Such a situation is clearly unacceptable.
Dirac is ingenious not just to invent this equation, but also to solve the
problem with the negative energy states. He proposed that all the negative
energy states are already filled in the “vacuum.” In his reasoning, the 1s
state cannot decay into any of the negetive energy states because they are
already occupied. It indeed makes the 1s state again absolutely stable. Now
the equation is saved again. The “vacuum” with all the negative energy
states (an infinite number of them) occupied is called the “Dirac sea.”
But there is a catch with the “Dirac sea.” We wanted to find a single-
particle wave function which is consistent with both relativity and probability
interpretation. The Dirac equation indees seems to be consistent both with
relativity and probability interpretation. But the correct implementation
calls for a multi-body state (actually, an infinite-body state)! We can’t just
talk about a single particle wave function ψ(~x) for a single electron, but
only a multi-particle one ψ(~x; ~y1 , ~y2 , · · ·) with an inifinite number of negative
energy electrons at positions ~yk . What it means is that we can’t talk about
single-particle wave mechanics in the end.
The hope for a good-old single-particle Schrödinger-like wave mechanics
is gone. We couldn’t do it with the Klein–Gordon equation because it didn’t
allow probability interpretation. We couldn’t do it with the Dirac equation
either because it ended up as a multi-particle problem. In the end, the only
way to go is the quantum field theory.

3.3 Dirac Field

Filling all negative energy states required a multi-body treatment of the elec-
tron, but this can be dealt with very easily within quantum field theory. It

7
is the same techinique when we dealt with Fermi-degenerate gas by inter-
changing the creation and annihilation operators. In other words, we will
talk about the “holes.”
Instead of talking about the Dirac equation as the probability wave, we
now talk about the Dirac field starting from the action
Z
d~xdt[ψ † ih̄ψ̇ − ψ † Ĥψ], (29)

where Ĥ is not longer the Hamiltonian acting on states, but rather a differ-
ential operator acting on the field ψ

Ĥ = −ih̄c~ ~ + mc2 β.
α·∇ (30)

By varying the action with respect to ψ † , we recover the Dirac equation

! !
∂ ∂ ~ − mc2 β ψ = 0.
ih̄ − Ĥ ψ = ih̄ + ih̄c~
α·∇ (31)
∂t ∂t

But the interpretation of the equation is now completely different, because

ψ is now an operator acting on the Hilbert space. The canonical anti-
commutation relation (remember the Dirac equation describes a spin 1/2
particle and we need Fermi statistics) is

{ψα∗ (~x), ψβ (~y )} = δ(~x − ~y ), (32)

where α, β indices refer to components (out of four) of the Dirac field. Using
the solutions to the Dirac equation we had obtained earlier, we expand the
Dirac field operator as
!
1 X X
u± (p)ei(~p·~x−Et)/h̄ au± (p) + v± (p)e−i(~p·~x−Et)/h̄ av± (p) .
X
ψ(~x, t) = 3/2
L p
~ ± ±
(33)
The creation and annihilation operators satisfy the usual anti-commutation
relations

{auλ (p), au† 0

λ0 (p )} = δp p0 δλ,λ0 ,
~,~ {avλ (p), av† 0
λ0 (p )} = δp p0 δλ,λ0 .
~,~ (34)

All other anti-commutators vanish.

8
 The trick we used in multi-body systems is that we can fill all states up
to the Fermi energy (which is the same as the chemical potential at zero
temperature) µ by defining new creation and annihilation operators
bv± (p) = av†
± (p), bv† v
± (p) = a± (p). (35)
In our case, the Fermi energy is zero to fill all negative energy states while
keeping all positive energy states unoccupied in the grounnd state. Then the
new creation/annihilation operators also satisfy the standard anti-commutation
relation
{bvλ (p), bv† 0
λ0 (p )} = δp p0 δλ,λ0 .
~,~ (36)
From this point on, I drop the superscript u, v with the understanding that
the a, a† operators refer to the positive energy solutions u, while the b, b† to
the negative energy ones v.
Then the expansion of the Dirac field is then
!
1 X X
u± (p)ei(~p·~x−Et)/h̄ a± (p) + v± (p)e−i(~p·~x−Et)/h̄ b†± (p) .
X
ψ(~x, t) = 3/2
L p
~ ± ±
(37)
The ground state (“vacuum”) is defined by
a± (p)|0i = b± (p)|0i = 0. (38)
The last requirement is the rephrasement of the fact that you can’t fill in any
more negative-energy electrons av† ± (p)|0i = 0 in the original notation.
The particles created by the operators a† are the normal electrons. What
are the states created by the operators b† , then? As in the case of degen-
erate Fermi gas, if you create a hole, by removing a particle that already
fills a state below the Fermi energy, the excitation behaves as a postively
charged particle. It has the momentum p~, because you have removed the
−i~
p·~
x/h̄
√ 2 2 −~p 2of4 the solution v(p)e
momentum , and also has a positive energy
E = p~ c + m c because you have removed the energy −E of the solution
v(p)e+iEt/h̄ . Therefore, the relation between the energy and the momentum
is precisely that of Einstein’s, and the “hole” behaves as a normal particle,
except that its charge is the opposite, and the same mass as the electron. It
is the anti-particle of the electron, namely the positron. Dirac theory hence
predicts the existence of an anti-particle for any spin 1/2 particles.1 The
1
Dirac himself, being afraid of predicting a non-existing particle, initially claimed that
this positively charged hole must be the proton. But other people pointed out that the
hole must have the same mass as the electron.

9
Fock space is constructed by acting electron creation operators a†± (p) and
positron creation operators b†± (p) on the vacuum.
Indeed the positron was discovered in cosmic rays by Anderson in 1932.
This was the first anti-particle.
One remark is that Fermi statistic is essential for the Dirac sea idea to
work. If we had tried to quantize the Dirac field as a boson (using com-
mutator instead of anti-commutator), we can never fill the negative energy
states enough, and system keep falling into lower and lower energy states
by creating more and more negative-energy electrons. This point shows an
amazing connection between spin and statistics: spin 1/2 particle must obey
Fermi statistics to obtain a consistent quantum fiedl theory.

3.4 Coupling to the Radiation Field

The gauge invariance discussed in 221A uniquely fixes the form of the in-
teraction between the Dirac field and the Radiation Field. It follows the
same rule in the Schrödinger theory p~ → p~ − ec A, ~ or equivalently, −ih̄∇ ~ →
~ − A.
−ih̄∇ e ~ Its Lorentz-covariant generalization also determines the time-
c
∂
derivative: ih̄ ∂t → ih̄ 1c ∂t
∂
− ec φ. (The relative sign difference is due to the fact
that Aµ = (φ, −A) ~ transforms the same way as the derivative ∂µ = ( 1 ∂ , ∇).) ~
c ∂t
Therefore, the Dirac action is now
!
Z
∂ ~ − e A)
~ − mc2 β ψ.
d~xdtψ † ih̄ − eφ − c~
α · (−ih̄∇ (39)
∂t c

The Dirac equation is again obtained by varying it with respect to ψ † ,

!
∂ ~ − e A)
~ − mc2 β ψ.
ih̄ − eφ − c~
α · (−ih̄∇ (40)
∂t c
When there is classical background field, such as an external Coulomb
potential in the hydrogen atom, we quantize the Dirac field by expanding
it in terms of the solutions to the Dirac equation in the presence of the
background field. Even though we do not regard Dirac equation any more
as the probability wave equation as in conventional quantum mechanics, but
rather a field equation, it is therefore necessary to solve the Dirac equation
in the presence of external fields. And for stationary solutions ψ ∝ e−iEt/h̄ ,
the proces is the same as solving the Dirac equation as if we are solving the
single-particle quantum mechanics problem.

10
 Therefore, we are interested in solving the equation

~ − eA
   
α · −ih̄∇
c~ ~ + mc2 β + eA
~ 0 ψ = Eψ. (41)
c
The way we will discuss it is by a sytematic expansion in ~v = p~/m. It
is basically a non-relativisic approximation keeping only a few first orders
in the expansion. Let us write Eq. (41) explicitly in the matrix form, and
further write E = mc2 + E 0 so that E 0 is the energy of the electron on top of
the rest energy. We obtain
~
~ − e A)
!
eφ c~σ · (−ih̄∇
~
~ − e A)
c ψ = E 0 ψ. (42)
c~σ · (−ih̄∇ c
−2mc2 + eφ
The solution lives mostly in the large components, i.e.. the upper two com-
ponents in ψ. The equation is diagonal in the absence of ~σ · (−ih̄∇ ~
~ − e A),
c
and we can regard it as a perturbation and expand systematically in powers
~ even though it must be
of it. To simplify notation, we will write p~ = −ih̄∇,
understood that we are not talking about the “momentum operator” p~ acting
on the Hilbert space, but rather a differential operator acting on the field ψ.
Let us write four components in terms of two two-component vectors,
!
χ
ψ= , (43)
η
where the large component χ is a two-component vector describing a spin
two particle (spin up and down states). η is the small component which
vanishes in the non-relativistic limit. Writing out Eq. (42) in terms of χ and
η, we obtain
e~
eφχ + c~σ · (~p − A)η = E 0χ (44)
c
e~
c~σ · (~p − A)χ + (−2mc2 + eφ)η = E 0 η. (45)
c
Using Eq. (45) we find
1 e~
η= c~σ · (~p − A)χ. (46)
E0 2
+ 2mc − eφ c
Substituting it into Eq. (45), we obtain
e~ 1 e~
eφχ + c~σ · (~p − A) c~
σ · (~
p − A)χ = E 0 χ. (47)
c E 0 + 2mc2 − eφ c

11
In the non-relativistic limit, E 0 , eφ  mc2 , and hence we drop them in the
denominator. Within this approximation (called Pauli approximation), we
find
~ 2
[~σ · (~p − ec A)]
eφχ + χ = E 0 χ. (48)
2m
The last step is to rewrite the numerator in a simpler form. Noting σ i σ j =
δ ij + iijk σ k ,
e~ 2 e e
[~σ · (~p − A)] = (δ ij + iijk σ k )(pi − Ai )(pj − Aj )
c c c
e~ 2 i e e
= (~p − A) + ijk σ k [pi − Ai , pj − Aj ]
c 2 c c
e ~ 2 ie
= (~p − A) + ijk σ k ih̄(∇i Aj − ∇j Ai )
c 2c
e ~ 2 eh̄ ~
= (~p − A) − ~σ · B. (49)
c c
Then Eq. (48) becomes
~ 2
(~p − ec A) eh̄
χ−2 ~ + eφχ = E 0 χ.
~s · B (50)
2m 2mc
In other words, it is the standard non-relativistic Schrödinger equation except
that the g-factor is fixed. The Dirac theory predicts g = 2! This is a great
success of this theory.

3.5 Tani–Foldy–Wouthuysen Transformation

One can extend the systematic expansion further to higher orders. It is done
usually with the method so-called Tani-Foldy–Wouthuysen transformation.
The basic idea is keep performing unitary basis transformation on ψ to elim-
inate small components at a given order in the expansion. The note here
is based on Bjorken–Drell, “Relativisc Quantum Mechanics,” McGraw-Hill,
1964.
Let us start with the free case, H = c~α · p~ + mc2 β. The problem is to
remove the mixing beween large and small components caused by the matri-
ces α
~ . Can we eliminate α completely from the Hamiltonian by a unitarity
rotation? The answer is yes. You choose the unitarity rotation as

ψ 0 = eiS ψ, (51)

12
where S = β~α · p~θ(~p) (it has nothing to do with the classical action). The
Hamitonian is also correspondingly unitarity transformed to
H 0 = eiS He−iS ! !
β~ α · p~ α · p~
β~
= cos |~p|θ + sin |~p|θ (c~ α · p~ + mc2 β) cos |~p|θ − sin |~p|θ
|~p| |~p|
mc2
!
α · p~ cos 2|~p|θ −
= c~ sin 2|~p|θ + β(mc2 cos 2|~p|θ + c|~p| sin 2|~p|θ).(52)
|~p|c
To eliminate the mixing between the large and small components, we choose
the parameter θ so that the first term vanishes:
|~p|
tan 2|~p|θ = . (53)
mc
√ √
Then we find cos 2|~p|θ = mc2 / c2 p~2 + m2 c4 , sin 2|~p|θ = c|~p|/ c2 p~2 + m2 c4 ,
and finally q
H 0 = β c2 p~2 + m2 c4 . (54)
This form correctly shows both positive and negative energy solutions.
Now we try to generalize this method in the presence of external radiation
field, starting again from the Hamiltonian
e~
α · (~p − A)
H = c~ + mc2 β + eφ. (55)
c
In this case, we must also allow ourselves to consider a time-dependent uni-
tarity transformation. The Dirac equation
ih̄ψ̇ = Hψ (56)
rewritten for the unitarity transformed field ψ 0 = eiS ψ is
" #
0 −iS ∂ −iS 0
ih̄ψ̇ = e HeiS
− ih̄e iS
e ψ = H 0ψ0 (57)
∂t
which defines the transformed Hamiltonian ψ 0 . (Notice the similarity to the
canonical transformations in the classical mechanics.)
We are interested in expanding H 0 up to O(p4 ). To this order, we find
1 i 1
H 0 = H + i[S, H] − [S, [S, H]] − [S, [S, [S, H]]] + [S, [S, [S, [S, H]]]]
2 6 24
i 1
−ih̄Ṡ − h̄[S, Ṡ] + h̄[S, [S, Ṡ]]. (58)
2 6
13
 |~
p|
In the free-particle case, we chose S = −iβ~
α ·~pθ with θ = 2|~1
p|
tan−1 mc 1
' 2mc .
e ~
Motivated by this, we can choose S = −iβ~ α · (~p − c A)/2mc. By calling
e ~
O=α ~ · (~p − A) and E = eφ, we find at this order
c

O2 O4
!
0 2 1 ih̄
H = β mc + − +E − [O, [O, E]] − [O, Ȯ]
2m 8m3 c2 2
8m c 2 8m2 c2
β O3 Ȯ
[O, E] −
+ + ih̄β . (59)
2mc 3m2 c 2mc
The last three terms still mix large and small components because they are
odd in O. At this point, we perform another unitarity transformation using
O0 O3
!
0 1 β Ȯ
S = −iβ = −iβ 2
[O, E] − 2 2
+ ih̄β . (60)
2mc 2mc 2mc 3m c 2mc2
Then the Hamiltonian is further transformed to
O2 O4
!
00 1 ih̄
H = β mc + − 2
+E − [O, [O, E]] − [O, Ȯ] + O00 ,
2m 8m3 c2 8m c2 2 8m2 c2
(61)
where O00 is still odd in α, but is suppressed by 1/m2 . Finally, using another
unitarity transformation with S 00 = −iβO00 /2mc eliminates the last term and
we find
O2 O4
!
000 21 ih̄
H = β mc + − + E − [O, [O, E]] − [O, Ȯ] (62)
2m 8m3 c2 8m2 c2 8m2 c2
to this order. Now we write it out explicitly and find
 
~ 2
(~p − ec A) p~4  eh̄ ~ ~
H 000 = β mc2 + − 3 2
+ eφ − βΣ · B
2m 8m c 2mc
ieh̄2 ~ ~ ~ − eh̄ Σ
2
~ × p~) − eh̄ ∇
~ · (E ~ · E.
~ (63)
Σ ·−(∇ × E)
8m2 c2 4m2 c2 8m2 c2
The first term is nothing but the rest energy, and the second the non-
relativistic kinetic term. The third term is the relativistic correction to the
kinetic energy. The Coulomb potential term is there as desired, and the
next term is the magnetic momentum coupling with g = 2 as we saw before.
~ ×E
∇ ~ = 0 for the Coulomb potential, while
eh̄ ~ ~ eh̄ ~ 1 dV eh̄ 1 dV ~ ~
− 2 2
Σ · (E × p~) = 2 2
Σ· (~x × p~) = Σ·L (64)
4m c 4m c r dr 4m2 c2 r dr
14
is the spin-orbit coupling (with the correct Thomas precession factor, where
eh̄2 ~ ~
I assumed a central potential). The last term − 8m 2 c2 ∇ · E is called Darwin

term, in honor of the first person who solved the hydrogen atom problem
exactly with the Dirac equation.2
The physical meaning of the relativistic correction, the spin-orbit couping, and the
magnetic moment coupling are probably familiar to you. What is the Darwin term? It is
attribtuted to a perculiar motion of a Dirac particle called Zitterbewegung (Schrödinger).
One way to see it is by using Heisenberg equation of motion (well, we shouldn’t use the
“Hamiltonian” α ~ · p~ + mβ too seriously because we abandoned the single-particle wave
mechanics interpretation, but it is still instructive). The velocity operator is
d
ih̄ ~x = [~x, H] = ih̄c~
α. (65)
dt
This is already quite strange. The velocity operator ~x˙ = c~ α has eigenvalues ±c and
velocities in different directions do not commute (i.e. not simultaneously observable).
Clearly, this velocity is not the motion of the particle as a whole, but something rather
different. To see this, we further consider the Heisenberg equation for the velocity operator
when the particle is at rest p~ = 0 (H = mc2 β). Then,
d2
ih̄ α, H] = 2mc3 α
~x = [c~ ~ β, (66)
dt2
while its further derivative is
d3 d
− h̄2 ~x = [2mc3 α
~ β, H] = 4m2 c5 α
~ = 4m2 c4 ~x. (67)
dt3 dt
Therefore,
d 2mc2 t 2mc2 t
~x(t) = c~
α cos − ic~αβ sin , (68)
dt h̄ h̄
which oscillates very rapidly with the period h̄/2mc2 = 6 × 10−22 sec. (Note that −i~ αβ is
hermitean, because (−i~ αβ)† = iβ~ α = −i~ αβ due to the anti-commutation relation.) The
position is then obtained by integrating it:
2mc2 t 2mc2 t
 
h̄
~x(t) = ~x(0) + α
~ sin + i~
αβ cos . (69)
2mc h̄ h̄
This rapid motion of an “electron at rest” is the Zitterbewegung, a peculiarity in the
relevatistic quantum mechanical motion of spin 1/2 particle. Because of this rapid motion
of the electron, the net electric field the electron experiences is averaged over its “blur,”
and hence is somewhat different from the electric field at the position itself. The averaging
of the electric field gives rise to the correction
1 ∂2V
hV i = h(δxi )(δxj )i i j , (70)
2 ∂x ∂x
2
Dirac himself did not do this. An anecdote I’ve read is that Dirac was so proud of his
equation that he was afraid of doing any tests which might falsify it. Of course we was
technically capable enough to solve it exactly, but he didn’t do it because of this fear.

15
where the isotropy tells us that h(δxi )(δxj )i = 13 δ ij h(δxi )2 i = δ ij (h̄/2mc)2 , where I used
the time average of the Zitterbewegung at the last step. Then the correction to the
potential energy is
1 h̄2 eh̄2 ~ ~
heV i = e ∆V = − ∇ · E, (71)
2 4m2 c2 8m2 c2
reproducing the Darwin term.

3.6 Hydrogen-like Atoms

Now let us specialize to the case of the hydrogen atom A ~ = 0 and eφ =
2
−Ze /r. The energy levels of a hydrogen-like atom are perturbed by the
additional terms in Eq. (63). The sum of the relativistic correction and the
spin-orbit coupling gives a correction
3
!
4 2 3 j(j + 1) − 3l(l + 1) − 4
(Zα) mc + (72)
8n4 2l(l + 1)(2l + 1)n3
Due to some magical reason I don’t understand, for both possible cases where
l = j ± 21 allowed by the addition of angular momenta, it simplifies to
!
4 2 3 1
(Zα) mc − . (73)
8n4 (2j + 1)n3
Therefore states with the same principal quantum number n and the total
angular momentum j, even if they come from different l, remain degenerate.
The case l = 0 is special and deserves attention. For this case, the spin-
orbit interaction vanishes identically. The relativisic correction to the kinetic
energy gives (Zα)4 mc2 ( 8n3 4 − n13 ). However, for s-waves only, the Darwin
term also contributes. For hydrogen-like atoms, the Darwin term is

eh̄2 ~ ~ eh̄2 Zαh̄3

− ∇ · E = Ze4πδ(~
x ) = 4πδ(~x). (74)
8m2 c2 8m2 c2 8m2 c
The first-order perturbation in the Darwin term gives

Zαh̄3 1
2
4π|ψ(0)|2 = (Zα)4 mc2 3 . (75)
8m c 2n
Therefore the sum of the relativistic correction and the Darwin term gives
!
3 1 3 1
 
4 2 4 2
(Zα) mc 4
− 3 = (Zα) mc 4
− , (76)
8n 2n 8n (2j + 1)n3

16
because j = 1/2 for l = 0, and hence happens to have the same form as
Eq. (73) which is valid for l 6= 0
In the case of hydrogen-like atoms, one can also solve the Dirac equation
exactly to find the energy levels. The derivation is discussed at this end of
this lecture note. We find
  2 −1/2
Zα
Enjlm = mc2 
1 +  q   . (77)
n − (j + 1/2) + (j + 1/2)2 − (Zα)2

The principal quantum numbers are n = 1, 2, 3, · · · as usual, and j + 1/2 ≤ n.

Expanding it up to O(Zα)4 , we find

(Zα)2 (Zα)4
" !#
2 3 1
Enjlm = mc 1− + − (78)
2n2 n3 8n 2j + 1

and agrees with the results based on Tani–Foldy–Wouthuysen transforma-

tion. One important point is the degeneracy between 2s1/2 and 2p1/2 states
(similarly, 3s1/2 and 3p1/2 , 3p3/2 and 3d3/2 , etc) persists in the exact solution
to the Dirac equation. This degeneracy is lifted by so-called Lamb shifts,
due to the coupling of electron to the zero-point fluctuation of the radiation
field. We will come back to this point later.

4 Klein–Gordon Field
We had abandoned the Klein–Gordon equation because it did not admit
probability interpretation. We instead went to the Dirac equation because
it did, but in the end the problem of negative energy solutions forced us to
regard the Dirac equation as a field equation which had to be quantized to
obtain the full Fock space. Then the issue with the probability interpretation
was basically moot. What don’t we reconsider the Klein–Gordon equation
now as a field equation rather than a probability wave equation?3
3
According to Tomonaga’s book “The Story of Spin,” which has many interesting
anecdotes in the development of quantum mechanics, Pauli felt beaten by Dirac who
could make Pauli’s ad-hoc spin matrices to a fully relativistically covariant theory. But
Pauli, together with Heisenberg, later showed that Klein–Gordon equation is as good as
Dirac equation once regarded as a quantum field theory. This was Pauli’s revenge.

17
4.1 Quantized Klein–Gordon Field
The action for the Klein–Gordon field is given by
2 2
" #
Z
1 ~ − m c φ∗ φ .
~ ∗ · ∇φ
S= d~xdt 2 φ̇∗ φ̇ − ∇φ (79)
c h̄2
By varying the action with respect to φ∗ , we recover the Klein–Gordon equa-
tion Eq. (6). The canonically conjugate momenta are
1 ∗ 1
π(~x) =
2
φ (~x), π ∗ (~x) = 2 φ(~x). (80)
c c
and the canonical commutation relations are
[φ(~x), ψ(~y )] = ih̄δ(~x − ~y ), [φ∗ (~x), ψ ∗ (~y )] = ih̄δ(~x − ~y ). (81)
Following the expansion we did for the radiation field, we expand the field
in momentum modes,
s
2πh̄c2 X 1 p·~
φ(~x) = 3 √ (a(~p)e
i~ x/h̄
+ b† (~p)e−i~p·~x/h̄ ) (82)
L p
~
ωp
s
2πh̄c2 X √
φ̇(~x) = 3
(−i ωp )(a(~p)ei~p·~x/h̄ − b† (~p)e−i~p·~x/h̄ ). (83)
L p
~

Unlike the radiation field, we regard φ(~x) as a complex field, and hence a(~p)
and b(~p) can be different. Together with the canonical commutation relations,
we find the usual commutation relations among creation and annihilation
operators
[a(~p), a† (~q)] = δp~,~q, [b(~p), b† (~q)] = δp~,~q, (84)
with all other combinations vanishing. The Fock space is constructed in the
usual way, starting from the vacuum
a(~p)|0i = b(~p)|0i = 0, (85)
and acting creation operators a† (~p) and b† (~p) on the vacuum.
The Hamiltonian of the Klein–Gordon field is
m2 c2 ∗
" !#
Z
∗ ∗ 1 ∗ ~ ∗ ~
H = d~x π φ̇ + π φ̇ − 2 φ̇ φ̇ − ∇φ · ∇φ − 2 φ φ
c h̄
2 2
" #
~ + m c φ∗ φ .
Z
= ~ ∗ · ∇φ
d~x π ∗ π + ∇φ (86)
h̄2

18
The Hamiltonian is manifestly positive and there is no concern with negative
energies. Rewriting it in terms of creation and annihilation operators, we
find Xq
H= p~2 c2 + m2 c4 (a† (~p)a(~p) + b† (~p)b(~p) + 1). (87)
p
~

The zero-point energy is present both for a and b operators and hence the
term 1 in the parentheses. We see no inconsistencies in the quantized Klein–
Gordon field. This is indeed the way we treat spinless bosons in the rela-
tivistic quantum field theory.
What is the difference between the a and b particles? It can be seen by
coupling the Klein–Gordon field to the radiation field:
1 ie ie
Z     
S = d~xdt 2 ∂t − A0 φ∗ ∂t + A0 φ
c h̄ h̄
2 2
#
ie ie m c
   
− ∇~ + A ~ φ∗ · ∇ ~ − A ~ φ− φ∗ φ . (88)
h̄c h̄c h̄2

The scalar potential A0 (here I avoided using φ not to be confused with the
Klein–Gordon field) must couple to the electric charge, which is read off from
the above action as
1 Z 1 i
d~x 2 (φ∗ φ̇ − φ̇∗ φ) = (a† (~p)a(~p) − b† (~p)b(~p)).
X
(89)
4π c h̄ p
~

Clearly, a and b particles have the opposite charges, while they have the same
mass as seen in the Hamiltonian. They are anti-particles of each other. This
is how we describe Yukawa’s charged pion π + and its anti-particle π − in the
quantum field theory.

4.2 Hydrogen Atom in Klein–Gordon Equation

It has been of interest to nuclear and atomic physicists to study bound states
of charged pions to nuclei, “pi-mesic atoms.” The point is that the pion is
more than 200 times heavier than the electron, and hence the “Bohr radius”
is correspondingly shorter, down to 10−11 cm level. Therefore pions probe
much deeper structure than electrons do. Their energy levels are obtained
by solving the Klein–Gordon equation in the Coulomb potential.
We will see below that the time-independent field equation for the radial
wave function φ = R(r)Ylm e−iEt/h̄ has the same form as the non-relativistic

19
Schrödinger equation for the hydrogen atom,

h̄2 d2 Ze2
" ! #
2 d λ(λ + 1)
− 2− + − R = R. (90)
2µ dr r dr r2 r

We write µ, λ,  in terms of E, m, and l starting from the Klein–Gordon

equation.
The field equation for φ = R(r)Ylm e−iEt/h̄ is
 !2 
Ze2 1 d2 2 l(l + 1)
!
1
 −E − h̄2 r − − m2 c2  R = 0. (91)
c2 r r2 dr2 r2

By reorganizing terms, we find

h̄2 c2 d2 l(l + 1) − Z 2 α2 Ze2 E 2 − (mc2 )2

" ! #
2 d
− 2− + − R= R. (92)
2E dr r dr r2 r 2E

α = e2 /h̄c is the fine-structure constant. By comparing to the Schrödinger-

like equation Eq. (90), we find

µ = E/c2 (93)
s
1 2 1
 
λ = − Z 2 α2 −
l+ (94)
2 2
2 2 2
E − (mc )
 = . (95)
2E
Eq. (90) has exactly the same form as the Schrödinger equation for the
hydrogen atom, except that λ is not an integer. Therefore the boundstate
eigenvalues are given by
1 Z 2 α2 µc2
=− ,
2 ν2
where the “principal quantum number” ν takes values ν = λ + 1, λ + 2, λ +
3, · · ·. This observation allows us to solve for E.

E 2 − (mc2 )2 1 Z 2 α2 E
=− . (96)
2E 2 ν2

20
Solving for E, we find4
mc2
E=q . (97)
1 + Z 2 α2 /ν 2
We now Expand E up to O(Z 2 α2 ) and see that it agrees with the result of
conventional Schrödinger equation including the rest energy. By expanding
Eq. (97) up to O(Z 2 α2 ), we find
1 Z 2 α2
!
2
E = mc 1− + O(Z 4 α4 ) . (98)
2 ν2
Note that λ = l +O(Z 2 α2 ). Therefore, ν = λ+k (k is a non-negative integer)
and hence ν is also an integer up to an O(Z 2 α2 ) correction. Neglecting
O(Z 4 α4 ) terms, we find the principal quantum number n = ν + O(Z 2 α2 ) and
hence
1 Z 2 α2 mc2
E = mc2 − + O(Z 4 α4 ). (99)
2 n2
The result agrees with conventional Schrödinger equation at this order.
We next expand E in Eq. (97) up to O(Z 4 α4 ), and find
1 Z 2 α2 3 Z 4 α4
!
E = mc2 1− + + O(Z 6 α6 ) . (100)
2 ν2 8 ν4
The difference between ν and n at O(Z 2 α2 ) cannot be ignored in the second
term because it gives rise to a term of O(Z 4 α4 ). By expanding λ up to
O(Z 2 α2 ),
Z 2 α2
λ=l− + O(Z 4 α4 ), (101)
2l + 1
we can write
Z 2 α2
ν =n− + O(Z 4 α4 ), (102)
2l + 1
and hence
1 Z 2 α2 Z 4 α4 3 Z 4 α4
!
2
E = mc 1− − + + O(Z 6 α6 ) . (103)
2 n2 (2l + 1)n3 8 n4
4
I’ve read somewhere that Klein–Gordon equation was the first equation considered
by Schrödinger. He actually solved the hydrogen atom problem with the Klein–Gordon
equation (or maybe the “original” Schrödinger equation), and found that the result does
not agree with data concerning the fine structure. He then abandoned it and took the
non-relativistic limit so that the equation and the data agree within the approximation.
It is sometimes a good idea to ignore the failure and forge ahead!

21
As before, the second term is the term we obtain in non-relativistic Schrödinger
equation.
The question is what are the next two terms. They are the so-called
“relativistic correction,” obtained by expanding the relativistic kinetic energy
q p~2 1 (~p2 )2
p~2 c2 + (mc2 )2 = mc2 + − + O(~p6 ). (104)
2m 8 m3 c2
Because |~p|/m = v = Zα in hydrogen-like atoms, O(~p6 ) ∼ O(Z 6 α6 ) and
these terms are beyond our interest. We can rewrite

Ze2 1 Z 2 α2 mc2
!
2
p~ |nlmi = 2m − , (105)
r 2 n2

and hence
!2
1 (~p2 )2 1 Ze2 1 Z 2 α2 mc2
hnlm| − |nlmi = − hnlm| − |nlmi. (106)
8 m3 c2 2mc2 r 2 n2

Using (see below for the derivation of these expectation values)

1 1 1 2
hnlm| |nlmi = 2 , hnlm| |nlmi = , (107)
r na r2 (2l + 1)n3 a2

with a = h̄2 /mZe2 = h̄/mcZα, we find

1 (~p2 )2 Z 4 α4 3 Z 4 α4
hnlm| − |nlmi = − + . (108)
8 m3 c2 (2l + 1)n3 8 n4

This precisely reproduces the O(Z 4 α4 ) terms in Eq. (97), and hence the rel-
4
ativistic correction − 18 mp~3 c4 is their origin. Obviously, there is no spin-orbit
coupling because the Klein–Gordon field does not have spin. What is more in-
teresting is that there is no Darwin term; the Klein–Gordon particle does not
do Zitterbewegung! In fact, if you take the square√ root of the Klein–Gordon
equation and consider the Hamiltonian to be H = c2 p~2 + m2 c√ 4 , the Heisen-

berg equation would give the velocity ~x˙ = [~x, H]/ih̄ = c2 p~/ c2 p~2 + m2 c4
which is perfectly normal, showing no sign of Zitterbewegung.
The energy levels of the Klein–Gordon equation in the Coulomb potential
is the starting point for the study of π-mesic atoms, i.e, the bound states of
negative pions π − to nuclei.

22
 We can derive Eq. (107) without suffering through generating functions for Laguerre
polynomials by using the Feynman-Hellman theorem which states5
∂H ∂E
hψ| |ψi = , (109)
∂λ ∂λ
quite generally when a Hamiltonian H, its eigenstates |ψi, and its eigenvalues E depend
on a parameter λ. (The eigenstates if degenerate must be diagonalized not to mix under
infinitessimal changes in λ.) To show equation (109) start with
∂ ∂
(H|ψi) = (E|ψi) (110)
∂λ ∂λ
∂H ∂ ∂E ∂
|ψi + H |ψi = |ψi + E |ψi, (111)
∂λ ∂λ ∂λ ∂λ
and act on the left with hψ|. Then hψ|H = hψ|E so that the unwanted terms drop out:
∂H ∂ ∂E ∂
hψ| |ψi + Ehψ| |ψi = hψ| |ψi + Ehψ| |ψi
∂λ ∂λ ∂λ ∂λ
∂H ∂E
=⇒ hψ| |ψi = . (112)
∂λ ∂λ
Now for the non-relativistic hydrogen atom,

h̄2 d2 Ze2
 
2 d l(l + 1)
H = − 2− + − , (113)
2m dr r dr r2 r
Z 2 α2 mc2
E = − . (114)
2n2
Mathematically, we can consider Z to be a continuous parameter and apply the Feynman-
Hellman theorem,

1 1 ∂H 1 ∂E 1 Zα2 mc2 1
hnlm| |nlmi = − 2 hnlm| |nlmi = − 2 = 2 = 2 , (115)
r e ∂Z e ∂Z e n2 n a
which is the first of (107). To find the second relation we can basically repeat the above
argument with l in place of Z, but there is one subtlety. For the Hamiltonian (113),
the radial eigenvalue problem is well-defined even for non-integer l. But when solving
for the radial wavefunction, we find a principle quantum number n = nr + l + 1 where
nr = 0, 1, 2, . . . must be an integer for the hypergeometric series to terminate and give a
normalizable radial wavefunction. When we differentiate with respect to l we must hold
nr , not n, fixed. In other words, ∂n∂l = 1. Then

1 2m ∂H
hnlm| |nlmi = hnlm| |nlmi
r2 h̄2 (2l + 1) ∂l
2m ∂E
=
h̄2 (2l + 1) ∂l
5
This derivation is by Ed Boyda.

23
 2m 2Z 2 α2 mc2 ∂n
=
h̄2 (2l + 1) 2n3 ∂l
2
= . (116)
(2l + 1)n3 a2

4.3 Hydrogen Atom in Dirac Equation

Now that you have seen how to obtain the energy levels for the Klein–Gordon
equation, you must be wondering what we do for the Dirac equation. Here
is how you do it. Starting from the Dirac equation
Ze2
" #
E+ α · p~ − mc2 β ψ = 0,
− c~ (117)
r
multiply by
Ze2
" #
E+ α · p~ + mc2 β
+ c~ (118)
r
from the left. Then you find
 !2 !
2 2
Ze ~ Ze  ψ = 0.
 E+ − c2 p~2 − (mc2 )2 + c~
α· −ih̄∇ (119)
r r

The anti-commutation relation {αi , αj } = 2δ ij , {αi , β} = 0 had been used in

simplifying the expression. Writing out the derivative acting on the Coulomb
potential, we find
 !2 
Ze2 Ze2
 E+ 2 2
α · ~ˆr 2  ψ = 0,
2 2
− c p~ − (mc ) + ih̄c~ (120)
r r

using the notation ~ˆr = ~r/r. At this point, we also rewrite p~2 using the
spherical coordinates,
 !2 
Ze2 1 d2 2
!
2 2 l(l + 1) 2 2 ˆr Ze  ψ = 0.
 E+ + c h̄ r − − (mc ) + ih̄c~
α · ~
r r dr2 r2 r2
(121)
We can block-diagonalize the matrix α ~ as
! !
0 ~σ ~σ 0
α
~= −→ . (122)
~σ 0 0 −~σ

24
 ~ · ~ˆr = ±~σ · ~ˆr.
Then depending on upper or lower two components, we have α
Then the equation becomes
  
Ze2 1 d2 l(l + 1) + Z 2 α2 ± iZα~σ · ~ˆr 
2
E − (mc ) + 2E 2 2
+ c2 h̄2  r − ψ = 0.
r r dr2 r2
(123)
The non-trivial point with this equation is to deal with the numerator l(l +
1) + Z 2 α2 ± iZα~σ · ~ˆr. The trick is to note that it commutes with J~ = L ~ +~σ /2.
Therefore, we can look at the subspace of the Hilbert space with fixed j and
hence l = j ± 1/2. On this space, the numerator has the form
!
(j + 21 )(j + 32 ) + Z 2 α2 ∓iZα
l(l+1)+Z α ±iZα~σ ·~ˆr =
2 2
.
∓iZα (j − 2 )(j + 12 ) + Z 2 α2
1

(124)
The eigenvlaues of this matrix are easily obtained, but we intentionally write
the eigenvalues as λ(λ + 1). The motiation to do so must be clear from what
we did with the Klein–Gordon equation. The two solutions are
" 2 #1/2 " 2 #1/2
1 2 2 1 2 2
λ+ = j+ −Z α , λ− = j+ −Z α − 1. (125)
2 2
Using λ, the Dirac equation is now
Ze2 1 d2
" !#
λ(λ + 1)
2
E − (mc ) + 2E 2 2
+ c2 h̄2 2
r− ψ = 0. (126)
r r dr r2
It has the same form as the Klein–Gordon equation except λ. By following
the same arguments, we find the energy eigenvalues
mc2
E=q , (127)
1 + Z 2 α2 /ν 2
with ν = λ + 1, λ + 2, · · ·. The solutions with both λ+ and λ− give the same
set of ν’s, except that the smallest ν is obtained only from λ− with j = 1/2.
This corresponds to the fact that n = 1 state has only l = 0 which does not
mix with an l = 1 state. The degeneracy of the eigenvalues for two solutions
is split only by Lamb shift. The principal quantum number is ν at the lowest
order in Zα, and hence
" 2 #1/2
1 1
 
2 2
ν =n+ j+ −Z α − j+ . (128)
2 2

25
We finally find the energy levels of the Dirac equation
 !2 −1/2
2 Zα
E = mc 1 +  , (129)
n − (j + 1/2) + [(j + 1/2)2 − Z 2 α2 ]1/2

showing Eq. (77).

26