Copyright c 2021 by Robert G.

Littlejohn

Physics 221A
Academic Year 2021–22
Notes 23
Bound-State Perturbation Theory†

1. Introduction

Bound state perturbation theory applies to the bound states of perturbed systems, for which the
energy levels are discrete and separated from one another. The system may also have a continuous
spectrum, but the interest is attached to the discrete states. The effects of perturbations on states
of the continuous spectrum, or on discrete states imbedded in the continuum, require different
techniques, such as time-dependent perturbation theory.
The version of bound-state perturbation theory that you are probably familiar with is called
Rayleigh-Schrödinger perturbation theory. In these notes, we will begin with a variation on Rayleigh-
Schrödinger perturbation theory, called Brillouin-Wigner perturbation theory. The Brillouin-Wigner
theory is simpler and less messy to derive than the Rayleigh-Schrödinger theory, and it also gives
more accurate answers in many circumstances. Its disadvantage is that the unknown energy levels
are given only implicitly, in terms of the solutions of nonlinear equations. But if explicit formulas
for the energy levels are desired, it is easy to expand the solutions in powers of the perturbation
parameter, whereupon the results of the Rayleigh-Schrödinger theory are recovered. In my opinion,
this is the easiest and most elegant way to derive Rayleigh-Schrödinger perturbation theory.

2. The Unperturbed and Perturbed States

We consider an unperturbed Hamiltonian H0 with eigenvalues ǫk and eigenstates |kαi, where

α is an index introduced to resolve degeneracies, so that

H0 |kαi = ǫk |kαi. (1)

See Fig. 1, in which the unperturbed levels ǫk are illustrated on a vertical axis. Although we use a
notation appropriate for a discrete spectrum of H0 , we actually allow it to have a mixed spectrum,
typically a continuous spectrum above some energy and a discrete one below that, as often happens
in practice. Sums over states, such as Eq. (4), must be interpreted as including the continuous part
of the spectrum, as necessary, as shown for example in Eqs. (1.109) and (1.110).

† Links to the other sets of notes can be found at:

http://bohr.physics.berkeley.edu/classes/221/2122/221.html.
2 Notes 23: Bound-State Perturbation Theory

We pick one of the discrete levels ǫn for study, so the index n will be fixed for the following
discussion. We denote the eigenspace of the unperturbed system corresponding to eigenvalue ǫn by
Hn , so that the unperturbed eigenkets {|nαi, α = 1, 2, . . .} form a basis in this space.

Energy
Hn⊥

E
ǫk ǫn

|ψi
Q|ψi
Hn
P |ψi
H0 H0 + H1

Fig. 1. The energy levels ǫk of the unperturbed system H0 Fig. 2. Subspace Hn is the unperturbed eigenspace for
are shown on a vertical axis. These levels are degenerate level ǫn , and Hn⊥ is the orthogonal subspace. Projectors P
in general. One of these, ǫn , is selected out for study. and Q project onto Hn and Hn⊥ , respectively. The exact
When the perturbation H1 is turned on, level ǫn may split eigenstate lies approximately in Hn , with a small correc-
and shift. The energy level E of the perturbed system tion Q|ψi orthogonal to this subspace.
H0 + λH1 is one that grows out of ǫn as the perturbation
is turned on.

We take the perturbed Hamiltonian to be H = H0 + λH1 , where λ is a formal expansion

parameter that we allow to vary between 0 and 1 to interpolate between the unperturbed and
perturbed system. When the perturbation is turned on, the unperturbed energy level ǫn may split
and shift. We denote one of the exact energy levels that grows out of ǫn by E, as shown in Fig. 1.
We let |ψi be an exact energy eigenket corresponding to energy E, so that

H|ψi = (H0 + λH1 )|ψi = E|ψi. (2)

Both E and ψ are understood to be functions of λ; as λ → 0, E approaches ǫn and |ψi approaches

some state lying in Hn . Equation (2) only determines |ψi to within a normalization and phase,
which can be functions of λ; we are free to choose as we wish these to simplify our work.
It helps to have a picture of the perturbation problem in terms of the geometry of Hilbert
space. We break the Hilbert space into the subspace Hn and its orthogonal complement which we
denote by Hn⊥ . The latter space is spanned by the unperturbed eigenkets {|kαi} for all k 6= n.
Since the perturbation is small, the exact eigenket |ψi presumably lies nearly in Hn , with only a
small component orthogonal to Hn , as illustrated in Fig. 2. The components of |ψi parallel and
perpendicular to Hn are conveniently expressed in terms of the projector P onto the subspace Hn
and the orthogonal projector Q, defined by
X
P = |nαihnα|, (3)
α
 Notes 23: Bound-State Perturbation Theory 3

and
X
Q= |kαihkα|. (4)
k6=n
α

These projectors satisfy P 2 = P , Q2 = Q, P Q = QP = 0 and P + Q = 1. They also commute with

H0 ,
[P, H0 ] = [Q, H0 ] = 0, (5)
since they project onto eigenspaces of H0 . In terms of these projectors, the components of |ψi
parallel and perpendicular to Hn are P |ψi and Q|ψi, respectively, as illustrated in Fig. 2.
The component P |ψi of the exact eigenstate |ψi lies in the unperturbed eigenspace, and thus
is itself an unperturbed eigenstate. It is therefore a linear combination of the known unperturbed
eigenstates {|nαi, α = 1, 2, . . .}. The orthogonal component Q|ψi, on the other hand, is harder to
find. In the following we will think of P |ψi as the easy part of |ψi and Q|ψi as the hard part, and
attempt to solve for the hard part Q|ψi (or else the total solution |ψi = P |ψi + Q|ψi) in terms of the
easy part P |ψi. It turns out it is possible to write a neat power series expansion for this solution,
Eq. (16) below.
As a first step in the derivation of this series, we rearrange Eq. (2) in the form,

(E − H0 )|ψi = λH1 |ψi. (6)

We would like to divide this equation by E − H0 to solve for |ψi, but before proceeding we must
examine (E − H0 )−1 to see if it is well behaved. The idea of dividing by E − H0 is closely related
to the use of Green’s functions. See Notes 37, where the basic idea is presented and developed.

3. The Inverse of E − H0 on Hn⊥

The operator (E − H0 )−1 is a function of the operator H0 , which according to Eq. (1.129) can
be written in terms of the eigenvalues and projectors of H0 ,
1 X |kαihkα|
= . (7)
E − H0 E − ǫk
kα

The resulting expression, however, is meaningless if the exact eigenvalue E should equal any of the
unperturbed eigenvalues ǫk . In that case, one or more denominators vanish, indicating that E − H0
does not possess an inverse. Even if none of the denominators vanishes, nevertheless the exact
eigenvalue E is presumably close to the unperturbed eigenvalue ǫn , and therefore the terms k = n
in Eq. (7) are large (and these terms diverge at λ = 0, where E = ǫn ).
In order to avoid small or vanishing denominators, and to work with operators that are guar-
anteed to be well defined, we define a new operator R which is Eq. (7) with the terms k = n
suppressed:
X |kαihkα|
R= . (8)
E − ǫk
k6=n
α
4 Notes 23: Bound-State Perturbation Theory

The operator R is closely related to the Green’s operators that we shall consider later in the course,
when we take up scattering theory.
It is possible that E −ǫn is not the only small or vanishing denominator in Eq. (7). For example,
if there are other unperturbed energy levels ǫk lying close to ǫn , then the perturbation could push
the exact energy E near to or past some of these other levels, and then other small denominators
would result in Eq. (7). This will certainly happen if the perturbation is large enough. For the time
being we will assume this does not happen, so that Eq. (8) is free of small denominators. This is
the situation illustrated in Fig. 1. When this is not the case we shall refer to “nearly degenerate
perturbation theory,” which is discussed in Sec. 7.
The operator R satisfies
P R = RP = 0,
(9)
QR = RQ = R,
and
R(E − H0 ) = (E − H0 )R = Q. (10)

These equations show that R vanishes on the subspace Hn , while it is the inverse of E − H0 on the
subspace Hn⊥ . Note that Q is just the identity operator on Hn⊥ .

4. A series for |ψi in terms of P |ψi

We return now to Eq. (6) and multiply through by R, using Eq. (10). This gives

Q|ψi = λRH1 |ψi, (11)

which expresses the “hard” part of the exact eigenstate Q|ψi in terms of the total exact eigenstate
|ψi. To turn this into something more useful, we first add P |ψi to both sides, obtaining

|ψi = P |ψi + λRH1 |ψi. (12)

This expresses the exact eigenket |ψi in terms of its projections onto Hn and Hn⊥ , of which the second
term is a small correction. It is not an explicit solution for |ψi since |ψi occurs on the right hand
side, but it can be converted into a power series solution by a method of successive approximation.
In particular, note that if we neglect the second term in Eq. (12), which is of order λ, we obtain
|ψi = P |ψi, which is correct to lowest order. To improve on the approximation, we substitute the
left hand side of Eq. (12) into the second term on the right hand side, obtaining,

|ψi = P |ψi + λRH1 P |ψi + λ2 RH1 RH1 |ψi, (13)

which is still an exact equation. Now all terms on the right hand side except the last one involve
the “easy” part P |ψi of the exact eigenstate, and only the final term, which is second order in the
perturbation, involves |ψi. If we neglect that term, we obtain an expression for |ψi in terms of P |ψi
that is valid to first order in λ. Continuing this way, we can substitute the left hand side of Eq. (12)
 Notes 23: Bound-State Perturbation Theory 5

into the last term in Eq. (13), to push the hard term to fourth order in λ, etc. The result is a formal
power series for the solution |ψi in terms of the easy part P |ψi.
Another way to get the same thing is to rearrange Eq. (12),

(1 − λRH1 )|ψi = P |ψi, (14)

so that  1 
|ψi = P |ψi. (15)
1 − λRH1
The inverse operator is then easily expanded in a series,
∞
X
|ψi = λs (RH1 )s P |ψi
s=0

= P |ψi + λRH1 P |ψi + λ2 RH1 RH1 P |ψi + . . . . (16)

This is the basic series out of which all the results of perturbation theory are constructed. The series
might not converge, but in perturbation theory we do not necessarily need convergent series.

5. Nondegenerate Perturbation Theory

In nondegenerate perturbation theory the level ǫn of H0 is nondegenerate. Then the index α

is not needed for the level ǫn , and we can write simply |ni for the corresponding eigenstate. We
retain α for the other levels k 6= n, since these may still be degenerate. Then since the space Hn
is 1-dimensional and is spanned by |ni, we must have P |ψi = c|ni, where c is some constant. It is
obvious from Fig. 2 that |ψi and P |ψi cannot both be normalized, since one is the base and the
other the hypotenuse of a right triangle. In the following it is convenient to choose the normalization
and phase of |ψi so that c = 1. Thus we have

P |ψi = |ni, (17)

where we assume that |ni and the other unperturbed eigenstates are normalized. This is the most
convenient normalization convention for nondegenerate perturbation theory, but it means that |ψi
is not normalized, in fact from Fig. 2 it is clear that hψ|ψi > 1, since the hypoteneuse is longer than
the base. After we are done with the perturbation expansion, we are free to normalize |ψi in the
usual way. With this normalization convention, we have

hn|ψi = 1. (18)

Now the series (16) becomes

∞
X
|ψi = λs (RH1 )s |ni
s=0

= |ni + λRH1 |ni + λ2 RH1 RH1 |ni + . . .

6 Notes 23: Bound-State Perturbation Theory

X hkα|H1 |ni
= |ni + λ |kαi
E − ǫk
k6=n
α

X X hkα|H1 |k ′ α′ ihk ′ α′ |H1 |ni

+ λ2 |kαi + ... (19)
(E − ǫk )(E − ǫk′ )
k6=n k′ 6=n
α α′

which expresses the exact eigenket |ψi as an infinite series in terms of the unperturbed eigenket
|ni. However, the operator R still involves the unknown energy level E, which appears in the
denominators, so Eq. (19) is still not an explicit solution for the exact eigenket |ψi.
To find an equation for E, we multiply Eq. (6) on the left by hn| and use Eq. (18) to get

hn|(E − H0 )|ψi = E − ǫn = λhn|H1 |ψi, (20)

and then we substitute Eq. (19) into this. Thus we find

E = ǫn + λhn|H1 |ni + λ2 hn|H1 RH1 |ni + λ3 hn|H1 RH1 RH1 |ni + . . .

X hn|H1 |kαihkα|H1 |ni
= ǫn + λhn|H1 |ni + λ2
E − ǫk
k6=n
α

X X hn|H1 |kαihkα|H1 |k ′ α′ ihk ′ α′ |H1 |ni

+ λ3 + .... (21)
′
(E − ǫk )(E − ǫk′ )
k6=n k 6=n
α α′

Since the unknown exact energy E still appears in the denominators on the right hand side of this
equation, this is not an explicit solution for E. But Eq. (21) does give E explicitly through O(λ),

E = ǫn + λhn|H1 |ni + O(λ2 ), (22)

which agrees with the usual result of Rayleigh-Schrödinger perturbation theory. If we want the
energy correct through O(λ2 ), we can substitute the zeroth order approximation for E, namely,
E = ǫn , into the denominator of the O(λ2 ) term in Eq. (21), whereupon we find
X hn|H1 |kαihkα|H1 |ni
E = ǫn + λhn|H1 |ni + λ2 + O(λ3 ), (23)
ǫn − ǫk
k6=n
α

which agrees with Rayleigh-Schrödinger theory through O(λ2 ). Similarly, if we wish the energy
correct to O(λ3 ), we can substitute the first order expression (22) for E into the denominator of the
O(λ2 ) term in Eq. (21), and the zeroth order expression E = ǫn into the O(λ3 ) term, etc. When
the denominators are expanded out in powers of λ, the results of Rayleigh-Schrödinger perturbation
theory are recovered. At third order and beyond, the expressions of the Raleigh-Schrödinger theory
are more complicated than those of the Brillouin-Wigner theory, because of the extra terms that
come from expanding the denominators.
 Notes 23: Bound-State Perturbation Theory 7

Once the energy E has been determined in this manner, we can go back to Eq. (19) and obtain
the eigenkets. For example, through first order we have
X hkα|H1 |ni
|ψi = |ni + λ |kαi + O(λ2 ), (24)
ǫn − ǫk
k6=n
α

which again is the usual result from Rayleigh-Schrödinger perturbation theory.

6. Degenerate Perturbation Theory

In the case that the unperturbed energy level ǫn is degenerate, the projection of the exact
eigenket |ψi onto Hn must be a linear combination of the unperturbed eigenkets |nαi,
X
P |ψi = |nαicα , (25)
α

where cα are the expansion coefficients. Since Q|ψi is orthogonal to all the kets |nαi, we have

hnα|P |ψi = hnα|ψi = cα . (26)

To obtain an equation for the cα ’s, we multiply Eq. (6) on the left by hnα| and use Eq. (26) to
obtain,
hnα|(E − H0 )|ψi = (E − ǫn )cα = λhnα|H1 |ψi. (27)
We substitute Eq. (16) into this, using Eq. (25) for P |ψi, and we find
Xh i
(E − ǫn )cα = λhnα|H1 |nβi + λ2 hnα|H1 RH1 |nβi + . . . cβ
β
" #
X X hnα|H1 |kγihkγ|H1 |nβi
2
= λhnα|H1 |nβi + λ + . . . cβ . (28)
E − ǫk
β k6=n
γ

This equation must be solved simultaneously for the eigenvalues E and the unknown expansion
coefficients cα .
We see that the expansion coefficients cα , which determine P |ψi by Eq. (25), are the eigenvectors
of a g × g matrix, where g is the degeneracy of the unperturbed eigenvalue ǫn , and that the energy
shifts E −ǫn are the eigenvalues. But this is not the standard problem of determining the eigenvalues
and eigenvectors of a matrix, because the unknown energies E appear not only on the left-hand side
but also in the denominators in the second and higher order terms of the matrix.
In practice, however, the most common type of problem is one in which we desire the energy
shifts to first order in λ. We can find these by truncating the series in Eq. (28) at first order,
whereupon the energy shifts E − ǫn are the eigenvalues of the matrix hnα|H1 |nβi, that is, they
are the eigenvalues of the matrix of the perturbing Hamiltonian inside the degenerate eigenspace of
unperturbed system. At this order there are no energy denominators to worry about and we can
8 Notes 23: Bound-State Perturbation Theory

just diagonalize the matrix. Also, the eigenvectors cα corresponding to the energy shifts determine
P |ψi via Eq. (25), so that the perturbed eigenkets are known to zeroth order. In many applications
this is sufficient (see, for example, the treatmant of the Stark effect on the n = 2 levels of hydrogen
in Sec. 24.12).
The first order matrix hnα|H1 |nβi may or may not have degeneracies itself. If it does not, then
all degeneracies are lifted at first order; if it does, the remaining degeneracies may be lifted at a
higher order, or may persist to all orders. Degeneracies that persist to all orders are almost always
due to some symmetry of the system, which can usually be recognized at the outset.
To proceed to second order, we may substitute the zeroth order solution E = ǫn into the denom-
inators of the second order term in Eq. (28) and truncate the series at this order. We then obtain
a new eigenvalue-eigenvector problem for the unknown energy corrections E − ǫn and coefficients
cα , which differs from the one we had at first order by the addition of a second order matrix as a
correction to the matrix hnα|H1 |nβi that we had at first order. We can evaluate the eigenvalues
and eigenvectors of the sum of these two matrices by standard matrix techniques, or we can regard
the new eigenvalue-eigenvector problem as a new problem in bound state perturbation theory, in
which the first order matrix hnα|H1 |nβi plays the role of a new unperturbed Hamiltonian, say, H̃0 ,
and the second order matrix plays the role of a new perturbation, say, H̃1 . We will have to use
either the nondegenerate or degenerate versions of this theory, depending on whether or not all
degeneracies of the original Hamiltonian H0 were lifted at first order. In any case, once the energies
E and coefficients cα have been determined to some order, they can be substituted into Eq. (25) to
determine P |ψi.
Finally, once E and P |ψi are known to some order, these can be substituted into Eq. (16) to
determine Q|ψi.

7. Nearly Degenerate Perturbation Theory

Now let us consider the case in which the unperturbed levels of H0 , while not technically
degenerate, are close to one another. This often happens in practice, because levels are often grouped
into multiplets of closely spaced levels, which are more widely separated from other multiplets.
Suppose to be specific that two levels, say, ǫn and ǫm , are close enough to one another that first
order perturbations will push the exact level E close to or onto the unperturbed level ǫm . Then the
denominators in the k = m terms in Eq. (8) are small, and that equation cannot be used.
In this case we may proceed as follows. Let us choose choose some energy, call it ǭ, which is
close to ǫn and ǫm . It could, for example, be one of the energies ǫn or ǫm themselves, or it could
be their average, etc. Then let us take the original unperturbed Hamiltonian and perturbation and
rearrange them in the form,
H = H0 + H1 = H0′ + H1′ , (29)
 Notes 23: Bound-State Perturbation Theory 9

where
X
H0 = ǫk |kαihkα|, (30)
kα
X X
H0′ = ǫk |kαihkα| + ǭ |kαihkα|, (31)
k6=n,m k=n,m
α α
X
H1′ = H1 + (ǫk − ǭ)|kαihkα|. (32)
k=n,m
α

In effect, we have merged the nearly degenerate levels ǫn , ǫm into a single degenerate level ǭ, and
thrown the correction terms into H1′ . Then standard degenerate perturbation theory may be applied.
We will call this procedure “nearly degenerate perturbation theory.”
An equivalent approach to nearly degenerate perturbation theory is to exclude both sets of
terms k = n and k = m from the definition of R, and to proceed from there. Then R becomes the
inverse of E − H0 on the space orthogonal to Hn ⊕ Hm .

Problems

1. This problem concerns the volume effect, which was discussed briefly in Sec. 17.8.

(a) Compute the shifts in the energy levels of a single-electron (hydrogen-like) atom of nuclear
charge Z due to the finite size of the nucleus. Treat the nucleus as a uniform sphere of charge of
radius r0 A1/3 , where r0 = 1.3 × 10−13 cm and A is the number of nucleons. Express your answer
as some number of eV times some function of Z, A, and the quantum numbers of the atomic state.
Only treat those atomic states for which the effect is the largest. This is called the volume effect.
You may use
 Z 3/2
Rn0 (0) = 2 , (33)
na0
valid in hydrogen-like atoms, where a0 = h̄2 /me2 is the Bohr radius.
The volume effect induces a splitting between the 2s and 2p levels of hydrogen. Compare this
to the Lamb shift (another splitting of the same two levels). The Lamb shift raises the 2s level
relative to the 2p level by approximately 1.05 GHz in frequency units.

(b) A muonic atom is one in which the electron has been replaced by a negative muon µ− , a particle
with a charge q = −e and a mass of 105 MeV/c2 . Compute ∆E/E for the volume effect via first
order perturbation theory for the 1s level of muonic uranium (238 U). Is this calculation reliable?

2. In this problem, set h̄ = 1 for simplicity. The Hamiltonian for a rigid body is
1  L21 L2 L2 
H= + 2+ 3 , (34)
2 I1 I2 I3
10 Notes 23: Bound-State Perturbation Theory

where (I1 , I2 , I3 ) are the principal moments of inertia and where (L1 , L2 , L3 ) = (Lx , Ly , Lz ) are
angular momentum operators satisfying the commutation relations,

[Li , Lj ] = i ǫijk Lk . (35)

You may assume that the angular momentum takes on only integer values. The Hamiltonian (34)
describes the rotational spectrum of molecules. Physically, the quantities Li are the negatives of the
components of the angular momentum with respect to the body frame.
In the case of a spherical top, such as methane (CH4 ), we have I1 = I2 = I3 = I. The
Hamiltonian is
L2
H= , (36)
2I
and the energy levels are
ℓ(ℓ + 1)
Eℓ = , (37)
2I
which is (2ℓ + 1)-fold degenerate since Eℓ is independent of m.
In the case of a symmetric top such as ammonia(NH3 ), we have I1 = I2 = I⊥ 6= I3 . The
Hamiltonian is
L2 − L23 L2
H= + 3, (38)
2I⊥ 2I3
and the energy levels are
Eℓm = A + Bm2 , (39)

where
ℓ(ℓ + 1) 1 1 1 
A= , B= − , (40)
2I⊥ 2 I3 I⊥
and where m is the usual quantum number of Lz .
Now consider the case of a slightly asymmetric top. (Water is an example of an asymmetric top
molecule.) Let
1 1+ǫ 1 1−ǫ
= , = , (41)
I1 I⊥ I2 I⊥
where ǫ is small. These equations are equivalent to defining

2I1 I2 I2 − I1
I⊥ = , ǫ= . (42)
I1 + I2 I1 + I2

Write the Hamiltonian as H = H0 + ǫH1 , where H0 is the symmetric top Hamiltonian in Eq. (38).
You may find the following relations useful:

hm|L2+ |m′ i = hm′ |L2− |mi∗ , (43a)

′
hm|L2+ |m′ i = (−1)m+m h−m′ |L2+ |−mi, (43b)

but if you use Eq. (43b) you must prove it.

 Notes 23: Bound-State Perturbation Theory 11

(a) Consider the state m = 0. Find the first order of perturbation theory at which the energy shift
does not vanish, and compute the energy shift at that order. In addition to the abbreviations (40),
use
1
C= , (44)
4I⊥
and express your answer in terms of the constants A, B and C.

(b) Find the first order of perturbation theory at which the energy shifts for the m = ±1 levels do
not vanish, and compute them at that order.

(c) Find the first order of perturbation theory at which the energy shifts for the m = ±2 levels do
not vanish, and compute them at that order. Express your answer in terms of

ℓ(ℓ − 1)(ℓ + 1)(ℓ + 2) (ℓ − 3)(ℓ − 2)(ℓ + 3)(ℓ + 4)

X= , Y = . (45)
4B 12B
Notice the appearance of a sum over “intermediate states.”

(d) What order of perturbation theory do you have to go to in order to find the first nonvanishing
correction to the energy in the case of m = ±3? m = ±4?