Phys 852, Quantum mechanics II, Spring 2009

Time-Independent Perturbation Theory

Prof. Michael G. Moore, Michigan State University
1 The central problem in time-independent perturbation theory:
Let H
0
be the unperturbed (a.k.a. background or bare) Hamiltonian, whose eigenvalues and eigenvectors
are known. Let E
(0)
n
be the n
th
unperturbed energy eigenvalue, and |n
(0)
be the n
th
unperturbed energy
eigenstate. They satisfy
H
0
|n
(0)
= E
(0)
n
|n
(0)
(1)
and
n
(0)
|n
(0)
= 1. (2)
Let V be a Hermitian operator which perturbs the system, such that the full Hamiltonian is
H = H
0
+V. (3)
The standard approach is to instead solve
H = H
0
+V , (4)
and use as for book-keeping during the calculation, but set = 1 at the end of the calculation.
The goal is to nd the eigenvalues and eigenvectors of the full Hamiltonian (4). Let E
n
and |n be the n
th
eigenvalue of the full Hamiltonian, and its corresponding eigenstate, respectively. They satisfy
H|n = E
n
|n (5)
and
n|n = 1. (6)
Because is a small parameter, it is assumed that accurate results can be obtained by expanding E
n
and
|n in powers of , and keeping only the leading term(s). Formally expanding the perturbed quantities
gives
E
n
= E
(0)
n
+E
(1)
n
+
2
E
(2)
n
+. . . , (7)
and
|n = |n
(0)
+|n
(1)
+
2
|n
(2)
+. . . , (8)
where E
(j)
n
and |n
(j)
are yet-to-be determined expansion coecients Inserting these expansions into the
eigenvalue equation (5) then gives
(H
0
+V )(|n
(0)
+|n
(1)
+
2
|n
(2)
+. . . .) = (E
(0)
n
+E
(1)
n
+
2
E
(2)
n
+. . .)(|n
(0)
+|n
(1)
+
2
|n
(2)
+. . . .). (9)
1
Because of the linear independence of terms in a power series, this equation can only be satised for
arbitrary if all terms with the same power of cancel independently. Equating powers of thus gives

0
: H
0
|n
(0)
= E
(0)
n
|n
(0)
, (10)

1
: (H
0
E
(0)
n
)|n
(1)
= V |n
(0)
+E
(1)
n
|n
(0)
(11)

2
: (H
0
E
(0)
n
)|n
(2)
= V |n
(1)
+E
(2)
n
|n
(0)
+E
(1)
n
|n
(1)
(12)
etc . . . (13)
This generalizes to

j
: (H
0
E
(0)
n
)|n
(j)
= V |n
(j1)
+
j

k=1
E
(k)
n
|n
(jk)
. (14)
Similiarly, we can expand the normalization equation in powers of , giving
1 = n
(0)
|n
(0)
+(n
(1)
|n
(0)
+n
(0)
|n
(1)
) +
2
(n
(2)
|n
(0)
+n
(1)
|n
(1)
+n
(0)
|n
(2)
) +. . . . (15)
Again, we require that all terms of the same power in cancel independently, resulting in the set of
equations:

0
: n
(0)
|n
(0)
= 1, (16)

1
: n
(1)
|n
(0)
+n
(0)
|n
(1)
= 0, (17)

2
: n
(2)
|n
(0)
+n
(1)
|n
(1)
+n
(0)
|n
(2)
, (18)
which readily generalizes to

j
:
j

k=0
n
(jk)
|n
(k)
= 0. (19)
Solving these equations can be simplied by choosing a convenient global phase-factor. In this case, we
can choose the phase of the full eigenstates to be such that n
(0)
|n is real-valued. To enforce our choice
of global phase, we can expand n
(0)
|n in power series, giving
n
(0)
|n = 1 +n
(0)
|n
(1)
+
2
n
(0)
|n
(2)
+. . . . (20)
Requiring that the r.h.s. be real-values for any real-valued requires that each term be independently
real-valued, so that
n
(0)
|n
(j)
= n
(j)
|n
(0)
. (21)
2
2 The non-degenerate case
We will now describe how to solve these equations in the case where none of the unperturbed energy levels
are degenerate.
Step #1: To obtain the j
th
correction to the n
th
energy eigenvalue, simply hit the
j
equation from the
left with the bra n
(0)
| and solve for E
(j)
n
, giving
E
(j)
n
= n
(0)
|V |n
(j1)

j1

k=1
E
(k)
n
n
(0)
|n
(jk)
, (22)
where we have used n
(0)
|(H
0
E
(0)
n
) = 0 and n
(0)
|n
(0)
= 1. Note that all quantities on the r.h.s are of
order < j, and are thus presumed to be known.
Step #2: To obtain the j
th
correction to the n
th
eigenstate, we hit both sides of the
j
equation with the
bra m
(0)
| where m = n. This gives
m
(0)
|n
(j)
=
m
(0)
|V |n
(j1)

E
(0)
m
E
(0)
n
+
j1

k=1
E
(k)
n
m
(0)
|n
(jk)

E
(0)
m
E
(0)
n
, (23)
which is the expansion coecient of the j
th
correction onto the m
th
bare eigenvector.
Step #3: To obtain the remaining unknown quantity, n
(0)
|n
(j)
, we simply solve the normalization
equation, giving
n
(0)
|n
(j)
=
1
2
j1

k=1
n
(jk)
|n
(k)
. (24)
Steps 1, 2, and 3 must be solved iteratively, starting with j = 1, and repeating for each order until the
desired accuracy is achieved.
Final Step: Once these quantities are computed, we can reconstruct the n
th
eigenvalue and eigenvector
via
E
n
=

j=0

j
E
(j)
n
= E
(0)
n
+E
(1)
n
+
2
E
(2)
n
+. . . , (25)
and
|n = |n
(0)

j=0
n
(0)
|n
(j)
+

m=n
|m
(0)

j=0
m
(0)
|n
(j)

= |n
(0)

_
1 +n
(0)
|n
(1)
+
2
n
(0)
|n
(2)
+. . .
_
+

m=n
|m
(0)

_
m
(0)
|n
(1)
+
2
m
(0)
|n
(2)
+. . .
_
.
(26)
3
2.1 First-order terms:
Introducing a more compact notation via
V
mn
= m
(0)
|V |n
(0)
. (27)
and
E
mn
:= E
(0)
m
E
(0)
n
, (28)
we can compute the rst-order terms from Eqs. (22), (23), and (24) with j = 1, yielding
E
(1)
n
= V
nn
, (29)
m
(0)
|n
(1)
=
V
mn
E
mn
, (30)
and
n
(0)
|n
(1)
= 0. (31)
2.2 Second-order terms:
By setting j = 2, and inserting the rst-order solutions, eqs (22), (23), and (24) give us
E
(2)
n
= n
(0)
|V |n
(1)

m=0
|V
mn
|
2
E
mn
, (32)
m
(0)
|n
(2)
=
m
(0)
|V |n
(1)

E
mn

V
nn
V
mn
E
2
mn
=

=n
V
mm
V
m

n
E
mn
E
m

V
mn
V
nn
E
2
mn
(33)
and
n
(0)
|n
(2)
=
1
2
n
(1)
|n
(1)

=
1
2

m=n
|V
mn
|
2
E
2
mn
(34)
4
2.3 Eigenvectors and eigenvalues to second-order:
Putting the rst- and second-order terms together then gives the results
E
n
= E
(0)
n
+V
nn

2

m=n
|V
mn
|
2
E
mn
+O(
3
), (35)
and
|n = |n
(0)

_
_
1

2
2

m=n
|V
mn
|
2
E
2
mn
+O(
3
)
_
_
+

m=n
|m
(0)

_
_

V
mn
E
mn
+
2
_
_

m

=n
V
mm
V
m

n
E
mn
E
m

V
mn
V
nn
E
2
mn
_
_
+O(
3
)
_
_
. (36)
Truncating the series, and setting = 1 then gives the approximations
E
n
E
(0)
n
+V
nn

m=n
|V
mn
|
2
E
mn
, (37)
and
|n |n
(0)

_
_
1
1
2

m=n
|V
mn
|
2
E
2
mn
_
_
+

m=n
|m
(0)

_
_

V
mn
E
mn
+

=n
V
mm
V
m

n
E
mn
E
m

V
mn
V
nn
E
2
mn
_
_
. (38)
5
2.4 The second-order Hamiltonian
With these results, we can reconstruct the full-Hamiltonian in the unperturbed basis to second-order via
H = H
0
+V
=

n
E
n
|nn|

n
_
_
E
(0)
n
+V
nn

m=n
|V
mn
|
2
E
mn
_
_

_
_
|n
(0)

_
_
1
1
2

m=n
|V
mn
|
2
E
2
mn
_
_
+

m=n
|m
(0)

_
_

V
mn
E
mn
+

=n
V
mm
V
m

n
E
mn
E
m

V
mn
V
nn
E
2
mn
_
_
_
_

_
_
_
_
1
1
2

m=n
|V
mn
|
2
E
2
mn
_
_
n
(0)
| +

m=n
_
_

V
nm
E
mn
+

=n
V
nm
V
m

m
E
mn
E
m

V
nn
V
nm
E
2
mn
_
_
m
(0)
|
_
_
(39)
Dropping all terms beyond second-order then gives
H

n
|n
(0)
n
(0)
|
_
_
E
(0)
n
+V
nn

m=n
|V
mn
|
2
E
mn
_
1 +
E
(0)
n
E
mn
_
_
_

m=n
|n
(0)
m
(0)
|
_
_
_
E
(0)
n
+V
nn
_
V
nm
E
mn

=n
E
(0)
n
V
nm
V
m

m
E
mn
E
m

n
+
E
(0)
n
V
nn
V
nm
E
2
mn
_
_

m=n
|m
(0)
n
(0)
|
_
_
_
E
(0)
n
+V
nn
_
V
mn
E
mn

=n
E
(0)
n
V
mm
V
m

n
E
mn
E
m

n
+
E
(0)
n
V
mn
V
nn
E
2
mn
_
_
+

m=n

=n
|m
(0)
m
(0)
|
_
E
(0)
n
|V
mn
|
2
E
2
mn
_
. (40)
With this Hamiltonian as the generator of the unitary time-propagator, we see how the perturbation will
introduce dynamical coupling between the bare eigenstates, as described by the o-diagonal terms.
6
2.5 Examples
2-Level system: The rst example we can consider is the two-level system. Here we have H
0
= S
z
and
V = S
x
, so that
H = S
z
+ S
x
. (41)
Here the Rabi-frequency will take the place of the perturbation parameter . Let the ground state of H
0
be | , with eigenvalue E

= /2, and let the excited state be | , with eigenvalue E

= /2, where
clearly we have assumed > 0. Let |0 be the ground state of the full Hamiltonian, with eigenvalue E
0
and let |1 be the excited state of the full Hamiltonian, with eigenvalue E
1
. The zeroth order terms in the
expansions of E
n
and |n are therefore E
(0)
0
= /2, E
(0)
1
= /2, |0
(0)
= | and |1
(0)
= | . Thus we
have
H
0
=

2
(| | | |) (42)
and
V =

2
(| | +| |). (43)
To second-order in perturbation theory we then nd the perturbed eigenvalues to be
E
0
= E

+ V

2
|V

|
2
E

=
_

2


2
4
+O(
3
)
_
(44)
and
E
1
= E

+ V

2
|V

|
2
E

=
_

2
+

2
4
+O(
3
)
_
(45)
This clearly indicates the phenomena of level repulsion. Regardless of the sign of , the leading-order
eect of the perturbation is to push the energy levels apart.
For the perturbed eigenstates we nd
|0 = |
_
1

2
2
|V

|
2
E
2

+O(
3
)
_
+|
_

+
2
_
V

E
2

_
+O(
3
)
_
= |
_
1

2
8
2
+O(
3
)
_
|
_

2
+O(
3
)
_
(46)
and
|1 = |
_
1

2
2
|V

|
2
E
2

+O(
3
)
_
+|
_

+
2
_
V

E
2

_
+O(
3
)
_
= |
_
1

2
8
2
+O(
3
)
_
+|
_

2
+O(
2
)
_
. (47)
7
Computing the normalization gives
0|0 =
_
1

2
8
2
+O(
3
)
_
2
+
_

2
+O(
3
)
_
2
= 1 +O(
3
) (48)
and
1|1 =
_
1

2
8
2
+O(
3
)
_
2
+
_

2
+O(
3
)
_
2
= 1 +O(
3
), (49)
which veries that the state is properly normalized, so that no re-normalization is required. Checking
orthogonality gives
0|1 =
_
1

2
8
2
+O(
3
)
__

2
+O(
3
)
_

2
+O(
3
)
__
1

3
8
2
+O(
3
)
_
= 0. (50)
8
Square well with delta-function For the second example we consider an innite square-well potential
as the unperturbed problem, and we perturb it by placing a delta-function potential at the center. Our
goal will be to compute the energy shifts to second order and the perturbed eigenstates to rst order. We
will attempt to explicitly evaluate all sums for the ground state only. The unperturbed Hamiltonian is
H
0
=
P
2
2M
+U(X), (51)
where
U(x) =
_
0 : 0 < x < L
: else
(52)
The form of the perturbation is
V (X) =

2
ML
(X
L
2
), (53)
so that the full Hamiltonian is
H = H
0
+V =
P
2
2M
+U(X) +

2
ML
(X
L
2
). (54)
The unperturbed energy levels are given by
E
(0)
n
=

2

2
2ML
2
n
2
; n = 1, 2, 3, . . . , (55)
with corresponding unperturbed wavefunctions

n
(x) x|n
(0)
=
_
2
L
sin
_
n
L
x
_
. (56)
The rst order energy shift is given by E
(1)
n
= V
nn
, where V
nn
n
(0)
|V |n
(0)
. Inserting the projector onto
position eigenstates gives
V
nn
=
_
dx

n
(x)V (x)
n
(x) =
2
2
ML
2
_
dx sin
2
_
n
L
x
_
(x
L
2
) =
_
2
2
ML
2
; n odd
0; n even
. (57)
The matrix element vanishes for even n because these wavefunctions have a node at x = L/2., the odd n
wavefunctions have an anti-node at x = L/2, so that the sin-function takes on its maximum value of unity.
The rst order correction to the wavefunction is given by the usual expression |n
(1)
=

m=n
|m
(0)

Vmn
Emn
.
Thus we must evaluate the o-diagonal matrix elements, giving
V
mn
=
2
2
ML
2
_
dx sin
_
m
L
x
_
(x
L
2
) sin
_
n
L
x
_
=
_
(1)
m+n
2
2
2
ML
2
: m and n odd
0; m and/or n even
. (58)
The energy denominator is E
mn
= E
(0)
m
E
(0)
n
=

2

2
2ML
2
(m
2
n
2
). The perturbed wavefunction to rst
order is dened as
n
(x) x|n
(0)
+x|n
(1)
, giving

n
(x) =
_
2
L
sin
_
n
L
x
_
+
4

2
_
2
L

m,n=1
m=n
sin
_
(2m1)
L
x
_
(1)
m+n1
(2m1)
2
(2n1)
2
. (59)
9
In fact, this sum can be performed analytically, although the result is not very illuminating. Computing
the rst-order perturbed ground-state (n = 1) wavefunction, for example, gives

1
(x)
_
2
L
_
1

2
_
sin(x/L) +
2
L
(x Lu(x L/2))
_
2
L
cos(x/L), (60)
where u(x) is the unit step function. The discontinuous derivative at x = L/2 is characteristic of eigenstates
with delta-potentials.
The second order correction to the energy-shift is computed from E
(2)
n
=

m=n
|Vmn|
2
Emn
, which yields
E
(2)
n
=
8
2

2
ML
2

m,n=1
m=n
1
(2m1)
2
(2n1)
2
. (61)
Here the sum can be computed exactly, giving
E
(2)
n
=
2
2

2
ML
2
1
n
2
. (62)
The resulting energy eigenvalue accurate to second order is
E
n


2

2
2ML
2
n
2
+

2
ML
2

2
2
2

2
ML
2
1
n
2
; n odd, (63)
and
E
n
=

2

2
2ML
2
n
2
; n even. (64)
10
3 The degenerate case
We now consider the case where some, but not all, of the unperturbed eigenvalues are the same. In this
situation, the non-degenerate formulas (22), (23), and (24) cannot be used because the E
mn
terms in the
denominator go to zero when levels m and n are degenerate. Our goal is still to nd the eigenvalues and
eigenstates of the full Hamiltonian
H = H
0
+V, (65)
where the bare eigenstates of H
0
are now given two quantum numbers, one which labels the energy level,
and the other which takes into account the degeneracy. Our bare basis is thus {|nm
(0)
}, where m labels
the degenerate sub-levels. They satisfy
H
0
|nm
(0)
= E
(0)
n
|nm
(0)
. (66)
In the case of degeneracy, the choice of basis which satised (59) is not unique. Provided that the pertur-
bation lifts the degeneracy, then a unique unperturbed basis is dened by
|nm
(0)
= lim
0
|nm. (67)
To see why this denes a unique basis, lets look at a very simple example. Consider a two-level Rabi
system, governed by the bare Hamiltonian H
0
= S
z
and the perturbation V = S
x
. For the case = 0, the
eigenstates of H
0
become degenerate, so that for every complex number z, we can dene a unique possible
basis choice via
|1
(0)
=
z|
z
+|
z

_
1 +|z|
2
, (68)
|2
(0)
=
|
z
z

|
z

_
1 +|z|
2
. (69)
On the other hand, for any non-zero , the eigenstates of H = H
0
+V are the S
x
eigenstates, given by
|1 =
|
z
|
z

2
, (70)
|2 =
|
z
+|
z

2
. (71)
Thus if we use |1
(0)
= lim
0
|1 = |1, and |2
(0)
= lim
0
|2 = |2, we have specied a unique bare
basis.
In fact, the good eigenstates will always be the eigenstates of V in the degenerate subspace. To prove this,
we start by introducing the projector into the degenerate subspace, I
n
, which clearly satises I
n
|nm
(0)
=
|nm
(0)
, I
n
H
0
= H
0
I
n
= I
n
H
0
I
n
= E
(0)
n
I
n
, and I
2
n
= I
n
, while for n

= n, we have I
n
|n

m = n

m|I
n
= 0.
The energy eigenvalue equation is
(H
0
+V E
nm
)|nm = 0. (72)
Since lim
0
|nm = |nm
(0)
, we can therefore write
lim
0
(H
0
+V E
nm
)|nm
(0)
= 0. (73)
11
Hitting this equation from the left with I
n
, and using |nm
(0)
= I
n
|nm
(0)
allows us to write the equation
as
lim
0
(I
n
V I
n
(E
nm
E
(0)
n
))I
n
|nm
(0)
= 0. (74)
Introducing V
n
= I
n
V I
n
, noting that lim
0
(E
nm
E
(0)
n
) = lim
0
E
(1)
nm
, and dividing both sides by
before taking 0 gives
(V
n
E
(1)
nm
)|nm
(0)
= 0, (75)
which tells us that the correct bare eigenstates are the eigenstates of V
n
, which satisfy V
n
|nm
(0)
=
v
nm
|nm
(0)
. Furthermore, we see that the rst-order energy shifts are just the corresponding eigenvalues,
E
(1)
nm
= v
nm
.
At this point, there are two ways to proceed. The rst is the straightforward method, we just take this
good basis and proceed with expanding the energy eigenvalue equation in powers of , and then solve the
equations associated with each power. The other method is conceptually trickier, but allows us to map
the degenerate problem onto the non-degenerate problem, and therefore use the previously derived results.
3.1 Approach #1
First, we will just approach the equations directly, one order at a time. The basic equation is now:
(H
0
E
(0)
n
)|nm
(j)
= V |nm
(j1)
+
j

k=1
E
(k)
nm
|nm
(jk)
, (76)
which together with the normalization constraint:
nm
(0)
|nm
(j)
=
1
2
j1

k=1
nm
(k)
|nm
(jk)
(77)
and the good basis condition:
nm
(0)
|V |nm
(0)
= v
nm

m
, (78)
allow us to compute the perturbed eigenstates and eigenvectors, dened via
E
nm
= E
(0)
n
+

j=1

j
E
(j)
nm
(79)
and
|nm =

j=0

j
|nm
(j)
. (80)
In the degenerate case, there are three possible actions we can perform on (76) to generate useable equations,
as opposed to two for the non-degenerate case. They are
I Hit with nm
(0)
| from the left
II Hit with nm
(0)
| from the left, where m

= m.
12
III Hit with n

m
(0)
| from the left, where n

= n.
The main thing we hope to prove is that the good basis condition is sucient to remove any singularities
due to vanishing energy denominators when degenerate levels are mixed.
3.1.1 First-order terms
With j = 1 we can start by computing nm
(0)
|nm
(1)
, giving
nm
(0)
|nm
(1)
=
1
2
0

k=1
nm
(k)
|nm
(jk)
= 0. (81)
The energy equation is then
(H
0
E
(0)
n
)|nm
(1)
= V |nm
(0)
+E
(1)
nm
|nm
(0)
. (82)
From action I, we get
0 = nm
(0)
|V |nm
(0)
+E
(1)
nm
, (83)
which gives
E
(1)
nm
= v
nm
. (84)
as expected.
From II, we get
0 = nm
(0)
|V |nm
(0)
(85)
which is just a re-statement of the good basis constraint (78).
From III, we nd
(E
(0)
n

E
(0)
n
)n

m
(0)
|nm
(1)
= nm
(0)
|V |nm
(0)
. (86)
Solving for n

m
(0)
|nm
(1)
gives
n

m
(0)
|nm
(1)
=
V
n

nm

n
, (87)
where we have introduced the new notation
n

n
= E
(0)
n

E
(0)
n
to avoid confusing the energy denominator
with the full energy level E
nm
.
At this point, we note that we have exhausted all of our equations, but have not found the components
nm
(0)
|nm
(1)
. We will, however, proceed to the case (j = 2), and hope for the best.
3.1.2 Second-order terms
With j = 2, we nd
nm
(0)
|nm
(2)
=
1
2
nm
(1)
|nm
(1)

13
=
1
2

m
(0)
|nm
(1)

2
=
1
2

=m

nm
(0)
|nm
(1)

1
2

=n
m

m
(0)
|nm
(1)

2
=
1
2

=m

nm
(0)
|nm
(1)

1
2

=n
m

|V
n

nm
|
2

2
n

n
, (88)
which we cannot evaluate further because the remaining coecients have not been determined.
The second-order eigenvalue equation is
(H
0
E
(0)
n
)|nm
(2)
= V |nm
(1)
+E
(1)
nm
|nm
(1)
+E
(2)
nm
|nm
(0)
(89)
From I, we obtain
0 = nm
(0)
|V |nm
(1)
+E
(2)
nm
, (90)
which leads to
E
(2)
nm
=

=n
m

V
nmn

m
n

m
(0)
|nm
(1)
+

V
nmnm
nm
(0)
|nm
(1)

=n
m

V
nmn

m
V
n

nm

n
(91)
where we have made use of (78) and (81) to eliminate the second term. This is essentially the same result as
that from non-degenerate perturbation theory, but with the singular terms excluded from the summation.
From action II on Eq. (89), we nd
0 = nm
(0)
|V |nm
(1)
+E
(1)
nm
nm
(0)
|nm
(1)
. (92)
Notice that this equation contains no second-order unknown quantities, but it does contain the remaining
rst-order unknown, nm
(0)
|nm
(1)
. Solving for this variable gives
v
nm
nm
(0)
|nm
(1)
=

=n
m

V
nm

m
n

m
(0)
|nm
(1)
+

V
nm

nm
nm
(0)
|nm
(1)

=n
m

V
nm

m
V
n

nm

n
+v
nm
nm
(0)
|nm
(1)
(93)
This then leads to
nm
(0)
|nm
(1)
=

=n
m

V
nm

m
V
n

nm
v
nm

n
, (94)
where we have introduced
v
nm

m
= v
nm
v
nm
. (95)
Intriguingly, we have obtained our remaining rst-order term by solving a second-order equation. That
the equation is second-order can be seen by the fact that there are two V s in the numerator. That the
14
resulting term is rst-order can be seen by the fact that there is a v in the denominator, so that the term
scales as

2

= .
This allows us to complete our expression for nm
(0)
|nm
(2)
, yielding
nm
(0)
|nm
(2)
=
1
2

=m

=n
m

=n
m

V
nmn

m
V
n

nm
V
nm

m
V
n

nm

n
v
2
nm

1
2

=n
m

V
nmn

m
V
n

nm

2
n

n
(96)
Lastly, from III, we obtain

n
n

m
(0)
|nm
(2)
= n

m
(0)
|V |nm
(1)
+E
(1)
nm
n

m
(0)
|nm
(1)
, (97)
which gives us
n

m
(0)
|nm
(2)
=

=n
m

V
n

m
V
n

nm

=m

=n
m

V
n

nm
V
nm

m
V
n

nm

n
v
nm

V
n

nm
v
nm

2
n

n
(98)
At this point we see that we have run out of second-order equations, yet have not obtained an expression
for the components nm
(0)
|nm
(2)
. It is logical at this point to assume that we can obtain these coecients
by solving a third-order equation, which is left as an exercise.
3.1.3 Results to second-order (almost)
Thus we have found
E
nm
= E
(0)
n
+v
nm

=n
m

V
nmn

m
V
n

nm

n
+O(
3
) (99)
as well as
|nm = |nm
(0)

_
1

2
2

=n
m

V
nmn

n
_
_
_
_

=m

=n
m

V
n

nm
V
nm

m
V
n

nm
v
2
nm

n
+
V
n

nm

n
_
_
_
_
+O(
3
)
_

_
+

=m
|nm
(0)

=n
m

V
nm

m
V
n

nm
v
nm

n
+O(
2
)
_

_
+

=n
m

|n

m
(0)

V
n

nm

n
_
1 +
v
nm

n
_
+
2

=n
m

_
_
V
n

=m
V
n

nm
V
nm

v
nm

m
_
_
V
n

nm

n
+O(
3
)
_

_
(100)
Note, in going from Eq. (98) to (100) it was necessary to make the change of dummy variables n

.
m

, and m

.
15
3.2 Approach #2
We begin by noting that the good bare eigenstates are eigenstates of H
0
+V
n
, with eigenvalues E
(0)
n
+
v
nm
. Armed with this, we can re-dene our bare Hamiltonian as

H
0
= H
0
+V
n
, and introduce the new
perturbation

V = (V V
n
). We will then use non-degenerate perturbation theory to solve
(

H +

V )|nm = E
nm
|nm, (101)
taking as our bare basis {|nm
(0)
}, with bare energies now given by E
(0)
nm
= E
(0)
n
+v
nm
. By setting = 1
at the end, we will recover the eigenstates of H = H
0
+ V . Note that even if some of the eigenvalues of
V
n
are degenerate, we can still use the non-degenerate theory, as the fact that nm
(0)
|

V |nm
(0)
= 0 will
kill all singular terms.
In using perturbation theory, we have to replace n with n, m, and the condition n

=n will go to n

=n &m

=m.
This is because we are elevating each |nm
(0)
state to the level of a non-degenerate state. After transforming
the indices in this manner, the second-order energy formula (32) become:
E
nm
= E
(0)
nm
+

V
nmnm

2

=n
d
n

=1
|

V
n

nm
|
2
E
n

nm

2
dn

=1
m

=m
|

V
nm

nm
|
2
E
nm

nm
+O(
3
), (102)
where E
n

nm
= E
(0)
n

E
(0)
nm
, and

V
n

nm
= n

m
(0)
|

V |nm
(0)
. Taking into account

V
nm

nm
= 0, this
simplies to
E
nm
= E
(0)
nm

2

=n
d
n

=1
|

V
n

nm
|
2
E
n

nm
+O(
2
) (103)
The next step is to set = 1, and express everything in terms of our original E
(0)
n
and V , giving
E
nm
= E
(0)
n
+v
nm

=n
d
n

=1

2
|V
n

nm
|
2

n
+v
n

nm
, (104)
where
n

n
= E
(0)
n

E
(0)
n
and v
n

nm
= v
n

m
v
nm
. As we only want the correct energies up to
O(
2
), we need to expand this expression in powers of and throw away terms of O(
3
) or higher. Since
(a +b)
1
=
1
a
(1
b
a
+
2 b
2
a
2
. . .), we see that keeping only terms up to second-order gives
E
nm
= E
(0)
n
+v
nm

2

=n
d
n

=1
|V
n

nm
|
2

n
+O(
3
). (105)
Thus Eq. (73) is the correct second-order expression for the energies in degenerate perturbation theory.
Similarly, translating the indices on the second-order state formulas (33) gives:
|nm = |nm
(0)

_
_
1

2
2

=n
d
n

=1
|

V
n

nm
|
2
E
2
n

nm
+O(
3
)
_
_
+

=n
d
n

=1
|n

m
(0)

_
_

V
n

nm
E
n

nm
+
2

=n
d
n

=1

V
n

V
n

nm
E
n

nm
E
n

nm
+O(
3
)
_
_
16
+
dn

=1
m

=m
|nm
(0)

_
_

=n
d
n

=1
V
nm

m
V
n

nm
E
nm

nm
E
n

nm
+O(
3
)
_
_
(106)
where we have again relied on

V
nm

nm
= 0 for simplications. Setting = 1 and expressing everything in
terms of V and E
(0)
n
gives
|nm = |nm
(0)

_
_
1
1
2

=n
d
n

=1

2
|V
n

nm
|
2
E
n

n
+(v
n

m
v
nm
)
_
_
+

=n
d
n

=1
|n

m
(0)

_
_

V
n

nm

n
+v
n

nm
+

=n
d
n

=1

2
V
n

m
V
n

nm
(
n

n
+v
n

nm
)(
n

n
+v
n

nm
)
_
_
+
dn

=1
m

=m
|nm
(0)

=n
d
n

=1

2
V
nm

m
V
n

nm
v
nm

nm
(
n

n
+v
n

nm
)
. (107)
The next step is to expand in powers of and keep only terms up to O(
2
), which gives
|nm = |nm
(0)

_
_
1
1
2

=n
d
n

=1

2
|V
n

nm
|
2

n
_
_
+

=n
d
n

=1
|n

m
(0)

_
_

V
n

nm

n
+

2
V
n

nm
v
n

nm

2
n

n
+

=n
d
n

=1

2
V
n

m
V
n

nm

n
_
_
+
dn

=1
m

=m
|nm
(0)

=n
d
n

=1
V
nm

m
V
n

nm
v
nm

n
. (108)
Notice that something very strange has occurred with the nal term. This term, which was originally
second order in , has become rst-order in . This means that a second-order calculation was necessary
to obtain the correct state to rst-order. Thus we can predict that a third-order calculation will similarly
yield second-order terms. This means that we can actually only trust our states (76) up to rst-order.
Dropping all terms greater than rst-order gives
|nm = |nm
(0)

_
1 +O(
2
)
_
+

=n
d
n

=1
|n

m
(0)

V
n

nm

n
+O(
2
)
_
+
dn

=1
m

=m
|nm
(0)

_
_

=n
d
n

=1
V
nm

m
V
n

nm
v
nm

n
+O(
2
)
_
_
. (109)
Equations (73) and (77) are the main results of this section, giving the energies to second-order and the
states to rst-order.
17
3.3 Examples
A three level system: The simplest quantum system which can require degenerate perturbation theory
is the three-state system. If a two-state system is degenerate, it must rst be diagonalized in the degenerate
subspace, after which the problem is solved exactly. Let us consider, therefore, a three-level system with
H
0
E
0
_
_
0 0 0
0 1 0
0 0 1
_
_
, (110)
and
V V
0
_
_
0

2 2

2 0 2
2

2 2 0
_
_
, (111)
both dened in the physical basis {|1, |2, |3}. The bare Hamiltonian has two energy eigenvalues E
(0)
1
= 0
and E
(0)
2
= E
0
. The rst eigenvalue is non-degenerate, with eigenvector |1, 1
(0)
= |1, while the second is
doubly degenerate, with a degenerate subspace spanned by |2 and |3. The projection operator for the
degenerate subspace is
I
2
= |22| +|33|
_
_
0 0 0
0 1 0
0 0 1
_
_
. (112)
This leads to the projection
V
2
= I
2
V I
2
= V
0
_
_
0 0 0
0 0 2
0 2 0
_
_
. (113)
This operator has eigenvalues 0, 2V
0
, and 2V
0
, with corresponding eigenvectors |1,
1

2
(|2 |3), and
1

2
(|2 + 3). To determine which two eigenvectors live in the degenerate subspace (pretending for the
moment that it is not completely obvious), we can apply I
2
to each, and discard any states which give
zero. Applying this test, we nd that the two good basis states for the degenerate subspace are |2, 1
(0)
=
1

2
(|2 |3) and |2, 2
(0)
=
1

2
(|2 +|3).
In the new basis {|1, 1
(0)
, |2, 1
(0)
, |2, 2
(0)
}, the bare Hamiltonian matrix is unchanged, and the perturba-
tion is diagonalized in the degenerate subspace giving
V
_
_
V
1111
V
1121
V
1122
V

1121
2V
0
0
V

1122
0 2V
0
_
_
. (114)
Computing the necessary matrix elements, we nd
V
1111
= 1, 1
(0)
|V |1, 1
(0)
= 1|V |1 = 0, (115)
V
1121
= 1, 1
(0)
|V |2, 1
(0)
=
1

2
(1|V |2 1|V |3) = V
0
, (116)
and
V
1122
= 1, 1
(0)
|V |2, 2
(0)
=
1

2
(1|V |2 +1|V |3) = 3V
0
, (117)
18
so that in the new basis,
V V
0
_
_
0 1 3
1 2 0
3 0 2
_
_
. (118)
We can now readily evaluate Eqs. (73) and (77), giving
E
11
= E
(0)
1
+v
11

2
_
|V
2111
|
2
E
21
+
|V
2211
|
2
E
21
_
+O(
3
)
=
2
10V
0
E
0
+O(
3
), (119)
E
21
= E
(0)
2
+v
21

2
|V
1121
|
2
E
12
+O(
3
)
= E
0
2V
0
+
2
V
2
0
E
0
+O(
3
), (120)
and
E
22
= E 2
(0)
+v
22

2
|V
1122
|
2
E
12
= E
0
+2V
0
+
2
9V
2
0
E
0
. (121)
For the rst-order states, we nd
|1, 1 = |1, 1
(0)
+
_
|2, 1
(0)

V
2111
E
21
+|2, 2
(0)

V
2211
E
21
_
+O(
3
)
= |1, 1
(0)
|2, 1
(0)

V
0
E
0
+|2, 2
(0)

3V
0
E
0
+O(
3
)
= |1 +|2

2V
0
E
0
+|3
2

2V
0
E
0
+O(
3
), (122)
|2, 1 = |2, 1
(0)
+|2, 2
(0)

V
2211
V
1121
v
2221
E
12
+|1, 1
(0)

V
1121
E
12
+O(
3
)
= |1, 1
(0)

V
0
E
0
+|2, 1
(0)
+|2, 2
(0)

3V
0
4E
0
+O(
3
)
= |1
V
0
E
0
+|2
1

2
_
1 +
3V
0
4E
0
_
+|3
1

2
_
1 +
3V
0
4E
0
_
+O(
3
), (123)
and
|2, 2 = |2, 2
(0)
+|2, 1
(0)

V
2111
V
1122
v
2122
E
12
+|1, 1
(0)

V
1122
E
21
+O(
3
)
= |1, 1
(0)

3V
0
E
0
|2, 1
(0)

3V
0
4E
0
+|2, 2
(0)
+O(
3
)
= |1
3V
0
E
0
+|2
1

2
_
1
3V
0
4E
0
_
+|3
1

2
_
1 +
3V
0
4E
0
_
+O(
3
). (124)
19
4 Practical Iterative Approach
While some energy was spent deriving analytic expressions for second-order and third-order terms, keep
in mind that in practice, the terms would be computed iteratively. In other words, instead of expressing
the j
th
-order terms in terms of zero
th
order terms, as is customary for terms up to second-order, it is more
ecient to express the j
th
-order terms in terms of (j 1)
th
order terms only, and just iterate the equations.
4.1 Non-degenerate iterative equations
For the non-degenerate case, the iterative equation (adopting uniform notation) for the energy level shifts
is
E
(j)
n
=

V
nn
n
(0)
|n
(j1)

j2

k=1
E
(k)
n
n
(0)
|n
(jk)
. (125)
For the expansion coecients, the iterative equations are
n
(0)
|n
(j)
=
1
2
j1

k=1
n
(k)
|n
(jk)
, (126)
and for n

= n,
n
(0)
|n
(j)
=

V
n

n
n
(0)
|n
(j1)
+
j2

k=1
E
(k)
n

n
n
(0)
|n
(jk)
. (127)
4.2 Degenerate iterative equations
Assuming that the correct bare eigenstates are in use, we have
E
(j)
nm
= v
nm
nm
(0)
|nm
(j1)
+

=n
m

V
nmn

m
n

m
(0)
|nm
(j1)

j2

k=1
E
(k)
nm
nm
(0)
|nm
(jk)
, (128)
and
nm
(0)
|nm
(j)
=
1
2
j1

k=1
nm
(k)
|nm
(jk)
, (129)
and for n

= n,
n

m
(0)
|nm
(j)
=

,m

V
n

n
n

m
(0)
|nm
(j1)
+
j1

k=1
E
(k)
nm

n
n

m
(0)
|nm
(jk)
. (130)
For n

= n, but m

= m, we have
nm
(0)
|nm
(j)
=

=n
m

V
nm

v
nm

m
n

m
(0)
|nm
(j)
+
j

k=2
E
(k)
nm
v
nm

m
nm
(0)
|nm
(jk+1)
(131)
20