Lecture 7: Stationary Perturbation Theory

In most practical applications the time independent Schrödinger equation

Hψ = Eψ (1)

cannot be solved exactly and one has to resort to some scheme of finding approximate solutions,
preferably by some method of iteration that allows one, at least in principle, to find the solution
with any desired accuracy. One of the most widely used such methods is perturbation theory.
Perturbation theory is applicable if there is a possibility of separating the Hamiltonian H
into a sum of H0 , say, for which one is able to find the exact solution of the corresponding
Schrödinger equation, and a term H 0 , the perturbation. The latter term must be in some sense
small for the method to work. The precise meaning of the smallness needed to make the method
useful has to be investigated in each case separately. However, in order to have a general picture
of this meaning one could think of the energy levels of a hydrogen atom perturbed by some
external force: if the shift of the energy levels is small compared with their separation in the
unperturbed state, then perturbation theory will be applicable.

1. The non-degenerate case

Perturbation theory proceeds somewhat differently in the presence or in the absence of de-
generacy. We shall consider in this section only the case of non-degenerate perturbation theory.
This will allow us to apply the method to most one-dimensional cases. Three-dimensional sys-
tems will usually have some degree of degeneracy; this will be taken into account in the next
section.
Thus we assume that the Hamiltonian can be written in the form

H = H0 + λV (2)

where I have written λV for H 0 . The parameter λ can be thought of as some quantity which
can be varied at will and can be reduced to zero, in which case the perturbation disappears and
all results must revert to the unperturbed case. If we consider the case of a hydrogen spectrum
which we observe with the atom placed in a magnetic field (Zeeman effect) or in an electric
field (Stark effect), then λ can be thought of as characterising the strength of the field over
which we have complete control.
The unperturbed Hamiltonian H0 is supposed to have been chosen such that the solution
of the corresponding Schrödinger equation is known: 1

H0 φn = En(0) φn (3)

with known eigenvalues En(0) and known eigenfunctions φn which form a complete set of or-
thonormal functions:
(φm , φn ) = δmn . (4)
1
The notation used here suggests the presence of discrete levels only. The modifications to include continuous
states are straightforward (see, e.g. L.I. Schiff, Quantum Mechanics).

1
 Before actually applying perturbation theory we shall rewrite the Schrödinger equation in
the form of a matrix equation. This can be done by using the completeness of the set {φn } to
express the nth eigenfunction of H in the form
X
ψn = cnk φk . (5)
k

Here the expansion coefficients are functions of λ: we must have

cnk = δnk for λ = 0 (6)

to ensure that ψn = φn for λ = 0.

If we now substitute (5) into (1) and take the scalar product with φm , then we get
(0)
X
(En − Em )cnm = λ cnk Vmk (7)
k

where Vmk = (φm , V φk ). Equation (7) is the Schrödinger equation in matrix form and is
equivalent to Eq. (1). Mathematically it is a linear simultaneous equation with the expansion
coefficients cnm and the energy eigenvalue En as the unknown quantities. If the system under
consideration had only a small number of energy levels one could solve this equation without
any approximations by standard methods of linear algebra.

Exercise: Assuming that the system under consideration has only two energy levels, show that
the eigenvalues are
s
1 1
E1,2 = (H11 + H22 ) ± (H11 + H22 )2 + |H12 |2
2 4
where Hmn = (φm , Hφn ).

In most cases of practical interest the number of eigenstates is large or even infinite. Then
the exact solution of Eq. (7) becomes either impractical or impossible. In these cases it may
be possible to find the solution by iteration using perturbation theory.
The essence of perturbation theory is to expand the energy eigenvalue and the eigenfunction
into a power series in λ 2
En = En(0) + λEn(1) + λ2 En(2) + ... (8a)
(1) (2)
cnk = δnk + λcnk + λ2 cnk + ... (8b)
Note that Eq. (8b) is equivalent with

ψn = φn + λψn(1) + λ2 ψn(2) + ... (8c)

2
Many authors, e.g. P.A.M. Dirac, The Principles of Quantum Mechanics, do not use such an expansion
parameter but assume each successive term to be of smaller order than the preceeding term. L.I. Schiff, Quantum
Mechanics, uses an expansion parameter λ but does not associate with it any physical significance: it appears
as an auxiliary variable which is set equal to one at the end of the calculation. It is strongly recommended to
study the approaches taken by different authors to see how many different ways lead to the same result.

2
 (i)
where ψn(i) = k cnk φk , i = 1, 2, ....
P

Substitution of Eqs. (8) into (7) yields

(En(0) − Em
(0)
+ λEn(1) + λ2 En(2) + ...) × (δnm + λc(1) 2 (2)
nm + λ cnm + ...)
(1) (2)
(δnk + λcnk + λ2 cnk + ...)Vmk
X
= λ
k

and comparing powers of λ we get the following set of equations:

(En(0) − Em
(0)
)δnm = 0 (9a)

(En(0) − Em
(0) (1)
)cnm + En(1) δnm = Vmn (9b)
(1)
(En(0) − Em
(0) (2)
)cnm + En(1) c(1) (2)
X
nm + En δnm = cnk Vmk (9c)
k
(2)
(En(0) − Em
(0) (3)
)cnm + En(1) c(2) (2) (1) (3)
X
nm + En cnm + En δnm = cnk Vmk (9d)
k
(0)
etc. Equation (9a) contains no new information: it is the statement that Em 6= En(0) for m 6= n.
From Eq. (9b) we get En(1) by putting m = n:

En(1) = Vnn (10)

and we get c(1)

nm by putting m 6= n:

Vmn
c(1)
nm = (0) (0) (m 6= n) (11)
En −Em

The coefficient c(1)

nn remains undefined by Eqs. (9) and will be chosen by the requirement that
the function ψn = φn + λψn(1) be normalised to the 1st order in λ. Thus, using Eqs. (8) and
(4), we get

(ψn , ψn ) = (φn , φn ) + λ[(ψn(1) , φn ) + (φn , ψn(1) )] + O(λ2 ) = 1 + λ(cnn + c∗nn ) + O(λ2 ) = 1

and it is evidently sufficient to put

c(1)
nn = 0 (110 )
to ensure the normalisation of ψn to 1st order in λ. This completes the solution of Eq. (1) to
first order of perturbation theory.
To find the 2nd order corrections we use Eq. (9c). Setting m = n and using Eqs. (11) we
get
Vkn Vnk |Vnk |2
En(2) = = (12)
P P
k6=n E (0) −E (0) k6=n E (0) −E (0)
n k n k

where in the last step we have used the hermiticity of V . The 2nd order correction to the wave
function can be found by putting m 6= n in (9c), and having found the 2nd order correction
one can proceed to find the 3rd order correction using (9d), etc. So, at least in principle the
solution can be found to any desired accuracy by successively increasing the number of terms.
An important result of our derivation is that the 1st order correction En(1) is linear in the
perturbation, the 2nd order correction En(2) is quadratic in the perturbation and so on. This

3
implies at least the possibility that for sufficiently small perturbations the series will converge.
However, it happens that even in cases where the perturbation series is known to diverge one
gets sometimes good agreement with experimental results if one takes only the lowest order
term (cf. P.A.M. Dirac, Principles of Quantum Mechanics, Section 42).
We summarise our results by writing down the energy eigenvalue En of the perturbed system
up to the 2nd order of perturbation theory:

|Hnk0 |2
En = En(0) + Hnn
0
+ (13)
P
k6=n E (0) −E (0)
n k

where H 0 = λV is the perturbation.

Exercise:
p̂2 1
1. A one dimensional harmonic oscillator with Hamiltonian Ĥ0 = + m ω 2 x2 is per-
2m 2
turbed by an additional potential energy λx. Find the energy eigenvalues to second order
in the perturbation, given the eigenvalues of the energy of the unperturbed harmonic
oscillator En = h̄ω(n + 21 ), and
s
h̄ √ √ 
(un , xuk ) = n + 1 δk,n+1 + n δk,n−1
2mω
where un (x) are the eigenfunctions of the unperturbed harmonic oscillator.

2. Show that the same result can be obtained without the use of perturbation theory by
solving the eigenvalue problem of the perturbed harmonic oscillator exactly.

3. (i) If the harmonic oscillator is perturbed by an additional potential energy V1 = g x3 ,

discuss the physical justification of using stationary perturbation theory to find the
perturbed energy levels. What is the lowest value of the total energy at which
perturbation theory is certainly going to fail?
(ii) Assuming that the use of perturbation theory is justified, explain the procedure of
calculating the perturbative correction, to which the x3 term gives rise, in the lowest
non-vanishing order.

4. Explain why the use of stationary perturbation theory is justified in the case of a pertur-
bation of the harmonic oscillator by V (x) = ax3 + bx4 and calculate the correction to the
energy in the lowest non-vanishing order.
Answer. The energy shift in the lowest non-vanishing order is
!2
3b h̄ 15 2 h̄2 11
 
(2n2 + 2n + 1) − a 3 4 n2 − n +
4 mω 4 m ω 30

Note that this result implies that the correction due to the x3 term is of 2nd order whereas
that due to the x4 term is of 1st order.

4
5. Calculate the energy shift of the ground state energy of a hydrogen-like atom (or ion)
caused by the finite size of the nucleus, assuming that the charge of the nucleus is uni-
formly distributed over its volume.
Answer. The P.E. of the electron in the field of a nucleus of charge Ze and radius R,
assuming uniform charge distribution, is given by

2

r2

− Ze

2R
3− R2
r≤R
V (r) = Ze 2
 − r≥R
r

2
and if we choose the unperturbed Hamiltonian to be H0 = T̂ + Vc (r), with Vc (r) = − Zer ,
then the ground state eigenfunction of H0 is the well known hydrogen wave function
1 −r/a
ψ0 = √ e , where a = aBohr /Z and aBohr = 5.3 × 10−11 m is the Bohr radius.
πa 3
(0) 1
The corresponding ground state energy is E1 = − (Zα)2 me c2 . The perturbation then
2
becomes (
V (r) − Vc (r) for r ≤ R
H0 =
0 for r ≥ R
(1)
The 1st order correction is given by E1 = (ψ0 , H 0 ψ0 ) and we note that the integration
over the polar angles gives just a factor of 4π leaving us with the integral over r. This
integral can be simplified if one realises that the upper limit of integration is the nuclear
radius R which even for the heaviest nuclei does not exceed ∼ 10−14 m. This implies that
over the entire range of r the exponential function is to a good approximation equal to
one. The remaining integration is straightforward and yields
2
R

(1) (0)
E1 = 0.8|E1 |
a
(1) (0)
Thus for hydrogen one finds E1 /|E1 | ∼ 10−9 , for oxygen (Z = 8) one gets ∼ 4 × 10−7 ,
and for uranium (Z = 92) ∼ 3×10−4 , where for the order-of-magnitude estimates we have
used a well known result from nuclear physics, R = R0 A1/3 , R0 = 1.3 × 10−15 m, together
with the approximate relation A ≈ 2Z between mass number A and atomic number Z.

6. Calculate the shift of the ground state energy of the hydrogen atom in a uniform electric
field (Stark effect).
Answer. The perturbation to the Hamiltonian of the hydrogen atom is in this case H 0 =
eEz if E is the electric field strength and we assume that the field is in the z direction.
The 1st order correction vanishes by symmetry. The 2nd order correction is given by

(2) X |(φnlm , H 0 φ100 )|2

E1 = (0) (0)
n>1,lm E1 − E n

Recall that the sum has to be extended over the complete set of eigenfunctions. In
the present case the complete set of eigenfunctions includes the eigenfunctions of the
continuum states for which the summation over n must be replaced by an integration

5
 over the energy E. But such a direct approach leads to an intractable mathematical
problem. A method that leads to the exact result is described in L.I. Schiff, Quantum
Mechanics. Evaluation of the above sum gives the result
∞
(2) n2
E1 = −2a3Bohr E 2
X
2
f (n)
n=2 n − 1

28 n7 (n − 1)2n−5
where f (n) = . Convergence of the above series is poor and one has to
3 (n + 1)2n+5
take a large number of terms for a reasonable approximation. Thus one finds
(2)
E1 = −1.832a3Bohr E 2

which is about 20% too small on account of the neglected continuum states. In spite
of the numerical deficiency of our result it is not without interest physically: it tells us
that the hydrogen atom in an electric field acquires an induced dipole moment d that is
proportional to the external electric field:
(2)
∂E
d=− 1 ≈ −4a3Bohr E
∂E

2. Perturbation theory in the presence of Degeneracy

A problem additional to the one solved in the previous section arises if the energy level, for
which we want to calculate the effect of the perturbation, is degenerate. Let us express the
degeneracy of the nth level of the unperturbed system by writing the wave functions which
correspond to the energy eigenvalue En(0) in the form

φnα α = 1, ..., f (14)

The label α has to be understood in a generic sense. A well known example of degeneracy is
found in the stationary states of the hydrogen atom which are characterised by three quantum
numbers: the principal quantum number n and the angular momentum quantum numbers l
and m. In this case α represents both l and m.
With the above notation the Schrödinger equation of the unperturbed system is

H0 φnα = En(0) φnα (15)

and the eigenfunctions form a complete set of orthonormal functions:

(φmα , φnβ ) = δnm δαβ . (16)

The characteristic novel feature that arises in the case of perturbation of a degenerate level
is that the perturbation must be expected to have a different effect on each of the f different
states φnα . Therefore the perturbed energies must also be distinguished by the label α. We

6
take this into account by writing the perturbation expansions corresponding to Eq. (8) in the
form
Enα = En(0) + λa1α + λ2 a2α + ...
(0) (1)
ψnα = ψnα + λψnα + λ2 ψnα
(2)
+ ... (17)
where for simplicity we have suppressed the label n on the corrections a.
Now, if the perturbation is switched off (λ → 0) then by virtue of Eq. (17) the energy
correctly takes on the corresponding value of the unperturbed system, but the wave function
(0)
becomes ψnα , and ab initio we do not know whether this will be any one of the unperturbed
states φnα or possibly some linear combination of these states. We must therefore consider the
most general possibility, namely that
(0)
X
ψnα = cαβ φnβ . (18)
β

Thus, in the case of degeneracy we have the additional problem to find the superposition
coefficients cαβ . We shall now show that this problem is solved automatically together with the
problem of finding the 1st order corrections a1α .
Let us write down the eigenvalue problem for the Hamiltonian H = H0 + λV in the form
(Enα − H0 )ψnα = λV ψnα (19)
and substitute the expansions (17). Then we get
(En(0) − H0 + λa1α + ...)(ψnα
(0) (1)
+ λψnα (0)
+ ...) = λV (ψnα (1)
+ λψnα + ...) (20)
and comparing powers of λ we get a set of equations similar to Eqs. (9). The first one of these
is again the trivial statement and need not be mentioned any more. Of particular interest in
the present case is only the equation that arises from the terms linear in λ:
(En(0) − H0 )ψnα
(1) (0)
+ a1α ψnα (0)
= V ψnα . (21)
Taking the scalar product with φnγ and using Eq. (18) yields
Pf (n)
a1α cαγ = β=1 cαβ Vγβ (22)

(n)
where Vγβ = (φnγ , V φnβ ), and since α is fixed and γ can take on f different values Eq.
(22) represents a system of f simultaneous linear homogeneous equations for the f unknown
coefficients cαβ . The 1st order energy shift a1α here plays the role of an eigenvalue. The
characteristic equation of (22) is a polynomial of order f in a1α so that we can expect up to
f different roots, each root corresponding to one value of α. If all roots are different one says
that the degeneracy is completely removed in 1st order. Often the degeneracy is only partly
removed in 1st order, i.e. not all roots of the characteristic equation are different.
Let us consider in detail the particular case of two-fold degeneracy, f = 2. Then Eq. (22)
takes on the form
a1α cα1 = V11 cα1 + V12 cα2
a1α cα2 = V21 cα1 + V22 cα2 (23)

7
which has the characteristic equation

a21α − T a1α + ∆ = 0

where T and ∆ are the trace and the determinant of the matrix Vαβ , respectively:

T = (V11 + V22 ), ∆ = V11 V22 − V12 V21

The two roots of the characteristic equation are

s
2
1 1

a1α = T± T − ∆.
2 2
Thus the degeneracy is removed in 1st order unless T 2 − 4∆ = 0. The separation of the split
energy levels is given by
q
∆a1 ≡ a1+ − a1− = (V11 − V22 )2 + 4|V12 |2 .

The split energy levels correspond to the zeroth-order wave functions

ψ± = c1± φ1 + c2± φ2

where the coefficients must be found from the simultaneous equations

(V11 − a1 )c1 + V12 c2 = 0

V12 c1 + (V22 − a1 )c2 = 0

together with the normalisation condition

|c1 |2 + |c2 |2 = 1.

After straightforward if somewhat tedious algebra one can get the coefficients in the following
form:
  1/2
 V12  V11 − V22 
c1± = 1± q 
 2|V12 |
(V11 − V22 )2 + 4|V12 |2 
  1/2
V21 
 V11 − V22 
c2± = ± 1∓ q 
2|V12 | (V11 − V22 )2 + 4|V12 |2 

Note that only the relative phase between c1± and c2± is fixed by the simultaneous equations;
the overall phase is arbitrary and was chosen such as to give the resulting formulæ a particularly
symmetric form.

Exercise: Calculate the splitting of the 1st excited state of the hydrogen atom in a uniform
electric field in 1st order of degenerate perturbation theory (Stark effect).

Answer. The 1st excited state corresponds to the principal quantum number n = 2; there are
four degenerate states with angular momentum quantum numbers l = 0, 1, m = −l, −l+1, ..., l.

8
The characteristic equation is a quartic equation in the 1st order energy correction which must
be computed from a 4 × 4 determinant. This seems a rather tedious task. However, with a
few very general considerations one can see that there is only one non-zero matrix element:
V01 = −3eEaBohr , and that the characteristic equation reduces to the simple form

a21 (a21 − |V01 |2 ) = 0

whose four roots are 0, 0, and ± |V01 |. Thus there is one two-fold degenerate unshifted level
and two non-degenerate levels shifted by equal√amounts up and down. The wave function
corresponding√to the upper level is (φ200 − φ210 )/ 2, and the wave function of the lower level is
(φ200 + φ210 )/ 2; the unshifted state remains some linear superposition of the states φ211 and
φ21−1 .
To derive these results consider the general matrix element (φ2lm , V φ2l0 m0 ) where the per-
turbation Hamiltonian is V = eEz. This Hamiltonian has cylindrical symmetry about the z
axis and therefore commutes with the angular momentum operator Lz . 3 Thus

(φ2lm , [Lz , V ]φ2l0 m0 ) = h̄(m − m0 )(φ2lm , V φ2l0 m0 ) = 0,

i.e.
(φ2lm , V φ2l0 m0 ) = 0 unless m − m0 = 0.
Next we note that
(φ2lm , V φ2lm ) = 0
by parity, which leaves only the matrix elements (φ2lm , V φ2l0 m ) with l0 6= l to calculate. But
there are only two matrix elements: V01 and V10 , and they are related by Hermiticity of the
Hamiltonian: V01 = V10∗ , i.e. effectively we need to evaluate only the matrix element V01 . To
do this we must substitute the hydrogen wave functions ψ200 (r, θ, φ) = R20 (r)Y00 (θ, φ) and
ψ210 (r, θ, φ) = R21 (r)Y10 (θ, φ) and carry out the integrations, hence
Z
V01 = eE d3 rR20 (r)Y00 (θ, φ)zR21 (r)Y10 (θ, φ)
= −3eEaBohr .

3
Formally one can see this by noting that z = r cos θ and Lz = −ih̄ ∂/∂φ.