16 The Variation Method

Here we will discuss a method which allows us to approximate the ground

state energy of a system without solving the Schrödringer equations. This
method is the known as the variation method and is based on the following
theorem
The variation Theorem For any normalized acceptable function φ
!
φ∗ Ĥφdτ ! E0 (578)

where E0 is the lowest eigenvalue of Ĥ. The variation theorem allows us to

calculate an upper bound for the system’s ground state energy. The prove the
theorem first expand the function φ in terms of the complete, orthonormal
set of eigenfunctions of Ĥ "
φ= ak ψk (579)
k
where
Ĥψk = EK ψk (580)
Substitution of this give
! ! " " ! " "
∗ ∗ ∗
φ Ĥφdτ = ak ψk Ĥ aj ψj dτ = ak ψk∗ aj Ĥψj dτ (581)
k j k j

using the eigenvalue equation and interchange the integration and summa-
tions gives
! ! " "
∗
φ Ĥφdτ = ak ψk∗ Ek aj ψj dτ (582)
k j
"" !
= a∗k aj Ej ψk∗ ψj dτ (583)
k j
""
= a∗k aj Ej δkj (584)
k j
" "
= a∗k ak Ek = |ak |2 Ek (585)
k k
2
Since Ek ! E0 and#|ak | is always positive we have |ak |2 Ek ! |ak |2 E0 for
every k which give k |ak |2 Ek ! k |ak |2 E0 , we therefore have
#
! " " "
φ∗ Ĥφdτ = |ak |2 Ek ! |ak |2 E0 = E0 |ak |2 (586)
k k k

70
Since φ is normalized we have
!
1 = φ∗ φdτ (587)
! "
= a∗k ψk∗ aj ψj dτ (588)
k
"" !
= a∗k aj ψk∗ ψj dτ (589)
k j
""
= a∗k aj δkj (590)
k j
"
= |ak |2 (591)
k

Therefore !
φ∗ Ĥφdτ ! E0 (592)

What if the wavefunction is not normalized. In this case we have

!
2
|N| φ∗ Ĥφdτ ! E0 (593)

where we know that

! !
2 ∗ 2
1 = |N| φ φdτ, |N| = 1/ φ∗ φdτ (594)

this gives us $ ∗
φ Ĥφdτ
$ ! E0 (595)
φ∗ φdτ
Therefore, if we can express our trial wavefunction in terms of a set of pa-
rameters {αi } we can find the optimum wavefunction by minimizing the
expection valued with respect to the parameters

∂ !E(α1 , α2 , · · · , αn )"
= 0, i = 1, 2, · · · , n (596)
∂αi

71
16.1 Variational treatment of the Hydrogen-like atom
Find the optimum form of a trial function ψ = exp(−kr) and the upper
bound to the ground state energy of the Hydrogen-like atom. The Hamilto-
nian of the system is
1 Z
Ĥ = − ∇2 − (597)
2 r
First we need to evaluate the normalization integral
! ∞
2! π
!
∗
ψ ψdτ = 4π exp(−2kr)r 2 dr = 4π = (598)
0 (2k)3 k3
Next we need to evaluate the repulsion integral
! ∞
∗1 1! π
!
ψ ψdτ = 4π exp(−2kr)rdr = 4π 2
= 2 (599)
r 0 (2k) k
and finally the kinetic energy integral
1 d d
! !
∗ 2
ψ ∇ ψdτ = ψ ∗ ( 2 r 2 ) exp(−kr)dτ (600)
r dr dr
1 d
!
= ψ ∗ ( 2 r 2 )(−k) exp(−kr)dτ (601)
r dr
1
!
= ψ ∗ ( 2 (−2kr + k 2 r 2 ) exp(−kr)dτ (602)
r
2k
!
= (k 2 − ) exp(−2kr)dτ (603)
r
2k
!
= 4π (k 2 − )r 2 exp(−2kr)dτ (604)
r
2! 1 π
= 4π(k 2 − 2k ) = − (605)
(2k)3 (2k)2 k
Therefore the variational ration is
π/2k − Zπ/k 3
!E" = = k 2 /2 − Zk (606)
π/k 3
To find the minimum we differentiate with respect to k
d !E"
=k−Z =0 (607)
dk
72
which gives k = Z. Therefore, the minimum ground state energy is

Z2
!E" = Z 2 /2 − Z 2 = − (608)
2
and the wavefunction is
√
φ= Z 3 π exp(−Zr) (609)

which is the exact solution for the Hydrogen-like atom.

16.2 Variational treatment of the Helium atom

Lets find using the variational method the optimum form of the wavefunction
for the Helium atom in the 1s2 configuration. In order to represent the
correct screening of the nuclei due the second electron we will replace Z with
a variable parameter ζ as
%
# ζ3
1s (1) = exp(−ζr1 ) (610)
π
Our nomalized trial function of the Helium atom is then given by

φ(1, 2) = 1s# (1)1s# (2) (611)

The average energy is

! ! ! !
∗
!E" = φ (1, 2)Ĥφ(1, 2)dτ1dτ2 = 1s# (1)1s#(2)Ĥ1s# (1)1s# (2)dv1 dv2
(612)
due to the integration over spin is normalized and Ĥ is spin-less. The Hamil-
tonian of the system is
1
Ĥ = ĥ(1) + ĥ(2) + (613)
r12
The integrals we need to evaluate is then
! ! !
# # # #
1s (1)1s (2)ĥ(1)1s (1)1s (2)dv1 dv2 = 1s# (1)ĥ(1)1s# (1)dv1(614)

= (ζ 2 /2) − 2ζ (615)

73
Which can be realized by rewritten the one-electron Hamiltonian as
1 2 1 ζ ζ −2
ĥ(1) = − ∇21 − = − ∇21 − + (616)
2 r1 2 r1 r1
where the first two parts gives a Hamiltonian which together with the 1s# is
2
the same as the hydrogen-like Schrödringer equation with eigenvalues E = ζ2 .
Similarly, for the integral over the second electron
! ! !
# # # #
1s (1)1s (2)ĥ(1)1s (1)1s (2)dv1 dv2 = 1s# (2)ĥ(2)1s# (2)dv1(617)

= (ζ 2 /2) − 2ζ (618)

The last integral we have already seen the solution to

1 5
! !
1s# (1)1s# (2) 1s# (1)1s# (2)dv1 dv2 = ζ (619)
r12 8
The average energy is therefore

ζ2 5 27
!E" = 2[ − 2ζ] + ζ = ζ 2 − ζ (620)
2 8 8
Minimizing the energy with respect to ζ gives

d !E" 27
= 2ζ − =0 (621)
dζ 8
which gives
27
ζ= = 1.6875 (622)
16
which is smaller than 2 as expected. Lets check the average energy. For ζ = 2
the average enegy give -2.75 a.u. and for ζ = 1.6875 we get the energy of
-2.8477 a.u. The exact results is 2.9037 a.u. The un-optimized wavefunction
therefore finds 95% of the total energy whereas the optimized wavefunctions
find about 98% of the energy. However, the energy difference is still around
35 kcal/mol in error!

74
16.3 The linear variation method
Lets assume that we have selected an appropriate set of basis functions {φj }
which are linear independent but not necessarily othonormal. We can then
expand our trial wavefunction in a finite number of basis functions as
n
"
ψ= cj φ j (623)
j

where the coefficients cj are parameters to be determined by minimizing the

variational integral. We will restrict the basis functions and coefficients to
be real. The variation integral is then
$ ∗
ψ Ĥψdτ
!E({φj , cj })" = $ ∗ (624)
ψ ψdτ

The normalization integral is

! n
! " n
" " n
n " ! n "
" n
∗
ψ ψdτ = cj φ j ck φk dτ = cj ck φj φk dτ = cj ck Sjk
j k j k j k
(625)
where we define the overlap integral Sjk as
!
Sjk = ψ ∗ ψdτ (626)

SImilarly, the expectation value integral is

! n
! " n
" n "
" n ! n "
" n
∗
ψ Ĥψdτ = cj Ĥφj ck φk dτ = cj ck φj Ĥφk dτ = cj ck Hjk
j k j k j k
(627)
where we have introduced the matrix elements of the Hamiltonian
!
Hjk = φj Ĥφk dτ (628)

The variational integral is therefore

$ ∗ #n #n
ψ Ĥψdτ j k cj ck Hjk
!E({φj , cj })" = Ē = $ ∗ = #n #n (629)
ψ ψdτ j k cj ck Sjk

75
or n "
n n "
n
" "
Ē cj ck Sjk = cj ck Hjk (630)
j k j k

We will not minimize the energy with respect to the ci coefficients

∂ Ē
= 0, ci = 1, 2, · · · , n (631)
∂ci
where we assume the basis functions to be fixed (or pre-optimized). We can
therefore differentiate partially with respect to each ci and get n equations:
n n n n n n
∂ Ē " " ∂ "" ∂ ""
cj ck Sjk +Ē cj ck Sjk = cj ck Hjk , i = 1, 2, · · · , n
∂ci j k ∂ci j k ∂ci j k
(632)
Lets consider the second term
n n n "n & ' n "
n & '
∂ "" " ∂ " ∂cj ∂ck
cj ck Sjk = (cj ck ) Sjk = ck + cj Sjk
∂ci j k j k
∂c i j k
∂c i ∂c i

(633)
Since all of the coefficients are independent variable we have
∂cj
= δij (634)
∂ci
Therefore, we have
n n n "
n n "
n n n
∂ "" " " " "
cj ck Sjk = ck δij Sjk + cj δik Sjk = ck Sik + cj Sji
∂ci j k j k j k k j
(635)
since we assumed real basis functions
n n n n n n n
∂ "" " " " " "
cj ck Sjk = ck Sik + cj Sij = ck Sik + ck Sik = 2 ck Sik
∂ci j k k j k k k
(636)
Replacing Sjk with Hjk we get
n n n
∂ "" "
cj ck Hjk = 2 ck Hik (637)
∂ci j k k

76
where we have used that Hji = Hij∗ = Hij since Ĥ is Hermitian and real
functions. Collecting the terms gives up
n
" n
"
2Ē ck Sik = 2 ck Hik , i = 1, 2, · · · , n (638)
k k

which gives us the following set on n simultaneous linear homogenous equa-

tions n
" ( )
(Hik − Sik Ē)ck = 0 (639)
k

For example for n = 2 we have

(H11 − S11 Ē)c1 + (H12 − S12 Ē)c2 = 0 (640)

(H21 − S21 Ē)c1 + (H22 − S12 Ē)c2 = 0 (641)

Similar to what we know from degenerate perturbation theory, the only non-
trivial solution is when
det(Hij − Sij Ē) = 0 (642)
The solution to this equation gives us n roots which we can order

Ē1 " Ē2 " · · · " E¯n (643)

Similar we can order the bound states of the real system as

E1 " E2 " · · · " En " En+1 " · · · (644)

From variational theorem we know that E1 " Ē1 and we prove that it holds
for all states
E1 " Ē1 , E2 " Ē2 , · · · , En " E¯n (645)
Example: Find the eigenvalues and normalized eigenvector of the fol-
lowing Hermitian matrix
* +
3 2i
A= (646)
−2i 0

The secular equation for the eigenvalues λ is det(aij − δij λ) = 0 or

, ,
, 3 − λ 2i , 2
, −2i λ , = 0, λ − 3λ − 4 = 0 (647)
, ,

77
This gives us eigenvalues of λ1 = 4 and λ = −1. For the first root the the
set of simultaneous linear equations is

(3 − λ1 )c11 + 2ic12 = (3 − 4)c11 + 2ic12 = −c11 + 2ic12 = 0 (648)

−2ic11 − λ1 c12 = −2ic11 − 4c12 = 0 (649)

using either of these give c11 = 2ic12 . Normalization gives

1 = |c11 |2 + |c12 |2 = 4|c12 |2 + |c12 |2 (650)

√ √
which gives c12 = / 5 and c11 = 2i/ 5. Similarly, for the second root we will
get √ √
c21 = −i/ 5, c22 = 2/ 5 (651)
Checking that if we use the normalized eigenvectors we can diagonalize the
Hermitian matrix
* +* +* +
−1 † −2i 1 3 2i 2i −i
C AC = C AC = 1/5 (652)
i 2 −2i 0 1 2
* +* + * +
−2i 1 8 i 20 0
= 1/5 = 1/5 (653)
i 2 4 −2 0 −5

Find the eigenvalues of the following matrices

* +
0 −1
A= (654)
3 2
√ √ √ √ √
Solution:
√ λ =√1+ 2i, √
ci = −1/ 12+i/ 6, c2 = 3/2 and λ = 1− 2i, c1 =
−1/ 12 − i/ 6, c2 = 3/2
* +
2 0
B= (655)
9 2

Solution: λ = 2, c1 = 0, c2 = 1, and λ = 2, c1 = 1, c2 = 0
* +
4 0
C= (656)
0 4

Solution: λ = 4, c1 = 1, c2 = 0, and λ = 4, c1 = 0, c2 = 1

78
16.4 The hydrogen atom in an external electric field
Lets consider a hydrogen atom in an external electric along the z-axis. Since
we polarize the wavefunction along the z direction an appropriate choice of
the trial wavefunction would be

ψ = c1 1s + c2 2pz (657)
and the Hamiltonian for the system is
1 1
Ĥ = Ĥhyd − F rcosθ = − ∇2 − − F rcosθ (658)
2 r
Since n = 2 we can write the secular equation as
, ,
, H11 − ĒS11 H12 − ĒS12 ,
, H21 − ĒS21 H22 − ĒS22 , = 0 (659)
, ,

Since the hydrogen-like functions are orthonormal the evaluation of the over-
lap integrals is trivial and the secular equation reduces to
, ,
, H11 − Ē H12 ,,
,
, H21 =0 (660)
H22 − Ē ,
We therefore need to consider the following integrals
1
H11 =< 1s|Ĥhyd|1s > −F < 1s|rcosθ|1s >= − (661)
2

1
H22 =< 2pz |Ĥhyd |2pz > −F < 2pz |rcosθ|2pz >= − (662)
8
and
H12 =< 1s|Ĥhyd |2pz > −F < 1s|rcosθ|2pz >= −F < 1s|rcosθ|2pz > (663)
! ∞ ! π
F 4
− √ r exp(−3r/2)dr cos2 θ sin θdθ (664)
2 2 0 o
F 4! 2
=− √ [ 5
][ ] = −0.745F (665)
2 2 (3/2) 3
Therefore, the secular equation reduces to
, − 1 − Ē −0.745F
, ,
,
2 ,=0 (666)
, −0.745F − 1 − Ē
,
,
8

79
Expanding this give
1 1 5
(− − Ē)(− − Ē) − 0.7452 F 2 = Ē 2 + Ē − 0.7452 F 2 = 0 (667)
2 8 8
for which the solution to the quadratic equation gives
-
5 9/64 + 2.22F 2
Ē = − ± (668)
16 2
Therefore, for F = 0 we get
-
5 9/64 5 3 1 1
Ē = − ± =− ± = − ,− (669)
16 2 16 16 2 8
which are the unperturbed solution as expected. For a field strength of
F = 0.1Z we get
5
Ē = − ± 0.202 = −0.5145, −0.1105 (670)
16
Next we need to solve for the coefficients c1 and c2 , the linear equations are

(H11 − Ē)c1 + H12 c2 = 0 (671)

H21 c1 + (H22 − Ē)c2 = 0 (672)

Substituting the solution for F = 0.1 into the equation gives

0.01425c1 − 0.0745c2 = 0 (673)

−0.0745c1 + 0.3893c2 = 0 (674)

which gives us c1 = 5.2275c2. Using the normalization constraints we get

c21 + c22 = 1 = (5.22752 + 1)c2 , c2 = ±0.1879 (675)

which for the positive root gives c1 = 0.98219

80