Chem 441 Quantum Chemistry Notes May2, 2003 rev

VI. Approximate Solutions

A. Variation Method and Huckel Molecular Orbital (HMO) Calculations

References: Liberles, Ch. 4, Atkins, Ch. 8, Pauling and Wilson

Streitweiser, "MO Theory for Organic Chemists"
Bauer, H; Roth, K. "The Numeric Solution of Eigenvalue Problems", J. Chem. Educ., 1980, 57, 422-424.
Consider the Schrodinger Equation: H ψ∼ = eψ∼ (1)
∼
Where ψ is an approximate solution to a particular problem and H is an accurate Hamiltonian operator for that problem.
A key difference in this approximate case is that the energy e is not constant (unless, of course, the approximate solution
happens to be exact!). Now, consider the following operations on (1):
∼ ∼ ∼ ∼ ∼ ∼ ∼
Multiply both sides by ψ * : ψ * H ψ = ψ * eψ = eψ *ψ (2)

(note that e and ψ∼* commute, but H and ψ∼* cannot commute.)
∼ ∼
ψ *ψ dτ
Now note that (3)
∼ ∼
∫ψ *ψ dτ
is the probability of finding a particle in the region dτ. (The integral in the denominator is over the entire region of space
∼ ∼
where the particle described by ψ could be found. It has the value 1 if ψ is normalized.)
∼ ∼
ψ *ψ dτ
Thus it is reasonable to claim that e represents the probability of finding the particle with energy e. Using
∼ ∼
∫ψ *ψ dτ
this idea, we can claim that the average value of the energy is the sum of such terms over all points in space:
∼ ∼ ∼ ∼ ∼ ∼
E =
∫ eψ *ψ dτ = ∫ψ *H ψ =
< ψ *H ψ >
(4)
∼ ∼ ∼ ∼ ∼ ∼
∫ψ *ψ dτ ∫ψ *ψ < ψ *ψ >

(The <...> symbols in < ψ∼ *ψ∼ > are simply a shorthand notation for integration over the entire space where the particle
∼
described by ψ might be found.)

An important theorem about <E>o is the Variation Theorem: < E0 > ≥ E0,exact (5)
∼
Proof: let ψ = ∑ cnψ n where ψ n is an exact solution with eigenvalue En.
n

     *  
∫  ∑ c ψ  H  ∑ cnψ n  ∫  ∑ cmψ m   ∑ cnH ψ n 
*
m m

Plugging this into (4) we have: E = n   n  =  n   n =

 *  
∫  ∑n cmψ m   ∑n cnψ n  ∑n cn ∫ψ nψ n
2 *

  
∫  ∑ c ψ   ∑ c E ψ ∑ E c ∫ψ ψ ∑E c
* 2 * 2
m m n n n  n n n n n n
n n = n
= n
(6)
∑ c ∫ψ ψ 2
n
*
n n ∑ c ∫ψ ψ2
n
*
n n ∑c n
2
n
n n

Page 1
which can be rearranged to: Eo ∑c 2
n = ∑ En cn2 where we have assumed that the approximate energy will be an
approximation to the lowest E level. Consider now:  Eo − Eo ,exact  ∑ cn2 = ∑  En − Eo, exact cn2 ≥ 0
Hence, <Eo> >_ Eo,exact QED.

Problem 1. Prove that all the steps in (6) are correct.

∼
Problem 2: The substitution: ψ = ∑ cnψ n is reasonable if it is a general way of
n

describing any possible approximate wave function. We can prove that this is the case by showing how to
obtain a general formula for ck. Show that:

ck = < ψ k* ψ∼ > (7)

(Hint: multiply the starting equation from the left by ψn* and use the
orthogonality of the basis set solutions ψn ).
Note: In a more careful treatment of this material, cn2 should be replaced by cn*cn. There is no difference between the
two cases when cn is real, as will be the case in the examples we examine.

Applications to Molecular Orbitals

Let us suppose that we are trying to find approximate energies and wave functions for the butadiene p electrons. The
usual way of picturing this is shown in Figure 1 on the following page, where the (+) and (-) symbols indicate relative
phases. The four pictures are relatively easy to construct without any algebra, and the assignments to π and π*
categories are also easy to rationalize. All four MO's also can be written in an algebraic way as:
ψ∼ = c ψ
MO ∑ n nwhere we take ψ to be ψ on carbon atoms 1 - 4 (n = 1,2,3,4).
n p
n

When we construct molecular orbitals of this type, we are using a linear combination of atomic orbitals (LCAO)
approach. The goal of this approximate procedure is to determine, first, the best set of energy levels from the LCAO
mo's, and second, the shapes of the associated wavefunctions.

Figure 1. π Molecular Orbitals for Butadiene

Page 2
First let us look at energies. We begin with the first version of (6):

   
∫  ∑ c ψ
 H  ∑ cnψ n  ∑∑ c c ∫ψ H ψ ∑∑ c c H
* *
m m m n m n m n mn
E = n   n = m n
= m n
(8)
 *   ∑∑ c c ∫ψ ψ *
∑∑ c c S
∫  ∑n cmψ m   ∑n cnψ n  m n m n m n mn
m n m n

where H mn = ∫ψ m* H ψ n and S mn = ∫ψ m*ψ n (9)

Our goal is, first of all, to find the minimum energy, which will be the best approximation to the correct Eo. We can
ensure that <E> is minimized by taking its derivative with respect to a particular constant ci and setting it equal to 0:
   
∂ E
2∑ cn H in  ∑∑ cm cn H mn  2  ∑ cn Sin  ∑n cn H in − E ∑n cn Sin
=0= n
−  m n   n  =2 (10)
∂ci ∑∑ c c Smn  
2
∑∑ cm cn S mn
 ∑∑ cm cn S mn 
m n
m n m n
 m n 

or: ∑c H
n
n in − E ∑c S
n
n in =0 (11)

 
(which is usually written: ∑c
n
n  H in − E

∑S
n
in =0

(12)

This equation must be true for all values of i. The set of all such equations is often expressed in matrix form:

 H11 − < E > S11 H12 − < E > S12 H13 − < E > S13 .... H1 j − < E > S1 j   c1   0 
 
 H 21 − < E > S21 H 22 − < E > S 22 H 23 − < E > S23 .... H 2 j − < E > S 2 j   c2   0 
   
H − < E > S H 32 − < E > S32 H 33 − < E > S33 .... H 3 j − < E > S3 j   c3   0 
 31 31
  =   (13)
 ....   ...   ... 
    
 ....   ...   ... 
 
H − < E > S
 i1 i1 H i 2 − < E > Si 2 H i 3 − < E > Si 3 .... H ij − < E > Sij   ci   0 

These equations are known collectively as the secular equations.

Some further simplifications:

1) Choose Sij = 1 if i = j, 0 if i =/ j

2) Call Hij = α if i = j (α is the energy of a non-bonding p orbital).

3) Call Hij = β if i = j +
_ 1 (β is known as the resonance integral).

4) Take Hij = 0 for all other i,j combinations.

Problem: Show that α does indeed correspond approximately to the energy of a p orbital in a carbon atom.

Page 3
For butadiene, using these assumptions, the secular matrix equation becomes:

α − E β 0 0   c1   0 
    
 β α−E β 0   c2   0 
= (14)
 0 β α−E β   c3   0 
    
 0 0 β α − E   c4   0 

where E = <E> has been used for notational simplicity. This equation can be simplified by dividing both sides by β and
α−E
then defining =x (Note that: E = α - βx)
β

x 1 0 0   c1   0 
    
1 x 1 0   c2   0 
When this is done, (14) reduces to: = (16)
0 1 x 1   c3   0 
     
0 0 1 x 
 c4   0 

This equation is valid either if all the ci are 0 (a solution of no physical interest) or if the determinant of the square matrix
is 0. Solving this determinant leads to four values of x, which in turn leads to four values of E.

Solving the Determinant:

x 1 0 0
x 1 0 1 1 0
1 x 1 0
=x1 x 1 − 1 0 x 1 = x ( x3 − x − x ) − ( x 2 − 1) = x 4 − 3 x 2 + 1 = y 2 − 3 y + 1 (17)
0 1 x 1
0 1 x 0 1 x
0 0 1 x
3± 5 3± 5
where y = x2. This has the roots: y = so x = = -1.618, -0.618, +0.618, +1.618 (18)
2 2
Problem: Write out the 4 energy values for E in terms of α and β using (15)

Note: MacDonald’s Theorem considers the relationship between the variational energy approximation for the nth E
level and the exact energy value for this level. The result is an extension of the variational theorem stated in (5) above:

En ≥ En exact (cf. Pauling and Wilson, p. 188) (19)

The MO's associated with the energy values are derived by the following recipe:

1) put a particular x value into the secular determinant, e.g. x = -1.618:

−1.618 1 0 0
1 −1.618 1 0
0 1 −1.618 1
0 0 1 −1.618

2) write down the cofactors of each column in the determinant:

−1.618 1 0
(1+1)
A = (−1)
11
1 −1.618 1 = −1.000
0 1 −1.618

Page 4
 1 1 0
A12 = (−1)(1+ 2) 0 −1.618 1 = −1.618
0 1 −1.618

1 −1.618 0
(1+ 3)
A = (−1)
13
0 1 1 = −1.618
0 1 −1.618
1 −1.618 1
(1+ 4)
A = (−1)
14
0 1 −1.618 = −1.000
0 0 1

A1i A1i
3) set ci = = giving c1 = 0.372 c2 =0.601 c3 = 0.601 c4 = 0.372 (20)
4 2.690
∑( A )
j =1
1j 2

Problem: Sketch a picture of the MO associated with these coefficients.

Problem: Derive the coefficients and pictures for the other three MO's

Problem: Show that the first MO is normalized.

Problem: Show that c112 + c122 + c132 + c142 = 1

where "11" refers to the coefficient for the first MO for carbon 1, "12" is the coefficient for the second MO for
carbon 1, etc.

More complex problems and computer codes

It is impractical to work out solutions "by hand" for matrix problems with a dimension greater than 4, and so we turn to
computer programs to solve matrix equations analogous to (16). There is a good discussion of the general method
("Jacobi rotation") in the J. Chem. Source cited at the beginning of this unit, and also in a programming source such as
"Numerical Recipies" by Press, et al. The procedures generally transform the square matrix of (16), with x set equal to
zero, by a series of Jacobi rotations into a diagonal form (that is, a set of transformations that makes all off-diagonal
elements arbitrarily close to zero). When the square matrix of (16) is "diagonalized" in this way, the diagonal elements
constitute the set of "x" eigenvalues, and the product of all the successive Jacobian rotation matrices is a square matrix
with the property that each row (or column) contains the set of LCAO coefficients. We have a FORTRAN
implementation of this procedure.

Applications of the simple HMO method

1) the 4n + 2 rule (cf. Streitweiser, MO Theory for Organic Chemists)

2) epr spectra for conjugated radical anions and cations of conjugated systems. We will do some spreadsheet examples.

Page 5
B. Approximate Solutions - Perturbation Theory

Perturbation theory is an alternative rather useful way of getting approximate results for complex problems in quantum
mechanics. It is developed as follows:

Assume that the Hamiltonian operator can be broken up into two parts: H = H o + λv (21)

where H 0 is the Hamiltonian for an exactly solvable problem and v is an additional potential energy term (a small one
we hope!) which makes the problem difficult or impossible to solve exactly.

We then assume that the eigenfunctions of H can be written in the following form:

ψ∼ = ψ o + λψ (1) + λ 2ψ (2) + λ 3ψ (3) + ... (22)

and that E = E o + λ E (1) + λ 2 E (2) + λ 3 E (3) + ... (23)

The ψ o and E o are solutions to the exact problem (v = 0). ψ (1) and E (1) are known as first-order perturbations, ψ (2) and
E (2) are second order perturbations, etc.. The theory is useful in cases where the series terms quickly get vanishingly
small. We also usually assume that λ = 1 for practical calculations, but we do not want to do this just yet. Now consider
the correct SE:
H ψ∼ = Eψ∼ which becomes, with substitution:
(H + λ v ) ψ + λψ
o o (1)
+ λ 2ψ (2) + λ 3ψ (3) + ...
(24)
− ( E + λE + λ E
o (1) 2 (2)
+ λ 3 E (3) + ... ) ψ o + λψ (1) + λ 2ψ (2) + λ 3ψ (3) + ... = 0
The game we play now involves collecting like terms in λ in the products of (24):

(H ψ o − E oψ o ) + λ  vψ o + H oψ (1) − E oψ (1) − E (1))ψ o 

o

+ λ 2  vψ (1) + H oψ (2) − E oψ (2) − E ( )ψ ( ) − E ( )ψ o 

1 1 2
(25)
+ λ 3 [...] + ... = 0
Then, since λ is a variable, we require the coefficient of each λ term to = 0 (we encountered this property earlier in the
term with the harmonic oscillator problem). This allows us to determine, in succession, E (1) ,ψ (1) , E (2) ,ψ (2 ) etc.

E (1) : Setting the coefficient of λ in (25) equal to 0 gives: vψ o + H oψ (1) − E oψ (1) − E (1)ψ o = 0 (26)

Rearrange this to: E (1)ψ o = vψ o + H oψ (1) − E oψ (1) (27)

and then multiplying both sides by ψ o* from the left and integrating over all space gives:

ψ o* E (1)ψ o = E (1) ψ o*ψ o = ψ o* vψ o + ψ o*H oψ (1) − ψ o* E oψ (1) (28)

There is a property of wavefunctions which we have not introduced or explored known as the Hermitian property
which is stated as follows:
*
ψ a*H ψ b = ψ b*H ψ a (29)
We can use this property to eliminate the last two terms:

E (1) ψ o*ψ o = ψ o* vψ o + ψ (1)*H oψ o − ψ (1)* E oψ o = ψ o* vψ o − 0 = ψ o* vψ o (30)

ψ vψ
o* o

so: E (1) = = ψ o* vψ o assuming normalization of ψ o . (31)

ψ o*ψ o

Page 6
ψ m(1) : for a particular energy level Em ψ m(1) is most often treated as a linear
combination of the exact solutions: ψ m(1) = ∑ cnmψ no (32)
n

If we look at equation (26) and focus on the middle terms, both of which involve ψ m(1) it is easy to show that if
ψ m(1') is a solution to (26), then ψ m(1) = ψ m(1') + aψ mo will be an equally good solution.

Problem: Prove that this claim is true.

In other words, ψ m(1) is not determined exactly by (26). We can take advantage of this by noting that:

ψ m(1)*ψ mo = ψ m(1')*ψ mo + a ψ mo*ψ mo = 0 (33)

ψ m(1)*ψ mo − ψ m(1')*ψ mo
if we choose the arbitrary constant as: a = (34)
ψ mo*ψ mo

This means that in general we can assume that ψ m(1) will be chosen so that it is orthogonal to ψ mo .

Let us now go back to (32) and solve it for cmn:

First we note that the orthogonality of ψ m(1) and ψ mo now requires cmm = 0.

Problem: Show that if cmm = 0 then ψ m(1) and ψ mo are in fact orthogonal.

Let us now plug (32) into (26). This leads to a formula for the cmn coefficients as follows:

vψ mo + ∑ cmnH ψ no − Emo ∑ cmnψ no − Em(1)ψ mo =

vψ mo + ∑ cmn Enoψ no − Emo ∑ cmnψ no − Em(1)ψ mo = (35)
vψ mo + ∑ cmn ( Eno − Emo )ψ no − Em(1)ψ mo

which rearranges to: ∑c (Emn

o
m − Eno )ψ no = vψ mo − Em(1)ψ mo (36)

To get a particular coefficient cmk, multiply both sides of (36) by ψ ko and integrate
over all space. Taking advantage of the orthogonality of the ψ no functions, we get:

∑c (Emn
o
m − Eno ) ψ ko*ψ no = cmk ( Emo − Eno ) = ψ ko* vψ no − Em(1) ψ ko*ψ no = ψ ko* vψ no (37)

ψ ko* vψ mo
and solving for cmk thus gives: cmk = (38)
(E o
m − Eko )
 ψ no* vψ mo 
Finally, this gives an explicit formula for ψ m(1) ψ m(1) = ∑ 
 ψ no (39)
n  ( E m − En ) 
o o
 
E : Setting the coefficient of λ = 0, we multiply by ψ once again from the left, followed by
(2) 2 o*

integration. The Hermitian property is now used to eliminate the second and third terms in the
λ2 coefficient in (25), giving:

Page 7
 ψ o* vψ (1) − E (1) ψ o*ψ (1) ψ o* vψ (1)
E (2) = = (40)
ψ o*ψ o ψ o*ψ o

Problem: Show the steps in the derivation of (40). (Orthogonality accounts for the numerator
simplification.)

If ψ (1) is written as an expansion once more (cf. (32)), equation (40) can be written as:

ψ mo* vψ m(1) ψ mo* v∑ cmnψ no  ψ o* vψ o 2 

= ∑ o 
m n
E (2)
= = (41)
ψ mo*ψ mo ψ mo*ψ mo n 

( Em − Eno ) 


Again, normalization of the ψ mo values has been assumed.

Applications:
1. Perturbed Harmonic Oscillator
2. Electronic transitions: e.g. IR spectroscopy of HCl

Page 8