Professional Documents
Culture Documents
Perturbation theory
6 lectures, (Tuesday and Friday, weeks 4-6 of Hilary term)
Chris-Kriton Skylaris
(chris-kriton.skylaris @ chem.ox.ac.uk )
Physical & Theoretical Chemistry Laboratory
South Parks Road, Oxford
Bibliography
All the material required is covered in Molecular Quantum Mechanics fourth edition
by Peter Atkins and Ronald Friedman (OUP 2005). Specically, Chapter 6, rst half of
Chapter 12 and Section 9.11.
Further reading:
Quantum Chemistry fourth edition by Ira N. Levine (Prentice Hall 1991).
Quantum Mechanics by F. Mandl (Wiley 1992).
Quantum Physics third edition by Stephen Gasiorowicz (Wiley 2003).
Modern Quantum Mechanics revised edition by J. J. Sakurai (Addison Wesley Long-
man 1994).
Modern Quantum Chemistry by A. Szabo and N. S. Ostlund (Dover 1996).
1
Contents
1 Introduction 2
1 Introduction
In these lectures we will study perturbation theory, which along with the variation theory
presented in previous lectures, are the main techniques of approximation in quantum
mechanics. Perturbation theory is often more complicated than variation theory but
also its scope is broader as it applies to any excited state of a system while variation
theory is usually restricted to the ground state.
We will begin by developing perturbation theory for stationary states resulting from
Hamiltonians with potentials that are independent of time and then we will expand
the theory to Hamiltonians with time-dependent potentials to describe processes such
as the interaction of matter with light. Finally, we will apply perturbation theory to
the study of electric properties of molecules and to develop Mller-Plesset many-body
perturbation theory which is often a reliable computational procedure for obtaining most
of the correlation energy that is missing from Hartree-Fock calculations.
(1) (1)
The terms n and En are called the rst order corrections to the wavefunction and
(2) (2)
energy respectively, the n and En are the second order corrections and so on. The
task of perturbation theory is to approximate the energies and wavefunctions of the
perturbed system by calculating corrections up to a given order.
Note 2.1 In perturbation theory we are assuming that all perturbed quantities are func-
tions of the parameter , i.e. H(), En () and n (r; ) and that when 0 we have
(0) (0)
H(0) = H (0) , En (0) = En and n (r; 0) = n (r). You will remember from your maths
course that the Taylor series expansion of say En () around = 0 is
dEn 1 d2 En 2 1 d3 En
En = En (0) + + 2 + 3 3 + (6)
d =0 2! d =0 3! d =0
By comparing this expression with (5) we see that the perturbation theory corrections to
(0)
the energy level En are related to the terms of Taylor series expansion by: En = En (0),
(1) (2) 2 (3) 3
En = dE n
| , En = 2!1 ddE2n |=0 , En = 3!1 ddE3n |=0 , etc. Similar relations hold for the
d =0
expressions (3) and (4) for the Hamiltonian and wavefunction respectively.
Note 2.2 In many textbooks the expansion of the Hamiltonian is terminated after the
rst order term, i.e. H = H (0) + H (1) as this is sucient for many physical problems.
To simplify the expressions from now on we will use bra-ket notation, representing
(0) (1)
wavefunction corrections by their state number, so n |n(0) , n |n(1) , etc.
n(0) |(H (0) En(0) )|n(1) = n(0) |(En(1) H (1) )|n(0) (12)
n(0) |H (0) |n(1) En(0) n(0) |n(1) = En(1) n(0) |n(0) n(0) |H (1) |n(0) (13)
En(0) n(0) |n(1) En(0) n(0) |n(1) = En(1) n(0) |H (1) |n(0) (14)
0 = En(1) n(0) |H (1) |n(0) (15)
where in order to go from (13) to (14) we have used the fact that the eigenfunctions of
the unperturbed Hamiltonian H (0) are normalised and the Hermiticity property of H (0)
which allows it to operate to its eigenket on its left
n(0) |H (0) |n(1) = (H (0) n(0) )|n(1) = (En(0) n(0) )|n(1) = En(0) n(0) |n(1) (16)
Lecture 1 5
So, according to our result (15), the rst order correction to the energy is
which is simply the expectation value of the rst order Hamiltonian in the state |n(0)
(0)
n of the unperturbed system.
Example 1 Calculate the rst order correction to the energy of the nth state of a har-
monic oscillator whose centre of potential has been displaced from 0 to a distance l.
The Hamiltonian of the unperturbed system harmonic oscillator is
h2 d2 1 2
H (0) = + kx (18)
2m dx2 2
while the Hamiltonian of the perturbed system is
h2 d2 1
H = 2
+ k(x l)2 (19)
2m dx 2
2 2
h d 1 1
= 2
+ kx2 lkx + l2 k (20)
2m dx 2 2
= H (0) + lH (1) + l2 H (2) (21)
where we have dened H (1) kx and H (2) 12 k and l plays the role of the perturbation
parameter . According to equation 17,
From the theory of the harmonic oscillator (see earlier lectures in this course) we know
that the diagonal matrix elements of the position operator within any state |n(0) of the
harmonic oscillator are zero (n(0) |x|n(0) = 0) from which we conclude that the rst
order correction to the energy in this example is zero.
(0) (0)
because we have assumed non-degeneracy of the zeroth-order problem (i.e. En Ek =
0 ).
To proceed in our derivation for an expression for |n(1) we will employ the iden-
tity operator expressed in the eigenfunctions of the unperturbed system (zeroth order
eigenfunctions):
|n(1) = 1|n(1) = |k (0) k (0) |n(1) (26)
k
Before substituting (25) into the above equation we must resolve a conict: k must be
dierent from n in (25) but not necessarily so in (26). This restriction implies that
the rst order correction to |n will contain no contribution from |n(0) . To impose this
restriction we require that that n(0) |n = 1 (this leads to n(0) |n(j) = 0 for j 1. Prove
it! ) instead of n|n = 1. This choice of normalisation for |n is called intermediate
normalisation and of course it does not aect any physical property calculated with |n
since observables are independent of the normalisation of wavefunctions. So now we can
substitute (25) into (26) and get
(1)
where the matrix element Hkn is dened by the above equation.
n(0) |H (0) En(0) |n(2) = n(0) |En(2) H (2) |n(0) + n(0) |En(1) H (1) |n(1)
0 = En(2) n(0) |H (2) |n(0) n(0) |H (1) |n(1) (28)
where we have used the fact that n(0) |n(1) = 0 (section 2.1.2). We now solve (28) for
(2)
En
En(2) = n(0) |H (2) |n(0) + n(0) |H (1) |n(1) = Hnn
(2)
+ n(0) |H (1) |n(1) (29)
which upon substitution of |n(1) by the expression (27) becomes
H (1) H (1)
En(2) = (2)
Hnn + nk kn
(0) (0)
. (30)
k=n En Ek
Example 2 Let us apply what we have learned so far to the toy model of a system
(0) (0)
which has only two (non-degenerate) levels (states) |1(0) and |2(0) . Let E1 < E2 and
assume that there is only a rst order term in the perturbed Hamiltonian and that the
Lecture 1 7
(1)
diagonal matrix elements of the perturbation are zero, i.e. m(0) |H (1) |m(0) = Hmm = 0.
For this simple system we can solve exactly for its perturbed energies up to innite order
(see Atkins):
1 (0) (0) 1 (0) (0) (1) 1
E1 = (E1 + E2 ) [(E1 E2 )2 + 4|H12 |2 ] 2 (31)
2 2
1 (0) (0) 1 (0) (0) (1) 1
E2 = (E1 + E2 ) + [(E1 E2 )2 + 4|H12 |2 ] 2 (32)
2 2
According to equation 30 the total perturbed energies up to second order are
(1)
(0) |H12 |2
E1 E1 (0) (0)
(33)
E2 E1
(1)
(0) |H12 |2
E2 E2 + (0) (0)
. (34)
E2 E1
These sets of equations show that the eect of the perturbation is to lower the energy
of the lower level and raise the energy of the upper level. The eect increases with the
(1)
strength of the perturbation (size of |H12 |2 term) and decreasing separation between the
(0) (0)
unperturbed energies ( E2 E1 term).
Lecture 2 8
We can now see that the above expression could be simplied signicantly if the sum
over k could be made to include n as this
would allow us to eliminate it by using the com-
pleteness (or closure) property (1 = k |k k (0) |) of the zeroth order wavefunctions.
(0)
(1) (1)
We achieve just that by adding and subtracting the Hnn Hnn /E term:
1 (1) (1) 1
En(2) Hnn (2)
Hnk Hkn + H (1) H (1) (36)
E k E nn nn
(2) 1 1
Hnn n(0) |H (1) H (1) |n(0) + H (1) H (1) (37)
E E nn nn
This approximation can only be accurate if n = 0 (the ground state) and all excited
states are much higher in energy from the ground state than their maximum energy
separation. This assumption is usually not valid. Nevertheless, from a mathematical
viewpoint, it is always possible to nd a value for E that makes the closure approx-
imation exact. To nd this value we just need to equate (37) to the righthand side of
(30) and solve for E to obtain
(1) (1)
Hnn Hnn n(0) |H (1) H (1) |n(0)
E = (1) (1)
. (38)
Hnk Hkn
k=n E (0) E (0)
n k
E6
E5
E4
E3
Energy E2
E1
0 O
where now we use two indices to represent each state: the rst index n runs over the
dierent energy eigenvalues while the second index i runs over the d degenerate states
(0)
for a particular energy eigenvalue. Since we have d degenerate states of energy En , any
(0)
linear combination of these states is also a valid state of energy En . However, as the
perturbation parameter is varied continuously from 0 to some nite value, it is likely
that the degeneracy of the states will be lifted (either completely or partially). The ques-
(0)
tion that arises then is whether the states n,i of equation (39) are the correct ones,
i.e. whether they can be continuously transformed to the (in general) non-degenerate
perturbed states. It turns out that this is usually not the case and one has to rst nd
the correct zeroth order states
(0)
d
n,j = |(n, i)(0) cij j = 1, . . . , d (40)
i=1
(0)
where the coecients cij that mix the n,i are specic to the perturbation H (1) and are
determined by its symmetry.
(0)
Here we will nd a way to determine the correct zeroth order states n,j and the
(0)
rst order correction to the energy. To do this we start from equation 9 with n,i in
(0)
place of n,i
(1) (1) (0)
(H (0) En(0) )n,i = (En,i H (1) )n,i (41)
(1)
Notice that we include in the notation for the rst order energy En,i the index i since the
Lecture 2 10
(0)
perturbation may split the degenerate energy level En . Figure 1 shows an example for a
hypothetical system with six states and a three-fold degenerate unperturbed level. Note
that the perturbation splits the degenerate energy level. In some cases the perturbation
may have no eect on the degeneracy or may only partly remove the degeneracy.
The next step involves multiplication from the left by (n, j)(0) |
(1) (0)
(n, j)(0) |H (0) En(0) |(n, i)(1) = (n, j)(0) |En,i H (1) |n,i (42)
(1) (0)
0 = (n, j)(0) |En,i H (1) |n,i (43)
(1)
0 = (n, j)(0) |En,i H (1) |(n, k)(0) cki (44)
k
where we have made use of the Hermiticity of H (0) to set the left side to zero and we
(0)
have substituted the expansion (40) for n,i . Some further manipulation of (44) gives:
(1)
[(n, j)(0) |H (1) |(n, k)(0) En,i (n, j)(0) |(n, k)(0) ]cki = 0 (45)
k
(1) (1)
(Hjk En,i Sjk )cki = 0 (46)
k
We now observe that as En,i occurs in every row, this determinant is actually a dth
degree polynomial in En,i and the solution of the above equation for its d roots will give
(1)
us all the En,i (i = 1, . . . d) rst order corrections to the energies of the d degenerate
(0) (1)
levels with energy En . We can then substitute each En,i value into (46) to nd the
corresponding non-trivial solution of cki (k = 1, . . . d) coecients, or in other words
(0) (1) (0) (0)
the function n,i . Finally, you should be able to verify that En,i = n,i |H (1) |n,i ,
i.e. that the expression (17) we have derived which gives the rst order energy as the
expectation value of the rst order Hamiltonian in the zeroth order wavefunctions still
holds, provided the correct degenerate zeroth order wavefunctions are used.
four n = 2 states (one 2s orbital and three 2p orbitals) have the same energy. The lifting
of this degeneracy when the atom is placed in an electric led is called the Stark eect
and here we will study it using rst order perturbation theory for degenerate systems.
Assuming that the electric eld E is aplied along the z-direction, the form of the
perturbation is
H (1) = eEz z (48)
where the strength of the eld Ez plays the role of the parameter . Even though we
have four states, based on parity and symmetry considerations we can show that only
elements between the 2s and 2pz orbitals will have non-zero o-diagonal H (1) matrix
elements and as a result the 44 system of equations (46) is reduced to the following
22 system (note that here all states are already orthogonal so the overlap matrix is
equal to the unit matrix):
2s|z|2s 2s|z|2pz c1 (1) c1
eEz =E (49)
2pz |z|2s 2pz |z|2pz c2 c2
The solution of the abovem system results in the following rst order energies and cor-
rect zeroth order wavefunctions
(0) 1 (0) 1
n,1 = (|2s |2pz ) , n,2 = (|2s + |2pz ) (52)
2 2
Therefore, the eect of the perturbation (to rst order) on the energy levels can be
summarised in the diagram of Figure 2.
Finally, we should mention that the energy levels of the hydrogen atom are also
aected in the presence of a uniform magnetic eld B. This is called the Zeeman eect,
and the form of the perturbation in that case is H (1) = 2m
e
(L + 2S) B where L is the
orbital angular momentum of the electron and S is its spin angular momentum.
Lecture 3 12
E(1)=3eEza0
l=1, m=-1, 0, 1
E(1)=0
l=0, m=0
4 degenerate
n=2 states
E(1)=-3eEza0
Ez
where the (0) (r), which depends only on position coordinates r, is the solution of the
energy eigenvalue equation (TISE)
and the expression for T (t) is derived by substituting the right hand side of the above
to the time-dependent equation 53. Finally we obtain
(0)
(0) (r, t) = n(0) (r)eiEn t/h
. (56)
(0)
Now let us consider the following linear combination of n
(0) (0)
(0) (r, t) = ak k (r)eiEk t/h (57)
k
where the ak are constants. This is also a solution of the TDSE (prove it!) because the
TDSE consists of linear operators. This more general superposition of states solution
of course contains (56) (by setting ak = nk ) but unlike (56) it is not, in general, an
(0)
eigenfunction of H (0) . Assuming that the n have been chosen to be orthonormal,
which is always possible, we nd that the expectation value of the Hamiltonian is
(0)
(0) |H (0) |(0) = |ak |2 Ek (58)
k
We see thus that in the case of equation 56 the system is in a state with denite energy
(0)
En while in the general case (57) the system can be in any of the states with an average
energy given by (58) where the probability Pk = |ak |2 of being in the state k is equal to
the square modulus of the coecient ak . Both (56) and (57 ) are time-dependent because
(0)
of the phase factors eiEk t/h but the probabilities Pk and also the expectation values
for operators that do not contain time (such as the H (0) above) are time-independent.
To simplify our discussion, in what follows we will only consider up to rst order per-
turbations in the Hamiltonian
We will use perturbation theory to approximate the solution (r, t) to the time-dependent
Schrodinger equation of the perturbed system.
ih = H(t) (61)
t
(0)
At any instant t, we can expand the (r, t), in the complete set of eigenfunctions k (r)
of the zeroth order Hamiltonian H (0) ,
(0)
(r, t) = bk (t)k (r) (62)
k
but of course the expansion coecients bk (t) vary with time as (r, t) does. In fact let
(0)
us dene bk (t) = ak (t)eiEk t/h in the above equation to get
(0) (0)
(r, t) = ak (t)k (r)eiEk t/h . (63)
k
Even though this expression looks more messy than (62), we prefer it because it will
simplify the derivation that follows and also it directly demonstrates that when the ak (t)
lose their time dependence, i.e. when 0 and ak (t) ak , (63) reduces to (57).
We substitute the expansion (63) into the time-dependent Schrodinger equation 53
(0)
and after taking into account the fact that the n = |n(0) are eigenfunctions of H (0)
we obtain
(0) dan (t) (0)
an (t)H (1) (t)|n(0) eiEn t/h = ih |n(0) eiEn t/h (64)
n n
dt
The next step is to multiply with k (0) | from the left and use the orthogonality of the
zeroth order functions to get
(0) dak (t) iE (0) t/h
an (t)k (0) |H (1) (t)|n(0) eiEn t/h
= ih e k (65)
n
dt
Solving this for dak (t)/dt results in the following dierential equation
dak (t) (1) (0) (0) (1)
= an (t)Hkn (t)ei(Ek En )t/h = an (t)Hkn (t)eikn t (66)
dt ih n ih n
The purpose now of the perturbation theory we will develop is to determine the
time-dependent coecients ak (t). We begin by writing a perturbation expansion for the
coecient ak (t) in terms of the parameter
(0) (1) (2)
ak (t) = ak (t) + ak (t) + 2 ak (t) + . . . (68)
where you should keep in mind that while and t are not related in any way, we take
t = 0 as the beginning of time for which we know exactly the composition of the
system so that
(0)
ak (0) = ak (0) (69)
(l)
which means that ak (0) = 0 for l > 0. Furthermore we will assume that
a(0)
g (0) = gj (70)
which means that at t = 0 the system is exclusively in a particular state |j (0) and all
other states |g (0) with g = j are unoccupied. Now substitute expansion (68) into (67)
and collect equal powers of to obtain the following expressions
(0) (0)
ak (t) ak (0) = 0 (71)
(1) (1) 1 t (0) (1) ikn t
ak (t) ak (0) = a (t )Hkn (t )e dt (72)
ih n 0 n
(2) (2) 1 t (1) (1) ikn t
ak (t) ak (0) = a (t )Hkn (t )e dt (73)
ih n 0 n
... (74)
We can observe that these equations are recursive: each of them provides an expression
(m) (m1) (1)
for af (t) in terms of af (t). Let us now obtain an explicit expression for af (t) by
rst substituting (71) into (72), and then making use of (70):
(1) 1 t (0) (1) if n t 1 t (1) if j t
af (t) = a (0)Hf n (t )e dt = H (t )e dt . (75)
ih n 0 n ih 0 f j
The probability that the system is in state |f (0) is obtained in a similar manner to
equation 58 and is given by the squared modulus of the af (t) coecient
Pf (t) = |af (t)|2 (76)
but of course a signicant dierence from (58) is that Pf = Pf (t) now changes with
time. Using the perturbation expansion (68) for af (t) we have
(0) (1) (2)
Pf (t) = |af (t) + af (t) + 2 af (t) + . . . |2 . (77)
Note that in most of the examples that we will study in these lectures we will conne
ourselves to the rst order approximation which means that we will also approximate
the above expression for Pf (t) by neglecting from it the second and higher order terms.
Lecture 3 16
Note 3.1 The previous derivation of time-dependent perturbation theory is rather rig-
orous and is also very much in line with the approach we used to derive time-independent
perturbation theory. However, if we are only interested in obtaining only up to rst order
corrections, we can follow a less strict but more physically motivated approach (see also
Atkins).
We begin with (67) and set equal to 1 to obtain
1 t (1)
ak (t) ak (0) = an (t )Hkn (t )eikn t dt (78)
ih n 0
This equation is exact but it is not useful in practice because the unknown coecient ak (t)
is given in terms of all other unknown coecients an (t) including itself ! To proceed we
make the following approximations:
1. Assume that at t = 0 the system is entirely in an initial state j, so aj (0) = 1 and
an (0) = 0 if n = j.
2. Assume that the time t for which the perturbation is applied is so small that the
change in the values of the coecients is negligible, or in other words that aj (t) 1
and an (t) 0 if n = j.
Using these assumptions we can reduce the sum on the righthand side of equation 78 to
a single term (the one with n = j for which aj (t) 1). We will also rename the lefthand
side index from k to f to denote some nal state with f = j to obtain
1 t (1) if j t
af (t) = H (t )e dt (79)
ih 0 f j
This approximate expression for the coecients af (t) is correct to rst order as we can
see by comparing it with equation 75.
Example 4 Show that with a time-dependent Hamiltonian H(t) the energy is not con-
served.
H H(t)
= |H(t)| + | | + |H(t)| (81)
t t t t
Lecture 3 17
Now using equation 80 to eliminate the t
terms we obtain
H H(t)
= | | = 0 (82)
t t
which shows (in contrast with the case of a time-independent Hamiltonian!) that the
time derivative of the energy is no longer zero and therefore the energy is no longer a
constant of the motion. So, in the time-dependent perturbation theory we develop here it
is pointless to look for corrections to the energy levels. Nevertheless, we will continue to
(0)
denote the energy levels of the unperturbed system as zeroth order, En , for consistency
with our previously derived formulas of time-independent perturbation theory.
Note 3.2 In this section we are not really applying perturbation theory: The two level
system allows us to obtain the exact solutions for the coecients a1 (t) and a2 (t) (up to
innite order in the language of perturbation theory).
(0)
The probability of the system being in the state 2 is
2 4|V |2 1 2
P2 (t) = |a2 (t)| = sin2 ( + 4|V |2 )1/2 t (87)
2
21 + 4|V |2 2 21
and of course since we only have two states here we will also have P1 (t) = 1 P2 (t).
Let us examine these probabilities in some detail. First consider the case where the
two states are degenerate (21 = 0). We then have
which means that the system oscillates freely between the two states |1(0) and |2(0)
and the only role of the perturbation is to determine the frequency |V | of the oscillation.
The other extreme is the case where the levels are widely separated in comparison with
2
the strength of the perturbation in the sense that 21 >> |V |2 . In this case we obtain
2
2|V | 1
P2 (t) sin2 21 t (89)
21 2
which shows that the probability of the system occupying state |2(0) can not get any
larger than (2|V |/21 )2 which is a number much smaller than 1. Thus the system remains
almost exclusively in state |1(0) . We should also observe here that the frequency of
oscillation is independent of the strength of the perturbation and is determined only by
the separation of the states 21 .
Lecture 4 19
If we assume that we will only examine times very long after the perturbation has
reached its nal value, or in other words kt >> 1, we obtain
(1) 1 (1) eif j t 1 1
af (t) = Hf j + (92)
ih if j k if j
and nally that the rate in which the perturbation is switched is slow in the sense that
k 2 << f2j , we are left with
(1)
(1) Hf j i t
af (t) = e fj (93)
hf j
The square of this, which is the probability of being in state |f (0) to rst order is
(1)
(1) |Hf j |2
Pf (t) = |af (t)|2 = . (94)
h2 f2j
where V does not depend on time (but of course it could be a function of coordinates,
e.g. V = V (x)). This in a sense is the most general type of time-dependent perturbation
as any other time-dependent perturbation can be expanded as a sum (Fourier series) of
harmonic terms like those of (95). Inserting this expression for the perturbation H (1) (t)
into equation 75 we obtain
t
i(f j +)t
(1) 1 it it if j t 1 e 1 ei(f j )t 1
af (t) = Vf j (e + e )e dt = Vf j + (96)
ih 0 ih i(f j + ) i(f j )
(1) i 1 ei(f j )t
af (t) = Vf j (97)
h (f j )
This equation shows that due to the time-dependent perturbation, the system can make
transitions from the state |j (0) to the state |f (0) by absorbing a quantum of energy h.
Now in the case where f j = exactly, the above expression reduces to
|Vf j |2 2
lim Pf (t) = t (99)
f j h2
which shows that the probability increases quadratically with time. We see that this
expression allows the probability to increase without bounds and even exceed the (max-
imum) value of 1. This is of course not correct, so this expression should be considered
valid only when Pf (t) << 1, according to the assumption behind rst order perturbation
theory through which it was obtained.
Our discussion so far for the harmonic perturbation has been based on the assump-
(0) (0)
tion that Ef > Ej so that the external oscillating eld causes stimulated absorption
of energy in the form of quanta of energy h. However, the original equation 96 for
Lecture 4 21
(1) 4|Vf j |2 1
Pf (t) = |af (t)|2 = 2 2
sin2 (f j + )t (100)
h (f j + ) 2
As the nal states form a continuum, it is customary to count the number of states
dN(E) with energy in the interval (E, E + dE) in terms of the density of states (E) at
energy E as
dN(E) = (E) dE (102)
Using this formalism, we can change the sum of equation 101 into an integral
(0)
Ef +E
(1)
P (t) = (E)|aE (t)|2 dE (103)
(0)
Ef E
where the summation index n has been substituted by the continuous variable E. Ac-
(0)
cording to our assumption E Ef so the above expression after substitution of (98)
becomes
E (0) +E 2 1 (0)
f |Vf j |2 sin 2 (E/h Ej /h )t
P (t) = 4 2 (0)
(E)dE (104)
(0)
Ef E h (E/h Ej /h )2
(0)
where the integral is evaluated in a narrow region of energies around Ef . The integrand
(0)
above contains a term that, as t grows larger it becomes sharply peaked at E = Ej +h
and sharply decaying to zero away from this value (see Figure 3). This then allows us
to approximate it by treating |Vf j | as a constant and also the density of states as a
Lecture 4 22
0.8
0.6
0.4
0.2
0
-10 0 10
2
Figure 3: A plot of sin x(xt/2)
2 as a function of x and t. Notice that as t increases the
function turns into a sharp peak centred at x = 0.
Lecture 4 23
(0) (0)
constant in terms of its value (Ef ) at Ef . These constants can then be taken out of
the integral. What remains inside the integral is the trigonometric function. We now
(0) (0)
extend the range of integration from [Ef E, Ef +E] to (, ) as this allows us
to evaluate it but it barely aects its value due to the peaked shape of the trigonometric
function. Evaluation of the integral then results in the following expression
2 (0)
P (t) = t|Vf j |2 (Ef ) . (105)
h
Its derivative with respect to time is the transition rate which is the rate at which the
initially empty levels become populated.
dP 2 (0)
W (t) = = |Vf j |2 (Ef ) (106)
dt h
This succinct expression, which is independent of time, is sometimes called Fermis
golden rule.
1
We will neglect the magnetic interaction of the radiation with atoms as it is usually small compared
to the interaction with the electric field
2
Actually the electric field oscillates both in space and in time and has the following form
where the wavelength of the radiation is = 2/|k| and its angular frequency is = c|k|. However,
here we work under the assumption that is very large compared to the size of the atom and thus we
neglect the spatial variation of the field. This approach is called the electric dipole approximation.
Lecture 4 24
where the sum over k runs over all the electrons, and the position vector of the kth
electron is rk . The nucleus of the atom is assumed to be xed at the origin of coordinates
(r = 0).
We can immediately see that the work of section 3.5 for a perturbation oscillating
with time according to a harmonic time-dependent potential applies here if we set V =
z Ez in equation 95 and all the expressions derived from it. In particular, equation 98
for the probability of absorption or radiation for transition from state |j 0 to the higher
energy state |f (0) takes the form
4|z,f j |2 Ez2 () 2 1
Pf j (t) = sin (f j )t . (110)
h2 (f j )2 2
You will notice in the above expression that we have written Ez as Ez () in order to
remind ourselves that it does depend on the angular frequency of the radiation. In fact
the above expression is valid only for monochromatic radiation. Most radiation sources
produce a continuum of frequencies, so in order to take this fact into account we need
to integrate the above expression over all angular frequencies
4|z,f j |2 Ez2 () 2 1
Pf j (t) = 2 sin (f j )td (111)
h (f j )
2 2
4|f j |2 Ez2 (f j ) sin2 [ 12 ( f j )t]
= d (112)
h2 ( f j )2
2t|f j |2 Ez2 (f j )
= (113)
h2
where we have evaluated the above integral using the same technique we used for the
derivation of Fermis golden rule in the previous section. The rate of absorption of
radiation is equal to the time derivative of the above expression and as we are interested
in atoms in the gas phase, we average the above expression over all directions in space.
It turns out that this is equivalent to replacing |z,f j |2 by the mean value of x, y and z
components, 13 |f j |2 , which leads to
2|f j |2 Ez2 (f j )
Wf j (t) = (114)
3h2
A standard result from the classical theory of electromagnetism is that the energy den-
sity rad (f j ) (i.e. energy contained per unit volume of space for radiation of angular
frequency f j ) of the electromagnetic eld is
rad (f j ) = 20Ez2 (f j ) (115)
which upon substitution into (114) gives
2|f j |2
Wf j (t) = rad (f j ) (116)
620h2
Lecture 4 25
Wf j = Af j + Bf j rad (f j ) (121)
where Af j is the Einstein coecient of spontaneous emission which does not need to be
multiplied by rad (f j ) as spontaneous emission is independent of the presence of the
radiation f j . The expression for this coecient is (see Atkins for a derivation):
hf3j
Af j = Bf j (122)
2 c3
As we saw, spontaneous emission was postulated by Einstein as it is not predicted by
combining a quantum mechanical description of the atoms with a classical description
of the electric eld. It is predicted though by the theory of quantum electrodynamics
where the eld is also quantized. The types of interaction of radiation with atoms that
we have studied here are summarized in Figure 4.
Lecture 4 26
Before: fj fj
fj
E(0)j E(0)j E(0)j
We should note that the Einstein coecients, while derived for thermal equilibrium,
are completely general and hold also for non-equilibrium conditions. The operation of
the laser (Light Amplication by Stimulated Emission of Radiation) is based on this
principle. The idea behind this is to have some means of creating a non-equilibrium
population of states (population inversion) where Nf > Nj . Then, from (117) and (121),
and under the assumption of negligible spontaneous emission (Af j
Bf j rad (f j )) we
will have
Nf Wf j rate of emission Nf
= >1 (123)
Nj Wf j rate of absorption Nj
which shows that the applied frequency f j will be amplied in intensity by the inter-
action process, resulting in more radiation emerging than entering the system. This
process will reduce the population of the upper state until equilibrium is re-established,
so the operation of a laser also depends on having a dierent process which maintains
the population inversion of the states. As Af j grows with the third power (see equa-
tion 122) of the angular frequency f j we can expect that spontaneous emission will
dominate at high frequencies leading to signicant uncontrolled loss of energy and thus
making population inversion dicult to maintain. A practical consequence of this is
that X-ray lasers are dicult to make.
Lecture 5 28
To simplify the notation in what follows we will always assume that the electric eld is
applied along the z-axis in which case the dot product of (124) becomes
H (1) = z Ez = Ez qi zi . (126)
i
You will notice that we are already using the notation of perturbation theory as we are
representing the above term as a rst order Hamiltonian. The role of the perturbation
parameter is played here by the z-component of the electric eld Ez .
Let us now write the energy as a Taylor series expansion (perturbation expansion) with
respect to the Ez parameter at the point Ez = 0:
(0) dE 1 d2 E 2 1 d3 E
E=E + Ez + E + E3 + (128)
dEz Ez =0 2! dEz2 Ez =0 z 3! dEz3 Ez =0 z
Lecture 5 29
where dE
dEz
is the rst derivative of the energy with respect to Ez evaluated at
Ez =0
Ez = 0, etc. Of course, the zeroth order term E (0) is the value of the energy at Ez = 0.
If we now dierentiate the above Taylor expansion with respect to Ez , and substitute
for the left hand side what we found in (127), we obtain an expression for the dipole
moment in non-zero electric eld
2
dE dE 1 d3 E
z = 2
Ez 3
Ez2 + (129)
dEz Ez =0 dEz Ez =0 2 dEz Ez =0
We usually write the above expression as
1
z = 0z + zz Ez + zzz Ez2 + (130)
2
where by comparison with (129) we dene following quantities as derivatives of the
energy with respect to the electric eld at zero electric eld (Ez = 0).
The permanent dipole moment
dE
0z = = 0(0) |z |0(0) (131)
dE Ez =0
which is the rst order energy correction to the ground state wavefunction.
The polarizability 2
dE
zz = (132)
dE 2 Ez =0
and the rst hyperpolarizability
d3 E
zzz = . (133)
dE 3 Ez =0
(2)
0(0) |z |n(0) n(0) |z |0(0)
zz = 2E0 = 2 (0) (0)
. (134)
n=0 E0 En
where we have dened the dipole moment matrix elements z,mn = m(0) |z |n(0) and
(0) (0)
the denominator En0 = En E0 . This compact form can be used to express the
mean polarizabilty which is the property that is actually observed when a molecule is
rotating freely in the gas phase or in solution and one measures the average of all its
orientations to the applied eld:
1 2 x,0n x,n0 + y,0n y,n0 + z,0n z,n0
= (xx + yy + zz ) = (136)
3 3 En0
n=0
2 0n n0
= (137)
3 n=0 En0
2 |0n |2
= (138)
3 n=0 En0
At this point we can also use the closure approximation (37) to eliminate the sum over
states and derive a computationally much simpler but also more approximate expression
for the polarizability.
2 2 2(2 2 )
0n n0 = 0n n0 00 00 = (139)
3E n=0 3E n
3E
h2 e2
fn0 (142)
me E 2 n=0
Lecture 5 31
This allows us to make use of the following standard result which is known as the
Kuhn-Thomas sum rule
fn0 = Ne (143)
n
where Ne is the total number of electrons in the molecule. Notice that the sum rule
involves a summation over all states, including n = 0, but this is compatible with (142)
as f00 = 0 by denition. We therefore obtain
h2 e2 Ne
(144)
me E 2
which again shows that the polarisability increases with increasing number of electrons
and decreasing mean excitation energy. We therefore expect molecules composed of
heavy atoms to be highly polarizable.
ih (0)
n(0) |xH (0) |a(0) n(0) |H (0) x|a(0) =n |px |a(0) (147)
m
ih (0)
(Ea(0) En(0) )n(0) |x|a(0) = n |px |a(0) (148)
m
ihn(0) |px |a(0)
n(0) |x|a(0) = (0) (0)
(149)
m(Ea En )
Now substitute this relation into the left hand side of (145) as follows
(En(0) Ea(0) )|n(0) |x|a(0) |2
n
1 (0)
= (En Ea(0) )(a(0) |x|n(0) n(0) |x|a(0) + a(0) |x|n(0) n(0) |x|a(0) )
2 n
ih (0)
= (a |px |n(0) n(0) |x|a(0) a(0) |x|n(0) n(0) |px |a(0) )
2m n
ih (0)
= a |px x xpx |a(0)
2m
ih (0) (0) ih (0) (0) h2
= a |[px , x]|a = a | ih|a =
2m 2m 2m
where in the last line we have made use of the well known commutator between momen-
tum and position, [px , x] = ih.
Having proved (145) it is straighforward to show that by rearranging it, multiplying
with appropriate coecients, generalising it to three dimensions and to Ne electrons
results in the Kuhn-Thomas sum rule (143).
which of course means that its zeroth order energies are the sum of the energies for the
isolated A and B species and its eigenfunctions are the products of the eigenfunctions
of the A and B species
(0) (0) (0) (0) (0) (0) (0) (0)
H (0) |nA nB = (En(0)
A
+ En(0)
B
)|nA nB , |nA nB = |nA |nB . (151)
Lecture 5 33
This Hamiltonian completely ignores all interactions between A and B. We will now add
to it the following rst order Hamiltonian
1
H (1) = (Ax Bx + Ay By 2Az Bz ) (152)
40 R3
which (can be proved using classical electrostatics) describes the interaction between a
dipole moment on A and a dipole moment on B, the two dipoles being a distance of R
apart. Note that we have implicitly assumed here the Born-Oppenheimer approxima-
tion which means that we are only working with the electronic wavefunctions and the
distance R is not a variable in our wavefunctions but it is just a parameter on which
our calculations depend. Here we will study only the ground state.
As we have assumed that A and B have no permanent dipole moments it is easy
(0) (0) (0) (0)
to show that the rst order correction to the energy 0A 0B |H (1) |0A 0B is zero (show
this!). We therefore turn our attention to the second order energy as dened by equation
30:
(0) (0) (0) (0)
0A 0B |H (1) |nA nB nA nB |H (1) |0A 0B
(0) (0) (0) (0)
E (2) = (0) (0)
(153)
nA ,nB =(0A ,0B ) E0A 0B EnA nB
(0) (0) (0) (0) (0) (0)
0A 0B |H (1) |nA nB nA nB |H (1) |0A 0B
(0) (0)
= (0) (0)
(154)
nA ,nB =(0A ,0B ) EnA 0A + EnB 0B
from which we can deduce that the interaction is attractive (E (2) < 0) and that the
interaction energy is proportional to 1/R6 .
We can do some further manipulations to obtain a more approximate yet physically
meaningful expression for E (2) . To proceed we apply the closure approximation by
Lecture 5 34
(0) (0)
replacing nA 0A + nB 0B with the an average value EA + EB and apply equation 37:
2
(2) 2 1 1
E (A,0A nA A,nA0A )(B,0B nB B,nB 0B )
3 40 R3 EA + EB
nA ,nB =(0A ,0B )
1 1
2 2 6
2A 2B
24 0 R EA + EB
(0) (0)
where 2A = 0A |2A |0A and there is no A 2 term since we assumed that the
permanent dipole moments of A and B are zero. Having reached this stage, we can
re-express the dispersion energy by using relation (139) between the mean square dipole
moment and the polarizability (in the absence of a permanent dipole moment, 2A
3
EA ) to obtain
2 A
(2) 3 EA EB A B
E 2 2
. (158)
32 0 EA + EB R6
Finally we approximate the mean excitation energy with the ionization energy of each
species EA IA to arrive at the London formula for the dispersion energy between
two non-polar species
(2) 3 IA IB A B
E 2 2
. (159)
32 0 IA + IB R6
This very approximate expression can provide chemical insight from back of the en-
velope calculations of the dispersion energy between atoms based on readily available
quantitites such as the polarizabilities and the ionization energies. Based on this formula
we expect large, highly polarisable atoms to have strong dispersion interactions.
Lecture 6 35
where xj = {rj , j } collectively denotes the space (rj ) and spin (j ) coordinates of
electron j.
We often choose to approximate the many-electron wavefunction as a product of
single-electron wavefunctions (spinorbitals). Such a simple product of spin orbitals (also
known as a Hartree product) is not antisymmetric. To overcome this limitation we dene
the wavefunction as a Slater determinant, which is antisymmetric as the interchange of
any of its rows, which correspond to its electron coordinates, will change its sign.
In Hartree-Fock theory, we assume that the many-electron wavefunction has the form
of a Slater determinant and we seek to nd the best possible such wavefunction (for the
ground state). To achieve this goal we use the variational principle which states that
the total energy for the optimum determinant which we seek is going to be lower than
the energy calculated from any other determinant
(0) (0)
E0HF = 0 |H|0 |H| (161)
(0)
where we have assumed that the Hartee-Fock solution 0 and all trial Slater determi-
nants are normalized. E0HF is the Hartree-Fock energy for the ground state which we
are seeking. The full Hamiltonian for the electrons in a material (e.g. a molecule or a
portion of solid) has the following form
h2
Ne
1
Ne NN Ne
2 ZI e2 e2
H = i + (162)
2me i=1 i=1 I=1
40|rI ri | 2 i,j i=j 40 |ri rj |
where we have assumed that the material consists of Ne electrons and NN nuclei. The
rst term is the sum of the kinetic energy of each electron and the second term is the
sum of the electrostatic attraction of each electron from the NN nuclei, each of which is
xed (Born Oppenheimer approximation) at position rI . The nal term is the repulsive
electrostatic (Coulomb) interaction between the electrons and consists of a sum over all
distinct pairs of electrons.
3
More generally, the postulate states that a wavefunction must be antisymmetric with respect to
interchange of any pair of identical fermions (=particles with half-integer spin quantum number such
as electrons and protons) and symmetric with respect to interchange of any pair of identical bosons
(=particles with integer spin quantum number, such as photons and -particles)
Lecture 6 36
which demonstrates the fact that the rst two terms are separable into sums of one-
electron Hamiltonians hi while this is obviously not possible for the last term as each
1/|ri rj | can not be broken into a sum of a term for electron i and a term for electron
e
j. The problem of the sum of one-electron Hamiltonians N i=1 hi is computationally triv-
ial as its solutions are antisymmetrised products (Slater determinants) of one-electron
wavefunctions (=molecular spinorbitals). In contrast, because of the non-separability
of the third term, such a simple solution is not possible for H. Its solution is extremely
complicated and computationally tractable only for very small systems (e.g. the hy-
drogen molecule). Thus this is a case where perturbation theory can be very useful for
approximating the solution to H. e
As a rst attempt to apply perturbation theory we may treat the N i=1 hi part of
(165) as the zeroth order Hamiltonian and the remaining part as the perturbation. This
is not a very good choice though as the perturbation is of similar magnitude to the zeroth
order Hamiltonian. Instead, we will dene the zeroth order Hamiltonian as follows
Ne
Ne
(0)
H = HF
hi + i = fi (166)
i=1 i=1
Lecture 6 37
as a sum of Fock operators fi for each electron i. The Hartree-Fock potential for electron
i is dened as
Ne
iHF = Ja (i) Ka (i) (167)
a=1
where Ne is the total number of occupied spinorbitals and Ja (i) is the Coulomb operator
made of spinorbital a acting on electron i, and the Exchange operator Ka (i) is dened
in a similar manner. With this choice of H (0) , the H (1) is given by
1
Ne Ne
e2
H (1) = H H (0) = HF (168)
2 i,j i=j 40|ri rj | i=1 i
where now we can see, at least in a qualitative manner, that H (1) dened in this way
is much smaller than H (0) and it is therefore plausible to treat it as a perturbation.
Many-body perturbation theory using this choice of H (0) is called Mller-Plesset (MP)
perturbation theory.
Here we will develop MP theory for the ground state. First of all we observe that
any Slater determinant made of Ne spinorbitals, each of which is a solution of the
Hartree-Fock eigenvalue equation 163, is an eigenfunction of H (0) according to (prove
this!)
(0)
H (0) 0 = H (0) |a (1)b (2) . . . z (Ne )| = (a + b + . . . + z )|a (1)b (2) . . . z (Ne )| .
(169)
(0)
where here we have used the ground state determinant 0 = |a (1)b (2) . . . z (Ne )|
which consists of the Ne lowest energy spinorbitals and as we can see from this equation,
the zeroth order energy for this state is the sum of the energies of these spinorbitals
(0)
E0 = a + b + . . . + z (170)
This means that we need to go beyond the rst order energy correction to obtain an
improvement to the Hartree-Fock energy, and more specically to recover (at least some
part of) the correlation energy. Here we will conne ourselves to second order MP theory
Lecture 6 38
which is often referred to as MP2. According to (30) the second order correction to
the energy is
(2)
(0) |H (1) |(0) (0) |H (1) |(0)
0 J J 0
E0 = (0) (0)
(173)
J=0 E0 EJ
(0) (0)
We need to evaluate the matrix elements 0 |H (1) |J . Using the orthogonality prop-
erty of dierent Slater determinants, we see that for J = 0
(0) (0) (0) (0) (0) (0) (0) (0)
0 |H (0) |J = 0 |EJ |J = EJ 0 |J = 0 (174)
which leads to
(0) (0) (0) (0) (0) (0) (0) (0)
0 = 0 |H (0) |J 0 = 0 |H H (1) |J 0 |H|J = 0 |H (1) |J
(175)
(1)
which shows that the matrix elements of the H and H operators are equal. Using this
(0) (0) (0) (0)
result, and the knowledge that 0 |H|J is nonzero only when J diers from 0
by two excitations (spinorbitals) we arrive at the following result
(2) 1 xy||rsrs||xy
E0 = (176)
4 x,y r,s x + y r s
where the x, y indices run over all occupied spinorbitals (from 1 to Ne ) while the indices
r, s run over all virtual (unoccupied) spinorbitals (from Ne + 1 upwards). The two-
electron integrals are dened as follows
x (x1 )y (x2 )r (x1 )s (x2 ) x (x1 )y (x2 )s (x1 )r (x2 )
xy||rs = dx1 dx2 dx1 dx2
|r1 r2 | |r1 r2 |
(177)
in terms of spinorbitals i (r). The second order energy correction as given by equation
176 is widely used in ab initio calculations which include electron correlation and is
available in many Quantum Chemistry software packages (e.g. gaussian, NWChem,
gamess etc.). Of course, by denition, before doing an MP2 calculation one needs to
have the Hartree-Fock solutions (spinorbitals) and their energies, so in practice MP2
calculations are performed as a post-processing step after a Hartree-Fock calculation.
Example 6 Equation 175 shows that the matrix elements of H (1) between Slater de-
terminants are the same with the matrix elements of the full Hamiltonian H. Given
(0) (0) (0) rs (0) (0)
that 0 |H|J is non-zero only when J = xy , or in other words when J
is constructed by replacing no more and no less than two of any x , y ground state
spinorbitals by any two excited state spinorbitals r , s respectively, derive (176) from
(0) rs(0)
(173). Also given is the value of the non-zero matrix element: 0 |H|xy = xy||rs.
Lecture 6 39
where the factors of 1/2 are introduced in order to make sure that each distinct pair
of indices is used only once (e.g. if we have the pair x=1 and y=5, we will also have
the same pair when y=1 and x=5, so we multiply with 1/2 to make sure we count this
distinct pair only once) while the cases where x = y and/or r = s lead to zero matrix
elements so it does not matter that they are included in the sum. We now substitute
(178) into the above expression to obtain
1
Ne
(2) xy||rsrs||xy
E0 = (0) rs(0)
(180)
4 x,y=1 r,s=Ne +1 E0 Exy
1
Ne
(2) xy||rsrs||xy
E0 = . (181)
4 x,y=1 r,s=Ne +1
x + y r s
MP2 calculations with their ability to include at least some of the correlation en-
ergy, are a denite improvement over HF calculations. Figure 5 demonstrates this with
some examples of bond lengths of small molecules calculated with the two methods and
compared with experiment.
We should observe however that MP theory is also qualitatively dierent from HF
theory. The total (perturbed) Hamiltonian in MP theory (165) is the exact one, in-
volving the true electron-electron interactions (the 1/|ri rj | terms). In contrast the
HF Hamiltonian (zeroth order, H (0) ) corresponds to a system of non-interacting parti-
cles that move in an eective (averaged) potential. Thus, MP theory includes electron
Lecture 6 40
M
ole
cu
M le
et
h od
CH4 NH3 H 2O FH
correlation and the perturbed wavefunction does take into account the instant interac-
tions between electrons: the modulus of the wavefunction (and hence the probability
distribution) decreases as a function of the positions of any pair of electrons when they
are approaching each other in space. This dynamical correlation is absent from a HF
wavefunction (Slater determinant). In section 4.5 we saw that dispersion interactions
between molecules are due to instantaneous uctuations of their electronic distributions.
We expect that HF calculations would be incapable of predicting dispersion interactions
while MP calculations should be able to. This is indeed the case. For example, a HF
calculation predicts zero binding between two Ne atoms, while an MP2 calculation pre-
dicts binding with an equilibrium distance of 6.06 a.u. and a binding energy of 2.3 meV.
The exact values for these quantities are 5.84 a.u. and 3.6 meV respectively. There
are numerous cases where dispersion interactions play a key role. In computational sim-
ulations involving such cases methods like MP theory need to be used. Figure 6 shows
some examples of materials with dispersion interactions.
Lecture 6 41
a b c