Reaction Dynamics
Reaction Dynamics
Contents
I. Introduction 2
V. Further Reading 29
VI. Appendices 30
A. The Dirac delta function 30
B. The Heisenberg picture 32
C. Liouville’s theorem 33
2
I. INTRODUCTION
The simplest possible continuum wave functions are the eigenstates of the free particle
Hamiltonian
p̂2
Ĥ0 = . (1)
2m
Since Ĥ0 commutes with p̂, its eigenstates can be chosen to be eigenstates of the momentum
operator. Calling these eigenstates |pi and the momentum eigenvalues p, one has both
and
p̂ |pi = p |pi , (3)
with E = p2 /2m.
It is straightforward to find the form of the momentum eigenstates in the position repre-
sentation, where
d
p̂ = −ih̄ (4)
dx
and
hx|pi = φp (x) (5)
Note that this also gives the form of the position eigenstates in the momentum representa-
tion,
φp (x)∗ ≡ hp|xi = N ∗ e−ipx/h̄ . (8)
The normalization of the momentum eigenstates is a little tricky, because the normal-
ization integral hp|pi is undefined. (In other words, the momentum eigenstate |pi is not
a “proper” state vector in the space of square integrable functions in one dimension, and
neither for that matter is the position eigenstate |xi. Both are nevertheless extremely useful
in quantum scattering theory as we are about to see.) The only way to make any sense
of the situation is to recognize that the overlap integral between two different momentum
eigenstates,
Z ∞
hp0 |pi = dx φp0 (x)∗ φp (x)
−∞
Z ∞
= dx hp0 |xi hx|pi
−∞
Z ∞
0
= |N |2 dx e−i(p −p)x/h̄ , (9)
−∞
0 1 Z∞ 0
δ(p − p) = dx e−i(p −p)x/h̄ . (10)
2πh̄ −∞
This suggests choosing N = (2πh̄)−1/2 in Eqs. (7) and (8), so that3
and
hp|xi = (2πh̄)−1/2 e−ipx/h̄ (12)
giving
hp0 |pi = δ(p0 − p). (13)
The momentum eigenstates form a complete set of states because they are the only
eigenstates of the free particle Hamiltonian H0 (which does not support any bound states).
This means that any state |ψi of a one-dimensional system can be expanded in terms of the
momentum eigenstates as
Z ∞
|ψi = dp |pi hp|ψi . (14)
−∞
4
or equivalently
Z ∞
1
ψ(x) = 1/2
dp e+ipx/h̄ ψ(p). (16)
(2πh̄) −∞
Thus the position eigenstates also form a complete set of states, and indeed setting |ψi = |x0 i
in Eq. (15) gives
Z ∞
hx|x0 i = dp hx|pi hp|x0 i
−∞
1 Z∞ 0
= dp e−ip(x −x)/h̄
(2πh̄) −∞
= δ(x0 − x). (20)
Equations (13), (14), (19) and (20) are the main results so far. They show that the
momentum eigenstates |pi and the position eigenstates |xi each form a complete set of
orthonormal states. These results are entirely analogous to those obtained in a discrete
basis of “proper” square integrable state vectors such as the eigenstates |ni of a harmonic
oscillator, in which one has the more familiar orthonormality condition
The only differences that arise in the momentum and position eigenstate bases are that
the Kronecker delta in Eq. (21) becomes a Dirac delta function and the sum over states in
Eq. (22) becomes an integral.
5
One other thing we shall need to know how to do is calculate the trace of an operator in
these two bases. In a basis set of harmonic oscillator eigenstates, the trace of an operator Â
is
h i ∞ D
X E
tr  = n|Â|n , (23)
n=0
and this is independent of the force constant that is chosen for the harmonic oscillator
because the transformation between any one (complete, orthonormal) basis and any other
is a unitary transformation. In the momentum and position eigenstate bases, one can write
either
h i Z ∞ D E
tr  = dp p|Â|p , (24)
−∞
or
h i Z ∞ D E
tr  = dx x|Â|x , (25)
−∞
both of which will again give the same result as can be seen from the following argument:
Z ∞ D E Z ∞ Z ∞ Z ∞ D E
dp p|Â|p = dp dx dx0 hp|xi x|Â|x0 hx0 |pi
−∞ −∞ −∞ −∞
1 Z∞ Z ∞ Z ∞
0
D E
= dp dx dx0 e−ip(x−x )/h̄ x|Â|x0
2πh̄ −∞ −∞ −∞
Z ∞ Z ∞ D E
= dx dx0 δ(x − x0 ) x|Â|x0
−∞ −∞
Z ∞ D E
= dx x|Â|x . (26)
−∞
Exercise 1. Any operator  that acts on the space of square integrable functions in one
dimension can be written in terms of the eigenstates |ni of a harmonic oscillator as
∞ X
∞ D E
|ni n|Â|n0 hn0 | .
X
 = (27)
n=0 n0 =0
(a) Why is this? (b) Use this expression to show that Eq. (25) will give the same result as
h i
Eq. (23) for tr  .
B. Scattering eigenstates
p̂2
Ĥ = Ĥ0 + V̂ = + V (x̂), (28)
2m
6
V(x)
in which the potential energy function V (x) tends to zero as x → ±∞. The scattering
eigenstates of this problem are the solutions of the Schrödinger equation
The reflection amplitude R(E) and the transmission amplitude T (E) can be obtained for any
potential V (x) by solving the Schrödinger equation in the interaction region. For example,
in the trivial case where the potential is zero, the only solution of the Schrödinger equation
that is compatible with both boundary conditions is ψp (x) = φp (x), giving R(E) = 0 and
T (E) = 1.
in which V0 > 0. Use the fact that ψp (x) and ψp0 (x) must both be continuous at x = 0 and
x = a to obtain explicit expressions for the reflection amplitude R(E) and the transmission
amplitude T (E) when E = V0 , and verify that your expressions satisfy
Equation (33) shows that the reflection and transmission probabilities for the problem
considered in exercise 2 sum to one (“unitarity”). This is a general result that can be shown
to hold for any potential V (x) and any scattering energy E as follows.
The flux (number of particles per unit time) passing through the point x = s is given in
quantum mechanics by the Heisenberg time derivative of the projection operator onto the
subspace in which x > s, namely (see Appendix B)
ih i
F̂ (s) = Ĥ, h(x̂ − s) , (34)
h̄
where
1, if x > s
(
h(x − s) = . (35)
0, if x < s
For a Hamiltonian of the form in Eq. (28), it is straightforward (and therefore left as an
exercise) to show that F̂ (s) can be written in the position representation as
( )
ih̄ d d
F̂ (s) = − δ(x − s) + δ(x − s) . (36)
2m dx dx
8
The flux through s for a general time-dependent wave function ψ(x, t) is therefore
∂ψ(s, t) ∗
( )
ih̄ ∗ ∂ψ(s, t)
j(s, t) ≡ hψ| F̂ (s) |ψi = − ψ(s, t) − ψ(s, t) , (37)
2m ∂s ∂s
and it is equally straightforward to show (another exercise) that this satisfies the quantum
mechanical continuity equation
∂ ∂
j(s, t) = − |ψ(s, t)|2 . (38)
∂s ∂t
The most important implication of these equations is that the flux j(s, t) will be a con-
stant (independent of both s and t) when ψ(s, t) = ψ(s, 0)e−iEt/h̄ is an eigenstate of the
Hamiltonian (a stationary state). If this stationary state is a bound state its wave function
ψ(s, 0) may be chosen to be real, and it follows from this that j(s, t) will be identically
zero for all s and t. So bound states do not contribute any flux, but scattering states do:
inserting the scattering eigenstate |ψp i into Eq. (37), and choosing s to be in the asymptotic
“product” region where the boundary condition in Eq. (31) applies, we obtain
Exercise 3. (a) Use Eqs. (28) and (34) to derive the expression for the flux operator
in Eq. (36). (b) Derive the continuity equation in Eq. (38) from Eq. (37) and the time-
dependent Schrödinger equation in the form
h̄2 ∂ 2
" #
∂ i
ψ(s, t) = − − + V (s) ψ(s, t).
∂t h̄ 2m ∂s2
(c) Verify the result in Eq. (40).
9
In the language of linear second order differential equations, the first term on the right-hand
side of Eq. (43) is a “complementary function” and the second is a “particular integral”; the
i in this integral is a tiny embellishment that is needed to make the inverse of the operator
(E − Ĥ0 ) well defined (see exercise 4 below).
In the position representation, Eq. (43) becomes an integral equation for the scattering
wave function ψp (x) ≡ hx|ψp i,
Z ∞
ψp (x) = χp (x) + dx0 G+ 0 0 0
0 (x, x )V (x )ψp (x ), (45)
−∞
where
0 −1
G+
0 (x, x ) = lim hx| (E + i − Ĥ0 ) |x0 i . (46)
→0
q
Letting p = 2m(E + i), and noting that Ĥ0 = p̂2 /2m, one can evaluate this free particle
Green’s function as follows:
0 −1
G+
0 (x, x ) = lim hx| (E + i − Ĥ0 ) |x0 i
→0
Z ∞
= lim dp0 hx| (E + i − p̂2 /2m)−1 |p0 i hp0 |x0 i
→0 −∞
Z ∞
= lim dp0 hx|p0 i (p2 /2m − [p0 ]2 /2m)−1 hp0 |x0 i
→0 −∞
0 0
2m Z ∞ 0 e+ip (x−x )/h̄
= lim dp 2
→0 2πh̄ −∞ (p − [p0 ]2 )
0 0
m Z ∞ 0 e+ip (x−x )/h̄
= − lim dp 0 . (47)
→0 πh̄ −∞ (p − p )(p0 + p )
10
Im p’
x−x’>0
+pε
Re p’ .
−pε
x−x’<0
q
For small positive , the poles in the integrand of Eq. (47) at p0 = ±p = ± 2m(E + i)
are shifted off the real integration axis into the first and third quadrants of the complex
p0 plane (see above). The integral is therefore well defined and can be evaluated using the
residue theorem. When x > x0 , the integration contour can be closed along a semicircle at
infinity in the upper half of the complex plane, enclosing the simple pole at p0 = +p . The
semicircle does not make any contribution to the contour integral and the residue theorem
gives
0
0 m m e+ip (x−x )/h̄
G+ = − lim
0 (x, x ) 2πiR(+p ) = − lim 2πi
→0 πh̄ →0 πh̄ 2p
im +ip (x−x0 )/h̄ im 0
= − lim e = − e+ip(x−x )/h̄ (x > x0 ), (48)
→0 h̄p h̄p
√
where p = lim→0 p = 2mE. When x < x0 , the contour can be closed along a semicircle
at infinity in the lower half of the complex plane (which again makes no contribution), and
the residue theorem gives
0
0 m m e−ip (x−x )/h̄
G+
0 (x, x ) = + lim 2πiR(−p ) = − lim 2πi
→0 πh̄ →0 πh̄ 2p
im −ip (x−x0 )/h̄ im −ip(x−x0 )/h̄
= − lim e =− e (x < x0 ), (49)
→0 h̄p h̄p
where the initial sign change arises because the contour is now clockwise. So overall
where φ±p (x) = (2πh̄)−1/2 e±ipx/h̄ and x< (x> ) is the lesser (greater) of x and x0 .
The final stage of the argument is to substitute this back into Eq. (45) and to use the
boundary conditions on ψp (x) in Eqs. (30) and (31) to determine the homogeneous solution
χp (x). As x → −∞, Eqs. (45) and (50) give
Z ∞
2πim
ψp (x → −∞) ∼ χp (x) − φ−p (x) dx0 φp (x0 )V (x0 )ψp (x0 )
p −∞
Z ∞
2πim
= χp (x) − φ−p (x) dx0 φ−p (x0 )∗ V (x0 )ψp (x0 ). (51)
p −∞
Comparing this with Eq. (30), and noting that the coefficients of φp (x) and φ−p (x) must be
the same in both equations, fixes
χp (x) = φp (x) (52)
and gives
2πim D E
R(E) = − −p|V̂ |ψp . (53)
p
Finally, as x → ∞, Eqs. (45), (50) and (52) give
Z ∞
2πim
ψp (x → ∞) ∼ φp (x) − φp (x) dx0 φ−p (x0 )V (x0 )ψp (x0 )
p −∞
Z ∞
2πim
= φp (x) − φp (x) dx0 φp (x0 )∗ V (x0 )ψp (x0 ), (54)
p −∞
2πim D E
T (E) = 1 − p|V̂ |ψp . (55)
p
Exercise 4. The need for the i in Eq. (43) arises because the resolvent operator (z − Ĥ0 )−1
has a branch cut along the positive real z axis, on which z = E is an eigenvalue of Ĥ0 . Prove
this by showing that when the real axis is approached from below, as it is in
G− 0
0 (x, x ) = lim hx| (E − i − Ĥ0 )
−1
|x0 i , (56)
→0
one obtains a result that differs from the one in Eq. (50). (The reason why we have chosen
to focus on G+ 0
0 (x, x ) here is that it is consistent with the boundary conditions on ψp (x)
and one can clearly define an analogous operator for the full scattering problem:
−1
Ĝ+ (E) = lim E + i − Ĥ . (58)
→0
These two operators are related by the fact that Ĥ = Ĥ0 + V̂ , which implies that
−1
Ĝ+
0 (E) = Ĝ+ (E)−1 + V̂ . (59)
These two equations are known as the Lippmann-Schwinger equations4 for Ĝ+ (E).
In view of Eqs. (43) and (52), the free particle Green’s operator Ĝ+
0 (E) provides a link
between the scattering eigenstate |ψp i and the free particle (momentum) eigenstate |pi:
and
h ih i
1 − Ĝ+
0 (E)V̂ 1 + Ĝ+ (E)V̂ = 1, (65)
13
where Eq. (64) follows from Eq. (60) and Eq. (65) from Eq. (61); therefore
h i−1 h i
1 − Ĝ+
0 (E)V̂ = 1 + Ĝ+ (E)V̂ (66)
and
h i
|ψp i = 1 + Ĝ+ (E)V̂ |pi . (67)
Note in passing that this expression implies that Eq. (55) can be written more symmetrically
as
2πim
T (E) = 1 − hp| T̂ (E) |pi (68)
p
where T̂ (E) is the transition operator4
Returning to Eq. (67), and recalling that V̂ = Ĥ − Ĥ0 and Ĥ0 |pi = E |pi, one finds that
the following formal manipulations1 give yet another expression for |ψp i:
h i
|ψp i = 1 + Ĝ+ (E)V̂ |pi
h i
= 1 − Ĝ+ (E)(E − Ĥ) |pi
h i
= lim 1 − (E + i − Ĥ)−1 (E − Ĥ) |pi
→0
h i
= lim 1 − (E + i − Ĥ)−1 (E + i − Ĥ − i) |pi
→0
The physical interpretation of this result is that the scattering eigenstate |ψp i can be obtained
by propagating the free-particle eigenstate |pi back into the infinite past under the influence
of the free-particle Hamiltonian Ĥ0 , and then propagating the result forward to time t = 0
under the influence of the full Hamiltonian Ĥ.
An important implication of Eq. (70) is that the full scattering eigenstates are normalized
in the same way as the free particle eigenstates:
If the Hamiltonian Ĥ does not support any bound states, this implies by analogy with
Eqs. (14) and (24) that any state |ψi can be expanded in terms of the scattering eigenstates
as
Z ∞
|ψi = dp |ψp i hψp |ψi , (73)
−∞
If Ĥ does support some bound states |ψb i, such that Ĥ |ψb i = Eb |ψb i with hψb0 |ψb i = δb0 b
and Eb0 , Eb < 0, these will be orthogonal to the scattering states as they are eigenstates of
the same Hermitian operator with different eigenvalues (recall that Ĥ |ψp i = E |ψp i with
E = p2 /2m > 0). It follows from this that each of the above equations will simply be
augmented by a bound-state contribution:
X Z ∞
|ψi = |ψb i hψb |ψi + dp |ψp i hψp |ψi , (75)
b −∞
and
h i X Z ∞
tr  = hψb |  |ψb i + dp hψp |  |ψp i . (76)
b −∞
Note finally that, according to the argument in Eq. (72), the operator Ω̂†+ Ω̂+ behaves like
a unit operator when acting on any free particle eigenstate:
Since the free particle eigenstates are complete (Ĥ0 does not support any bound states), this
implies that Ω̂†+ Ω̂+ is a unit operator:
If it were also true that Ω̂+ Ω̂†+ = 1, then Ω̂+ would be unitary. However, this is not true
when the Hamiltonian Ĥ supports any bound states, as can be seen from the following
argument:
hp| Ω̂†+ |ψb i = hψb | Ω̂+ |pi∗ = hψb |ψp i∗ = 0. (79)
Since this is true for all |pi, and the free particle eigenstates are complete, it implies that
for all bound states |ψb i, and it is immediately apparent from this that
(in fact, Ω̂+ Ω̂†+ is just a projection operator onto the subspace of scattering eigenstates,
as shown in exercise 5 below). An operator Ω̂+ that satisfies Eqs. (78) and (81) is norm-
preserving (or isometric), but it is not unitary.4
h i
Exercise 5. Starting with Eq. (62) in the form |pi = 1 − Ĝ+
0 (E)V̂ |ψp i, and recalling that
V̂ = Ĥ − Ĥ0 and Ĥ |ψp i = E |ψp i, show by following the steps in Eq. (70) that |pi = Ω†+ |ψp i
where Ω†+ = limt→∞ e−iĤ0 t/h̄ e+iĤt/h̄ . This implies that Ω̂+ Ω̂†+ |ψp i = Ω̂+ |pi = |ψp i for any
|ψp i, which in conjunction with Eq. (80) shows that Ω̂+ Ω̂†+ is a projection operator onto the
subspace of scattering eigenstates; a simpler way to see this is to note that
Z ∞ Z ∞
Ω̂+ Ω̂†+ = Ω̂+ 1̂ Ω̂†+ = dp Ω̂+ |pi hp| Ω̂†+ = dp |ψp i hψp | .
−∞ −∞
The one-dimensional barrier transmission problem illustrated in the figure on page 6 is the
simplest possible model for a bimolecular chemical reaction, with the reactant asymptote at
16
Exercise 6. Verify that k(T ) in Eq. (82) has the correct dimensions for a bimolecular (sec-
ond order) rate coefficient in one-dimensional space, where “concentrations” are measured
in terms of the number of particles per unit length.
Z ∞ !
m
= dE hψp | e−β Ĥ/2 F̂ e−β Ĥ/2 |ψp i , (86)
0 p
where we have used the fact that Ĥ |ψp i = E |ψp i and chosen to split the Boltzmann operator
symmetrically around the flux operator for later convenience.
We would now like to change the integration variable from energy to momentum in
Eq. (86), so as to obtain an expression that more closely resembles an operator trace [see
√
Eq. (24)]. The correct way to do this is to set p = + 2mE, which along with dE = (p/m)dp
gives
Z ∞
k(T )Qr (T ) = dp hψp | e−β Ĥ/2 F̂ e−β Ĥ/2 |ψp i . (87)
0
The steps from here to Miller’s trace formula are a series of straightforward manipulations
that begin by replacing |ψp i with Ω̂+ |pi:
Z ∞
k(T )Qr (T ) = dp hp| Ω̂†+ e−β Ĥ/2 F̂ e−β Ĥ/2 Ω̂+ |pi
Z0∞
= dp hp| Ω̂†+ e−β Ĥ/2 F̂ e−β Ĥ/2 Ω̂+ |pi h(p)
−∞
Z ∞
= dp hp| Ω̂†+ e−β Ĥ/2 F̂ e−β Ĥ/2 Ω̂+ h(p̂) |pi
−∞
h i
= tr Ω̂†+ e−β Ĥ/2 F̂ e−β Ĥ/2 Ω̂+ h(p̂)
h i
= tr e−β Ĥ/2 F̂ e−β Ĥ/2 Ω̂+ h(p̂)Ω̂†+ . (88)
The physical interpretation of this result is that the rate coefficient can be obtained by
correlating the thermal flux e−β Ĥ/2 F̂ e−β Ĥ/2 through the dividing surface at time t = 0 with
the projection
onto states with positive momentum (which are incident on the potential from the reactant
side) in the infinite past.1
As it stands, Eq. (88) is already quite appealing. Since the trace can be evaluated in
any convenient basis, we are no longer obliged to solve the Schrödinger equation for the
scattering eigenstates |ψp i in order to calculate the reaction rate. And since the equation is
18
guaranteed to give the same result for any choice of dividing surface x = s, we can make this
choice in such a way as to simplify the calculation (as will become apparent when we come on
to consider the classical limit of the equation and its transition state theory approximation
below). However, before we get on to these things, let us spend some time seeing if we can
make Eq. (88) look even prettier than it already is.
The first step in this direction is to note that the projection operator h(p̂) can be written
equivalently as either2
h(p̂) = lim e−iĤ0 t/h̄ h(s − x̂)e+iĤ0 t/h̄ , (90)
t→∞
or
h(p̂) = lim e+iĤ0 t/h̄ h(x̂ − s)e−iĤ0 t/h̄ , (91)
t→∞
for any choice of the dividing surface x = s. In other words, positive momentum states
are those that evolve under the influence of the free particle Hamiltonian from the left of
the dividing surface in the infinite past [Eq. (90)] to the right of the dividing surface in the
infinite future [Eq. (91)]. This clearly makes good physical sense, and it can be shown to
be correct by proving that the matrix elements of both sides of Eqs. (90) and (91) are the
same. For example, Eq. (90) can be verified as follows.2
Choosing the position eigenstates as a basis, the matrix elements of h(p̂) are simply
Z ∞
0 1 Z∞ 0 0
hx| h(p̂) |x i = dp hx|pi h(p) hp|x i = dp e+ip(x−x )/h̄ . (92)
−∞ 2πh̄ 0
To check that the right-hand side of Eq. (90) has the same matrix elements, we can begin
by writing
Z ∞
hx| e−iĤ0 t/h̄ h(s − x̂)e+iĤ0 t/h̄ |x0 i = dx00 hx| e−iĤ0 t/h̄ |x00 i h(s − x00 ) hx00 | e+iĤ0 t/h̄ |x0 i . (93)
−∞
This contains two free particle propagator matrix elements, which can be shown (see exer-
cise 7 below) to be
1/2
m
−iĤ0 t/h̄ 00 00 )2 /(2h̄t)
hx| e |x i = e+im(x−x , (94)
2πih̄t
and
1/2
im
00 0 00 −x0 )2 /(2h̄t)
hx | e +iĤ0 t/h̄
|x i = e−im(x . (95)
2πh̄t
Substituting these results into Eq. (93) and rearranging gives
and changing the integration variable to p = m(s − x00 )/t converts this into
0 0 1 Z∞ 0
hx| e−iĤ0 t/h̄ h(s − x̂)e+iĤ0 t/h̄ |x0 i = e+im(x+x −2s)(x−x )/(2h̄t) dp e+ip(x−x )/h̄
2πh̄ 0
0 0
= e+im(x+x −2s)(x−x )/(2h̄t) hx| h(p̂) |x0 i . (97)
Hence
lim hx| e−iĤ0 t/h̄ h(s − x̂)e+iĤ0 t/h̄ |x0 i = hx| h(p̂) |x0 i (98)
t→∞
(for all x, x0 and s), and since the position eigenstate basis is complete this completes the
proof of Eq. (90).
Of Eqs. (90) and (91), Eq. (90) is the more convenient for us, owing to the presence of
the Møller operators Ω̂+ and Ω̂†+ in Eq. (89). Indeed combining Eqs. (89) and (90) gives
Ω̂+ h(p̂)Ω̂†+ = lim e−iĤt/h̄ e+iĤ0 t/h̄ e−iĤ0 t/h̄ h(s − x̂)e+iĤ0 t/h̄ e−iĤ0 t/h̄ e+iĤt/h̄
t→∞
This is almost in the neatest possible form, but there is one more useful modification we
can make to it, based on the observation that
the unit operator in Eq. (101) does not contribute to k(T )Qr (T ), and we are left with
h i
k(T )Qr (T ) = − lim e−β Ĥ/2 F̂ e−β Ĥ/2 e−iĤt/h̄ ĥe+iĤt/h̄ , (103)
t→∞
20
Im p
P
Re p .
−P π/4
where
ĥ = h(x̂ − s) (104)
ih i
and F̂ = Ĥ, ĥ is the Heisenberg time-derivative of ĥ.
h̄
can be evaluated along the contour illustrated above to give the result in Eq. (94). (b) What
is the appropriate contour for the integral in Eq. (95), assuming again that t > 0?
Exercise 8. Explain carefully why the sum and the integral in the last line of Eq. (102)
both vanish.
i
Since F̂ = [Ĥ, ĥ] is the Heisenberg time derivative of ĥ, one might expect this to be related
h̄
to the following side-side
h i
css (t) = tr e−β Ĥ/2 ĥe−β Ĥ/2 e+iĤt/h̄ (1 − ĥ)e−iĤt/h̄ (108)
and flux-flux
h i
cf f (t) = tr e−β Ĥ/2 F̂ e−β Ĥ/2 e+iĤt/h̄ F̂ e−iĤt/h̄ (109)
correlation functions, and indeed it is. Differentiating Eq. (107) with respect to t immediately
gives
d h i
cf s (t) = tr e−β Ĥ/2 F̂ e−β Ĥ/2 e+iĤt/h̄ F̂ e−iĤt/h̄ = cf f (t), (110)
dt
and differentiating Eq. (108) with respect to t and rearranging gives
d h i
css (t) = −tr e−β Ĥ/2 ĥe−β Ĥ/2 e+iĤt/h̄ F̂ e−iĤt/h̄
dt h i
= −tr e−iĤt/h̄ ĥe+iĤt/h̄ e−β Ĥ/2 F̂ e−β Ĥ/2
h i
= −tr e−β Ĥ/2 F̂ e−β Ĥ/2 e−iĤt/h̄ ĥe+iĤt/h̄ = −cf s (−t), (111)
where we have used the fact that the Boltzmann operator e−β Ĥ/2 commutes with the evolu-
tion operators e±iĤt/h̄ and that one can cyclically permute the operators within a trace.
These same manipulations reveal that both css (t) and cf f (t) are real and even functions
of t. For example,
h i
css (t)∗ = css (−t) = tr e−β Ĥ/2 ĥe−β Ĥ/2 e−iĤt/h̄ (1 − ĥ)e+iĤt/h̄
h i h i
= tr e−β Ĥ ĥ − tr e−β Ĥ/2 ĥe−β Ĥ/2 e−iĤt/h̄ ĥe+iĤt/h̄
h i h i
= tr e−β Ĥ ĥ − tr e+iĤt/h̄ ĥe−iĤt/h̄ e−β Ĥ/2 ĥe−β Ĥ/2
h i h i
= tr e−β Ĥ ĥ − tr e−β Ĥ/2 ĥe−β Ĥ/2 e+iĤt/h̄ ĥe−iĤt/h̄
h i
= tr e−β Ĥ/2 ĥe−β Ĥ/2 e+iĤt/h̄ (1 − ĥ)e−iĤt/h̄ = css (t), (112)
and similarly for cf f (t). In view of Eq. (111), this implies that cf s (t) is a real and odd
function of t, and combining this with Eqs. (106) and (110) gives
Z ∞
d
k(T )Qr (T ) = lim css (t) = lim cf s (t) = dt cf f (t). (113)
t→∞ dt t→∞ 0
These final results were originally obtained by Miller et al.2 in 1983. They are entirely rigor-
ous, they apply equally well to more complicated reactions than the simple one-dimensional
22
model problem we have considered here, and they form the basis of all modern approaches
to the calculation of chemical reaction rates.
Exercise 9. (a) Show that css (t) in Eq. (108) can be evaluated as
Z ∞ Z s 2
css (t) = dx dx0 hx0 | e−iĤtc /h̄ |xi , (114)
s −∞
where tc = t − iβh̄/2. (b) Now consider the special case where Ĥ = Ĥ0 and assume that
t > 0 so that hx0 | e−iĤ0 tc /h̄ |xi can be evaluated using Eq. (94). Show that in this case
m Z∞ Zs 0 2 2
css (t) = dx dx0 e−βm(x −x) /(2|tc | ) . (115)
2πh̄|tc | s −∞
There is a straightforward and well-defined procedure for obtaining the classical limit of
a quantum mechanical trace expression such as the definition of cf s (t) in Eq. (107): One
simply replaces the trace with a classical phase space average and all of the operators within
the trace by the corresponding classical functions of positions and momenta.
For example, the classical (high temperature) limit of the partition function for a particle
in a one-dimensional box of length a, which will be familiar from elementary statistical
mechanics courses as
∞ ∞
2 π 2 h̄2 /(2ma2 )
h i
Q(T ) = tr e−β Ĥ = e−βEn = e−βn
X X
n=1 n=1
!1/2 !1/2
Z ∞
−βn2 π 2 h̄2 /(2ma2 ) 1 2πma2 a 2πm
∼ dn e = = , (117)
0 2 βπ 2 h̄2 2πh̄ β
23
can be calculated more directly (without using the quantum mechanical energy levels) as
1 Z∞ Z ∞
Q(T ) = dp dx e−βH(p,x) (118)
2πh̄ −∞ −∞
where
p2
H(p, x) = + V (x) (119)
2m
with
0, 0 < x < a,
(
V (x) = (120)
∞, x < 0, x > a.
Indeed substituting Eqs. (119) and (120) into Eq. (118) immediately gives
!1/2
1 Z∞ Z a
2 a 2πm
Q(T ) = dp dx e−βp /2m = . (121)
2πh̄ −∞ 0 2πh̄ β
The classical analogue of the flux operator F̂ in cf s (t) is given by Eq. (36) as
1 p
F (p, x) = {δ(x − s)p + p δ(x − s)} = δ(x − s), (122)
2m m
is clearly
h(p, x) = h[xt − s], (124)
where xt ≡ xt (p, x) is obtained from p and x by solving the classical equations of motion
∂H(pt , xt ) dV (xt )
ṗt = − =− (125)
∂xt dxt
and
∂H(pt , xt ) pt
ẋt = + = (126)
∂pt m
subject to the initial conditions p0 = p and x0 = x. The classical limit of the rate coefficient
of our model one-dimensional reaction is therefore given by Eq. (113) as
1 Z∞ Z ∞
p
ccl
f s (t) = dp dx e−βH(p,x) δ(x − s)h(xt − s)
2πh̄ −∞ −∞ m
1 Z∞ Z ∞
p0
≡ dp0 dx0 e−βH(p0 ,x0 ) δ(x0 − s)h(xt − s). (128)
2πh̄ −∞ −∞ m
As in the quantum mechanical case, this classical flux-side correlation function is a real
and odd function of t, and it can be obtained from the time derivative of a classical side-side
correlation function
1 Z∞ Z ∞
ccl
ss (t) = dp0 dx0 e−βH(p0 ,x0 ) h(x0 − s) [1 − h(xt − s)] . (129)
2πh̄ −∞ −∞
d cl 1 Z∞ Z ∞
pt
css (t) = − dp0 dx0 e−βH(p0 ,x0 ) h(x0 − s) δ(xt − s), (130)
dt 2πh̄ −∞ −∞ m
and the fact that classical dynamics conserves the Boltzmann factor
d cl 1 Z∞ Z ∞
pt
css (t) = − dpt dxt e−βH(pt ,xt ) h(x0 − s) δ(xt − s)
dt 2πh̄ −∞ −∞ m
1 Z∞ Z ∞
p0
≡− dp0 dx0 e−βH(p0 ,x0 ) h(x−t − s) δ(x0 − s)
2πh̄ −∞ −∞ m
cl
= −cf s (−t), (133)
Exercise 10. Use Eqs. (131) and (132) to prove that ccl
ss (t) is a even function of t. It
d cl
k cl (T )Qr (T ) = lim c (t); (135)
t→∞ dt ss
The classical limit of the flux-side correlation function introduced above is clearly very
similar to its quantum mechanical counterpart. Both are real and odd functions of t and both
are independent of the location of the dividing surface in the limit as t → ∞. However, there
is one essential difference between the two that leads to a well-defined classical transition
state theory: ccl
f s (t) is discontinuous at t = 0 and tends to a non-zero value as t → 0 from
we have
1 Z∞ Z ∞
p0
lim ccl
f s (t) = dp0 dx0 e−βH(p0 ,x0 ) δ(x0 − s)h(p0 )
t→0+ 2πh̄ −∞ −∞ m
1 Z∞ p0 −βp20 /2m Z ∞
= dp0 e dx0 e−βV (x0 ) δ(x0 − s)
2πh̄ 0 m −∞
1 −βV (s)
= e , (137)
2πβh̄
theory result
1 1 kB T 1
kclTST (T ) = lim ccl
f s (t) = e−βV (s) ≡ e−βV (s) . (138)
Qr (T ) t→0+ 2πβh̄Qr (T ) h Qr (T )
Note that:
(a) In undergraduate text books, the transition state theory rate coefficient is usually
written as
kB T Q‡ (T ) −∆E0‡ /kB T
k TST (T ) = e , (139)
h Qr (T )
where Qr (T ) is the partition function of the reactants (per unit volume in three di-
mensions or per unit length in one dimension), Q‡ (T ) is the partition function of the
transition state excluding the contribution from motion along the reaction coordi-
nate, and ∆E0‡ is the difference in zero point energies between the reactants and the
transition state. For the simple model reaction we are considering here, the reaction
26
(b) With this choice of dividing surface, the transition state theory approximation in
Eq. (138) will actually give the same result (for our model one-dimensional reaction)
as the full classical rate in Eq. (127). This can be seen by setting s = x‡ in Eq. (128):
1 Z∞ Z ∞
p0
ccl
f s (t) = dp0 dx0 e−βH(p0 ,x0 ) δ(x0 − x‡ )h(xt − x‡ ). (140)
2πh̄ −∞ −∞ m
If p0 is positive, a classical trajectory starting at the top of the reaction barrier will
clearly evolve to the right of the barrier as time increases and so make a positive
contribution to the integral in Eq. (140) for t > 0. However, if p0 is negative, the
trajectory will evolve to the left of the barrier and contribute nothing. For any positive
time t, we can therefore replace the step function h(xt − x‡ ) in the integrand with a
step function in the initial momentum,
1 −βV (x‡ )
ccl
f s (t > 0) = e (142)
2πβh̄
and therefore
1 1 ‡ kB T 1 ‡
k cl (T ) = lim ccl
f s (t) = e−βV (x ) ≡ e−βV (x ) . (143)
Qr (T ) t→∞ 2πβh̄Qr (T ) h Qr (T )
(c) For any other choice of dividing surface, we will have V (s) ≤ V (x‡ ), and therefore
‡
e−βV (s) ≥ e−βV (x ) and [compare Eqs. (138) and (143)]
kclTST (T ) ≥ k cl (T ). (144)
This variational aspect of transition state theory carries over to reactions with more
degrees of freedom and enables one to find the optimum dividing surface by minimizing
27
the transition state theory rate (at least in principle; in practice this is quite difficult
to do properly for reactions with more than a handful of degrees of freedom). The
one peculiarity of the one-dimensional case is that the resulting optimized kclTST (T ) is
exactly equal to k cl (T ) (as shown above); this is not generally true for reactions with
more degrees of freedom.
Exercise 11. Show by combining Eq. (138) with Eq. (83) that the optimum transition
state theory rate coefficient can be written as
1 ‡
kclTST (T ) = h|ẋ|i e−βV (x ) , (145)
2
where
Z ∞ Z ∞
1 p 2 2
h|ẋ|i = dp e−βp /2m / dp e−βp /2m (146)
2 0 m −∞
is the thermal expectation value of the classical flux through the dividing surface and the
‡
factor of e−βV (x ) is the probability that a thermal fluctuation will bring the classical system
to the top of the reaction barrier.
We have just seen that classical transition state theory is obtained by taking the short-
time limit of the classical flux-side correlation function. This gives a well-defined approx-
imation to the classical rate coefficient because ccl
f s (t) is discontinuous at t = 0. However,
one cannot do the same thing in quantum mechanics because the exact quantum mechan-
ical cf s (t) is a real, odd, and continuous function of t (as you have shown for the case of
free motion in exercise 9). Thus there is no true quantum mechanical analogue of classical
transition state theory (see below).
The ubiquitous (and generally quite successful) use of transition state theory to calculate
reaction rates in the classical limit and the need to allow for quantum mechanical zero
point energy and tunneling effects in reactions involving hydrogen atoms have nevertheless
led to the development of a number of transition state theory-like approximations to the
quantum rate coefficient. The simplest of these is the undergraduate transition state theory
expression in Eq. (139). For a multi-dimensional reaction, this expression assumes that the
motion along the reaction coordinate is separable from the motion on the dividing surface
28
cfs(t)/Qr(T) cfs(t)/Qr(T)
exact
rate
TST limit classical
No TST rate .
.
t t
quantum classical
.
and treats the former classically and the latter quantum mechanically. It therefore includes
zero point energy in the modes orthogonal to the reaction coordinate but it does not allow
for tunneling through the reaction barrier. More sophisticated approximations remove the
separability assumption and include the effect of tunneling, typically in such a way as to
give the exact result for a parabolic barrier. We do not have the time to discuss these
approximations here but some references to them are given in the further reading.1,5,6 The
starting point for the development of new approximations is the exact flux-side correlation
function expression for the quantum rate coefficient in Eq. (113), which underlies most of
the modern research into the theory and calculation of chemical reaction rates.
29
V. FURTHER READING
The material described in these lecture notes is based on two classic papers by Bill Miller,
both of which indicate how the theory can be generalized to treat more complicated reactions
than the simple one-dimensional model problem we have considered here:
The two books I found most useful while preparing the lectures were:
6. W. H. Miller, Y. Zhao, M. Ceotto and S. Yang, J. Chem. Phys. 119, 1329 (2003).
And for what it is worth, my own preferred way to calculate chemical reaction rates (includ-
ing quantum mechanical zero point energy and tunneling effects and allowing for recrossing
of the transition state dividing surface) is described in the following two papers, both of
which should be quite accessible to anyone who has understood these lecture notes:
fb(p’-p)
b/π
p’-p
π/b
VI. APPENDICES
The Dirac delta function δ(p0 − p) in Eq. (10) is not a proper mathematical function,3
but it can at least be written as a limit of something that is:
1 Z∞ 0
δ(p0 − p) = dx e−i(p −p)x/h̄
2πh̄ −∞
1 Za 0
= lim dx e−i(p −p)x/h̄
a→∞ 2πh̄ −a
0 #a
e−i(p −p)x/h̄
"
1
= lim
a→∞ 2πh̄ −i(p0 − p)/h̄
−a
1 sin(p0 − p)a/h̄
= lim
a→∞ π (p0 − p)
= lim fb (p0 − p), (A1)
b→∞
where
1 sin(p0 − p)b
fb (p0 − p) = . (A2)
π (p0 − p)
This function is plotted above. It consists of a sharp peak centred on p0 = p with a height
proportional to b and a width inversely proportional to b, along with a rapid oscillation with
a period of 2π/b that extends to larger values of |p0 − p|. In view of this, the delta function
δ(p0 − p) = limb→∞ fb (p0 − p) can be envisaged as an infinitely high and infinitely narrow
31
spike at p0 = p and nothing else, the period of the rapid oscillation having shrunk to zero in
the limit as b → ∞.
The area under fb (p0 − p) remains well defined in this limit and can be calculated as
follows:
Z ∞
0 1 Z ∞ 0 sin(p0 − p)b
0
dp δ(p − p) = lim dp
−∞ b→∞ π −∞ p0 − p
Z ∞
1 sin x
= dx
π −∞ x
= 1, (A3)
where we have made the substitution x = (p0 − p)b in the second line and evaluated a
standard contour integral to get the third (exercise). In fact, the same argument shows
more generally that, for any continuous function f (p),
Z ∞
1Z∞ 0 sin(p0 − p)b
dp0 f (p0 )δ(p0 − p) = lim dp f (p0 )
−∞ b→∞ π −∞ p0 − p
Z ∞
1 sin x
= lim dx f (p + x/b)
b→∞ π −∞ x
1Z∞ sin x
= dx f (p) = f (p). (A4)
π −∞ x
This is one of the most important properties of the delta function and it is used in various
places in the text [for example in Eq. (26)].
Another generalization of the argument in Eq. (A3) shows that, for any > 0,
Z p−
1 Z −b sin x
dp0 δ(p0 − p) = lim dx = 0, (A5)
−∞ b→∞ π −∞ x
Z p+
0 1 Z +b
0 sin x
dp δ(p − p) = lim dx = 1, (A6)
p− b→∞ π −b x
Z ∞
1Z∞ sin x
dp0 δ(p0 − p) = lim dx = 0, (A7)
p+ b→∞ π +b x
and thus that the integral of the delta function is a step function:
Z p00 1, if p00 > p
(
0 0 00
dp δ(p − p) = h(p − p) = . (A8)
−∞ 0, if p00 < p
We have also made extensive use of this result in these notes [for example in going from
Eq. (128) to (129)].
32
In the text we have asserted that F̂ = (i/h̄)[Ĥ, ĥ] is the Heisenberg time derivative of ĥ,
and that limt→∞ e−iĤt/h̄ h(p̂)e+iĤt/h̄ is the projection onto states with positive momentum
in the infinite past. These assertions are based on the Heisenberg picture of quantum
mechanics, in which state vectors are regarded as time-independent and the time-dependence
is transferred to the operators.
Consider a system with a time-independent Hamiltonian Ĥ and a state vector |ψ(t)i that
satisfies the time-dependent Schrödinger equation
d
ih̄ |ψ(t)i = Ĥ |ψ(t)i . (B1)
dt
Assuming for simplicity that the state |ψ(t)i is normalized, the expectation value of an
observable A corresponding to an operator without any explicit time dependence (such as
the projection operator  = h(p̂)) is given at time t by
This is in the Schrödinger picture where the state |ψ(t)i evolves in time and the operator Â
remains constant.
However, since the Schrödinger equation in Eq. (B1) can be integrated to give
and therefore
hψ(t)| = hψ(0)| e+iĤt/h̄ , (B3)
changes with time. This is the Heisenberg picture, in which Â(t) is the Heisenberg-evolved
operator at time t. So for example
can be interpreted as the projection operator onto states with positive momentum (h(p̂) = 1)
in the infinite past (t → −∞). And since Ĥ commutes with the evolution operators e±iĤt/h̄ ,
we can differentiate Eq. (B5) with respect to t to obtain
d i h i
Â(t) = e+iĤt/h̄ Ĥ Â − ÂĤ e−iĤt/h̄
dt h̄
i h i
= e+iĤt/h̄ Ĥ, Â e−iĤt/h̄
h̄
ih i
= Ĥ, e+iĤt/h̄ Âe−iĤt/h̄
h̄
ih i
= Ĥ, Â(t) , (B7)
h̄
thus confirming that F̂ (t) = (i/h̄)[Ĥ, ĥ(t)] is the Heisenberg time derivative of ĥ(t) (at any
time t including t = 0).
C. Liouville’s theorem
We have made use of Liouville’s theorem in Eq. (131). This theorem states that phase
space volume is conserved in classical mechanics,
or equivalently that the Jacobian of the dynamical transformation from (p0 , x0 ) to (pt , xt )
is unity:
∂pt /∂p0 ∂xt /∂p0
! ! ! !
∂(pt , xt ) ∂pt ∂xt ∂xt ∂pt
Jt = = = − = 1. (C2)
∂(p0 , x0 ) ∂pt /∂x0 ∂xt /∂x0 ∂p0 ∂x0 ∂p0 ∂x0
Since it is evidently true that J0 = 1, we can prove the theorem by showing that J˙t = 0 for
all t, and this follows directly from Hamilton’s equations of motion in Eqs. (125) and (126):
! ! ! ! ! ! ! !
∂ ṗt ∂xt ∂pt ∂ ẋt ∂ ẋt ∂pt ∂xt ∂ ṗt
J˙t = + − −
∂p0 ∂x0 ∂p0 ∂x0 ∂p0 ∂x0 ∂p0 ∂x0
! ! ! ! ! ! ! !
∂ ṗt ∂xt ∂ ṗt ∂xt ∂ ẋt ∂pt ∂ ẋt ∂pt
= − − +
∂p0 ∂x0 ∂x0 ∂p0 ∂p0 ∂x0 ∂x0 ∂p0
!" ! ! ! !#
∂H ∂xt ∂xt ∂xt ∂xt
=− −
∂xt ∂p0 ∂x0 ∂x0 ∂p0
!" ! ! ! !#
∂H ∂pt ∂pt ∂pt ∂pt
− −
∂pt ∂p0 ∂x0 ∂x0 ∂p0
= 0. (C3)