PH4211 Statistical Mechanics

Problem Sheet 5 — Answers

bg b g
1 A random quantity has an exponential autocorrelation function G t = G 0 e −γt .
Calculate its correlation time using the usual definition.

The correlation time is the ‘width’ of the correlation function – the area divided by the
height. Thus the definition
∞
1
G ( t ) dt .
G ( 0 ) ∫0
τc =

So for the exponential correlation function

∞
τ c = ∫ e −γ t dt
0
∞
e −γ t 1
= = .
−γ 0
γ
The correlation time is simply 1/γ – the time constant of the exponential.

2 Show that the autocorrelation function of a periodically varying quantity

m ( t ) = m cos ωt is given by
m2
G (t ) = cos ωt .
2
Show that the autocorrelation function is independent of the phase of m(t). In other
words, show that if m ( t ) = m cos (ωt + ϕ ) , then G(t) is independent of ϕ.

The autocorrelation function is defined by

G ( t ) = m (τ ) m (τ + t ) .
It is convenient here to regard the average as a time average. That is, we average over
τ so that G(t) is calculated as
T
1
G (t ) = m (τ ) m (τ + t ) dτ .
2T −∫T
But actually we don’t need to do any integration, as we shall see. We star from
m ( t ) = m cos ωt , so that
G ( t ) = m (τ ) m (τ + t )
= m cos (ωτ ) m cos ω (τ + t ) .
We use the trig identity cos ( x + y ) = cos x cos y − sin x sin y , so that
G ( t ) = m 2 cos 2 ωτ cos ωt + m 2 cos ωτ sin ωτ sin ωt .
The average is evaluated over τ, thus
G ( t ) = m 2 cos ωt cos 2 ωτ + m 2 sin ωt cos ωτ sin ωτ .

Answers 5-1
The average of cos 2 is ½; the function varies smoothly between 0 and 1. The average
P P

of cos sin is zero; this function varies smoothly between –1 and +1. Thus we conclude
that
m2
G (t ) = cos ωt
2
as required.

Now consider the case m ( t ) = m cos (ωt + ϕ ) . Then

G ( t ) = m (τ ) m (τ + t )
= m cos (ωτ + ϕ ) m cos (ωτ + ϕ + ωt ) .
We use the same trig identity to obtain
G ( t ) = m 2 cos 2 (ωτ + ϕ ) cos ωt + m 2 cos (ωτ + ϕ ) sin (ωτ + ϕ ) sin ωt .
The average is evaluated over τ, thus
G ( t ) = m 2 cos ωt cos 2 (ωτ + ϕ ) + m 2 sin ωt cos (ωτ + ϕ ) sin (ωτ + ϕ ) .
Then, once again, the average of cos 2 is ½ and the average of cos sin is zero. And so
P P

in this case also we obtain

m2
G (t ) = cos ωt ;
2
this shows that G(t) is independent of φ as required.

3 The dynamical response function X(t) must vanish at zero times, as shown in
Fig. 5.13. What is the physical explanation of this? What is the consequence for the
step response function Φ(t)? Is this compatible with an exponentially decaying Φ(t)?

The step response function Φ(t) is proportional to the autocorrelation function of the
response variable
Φ (t ) ∝ M ( 0) M (t ) ,
from the Onsager hypothesis or the linear response derivation. Since, certainly in the
classical case, M(0) and M(t) commute, we may swap these around so that
Φ ( t ) = Φ ( −t ) .
Thus Φ(t) is an even function and its odd derivatives must vanish at t = 0. And then
since X(t) is the first derivative of Φ(t), it follows that X(0) = 0 as required.

If the odd derivatives of Φ(t) vanish at the origin, this is clearly incompatible with an
exponential decay. More precisely Φ(t) cannot decay exponentially in the vicinity of
t = 0; it can elsewhere.

4 In Section 5.3 we examined the form of the dynamical susceptibility χ(ω) that
followed from the assumption that the step response function Φ(t) decayed
exponentially. In this question consider a step response function that decays with a
gaussian profile, Φ ( t ) = χ 0 e−t 2τ . Evaluate the real and imaginary parts of the
2 2

dynamical susceptibility and plot them as a function of frequency. The real part of the
susceptibility is difficult to evaluate without a symbolic mathematics system such as

Answers 5-2
Mathematica. Compare and discuss the differences and similarities between this
susceptibility and that deduced from the exponential step response function (Debye
susceptibility).

The response function X(t) is given by minus the derivative of Φ(t), thus
d
X ( t ) = − χ 0 e −t 2τ
2 2

dt
t 2 2
= χ 0 2 e −t 2τ .
τ
The dynamical susceptibility is the Fourier transform of this
∞
χ (ω ) = ∫ X (t ) e
iωt
dt
−∞

so that
∞
χ
χ (ω ) = 20 ∫ te
− t 2 2τ 2 iωt
e dt .
τ −∞

Upon integration we obtain the real and imaginary parts as

⎧⎪ π ωτ ⎫⎪
χ ′ (ω ) = χ 0 ⎨1 −
2 2
ωτ e −ω τ 2 erfi ⎬
⎩⎪ 2 2 ⎭⎪
π
χ ′′ (ω ) = χ 0
2 2
ωτ e−ω τ 2 .
2
These are plotted in the figure.

χ /χ 0
1 χ'(ω)

0.8

0.6
χ''(ω)
0.4

0.2

1 2 3 4 ωτ
- 0.2

The imaginary part of the susceptibility is always positive, as required by energy

considerations. In this case the real part of the susceptibility goes negative. This is in
contrast to that of the Debye susceptibility, that remains positive at all frequencies.

Answers 5-3
5 The Debye form for the dynamical susceptibility is
1
χ ′ (ω ) = χ 0
1 + ω 2τ 2
ωτ
χ ′′ (ω ) = χ 0 .
1 + ω 2τ 2
Plot the real part against the imaginary part and show that the figure corresponds to a
semicircle. This is known as a Cole-Cole plot.

The Cole-Cole plot of the Debye susceptibility is shown below.

χ''(ω)

0.4

0.2

0.2 0.4 0.6 0.8 1 χ'(ω)

- 0.2

- 0.4

The upper half circle corresponds to positive frequencies and the lower half
corresponds to negative frequencies.

The real and imaginary parts pf the Debye susceptibility are seen to satisfy the
equation
2 2 2
⎛ χ ′′ ⎞ ⎛ χ ′ 1 ⎞ ⎛ 1 ⎞
⎜ ⎟ +⎜ − ⎟ = ⎜ ⎟ .
⎝ χ0 ⎠ ⎝ χ0 2 ⎠ ⎝ 2 ⎠
This is corresponds to a circle of radius ½ centred on χ ′ χ 0 = 1 2 .

6 Plot the Cole-Cole plot (Problem 5.5) for the dynamical susceptibility considered in
Problem 5.4. How does it differ from that of the Debye susceptibility.

In Problem 5.4 we obtained the expressions for the real and the imaginary parts of the
dynamical susceptibility
⎧⎪ π ωτ ⎫⎪
χ ′ (ω ) = χ 0 ⎨1 −
2 2
ωτ e −ω τ 2 erfi ⎬
⎪⎩ 2 2 ⎭⎪
π
χ ′′ (ω ) = χ 0
2 2
ωτ e−ω τ 2 .
2

Answers 5-4
From these we can make the Cole-Cole plot:
χ''(ω)

0.6

0.4

0.2

- 0.2 0.2 0.4 0.6 0.8 1 χ'(ω)

- 0.2

- 0.4

- 0.6

The main difference from the Debye form is that this curve passes to the left of the y
axis (high frequencies), where the real part of the susceptibility becomes negative.
Since in the high frequency limit the real and imaginary parts of the susceptibility
must both vanish, this gives the cardioid shape to the plot.

7 The full quantum-mechanical calculation of the Johnson noise of a resistor gives

hf
v 2 = 4 R hf kT ∆f .
∆f e −1
Show that this reduces to the classical Nyquist expression at low frequencies. At what
frequency will there start to be serious deviations from the Nyquist value? Estimate
the value of this frequency.

At low frequencies such that hf << kT we can expand the exponential so that
hf
v2 = 4R ∆f
∆f 1 + hf kT + … − 1
hf
= 4R ∆f
hf kT + …
and in the low frequency limit the hf cancels, giving
v2 = 4kTR∆f ,
∆f

the classical Nyquist expression.

For the Nyquist expression to be valid we require the frequency to satisfy f << kT/h,
so at room temperature (T ~ 300K) this means

Answers 5-5
 1.4 ×10−23 × 300
f << −34
= 6.4 ×1012 Hz .
6.6 × 10

8 Show that for the Debye susceptibility, the relation

1 χ ′′ (ω )
∞
χ0 = ∫ dω
π −∞
ω
holds. Demonstrate that χ ′′ vanishes sufficiently fast at ω = 0 so there is no pole in the
integral and there is thus no need to take the principal part of the integral in the
Kramers-Kronig relations.

In the Debye case

ωτ
χ ′′ (ω ) = χ 0
1 + ω 2τ 2
so that the integrand is
χ ′′ (ω ) τ
= χ0 .
ω 1 + ω 2τ 2
The ω in the denominator has been cancelled by the ω in the numerator. Thus there is
no pole in the integral.

We wish to evaluate the integral

1
∞
χ ′′ (ω ) 1
∞
τ
I= ∫
π −∞ ω
dω = χ 0 ∫
π −∞ 1 + ω 2τ 2
dω .

We can remove the τ by changing variables to x = ωτ, so that dω = dx/τ, and

∞
1 dx
I = χ0
π ∫ 1+ x
−∞
2
.

The integral is standard; its value is π. Thus we have shown that I = χ 0 ; we have
B B

demonstrated that the Kramers-Krönig sum rule holds in the Debye case.

5.9 In Section 5.3.6 we considered an electrical analogue of the Langevin Equation

based on a circuit comprising an inductor and a resistor. In this problem we shall
examine a different analogue: a circuit of a capacitor and a resistor. Show that the
equation analogous to the Langevin equation, in this case, is
dV ( t ) 1
C + V (t ) = I (t ) .
dt R
Hence show that the fluctuation-dissipation result relates the resistance to the current
fluctuations through
∞
1 1
=
R 2kT −∞ ∫ I ( 0 ) I ( t ) dt .

Answers 5-6