Correlation and Coherence Functions

Jhih-Sheng Wu
2022

1 Correlation Functions

A correlation function is a mathematical tool used to measure the dependence between two or
more variables. In physics, these variables are physical quantities depending on positions and
time. For example, let s1 (r, t) and and s2 (r, t) be the two amplitudes of two scalar waves. The
first order correlation function is
C (1) (∆r, ∆t) = hs1∗ (r1 , t1 )s2 (r2 , t2 )i, (1)
where h...i denotes an ensemble average.

The correlation functions are used to describe spatial and temporal coherences of waves. The
superposition of two waves is
|s1 (r1 , t1 ) + s2 (r2 , t2 )|2 = |s1 (r1 , t1 )|2 + |s2 (r2 , t2 )| + 2Re[s1 (r1 , t1 )s2∗ (r2 , t2 )]. (2)
The ensemble-averaged interference is
h2Re[s1 (r1 , t1 )s2∗ (r2 , t2 )]i = 2Re[C(∆r, ∆t)]. (3)
Correlation functions are called auto-correlation functions if s1 and s2 are the same variables.
If s1 is the same source as s2 and ∆r = 0, the correlation functions measure the temporal
coherence. If s1 is the same source as s2 and ∆t = 0, the correlation functions measure the
spatial coherence. We can define the dimensionless correlations function g (1) , 1 called the
first-order correlation function or normalized correlation function,

(1) hs1∗ (r1 , t1 )s2 (r2 , t2 )i

g (∆r, ∆t) = p . (4)
h|s1 (r1 , t1 )|2 ih|s2 (r2 , t2 )|2 i

According to the Schwartz inequality, 0 ≤ g (1) ≤ 1. The coherences are related to g (1) by

|g (1) | = 1, coherent, (5)

(1)
0 < |g | < 1, partially coherent, (6)
(1)
|g | = 0, completely incoherent. (7)

The coherence function g (1) typically decreases at time goes or traveled optical length increases.
The process is called decoherence. Two main sources of decoherence are (a) non-monochromatic
light and (b) noises due to collisions or scatterings. We might model the decoherence as
1 In the literature, people use γ (1) for classical cases and g (1) for quantum cases. Here, I use g (1) for both the
cases.

1
 !
(1) (1) t
g (t) = g (0) exp − , (8)
τc

where τc is the coherence time. If a light source is not monochromatic and has a band width
∆ω, the coherent time is τc ∼ ∆ω
1
.

In theories of probability and statistics, we can specify a probability distribution of a variable X

if we know all the moments of X, i.e., hXi , hX 2 i, hX 3 i,... If we want to fully specify the relation
between X and Y , we need to know not only hXY i but also the higher order terms such as
hX 2 Y 2 i, hX 3 Y 3 i. One can define the high-order autocorrelation functions are defined as

C (2) (x1 , x2 , x3 , x4 ) = hs∗ (x1 )s∗ (x2 )s(x3 )s(x4 )i (9)

ans so on. Here, xn denotes (rn , tn ). One useful case is

C (2) (x1 , x2 , x2 , x1 ) = hI(x1 )I(x2 )i, (10)

which is called the intensity-intensity correlation function. The order of xn in the correlation
function does not matter for classical fields, but as we will see soon, the order for quantum
fields does matter. The second-order coherence function g (2) is defined as

C (2) (x1 , x2 , x2 , x1 )
g (2) (x2 − x1 ) ≡ g (2) (x1 , x2 , x2 , x1 ) = . (11)
C (1) (x1 , x1 )C (1) (x2 , x2 )

In quantum optics, waves are electromagnetic fields. The scalar field s(x) is replaced by E ≡ E · ê,
where ê is a unit vector. For example, the first-order coherence function becomes

hE1∗ (r1 , t1 )E2 (r2 , t2 )i

g (1) (∆r, ∆t) = p . (12)
h|E1 (r1 , t1 )|2 ih|E2 (r2 , t2 )|2 i

1.1 Definitions in Quantum Optics

When defining correlation functions for quantum optics, we have the following term

hE1∗ (r1 , t1 )E2 (r2 , t2 )i. (13)

Classically, the order of the product in the average does not matter. But quantumly, we have
to deal with the order carefully. Physically, correlations are measured quantities. Measurement
processes consist of absorptions of photons by the detectors. Say, first the detector absorbed
one photon at t1 and another photon at a latter time t2 . This process is described by two
annihilation operators

a(t2 )a(t1 )|ii, (14)

2
where |ii is the initial state. The probability of the process is proportional to the norm of
Eq. (14).

hi|a† (t1 )a† (t2 )a(t2 )a(t1 )|ii. (15)

Thus, we have the following summary

• All annihilation operators are on the right.

• All creation operators are on the left.

• An annihilation operator at an earlier time is on the right.

• An creation operator at an earlier time is on the left.

The correlation functions are given by the density-matrix approach

C (1) = Tr[ρE ∗ (t2 )a† (t2 )E(t1 )a(t1 )] (16)

∗ †
= E (t2 )E(t1 )Tr[ρa (t2 )a(t1 )], (17)
(2)
C ∗ † ∗ †
= Tr[ρE (t1 )a (t1 )E (t2 )a (t2 )E(t2 )a(t2 )E(t1 )a(t1 )] (18)
∗ ∗ † †
= E (t1 )E (t2 )E(t2 )E(t1 )Tr[ρa (t1 )a (t2 )a(t2 )a(t1 )] (19)
˜ 1 )I(t
= I(t ˜ 2 )Tr[ρa† (t1 )a† (t2 )a(t2 )a(t1 )]. (20)

We should use these correlation function to calculate the quantum coherence functions defined
in Eqs. (11) and (12).

3
2 Young’s Interference and First-Order Coherent Function

Figure 1: Young’s interference.

Let the source generated single-mode photons whose annihilation operator is a. The photons
then pass the two slits. The two slits are regarded as the light sources whose annihilation
operators are a1 and a2 . We assume the two slits are equal such that
1
a = √ (a1 + a2 ) . (21)
2
The intensity on the screen is indeed the first-order correlation function
˜
I(t) = I(t)Tr[ρa †
(t)a(t)] (22)
˜ n
I(t) o
= Tr[ρa†1 (t)a1 (t)] + Tr[ρa†2 (t)a2 (t)] + Tr[ρa†1 (t)a2 (t)] + Tr[ρa†2 (t)a1 (t)] (23)
2
˜ n
I(t) o
= Tr[ρa†1 (0)a1 (0)] + Tr[ρa†2 (0)a2 (0)] + Tr[ρa†1 (0)a2 (0)]eiΦ + Tr[ρa†2 (0)a1 (0)]e−iΦ ,
2
(24)
where Φ is the phase difference due to the optical length. When the incident light is a one-
photon state, we have the state after two slits
1   1
a† |0i = √ a†1 + a†2 |0i = √ (|10i + |01i). (25)
2 2
The first-order correlation function is
!
˜ 1 + cos Φ
I(t) = I(t) . (26)
2

4
When the incident light is a two-photon state, we have the state after two slits
1 † 2 1 √
(a† )2 |0i = a1 + a†2 |0i = (|20i + 2|11i + |02i). (27)
2 2
The first-order correlation function is
!
˜ 1 + cos Φ
I(t) = 2I(t) . (28)
2
For a n-photon state, we have
!
˜ 1 + cos Φ
I(t) = nI(t) . (29)
2
It can be shown that for a coherent state, we have
!
˜ 1 + cos Φ
I(t) = n̄I(t) . (30)
2
We have the following notes

• Interference occurs even when for a single-photon state.

• First-order correlation functions describe interferences.
• First-order correlation functions can not distinguish what kind of photon state the light
is. For number states and coherent states, the interferences are the same as cos Φ.

3 Hanbury-Brown and Twiss Experiment and Second-Order

Coherent Function

Figure 2: Hanbury- Brown and Twiss experiment.

5
First-order coherence in Young’s experiment can determine how monochromatic a light is, or
noise levels by measuring coherent lengths. The properties revealed by first-order coherence are
related to the modes and the environments (noise). However, first-order coherence does not tell
which photon state it is, that is, if two states are from the same mode, first-order coherence can
not distinguish their photon distribution. Say, to distinguish a number state |ni and a coherent
state |αi from the same mode, we have to use the second-order coherence function g (2) .

In the 1950s, Hanbury Brown and Twiss developed an experiment to measure intensity-intensity
correlations (see Fig. 2). The main difference from a Young’s interference is that there are two
intensity detectors instead of one. The two optical paths after the beam splitter lead to a time
delay τ. The rates of coincident events is measured by the coincidence counter and proportional
to

C (2) = hE ∗ (t1 )a† (t1 )E ∗ (t2 )a† (t2 ).E(t2 )a(t2 )E(t1 )a(t1 )i (31)

3.1 Classical Regime

Classically, we can write the second-order correlation function as

C (2) = hI(t + τ)I(t)i, (32)

and the second-order coherent function g (2) as

hI(t + τ)I(t)i
g (2) = . (33)
hI(t + τ)ihI(t)i

For a stationary light, hI(t)i is time-independent. The second-order coherent function g (2)
becomes
hI(t + τ)I(t)i
g (2) = . (34)
hI(t)i2
The following properties can be shown (only in the classical regime),

1 ≤ g (2) (0) < ∞, (35)

(2) (2)
g (τ) ≤ g (0). (36)

For a chaotic light source, it can be shown that

g (2) (τ) = 1 + |g (1) (τ)|2 , (37)

− ττ
and if g (1) (τ) = e c ,
− 2τ
g (2) (τ) = 1 + e τc . (38)

For chaotic lights, it is shown g (2) (0) = 2. This is called the photon bunching-effects.

6
3.2 Quantum Regime

The orders of operators in the correlation functions should be treated carefully. For a single
mode, the second-order coherent function is

ha† (t)a† (t + τ)a(t + τ)a(t)i

g (2) (τ) = (39)
ha† (t + τ)a(t + τ)iha† (t)a(t)i
ha† a† aai
= (40)
ha† ai2
σ 2 (n) − n̄
= 1+ (41)
n̄2
We can show that

2, chaotic,

g (2) (0) =  (42)

1, coherent,


and for a number state


0, n = 0, 1,

g (2) (0) =  (43)

n−1
 n , else.


As time increases, light becomes incoherent. We may use the result for chaotic light,

g (2) (τ) = 1 + |g (1) (τ)|2 , (44)

and g (1) (τ) → 0 as τ → ∞. Thus as τ → ∞, g (2) (τ) = 1.

Figure 3: g (2) (τ) of various photon ensembles.

7
 Figure 4: Photon counting statistics of various photon ensembles.

Figure 5: Experiment of single photon measurement.

Exercise 1: Coherence Functions

Calculate g (2) (0) for the following cases:
(a) a state
1
|ψi = √ (|0i + |1i) . (45)
2
(b) an ensemble
1 1
ρ = |0ih0| + |1ih1|. (46)
2 2
(c) an ensemble
1 1 i −i
ρ = |0ih0| + |1ih1| + |0ih1| + |1ih0| (47)
2 2 2 2

8
Hint: use
ha† a† aai
g (2) (0) = (48)
ha† ai2

Summary 1: Correlations of Light

1. Coherent light has g (1) (t) = 1 and g (2) (t) = 1. The photon number distribution is a
Poisson distribution.

2. A chaotic light has a g (1) = 0. There are many kinds of chaotic lights, for example,
thermal lights, unpolarized lights, and so on. A chaotic light has g (2) (0) = 2, which
is the photon bunching. One can distinguish different kinds of chaotic lights by
n~ω
photon counting statistics. The pn of a thermal light is proportional to e− kT . In
general, pn of a chaotic light can be arbitrary. It is possible that pn of a chaotic light
is a Poisson distribution. Hence, a Poisson distribution of pn does not conclude that
the light is coherent.

3. A single photon state has g (1) (t) = 1 and g (2) (0) = 0, which is a result of the photon
anti-bunching.