Creation/annihilation Operators

There is a correspondence1 between classical canonical formalism and quantum mechanics. For
the simplest case of just one pair of canonical variables,2 (q, p), the correspondence goes as follows.
Consider classical Hamiltonian H(q, p), introduce a pair of Hermitian operators, q̂ and p̂, quantum
counterparts of q and p, such that their commutator equals i~:

[q̂, p̂] = i~. (1)

The quantum-mechanical counterpart of the classical Hamiltonian H(q, p) is the operator3

Ĥ = H(q̂, p̂).

Let us use this correspondence as an inspiration for developing a very important formalism. In
classical mechanics, very useful are complex canonical variables. The quantum counterpart of a
classical complex canonical variable a associated with the pair (q, p) is the operator

â = αq̂ + iβ p̂, (2)

where α and β are certain complex numbers. In classical mechanics, the two real variables, q and
p, are expressed as linear combinations of linear independent variables a and a∗ . The operator
analog of a∗ is the operator ↠≡ â+ :

↠= α∗ q̂ − iβ ∗ p̂. (3)

The expressions for q̂ and p̂ in terms of â and ↠readily follow from (2)–(3):

q̂ = (β↠+ β ∗ â)/(αβ ∗ + α∗ β), (4)

p̂ = i(α↠− α∗ â)/(αβ ∗ + α∗ β), (5)

revealing, in particular, the necessity of the condition

αβ ∗ + α∗ β 6= 0.

From (2)–(3), we also find

[â, ↠] = ~(αβ ∗ + α∗ β). (6)
Let us optimize the choice of α and β by requiring that

αβ ∗ + α∗ β = 1/~. (7)

This leads to a simple and universal form of the commutator (6):

[â, ↠] = 1, (8)

1
Often referred to as canonical quantization.
2
The generalization to an arbitrary number of canonical pairs is straightforward.
3
Non-commutativity of q̂ and p̂ involves a certain ambiguity: The form of Ĥ depends on the order of operators.
The standard way of dealing with this ambiguity is to impose the so-called normal-order condition requiring that
the expression for Ĥ in terms of (below-introduced) creation and annihilation operators—replacing corresponding
classical complex canonical variables—be such that all the creation operators are to the left of all the annihilation
operators.

1
and also implies that
q̂ = ~(β↠+ β ∗ â), (9)
† ∗
p̂ = i~(αâ − α â). (10)
With such a choice of parameters α and β, the operators ↠and â are called, respectively, creation
and annihilation operators. A crucial part in our theory is played by the so-called number operator4

n̂ = ↠â. (11)

The number operator is Hermitian. And its name reflects the fact that its spectrum consists of
nonnegative integers (corresponding eigenvector is denoted as |ni):

n̂|ni = n|ni (n = 0, 1, 2, . . .). (12)

To establish this result along with other important relations, we observe that (8) implies

[n̂, ↠] = ↠, [n̂, â] = −â. (13)

Now let |λi be an eigenvector of the number operator corresponding to the eigenvalue λ:

n̂|λi = λ|λi. (14)

Combining (14) with (13), we get

n̂ ↠|λi = (λ + 1) ↠|λi, (15)

n̂ â|λi = (λ − 1) â|λi. (16)

Hence, by acting with the creation operator on the vector |λi we get (up to a certain factor) the
vector |λ + 1i, or a zero vector. Similarly, by acting with the annihilation operator on the vector
|λi we either get the vector |λ − 1i (up to a certain factor), or a zero vector.
Our next observation is that λ is nonnegative:

λ = hλ|n̂|λi = hλ|↠â|λi = hâλ|âλi ≥ 0, |âλi ≡ â|λi. (17)

Furthermore, combining (17) with (16) we see that (up to a phase factor that can be absorbed
into the eigenvector |λi) 
0, if λ = 0,
â|λi = √ (18)
λ |λ − 1i, if λ =6 0.
It is worth emphasizing that Eq. (18) does not explicitly tell us that 0 is a legitimate eigenvalue.
It only states that if that is the case, then acting with operator â on corresponding state |0i
will necessarily result in a zero vector, while acting on any state |λi with nonzero λ necessarily
produces a nonzero eigenvector proportional to the eigenvector |λ − 1i. Nevertheless, that is all
we need to readily conclude that λ has to be not only nonnegative, but also integer. Otherwise,
applying relation (18) iteratively, we will arrive at a certain value 0 < λ < 1, for which (18) would
imply the existence of the state |λ − 1i with (λ − 1) < 0, which would contradict inequality (17).
4
By construction, the number operator is a quadratic polynomial of operators q̂ and p̂. And that is the reason
why our theory proves very powerful for all sort of problems with (generalized) quantum harmonic oscillators.

2
 In an analogy with (17), we also have

λ + 1 = hλ|n̂ + 1|λi = hλ| â ↠|λi = h↠λ|↠λi, |↠λi ≡ ↠|λi. (19)

Combining (19) with (15), we see that (absorbing the phase factor into |λ + 1i)
√
↠|λi = λ + 1 |λ + 1i. (20)

Hence, when acting on an eigenvector |λi of the number operator, the creation operator can never
produce a zero vector; it always creates the eigenvector |λ + 1i.
Summarizing, we have proven the relation (12) and also established that
√
â|0i = 0, â |ni = n |n−1i (n = 1, 2, 3, . . .), (21)
√
↠|ni = n + 1 |n + 1i (n = 0, 1, 2, . . .). (22)
With Equation (22) we can directly relate the state |ni to the state |0i:

(↠)n
|ni = √ |0i. (23)
n!
A remark is in order here. Despite the inspiration coming from the correspondence between
classical and quantum mechanics, the theory we developed so far is a set of absolutely self-contained
purely mathematical relations between certain linear operators. Therefore, the presence of ~ in
the r.h.s. of (1) and subsequent formulas might seem somewhat confusing. Note, however, that
within this section, ~ was playing a role of a free positive constant. In particular, we could set it
equal to unity, or, equivalently, absorb it into q̂ or p̂.

Application to generalized harmonic oscillator

Consider the Hamiltonian of a generalized harmonic oscillator

q̂ 2 p̂2
H = A +B (A > 0, B > 0), (24)
2 2
where q̂ or p̂ are certain Hermitian operators satisfying the commutation relation (1); such oper-
ators are called canonically conjugate operators. Introducing creation-annihilation operators and
expressing H in terms of them, we get four different terms: (i) a constant, (ii) a term proportional
to n̂, (iii) a term proportional to â2 , and (iv) a term proportional to (↠)2 (the last two terms being
related to each other by Hermitian conjugation). Let us use our freedom of choosing α and β—
while still obeying (7)—to get rid of the term ∼ â2 (by Hermitian conjugation, this automatically
guarantees the absence of the term ∼ (↠)2 . As is readily seen by direct substitution of (9)–(10)
into (24), our α and β has to satisfy the requirement
p
α/β = A/B. (25)

Combining (25) with (7), we find

eiϕ eiϕ
α = √ (A/B)1/4 , β = √ (B/A)1/4 , (26)
2~ 2~

3
where ϕ is an arbitrary fixed phase. The freedom of choosing the phase ϕ corresponds to the U(1)
canonical transformation5 of the creation/annihilation operators,

â → eiϕ â, ↠→ e−iϕ ↠. (27)

Our Hamiltonian becomes

Ĥ = ~ω (↠â + 1/2) = ~ω(n̂ + 1/2), (28)

where √
ω = AB. (29)
The Hamiltonian does not depend on ϕ. This is an example of U(1) symmetry. In the context of
our problem, U(1) symmetry is the symmetry between dynamic roles played by the operators q̂ and
p̂, and their linear combinations. To appreciate this symmetry, see what happens to Eqs. (9)–(10)
upon the transformation (27) with ϕ = π/2.
With the Hamiltonian written in the form (28), the problem of the eigenstates and eigenenergies
is immediately solved by the theory developed in the previous section:

Ĥ |ni = En |ni, En = ~ω(n + 1/2) (n = 0, 1, 2, . . .). (30)

Thermal averages

The theory of creation/annihilation operators yields a powerful tool for calculating thermodynamic
averages of q̂- and p̂-dependent observables, like, q̂ 2 , p̂2 , q̂ 4 , p̂4 , etc. (Note that from the properties
of creation and annihilation operators it is easily seen that the thermal averages of odd powers of
q̂ and p̂ are zero.) Indeed, using the explicit expression for the statistical density matrix
P∞ −~ωn/T hn| ∞
e−Ĥ/T n=0 |ni e −~ω/T
X
ρ̂ = = P∞ −~ωn/T
= (1 − e ) |ni e−~ωn/T hn|, (31)
Tr e−Ĥ/T n=0 e n=0

relations (9) and (10), and the properties of creation/annihilation operators established in the
previous section, one obtains the answer for the average of an even power of q̂ or p̂ as a linear
combination of the following standard sums (λ = ~ω/T ):
∞
X  −1
e−λn = 1 − e−λ ,
n=0

∞ ∞ −1
X
−λn d X −λn d 
ne = − e = − 1 − e−λ ,
dλ dλ
n=0 n=0
∞ ∞
X
2 −λn d2 X −λn d2  −λ
−1
n e = e = 1 − e ,
dλ2 dλ2
n=0 n=0
5
A canonical transformation is the transformation that preserves the commutation relation (8).

4

−1
and so on, the function 1 − e−λ playing the role of generating function for the averages hn̂s i:
  ds  −1
s s −λ −λ
hn̂ i = (−1) 1−e 1−e (s = 1, 2, 3 . . .) .
dλs
Indeed, any polynomial function of q̂ and/or p̂ reduces to a polynomial function of creation and
annihilation operators. Such a polynomial consists of terms of the following two categories: (i) the
terms with equal powers of creation and annihilation operators and (ii) the terms with different
powers of creation and annihilation operators. Thermal averages of the terms of the category (ii)
are zero [see (31) and the properties of creation-annihilation operators]. The terms of the category
(i) can be represented as polynomial functions of the number operator, which reduces the averag-
ing to the above-mentioned formulas.

Problem 31
With the statistical density matrix (31), find the following thermal averages

E = hĤi = Tr (Ĥ ρ̂), hq̂ 2 i = Tr (q̂ 2 ρ̂), hp̂2 i = Tr (p̂2 ρ̂).

Observe that all the three averages are similar to each other: the answer naturally decomposes into
a sum of the ground-state and thermal parts, the thermal part being proportional to a thermal
average of the number operator.

Heisenberg picture

Yet another convenient tool comes with the relations

â e−iωn̂t = e−iωt e−iωn̂t â, ↠e−iωn̂t = eiωt e−iωn̂t aˆ† . (32)

Problem 32
Derive (32). Hint: Expand the exponential and use (13). You may find it useful to observe that
there is a general—and interesting on its own—relation

â f (n̂) = f (n̂ + 1) â,

where f is any analytic function. This relation can be also written in the following three forms:

f (n̂) â = â f (n̂ − 1),

f (n̂) ↠= ↠f (n̂ + 1),

↠f (n̂) = f (n̂ − 1) ↠.

Equations (32) imply simple expressions for the creation/annihilation operators in the Heisenberg
picture (note the analogy with complex canonical variables of a classical harmonic oscillator):

â(t) = e−iωt â(0), ↠(t) = eiωt aˆ† (0) (Heisenberg picture). (33)

5
With (9) and (10), this translates into simple expressions for Heisenberg-picture representations
for the operators q̂(t) and p̂(t), and their integer powers.

Working in q-representation: back to linear harmonic oscillator

In quantum mechanics, we are free to choose the representation for the vector of state. Let us
choose the q-representation,6 where the action of the operator q̂ on the vector of state |ψi—the
wave function ψ(q)—reduces to multiplying it by the factor q, and the operator p̂ is a differential
operator (we set ~ = 1):

|ψi ≡ ψ(q), q̂ |ψi ≡ q ψ(q), p̂ |ψi ≡ −iψ 0 (q).

It is convenient to rescale the operators q̂ and p̂ to render them dimensionless. Below ξˆ denotes an
appropriately nondimensionalized q̂; in such a way that a generic harmonic oscillator is described
by the Hamiltonian
∂
Ĥ = ξ 2 + p̂2 , p̂ = −i .
∂ξ
Comparing this to (24), we have
√
A=B=2 ⇒ α = β = 1/ 2,

and for the creation/annihilation operators we have:

   
1 ∂ ˆ†
1 ∂
â = √ ξ+ , a = √ ξ− .
2 ∂ξ 2 ∂ξ

Equations (28) and (30) now become

Ĥ = 2↠â + 1 = 2n̂ + 1, (34)

Ĥ |ni = n |ni, n = 2n + 1 (n = 0, 1, 2, . . .). (35)

In coordinate representation,
|ni ≡ ψn (ξ),
taking us back to our original consideration of the linear harmonic oscillator, but now with a new
and very powerful toolkit. The basic tools are as follows.

Simple way to find the ground state:

 
d
â|0i = 0 ⇔ ξ+ ψ0 (ξ) = 0.
dξ
2 /2
ξψ0 (ξ) + ψ00 (ξ) = 0 ⇒ ψ0 (ξ) ∝ e−ξ .

6
Because of the symmetry between q̂ and p̂, working in the p-representation would be essentially the same!

6
Simple explicit relation for the properly normalized energy eigenstates:
n
(↠)n π −1/4

d 2 /2
|ni = √ |0i ⇔ ψn (ξ) = √ ξ− e−ξ . (36)
n! 2n n! dξ

Simple recurrence relation for the Hermite polynomials:

 
d
Hn+1 = 2ξ − Hn (ξ). (37)
dξ

Problem 33. Derive (37).

More sophisticated tools come with the notion of coherent states introduced in the next section.

Coherent states

Definition and basic properties

Consider the states7

∞
2 /2
X an
|ai = e−|a| √ |ni (a 6= 0), |a = 0i = lim |ai = |n = 0i. (38)
n=0 n! a→0

Here a is a complex number; the exponential pre-factor is a normalization constant guaranteeing

ha|ai = 1. As can be straightforwardly checked, the state |ai is an eigenvector of the annihilation
operator with the eigenvalue a:
â |ai = a |ai. (39)
It can be also readily established that any (normalized to unity) eigenvector of the operator â
necessarily has the form (38) times the phase factor. The states (38)–(39) are called coherent
states.8
From (38) we find the expression for the inner product of two coherent states:
2 +|b|2 −2a∗ b)/2
ha|bi = e−(|a| . (40)

Note also
2
|ha|bi|2 = e−|a−b| . (41)
We see that coherent states are not orthogonal. This is not surprising given that the operator â
is not Hermitian.
7
We can also write it as
∞
!
−|a|2 /2
X an
|ai = e |0i + √ |ni ,
n=1 n!
so that the a = 0 case becomes trivial.
8
Coherent states of a particular annihilation operator â, to be more accurate.

7
 An important property of coherent states is the completeness relation:
Z
1
da |aiha| = 1̂. (42)
π
Here the integration is over the complex plane of a:

da ≡ da1 da2 , a = a1 + ia2 .

Relation (42) is checked by substituting (38) for |ai and ha|, and then performing the integration
in polar coordinates:
da ≡ (1/2) dϕ d|a|2 , a = |a|eiϕ .
We first integrate over dϕ, which renders the operator expression diagonal—nullifies
P∞ all the terms
2
|n1 ihn2 | with n1 6= n2 . Subsequent integration over d|a| readily produces n=0 |nihn| thus
completing the proof.
The completeness relation allows one to expand a generic density matrix ρ with respect to
coherent states. Starting with the identity

ρ̂ ≡ 1̂ ρ̂ 1̂

and then using (42) for both 1̂’s, we get

Z
ρ̂ = da db C(a, b) |biha|, C(a, b) = π −2 hb|ρ̂|ai. (43)

The normalization condition Trρ̂ = 1 reads:

Z Z
2 +|b|2 −2a∗ b)/2
Trρ̂ = 1 ⇔ da db C(a, b)ha|bi = 1 ⇔ da db C(a, b)e−(|a| = 1. (44)

This representation is particularly convenient for calculation of the averages of polynomial

expressions (or infinite series) involving creation and annihilation operators. Let

 = A(↠, â)

be such an expression. Without loss of generality, we adopt a convention that the expression
A(↠, â) has the normal-order form, where all the creation operators are to the left of all the
annihilation operators. Then, taking into account that

â |ai = a |ai ⇔ ha| ↠= ha| a∗ ,

we have
2 +|b|2 −2a∗ b)/2
ha|Â|bi = ha|A(↠, â)|bi = A(a∗ , b)ha|bi = A(a∗ , b) e−(|a| .
Hence,
Z Z
2 +|b|2 −2a∗ b)/2
hÂi = TrÂρ̂ = da db C(a, b) ha|Â|bi = da db C(a, b) A(a∗ , b) e−(|a| . (45)

This relation has two major applications. In quantum field theory, it gives rise to functional-
integral representation of correlators of bosonic fields, both in and out of the equilibrium. In

8
its turn, the functional-integral representation allows one to introduce advanced field-theoretical
techniques, such as, e.g., Feynman diagrammatics. The second crucial application is the classical
limit, which we discuss below.

Coherent states and the classical limit

The classical limit corresponds to the regime9

|a|, |b|  1 ⇒ |a − b| ∼ 1  |a|, |b| (classical limit).

Here we have b ≈ a, and (45) acquires the form of classical averaging over the phase-space10 of
the observable A = A(a∗ , a) with a certain classical distribution function Ccl (a):
Z Z
∗ 2 2 ∗
hAi = A(a , a) Ccl (a) da, Ccl (a) = db C(a, b) e−(|a| +|b| −2a b)/2 .

Note that the normalization of Ccl (a) is guaranteed by (44).

Yet another perspective on the classical limit emerges when studying how operator ↠acts on
the coherent state |ai. Consider the vector

|xi = ↠|ai − a∗ |ai.

By a straightforward calculation, observe that, independently of a,

hx|xi = 1.

From the identity  

† ∗ 1
â |ai = a |ai + ∗ |xi
a
we then see that, at |a|  1,
↠|ai = a∗ [ |ai + O(1/|a|) ] ,
and conclude that the coherent state |ai is almost an eigenstate of the creation operator,11 with the
eigenvalue a∗ . Hence, the coherent states with |a|  1 play the role of the quasi-classical states, the
annihilation operator behaving as classical complex canonical variable a and the creation operator
behaving as a∗ .
So far, our discussion of coherent states was free of any direct connections to harmonic os-
cillators, though in both cases the creation/annihilation operators were playing the central part.
In the next section, we will reveal the relevance of the coherent states to the theory of harmonic
oscillator.

9
Note that condition |a − b| ∼ 1 is guaranteed by the inner-product exponential; see (41).
10
Which in our case is nothing but the complex plane a.
11
It is good to keep in mind that there are no exact eigenstates of the creation operator: Trying to construct one
in the number basis, we end up with the requirement that all the expansion coefficients be zero.

9
Coherent states of a harmonic oscillator

Consider the harmonic oscillator Hamiltonian in the n̂-representation:

H = ~ωn̂ + E0 .

[Here we deliberately write the constant term in a general form rather than ~ω/2.] Let the initial
state of the system be a certain coherent state:

|ψ(t = 0)i = |ai.

The time evolution of this state can be established from (38):

∞
−iHt/~ −iE0 t/~ −|a|2 /2
X an e−inωt
|ψ(t)i = e |ai = e e √ |ni. (46)
n=0 n!

Comparison with (38) then brings us to quite an amazing conclusion:

|ψ(t)i = e−iE0 t/~ |ae−iωt i. (47)

That is, the system stays in a coherent state, but the coherent state does not stay the same: It
evolves following the pseudo-classical law:

a(t) = a(0) e−iωt .

Despite the analogy with classical oscillator for any value of |a|, it is only the |a|  1 case that is
semi-classical. This is readily seen by calculating the survival probability12

p(t) = |hψ(0)|ψ(t)i|2 .

Combining (47) with (41), we get

ωt
2 sin2
p(t) = e−4|a| 2 .
We see that at |a| ∼ 1, the survival probability stays on the order of unity, and at |a|  1 it is
always close to 1:
ωt
p(t) ≈ 1 − 4|a|2 sin2 (|a|  1).
2
At |a|  1, the survival probability becomes exponentially small at the characteristic time t ∼ 1/ω,
which is consistent with the semi-classical behavior expected in this regime.

Coherent states in coordinate representation

Here we work in the representation (and units) where

∂
ξˆ = ξ, p̂ = −i ,
∂ξ
 
1 1 ∂
â = √ (ξˆ + ip̂) = √ ξ+ .
2 2 ∂ξ
12
The survival probability is the probability to find the system in the initial state at a given time of evolution.

10
Finding a coherent state |ai in the ξ-representation, |ai ≡ ψa (ξ) thus amounts to solving the
equation:
√
 
d
ξ+ ψa = 2 a ψa . (48)
dξ
It is very convenient13 to decompose a in the r.h.s of (48) into a real and imaginary parts; also
making an observation that the two are proportional to the expectation values of coordinate and
momentum:
1
a = √ (ξ¯ + ip̄), ξ¯ = ha|ξ|ai,
ˆ p̄ = ha|p̂|ai.
2
Equation (48) is readily solved by the substitution

ψa (ξ) = CeΛ(ξ) ,

where C is a normalization constant. The final result for the normalized solution reads:
1 2 /2+ip̄ξ+iθ
ψa (ξ) = π − 4 e−(ξ−ξ̄) , (49)

where θ is an arbitrary phase. Note that

1 2
|ψa (ξ)|2 = π − 2 e−(ξ−ξ̄) ,

which corresponds to the probability density of the ground-state wave function of the harmonic
oscillator
Ĥ = ξ 2 + p̂2 ,
¯
shifted by the distance ξ.
In view of the very important result (47), applying (49) to the harmonic oscillator amounts to
introducing time dependence:
√ √
ξ¯ = 2|a| cos(ωt), p̄ = 2|a| sin(ωt). (50)

The phase θ also becomes a function of time. Its form is readily found by direct substitution of
(49)–(50) into time-dependent Schrödinger equation for harmonic oscillator.

Functional-integral representation of evolution and statistical operators

Evolution operator

In a direct analogy with (43), the evolution operator can be written in a coherent-state basis as
(~ = 1) Z
−iĤt
e = da db C(a, b) |biha|, (51)

with
C(a, b) = π −2 hb|e−iĤt |ai. (52)
13
And especially for future application to harmonic oscillator.

11
Now represent evolution operator as product of N → ∞ terms:

e−iĤt = (1 − iεĤ)(1 − iεĤ)(1 − iεĤ) · · · (1 − iεĤ), ε = t/N,

and identically rewrite it as

e−iĤt = (1 − iεĤ)1̂N −1 (1 − iεĤ)1̂N −2 · · · 1̂2 (1 − iεĤ)1̂1 (1 − iεĤ), ε = t/N, (53)

where Z
1
1̂j = daj |aj ihaj |, j = 1, 2, 3, . . . , (N − 1) (54)
π
are identity operators [see Eq. (42)]. Then substitute it into (52).
Without loss of generality, we assume that the Hamiltonian is represented by the normal-order
function of creation and annihilation operators:

Ĥ = H(↠, â), (55)

in which case

haj+1 |[1 − iεĤ]|aj i = haj+1 |[1 − iεH(a∗j+1 , aj )]|aj i = haj+1 |aj i[1 − iεH(a∗j+1 , aj )].

Now we can exponentialize the factors:

∗
[1 − iεH(a∗j+1 , aj )] → e−iεH(aj+1 ,aj ) ,

and use (40) for the inner product:

∗ ∗ ∗
haj+1 |aj i = e[aj+1 (aj −aj+1 )+aj (aj+1 −aj )]/2 .

This brings us to the final result (in which we restore Planck’s constant)
Z N −1
−iĤt/~ i
S({a})
Y daj
hb|e |ai = lim e ~ , (56)
N →∞ π
j=1

where
N −1  
X i~ ∗ ∗ ∗ ∗
S({a}) = [a (aj+1 −aj )−aj (aj+1 −aj )] − εH(aj+1 , aj ) , (57)
2 j+1
j=0

with the convention that

a0 ≡ a, aN ≡ b.
Condition of extremal action:
∂S
= 0
∂a∗j
implies
∂H(a∗j , aj−1 )
i~ (aj − aj−1 ) = ε . (58)
∂a∗j

12
Continuous limit for the extremal solution14 ,

aj → a(tj ), (aj − aj−1 ) → εȧ(tj ), tj = j(t/N ),

corresponds to the classical Hamiltonian dynamics:

∂H(a∗ , a)
i~ ȧ = . (59)
∂a∗
One should not be confused about the presence of Plank’s constant in this purely classical limit.
Here ~ reflects nothing but the units in which the classical canonical variable a is measured. The
Plank’s constant can be readily absorbed into a by re-scaling transformation
√
a → a/ ~.

For the harmonic oscillator this transformation would yield (omitting irrelevant constant term)

∂H(a∗ , a)
iȧ = , H(a∗ , a) = ω|a|2 (classical harmonic oscillator).
∂a∗

Statistical operator and thermal averages

Consider thermal average

Tr Âe−β Ĥ Tr e−β Ĥ Â
hÂi = = .
Tr e−β Ĥ Tr e−β Ĥ
Use completeness relation (42) to represent hÂi as

da ha|e−β Ĥ Â|ai
R
hÂi = R . (60)
da ha|e−β Ĥ |ai

In a direct analogy with (53)–(54), factorize

e−β Ĥ Â = (1 − εĤ)1̂N (1 − εĤ)1̂N −1 · · · 1̂3 (1 − εĤ)1̂2 (1 − εĤ)1̂1 Â, ε = β/N, (61)

Z
1
1̂j = daj |aj ihaj |, j = 1, 2, 3, . . . , N. (62)
π
The next steps are essentially the same as in the case of evolution operator. Assuming the normal-
order form for both Ĥ and Â, we get
∗
haj+1 |[1 − εĤ]|aj i = haj+1 |[1 − εH(a∗j+1 , aj )]|aj i → haj+1 |aj ie−εH(aj+1 ,aj ) ,
2 +|a |2 −2a∗
haj+1 |aj i = e−(|aj+1 | j j+1 aj )/2 .
14
There is no continuous limit at the level of action, because typical difference between aj+1 and aj is controlled
only by the inner product haj+1 |aj i and thus |aj+1 −aj | ∼ 1. Physically, this fact comes from non-commutativity of
creation and annihilation operators, resulting in inevitable quantum fluctuations. By contrast, the extremal solution
approaches a smooth function in the limit of N → ∞.

13
This brings us to the representation (central for advanced field-theoretical methods):
Z N
PN
H(a∗j+1 ,aj ) −
PN 2 +|a |2 −2a∗ daj
A(a∗1 , a) e−ε j=0 (|aj+1 |
Y
−β Ĥ j+1 aj )/2
ha|e Â|ai = j=1 e j
, (63)
π
j=1

where we use the convention

a0 ≡ aN +1 ≡ a. (64)
As a particular case, this formula works for the denominator in (60) by replacement → 1. A(a∗1 , a)
Let us see how representation (63) turns into the Gibbs distribution of classical mechanics
under the condition of high enough temperature, when typical magnitude of a is much larger than
unity. Given that |aj+1 − aj | ∼ 1 (because of the inner-product exponentials, while |a|  1, we
conclude that up to order-unity corrections, we have

aj ≈ a.

This allows us to replace

H(a∗j+1 , aj ) → H(a∗, a), A(a∗1 , a) → A(a∗, a)

and arrive at the following result

∗
ha|e−β Ĥ Â|ai = A(a∗, a) e−βH(a , a) I0 ,
Z N
PN 2 +|a |2 −2a∗ daj
e− j=0 (|aj+1 |
Y
I0 = j j+1 aj )/2 .
π
j=1

It is easy to see that I0 = 1. Indeed,

I0 = ha|1̂N . . . 1̂2 1̂1 |ai = ha|ai = 1.

The bottom line is the Gibbs distribution of classical Hamiltonian mechanics:

∗
A(a∗, a) e−βH(a , a) da
R
hÂi = (classical limit).
e−βH(a∗, a) da
R

If necessary, we can re-write it in terms of canonical coordinate and momentum. The Jacobian
of the tranformation from a to (q, p) is constant and thus cancels between the numerator and
denominator.

Shifted/shaken harmonic oscillators

Consider the Hamiltonian (~ = 1)

Ĥ = ω↠â + b∗ â + b ↠, (65)

where b is a certain complex number. Performing a canonical transformation of shift by a complex

number c:
â = â1 + c, ↠= â†1 + c∗ . (66)

14
(It is trivial to check that [â1 , â†1 ] = 1, meaning that â†1 and â1 are creation/annihilation operators.)
The transformed Hamiltonian is

Ĥ = ωâ†1 â1 + (ωc∗ + b∗ )â1 + (ωc + b)â†1 + ω|c|2 + b∗ c + bc∗ . (67)

In particular, with
c = −b/ω (68)
we get
Ĥ = ωâ†1 â1 − |b|2 /ω, (69)
which is a standard harmonic oscillator.
This simple fact becomes quite important the moment we realize that Hamiltonian (65) de-
scribes a harmonic oscillator with the potential shifted by a certain distance in space; the descrip-
tion being in terms of creation/annihilation operators corresponding to the original (non-shifted)
oscillator. The canonical transformation (66) then establishes the relationship between the cre-
ation/annihilation operators of the two oscillators. The key observation now is that the two
oscillators have exactly the same coherent states, but with different (shifted) eigenvalues:

â|ai = a|ai ⇔ â1 |ai = (a − c)|ai. (70)

This reveals a very peculiar property of coherent states: They are robust with respect to shifts of
the potential. The shifts can be instant or gradual—the coherent state remains intact, only its time
evolution depends on the details of the shifting protocol. In particular, to create a coherent state
with an arbitrary real a, it is sufficient to prepare the ground state—which is the coherent state with
the eigenvalue 0—and then instantly shift the potential by an appropriate distance. Subsequent
time evolution will then change the phase of a; see Eq. (47). Alternatively, if the system is in a
certain coherent state, we can readily drive it into any other coherent state (including the ground
state of the system). Let us elaborate on the quantitative aspects of corresponding protocols.
For a static external potential U (x) we introduce its “shaken” generalization:

U (x) → U (x − x0 (t)). (71)

The shaken potential has exactly the same shape as the original one, up to time-dependent shift
in space controlled by the function x0 (t). Now if the potential U (x) is harmonic,15 then its shaken
generalization corresponds to the Hamiltonian (65) with a certain time-dependent parameter b,
and a certain time-dependent additive constant.16
If, at a certain time t, the state of the system is a coherent state of the operator â, then the
state remains coherent during all the evolution governed by the Hamiltonian (65) with b = b(t),
obeying the law
â|ψ(t)i = a(t)|ψ(t)i, iȧ(t) = ωa(t) + b(t). (72)
To derive (72), consider the evolution during an infinitesimal time step ∆t. From the Shrödinger
equation we have
|ψ(t + ∆t)i = (1 − i∆t Ĥ)|ψ(t)i + O[(∆t)2 ]. (73)
15
It is worth recalling (cf. Problem 30) that for a harmonic potential, shaking protocol (71) is equivalent to the
protocol of adding a linear-in-x time-dependent term: U (x) → U (x) − F (t) x. By analogy with classical mechanics,
we can interpret this term as a time-dependent force F (t).
16
As always, this constant is dynamically irrelevant.

15
Acting on both sides with â, and neglecting the O[(∆t)2 ] term, we arrive at (72).

Problem 34. For a harmonic potential:

(a) Establish the relationship between x0 (t) and b(t).
(b) Derive (72) from (73).

Remarkably, the form of the first equation in (72) is exactly the same as the equation of mo-
tion of the classical harmonic oscillator in a shaken potential,17 meaning that identically the same
experimental protocol can be used to control both systems, provided in the quantum case we start
with a coherent state.

For a generic—neither coherent, nor even pure—initial state of the system described by Hamil-
tonian (65) with time-dependent b, the result (72) is still very useful. Indeed, representing the
initial state as a double integral (43) and then applying the evolution operators, we see that
Eq. (72) solves the problem of time evolution of a generic initial state.

Coherent states and the uncertainty relation

The famous uncertainty relation for two Hermitian operators, A and B, reads
1
σA σB ≥ |h[A, B]i| , (74)
2
where σA and σB are uncertainties of the two operators defined in terms of their dispersions (see
the Language and Axioms chapter):
p
σA = (∆A)2 , (∆A)2 = h (A− Ā)2 i, Ā = h A i, (75)

and the same for σB . The averaging in (74) and (75) is over a certain pure or mixed state. While
you probably saw it in your undergraduate course of Quantum Mechanics, it will not hurt to
review a short but nontrivial derivation of relation (74).
For any operator Ô, there takes place the inequality [cf. Eq. (17)]

h Ô+ Ô i ≥ 0, (76)

which is obvious in the basis of eigenstates of the density matrix:

X X
h Ô+ Ô i = Tr ρ̂ Ô+ Ô = Tr Ôρ̂ Ô+ = Tr wj Ô|ψj ihψj |Ô+ = wj hÔψj |Ôψj i ≥ 0 .
j j

With
Ô = (A− Ā) + iλ(B − B̄), (77)
where λ is an arbitrary real variable, the inequality (76) becomes

λ2 σB + iλh[A, B]i + σA ≥ 0. (78)

17
We can also think of a classical harmonic oscillator with a time-dependent driving force: the equation for a(t)
is equivalent to Newton’s Second Law for such a system.

16
As a function of λ, the l.h.s. of (78) is a real18 second-order polynomial. The necessary condition
for such a polynomial to be nonnegative—to have no more than one real roots—immediately brings
us to (74).
For the pair of operators q̂ and p̂ of Eq. (1), relation (74) simplifies to

~
σq σp ≥ , (79)
2
thus setting the lower bound of ~/2 on the product σq σp , and naturally bringing the problem of
finding the minimum-uncertainty states, such that
~
σq σp = (in a minimum-uncertainty state). (80)
2

Problem 35. Check that any coherent state of a harmonic oscillator is a minimum-uncertainty
state. To appreciate the contrast, also calculate the product σq σp for the number state |ni of the
harmonic oscillator.

18
The value of h[A, B]i is imaginary because the operator [A, B] is anti-Hermitian.

17