Introduction to the Rotating Wave Approximation

(RWA) : Two Coherent Oscillations

Kazuyuki FUJII
arXiv:1301.3585v3 [quant-ph] 23 May 2014

International College of Arts and Sciences

Yokohama City University

Yokohama, 2360027
Japan

Abstract

In this note I introduce a mysterious approximation called the rotating wave approx-
imation (RWA) to undergraduates or nonexperts who are interested in both Mathe-
matics and Quantum Optics.
In Quantum Optics it plays a very important role in order to obtain an analytic
approximate solution of some Schrodinger equation, while it is curious from the math-
ematical point of view.
I explain it carefully with two coherent oscillations for them and expect that they
will overcome the problem in the near future.

Keywords : quantum optics; Rabi model; rotating wave approximation; coherent oscillation.

Mathematics Subject Classification 2010 : 81Q05; 81V80

E-mail address : fujii@yokohama-cu.ac.jp

1
1 Introduction

When undergraduates study Quantum Mechanics they encounter several approximation

methods like the WKB, the BornOppenheimer, the HartreeFock, etc. In fact, exactly
solvable models are very few in Quantum Mechanics, so (many) approximation methods
play an important role. As a text book of Quantum Mechanics I recommend [1] although it
is not necessarily standard.
When we study Quantum Optics we again encounter the same situation. We often use a
method called the rotating wave approximation (RWA), which means fast oscillating terms
(in effective Hamiltonians) removed. Because
Z
in in ein
e = e d = 0
in

holds if n is large enough. We believe that there is no problem on this approximation.

However, in some models slow oscillating terms are removed. Let us show an example.
The Euler formula gives

ei = cos + i sin = 2 cos = ei + ei .

From this we approximate 2 cos to be

2 cos = ei + ei = ei (1 + e2i ) ei

because ei goes away from ei by two times speed, so we neglect this term. In our case n
is 2 ! Read the text for more details.
Why is such a rude method used ? The main reason is to obtain analytic approximate
solutions for some important models in Quantum Optics. To the best of our knowledge we
cannot obtain such analytic solutions without RWA.
In this review note I introduce the rotating wave approximation in details with two
models for undergraduates or nonexperts. I expect that they will overcome this high wall
in the near future.

2
2 Principles of Quantum Mechanics

One of targets of the paper is to study and solve the time evolution of a quantum state
(which is a superposition of two physical states).
In order to set the stage and to introduce proper notation, let us start with a system of
principles of Quantum Mechanics (QM in the following for simplicity). See for example [1],
[2], [3] and [4]. That is,

System of Principles of QM
1. Superposition Principle
If |ai and |bi are physical states then their superposition |ai + |bi is also a physical state
where and are complex numbers.

2. Schrodinger Equation and Evolution

Time evolution of a physical state proceeds like

|i U(t)|i

where U(t) is the unitary evolution operator (U (t)U(t) = U(t)U (t) = 1 and U(0) = 1)
determined by a Schrodinger Equation.

3. Copenhagen Interpretation1
Let a and b be the eigenvalues of an observable Q, and |ai and |bi be the normalized
eigenstates corresponding to a and b. When a state is a superposition |ai + |bi and we
observe the observable Q the state collapses like

|ai + |bi |ai or |ai + |bi |bi

where their collapsing probabilities are ||2 and ||2 respectively (||2 + ||2 = 1).
This is called the collapse of the wave function and the probabilistic interpretation.
1
There are some researchers who are against this terminology, see for example [4]. However, I dont agree
with them because the terminology is nowadays very popular in the world

3
4. Many Particle State and Tensor Product
A multiparticle state can be constructed by the superposition of the Knonecker products of
one particle states, which are called the tensor products. For example,

|ai |ai + |bi |bi |a, ai + |b, bi

is a two particle state.

These will play an essential role in the later sections.

3 TwoLevel System of an Atom

In order to treat the twolevel system of an atom we make a short review of the two
dimensional complex vector space C2 and complex matrix space M(2; C) within our neces-
sity. See for example [5].
First we introduce the (famous) Pauli matrices {1 , 2 , 3 } defined by

0 1 0 i 1 0
1 = , 2 = , 3 = (1)
1 0 i 0 0 1

and set the unit matrx 12 by

1 0
12 = .
0 1
Moreover, we set

1 0 1 1 0 0
+ = (1 + i2 ) = , = (1 i2 ) = .
2 0 0 2 1 0

Note that 1 = + + . Then it is easy to see

1 1 1
[ 3 , + ] = + , [ 3 , ] = , [+ , ] = 2 3 . (2)
2 2 2

Comment The Pauli matrices {1 , 2 , 3 } are generators of the Lie algebra su(2) of the
special unitary group SU(2) and {+ , , 12 3 } are generators of the Lie algebra sl(2; C) of

4
the special linear group SL(2; C). For the sake of readers we write a Liediagram of these
algebras and groups.

sl(2; C) SL(2; C)

su(2) SU(2)

Next, we define {|0i, |1i} a basis of C2 by use of the Diracs notation

1 0
|0i = , |1i = . (3)
0 1

Then, since 1 satisfies the relation

1 |0i = |1i, 1 |1i = |0i

it is called the flip operation.

Note If we define {+ , , 12 3 } as above then {|0i, |1i} should be chosen as

0 1
|0i = , |1i =
1 0

instead of (3). Because,

|0i = 0, |1i = |0i, + |0i = |1i.

However, I use the conventional notations in this note.

For the later convenience we calculate the exponential map. For a square matrix A the
exponential map is defined by

X
X
A (A)n n
e = = An , A0 = E,
n=0
n! n=0
n!

where E is the unit matrix and is a constant.

5
 Here, let us calculate ei1 as an example. Noting
2
0 1 1 0
12 = = = 12
1 0 0 1

we obtain

X
i1 (i)n 1 n
e =
n=0
n!
X
(i)2n 2n X (i)2n+1 2n+1
= 1 + 1
n=0
(2n)! n=0
(2n + 1)!

X
X
(1)n 2n (1)n 2n+1
= 12 + i 1
n=0
(2n)! n=0
(2n + 1)!
= cos 12 + i sin 1

cos i sin
= . (4)
i sin cos

Exercise Calculate
ei2 and ei3 .

We discuss an atom trapped in a cavity and consider only two energy states, namely
(in our case) the ground state and first excited state. That is, all the remaining states are
neglected. This is usually called the twolevel approximation. See for example [6] as a
general introduction.
We set that energies of the ground state |0i and first excited state |1i are E0 and E1 iE0 <
E1 jrespectively. Under this approximation the space of all states is twodimensional, so there
is no problem to identify {|0i, |1i} with (3).
Then we can write the Hamiltonian in a diagonal form like

E0
H0 = .
E1

6
For the later convenience let us transform it. For = E1 E0 the energy difference we have

E0 +E1 E1 E0
E 2
0 = 2
E0 +E1 E1 E0
E1 2
+ 2
E0 + E1 E1 E0
= 12 3
2 2
E0 + E1
= 12 3 . (5)
2 2

To this atom we subject LASER (Light Amplification by Stimulated Emission of Radiation)

in order to controll it. As an image see the following figure.

E1 |1i

E0 |0i

In this note we treat Laser as a classical wave for simplicity, which is not so bad as shown
in the following. That is, we may set the laser field as

A cos(t + ).

By the way, from several experiments we know that an atom subjected by Laser raises
an energy level and vice versa. This is expressed by the property of the Pauli matrix 1

1 |0i = |1i, 1 |1i = |0i,

so we can use 1 as the interaction term of the Hamiltonian.

As a result our Hamiltonian (effective Hamiltonian) can be written as

2
2g cos(t + )
H = 3 + 2g cos(t + )1 = (6)
2 2g cos(t + )
2

7
where g is a coupling constant regarding an interaction of between an atom and laser, and
E0 +E1
A is absorbed in g (gA g). We ignore the scalar term 2
12 for simplicity. Note that
(6) is semiclassical and timedependent.
Therefore, our task is to solve the Schrodinger equation

ih = H (7)
t

exactly (if possible).

4 Rotating Wave Approximation

Unfortunately we cannot solve (7) exactly at the present time. It must be nonintegrable
although we dont know the proof (see the appendix). Therefore we must apply some ap-
proximate method in order to obtain an analytic approximate solution. Now we explain a
method called the Rotating Wave Approximation (RWA). Let us recall the Euler formula

ei = cos + i sin = 2 cos = ei + ei .

From this we approximate 2 cos to be

2 cos = ei + ei = ei (1 + e2i ) ei (8)

because ei goes away from ei by two times speed, so we neglect this term ! We call this
the rotating wave approximation.

ei

|z| = 1 ei

8
Problem In general, fast oscillating terms may be neglected because
Z
ein
ein d = 0
in

if n is large. Our question is : Is n = 2 large enough ?

By noting that the Hamiltonian should be hermitian, we approximate

0 2 cos(t + ) 0 ei(t+)
2 cos(t + )1 = ,
i(t+)
2 cos(t + ) 0 e 0

by use of (8), so (6) is reduced to

2 gei(t+)
e =
H . (9)
i(t+)
ge 2

As a result our modified task is to solve the Schrodinger equation

e
ih = H (10)
t

exactly. Mysteriously enough, this equation can be solved easily.

Note For the latter convenience let us rewrite the method with formal notations :

2 cos 1 = 1 2 cos = (+ + ) (ei + ei )

= + ei + + ei + ei + ei + ei + ei .

In order to solve (10) we set h = 1 for simplicity. From (9) it is easy to see
(t+)
(t+)

2 gei(t+) i
e 2 2 g ei 2
= ,
i(t+) i (t+) i (t+)
ge 2
e 2 g 2
e 2

so we transforn the wave function in (10) into

(t+)
(t+)

i i
e 2 e 2
= = . (11)
i (t+) i (t+)
e 2 e 2

9
Then the Schrodinger equation (10) becomes


2
g
i = (12)
t g
2

by a straightforward calculation.

Here we set the resonance condition

= ( E1 E0 = h precisely). (13)

Namely, we subject the laser field with equal to the energy difference . See the following
figure.

E1 |1i
E1 E0 = h
E0 |0i

Then (12) becomes

0 g
i = = g1
t g 0
and we have only to solve the equation

= ig1 .
t
By (4) ( = gt) the solution is

cos(gt) i sin(gt)
(t) = eigt1 (0) = (0),
i sin(gt) cos(gt)
and coming back to (from ) we obtain
(t+)
i 2
e cos(gt) i sin(gt)
(t) = (t+)
(0)
i 2
e i sin(gt) cos(gt)

(t+) 1 cos(gt) i sin(gt)
= ei 2 (0)
i(t+)
e i sin(gt) cos(gt)

cos(gt) i sin(gt)
= (0) (14)
i(t+) i(t+)
ie sin(gt) e cos(gt)

10
 (t+)
by (11) ((0) = (0)) because the total phase ei 2 can be neglected in Quantum Me-
chanics.
As an initial condition, if we choose

1
(0) = |0i =
0

we have

cos(gt) cos(gt)
(t) = =
i(t+) i(t++/2)
ie sin(gt) e sin(gt)
= cos(gt)|0i + ei(t++/2) sin(gt)|1i (15)

by (3). That is, (t) oscillates between the two states |0i and |1i. This is called the coherent
oscillation or the Rabi oscillation, which plays an essential role in Quantum Optics.
Concerning an application of this oscillation to Quantum Computation see for example
[4].

Problem Our real target is to solve the Schrodinger equation

ih = H
t

with
2 2g cos(t + )
H = H(t) = .

2g cos(t + ) 2

Present a new idea and solve the equation.

5 Quantum Rabi Model

In this section we discuss the quantum Rabi model whose Hamiltonian is given by

H = 3 1 + 12 a a + g1 (a + a )
2

= 3 1 + 12 N + g(+ + ) (a + a ) (16)
2

11
where 1 is the identity operator on the Fock space F generated by the Heisenberg algebra
{a, a , N a a}, and and are constant, and g is a coupling constant. As a general
introduction to this model see for example see [6].
Let us recall the fundamental relations of the Heisenberg algebra

[N, a ] = a , [N, a] = a, [a, a ] = 1. (17)

Here, the Fock space F is a Hilbert space over C given by

F = VectC {|0i, |1i, , |ni, }

where |0i is the vacuum (a|0i = 0) and |ni is given by

(a )n
|ni = |0i for n 0.
n!

On this space the operators (=infinite dimensional matrices) a , a and N are represented as

0 1 0

0 2 1 0


a= 0 3 , a =
2 0 ,


..
0 . 3 0

.. .. ..
. . .

0

1


N =a a=

2
(18)


3

..
.

by use of (17).

Note We can add a phase to {a, a } like

b = ei a, b = ei a , N = b b = a a

12
where is constant. Then we have another Heisenberg algebra

[N, b ] = b , [N, b] = b, [b, b ] = 1.

Again, we would like to solve Schrodinger equation (h = 1 for simplicity)

 

i |i = H|i = 3 1 + 12 N + g(+ + ) (a + a ) |i (19)
t 2

exactly. To the best of our knowledge the exact solution has not been known, so we must
use some approximation in order to obatin an analytic solution.
Since
(+ + ) (a + a ) = + a + + a + a + a ,

we neglect the middle terms + a + a and set

e = 3 1 + 12 N + g(+ a + a ).
H (20)
2

This is called the rotating wave approximation and the resultant Hamiltonian is called the
Jaynes-Cummings one2 , [7].
Therefore, our modified task is to solve the Schrodinger equation
 
e
i |i = H|i = 3 1 + 12 N + g(+ a + a ) |i (21)
t 2

exactly. Mysteriously enough, to solve the equation is very easy.

For a unitary operator U = U(t) we set

|i = U|i.

Then it is easy to see

 
e 1 U 1
i |i = U HU +i U |i
t t
by (21). If we choose U as


eit(N + 2 )
U(t) = eit 2 3 eitN =
it(N 2 )
e
2
In [6] it is called the Jaynes-Cummings-Paul one

13

(we use 2
in place of 2 1 for simplicity), a straightforward calculation gives


U 1 2
ga
e 1 + i
U HU U = (22)
t ga
2

and we have a simple equation

2
ga
i |i = |i. (23)
t ga
2

Note that in the process of calculation we have used the relations

eitN aeitN = eit a, eitN a eitN = eit a .

The proof is easy by use of the formula

1 1
eX AeX = A + [X, A] + [X, [X, A]] + [X, [X, [X, A]]] + (24)
2! 3!

for square matrices X, A and (17).

Here we set the resonance condition

= , (25)

then (23) becomes

ga a
i |i = |i = g |i.
t ga
a

Let us solve this equation. By setting

a
A=
a

we calculate the term eigtA . Noting

aa N +1
A2 = = (= [a, a ] = 1)

aa N

14
we have

X (igt)n An
eigtA =
n=0
n!
X
(igt)2n 2n X (igt)2n+1 2n+1
= A + A
n=0
(2n)! n=0
(2n + 1)!

X n X n
(1) (gt) (N + 1)
n 2n
i
n
(1) (gt) 2n+1

(N + 1) a

=
n=0
(2n)! N n
n=0
(2n + 1)! n
N a

cos( N + 1gt) 1 sin( N + 1gt)a
= i
N +1
1
cos( N gt) sin( N gt)a
N

1
cos( N + 1gt) i N +1 sin( N + 1gt)a

= . (26)
1
i N sin( N gt)a
cos( Ngt)
Therefore, the solution is given by

cos( N + 1gt) i N1+1 sin( N + 1gt)a
|(t)i = |(0)i
1
i N sin( N gt)a cos( Ngt)

and coming back to |i (from |i) we finally obtain

eit(N + 2 ) cos( N + 1gt) i N1+1 sin( N + 1gt)a
|(t)i = |(0)i
it(N 2 ) 1
e i N sin( N gt)a cos( Ngt)
(27)

where |(0)i = |(0)i.

As an initial condition, if we choose

1 |0i
|(0)i = |0i =
0 0
we have

eit 2 cos(gt)|0i
|(t)i = (28)
it 2
ie sin(gt)|1i
because a|0i = 0 and a |0i = |1i, or

1 0
|(t)i = cos(gt) |0i + ei 2 sin(gt) |1i
0 1

15

where the total phase eit 2 has been removed.

Problem Our real target is to solve the Schrodinger equation

ih |i = H|i
t

with

H = H(t) = 3 1 + 12 N + g(+ + ) (a + a )
2
Present a new idea and solve the equation.

As a developed version of the Jaynes-Cummings model see for example [8] and [9].

6 Concluding Remarks

In this note I introduced the rotating wave approximation which plays an important role
in Quantum Optics with two examples. The problem is that the method is used even in
a subtle case. As far as I know it is very hard to obtain an analytic approximate solution
without RWA.
I dont know the reason why it is so. However, such a temporary method must be
overcome in the near future. I expect that young researchers will attack and overcome this
problem.
Concerning a recent criticism to RWA see [10] and its references, and concerning recent
applications to the dynamical Casimir effect see [11], [12] and [13], [14] ([13] and [14] are
highly recommended).

Appendix

[A] Another Approach

16
Let us give another approach to the derivation (4), which may be smart enough. It is easy
to see the diagonal form
1 = W 3 W 1

where W is the WalshHadamard matrix (operation) given by

1 1 1
W = O(2).
2 1 1

Note that
W 2 = 12 = W = W 1 .

Then we obtain

ei1 = eiW 3 W = W ei3 W 1

1

i
1 1 1 e 1 1
=
2 1 1 e i
1 1

ei +ei ei ei
2 2
=
ei ei ei +ei
2 2

cos i sin
= .
i sin cos

Readers should remark that the WalshHadamard matrix W plays an essential role in
Quantum Computation. See for example [15] (note : W UA in this paper).

[B] Tensor Product

Let us give a brief introduction to the tensor product of matrices. For A = (aij ) M(m; C)
and B = (bij ) M(n; C) the tensor product is defined by

A B = (aij ) B = (aij B) M(mn; C).

17
Precisely, in case of m = 2 and n = 3

a11 a12 a11 B a12 B
AB = B =
a21 a22 a21 B a22 B

a b a b a b a12 b11 a12 b12 a12 b13
11 11 11 12 11 13

a11 b21 a11 b22 a11 b23 a12 b21 a12 b22 a12 b23


a11 b31 a11 b32 a11 b33 a12 b31 a12 b32 a12 b33

= .
a b
21 11 a21 b12 a21 b13 a22 b11 a22 b12 a22 b13


a21 b21 a21 b22 a21 b23 a22 b21 a22 b22 a22 b23

a21 b31 a21 b32 a21 b33 a22 b31 a22 b32 a22 b33
When I was a young student in Japan this product was called the Kronecker one. Nowa-
days, it is called the tensor product in a unified manner, which may be better.
Note that
b b b
11 12 13

b21 b22 b23


b31 b32 b33

12 B = ,
b11 b12 b13



b21 b22 b23

b31 b32 b33
while
b b12 b13
11

b11 b12 b13


b21 b22 b23

B 12 = .
b21 b22 b23



b31 b32 b33

b31 b32 b33
The blanks in the matrices above are of course zero.
Readers should recognize the difference. See for example [5] for more details.

[C] Beyond the RWA

18
Let us try to solve the equation (7). For the purpose it is convenient to assume a form for
some solution

(t) = eiF (t)+ eiG(t)3 eiH(t) (0), F (0) = G(0) = H(0) = 0

where we set 3 = (1/2)3 for simplicity. Note that this form called the disentangling form
(a kind of Gauss decomposition of some matrices) is very popular in Quantum Physics.
By setting h = 1 for simplicity in (7) we must calculate

i = {3 + 2g cos(t + )(+ + )}
t
= {2g cos(t + )+ 3 + 2g cos(t + ) }

where 3 = (1/2)3 .
Then we have
 iF (t)+ iG(t)3 iH(t)
i = i e e e (0)
t nt o
iF (t)+ iF (t)+ iF (t)+ iG(t)3 iG(t)3 iF (t)+
= F (t)+ + G(t)e 3 e + H(t)e e e e .

From (2)
[3 , + ] = + , [3 , ] = , [+ , ] = 23

and the formula (24) it is easy to see

eiF (t)+ 3 eiF (t)+ = 3 + iF (t)+ ,

eiG(t)3 eiG(t)3 = (1 + iG(t)) , eiF (t)+ eiF (t)+ = 2iF (t)3 + F (t)2 + .

Therefore
n o
i = F (t)+ + G(t)(3 + iF (t)+ ) + H(t)(1 + iG(t))( 2iF (t)3 + F (t)2 + )
t hn o
2
= F (t) + iF (t)G(t) + (1 + iG(t))F (t) H(t) + +
n o i
G(t) 2i(1 + iG(t))F (t)H(t) 3 + (1 + iG(t))H(t) .

By comparing two equations above we obtain a system of differential equations

F (t) + iF (t)G(t) + (1 + iG(t))F (t)2 H(t) = 2g cos(t + ),


G(t) 2i(1 + iG(t))F (t)H(t) = ,




(1 + iG(t))H(t) = 2g cos(t + ).

19
By deforming them we have


F (t) iF (t) 2g cos(t + )F (t)2 = 2g cos(t + ),


G(t) 4ig cos(t + )F (t) = ,




(1 + iG(t))H(t) = 2g cos(t + ).

This is a simple exercise for young students.

If we can solve the first equation then we obtain solutions like

F (t) = G(t) = H(t).

The first equation

F (t) 2g cos(t + ) iF (t) 2g cos(t + )F (t)2 = 0

is a (famous) Riccati equation of general type. Unfortunately, we dont know how to solve
it explicitly at the present time.

[D] Full Calculation

Let us give the full calculation to the equation (23). We set


ga
B= 2

ga 2

and calculate eitB without assuming = in (25). Again, noting

2 2 2
2
+ g aa
2
+ g2N + g2
2
B = =


2 2


2 2
2
+ g a a 2
+g N

(aa = a a + 1 = N + 1) we obtain


ga
itB 2
e = exp it
ga
2
p
i sint +g 2 sin t +g 2
2
cos t + g 2 +g 2
ig
+g 2
a
=
sin t sin t
ig a cos t + i2

20
where we have set
2
, + g2N
4
for simplicity. See (26). This is a good exercise for young students.

References

[1] H. S. Green : Matrix Mechanics, P. Noordhoff Ltd, Groningen, 1965.

[2] P. Dirac : The Principles of Quantum Mechanics, Fourth Edition, Oxford Uni-
versity Press, 1958.

[3] Asher Peres : Quantum Theory : Concepts and Methods, Kluwer Academic
Publishers, 1995.

[4] Akio Hosoya : Lectures on Quantum Computation (in Japanese), SGC Library
4, Saiensu-sha Co., Ltd. Publishers (Tokyo), 1999.

[5] K. Fujii and et al : Treasure Box of Mathematical Sciences (in Japanese), Yusei-
sha, Tokyo, 2010.

[6] W. P. Schleich : Quantum Optics in Phase Space, WILEYVCH, Berlin, 2001.

[7] E. T. Jaynes and F. W. Cummings : Comparison of Quantum and Semiclassical Radi-

ation Theories with Applications to the Beam Maser, Proc. IEEE, 51 (1963), 89.

[8] K. Fujii and T. Suzuki : An Approximate Solution of the JaynesCummings Model

with Dissipation, Int. J. Geom. Methods Mod. Phys, 8 (2011), 1799, arXiv : 1103.0329
[math-ph].

[9] K. Fujii and T. Suzuki : An Approximate Solution of the JaynesCummings Model

with Dissipation II : Another Approach, Int. J. Geom. Methods Mod. Phys, 9 (2012),
1250036, arXiv : 1108.2322 [math-ph].

21
[10] J. Larson : On the rotating wave approximation in the adiabatic limit, Physica Scripta,
2013, 014040, arXiv : 1208.1891 [quant-ph].

[11] C. K. Law : Effective Hamiltonian for the radiation in a cavity with a moving mirror
and a timevarying dielectric medium, Phys. Rev. A 49 (1994), 309.

[12] A. V. Dodonov and V. V. Dodonov : Approximate analytical results on the cavity

Casimir effect in the presence of a twolevel atom, Phys. Rev. A 85 (2012), 063804.
arXiv : 1112.0523 [quant-ph].

[13] K. Fujii and T. Suzuki : An Approximate Solution of the Dynamical Casimir Effect
in a Cavity with a TwoLevel Atom, Int. J. Geom. Methods Mod. Phys, 10 (2013),
1350035, arXiv : 1209.5133 [quant-ph].

[14] K. Fujii and T. Suzuki : Rotating Wave Approximation of the Laws Effective Hamil-
tonian on the Dynamical Casimir Effect, Int. J. Geom. Methods Mod. Phys, 11 (2014),
1450003, arXiv : 1209.5133 [quant-ph].

[15] A. Barenco : Quantum Physics and Computers, Contemp. Phys. 37 (1996), 375,
quant-ph/9612014.

22