Anda di halaman 1dari 21

Practical Guide to Solving Coupled Oscillator Problems

November 8, 2013
1 Two and Three Coupled Oscillators
1.1 Formal Preliminaries
Consider two particles that are constrained to move in one dimension (for example on an air track).
Their positions are specied by their respective distances x
1
and x
2
from their equilibrium positions.
These two positions can be represented as a two-component vector
x =
_
x
1
x
2
_
(1)
that can be represented as a point in a two-dimensional plane as shown in Fig. (2). An alternative
but equivalent representation of x(t) relies on the introduction of vectors e
1
and e
2
aligned along
the x
1
and x
2
directions:
e
1
=
_
1
0
_

_
e
1,1
e
1,2
_
; e
2
=
_
0
1
_

_
e
2,1
e
2,2
_
, (2)
where e
1,1
= 1, e
1,2
= 0, e
2,1
= 0 and e
2,2
= 1 or, equivalently, e
i,j
=
ij
, where
ij
is the Kroncker
delta that is equal to 1 if i = j and zero otherwise. These vectors form an orthonormal pair , i.e.,
they are of unit length and they are perpendicular to each other. They satisfy
e
i
e
j

k
e
i,k
e
j,k
=
ij
i, j = 1, 2 (3)
and

i
e
i,j
, e
i,k
=
jk
. (4)
By construction any vector x can be uniquely expressed as
x = x
1
e
1
+ x
2
e
2
x
i
e
i
, (5)
where the summation convention over i is understood. Any set of vectors (e

1
, e

2
) that provide a
unique representation
x = x

1
e

1
+ x

2
e

2
x

i
e

i
, (6)
of any vector x is is called a basis. If in addition the vectors satisfy the orthogonality relation
e

i
e

j
=
ij
the basis is orthonormal. The vectors (e
1
, e
2
) form a special basis called the standard
basis. The jth component x

j
, j = 1, 2 of x with respect to the (e

1
, e

2
) basis is simply
x

j
= e

j
x = e

j
(x

i
e

i
) = x

ij
= e

j,i
x

i
. (7)
1
m m
k
a
k
b
k
c
Figure 1: Two-mass oscillator
In addition to satisfying the orthogonality relation of Eq. (3), the vectors of an orthonormal basis
satisfy the completeness relation,

i
e

i,k
, e

i,j
=
kj
. (8)
Note that the sum is over the rst index, indicating the particular vector, rather than over the second
index, indicating the component of the vector, as in the orthogonality relation. The completeness
relation follows from the fact that (e

1
, e

2
) form a basis and any vector can be uniquely represented
in the from of Eq. (6). The j-component of x in the prime basis is
x

j
= x

i
e

j,i
= (x e

i
)e

ij
= x

k
e

i,k
e

i,j
. (9)
The two sides can be equal only if the completeness relation, Eq. (8) is satised. (Remember that
the summation convention is used on repeated indices.)
If the particles are coupled with three springs with respective spring constants k
a
, k
b
, k
c
as
shown in Fig. 1, their equations of motion are
m x
1
(t) = k
a
x
1
k
b
(x
1
x
2
)
m x
2
(t) = k
b
(x
2
x
1
) k
c
x
2
. (10)
This equation can be written in a compact form with the introduction of the stiness matrix,
kkk =
_
k
a
+ k
b
k
b
k
b
k
b
+ k
c
_

_
k
12
k
12
k
12
k
22
_
(11)
where k
ij
, i, j = 1, 2 are the components of the symmetric stiness matrix with k
ij
= k
ji
. Then
x(t) =
1
m
kkk x(t) DDDx(t), (12)
where
DDD =
_
(k
a
+ k
b
)/m k
b
/m
k
b
/m (k
b
+ k
c
)/m
_

2
a
+
2
b

2
b

2
b

2
b
+
2
c
_
(13)
is the dynamical matrix.
We expect that the masses will undergo simple harmonic oscillations with two distinct frequen-
cies, so we assume that x(t) has the form
x(t) = e
it
a = e
it
_
a
1
a
2
_
, (14)
where a
1
and a
2
are complex amplitudes with independent magnitudes and phases for a total of
four independent variables. The physical solutions are obtained by taking the real part of or x(t).
The second derivative with respect to time brings down a factor of
2
so that Eq. (13) becomes

2
a = DDDa or (DDD
2
III) a = 0, (15)
2
a homogeneous equation that can only be satised when
2
is an eigenvalue and a is an associated
eigenvector of the the dynamical matrix DDD. III is the identity matrix with components I
ij
=
ij
. It
has the property that IIIa = a for any vector a.
A fundamental theorem of algebra is that a system of homogenous linear equations has a solution
if an only if the determinant of the matrix of its coecients is zero. To see how this result comes
about in the simplest form, consider an arbitrary 2 2 matrix
CCC =
_
C
11
C
12
C
21
C
22
_
. (16)
The homogenous equation CCCa = 0 in component form is
C
11
a
1
+ C
12
a
2
= 0 (17)
C
21
a
1
+ C
22
a
2
= 0. (18)
If C
22
> 0, then from Eq. (18),
a
2
=
C
21
C
22
a
1
, (19)
which when used in Eq. (17) yields
C
11
a
1
+ C
12
a
2
=
_
C
11

C
12
C
21
C
22
_
a
1
=
1
C
22
(C
11
C
22
C
12
C
21
)a
1
=
a
1
C
22
det CCC = 0, (20)
where
det CCC = C
11
C
22
C
12
C
21
(21)
is the determinant of CCC. Thus either a
1
= 0 or det CCC = 0. If a
1
= 0, then a
2
= 0 provided C
21
= 0.
This is an uninteresting case in which a
1
= a
2
= 0. If C
21
= 0, then C
22
a
2
= 0 and either C
22
= 0
or a
2
= 0. The latter case is the uninteresting one with a
1
= a
2
= 0 again. If C
22
= 0 and C
21
= 0
and det CCC = 0. A similar line of argument can be used if C
11
= 0. The end result is that in all
cases, either a
1
= a
2
= 0 or det CCC = 0. Clearly the latter case is the only interesting one.
Thus to nd the eigenvalues of the dynamical matrix, we need to solve the equation
det(DDD
2
III) = 0. (22)
This determinant is a second-order polynomial, P(
2
), in
2
, called the characteristic polynomial
of DDD, whose roots, which are the eigenvalues, can be calculated for general DDD from the quadratic
formula.
1.2 A Simple Two-particle Example
To keep our discussion simple for the moment, we will rst consider simple special cases. In
particular, we consider the case in which
2
c
=
2
b
. The characteristic equation in this case is
P(
2
) = det
_

2
a
+
2
b

2

2
b

2
b

2
a
+
2
b

2
_
= (
2
a
+
2
b

2
)
2

4
b
(23)
Its roots are
(
2

2
a

2
b
) =
2
b
, (24)
3
e
1
e
2
2
e
1
e
e
1
e
2
x
2
e
1
e
x
1
x
2
1
x
2
x

1,1
e
2,1
e
1,2
e
2,2
e
(a) (b)
Figure 2: (a) The standard basis vectors e
1
and e
2
and the eigenvectors e
1
and e
2
for the case
in which the rotation angle is 45

. The components e
,1
and e
,2
of e

, = 1, 2 are simply the


projects of these vectors onto the e
1
and e
2
axes. (b) The standard and rotated basis vectors for
an arbitrary rotation angle showing how the components of an arbitrary vector can be represented
with respect to either basis.
or

2
1
=
2
a

2
2
=
2
a
+ 2
2
b
. (25)
These are the same frequencies we obtained by guessing the form of the displacements with a xed
frequency from the air track demonstration in class.
Associated with these eigenfrequencies are eigenvectors a
1
= (a
1,1
, a
1,2
) and a
2
= (a
2,1
, a
2,2
),
which are obtained by requiring DDDa
1

2
1
a
1
= 0 and DDDa
2

2
2
a
2
= 0. We start with the equation
for a
1
:
(
2
a
+
2
b

2
1
)a
1,1

2
b
a
1,2
=
2
b
a
1,1

2
b
a
1,2
= 0

2
b
a
1,1
+ (
2
a
+
2
b

2
1
)a
1,2
=
2
b
a
1,1
+
2
a
a
1,2
= 0. (26)
Both of these equations yield a
1,2
= a
2,1
. Of course if they did not yield the same result, we would
be in trouble, but we are guaranteed that they do because det DDD
2
1
III = 0 guarantees that the two
equations are linearly proportional to each other. Similarly, keeping only one equation,
(
2
a
+
2
b

2
2
)a
2,1

2
b
a
2,2
=
2
b
a
2,1

2
b
a
22
= 0 (27)
and a
2,1
= a
2,2
. If a

( = 1, 2) is an eigenvector, then any constant times a

is an eigenvector.
We will denote the normalized version of a

as e

, where
e
1
=
1

2
_
1
1
_
=
_
cos
sin
_
, e
2
=
1

2
_
1
1
_
=
_
sin
cos
_
, (28)
where for this case = 45

. For more general values of the elastic constants, can take on any value
as depicted in Fig. 2. These two vectors are orthogonal, i.e, e
1
e
2
= 0, and they were constructed
to have unit length. Thus the form an orthonormal pair with
e

=
,
, , = 1, 2 (29)
where are before,
,
is the Kronecker delta that is equal to one if = and zero otherwise.
The components e
,i
of e

are the coecients of e


i
in the representation of e

in the standard
basis:
e

= e
,i
e
i
= U
,i
e
i
, (30)
4
where as usual the summation convention on repeated indices is understood, U
,i
= e
,i
are the
components of the matrix
UUU =
_
e
1,1
e
1,2
e
2,1
e
2,1
_
=
1

2
_
1 1
1 1
_
=
_
cos sin
sin cos
_
. (31)
The set ( e
1
, e
2
) is merely a rotated version of the basis (e
1
, e
2
) as shown in Fig. 2. In this case, the
rotation angle is 45

, but in more general problems, it can be any angle, as we shall see shortly.
The matrix U is a rotation matrix.
The inverse AAA
1
of any matrix AAA is dened by
AAA
1
AAA = AAAAAA
1
= III (32)
Rotation matrices like UUU with components U
,i
have the property that their inverse is simply their
transpose UUU
T
whose components are U
T
i,
= U
,i
. Thus
UUU
T
=
1

2
_
1 1
1 1
_
=
_
cos sin
sin cos
_
. (33)
and
UUU
T
UUU =
1
2
_
1 1
1 1
__
1 1
1 1
_
=
_
1 0
0 1
_
= III. (34)
ThusUUU
T
is the inverse of UUU.
It is clear from the geometry in Fig. 2 that any vector x can be represented in terms of its
components (x
1
, x
2
) with respect to the the standard basis or in terms of its components ( x
1
, x
2
)
relative the e
1
and e
2
. In other words, the set ( e
1
, e
2
) form a basis, and
x = x
1
e
1
+ x
2
e
2
x
i
e
i
= x
1
e
1
+ x
2
e
2
x

. (35)
Taking the dot produce of both sides of this equation with e

yields
x

= x e

= x
i
e
i
e

= x
i
e
,i
= U
,i
x
i
. (36)
Dening x to be the column vector with components x

, this equation can be rewritten as


x = UUUx. (37)
Now what can we do with all of this? First, lets return to the original dynamical equation,
Eq. (12), and express x(t) in the tilde basis ( e
1
, e
2
):
x =

x

= DDD x

(38)
where the last step follows because e

is the eigenvector of DDD with eigenvalue


2

. Taking the dot


produce of both sides of the equation with e

, = 1, 2 leads to

=
2

, = 1, 2. (39)
In other words x
1
and x
2
obey the equations of a simple harmonic oscillators with respective
frequencies
1
and
2
. We have reduced the coupled oscillator equations to uncoupled single
oscillator equations. x
1
and x
2
are the normal modes of the system, often denoted as q
1
(t) and
5
q
2
(t). The solutions to the uncoupled equations are the familiar solutions to the simple oscillator,
i.e,.
x

(t) q

(t) =

A

e
it
or (40)
q

(t) = B

cos

t + C

sin

(41)
= A

cos(

) (42)
We now have the general solution for the coupled oscillator problem:
x(t) =

(t) e

_
B

cos

t + C

sin

_
e

(43)
x(t) =

sin

t + C

cos

t) e

. (44)
The unknown coecients

A

(or B

and C

) are determined by boundary conditions that are


usually specied at t = 0. Setting x(0) = x
0
= (x
0,1
, x
0,2
) and

x(0) = v
0
= (v
0,1
, v
0,2
), the
boundary conditions can be expressed as
x
0
= B
1
e
1
+ B
2
e
2
; v
0
= C
1
e
1
+ C
2
e
2
. (45)
The orthonormal property of e

allows us to calculatesd the unknown coecients by simply taking


the dot product of these two equation with e
1
and e
1
:
B
1
= x
0
e
1
=
1

2
(x
0,1
+ x
0,2
) (46)
B
2
= x
0
e
2
=
1

2
(x
0,1
+ x
0,2
) (47)
C
1
= v
0
e
1
=
1

2
(v
0,1
+ v
0,2
) (48)
C
2
= v
0
e
2
=
1

2
(v
0,1
+ v
0,2
). (49)
Putting these terms back into Eq. (43), we obtain
x
1
(t) = q
1
(t) e
1,1
+ q
2
(t) e
2,1
=
1
2
(x
0,1
+ x
0,2
) cos
1
t +
1
2
(x
0,1
x
0,2
) cos
2
t +
1
2
(v
0,1
+ v
0,2
)
sin
1
t

1
+
1
2
(v
0,1
v
0,2
)
sin
2
t

1
(50)
x
2
(t) = q
2
(t) e
1,2
+ q
2
(t) e
2,2
=
1
2
(x
0,1
+ x
0,2
) cos
1
t
1
2
(x
0,1
x
0,2
) cos
2
t +
1
2
(v
0,1
+ v
0,2
)
sin
1
t

1
+
1
2
(v
0,1
v
0,2
)
sin
2
t

1
(51)
This equation can easily be applied to special cases. For, example if x
0,1
= x
0
and x
0,2
= 0,
v
0,1
= v
0,2
= 0, the solutions are
x
1
(t) =
1
2
x
0
(cos
1
t + cos
2
t)
x
2
(t) =
1
2
x
0
(cos
1
t cos
2
t) (52)
6
1.3 A More Complex Two-Particle example
In the example we just studied, the normal modes were simple. In one mode the two masses moved
in the same direction the same amount, and in the second, they moved in opposite directions
the same amount. We now consider a somewhat more complicated realization of the two-mass,
two-spring problem in which k
b
= 2k
a
and k
c
= 4k
a
. From Eq. (13),
DDD =
2
a
_
3 2
2 6
_
, (53)
where
2
a
= k
a
/m. Clearly the eigenvalues of DDD will be proportional to
2
a
, so we set
2
=
2
a
s.
The characteristic polynomial is then
P(
2
) =
4
0
[(3 s)(6 s) 4] =
2
0
(s
2
9s + 14), (54)
and its roots are s
1
= 2 and s
2
= 7 or
2
1
= 2
2
a
and
2
2
= 7
2
a
. The eigenvectors a
1
= (a
1,1
, a
1,2
)
and a
2
= (a
2,1
, a
2,2
) are determined by
(3
2
a
2
2
a
)a
1,1
2
2
a
a
1,2
= =
2
a
[a
1,1
2a
1,2
] = 0 (55)
(3
2
a
7
2
a
)a
2,1
2
2
a
a
2,2
=
2
a
[4a
2,1
2a
2,2
] = 0. (56)
Thus the normalized eigenvectors are
e
1
=
1

5
_
2
1
_
=
_
cos
sin
_
, e
2
=
1

5
_
1
2
_
=
_
sin
cos
_
, (57)
where = tan
1
(1/2) = 26.6

. These vectors clearly satisfy the orthonormality and completeness


relations e

=
,
and

e
,i
e
,j
=
ij
.
Set boundary conditions x
0
= x
0
(1, 1) and x(0) = v
0
= v
0
(1, 1), then
B
1
= x
0
e
1
= x
0
_
2

5
+
1

5
_
=
3

5
x
0
B
2
= x
0
e
2
= x
0
_

5
+
2

5
_
=
1

5
x
0
C
1
= v
0
e
1
= v
0
_
2

5
_
=
1

5
v
0
C
2
= v
0
e
2
= v
0
_

5
_
=
3

5
v
0
(58)
and
x
1
(t) =
1
5
x
0
(6 cos
1
t cos
2
t) +
1
5
v
0
_
2
sin
1
t

1
+ 3
sin
2
t

2
_
x
2
(t) =
1
5
x
0
(3 cos
1
t + 2 cos
2
t) +
1
5
v
0
_
sin
1
t

1
6
sin
2
t

2
_
. (59)
7
m m m
k
a
k
b
k
c
k
d
Figure 3: Three-mass oscillator
1.4 A Three Oscillator Problem
In the previous sections, we set up a formalism for solving the eigenvalues and eigenvectors of the
dynamical matrix. We focussed on simple two-oscillator examples, but the formalism is valid for
oscillator systems with an arbitrary number of degrees of freedom. Basically, each particle has
one translational degree of freedom per spatial dimension. Thus, if there are N particles in d
dimensions, there will be dN degrees of freedom, and the dynamical matrix will a dN dimensional
square matrix (i.e., a dN dN matrix). Here we will study a simple three-oscillator system in
one-dimension with three degrees of freedom. The system consists of three masses and four springs
as shown in Fig. 3.
The springs from left to right have k
a
= 4k, k
b
= 3k, k
c
= 4k, and k
d
= 3k. The equations of
motion are
x
1
(t) = 4
2
0
x
1
3
2
0
(x
1
x
2
)
x
2
(t) = 3
2
0
(x
2
x
1
) 4
2
0
(x
2
x
3
)
x
3
(t) = 4
2
0
(x
3
x
2
) 3
2
0
x
3
, (60)
where
2
0
= k/m. The dynamical matrix is then
DDD =
2
0
_
_
7 3 0
3 7 4
0 4 7
_
_
, (61)
and setting
2
= s, the characteristic polynomial, det(DDD
2
0
s) is
p(s)
P(s
2
0
)

6
0
= (7 s)[(7 s)
2
16] 9(7 s) = (7 s)[(7 s)
2
25] = 0 (62)
The eigenvalues (in units of
2
0
) are s = 7, s = 7

25 or
s
1
= 2, s
2
= 7, s
3
= 12. (63)
The associated eigenvectors a

= (a
,1
, a
,2
, a
,3
), = 1, 2, 3, are determined by solving (DDD
s

2
0
)aaa

= 0. We solve these equations in order:


s = 2 ,
1

2
0
(DDD2
2
0
)a
1
=
_
_
5 3 0
3 5 4
0 4 5
_
_
_
_
a
1,1
a
1,2
a
1,3
_
_
=
_
_
5a
1,1
3a
1,2
3a
1,1
+ 5a
1,2
4a
1,3
4a
1,2
+ 5a
1,3
_
_
=
_
_
0
0
0
_
_
(64)
The rst and third columns yield a
1,1
= (3/5)a
1,2
and a
1,3
= (4/5)a
1,2
. The equation from the
second column is automatically satised by these relations. The rst normalized eigenvector is,
therefore,
e
1
=
1
5

2
_
_
3
5
4
_
_
(65)
8
The second eigenvector is next:
s = 7 ,
1

2
0
(DDD 7
2
0
)a
1
=
_
_
0 3 0
3 0 4
0 4 0
_
_
_
_
a
2,1
a
2,2
a
2,3
_
_
=
_
_
3a
2,2
3a
2,1
4a
2,3
4a
2,2
_
_
=
_
_
0
0
0
_
_
. (66)
Thus a
2,2
= 0, and a
2,3
= (3/4)a
2,1
, and the normalized second eigenvector is
e
2
=
1
5
_
_
4
0
3
_
_
. (67)
Finally, the third eigenvector
s = 12 :
1

2
0
(DDD 12
2
0
)a
1
=
_
_
5 3 0
3 5 4
0 4 5
_
_
_
_
a
3,1
a
3,2
a
3,3
_
_
=
_
_
5a
3,1
3a
3,2
3a
3,1
5a
3,2
4a
1,3
4a
3,2
5a
3,3
_
_
=
_
_
0
0
0
_
_
.
(68)
Thus, a
3,1
= (3/5)a
3,2
, a
3,3
= (4/5)a
3,2
, and the normalized third eigenvector is
e
3
=
1
5

2
_
_
3
5
4
_
_
(69)
It is a straightforward exercise to verify the orthogonality and completeness relations: e

=
,
and

3
=1
e
,i
e
,j
=
ij
.
The general solution for this set of oscillators is thus
x(t) =
3

=1
_
B

cos

t + C

sin

_
e

(70)
If the initial conditions x
0
and v
0
are given for the displacement and velocities, then
B

= x
0
e

, C

= v
0
e

. (71)
If x
0
= (x
0
, 0, 0) and v
0
= (0, 0, 0),
B
1
=
3
5

2
x
0
, B
2
=
4
5
x
0
, B
3
=
3
5

2
, (72)
C
1
= C
2
= C
3
= 0, and
_
_
x
1
(t)
x
2
(t)
x
3
(t)
_
_
= x
0
_
_
1
50
(9 cos
1
t + 32 cos
2
t + 9 cos
3
t)
3
10
(cos
1
t cos
3
t)
6
25
(cos
1
t 2 cos
2
t + cos
3
t)
_
_
(73)
1.5 Dierent Masses
So far, we have only considered systems in which the masses are equal. When they are no equal,
the most direct approach leads to dynamical matrices that are not symmetric. This matrix, as
we shall see, still has real eigenvalues as it must, but its right an left eigenvectors are dierent.
This is something of a nuisance, and it is advantageous in general to formulate the problem so that
9
the dynamical matrix is symmetric. We rst look at the the problem of two coupled oscillators
with dierent masses using the direct approach, and then we look at it using the approach with
a symmetric dynamical matrix. Consider the system depicted in Fig. (1) but with the right mass
having mass m
1
, the left mass having mass m
2
, and k
c
= k
a
. The equations of motion are
m
1
x
1
= k
a
x
1
k
b
(x
1
x
2
)
m
2
x
2
= k
b
(x
2
x
1
) k
a
x
2
. (74)
In compact form, this equation is
mmm x = kkkx (75)
x = mmm
1
kkkx DDD

x, (76)
where the stiness matrix kkk is the same as Eq. (11) and the mass matrix is
mmm =
_
m
1
0
0 m
2
_
. (77)
so that
DDD

=
_
ka+k
b
m
1

k
b
m
1

k
b
m
2
ka+k
b
m
2
_
=
_

2
a
+
2
b

2
b

2
b
(
2
a
+
2
b
)
_
(78)
where
2

= k

/m
1
for = a, b, c and = m
1
/m
2
. The eigenvalues, which are the roots of the
characteristic polynomial,
P(
2
) = det
_

2
III +DDD

_
= det
_

2
a
+
2
b

2

2
b

2
b
(
2
a
+
2
b
)
2
_
(79)
= [
2
a
+
2
b

2
][(
2
a
+
2
b
)
2
]
4
b
, (80)
are
P(
2
) =
1
2
_
(1 + )
2
a
+
2
b

_
[(1 )
2
(
2
a
+
2
b
)
2
4
4
b
_
. (81)
The matrix DDD

is not symmetric. Though it is not necessary for the dynamical matrix to be


symmetric, there are advantages to setting up a formalism in which it is. These include dealing
with a matrix form that is guaranteed to have real eigenvalues and one whose right and left
eigenvectors are transposes of each other. If an nn square matrix AAA has eigenvalues

, = 1, . . . n,
it has right eigenvectors r

such that AAAr


n
=
n
r
n
and left eigenvectors l

such that AAAl

. For
symmetric matices, l

= r
T

. To convert the equation of motion, Eq. (75) to one with a symmetric


dynamical matrix, we introduce the square root, mmm
1/2
, and the inverse square root, mmm
1/2
, of the
mass matrix:
mmm
1/2
=
_
m
1/2
1
0
0 m
1/2
2
_
mmm
1/2
=
_
m
1/2
1
0
0 m
1/2
2
_
. (82)
It is straightforward to verify that mmm
1/2
mmm
1/2
= mmm and that mmm
1/2
mmm
1/2
= III. We next introduce the
vector
y = mmm
1/2
x =
_
m
1/2
1
0
0 m
1/2
2
_
_
x
1
x
2
_
=
_

m
1
x
1

m
2
x
2
_
. (83)
This equation can be inverted to give x in terms of y:
x = mmm
1/2
y =
_
y
1
/

m
1
y
2
/

m
2
_
. (84)
10
With these denitions, we can reexpress mmmx = mmm
1/2
y and kkkx = kkkmmm
1/2
mmm
1/2
x = kkkmmm
1/2
y. Thus
mmm x = mmm
1/2
y = kkkx = kkkmmm
1/2
y or
y = DDDy, (85)
where
DDD = mmm
1/2
kkkmmm
1/2
=
_
k
11
m
1
k
12

m
1
m
2
k
12

m
1
m
2
k
22
m
2
_
=
_

2
a
+
2
b

2
b

2
b
(
2
a
+
2
b
)
_
(86)
is the now symmetrized dynamical matrix. The only price we had to pay for this transformation is
the introduction of the vector y. It is clear that characteristic polynomial associated with DDD is the
same as that associated with DDD

.
Our next step is to determine the eigenvectors of D. Though we can do this analytically for
general
2
s
, s = a, b, c and , the expressions are messy, so we consider only a special case in which

2
a
= 25
2
0
,
2
b
= 7
2
0
, and = 1/4. Then
DDD =
2
0
_
32 7/2
7/2 8
_
andDDD
2
III =
2
_
32 s 7/2
7/2 8 s
_
(87)
where
2
=
2
0
s and the characteristic polynomial is
P(s
2
0
) =
2
0
[(32 s)(8 s) (49/4)] =

0
2
[s
4
40s + 8 32 (49/4)]. (88)
The roots of P give the eigenvalues:

2
=

2
0
2
_
40
_
40
2
32
2
+ 49
_
=

2
0
2
[40 25] (89)
or

2
1
=
15
2

2
0
,
2
2
=
65
2

2
0
. (90)
The eigenvectors are calculated as in previous examples. They are
e
1
=
1

50
_
1
7
_
=
_
cos
sin
_
e
2
=
1

50
_
7
1
_
=
_
sin
cos
_
(91)
where cos = 1/

50 and sin = 7/

50. The nal form in terms of cos and sin is valid for any
values of k
a
, k
b
, k
c
, m
1
and m
2
but with dierent values of .
Solving initial value problem proceeds exactly as before, except we have to keep straight factors
of

m
1
and

m
2
. Consider rst the case in which x(0) = (x
0
, 0) and x(t = 0) = (0, 0). Then
y(0) = (

m
1
x
0
, 0) and y(t = 0) = (0, 0), and
y(t) = q
1
(t) e
1
+ q
2
(t) e
2
= B
1
cos
1
t e
1
+ B
2
cos
2
t e
2
(92)
so that
y(0) = B
1
e
1
+ B
2
e
1
(93)
This implies
B
1
= y(0) e
1
=

m
1
x
0
e
1,1
=
_
m
1
50
x
0
B
2
= y(0) e
2
=

m
1
x
0
e
2,1
= 7
_
m
1
50
x
0
11
To nd x(t), we simply multiply y(t) by mmm
1/2
to obtain
x
1
(t) = m
1/2
1
y
1
(t) = x
0
( e
2
1,1
cos
1
t + e
2
2,1
cos
2
t) = x
0
(cos
2
cos
1
t + sin
2
cos
2
t)
=
x
0
50
(cos
1
t + 49 cos
2
t) (94)
x
2
(t) = m
1/2
2
y
2
(t) =
_
m
2
m
1
x
0
( e
1,1
e
1,2
cos
1
t + e
2,1
e
2,2
cos
2
t)
=
_
m
2
m
1
x
0
cos sin (cos
1
t cos
2
t)
=
7x
0
50
(cos
1
t cos
2
t). (95)
This result should be contrasted with that we obtained with equal masses. There cos = sin =
1/

2 so that
x
1
(t) =
1
2
x
0
(cos
1
+ cos
2
t) = x
0
cos
av
t cos t (96)
x
2
(t) =
1
2
x
0
(cos
1
cos
2
t) = x
0
sin
av
t sin t, (97)
where
av
= (
1
+
2
)/2 and = (
2

1
)/2. These equations describe a beating pattern
in which the lower frequency parts sin t and cos t provide an envelope for the more rapidly
varying parts oscillating with frequency
av
. When sin t = 0 (i.e., t = n/ for any integer n,
x
2
(t) = 0 and the envelope of x
1
(t) is a maximum. Similarly, when the cos t = 1, the envelope
of x
1
(t) is zero that of x
2
(t) is a maximum. The result as we say in class is that in the beginning,
mass 2 appears to be at rest, then both masses move, then mass one appears to be at rest and so
on periodically as shown in Fig. 4 (a) and (c). In the unequal mass case, cos = sin , and the
beating is less obvious as can be seen by the following alternative (but equivalent) expressions for
x
1
(t) and x
2
(t):
x
1
(t) =
1
2
x
0
[cos
av
t cos t + cos 2(sin
av
t sin t)] (98)
x
2
(t) =
1
2
x
0
sin 2 sin
av
t sin t (99)
Clearly the envelope of x
1
(t) never goes to zero as shown in Fig. 4 (b) and (d).
2 N-coupled oscillators
2.1 Fixed Boundary Conditions
Consider N particles of mass m connected by springs and constrained to move in one-dimension as
shown for three particles in Fig. 3. For the moment, we consider only the case in which all springs
have the same spring constant k. There are two boundary conditions on these oscillators; the left
end of the rst spring is clamped at the left wall and the right end of the last spring is clamped at
the right wall. These boundary conditions are referred to as xed boundary conditions.It is useful
to imagine that the walls are clamped particles, which we label as particle 0 and particle N + 1,
respectively. The set-up is now that there are N +2 particles 0, 1, ...N, N +1, where the positions of
the rst and last particles are clamped at the walls. The equations of motion for particle 2, ..., N1
all have the same form because they have identical environments, being connected via springs to a
12
5 10 15 20 25 30
-1.0
-0.5
0.5
1.0
(a)
5 10 15 20 25 30
-1.0
-0.5
0.5
1.0
(b)
5 10 15 20 25 30
-1.0
-0.5
0.5
1.0
(c)
5 10 15 20 25 30
-0.5
0.5
(d)
Figure 4: (a) x
1
(t) and (c) x
2
(t) for equal-mass coupled oscillators; (b) x
1
(t) and (d) x
2
(t) for
unequal mass coupled oscillators. Note in the equal-mass case, x
1
(t) oscillates near zero when x
2
(t)
oscillates near its max and vice versa. In the unequal mass case, the envelope of ov x
1
(t) never
goes to zero.
mobile particle on the right and one on the left. Particles 1 and N, on the other hand are connected
to an immobile wall on one side, and their equations of motion look dierent:
x
1
= k(x
1
x
0
) k(x
1
x
2
) = 2kx
1
+ kx
2
x
2
= k(x
2
x
1
) k(x
2
x
3
) = k(2x
2
x
3
x
1
)
... ...
... ...
... ...
x
p
= k(x
p
x
p1
) k(x
p
x
p+1
) = k(2x
p
x
p+1
x
p1
)
... ...
... ...
... ...
x
N
= k(x
N
x
N+1
) k(x
N
x
N1
) = 2kx
N
+ kx
N1
(100)
The usual strategy for determining the eigenvalues and eigenvectors of this system of equations is
to guess (ansatz ) the form of the eigenvectors and use their free degrees of freedom to ensure
that the rst and last equations are satised. We, therefore, make the following ansatz:
x
q,p
(t) = A
q
e
it
sin(qp +
q
) (101)
where q is an index (whose form is to be determined) specifying the particular eigenmode, p = 1, ...N
labels the position of particle p, and
q
is a phase which we will use to set boundary conditions. In
13
vector notation, this ansatz corresponds to an vector
x
q
(t) = A
q
e
it
_
_
_
_
_
_
_
_
sin(q +
q
)
sin(2q +
q
)
.
.
.
sin(Nq +
q
)
_
_
_
_
_
_
_
_
. (102)
If this ansatz is right, then it must be that (2x
p
x
p+1
x
p1
) be proportional to x
p
for every p.
Consider, therefore,
2 sin(qp +
q
) sin(q(p + 1) +
q
) sin(q(p 1) +
q
)
= 2 sin(qp +
q
) (sin(qp +
q
) cos q + cos(qp +
q
) sin q + sin(qp +
q
) cos q cos(qp +
q
) sin q)
= 2(1 cos q) sin(pq +
q
). (103)
Thus, provided we satisfy the boundary equations [p = 1 and p = N], we have (using x
q,p
=

2
x
q,p
), each term in Eqs. (100) have the same form

2
x
q,p
= 2(k/m)(1 cos q)x
q,p
= 4(k/m) sin
2
(q/2) (104)
Note that this result is independent of
q
. This phase will be at our disposal to match dierent
boundary conditions. To each index q, there corresponds the eigenvalue

2
q
=
2k
m
(1 cos q) =
4k
m
sin
2
(q/2). (105)
Its associated eigenvectors is
e
q
= C
q
_
_
_
_
_
_
_
_
sin(q +
q
)
sin(2q +
q
)
.
.
.
sin(Nq +
q
)
_
_
_
_
_
_
_
_
, (106)
where the coecient C
q
is chosen to normalize the eigenvector so that e

q
e
q
= 1. We will return
to this normalization and the demonstration that the eigenvectors are orthogonal shortly.
Now we have to deal with the boundary conditions. In the current problem, we require that
the left and right walls be stationary, i.e., that x
0
= 0 and x
N+1
= 0. To satisfy the rst condition,
we merely chose
q
= 0. The displacements are then proportional to sin qp, which clearly equals
zero when p = 0. To match the second boundary condition, we set
sin q(N + 1) = 0. (107)
This constraint quantizes the values of q:
q =
n
N + 1
; n = 1, 2, ..., N. (108)
Notes that n = 0 and n = N + 1 are not counted because sin qp = 0 for every p for these values.
There are nonetheless N independent modes as required.
14
Why dont we have to consider qs that are outside the range 0 to ? It is because qs outside
this range describe exactly the same situations as those for qs within this range. Consider rst the
frequency of Eq. (104), and let q = q, then
sin
q
2
= sin
_

2

q
2
_
= cos
q
2
(109)
and sin[(/2) + (q/2)] = sin[(/2) (q/2)]. For each q in the interval (, 2), there is a
companion q in the interval (0, ) that that has the same frequency. Similarly for each q in the
interval (, 2), there is a companion q in the interval (0, ) with the same eigenfunction modulo
a minus sign as can be seen from the relation,
sin( q)p = cos p sin q p = (1)
p
sin q p, (110)
or
sin( + q)p = sin( q)p. (111)
Now that we have the quantized values of q, we can demonstrate that the dierent eigenvectors
are orthogonal and complete, i.e., we need to show that
e

q
e
q
=
N

p=1
e
q,p
e

,p
=
q,q
(112)

q
e

q,p
e
q,p
=
p,p
. (113)
To show that the rst of these relations is true, we dene
J
q,q
=
N

p=1
sin qp sin q

p =
1
4
N

p=1
_
e
i(qq

)p
e
i(q+q

)p
+ c.c.
_
(114)
=
1
2
(F
qq
F
q+q
), (115)
where c.c. signies the complex conjugate and
F
k
= Re
N

p=1
e
ikp
= Re
1 e
ik(N+1)
1 e
ik
1 (116)
= Re
e
ik/2
(1 e
ik(N+1)
)
2i sin(k/2)
1 =
sin(k/2) + sin(N +
1
2
)k 2 sin(k/2)
2 sin(k/2)
(117)
=
sin(N +
1
2
)k sin(k/2)
2 sin(k/2)
=
sin(N + 1)k cos(k/2) cos(N + 1)k sin(k/2) sin(k/2)
2 sin(k/2
(118)
=
1
2
[cos(N + 1)k + 1] +
cos(k/2) sin(N + 1)k
2 sin(k/2)
(119)
=
1
2
(1 cos k(N + 1)) +
1
2
cos(k/2) sin k(N + 1)
2 sin(k/2
(120)
=
1
2
(1 + (1)
n
) + (N + 1)
k,0
(121)
15
where we used various trigonometric identities and the fact that k = n/(N +1) for some integer n,
so that cos k(N+1) = cos n = (1)
n
, sin k(N+1) = sin n = 0 and lim
k0
sin k(N+1)/ sin(k/2) =
2(N + 1) Thus,
J
q,q
=
1
2
_

1
2
(1 + (1)
nn

) + (N + 1)
qq

,0
+
1
2
(1 + (1)
n+n

) + (N + 1)
q+q

,0
_
=
1
2
(N+1)
q,q

(122)
where q = n/(N + 1) and q

= n

/(N + 1). The choice C


q
=
_
2/(N + 1) then provides the
orthonormal eigenvectors we need.
e
q
=
_
2
N + 1
_
_
_
_
_
_
_
_
sin q
sin 2q
.
.
.
sin Nq
_
_
_
_
_
_
_
_
. (123)
For N = 5, the ve eigenvectors { e
n
} are
e
1
=
_
_
_
_
_
_
_
1
2

3
1
2
1

3
1
2
1
2

3
_
_
_
_
_
_
_
, e
2
=
_
_
_
_
_
_
1
2
1
2
0
1
2
1
2
_
_
_
_
_
_
, e
3
=
_
_
_
_
_
_
_
1

3
0
1

3
0
1

3
_
_
_
_
_
_
_
, e
4
=
_
_
_
_
_
_
1
2

1
2
0
1
2

1
2
_
_
_
_
_
_
, e
5
=
_
_
_
_
_
_
_
1
2

1
2
1

3
1
2
1
2

3
_
_
_
_
_
_
_
(124)
You should verify directly that these vectors do indeed form an orthonormal set.
2.2 An example with initial conditions
We now have all of the apparatus necessary to solve initial value problems for N coupled oscillators
with the geometry of Fig. 3. Suppose we impose initial conditions that at t = 0, all velocities are
zero and that particle p is at position x
p,0
. Then we can write
x(t) =

q
A
q
e
q
cos
q
t (125)
where we can choose the cos solutions because we have set the velocities equal to zero. The positions
at t = 0 are, however, non-zero:
x(t = 0) = x
0
=
_
_
_
_
_
_
_
_
x
1,0
x
2,0
.
.
.
x
N,0
_
_
_
_
_
_
_
_
. (126)
We thus have
x
0
=

q
A
q
e
q
. (127)
Our goal is to determine the values of the amplitudes A
q
. Once we know them, we know the time
evolution for all times t > 0. Equation 127 is a set of N coupled equations, and because the e
q
16
k
m m m m m m
k k k k k k
m m m m m m
k k k k k
(a)
(c)
(b)
m m
m
m m
m
Figure 5: Schematic of (a) free, (b) hybrid, and (c) free boundary conditions.
form a linearly independent set, this equation has a solution. It is straightforward to solve using
the orthogonality relations of the basis vectors e
q
. Simply take the dot product of both sides of the
equation with a particular member of this basis e

s
:
e

s
x
0
=

q
A
q
e

s
e
q
=

q
A
q

q,s
= A
s
, (128)
and the problem is solved. For all time t > 0,
x(t) =

q
(x
0
e

q
) e
q
cos
q
t. (129)
2.2.1 Other Boundary Conditions
There are three other boundary conditions that are of physical interest: (1) The free chain with free
boundary conditions in which there are no terminal springs conning the particles. There are then
N particles connected by N 1 springs; (2) A chain with hybrid boundary conditions with one end
(say the left end) clamped to the wall but the nal particle with no spring to its right; (3) Periodic
boundary conditions in which particle N is constrained to have exactly the same displacement as
particle 0. This condition can be thought of as tying the string of particle in a circular loop and
with particle 0 attached to particle N 1. (Note there are n particle because counting starts at 0,
not at 1.)
The rst observation is that the equations for particles not at the boundaries are the same for
all of these conditions and for the one with xed boundaries we just considered, so they will all have
the same dependence of eigenfrequencies
q
on q. The solutions with dierent boundary conditions
will dier in the values of q that are allowed (quantization condition) and possibly on the value of
the phase
q
, which was zero in the case of xed boundary conditions.
2.2.2 Free Boundary Conditions
We label the rst particle as 0 and the Nth particle as N
1
. The equations of motion for these two
particles are
m x
0
= k(x
0
x
1
)
2
x
0
=
2
0
(x
0
x
1
) (130)
m x
N1
= k(x
N1
x
N2
)
2
x
N1
=
2
0
(x
N1
x
N2
), (131)
where
2
0
= k/m. We try a solution in which x
p
cos(qp+
q
). This form corresponds to changing
the
p
dened earlier to (/2) +
q
. Inserting this form into the two equations above yields:

2
q
cos
q
=
2
0
[cos
q
cos(q +
q
)] (132)

2
q
cos[q(N 1) +
q
] =
2
0
[cos[q(N 1) +
q
] cos[q(N 2) +
q
]. (133)
17
We solve the rst equation rst. With some trivial trigonometric manipulation and using Eq. (105),
we nd
(1 2 cos q) cos
q
= cos(q +
q
) = cos q cos
q
+ sin q sin
q
(134)
tan
q
=
1 cos q
sin q
= tan q/2. (135)
Thus,
q
= q/2. We now use this result in the second boundary equation:
0 =
2
q
cos q(N
1
2
) +
2
0
[cos q(N
1
2
) cos q(N
3
2
)] (136)
= (
2
q
+
2
0
) cos(N
1
2
)q
2
0
(cos q cos(N
1
2
)q + sin q sin(N
1
2
)q) (137)
=
2
0
[(1 cos q) cos(N
1
2
)q + sin q sin(N
1
2
)q sin q) (138)
=
2
0
[cos(N
1
2
)q cos(N +
1
2
)q] = 2
2
0
sin qN sin(q/2) = 0. (139)
The quantization constraints are, therefore,
q =
n
N
, n = 0, 1, ..., N 1 (140)
Note that q = 0 is include in the set. It must be there because we know that there must be a
pure translation mode with zero frequency. When q = 0,
q
= 0 as required, and in addition, the
amplitude at article p, cos(q(p+(1/2)) 1, is the same for every p as it is in a uniform translation.
2.2.3 Mixed boundary conditions
As shown in Fig. (5)(b) the left most particle (particle 1) is attached to a spring that is clamped in
a xed position, but the nal particle has no spring to its right. To satisfy the boundary condition
that the rst spring is clamped, we use the same ansatz as we used for the spring with xed
boundary conditions at both ends, i.e., x
q,p
= A
q,p
sin qpe
iqt
. The nal mass satises
m x
n
= k(x
N
x
N1
), (141)
implying

2
q
sin qN =
2
0
[sin qN sin q(N 1)] =
2
0
[sin qN sin qN cos q = cos Nq sin q] (142)
=
2
0
(1 cos q) sin qN +
2
0
cos Nq sin q. (143)
Using
2
q
=
2
0
(1 cos q), we obtain
0 = (1 cos q) sin Nq cos Nq sin q = 2 sin
2
(q/2) sin Nq 2 sin(q/2) cos(q/2) cos Nq (144)
0 = sin(q/2) sin Nq cos(q/2) cos Nq = cos(N +
1
2
)q, (145)
which implies
(N +
1
2
)q = (n +
1
2
) (146)
where k = 0, 1, N 1 is an integer. Thus
q =
(2n + 1)
2N + 1
. (147)
18
2.2.4 Periodic boundary conditions
There is another boundary condition that is used extensively that in some sense eliminates bound-
aries altogether. This is the periodic boundary condition in which displacements satisfy the peri-
odicity condition
x
p
= x
p+N
(148)
for some integer N for all positions p. You can think of this boundary condition as arising from
wrapping the line of springs around a ring so that the Nth particle and the zeroth one are actually
identical. The periodicity condition is trivially satised if
x
p
= A
q,p
e
ipq
= x
p+N
= A
q,p
e
iq(p+N)
(149)
provided
q =
2n
N
(150)
for n = 0, 1, (N 1) . Note that n = 0, which corresponds to q = 0 and to a zero-frequency
mode (
q=0
= 0). This is not surprising because q = 0 implies that all particles move the same
amount, e
iqp
= 1 when q = 0, implying that no springs are stretched or compressed.
For all of the other boundary conditions, q is restricted to lie between 0 and . For periodic
boundary conditions, q lies between 0 and 2. What is going on? It turns out that the eigenmodes
under periodic are traveling waves that can move either to the right or left whereas the eigenmodes
under the other boundary conditions are stationary. To see why this is, we rst note that periodic
eigenfunction satises e
iqp
= e
i(q+2)p
for every p, so rather than restricting q to lie between 0 and
2, we could equally well restrict it to lie between and . Or in equivalently restrict n to lie
between N/2 and N/2. Or course, we have to distinguish between systems with N odd and N
even. In the former case, n =
1
2
(N 1),
1
2
(N 3), , 1, 0, 1, ,
1
2
(N 1), and in the latter
case, n =
1
2
N + 1,
1
2
N + 2, , 1, 0, 1, ,
1
2
N 1,
1
2
N. Thus, q can be either positive or
negative. Now, consider the full time dependence of the displacements for a particular mode q:
x
q,p
(t) = Re(A
q
e
i(qpqt)
) = |A
q
| cos(qp
q
t
q
), (151)
where
q
is the phase of the complex amplitude A
q
. This function is a maximum whenever
qp
q
t
q
= 0; or p = p
0q
+

q
q
t. (152)
where p
0q
=
q
/q. At time t = 0, the maximum occurs at p = p
0
. Of course p
0q
is not in general
an integer, but the function x
q,p
(t) makes sense even when p and p
0q
are not integers. At time t
later the maximum that was originally at p = p
0
has moved to a new position p
0q
+
q
q
t. In other
words, the maximum move with a constant velocity v
q
=
q
t. The particles themselves do not
move with this velocity: they oscillate back and forth (or up and down) with frequency
q
. It is
only the peak in the amplitude that moves. When q > 0, the peak moves to the right; but it q < 0,
the peak positions is at
p = p
0q


q
|q|
t. (153)
Thus the peak that was originally at position p
0q
not moves to the left rather than to the right.
19
m m m m m m
k
a
k
a
k
a
k
a
k
b
k
b
k
b
1 2 2 2 1 1
Figure 6: A linear ball and spring system in with springs of alternating spring constants k
a
and k
b
.
There is a repeated unit cell (enclosed in the dashed box) with two balls, each of mass m.
It is instructive to return to the other boundary conditions. Their time-dependent displacements
in mode q are
x
q,p
(t) = Re
_
A
q
sin(qp +
q
)e
iqt

= B
q
sin(qp +
q
) cos(
q
t
q
)
= |A
q
|Re
_
1
2i
_
e
i(qp+q)
e
i(qp+q)
_
e
i(qtq)
_
=
1
2
|A
q
|[sin(qp
q
t +
q+
) + sin(qp +
q
t +
q
)] (154)
where
q,
=
q

q
. The rst sin function in the last equation describes a wave traveling to the
right, and the second to a wave traveling to the left: the standing waves are a linear combination
of right and left-going waves.
2.3 A periodic chain with alternating spring constants
In our study of vibrations of a small number of coupled oscillators, we considered situations in
which dierent springs and particles had dierent spring constants and masses. As the number
or particles grows, the number or arrangements of dierent springs and masses also grows, and
it is an almost impossible task to obtain general expressions for the eigenfrequencies that applies
to all systems. There are, however, special arrangements of dierent springs and masses that are
physically important. Solids composed of more that one type of atom, for example, have periodically
repeated unit cells composed of two or more dierent atoms. Here we consider the simplest model
of such a system: a ball-and-spring model connected by springs whose spring constants alternate
between k
a
and k
b
as shown in Fig. (6). This give us a repeated pattern of cells consisting of two
equal-mass balls. We label the cell by an integer p and the two balls within each cell by 1 ad 2.
The displacement from equilibrium of ball 1 in cell p is x
p,1
and that of ball 2 is x
p,2
.
The equations of motion for the two balls are
d
2
x
p,1
dt
2
=
2
a
(x
p,1
x
p1,2
)
2
b
(x
p,1
x
p,2
) (155)
d
2
x
p,2
dt
2
=
2
a
(x
p,2
x
p+1,1
)
2
b
(x
p,2
x
p,1
), (156)
where
2
a
= k
a
/m and
2
b
= k
b
/m. We consider periodic boundary conditions and take
x
p,
= A
q,
e
i(qpt)
, = q, 2, (157)
where to satisfy the periodic boundary conditions that x
p+Nc,
= x
p,
, where N
c
is the number of
cells, q = 2n/N
c
with n an integer. With this form, the equations of motion become

2
A
q,1
= (
2
a
+
2
b
)A
q,1

2
a
e
iq
A
q,2

2
b
A
q,2
(158)

2
A
q,2
= (
2
a
+
2
b
)A
q,2

2
a
e
iq
A
q,1

2
b
A
q,1
. (159)
20
-3 -2 -1 1 2 3
0.5
1.0
1.5
2.0
2.5
3.0
q
2
b

2
a

( ) q
Figure 7: The vibration (phonon) spectrum (frequency as a function of q) of a two-particle unit
cell one-dimensional ball-and-spring model. Note the bandgap at q = .
This equation can be rewritten in matrix form as

2
A
q
= DDD(q)A
q
, (160)
where
DDD =
_

2
a
+
2
b
(
2
b
+
2
a
e
iq
)
(
2
b
+
2
a
e
iq
)
2
a
+
2
b
_
. (161)
is the q-dependent dynamical matrix. Note that the o-diagonal entries are complex, but the 21
component is the complex conjugate of the 21 component. This means that the eigenvalues of DDD(q)
are real. To determine the, we proceed as usual and calculate the roots of the polynomial,
det[DDD(q)
2
III] = (
2
a
+
2
b

2
)
2
(
2
b
+
2
a
e
iq
)(
2
b
+
2
a
e
iq
) (162)
= (
2
a
+
2
b

2
)
2
(
4
a
+
4
b
+
2
a

2
b
cos q) (163)
The roots, which are plotted in Fig. (7), are

(q) =
2
a
+
2
b

_
(
2
a
+
2
b
)
2
4
2
a

2
b
sin
2
(q/2). (164)
The limits of these two solutions as q 0 and q are informative:


_

2
a

2
b
2(
2
a
+
2
a
)
q
2
as q 0
2
2
b
as q ;
(165)

2
+

_
2(
2
a
+
2
b
) as q 0
2
2
a
as q .
(166)
(167)
The solutions assume
a
>
b
; the two frequencies are interchanged is the opposite is true. Thus

(
q) tends linearly to zero with q, yielding a zero frequency modes at q = 0 as it most.
+
(q = 0) on
the other hand is nonzero. At q = the two frequencies are dierent:
+
() =

2
a
>

() =

2
b
. There is a gap in the spectrum. When
b

a
, the gap vanishes and the spectrum is
continuous. The existence of such band gaps, particularly in electronic as opposed to vibrational
properties, and being able to control them is essential to the electronics industry.
21

Anda mungkin juga menyukai