Chapter 3

Basic Floquet Theory

3.1 General Results

If we have a problem of the form
x = A(t)x (3.1)
where A(t) is periodic with period T , then x need not be periodic, however it
must be of the form
et p(t) (3.2)
where p(t) has period T . Additionally, it has n such j and together they satisfy
!
Z T
e1 T e2 T en T = exp tr (A(s)) ds . (3.3)
0

The following theorems prove those results. We follow Ward [28].

Definition (Fundamental Matrix). Let x1 (t), . . . , xn (t) be n solutions of x =
A(t)x. Let

X(t) = x1 xn (3.4)

so that X(t) is an n n matrix solution of X = AX.

If x1 (t), . . . , xn (t) are linearly independent, then X(t) is non-singular and is
called a fundamental matrix. If X(t0 ) = I, then X(t) is the principal fundamen-
tal matrix.
Lemma 3.1. If X(t) is a fundamental matrix then so is Y(t) = X(t)B for any
non-singular constant matrix B.
Proof. Since X(t) and B are non-singular then the inverse of Y(t) is B1 X1 (t)
and so Y(t) is non-singular. Also,
Y = X B = AXB = AY (3.5)

49
so that Y (t) = AY(t).
Lemma 3.2. Let the Wronskian W (t) of X(t) be the determinant of X(t).
Then Z t 
W (t) = W (t0 ) exp tr (A(s)) ds . (3.6)
t0

Proof. Let t0 be some time. Expanding in a Taylor series,

 
2
X(t) = X(t0 ) + (t t0 ) X (t0 ) + O (t t0 ) (3.7)
 
2
= X(t0 ) + (t t0 ) A(t0 )X(t0 ) + O (t t0 ) (3.8)
 
2
= [I + (t t0 ) A(t0 )] X(t0 ) + O (t t0 ) (3.9)

so that

det (X(t)) = det [I + (t t0 ) A(t0 )] det (X (t0 )) (3.10)

W (t) = det [I + (t t0 ) A(t0 )] W (t0 ). (3.11)

Now since
det (I + C) = 1 + tr (C) + O 2 ,


(3.12)
we have that
W (t) = W (t0 ) (1 + (t t0 ) tr (A(t0 ))) . (3.13)
Now by expanding W (t) in a Taylor series, we obtain that
 
2
W (t) = W (t0 ) + (t t0 ) W (t0 ) + O (t t0 ) (3.14)

so that

W (t0 ) = W (t0 ) tr (A(t0 )) . (3.15)

Since we have not made any assumptions about t0 , we can the write

W (t) = W (t) tr (A(t)) . (3.16)

We know that the solution to this equation is

Z t 
W (t) = W (t0 ) exp tr (A(s)) ds (3.17)
t0

Theorem 3.3. Let A(t) be a T -periodic matrix. If X(t) is a fundamental

matrix then so is X(t + T ) and there exists a non-singular constant matrix B
such that
i. X(t + T ) = X(t)B for all t
R 
T
ii. det (B) = exp 0 tr (A(s)) ds

50
Proof. Begin by showing that X(t + T ) is also a fundamental matrix. Let
Y(t) = X(t + T ). Then
Y (t) = X (t + T ) = A(t + T )X(t + T ) = A(t)X(t + T ) = A(t)Y(t) (3.18)
and so X(t + T ) is a fundamental matrix.
i. Let B(t) = X1 (t)Y(t). Then
Y(t) = X(t)X1 (t)Y(t) (3.19)
= X(t)B(t) (3.20)
Let B0 = B(t0 ). We know by lemma 3.1 that Y0 (t) = X(t)B0 is a fun-
damental matrix, where, by definition, Y0 (t0 ) = Y(t0 ). Since these are
both solutions to X = AX, by the uniqueness of the solution, we must
then have Y0 (t) = Y(t) for all time. As a result, B0 = B(t) and so B is
time-independent.
ii. From lemma 3.2, we have that
Z t 
W (t) = W (t0 ) exp tr (A(s)) ds (3.21)
t0
!
Z t Z t+T
W (t + T ) = W (t0 ) exp tr (A(s)) ds + tr (A(s)) ds (3.22)
t0 t
!
Z t+T
W (t + T ) = W (t) exp tr (A(s)) ds (3.23)
t
!
Z T
W (t + T ) = W (t) exp tr (A(s)) ds . (3.24)
0

We also know that

X(t + T ) = X(t)B (3.25)
det (X(t + T )) = det (X(t)) det (B) (3.26)
W (t + T ) = W (t) det (B) (3.27)
and so !
Z T
det (B) = exp tr (A(s)) ds (3.28)
0

Remark. Since B is time-independent, it can be computed by setting t = 0,

so that B = X1 (0)X(T ). If we took the initial conditions X(0) = I, then
B = X(T ).
Definition (Characteristic Multipliers and Exponents). The eigenvalues 1 , . . . , n
of B are called the characteristic multipliers for X (t) = A(t)X(t). The char-
acteristic exponents or Floquet exponents are 1 , . . . , n satisfying
1 = e1 T , 2 = e2 T , ... n = en T . (3.29)
Note that j for j N may be complex.

51
Properties.
i. The characteristic multipliers (eigenvalues) 1 , . . . , n of B = X(T ) with
X(0) = I satisfy
!
Z T
det (B) = 1 2 n = exp tr (A(s)) ds . (3.30)
0

This follows from theorem 3.3ii.

ii. Since the trace is the sum of the eigenvalues, we also have

tr (B) = 1 + 2 + + n . (3.31)

iii. The characteristic exponents are not unique since if j = ej T , then j =

e(j +2i/T )T .
iv. The characteristic multipliers j are an intrinsic property of the equation
X (t) = AX and do not depend on the choice of the fundamental matrix.

Proof. Suppose X(t) is another fundamental matrix. Then
+ T ) = X(t)
X(t B. (3.32)

We have showed in the proof of theorem 3.3 that since X(t) and X(t) are
fundamental matrices then there is a constant non-singular matrix C such
that

X(t) = X(t)C (3.33)
so that
+ T ) = X(t + T )C
X(t (3.34)
 
B
X(t) = (X(t)B) C (3.35)
= X(t)BC
X(t)CB (3.36)
= BC
CB (3.37)
1 = B
CBC (3.38)

are the same.

so the eigenvalues of B and B

Theorem 3.4. Let be a characteristic multiplier and let be the corresponding

characteristic exponent so that = eT . Then there exists a solution x(t) of
x = A(t)x such that
i. x(t + T ) = x(t)
ii. There exists a periodic solution p(t) with period T such that x(t) = et p(t).

52
Proof.
i. Let b be an eigenvector of B corresponding to eigenvalue . Let x(t) =
X(t)b. Then x = Ax and

x(t + T ) = X(t + T )b (3.39)

= X(t)Bb (3.40)
= X(t)b (3.41)
= x(t) (3.42)

so that x(t + T ) = x(t).

ii. Let p(t) = x(t)et . We now need to show that p(t) is T -periodic.

p(t + T ) = x(t + T )e(t+T ) (3.43)

(t+T )
= x(t)e (3.44)

= T x(t)et (3.45)
e
= x(t)et (3.46)
= p(t) (3.47)

As a result, we have a solution of the form x(t) = et p(t) where p(t) is

periodic with period T .
Remarks.
i. If is replaced by + 2i/T , then we get

x(t) = et p(t)e2it/T (3.48)

where p(t)e2it/T is still periodic with period T . As a result, the fact that
is not unique does not alter our results.
ii. We have that

xj (t + T ) = j xj (t) (3.49)
xj (t + N T ) = N
j xj (t). (3.50)

Each characteristic multipliers falls into one of the following categories:

t
(a) If || < 1, then Re () < 0 and so x(t) 0.
(b) If || = 1, then Re () = 0 and so we have a pseudo-periodic solution.
If = 1, then the solution is periodic with period T .
(c) If || > 1, then Re () > 0 and so x(t) as t .
The entire solution is stable if all the characteristic multipliers satisfy |j |
1.

53
iii. As for the general solution, suppose that b1 , . . . , bn are n linearly inde-
pendent eigenvectors of B corresponding to distinct eigenvalues 1 , . . . , n .
Then there are n linearly independent solutions to x = Ax, which by the
above theorem are given by

xj (t) = ej t pj (t) (3.51)

where pj (t) is T -periodic. As a result, we can define

X0 (t) = x1 xn , P0 (t) = p1 pn , (3.52)

e1 t

1 0 0
D0 (t) =
.. ,

Y0 (t) =
.. ,

(3.53)
. .
n t
0 n 0 e

such that

X0 = P0 Y0 , Y0 = D0 Y0 . (3.54)

iv. Now consider what happens if < 0. Suppose < 0 real, so that we can
write
= e(+i/T )T (3.55)
where
= eT . (3.56)
Then we obtain

x(t) = et p(t) (3.57)

t it/T
=e e p(t) (3.58)
t
= e q(t), (3.59)

where q(t) has period T since p(t) has period T . Since we can choose x
to be real, without loss of generality, we can also choose q to be real. For
the general solution, if j < 0, we can replace pj with qj and j with j so
that
1 T
e 0
..

.

j T
P0 = p1 qj pn , Y0 = e (3.60)


. ..


n T
0 e

and
X0 (t) = P0 (t)Y0 (t). (3.61)

54
 v. Suppose now that is complex. Then since is an eigenvalue of the real
matrix B, is as well. The characteristic exponents are and . Let
= + i, p(t) = q(t) + ir(t) (3.62)
where q(t) and r(t) must both have period T since p(t) does. Since x(t) =
et p(t) is a solution to x = A(t)x, then by taking the complex conjugate,
so is x(t) = et p(t). We can write these as
x(t) = e(+i)t (q(t) + ir(t)) (3.63)
t
= e [(q cos (t) r sin (t)) + i (r cos (t) + q sin (t))] (3.64)
and
x(t) = e(i)t (q(t) ir(t)) (3.65)
t
= e [(q cos (t) r sin (t)) i (r cos (t) + q sin (t))] . (3.66)
We can alternately write the linearly independent real solutions
xR = Re et p(t) = et [cos (t) q(t) sin (t) r(t)] ,
 
(3.67)
xI = Im et p(t) = et [sin (t) q(t) + cos (t) r(t)] ,
 
(3.68)
so that

X0 = x1 xR xI xn , (3.69)

P0 = p1 qr pn , (3.70)

e1 T

0
..

.

et cos (t) et sin (t)
Y0 = (3.71)

et sin (t) et cos (t)

..
.
0 en T
and
X0 (t) = P0 (t)Y0 (t). (3.72)

3.1.1 Example
For example, consider
 
cos (t)
x1 = 1+ x1 (3.73)
2 + sin (t)
x2 = x1 x2 . (3.74)

55
Here, we know that the solution is in general

x1 = c1 et (2 + sin (t)) (3.75)

 
1 1
x2 = c1 et 2 + sin (t) cos (t) + c2 et (3.76)
2 2

which we can write as

   
2 + sin (t) 0
x = c1 et + c2 et . (3.77)
2 + 12 sin (t) 12 cos (t) 1

Using all the above definitions, the fundamental matrix is

et (2 + sin (t))
 
0
X(t) = (3.78)
et 2 + 21 sin (t) 12 cos (t) et

so that

B = X1 (0)X(2) (3.79)
1 
2e2
 
2 0 0
= 3 3 2 (3.80)
1 e e2
2  2 2 
1 1 0 2e 0
= (3.81)
2 3 2 3 2
e e2
 2 2  2
e 0
= (3.82)
0 e2

As a result 1 = e2 , 2 = e2 and so 1 = 1 and 2 = 1. Theorem 3.4 then

tells us that there is a solution of the form

x1 (t) = et p1 (t), x2 (t) = et p2 (t) (3.83)

where p1 (t) and p2 (t) are periodic with period 2. We know that in fact
   
2 + sin (t) 0
p1 (t) = , p2 (t) = . (3.84)
2 + 21 sin (t) 12 cos (t) 1

3.1.2 Periodic Solution

Consider a problem of the form x = f (x) with x Rn where there is a periodic
solution x(t) = (t) with period T . Linearise the solution about by writing
x = + v. We then obtain
v = A(t)v (3.85)

fi 
where A(t) is the Jacobian of f (so Aij (t) = x j
 ). Since (t) has period
(t)
T , so does A(t). Now let X(t) be the principal fundamental matrix of v = Av
(so that X(0) = I). Then B = X(T ).

56
 Now by definition,
(t) = f ((t)) (3.86)
so

fi 
(t) = (t) (3.87)
xj (t)
(t) = A(t) (t) (3.88)

If we let v = , then
v (t) = A(t)v(t) (3.89)
where, since (t) has period T by assumption, v(t) must also, and so the corre-
sponding characteristic multiplier is 1. As a result, for a nonlinear system with
a periodic solution, one characteristic multiplier is always = 1.

3.2 General Results for n = 2

3.2.1 Stability of Periodic Solution
Consider a problem of the form x = f (x) with x R2 where there is a periodic
solution x(t) = (t) with period T . We know from 3.1.2 that we must have
1 = 1 and we know from theorem 3.3ii that
Z T !
1 2 = exp tr (A(s)) ds (3.90)
0
!
Z T
2 = exp tr (A(s)) ds . (3.91)
0

From remark (ii) on page 53, we know that for the perturbation to be bounded
and hence for the solution to be stable, we must have 1 1 and 2 1 and
so, since we know 1 = 1 and we wish 1 and 2 to be distinct, we must have
Z T
0> tr (A(s)) ds (3.92)
0
Z T  !
fi 
0> tr ds (3.93)
0 xj (s)
Z T  
f1 f2 
0> + ds (3.94)
0 x1 x2 (s)
Z T
0> f |x= ds. (3.95)
0

We get instability when

Z T
0< f |x= ds. (3.96)
0

57
3.2.2 Example
Consider

x = x y x x2 + y 2


(3.97)
2 2


y =x+yy x +y . (3.98)

Let

x = r(t) cos ((t)) (3.99)

y = r(t) sin ((t)) (3.100)

so that our problem becomes

sin () (r r ) = cos () r r3 r


(3.101)
3


cos () (r r ) = sin () r r r . (3.102)

By squaring and adding these equations, we obtain that

2
2
(r r ) = r r3 r (3.103)

so we can write

a = r r (3.104)
3
sa = r r r (3.105)

where s = 1. Our equations then become

a sin () = sa cos () (3.106)

a cos () = sa sin () (3.107)

which can be rewritten as

a sin () = sa cos () (3.108)

2
s a sin () = sa cos () (3.109)

so that we must have

a sin () = a sin () (3.110)

a sin () = 0. (3.111)

As a result, we have that

a sin () = sa cos () = 0 (3.112)

so that we must have a = 0. This means that

r r = r r3 r = 0. (3.113)

58
We have that

r = r 1 r3


(3.114)

and so we have a solution of constant radius when r = 0 (the trivial case) and
r = 1. Without loss of generality, choose r = 1. Then since

r = r, (3.115)

we have that = 1, so = t + C. As a result, our solution has period T = 2.

Now
 
f1 f2
f |r=1 = + (3.116)
x y r=1
= 1 3x y 2 + 1 x2 3y 2 r=1
2



(3.117)
= 2 4r2 r=1
 
(3.118)
= 2 (3.119)

so that
!
Z T
2 = exp tr (A(s)) ds (3.120)
0
Z 2 
= exp 2 ds (3.121)
0
= e4 (3.122)
< 1. (3.123)

As a result, the limit cycle with radius r = 1 is stable.

3.2.3 Stability of Second-Order ODE

Consider the second-order ODE

x + a(t)x = 0 (3.124)

where a(t) is periodic with period T . Letting x1 = x and x2 = x1 , this can be

rewritten as     
x1 0 1 x1
= (3.125)
x2 a(t) 0 x2
By choosing the initial condition
   
x1 (0) 1
= (3.126)
x2 (0) 0
we obtain a solution of the form
" #
(1)
x1 (t)
(1) . (3.127)
x1 (t)

59
Likewise by choosing the initial condition
   
x1 (0) 0
= (3.128)
x2 (0) 1
we obtain a solution of the form
" #
(2)
x1 (t)
(2) . (3.129)
x1 (t)
As a result, we have chosen X(0) = I so that
" #
(1) (2)
x1 (T ) x1 (T )
B = X(T ) = (1) (2) . (3.130)
x1 (T ) x1 (T )
Now we have from property (i) on page 52 that
Z T !
1 2 = exp tr (A(s)) ds (3.131)
0
!
Z T
= exp 0 ds (3.132)
0

=1 (3.133)
and from property (ii) that
1 + 2 = tr (B) (3.134)
(1) (2)
= x1 (T ) + x1 (T ). (3.135)
Let = tr (B) /2 so that
1 2 = 1 (3.136)
1 + 2 = 2. (3.137)
Solving these, we obtain that
p
= 2 1. (3.138)
We can rewrite i as exp(i T ), so that
1 + 2 = 0 (3.139)
and so
e1 T + e2 T = 2 (3.140)
e1 T + e1 T = 2 (3.141)
1 T 1 T
e +e
= (3.142)
2
cosh (1 T ) = . (3.143)
Consider the following cases.

60
 I. Let 1 < < 1. We can then define by = cos (T ), where, without
loss of generality, 0 < T < , so that
p
= 2 1 (3.144)
= cos (T ) i sin (T ) (3.145)
iT
=e (3.146)

As in remark (v) on page 55, we can write the general solution as

x(t) = c1 Re eit p(t) + c2 Im eit p(t)

(3.147)

and since |1 | = 1 and |2 | = 1, then from remark (ii) on page 53, the
solution is stable and pseudo-periodic.
Now eit has period T = 2 . Now since 6= 1 and 6= 1, we must have

T =6 m (3.148)
2
T =6 m (3.149)
T
2T
6 T
= (3.150)
m

so that T 6= 2T, T, 23 T, . . .
Note that for T to equal nT , we must have
2
= (3.151)
nT
for n 6= 1, 2 from above.
p
II. Let > 1. Then since = 2 1, we must have 1 > 1 and since
1 2 = 1, we must have 1 > 1 > 2 > 0 and 2 = 11 means 2 = 1 .
Our solution must therefore be of the form

x(t) = c1 e1 t p1 (t) + c2 e1 t p2 (t) (3.152)

where p1 (t) and p2 (t) are both periodic with period T . As a result, the
solution is unstable.

III. Let = 1. Then 1 = 2 = 1. Here, theorem 3.4 only guarantees that

we will have one solution x(t) of the form et p(t). If B has two linearly
independent eigenvectors, we can find two linearly independent p1 (t) and
p2 (t) so that the two solutions are both in the standard form. However, if
B only has one eigenvector, we will end up with one solution of the form
p1 (t) (since = 1 in this case) and the other of the form tp1 (t) + p2 (t).
To see this, we replace  
1 0
(3.153)
0 2

61
 with the Jordan block  
1
. (3.154)
0
As a result, instead of our solution being of the form
  
1 0
X(t) = P(t) exp t (3.155)
0 2
 t 
e 1 0
= P(t) (3.156)
0 e2 t
P1 e1 t P2 e2 t
 
= , (3.157)
P3 e1 t P4 e2 t

it will be of the form

  
1
X(t) = P(t) exp t (3.158)
0
 t
tet

e
= P(t) (3.159)
0 et
P1 et P1 tet + P2 et
 
= . (3.160)
P3 et P3 tet + P4 et

See the papers by Akhmedov [1] and Wiesel and Pohlen [30].
p
IV. Let < 1. Since = 2 1, we must have 1 < 1 and since
1 2 = 1, we must have 1 < 1 < 2 < 0 and 2 = 11 means 2 = 1 .
Now we can write 1 = iT + so that our solution must be of the form

x(t) = c1 et eit/T p1 (t) + c2 et eit/T p2 (t) (3.161)

where p1 (t) and p2 (t) are both periodic with period T and so eit/T p1 (t)
and eit/T p2 (t) are both periodic with period 2T . As a result, the solution
is unstable.

V. Let = 1. Then 1 = 2 = 1. As in the case when = 1, we have one

solution which is periodic (this time with period 2T ),

x1 (t) = eit/T p1 (t) (3.162)

and the other which grows linearly with time,

x2 (t) = teit/T p1 (t) + eit/T p2 (t). (3.163)

We summarise these results in figure 3.1. For > 1, we have an unstable

solution of the form

x(t) = c1 e1 t p1 (t) + c2 e1 t p2 (t). (3.164)

62
 1.5

0.5
Im()

0.5

1.5
4 3 2 1 0 1 2 3 4
Re()
>1
=1
1<<1
=1
<1

Figure 3.1: The range of 1 , 2 for different values of real. In the region > 1,
the sample point has = 1/3, 3; for = 1, we have = 1. In 1 < < 1, the
sample point shown is = 2/3 i 5/3; for = 1, we have = 1 and in the
region < 1, we show = 1/3, 3.

For = 1, we have an unstable solution of the form

x(t) = (c1 + tc2 ) p1 (t) + c2 p2 (t). (3.165)
For 1 < < 1, we have a stable pseudo-periodic solution of the form
x(t) = c1 Re eit p(t) + c2 Im eit p(t) .



(3.166)
For = 1, we have an unstable solution of the form
x(t) = (c1 + tc2 ) q1 (t) + c2 q2 (t). (3.167)
Finally, for < 1, we have an unstable solution of the form
x(t) = c1 et q1 (t) + c2 et q2 (t) (3.168)
where pi (t) represents a function that has period T and qi (t) represents a func-
tion that has period 2T .

3.2.4 Application to Hills Equation

Consider Hills equation
x + ( + b(t)) = 0 (3.169)

63
where b(t) has period T . If = 0, the solution is stable, however, there are some
values of for which the solution is only marginally stable, according to the
above criteria. As a result, we expect that for small but nonzero near those
values of , we will get the beginning of a region of instability. We wish to find
those values of .
For = 0, if X(0) = I, then
   
cos t 1 sin t

X(t) =    (3.170)
sin t cos t

and so
   
cos T 1 sin T

B = X(T ) =     . (3.171)
sin T cos T

As a result,
tr (B)  
= = cos T . (3.172)
2
If = 1, then

T = 2m (3.173)
 2
= 2m (3.174)
T

where m is a positive integer since > 0. If = 1, then

= (2m + 1) (3.175)
 2
= (2m + 1) . (3.176)
T
Now we have from the previous section that = 1 corresponds to the existence
of a periodic solution of period T and = 1 corresponds to the existence of
a periodic solution of period 2T . As a result, we will have the border between
stability and instability breaking off from = 0 at
 2
= 2m (3.177)
T
corresponding to solutions with period T and breaking off from = 0 at
 2
= (2m + 1) (3.178)
T
corresponding to solutions with period 2T .

64
3.3 Stability Boundary of Mathieus Equation
3.3.1 Undamped Case
We have from 3.2.3 and 3.2.4 that on the edge of the region of stability, we
have either = 1 or = 1. The former corresponds to the existence of a
periodic solution with period T and the latter to a periodic solution with period
2T . In order to determine the region of stability of the Mathieu equation in
the - plane, we then need to determine the conditions on and required in
order to have a solution which is periodic with either period or 2. We follow
McLachlan [17] and Ward [28].

Functions of Period
We can write a general function of period as

X
X
x= an cos (2nt) + bn sin (2nt) . (3.179)
n=0 n=1

We then obtain

0 = x + ( + cos (2t)) x (3.180)

X
X
4n2 an cos (2nt) + 4n2 bn sin (2nt)



0=
n=0 n=1

X
X
+ an cos (2nt) cos (2t) + bn sin (2nt) cos (2t) . (3.181)
n=0 n=1

Using the identities

1
cos (A) cos (B) = (cos (A B) + cos (A + B)) (3.182)
2
1
sin (A) cos (B) = (sin (A B) + sin (A + B)) (3.183)
2
this becomes

X
X
4n2 an cos (2nt) + 4n2 bn sin (2nt)



0=
n=0 n=1

X
+ an (cos (2(n + 1)t) + cos (2(n 1)t))
2 n=0

X
+ bn (sin (2(n + 1)t) + sin (2(n 1)t)) (3.184)
2 n=1

65
and so we must have

X
4n2 an cos (2nt)


0=
n=0

X
+ an (cos (2(n + 1)t) + cos (2(n 1)t)) (3.185)
2 n=0
   
0 = a0 + a1 cos (0) + ( 4) a1 + (2a0 + a2 ) cos (2t)
2 2

X 
4n2 an + (an1 + an+1 ) cos (2nt)


+ (3.186)
n=2
2

and

X
4n2 bn sin (2nt)


0=
n=1

X
+ bn (sin (2(n + 1)t) + sin (2(n 1)t)) (3.187)
2 n=1
 
0 = ( 4) b1 + b2 sin (2t)
2

X 
4n2 bn + (bn1 + bn+1 ) sin (2nt) .


+ (3.188)
n=2
2

By orthogonality of the sine and cosine, these can be rewritten as

2 0
0 a0
2

0 41 2 a1


0 2 a2
4 2

= (3.189)

2 2

0 2 a3

..

2 4 3 2 ..

. .

.. .. ..
0 . . .
and


4 12 0
0 2 b1

4 22

0 2 2
b2
= (3.190)

0
4 32 b3

..

2 2
..

. .. .. .. .
0 . . .
In order to have a non-zero solution, the determinant of at least one of these
(infinite) matrices must be zero. This gives us the requirement that and
must satisfy in order to be on the borderline between stability and instability.
We can approximate the determinants of these matrices by the determinants of
the finite n n matrices of the same form. The resultant curves in the - plane
for different values of n are shown in figure 3.2.

66
 60

50

40

30

20

10

0
5 0 1 4 5 9 10 1516 20

n=3
n=5
n = 20

Figure 3.2: The approximation to the border of the region of stability of the
Mathieu equation (determined by equations 3.189, 3.190, 3.198, 3.199) where
each infinite matrix is approximated by its n n counterpart.

67
Functions of Period 2
We now perform a similar analysis for functions of period 2. We can write a
general function of period 2 as

X
X
x= an cos (nt) + bn sin (nt) . (3.191)
n=0 n=1

We then remove from this all the terms which also have period since we have
already dealt with those. If we included them, we would obtain the lines in the
- plane where we obtain solutions that either have period or have period
2. As a result, we have

X
X
x= an cos (nt) + bn sin (nt) . (3.192)
n=1 n=1
n odd n odd

so that we obtain

0 = x + ( + cos (2t)) x (3.193)

X
X
n2 an cos (nt) + n2 bn sin (nt)



0=
n=1 n=1
n odd n odd

X
X
+ an cos (nt) cos (2t) + bn sin (nt) cos (2t) (3.194)
n=1 n=1
n odd n odd

X
X
n2 an cos (nt) + n2 bn sin (nt)



0=
n=1 n=1
n odd n odd

X
+ an (cos ((n + 2)t) + cos ((n 2)t))
2 n=1
n odd

X
+ bn (sin ((n + 2)t) + sin ((n 2)t)) . (3.195)
2 n=1
n odd

We must then have


X 
0= ( 1) a1 + (a1 + a3 ) cos (t)
n=1
2
n odd

X 
n2 an + (an2 + an+2 ) cos (nt)


+ (3.196)
n=1
2
n odd

68
and

X 
0= ( 1) b1 + (b1 + b3 ) sin (t)
n=1
2
n odd

X 
n2 bn + (bn2 + bn+2 ) sin (nt)


+ (3.197)
n=1
2
n odd

which we can write as

12 + 0
0 2 2 a1

32

0 2 2
a3
0 = (3.198)


52 a5

..

2 2
..

. .. .. .. .
0 . . .

and


12 0
0 2 2 b1

32

0 2 2
b3
= (3.199)

0
52 b5

..

2 2
..

. .. .. .. .
0 . . .

As before, in order to obtain a nonzero solution, we must have the determinant

of at least one of the matrices being zero. This constrains and .
The resultant region of stability is shown in figure 3.3

3.3.2 Undamped Case with small

Consider now when is small. We have from 3.2.4 that for small, we will
have the border between stability and instability near
2
= (2m) (3.200)

and
2
= (2m + 1) . (3.201)
2
As a result, we seek periodic solutions near = n to the equation

x + ( + cos (2t)) x = 0. (3.202)

Let

x = x0 (t) + x1 (t) + 2 x2 (t) + . . . , (3.203)

2 2
= n + 1 + 2 + . . . (3.204)

69
 60

50

40

30

20

10

0
5 0 1 4 5 9 10 1516 20

Figure 3.3: The region of stability of the Mathieu equation.

Substituting these into Mathieus equation, we obtain

x0 + n2 x0 = 0 (3.205)
2
x1 + n x1 = 1 x0 x0 cos (2t) (3.206)
x2 + n2 x2 = 1 x1 2 x0 x1 cos (2t) . (3.207)

For n 6= 0, the solution to equation 3.205 is

x0 = a cos (nt) + b sin (nt) . (3.208)

Inserting this into equation 3.206, we obtain

x1 + n2 x1 = 1 x0 x0 cos (2t) (3.209)

= 1 (a cos (nt) + b sin (nt))
(a cos (nt) + b sin (nt)) cos (2t) (3.210)
= 1 a cos (nt) 1 b sin (nt)
a a
cos ((n + 2) t) cos ((n 2) t)
2 2
b b
sin ((n + 2) t) sin ((n 2) t) (3.211)
2 2
Under the assumption that n 6= 1, in order to eliminate secular terms, we must
have

1 a = 0, 1 b = 0. (3.212)

70
As a result, in order to avoid x0 being the zero solution, we must have 1 = 0.
We then have
a a
x1 + n2 x1 = cos ((n + 2) t) cos ((n 2) t)
2 2
b b
sin ((n + 2) t) sin ((n 2) t) . (3.213)
2 2
Letting

X
x1 = ci sin (it) + di cos (it) , (3.214)
i=0

this becomes

X
X
ci i2 sin (it) + di i2 cos (it) + ci n2 n2 sin (it) + di n2 cos (it)
i=1 i=0
a a
= cos ((n + 2) t) cos ((n 2) t)
2 2
b b
sin ((n + 2) t) sin ((n 2) t) . (3.215)
2 2
Equating coefficients of the sines and cosines, we obtain that
b a
cn2 = , dn2 = (3.216)
8 (n + 1) 8 (n + 1)
b a
cn+2 = , dn+2 = . (3.217)
8 (n + 1) 8 (n + 1)

We can assume that all the sin (nt) and cos (nt) component is already in x0 , so
we can choose cn = 0, dn = 0. All remaining ci and di are zero. As a result,
b b
x1 = sin ((n 2) t) + sin ((n + 2) t)
8 (n + 1) 8 (n + 1)
a a
+ cos ((n 2) t) + cos ((n + 2) t) . (3.218)
8 (n + 1) 8 (n + 1)

Finally, inserting this into equation 3.207, we obtain that

x2 + n2 x2 = 2 (a cos (nt) + b sin (nt))

b
(sin (nt) + sin ((n 4) t))
16 (n + 1)
b
(sin ((n + 4) t) + sin (nt))
16 (n + 1)
a
(cos (nt) + cos ((n 4) t))
16 (n + 1)
a
(cos ((n + 4) t) + cos (nt)) . (3.219)
16 (n + 1)

71
Under the assumption that n 6= 2, in order to eliminate the secular terms, we
must have
a a
0 = 2 a , (3.220)
16 (n + 1) 16 (n + 1)
b b
0 = 2 b , (3.221)
16 (n + 1) 16 (n + 1)

which can be rewritten as

 
1
0 = a 2 , (3.222)
8 (n2 1)
 
1
0 = b 2 . (3.223)
8 (n2 1)

As a result, in order to avoid a nonzero x0 (i.e., making sure that we dont

simultaneously have a = 0 and b = 0), we must have
1
2 = . (3.224)
8 (n2 1)

Case n = 2
In the case n = 2, eliminating the secular terms in equation 3.219 tells us that
a a
0 = 2 a + , (3.225)
8 48
b
0 = 2 b 0 , (3.226)
48
which become
 
5
0 = a 2 , (3.227)
48
 
1
0 = b 2 + . (3.228)
48

As a result, for n = 2 we must have either

1
a = 0, 2 = (3.229)
48
or
5
b = 0, 2 = . (3.230)
48
As a result, for n = 2, we either have
1
= 4 2 + O 3


(3.231)
48

72
or
5
= 4 + 2 + O 3 .


(3.232)
48
We also have either
b
sin (4t) + O 2


x = b sin (2t) + (3.233)
24
or  a a 
cos (4t) + O 2 ,


x = a cos (2t) + + (3.234)
8 24
which both have period , as expected.

Case n = 1
In the case n = 1, eliminating the secular terms in equation 3.211 tells us that
a
0 = 1 a , (3.235)
2
b
0 = 1 b + (3.236)
2
and so we must either have
1
1 = , b=0 (3.237)
2
or
1
1 = , a = 0. (3.238)
2
In either of these cases, equation 3.211 becomes
a b
x1 + x1 = cos (3t) sin (3t) . (3.239)
2 2
As before, we let

X
x1 = ci sin (it) + di cos (it) (3.240)
i=1

and find that

b a
c3 = , d3 = . (3.241)
16 16
As a result,
b a
x1 = sin (3t) + cos (3t) . (3.242)
16 16

73
Then equation 3.207 becomes
 
b a
x2 + x2 = 1 sin (3t) + cos (3t)
16 16
2 (a cos (t) + b sin (t))
 
b a
sin (3t) + cos (3t) cos (2t) (3.243)
16 16
 
b a
= 1 sin (3t) + cos (3t) 2 a cos (t) 2 b sin (t)
16 16
b b a a
sin (t) sin (5t) cos (t) cos (5t) . (3.244)
32 32 32 32
In order to eliminate the secular terms, we must have
 
1
0 = a 2 + (3.245)
32
 
1
0 = b 2 + . (3.246)
32

As a result, 2 = 1/32, so that either

1 1
= 1 2 + O 3


(3.247)
2 32
or
1 1
= 1 + 2 + O 3 .


(3.248)
2 32
We also have either
a
cos(3t) + O 2


x = a cos(t) + (3.249)
16
or
b
sin(3t) + O 2 ,


x = b sin(t) + (3.250)
16
which are periodic with period 2, as expected.

Case n = 0
In the case n = 0, we get
x0 = a + bt. (3.251)
Now we expect a periodic solution, so b = 0. As a result, equation 3.206 becomes

x1 = 1 a a cos (2t) . (3.252)

In analogy with before, when we eliminated secular terms, we must have 1 = 0.

As a result, we have
x1 = a cos (2t) (3.253)

74
 20

15

10

0
5 0 5 10 15 20

Figure 3.4: The quadratic approximations to the boundary between stability

and instability of the Mathieu equation, in comparison with the approximation
from 3.3.1, with n = 20.

so that
a
x1 = cos (2t) (3.254)
4
and equation 3.207 becomes
a
x2 = 2 a cos (2t) cos (2t) (3.255)
4
a a
= 2 a cos (4t) (3.256)
8 8
so that we must have  
1
0 = a 2 + . (3.257)
8
so that 2 = 1/8 and
1
= 0 2 (3.258)
8
with
a
x = a + cos (2t) + O 2 ,


(3.259)
4
which is again periodic with period , as expected.
These approximations to () for small are compared to the approximation
in the previous section (which is valid for both small and large ) in figure 3.4.

75
3.3.3 Damped Case
We follow Richards [24]. Our equation is
x + kx + ( + cos (2t)) x = 0. (3.260)
If we let
k
y(t) = e 2 t x(t), (3.261)
we obtain that
y + (a + cos (2t)) y = 0 (3.262)
where
k2
a= . (3.263)
4
Now equation 3.260 isnt of the form of equation 3.124 (3.2.3), but equation
3.262 is. As a result, we know that the solution to equation 3.262 is of the form
y(t) = e1 t p1 (t) + e2 t p2 (t) (3.264)
where 1 and 2 satisfy
p
e = = 2 1 (3.265)
where is half of the trace of B for y(t) above when we use the initial conditions
X(0) = I. As a result, the largest (the one most likely to cause instability)
satisfies p
e = = + 2 1 (3.266)
so that
 p 
= ln + 2 1 (3.267)
= cosh1 () (3.268)
1
cosh ()
= . (3.269)

Now in order for x(t) to be stable, we must have
 
k
0 Re (3.270)
2
k
Re () (3.271)
2
with as above. This can be used to numerically determine the stability of the
damped equation. The result for k = 0.2 is shown in figure 3.5.

3.3.4 Damped Case with small

Consider the damped Mathieu equation
x + kx + ( + cos (2t)) x = 0. (3.272)

76
 20
k=0
k = 0.2

15

10

0
5 0 5 10 15 20

Figure 3.5: The border of the region of stability of the Mathieu equation, in the
damped case.

Near = 1
Suppose that k is of order . Then we can write k = k1 and expand near = 1,

= 1 + 1 + . . . (3.273)
x = x0 + x1 + . . . (3.274)

Plugging this in and equating terms of equal order, we obtain

x0 + x0 = 0 (3.275)
x1 + x1 = k1 x0 cos (2t) x0 x0 . (3.276)

This tells us that

x0 = a cos (t) + b sin (t) (3.277)
so that

x1 + x1 = k1 (a sin (t) + b cos (t)) cos (2t) (a cos (t) + b sin (t))
1 (a cos (t) + b sin (t)) (3.278)
a
= k1 a sin (t) k1 b cos (t) (cos (t) + cos (3t))
2
b
( sin (t) + sin (3t)) 1 (a cos (t) + b sin (t)) (3.279)
2

77
In order to eliminate secular terms, we must have
b
1 b = 0
k1 a + (3.280)
2
a
k1 b 1 a = 0 (3.281)
2
which can be written as

1
k1 1
   
2 a 0
= . (3.282)
1
1 k1 b 0
2

In order for this to have a nonzero solution, the determinant of the matrix must
be zero, so we must have
  
1 1
0 = k12 + + 1 1 (3.283)
2 2
1
0 = k12 + 12 (3.284)
r 4
1
1 = k12 (3.285)
4
so that

= 1 + 1 + O 2


(3.286)
r
2
k 2 + O 2 .


=1 (3.287)
4

Near = 4
For larger values of , in order to still be small at the edge of stability, we
must have k quite a bit smaller. As a result, near = 4, we choose k to be of
order 2 . Then we can write k = 2 k1 and expand near = 4,

= 4 + 1 + 2 2 + . . . (3.288)
2
x = x0 + x1 + x2 + . . . (3.289)

We need to expand these to order 2 because it will turn out that 1 = 0.

Plugging this in and equating terms of equal order, we obtain

x0 + 4x0 = 0 (3.290)
x1 + 4x1 = 1 x0 cos (2t) x0 (3.291)
x2 + 4x2 = k1 x0 1 x1 2 x0 cos (2t) x1 . (3.292)

This tells us that

x0 = a cos (2t) + b sin (2t) (3.293)

78
so that

x1 + 4x1 = 1 x0 cos (2r) x0 (3.294)

= 1 (a cos (2t) + b sin (2t))
cos (2t) (a cos (2t) + b sin (2t)) (3.295)
= 1 a cos (2t) 1 b sin (2t)
a a b b
cos (4t) sin (4t) 0 (3.296)
2 2 2 2
In order to eliminate secular terms, we must have

1 a = 0, 1 b = 0 (3.297)

so we must have 1 = 0. As a result, we have

a a b
x1 + 4x1 = cos (4t) sin (4t) . (3.298)
2 2 2
Expanding x1 in terms of sines and cosines and equating coefficients, we find
that
a a b
x1 = + cos (4t) + sin (4t) . (3.299)
8 24 24
As a result, we have that

x2 + 4x2 = k1 x0 1 x1 2 x0 cos (2t) x1 (3.300)

= k1 (2a sin (2t) + 2b cos (2t)) 0
2 (a cos (2t) + b sin (2t))
 
a a b
cos (2t) + cos (4t) + sin (4t) (3.301)
8 24 24
 
b
= 2k1 a 2 b sin (2t)
48
 a a
+ 2k1 b 2 a + cos (2t)
8 48
a b
cos (6t) sin (6t) . (3.302)
48 48
In order to eliminate secular terms, we must have
b
0 = 2k1 a 2 b (3.303)
48
5b
0 = 2k1 b 2 a + (3.304)
48
which can be written as

1
2k1 2
   
48 a 0
= . (3.305)
2 + 5
2k1 b 0
48

79
 20

15

10

0
5 0 5 10 15 20

Figure 3.6: The approximation to the boundary between stability and instability
of the Mathieu equation, in comparison with the numerical result from 3.3.3,
with k = 0.2.

In order to have a nonzero solution to this, we must have that the determinant
of the matrix is zero. As a result,
  
2 1 5
0 = 4k1 2 + 2 (3.306)
48 48
q
1 1 5 2


12 144 4 482 + 4k1
2 = . (3.307)
2
where

= 1 + 2 2 + O 3 .


(3.308)

These approximations are compared to the result from 3.3.3 in figure 3.6.

3.3.5 Hills Equation

Consider Hills equation, which is a generalised version of the Mathieu equation

x + ( + b(t)) x = 0 (3.309)

where b is periodic with period . Let us assume that

Z
b(t) dt = 0 (3.310)
0

80
and that we can expand b(t) as

X
b(t) = cn cos (2nt) + dn sin (2nt) . (3.311)
n=1

We wish to determine an expansion for the solution where is small. Now we

know this occurs near = m2 for positive integers m, so we expand
= m2 + 1 + . . . (3.312)
x = x0 + x1 + . . . (3.313)
Then we obtain that
x0 + m2 x0 = 0 (3.314)
2
x1 + m x1 = 1 x0 b(t)x0 (3.315)
so that
x0 = a cos (mt) + b sin (mt) (3.316)
and
x1 + m2 x1 = 1 x0 b(t)x0 (3.317)
= 1 (a cos (mt) + b sin (mt)) (a cos (mt) + b sin (mt))

X
(cn cos (2nt) + dn sin (2nt)) (3.318)
n=1
= 1 a cos (mt) 1 b sin (mt)

X acn
+ (cos ((2n + m) t) + cos ((2n m) t))
n=1
2
adn
(sin ((2n + m) t) + sin ((2n m) t))
2
bcn
(sin ((2n + m) t) sin ((2n m) t))
2 
bdn
( cos ((2n + m) t) + cos ((2n m) t)) (3.319)
2
To eliminate secular terms, if m = 0, we must have 1 a = 1 b = 0, and so
1 = 0. As a result for m = 0, we must expand everything to second order. We
will return to this later. For m 6= 0, we must have
acm bdm
0 = 1 a (3.320)
2 2
adm bcm
0 = 1 b + (3.321)
2 2
which we can rewrite as

1 c2m dm    
2 a 0
= . (3.322)
d2m 1 + cm b 0
2

81
As a result, we must have
1 2
12 = cm + d2m


(3.323)
4
and so
p 2
= m2 cm + d2m . (3.324)
2

Case m = 0
Recall that we determined that in the m = 0 case, we must expand everything
to second order. As a result, we expand

= 1 + 2 2 + . . . (3.325)
2
x = x0 + x1 + x2 + . . . (3.326)

From before, plugging in m = 0, we have x0 = a and 1 = 0, so that

X
x1 = 1 a cn a cos (2nt) + dn a sin (2nt) (3.327)
n=1

X
x1 = cn a cos (2nt) + dn a sin (2nt) (3.328)
n=1

X cn a dn a
x1 = 2
cos (2nt) + 2 sin (2nt) . (3.329)
n=1
4n 4n

The second-order equation gives us

x2 = 1 x2 2 x0 b(t)x1 (3.330)

!
X
= 2 a ci cos (2it) + di sin (2it)
i=1


X cj a dj a
2
cos (2jt) + 2 sin (2jt) (3.331)
j=1
4j 4j

In order to eliminate the secular-like terms, we must have

2
X c a i d2i a
0 = 2 a + (3.332)
i=1
8i2 8i2

1 X c2i + d2i
2 = (3.333)
8 i=1 i2

so that
2
21
X c + d2
i i
= . (3.334)
8 i=1
i2

82
3.4 Applications of Mathieus Equation
3.4.1 Pendulum with Oscillating Pivot
Suppose we have a mass m attached at the end of a massless pendulum of length
L. Suppose the pivot point P oscillates in the vertical direction according to
some function p(t). Then the angle from the vertical to the pendulum obeys

g + p (t)
 

+ sin () = 0. (3.335)
L

We choose to measure the angle such that when the pendulum is vertical,
pointed upward (at what is usually the unstable stationary solution), = .
When the pendulum is near the top, . Let x = so that |x| 1.
Then our model is approximately

g + p (t)
 
x + (x) = 0. (3.336)
L

Let p(t) = A cos (t) to obtain

g + A 2 cos (t)
 
x + x = 0. (3.337)
L

Now let 2 = t so that

 
4g 4A
x
+ 2 + cos (2 ) x = 0. (3.338)
L L

We can finally let

4g 4A
= , = (3.339)
2 L L
to obtain
x
+ ( + cos (2 )) x = 0 (3.340)
where will be small if the amplitude of oscillations of the pivot is small com-
pared to the length of the pendulum.
We wish to determine an and , and hence an A and , such that the
solution to the above equation (Mathieus equation) is stable for x small. Notice
that the usual problem (A = 0 so = 0) is unstable; near x = 0 the solution
grows exponentially in time.

3.4.2 Variable Length Pendulum

Consider now a pendulum with an oscillatory length. This time, the pendulum
is pointed downward.

83
Derivation of Model
Suppose that there is some force F on the mass along the pendulum. Then the
forces on the mass at the end of the pendulum are given by

mx = F sin () (3.341)
my = F cos () mg, (3.342)

where

x = L sin () (3.343)
y = L cos () . (3.344)

By letting z = x + iy = iLei , we obtain

z = 2L + L iL + iL2 ei


(3.345)

so that
m 2L + L + iL2 iL = iF imgei .


(3.346)
By equating real parts, we then obtain

2L + L + g sin () = 0. (3.347)

Letting = L, this becomes

L
 

+ g sin = 0. (3.348)
L L
For 1, this is approximately
g L
 
+ = 0. (3.349)
L

Transformation to Mathieus Equation

Let
L = L0 (1 + cos (t)) (3.350)
for 1. Then we obtain
g L
 
0 = + (3.351)
L
L0 2 cos (t)
 
g
0 = + (3.352)
L0 (1 + cos (t)) L0 (1 + cos (t))
 
g
0 = + (1 cos (t)) + 2 cos (t) (3.353)
L0
   
g g
0 = + + 2 cos (t) . (3.354)
L0 L0

84

Figure 3.7: The physical ion trap, for z0 = 1, r0 = 2.

Letting 2 = g/L0 , this becomes

+ 2 + 2 2 cos (t) = 0.



(3.355)

Letting

42 2
 

= t, = 2 , = 4 1 2 (3.356)
2

this becomes Mathieus equation:

+ ( + cos (2 )) = 0. (3.357)

3.4.3 Ion Traps

As in the honours thesis by Fischer [8], we consider an ion trap as shown in
figure 3.7. The side walls are described by

r2 = 2z 2 + r02 (3.358)

where r0 is the radius at the narrowest point. The end caps are described by

r2
z2 = + z02 (3.359)
2

85
where 2z0 is the shortest distance between the two end caps.
Now if we apply a potential difference A between the side walls and the end
caps, taking the end caps to be ground, we obtain a potential of

r2 2 z 2 z02


V (z, r) = A (3.360)
r02 + 2z02

and hence an electric field of

A
E = V = (2rr + 4z
z) . (3.361)
r02 + 2z02

As a result, in the z-direction, we have

4QA
mz = z (3.362)
d20

where prime denotes differentiation with respect to t and we have let d20 =
r02 + 2z02 . If
A = U0 V0 cos (t) , (3.363)
as in the thesis of King [15], our problem then becomes

4Q
z (U0 V0 cos (t)) z. (3.364)
md20

Following King [15], we can then make the substitutions

16QU0 16QV0
= t, = , = (3.365)
2 md20 2 md20 2

so that our equation once more takes the familiar form of Mathieus equation:

z + ( + cos (2 )) z = 0. (3.366)

Stability for U0 = 0
In the case that U0 = 0, our equation becomes
4QV0
mz = cos (t) z. (3.367)
d20

We follow King [15]. We assume that the solution is composed of two parts:
one which has large amplitude and small acceleration, the other which has small
amplitude but large acceleration (something small but quickly oscillating). We
approximate z = zM + z so that we can approximate our equation by

4QV0
mz = cos (t) zM (3.368)
d20

86
so that
4QV0
z cos (t) zM . (3.369)
md20 2
As a result, we obtain
4QV0
mz = cos (t) z (3.370)
d20
4QV0
zM + z = cos (t) (zM + z ) (3.371)
md20
4QV0 4QV0 16Q2 V 2

zM 2 cos (t) zM = 2 cos (t) zM 2 4 02 cos2 (t) zM (3.372)
md0 md0 m d0

Averaging over one period, this becomes

8Q2 V02
zM = zM , (3.373)
m2 d40 2

which is a harmonic oscillator with frequency

2 2QV0
. (3.374)
md20

As a result, for U0 = 0, the ion trap acts like a harmonic oscillator, trapping
the ion at its centre.
See King [15] and Brewer et al. [4] for further reference.
A physical analogy to the trap is shown in figure 3.8. If one constantly
rotates the base at the correct frequency, the ball will be not roll down the base
[25, 27].

87
Figure 3.8: A physical analogy to the ion trap.

88