Mathematics 3F03 Advanced Dierential Equations, Fall 2012

INSTRUCTORS SOLUTIONS TO ASSIGNMENT 2

October 17, 2012 @ 23:04
1 Pages 5758, Exercise 2(i), (iii) and (v).
For each of the following systems of the form X

= AX,
(a) Find the eigenvalues and eigenvectors of A.
(b) Find the matrix T that puts A in canonical form.
(c) Find the general solution of both X

= AX and Y

= (T
1
AT)Y .
(d) Sketch the phase portraits of both systems.
(i) A =
_
0 1
1 0
_
(iii) A =
_
1 1
1 0
_
(v) A =
_
1 1
1 3
_
Solution: We obtain the eigenvalues of A by solving for in the characterisitic equation,
det (A I) = 0. For each eigenvalue , we nd associated eigenvectors by solving the linear
equation (AI)v = 0 for v R
2
. If the eigenvalues are real and distinct, as in parts (i) and
(v), then there are two linearly independent real eigenvectors, the Jordan canonical form of
A is J = diag(
1
,
2
), and the similarity transformation T that converts A to J has columns
equal to linearly independent eigenvectors of A. If the eigenvalues are complex (

= i,
, R, = 0), as in part (iii), then the real Jordan canonical form of A is
J =
_


_
and the similarity transformation T has columns equal to the real and imaginary parts of
either of the complex eigenvectors.
Carrying out the computations for the matrices in question yields the results summarized
in the table below. The type of phase portrait is determined by the eigenvalues in these cases:
with two real eigenvalues of opposite sign we have a saddle; complex eigenvalues with positive
real part yield a spiral source.
A Eigenvalues J T Type
(i)
_
0 1
1 0
_
1
_
1 0
0 1
_ _
1 1
1 1
_
saddle
(iii)
_
1 1
1 0
_
1
2
(1 i

3)
_
1
2

3
2

3
2
1
2
_
_

1
2

3
2
1 0
_
spiral source
(v)
_
1 1
1 3
_
1

3
_
1

3 0
0 1 +

3
_ _
2 +

3 2

3
1 1
_
saddle
1
Before going any further, it is important to check that the computations above are actually
correct. Recalling the inverse of a 2 2 invertible matrix T is
T
1
=
1
det T
_
d b
c a
_
,
we can easily check that the table above is correct by verifying in every case that
J = T
1
AT .
To obtain the general solutions, we use the important result (proved in class, but which
you should be able to prove yourself) that e
t
V
0
is a solution of X

= AX if V
0
is an
eigenvector of A associated with eigenvalue . This led us to the conclusion that in the case
of distinct real eigenvalues
1
and
2
, the general solution to X

= AX is
X(t) = c
1
e

1
t
V
1
+ c
2
e

2
t
V
2
,
where c
1
, c
2
R and V
1
, V
2
are (linearly independent) eigenvectors of A. The table above
therefore tells us everything we need in order to write down the solutions in the case of the
saddles (since the columns of T are linearly independent real eigenvectors of A). Thus, we
have
(i) X(t) = c
1
e
t
_
1
1
_
+ c
2
e
t
_
1
1
_
Y (t) = c
1
e
t
_
1
0
_
+ c
2
e
t
_
0
1
_
.
Note here that the only dierence between the form of the solution in original and canonical
coordinates is that in the latter the eigenvectors are the coordinate axes. Similarly, we have
(v) X(t) = c
1
e
(1+

3)t
_
2 +

3
1
_
+ c
2
e
(1

3)t
_
2

3
1
_
Y (t) = c
1
e
(1+

3)t
_
1
0
_
+ c
2
e
(1

3)t
_
0
1
_
.
In the case with complex eigenvalues, we have

=
1
2

3
2
and the general solution is
the complex function
Z(t) = c
1
e

+
t
V
+
+ c
2
e

t
V

, ()
where V

are complex eigenvectors associated with

,
V

=
_

1
2
i

3
2
1
_
=
_

1
2
1
_
i
_
3
2
1
_
.
2
Inserting this in (*), and using the standard notation [z] for the real part of z, we have
(iii) X(t) = [Z(t)]
=
_
c
1
e
(
1
2
+i

3
2
)t
__

1
2
1
_
+ i
_
3
2
1
__
+ c
2
e
(
1
2
i

3
2
)t
__

1
2
1
_
i
_
3
2
1
___
= e
t/2

_
c
1
e
i

3
2
t
__

1
2
1
_
+ i
_
3
2
1
__
+ c
2
e
i

3
2
t
__

1
2
1
_
i
_
3
2
1
___
= e
t/2

_
c
1
_
cos

3
2
t + i sin

3
2
t
_
__

1
2
1
_
+ i
_
3
2
1
__
+ c
2
_
cos

3
2
t + i sin

3
2
t
_
__

1
2
1
_
i
_
3
2
1
__
_
= e
t/2
_
(c
1
+ c
2
) cos

3
2
t
_

1
2
1
_
+ (c
1
+ c
2
) sin

3
2
t
_
3
2
1
_
_
= e
t/2
_
C
1
cos

3
2
t
_

1
2
1
_
+ C
2
sin

3
2
t
_
3
2
1
_
_
where we have renamed the arbitrary constants in the nal step. In canonical coordinates
we have
(iii) Y (t) = e
t/2
_
C
1
cos

3
2
t
_
1
0
_
+ C
2
sin

3
2
t
_
0
1
_
_
.
Phase portraits in original and canonical coordinates are shown below.
3
x
y
Original Coordinates
(i) x
y
Canonical Coordinates
x
y
(iii) x
y
x
y
(v) x
y
2 Page 58, Exercise 7.
Consider the system X

= AX, where
A =
_
1
0
_
,
with = 0. Show that all solutions tend to (respectively, away from) the origin tangentially
to the eigenvector (1, 0) when < 0 (respectively, > 0).
Solution: The matrix A is already in Jordan canonical form. The unique eigenvalue is
and the 1 above the main diagonal tells us that we are in the situation in which the repeated
eigenvalue has geometric multiplicity 1 (there is only one eigendirection).
4
Writing out the system X

= AX in coordinates, we have
x

= x + y (1a)
y

= y (1b)
The second equation is independent of the rst and can be solved immediately:
y(t) = y
0
e
t
, (2)
where y
0
= y(0). Given this expression for y(t), the rst equation can now be written as a
simple non-autonomous linear rst order ODE in one variable,
x

= x + y
0
e
t
, (1a

)
which is a type of equation studied ad nauseam in previous courses. One solution method
is to guess a solution of the form
x(t) = c
1
e
t
+ c
2
te
t
, (3)
and then solve for c
1
and c
2
by plugging this guess into the dierential equation. First note
that Equation (3) implies x(0) = c
1
, so we can write
x(t) = x
0
e
t
+ c
2
te
t
. (2

)
The derivative of this function is
x

(t) = x
0
e
t
+ c
2
te
t
+ c
2
e
t
(4a)
= [x
0
e
t
+ c
2
te
t
] + c
2
e
t
(4b)
= x(t) + c
2
e
t
. (4c)
Plugging this expression for x

into Equation (1a

) and simplifying, we nd
y(t) = c
2
e
t
,
and hence Equation (2) implies
c
2
= y
0
.
Thus, the general solution of X

= AX is
X(t) =
_
x(t)
y(t)
_
=
_
x
0
e
t
+ y
0
te
t
y
0
e
t
_
(5a)
= x
0
e
t
_
1
0
_
+ y
0
e
t
_
t
1
_
(5b)
If y
0
= 0 then y(t) = 0 t, so the solution starts and remains on the x-axis (the eigendirec-
tion); if < 0 then x(t) 0 whereas if > 0 then x(t) sign(x
0
) (and thus travels
away from origin along the eigendirection regardless of the sign of x
0
). If x
0
= 0 then we
can write the solution
X(t) = y
0
te
t
_
1
1/t
_
,
5
i.e., X(t) is proportional to the vector (1, 1/t), which approaches the eigenvector (1, 0) as t
. Moreover, the proportionality factor y
0
te
t
0 as t if < 0 and sign(y
0
)
if > 0. If x = 0 and y
0
= 0 then the solution is the sum of the two special cases just
considered, hence retains the properties to be established.
Although not required, the graph below shows the phase portrait of the system. The
direction of ow is not marked, but will be inwards if < 0 and outwards if > 0. This
is an example of special sink or special source in which there is only one invariant line
(the x-axis in this case).
x
y
3 Page 59, Exercise 9.
Determine a computable condition that guarantees that, if a matrix A has complex eigen-
values with nonzero imaginary parts, then solutions of X

= AX travel around the origin in

the counterclockwise direction. Note: Although not stated in the question, assume that A is
a 2 2 real matrix.
6
Solution: Recall the meaning of the RHS of a system of ODEs X

= F(X). F(X) is a
vector eld that points in the direction of motion at any point X. In our present situation,
F(X) = AX, with A a 2 2 matrix. Consider X = (1, 0)
T
and note that AX is then equal
to the rst column of A. So establishing the direction of rotation in the phase plane reduces
to the question of whether the rst column of A is a vector that points down (clockwise) or
up (counter-clockwise), which is determined by the sign of the bottom left entry of A:
a
21
< 0 = clockwise ,
a
21
> 0 = counter-clockwise .
Note that this implies that multiplying A by 1 (which does not change the eigenvalues)
will change the sense of rotation of the phase portrait. Consequently, the sense of rotation
in canonical coordinates is not necessarily the same as the sense of rotation in the original
coordinates.
4 Page 59, Exercise 10.
Consider the system X

= AX, where
A =
_
a b
c d
_
,
where a + d = 0 but ad bc = 0. Find the general solution of this system and sketch the
phase portrait.
Solution: The characteristic polynomial of the matrix A is
p() = det(A I)
=
2
(a + d) + (ad bc)
=
2
(a + d)
= [ (a + d)],
so the eigenvalues of A are
1
= 0 and
2
= a + d = 0. Since there are distinct real
eigenvalues, there is a basis of real eigenvectors; writing V

for an eigenvector associated

with eigenvalue , one eigenbasis is {V
0
, V
a+d
} where
V
0
=
_
b
a
_
, V
a+d
=
_
a
c
_
. ()
Indeed, using the fact that ad bc = 0, we have
AV
0
=
_
0
ad bc
_
=
_
0
0
_
= 0 V
0
,
AV
a+d
=
_
a
2
+ bc
c(a + d)
_
=
_
a
2
+ ad
c(a + d)
_
= (a + d)
_
a
c
_
= (a + d)V
a+d
.
7
All seems well, but we have made an implicit assumption in (*) that V
0
and V
a+d
are not
zero vectors. So a bit more care is actually required. If a = b = 0 then V
0
= (1, 0) is an
eigenvector of = 0. If a = c = 0 then V
a+d
= (b, d) is an eigenvector of = a +d (and note
that d = 0 here because we cannot have both a = 0 and d = 0 given that a + d = 0).
With the understanding that V
0
and V
a+d
are dened as above in the special cases where
some entries of A are zero, the general solution of X

= AX is
X(t) = c
1
V
0
+ c
2
e
(a+d)t
V
a+d
.
If a + d < 0 we have a repelling line, whereas if a + d > 0 we have an attracting line. See
the example phase portraits below.
Note that this question occurs in the textbook before the discussion of the trace-determinant
plane, but it is really asking us to determine the nature of solutions on the trace axis of the
trace-determinant plane (ignoring the origin).
x
y
A =

1 2
3 6

x
y
A =

1 2
3 6

END OF ASSIGNMENT
8