Linear Continuous Dynamical Systems and the Matrix Exponential

1. If M is a square matrix, we’ve defined the matrix exponential

M2 M3
eM = I + M + + + ··· .
2! 3!
On Daily Problem Set 23, #3, you found the matrix exponential of a diagonal matrix D. We want to
use your result to find the matrix exponential of any diagonalizable matrix.

(a) If M = SBS −1 where S is invertible, express eM in terms of S and B.

Solution. If M = SBS −1 , then
M2 M3
eM = I + M + + + ···
2! 3!
−1 2
(SBS ) (SBS −1 )3
= I + SBS −1 + + + ···
2! 3!

Since (SBS −1 )k = SB k S −1 , we can simplify this:

SB 2 S −1 SB 3 S −1
= I + SBS −1 + + + ···
2! 3!
B2 B3
 
=S I +B+ + + · · · S −1
2! 3!
= SeB S −1

 
1 3
(b) If M = , find eM .
−2 6

Solution. Let’s try to diagonalize M . After all, if we can write M = SDS −1 where S is invertible
and D is diagonal, then it will be easy to evaluate eD (using Daily Problem Set 23, #3), and then
we’ll know by (a) that eM = SeD S −1 .
To diagonalize M , we first find its eigenvalues and eigenvectors. Its characteristic polynomial is
λ2 − 7λ + 12 = (λ − 3)(λ − 4), so the eigenvalues are 3 and 4.
   
−2 3 3
The 3-eigenspace is ker(M − 3I) = ker = span . The 4-eigenspace is ker(M − 4I) =
−2 3 2
       
−3 3 1 3 1 3 0
ker = span . So, if we let S = and D = , then M = SDS −1 . Then,
−2 2 1 2 1 0 4
 3 
M D −1 D e 0
#1 says that e = Se S , and Daily Problem Set 23, #3, says that e is simply . So,
0 e4

eM = SeD S −1
−1
3 1 e3 0 3 1
  
=
2 1 0 e4 2 1
 3
3e − 2e4 −3e3 + 3e4

=
2e3 − 2e4 −2e3 + 3e4

1

c Harvard Math 21b
  
1 3
(c) If M = (the same as in (b)), find eM t .
−2 6
   
3 1 3 0
Solution. In (b), we already diagonalized M as M = SDS −1 with S = and D = .
2 1 0 4
Therefore, M t diagonalizes as S(Dt)S −1 , and

eM t = SeDt S −1
−1
3 1 e3t 0
  
3 1
=
2 1 0 e4t 2 1
 3t
3e − 2e4t −3e3t + 3e4t

=
2e3t − 2e4t −2e3t + 3e4t

 
d~x 1/2 0
2. Consider the continuous dynamical system = A~x, where A = .
dt −1/12 1/4
 
1
(a) Solve the system with the initial condition ~x(0) = ; express your answer in closed form.
0
d~x
Solution. In general, the solution of = A~x is always eAt ~x(0); to compute the matrix exponen-
dt
tial eAt in closed form, we want to diagonalize A. So, weshould first find the  eigenvalues and eigen-
1/2 − λ 0
= 21 − λ 41 − λ , so



vectors of A. The characteristic polynomial of A is det
−1/12 1/4 − λ
the eigenvalues of A are 12 , 14 .
   
0 0 3
The 12 -eigenspace is ker A − 21 I = ker . The 41 -eigenspace is


= span
    −1/12 −1/4 −1
1/4 0 0
ker A − 14 I = ker


= span .
−1/12 0 1
   
−1 3 0 1/2
So, we can write A = SDS where S = and D = . Then,
−1 1 1/4

~x(t) = eAt ~x(0)

= SeDt S −1 ~x(0)
3 0 et/2
     
1 1 0 1
=
−1 1 et/4 3 1 3 0
   t/2  
1 3 0 e 1
= t/4
3 −1 1 e 1
   t/2 
1 3 0 e
=
3 −1 1 et/4
3et/2
 
1
=
3 −et/2 + et/4

Note: If we want to sketch

 the trajectory,
  it’s more convenient in the last step to instead write
1 t/2 3 0
the solution as e + et/4 .
3 −1 1

2

c Harvard Math 21b
(b) Sketch the phase portrait of the system. Be sure to indicate all straight-line trajectories.
   
3 0
Solution. Let’s call the eigenvectors we found in #2 ~v1 = and ~v2 = . From Prep
−1 1
d~x
Video 23, the general solution of = A~x is ~x(t) = c1 et/2~v1 + c2 et/4~v2 . Let’s first think about
dt
the straight line trajectories:

• If c2 = 0, then the solution trajectory ~x(t) = c1 et/2~v1 is a straight line away from the origin
along the line span(~v1 ). (We can tell that it moves away from the origin because et/2 grows
as t increases.)

• Similarly, if c1 = 0, then the corresponding solution trajectory ~x(t) = c2 et/4~v2 is a straight

line away from the origin along the line span(~v2 ).

Finally, suppose we have a general trajectory ~x(t) = c1 et/2~v1 + c2 et/4~v2 where c1 , c2 6= 0. This
trajectory will curve; to understand how, let’s think about what happens when t is really large or
really negative:

• When t is very large, et/2 is much larger than et/4 , so ~x(t) ≈ c1 et/2~v1 . That is, as t increases,
~x(t) approaches the line span(~v1 ).

• When t is very negative, et/2 is much smaller than et/4 , so ~x(t) ≈ c2 et/4~v2 . So, when the
trajectory is near the origin, it almost lies along span(~v2 ).

That gives us the following phase portrait.

y
4

x
-4 4

-4

Note: From this picture, you can see clearly that the trajectories are almost on span(~v2 ) near the
origin, but it doesn’t really look like the trajectories eventually approach span(~v1 ). To see that,
we really have to zoom out on the picture:

3

c Harvard Math 21b
 y
400

x
-400 400

-400

 
d~x −4 −3
3. Which of the following is the phase portrait of = ~x?
dt 2 −11
y y y
4 4 4

3 3 3

2 2 2

1 1 1

x x x
-4 -3 -2 -1 1 2 3 4 -4 -3 -2 -1 1 2 3 4 -4 -3 -2 -1 1 2 3 4
-1 -1 -1

-2 -2 -2

-3 -3 -3

-4 -4 -4

(A) (B) (C)

 
−4 −3
Solution. The characteristic polynomial of is λ2 + 15λ + 50 = (λ + 5)(λ + 10), so the
2 −11
eigenvalues are −5 and −10. The eigenspaces are:
   
1 −3 3
E−5 = ker = span
2 −6 1
   
6 −3 1
E−10 = ker = span
2 −1 2
     
d~x −4 −3 −5t 3 −10t 1
From Prep Video 23, the general solution of = ~x is ~x(t) = c1 e + c2 e . In
dt 2 −11 1 2
particular:
 
−2t 3
• When c2 = 0, we have straight-line trajectories ~x(t) = c1 e ; these tend toward the origin
1
 
3
along the line span (because e−5t → 0 as t increases).
1
 
−10t 1
• When c1 = 0, we have straight-line trajectories ~x(t) = c2 e ; these tend toward the origin
  2
1
along the line span .
2

4

c Harvard Math 21b
 Therefore, the correct phase portrait is (B) .

4. (Weekly PSet 9, #4) Drug metabolism.(1) Lidocaine is a drug commonly used to treat irregular
heartbeat, particularly when a patient is experiencing a heart attack. Because the drug is absorbed
differently by the bloodstream and body tissues, a typical model of how lidocaine is metabolized looks
at two functions,

B(t) = concentration of lidocaine in the bloodstream (in ng/mL) at time t

T (t) = concentration of lidocaine in body tissues at time t

where t is measured in minutes. Based on experimental data, these functions roughly satisfy the
differential equations
dB
= −0.8B(t) + 0.11T (t)
dt
dT
= 0.56B(t) − 0.11T (t)
dt
Suppose a patient is given an intravenous dose of 8500 ng/mL of Lidocaine at time t = 0; because this
enters the patient’s bloodstream, we have initial conditions B(0) = 8500, T (0) = 0.

(a) Find B(t). Hint: One of the eigenvalues is −0.88; from this, you should be able to do everything
else by hand.

(b) In order for lidocaine to be effective in preventing irregular heartbeats, the concentration in the
bloodstream must be at least 1200 ng/mL. After 10 minutes, does this patient have an effective
concentration of lidocaine in his bloodstream? (You’ll need a calculator to answer this.)

(1) The information and data in this problem are referenced in the weekly problem set; the numbers in the problem are

approximated from the actual data to be more convenient.

5

c Harvard Math 21b