Sie sind auf Seite 1von 26

Solutions to the Exercises on

Differential Equations

Laurenz Wiskott
Institut fur Neuroinformatik
Ruhr-Universitat Bochum, Germany, EU

4 February 2017

Contents

1 Basic definitions and properties 2

1.1 Classification of differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.1 Exercise: Classification of differential equations . . . . . . . . . . . . . . . . . . . . . . 2

1.1.2 Exercise: Classification of differential equations . . . . . . . . . . . . . . . . . . . . . . 3

1.1.3 Exercise: Verify solutions of differential equations . . . . . . . . . . . . . . . . . . . . . 3

1.2 Stability of fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Linear differential equations with constant coefficients 4

2.1 Single linear differential equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 System of linear differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2.1 Exercise: ODEs in standard form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2.2 Exercise: ODEs in standard form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.3 Superposition principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.4 Relation to iterated maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.5 Hermitean systems (real eigenvalues) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5


2017 Laurenz Wiskott (homepage https://www.ini.rub.de/PEOPLE/wiskott/). This work (except for all figures from
other sources, if present) is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License. To view
a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/. Figures from other sources have their own
copyright, which is generally indicated. Do not distribute parts of these lecture notes showing figures with non-free copyrights
(here usually figures I have the rights to publish but you dont, like my own published figures).
Several of my exercises (not necessarily on this topic) were inspired by papers and textbooks by other authors. Unfortunately,
I did not document that well, because initially I did not intend to make the exercises publicly available, and now I cannot trace
it back anymore. So I cannot give as much credit as I would like to. The concrete versions of the exercises are certainly my
own work, though.
These exercises complement my corresponding lecture notes available at https://www.ini.rub.de/PEOPLE/wiskott/

Teaching/Material/, where you can also find other teaching material such as programming exercises. The table of contents of
the lecture notes is reproduced here to give an orientation when the exercises can be reasonably solved. For best learning effect
I recommend to first seriously try to solve the exercises yourself before looking into the solutions.

1
2.6 Nonhermitean systems (complex eigenvalues) . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.7 Nonhermitean systems (multiple eigenvalues with too few eigenvectors) . . . . . . . . . . . . 5

2.7.1 Exercise: Inhomogeneous ODE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.7.2 Exercise: Inhomogeneous ODE for a transformed solution . . . . . . . . . . . . . . . . 6

2.7.3 Exercise: System of linear differential equations . . . . . . . . . . . . . . . . . . . . . . 7

2.7.4 Exercise: System of linear differential equations . . . . . . . . . . . . . . . . . . . . . . 8

2.7.5 Exercise: System of linear differential equations with too few eigenvectors . . . . . . . 9

2.7.6 Exercise: Damped oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.7.7 Exercise: Metabolism of a pharmaceutical substance . . . . . . . . . . . . . . . . . . . 15

2.7.8 Exercise: Metabolism of a pharmaceutical substance revisited . . . . . . . . . . . . . . 18

3 Nonlinear differential equations 21

3.1 Fixed points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Qualitative analysis in 2D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3 Linear stability analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3.1 Exercise: Linearisation around a Fixpoint . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3.2 Exercise: Pitchfork bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3.3 Exercise: Qualitative analysis of a nonlinear ODE in 2D - parabola . . . . . . . . . . . 21

3.3.4 Exercise: Qualitative analysis the of a nonlinear ODE in 2D - FitzHugh-Nagumo model 22

1 Basic definitions and properties

1.1 Classification of differential equations

1.1.1 Exercise: Classification of differential equations

Classify the following differential equations. Determine their order and decide whether they are ordinary or
partial, linear, homogeneous, and/or autonomous.

(a) y = y . (1)

Solution: This differential equation is first order, ordinary, linear, homogeneous, and autonomous.

(b) y + 2 y + 2 y = f (t) . (2)

Solution: This differential equation is second order, ordinary, linear, inhomogeneous, and not au-
tonomous.

(c) 1 2 h(x, t) 2 h(x, t)


+ = f (x, t) . (3)
c2 t2 x2
Solution: This differential equation is second order, partial, linear, inhomogeneous, and not au-
tonomous.

2
(d) (y + y)(y 2) = 0 . (4)

Solution: This differential equation is second order, ordinary, nonlinear, homogeneous (if one wants
to make such a statement at all, since this is a nonlinear differential equation), and autonomous.

1.1.2 Exercise: Classification of differential equations

Classify the following differential equations. Determine their order and decide whether they are ordinary or
partial, linear, homogeneous, and/or autonomous.
(a)
y + 4 = 0 (1)

Solution: This differential equation is first order, ordinary, linear, inhomogeneous, and autonomous.
(b)
(1 + y 2 )y + ty + y = et (2)

Solution: This differential equation is second order, ordinary, nonlinear, inhomogeneous (if one wants
to make such a statement at all, since this is a nonlinear differential equation), and not autonomous.
(c)
g(x, y) g(x, y)
2 = 0 (3)
x y
Solution: This differential equation is first order, partial, linear, homogeneous, and autonomous.
(d)
dy(x)
= x. (4)
dx
Solution: This differential equation is first order, ordinary, linear, inhomogeneous, and not au-
tonomous.

1.1.3 Exercise: Verify solutions of differential equations

Show that the given functions yi are indeed solutions to the given differential equations.

(a) y 4 y + 3 y = 0 , y1 (t) = et , y2 (t) = e3t . (1)

Solution: With

y1 (t) = et , y1 (t) = et , (2)


3t 3t
y2 (t) = 3 e , y2 (t) = 9 e , (3)

we find

y1 4 y1 + 3 y1 = et 4 et + 3 et = (1 4 + 3) et = 0 , (4)
t t t t
y2 4 y2 + 3 y2 = 9 e 12 e + 3 e = (9 12 + 3) e = 0 . (5)

Thus y1 and y2 are solutions of the ODE.



(b) 2x2 y 00 + 3xy 0 y = 0 , y1 (x) = x, y2 (x) = x1 . (6)

Solution: With

y10 (t) = (1/2) x1/2 , y100 (t) = (1/4) x3/2 , (7)


y20 (t) = x 2
, y200 (t) = 2x 3
, (8)

3
we find

2x2 y100 + 3xy10 y1 = 2x2 (1/4) x3/2 + 3x(1/2) x1/2 x1/2 (9)
1/2 1/2 1/2
= (1/2)x + (3/2)x x (10)
= (1/2 + 3/2 1)x1/2 = 0 , (11)
2x2 y200 + 3xy20 y2 = 2
2x 2 x 3
+ 3x (x 2
)x 1
(12)
1 1 1
= 4x 3x x (13)
1
= (4 3 1)x = 0. (14)

Thus y1 and y2 are solutions of the ODE.

1.2 Stability of fixed points

2 Linear differential equations with constant coefficients

2.1 Single linear differential equation

2.2 System of linear differential equations

2.2.1 Exercise: ODEs in standard form

Write the following higher-order differential equations in standard form, i.e. as a system of first-order differ-
ential equations with exactly one derivative without prefactor per equation.

(a) y y 6 y = 0 . (1)

Solution: With y0 := y and y1 := y the ODE above is equivalent to

y0 = y1 (2)
y1 = 6 y0 + y1 . (3)

(b) d3 y d2 y 1 dy
3
+3 2 +2 = 0. (4)
dt dt 2 dt

Solution: With y0 := y, y1 := y, and y2 := y the ODE above is equivalent to

y0 = y1 (5)
y1 = y2 (6)
y2 = 3 y2 + y1 /2 2 . (7)

(c) y1 = sin(t y1 ) + 4 y2 (8)


y2 = y1 3 y12 . (9)

Solution: With y10 := y1 , y11 := y1 , y20 := y2 , and y21 := y2 the ODE above is equivalent to

y10 = y11 (10)


y20 = y21 (11)
y11 = sin(t y10 ) + 4 y21 (12)
2
y21 = y11 3 y10 . (13)

4
2.2.2 Exercise: ODEs in standard form

Write the following higher-order differential equations in standard form, i.e. as a system of first-order differ-
ential equations with exactly one derivative without prefactor per equation.

...
(a) 3 z 2z + z = 3
 1
(b) 2 x 3y = x + y
6y + x = 2x 1

x + 2y = x + 2
(c)
2y + 4x = 2x + y

Solution: Not available!

2.3 Superposition principle

Spezielle Losung des inhomogenen Systems

Allgemeine Losung des homogenen Systems

Allgemeine Losung des inhomogenen Systems

2.4 Relation to iterated maps

2.5 Hermitean systems (real eigenvalues)

2.6 Nonhermitean systems (complex eigenvalues)

2.7 Nonhermitean systems (multiple eigenvalues with too few eigenvectors)

2.7.1 Exercise: Inhomogeneous ODE

(a) Find the general solution of the ordinary differential equation

y = y + cos(t) (1)

with positive .
Hint: At some point it may help to make the ansatz y(t) = s sin(t) + c cos(t).
Solution: The differential equation is inhomogeneous. In order to find the general solution we must
find one special solution and the general solution of the corresponding homogeneous differential equa-
tion.
Finding the latter is easy, since we already know that

y = y (2)

has the general solution


y = a exp(t/ ) , (3)
as can be easily verified. This solution decays exponentially to y = 0.

5
One would expect that the inhomogeneous term cos(t) somehow keeps the system oscillating at fre-
quency 1. This motivates the ansatz

y(t) = s sin(t) + c cos(t) (4)


= y(t) = s cos(t) c sin(t) . (5)

for a special solution of the inhomogeneous system. With this ansatz we find
(1)
y = y + cos(t) (6)
(5,4)
(s cos(t) c sin(t)) = (s sin(t) + c cos(t)) + cos(t) (7)
0 = ( c s) sin(t) + ( s c + 1) cos(t) (8)
0 = c s (9)
s = c (10)
(15) 2
= /(1 + ) (11)
0 = s c + 1 (12)
(10)
= 2 c c + 1 (13)
2
= (1 + )c + 1 (14)
2
c = 1/(1 + ) (15)
sin(t) + cos(t)
y(t) = . (16)
1 + 2

Thus, the general solution is

sin(t) + cos(t)
y(t) = + a exp(t/ ) (17)
1 + 2
with the single free parameter a.
(b) Discuss how the solution depends on .
Solution: The exponential part obviously decays with time constant , i.e. the smaller the faster
this part decays.
Now consider the induced oscillation (16). For 0 the special solution of the inhomogeneous ODE
becomes y(t) = cos(t), thus the function simply follows the forcing function cos(t). As gets larger,
the sin(t) part kicks in and finally dominates the cos(t) so that the phase difference to the external
force cos(t) gets larger and larger until it is /2 = 90 . At the same time the amplitude becomes
smaller and smaller due to the 2 in the denominator.
As an intuitive picture consider a ping-pong ball in honey that is fixed in a frame by two elastic springs.
The frame oscillates back and forth like cos(t). If the springs are very stiff, i.e. = 0 is very small, the
ping-pong ball follows the frame precisely. If the springs are weak, the ball gets draged through the
honey with some lag and the amplitude is smaller than that of the frame. If the springs disappear, i.e.
, the ball does not move at all.

2.7.2 Exercise: Inhomogeneous ODE for a transformed solution

Assume a solution y(t) is known for the inhomogeneous ODE

y(t) = (t)y(t) + f (t) . (1)

Find a differential equation for which the scaled version y(t) would be a solution.

6
dy(t)
Solution: First we write the ODE with dt instead of y(t) for greater clarity and then we scale the whole
equation

dy(t)
= (t)y(t) + f (t) (2)
dt
dy(t0 )
= (t0 )y(t0 ) + f (t0 ) (3)
d(t0 )
=1
z }| {
dy(t ) d(t0 ) dt0
0
= (4)
d(t0 ) dt0 d(t0 )
dy(t0 ) 1
= 0
(5)
dt
y(t0 ) = (t0 )y(t0 ) + f (t0 ) . (6)

Thus, y(t) is a solution to the ODE (6).

2.7.3 Exercise: System of linear differential equations

Find the solutions of the following system of linear differential equations and sketch their trajectories.

x = 3x 2y (1)
y = 2x 2y . (2)

Solution: Written in standard form the system becomes

x = Ax (3)
 
3 2
with A = . (4)
2 2

First we determine the eigenvalues.


!
0 = |A 1| (5)
= (3 )(2 ) 2(2) (6)
2
= 6 3 + 2 + + 4 (7)
2
= 2 (8)
2
= 2(1/2) + 1/4 1/4 2 (9)
2
= ( 1/2) 9/4 (10)
p
(1/2 1/2) = 9/4 (11)
1/2 = 1/2 3/2 (12)
1 = 2 2 = 1 . (13)

This tells us already a lot about the system. There is one solution that grows with a rate of 2 and one
solution that decays with a rate of -1. For general initial conditions, only the first solution will survive.

7
Then we determine the corresponding eigenvectors.

Av1 = 1 v1 (14)
    
3 2 x1 x1
= =2 (15)
2 2 y1 y1
3x1 2y1 = 2x1 (16)
2x1 2y1 = 2y1 (17)
x1 = 2y1 (18)
x1 = 2y1 (19)
= x1 = 2 y1 = 1 (20)
 
2
x1 = . (21)
1

Av2 = 2 v2 (22)
    
3 2 x2 x2
= = 1 (23)
2 2 y2 y2
3x2 2y2 = 1x2 (24)
2x2 2y2 = 1y2 (25)
2x2 = y2 (26)
2x2 = y2 (27)
= x2 = 1 y2 = 2 (28)
 
1
x2 = . (29)
2

Putting eigenvalues and eigenvectors together we obtain the two linearly independent solutions of the system.
 
2
x1 (t) = exp(2t) , (30)
1
 
1
x2 (t) = exp(1t) . (31)
2

The general solution grows along x1 and decays half as quickly along x2 .

Sketch not available!

2.7.4 Exercise: System of linear differential equations

Find the solution of the following system of linear differential equations and sketch its trajectories.

x = +x 4y (1)
y = x + y . (2)

Solution: Written in standard form the system becomes

x = Ax (3)
 
1 4
with A = . (4)
1 1

8
First we determine the eigenvalues.
!
0 = |A 1| (5)
2
= (1 ) 4 (6)
1 = 2 (7)
= 1 2 (8)
+ = 1 = 3 . (9)
This tells us already a lot about the system. There is one solution that grows with a rate of 3 and one
solution that decays with a rate of -1. For general initial conditions, only the first solution will survive and
actually grow infinitely.

Then we determine the corresponding eigenvectors.


Ax+ = + x+ (10)
    
1 4 x+ x+
= = (11)
1 1 y+ y+
x+ 4y+ = x+ (12)
x+ + y+ = y+ (13)
x+ = 2y+ (14)
 
2
= x+ = . (15)
1

Ax = x (16)
    
1 4 x x
= =3 (17)
1 1 y y
x 4y = 3x (18)
x + y = 3y (19)
x = 2y (20)
 
2
= x = . (21)
1

Putting eigenvalues and eigenvectors together we obtain two linearly independent solutions of the system.
 
2
x+ (t) = exp(t) , (22)
1
 
2
x (t) = exp(3t) . (23)
1

The general solution is therefore


x(t) = c+ x+ (t) + c x (t) (24)
with arbitrary constants c+ , c R. It decays along x+ and grows trice as quickly along x .

Sketch not available!

2.7.5 Exercise: System of linear differential equations with too few eigenvectors

Find the general solution of the following system of linear differential equations and sketch its trajectories.
x 3x 2y = 0 (1)
y y + z = 1 (2)
z z = 0 . (3)

9
Solution: Written in standard form (subtract second from third equation first) and matrix notation the
system becomes

x = Ax b (4)

3 2 0
with A = 0 1 1 (5)
0 0 1

0
and b = 1 . (6)
0

This is now an inhomogeneous system of linear differential equations. Thus, we first determine the special
solution x0 by setting x = 0.
!
0 = x0 = Ax0 b (7)
Ax0 = b (8)

3 2 0 x0 0
= 0 1 1 y0 = 1 (9)
0 0 1 z0 0
3x0 + 2y0 = 0 (10)
y0 z0 = 1 (11)
z0 = 0 (12)
x0 = 2/3 (13)
y0 = 1 (14)
z0 = 0 (15)

2/3
x0 = 1 . (16)
0

Once we have a special solution of the inhomogeneous system we only need to consider the associated
homogeneous system

x = Ax . (17)

To find the general solution of this system we first have to determine the eigenvalues of matrix A by finding
the roots of the characteristic polynomial (D: charakteristisches Polynom).
!
0 = |A 1| (18)
2
= (3 )(1 ) (19)

= 1 = 3 , 2 = 1 , 3 = 1 . (20)

10
Then we determine the corresponding eigenvectors.

Av1 = 1 v1 (21)

3 2 0 x1 x1
= 0 1 1 y1 = 3 y1 (22)
0 0 1 z1 z1
3x1 + 2y1 = 3x1 (23)
y1 z1 = 3y1 (24)
z1 = 3z1 (25)
x1 = x1 (26)
y1 = 0 (27)
z1 = 0 (28)

1
= v1 = 0 . (29)
0

Ax2 = 2 v2 (30)

3 2 0 x2 x2
= 0 1 1 y2 = 1 y2 (31)
0 0 1 z2 z2
3x2 + 2y2 = x2 (32)
y2 z2 = y2 (33)
z2 = z2 (34)
x2 = y2 (35)
z2 = 0 (36)
0=0 (37)

1
= v2 = 1 . (38)
0

Since 3 = 2 , we would arrive at the same equations for x3 as for x2 , but there is no other solution that
would be linearly independent of x2 . Thus, there is only one eigenvector for eigenvalue 3 = 2 = 1.

The next step, therefore, is to find a solution of the form (polynomial in t) exp(3 t). Thus, we make the
following ansatz for the homogeneous system:

x(t) = (a + b t) exp(t) . (39)

11
With this ansatz we find
(17) x = Ax (40)
(39)
b exp(t) + (a + b t) exp(t) = A(a + b t) exp(t) | exp(t) (41)
b + (a + b t) = A(a + b t) (42)
(a + b Aa) + (b Ab)t = 0 (43)
Aa = a + b (44)
Ab = b (45)
(3038)
= Aa = a + b (46)

1
b = 1 (47)
0
(5)
3a1 + 2a2 = a1 + 1 (48)
a2 a3 = a2 1 (49)
a3 = a3 (50)

1
b = 1 (51)
0
a1 = 1/2 a2 (52)
a3 = 1 (53)

1
b = 1 (54)
0

1/2
= a= 0 (55)
1

1
b = 1 (56)
0
Note that in (47) setting b = 0 would be valid but useless, because it would not yield a new solution.
In (55) the decision for a is somewhat arbitrary, but it can be compensated for by the eigenvector of the
corresponding (double) eigenvalue.

In summary, we have found one special solution of the inhomogeneous system and three linearly independent
solutions of the homogeneous system:

2/3
(16)
x0 = 1 (57)
0

1
(2129)
x1 (t) = 0 exp(3t) (58)
0

1
(3038)
x2 (t) = 1 exp(t) (59)
0

1/2 1
(3956)
x3 (t) = 0 + 1 t exp(t) . (60)
1 0

The general solution then is


x(t) = x0 + c1 x1 (t) + c2 x2 (t) + c3 x3 (t) (61)

12
with arbitrary constants ci .

We just sketch the flow field in certain planes centered at x0 . On the left is the sketch in the x-y-plane for
c3 = 0, which is pretty straight forward. On the right we display the a-b-plane for c1 = 0. First construct
the main trajectories for c2 = 0 or c3 = 0 (thick lines) - one is simply the extension of b. The other one is
starting in the direction of a but gets shifted towards b with time. The direction of the trajectory comes
near to that of b, but never is equal since a also has influence. The other trajectories are in between.

Sketch not available.

2.7.6 Exercise: Damped oscillator

Consider the ODE


s + 2 s + 2 s = 3 . (1)

(a) Write the ODE in standard form and find its general solution.
Solution: First we bring the equation in standard form and matrix notation. With x := (x, y)T :=
(s, s)T we get

x = y (2)
2
y = 2y x + 3 (3)
x = Ax b (4)
   
0 1 0
=: x . (5)
2 2 3

This is an inhomogeneous system of linear differential equations. Thus first find the special solution
x0 by setting x = 0.
   
! 0 1 0
0 = x0 = x 0 (6)
2 2 3
    
0 0 1 x0
= (7)
3 2 2 y0
y0 = 0 (8)
2
x0 + 2y0 = 3 (9)
y0 = 0 (10)
3
x0 = 2 (11)

3/ 2
 
x0 = (12)
0

Now we need to find the solution of the associated homogeneous system


 
0 1
x = x. (13)
2 2

First determine the eigenvalues of matrix A:


!
0 = |A 1| (14)
= (2 ) ( 2 ) = 2 + 2 + 2 (15)
p
1,2 = 2 2 (16)

13
Then determine the corresponding eigenvectors and start with v1 .

Av1 = 1 v1 (17)
    
0 1 x1 p x1
= = ( + 2 2 ) (18)
2 2 y1 y1
p
y1 = ( + 2 2 )x1 (19)
p
2 x1 2y1 = ( + 2 2 )y1 (20)
 
1
= v1 = . (21)
+ 2 2

We check that v1 actually solves equation (20).


p
2 x1 2y1 = ( + 2 2 )y1 (22)
p
2
x1 = (+ + 2 2 )y 1 (23)
(21)
p p
2 = (+ + 2 2 )( + 2 2 ) (24)
2 2 2
= + ( ) (25)
0 = 0. (26)

For v2 we obtain similarly

Av2 = 1 v2 (27)
    
0 1 x2 p
2 2 x2
= = ( ) (28)
2 2 y2 y2
p
y2 = ( 2 2 )x2 (29)
p
2 x2 2y2 = ( 2 2 )y2 (30)
 
= v2 = 1 . (31)
2 2

The general solution is then


x(t) = x0 + c1 v1 e1 t + c2 v2 e2 t (32)
with
3
 
x0 = 2 , (33)
0
 
v1 = 1 , (34)
+ 2 2
 
v2 = 1 , (35)
2 2
p
1 = + 2 2 , (36)
p
2 = 2 2 (37)

and arbitrary constants c1 and c2 .


(b) Discuss the qualitative behavior of the general solution as a function of > 0 on the basis of the
eigenvalues. Can you tell why it describes a damped oscillator?
Solution: For the eigenvalues are real and negative and the solution is a linear combination
of two decaying exponential functions and a constant. If 0 < < the eigenvalues are (conjugate)
complex numbers and the solution is oscillating. However, there is still a negative real part , so that
the solution decays. It therefore describes a damped oscillator.

14
2.7.7 Exercise: Metabolism of a pharmaceutical substance

Consider the metabolism of a pharmaceutical substance in the body. After oral intake the substance is
absorbed in the digestive tract and goes from there into the blood. From there it diffuses into the tissue and
is used or otherwise broken down (we dont care how this happens exactly).

The goal is to define a simple mathematical model for the time course of the amounts of substance in the
digestive tract, the blood, and the tissue. Make the following assumptions:

The substance in the digestive tract is absorbed into the blood with a rate of about 20% per hour.

The substance in the blood diffuses into the tissue with a rate of about 90% per hour.

The tissue uses or breaks down the substance with a rate of about 5% per hour.

(a) Define a system of linear differential equations that describes the dynamics of the amounts of substance
in the three different areas (digestive tract, blood, and tissue) and find the general solution.
Solution: We have to consider the 3 compartments digestive tract, blood, and tissue. We denote
the amount of substance in each compartment with x1 , x2 , and x3 , respectively. The transition rates
are: k1 = 0.2 , k2 = 0.9 , k3 = 0.05. The system can then be described by the following system of
differential equations:

x1 = k1 x1 , (1)
x2 = k1 x1 k2 x2 , (2)
x3 = k2 x2 k3 x3 , (3)

k1 0 0
x = Ax = k1 k2 0 x. (4)
0 k2 k3

This is a homogeneous system. To find the general solution first determine the eigenvalues. Since the
characteristic polynomial is simply (k1 )(k2 )(k3 ) the eigenvalues are obviously

1 = k1 , 2 = k2 , 3 = k3 . (5)

To find the corresponding eigenvectors vi we solve (A i 1)vi = 0 for i = 1, 2, 3.


For 1 = k1 :

0 0 0 x1
0 = k1 (k2 + k1 ) 0 y1 (6)
0 k2 (k3 + k1 ) z1
0 = k1 x1 + (k1 k2 )y1 (7)
0 = k2 y1 + (k1 k3 )z1 (8)
k1 2
y1 = x1 = x1 (9)
k2 k1 7
k2 k1 k2 12
z1 = y1 = x1 = x1 (10)
k3 k1 (k2 k1 )(k3 k1 ) 7

1
= v1 = 2/7 . (11)
12/7

15
For 2 = k2 :

(k1 + k2 ) 0 0 x2
0= k1 0 0 y2 (12)
0 k2 (k3 + k2 ) z2
0 = (k2 k1 )x2 (13)
0 = k1 x2 (14)
0 = k2 y2 + (k2 k3 )z2 (15)
0 = x2 (16)
k2 18
z2 = y2 = y2 (17)
k k2 17
3
0
= v2 = 1 . (18)
18/17

For 3 = k3 :

(k1 + k3 ) 0 0 x3
0= k1 (k2 + k3 ) 0 y3 (19)
0 k2 0 z3
0 = (k3 k1 )x3 (20)
0 = k1 x3 + (k3 k2 )y3 (21)
0 = k2 y3 (22)
x3 = 0 (23)
y3 = 0 (24)

0
= v3 = 0 . (25)
1

Therefore, the general solution is



1 0 0
x(t) = c1 2/7 e0.2t + c2 1 e0.9t + c3 0 e0.05t . (26)
12/7 18/17 1

(b) Assume that at a certain time point a certain amount has been taken in orally and that there has
been no intake ever before. Determine and sketch the time course of the amounts in the three different
areas.
Solution: We know that at time t = 0 there is an oral intake p which is the initial value for the
digestive tract. Since there has been no intake before the other 2 compartments have an initial value
of 0. Now
we candetermine the parameters c1 , c2 , and c
3 of the general solution
x1 (t) p
x(t) = x2 (t) for the initial conditions x(0) = 0 .
x3 (t) 0

16

p 1 0 0 c1
x(0) = 0 = 2/7 1 0 c2 (since e0 = 1) (27)
0 12/7 18/17 1 c3
c1 = p (28)
0 = 2/7c1 + c2 (29)
0 = 12/7c1 18/17c2 + c3 (30)
c1 = p (31)
c2 = 2/7c1 = 2/7p (32)
c3 = 12/7c1 + 18/17c2 (33)
= 12/7p + 18/17 2/7p = (12/7 36/119)p = 24/17p (34)

1 0 0
(26) 2 24
x(t) = p 2/7 e0.2t p 1 e0.9t + p 0 e0.05t (35)
7 17
12/7 18/17 1

1 0 0
= p 2/7 e0.2t + 2/7 e0.9t + 0 e0.05t . (36)
12/7 36/119 24/17

For an initial amount of 100 we get the following time course:

Figure: (Wiskott group, 2017) unclear.

(c) What happens if you give the substance as a continuous blood infusion? How much do you have to
give per hour to obtain a constant amount of 200mg in the tissue?
Solution: With a blood infusion of amount C the system becomes inhomogeneous.

0.2 0 0 0
x = Ax b = 0.2 0.9 0 x C (37)
0 0.9 0.05 0

Additionally we have the initial condition x1 (0) = 0 (since the substance is not passing the digestive
tract).
We have already determined the general solution (equation (26)) for the corresponding homogeneous
system. Therefore we only need to determine a special solution of the inhomogeneous system in order
to find the general solution.

17
!
0 = x = Ax0 b (38)
Ax0 = b (39)

0.2 0 0 x0 0
= 0.2 0.9 0 y0 = C (40)
0 0.9 0.05 z0 0
0.2x0 = 0 (41)
0.2x0 0.9y0 = C (42)
0.9y0 0.05z0 = 0 (43)
x0 = 0 (44)
0.9y0 = C (45)
0.05z0 = 0.9y0 (46)
x0 = 0 (47)
y0 = C/0.9 (48)
z0 = C/0.05 (49)

0
x0 = C/0.9 (50)
C/0.05

The general solution of the inhomogeneous system is then:



0 1 0 0
x(t) = C/0.9 + c1 2/7 e0.2t + c2 1 e0.9t + c3 0 e0.05t (51)
C/0.05 12/7 18/17 1

With our initial condition we easily get c1 = 0. Since x1 (t) only contains the component with factor
c1 we can reduce the solution to:

x1 (t) = 0 (52)
       
x2 (t) C/0.9 1 0
= + c2 e0.9t + c3 e0.05t (53)
x3 (t) C/0.05 18/17 1

Now the amount of substance in the tissue should be 200mg.

C 18 !
x3 (t) = c2 e0.9t + c3 e0.05t = 200mg (54)
0.05 17
c2 = c3 = 0 (for a constant solution) (55)
C
= 200mg (56)
0.05
C = 10mg (57)

This means you have to give 10mg of the substance per hour to obtain the constant amount of 200mg
in the tissue.

2.7.8 Exercise: Metabolism of a pharmaceutical substance revisited

Consider again the model for the metabolism of a pharmaceutical substance we had recently.

18
(a) Describe in words what happens if over years a patient gets a pill every 48 hours. Find an estimation
for the amount of substance in the patients tissue in the long run.
Hint:
P Use the previous results from the model, the superposition principle, and the geometric series
k x
k=1 x = 1x .
Solution: 48 hours after taking a pill there will be some substance left in the tissue. Therefore the
amount of substance in the patients tissue at the time of taking the pill will rise with each pill until it
has reached a steady state. In this state the amount of substance in the tissue 48 hours after taking
the pill will be the same as it was when taking the pill.
To calculate the amount of substance in the tissue at this steady state we use the general solution of
the system we have calculated earlier:

1 0 0
x(t) = c1 2/7 e0.2t + c2 1 e0.9t + c3 0 e0.05t (1)
12/7 18/17 1

Furthermore we already know that for the initial condition x(0) = (p , 0 , 0)T , with p indicating the
amount of substance in one pill, the values of the parameters are:

c1 = p , c2 = (2/7)p , and c3 = (24/17)p . (2)

Therefore the general solution for the amount of substance in the tissue for the given case is:

x3 (t) = (12/7)pe0.2t + (36/119)pe0.9t + (24/17)pe0.05t . (3)

From (1) and (2) we can calculate the amount in the different compartments 48 hours after taking the
first pill:

1 0 0
x(48) = p 2/7 e0.248 + (2/7)p 1 e0.948 + (24/17)p 0 e0.0548 (4)
12/7 18/17 1
5

6.8 10
= 1.9 105 p (5)
0.13

From linear differential equations we know that the sum of solutions (with an initial input) is itself a
solution (with the sum of the initial inputs). Therefore the solution for a pill every 48 hours is the sum
of the solutions for a pill at t = 0, a pill at t = 48, a pill at t = 96,. . . . An estimation of the amount
of substance in the tissue at the time the pill has to be taken is therefore:

x3 = x3 (48) + x3 (48 2) + x3 (48 3) + . . . (6)


X
= x3 (48 k) (7)
k=1
X
(3)
= (12/7)pe0.248k + (36/119)pe0.9t48k + (24/17)pe0.0548k (8)
k=1
X
= p (12/7)k + (36/119) k + (24/17) k (9)
k=1
(with = e0.248 , = e0.948 , = e0.0548 )
 X X X 
= p (12/7) k + (36/119) k + (24/17) k (10)
k=1 k=1 k=1
 
= p (12/7) + (36/119) + (24/17) (11)
1 1 1

4 20
= p 1.1 10 + 5.2 10 + 0.14) (12)
0.14p . (13)

19
That means after years (when the patient has reached the steady-state) the amount of substance in
the tissue at the time of taking the pill is approximately 14% of the amount of substance in one pill.
(b) Reduce the current 3-dimensional model of this metabolic process to a 2-dimensional model. Perform
the reduction such that the predictions of the 2-dimensional model (measured by the amount of the
substance in the tissue) are as faithfull as possible.
Solution: To keep the predictions of the model valid as far as possible one would neglect the eigenvector
of the smallest eigenvalue since the largest eigenvalues influence the system most. In our case the
smallest eigenvalue is -0.9 and its corresponding eigenvector is found in compartment 2 (blood). We
can interprete the decision as neglecting the component that is least important for the system - e.g.
where the substance is transportet fastest. The parameter of 0.9 means that the substance is passed
nearly instantaneous.
Therefore the new system would be:

x0 = 0.2x0 , (14)
x2 = 0.2x1 0.05x2 , (15)
 
0.2 0
x = Ax = x. (16)
0.2 0.05

(c) Give an estimate for how much the amount of substance in the patients tissue is fluctuating in the 48
hours between the two pills?
Solution: To determine the fluctuation of the amount we calculate the solution with the inital values of
the steady-state-condition. Since after 48 hours (directly before taking the new pill) the amount of sub-
stance in the digestive tract and in the blood is near to 0 (see (5)) we can assume x0 = (p , 0 , 0.14p)T .
Since the parameters c1 and c2 are not dependent on x3 we can use the values in (2). With (3) we can
determine c3 for the given initial value:

x3 (0) = (12/7)p + (2/7)(18/17)p + c3 = 0.14p (17)


c3 = 0.14p + 1.71p 0.30p (18)
c3 = 1.55p (19)

x3 (t) = 1.71pe0.2t + 0.3pe0.9t + 1.55pe0.05t . (20)

The sketch shows how the amount is fluctuating within the 48 hours from one pill to the next - it
varies from 14% to 70% of the amount in 1 pill.

Figure: (Wiskott group, 2017) unclear.

To determine the maximum one would have to determine the root of the derivative and calculate the
value of the function at this point. This is only possible numerically.

20
One could calculate an approximation analytically. Since the substance is passed from blood to tissue
nearly instantaneously one could omit the second term (the blood compartment). Mathematically this
can be based on the fact that the corresponding eigenvalue is the smallest one. Then the root would
be at 9.65 (x3 (9.65) = 0.71). The true maximum is at 9.85 (x3 (9.85) = 0.71) but the function
value is nearly the same.

3 Nonlinear differential equations

3.1 Fixed points

3.2 Qualitative analysis in 2D

3.3 Linear stability analysis

3.3.1 Exercise: Linearisation around a Fixpoint

Consider the classic Lotka-Volterra predator-prey model:



x = x(p sy)
y = y(q + cx)
First calculate the fixpoints. Then linearize the differential equations around the fixpoints which are not
equal to zero.

3.3.2 Exercise: Pitchfork bifurcation

Consider the following differential equation:


x + x + x3 x = 0 mit > 0
Bring it into standard form, find the fixpoints depending on the parameter and investigate their stability.

3.3.3 Exercise: Qualitative analysis of a nonlinear ODE in 2D - parabola

Consider the system of nonlinear differential equations


x = px (1)
2
y = 0.1(x y) , (2)
where p denotes a parameter of the system. Analyze the system as follows:

(a) Sketch the nullclines and determine the only fixed point.
Solution: The two nullclines are characterized by x = 0 and y = 0. For x we find
! (1)
0 = x = p x (3)
x = p. (4)
The x-nullcline therefore is a vertical line at x = p.
For the y-nullcline we find
! (2)
0 = y = 0.1(x2 y) (5)
2
y=x , (6)

21
which is simply a parabola. The two nullclines intersect at (x0 , y0 ) = (p, p2 ), which is the only fixed
point of the system.
Sketch not available!
(b) Sketch the flow field.
Solution: To qualitatively draw the flow field, we need to know the sign of x and y on the two sides
of their respective nullclines. From (1) we can see that x < 0 if x > p and x > 0 if x < p. From (2) we
can see that y < 0 if y > x2 and y > 0 if y < x2 . With this information it is easy to sketch the flow
field on the nullclines and then interpolate it for the whole x-y-plane.
The prefactor 0.1 means that the dynamics of y is somewhat slower than the dynamics of x unless the
system is close to an x-nullcline. Thus, away from the x-nullcline the flow field is primarily vertical.
Note that the trajectories crossing the parabola twice change curvature.
Sketch not available!
(c) Derive a linear approximation to the nonlinear differential equations around the fixed point. (If you
have not found a fixed point, linearize the differential equation around an arbitrary point.)
Solution: To analyze the stability of the fixed point (x0 , y0 ) = (p, p2 ) we have to linearize the differ-
ential equations (1,2) with respect to x and y around (x0 , y0 ), indicated by |0 . This yields

x(x, y) x(x, y)
x(x, y)|0 x(x0 , y0 ) +(x x0 ) + (y y0 ) (7)
| {z } x 0 y 0
=0
(1)
= (x x0 ) (1) + (y y0 ) 0 , (8)

y(x, y) y(x, y)
y(x, y)|0 y(x0 , y0 ) +(x x0 ) + (y y0 ) (9)
| {z } x 0 y 0
=0
(1)
= (x x0 ) 0.2x0 + (y y0 ) (0.1) . (10)

Thus, we can approximate the system of nonlinear differential equations around (x0 , y0 ) by
 0    0 
x 1 0 x
, (11)
y 0 0.2x0 0.1 y0

whith x0 := x x0 and y 0 := y y0 indicating relative coordinates. Note that x0 = x and y 0 = y.


(d) Analyze the stability of the fixed point. (If you were not able to linearize the system above around its
fixed point, describe at least how you would analyze the stability of a fixed point and give a simple
example.)
Solution: To analyze the stability of the fixed points, we need the eigenvalues of the matrix of the
linearized system. Thus, we set the characteristic polynomial to zero and find

! 1 0
0= (12)
0.2x0 0.1
= (1 )(0.1 ) (13)
1 = 1 2 = 0.1 . (14)

Since all eigenvalues are negative, the fixed point is always stable, which is consistent with the quali-
tative flow field analysis above.

3.3.4 Exercise: Qualitative analysis the of a nonlinear ODE in 2D - FitzHugh-Nagumo model

Consider the following FitzHugh-Nagumo model of a neuron, also known as Bonhoeffer-van der Pol oscillator:

V = V V 3 /3 W + I (1)
W = 0.08(V + 0.7 0.8W ) . (2)

22
V denotes the membrane potential, W is a recovery variable, and I is the stimulus current. Analyze the
system as follows:

(a) Sketch the nullclines for I = 0. How do they change if I changes?


Solution: The two nullclines are characterized by V = 0 and W = 0. For V we find
! (1)
0 = V = V V 3 /3 W + I (3)
3
W = V V /3 + I . (4)

The V -nullcline can therefore be written as a function W (V ). This function goes to


for V .
For I = 0 the function is odd, it is relatively easy to see that the roots are 0 and 3, and the local
extrema can be found as follows:
! dW (4)
0= = 1V2 (5)
dV
V = 1 , (6)
3
W (1) = 1 (1) /3 (7)
= 1 1/3 (8)
= 2/3 . (9)

Thus for I = 0 the V -nullcline has a local minimum at (V, W ) = (1, 2/3) and a local maximum at
(V, W ) = (+1, +2/3). For I 6= 0 the curve is simply shifted vertically by I. It is also interesting to
note that the slope of the nullcline at V = 0 is 1, which is the maximum slope between the two local
extrema. With this information we can easily sketch the V -nullcline.
For the W -nullcline we find
! (2)
0 = W = 0.08(V + 0.7 0.8W ) (10)
W = (V + 0.7)/0.8 , (11)

which is simply a linear function going through the points (V, W ) = (0, 0.875) and (V, W ) = (0.7, 0)
and can be drawn directly. Note that the W -nullcline has a slope greater than 1, so that it should
cross the V -nullcline only once.

Figure: (Scholarpedia FitzHugh portrait.gif 2011-07-13)1 CC BY-NC-SA 3.0.

(b) Sketch the flow field.


Hint: Note the prefactor 0.08. What does that mean for the relative speed of the dynamics of V and
W.

23
Solution: To qualitatively draw the flow field, we need to know the sign of V and W on the two sides
of their respective nullclines. From (1) we can see that V < 0 if W > WV =0 and V > 0 if W < WV =0 .
From (2) we can see that W < 0 if W > WW =0 and W > 0 if W < WW =0 . WV =0 and WW =0 indicate
the W -values on the V - and W -nullclines, respectively. With this information it is easy to sketch the
flow field on the nullclines and then interpolate it for the whole V -W -plane.
The prefactor 0.08 means that the dynamics of W is much slower than the dynamics of V unless the
system is close to a V -nullcline. Thus, away from the V -nullcline the flow field is primarily vertical.
See figure above.
(c) Discuss the qualitative behavior for different values of I and also for cases where I changes suddenly.
Solution: Not available!
(d) Determine the fixed points analytically. How many are there?
Hint: It might be useful not to consider I as the independent parameter but to write I as a function
of another variable.
Solution: The fixed points (Vi , Wi ) are simply the intersection points of the two nullclines. Thus the
condition is:

WV =0 (Vi ) = WW =0 (Vi ) (12)


(4) (11)
= Vi Vi3 /3 + I = (Vi + 0.7)/0.8 (13)
I= Vi3 /3 + 0.25Vi + 0.875 . (14)

Since this is a cubic equation in Vi , the values of Vi given I cannot be determined analytically. But we
can determine I as a function of Vi . Furthermore, since
dI
= Vi2 + 0.25 (15)
dVi
> 0, (16)

I(Vi ) is a bijective function. This means there is exactly one fixed point for any given value of I, which
we will refer to as V0 .
The corresponding W0 -values can be computed with (4) or, much simpler, with (11).
Thus, the fixed point is (V0 , W0 ) = (V0 , (V0 + 0.7)/0.8) and the input current to realize this fixed point
is I = V03 /3 + 0.25V0 + 0.875. From now on we will consider V0 as the independent parameter and not
I.
(e) Derive a linear approximation to the nonlinear differential equations around the fixed points.
Solution: To analyze the stability of the fixed point (V0 , W0 ) we have to linearize the differential
equations (1,2) with respect to V and W around (V0 , W0 ), indicated by |0 . This yields

V (V, W ) V (V, W )
V (V, W )|0 V (V0 , W0 ) +(V V0 ) + (W W0 ) (17)
V W

| {z }
=0 0 0
(1)
= (V V0 ) (1 V02 )
+ (W W0 ) (1) , (18)

W (V, W ) W (V, W )
W (V, W )|0 W (V0 , W0 ) +(V V0 ) + (W W0 ) (19)
V W

| {z }
=0 0 0
(1)
= (V V0 ) 0.08 + (W W0 ) (0.064) . (20)

Thus, we can approximate the system of nonlinear differential equations around (V0 , W0 ) by
 0    0 
V 1 V02 1 V
, (21)
W 0 0.08 0.064 W0

with V 0 := V V0 and W 0 := W W0 indicating relative coordinates. Note that V 0 = V and W 0 = W .

24
(f) Analyze the stability of the fixed points at V0 {0, 1, 2}.
Solution: To analyze the stability of the fixed points, we need the eigenvalues of the matrix of the
linearized system. Thus, we set the characteristic polynomial to zero and find
2

! 1 V0 1
0 = (22)
0.08 0.064
= (1 V02 )(0.064 ) 0.08(1) (23)
2
= + 2(V02 1 + 0.064)/2 + 0.064(V02 1) + 0.08 (24)
2
= + 2 (V02 /2 0.468) + (0.064V02 + 0.016) (25)
| {z } | {z }
=:b =:c

= ( + b)2 b2 + c (26)
p
= b b2 c (27)
q
(25)
= (V02 /2 0.468) (V02 /2 0.468)2 (0.064V02 + 0.016) (28)
q
2
= (V0 /2 0.468) (V02 /2 0.532)2 0.08 (see 3539) . (29)

With this equation we can calculate the eigenvalues for particular values of V0 .

0.918...
V0 = 0 = +0.468 0.450... = (30)
0.0174...
V0 = 1 = 0.032 i 0.281... (31)

0.0915...
V0 = 2 = 1.532 1.440... = . (32)
2.972...

Thus, for V0 = 0 the fixed point is unstable, for V0 = 1 the fixed point is stable and oscillatory, and
for V0 = 2 the fixed point is stable and not oscillatory.
(g) If you like, analyze the stability of the fixed points for any value of V0 .
Solution: Now we want to do the stability analysis not only for particular values of V0 but for all
values. The first critical question is for what values of V0 the argument of the square root is negative,
so that the eigenvalues is complex.

=( ) 6= 0 (i.e., fixed point is oscillatory) (33)


(27)
0 > b2 c (34)
(25)
= (V02 /2 0.468)2 (0.064V02 + 0.016) (35)
= (V02 /2)2 2 0.468V02 /2 + 0.4682 2 0.064V02 /2 0.016 (36)
= (V02 /2)2 2 0.532V02 /2 + 0.4682 0.016 (37)
= (V02 /2 0.532)2 0.5322 + 0.4682 0.016 (38)
= (V02 /2 0.532)2 0.08 (39)

(V02 /2 0.532)2 < 0.08 (40)



0.08 < V02 /2 0.532 < + 0.08 (41)

0.532 0.08 < V02 /2 < 0.532 + 0.08 (42)

1.064 0.32 < V02 < 1.064 + 0.32 (43)
1.276... < V0 < 0.705...
+0.705... < V0 < +1.276... . (44)

If the argument of the square root is negative the square root is imaginary and the real part is only
determined by b. So the next question is, for which values of V0 is b positive, so that the fixed point

25
is unstable given the eigenvalues are complex.

Given =( ) 6= 0 : (45)
?
<( ) > 0 (46)
(27)
0>b (47)
(25)
= V02 /2 0.468 (48)
V02 < 0.936 (49)
0.967... < V0 < +0.967... . (50)

If, on the other hand, the argument of the square root is positive, then the eigenvalues are real and
the square root has to be taken into account as well to determine the real part of the eigenvalues. For
determining the stability of the fixed point it is sufficient to determine whether there is at least one
eigenvalue with a positive real part. Since + if =( ) = 0, we only consider + .

Given =( ) = 0 : (51)
?
<(+ ) > 0 (52)
(27)
p
b2 c > b (53)
| {z }
0

(true if b < 0, false if b 0 since 0 < c b2 , see (25, 51))


0>b (54)
(25)
= V02 /2 0.468 (55)
V02 < 0.936 (56)
0.967... < V0 < +0.967... . (57)

This is the same result as for the case =( ) 6= 0. Thus, in any case
?
<(+ ) > 0 (i.e., the fixed point is unstable) (58)
(50,57)
0.967... < V0 < +0.967... . (59)

As a rule of thumb, we can say that the fixed point is unstable between the two local extrema (see 6,
59) of the V -nullcline and it is oscillatory around the two local extrema (see 44).

Notes

1 Scholarpedia FitzHugh portrait.gif 2011-07-13, CC BY-NC-SA 3.0, http://www.scholarpedia.org/article/File:FitzHugh_


portrait.gif

26

Das könnte Ihnen auch gefallen