Systems modeling

Ordinary Differential Equations

Master of Interdisciplinary & Innovative Engineering

2023

Departament de mecànica de fluids

Escola d’Enginyeria de Barcelona Est (EEBE)
Universitat Politècnica de Catalunya (UPC)

This work is licensed under the Creative Commons

Attribution-ShareAlike 4.0 International License.
Ordinary Differential Equations (ODE)

Index:
• Definitions and basic concepts.

• Separation of variables. Population models in Biology.

• Homogeneous differential equations. Aircraft with side-wind.

• Linear ODE. Chemical dilutions.

• Exact differential equations. Building a radio telescope.

• Qualitative theory of differential equations.

• Basics.
• Bifurcations.

1/30
Definitions and basic concepts

Deterministic process: Its entire future and past are uniquely

determined by the present state. The set of all states is called phase space.
Examples: Classical mechanics, population models in biology, chemical
reactions, etc...
1st order ordinary differential equations (ODE): They are given
in explicit form:
ẋ = f (t, x), (1)
n dx n n
where t ∈ R, x ∈ R , ẋ = and f : R × R → R is a vector field.
dt ,
Alternatively, ODE may be also given in implicit form:

F (t, x, ẋ) = 0, (2)

where F : R × Rn × Rn → Rn . If f ≡ f (x) the ODE is called autonomous.

dy
The notation y ′ = f (x, y), where x ∈ R, y ∈ Rn , y ′ = dx , and
f : R × Rn → Rn , instead of Eq. (1), is quite common and may be used.
2/30
Definitions and basic concepts

A function ϕ : I ⊂ R → Rn is a solution of eq. 1 if

dϕ
= f (t, ϕ(t)), ∀t ∈ I.
dt
Then, the graph of the solution ϕ of an ODE is an integral curve of the
vector field f .

The initial condition of an ODE specifies the time and the phase space point
(t0 , x0 ) of a solution. Then ϕ(t0 ) = x0 .

ODE may involve higher derivatives and definitions of Eqs.(1-2) are easily
generalised. Explicit high order ODE with autonomous vector field can be
always reduced to a system of 1st order ODE.

Reduction to quadratures is used to solve some particular ODE but a

general method does not exist. The only option for non-integrable ODE is
to employ numerical methods.
3/30
Definitions and basic concepts

Example: Small oscillations (Hooke’s law)

f (x, y) = (y, −x), x2 + y 2 = R 2
4
The restoring force is proportional to
5
the displacement:
3

2
4 ẍ = −kx, k>0

Vector Length
1
3
0 is equivalent to the system
y

−1 2
ẋ = y
−2

−3
1
ẏ = −kx.
−4 0
−4 −3 −2 −1 0
x
1 2 3 4 with vector field f (x, y) = (y, −kx).
Note that scaling time we can generically assume that k = 1. The integral
curves are circles centered at the origin x2 + y 2 = R2 . If at t0 = 0 the
position and the velocity of the particle are (x0 , ẋ0 ) = (x0 , y0 ) = (1, 0) the
solution is ϕ(t) = (cos t, sin t) (inner circle of the figure).
4/30
Separation of variables

Consider a differential equation that can be written as

g(x)
y′ = ⇐⇒ h(y)dy = g(x)dx,
h(y)
i.e the variables involved can be separated on both sides of the equation.
By integrating both sides:
Z Z
h(y)dy = g(x)dx + C ⇐⇒ H(y) = G(x) + C,

where H and G are primitives of h and g, respectively. Imposing an initial

condition y(x0 ) = y0 provides the value of the constant C.

Example:
Solve y ′ − 2xy = x with y(0) = 5.
11 x2
Solution: y(x) = 2 e − 12 .

5/30
Example: Population models in biology

The rate of change of individuals ẋ in a given population of x individuals

depend on the birth and death rates and migration process. On isolated
populations Malthus (1798) considered the normal reproduction equation
ẋ = ax, for a > 0, which is valid when there’s enough food for all the
individuals (no competition). Using separation of variables the general
solution is x(t) = keat with k > 0 being the initial t = 0 population.
f (t, x) = (1, x), (t, ket )
4
4 The normal reproduction equation is
3.5
transformed (by scaling time) to ẋ = x
3
without loss of generality. The latter
3
Vector Length

ODE can be written as a system as

2
x

2.5
ṫ = 1
2
1 ẋ = x.
1.5

0 1
with vector field f (t, x) = (1, x).
0 1 2 3 4 6/30
t
Example: Population models in biology

Verhulst (1836) improved Malthus’ model, taking into account finite available
food:
ẋ = x(a − bx), with a, b > 0.
By scaling t and x the general logistic equation ẋ = x(1 − x) is obtained.
Note that for x = 1, the population remains constant since ẋ = 0. Moreover,
since x > 0 then the population increases (ẋ > 0) if x < 1. Conversely, the
population decreases.
Considering x0 = x(0) (the initial population) and by separation of variables
the population at each time is
 t
ke
 1+ket if x0 < 1


x(t) = 1 if x0 = 1

 −ket

if x0 > 1
1−ket
x0
where k = |1−x0 | .

7/30
Example: Population models in biology
f (t, x) = (1, x(1 − x)), (t, et /(1 + et )), (t, 1), (t, −5et /2(1 − 5et /2))
2 2.2

Vector Length
1.8

1
x

1.6

1.4

1.2

0 1
0 1 2 3
t

(x, ẋ) = (x, x(1 − x))

0.4
Phase portrait (x, ẋ) displaying the
fixed points (ẋ = f (t, x) = 0). The
ẋ

0.2
point x = 1 is stable and x = 0 is
0•
0 0.5
•1 1.5
unstable.
x 8/30
Example: Population models in biology

Exercice:
Consider now that part of the individuals are emigrating at a constant rate
c > 0. By scaling t and x the general migration equation (also called
harvesting)
ẋ = x(1 − x) − c
is obtained.
Determine the fixed points and plot the corresponding vector field with the
integral curves as well as the phase portrait. Discuss the results as function
of the migration parameter c.
Perform the same analysis assuming now the migration rate is relative to
the population:
ẋ = x(1 − x) − cx

9/30
Homogeneous differential equations

A differential equation of the form P (x, y)dx + Q(x, y)dy = 0 is called

homogeneous if P and Q are homogeneous functions of the same degree d:

P (λx, λy) = λd P (x, y), Q(λx, λy) = λd Q(x, y), ∀λ.

Using the change of variable y = xu, this definition is equivalent to

dy y
y′ = =f ,
dx x
P (1,y/x)
with f (y/x) = − Q(1,y/x) , and the solution is given by separation of variables:

du dx
  =−
u+ P (1,u) x
Q(1,u)

10/30
Example: Aircraft with side-wind

Determine the aircraft trajectory assuming a constant modulus of the aircraft

velocity v, which is always pointing to the origin of coordinates. The aircraft
is subjected to a lateral wind of constant velocity w directed to the positive
x-axis.

Q •S

11/30
Example: Aircraft with side-wind

The tangent to the trajectory at a given point P = (x, y) is parallel to

−→
⃗ The intersection of the tangent line (with slope y ′ ) with the
P S = ⃗v + w.
X-axis gives the point X0 = x − yy′ .

Since the triangles P[

QS and P\
OX0 are similar:

x − yy′ w
−−→ = v .
|P O|

w −−→ p
Considering α = v and with |P O| = x2 + y 2 an homogeneous ODE is
obtained:
y p  p 
2 + y 2 =⇒ ydx + α x2 + y 2 − x dy = 0
x− = α x
y′

12/30
Example: Aircraft with side-wind

The trajectory is best described as x ≡ x(y) rather than y ≡ y(x) (see

figure). Then we use the change x = uy to obtain

dy du
=− √
y α 1 + u2
√
which by integration gives (cy)−α = u + 1 + u2 .
Finally, since
 p  p  p
u + 1 + u2 u − 1 + u2 = −1 =⇒ u − 1 + u2 = −(cy)α ,

and then the solution is

(cy)−α − (cy)α k −1 y 1−α − ky 1+α
u= =⇒ x = with k = cα
2 2

13/30
Linear ODE

Given a(x) and b(x) an ODE of the form y ′ + a(x)y = b(x) is called linear.
The solution with initial condition y0 = y(x0 ) is:
 Z x 
−A(x) A(t)
y(x) = e y0 + e b(t)dt , (3)
x0

where Z x
A(x) = a(s)ds. (4)
x0

Example: Solve 2x2 y ′ + 4xy = 7x4 with y0 = y(x0 ). This ODE is equivalent
to y ′ + x2 y = 72 x2 . We first find A(x) = 2 ln xx0 to solve the integral
Z x Z x 
2 ln xt 7 2 7

A(t)
e b(t)dt = e 0 t dt = (x5 − x50 ).
x0 x0 2 10x20
The solution is then
 
7 7 3 c 7 5

− 2 ln xx
y(x) = e 0 y0 + (x5 − x50 ) = x + 2, c = y0 x20 − x
10x20 10 x 10 0

14/30
Example: Chemical dilutions

A dilution is contained in a tank with a mixer to maintain uniform the dilution.

The initial volume and concentration are V0 (l) and C0 (kg/l), respectively. A
dilution with concentration Ce (kg/l) is introduced in the tank with velocity ve
(l/min) and at the same time the tank is emptied with velocity vs (l/min). Let
be V (t), C(t) and s(t) the volume, concentration, and solute quantity,
respectively, at a given instant t.
The rate of change of the solute quantity is
ṡ(t) = ve Ce − C(t)vs (kg/min) (5)
and the rate of volume change is V̇ (t) = ve − vs . Using separation of
variables we first find the volume V (t) = V0 + (ve − vs )t.
s(t) s(t)
Since C(t) = V (t) = V0 +(ve −vs )t we substitute in eq. (5) to obtain the linear
ODE
s(t)
ṡ(t) = Ce ve − vs
V0 + (ve − vs )t

15/30
Example: Chemical dilutions

Using Eqs. (3-4) we find the solute quantity at each instant t assuming
ve ̸= vs . If ve = vs (V (t) = V0 , ∀t) the ODE for the solute quantity can be
solved by separation of variables. If s0 is the initial (t = 0) quantity of the
solute, the general solution is
  vs
 V0 ve −vs (s − C V ) + C V (t) if v ̸= v
V (t) 0 e 0 e e s
s(t) =
(s − C V ) e −v V0
et
+C V if v = v
0 e 0 e 0 e s

where V (t) = V0 + (ve − vs )t.

• If ve > vs and assuming a tank with very large volume the solute
quantity s(t) tends to Ce V (t) for large t, i.e, C(t) → Ce .
V0
• If ve < vs then the tank will be empty after t = vs −ve minutes.
• If ve = vs the solute quantity s(t) tends to Ce V0 for large t, i.e,
C(t) → Ce .

16/30
Exact differential equations

∂P ∂Q
The ODE P (x, y)dx + Q(x, y)dy = 0 is exact if and only if ∂y = ∂x .

The solution is the family of curves u(x, y) = C, where C is a constant, that

verifies ∂u ∂u ⃗
∂x = P and ∂y = Q. Note that then the vector field F = (P, Q) is
conservative, since F⃗ = ∇u, with u(x, y) ∈ R a potential function.
The general solution is:
Z x Z y
u(x, y) = P (v, y)dv + Q(x0 , v)dv = C. (6)
x0 y0

Example: Solve xdx + ydy = 0. It is exact since P (x, y) = x and Q(x, y) = y

∂Q
verify ∂P
∂y = ∂x = 0. Using the above formula the family of curves is
x + y = R , with R2 = 2C + (x20 + y02 ). They are circles centered into the
2 2 2

origin.

17/30
Example: Building a radio telescope

A radio telescope is designed as a revolution surface of a plane curve with

an specific property. The plane curve has to concentrate a beam of parallel
rays, which reflect on it, into a point.
We assume that the point is the origin and the rays are parallel to the
horizontal X-axis.

The angle α of the incident ray with

respect the tangent to the curve is the
same than the angle β − α of the
reflected ray. Thus
tan(β)−tan(α)
tan(α) = tan(β − α) = 1+tan(α) tan(β) .
Since tan(α) = y ′ and tan(β) = xy the
ODE is: y
− y′
y′ = x ′ y . (7)
1+y x

18/30
Example: Building a radio telescope

Solving Eq. (7) for y ′ (2nd degree equation) the ODE is rewritten as
p
′ −2x ± 4x2 + 4y 2 p
y = =⇒ ydy + xdx ∓ x2 + y 2 dx = 0 =⇒
2y
!
x y
=⇒ ∓ p + 1 dx ∓ p dy = 0,
x2 + y 2 x2 + y 2
which is an exact ODE with
!
x y
P (x, y) = ∓p +1 and Q(x, y) = ∓ p ,
x2 + y2 x2 + y2

verifying
∂P ∂Q 3
= = ±xy(x2 + y 2 )− 2
∂y ∂x

19/30
Example: Building a radio telescope

The expression for the general solution (eq. (6)) is:

p q
u(x, y) = ∓ x2 + y 2 ± y02 + x20 + x − x0 ,

and since u(x, y) = c, the family of curves is

p q
± x2 + y 2 = x − c − x0 ± y02 + x20 .

Squaring both sides leads to

q
x2 + y 2 = (x + k)2 , with K = ± x20 + y02 − c − x0 ,

which is equivalent to
y2 k
x= − .
2k 2
This family of curves correspond to parabolas with their focus in the origin.

20/30
Bibliography recommended

• V. I. Arnold Ordinary Differential Equations, Springer-Verlag, 1992.

• R. K. Nagle & E. B. Saff & A. D. Snider. Fundamentals of Differential

Equations, Adison-Wesley, 2012.

• E. Hairer & S. P. E. Norsett & G. Wanner. Solving Ordinary

Differential Equations I, Springer, 1987.

• E. Hairer & G. Wanner. Solving Ordinary Differential Equations II,

Springer, 1996.

• L. N. Trefethen & A. Birkisson & T. A. Driscoll. Exploring ODEs,

Society for Industrial and Applied Mathematics, 2018.

21/30
Qualitative Theory of ODE. Basics

Given the system of ODEs

ẋ = f (x, p), (x, p) ∈ U ⊂ Rn × R, (8)
depending on the parameter p, and its solution ϕ(t, x, p) with initial condition
x at t = 0:
ϕ(0, x, p) = x (9)

A point xe ∈ Rn is called an equilibria (also fixed point or steady state) if

f (xe , p) = 0.

A trajectory ϕ(t, x∗ , p), starting at

x∗ ∈ Rn , is called a periodic orbit if
there’s a minimal period T > 0
verifying

ϕ(T, x∗ , p) = x∗
22/30
Qualitative Theory of ODE. Basics

A particular periodic orbit satisfies the system of equations:

x − ϕ(T, x, p) = 0
g(x, p) = 0,

and x is selected by a phase condition g(x, p) = 0.

Usually, g(x, p) = 0 is an hyperplane transverse to the periodic orbit. The
case
g(x, p) = f (x0 , p0 ), x − x0 ,

where x0 is an approximation of x for p = p0 is called Poincaré condition.

The computation of equilibria and periodic orbits as functions of parameters,
and the study of their stability, imply solving nonlinear systems of equations
and eigenvalue problems.

23/30
Qualitative Theory of ODE. Basics

A fixed point xe is Lyapunov stable if or every neighborhood U of xe , it exist

a neighborhood V ⊂ U of xe such that if x ∈ V then ϕ(t, x, p) ∈ U for all
t ≥ 0. Otherwise is unstable.
A fixed point xe is asymptotically stable if it is Lyapunov stable and there is
a neighborhood U of xe such that if x ∈ U then limt→∞ ||ϕ(t, x, p) − xe || = 0.
A fixed point xe is exponentially stable if it is asymptotically stable and
there exist α > 0 and β > 0 and a neighborhood U of xe such that if x ∈ U
then ||ϕ(t, x, p) − xe || ≤ α||x − xe ||e−βt .
If f ∈ C 1 and xe is a fixed point such that Df (xe ) has all the eigenvalues λi
with Reλi < 0 then xe is exponentially stable. If exist j such that Reλj > 0
then xe is unstable =⇒ The eigenvalues of Df (xe ) of largest real part
must be computed to know the stability of fixed points.
A set W is invariant under the flow ϕ(t, x, p) if for each x ∈ W ,
ϕ(t, x, p) ∈ W for any t. That is ϕ(t, W, p) = W for all t.
Example: A periodic orbit ϕ(t, x∗ , p) ∈ Rn , t ≥ 0, is an invariant set. 24/30
Qualitative Theory of ODE. Basics

The invariant set W is stable if for every neighborhood U of W , it exist a

neighborhood V ⊂ U such that if x ∈ V then ϕ(t, x, p) ∈ U for all t ≥ 0.
Otherwise is unstable.
The invariant set W is asymptotically stable if it is stable and there is a


neighborhood U of W such
 that if x ∈ U then limt→∞ ρ ϕ(t, x, p), W = 0,


with ρ x, W = inf y∈W ||x − y|| .

The first variational equation about a periodic orbit ϕ(t, x∗ , p) of period T is

Ṁ = Df (x(t))M, with initial condition M (0) = Id.

The solution at time T is called the monodromy matrix M (T ) and its

eigenvalues are the Floquet multipliers of the periodic orbit.
The monodromy matrix M (T ) always has a unit eigenvalue with
eigenvector ẋ(0) = ẋ(T ) = f (x(0)). This eigenvalue is called the trivial
eigenvalue of the periodic orbit.
25/30
 Qualitative Theory of ODE. Basics
11
A periodic orbit ϕ(t, x∗ , p) ∈ Rn , t ≥ 0, with ϕ ∈ C 2 is linearly asymptotically
stable if its monodromy matrix has all the eigenvalues inside the unit circle
(except the trivial one). 1 0.7
P3 Ra1
P1 Ra2
If a periodic orbit 0.6 Ra3
ϕ(t, x∗ , p) ∈ Rn , t ≥ 0, Ra11
3
max max
0.5
is linearly 0.5
max max
asymptotically stable,
then it is 0.4
ℑ(λ)

asymptotically stable. 0
max max
0.3
The eigenvalues of
max max

M (T ) of largest 0.2
max max
-0.5
modulus have to be
max max
computed to asses the 0.1
stability of periodic (a) (b)
orbits. -1 0
0 0.5 1 0.7 0.8 0.9 1
26/30
ℜ(λ) ℜ(λ)
Qualitative Theory of ODE. Bifurcations

A bifurcation is a sudden qualitative or topological change of the solutions

of an ODE due to an infinitesimal smooth perturbation of the parameter.
Local bifurcations: Changes of stability of local equilibria, periodic orbits or
other invariant sets as a the parameter is varied beyond a critical parameter.
Global bifurcations: The topological structure of the phase space changes
dramatically as the parameter is varied. Is usually associated to a collision
of invariant sets at a critical parameter.
A bifurcation diagram shows the values of a measured property of the
asymptotic state (fixed points, periodic orbits, quasiperiodic orbits or chaotic
attractors) of a system as a function of a bifurcation parameter in the
system.
An attractor is an asymptotic state towards which the system tends when
evolving a wide range of initial conditions of the system.

27/30
ments
Qualitative
10
(a)
Theory of ODE. Bifurcation Diagrams
1

Horizontal Axis: A parameter

P1
(or a function of the
0 P2
Pe

10 parameter) of which the

∝
√
Ra − Rac
system of ODE depends on.
P3
Weak bran h Vertical Axis: The measured
-1
10 -3 -2 -1 0
properties are usually time
10 10 10 10
f = Ra/Rac − 1
Ra averaged and can be local (a
ments 4 single space variable of the
(b)
P3
solution) or global (a property
involving all the space
3
103 ωE

P2
P1
variables of the solution).
Example: Bifurcation
k bran h 2
diagrams in a problem of
Rayleigh-Bénard rotating
1
10
-3
10
-2
10
-1
10
0 convection in a sphere.
f = Ra/Rac − 1
Ra 28/30
Qualitative Theory of ODE. Fold and Hopf Bifurcations

Consider a fixed point xe and the eigenvalues λi ∈ C of df (xe ). Let n− , n0 ,

and n+ the numbers of eigenvalues with negative, zero, and positive real
part, respectively.
A fixed point xe is hyperbolic if n0 = 0 (there are no eigenvalues of df (xe )
on the imaginary axis). If n− , n+ ̸= 0 the fixed point is a hyperbolic saddle.
Let be an hyperbolic fixed point x0 at a given parameter p = p0 . By varying p
the hyperbolicity can be only lost in two ways: either exist j such that λj = 0
or λj,j+1 = ±ω0 i, with ω0 > 0.
The bifurcation associated with the appearance of λj = 0 is a fold
bifurcation. It is also called tangent, saddle-node or turning point
bifurcation. Example: ẋ = p + x2 .
The bifurcation associated with the appearance of λj,j+1 = ±ω0 i, with
ω0 > 0, is a Hopf bifurcation. Example: ż = (p + i)z ± z|z|2 , z ∈ C.

29/30
Bibliography recommended

• Y. A. Kuznetsov Elemets of Applied Bifurcation Theory,

Springer-Verlag, 1995.

• J. Guckenheimer & P. Holmes. Nonlinear Oscillations, Dynamical

Systems and Bifurcations of Vector Fields, Springer-Verlag, 1985.

• J. D. Crawford. Introduction to Bifurcation Theory, Reviews of

Modern Physics, Vol. 63, 1991.

• S. Wiggins. Global Bifurcations and Chaos, Springer-Verlag, 1988.

• M. Golubitsky & I. Stewart. & D. G. Schaeffer Singularities and

Groups in Bifurcation Theory, Springer-Verlag, 1988.

30/30