Bicout Modelisation M1 2013

Linear Stability Analysis
Dominique J. Bicout
Biomathématiques et Epidémiologies, EPSP - TIMC, UMR 5525,

UJF - VetAgro Sup, Veterinary campus of Lyon. 69280 Marcy
l’Etoile, France
UE MMED4100, Joseph Fourier - Grenoble University

Introduction Linear Stability Analysis Illustrative Examples
Linear System
The Exponential
Consider the following ordinary differential equation (ODE):
dx
= F (x) = ax + b for t ≥ 0 , (1)
dt
with the initial condition: x(t = 0) = x0 .
What can we learn out of this?
D. J. Bicout
Linear System
The Exponential
general solution of Eq.(1)

b
x(t) = x0 exp{at} + [exp{at} − 1] (2)
a
stability: eigenvalue a steady state (for a < 0)

b
F (x ∗ ) = 0 =⇒ x ∗ = −
 a
 ∞ ; a ≥ 0 unstable
lim x(t) = b
t→∞  − ; a<0 stable
a
D. J. Bicout
Linear System
Objectives
What happen when F (x) is non-linear?
How to deal with higher dimensional systems where F(x) is

now (either a linear or non-linear) matrix?
D. J. Bicout
General theory
n−dimensional Dynamical System
Consider
 the  n−dimensional dynamical system, represented by the vector
x1
 x2 
x =  . , and described by the ODE:
 
 ..  
dx 1
 = F1 (x1 , x2 , · · · , xn )
xn




 dt


dx  dx2

= F(x) ⇐⇒ = F2 (x1 , x2 , · · · , xn ) for t ≥ 0 , (3)
dt dt



 .
..


 dxn = Fn (x1 , x2 , · · · , xn )



dt
with the initial conditions, x(t = 0) = x0 , and F(x) is a matrix nonlinear
function of x.
Aim: use what we have learned from the exponential system
D. J. Bicout
General theory
Taylor’s Expansion

∂F
F(x) ≈ F(x∗ ) + (x − x∗ ) + · · · (4)
| {z } ∂x
|{z}
=0
J(x) x=x∗
fixed points Jacobian matrix
 ∂F1 ∂F1 ∂F1 

F1 (x1 , x2 , · · · , xn ) = 0


 ∂x1 ∂x2
··· ∂xn

  
  
 F (x , x , · · · , xn ) = 0

∂F2 ∂F2 ∂F2
2 1 2
 
∗ ···
F(x ) = 0 =⇒ (5) 
∂x1 ∂x2 ∂xn

. J(x) =  (6)
 
.

 

 .
 . . . 

  . . . 
Fn (x1 , x2 , · · · , xn ) = 0 . . ··· .
  
 
∂Fn ∂Fn ∂Fn
∂x1 ∂x2
··· ∂xn
D. J. Bicout
General theory
Linear approximation (close to fixed points)
Setting δx = x − x∗ , the Eq.(3) becomes,

d(δx)
= J(x∗ ) for t ≥ 0 , (7)
dt
with the initial conditions, δx(t = 0) = x0 − x∗ , and J(x∗ ) is a constant
matrix independent of x.
general solution (close to fixed points)
n
∗
X λi = eigenvalues
x(t) = x + ai vi e λi t
;
of J(x∗ ) (8)
vi = eigenvectors
i=1
det |J(x∗ ) − λ 1| = 0 ; characteristic equation (9)
ai are constants determined by using the initial conditions.

D. J. Bicout
Take home messages
About stability?
Procedure for stability analysis at fixed points
1 Determine the fixed point vector, x∗ , solving F(x∗ ) = 0

∂F(x)
2 Construct the Jacobian matrix, J(x) =
∂x
∗ ∗
3 Compute eigenvalues of J(x ): det |J(x ) − λ 1| = 0
4 Stability or instability of x∗ based on the real parts, <(λ), of

eigenvalues
D. J. Bicout
Take home messages
What do eigenvalues tell us about stability?
stability of x∗ based on the real parts of eigenvalues
1 all eigenvalues have real parts less than zero =⇒ x∗ is stable

2 at least one of the eigenvalues has a real part greater than zero
=⇒ x∗ is unstable
3 Otherwise : there is no conclusion (borderline case between stability
and instability; require an investigation of the higher order terms)
case of λm = 0 =⇒ existence of an equilibrium

vm = (vm,1 vm,2 , · · · , vm,n ) = fraction of the system in each dimension
D. J. Bicout
Take home messages
What do eigenvalues tell us about stability?
Eigenvalues allow
1 stability analysis
of linear
dynamical
systems
2 local stability
analysis of
nonlinear
dynamical
systems
D. J. Bicout
One Dimension (one variable): Linear Systems
Procedure for F (x) = ax + b
1 Determine the fixed point, x ∗ , solving F (x ∗ ) = 0

dF (x)
2 Construct the Jacobian, J = = a, constant
dx
3 Compute eigenvalues of J: solving J − λ = 0 =⇒ λ = J = a
4 As J = constant, stability or instability of of the whole system based

on the sign of λ: λ < 0 =⇒ the system is stable while λ ≥ 0 =⇒
the system is unstable
D. J. Bicout
One Dimension (one variable): Linear Systems
Example 1: Exponential decay
For the following kinetics described by the ODE below,
dC
= f (C ) = −kel C + R for t ≥ 0 , (10)
dt
with the initial condition: C (t = 0) = C0 .
R
Fixed point = Steady State: f (C ∗ ) = 0 =⇒ C ∗ =
kel
Eigenvalue: λ = −kel < 0 =⇒ stable system
C (t) = C ∗ + (C0 − C ∗ ) exp{−kel t}
D. J. Bicout
One Dimension (one variable): Non-Linear Systems
Procedure for F (x) 6= ax + b
1 Determine all fixed points, x ∗ , solving F (x ∗ ) = 0

dF (x)
2 Construct the Jacobian, J(x) =
dx
3 For each x , compute the eigenvalue: λ = J(x ∗ )
∗
4 Conclude on the stability or instability of each x∗ based on the real

parts of λ.
D. J. Bicout
Example 2: Modeling the population growth (P.-F. Verhulst, 1838)
Let N represents the population size, the

population growth is described by the
Verhulst-Pearl equation:

dN N
= rN 1 − (11)
dt K
where r defines the growth rate and K is the

carrying capacity. Setting x = N/K gives the
differential equation (logistic equation),
Pierre-Francois Verhulst
dx (1804 - 1849)
= rx(1 − x)
dt
D. J. Bicout
Example 2: The logistic growth
Consider the logistic differential equation:

dx
= F (x) = rx(1 − x) for t ≥ 0 , (12)
dt
with the initial condition: x(t = 0) = x0 .
Aim: use the linear stability analysis
D. J. Bicout
Example 2: The logistic growth: F (x) = rx(1 − x)
fixed points of F (x) Jacobian function

x1∗ = 0 dF
F (x ∗ ) = 0 =⇒ (13) J(x) = = r (1 − 2x) (14)
x2∗ = 1 dx
eigenvalues of J(x)
J(x1∗ ) = r ; x1∗ = 0

∗
det |J(x ) − λ 1| = 0 =⇒ λ = (15)
J(x2∗ ) = −r ; x2∗ = 1
D. J. Bicout
)(local) linear approximation of the logistic
d(x − x ∗ ) ; x → x1∗ = 0

r unstable
= λ(x − x ∗ ) ; λ = (16)
dt −r ; x → x2∗ = 1 stable
x → x1∗ = 0, unstable: exponential growth
x(t) = x0 e rt , (Malthus, 1798) (17)
x → x2∗ = xss = 1, stable: exponential decay

Thomas Robert Malthus
x(t) = xss + (x0 − xss )e −rt (18)
(1766 - 1834)
D. J. Bicout
x0 e rt
x(t) = (19)
1 + x0 (e rt − 1)
or,
KN0 e rt
N(t) = (20)
K + N0 (e rt − 1)
Example: N0 = 1, r = 1.0 and K = 1, 500.
D. J. Bicout
Two Dimensions (two variables): Linear Systems
Procedure for F(x) = J x
dx
1 Rewrite the system of ODE in the matrix form as, = Jx
dt
∂F(x)
2 The Jacobian, J = , is constant and has the form:
∂x

a b
J= (21)
c d
3 Compute eigenvalues of J as:
det |J − λ 1| = 0 =⇒ λ2 − (a + d)λ + ad − bc = 0 (22)

λ1 1 h p i
= a + d ± (a + d)2 − 4(ad − bc) (23)
λ2 2
4 As J = constant: <(of all λ) < 0 =⇒ the system is stable while
<(of at least one λ) ≥ 0 =⇒ the system is unstable
D. J. Bicout
Example 3: Pharmacokinetics

dC1
 dt = −(k12 + kel )C1 + k21 C2 + R



(24)

 dC2

 = k12 C1 − k21 C2
dt
Q: develop the linear stability analysis
D. J. Bicout
Example 3: Pharmacokinetics
matrix J Stability

−(k12 + kel ) k21

J= (25) λ1
k12 −k21 = <(λ) ± i=(λ) (26)
λ2
<(λ) ∼ − (k12 + kel + k21 ) (27)
=⇒ stable system
eigenvalues of J
1

λ1
q
=− k12 + kel + k21 ± (k12 + kel + k21 )2 − 4(k12 + kel )k21 + 4k21 k12 ) (28)
λ2 2
D. J. Bicout
Two Dimensions (two variables): Non-Linear Systems
Procedure for F(x) 6= J x

1 Determine all fixed point vectors, x∗ , solving F(x∗ ) = 0
∂F(x)
2 Construct the Jacobian matrix, J(x) =
∂x
3 For each x∗ , the Jacobian J(x∗ ) has the form:

a b
J(x∗ ) = (29)
c d
4 For each x∗ , compute eigenvalues of J(x∗ ) as:
det |J(x∗ ) − λ 1| = 0 =⇒ λ2 − (a + d)λ + ad − bc = 0 (30)

λ1 1 h p i
= a + d ± (a + d)2 − 4(ad − bc) (31)
λ2 2
5 <(of all λ) < 0 =⇒ x∗ is stable while
<(of at least one λ) ≥ 0 =⇒ x∗ is unstable
D. J. Bicout
Example 4: Lotka - Volterra Equation for Predator - Prey Systems
Let x and y represent the number of preys (e.g., rabbits)

and of predators (e.g., foxes), respectively. The population
dynamics is described by the system of Lotka-Volterra
equations: 
 dx
 dt = ax − bxy


(32)

 dy

 = −cy + dxy
dt
rabbit & fox
a (c) is the net growth (death) rate of preys (predators) in
the absence of predators (preys), b the rate of predation
affecting the prey population, and d the growth rate of
predators proportional to the food intake.
Lotka & Volterra
Q: develop the linear stability analysis

D. J. Bicout
fixed points of F(u∗ ), u∗ = (x ∗ , y ∗ )
 c 
d

ax − bxy = 0 0
 
∗ ∗
=⇒ u1 = and u2 =  (33)
 
−cy + dxy = 0 0

 a 
b
Jacobian J(x)
 
a − by −bx
J(x) =   (34)
dy −c + dx
D. J. Bicout
Jacobian J(x)
 
a − by −bx
J(x) =   (35)
dy −c + dx
Stability anlaysis close to u∗

1
 
a 0
∗  =⇒ (a − λ)(c + λ) = 0 =⇒ λ1 = a
J(u1 ) =  (36)
λ2 = −c
0 −c
λ1 > 0 and λ2 < 0 =⇒ u∗

1 is unstable
D. J. Bicout
Jacobian J(x)
 
a − by −bx
J(x) =   (37)
dy −c + dx
Stability anlaysis close to u∗

2
 bc 
0 − √
d

λ1 = −i
√ ac
 
∗ 2
J(u2 ) =   =⇒ λ + ac = 0 =⇒ (38)
 
 ad  λ2 = i ac
0
b
λ1 and λ2 are imaginary =⇒ u∗

1 is oscillatory
D. J. Bicout
Three Dimensions (three variables): Non-Linear Systems
Example 5: SIRS Epidemic model
dS


 = −βSI + γR



 dt


dI

= βSI − αI with S + I + R = N (39)


 dt


 dR = αI − γR



dt
Kermack & McKendrick
Q: develop the linear stability analysis and determine R0
D. J. Bicout
fixed points of F(x∗ ), x∗ = (S ∗ , I ∗ , R ∗ )
  
 −βSI + γR = 0 N
∗ ∗ ∗
βSI − αI = 0 =⇒ x1 =  0  and x2 6= x1 (40)
αI − γR = 0 0

Determination of R0
 N 
S∗ =
 2 R0 
 
 
γ R0 − 1 βN
 
∗  ∗  ∗
 I2 = α + γ
x2 =  N  where R0 = ; x2 > 0 =⇒ R0 > 1 (41)
 R0 
 α
 
 
α R0 − 1

R2∗ =
 
N
α+γ R0
D. J. Bicout
Jacobian J(x)
−βI −βS γ
 
 
J(x) =  βI βS − α 0 (42)
 

 
0 α −γ
Stability anlaysis close to x∗

1
0 −βN γ
 

   λ1 = 0
∗
J(x1 ) = 0 βN − α 0  =⇒ λ(γ + λ)(βN − α − λ) = 0 =⇒ λ2 = −γ (43)
 
λ3 = α(R0 − 1)
  
0 α −γ
x∗
1 is stable for R0 < 1 and instable for R0 > 1
D. J. Bicout

Jacobian J(x)
−βI −βS γ
 
 
J(x) =  βI βS − α 0 (44)
 

 
0 α −γ
Stability anlaysis close to x∗

2 (in the limit R0 > 1)
∗
−βI −α γ
  
2
λ1 = 0

 

   
 ∗
 J(x2 ) =  βI2 ∗
0 0 
 1 h q i
γ + βI2∗ + (γ + βI2∗ )2 − 4αβI2∗
   
 λ2 = −
 


2
 
=⇒ (45)
 0 α −γ 
1 h

 
 q i
γ + βI2∗ − (γ + βI2∗ )2 − 4αβI2∗
 


 h i  λ3 = −

 =⇒ λ λ2 + (γ + βI ∗ )λ + αβI ∗ = 0
 2
2 2
λ2 and λ3 are < 0 and λ1 = 0 =⇒ x∗

2 is stable and steady state
D. J. Bicout

Bicout Modelisation M1 2013

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Bicout Modelisation M1 2013

Transféré par

Droits d'auteur :

Formats disponibles

Linear Stability Analysis

Biomathématiques et Epidémiologies, EPSP - TIMC, UMR 5525,

UE MMED4100, Joseph Fourier - Grenoble University

Consider the following ordinary differential equation (ODE):

What can we learn out of this?

general solution of Eq.(1)

stability: eigenvalue a steady state (for a < 0)

What happen when F (x) is non-linear?

How to deal with higher dimensional systems where F(x) is

n−dimensional Dynamical System

fixed points Jacobian matrix

 ∂F1 ∂F1 ∂F1 

Linear approximation (close to fixed points)

Setting δx = x − x∗ , the Eq.(3) becomes,

ai are constants determined by using the initial conditions.

Take home messages

Procedure for stability analysis at fixed points

1 Determine the fixed point vector, x∗ , solving F(x∗ ) = 0

4 Stability or instability of x∗ based on the real parts, <(λ), of

Take home messages

What do eigenvalues tell us about stability?

stability of x∗ based on the real parts of eigenvalues

1 all eigenvalues have real parts less than zero =⇒ x∗ is stable

case of λm = 0 =⇒ existence of an equilibrium

Take home messages

What do eigenvalues tell us about stability?

One Dimension (one variable): Linear Systems

Procedure for F (x) = ax + b

1 Determine the fixed point, x ∗ , solving F (x ∗ ) = 0

4 As J = constant, stability or instability of of the whole system based

One Dimension (one variable): Linear Systems

Example 1: Exponential decay

For the following kinetics described by the ODE below,

One Dimension (one variable): Non-Linear Systems

Procedure for F (x) 6= ax + b

1 Determine all fixed points, x ∗ , solving F (x ∗ ) = 0

4 Conclude on the stability or instability of each x∗ based on the real

One Dimension (one variable): Non-Linear Systems

Example 2: Modeling the population growth (P.-F. Verhulst, 1838)

Let N represents the population size, the

where r defines the growth rate and K is the

One Dimension (one variable): Non-Linear Systems

Example 2: The logistic growth

Consider the logistic differential equation:

Aim: use the linear stability analysis

One Dimension (one variable): Non-Linear Systems

Example 2: The logistic growth: F (x) = rx(1 − x)

fixed points of F (x) Jacobian function

One Dimension (one variable): Non-Linear Systems

Example 2: The logistic growth

)(local) linear approximation of the logistic

x → x1∗ = 0, unstable: exponential growth

x(t) = x0 e rt , (Malthus, 1798) (17)

x → x2∗ = xss = 1, stable: exponential decay

One Dimension (one variable): Non-Linear Systems

Example 2: The logistic growth

general solution of Eq.(12)

Two Dimensions (two variables): Linear Systems

Procedure for F(x) = J x

Two Dimensions (two variables): Linear Systems

Q: develop the linear stability analysis

Two Dimensions (two variables): Linear Systems

Two Dimensions (two variables): Non-Linear Systems

Procedure for F(x) 6= J x