03 Dynamical Systems Tutorial

Dynamical systems
tutorial
Gregor Schner, INI, RUB
Dynamical systems: Tutorial
the word dynamics

time-varying measures

range of a quantity

forces causing/accounting for movement => dynamical

systems

dynamical systems are the universal language

of science

physics, engineering, chemistry, theoretical biology,
economics, quantitative sociology, ...
time-variation and rate of change
variable x(t);

rate of change dx/dt
dynamical system
dx/dt = f(x)
x
dynamical system
dx/dt=f(x)
x
dynamical system
present determines the future

given initial condition

predict evolution (or predict the past)
dx/dt=f(x)
predicts
future initial
evolution condition x
dynamical systems
x: spans the state space (or phase space)

f(x): is the dynamics of x (or vector-field)

x(t) is a solution of the dynamical systems to
the initial condition x_0

if its rate of change = f(x)

and x(0)=x_0
numerics
x = f (x)
sample time
discretely

ti = i t; xi = x(ti)
compute
dx x xi+1 xi
solution by x = =
iterating through dt t t
xi+1 = xi + t f (xi)
time
[forward Euler]
linear dynamics
=> simulation
attractor
fixed point, to which neighboring initial conditions

converge = attractor
dx/dt=f(x)
attractor
fixed point
is a constant solution of the dynamical system

x = f (x)
x = 0 f (x0) = 0
stability
mathematically really: asymptotic stability

defined: a fixed point is asymptotically stable,
when solutions of the dynamical system that
start nearby converge in time to the fixed
point
stability
the mathematical concept of stability (which

we do not use) requires only that nearby
solutions stay nearby

Definition: a fixed point is unstable if it is not
stable in that more general sense,

that is: if nearby solutions do not necessarily stay nearby (may
diverge)
linear approximation near attractor
dx/dt = f(x)
non-linearity as a small
perturbation/
deformation of linear
system
x
=> non-essential non-

linearity
stability in a linear system
dL/dt=f(L)
if the slope of the
linear system is
negative, the fixed
point is L
(asymptotically
stable)
dL/dt=f(L)
if the slope of the
linear system is
positive, then the L
fixed point is
unstable
dL/dt=f(L)
if the slope of the linear
system is zero, then the
system is indifferent
(marginally stable: stable L
but not asymptotically
stable)
stability in linear systems
generalization to multiple dimensions

if the real-parts of all Eigenvalues are negative: stable

if the real-part of any Eigenvalue is positive: unstable

if the real-part of any Eigenvalue is zero: marginally stable in that

direction (stability depends on other eigenvalues)
stability in nonlinear systems
stability is a local property of the fixed point

=> linear stability theory

the eigenvalues of the linearization around the fixed point determine
stability

all real-parts negative: stable

any real-part positive: unstable

any real-part zero: undecided: now nonlinearity decides (non-

hyberpolic fixed point)
dL/dt = f(L)
all real-parts negative: stable

L
any real-part positive: dL/dt = f(L)

unstable
L
dL/dt = f(L) dL/dt = f(L)
L
any real-part zero:
undecided: now
nonlinearity decides dL/dt = f(L) dL/dt = f(L)
(non-hyberpolic fixed
point)
L L
bifurcations
look now at families of dynamical systems, which

depend (smoothly) on parameters

ask: as the parameters change (smoothly), how do
the solutions change (smoothly?)

smoothly: topological equivalence of the dynamical systems at
neighboring parameter values

bifurcation: dynamical systems NOT topological equivalent as

parameter changes infinitesimally
bifurcation
dx/dt=f(x)
ATTRACTOR ATTRACTOR
bifurcation
bifurcation=qualitative change of dynamics (change in
number, nature, or stability of fixed points) as the
dynamics changes smoothly
dx/dt=f(x)
ATTRACTOR ATTRACTOR
tangent bifurcation
the simplest bifurcation (co-dimension 0): an attractor collides
with a repellor and the two annihilate
dx/dt=f(x)
ATTRACTOR ATTRACTOR
local bifurcation
dx/dt=f(x)
ATTRACTOR ATTRACTOR
reverse bifurcation
changing the dynamics in the opposite direction
dx/dt=f(x)
ATTRACTOR ATTRACTOR
bifurcations are instabilities
that is, an attractor becomes unstable before

disappearing

(or the attractor appears with reduced stability)

formally: a zero-real part is a necessary condition
for a bifurcation to occur
tangent bifurcation
normal form of tangent bifurcation

!x = x2
(=simplest polynomial equation whose flow is

topologically equivalent to the bifurcation)
p
dx/dt fixed point x0 =
stable
A positive
x A
unstable
A =0
A negative
Hopf theorem
when a single (or pair of complex conjugate)

eigenvalue crosses the imaginary axis, one of four
bifurcations occur

tangent bifurcation

transcritical bifurcation

pitchfork bifurcation

Hopf bifurcation
transcritical bifurcation
normal form x = x x2
dx/dt fixed point
A negative A positive stable A

x
unstable
A =0
pitchfork bifurcation
normal form x = x x3
fixed point
dx/dt
p
x = 2x0x = 2 x
x stable A
unstable
A positive
A negative A =0
Hopf: need higher dimensions
2D dynamical system:
vector-field
x2
x1 = f1(x1, x2)
x2 = f2(x1, x2)
x1
vector-field
x2
initial
condition
x1 = f1(x1, x2)
x2 = f2(x1, x2)
x1
fixed point, stability, attractor
x2
initial
condition
x1 = f1(x1, x2)
x2 = f2(x1, x2)
x1
Hopf bifurcation
r = r r3
normal form =!
dr/dt
x
r
stable A
A
A
A =0 unstable
d F /dt y
F
higher bifurcations
e.g., degenerate (non-linear terms

simultaneously zero with real-part of an
eigenvalue)

e.g. higher co-dimension ...
the center manifold
at bifurcation: real part of one z

eigenvalue=0

corresponding eigenvector: critical direction
cen
man
near bifurcation: real part of one

x
eigenvalue close to zero

ATTRACTORABOUT
=> Center manifold theorem y
TOBECOMEUNSTABLE
center manifold
= manifold which is locally z

tangent to eigenvector of
zero real part eigenvalue

center
CM theorem: dynamics manifold
within the CM is
x
topologically equivalent to
original dynamics
ATTRACTORABOUT
y
TOBECOMEUNSTABLE
=> enormous dimensional
reduction
center manifold
=> the 4 elementary bifurcations are fair

representatives of any dynamical system with
the same qualitative flow
forward dynamics
given known equation, determined fixed points /

limit cycles and their stability

more generally: determine invariant solutions (stable, unstable
and center manifolds)

use CMF to simplify dynamics

inverse dynamics
given classification of solutions (stable states)

and their dependence on parameters,
determine the dynamical system! Thats the
modelers job

practical approach

identify the class of dynamical systems using the 4 elementary
bifurcations

and use normal form to provide an exemplary representative

of the equivalence class of dynamics

03 Dynamical Systems Tutorial

Uploaded by

Copyright:

Available Formats

03 Dynamical Systems Tutorial

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

03 Dynamical Systems Tutorial

Uploaded by

Copyright:

Available Formats

Dynamical systems

forces causing/accounting for movement => dynamical

dynamical systems are the universal language

predict evolution (or predict the past)

x: spans the state space (or phase space)

fixed point, to which neighboring initial conditions

is a constant solution of the dynamical system

mathematically really: asymptotic stability

the mathematical concept of stability (which

=> non-essential non-

generalization to multiple dimensions

if the real-part of any Eigenvalue is positive: unstable

if the real-part of any Eigenvalue is zero: marginally stable in that

stability is a local property of the fixed point

all real-parts negative: stable

any real-part positive: unstable

any real-part zero: undecided: now nonlinearity decides (non-

all real-parts negative: stable

any real-part positive: dL/dt = f(L)

look now at families of dynamical systems, which

bifurcation: dynamical systems NOT topological equivalent as

changing the dynamics in the opposite direction

that is, an attractor becomes unstable before

(=simplest polynomial equation whose flow is

when a single (or pair of complex conjugate)

dx/dt fixed point

A negative A positive stable A

e.g., degenerate (non-linear terms

at bifurcation: real part of one z

= manifold which is locally z

=> the 4 elementary bifurcations are fair

given known equation, determined fixed points /

use CMF to simplify dynamics

given classification of solutions (stable states)

and use normal form to provide an exemplary representative

You might also like