Anda di halaman 1dari 78

10/25/2014

SE517Nonlinear
Systems
Lecture2
DynamicBehavior

Example3:HIVDrug
Administration

F.Doyleetal./JournalofProcessControl17
(2007)571594

10/25/2014

Example4:PredatorPreyModel
VolterraLokta PredatorPreyModelforSmallfishintheAdriatic
M1

M2

Example5:SteeringProblem
BicycleModel

ChihLyang etal.2009

10/25/2014

Example6:Vectoredthrustaircraft

Example7:InternetCongestion
Control

10/25/2014

Objectives
Studythedynamicbehaviorofnonlinearsystems.
Introducetheconceptofequilibriumpoint,stability,andlimit
cycles.
Learnhowtosolvedifferentialequations.
Learnhowtoconstructthephaseportrait.
Learnhowtoevaluatethestabilityofthesolution.
Understandthedifferencebetweenglobalbehaviorandlocal
behavior.

ExpectedOutcomes
Usetoolstostudythedynamicbehaviorofnonlinearsystems.
Grasptheconceptofequilibriumpoint,stability,andlimit
cycles.
Solvedifferentialequations.
Constructthephaseportrait.
Evaluatethestabilityofthesolutionofadifferentialequation.
Differentiatebetweenglobalbehaviorandlocalbehavior.

10/25/2014

Structure
Recall/usesomeofthemotivatingexamples.
Solutionofdifferentialequations.
Useofmodernsoftwaretoolstosolvesuchequations.
Examples.
QualitativeAnalysis
Phaseportrait,equilibriumpoints,andlimitcycles.
Examples
Stability
Definition.
Stabilityoflinearsystemsandstabilityanalysisvialinearapproximation.
Lyapunov stabilityanalysis.
Parametricandnonlocalbehavior.
Examples

GeneralStateSpaceModel

InGeneralifthesystemisnonlinear

10/25/2014

GeneralStateSpaceModel
Define

then

Introduction(Summaryof
PreviousLecture)
StatespaceModelsaresystemsofODE.
StatespaceModelcanbeeither
Linearornonlinear.
TimevariantorTimeinvariant.
AutonomousorNonautonomous
DeterministicorStochastic
Themodelcancontainlags(delaysattheleveloftheinputor
output)
Themodelcancontaindelaysatthelevelofthestates.

10/25/2014

Objectives
Studythedynamicbehaviorofnonlinearsystems.
Introducetheconceptofequilibriumpoint,stability,andlimit
cycles.
Learnhowtosolvedifferentialequations.
Learnhowtoconstructthephaseportrait.
Learnhowtoevaluatethestabilityofthesolution.
Understandthedifferencebetweenglobalbehaviorandlocal
behavior.

ExpectedOutcomes
Usetoolstostudythedynamicbehaviorofnonlinearsystems.
Grasptheconceptofequilibriumpoint,stability,andlimit
cycles.
Solvedifferentialequations.
Constructthephaseportrait.
Evaluatethestabilityofthesolutionofadifferentialequation.
Differentiatebetweenglobalbehaviorandlocalbehavior.

10/25/2014

Comparison:StateSpace
NonlinearSystem
Representation
LinearSystem

1.PhenomenaofNonlinear
Dynamics
Linearvs.Nonlinear
Input

Output
System
state,

Definitions :

Linear : when the superposition holds


Nonlinear : otherwise

10/25/2014

Linearity
Asystemissaidtobelinearifitsatisfiesthesuperposition
theorem(addition)andalsohomogeneity
Ify1istheoutputduetou1
Ify2istheoutputofu2
Letu=u1+u2thenewoutputy=y1+y2.
Letu= u1thenthenewoutputy= y1

Superposition
* Superposition

Sys.
=

Sys.

Sys.

Is the system linear ?

So is it linear? No, under zero initial conditions only.

10/25/2014

Linearity
What is the linearity when

A mnemonic rule for linear system :All functions in RHS of a differential


equation are linear. System is linear at least at zero input or zero initial
condition
Ex:

Timeinvariantvs.Timevarying
Timeinvariantvs.Timevarying
System(1)istimeinvariant parametersareconstant
Lineartimevaryingsystem

System(2)istimeinvariant nofunctionhastasitsargument.
Nonlineartimevaryingsystem

10

10/25/2014

Autonomous&Non
Autonomous
Timeinvariantsystemarecalledautonomousandtime
varyingarecallednon autonomous.Inourbook,
autonomousisreservedforsystemswithnoexternalinput,
i.e.,
Ex:

Thusautonomousaretimeinvariantsystemswithnoexternal
input.

Summary
StatespaceModelsaresystemsofODE.
StatespaceModelcanbeeither
Linearornonlinear.
TimevariantorTimeinvariant.
AutonomousorNonautonomous
DeterministicorStochastic
Themodelcancontainlags(delaysattheleveloftheinputor
output)
Themodelcancontaindelaysatthelevelofthestates.

11

10/25/2014

Stability&Outputofsystems
Stabilitydependsonthesystemsparameter(linear)
Stabilitydependsontheinitialconditions,inputsignalsas
wellasthesystemparameters(nonlinear).
Outputofalinearsystemhasthesamefrequencyasthe
inputalthoughitsamplitudeandphasemaydiffer.
Outputofanonlinearsystemusuallycontainsadditional
frequencycomponentsandmay,infact,notcontainthe
inputfrequency.

EquilibriumPoint
EquilibriumPoint
Westartwithanautonomoussystem.

Definition:
isanequilibriumpoint(orasteadystate,
orasingularpoint)

Ifdet(A)0,theautonomoussystemhasauniqueequilibrium
point,
(LinearSystem).

12

10/25/2014

Timevariantsystems
x*issaidtobeanequilibriumpointfor

If

Comparison:Equilibriumrelations
Example1:MassDamper

13

10/25/2014

Example2:Pendulum

Example3:PredatorPreyM1

14

10/25/2014

PredatorPreyM2

Comparison:
LinearSystem

EquilibriumLinearSystem

NonlinearSystem

EquilibriumnonlinearSystem

15

10/25/2014

Simulation
mass_spring_sim.

Pendulum_sim.m

clc

clc

closeall

closeall

clearall

clearall

globalmkgrB

globallmgFB

m=1;k=2;g=9.8;r=0;B=0.5;

l=2;m=1;B=2;g=9.8;F=2;

X0=[2,8];

X0=[pi/4,0];

[t,Y]=ode45(@mass_spring,[050],
X0);

[t,Y]=ode45(@pendulum,[050],X0);

figure(1)
plot(t,Y(:,1))
grid
figure(2)
plot(Y(:,1),Y(:,2))

figure(1)
plot(t,Y(:,1))
grid
figure(2)
plot(Y(:,1),Y(:,2))
grid

grid

ODE45Code
function
[dXdt]=mass_spring(t,X)

function[dXdt]=pendulum(t,X)

globalmgrkB
dXdt(1,1)=X(2);
dXdt(2,1)=k/m*X(1)
B/m*X(2)+r/m;

dXdt(1,1)=X(2);

globallmgFB

ifabs(sin(X(1)))<0.000001,
temp=0;
else
temp=sin(X(1));
end
dXdt(2,1)=g/l*temp
B/(m*l^2)*X(2)+F/(m*l);

16

10/25/2014

AdditionalrealExample:Tunnel
Diode

Applicationsfortunneldiodesincludedlocal
oscillatorsforUHFtelevisiontuners,trigger
circuitsinOscilloscopes,

17

10/25/2014

h(x1)
Step1:computetheequilibriumpoint

LetsSummarize
LinearSystem

NonlinearSystem

UniqueequilibriumPoint

Multipleequilibriumpoints.

StableLinearsystemunder
harmonicinputproducesan
outputwiththesame
frequency.

Nonlinearsystemunder
harmonicinputproducesan
outputcontainingharmonics
andsubharmonics

Singlemodeofbehavior

Multiplemodesofbehavior.

18

10/25/2014

Summary
LinearSystem

NonlinearSystem

UniqueequilibriumPoint

Multipleequilibriumpoints.

StableLinearsystemunder
harmonicinputproducesan
outputwiththesame
frequency.

Nonlinearsystemunder
harmonicinputproducesan
outputcontainingharmonics
andsubharmonics

Singlemodeofbehavior

MultiplesteadyStatemodes
ofbehavior.

Infiniteescapetime.

Finiteescapetime.

Solutionofthe
DynamicEquations

19

10/25/2014

LinearAutonomousSystems
linearautonomoussystem
where
For1dimsys.

For2dimsys.

LinearAutonomousSystems
(Contd.)

20

10/25/2014

SolutionofLinearsystems
Solutionalwaysexistslocally.
Solutionalwaysexistsglobally.
Solutionisuniqueeachinitialconditionproducesadifferenttrajectory.
Solutioniscontinuouslydependentoninitialconditionsforeveryfinitet,

Equilibriumpointisunique(whendet A0).

LinearSystemPeriodicSolution
Ifthereisoneperiodicsolution,thereisaninfiniteset
ofperiodicsolutions. (Thereisnoisolatedclosedsolution.)
Ex:

( manyperiodicsolutions,w.r.t.I.C.)

21

10/25/2014

Non linearAutonomousSystem
Basicallyeverything.

Asolutionmaynotexist,evenlocally.

Herethesolutionischattering,becauseTherefore,nodifferential
functionsatisfyingtheequationexists.

Solutions
Solutionmaynotexistglobally.

Assume

finiteescapetime
(= :linearsystem)

Solutionmaynotbeunique.

22

10/25/2014

Example:
Considerthefollowingsystem

Apossiblesolutionis

ForK=1

Example:Escapetime

Consideranonlinearspring

Thenthemassspringmodel
becomes
www.tokyomodel.com.hk/ecshop/goods.php
?id=2241)

Asolutionforthissystem
willbedonenumericallylateron

ButifweassumeMverysmall;r(t)=0andy(0)=1,then
andisapossiblesolution.

23

10/25/2014

Simulation
clc
%closeall
clearall
globalmkgrB
m=1;k=0.2;g=9.8;r=1;B=0.5;
X0=[2.20];%initialvalue

[t,Y]=ode45(@mass_nonlinear
_spring,[050],X0);
figure(2)
plot(t,Y(:,1))
grid
figure(1)
plot(Y(:,1),Y(:,2))
grid

PeriodicSolutions
Nonlinearsystemmayhaveisolatedclosed(periodic)
solutions. Ex:

24

10/25/2014

Isolatedclosedsolution

Isolatedclosedsolution( onlyoneperiodicsolution.) Isolatedattractive


periodicsolution

Chaos
Forstablelinearsystems,smalldifferencesininitialconditions
canonlycausesmalldifferencesinoutput.
Nonlinearsystems,however,candisplayaphenomenoncalled
chaos,bywhichwemeanthatthesystemoutputisextremely
sensitivetoinitialconditions.(Chaoticregimes nonperiodic,bounded
behavior)

Example:

25

10/25/2014

Pitchforkbifurcation
Example:Undamped Duffings equation

Asthecoefficientvariesfrompositiveto
negative,one
equilibriumpointsplitsintothreepoints(xe =0,
,)

AHopf bifurcation
involvestheemergenceoflimitcyclesasparametersare
changed

26

10/25/2014

VideoLifeandChaos

PhasePortraits

27

10/25/2014

Qualitativeanalysis
Limitedtosecondorder
systems(planarsystems)
PhasePortraits
VectorField
Isocline

Equilibriumpoints
Stablevs unstable.

LimitCycles
Stablefocus
Unstablefocus
Center

Stability
Linearsystem
Lyapunov

ParametricandNonlocal
Behavior.

Phaseportraits
General2nd ordersystemStateSpaceModelformulation.

28

10/25/2014

VectorField
x2

X
x1

Example:supposewehave

VectorField

Example:supposewehave

29

10/25/2014

PhasePortrait:Isocline Method

example

30

10/25/2014

Isocline:Pendulumwithout
friction

31

10/25/2014

CaseB=0

CaseB0

32

10/25/2014

ExamplesofPhasePortrait

Example:
Consideranonlinearspring
Thenthemassspringmodel
becomes
www.tokyomodel.com.hk/ecshop/goods.php
?id=2241)

33

10/25/2014

PhasePortrait

PhasePlaneAnalysisofLinear
Systems
Thegeneralformofalinearsecondordersystemis

Whichcanbewrittenas
Oringeneral
Thecharacteristicequationofthissystemisgivenby

34

10/25/2014

4cases
Polesarebothrealandthesamesign.
Polesarebothrealbuthaveoppositesigns.
Polesarecomplexconjugatewithnonzerorealparts.
Polesarecomplexconjugateswithrealpartsequaltozero.

bothrealandhavethesamesign
(positiveornegative)
Stableorunstablenode

35

10/25/2014

Example

Polesarebothrealbuthave
oppositesigns.
Saddlepoint

36

10/25/2014

Example

Example:Repeatedeigenvalues

37

10/25/2014

Polesarecomplexconjugatewith
nonzerorealparts
Stableorunstablefocus

Polesarecomplexconjugateswith
realpartsequaltozero
Centerpoint

38

10/25/2014

Localbehaviorofnonlinear
systems
Linearizing thesystemaroundtheequilibriumpointsleadsto

InthevicinityoftheequilibriumpointtheH.O.Tg1andg2are
negligible.
thenonlinearsystemtrajectoriesessentiallysatisfythe
linearized equation

ExistenceofLimitCycles
LetNbethenumberofnodes,centers,andfocienclosedbyalimit
cycle.
LetSthenumberofenclosedsaddlepoints.
Theorem2.1(Poincarealsocalledindextheorem.)Ifalimitcycle
existsinthesecondorderautonomoussystem

thenN=S+1.
Onesimpleinferencefromthistheoremisthatalimitcyclemust
encloseatleastoneequilibriumpoint.

39

10/25/2014

Limitcycles
Theorem2.2(PoincareBendixson)Ifatrajectoryofthesecond
orderautonomoussystemremainsinafiniteregionQ,then
oneofthefollowingistrue:
(a)thetrajectorygoestoanequilibriumpoint
(b)thetrajectorytendstoanasymptoticallystablelimitcycle
(c)thetrajectoryisitselfalimitcycle

Limitcycles
Theorem2.3(Bendixson):
Forthenonlinearsystem(2.1),nolimitcyclecanexistina
regionQ.ofthephaseplaneinwhich

doesnotvanishanddoesnotchangesign.

40

10/25/2014

Example:
Considerthenonlinearsystem

Limitcycles
alimitcycleisdefinedasanisolatedclosedcurve.The
trajectoryhastobebothclosed,indicatingtheperiodicnature
ofthemotion,andisolated,indicatingthelimitingnatureofthe
cycle(withnearbytrajectoriesconvergingordivergingfromit).
Threekindsoflimitcycles

41

10/25/2014

ExamplePendulumwithout
Friction

PendulumwithFriction

42

10/25/2014

Examples
Considerthefollowingnonlinearsystems

Aretheystable,unstable,andsemistablelimitcycles?

StabilityinthesenseofLyapunov
Forthenonlinearmassspring,weobservethatifwestartwith
aninitialconditionclosetotheequilibriumpoint,thetrajectory
staysclosetothatpoint.
Inthiscasewesaythattheequilibriumpointisstable.
Iftheinitialconditionisnottheequilibriumpoint,wesaythat
thesolutionisstable.
Definition:Letx(t;a)beasolutiontothedifferentialequation
withinitialconditiona.Asolutionisstableifothersolutions
thatstartneara stayclosetox(t;a).Formally,asolutionx(t;a)
isstableifforall>0,thereexistsa>0suchthat

43

10/25/2014

AsymptoticStability
Asolutionx(t;a)isasymptoticallystableifitisstableinthe
senseofLyapunov andalsox(t;b)x(t;a)astforb
sufficientlyclosetoa.

UnstableSolutions
Asolutionx(t;a)isunstableifitisnotstable.
A solutionx(t;a)isunstableifgivensome >0,theredoesnot
exista>0suchthatif||ba||<,then||x(t;b)x(t;a)||
< forallt.

44

10/25/2014

Stabilityoflinearsystem
Recallalinearsystemisdescribedbythefollowingstatespace
model

GeneralTheoremforLTISystem

45

10/25/2014

ExponentialStability

Example:LinearMassSpring
example

Theeigenvalues ofAaretherootsofthefollowingequation
characteristic
polynomial

WheresisacomplexnumberandIistheidentitymatrix
havingthesamedimensionsasA.
Examplewillcontinuein
Class

46

10/25/2014

Toolstostudythesignsofthe
eigenvalues
Matlab:>>eig(A)
RouthHurwitzCriteria(review)

SummaryStabilityAnalysisvia
Linearapproximation
Ideaistoapproximatethesystemnonlineardynamicsbya
linearsystemaroundtheequilibriumpointusingTaylorseries
oforder1.(H.O.T removed)
Theobtainedsystemisnowlinearandtheoremcanbeapplied
Taylorseriesreview(willbedoneinclass)
Example(Nonlinearmassspringsystemwillbedoneinclass).

47

10/25/2014

Lyapunov Stability(Stabilityof
nonlinearsystem)
Definitions:
AcontinuousfunctionVispositivedefiniteif

Afunctionisnegativedefiniteif
AcontinuousfunctionVispositivesemidefinite (negativesemidefinite)if
itispositivedefinite(negativedefinite)andV(x)=0forx0

Lyapunov StabilityTheorem
LetVanonnegativeLyapunov function

IfVsatisfiestheseconditions,VissaidtobealocalLyapunov
function.

48

10/25/2014

Geometric interpretation

49

10/25/2014

Step2:Rewritethedynamicequationaroundanequilibrium
point.

Theoriginistheequilibriumpointforthesesystems

50

10/25/2014

Step3:SelectaLyapunov function.

V(z)>0;V(z)=0atx=0.V(z)ispositivedefinite
Step4:ComputedV(z)/dt
ContinueinClass

ContinueinClass

51

10/25/2014

Example:Pendulum
Recallthestatespacemodelforpendulum

Lyapunov functionasenergyfunction:

GlobalStability

52

10/25/2014

Example:firstorder
a
a

Whatif

Example1:

53

10/25/2014

Example2:

Example3:

54

10/25/2014

Lyapunovs Theorem
Theoriginisstableifthereisacontinuouslydifferentiable
positivedefinitefunctionV(x)sothatdV(x)/dt isnegative
semidefinite,anditisasymptoticallystableifdV(x)/dt is
negativedefinite.
Itisgloballyasymptoticallystableiftheconditionsfor
asymptoticstabilityholdgloballyandV(x)isradially
unbounded

Lyapunov surface
AcontinuouslydifferentiablefunctionV(x)satisfyingthe
conditionsforstabilityiscalledaLyapunov function.Thesurface
V(x)=c,forsomec>0,iscalledaLyapunov surfaceoralevel
surface

55

10/25/2014

Lyapunov forLinearSystem
Supposethat

ApossibleLyapunov functionis
Quadraticform
WhenPissymmetricpositivedefinitematrixP>0thenV(x)>0

IfQ>0,thenAisHurwitz
Thus,tofindaLyapunov functionforalinearsystemitis
sufficienttochooseaQ>0andsolvetheLyapunov equation.
ForlinearsystemwearealwaysabletoselectQ>0andfindP>0
thatsatisfiesthefollowingequation:

56

10/25/2014

LinearSystemStability

QuadraticForms

P0(Positivesemidefinite)ifandonlyifeigenavlues (P)0
P>0(Positivedefinite)ifandonlyifeigenavlues (P)>0
V(x)ispositivedefiniteifandonlyifPispositivedefinite
V(x)ispositivesemidefinite ifandonlyifPispositivesemidefinite
P>0ifandonlyifalltheleadingprincipalminorsofParepositive

57

10/25/2014

PositiveDefiniteMatrix
Asquaresymmetricrealmatrixispositivedefinite(positive
semidefinite)ifandonlyifallofitseigenvalues are>0( 0)
Example:
A1=[2.33330.6667
0.66670.6667]
A2=[2.33330.6667
0.66670.6667]
A3=[0.33330.6667
0.66671.3333];

Example
isthefollowingmatrixpositivesemidefinite?

58

10/25/2014

Example:
RecalltheLTI massspringsystem

LetTheproblemistofindP(pis)suchthat

Example

59

10/25/2014

Example
Consider

Lyapunov AnalysisofLinearTime
InvariantSystems
Givenalinearsystemoftheformletusconsidera
quadraticLyapunov functioncandidate
ThereexistasymmetricmatrixQ

60

10/25/2014

Example
Considerthefollowingsystem

Computetheeigenvalues
AssumeP=IcomputeQ.
ThematrixQisnotpositivedefinite.Therefore,
noconclusioncanbedrawnfromtheLyapunov
functiononwhetherthesystemisstableornot

Stability
chooseapositivedefinitematrixQ
solveforPfromtheLyapunov equation
checkwhetherPisp.d.
IfP.isp.d.thesystemisgloballyasymptoticallystable.
Theorem3.6AnecessaryandsufficientconditionforaLTl system
tobestrictlystableisthat,foranysymmetricp.d.matrixQ,the
uniquematrixPsolutionoftheLyapunov equationbesymmetric
positivedefinite.

61

10/25/2014

Example
Let

and

UsingMatlab:
Lyap
lyap solvesthespecialandgeneralformsoftheLyapunov
matrixequation.
P=lyap(A,Q)

62

10/25/2014

Lyapunov Functionfornonlinear
system
Step1findtheequilibriumpoint
UseofTaylorseriesaroundtheequilibriumtofind:

SincewearearoundtheequilibriumF(0)=0andcontains
H.O.T

Example

Matlab/SIMULINK
>>doctrim
>>doclinmod
>>docdlinmod

63

10/25/2014

Example2:revisited

Remarks
AsystemcanhavemanyLyapunov Functions.IfVisaLyapunov Functionthan

Foraparticularsystem,aspecificchoiceofVmayleadtomorepreciseresultsthan
others.
Iftheequilibriumpointisstableforthelinearized systemthenitisstableforthe
nonlinearone.
Iftheequilibriumisunstablethanitisunstableforthenonlinearsystem
Ifitismarginallystablewecannotconclude.

64

10/25/2014

Example:PendulumwithFriction
Anewtermfunctionofx2 isaddedsinceBisdifferentfromzero

65

10/25/2014

Summary
Thedynamicbehaviorofnonlinearsystemsisveryrich.
Anonlinearsystemcanhavemanyequilibriumpoints,different
stabilitycharacteristicsandlimitcycles.
Qualitativeanalysisofthesolutionofdifferentdifferentialequations
canbedonegraphicallyusingPhaseportrait.
Isocline andvectorfieldcangiveaninsightonthephaseportrait.
Lyapunov stabilitycanbeusedtoanalyzethestabilityofasolution.
Lyapunov functionisnotunique.
Alocalstabilitycannotdeterminetheglobalbehavior.

InvariantSets
Slotine andLiChapter3

66

10/25/2014

Invariantsetforadynamicsystem
AsetGisaninvariantsetforadynamicsystemifeverysystem
trajectorywhichstartsfromapointinGremainsinGforallfuture
time.
Example:
anyequilibriumpointisaninvariantset.
Thedomainofattractionofanequilibriumpointisalsoaninvariant
set.
Atrivialinvariantsetisthewholestatespace.
Foranautonomoussystem,(dx/dt=f(x))anyofthetrajectoriesin
statespaceisaninvariantset.
Sincelimitcyclesarespecialcasesofsystemtrajectories(closed
curvesinthephaseplane),theyarealsoinvariantsets.

Theorem3.4(Slotine andLi)
(LocalInvariantSetTheorem)
Consideranautonomoussystemwithfcontinuous,andletV(x)
beascalarfunctionwithcontinuousfirstpartialderivatives.
Assumethatforsomel>0,theregionl definedbyV(x)<lis
bounded
dV(x)/dt 0forallxinl
LetRbethesetofallpointswithinl wheredV(x)/dt =0,and
MbethelargestinvariantsetinR.Then,everysolutionx(t)
originatinginl tendstoMast

67

10/25/2014

Corollary:Localinvariantsets
Considertheautonomoussystemdx/dt=f(x),withfcontinuous,and
letV(x)beascalarfunctionwithcontinuouspartialderivatives.
Assumethatinacertainneighborhood oftheorigin
V(x)islocallypositivedefinite
dV/dt isnegativesemidefinite
ThesetRdefinedbydV(x)/dt =0containsnotrajectoriesother
thanthetrivialtrajectoryx=0

Then,theequilibriumpoint0isasymptoticallystable.
Furthermore,thelargestconnectedregionoftheforml
(definedbyV(x)<l)within isadomainofattractionof
theequilibriumpoint.

68

10/25/2014

Example:Domainofattraction
Considerthesystem:Showthat0isasymptoticallystable.

Practice
Canwesaythesamefora?

69

10/25/2014

Example: AttractiveLimitCycle
Isaninvariantset

Themotiononthisinvariantsetisdescribed
(equivalently)by

Remarks
Theabovecorollaryreplacesthenegativedefinitenessconditionon
dV/dt inLyapunov's localasymptoticstabilitytheorembyanegative
semidefinitenessconditionondV/dt combinedwithathirdcondition
onthetrajectorieswithinR.
Thelargestconnectedregionoftheform;within l isadomainof
attractionoftheequilibriumpoint,butnotnecessarilythewhole
domainofattraction,becausethefunctionVisnotunique.
Theset itselfisnotnecessarilyadomainofattraction.Actually,the
abovetheoremdoesnotguaranteethatisinvariant:some
trajectoriesstartinginbutoutsideofthelargestl mayactually
endupoutside.

70

10/25/2014

GlobalInvariantsets

Globalinvariantset
GlobalInvariantSetTheorem)Considertheautonomous
system,withfcontinuous,andletV(x)beascalarfunctionwith
continuousfirstpartialderivatives.Assumethat

LetRbethesetofallpointswhereandMbethelargest
invariantsetinR.Thenallsolutionsgloballyasymptotically
convergetoMas

71

10/25/2014

Considerthesystem

TheLyapunov function

Thisfunctionhastwominima,atandalocal
maximuminx(asaddlepointinthestatespace)at

Thus,theinvariantsettheoremindicatesthatthesystem
convergesgloballytoor

72

10/25/2014

Remark
Lyapunov functionisnotunique.
SeveralLyapunov functionsmayexistforagivensystem,and
thereforeseveralassociatedinvariantsetsmaybederived.
Thesystemthenconvergestothe(necessarilynonempty)
intersectionoftheinvariantsetsMi,whichmaygiveamore
preciseresultthanthatobtainedfromanyoftheLyapunov
functionstakenseparately.
Equivalently,onecannoticethatthesumoftwoLyapunov
functionsforagivensystemisalsoaLyapunov function,whose
setRistheintersectionoftheindividualsetsRi.

Krasovskii's Method(Lyapunov for


NLsystem)
Krasovskii's methodsuggestsasimpleformofLyapunov function.
Theorem3.7(Krasovskii)Considertheautonomoussystem
definedbydx/dt=f(x),withtheequilibriumpointofinterestbeing
theorigin.LetA(x)denote,theJacobian matrixofthesystem,i.e.,
Ifthematrixisnegativedefiniteinaneighborhood
thentheequilibriumpointattheoriginisasymptoticallystable.A
candidateLyapunov functionforthissystemis
If istheentirestatespaceand,inaddition,
thentheequilibriumpointisgloballyasymptoticallystable.

73

10/25/2014

Example:
Considerthenonlinearsystem

Theequilibriumstateattheoriginisglobally
asymptoticallystable.

Practice
Let

findaLyapunov functionforc.

74

10/25/2014

GeneralizedKrasovkii theorem

TheVariableGradientMethod
ApproachtoconstructingLyapunov functions.
Thewayitworks:
Define

InordertorecoverauniquescalarfunctionV(x)from
thegradient,thegradientfunctionhastosatisfy
thesocalledcurlconditions

75

10/25/2014

Step1
Theprincipleofthevariablegradientmethodistoassumea
specificformforthegradient,insteadofassumingaspecific
formfortheLyapunov functionVitself.
Usualassumption
aij aretheunknowns

Step2
Solveforaij

Example
FindtheLyapunov functionofthefollowingsystem

WeassumethatthegradientoftheundeterminedLyapunov
functionhasthefollowingform

Thecurlequationis

76

10/25/2014

Step3:restrictthecoefficientsinsothatdV/dt isnegative
semidefinite(atleastlocally)
ThecoefficientsshouldselectedsuchdV(x)/dt<0.
Thecoefficientsareconsideredconstantforexample

Step4:computeVfrombyintegration

Lyapunovs Instabilitytheorem
(Chetaevs Theorem)
Ifinacertainneighborhood oftheorigin,thereexistsa
continuouslydifferentiable,decrescent scalarfunctionV(x,t)
suchthat
V(0,t)=0 t>t0
V(x,t0)canassumestrictlypositivevaluesarbitrarilycloseto
theorigin
dV(x,t)/dt ispositivedefinite(locallyin.)
Thentheequilibriumpoint0attimet0 isunstable.

77

10/25/2014

Practice
Canwesaythesamefora?

78

Anda mungkin juga menyukai