Lyapunov stability theory Consider the nonlinear system

= f ( x) x

(1)

Let us assume that xeq = 0 is an equilibrium point of (1). Q.: What if the equilibrium point xeq 0? .: !here is no loss of "enerality by this assumption. We #an al$ays #hoose a shifted #oordinates system in the form y = x % xeq. !he deri&ati&e of y is "i&en by
=x = f ( x ) = f ( y + xeq ) = g ( y ) = 0 y

!he system des#ribed by the means of the ne$ &ariable has the equilibrium at the ori"in.

'efinition !he equilibrium point x = 0 of (1) is stable( if for ea#h ) 0( there is = () ) 0 su#h that
x (0) < x(t ) < ( t 0

unstable( if not stable asymptoti#ally stable( if it is stable and #an be #hosen su#h that
x(0) < lim x(t ) = 0
t

mar"inally stable if it is stable but not asymptoti#ally stable

symptoti#ally stable

xeq

+ar"inally stable

unstable

Lyapunov First Method (The indirect method) ##ordin" to the basi# definitions( stability properties depend only on the nature of the system near the equilibrium point.

Let us lineari,e the system des#ription-

.or small de&iations from the equilibrium point( the performan#e of the system is appro/imately "o&erned by the linear terms. !hese terms dominate and thus determine stability % pro&ided that the linear terms do not &anish. !he idea of #he#0in" stability by e/amination of a lineari,ed &ersion of the system is referred to as Liapunovs first method or Liapunovs indirect method.

Theorem Let x = 0 be an equilibrium point of a nonlinear system

$here f : D Rn is #ontinuously differentiable and D is the nei"hborhood of the equilibrium point. Let i denote the ei"en&alues of the matri/
A= f x
x= 0

1. 2f 3e i 4 0 for all i then x = 0 is asymptoti#ally stable for the nonlinear system. *. 2f 3e i ) 0 for one or more i then x = 0 is unstable for the nonlinear system. 1. 2f 3e i 0 for all i and at least one 3e j = 0 then x = 0 may be either stable( asymptoti#ally stable or unstable for the nonlinear system Con#lusion: 5/#ept for the boundary situation( the ei"en&alues of the lineari,ed system #ompletely re&eal the stability properties of an equilibrium point of a nonlinear system. 2f there are boundary ei"en&alues( a separate analysis is required

An example K M B
d * y (t ) dy (t ) +B + Ky (t ) = f (t ) * dt dt

f y

+oreo&er( sin#e $e are interested in stability properties( f(t) = 0.

equilibrium point : y = 0

7tate &ariables

x1 (t ) = y(t ) (t ) x* (t ) = y
1 = y = x* x K B x = y = x1 x* * M M

!he total stored ener"y is "i&en by

$hi#h ha&e the follo$in" properties: positi&e for all non,ero &alues of x1(t) and x1(t) equals ,ero $hen x1(t) = x1(t) = 0 !he time deri&ati&e of V(t) is "i&en by:
dV (t ) V (t ) V (t ) 1 + * = x x dt x1 x* dV (t ) * = Bx* dt

dV9dt is ne"ati&e the state must mo&e from its initial state in the dire#tion of smaller &alues of V(t)

V V V

!1 !* !1

x*

x1

!1 4 !* 4 !1

Lyapunov Second Method (The direct method) Theorem Let x = 0 be an equilibrium point of a nonlinear system
= f ( x) x

Let V : D R be a #ontinuously differentiable fun#tion on a nei"hborhood D of x = 0" su#h that V(0) = 0 and V(x) ) 0 in D % ;0<(
( x) 0 V

in D

!hen( x = 0 is stable.
( x ) < 0 in D % ;0< then x = 0 is asymptoti#ally stable +oreo&er( if V

!he tas0: !o find V(x)( #alled a Lyapunov function( $hi#h must satisfy the follo$in" requirements: V is #ontinuous V(x) has a unique minimum at xeq $ith respe#t to all other points in D lon" any tra=e#tory of the system #ontained in D the &alue of V ne&er in#reases

>

What if the stability of x = 0 has been established? !he first Lyapuno& method determines stability in the immediate &i#inity of the equilibrium point. !he se#ond Lyapuno& method allo$s to determine ho$ far from the equilibrium point the tra=e#tory #an be and still #on&er"e to it as t approa#hes

region of asymptotic sta#i$ity (region of attraction( #asin) Let (t?x) be the solution of the system equation that starts at initial state x at time t = 0. !hen the region of attraction is defined as the set of all points x su#h that limt (t?x) = 0 2f # = ; x Rn @ V(x) c < is bounded and #ontained in D( then e&ery tra=e#tory startin" in # remains in # and approa#hes the equilibrium point as t . !hus( # is an estimate of the re"ion of attra#tion.
A

!ypes of stability $ith referen#e to the re"ion of attra#tion: $oca$ sta#i$ity (sta#i$ity in the sma$$) % $hen a system remains $ithin an infinitesimal re"ion around the equilibrium $hen sub=e#ted to small perturbation finite sta#i$ity % $hen a system returns to the equilibrium point from any point $ithin a re"ion R of finite dimensions surroundin" it g$o#a$ sta#i$ity (sta#i$ity in the $arge) % if the re"ion R in#ludes the entire state spa#e Theorem Let x = 0 be an equilibrium point of a nonlinear system
= f ( x) x

Let V : Rn R be a #ontinuously differentiable fun#tion su#h that V(0) = 0 and V(x) ) 0 x 0(

@@x@@ V(x)
( x) 0 V

x 0

then x = 0 is "lobally asymptoti#ally stable

Another example (a pursuit problem) 7uppose a hound is #hasin" a rabbit (in su#h a $ay that his &elo#ity &e#tor al$ays points dire#tly to$ard the rabbit). !he &elo#ities of the rabbit and the hound are #onstant and denoted by R and %( respe#ti&ely (see the pi#ture)

Let xr( yr( and xh( yh denote the x and y #oordinates of the rabbit and hound( respe#ti&ely. !hen
r = R x y r = yr = 0
* * h h x +y = %*

10

!he fa#t that &elo#ity &e#tor of the hound al$ays points to$ard the rabbit means that
h = & ( xh xr ) x y h = & ( y h y r )

& % a positi&e #onstant

7o
h = x h = y % ( xh xr )
* ( xh xr ) * + y h

%y h
* ( x h xr ) * + y h

Let us introdu#e the relati&e #oordinates % the #oordinates of the differen#e in position of the hound and the rabbit:
x = xh x r y = yh

= x h = y

%x x* + y * %y

(C)

x* + y *

Will the hound al$ays #at#h the rabbit?

Will a tra=e#tory $ith an arbitrary initial #ondition e&entually "et to the point $here the relati&e #oordinates are ,ero?

We #an #onsider the ori"in as an equilibrium point

What are #onditions for "lobal stability of the system?


We ha&e to find a suitable Lyapuno& fun#tion for the system "i&en by (C) Let us #hoose as a Lyapuno& fun#tion V(x(y) = x* D y* !hen
( x( y ) = * % V x * + y * * Rx

2f % ) R:
( x( y ) < 0 if x = 0 and y 0( it is #lear that V

if x 0 then
% x * + y * Rx < ( % R ) x < 0

( x( y ) < 0 for all x( y e/#ept the ori"in. !hus( V

2f the hound runs faster than the rabbit( he al$ays #at#hes the rabbit

1*

Comments on the se#ond Lyapuno&Es method: determines stability $ithout a#tually ha&in" to sol&e the differential equation #an be applied e&en if the system model #annot be lineari,ed allo$s to estimate the stability re"ion in some #ases there are natural Lyapuno& fun#tion #andidates( li0e ener"y fun#tions in ele#tri#al or me#hani#al systems

the stability #onditions are suffi#ient( but not ne#essary

there is no systemati# method for findin" Lyapuno& fun#tions % sometimes a matter of trial and error a Lyapuno& fun#tion for any parti#ular system is not unique

11

Nonlinear phenomena: 'inite escape time: !he state of an unstable linear system "oes to infinity as time approa#hes infinity? a nonlinear systemEs state( ho$e&er( #an "o to infinity in finite time. Mu$tip$e iso$ated equi$i#ria: a linear system #an ha&e only one isolated equilibrium point? hen#e it #an ha&e only one steadyF state operatin" point $hi#h attra#ts the state of the system irrespe#ti&e of the initial state. nonlinear system #an ha&e more than one isolated equilibrium point. !he state may #on&er"e to one of the se&eral steadyFstate operatin" points( dependin" on the initial state of the system. Limit cyc$es: .or a linear timeFin&ariant system to os#ilate( it must ha&e a pair of ei"en&alues on the ima"inary a/is( $hi#h is a nonrobust #ondition that is almost impossible to maintain in the presen#e of perturbations. 5&en if $e do( the amplitude of the os#illation $ill be dependent on the initial state. 2n real life stable os#illation must be produ#ed by nonlinear systems. !here are nonlinear systems$hi#h #an "o into an os#illation of fi/ed amplitude and frequen#y( irrespe#ti&e of the initial state (so #alled $imit cyc$e)
16

 (u#harmonic" harmonic or a$most)periodic osci$$ations: stable linear system under a periodi# input produ#es an output of the same frequen#y. e/#itation #an os#illate nonlinear system under periodi# $ith frequen#ies $hi#h are

submultiples or multiples of the input frequen#y. 2t may e&en "enerate an almost%periodi# os#illation( an e/ample of $hi#h is the sum of periodi# os#illations $ith frequen#ies $hi#h are not multiples of ea#h other !haos: nonlinear system #an ha&e more #ompli#ated steadyF

state beha&ior that is not equilibrium( periodi# os#illation or almost%periodi# os#illation. 7u#h beha&ior is usually referred to as #haos. 7ome of these #haoti# motions e/hibit randomness( despite the deterministi# nature of the system Mu$tip$e modes of #ehavior: 2t is not unusual for t$o or more modes of beha&ior to be e/hibited by the same nonlinear system. #y#le. n unfor#ed system may ha&e more than one limit for#ed system $ith periodi# e/#itation may e/hibit

harmoni#( subharmoni# or more #ompli#ated steadyFstate beha&ior( dependin" upon the amplitude and frequen#y of the input. 2t may e&en e/hibit a dis#ontinuous =ump in the mode of beha&ior as the amplitude or frequen#y of the e/#itation is smoothly #han"ed
18