Hassan K. Khalil - Complete Solution Manual For Nonlinear Systems-Pearson - Prentice Hall PDF

Nonlinear Systems
Third Edition
非线性系统
（第三版）
Solutions Manual
Hassan K. Khalil
l[iij] 1' :,. .l. ~ ~ }ti 1;J•.

~fflmQ House c f l:lffl~ftiCS 1ncuwry
~---11!!!· htlp-JIWW\ll',phii.OOi'ii'i.ffl
《非线性系统（第三版）》习题解答
Chapter 1
• 1.1 Let Zt = 11, %2 = 11<1), •.• , Zn= 1l(n-l)_
Xn-1 = Zn
Zn = g(t,x1, ... ,xn, u)
11 = %1
•
.,. -_ fl (1) , • • •,
1 •2 Let Zl -_ Y, ..,2 _
Xn-1 - 'I/ (n-3) , •n
,.. _ 1 ,(n-1) _
- 11 fh (t , 1/tJ/{1) , • • • 1 'I/ (n-2))
· U,
Zn-2 =
Zn-1 =
Zn =
=
'I/ = Z1
.,. -- 1/ I ...
• 1 •8 Let *l ,.. - 11 (1)
2 - 0
.,. - y(n-1) ...
• • • 7 '"'" -
_ ..
0 *n+l - '"'1 • •' 0
.. _ z(m-1)
'"'f'l+na - •
:i1 = Z2
Xn-1 = Zn
Zn = g(z1,.,. , Xn,Zri+t,, ., , Zn+m, U)
tn+1 = Zn+2
Zn+m-1 = Zn+m
Zn+m = u
'JJ = :i:1
- 3
• 1.4 Let z1 = q, z2 = q,· z =[ z.z12 ] E Ram ·
Z1 = Z2
z2 = q = M-1(:i:i)(u - C(.z1,.z2):t:2 - D:t:2 - g(.zi))
%1 = x2
MgL. k
z2 = - -smz1
I
- -(z1 - :t:3)
I
%3 = x,
k 1
x, = -(.z1 - zs) + -u
J J
:i:1 = z2
:i:2 = -M-1(.z1)[h(z1, z2) + K(z1 - za)]
%3 = z,
%4 = J- 1 K(z1 - za) + 1-1 v
• 1.7 Let
:i: = Az + Bu, fl = Cz
be a state model of the linear system.
u = r-,J,(t,11) = r- ¢(t, Cz)
Bence
x = h - B1/,(t,Cz) + Br, i =Cz
• 1.8 (a) Let 2:1 = 6, z2 =6, zs =E9 , and u = E,o.
z1 = z2
p D 'h .
= -M - -2:2
:i:2
M - -z1smz1
M
'1'l 'II 1
., +-cosz1 + -u
%3 = - -2:3 ., .,
(b) The equilibrium points are the roots of the equatiom
0 = .Z:z
0 = 0.815- Dz2 - 2.0zs sinz1
0 :: -2.7.zs + 1.7cosz1 + 1.22
0.4076
Z2 = 0 => Zs = -:---
IID %1
Substituting 2:3 in the third equation yields
(1.22+ l.7cosz1)sinz1 -1.l~ =0
The foregoing equation has two roots z1 = 0.4067 and z1 = 1.6398 in the interval -1t S z1 ::s;: ,r. Due to
periodicity, 0.4067 + 2mr and 1.6398 + 2mr are also roots for n = :H,:1:2, .... Ea.ch root z 1 = z gives an
- 4
equilibrium point (z, 0, 0.4075/ Binz).
(c) With E, =constant, the model reduces to

Z1 = X2
z2 = ~ - ix2 - ZE,sinx1
which is a pendulum equation with an input torque.
• 1.9 (a) Let Z1 =t/>L, Z2 = Ve,
(b) Let X1 = iL, Z2 =VC,

z1 = Iokooslc,pL ¢L = 1cJJl-iivc
= z2k../Jj - :it
:i:2 = ~ [i• - z1 - !z2]
The model of (a) is more famUiar lince it is the pendulum equation.

• 1.10 (a) Let z1 = 'PL, X2 = vc.
:i:1 = 4'1, = VL =VO =:t2
. . 1.
z2 = vc = c•c = C1 [·•• - tlC
]f -
. ]
IL
= · - LX1
1 [••
C - l'Z1s - RZ2
1 ]
(b) :r2 =0 • Lz1 + µzf = 0 ~ z1 = 0. There is a unique equilibrium point at the origin.
• 1.11 (•)
.i = Az+ Bu, 'JI= Cz, u= sine
i. = 81. - Bo = -Bo = -11 = -Cz
i=Az+Bsine, e=-Cz
. (b)
O=Az+Bsine ~ z=-A- 1 Bsine
O= Cz => -cA- 1 Bsine = G(O)Bine = 0
G(O)~O => sine=O => e=±mr, n=0,1,2,· .. and z=O
(c) For G(s) = 1/(,,-, + 1), take A= -1/'r, B = 1/T and C = 1. Then
. 1 1 . .
z=- -z+-sme e=-z
'T T '
Let z1 = e, z2 = -z.
. 1 1 .
:t2 =- -%2 -
T
- smz1
T
- 5
• 1.12 The equation of motion is
Let %1 =11 8.11.d %2 =fl.
• 1.13 (a)
mjj = -(k1 + k2)f - cy + h(t.io -ti)
where c > 0 is the viscous friction coefficient.
(b) h(v) ~ h(Uo) - h'(t.io}Ji.
mjj = -(k1 + ~)II - [c+ h'(t1o)]ii + h(t.io)
(c) To obtain negative friction, we want c+ h'("n) < 0. This can be achieved with the friction characteristic
of Figure 1.5(d) if 110 is in the range where the slope is negative and the magnitude of the negative slope is
greater than c.
where k1 , k2 , and l:3 are positive constants. Let :c =v, u =F, and w =sin 8.
:t = ! [-k1sgn(x) - .k2x - ksx2 + u] - gw
• 1.115 (a)
al d ' - rn2
B = m dt:, (11 + Lsin6) = m dt (Ji+ Ucod} = m(j + UcoalJ - ur sin8)
al d • .;
V =mdii(Lcos8) + mg= mdt(-L8sin8) rA2
+mg= -mL8sin6-m.tA"" cosl+mg
Substituting V and H in the 8-equation yields
16 = VLliDI-HLcosl
=: -mL26(1in.8) 2 -mL2 e2sin9coslJ+mgLsin6
-ml,j cod - mL21(cod)2 + mL2 e2 siu6C088
= -mL2i[(sin8)3 + (cos8)2 ) + mgLsin9- mLficos9
= -mL26+ mgL sin8 - mi,icoe8
Substituting Hin the ii-equation yields
Mii = F-m(jj + L8cos8-LiJ2 sin8) -kj
(b)
jj
D(D) [ y
1= [ F+mU'sin9-~
mgLsin9 ]
where
D(B) = [ I+ mL2 mLcoe9 ]
mLcos9 m+M
det(D(6)) = (I+ mL2 }(m + M) - m 2L 2 coa2 9 = ~{0)
- 6
Hence,
_1 1 [ m+ M -mLcos(J]
D (fJ) = .6.(6) -rnLcos9 I+ rnL2
[ ii9 ] = 1 [ m +M -mL cos 9 ] [ mgL sin 9 ]

.6(9) -mLcos8 I+mL 2 F+mLe2sin8-ky
(c)
%1 = X2
z, = 6 = Ll~B) [(m + M)mgLsin9-mLcos8(F +mL.82 sin8- k1i))

= .6.{~i) [(m + M)mgLsinz1 - mLcou:1 (u + mL~ sinz1 - kz.))
%3 = x.
=· = ii= Ll~B) [-m2 L2 gsin8cos8+(l+mL2 )(F+m.l.82sin9-kti)]
= .6.(~i) [-m2L2gsinz1 cosz1 +(I+ mL2)(u + mLzi sinz1 - b,.)J
• 1.16 (a)
<fl d . .. .
F:ri = m dt2 (Xe+ Lsin.9) = m dt (zc + L8coa9) =m(ftc + L8oos6- L82 sin8)
<fl d . - .. ~
F, = mdt' (L cos I)= m dt (-L8sin8) = -mL8sinfJ- m.t.«r 0088
Substituting F. and Fr in the 8-equation yields
Ii = u+Fs,Lsin8-F.Lcos9
= v - mL2i(sin9)2 -mL2i2 sinlcosfJ
-mLfcma9-mL2i(e019)2 + mL2f' sinlcosl
= v-mL2i[(ain9)~ +(co8')2 J- mLftccosl
= 2 ..
u-mL 9-mLzecosB
Substituting F. in the Zc-equ&UOD yields
M!e = -m!c - mL8 coe9 + mLIP sin 8 - kxc
Thus,
where
D(B) =[ I+ mL2 mLcoa8 ]
mLcos9 m+M
(b)
det(D(S)) = (J + mL2 )(m + M) - m2L 2 oor 9 = .6.(9)
Hence,
D-1(8) _ _!_ [ m +M -mLcos9 ]
- .6.(9) -mL cod I + mL"
- 7
jj ]
[ fie 1 [ m+M -mLcos9 ] [ u ]
= ~(9) -mLcos8 I+ mL2 mLe2 sin8 - kxe
(c)
z1 = x,
.i'2 = i = ~~S) [(m + M)u - mLC01!8(m.L82 sin8- kzc)]
= d(~i) [(m+M)u-mLcosx1{mLxisinx1 -kx3 )].
:ts =
z, = fie z: ~~B) [-mLucos9 +(I+ mL2 )(mL82 sin8- kxc)]
= ~(:i) [-mLucosx1 +(I+ mL2)(mL~sinz1 - b 3)]
(d) Take u = constant. Setting the derivatives z, =0, we obtain z 2 = z, =0 and

0 =
(m + M)u + mkL:ca cosz1
0 ;:: -mLucosz1 - k(I + mL2)xs
Eliminating x 3 between the two equations yields
u[(m + M)(/ + mL2 ) - m2 L 2 cos2 x 1 ] = u 4(:ri) = 0
Since 4(x1) > 0, equilibrium can be maintained only at u = 0. Then, zs = O. Thus, the system has an
equilibrium set {x, =:rs= = O}. x•
• 1.17 (a) Let z1 = i1, X2 = i8, and %3 = w.
z1 = - !!t.%1 + !L
L1 L1
R..
i::, = --z2- ct "•
-z1z1+-
L. L. La
%3 = - J
cs C,
-zs+-z1z2
J
(b) Take "• = v. = constant and "I = u.
(c) 'lake v1 = v, = constant and "• = u.A constant field voltage implies that (at steady state) i1 =
V1/R1 ~ 11 = constant. Hence, the model reduces to ihe aec:ond-oder linear model
z2 Ro ctl1 zs+-
u
= - -x2-
La La L0
Cs t:211
.i's = - -xs+-x2
J J
(d) Let"= u.
R. +R1 u
.i'1 = - Z1 +-
L1 L1
Re C1 U
±2 = --x2- -x1xs+-
L. La L.
Cs £2
zs = - -xs+-x1z2
J J
- 8
• 1.18 (a) z1 =11, z2 ='ii, zs = i, u = v.

:t1 = z2
%2 = fl.. =- k . + g + -1 F(11, t')
-y
m m
k M
1
= g- mz2 - ffl . 2a(l+z1/a)2
k Lo~
= g- -z2 -
m 2m(a+x1)2
%3 = di
dt = d[']
L(r,)
dt
~ ti> dL.
= L(i,) - L 2(11) . di/
1 ( . ef> Lo .
= L(s,) v - Ri) + £2(s,) . a(l + s,/a)211
= _!__ [-Rz3 + Looz2za2 +u]

L(xi) (a+ :i:1 )
(b) The equilibrium equations a.re
0 = .t2
0 - g- !.~- Lo~
m 2m(a+!1)2
0 = -& + Lod2~3 + U
I (a+ ~1)2
Set M1 = r, ~, =1,., and tl = V.,. Then
I.. -- (2mg(a+r}2)1/2
Loa • v.. =RI..
• 1.19 (a)
! (Lh d).) A(>.) = Wi - k,/pih
A(h)h = u- k.Jijih
Let z = h.
z= Atx) [u - A:,,lpji), 11 =:,:
(b) X =p-p. = pgh.
(c) At equilibrium,
0 = u., - k..,/pgz.. , a,,,= z,, = r
Hence, u,, = k.,/Pir
- 9
• 1.20 (a) From the equations v =Wi -wo and p =Po+ (pg/A)v, we have
p= 1v = PJ (wi -1110) = 1[,r (~p)- ky'Ap)
1
Using z = ll.p as the state variable, we obtain

z = 1[,t,-1 (z)-k/.i)
(b) At equilibrium we have
f)-1 (.f) =k#
Writing!= ,p(w,), we can rewrite the previous equation as
Wi = ky'(/J(tDi)
Henc~,
(~;) 2 = 4>(wi)
The solutions ofthis equation are giftm by the intersection of the curve (w,/k) 2 with the curve t/>{wi), which
is shown in Figure 1.29 of the text. From the figure, it is clear that there is only one intersection point.
• 1.21 (a) We have
Vi =Wp-W1, V2 =W1-W2
. pg. . pg.
Pl = A1 vt, P2 = 1;va
W1 = k1 V.,_Pl___P2_, "'2 • k2v'P2 - Pa, Pl - Pa =t/>(wp)
Let :Z:1 z Pl - Pa and :z:2 =P2 - P•·
.t1 = 'Pl = ~ {w,, -111i) = ~~ [9',-1(:z:1)- k1v':z:1 -:z:2]
z2 = Pl =: (to1 -w,) = : [kiv'z1 -:z:2 - k:av'ii]
(b) The equilibrium equa.tions are
t/>- 1(z1) = k1 v'~1 - ~2
k1 v'!1 - .f2 = ~vii'
· From the second equation, we have
Writing !1 =,f,(w,,), we can rewrite the previous equation as

w,, =k_,J;(w,,)
Bence,
(£!) 2 = ~(w,,)
The solutions of this equation are given by the intersection of the curve (w,/k.q)2 with the curve ;(wp),
which is shown in Figure 1.29 of the text. From the figure, it is clear that there is only one intersection
point.
- 10
• 1.22 (a)
z1 = dz11 -dz1 +r1

i2 = dx21 - dz2 - r2
The assumptions r1 = JJX1, r2 = rifY = JJZ1/Y, and z11 =0 yield

:i:1 = (µ ~ d)z1
z2 = d(z21 - z2) - JJX1 /Y
(b) When µ = Jlm:Z2/(km + z2), the equilibrium equatiOllS are

0 l'ffl!2
= ( ./cm +!2
- d) .f l
l'ffl!1!2
0 = d(z21 - !2) - Y(km + i2)
from the fint equation,

f1 =0 PfflZz
or - - - •
d -=
~ ..,2 =-k,,.d
-
km +!2 Pm-d
Substituting ! 1 = 0 in the second equilibrium equation yields !2 =z 2,. Substituting f 2 = /.md/(JJ.m -d) m
the second equilibrium equation yields
f1 = Y (z21 - #'n&km~d)
Bence, there are two equilibrium points at
( y ( ~2/ - µ,,_~ d) t #'m~ d) &Dd (0, Z2/}
(c) Whenµ= µ,,.z2/(km + z2 + k1zl}, the equilibrium equations are
from the fint equation, !1 = 0 or !!2 is the root of d =- µ(!!2), Since d < ma.x.2~o{µ(z 2)}, the equation
d = µ(.:f2) has two roots. Denote these roots by !2a and!». Subetituting z1 = 0 in the second equilibrium
equation yields ~2 = :i:21. Substituting !2 = !2. in the second equilibrium equation yields
d(z21 - !2a) - µ(!!2a)!1/Y =0 ~ z1 =Y{z21 - !2a)
since µ(:!2a} = d. Similarly, substituting f2 = fn in the second equilibrium equation yields ! 1 = Y(:i:21 -
!2111). Thus, there are three equilibriUUl points at
(Y(z21-z2.),f2c), (Y(z21-!2•),!n), and (0,z21)
- 11
-12
Ch~pter 2
• 2.1 (I)
0 = -xi + 2xf + :i:2, 0 = -z1 - :t2
x2 = -x1 => 0 = 2x1(ri - l} => X1 = 0, 1, or - 1

There are three equilibrium points at (0, 0), (1, -1), and (-1, 1). Determine the type of each point using
linearization.
!JJ -[-1+6z¥ 1]
8x - -1 -1
8,, = r-1 !
Ox (o,o) -1 1] => .\1,2 = -1 :I:: j => (O, 0) is a stable focus
: 1(1,-1) = [ !1 !1 ] -> >.1,2 = 2 ± v'8 => (1, -1) is a saddle

Similarly, (-1, 1) is a saddle.
(2)
0= X1 (1 + Z2), 0 = -.:i::2 + xi + Z1X2 - xf
0 = z1(l +%2) => z1 =O or z2 = -1
%1 = 0 => 0 = -%2 + ~ => = 0 or X2 = 1
:Z:2
X2 = -1 :::> 0 = 2 - X1 - z? => Z1 = 1
There are three equilibrium points at (O, 0), (O, 1), uid (1, -1}. Determine the type of eacli point using
linearization.
8/ [ 1 +z2 ~1 ]
lJz = X2 - 3zf -1 + 2z2 + X1
: I =[ ~ !
(O,O)
1] => .\1,2 == l, -1 => (O, 0) is a saddle
l)J
fJx I [ oJ
(0,1)
= l2 l => >.1,2 = 2, 1 => (0, 1) is unstable node
8/1 = [ _04
8 _I2 ] . ~ => (1, -1) ia a stable focus
-> >.1,2 = -1 ±3v3
z (1.-1)
(3)
- 13
From the second equation, z2 = 0 or x2 = 2(1 + x1),
z2 = 0 => =0 or x1 = 1
X1
x2 =2(1 + :r:i} => 0 = (x1 + 3):r:1 => X1 = 0 or :1:1 = -3

There are four equilibrium points at (0,0), (1,0), (0, 2), and (-3, -4). Notice that we have assumed 1 +:r:1 :/:,
O; otherwise the equation would not be well defined.
a, = [
In
i- 2%1 - 11:~>' -2c1-f~1> ]
213 2
(1+:115 2- ~
:I •=(O,O)
= [ ~ ~ ] i Eigenvalue.s : 1, 2 => (O, 0) is UDStable node
:I -(1,0)
= [ 01 -;1 ] ; Eigenvalues : - 1, 2 => (1, 0) is a saddle
:I 111=(0,J)
= [ ~3 !2 ] ; Eigenvalues: - 3, -2 => (0, 2) is a stable node
!1I
v"X a,a:(-3,_.)
= [ : . =! ];Eigenvalues : 7.722, -0.772 => (-3, -4) is a saddle
(4)
0 = Z2, 0 = -z1 + 2:2(1 -xf + 0.lzf}
There is a UDique equilibrium point at (O, O). Determine its type using linearization.
lJf [ 0 1 ]
. fk = -1 - 2%1:r:2 + 0.4:r:j:r:2 1 -%f + O.b1
: 11
X (0,0)
=[ !1 ~] =>
.
A1,2 =(1/2)±j../s/2
. .
=* (O,O) is unstable focus
,
(5)
0 = (x1 - x2)(l - xf - ~), 0 =(:r:1 + x2)(l - zf - ~)
{zf + zl = l} is &11 equilibrium 98t and (0,0) man isolated equilibrium point.
a, I
fh irs(0,0)
=[ 1-
1-
azr -~ +
3zf - ~ -
2x1z2
2x1%2
-2x1:t2 -
-2x1X2
1 + ~ + 3zi
+ 1 - ZJ - ~
J =[1
(0.0) 1
Eigenvalues are 1 ± j; hence, (0, 0) is unstable focus.
(6)
0 = -zt + :r:2, 0 = z1 - xi
Z2 =:r:f => Z1 (1 - zf) =0 => Z1 = 0 or zr = 1
The equation :,;f = 1 has two real rooia at .z1 = :U. Thus, there are three equilibrium points at (0, O), (1, l},
(-1,-1).
- 14
:1 z•(0,0)
=[~ ~]; Eigenvalues:l,-1 => (0,0)isasaddle
IJ/1:i:=(1,1) =
8x
r-a 1
1
-3
] .
I
Eigenvalues:-2,-4 => {1,l)isastablenode
Similarly, (-1,-1} is a stable node.

• 2.2 (1)
Q =%2, 0 = _,XI + (l/16)zt - X2
z2 = 0 => 0 = z1 (zt - 16) => z1 = 0, 2, or - 2

There are three equilibrium points at (0,0), (2,0), and (-2,0). Determine the type of each point UBing
liDearizatioD.
lJJ [ 0 1 ]
EJx = -1 + (5/16)zf -1
81 / - [ -1
O 1 ] => ,\1,2 = -(1/2) ±j./3/2 => (0,0) is a stable focus
/h (0.0) - -1
8/1
{)z (2 .~> =
[ 40 !1 ] ., .\1,2 = -(1/2) :l:: .fiT/2 .- (2,0) is a saddle
Similarly, ( -2, 0) is a saddle.
(2)
0 = 2%1 - i1z:a, 0 = 2ri - z2
Z1 (2 - Z2) =0 * Z1 = 0 or Z2 = 2
Z1 = 0 => Z2 = 0, Z2 = 2 . => zf ,;: 1 .. Z1 = 1 or - 1
There are three equilibrium points at (0,0), {1,2), and (-1,2). Determine the type of each point using
linearlzaacm.
8/,
{Jz (O,D) =
[ 2 ~l ]
0
=> .\1,2 = 2, -1 => {O, 0) is a saddle
8/1
8z =[ 0 -1]
4 -1
(1,2)
=> ,\1,2 • -(1/2) ±j,/ff,/2 ., (1, 2) is a stable focus
~Ll,2) = f ! !1 J .- .\1,2 = -(1/2) :l:: ;'1i5/2 => {-1, 2) is a stable focus

(3)
0 = z2, 0 = -z2 -tt,(z1 - :r::2)
z2 = 0 *
tJ,(z1) O => :r::1 0 = =
There is a UDique equilibrium point at (0, 0). Determine its type by linearization.
8/1
& (O.O} =
[ -3(x1 - 0 z2) 2 - 0.5 -1 + 3(z1 -
1
z2)2 + 0.5
]
(O.O) =[ 0
-0.5
The eigenvalues are -(1/4) :I: j./7/4. Hence, (0, 0) is stable focus.
- 15
• 2.3 (1) The system has three equilibrium points: (O,O) is a Btable focus, {1,-1) and (-1,1) are saddle
points. The phase portrait is shown in Figure 2.1. The stable trajectories of the saddle form a lobe around
the stable focus. All trajectories inside the lobe converge to the stable focus. All trajectories outside it
diverge to infinity.
(2) The system has three equilibrium points: (0, O) is a saddle, (0, 1) is unstable node, and (1, -1) is a stable
= =
focus. The phase portrait is showu in Figure 2.2. The z2-axis is a trajectory itself since z1 (t) 0 => z1 (t) 0.
The x 2-axis is a separatrix. All trajectories in the right half converge to the stable focus. All trajectories in
the left have diverge to infinity. On the z2-a:x:is itself, trajectories starting at z 2 < 1 converge to the origin,
while trajectories starting at z2 > 1 diverge to infinity.
0 2 0
"1 •1
Figure 2.1: Exercise 2.3(1). Figure 2.2: Exercise 2.3(2).
(3) The system has four equilibrium poiD.ta: (0, 0) ia umtable aode, (1, O) ii a saddle, (0, 2) is a stable node,
and (-3,-4) is a saddle. To avoid the condition z1 + l = 0, we limit our analysis to the right half of
the plane, that is, {z1 :2:: O}. This makes aenae in view of the fact that the zraxis is a trajectory since
=
:i:1(0 i!E O => :i1(t) O. Hence, trajectories starting in {z1 :2:: O} ltay there for all time. The phaae por-
trait ill shown in Figure 2.3. Notice that the z1-axis ia a trajectory since :i:2(t) !E O *
z2(t) =
O. It ia a
separam,c that divides the half plane {z1 ~ O} into two quarters. All trajectories starting in the quarter
{z1 > 0, z 2 > O} converge to the stable Dode (0, 2). All uajectoriea nartiDg in the quarter {z1 > 0, z2 < O}
diverge to infinity. 'lrajectories starting on the z 2-axis approach the stable node (0, 2) if z2(0) > 0 and
diverge to infinity if z2(0) < 0. 'lrajectoriea starting on the z1•axil with z1(0) > 0 approach the addle
(1,0).
(4) There is a unique equilibrium point at the origin, which js unstable focus. The phase portrait is
shown in Figure 2.4. There are two limit cycles. The inner cycle is stable while the outer one ii unstable.
Al trajectories atartil1g inside the stable limit cycle, except the origin, approach it as i tends to infinity.
Trajectories starting in the region between the two limit cycles approach the stable Umit cycle. Trajectories
starting outside the unstable limit cycle diverge to infinity.
(5) The system has an equilibrium set at the unit circle and unstable focus at the origin. The phue portrait
is shown in Figure 2.5. All trajectoriel, except the origin, approach the unit circle at t tends to infinity.
(8) The system has three equilibrium points: a saddle at (0,0) and stable nodes at (1,1) and (-1,-1).
The phase portrait is shown in Figure 2.6. The stable trajectories of the saddle lie on the line z 1 + :r2 = 0.
All irajectorie& to the right ol this line converge to the stable node (1, 1) and all trajectories to its left
=
converge to the stable node (-1, -1). Trajectories on the line :r1 + z2 0 itself converge to the origin.
- 16
~0
-2
-4
-8
x, -5 0 5
x,
Figure 2.3: Exercise 2.3(3).
-1
~ -1 0 1 2
x,
Figure 2.5: Exercise 2.3(5). Figure 2.6: Extrcise 2.3(6).
• 2.4 (.1) The system bu three equilibrium points at (0,0),-(a,O), and (-a,0), where a is the root of
a = tan(a/2) =t- . a Aj 2.3311

The Jacobian matrix is
8/ [ O 1 )
lk = 1 - 2/[1 + (~1 + z2) 2] -2/[1 + (z1 + z2)2J
8/1
/jz (O,O)
= [ 0 1 ]
-1 -2
* A1,2 = -1, -1
Although we have multiple eigenvalues, we can conclude that the origin is a stable node because /(z) is an
analytic function of z in the neighborhood of the origin.
:t2.3311,0) = f o.~2 -o.!10s ] => >.1,2 :r: 0.6892, -1 ~ (2.3311,0) is a saddle
Similarly, (-2.3311,0} is a saddle. The phase portrait is Bhown in Figure 2.7 with the arrowheads. The
stable trajectories of the two saddle points forms two aeparatrices, which divide the plane into three regions.
AD trajectories in the middle region converge to the origin u t tends to infinity. All trajectories in the outer
- 17
regions diverge to infioity.
(2)
0 = x1(2 - x2}, 0= 2x? - :&2
From the first equation, z1 =0 or x2 =2.
%1 = 0 => %2 =0
%2 =2 => ~ = 1 :> %1 = ±1
There are three equilibriwn points at (0,0), (1,2), and (-1, 2).
IJJ I . = [ o·
a 2 o
-1 J :> >..1,2 = 2. -1 .. (0,0) .
18 a saddle
% (0,0)
{)JI
8x (l,ll) =
[ 40 =~ ] => >..1,2 = -(1/2) ±jv'l5/2 * (1, 2) is a stable fOCU8
8'1
{jz (-1,2) = ( 0
1 ]
-4 -1
=> >.1,2 = -(1/2) ±jv'U/2 => (-1,2) is a stable focus
The phase ponrait is showD iD Figure 2. 7 with the arrowheads. The stable trajectories of the saddle lie
on the :i:raxis, They form a separatrix that divides the plane in two halves. 'bajectories iD the right half
converge to the stable focus (1, 2) and thoee in the half converge to the stable focus (-1, 2).
(3) There is a unique equilibrium point at the origin.
:1 (0,0)
= [ !1 !] => .\1,2 = (1/2) :j../3/2 * (-1,2) is unstable focus
The phase portrait is shown in Figure 2.7 with the arrowheads. There is a stable limit cycle a.round the
origin. All trajectoriea, except the origin, approach the limit cycle u t teuds to mfinlty.
(4) The equilibrium points are gmm by the real roota of the equation
0 = 1/4 - 2112 + II
where :t1 = ti and :t2 = l-11. It can be seen that the equatiou hu four roots at,:; 0, 1, (-1:v'S)/2. Bence,
there are four equilibrium points at (0, 1), (1,0), ((3 - ../5)/2, (3- -v'S)/2), and ((3 + ../5)/2, (3 + ../5)/2).
The following table shows the Jacobian matrix and the type of each point.
Point Jacobian matrix Eigenvalues Type
(0, 1) r-2~I]
-1
-2.4142, 0.4142 saddle
(1, 0) [ !1 -2
-1] 0.4142, -2.4142 saddle
[ -1.2361 -1 ] -0.2361, -2.2361 stable node

((3 - ../5)/2, (3 - ./5)/2) -1 -1.2361
((3 + ../5)/2, (3 + ./5)/2) [ 3.2361 -1 ] 4.2361, 2.2361 UDStable node

-1 3.2361
- 18
The phase portrait is shown in Figure 2. 7 with the arrowheads. The stable trajectories of the Addle points
divide the plane into two regions. The region that contains the stable focus has the feature that all trajectories
inside it converge to the stable focus. All trajectories in the other region diverge to infinity.
-5L----...L.-----J
-2 0 2 4 -5 0 5
(1) (2)
2
0
x, . . 1
-1
-2
-2 -1 0 1 2 0 1 2 3
(3) (4')
Figure 2. 7: Phase pOrtraits of Exercise 2.4.
• 2,5 (a)
-fJ(:r) + 2:io(z} ]
-1- X1X2ct{z)
where
1 l
a(z) = (zf + zl)(ln /4 +zip' /J(z) = In v'zf + 4
Noting that lims-+0X;%Jc:t(z) = 0 for i,; = 1, 2 and lim.~ /J(z) = o. it can be seen that
a,, . [
& .-o =
-1
0
o ] =>
_1
th . . . .... _1..1 od
e ongm 18 a Hau e n e
(b) Transform the state equation into the polar coordinates
r= {zt +~, 8 =tan- 1 (::)
to obtain
r = -r => r(t) = r0 e-'
- 19
and, for O < ro < 1,

-
9 = lnr1 =lnro1 _ t * 9(t) =9o -ln(llnrol + t) + ln(llnrol)
Hence, for O < r 0 < 1, r(t) and 8(t) a.re strictly decreasing and limt~oo r(t) = 0, limt-,o B(t) = -oo. Thus,
the trajectory spirals clockwise toward the origin.
(c) /(z) is continuously differentiable, but not analytic, in the neighborhood of z = O. See the discussion
on page 54 of the text.
• 2.6 (a) The equilibrium points are the real roots of
0 = -z1 + a:r:2 -bz1z2 + zi, 0 =-(a+ b)z1 + bzf - z1z2

From the second equation we have
X1 [-(a+ b} + b:t1 - X2] =0 * Z1 =0 or Z2 = -(a+ b) + bzl

Substitution of x1 = 0 in the first equation yields
z2(z2 + a) =0 * z2 =0 or z2 = -a
Thus, there are equilibriuni points at (OtO) and (0, -a). Substitution of z2 = -(a+ b) + b::1:1 in the first
equation yields
Hence, there is an equilibrium point at ( ~$2, -1~t,•>).

(b)
8/ [ -l-b::1:2
8z =
-(a+b)+2bz1-Z2
1. z s (0;O)
-1
A= [ -(a +b) 0
a]
The eigenvalues of A are
,\: -1:l::y'l-4a(a+b)
2
The equilibrium point (0, 0) is a stable focus if 4a(a + b) > 1, a stable node if O < 44(a + b) < 1, and
a saddle if a(a + b) < 0.
2. :z: = (0,-a)
A = [ -1-~ ob 0a ]
The eigenvalues of A are ~ = ab and ,\ = -1. The equilibrium point (0, -a) is a saddle if b > 0 and a
stable node if b < 0.
3. z = ( 1~t:>' -1':rf1)
1 [ -1 + ab -,,S - a- 2b ]
A ::; 1 + b2 (a+ b)b2 -b(a + b)
- 20
The eigenvalues of A are

.\ = -1 ::I: y'l - 4b(a + b)
2
The equilibrium point { ~~t:>, -J:tf>) is a stable focus if 4b(a+b) > 1, a stable node if O < 4b(a+b) <
1, and a saddle if b(a + b} < 0.
The variOUI cues are summarized in the following table.
(O,O) {0,-a) (1:"t:', -lw).)

b > 0, 44· a+b) > 1, 4b(a+b) > 1 stable focus saddle stable focus
b > 0, 4a a+b) > 1, 4b a+b )<1 stable locus saddle liable node
b > o, 4 l stable node saddle stable focua
b > 0, 4a a+b) < 1, 4b o+ b1<1 stable node saddle stable node
b < 0, 4 + b > 0, 4a a+b) > 1 stable fOCU8 stable node saddle
b < 0, a + b > 0, 4a a+b) < 1 stable node stable node saddle
b < 0, a+ b < 0, 4h a+b) > 1 saddle stable node stable focus
b < 0, a+ b < O, 4a a +b) < l saddle stable node stable node
If any one of the above conditions holds with equality rather than inequality, we end up with multiple
eigenvalues or eigenvalues with uro real parts, in which cue linearization fails to determine the type of the
equilibrium point of the DODlinear system.
(c) The phase portraits of the three cases are shOWD in Figures 2.8 through 2.10.
i a = 1, = 1. The equilibrium points are
(0, O) stable focus
{O, -1) saddle
(1, -1) stable i>cus
The linearization at the saddle ill A = [ _~ -~ ] . The stable eigenvector is [ !] aiid the umtable
eigenvector is [ - ~ ]. They are 11led to generate· the stable and umtable trajectories of the saddle.
The stable trajectorie9 form a eeparatrix that.divides.the plane into two halves, with all trajectories
in the right half aprPO&Cbing (1, -1) and all trajectories in the left half approacl,ing (0,0).
ii a= 1, b == -½, The equilibrium points are
(O, 0) stable focus
(0,-1) stable node
(i11 i1) saddle
The linearizat.ion at tbe saddle is A, where (1 + b2)A • [ -~:~:~ ~!~:~ ]. The stable eigenvector
is [ -~::: ] and the UDStable eigenvector is [ ~::: ] . They are uaed to geuerate the stable and
UJL1tab1e trajectories of the saddle. The stable trajectories form a separatrix in the form of a lobe. All
trajectories outside the Jobe approac:h (0, -l)i all trajectories inside the lobe approach (0, 0).
ill a= 1, b: -2. The equilibrium points are
(0,0) saddle
(O, -1) stable node
(f, l) stable focus
The linearization at the saddle is A = [ -! ~ ]. The stable eigenvector is [ -~:=; ] and the
- 21
unstable eigenvector is [ ~::; ] . They are med to generate the stable and unstable trajectories of
the saddle. The stable trajectories form a separatrix mthe form of a lobe. All trajectories outside the
lobe approach (O. -1); all trajectories inside the lobe approach (l, ¼) .
..
2
,,r 0
-2
..._.. --2 0 2 0 5
X, x1
Figme 2.8: Exercise 2.6(i). Figure 2.9: Exercise 2.6(ii).
5...--.........- - - - - - - - - - .
,r 0
0 5
X,
Figure 2.10: Exercise 2.6(iii).
• 2.7 The system

z1 = z2, z2 = -z1 - z2 {-1 + ~ - zt + jgzn ·
has a unique equilibrium point at the origin. Linearization at the origin yields A = [ _~ !] whose
etgeovalues are 0.5 ¼ 0.866j. Hence the origin is unstable focus. The phase portrait is shown in Figure 2.11.
There are three limit cyclea. The inner limit cycle is stab~ the middle one is UD8ta.ble1 and the outer one is
stable. AU trajectories starting inside the middle limit cycle, other than the origin, approach the inner limit
cycle as t tends to infinity. AU trajectories starting outside the middle limit cycle approach the outer limit
cycle as t tends to ioftnity. Trajectories starting at the unstable focus or on the unatable limit cycle remain
there.
O=z2,
- 22
11-----
.
-111i..--..a...._....
0
., II
Figure 2.11: Exercise 2.7.
Hence
z1(l6-zf)=O => z1=0, ±2
There are three equilibrium points at (0, 0), (2, 0), and (-2, O).
: =[ _1:/i~ ~I]
0 1 ] , _ -1 ::t:jv'i
z e (0, 0) • A a: [ => "' - 2
. -1 -1
(0, 0) is a stable focus.
1
z = (2, 0) or (-2, 0) => A - [ O ] => A= -l±.Ji7
- 4 -1 2
(2,0) and (-2,0) are saddle points.
(b) The phase portrait can be sketched by constructing a vector field diagram and using the informa-
tion about the equilibrium points, espedal]y the directions of the stable and unstable trajectories &t the
aaddJe point.s. The stable and umtable eigenvectors of the linearization at the saddle points are
Vn.we ~ [
-tn
- 1
l ] = [ 1
-2.5616
] , 11-unnaMe =[
-yu
-1
1 ] =[ 1
1.5616
]
Find the directions of the vector fields on the two axes. On z 1 = 0, J. = [ -:t2
z2 ] . Bence the vector
field makes an angle -45 deg with the z1 axis and its magnitude increues with l:t:al- On z:a = 0, / =
- 23
[ O s• ] · Hence the vector field is parallel to the z2-a.xis, The sketch can be improved by finding the
-xi+ it .
vector field at other points.
• 2,9 (a)
(b) At the equilibrium points, we have
From the first equation, z2 = vc,, and from the second one, z1 =(Kc+ K1v1. + K,,v~)/KJ' This is the only
equilibrium point. Linearization at· the equilibrium point yields
[ 0 -1 ]
A= K,/m -(K1 + 2Katid + K,,)/m
whose eigenvalues a.re
~ _ -(K1 + 2K.v, + K,,)/m ± J(K1 + 2K.v4 + K,,)2/m2 - 4Kifm
- 2
H (Ki+ 2Kavd + K,,) 2 > 4mK1 , the equilibrium is a stable node, and if (Ki+ 2Kavd +K,,) 2 < 4mK,, the
equilibrium is a stable focus.
(c) For the given numerical values, the eigenvalues of the linearizatioJi are -0.0289 and -0.3461. Hence, the
equilibrium point is a stable node. The phase portrait is shown in Figure 2.12. All trajectorie.s approach the
stable node along the a1ow eigenvector of the node, which bu a small elope. Starting &om different initial
speeds, the trajectory reaches the desired speed witll no (or -very little) overshoot.
(d) The eigenvalues of the linearization are -o.1s1:;:1:o.2646;; hence the equilibrium point is a stable focus.
The phue portrait is lhown in Figure 2.13. All trajectories approach the stable focus. Notice the inc:reued
ovenboot compared with the previous cue. For example, starting at the iDitial state (z1 = 15, z2 = 10),
the speed reaches about 36 m/ ,u
before approaching the steady•nate of 30 m/ ,u.
(e) The phue portrait is shown in Figure 2.14. The local behavior near the equilibrium point is not
affected lince saturation will not be effective. However, far from the equilibrium point we can aee that the
state of the integrator, z1, '8.bs large values during saturation, resulting in au increaaed overshoot.
• 2.10 (a) Using the same aca1ing u in Example 2.1, the state equation is giW!D by
i-1 = O.S{-h(z1) + z2), i2 = 0.2(-:1 - 0.2z:, + 0.2}
where h(z1 ) is given in Example 2.1. The equilibrium points are the intenection points of the curves
=
z 2 h(z1) and : 2 = 1..:. 52:1 , Figure 2.15 shows that there is a unique equilibrium point. Uling the "roots"
commUJd of MATLAB, the equilibrium point was determined to be~= (0.057,0.7151).
8/ = [ -0.5h'(z1 ) 0.5 ]
lJz -0.2 -0.04
lJ f
In
I•=t = [
-0.2
- 1.046l _ 00·_504 ] => Eigenvalues= -0.9343, -0.1518 => stable node
(b) The phase portrait is shown in Figure 2.16. All trajectories approach the stable node. This circuit is
known as "monostable" becaUle it hu one steady-state operating point.
- 24
0 20 40 90 ID 100
~
Figure 2.12: Exercise 2.9(c). Figure 2.13: Exercise 2.9(d).
OL---------'
0 100 200 300
Figure 2.14: Exercise 2.D(e).
• 2,11 {a) Using the same sc.aling as in Example 2.1, the state equation is given by
t1 a 0.5[-h(z1) + :t,], ZJ = 0.2(-zi - 0.2z2 + 0.4) ·
where h(!:1 ) i8 given in Example 2.1. The equilibrium points are the iDienectioD points of the curve1
z2= h(~1) and z2 s 2-5%1. Figure 2.17 sbowa that there i8 a unique equilibrium point. Using the "roots"
oornrnanA of MATLAB, the equilibrium point..,. determined to be%= (0.2582,0.7091).
:! '·=- = [ !t.~1 3
-~-~ ] => Eigell~ues = 1.7618, 0.0155 ~ uutable node
(b) The phase portrait is shown iD Figure 2.18. The circuit has a stable limn cycle. All trajectories, acept
the amstant solution at the equilibrium point, approach the limit cycle. Thia circuit is known u "anable."
-o.:g.5
0.5 v- 1 0 0.5
Figure 2.15: Exercise 2.10: equilibrium point. Figure 2.16: Exercise 2.10: phase portrait.
- 25
.
,t
0.5 0 0.5 1 1.5

11 1
Figure 2.17: Exercise 2.11: equilibrium point. Figure ~-18: Exercise 2.11: phase portrait.
• 2.12 (a) Note that T12 = T21 =I .. .J-

.n1~
= llil = 1 and Tu = T22 =0
_L .. "Jlu
_L = Jtii = 0. Hence the
__L
state equation is given by
z1 = h(z1)[:r2 - 2f}(z1)] ~ fi(z), z2 = h(z2)[z1 - 27J(z2)J ~ h(z)

where h(z) = '>..car(wz/2) and 71(:r) = g-1 (z) = (2/,r.\) tan(.-z/2). Equilibrium ,om.ta are the intersection
pointaofthecurvesz2 = 217(2:1) andz1 = 2T](z2), Notethatf1'(0) = l/..\and71'(z) = (1/.\)sec2(1rz/2) ~ 1/'>...
Therefore, for ,\ 5 2, the two curves intersect only at the origin (0, 0). For ..\ > 2, there are three intersection
points &t (0, 0), (a, a) and (-a, -a) where O < a < 1 depends on .\. This fact can be seen be sketching the
curves and uaiDg aymmetry; see Figure 2.19. '!he partial derivatives of Ii and /2 are given by
: : = h'(z1)[z2 - 217(z1)] - 2h(z1}fJ'(z1), : : = h(z1)
~~2 = h(z2),
v.c;1
812
8:t2 = h'(:r2)[:r1
.
- 2T](z2)] - 2h(z2)11'(z2)
At equilibrium points, [%2 -2fJ(z1)J = 0 and-[z1 -211(z2)1 = O. Therefore, the Jacobian matrix reduces to
:!I ••(U)
=h(b) [ -2r(b) -2,,1'(b) l
where b = 0, a, or -a, depending on the equilibrium point.
~,
••(0,0)
= [ -,.2 !2 ] => Eigenvalues= -2± ,\
For .,\ < 2, the unique equilibrium point at (O, 0) is a stable node. For ..\ > 2, the equilibrium point (0, 0) is
a saddle. For A > 2 there are two other equilibrium pob:"8 at (a, a) and (-a, -a).
~JI
vZ ••(a,G)
= h(a) [ - 2,r~a) _ 2 1,(a) ] => Eigenvalues= h(a)[-2q'(a) ± 1)
fJ
It ii not hard to .ee from the sketch of the curves 2:2 =

2'1(zi) and z1 =
2TJ(z2) that at the intersection
point (a,a), the slope 2fJ'(o) > 1. Bence, (a,a) is a stable node. Similarly, it can be shown that (-a1 -a) is
a stable node.
(b) The phase portrait is shown in Figw-e 2.20. The stable trajectories of the saddle point at the ori-
gin form a aeparatrix that divides the plane into two regions. Trajectories in each region approach the stable
node in that region.
- 26
-0.5 0 0.5
Figure 2.19: Exercise 2.12: equilibrium point. Figure 2.20: Exercise 2.12: phase portrait.
• 2.lS (a) Using Kirchhoff's Voltage law, we obtain
,:h = -C1R1
1-[V2 - t11 - g(v,)]
Using Kirchhoff's Current law, we obtain
:i:2 = _!_
C2
{_..!:!.
R2 R1
- _!_fv, - Vi - g(t12)l}
Thus, the state equation is

1
X1 = CiRi f-z1 + z2 - g(z2)]
1 l 1 1
i-2 = C2R1 :r1 - C2R2 z:i - C2R1 :tz + 02R1 g(z2)
(b) For ~e given data, the state equation is given by
%1 = -%1 + Z2 - g(:r2), i2 = Z1 - 2z2 + g(z2)
g(:r2) =3.2:wz2 - 2.19sxl + o.666z~
The system has a Wlique equilibrium point at the origin. The Jacobian at the origin is given by
:1 21 ==O;sa :mO
= f ~l ~~22:] => Eigenvalues =0.117±0.993j
Hence the origin is an UD8table focus. The phase portrait is shown in Figures 2.21 and 2.22 using two different
scales. The system has two limit cycles. The inner limit cycle is stable, while the outer one is unstable. AU
trajectories starting inside the outer limit cycle, except the origin, approach the inner one. All trajectories
starting outside the outer limit cycle diverge to infinity.
• 2,14 The system is given by
:i:1 = z2
±2 = -kz1 - c:i:2 - 11(:r1, z2)
where
for lz2I >0
for :r2 = 0&jz1 I :$ µ.mg/k
for %2 = 0&lz1I > µ.mg/k
- 27
6---.--~---.. . . . .
4
o.s
-2
-0.5
-4
_,L-...;:::,,,_ _.....i:::::...._ _ __J
-6L....._ _ _ _,..__....__......_.
-1 -0.5 0 0.5
-5 0 5
Figure 2.21: Exercise 2.13. Figure 2.22: Exercise 2.13.
For x 2 > 0, the state equation is given by

:i:1 = :r2
:i:2 = -kx1 - c:r2 - µ1img
while, for x2 < 0, it iB given by

:i:1 = x2
:i:2 = -kx1 - c:r2 + µ1emg
In each half, we C&D determine the trajectories by studying the respective linear equation. Let us start with
.z2 > O. The linear state equation has an equilibrium point at (-µ1,mg/k,0). Shifting the equilibrium to
the origin, we obtain a linear state equation with the matrix [ ~k ~c ] , whose characteristic equation is
,\2 + c,\ + k = 0, where k ~ c are positive constants. The equilibrium point is a stable focus when 4k > cl
and a stable node when 4k $ d'. We shall continue our discussion assuming 4k > cl. 'Irajectories would tend
to spiral toward the equilibrium point (-µ1,mg/k,O). It will not actually spiral toward the point becaUBe
the equa.tion is valid only for z2 > 0. Thus, for any point in the upper half, we can solve the linear equation
to find the trajectory that should spiral toward the equilibrium point, but follow the trajectory only UJ1til it
hits the :i: 1-axis. For z 2 < 0, we have a similar situation·e:x:cept tnat trajectories tend t.o spiral toward the
point (µ1,mg/k, 0). On the x1 -axis itself, we should distinguish between two regions. If a trajectory bits the
x 1 -axis within the interval [-µ.mg/k,µ.mg/k), it will rest at equilibrium. If it bits outside this interval,
it will have z2 '# O and will continue motion. Notice that trajectories reaching the 2:1-axis in the interval
x 1 > µ,mg/k will be coming from the upper half of the plane and will continue their motion into the lower
half. By symmetry, trajectories reaching the :r1-axis in the interval z1 < -µ. mgI k will be coming from the
lower half of the plane and will continue their motion into the upper half. Thus, a trajectory starting far
from the origin, will spiral toward the origin, until it hits the xi-axis within the interval (-µ.mg/k, µ.mg/k].
The phase portrait is sketched in Figure 2.23.
• 2.15 The solution of the state equation
:i:1 = z2, z1 (0) =z10

i-2 = k, z2(0) = 2:20
where k = ±1, is given by

z2(t) = kt + z20
Z1(t) = ½kt2 + Z30t + %10
- 28
Eliminating t b_etween the two equations, we obtain
X1 = 2lkz22 +c
=
where c :z:10 -4/(2k). This is the equation of the trajectories in the :z:1-:z:2 plane. Different trajectories
correspond to different_ values of c. Figures 2.24 and 2.25 show the phase portraits for u 1 and u -1, = =
respectively. The two portraits a.re superimposed in Figure 2.26. From Figure 2.26 we see that trajectories
can reach the origin through only two curves, which are highlighted. The curve in the lower half corresponds
to u = 1 aod the curve in the upper half corresponds to u = -1. We will refer to these curves a., the
switching curves. To move any point in the plane to the origin, we can switch between ±1. For example,
to move the point A to the origm, we apply u = -1 until the trajectory hits the switclling curve, then we
switch to u = 1. Similarly, to move the point B to the origin, we apply u = 1 until the trajectory hits the
=
switching curve, then we switch to u -1. When the trajectory reaches the origin we can keep it there by
=
switching to u 0 which makes the origin an equilibrium point.
--2 0 2 4 0 2
x1 X, '
Figure 2.24: Exercise 2.15. Figure 2.25: Exercise 2.1S.
• 2.16 (a) The equilibrium points are the roots of
0 = x1 (1 - X1 - ax2), 0 = bx2(z1 - z2)
=
From the first equation, we have x 1 0 or x1 = 1- az2. Substitution of :1:1 =
0 in the second equation results
=
in x2 0. Substitution of X1 =
1- a:i:2 in the second equation results in .:r:2(1- az2 - x 2 ) = 0 which yields
- 29
_,
~
--4
--_-.,,,. -
..,.
-3 .a -1 0 1 2 3 ..
"•
x2 = 0 or X:z = 1/(1 +a). Thus, there are three equilibrium points at (0,0), (1,0), and (1/(l+a), 1/(1 +a)).
The Jacobian matrix is given by
aI = [
8z
1 - 2x1 - a:r2
bz2
-a:r1
6%1 - 2bx2
J
:1(0,0) = [ ~ g]. :L) = [~I .-,,-]. :!L. y¼,) = 1~.[ ~I =: l

At the equilibriUD1 point (0, 0) the matrix has a zero eigenvalue; hence linearization fails to determine the
type of the equilibrium point. At (1, 0), the equilibrium point is a saddle. At { r.f:;;-, 1
are
the eigenvalues !,.),
A _ -(1 + b) ::l: v'l - 2b + 1,2 - 4ab
1' 2 - 2(1 + a)
Hence, (cl;, r:f:a) is a stable node if 1 - 2b + b2 - 4ab > 0 and a st.able focus if l - 2b + b2 - 4ab < O.
½)
The phase portrait is shown in Figure 2.27. The equilibrium point (½, is a stable focus that attracts all
trajectories in the first quadrant except trajectories on the xi-axis or the :z:2 axis. Trajectories on the x 1-axis
move on it approaching the saddle point at (1,0). Motion on the x1-ax:is corresponds to the case when there
are no predators, iD which case the prey population settles at x1 = 1. Motion on the z2: a.xis corresponds
to the case when there are no preys, in which the case the predator population settles at :r:2: = O; i.e., the
predators vanish. In the presence of both preys and predators, their populations reach a balance at the
equilibrium point ( ½, ½) .
• 2.l'T (1) Assume & > 0 and let X1 = y, z2 - 'ii, and V(:c) = xr + z~.
f(z) · VV(z) = 2ez~(l - :cf - :z:~) = 2e~(l - V)
- 30
1.5 2
x,
Hence, J(z) · VV(z) ~ 0 for V(x) ~ 1. In particular, all trajectories starting in M =

{V(z) ~ 1} stay in M
for all future time. M contains only one equilibrium point at the origin. Linearization at the origin yields
the matrix [ ~l !]. Bence, the origin is unstable node or unstable focus. By the Poincare-Bendixson
criterion, there is a periodic orbit in M.
(2) Let V(z) = ~ + z~.

/(z) · VV{z) = 2z~(2 - 3zf - 2x~) = 4z~(l - zf - ~) - 2z~z~ ~ 4z~(i - zf - z~)
Hence, J(z) · VV(z) ;$ 0 for~+ zi 2:: l. In particular, all trajectories starting in M = {V(z) ;$ 1} stay
in M for all future time. M contains only one equilibrium point at the origin. Linearization at the origin
yields the matrix [ ~l ! ], whose eigenvalues are 1 and 1. Since /(z) is an analytic function of x, we
conclude that the origin iB unstable node. By the Poincare-Bendixson criterion, there is a periodic orbit in M.
(3) Let V(z) = 3zf + 2z1z2 + ~-

J(z) · VV(.z) = (6z1 + 2z2)z2 + (2z1 + 4z2)(-z1 + X2 - 2(x1 + 2:r2)~]
= -2z~ + 4%1%2 + 6z: - 4(:r:1 + 2z2)2z~
:::: -2(x? + ~) + 4x2(z1 + 2z2) - 4{z1 + 2z2)2x~
= -2(.z~ + ~) + 1 - [1 - 2z2(z1 + 2z2)]2
S -2(.zf + ~) + 1 :$ 0, for z~ + :i:~ ~ ½
Choose a constant c > 0 sudl that the surface V(z) = c contains the circle {zf + ~ = ½} in its interior.
Then, all trajectories starting in M = {V(:r:) $ c} stay in M for all future time. M contains only one
equilibrium point at the origin. Linearization at the origin yields the matrix [ ~l ! ],
whose eigenvalues
are (1 ±j./3)/2. Hence, the origin is unstable focus. By the Po~Bendixson criterion, there is a. periodic
orbit in M.
(4) The equilibrium points are the roots of

0 = x1 + x2 - z1 max{lz1I, lx2I},
From the first equation, we have z2 = z1 (max{lxd, lz21} -1). Substitution in the second equation results in
0 = -2x1 - xi (max{lz1 I, lx2I} - 1)2 = -z1[2 + (max{lz1 I, lz2I} - 1) 2) ~ :t1 = 0 ~ z2 =0
- 31
Hence there is a unique equilibrium point at the origin. Lineariz&tion at the origin yields the matrix
A = [ ~ !]
2 whose eigenvalues are 1 ± jl.4142. Hence, the origin is an unstable focus. Now consider
V(x) = x? +~.
VV · f = 2x1 (x1 + X2 - .1:1 max{lx1I, lx21}) + 2z2(-2x1 + x2 - x2 ma.x:{lx1I, lz21})
= 2x~ - 2:t1.z2 + 2z~ - 2(z~ + x~) max{lx1I, lz21}
= 2 [:i? Px - Jlxll~ max{lx1 I, lz2IH, where P = [ -~.S -~- 5 ]

The matrix P is positive definite with maximum eigenvalue 1.5. Therefore,
VV · f ~ 2(1.Sllxll~ -Uxll~max{lz1l,lz21}] < 0, for ma.x:{lz1l,lx21} > 1.5
Thus, by choosing c large enough, VV · f will be negative on the surface {V(z) = c}. Hence, all trajectories
starting in the set M = {V(x) ~ c} stay in M for all future time and M contains a single equilibrium point
which is unstable focus. It follows from the Poincare-Be ndixson's criterion that there is a periodic orbit in
M.
• 2.18
(a)
V =z2±2 + g(z1)±1 = -x2g(z1) + X2g(x1) = 0
=
(b) For small c > 0, the equation V(z) c defines a closed curve that encloses the origin. Since V = 0, a
trajectory starting on the curve must remain on the curve for all t. Moreover, from :t1 = z2, we see that the
trajectory can only move in the clockwise direction. Hence, a trajectory starting at any point on the closed
=
curve V(z) c must move around the curve until it comes back to the starting point. Thus, the trajectory
is a periodic orbit.
(c) Extension of (b) because V(z) = c is a closed curve for all c > 0.
(d) V(z) = ½zi +G(z1) = coustant. At z = (A,0), V = G(A). Thus
½x~(t) + G(z1 (t)) =G(A) =*' z2(t) =±y'2[G(A) - G(z1 (t))]
(e) Starting from i'1 =z2, we have
dt = dx1
J2[G(A) - G(z1)]
for z 2 ~ 0. Calculating the line integral of the right-hand side in the upper half of the plane from (-A, 0)
to (A, 0), we obtain
! - IA dy =*' T - 2v'2 / A dy
2- /_A vf2[G(A)-G(y)J - Jo v'G(A) - G(y)
where we h&ve used the fact that G(z1) is an even function.
(f) We can generate the trajectories using the equation in part (d). For each value of A, we solve the
equation to find x 2 as a function of z1. The function G(x1) has a minimum, a maximum, or a point of
inflection at each equilibrium point of the system. In particular, It has a miDimum at z 1 = 0 correspondi ng
to the equilibrium point at the origin. Starting from small values of A, the equation will have a solution
defining a closed orbit. As we increase the value of A, the equation will continue to define a closed orbit until
A reaches the level of a maximum point of G(z1). Fbr values of A higher than the maximum, the curves
will not be closed. Depending on the shape of G(x1 }; the equation may have multiple solutions defining
trajectories in different parts of the plane. The conditions of part(c) ensure that G{xi) will have a global
minimum at z1 0. =
- 32
• 2.19 The phase portraits can be generated by solving the equation of the previous exercise either graph-
icaJly or using a computer. We will only give the function G(xi) and calculate the period of the trajectory
through (1,0).
(1)
G(y) = fou sinz dz= 1- cosy
T=2V2 (1 dy
lo Joos 11 - cos 1
(2)
G(y) = (fl (z + z 3 ) dz= ½rl + ¼Y4
lo ·
G(y) -+ oo as IYI -+ oo and zg(z) 2: 0 for all z. Hence, every solution is periodic.
T == 2./2 (1 ~==d==y= =
lo J¾-b2-ty4
(3)
G(y) = lo" zs dz = is,4
G(y) -+ oo as 1111 -+ oo and zg(z) ~ 0 for all z. Hencet every solution is periodic.
T=2V2 /1 dy
lo /¼- ¼v4
• 2.~o
(1)
8/i + 8h =-l+o~O
8x18x2
By Bendixson's criterion, there are no periodic orbits.
0 = z1(-l + xf + x:), 0 =:z:2(-l +x~ + z;)

The system has an isolated equilibrium point at the origin and a continuum of equilibrium points on the
unit circle ~ =
+ ~ 1. It can be checked that the origin is a stable node. 'Iransform the system into the
polar coordinates z 1 = r oos 8, x2 = r sin 9. It can be verified that
r = -r(l -r2)
For r < 1, every trajectory start~ inside the unit circle approaches the origin as t-+ oo. For r > 1, every
trajectory starting outside the unit circle escapes to oo as t -+ oo. Thus, there are no limit cycles.
(3) The equilibrium points of the system are the roots of
0 = 1- Zt X~, 0= X1
These equations have no real roots. Thus, there are no equilibrium points. Since, by Corollary 2.1, a closed
orbit must enclose an equilibrium point, we conclude that there are no closed orbits.
- 33
(4) The z 1-axis iB au equilibrium set. Therefore, a periodic orbit cannot cross the :z:1-axis; it must lie
entirely in the upper or lower halves of the plane. However, there are no equilibrium points other than the
z 1-axis. Since, by Corollary 2.1, a periodic orbit must enclose an equilibrium point, we conclude that there
are no periodic orbit.s.
The equilibrium point are (±mr,O) for n = 0, 1,2, · · ·. Linearization at the equilibrium points yields the
matrix ( ~ ~ ] where a = ±1. Hence, all equilibrium points are saddles. Since, by Corollary 2.1, a
periodic orbit must enclose equilibrium points such that N - S = 1, we conclude that there are no periodic
orbits.
• 2.21
(a) Let
%1 +b
V(.z) = Z2 - --
:r:1 +a
The function V(z) is negative in D and the curve V(z) == 0 is the boundary of the set D.
f(:r:) • VV(z) = -c:z:1(.z1 +a)+ (Z1(b +a

- 0 ~2 [-z1 + z2(.z1 + a) - b]
Evaluating /(z) · VV(z) on the curve V(z) =0 yields

/(z) · VV{.z)lv(a)-O = -c:z1(z1 +a)< 0, 'v :r; E lJD
Hence, trajectories on the boundary of D must move into D, which shows that trajectories starting in D
cannot leave it.
(b)
8lJfi + 8812 = -1 + z2 < 0, 'f/ z eD
z1 z2
By Bendixson 's criteriOD, there can be no closed orbits entirely in D. Since trajeciories starting in D cannot
leave it, a closed orbit through any point in D must lie entirely in D. Thus, we conclude that there are no
closed orbits through any point in D.
• 2.22 (a) The value of :i:2 on the zi-axis ia .i'2 =l,zf ~ 0. Thus, trajectories starting in D cannot leave it.
(b)
l)Ji 8h
-8:c1 + 8z2
- =a- z2 - c ~ -(c - a) < 0, 'r/ z e D
By Bendixson's criterion, there can be no closed orbits entirely in D. Since trajectories starting in D cannot
leave it, a closed orbit through any point in D must lie entirely in D. Thus, we conclude that there are no
closed orbits through any point in D.
• 2.23
(ll)
~
lJz1
¾
+ lh2 = -a[2b- g(zi)] = {-2~
-a(2b- k)
for lz1I > 1
for lz1I 5 1
k < 2b => {)Ji + 8 /2 < 0, 't/ X

8:z:1 8x2
- 34
flz-0.15 11•-0.1
-0.5 0 0.5 -0.5 0 0.5

X1 x1
11=-0.05 I' •O.06
0.6 0.8
0.4 0 .4
0.2 0.2
.,r 0 J" 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-o.s 0 0.5 -0.5 0 0.6
xt x1
Hence, the origin is a stable focus for µ < 0 and unstable focus for µ > O. The phase portrait for different
values of µ is shown in Figure 2.29. For µ < O, there is a stable focus at the origin and unstable limit cycle
around the origin. The size of the limit cycle shrinks as µ tends to zero. For µ > 0, the origin is an unstable
focus and the limit cycle disappears. Hence, there is a subcritical Hopf bifurcation atµ= O.
(3) The equilibrium points are the real roots of

0 = z2, 0 = µ. - z2 - x~ - 2x1x2
x2 ;:;;Q => xf = µ
Whenµ.< O, there are no equilibrium points. Whenµ,> 0, there are two equilibrium points at (.,fii,0) and
(-Jji,0)
!!Lr,,,o} == [ -2~ -(1 +\.,/ii) ] => -\1,2 - -1, -2..jµ => (..jµ,O) is astable node
fJfl - [ 2-/µ
lJz (-v'ji,O) -
O 1
-(} - 2v'ji}
] => .\1,2 = -1, 2./µ => (,/µ,O) is a saddle
There is a saddle-node bifurcation at µ. = O.

(4) The equilibrium points a.re the real roots of
0 = X2, 0 = -(1 + µ.2).z1 + 2µ:t2 + JJ,X~ - X~X2
- 35
• 2.24 Suppose M does not contain an equilibrium point. Then, by the Poincar~Bendixson criterion, there
is a periodic orbit in M. But, by Corollary 2.1, the periodic orbit must contain an equilibrium point: A
contradiction. Thus, M contains an equilibrium point.
• 2.25 Verifying Lemma 2.3 by examining the vector fields is simple, but requires drawing several sketches.
Hence, it is skipped.
• 2.26
(1) Linearization at the origin yields [ ~ ~l ] . Hence, the origin is not hyperbolic. The index of the
=
origin is zero. This can be ea.sily seen by noting that Ji xf is always nonnegative. Clearly, the vector field
cannot make a full rotation as we encircle the origin because this will require Ji to be negative.
(2) Linearization at the origin yields [ ~ ~ ] . Hence, the origin is not hyperbolic. The index of the origin
is two. This can be seen by sketching the vector field along a closed curve around the origin.
• 2.27 (1) The equilibrium points are the real roots of
0 = z2, 0 = µ(.:z:1 + z2) - X2 - x~ - 3:r:fx2
:Z:2 =0 => 0 = X1 (µ - xn => X1 =0 Or X~ =µ

Forµ> 0, there are three equilibrium points at (0,0), (Jµ,O), and (-y"µ",O).
8/ [ O 1 ]
(}z = µ - ~ - 6z1z2 µ - 1- 3zf
81
8x I -[
(o,o) -
O 1 ]
µ µ- 1
=> ..\1,2 = -1, µ ~ (0,0) is a saddle
fJJI - [ -2µ
8:,; (y1i,O) -
O 1
-1-2µ
] ~ A1,2 = -2µ, -1 ~ {y'µ,O) is a stable node
Similarly, (-fo, 0) is a stable node. For µ < 0, there is a unique equilibriUJP point at (0, O).
81 /
8Z (D,O) = [ µO µ : 1 ] => ).1,2 = -1, µ ~ (0,0) is a stable node
Thus, there is superaitical pitchfork bifurcation at µ = 0.

0 = -zf + z2, 0 = -(1 + µ 2)x1 + 2µz2 - µx: + 2(z2 - µzi) 3

x2 = x: ~ 0 = {-1 + (xJ - µ)[µ + 2zUx~ - µ) 1}
X1 2
For all values ofµ, there is an equilibrium point at (O, 0). Atµ= 0, there are two other equilibrium points
=
at (a, a 3 ) and (-a, -a3 ), where a 8 0.5. It can be checked that these two equilibrium points are saddles.
By continuous dependence of the roots of a polynomial equation on its parameters, we see that there i! a
range of values of µ around zero for which these two saddle points will persist. We will limit our attention
to such values of µ and study local bifurcation at µ = O.
01 I =[
{J:z (0,0)
o 1 ]
-(1 + 1,i) 2µ => ..\1,2 = µ ± J.
- 36
flz-0.15 11•-0.1
-0.5 0 0.5 -0.5 0 0.5

X1 x1
11=-0.05 I' •O.06
0.6 0.8
0.4 0 .4
0.2 0.2
.,r 0 J" 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-o.s 0 0.5 -0.5 0 0.6
xt x1
Hence, the origin is a stable focus for µ < 0 and unstable focus for µ > O. The phase portrait for different
values of µ is shown in Figure 2.29. For µ < O, there is a stable focus at the origin and unstable limit cycle
around the origin. The size of the limit cycle shrinks as µ tends to zero. For µ > 0, the origin is an unstable
focus and the limit cycle disappears. Hence, there is a subcritical Hopf bifurcation atµ= O.

0 = z2, 0 = µ. - z2 - x~ - 2x1x2
x2 ;:;;Q => xf = µ
Whenµ.< O, there are no equilibrium points. Whenµ,> 0, there are two equilibrium points at (.,fii,0) and
(-Jji,0)
!!Lr,,,o} == [ -2~ -(1 +\.,/ii) ] => -\1,2 - -1, -2..jµ => (..jµ,O) is astable node
fJfl - [ 2-/µ
lJz (-v'ji,O) -
O 1
-(} - 2v'ji}
] => .\1,2 = -1, 2./µ => (,/µ,O) is a saddle
There is a saddle-node bifurcation at µ. = O.

(4) The equilibrium points a.re the real roots of
0 = X2, 0 = -(1 + µ.2).z1 + 2µ:t2 + JJ,X~ - X~X2
- 37
=0
=> 0 = -(1 + µ 2 )2:1 + µx~
X2
For all values ofµ, there is an equilibrium point at (0, 0). For µ > 0 there are two other equilibrium points
at (a,O) and (-a,O), where a= v'(l + µ 2}/µ.
~ tO,O) = [ -(1 ~ µ2) 2~ ] => At,2 =µ ± j

Hence, the origin is a stable focus for µ < 0 and UDStable focus for µ > 0.
8f I [
{):r; (±c,0) = 2(1+µ2}
0 I
(-1+µ2)/µ
]
=> .>.1,2
1 [
=2 - (I-
-µ-
µ2) ± ( 1 ~µ 2) 2+8(1 +µ 2)
Hence, (a, O) and (-a, 0) are saddle points. The phase portrait for different values of µ is shown in Fig-
ure 2.30. For µ < 0, there is a stable focus at the origin. For µ > 0, the origin is an unstable focus and
there is a stable limit cycle around the origin. The size of the limit cycle shrinks as µ tends to zero. Hence,
=
there is a supercritical Hopf bifurcation at µ 0. Also, as µ becomes positive, the saddle points appear on
=
the :i:1-axis at :i:1 ±v'(l + µ 2)/µ. The saddle points start at infinity and they move toward the origin as
µ increases, until they reach ±2 atµ= 1. Then they move again toward infinity. For p. = 0.2, the phase
portrait is shown in a larger area that includes the saddle points.
1& • -o. t 1& • 0.1

2 2
1 1
.t' 0 .,. 0
_, ~,
_,
-2
-2
., 0 1 2 -1
.,
0 2
11-0.2 1&•02
-1 0 2 0 s
., x,
Figure 2.30: Exercise 2.27(4}.
{5} The equilibrium points are the real roots of
0 = :i:2, 0 =µ(z1 + z2) - :r:2 - .z~ + 3xfz2
- 38
X2 =0 => 0 =%1 (µ - xn
For all values ofµ, there is an equilibrium point at {O, 0). Forµ> 0 there are two other equilibrium points
a.t (y"µ,O) and (-Jµ,O).
88! I
X (0,0)
= [ µ0 µ
~1 ] ~ A1,2 = -1, µ
Hence, the origin is a stable node forµ< 0 and a saddle forµ> 0.
a,, -[
-ax (±yii,0)
-
o
-2µ
(4µ -1)
1 ] .
=>
\
"'12 -
'
_ -(I-4µ)±J(1-4µ) 2 -sµ
2
The following table gives the type of the equilibrium points(,[µ, O) and (-Jµ, 0) for various positive values
ofµ..
Range Type
0 < µ < 0.067 Stable node
0.067 < µ < 0.25 Stable focus
0.25 < µ < 0.933 Unstab]e focus
0.933 < µ Unstable node
Thus, there is a supercritical pitchfork bifurcation atµ =

0. We also examine /J 0.25, where the equi- =
librium points (y'ii,0) and (-vP",0) change from stable focus to unstable focus. The phase portraits for
=
µ 0.24 and JJ = 0.26 are shown in Figure 2.31. As µ crosses 0.25 new stable limit cycles are created around
the points (.Jji,O) and (-v'Ji,O). Thus, there is a supercritical Hopf bifurcation atµ= 0.25.
~=0.26
1--------
-1 ' - - - - - - - - -
-1 -0.5 0 0.5 -1'--------'
-1 -0.6 0 0.5 1
X1 x1
( 6) The equilibrium points are the real roots of
Z2 = 0 => 0 = Z1 (µ - Z1)
There are two equilibrium points at (0, 0) and (µ, 0).
ail
{)z {O.Ol
= [ µo i ]
µ-1
=> -X1,2=-1, µ
- 39
Hence, the origin is a stable node forµ< 0 and a saddle forµ> D.
:!t,,,O) = [ ~µ -(,/+ l) ] => .\1,2 = -1, -µ

Hence, (µ, O) is a saddle forµ < 0 and a stable node forµ > 0. There is a. tra.nscritical bifurcation atµ= O.
1 l
0 =- -z1
T
+ tanh(.\z1) - tanh(..\z2), 0 = - -z2 + tanh(>.%1) + tanh(Ax2)
T
By adding and subtracting the two equations, we see that the equilibrium points are the intersections of the
two curves
z2 = -z1 + 2-rta.nh(Axi), z1 =X2 - 2rta.nh(Ax2)
Clearly there is an equilibrium point at the origin (O, 0). By plotting the two curves for different values of AT
(see Figure 2.32), it can be seen that the origin is the only intersection point. In fact, the two curves touch
each other asymptotically as AT ~ oo. Thus, we conclude that the origin is the only equilibrium point. Next
we use linearization to determine the type of the equilibrium point.
81 =
8z
r- ¼+ ,\co.1A,\s1)
,\ 1
-,\COlh,(..Xi,>
1 ,\ 1
]
c:osh"(>.z1} - -; + coah•(Xz3)
8/1
8x
·=[-¼+-'
z:=O ,\
-,\
-¼'+>. •
] Eigenvalues: (~T - l) :i: j>..
T
< 1 and unstable focus for >.r > l . To apply the Poincare-Bendixson
Hence, the origin is a stable focus for >.T
criterion when Ar> 1, we need to find a set M that satisfies the conditions of the criterion. We do it by
transforming the equation into the polar coordinates
,. = vri + x';. 8 = tan-l (::)

r=- !.r + cos(tJ)[tanh(,\r oos(9)) - tanh(,\r sin(B))] + sin(9)[tanh(,\rcos(I)) + tanh(,\r sin(B))J
T
Using ltanh(·)I $ 1, fcos(·)I $ 1, and lsin(·)I $ 1, we see that
r<
-
- .!.r +4
1'
Choosing r = c > 4-r, we conclude that on the circler== c, r < o. Hence, vector fields on r c point to the =
=
inside of the circle. Thus, the set M {r $ c} has the property that every trajectory starting in M stays
in M lot all future time. Moreover, Mis closed, bounded, and cont.aim only one equilibrium point which is
unstable focus. By the Poincare-Beodixson criterion, we conclude that there is a periodic orbit in M.
(b) The phase portrait is shown in Figure 2.33. The origin is an unstable focus and there is a stable
limit cycle around it. All trajectories, except the trivial solution z = 0, approach the limit cycle asymptoti-
cally.
(c) The phase portrait is shown in Figure 2.33. The origin is a. stable focus. All trajectories approach
the origin asymptotically.
(d) For ,\r < 1, there a stable focus at the origin. For ..\T > 1, there is an unstable focus at the origin
and a stable limit cycle around the origin. Hence, there is a supercritical Hopf bifurcation at ...\r = 1.
- 40
h•O.S h•2
3 3
I
2 I 2
J
f
,/" 0
_,
I
,
-2 I -2
,,
-3 -3
-2 0 z -2 0 2
", X,
1,. ,o h•30
3 , 3 ;
-·
-2
-3,__-~_ _ __,
-2 0 2
-2
_,....__...._
-2
___ 0 ll
_,
", .,
-1
-1 1 . 2
0 :::: 0 - Z1 - i:'4x1x2
"'+"'if, 0 = b:1:1 ( 1 -
From the second equation we have x1 :::: 0 or x2 = 1 + xf. The first equation cannot be satisfied with z1 = 0 .
. Substitution of x2 = 1 + xf in the first equation results in x1 a/5. Thus, there is a unique equilibrium=
point at ((a/5),_1 + (a/5) 2 ). The Jacobian matrix is given by
8/
fk = -::i]
- 1+:!' 1
OJI _ 1 [ -5 + 3(a/5)2 -4(a/5) ] _ 1 B

&x ((a/S),l+(a/S)3) - 1 + (a/5) 2 2b(a/5)2 -b(a/5) - 1 + (a/5)2
- 41
The characteristic equation of B is

s2 + fJs + 'Y = 0
where
/3 = 5 - 3(a/5) 2 + b(a./5), -y = 5[1 + (a/5) 2]b(a/5)
We have 'Y > 0. H fJ < 0, the eigenvalues will be real and positive or complex with positive real parts; hence
the equilibrium point will be unstable node or unstable focus. /3 is negative if b < 3(a/5) - 25/o.. To apply
the Poincare-Bendixson criterion, we need to choose the set M. Figure 2.34 sketches the two curves whose
intersection determines the equilibrium point. On the sketch we identify a rectangle with vertices at A, B,
C and D. On the line AB, ±2 > O; hence the vector fields point upward. On the line BC, ±1 < O; hence the
vector fields point to the left. On the line CD, ±2 < O; hence the vector fields point downward. On the line
DA, ±1 > O; hence the vector fields point to the right. Thus, taking the set M to be the rectangle ABCD, we
see that every trajectory starting in M stays in M for all future time. Moreover, M is closed, bounded, and
contains only one equilibrium point which is unstable node or unstable focus. By the Poincare-Bendixson
criterion, we conclude that there is a periodic orbit in M.
D ----~---,---~---,- C
I
'1 >0
,. ...____,.-:_z ---> ----
•
=
(b) For a= 10 and b 2, we have b < 3a/5 - 25/o.. The equilibrium point is {2,5) and it is unstable
focus. The phase portrait is shown in Figure 2.35. The system has a stable limit cycle. All trajectories,
=
except the equilibrium solution x (2, 5), approach the limit cycle asymptotically
(c) For a = 10 and b = 4, we have b > 3a/5 - 25/a. The equilibrium point is (2, 5) and it is a stable
focus. The phase portrait is shown in Figure 2.35. AD trajectories approach the equilibrium point asymp--
totically.
(d) For b < 3a/5 - 25/a, fJ is negative. Moreover, when l, is close to 3o./5 - 25/o., /3 will be close to
zero. Hence, 4-y > {J2 and the equilibrium point is unstable focus. As we saw from the phase portrait, there
- 42
,.,10,b-2
5 10 10
x1
is a stable limit cycle around the equilibrium point. For b > 3a/5 - 25/a, f) is positive. Once again, when b
is close to 3o/5 - 25/a, /3 will be close to 7.el"O. Hence, 4-y > ~2 and the equilibrium point is a stable focus.
There is a supercritical Hopf bifurcation at b =3a/5 - 25/a.
• 2.30 (a) The equilibrium points are the roots of
The first equation has two solutions: z1 = 0 or the solution of d = /Jm/£2/(k,,,. + z 2), which we denote by a.
When d < I'm, there is a unique solution ct. Substitution of z 1 = 0 in the second equation yields z 2 = z 21 •
=
Substitution of :t2 = ct in the second equation yields x 1 Y(x21 - o), which will be a feuible solution if
a S x21; that is, d $ 1JmX2f /(k,,,. + z21 ). Thus, when d S µ.,,.x2,l(km + X2J ), there are two equilibrium
points at (0, :i:21) and (Y(x:11 - o), a}. When d > J.'mX2J/(km + z21 ), there is a unique equilibrium point at
(O,x21),
Cml;nZt, _
(li,.+a:2 ]
- d- b:JhnZI
Y .. +z2l'
~-d
IJJ [
1s.. +z21
lJx to,z21) = -~... z,,

Y(4,,.+za1)
Hence, (O, Z2/) is a saddle if d < f.Sm,Z2/ /(km+ z21) and a stable node if d .> µ.,,.x21 /(k,,,. + z21 ).
{J'
lJx
I (Y(z21-a),o)
[
= _ -j
0
The eigenvalues are -d and -km1Jm(z21 - a)/(km + o)2. Ford< f.Sm,X21/(km. +:r:21), (Y(z21 -a),a) is a
stable node. For the given numerical data, 1,J,mZ2/ /(km+ z21) = 0.4878. When d > I'm, there is a unique
equilibrium point at (0, :e21 ), which is a stable node.
(b) The bifurcation diagram is shown in Figure 2.36. As d increases toward 0.4878, the saddle at (O, x 21 )
and the stable node at (Y(z21 - o),cr) collide and bifurcate into a. stable node at (O,z21 ).
- 43
s-------.---.-----,
Saddle Stable node
--------
Stable node
1 2 3 4 5
d 0.4878 x1
Figure 2.36: Exercise 2.30: Bifurcation diagram. Figure 2.37: Exercise 2.30.
(c) For d =
0.4, a =kmd/(fJm - d) =
0.4. Since d = 0.4 < 0.4878, there is a saddle at (0,4) and a
stable node at (1.44, 0.4). The phase portrait is shown in Figlll'e 2.37.
• 2.31 (a) Let
The equilibrium points are the roots of
From the first equation, z1 = 0 or µ(:t2} =d.
µ(z2) =d * ·:t1 = Y(z21 - z2)

By sketching the function µ(z2), it can be seen that if d < m.ax.s1 ~{µ(z2)}, the equation d = µ(:t2) will have
two solutions. Let 118 denote them by ct1 and ct2. In this cue, there are three equilibrium points at (0, z21 ),
(Y(:c21 -ai),a1), and (Y{:i:21-02), 02), provided z21 -01 and z21-a2 are nonnegative numbers. If one of
these numbers is negative, the corresponding equilibrium point is not feasible. If d > ma.xs,~o{µ(z 2)}, the
equation d = µ(z 2 ) has no solutions. In this case, there is only one equilibrium point at (0, :t21 ). The plot of
µ as a function of z2 is shown in Figure 2.38. By differentiation, it can be seen that µ has a. maximum value
(µ../km/lc1)/(2km + ,/km/k1) = 0.3455 at z2 = ./km/k1 = 0.4472. When d > 0.3455, there is a unique
equilibrium point at (0,4), and when d < 0.3455 there are three equilibrium points at (0,4), (0.4(4-a1),a1 ),
and (0.4(4- o 2),o2), where a1 < 0.4472 and 02 > 0.4472 are the solutions of d = µ(x2), In the case of a 2,
the equilibrium point (0.4(4- a2), a,) is not feasible if a 2 > 4. It can be checked thatµ= 0.1653 at z 2 4. =
Hence, ford< 0.1653, there are only two equilibrium points at (0,4) and {0.4(4- o 1 ),a1}. The Jacobian
matrix is given by
At z = (0, 4), the Jacobian matrix is

[ -d +*0.1653 0 ]
-d
- 44
When d > 0.1653, the equilibrium point is a stable node. When d < 0.1653, it is a saddle. At z =
{0.4(4- 01),oi), the Jacobian matrix is
<P{JY ]
-d-<PP
=
where fj (x21 - ai)(k,,, - k 1aD/µmoi > 0. The eigenvalues of this matrix are -d and -tPp. Hence, the
equilibrium point is a stable node. At z = (0.4(4- a-2),02) with 02 < 4, the Jacobian matrix is
d2-yY ]
-d- d2-y
=
where 'Y (x 21 - a:i)(km - . k1~)/"1na~ < O. The eigenvalues of this matrix are -d and -d2..,. Hence, the
equilibrium point is a saddle. In summary, we have the following three cases:
• When d > 0.3455, there is one equilibrium point at (0, 4) which is a stable node.
• When 0.1653 < d < 0.3455, there are three equilibrium points: a stable node at (0,4), a stable node
at (0.4(4 - a-1), ai), and a saddle at (0.4(4 - 02), 02)-
• When d < 0.1653, there are two equilibrium points: a saddle at {0,4) and a stable node at (0.4(4 -
et1),01),
Sl!_dqte _______S=ta=b=le__node=;;,._-
\
\
\
~ 0.2 ,8adcte
1E:o.1s \
~
0.1 ''
0.05 Stable node ..-,/
' \
0'---------------
0 2 4 6 8 10
0.1653 0.3455 d
Figure 2.38: Exercise 2.31. Figure 2.39: Exercise 2.31: Bifurcation diagram.
(b) The bifurcation diagram is shown in Figure 2.39. There is a saddl~node bifurcation at d = 0.3455. At
=
d 0.1653 there is a type of bifurcation that is not shown in Figure 2.28. A saddle point bifurcates into a
stable node and a new saddle is created.
=
(c) When d 0.1, there are two equilibrium points: (0,4) is a saddle and (1.59,0.0251) is a stable node.
The phase portrait is shown in Figure 2.40. The stable trajectories of the saddle are on the z 2.axis. All
trajectories in the first quadrant approach the stable node.
=
(d) When d 0.25, there are three equilibrium points: (0,4) is a stable node, {1.5578,0.1056) is a stable
node, and (0.8426, 1.8936) is a saddle. The phase portrait is shown in Figure 2.41. The stable trajectories
of the saddle form a separatrix which divides the first quadrant into two halves. All trajectories in the right
half approach the stable node {1.5578, 0.1056), while all trajectories in the left half approach the stable node
(0,4).
- 45
ds0.1 d .. 0.25
0 1 2 3 4 5
x1
Figure 2.40: Exercise 2.3l(c). Figure 2.41: Exercise 2.31(d).
d•0.5
0 2 3 4 5
~
Figure 2.42: Exercise 2.31{e).
(e) When d = 0.5, there is one equilibrium point a.t (0,4) which is a stable node. The phase portrait
is shown in Figure 2.42. All trajectories in the first quadrant approach the stable node.
- 46
Chapter 3
• 3.1 (1) The term lxl is not continuously di1ferentiable at x = 0, but it is globally Lipschitz. The term
x 2 is continuously differentiable, but its partial derivative is not globally bounded. Thus / = x2 + l:rl is
not continuously differentiable at z = 0. It is continuously differentiable on a domain that does not include
x =O. It is locally Lipschitz, hence continuous, but not globally Lipschitz..
(2) The term sgn(x) is discontinuous at x = 0. Thus, /(z) ::;;: x + sgn(x) does not have any of the four
properties in a domain that contains z = O.
(3) /(z) = sin{z) sgn(x) is globally Lipschitz. This can be seen as follows .. If both x and y are nonnegative,
we have
I/(z) - /(y)I = I sin(z) - sin(y)I $ Ix - YI
H z ~ 0 and y 5 0, we have
l/(z)-/(11)1 = lsin(x)+.sin(y)I = l2sin(½(z+11))cos(½(z-y))I $ l.z-yl

Other cases can be dealt with similarly to conclude that 1/(z) - /(y)I $ lz - 111 for all x,y. It follows
that J is both locally Lipschitz and continuous. It is not continuously differentiable at x = 0 because
lim3 -+o+ /'(z) = +l while lims-.o- /'{:r) = -1.
=
(4) /(z) -x + asinx is continuously differentiable. Hence, it is locally Lipschitz and continuous. f; =
-1 + a cou: is globally bounded. Hence, it is globally Lipschitz.
=
(5) /(x) -x + 2lxl is not continuously differentiable. It is globally Lipschitz because both x and lxl are
so. Hence, it is locally Lipschitz.
(6) /(z) = tan(x) is continuously difi'erentiable in the open interval -1r/2 < z < 1r/2. Hence, it is locally
Lipschitz and continuous in the same interval. Its derivative sec2 (x) is not globally bounded; hence, it is not
globally Lipschitz.
(7) The function tanh(y) is continuously differentiable and its derivative l/cm,h2 {y) is globally bounded;
hence it is globally Lipschitz. Clearly, the lineac function 11 is both continuously differentiable and globalJy
Lipschitz. Hence, f has all four properties.
(8) J is not continuously differentiable due to the term lx2I in Ji. Check the Lipschitz property component
by component. Ii is globally Lipschitz as can be easily checked. '2 is continuously differentiable, but its
partial derivatives are not globally bounded. Hence h is locally Lipschitz but not globally 80. Since both Ii
and '2 are locally Lipschitz, so is /. Since f is locally Lipschitz, it is continuous. / is not globally Lipschitz
since h is not so.
• 3.2 (1)
f(:i:) = [ . :Z:2 A: ] ~ 8/ = [
-
0 1A: ]
- f sm:z:1 - ;:z:2 + ;:,iT 8x - f cos:z:1 - iii
[8/ /th] is globally bounded. Hence/ is globally Lipschitz, which implies that it is locally Lipschitz on D,,
for any r > O.
(2)
- 47
[8J /8x] is continuous everywhere; hence it is bounded on the bounded set Dr. Thus/ is locally Lipschitz
on Dr for any finite r > 0. It is not globally Lipschitz since [8/ /oz] is not globally bounded.
(3)
f'/(X1, x2) is discontinuous at x2 = 0. Hence it is not locally Lipschitz at the origin. This means it is not
locally Lipschitz on Dr for any r > O.
(4)
/( ) _ [ Z2 ] 8/ _ [ 0 1 ]
Z - -xi - .s(l -tj)z2 ~ ax - · -1- 2ez1z2 -e:(l - xi)
[81/Dx] is continuous on D,.; hence / is locally Lipschitz on D,. for any r > O. [8J /8x] is not globally
bounded; hence / is not globally Lipschitz.
(5) Let X = [ea,~1.~JT.
0mz1 + k,,z2r(t) + kpx3(X1 + Ym(t)) ] lJf [ Gm+ k,,za kpr(t) k,,{z1 + Ym(t)) ]
J(t,x) =[ -")'X1r(t) => - = -1r(t) 0 0
-"}':t1(z1+S,m(t)) Bx -")'(2:r1+S,m(t)) 0 0
[8//lJx] is continuous and bounded on D,. for bounded r(t) and Ym(t). Hence/ is locally Lipschitz. It is
not globally Lipschitz since [8//8z] is not globally bounded.
(6)
/(z) = Ax - B,p(Cx)
where ¢(·) is a dead-zone nonlinearity. The dead-zone nonlinearity is globally Lipschitz. Hence f is globally
Lipschitz, which implies that it is locally Lipschitz on D,. for any r > 0.
• 3.3 For each :i:o E R, there exist positive constants r, L1, ~, k1 , and k2 such that
lfi(z) - /i(s,)I S L1lz -yl, 1/:i(z) - '2(s,)j S ~l:r -111, 1/i(z)I $ k1, lh(z)I $ k2
for all x,i, E {:r E .RI jz - zol < r}. For f =Ii+ /2, we have
1/(x)- f(y)I = 1/i(z) - /i(y) + /2(:r) - '2(11)1 $ L1l:r - 111 + L,lx - 111 $(Li+ ~)li-111
For f =/1'2, we have
1/(:r) -/(y)J = lfi(z)/2(:r) - /i(11)'2(11)l = lfi(x)f:a(z) - Ji(x)'2(11) + fi(x)'2(11) - /i(11)h(y)I
5 1/i(z)l lh(x) - h(11)I + lh(s,)j lfi(:r)- /i(y)/ 5 k1l.,lx - YI+ k,L1lx -yl
$ (k1!,, + k,L1)lz -111
For/= '2 o fi, we have
lf(x) - /(11)1 = l'2(!1(z)) - /2(/1(11))1 $ ~1/i(:r) - fi(y)I $ L2L1lz -yl
• 3.4 The function / can be written as /(:r) = g(x)K:rh(tp(z)) where
h(1/I) = {;, l and ¢(:r) = g(z)IIK:rll

~· if,t,<µ
The norm function IIK:rll is Lipschitz since

I IIK:rlJ - IIK111! I$ IIKx - Kyll S IIKII llz -11II
- 48
Using the previous exercise, we see that ¢(z) is Lipschitz on any compact set. Furthermore, g(x)Kx is also
Lipschitz. Thus, J(x) will be Lipschitz on any compact set if we can show that h('f/J) is Lipschitz in 'f/J over
any compact interval {O, bJ. Now if 1/)1 ~ µ and 1/12 ~ µ, we have
H ¢2 2: µ a.nd 1/11 < µ , we have

T/>z-µ ¢2-'¢1
lh(¢2) - h(t/11)1 = 1-~1 - -µ111 -1 = -
= -µ - ~ - < ----.........:..
µ'¢2 - ~
1
< -1""2 - 1/J1 I
-~
U ¢ 1 < µ and 1P2 < µ, we have
lh(t/12) - h(¢i)I = I.!.µ - 11=0:52-11/12

µ µ2 - T/J1I
Thus h(1/J) is Lipschitz with a Lipschitz constant 1/µ.2 .
• 3.5 There are positive constants c1 and c2 such that
Suppose
IJ/(y) - /{x)llcr :5 LallY - xlla
Theo
llf (y) - /(:c)ll1 :5 c2ll/(y) - f (x)lla :5 e:iLollY - xlla :5 c2c~0 IIY - zllB
Similarly, it can be shown that if f is Lipschitz in the /J-norm, it will be Lipschitz in the a-norm.
• 3.6 (a)
z(t) = :ro + £ /(-r,:r(-r)) d-r
llx(t)II :5 IIXoll + It llf(r,x.(T))II dr

Ito
:S llxoll + t [k1 + k2llz(T)II) dT
J,o
= llzoll + k1 (t - to) + ~ L
llz(T)II dr
By Gronwall-Bellman inequality
llx(t)II :5 llzoll + k1 (t - to) + £ [llzoll + k1 (s - to))it2eA:a(t-•) ds
Integrating by parts, we obtain
k1
Uz(t)ll :5 llxollexp(k2(t - to)J+ k 2 {exp[k2(t - to)]- 1}, Vt~ ta
(b) The upper bound on Jlx(t)II is finite for every finite t. It tends to oo as t-+ oo. Hence the solution of
the system cannot have a finite escape time.
- 49
• 3.T It can be easily verified that /(z) is continuously differentiable. Hence, local existence and uniqueness
follows from Theonmi 3.1. Furthermore,
llg(z)ll2 1
II/(z)ll2 = 1 + llg(z)II~ $ 2
Hence
1
]lx(t)ll2 :5 llzol12 + 2(t - to)
which shows that the solution is defined for all t 2'. t0 •
• S.8 It can be easily seen that /(z) is continuously differentiable and
for some positive constants k1 and k2. Apply Exercise 3.6,
• 3.9 Due to uniqueness of solution, trajectories in the plane cannot intersect. Therefore, all trajectories
starting in the region enclosed by the limit cycle must remain in that region. The closure of this region is a
compact set. Therefore, the solution must stay in a compact set. Apply Theorem 3.3.
• 3.10
where R = 1.5, u = 1.2, and the nominal values of C and L are 2 and 5, respectively. Let ). = [C, LJT. The
Jacobian matrices [8/ /8z) and [IJ/ /8.\], are given by
8/ =[ - -h\(zi) bR ] , 8/ =[ - b[-h(z1) + z2] 1 0 ]

/Jz - r - r 8>. 0 - p(-x1 - .Rz2 + u)
Evaluate these Jacobian matrices at the nominal values C = 2 and L = 5. Let
s- -
8z I -[
{}). nomln,t,I -
zs
z, Ze
Z& ]
Then
s= 8 'j
lJz aamlaal
81 j
s+ lJ). aamlaal '
S(O)=o
The augmented equation (3.7) is given by
%1 = 0.5[-h(zi) + z2)
:i:2 = 0.2(-.xi - l.Sx2 + 1.2)
:i:3 = o.s[-h'(:t:1):t3 + z,] - 0.25[-h(zi) + z2)
:i:, = 0.2(-zs - l.Sz,)
:h = O.S[-h'(z1)Z& + xs]
ze = 0.2(-:r11 - I.Sze) - 0.04(-:i:1 - 1.Sx:;i + 1.2)
with the initial conditions
- 50
• 3.11
.i1 =x2, X2 =-X1 + e(l - xnx2
Denote the nominal values of Eby e0 • The Jacobian matrices [8J /8x] and [8/ /&], are given by
8/ [ 0 1 ] 8/ [ 0 ]
8x = -1 - 2Ez1x2 e(l - zr) ' 8e :;::: (1 - ~)z2
Let
Then
s = 8x
{)JI
nominal
s + 81 1
8£ nominal
' S(D) = 0

Z1 = X2
i:2 = -%1 + eo(l - xnx2

Z3 = X4
i:4 = - (1 + 2EoX1X2] %3 + Eo{l - X~)X4 + (1 - :enz2

• 3.12
. =
z2 -E ( :r.:1 -x2+ 1~
3 3)
Denote the oominal values of e by eo. The Jacobian matrices [8//&) and [8//&], are given by
{}J [ 0 ¼. ] 8/ ·[ - :r:r:2 ]
Bx = -E e(l - ~) ' Oe = - (xi - x2 + ½xi)
S-
- 8E
8zl nominal -
_[ Z3] X4
Then
8= {}JI S+ fJJI , 8(0)=0
{Jx nominal lJE nomiu.J
The augmented equation (3. 7) is given by
1
:i:1 = -z2
co
:i:2 = -Eo ( X1 - Z2 + ~x:)
1 1
:i:3 = -x.-
Eo
- -z2
t1
X4 = -eozs + eo(l - x~)x.- - (i:1 - z2 + ~zf)
- 51
• 3.13
Let >. = [a,b,c]T. The nominal values are ao = 1, bo = 0, and eo = 1. The Jacobian matrices [8//IJx) and
[8/ /8>.], a.re given by
l}j = [ 1+:,s~ -x2 -xi]' BJ= [ 1+!i.f O O ]

{n 2bz1 -C a>. 0 zi -X2
Let
Then
s= 81 j
{Jz nominal
81 /
s + 8:A nominal
, sco> = o
:i:1 = tan- 1 {xi) - z1z2

z2 = -x2
%3 = ( l2
1
+x1
- Z2
) X3 - X1 X4 +~
X1
+x1
=
(1:
%4 -:&4
z11 = xf - .x2) X& - X1Xe

.ia = -xe +x~
:i:7 = (l: z¥ - Z2) X7 - X1X1
:i:s = -za -:z:2
z1(0) = X10, z2(0) = z20, zs(O) = Z4(0} =z11{0) = xa(O) = z1(0) = zs(O) =0
• 3.14 (a) Let p = [ ~ ] be the vector of parameters.
B =~ = [ -(l/r2)z1 z1/ cosh2(AXi) - x,/ cosh2(>.z2) ]

8p -(l/r3)z2 z1/ cosh2{.\z1) + x2/ cosh2(AZ2)
The sensitivity equation is given by
S = AoS + Bo, S(O) = 0
where Ao and Bo are evaluated at the nominal parameters. This equation should be solved simultaneously
with the nominal state equation.
- 52
(b)
rr = X1X1 + X2X2
= -(1/T)r2 + xi[tanh{,\x1) - tanh(.\x2 )] + x2[tanh(Az1) + tanh(.\xi)]
= -(1/r)r2 + rcos(6)(tanh(,\xi) - tanh(>.x2)]
+ rsin(8)[tanh(.\x1) + tanh(-\xi)]
< -(1/'r)r2 + 2r(I cos(B)I + Isin(8)1)
$ -(1/r)r2 + 2V2r
(c) By the comparison lemma, r(t) :$ u(t) where u satisfies the scalar differential equation
u = -(1/r)u + 2\/'2, u(O) = r(O) = llx(O)ll2

The solution of this differential equation is
u(t) = exp(-t/T)llz(O)ll2 + lot e~[-(t - u)/T]2v'2 du

= exp(-t/r)llz(0)ll2 + 2v'2r(l - exp(-t/r)]
• 3.15 Let V = llxlli = z? + Xi· Then

V
'
= 2x1:i:1 +2x2:i:2
~-2 2 U1Z2 4x1X2
= -=- 1 - 2x 2 + - - + ~
l+z~ l+Xi
lz2I lz1I
S -2V+4lx11 1 +~ +4lz21 1 +zi
< -2V + 2lx1I + 2lz2I ( since 1 !~ 2 S ~)
< -2V + 2\/'2fl (since 11:z:lh $ v'fillzll2}

Taking W = v'V = llxll2, we see that for V # 0,
.
W=-<-W+v2
v rn
2.;v-
At V = 0, we have
IW(t + h) - W(t)I = IW(t + h)I = .!.nx(t + h)ll2
h h h
Similar to Example 3.9 of the textbook, it can be shown that
11t+h
lim -h
h-+O+ t
llf(z(r))ll2 dr =0
Thus D+W(t) $ -W(t) + v'2 for all t ~ 0. Let u(t) be the solution of the differential equation
u = -u + ./2, u(O) = llx(O)ll2

By the comparison lemma,
- 53
• 3.16 Let 11 = x 2 •
. 2xx=-
. 2z 2 +-
2xsin t 2 1
v= 1- - 2 <- v+
+x -
let u(t) be the solution of the differential equation
u = -2u + 1, u(O) = 2
Then
v(t) :'.5 u(t) = 2e-21 + fo
t e-_2(t-T) dr =
1 + 3e-2t
2
Thus
• 3.17 (a)
!!._XT X
dt
= 2zT:X = 2zTf(t,x)
l!zTzl ::5 21lzll2ll/(t,z)ll2 ::5 2Lllzll~
(b) Let V(t) = zT(t)x(t) and Vo= zZ'zo, then from part (a) we have
-2LV(t) ::5 V(t) ::5 2LV(t)
Divide through by V(t), multiply by dt, and iiltegrate to obtain
- 1to
t 2Ldt$
Vo
1v 1' dV
V$
to
2Ldt
-2L(t - to) ::5 ln ( V(t))

Vo ::5 2L(t - to)
Vo exp(-2L(t - to)] :S V(t) :S Vo exp[2L(t - to)]
Taking the square root yields
Uzoll2 exp[-L(t - to)] :5 llx(t)ll2 :$ llzolb exp[L(t - to)]

• 3,18 Let z(t) = y(t)e'"(t-to). Then
z(t) ~ A:1 + [ e4 <.,.-to>[~y(r) + k3] dr = k1 + l [k-iz(r) + k3e°'<r-to)] dr
= k1 + "21.t z(r) d-r +

to
(k3/a) [expcr(t-te) -1]
From Gronwall-Bellman inequality,
By evaluating the integral, it can be shown that
z(t) ::5 k3 ea(t-to) +

0
(1. ks)
a
1 _ ei:t(t-t0 ) + k2k3
o(o-k2)
[ea(t-te) _ ei:t(t-to)]
- 54
Hence
y(t) = z(t)e-01(t-to) < ks (1 + k2 ] + [k1 - ka - k2ks ] e(A:2-a)(t-to)

- a (a - k2) o o(o - k2)
= ks + [k1 - ks ] e(A:2-a)(t-to)
(o-k2) (a-k2)
= k -(a-ki)(t-to)
1e
+ k3
(a_~)
[1 _e -(a-ll2)(t-to)]
• 3.19 Choose the covering of S as described in the hint. Within each neighborhood we have
If x,y e SnN(ai, r.) for some i, then the Lipschitz condition holds with L = L,. Otherwise, llx-yll ~ min; Ti.
Since f (x) is uniformly bounded on S, we have
IIJ(x) - /(y)Jl :SC, V x,y ES, C > 0
Therefore, whenever llx -yll ~ min,r,, we have

C
llf(x) - /(11)11 :S --:--llx -yll
mm,r,
Take
• 3.20 We have
llf (x) - /(11)11 :S Lllx - ylJ, V x, 11 E W
Given c > 0 1 let o = e/L. For all llx - 1111 < 6, we have 11/(z) - /(t,t)II < U = e:, which implies uniform
continuity.
• 3.21 The vector 1J is defined only for x :/: O. For z =0, we can take y as any vector with IIYII, = 1. Now,
for x :/: ·o we have
For p = oo, take

=
if i argmax lx,I
otherwise
• 3.22 If
11:~(t,z)II :SL, V (t,:z:) E (a,b] x R,R
then, from Lemma 3.1,

11/(t,y) - f(t, x)II :S Lllx - i,11, 'Ir/ (t,x) E [a,b] x Rn
- 55
Alternatively, suppme / (t, x) is globally Lipschitz. By the mean value theorem
fi(t, y) - /i(t, x) a;
= 8J, (t, z) (y - .z)
where z =ax+ (1-o)y and O < c:r < l. Then
fl :(t,z} (y - x)II = llfi(t, y) - f;(t,x)II $ L,IIY - xii, V x, y ER", Vt E [a,b]
Hence
jj¥z,-(t,z) (y-x)II
llu-xll ~Li, 'vx,yeR",'vte[a,b]
Taking 11 = /Jx with /J > 1, we have z = [cr + {l - o).8]x ~f "'fX, and
,,~(t,yx) (/J- l)xll

(,8 _ l)llzll '5 Li, V z E R", V t E (a, b)
Thus
fl~(t,-yx) zjJ
llxll ~ L;, 'v x E R'\ V t E [a, b]
By letting /J approach 1, we conclude that
which shows that

j ~~(t,z)II $ L;, 'v z e R", Vt E [a,b)
Since this inequality holds for every 1 $ i :S n, we conclude that the Jacobian matrix [8/ /8x] is globally
bounded.
• 3.23 Set g(u) = J(rrx) for O :S u :S 1. Since D is convex, ux E D for O :S u :S 1.
g'(u) = l:Jf (uz) &x = 81 (uz) z

&x 8u 8:i
f(x) = f(x) - /(O) =g(l) - g(O) = fo 1 g'(u) de,= fo 1 !! (o-z) de, x
• 3.24 (a)
V(t,x) IJx (t,o-x) d4 x :S lor1 11 av

= Jofl {JV /1 c.u de, llxll2 :S 21 C4llxll 2
&x (t,o-x) II llxll d<, '5 Jo
(b) Since
we must have c1 :S ½c...

(c) Consider two points z1 and z2 such that oz1 + (1 - o)z2 ~ 0 for all O :So :S I; that is, the origin does
- 56
not lie on the line connecting z1 and z2. The Jacobian [lJW/8x} is defined for every x = oz1 + (1 - o):e2
and given by
aw = ---:i==-(t
- ( t x)
1 av x)
ox ' 2y"V(t,z) 8x '
By the mean wlue theorem, there is o• E (0, 1) such that, with z = a*x1 + (I - a•)x 2 ,
W(t,x2)- W(t,xi)
aw
= -;-(t,z) (x2 -xi)= ~ -
1 av (t,z) {x2 -xi)
ux 2y V(t,z) 8 x
Hence
1 4
IW(t,:t2) - W(t,.2:1)1 :$ 2Jci" llzll c.&llzll ll:c2 -z1II S 2v'ci' llx2 - xiii
Consider now the case when the origin lies on the line connecting z1 and x2; that is, 0 =a 0x 1 + (1- ao)x2
for some Qo E [_O, 1). We have
IW(t,x2) - W(t,O)I = IW(t,x2)I = y"V(t,z2) S fi11x2H
fW{t,zi) - W(t,O)I = IW(t,z1)I = v'V(t,x1) S J¥11x11l
IW(t,x2) - W(t,z1)I = IW(t,:z:2)- W(t,O) + W(t,O) -W(t,xi)I s flrnx2II + 11:z:ill)

Since the origin lies on the line connecting x1 and z2, we have llz2II + 1lx1II = ll:i:2 - z1II, We also have
1 :5 Jc./2ci- Therefore,
Thus, the preceding inequality is satisfied for all x 1,z2 ED.
• 3.25
(a)
x(t) =z(o) + Lt f(T, x(r)) dr, V [o, t) C [to, T)
Since f(t,x} is piecewise continuow in t and continuous in x, there exists a constant M > O such that
11/(t,x{t))II SM for all t E (to,T). Therefore
llx(t)-.z(a)II = lllt /(r,x(r)) dr/1 S Lt M dr = M(t - a)
which shows that x(t) is uniformly continuous on [to, T).

(b)
x(T) = z(to) + Um {' f(-r,z(T)) d,,-
t-+T J,D
= x(to) + lt,IT J(r,x(r)) dr
since z(t) is uniformly continuous. Thus
z(t) = z(t0 ) + l.t'• /(1'1 :z(-r)) dT, Vt E {to,T}
is a solution on[to,TJ. Since Wis closed, z(T) E W.

(c) Apply the local existence and uniqueness theorem at (T, z(T)).
- 57
• 3.26 Suppose there is no such t. Then, y(t) e W for all t E [to,T). From the previous exercise we can
extend the solution beyond T, which contradicts the claim that (to, T) is the maximal interval of existence.
• 3.27 Set y(t) = z1 (t) - x2(t) and µ = Pi + µ2.

lbi(t)ll = llx1 (t) - x2(t)II
= llz1(t) - Ji{t,z1(t)) - d:2{t) + /:i(t,z2(t)) + fi(t,zi(t))- h(t,z2(t))II
< µ1 + P2 + L!lz1 (t) - z2(t)U = µ + llt1(t)II
lly(t)II = ll11(to) + £Ji(s) dsll ~ "f + [ ll1i(a)II ds
~ 1' + µ(t - a)+ 1t Llly(,)11 d,
Application of Gronwall-Bellman inequality yields
1
llu(t)II ~ 1' + µ(t - a)+ ..t h + µ(s - a)]Le1•• L d
T els
After integrating the right-hand side by parts, we obtain
lly(t)II $ -yeL(t-a) + I [eL(t-11) - 1]

• 3.28 Let
z(t) = zo + l f(a,x(s)) ds, y(t) = Xo + rt f(s,y(,)) ds
lto
where fo ~ to- Then
x(t) - y(t) = 1°
a.
~
/{s, x(s}) ds + f
ft~
t
[f(s,s,(1)) - /(s, x(s))] ds
We have 11/(a,z(s))JI $ M for all t E [to, t1], and IIJ(a, v) - J(a,z)II S Lllz- i,11. Therefore
lfx(t) - y(t)ll 5 M(t~ - to)+ {' Lllz(s) - y(,)11 d8

J,, 0
Hence, over any compact interval of time, we have
llx(t) - y(t)II :5 K(t:, - to)

• 3.29
z = J(t,z), :z:(to) = '1
Setting y =x - 1/, we obtain
'ii = /(t, 11 + TJ), y(to) = 0
Since f is continuously duferentiable in z, the solution will be continuously differentiable in fl. Let
y,,(t, 71) = by~ '1) = Oz~ '11) - I== x,,(t, '1) - I
- 58
From (3.4)
where
A(t, TJ)
aJ (t, y(t, '7) + 27) = at
= 8y ax (t, x(t, TJ))
B(t, TJ) = a, Bf
81] (t, y(t, 17) + 11) = 8x (t, x(t, 77)) = A(t, 71)
Thus, x,,(t, 11) satisfies the variational equation
• 3.30
z(t,a,71) = 1J + 1' J(s,x(s,a,71)) ds
Therefore
Za(t) + z,,(t)J(a, 71) = 1t .{ ~ (s, z(s, a, 77))(:i:G{s) + x,,(s)/(a, 71))} ds
Differentiating with respect to t, we see that Za(t) + x,,(t)f(a, 17) satisfies the differential equation
with initial condition

z.(a) + z,,(a)f(a, 17) = -/(a, 11) + /(a, 7}) = 0
Thus
• 3.31 Put
z(t) =z(a) + lt f(s,y(s)) ds
so that z(a} = z(a) and z(t) :$ r,(t) for a:$ t :S b.

i = f (t, y(t)) :$ J(t, z(t)}
From the comparison lemma, we conclude that
z(t) :$ .:r:(t) => u(t) :$ x(t), 'ti/ a::; t :Sb
- 59
-60
Chapter 4
• 4.1
(1) asymptotically stable (2) unstable (3) asymptotically stable
(4) unstable (5) stable (6) unstable
• 4.2 Let /(x) = azP + g(x). Near the origin, the term axP is dominant. Hence, sign(/(x)) =
sign(axP) .
Consider the case when a < 0 and pis odd. With V(x) = ½z2
as a Lyapunov function candidate, we have
V = z[azP + g(x)J :S az'*1 + klzl'*2

Ne&r the origin, the term ax'*1 is dominant. Hence, V(x) is negative definite and the origin is asymptotically
stable. Consider now the case when a > 0 and pis odd. In the neighborhood of the origin, sign(J{x)) =
sign(x). Hence, a trajectory starting near z = 0 will be always moving away from z =
0. This sh01VS that
the origin is unstable. When p is even, a similar behavior will take place on one side of the origin; namely,
on the side x > 0 when a > 0 and x < 0 when a < 0. Therefore, the origin is unstable.
• 4.3 (1) Let V(x) = {l/2)(zf + xJ).

V = z1(-z1 + z1z2} -z;
In the set {llzll2 :S ~}, we have lz1I :Sr. Hence,
Vis negative definite for r < 2. Thus, the origin is asymptotically stable. To investigate global asymptotic
=
stability, note that the solution of the second equation is z 2 (t) exp(-t)x2 (0), which when substituted in
the first equation yields
i1 = (-1 + exp(-t)z2(0)}z1
This is a linear time--varying system whose solution does not have a finite escape time. After some finite time
=
the coefficient of x 1 on the right-hand side will be less than a negative number. Hence, limt-+oox1(t) 0 .
Thus, the origin is globally asymptotically stable.
=
(2) Let V(x) (1/2)(zf + ~)-
V= -(xf + x~)(l - xf - x~) = -2V(1-2V)
In the region V(:e) < 1/2, Vis negative definite. Hence, the origin is asymptotically stable. For V > 1/2, V
is positive. Hence, trajectories starting in the region V(x) > 1/2 cannot approach the origin. In fact, they
grow unbounded. Thus, the origin is not globally asymptotically stable.
(3) Let V(x) == :eTPx =Puz?+ 2P12X1z2 + JJ22Z~, where Pis a positive definite symmetric matrix.
'
V = -2PJ.2Z12 + 2(pu 2
- P12 - P2,)x1x2 - 2(P'l2 - P12)x2 + Higher order terms
- 61
Near the origin, the quadratic term dominates the higher-order terms. Thus, V will be negative definite in
the neighborhood of the origin if the quadratic term is negative definite. Choosing Pi 2 = =
1, PoJ2 2, and
Pu= 3 makes V(z) positive de.finite and V(z) negative definite. Hence, the origin is asymptotically stable.
=
It is not globally asymptotically stable since the origin is not the unique equilibrium point. The set { xf 1}
is an equilibrium set.
=
(4) Let V(z) xf + (1/2)~.
Hence, the origin is globally asymptotically stable.
• 4.4 (a) Take V(w) = (1/2){J1wf + J2w~ + J3wi} as a Lyapunov function candidate.
V = J1w1w1 + J2w2~ + J3w3w,
= {J2 - J3)W1"'2Wa + ()3 - Ji)w1W:,Wa + (J1 - J2)W1W:,W3
= 0
The origin is stable. It is not asymptotically stable since V is identically zero.

(b) The closed-loop state equation is
J1w1 = (J2 - J3)w:iwa - k1Wt

J2W<J = (Js - Ji )wa"-'1 - /ciw:,
J3w3 = (J1 - J2)wt"'2 - ksWs
Using the same function V(w) as in part (a), we obtain

. .2 2 2
V -= -k1"'1 - kiw2 - kaws
Thus, the origin is _globally asymptotically stable.
• 4.5 Let g(z) = VV; 9i(z} = 8V/8z,.
lJgi 8 8V a2V
8:cj = 8z; 8z, = 8z;8z,
Similarly
8gj lJ 8V lJ2V
8:t, = /Jz, 8z; = 8zin;
Hence
Alternatively, supp08e
/Jg; - 8g;
Vi,j= l, ... ,n
/Jz; - 8:c,'
Define
- 62
8V
ln1
Similarly, it can be shown that
• 4.6 Try
To meet the symmetry requirement, take -y = /3.

V(x) = (az1 + .Bz2)X2 - (/Jz1 + oz2)[(z1 + x2) + h(x1 + x2)]
Take o= fi. •
V(z) = -/h:21 + (a - 2/3)z1z2 - /J(z1 + z2)h(.z1 + z:,)
Taking a = 2/J and p > 0 yields
. :,
= -/Jz1 -
.
V(z) /J(x1 + z2)h(z1 + z2)

which is negative definite for all z E R 2 • Now
g(z) = /3 [ i ~] ~ z Px ~ V(z) ~ Le gT(y) d11 = ~ZTPx
where Pis positive definite. Thus, V(x) is a radially unbounded Lyapunov function and the origin is globally
asymptotically stable.
= =
• 4.1 (•) Let VV(x) g(z). Then, V := -gT(z)Qt/>(x). Choose g(x) Px so that V(z) = (1/2)zTPz. We
need to choose P= pT > 0 such that V = -xTPQt/J(:z) is negative definite. Choosing P = Q-1 yields
V is negative definite in the neighborhood of the origin because yrp(y) > 0 for y :/: 0. Hence, the origin is
(b) The function V(:z) is radially unbounded. The origin will be globally asymptotically stable if V is
negative definite for all x. This will be the cue if y~(r,) > 0 for all r, ¥ 0.
(c) The function ~ sa.tisfies the condition Jlrl>i(V) > 0 for all :i, =F O. The function ¢>i satisfies the condition
= =
only near y 0 because t/J1 (11) vanishes at JI 1. Thus, we can only show asymptotic stability of the origin
using the Lyapunov function V(z) z 7 Q- 1z= = zl
+ 2z1z2 + 2~.
- 63
• 4.8
(a) V(O) =0 and V(x) > 0 for x :/: 0.
V. ( )
X =
2x1{l + xn- 2.xf.
(l + :tf) 2 Z1 + 2X2X2
.
2x1 r-6:t1 2x ] 4x2(x1 + .z2)

= u2 ~+ 2 - u2
12.xl 4xi
= - - 4- - - 2 < 0, V X E R 2 • :r :/: 0
u u
(b) The slope of the tangents to the hyperbola is given by
dx2 -2 -1
------
dx1 - (:t1 - v'2) 2 - !~ - v'2x1 + 1
On the other hand, ·
hi
/1 hypvbol•
= -2(.i-1 + x,)
-6x1 + 2x2{l + ~) 2
I•i=~
-1
= 2ri + 2T2x1 + 1
2~ + 2v'2x1 + 1 > ½x? - V2x1 +I

Hence
1:,2 I
l hyperbola
> slope of tangents
Moreover
JI _ -6x1 + 4 _ 4 + 6v'2:i:1 + 2xf + uf 0
1 byperbola - u2 z1 - v'2 - u2(z1 - V2) >
Hence, the vector field on the branch of the hyperbola in the first quadrant always points to the right of the
hyperbola.
(c) Since trajectories starting to the right of the hyperbola do not reach the origin, the origin is not globally
• 4.9 (a)
(b) On the line :r2 = z1, we have
• 4.10 (a)
xTPJ(z)+/T(x)Px = zTp fo !!<0-x)xdu+ fo xr[;!cux)]T duPx

1 1
= XT fo1{P:(ux)+ [!!<uz)]T p} dax $ -XTX
- 64
(b) V(x) = JT(x)Pf(x) is positive semidefinite. To show tha.t it is positive definite, we need to show that
V =0 ~ x = =
0. Since Pis positive definite, we need to show that f(:t) 0 if and only if x O; that is, =
the origin is the unique equilibrium point. Suppose, to the contrary, that there is p ':I- 0 such that /(p) = O.
Then,
pTp,s-[pTPf(p}+fT(p)PpJ=O ~ p=O
which is a contradiction. Hence, the origin is the unique equilibrium point. To show that V is radially
unbounded, we note that for any x E R"
xTPJ(x) _ 1 T T 1
llxll~ - 2Uxll~ [x Pf (x) +f (x)Px] 'S - 2, V x E Rn
Suppose now that 11/(x)lb ~ c as llxll2 -+ oo. Then
11:z:TPf(:i:)112 < llzTll:illPIJ2c < JIPll:ic-+ 0 88 llxll -+ 00

llxll~ - Uxll~ - llxll2 2
But this is & contradiction since

zTPJ(x) 1
llzll~ ~ - 2' V z E RTI
Thus, as ll:r:ll2 -+ oo, the magnitude of at least one component of /(z) must approach oo. This shows that
V(x)-+ oo as llzll2-+ oo.
(c) We have shown that V(z) is positive definite and radially unbounded.
V(z) = .f'(z) { P [ :! (z)] + [ :! P}(zf /(z) S -II/(:r)lll
Since J(x) = 0 <=> x = 0, V(z) < 0, for all z ER!', x :/:, 0. Thus, the origin is globally asymptotically stable.
• 4.11 Since Vi (z) is not negative semidefinite, there exists a point zo arbitrarily close to the origin such
that Vi(:r:0 ) > O. Let U = {x EB.,. I Vi(z) > O}. where B.,. CD. Since Vi(z) is positive definite, we have
Vi (x) > 0 for all z E U. Hence, the origin is UDStable.
• 4.12 Since Vi(:t) is not negative semidefinite, there exists a point z 0 arbitrarily close to the origin such
that V..(zo) > O. Let U = {x EB,. I V..(z) > 0}, where B,. CD. Since W(z} ~ 0 for all z ED, we have
Vi{z) = W(x) +AVi(x) > 0, V x EU

Hence, the origin is unstable.
• 4.13 (1) Apply Chetaev's theorem with V(z) = (1/2}(:tf - xj) . The function V is positive at points
arbitrarily close to the origin on the z1 -axis.
V(x) = X1(x~ +xrx2) -x2(-x2 +x~ +x1X2 - x~)

= (xf + z1z2) 2 + zi{l - z2 - z1 - zf)
For any O < c < 1, there is a domain around the origin where
1- z2 - :r:1 - xf > c> 0

Hence, in this domain, we have
- 65
1 Xz.
0.8
0.6
0.4
0.2
0
0 0.5 1
Figure 4.1: Exercise 4.13 (2).
The right-hand side of the preceding inequality is positive definite; hence all the conditions of Chetaev's
theorem are satisfied and the equilibrium point at the origin is unstable.
(2) The system
i1 = -zf + z2 = /i(z), ±2 =xt - x~ = h(x)
zU
has two equilibrium points at (0, 0) and (1, 1). The set r = {0 .$ :r:1 .$ l} n {z2 2: :r:f} n {z2 ~ is shown
=
in Figure 4.1. On the boundary z2 = zf, '2 0 and ft > 0; hence, all trajectories on this boundary move
=
into r. On the boundary :r:2 zf, /i == 0 and h > 0; hence, all trajectories on this boundary inove into
r. Thus, r is p06itively invariant. Insider, both / 1 and h are positive. Thus all trajectories move toward
the equilibrium point (1, 1). Since this happens for trajectories starting arbitrarily close to the origin, we
conclude that the origin is unstable.
• 4.14
Therefore,
V(x) ~ ½:if+ :z:1:z:2 + xJ = ½:r:T [ !~] X
The matrix of the quadratic form is positive definite. Hence, V(x) is positive definite for all :z:, and radially
unbounded.
V = :r:1g(z1)i:1 + X1X2 + :1;2±1 + 2:r:2±2

= g(.z1)(x1:r:2 - zf - z1z2 - 2x1:r:2 - 2x~) + x~
= -g(x1)(xf + 2z1z2 + 2x~) + zi
= -g(xi)xTQx + x~
where Q =[ ! !] is positive definite. Since g(:r:1) ~ 1 and zTQz ~ 0, we have

V .$ -(xf + 2x1x2 + 2:r:~) + x~ = -(x~ + 2:r:1:r:2 + x:) = -(:r:1 + :r:2) 2
This shows that V is negative semidefinite. We need to apply the invariance principle.
V =0 ~ 0 ~ -(z1 + z2}2 => 0 ~ (z1 + z2)2 ~ z1 + z2 = 0
- 66
x1(t) + z2(t) =0 => z1(t} + X2(t) =0 => X2(t) =0 => zi(t) =0

Since V (x) is radially unbounded and all the assumptions hold globally, we conclude that the origin is
globally asymptotically stable.
• 4.15 (a) The equilibrium points are the roots of the equations
0 =x2, 0 =-h1 (xi).- X2 - h2{X3), 0 = X2 - Z3
X2 = 0 => X3 =: 0 => h1 (zi) = 0 ::::} Xt = 0

Hence, there is a. unique equilibrium point at the origin.
(b) V(x) is the sum _of nonnegative terms; hence V(x) ~ O. To show that it is positive definite for all x,
we need to show that V(x) = 0 => x = 0. Since yhi(Y) > 0 for ally-:/: 0, the integrals 1 h (y) dy and
1 J:
J:' h2 (y) dy vanish only at Xi = 0 and x 3 = 0, respectively. Hence, V(x) is positive definite.
(c)
•
V =h1(x)x2 + x2[-h1(x1) - 2
x2 - ~(x3)] + h2(z3)(x2 - :i:3) = -x2 - zsh2(X3)
V(x) is negative semidefinite for all z, but not negative definite because Y(z) = 0 when :z:2 =zs = O for any
z 1 • We apply the invariance principle.
Hence, the origin is asymptotically stable.

(d) To show global asymptotic stability we need V(x) to be radially unbounded. This will be the case if the
integrals J:h;(y) dy, i == 1, 2, tend to infinity a.s lzl -+ 00.
• 4.16 ·1..et V(x) =(1/4).zf + (1/2)~.
V(x) is negative semidefinite for all .z.
x2(z) =0 => x~(t) =0 => x1 {t) =0

By the invariance principle, the origin is globally asymptotically stable.
• 4.11
(a)
At equilibrium,
x2 = 0 & g(z1) + h(zi)x2 = 0 => z2 = 0 & g(:z:i) = 0
Assume that g(z1) = 0 has an isolated root at the origin. Then, the origin is an isolated equilibrium point.
(b) With V(x) = ft g(y) dy + ½~, we have

• 2 2
V(x} = g(xi)x2 - x2g(z1) - h(z1)z2 = -h(z1)z2 :$ 0
Assume that h(z1) > 0 V z1 ED (a domain that contains the origin}. Then, V :$ 0 and
VEO => h(z1(t))z~{t):::0 => z2(t):::0 => g(x1(t))=0 => z1(t)sO
- 67
Hence, by LaSalle's theorem (Corollary 4.1), the origin is asymptotically stable.
(c) With V(x) = ½[:r:2 + ft h(y) dy]2 + J;1 g(y) dy, we have
v = [x2 + foz h.(11> t4t] [x2 + h(x1)±1] + ±1o(xd
1
= [x2 + lzi h(y) dy] [-g(x1) - h(x1)x2 + h(z1)x2] + x2g(x1)
= -g(xi} k:z:i h(y) dy

Assume that g(x1) J; 1 h(11) dy ~ 0 and J0z 1 h(t,) dy :# O.
.
Then , V :$ 0 and
V =0 => g(z1(t)) =0 ~ x1(t) =0 ~ x2(t) =0

Hence, by LaSalle's theorem (Corollary 4.1), the origin is asymptotically stable.
• 4.18 The system has an equilibrium point at 11 =Mg/Ir.and;= O. Let z1 = s,- Mg/Ir. and x 2 = fi.
Take V(x) = axf + bzj, wit~ a, b > 0. V(x) is positive definite and radially unbounded.
..
. (bk ) . 2bc1 2 2bc, 2
V{z) = 2 a - M Z1Z2 - - M X2 - M x2lz2I
Taking a= '/r./2 and b = M/2, we obt;ain
Moreover,
V = => 0 x2(t) = 0 => x1{t) =0
Using LaSalle's theorem (Corollary 4.2), we conclude that the origin is globally asymptotically stable.
M(q)ij + C(q, q)q + Dq + g(q) =u

where M(q) = MT(q} is positive definite, M - 2C is skew symmetric, D = DT is positive semidefinite,
g(q)= O has an isolated root at q = 0, g(q) = (8P(q)/8qJT, and P(q) is positive definite. The 2m-
dim~sional state vector can be taken as z = [ : ].
(a) Let u =0 and V =½4TM(q)q + P(q). V is a positive definite function of z.
V = qTMq+ ~qTMtj + gTtj = ~qT(M - 2C)q - ,I7Dq - q_Tg + gT q = -qTDq :$ 0
Hence, the origin is stable.
(b) With u = -Ktt4, we have V = -qT(Kd + D)tj $ 0. Moreover,
v =o ~ q =o => ii =o ~ g(q) =o => q =o
- 68
Hence, by LaSalle's theorem (Corollary 4.1}, the origin is asymptotically stable.
(c) With u =9(q) - K,,(q - q•) - K,1q, the equation of motion is given by
M(q)ij + C(q, q)q + Dq + K,,(q - q•) + K,1q =0
There is an equilibrium point at q = q• and q= O. Take
V = ½eTMe+ ½eTKpe is positive definite.

v = eTMii + ½eTMe + erKpe = ½eT(M - 2C)e - e.TKpe - eT(K" + D)e + eTKpe = -e7(K,1 + D)e :5 o
ti =o => e =o => e=o => Kpe =o => e =o
Hence, by LaSalle's theorem (Corollary 4.1), the equilibrium point (q*,0) is asymptotically stable.
• 4.20 According to ~alle's theorem, x(t) approaches M as t -+ oo. Equivalently, given E > 0 there is
T > 0 such that
inf llz(t) - 1111 < c, Vt> T
s,EM
Choose E so small that the neighborhood N(p,2£) of p EM contains no other points in M.

Claim: 11.z(t) -pl[< E, for all t > T, for some p EM.
The claim can be proved by contradiction. At t = t1 > T, let Pi EM be a point for wbicli 11.z(ti)-Pill < t.
Suppose there is time t2 > t1 such that llz(t2) - Pill= e. Let p =# Pl be any other poiDt of M. Then
Uz(t2) - PII = ll%(t2) - Pl + Pl - PII

2:: IIPl - PII - llz(t2) - Pill ~ 2£- e = e
=> 11¥,,, llz(t) - 11II 2:: E
The last statement contradicts the fact that inf,eM 11.z(t) -yll < E for all t > T, which proves the claim.
Since this claim is true for any, sufficiently small, e > 0, it is equivalent to z(t) -+pas t -+ oo, for some
pEM.
• 4.21
(a)
·
V(z) = 8V T
Bz z = -(VV) (VV) ~ O
V(x) = O<=> VV(x) = O <:> z =o

=
Hence, V(x) 0 if and only if xis an equilibrium point.
(b) Every solution starts in a set Oe with c 2:: V(.xo). Since V ::5 0 in Oc, the solution remains in !le for all
t > 0. Since Oc is compact, we conclude by Theorem 3.3 that the solution is defined for all t > O.
=
(c) By LaSalle's theorem, x(t) -+ M {Pt, ... ,Pr} as t-+ oo. Since the points Pl, . . . ,PY. ar~ isolated., we
conclude from Exercise 4.20 that z(t)-+ Pi as t-+ oo for some Pi EM.
• 4.22
Sufllciency: Suppose there is P = pT > 0 such that
PA+.ATP = -cTc
- 69
V(x) = -xTcTcx ~ 0
V(x) = D=> Cx(t) =0 => Cexp{At)x =0 => xo =D
0
since the pair (A, C) is observable. By LaSalle's Theorem (Corollary 4.2), the origin is globally ~ptotically
stable.
Necesaity: Suppose A is Hurwitz. Let
P is symmetric and positive semidefinite by construction. To show that it is positive definitet supp0&e it is
not so. Then, there is x ~ 0 such that zT Px = 0. Consequently
fu 00
xT exp(ATt)CTCexp(At)x dt = 0 => Cexp(At)x E O=> x = D
Hence, P is positive definite. Similar to the proof of Theorem 4.6, it can be shown that P satisfies the
equation
and that it is the unique solution.
• 4.23
(1) Let V(x) = zTPz.
Using the Riccati equation, we obtain
V(x) = -zT(Q+PBR-1BTP)z
Q > 0=> Q+ PBR- 1 BTP > 0
Hence, V(x) is negative definite and the origin is ~obally_ asymp~cally stable.
(2) When Q = crct we can only conclude that V(:z:) is negative semidefinite. But
V(x) =0 => :rT(t)(Q +PBR-1BTP)z(t) =0

::;,, Cx(t) =0 and R- BTPx(t) = 0
1
Due to the second identity, the state equation simplifies to
z = Ax-BR-1 BTPx = Ax
and it solution is given by x(t} = C exp(At)zo. Thus

Cexp(At)xo =0 => zo =0
since (A, C) is observable. By LaSalle's theorem (Corollary 4.2), we conclude that the origin is globally
• 4.24
- 70
Substitute for [8V/8z}J using the Hamilton-Jacobi-Bellman equation.
·
V(x) = -q(:r:) - av
(k - ¼) Bx G(x)R-1 (:r:)GT(:r:) (av)T
ax
H q(x) is positive definite and k ~ 1/4, we conclude that V(z) is negative definite; hence, the origin is
asymptotically stable. If q(z) is only negative semidefinite and k > 1/4, we can only conclude that V(z) is
negative semidefinite. But
.
V(x) =0 => q(r) and GT(x)
(av)T
lJ:r: = 0
·
=> z = /(z)
Since the only solution of x = J(x) can stay identically in the set {q(x) = O} is the zero solution, we see that
q(x(t)) =0 • x(t) =0
By LaSalle's theorem (Corollary 4.1), we conclude that the origin is asymptotically stable. The origin will
be globally uymptotically stable if all the assumptions bold globally and V(:r:) is radially unbounded.
• 4.25 Since (A, B) is controllable, the controllability Gramian W = J; e-At BBT [e-At] T dt is positive
definite. Hence, w-1 is positive definite. Note that
AW+ WAT = 1,,. {Ae-AtBBT [e-Atj7 +e-AtBBT [e-At( AT} dt
= - fa"'~ { e-AtBBT [e-Atf} dt = BBT - e-A-r BBT [e-A-rf
Hence,
(A - BK)W + W(A - BK)T =AW+ WAT - 2BBT = -e-A-rBBT [e-A-rf - BBT

With V(x} = :r:Tw-1:r:, we have
V(:r:) = xTw- 1((A- BK)W + W(A -BK)TJw-1:r: = -:r:Tw- 1 { e~ A-r BBT fe-Ar]7 + BBT} w-1z ~ 0
Bence, the origin of z = (A - BK} is stable; all eigenvalues of (A - BK) satisfy R.e[~] :S 0. Now we want
to show that Re[A} < 0. Let ). have Re[).]= 0 and let 11 be the left eigenvector of (A - BK} corresponding
to.\. Then
We have
(A - BK)W + W(A - BK)T = -e-A-rBBT [e-ATf - BBT
11·(A-BK)W11 + 11•w(A - BK)TII= -11•e-M BBT [e-A-rf II - v• BBTv
=> ).v*W11 + ~·v'"Wv = -v•e-A"' BBT [e-""']T v - v•BBTv

• 2(Re[~J)v'"Wv = -v•e-A-rBBT fe-A-rf v- v• BBT 11
Thus
Re[.\] =0 => v• BBTv = 0 => 11'" B =0
which contradicts the controllability of (A, B). Thus, all the eigenvalues of {A - BK) have negative real
parts.
- 71
• 4.26
(a) Suppose z = 0 is an isolated equilibrium point. Clearly z =
0 is an equilibrium point. We can show
that it is isolated by contradiction. Suppose it is not isolated. Then there is z ¥ 0, arbitrarily close to 0,
=
such that j(z) O. Define x =T- 1 (.z). Then, f(x) ={8'.l'/8x]-1 i(z) = O; that is, xis an equilibrium point.
By continuity of T-1 ( -) , we can make x arbitrarily close the origin, which contradict the fact that the origin
is an isolated equilibrium point. Clearly the argument works the other way around. Hence, z = O is an
isolated equilibrium point if and only if z = 0 is an isolated equilibrium point.
(b) Suppose z = 0 is a stable equilibrium point. Then, given e 1 > 0 there is o1 > 0 such that
Ux(O)II < 61 => llz(t)II < E1, V t 2= 0
By continuity of T(·): Given e2 > 0 there is r > 0 such that
Thus, there exists 6 > 0 such that
llz(O)II < 6 => llz(t)II < r => Uz(t)II < E2, V t 2'.: 0
By continuity of T-1 (· ), there is ~ > 0 such that

llzll < ~ => llxll < o
Hence
llz(O)II < ~ => llz(O)II < 6 ~ llx(t)U < r ~ llz(t)II < £2, Vt 2'.: 0
Thus, z = 0 is a stable equilibrium point.
Suppose now that z = 0 is an asymptotically stable equilibrium point. Then
x(t) -+ 0 as t -+ oo
Given E1 > 0 there is T1 > 0 such that llz(t)II < E1 for all t > Ti. By continuity of T(·): Given £ 2 > 0 there
is r > 0 such that
There exists T2 > 0 such that
Hence
z(t) -+ 0 as t --+ oo
and z ; 0 is asymptotically stable. The opposite direction of the proof is done similarly. Now
x = 0 is stable ~ z = 0 is stable
is equivalent to
z = 0 is unstable # z == 0 is unstable
0 =- X2X3 + 1, 0 = Z1Z3 - Xz, 0 = x~(l - zs)
From the third equation, :,;3 = 0 or z3 = l. The first equation cannot be satisfied with xs = 0.
- 72
Hence, there is a unique equilibrium point at (1, 1, 1).

(b)
-aJ
ox
I :r=(l,1,1)
= [ 0
:Z:3
0
-x3
-1
0 2%3 -
-x2
X1
3:rf
]
:r:m(l ,1,1)
=
[ 0 -1 -1
1
O
-1
O
1
-1
i
The eigenvalu~ are -1 and (-1 ± j "113) /2. Hence, the origin is asymptotically stable. The third state
equation bu equilibrium at X3 = 0. Starting with the initial condition x 3 (0} = 0, we have x 3 (t} :E 0. Then,
±1 = 1 and x 1 (t) grows unbounded. Thus, the equilibrium point is not globally asymptotically stable.
• 4.28
(a)
· O=z1,
Substitution of :z:1 = 0 in the second equation yields
-z2(l + xn = Q => X2 =0
Hence, the origin is the unique equilibrium point.
(b)
Hence, the origin is a.,ymptotically stable.

(c) Let V(x) = Z1X2,
V(z) = z1±2 + :i:1z2 = (z1z2 - l)z1z~ + (z1z2 -1 + zi)z1z2 - z1.z2
V(z)j = ui > o
, Z1:ll2=2
which implies that r is a positively i n ~ t set.

( d) The origin is not globally asymptotically stable since trajectories starting in r do not converge to the
origin.
• 4.29 (a) The equilibrium points az-e the roots of the eqlla!tions
. 3
0 = z1 - x 1 + z2, 0 =3x1 - x2
X2 = 33:1 =* X1 ( 4 - zn = 0
The equilibrium points are (0, 0), (2, 6), and (-2, -6}.
(b)
8/ = [ 1 - lz? 1 ]
8x 3 -1
IJJ I -[
{Jz :a:=(O,o) -
1 1
3 -1
J => ~2 - 4 = 0 => ~ = ±2
The equilibrium point (O, 0) is unstable (saddle).
~fl
uX z=(2,6)
= [ -:l !1 ] ~ ..\2 + 12..\ + 8 = 0 ~ A= -11.29, -0.71
The equilibrium point (2,6) is asymptotically stable (stable node).
a1
8x
I
z=(-2.-e) =
[ -11
3
1
-l
J
- 73
-:l _:1]
The equilibrium point (-2, -6) is asymptotically stable (stable node).
(c) Let A= [ and P be the solution of PA+ ATP= -I. Using Matlab, Pis found to be
P = [ g:~~ ~:!~! ]. The eigenvalues of Pare Ama:(P) = 0.7266 and ~in(P) = 0.0442. To estimate
the region of attraction of (2, 6), shift the equilibrium point to the origin via the change of variables
The state equation in the new coordinates is given by
.i1 = -lli1 + i2 - 6zf - i~

i:1 = 3i1 - i2
We use V = fl'1' Pfi as a Lyapunov function candidate. The derivative V is given by

V = -iTi- 2(p11f1 + JJ12i2)(6 + i1)i~
< -llzlli - 12(PJ.1i1 + P12i2)if - 2Pt2i~f2
< -llill~ + 12VYJ1 + .Pl2ll!fl~ + P121filf~
~ -(1- 2.4r - 0.177lr2)1JilJ~, for llill2 ::Sr
Taking r = 0.4, we see that V(i) is negative in {llill2 ::Sr}. Choosing c < Am1n(P)r2 = 0.00707, ensures
=
that {V(i) ::S c} C llilb ~ r because .\min(P)llzll; ::S V(f). Take c 0.007. Thus, the region of attraction
is estimated by {zT Pi~ 0.007}. The estimate of the region of attraction of (-2, -6) is done similarly and
the constant c is chosen to be 0.007. A less couervative estimate of the region of attraction can be obtained
graphically by plotting the contour of V(i) = 0 in the :ti-.i:2 plane and then choosing c a.nd plotting the
=
surface V(x) c, with increasing c, until we obtain the largest c for which the surface V(z) = c is inside the
region {V(i) < O}. The comtaDt c is determined to be 0.1. The two estimates of the region of attraction
are shown in Figure 4.2.
( d) The phase portrait is shown in Figure 4.3 together with the estimates of the region of attraction obtained
in part (c). The stable trajectories of the saddle form a aeparatrix that divides the plane into two halves,
with the right hatf as the region of attraction of (2,6) and the left half the region of attraction of (-2, -6).
Notice that the estimates of the regions of attraction are much smaller that the regions themselves.
There are three equilibrium points at(-½,-½), (0,0), and(½,½).

(b)
fJJ _ [ -(1r/4)sec2(rx1/2) 1 ]
&x - 1 -{,r/4) sec2 (1rx2/2)
8/ I _ [ -(,r/2) 1 ]
8x (-½,-i) - 1 -(,r/2) , Eigenvalues are - (1r/2) ± 1
lJ/1 = [ -(1r/ 4) 1 ] Eigenvalues are - (11'/4) ± 1

ax (0,0) 1 -(1r/4) '
{)fl
8x (½,½)
= [ -(1r/
1
2) 1 ]
-(1r/2) '
Eigenvalues are - (1r/2) ± 1
- 74
a.----------.
a
7 6
4
2
:J" 0
_______
I
s I -2
.. .__ _.
"""
-6
0 2 3 4
x, -8
-5 0 5
x,
Figure 4.2: Exercise 4.29. The dotted line is the
=
contour of V(z) 0, the dash-dot line is the contour Figure 4.3: Exercise 4.29. Phase portrait with esti-
=
of V(x) 0.007, and the solid line is the contour of mates of the regions of attraction.
V(z) = 0.1.
Thus, the equilibrium points ( -½, -½), and ( ½, ½) are asymptotically stable (stable nodes) while the equi-
librium point (0, 0) is unstable (saddle).
( c) Let us start with the equilibrium point ( ½, ½) . Define Ill = x1 - ½and 112 = z2 - ½. In the y-coordinates
the system is represe11ted by
ii= Ay + g(y), where A= [ -(~/2) -(~/2) ]
( ) _ r-(l/2)tan('ltY1/2+1'/4)+11'!11/2+ 1/2]
9 Y - -(1/2) tan(1t712/2 + 1r/4) + 1'Y2/2 + 1/2
The solution of the Lyapunov equation PA + ATP = -I is
p = (,r2 ~ 4) [ 2}1r 2{7r ]

Using V(y) =yTPy as a Lyapunov function candidate, we obtain
V(y) = -yTy + 2yTPg(11)
V(y) is negative in some neighborhood of y = 0. Using the "contour" command of matlab, we have plotted
the contour of V(y) = 0 and the contours of V(y} =
c for different values of c. We found that the choice
=
c 0.07 results in a set {V(11) ~ 0.07} which is a subset of the region where V(y) is negative as well as the
rectangle {-1 < z, < l}; see Figure 4.4. Similarly, the region of attraction of(-½,-½) is estimated by the
same Lyapunov surface V(11) = 0.07, except for the fact that now Y1 = %1 +½and 112 = z 2 + ½·
(d) The phase portrait and the estimates of the regions of attraction are shown in Figure 4.5. The ex-
act regions of attraction are the two halves of the plane separated by the separatrix formed of the stable
trajectories of the saddle equilibrium point.
• 4.31
(I) 8/1 = [ -1 +o X2 -1
a Z1 ] = [ -1
o -1 0 ]
,
).1 2
,
= - l, - l
I _[
X s=O z:,,:O
(2) 8I -1 + 3.1:f + z~ -1 + 2:r1 x 2 ] _ [ -1 -1 ] .>.1,2 = -1 ±j

-/Jr 3=0 - 1+2X1X2 2
-1 2
+ Z1 + 3z2 z=O - 1 -1 '
- 75
-- --
0.5
~ 0 /
~ ' ~ 0
-0.S
I
t, __ ... " /
I -0.5
I
-0.5 0 0.5
-0.5 0 0.5
)11
X1
Figure 4.4: Exercise 4.30. The dotted line is the

Figure 4.5: Exercise 4.30. Phase portrait with esti-
=
contour of V(y) 0 and the solid line is the contour
mates of the regions of attraction.
of V(y) = 0.07.
(3) a, I [
lJz._,,=
-2x1z2 1- z?
-1+3zl+2z1z2 -(1-~)
Jz=-0=
[ -1o 1
-1
l - .
In
'A1,2=(-l±1v3)/2
(4) a,
-8 1
% z=O
= [ -12 -1
-azi
2
l
•=O
= [ -1
2
-1
o J, .X1,2 =(-1 ±jv7)/2
r.;
• 4.32 We investigate stability of the origin using linearization.

(1)
a,
In =
r-1 +0 2x1 -1
-2x1
0
0
0
2xs
1
J, A = In
8/,
-0
= [-1
0
0
0 0o
-1
0
1
J
A has an eigenvalue at l; hence, the origin is unstable.
(2) Near the origin, sat(y) = fl which implies that
~ = [ z +2
°4x1z 10!1z
5
2 -8z1z ~ COSZ3 ] ,
-4
A = 881 I
z •=0
= [ ~ ~ !1 ]
2 5 -4
The eigenvalues of A are -1, -1, -2; hence, the origin is asymptotically stable.
(3)
-8/ =
Oz
1- 2 +lzf O O ]
2z1
O
-1 0
O -1
, A= -
{)z ....o
8/1 = [-2
O
O
0
-1 00 ]
O -1
The eigenvalues of A are -2, -1, -1; hence, the origin is asymptotically stable.
(4)
~ = [-1-: X2
X3
l
!1 -1 O~
+ Zt
X1 ] ' A= :1 z=-0
=[ =! ~1 -0°1 ]
0 1
The eigenvalues of A are -1, -½ ± j ½v'3; hence, the origin is asymptotically stable.
- 76
• 4.SS Since /(z) is twice continuously differentiable, we can represent it in some neighborhood of the origin
by f(x) =-Bx+ g(x), where 1lg(x)Jl2 $ kllzlll- The fact that the surface {V(x) =
c} will be closed for
sufficiently small c follows from continuity of V(x).
av
ax f(x) = av
ox [-Bx+ g(x)J = -xT(PB +BTP)x+2xTPg(x) = xT x + 2xTPg(x)
Using
we obtain
; :J(x) ~ llxll~ - 2kllPll2llxll~ ~ ½llxll~ for llxll2 $ 4kll~ll2
Choosing c small enough so that the closed surface {V(x) = c} is contained in the ball {llxll 2 ~ l/(4kllPll2)},
we see that [8V/8x]f(x) > 0 for all x E {V(x) c}. =
• 4,34 Proof of Lemma 4.2.
• Since a 1 (·) is continuous and strictly increasing on [O, a), for each given y E [O, o 1 (a)) there is a unique
=
x such that ct1(x) = y. Define x 0- 1(11). The function 0-1 (-) is continuous, vanishes at the origin,
and is strictly increasing. Hence, a-1 (-) is a class K. function.
• From the previous item, a 31 ( ·) is a class IC function. Moreover, since as (·) is a class /C00 function,
a;1 0 is defined on [O,oo) and a; 1 {r)-+ 00 88 r-+ 00, Hence a;1 (·) is a class K.co function.
• Let a(r) =a1(02(r)). Then

a(O) = O; o:(r) > 0, for r >0
r::.i > r1 ~ a2(r2) > 02(r1) ~ a1(a2(r2)) > a1(a2(ri))
Therefore a(·). is strictly increasing. It is also continuous. Hence, it is a class K, function.
• From the previous item, a(r) = a3(04(r)) is a class K. function. Moreover, it is defined for all r 2'. 0
and
r-+ oo ~ a 4 (r)-+ oo ~ cr(r)-+ oo
Hence, a 3 (o,.(·)) is a class K.00 function.
• For each fixed s, .8(02(r), s) is a class K. function of r. Thus a 1 (.8(a2(r), s)) is a class K. function of
r. Now, for each fixed r, ,8(02(r),s) decreases ass increases. Hence, a1(.8(a2(r),a)) decreases ass
increases. Moreover, a1(P(a2(r), a))-+ 0 as a-+ oo. Hence, a1(.8(a2(r),s)) is a class K,.C, function.
Thus, the inequality
is always satisfied.
• 4.36 This exercise is already dealt with in Example 4.18.
- 77
• 4.37 (1) Let V(:r:) = ½(xj + ~)-

V = -x? + a(t)x1z2 + o(t)x1x2 - 2x~ :S -x~ - 2x~ + 2jz1I lx2I
lx1I ]T[
= -[ lx2 I -1
1 -1] [lx1I] <-o.as2(x2+x2)
lx2 I -
2 1 2
where 0.382 is the minimum eigenvalue of the 2 x 2 matrix. By Theorem 4.10, the origin is exponentially
stable.
(2) Let V(x) = ½(zi + z~).
V' == -x21 + a(t)x1x2 - a(t)x1:z:2 - 2x22 = -x1

2 -2
- 2~
By Theorem 4.10, the origin is exponentially stable.

(3) Take V(x) = pux? + 2p12x1z2 + P22~, where Pti.P22 - ri2 > 0, and Jet k be an upper bound on la(t)j.
· ' 2 ~ 2
V = -2PJ,2.z1 + 211'11 - P22 - a(t)P12)x1x2 + 2!Pi.2 - a(t)PJ2)z2
Take Pn =P22 =p > I and P12 = 1.
V :S -2xf + 2klz1I lz21-2(2p- I)x: =- [ J:~/ ]T [ _2k -k
2(2p-1)
] [ lx1I ]
!x2I
Choose p > (I+ k2 /4)/2 so that 4(2p - 1) - k2 > 0. Then, the 2 x 2 matrix is positive definite and, by
Theorem 4.10, the origin is exponentially stable.
=
(4) Take V(x) pux? + 2P12Z1z2 + P22xi, where .P11P22 - ~ 2 > 0, and let k be an upper bound on lo(t)I,
. 2 2
V = -2lPi1 - a(t)p12]X1 + 2[-Pn + a(t)Pon - 2PI2].z1z2 - 4P22X2
Take P12 = 0 and P22 = 1.
V :S -2puzf + 2klz1I lz21-4z~ =- [ 1:~f ]T [ 2~ 1
Choose p11 > k 2 /8 so that 8pu -k2 > O. Then, the 2 x 2 matrix is positive definite and, by Theorem 4.10,
the origin is exponentially stable.
• 4.38
(a) Using the bounds on L(t), C(t) and R(t), we can show that
2k1 2L(t) 2k2
ka + kek. S R(t) + R(t)C(t) :S ke + k5ks
2 2 2
-
ke < -<
- R(t) -
- ~
Using the upper bounds, it is clear that V(t, z) is decrescent. H we try to use the lower bounds to show that
V(t,z) is positive definite, we will have to restrict the constants to satisfy
2~ 4k1
~ + klk, -1 > 0
Instead of making this restriction, we work directly with V(t, z), which can be written as
V(t , x ) -_XTp X -_ X
T [ R(t) + }- 12
1lm
l Z >
_ ZT [ R(t)
l 1
_L
Rm
] ct!.f X Tp-X
X -
- 78
It can be shown that the minimum eigenvalue of P is given by
-Xmin(.P} = ½(a - ../o.2 - 4}
where
2
a=R+ R
It can be easily seen that there are positive constants c 1 and ei such that a 2 - 4 2: c1 and Amin(P(t)) ~ c2 ,
for all t .2: 0, which shows that V(t,x) is positive definite.
(b)
· .z} ;::: - -2 [ 1 + R· ( -
V(t LC - -i]
L C) + -RC 2
-2 LR)
( 1+ - 2
I C R2 - -2 R %1 -
L R2 X2
Suppose l, 6, and R satisfy

1 + R (~ -
R2
C) + RC
2
LC - iR-> C1
LR ~c2
l+ R2
for all t .2: 0, for some positive constants c1 and c2 . Then
. 2c1 2 2C2 2
V(t,:r:} :5 - kax1 - ki:z:2
Hence, V(t, x) is negative definite. This shows that the origin is uniformly asymptotically stable. By linearity,
it is exponentially· stable. Alternatively, we can conclude that the origin is exponentially stable by noting
that V(t,x) satisfies the conditions of Theorem 4.10.
• 4.39
(a) Since g(t} 2: a> a,
1 + ag(t) - a 2 2: 1 + aa - a2 >1
Hence
2: ½(asinz1 + :c2)2 + 1 - cos:z:1
V(t,x)
which shows that V is positive definite in the region lx1 I < 2,r. Since g(t) :5 /3,
V(t,x) :5 ½(asinx1 + :t2)2 + (1 + a,8 + 0 2)11 - cosz1 I

which shows that V(t, x) is decrescent.
(b)
V = -[g(t) - acosxiJ~ - asin2 x1 - a 2(1- cosxi)x2 sinx1 + ag(t){l -coszi)

< -(et - a)x~ + a-y(l - cos.xi) - asin2 z1 + O(ll:cll 1)
= -(et - a)x~ - a(2 - 'i'){l - coszi) + a[2(1 - cosz1) - sin2 zi) + O(llzll 3)
The term [2(1 - cosz1) - sin2 xi} is O(l:c11 3). Hence
V ~ -(a- - a)~ - a{2- 'Y)(l - cosx1) + O(llxll3)
(c) The preceding inequality shows that Vis negative definite, since near the origin the negative definite
quadratic term dominates the cubic order term. Thus, the origin is uniformly asymptotically stable.
- 79
• 4.40
(a)
P(t + T) = exp(Bt + BT)+(O, t + T)
= exp(Bt) exp(BT)+(O, T)t(T, t + T) = exp(Bt)+(T, t + T)
since t(O, T) = (•(T, O}J-1 = fexp(BT)]- 1 .
A(t + T) = A(t) => t(T, t + T) = +(O, t) => P(t + T} = exp(Bt)t(O, t) = P(t)
(b)
p-t (t) exp((t - r)~JP(r) = +(t, 0) exp(-Bt)exp(Bt - Br) exp(BT)t(0, r)

= +(t, O)t(O, r) =
c)(t, r)
(c) P(t) and p-1 (t) are continuous functions oft. Hence, they are bounded on [O,TJ. Since they are
periodic, they are bounded for all t ~ O. Let IIP(t)II ~ k1 and IIP-1 (t)ll ~ "2. Then
llt(t, T)II ~ k1k2II exp((t - r)BII
and
II exp[(t - T)BII :5 k1k2ll+(t, 'T)II
Hence, U+(t, T)II is bounded by ke--r(t-T) if and only is Bis Hurwitz.
• 4.41
(a)
Z2=1 =z1
2z1z2 + 3t + 2 - 3z1 - 2(t + l)z2. =2t + 3t + 2 - 3t - 2{t + 1) = O= z2
Thus, z(t) = [ !] is a solution. _
(b) Recall from the discussion at the beginning of Section 4.5 that to show asymptotic stability of a solution
we shift it to the origin and then show asymptotic stability of the origin. Let z1 = ~1 - t and z2 = x 2 - 1.
Thm ·
i1 = z:i, i2 = 2z1z2 - z1 - 2z2
We need to show that the origin z = 0 is uniformly asymptotically stable.
:: = [-1~2z2 -2~2z1]' A= ~/..0= [ !1 ~2]
The matrix A is Hurwitz; hence, the origin is uniformly asymptotically stable.
• 4.42 Let V(x) = ½zTz.

V = zTx = -a[zTx + z~S(z)x + (zTz)2J= -a(l + zTz)zTz $ -a :z;Tx
where we used the property zTS(x)z = O. By Theorem 4.10, the origin is globally exponentially stable.
• 4,43 The closed-loop system is :t = f(z) - uG(z)GT(z)Pz. Use V(z) = zTpz as a Lyapunov function
candidate.
V = 2xTPt = 2zTPJ(x) - 2uzTPG(x)GT(z)Pz 5 --yzTPz - W(z) ~ --yV(x)

By Theorem 4.10, the origin is globally exponentially stable.
- 80
• 4.44 Let V(x) = ½(zj + zi).

v = -xf + x1z2 + z1 (~ + x~) sin t - x1x2 - x; + x2(x~ + x~) cost
= +
-(x~x~) + (xf + x~)(.x1 sin t + X2 cost)
< -Uxfl~ + llxflh/(sin t) 2 + {cos t) 2 = -llxll~ + llzll~

~ -(1 - r}llxll! , V llxll2 :S: r, for any r < 1
Hence, by Theorem 4.10, the origin is exponentially stable. Since V(x) = ½llzll~, the region of attraction
can be estimated by the set {llxll2 ~ r} for any r < 1.
• 4.45 (a) Use V(x} = ½(ri + Xi) as a Lyapunov function candidate.
V = x1[h(t)x2 - g(t)zf] + z2[-h(t)x1 - g(t)x~] = -g(t)(x1 + xi) :S: -k(x1 + x~)
Hence, the origin is uniformly asymptotically stable. Since all assumptions hold globally and V(x) is radially
unbounded, the origin is globally uniformly asymptotically stable, which answers part ( c).
(b) The conditions of Theorem 4.10 are not satisfied. Let us linearize the system
A(t) = 8/
8x (t, O) =
[ O
-h(t}
h(t) ]
O
Use V(x) = ½(zi + x~) as a Lyapunov function candidate for the linear system.
V =x1 h(t)x2 - x2h(t)x1 = 0
This shows that solutions starting on the surface V(x) = c remain on that surface for all t. Hence, the origjn
of the linear systeDl is not exponentially stable. This implies that the origin of the nonlinear system is not
exponentially stable.
(d) No.
• 4.46 Linea.rization at the origin yields the matrix
a1 I
{Jz ,r;=O
= [ -1 + a~ + ~
1+2z1X2
-1 + 2z1x2
-l+zf +~
l z•O
= [ -1
1
-1]
-1
whose eigenvalues are -1 ± j. Bence, the matrix is Hurwitz and the origin is exponentially stable. Co~
quently, it is asymptotically at.able.
• 4.41 Let V(z) = ½(bxi + ~)-

V = -bef»(t)(z1 - az1) 2 - actJ,(t)z~ :S: -~(z1 - az2) 2 - ac¢oxi ~ -W3(x)
It can be verified that Ws(z) is positive definite for all x. Hence, by Theorem 4.9, the origin is globally
uniformly asymptotically stable. Lineariza.tion at the origin yields the linear system
Yi = -4>(t)111 + a;(t)112, ih = bt,l)(t)111 - abef»(t)112

The invertible change of variables
[::]=[~ !][:]
results in the system
.i-1 = 0, .i2 = ,(t)z1 - ¢(t)(l + ab)z2
which has the solution z1 ( t) =
constant. This shows that the linear system is not exponentially stable.
Therefore, the origin of the nonlinear system is not exponentially stable.
- 81
• 4.48 Let A1 = ~(O) be the linearization of x = f(x). To find the lineam.ation of z= h(x)f(x), set
g(:z:) = h{z)/(z). Then
Hence
ogi (0) = h(O) fJh (O) + fJh (0)/i(O) =h(O) lJ/i (0)
OXj ax, 8x; {)zj
Hence
8g
A2 = 8:,; (O) = h(O)A1
Since h(O) > 0, A1 is Hurwitz if and only if A2 is Hurwitz. By '}:'heorem. 4.15,
x = J(x) is exp. stable ~ A1 is Hurwitz ¢> A2 is Hurwitz ¢> x = h(z)f(x) is exp. stable
• 4,49 Equilibrium points:
0 = -a.i:1 + b => f1 = b/a

O = -ex,+ f1 (a - Pfi1fi2)
Substituting ~1 = b/a in the second equation, we obtain
_ ab/a
z 2 = c-+.......,fi~(b...,/...,a)-=-
2
Thus the system has a unique equilibrium point.at (~1 ,f:a). Let
It can be verified that
Let
V = 2l_.2 "'f 2 1 4
1'1 + 2112 + 4Y1 , 'Y >0
V = -a11? - -y(c + fJ(Y1 + z1)2)y~ +-y(a - 2/Ji1f2)111Y2 - /J1f211f:i.r2 - ay:

:5 -ar,f - 'YCY'~ + 1c11111 I IJ/2 I+ "'fC2t1f 1112 I - ay:, c1 > 0, C2 > 0
= _ ~!fi-
1 2
')'CJ.'2 _
2
2
~t/4 - [
4 1
11111
li,,I
]T Q1 [ 11121
IJ11I ]- [ 11111 2
lu,I
]T Q2 [ IJ.1112]
IY21
where
_ [ a/2 -1c1/2 ] _ [ 3a/4 --rc2/2 ]
Qi - -'Yci/2 "fC/4 ' Q2 - -1c2/2 7c/4
Choosing 'Y < min{ac/2q, 3ac/4~} ~ures that Q 1 and Q 2 are positive definite. Then
. a 2 7c 2 a 4
V :5 - -Yi
2
- -312 - -111
2 4
~ . {
-mm a,c
} ( 2111
1 2+ 2Y2
'Y 2 1 4)
+ 4111
= -aoV, ao = min{a,c}
which shows that the origin is globally exponentially stable.
- 82
• 4.50
(a) Since the origin of the linearization is exponentially stable, by Theorem 4.13, the origin is exponentially
stable. Therefore, there exist positive constants k1 , 7, and c 1 such that
11.x(t)II ~ k1e-'Y(t-to)llx(to)II, V llx(to)II 5 c1
Suppose that c1 < llx(to)II < c. Then
llx(t)II 5 P{llx(to)II, t - to)
Since P(·,t) is a decreasing function oft, there is T > 0 such that tJ(c,T-t0 ) = c1 . Thus
,8(11.x(to)II, t - to), for to :5 t 5 T

llx(t)II s{
llx(T)llk1e-.,<t-T), fort> T
which implies that
Ux(t)II S ,8(11z(to)II, O}k1e-vCT-to)e--v<t-to), Vt ~ to, 'v c1 < llx(to)II < c
Notice that the foregoing inequality is also valid for llz(to)ll 5 c1 since P(llz(to)II, 0) 2!: 1. Define
a(r) = k1e-y(T-to).,B(r, 0)
Clearly, a(·) is a class IC function and
llz(t)II $ a(llz(to)ll}e--v<t-to), 'v t 2: to, 'ti llx(to}II <c

(b) Let M1 = DlaXe <r<c{a(r)}. Then,
1
For 11.x{to)II S c1, we have
For c1 < llx(to)ll < c, we have
llx(t)II 5 o(llx(to)ll)e--y(t-to) :5 ! 1 llx(to)lle--v(t-to)
Taking M = max{k1, Mi/c1 }, we obtain
llx(t)II S Mllx(to)lle~-y(t-&e>, Vt 2: to, V Ux(to)II < c

(c) The answer is no, because the constant M1 of part (b) depends on c and could approach oo as c -+- oo.
• 4.51 Using the inequalities
k1llzlla 5 V 5 k21l:rlla and V $ -kslJzW\ V 11:i:II ~ µ > 0

we see that as long as V ~ k2µa, we have llzll ~ µ and
V 5 - :: V => V(t, z(t)) :S e-Cts/h)(t-t0 >V(t0 , :r(to))
- 83
Hence,
llz(t)II $ (V(\:(t)) r/a (11 e-(ts,J•,)(t-ta>v(to,x(to))r/a

$
< ( ~ e-(A:3/A:2 )(t-tc,) k2llx(to)IJ•) l/B
= (!:) 1/a e-(ia/GA:a)(t-to)ll:z:(to)II = ke--r(t-to)llz(to)IJ
The foregoing inequality will hold over the interval [to,to + T] during which V 2: k2 µG. Fort 2: t 0 + T, we
have
• 4.52 Let E = max:w.(.i:)S,- V(x) . The assumption E < censures that {W4 (x) ::5 µ} is in the interior of
=
{V(x) ::5 c}. Hence, Vis negative on the boundary {V(z) c} and the set {V(x) ::5 c} is positively invariant.
Let A= {c $ V(z) ~ c}. As argued on page 170 of the text, we can show that there is a finite T >Osuch
that for to ::5 t ::5 to+ T, z(t) will be in A, while fort ~ to+ T, it will be in {V(x) ::5 t }. Hence, in [to, to+ TJ,
x(t) satisfies inequality (4.42}. Fort ~ t0 + T, %(t) e {V(z) 5 e}. Since ]Xliitive definite functions are _
bounded from below a.nd above by class K. functions (Lemma 4.3), any x in {V(x) ::5 e} can be bounded by
a class K. function ofµ.. In general, the function will not be a 11 (a 2(µ)) aa in (4.43).
• 4.53 The proof is exactly the same as that of Theorem 4.18. Notice that the argument used in the proof
of Theorem 4.18 holds outside a ball that contains the origin.
• 4.54 (1) The system is not input.to-state stable since with u(t) = c > 1 and x(O} > 0, z(t) -+ 00 as
t ~ 00.
(2} Let V(x) :ic ½x2.

V = -x4 + uz4 - x" $ -z4, 'r/ lxl ~ VU
By Theorem 4.19, the system is input-to-state stable.
(3) The system is not input.tcrstate stable since with u(t) =1 and x(O) > 0, x(t)-+ oo as t-+ oo.
(4) With u =0, the origin of z = x - x 3 is unstable. Hence, the system is not input-~state stable.
• 4.1515 (1) Take V = ½(xf + xj).

• _2 2 2
V = - (ri: + z2) + :i:2u $ -llxfb + Uxll2lul
= -(1 - 9)llxll: - 811:clli + llxlf2 lul :$ -(1 - 8)11.xlll, 'r/ ll:cll2 ~ lul/8
where O < 8 < 1. Hence, the system is input-~state stable.
=
(2) Take V (1/4)xf + (l/2)xf.
V = -x1 - x~ + x:iu = -x1 - (1 - 8)~ - 9xf + z2u ::5 -.xf - (1 - 9)xt for lx2I ~ lu]/8
where 0 < 8 < 1. When j.x2( $ luf/8, we have
V :s; -(1- 8}x1 - 8x1 - .x~ + u 2 /8 :$ -(1- 8)z1 - x~, for lx1I ~ /iui78
- 84
Thus, V ~ -(1- O}[xt +~]for all llzlloo ~ p(lul), where p(r) = max {r/8, Jr/8}. Hence, the system is
input-terstate stable.
(3) Take V = x~ + 2z-1z2 + 2x~ + z1.
V = -2x~ - 2x~ + 2z1u + 4x2u :$ -2xf - 2x~ + 2lx1I lul + 4jx2I lul
We have
Using Young's inequality, we have
Thus
V :$ -xt - x; + (21ul)'13 + 4lul2
Define the class K, function cp(r) = (2r)"/3 + 4r2 • Then,
V• ~ -x1' - x22 + q,(lul)
Since xt
+ ~ is positive definite and radially unbounded, by Lemma 4.3, there exists a class 1C00 function
03 such that
V ~ -03(11:i:II) + Tti(lul) = -(1- 9)<:ts(Uzll) - 8as(llzll) + ¢(1ul) :$ -(1- 8}03(11zll), V lxl ~ o:31 ( </>(~ul))
where O < 8 < 1. It follows from Theorem 4.19 t~t the system is input-to-state stable.
(4) With u = 0, the system
±1 = (z1 - x2)(.z~ - 1), ±2 = {.z1 + x2)(xi - 1}
z¥
has an equilibrium set { = 1}. Hence, the origin is not globally asymptotically stable. It follows that the
system is not input-to-state stable.
(Ii) The unforced system {with u = 0) has equilibrium points at (0,0), (1,1) and (-1,-1). Hence, the
origin is not globally asymptotically stable. Consequently, the system.is not input-~state stable.
(6) Let V(x) = ½(ri + zi),
v = -xi - X1X2 + X1U1 + X1Z2 - zt + X21'2
= -xi - X~ + X1 U1 + .Z21'2
< -(1 -O)xi - (1- 9)z1 - Ox? -8x~ + l.z1I IJullaa + lx2I Uulloo, 0 < 8 < 1
113
~ -(1 - 9):ri - (1 - 9)x;, for lx112: llu!l 00
and l:r2I 2: ( llu: 00
)
For lx,I < (llulloo/8)113 , we have
V < -x~ - x~ + lz1 I llulloo + (llu:;;t/s
= -(1 - 8)xf - :i:i - Bx~ + lx1 I llullac + (llu~1 '7t 13
Let p1 ( r) be the largest positive real root of the polynomial equation

r.C/3
-8y2 + ry + 9113 =0, r 2: 0
- 85
It can be seen that Pt(r) is a class ICoc function of rand Pt(r) ~ r/8. Hence, for lz2I < (llull00 /9) 1l 3 , we
have
For l:r:1 I < llulloo/9, we have
V < -x~ - x~ + llull~

e + lz2l llulloo
= -xf - (l -9)z~ - Ox~+ lz2l llull00 + IJu:~
Let PoJ(r) be the largest positive real root of the polynomial equation
• r2
-By + ry + 7i = 0, r~O
It can be seen that PoJ(r) is a class K.oo function of rand P'l(r) ::: (r/6) 113 • Hence, for lz1I < llull00 /8, we
have
V$ -x? - (1 - B)z~, for lz2I ~ PoJ(llulloo)
Define p(r) = max{Pl(r), PoJ(r)}. Then, p(r} is a class K.oo function of rand
V :S - (1 - 9)z1 - (1 - 9):i:~, V llzlloc ~ p(llulloo)
From Theorem 4.19, we conclude that the system is input-to-state stable.

(7) Let V(z) = ½zf
+ J;1 a(y) dJI + ½~-
V = [z1 + o-(z1)}(-z1 + z2) + z2[-z1 - o-(z1) - z2 + u]
~ -:z:f - x1a(z1) - z~ + z2u $ -llxll~ + llzl12lul ~ -(l - 8)llzll2, V llzll2 2: lul/8
where O < 8 < 1. From Theorem 4.19, we conclude that the system is input-to-state stable.
• 4.58 The system Z1 = -2:r

+ Z2, with Z2 as input, is input-to.state stable. The system :i2 has an = -~
asymptotically stable equilibrium point at the origin. It follows &om Lemma. 4. 7 that the origin of the full
system ia globally asymptotically stable.
• 4.57 tJ,(u) :S lw(u)I, where l'I/J(u)I is continuoUB and poaitive definite. It follows from Lemma 4.3 that
there is a cl&111 K. function ¢> such that j¢(u)I :S <l>(llull). Hence
V < -(1 - 8)as(llzll) - 8a3(jµ:jl) + (/>(llull), 0 < e< 1

$ -(1 - 8)as(tlzll), V llxll ~ a 3 1 ( q,(1~11))
Now we can apply Theorem 4.19 to conclude that the system is input-to-state stable.
• 4.58 Given e > 0, find e1 > 0 such that '"f(Ei) $ E/2. Since limHoo u(t) = 0, given Et there is T1 > 0 such
that llu(t)II $ E1 for all t ~ T1. Take to 2:: T1. Fort~ to, we have
llx(t)II :S .B(llx(to)II, t - to)+ -y(e1) :S .B(c, t - to)+ e/2

for some c > 0, where we used the fact that z(t) is bounded. Since /J(c, t-to) ~ 0 as t ~ oo, there is T2 >O
such that P(c, t -to)~ e/2 for all t ~ T2. Thus,
11:t(t)IJ $ e, Vt ~ T =max{T1, T2}

which shows that limHoo z(t) = O.
- 86
• 4.59 It follows from Exercise 4.58, but an indepenent solution is given next. Let V(x) = ¼z".
Starting from any time to, we have e-t :5 e-to for all t ~ to. Hence
where O < fJ < I. It follows from Theorem 4.18 that there is a finite time t1 ~ to such that
llx{t)II $ (T-to) , 1/3

"I t ~ t1
=
Given O < e < 1, take to ln(l/9e3 }. Then there exists a finite time T such that llz(t)II ::; f £or all t ~ T.
This shows that z(t) converges to zero as t tends to infinity.
• 4.60 Applying the non-global version of Theorem 4.18 shows that for any z(to) and any input u(t) such
that
the solution x(t) exists and satisfies
11:r:(t)II :5 P(llz(to)ll, t - to)+ 'Y (~~f. llu(r)II) , Vt;;:: to

Since the solution z(t) depends only on u(-r) for to :5 TS t, the supremum on the right-hand side can be
taken over [to,t], which yields (4.47).
• 4.61
(a) Asymptotic stability can be shown by linearization which yields the Hurwitz matrix [8//8z)(O) = -/.
Global asymptotic stability can be shown as follows. First we note that there is no finite escape time because
IIJ(z)II :5 kllzll; see Exercise 3.6. We have z2(t) = e-'z2(0). ·Therefore, there exists a finite time T > 0 such
that
Consequently
z1z1 S -(1 -P)z~, 'v t 2: T => z 1 (t)-+ Oas t-+ oo
(b) The linear system i2 = -x2 +u is input-to-state stable. Hence, for any bounded u(t), :r2 (t) is bounded.
z, (t) = z, (0) + f. z, (T) [ (,in .-z;(,) )' - I l dT
lx1(t)I ~ lz1(0)I + Lt lx1(-r)I d-r
Let T be as defined in part (a).
z1:i:1 $-(l-P)xL 'vt?:_T ~ lx1(t)l$lx1(T)I, 'vt?:.T ~ l:z:1(t)[$lx1(0)leT, 'v't?:_O
- 87
which shows that z 1 (t} is bounded.
(c) With u(t) =1 and x2(0) = 1, we have x2(t) =1. Then

.±1 = X1 [ (sin if -1] =0=> X1{t) =X1(0) = a
( d) Suppose the system is input--to-state stable, then there exist a class K.£ function f3 and a class K. function
"Y such that
lfz(t)II S P(llz(O)II, t) + "'f (supUu(t)II), Vt 2: 0
t20
Applying this inequa.lity to the solution of part (c), we obtain
11.:z:(t)II $ .B(llx(O}II, t} + -y(l), Vt~ 0
Since ,B(llx(O)II, t) tends to zero as t tends to infinity, there exists a finite time T1 such that
llx(t)ll S 2-y(l), 'r:/ t 2: Ti
= n+T
But 11.:z:(t)lli can be made arbitrarily large by choosing a large enough. This is a contradiction.
hence, the system is not input-~state stable.
• 4.62 The equilibrium point x =0 of x(k + 1) = J(z(k)) is

• stable, if for each E > 0, there is 6 > 0 such that
11:i:(O)II < 6 ~ llx(k)II < E, 'r:/ k 2: 0
• unstable, if not stable.
• asymptotically stable, if stable and 6 can be chosen such that
. llz(O)IJ < 6 => lim z(k) = 0

A:-+oo
• 4.63 The proof proceeds exactly as in the proofs of Theorems 4.1 and 4.2.
• 4.64 Since 4'iV $ -esllzll 2 S -(cs/ci)V, v. ~ V(x(k)) satisfies the inequality VH1 - v. $ -(c3 /c2 )Vi:-
By applying this inequality recursiwly, it can be shown that V.t ~ ( 1 - : ) • Vo. It can be seen that the
constant (C"J/c2) < 1, for if this was not the case Vt would go negative which is impossible. Therefore,
(1-:)<l.Now
llz(k)II S ( V(1k)) r $ [ ~ (I- : )' c,Hz(O)II' r = (: r( 1 - : )"' Hz(O)II
Seto= (~f/c and..,-; (1 - :)1/c to obtain 11:i:(k)II :$ ollz(O)lb.t, where o ~ 1 and O < -y < 1. The
origin will be globally exponentially stable if the inequalities satisfied by V and .dV hold globally.
• 4.65 The proof proceeds as in the proof of LaSalle's theorem (Theorem 4.4).
- 88
• 4.66
(3) ~ (1) Consider the Lyapunov function candidate V(x) = xT Px. Then
-6.V(x) = xT(ATPA-P)x = -xTQx < 0, 'v x ¥- 0
Hence, the origin is asymptotically stable.
(1) <=> (2) The solution of the state equation is
x(k) =A"z(O) = M J-,,, M-1x(O)
where
J = block dia9[J1, J2, ... , JNJ
is the Jordan form of A. If Ji = .>.i, then
Jt = >.f-+ 0, u k-+ 00 <==> l>-il < 1
If Ji= >.d + N, where N is a Nilpotent matrix, then
Jf =(>.ii+ N)1c =}: ( ~ ) >.fN"- -+ 0, ask-+ oo

l=O
1 {:::::::>
.
l.>.il < 1
Thus
x(k) = A1 x(O) = M J" M-1x(O)-+ 0, ask-+ oo ~ j.>.il < 1
for all eigenvalues of A.
(2) * (3) Let
From (2), IIAII ~ C-r", £or O < -y < 1 and C > O. Therefore
IIPIJ $ :E IIQIIIIAll :S 'E IIQIIC

k=O
2•
Al=O
2 121: $ !~ic:
1
"/
Hence, the infinite summation exists. On the other hand
V(x) = xT Px = xTQx + x l~ (A")T Q (A-,,,)] x ~ xTQx

Hence, P is positive definite. Now
00 00
ATPA - p = L (At+if Q (A*+l) - E (A·( Q (Ale)

4=0 i=O
00 00
= L (Arf Q(A'") - :E (At)T Q (A"} = Q

r=l k-0
• 4,67 Suppose the eigenvalues of A have magnitudes less than one, and let P be the solution of ATPA- P =
-Q, where Q =QT> 0. It follows from the previous exercise that P = pT > O. Take V(z) = xTPx as a
Lyapunov function candidate.
.1.V(z) = V(/(x)) - V(x) = /T(x)P/{x) - xTPx

= [Ax+ g(x)f P[Az + g(z)] - zTpz
= -xTQz + 2gT(z)PA:z: + gT(x)Pg(:z:)
- 89
Given 1 > 0, there is r > 0 such that llg(x)ll2 :S 'Yllxll2 for all llxll2 $ r. Hence,
~V(x) :5 -..\min(Q)llxll~ + 2-yjlPAll2llzU~ +",21f Pll2llzll~
Choose 1' small enough that
Then
1
~ V(x) .$ - 2>.min(Q)llxll~
which shows that the origin is asymptotically stable.
• 4.68 Let ,p(k, x) be the solution of x(k + 1) = J(x(k)) that starts at z at time k = 0. Let
N-l
V(x) =L </>T(k,x)<f>(k,x)
A:=O
Then
N-1
V(x) =xTx+ L <J>T(k,x)<J,(k,x) ~ xTz
1:::::1
On the other hand
N-1 N-1
~V(x) = V(J(x)) - V(x) = L q,T(k + l,x)(j)(k + l,x) - L <J>T(k,x)t/>(k,z)
l:=O p:O
N N-1
= L'PT(i,:z:)tf>(j,x) - L t/>T(k,x)t/>(k,x) = tt,T(N,x)4>(N,x)-xTx
j:;;;l 1:..0 .
:5 0 21'2 Nllxll~ - llxlll = -(1 - C 2-y2 N)llxll~

where we have used the fact that
1p(k,f(x)) = tf>(k,tf,(l,x)) = tf,(k + 1,x)

Choose N large enough to ensure that 1 - C 2,rN > 0. Thus, the second inequality is satisfied with
c3 = I - 02-y2N > O. Finally, from the continuous differentiability of f(x) we know that /(x) is Lipschitz
over the bounded domain D . Let L be a Lipschitz constant sucli that
11/(x) - /{11)!12 $ Lllx - 11112

Then
11</>(k + l,x) - rf>(k + l, 11)112 = 11/(t/>(k, x)) - f(<,b(k, y))ll2 $ Lll</>(k, x) - <J>(k, y)ll2
and by induction we obtain
- 90
Consider
N-1
IV(x) - V(11)I = L [tl>T(k,x)tf>(k,x} - ,t,T(k,y),p(k,y)]
N-1
= L {lf»T(k,x)[¢(k,z)- ¢(k,y)] + q,T(k,y)[q,(k,x) - t,6(k,y)J}
11:::aO
N-1
~ L [lltJ,(k,z)!l2ll¢,(k,z) - ,p(k,11)!12 + ll4>(k,y}ll2llt/>(k,z) -t,6(k,y)IJ2]
N-1
< L [ll9'>(k,x)ll2 + ll¢>(k,11)1l2] L llz -11ll2
.11-0
11
N-1
< L (C·r1"/lxll2 + C·r"IIY/12) Lillz - Ylh

11=0
:S
ThWI the last inequality is satisfied with Cc = ( E:;;,1 c..,• Lie).
- 91
-92
Chapter 5
• 5.1 Let (u1, yi) and (u2, 112) be the input-output pairs of the two systems. We have u = u 1, y = y2,
Then
1l1/2rll, $ 0 2 (IIJ/1rll_c) + /32

$ a2 (cr1 (llu1rllL). + f3t) + /32
< 02 (201 {llu1rll.c)) + 02(2/Ji) + f3i
where we have used Exercise 4.35. Set a:= 02 o 2a1 and /J = a 2(2/31) + /32, to obtain
To show finite-gain £ stability, start from
IIY,rll.c $ 'Yi lluiTll.c + P,, i = 1, 2

In this case
11:u:irll..c $ "'Y2 hiJlu1rll..c" + .81] + .82 $ 'Yl 'l'2 llu1,-IJ.c + 12.81 + /32
Set")'= 'Y1"Y2 and fJ = "Y2/Ji + /J2, to obtain
IIYrll..c $ ")' llurll..c + ,8
• 5.2 Let (u1,111) and (u2,112) be the input-output pairs of the two systems. We have u 1 = u2 = u,
11 = Yi + J/2, and
Then
IIYrll.c :$ IIY1-rll.c + IIY2rll.c :S 01 (llu,.llc) + /31 + 02 (llu,.11,c) + /32
Seto= 01 + 02 and /J = /31 + /32, to obtain
To show ~ga.in l stability, start from
Jlyi,.Jl..c $ -Yi IIUirll.c + /3,., i = 1, 2

In this case
IIYr 11.c $ 'Yl l!urll.c + fJ1 + 'Y2 Uu,.11, + fJ:i
Set -y =11 + "12 and /3 = /31 + /:h, to obtain
- 93
• 5.3
(a) Let a(r) = r 1l 3 ; a is a cws K:.00 function. We have
IYI :5 lul 113 * 1111.-ll.c ... $ (llu.-ll.c.,) 113 * IIY.-llt.., $ o (llu.-llt,.J

Hence, the system is £or:i stable with zero biM.
(b) The two curves IYI :::: lul 113 and ]YI = alul intersect at the point lul = (l/0) 312. Therefore, for
lul :5 {1/a)3l 2 we have IYI :5 (l/a) 1l 2 , while for lul > (1/a) 312 we have IYI $ aJul. Thus,
IYI $ ajul + (l/a) 112 , Y lul ~ 0
Setting 'Y = a and /3 = (l/a)1l 2 , we obtain
IIY.-11.c.., $ 'Y llu.-Ut_ + fJ
( c) To show finite-gain stability, we must use nonzero bias. This example shows that a nonzero bias term
may be used to achieve finite-gain stability in situations where it is not possible to have finite-gain stability
with zero bias.
• 5.4
(1} h(O) = 0 ~ jh(u)I $ Llul, Vu. For p = oo, we have
sup ly(t)I $ Lsup lu(t)j
t:::;O t:::;D
which shows that the system is finite-gain C.00 stable with zero bias. For p E [l, oo), we have
lor lr,(t)I" dt $ V
lor lu(t)IS' dt
. * llr-,-11.cl' $ L llu,-ll4
Hence, for each p E [1, oo), the system is finite-gain .C.., stable with zero bias.
(2) Let lh(O)I = k > O. Then, lh(u)I $ Liu!+ k. For p = oo, we have
suply(t)I $ Lsuplu(t)I + k.
t:::;O t:::;O
which shows that the system is finite-gain COG stable. For p e [l, co), the integral (Llu(t)I +k)" dt diverges J;
as.,.~ oo. The system is not £p stable for p E {l,oo), as it can be seen by taking u(t) 0. =
• 5.5 The relay characteristics of pa.rts (a), (b), and (d) satisfy lr(t)I :S kju(t)I for some /c > 0. Therefore,
sup,~0 Jy(t)l :S hup,~0 ju(t)I and ft
r 2 (t) dt 5 lrJ fa°° u2 (t) dt. Thus, in these three cases the s ~ is
both finite-gain C.,:,o and finite-gain £2 stable. In case (c), the system is clearly Cao stable since the output
is always bOUDded. However, it is not C.2 stable. For example, the £ 2 input u(t) = e-t produces the output
=
y(t) 1 which is not £2.
• 5.6 Let V(t,z(t)} = 0.
D+W = Jim sup -h1 [W(t+h,x(t+h))-W(t,z(t))]
,._o+
= lim sup -h1 y'V(t + h,z(t + h))

h-o+
We have
V(t + h,x(t + h)) :S ~ 11:r:(t + h)ll2
- 94
x(t + h) = h[f(t, O) + g(t, O)] + o(h) => llx(t + h)ll2 = h2 llg(t, 0)11 2 + ho(h)
; 2 V(t + h, z(t + h)) ~ ~ llg(t, 0)11 2 + 1;') : ; ~ 6 (t) + o~)
2
fun h ~ ¾Jv(t + h,x(t + h)) ~ fl6(t) $ {t[;fl6(t)

since / C4 /2c1 ~ 1. Thus
D+ W .:$ 2~6(t)
which agrees with the right hand side of (5.12) at W = O.

• 5. 7 Following the proof of Theorem 5.1, it can be shown that
.
llx(t)II S 'Y1 supllu(t)II + .81. where "Yl
c~o
= --,
C2C4L
C1C3
/3i = ~2
-llxoll
C1
Consequently,
ll11(t)II s ('71 "YI + '12) sup llu(t)II + .81 + '13
t~O
which shows that (5.11) is satisfied with p = oo and
'hC2C4L
"'I = 112 + - -, /3 = 1h IIZoll~-C1
2
+ '1l
C1C3
• 5.8 Following the proof of Theorem 5.1, it can be shown that
~ 'Y1 sup llu(t)II + /31, where 'Yt = c,c.,L, /31 = i§'llzoll

flz(t)fl
~ ~~ v~
Using (5.20), we obtain
l]y(t)II $ 01 (-r1 sup Uu(t)II + !3i) + 02 (sup ll1'(t)11) + f/

t~o c~o
$ 01 (211 sup
c~o
llu(t)II) + 01 (2.Bi) + a, (sup llu(t)II) + '1
,~o
which shows that (5.22} is satisfied with
'Yo{r) = a1(2-y1r) + et2(r), Po= 01(2.81) + 'I
• 5.9 Consider the system
:i: = A(t)z + B(t)u
11 = C(t):r + D(t)u
Assume that all matrices are uniformly bounded; that is,
IIA(t)II ~ c1, IIB(t)II $ c2, IIC(t)ll $ C3, IID(t)II $ C., 'v t ~ to
and the origin of :t = A(t)z is exponentially stable so that the transition matrix +(t, to) satisfies
ll~(t, to)II s; ke-a(C-Co), 'ff t ~ to
It can be easily shown that
lfy(t)II $ c3ke-a(,-ta)llxoll + £t.

1
e:,ic3ke-aU-T)IJu{r)II dr + C4llu(t)ll
From this point on, proceed as in Corollary 5.2.
- 95
• 5.10 (1) Let V(x) = ½r,

V = -(1 + u)z' $ -(1 - r.)x4, V jul $ r 16 <1
By Theorem 5.2, we conclude that the system is small-signal lOQ stable for sufficiently small jx(O)I, Taking
u(t) = -2, it can be seen that the system is not l 00 stable. The origin of the unforced system in not
exponentially stable. However, the origin of the forced system is asymptotically stable for lul < 1, which
implies that l11(t)I $ P(lzol,t). Therefore, the system is small-signal finit~gain C 00 stable.
(2) We saw in Exercise 4.54 that the system is inpu~to--state stable. Using Theorem 5.3, we conclude that
the system is £ 00 stable. The origin of the unforced system in not exponentially stable. However, the
origin of the forced system is asymptotically stable for lul < 1, which implies that jy(t)I :5 ,B(lxol, t) + lu(t) I
Therefore, the system is small-sign~ finite-gain loo stable.
(3) Since 1!11 $ ½, the system is finite-gain Coo stable.
½x
(4) With V = 2 , we have
V = -x2 - x'+ :e3 u $ -x2 , V lxl ~ lul
By Theorem 4.19 we conclude that the system is input-to-state stable. Using IYI = lxsin(u)I $ lxl, we
conclude by Theorem 5.3 that the system is Coe stable.
• 5.11 (1) We saw in Exercise 4.55(1) that the system is input-u;state stable. By Theorem 5.3, it is C00
stable. Take V ::: ½(xf + ~) and set u =
O. Then, V = -2V, which shows that the origin is globally
exponentially stable. All the assumptions of Theorem 5.1 are satisfied globally. Therefore, the system is
finite-gain £ 00 stable.
(2) We saw in Exercise 4.55(2) that the system is input-Urstate stable. By Theorem 5.3, it is C00 stable.
By linearization, it can be seen that the origin of the unforced system is exponentially stable and all the
assumptions of Theorem 5.1 are satisfied locally. -~ce, the system is small-signal finite-gain £ 00 stable for
sufficiently small Uz(O)II,
= =
(3) Let V ½(zf + ~). When u 0, we have
V = 2V(2V - 1) > o when V > 1/2
Thus, solutions starting in ll:.i:ll2 > 1 grow unbounded. This shows that the systelll is not C00 stable.
assumptions of Theorem 5.1 are satisfied locally. Hence, the system is small-signal finite-gain l 00 stable for
sufficiently~ llz(O)II.
(4) We saw in Exercise 4.55(6) that the system is input-to-state stable. By Theorem 5.3, it is £go stable.
By linearization, it can be seen that the origiD of the unforced system is exponentially stable and all the
assumptions of Theorem 5.1 are satisfied locally. Hence, the system is small-signal finite-gain £go stable for
sufficiently small llz(O)II.
(5) With u = 0, the system has three equilibrium points at (0,0), (1,1) and (-1,-1). By linearization,
it can be seen that the equilibrium points at (1, 1) and (-1, -1) are saddles. By sketcliing the phase
portrait, we can see that there are trajectories that diverge to infinity. Hence, the system is not l 00 stable.
aaawnptions of Theorem 5.1 are satisfied locally. Hence, the system is small-signal finite-gain l 00 stable for
sufficiently small llx(O)ll-
(6) We saw in Exercise 4.55(3) that the' system is input-to-state stable. By Theorem 5.3, it is £ 00 stable.
By linearization, it can be seen that the origin of the unforced system is not exponentially stable. Therefore,
we cannot apply Theorem 5.1.
(7) We view the system as a cascade connection of the linear system
and the delay element y(t) = 1,11 (t - T}. The linear system has a Hurwitz matrix. Hence, by Corollary 5.2,
it is finite-gain £ 00 stable. The time-delay element is finite-gain C00 stable. Hence, the cascade connection
is finite.gain £ 00 stable.
- 96
• 5.12 We start by investigating stability of the unforced system
:i:1 = Z2, :i:2 = -(x1 + X2) - h(z1 + X2)

=
We use the variable gradient method to.find a Ly&punov function. V gT(x)/(x) = g1 (x)fi(x)+92(x)'2(x).
Since h(x) =-(xi+ z2) - h(x1 + x2), let us take !/2 = x1 + z2. From the symmetry condition 8gif8x2 =
=
{)92/8x 1 1, we take 91 = bx1 + x2. Then V(x) = ½bx~ + z1z2 + ½z~. The quadratic function V(x) is
positive definite for b > 1.
V
'
= (bz1 + x2).x2 - (x1 + X2) 2 - (xi + z2)h(z1 + x,) = -x12 + (b - 2).x1x2 - (z1 + .:z:2)h(x1 + x2)
Taking b = 2 yields
V = -Xi - (z1 + z2)h(z1 + z2) ~ -xf - a(xi + z2) 2 = -xTQx
where Q = [ 1-a
+a -a ] . The matrix Q is positive definite; hence, the equilibrium point at the origin is
a
globally exponentia.lly stable. Now consider the forced system. The Lyapunov function V satisfies inequalities
(5.6)-(5.8) globally. It is easy to see tha.t (5.9) and (5.10) are satisfied globally. It follows from Theorem 5.1
that the system is £, stable for each p E [l, ooJ.
• 5.13 Since W(x) is positive definite and radially unbounded, it follows from Lemma 4.3 that there is a
class K.010 function a such that W(x) ~ a(llxll) for all z E R". Since l?Ji(u)I is positive definite, it follows
from Lemma 4.3 that there is a class JC function Po such that l,J,(u)I 5 PoCllull) for all u E Jl!R. Hence,
w
fJx f(z, u) $
l
-o(llxll) + po(llull) S - 2o(llxll), 'r:/ llxH ~ a
- 1 (2po(llull))
We conclude &om Theorem 4.19 that the system is input-to-state stable. Furthermore, llh(x, u)II is a positive
definite fun~on of [ : ] . It follows from Lemma 4.3 that there is a class K.oo function />1 such that
1/h(z,u)II $ Pi (II : ID $ Pt{2llxll) + P1(2llull), 'r:/ (z,u)
Thus, (5.23) is satisfied globally with r, = O. We conclude from Theorem 5.3 that the system is £ 00 stable.
• 5.14 FroDl Example 5.2, we know that
IIJ,,.11.c, :$ llhll.c1 llu,.11.c,

This inequality implies that the £2 gain is less than or equal to Hhll,1 = fa°'° lh(t)I clt. From Theorem 5.4,
we know that the £2 gain is SUP.,,eRIH(jw)I, Hence,
SUPweRIH(jw)I $ fo 00
lh(t)I dt
• 5.15 (1)
J(z) = [ -asmx1
. x2 - L
~x2 ] , G = [ 01 ] , h(x) = z2
Let W(z) = a(l - cos:i:1 ) + ½z~. W(x) 2'.: 0 for all x E R 2 •
a;: f(x) = [ asinx1
- 97
-aw
/jz
0 =[ .
OSU1Z1
Thus, W(:r) satisfies (5.32)-(5.33) globally. It follows from Example 5.9 that the system is finite.gain C,2
stable and its £2 gain is less than or equal to 1/k.
(2)
f(x) =[ -x2 ]
z1 - x2 sa.t(x~ - x:) ,
X3 Sat(:ri - xn G =[ !
-x3
l· h(:r) = xJ - x~
Let W(x) = ½zTx.
8W
-/(:,;)
8x
= [ X1 x2 X3 ] [
-X2
:i:1 - X2 sat(:r~ - xD
]
= -(x~ - xi) sat(x~ - xi) = -h sat(h)
:rs sat(z~ - zf)
a;: G = [ z1 z2 :z:s ] [ ! ]=
-:rs
x~ - z~ = h(x)
Let D = {.z e R 2 I lh(x)I ~ 1}. W(x) satisfies (5.32)-(5.33) in D with k = 1. Taking V(x) = W(x) and
-y = 1, it can be verified that (5.28) _is satisfied in D. Consider now the unforced system.
h(:r(t)) =0 => is(t) =0 => xa(t) =constant => i2(t) =0

=> z1(t) =0 * x1(t) =0 * z2{t) =0 * zs{t) =0
Using of Lemma 5.2, we oonclude that, for sufficiently small IIZo!I, the system is small-signal finite-gain £2
stable and its £ 2 gain is less than or equal to 1.
=
(3) Let D {:c E R 2 112:ri + z2I < 1}. For x ED, the system is given by
:t = Ax + Bu, y = Cx
where
A= [ ~l !1 ] , B= [ ~ .] , C~ ( 1 O]
Since A is Hurwitz, there exist positi~ constants k1 and k2 such that for all llz(O)II < k1 and SUPe>o ju(t}I :S
k2, x(t) E D for all t ~ O. The system is linear time--invariant and its £ 2 gain can be determmed using
Theorem 5.4. The transfer function is
H(s) = a
.2
l
+ 8 + 1 = 8 2 + 2/"-,WnB
n
+ Wn2'
~
Wn =1, '= 0.5
suplH(jw)I=
weR
R= .~
2( 1 - ( 2 3
vi>
Thus, for sufficiently small llxoll, the system is small-signal finite-gain £2 stable and its C,2 gain is 2/../3.
(4)
Let W(z) = ¼xf + ½xi.
~! f(x) = [ zf X2 ] [ -(l + ~)x2 _ xf ] = -(1 + xf )xi ~ -z~z~ = -h2 (:s:)
- 98
~: G = [ xf X2 ] [ ! ]= x1x2 = h(x)
Thus, W(.:i::) satisfies {5.32)-(5.33) globally with k = l. It follows from Example 5.9 that the system is
finite-gain £ 2 stable and its £2 gain is less than or equal to 1.
• 5.16 (a) Let V(x) = ft u(y) dy + ½(xf + x~).

V = -x~ - x1u(xi) - x~ + x2u S -xi -
x~ + x2u
~ -(1 - 9)11.:i::II~ - 9llxJI~ + Jlxll2lul $ -(1 - 6)11.:i::llt· V lf.:i::ll2 ~ lul/9
where O < tJ < l. It follows from Theorem 4.19 that the system is input-to-state stable. Since IYI = lx2I S
llxll2 , we conclude from Theorem 5.3 that the the system is C00 stable.
=
(b) Let V(x) a[Jt u(y) dy + ½(x~ + z~)].
11. = 8V f + _1_ 8V GGT (8V) T + !hTh

ax 2-y2 ax 8x 2
= 2
a[-x1 - z10-(x1) - x 2
2]
+ 2a2 2 21
")'2 X2 + 2x2 S
( o:2 .1)
-a+ 21'2 + 2 X2
2
Choosing Cl = -y = 1 yields 1(. $ 0. Hence, the system is finite-gain £ 2 stable with £2 gain less than or equal
to one.
• 5.17
8V I+ 8V Gu
8x 8x
1{. $ 0 implies
8V I+ av Gu< !-y2uTu- \,Ty
8x ax -2 2
From this point on, proceed as in the proof of Theorem 5.5 (starting from (5.29)).
• 5.18 The closed·loop system is given by
x =f - GGT (av)T
Bx + Kw
The closed-loop map from w to [ : ] is given by
:i: = fe(x) + Gc(x)w

Ye = hc(x)
- 99
where
11.c =
=
=
<
From Theorem 5.5, we conclude that the closed-loop map from w to ( !] is finite-gain £ 2 stable with C 2
gain less than or equal to -y. .
• 5.19 (a) The existence of~ follows from the fact that
lim l - t./2 - ./"J = 1 - ~ > 1 - E

cf~O v'l=-1 2
Thus, by choosing 6 small enough, we can make (1- E/2- -Jl)/../f='S ~ 1- t.

(b) By definition of the £2R gain, we can find u E l2R such that llull.c:111 =
1 and ll111l~1t ~ "Y2R(l - t./2).
=
If there was no such u, we would have Urtll.cut :5 'Y2R(l - t./2) for all llulli::ut 1, which contradicts the
claim that 1'2R is the £2R gain. By choosing o of part (a) less than one, there exits time t 1 such that
t1«, uT{t)u(t) dt = 6.
(c) We have
1-: uf(t)½(t) dt = [ : uT(t)u(t) dt = 6 => llu2ll.c

2 1t = ,/6
/
J_oo
00
ui(t)u1(t) dt = £ 00
ti
uT(t)u(t} dt = 1-6 => llu1ll.cu = vf=1
Consequently,
and
Therefore,
ll1J1ll.cu, > 1 - E/2 - yd > (1 )
llu1 lieu - ~ 'Y2R - - E 'Y2R
(d) Let u(t) = ui(t + t1) and 71(t) = 111(t + t1)
{oo uT(t)u(t) dt = [oo uf(t + t1)U1 (t + t1) dt = (':xi U1(Tf U1('r) d,,-
lo lo lt1
Which implies that Uull£2 = llu1ll.cu- Simiwly, IIPll.c2 =
ll111ll.c 2.11· The fa.ct that y(t) is the output
corresponding to u(t) follows from linearity. Finally, IIYll.c2 ~ (1 - E)"f2Rllull£2 follov,s from part (c).
- 100
l[
• 5.20 The closed-loop transfer functions are given by
][:: l
-(,+1)
y,1 •+2 ·
,-1 •+2
-1 ] [ Ul E1 ] [ •+2
.it (•-1)(•+2)
[ ] [
Y2 = .!2 (•-~TI~+2) U2 ' E, = ::;~ .!±!
i+2
The closed-loop transfer function from (u1, u2) to (Y1, 112) ( or (e1, e2)) has four components. Due to pole-zero
cancellation of the unstable poles= 1, three of these components do not contain the unstable pole; thus,
each component by itself is input-output stable. If we restrict our attention to any one of these components,
we miss the unBtable hidden mode. By studying all 'four components we will be sure that unstable hidden
modes must appear in at least one component.
• 5.21
11111.,.ll.c :5 1'1lle1.,.ll.c + fJ1

< 1'1 ( 1 1 (llu1rll.c + 'Y211u~II£ + /32 + 12/31)] + /31
-1'1l2
= "YI llu1.,.ll.c+ 'Y11'2 llu2.,.1l,+ "Y1(P2+1'2P1) +/31

1-~1'2 1-~1'2 1-~1'2
11 'Yl "Y2 -Y1P2+ fli
= 1 - "YI 1'2 llul'Tll.c + 1 - "Yl12 llu2.-ll.c + 1 - 1'1 1'2
The expression for lb12rll.c can be derived in the same way.

• 5.22 (a) Let the underlying space be C 00 • We have "2, d2 E C 00 • From the analysis of Example 5.14, we
have e1, e2,x E Coo, provided E < 1/'Yl'Y/· From the equation
ai = A11 + EA-1 Be2

we see that 'I E C00 • Thus, for sufficiently small E, the state of the closed-loop system is uniformly bounded.
(b) From (5.44}, we see that one of the terms on the right-h,!!,D.d side is E"YJlldill.c_, For d,(t) = a sin~, this
term is given by
When the product EW is small, the term is negligible. However, as we increase w, the product EW will no
longer be small. At frequencies of the order 0(1/E), the product w will be of order 0(1). While the system
remains C00 stable, the bound on x, given by (5.44), will not be close to the bound 1'lldll .. + /J + "Y/32, as
concluded in the example. The conclusion of the example is valid only for frequencies of the order 0(1) (or
frequencies for which fl,;) is small).
• 5.23 (a) Using

1 3
e1 = tL1 - 1/2 = tL - 2X3
e2 = u2+111 = x2
we obtain
i1 = -xi +z2
x2 = 3 1 3
-X1 - X2 - 2X3 + 'U
zs = x2-x~
y = X2
- 101
(b) We search for functions Vi and V2 that satisfy the Hamilton-Jacobi inequality for the systems H 1 and
H 2 , respectively. For H1, let Vi= ct1(¼x1 + ½~), 01 > 0.
1f. = -a1(X14 +.X22) + 2a~1':Zi2 + 21x22 !5 ( -01 + 2-yf

a~
+ 21) :l:22
Choosing 01 to minimize 1'1, we end up with 01 =')'1 = l. For H2, let ½ = (02/4).x~, o, > 0.
Choosing a 2 to miniroize "'f2, we end up with a2 "Y2 = =

¼- Since 'Yl"Y2 = ¼ < 1, we conclude by the
small-gain theorem that the closed-loop system is finite-gain £2 stable. To find an upper bound on the £ 2
gain, we search for a function V that satisfies the Hamilton-Jacobi identity. We consider
V(x) =Vi+½= a1 (¼z1 + ½zi) + ~2z:

and we allow ourselves the freedom to change the choice of the positive constants cr1 and a 2 •
8V a1 a
8:r I=
4
-01%1 -
2
0 1::i:2 - 2Z2Z3 + Q2%2Z3s - e
Q2Z3
Choose 02 =01 /2 to cancel the cross product term.

av
fh; G =a1x2
1f. = -01%1_. - Q1Z2 -

2 01 G
-%3
2
+ -2-y2 2 -
1
of + -%22 < (
-01
~ 2 +-
+ -%2
2-y2 2
1) 2
Z2
=
Choosing 01 to minimir.e 'Y, we end up with 01 'Y = 1. Thus, the £2 gain is less than or equal to one. Note
that a more comervative upper bound can be obtained by applying (5.40). According to {5.40}, an upper
bound on the C.2 gain from u = u1 to 11 = 111 is given by
')'1 1 4
1- ')'1'Y2 = 1 - ¼= 3
-102
Chapter 6
• 6.1 Define u, u, and y as shown in the next figure. Then,
y=h(t,u)-K1u, u=Ku, u=ii+ii

From
we have
u + jj
y = h(t,u)
+
Figure 6.1: Exercise 6.1
• 6.2
V. = ah(:r)x = h(:e) [ -:r + kh(z)

1 + u ] = kh(z)[h(%)
1 - k.z] + h(z)u ~ yu
• 6.3 Take V(x) = {[ka2 :rf + 2kaz1.:r2 + 4] + o J; 1 h(y) dy, where o > 0 and O < k < 1. V(z) is positive
definite and radially unbounded.
v = o[ka2; 1 + ka.z2 + h(x1)]x2 + 6(kax1 + .:r2)[-h(:r1) - ax2 + u]

= -oka:r1h(zi) + 6(ka - a)x~ + &(ka:r1 + z2)u
. 2
yu - V = (cu:1 + z2)u + &kaz1 h(x1) -6(ka - a)x2 -6(kax1 + x2)u
Take 6 = 1 and k = ofa< 1.
yu - V = az1h(xi) + 6(a - a)z~
The right-hand side is a positive definite function. Hence, the system is strictly passive.
- 103
• 6.4 Since O < 1112 < ok/2, Pis positive definite.

V = + 2xTPx
kh(x1 ):i:1
= kh(x1)X2 + (2aPJ.2X1 + 2p12x2)x2 + (2PJ.2X1 + b2)1-h(xi) - a.z2 + u]
= 2P12X~ - 2p12z1h(x1) + 2PJ.2X1 u - kax~ + kx2u
Hence,
.
yu = V + u2 - 2 2
2P12X2 + 2p12:z:1h(zi) - 2PJ.2x1u + kax2
• 2 --2 2 __ 2
= V .P12X1) - P12Xi + 2P12X1h(.zi) + (ka - 2P12)~
+ (u -
. 2 2 2 __ 'l
~ V - P12X1 + 2a1P12X1 + (ka - 2P12)~
.
V + tJ,(x) =
Since P12 < min{2o1, ak/2}, tJ,(x) is positive definite. Hence, the system is strictly passive.
• 6.5 We have X(s) = (Ms+K)- 1U(s) or (Ms+K)X(s) = U(s) . Thus, the state model for the dynamical
system is
Mx= -Kx+u
Let V(x) = f0~ hT(u)M du ~ O.
V = hT(x)Mx = hT(x)(-Kx + u) $ -hT(x)h(x) + hT(z)u = -yT71+ 'VTu
Hence, the system is output strictly passive.
• 6.6 We have u1 =u2 = u and y =!11 + f/2· Let

T OVi T T
u !11 ~8- /i(:r:1,u)+u 1P1(u)+111P1(J.11)+1P1(.z1)
%1
UT1/2 2:: f)8V, /2(.z2, u) + tJT1i02(u) + !IIP2(112) + ¢:i(.z2)

%2
where Vi = Vi(x1) 'and l'2 = V2(x2). Then,

8V
UT'V = UT (!/1 + 1/2) 2:: lJx f(x, u) + UTcp(u) + 'ViPl (rJ1) + J!lP2(7/2) + tp(z)
where V(z) = Vi(zi) + l'2(x2), i,,,(u) = 'P1(u) +~(u), ,p(z) = fPi(zi) +1P2(z2), and /(z,u) = [
The proof follows from this inequality for the cases of passivity, strict passivity, and input strict passivity.
~:~=~::~ ).
For output strict passivity, we require 11TPi(lt;) 2:: 6wl'lli, for all Vi, for some 6, > O. Then
Yt Pl (J.11) + YiP2(1/2) ~ ½min{61 ,6:,} 1/T11

where we used the fact that
(yi + n)T (y1 + 1,12) ~ 2(7if'Yi + Yi Y2)
• 6.7 G(s) is Hurwitz if and only if 01 and a2 are positive.
Re[GUw)J =Re [ bi+ jbow ] = bia2 + (boa1 - bi)w2

a2 - w 2 + ja1w (02 - ~) 2 + a~w2
Re[GUw)] > 0 for all w ER if and only if b1 > 0 and boa1 ~ b1.
lim w 2Re{G(jw)]:;:; boa1 - b1

c.,-+oc
Thus, G ( s) is strictly positive real if and only if all coefficients are positive and b1 < boa1 .
- 104
• 6.8 For any strictly positive real transfer function, D + DT ~ 0. Since, D + DT is nonsingular, we have
D + Dr > 0. Hence, W is a square nonsingular matrix. Thus,
LTL = LTW(D+DT)- 1WTL

Substituting for wr L from (6.15) we obtain
LTL= (CT -PB)(D+DT)- 1 (C-BTP)
Substituting this expression in (6.14} yields
P((c/2)/ +A]+ [(e/2)1 + AfP + (CT - PB)(D + DT)- 1 (C - BT P) =0

Factoring out the quadratic term, we obtain the given Riccati equation.
• 6.9 Since the system is input strictly passive with .p(u) =ru, we have
UT y
av
2:: ax f (x, u) + EUTu, E > 0
Since the system is finite-gain £2 stable, we have
{TJ t?(t)y(t) dt $ 'Yi f-r, uT(t)u(t) dt + fh, 'Y1 > 0, fJ1 ~0

JTJ lr1
To arrive &t the desired inequality, we need to assume that fJ1 = 0. From the first inequality, we have
V(z(r2))- V(z(r1)) $ lTJ [uT(t)r,(t) -euT(t)u{t)] dt
= {T-J uT(t)11(t) dt-

~
i k1f'1'2 vT(t)u(t) dt - ; 1'1'2 uT(t)u(t) dt
T)
< fTJ uT(t)y(t)

I.,.I
dt-; lri[1'2 uT(t)u(t) dt - f'Yl 11'.,..2 yT(t)y(t) dt
Since this inequality is valid for all T2 ~ Tt ~ 0, we have
8V ( T E T E T T 8V e T e T
-/zu) u 112:: t:1,..f(z,u)+-
2 u u+ 2,.,1 1111
ax 1
- 2 2"(1 V""' I
• 6.10 (a) The equations of motion are
M(q)ij + C(q,q)q + Dq + g(q) = u, y =q

The derivative of V = ½4-TM(q)q + P(q) is given by ·
V = q•TM()-+ 1 .TM· 8P.
q q 2q q + f)q q
= qr[u -C(q,q)q- Dq - g(q)] + ~qTMq + gT(q)q = YT u-yT Dy :5 YT u
where we have used the property that M - 2C is a skew-symmetric matrix. The inequality V :5 yT u shows
that the system is passive.
(b) In this case
- 105
which shows th&t the map from v to y is output strictly passive.

=
(c) With v 0, we have ·
V $ -yTKt1,y = -i?K,ti1 $ 0
v = o '* 11(t) =o ~ ii(t) =o ~ g(q(t)} =o => q(t) =o
Hence, the origin is asymptotically stable. It will be globally asymptotically stable if q =0 is the unique
=
root of g(q) 0 and P(q) is radially unbounded.
• 6.11 (a) Let V = ½J1w? + ½J2wf + ½J3wj

V = J1w1w1 + J2W2W2 + }3W3WJ
= (J2 - J3)W1!JJ2W3 + W1U1 + (J3 - Ji)WJ~~ + W2U2 + {Jl - J2)w1"'2Wa + W3U3 = WTU
which shows that the system is lossless.
(b) With u = -Kw+v, we have
V = -wTKw + vTw => VTw ~ V + Amin(K)llwll~
Hence, the map from v tow is finite gain £2 stable with £2 gain less than or equal to 1/A-m1n(K}.
(c) With u = =
-Kw, we have V -wTKw. Vis p05itive definite and radially unbounded. and Vis negative
definite for all w. Hence, the origin is globally asymptotically stable.
• 6.12
e1 = u1 - Y2 = u1 - h2(.:z:2} - J2(z2)e2
e2 = u2+111 = u2+h1(z1)+J1(z1)e1
Substitute e 2 from the second equation into the first one.
e1 = u1 - ~(z2) - J2(.:z:2)[u2 + h1(z1) + J1(z1)e1]

[J + J2(z2)J1(:z:i)Je1 = u1 - ~{z2) - J2(z2)u2 - J2(.:z:2)h1(z1)
e1 =[I+ J2(z2)Ji(z1)t 1{u1 - h2(z2) - J2(z2)u2 - J2(z2)h1(z1)]
=
e2 u2 + h1 (x1) + J1 (.zi){/ + J2(z2)J1 (z1 ))-1[u1 - h2(z2) - J2(z2}u2 - J2(z2)h1 (xi)]
Similarly, Substituting e1 &om the first equation into the HCOnd one, we obtain
e2 = [J + J1(:i:1)J2(z2)]-1[u2 + h1(x1) + J1(z1)u1 - J1(zi)~(z2)]
Substitute the expressions for e 1 and e2 into the equations
±1 = /i(:r:1) + G1(z1)e1,
• 6.13 Let us start with {6.26)-(6.27).
e1 = u1 - h2(z2, e2),
substitute e2 from the second equation into the first one.
e1 := u1 - h2(z2,u2 + h1(.z1))
The pair (e1, e2) is uniquely determined. Consider now (6.30)-(6.31).
substitute e 2 from the second equation into the first one.
e1 =u1 - h2(t, u2 + h1 (xi))

The pair (e1, e2) is uniquely determined.
- 106
• 6.14 (a) Take Vi = (l/2)(xf + x~) and V2 = J;• ~(s) ds.

Vi= -.:z:2h1(.:z:2) + x2e1 => 111e1 = V1 + Y1h1(11i)
Since h1 E (0, oo], H1 is output strictly passive.
V2 = -z3~(2:3) + e2h2(z3) ~ 112e2 = V2 + x3h2(xs)

Since h 2 E (O, oo], H2 is strictly passive. Thus, the feedback connection is passive.
(b) With e1 = 0, we have
Hence, H 1 is zero-state observable and the origin of the feedback connect.ion is asymptotically stable. It is
globally asymptotically stable if Vi and V2 are radially unbounded. Vi is radially unbounded because it is a
quadratic form and ½ is radially unbounded because
rsa
h2(B) ds ~ fo
rs, s 1
lo 1 + s2 ds = 2 ln(l + x~) -+ oo as j.:z:31-+ oo
Thus, the origin is globally asymptotically stable.
• 6.15 (a) Let V1 = ¼xf + ½~- 2

•
l/i = -.:Z:1 -
4
X2 + Y1e1
Hence, H1 is strictly passive. Let½= ¼xt.
. = -.zs' + Y2e:i
l'2
Hence, H:i is strictly passive. It follows from Theorem 6.1 that the feedback connection is passive.
(b) Since both systems are strictly passive with radially unbounded storage functions, it follows from The--
orem 6.3 that the origin is globally asymptotically stable. ·
• 6.16 To study asymptotic stability, take u1 = 0 and u2 = 0. Then, e1 = -112, e2 = 1/1 · Suppose H2 is
input strictly passive.
$ e2 tn - e; <P2(e:z) = Yt 1/2 - e; <,02(e2),

. T
Vi $ e11/1 = -112T 1/1, V2
. TT TT T
~ <P2(e2) > 0 'v e2 -# 0
Take V = Vi + Vi.
By zero-state observability of H 1 (-H2 ),
Yt (t} =0 ~ z(t) =0
Hence, by LaSalle's invariance principle, the origin is asymptotically stable. Similarly, suppose H1 is output
strictly passive.
·TT TT . TT T
Vi S e1 Y1 -Yi P1(Y1) = -J.12 Y1 - J.11 P1(J11), ½ $ e2J12 = Y1 l/2, l/1 P1(Y1) >0 V Y1-# 0
Take V = Vi + V2. . T .
V S -:111 Pl (111) !:: 0 and V =0 => l/1 =0
By zero-state observability of H1 (-H2),
111 (t) =0 :::::,. x(t) =0
- 107
• 6.11 To study asymptotic st&bility, take u2 =0 and u1 = 0. Then, e2 = 111, e1 = -y2. Suppose H 1 is
input strictly passive.
. TT.TT TT T
½ $ e2112 = !Ii 1h, Vi$ e1 Y1 - e1 <P1(ei} = -112111 - e1 cpi(e1), e 1 <p1(e1) >0 V e1 #0
Take V =½+Vi-
vs: -ef'P1 (e1) $ o a.nd v =o => ex = o => 112 = o
By zero-state observability of H2H1,
1n(t) =0 ~ x(t) =0
Hence, by LaSalle's invariance principle, the origin is asymptotically stable. Similarly, suppose H 2 is output
strictly p&SSive.
Vi :5 eft12 - Yi P2(112) =Yi112 - 1/iP2(112), V1 :5 ef1/1 = -yf1/1, 11TP2(112) > 0 'v 112 ~0
Take V =½+Vi- . T .
V S: -y2 P2(112) S: 0 and V = O => !12 =0
By zero-state observability of H2H1,
112(t) =O * z(t) :0
• 6.18 To study asymptotic stability, take u2 = 0 and u1 = 0. Then, e2 =111, e1 = -y, .
Vi. S: Y1T Q1111 + 2111T S1e1 + e1T R1e1, 'T T T
V2 S: y2 Q2112 + 2112 S2e2 + e2 R2e2
Take V = Vi + 0V2, a > 0.
V$ 11[Q1111 - 231[81112 + yfR1112 + 0(11fQ2112 + 2yf82111 + 11[R2111)
If the matrix is negative semidefinite, the origin is stable. If the matrix is negative definite,
V=O => 11=0
By zercrstate observability
y(t) =0 => z(t) =0
= = =
• 6.19 When u 0, we have e1 u1 -112 = -112, e2 u2 + 1h = 112, and ef111 + ef312 = uf111 + uf y, = O.
=
Use V Vi + ¼ as a Lyapunov function candidate for the closed-loop system.
V ~ er111 - er<Pl (e1) - 11TPl (111) + eIY2 - ef',02(e2} - Yip,z(112)

= -[YiPt(111)+11i<P2(111)]-I11fp,z(V2)-11ft.P1(-112)] $ 0
V=O => 111(pi(111)+cp,(r1i)]=O and 1,l(P2(y,)-cp,(-112))=0 ~ 1,11=0 and 1n=O
Now
By zero-state observability,
111(t) =0 => x1 (t) =0 and y2(t) =0 ~ z2(t) =0
Hence, by the invariance principle, the origin is asymptotically stable. It will be globally asymptotically
stable if Vi and V2 are radially unbounded.
- 108
Chapter 7
• 7 .1 (1) The Nyquist plot of G(s) for w > 0 is shown in Figure 7.1. It is a. circle centered at O) with(-½,
radius equal to ½.
The plot for w < 0 is identical; that is, the Nyquist plot traverses the circle twice as
w changes from-oo to oo. The transfer function has two poles in the right-half plane. Thus, for absolute
stability, the disk D(a,P) must be inside the circle, so that the Nyquist plot encircles the disk twice in the
counterclockwise direction. Clearly, the largest disk is the circle itself. Thus taking
-1 -1 -1
- = -c2, and -
fJ a
=- -
1 + €1
where c 1 > 0 and e2 > 0 are arbitrarily small, we conclude that the system is absolutely stable for the sector
[l + E1, l/E2).
(2) The Nyquist plot of G(s) for w > 0 is shown in Figure 7.2. Since G(s) is Hurwitz we can apply case 2
or case 3 of the circle criterion. For case 2, we see that the Nyquist plot lies to the right of the vertical line
through (-0.021, O). Thus the system is absolutely stable for the sector [0, 47.62). For case 3, the Nyquist
plot should be contained inside the disk D(a, P). We chose a circle of radius 0.114 centered at 0.0732.
Thus, a: = -1/0.1871 =
-5.34, fJ =
1/0.0408 =
24.51, and the system is absolutely stable for the sector
l-5.34, 24.51).
Nyquilt Diagram&
0.6
0.15
0.4
to 0.2 .
i
0.1
0.05
0 ~
l-0.2 .!
-0.4 l -0.05
-0.6 -0.1
-1.5 -1 -0.5 0 0.5 -0.1 0 o., 0.2

Real Axis RHIAxls
Figure 7.1: Exercise 7.1(1). Figure 7.2: Exercise 7.1(2) .
(3) The Nyquist plot is shown in Figure 7.3. Since G(s) is Hurwitz we can apply case 2 or case 3 of the circle
criterion. For case 2, we see that the Nyquist plot lies to the right of the vertical line through (-0.35, O).
Thus the system is absolutely stable for the sector [O, 2.857]. For case 3, the Nyquist plot should be contained
inside the disk D(a,tJ). We chose a circle of radius 1.07 centered at 0.35. Thus, a= -1/1.42 =
-0.704,
fJ = 1/0.72 = 1.389, and the system is absolutely stable for the sector [-0.704, 1.389].
-
109
(4) The transfer function can be simplified to
s-1
G(s) = s2 + 1
It has poles on the imaginary axis. We dose the loop around G(s) with a. the transformed transfer function
G(s) 1-s
GT(s) = 1 + aG{s) = s2 + (a - l)s + 1
is Hurwitz for O <a< 1. Take a= 0.1. The Nyquist plot of GT(s) is shown in Figure 7.4. It lies to the
.
right of the vertical line through (-:2.76,0). Thus the system is absolutely stable for the sector [0.1,0.4632] .
: ,->-·..,, .. Nyquiat Diagrams
0.5
.,...·r ,-.:.:~--
_-)
_.·, :
: \ ··.
\ ·.
en
: \ :. \ ·:
i 0
: \ .
I
-0.5
-0.5 0 0.5 0 5 10
Real Axis RelllAlda
Figure 7.3: Exercise 7.1(3). Figure 7.4: Exercise 7.1{4).
(5) The Nyquist plot is shown in Figure 7.5. The transfer function is Hurwitz. Therefore, we can apply cases
2 and 3 of the circle aiterion. We start with case 2. From the Nyquist plot we find that min{R.e(G(jw)]} >
-0.6. So we chooee fJ = 1/0.6 = 1.67 and conclude that the system is absolutely stable for the sector (0, 1.67].
Now we apply case 3 of the circle criterion. The Nyquist plot should be inside the disk D(a, P). We locate
the center of the disk at (0.2,0) and take the radius as 0.9. The circle is shown in Figure 7.5 (dotted line).
=
Thus, I/a= -1.l => a= -0.91, 1//3 = 0.7 => /3 1.43, and the system is absolutely stable for the sector
[-0.91, 1.43].
(6) The Nyquist plot is shown in Figure 7.6. Since G(s) bas one pole in the right-half plane, we apply case
1 of the circle criterion. The Nyquist plot should encircle the disk D(a,fJ) once CCW. We cb.oee a circle of
= = =
radius 0.09 cent.ered at (-0.16,0). Thus, a= 1/0.25 4, fJ 1/.07 14.29, and the system is absolutely
stable for the sector [4, 14.29).
(7) The Nyquist plot is shown in Figure 7.7. The transfer function is Hurwitz. Therefore, we can apply cases
2 and 3 of the circle criterion. We start with case 2. From the Nyquist plot we find that min{R.e(G(jw)J} >
=
-0.341. So we choose /j = 1/0.341 2.93 and conclude that the system is absolutely stable for the sector
[O, 2.93]. Now we apply case 3 of the circle criterion. The Nyquist plot should be inside the disk D(a, {3). We
locate the center of the disk at (0.3,0) and take the radius as 0.8. The circle is shown in Figure 7.7 (dotted
line). Thus, 1/a = -1.1 => a = = =
-0.91, 1//3 0.5 => /3 2, and the system is absolutely stable for the
sector [-0.91, 2].
(8) The Nyquist plot is shown in Figure 7 .8. The transfer function is Hurwitz. Therefore, we can apply cases
2 a.nd 3 of the circle criterion. We start with case 2. From the Nyquist plot we find that min{Re{G(jw)]} >
-0.08. So we choose /3 =
1/0.08 = 12.5 and conclude that the system is absolutely stable for the sector
(0, 12.5]. Now we apply case 3 of the circle criterion. The Nyquist plot should be inside the disk D(a, /3) .
- 110
1 1 ,--...-----------------
.,......-:..+···· ' ·~~.
'
_.; :. <·.
- -:-:-- ..·""':"'.:--- -
/
0.5 ."J \ ·._ 0.5

\\ ,, ,,
! 0 0
'
..... .. '. . ' .... ·+ · .. ... ·: · .....'.
i
-0.5 -0.5
-1..___ _ ~ _ _ ___._______ -1,.__________.__ __.__ __.__J
-1 -0.5 0 0.5 1 -2 -1.5 -1 -0.5 0

Real Axis Real Axis
We locate the center of the disk at (0.1, 0) and take the radius as 0.21. The circle is shown in Figure 7.8
=
(dotted line). Thus, 1/o. -0.31 ~ o. = -3.223, 1/{J = 0.11 ~ P = 9.1, and the system is absolutely
stable for the sector [-0.31, 9.1]. ·
Nyquilt Diagrams
0.2
0.5
I ! 0.1
I 0 - -:-
' -1).5
-1 -0.5 0 0.5
l -0.1
-0~.__ _ _ _......;.;_ _ _ _
-0.2 0.2
~
• 7.2
(a) The Nyquist plot is shown in Figure 7.9. The plot lies entirely in the right-half plane. By case 2 of the
circle criterion, we conclude that the system is absolutely stable for the sector (0, P], where /3 can be made
arbitrarily large. Consequently, the system is absolutely stable for the sector [O, 1].
(b) The nonlinearity sat(y) belongs to the sector (0, 1]. Hence, the system has a globally exponentially stable
equilibrium point at the origin. This implies that there can be no periodic solutions, since every solution
must converge to the origin.
0 = -xi - h(x1 + ~2), 0 = xi - xi - 2h(x1 + x2)

Multiply the first equation by -2 and add to the second equation to obtain the equation
0 = -(x1 + x2) - 4h(x1 + xz)
-
111
0.5
(I)
~
Q 0
]
-0.5
0 0.5 1 1.5 2
Real Axis
which has the unique solution x1 + z2 = 0 due to the properties of h.

Xt + X2 = 0 => h(x1 + z2) = 0 => Z1 = 0 => Z2 = 0
Hence, the origin is the unique equilibrium point.
(b) Let y = x1 + z2 and ti= -h(z1 + z2). Rewrite the system equation as
z1 = -z1 +ti
±2 = x1 -z2 + 2u
y = Z1 +z2
u = -h(z1 +z2}
The system is now model~ as & feedback connection with the transfer function
3s+4
G(s) = (s + 1)2
and the nonlinear element tp(y) =
h(y). The Nyquist plot of G(,) lies completely in the right-half plane.
Hence, by the circle criterion, the system is absolutely stable for the sector [O, P), where /3 can be arbitrarily
large. This sector includes the nonlinearity h. Thus, the origin is globally asymptotically stable.
• 7.4 Let y = .z1 and ti= -qcoswt z1. Rewrite the system equation as
z1 = Z2
±2 = -(p2 + a2)z1 - 2µz2 + u
1J = :i:1
u = -qcoswt y
The system is now modeled as a feedback connection with
1
G(s) - ...,,....---...-.,,.
- s2 + 2µ 8 + µ2 + 02
and ¢(t,y) = qcoswt y. The function tJ, belongs to the sector [-q, q]. We will apply the result of Example 7.1.
We have ltJ.,(t,y)j ~ 1'2IYI with -r, = q. We also have
IG(iw)l2 = (µ2 + a2)2 + Z:2(µ2 _ 0 2) + w"
- 112
It can be verified that l

,..2+4111 forµ~ a
1'1 = sup IG(iw)I
wER
={ 1
forµ< a
21,611'
The system is absolutely stable if q11 < 1. Since the system is linear, this condition ensures that the origin
is exponentially stable.
• 7 .5 Represent the system as
z =Ax+ Bu, y == Cx, u::;: E(t)Cx = E(t)y ~ -'¢(t,y)
This problem is a special case of Example 7.1, with
ll1/l(t,y)ll2 $ IIE(t)ll2IIYlh $ IIYll2 => 1'2 =1

and
'Y1 = supa,n ,l:[G(jw)] < 1
11
wER
The condition for absolute stability (1'11'2 < 1) is satisfied. Hence, the origin is uniformly asymptotically
stable.
• 7.6
(a) The closed-loop state equation is
x = Az + BLsat(-Fz/L) = (A- BF)x + B[Fx - Lsat(Fz/L)]
Setting y = Fz and t/J(y) = Lsat(y/L} -y, we represent the system as

± = (A - BF):z - B1/l(y}, y = F:z
which is in the form of Figure 7.1 of the text, with G(s) = F(sl - A+ BF)- 1 B.
(b) We have
Hence
This is a special case of Example 7.1, with
"Y2 = 1 r, 6 , and 'Y1

+
=wER
sup 1lG(jw}ll2
Hence, the system is absolutely stable with a finite domain if
6 r, sup IIG(jw)ll2 < 1

1
+ "'ER
(c)
G(s) _ 2s+ 1
- s 2 +s+0.5
It can be checked that sup IG(jw}I $ 2.55. Hence, the system is absolutely stable with a finite domain if
6 l
1 + r, < 2.55 => r, < 0.6452
- 113
To estimate the region of attraction, we apply the loop transformation of Figure 7.3 of the text.
u =u - K1y = ii+ "12Y,

The state model of the transformed system is
:i: =
(A - BF+ "t2BF)x + Bu
y ::: 2"f2FZ + u
u = -',b(y)
where'¢ E {0, ooJ. We need to find a matrix P that satisfies (7.6)-(7.8) for the transformed system. Equations
(7.6)-(7.8) take the form
P(A- BF+ ,-,2 BF) +(A-BF+ 72BFf P = -LTL- eP

PB= 2-y2FT - ./2.LT
0.4638 0.3757]
E = 0.002, p = [ 0.3757 2_2338 , and L = [ 0.2859 -0.4764 ]
satisfy (7.6)-(7.8). The region of attraction can be estimated using V(x) = zTPx, as in Example 7.4.
• 7.7 (1) The transfer function has poles in the right-half plane. Therefore, we start by applying a loop
transformation. We close the loop around G(a) with a. The transformed transfer function is
G (•) _ G(s) _ s
T - l + aG(s) - , 2 + (a - l)a + 1
Take a= 1.1. The Popov plot of GT(•) is shown in Figure 7.10. The largest sector is obtained with 'Y = 0
so that the line of slope l h is vertical and intersects the horizontal axis at a point arbitrarily cloae to zero.
The transformed system is absolutely stable for the· sector {0, k], where k can be arbitrarily large. Hence,
the original system is absolutely stable for the sector (1.1,,8], where /3 can be arbitrarily large.
(2) G(a) is Hurwitz. Its Popov plot is shown in Figure 7.11. The Popov plot approaches the origin tangent
to a line wh08e slope is 4.926. We take-,= 1/4.926::: 0.203 and k > 0 can be arbitrarily large. Let us verify
that -y = 0.203 is acceptable. Since G(a) has relative degree two, the product CB is zero in any state-space
realization of G(s). Thus, the condition 2/k+...,CB +-yBTcT ~ 0 is always satisfied. The poles of G(s) are
=
-2 and -3. Thus the condition 1 + -y..\ ::f: 0 is satisfied for 1' 0.203. We conclude that the given systeDJ. is
absolutely stable for the sector [O, k] where k is arbitrarily large. 1
(8) G(s) is Hurwitz. Its Popov plot is shown in Figure 7.12. The Popov plot approaches the origin tangent
to a line whose slope is 1. We take 'Y = l and k > 0 can be arbitrarily large. Let us verify that 'Y = l is
acceptable. Since G(s) has relative degree two, the product CB is 7.erO in any state-space realization of G(s).
Thus, the condition 2/k + -yCB + "(BTCT ~ 0 is always satisfied. The poles of G(a) are -0.5 ± j0.5v3.
Thus the condition 1 + ")'A ¥ 0 is satis&d for -y =
1. We conclude that the given system is absolutely
stable for the sector [0, k] where k is arbitrarily large. We can obtain a sector with o < 0 by applying a
loop transformation. Using the Routh-Hurwitz criterion, it can be shown that GT(s) = G(s)/[1 + aG(s)] is
= =
Hurwitz if o > -1. Take o -0.9 to obtain GT(,) 1/(a2 + s + 0.1). The Popov plot of GT(s) is shown in
Figure 7.13. The Popov plot approaches the origin tangent to a line whose slope is 1 We take ""f ::;;: 1 and k > 0
can be arbitrarily large. Let us verify that "Y =
1 is acceptable. Since GT(s) has relative degree two, the
product CB is zero in any ~space realization of GT(s). Thus, the condition 2/k+-yCB +-.,BTCT ~ 0 is
always satisfied. The poles of Gr(s) are -0.1127 and -0.8873. Thus the condition 1 +-y..\ ::p; 0 is satisfied for
'Y = 1. We conclude that the given system is absolutely stable for the sector [-0.9, k] where k is arbitrarily
large. Compare these sectors with the ones obtained using the circle criterion in Exercise 7.1.
- 114
0...--.........- ~ - ~ - - ~ ~
4 ...... ..
-0.05
r a
0
a-
ra -a.1 :
.
-2
-0.15 :
0 2 4 6 8 10 -0.05 0 0.05 0.1 0.15 0.2

Re(GJ Re{GJ
0 ··· :·· ··· . . . .. .. ... ..... . . ...

. .. . .
.. .
..
.
3 ... . .. . .. .. . . . .. .. ~ ·· ·- -· ~· ···· ·~ ~· -·, · · ·
. . . .
-0.2 . .. : . .. . .:. .... ·; .... : .. .. . : .. ....:. . . . 2 . •.· . . , , . · ·: ·· , .. ' ' ; . . ''.'. '. ' , . . , ' ' '. . . .. . '
. . .
1 ..: . . .. .: . . .. . .. . . . .. ; . . . .. .. , . ... .
.
. . . .
i:j'-0.4 . ··: · .. ···:· · ·· ·<..... . , . ... . .:,. · · ·:··
I : .. ; ; . i's-1o~v[:.··~~;--- ~·-~·.·~·.~· ~C.J
":" " '" ' ... .. ..:.. ..... , ... ...
9 -0.6 ··· !·· .. " :· · .. ·:· ····· ;····· :··· ..:......
.. . .
.
.. ... .. .. -2 ·· · · ·'. · · · · · .; .. . .... _; . . ... .. ;- .. , . .. .
. ' .
. ', .
-0.8 ··> ····;- · -3 . ·:· · · · ·· •·:·
.
· · ··· · ,· · · ···· :.····· ·· ; ... . .. .
-.c .·:· .. " ...:... .. .. :·" .. ' 'i .. .. .. .: .. .. . ..
-0.2 0 0.2 o..c 0.6 0.8 0 2 ,4 6 8 10
Rl(G) Ae(G]
(4) The transfer function can be simplified to

s-l
G(s) = ~ l
8 +
It has poles on the imaginary axis. We close the loop around G(a) with a. the transformed transfer function
GT(s) = G(s) = 1- s
1 + oG'(s) a2 + (a - l)s + 1
is Hurwitz for O < a < 1. Take a = 0.1. The Popov plot is shown in Figure 7.14. The Popov plot lies to the
right of the vertical line through (-2.76,0). Thus, we take "'1 = 0 and k = 1/2.76 = 0.3623. Therefore, the
system is absolutely stable for the sector [0.1, 0.4623). This is the same sector obtained by using the circle
criterion because -y = 0.
(S) The Popov plot is shown in Figure 7.15. It lies to the right of (and tangent to) a line through (-0.5,0}
whose slope is 3. Thus, we take "Y = 1/3 and k = 1/0.505 = 1.98. Let us verify that "f = 1/3 is acceptable.
Expanding G(s) as a power series in 1/s, it can be seen that the coefficient of 1/s is -1. Therefore, in any
state-space realization of G(s), CB= -1. Thus, 2/k+ 7GB + 7BTCT = 0.3433 > 0. The poles of G(s) are
-1 and -1. Thus, 1 + -y,\ = 2/3 # 0. Hence, the system is absolutely stable for the sector {O, 1.98).
1We can alao show absolute stability with a aector (a,.8] where a ill negative by applyi11g a loop uauaforma.tion. See the
soluiion of part (3).
- 115
0.8 .. . . ...... .. · ... . . . . .. ·· · · · ··· · .. .

I :
8 · i-- ·· ·:··· · . ... 0.6 ····· · .... . . · ·· · ··· · · ·· . . , .. . . ... . . .
a- 6 ... 1 • .... ; . • •. • • .. . . . •• . .. . ...• • .. . . . 8' 0.4 . ... . . . .. . . . .. . . . .. . . · · · ·· ··· · . . . ,
:[ . I 0.2 · ·· · · . . . ..... .. . . ... .. .. . .
8 4 . .. ·· · · · :· · · · · · · ·
s
2 .. ·r .... : ... .. ... .:" . .. ... .. :· . .. .. .
0 · ··'·· · . . ; . .... . ... . .:.. . .. . . .... · . . .... . .
I : :
-2 . . .l . . .. ; . .
0 5 10 ~-!5 0 0.5
Re[G) Rll{GJ
Figure 7.14: Exercise 7.7(4). Figure 7.15: Exercise 7.7(5}.
(6) G(s) is not Hurwitz. We need to apply a loop transformation with o > 0 such tha.t GT(.s) G(a)/[1 + =
oG(a)} is Hurwitz. Using the Routh·Burwitz criterion, it can be shown that GT(•) is Hurwitz if o > 4. Take
a= 4.1 to obtain GT(•)= (1 + 1)/(s3 + 3,2 + 4.ls + 0.1). The Popov plot of GT(s) is shown in Figure 7.16.
The Popov plot approaches the origin tangent to a line whose slope is 2. We take 'Y = 0.5 and le > 0 can be-
=
arbitrarily large. Let us verify that 'Y 0.5 is acceptable. Since GT(B) has relative degree two, the product
CB is zero in any state-space realization of GT(,). Thus, the condition 2/k +"{CB+ "(BTCfI' ~ 0 is always
satisfied. The poles of GT(a) are -0.0248 and -l.4876±jl.34 66. Thus the condition 1 + "Y>. :/: 0 is satisfied
for"/= 0.5.We conclude that the given system is absolutely stable for the sector (4.1, k] where le is arbitrarily
large. Compare this sector with the one obtained using the circle criterion in Exercise 7.1.
(7) The Popov plot is shown in Figure 7.17. It lies to the right of (and tangent to) a line through (-0.25, 0)
whose slope is· 1. Thus, we take -y = 1 and k = =
1/0.2505 3.99. Let us verify that -y = 1 is acceptable.
Since G(a) has relative degree four, the product CB is zero in any state-space realization of G(s). Thus, the
condition 2/'/r. +,.,CB+ 7BTCT ~ 0 is always satisfied. The poles of G(a) are at -1. Hence, .., = 1 does not
satisfy the condition l +'Y.>. ~ 0. We take -y = 1.01. It can be verified that the Popov plot lies to the right of
=
the line through (-0.2505, O) wh011e slope is 1/1.01 0.99 (see Figure 7.17). Thus, the system is absolutely
stable for the sector (0, 3.99).
0.5 - - - - . . . - - - - - - ,
o~---------------
0.2 . .. . . .. . . . . .t ; ' .. . .. . . ... . . ; ... . . . .. . .. . .
I : :
0.1 ....... . ./ ; .. . . . . . ..... . . . . .. . .... ..
I
a
o .. ..... .... .
-0.3 .. .. ........~~ ··· ··: · · -· ··· ···· ·

~-4 _ _ _ _ _ _ _...____ __,
0 0.5 ~-5 0 0.5
Re(GJ Re(G]
(8) The Popov plot is shown in Figure 7.18. It lies to the right of {and tangent to) a line through (-0.0556, O)
=
whose slope is 1.6. Thus, we take 1' = 1/1.6 0.625 and le= 1/0.056 17.86. Let us verify that 1' 0.625 = =
- 116
is acceptable. Since G(s) has relative degree four, the product CB is zero in any state-space realization of
G(s). Thus, the condition 2/k + -yCB + -yBTCT ~ 0 is always satisfied. The poles of G(s) are at -1 and
-2. Hence, the condition 1 + -y>.-:/- 0 is satisfied for 'Y = 0.625. Thus, the system is absolutely stable for the
sector {O, 17.86].
0.05------------
.. ......... ., ... .... .. .....
. ..... .
. .
!6' -0.05 ' . '

. .. . . ' . . -:- . . . . ' .. . ; ; . ' ' ' ... •:- . " . '
8
-0.1
-0.15 ..__ ____._ _ _ _ _ _ ____.

-0.1 0 0.1 0.2 0.3
Re{G]
• 7.8 (a) Write V as

V ~ zT Pz + ad2 v2 + bfo'" t/J(u) du
Since p = pT > 0 and ~ > 0, the first two terms form a positive definite quadratic function of {z, v).
Since b ~ 0 and 1/J is a first quadrant-third quadrant nonlinearity, the integral term is nonnegative. Hence,
Vis positive definite. It is also radially unbounded since V(z,v) ~ zTPz+ adv2 .
(b)
v = zT(PA + AT P)z - 2zTPB¢- 2adlv¢ + btp(CAz -CBt/)-d,p)

2ad
$ zT(PA + ATP)z-2zTPB1J,- 2ad(y- Cz)'IJ) + btpCAz -"bCB1J}2 -bd,/J2 - Tt/J('¢- ky)
= zT(PA+ATP)z-2zT(PB-½bATCT-adCT),/-(bCB+bd+ 2:d)tJ,2
= zT(PA + ATP)z- 2zT(PB-w)1/J- -y1/i
where
2ad
w = ½bATcr + adcr, 1 = bCB + bd + T
Choose b such that 'Y ~ 0. Suppose there are P = pT > 0, L and E > 0 which satisfy
PA+ATP = -L7 L-eP
PB = w-LT._n
Then
V :$ -EzTPz-zTLTLz+2zTLTv""f11J--y'l/;2
= -u7 Pz- (zTLT - ,nt/J)(Lz - ../y'I/J) :$ -ezTpz
- 117
Hence, V is negative semidefinite.

V = 0 :::} z(t) =0 => ¢(y(t)) =0 => y(t) =0 :::} v(t) =0
Thus, by LaSal.le's theorem, the origin is globally asymptotically stable. Now, by Lem.ma 6.3, the existence
of P, L, and ~ satisfyjng the foregoing equations is equivalent to the strict positive realness of
Z(s) = wT(sl -A)- 1 B + h·
= ad [(c +rJCA}(sl - A)- B + ¼+ 11(d + CB)]
1
where,,== b/2ad. By Lemma. 6.1, we want Re[Z(jw)] > 0 for all w, that is,
¼+ Re[(C + T]CA)(jwl - A)- 1 B + '1(d +CB}]> 0

Noting that
(1 +'ls)C(sl - A)- 1 B = C(sl -Ar 1 B +f'JsC (!1 +_!_A+···] B

2B s
= C(sl - A)- 1 B + f'/CB + ,.,CA [!1

8
+_!_A+
.,2
···] B
= (C + 11CA)(.,J - A)- 1 B + 11GB
a.nd
we obtain
Re[(C + ,,CA)Uwl - A)-1 B + 17(d + CB)] = Re { (1 + iWTJ) [c(jwl - A)-1 B + i~]}

and the condition for absolute stability reduces to
¼+ Re[(l + jWTJ)G(jw)] > 0, Vw ER

• 7.9 (a) From the block diagram, we have
I
E(s) = -Y(s) - kU1(s) = -(H(s) + k]U1 (s) = -[H(s) + k]-U(s)
8
Hence, the system can be represented as the feedback connection of Figure 7.1 of the text with
G(a} = ![H(s) + k] = s + 6 + k(s + 2)(s + 3)

s s(s + 2)(s + 3}
(b) From the Popov criterion (Exercise 7.8), the system is absolutely stable if
j + Re[(l + jW7J)G(jw)] > 0, Vw e R
To simplify the calculations, chooser,= ½· Then
Re[(l + jwt7)G(jw)J = ~~9+:~;) 3

which is pOBitive for k > f · Thus, we take kc = ½·
- 118
• 7.10
(1)
IJl'(a) = -1
11'0
1,.. a sin 9 dD
0
5 6
sin6 6 = [½(l - cos 28}] 3 = Hl - 3 cos 29 + 3cos 2 20 - cos3 20} = t (1 - 3cos 28 + J+ Jcos 49 - cos3 28)
The terms cos 28, cos3 26, and cos 48 integrate to zero over O to 1r. Thus
111(a) == -
21r
1ra O 16
5
-5a dD = -Sa'
8
(2)
'l'(a) = -21,r a°'
,ra o
sin6 8 dD
-sed8 = -
/ sm 5i.,e
sm cos B - 4.e 6 Be
15 sm cos - 15 cos
Upon substituting the limit from Oto 1r, the first two terms vanish and the third term gives (16/15). Thus
it(a) = 32a3
15,r
(3) We can write the given function as ,p(y) =k +A sgn(s,). The describing function of the signum
nonlinearity is given in Example 7.6. Thus
4A
t'(a) = k+ -
1(0,
(4) For a SA, tt,(asin8) =0 => "t(a) = O. For a > A, 1ji(asin8) is given by
for OS8 S P &1t - /3 $. 8 5. 1r
,p(a sin 8) = { ~. for/3<8<1f-/3
where .8 = sin-1 (A/a). Thus
•(a)=-
,ra
41""
fJ
12 4B oosfJ=-
Bsin8d8=-·
,ra
4Bg
1ra
1--
2
a2
(5) The given function can be written as ¢(y) = 1P1(11) + ,h(y), where
ifO~y<A ifOSyB
if y_
¢ 1 and 1/J, are special cases of the function treated in Example 7.7. Using the result of that Example, we
obtain
for a< A { 0, fora SB
for a> A ' i' 2 (a) = -k + kN (j) fora>B
The sum +-1(a)+ 1'2(a} gives the desired de$Crlbing function.

• 7.11 (1)
G (j ) = 1 - jw = I - w 2 - 2jw
w jw(l + jw) jw(l + w2 )
- 119
. -2 . -l+w2
Re[G(jw)] = 1 + w 2 , Im(G(jw)] = w{l + w 2 )
Im[G(jw)) = 0 => w = 1, Re[G(j)J = -1
= =
For 1P(J1) y 5 , we have •(a) 5a4 /8. The equation 1 - IJl(a) = 0 has a unique solution a = (f) ¼= 1.125.
There is a possibility of a periodic solution of amplitude close to 1.125 and frequency close to 1 rad/sec.
(2) The transfer function is the same as in part (1). The describing function is given by
0, for a$ 1
ik(a) = { 2[1 - N{a)], for 1 $a$ 1.5
2[N(a/l.5) - N(a)J, for a::! 1.5
By using Matlab, it can be checked that the equation 1 - i'(a) = 0 has no solution. So, we expect that the
system will not have sustained oscillation.
(3)
G(jw)
= 1 _ (1-jw) 6
(1 + jw) - (1 + w2 ) 8
8
1 + 6(-jw) + 15(-jw)2 + 20{-jw) 3 + 15(-jw)4 + 6(-jw)5 + (-jw)11

= (1 + w2)'
= 2 4
1 - 15w + 15w - w + j[-6w + 2(x.;3 - &.,5)
6
(1 + w2)6
Im[G(jw)J = 0 => -6 + 20w2 - &,' = 0 => ,.,.,2 =3 or w 2 = !3

Re[G(j3)] = !, Re {G (i½)] = - :
=
From Example 7.6, we know that iJ'(a) 4/wa.. For w2 = 3, the equation 1 + •(a)Re[G) = 0 has no solution.
For w2 = ½, the equation has a unique root a =
27/lfnr. TbUB, we expect that the system will have a
periodic solution with amplitude close to 27/161r and frequency close to 1/,/3 rail/sec.
(4)
G(j~) = jw + 6 =
= -j(6 + jw)(2 - jw)(3 - jw) -w(24 + w2) - j(36- w2 )
jw(jw + 2)(jw + 3) w(4 + w2)(9 + w2) w(4 + w2)(9 + w2)
Im(G(jw)] = 0 => 36 - w 2 = 0 => w = 6
= =
Re[G(j6)] -1/30 and the equation l + tt(a)ReG 0 has the unique solution o 2/151r. Thus, we expect=
that the system will have a periodic solution with amplitude close to 2/151r and frequency close to 6 rtw./sec.
(5)
G(jw) = jw = jw = -w2 + jw(l - w2)
-w2 - jw + 1 1 - w2 + jw (1 - w 2 ) 2 + w2
Im[G(jw)] = 0 => w =1
=
Re[G(j)] = -1 and the equation 1 + '1t(a)ReG 0 has the unique solution a= (8/5) 114 • Thus, we expect
that the system will have a periodic solution with amplitude close to (8/5) 11' and frequency close to 1
ran/sec.
(6)
G(jw) = 5(1 + j4w) = 5(1 + 4jw)(2 - jw)2 = 5(4 + 15w2 ) + j20w(3- w2)
4(jw)2(2 + jw)2 -4w2(4 + w2)2 -4'.,2(4 + w2)2
Im[G(jw)} = 0 => w(3 - w2 ) =0 => w =v'3
- 120
Re[G(jvl}] = -5/12. The equation 1 + it(a)ReG =0 yields l'(a) = 12/5.

4 12 10
w(a) = 2 + 1r0 = 5 ~a=-;-
Thus, we expect that the system will have a periodic solution with amplitude close to l0/1r and frequency
close to 1 raJJ/sec.
(7) From part (6), the equation 1 + '.lf(a)ReG = 0 has the unique solution a= (8/5) 114 •
for a :5 1
for a~ l
~(a) :s: 4/,ro. Hence, it(a) = 12/5 has no solution. Thus, we eJg>ect that the system will not have sustained
oscillation.
(8) From part (6), the equation 1 + i'(a)ReG = 0 has the unique solution a= (8/5) 114 . The describing
function is given in the solution of part (2). By using Matlab; it can be verified that t)(a) = 12/5 has no
solution. Thus, we expect that the system will not have sustained oscillation.
(9)
. 1 1 - 3w2 - jw(3 - w 2 )
G(jw) = (1 + jw)S = (1 +w2)3
Im(G(jw)] = 0 ~ w(3 - w2) =0 ~ w = v'3

=
Re[G(jv'3)] -1/8. The equation 1 + ~(a)Re[G] = 0 yields ll(a) = 8. From Example 7.6, the describing
function of the signum function is i'(o) = 4/(rra). The equation ~(a) = 8 bu the solution a = 1/(2,r).
Thus, we expect that the system will have a periodic solution with amplitude close to 1/(211') and frequency
close to v'3 r&Ai/sec.
(10) From part (9), 'lt{a) = 8. From Example 7.7. the describing function of the saturation function satisfies
t(a) :$ 1. Thus, •(a)= 8 bas no solution and we expect that the .system will not have sustained oscillation.
• T.12 With reference to Section 1.2.4, represent the negative resistance oscillator in the z--coordinates
.i1 =z2, i2 = -z1 - h(z2)

where h(v) = -v+ v3 - fv
11 • including the -v term with the linear system, we can represent the system in
the form of Figure 7.1 with
A= [ ~l !], B = [ ~], C= [ 0 1 J, v,(y) = 1/ - ~r/i

-1 s . -w2 + jw(l - w2)
G(s) = C(sl - A) B = , G(Jw) = (l 2)2
s -s+ 1
2
-w +w2
Im G(jw) => w =1
=0
Re G(j) = -1 => ~(a) = 1
On the other hand,
IP(a) = :a 1"" [(asin8) 3 - &(asin8) 11] sin9 d8 = 3: 2 - ~

4
2 4 /n
The equation 3: - 08 = I has two ~lutions: a = 2 and a = v 2. Thus the harmonic balance equation
has two solutions: (w., a.}= (1, 2) and (w., a.)= (1, v'2). We expect that the system will have two periodic
solutions, the first one has amplitude close to 2 and frequency close to l ratl/sec, and the second one has
- 121
amplitude close to ..fa and frequency close to 1 rad/sec. The phase portrait (Figure 7.19) shows that the
system has two limit cycles. The amplitude of z2 in the inner limit cycle is about 1.4 while that of the
outer limit cycle is about 2. These· numbers correlate reasonably well with the estimates of the amplitude
of the first harmonic obtained using the describing function method. From Figure 7.20, we can estimate
the frequency of oscillation by 0.9657 rad/sec, which is close to the frequency of oscilla.tion predicted by the
describing function method.
2..---------~-----,
-2....__ _ _ _ _ _ _ _ ____.
-4-----------
-4 -2 0 2
90 9:2 96 98 100
z,
Figure 7.19: Exerdse 7.12. Figure 7.20: Exercise 7 .12.
• 7.13
G(jw) = 2bjw = -2b w2 + j2bw{l -
2 w2 )
1 - w2 - jbw (1 - w2)2 + b2w2
Im(G(jw)J == 0 ::::> w =1
=
R.e[G(j)) = -2. The equation 1 + 9(a)ReG 0 yields 'il(a) = ½- The describing function of the saturation
nonlinearity is given in Example 7.7. By using Matlab, it can be verified that 'i(a) = ½has a solution
a= 2.47. Thus, we expect that the system will have a periodic solution with amplitude close to 2.47 and
frequency close to 1 ra4/sec. To confirm this conjecture, we apply Theorem 7.4. The saturation nonlinearity
satisfies the slope restriction with o = 0 and /J 1. =
_I_ = 1- w2 - jbw = _! _ . (. 1 - w2 )
G(jw) 2bjw 2 1 2bw
The inverse Nyquist plot is a vertical line through the point (-0.5,0). The critical circle has its center at
the point (-0.5, 0) and its radius is 0.5. The inverse Nyquist plot intersects the critical circle at the points
(-0.5, -0.5) and (-0.5, 0.5). The frequencies at these two points a.re given, approximately, by w = 1 - b/2
and w = 1 + b/2. So, we limit our analysis to the frequency band w E [1 - b/2, 1 + b/2].
l o+/3 + G(jw)
-2-
1 I= 11-w21j2bw
For w > 1, the right-hand side is a monotonically increasing function of w. Hence, p(w) is calculated at k =3
to be
9w2- l
p(w) = 6bw
For 1 :$ w 5 1 + b/2,
4 9(1 + b/2) 2 - I 4
3b $ p(w) $ 6b(1 + b/2) :::: 3b
(1 ++
1
9bf8)
b/2
- 122
Thus, for small b, p(w) is of order 0(1/b). Consequently,

(0.5) 2 3b
a(w) = p(w) - 0.5 i:::i 16 =O(b)
The uncertainty band will be very narrow for sufficiently small b. It can be verified that the uncertainty
band defines a complete intersection. Theorem 7.4 confirms that there is a half-wave symmetric periodic
solution. The amplitude of oscillation will be O(b) close to 2.47 and the frequency of oscillation will be O(b)
close to 1 rad/sec .
• 7.14 (a) 'PE (0, b). Since G(s) is Hurwitz, we apply case 2 of the circle criterion. From the Nyquist plot,
we see tha.t Re[G(jw)] 2: -0.085. Thus, the largest bis b 1/0.085 = 11.76. =
(b) The Popov plot lies to the right of a line having slope = 1.2 and intersecting the real axis at -0.06.
The slope correspondl! to "f = =
1/1.2 0.833. Since the eigenvalues of A are at -1 and -2, the condition
(1 + ,\1') ~ O is satisfied. Also, since G(s) bas relative degree higher than one, CB =
0. Hence, the
condition 2 + 2"'(kCB > 0 is satisfied for any -y: ThWI, the choice 'Y = 0.833 is acceptable and the largest bis
=
b = 1/0.06 16.67.
(c) ¢ is a special case of the piecewise linear function of Example 7.7 with s1 = b, s2 = 0, and d = 1/b. The
= -;; [ ( ) F&])
describing function is
2
2b . -1 1 1 1
')(a) sm ab + ab 1- ab
.
GUw)-~--..;;,_.;..;....- w2 - .....;...~
(1 - w 2 - 2jw)(4 ___ 4jw)
- (1 + w2)2(4 + w2)2
Im[G(jw)) = -2w(6-3w')
(·) == 0 => w = ../2
1 + ii'(a).Re[G(j./2)] = 0 => ~(a) = 18
= =
Because tJ,(o) starts from b at a 0 and decreases after a 1/b, the equation 'f/J(a) = 18 has a solution if
b > 18. The &equency of oscillation will be clme tow= \1'2.
=
(d) For b = 10, the slope restrictions are a= 0 and fJ 10. The inverse Nyquist plot and the critical circle
are shown in Figure 7.21. The inverse Nyquist plot is plotted only for w e [O, 2). It keeps moving farther
away for higher frequencies. Because the inverse Nyquist plot is always outside the critical circle, there is no
oscillation.
(e) For b = 30, the slope restrictions are a= 0 and fJ = 30. The inverse Nyquist plot and the critical circle
are shown in Fi e 7 .23. The error circle are plotted for the following points:
w 1 G(jw a(w
1.3 (-15.1139,2.418 0.7536
1.389 (-17.3589,0.589) 0.5911
1.436 (-18.555,-0.535) 0.5231
1.45 (-18.912,-0.8917) 0.5048
1.5 (-20.1875,-2.25) 0.4456
The circles tangent to the real axis correspond to w 1.389 and w = 1.436. Thus, w1 = =
1.389 rad/se,c
and W2 = 1.436 raJJ/sec. The boundaries of the band of uncertainty intersects the real axis &t -17 and
-18.75. By plotting ti(a) versus a, the values of a1 and 02 are determined as a1 = .0649 and a2 = 0.0721.
It can be verified that the regularity conditions are satisfied. Therefore, there is oscillation of frequency
we [1.389, 1.436] and the amplitude of the first harmonic of the output is a E (0.0649,0.0721]. At w = 1.389
and a= 0.0721, we have
;£ 211 /"' 1/~(t) dt ::S [2 x 0.591!x 0.0721] 2 = 8 x 10_ 8
- 123
The energy content in the first harmonic is
W
-
121'/w a 2 sin2 (wt) dt = a2 = (0.0721)2 = 0.52 x 10-2
7r 0
the ratio is 15.38 x 10-4 • Taking the square root, the percentage of the higher harmonics to the first one
is 3.9%. The simulation is shown in Figure 7.22. The simulation shows a period of oscillation of about 4.4
sec., that is, w ~ 1.428 rad/sec.. The amplitude of oscillation is about 0.068. These results are consistent
with the ones we obtained using the describing function method.
10---------------, 0.1....--------------.
! : ::: ::
i -5 . . ....... ; .. . .... . . ;.
~ .
8 . .. ; ... .
.
>, 0
J-10 .... - .. .\. .... . . .... : ....

-0.05
-20L-------------'
-30 -20 -10 0 -0.10
5 10 15 20
RulAxls t
Figure 7.21: Exercise 7.14. Figure 7.22: Exercise 7.14.
• 7.16 (a) The nonlinearity belongs to the sector [O,b]. By using Matlab to find the Nyquist plot of
G(jw), we found that :minRe(G(jw)J = -1.1638. Hence, the origin is globally asymptotically stable for
=
b < 1/1.1638 0.8593.
(b) By using Matlab, we found that the Popov plot intersects the real axis at -0.5564. With-,= 1/1.1 we
can draw a line of slope 1/1 which paases through the point (-0.5564, O) and is to the left of the Popov plot.
The choice..,= 1.1 satisfies the condition 1 + 'Y~.,.
0 since~= -1 or -2. The condition 1 +-yJcCB > O i8
satimed !or any-, aiDce CB= 0 (the tramfer function baa relative degree two). Thus, the origin is globally
asymptotically stable for b < 1/0.5564 = 1.7972.
(c) t/J is a special case of the piecewise linear function of Example 7.7 with , 1 = b, , 2 = 0, and 6 1/b. The =
describing function is
The harmonic balance equation is
1
1 + G(jw)jl(a) = 0 ~ G(jw) + i'(a) = 0
1 2
10 [2(1- 2w ) =O+ jw(5 - w2 )] + fl(a)
The imaginary part yields w 2 = 5; then the real part yields t,(o) = 1.8. Because 1/J(a) starts from b at
a= O and decreases after a= 1/b, the equation t/J(o.) = 1.8 has a solution if b > 1.8. Thus, the describing
function method predicts that there will oecillation for b > 1.8. The frequency of oscillation will be close to
w = =
VS 2.2361.
- 124
10
I ot---+-----"""""2~,r;,_---------+-......~
I~
-10
-15
-20~-----------------.__ __.-5___.___
-35 -30 -25 -20 -15 -10 0
~
5
RNIAm
• 7.16 (a)
G(jw) = (1 - 3w2) + jw(w2 - 3)
(1 +w2)3
Im G(jw) =0 => w = v'3
Re G(jv'3) =- ~ => +(a) =8
With s 1 ;,., 10, s 2 = 0 and 6 =0.1, the describing function is given by
if(a) = 20 [sin-I (2-)
1r 10a
+ _l Jl _
IOa
( 1 )']
10a
Plotting 1,b(a) using Matlab, we determined that IJl(a) = 8 corresponds to a = 0.146. Thus we predict that
there is an oscillation of frequency close to v3 rad/s and amplitude of the first harmonic close to 0.146.
(b)
--. 1 - = 1 - 3w2 + JW
. (
3 - W 2)
G(Jw)
The inverse Nyquist plot is shown in Figure 7.24, together with the critical circle and the band of uncertainty.
It defines a complete intersection with w1 =
1.699, W2 == 1.76, 01 == 0.139, and a2 =
0.1535. Thus, by
Theorem 7.4 we conclude that there is an oscillation of frequency w E (1.699, l.76) and amplitude of the
first harmonic a E (0.139, 0.1535).
(c) Figure 7.25 shows the inverse Nyquist plot with three critical circle corresponding to k = 4, 7.5, 10. With
-125
k = 4, the inverse Nyquist plot lies outside the critical circle. It follows from the first part of Theorem 7.4
that there is no oscillation. This however is a conservative estimate because we can use the second part of
the theorem to conclude that there is no oscillation if the inverse Nyquist plot enters the critical circle but
for each point inside the critical circle the error circle does not intersect the real axis. The critical circle with
k = 7.5 captures this situation. Therefore, we conclude that the largest slope for which the system does not
oscillate is about 7.5.
5 . ... . ..- . . ·;..· :.::- ·--.- ·- ·~

:
·;,: . ·:·
.... ~
4 . .... ; ......,i~. ..
: : .,,
3 .. .
t:
1..' .,. • •• • ,j ,
I :
2 / ·· :
I
I
... :.. . .. . ,: .... . .'. ..... . -----1 -·
I
..... .,:..
/'.
I :
--2 ·\··. . ·: · · · · · ·:· · ·· · · ·;· •· . ••. . ·• • . • • • . ••• .. . . , .. I.
' :,
-3 ... . ~s.--.
; '. I
·· --:,( ·
: ' : I ;
'-~,---
: ........
... ,:. .. ....... . ... ~~ ... ; ..
·· · -:--.
: ~, : .
-5 .. .. . ....... , . - -.~~~ ·--l----~ -~- -~~
-ti) -f ... -7 -t ,...,_
-II -4 -3 -.2 _, 0
• 'T.l 7 (I)
. 2(jw - 1) -2j(l - jw)2 · -4w - 2j(l - w 2 )
G(jw) = (jw)B(jw + 1) = = w3 (1 + w2 ) w3 (1 +w2 }
Im[G(jw)] = O~ w = 1
=
R.e[G(j)) = -2 and the equation 1+1li'(a)Re[G) 0 yields '1(a) = ½- The describing function of the saturation
nonlinearity is given in Example 7.7. The equation ~(a) = ½has a unique solution a 2.47. Thus, we =
expect the system to have a periodic solution with amplitude close to 2.47 and frequency close to 1 ran/sec.
To confirm this conjecture, we apply Theorem 7.4. The uncertainty band defines a complete intersection.
The sketch in Figure 7.26 shows the details of the complete intersection. The two circles (almost) tangent
to the real axis have w1 = 0.948, i'(ai) = 0.4255 and W2 = 1.032, ~(a:,) = 0.5494. From the describing
function of the saturation nonlinearity, we find that ~(a)= 0.4255 gives a= 2.933 and 1li'(a) 0.5494 gives =
=
a 2.238. Thus, Theorem 7.4 confirms that there is a half-wave symmetric periodic solution with frequency
in the range [0.948, 1.032] and the amplitude of the first harmonic is in the range [2.238, 2.933].
(2) From Example 7.14, we know that the harmonic balance equation has a solution with w 2"'2 and =
=
'11(0) 0.8. The inverse Nyquist plot is shown in Figure 7.21 of the text. For the nonlinearity ,p(y) = ½sin 11,
the slope restriction is satisfied with a = -½
and /3 = ½. Therefore, the critical circle is centered at the
origin and has a radius of 0.5. Neither the inverse Nyquist plot nor the uncertainty band enter the critical
circle. Hence, by the first part of Theorem 7.4, we conclude that there is no oscillation.
(3) From Example 7.14, we know that the harmonic balance equation has a solution with w 2"'2 and =
ll(a) = 0.8. The inverse Nyquist plot is shown in Figure 7.21 of the text. The nonlinearity satisfies the slope
restriction with a = -1 and /J = 1. Therefore, the critical circle is centered at the origin and has a radius
- 126
.,
-~·~;:;,·~w·--r·-.·t ·:..·:.:; . .
.. ... ... , .
,, ,
;~..
:
'
.
. -
·.· .. ...:-. .... ....
- ~ - ...: '' :
.,,,,. -:-
3 .. .. . /... .. ....:..... ; ..: -~ .. . .. --~ ., .,~'i .
,': /' : - \ \
I . ~ '\ \
2 i ·/ .. ....... ·•·· -;_., - , "\'1/G(JOII
..
J1 "
~
' ~ .
...,.. .... ·:: , ... ., .... ..........
; " . '\I: .
: ' .
,
..
t: .... ( ... ,
~ : .
: .... .cU1.&.'. ..... .: .... . : .... /t;
: \ ;
"'_S ... .-:... ... ~~-. I,~_]
. _____ _
If :
I ·:·
. ...
: ', : j : 1:1 .
.. ~:., .. ' .·i·...."L:-,.-".'/ ·
: '' . : " ... :. . - -~- - ~ - ""'! , , !
.,-
.. . . .. .... . · -'··· · · · ·· ..... . . ·., . . , . .•. . .. . .. :•.... . . . ...-: . .... ., . . • . , ! - · ·
~
-5 . .:. . ~-~"" -- -:- --~-":-~(.
-10 _. -I -7 -4 ~
._.,... ~ -3 -2 -1 0
J o.05 m=0.948
i 01---~--,.--=-~-------i
tE -0.05
1/G(j co)
-0.6 -0.5 -0.4 -0.3
Real/oos
of 1. The describing function is given by
IJ(a) = I1
(¼) sin- 1 (!) + !;~ (2/a)2 -1

~ [sin- 1 (¾) - sin-1 (¾)] J1- (2/a) 2 - !! Jl - (3/o.)2 + 1
+: 0
It is shown iD Figure 7.27. The equation •(a) = 0.8 has two roots at a = 2.48 and a = 12.47. Thus, we
expect that there are two periodic solutions, one with amplitude close to 2.48 and frequency close to 2/2
raA/sec. and the other with amplitude close to 12.47 and frequency close to 2\/'2 rail/sec. Next, we apply
Theorem 7.4. It is clear from Figure 7.27 that •'(a) is different than zero at the roots a•. Thus, all the
conditions of Theorem 7.4 (case 3) are satisfied and we confirm the existence of the periodic solutions. By
calculating and plotting the uncertainty band, it can be seen that the boundaries of the band intersect the
real axis a.t -0.6481 and -0.9519. The frequencies of the error circles closest to the real axis a.re w = 1.75
- 127
and w = 2.905. Using the graph of ~(a), the set f corresponding to each solution is given by
I'1 = {(w, a) I 2.75 :5 w $ 2.905, 2.1606 $a$ 2.803}
r2 = {(w , a) j 2.75 :$ w :5 2.905, 6.7065 :5 a~ 52.8815}
0.9
0.8
'ii'
~0.7
a..
0.6
0.5
0.4
0 5 10 15
a
(4) The harmonic balance equation has the solution w = J3 and 'i(a) =
1. For the given nonlinearity,
~(a) = 1 yields the unique solution a= 0.8165. Application of Theorem 7.4 succeeds and gives
I'= {(~,a) I 1.711 $ w $ 1.749, 0.7746 $ a S 0.8485}
- 128
Chapter 8
• 8.1 As in the proof of Theorem 8.1, take .v(y, w) = V(y) + v'wT Pw. Then
11 $ : [A1Y + Y1 (y, h(y))] - 4y'A~=(P) llwll2 $ 0
Hence, the origin is stable.
• 8.2 From Theorem 8.2, we know that asymptotic stability of the origin of the reduced system implies
a.,ymptotic stability of the origin of the full system. The opposite statement can be proved by contradiction.
Suppose the origin of the full system is asymptotically stable but the origin of the reduced system is not so.
Asymptotic stability of the origin of the full system implies its stability, which in turn implies stability of
the origin of the reduced system. H the origin of the reduced system is stable but not asymptotically stable,
then there is a bounded solution y(t) (with lly(O)I] arbitrarily small) which does not converge to zero as t
tends to infinity. Then, the corresponding solution (y{t), 0) of the full system does not converge to zero,
which contradicts asymptotic stability of the origin of the full system.
= = =
• 8.3 If 92(l', 0) 0, then h(y) 0 satisfies the PDE (8.8). If 91(y,0) 0 and A1 0, the reduced system =
is fl === 0. The origin of the reduced system is stable and the Lyapunov function V(y) = yT'ti satisfies the
condition of Corollary 8.1. Thus, the origin of the full system is stable.
• 8.4 .In Example 8.1, A1 = 0 and a= 0 yields 91(11,z) = -b(yz + z2 ) and 92(1,1,z) = b(yz + z2 ) . Thus,
=
91 (11, 0) 92(11, O) = 0 and the conclusion follows from the previous exercise.
• 8.5
(a) The function /1,(xe,z1,) can be expanded as
/1>(Xa,Zb) = -88/b
Zo
8/b -
(O,O)za + -8 (O,O)zb + /b(x
XI,
11 ,x1,)
8/b
= ~(0,0)xe
vXo
-
+ /•(:ta,Zb)
where j,,(xe, z1,) vanishes at (0, 0) together with its first partial derivatives. Since /1t(z0 , 0) = 0 in the
neighborhood of Xe= 0, ~(0,0) = O. Let x = [ Xa ] and /(x)
• X1,
= [ A1,Xbla+(x;,(z,,)
JII Xa, Zit
) ] . Then
of [ ~(0,0) ~(0,0) ] [ ~(0,0) ~(O,O)]

-{O) = =
Bx ~{O, 0) A&+ ~(0,0) 0 A•
H the origin of Xe = f11(z0,0) is exponentially stable, then (from the Theorem 4.15) the matrix l!!(0,0)
is Hurwitz. Consequently, the matrix ~(O) is Hurwitz and the origin z =
0 is an exponentially stable
equilibrium point of the full system.
- 129
(b) In this case, the matrix M!,

(0, 0) has some eigenvalues with zero real parts, with the remaining eigenval-
ues (if any) having negative real parts. By a linear change of coOl"dinates, the full system can be transformed
into the form
~1 = F1111+Q1('11,"12,x11)
'72 = F2112+G2x6+fb('J1,"12,Z11)
X6 = AbX6 + /11('11, '12, z11)
with F 2 having all eigenvalues with negative real parts and F 1 having all eigenvalues with zero real parts.
Moreover, the functions Q1 and 'h vanish at (0,0,0) together with their first partial derivatives. By assum1r
=
tion, the origin (1/1, 112) (0, 0) is an asymptotically stable equilibrium point of the system
Let f1'J =1r:i'11) be a center manifold for this system. Then, 1r2 satisfies the PDE
lhri
lhll [F1'71 + gi('71, 1r2(1,i}, O)] = F11r2('11) + '12('71, '11"2(171 ), 0)
From Corollary 8.2, we conclude that the origin '71 =0 is Bll uymptotically stable equilibrium point of the
reduced system
'11 = F1'11 +q1('11,'ll"2(1J1),0)
Consider now the full system. The center manifold for the fuD system is the pair
that satisfies the PDE's

8h-J
8rJi [F1'11 + 91('11, ~('11),h1 ('11))) = F2"2('11) + G2h1('11} + 112('11,h2('11), h1 ('11)) .
8h1
8rti [F1'11 + 91('11,hi('h), h1{'11))] = A11h1("1) +·'6(111,"2("1),h1('11))
It can be verified that these equations are solved by h2('11) = ,r2('11) and h 1('1t) = O. Thus, the reduced
system (of the full problem) is
'11 = F1'11 +q1('11,1r2('h),O)
Since the origin of this system is asymptotically stable, we conclude from Theorem 8.2 that the origin of the
full system is asymptotically stable.
• 8.6 (1) A= UL=0 = [ g !1 ] · Take 11 =z1 and z = z2.

'ii= -z2 , z = -z + y2 + yz
N(h(y)) = h'(y)[-h:t(y)] + h(y) - 11" -yh(y) = 0, h(O) = 0, h'{O) =O
=
Try h(y) 0(1111 2 ). It yields the reduced system 'fl = -y2 0(1yj2 ) . We cannot reach any conclusion using
=
this equation. Next, try h(y) h-J1l + O(IYl 3 ). Substituting this expression in the center manifold equation
and matching the coefficients of ,r,
we obtain h2 = 1. The reduced system is
'JI= -i,4 + 0(11,115 )
- 130
The origin of t.his system is unstable. Hence, the origin of the full system is unstable.
(2) A = ~L=o = [ ~ ~I ] . Take y = x1 and z = x 2 •
y = ay2 - z 2, i = -z + y 2 + yz
N(h(y)) = h'(y)[ay 2 - h 2 (y)] + h(y) - y 2 - yh(y) = 0, h(0) = 0, h'(0) ""O

Try h(y) = O(fyj 2 ) . It yields the reduced system .
ii = ay2 + O(lyl 4 ), a # 0
The origiD of this system is unstable. Hence, the origin of the full system is unstable.
(3)
A= 8/1
OX :t:=O
=[ ~
O
~l
Q
~
-1
]
Take y = f x1 x2 JT and z = x3.
'l/1 = -112 + Y1Z, iJ2 = Y1 + Y2Z , i = -z - (yf + 11?) + z 2

oh ah &h oh
= &y 1 (-Y2+Y1h(y)J+ ~ [111 +112h(y)]+h(y)+y~+y~-h (y) = 0, h(O) = 0, {}y1 (O) = 0, fJv2 (0) = 0
N(h(y)) 2
Trying h(y) = O(llyll2 ) we reached no conclusion. So we take
h(y) = ¢,(y) + O(llyll 3), where 4>(y) = YT Py= PnY~ + 2P:12!11Y2 + P22Yi
84> 8¢,
ay1 = 2P11Y1 + 2P12Y2, 8y2 = 2P12Y1 + 2P'l2Y2
Substituting these expressions in the center manifold equation and matching the coefficients of i, y1y2, and
y~, we obtain
Pu + 1 + 2p12 = 0, -2pu + 2J>J2 + 2P12 = 0, -2P12 + 1>22 + 1 = 0
whose solution is Pu= P22 = -1 and P12 = 0. Hence we have h(y): -(yf +Yi)+ 0(111111 3) and the reduced
system is
Try V(y) = ½(11? + yJ) as a Lyapunov function candidate.
Hence V(y) is negative definite in some neighborhood of the origin and the origin is asymptotically stable.
Thus, the origin of the full system is asymptotically stable.
(4) A= ~L=o = [ 6 ~l ] · Take y = x1 and z == x 2 •
.N(h(y)) = -h'(y)y 2 h(y) - h(y) - y 3 = 0, h(O) = 0, h'(O) = 0

=
Try h(y) O(Jyl 2 ). It yields the reduced system y = y 2 O(lyl 2 ). We cannot reach any conclusion using this
=
equation. Next, try h(y) h2y 2 + O(ly/ 3 ). Substituting this expression in the center manifold equation and
matching the coefficients of y 2 , we obtain h2 =
0. Try h(y) = h3y3 + O(lyl4 ). Substituting this expression
- 131
in the center manifold equation and matching the coefficients of y3 , we obtain h3 = -1. The origin of the
reduced system
'JI = -y5 + O(IYl 6 )
is asymptotically stable. Hence, the origin of the full system is asymptotically stable.
(5) A= UL=o = [ ~ ~l ]- Take y = xi and z = x 2.
y=yz3 , i=-z-y 2 +2y8
N(h(y)) = h'(y)[yh3 (y)] + h(y) + y2 - 2y8 = 0, h(O) = 0, h'(O) =O

It can be seen that h(y) =
-y2 satisfies this PDE. The origin of t4e reduced system 'fl = -y7 is asymptotically
stable. Hence, the origin of the full system is asymptotically stable.
(6) A= UL=o = [ ~I ~ ] - Take y = x2 and z = x 1 .
fl= y 3 (z + y - 1), z = -z -y2 + y 3 (z + y - 1)
N(h(y)) = h (y)[y3 (h(y} + y -

1 l)J + h(y) - y 3 (h(y) +y- 1) = 0, h(O) = 0, h'(O) =O
Try ¢,(y) = O. Then h(y) = 0(1111 3 ) and the reduced system is ii= -y3 + O(IYl'}: Tbe origin of this system
is asymptotically stable. Hence, the origin of the full system is asymptotically stable.
(7) A = UL:i:0 = [ ~ !1 ] · Take y = x1 + x2 and z = x2.
. (y - z)3 z = -z + a (y - z) 3
y = a 1 + (y- z)2' l+(y-z)2
(1/ - h(y)) 3 ] (y - h(y)) 3

•I
1v (h(y)) =h (y)
I [
a 1 + (y _ h(y)) 2 + h(y} - a 1 + (y _ h(y)) 2 = 0, h(O) =0, h'(O) = 0
Try <J,(y) =0. Then h(y) = 0(1111 3) and the reduced system is
3
'ii= a -1 y 2 + O(IYl4 )
+v
= a1,i3 + O(IYl 4 )
The origin of the full system is asymptotically stable if a < 0 and unstable if a > O.
(8)
A- - 8/1 -
- 8x 2:=o -
[-2 -3 ~I]I
0 0
1
Take y = X3 and z = [
X1 +
X3 ]
.
· X2 -X3
N1 (h(y)) = h~ (y)[h1 (y) - y]2 + 2h1 (y) + 3h2(Y) - y2 - [h1(y) - y) 2

N 2(h(y)) = '4(y)[h1 (y) - y) 2 - h1 (y) - h2(y)
~(O) = 0, h~(0) = 0, i = 1, 2
Try ¢(y) = o.
N1(0) = 0(1111 2 ), N2(0) =o
The origin of the reduced system iJ = y 2 +O(lyl 3 ) is unstable. Hence, the origin of the full system is unstable.
- 132
• 8.7 A= lf L=o = [ ~ ~I ] . Take y = x1 and z = x2.

ii= yz + ay3 + byz2 , z = -z + cy2 + d·y2z
N(h(y)) = h (y)[yh(y) + ay3 + byh2(y)J + h(y} -
1 cy 2 -d1/h(y) = 0, h(O) = 0, h'(O) =0
We reach no conclusion by trying h(y) = =
O(lyl 2 ) . So we take h(y) h 2 y2 + O(lyl 3 ). Substitution in the
center manifold equation and matching of the coefficients of y 2 yield h2 = c. The reduced system is
ij =(a+ c)y 3 + O(IYl 4 )
a +c > 0 =* The origin is unstable

a+c < 0 ==> The origin is asymptotically stable
When a+ c = 0 we can reach no conclusion. we seek a higher-order approximation of h(y). We take
=
h(y) cy2 + h 3 y 3 + O(lyl 4 }. Substitution in the center manifold equation and matching of the coefficients
of y3 yield h3 = 0. So we take h(y) =
cy2 + h4 y 4 + O(lyJ 5 ) and repeat the process to obtain h,. = cd by
matching the coefficients of y • The reduced system is
4
'ii= (a+ c)y3 + (cd + bc2)y5 + O(IYl6 ) = (cd + bc2)y5 + 0(1111 6 )

since a + c = O.
a + c = 0 and cd + bc2 > 0 '* The origin is unstable
a + c = 0 and cd + bc2 < 0 => The origin is asymptotically stable
If a+ c = cd + bc2 = 0, we cannot reach a conclusion and we seek a higher-order approximation of h.
Proceeding as before, we obtain
h(y} = cy2 + cdy4 + cd2y6 + O(lr,f)

The reduced ·system is
ii= (a+ c)y3 + (cd + bc2)ys + (cd2 + 2bc2d)111 + O(ly/s) = -cd.2y1 + 0(~1118)
a+c = cd + bc2 = 0 and c.d2 < 0 ::::> The origin is unstable
= c.d + bc2 = 0 and cd2 > 0 =} The origin is asymptotically stable
a+c
The only case left is a+ c = cd + bc2 = cd2 = 0. If c = 0, we have a = 0 and h(y) = 0 solves the center
manifold equation exactly, resulting in the reduced system fl = 0. H c ;/; 0, we must have b d O and = =
=
h(y) cy2 solves the center manifold equation exactly, resulting in the reduced system 'fl= 0. In either case,
the conditions of Corollary 8.1 are satisfied; hence, the origin of the full system is stable.
• 8.8 A= UL=o = [ ~ ~l ] · Take y = xi and z = x 2 •
ij = ay3 + y 2 z, i = - z + z 2 + zy - y3
.N(h(y)) = h'(y)[ay3 + y2 h(y)] + h(y) - h2 (y) - yh(y) + y3 = 0, h(O) = h'(O) = 0

'Iry h(y) = O(Jyl 2). The reduced system is
a > 0 => The origin is unstable
- 133
a < 0 => The origin is asymptotically stable

Ha= 0, we reach no conclusion. We take h(y) =
h 2 y 2 + O(ly13 ). Substituting h(y) in the center manifold
= =
equation and matching the coefficients of y3, we obtain h2 0. We take h{y) h3y 3 + O(IYI') and repeat
the process to obtain h 3 = -1. The reduced system is
The origin is asymptotically stable.
y=ayz-y3 , i=-z+byz+c:y2
N(h(y)) = h (11)[a11h(11) -113] + h(11) -

1 byh(y) - cy 2 = O, h(O) = 0, h'(O) =O
We reach no conclusion by trying h(y) = 0(1111 2 ). So we take h(y) = hzy2 + O(lyl3 ). Substitution in the
=
center manifold equation and matching of the coefficients of y 2 yield h2 c. The reduced system is
'ii= (ac - l)y3 + O(lyl')

ac - 1 > 0 => The origin is unstable
ac - 1 < 0 => The origin is asymptotically stable
When ac - 1 = 0 we cannot reach a conclusion. We seek a higher-order approximation of h(t,). We take
=
h(y) cy2 + h 3 y 3 + O(lyl 4 ). Substitution in the center manifold equation and matching of the coefficients
=
of y3 yield h 3 be. The reduced system is
=
H b ¢ O, the origin is unstable. The only case left is the case ac = 1 and b 0. In this case, the center
manifold equation is solved exactly by h(y) = cy2 • The reduced system is ii = 0. Its origin is stable and the
Lyapunov function V(y) = 'If satisfies the condition of Corollary 8.1. Hence, the origin of the full system is
stable.
• 8.10 In the neighborhood of the origin, V(.:z:) is positive definite and V(x) is negative definite. Hence, the
origin is asymptotically stable. To show that G is exactly the region of attraction, we need to show that
.:z:(O) E G => x(t) -+ 0 as t -4 oo and x(O) tG => x(t) f+ 0 as t -+ oo

Let x(O) € G. Then V(x(O)) < 1. Choose c > 0 such that V(x(O)) :;:'; c < l. Consider the set n {x E =
G I V(x) ~ c}. n is in the interior of G. Moreover, n is bounded since limnzll--+oo V(x) 1 and c < 1. n is =
positively invariant since V(x) is negative definite. For x E G, we have
V(x) = 0 => h(x)[l - V(x)] = 0 => h(.:z:) = 0 => x = 0

where we have used the fact that 1 - V(x) > 0 for all x E G. Thus, by LaSalle's theorem, x(t) -4 0 as
t -t oo. Now, let :z;(O) ¢ G and suppose x(t} -+ 0 as t -+ oo. Then x(t) must enter the set G. Hence, there
are finite times t 1 and t2 > t1 such that .:z:(ti) e &G and x(t) E G for all t E (t1 ,t2]. Let W{.:z:) = 1- V(.:z:).
W(x) = -V(x) = h(x)(l - V(x)) = h(x)W(x)
-134
1wdW = lt
Wo W to
h(x(s)) ds
W(x(t)) = W( x {t )) e
0
J: h(z(•}) da
10
1 - V(x(to)) = [1 - V(x(t))]e - J,: h(z(a)) da

Take t = t 2 and let to -t t1.
lim L.H.S. = lim [l - V(x(t0 ))] =0 and lim R .H.S. >0 Contra.diction
to~t,
Thus if x(O) ,t. G, x(t) f+ 0.

• 8.11 Let
Using the hint, (8.17) takes the form
( ~ :1 + 91(x1)W1(x1)) W2(x2)h(xi} + ( ~:2 Wi(xi)g1(xi) - ~:1 W2(x2)92(x2)) =0

The equation is satisfied by W1 and W2 which satisfy
8W1 8W2
~
vX1
= -g1(xi)W1(x1), and ~
vX2
= -.92{x2)W2(x2)
Thus, equation (8.17} is satisfied with
V(x) = 1- exp (- foz' B1(a) oo - fo;i: g2(a) da)

2
The function V(x) has the properties: V(O) = 0, 0 < V(x) < 1 for all x E G, V(x) -+ 1 as x-+ lJG.
V(x) = -01(x1)h(x1)[l - V(x)J :S 0
All the conditions of Zubov's theorem are satisfied except that V(x) on only negative semidefinite.
V=O => 91(xi)h(xi}=O => x1 =0
=
X1(t) 0 => 92(x2(t)) := 0 => X2(t ) =0
Hence, by application of LaSalle's theorem (to the set G) and Zubov's theorem, we conclude that G is the
region of attraction.
• 8.12 The system is a special case of the one treated in the previous exercise with h 1 (x1 ) x 1 , g 1 (xi) = =
tanx1, a.nd .92(x2):..: x2. It can be verified that all the conditions of the previous exercise are satisfied in the
set G = { x E R2 I lx1 I < f} . It follows from the previous exercise that G is the region of attraction.
• 8.13 Since n is positively invariant and every trajectory inn approaches the origin as t-+ oo, every point
in n is connected to the origin by an arc. Consequently, any two points in O are connected by an arc that
is formed by connecting the two arcs that connect them to the origin. This shows that O is connected.
• 8.14 Let c = min::EIJD V(x). Then Oc = {V(x) < c} is an estimate of the region of attraction. We have
min { Xi + xn = 1, min {Xi + xn ::: }
:z:3-l;O$:z:1 $2 z2=-l ;-2S:i:1 ;SO
and
min
z1-z2=-1;-2$z1$0
{xf + xn = min {x~
-2::S:z1SO
+ (x1 + 1)2} = !2
Hence c = ½and the region of attraction is estimated by {x E R2 I xf + x~ < ½}.
- 135
• 8.15
(a) Using V{x) = 5x? + 2x1x2 + 2xt we have
V(x) = (IOx1 + 2x2)x2 + (2x1 + 4x2}[-x1 - X2 - (2x2 + X1 )(1- xrn
= -2x~ + 4x1x2 - 2x~ - 2(x1 + 2.'l:2)2(1 - x~)
For lx2I :$ 1, we have . 2

V(x) $ -2(x1 - 2x1x2 + x 22 ) = 2
-2(x1 - x2) ::; 0
V(x) =0 => x1(t)- x2(t) =O ~ x1(t) - ±2(t) =0 => 3x2(t)(2 - x~(t)) =O ~ x 2(t) =O
Hence, by La.Salle's Theorem (Corollary 4.1), we conclude that the origin is asymptotically stable.
(b) Notice that V(x) $ 0 for all x E S. To show that S is an estimate of the region of attraction we
need to show that S is positively invariant. The boundary of S consists of four pieces, as shown in Fig-
ure 8.1. The boundaries BC and DA are parts of the Lyapunov surface V(x) = 5. Since V(x) :5 0, the
trajectories cannot leave S through BC or DA. To show a similar property for the boundaries AB and CD,
notice that
It can be easily seen that the right-hand side is nonpositive on the boundaries AB and CD. Thus, the
trajectories cannot leave S through AB or CD. Hence, S is positively invariant. It is also compact and
V(x) $ 0 in S . Thus, by LaSalle's theorem, all trajectories starting in S must approach the largest invariant
=
set in {x E S I V(x) O}. It is clear from part (a) that the largest invariant set is the origin. Hence, S is
an estimate of the region of attraction.
1.5
0.5 V(x)=
0 t---,,__--+-_ _,,__---1
x,
-0.5 V(x) • 5
-1
D C
-1.5
-1 0
• 8.16
±1 = x2 , ±2 = -x2 - (x1 - xD
Consider the Lyapunov function candidate
V(x) = ½x~ + foz 1

(y - y 3 ) dy = ½x~ + ½xi - ¼x1
V(x) is positive definite in the region lxil < ./2.
•
V(x) = (xi -x13 )x2 +.z2{-.z2 -{x1 -x13 )] = -x22
- 136
Moreover
V(x) = 0 => x2(t) =0 => x1{t) - x~(t) =0 => x1(t) for lxd < 1
Hence, all the conditions of Corollary 4.1 are satisfied in the domain D = {lx1 1 < l}. We conclude that
the origin is ~mptotically stable and Oc = {x E R 2 I V(x) $ c} is an estimate of the region of attraction,
provided c is chosen small enough to ensure that !le is compact and Oc C D . The condition Oc C D is
=
satisfied by choosing c < ¼- For any O < c < ¼, the surface V(x) c is closed. Hence, Oc with O < c < ¼is
an estimate of the region of attraction. Equivalently, we can estimate the region of attraction by the open
set {x E R 2 I V(x) < ¼}.
• 8.17 Since V(x) is negative, the vector field must point to the inside of the surface V(x) = c. Therefore,
the directions (2} and (3) are possible, while {l) and (4) are impossible.
• 8.18 (a)
Hence
sinx1 +2sat(xi) =O => x1 =0
The origin is the unique equilibrium point.
(b}
The eigenvalues of A are -3/2 ± j,,/3/2. Hence, A is Hurwitz and the origin is asymptotically stable.
(c)
11i:r = er[- smx1 - 2sat(u)] :S 10'1 - 21111 =
-lul, for 10'1 ~ 1
(d)
V(x) = (2:c1 + sinxi)x2 + x2[-x2 - sinx1 - 2sat(x1 + x2)]

= -3x~ $ 0, for lul $ 1
The set Mc is closed and bounded. Its boundary is formed of four parts, two lying on the surface V(x) c, =
one lying on the line a = 1, and one lying on the line a =
-1. Since V(x) $ 0 ill Mc, trajectories cannot
leave the set through the parts of the boundary on the surface V(x) = c. Using the result of part (c), we see
=
that the trajectories cannot leave Mc through the lines O' ±1 since on those lines lul must be decreasing.
Hence, every trajectory starting in Mc will remain in Mc for all future time. Let us find the largest invariant
set in E = {x E Mc I V(:r) = O}.
V(x) = 0:::,. x2(t) =0 => sinx1 (t) - 2x1 (t) = =>0 X1 (t) =0
Hence, the origin is the largest invariant set in E. We conclude, by LaSalle's theorem, that every trajectory
starting in Mc approaches the origin as t ~ co.
(e) Since qif :S -lt1I for 10'1 2: 1, every trajectory starting outside the region 10'1 $ 1 must reach this region
in finite time. Suppose, for example, that the initial point is in the region a > 1. Then
air :S -11 ;:> u$ -1
Therefore u(t) :S a(O) - t, which implies that the trajectory reaches the region luf $ 1 in time less than
or equal to u(O) - 1. A similar argument can be made if the trajectory starts in t1 < -1. Once inside the
region 1111 $ 1, the trajectory belongs to a set Mc for some c > 0. It follows from part (d) that the trajectory
reaches the origin as t ~ co. Therefore, the origin is globally asymptotically stable.
- 137
• 8.19 (a)
Equilibrium points:
1
O=x2, 0 = -[-'73
T
COSX1 - 712x3 + EFD]
Substituting x 3 from the third equation into the second one, we obtain
0.815 = 2\(l.7cosx1 + 1.22) sinx1

Using "fzero" of MATLAB, we found two roots in the region -1f'::; x 1 $ 1r. The corresponding equilibrium
points are:
P = [ o.4g67 ] • q = [ 1.6i98 ]
1.301 0.4085
!f = [-('11/Mfx3
x
COSX1
-(7':J/r) sin x 1
1
-D /M
0
-(111/;) sinx1
-(772/'r)
]
81 1
8x :r.=p
= [-128~7204
-0.1019
:4 0
-53~821]
-0.4091
=> Eigenvalues= -2.0215±jll.171, -0.366
::;> p is asymptotically stable
Bf
ax
I
:r.:::q
= [ 3.~29
-0.257
_\
0
-135~7311 ]
-0.4091
::;> Eigenvalues= -3.4381 ± jl.7179, 2.4671 => q is unstable
(b) Estimate the region of attraction of p. Use V(x) = yT Piy, where y = x - p and Pi is the solution of
=
the Lyapunov equation P1A + AP1 -_!. Using MATLAB, Pi was found to be
16.039 -0.0013 6.6042 ]

Pi = [ -0.0013 0.1247 -0.0239
6.6042 -0.0239 4.3642
The eigenvalues of Pi are 19.0159, 1.3877, 0.1243. Set 9(x) = /(x) - A (x - p). We have
g1(x) = 0
gi(x) = (771/M)[-x3 sinx1 + .P3 sinp1 + Ps(z1 - p1) COSPt + (x3 - P3) sin PI)
93(x) == (113/r)[cosx1 - COSPl + (x1 - p1) sin pi)
Using the mean value theorem, we obtain
We have
- 138
P11
Pig = [ P12 P22
Pia
P12
P23
Pt3
P23
1>33
l[ l [ 0
92
93
= Pt2Y2
P:2292
P32U2
+ Pt393
+ P2393
+ P3s9a
l
l(Pig)1I < (111 IP121/M)(2IY1 I IY31 + p3yf) + (r,3IPt3l/r)yf 1;f a1y~ + bily1 I IYJI
l(P19hl $ (1111P221/M)(2IY1I lval + P3YD + (11:ilP23i/T)y~ ~r a2y~ + ~l11d IYJI
l(Pig)3I < (1111Plsl/M)(21111I IY31 + P3Yf) + (11311'331/r)y~ ~f a3y~ + bslY1I IYsl
Hence
IIP1ull~ ~ ayt + 2bl11113 IYsl + dy~yJ =11l [ 1:~1 ] T N [ l!~i ]
where a= ai +a~+ a~, b = a1b1 + a2b-i + a3~, d = b~ + b~ + b~, and N = [ : : ] . Thus,
• 2 S
V::; -11111'2 + 2v').m,u:(N)l111ll2
=
Let r = 1/2./'Amru;(N) and c = r2 Amin(Pi) Amin(Pi)/4>.mGz(N). It can be verified that c = 1.8175x 10-5 _
The set {V (y) < c} is an estimate of the region of attraction.
• 8.20
(a) Take V(z) = ½(x~ + x~). Then
Let
A(t) = [ ~l -:(t) ] , A= [ ~l ~], C(t) = [ O y'g(ij ]
Using g(t) 2;: k1 , it can be shown by calculating the observability Gra.mian that the pair (.A, C(t)) is uniformly
=
observable. Since A(t) A-CT(t)C(t} and C(t) is uniformly bounded, we conclude that the pair (A(t), C(t))
is uniformly observable. It can be easily verified that P = ½I satisfies the equation
-P(t) = P(t)A(t) + AT(t)P(t) + CT(t)C(t)

It follows from Theorem 8.5 and Example 8.11 that the origin is exponentially stable.
(b) With g(t} = 2 +et, the state equation is
±1 = x2, :i;i = -x1 - (2 + et)x2

· It can be verified that
x1 (t) = -(1 + e-t)k, x2(t) = e-tk
is a solution for any constant k. Thus
xi(t) --+ -k and :r2(t) --+ 0 as t --+ oo
This shows that the origin is not asymptotically stable. Hence, the boundedness of g(t) is necessary
• 8.21 Linearize the system at :r = 0. The linearized system is studied in the previous exercise. The origin
of the linearization is exponentially stable. Thus, by Theorem 4.13, the origin is an exponentially stable
equilibrium point for the nonlinear system.
- 139
• 8.22
x-
-
[
-1
1
a(t)
-1
0
0
-o{t)
0
0
l x = A(t)x
With V(x) = xTx, we have
V(x) = xT[A(t) + AT(t)Jx = -xf = -x7 CTCx
where C = (1,0,0J. Taking K(t) = [-1,1,a(t)]T yields
0 -~1 -a~(t) ]
A(t) - K(t)C =[ ~
The transition matrix of [A(t) - K(t)C] is
=[
I -(t - T) - J: a(u) cw ]
+(t, r) 0 1 0 =* C+(t,T) =[ I -(t-r) -J:o(u) du J
0 0 1
With o{t) = sin t + sin 2t, it can be seen that the pair (A - KC, C) is uniformly observable. Exponential
stability of the origin follows from Example 8.11.
• 8.23 (a)
ei = Xi - r(i) = Xi+l - r(i) = eHl, 1 S i S n - I
en = :i,.-r(n.) = fo(x)+(i9*)Tfi(x)+9o(x)u-r<n)
Hence, e satisfies the equation
e = Ae + B[/o(x) + (8•) T Ii (x) + 9o(x)u - r(n.)]
where (A, B) is a controllable canonical pair that represents a chain of n integrators.
(b)
V = 2eTPJ+2<17r- 1¢
= 2eTP{(A- BK)e + B[fo(x) + (e·ffi(x) + 9o(x)u- r(n) - Ke]}+ 2,pTr- 1 rJ1(x)eTPB
= e7 [P(A- BK)+ (A- BKfP)e + 2e7 PB[-¢,T fi(x)J + 2¢7 ft(x)eTPB = -eTe
By Theorem 8.4, all state variables are bounded and limt-+oo e(t) =O.
(c) The closed-loop system can be represented as the linear time-varying system
X = [A.(t) + B(t)JX. e{t) = CX
where
A(t) = [ r!(~)!fp -B~T(t)] ,B(t) = [ r(w(t) -!(t))BTP -B(w(t)o-w(t))T ] X = [:]

C =[ I O ] , w(t) =
fi(x(t)), and w(t) = fi(1l(t)). We want to show that the origin is exponentially
=
stable. Since limHoo e(t) 0, it is sufficient to prove exponential sta.bility of the system (see Example 9.6)
x = .A(t)X, e(t) = ex
It follows from Theorem 8.5 and Example 8.11 that the origin of the closed~loop system will be exponentially
stable if the pair {A(t), C) is uniformly observable. Setting A::(t) = [ r!;)!fp ] , It can be seen that
A(t) = A(t) - K:(t)C. Since uniform observability of (A(t), C) is equivalent to uniform observability of
(A(t} - K(t)C, C) for any piecewise continuous bounded K(t}, we conclude that the origin of the closed-loop
system will be exponentially stable if (A(t), C) is uniformly observable. In this case, limt-+oo ¢(t) O. =
- 140
Chapter 9
• 9.1 (a) If P and Q satisfy the Lyapunov equation, then kP and kQ satisfy the same equation for any
positive constant k. Thus
µ(kQ) = Amin(kQ) = Am,n(Q) = µ(Q)
Ama:,:(kP} Amaz{P)
(b) let
Pi =L°" exp(ATt) exp(At) dt, P2 = fo 00
exp(ATt)Qexp(At) dt
Then
Pi - P2 = fo 00
exp(ATt)[I - Q] exp(At) dt
Q~ I ~I - Q ::5 0 ~ Pi - P2 $ 0 ~ P2 ~ P1
Hence, >-maz ( P2) ~ Ama.z (Pi) and
µ(.~) = Amin(!) X AmGZ(~2) > l

µ(Q) Amoz(Pi) ,\min(Q) -
(c) For any Q =QT> 0, let k = 1/>.min(Q). Define Q = kQ so tha.t ;\m,n(Q);::: 1. Thus
µ(I) ~ µ(Q) ~ µ(Q), "r/ Q = QT > 0
• 9.2 (a) The function l~(v)I is sketched in Figure 9.1. From the sketch it is clear that
6
jh.(v)I ::5 1 + c,lvil, "r/ /vii$ L(l + 6)
(b} The derivative of V(x) = xTPx along the trajectories of the system satisfies
V(x) =-xTx-2xTPBh(Fx) ::5 -!lxll~+2llxll2IIPBll2llh(Fx)H2
o 26
< -llxll~ + 2llxll2IIPBll2 1 + 0 IIFxll2 ::5 -JI.xii~+ 1 + 0 IIPBll2IIFll2llxll~
6 1
< O, for 1 + cS < 2IIPBll2IIFll2
(c) The origin is asymptotically stable by Theorem 4.1. To estimate the region of attraction, find a constant
c > 0 such that the set flc = {x E R" I xT Px ::5 c} is inside the region l(Fx)il ::5 L{l + 6) for all i . The
18l'gest such constant is obtained by minimizjng V(x) on the boundary of the region.
(d) We have
0
A- BF= [ -0.5 I ] => P
-1 =[ l.l75 1.5
1 J~ 2IIPBll2IIFll2 = 8.0623
- 141
lh.(v)I
I
6 1 1
l + 6 < 8.0623 =* a< 7.0623 = 0 ·1416
The region of interest is given by IFxl < 1.1426. Take c < (l.1426) 2 /[FP- 1 FT] = O. 724. The region of
attraction is estimated by {xT Px 5 0.723}.
• 9.3
. av av av
V(t, x) :::: 8t + ax f(t, x) - ax Bg(t,Cx) + c..Jlxll2IIBll2llg(t, Cx)ll2
:$ -c3llxll~
~ -c31/xll~ + c.tllxl12IIBlb1'1lCl'2llxll2 = -[ca - 1'C4IIBlhllCl'2Jllxl]~

For a.ll "'f < 7• = ca/(C4IIBll2IIOll2), the origin is globally exponentially stable.
• 9.4 The closed·loop system is given by
x = (A - BBT P)x + Bg(t,x)
Rewrite the Riccati equation as
P(A-BBTP) + (A-BBTPf P+ Q+PBBTP+2o:P= 0
Consider V{x) = xT Px as a Lyapunov function candidate.
V(t, x) = xT[P(A - BBTP) + (A - BBT P)TP)x + 2xT PBg(t, x)
= -xT[Q + PBBTP+ 2aP]x + 2xTPBg(t , x)
5 -k 2 llxll~ - llwJI~ - 2o:A.nin(P)llxlf~ + 2kllw1'211xl'2, where w = BTPx
= -[k1lxll2 - llwll2]2 - 2a>.min{P)llxlJ~ ~ -2a>.min(P)llxll~
Hence, the origin is globally exponentially stable.
• 9.5 The closed-loop system is given by
x = ( A- 2~BBTP) x + Dg(t,y)
Let V(x) = xTPx.
lf(z) = ,r
I
[p (A - !ssTp) + ( A - !BBTP
l
r P] x +2zTPDg(t,y)
= - -xTPDDTPx- -xTcTcx-exTQx+2xTPDg(t,y)
')' 1'
:5 - .!.xr PD DT Px -
1'
.!. IIYII~ -
1'
exT Qx + 2ll xT P Dlf2kllvll2
- 142
V(x) < - .!.llzlli - .!.IIYlli- txTQx+ 2k!lzl12IIYll2

1' 1
< - _!_llzll~ - _!_IIYII~ + k(llzll~ + IIYII~) - exTQx
1 1
< -exTQx
for 1 < 1/k. Hence, the origin is globally exponentially stable.
• 9.6 (a) The perturbed system can be written as
x• = Ax+ !lxll22 Bx, where A = [ -o -w ] , B __ [ /3 -1 ]

w -a 1 13
V(x) =
xTx is a Lya.punov function for the nominal system since A+ AT= -2ol. The derivative of V
along the trajectories of the perturbed system satisfies
For llxlJ2 ~ r, we have
(b) Calculating the derivative of V directly (without viewing llxll~Bx as a perturbation term), we obtain
V = -2ollxll~ + 2.Bllxll~
When /3 $ 0, V ~ -2allxll? and the origin is globally exponentially stable. When f3 > 0, we have
V ~ -2(a - /1r 2 )11xllt 'r/ llx!l2 ~ r
Hence, Vis negative definite when r 2 < o:/P. Since V(x) = llxll~, we conclude that the set {llxl/~ < o//3} is
included in the region of attraction.
(c) Clearly, the results of (b) are less conservative than those of (a).
• 9,7 f(x) can be written as f(x) =Ax+ /(x), where /(x) and its first-order partial derivatives vanish
at x == 0. Take g(x) == Bx so that the perturbed system is given by x = (A+ B)x + i(x). Since A has
eigenvalues with zero real parts, for any 1 > 0, there is B with IIBII ~ 1' such that at least one eigenvalue of
(A+ B) has a positive real part. Hence, the origin of the perturbed system is unstable.
• 9.8 (a) The derivative of V along the trajectories of the perturbed system is given by
. av av
V = ox f (x) + ox g(x) ~ -os{llxll) + a4(11xl!hll.xll
Let
a1 = min 03(/lxll) > 0 and a, = max /lxllo4(ll.xll) > 0
:z:E{V(:i:)=c} :iEfl
Then, for all x E {V(x} =c}, we have
-143
Thus, 0 is positively invariant for sufficiently small 1. Inside n, we have
V· $ -a3(11xll) + 1a2 $ -(1 - 0)a3{llxl1), 'v llxll ~ a 31 ( e

1a2)
where O < 8 < 1. The preceding inequality is valid as long as a 3 (11xll) $ -ya 2 /8 for all x E 0, which is the
case for sufficiently small "f· It follows from Theorem 4.18 that a.lJ solutions starting in n are ultimately
bounded by a class JC function of -y.
=
(b) Since A= ~(O) is Hurwitz, the origin of x f(x) is exponentially stable. By Theorem 4.14, there is
r 0 > 0 and a Lyapunov function V(x) such that
av
ax f(x) $ -c3llxll , 2
I ~: I $ C4llx!I
for all llxll $ r 0 , where c1 to c_. are positive constants. The derivative of V along the trajectories of th_,
perturbed system is given by
V
. = av av
8x f(x) + &x g(x) $ -(ca - 1ec)llxl12
Hence, there is Co > 0 (independ~t of -y) such that for all -y < c3/C., the set O = {V(z) $ eo} is positive
invariant and every trajectory in n converges to the origin as t tends to infinity. For sufficiently sma.ll -y,
the ultimate bound of part (a) will be small enough to ensure that every trajectory starting inn enters 0
in finite time. Hence, every trajectory starting in n converges to the origin as t tends to infinity.
(c) If A is not Hurwitz, we cannot ensure tha.t the origin of the perturbed system is exponentially stable.
Hence, the proof of part (b) falls apart; that is, we can show uniform ultimate boundedness but we cannot
show that the trajectories will converge to the origin. In fact, the example given in the problem statement
is a counter example where the trajectories do not converge to the origin for all x(O). For this example, we
have
A= -
bx
fJf I .= [o
z=D
1
O
-1
0
1
o]
0
0
which is not Hurwitz since a.lJ its eigenvalues are on the imaginary axis. On the other hand, the derivative of
V(x) = Xi+½x~+½xi+x1zs along the trajectories of the nominal system is given by V(x) = -(2x1+z3)4 $ O.
V(x} = 0 => 2x1(t} + xs(t) =O => 2xi(t} + x3(t) =0 => x 2 (t) =O
Hence, by LaSalle's theorem, the origin is asymptotically stable. It follows from Theorem 4.16 that there is
a Lyapunov function V(x) that satisfies the given inequalities in some domain a.round the origin. Finally,
the linearization of the perturbed system at the origin is given by
au+ g)
iJX
I :i:::=O
=[ ~
O
~1
1
~a ]
0
and its characteristic equation is .,\3 - a.A 2 +.A+ a= 0. By the Routh-Hurwitz criterion, it can be seen that
this matrix hM at least one eigenvalue in the right-half plane. Bence, the origin of the perturbed system is
unstable.
• 9.9 (a)
•
Xt = -Xl3 + X2,5 ·
X2 = -Xi3 - 5
X2
Linearization at the origin yields the matrix A = 0 which is not Hurwitz. Hence, the origin is not exponen-
=
tially stable. 'Let V(x) ¼xt + ¼x~.
-144
Hence, the origin is globally asymptotically stable.

(b) Linearization at the origin yields the matrix [ ~ ~'Y ] whose eigenvalues are 1/2 ± j-y../3/2. Hence,
the origin is unstable. The derivative of V along the trajectories of the perturbed system is given by
V. =
6 10 3
-Xi - X2
5 6
+ "fX1X2 + "fX2
- 'YX1X2
< -x6l _ x10

2
+ 1(x6
21 + x2)2
+ 1(x2
21 + x10)
2
+-yx62
= - ( 1 - 2'Y) X16 - ( 1- '2Y) X210 + 2'Y xl2 + 2"Y x22 + "fX26

$ -
3 6
4Xl -
3
4X2
10
+ ri:1
'Y 2 'Y
+ 2X2 +
2 6
'YX2
< - 41 (xt + x~0 ), for lx1I 2: -y114 and lx2I 2: [ ~]l/4

-y + y-y 2
+ 'Y
For lx1 I < -y 1 / 4 , we have
Let p1 ( 1') be the largest positive real root of the polynomial equation
1 ,y3/2 ..,
_ -ylO + _,_ + ..!.y2 + 'YY6
2 2 2
P1('Y) is a class JC function of-)· and P1b) 2: ['Y + .../72 +-yr;.,_ Hence, for lx1I < -yl/4 , we have
V ~ - ~(x~ + x~0 ), V lx2I 2: P1 (-y)
For Ix;!< [-r + y'-y2 + ,r 1', we have
. 3 6 3
V$-rc1- 4X2 +2.x1+2 "Y+v,2+,
10 'Y 2 i [ ~11/2 +-y [,+y72+-y
~]3/2
Let P2 (1 ) be the largest positive real root of the polynomial equation
1 i i [
- 2Y6 + 2Y2 + 2 'i' + V~ 11/2
-y2 + 1' + 1' ["Y + ~13/2
V ,2 + 'Y
pz(-y) is a class 'K, function of7 and Pzb) 2: -y1 / 4 _ Hence, for lx2I < [7 + J-y +..,.J114 , we have
2
. l 6
V $ - 4(:r1 + X210 ), V lxd 2: {)2(1')
settingµ.= max{P1h), pz('y)}, we see that the inequality V$ - ¼(xr + x!0 ) is satisfied for all llxlloc ~ µ.
The conclusion follows from Theorem 4.18.
• 9.10 (a) With q = 0, we have
V = -acx1 sin x1 - bcxf - ex~ ~ -(b- a)cxf - ex~
which is negative definite for all x, since b > a.
(b) With q # 0, we have
V = -acx1 sin X1 - bcxf - ex~ + (cx1 + 2x2)q(t) cos.xi S -(b - a)cxf - ex~+ kVc2 + 4 Uxll2
Hence, the solutions are globally uniformly ultimately bounded a.nd the ultimate bound is proportional to
k. An estimate of the ultimate bound ca.n be obtained using Theorem 4.18.
- 145
• 9.11 (a) With b = 0, we have
Linearization at the origin yields the Hurwitz matrix A = - I; hence the origin is exponentially stable.
Global asymptotic stability can be seen by solving the equations to obtain
X1(t) = x1(0)exp [Lt (sin2(e-2rx2(0))- 1) dT], x2(t) = e- tx2(0)
It is clear that for all initial states, the solution x(t) tends to zero as t tends to infinity.
(b) Exponential stability for sufficiently small b follows from ~ a 9.1. To show that the origin is not
globally asymptotically stable, show that there are other equilibrium points. In particular, there is an
equilibrium point at x1 = -(1 + b)1r/2b, x2 1r/2. =
(c) While we cannot apply Lemma 9.1 because the Jacobian matrix is not globally bounded, the lemma.
hints tha.t the origin of the nominal system is not globally exponentially stable because if it was so we would
have expected the origin of the perturbed system to be globally exponential stable as well. The fact that the
origin of the nominal system is not globally exponentially stable can be seen from the closed-form solution
given in part (a).
• 9.12 (a) Let b = 0. Tty V(x) = ½xf + ½zi-

• 2 2
V{x} = -x1 + x1z2(z1 + a) - z1x2(z1 + a) = -x1
V{x) =0 => x1(t) ~ 0 ~ ax2(t) = 0 ~ x2(t) =0

Thus, the origin is globally asymptotically stable. To investigate exponential stability, linearize at x = 0 .
A = aJI = [ -1 + z2 .r1 +a ] = [ -a
-1 a ]
8x z:=D -2z1 - a 0 z:=O
0
The characteristic equation of A is A2 + A + a 2 = 0. Hence, A is Hurwitz and the origin is exponentially

stable.
(b) Let b > o. A= *L=O = [ =! : ].The characteristic equation of A is xi+ {1 - b)A + a 2 - b= o.
Hence, A is Hurwitz if b < min{l,a2 }.
(c) For b > 0, the equilibrium points are
+
r. -a../b
(0,0), ( -a+ vb, vb) , (-".. _. r.b a +
vo, ./f, vb)
Since the system has multiple equilibria, the origin is not globally asymptotically stable.
(d) While we cannot apply Lemma 9.1 bec&US(! the Jacobian matrix is not globally bounded, the lemma
hints that the origin of the nominal system is not globally exponentially stable because if it was so we would
have expected the origin of the perturbed system to be globally exponential stable as well. It is possible to
show that the origin of the nominal system is not globally exponentially stable by a contradiction argument
that involves deriving a special converse Lyapunov function for this example. The argument is long and can
be found in the solution manual of the first edition (See Exercise 4.25).
• 9.13 (a) V(.r) = x4 satisfies (9.11) with o 1 (r) = o 2 (r) = r'.

. 3 4x4 . 4r4
V(x) == 4x x = - -1 - 2 => V(x) < -03(1.rl), with 03(r) = -1 - -
+x - +r2
- 146
(b) o 1 ( -) , o 2 O, and 04 ( ·) are clearly class Koo functions. 03 ( r) is monotonically increasing and o 3 ( r} -t oo
as r -t oo. Hence, o 3 ( ·) belongs to class /C 00 •
(c)
(d)
I xz I=
1: 1 ~~2 S ½, V lxl
1 -x
6> 2 => x = 1 + z2 + t, > 0
Hence, the solution x(t) escapes to oo for any initial state x(O).
• 9.14 Let V(t,x(t)) = 0.

D+W = lim sup _hl [W(t + h,x(t + h)) - W(t,x(t)))
h--+O+
= 1 y'V(t + h, x(t + h))
lim sup -h
h--+O+
We have
V(t + h, x(t + h)) S ~ llx(t + h)ll 2
x(t + h) =h[/(t,O) + g(t, O)) + o(h} => IJx(t + h)ll 2 =h2 llg(t,O)ll 2 + ho(h)
; 2 V(t + h, x(t + h)) S ~ llg(t, 0)11 2 + o~) S ~ 62 (t) + o~)

. lv'
lim hS:f+,;, V(t + h,x(t + h)) s y~ [Ii re;
2o(t) s y ~v 2o(t)
since JC4/2c1 ~ 1. Thus, D+W S c.,,5(t)/(2,/ci), which agrees with the right hand side of (9.17) at W = 0.
• 9.15 As in Example 9.6, the perturbation term B(t)x satisfies (9.15) with -y(t) = IIB(t)ll2 and 6(t) = 0.
1
t:,.
lt+.6 -y(-r)
t dr
1 [ 00
S t:,. lo ")'(r) dr S t:,.
k
where k is independent of .6.. Given e > 0, we can choose A large enough so that (k/ A) < ~- We conclude
from Corollary 9.1 and the third case of Lemma 9.5 tha.t the origin is globally exponentially stable.
• 9.16 As in Example 9.6, the perturbation term B(t)x satisfies (9.15) with -y(t) = IIB(t)ll 2 and 6(t) = 0.
1
.6.
lt+.6 -y(-r) dr
t
where k is independent of~- Given e > 0, we can choose A large enough so that (k/../A) < e:. We conclude
from Corollary 9.1 and the third case of Lemma 9.5 that the origin is globally exponentially stable.
- 147
• 9.17 Write A(t) = A+ B(t), where B(t) = A(t) - A. Since A is Hurwitz and B(t) -+ 0 as t -+ oo, the
result follows from Example 9.6.
• 9.18 Using the third case of Lemma 9.6, we conclude that x(t) -+ 0 as t-+ oo.
=
• 9.19 We view the system x f(t,x) as a perturbation of the nominal system x f(O,x). Noting that =
=
A+ AT= -2o:l and B +BT= 2/31, we use V(x) xTx. The derivative of V along the trajectories of the
nominal system is given by
V = xT(A + AT)x - llxll~xT(B + BT)x = -2a1Jxll~ - 2Pllxll~ S -2a:ll:z:II~

Thus, V satisfies (9.3)-(9.5) globally. Since ')'(t) -+ 0 as t -+ oo, we conclude from Corollary 9.1 and the
second case of Lemma 9.5 that the origin is globally exponentially stable.
• 9.20 Let V(x) = zT Px.

V = 2xT P[f(x) - aG(x)GT (x)P.i-J + 2xT Pw(t} = 2xTPf(x) - 2uxT PG(x)GT(x)Px + 2xT Pw(t)
::; -"';XTPx - W(x) + 2xT Pw(t) < -1'V(x) + 211:i:%>.max(P)(a + ce- 1)
Let U(t) = JV{x(t)).
U::; -lU(t) + Amax(P) (a+ ce-t)
2 \l'Amin(P)
Application of the comparison lemma yields
U(t) $ exp(-1't/2)U(O) + t exp[-(-y/2)(t - r)] .\ma.x(P) (a+ ce-r) dr

lo ___ J.\nun(P)
The right·hand side approaches (2ah)[>.max(P)/v'Alllin(P)] as t tends to infinity. Hence, an ultimate bound
on U(t) can be taken as (2kah)[Amax(P)/v'Amin(P)J fork> 1. Since llxlJ2 $ y'V/>.min(P), an ultimate
bound on llxll2 can be taken as (2ka/,)[Amax(P)/ Amin(P)].
• 9.21 The perturbation term satisfies (9.15) with -y = 0. Therefore, (9.18) takes the form
W(t) < e-a(t-tolW(to) + t

J,o
e-a(t_-r) k6(r). dT, Vt ~ to
W(t 0 + (i + l)A) < e-a.6.W(to + iA) + l. to+{i+l).6.
to+i.6.
e-u(to+(Hl).6.-r)U(r) dT
::5 e-a.o.W(to + iA) + l to+(i+l).6.
to+~
kt5(T) dr $ e-a.6.W(to + iA) + k11A
Hence
i-1
W(to +iA} ~ (e-u 6 )iW(to) + L (e-".6.)i-;i-I krJA::; W(to) + 1 ~:~...6.
j=O
Between any two sampling points to+ iA aod to+ (i + l)A, we have
W(t)::; e-u(t-to-i.6.}W(to + iA) + t

lto+i.6.
e-u(t-r)k6(r) dr :S W(to + iA) + krJ6.
Thus
W(t) ::5 W(to) + 1 k!'J~a-.6. + kr,6., 'rt to + iA :S t :S to+ iA + 6.
-e
- 148
Setting k 1 = k(2 - e-uA)/(1 - e-aA), we can rewrite the foregoing inequality as
W(t) S W(to) + k111A, V to+ i6 :'.St :'.S to+ i~ + ~

Recall that W(t) = JV(t, x(t)). Hence
W(to) ~ Jc;'llx(to)lb, and W(t) ~ v'cillx(t)ll2
Thus
Given~> 0 and~> 0, choose tJ and p small enough to satisfy
• 9.22 Prove by induction that all the principal minors of A are positive. It is clear that a 11 > O and
Now assume that the mth-order principal minor det Am > 0 and write the (m + l)th-order principal minor
as det Am+l, where
Am+1 =[ !¥' : ]
and Am, b, cT, and d have dimensions m x m, m x I, 1 x m, and 1 x I, respectively. All the elements of b
and c are nonpositive and d > O. We have
det Am+l = det Am det(d - cT A;;-,1 b)
From the diagonaJ dominance property, a1l the elements of A~1 are nonnegative. It follows that all the
elements of (d- cT A;;/b) are nonnegative. To see this point consider the equation
Amv + b = W1
CTV +d = W2
where v is a vector of ones. Due to diagonal dominance, w1 a.nd w2 have positive components. After a simple
manipulation we write
d- cT A;;/b = w2 - cT A;;/w1
Since c has nonpositive elements, A;1 has nonnegative elements, and w 1 has positive elements, we conclude
that -cT A;;/w1 ,?: 0. Thus, d- cT A;;,1b > 0. Therefore, detAm+l > 0 and the proof is complete.
• 9.23 H ¢i(x1) = //xiii, then
V(t,x) .$ -½Amin(DS + sTD) L ¢:(xi) .$ -c3j/xll 2
for some C3 > 0. If
then
V(t,x) = Ldi½(t , xi) i I:d,cdlxill2 :5 c2llxl1 2
and, similarly, V(t,x) ~ cillxll 2 . The conclusion follows from Theorem 4.10.
- 149
• 9.24 A direct application of Theorem 9.2 will not be conclusive. Therefore, we proceed to perform the
composite Lyapunov analysis as follows. Write the state equation as
±1 = -x~ + g1 (x), ±2 = -x~ + g2(x)

where 91(:t) = -l.Sxdx21 3 and 92(x) = x~x~. We try Vi(xi) = ½x? and Vi(x2) = ½x~ as Lyapunov function
candidates for the isolated subsystems.
: (-x:) = -x1 = -¢f(xi), where 4>1(x1) = x~
: (-x;) = -x~ = -¢>~(x2), where 4>2(x2) = lx213

Next, we study the interconnection terms.
: g1(x) = -l.5x~lx21 = -1.5¢1¢>2

3
dV2 ( 2 3
d,x2 9z x) = X1X2 $ r/)1</>2
Consider the composite Lyapunov function V(x) = di Vi (xi)+ "2·½(z2 ).
For V to be negative definite, we want the 2 x 2 matrix to be positive definite. Choosing d1 = "2 = I results
in a positive definite matrix. Bence, the origin is asymptotically stable.
• 9.25 Use the notation of Section 9.5. Represent the system as an interconnection of two subsystems.
[!~ ] = [ ~I :1 ][:~ ]+ .[ x:!: ]

:i3 : [ 1 1 ] [ :: ] - x~
Use
V, (
1 X1,X2
)
=[ .Z1 ] T [
X2
l.5 0.5 ] [
0.5 }
XX 1 ]
2
as a Lyapunov function candidate for the first isolated subsystem.
Vil uolmted
. = -x~ -x~ => 01 = 1, ¢1 = /xi +x~
We also have /31 = 1.809. Use V2(xa) = ¼xi as a Lyapunov function candidate for the second isolated
subsystem.
V2j.••oltitr:d = -zg => 02 = 1, ¢>2 = lx3 j3
We also have /j2 = 1. The interconnection terms satisfy
-150
The matrix S is given by

S =[ 1- l .809V2c1 - l .809'\1'2c2 ]
-v'2 1
Sis an M-matrix if
1 - l.809V2c1 - 3.618c2 >0
which can be ensured by choosing c1 and c2 small enough. Thus, the origin is asymptotically stable.
• 9.26 Use the notation of Section 9.5. Let P; be the solution of the Lyapunov equation P;Aii+AiiP; = -Ji
and use ¼(xa:) = xT P.x; as a Lyapunov function foi: the ith isolated system.
aa¼ A;,z; == -llx;IJ~ * a; = 1, (/>; = llz;IJ2

Z;
The interconnection tenns satisfy

ffl
llu,(x)ll2 $ L IIA,jll2llx;ll2 *
j=l
'Yij = IIA;;ll2
The matrix S is defined by
1, for j =i
S;j = { -2IIP;ll2IIA,;ll2, for j i=i
The origin is asymptotically stable if Sis an M-matrix.
• 9.27 Since O $ e;, ~ 1, we have
m m
llg;(t,eux1, . .. ,eimXm)II $ I:e,j'Y;;</>;(x,) :SL -Y,;<P;(x,)
j=l j=l
Thus, the conditions of Theorem 9.2 are satisfied irrespective of the values of e,j.
• 9.28 (a) The closed-loop system is given by
[! ]= A [ =] + [ ~ ] r(t)
where .A= [ A -=_gFi -~F2 ] is Hurwitz. When r is constant, the system has a unique equilibrium point
(x(r),i{r)), where Cx = r. Moreover, this point is exponentially stable. Hence x(t)-+ x and z(t)-+ z as
t-+ oo. Consequently, y(t) = Cx{t) approaches Ci r as t -+ oo. =
(b) Suppose that llr(t)IJ.:::: e for all t ~ 0. The equilibrium point (i(r),z(r)) is a linear function ofr. Define
x = x - x and i = z - z to shift the equilibriwn point to the origin.
Consider the Lyapunov function V (i, i) =[ !] T P [ ; ] , where P is the solution of the Lyapunov equation
PA+ ATP= -I. The function V satisfies the inequalities (9.41)-(9.44) of the text with c1 = ,\min (1'),
c2 = >.ma: (P), C3 = 1, C. = 2>.ma:i- (P), and c5 = 0 (since V
is independent of r). Thus, all the assumption
of Theorem 9.3 are satisfied globally. Therefore, the solutions (x(t), z(t)) are wiiformly ultimately bounded
with an ultimate bound that is proportional toe. It follows that the norm of the tracking error is smaller
than ke for some k > O. Moreover, if r(t) -+ 0 as t --+ oo, then the tracking error tends to zero.
- 151
• 9.29 (a) Since (x, z) is exponentially stable, for initial states sufficiently close to the equilibrium point,
=
x(t) -+ x and z(t) -+ z as t -+ oo. Consequently, y(t) h(x(t)) approaches h(x) r as t-+ oo. =
(b) Suppose llr(t)i12 SE for all t;?: 0. It can be ell.'lily verified that all the assumptions of Lemma. 9.8 are
satisfied in some domain around the equilibrium point. Hence, there is a Lyapunov function that satisfies
inequalities (9.41)-(9.44) of the text. Now, it can be seen that a.ll the assumptions of Theorem 9.3 are
satisfied in some domain around the equilibrium point. It follows from the theorem that for sufficiently small
e, the solutions will be uniformly ultimately bounded to a ball around the equilibrium point. The radius
of this ball is proportional to e. Hence, the norm of the -tracking error is smaller than ke for some k > O.
Moreover, if r(t) -t Oas t-+ oo, then the tracking error tends to zero.
• 9.30 The solution of this exercise is identical to the solution of Exercise 9.29, except for talcing the state
vector as (x, z1, Zz).
• 9.31 Following the proof of Lemma 9.8, we see that
for some positive constants k1 and"/, and P(t)A(t) + ATP(t) = -I has a unique positive definite solution
P(t) for each t 2'. 0. The matrix l'(t) satisfies
P(t)A(t) + AT(t)l'(t) = -Q(t)
where Q(t) = ,-F(t)P(t} + P(t)A(t). Therefore,
P(t) = Loo [e-rA(t)r Q(t) [e-rA(t)J dr
IIP(t}II ~ UQ(t)ll 1
00
kte-Z-,-r dr :5 k2IIA(t)II
Let V(t, x) = xT P(t)x. Then
V = .xT[P(t)A(t) + AT(t)P(t)]x + xT P(t)x

= -xTx + xT P(t)x ~ -ci V + c2II.A.(t)IIV
Using the comparison lemma, we obtain
V(t,x(t)) ~ V(to,x(to))exp [i: (-c1 + c2IIA.(T)II) dT]

By Cauchy-Schwartz inequality, we have
L t
ll~(r)H dr::;
(
LnA(r)ll
t
2 dr
) 1/2
v1f=to :5 Vp(t - to)
Thus
V(t,x(t)) ~ V(to,x(to)) exp [-c1(t -to)+ c2Jp(t- to)]~ c3e-c4 (t-to>v(t 0,x(t0 ))
which shows that the origin of x = A(t)x is exponentially stable.
- 152
Chapter 10
• 10.1 6(e) ;::; O(e) => 16(e)I $ lcie:I, for ltl < c. Thus, 16(s)I :S kye.,fi ~f k1 ,fi, for lel < c. Thus, 6(s) is
0( ,/i). It is not O(lc:I I).
• 10.2 The answer is no. If there was such N, then
€¼ $ kEN => I :5 k"'en.N

The second inequality is impossible since en.N -t O as e -t 0.
• 10.3 (a) Set e = 0.

:i:10 ;::; -O.2x10 + i - tan-1 X10, x1o(O) = 111
±20 = -0.2x20 + ¾- tan- 1 Xzo, X2o(O) = T/2
(b) Substitute xi = x10 + exu + · · ·, and x2 = x:io + Exn + · · · in the state equations and match the
coefficients of e. For the first equation we have
11"
= -(0.2 + e)(x10 + Exu + · · ·} + 4 - tan- 1 (x10 + EX11 + · · ·) + e-tan- 1(x20 + tx 21 + · · ·)
= -O.2x10 + e(-x10 -O.2.xu + · · ·) + ~ - tan- 1 X10 - £ 2 xu + ·· · + ttan-1 X20 + · ·
4 1 + X 1o
Matching coefficients of E, we obtain
±11 = - (0.2 + l +ltjo) X11 - X10 + tan-I X20, X11 (0) =0
Similarly
X-21 = - (0.2 + l +lxti) X21 - X20 + tan- 1 X10, X21 (O) =0
(c} The nominal system has a unique equilibrium point (p,p), where p satisfies the equation 0.2p+tan- 1 p =
1r J4. Linearization at this point yields the matrix
8J
8x
l _[
(p,p) -
-0.2 -
0
ri:;z O
-0.2 - 1 .f_;i
]
The equilibrium point (p,p) is exponentially stable. Hence, Theorem 10.2 applies and we can conclude that
there is a neighborhood of (p,p) such that for all initial states in that neighborhood, the approximation is
valid on the infinite time interval.
(d) Write the equation for X10, X20, :tu, and x21 a.s four simultaneous autonomous equations and solve using
a computer program. The simulation results are shown in Figure 10.1.
- 153
~n---~----------. 1.8----------~--~
0.7 u
0.65 1.2
.,.- ,t'
0.6
Exact
0.55 1st order approx 0.8
2nd order appro . 0.6 .___ _ _ _ _ _ _ _ ___,
2 3 0 2 3
• 10.4 (a)
±10 = x20, X1o(O) = '71
.X20 -X10 - X20, x20(D) = 112
(b)
±11 = x:n, xu(O) =0
±21 = -xu - z21 + xfo, X21(0) =0
(c) The nominal system is a linear one with a Hurwitz matrix. Hence, the origin is exponentially stable
and Theorem 10.2 applies.
(d) The simulation results are shown in Figure 10.2.
- Exact 0.1
1st order approx
2nd order appro 0
0.!5
-0.1
.,.- ....-0.2
-0.3
0
-0.'4
-0.5
--0.5
0 2 3 4 5 0 2 3 4 5
• 10.5 (a)
:i:10 = -X10 + X20, X10(0) = fh
.i-20 = -x20 - ½Xio, x20{0) = TJ2
(b)
:i:11 = -xu +x21, xu(O) = 0
:i:21 = X10 - X21 - X~oX21, X21(0) = 0
(c) The nominal system has a unique equilibrium point a.t the origin. The Jacobian matrix Mj
&(0,0)
=
[ 01 !1 ] is Hurwitz. Hence, the origin is exponentially stable and Theorem 10.2 applies.
(d) The simulation results are shown in Figure 10.3. Note that x 20 (t) =0.
-
154
0.8
Exact
1st order approx
0.6 2nd order approi
..-
0.-4
0.2
0'-------------=-
0 2 3 4 2 3 4
• 10.6 (a)
±10 = X10 - Xio, X1o(O) = 1J1
X20 = 2:t20 - X~o, X20(0) = T/'l
(b)
Xu = Xu - 2x10X11 + X10X20, xu(O) =0
X21 = 2x21 - 2x20X21 - X10X20, X21(0) =0
(c) The nominal system ha.s four equilibrium points at (0, 0), (0, 2), (1, 0), and (1, 2). The Jacobian matrices
at these points are
8/1
8x (o,o) =
[0 0]
2 '
1 8/1
8x (o,2 ) =
[ 01 0 ]' 8/1
-2 8x {l,O) =
[ 0 0]
2 '
-1 8/1
8x (l, 2) =
[ -10 0 ]
-2
Theorem 10.2 applies only to the equilibrium point (1, 2).

(d) The simulation ·results are shown in Figure 10.4.
2-----~--._.., . ,. ______
1.-4.---------------,
---
1.2 1.8 ,· , -- - - - - - - -
I "'
..- ~-~.------~--,
.---·- --·- · ..
.,,_
1.8
I
I
I
0.8 1.4
Exact
0.6 1st order approx
2nd order appro
2 3 4 2 3 4
I I
• 10.7 (a)
x10 = -x10 + X2o(l + x10), X1o(O) = 111
X20 -X1o(X10 + 1), X2o(O) = 1/2
(b)
±11 = -Xu + (1 + X1o)X21 + X20X11 + 1 + X~o, xn(O) 0=
:i-21 = -(1 + X1o)zu - X10X11, X21(0) = 0
-
155
( c) The nominal system has a unique equilibrium point at the origin. The Jacobian matrix !li I -
7fi (0,0) -
[ =~ ~ ] is Hurwitz. Hence, the origin is exponentially stable and Theorem 10.2 applies.
(d) The simulation results are shown in Figure 10.5.
t:=-0.05 £ = --0.06
2 3
Exact 2
1 1st order approx
1
x- 2nd order appro )(
C'II
0
0
-1
-2
1 2 3 4 5 0 1 2 3 4 5
t:=-0.1 £ = --0.1
2 3
2
1
-- 0
'I X
N
0
' . .-:. =·=-=-~--=--.:· ,:.=.
-------- -1
-2
1 2 3 4 5 0 1 2 3 4 5
t:•-0.2 £ = -0.2
2 3
2
N
1
K,... X
0
0
-1
.' .X.._~ _-.=.-:. -:-_-:-_-. -- . -- .
-2
1 2 3 4 5 0 2 3 4 5
• 10.8 The nominal solution is given by

±10 = -X10, X1o(O) =11 => = r,e-t
x1o(t)
±20 = -X20, X2o(O) = fJ => X2o(t) = qe-t
By calculating the state transition matrix as a function of f:, the exact solution is determined as
.z1 (t) = 17e-t(cosct + sin et)
.z2(t) = qe-t(cosf:t - sin et)
Thus
.z1(t)-x1o(t)= = ,.,e-t(-l+cos.!t+sinet)
x2(t) - X2o(t) = = 1Je-t(-1 + cos et - sin et)
- 156
It can be easily seen that 11- cose:tj $ e:t and I sinetj $ et. It can be also seen that te-t $ 1/e for all t ~ 0.
Thus
These error bounds are consistent with Theorems 10.1 and 10.2. They confirm that the approximation error
is bounded by kE for some constant k. Notice, however, that the constant k depends on the initial state
of the system. Thus for this quantity to be O(e), the initial states should by 0(1); that is, they should be
uniformly bounded as e ~ O. When we have e = 0.1 and T/ = 1, we should expect the error to be indeed
of the order of 0.1. But when e = O.l and T/ = 10, then actually 'f/ = O(1/e) and we should not expect the
error to be of the order of 0.1, as we can see in the current case from the explicit error bounds. Numerical
calculations will confirm these observations. The point of this example is to illustrate that when we use the
perturbation method to approximate the solution of a differential equation for some small value of e, then
that value puts a limit on the magnitude of the initial conditions.
• 10.9 (1)
fc.v(x) -= - 11,r
~ 0
(x - x 2 ) sin2 t dt = - 11:ir
~ 0
sin 2 t dt (x - x 2 ) = ½(x - x 2 )
=
The average system x ½e(x - x 2 ) bas equilibrium points at x = 0 and x 1. The Jacobian function at =
these points are e/2 and -e/2. Thus, the equilibrium poin~ x = I is exponentially stable. By Theorem 10.4,
we conclude that, for sufficiently small e, the system has an exponentially stable periodic solution of period
1r in an 0(£) neighborhood of x =
l. However, x = 1 is an equilibrium point of the original system. Thus,
=
the periodic solution is the trivial solution x(t) 1, and the equilibrium point x = 1 is exponentially stable
for sufficiently small£. Moreover, for initial sta.tes sufficiently near x = 1, x(t,£) = Xa.v(t,e:) + O(e) for all
t ~ o.
(2)
!a.v(x) = - 11"'
~ 0
(x cos 2 t - ½x2 ) dt = ½(x - x 2)
This is the same average function M pa.rt (1). The rest of the solution is similar to pa.rt (1), except that in
the current case the periodic solution is nontrivia.l. ·
(3)
fa..,,(x) = -1111' (-x + cos 2 t) dt = -x+ ½
7r 0
The average system x = -£(x - ½) has an equilibrium point at x = ½which is exponentially stable. By
Theorem 10.4, we conclude that, for sufficiently small£, the system has an exponentially stable periodic
solution of period~ in an 0(£) neighborhood of x = ½- Moreover, for initial states sufficiently near x = ½,
x(t,e) = Xc. 11 (t,E) + O(e) for all t ~ 0.
(4)
!av(x) = 2~ fa 2
"' (-zcost) dt =0
The average system is :i; = 0. We can only conclude that
x(t,£) = Xa.v(t,£) + O(e) = x(O} + O(E}, 'tr/ t E [O,b/£)
for finite b > 0.
• 10.10 (1)
la.v(x)
1
= 2,r 1O
2,r [
X2
-(1 + 2 sin t)x2 - (1 +cost) sin:i:1
]
·
=
[
X2
-x2 - sinx1
]
- 157
The average system has an equilibrium point at the origin. Linearization at the origin yields £:v I =
8
z=O
[ ~l !1 ] , which is Hurwitz. Thus, the origin of the average system is exponentially stable. It follows
that, for sufficiently small e, the original system has a unique exponentially stable periodic solution in the
neighborhood of the origin. But the origin is a.n equilibrium point of the original system. Hence, the periodic
solution is the trivial solution, which shows that the origin is exponentially stable.
(2)
fav(x)
1 / 21r[ (-l+l.5cos2 t)x 1 +(1-l.5sintcost)x2 ]
= 21r } 0 (-1 - 1.5sintcost)x1 + (-1 + l.5sin2 t)x2 =
[-¼-1 1 ] [ x1]
-¼ X2
The matrix on the right-hand side is Hurwitz. Hence, the origin of the average system is exponentia.lly
stable. Noting that the origin is also an equilibrium point of the original system, we conclude, as in part (1),
that the origin is exponentially stable for sufficiently small e.
( 3) View the given system as a perturbation of the system
x = e(-xsin2 t + x 2 sin t)
Apply the averaging method to this system.
1 /2 1( 1
lau(x)= 21rJo [-xsin2 t+:z:2sint] dt=- 2x
The matrix on the right-hand side is Hurwitz. Hence, the origin of the average system is exponentially stable.
Noting that the origin is also an equilibrium point of the original system, we conclude, as in -pa.rt (1), that the
origin is exponentially stable for sufficiently small e. Since the perturbation term satisfies leze-tl::; clxle- 1
and e- 1 -+ Oas t-+ oo, we conclude from the second case of Lemma 9.5 that the origin of the given system
is exponentially stable for sufficiently small e.
• 10.11 View the system as a perturbation of the system treated in Exercise 10(2}, with e-t as the per-
turbation term. We saw in Exercise 10(2) that the origin of the nominal system is exponentially stable
for sufficiently small e. Since the system is linear, it is globally exponentially stable. From part (3) of
Lemma 9.6, we conclude that z(t) tends to zero as t tends to infinity.
• 10.12 (a)· Let A = M JM- 1 , where J is the real Jordan form of A. Then, exp(At) = M exp(Jt}M- 1
and exp(- At) = Mexp(-Jt)M- 1 • Since A has only simple eigenvalues on the imaginary axis, exp{Jt) and
exp(-Jt) are formed of sinusoidal functions oft. Thus they are bounded, which implies that ex:p{At) and
exp(-At) are bounded for all t ~ 0.
(b)
:i: = -Aexp(-At)y + exp(-At)[Ay +Eg(t,y,e)J = texp(-At)g(t,exp(At}x,e) ~f ef(t,x,e)
• 10.13 Let z1 = y and z2 = 'fl. Then
z. = [ -l0 1 ] z +e [ -2z10cos2t ]
O = Az +eg( t,z )
def
Use the change of variables z = exp(At)x to transform the system into x = ef(t,x), where
f(t,x)
cost
= exp(-At)g(t,exp(At)x) = [ sint -sint][
cost
0
-2(x1 cost+ x 2 sin t) cos2t
]
: [ ½sin 4t 2 sin 2 t cos 2t ]

~ F(t)x
-2 cos2 t COS 2t -½sin4t X
- 158
The average system is given by i; =eF ,,x, where

4
1 {"
Fci,, = :;;: lo F(t) dt = [ -½
0
The matrix F011 is not Hurwitz. Thus, we can only conclude that
x(t,E:) = Xa,,(t,E:) + O(e) = exp(tF vt)x v(O) + O(t),

0 0 fort E [O,b/t]
for a finite b > 0. Transforming this expression to the original variables, we obtain
z(t,e) =exp(At)exp(tF,.,.t)z(O) +O(e), fort E [O,b/£]
• 10.14 Let z1 = y and z2 :::: iJ. Then

i = [ ~1 ~] z+c: [ Bz~~ost] ~r Az+tg(t,z)
Use the change of variables z = exp(At)x to transform the system into x = Ef(t, x), where
f (t, x) = exp(-At)g(t, exp(At)x) = [ ~:: -~~ t ] [ 8(-x1 sin t +~ 2 cos t)2 cost ]
== [ -Ssin t cost (-xi sin t + z2 cos t) 2 ]
8 cos2 t (-xi sin t + x2 cos t) 2
The average system is given by x = c:/ v{x), where

4
f av(x) = [ xl~~!~ ]
The average system has a unique equilibrium point at the origin, but the origin is not exponentially stable
as it can be seen by linearization. Thus, we can only conclude that
x(t, e) =x 0 ,,(t, c) + O(c), for t E [O, b/c]

for a finite b > 0. Tuansforming this expression to the original va.ria.bles, we obtain
z(t,e') = exp(At)x v(t,E) + O(c),

0 fort E (O,b/cJ
• 10.15 (1) g(y, y) = -y(l - y2 ).
i r2•
lav(r) = 211' lo [-rcost/>(1- r 2 sin2 ti>) cos¢] d¢ = -½r + lr3
The average system ~; = c/,n,(r) has equilibrium points a.tr= 0 and r = 2.
dfav I __ -
l
2•
dfo., I _ -
l
dr r=O dr r=2
The equilibrium point r = 0 is exponentially stable. Noting that r =

0 is also an equilibrium point of the
system
dr 9(·,·)cos<J,
,1,. = e/{tf,,r,e),
d ¥' where f =
1 - (E / r )g ( ·, ·) sin
.

we conclude that, for sufficiently small E, r = 0 is an exponentially stable equilibrium point of the foregoing
system. Noting further that J, =
1 + O(e), we can carry this conclusion over to the second-order system
- 159
ii+ y =
e9 and say that its origin is exponentially stable for sufficiently small c. We can also show that there
is an unstable limit cycle in an O(e) neighborhood of r = 2. To do this, reverse the time variable in the
=
original equation to obtain 'ii+ y ey(I - y 2 ), which. is the Van der Pol oscillator studied in Example 10.11.
Since Va.n der Pol oscillator has a stable limit cycle near r = 2, the current system has an unstable limit
cycle near r = 2. The same conclusion is also obvious from the foregoing analysis since reversing time has
the effect of reversing the independent variable of the average system.
=
(2) g(y,y) y{l -y2) -y3 .
The average system is the same as the Van der Pol Oscillator of. Example 10.11. The rest of the solution is
the same as in the example.
=
(3) g(y, y) -[l - (311' /4)jyl]y.
JB.,(r) = 2~fo [-1 + 3:

2
1f rj sin¢1] rcos2 ¢,dip= -½r + ½r 2
The average system ~ == e/4.,(r) has equilibrium points at r =0 and r = 1.

dfGtJ
dr
I
,.=0
= _!2' dfo.,
dr
I_
,.=1 -
!
2
=
The equilibrium point r 0 is exponentially stable. Similar to part (1), we can conclude that, for sufficiently
small e:, the origin of the second~order system is exponentially stable. By reversing time, we can show, a.gain
as in part (1), that for sufficiently small e there is an unstable limit cycle in an O(t) neighborhood of r = 1.
=
(4) g(y,y) -(1- (31r/4)11i1Jti,
Proceed a.s in Case (3).

=
(5) 9(y,iJ) -(y -y3 ).
/o.,(r) = 2l11' 12,r

0
(-rcos¢ + r 3 sin3 ¢) cost/) d¢, = -½r
=
Th~ i •,erage system -$ E/o.,(r) has one equilibrium point at r = O. The equilibrium point is exponentially
stable. Repeating the argument of part (1), we conclude that, for sufficiently small e:, the origin of the
second-order system is exponentially stable.
(6) g(y, ti) = y(l - y2 - r/).
l / 2 7r l /2'1(
f B.,(r) = 21r Jo (1 - r 2 sin2 4> - r 2 cos 2 ¢] r cos2 </> d¢ = 2'11' Jo (1 - r 2)r cos2 <f, d<f, = ½r(l - r 2)
The average system i =e/G'l/(r) has equilibrium points at r = 0 and r = 1.

dfa..,I
dr r=O
=!
2'
d/"'tJI =-1
dr r=l
For sufficiently small e:, there is a stable limit cycle in a.n O(e:) neighborhood of r = I.
- 160
• 10.16 (a) Apply the averaging method of Section 10.5 with g(y, y) = y + y(l - y2 - y2 ).
1 r21r
/av(r) = 2,r Jo [rsin¢, + (1- r 2 )rcos¢,] cos¢, def>= ½(r - r 2)
The average system has equilibrium points at r = 0 and r = l. {d/av/dr](l) = -1. For sufficiently small c,
there is a stable limit cycle in an O(c) neighborhood of r = 1.
(b) There is a unique equilibrium point at x 0. =
ofl =[ -10+ e
8x
1]
e '
,\
1' 2
=s±Je2-4(1-e)
2
:i:=O
For c > 1, the equilibrium point is a saddle. Hence, there are no periodic orbits.
• 10.17 (a)
ii = W)2cPY2 =
dt
~ tPu
ku• dt 2
= _l
ku•
{-kv. + ,\ [i _ (du)
dt
o:
2
] du}
dt
= -y+ :. [1-a(~:r ]1i =-y+e(l-½1/)y

!i2
(b)
f,n,(r) = 2,r lo
1 r27r (r cos¢ - ½r3 cos3 ¢)cos¢> d¢ = ½r - ½r3
Proceed similar to Example 10.11.

( c) The phase portrait should show a stable limit cycle which approaches the circle r 2 as c --t 0. =
(d) The simulation results are shown in Figure 10.6. It as clear that, as c approaches zero, the stable limit
cycle approaches a circle of radius two.
,.1 lt•0.1
-2 2
• 10.18 (a)
dx2 [ k 2 3
- = E : --(x1+axi)-
c A
-x2+-cosr]
dr m m m
(b) The average system is given by
d:c2a11
~ =e [ - m
k (X1av + 2 3
0 X1 011
}
-
C
m X2011
]
- 161
The average system has a unique equilibrium point at the origin. Linearization at the origin yields a
Hurwitz matrix. Thus, for sufficiently small e (equivalently, sufficiently large w), the system has a unique
exponentially stable periodic solution in an O(t) (equivalently, O(1/w)) neighborhood of the origin. The
frequency of oscillation is close tow.
• 10.19 Start from (10.43) of the text. For t $ l/ Jij, we have
{.,/ii
11llw(t)II < k~u(O) + kO'(O) fo e-•s ds
< k~o-(0) + ko-(0) (-,/ije-..,r,; + I - e--.lii) = ku(O){l + -Jij) ( I - e-v'ii) ~r o-1 (fJ)
It can be checked that 01 (17) is a class JC function. Fort> I/Jij, we have
fJllw(t)II $ !u ( ~) + kri [1l-.fii e-'l>.ri-Xu(>.) d>. + kf7 ft e-rJ>-ri>.o-(>.) d.\

e vTJ lo 111..,r,;
< )r;) + ku(O} (-,/ije--.fii + l - e-..,r,;) + k 1; e-•
~u ( s dsu ( ~ )
< (;+vii +1) u ( ~ ) + ku(O) (1 -e-..,r,;) 02(11) ~f
where we have used te-t $ 1/e. It can be checked that 0 2 (77) is a class JC function. Choose a class JC function
0{17) such that a(r,) ~ max{a1(1J), a:z(TJ)}.
• 10.20 The average system is :i: = eA1.11,z, where
fun -T
11t+T (sin 2T + sin 1.5-r + e-r) dr
T~ao t
= T~ ~ { !T- ¼{sin(2t + 2T) - sin2t] - 1\[cos(l.5t + I.ST) - cos 1.5t)- [e-(t+T) - e- 1]} =½
The average system is i =· ½cz. From the above limit calculation, it is not hard to see that the convergence
function can be taken as a(T) =
1/(T + 1). Hence, the class K Junction 0(11) of Theorem 10.5 can be taken
as 0(17) = kr,. The origin of the average system is not exponentially stable. We can only conclude that
z(t,e) = Xai,,(t,_e:) + O(e) = ete/ 2~(0) + O(~), Vt E [O, b/e]
for a finite b > 0.
• 10.21 (a) We have iJ = -ee(t)w(t) = -ew(t)e(t) and e(t) = (8(t) - 9•]7w(t) ::; ¢,T(t)w(t) = wT(t)(/)(t) .
Then
where A(t) = -w(t)wT(t).

(b) Suppose that w(t) is bounded and smooth enough to ensure that A(t) a.nd its derivatives up to the second
order are continuous and bounded. From Example 10.13 we see that the origin of the system¢= eA(t)¢ is
exponentially stable for sufficiently small E if
Aaiv:;: - lim -T
T~oc
11t+T w(r)wT(r) d-r
t
is Hurwitz. The matrix A,w is negative semidefinite by construction. Hence, it is Hurwitz if and only if it is
negative definite. Thus, we require that
l
t
t+T
w(-r)wT(-r) dr ~ cl, Vt~ O
for some c > 0. This is a persistence of excitation condition.
- 162
Chapter 11
• 11.1 The circuit equations are
=
Let e G2/G1 and R1 = R2 = R. Multiplying the second equation through by 1/R2G1, we can rewrite the
equations as
tii = R~i (-vi + V;i), ev2 = R~i (vi - 2v2 + u)
Setting E =0 results in the equation O = v1 - 2t12 + u, which has the unique root v2 = ½(v1 + u). Hence, the
system is in the standard singularly perturbed form.
• 11,2 The circuit equatjons are given above. Let E = R1 /R2 and C1 = C2 = C. We can rewrite the
equations as
eir1 = R:C(-vi + "2), e1>2 = R~C[tri - (e + l)v2 + eu]
Setting c = 0 yields the equation -vi + 1-'2 =
0, whose roots v2 =
v1 are not isolated. Now, similar to
Example 11.3, take x = ½(v1 + v2). We can take z as in the example, or simply take z = ?12, We obtain
x = 21c (-z + u), ei = R~C[2x - (2 + e)z + eu]

2
Setting e = 0 yields the equation 2x - 2z = 0 which has a unique root z = x. Hence, the system is in the
standard singularly perturbed form.
• 11.3 Recall from Section 1.2.2 that the circuit is represented by
When Lis small, the inductance current x 2 is the candidate to be the fast variable z. Let x = xi, z x 2 , =
and e = L/CR2 •
::i; = ~ [-h(x) + z], d = - 1-[-x - Rz + u]
CR2
Setting e = 0 yields the equation -x - Rz + u = 0 which has a unique root z = (-x + u)/ R . Hence, the
system is in the standard singularly perturbed form.
• 11.4 From the block diagram, we obtain
i = z, i = -z + k[u - x - ,t,(z)]
- 163
Let c: = I/k and multiply the second equation through by E, to obtain

X = Z, C:Z = -E:Z + U - X - tµ(z)
=
Setting c 0 yields the equation u - x -1/J(z) = 0. Hence, the system is in the standard singularly perturbed
form provided ¢-1 {·) is defined in some domain of interest.
• 11.5 Consider equation (11.14) of the text:
dy
dT = g(t, x,y + h(t,x),O)
where (t,x) E [O,t1] x Dz are treated as constant parameters. The equation ha.s equilibrium at y o. =
Linearization at the origin results in the matrix A(t ,x) = t(t,x, h(t, x), 0). We can rewrite the equation as
a perturbation of its linea.riza.tion at 11 0. =
dy
d,,- = A(t,x)y + 1/l(t,x,y)
where
ag
tj)(t , x,y)= [&y(t,x,oy+h(t,x),0)- 8y(t,x,h(t,x),O) y
ag ]
for some 0 < a < 0. Assuming that the Jacobian matrix is Lipschitz in 'Y, we know that there exists a
constant k, independent of (t,x) such that
llt/J(t, x, 11)112 :5 kllvll~, 't/ (t , x) E [0, t1J X D:,;
The matrix A(t, x) is a bounded function in (t,x) and satisfies (11.16). Suppose further that the partial
derivatives of the elements of A(t, x) with respect to t and x are bounded. Then, from Lemma 9.8, we
conclude that there is a Lyapunov function V(t, x, y) that satisfies .inequalities
&V
fJy A(t, x}y $ -c:sllYII~,
where the constants c1 to c4 are independent of (t, x). The derivative of V with respect to (11.14) satisfies
av av 2
l)yg(t,x,y+h(t,x),0) = aylA(t,x)y+?J,(t,x,y)] $-csllvlb+C4kll11II~ $ -½c3IIYII~
for IIYll2 ~ ca/2kC4 , Thus, V(t, x) satisfies inequalities (11.17)-(11.18}. Going from here to inequality (11.15)
is the subject of the next exercise.
• 11.6 Suppose V satisfies (11 .17)-(11.18). The derivative of V with respect to (11.14) is given by
av av 2 c3
dT = 8y g $ -c3 IIYII ~ - c2 V
Hence, (11.15) is satisfied with k = ,/c2/c1 and -y = c3/2c2, To ensure that lly(t)II x(t) = ~ tan (tv2 + tan- 1 ( e\1"2))
The boundary-layer model is
~: = -y, y(O) = T} - (1 + (2) => y(T) = [17 - (I+ {2)]e_,.
Thus,
x(t,e) = ~ tan (tv'2 + tan- 1 ( ev'2)) + O(t:)

z(t,c) = l+½[tan(tv'2+tan- 1 (ev2))r +{17-(l+e2)]e-t/e +O(c)
(b) The simulation results are shown in Figure 11.l.
, .. 0 .1
2.-------------, '"' 0.1
3.5
1.5 ,~ =xunatel I
2.5
3
/
I I
2
>< 1 I N ,/
1.5
0.5
0.2 o.• O.ll 0.8 OA 0.6 0.8
&•0.05 ,-o.o=;
2.....----------- . . . 3.5 .....-- ---- - - -- - ---,
3
1.5
u I
I I
2
>C 1 N
1.5
0.5
--
0.2 0.• 0.6 0.8 0.2 O. i(t) = r,e-t
dy = - -2 tan -1 (11'
-dT -y) , y(O) ='1 + 2{
'lt 2
Denote its solution by y('r) . Thus,
x(t,t:) = r,e-t + O(e), z (t,e) = -211e-t + y(t/e) + O(e)
(b) The simulation results are shown in Figure 11 .2. As you decrease £, the exact a.nd approximate solutions
will come closer to each other.
- 165
t•0.2 E" 0.2
2....-----------,
0
, -1
.... --- -
" N
0.5 -- -- -- -2 - Exact
- - Approximate
0 -3
0 0.2 0.4 0.8 0.8 0 0.2 0.4 0.11 0.8
t• 0.1 E •0.1
1.6
u
1.2 0
. 0.8 ' '-

N
-o.s
-1
0.11 -1.5
0.4
0.2
-- -2
0 0.2 0.4 0.11 0.11 0.2 0.4 o.e 0.11
• 11.9 Setting€ = 0 yields h(t, x) = ln{l + u(t) - x ), assuming that (I+ u(t) - x) > 0. The reduced model
is
x = In(l + u(t) - x)
dy = -(e11 - 1)(1 + u(t) - x)
dr
Linearization of the boundary-layer model at y = 0 results in the linear system
dr
dy = -(1 + u(t) - x)y
Assuming O < c1 ~ [I + u(t) - x] ~ c2 < oo ensures that the origin of the boundary-layer model is
exponentially stable uniformly iD (t, x).
=
• 11.10 (a) Setting e = 0 yields O (z + xt)(z - 2)(z - 4) which has three roots h 1 (t, x) = -xt, h 2 = 2,
and h 3 = 4. We assume that xt > 0 is the domain of interest. Then, the three roots are isolated. Each root
could give rise to a reduced model.
=
(b) For h 1 (t,x) -xt, the boundary-layer model is
dy
-dr = -y(y - xt - 2)(y - xt - 4)
The Jacobian {}g/8y at y = 0 is given by
~,
vy tt=O
= -(xt + 2)(xt + 4) < -8
- 166
Hence, the origin of the boundary-layer model is exponentially stable wiiformly in (t, x}. It is also clear that
y < xt + 2 is the region of attraction. Transformed to the z variable, the region of attraction is z < 2. For
hz = 2, the boundary-layer model is
!= -y(y + xt + 2)(y - 2)
The Jacobian og/{)y at y = 0 is given by
~ t,=o = 2(2 + xt) > 4

Hence, the origin of the boundary-layer model is unstable.. For h3 = 4, the boundary-layer model is
dy
dr = -y(y + xt + 4)(y + 2)
The Jacobian 8g/8y at y = 0 is given by
89 1 = -2(xt +4) < -8
oy ~==O
Hence, the origin of the boundary-layer model is exponentially stable uniformly in (t, :i;). It is also clear that
y > -2 is the region of attraction. Transformed to the z variable, the region of attraction is z > 2.
(c) For a< 2, z(0) belong to the region of attraction of the root z;:; hi, while for a> 2, it belongs to the
=
region of attraction of the third root z h3. Thus, for a< 2, the reduced model is
:t = -x, x(O) =I
and for a> 2, the reduced model is
x = ¼z2t, x(O) =1
An O(e) approximation can be obtained by using expressions (11.20)-(11.21) of Tikhonov's theorem.
• 11.11 Setting E = 0 results in z = sin t. The reduced model is :i: = -x and the boundary-layer model is
dy/dT = -y. Clearly all the assumptions of Theorem 11.2 are satisfied. Hence, the O(e) approximation
x{t,e) = e-tx(O) + O(c:), z(t,c:) = sin t + e-tz(O) + O(c:)
holds uniformly in t for all t ~ 0.
• 11.12 The manifold equation is
-H - x•IJ + ic:xl6/3 - E 8H xH3 =0

3 8x
It can be easily checked that H = -x4 l 3 satisfies this equation.
• 11 .13 The manifold equation is
-(H - sin2 x)(H - e'*z)(H- 2e20z) +e~! xH =0

At c = O we have three equilibrium manifolds
The Jacobian og/oz is negative on H1 and H 3 and positive on H 2 . Hence, H 1 and H 3 are attractive.
- 167
• 11.14 (a) The manifold equation is
Let 'H. = -Lx.

[(A21 - AnL} + eL(Au - A12L)Jx =0
Thus, if L satisfies the equation
(A21 - A22L) + EL(Au - A12L) =0

z = -Lx will be an exact slow manifold.
(b) Let r, = z + Lx. ·Then
x = Aux+ A12(77 - Lx) == (Au - A12L)x + A121J

E"TJ = EZ + EL±
= A21x + A22(1, - Lx} + EL[(A11 - A12L)x + A1211]
= (A22 + eLA12)'1
where we have used the equation satisfied by L. The system is in a block triangular form.
( c) Since the system is in block triangular form, its eigenvalues are the eigenvalues of the diagonal blocks
(Au - A12L} and (A22 + eLA12)/t. Thus, there are n eigenvalues of order 0(1) and m eigenvalues of order
0(1/c).
(d) Let~= x - eH'f/. Then
f = (Au - A12L)c + [e(Au - A12L)H - H(A22 + eLAu1) + A12]'7

which reduces to
when H satisfies the equation
e:(Au - A12L)H - H(A22 + eLA12) + A12 =0

Thus, the change of variables
{=x-EHTJ = x -EH(z + Lx) = (1-EHL}x-EHz, 17 = Lx +z

transforms the system into the bloc.k diagonal form
{ = A.(t)(, where A. = Au - A12L

E"~ = A1(£)17, where A1 = A22 + eLA12
(e) The inverse of the foregoing transformation is given by
x = {+EH17, z = -Lx+ (1-ELH)'IJ
From the first equation it is clear that the component of the fast mode '7 in xis O(e:).
(f) Setting E = 0 in { = A.(e)~ gives the reduced model x = (Au -A12Ail A21 )x. Hence, e(t) = z(t) +O(e).
= =
Setting e 0 in d11/dr = A1(e)11 gives the boundary·layer model dy/d-r A2 2 y. Thus, 77(T) = y(-r) + O(e).
Therefore
x = { + eH11 = x + O(e)
z = -Lx + (I - eLH)11 = -A2l A21ft + y + O(e-)
l
where we have used L = A 2 A21 + O(e).
-168
• ll.15 Application of the transformation of Exercise ll.14; namely,
results in
X = A,.X+B,u,
where
A. = Au - A12L = Ao + O(e)
AJ = A22 + eLAu =
A22 + O(e)
B, (I - eHL)B1 - HB2 =
Bo + O(e)
B1 = B2 + eLB1 = B2 + O(e)
Furthermore, x =
X +e:HZ. Since A22 is Hurwitz, so is Ai, for sufficiently small e:. Thus, Z(t,e) is uniformly
bounded for all t ~ 0. Therefore,
x(t, e) - X(t,e) = O(e:), Vt~ 0
(a) Application of Theorem 3.4 shows that X(t, e:) - f(t) = O(e) on any compact time interval. This implies
that x(t,e:) - x(t) = O(e:) on any compact time interval.
=
(b) Let e X - x.
=
e A,.X + B,u - Aox - Bou:: (Ao+ O(e)]e + O(e)
Since Ao is Hurwitz and e(O,e) == O(E), we conclude from Theorem 9.1 that e(t,e) = O(e) for all t 2: 0.
Consequently, x(t,e) - x(t) :: O(e:) for all t ~ O.
• 11.16 (a) Setting e: = 0 yields the equation ian-1 {1 -x1 - z):: 0, which has a unique root z:: 1- x 1 ~
h{x). The reduced model is

dy . -1
dr = -tan y
(b) Let W(y) = b 2 be a Lyapunov function candidate for the boundary-layer model. Then
dW
dy (-tan- 1 y) = -ytan- 1 y < O, Y y # O
Hence, the origin of the boundary-layer model is globally asymptotically stable. For IYI ::5: a, we have
dW(
dy - tan
-1 )
y ~-
1
a
-1( ) 2
tan a y
which shows that the origin is exponentially stable. Of course these stability properties are uniform in x
since the bondary-Jayer model is independent of :r.
(c) Let x(t) be the solution of the reduced model starting from initial conditions x1(0) = z2(0) = 0. Let y(r}
=
be the solution of the boundary-layer model starting at y(0} = z{O)-l+:r:1(0) -1. An O(e) approximation
can be obtained by using expressions (11.20)- (11.21) of Tikhonov's theorem. Simulation results are shown
in Figure 11.3.
(d) The conditions of Theorem 11.2 are satisfied.
(e) The system ha.s a unique equilibrium point at x 1 = 1, .:r2 :: z = 0. Applying a change of variables to
shift the equilibrium to the origin, we can rewrite the state equation as
x =Ax+ B(z + Cx), ez = -tan- 1 (z + Cx)
- 169
1.4 0.6 1
1.2 0.8
0.4
0.6
0.8 0.4
)( x"" 0.2 N
0.6 0.2
- Exact 0 0
0.2 - Approximate -0.2
0 --0.2 --0.4
0 5 10 0 5 10 0 5 10
t t
Figure 11 .3: Exercise 11.16.
where
A= [ ~l !1 ] , B =[~ ], C =[ 1 O]
Use the notation of Section 11.5. The matrix A is Hurwitz. Let P be the solution of the Lyapunov equation
PA+ ATP= -1. It is given by P = [ ~:: 0/]. Let V(x) = xT Px and W(y) = ½y2 . Then
av
ax f(x, h(:i:)) = -llxll~ ~ 01 = 1, 'Pl = llzll2
: g(x,y + h(x)) = -yta1C 1 y 5 -a21Yl 2 , Y IYI $ a, where 02 = ~ tan- 1 (a), "'2 = IYI
~~ [f(x,y+ h(x)) - f(z,h(x))J = 2xTPBy ~ 2IIPBll2llzll2IYI = v'sl/:rll2IYI ~ /31 = v15
aw - awah]
[_Bx &y ox f(x , y+h(x)) = yCJ(x,y+h(x)) = yCAx+yCBy = yCAx
:5 UCAll2llzU2b1I = llxll2IYI => fh = 1, "I = 0
Hence, the origin is asymptotically stable for e < c•, where
• a 1 ct2 1 1
e = an+ /31/32 = aVS tan- a
It is also exponentially stable (see Exercise 11.24). Based on this analysis we cannot conclude global asymp-
totic stability.
=
• 11.17 (a) Setting e = 0 yields the equation x -x2 - z 0, which bas the unique root z == x- x 2 ~f h(x).
The reduced model is i =
-x. The bondary-layer model is ~ -y. =
(b) The boundary.layer model is linear and independent of x. Its origin is clearly globally exponentially
stable.
(c) Let x(t) be the solution of the reduced model starting from the initial condition x(0) l. It can =
be verified that x(t) = e- t. Let y(T) be the solution of the boundary-layer model starting at y(O) ::::
=
z(O) - x(0) + x 2 (0) = l. It can be verified that ij(T) e_.,.. An O(e) approximation is given by
x(t, e) = e-t + O(e),
- 170
0.8
0.6 0.6
>C N
0.4 0.4
0.2 0.2
" ..,
"
0
2 4 6 e 10 0 2 4 6 B 10
Simulation results fore= 0.1 are shown in Figure ll.4.

(e) The system has a unique equilibrium point at the origin. Let V(x) = ½x2 and W(y) = b 2• Then
av
fJx J(x, h(x)) = -lxl 2 ==> ·a1 = 1, 'I/J1 = lxl
aw
oy g(x, y + h(x)) = -y2 , ==> a2 = 1, t/J2 = IYI
av
IJx [/(x, 11 + h(x)) - f(x, h(x))] = xy ~ jxllyl ==> /31 =1
[aw awah]
&x - {Jy fJx f(x,y + h(x)) = -y(l - 2x)(-x + y)
< (1 +2a)lxlb,I + (1 ·+ 2a)IYl2 , 'v lxl ~ a, ==> /32 = -y = (1 + 2a)

Hence, the origin is asymptotically stable for e < e• = 0102/(oyy + /31/J,i) 1/(2(1 + 2a)] It is also =
exponentially stable (see Exercise 11.24). Based on this analysis we cannot conclude global asymptotic
stability.
• 11.18 (a) Setting,= Oyieldstheequation-2x4 13 -2z = 0, whichhastheuniquerootz -x4/ 3 ~f h(x). =

The reduced model is x = -x:.. The bondary-layer model is~= -2y.
(b) The boundary-layer model is linear and independent of x. Its origin is clearly globally exponentially
stable.
(c) Let x(t) be the solution of the reduced model starting from the initial condition x(O) = I. It can
be verified that x(t) =
1/(1 + 4t) 114 . Let y(r) be the solution of the boundary-layer model starting at
= =
y(O) z(O) + x 4 13 (0) = 2. It can be verified that ii(r) 2e- 2 T. An O(c) approximation is given by
1
x(t,e) = (l + 4t)1I' + O(c), z {t,e ) -- - (1 +l 4t)3 + 2e -'lt/c + 0(c )
Simulation results for c = 0.1 are shown in Figure 11.5.
(e) The system ha.c; a unique equilibrium point at the origin. Let V(x) = ¼x6 and W(y):::. ½Y2 , Then
8V
OX/(x,h(x)) = -lxllO => Ot = l,t/Jl = lxl5
- 171
1.2 1
- Exact
- - Approximate 0,5
.. 0.8
0.6
N 0
0.4
0.2
0 2 4 6 8 10 2 4 6 a 10
8W
8y g(x,y + h(x)) = -2y2 , => a2 = 2, 1P2 = IYI
av
ax [f (x, y + h(x)) - f(x, h(x))J = x 6 [(y - x 4
13
)3 + x 4]
= x6(y3 + 3yx8/3 _ 3y2x4/3) < xs(y3 + 3yx8/3)
< /311/Ji1P2
f
Y ( xl/3) x(y - x'fs)s
= _ !xlf3yxs + !f,2x'fs(11 2 + 3xs/s _ 3yx•fs)

$ /32'Pl 'IP2 + 'Y'P~
where the last two inequalities hold on compact sets. For example, on the set {lxl :5 a, 1111 :5 b}, the
constants th, /32 and -y are given by
/31 = a(b2 + 3a813 ), /32 = !a1l 3 , !a413 (b2 + 3a813 + 3ba413 )
Hence, the origin is asymptotically stable fore< e• = 0 1 0:2/(0:1 ; + /Ji/32 ). The Lyapunov functions V
and
W do not satisfy the exponential stability conditions of Exercise 11.24. Linearization at the origin yields
the matrix [ ri -~/E ] which is not Hurwitz. Hence, the origin is not exponentially stable. Based on this
analysis we cannot conclude global asymptotic stability.
• 11.19 (a) Setting~= 0 yields the equation -z - x = 0, which has the unique root z = -x ~r h(x) . .The
reduced model is x = -x3 - taiC 1 x. The boundary-layer model is ~ -y. =
(b) The boundary-layer model is linear and independent of x. Its origin is clearly globally exponentially
stable.
(c) Let x(t) be the solution of the reduced model starting from the initial condition .z(O) = -1. Let
=
y(T} be the solution of the boUDdary-layer model starting at y(O) z(O) +x(O} 1. An O(E) approximation =
can be obtained by using expressions (11.20)-(11.21) of Tikhonov's theorem. Simulation results for E = 0.1
are shown in Figure 11.6.
J;
(e) The system has a unique equilibrium point at the origin. Let V (x) = (y 3 +tan- 1 y) dy and W (y) = ½y 2 •
Then
oV f(x,h(x)) = -(x3 + tan- 1 x) 2 => 01 = l,1P1 = lx3 + tan- 1 xi
ax
- 172
0 2
~
1.5
Exact
Approximate
N 1
"'
-0.8
-1
0 2 4 6 e 10 2 4 6 8 10
aw
8y g(x,y+h(x)) = -y2 , ::::} 02 = l,1,b2 = IYI
oV
ox [! (x, y + h(x)) - f(x, h(x))} = (x 3 + tan- 1 x)[tan- 1 (x) - tan- 1 (x - y)]
::S 1/11 "'2 ::::} P1 = 1
[aw ow oh]
ax - fJy ox J(x, y + h(x)) =
= y [-x 3 -tan- 1 x + tan- 1 x - tan- 1 (x -y}]
::S tP1 '¢2 + 1Pi =->- P2 = -r = 1
where all inequalities hold globally. Hence, the origin is globally asymptotically stable for e < g• ½· =
The Lyapunov functions V and W do not satisfy the exponential. stability conditions of Exercise 11.24.
Linearization at the origin yields a HUJ'Witz matrix. Hence, the origin is exponentially stable.
• 11.20 (a) Setting c = 0 yields the equations

0 = .Zt, 0 = - Z2 - (X + Zt + XZ1)
Thus, h(x) = [ _ox ] . The reduced model is x = -2x. The bondary-layer model is
dy -1 0 ]
dr = A(x)y, A(x) =[ -(1 + x) -1
(b) The condition (11.16) is satisfied on any compact set Br- Therefore, the boundary-layer model is expo-
nentially stable, uniformly in x, for all x E Br-
(c) Let x(t) be the solution of the reduced model starting from the initial condition x(0) = l. Let y(r) be the
solution of the boundary-layer model starting at y(O) = [ z2 (~1f ~(O) ] = [ ; ]. An O(e) approximation
can be obtained by using expressions {11.20)-(11.21) of Tikhonov's theorem. Simulation results fore= 0.1
are shown in Figure 11. 7.
(e) The system has a unique equilibrium point at the origin. By linearization, it can be shown that the
- 173
- Exact
0.8 0.5
- Approximate
0
,. N
N
-1
oL---~----.J -1.5
0 2 3 o.s 0 1 2
I t
origin of the reduced system is exponentially stable and the origin of the boundary-layer system is exponen-
tially stable uniformly in x on any compact set of z. It follows from Theorem 11.4 that the origin of the
full system is exponentially stable for sufficiently small e. An upper bound c'" can be calculated by using
the Lyapunov functions V(x) = ½x2 and W(x,y) = yTP(x)y where P(x) is the solution of the Lyapunov
equation P(x}A(x) + A7 (x)P(x) = -1. ·
• 11.21 (a) The equilibrium points satisfy the equations
U = R1l1, Va = c1l1fl + Raio, 0 = c2I1lo - c30
From the first equation, we obtain 11 = U/R,. Substituting for O from the second equation into the third
one, we obtain
This equation has the unique root
l _ csV11 => n- c2V11 U/R1

0
- c3Ra,+c1c2U2/R} - c3.R,,+c1c2U2/R'
(b)
Ta dz Ta dia. 1 [ c1l1!l 1 ]
eZ = TJ Ti dt ~ Io dt :::; lo -la.Z - ~ X 1 X 2 + Ra, Va
c1I1n Va c1c2U 2 Va
= -z - IoRa. X1X2 + IaRa. = -z - cs.R,,R} X1X2 + IoRa
(c) Setting e =0 yields O= -z - bx1x2 + c. Hence, h(x) = -bx1x2 + c. The reduced model is
z1 = -xi + u, ±2 = -ax2 - abz~x2 + ac:t1
The boundary-layer model is ~ = -y.
(d) The boundary-layer model is linear and independent of x. Its origin is clearly globally exponentially
- 174
stable.
(e) Use expressions (11.20)-(11.21) of Tikhonov's theorem.

=
(f) For any constant input u{t) Uo, the reduced system has a unique equilibrium point. Using Lineariza-
tion, it can be shown that this point is exponentially stable. Hence, the conditions of Theorem 11.2 are
satisfied.
(g) Simulation results fore= 0.2 are shown in Figure 11.8.
1.2~----~~
0.8 -- 2
1.5
0.6
)( ,J"
- Exact
- Approximate
0.5
0 0
0 2 4 2 4 0 2 4
=
• 11.22 Setting e = 0 yields z -x/a. The reduced model is i = -11(x)-x and the boundary-layer model
is dy/dr = -y. Let V(x) = ½x2 + foz fJ(u) dn and W(y) = ½y 2 • Then
av
axf(x,h(x)) = -lx+77(x)l 2 , V x E (-oo,bJ, ~ 01 = l,1J,,1 = lx+77(x)I
aw
8y g(x,y+h(x)) = -y2 , ~ 02 = l,1J,,a = IYI
av
ox lf(x, Y + h(x)) - f(x, h(x))] = axy ~ a'¢1'1P2 ~ fj1 =a
[aw aw ah] 1
8x - 8y 8x f(x,y + h(x)) = ~y[-71(x) + ay - x] $;
1
~/Yllx + 71(x)I + IYl2 ~ /12
1
= ~,-Y =1
Hence, the origin is asymptotically stable for all e < e• = ½.
• 11.23 (a) LiDearization at the origin yields the matrix [ ~ -~/e ] which shows that linearization fails.
(b) Setting t: = 0 yields z = tan- 1 x 3 . The reduced model is :i: = -2x3 + (tan- 1 :z: 3 ) 2 and the boundary-layer
model is * = x 3 - tan(y + tan- 1 x 3 ). Let D = {lxl ::; a and IY + tan- 1 (x3 )1 :$ b < 1r/2} All our analysis
will be restricted to the set D. Let V(x) = ¼z4 and W(y) = ½Y2 . Then .
where we have used the property z(tan- 1 z) 2 ~ z 2 .
aw
8y g(x, y + h(x)) = yx 3 - y tan(y + tan- 1 x 3 ) ~ -y2 ==> 02 = 1, 1/J2 = IYI
- 175
where we have used the property z(tan(z + b) - tan bJ ~ z2 .

av _1
ax [f(x, Y + h(x)) - f(x, h(x))] = x 3y(y + 2tan x3 ) $ P1 1xl3 !yl
[-aw - ~~
aw 8h]
f(x,y + h(x)) = - -
3x2y -
-6 [-2x 3 + (tan 1 x3 )2 + 2ytan- 1 x 3 + y2] $ .821xl 3 Jyl + 1'1YI
& ~~ l+x
where /31 , /32 and "I depend on the constants a and b of the set D. Thus, the origin is asymptotically stable
for all e < e., where e* is given by (11.46). The Lyapunov function is v(x,y) = (I - d*)V(x} + d·W(y),
where d" =
fh/(/31 + /h). The region of attraction can be estimated by the set {v(x,y) ~ c}, where
c = minav{v(x,y)}.
• 11.24 The inequalities satisfied by V and W imply that v(x,y) = (1 - d}V(x) + dW(y) satisfies the
inequality
for some positive constants c1 a.nd c2, When 't/!1 {x) = llxll and '!Ji2(y) = IIYII, the inequality i, $ -,µT A'ljJ, for
some positive definite matrix A, implies
for some positive constant c3. Then, by Theorem 4.10, the origin is exponentially stable.
• 11.25 (a) By smoothness, there is a positive constant L such that
11 = (1 - d}V(x) + dW(y)
av d8W
;., = e
(I - d) ax /(x, y) + l)y [Ay + egi(x, y)j
= (1 - d) 08V f(x, 0) + (1 - d) ~V [f (x, y) - f (x, O)] - ~YT y + 2d1? Pg1 (x, y)

x v:i: e
~ -(1 - d)a1</J + (1 - d)k2(/)112 IIYll2 - ~y·TY + 2dkillPll24'112 IIYll2 + 2dLIIPJl:idlYII~
c
= 2 d 2 2
-(1 - d a1¢1 + (1 - d)/J1'1/J11h + df32t/J11/J2 - -'lj.,2 + d--y'l/)2
)
£
where /31 = =
k2, /32 2k1 IIPl'2, 1' = 2LIIPll2, ¢1 = ¢,112 , and ""2 = IIYll2• Hence, the origin is asymptotically
stable for sufficiently small e.
(b)
v = V(x) + (YT Py)..,

it = ~~f(x,0)+ ~:[f(x,y)-J(x,O)}+-y(7,,-TPyf"-I [-¾yTy+2yTPg 1 (x,y)]
$ -01</, + k4</J''IIYll2 - ~ (Amin(P))..,_ 1 111111~-r + 'Y (Amu(P))i'-l 1111/1~-,- 2 X 2llvll2IIPl'2(k3q,C + Lll11ll2)

= -a1<J, + k,¢1tllYII~ - 2£ (Amtn(P))-,- 1 IIYII~.., + 2-yk3IIPll2¢> IIYll~-,-l + 2')'L/IPll2ll!1ll:-r
0
Using Young's inequality with p = 1/b and arbitrary µ 1 > 0, we obtain
¢l'IIYll2 $ :1 + µtl(b-l)IIYll~'Y
- 176
Using Young's inequality with p = 1/a = 2"}' and arbitrary µ,1 > 0, we obtain
Hence,
where
k4 2')'k3JIPll2
C1 = 01 - -
µ1
-
µ2
C2 = 1'. (.Amin(P))"'-l - k4µ:/(b-l) - 2')'k3IIPl/2µ~/(2-y-l) - 2"}'LIIPll2
e
Choose µ 1 and µ2 large enough to make c1 > O; then choose e• small enough such that c2 > 0 for all e < e•.
( c) Consider t\Je second-order system
x= -x+y, ci):;;: -y + eg1 (x)

Part (a) requires 191 (x)j :S: k1 lxl, while part (b) requires jgi(x)I :S: k31xl 20 , 0 <a< ½- Ta.lee 91 (x) = vfxT.
• 11.26 Let y == z + A- 1 71(.x) and rewrite the equation as
x = f(x,y- A- 1 77(x)), eD'[j = Ay + eDA- 1 8x

81]
The derivative of v(z,y) = (1- d)V(x) + dyTPDy along the trajectories of the system is given by
i,
av _ av _
= (1 - d) Bx J(x, -A 1 71(x)) + (1 - d} ax [J(x, y - A 1 11(.x)) - J(x, -A- 111(x))]
- ~yTQy + 2dyT PDA- 1 8
8rJ f(x, y - A- 1 17(.x))
E X
Suppose that
8V
8xf (x,-A- 1 17(.:z:)) :S: -0:1¢1(.:z:)
2
av _ _1
ax [f(x,y-A 1 11(x)) - f(x,-A 11(.x))J :S: ~11/Ji(x)ll11ll2
11::f(x, y -A- 111(x))"2 :S: k11P1 (x) + k2ll!1!12 ~ 21,,TPDA- :J(x,y - 1 A- 1 r,(x)) $ i92t/Ji(x)IIYll2 + "YIIYII~
where the constants /32 and I are chosen to independent of the elements of the matrix D which is possible
since the elements of Dare bounded by one. Now setting 02 = >.min(Q) and 1P2(y) IIYll2, we see that 'ii =
satisfies the same inequality ii :S: - t/;T A'lj; that appears on page 452 of the text. · Thus, we can continue to
prove that the origin is asymptotically stable for sufficiently small e.
• 11.27 Setting e =0 results in z = -x1x2. The reduced model is
The boundary-layer model is*= -y. We analyzed the reduced model in Example 4.10. There we saw that
the reduced model has an equilibrium set {x1 = O}. Using the Lyapwiov function V{x) = ½x~ + (1/2b){x2 -
k) 2 , k > a, we applied LaSalle's theorem to show that all solutions approach the equilibrium set as t --too.
In the current singularly perturbed system, it can be checked that {x1 = 0, z = O} is an equilibrium set.
- 177
We will prove that all solutions of the singularly perturbed system approach this set as t -+ oo. Apply the
change of variables y = z + x1 x2. Then,
X1 = -(x2 - a)x1 + 2y
.
X2 = bX12
c'f; = -y + c[-(x2 - a}x1x2 + 2yx2 + bx~

Let W (y) = ½Y2 . The following inequalities can be easily checked.
ov .
ax f(x, h(x)) == -(k - a)x~ = -n-1,Ji?(x), where o 1 = k - a> 0, ,j,1 (x) = lx1 )
aw
8y g(x,y + h(x)) = -y2 = o-2t/J~(y}, where 02 = 1,1/¼ = IYI
av
-8 [/(x, Y + h(x)) - f(x, h(x))] = 2x1y <- /J1¢1(x)'¢2(y), where f)i =2
X
[8W 8W8h]
lJx - 8y ax J(x,y + h(x)} = -yx1x2(x2 - a)+ 2z2y2 + bx~y :$ ,82v,1(x)¢2(y) + 1'?Pi(Y)
where /32 and -y are positive constants such that the domain of interest is a subset of the set D defined by
D = {2x2 :S "Y and lbxf- x2(x2 - a)I :S '92}

By comparison with the proof of Theorem 11.3, it is now clear the the derivative of 11(x, y) = (1- d)V(x) +
dW(y) along the trajectories of the system satisfies the inequality v :$ -,j,7 At/;, where the matrix A is
positive definite for sufficiently small c. Hence
ii~ -,\('¢r(x) + "1i(Y)), ;\ > 0

The difference from Theorem 11 .3 is that in the current case the foregoing inequality shows only that ii is
negative semidefinite because 1P1(x) =
l:ril is not a positive definite function of .z; that is, it vanishes at
=
x 1 0 for all values of x2. It is clear though that all solutions starting in {v(x,y) :$ c} CD will remain in
this set for all t 2: 0. Moreover, by Theorem 8.4, we conclude that x 1 (t) and :r,(t) tend to zero as t -t oo.
=
Consequently, z(t) y(t) - x1(t)x2(t) tends to zero as t-+ oo.
• 11.28 View the given system as a perturbation of the system
±1 = x2, ±2 = -x2 + z, ci = -(:z:1 + z) - (x1 + z)3

The perturbation term satisfies je-tzl :$ e-tlzl. It follows from Corollary 9.1 and part 2 of Lemma 9.5 that
if the origin of the nominal system is globally exponentially stable, so is the origin of the perturbed system.
Thus, we only need to show that the origin of the nominal system is globally exponentially stable. We use
the singular pertmbation stability analysis. Setting e: = 0 results in z = -x1 . The reduced model is
This is a linear time-invariant system with a Hurwitz matrix A = [ ~l _:1 ] . A Lyapunov function
=
V(x) can be taken as V(.:r) xT Px, where P is the solution of the Lyapunov equation PA+ ATP -I. =
= = =
Inequality (11.39) is satisfied with 'I/J1 (x) Uxll2 and o 1 1. Define y z + x 1 . The boundary-layer model
is ~ = =
-11 - y 3 • The function W(y) ½:r, 2 satisfies
&W(
8y -y - y 3) = -y2 - y4 ~ - y2
- 178
Inequality (11.40} is satisfied with T/J2{y} = IYI and a2 = l. Let us verify the interconnection inequalities
(11 .43) and (11 .44).
av
ax [f(x, y + h(x) - f(x, h(x))] = 2xTP [Y o] ~ 2IIPBll2llxll2IYI
- ~
uy
aah f(x,y + h(x) = -y [ -1
X
0] [
-X1 -
x2X2 + Y ] = YX2 ~ 1Jxll2IYI
where B = [ ~ ] . Thus, all the conditions of Theorem 11.3 are satisfied globally with 01 = 1, 02 = 1,
fJ1 = 2JIPBll 2 , /32 = l, and '"Y = 0. Since V(x) and W(y} are quadratic functions and ¢i(x) = llxll2 and
1h(11) = IYI, we conclude that the origin is globally exponentially stable for O< c < c· = l/2IIPBll2, Notice
that we cannot conclude global exponential stability from Theorem 11 .4 because the term (x 1 + z) 3 is not
globally Lipschitz.
• 11.29 (a) Note that tan- 1 (z) is globally Lipschitz with a Lipschitz constant equal to one because
d _ 1
d2 tan I(z) = 1 + z2
Withy= x + z, the unforced system is given by
ey = -y + e(-x + tan- 1 (-x + y)l
Let v= ½x2 + ½Y2 -
2
i, = -x2 +xtan- 1 (-x+y)-!..+y(-x+tan- 1 (-x+y)]
€
2
= -x2 - xta.n- 1 (x) + x[tan- 1 {-x + y) - tan- 1 (-x)) - !.. - xy + y tan- 1 (-x + y)
e
2
= 'J/
-x2 -xta.n -1( x ) -7+11 [ X
l+(-x+(y) 2 -x+tan
-l ]
(-x+y), 0<,<l
y2
< -x2 - -
e
+ IYI lxl + lvl I - x + YI
< [ !xi ] T [ I -1 lxl ]
] -[
- IY! -1 ¼- 1 IYI
The 2 x 2 matrix is positive definite fore < e• = ½-
Since v and the upper bound on ii are quadratic and
the change of variables y = z + x is linear, we conclude that the origin is globally exponentially stable for
all c < ½,
(b) Input-to-state stability follows from Lemma 4.6.
• 11.30 (a) The dosed-loop system is given by
Setting c = 0 results in z = -1JJ(2x1 + x2) = h(x). The reduced model is
.:i:1 = x2, ±2 = -z1 - 2x2 - ¢(2x1 + x2)

The boundary-layer model is ~ -y. =
(b) The reduced system can be represented as a feedback connection of the linear system
- 179
and the nonlinear function u = -'lj,(y). The transfer function of the linear system is G(s) = (s + 2)/(s + 1)2 .
For any k > 0, the transfer function 1 + kG(s) is strictly positive real because
1 + kRe{G(jw}] = 1 + (l : :2 ) 2 > 0, 'r/ w ER
and 1 + kG(oo) =
1 > 0. Thus, by Theorem 7.1, the system is absolutely stable for any finite k > 0.
Moreover, a Lyapunov function is given by V{x) = ½xT
Px, where P satisfies equations (7.6)-(7.8). We
replace e in (7.6) byµ > 0. Thus, V(x) satisfies (11.39) with a1 = ½µAmin(P) and ¢1(x) llxll:i- For =
the boundary-layer system, W(y) ;;:; ½v2 satisfies (11.40) with a2 =
1 and t/i2(y) = lYI- Let us check the
interconnection conditions (11.43) and (11.44).
!: !J(x, y + h(x)) - J{x, h(x))J = xTP [ ~ ] ~ IIPBll2llxll2IYI, where B =[ ~ ]

Thus, {11.43) is satisfies with /31 = IIPBll2-
8W 8h ,
- 8y lJx f(x,y + Ji(z)) = ytp (2x1 + X2}[2x2 - X1 - 2x2 + y -t/J(2x1 + x2)]
Suppose that 1J,' ( ·) is globally bounded. Then,
=
where Lis a Lipschitz constant for(/>. Hence, (11.44) is satisfied with /32 = L + i:i..;g and 'Y L. Thus, all
the conditions of Theorem 11.3 are satisfied globally and we can conclude that, for sufficiently small e, the
origin is globally asymptotically stable; that is, the system is absolutely stable. If '¢ is not globally Lipschitz,
it will be Lipschitz on any compact set, due to smoothness. In this case, we can conclude absolute stability
with a finite domain.
- 180
Chapter 12
• 12.1 We represent the closed-loop system as a perturbation of its linea.rization at the origin.
:i = (A - BK)x + Bg(x1)
where
A-BK= [ -l0( 2_5: l/v'2) -~o], B = [ ~], g(xi) = -10 (sin (x1 + i)- ~- ~]
It can be easily seen that lg(x1)I ~ 5xf We use V(x) = xT Px as a Lyapunov function, where P =
[ ~:~~: ~:~~~: ] is the solution of the Lyapunov equation P(A - BK)+ (A - BK)T P = -I.
V(x) = -xTx+2xTPBg(x1) ~ -llzll~+2llzl'211PBll25xf
~ -[l - 0.5386lz1 IJllxll~ < 0, V lx1 I < 1.8567
Hence, V(x) is negative definite in the region {lzd < 1.8567}. We estimate the region of attraction by
rle = { V(x) ~ c} C {lx1I < 1.8567}. It can be verified that
min {zTPz} = (l.S5671)2 } = 6.2216

lz1f=l.8S67' LTP- L
where LT= [I OJ. Thus, we estimate the region of attraction by the set {V(x) 5 6.22}.
• 12.2 (1) Linearization at the origin yields
(a) The state feedback control u = -Kx = - [7 4 ) x assigns the closed-loop eigenvalues at -1 and -2.
(b) Use the observer-based controller
i=(A-BK-HC)x+Hy, u=-Kx
with K as in part (a) and H =[ ~i ], which assigns the observer eigenvalues at -5 and -6.
( 2) Linearization at the origin yields
1 1 0 0
A= a1 (0, 0) = [
ax ~I ~ ~ ] , B af (0, 0) = [
= 8u ~] , C ah (0) = [ 0
= ax I O]
-181
(a) The state feedback control u = -Kx = - [ 16 17 7 ] x assigns the closed-loop eigenvalues at -1,
-2, and -3.
i::::: (A - BK - HC)x + Hy, u = -Kx
with K as in pa.rt (a.) and H = [

-335
19
-210
l , which assigns the observer eigenvalues at -5, -6, and -7.
( 3) Linearization at the origin yields
A= aJ
ax (0,0) = r- ~
1
l
~l ~O2 l , B = 8f
8u (0,0) = [ ~0 l , C = 0/Jh/0) = [ 1 0 O)
(a) The state feedback control u = -Kx = - [ 1 2 0 ] :r assigns the closed-loop eigenvalues at -1 , -2,
and -3.
¼= (A-BK-HC)x+Hy, u =-Ki
with K as in part (a) and H = [ {1 l· which assigns the ohse,ver eigenw.lues at -5, -6, and -2.
• 12.3 Use the PBH controllability test.
rank [ sln+p - A B ] = rank [ sl~C A 0

sl,, ~]
sf,. - A
For s # 0, rank [ -C
0
sip ! ]= + p rank [ Bin - A B]
For s =0, sin -A

rank [ -C
0
sip !]=rank[~
Hence, (A, B) is controllable if and only if (A, B) is controllable and the rank condition (12.23} is satisfied.
• 12.4 (a) A state feedback integral controller is designed in Example 12.4. For output feedback, use the
observer
i = (A - HC)x +Bu+ Hy
where y =8 - 6, and replace iJ by i2 in the control law. The observer gain His designed such that
A - HG= [ -hi 1 ]
-h2 - a coso -b
is Hurwitz. Using the Routh-Hurwitz criterion, it ca.n be shown that A - HG is Hurwitz if
h1 > -b, h2 > -acos6 - h 1 b

(b) A state feedback integral controller is designed in Example 12.4, for a. fixed (),. = 5, as
ti= -k1 (6)(8 - o) - k2(6)8 - k3(o)a, (1 =9 - 0
o.
where the gains k1, k2, and k3 are, in general, functions of In a gain scheduled controller, ois replaced by
the scheduling variable 8r- To ensure that the linearization of the closed-loop system at each operating point
- 182
is the same as the feedback connection of the parameterized linear system and the corresponding controller,
we modify the gain scheduled controller to
In this modification, we kept the first two terms since both () - Or and 9 vanish at the equilibrium point.
(c) The observer-based integral controller designed in Exercise 12.4 and part (a) is given by
u = -ki(<>)(e - <>) - k2(6).x2 - k3(o)CT, a= e - c5
i = [A(h) - H(c5)C].x +Bu+ H(c5)y = (A(c5) - BK(6) - H(c5)C]x - Bk3(0)11 + H(tS)y

The controller takes the form of the original block diagram of Figure 12.4 with F = A - BK - HG,
G = [-Bk3 H], L = (0 - k2], M = [-k3 OJ, M3 = -k1, and Ym = y. Applying the modification of
Figure 12.4, we obtain the gain scheduled controller
i = [A(8r) - BK(Br) - H(Or)C]z - Bks(IJ,.)(8 - Br)+ H(Or)il

V = -k2(8r)Z2 -ka(Or)(O-Br)
1J = V
u = 11- k1(8)(8 - (),.)
where y is obtained from y =8 - 8,. by using the filter (12.55)-(12.56).
• 12.S (a) Since c 2 cr :5 e-2c :5 1, 'v O:5 u :5 T, we have

11
Q(a) = for e- c" exp(-A(o)uJB(o)BT(a}exp[-AT(o:)u) du

2
> e-2 ",. l,. exp[-A(o)u]B(o)BT (o) exp[-AT (o)o) de, = e- 2CTW(a) ~ e-2cr c1 /
Q(a) = 1,. e-2co- exp[-A(o)u]B(a)BT(o)exp[-AT(o)u) du
< Lr exp[-A(a)u]B(a)BT (o) exp[-,-AT (a)<1) da = W(a) :$ c2/

(b)
V = :r;T { P(o)[A{a) - B(a)K(a)J + [A(a) - B(a)K(a)f P(o)} x

= xT{P(a)A(a) + AT(a)P(a) - P(o)B(a)BT(a)P(a)Jx
= :r;TP(a)[A(a)Q(a) + Q(a)AT(o:) - B(a)BT(a)}P(a)x
A(a)Q(o) + Q(a)AT(a) = 1,. e- 2 co- A(cr) expf-A(a)u)B(a)BT (a) exp[-AT(o:)11] du
+ la,. e-2c., exp[-A(a)u]B(o)BT (a) exp[-AT(a)u]AT (a) du

= - Jo r e- d
2ca do {exp[-A(a)a]B(a)BT(a)exp[-AT(a)aJ} dtT
Integrating the right-hand side by parts, we obtain
- 183
Therefore
A(o)Q(o) + Q(a)A 7 (o) - B(a)BT (a) :5 -2l'Q(o}
Hence
{c)
V ~ -2cV
where k1 = c1 e-zc.- and k2 = c2. From the comparison lemma,
V(t) S: e- 2ctV(O)
Jlx(t)ll2 ~ vk-y'V(t) ~ :;;VV(O) ~ ~e-ctlfx(O)ll2
• 12.6 Since P is block triangular, it is nonsingular if and only if [ ~: f] is nonsingular. By inter-
changing rows, the latter matrix is nonsingular if and only if [ ¾: ~ ] is nonsingular. It is argued on
page 484 that [ -:t: ~ ] is nonsingular. Hence, P is nonsingular. Verifying that P satisfies (12.51) is
straightforward.
= = =
• 12. 7 Let x 1 1/J - 1/J,., X2 ,,ji, and u 6. Augment the state equation with the integrator i, = x 1 • Define
x =
[x 1 , x 2 , o-JT. The augmented state model is given by :r = Az + Bu, where
A= [0 -1/r
1 0]
0
1 0
O
0
,
B= [ +]
where 1/r = (1/ro)(v/vo) and k/r = ('k-O/-ro)(v/vo) 2 . The state feedback control
u=-12(i)-[9(i)-(i)]-6(i)
assigns the closed-loop eigenvalues at -7.5082, -0.7459±0.4927j. the closed·loop transfer function from 1/).,.
to 'I/,, is given by
6(2s + 1)
s 3 + 9s2 + 12s + 6
and has a step response with 20% overshoot and 6.4 sec. settling time. A gain scheduled controller can be
taken as
Since the scheduling variable is not the reference input, we don't need the modification of Figure 12.4.
• 12.8 (a) The equilibrium equations are
k _ Loax~
0 =a: g - -x2 - 2 ( _) ,
m m a+x1 2
Set X1 = y.. = r, X3 = las, ii= 1.1.u = v.. , and Xas = x. Then,
lss = (2mg(a + r)2 ) 112

,
Vu =lu
Loa
- 184
(b)
8/l
-
ox
[
.z:=.z:..
0
C1
O
1
where the positive constants c1 t hrough cs are given by

Loa/;. k Loalse Loalas R
C1 = m(a+r)3' C2 = -,
m
C3 = m(a + r} 2 , C4 = L(r)(a + r} 2 , cs= L(r)
The characteristic equation
s 3 + (c2 + c5)s 2 + (c2c5 + c3c4 - ci)s - c1 cs =0

has a negative coefficient -c1c:s. Hence, by the Routh-Hurwitz criterion, the matrix has eigenvalues in the
right-half plane. The equilibrium point is unstable.
(c) The linearized system is given by
Xo = Ax.s + Bu0
=
where x 0 : x - Xu, u11 u - ua, A is given in part (b), and B =
(0, 0, 1/L(r)JT. It can be verified that
(A, B) is controllable. K is designed to assign the eigenvalues of (A - BK) at -10, ·_ 10 ± jlO.The control
u is given by
=
u 62.9295(.xi - r) + 4.4432x2 + 0.2502(x3 - lu) + V.1
where V.1 = lu = 6.2642.
=
(d) The step response of y and u for y(O) 0 and y{O) = 0.07 (with other initial states equal to zero)
are shown in Figure 12.1. To account for the constraint O ~ u ~ 15, a limiter is included in the Simulink
simulation model. The response is considered feasible only if y belongs to the interval [O, 0.1]. Using this
criterion, 0.07 is the largest acceptable initial position.
\'(O)•O y(O) •O
a.-----------,
0.01 2
1
0.2 0.4 0.8 0.8 0 0.2 0.4 O.B 0.8
y{D) • 0.07 y(O) •0.07

0.1 11
10
0.08 II
>-0.07 ., 8
0.08 7
0.05 8
0.04 5
D D.2 0.4 0.8 0.8 0 0.2 0.4 0.6 0.8
(e) Figure 12.2 shows the response for different values of m. These values are the extreme values for which
- 185
y(O) =0, m =0. 1 y(O) = 0.07, m = 0.1

0.06 0.1
0.04 0.08
>, >,
0.02 0.06
0.04
0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 1
y(O) "" O. m -= 0.046 y(O) = 0.07, m =0.046
0.02 0.08
0.06
>- 0.01 >-0.04
0.02
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8

y(O) • 0, m = 0.11 =
y(O) z 0.06, m 0.11
0.08-------.--- ..------ 0.12
0.1
>,
0.08
0.06
0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8
t
Figure 12.2: Exercise 12.8 .
a feasible response is obtained for the given initial position. There is a large steady-state error.
(f) The integrator iT = x1 - r is augmented with the state equation. A matrix K = lk1 k2 k3 A:,.] is designed
to assign the eigenvalues of
0 1 0
[ 196.2 -0.01 -3.1321
k1 k2 + 12.5284 k3 -40
1 0 0
at -8, -10 ± jlO, -12. The control is given by
u = 109.8789x1 + 6.805x2 + 0.00025x3 + 153.25220'

The step response is shown in Figure 12.3 for the nominal mass and in Figure 12.4 for different values of
m. It is clear that the control achieves zero steady-state error despite the perturbations in m. The transient
response is, in general, worse than the design without integral control. For example, in the case m 0.1 =
and y(O) =0, the integral controller has about 60% overshoot, compared with 10% overshoot in the ca.se
without integral control.
The solutions of the following four parts do not repeat the simulations of parts (d) and (e).
The simulations should show a trend similar to what we have seen before.
- 186
y(O) -o y(O) • O
0.1 0.1
0.08 0.08
0.lll)
::,
""
0.04
0.02 0.02
0.2 o., 0.8 0.8 0.2 0.4 o.s 0 .1
y(O) ~ 0.09 y(D)•0.09

0.12 14
0.1
0.08
::,
"'
0.08
0.0<l
0.02
0 0.2 u 0.6 0.8
(g) Use the observer-based controller

_ i =Ax+ Bu+ H(y - Cx), u = 62.9295(x1 - r) + 4.4432x2 + 0.2502(x3 - I•• )+ V.ss
where C = (1 0 0] and H = [ 9;g_8 ] a.ssigns the eigenvalues of (A - HG) at -20, -20 ± j20.
4857.7
(h) Use the observer-based controlJer
where C
i;;:; Ax+ Bu+ H(y - Ci),
=[ l O O] [ 2.2405 3.0131
0 0 1 and H = 201.2620 60.2870
3.0131 -0.2505
l
u = 62.9295(x1 - r) + 4.4432x2 + 0.2502(x3 - In)+ V.,
assigns the eigenvalues of ( A-HC) at - 20, -20±
j20.
(i) Use the observer of part (g) and replace x2 and x 3 in the control law by their estimates z2 and .x3 .
(j) Use the observer of part (h) and replace x2 in the control law by it s estimate x2 .
(k) For state feedback, we solve the pole placement problem at r = 0.03, 0.05, and 0.07 to assign the
closed-loop eigenvalues at -8, -10 ± jlO, -12. The resulting feedback gains are
Ko3 = [ -105.4498 -5.9167 0.0497 -128.7319 ]
Kos = [ -109.8789 -6.8050 -0.00025 -153.2522 J
Ko1 = [ -115.3518 -7.6934 -0.0336 -177.7726]
We use linear interpolation to determine K for other values of r:
Kr = Ko3 + (r ~.i;Ol) (Kos - Koa), for 0.03 ~r S

0.05
K,,. = Kos+ ( r ~.~;05 ) (Ko1 - KosJ. for 0.05 ~ r S 0.07
- 187
y(0) =0, m =0.1 y(O) = 0.09, m = 0.1

0.1
>-o.os
0 - -.........- - - - ~ ~ - - - - ' - - - - - - '

0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y(O} = 0, m = 0.025 y(O) :: 0.09, m = 0.025
0.06
0.04
>,
>-0.05
0.02
0.5 1 1.5 2 0.5 1 1.5 2

y(O) = 0, m = 0.115 y(O) "" 0.08, m "" 0.115
0.1
0.08
>-0.06
0.04
0.02 .___ _ _ _ _......__ _ _.______
1 1.5 2 0 0.5 1 1.5 2
t t
Linear interpolation does not guarantee that Kr will stabilize the system at the corresponding value of r .
Therefore, we calculate the closed•loop eigenvalues for r =
0.03, 0.035, 0.04, 0.045, 0.05, 0.055, 0.06, 0.065,
and 0.07. The eigenvalues, plotted in Figure 12.5, indeed lie in the neighborhood of the desired eigenvalues.
By the continuous dependence of the eigenvalues of a matrix on its parameters, we expect the same to he
true for other values of r. For the observer, we solve the pole placement problem at r = 0.03, 0.05, and 0 .07
Ho3
.
=
[ 2.1700
248.8284
2.53571.7247
2.5357
49.4564
l [
to assign the eigenvalues of (A - HG) at -20, -20 ± j20. The resulting observer gains are
, J:lor, =
2.2405
201.2620
3.0131
3.0131 ]
60.2870
-0.2505
, Ho7 =
[ 2.3229
170.2966
3.4876
3.4876
70.7955
-1.7122
l
We use linear interpolation to determine H for other values of r:
Hr = Ho3 + ( r -0.03)
0 _02 (Hor, - Ho3), for 0.03 $ r $ 0.05
H,. Ho5 + (r ;_g; 0 5 ) (Ho7 - Hor.), for 0.05 $ r $ 0.07
The observer eigenvalues are checked for r = 0.03, 0.035, 0.04, 0.045, 0.05, 0.055, 0.06, 0.065, and 0.07. and
-188
plotted in Figure 12.5.
RegulalOI' Eigenvalues ot>Mrver EigenYalues

15 . - - - - - - . . . . . - - ~ - - - - - , 90~~-~----,.--,..._-~
10 20
i 5 J 10
~ 0
i- 0 .!i
a
! -5 j -10
-10 -20
-15,..__---~-------' -30'---~----~--'
-14 -12 -10 -8 -6 -20 -15 -10 -5 0
R8aJ Axis Real Axil
Figure 12.5: Exercise 12.8. Closed-loop eigenvalues for different values of r .
(I) y = x 3 = [O, 0, l]x = Cx. It can be checked that (A, C) is observable. Thus, we can design a. linear
output feedback control law to stabilize the ball at y = r. We cannot design an integral controller because
one of the conditions of integra.l control is that the controlled output (x1 in this case) should be measured.
• 12.9 (a) The equilibrium equations are
0 = -01z1 - 82ru,.. + 83, 0 = -94r + 95X1Uu
Substituting .t1 from the first equation into the second one, we obtain
2
u 81 - au11 + b =0, where a -- ~, b -- 8184
82r 8285
< 8l8ri
=> 4b < a2
T2
4l/i928.c
Hence, the quadratic equation has two real roots. To obtain the steady-state value of :i:1 in the range
x 1 > 93/281, we choose the real root for which u.a < a/2; that is u.1 = (1/2) {a - v'a2 - 4b]. Linearization
at the chosen equilibrium point results in the matrix A = [ 8~:: -~~~- ] . It can be verified that A is
Hurwitz. Therefore, the equilibrium point is exponentially stable.
(b) The maximum value of r is ../(9i85/4l}i9284 ) = 298.14. We simulated the step response for increasing
=
values of r in the range O $ r $ 298 and with initial conctitions .z1 (0) Bs/91 0.6667 a.nd x2(0) = 0. The =
motor always reached steady state at the desired speed, without x 1 violating the constraint x 1 > 83 /291 :;;;-
0.3333.
(c) Changes in the rotor inertial change the constants 84 and 8 6 , which are inversely proportional to the
moment of inertia. It is clear, however, that multiplying 94 and 85 by the same constant does not affect the
steady-state calculation. Simulation confirmed that the motor always reached steady state at the desired
speed for O $ r :S 298.
(d) Linea.rization at the desired equilibrium point results in the pair (A, B) where A is given in part (a) and
B = [ ~~~~r l·The controlled output is y = X2 => C = [O, 1).
! ]= rank [e~;:
The determinant of this 3 x 3 matrix is given by
det(·) = 8185£1 - 828r,u81 r = 85((J3 - 282uur)
- 189
Substituting for u.., using its expression from part (a), we obtain det(,) =
9582rJa2 - 4b =;I: 0. Therefore,
the rank condition (12.23) is satisfied. It can be also verified that (A, B) is controllable. Now Set A =
[i ~ ], = [ ! ]
B and proceed to design K such that A - BK is Hurwitz, using any pole placement
algorithm.
(e) It can be verified that (A, C) is observable. Design H such that A- HC is Hurwitz a.nd use an observer-
based controller.
• 12.10 (a)
Xt -' X2
±2 = 0 == A~B) [(m + M)mgLsinB -mLcosB(F + mLiJ2sinB-ky)]

l
= A(xi) ((m + M)mgLsinx1 - mLcosx 1 (u + mLx~ sinx 1 - kx,)]
X3 = X4
i:4 = 'ii== A~B) [-m L gsinllcosO+(I+mL

2 2 2 )(F+mLtPsin8-ky)J
= A(~i) [-m2L 2 gsinx1 cosx1 +(I+ mL2 )(u + mLx~sinx 1 -kx4}]
(b) Set u = 0 and±,= O.

0 === x2
0 == (m + M)mgLsinx1 - mLcosx 1 (mLx~ sinx1 - kx,.)
0 = X,a
0 = -m2 L 2gsinx1 COSX1 +(I+ mL 2)(mL.fi sw.f1 - kx,a)
The equilibrium points are given by (.f1, 0, xa, 0) where x 1 = 0, :1r, .. ., and £3 is arbitrary.
(c) Take x1 = 0. Linea.rization at x = x and u = 0 results in the matrix
[~~~.!]
0
-C3
0 0
0 0 -C4
1
where the positive constants c1 to c, a.re given by

(m + M)mgL mLk m 2 L 2g k(I +mL 2 )
c1 = .:1(0) , c2 = A(O), cs = A(O) , C. = A(O)
The characteristic equation is given by
s [s3 + C4S2 - C]S + C2C3 - C1c,) =0
From the Routh-Hurwitz criterion, we can see that the matrix has eigenvalues in the right-half plane. Hence,
the equilibrium point is unstable.
(d) The linearized system is given by

X6 = A:c6 + Bu6
=
where xs x - x, U6 = u, A is given in pa.rt (c) and B =
[O, -mL/A(O), 0, (J +m£2 )/A(O)JT. It can be
verified that (A, B) is controllable. Proceed to design K to stabilize the matrix (A+ BK) using any pole
placement algorithm.
- 190
Chapter 13
• 13.1 (1)
Y = X1
'!/ = X2
ii = -al(l + X3) sin(x1 + 6) - sin6] - bx2
y< 3> == -a(l + X3) cos(x1 + &) · x2 - b{-a[(l + X3) sin(x1 + &) - sin o) - bx2}
- a sin(x1 + 6){ -cz3 + d[cos(x1 + 6) - cos 6) + u}
Let Do = {x E R 3 I O < x1 + o < 11'}. The system has relative degree 3 in Do- It is transformed into the
normal form via the change of variables
which is invertible in Do.

(2)
y = Xl + ')'X2
y = x2 + -y{-a[(l + x3) sin(x1 + 6) - sin o) - bx2}
jj = --ya(l + x3) cos(x1 + 6) · :i:1 + (1 - lry}±2 - ')'a sin(x1 + 6) · :i:3
= --ya.z2(l + Z3) cos(x1 + &) + (1 - lry){-al(l + xa) sin(x1 + 6) - sin6] - bx2}
- -yasin(x1 + '5){-cx3 + d(cos(x 1 + 6) - coso] + u}
The system has relative degree 2 in Do, We have
Find <f>(x) such that [8¢/&x]g = 0 and T(x) = [

</>(x)
h(x)
L1h(x)
l is invertible in D 0 . It can be verified that
<J,(x) = x 1 meets these conditions. With 1) = ef>(x), we have

T/ = :i:1 = X2 = ({1 - 77)/-y
y(t) =O ~ 6(t) =0 ~ r, = -(lh)TJ

The system is minimum phase if "Y > 0.
-191
• 13.2 (a)
8¢1 + 8<1>1 =0
8x2 oxa '
Take ¢>1 ;; xi and r/>2 == x2 - :z:3, to obtain
T(x) is a global diffeomorphism. The normal form is given by
771 = -'71 + '12, T/2 = f'/1 - '12 - f. - '71.~, { = -111 + u

The zero dynamics are
. 1 ]
1J =[ -1
1 -1 f7
The origin is stable but not asymptotically stable. Therefore, the system is not minimum phase.
• 13.S
ii= a~O) [(m + M)mgLsin8 - mLcos(8)(u + mLB2sin6 -kit)]
The system has relative degree 2 in the domain {181 < ,r/2}. Therefore, it is input-output linearizable. To
find the zero dynamics, take 6(t) 0. =
9(t) =0 ~ 9(t) =0 ~ 9(t) =0 ~ u(t) - ky(t) =0
Thus, the zero dynamics are given by 'ii = 0. The origin is stable but not asymptotically stable. Therefore,
the system is not minimum phase.
• 13.4
'!/ = X1 + U
The system has relative degree 1 in R 2 . Therefore, it is input-output linearizable. To find the zero dynamics,
=
take y(t) 0. The zero dynamics are given by x1 = tan x1. The origin is unstable. Therefore, the system is
not minimum phase.
• 13.5 For,/>;= (i, 1 5 i :!Sm -1, the PDE (13.26) is clearly satisfied since <Pi is independent of (m and {n-
For IPm :::::: (m - f.n/q(x), we have
O<Pm 8</Jm (
B(,m + 8 f.n q x) =1 - q(x)/q(x) =0
- 192
• 13.6 Taking Xi= wi for i = 1,2,3, the state equation is given by± = J- 1 1u - o{x)], where J =
. [ {J2 - J3)X2X3 ]
diag[J1, J 2, J2] and a(x} = - (J3 - J1)x3X1 . The matrix J is nonsingular. This system is in the form
(J1 - J2)X1X2
(13.6) with A = 0, B = J- 1 . The pair (A, B) is controllable. Therefore, the system is feedback linearizable.
• 13.7
M(q)ij + C(q, q)q + Dq + g(q) =u
Let x1 = q, x2 = q, and x = [ :: ] · Then
x = Ax+ B-y(x)[u - a(x)]

where
A= [ ~ ~ ] , B = [ ~ ] , -y(x) = M- 1(x1), a(x) = C(x1,x2)x2 + Dx2 + g(x1)
The inertia. matrix M(x1) is nonsingular by assumption. The pair (A,B) is controllable. The system talces
the form (13.6} of the text; hence it is feedback linearizable.
• 13.8 Let ..X = (f,g].
L1L9 h(x) - L 9 L1h(x) = L1 (::g) - Lg(::/)

8h8gj T{Phj 8h8f ITEJ2h
= 8x 8:c + 9 8x2 8x 8x g 8x 2 g
= 8h 8g J _ 8h 8/ 9 = L,.h
8x8x 8x8x
where we have used the property that the Hessian matrix S is symmetric.
• 13.9 Necessity follows from the fact, that Ji to fr are vector fields of ti.. To prove sufficiency, let g 1 and
g2 be any two vector fields of fl.. Then, they can be expressed {locally) as
,. ,.
91(x) = LCi(x)fi(x), 92(x) = Lt4(x)J.(x)
i=l i=l
r r
= I: L {c.d;{J;, f;] + Ci (Lj;dj) ]j - d; (LJ;Ci) f;}
i=l j=l
-193
Since (/i, /j] EA, we conclude that [91 , 92] EA.
uT~ n[I]-U [ [ n[T] u]

• 13.10
[t,,f,] = =
Since [h, f i) = -[/1 , h], we conclude that A is involutive.

• 13.11 (a)
y. = - 3X1X1 . = - 3X12(X1 + X2 ) + 3X1X2
2. + X2 2 + X 1 + U = - 3X13 + X1 + U
The system has relative degree 1 in R 2 . Hence, it is input-output linearizable.
(b) Find ¢(x) such that (lJ~/8x)g = (8¢>/8x2) = 0 a.nd T(x} = [ -Xl~(x}
+x'l
] is a diffeomorphism. With
if>(x) = x 1 , T(x) is a global diffeomorphism. The change of variables
Tl= xi, { = -xf +x2

transforms the system into the globally~defined normal form
;, = 11 + 1]3 + {, { = -3173 + 77 + u
(c) The zero dynamics are i/ = q + 713 • Hence, the system is not minimum ph~.
(d)
I (x) = [ 3~~;;~1 ] , g =[ ~ ]
adiu = - !!g = [ 6x x~ + 1 3;f ] [ ~ ] = [ a!f
1 ]
g = [ g ad1g J= [ ~ 3!f ], det(Q) = -1 # O
Hence, the system is feedback linearizable.
(e) Find h(x) such that L,h = 0 and LgLJh # 0. It can be verified that h(x) = :z:1 satisfies these conditions
and results in the global cliffeomorphism T(x) = [ Xt x+i X2 ] . The transformed system is
Z1 = Z2, Zz = Z2 + 3z~ (Z2 - ZJ) + ZJ + tJ

u = -z1 - z2 - 3zr(z2 - zi) yields
• 13.12 (a)
y = X1 + %2
y = x1 + z2 = -x1 + X1x2 + x2 + X3
y = (-1 + x2):i:1 + (x1 + l):i:2 + :i:3 = (-1 + x2)(-.:z:1 + x1x2) + (x1 + l)(x2 + xs) + d(x) + u
:;;::(:::·::,~=:):·::~::z:;::::~:;:·rlineariu:tt~,
phism. With ef>(x)
.
= x1 , T(x) is a. globa.l diffeomorphism.
-X1 + .X1X2 + X2 + X3
The change of variables
] is a diffeomor~
- 194
transforms the system into the globally-defined normal form
T/ = -11 + 77({1 - 1/), {i ={2, e3 = (-1 + X2)(-X1 + X1X2} + (x1 + l)(z2 + X3) + 8(x) + ti
(c)
y(t) :: 0 ~ ~I (t) := 0 and {2(t) : Q ~ rJ = -rJ - 1J2
The origin of the zero dynamics is asymptotically stable. Hence, the system is minimum phase.
(d)
ad1g=[f,g]::::: -[-1;x2
*
xl1 * (&6/~xa)
~ l[~ l 1
=[ ~l
-(8&/8xa)
l
(g,ad1 g) =[ ~* ~* 6
-(82 5/8~)
] [6
1
l =[
-({J26/8x~)
g ] = -(8 6/8x~)g 2
Hence, the distribution {g,ad1g} is involutive.
g =[
. 1
~ ~l
•
X1
*
*
l , det(Q) = X1
Hence, the system is feedback linearizable in the domain {x1 > O} or the domain {x1 < O}.
( e) Find h( x) such that
L 9 h 0, =
L,Lth 0, =
8h
L 9 h == -8
X3
=0 => h = h(x1 ,x3 )
&h 8h
L1h = -8 (-x1 + X1X2) + -8 (x:z + xa)
Xt X2
L,L,h = 8(L1h) = 0 ~ 8h =0
8x3 8x2
1
Take h(x) =x 1• It can be verified that T(x) = [ -xi : x1x2 ] is a diffeomorphism in the
(-1 + x2) 2 x1 + xi{x2 + xa)
domain {x1 > O} or the domain {x1 < O}. The transformed system is
i1 = z2, i2 = z3, za = x1[u - o(x)]
Take u = o:(x) + v/x1.

• 13.13 Restrict u to the domain lul < rr/2 and x to the domain D = {lx2I < 1r/2, lxal < rr/2}. Then,
cos(x 2 ) ¥- 0, cos(x3 ) ~ 0, and tan(u) is invertible. Apply the input transformation w == tan(u)/(bcos(x 2 ) cos(x3 )),
to obtain
x ::a::: f(x) + gw
!l
where
9= [
- 195
a:g) g= [ a2 hfOox~
[g, ad19] = 8(ad
-82 '2/8x5
l
Since 8h/8x 2 = - sec2 (x2)/(acos(xs)) 1' 0 in D, we have
82 h/8x~
[g, ad1gJ = 8'2/8x2 ad.19
Therefore, the distribution {g, adI g} is involutive.
9 = [g, ad1g, ad}g] =[

0
1
0
0
(oh/8x2)
-(8'2/8x2)
(8fi/8x3)(fJh/8x2)
*
*
l
det(9) = -(8fi/8x3}(8h/8x2)2
Since ofl/8x3 =-
sec2(::z:3) f:. 0 in D, det(Q) ¥ 0 in D. Thl18, the system is feedback linearizable. To find
the domain.of validity of the linear model, we need to find h(x) such that
8h 8(L1h) _ O 8(L}h)
h(O) = 0, 8x9 = O, 8x g- ' 8x g#O
8h D = O => 8h =O
8x 8x2
We must take h independent of x2, With (8h/8x 2 ) =0, we have
8(L1h) _ O 8(L1h) _ 0 oh (secx2) 2 = O => 8h =O

8:,; n- => x2 a -
=> xs a cos xs a ( ) 8xs
So, we choose h independent of xs. Therefore,
8(L}h) g = 8h (secx3)2(secx2) 2 'F 0

ox 8x1 a cos(x3)
We must require
Take h = x 1 , which yields

L1h = tanxa, L}h = (1/a)(secx3 ) 3 tanx 2
The domain of validity of the linear model is x ED and u E { lul < 1t/2}.
- 196
• 13.14 {a)
(g, adtg] = 8(ad1g) 9 = [

8x
g]
0
l
Therefore, the distribution {g, ad1g} is involutive.
ad}g = [/, ad1g) =

a(ad
a: g)
f -
of
ax ad.Jg= [ -(l + xi)
I+ x~ - X3
g = [g, ad,g, adjg) =[

0
I O
0
1 1 + X1
-(1 +xi)
1 + X2 - X3
0
l ' det(g) =(I+ xi) 2
Thus, the system is feedback linearizable in the set {1 + x 1 # O} .

(b) Find h that satisfies
ah 8(L1h) O 8(LJh)
h(O) = 0, 8x9 = 0, 8x g= ' 8x g#O
&h
8xg=0 ~
Try h = x1 . Then, L1h =-xi+ z2 - Zs ,

8(L1h) O
ax g =
L}h = -(-x1 + X2 - X3) + (-x1X3 - X2) + X1 = 2x1 - 2x2 + X3 - X1X3
8(LJh)
OX 9 = -2 + 1 - X1 = -(1 + xi)
Restrict X to the domain {1 + X1 > O}. Thus, the transformation T(x) is given by
A linearizing state feedback control is given by
u = a(x) + (3(x)v
where P = -1/(l + x1) and a can be calculated using {13.36) of the text.
• 13.15
z = T(x) = [
8x~
20x2(k - bx2)
+ ex~ + c2
26(k - 2bx2}(-bx2 + k - cx1x3)
l
Clearly T(x) is differentiable. Find y- 1 (z). Start with the equation
z2 = 28x2(k - bx2}
- 197
For x2 < k/2b, this equation has a unique solution x2 = <f,(z2), Consider next the equation
which yields
in T(D0 )
The last equation is

Z3 = 28(k - 2bz2)(-bx2 +k - CX1X3)
Since (k - 2bx2) > 0 and Z3 > 0 in Do, this equation uniquely determines x 1 as a function of z 1 , z 2 and
z 3 . Thus the map r- 1 (z) is well defined on T(D0 ). It can be easily checked that r- 1 (z) is differentiable in
T(Do) -
• 13.16
±1 = X2, ±2 = -10 [sin (x1 + i)- ~] + IOu

The state feedback control of Exercise 12.1 is
1 .
u == -2.5x1 - x2 => T = v'2 - 2.5(8- 1r/4) - 8
The closed-loop eigenvalues of the linaerized model are -5 ± 2.659lj. Using feedbaclc linearization, we obtain
the control law
To assign the closed-loop eigenvalues at -5 ± 2.6591j. we choose k1 = 3.20711 and k2 = 1. Thus,

T = sin8 - 3.20711(8 -1r/4) -8
Simulation of the two controllers, for different initial conditions, are shown in Figure 13.1. For the case of
zero initial conditions, the responses are almost identical. For the initial condition (J =
51r /4 and iJ = O,
there is a difference where 8 reaches steady state faster under feedback linea.rization, at the expense of more
control effort and a larger transient of /J. The difference between the two controllers is not dramatic. We
would have seen a more dramatic difference had the feedback gains been smaller, for then the control based
on linearization would have produced a closed-loop system with multiple equilibria.
• 13.17
u=[!]
ad19 =[/,g) = [ :t, ], [g,ad1g) = [ ~]
Hence, the distribution {g, ad1g} is involutive.
- 198
2 2.5
2
1.5
1.5
,l' 1 ~
0.5
0.5
2 2 2
0
,
I
-2
..
"-4 I
~
''I I
I
--6 I
-6
\ , -8 I
.,
o...__,_ _ _ _ __,
0 2
-8
0 2
-10,----------
0 2
Figure 13.l; Exercise 13.16: Response under linearization (solid) and under feedback linearization (dashed).
Q= [ ~ ~l
2
-. ] , det(Q) =1
0 -xi 1 - 2x1
Bence, the system is feedback linearizable. Find h(x) that satisfies
8h 8(Lih) _ 8(L}h)
h(O) = 0, 8xg = 0, 8x 9 - 0, 8x g ::/: 0
8h 8h
-g=O => - =0 => h=h(x1,xs)
ax 8x2
8h 8h 8h
L1h = - f = -(x2 - x1) + -(x1 +x1:r::2 - 2:r;3)
8x 8x1 8xs
8(L1h) g = O => 8h + X1 8h =O
ax OX1 8x3
Try h = x? - 2x3.
2
L1h= 8(L
Bx1h) /=-22x1+l
( ){x2-x1 ) +4x1+x1x2-2xs
( ) =4x21 +6xi-2x2 -8x3
8(L}h)
ax g = -2 :/, 0
- 199
is a global diffeomorphism. In particular, r- 1 is given by
The transformed system is given by

i = Acz - 2Bc[u - a:(x)]
Design K such that Ac - BcK is Hurwitz and take u = a(x) + (I/2)Kz.
• 13.18 (a) The system is already in the form (13.1), with
A= [ ~ ~ ]. B= [ ~], 1(x) = -bcosx1, ct (X )

asinx 1
;:::;; - , - - - -
b COS X1
Therefore, it is feedback linearizable provided cos .:z:1 # 0, which is the case in the domain D = {lx1 I < 1f/2}.
(b) Take
I .
u=- b (-asmx1 + v)
COSX1
to obtain
To stabilize the system at x1 ;:::;; 8, take
The closed-loop system

X1 = Z2, X2::::; -k1(X1 -8) - k2X2
=
has an asymptotically stable equilibrium point at x (8, O). The domain of validity of this control is limited
to lx1 I < 1r /2. Therefore, we cannot make the equilibrium point globally asymptotically stable.
• 13,19 The map T(x) is given in Example 13.14. Using (13.36), we can calculate o(x) and "Y(x). These
three functions depend on the parameters a, b, c, and d. Define· T(x), &(x), and i'(x) by replacing the
parameters with their estimates, and set P(x) =
1/i'(x). The state feedback control is given by u =
=
a(x) - P(x)KT(x). The closed-loop system can be represented by .i (Ac - BcK)z + B.,o(z), where o(z) is
defined in Section 13.4.l.
• 13.20 (a)
'7 = /0(17,{i), i =A.,~+ BcP-1 (x)(u- a(x)]
Ta.ke u = a:(x) + P(x)v, to obtain
Apply the change of variables Zi = e(i-l)ei and w = e;Pv.

ez1 == e{1 = e{2 = z'l
2. 2
ei2 = E6=EQ=Z3
E:Zp-1 = e p-lfp-1 = C: p-1{p =Zp

e:i,, = eP{p = ePv = w
- 200
Thus,
(b) With w = -Kz, we have

Let V('1, z) = Vo(JJ) + J"iT Pz.
V = 8~ /0('1, z1) + ~zT[P(Ac - BcK} + (Ac - BcK? P]z
v,, 2e zTPz.
8Vo 8Vo 1 T
::::: Of/ /o(7J,0) + &r, [/0('1,zi) -10(11,0)) - 2c'-lz7'Pzz z
Let c be any positive constant and define the compact set n by n = {V(11,z) :Sc}. On the compact set n,
we have
oVo
Ch] [fo('7,z1) - /0(77,Q)j :5 klfal
where k depends on c. Hence,
Choosing e small enough, we obtain
which shows that, for sufficiently small e, the origin is asymptotically stable and the set !l. is contained in
the region of attraction.
(c) Without loss of generality, take E $ 1. For any compact set in Rn, we have IITJoll2 :S c1 and ll€0ll2 :S c2.
Then, llz(O)ll2 :S 11~0ll2 :S c2. Choose c > 0 such that {11771!2 $ c1 and llzl12 :S c2} C {V(77,z) :Sc}. Then,
initial states (1Ji>,{o) in any compact set in R"' are included in the region of attraction. which shows that the
feedback control achieves semiglobal stabilization.
(d) Notice that for any bounded {(O), the initial state z{O) is bounded uniformly in e. Therefore, the
solution (77(t),z(t)) is bounded uniformly in E. However, peaking is present in {2 to {p, as seen from the
scaling transformation. There is no peaking {1 since e1 =
z1. The function /o depends only on z1, which
explains the special nature of the current system.
• 13.21 (a)
=> 'ii = ±1 => 'ii =.:i:2 = x1 x2 - +u
:::~:::::(::iv[e1~T2]m~:::~:::.z:::;:::):z:,~: ~:~~=
y = X1 - X2 - .:i:2 = x2 x~
L 1h(x)
the PDE and makes T(x) a global diffeomorphism. The normal fonn is given by
This is a special normal form because the r,-equation depends only on 6.

(b) The zero dynamics are given by iJ = -7]3 • The origin of this system is globally asymptotically stable.
=
(c) Taking Vo(11) ½112 , we have
- 201
on the set {1111 $ k1}. Therefore, all the assumptions of Exercise 13.20 are satisfied, and we can design
a semiglobally stabilizing state feedback control following the procedure outlined there. In particular, Let
K = [1 l]; then, Ac - BcK is Hurwitz. A semiglobally stabilizing state feedback control can be taken as
1 1 1
u = -66 + -(-z1
e2
- z2) == -l16 - -fa
e2
- -6
e
with sufficiently small ~-
• 13.22
i] = /0(1/,{), ~=(A - BK){+ Bt5(z)
By Theorem 4.16, there is a Lyapunov function Vo(11} that satisfies
8Vo .
a;j' fo(TJ, 0) :5 -a3(Jlr,II)
in some neighborhood of f'J =
0, where 03 is a class K, function. Let P = pT > 0 be the solution of the
Lyapunov equation P(A - BK}+ (A - BK)T P = -I. Take
V(71, €) = Vo(T/) + >.../f!Pi,, >. > 0

as a Lyapunov function candidate for the full system.
V = ~o !0(11,0)+ ~[!(11,e)-!0(11,0)J+ 2 ~[-(7e+2ePBo)

VI/ VI/ { P{
< -03(1117ll)+k1ll{ll->..k2ll{ll2+>..kse
-
where k1 to k3 are positive constants. In arriving at the foregoing inequality we have used the fact that / 0
is locally Lipschitz and [8Vo/8'7] is bounded in some neighborhood of 1/ = 0. Choosing>.= 2k1 /k2 yields
v :5 -a3(11'111) - k1ll{ll2 + (2k1k3/~)t
The conclusion follows from Theorem 8.18.
• 13.23
iJ = fo(fJ, {), t = (A - BK)€+ BtS(z)
Without loss of generality, assume that 11611 $ kllzll is satisfied in the 2-norm. By Theorem 4.14, there is a
Lyapunov function Vo(1J) that satisfies
c1ll1Jll22 S Vo(q) S c2ll11lb,

2 Of/ fo(TJ, 0) S -csll11l'2,
8Vo 2 8rJ !I ~ C4IITJll2,
118Vo "I TJ
Let P = pT > 0 be the solution of the Lyapunov equation P(A - BK) + (A - BK)T P = -I. Take
V(17, ~} = Vo(T/) + >.eT Pe, ). > 0
as a Lyapunov function candidate for the full system.
V =
8Vo
<711 fo(fJ, 0)
8Vo
+ a;,[!(17, e) - /o (77, O)] - ,\e{+ 2,\{TPB6
< -c3l117II~ + C4Lllt1U2ll!ll2 - >.U{II~ + 2>.kllPBll2IIT/ll2ll{lb + 2>.kllPBll2ll{II~
where Lis a global Lipschitz constant for Jo with respect to{. Let k• = min{l/(4IIPBll 2), (C4L}/{2AIIPB11 2 )}.
Then, for all k ~ k·, we have
·
V $ -
[ 1111/12
11{112
]T [ -c4L
C3 -c.L] [ 11111'2 ]
)../2 ll{ll2
=
Taking ). ).* ~f 4(c.L) 2 /c3 ensures that the 2 x 2 matrix is positive definite. Consequently, the origin is
globally exponentially stable for all k .$ min{I/(4IIPBll2), (C4L}/(2).·IIPBlb}}.
- 202
• 13.24
iJ = /0(11,{), {=(A - BK){+ B6(z)
Without loss of generality, assume that 11811 $ kll{II + W(77) is satisfied in the 2-norm. The derivative of
V(11, {) = Vo(11) + >.JeP{ is given by
v = ~o !0(11, 0) +
VI/
~ [/(11, {) -
VI/
!0{17,0)) + ~ [ - e { + 2e PBo]
2 { P{
< -W(71) + k111{112 - Ak2ll{lf2 + ..l.kal16!12
< -W(17} + k11i{ll2 - Ak2ll{lb + .\kakll{ll2 + Ak3kW(11)
= -(1 - >.kak)W(11} - (..l.k2 - k1 - .\kak)ll{ll2
where k1 to k3 are positive constants. In arriving at the foregoing inequality we have used the fact that
/ 0 is locally Lipschitz and [8Vo/87JJ is bounded in some neighborhood of fJ =
0. Ta.Ice ,\ = 2k1 /k2 and
k· = min{l, k1 }/(2).k3). Then, for all k $ k·, we have
· I 1
V $ - 2W(11} - 2kdl{ll2
which shows that the origin is asymptotically stable.
• 13.25
y =X1 '* ii= X2 + 2x~ '* ii= X3 + u + 2:i:1 (x2 + :z:f)
Therefore, the system has relative degree 2 in R 3 . Let us check the minimum-phase property.
y(t) =0 -* :i:3 = -:z:3

Hence, the system is minimum phase. Let e = y - r.
e = jj - r = X3 + U + 2x1 (X2 + xi) - r
Ta.Ice
U = -X3 - 2x1{X2 +xi) +r - k1e- k2e

= -x3 - 2:z:1 (z2 + xf) + f - k1 (x1 - r) - k2(x2 + 2x~ - r)
where k 1 and k2 a.re positive constants. The tracking error e satisfies the equation e+ k2e + k1 e = 0, which
shows that e(t) -t Oas t -t oo.
• 13.26
Therefore, the system has relative degree 2 in R 2 • Let e =y - r.
Ta.Ice
ti= -X1X2 - (sinx1 + X1 COS:t1Hx2 + X1 sin xi)+ r -
k2e k1e -
where k1 and k2 are positive constants. The tracking error e satisfies the equation e + k2 e+ k 1 e = 0, which
shows that e(t) -t O as t -+ oo.
- 203
• 13.27 (a) Denote the gravitational constant by go and write the state equation as x = f(x) + g(x)u.
o0 o][fil
0 h - [o * 1* o ][o]
(0'2/ox3) 0
0 0 '3 * • (8h/{)x3) 93
The foregoing representation of ad1g is valid in the domain D = {a+ x 1 > 0 and x 3 > O} since in this
domain
8h Loaza
Iha =- m(a + x 1 ) 2
is well defined and different than zero. Hence, the distribution {g,ad1g} is involutive.
8(ad,g) 8/
ad}g = [/, ad19] = ax f - OX (ad,g)
= [ ~• ~*.o~ l[~: ]-[~ ~

h • •
0
g = [ g ad1g ad}g ] = [ 0
93
0
-(8h/8xa)ga
•
-(8h/:8xa)!/3 l
. .
det(9) = -93 ( 8h
oxa 9a
)2
=-
£2a2x2
L3(xi),:2 (a ~ xi)• =IO for x ED
Thus, the system is feedback linearizable in D.
(b) Find h(x) such that
8h 8(L}h)
h(O) = 0, oxg = o, 8x g=l,O
Try h = h(xi),
8h 8h
h=h(x1) => oxa =0 => axg=O
L1h = 8h J = 8h X2 => 8(L1h) 9 = 0

ax 8x1 ax
L 2 h - 8(L1h) 1 _ fflh 2 8h ( k aL0 x~ )
1 - ax - 8x~x2 + 8x1 90 - mx 2 - 2m(a + x 1)2
8(L~h) 8h aL0 x 3
OX3 =- 8x1 m(a + x1) 2
- 204
Then, [o(L}h)/8x]g '# 0 for all x ED. Take h;::;; x1, to obtain
It can be verified that T(x) is a diffeomorphism in D. The change of variables z ;::;; T(x) transforms the
system into
Take
to obtain
. k
Z3 =- -Z3 +v
m
To stabilize the ball at y = r, take
The overall control is
The feedback gains are chosen as k1 = 2000, k2 = 400, and ka = 30 to assign the eigenvalues of
[ -k1o0 01 o1
-k:i -(ks + k/m}
l [o
:::::: 0
-k1
0l
-k2
O
1
-ks
l
at -10, -10 ± jlO. The step response of y and u for y(O) =
0 and y{O) =
0.089 (with other initial states
equal to zero) are shown in Figure 13.2. To account for the constraint 0 ~ u ::5 15, a limiter is included in
the Simulink simulation model. The response is considered feasible only if y belongs to the interval [0, 0.1].
Using this criterion, 0.089 is the largest acceptable initial position. Comparing these results with those of
Exercise 12.8, we see that the response has less overshoot in the current design. Consequently, the maximum
deviation of y(O) is 0.089 compared with 0.07 in Exercise 12.8. Figure 13.3 shows the response for different
values of m. These values are the extreme values for which a feasible response is obtained for the given
initial position. There is a large steady-state error. The smallest mass we could work with was m 0.067,=
compared with m == 0.046 in Exercise 12.8. This could be an indication that feedback linearization is more
sensitive to parameter perturbations. The large steady-state error can be handled by the use of integral
control .
(c) From pa.rt (b), we know that the system is feedback linearizable with h(x) =
:z: 1 • This shows that,
=
with y x 1 as the output, the system is input-output linearizable. Proceed to design the tracking control
following the steps of Section 13.4.2.
- 205
y(0)• 0
o.oe...--~---------, 1.------------, y(0)=0
o.ot
3
0.2 o., 0.8 0.8 0 0.2 0.4 O.B 0.8
y(0)"0,089 y(O) • O.Cl89

16
1,
12
:,
0.2 o., 0.11 U 02 o., 0.6 0.8
• 13.28 (a) -
y -== xz => ii = -94x2 + 8:,x1 u
Let D = {x E R2 I X1 > Oa/'281}- The system has relative degree 1 in D. Therefore, it is input-output
linearizable.
(b)
The zero _dynamics equation has a globally exponentially stable equilibrium point at .x1 =63 /91 • Therefore,
the system is minimum phase.
(c) Let e = y - r .
The control
u = 62-[84.x2
,5%1
+r - keJ, k > o
results in e = -ke. Hence, e(t)-+ 0 as t ~ oo.
(d)
(1) Figure 13.4 shows the response for 'T = 0.1 and k = l. The settling time is about 0.5.
(2) To decrease the settling time, -r w~ reduced to -r = 0.02. The response didn't improve because reducing
T increases f while the control is limited to ±0.05. This situation can be improved by increasing the feedback
gain k. Figure 13.5 shows the reiiponse for r = 0.02 and k = 10. The settling time is about 0.1. Notice,
however, that the contro1 saturates as a result of the increase inf.
(3) The response with a 50% decrease and 50% increase in J is shown in Figure 13.6. The settling time
deteriorates.
(4) Increasing k to k::;;: 10 improves the response a lot, as shown in Figure 13.6.
- 206
y(O) =0, m =0.1 y(O} = 0.089, m = 0.1

0.06 0.1
0.04 0.08
>- >,
0.02 0.06
0.04
0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
y(O) = 0, m = 0.067 y(O) = 0.089, m = 0.067
0.1
0.01
;i,,, >-0.05
0.005
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

y(O) = 0, m = 0.11 y(O) = 0.083, m = 0.11
0.08 0.12
0.06
0.1
>,
>-0.04
0.08
0.02
0.06
0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
100 0.04
80
0.03
60
>-
Output :::1 0.02
Reference
40
0.01
20
0 0
12 1.4 1.6 1.2 1.4 1.6
I I
- 207
100
0.05
80
0.04
so ·0.03
>- :::,
Output
40 Reference 0.02
2D 0.01
0 0
0.9 1 11 1.2 1.3 1.2 1.4
t I
50% lnel'tia Reduction, k = 1 50% I.,.rtia lncrvue, k = ,
150 100 - -- - - - - -- .... --

BO
>-
too
50
0
I
I
I
,,...,
,~ ~~nee, ,._ 60
40
20
0
1 1.5 2 2.5 3 1 2 3 4 5
t
50% Inertia Reduotion, k = 10 50% Inertia lncreue, k • 1 O

100
,,,. .., -- 100
ao BO
60 60
>- ,._
40 40
20 20
0 0
u 1.4 1.6 1.2 1.4 1.6 1.8 2
I
-
208
Chapter 14
=
• 14.1 (a) Let s a1x1 + x2, where a1 > 0. On the sliding surfaces =0, we have ±1 = -a1x 1 + sinx1.
Choosing a 1 > I ensures that the origin X1 =
0 is asymptotically stable.
s = 01±1 + ±2 = a1(x2 + sin xi)+ 81xi +(I+ 02)u
Ta.king u= -a1 (x2 + sin x1) + v yields
s = 81Xi - 82a1 (x2 + sinx1) +(I+ 82)v
ss = 81sxi - 02a1s(x2 + sin xi)+ (1 + 82)sv :5 2x~lsl + (ai/2)lx2+sinx111s1 +{I+ 82)sv
Take v = -P(x)sat(s/1:), 1: > 0 . For Isl ~ 1:, we have
ss $ 2x~lsf+-(a1/2)lx2 + sinxillsl - (1/2)/1(:1:)lsl

Take {J(x) = 4x~ +a1lx2 +sinx1I + /Jo, /3o > 0. Then ss :'.S: -.Bolsl/2, which shows that the trajectories reach
the boundary layer {Isl ~ e} in finite time. Inside the boundary layer we have
Let Vi = ½xr
Take 01 = 2. . 2 2
Vi $ -x1 + lx1le $ -(l/2)x1, V lx1I ~ 21:
Thus, the trajectories reach the set !le- = {lx1 I $ 2e, Isl $ 1:} in finite time. Inside !le-, the system is
represented by
±1 = -2x1 + sinx1 + s, s = 81x~ - 282(s - 2x1 + sin xi) - (1 + 92)(s/e)

Let V2 = ½(x? + s2 ).
¼ =
<
=
The 2 x 2 matrix is positive definite for sufficiently small e. Hence, all trajectories in !le- approach the origin
at t tends to infinity.
=
(b) Let e1 X] -rand e2 = e1 =±1 -r X2 +sinx1 -r. =
e1 = e2, e2 = f)ix~ + (1 + 92)u + (x2 + sinx1) cosx1 - r
- 209
Let the sliding manifold be e2 = -a1 e1, a1 > 0, and set s = e2 + a 1e1.
s = B1xi + (1 + 82)u + (x2 + sinx1) cosx1 - r + a1e2
Take
u = -(x2 + sin xi) cos x 1 + f - a1 e2 + v
s = v + .i, where .i = B1xf + 92[-(x2 + sin xi) cosx1 + r - a1e2J + B2 v
i
I.ii :5 2xf + ~ l(x2 + sin xi) cosx1 + a1e2I + Iii+ ½lvl
Take
and
v = -P(x) sat(~)
For Isl 2: e,
ss :5 -.Bois!
which implies that the trajectories reach the boundary layer {Isl $ e} in finite time. Inside the boundary
layer,
Take a1 = 2.
Thus, lei (t)I :5 e after a finite time.
• 14.2 (a)
(b)
'ii= z2, ii= -ax2lz2I + u
The system has relative degree 2 in R 2. Hence, it is input-output linearizable. In fact, the system is globally
=
input-output linearizable since the control u ax2lz2I + v results in a chain of two integrators form v toy.
=
(c) Let e1 x1 - ¢.,. and e2 = x2 -1/i,,..
e1 = e2, e2 = -ax2lx2I + u -,ii..

Taking
results in the linear error equation

e1 = e2, e2 = -e1 - e2
whose matrix A = ( ~l ~1 ] is Hurwitz. Hence, the control achieves global tracking.

(d) When a::/ a, the error equation is
e1 = e2, e2 = -e1 - e2 - (a - a}x2lx2I
The perturbation term satisfies
(a - a)x2lx2I :5 0.0lx~ = O.Ol(e2 + 2 cos2t) 2 :5 O.Ol(e~ + 4le2l + 4)
- 210
To show ultimate boundedness, we need to bound the perturbation term by a linear growth bound. Restrict
e2 to le2 I S k2. Then,
(a - a)x2lx2 I S 0.Ol(k2 + 4}le2I + 0.04
The solution of the Lyapunov equation PA+ ATP = -I is P = [ ~:: Oi5 ] . Use V =eT Pe as a Lyapunov
function .
V = -llell~ - (e1 + 2e2)(a - a)x2lx2I S -llell~ + v'5Uell2[0.0l(k2 + 4)11e!l2 + 0.04)

= -[1- 0.01-15(4 + k2}Jllell~ + 0.04v'Sl)ell2
Take k2 = 4.
V = -0.82lllell~ + 0.0891lel12 = -0.00llfell~ - 0.82l!ell~ + 0.089llell2
S -0.00lllell~, V lle1'2;?: ~~~ ~ µ

It follows from Theorem 4.18 that the ultimate bound is given by
Amu(P) Jl1ID9
b =µ >.min(P) = 0.1085V ~ = 0.176
This ultimate bound is not valid globally because we restricted the analysis to the set {le2 1 $ 4}.
=
( e) Use sliding mode control with s e 1 + e2 .
s = e2 - ax2lx2I + u -../,r = e2 - x2lx2I - (a - ii)x2!x2I + u-../,r
Take u = -e2 + x2lx2I + ;J.,r + v.

ss = -(a - ii)x2lx2ls + sv $ 0.0lx~lsl + sv
Take v = -(0.0lx~ + 0.1) sat{s/c). After some finite time, Isl $ E. From the equation e1 = -e 1 + s , we see
that after some finite time /e1 I ~ E /8, where O < 8 < 1. Choose E < 0.01 to ensure that the ultimate bound
on le1I is 0.01.
• 14.3
(a) For u = 1, this is the standard van der Pol oscillator which is known to have a stable limit cycle. The
fact that the limit cycle is outside a circle of radius 1/µ in the plane (x 1 , x 2 /w) can be shown by transforming
the equation into polar coordinates. Let
p2 = + x~/w2xr
pp= ~(1 - µ 2 x~)x~
w
On the circle p = 1/µ, we have jx1I $ 1/µ, which implies that {I - µ 2 xf) ? 0. Hence, all trajectories on
this circle must be moving to the outside. Therefore, the stable limit cycle must be outside the circle. For
u = ~ 1, we can show the existence of the unstable limit cycle by reversing time and scaling the state variable
to arrive again at the standard van der Pol oscillator.
(b) We have
Now
s(t) =0 :::;,. s(t) =0 :::;,. u(t) =0
- 211
Therefore, the state equation reduces to a harmonic oscillator

.
Xz = -W2X1
(c)
Fors-::/: 0, take u : -sgn(s).
Inside the band /.x 1 1 < 1/µ., the term (1- µ. 2.xf) is positive. Therefore, ss $ 0. In fact, ss < 0 except on the
line x 2 =::: 0. Since no trajectory (other than x(t) =
0) can stay identically in the set x 2 == 0, we conclude
that s(t) must reach zero; that is, the trajectory must reach the sliding manifold.
(d) Simulation results for u = - sat(s/0.01) are shown in Figure 14.1.
1 .---------.......----...
0.4
0.5
0.2
x.... 0
-02
-0.5
-0.4
0 10 20 30 -0.5 0.5
Figure 14.l: Exercise 14.3.
• 14.4 Let l/i = ½x? be a Lyapunov function candidate for :i:1 = x 2 + ax 1 sinx1 •
The origin x1 = 0 can be globally stabilized with z2 = -3z1 since V1 ~ -:zf Takes= 3x1 + xz.
s = 3:i:1 + ±2 = 3(x2 + ax1 sin.xi)+ bx1x2 + u
Take
u = -3{x2 + x1 sinx1) - x1x2 +v
iJ = -3(a - l)x1 sinx1 + (b- l)x1x2 + v
I - 3(a - I)x1 sinx1 + (b- l)x1x2I $ lxil(3 + 2lx21)
Talcing
v = -[l + jx11(3 + 2lx21)] sat G)
=
ensures that s(t) reaches the boundary layer {lsj $ e} in finite time. Since the origin of :i:1 -3x 1 +ax 1 sinx 1
is exponentially stable, it follows from Theorem 14.2 that the origin x = 0 is globally asymptotically stable.
- 212
• 14.5 Let x 1 = 8, x2 = iJ, u = T, a;;; 9/l, b = k/m, c = I/me2 , and ((t) = h(t)/l, to obtain
±1 = x2, :i:2 = -asinx1 - bx2 +cu+ ((t) cosx1
Ta.ke s = x1 + x2. Then,
s = x2 - asinx1 -bx2 +cu+ ((t)cosx1::;: c[u + o]
where
6 = ! [x2 - asinx1 - bx2 + ((t) cosxi]
C
161 Sl~l lx1I +II: bl +I(~) ISlx2I 16.186512:d + l.815lx2I + 1.1111

Ta.ke
u = -[16.1865lxd + l.815lx2I + 2) sat(;)
The trajectory reaches the boundary layer {Isl S c} in finite time. Inside the boundary layer, we have
x1= -xi + s. Talcing Vi = x~ /2, we obtain
•
V1 =-Xi+ X1Z
2 2
s; -Xi+ 2
l:t1le s; -(1- 8)x1, V lx1I 2!: e
E
where O < IJ < l. Thus, the trajectory reaches the set n, =. {lx1I :5 E/8, lx1 +x2I :5 c} in finite time. Inside
this set,
lx2I = lx1 + x2 - xii$ lx1 + x2I + lx1I :S (I+ I/8}c
For 8 = 0.9, we have Jx2l ~ 2.lk Choose e small enough that 2.llc :S 0.01. In particular, ta.ke c = 0.004.
• 14.6 (a) The system ±1 = x1x2, with x2 viewed as the control input, can be globally stabilized by
X2 = -xr. Take S = xJ + X2-
Take
U = -2xfX2 - X1 - sat (;)
ss::;: -Isl, for Isl 2: e
Hence, the trajectories reach the boundary layer {Isl S E} in finite time. Inside the boundary layer, we have
. 3
XJ =-Xi+ X18
¥1 = -xt + x~s :S -xf + xfE :5 -(1/2}x1, V lx1I ~ hi

Thus the trajectories reach the set n, = {Jx1 I :5 -..fii, Isl :S E} in finite time. Inside n£ the system is
represented by
Xt = -x~ + XtS, s =-sf.:
The foregoing analysis shows that the system Z1 = -xr
+ XJS is input-to-state stable. It follows from
Lemma 4. 7 that the origin of the cascade system is globally asymptotically stable. Thus, every trajectory
inside n,! converges to the origin as t tends to infinity. Since all trajectories outside n.. reach n~ in finite
time, we conclude that the origin is globally asymptotically stable.
(b)
-X1 ]
ad1g = [ 1 , det Q = det[g,ad1g] = x1
Hence, g is singular at x = 0. From Theorem 13.2, we see that the system is not feedback linearizable in
the neighborhood of the origin.
- 213
• 14, 7 The system is in the regular form with TJ = [ :: ] and l = X3. Design X3 = q,(x1, x2) such that the
origin of
is globally asymptotically stable. Take <J, = -2x2 and V = (1/2}(xr + axD, where a> 0 is to be chosen.
J::: ]T [ !½ -} ] [ l::I ]
V = -xf +x1 tanbx2 - ax~:; -x~ + !xii lx2I -ax~== - [
where we have used ltanhx2I :$ lx2I- Choose a> 1/4 such that Q = [ !½ -} ] positive definite. Take is
a= 1. Then V = (l/2)(x~ + x~). Now, take s = X3 - <J, = X3 + 2x2.
s = u + 6(x) + 2x2 + 2x3
Take
u = -2x2 - 2x3 - (,8(:i;) + .Bo] sat(~), /3o >0
ss = so - (/3 + /3o)s sat (~)
For Isl 2'.: c,

ss :$ .Bisi - .B1s1 - .Bolsl = -f3olsl
Hence, s reaches the boundary layer {Isl :$ c} in finite time. Inside the boundary layer,
• 2 1 2· 1 2 2
V :$ -,\min(Q)ll'7112 + lx2I Isl :S - 211111'2 + 1111112 Isl :S - 2(1 - 8)11771'2, "fl 1'711z 2'.: 911sll
where 0 < 8 < 1. Inequalities (14.14) a.nd (14.15) of the text a.re satisfied with II · II II· l12, 0:1 (r) 02(r) = = =
(l/2)r2 , -y(r} =
(2/8)r. The constant k 1 =
1 since sis scalar. Thus a(r) a-2 (-y(r)) =- (1/2}(2/9) 2r 2 = =
(2/02)r . The sets O and 01! are given by
2
1 2
fl = {2(x~ + x~) :$ c.o} x {Isl ::: c}, Co ;::: 82 c2
n£ = {21 (x~ + x~),:; 022 e2 } x {Isl:$ e}
Choose c and co such that {llxllQO s k} c n.

1
llxlloo ::s k ~ !xii ::s k, for i = 1, 2, 3 ~ 2(xr + x~) ::;: k 2 and Isl ~ 3k
Take c : 3k, 92 = 0.9 and Co = 20k2 . We need to estimate the ultimate bound on x1.
2
+ x22 $ 4 2
~
Ix1I ~
:$ .Jo.g = 2.1082e
x1 0 _9 e
Choose£ small enough that 2.1082e ::;: 0.01 or e ::;: 0.0047.
• 14.8
i, = e, e = A(ly ) (u - cJy) =~ ( u - c.jy) + 6

A(y)
- 214
where
1
6 == ( A(y) I )
- A(y} u -
( c c )
A(y) - A(y) v'Y
Let the sliding manifold be e =- k1 a, k1 > 0, and set s == e + k1 u.
s = A~y) (u. - cJy) + & + k 1e
Take u = c..,jy- A(y)k1e + A(y)v. Then
Take
f:J(x) 1 [1!3 ./Y + l,14k1lel] + t3o,
= -1 -l,14 .Bo> 0
l!l
and
v = -{J(x) sat(~}
• 14.9 Part (a) follows from elementary matrix operations, and pa.rt (b) is a straightforward application of
Frobenius theorem.
• 14.10 It can be verified that s = Gv + a. Therefore, the control can be taken as in (14.11) with e(x)
replaced by e(t,x). The analysis proceeds exactly as in the autonomous case.
• 14.11 (b}
sisi = siai - Bigi{Jq(s,/e) ~ g, [ ~.· s. - .Bsiq(si/e}]
For ls,I?: £, s;u(sde) 2:: ls;lu(l). Hence,
Bisi < g; [(u + Ko.B)lsil - Pls;la(l)] = Yi [t.>ls;I - (u(l) - "O),Bls;I]

< g; [els;I - els.I - (u(l) - Ko).Bols;I] 5 -go.Bo["(!) - Ko]ls;I
(c) The foregoing inequality shows that the trajectories reach the boundary layer { ISi I ~ e} in finite time
and stay therein for all future time. From this point on we can repeat the proof of Theorem 14.1. To prove
=
a result similar to Theorem 14.2, assume e(O) 0 and Ko= 0. For lsil :::; £,
Therefore,
Using the Lyapunov function W = Vo(11) + ½Ef= 1 sf as in the proof of Theorem 14.2, it can be shown that
the origin is asymptotically stable for sufficiently small E.
- 215
• 14.12 (a)
s;si = s;:Ai - 9iSi/3isgn(si) = 9i [ !i' Si - .P,lsil]
< . 9i [lsill?; + lsil t

J=l
ltijlv;f - .Bilsi/] :5 g; [fs;fUi + lsif t
J=l
"'i;/3; - Pifsifl
= g; [lsil!?i + lsil t
J=l
K-;;(u, + w,) - (i:T; + w,;)lsil]
p
(J - X:.)u = e => CTj - L K.ijCTj = Q;
i=l
p
(I - IC)w = b => wi - L "-iiw; = b;
j=l
Si.Si= g, [ls,jq, - Uils,f - bils,1] $ -g,bils,/ $ -9ob,lsd

(b) The condition }::=1 ,,.,, :5 Ko < I implies tha.t I - I( is diagonally dominant; hence, it is an M-matrix.
Take
p p
Ui = U; - L "-ijUj = (I - L K-i;)/3 2'.: (1 - K.o)/3

i=l i=l
Taking /3 2'.: u/(1 - K.o) implies that Q, 2'.: q.
• 14.13 First we show thats reaches the boundary layer Isl SE:. in finite time. Then, we use the equation
=
fr-2 = e,._1
er-1 = -k1e1 · · · - kr-lTr-1 +Z
to show that (e1, . .. , er-d is ultimately bounded with an O(E:) ultimate bound. Therefore, e1 has an O(e)
ultimate bounded. ·
• 14.14 Let a, b, and c be nominal values of a, b, and c, respectively. The state equation can be written as
where
- 216
We have 0.551 ~ c ;5; 2.4691 and we need to choose c such that l{c - c)/cl < 1. We take m = 0.66 and
= = =
i 1 so that c 1.152 results in l(c - c}/cl ~ 0.64. We take k 0.1. These choices of m, i and k result in
a = 9.81 and b = 0.152. For the nominal design, we take
u = ~(iisinx1
C
- k1x1 - k2x2)
which results in the nominal closed-loop system :i; = Aox, where Ao= [ -~1 -(k21+ b) ] . We take k1 = 1
and k 2 = 2 - b= 2- 0.152 = 1.848 to assign the eigenvalues of Ao at -1 and -1. It can be verified that
P = [ ~:~ ~:~ J is the solution of the Lyapunov equation PA0 + AfP = -I. Thus, V(x) = xT Px is
a Lyapunov function for the nominal closed-loop system. Consequently, w = [8V/8x]G(x) = c(x1 + x 2 ).
Substitution of
in 5 results in
6 = iI [(ac -ac)
c sinx1 - (cT-c) . (xi +2x2) + (be
-c- be)
- x2 + ({t)cosx1 l
Hence,
where k = 0.64, P2 = 0.66, and Pl is an upper bound on
let f'} = P1 llxll2 + P2 and take

v ={ - A sgn(w),
- L
1-k
. 3£ I!' if 11lwl <e
The derivative of V(x) = xT Px along the trajectories of the perturbed system satisfies
. T t
V <-xx+-
- 4
Thus,
V ::; -(1 - B)xT x, "i llxll2 2 vfo

where 0 < () < 1. For D = 0.9, llxlb is uniformly ultimately bounded by
To ensure that lx1 I :S 0.01 and lz2 I :S 0.ol, choose e such that l.2724J! :S 0.01. In particular, take
=
f; 0.6 X 10- 4 ,
- 217
• 14.15 Take
e = A;y) [cy,y + A(y)(-2(wne - w~o- + v) - cJy]
= A;y) [<c-c)y"y+[A(y)-A(y)J(-2(wne-w~u+v)] +(-2(w,..e-w!a+v)

= -2(wne - w!o- + v + 5
where
1 _ A(y) - A(y) 2
6 = A(y) [(c - c).jy] + A(y) (-2(wne -wnq + v)
l6J < P3
U1
VY+ U412(wne + w~c,/ + U4lvl
< p(e,y,u) + 1!4lvl
where
The closed·loop system is given by
We can now design v a.s described in Section 14.2.1 to achieve ultimate boundedness with an ultimate bound
that can be made arbitrarily small by choosing c small enough. Therefore, we can ensure that the tracking
error IY - rl <µfort~ T , for any specifiedµ.
• 14.16
• 14.17 (1)
x = -x + x2[u + 6)
It takes the form (14.47) with f = -x, G = x 2 , and r = i. Take 1/) = O and V = ½x2 . Then, w = x 1 .
0
From
=
(14.48), V --yx3 , "'f > 0.
(2)
:i; = x2
x[u - x«S)
-
It takes the form (14.47) with f = x 2 , G = -x, and r = -x. Take T/) =x +x 2 and V = ½x2 . Then,
=
w = -x2 • From {14.48), v -yx\ "f > 0.
• 14.18
iJ = fo(TJ, l), i =A,,{+ B.,-y(x)[u - o(x)), y = Cc{

Let
e1 == y - r, e2 = y - r, ...
Then
Take
u = o(x) + "f(X)
~[r<P) - Ke+ v]
- 218
where K is designed such that Ac - BcK is Hurwitz.

e = (Ac - BcK)e + Bc(v + 6)
where
5 = 1(0 - ct] + 1 ; 1 (r(Pl - Ke+ v)
lt51 .$ Po(x) + klr(P)I + kllKll2llell2 + klvl ~f p(t,x) + klvl
We can now design v as described in Section 14.2.1 to achieve ultimate boundedness with an ultimate bound
that can be made arbitrarily small by choosing E small enough. Therefore, we can ensure that the tracking
error IY - rl < µ for t 2: T, for any specified µ .
• 14.19 The solution is similar to the previous exercise.
• 14.20 (a) The nominal system

X1 = X2, X2 = -X1 - X2
has a Lyapunov function V(x) = :r7 Px = l.5zj + X1X2 + x~, where P satisfies the Lyapunov equation
p A + ATP = -I. With this Lya.punov function, we have w 2xT PB = x1 + 2x2. =
V:::: -llxll~ + wT (v + 6) :S -llxll~ - P1 llxll2lwl + Pl llxl12lwl $ -llxll~, for P1 !lxlhlwl ;::: E
On the other hand, for P1llxll2lwl < E, we have
VS -!lxll~ + [-¼(llxll2P1lwl)2 + llxll2P1lwl] $ -11.:z:II~ + i :S -½llxllt for llxJI~ 2: i

(b) Let 6(x) = 2(x1 + x2). Then, 161 $ 2V2llxll2- Choose Pl = 2v'2. The closed-loop system is given by
where vis of the order O(llxll~)- Hence, linearization at the origin yields the matrix [ ~ i ],which has an
eigenvalue in the right-half plane. Thus, the origin is unstable.
• 14.21 (a)
v $ -a3(11xlloo) + WT V + WT 6
2
< -a3(11xll00) - L 1j_wl - L r,lwil + llwll1 (p + l'.()llvll00)
iE/ E if.I
:5 -a3(11xlloo) - L (f7lwil) L 11lw1I + [fJ{l - Ko)+ Ko'l]

iEl E
2
-
if/./
t
i=l
lw1I
= -a3(11xll00) + L [- (T/lw,1) 2 + 11lw1I] + L [-qjwil + 11lwil]

iE/ ! iil
= -as(llxlloo) + L [- (7Jlw;1) 2
+ 11lw1I]
iEl e
(b)
. ~e
V ~ -a3(11xlloc) + ~ 4 ~ -oa(llxlloo) + k1e, k1 >0
iE/
From this inequality, we can show ultimate boundedness as in Theorem 14.3.
- 219
• 14.22 (a) Repeating the calculation of the previous exercise, we have
y < -q,2 + I: [- (11lwil) 2

+ lwilPi.P + K-01f1lwil 2 ]
iEI £ £
= -,t,2 + L [-(1- tco/ 77l:,l)2 + lwilP1¢]

iE/
$ -q,2 + L [-(1 - "o) (1J0lwil)2 + jw;IP1¢]

iEJ c
(b)
-(1 - "o) ('7olWil)

2
V· 5 - 2
1 4>2 + { - 21¢2 + 7ef
" [
, + lw,IJ>i<J,] }
The quadratic form in the braces can be made negative definite by choosing e small enough. Therefore,
V$ -½¢
2 , which shows that the origin is asymptotically stable.
• 14.23 (a}
v $ -<ti - '1o +7Jt¢llwll~ +(Po+ P1<l>)llwll2 + Ito fJo + 'M''llwll~

e e:
= -¢2 - .!.(1- tcoH'7o + 7J14>)11wll~ +(Po+ P1<l>)llwll2

E
< -q,2 + (Po + P1 t/,) [-~ llwlJ~ + llwll2]

.$ -</>2 + e(Po+ Pi</>)
4
= _!¢2 _ ! [¢- ep1] 2+ c2pf + ePo
2 2 4 32 4
1 e2p2 ePo
< -203(Uxll2) + 321 +4
• 14.24 · The closed-loop system is
x = f + a (¢ - -yw + o)
Taking -y > PU4(1 - tco) ensures that V is negative definite.
• 14.25 Starting with the system

± =f + G(v. + tS)
add and subtract the term t/J to obtain
:i: = (J + Gv,) + G(u + t5 - v,}
View .S -1/J as the perturbation term and proceed to design u = v.
- 220
• 14.26
x = (f +Gt/I)+ 6 + G(v + 6)
Outside the boundary layer, we have
. T &V 2
V ~ -¢>2 +w (v + &) + &x ~ ~ -(1 - µ)</>
The right-hand side is negative definite for µ < 1. Inside the boundary layer, we have
.
V :$ -(1- µ)</J
2
+ P1¢llwll2 - 1/6
(1- K.o)7llwf12
2
The quadratic form can be made negative definite by choosing e < 4(1 - µ)(l - Ko)r(5/ pr, Therefore, the
origin is asymptotically stable.
• 14.27
(a) Take u = ¢(x) + v, where
_ '71e2(z}+,zow
{ llzlb
v = -[711p2(x) + 77o)w
where wT = ~G, a.nd TJo, 1J1 are positive constants to be chosen. Use Vas a Lyapunov function candidate.
For llxll2 ~ 1, we have
&V
V = ox(/+ Gt/,)+ wT(v + &)
II x 11 22 - 111r? + ~ ,1 2 T/lp2 + 1}0 2
S -c3 U·xU wll2 + Pllwll2 + l£o Hxll llwlb
fi:~
2 2
= -cs llxll~ - (1 - K.o) 111 110 llwll~ + pllwlb
2
< - ~ llxll~ - ~ llxll~ - (1 - K.o) ~~ 2 llwll~ + pJlwll2
Using llxll~ ~ llxll2 for llxll2 ~ l, we obtain
V :S - ~ Uxll~ - ll:lh [(1 - K-0)1/1 (pllwll2) 2 - (pllwll2)llxll2 + ~ llxll~]
Choosing Th > l/[2c3(l - Ko)) yields
For llxll 2 < 1, note that there is p1 > 0 such that p(x) ~ pi/lxl'2 for all llxll2 S 1.
V
av
= ax (f + G1/;) + w T ( v + &)
< -csllxll~ - (111l + '7o)llwll~ + P1 llwll2llxlb + Ko(1hp2 + ~)llwll~
= -c31lxll~ - (l - Ko)(1'/l.P2 + 1Jo)llwll~ + P11lwlhHxll2
C3 llxl/22 - (I -
< -2 '-o) ( f1o jlw 1122 -
C3 11:rlb
P1llxl!illwll2 + 2 2)
Choosing TJo > PU[2c3(l - t1:o)) yields
- 221
Therefore, the origin is globally exponentially stable.
(b) Set
f (x) = [ x~: x~ ] , G = [ ~ ] ,&(x, u) = a1 (x~ + x~) + a2u

Take 1/J(x) = -(x~ + x~) - X1 - 2x2. Then
x = f(x} + G(x)¢(x) = [ ~l : 2 ]x
is globally exponentially stable with V(:r) = :rT Px = xT [ ~~; !~; ] x and c1 = Amin(P) =1- 1/./2,
C2 = Ama:(P) = 1 + 1/./2., and C3 = 1.
&(x,1/J(x) + v) = 01(.x~ + x~) + a2[-(xr + x~) - (x1 + 2x2} + v} = (a1 - a2)(x~ + x~) -a2(x1 + 2x2) + a2v
Take
Clearly,
3 ./5
2llxll~ + 2 llxll2
p(x) $
Therefore, for· Jlxll2 $ 1, we have p(x) :$ P1llxll2, where Pl = (3 + ../5)/2. Choose 1Jo and '71 such that
77o > PU[2c3(l - 1£o)] and '71 > 1/[2c,(l - ~)]. Take 7/0 = IO and '71 = 2. The design is now complete.
• 14.28 (a) Consider i1 = x2 + sinx1 and view x2 as the control input. With x2 = -2x1 and V; ½x~ we
have V :5 -xf Now set z = x2 + 2x1 and Va = ½x? + ½z 2 •
Va = x1(x2 + sinx1) + z[81x~ + (1 + 62)u + 2x2 + 2sinx1]

= X1 (-2x1 + sin xi)+ z[x1 + 91:rf + (1 + 82}-u + 2x2 + 2 sinx1)
Take u = -x1 - 2:r2 - 2 sin :r1 - z + v to obtain

•
Va $-x21 -z2 +z(v+&)
where
= 81x~ -82(3x1 + 3x2 + 2sinxi) + 92v $ p(x) + ~lvl
fJ
where p(x) = 2xi + ½l3x1 + 3x2 + 2sinx1I- Take 17(x) ~ p(x)/(1 - 0.5) = 2p(x) such that 17(x) ~ TIO > 0.
Using Lyapunov redesign, take
v ={ -17 :gn(z),
- !l:.z
I! if TJ/zl < c
which yields
• 2 2 e(l - 0.5)
V0 ~ -X1 - Z + 4 = -Xi2 - Z
2 E:
+S
- 222
Near the origin, jp(x)I ~ k1lx1I + k2lzl. This can be used to show that the origin is asymptotically stable for
sufficiently small e. In particular, near the origin
2
v~ < -x~ - z2 + Jzl(k1lxil + k2lzl)- ~~lzl 2
2
< -xf - z2 - ~~lzl 2 + lzl(kilx1I + k2izl)
The RHS is negative definite for sufficiently small e.
= =
(b) Define e1 x1 - rand e2 x1 - f x 2 + sinx 1 = - r to obtain
e1 = e2, e2 = 01 x~ + (1 + B2 )u + (x2 + sin X1) cos X1 - r
Proceed as in part (a) to design a state feedback control that drives e towards zero. This time however you
can only achieve ultimate boundedness.
• 14.29
x1 = x2, X2 :;; -a sinx1 - bx2 +cu+ ((t) cosx1
Consider the equation ±1 =
X2 and view X2 as the control input. The origin x 1 ='O can be stabilized by
x2 = -x1 with the Lyapunov function Vi = x? /2. Apply the change of variables z1 = x 1 and z2 = x 2 + x 1 ,
to obtain
= -z1 + Z2,
i1 i2 = -asinx1 - bx2 +CU+ ((t) COSX1 - X1 + Z2
Take V = xU2 + zif';, = z;/2 + zU2.
\I= -zf + z2lx1 - asinx1 - bx2 +CU+ ,ct) COSX1 - X1 + z2}

Take u = -4z2 + v, to obtain
V = -z~ - (4c - 1)-Zi + z2c[& + v] :s; - z f - z~ + z2c[6 + v]
where we used the fact that (4c - 1) ~ (4 x 0.551- 1) > 1, and
o= !{-asinx1 - bx2 + ({t) cos x1)

C
satisfies the bound

l6J $ 16.1865lz1J + 0.242lz2I + 1.1111 ~f fJ
Take
v = { - r, sgn(z2),
- ~ ! ,
which yields
•
V $ -z1
2
-
2
z2
e
+4 $ -(1 -
2
B)llzlb, "I llzll2 2: V/e
48
where O < 8 < 1. For B = 0.9, llz)b is uniformly ultimately bounded by
1 r;
2y O= o.s21Ji
Using the fact that
x= [ ! 1 ~]z ~ !lxll2 ~ l.6l8llzlf2
choose c such that 0.527yg ~ 0.01/1.618 to ensure that lfxll2 ~ 0.01. In particular, take c = 10- 4 .
- 223
• 14.30 We start with the scalar system ±1 = xix2 and view x2 as the control input. We can globally
stabilize the origin with x2 = -x? which results in ±1 = -:ri- A Lyapunov function can be taken as
V(xi) = ½xf Now consider
Va = x1i-1 + {x2 + x~)(x2 + 2x1i-i) = xi(x1x2) + (x2 + x~){x1 + u + 2xfx2)

Add and subtract -x1.
Take
u = -{xf + x1 + 2x~x2) - (x2 + 2x~) = -(x1 + x2 + 2xix2)
V = -xi - (x2 + xi)
. 4 2 2
which shows that the origin is globally a.symptotically stable.
• 14.31 The system is in the form (14.53)-(14.54) with
J =a+ (x1 - a 113 )3, g = 1, fo; X1, Do= l
Take ¢(xi) = -a - (xi - a 1l 3) 3 - xi and V = ½:zf Use (14.56).
• 14.32 See Section 2.4.3 of {108].
• 14.33 (a) The system is feedback linearizable. Take u = -2x1 + xf - x2 ~r <f,(x). Then, x = Ax where
A= [ ~I !1 ] is Hurwitz. Hence, the origin is globally exponentially stable.
(b) Let,€= z - q,(x).
. aq,
x =Ax+B{, {=v- -(Ax+B!)
8x
where B == [ ~ ] . Let V = xTPx + ½{2.
V = -xTX + 2xTPB,€ + e[v - ::(Ax+ B,€)]
Take
a
v -(Ax+ B{) - 2xT PB - e
= 8:r
V' = -xT x-,€2
which shows that the origin [in the coordinates (x,,€}] is globally exponentially stable. Note that in the
original coordinates (x,z) we can only conclude global asymptotic stability due to the nonlinear change of
variables { =z - <J,(x).
• 14.34 Start with .ii= -xi + x2 and view x2 as the control input. We can take t/)i(x) = 0 which results
=
in x1 -xi. A Lyapunov function can be taken as Vi = ½xr
Now, let V0 = ½(zj + x~)-
- 224
which implies that the origin of the second-order system
is globally exponentially stable. From the third equation
it is clear that the origin of the full system is globally asymptotically stable.
• 14.35 Treat fJ as an unknown parameter, but assume that a will be available on line. Our design, however,
should work with any a that satisfies lal :::; I. Start with z1 =
x2 + Ox~ and view x 2 a.s the control input.
Let e1 = x1 - a sin t. Then,
el =
X2 + 8xf - a COS t
Take x2 = -k1 e1 + a cost ~f c/>1 (e1, t) where k1 > 0. With Vi = ½e~, we have
V1 -k1e~+8e1x?:::; -k1e~+2ledlx11 2
~ -k1e? + 2leil(le1/ + a) 2 ~ -kfe? + 2le1/ 3 + 4leil 2 + 2le1I
By choosing k1 sufficiently large, we can make e1 ultimately bounded and the ultimate bound can be
made arbitrarily small. Now, proceed to the second step of the backstepping procedure. Let z 1 ei and =
z2 = =
x2 - ',61 ( e1 , t) X2 + k1 e1 - a cos t.
i1 = -k1z1 +ex?+ z2, i2 = X3 + u + k1 (-k1z1 +Ox~+ z2) + a sin t

Let Va = ½z: + 6; zi. The rea.wn for including the constant b will be revealed shortly.
Take
Va = -k1z; + th1x~ + b2 k18x?i2 - k2z~ ·

< -k1Zi + 2lz1l(lz1I + a) 2 + 2b2 k1(1z1I + a) 2 lz2I - k2~
< -k1zi + 2lz1/3 + 4lz11 2 + 2lz1I + 2b2 k1(1z1/ 2 + 2lzil + l)lz2I - k2z~
Due to the cubic terms on the right-band side, we need to limit our analysis to a compact set. Let n=
{Va{z) $ c}. Choose c > 0 such that z(0) E 0. Using the fact that llx(O)ll 00 $ 1, we have
lz1(0)I = lx1(0)I ~ 1
lz2(0)I = lx2(0) + k1e1 (O} - al ~ 1 + k1 +a~ 2 + k1
The initial state :t2 (0) depends on k1. To be able to choose c independent of k1 , we have included the
=
coefficient b2 in the Lyapunov function. Take b 1/(2 + ki). Then
V11 (z(O)) = ½z;(O) + 2(2 : ki )2 z~(O) $ ½+ ½== 1

Therefore, we ta.lee c = I and limit our analysis to the set O = {V (z) ~ 4 l}. Notice that lzil ~ J2 in n.
Therefore
- 225
2b2 k1(lzd 2 + 2lz1I + l)lz2I :5 (2 + v'2)1zil lz2I + lz2I

where we have used the fact that k1/(2 + ki):;? :5 1/2. Thus,
"
Va ~ -k1z1
2 1n
+ (2v2 + 4)z12 + 2lz1I - k2z2
2
+ (vIn2 + 2)lz1I lz2I + Jz2I
Take k1 = 2v'2 + 4 + 2a: and k2 = 2o, where o > 0. Then

VG ~ -ollzll~ + vSllz!l2 + [-allzll~ + (v'2 + 2)lz1I lz21]
By choosing o large enough, we can make the bracketed term negative definite. Hence,
Yo$ -crllzll~ + v'5llzlb 5 -(1 - .B)ollzll~, V llzlb 2: ./s/a:/3, 0 < fJ < 1

which shows that z is uniformly ultimately bounded and the ultimate bound is proportional to 1/o:. By
ct , ing a: sufficiently large, we can make the ultimate bound arbitrarily small. Finally, notice from the
i: 3 , 1uation that boundedness of x 1 implies boundedness of x 3 •
• 14.36
• 14.37 Apply ba.ckstepping to deal with the nominal system and then Lyapunov redesign to take care of
the uncertain tenn 6. Start :with Z1 =
-z1 + x1z2. Using Vi =
½xf, we take .2:2 = 0 to obtain -xi, Vi =
Now consider
Take xs = -x? - ~ to obtain
Now consider the whole system with
V = rr• + 2X2 + 2{x3 +x1 + 2z2

1 2 1 2 1 2 )2
V = -x? - x~ + (.2:3 + xf + x~)[-a(x) + 6(x) + u)

where a:(.x) is known. Take
u = o(x) - (x3 + x~ + x~) + v
to obtain
ti= -x~ -x~ - {x3 + x~ + xn2 + w(6 + v)
where w = x3 + xi + ~l Now apply Lyapunov redesign to finish the problem.
• 14.38 (a) Consider the system
and view x 3 as a control input. We saw in Exercise 14.7 that X3 = -2x2 globally stabilizes the origin with
V = (1/2)(x? + z~) as a Lyapunov function. To backstep, let Z3 = xs + 2x2 •
Let V0 = {1/2)(xf + ~ + z~).
- 226
(b) With o# 0, we have

i1 = -x1 + tanh x2,
. 1 2 2
Vo ::; - 2(xl + Xz) + X2Z3 + Z3(2X2 + 2X3 + U + 0)
. 1 2 2 2
U = -3X2 - 2X3 - Z3 + V => Va S - 2(x1 + X2) - Z3 + Z3(V + 0)
Take
v ={ -p(x) sgn(z3), if p(x) lzsl ~ E:
- p2(::).i3
I! '
For p(x)lz3 1 ~ E:, z3(v + o)::; 0, while for p(x)lz3 1 < e:, z 3(v + o) S e/4. thus,
V.Q $ - 21(X12 + X22) - 2

Z3
e
+4 1 -
:::; -2(1 B)Uz 12
'2, 'fl {2;
Uzll2 ~ Ve
where z = {x1, x2, zs]T and O < 8 < l. The conditions of Theorem 4.18 are satisfied with II· II = 11-11 2,
a1(r) = a2(r) = (1/2)r2, and µ = J(2e:)/9. Thus, llzll2 is ultimately bounded by a 11(a 2(µ)) = µ.
Consequently, lx1I is ultimately bounded byµ. Choose c such thatµ S 0.01.
V{2e
8 s 0.01 ~ E $
I}
2 X 10
-
4
Take () = 0.9 and e $ 0.45 x 10-4 •

• 14.39 (a) Let x1 = y - rand x2 = i;.
. . k
X1 = Xz, X2 =g - -X2
m
-1F + -mld
+m
x =Ax+ B(v + d), where A= [ O

-~
1
_{J:~A:}
l· B =[ O]
-;;
The matrix A is Hurwitz. Let P = pT > 0 be the solution. of the Lyapunov equation PA+ ATP = -1.
=
Take V == xT Px, w [BV/8x]B, and v -10w. It can be shown that =
vs -I/xii~+ 441'0
Therefore, x is uniformly ultimately bounded by the ultimate bound
Given any µ, we can choose 1'0 large enough to ensure that
Amo.~(P)
---<µ
>.min(P)
- 227
In summary
:i; = Ax+ B(z + d- '}'ow)

8F . + -{}F . O"f . 8"{ .
X3 + -X1 + -x2 = 0'1 (· ) + 0'2 (· ) U + 03 (· )d
i - -X1
8x1 8xs 8x1 8x2
where the functions a:i(-) are known and 02(·) fa 0. Let J.:'o = xTPx + ½z2.
.
V0 = -xT x+wz+wd-')'oW2 +z[a1(·)+02(·)u+o3(·)d]
Take u =- a 210 [ct1 (·) +w- 11z].
•
V,. :;;; -x x
T 4
+ 410 - ··11 lzl
2
+ zo3(·)d
It can be shown that by choosing ,o
and 11 large enough, we can achieve ultimate boundedness and the
ultimate bound can be made smaller than µ.
• 14.40 Start with the scalar system x1 = -xi + x~[z2 + 6(t)) and view z2 as the control input. This
=
system takes the form (14.47) with J = -xi, G x~, and r 1. Take ,t, 0 so that f + G,p = -x and = =
V = ½~-Then, w = xf. Using (14.48), the control is taken as
X2 = -')'1X~ ~f q'>(zi), 'Yl >0

Now Let
.Z1 = + z?[-")'1Z: + 6(t)] + zrz2

-z1
i2 = u + 3-rizl {-z1 + z:f-11zf + o(t)] + zf z2}
Consider
1 2 1 2
Vo(x) = 2z1 + 2~
V
0 = -zf + z~[-11z~ + o(t)] + zf z2 + uz2 + 311zfz2{-z1 + zi[--·rvl + 6(t)] + z~z2}
Take
u -- -z1s + 31'1•13
-Y1Z1 - 311z1z2
+ 327 4 +v
Vo= -z? + zf{-11zf + o(t)] + VZ2 + 3"(1ztz26(t)
Take
V == -1'2Z2(3')'1zt) 2 - Z2, ')'2 >0
v.. = -z12 -
0 _ -z12 - z 22 + -ki + -~
., ,2112z22 z 18 + 3")'1z14 z2u.[( t ) <
z22 + +z13[-11z13 +uJ:(t)) - a...,
411 4")'2
Therefore, the state is uniformly bounded.
• 14.41 The procedure is similar to the previous example. In the first step, we take x 2 = x 1 + x~ and
V = ½x~.
The final control and Lyapunov functions are
u = (1 + 2z1 + 4z:)(-z; - ')'1zf - z1z2) + zf- z2 -12z2(l + 2z2 + 4--y1zf)2zf
1 2 1 2
Vo= 2z1 + 2z2
where z1 = x 1, z2 = X2 - X1 - x~ -11xf, 'Y1 > 0, and 1'2 > O.
- 228
• 14.42 Let C = BT P and define the system

x = Ax + Bu, y = Cx
V = xT(PA + ATP)x + xT PBv. ~ yT u

Hence, the system is passive. It is also zero-state observable since the pair (A, C) is observable. By Theo-
rem 14.4, the origin is globally stabilized by Ui -k sat(y,). =
• 14.43 Consider the system
.
X2 = -Xi +v,3
V
0
= x31 x2 - x31x2 + x2v = yv

Hence, the system is passive. With v = 0,
y(t) : 0 => X2(t) :: 0 => X2(t) :: 0 => X1 (t) =0
Thus, the system is zero-state observable. Therefore, we can globally stabilize the system by w = -t/;(y),
which shows that the system
can be globally stabilized by u = -y = -x2.

• 14.44. With u = -(b + v)/a, the system can be transformed into
TJ = fo(TJ, O) + F(11, y)y, ii= v
The representation /0(11,y) = /o(r,,O) + F(11,y}y is always possible. The system y = v with output y is
passive. Thus, the system takes the cascade form (14.80)-(14.82). It can be verified that all the assumptions
of Theorem 14.5 are satisfied. Thus, the origin can be globally stabilized by the feedback control of (14.84).
• 14.45 With u = -x1 + x2 + x1x3 + v, the system can be written as

[ d:1
X3
]
=[ -
X1 -
X1
2xs
]
+
[ 1 ]
x1 x2'
x2 =v
which takes the cascade form (14.80}-(14.82) with y = x2 as the output. Consider the (unforced) driven
system
and take W = (,\/2)xi + (l/2)x~.

A
] T [ -l/2
= -AX12 + X1X3 - 2
=- X1
0
W 2X3 [ .X3
The 2 x 2 matrix is positive definite when >. > 1/8. Ta.ke A= 1. Thus, all the conditions of Theorem 14.5
are satisfied and, by (14.84), a globally stabilizing state feedback control is given by
- 229
• 14.46 With u = -x~ + 1 - x2 + v, the system can be written as
X1=-(l+y)x~, [±X3 2 ]-[

-
Xs]
-x2 + v '
y-x3
-
which takes the cascade form (14.80)-(14.82} with z = X1. The driving system is passive, as ca.n be seen
from V =
½(x~ + x~)-
It is also zero-state observable since
y(t) =0 :::} .:i;3(t) =0 . =0

:::} X2(t)
For the driven system, we have W =

¼xf. Thus, all the conditions of Theorem 14.5 a.re satisfied and, by
(14.84), a globally stabilizing state feedback control is given by
V = X~ - X3 :::} U = -X~ +}- X2 + xr - X3

• 14.47 (a)
y = b + au = b - ; - a~ sat (~)
•Y;a [ (~ -i) •-Pu sat{n]

For IYI 2::: e,
yiJ ~ a [u(y)lyl - .BIYI] ~ -ao.BolYI
Hence, y reaches the boundary layer {1111 ~ e} in finite time. The rest of the analysis is similar to the proof
of Theorem 14.1.
(b) Similar to the proof of Theorem 14.2.
(c) b(rJ, y) and b(y) are continuous functions. Hence, they are bounded on any compact set. a and a are
bounded from below.
( e) The system has relative degree ooe since Ji ; -x, -<r, - u.

transforms the system into the globally defined normaJ form
The chauge of variables T(x) ; [ x,:, x, l
.
= -7'/J. + y 3 ' "12· = -'b + 'f/12 - ( '72 - y )3 ' '/I= -y - (Tn - y) 3 - U
'11
It can be verified V(1J) = ¼11t + ½~ satisfies the inequalities stated in the sta.tement of the exercise. The
functions a and b are given by a = 1 and b = -y + (172 - y )3 . We take a = 1 and b= -y - y3 . The control
can he taken as
u = - [Y - y 3 - (e + /3o) sat(~)]
• 14.48 (a) Let A, B, and K be diagon:al matrices whose diagonal elements are 0-i, b;, and k;, respectively.
The closed-loop system is given by
x = f(x, -Kz), i =-Az + by, y = h(x)

The derivative of W = V + ½zTK B-1 z is given by
W = V + zTKB- 1 i :s; -zTKy- zTKB- 1 Az +zTKy = -zTKB- 1 Az $ O
z(t) := 0 => y(t) = 0 => x(t) =0
- 230
due to zero-state observability. By the invariance principle, the origin is globally asymptotically stable.
(b) Repeat the analysis of part (a). For global asymptotic stability you need
1z, ,Pi(a) da -+ oo as lz•I -+ oo

so that the Lyapunov function will be radially unbounded.
(c) Take u = mg sin8 - k11e + v, where e = () - 61 and kp > 0.
mfe+/.pe = V
The conditions are satisfied with V = ½mle2 + ½ki>e2 • The control is given by v = -kz where
z(s) bs
e(s) - ~
• 14.49 It follows from Lemma. 4. 7
• 14.50
• 14.51 (a)
Assuming~> 0, we can divide through by >.c1, to obtain
(b) For constant w, iq, and Ad, p is constant. Whenever /J ¢: 0, p will grow unbounded. However, the
transformation depends only on sin p and cos p, which are always bounded.
(c} Flux regulation: the closed-loop system is given by
).d =- R,. Ad+ R,.M [>.r - k(>.d - Ar)]

Lr L,. M
= - (k + 14
L,.
) (>.d - ).,.) (14.1)
Hence, Ad approaches,\,. monotonically. The time constant can be controlled by choosing k.

(d) The closed-loop system is
;.
Ad = - -R,. \
Ad
R,.M l
+- -d sat
('A,./M - k(>-.d - >.,.))
L,. L,. Ia
- 231
At the initial time, .\,, - Ar < 0. There are two possible cases:
In the first case, i, = I, and the closed-loop system reduces to
>., will exponentially approach Mld. with the time constant L,./R,.. Consequently, >.,./M - k(>..d - >.,.) will
exponentially approach >..,./M - kM(Jd - >.,./M) < >...,./M < 111 • Therefore, there must be some finite time
ti such that at t = t1, >.,./M - k(>.d - ,\r) = Id and the closed-loop system is described by (14.1) fort~ t 1 .
>., will exponentially approach>.,. fort ;?:: t1 with the time constant L,./(R,. + kL,.) which is smaller than the
time constant L,. / L,.. Therefore, the overall settling time is less than 4Lr / R,. .
If, on the other hand, >.r/M - k[,\d(O} - >.,.] < Id, the closed-loop system will be represented by (14.1) for
all t ~ 0 and the settling time will be 4L,./(R,. + kL,.). With the choice k = 0, the second case will apply
and the settling time will be 4£,. / R,.
(e) Speed regulation: assume >.d =>.,.and consider the equation
Jw = kt>.riq - Ti ~r Jµ,i 9 - Ti
whereµ= kt,>.,./J. Let e = w -w,. ..
e=w-w,. =µ,i 9 -T /J-<1>(w)/J-w,.
0
Assume
ira/J+4>~)/J+w,.I $1q-b, b>O
over the set {lel $ c}. Take iq = -I,, sat(e/e) . Then

ee = e [-µIq sat G)-Ta/J - <J,(w)/J -w,.]
:S µ [-Iqe sat(~) + (1 0 - b)lel}
For lel ~ e, ·
ee s -µblel
Hence, for all le(O)I $ c, e(t) reaches the set {lel Se} in finite time.
(f) Let eo = e and z = koeo + e. Then
i = koe + µi" - T0 / J - 4,(w)/J - w,.

Assume
I koe - Ta/ J ~</J(w) / J - w,. t S Iq _ b, b>0
over the set n = {leol S c/ko, lzl Sc}. Take i 0 = -lq sat(e/e). Then, for all (eo,e) En and lzl ~ E:, we
have
zi = z [-µIq sat(;)+ koe -To/J -<P(w)/J -w,.] S -µblzl
which shows that all trajectories starting n reach n~ = {leol S E:/(ko01), lzl S e} in finite time, where
0 < 81 < 1. Inside n~ we want to show that the trajectories asymptotically approach an equilibrium point
where e = 0. The equilibrium point will correspond to the steady-state condition w = w,., w = 0 and
T0 = constant. The equilibrium equations are
- 232
Hence, at equilibrium, e = O and i = koeo = -(c/µJ/q)i¢(w) -T 0 ]. Let
eo = eo - eo, i=z-i
The new state variables (eo, z) satisfy the equations
~o = -koeo + z, i = ko(i - koeo) - (µlq/c)i + t/;(e, t) + v(t)

where l/J = [<J>(w) - ¢,(wr)]/J and v = -[¢(wr )-</>(wr)]/ J -wr. Assuming that¢, is Lipschitz with a Lipschitz
constant l, it can be seen that 11/JI $ {l/J)lel = (l/ J)li-koiol and limt-+oo v(t) = 0. Using V = (1/2}(e~+i2 )
as a Lyapunov function candidate, we have
V = -koe5 + eoi + koi kSieo - (µlq/E)i 2 + i'I/J + iv

2 -
< -koe~ + leol Iii+ koi 2 + k8lil leol- (µI,i/e)i 2 + (l/J)lil 2 + (kol/J)lzl leol + iv
= [ leol ]T [ ko -0.5(1 + k~ + kol/J) ] [ leol ] _
- Iii -0.5{1+~+.kol/J) (µIq/e)-ko-l/J Iii +zv
Choose € small enough that the 2 x 2 matrix is positive definite; that is,
ko[(µlq/c) - ko - l/J]- 0.25(1 + k5 + kof.JJ) 2 > 0
Then, it can be shown that (eo(t),z(t)) tend to zero at t tends to oo.

(g)
= k1Ad(i11cosp-iasinp)
-(X11 -.\)][c?3p
11
-sinp]
sm p cos p
[~d]-r
i 11 L
Assuming Ad> 0, we can divide through by Ad, to obtain
- 233
ed = [-(Xo - ..\o)sinp+ (A11 - >.11)cosp],o + (Ao - ..\ 0 )cosp + (111 -A11)sinp

(/4-R,.). (Rr-R,.) · ]
= eqp +
O [
- R,. A
L,. (>.a - >.a) -

A
pw().b - ..\11) - Lr Ao+ Lr Mit1 cosp
R,. .
+ [ - Lr (>.11 - Ab)+ pw(Aa - >.,.) -
- {/4-R,.)-
Lr Ab+
(R,.-R,.)
Lr Mi11 sinp
l
. R,. (/t..-R,.) (Rr-R.-) .
= eqp - y-ed - pweq -
r
L
r
>.d + L Mid
r
R,. R,.Mi9 (R,. - R,.) .

= - L,. ed + Lr>.d eq + Lr (Mid - >.d)
. .
e.9 = [-(Xo -
= -edj,- [ -
>.o)cosp-(.X1,- >.1>)sinpJp- (Aca --\,a,}sinp+ (X11- .\1>}cosp
R,..
-(Aa
4
•
- Ao) - p:.,(A1> (R,.-R,.). (Rr-R,.)
- .>..)- - - - A o + -----Mia
4 4
l
sinp
R,..
+ [ - Lr (Ab - A1>)
•
+ p:..,(..\o - >..,) -
(Rr-Rr)·
Lr _A1> +
(R,.-R,.)
L,. Mi1> cosp
l
. R,. (R,.-R,.}M.
= -edp - Lr eq + p..;ed + Lr lq
R,. R,.Miq (R,. - R,.) .

= - L,. eq - Lr>.d ed + L,. M Zq
{h) Repeat parts 3 and 4.
(i) With V = (1/2)(e~ + e:), we have
V. • R,. ( 2 2) (R,. - R,.) ·

= eded + e 0 e 0 =- L,. ed + eq + L,. Me9 i 0
< - R,.(l - a) llell2 - R,.o llell 2 + (/4 - R,.) Mflelllq

L,. L,. L,.
< - R,.(~:a)Bell 2 , \f llell;?: l(R,.a~R,.) Mlq
where O < a < I. Using Theorem 4.18, it can be shown that
fle(t)ll < exp [- R,.(~: a) t] fle(O)II + I(l4a~Rr) IMIo

= Ar [exp(--yt) (lle(O)II) + (R,. - R,.) M/9 ]
>.,. R,. a>.,.
where -y = (1-a}R,./L,..
(j)
i = koe + i£~ [(,\r - ed)i0 + e0 >.,./M] -TL/J-w,.
Restrict analysis to a region where ,\,. - ed > 0.
. 2pM(' ) [ (z) I I feq>.,.JM + (3LJ/2pM)(koe -T1/J - w,.)f]
zz ;:; 3LJ 11,- - ed -z1q sat € + z (>.,. _ ed)
- 234
The controller will work if

le,,>.r/M + (3LJ/2pM)(koe-TL/J - wr)J < l _ b
(Ar - ed) - q
• 14.52 (a)
s = Ae + e = Ae + ij - iir
= Ae+M- 1 (-Ctj-Dq-g+u)-ijr
= Ae +M- 1 [-Cq- Dq- g+ Mv + L(Cq +g+Miir -MAe)]-iir
= v+L\
where
L\ = (M- 1 M - I)v + Ae + M- 1 [-Ctj -Dq- g + L(Cq + g + Miir - MAe)J - iir
When L =O,
L\ = (M- 1 M - I)v + Ae + M- 1 [-Cq - Dq- g] - iir
When L= I,
L\ = (M- 1 M-l)v + Ae +M- 1 [-Cq- Dq-g + (Cq+ g + Miir -MAe))-iir

= (M- 1 M - I)(v - Ae + iir) + M-1 [(6 - C)tj - Dq + g - g)
(b) Write L\ as L\ = (M- 1 M -J)v + 8, where 8 is independent of v. Let p be an upper bound on 11811 00 ,
(c)
Choose
f3 ?. -1 p
- .Ito
+ /Jo, /3o > 0
SiBi $ -(1 - Ko),Bjsil + (I - Ko)(/3 - .Bo)lsil ~ -(1- Ko)Polsil
Hence, si reaches the boundary layer {js,I ::::; c} in finite time. Inside the boundary layer, we have e = -Ae+s.
Since A is diagonal (A= diag(>.1, . .. , >.m)),
eiei :S ->.;ef + e-le,I $ -(1 - 0)J..;e~, '.i)

where O < e < 1. Hence, e; reaches the set {leil ~ c/{0>.;}} in finite time. The ultimate bound on e can be
estimated by (c/B)JI:~ 1 (I/>.i) 2 since
Lm(e~- )2 = ~ I:m( ~)2

1
i=l I i=l 1
( d) We recover the previous result with a finite region of attraction because the analysis has to be limited
to a set where the inequality f3 ~ Po+ pf (I - K.o) is satisfied.
- 235
• 14.53 (a)
s = Ae + M- 1 (-Cq - Dcj - g + u) - ij.,.
W = srMs+ ½sTMs
= sTMAe + sT[-C(s -Ae + cj.,.) - Dq - g+ u] - sT Mij.,. + ½sTMs
= +½sr(M - 2C)s +sr[MAe + C(Ae -q.,.) - Dq - g + u - Mq.,.]
= sT[MAe + C(Ae - q.,.) - Dcj- g·- Mij,. +u]
(b)
W ::::: sT(MAe + C(Ae - q.,.) - Dq- g- Mij,. +v + L(-MAe- C(Ae- q,.) - g + Mtir)]
= sT(v+ A)
where
A = M Ae + C(Ae - Qr) - Dq - g - M ij,. + L( -MAe - C(Ae - q,.) - f; + M;;,.)
WhenL=O,
..6. = M Ae + C(Ae - tir) - Dq - g - M ij,.
When L=I,
A =MAe+C(Ae-q,.)-Dq-g- Mij,.- MAe-C(Ae-q,.)-iJ+ Mij,.
(c) For llsll2 ~ e, rp(s/c) = s/llsll2, and for llsll2 < e, rp(s/e) = s/e. Thus, for 11s112 2:: £, we have
W = sT (-/3 11 ; ,2 +A) $ -,8llsll2 + llsll2IIAll2

Choose /3 ~ 11..6.ll2 + Po, with /3o > 0, to obtain
W $ -.Bolls/12 $ -/Joe, for W > >.Me2

This inequality shows that the trajectories reach the positively invariant set {W $ AMe2 } in finite time.
=
Inside the set, we have 11s112 $ ey').M/>.m. Using the equation e -Ae + s we can calculate an ultimate
bound one. In particular,
which shows that the ultimate bound on llell2 is proportional tot.

(d) We recover the previous result with a finite region of attraction because the analysis has to be limited
to a set where the inequality f3 ~ IIAll2 + /Jo is satisfied.
• 14.54 Simulation results for the different control laws are given in Figures 14.2 to 14.5. The controller
para.meters a.re A = 101, e = 0.05, and P = 30. The control saturation levels are U1 6000 and U2 5000. = =
• 14.55 Simulation results for the different control laws are given in Figure 14.6. The controller parameters
=
are Kp = Kd 70001. The control saturation levels are U1 = 6000 and U2 = 5000.
• 14. 56 The equation of motion is given by
D(8) [ .1 ] = [ m u·2 Slnt1

·u l) _ k ] , where D(O) = [ / + mL2 mL cos (J ]
Xe Xe mLcose m+M
-
236
104
-
X
2 - - - -.........- - - - - - - - - - - , 1
1 0
o""'-,.:;____...________.._____,
LJ
-1
0 2 3 0 1 2 3
2.--------.----------, 5000
. 0
-5000,.._---~-------'------'
2 3 0 1 2 3
Time Time
(a) Let
V = ½vT D(B}v + ½kx~, where v = [ :r: ]
· 0 -mL sinB ] · k .
= v T D ( B)v. + 2vT [
1
V -mL sin f) 0 v 8 + XcXc
= [ 8 Xe. ] [ mLiJ2 sinu IJ - kxc

]
+2
1 [ ·
(J
.
Xe
] [ -9i'cmLsin8 ]
-(J2mL sin (J
.
+ kxcXc
= /Ju
Hence, the system is passive. Check zero-state observability with y = 8. When ti= O,
y(t) =0 => O(t) =0 => 9(t) = 0 and 8(t) =D(constant)

8(t) =0 and O(t) =0 => Xc(t) cos()= 0 and (m + M)!ic = -kxc(t)
cos0-:/,: 0 => Zr:(t) =0 ==> Xc(t) =0 ==> xc(t) =0
Thus, y(t) =0 does not imp}y that fJ(t) =0. The system is not zero,.st&te observable.
(b) Let U = -¢,1 (IJ) + w.
· v = B[-<t>i(B) +w) + ¢-1(1J)e = Ow

Hence, the system is passive. Check zero-state observability with y = 8. When w = 0,
y(t) =0 => O(t) =0 ==> 9(t) =0 and fJ(t) =O(constant)
O(t) =O and B(t) =o => mLxc(t) cos8 = -¢1(8) and (m + M)fic = -k:t'c(t)
cos() 'f- 0 implies that Xc(t) is identically zero. Then,
mLic cos8 = -4>1 (IJ) => ¢1 (0) == 0 => fJ =0
- 237
X 10'4
2 1
u1
1 q1 0
0 -1
0 1 2 3 0 1 2 3
2
u2
1
q2
-5000
1 2 3 0 1 2 3
Time Time
Hence, the system is zero-state observable.

(c) Since the system from w to iJ is passive (with a radially unbounded storage function) and zero-state
observable, it can be globally stabilized by w = -<h(8}. The overall control is u = -4>1 (8) - ¢>2(8).
(e) Simulation results are shown in Figure 14.7 for u,, = U., =
0.05 and K,, = K'IJ 0.1. The settling time =
is about 30 sec.
=
(f) Simula.tion results are shown in Figure 14.8 for Up= U., 0.05, K,, K 0 0.1, and two different values = =
of!: 0.S and 0.1. As we reduce~. we l'ecover the performance under state feedback.
• 14.57 The equation of motion is given by
D (8) r· Zc
-~ ] = [ m.i.n-
r Li2 • u n 1.. ] ,
where D(0) = [ I +mL2 mLcosfJ]
sm u - ,.:z;e mLcose m+M .
A(fJ) = det(D(8)} .
(a)
mLD
th = xe + m+ M cos e = T/2
mLO mL92
th = Xe+
m+ Moos(} - M sin(}
m+
1] 1 [ m +M -mL cos (J ] [ u ] mLiJ2 . (J
= [ L
m~McosO A(fJ) -mLcosfJ l+mL 2 mL02 sinO-kxc - m+M st0
1 [m LA2 . fJ - k Xe ] - mL82 sm
rr sm . 9
m+ M m+M
k k ( mL . )
= m + Mxc = - m + M 1Jl - m + M sw 113
773 = e={
{ = jj = A~O) [(m + M)u - mLcos8(mL62 sinO - kxe)]
The transformed system takes the regular form (14.4)-(14.5).
-
238
X 104
2
q1 0
1
-
LJ
-1
1 2 3 0 1 2 3
2 5000
1 q2 0
1 2 3
-5000'-----------------'
1 0 2 3
Time nme
(b) Consider the system
iJ1 = 772,
with the Lyapunov function Vo(77) .
. Vo.: - (k2173 - k1 ( 171 - m :LM sin fJJ) cos113] = -¢l-(r,) :5 O

2
Vo =0 =? q,(r,(t)) =0 ~ TJ"J(t) =0 ~ 113(t) =113(constant)

cos 713 # 0 * ='11 (constant) ~
T/1 (t) 111 (t) =0 => 712{t) = 0
712(t) =0 ~ f'h(t) =0 * mL sin°113 = 0 *

f/1 - · T/3 =0 ::? 711 = 0
m+M
By the invariance principle, the origin T/ = 0 is globally asymptotically stable. Thus, the sliding surface can
be taken as s = 0, where
(c)
s = jj + k28- k1Xc + k1Xc8sin8
= A:O) l(m + M)u - mL cos8(mL02 sin 8 - kxc)] + kl} - k 1xc cos(}+ k1xcOsinO
= mA:(/7 {u + m: M [-mLcos8(mLe 2 sinO- kxc) + A(8}(k29-k1 xccos8+ k1 xc8sin8)]}
Choose f3(x) to satisfy
/J(x) 2'. /3o + m: M 1-mL cos8(mL02 sin 8 - kxc) + A(O)(k28 - k1Xc cos9 + k1 xc8sin9)j
- 239
X 104
2~-------.-----------, 1.--------.------------,
Q'"'-':::_ _ _.......,_ _ _ _ _ _ _ ___.
0 1 2 3
-1 "-----------------....J
0 1 2 3
2 .---------.------r------, sooo,.,....------.------'--------,
-5000...__ _ ___.__ _ _ _..__ _ __,

1 2 3 0 1 2 3
Time Tme
for some /30 > 0. Then, the control law u = -/3(x) sat(s/µ) guarantees that
ss S -{Jo "'a7()7 Isl, for Isl 2:: µ
Hence, all trajectories reach the boundary layer {Isl $ µ} is finite time. Assuming that the system
7j1 = T/2, ~= m: M ( 'h - m :LM sin 113) , 173 = </,(71) + s
is input-to-state stable, we can show, as in the proof of Theorem 14.1, that all trajectories reach a set fl:€
in the neighborhood of the origin. Local analysis will then show that for sufficiently small µ the origin is
locally exponentially stable. Hence, for sufficiently small µ, all trajectories converge to the origin as t tends
to infinity.
(d) Since
w.nishes at the origin, for any /3 > /3o, the inequality
{) ~ f3o + m: M 1-mL cos0(mL82sin9 - kxc) + ~(9)(k29 - k1±c cos(J' + k1xc6 sin0}j

is satisfied in some neighborhood of the origin. Limit the analysis to such neighborhood.
= =
(f) Simulation results a.re shown in Figure 14.9 for k1 1000, k2 1, fJ = 0.1, andµ= 0.05. The settling
time is about 4 sec.
= = =
(g) Simulation results are shown in Figure 14.10 for k1 1000, k2 1, f) 0.1, µ 0.05, and,= 0.1. The =
response under output feedback is very close to the response under state feedback .
- 240
X 104
2 1
0
------
q1
-1
0" u1
0 2 4 6 0 1 2 3
2 5000
------ u2
0 \_
1
q2
-5000
2 4 6 0 1 2 3
Time Time
4
0.01
3 0.02
Xe 0.005
2 8
0 0
1 \. -0.005
0 -0.02
-0.01
-1
0 20 40 0 20 40 0 20 40
Time Time Time
Figure 14.7: Exercise l4.56(e).
- 241
4----------, 0.05 . . - - - - - - - - - - ,
2 u
8
-2
-0.02
_4L---------' -0.05 .....__ _ _ _ _ _ __,
0 20 40 0 20 40 0 20 40
4---------, 0.1 . - - - - - - - - - -
0.02 E =0.1
0.05 u
2 9
0 a•--------~
o•--------1
-0.05
-0.02
-2L.---------'
0 20 40 0
-0.1 ' - - - - - - - - - - '
20 40 0 20 40
Time Time Time
Figure 14.8: Exercise 14.56(f).
4---------,
0.1
3 0.02
u
0.05
2
0 0
1
-0.05
0 -0.02
-0.1
-1 L---------'
0 2 4 0 2 4 0 2 4
Time Time Time
Figure 14.9: Exercise 14.57(f).
4---------,
0.1
0.02
0.05
0
-2 -0.05
-0.1
-4L.---~-----
o 4 2 0 2 4 o 2 4
Time Time Time
Figure 14.10: Exercise 14.57(g).
- 242

Hassan K. Khalil - Complete Solution Manual For Nonlinear Systems-Pearson - Prentice Hall PDF

Uploaded by

Document Information

Original Title

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

Hassan K. Khalil - Complete Solution Manual For Nonlinear Systems-Pearson - Prentice Hall PDF

Uploaded by

Copyright:

Available Formats

Nonlinear Systems

l[iij] 1' :,. .l. ~ ~ }ti 1;J•.

• 1.1 Let Zt = 11, %2 = 11<1), •.• , Zn= 1l(n-l)_

• 1.4 Let z1 = q, z2 = q,· z =[ z.z12 ] E Ram ·

u = r-,J,(t,11) = r- ¢(t, Cz)

(1.22+ l.7cosz1)sinz1 -1.l~ =0

equilibrium point (z, 0, 0.4075/ Binz).

(c) With E, =constant, the model reduces to

(b) Let X1 = iL, Z2 =VC,

The model of (a) is more famUiar lince it is the pendulum equation.

• 1.12 The equation of motion is

Let %1 =11 8.11.d %2 =fl.

• 1.14 The equation of motion is

Mii = F-m(jj + L8cos8-LiJ2 sin8) -kj

[ ii9 ] = 1 [ m +M -mL cos 9 ] [ mgL sin 9 ]

z, = 6 = Ll~B) [(m + M)mgLsin9-mLcos8(F +mL.82 sin8- k1i))

= .6.(~i) [-m2L2gsinz1 cosz1 +(I+ mL2)(u + mLzi sinz1 - b,.)J

Substituting F. and Fr in the 8-equation yields

Substituting F. in the Zc-equ&UOD yields

M!e = -m!c - mL8 coe9 + mLIP sin 8 - kxc

(d) Take u = constant. Setting the derivatives z, =0, we obtain z 2 = z, =0 and

• 1.18 (a) z1 =11, z2 ='ii, zs = i, u = v.

= _!__ [-Rz3 + Looz2za2 +u]

! (Lh d).) A(>.) = Wi - k,/pih

(b) X =p-p. = pgh.

Using z = ll.p as the state variable, we obtain

Writing !1 =,f,(w,,), we can rewrite the previous equation as

z1 = dz11 -dz1 +r1

The assumptions r1 = JJX1, r2 = rifY = JJZ1/Y, and z11 =0 yield

(b) When µ = Jlm:Z2/(km + z2), the equilibrium equatiOllS are

from the fint equation,

( y ( ~2/ - µ,,_~ d) t #'m~ d) &Dd (0, Z2/}

(c) Whenµ= µ,,.z2/(km + z2 + k1zl}, the equilibrium equations are

d(z21 - !2a) - µ(!!2a)!1/Y =0 ~ z1 =Y{z21 - !2a)

(Y(z21-z2.),f2c), (Y(z21-!2•),!n), and (0,z21)

x2 = -x1 => 0 = 2x1(ri - l} => X1 = 0, 1, or - 1

: 1(1,-1) = [ !1 !1 ] -> >.1,2 = 2 ± v'8 => (1, -1) is a saddle

From the second equation, z2 = 0 or x2 = 2(1 + x1),

x2 =2(1 + :r:i} => 0 = (x1 + 3):r:1 => X1 = 0 or :1:1 = -3

Eigenvalues are 1 ± j; hence, (0, 0) is unstable focus.

Similarly, (-1,-1} is a stable node.

z2 = 0 => 0 = z1 (zt - 16) => z1 = 0, 2, or - 2

Similarly, ( -2, 0) is a saddle.

~Ll,2) = f ! !1 J .- .\1,2 = -(1/2) :l:: ;'1i5/2 => {-1, 2) is a stable focus

Figure 2.1: Exercise 2.3(1). Figure 2.2: Exercise 2.3(2).

a = tan(a/2) =t- . a Aj 2.3311

:t2.3311,0) = f o.~2 -o.!10s ] => >.1,2 :r: 0.6892, -1 ~ (2.3311,0) is a saddle

regions diverge to infioity.

(3) There is a unique equilibrium point at the origin.

Point Jacobian matrix Eigenvalues Type

[ -1.2361 -1 ] -0.2361, -2.2361 stable node

((3 + ../5)/2, (3 + ./5)/2) [ 3.2361 -1 ] 4.2361, 2.2361 UDStable node

Figure 2. 7: Phase pOrtraits of Exercise 2.4.

(b) Transform the state equation into the polar coordinates

r= {zt +~, 8 =tan- 1 (::)

and, for O < ro < 1,

• 2.6 (a) The equilibrium points are the real roots of

0 = -z1 + a:r:2 -bz1z2 + zi, 0 =-(a+ b)z1 + bzf - z1z2

X1 [-(a+ b} + b:t1 - X2] =0 * Z1 =0 or Z2 = -(a+ b) + bzl

Hence, there is an equilibrium point at ( ~$2, -1~t,•>).

The eigenvalues of A are

(O,O) {0,-a) (1:"t:', -lw).)

Figme 2.8: Exercise 2.6(i). Figure 2.9: Exercise 2.6(ii).