IEEE TRANSACTIONS ON CIRCUIT THEORY, VOL. CT-18, NO.

3, MAY 1971 375

Synthesis of Nonlinear Systems with

Prescribed Singularities
LEON 0. CHUA, SENIOR MEMBER, IEEE

Abstract-This paper presents an affirmative answer to the operating points can be “frozen” simultaneously. The
heretofore unresolved question as to whether an autonomous non-
linear system can always be synthesized to possess a finite number of
trajectories in the vicinity of eachsingular point would then
prescribed singular points, as wall as a prescribed sat of aiganvalues correspond to the zero-input response.
associated with each singular point. The answer is given in the form Since second-order nonlinear systems can be readily
of a new canonic nth-order nonlinear system containing (3n-2)
single-valued functions. In the case n=Z. the canonic system is shown analyzed by geometric concepts [4]-[7], we will present the
to possess many nice general properties which are useful for detar- canonic system for this case first. The generalization to
mining aspects of the system’s global qualitative behavior, such as nth-order systemswill be presentedin Section IV.
global stability.
II. SECOND-ORDER CANONIC REALIZATION
I. INTRODUCTION
Our solution for the case n = 2 is given by the following
HIS PAPER is concerned with the problem of canonic system of two normal form equations :
synthesizing an nth order autonomous nonlinear
T
system [l] 21 = Sl(X2)

i = f(x) )i-2 = fi(x2k72W + h2b2) (2)

(1)
with a prescribed set of q isolated simple singular points’ where the four functions gl( *), g2( *), f2(. ), and h,( *:$are
such that the Jacobian matrix af(x) evaluated at each assumedto be of class C’ [8]. The Jacobian matrix associ-
singular point has n prescribed eigenualues.2 A partial solu- ated with (2) is, therefore, well defined and is given by
tion to this problem was given by Vallese [2] for n=2, but
even then only singular points located along the x2 = 0 axis
could be prescribed.The main difficulty lies not so much in
J(x)-
a
[ I[
b
c d =
fi(X2kml)
0

where the primes denote differentiation with respect to the

f;(x2)g2w+~2(X2)
d&2) ’
1
(3)

realizing the prescribed singular points, for it is easy, at

least for n=2, to construct two functions fi(x,, x2) and respectivearguments. Let xCk)be the kth prescr&d singular
f2(x1, x2) such that upon setting them to zero a set of solu- point with coordinates (x(:), x’,)), k = 1, 2, . * *, q. For con-
tion points with the prescribed location is obtained. Un- venience, we will assume that no two singular points lie
fortunately, the problem becomes much more formidable along a vertical or a horizontal line.3 The two critical
if the eigenvaluesassociated with each singular point are parameters which completely characterize the type of each
also prescribed. This is becausein trying to satisfy both singular point x(k)are of course the two eigenvalueslik) and
requirements, extraneous singular points are almost always AT)of J(xck’).For second-ordersystemsit is more convenient
generatedin addition to those that were actually prescribed. to work with the following equivalent set of parameters,
The main contribution of this paper consists of a con- namely,
structive solution to the preceding problem in the form of a Tck)=a+d=I$k)+@=f;(x(Zk))g2(x\k))+h;(x(Zk)) (4)
new canonic system of n state equations. The solution to this
Atk’= ad - bc = &“‘A$‘) = - f2(x$‘))g;(xfk’)g’ (x\~)). (5)
problem implies that a nonlinear system can be synthesized
such that the behavior of the trajectories in the vicinity of We will now devise an algorithm for synthesizing the four
each singular point can be prescribed. This observation is unknown functions so that the resulting system will possess
somewhat analogous to the “frozen-operating point” preciselythe q prescribedsingular points with a correspond-
concept introduced by Desoer and Peikari [3]. The basic ing pair of eigenvalues(A’:‘, A’,“‘)for each singular point.
difference is that our system is autonomous, and hence all q I) Synthesizing the Function gl(x2): Let g1(x2) be any
function of class C’ which intersects the x2 axis at x(:), xi2),
Manuscript received June 8, 1970; revised October 15, 1970.This work x2
(3)
>-..v x2’. For simplicity let us choose the polynomial
was supported in part by the National Science Foundation under Grant
GK 2988.
The author is with the School of Electrical Engineering, Purdue 91(x2) = fi (x2 - x1”‘). P-5)
k=l
University, Lafayette, Ind. 47907.
i A point f is said to be a singular point of (1) iff(f) = 0. Otherwise, it is
said to be a regular point. A singular point f is isolated if there exists a 3 There is no loss of generality in this assumption becauseif indeed it is
neighborhood of P containing only regular points. A singular point f is necessaryto realize two or more singular points with an identical xi or x2
simple if the Jacobian determinant det [af(x)/ax],,,#O. coordinate, then one can always carry out a rotational transformation,
* These eigenvaluesneed not be distinct. Complex eigenvalues are as- synthesize the system by the technique presented in this paper, and carry
sumed to be specified in complex-conjugate pairs. out an inverse rotational transformation to obtain the final system.

Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
 376 IEEE TRANSACTIONS ON CIRCUIT THEORY, MAY 1971

2) Synthesizing the Function g2(x1): Let g2(x1) be any where g; ‘(. ) denotes the inverse of g.J. ) and where de-
strictly monotonically increasing and onto function of notes the composition operation. Sinceg2( . ) is a one-to-one
class C’. For example, one could choose g2(x1)=x1. function, only one value of xi can correspond to x(,),
3) Synthesizing the Function f2(xJ: With gl(x,) and namely, x1 =x ‘1”‘.This contradicts our earlier assumption
gZ(xl) already determined, the function fi(xZ) is chosen to that x* is extraneous, and hence, the above algorithm
be any function of class C’ which passesthrough the points realizes the specification exactly.
f,(+$)), k = 1,2, . . . , q where We will illustrate the above synthesisprocedure with the
- A(‘4 following two examples.
f2W) =
g;(xy)g;(xp))
(7) Example 1
=W
k.

In particular, we can apply the well-known interpolation Synthesize a second-order system with two singular
theorem to construct a (q-1)-degree polynomial [9], points located at xl’) = (0, 0) and xc2)= (3, - 1). Let each
namely, singular point be a stable focus with identical eigenvalues
~',')=;1':)=-l+jJZand~11)=1(,2)=-l-j~.

f2(xJ = t c(,x;-? Corresponding to the prescribed eigenvalues,we obtain

(8)
fl=l the trace T(l) = Tc2)= - 2 and the determinant A(l) = Ac2)= 3.
If we let a and w be q x 1 vectors with components equal to From (6) we obtain
CI,and w,, respectively, where n = 1,2, . . . , q, then it is easily gl(x2)=(x2-xx(21))(x2-x(zz))=(x2-O)(x2+1)=x~+x2. (16)
shown that
For simplicity, let us choose
a = M-lw (9)
g2b1) = Xl. (17)
where M is a q x q matrix whosejkth element mjk is equal to
[s’:“]“- 1, Substituting (16) and (17) into (7), we obtain w1= -3 and
4) Synthesizing the Function h,(x,): Let h,(x,) be any
w2= 3. From (9) we obtain
function of class c’ which passesthrough the points
h,(x’,k’) = - fi(x’z”‘)gZ(x’lk’) G =k, k = 1,2;.., q (10)
with the following slopes:
K2(x’2k’)
= T(k)- f;(x’z))g2(x’:‘) E pk, k= 1,2, . . . , q. (11)
For simplicity, let us choose. a Hermite interpolation Hence, substituting a, and ~1~from (18) into (S),we obtain
polynomial of degree(2q - 1) [9] : f2(x2) = a1 + a2x2 = - 3 - 6x,. (19)
h,(x,) = 2 /?,,x;-~. Substituting (17) and (19) into (10) and (ll), we obtain
(12)
n=l cl = 0, c2 = - 9, and pl = - 2, p2 = ‘16.From (13) we obtain
If we let B be a 2q x 1 vector with components equal to /I,,
n=l, 2,.-e, 2q and z be a 2q x 1 vector whose first q com-
ponents are equal to g,,, n = 1,2, * * . , q, and if the remaining
q components are equal to P,,, n= 1, 2, . . . , q, then, it is
easily shown that
/?=p-lz (13)
-2
where P is,a2q x 2q matrix whosejkth elementpjk is given by
-15 . cw
pj,,= [x$q”- 1, 1IjSq -41
= (k - 1)[x(2j)lk-‘, q+lljI2q. (14) Substituting pl, p2, p3, and b4 from (20) into (12),we obtain
An examination of the above construction procedure h2(x2)=~1+~2~2+~3~;+~4x;= -2x,-15x:-4x;. (21)
shows that x(l), xc’), . . . , xc4)are indeed the singular points,
and that the trace Tck)and determinant Ack)associatedwith Finally, substituting (16), (17), (19), and (21) into (2), we
each of theseq singular points are also precisely realized. It obtain the desired normal form equations
remains for us to prove that there are no extraneoussingular i, = x: + x2
points. Our choice of gl(x,) in (6) guarantees that the x2
coordinate of all singular points can only assumeone of the 1, = - 4x; - 15x; - 6x,x, - 3x, - 2x,. (22)
q prescribed values. Hence, if there exists an extraneous It is easily verified that (22) does realize the specification. It
singular point x*, it must assumethe coordinates (xl, XT)) might appear somewhat paradoxical to be able to find a sys-
where tem with only two singular points, both of which are stable.
x1 = g;‘n [ - Mx’zk’Yfi(x!?)] (15) Indeed, all known bistable systems such as switching cir-
Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
CHUA : SYNTHESIS OF NONLINEAR SYSTEMS 311

lO.OI--STRBLE FOCUS 13.-l I--SIRBLL FOCUS

,“.U,--SIHHLE FOCIJS I 3.-l I--5TRBLE FOCUS

Separatrtx

lO.OI--5TRBLE FOCUS 13.-l l--STABLE FOCUS lO.OI--STRElLE FOCUS [3.-l I--STRBLE FOCUS

UXIII = XIPl*Xlzl + Xl21 OX(I1 = Xl21~XIZl + Xl21

Ox~zl:~-6.~X~zI-3.I.XI,l-~..X(zl..3-l5.~X~zl**z-z..xl~l OX121sl-6..Xlzl-3.l~XlI1-~.=xlzl~~3-ls.nXIzl99z-z.~xlzl

(4 (4
Fig. 1. Phaseportraits associatedwith second-order systemin Example 1.

cuits [7], [lo], [ 111have at least one unstable singular point to locate the separatrix and to study the qualitative proper-
(usually a saddlepoint) separatingthe two stable points. To ties of this system, a phase portrait, corresponding to dif-
i-esolvethis paradox, one must take the singular points at ferent regions of interest in the phase plane, is constructed
injinity [5], [6], [12] into consideration. In the next section using the computer program CAPPA [13], as shown in Fig.
we show that there are only two singular points at infmity 1W(d).
associated with the canonic system (2). For this example
both of theseare unstable. Hence, the trajectories emerging ExamP1e2
from infinity must converge toward either one of the two Synthesize a- second-order system with three singular
stable singular points, depending on which side of the points located at x(l) = (3, l), x(‘) = (2, 2), and xC3)= (1, 3).
imaginary separatrh the trajectories originated. In order Let the first singular point be a stable node with eigenvalues
Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
 378 IEEE TRANSACTIONS ON CIRCUIT THEORY, MAY 1971

/z\” = - 1 and &i) = - 2. Let the secondsingular point be an

unstable node with eigenvaluesA’:)= 1 and ;1i2)=2. Let the
third singular point be a saddle point with eigenvalues
1:“) = - 1 and ;1i3)=2.
Corresponding to the prescribed eigenvalues at each
singular point, we obtain T(l)= -3, A”‘=2 for the stable
node Tc2’= 3, Ac2’= 2 for the unstable node, and Tc3)= 1,
Ac3)i - 2 for the saddle point. Upon carrying out the above
synthesisprocedure, we arrive at the following second-order
system :
rn.4) Isocline
il = x2” -6x; + 11x, - 6
s-2,= 0.5x; - 4.5x; + 12.5~; - 2x,x; - 2.5x;
+ 9x,x, - 8x, - 35x, + 32. (23)

III. QUALITATIVE PROPERTIES OF THE SECOND-ORDER

CANONIC SYSTEM
The system (2) is said to be canonic because it is the
simplest system possessingthe full capability of realizing
any prescribed set of singular points.4 In view of the gen-
erality of this system, it is desirable to uncover the qualita-
tive properties which remain invariant, i.e., properties in-
dependent of the specific form of the four nonlinear func-
tions. These properties are extremely helpful in predicting
the globul behaviors of the trajectories. As an application of Fig. 2. Qualitative properties of system in Example 1
these properties, we will prove in the latter part of this
section that the system synthesizedin Example 1 is globalZy Proof: Each horizontal line x2 = ~1”)is clearly an isocline
stuble. with m= co since gl(x(2k))=0 and dx,/dx, = co in (24). It
The phase-plane equation associated with the canonic follows therefore that the half-line to the right or to the left
system (2) is of the singular point is a segment without contact. This
implies that all trajectories crossing the kth horizontal line
dx, Qh ~2) = .h(x&z(x1) + h,h).
-= (24)
on one side of the kth singular point must point to the same
dx, P(Xl, x2) 91(x2) direction. The sign of dx,/dt on this line asxi + co is identical
to the sign of f,(x$‘). Therefore, for all trajectories crossing
The isocline with slope m is obtained by equating (24) to m the kth horizontal line on the right of (x(lk),xp)), we have
and solving for x1 : dx,/dt<O if fi(x$‘)<O, and dx,/dt ~0 if J2(x$))>0.
Q.E.D.
x1 = g;’
[
mgl(x2) - Ux2)
“f&2) 1 .
(25) Applying Property 1 to (22) of Example 1, we conclude
immediately that the two horizontal lines x2 =0 and
x2 = - 1 are isoclines with m = co. Since fi(x$i))<O and
Property I
f2(x(22))>0,we conclude that the direction of the vertical
The horizontal line x2 =x(,k)passing through each of the directors must be as shown in Fig. 2.6
4 singular points (x(lk),x’,“))is an isocline with slope m= co.
The singular point (x(:), xik’) partitions this line into two Property 2
segments without contact.’ In particular, the direction of the The 4 horizontal lines representing the m= co isocline
trajectories to the right of the singular point must be’ partition the phase plane into (4 + 1) horizontal strips such
downward if wk= f,(x$‘)) < 0, and upward if wk= f,(x’,))> 0, that within each strip the sign of dx,/dt remains unchanged.
where wkz f,(xik)) is defined in (7). Moreover, the sign of dx,/dt alternates between adjacent
strips, starting with dx,/dt >O in the uppermost strip.
4 The adjective “simplest” is used loosely here to mean the minimum Proof: Without loss of generality, let us assumi:
number of nonlinear functions needed to specify the system. By a rather x(:)>xy.. . ~~(24).Then it follows from (6) that dx,/dt >O
tedious though straightforward analysis, it can be shown that it is not pos- for x,>x(,l), dx,/dt<O for x(~~)-cx~<x(~~), dx,/dt>O for
sible to find a system of equal generality with fewer than four nonlinear
functions. It can also be shown that there exist three other canonic forms ~(2~)
<x2 <x(,z), etc.
containing four nonlinear functions with the same capability as (2). How- Q.E.D.
ever, we do not present them here becausethey are obtained simply by a
permutation of the four functions in (2).
5 A segment in the phase plane is said to be a “segment without con- ’ A director is a short line segment on the phase plane such that a tra-
tact” if no trajectories are tangent to it [6]. jectory must be tangent to it.
Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
CHUA: SYNTHESIS OF NONLINEAR SYSTEMS 379

In view of Property 2, the sign of dx,/dt at any point on

the phase plane can be determined by inspection. For
example, the phase plane in Fig. 2 must contain three
horizontal strips, with dx,/dt >O for x2 >O, dx,/dt <O for
-l<x;tO,anddx,/dt>Oforx,<-1.
Property 3
The m=O isocline is made up of (p + 1) disconnected
branches of the equation

x I =g;’ ~- M-d (26)

[ f2W 1
where p<q - 1 is the number of distinct real roots of
fi(xZ)=O. Each branch j of this isocline is asymptotic to
the horizontal line x2 = tj where fi(tj) =O. Moreover, the
(p+ 1) branches of the m=O isocline partition the phase
plane into mutually exclusive regions such that within each
region the sign of dx,/dt remains unchanged,and such that
the sign changesonly when the trajectory crossesa branch
of the m = 0 isocline.
Proof: Equation (26) follows immediately from (25) with
m=O. The function - h,(x,)/f,(x,) is clearly made up of Fig. 3. Stereographic projection showing that the canonic system has
two and only two singular points at infinity corresponding to the four
(p+ 1) branches,corresponding to the p zeros of f2(x2)=0. singular points A’, A”, B’, and B” on the equator.
Sincef2(x2) is a (q - 1)-degreepolynomial, we have p I q - 1.
Sinceg; ’ ( . ) is a one-to-one function, (26)representsa total roots of f2(x2)=0. This observation is usually sufficient to
of (p+ 1) disconnected branches. The last part of Property establish the nonexistence of limit cycles. For example, an
3 is obvious sincedx,/dt = 0 on eachof these(p + 1)branches. inspection of the phase-plane analysis shown in Fig.. 2
Q.E.D. clearly shows that this system cannot support a limit cycle.
In view of Property 3, one only needs to sketch the loci In view of the above properties, we can now assert.tbat
of (26) to obtain the (p+ 1) branches of the m=O isocline. the canonic system(2)exhibits two distinct types of isochnes.
These branches then partition the phase plane such that There are q horizontal isoclines corresponding to m= co
the sign of dx,/dt at any point can be determined by inspec- which pass through the q singular points and (p+ 1) hori-
tion. For the example shown in Fig. 2, the different regions zontal isoclines with slope m given by (27). The remaining
are indicated by different shadings. isoclines consist of (p + 1) disconnectedbranches as defined
1%
c
by (25).
Property 4
Property 5
The horizontal line x2 = tj is an isocline with slope
The canonic system(2) has exactly two nonsimple singular
m _ h2(t.i)

\ I
points at infinity.7 One singular point lies along the x1 axis
gl(tj) and the other singular point lies along the x2 axis.
where cj is any real root of fi(x2)=0. Moreover, this hori- Proof: In order to investigate the behavior of trajectories
zontal isocline is a segment without contact; i.e., all tra- at infinity, it is convenient to consider the stereographic
jectories crossing this line must point in the samedirection. projection of points on the phase plane onto the surface of
Proof: Let x2 = tj be any real root of f2(x2) = 0. Then (24) a unit sphere [5], [12]. For example, Fig. 3 shows the
reduces to (27) which is a constant independent of xi. projection of a point P(x,, x2) on the phaseplane into corre-
Hence, the horizontal line x2 = tj is an isocline with slope sponding points P’ and P” on the lower and upper hemi-
m given by (27). Moreover, cj#x’,k’ implies that the line spheres,respectively. The point at infinity along any fSty
x2 = 5j lies within a horizontal strip described in Property 2 x2 = kx, in the phaseplane then corresponds to two points
and is, therefore, a segment without contact. on the equator (z=O). In particular, the point at infinity
Q.E.D. along the xi axis corresponds to the two points A’ and’;?“.
Similarly, the point at infinity along the x2 axis corresponds
For the example shown in Fig. 2, this horizontal isocline to the two points B’ and B”. What Property 5 assertsis t$at
correspondsto x2 = - 0.5 and m= 9. Observe that all direc- thesefour points on the equator are actually singular points
tors along this line point in a downward direction (as they of (2)at infinity.
should since x2 = -0.5 is a segment without contact).
One immediate consequenceof Property 4 is that if the
system has one or more limit cycles, they must necessarily ’ The behavior of nonsimple singular points at infinity is quite complex
[5], [12]. For example, a nonsimple singular point could have more than
be bounded betweenthe p horizontal isoclines with slope m two separatrices. Hence, Property 5 does not imply that there is only one
given by (27), where plq- 1 is the total number of real integral curve which leavesinfinity and comes back to infinity.
Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
 380 IEEE TRANSACTIONS ON CIRCUIT THEORY, MAY 1911

To prove that A’ and A” correspond to the singular point system.It is well known that if the point at infinity is stable,
at infinity along the x1 axis, let then trajectories outside a small neighborhood of stable
1
singular points in the finite plane may neverthelesstend to
Y
x1 =-; X 2 =-. infinity. This global instability behavior is clearly unde-
z z
sirable in practical systems. While theorems on global
This transformation. is equivalent to projecting the trajec- stability exist [l], the hypothesis is usually too strong for
tories on the surface of the sphere onto a vertical plane systems characterized by nonmonotonic functions. For-
tangent to point A’ [6], [12]. Substituting (28) into (24), we tunately, our canonic system is endowedwith the preceding
obtain after some algebraic manipulations the following nice general properties; therefore it is a simple matter to
phase-planeequation in terms of the transformed variables test for global stability. First, we draw horizontal lines
zand y: through each singular point to obtain the m= co isoclines.
Next we sketch the m = 0 isocline as given in (26). We then
dz - zg 1(Y/Z)
-- (29) invoke properties 1, 2, and 3 to determine the “sense” of
6 - .f2(Y/4cl,Ul~) + h,(YlZ) - YY,(Y/Z)’ the trajectories within each region bounded by these iso-
clines.’ For the systemof Example 1, theseregions are indi-
Following the procedure described in [12], we conclude cated with different shadings in Fig. 2. An examination of
that the singular points at infinity are the zeros of this figure readily reveals that no, trajectory can go to in-
finity. For example, any trajectory originating in the fourth
z2q-l[zg1(y/z)] = 0 (30) quadrant under the m = 0 isocline must eventually inter&t
and this isocline and “bend downward.” Consequently, no
trajectory originating in the uppermost strip of Fig. 2 could
~2q-‘[f2(Y/4Y2w) + h,(Yl4 - Ygl(YM] = 0. (31) tend to infinity in the direction of x2. By uniquenessargu-
ments [6], no trajectories originating in the uppermost strip
The multiplication factor z2q-1 is introduced in order to in Fig. 2 could tend to infinity along the x1 direction either
remove the denominator in z from the bracketed expres- becausethey all must eventually intersect the m = co isocline
sions in (30) and (31).* A careful analysis of these two
and “bend towards the left.” A similar analysis of trajec-
equations show that the only values of z and y satisfying
tories originating in the lowest strip reveals that no trajec-
(30) and (31) are z =0 and y=O. These correspond to the tories could tend to minus infinity. Consequently, this sys-
singular point at infinity along the xi axis, or in terms of the tem is globally stable and all trajectories must eventually
unit sphere, this singular point corresponds to points A’ converge to either one of the two stable singular points.
and A” on the equator. This conclusion is equivalent to saying that both singular
To prove that B’ and B” correspond to the singular point
points at infinity are unstable.
at infinity along the x2 axis, we must define another trans-
formation projecting the trajectories on the sphere onto a
vertical plane tangent to point B’, namely, IV. IZTH-ORDERCANONIC REALIZATION
We will now generalizethe canonic second-order system
Xl = ylz; x2 = l/z. (32) (2) t o an nth-order system such that the location and the
associatedeigenvaluesof a finite number of singular points
The transformed phase-planeequation is now given by can be prescribed. This canonic system of equations now
takes the following form :
dz -Zf2Wk72(YI4 + h2W ‘.
dy = g,UM - Y["Mw)g,(Yl4 + h,W] (33)
11 = Sl(X2)

i2 = .Mx2)gz(x3) + h,(x,)
A similar analysis will show again that z=O and y=O
13 = f3(x3)g3(x4) + h,(x,)
correspond to the only singular point for (33). But this . .
corresponds to the point at infinity along the ‘x2 axis, or . .
points B’ and B” on the equator. . .
Finally, a straightforward but lengthy analysis similar to )I-,-1 = f,- 1(x,- &In- lb,) + L 1(x,- 1)
that given in [12] shows that the Jacobian determinant is %I = L&J&h) + kdx,) (34)
zero at both singular points. Therefore, the two singular
points at infinity are nonsimple. where the (3n-2) functions yj( ), fj( . ), and hj( . ) are
Q.E.D. assumed to be of class C’.
In order to derive a constructive procedure for identifying
We will now illustrate some applications of the preceding
the (3n- 2) unknown functions, let us first obtain the follow-
properties in determining the global stability of our canonic
9 It would also help to sketch the isocline specified’by (27) in Property 4.
’ Recall that the highest degree of the polynomial functions in (29) is However, this is often unnecessaryif one is interested only in determining
(2q- 1) corregponding to the degree of h,(x,) as defined in (12). global stability.

Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
CHUA : SYNTHESIS OF NONLINEAR SYSTEMS 381

ing characteristic polyno,mial associatedwith (34):

P-4 Bl 0 0 .. 0 0
0 (a2 - 4’ B2 0 .. 0 0
0 0 (a3 - 4 B3 .. 0 0
0 h - 4
.. .
D(A) = ..

0 ..@n-1 - 4 A-1
in 0 0 .. 0 (%I - 4

= - qcr, - A)@, - 4. . . (a, - 4 z!c 8182 . . P, (35)

‘where the sign for the last term is + if n is odd and - if n Step 3: Construct the characteristic polynomial associ-
is even, and where ated with each of the 4 singular points, namely,
aj E f>(xj)gj(xj+l) + h>(xjh j = 2, 3, .. . , n - 1 (36) - DW(;l)= (a _ )$;))(A- @),’. . . (A _ ;i;k))
~1,= f%,)g,h) + h&J (37) =i”+A~k~l~n-1+A~~23,n-2+ ... +Ay%+Ag’,
/31= dl(x2) (38) k= 1, 2,. . . , q. (46)
Bj ZEfj(xjkl)(xj+ 11, j = 2,3, ... , n - 1 (39) Step 4: Equate the constant term in (41) and (46) as
P, = f,(x,)g’,(x1). (40) follows.
The eigenvaluesof (34) are simply the roots of the charac- a$’ = T [f2(x$k9f3(x$k9 . .f,(xkk))] [g;(xf))g;(x$k’)
teristic equation . . : g;- ,(xf’)g;(x’:‘)] = At’, k = 1, 2,. . . , ~1. (47)
--D(A) = 2 + anvllnT1 + a,-23J-2 + ... Assigning arbitrary nonzero valuesfor f3(xik’), j,(xF’), . . . ,
+ al/Z + a, = 0 (41) and f(xtk))
” ” in (47) we obtain

fz(x(zk))
= [f3(xs”‘,f,(xk”‘, . . fn(x;W)][g;~;~,g;(x~‘) . . g;- , (x;4)g;(x\!4)’ k = 1, 2, . . . , q.

where Step 5: From Step 4, we obtained the values of fj(xy’),

ji(xy’), . . , jj(xp’), . . . , jJx$@)Q= 2, . . . ) n). Therefore, each
a0 = T PIP2 . . . A = T [f2(~2).f3(~3). . ~.f&,)l
function fj(xj) is simply a (q-- 1)-degreepolynomial in xj
[d(x2k7;(x3)~ . .g;- ,c4g341 (42) passing through the above set of points and assumes,
and therefore, the same form as (8).
Step 6: It remains to construct the functions hj(xj).
aj = Fj(c12, u3, " ' ) %A j = 1,2;..,n - 1. (43) Equating (34) to zero, we obtain
We are now ready to exhibit the synthesisprocedure. Let
#), x(2) . . . , xcq) be the prescribed locations of the 4 h,(x$“‘) = - fj(xy))gj(x$ J, j = 2,3,. . , n - 1
singular bbints. Let = - j;(x$ygj(.Yy’), j = n. (49)
A!k)= Re A!k)+ j Im J.Ck)
I ) i= 1,2;..,n
(44) Next, we equate the coefficients of (41) with the corre-
be the n prescribed eigenvalues associated with the kth sponding coefficients in (46) and substitute (43) for aj to
. . . , xik’. The obtain the following system of (n - 1) nonlinear equations :
singular point Xck)with coordinates (x(:), x(zk),
synthesisprocedureconsistsof the following steps. a(b)= Fj(~ik),ay), . , @) = A’!’J , ,j = 1,2, ..., n - 1. (50)
J
Step I: Choose
Let a(,k),at), . . . , t&k)
n be a solution of (50).Then from (36)and
gl(x,) = fi (x, - XT’). (45) (37),we obtain
k=l
h’.(x’?) = a’!’ - f”.(X’k’)g.( !k) ) j = 2,3, ’ . . , n - 1
Step 2: Choose g2(x3), g3(xq),... , g,(x,) to be any J J J J xJ+l 2

strictly monotonically increasing onto function of class C’. = ,$’ - f;(x;!qgj(xy’b j = n. (51)

Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.
 3i2 IEEE TRANSACTIONS ON ClRCUlT THEORY, MAY 1971

Eqations (49) and (51) give us the numerical values of ciently small neighborhood of each singular point must re-
main invariant because the first-order terms completely
j;+, 1,2, . . .-,ni corresponding to eachof the q-singularpoints. characterize the local behavior through their associated
Therefore, hj(xj) is simply a (2q- 1)-degreeHermite poly- eigenvalues.This observation suggeststhat the introduc-
nomial passing through the above set of points with the tion of the concept of locul equivulence in nonlinear systems
pfescribed slopes.Clearly, eachfunction hj(xj) is of the same might lead to a fruitful area of future research.Another in-
f&m as (12). teresting researchproblem along the theme of this paper is
‘;.In view of the above construction procedure, it follows to determine what kind offlows can be synthesized [14].
that (34) will realize the set of 4 prescribed singular points
and eachof theseq singular points will correspond to exactly ACKNOWLEDGMENT
the n prescribed eigenvalues.It remains therefore for us to The author wishes to thank Prof. C. A. Desoer, Uni-
prove that there are no extraneoussingular points generated versity of California, Berkeley, for his suggestionsand en-
by this procedure. Again, the proof is identical to that given couragement; and D. C. Mossman, a former student, for
earlier in Section II and dependsprimarily on the fact that his contribution of some preliminary results during the
the functions g2(xgj, gs(xJ, . . . , g,(x,) are one-to-one and early phase of this research.
onto functions. Therefore, no extraneous singular points are
present. REFERENCES
[l] T. E. Stern, Theory qf Nonlinear Networks and Systems. Reading,
V. CONCLUDINGREMARKS Mass. : Addison-Wesley, 1965.
[2] L. M. Vallese, “On the synthesis of nonlinear systems,” in Proc.
1The canonic system presented in this paper provides an Symp. Nonlinear Circuit Analysis. New York: Polytechnic Institute
affirmative answer to the basic question of whether a system of Brooklyn Press, 1953,pp. 201-214.
Canbe found to exhibit a prescribed set of singular points [3] C. A. Desoer and B. Peikari, “The ‘frozen operating point’ method of
small-signal analysis,” IEEE Trans. Circuit Theory (Corresp.), vol.
tiith a prescribed set of eigenvalues.Although this author is CT-17, May 197d, pp. 259-261.
uuable to prove that (3n- 2) functions are necessary in t41 V. V. Nemytskii and V. V. Stepanov, Qualitatiue Theory ofDl$feren-
general for n >2, he has so far not been able to exhibit a tial Equations. Princeton, N. J. : Princeton University Press, 1960.
[5] G. Sansone and R. Conti, Nonlinear Differential Equations. New
simpler realization without introducing extraneous singu- York: Macmillan, 1964.
iarities. Observe that the term “necessary” is used in this [6] A. A. Andronov, A. A. Vitt, and S. E. Khaikin, Theory of Oscillators.
paper to mean that violating the necessarycondition implies Reading, Mass. : Addison-Wesley, 1966.
[7] L. 0. Chua, Introduction to Nonlinear Network Theory. New York:
that there exists at least one set of specifications for which McGraw-Hill, 1969.
the realization is impossible. This, of course, doesnot imply [8] W. H. Fleming, Functions of Several Variables. Reading, Mass.:
that for some given specifications, a simpler realization is Addison-Wesley, 1965.
[9] P. J. Davis, Interpolation and Approximation. New York: Blaisdell,
not possible. However, from a system theoretic stand- 1963.
point, it is imperative that a canonic realization, such as the [lo] J. K. Moser, “Bistable system of differential equations with applica-
ohe presentedin this paper, be found so that one is guaran- tions to tunnel diode circuits,” IBM J. Res. Deuelop., vol. 5, no. 3,
196 I, pp. 226240.
teed that there exists at least one method that always works. [l l] L. 0. Hill, D. 0. Pederson, and R. S. Pepper, “Synthesis of electronic
although the generalform of the canonic systemremains bistable circuits.” IEEE Trans. Circuit Theorv. ,l vol. CT-IO.1 Mar..1
unchanged, the individual functionsfj( . ), gj( . ), and hj( . ) 1963,pp. 25-35.’
[12] T. V. Davies and E. M. James, Nonlinear Differential Equations.
can vary widely for a given specification. In fact for n>2, Reading, Mass. : Addison-Wesley, 1966, pp. 42-52.
there appearsto be somedegreeof freedom left in choosing [13] L. 0. Chua and Y. F. Lam, “Programming and user’s manual for
CAPPA (computer-aided phase plane analysis),” Purdue University,
these functions to meet additional specifications, say the Lafayette, Ind., Rep. TR-EE 70-3, Jan. 1970.
eigenvectors. In other words, the realization is not unique. [14] R. Abraham and J. Robbin, Transversal Mappings and Flows. New
However, the local behavior of the trajectories in a suffi- York: W. A. Benjamin, 1967.

Authorized licensed use limited to: Indian Instt of Engg Science & Tech- SHIBPUR. Downloaded on March 10,2025 at 06:06:35 UTC from IEEE Xplore. Restrictions apply.