Systems & Control Letters 19 (1992) 467-473 467

North-Holland

Homogeneous Lyapunov function for

homogeneous continuous vector field
Lionel Rosier
Laboratoire d'Analyse Num#rique, Universit#Paris-Sud, Orsay, France

Received 6 December 1991

Revised 16 July 1992

Abstract: The goal of this article is to provide a construction of a homogeneous Lyapunov function P associated with a system of
differential equations J = f(x), x ~ R~ (n > 1), under the hypotheses: (1) f ~ C(Rn, ~ ) vanishes at x = 0 and is homogeneous; (2)
the zero solution of this system is locally asymptotically stable. Moreover, the Lyapunov function V(x) tends to infinity with 1[x [I,
and belongs to C=(R~\{0}, R)n CP(~ ~, ~), with p E [~* as large as wanted. As application to the theory of homogeneous systems,
we present two well known results of robustness, in a slightly extended form, and with simpler proofs.

Keywords: Local asymptotic stability; Lyapunov function; homogeneity; non-differentiable feedback.

1. Introduction

Kurzweil, in [9], proved the converse of Lyapunov's second t h e o r e m in the quite general framework of
a system of differential equations k = f ( x , t), where f ~ C(G x [0, + ~ ) , E~), G being an o p e n set in R ~,
and C(G x [0, + oo), E~) being the set of continuous maps from G x [0, + ~) into ~n. In the particular
case of a u t o n o m o u s systems on E~, he obtains the following result:

Theorem 1 (Kurzweil). If f ~ C ( ~ ", ~ ) is such that f(0) = 0, the trivial solution of the equation fc = f ( x ) is
strongly stable in ~ (see below for a definition) if and only if there exists a (so-called Lyapunov) function
V ~ C ~ ( ~ ~, ~) such that V(O)=O, V ( x ) > O for all x 4 : 0 , V ( x ) ~ +oo as I l x l l ~ +oo and
17V( x) " f ( x ) < 0 Vx 4: O.

A n important class of functions f for which one would like to study the differential system ~ = f ( x ) is
the class of h o m o g e n e o u s functions. A function g : Nn __. N (resp. g : ~n ~ ~ n ) is said to be homogeneous
if there exist (r 1. . . . . r~) ~ ((0, + ~ ) ) " and ~" ~ R such that Vx = (xi)i=l, n ~ R ~ \ { 0 } , Vt > 0,
g(t~lXl,..., tr, xn) = U g ( x ) (resp. Vi, Vx 4: 0, Vt > 0, gi(Ulxl . . . . . trnXn) = U+~'gi(x)).
It is natural to ask if h o m o g e n e i t y of V may be imposed when f is assumed to be h o m o g e n e o u s . Many
authors have replied in the affirmative w h e n f is in addition assumed to be of class C I (see [4, T h e o r e m
57.4; 10, T h e o r e m 36; 7, Proposition p.1246]). Extending this result to the m u c h m o r e general framework
where only continuity of f is supposed necessitates proceeding differently. To convince ourselves of this,
let us examine in detail the hypotheses used by H a h n (see [4]) in proving his theorem.
(i) First to define his L y a p u n o v function, H a h n needs uniqueness of the trajectory for a given initial
data. Indeed, he puts V(x):= f ~ l p ( ~ c , x ) l " d z where z ~ , p ( r , x) is the solution of ~ = f ( x ) starting
from x at t -- 0, and a a suitable exponent.
(ii) T h e n to prove that V is of class C ~, he must assume Of/Ox i exists and is b o u n d e d for each
i ~ {1 . . . . . n}. This implies that ~- > 0.

Correspondence to: L. Rosier, Laboratoire d'Analyse Num6rique, Universit6 Paris-Sud, B~,timent 425, 91405 Orsay, France. E-mail:
[email protected].

0167-6911/92/$05.00 © 1992 - Elsevier Science Publishers B.V. All rights reserved

468 L. Rosier / Homogeneous Lyapunov function

We remark that in his definition of homogeneity the variable t ranges over the whole set N, instead of
the interval (0, + ~). In the restricted case where r~ = 1 for all i, he can only treat values for ~" of the
form p / ( 2 q + 1), p, q ~ ~.
These hypotheses will not be required in the following theorem, which is the principal result of this
paper.

Theorem 2. Let f be a function satisfy&g :

(a) f ~ C(R", ~ ' ) , f(0) = 0;
(b) f is homogeneous: there exist (r I. . . . . r,) c ((0, +oo)) ~ and "c ~ ~ such that:

Vx:(xi)i=l., E[~n\{o}, Vt>0 fi(trlXl . . . . . tr"Xn) =t~'+rifi(x 1. . . . . Xn) ; (a)

(c) the trivial solution x = 0 of system k = f ( x ) is locally asymptotically stable.
Let p be a positive integer, and k be a real number larger than p.maxl<_i<_nr i. Then there exists a
f u n c t i o n P : ~ ~ --* ~ such that:
(i) V ~ CP(R ", ~) m C~(~" \ {0], ~);
(ii) V(0) = 0, V(x) > 0 for all x ~ 0 and ~'(x) ~ + oo asJI x [I ~ + oo;
(iii) ~" is homogeneous: Vx = (xi)i=l, . ~ ~"\{0}, Vt > 0 v(trlXl . . . . . tr"x,) = tkV(x);
(iv) VX ~ 0 I 7 ~ ( X ) " f ( x ) < O.

Remarks. (a) Using the homogeneity of f we shall prove that the zero solution of the system J = f ( x ) is
strongly stable in E" if and only if it is locally asymptotically stable. Thus, with the minimal hypothesis
concerning the stability at the origin we will obtain what we need to start our construction of V (namely
a Lyapunov function V) by using Kurzweil's theorem.
(b) In addition to the fact that the Lyapunov function P is homogeneous (with the same r i as for f ) it
can be as smooth as we want, except that, with our method, we cannot assert that V E C~(E ", E). (A C ~
P would necessarily be a polynomial: indeed, doing s ~ 0 in (7) (see below), we see that O " P = 0 for I a l
large enough.) Furthermore V(x) tends to infinity as [Ix II tends to infinity.
(c) Extension of the existence of a homogeneous Lyapunov function to the framework of continuous
(homogeneous) vector valued functions is not devoid of interest. Indeed, many smooth nonlinear systems
2c=g(x, u) can be stabilized by only continuous feedback laws x ~ u ( x ) (e.g. [2,3,8]). In that case
f ( x ) := g(x, u(x)) is only continuous.

We now recall how Kurzweil defines the different notions of stability in the continuous framework,
without requiring uniqueness of the trajectory for a given initial data. Let G be an open subset of ~"
which contains the origin, F the closed set ~ \ G, and oJ the function defined on G by:

w(x) =
I( 1
max IIx[I, d ( x . F )
2)
d(O,F)
if F is nonempty,

kllxl[ if F is empty.

Let f : G ~ ~n be a continuous function such that f(0) = 0. The zero solution of the equation

.~ = f ( x ) , x ~ G, (2)

is said to be:
(a) locally stable if for every e > 0 there exists a 6 = 6(e) > 0 such that, for any solution x ( t ) of the
equation (2) with oo(x(O)) < 6, defined for 0 < t < T, 0 < T < +o% there is a solution y(t) of the equation
(2), defined for all t >_ 0, such that

x(t)=y(t) for0<t<T and w ( y ( t ) ) < e fort>0;

 L. Rosier / Homogeneous Lyapunovfunction 469

(b) locally asymptotically stable if it is locally stable and if, in addition, there exists a 60, 0 < ~0 < 6(1),
such that if x ( t ) is a solution of the equation (2) with w(x(O)) < 6 o, defined for all t > 0, then x ( t ) ~ 0 as
t~ +~;
(c) strongly stable in G if there exist two functions B : (0, + oo) ---, (0, + oo) and T : (0, + oo)2 ~ (0, + ~),
with B increasing and l i m ~ o B ( / 3 ) = 0, such that for all /3 > 0 and e > 0, for every solution x ( t ) of (2)
defined on an interval [0, tl), w h e r e 0 < t 1 < + ~ , such that t o ( x ( 0 ) ) < / 3 , there exists a solution y ( t ) of
(2) defined on [0,+ oo) such that y(t) = x ( t ) for 0 < t < t 1, w(y(t)) < B(/3) for t >_ 0, and w ( y ( t ) ) < e for
t > T(/3, e).

R e m a r k s . (a) By the extension t h e o r e m (see [5, Th. 3.1, Chap. II]), one proves easily that the zero
solution of ( 2 ) / s locally stable if and only if for every e > 0 there exists ~ > 0 such that for every solution
x ( t ) of (2) with oo(x(O)) < ~, defined for 0 < t < T (0 < T < +oo), we have oo(x(t)) < e Vt ~ [0,T).
(b) In the definition of local stability and local asymptotic stability, to m a y everywhere be replaced by
[1 [[" Indeed, oJ(x) = [Ix 11 for all x such that IIx [[ < ½d(O, F).

T h e region of asymptotic stability A will be the set of points x 0 ~ G for which the following is true: if
x ( t ) is a solution of (2) with x(0) = x 0, defined for 0 _< t < T (0 < T < + ~ ) , then there exists a solution
y ( t ) defined for all t > 0 such that y(t) = x ( t ) for 0 < t < T and y(t) ---, 0 as t ---, +oo. It is clear that every
solution of J = f ( x ) , x ( t ) ~ G coming from a point of A at t 0 = 0, will remain in A for all times greater
than 0.
Let us assume that the zero solution of (2) is locally asymptotically stable. In this case Kurzweil proved
also (in [9, pp. 69-71]) that A is an open set containing 0, in which the zero solution of J = f ( x ) (x ~ A )
is strongly stable.
Before giving the p r o o f of T h e o r e m 2, we recall the main properties of h o m o g e n e o u s functions which
are used here. Suppose V : ~ " ~ It~ is a s m o o t h function which is h o m o g e n e o u s , m o r e precisely

VX=(Xi)i=l,nE~n\{o} V t > O ~ ' ( t r ' x , , . . . , t r " x n ) = t k k ' ( X , . . . . . Xn) (3)

w h e r e r~ . . . . , r~ are some positive real numbers, and k is a non-negative real number.

T h e first p r o p e r t y is that P is entirely defined by its values on S n- i := {x [11x II = 1}. I n d e e d the m a p
& :(0, +a¢) × S ~-1 -~ ~ \ { 0 } given by ~b(t,(y 1. . . . . y~)) = (tr~y 1. . . . . tr"y n) is surjective. As this function
will a p p e a r again, we p r e s e n t some of its properties in the following lemma.

L e m m a 1. (1) The map

&:(0,+oo) XS "-l~"\{0}, (t,(y, .... ,y,))~(tr'y,,...,ff"y~)

is a bijection, and its inverse function qb- 1, which we write ~ = (~bo,qq. . . . . ~b,), is of class C ~.
(2) The function q'o satisfies

lim ~bo(X)=0 and lim ~ b 0 ( x ) = + ~ .

x Ox~O Ilxll-" +~

W e give a sketch of the p r o o f of this lemma. Since the m a p t ~ ~ni=1 x 2i / t 2ri from (0, +oo) into itself
(for x ~ ~ \ { 0 } fixed) is decreasing and onto, it follows that ~b is bijective. M o r e o v e r the implicit
function t h e o r e m applied to g(t, x):= -,iv'n=lXi2/'2r'/° _ 1 shows that ~bo, and thus ~bi(x) =xi/qJo(X) r~ for
i = 1 . . . . . n, are C ~ maps. T h e remaining p r o p e r t i e s of ~b0 are easy to prove.
T h e second p r o p e r t y is that the partial derivatives 8F'/Ox i are also h o m o g e n e o u s . M o r e precisely

VX=(Xi)i=l,nE~n~{o}, Vt>O - - ( t r l X l . . . . . tr'xn) = t k-ri (X 1. . . . . Xn). (4)

~Xi ~Xi
470 L. Rosier / Homogeneous Lyapunov function

Indeed, by differentiating each m e m b e r of equation (3) with respect to xi, we get that

Vx:g0, Vt>0 t r ' - - ( t r ~ x l .... ,t rnX , ) = t k z - - - ( X l . . . . . X,)

~X i OXi

This property implies that V F ' ( x ) . f ( x ) < O on I~n\{0} whenever 1 7 V ( x ) . f ( x ) < O holds on S" 1.
Indeed, using (1) we get that for all y ~ S ~- ~ and t > 0,

17~'( tr'yl . . . . . tr"yn) "f ( t~'y~ ..... tr"yn) = t¢+klT~'( y ) "f ( y ) (5)

and, since the map 4~ is onto, we obtain VV(x) "f(x) < 0 for all x ~ 0.
For the convenience of the reader, we now describe the organization of the paper.
In Section 2 we give the proof of T h e o r e m 2. First we establish a preliminary proposition, which
claims that the zero solution of J = f ( x ) is strongly stable in ~ ' . To accomplish this we remark that the
set A, which is a neighborhood of the origin, is invariant under the action of the dilation group, and then
is all of ~ . We will denote by V the (non-homogeneous) Lyapunov function associated with__f, given by
T h e o r e m 1 (Kurzweil). Afterwards we construct, from V (and another function a) a function V satisfying
all the properties (i)-(iv) of T h e o r e m 2. The expression of P is given in Proposition 2.
In Section 3 we study two problems of robustness for homogeneous stable systems, namely: is the
asymptotic stability of the origin preserved after adding an integrator or some perturbing term? Two
results, recently obtained by Coron and Praly (see [1, Proposition 3]) and H e r m e s (see [6, Theorem 1])
are easily proved by means of T h e o r e m 2, and slightly generalized.

2. Proof of Theorem 2

We first prove the following result.

Proposition 1. Under the assumptions of Theorem 2, the region of asymptotic stability is all of ~ , and so
the trivial solution of (2) & strongly stable in ~ .

Proof. Let A be the region of asymptotic stability, A is a subset of G = E". Let x 0 be in En, and let x(t)
be a solution of (2), defined on [0,T), where 0 < T < + ~, such that x(0) = x 0. Let e be a positive real
number such that (sr'Xl(O),..., er"xn(O)) CA. (Such an e exists since A is an open set which contains 0.)
For i = 1. . . . . n and t ~ [T/e ~) we set ~i(t) = Erixi(E~"t). Then Y(t) := (Yl(t) . . . . . £n(t)) is a solution of
(2). Indeed, we have

x i ( t ) = Er,+~fCi(Ert)

= e~'+~fi(xl(e~t),..., x , ( e ' t ) ) (since x is a solution of (2))

=fi(er'xl(s~t) . . . . . e r " X n ( e r t ) ) (since f is homogeneous)

= f i ( £ , ( t ) .... , £ , ( t ) ) .
Moreover Y ( 0 ) c A , and so there exists a solution ~ of (2), defined on [0, +oo), which extends ~f, and
which tends to 0 as t tends to infinity. It follows that the function y defined by yi(t) = (1/e)rif~i((1/e)~t)
for t > 0, is a solution of (2) such that
y(t)=x(t) for0<t<T and y(t)~O ast--* +~.
Thus, x 0 c A , and the proof of A = ~ is complete. Since we know the zero solution of J = f ( x ) , x c A ,
is strongly stable in A (see [9, proof of T h e o r e m 12]), we can apply T h e o r e m 1, thereby obtaining a
Lyapunov function V associated with f which is of class C ~ on ~n and which tends to infinity with IIx [[.
Now, a candidate for the Lyapunov function V of T h e o r e m 2 is proposed by the following proposition.
 L. Rosier / Homogeneous Lyapunov function 471

Proposition 2. Let f and p be as in Theorem 2. Let a ~ C ~ (~, ~) be such that

and a' > O on R.

on [2, oo),

Let k be a positive integer. Then the function

+~ 1
V ( x ) :=

{fo t--~'~'-f(aoV)(trax 1. . . . . tr"Xn) dt

is well de_fined, of class C ~ on En \{0}, and satisfies

ifxE~n\{o},

ifx = O,

(a) V V ( x ) ' f ( x ) < O,

( b ) g ( s r l x l , . . . , s r n x n ) =Sk~'(Xl . . . . . Xn) ,
for all (x 1. . . . . x , ) ~ IW'\{O} and s > O.
Moreover, if k > p" max{r i [1 <_i <_n}, then ~" is C ° at O.

Proof. Since V ( x ) tends to infinity with Hx II and vanishes at 0, the function P is well defined. Moreover,
we may find two numbers l > 0 and L > 0 such that

v(trlxl . . . . . trnXn) < 1 for all Ilxll ~ [1, 2], t < l ,

v ( t r l x l , . . . , t r " X n ) >__2 for all l[xll ~ [a, 2], t >__Z.

Then, for all x ~ E" such that IIx II (½, 2),

L 1 1
k ' ( x ) = f/ -t - Y ~ ( a ° V ) ( t r ' x l . . . . . tr"Xn) dt +
kL-----~ .
Clearly, V is C ~ on {x 111x H~ (½, 2)}, and we may write, on this set

0~" = fL t r' , 13V

-~x~(x) Jt --~--~a
t (v(trlxl'''''tr°x"))~-~x~(tr'x~ .... ,tr°x.) tit. (6)

Hence, by (1),

~-,fi-~x.. = f tz+k+la'(V(trlxl,'",t rnXn)) fi ( t r l x l , "', tr"X,) dt,

i=1 Xi i
and since a ' ( s ) > 0 for some s ~ (1, 2), V V ( x ) . f ( x ) < 0 holds for ~1 < I[x H< 2.
By an obvious change of variable of integration, we obtain at once (b). Hence, P is C = on ~" \{0} and
(a) is true. (See (5).)
Now, let k > p . max{rill < i < n}. For a multiindex a = ( a 1. . . . . a , ) let ~" denote 0~+ ' +~,/Ox~ • • •
0x~',. Then, by a straightforward induction argument using (4), we get, for ~"> 0 and y ~ S"-~,

Oa~'( srlyl ..... sr"yn) = sk-a'r Oa~'( Yl . . . . . Y,)" (7)

Hence, by Lemma 1, for 0 <a~ + "-. + a , < p , 0 ~ V ( x ) ~ 0 as Ilxll-" o. It follows that P is C v at the
origin of ~". This completes the proof of Proposition 2, as well as Theorem 2.

3. Robustness of stable homogeneous systems

The theorem we just proved will be used now to give simpler proofs of two results of control theory.
By the way we shall weaken somewhat their hypotheses: the smoothness of f in [1, Proposition 3] and
472 L. Rosier / Homogeneous Lyapunot, function

the uniqueness of solutions in [6, Theorem 1] will no longer be required. The first result claims that a by
means of homogeneous feedback stabilizable homogeneous system gives rise, after adding an integrator,
to a stabilizable system.

Proposition 3. L e t f = ( f i ) i - i,,, : ~ " × ~ ~ ~" be a C ~ m a p such that

V i = I . . . . . n, V x = ( x i ) i _ l , n ~ n, V t > 0 , V u ~
fi( tr'xl . . . . . tr"Xn, tr"+lU)=lT+r'fi( x 1. . . . . Xn, U)
f o r some r i > O, 1 <_i < n + 1, and some r ~ ( - m i n i { r i } , ~). A s s u m e that the system fc = f ( x , u) is locally
asymptotically stabilizable with a continuous feedback law u : ~" ~ ~ such that

u(trlxI . . . . . tr"Xn) =tr"*lU(Xl . . . . . Xn) ~X=(Xi)i=l,nE~ n, Vt>_O.

Under these conditions, the system fc = f ( x , y), y = z~ is globally asymptotically stabilizable with a continuous
feedback law v : ~ x ~ --* ~ such that

V X = ( X i ) i l,nE~ n, ~ y ~ , Vt>__O l.r(t~Xl . . . . . tr,,x,,, t~,,+,y)=t:+ .... ~t~(X I . . . . . X,,, y).

Proposition 3 is proved in [1] under the extra assumption that f is of class C ~.

The following lemma, which provides a smooth homogeneous stabilizing control law ~ and a
homogeneous Lyapunov function associated with the system J = f ( x , ~ ( x ) ) is a key technical step in the
proof of Proposition 3. We shall only simplify the proof of this lemma, the proof of Proposition 3 being
the same as in [1]. Note that in this case no smoothness assumption on f is needed.

Lemma 2 (Lemma 3 in [1]). Under the assumptions o f Proposition 3, there exist

(i) a stabilizing control law ~ : ~" --* R in c l ( ~ " \ { 0 } ) A C°(~ ") which satisfies
~l(trlxl . . . . . tr"xn) =tr"+lU(Xl . . . . . Xn) V X = ( X i ) i = l , n ~ff~ n, Vt>_0;
(ii) a C l function P :~" ~ ~ which # positit;e definite (see [4, Def. 24.3]), radially unbounded (see [4,
Def. 24.5]) and satisfies

~'(trlxl . . . . . t r , , x , , ) = t k V ( x l , . . , ,
x,,) Vx=(xi)i=l.,,e~", V t >_ 0
where k is a real number satisfying: k > r i, Vi ~ {1. . . . . n};
such that

VP(x)'f(x,~(x)) <0 Vx~"\{0}. (8)

Proof. Using Theorem 2, we know there exists, for the system J = f ( x , u(x)), x ~ ~", a Lyapunov
function P of class C ~, which is homogeneous (with the same r~, for 1 < i < n, as for f ) . The restriction of
u to S" i is easily approximated by a smooth function ~ satisfying V P ( x ) . f ( x , ~ ( x ) ) < 0 for all
x ~ S " - ~. Then it is sufficient to extend ~ to ~" as a homogeneous function, i.e. to set

fi(x) = [(~o(X))~"+'K(t#,(x) ..... O,(x)) if x * 0 ,

[0 if x = 0 ,
for getting (i) and (ii). ((8) follows from (5).)

The second result, which claims that a stable homogeneous system remains stable after adding higher
order perturbing terms, comes from [6], and is proved by the author of this article under the extra
assumption that the solution of J = f ( x ) , x(0) = x 0 is unique on [0, + ~ ) for any x 0 ~ ~".

Theorem 3 (Theorem 1 in [6]). Let f , 7, r 1. . . . . r. be as in Theorem 2 and g be a continuous vector valued

such that, for all i ~ {1. . . . . n}, gi(t~JXl . . . . , t r " x . ) / t ~+r' ~ 0 uniformly on S " - l as
function defined on ~
 L. Rosier / Homogeneous Lyapunov function 473

t ~ O. Then if the zero solution o f k = f ( x ) is locally asymptotically stable, the same is true for the zero
solution of So= f ( x ) + g( x ).

Proof. We denote, as Hermes, 67x = ( t r x x l . . . . , trnXn ) for t > 0 and x 4: 0. Let V be a Lyapunov function
for the system J = f ( x ) , which is of class C t and homogeneous (see T h e o r e m 2). Thus, for some k ~ R
we have Vt > 0, Vx 4:0 F'(6tx) = tkF'(x). So, for x ~ S n-1 and t ~ (0, 1) we get (see (5))

f ( ~tx) " Vk'( ~tx ) = te+~f( x) • V~'( x).

Let A be the negative n u m b e r maXx~s,-lf(x). VV(x). Since g(~Tx)" VF'(67x)= o(t k+~) uniformly on
S n- ~ as t ~ 0, there exists t o in (0, 1) such that, for 0 < t < t o and x ~ S n- 1, we have

[g( Stx) " Vk'( 8;x) l <_ ½tk+~lAI.

Therefore, for 0 < t < t o and x ~ S ~-1, we have ( f ( 6 t x ) + g ( 6 t x ) ) . V V ( ~ t X ) _< 7t
- r
1 k+'r A < 0. Now the set
{67x, 0 < t < to, x ~ S "-t} u {0} is a neighborhood of the origin in ~", since by L e m m a 1, 2) there exists
r 0 > 0 such that 0 < IIx II < r0 implies 0 < So(X) < t 0. Thus the function P is, in a neighborhood of 0, a
Lyapunov function for f + g . It follows that the trivial solution of the system ]c = ( f + g ) ( x ) is locally
asymptotically stable (see [5, Th. 8.2, Ch. III]).

Acknowledgements

We would like to thank professor Jean Michel Coron for having pointed out this question to us and
for fruitful discussions, professor W.P. Dayawansa for having substantially simplified our proof of the
main result (the proof given here is based on his suggestions), and professor Fred Weissler for useful
remarks.

References

[1] J.M. Coron and U Praly, Adding an integrator for the stabilization problem, Systems Control Lett. 17 (1991) 89-104.
[2] W.P. Dayawansa and C.F. Martin, Asymptotic stabilization of two dimensional real analytic systems, Systems Control Lett. 12
(1989) 205-212.
[3] W.P. Dayawansa, C.F. Martin and G. Knowles, Asymptotic stabilization of a class of smooth two-dimensional systems, SIAMJ.
Control Optim. 28 (1990) 1321-1349.
[4] W. Hahn, Stability of Motion (Springer-Verlag, Berlin, 1967).
[5] P. Hartman, Ordinary Differential Equations, Second Edition (Birkhauser, Boston, MA, 1982).
[6] H. Hermes, Homogeneous coordinates and continuous asymptotically stabilizing feedback controls, in: S. Elaydi, Ed.,
Differential Equations Stability and Controls, Lecture Notes in Applied Math. Vol. 109 (Marcel Dekker, New York, 1991)
249-260.
[7] M. Kawski, Stabilization and nilpotent approximations, Proc. 27th IEEE Conference on Decision & Control H (1988),
1244-1248.
[8] M. Kawski, Stabilization of nonlinear systems in the plane, Systems Control Lett. 12 (1989) 169-175.
[9] J. Kurzweil, On the inversion of Lyapunov's second theorem on stability of motion, Ann. Math. Soc. TransL Ser. 2 24 (1956)
19-77.
[10] V.I. Zubov, Methods of A.M. Lyapunov and their Applications (Noordhoff, Leiden, 1964).