A note on exponential stability of partially slowly time-varying nonlinear systems Joan Peuteman Dirk Aeyels SYSTeMS Universiteit Gent Technologiepark-Zwijnaarde, 9 9052 GENT (Zwijnaarde) BELGIUM e-mail: Joan.Peuteman@rug.ac.be e-mail: Dirk.Aeyels@rug.ac.be Preliminary version Abstract 1 Introduction The classical Liapunov approach to exponential stability of an equilibrium point of a dynamical system x_ = f (x; t) requires the existence of a positive denite decrescent Liapunov function V (x; t) with vminxT x V (x; t) vmax xT x (vmin; vmax 2 R+0 ) such that for some 2 R+0 : V_ (x; t) ? kxk2. In general it may be hard to construct a Liapunov function which satises this condition. A weaker condition, still guaranteeing exponential stability of the equilibrium point has been proved in [1]. The main theorem of [1] proves that the origin of a system is exponentially stable under the condition that for a positive denite decrescent V (x; t) with vmin xT x V (x; t) vmax xT x (vmin ; vmax 2 R+0 ), 9T > 0 and a sequence of times tk such that V (x(tk+1 ; tk ; x(tk )); tk+1 ) ? V (x(tk ); tk ) ? kx(tk )k2 with 2 R+0 and tk+1 ? tk T 8k and tk ! +1 as k ! +1 and tk ! ?1 as k ! ?1. This theorem allows to establish, by means of a Liapunov approach, that the system x_ = f (x; t), with suciently large, is exponentially stable when its Consider a system x_ = f (x; t; t ) with a time- varying vectoreld which contains a regular and a slow time scale ( large). Assume there exists ( ) such that kx (t; t0 ; x0 )k K ( )kx0 ke( )(t?t0) where x (t; t0 ; x0 ) is the solution of the system x_ = f (x; t; ) with initial state x0 at t0 . We show that for suciently large, x_ = f (x; t; t ) is exponentially stable when the average of ( ) is negative. This result can be used to extend the circle criterion i.e. to obtain a sucient condition for exponential stability of a feedback interconnection of a slowly time-varying linear system and a sector nonlinearity. An example is included which shows that the technique can be used to obtain an exponential stability result for a pendulum with a nonlinear partially slowly time-varying friction attaining positive and negative values. Keywords: Nonlinear dynamics, exponential stability, circle criterion, pendulum. 1 averaged system is exponentially stable. This averaging result allows to study exponential stability of fast time-varying nonlinear systems. In [5], the averaging approach is generalized. This generalization, which we call `partial averaging', allows to study exponential stability of a system x_ = f (x; t; t) with suciently large (see also [2], pp. 194-195). In the present paper, the main theorem of [1] is used to investigate exponential stability of slowly timevarying nonlinear systems. The investigation of exponential stability of slowly time-varying systems is already discussed in [3, 6, 7]. Rugh [6] and Solo [7] only deal with linear systems. Rugh [6] considers only systems whose eigenvalues are pointwise in the open left half-plane. Solo [7] generalizes this result and considers linear systems with eigenvalues which are allowed to `wander' into the right half-plane as long as `on average' they are strictly in the left halfplane. Khalil ([3], pp. 239-250) discusses exponential stability of slowly time-varying nonlinear systems with the restriction that the `frozen' systems, i.e. the time-invariant system where t is treated as a `frozen' parameter, are all exponentially stable. An extension of the results of Solo [7] and Khalil ([3], pp. 239-250) is the study of a nonlinear slowly timevarying system x_ = f (x; t ) with suciently large where the `frozen' systems are exponentially stable or exponentially unstable. The main theorem of the present paper does not only allow to study exponential stability of slowly timevarying systems x_ = f (x; t ). The main theorem also studies exponential stability of partially slowly timevarying systems x_ = f (x; t; t ). Let ( ) and K ( ) be chosen such that 8t0 8x0 : kx (t; t0 ; x0 )k K ( )kx0 ke( )(t?t0) where x (t; t0 ; x0 ) is the solution of the system x_ = f (x; t; ) with initial state x0 at t0 . We prove that for suciently large x_ = f (x; t; t ) is exponentially stable if the average of ( ) is negative. It is important that ( ) is allowed to take positive and negative values. This result can be used to extend the circle criterion i.e. to obtain a sucient condition for exponential stability of a feedback interconnection of a slowly time-varying linear system and a sector nonlinearity. The classical circle criterion only gives a sucient condition for exponential stability of a feedback interconnection of a time-invariant linear system and a sector nonlinearity. An example is included which demonstrates that the technique can be used to obtain an exponential stability result for a pendulum with a nonlinear partially slowly time-varying friction attaining positive and negative values. 2 Exponential stability We recall the main theorem for exponential stability of [1] since the proof of Theorem 2 relies on it. Consider x_ (t) = f (x; t) (1) with f : W R ! Rn , W open, W Rn , f (x; t) is measurable for each xed x, 0 2 W and f (0; t) = 0 8 t 2 R. Furthermore we assume that conditions are imposed on (1) such that existence and uniqueness of its solutions is secured. These conditions are imposed on all the dierential equations mentioned in the present paper and of these conditions, we single out the local Lipschitz condition: f is locally Lipschitz on W , i.e. for 8x 2 W , 9 a neighborhood N (x) W , such that the restriction f jN (x) is Lipschitz with Lipschitz function lx(t). We assume that lx(t) is bounded 8 t 2 R. Theorem 1. Consider a V : U R ! R, with U W an open neighborhood of 0. We assume that the following conditions are satised: Condition 1: V (x; t) is positive denite and decrescent (V (0; t) = 0 8t) and there exist positive numbers vmin and vmax such that 8x 2 U vmin xT x V (x; t) vmax xT x. Condition 2: There exists an increasing sequence of times tk (k 2 Z) with tk ! 1 as k ! 1 and tk ! ?1 as k ! ?1, 9 nite T > 0 : tk+1 ? tk T (8k 2 Z), 9 > 0 and an open set U 0 U which contains the origin 2 ( ) > 0 and 8r0 : 0 < r0 Kr , 8t0, 8t 2 [t0 ; t0 + 1 ln( Krr 0 )]: 8x0 2 B r0 (0): such that the Liapunov function V (x(t); t) has the property that 8 k 2 Z and 8 x(tk ) 2 U 0 nf0g V (x(tk+1 ; tk ; x(tk ));tk+1 ) ? V (x(tk ); tk ) (2) ? kx(tk )k2 < 0 where x(tk+1 ; tk ; x(tk )) is the solution of (1) at tk+1 with initial condition x(tk ) at tk . then the equilibrium point x = 0 of (1) is exponen- kx (t; t0 ; x0 )k K ( )kx0 ke( )(t?t0) (4) Here x (t; t0 ; x0 ) is the solution at t of the system x_ = f (x; t; ) with initial condition x0 at t0 . { K ( ) is bounded i.e. 9Kmax 1 such that 8 : K ( ) Kmax. Condition 2: 9 Tm > 0, 9 TM > 0 and a sequence of times k with the property that 8k 2 Z : Tm k+1 ? k TM . The sequence k has the additional properties that { For some > 0 and 8k tially stable. Proof. For a proof, the reader is referred to [1]. Remark 1. When x_ = f (x; t) is globally Lipschitz and the conditions of Theorem 1 hold globally then the equilibrium point x = 0 is globally exponentially stable. 3 Partially slowly time-varying systems Z k+1 k ( )d ?(k+1 ? k ) (5) { f (x; t; ), for each xed x 2 B r (0) and each xed t, has nd (k) ? 1 discontinuities k; 1 ; :::; k;n d (k)?1 in ]k ; k+1 [ and 9nd 2 N such that 8k: nd(k) nd . In the present section, we will investigate exponential stability of partially slowly time-varying nonlinear systems i.e. systems of the form x_ = f (x; t; t ) with suciently large. Theorem 2. Consider the system x_ = f (x; t; t ) which satises the assumptions made in Section 2: f is locally Lipschitz on W , i.e. for 8x 2 W , 9 a neighborhood N (x) W , such that the restriction f jN (x) is Lipschitz with Lipschitz function lx (t; t ) which is assumed to be bounded on R with a bound independent of . If 9r > 0 (B r (0) W ) such that f (x; t; ) is allowed to have discontinuities at k . { We require that 80 9r (0) such that j1 ? 2 j r (0 ) with the additional condition that 9k 2 Z and 9i 2 f0; :::; nd(k) ? 1g such ; [ and 2 2 [ ; [ that 1 2 [k;i k;i+1 k;i k;i+1 implies that 8x 2 B r (0), 8t: jf (x; t; 1 ) ? f (x; t; 2 )j 0 kxk. { We require that 80 9(0) such that j1 ? 2 j (0 ) with the additional condition Condition 1: that 9k 2 Z and 9i 2 f0; :::; nd(k) ? 1g such ; [ and 2 2 [ ; [ that 1 2 [k;i k;i+1 k;i k;i+1 { 8 , 9K ( ) 1 and 9( ) such that either implies that j(1 ) ? (2 )j 0 . ( ) 0 and 8t0 , 8t t0 : 8x0 2 then there exists 1 > 0 such that 8 1 the equiB r0 (0): librium point x = 0 of x_ = f (x; t; t ) is exponentially ( )( t ? t ) 0 kx (t; t0 ; x0 )k K ( )kx0 ke stable. (3) Proof. The proof relies on Theorem 1 where with r0 = Kr . V (x; t) = kxk2. For suciently large, Condition 3 2 of Theorem 1 is satised with tk = k . The entire proof of the theorem will be submitted for publication elsewhere. Remark 2. Partially slowly time-varying systems x_ = f (x; t; t ) with suciently large are a generalization of slowly time varying systems x_ = f (x; t ). Remark 3. When x_ = f (x; t; t ) is globally Lipschitz and when the conditions of Theorem 2 hold globally than the system is globally exponentially stable. When there exists a positive denite symmetric P , a n-dimensional L and a constant 0 2 R such that PA + AT P = ?LT L + 0 P p PB = C T k ? 2LT then V_ (x; t) 0 xT Px (14) p p T ? (Lx ? 2k(x; t)Cx) (Lx ? 2k(x; t)Cx) 0 xT Px (15) 4 Systems of the Lure type q max (P ) that This implies that with K = min (P ) 0 kx(t; t0 ; x0 )k K kx0 ke(t?t0 ) 2 where x(t; t0 ; x0 ) denotes the solution of (9) at t with initial condition x0 at t0 . This result is a generalization of the classical circle criterion in the sense that 0 is allowed to be positive or negative whereas the classical circle criterion is only interested in negative 0 . By the Kalman-Yakubovich-Popov lemma ([3], pp. 468-469), (12) and 0 (13) are satised if and only if 1 + kC (sI ? A + 2 I )?1 B is positive real. Positive 0 realness of 1 + kC (sI ? A + 2 I )?1 B is by denition equivalent with 4.1 Lure systems with time-invariant linear part Consider a SISO system which is built up as a feedback interconnection of a linear time-invariant system, with (A; B ) controllable and (A; C ) observable and a nonlinear time-varying element. When the system is unforced, the behavior is represented by x_ (t) = Ax(t) + Bu(t) (6) y(t) = Cx(t) (7) u(t) = ?k(x(t); t)y(t) (8) where for some k > 0: 0 k(x; t) k 8x 2 Rn 8t and u; y 2 R. Taking (6), (7) and (8) into consider- 0 Re(1 + kC ((j! + 2 )I ? A)?1 B ) 0 8! 2 R ation one obtains that x_ (t) = Ax(t) ? Bk(x(t); t)Cx(t) (12) (13) (16) (9) Therefore, the ow of the system (9) satises 0 ( t ? t ) 0 kx(t; t0 ; x0 )k K kx0 ke 0 2 = K kx0 ke(t?t0) when (16) is satised ( = 2 ). Consider V (x; t) = V (x) = xT Px such that along the ow of (9) V_ (x; t) = xT (PA + AT P )x ? 2xT PBk(x; t)Cx (10) Notice that k(x; t)y(k(x; t)y ? ky) 0 such that ?2k(x; t)Cx(k(x; t)Cx?k Cx) 0. One obtains that V_ (x; t) is less or equal than xT (PA + AT P )x + 2xT (C T k ? PB )k(x; t)Cx ? 2xT C T k2 (x; t)Cx (11) 4.2 Lure system with time-varying linear part Consider a SISO system which is built up as a feedback interconnection of a linear time-varying system, with A(t), B (t) and C (t) time-variant, and a nonlinear time-varying element. We assume that A(t), B (t) and C (t) are bounded, periodic and piecewise continuous with each period T a nite number of discontinuities. Here (A(t); B (t)) is controllable 8t 4 and (A(t); C (t)) is observable 8t. When the system is unforced, the behavior is represented by the discontinuities (if any) of A(t), B (t) and C (t) then Condition 2 of Theorem 2 is satised with a sequence k := kT (k 2 Z). Theorem 2 implies that 91 > 0 such that 8 1 , the partially slowly time-varying system x_ (t) = A(t)x(t) + B (t)u(t) (17) y(t) = C (t)x(t) (18) u(t) = ?k(x(t); t)y(t) (19) x_ (t) = A( t )x(t) ? B ( t )k(x(t); t)C ( t )x(t) (26) where for some k > 0: 0 k(x; t) k 8x 2 Rn is exponentially stable. 8t and u; y 2 R. Taking (17), (18) and (19) into Remark 4. In Section 5, we show by means of an consideration one obtains that example how piecewise continuity of ( ) and the x_ (t) = A(t)x(t) ? B (t)k(x(t); t)C (t)x(t) (20) existence of a Kmax may be established. Consider now for every the system x_ (t) = A( )x(t) ? B ( )k(x(t); t)C ( )x(t) with 0 ( ) 2 R such that 0 1 + kC ( )((s + ( ) )I ? A( ))?1 B ( ) 2 5 Pendulum with nonlinear partially slowly time-varying friction which attains positive and negative values (21) (22) Consider the pendulum described by is positive real. x_ (t) = M (t; t ; x)x (27) Recall that by the Kalman-Yakubovich-Popov lemma, positive realness of (22) is equivalent with with the existence of an 0 ( ) 2 R, a symmetric positive t 0 1 denite matrix P ( ) and a n-dimensional L( ) such (28) M (t; ; x) = ?1 ?f (t; t ; x) that t t P ( )A( ) + AT ( )P ( ) = ?LT ( )L( ) + 0 ( )P ( ) and f (t; ; x) = 1+b( 2)k(x; t). Here, for some k > 0, k(x; t) k 8x 2 R 8t. (23) 0Notice that the system (27) can be written in the p P ( )B ( ) = C T ( )k ? 2LT ( ) form (26) with (24) A = A( t ) = ?01 ?11 (29) By Section 4.1, this positive realness implies that the ow of0 (21) satises kx (t; t0 ; x0 )k and B T ( t ) = [0 b( t )] and C = C ( t ) = [0 1]. K ( )kx0 ke(t?t0) 2( ) with We consider the case where k = 1 and we assume that b( ) is a continuous function with period T and s that 8 : b( ) 2 [?1:5; ?0:5]. Notice that 8 > 0, ( P ( )) max (25) 8t: (A; B ( t )) is controllable and (A; C ( t )) is obK ( ) = (P ( )) min servable. When there exists a Kmax 1 such that 8 : Kmax In order to apply the results of Section 5.2, we investigate the positive realness of (22) i.e. K ( ) then Condition 1 Rof Theorem 2R is satised. 0 When 9 > 0 such that 0T ( )d = 0T 2( ) d ? 0 ( ) )I ? 0 1 ]?1 0 0 0 1 1 + [( s + b( ) ?1 ?1 2 ?T and ( ) = 2( ) is bounded and piecewise con(30) tinuous with discontinuities which synchronize with 5 6 Acknowledgement Recall that (30) is positive real when 8! 2 R its real part is nonnegative. After some calculations, one obtains that this is satised when This paper presents research results of the Belgian Programme on Interuniversity Poles of Attraction initiated by the Belgian State, Prime Minister's Ofce for Science, Technology and Culture. The scientic responsability rests with its authors. p 0 ( ) ?2 (b( ) + 2)(b( ) ? 3) + 2jb( )j 3 ? b( ) (b( ) ? 2)(b( ) + 6) (31) Since b( ) 2 [?1:5; ?0:5], one obtains by taking 0 ( ) equal to the right hand side of inequality (31) that 0 ( ) 2 [?0:246; +0:262] and attains positive and 2 negative values. This implies that for 0 ( ), (30) is positive real such that x_ (t) = M (t; ; x)x has solutions satisfying 0 kx (t; t0 ; x0 )k K ( )kx0 ke(t?t0) 2( ) References [1] D. Aeyels and J. Peuteman, On exponential stability of non-linear time-varying dierential equations, submitted for publication. [2] J.K. Hale, Ordinary dierential equations, Robert E. Krieger Publishing Company, (1980), pp. 190-194. [3] H.K. Khalil, Nonlinear systems, Printice Hall, (1996). [4] P. Lancaster and L. Rodman, Algebraic Riccati Equations, Oxford Science Publications, (1995). [5] J. Peuteman and D. Aeyels, Exponential stability of non-linear time-variant dierential equations and partial averaging, in preparation. [6] W.J. Rugh, Linear system theory, Prentice Hall New Jersey, (1993). [7] V. Solo, On the stability of Slowly Time-Varying Linear Systems, Mathematics of Control, Signals, and Systems, (1994) pp. 331-350. (32) We assumed that b( ) is periodic and continuous which implies by (31) (and 0our choice of 0 ( )) continuity and periodicity of 2( ) = ( ). By the Kalman-Yakubovich-Popov lemma, positive realness of (30) is equivalent with (23) and (24) when A( ), B ( ) and C ( ) take the values indicated in the present section. By eliminating L( ) in (23) and (24), one obtains a Riccati equation with solution P ( ). Relying on the continuity properties of A( ), B ( ), C ( ) and 0 ( ), it is possible to establish continuity of P ( ) ([4], pp. 260-266) since (?A( ) + 1 T 2 B ( )C ( )+ ( )I; B ( )B ( )) is c-stabilizable 8 . This implies the existence of a max > 0 and a min > 0 such that 8 : max max (qP ( )) and min min (P ( )). Therefore, Kmax = max min 1 such that 8 : K ( ) Kmax. One obtains by Theorem 2 that 91 such that 8 1 , (27) is exponenially stable when T Z 0 0 ( )d < 0 (33) or equivalently Z 0 T (b( ) + 2)(b( ) ? 3) + 2jb( )jp3 ? b( ) (b( ) ? 2)(b( ) + 6) d > 0 (34) 6