IEEE TR.4KShCTIONS OX AVTOBUTIC CONTROL, VOL. AC-19, XO. 3, JUNE D. G. Luenberger, ‘fXat,henlaticalprogramming and control theory: Trends of interplay,”Presented a t t,he 7thIntern. Symp. Math. Programming, The Hague, Ketherlands,Sept. 1970. __ , “Introduction t,o Linear and Nonlinear Programming.” Reading, M a s . : Addison-Wesley, 1973. -, “Convergence rate of a penalty function scheme,” J . Optimiz. Theory AWL, vol. 7, no. 1, pp. 39-51, 1971. E. Polak, Compututional Methods in Optimization. Kew York: Academic, 1971;‘ B. T. Pofyak,The c0njugat.e gradient met.hod in extrema1 problem, USSR Comput. N a f h . Math. Phys., vol. 9, no. 4, 1969 (English Translation). W . I. Zangwill, Xmlinear Programming: A U n i f i d Approach. Englewood Cliffs, X. J.: Prentice-Hall, 1969. I). K. Faddeev and V. N. Faddeeva, Comptutional Methods of Linear A7gebra. San Francisco, Calif.: Freeman, 1963. G. E. Forsvt,he. “On the asvrmtotic directions of the sdimensional optiG1um gradient mgthbd,” A’um.er. Nath.., vol. 11, pp. 57-76. 1Qm. G. P. McCornlick, “8nt.i-zig-zagging by bending,” Management Sci., V O ~ . 15, pp. 315-319, 1969. A. A. Goldstein, “Convex programming in Hilbert space,” Bull. Amer. Math. Soc., vol. 70, no. 5, pp. 709-710, 1964. E. S. Le7,itin and B. T. Polyak, “Const.rained minimization methods, Zh. Vychisl. X a t . M a t . Fiz., vol. 6, no. 5 , pp. 787823, 1966. --7 -II- 1974 217 Dimitri P. Bertsekas mas born in Athens, Greece, in 1942. He received the Mechanical and Electrical Engineering Diploma from the National TechnicalUniversity of Athens, Athens, Greece, in 1965, the N.S.E.E. degree ’, , from George WashingtonUniversity, Washington, D.C., in 1969, and the Ph.D. degree in system science from theMassachusetts Institute of Technology, Cambridge, in 1971. From 1966 to 196’7, he performed research atthe National Technical Vniversity of Athens, and from 1967 to 1969, he was with t.he U. S. Army Research Laboratories, Fort. Belvoir, Va. In the summer of 1971, he worked at Systems Control, Inc., Palo Alto, Calif. From September 1971 to December 1973, he was in the faculty of the Engineering-Economic SystemsDepart.ment of St.anford University %.here hetaught courses inoptimization by vectorspacemethods,nonlinear programming, optinlizat,ion under uncertainty,and convex analysis. In January 1974, he joined t,he faculty of t.he Electrical Engineering Department, Universit.y of Illinois, Urbana, where hecurrently holds the rank of Assist.ant.Professor. His research int.erests include the areas of estimation and control of uncertain syst.ems, dynamic programming, optimixat.ion theory, and nonlinear programming algorithms. ,I L-Stability of Nonstationary Feedback Systems: Frequency-Domain Criteria MALUR K. SUKDARESHAN Abstract-A frequency-domain positivity condition is derived for develop & linear time-varying operators in L andisusedto stability criteria for linear and nonlinear feedback systems. These criteria permit the useof a very general classof operators inL with nonstationary kernels, a s multipliers. More specific resultsare a timeobtained by using a first-order differential operator with varying coefficient a s multiplier. Finally, by employing periodic multipliers, improved stability criteria are derived for the nonlinear damped Mathieu equation with a forcing function. I. IKTHODUCTIOS T HE employment of an operator theoretic framework stabi1it.y studies hasresulted in the development of the posit,ivit.y theorem (due t o Zames [l])as a versatile tool for the input-ou6put sta.bility analysisof feedback syst.ems. Many useful criteria for t,he L-stability of t.ime-invsriant Imdinear syst,ems [2] and syst.ems lyith isolatedtimeva.rying gains[3],[4],have been developed in the last Manwcript received October 30, 1972; revised March 2.3, 1973. Paper recommended by J. C. ‘A7illem, Past Chairman, and R. A. Skoog, Chairman of t h e I E E E S C SStability, Konlinear, and Distributed Systems Conunittee. The authors are with the Department of Electrical Engineering, Indian Institute of Science, Bangalore, India. AND &I. A. L. THATHACHAR few years, following t,his approach. However, the derivat,ion of sindar resu1t.s in the more general situa,tion of systemscontaininglineartime-varyingoperators which do not, admit, a separation of time-va.ria.tions, has not received much at,tention. Recalling that t.he application of t,he positivity t,heorem requires t.he open-loop to be fact.ored into a composition of t*wopositive operators, the unpopularit,; of this problem ma>- be atkribut.ed t.0 t,he difficult,; of obt.aining positivity conditions for a.rbitrary t,ime-varying operators.Recently, by using aninternal (state-space)description of t,heoperat,or, &’illems [51, for Estrada and Desoer [ 6 ] ,and the authors[7] have obtained posit,ivit,y condit,ions in thc time-domain (and from t,hese L?-stability criteria [’i] for feedback system5 containing a time-varyinglinear part). Theseposit>ivityconditions require the solut,ion of certain associat.ed Riccat,i equations and are not easy t o check, except in a few simplc cases. This paperpresentssimplerfrequency-domain conditions for the positivity and L2-stability of t.ime-varying systems, derived with the imposition of certain additional c0nstraint.s of differentiabilit,g, null initial condit.ions, etc. The nwthod draws inspirat.ion from a recent paper due to BIodgett, and Young [SI, which gives an absolutc sta.bility 218 IEEE T W S A C T I O N S ON AUTOMATIC CONTROL, JUNE 1974 criterion for a zero-input feedback system with a. time- K , and a being positive const.ants and 11- IIB denoting the varying linear part anda Popov-type nonlinea.rit,y,using a Euclidean norm. mult.iplier (1 qs), y > 0. It should however be mentioned 4) H has zero initial condit,ions, i.e., ~ ~ ( =0 0.) t,hat, although [SI cont,ains ma.ng valuable ideas, the main Because of propert.- 3 above, it is simple to observe t,hat, stability t.heorem appears to be incorrect. A corrected H E O I P A implies H E O E with the kernel restricted by version of the st.abilitg criterion of [SI may be obtained as Ih(f,T)1 6 k ‘ esp(--a(f - T)), which in turn implies that a special case of the criteria derived in this paper, which H:L2 + L. permit. the use of more general time-varying multipliers. An operat.or H E O E is said t o have “finite ga.in” if + 11. PROBLEM FORMULATION Notations and D e j h i t i m s A cert.ain familiarity with the not,ions of L,-spaces will be assumed. Let R, Ri, and R“ denote, respect.ively, the real numbers, t,henonnega.tive real numbers, and .n-dimensional Euclidean space. The concept of the extended L2-space (L?,)is defined by, LPe = {x(.):x*(.) E Lp, T ‘+ T E R+) (2.1) where x*(.) is the t.runcation of x(.), x,(t) = z(t), ++ t E [O,T], and is zero otherwise. An operator H in L, (LBe)is defined a$ n single-valued mapping of L,(L?,) into it.self. H is said to be a, “causal” operat,or in L2 (Lze)if ( H z ( . ) ) , = (Hz,(.)), ++ x(.) E L2 (L.28)and ++ T E R+. Let. OB den0t.e t.he class of linear causal operators H in L2, withanexternal(input-output)) deecript-ion, i.e., if H E OB, t,hen there exists a map (t.he kernel of H ) h ( . 7 - ) : R+ X R+ R such that, - ?/B(t) = HuH(t) = 1- h(t,T)UH(T) dT v uH(’) the normsindicat,edbeing t.he Lz-norms. Xotethat H E O I P A => r(H) < An operator H E ee is said to be“positive(e)” [strongly posit,ive(e)] if t.he inequality, EL and V T E R+ holds Xvith e = O[E > 01. If in a,ddition H E Or, then y H ( . ) given by (2.3) may beusedin theplace of in (2.5). If H is a causal operator in Lf,then H positive(e) (=) ( u H ( . ) , H u H ( . )3) 0, ‘4 E L2. (2.6) System D e s c r i p f i m The system under consideration ha3 t.he configuration as in Fig. 1 and is described by the input,-output relations, e el(t) - (2.2) . Z~N(.) = ul(t) - 1c2(t) + e?(t) = u&) zc1(t) (2.7) where 2tH(.):R+ -+ R is t.he input to H , yH(.):R+ R rrl(t) = Gel(t) and m ( t ) = Fez(t) is t,he output of H and h ( t , ~ = ) 0 V T > t. Let OI den0t.e the class of linear causal operators H in with the following assumptions: Lze withaninternal (state-space) description, i.e., if ) L2e. H E Or, thenthere exists aquadruple { A H ( - ) , b B ( . ) , Assumptim 1: t c 1 ( . ) , u 2 ( . ) E b,and e l ( - ) , e B ( .E Asrumptiott. 2: G E oE n eI. c H ( . ) ,and d R ( . ) f and a positive integer 71, where A H ( . ) : Assumptiott. 3: F is a time-invariant. Popor-type nonR + +- R” X R”, b,(.):R+ -+ R”,c H ( . ) : R + + R” and linear mcmoryless operator in I,, defined b>-: d H ( .) :R R, such t.hatH is described by t h r dynamical equat,ions, + - + bx(t) CR’(t) XH(t) + *x(O = A x ( Q zH(0 Y&) = ~ . H ( O dH(t) UH(t) (2.3) Let us denot>c the class of operators F satisfying 5p. (2s)by where cH‘(t) d e n o h the t.ranspose of c H ( t ) , x H ( - ) : R + - R“ is the stateof H and u H ( . ) , yx( - ) are defined as earlier. The 31aitl Problem. Let. O I p A be t.he subset of aI consist,ing of those operators Find conditions on C a,nd F which ensure that the sysH which satisfy the following requirements: tem drscribed by(2.7) is L,-stablr, i.e., Z L ~ (-),?le(.) E L? => 1) The internal dcscriplion of H is in the phase-variable e l ( . ) , e d . ) E h . canonical form. 111. SOLCTIOS OF THE 1 \ I ~ r sPROBLEM 2) The elements of A,(t) and cH(t) are ?)-times differentiable n i t h respcct to f, 71 being t,he dimension of the In this section, it is proposed to provide a solution to state-vector of H . t,hr a,bove problem byapplying the po3itivitytheorem 3) H is exponentiallyasymptoticallystablewith zero [1],[2] after transforming the system [9] n-ith t,he introinput,, i.e., I’xH(t)I’g K llzH(to) ;IE esp(-a(t - to)) v t 3 to, duction of “multiplicrs” (as shown in Fig. 2 ) . Recall that < 219 SUNDIRESHAN AND THATHACHAR: NONSTATIONARY FEEDBACK SYSTEMS 5 a = i ~ * ( ‘ -2~ )0, i = l ++ w ER (3.4) , are satisfied. If, in addition t,o t.he above hypotheses, &(t) Y t E R+,then H is st,rongly positive(e). Fig. 1. The feedback system under consideration. 3 E >0 Proof: Po&ivity(e) of H: Since H E OIPA, xHi(t)= 0 at i = 1,2,-. *,nandlim xHi(t) = 0 ++ i = 1 , 2 , - . . , n l t = 0 ‘4 t- m Now, since H E O I P A =) H is a causal operator in L, it, is sufficient,to prove, in view of (2.6), t.hat 3 v 0, (ZLH(.),YH(.)) UH(.) E L?. (3.5) Sow, Left-hand side of (3.5) t,he application of the theorem requires a. factorization of t,he loop into a composition of two operators, one of which isstrongly positive(e) witjh finitegain and t.he other is positive(e). = = (UH(.),CIf’(.) ZH(.)) + M - ) ud-)),from (2.3) + ?&I(.)). (3.6) (%I(.)rdH(.) Second term on t.he right-ha.nd side (RHS) of (3.6) A F.requency-Do?~~.ailz Positivity Cmzdition for Time-Varying Linear Operators h criterion for the positivit.y(e) of linear time-varying operat.ors H E O I p A 15-ilI now beenunciat,ed. This is a sufficient. condit,ion and requires the establishment of the nonnegativit,y of a.n evendegree polynomial whose coefficients reault. from the minimizat,ion of certain combinat.ions of aHi(t) and c H j ( t )and their derivat.ives. Lenzma.: Let, { a H i }i, = 1,2,. . . , ? I . be an .n-tuple of const.ant.sassociated wit,han operator H E O I p a and defined by, (u.d.),cH’(.) ax(.) = lrn dt dH(U t )H Z ( t ) > 0, a ?AH(.) E Lz (3.7) since cl,(t) 3 0, Y t E R’. Further, since H E 01,,, subst,ituting for u H ( . ) from (2.3), we ha,ve, first tern1 on the R.HS of (3.6) (3.1) where, -aHi(t), i = 1,2; . .,?Aare the elements of the nt,h row of A H ( t ) ,c H j ( t ) , i = 1,2, . . .,n.are t.he element,s of cH(t) a.nd Pt,iare the coefficients defined by, S(j) + k S ( j - 1) 1 - l)! + kh.(j - 2 ) (k(x:--j 2j)!.j! h(k - 2 j ) (3.2) + f ( t >y i ( t ) y i + k ( t ) dt = (the sL1perscript.s d h i n the brackets in (3.1) den0t.e the order of the derivative with respect to time; terms with negative superscripts should be discarded). Then H is positiw(e) if 1) cl,(I) 3 0, Y t E R+ and 2) the frequency-domain inequality, (3.3) (3.8) We will next simplify t,he first and third terms on the RHS of ( 3 3 ) bg: repeated integration by parts. With this motive, let us now state a simple result. Proposition: If f ( t ) is a real-valued funct.ion different.iable k-times aud if yi(t), i = 1,2; . . arc a family of timefunctions sat.isfying 1) j i ( t ) = yi+](t), i = 1,2,. . . and 2 ) yi(tl) = gi(f.2) = 0, Y i = 1,2,. . . and some t l , t2 E R+,hhen j’r where aHl,(t.)~~i(~)]~~i(t).~H~(t) clt. J t? ti k pp,j(f(t))‘”-’i’yi+j2(f) dt, (3.9) j = O ‘This last assertion is a result. of the following arguments. If h i ( t , ~ are ) the kernels relating the components of the state-vector to the input,, i.e., x ~ i ( t = ) ~ ~ ( L , T ) U H ( T ) dr, ’ ~ ~ i ( f ) ! [hi(t,T)’ Il(H(r)i d s ICs; e - a ( t - r ) lu~(s)I d7 since HEBrpn; noting that t.he last term is a convolution oft.mo Icfunctions, a st.raight.forward application of t,he Riemann-Lehesgue theorem gives the intended result. < < st where Br ,jare t,he coefficients defined as in (3.3) (the superscriptwithin the bracketsin (3.9) denoting, as earlier, the order of the time-derivative). The proof of this resultissimplyestablishedby repeated integration by parts and is omitt.ed as it is readily available in [SI. i\Tow, simplifying t.he first, and third terms on the RHS of (3.5)from an a.pplicat.ionof the above result2, we have, RHS of (3.S) ing condit.ions are satisfied: A Few Remarks Rem.a.rk 1: I n compa.risonwith tjhe esisting results for the stability of similarsystems,itnlaybe observed that Theorem 1 pernut,s t>heuse of a. very gcncral class of timeaHI:(t)CHj(t)](L+j--2i)XHi?(1) dt varying mult,ipliers, much like the results of [ i ] However, . 2 aH+xHi2(t)df, aHibeing defined as in (3.1) it should be not.ed t,hat, while ['i] uses a minimal realizai = l tion t.0 givestabilit.y criteria. in the time-domain which require t.he solution of ccrt,ain matrix Riccati equat>ions, (3.10) t.hcpresentresults startwith a canonical realization sat,isfying additional differentiabilit.!- requirements, t.0 by an application of Parscval's t.heorem, where X H i ( j o )is >-ield simplerfrequcncy-domain conditions. The not,cthe Fourier-transform of .rHi(f). worthy feat.ureof the present results isthe ease in checking Xow, eincv r H j ( t= ) anci-l)(t), V i = l!?; . . $ u , XHi(ju). the stabi1it.y conditions, which has rcsulted from the im= j~S,,,-~,(j6~) nnd by induction, X H , ( j w ) = ( j ~ ) ~ - l position of certain addit.ionn1constraints on the system. X H i ( j u ) ,V i = 1.2,. . .,v. Remark 2: A stabi1it.y criterion for nonlinear system, Hence, on substitution, we have similar to the Theorem 1, may be formulat.ed by using a time-varyingmultiplier :If which isdrcomposible int.0 X =E Z with jZ'I < 1 (the norm used is the opcrat,or norm) andrestricting the nonlinea.rity t o be odd and 3 0, from (3.4). (3.11) monot.onically nondecreasing. The statement. of the rcsult will br omit,ted, i n view of t.hc popularity (see Zames xnd Thus, combining (3.7) and (3.11), (3.5) results. Stro,l!/po~jtil'ity(e)o f H : It is sinlplc to observe that! Fnlb [2] or Willems and Brockett [lo]) of such methods. H is stronglypositirc(e) if ( H - € E ) is positivc(e) for I t should. however. bc nwntioned that the conditions inlsome e > 0. E being the identit.! operator in I+. Further, posed by this criterion on the linear part of the system H E @Ip.-l => ( H - EA') E @ I P A with the. st:Lte-equation would be simplrr and nlorc explicit than the continuousunchanged, but the output cquation nlodificd into, time version of the results of [lo]. Remark 3: The difficult step in thr application of the B H ( . ) = cH'(-).cH(.) [(I,(.) - e ] u H ( . ) . H t ~ l c c !t h c . stability criterin of the present type lics in ensuring that positivity(e) of ( H - € E )ma?- b(3 cstablisllcd by working the composite oprrator L E Oyp2,(in fact?thi!: is the only as bcforc, [reprating thr stvps from (X.;) to (:<.11)] and difficult step; oncc such n rcprcsentntiotl is obtnincd, the noting that d H ( f ) 3 E, +f t E R'. Q.E.D. determination of theconstants { cyLi\ and { although at first. glance appears t.o be difficult, is quite simple Frequellcy-Donlaitl Stability Criteria The positivity condition obtnincd in the previous srct.ion irrespective of the dimensions of (; and U ) . Even in this mnkwthe proof of the following stnbilit?-criterion for stcg, thr difficult part is only to ensure R phase-variable linrar system :I straightforward application ofthcl posi- rcalizntion of I,, sincc. withsuch a representation being ensured, tllc other requirements result in a simplc manner tivity tllcorem. from Tlreorem 1: If t h t w exists an operator AI E o ~ such ~ . ~ the imposition of suitablr restrictions on the describdifficulty isgreatlyreducedby the that J/ is invertible in I,?. L = J I ( ; E O I p . L nnd the follow- ingelements.This availabi1it.y of 11-rstablished algorithms [ 111. [12] m-hich ensure for any uniformly controllable and differentiable realizat,ion of L , thc c.xistencc of n uniquenonsingular 2 Sote that all the requirements of the proposition are fulfilled A a H i ( f )and c ~ i ( tare ) differentiablen-time. and since H f ~ I P implies u L x u L matrix, which transforms the realization of L rff;(t)= 0 at f = 0 and f = a.Further, r,r;(f) being the state-vector to the phnsc-variable form. components of H , i,;(t) = ~ H , . i + l ) ( f ) , V i = 1,2,. . .,(n - 1). + 2 0 + + ~~--c. 221 SUNDARESHAN A N D THATHACHAR: NONSTATIONARY FEEDBACK SYSTEAfS being the coefficients defined as in (3.2), n. being the dimension of G and cGO(t)= 0. Then the syst,em described by (2.7), mith the additiona.1 The Case of the Time-carying Popoc ll4ultipdiw assumption uz(.) = uzf(.) 3 hf(.) E L2, is &-stable for Alt.hough it is evident that,t,he stability criteria. given inall F E SP. The proof of t.his thcorenl is omitt.ed due t.o space rethe previous sect.ion, which permit, the use of a very general class of t,ime-varying operat,ors in Lz as ndtipliers, quirements. The general patt,ernof development, however, are more general t,hanthe existing results,the advantage of folloms very closely the proof of the lemma in Section 11, using a t,ime-varying mult,iplier is yet. to be decisively dem- in establishing the st.rong positivity(e) of MG. 4 complete onst.rated. Inthis sect.ion, we propose t.o do t>his by proof may be found in [ 131. using asimple first.-order differentialoperator asthe nlultiplierandderiveastabilitycriterion which will be Discussion of the Result compared d h the existing results that. employ similar but. 1) AhhoughTheorem 2 contains a. frequency-domain time-invaria,nt multipliers. As an a.pplicat,ion of the resultsderived, L2-st.a.bilityconditions for the nonlinear inequality similar to t,hose appearing in Theorem 1, the da.mped Jlathieu equation arc obtained in a subsequent. conditions are explicitJy on the e1ement.s describing G' section and compared qith the results of t,he earlier in- (note in comparison that, Theorem 1 imposes condit.ions on the composite operat.or L = JIG) and hence, are far vestigat.ors. simplcr t.0 check. The complexity of the expression for t,he The multipliers that will be considered in this srction are operators E OE having a decomposition M = Q D coefficients { a i l nrednot beaseriousdrawback,since where Q is atime-varying gain defined by Qz(t)= q ( t ) x ( t ) , the evaluat,ion of these is very simple even for systems of 'd: x(.) E Lz,, q(t) being a nonnegativefunction on R+ large dimensions. 3) The advantage of using a time-varying multiplier is diffrrentiable almost everywhere, and D, t,he differential opera.tor in Lf,, D x ( t ) = i ( t ) , ++ x(.) E L2,.Let us denote illustrated in t,he following cxa.mplr. Excmple: Consider the syst.em Tvith t.hr linrar part G t,he class of such operators by a?. It is sinlple to observe . that X E -mp=) M - I exists as an operator in Lfe. One gowrned by t.he nona,utonomousdifferrntialequation, particular point needs to be emphasized when the use of multipliers M E is cont,emplated.Since M = Q D is not bounded operat,or in L2? the familiar introduction of &I into the loop (see Fig. 2 ) will change the nature of theinputsto t.hcsystem(notet.hat M u ? E L?). This where @(t)is a gain?the bounds on whosr rate of variation difKculty may however be ovwcome, following Zanxs [I], arc. to be determined for the sta,bility of t.he feedba.ck loop by restricting u.l(.) t.0 be a fixed funct,ion tizf( .) in & such comprising of a nonlinear operator F E 5 pin cascade wit.h (i. that GI(.) E LI. Theorem 9: If G' E O I p A and t>here exists an operat,or Rcpresent.ing G in the phase-variable canonical form, we ME%, satisfying the following conditions: have the describing q ~ a d r u p l e , ~ 117. SIMPLIFIED CRITERIAUSING FIRST-ORDER &,j MULTIPLIERS + + r, Using Brockett,'s result! [14], it maybeobserved that. G' E o~~~~ if 0 < e' < @(t) 6 11.5. For a more rest.ricted form of the syst,em (i.e., with dG(t) 0 and no inputs into the feedback loop), Blodget,t (4.3) + kGj(t) + CG(j-I)(t) - cGn(t)aGj(t))]'"+'-"' The considerat,ion of t,he particular form of $ ( u ) in (4.1) is merely t.o facilitate a simple phasevariable realization of G. 222 IEEE TRANSACTIOSS ON AUTOMATIC CONTROL, JUNE 1974 I € 1 Fig. 3. Stability regions for t.he hiat.hieu equation from (4.9). + G E f l I p A is 0 < ( b E cos p t ) 6 11.5; this will be satisfied if b and E arerestricted by 0 < ( b f e ) 6 11.5. [Sote howevcr that this bound does not make US(> of the fact. that the time varying coefficient is periodic with period 5 ~ : p ;more relaxed bounds, for larger values of p , may be obtained by using 1J-illrms' [15] result for periodic systems. ] In the follon-ing, two diffcrcnt choices of q ( t ) are made and corrwponding stability regions are obtained in order to dcmonstratc tllnt diffcrcnt choice::of q(t) often Irad to improved results. Clroice X o . 1: Let so as to satisfy (4.G). Substituting this in ( 4 . i ) and simplifying, we have Sot(. that (4.G) is satisfied. Substituting in (4.7) and simplifying, we 1 1 : ~ ~ 223 S U N D A R E S W N A N D T H L T H A C W R : XONSTATIONARY FEEDBACK SYSTEMS Fig. 4. Stability regions for the Mathieu equation from (4.11). + ( e / p ) sin p t ) gives a larger that the stability region obt,a.ined presentlyis close to p = 1, the choice of q ( t ) = (1 Parks' region in the range -2 < E < +S. Further, it may region in the rm1ge.b 6 0.5 than the choice of q(t) = be observed from Fig. 4 that the condit,ion (4.11) gives (1 - ( E , / ? ) cos kt), while the reverse is true for b > 0.5. t,hesame regions asParks*intheparameter ra.nges However, since the stability results obtained through the b 6 0.5 for p = 1 and b 1.3 for p = 2: whereas for use of these periodic mukipliers are only sufficient. conditions, the bounds on the parameters of t,he 11at.hieu equap = 5, the region presently obtained is hrger in therange tion (4.5) which ensure st.ability of t.he feedback sgst.enl b 4.5. 2 ) I n comparison with the L,-stabilit,\T crit.erion of Wu ma.y correspond to either of these two regions. I n other and Desoer [17] (for t.he problem present.ly considered, words, a union of t.hc twostability regions is itself a i.e., damping coefficient. = 2, t.he condition from [17] is stabilitJ7 region for tjhe syst,em. Continuing on this theme, b > I E ~ 0.75), it. may be seen that. (4.9) gives a la.rger one ma.y conclude that the present, met.hod of analysis region in t.he pa.rameter ranges, b 2.75 for p = 1 and involving t.he use of periodic multipliers, has opened a new b 6 2.25 for p = 2 , while (4.11) gives a larger region in avenue for obtaining enlarged stability regions t,hrough a t,he ranges, b 6 1.5 when p = 1, b 6 2.25 when U, = 2 and judicious choice of different periodic functions q(1) in the b 6 4.75 when p = 5. Further, it may be noted that in mult,iplier. t.he latter case, the improvement, in $he stabilit,J: region V. CONCLUSIONS increases nit.h increased p. Thus one of the st,rong points of t.he present criteria is that information on the value of The problem of developing L d a b i l i t y criteriafor linear and nonlinear feedback systems containing a timethe parameter p is exploit,ed to obtain enlarged stability regions. varying linear operatorin Lz, has been trea,t.ed. A frepositivity condition for t.ime-varying 3) It should however be noted that the above compari- quency-domain son is notactuallyfair to the presentcriteria;for, the 1inea.r operat,orsisderived andis used to generate LZresults of Parks [16] a,nd Wuand Desoer [17] were stability criteria, which permit t.he use of a very general class of mult,iplicrs withnonstationarykernels. originally derived for the Mathieu equation without any For the feedback nonlinearity. If the latter results are t,o remain case of feedba.ck systems containing a Popov-type nonvalid in the nonlinear case, the nonlinearity should be linearity, more explicit. conditions in the frcquencyconfined t.0 the sector [ O , E ] whereas t.he present critcria domain, involving a firsbordertime-varyingdifferentmid permit it to bein the infinit,e sect.or. Thus in spiteof con- multiplier are derived and a.re shown t.o improve upon the sidering a more general syst.em, the present. results give existing results. The use of periodic multipliers for the Lzcertain st.ability rrgions not contained in t,hc earlier ones.stability of nonlinear damped3Iat.hieuequationwitha A Ge)lera.l Remark: A comparison of the stabi1it.y regions forcing function,issuggested and is shown to result in regions not. contained in those given by the existobtained in t.his section through the choice of t.wo diffrrcnt stnbilit>~ periodic functions q ( f ) , shown in Figs. 3 and 4, reveals an ing critcria even for the case of the linear equa,tion. The int.erest,ing point.. Note that for the part,icular case of positivity criterion obtained in this paper, although motivated from a desire t o use it in t.he stability analysis, is of independent interest and has applications in various 4 Parks' result is derived by the use of the Circle crit.erion and other area5 of IIathrmatical System Theory. hence does not, give different conditions for different. values of p . < < + < 224 IEEE TRANSACTIOXS ON AUTOMATIC CONTROL, JUNE REFEREXCES G. Zames, “On the input-output stabilitlr of timevarying nonlinearfeedbacksystem-Paxt I: Condition derived using concepts of loop gain, conicity, and positivity,” IEEE Trans. dutonmt. Contr., vol. AC-11, pp. 228-2335, Apr. 1966-“On the input-output stability of timevarying nonlinear feedback 11: Conditions involving circle in the fresystem-Part q-uency plane and sector nonlinearity,’’ pp. 465476. Jnly 1966. G. Zames and P. L. Falb,“Stability conditions for systems with monotone and sloperestricted nonlinearities,” S I A N J . Contr., vol. 6, pp. 89-108, 1968. AI. K. Sundareshan and AI. A . L. Thathachar, “&stability of linear time-varying system-Conditions involving noncaosal multipliers,” I E E E Trans. Automat. Contr., vol. AC-17, pp. .504-.510, Bug. 1972. 31. I. Freedman and G. Zames, “Logarithmic variation criteria for the stability of systems with time-varyinggainr,“ SZAJl J . Contr., vol. 6, pp. 4Si-507, 1968. J. C. Willems, “Least squares stationary optimal control and the Algebraic Iticcati equation,” I E E E Trans. Automaf. Contr., vol. ric-16, pp. 621434, Dee. 19’71. K. F. Estrada and C. A . Ileaoer, “Passivity and stability of systems with a state repre*entation,” Int. J . Contr., vol. 13, pp. 1-26, 19i1. 31. K. Sundareshan and 31. .I. L. Thathachar, “Time-domain criteria for the L-stabilitl- of nonstationary feedback s>-stetns,“ IEEE Trans.dutonmt. Conir. (Tech.Sote.and Corresp.), vol. XC-135, pp. 80-81, Feb. 1973. I<. E. Blodgett and K. P. Young, “A combined timefrequencJcondition for stability of time-varying systems with one 11011linearity,” Trans. ASJIE (J. Dynamic Syst. 31eas. Contr.!, vol. 93, series G , pp. 261-268, Dec. 1971. J . C. Willen1s, The Analysis of Feedback S y s t e m . Cambridge, Mass.: 31. 1. T. Press, 1970. J. C. Willems and R. W. Brockett, “Some nex rearrangement inequalities havingapplication in stabilitv analysis,“ IEEE Trans. A u t o m a t . Cotztr., vol. AC-13, pp: 5 3 - 5 4 9 , Ort. 196s. I l l 1 L. 11.Silverman, “Transfornlation 01 tmle-v:triable systems to canonical (phasevariable) form,” I E E E Trans. dcrto?nnt. Contr.. vol. AC-11, pp. 3OC-303, Apr. 1966. [121 B. Rdnlasn-ami and K. Ramar, “On the transformation of timevariable s-stems to the phase-variable canonical form,‘‘ IEEE Trans. Autonzat. Contr. (Corresp.1, vol. Ac-14, pp. 415419, - h g . 1969. 31. K. Sundareshan and 11. A . L. Thathachar. “Sew results for the Lrstability of nonstationary feedback system^," Ilept. Elec. Eng.,IndianInstitute of Science, Bangalore, Tech. Hep. EE-26, Oct. 1972. [141 R . kV. Brockett, “Optimization theory and the converse of the Circle Criterion,” Proc. :\-E(?, pp. 697-701,196.5.. 1974 [I51 J. C. Willems, “On the asymptoticst.ability of the null solution of linear differential equations with periodic coefficients,” IEEE Trans. Azctomal. Coontr., vol. AC-13, pp. 65-72, Feb. 1968. [IF] P. Parks, “Circle criterion and the damped 3Iathieu equation,” Electron. Lett., vol. 2, no. 8, p. 315, Aug. 1966. [I?] 31. \-. Wu and C. A. Ilesoer, “LP-stability (1 _< p _< a ) of nonlinear timevarying feedback system,” SIB.11 J . Coontr., ~ 0 1 7, . pp. 356-364, 1969. IC. Sundareshan wa5 born in ltobertsonpet, Kolar Gold Fields, India, on June 16, 1946. He received the B.E. degree i n electrical engineering from the Bangalore University,Bangalore, India in 1966 andthe 3I.E. and Ph.1). degrees in control s;?-stems engineering from the Indian Institute of Science, Bangalore, India in 1969 and 1Yi3, respectively. Since January 1973, he has been with the School of Automation, IndianInstitute of Science aud is engaged i n the design and development of systems for flight path control. Hi:: current research interests lie i n the areas of linesr and nonlinear 2;>-stemtheor>-, optimal performance of flight control sh-stemsand computational methods. Malur M. A. L. Thathachar w s born in 31ysore City,India in 1939. He received the B.E. degree in electrical engineering from the University of Llysore, 313-sore, India, in 1959 andthe 3I.E. and Ph.11. degrees from the Indian Institute of Science, Bangalore i n 1061 and 1968, respectively. He has been on the faculty of the Indian Institute of Technology, 31adra (during 1961-64) and of theIndianlnstitute of Science. Bangalore (since10611, where heis currently -k+ociate Profesor. He spent a Sabbatical year in 19T3-54 at Tnle Illriversit>-,S e a Haven, Conn., and at Sir George \I-illiams Vniversity, Jlontreal,Canada. His research interestsare in the areas of Stability Theory and Learning Systems.