Robustness in the Gap Metric and Coprime Factor Perturbations for LTV Systems Seddik M. Djouadi Abstract— In this paper, we study the problem of robust stabilization for linear time-varying (LTV) systems subject to time-varying normalized coprime factor uncertainty. Operator theoretic results which generalize similar results known to hold for linear time-invariant (infinite-dimensional) systems are developed. In particular, we compute a tight upper bound for the maximal achievable stability margin under TV normalized coprime factor uncertainty in terms of the norm of an operator with a time-varying Hankel structure. We point to a necessary and sufficient condition which guarantees compactness of the TV Hankel operator, and in which case singular values and vectors can be used to compute the time-varying stability margin and TV controller. D EFINITIONS AND N OTATION • B(E, F ) denotes the space of bounded linear operators from a Banach space E to a Banach space F , endowed with the operator norm kAk := sup kAxk, A ∈ B(E, F ) x∈E, kxk≤1 • `2 denotes the usual Hilbert space of square summable sequences with the standard norm kxk22 := ∞ X ¡ ¢ |xj |2 , x := x0 , x1 , x2 , · · · ∈ `2 j=0 • • Pk the usual truncation operator for some integer k, which sets all outputs after time k to zero. An operator A ∈ B(E, F ) is said to be causal if it satisfies the operator equation: Pk APk = Pk A, ∀k positive integers • tr(·) denotes the trace of its argument. The subscript “c ” denotes the restriction of a subspace of operators to its intersection with causal (see [19], [6] for the definition) operators. “⊕” denotes for the direct sum of two spaces. “? ” stands for the adjoint of an operator or the dual space of a Banach space depending on the context [4], [5]. I. I NTRODUCTION The gap metric was introduced to study stability robustness of feedback systems. It induces the weakest topology in which feedback stability is robust [1], [20], [18], [22], [24]. In [20] Georgiou showed the relationship between the gap metric and a particular two-block H ∞ problem. In [18], the authors showed that feedback optimization in the gap metric is equivalent to feedback optimization S.M. Djouadi is with the Electrical Engineering & Computer Science Department, University of Tennessee, Knoxville, TN 37996-2100. djouadi@ece.utk.edu with respect to normalized factor perturbations. They computed the largest possible uncertainty radius such that robust stability is preserved. Extensions to time-varying systems have been proposed in [26], [21] where a geometric framework for robust stabilization of infinite-dimensional time-varying systems was developed. The uncertainty was described in terms of its graph and measured in the gap metric. Several results on the gap metric and the gap topology were established, in particular, the concept of a graphable subspace was introduced. In [6], some of the results obtained in [20] were generalized, in particular, the gap metric for time-varying systems was generalized to a two-block time varying optimization analogous to the two-block H ∞ - optimization proposed in [20]. This was achieved by introducing a metric which is the supremum of the sequence of gaps between the plants measured at every instant of time. The latter reduces to the standard gap metric for linear time-invariant (LTI) systems. In [7], [6], [8] using the time-varying gap metric it is shown that the ball of uncertainty in the time-varying gap metric of a given radius is equal to the ball of uncertainty of the same radius defined by perturbations of a normalized right coprime fraction, provided the radius is smaller than a certain quantity. In [8] tight lower and upper bounds are derived for computing the maximal stability margin (quantified in the TV gap metric), in terms of using coprime factorizations. These bounds are equal to the maximal stability margin for LTI systems. In [17], the authors showed that the time varying (TV) directed gap reduces to the computation of an operator with a TV Hankel plus Toeplitz structure. Computation of the norm of such an operator can be carried out using an iterative scheme known to hold for standard two-block H ∞ problems [17]. The minimization in the TV directed gap formula was shown to be a minimum using duality theory. In this paper, we use the equivalence between uncertainty quantified in TV gap metric balls and coprime factor uncertainty for LTV systems. We study the problem of robust stabilization for time-varying normalized coprime factor perturbations and obtain operator theoretic results, which generalize similar results in [11], [18], [22] known to hold for LTI systems. In particular, we compute a tight upper bound on the maximal achievable stability margin under TV normalized coprime factor uncertainty in terms of the norm of a TV Hankel operator. The upper bound reduces to the maximal stability margin for LTI systems. We point to a necessary and sufficient condition which guarantees compactness of the TV Hankel operator, and in which case singular values and vectors can be used to compute the TV optimal stability margin and TV controller. Therefore, generalizing similar results obtained in [22] for LTI systems to TV ones. The technique developed to compute the upper bound also applied to the lower bound on the maximal achievable stability margin in the TV case. To this end, it suffices to restrict the relevant operators to a particular subspace. The rest of the paper is organized as follows. In section II the gap metric is introduced. In section III the relation between the TV gap metric and coprime factorization is discussed, and computations in terms of operator theory are developed. In section IV the computation of the optimal controller via the Youla parameter is presented. We conclude with a summary of the paper contributions in section V. II. T HE T IME -VARYING G AP M ETRIC LTV systems may be regarded as causal linear (possibly unbounded) operators acting on `2 as multiplication operators. To each plant P we associate its domain D(P ) = {u ∈ `2 : P x ∈ `2 } (1) A LTV plant P is stabilizable if there exists a LTV controller C such that the operator [27], [6] µ I −P −C I ¶ : D(P ) ⊕ D(C) −→ `2 ⊕ `2 (2) is invertible with bounded inverse. The inverse is given by µ −1 (I − CP ) P (I − CP )−1 −1 C(I − P C) I(I − P C)−1 ¶ (3) It is readily seen from (3) that P is stabilizable if all its entries belong to Bc (`2 , `2 ). The LTV plant P has a right coprime factorization if there exist operators M and N both in Bc (`2 , `2 ), such that, P = N M −1 , and a left coprime factorization if there exist M̂ , N̂ ∈ Bc (`2 , `2 ), such that P = M̂ −1 N̂ . In addition there exist X, Y ∈ Bc (`2 , `2 ), and X̂, Ŷ ∈ Bc (`2 , `2 ), such that, XN + Y M N̂ X̂ + M̂ Ŷ = I = I (4) Such factorizations exist if and only if P is stabilizable [23]. That is, The LTV plant P is stabilizable if and only if it has left and right coprime factorizations P = M̂ −1 N̂ = N M −1 (5) There exist causal bounded linear operators U, V, Û , and V̂ such that µ ¶µ ¶ V̂ −Û M U N V −N̂ M̂ µ ¶µ ¶ M U V̂ −Û = N V −N̂ M̂ µ ¶ I 0 = (6) 0 I and the all P stabilizing LTV controllers C can be parameterized as C = (U + M Q)(V + N Q)−1 = (V̂ + QN̂ )−1 (Û + QM̂ ), Q ∈ Bc (`2 , `2 ) (7) Following [18], [6] we µ are interested in normalized coprime ¶ M factorizations. That is, is an isometry from `2 into N `2 ⊕ `2 , in this case, M ? M + N ? N = I as is in the LTI case, where M ? , N ? are the adjoint operators of M and N , respectively [6]. Suppose that two LTV plants Gµ 1 and G ¶2 have normalized µ ¶ M1 M2 right coprime factorizations and , N1 N2 respectively. Denote by Π1 and Π2 the orthogonal projections on their ranges. Then the gap between between G1 and G2 is defined by [6] δ(P1 , P2 ) = kΠ1 − Π2 k and the directed gap → − δ (G1 , G2 ) = k(I − Π2 )Π2 k The gap is then [18], [21] ¡→ ¢ − − → δ(P1 , P2 ) = max δ (G1 , G2 ), δ (G2 , G1 ) 9) (8) (9) (10) It is well known that if δ(G1 , G2 ) < 1, then all these quantities are equal. µ ¶ µ ¶ M1 M2 Normalized coprime factorizations and N1 N2 imply that µ ¶ µ ¶ M1 M2 (I − Pn ), (I − Pn ) (11) N1 N2 are isometries on (I − Pn )`2 with range in (I − Pn )`2 ⊕ (I − Pn )`2 [7], [6]. Following [7], [6] µ Let Π1n¶denote the M1 orthogonal projection on the range of (I − Pn ), N1 Π2n is defined similarly. Define °½µ ° ¶ ¾ ° ° − → I − Pn −0 ° δn (G1 , G2 ) = ° − Π2n Π1n ° ° (12) 0 I − Pn and δn (G1 , G2 ) = kΠ1n − Π2n k ¡→ ¢ − − → = max δn (G1 , G2 ), δn (G2 , G1 ) (13) Lemma 1: [7] °µ µ ¶ ¶ ° M2 ? ? ° δn (G1 , G2 ) = ° I − (I − Pn )(M2 , N2 ) N2 ° µ ¶ ° M1 (I − Pn )° ° (14) N1 The following result generalizes the time-invariant counterpart derived in [20]. Theorem 1: [6] °µ ¶ µ ¶ ° ° M1 ° − → M2 sup δn (G1 , G2 ) = inf 2 2 ° − Q° ° ° (15) N1 N2 Q∈Bc (` ,` ) n≥0 The directed time varying gap between G1 and G2 is then defined as − → → − α (G1 , G2 ) := sup δn (G1 , G2 ) (16) n≥0 and the time varying gap [6] ´ ³ → → α(G1 , G2 ) = max − α (G1 , G2 ), − α (G2 , G1 ) (17) The function α is a metric and for time-invariant systems reduces to the standard gap metric δ [7]. Note that in the time-invariant case Theorem 1 reduces to the two-block H ∞ optimization derived for the gap metric in [20]. In [17] − → α (G1 , G2 ) was solved in terms of a time-varying Hankel operator, and in fact using duality theory the infimum in (15) is shown to be attained by some Q in Bc (`2 , `2 ). Denote by Bs (P, r) the set of all P1 with right coprime factorization N1 M1−1 for which °µ ¶° ¶ µ ° M M1 ° ° °<r − (18) ° M N1 ° and denote by B(P, r) the set of all plants P1 such that α(P, P1 ) < r. then the following result in [8] holds. Theorem 2: [8] B(P, r) = Bs (P, r) (19) Theorem 2 relates coprime uncertainty balls to balls defined in the time-varying gap metric. In particular, maximizing the uncertainty radius for coprime uncertainty results in maximizing the uncertainty radius quantified in the TV gap metric. In the next section, we study the problem of robust stabilization for time-varying normalized coprime factor perturbations and obtain operator theoretic results, which generalize similar results in [11], [18], [22] known to hold for LTI systems. III. ROBUST S TABILIZATION U NDER C OPRIME FACTOR U NCERTAINTY In [9] the authors proved a robust stabilization result under proper coprime factor uncertainty for LTI systems. This result was generalized to LTV systems in [8], and is summarized in the following theorem. Theorem 3: Let C be a controller with right coprime representation C = (U + M Q)(V + N Q)−1 Then the following are equivalent 1) C stabilizes all plant P1 with normalized left coprime factorization ¡ ¢−1 (20) P1 = M̂ + ∆M̂ (N̂ + ∆N̂ ) where ∆M̂ , ∆N̂ ∈ Bc (`2 , `2 ) and k(∆M̂ , ∆N̂ k < r, for some r > 0. 2) °µ ¶° ° U + MQ ° 1 ° ° (21) ° V + NQ ° ≤ r A similar result holds for (TV) normalized right coprime factorizations. If ropt is the supremum over all r such that C stabilizes Bs (P, r), then for TV systems the following inequality holds [8] °µ °¾ ¶ ½ ° U + MQ ° ° (I − Pn )° inf inf2 2 ° ° V + N Q n≥0 Q∈Bc (` , ` ) °µ ¶° ° U + MQ ° 1 ° ° ≤ ≤ inf V + NQ ° ropt Q∈Bc (`2 ,`2 ) ° (22) For time-invariant systems these numbers are equal. In the same vein as [18], we define ro as the supremum over all r such that there exists a ¡fixed controller ¢−1 C which stabilized all plants P1 with P1 = M̂ + ∆M̂ (N̂ + ∆N̂ ), ∆M̂ , ∆N̂ ∈ Bc (`2 , `2 ) and k(∆M̂ , ∆N̂ k < r. It follows that °µ ¶ µ ¶ ° ° U ° M −1 ° inf ro = + Q° (23) ° V N Q∈Bc (`2 ,`2 ) ° The infimum is achieved for some Q ∈ Bc (`2 , `2 ) [16], [17]. In the sequel we are concerned with solving the optimization (23) which reciprocal is an upper bound for the optimal robustness radius. The solution proposed applies as well to the lower bound in (22) by a restriction to a subspace, that is, by restricting the domain of definition of the operators introduced in the sequel to the range of (I − Pn ). Now observe that (as in the time-invariant case) the operator µ ¶ M? N? Ž := ∈ B(`2 , `2 × `2 ) (24) −N̂ M̂ and the operator induced norm °µ °µ ¶° ¶ ° ° A ° ° A ° ° ° := ° sup f° ° B ° ° ° B f ∈`2 , kf k2 ≤1 2 ³ ´ 12 = kAf k22 + kBf k22 µ ¶ A ∈ B(`2 , `2 × `2 ) B (25) (26) is unitarily invariant, that is, for any unitary operator U ∈ B(`2 , `2 × `2 ), we have ° µ ¶° °µ ¶° ° ° ° A ° °U °=° A ° ° B ° ° B ° In particular, for U = Ž, we have °µ ° µ ¶ µ ¶ ° ¶ µ ¶ ° ° U ° ° ° M U M ° ° ° + Q° = °Ž + Q° ° V ° N V N °µ ¶° ° R+Q ° ° = ° ° ° I ¡ ¢1 = 1 + kR + Qk2 2 (27) where R := M ? U + N ? V ∈ B(`2 , `2 × `2 ). Next, we show that ¡ ¢− 1 ro = 1 + kHR k2 2 (28) where HR is the time-varying Hankel operator with symbol R. Note that as in the LTI case 0 < ro ≤ 1. To define the operator HR , we need some mathematical preliminaries. Let Qn := I − Pn , for n = −1, 0, 1, · · · where Pn is the standard truncation operator, which also an orthogonal projection that sets all outputs after time n to zero, and P−1 and P∞ := 0 := I The quantity ro is a tight bound on the maximal achievable TV stability margin under coprime factor uncertainty for LTV systems, and reduces to the maximal stability margin for LTI system. Proof. Let Π1 be projection on the subspace µ the orthogonal ¶ M (A2 ⊕ A2 ) ª A2 the orthogonal complement of N µ ¶ M A2 in the operator Hilbert space A2 ⊕ A2 under N the inner product (A, B) := tr(B ? A), A, B ∈ A2 ⊕ A2 Then Qn is a projection, and we associate to it the following nest N := {Qn `2 , n = −1, 0, 1, · · · } The orthogonal projection is then defined on the spaces µ ¶ M Π1 : A2 ⊕ A2 7−→ (A2 ⊕ A2 ) ª A2 N Call The space of causal bounded linear operators Bc (`2 , `2 ) can be viewed as a triangular or nest algebra, which leaves invariant every subspace N ∈ N , that is, T N ⊆ N , ∀T ∈ Bc (`2 , `2 ). In fact, Bc (`2 , `2 ) can be written as Bc (`2 , `2 ) = = where HR = (I − P)(M ? U + N ? V ). Formula (32) is the time-varying analogue of the standard Nehari problem. The following expression for ro was obtained in the LTI case was obtained in [11] using state space techniques, and in [18] using operator theory. in [11]. We will give its time-varying counterpart and give an operator theoretic proof along the lines of [18]. Theorem 4: ° °2 ° ° ro−2 = 1 − °H (33) ° M̂ ? N̂ ? {A ∈ B(`2 , `2 ) : Pn A(I − Pn ) = 0, ∀ n} {A ∈ B(`2 , `2 ) : (I − Qn )AQn = 0, ∀ n} (29) Now, call C2 the class of compact operators on `2 called the Hilbert-Schmidt or Schatten 2-class [10], [3] under the norm, ³ ´ 12 kAk2 := tr(A? A) µ S := (A2 ⊕ A2 ) ª (30) Then A2 is the space of causal Hilbert-Schmidt operators. Define the orthogonal projection P of C2 onto A2 . P is the lower triangular truncation or nest projection. Following [15] an operator X in B(`2 , `2 ) determines a Hankel operator HX on A2 if HX A = (I − P)XA, 2 2 for A ∈ A2 (31) 2 For R ∈ B(` , ` × ` ) the TV Hankel operator solves the minimization [16] inf Q∈Bc (`2 ,`2 ) kR + Qk = kHR k (32) ¶ A2 Define the operator Ξ : Ξ := A2 µ 7−→ A¶2 ⊕ A2 U Π1 V (34) Using the Commutant Lifting Theorem [6], it follows ro−1 = kΞk (35) Next, let Γ be the left multiplication operator Γ Γ Define the space A2 := C2 ∩ Bc (`2 , `2 ) M N : S 7−→ A2 := (−N̂ , M̂ )|S Then, for X ∈ A2 we have µ ¶ µ ¶ U M ΞX = X− Y ∈S V N for some Y ∈ A2 and ·µ ¶ µ ¶ ¸ U M Γ(ΞX) = (−N̂ , M̂ ) X− Y =X V N therefore ΓΞ = I on S. Moreover, for s ∈ S, we have Γs = = (−N̂ , M̂ )s ·µ ¶ µ ¶ ¸ s1 M (−N̂ , M̂ ) − s3 s2 N (36) ° ° its norm °H for some s1 , s2 and s3 in A2 . Now, apply Ξ to get µ ¶ µ ¶ U M Ξ(Γs) = (−N̂ , M̂ )s − (−N̂ , M̂ )s V N and thus Ξ(Γs) = s showing that ΞΓ = I on S, ΓΞ = I on A2 , that is Γ is the inverse of Ξ, Γ = Ξ−1 . This implies inf s∈S, ksk2 ≤1 kΓsk−1 =: τ (Γ)−1 (37) ? Now, note that the adjoint operator of Γ, Γ is defined by Γ? Γ? X : A2 − 7 →S µ ¶ −N̂ ? = Π1 X, X ∈ A2 M̂ ? and the operator Υ defined by is equal to the maximal singular M̂ ? N̂ ? value. A necessary and sufficient condition for any Hankel operator HX to be compact is that X belongs to a triangular algebra A plus the space of compact operator denoted K, that is, X ∈ A + K [15]. In our case, if we define K as the space of compact operator from A°2 into C2 and ° ° ° A := B(A2 , A2 ⊕ A2 ), then °H ° is compact if ? M̂ N̂ ? and only if µ ¶ M̂ ? ∈A+K N̂ ? which by (6) is equal to µ ¶ µ ¶ U M Ξ(Γs) = (−N̂ , M̂ )s − (V̂ , Û )s V N kΞk = ° ° ° (38) This implies that H attains its norm at some X ∈ M̂ ? N̂ ? A2 . As in the LTI case [22], this has implications for the operators Ξ and Γ, in particular Ξ attains its norm. This is summarized in the following theorem. Theorem 5: Let 0 < λ < 1. Then the following are equivalent 1) λ is a singular value of Υ. A2 7−→ C2 ª A2 =: A⊥ ? ⊥ 2 2) There exists X µ ¶ µ 6= 0¶ ∈ A2 , Y 6= 0 ∈ A2 , W ∈ −N̂ ? M Υ = (I − P) A2 such that (A2 ⊕ A2 ) ª M̂ ? N ¶ µ µ ¶ −N̂ ? 1 −N̂ ? defined Consider the left multiplication operator X − (1 − λ2 ) 2 W = λY ? M̂ ? M̂ ? from A2 into C2 which is isometric since we are using (−N̂ , M̂ )Y ? = λX (41) normalized coprime factorizations. Then for any X ∈ A2 1 we have 3) (1 − λ2 ) 2 is a singular value of Γ. µ ¶ µ ¶ µ ¶ ? ? ? ? ⊥ −N̂ −N̂ −N̂ 4) There exists X µ 6= 0¶ ∈ A2 , Y 6= 0 ∈ A2 , W ∈ X = P X + (I − P) X ? ? ? M M̂ M̂ M̂ A2 such that (A2 ⊕ A2 ) ª µ ¶ µ ¶ N ? −N̂ −N̂ ? = Π1 X + (I − P) X 1 M̂ ? M̂ ? (−N̂ , M̂ )W = (1 − λ2 ) 2 X (42) ? µ ¶ = Γ X + ΥX 1 −N̂ ? X − λY ? = (1 − λ2 ) 2 W (43) M̂ ? But since ΓΓ? + Υ? Υ = I, that is, the operator Γ? + Υ is an isometry, and τ (Γ? )2 + kΥk2 = 1. Further, Ξ? Γ? = I 1 1 5) (1 − λ2 ) 2 is a singular value of Ξ. and so kΓ? k ≥ kΞk , i.e., Γ? is bounded below implying Proof. Follows as in the LTI case [22]. kτ (Γ)k = τ (Γ? )k. By definition IV. O PTIMAL ROBUSTNESS IN THE TV G AP M ETRIC ° ° ° ° (39) kΥk = °H ° ? M̂ In this section we give a necessary and sufficient condition and generalizes for λ to be a singular value of H N̂ ? M̂ ? This yields N̂ ? its LTI counterpart. °2 ° ° ° (40) ° +kΞk−2 = 1 °H Theorem 6: Let 0 < λ < 1. Then λ is a singular value ? M̂ of Υ. if and only if there exists W 6= 0 ∈ A2 such that ? N̂ ³µ −N̂ ? ¶ ´ 1 which implies the result of the theorem (33). (−N̂ , M̂ ) − (1 − λ2 ) 2 I W ∈ A2 (44) ? M̂ is compact, then µ ¶ If the Hankel operator is H M̂ ? M Moreover, if (44) holds then W ∈ (A2 ⊕ A2 ) ª A2 N̂ ? N Υ : and Theorem 5, 3) and 4) hold for [9] M. Vidyasagar and H. Kimura, Robust stabilization for uncertain linear multivariable systems, Automatica, vol. 22, pp. 85-94, 1986 1 [10] Schatten R. Norm Ideals of Completely Continuous Operators, X = (− N̂ , M̂ )W (45) 1 Springer-Verlag, Berlin, Gottingen, Heidelberg, 1960. (1 − λ2 ) 2 [11] K. Glover and D. McFarlane, Robust Stabilization of Normalized µ ¶ ³ ´ ? 1 Coprime Factor Plant Descriptions with H ∞ -Bounded Uncertainty, −N̂ 2 12 Y? = I W (− N̂ , M̂ ) − (1 − λ ) 1 IEEE Trans. Automatic Contr., vol. 34, pp. 821-830, 1989. M̂ ? λ(1 − λ2 ) 2 [12] Francis B.A., Doyle J.C. Linear Control Theory with an H∞ Opti(46) mality Criterion, SIAM J. Control and Optimization, vol. 25, (1987) 815-844. The optimal operator Qo ∈ Bc (`2 , `2 ) which achieves the [13] Francis B.A. A Course in H ∞ Control Theory, Springer-Verlag, 1987. infimum in (23) and exists by a result in [16], i.e., [14] Zhou K., Doyle J.C., Glover K. Robust and Optimal Control, Prentice °µ ¶ µ ¶ ° Hall, 1996. ° U ° M [15] Power S. Commutators with the Triangular Projection and Hankel ° ° + Q (47) o ° V ° N Forms on Nest Algebras, J. London Math. Soc., vol. 2, (32), (1985) °µ ¶ µ ¶ ° 272-282. ° ° U M [16] S.M. Djouadi and C.D. Charalambous, On Optimal Performance for + Q° = inf 2 2 ° ° ° V N Linear-Time Varying Systems, Proc. of the IEEE 43th Conference Q∈Bc (` ,` ) on Decision and Control, Paradise Island, Bahamas, pp. 875-880, = kΞk December 14-17, 2004. [17] S.M. Djouadi and Y. Li, On Robust Stabilization in the Gap Metric and can be computed from the identities for LTV Systems, Proceedings of the IEEE Conference on Decision and Control, San-Diego December 2006. ? ΞXo = kΞkWo , Ξ Wo = kΞkXo (48) [18] T.T. Georgiou and M.C. Smith, Optimal Robustness in the Gap Metric, IEEE Trans. on Automatic Control, vol. 35, No. 6, pp. 673-686, 1990. that is, Xo and Wo is the Schmidt pair corresponding to the [19] Feintuch A., Saeks R. System Theory: A Hilbert Space Approach, Academic Press, N.Y., 1982. maximum singular value λ = kΞk, and satisfies the previous [20] Georgiou T., On the Computation of teh Gap Metric, Systems and Theorems. And the following operator identity gives Qo , Control Letters, vol. 11, pp. 253-257, 1988 . µ ¶ µ ¶ [21] Foias C., Georgiou T. and Smith M.C., Robust Stability of Feedback U M Systems: A Geometric Approach Using The Gap Metric, SIAM J. Xo + Qo Xo = ΞXo = kΞkWo V N Control and Optimization, vol. 31, No.6, pp. 1518-1537, 1993. ¡ ? ¢ ¡ ? ¢ [22] T.T. Georgiou and M.C. Smith, Robust Stabilization in the Gap Metric: ? ? ⇐⇒ Qo Xo = − M U + N V Xo + kΞk M , N Wo Controller Design for Distributed Plants, IEEE Trans. on Automatic Control, vol. 37, No. 8, pp. 1133-1143, 1992. which solution is guaranteed to exist by Theorem 6. [23] Dale W.N. and Smith M.C., Stabilizability and Existence of System Representation for Discrete-Time-Varying Systems, SIAM J. Control V. C ONCLUSION and Optimization, vol. 31, No.6, pp. 1538-1557, 1993. S.Q. Zhu, M.L.J. Hautus and C. Praagman, Sufficient Conditions for In this paper we considered the problem of robust stabi- [24] Robust BIBO Stabilisation: Given by the Gap Metric, Systems and lization of LTV systems in the gap metric and its connection Control Letters, vol. 11, pp. 53-59, 1988. with coprime factor uncertainty. We studied the problem of [25] M. Cantoni and G. Vinnicombe, Linear Feedback Systems and the Graph Topology, IEEE Trans. on Automatic Control, ol. 47, No. 5, computing the optimal controller and characterize the radius May 2002. of the maximal TV gap ball about the plant which can be [26] Foias C., Georgiou T. and Smith M.C., Geometric Techniques for Robust Stabilization of Linear Time-Varying Systems, Proc. of he 29th stabilized by a fixed controller. We introduced TV Hankel pp. 2868-2873, December 1990. and some related operators along with their singular values [27] CDC, Zhou K. and Doyle J.C., Essentials of Robust Control, Prentice Hall, and vectors that play a central role in optimizing the TV gap 1998. and in the computation of the robust stabilizing controller. The maximal stability margin under coprime factor uncertainty for LTV systems is characterized as the norm of these operators. Our results generalize similar results obtained for the case of LTI systems in [22], and in fact reduce to them in that case. R EFERENCES [1] G. Zames and A.K. El-Sakkary, Unstable Systems and Feedback: The Gap Metric, in Proc. of the Allerton Conference, pp. 380-385, 1980. [2] Arveson W. Interpolation problems in nest algebras, Journal of Functional Analysis, 4 (1975) 67-71. [3] Davidson K.R. Nest Algebras, Longman Scientific & Technical, UK, 1988. [4] Douglas R.G. Banach Algebra Techniques in Operator Theory, Academic Press, NY, 1972. [5] Luenberger D.G. Optimization by Vector Space Methods, John-Wiley, NY, 1968. [6] Feintuch A. Robust Control Theory in Hilbert Space, Springer-Verlag, vol. 130, 1998. [7] A. Feintuch, The Gap Metric for Time-Varying Systems, SYstems and Control Letters, v. 16, pp. 277-279, 1991. [8] A. Feintuch, Robustness for Time-Varying Systems, Math. Control Signals Systems (1993), 6:247-263.