Robust output feedbackstabilization via risksensitive control


 Wilfrid Holmes
 3 years ago
 Views:
Transcription
1 Automatica 38 22) Robust output feedbackstabilization via risksensitive control Valery A. Ugrinovskii, Ian R. Petersen School of Electrical Engineering, Australian Defence Force Academy, Canberra AC 26, Australia Received 2 February 2; received in revised form 24 October 2; accepted 2 November 2 Abstract We consider a problem of robust linear quadratic Gaussian LQG) control for discretetime stochastic uncertain systems with partial state measurements. For a nitehorizon case, the problem was recently introduced by Petersen et al. IEEE rans. Automat. Control 45 2) 398). In this paper, an innite horizon extension of the results of Petersen et al. IEEE rans. Automat. Control 45 2) 398) is discussed. We show that for a broad class of uncertain systems under consideration, a controller constructed in terms of the solution to a specially parameterized risksensitive stochastic control problem absolutely stabilizes the stochastic uncertain system.? 22 Elsevier Science Ltd. All rights reserved. Keywords: Robust control; LQG control; Stochastic control; Stochastic risksensitive control. Introduction For systems with partial state measurements, the linear quadratic Gaussian LQG) methodology has proved to be a useful technique for designing output feedbackcontrollers. However, the development of a robust version of the LQG technique has been a challenging problem. It is known that LQG controllers may lead to very poor robustness in terms of gain and phase robustness margins Doyle, 978). herefore, considerable eorts have recently been undertaken to develop a robust version of the LQG control synthesis methodology. For dierent uncertainty models, this problem has been attacked using the H and mixed H 2 =H control approach e.g., see Mustafa & Bernstein, 99; Haddad et al., 99; Basar & Bernhard, 995), the guaranteed cost and quadratic stabilization approach Fu, de Souza, & Xie, 99), the minimax optimization approach based on sum quadratic constraints Moheimani, Savkin, & Petersen, 995). It has recently been demonstrated in Petersen, James, and Dupuis 2a) that one possibility to enhance the his workwas ported by he Australian Research Council and he Defence Science and echnology Organization. his paper was not presented at any IFAC meeting. Corresponding author. el.: ; fax: addresses: V.A. Ugrinovskii), I.R. Petersen). robustness of an LQG controller is to use a risksensitive control approach to the controller synthesis. he risksensitive control approach has a number of attractive features. Apart from the enhanced robustness of a controller, this approach leads to a tractable minimax design procedure which can be viewed as a direct extension of the existing LQG control technique, see Petersen, Ugrinovskii, and Savkin 2b). Note that the results in Petersen et al. 2a) which concern the output feedbackrobust control of linear systems, are incomplete in that they address the nitehorizon control problem and lead to timevarying controllers, whereas in practical robust controller design problems, timeinvariant controllers are most useful. he derivation of timeinvariant output feedbackrobust controllers requires that an innitehorizon version of the partial information minimax optimal control problem of Petersen et al. 2a) be considered. Although a steadystate version of the minimax optimal controller was discussed by Petersen et al. 2a), the stabilizing properties of this controller were not addressed. In this paper, we rigorously address an innitehorizon robust control problem which can be regarded as an extension of the problems considered by Petersen et al. 2a) to the innite time horizon case. he contribution of this paper is to show that the resulting optimal control schemes guarantee the absolute stability of the closed loop system against the class of admissible uncertainty perturbations under consideration; see also Ugrinovskii and Petersen 2a), Petersen et al. 2b) and Dupuis et al. 998) in the continuoustime case. 598/2/$  see front matter? 22 Elsevier Science Ltd. All rights reserved. PII: S 598)2886
2 946 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) he main result of the paper is a robust LQG control synthesis procedure based on a pair of discretetime algebraic Riccati equations arising in a risksensitive optimal control; see Glover and Doyle 988). We show that solutions to a certain specially parameterized risksensitive control problem provide us with a controller which guarantees an optimal upper bound on the timeaveraged performance of the closedloop system in the presence of admissible uncertainties. his result is analogous to the corresponding continuoustime result of Ugrinovskii and Petersen 2a). 2. Denitions We adopt the stochastic uncertain system model introduced in Petersen et al. 2a). his model includes a nominal system model and also a description of a class of admissible uncertain perturbations. In this section, we present a modication of the model of Petersen et al. 2a) which is required in order to obtain an innitehorizon minimax LQG control result. he nominal system and measurement dynamics are described by the following equation: x t+ = Ax t + B u t + B 2 w t+ ; z t = C x t + D u t ; y t+ = C 2 x t + v t+ : ) In the above equations x t R n is the state, z t R p is the uncertainty output, and y t R q is the measured output. Also, w := {w t } t= ;v := {v t } t= are two sequences of i.i.d. Gaussian variables with zero mean and covariance matrices W ;, respectively. It is assumed that w t R r ;v t R q ;t=; 2;:::.WeletP w ;P v denote marginal probability measures of w t ;v t. he sequence w corresponds to the system noise, while the sequence v is referred to as the measurement noise. For simplicity, we pose that the reference system noise and the reference measurement noise are independent. All coecients in Eqs. ) are assumed to be constant matrices of corresponding dimensions. Also for the sake of simplicity, we assume that C D =. he initial state of system ) is a Gaussian random vector x R n with mean x and nonsingular covariance matrix Y. he corresponding Gaussian probability measure associated with x is denoted P x. he random variable x is assumed to be independent of the reference noise sequences w ;v. he joint probability measure on the initial condition vector x and the reference noise sequences w and v is hroughout the paper, t ; t will denote sequences of random variables as follows t := { t; t+ ;:::; }; t := { t; t+ ;:::}. denoted by P : P dx dw dv ) = P x dx ) P w dw t ) P v dv t ): t= his probability measure is dened on measurable sets of the ltration {F ;=; ;:::}. For each =; ;:::;the algebra F is generated by the initial condition x of system ) and the noise sequences w, v and is completed by including all sets of P probability zero. he sets of the algebra F are subsets of the noise space W, W = {x ;w ;v )}: From Kolmogorov s theorem, the family of probability measures P gives rise to a probability measure P dened on measurable space W; F); F = {F ;=; ;:::}. 2.. Stochastic uncertain system We now introduce an uncertainty model. he uncertainty will be modeled in terms of perturbations of corresponding joint probability distributions of the noise inputs w, v and the initial condition of the system. It is demonstrated in Section 2.3 that such a model provides a meaningful description of stochastic uncertain systems modeled using a linear fractional transformation LF); also, see Petersen et al. 2a,b) and Ugrinovskii and Petersen 999). he rigorous denition of our uncertainty model is the following. Let Q := {Q ;Q 2 ;:::} be a collection of conditional probability measures referred to as an uncertainty; i.e., for each t =; 2;:::; Q t dw t dv t x ;w t ;v t ) is a probability measure dened on measurable subsets of R r R q.itisassumed that Q t dw t dv t x ;w t ;v t ) Pdw t dv t x ;w t ;v t ); t =; 2;::: 2) Here, the notation means that the probability measure is absolutely continuous with respect to the reference probability measure. Using the collection Q, the joint probability measure of the initial condition vector x and the perturbed noise sequences w and v is dened by Q dx dw dv ) = P x dx ) t= Q t dw t dv t x ;w t ;v t ): t= It follows from Eq. 2) that Q P. Also, for any F measurable set W; Q )=Q ). As in Petersen et al. 2a), we will require that each uncertainty Q = {Q ;Q 2 ;:::} has the property hq P ) for all :
3 Here, hq P ) is the relative entropy functional; e.g., see Dupuis and Ellis 997). his requirement will allow us to rule out the possibility of singular uncertain probability measure perturbations. he class of uncertain systems considered in this paper comprises the systems of the form ) in which the marginal conditional probability measures of the noise inputs w, v are required to satisfy the above requirements. he set of such uncertainties Q will be denoted as Q. From the denition and properties of the relative entropy functional, it follows that the set Q is convex. V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) Relative entropy constraint on stochastic uncertainty LF uncertainty models require that a magnitude constraint be imposed on the uncertainty. For example, it is often required that the unmodeled dynamics transfer function has a bounded H norm. As shown in Ugrinovskii and Petersen 999, 2a) and Petersen et al. 2b), for stochastic LF systems with additive noise, the constraint on the H norm of the uncertainty transfer function can be replaced by a constraint on the relative entropy of the associated perturbation probability measure. For discrete time systems, this leads to an extension of the sum quadratic constraint uncertainty description Moheimani et al., 995); see Section 2.3. Let E Q denote the expectation with respect to the probability measure Q. Denition. Let d be a given positive constant. A collection of conditional probability measures Q Q is said to dene an admissible uncertainty if the following stochastic uncertainty constraint is satised: hq P ) 6 d + 2 EQ z s 2 + ) 3) for all =; 2;:::.In3);z t is the uncertainty output dened by Eq. ). Also; ) as. We denote the set of uncertainties Q Q satisfying condition 3) by. he corresponding probability measures Q are also called admissible probability measures. Observe that the set is not empty. Indeed, consider the reference probability measure P. Since the Gaussian random variables w t, v t are independent, the corresponding reference conditional probability measures are simply the marginal probability measures: P t dw t dv t x ;w t ;v t )=P w dw t ) P v dv t ): Also, the fact that hp P ) = and the condition d imply that the corresponding collection of conditional probabilities P := {P ;P 2 ;:::} is admissible, P. Note that in this case, constraint 3) is satised strictly. Hence, P is an interior point of the set. Fig.. An uncertain control system A connection between uncertainty input signals and uncertain probability measures In order to give a further insight into Denition, we show that the uncertainty class introduced in Section 2.2 includes uncertainty models which often arise in control systems such as H normbounded uncertainty and bounded exogenous uncertainty. Consider an uncertain system shown in Fig.. he system is described by the equations x t+ = Ax t + B u t + B 2 t z t = C x t + D u t ; y t+ = C 2 x t + t + ex t + w t+ ); + ex t +ṽ t+ ); 4) and is driven by the system and measurement noises w t, ṽ t, exogenous disturbance processes ex t, ex t and uncertainty inputs t, t which are generated by a stable linear timeinvariant uncertainty. he latter is due to the presence of unmodeled dynamics which are described by a stable transfer function z). It is assumed that! ; W =2 e j! ) =2 6 2 : 5) Condition 5) constitutes a H type norm bound on the size of unmodeled dynamics in system 4). System 4) is considered in a complete probability space W; F; Q); w,ṽ are Gaussian white noise processes in this probability space, E Q w t w t =W, E Qṽ t ṽ t =. It follows from 5) that the processes, are adapted to the ltration {F ;=; ;:::;} generated by the noise inputs
4 948 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) and W =2 t 2 + =2 t 2 ) 6 2 E Q z t 2 : E Q Also, it is assumed that the exogenous disturbances ex, ex are adapted to the ltration {F ;=; ;:::;} and E Q W =2 ex t 2 + =2 ex t 2 ) 6 d + ); 6) where d is a given constant, and ) as. Condition 6) imposes a bound on the size of uncertain exogenous perturbations in system 4). In particular, one can see the role of the constant d from Denition in this example. he introduction of this constant reects the fact that the power seminorm of the input disturbances does not exceed d. he above constraints on the exogenous disturbance and unmodeled dynamics lead to an uncertainty constraint which can be represented in the form of the relative entropy uncertainty constraint 3). Indeed, the above assumptions imply that the processes ex, ex t := t + ex t ; t := t + ex t satisfy the condition 2 E Q W =2 t 2 + =2 t 2 ) 6 d + 2 E Q z t 2 + ): 7) In order to express constraint 7) in the form of condition 3), we introduce the probability measure transformation dp = exp { d Q tw w t+ 2 } F tw t exp { t ṽ t+ 2 } t t : 8) Using the discretetime version of the Girsanov s theorem Elliott, Aggoun, & Moore, 994) we conclude that under the probability measure P the sequences w ;v dened by the equations w t+ = w t+ + t ; v t+ =ṽ t+ + t ; t =; ; 2;::: : 9) are the sequences of i.i.d. Gaussian random variables. Furthermore, when considered with respect to the probability measure P, system 4) becomes a system of the form ) driven by the i.i.d. Gaussian white noises w t ;v t. Finally, using the chain rule Dupuis & Ellis, 997), we obtain h Q P )= 2 E Q W =2 t 2 + =2 t 2) : Here, Q and P denote the restrictions of the probability measures Q and P to W; F ). hus, from 7), condition 3) of Denition 2 follows. herefore, the uncertain system considered in this section belongs to the class of uncertain systems dened in Denition. It can be shown in a similar fashion that Denition encompasses some other important classes of uncertainty arising in control systems such as, for example, the conebounded uncertainty Absolutely stabilizing control In this paper, our attention will be restricted to linear outputfeedbackcontrollers of the form ˆx t+ = A c ˆx t + B c y t+ ; u t = C c ˆx t ; ) where ˆx R ˆn is the state of the controller, A c R ˆn ˆn, B c R ˆn q, and C c R m ˆn. Let U denote this class of linear controllers. Note that controller ) is adapted to the ltration {Y t ;t } generated by the observation process y; Y t = {y s ;s=;:::;t}. he closedloop system corresponding to system ) and controller ) is described by a linear dierence equation of the form x t+ = A x t + B w t+ ; z t = C x t ; u t = C c x t : ) In Eq. ), x =x ˆx R n+ˆn and w t =w t v t are the state and the noise input of the closed loop system. Also, the following notation is used: A B C c B2 A = ; B = ; B c C 2 A c B c C =C D C c : 2) In the sequel, we will consider a subclass of controllers ) which satisfy some additional stabilizability and observability requirements. We say that a controller K belongs to the class U U if for this controller, the corresponding matrix pair A; B) is stabilizable 2 and the pair A c ;C c )is observable. It was shown in Petersen et al. 2a) that in the steadystate case, the minimax optimal LQG controller has the structure given in equation ). As mentioned above, we wish to investigate the stabilizing properties of this optimal controller as. First, we introduce a denition 2 For example, this condition holds if the matrix pairs A; B 2 ) and A c;b c) are stabilizable.
5 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) of stabilizability which allows for uncertain control systems driven by additive noise whose solutions are not square summable. Denition 2. A controller K of the form ) is said to be an absolutely stabilizing controller for the uncertain system ) with uncertainty satisfying the relative entropy constraint 3); if the system state process x ; the controller state process ˆx and the control process u dened by the closedloop system ) corresponding to this controller satisfy the following condition. here exist constants c ; c 2 such that for any admissible uncertainty Q ; x t 2 + ˆx t 2 + u t 2 ) E Q +hq P ) 6 c + c 2 d: 3) In particular, for the uncertain system shown in Fig. above property means that in a stable closedloop system, timeaverage norms of the system state, controller state as well as control input and uncertainty inputs are bounded regardless the value of the transfer function z) and disturbances ex, ex. In the special case of the uncontrolled uncertain system ), 3), the above denition reduces to the property of absolute stability the system. hat is, the uncertain system ), 3) with u t = is absolutely stable if for any admissible uncertainty Q, x t 2 + hq P ) 6 c + c 2 d: E Q 4) Lemma. Suppose the stochastic nominal closedloop system ) is mean square stable; i.e. E x t 2 : 5) Also; pose the pair A; B) is stabilizable. hen; the matrix A must be stable. he proof of this lemma is identical to the proof of the corresponding continuoustime result in Ugrinovskii and Petersen 2a,b) and Petersen et al. 2b). 3. he main results he steadystate minimax LQG control problem considered in Petersen et al. 2a) was to nd a controller which attained an upper value inf J K; Q) 6) K Q of the cost functional J K; Q) = 2 EQ Fx t ;u t ); Fx; u):=x Rx + u Gu; 7) R and G are symmetric positive denite matrices, R R n n, G R m m. Here, x t denotes the solution to system ) corresponding to a given controller K and an admissible uncertainty Q. he minimax optimal controller for problem ), 3), 7) proposed in Petersen et al. 2a) was constructed using a pair of parameter dependent algebraic Riccati equations X = A X + B G + D D ) B B 2WB 2) A + R + C C ; 8) Y = A Y + C 2 C 2 ) R C C A + B 2 WB 2: 9) Here, is a positive constant which was chosen as follows. For each such that the Riccati equations 8), 9) admit positive denite stabilizing solutions satisfying the conditions Y + C 2 C 2 R C C ; X B 2WB 2 ; X Y 2) dene the quantity V = det 2 log I ) Y R + C C ) det 2 log I X I ) ) Y X ; 2) where := K + C 2 I ) ) Y R + C C ) Y C K ; K := A Y + C 2 C 2 ) R C C C 2 : 22) he set of the parameters satisfying the above requirement is denoted. In order to obtain the minimax optimal LQG controller, the parameter is chosen to achieve the inmum in inf V + d): 23) Using this optimal value of, the corresponding minimax optimal steadystate controller K is dened by the
6 95 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) equations u t = L ˆx t ; 24) L := G + D D ) B X + B G + D D ) B B 2WB 2 A I ) Y X ; where ˆx t was generated by the lter ˆx t+ = A ˆx t + B u t + Ky t+ C 2 ˆx t ) + A Y + C 2 C 2 ) R C C R + C C )ˆx t ; ) ˆx =x : 25) he above solution to the minimax optimal LQG control problem was obtained in Petersen et al. 2a) by letting in the corresponding nite horizon problem. However, as we let approach, two important issues arise which were not addressed in Petersen et al. 2a). ) he ability of an optimal controller to stabilize the system is an important issue in optimal control on an in nite time interval. In our case, it is desired that the resulting minimax optimal controller stabilizes the system ) for all admissible uncertainties Q. 2) It was shown in Petersen et al. 2a) that on a nite interval, the minimax optimal LQG controller exists if and only if associated parameter dependent dierence Riccati equations have positive denite solutions. In the innite horizon case, the design methodology outlined in Petersen et al. 2a) provides only a sucient condition for the minimax optimal controller to exist. However, it is often useful to know that the only if claim holds as well. Necessary conditions usually give a good indication as to how conservative the proposed controller is. hese issues are addressed in the statements given below. heorem shows that the existence of stabilizing solutions to the algebraic Riccati equations 8) and 9) is a sucient as well as a necessary condition for the innitehorizon minimax optimal LQG control problem 26) to have a solution. Furthermore, heorem 2 shows that the minimax LQG controller proposed in Petersen et al. 2a) is an absolutely stabilizing controller. heorem. i) If the set is nonempty, then the minimax optimal control problem inf J K; Q) 26) K U Q has a nite value. ii) Conversely, if there exists an absolutely stabilizing controller K U which attains the inmum in 26); then the set is not empty. he innitehorizon version of the minimax optimal LQG control result of Petersen et al. 2a) now follows from heorem. heorem 2. Suppose that and attains the inmum in 23). hen; the corresponding controller K = K dened by 24); 25); guarantees that the worst case of the cost functional 7) does not exceed the value 23). Furthermore; this controller is an absolutely stabilizing controller for stochastic uncertain system ) satisfying relative entropy constraint 3). Remark. Note that the controller resulting from heorem 2 is the steadystate limit of the nite horizon minimax optimal controller proposed in the reference Petersen et al. 2a). Also; the worstcase performance 23) is the steadystate limit of the performance guaranteed by the nitehorizon minimax optimal controller. he proofs of the above theorems will be given in Sections 3.3 and Preliminary remarks he proof of heorem relies on a duality relationship between free energy and relative entropy established in Dupuis and Ellis 997) and Dai Pra, Meneghini and Runggaldier 996). Associated with system ), consider the parameter dependent risksensitive cost functional I ; K)= {exp log E 2 } F x t ;u t ) ; 27) where is a given constant, K is a controller dened by Eq. ). Also, F x; u):=x R + C C )x + u G + D D )u: 28) When applied to system ) and the risksensitive cost functional 27), the duality result states that for each admissible controller K, I ; K) = Q : hq P ) 2 EQ F x t ;u t ) : hq P ) 29) he use of the above duality result is a key step in this paper in that it enables us to replace a guaranteed cost control problem by the following risksensitive optimal control problem:
7 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) Find an output feedbackcontroller of form ) which attains the inmum in inf lim I ;K): 3) K In this paper, we use the solution to this risksensitive control problem presented in Whittle 99). Note that similar results concerning problem 3) were obtained in Iglesias, Mustafa, and Glover 99) and Glover and Doyle 988) 3 where the problem was shown to be equivalent to a problem of constructing an output feedbackcontroller which maximizes the entropy functional I d K;)= log deti H e j! )H e j! )) d! 2 3) over the set of stabilizing output feedbackcontrollers satisfying the following bound: H := H e j! )H e j! ) : 32)! ; Here, H s) is the transfer function of the closed loop system dened by the equations x t+ = Ax t + B u t +B 2 W =2 t+ ; 33) R =2 z t = x t + G =2 u t; C D y t+ = C 2 x t + =2 t+ ; and a stabilizing linear controller u t =Ky t ). In particular, for a linear controller of the form ), the transfer function H s) is given by R =2 H s) := G =2 C c si A) C D C c B2 W =2 : 34) B c =2 It follows from the results of Iglesias et al. 99), Glover and Doyle 988) and Whittle 99) that the solution to the above maximum entropy problem and equivalently, the solution to the risksensitive control problem 3) is given by the central solution to the corresponding H control problem; see Basar and Bernhard 995). 4 his controller is dened by Eqs. 24) and 25) and involves positive denite solutions 3 Iglesias et al. 99), Glover and Doyle 988) consider a slightly dierent information pattern which allows the control u t to instantaneously access y t+. his leads to an optimal controller which is not strictly causal. 4 Also, the links between the problem considered in this paper and H control were discussed in Petersen et al. 2a). to the Riccati equations 8) and 9). Furthermore, the minimum risksensitive cost corresponding to this controller is given by Eq. 2) Absolutely stabilizing properties of risksensitive control In this section, we establish the absolutely stabilizing properties of the risksensitive controller. Lemma 2. Let K be a controller which guarantees a nite risksensitive cost: V K := lim I ;K) : hen; the controller K is an absolutely stabilizing controller for the stochastic uncertain system ) satisfying the relative entropy constraint 3). Furthermore; J K; Q) 6 V K + d: 35) Q Proof. We wish to prove that the controller K satises condition 3) of Denition 2. Consider an uncertain system with an admissible uncertainty Q governed by the controller K. Recall that the system is described by Eq. ) where the control process u is generated by the given controller K. Also, the joint probability distribution of the noise inputs w, v and the initial condition x is dened by the given collection of conditional probability measures Q. Note that since the uncertainty Q is admissible, then the corresponding probability measures Q have the property hq P ) for every =; 2;:::. his allows us to apply duality condition 29). From this equation, we obtain + { 2 EQ 2 EQ Fx s ;u s ) } z s 2 hq P ) 6 V K : 36) Furthermore since Q, satisfaction of Eq. 35) follows from 36) and 3). We now show that the controller K is absolutely stabilizing. Since the matrices R and G are positive denite, then inequality 35) implies EQ x s 2 + u s 2 ) 6 V K + d); 37) where is a constant which depends only on R and G. Next, we show that there exist constants c ;c 2 such that hq P ) c + c 2 d: 38)
8 952 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) o this end, we note that for any Q, t t hqt P t ) 6 d + ˆ + t 2t EQ z s 2 : 39) t t Here, ˆ := t t) as. Inequality 39) takes into account the fact that the probability measure Q t satises the relative entropy constraint 3). herefore, inequalities 37) and 39) imply hq P ) 6 d + 6 c + c 2 d; 2 EQ z s 2 where the constants c, c 2 are dened by V K,,, C, D, and hence independent of Q. o complete the proof, it remains to prove that the process ˆx satises condition 3) of Denition 2. Indeed, consider the uncertain closed loop system ) corresponding to the controller K and an admissible uncertainty Q. Using the duality relation between free energy and relative entropy established in Dupuis and Ellis 997) and Dai Pra et al. 996), we obtain Q 2 EQ x t 2 hq P ) 6 Q :hq P ) 2 EQ x t 2 hq P ) { } =log E exp x t 2 : 2 From this equation, it follows that for any Q, 2 EQ x t 2 6 hq P ) { } + lim log E exp x t 2 : 4) 2 Note that the process x on the lefthand side of Eq. 4) corresponds to the uncertain system ) while the process x that appears in the second term on the righthand side of Eq. 4) is generated by the nominal closed loop system ). Since this nominal system is driven by the Gaussian process w, one can choose suciently small such that the second term on the righthand side of Eq. 4) is nite; e.g., see Glover and Doyle 988) and Whittle 99). his term is independent of Q. Since we have already proved 38), then condition 3) follows. Remark 2. It follows from Lemma 2 that the controller solving the risksensitive control problem 3) is an absolutely stabilizing controller for the uncertain system under consideration. Lemma 3. Suppose 5 that for any K U J K; Q)= : 4) Q Q If there exists an absolutely stabilizing controller K U such that J K; Q) c ; 42) Q then there exists a such that V. Proof. Since the given controller K U absolutely stabilizes the uncertain system ); 3); then condition 3) of Denition 2 is satised. Note that it follows from 3) and Lemma that the matrix A corresponding to the controller K is stable. Also; condition 3) implies that there exists a positive constants c ; such that for all Q ; lim inf 2 EQ + lim inf Fx s ;u s ) 2 EQ x s 2 + ˆx s 2 ) 6 c: 43) Consider the functionals G Q) := c lim inf lim inf G Q):= d lim inf 2 EQ 2 EQ Fx s ;u s ) x s 2 + ˆx s 2 ); 2 EQ z s 2 hq P ) : 44) It is readily proved that satisfaction of condition 43) implies that the following condition is satised: If G Q) 6 then G Q) : 45) he proof of this fact follows along the same lines as the proof of the corresponding fact in Ugrinovskii and Petersen 2a); see also Petersen et al. 2b). Furthermore; the set of uncertainties satisfying the condition G Q) 6 has an interior point; see the remarkfollowing Denition. Also; it follows from the properties of the relative entropy functional that the functionals G ) and G ) are convex. 5 Condition 4) is similar to a corresponding condition in Petersen et al. 2a). his condition was introduced in Petersen et al. 2a) to rule out the case = in the proof of heorem 3:. In the continuoustime case, this condition holds if the pair A; B) corresponding to the given controller K U is controllable; e.g., see Ugrinovskii and Petersen 2a) and Petersen et al. 2b). Obviously, in the discrete time case, condition 4) can be replaced by a similar controllability condition.
9 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) We have now veried all of the conditions needed to apply the Lagrange multiplier result e.g.; see Luenberger; 969). Indeed; heorem on p. 27 of Luenberger 969) implies that there exists a constant such that lim inf + EQ lim inf + lim inf 2 Fx t;u t ) 2 EQ 2 EQ x t 2 + ˆx t 2 ) z t 2 hq P ) 6 c d: 46) for all Q Q. Also; condition 4) rules out the possibility that =. hus;. Condition 46) implies the satisfaction of condition 32) for the transfer function H s) corresponding to the system 33) and the given controller K. his claim can be established using the same arguments as those used in proving the corresponding fact in Ugrinovskii and Petersen 2b) and Petersen et al. 2b). For the sake of brevity, we only outline the proof. Consider an augmented version of the closed loop system corresponding to system 33) and the given controller K: B2 x t+ = W =2 A x t + t ; 47) B c =2 R =2 G =2 C c z ;t = C D C c x t : I I his system is posed to be driven by a deterministic input disturbance which has a nite autocorrelation matrix and also has a power spectral density function Makila, Partington, & Norlander 998; Ljung, 987). We denote the set of such signals by P +. he transfer function of system 47) is denoted H ; z). We will establish condition 32) by showing that H ; 6 : 48) he proof is by establishing a contradiction. First, we observe that the failure of condition 48) to hold must lead to the existence of a sequence of deterministic inputs { N ;N =; 2;:::} P+ such that z ;t 2 t N 2 ) N 49) 2 for all ; N), where is a suciently small constant. hen, we show that the satisfaction of condition 49) must lead to a contradiction with 46). Indeed, for each input N, a collection of conditional probability measures Q N = {Q N ;QN 2 ;:::} Q can be dened using a corresponding probability measure transformation; see Section 2.3. o this end, the input N has to be partitioned as follows: N t W =2 = N =2 t : N t Also as in Section 2.3, the probability measures Q N; are dened, hq N; P )= 2 W =2 N t 2 + =2 N t 2 : 5) Next, it is shown using 49) that the closed loop system ) driven by the deterministic inputs N, N and considered on the probability space W; F ;Q N; ), satises the following condition: N; M N := lim 2 EQ Fx t;u t )+ x t 2 Q Q + z t 2 W =2 N t 2 =2 N t 2 N: 5) Letting N in Eq. 5) and using Eq. 5), we obtain the following contradiction with 46): lim inf EQ 2 Fx t;u t ) + lim inf + lim inf 2 EQ 2 EQ x t 2 z t 2 hq P ) M N = : N he above contradiction shows that condition 48) must hold. Since condition 48) implies 32), the lemma now follows from condition 32) and the results of Glover and Doyle 988) and Whittle 99) Proof of heorem Part i) of the theorem follows from Lemma 2. Indeed, if, then the corresponding Riccati equations 8) and
10 954 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) ) have stabilizing solutions. his ensures that the corresponding controller 24) solves the maximum entropy control problem and equivalently, the risksensitive control problem 3). he corresponding value of the risksensitive cost in 3) is therefore nite. hus, controller 24) corresponding to also satises the conditions of Lemma 2. From this lemma, part i) of heorem follows. Part ii) follows from Lemma Proof of heorem = = = It has been observed that for each, controller 24), 25) solves the corresponding risksensitive optimal control problem. Hence from Lemma 2, this controller is an absolutely stabilizing controller. Also from Eq. 35) in Lemma 2, it follows that the risksensitive controller corresponding to guarantees that the cost is not greater than the value dened by Eq. 23). 4. Illustrative example o illustrate our theory, we consider an example discussed in Petersen et al. 2a). In this example, an uncertain system was considered which had the structure shown in Fig.. he system is dened by the equations :52 :5 :53 x t+ = :52 x t + :52 u t + :53 :52 t + w t+ ); x 4 Fig. 2. he graph of xt 2 versus. equations 8), 9) is given by the transfer function 76:85z +78:3 Hz)= z 2 53) +:48z +:386 which corresponds to =:8. o verify stabilizing properties of this controller, the closed loop system 52), 53) was simulated with dierent values of and the graphs of x t 2 ; u t 2 ; t 2 were plotted. It was observed that the above quantities showed a tendency to remain bounded over the time of simulation. One of these graphs is shown in Fig. 2. z t =:5u t ; y t+ = x t + t +ṽ t+ ): 52) he reference noise signals w t, v t are assumed to be Gaussian white noise signals with covariances W =, = 4, respectively. he uncertainty is assumed to satisfy the stochastic sum constraint 7) with d = 8. Also in Petersen et al. 2a), cost functional 7) was considered in which R = ; G = 4 : For this system, the robust LQG controller designed in Petersen et al. 2a) using the steadystate Riccati References Basar,., & Bernhard, P. 995). H optimal control and related minimax design problems: a dynamic game approach. 2nd edition. Boston, Birkhauser. Dai Pra, P., Meneghini, L., & Runggaldier, W. 996). Connections between stochastic control and dynamic games. Mathematics of Control, Systems and Signals, 94), Doyle, J. C. 978). Guaranteed margins for LQG regulators. IEEE ransactions on Automatic Control, 23. Dupuis, P., & Ellis, R. 997). A Weak Convergence Approach to the heory of Large Deviations. New York: Wiley. Dupuis, P., James, M. R., & Petersen, I. R. 998). Robust properties of risksensitive control. In Proceedings of the IEEE Conference on Decision and Control, Vol. 2. pp ), ampa, FL, December. Elliott, R. J., Aggoun, L., & Moore, J. B. 994). Hidden Markov models. Estimation and Control. New York: Springer.
11 V.A. Ugrinovskii, I.R. Petersen / Automatica 38 22) Fu, M., de Souza, C. E., & Xie, L. 99). Quadratic stabilization and H control of discretetime uncertain systems. In Proceedings of the International Symposium on Mathematical heory of Networks and Systems, Kobe, Japan. Glover, K., & Doyle, J. C. 988). Statespace formulae for all stabilizing controllers that satisfy an H norm bound and relations to risksensitivity. Systems & Control Letters,, Haddad, W. H., Bernstein, D. S., & Mustafa, D. 99). Mixednorm H 2 =H regulation and estimation: he discretetime case. Systems & Control Letters, 64), Iglesias, P. A., Mustafa, D., & Glover, K. 99). Discrete time H controllers satisfying a minimum entropy criterion. Systems & Control Letters, 4, Ljung, L. 987). System Identication:heory for the User. EnglewoodCli, NJ: PrenticeHall. Luenberger, D. G. 969). Optimization by Vector Space Methods. New York: Wiley. Makila, P. M., Partington, J. R., & Norlander,. 998). Bounded power signal spaces for robust control and modeling. SIAM J. Control Opt., 37), Moheimani, S. O. R., Savkin, A. V., & Petersen, I. R. 995). A connection between H control and the absolute stabilizability of discretetime uncertain systems. Automatica, 38), Mustafa, D., & Bernstein, D. S. 99). LQG bounds in discretetime H 2 =H control. ransactions of the Institute of Measurement and Control, 3, Petersen, I. R., James, M. R., & Dupuis, P. 2a). Minimax optimal control of stochastic uncertain systems with relative entropy constraints. IEEE ransactions on Automatic Control, 45, Petersen, I. R., Ugrinovskii, V., & Savkin, A. V. 2b). Robust Control Design using H Methods. Berlin: Springer. Ugrinovskii, V. A., & Petersen, I. A. 999). Finite horizon minimax optimal control of stochastic partially observed time varying uncertain systems. Mathematics of Control, Signals and Systems, 2), 23. Ugrinovskii, V. A., & Petersen, I. R. 2a). Minimax LQG control of stochastic partially observed uncertain systems. SIAM J. Control Optim., 44), Ugrinovskii, V. A., & Petersen, I. R. 2b). Robust stability and performance of stochastic uncertain systems on an innite time interval. Systems & Control Letters, 444), Whittle, P. 99). Risksensitive optimal control. Chichester, UK: Wiley. Ian R. Petersen was born in Victoria, Australia in 956. He received the Bachelor of Engineering Electrical) degree from the University of Melbourne in 979. He received a Master of Science degree in 984 and a Ph.D in Electrical Engineering in 984 both from the University of Rochester. From 983 to 985 he was a Postdoctoral Fellow in the Department of Systems Engineering, Australian National University. In 985 he was appointed as a Lecturer in the Department of Electrical Engineering, Australian Defence Force Academy and he is currently a Professor in this department. In 989, he was a visitor in the Department of Engineering, Cambridge University. He served as an Associated Editor for the IEEE ransactions on Automatic Control and Systems and Control Letters. Currently he is an Associate Editor for Automatica and SIAM Journal on Control and Optimization. His main research interests are in robust control theory, H control, robust ltering, and optimal control theory. Valery A. Ugrinovskii was born in Ukraine in 96. He received the M.Sc. degree in Applied Mathematics in 982 and a Ph.D in Physics and Mathematics in 99 both from the State University of Nizhny Novgorod, Russia. From 982 to 995 he held research positions with the Radiophysical Research Institute, Nizhny Novgorod. From 995 to 996 he was a Postdoctoral Fellow at the University of Haifa. Since 996, he has been a Research Associate and Lecturer in the School of Electrical Engineering, Australian Defence Force Academy, Canberra. He is the coauthor of the research monograph Robust Control Design using H Methods, Springer, London, 2, with Ian R. Petersen and Andrey V. Savkin. His current research interests include stochastic control theory, robust control theory and H control.
CONTROL SYSTEMS, ROBOTICS, AND AUTOMATION  Vol. V  Relations Between Time Domain and Frequency Domain Prediction Error Methods  Tomas McKelvey
COTROL SYSTEMS, ROBOTICS, AD AUTOMATIO  Vol. V  Relations Between Time Domain and Frequency Domain RELATIOS BETWEE TIME DOMAI AD FREQUECY DOMAI PREDICTIO ERROR METHODS Tomas McKelvey Signal Processing,
More informationMixedÀ¾ Ð½Optimization Problem via Lagrange Multiplier Theory
MixedÀ¾ Ð½Optimization Problem via Lagrange Multiplier Theory Jun WuÝ, Sheng ChenÞand Jian ChuÝ ÝNational Laboratory of Industrial Control Technology Institute of Advanced Process Control Zhejiang University,
More informationDuality of linear conic problems
Duality of linear conic problems Alexander Shapiro and Arkadi Nemirovski Abstract It is well known that the optimal values of a linear programming problem and its dual are equal to each other if at least
More information2.3 Convex Constrained Optimization Problems
42 CHAPTER 2. FUNDAMENTAL CONCEPTS IN CONVEX OPTIMIZATION Theorem 15 Let f : R n R and h : R R. Consider g(x) = h(f(x)) for all x R n. The function g is convex if either of the following two conditions
More informationAn interval linear programming contractor
An interval linear programming contractor Introduction Milan Hladík Abstract. We consider linear programming with interval data. One of the most challenging problems in this topic is to determine or tight
More informationLinearQuadratic Optimal Controller 10.3 Optimal Linear Control Systems
LinearQuadratic Optimal Controller 10.3 Optimal Linear Control Systems In Chapters 8 and 9 of this book we have designed dynamic controllers such that the closedloop systems display the desired transient
More informationChapter 15 Introduction to Linear Programming
Chapter 15 Introduction to Linear Programming An Introduction to Optimization Spring, 2014 WeiTa Chu 1 Brief History of Linear Programming The goal of linear programming is to determine the values of
More information5. Convergence of sequences of random variables
5. Convergence of sequences of random variables Throughout this chapter we assume that {X, X 2,...} is a sequence of r.v. and X is a r.v., and all of them are defined on the same probability space (Ω,
More informationLoad Balancing and Switch Scheduling
EE384Y Project Final Report Load Balancing and Switch Scheduling Xiangheng Liu Department of Electrical Engineering Stanford University, Stanford CA 94305 Email: liuxh@systems.stanford.edu Abstract Load
More informationContinued Fractions and the Euclidean Algorithm
Continued Fractions and the Euclidean Algorithm Lecture notes prepared for MATH 326, Spring 997 Department of Mathematics and Statistics University at Albany William F Hammond Table of Contents Introduction
More informationSolving nonlinear equations in one variable
Chapter Solving nonlinear equations in one variable Contents.1 Bisection method............................. 7. Fixed point iteration........................... 8.3 NewtonRaphson method........................
More informationLecture 7: Finding Lyapunov Functions 1
Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science 6.243j (Fall 2003): DYNAMICS OF NONLINEAR SYSTEMS by A. Megretski Lecture 7: Finding Lyapunov Functions 1
More informationThe Heat Equation. Lectures INF2320 p. 1/88
The Heat Equation Lectures INF232 p. 1/88 Lectures INF232 p. 2/88 The Heat Equation We study the heat equation: u t = u xx for x (,1), t >, (1) u(,t) = u(1,t) = for t >, (2) u(x,) = f(x) for x (,1), (3)
More informationAnalysis of a Production/Inventory System with Multiple Retailers
Analysis of a Production/Inventory System with Multiple Retailers Ann M. Noblesse 1, Robert N. Boute 1,2, Marc R. Lambrecht 1, Benny Van Houdt 3 1 Research Center for Operations Management, University
More informationA NOTE ON AN LQG REGULATOR WITH MARKOVIAN SWITCHING AND PATHWISE AVERAGE COST*
A NOTE ON AN LQG REGULATOR WITH MARKOVIAN SWITCHING AND PATHWISE AVERAGE COST* Mrinal K. Ghosh, Aristotle Arapostathis and Steven I. Marcus Abstract. We study a linear system with a Markovian switching
More informationFurther Study on Strong Lagrangian Duality Property for Invex Programs via Penalty Functions 1
Further Study on Strong Lagrangian Duality Property for Invex Programs via Penalty Functions 1 J. Zhang Institute of Applied Mathematics, Chongqing University of Posts and Telecommunications, Chongqing
More informationOn closedform solutions of a resource allocation problem in parallel funding of R&D projects
Operations Research Letters 27 (2000) 229 234 www.elsevier.com/locate/dsw On closedform solutions of a resource allocation problem in parallel funding of R&D proects Ulku Gurler, Mustafa. C. Pnar, Mohamed
More informationInternational Journal of Information Technology, Modeling and Computing (IJITMC) Vol.1, No.3,August 2013
FACTORING CRYPTOSYSTEM MODULI WHEN THE COFACTORS DIFFERENCE IS BOUNDED Omar Akchiche 1 and Omar Khadir 2 1,2 Laboratory of Mathematics, Cryptography and Mechanics, Fstm, University of Hassan II MohammediaCasablanca,
More informationTopic 1: Matrices and Systems of Linear Equations.
Topic 1: Matrices and Systems of Linear Equations Let us start with a review of some linear algebra concepts we have already learned, such as matrices, determinants, etc Also, we shall review the method
More information15.062 Data Mining: Algorithms and Applications Matrix Math Review
.6 Data Mining: Algorithms and Applications Matrix Math Review The purpose of this document is to give a brief review of selected linear algebra concepts that will be useful for the course and to develop
More informationMath 313 Lecture #10 2.2: The Inverse of a Matrix
Math 1 Lecture #10 2.2: The Inverse of a Matrix Matrix algebra provides tools for creating many useful formulas just like real number algebra does. For example, a real number a is invertible if there is
More informationOutput feedback stabilization of the angular velocity of a rigid body
Systems & Control Letters 36 (1999) 181 192 Output feedback stabilization of the angular velocity of a rigid body A. Astol Department of Electrical and Electronic Engineering and Centre for Process Systems
More informationLOOP TRANSFER RECOVERY FOR SAMPLEDDATA SYSTEMS 1
LOOP TRANSFER RECOVERY FOR SAMPLEDDATA SYSTEMS 1 Henrik Niemann Jakob Stoustrup Mike Lind Rank Bahram Shafai Dept. of Automation, Technical University of Denmark, Building 326, DK2800 Lyngby, Denmark
More informationInformation Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay
Information Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay Lecture  17 ShannonFanoElias Coding and Introduction to Arithmetic Coding
More informationStochastic Linear Control Over a Communication Channel
Stochastic Linear Control Over a Communication Channel Sekhar atikonda Yale University New Haven, C 06520 sekhar.tatikonda@yale.edu 203432474 Anant Sahai University of California, Berkeley Berkeley,
More informationCommunication on the Grassmann Manifold: A Geometric Approach to the Noncoherent MultipleAntenna Channel
IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 48, NO. 2, FEBRUARY 2002 359 Communication on the Grassmann Manifold: A Geometric Approach to the Noncoherent MultipleAntenna Channel Lizhong Zheng, Student
More informationIterative Techniques in Matrix Algebra. Jacobi & GaussSeidel Iterative Techniques II
Iterative Techniques in Matrix Algebra Jacobi & GaussSeidel Iterative Techniques II Numerical Analysis (9th Edition) R L Burden & J D Faires Beamer Presentation Slides prepared by John Carroll Dublin
More information10. Proximal point method
L. Vandenberghe EE236C Spring 201314) 10. Proximal point method proximal point method augmented Lagrangian method MoreauYosida smoothing 101 Proximal point method a conceptual algorithm for minimizing
More informationDiscussion on the paper Hypotheses testing by convex optimization by A. Goldenschluger, A. Juditsky and A. Nemirovski.
Discussion on the paper Hypotheses testing by convex optimization by A. Goldenschluger, A. Juditsky and A. Nemirovski. Fabienne Comte, Celine Duval, Valentine GenonCatalot To cite this version: Fabienne
More informationTHE NUMBER OF GRAPHS AND A RANDOM GRAPH WITH A GIVEN DEGREE SEQUENCE. Alexander Barvinok
THE NUMBER OF GRAPHS AND A RANDOM GRAPH WITH A GIVEN DEGREE SEQUENCE Alexer Barvinok Papers are available at http://www.math.lsa.umich.edu/ barvinok/papers.html This is a joint work with J.A. Hartigan
More informationCOMPLETE MARKETS DO NOT ALLOW FREE CASH FLOW STREAMS
COMPLETE MARKETS DO NOT ALLOW FREE CASH FLOW STREAMS NICOLE BÄUERLE AND STEFANIE GRETHER Abstract. In this short note we prove a conjecture posed in Cui et al. 2012): Dynamic meanvariance problems in
More information1 Limiting distribution for a Markov chain
Copyright c 2009 by Karl Sigman Limiting distribution for a Markov chain In these Lecture Notes, we shall study the limiting behavior of Markov chains as time n In particular, under suitable easytocheck
More informationThe Method of Least Squares
The Method of Least Squares Steven J. Miller Mathematics Department Brown University Providence, RI 0292 Abstract The Method of Least Squares is a procedure to determine the best fit line to data; the
More informationTD(0) Leads to Better Policies than Approximate Value Iteration
TD(0) Leads to Better Policies than Approximate Value Iteration Benjamin Van Roy Management Science and Engineering and Electrical Engineering Stanford University Stanford, CA 94305 bvr@stanford.edu Abstract
More informationOptimal Investment with Derivative Securities
Noname manuscript No. (will be inserted by the editor) Optimal Investment with Derivative Securities Aytaç İlhan 1, Mattias Jonsson 2, Ronnie Sircar 3 1 Mathematical Institute, University of Oxford, Oxford,
More informationAdaptive Online Gradient Descent
Adaptive Online Gradient Descent Peter L Bartlett Division of Computer Science Department of Statistics UC Berkeley Berkeley, CA 94709 bartlett@csberkeleyedu Elad Hazan IBM Almaden Research Center 650
More informationOPTIMIZATION AND OPERATIONS RESEARCH Vol. IV  Markov Decision Processes  Ulrich Rieder
MARKOV DECISIO PROCESSES Ulrich Rieder University of Ulm, Germany Keywords: Markov decision problem, stochastic dynamic program, total reward criteria, average reward, optimal policy, optimality equation,
More informationarxiv:1112.0829v1 [math.pr] 5 Dec 2011
How Not to Win a Million Dollars: A Counterexample to a Conjecture of L. Breiman Thomas P. Hayes arxiv:1112.0829v1 [math.pr] 5 Dec 2011 Abstract Consider a gambling game in which we are allowed to repeatedly
More information( % . This matrix consists of $ 4 5 " 5' the coefficients of the variables as they appear in the original system. The augmented 3 " 2 2 # 2 " 3 4&
Matrices define matrix We will use matrices to help us solve systems of equations. A matrix is a rectangular array of numbers enclosed in parentheses or brackets. In linear algebra, matrices are important
More informationECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE
ECON20310 LECTURE SYNOPSIS REAL BUSINESS CYCLE YUAN TIAN This synopsis is designed merely for keep a record of the materials covered in lectures. Please refer to your own lecture notes for all proofs.
More informationExample 4.1 (nonlinear pendulum dynamics with friction) Figure 4.1: Pendulum. asin. k, a, and b. We study stability of the origin x
Lecture 4. LaSalle s Invariance Principle We begin with a motivating eample. Eample 4.1 (nonlinear pendulum dynamics with friction) Figure 4.1: Pendulum Dynamics of a pendulum with friction can be written
More informationPolePlacement Design A StateSpace Approach
TU Berlin DiscreteTime Control Systems 1 PolePlacement Design A StateSpace Approach Overview ControlSystem Design Regulation by State Feedback Observers Output Feedback The Servo Problem TU Berlin
More informationVector and Matrix Norms
Chapter 1 Vector and Matrix Norms 11 Vector Spaces Let F be a field (such as the real numbers, R, or complex numbers, C) with elements called scalars A Vector Space, V, over the field F is a nonempty
More informationAbsolute Value Programming
Computational Optimization and Aplications,, 1 11 (2006) c 2006 Springer Verlag, Boston. Manufactured in The Netherlands. Absolute Value Programming O. L. MANGASARIAN olvi@cs.wisc.edu Computer Sciences
More informationMathematical Finance
Mathematical Finance Option Pricing under the RiskNeutral Measure Cory Barnes Department of Mathematics University of Washington June 11, 2013 Outline 1 Probability Background 2 Black Scholes for European
More informationThe Dirichlet Unit Theorem
Chapter 6 The Dirichlet Unit Theorem As usual, we will be working in the ring B of algebraic integers of a number field L. Two factorizations of an element of B are regarded as essentially the same if
More informationDecember 4, 2013 MATH 171 BASIC LINEAR ALGEBRA B. KITCHENS
December 4, 2013 MATH 171 BASIC LINEAR ALGEBRA B KITCHENS The equation 1 Lines in twodimensional space (1) 2x y = 3 describes a line in twodimensional space The coefficients of x and y in the equation
More informationMA651 Topology. Lecture 6. Separation Axioms.
MA651 Topology. Lecture 6. Separation Axioms. This text is based on the following books: Fundamental concepts of topology by Peter O Neil Elements of Mathematics: General Topology by Nicolas Bourbaki Counterexamples
More informationTHE FUNDAMENTAL THEOREM OF ALGEBRA VIA PROPER MAPS
THE FUNDAMENTAL THEOREM OF ALGEBRA VIA PROPER MAPS KEITH CONRAD 1. Introduction The Fundamental Theorem of Algebra says every nonconstant polynomial with complex coefficients can be factored into linear
More informationMathematics Course 111: Algebra I Part IV: Vector Spaces
Mathematics Course 111: Algebra I Part IV: Vector Spaces D. R. Wilkins Academic Year 19967 9 Vector Spaces A vector space over some field K is an algebraic structure consisting of a set V on which are
More informationOnline Convex Optimization
E0 370 Statistical Learning heory Lecture 19 Oct 22, 2013 Online Convex Optimization Lecturer: Shivani Agarwal Scribe: Aadirupa 1 Introduction In this lecture we shall look at a fairly general setting
More informationModern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh
Modern Optimization Methods for Big Data Problems MATH11146 The University of Edinburgh Peter Richtárik Week 3 Randomized Coordinate Descent With Arbitrary Sampling January 27, 2016 1 / 30 The Problem
More informationOverview of Violations of the Basic Assumptions in the Classical Normal Linear Regression Model
Overview of Violations of the Basic Assumptions in the Classical Normal Linear Regression Model 1 September 004 A. Introduction and assumptions The classical normal linear regression model can be written
More informationTHE FUNDAMENTAL THEOREM OF ARBITRAGE PRICING
THE FUNDAMENTAL THEOREM OF ARBITRAGE PRICING 1. Introduction The BlackScholes theory, which is the main subject of this course and its sequel, is based on the Efficient Market Hypothesis, that arbitrages
More informationProbability and Random Variables. Generation of random variables (r.v.)
Probability and Random Variables Method for generating random variables with a specified probability distribution function. Gaussian And Markov Processes Characterization of Stationary Random Process Linearly
More information15 Limit sets. Lyapunov functions
15 Limit sets. Lyapunov functions At this point, considering the solutions to ẋ = f(x), x U R 2, (1) we were most interested in the behavior of solutions when t (sometimes, this is called asymptotic behavior
More informationRobust output feedback stabilization of the angular velocity of a rigid body
Systems & Control Letters 39 (2000) 203 20 www.elsevier.com/locate/sysconle Robust output feedback stabilization of the angular velocity of a rigid body F. Mazenc, A. Astol Centre for Process Systems Engineering,
More informationModeling and Performance Evaluation of Computer Systems Security Operation 1
Modeling and Performance Evaluation of Computer Systems Security Operation 1 D. Guster 2 St.Cloud State University 3 N.K. Krivulin 4 St.Petersburg State University 5 Abstract A model of computer system
More informationA Direct Numerical Method for Observability Analysis
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL 15, NO 2, MAY 2000 625 A Direct Numerical Method for Observability Analysis Bei Gou and Ali Abur, Senior Member, IEEE Abstract This paper presents an algebraic method
More informationStability of the LMS Adaptive Filter by Means of a State Equation
Stability of the LMS Adaptive Filter by Means of a State Equation Vítor H. Nascimento and Ali H. Sayed Electrical Engineering Department University of California Los Angeles, CA 90095 Abstract This work
More informationNumerical Analysis Lecture Notes
Numerical Analysis Lecture Notes Peter J. Olver 5. Inner Products and Norms The norm of a vector is a measure of its size. Besides the familiar Euclidean norm based on the dot product, there are a number
More informationALMOST COMMON PRIORS 1. INTRODUCTION
ALMOST COMMON PRIORS ZIV HELLMAN ABSTRACT. What happens when priors are not common? We introduce a measure for how far a type space is from having a common prior, which we term prior distance. If a type
More informationThe Goldberg Rao Algorithm for the Maximum Flow Problem
The Goldberg Rao Algorithm for the Maximum Flow Problem COS 528 class notes October 18, 2006 Scribe: Dávid Papp Main idea: use of the blocking flow paradigm to achieve essentially O(min{m 2/3, n 1/2 }
More informationLecture 8: Random Walk vs. Brownian Motion, Binomial Model vs. LogNormal Distribution
Lecture 8: Random Walk vs. Brownian Motion, Binomial Model vs. Logormal Distribution October 4, 200 Limiting Distribution of the Scaled Random Walk Recall that we defined a scaled simple random walk last
More informationBy W.E. Diewert. July, Linear programming problems are important for a number of reasons:
APPLIED ECONOMICS By W.E. Diewert. July, 3. Chapter : Linear Programming. Introduction The theory of linear programming provides a good introduction to the study of constrained maximization (and minimization)
More information1 Norms and Vector Spaces
008.10.07.01 1 Norms and Vector Spaces Suppose we have a complex vector space V. A norm is a function f : V R which satisfies (i) f(x) 0 for all x V (ii) f(x + y) f(x) + f(y) for all x,y V (iii) f(λx)
More informationIdeal Class Group and Units
Chapter 4 Ideal Class Group and Units We are now interested in understanding two aspects of ring of integers of number fields: how principal they are (that is, what is the proportion of principal ideals
More informationLabeling outerplanar graphs with maximum degree three
Labeling outerplanar graphs with maximum degree three Xiangwen Li 1 and Sanming Zhou 2 1 Department of Mathematics Huazhong Normal University, Wuhan 430079, China 2 Department of Mathematics and Statistics
More informationA LogRobust Optimization Approach to Portfolio Management
A LogRobust Optimization Approach to Portfolio Management Dr. Aurélie Thiele Lehigh University Joint work with Ban Kawas Research partially supported by the National Science Foundation Grant CMMI0757983
More informationPROVING STATEMENTS IN LINEAR ALGEBRA
Mathematics V2010y Linear Algebra Spring 2007 PROVING STATEMENTS IN LINEAR ALGEBRA Linear algebra is different from calculus: you cannot understand it properly without some simple proofs. Knowing statements
More informationFuzzy Probability Distributions in Bayesian Analysis
Fuzzy Probability Distributions in Bayesian Analysis Reinhard Viertl and Owat Sunanta Department of Statistics and Probability Theory Vienna University of Technology, Vienna, Austria Corresponding author:
More informationLecture 13 Linear quadratic Lyapunov theory
EE363 Winter 289 Lecture 13 Linear quadratic Lyapunov theory the Lyapunov equation Lyapunov stability conditions the Lyapunov operator and integral evaluating quadratic integrals analysis of ARE discretetime
More informationSeparation Properties for Locally Convex Cones
Journal of Convex Analysis Volume 9 (2002), No. 1, 301 307 Separation Properties for Locally Convex Cones Walter Roth Department of Mathematics, Universiti Brunei Darussalam, Gadong BE1410, Brunei Darussalam
More informationThis article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE/ACM TRANSACTIONS ON NETWORKING 1 A Greedy Link Scheduler for Wireless Networks With Gaussian MultipleAccess and Broadcast Channels Arun Sridharan, Student Member, IEEE, C Emre Koksal, Member, IEEE,
More informationMath 2331 Linear Algebra
2.2 The Inverse of a Matrix Math 2331 Linear Algebra 2.2 The Inverse of a Matrix Jiwen He Department of Mathematics, University of Houston jiwenhe@math.uh.edu math.uh.edu/ jiwenhe/math2331 Jiwen He, University
More information24. The Branch and Bound Method
24. The Branch and Bound Method It has serious practical consequences if it is known that a combinatorial problem is NPcomplete. Then one can conclude according to the present state of science that no
More informationNotes on Factoring. MA 206 Kurt Bryan
The General Approach Notes on Factoring MA 26 Kurt Bryan Suppose I hand you n, a 2 digit integer and tell you that n is composite, with smallest prime factor around 5 digits. Finding a nontrivial factor
More informationOPRE 6201 : 2. Simplex Method
OPRE 6201 : 2. Simplex Method 1 The Graphical Method: An Example Consider the following linear program: Max 4x 1 +3x 2 Subject to: 2x 1 +3x 2 6 (1) 3x 1 +2x 2 3 (2) 2x 2 5 (3) 2x 1 +x 2 4 (4) x 1, x 2
More informationNonparametric adaptive age replacement with a onecycle criterion
Nonparametric adaptive age replacement with a onecycle criterion P. CoolenSchrijner, F.P.A. Coolen Department of Mathematical Sciences University of Durham, Durham, DH1 3LE, UK email: Pauline.Schrijner@durham.ac.uk
More informationWeakly Secure Network Coding
Weakly Secure Network Coding Kapil Bhattad, Student Member, IEEE and Krishna R. Narayanan, Member, IEEE Department of Electrical Engineering, Texas A&M University, College Station, USA Abstract In this
More informationLecture 13: Martingales
Lecture 13: Martingales 1. Definition of a Martingale 1.1 Filtrations 1.2 Definition of a martingale and its basic properties 1.3 Sums of independent random variables and related models 1.4 Products of
More informationLeast Squares Estimation
Least Squares Estimation SARA A VAN DE GEER Volume 2, pp 1041 1045 in Encyclopedia of Statistics in Behavioral Science ISBN13: 9780470860809 ISBN10: 0470860804 Editors Brian S Everitt & David
More informationPattern Analysis. Logistic Regression. 12. Mai 2009. Joachim Hornegger. Chair of Pattern Recognition Erlangen University
Pattern Analysis Logistic Regression 12. Mai 2009 Joachim Hornegger Chair of Pattern Recognition Erlangen University Pattern Analysis 2 / 43 1 Logistic Regression Posteriors and the Logistic Function Decision
More informationSolution Using the geometric series a/(1 r) = x=1. x=1. Problem For each of the following distributions, compute
Math 472 Homework Assignment 1 Problem 1.9.2. Let p(x) 1/2 x, x 1, 2, 3,..., zero elsewhere, be the pmf of the random variable X. Find the mgf, the mean, and the variance of X. Solution 1.9.2. Using the
More information3.4. Solving Simultaneous Linear Equations. Introduction. Prerequisites. Learning Outcomes
Solving Simultaneous Linear Equations 3.4 Introduction Equations often arise in which there is more than one unknown quantity. When this is the case there will usually be more than one equation involved.
More informationMultivariate Normal Distribution
Multivariate Normal Distribution Lecture 4 July 21, 2011 Advanced Multivariate Statistical Methods ICPSR Summer Session #2 Lecture #47/21/2011 Slide 1 of 41 Last Time Matrices and vectors Eigenvalues
More informationOn the second part of Hilbert s 16th problem
Nonlinear Analysis ( ) www.elsevier.com/locate/na On the second part of Hilbert s 6th problem Elin Oxenhielm Department of Mathematics, Stockholm University, Stockholm 69, Sweden Received 3 July 23; accepted
More informationMaster s Theory Exam Spring 2006
Spring 2006 This exam contains 7 questions. You should attempt them all. Each question is divided into parts to help lead you through the material. You should attempt to complete as much of each problem
More informationCHAPTER II THE LIMIT OF A SEQUENCE OF NUMBERS DEFINITION OF THE NUMBER e.
CHAPTER II THE LIMIT OF A SEQUENCE OF NUMBERS DEFINITION OF THE NUMBER e. This chapter contains the beginnings of the most important, and probably the most subtle, notion in mathematical analysis, i.e.,
More informationMATH 304 Linear Algebra Lecture 8: Inverse matrix (continued). Elementary matrices. Transpose of a matrix.
MATH 304 Linear Algebra Lecture 8: Inverse matrix (continued). Elementary matrices. Transpose of a matrix. Inverse matrix Definition. Let A be an n n matrix. The inverse of A is an n n matrix, denoted
More informationSPARE PARTS INVENTORY SYSTEMS UNDER AN INCREASING FAILURE RATE DEMAND INTERVAL DISTRIBUTION
SPARE PARS INVENORY SYSEMS UNDER AN INCREASING FAILURE RAE DEMAND INERVAL DISRIBUION Safa Saidane 1, M. Zied Babai 2, M. Salah Aguir 3, Ouajdi Korbaa 4 1 National School of Computer Sciences (unisia),
More informationLinear Systems. Singular and Nonsingular Matrices. Find x 1, x 2, x 3 such that the following three equations hold:
Linear Systems Example: Find x, x, x such that the following three equations hold: x + x + x = 4x + x + x = x + x + x = 6 We can write this using matrixvector notation as 4 {{ A x x x {{ x = 6 {{ b General
More informationTangent and normal lines to conics
4.B. Tangent and normal lines to conics Apollonius work on conics includes a study of tangent and normal lines to these curves. The purpose of this document is to relate his approaches to the modern viewpoints
More informationRolle s Theorem. q( x) = 1
Lecture 1 :The Mean Value Theorem We know that constant functions have derivative zero. Is it possible for a more complicated function to have derivative zero? In this section we will answer this question
More informationMATRIX ALGEBRA AND SYSTEMS OF EQUATIONS. + + x 2. x n. a 11 a 12 a 1n b 1 a 21 a 22 a 2n b 2 a 31 a 32 a 3n b 3. a m1 a m2 a mn b m
MATRIX ALGEBRA AND SYSTEMS OF EQUATIONS 1. SYSTEMS OF EQUATIONS AND MATRICES 1.1. Representation of a linear system. The general system of m equations in n unknowns can be written a 11 x 1 + a 12 x 2 +
More informationMarkov random fields and Gibbs measures
Chapter Markov random fields and Gibbs measures 1. Conditional independence Suppose X i is a random element of (X i, B i ), for i = 1, 2, 3, with all X i defined on the same probability space (.F, P).
More informationIRREDUCIBLE OPERATOR SEMIGROUPS SUCH THAT AB AND BA ARE PROPORTIONAL. 1. Introduction
IRREDUCIBLE OPERATOR SEMIGROUPS SUCH THAT AB AND BA ARE PROPORTIONAL R. DRNOVŠEK, T. KOŠIR Dedicated to Prof. Heydar Radjavi on the occasion of his seventieth birthday. Abstract. Let S be an irreducible
More informationMathematical Background
Appendix A Mathematical Background A.1 Joint, Marginal and Conditional Probability Let the n (discrete or continuous) random variables y 1,..., y n have a joint joint probability probability p(y 1,...,
More informationResearch Article Stability Analysis for HigherOrder Adjacent Derivative in Parametrized Vector Optimization
Hindawi Publishing Corporation Journal of Inequalities and Applications Volume 2010, Article ID 510838, 15 pages doi:10.1155/2010/510838 Research Article Stability Analysis for HigherOrder Adjacent Derivative
More information1 Short Introduction to Time Series
ECONOMICS 7344, Spring 202 Bent E. Sørensen January 24, 202 Short Introduction to Time Series A time series is a collection of stochastic variables x,.., x t,.., x T indexed by an integer value t. The
More information