A minimax optimal control problem

A control system x=f(t,x,u) is considered, and a cost functional ess supT ₀tT ₁ G(t, x(t),u(t)) is to be minimized. Necessary conditions for optimality (maximum principle and transversality conditions) are derived. It is also shown that an optimal control is optimal for the corresponding problem on a subinterval of [T ₀,T ₁], if a certain controllability condition is satisfied.     

Minimal time for the approximate bilinear control of Schrödinger equations

We consider a quantum particle in a potential V(x) subject to a time‐dependent (and spatially homogeneous) electric field E(t) (the control). Boscain, Caponigro, Chambrion, and Sigalotti proved that, under generic assumptions on V, this system is approximately controllable on the unit sphere, in sufficiently large time T. In the present article, we show that, for a large class of initial states (dense in unit sphere), approximate controllability does not hold in arbitrarily small time. This generalizes our previous result for Gaussian initial conditions. Furthermore, we prove that the minimal time can in fact be arbitrarily large.     

Global null controllability of nonlinear delay equations with controls in a compact set

The question of controlling some nonlinear retarded functional differential equations from an initial function to the zero function is considered. The control setsM() are square integrable functions with values in the unit closed sphereL ₂([t ₀, ),E ^m ) with center the origin. Assuming that the linear approximation of the nonlinear equation is null controllable with some integrable controls on some interval [t ₀,t ₁–2r], wheret ₀ is sufficiently large and wherer>0 is the delay, and assuming that the nonlinear system, withu=0, is uniformly globally asymptotically stable, we show that the nonlinear control process is globally null controllable with controlsuM(). The paper gives conditions which guarantee the stability assumptions, and also indicates conditions which yield the null controllability assumptions of the linear approximation. Our research extends known results on ordinary differential processes.     

Complete controllability

A linear autonomous control system in ⁿ is said to be completely controllable iff there existsT>0 such that eachx ⁿ can be steered to anyy ⁿ in timeT. This paper presents a geometric characterization of this property in the case in which there are constraints on the values which the control maps can assume. A necessary and sufficient condition to get instant controllability (i.e., complete controllability for anyT>0) is also derived. This condition generalizes the well-known Kalman condition to the constrained case.     

Approximation and regular perturbation of optimal control problems via Hamilton-Jacobi theory

We present two convergence theorems for Hamilton-Jacobi equations and we apply them to the convergence of approximations and perturbations of optimal control problems and of two-players zero-sum differential games. One of our results is, for instance, the following. LetT andT _h be the minimal time functions to reach the origin of two control systemsy = f(y, a) andy = f _h (y, a), both locally controllable in the origin, and letK be any compact set of points controllable to the origin. If f _h –f Ch, then |T(x) – T _h (x)| C _K h , for all x K, where is the exponent of Hölder continuity ofT(x).     

Sufficient conditions for asymptotic optimal control

For a selected family of Lagrange-type control problems involving a nonnegative integral costJ _T (y,u) over the interval [0,T], 0<T<, with system conditions consisting of differential inequalities and/or equalities, the following material is treated: (i) a resumé of relevant necessary conditions and sufficient conditions for a pair (y _T ,u _T ) to minimizeJ _T (y,u); (ii) conditions sufficient for the convergence asT of minimizing pairs (y _T ,u _T ) over [0,T] to a limit pair (y ,u ) over the infinite-time interval [0, ); (iii) conditions sufficient for (y ,u ) to minimize the costJ (y,u) over [0, ); and (iv) conditions sufficient for the optimal cost per unit timeJ _T (y _T ,u _T )/T to have a limit asT.     

Control problems with random and progressively known targets

The situation considered is of optimally controlling a deterministic system from a given state to an initially unknown targety in a fixed time interval [T ₀,T]. The target will be certainly known at a random time in [T ₀,T]. The controller knows the distributions ofy and . We derive the Bellman equation for the problem, prove a verification theorem for it, and demonstrate how the distribution influences the optimal control. We show that, in the linear-quadratic case, the optimal control is given by a feedback law that does not depend on the distribution of .The author wishes to express his gratitude to Prof. Wendell H. Fleming and Prof. Steven Orey for fruitful discussion concerning this work.This research was supported in part by the Institute for Mathematics and Its Applications with funds provided by the National Science Foundation.     

The entropy theory of symbolic extensions

Fix a topological system (X,T), with its space K(X,T) of T-invariant Borel probabilities. If (Y,S) is a symbolic system (subshift) and :(Y,S)(X,T) is a topological extension (factor map), then the function h_ext on K(X,T) which assigns to each the maximal entropy of a measure on Y mapping to is called the extension entropy function of . The infimum of such functions over all symbolic extensions is called the symbolic extension entropy function and is denoted by h_sex. In this paper we completely characterize these functions in terms of functional analytic properties of an entropy structure on (X,T). The entropy structure is a sequence of entropy functions h_k defined with respect to a refining sequence of partitions of X (or of X×Z, for some auxiliary system (Z,R) with simple dynamics) whose boundaries have measure zero for all the invariant Borel probabilities. We develop the functional analysis and computational techniques to produce many dynamical examples; for instance, we resolve in the negative the question of whether the infimum of the topological entropies of symbolic extensions of (X,T) must always be attained, and we show that the maximum value of h_sex need not be achieved at an ergodic measure. We exhibit several characterizations of the asymptotically h-expansive systems of Misiurewicz, which emerge as a fundamental natural class in the context of the entropy structure. The results of this paper are required for the Downarowicz-Newhouse results [DN] on smooth dynamical systems. Mathematics Subject Classification (2000) Primary: 37B10; Secondary: 37B40, 37C40, 37C45, 37C99, 37D35     

A discrete method of optimal control based upon the cell state space concept

A discrete method of optimal control is proposed in this paper. The continuum state space of a system is discretized into a cell state space, and the cost function is discretized in a similar manner. Assuming intervalwise constant controls and using a finite set of admissible control levels (u) and a finite set of admissible time intervals (), the motion of the system under all possible interval controls (u, ) can then be expressed in terms of a family of cell-to-cell mappings. The proposed method extracts the optimal control results from these mappings by a systematic search, culminating in the construction of a discrete optimal control table.The possibility of expressing the optimal control results in the form of a control table seems to give this method a means to make systems real-time controllable.Dedicated to G. LeitmannThe material is based upon work supported by the National Science Foundation under Grant No. MEA-82-17471. The author is also indebted to Professor G. Leitmann for his many helpful comments.     

On Global Optimality Conditions for Nonlinear Optimal Control Problems

Let a trajectory and control pair maximize globally the functional g(x(T)) in the basic optimal control problem. Then (evidently) any pair (x,u) from the level set of the functional g corresponding to the value g( (T)) is also globally optimal and satisfies the Pontryagin maximum principle. It is shown that this necessary condition for global optimality of turns out to be a sufficient one under the additional assumption of nondegeneracy of the maximum principle for every pair (x,u) from the above-mentioned level set. In particular, if the pair satisfies the Pontryagin maximum principle which is nondegenerate in the sense that for the Hamiltonian H, we have along the pair on [0,T], and if there is no another pair (x,u) such that g(x(T))=g( (T)), then is a global maximizer.     

Periodic solutions with prescribed minimal period for convex autonomous hamiltonian systems

Clarke has shown that the problem of findingT-periodic solutions for a convex Hamiltonian system is equivalent to the problem of finding critical points to a certain functional, dual to the classical action functional. In this paper, we relate the Morse index of the critical point to the minimal period of the correspondingT-periodic solution. In particular, we show that if the critical point is obtained by the Ambrosetti-Rabinowitz mountain-pass theorem the corresponding solution has minimal periodT, that is, it cannot beT/k-periodic withk integer,k2. As a consequence, we prove that if the Hamiltonian is flat near an equilibrium and superquadratic near infinity, then for anyT>0, the corresponding Hamiltonian system has a periodic solution with minimal periodT.     

A Finite Algorithm for Solving Infinite Dimensional Optimization Problems

We consider the general optimization problem (P) of selecting a continuous function x over a -compact Hausdorff space T to a metric space A, from a feasible region X of such functions, so as to minimize a functional c on X. We require that X consist of a closed equicontinuous family of functions lying in the product (over T) of compact subsets Y _t of A. (An important special case is the optimal control problem of finding a continuous time control function x that minimizes its associated discounted cost c(x) over the infinite horizon.) Relative to the uniform-on-compacta topology on the function space C(T,A) of continuous functions from T to A, the feasible region X is compact. Thus optimal solutions x ^* to (P) exist under the assumption that c is continuous. We wish to approximate such an x ^* by optimal solutions to a net {P _i }, iI, of approximating problems of the form min_{xX i} c i(x) for each iI, where (1) the net of sets {X _i } _I converges to X in the sense of Kuratowski and (2) the net {c _i } _I of functions converges to c uniformly on X. We show that for large i, any optimal solution x ^* _i to the approximating problem (P _i ) arbitrarily well approximates some optimal solution x ^* to (P). It follows that if (P) is well-posed, i.e., limsupX _i ^* is a singleton {x ^*}, then any net {x _i ^*} _I of (P _i )-optimal solutions converges in C(T,A) to x ^*. For this case, we construct a finite algorithm with the following property: given any prespecified error and any compact subset Q of T, our algorithm computes an i in I and an associated x _i ^* in X _i ^* which is within of x ^* on Q. We illustrate the theory and algorithm with a problem in continuous time production control over an infinite horizon.     

A new control chart in contaminated data of t‐Student distribution for individual observations

Control charts are the most popular tool for monitoring production quality. In traditional control charts, it is usually supposed that the observations follow a multivariate normal distribution. Nevertheless, there are many practical applications where the normality assumption is not fulfilled. Furthermore, the performance of these charts in the presence of measurement errors (outliers) in the historical data has been improved using robust control charts when the observations follow a normal distribution. In this paper, we develop a new control chart for t‐Student data based on the trimmed T² control chart () through the adaptation of the elements of this chart to the case of this distribution. Simulation studies show that a control chart performs better than T² in t‐Student samples for individual observations. Copyright © 2012 John Wiley & Sons, Ltd.     

Approximate solution of a class of singular control problems

A class of singular control problems involving amplitude constraints on the controls is examined. IfL is the space of control functionsU, the control constraint setS can be identified with the unit ball inL . Now, for anyn (1, ), an analogous problem may be set up withL ⁿ forU and the unit ball inL ⁿ forS. This modified problem is necessarily nonsingular for controllable systems. It is shown that, by takingn sufficiently large, the solution to the modified problem also solves the original problem arbitrarily closely (in a sense made precise). Behavior asn is investigated.This research was supported by the Science Research Council of Great Britain and the Commonwealth Fund (Harkness Fellowship).     

On nearly self-optimizing strategies for a discrete-time uniformly ergodic adaptive model

We control a discrete-time uniformly ergodic system, which depends on an unknown parameter ⁰ A, a compact set. Our purpose is to minimize the long-run average-cost functional. We estimate the unknown parameter using the biased maximum likelihood estimator and apply the control which is almost optimal for the value of estimation. This way we construct strategies such that the value of the cost functional can be arbitrarily close to the optimal value obtained for ⁰.     

On the strong controllability of the diffusion equation

Let (x,t)y (x,t),x[0, 1],t[0,T], be the solution of the diffusion equation in one spatial variable corresponding to zero initial conditions and boundary controluL ₂(0,T). GivenfL ₂(0, 1), it is not possible, in general, to find a controlu such thaty(·,T)=f. We extend the space of controls in such a manner thatL ₂(0,T) can be considered to be a subset of a new spaceS of control elements; this space contains elements which do provide a solution to the problem of moments associated with the problem of makingy(·,T)=f inL ₂(0, 1). We show then that the action of the elements ofS can be approximated by that of control functions inL ₂(0,T) in a suitable manner.     

Ergodicity and weak-mixing of homogeneous extensions of measure-preserving transformations with applications to Markov shifts

Given a homogeneous extensionS of a measure-preserving transformationT, we provide necessary and sufficient conditions for the ergodicity and weak-mixing ofS in terms of functional equations. We then apply our findings to the case whenT is a Markov shift and the associated skewing function ofS depends on a finite number of coordinates. In this case, we obtain a simplification to the appropriate functional equations.     

Time-average and asymptotically optimal flow control policies in networks with multiple transmitters

We considerM transmitting stations sending packets to a single receiver over a slotted time-multiplexed link. For each phase consisting ofT consecutive slots, the receiver dynamically allocates these slots among theM transmitters. Our objective is to characterize policies that minimize the long-term average of the total number of messages awaiting service at theM transmitters. We establish necessary and sufficient conditions on the arrival processes at the transmitters for the existence of finite cost time-average policies; it is not enough that the average arrival rate is strictly less than the slot capacity. We construct a pure strategy that attains a finite average cost under these conditions. This in turn leads to the existence of an optimal time-average pure policy for each phase lengthT, and to upper and lower bounds on the cost this policy achieves. Furthermore, we show that such an optimal time-average policy has the same properties as those of optimal discounted policies investigated by the authors in a previous paper. Finally, we prove that in the absence of costs accrued by messages within the phase, there exists a policy such that the time-average cost tends toward zero as the phase lengthT.     

Partition deformations in hierarchical hybrid control systems

Hierarchical hybrid control (HHC) (Caines & Wei, submitted)needs to be applicable to systems in which the dynamics of thesystem, the nature of the controllable flows, and the positionsof the boundaries of any partition can only be approximatelyspecified. In this paper, conditions are given under which thedynamics of a high-level HHC system M, and its associated setof control laws, are robust (that is to say, insensitive) withrespect to small perturbations of the partitions defining Mand small perturbations of the controllable flow. This analysisemploys two notions of the deformation of a partition specifiedrespectively by (i) maps of the boundaries and (ii) maps ofthe blocks of . Further, the robustness properties of the partitionmachine M are investigated for deformations of the controllableflow and, finally, for certain mechanical systems.     

Fcfs-scheduling in a hard real-time environment under rush-hour conditions

We investigate some real-time behaviour of a (discrete time) single server system with FCFS (first come first serve) task scheduling under rush-hour conditions. The main result deals with the probability distribution of a random variable SRD(T), which describes the time the system operates without violating a fixed task service time deadlineT.Relying on a simple general probability model, asymptotic formulas concerning the mean and the variance of SRD(T) are determined; for instance, if the average arrival rate is larger than the departure rate, the expectation of SRD(T) is proved to fulfilE[SRD(T)]=c ₁+O(T ^–3) forT, wherec ₁ denotes some constant. If the arrival rate equals the departure rate, we findE[SRD(T)]c ₂ T ⁱ for somei2.