CONTROL AND STABILIZATION OF THE KORTEWEG-DE VRIES EQUATION: RECENT PROGRESSES

Similar documents
Solitons : An Introduction

Nonlinear Optics (WiSe 2015/16) Lecture 7: November 27, 2015

Nonlinear Optics (WiSe 2017/18) Lecture 12: November 28, 2017

Nonlinear Optics (WiSe 2018/19) Lecture 7: November 30, 2018

Shock Waves & Solitons

Convergence of operator splitting for the KdV equation

Solitons in optical fibers. by: Khanh Kieu

Dark & Bright Solitons in Strongly Repulsive Bose-Einstein Condensate

FORCED OSCILLATIONS OF A CLASS OF NONLINEAR DISPERSIVE WAVE EQUATIONS AND THEIR STABILITY

Soliton Molecules. Fedor Mitschke Universität Rostock, Institut für Physik. Benasque, October

Exponential Energy Decay for the Kadomtsev-Petviashvili (KP-II) equation

Forced Oscillations of the Korteweg-de Vries Equation on a Bounded Domain and their Stability

Partial differential equation for temperature u(x, t) in a heat conducting insulated rod along the x-axis is given by the Heat equation:

Stability of an abstract wave equation with delay and a Kelvin Voigt damping

12 Lecture 12. Lax pair and Toda lattice

Stabilization of a Boussinesq system of KdV KdV type

What we do understand by the notion of soliton or rather solitary wave is a wave prole which is very stable in the following sense

Dispersion relations, stability and linearization

Partial Differential Equations

An introduction to Mathematical Theory of Control

Deterministic Dynamic Programming

KdV soliton solutions to a model of hepatitis C virus evolution

Lucio Demeio Dipartimento di Ingegneria Industriale e delle Scienze Matematiche

ATTRACTORS FOR SEMILINEAR PARABOLIC PROBLEMS WITH DIRICHLET BOUNDARY CONDITIONS IN VARYING DOMAINS. Emerson A. M. de Abreu Alexandre N.

Applied Analysis (APPM 5440): Final exam 1:30pm 4:00pm, Dec. 14, Closed books.

On some weighted fractional porous media equations

On the bang-bang property of time optimal controls for infinite dimensional linear systems

WELL-POSEDNESS OF THE TWO-DIMENSIONAL GENERALIZED BENJAMIN-BONA-MAHONY EQUATION ON THE UPPER HALF PLANE

Math 575-Lecture 26. KdV equation. Derivation of KdV

EXISTENCE OF SOLUTIONS TO ASYMPTOTICALLY PERIODIC SCHRÖDINGER EQUATIONS

Presenter: Noriyoshi Fukaya

ASYMPTOTIC BEHAVIOR OF THE KORTEWEG-DE VRIES EQUATION POSED IN A QUARTER PLANE

z x = f x (x, y, a, b), z y = f y (x, y, a, b). F(x, y, z, z x, z y ) = 0. This is a PDE for the unknown function of two independent variables.

Artificial boundary conditions for dispersive equations. Christophe Besse

u xx + u yy = 0. (5.1)

Local null controllability of the N-dimensional Navier-Stokes system with N-1 scalar controls in an arbitrary control domain

OBSERVABILITY INEQUALITIES AND MEASURABLE SETS J. APRAIZ, L. ESCAURIAZA, G. WANG, AND C. ZHANG

Exponential stability of abstract evolution equations with time delay feedback

Lecture Notes on PDEs

Conservative Control Systems Described by the Schrödinger Equation

Separation of Variables in Linear PDE: One-Dimensional Problems

NONTRIVIAL SOLUTIONS TO INTEGRAL AND DIFFERENTIAL EQUATIONS

GENERATORS WITH INTERIOR DEGENERACY ON SPACES OF L 2 TYPE

Lecture No 1 Introduction to Diffusion equations The heat equat

1 The Observability Canonical Form

Functional Analysis. Franck Sueur Metric spaces Definitions Completeness Compactness Separability...

Applied Math Qualifying Exam 11 October Instructions: Work 2 out of 3 problems in each of the 3 parts for a total of 6 problems.

Sobolev Spaces. Chapter 10

Inverse scattering technique in gravity

SEMILINEAR ELLIPTIC EQUATIONS WITH DEPENDENCE ON THE GRADIENT

The incompressible Navier-Stokes equations in vacuum

SOURCE IDENTIFICATION FOR THE HEAT EQUATION WITH MEMORY. Sergei Avdonin, Gulden Murzabekova, and Karlygash Nurtazina. IMA Preprint Series #2459

A Concise Course on Stochastic Partial Differential Equations

Reflected Brownian Motion

CLASSIFICATION AND PRINCIPLE OF SUPERPOSITION FOR SECOND ORDER LINEAR PDE

6 Non-homogeneous Heat Problems

Control, Stabilization and Numerics for Partial Differential Equations

Nonlinear Dynamical Systems Lecture - 01

Traces, extensions and co-normal derivatives for elliptic systems on Lipschitz domains

13 PDEs on spatially bounded domains: initial boundary value problems (IBVPs)

Research Article On the Blow-Up Set for Non-Newtonian Equation with a Nonlinear Boundary Condition

Liquid crystal flows in two dimensions

LECTURE 1: SOURCES OF ERRORS MATHEMATICAL TOOLS A PRIORI ERROR ESTIMATES. Sergey Korotov,

LECTURE # 0 BASIC NOTATIONS AND CONCEPTS IN THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS (PDES)

A higher-order Boussinesq equation in locally non-linear theory of one-dimensional non-local elasticity

Diffusion on the half-line. The Dirichlet problem

CUTOFF RESOLVENT ESTIMATES AND THE SEMILINEAR SCHRÖDINGER EQUATION

Nonlinear Modulational Instability of Dispersive PDE Models

Spectrum and Exact Controllability of a Hybrid System of Elasticity.

Dispersion relations, linearization and linearized dynamics in PDE models

Abstract In this paper, we consider bang-bang property for a kind of timevarying. time optimal control problem of null controllable heat equation.

Some SDEs with distributional drift Part I : General calculus. Flandoli, Franco; Russo, Francesco; Wolf, Jochen

Eigenvalues and Eigenfunctions of the Laplacian

BOUNDARY VALUE PROBLEMS IN KREĬN SPACES. Branko Ćurgus Western Washington University, USA

Model Equation, Stability and Dynamics for Wavepacket Solitary Waves

OBSERVABILITY INEQUALITY AND DECAY RATE FOR WAVE EQUATIONS WITH NONLINEAR BOUNDARY CONDITIONS

Scientiae Mathematicae Japonicae Online, Vol. 5, (2001), Ryo Ikehata Λ and Tokio Matsuyama y

NONTRIVIAL SOLUTIONS FOR SUPERQUADRATIC NONAUTONOMOUS PERIODIC SYSTEMS. Shouchuan Hu Nikolas S. Papageorgiou. 1. Introduction

REACTION-DIFFUSION EQUATIONS FOR POPULATION DYNAMICS WITH FORCED SPEED II - CYLINDRICAL-TYPE DOMAINS. Henri Berestycki and Luca Rossi

Global well-posedness for KdV in Sobolev spaces of negative index

ELEMENTS OF PROBABILITY THEORY

ASYMPTOTICALLY NONEXPANSIVE MAPPINGS IN MODULAR FUNCTION SPACES ABSTRACT

Controllability of the linear 1D wave equation with inner moving for

REAL AND COMPLEX ANALYSIS

Convexity of the Reachable Set of Nonlinear Systems under L 2 Bounded Controls

Existence and Multiplicity of Solutions for a Class of Semilinear Elliptic Equations 1

1.3.1 Definition and Basic Properties of Convolution

Lecture 12: Detailed balance and Eigenfunction methods

DISPERSIVE EQUATIONS: A SURVEY

ORBITAL STABILITY OF SOLITARY WAVES FOR A 2D-BOUSSINESQ SYSTEM

CONTROL AND STABILIZATION OF A FAMILY OF BOUSSINESQ SYSTEMS

3 (Due ). Let A X consist of points (x, y) such that either x or y is a rational number. Is A measurable? What is its Lebesgue measure?

Nonlinear stabilization via a linear observability

Inégalités spectrales pour le contrôle des EDP linéaires : groupe de Schrödinger contre semigroupe de la chaleur.

Recent result on porous medium equations with nonlocal pressure

Are Solitary Waves Color Blind to Noise?

There are five problems. Solve four of the five problems. Each problem is worth 25 points. A sheet of convenient formulae is provided.

Global Solutions for a Nonlinear Wave Equation with the p-laplacian Operator

NONLINEAR DECAY AND SCATTERING OF SOLUTIONS TO A BRETHERTON EQUATION IN SEVERAL SPACE DIMENSIONS

ON WEAKLY NONLINEAR BACKWARD PARABOLIC PROBLEM

Transcription:

Jrl Syst Sci & Complexity (29) 22: 647 682 CONTROL AND STABILIZATION OF THE KORTEWEG-DE VRIES EQUATION: RECENT PROGRESSES Lionel ROSIER Bing-Yu ZHANG Received: 27 July 29 c 29 Springer Science + Business Media, LLC Abstract The study of the control and stabilization of the KdV equation began with the work of Russell and Zhang in late 198s. Both exact control and stabilization problems have been intensively studied since then and significant progresses have been made due to many people s hard work and contributions. In this article, the authors intend to give an overall review of the results obtained so far in the study but with an emphasis on its recent progresses. A list of open problems is also provided for further investigation. Key words Exact controllability, Korteweg-de Vries equation, smoothing property, stabilizability, unique continuation. Acknowledgements BZ had spent two wonderful years (1985 1986) at the Institute of System Science, AMSS, CAS as a graduate student. He would like to take this opportunity to express his deepest thanks to his advisors, Professor Chen Hanfu and Professor Feng Dexing, for their constant support and encouragement. He is grateful for their patience, understanding, and kindness during his years at the Institute. BZ also would like to express his sincere thanks to his then graduate student fellows and roommates, Guo Lei and Huang Yunqing, who helped him in many different ways during his years at the Institute. Their friendship has made his years at the Institute more enjoyable and unforgettable. BZ is partially supported by the Charles Phelps Taft Memorial Fund of the University of Cincinnati and the Chunhui program (State Education Ministry of China) under Grant No. 27-1-616. 1 Introduction To tell any story about the Korteweg-de Vries equation one cannot help but begin with John Scott Russell s vivid description of his discovery of solitary waves in his Report on Waves: Lionel Rosier Institut Élie Cartan, UMR 752 UHP/CNRS/INRIA, B.P. 239, F-5456 Vandøeuvre-lès-Nancy Cedex, France. Email: rosier@iecn.u-nancy.fr. Bing-Yu ZHANG Department of Mathematical Sciences, University of Cincinnati, Cincinnati Ohio 45221-25, USA. Email: bzhang@math.uc.edu. The paper is dedicated to the Institute of System Science, Academy of Mathematics and Systems Science, Chinese Academy of Sciences for its 3th anniversary. Report of the 14th meeting of the British Association for the Advancement of Science, York, September 1844 (London 1845), pp 311 39, Plates XLVII-LVII

648 LIONEL ROSIR BING-YU ZHANG I was observing the motion of a boat which was rapidly drawn along a narrow channel by a pair of horses, when the boat suddenly stopped - not so the mass of water in the channel which it had put in motion; it accumulated round the prow of the vessel in a state of violent agitation, then suddenly leaving it behind, rolled forward with great velocity, assuming the form of a large solitary elevation, a rounded, smooth and well-defined heap of water, which continued its course along the channel apparently without change of form or diminution of speed. I followed it on horseback, and overtook it still rolling on at a rate of some eight or nine miles an hour, preserving its original figure some thirty feet long and a foot to a foot and a half in height. Its height gradually diminished, and after a chase of one or two miles I lost it in the windings of the channel. Such, in the month of August 1834, was my first chance interview with that singular and beautiful phenomenon which I have called the Wave of Translation. Following this discovery, John Scott Russell built a 3 feet water tank in his back garden and designed experiments generating long waves to investigate the phenomenon he had observed. He studied the form of the waves, their speed of propagation and stability. Then he challenged the mathematical community to prove theoretically the existence of his solitary wave ( the Wave of Translation ) and to give an a priori demonstration a posteriori, i.e., to show the possible existence of a stable solitary wave propagating without change of form. However, while Russell believed strongly that his solitary wave was of fundamental importance, he encountered resistance from established convictions. In particular, several prominent scientists and mathematicians, including Airy and Stokes, were convinced that the existence of such solitary waves was impossible. The issue kept unresolved for several decades until the positive answer was finally given, first by J. Boussinesq in 1871 (see, for example, [1 4]), some time later in 1876, by Lord Rayleigh [5] and in order to remove still existing doubts over the existence of the solitary wave by G. de Vries [6] and by D. J. Korteweg and G. de Vries [7] in 1895. In their famous paper [7], D. J. Korteweg and his PhD student G. de Vries derived the following equation (now known as the KdV equation) t u + u x u + 3 xu = to model water waves propagating along a shallow canal. Its original form is t η = 3 g ( 1 2 l x 2 η2 + 2 3 αη + 1 3 xη) σ 2, where η is the surface elevation above the equilibrium level l, α is a small constant related to the uniform motion of the liquid, g is the gravitational constant, and σ = l3 3 T l ρg with surface capillary tension T and density ρ. Korteweg and de Vries [7] proved that Russell was correct by finding explicit, closed-form, traveling-wave solutions to their equation that moreover decay rapidly and so represent a highly localized moving hump and therefore completely settled the long-standing open problem. Nevertheless, the KdV equation did not receive significant further attention and its study kept dormant for more than half century until 1965, when N. Zabusky and M. Kruskal [9] published the results of their numerical experimentation with the equation. Their computer generated approximate solutions to the KdV equation indicated that any localized initial profile that was The interested readers are referred to [8] for an excellent article regarding the origin of the KdV equation.

CONTROL AND STABILIZATION OF THE KDV EQUATION 649 allowed to evolve according to the KdV equation eventually consisted of a finite set of localized traveling waves of the same shape and furthermore, when two of the localized disturbances collided, they would emerge from the collision again as another pair of traveling waves with a shift in phase as the only consequence of their interaction. Since the solitary waves which made up these solutions seemed to behave like particles, Zabusky and Kruskal [9] coined the term soliton to describe them and thus opened the era of soliton. Shortly after that, another remarkable discovery was made concerning the KdV equation. C. Gardner, J. Greene, M. Kruskal, and R. Miura [1] demonstrated that it was possible to write many exact solutions to the equation by using ideas from scattering theory. It turns out that their method, now called the inverse scattering transform or the nonlinear Fourier transform, not only can be used to solve the KdV equation, but also can be adapted to solve effectively many other important nonlinear partial differential equations. Because of the discovery of solitons and the invention of the inverse scattering transform, the KdV equation returned to the center stage of nonlinear sciences and has been intensively studied in the past fifty years from various aspects of both mathematics and physics. The equation has become the source of important breakthroughs in mechanics and nonlinear analysis and of many developments in algebra, analysis, geometry, and physics. It has been found that the equation is not only a good model for some water waves but also a very useful approximation model in nonlinear studies whenever one wishes to include and balance a weak nonlinearity and weak dispersive effects [11]. In particular, the equation is now commonly accepted as a mathematical model for the unidirectional propagation of small-amplitude long waves in nonlinear dispersive systems. In this paper, we are mainly concerned with the study of the KdV equation from control point of view. Our attention, in particular, is given to the following three distributed parameter control systems described by the KdV equation. (a) The KdV equation posed on a periodic domain T (a unit circle in the plane) with a forcing function f = f(x, t) added to the equation: t u + u x u + 3 xu = f, u(x, ) = φ(x), x T, t R, (1.1) where f is assumed to be supported in a given open set ω T; (b) The KdV equation posed on a finite interval (, L) with the Dirichlet boundary conditions: t u + x u + u x u + xu 3 =, u(x, ) = φ(x), x (, L), t >, (1.2) u(, t) = h 1 (t), u(l, t) = h 2 (t), x u(l, t) = h 3 (t). (c) The KdV equation posed on the half line R + : t u + x u + u x u + 3 xu =, x R +, t >, u(x, ) = φ(x), u(, t) = h(t), x R +, t >. In these systems, the external forcing terms f, h j, j = 1, 2, 3, and h, are considered as control inputs. The purpose is to see whether one can force the solutions of those systems to have certain desired properties by choosing appropriate control inputs. The following two problems are fundamental in control theory. This is equivalent to imposing the periodic boundary conditions: u(, t) = u(2π, t), xu(, t) = xu(2π, t), 2 xu(, t) = 2 xu(2π, t). (1.3)

65 LIONEL ROSIR BING-YU ZHANG Exact control problem Given an initial state φ and a terminal state ψ in a certain space, can one find an appropriate control input f so that the equation (1.1) admits a solution u which equals φ at time t = and equals ψ at time t = T? If one can always find a control input f to guide the system described by (1.1) from any given initial state φ to any given terminal state ψ, then the system (1.1) is said to be exactly controllable. Stabilization problem Can one find a feedback control law: f = Ku so that the resulting closed-loop system t u + u x u + 3 xu = Ku, u(x, ) = φ(x) x T, t R + is asymptotically stable as +? If such a feedback control law exists, then the system (1.1) is said to be stabilizable. The same exact control and stabilization problems hold also for the systems (1.2) and (1.3). The study of control and stabilization of the KdV equation began with the work of Russell [12] and Zhang [13] in late 198s. Both exact control problem and stabilization problem have been intensively studied since then. Significant progresses have been made due to many people s hard work and contributions in the past twenty years. In this article we intend to give an overall review of the results obtained so far in the study but with an emphasis on its recent progresses. The paper is organized as follows. In Section 2, we consider distributive control of the KdV equation posed on the periodic domain T and demonstrate how harmonic analysis and micro-local analysis play important roles in establishing exact controllability and stabilizability of the system (1.1). In Section 3, consideration is given to the boundary control system (1.2), which has some amazing properties. It is only null controllable if control input is only allowed to apply to the left end of the spatial domain; The system therefore behaves like a parabolic system. But the system is exactly controllable if control inputs are allowed to act on the right end of the spatial domain; The system thus behaves like a hyperbolic system. Moreover, depending on the length of the spatial domain, the associated linear system may not be controllable, but the nonlinear system is still exact controllable. In Section 4, the boundary control system (1.3) will be discussed. Some general concluding remarks together with a list of open problems for further investigation will be provided. 2 The KdV Equation on a Periodic Domain Consideration in this section is given to the KdV equation posed on the periodic domain T: t u + u x u + 3 xu =, x T, t R. (2.1) The equation is known to possess an infinite set of conserved integral quantities of which the first three are I 1 (t) = u(x, t)dx, and I 3 (t) = I 2 (t) = T T T u 2 (x, t)dx, ( u 2 x(x, t) 1 ) 3 u3 (x, t) dx.

CONTROL AND STABILIZATION OF THE KDV EQUATION 651 From the historical origins [4,7,14] of the KdV equation, involving the behavior of water waves in a shallow channel, it is natural to think of I 1 and I 2 as expressing conservation of volume (or mass) and energy, respectively. The Cauchy problem of the Equation (2.4) has been intensively studied for many years (see [15 18] and the references therein). The best known result [19] so far is that the Cauchy problem is well-posed in the space H s (T) for any s 1: Let s 1 and T > be given. For any u H s (T), the equation (2.1) admits a unique solution u C([, T ]; H s (T)) satisfying u(x, ) = u (x). Moreover, the corresponding solution map (u u) is continuous from the space H s (T) to the space C([, T ]; H s (T)). In this section, the Equation (2.1) will be studied from control point of view with a forcing function f = f(x, t) added to the equation as a control input by which solutions are to be influenced: t u + u x u + 3 xu = f, x T, t R, (2.2) where f is assumed to be supported in a given open set ω T. In order to keep the mass I 1 (t) conserved, the control input f(x, t) is chosen to be of the form ( ) f(x, t) = [Gh](x, t) := g(x) h(x, t) g(y)h(y, t)dy, (2.3) where h is considered as a new control input and g(x) is a given nonnegative smooth function such that ω = g > } and 2π[g] = g(x)dx = 1. For the chosen g, it is easy to see that d u(x, t)dx = dt for any solution u = u(x, t) of the system T T T f(x, t)dx =, T for any t R t u + u x u + 3 xu = Gh, (2.4) the mass of the system is therefore conserved. Exact control problem Given an initial state u and a terminal state u 1 in a certain space, can one find an appropriate control input h so that the Equation (2.4) admits a solution u which coincides with u at time t = and with u 1 at time t = T? Stabilization problem Can one find a feedback control law: h = Ku so that the resulting closed-loop system t u + u x u + 3 xu = GKu, x T, t R + is asymptotically stable as +? Depending on the support of the function g(x) in the domain T, there are two different control situations. If the support of the function g is the whole spatial domain T, then our control acts on the whole domain and we refer to it as global control. If the support of the function g is a proper subset of the spatial domain T, our control acts only on a sub-domain and we refer to it as local control. Obviously, we have more control power in the global control If s > 1, this solution map is, in fact, analytic. 2

652 LIONEL ROSIR BING-YU ZHANG situation than that in the local control case. On the other hand, the local control situation includes more cases of practical interest and is therefore more relevant in general. These problems for the KdV equation were first studied by Komornik, and Russell and Zhang in [2], Russell and Zhang in [21 22]. In the case of global control, they showed that the system is globally exactly controllable and globally exponentially stabilizable. Theorem 2.1 [22] Let T > and s 1 be given. Assume that g(x) β > x T. Then, for any u, u 1 H s (T) with [u ] = [u 1 ], there exists a h L 1 (, T ; H s 1 (T)) such that the system (2.4) admits a solution u C([, T ]; H s (T)) satisfying u(x, ) = u (x), u(x, T ) = u 1 (x). Theorem 2.2 [2] Let k be a given integer. There exists a γ > such that for any u H k (T), the closed-loop system t u + u x u 3 xv = GG u, u(x, ) = u (x), x T, t R admits a unique solution u C(R + ; H k (T)) which, moreover, satisfies u(, t) [u ] H k (T) α k ( u Hs (T))e γt for any t, where α : R + R + is a nondecreasing continuous function. The case of local control is much more challenging. In this situation, Russell and Zhang [21] considered first the associated linear open loop control system: Let A denote the operator t v + 3 xv = Gh, v(x, ) = v (x), x T, t R. (2.5) Aw = w with its domain D(A) = H 3 (T). The operator A generates a strongly continuous group W (t) on the space L 2 (T); The eigenfunctions are simply the orthonormal Fourier basis functions in L 2 (T), φ k (x) = 1 e ikx, k =, ±1, ±2,. 2π The corresponding eigenvalue of φ k is λ k = ik 3, k =, ±1, ±2,. Note that A = A, G = G, and that W ( t) = W (t) for any t R. The linear system (2.5) is shown by Russell and Zhang [22] to be exactly controllable in H s (T) for any s. Theorem 2.3 [22] Let s and T > be given. There exists a bounded linear operator Φ : H s (T) H s (T) L 2 (, T ; H s (T)) such that for any v, v 1 H s (T), with [v ] = [v 1 ], T W (T )v + W (T t)g(φ(v, v 1 ))(t) dt = v 1

CONTROL AND STABILIZATION OF THE KDV EQUATION 653 and Φ(v, v 1 ) L 2 (,T ;H s (T)) C( v s + v 1 s ), where C > depends only on T and g s. The following estimate is a direct consequence of Theorem 2.3. Corollary 2.4 Let T > be given. There exists δ > such that T GW (t)φ 2 dt δ φ 2 for any φ L 2 (T) with [φ] =. In addition, choosing the following simple feedback law h(v) = G v. Russell and Zhang [22] showed that the resulting closed-loop system t v + 3 xv = GG v, v(x, ) = v (x), x T (2.6) is exponentially stable. Theorem 2.5 [22] Let s be given. There exists a number κ > independent of s such that for any v H s (T), the corresponding solution v of (2.6) satisfies v(., t) [v ] s Ce κt v [v ] s for any t where C > is a constant depending only on s. Note that in the above theorem, the decay rate is κ = inf Re λ, λ σ(a GG )}. (2.7) However, it is possible to choose an appropriate linear feedback law such that the decay rate of the resulting closed-loop system is as large as one desires. For given λ >, define D λ φ = 1 e 2λτ W ( τ)gg W ( τ)φ dτ for any φ H s (T). Clearly, D λ is a bounded linear operator from H s (T) to H s (T). Moreover, D λ is a self-adjoint positive operator on L 2 (T) = u L 2 (T); [u] = }, and so is its inverse D 1 λ. Choose the feedback control The resulting closed-loop system reads: with h = G D 1 λ (v [v]). t v + 3 xv = K λ (v [v]), v(x, ) = v (x), x T (2.8) K λ := GG D 1 λ. If λ =, we define K = GG. As a direct application of Theorem 2.1 in [23] due to Slemrod, we obtain the following result.

654 LIONEL ROSIR BING-YU ZHANG Theorem 2.6 Let s and λ > be given. For any v H s (T), the system (2.8) admits a unique solution v C(R + ; H s (T)). Moreover, there exists M = M s depending on s such that v(., t) [v ] s M s e λt v [v ] s for any t. To extend the above linear results to the nonlinear system (2.4), we rewrite (2.4) in its integral form u(t) = W (t)u + t W (t τ)[gh](τ)dτ For given T > and u, u 1 H s (T) with [u ] = [u 1 ], define t W (t τ)[u x u](τ)dτ. Select and consider the map ω(t, u) = T W (T τ)[u x u](τ)dτ. h = Φ(u, u 1 + ω(t, u)) Γ (u)(t) = W (t)u + t for any u C([, T ]; H s (T)). Note that W (t τ)[gφ(u, u 1 + ω(t, u))](τ)dτ Γ (u) t= = u, Γ (u) t=t = u 1 + ω(t, u) ω(t, u) = u 1. t W (t τ)[u x u](τ)dτ Thus, if one wants to show that the nonlinear system (2.4) is exactly controllable from the given initial state u to the given terminal state u 1, it suffices to prove that the map Γ is a contraction in an appropriate Banach space X T s which meets the following three requirements: (a) X T s C([, T ]; H s (T)); (b) There exists a constant C such that and t W (t)φ X T s C φ s for any φ H s (T) W (t τ)f(τ)dτ C f X T L 2 (,T ;H s (T)) for any f L 2 (, T ; H s (T)); s (c) There exists a constant C such that t W (t τ) x (uv)(τ)dτ C u X T X T s v X T s for any u, v X T s. s Theorem A Let s and T > be given and assume that (a), (b), and (c) hold. There exists a δ >. If u, u 1 H s (T) with [u ] = [u 1 ] satisfying u s δ, u 1 s δ, then one can find a control input h L 2 (, T ; H s (T)) such that the system (2.4) admits a solution u C([, T ]; H s (T)) satisfying u(x, ) = u (x), u(x, T ) = u 1 (x).

CONTROL AND STABILIZATION OF THE KDV EQUATION 655 Proof Consider the nonlinear map Γ (u)(t) = W (t)u + t for u X T s. Note that, by its definition, W (t τ)[gφ(u, u 1 + ω(t, u))](τ)dτ t ω(t, u) s t C W (t τ)[u x u](τ)dτ) t W (t τ)[u x u](τ)dτ) C([,T ];Hs (T)) X T s W (t τ)[u x u](τ)dτ C u 2 X T s for some constants C, C >. Thus, there exist positive constants C 1 and C 2 such that Γ (u) X T s C 1 ( u s + u 1 s ) + C 2 u 2 X T s for any u X T s. Let Then Γ (u) X T s R 1 = 2C 1 ( u s + u 1 s ). ( 1 2 + C 2R 1 ) R 1 3 4 R 1 if u X T s R 1 and 1 2 + C 2R 1 3 4. Consequently, if we choose δ > such that and, in addition, require that then and 1 2 + 2C 1C 2 δ 3 4 u s + u 1 s δ, Γ (u) X T s R 1 Γ (u) Γ (v) X T s α u v X T s if u, v Xs T with u X T s R 1 and v X T s R 1, where < α < 1 is a constant independent of u and v, that is to say, the map Γ is a contraction whose fixed point is the desired solution. Thus, attention should turn to find such an appropriate space X T s. However, at the time when Russell and Zhang were attacking the problem around 199, it was a common opinion then that the selection of such a space X T s was extremely difficulty if not impossible. Indeed, one cannot choose = C([, T ]; H s (T)) X T s because the requirement (c) would not be satisfied for any s. This is due to the fact that f L 2 (, T ]; H s (T)) only implies that u = t W (t τ)f(τ)dτ belongs to the space

656 LIONEL ROSIR BING-YU ZHANG C([, T ]; H s (T)). So, when u C([, T ], H s (T)), the function f = u x u belongs to the space L 2 ([, T ]; H s 1 (T)), one can only have t W (t τ)[u x u](τ)dτ C([, T ]; H s 1 (T)) at most. Certain smoothing property of the KdV equation on a periodic domain T is needed in searching such a space X T s. Many failed attempts to find smoothing properties of the KdV equation on T led people to believe that such a smoothing property may not exist. It thus came as a big surprise that, in establishing the well-posedness of the Cauchy problem, t u + u x u + 3 xu =, x T, t R, u(x, ) = u (x) in the space H s (T) for any s, Bourgain [17] discovered a very subtle smoothing property of the equation on T. In particular, Bourgain constructed a Banach space Xb,s T, now called Bourgain space associated to the KdV equation, that has the property t W (t τ)[u x u](τ)dτ C u 2 X T X T 1 1 2,s,s 2 for any s! Because of the usefulness of the Bourgain space, let us have a more detail discussion. For a function w : T R R, define the quantity ( Λ b,s (w) := k= R k 2s τ k 3 + µk 2b ŵ(k, τ) ) 1 2 2 dτ for given b, s R, where ŵ(k, τ) denotes the Fourier transform of w with respect to the space variable x and the time variable t (by contrast, ŵ(k, t) denotes the Fourier transform in space variable x) and = 1 +. 2. Moreover, denote by D r the operator defined on D (T) by D r u(k) = k rû(k), if k, û(), if k =. (2.9) The Bourgain space X b,s associated to the KdV equation posed on T is the completion of the space S(T R) under the norm := Λ b,s (v) v Xb,s for any v S(T R). For a given interval I, let X b,s (I) be the associated restriction space of X b,s to the interval I with the norm Note that for any u X b,s, w Xb,s (I) = inf w Xb,s w = w on T I }. u Xb,s = W ( t)u Hb (R,H s (T)). The following properties of the space X b,s are easily verified: (i) X b,s (I) is a Hilbert space. (ii) D r u X b,s r (I) for any u X b,s (I). (iii)if b 1 b 2 and s 1 s 2, then X b2,s 2 is continuously imbedded in the space X b1,s 1.

CONTROL AND STABILIZATION OF THE KDV EQUATION 657 (iv) For a given finite interval I, if b 1 < b 2 and s 1 < s 2, then the space X b2,s 2 (I) is compactly imbedded in the space X b1,s 1 (I). For simplicity, we denote X b,s (I) by Xb,s T if I = (, T ). The space XT b,s possesses the following important properties. Lemma 2.7 Let b, s R with b > 1/2 and T > be given. There exists a constant C > depending only on b such that (i) for any φ H s (T), W (t)φ X T b,s C φ s ; (ii) For any f X T b 1,s, t W (t τ)f(τ)dτ C f X T X T b 1,s. b,s Lemma 2.8 (Bilinear estimates) Let s, T >, and u, v X T 1 2,s L2 (, T ; L 2 (T)). Then, the function f(t) = t W (t τ)(uv) x(τ) dτ belongs to the space C([, T ]; H(T)) s X T 1,s, 2 and there exist some constants θ > and C > independent of T (for T ) and u, v such that (uv) x X T CT θ u X 1 T 2,s 12,s v X T 12, (2.1),s t W (t τ)(uv) x (τ) dτ C u X C([,T ];Hs (T)) X T T 12 v X T 1 2,s,s 12 (2.11),s. In the above lemma, H s (T) = f H s (T); [f] = } and H (T) = L 2 (T). Thus, for given s and T >, the space X T s = X T 1 2,s C([, T ]; Hs (T)) meets all the three requirements (a), (b) and (c) listed earlier. The following local controllability result due to Russell and Zhang [22] follows consequently. Theorem 2.9 [22] Let s and T > be given. There exists a δ >. If u, u 1 H s (T) with [u ] = [u 1 ] satisfying u s δ, u 1 s δ, then one can find a control input h L 2 (, T ; H s (T)) such that the system (2.4) admits a solution u C([, T ]; H s (T)) satisfying u(x, ) = u (x), u(x, T ) = u 1 (x). In order to stabilize the system (2.4), Russell and Zhang [22] selected h(x, t) = G u(x, t). (2.12) The resulting closed-loop system t u + u x u + xu 3 = GG u, x T, t R (2.13) is locally exponentially stable.

658 LIONEL ROSIR BING-YU ZHANG Theorem 2.1 [22] Let s be given. There exist positive constants M, δ, and γ such that if u H s (T) satisfies u [u ] s δ, (2.14) then the corresponding solution u of (2.13) satisfies u(, t) [u ] s Me γt u [u ] s for any t. As in the proof of Theorem 2.9, the Bourgain space X T s plays a key role in the proof of Theorem 2.1. Thus, one can always find an appropriate control input h to guide the system (2.4) from a given initial state u to a terminal state u 1 so long as their amplitude are small and [u ] = [u 1 ]. A question arises naturally. Question 2.1 Can one still guide the system by choosing appropriate control input h from a given initial state u to a given terminal state u 1 when u or u 1 have large amplitude? As for the closed-loop system (2.13), its small amplitude solutions decay at a uniform exponential rate to the corresponding constant state [u ] with respect to the norm in the space H s (T) as t. One may wonder: Question 2.2 Do the large amplitude solutions of the closed-loop system (2.13) decay exponentially as t? A further question is Question 2.3 For any given number λ >, can we design a linear feedback control law such that the exponential decaying rate of the resulting closed-loop system is λ? All these three questions have been studied recently by Laurent, Rosier and Zhang and positive answers have been provided in [24]. Theorem 2.11 [24] Let s and R > be given. There exists a T >. If u, u 1 H s (T) with [u ] = [u 1 ] satisfy u s R, u 1 s R, then one can find a control input h L 2 (, T ; H s (T)) such that the system (2.4) admits a solution u C([, T ]; H s (T)) satisfying u(x, ) = u (x), u(x, T ) = u 1 (x). Remark 2.12 In Theorem 2.9, the control time T is independent of the initial state u and the terminal state u 1 and can be, in fact, chosen arbitrarily small. By contrast, in Theorem 2.11, the control time T depends on the size of the initial state u and the terminal state u 1 in the space L 2 (T). Such a global controllability is called large time global controllability. Theorem 2.13 [24] Let s be given. There exists a constant κ > such that for any u H s (T), the corresponding solution u of the system (2.13) satisfies u(, t) [u ] s α s ( u )e κt u [u ] s for all t, (2.15) where α s : R + R + is a nondecreasing continuous function depending on s. The decay rate κ in this theorem has an upper bound where A G is the operator defined by κ inf Re λ : λ σ p (A G )}, A G v = v GG v

CONTROL AND STABILIZATION OF THE KDV EQUATION 659 with its domain D(A G ) = H 3 (T). In order to have the decay rate κ being arbitrarily large, a different feedback control law is needed. Theorem 2.14 [24] Let λ > and s be given. There exists a number δ > and a linear bounded operator Q λ from H s (T) to H s (T) such that if one chooses the feedback control law h = Q λ u in system (2.3) (2.4), then the solution u of the resulting closed-loop system satisfies t u + u x u + 3 xu = GQ λ u, u(x, ) = u (x), x T (2.16) u(, t) [u ] s Ce λt u [u ] s for all t, whenever u s δ, C > denoting a constant independent of u. Note that this is still a local stabilization result. However, the feedback laws in Theorems 2.13 and 2.14 may be combined into a time-varying feedback law (as in [25]) ensuring a global stabilization with an arbitrary large decay rate. Theorem 2.15 [24] Let λ > and s be given. There exists a smooth map Q λ from H s (T) R to H s (T) which is periodic with respect to the second variable, and such that the solution u of the closed-loop system satisfies t u + u x u + 3 xu = GQ λ (u, t), u(, ) = u u(, t) [u ] s α s,λ ( u s )e λt u [u ] s for all t, where α s,λ : R + R + is a nondecreasing continuous function depending on s and λ. Remark 2.16 While the decay rates κ in Theorem 2.13 and λ in Theorem 2.15 are independent of u, the constants α s ( u ) or α s,λ ( u s ) are likely not uniformly bounded; i.e., it may happen that lim α s(r) = or lim α s,λ(r) =. r r As explained earlier, for the control problems discussed in this article for KdV equation, how to extend the results for the linear systems to the corresponding nonlinear systems turned out to be a challenging task. Indeed, after having published their linear results [21], Russell and Zhang had to wait for several years to present their proofs of Theorems 2.9 and 2.1 for the nonlinear systems (2.4) and (2.13) until Bourgain [17] discovered a subtle smoothing property of solutions of the KdV equation posed on a periodic domain T. This then newly discovered smoothing property of the KdV equation has played an indispensable role in the proofs of Theorems 2.9 and 2.1. By contrast, to establish the global exact controllability and stabilizability for the nonlinear system (2.13) is even more challenging. After all, the results presented in Theorems 2.9 and Theorem 2.1 are essentially linear in nature; they are more or less small perturbations of the linear results. The global results presented in Theorems 2.11, 2.13, and 2.15 are truly nonlinear and their proofs demand new tools. The needed help turns out to be certain propagation properties of compactness and regularity for the KdV equation in the framework of Bourgain space as given by the propositions below, which are inspired by those established by Laurent in [26] for the Schrödinger equation. Proposition 2.17 (Propagation of compactness) Let T > and b b 1 be given (with b > ) and suppose that u n Xb, T and f n X b, 2+2b T satisfy t u n + 3 xu n = f n

66 LIONEL ROSIR BING-YU ZHANG for n = 1, 2,. Assume that there exists a constant C > such that and that Then u n X T b, C for all n 1, (2.17) u n X T + f n b, 2+2b X T + u n b, 2+2b X T as n. (2.18) b, 1+2b In addition, assume that for some nonempty open set ω T, it holds u n strongly in L 2 (, T ; L 2 (ω)). u n strongly in L 2 loc ((, T ); L2 (T)). Proposition 2.18 (Propagation of regularity) be given. Let u Xb,r T be a solution of t u + 3 xu = f. Let T >, b < 1, r R and f X T b,r If there exists a nonempty open set ω of T such that u L 2 loc ((, T ), Hr+ρ (ω)) for some ρ with < ρ min 1 b, 1 }, 2 then u L 2 loc ((, T ), Hr+ρ (T)). The proofs of Propositions 2.17 and 2.18 can be found in [24]. Corollary 2.19 Let u X T 1 be a solution of 2, t u + 3 xu + u x u = on T (, T ). (2.19) Assume that u C (ω (, T )), where ω is a nonempty open set in T. Then u C (T (, T )). Corollary 2.2 Let ω be a nonempty open set in T and let u X T 1 be a solution of 2, t u + xu 3 + u x u =, on T (, T ), u = c, on ω (, T ), where c R denotes some constant. Then u(x, t) = c on T (, T ) In order to demonstrate how the properties of propagation of compactness and regularity together with the Bourgain space analysis play key roles in establishing the global results as described in Theorems 2.11, 2.13, and 2.15, we provide a sketch of the proof of Theorem 2.13 in the case of s =. The interested readers are referred to [24] for the proofs of the others. Note first that Theorem 2.13 with s = is a direct consequence of the following observability inequality. Proposition 2.21 Let T > and R > be given. There exists a constant β > 1 such that for any u L 2 (T) satisfying u R, the corresponding solution u of (2.13) satisfies T u 2 β Gu 2 (t)dt. (2.2)

CONTROL AND STABILIZATION OF THE KDV EQUATION 661 that Thus Indeed, if (2.2) holds, then it follows from the energy estimate u(, t) 2 = u 2 t u(, T ) 2 (1 β 1 ) u 2. u(, mt ) 2 (1 β 1 ) m u 2, Gu 2 (τ)dτ, t (2.21) which gives (2.15) by the semigroup property. Now, we present a sketch of the proof of Proposition 2.21. The sketch of the proof of Proposition 2.21 We prove the estimate (2.2) by contradiction. If (2.2) is not true, then for any n 1, (2.13) admits a solution u n X T 1 satisfying 2, and T u n () R Gu n 2 dt < 1 n u,n 2, (2.22) where u,n = u n (). Since α n := u,n R, one can choose a subsequence of α n }, still denoted by α n }, such that lim n α n = α. There are two possible cases: (i) α > and (ii) α =. (i) α > Note that the sequence u n } is bounded in both spaces L (, T ; L 2 (T)) and X T 1 2,. By Lemma 2.8, the sequence x (u 2 n)} is bounded in the space X T 1 2,. On the other hand, the space X T 1 2, is compactly imbedded in the space XT, 1. Therefore, we can extract a subsequence of u n }, still denoted by u n }, such that u n u weakly in X T 1 2,, and strongly in XT, 1, 1 2 x(u 2 n) f weakly in X T 1 2,, where u X T 1 2, and f XT 1 2,. Furthermore, it can be shown that one can extract a subsequence of u n }, still denoted by u n }, such that 1 2 x(u 2 n) converges to f strongly in X T 1, 1. 2 It follows from (2.22) that T Gu n 2 dt T Gu 2 dt =, which implies that u(x, t) = c(t) on ω (, T ) for some function c(t). Thus, passing to the limit in (2.13), we obtain t u + xu 3 = f, on T (, T ), (2.23) u = c(t), on ω (, T ). Let w n = u n u and f n = 1 2 x(u 2 n) f K u n. One can show that

662 LIONEL ROSIR BING-YU ZHANG and f n t w n + 3 xw n = f n, w n, X T 1 L 2 (,T ;L 2 ( ω)) 2, 1 where ω := g > g L (T)/2}. Applying Proposition 2.17 with b = 1 2 and b = yields that w n. L 2 loc ((,T );L2 (T)) Consequently, u 2 n tends to u 2 in L 1 loc ((, T ); L1 (T)) and x (u 2 n) tends to x (u 2 ) in the distributional sense. Therefore, f = 1 2 x(u 2 ) and u X T 1 satisfies 2, t u + xu 3 + 1 2 x(u 2 ) =, on T (, T ), u = c(t), on ω (, T ). The first equation gives c (t) = which, combined to Corollary 2.2, yields that u(x, t) c for some constant c R. Since [u] =, c =, and u n converges strongly to in L 2 loc ((, T ), L2 (T)). We can pick some time t [, T ] such that u n (t ) tends to strongly in L 2 (T). Since t u n () 2 = u n(t ) 2 + Gu n 2 dt, it is inferred that α n = u n (), which is a contradiction to the assumption α >. (ii) α =. Note first that α n > for all n. Set v n = u n /α n for all n 1. Then and Because of t v n + 3 xv n + µ x v n + K v n + α n 2 x(v 2 n) = T Gv n 2 dt < 1 n. (2.24) v n () = 1, (2.25) the sequence v n } is bounded in both spaces L (, T ; L 2 (T)) and X T 1 2,. We can extract a subsequence of v n }, still denoted by v n }, such that v n v weakly in the space X T 1 and 2, strongly in the spaces X T 1 2, 1 and XT, 1. Moreover, the sequence x (vn)} 2 is bounded in the space X T 1,, and therefore α n x (vn) 2 tends to in the space X T 2 1,. Finally, T Gv 2 dt =. 2 Thus, v solves t v + xv 3 =, on T (, T ) (2.26) v = c(t), on ω (, T ). Once again, we infer that v(x, t) = c(t) = c by Holmgren Theorem, and that c = because of [v] =.

CONTROL AND STABILIZATION OF THE KDV EQUATION 663 According to (2.24), T Gv n 2 dt and so K v n converges strongly to in X T 1 2, 1. Then, an application of Proposition 2.17 as in (i) shows that v n converges to in L 2 loc ((, T ), L2 (T)). Thus, we can pick a time t (, T ) such that v n (t ) converges to strongly in L 2 (T). Since t v n () 2 = v n(t ) 2 + Gv n 2 dt, we infer from (2.24) that v n (), which is a contradiction to (2.25). 3 The KdV Equation on a Finite Interval (, L) In this section our attention is focused on the KdV equation posed on a finite interval (, L) with the Dirichlet boundary conditions: t u + x u + u x u + xu 3 =, x (, L), t >, (3.1) u(, t) = h 1 (t), u(l, t) = h 2 (t), x u(l, t) = h 3 (t). Considering the boundary value functions h j, j = 1, 2, 3, as control inputs, we are mainly concerned with the controllability of the control system (3.1). The boundary control problem of the KdV equation was first studied by Rosier [27] who considered the system (3.1) with only one boundary control input h 3 (with h 1 = h 2 ) in action. He showed that the system (3.1) is locally exactly controllable in the space L 2 (, L). Theorem 3.1 [27] Let T > be given and assume j2 + l L / N := 2π 2 + jl ; j, l N }. (3.2) 3 There exists a δ >. If φ, ψ L 2 (, L) with φ L 2 (,L) + ψ L 2 (,L) δ, then one can find a control input h 3 L 2 (, T ) such that the system (3.1) (with h 1 = h 2 ) admits a solution u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x). The theorem was first proved for the associated linear system using the Hilbert Uniqueness Method without the smallness assumption on the initial state φ and the terminal state ψ. The linear result was then extended to the nonlinear system to obtain Theorem 3.1 by using the contraction mapping principle. Remark 3.2 In the case of L N, Rosier [27] demonstrated that the associated linear system t u + x u + 3 xu =, x (, L), t >, u(, t) =, u(l, t) =, x u(l, t) = h(t) (3.3)

664 LIONEL ROSIR BING-YU ZHANG is not exactly controllable; There exists a finite dimensional subspace M of L 2 (, L) which is not reachable by the system (3.3) when starting from the origin. More precisely, for any ψ M, the solution u of (3.3) satisfies u(x, ) =, u(x, T ) ψ(x) for any control input h L 2 (, T ). From now on a spatial domain (, L) is called critical if its domain length L N. Remark 3.3 However, if one is allowed to use both h 2 and h 3 as control inputs, L. Rosier has pointed out in [27] that both the linear system t u + x u + xu 3 =, x (, L), t >, u(, t) =, u(l, t) = h 2 (t), x u(l, t) = h 3 (t) and the corresponding nonlinear system t u + x u + u x u + 3 xu =, x (, L), t >, u(, t) =, u(l, t) = h 2 (t), x u(l, t) = h 3 (t) are locally exactly controllable in the space L 2 (, L) even if L N. When the spacial domain (, L) is critical, as the linear system (3.3) is not exactly controllable, one usually would not expect the corresponding nonlinear system t u + x u + u x u + xu 3 =, x (, L), t >, (3.4) u(, t) =, u(l, t) =, x u(l, t) = h(t) to be exactly controllable. However, Coron and Crépeau [28] showed surprisingly that the system (3.4) with L = 2kπ N (k = 1, 2, ) is locally exactly controllable in the space L 2 (, L) though its associated linear system (3.3) is not exactly controllable! Theorem 3.4 [28] Let T > and L = 2kπ N for some positive integer k. There exists a δ >. If φ, ψ L 2 (, L) with φ L 2 (,L) + ψ L 2 (,L) δ, then one can find a boundary control input h L 2 (, T ) such that the system (3.4) admits a solution u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x). Note that when the spacial domain length L = 2kπ N for some integer k >, the unreachable subspace M of the linear system (3.3) is only one-dimension. Later, Cerpa [29] considered the case where the unreachable space M of the linear system (3.3) is two-dimensional, and showed that the nonlinear system is also locally large time exactly controllable in the space L 2 (, L). Let N k2 + kl + l := 2π 2 ; (k, l) N N satisfying k > l and 3 m, n N \ k}, k 2 + kl + l 2 m 2 + mn + n 2 }.

CONTROL AND STABILIZATION OF THE KDV EQUATION 665 Obviously, N N. Moreover, the corresponding unreachable subspace M is two-dimensional if L N. Theorem 3.5 [29] Let L N be given. There exists T L > such that for any T > T L, there is a δ >. If φ, ψ L 2 (, L) with φ L2 (,L) + ψ L2 (,L) δ, then one can find a boundary control input h L 2 (, T ) such that the system (3.4) admits a solution u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x). Note that in Theorem 3.4, the control time T needed can be arbitrarily small while in Theorem 3.5, the needed control time T has to be large enough. That is why we call the exact controllability presented in Theorem 3.5 as large time controllability. According to Theorems 3.4 and 3.5, the system (3.4) is still locally exactly controllable when the unreachable subspace M of the associated linear system is either one- or two-dimensional. However, the dimension of M though finite can be as large as one wants by choosing appropriate L N. A question left open in [28 29] is if the system (3.4) is still locally exactly controllable for any other L N. An affirmative answer is provided by Cerpa and Crépeau [3] recently. Theorem 3.6 [3] Given L N, there exists a T L > such that for any T > T L, there is a δ > for which for any φ, ψ L 2 (, L) with φ L 2 (,L) + ψ L 2 (,L) δ, one can find a boundary control input h L 2 (, T ) such that the system (3.4) admits a solution satisfying u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) u(x, ) = φ(x), u(x, T ) = ψ(x). Remark 3.7 Not like the local controllability result as stated in Theorem 3.1, which is a small perturbation of the linear controllability and therefore essentially a linear result, the controllability results presented in Theorems 3.4 3.6 are truly nonlinear. Now, we return to consider the system (3.1) with more boundary control inputs in action. In [31], Zhang studied the nonlinear system (3.1) with all three boundary control inputs in action and showed the system is locally exactly controllable around any smooth solution v of the KdV equation posed on the spacial domain ( ɛ, L + ɛ): t v + x v + v x v + 3 xv =, x ( ɛ, L + ɛ), t. Theorem 3.8 [31] Let L >, T > and s be given. Suppose that v C (( ɛ, L + ɛ) ( ɛ, T + ɛ)) solves the KdV equation on the region ( ɛ, L + ɛ) ( ɛ, T + ɛ): t v + x v + v x v + 3 xv =, (x, t) ( ɛ, L + ɛ) ( ɛ, T + ɛ).

666 LIONEL ROSIR BING-YU ZHANG There exists a δ >. If φ, ψ H s (, L) satisfying φ( ) v(, ) Hs (,L) δ, ψ( ) v(, T ) Hs (,L) δ, then one can find control inputs h j C([, T ]) for j = 1, 2 and h 3 L 2 (, T ) with h 1 (t) = w(, t), h 2 (t) = w(l, t), h 3 (t) = x w(l, t) for some w C([, T ]; H s (R)) L 2 (, T ; H s+1 loc (R)) such that the system (3.1) admits a solution u C([, T ]; H s (, L)) L 2 (, T ; H s+1 (, L)) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x), on the interval (, L). To prove his theorem, Zhang considered an alternative initial value control problem of the KdV equation posed on the domain R R: t w + x w + w x w + xw 3 =, (x, t) R R, (3.5) w(x, ) = h(x), x R. Let T > and s be given. For any φ, ψ H s (R), can we find a h H s (R) such that the corresponding solution u the IVP (3.5) satisfies u(x, ) = φ(x), u(x, T ) = ψ(x) on the interval (, L)? An affirmative answer to this initial value control problem leads to a positive answer to the boundary control problem asked earlier. Indeed, for any given φ, ψ H s (, L), we can extend φ and ψ to be functions in H s (R) (let us still write their extensions as φ and ψ). If we can find a h H s (R) such that the corresponding unique solution u of the IVP (3.5) satisfies w(x, ) = φ(x), w(x, T ) = ψ(x) on the interval (, L), then we can just choose h 1 (t) = w(, t), h 2 (t) = w(l, t), h 3 (t) = x w(l, t) to guide the boundary control system (3.1) from the initial state φ to the terminal state ψ. The following theorem for the initial value control problem was established in [31]. Theorem 3.9 [31] Let T > and s be given. Suppose that v C (R R) is a given function satisfying t v + x v + v x v + xv 3 = for (x, t) [, L] [, T ]. There exists a δ >. If φ, ψ H s (R) with φ( ) v(, ) H s (,L) δ, ψ( ) v(, T ) H s (,L) δ, then one can find h H s (R) such that the corresponding solution w of the IVP (3.5) satisfies w(x, ) = φ(x), w(x, T ) = ψ(x) on the interval (, L). Instead of considering the initial control problem of the KdV equation posed on the whole line R, one may consider the following distributive control of the KdV equation posed on the whole line R: In fact, h 1, h 2 H s+1 3 (, T ), and h 3 H s 3 (, T ).

CONTROL AND STABILIZATION OF THE KDV EQUATION 667 t w + x w + w x w + 3 xw = f(x, t)g(x), x R, t (, T ), (3.6) where g C (R) is a given function satisfying, if x [, L]; g(x) = 1, if x / [ 1, L + 1], and f = f(x, t) is considered as a control input: Let T > and s be given. For any φ, ψ H s (R), can we find a function f L 2 (, T ; H s (R)) such that the equation (3.6) admits a solution w C([, T ]; H s (R)) satisfying w(x, ) = φ(x), w(x, T ) = ψ(x) on the interval (, L)? An affirmative answer to this distributive control problem leads obviously to a positive answer to the boundary control problem asked earlier. There are two major advantages in these approaches dealing with boundary control problem of the KdV equation. First, by considering the KdV equation on the whole line R, one may benefit a lot from various smoothing properties of the KdV equation discovered in the past twenty years, in particular, the Bourgain smoothing property [17] for the KdV equation discovered in early 199s. Secondly, one can avoid some difficult trace regularity problems associated with the boundary value problem. On the other hand, there is a drawback to these approaches: one has to use all the three boundary control inputs. In Theorem 3.8, all three boundary control inputs have been employed. It is interesting to study the cases that less control inputs are used. In [33], Rosier considered the system (3.1) with only one boundary control input h 1 (t) acting on the left end of the spatial domain: t u + x u + u x u + xu 3 =, x (, L), t >, (3.7) u(, t) = h 1 (t), u(l, t) =, x u(l, t) =. Using a new Carleman estimate for KdV, he showed that the system (3.7) is locally exactly trajectory controllable and, in particular, is locally exactly null controllable. Theorem 3.1 [33] Let L > and T > be given. Let v C([, T ]; H 2 (, L)) C 1 ([, T ]; L 2 (, L)) H 1 (, T ; H 1 (, L)) be a given function satisfying t v + x v + v x v + xv 3 =, x (, L), t (, T ), v(l, t) = x v(l, t) =, t (, T ), v(x, ) = v (x), x (, L). Then, there exists a δ > such that for any φ H 3 (, L) with φ(l) = φ (L) = and there exists a function φ v H 3 (,L) δ, u L 2 (, T ; H 3 (, L)) H 1 (, T ; H 1 (, L)) This is the approach used recently by Rosier and Zhang [32] to establish the exact boundary controllability of the nonlinear Schrödinger equation posed on a bounded domain Ω in R n.

668 LIONEL ROSIR BING-YU ZHANG which solves t u + x u + u x u + xu 3 =, x (, L), t (, T ), u(l, t) = x u(l, t) =, t (, T ), u(x, ) = φ(x), u(x, T ) = v(x, T ), x (, L). Note that only two boundary conditions have been specified in system (3.8). To have a closed system, we have to add to these boundary conditions a third boundary condition involving the control, for instance u(, t) = h 1 (t). In that case, the control input h 1 may be taken in the class H 1 (, T ). On the other hand, if we are interested in the generation of waves by a wavemaker, a physically relevant boundary condition is ( u 1 6 u2 + u xx ) x= = h. The control input h can then be chosen in C([, T ]) (see [33]). The theorem was improved later by Glass and Guerrero [34] as far as the regularity of the trajectory is concerned. Theorem 3.11 [34] Let L > and T > be given. Let v C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) be a given function satisfying t v + x v + v x v + xv 3 =, x (, L), t (, T ), v(, t) = v(l, t) = x v(l, t) =, t (, T ), v(x, ) = v (x), x (, L). Then, there exists a δ > such that for any φ L 2 (, L) with φ v L2 (,L) δ, there exists h 1 H 1 2 ɛ (, T ) such that the system (3.7) admits a solution satisfying u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) u(x, ) = φ(x), u(x, T ) = v(x, T ). Glass and Guerrero also considered in [34] the system (3.1) using two control inputs h 1 (t) and h 2 (t): t u + x u + u x u + xu 3 =, x (, L), t >, (3.9) u(, t) = h 1 (t), u(l, t) = h 2 (t), x u(l, t) =. They showed that the system (3.9) is locally exactly controllable in the space L 2 (, L) even if the spacial domain length L is critical. Theorem 3.12 [34] Let L > and T > be given. There exists a δ > such that for any φ, ψ L 2 (, L) with φ L 2 (,L) + ψ L 2 (,L) δ, there exist h 1, h 2 L 2 (, T ) such that the system (3.9) admits a solution u C([, T ]; H 1 (, L)) L 2 (, T ; L 2 (, L)) (3.8) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x).

CONTROL AND STABILIZATION OF THE KDV EQUATION 669 More recently, Glass and Guerrero [35] considered the system (3.1) with only a single boundary control input acting on the right end of the spacial domain: t u + x u + u x u + xu 3 =, x (, L), t >, (3.1) u(, t) =, u(l, t) = h 2 (t), x u(l, t) = and showed that if the spacial domain length L / N : = L R + ; (a, b) C 2 such that ae a = be b = (a + b) and L 2 = (a 2 + ab + b 2 )}, then the system (3.1) is locally exactly controllable in the space L 2 (, L). Theorem 3.13 (Glass and Guerrero) Let T > and < L / N be given. There exists a δ > such that for any φ, ψ L 2 (, L) with φ L2 (,L) + ψ L2 (,L) δ, there exists h 2 H 1 6 (, T ) such that the system (3.1) admits a solution u C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) satisfying u(x, ) = φ(x), u(x, T ) = ψ(x). If only boundary control input h 1 (t) on the left end of the spatial domain is allowed to use, then the system (3.1) is locally null controllable in the space L 2 (, L) as demonstrated in Theorem 3.11, i.e., by choosing appropriate control input h 1, one can guide the system (3.1) from a given small initial state to zero in a given time T. One may wonder if one can guide the system from a given small initial state to another given small terminal state or more precisely, is the system (3.1) with h 2 = h 3 locally exactly controllable? It was noticed by Rosier in [33] that the answer is negative for the linearized system (i.e., without the nonlinear term u x u in KdV). The answer is still negative for the nonlinear system. Theorem 3.14 If only the boundary control input h 1 (t) is allowed to use, then the system (3.1) (with h 2 = h 3 ) is not locally exactly controllable in the space L 2 (, L). The theorem follows as a corollary from the result given below, which was obtained recently by Glass and Zhang [36], regarding interior regularity of solutions of the IBVP of the KdV equation on (, L): t u + x u + u x u + xu 3 =, x (, L), t (, T ), u(x, ) = φ(x), (3.11) u(, t) = h 1 (t), u(l, t) = x u(l, t) =. Theorem 3.15 [36] Let L > and T > be given. For any φ L 2 (, L) and h 1 H 1 3 (, T ), the unique solution u of IBVP (3.11) in the space C([, T ]; L 2 (, L)) L 2 (, T ; H 1 (, L)) also belongs to the space C ((ɛ, L) (ɛ, T )). Remark 3.16 Thus, the boundary control system (3.1), if only the left control input h 1 is in action, behaves like a parabolic system and is only null controllable. However, if the system is allowed to control from the right end of the spacial domain, then the system behaves like a hyperbolic system and is exactly controllable. Next, we turn to consider stabilization problem of the system (3.1). First, consider the case that no boundary control inputs are in action: t u + x u + u x u + xu 3 =, x (, L), t (, T ), u(x, ) = φ(x), (3.12) u(, t) =, u(l, t) = x u(l, t) =.