[IEEE 2009 17th Mediterranean Conference on Control and Automation (MED) - Thessaloniki, Greece...

6
1 Scanning control for the string equation erald Tenenbaum, Marius Tucsnak I. I NTRODUCTION It is well known that for pointwise control problems we generally have a lack of robustness with respect to the location of the actuator. More precisely, any open subset of the considered domain ([0] in our case) contains points for which controllability fails, see, for instance, [1] and references therein. A remedy which has been proposed in Berggren [2] is to consider an actuator which moves according to a prescribed law. To describe the problem introduced in [2], let α, β and ω be positive numbers and define %(t) := α + β sin(ωt) (t R). (1.1) We consider the initial and boundary value problem 2 w ∂t 2 = 2 w ∂x 2 + u(t)δ %(t) (0 < x < π, t > 0), (1.2) w(0,t)= w(π,t)=0 (t> 0), (1.3) w(x, 0) = 0, ˙ w(x, 0) = 0 (0 <x<π), (1.4) where, for every real a, δ a stands for the Dirac measure at a. The above system describes the linear vibrations of an elastic string with a pointwise scanning actuator. This means that, for very t > 0 the actuator is positioned at %(t) at instant t. It is easy to see that if α π Q and β =0 (i.e., for some fixed actuators) the above system is not approximately controllable. Based on numerical experiments, it has been conjec- tured in [2] that the system (1.2)-(1.4) is controllable (in some sense) for every α (0) and β, ω 6=0. From a theoretical point of view, related questions have been studied in Khapalov [3], [4]. More precisely, [3] gives an algorithm defining in an implicit a man- ner a class of functions % ensuring controllability of (1.2)-(1.4) and of its generalization to several space dimensions. In the one dimensional case, it has been shown in [4] that for a simpler function % (but still dif- ferent of ours), approximate controllability and weak stabilizability of (1.2)-(1.4) holds. We mention that the stabilizability of systems related (1.2)-(1.4) was previously studied in [5] and [6]. A related field which is not tackled in this work is the controllability of heat equations. In this case we cannot expect exact controllability so that the natural questions are the approximate and null-controllability. The approximate controllability for the heat equations with pointwise moving control has been studied in [7]. As far as we know, establishing the corresponding null- controllability properties is still an open question. Throughout this work we explicitly assume that βω < 1, or we make assumptions implying this inequality. This is required in our approach inasmuch we use the fact that the map t 7t + %(t) is a smooth diffeomorphism from R onto R. Our first main result shows that the system (1.2)- (1.4) is well posed in the natural energy space. Theorem 1.1: Let ω, α, β denote positive numbers such that βω < 1 and assume that α, α ± β ]0[. Then, for every w 0 H 1 0 ([0]), w 1 L 2 ([0]) and u L 2 ([0]), the initial and boundary value problem (1.2)–(1.4) admits a unique solution w C([0]; H 1 0 ([0]) C 1 ([0]; L 2 ([0])). The second main result asserts that the considered system is controllable in a quite weak sense. Theorem 1.2: There exists an absolute constant A such that, for all ε ]0, 1 2 [, β> 0, ω> 0, T> 0 satisfying βω < ε A , T > 1 β 4 ω 4 , the system (1.2)–(1.4) is approximately controllable in infinite time (in the sense that for every w 0 H 1 0 (0),w 1 L 2 (0) there exists u L 2 (0, ) such that lim τ →∞ w(·)= w 0 , lim τ →∞ ˙ w(·)= w 1 ). The remaining part of this work is organized as follows. In Section II we prove some trigonometrical inequalities which play a central rˆ ole in this work. Section III is devoted to the proof of the main results. 17th Mediterranean Conference on Control & Automation Makedonia Palace, Thessaloniki, Greece June 24 - 26, 2009 978-1-4244-4685-8/09/$25.00 ©2009 IEEE 504

Transcript of [IEEE 2009 17th Mediterranean Conference on Control and Automation (MED) - Thessaloniki, Greece...

1

Scanning control for the string equationGerald Tenenbaum, Marius Tucsnak

I. INTRODUCTION

It is well known that for pointwise control problemswe generally have a lack of robustness with respect tothe location of the actuator. More precisely, any opensubset of the considered domain ([0, π] in our case)contains points for which controllability fails, see, forinstance, [1] and references therein. A remedy whichhas been proposed in Berggren [2] is to consider anactuator which moves according to a prescribed law.To describe the problem introduced in [2], let α, βand ω be positive numbers and define

%(t) := α+ β sin(ωt) (t ∈ R). (1.1)

We consider the initial and boundary value problem

∂2w

∂t2=∂2w

∂x2+ u(t)δ%(t) (0 < x < π, t > 0), (1.2)

w(0, t) = w(π, t) = 0 (t > 0), (1.3)

w(x, 0) = 0, w(x, 0) = 0 (0 < x < π), (1.4)

where, for every real a, δa stands for the Dirac measureat a. The above system describes the linear vibrationsof an elastic string with a pointwise scanning actuator.This means that, for very t > 0 the actuator ispositioned at %(t) at instant t. It is easy to see thatif α

π ∈ Q and β = 0 (i.e., for some fixed actuators)the above system is not approximately controllable.

Based on numerical experiments, it has been conjec-tured in [2] that the system (1.2)-(1.4) is controllable(in some sense) for every α ∈ (0, π) and β, ω 6= 0.

From a theoretical point of view, related questionshave been studied in Khapalov [3], [4]. More precisely,[3] gives an algorithm defining in an implicit a man-ner a class of functions % ensuring controllability of(1.2)-(1.4) and of its generalization to several spacedimensions. In the one dimensional case, it has beenshown in [4] that for a simpler function % (but still dif-ferent of ours), approximate controllability and weakstabilizability of (1.2)-(1.4) holds. We mention that

the stabilizability of systems related (1.2)-(1.4) waspreviously studied in [5] and [6].

A related field which is not tackled in this work isthe controllability of heat equations. In this case wecannot expect exact controllability so that the naturalquestions are the approximate and null-controllability.The approximate controllability for the heat equationswith pointwise moving control has been studied in [7].As far as we know, establishing the corresponding null-controllability properties is still an open question.

Throughout this work we explicitly assume thatβω < 1, or we make assumptions implying thisinequality. This is required in our approach inasmuchwe use the fact that the map t 7→ t+ %(t) is a smoothdiffeomorphism from R onto R.

Our first main result shows that the system (1.2)-(1.4) is well posed in the natural energy space.

Theorem 1.1: Let ω, α, β denote positive numberssuch that βω < 1 and assume that α, α ± β ∈]0, π[.Then, for every w0 ∈ H1

0 ([0, π]), w1 ∈ L2([0, π]) andu ∈ L2([0, τ ]), the initial and boundary value problem(1.2)–(1.4) admits a unique solution

w ∈ C([0, τ ];H10 ([0, π]) ∩ C1([0, τ ];L2([0, π])).

The second main result asserts that the consideredsystem is controllable in a quite weak sense.

Theorem 1.2: There exists an absolute constant Asuch that, for all ε ∈]0, 1

2 [, β > 0, ω > 0, T > 0satisfying

βω <ε

A, T >

1β4ω4

,

the system (1.2)–(1.4) is approximately controllablein infinite time (in the sense that for every w0 ∈H1

0 (0, π), w1 ∈ L2(0, π) there exists u ∈ L2(0,∞)such that limτ→∞ w(·, τ) = w0, limτ→∞ w(·, τ) =w1).

The remaining part of this work is organized asfollows. In Section II we prove some trigonometricalinequalities which play a central role in this work.Section III is devoted to the proof of the main results.

17th Mediterranean Conference on Control & AutomationMakedonia Palace, Thessaloniki, GreeceJune 24 - 26, 2009

978-1-4244-4685-8/09/$25.00 ©2009 IEEE 504

II. SOME TRIGONOMETRICAL INEQUALITIES

Let α, β, ω and T be positive numbers. For t ∈ Rand ak∞k∈Z ∈ `2(Z,C), we define the function

f(t) :=∑k∈Z

akeikt sin (k%(t)) (t ∈ [−T, T ]). (2.1)

We will show, in particular, that if f vanishes inL2([−T, T ]) then all the terms of the sequence (ak)vanish.

We first introduce some notation. First recall thatthe Fourier transform of ϕ ∈ L1(R) is given by theformula

ϕ(ξ) :=∫

Rϕ(x)eixξ dx.

The above definition is extended in the usual way forϕ ∈ L2(R). We also set

g(t) :=∑k∈Z∗

akeikt (t ∈ R),

so that, for every t ∈ R we have

f(t) :=∑k∈Z

akeikt sin(k%(t)) =g(t2)− g(t1)

2i,

where we have put tj = tj(t) := t + (−1)j%(t) forj ∈ 1, 2. For the sake of notational simplicity, weomit, here and in the remaining part of this work, thevariable t in the symbols t1 and t2.

For t ∈ R and t1, t2 defined as above, we put

Jk(T ) :=∫ T

−T|g(tk)|2 dt (k = 1, 2, T > 0).

We need the following technical result:Proposition 2.1: For every bounded interval I ⊂ R

there exist two continuous functions µ±I ∈ L1(R) suchthat

µ−I (x) 6 1lI(x) 6 µ+I (x) (x ∈ R), (2.2)

µ±I (ξ) =

|I| ± 2π if ξ = 0,0 if |ξ| > 1

(ξ ∈ R), (2.3)

where 1lI is the characteristic function of I .Proof. Following Beurling [8], we defined in [9] a

pair of functions B(x), b(x) such that

b(x) 6 sgn (x) 6 B(x) (x ∈ R)∫RB(x)− sgn (x) dx =

∫Rsgn (x)− b(x) dx = 1.

For normalisation purposes, we set here B∗(x) :=B(x/2π), b∗(x) := b(x/2π). If I := [s, t] is a realinterval, we put µ−I (x) := 1

2b∗(t − x) + b∗(x − s),

µ+I (x) := 1

2B∗(t − x) + B∗(x − s). This immedi-

ately yields (2.2). Moreover, in view of the vanishingproperties of the involved Fourier transforms recalledin [9], we see that (2.3) also holds.

We next show that, for large T , the integrals J1(T )and J2(T ) behave like 2T .

Proposition 2.2: With the above notation, assumethat β, T > 0 and that

0 6 ω < min(

1β,π

T

).

Then

|J1(T )− 2T | 6 8π1− %′(T )

=8π

1− βω cos(ωT ), (2.4)

|J2(T )− 2T | 6 8π1 + %′(T )

=8π

1 + βω cos(ωT )· (2.5)

Proof. Define

ϕ1(t) :=∫ t

0

|g(s−%(s))|21−%′(s) ds (t ∈ R),

(2.6)ψ1(t) := ϕ1(t)− t1 − α (t ∈ R), (2.7)

so that

|g(t1)|2 = 1 +ψ′1(t)

1− %′(t)(t ∈ R). (2.8)

From (2.6) and (2.2), it follows that, for all real,positive t, we have

ϕ1(t) =∫ t1

−α|g(w)|2 dw >

∫R|g(w)|2µ−[−α,t1](w) dw

=∑j,k∈Z

aj ak µ−[−α,t1](j − k),

and therefore, in view of (2.3),

ϕ1(t) > t1 + α− 2π (t > 0).

Similar calculations using µ+[−α,t1] instead of µ−[−α,t1]

yieldϕ1(t) 6 t1 + α+ 2π (t > 0).

505

Applying an analogous treatment for negative t, weobtain that the last two estimates are actually valid forall real t. From (2.8), we may therefore deduce that

|ψ1(t)| 6 2π (t ∈ R). (2.9)

Now, it follows from (2.8) that

J1(T ) = 2T +∫ T

−T

ψ′1(t)1− %′(t)

dt

= 2T +[

ψ1(t)1− %′(t)

]T−T

+∫ T

−T

ψ1(t)%′′(t)1− %′(t)2

dt.

Using (2.9) we obtain that

J1(T ) 6 2T +4π

1− %′(T )+ 2π

∫ T

−T

%′′(t)1− %′(t)2

dt

(2.10)Since βT 6 π, we have %′′(t) = −βω2 sin(ωt) 6 0for t ∈ [−π, π] and so, using (2.9),∫ T

−T

|ψ1(t)%′′(t)|1− %′(t)2

dt 6 −2π∫ T

−T

%′′(t)1− %′(t)2

dt

=4π

1− %′(T )·

This and (2.10) imply (2.4). Estimate (2.5) is derivedsimilarly.

The result below gives conditions guaranteeing thatthe right-hand side of (2.1) does indeed define afunction in L2([−T, T ]).

Proposition 2.3: Let T, ω, α, β denotepositive numbers such that βω < 1 and letak∞k∈Z ∈ `2(Z,C). Then the series (2.1) convergesin L2([−T, T ]) to a function f satisfying

12T

∫ T

−T|f(t)|2 dt

6

(1 +

4πT1− (βω cosωT )2

) ∑k∈Z∗

|ak|2.

Proof. With no loss of generality we may assumethat

∑k∈Z |ak|2 = 1. We clearly have∫ T

−T|f(t)|2 dt = 1

4 J1(T ) + J2(T )−Q(T ) ,(2.11)

where

Jk(T ) :=∫ T

−T|g(tk)|2 dt (k = 1, 2),(2.12)

Q(T ) := 2<e∫ T

−Tg(t1)g(t2) dt. (2.13)

The above formulas clearly yield∫ T

−T|f(t)|2 dt 6 1

2

J1(T ) + J2(T )

.

The stated conclusion now follows from (2.4) and(2.5).

The main result of this section is the following.Proposition 2.4: Let M > 0, ε > 0 and letak∞k∈Z ∈ `2(Z,C) satisfy∑

k∈Zk2|ak|2 6 M.

Then there exists A = AM > 0 such that if βω < ε/Aand if T > 1/β4ω4 we have(

12 − ε

) ∑k∈Z∗

|ak|2

61

2T

∫ T

−T

∣∣∣∣∑k∈Z

akeikt sin (k%(t))∣∣∣∣2 dt. (2.14)

Proof. As in the proof of Theorem 2.3, we assumethat

∑k∈Z |ak|2 = 1 and use formula (2.11). We

appeal to (2.4) and (2.5) to bound J1(T ) and J2(T )from above or from below and so only need to considerQ(T ). Expanding g(t1) and inverting summations, weget

Q(T ) =∑k∈Z

ak

∫ T

−Teikt1 g(t2) dt. (2.15)

For j = 1, 2 and s ∈ R we write ϑj for the inversefunction of t 7→ tj . In other words, ϑj(s) is defined bythe formula ϑj + (−1)j%(ϑj) = s. Setting Tj := T +(−1)jβ sin(ωT ), we hence see that ϑ1(s) ∈ [−T, T ]whenever s ∈ [−T1 − α, T1 − α]. It follows that∫ T

−Teikt1g(t2) dt =

∫ T1+α

−T1−αeiksH(s) ds, (2.16)

where

H(s) :=g(t2(ϑ1(s)))1− %′(ϑ1(s))

(−T1 6 s+ α 6 T1).

(2.17)Now, let us assume that T1 = 2qπ for some q ∈ N.Then H , considered as initially defined on ] − T1 −α, T1−α], may be extended to a 4πq-periodic functionon R and consequently written as a Fourier series

H(s) =∑k∈Z

hkeiks/(2q) (s ∈ R).

506

This series converges in L2(I) for every boundedinterval I . Using (2.16) and Parseval’s formula weobtain that∑k∈Z

∣∣∣∣∣∫ T

−Teikt1g(t2) dt

∣∣∣∣∣2

= 4T 21

∑k∈Z|h2kq|2. (2.18)

It is well-known that the h2kq are the Fourier coeffi-cients of the Riemann sum

Hq(s) :=12q

∑−q6j<q

H(s+ 2πj). (2.19)

Indeed

Hq(s) =12q

∑k∈Z

hk eiks/(2q)∑

06j<2q

eiπjk/q

=∑k∈Z

h2kqeiks. (2.20)

This and (2.18) imply that

∑k∈Z

∣∣∣∣∣∫ T

−Teikt1g(t2) dt

∣∣∣∣∣2

= 2T1

∫ T1−α

−T1−α|Hq(s)|2 ds,

which, combined with (2.15), yields

|Q(T )|2 6 2T1

∫ T1−α

−T1−α|Hq(s)|2 ds.

Using the change of variables s = t1 = t1(t), weobtain that

|Q(T )|2 6 2T1

∫ T

−T|Hq(t1)|21− %′(t)dt. (2.21)

For t ∈ [−T, T ], j ∈ Z and tk = tk(t) (k = 1, 2)defined as before, we put

ξt(j) := t2 (ϑ1 (t1 + 2πj + 4πεmjq))− t2,

where m is the unique integer in [−q, q[ such that2mπ < t1 + α < 2(m+ 1)π and

εmj :=

−1 if q −m 6 j < q,

0 if − q −m 6 j < q −m,1 if − q 6 j < −q −m.

With the above notation, we have

Hq(t1) = Gq(t)

:=12q

∑−q6j<q

g(t2 + ξt(j))1− %′(t2 + ξt(j))

(−T < t 6 T )

so that, from (2.21),

|Q(T )|2 6 2T1

∫ T

−T|Gq(t)|21− %′(t) dt.

The 1-periodic function g has zero mean-value and%′ is small when ωT is small. Our strategy is to showthat, for each t, the sequence ξt(j)q−1

j=−q is well-distributed in the torus, so as to infer that Gq(t) issmall in size. To this end, we introduce the Weyl sums

σh(t) :=12q

∑−q6j<q

eihξt(j) (h ∈ Z).

We set v := j + 2εmjq, so that |t1 + 2πv| 6 T forevery t ∈ [−T, T ]. It is easily seen that

4πβω cos(ωT )1− βω cos(ωT )

6 ξ′t(j)− 2π

=4π%′(ϑ1(t1 + 2πv))

1− %′(ϑ1(t1 + 2πv))6

4πβω1− βω

·

Using the Kusmin–Landau inequality (see, for in-stance, [10], Theorem I.6.7) it follows that

|σh(t)| 6 3πqβωh

=6

T1βωh, (2.22)

provided that

ωT 6π

3, βω 6

15, 1 6 4h+ 1 6

1βω·

Write

gK(s) :=∑k6K

ak cos(ks), rK(s) := g(s)− gK(s).

Then

Gq(t) := PK(t) +RK(t) + V (t), (2.23)

where

PK(t) :=12q

∑−q6j<q

gK(t2 + ξt(j)), (2.24)

RK(t) :=12q

∑−q6j<q

rK(t2 + ξt(j)),(2.25)

(2.26)

V (t)

:=12q

∑−q6j<q

g(t2 + ξt(j))%′(t2 + ξt(j))1− %′(t2 + ξt(j))

·

507

Let K > 0 be an integer such that 4K + 1 6 1/βω.Inverting summations yields

PK(t) =∑

16|k|6K

akσk(t)eikt2 ,

and so, using (2.22),

|PK(t)|2 662

T 21 β

2ω2

∑16|k|6K

|ak|2∑

16|k|6K

1k2

612π2

T 21 β

2ω2·

It follows that∫ T

−T|PK(t)|2 dt 6

24π2T

T 21 β

2ω2· (2.27)

On the other hand, we derive from (2.25) and (2.5)that∫ T

−T|RK(t)|2 dt

6 (2T + 8π)∑|k|>K

|ak|2 62M(T + 4π)

K2· (2.28)

Finally, we plainly infer from (2.26) that, for a suitableabsolute constant C, we have∫ T

−T|V (t)|2dt 6 Cβ2ω2T. (2.29)

Choosing K equal to the integer part of 1/(4βω), T >K4 and βω/ε small enough, it follows from (2.23)combined with (2.27)–(2.29) that Q(T ) 6 εT .

III. PROOF OF THE MAIN RESULTS

Consider the following homogeneous initial andboundary value problem associated to (1.2)-(1.4)

∂2η

∂t2(x, t) =

∂2η

∂x2(x, t) (0 < x < π, t > 0), (3.1)

η(0, t) = η(π, t) = 0 (t > 0), (3.2)

η(x, 0) = η0(x), η(x, 0) = η1(x) (0 < x < a).(3.3)

The above system models the linear vibrations of anelastic string occupying the interval [0, π] and fixedat both ends. The functions η and η stand for thetransverse displacement and the transverse velocity ofthe string, respectively.

It is well-known that for every η0 ∈ H2(0, π) ∩H1

0 (0, π) and η0 ∈ H10 (Ω), the initial and boundary

value problem (3.1)–(3.3) admits a unique solution

η ∈ C([0,∞);H2(0, π) ∩ C1([0,∞);H1

0 (0, π))).

We consider the observation problem which is basedmeasuring, at each t > 0 of the transverse velocity ηat x = %(t), where % : [0,∞[→ [0, π] is the functiondefined in (1.1), i.e. we associate to (3.1)-(3.3) theoutput law

y(t) := η(%(t), t). (3.4)

The main ingredient of the proofs of our main resultsis the following proposition.

Proposition 3.1: Let τ, ω, α, β denote positive num-bers with such that βω < 1 and assume that α, α±β ∈]0, π[. Then there exists a positive constant K =K(τ, ω, α, β) such that for any η0 ∈ H2(0, π) ∩H1

0 (0, π), η1 ∈ H10 (0, π) the solution η of (3.1)–(3.4)

satisfies∫ τ

0

|y(t)|2dt 6 K(‖η0‖2H1

0 (0,π) + ‖η1‖2L2[0,π]

)(3.5)

Moreover, there exists a positive constant ω0 suchthat for any ω ∈]0, ω0[, the system (3.1)–(3.4) isapproximately observable in infinite time, i.e. if y(t) =0 for every t > 0 then η0 = η1 = 0.

Proof Let X := H10 (0, π) × L2[0, π]. Is is well

known that X is a Hilbert space if endowed with thescalar product⟨[

f1g1

],

[f2g2

]⟩=∫ π

0

df1dx

(x)df2dx

(x)dx

+∫ π

0

g1(x)g2(x)dx.

For k ∈ Z∗ and x ∈ [0, π], we write ϕk(x) :=√2/π sin(kx). Since (ϕk)k∈N∗ is an orthonormal ba-

sis in L2[0, π], simple calculations yield that the family(ϕk)k∈Z∗ defined by

ϕk =1√2

[ϕk/(ik)ϕk

](k ∈ Z∗), (3.6)

is an orthonormal basis in X . Therefore, it sufficesto prove (3.5) under the assumption that there existsN0 > 0 for such that:⟨

dη0dx

,dϕkdx

⟩L2[0,π]

= 〈η1, ϕk〉L2[0,π] = 0 (k > N0).

508

In this case the solution η of (3.1)–(3.3) is given by[η(·, t)η(·, t)

]=

1√2

∑k∈Z∗

eikt(i

k

⟨dη0dx

,dϕkdx

⟩+ 〈η1, ϕk〉

)ϕk.

It follows that

η(ρ(t), t)

=∑k∈Z∗

eikt√π

(i

k

⟨dη0dx

,dϕkdx

⟩+ 〈η1, ϕk〉

)sinkρ(t).

The above formula, (3.4) and Theorem 2.3 imply (3.5).Finally, the fact that, for small enough ω0 > 0, the

unique continuation property stated holds follows fromProposition 2.4.

We are now in a position to prove our main results.Proof of Theorem 1.1 The result follows from the

first part of Proposition 3.1 by applying the transposi-tion method of Lions and Magenes.

Proof of Theorem 1.2 The result follows from thesecond part of Proposition 3.1 by applying the HUMmethod of Lions.

REFERENCES

[1] M. Tucsnak, “Regularity and exact controllability for a beamwith piezoelectric actuator,” SIAM J. Control Optim., vol. 34,no. 3, pp. 922–930, 1996.

[2] M. Berggren, “Optimal control of time evolution systems:Controllability investigations and numerical algorithms,” Ph.D.thesis, Rice University, 1995.

[3] A. Y. Khapalov, “Controllability of the wave equation withmoving point control,” Appl. Math. Optim., vol. 31, no. 2, pp.155–175, 1995.

[4] ——, “Observability and stabilization of the vibrating stringequipped with bouncing point sensors and actuators,” Math.Methods Appl. Sci., vol. 24, no. 14, pp. 1055–1072, 2001.

[5] A. Bamberger, J. Jaffre, and J. P. Yvon, “Punctual control ofa vibrating string: numerical analysis,” Comput. Math. Appl.,vol. 4, no. 2, pp. 113–138, 1978.

[6] J. R. McLaughlin and M. Slemrod, “Scanning control of avibrating string,” Appl. Math. Optim., vol. 14, no. 1, pp. 27–47, 1986.

[7] C. Castro and E. Zuazua, “Unique continuation and controlfor the heat equation from an oscillating lower dimensionalmanifold,” SIAM J. Control Optim., vol. 43, no. 4, pp. 1400–1434 (electronic), 2004/05.

[8] A. Beurling, “Mittag-Leffler lectures on harmonic analysis,” inThe collected works of Arne Beurling. Vol. 2, ser. Contempo-rary Mathematicians. Boston, MA: Birkhauser Boston Inc.,1989, pp. xx+389, harmonic analysis, Edited by L. Carleson,P. Malliavin, J. Neuberger and J. Wermer.

[9] G. Tenenbaum and M. Tucsnak, “Fast and strongly localizedobservation for the Schrodinger equation,” Transactions of theAmerican Mathematical Society, vol. to appear, 2008. [Online].Available: http://hal.archives-ouvertes.fr/hal-00140540/en/

[10] G. Tenenbaum, Introduction a la theorie analytique et proba-biliste des nombres, 3rd ed., ser. Coll. Echelles. Paris: Belin,2008.

509