On Weak Observability For Evolution Systems with Skew-Adjoint Generators
aa r X i v : . [ m a t h . A P ] D ec ON WEAK OBSERVABILITY FOR EVOLUTION SYSTEMS WITH SKEW-ADJOINTGENERATORS
KA¨IS AMMARI † AND FAOUZI TRIKI ‡ Abstract.
In the paper we consider the linear inverse problem that consists in recovering the initial state ina first order evolution equation generated by a skew-adjoint operator. We studied the well-posedness of theinversion in terms of the observation operator and the spectra of the skew-adjoint generator. The stabilityestimate of the inversion can also be seen as a weak observability inequality. The proof of the main resultsis based on a new resolvent inequality and Fourier transform techniques which are of interest themselves.
Contents
1. Introduction 22. Main results 33. Proof of the main Theorem 2.1 44. Sufficient conditions for the spectral coercivity. 105. Application to observability of the Schr¨odinger equation 13Appendix 16Acknowledgements 17References 17
Date : December 20, 2018.2010
Mathematics Subject Classification.
Key words and phrases.
Conditional stability, weak observability, resolvent inequality, Hautus test, Shr¨odinger equation. † AND FAOUZI TRIKI ‡ Introduction
Let X be a complex Hilbert space with norm and inner product denoted respectively by } ¨ } X and x¨ , ¨y X .Let A : X Ñ X be a linear unbounded self-adjoint, strictly positive operator with a compact resolvent.Denote by D p A q the domain of A , and introduce for β P R the scale of Hilbert spaces X β , as follows: forevery β ě X β “ D p A β q , with the norm } z } β “ } A β z } X (note that 0 R σ p A q where σ p A q is the spectrumof A ). The space X ´ β is defined by duality with respect to the pivot space X as follows: X ´ β “ X ˚ β for β ą A can be extended (or restricted) to each X β , such that it becomes a bounded operator(1) A : X β Ñ X β ´ @ β P R . The operator iA generates a strongly continuous group of isometries in X denoted p e itA q t P R [28].Further, let Y be a complex Hilbert space (which will be identified to its dual space) with norm andinner product respectively denoted by || . || Y and x¨ , ¨y Y , and let C P L p X , Y q , the space of linear boundedoperators from X into Y .This paper is concerned with the following abstract infinite-dimensional dual observation system with anoutput y P Y described by the equations(2) $&% z p t q ´ iAz p t q “ , t ą ,z p q “ z P X,y p t q “ Cz p t q , t ą . In inverse problems framework the system above is called the direct problem, i.e, to determine the ob-servation y p t q “ Cz p t q of the state z p t q for given initial state z and unbounded operator A . The inverseproblem is to recover the initial state z from the knowledge of the observation y p t q for t P r , T s where T ą z P C p IR , X q defined by: z p t q “ e itA z . (3)If z is not in X , in general z p t q does not belong to X , and hence the last equation in (2) might not bedefined. We further make the following additional admissibility assumption on the observation operator C : @ T ą D C T ą @ z P X , ż T } Ce itA z } Y dt ď C T } z } X . (4)We immediately deduce from the admissibility assumption that the map from X to L loc p IR ` ; Y q that assigns y for each z , has a continuous extension to X . Therefore the last equation in (2) is now well defined for all z P X . Without loss of generality we assume that C T is an increasing function of T (if the assumption isnot satisfied we substitute C T by sup ď t ď T C T ).Since A is a self-adjoint operator with a compact resolvent, it follows that the spectrum of A is given by σ p A q “ t λ k , k P N ˚ u where λ k is a sequence of strictly increasing real numbers. We denote p φ k q k P N ˚ theorthonormal sequence of eigenvectors of A associated to the eigenvalues p λ k q k P N ˚ . EAK OBSERVABILITY 3
Let z P X zt u Ă X ÞÝÑ λ p z q P R ` be the A -frequency function defined by λ p z q “ x Az, z y X } z } ´ X , (5) “ `8 ÿ k “ λ k x z, φ k y X ˜ `8 ÿ k “ x z, φ k y X ¸ ´ . (6)We observe that z ÞÝÑ λ p z q is continuous on X zt u , and λ p φ k q “ λ k , k P N ˚ .Let C be the set of functions ψ : R ` Ñ R ˚` continuous and decreasing. Recall that if ψ P C is not boundedbelow by a strictly positive constant it satisfies lim t Ñ`8 ψ p t q “ Definition 1.1.
The system (2) is said to be weakly observable in time T ą if there exists ψ P C such thatfollowing observation inequality holds: @ z P X , ψ p λ p z qq} z } X ď ż T } Ce itA z } Y dt. (7) If ψ p t q is lower bounded, the system is said to be exactly observable. Remark 1.1.
If the system (2) is weakly observable in time T ą , it is weakly observable in any time T larger than T . The function ψ appearing in the observability inequality (7) may depends on the time T . Most of the existing works on observability inequalities for systems of partial differential equations arebased on a time domain techniques as nonharmonic series [1, 16], multipliers method [20, 21], and microlocalanalysis techniques [10, 17]. Only few of them have considered frequency domain techniques in the spirit ofthe well known Fattorini-Hautus test for finite dimensional systems [12, 13, 11, 25, 31].The wanted frequency domain test for the observability of the system (2) would be only formulated interms of the operators A , C . The time domain system (2) would be converted into a frequency domain one,and the test would involve essentially the solution in the frequency domain and the observability operator C . The frequency domain test seems to be more suitable for numerical validation and for the calibrationof physical models for many reasons: the parameters of the system are in general measured in frequencydomain; the computation of the solution is more robust and efficient in frequency domain.The objective here is to derive sufficient and if possible necessary conditions on(i) the spectrum of A , and(ii) on the action of the operator C on the associated eigenfunctions of A ,such that the closed system (2) verifies, for some T ą , sufficiently large, the inequality (7). The aim ofthis paper is to obtain Fattorini-Hautus type tests on the pair p A, C q that guarantee the weak observabilityproperty (7).The rest of the paper is organized as follows: In section 2 we present the main results of our paper relatedto the weak observability. Section 3 contains the proof of the main Theorem 2.1 based on new resolventinequality and Fourier transform techniques. In section 4 we study the relation between the spectral coercivityof the observability operator and his action on vector spaces spanned by eigenfunctions associated to closeeigenvalues. Finally, in section 5 we apply the results of the main Theorem 2.1 to boundary observability ofthe Schr¨odinger equation in a square. 2. Main results
We present in this section the main results of our paper.
Definition 2.1.
The operator C is spectrally coercive if there exist functions ε, ψ P C such that if z P X z t u satisfies (8) } Az } X } z } X ´ λ p z q ă ε p λ p z qq , KA¨IS AMMARI † AND FAOUZI TRIKI ‡ then (9) } Cz } Y ě ψ p λ p z qq} z } X . Remark 2.1.
We remark that the following relation (10) 0 ď }p A ´ λ p z q I q z } X } z } ´ X “ } Az } X } z } X ´ λ p z q holds for all z P X zt u . In addition, the equality } Az } X } z } X ´ λ p z q “ is satisfied if and only if z “ φ k forsome k P N ˚ . Now, we are ready to announce our main result.
Theorem 2.1.
The system (2) is weakly observable iff C is spectrally coercive, that is the following twoassertions are equivalent. (1) There exist ε, ψ P C such that if z P X z t u satisfying ď } Az } X } z } X ´ λ p z q ă ε p λ p z qq , then } Cz } Y ě ψ p λ p z qq} z } X . (2) The following weak observation inequality holds: @ z P X , θ ψ ˆ θ ˆ T ` λ p z q ˙˙ } z } X ď ż T } Cz p t q} Y dτ (11) for all T ě T p λ p z qq , where T p λ p z qq is the unique solution to the equation T ε ˆ θ ˆ T ` λ p z q ˙˙ “ θ , (12) and ε, ψ P C are the functions appearing in the spectral coercivity of C . The strictly positives constants θ i , i “ , , , do not depend on the parameters of the observability system. In addition, the function λ ÞÑ T p λ q is increasing. The above theorem can be viewed as a extension of several results in the literature [13, 11, 25, 31, 24].3.
Proof of the main Theorem 2.1
In order to prove our main theorem, we need to derive a sequence of preliminary results. We start withthe main tool in the proof of the theorem which is a generalized Hautus-type test.
Theorem 3.1.
The operator C P L p X , Y q , is spectrally coercive, if and only if there exist functions ψ, ε P C ,such that the following resolvent inequality holds } z } X ď inf " } Cz } Y ψ p λ p z qq , }p A ´ λI q z } X p λ ´ λ p z qq ` ε p λ p z qq * , @ λ P IR , @ z P X zt u . (13) Proof.
Let z P X zt u be fixed. A forward computation gives the following key identity: }p A ´ λI q z } X “ p λ ´ λ p z qq } z } X ` }p A ´ λ p z q I q z } X . (14)We remark that the minimum of }p A ´ λI q z } X for a fixed z with respect to λ P IR is reached at λ “ λ p z q . We first assume that C P L p X , Y q , is spectrally coercive and prove that (13) is satisfied. Let now ε, ψ P C the functions appearing in the spectral coercivity of the operator C in Definition 2.1, and consider the fol-lowing two possible cases: EAK OBSERVABILITY 5 (i) The inequality } Az } X ´ λ p z q} z } X ă ε p λ p z qq} z } X is satisfied. Then by the spectral coercivity of C ,we deduce } Cz } Y ě ψ p λ p z qq} z } X . (15)(ii) The inequality } Az } X ´ λ p z q} z } X ě ε p λ p z qq} z } X holds. Then, the identity (14) implies }p A ´ λI q z } X ě ` p λ ´ λ p z qq ` ε p λ p z qq ˘ } z } X . (16)By combining both inequalities (15) and (16), we obtain the resolvent inequality (13).We now assume that (13) holds and, we shall show that C P L p X , Y q , satisfies the spectrally coercivityin Definition 2.1. Let ε, ψ P C the functions appearing in (13), and assume that z P X zt u satisfies }p A ´ λ p z q I q z } X “ } Az } X ´ λ p z q} z } X ă ε p λ p z qq} z } X . (17)Then, we have two possibilities(i) The inequality } Cz } Y ψ p λ p z qq ď }p A ´ λI q z } X p λ ´ λ p z qq ` ε p λ p z qq holds for some λ P IR. Consequently the following spectral coercivity } Cz } Y ě ψ p λ p z qq} z } X can be trivially deduced from the resolvent identity (13).(ii) The inequality } Cz } Y ψ p λ p z qq ą }p A ´ λI q z } X p λ ´ λ p z qq ` ε p λ p z qq is valid for all λ P IR. We then deduce from the identity (14) the following inequality } Cz } Y ψ p λ p z qq ą p λ ´ λ p z qq } z } X ` }p A ´ λ p z q I q z } X p λ ´ λ p z qq ` ε p λ p z qq , @ λ P IR . Taking λ to infinity we get the wanted inequality, that is } Cz } Y ψ p λ p z qq ě } z } X , which finishes the proof of the Theorem. (cid:3) Next we use a method developed in [11] to derive observability inequalities based on resolvent inequalitiesand Fourier transform techniques. Our objective is to prove the equivalence between the resolvent inequality(13) and the weak observability (11). The proof of the Theorem is then achieved by considering the resultsobtained in Theorem 3.1.We further assume that the resolvent inequality (13) holds and shall prove the weak observability.Let χ P C p IR q be a cut off function with a compact support in p´ , q . For T ą
0, we further denote χ T p t q “ χ ˆ tT ˙ , t P R . (18)Let z P X zt u . Set z p t q “ e itA z , x “ χ T z and f “ x ´ iAx . Since z ´ iAz “
0, we have f “ χ T z . TheFourier transform of f with respect to time is given by p f p τ q “ p iτ ´ iA q p x p τ q , KA¨IS AMMARI † AND FAOUZI TRIKI ‡ where p x p τ q is the Fourier transform of x p t q . Applying (13) to p x p τ q P X zt u for λ “ τ , we obtain(19) } p x p τ q} X ď inf } C p x p τ q} Y ψ p λ p p x p τ qqq , } p f p τ q} X p τ ´ λ p p x p τ qqq ` ε p λ p p x p τ qqq + . We remark that since p x p τ q “ , we have λ p p x p τ qq “ `8 , and the inequality (19) is well justified. Next,we study how do the frequency λ p p x p τ qq behave as a function of τ . We expect that λ p p x p τ qq that is close to λ p z q , the frequency of the initial state z , and reach increases when | τ | tends to infinity.To simplify the analysis we will make some assumptions on the cut-off function χ p s q . We further assumethat χ P C p IR q satisfies the following inequalities: χ P H p´ , q , κ ` τ ď | p χ p τ q| ď κ ` τ , τ P IR , (20)where κ ą κ ą τ . We will show in the Appendix theexistence of a such function. Theorem 3.2.
Let z P X zt u , and let z p t q “ e itA z , and let p x p τ q be the Fourier transform of x p t q “ χ T p t q z p t q , where χ T p t q is the cut-off function defined by (18) , and satisfying the inequality (20) .Then, there exists a constant c “ c p χ q ą such that the following inequality λ ď λ p p x p τ qq ď | τ | ` c λ p z q , (21) holds for all τ P IR . Proof.
Recall the expression of the frequency function: λ p p x p τ qq “ x A p x p τ q , p x p τ qy X } p x p τ q} ´ X , @ τ P IR . (22)Let z “ ř `8 k “ z k φ k P X . Hence(23) p x p τ q “ `8 ÿ k “ p χ T p τ ´ λ k q z k φ k . Hence(24) λ p p x p τ qq “ `8 ÿ k “ λ k | p χ T p τ ´ λ k q| z k ˜ `8 ÿ k “ | p χ T p τ ´ λ k q| z k ¸ ´ . We first remark that λ p p x p τ qq ě λ for all τ P R , and it tends to λ p z q when T approaches 0. In order tostudy the behavior of λ p p x p τ qq when τ is large we need to derive the behavior of p χ T p s q when s tends to infinity.We start with the trivial case where τ is far away from the spectrum of A , that is τ ă λ .Let K P R ` be large enough, and set `8 ÿ k “ | p χ T p τ ´ λ k q| z k “ ÿ | τ ´ λ k |ď K | p χ T p τ ´ λ k q| z k ` ÿ | τ ´ λ k |ą K | p χ T p τ ´ λ k q| z k “ I ` I . We claim that there exists K τ ą I ď I , for all K ě K τ . We first observe that there exists r ą κ ÿ λ k ą r z k ď κ ÿ λ k ď r z k , or equivalently EAK OBSERVABILITY 7 ˆ κ κ ` ˙ ÿ λ k ą r z k ď } z } X . In fact, we have ÿ λ k ą r z k ă r ÿ λ k ą r λ k z k ď λ p z q r } z } X . (27)Hence the inequality (26) holds if(28) r “ ˆ κ κ ` ˙ λ p z q . Now by taking K “ | τ | ` r , and using the bounds (20) with p χ T p s q “ T p χ p T s q in mind, we get2 I ď κ T p ` K T q ÿ λ k ą K ` τ z k , (29) I ě κ T p ` K T q ÿ λ k ď K ` τ z k . (30)Since K ě r , inequalities (26), (29) and (30) imply2 I ď κ T p ` K T q ÿ λ k ď K ` τ λ k z k ď I . (31)Then, inequality (25) is valid for K τ “ max p τ, r q . Consequently the inequalities12 I ď } p x p τ q} X “ `8 ÿ k “ | p χ T p τ ´ λ k q| z k ď I , (32)holds for all K ě K τ .Considering now identity (24), and inequalities (32), we obtain λ p p x p τ qq ď ¨˝ ÿ | τ ´ λ k |ď K λ k | p χ T p τ ´ λ k q| z k ˛‚¨˝ ÿ | τ ´ λ k |ď K | p χ T p τ ´ λ k q| z k ˛‚ ´ ` ¨˝ ÿ | τ ´ λ k |ą K λ k | p χ T p τ ´ λ k q| z k ˛‚¨˝ ÿ | τ ´ λ k |ď K | p χ T p τ ´ λ k q| z k ˛‚ ´ “ J ` J . On the other hand we have(33) J ď p τ ` K q . In addition, using again the bounds (20), we obtain(34) J ď ˜ ÿ λ k ą τ ` K λ k z k ¸ ˜ ÿ λ k ď K ` τ z k ¸ ´ . Since K ` τ ě r , inequality (26) gives(35) ÿ λ k ď K ` τ z k ě ˆ κ κ ` ˙ ´ } z } X . Hence
KA¨IS AMMARI † AND FAOUZI TRIKI ‡ (36) J ď ˆ κ κ ` ˙ ˜ `8 ÿ k “ λ k z k ¸ ˜ `8 ÿ k “ z k ¸ ´ “ ˆ κ κ ` ˙ λ p z q . Combining inequalities (33), (36)and (35), we get λ p p x p τ qq ď | τ | ` K ` ˆ κ κ ` ˙ λ p z q . for all K ě K τ .Consequently, the proof is achieved by taking c “ κ κ ` κ κ ` . (cid:3) Remark 3.1.
The upper bound of λ p p x p τ qq obtained in Theorem 3.2 is not optimal since λ p p x p τ qq “ λ k “ λ p z q if z “ φ k . Moreover when λ max p z q “ max t λ k , k P N ˚ , x z , φ k y X “ u ă 8 , we can easily show that λ p p x p τ qq ď λ max p z q . We remark that in both cases the bounds of λ p p x p τ qq are independent of the Fourierfrequency τ . Lemma 3.1.
Let c “ } χ } L p´ , q } χ } L p´ , q , z P X zt u , and let z p t q “ e itA z , and let p x p τ q be the Fourier transformof x p t q “ χ T p t q z p t q , where χ T p t q is the cut-off function defined by (18) .Then, the following inequality ˆ ´ R ˆ c T ` λ p z q ˙˙ } z } X ď } χ } ´ L p´ , q ż R ´ R } p x p τ q} X dτ (37) holds for all R ą c T ` λ p z q . Proof.
Recall that x “ f ` iAx where f “ χ T z . By integration by parts we then have p x p τ q “ ´ iτ ´ p f p τ q ` iA p x p τ q ¯ . Consequently } p x p τ q} X “ x´ iτ ´ p f p τ q ` iA p x p τ q ¯ , p x p τ qy X . Then for any R ą
0, by Fourier-Plancherel Theorem, we have } χ } L p´ , q } z } X ď ż R ´ R } p x p τ q} X dτ ` R ˆ T } χ } L p´ , q } χ } L p´ , q ` λ p z q} χ } L p´ , q ˙ } z } X . Hence for R large enough we have ˆ ´ R ˆ T } χ } L p´ , q } χ } L p´ , q ` λ p z q ˙˙ } z } X ď } χ } ´ L p´ , q ż R ´ R } p x p τ q} X dτ, which finishes the proof of the lemma. (cid:3) Back now to the proof of the theorem. Combining inequalities (19) and (37), we find ˆ ´ R ´ c T ` λ p z q ¯˙ } z } X ď } χ } ´ L p´ , q ˜ż R ´ R } C p x p τ q} Y ψ p λ p p x p τ qqq dτ ` ż R ´ R } p f p τ q} X ε p λ p p x p τ qqq dτ ¸ . (38) EAK OBSERVABILITY 9
Applying the upper bound λ p p x p τ qq derived in Theorem 3.2, and considering the monotony of the functions ψ and ε in C , we obtain ˆ ´ R ˆ c T ` λ p z q ˙˙ } z } X ď ψ p R ` c λ p z qq } χ } L p´ , q } χ } L p´ , q ż T | Cz p t q} Y dτ ` T ε p R ` c λ p z qq } χ } L p´ , q } χ } L p´ , q } z } X , for all R ą c T ` λ p z q .Now, by taking R “ ` c T ` λ p z q ˘ , and θ “ max p c , ` c q , we find ˜ ´ T ε ` θ ` T ` λ p z q ˘˘ } χ } L p´ , q } χ } L p´ , q ¸ } z } X ď ψ ` θ ` T ` λ p z q ˘˘ } χ } L p´ , q } χ } L p´ , q ż T } Cz p t q} Y dt. Let θ “ } χ } L p´ , q } χ } L , q , and θ “ } χ } L p´ , q } χ } L , q .Then, for T ε p R ` c λ p z qq ě θ , we finally get the wanted estimate: θ ψ ˆ θ ˆ T ` λ p z q ˙˙ } z } X ď ż T } Cz p t q} Y dt. (39)Simple calculation shows that the function T ÞÑ T ε ` θ ` T ` λ p z q ˘˘ is increasing, tends to infinity when T approaches `8 , and tends to 0 when T approaches 0. Then there exists a unique value T p λ p z qq ą λ ÞÑ T p λ q is increasing. Finally, the inequality (39) isvalid for all T ě T p λ p z qq .Now, we shall prove the converse. Our strategy is to adapt the proof of Theorem 1.2 in [29] for the clas-sical exact controllability to our settings (see also [11, 24]). We further assume that the weak observabilityinequality (11) holds for some fixed ψ and ε in C . Our goal now is to show that C is indeed spectrally coercive.Let z P X , and x : “ p iA ´ iτ I q z for some τ P IR. Define x p t q “ e itA x and z p t q “ e itA z .A forward computation shows that z p t q solves the following z p t q ´ iτ z p t q “ x p t q , @ t P IR ˚` ,z p q “ z . Then z p t q “ e iτt z ` ż t e iτ p t ´ s q x p s q ds. Applying now the observability operator both sides gives Cz p t q “ e iτt Cz ` ż t e iτ p t ´ s q Cx p s q ds, whence } Cz p t q} Y ď } Cz } Y ` ż t } Cx p s q} Y ds. † AND FAOUZI TRIKI ‡ Integrating the inequality above both sides over p , T q , we obtain ż T } Cz p t q} Y dt ď T } Cz } Y ` T ż T } Cx p s q} Y ds. We deduce from the admissibility assumption (4) that ż T } Cz p t q} Y dt ď T } Cz } Y ` T C T }p A ´ τ I q z } X . Applying the weak observability inequality (11) for T “ T p λ p z qq , leads to θ ψ ˆ θ ˆ T p λ p z qq ` λ p z q ˙˙ } z } X ď T p λ p z qq} Cz } Y ` T p λ p z qq C T p λ p z qq }p A ´ τ I q z } X , for all τ P R .Since T p λ q ě T “ T p q , for all λ ě
0, we have θ ψ ˆ θ ˆ T ` λ p z q ˙˙ } z } X ď T p λ p z qq} Cz } Y ` T p λ p z qq C T p λ p z qq }p A ´ τ I q z } X , Taking τ “ λ p z q in the previous inequality implies θ T p λ p z qq C T p λ p z qq ψ ˆ θ ˆ T ` λ p z q ˙˙ } z } X ď C T p λ p z qq } Cz } X ` }p A ´ λ p z q I q z } X . Let r ψ p λ q “ θ T p λ q ψ ˆ θ ˆ T ` λ ˙˙ , r ε p λ q “ θ T p λ q C λ ψ ˆ θ ˆ T ` λ ˙˙ . We deduce from the monotonicity properties of ψ p λ q , C λ , and T p λ q that r ψ p λ q , r ε p λ q P C .Consequently C becomes spectrally coercive with the functions r ψ p λ q , r ε p λ q , that is0 ď } Az } X } z } X ´ λ p z q ă r ε p λ p z qq , implies } Cz } Y ě r ψ p λ p z qq} z } X , which finishes the proof of the Theorem.4. Sufficient conditions for the spectral coercivity.
In this section we study the relation between the spectral coercivity of the observability operator C givenin Definition 2.1, and the action of the operator C on vector spaces spanned by eigenfunctions associated toclose eigenvalues.For λ P R ` and ε ą
0, set(40) N ε p λ q “ t k P N ˚ such that | λ ´ λ k | ă ε u , to be the index function of eigenvalues of A in a ε -neighborhood of a given λ . EAK OBSERVABILITY 11
Definition 4.1.
The operator C is weakly spectrally coercive if there exist a constant ε ą and a function ψ P C such that for all λ P R , the following inequality (41) } Cz } Y ě ψ p λ q} z } X , holds for all z “ ř k P N ε p λ q z k φ k P X z t u . Lemma 4.1.
The operator C is weakly spectrally coercive iff there exist a constant ε ą and a function ψ P C such that the following inequality (42) } Cz } Y ě ψ p λ n q} z } X , holds for all z “ ř k P N ε p λ n q z k φ k , and for all n P N ˚ .Proof. Assume that C is weakly spectrally coercive. By taking λ “ λ n in (41), inequality (42) immediatelyholds. Conversely, assume that inequality (42) is satisfied, and let λ P R . One can easily check that the set N ε p λ q is either empty or it contains at least an element n P N ˚ . Since N ε p λ q Ă N ε p λ n q , we have } Cz } Y ě ψ p λ n q} z } X , holds for all z “ ř k P N ε p λ q z k φ k P X z t u . On the other hand the fact that ψ is non-increasing implies } Cz } Y ě ψ ´ λ ` ε ¯ } z } X , holds for all z “ ř k P N ε p λ q z k φ k P X z t u , which shows that C is weakly spectrally coercive with theconstant ε ą ψ p λ q : “ ψ p λ ` ε q P C . (cid:3) The Lemma 4.1 has been proved in [25] for the particular case where ψ is a constant function. Theorem 4.1.
Let ε ą be a fixed constant and let ψ P C . If C is spectrally coercive with ε, ψ , then itis weakly spectrally coercive. Conversely, if C is weakly spectrally coercive with ε, ψ , then C is spectrallycoercive.Proof. Let λ P R ` , and β ą z “ ÿ k P N β p λ q z k φ k , we have p λ p z q ´ λ q} z } X “ ÿ k P N β p λ q p λ k ´ λ q z k . Hence | λ p z q ´ λ | ă β. On the other hand } Az } X ´ λ p z q} z } X “ }p A ´ λ p z q I q z } X ď }p A ´ λI q z } X ` | λ ´ λ p z q| } z } X ă β } z } X . Then, we deduce from the spectral coercivity in Definition 2.1 that (41) holds if we choose β such that2 β ă ε .Now, we shall prove the opposite implication. Assume that (41) is satisfied for all λ P R ` , and let z “ `8 ÿ k “ z k φ k , being in X zt u , and satisfying the inequality † AND FAOUZI TRIKI ‡ } Az } X } z } X ´ λ p z q ă β p λ p z qq , (43)where β will be chosen later in terms of ε and ψ .Set pp A ´ λ p z q I q z “ f. (44)We deduce from (43), the following estimate } f } X ď β } z } X . (45)We now introduce the following orthogonal decomposition of z : z “ z ` r z, (46)with z “ ÿ k P N ε p λ p z qq z k φ k , r z “ ÿ k R N ε p λ p z qq z k φ k . (47)We deduce from (43), (44) and (45) the following estimate } r z } X “ ÿ k R N ε p λ p z qq z k “ ÿ k R N ε p λ p z qq f k p λ p z q ´ λ k q ď ε } f } X ď βε } z } X . (48)On the other hand the inequality (41) for λ “ λ p z q implies } z } X ď } Cz } Y ψ p λ p z qq . (49)The following result has been proved for admissible operator C first on p , `8q in [29], and on p , T q in[25]. Proposition 4.1.
For each ε ą and λ P IR ` , we define the subspace V p λ q Ă X by V p λ q : “ t φ k : k R N ε p λ qu , and we denote A λ : V p λ q X X Ñ X , the restriction of the unbounded operator to V p λ q .Then, there exists a constant M ą , such that } C p A λ ´ λI q ´ } L p V p λ q ,Y q ď M, @ λ P IR ` . (50)We deduce from (44) and (46), the following inequality } Cz } Y ď } Cz } Y ` } C r z } Y ď } Cz } Y ` } C p A λ p z q ´ λ p z q I q ´ f } Y . (51)Applying now the results of Proposition 4.1 on (51), we get(52) } Cz } Y ď } Cz } Y ` M } f } X . Inequalities (45) and (52), give } Cz } Y ď } Cz } Y ` M β } z } X . Now, using the inequality (49), we get
EAK OBSERVABILITY 13 } z } X ď } Cz } Y ψ p λ p z qq ` M βψ p λ p z qq } z } X . (53)Combining (48) and (53), we obtain } z } X “ } z } X ` } r z } X ď ρ p λ p z qq} z } X ` } Cz } Y ψ p λ p z qq , with(54) ρ p λ p z qq : “ ˆ Mψ p λ p z qq ` ε ˙ β p λ p z qq . By taking(55) β p λ p z qq : “ ˆ Mψ p λ p z qq ` ε ˙ ´ , we find 14 ψ p λ p z qq} z } X ď } Cz } Y . One can check easily that β p λ q belongs to C . Then C becomes spectrally coercive with the functions β p λ q , ψ p λ q P C . (cid:3) Remark 4.1.
Theorem 4.1 shows that the results of the paper [25] by M. Tucsnak and al. correspond to theparticular case of spectral coercivity where ε and ψ are constant functions. Finally, applying Proposition 4.1is not necessary to prove the theorem. In fact we can bound in inequality (51) , C by } C } p λ p z q ` ε q βε } z } X where } C } is the norm of C in L p X , Y q . Applying the results of Proposition 4.1 improves the behavior of ε p λ q for large λ . Application to observability of the Schr¨odinger equation
Let Ω “ p , π q ˆ p , π q , and B Ω be its boundary. We consider the following initial and boundary valueproblem:(56) $&% z p x, t q ` i ∆ z p x, t q “ , x P Ω , t ą ,z p x, t q “ , x P B Ω , t ą ,z p x, q “ z p x q , x P Ω . Let Γ be an open nonempty subset of B Ω. Define C to be the following boundary observability operator(57) y p x, t q “ Cz p x, t q “ B ν z | Γ , where ν is the outward normal vector on B Ω , and B ν is the normal derivative.We further show that the observation system (56)-(57) fits perfectly in the general formulation of thesystem (2).Let X “ H p Ω q be the Hilbert space with scalar product x v, w y X “ ż Ω ∇ u ¨ ∇ v dx. Therefore A “ ´ ∆ : X Ă X Ñ X , is a linear unbounded self-adjoint, strictly positive operator with acompact resolvent. Hence the operator iA generates a strongly continuous group of isometries in X denoted p e itA q t P R . Moreover for β ě X β “ D p A β q is given by † AND FAOUZI TRIKI ‡ X β “ ! φ P H p Ω q : p´ ∆ q β φ P H p Ω q ) . Then the observability operator C : X Ñ Y : “ L p Γ q , defined by (57), is a bounded operator. In additionit is known that C is an admissible observability operator, that is for any T ą C T ą
0, such that the following inequality holds ż T ż Γ |B ν z | ds p x q dt ď C T ż Ω | ∇ z | dx, for all z P X .The eigenvalues of A are(58) λ m,n “ m ` n , m, n P N ˚ . A corresponding family of normalized eigenfunctions in H p Ω q are(59) φ m,n p x q “ π ? m ` n sin p nπx q sin p mπx q , m, n P N ˚ , x “ p x , x q P Ω . Next we derive observability inequalities corresponding to different geometrical assumptions on the ob-servability set Γ.
Assumption I : We assume that Γ contains at least two touching sides of Ω.In this case it is known that Γ satisfies the geometrical assumptions of [10], and the exact controllabilityis reached [19]. We will show that it is indeed the situation by applying our coercivity test.Consider the Helmholtz equation defined by(60) $&% ∆ u ` k u “ f, x P Ω ,u “ , x P B Ω z Γ , B ν u ´ iku “ , x P Γ , where g P L p Γ q and f P L p Ω q .It has been shown using Rellich’s identities (which are somehow related to the multiplier approach inobservability [20, 21]) the following result [14]. Proposition 5.1.
Under the assumptions I on Γ , a solution u P H p Ω q to the system (60) satisfies thefollowing inequality (61) k } u } L p Ω q ` } ∇ u } L p Ω q ď c ` } f } L p Ω q ` } g } L p Γ q ˘ , for all k ě k , where k ą and c ą are constants that only depend on Γ . We deduce from Proposition 5.1 the following inequality(62) } z } X ď c p} Az ´ λ p z q z } X ` } Cz } Y q , for all z P X zt u , where λ p z q is the A -frequency of z , and c ą ε p λ q “ c , we find that C is spectrally coercive with ψ p λ q “ c , which implies in turn that thesystem (56)-(57) is exactly observable. Theorem 5.1.
Under the assumptions I on Γ , the system (56) - (57) , is exactly observable.Assumption II : We assume that Γ in a one side of Ω. Without loss of generality, we further assume thatΓ “ p , π q ˆ t u .The following result has been derived partially in [8]. EAK OBSERVABILITY 15
Proposition 5.2.
Under the assumptions II on Γ , a solution u P H p Ω q to the system (60) satisfies thefollowing inequality (63) k } u } L p Ω q ` } ∇ u } L p Ω q ď c k ` } f } L p Ω q ` } g } L p Γ q ˘ , for all k ě k , where k ą and c ą are constants that only depend on Γ . We again deduce from Proposition 5.2 the following resolvent inequality(64) } z } X ď c p ` a λ p z qq p} Az ´ λ p z q z } X ` } Cz } Y q , for all z P X zt u , where λ p z q is the A -frequency of z , and c ą ε p λ q “ c p ` λ q , we find that C is spectrally coercive with ψ p λ q “ c p ` λ q . This implies inturn that the system (56)-(57) is weakly observable: there exists a constant T ą ψ p λ p z qq} z } H p Ω q ď ż T ż Γ |B ν z | ds p x q dt, for all z P X , and for all T ě T . Theorem 5.2.
Under the assumptions II on Γ , the system (56) - (57) , is weakly observable for any z P X .Assumption III : We assume that Γ is included in a one side of Ω. Without loss of generality, we furtherassume that p α, β q ˆ t u Ă Γ Ă p , π q ˆ t u , with 0 ă α ă β ă π. Then, we have the following weakobservability inequality.
Theorem 5.3.
Under the assumptions
III on Γ , the system (56) - (57) , is weakly observable for any z P X with r ε p λ q “ Mδ Γ λ ` and r ψ p λ q “ δ Γ λ , where δ Γ ą is a constant that only depends on Γ , and M ą is theadmissibility constant appearing in Proposition 4.1. Different from the proofs in the two first cases, the proof of the weak observability in the theorem aboveis based on intrinsic properties of the eigenelements of A and the operator C . We first present the followinguseful result. Lemma 5.1.
The operator C is weakly spectrally coercive, that is, the following inequality (66) } Cz } Y ě ψ p λ m,n q} z } X , holds for all z “ ř k P N p λ m,n q z k φ k where ψ p λ q “ δ Γ λ , with δ Γ ą is a constant that only depends on Γ .Proof. Let λ m,n “ m ` n be fixed eigenvalue, and let z “ ř k P N p λ m,n q z k φ k be a fixed vector in X zt u .It is easy to check that(67) N p λ m,n q “ t k “ p p, q q P N ˚ ˆ N ˚ : p ` q “ m ` n u . Therefore } Cz } Y “ ż Γ ˇˇˇˇˇˇˇ ÿ k P N p λ m,n q z k Cφ k p x q ˇˇˇˇˇˇˇ ds p x q , ě π ż βα ˇˇˇˇˇˇ ÿ p ` q “ m ` n q p p ` q q z p,q sin p px q ˇˇˇˇˇˇ dx , (68)Based on techniques related to nonharmonic Fourier series, the following inequality has been proved inProposition 7 of [25]. † AND FAOUZI TRIKI ‡ ż βα ˇˇˇˇˇˇ ÿ p ` q “ m ` n q p p ` q q z p,q sin p px q ˇˇˇˇˇˇ dx ě ˜ δ α,β ÿ p ` q “ m ` n q p ` q | z p,q | , (69)where ˜ δ α,β ą α and β .Combining now inequalities (68) and (69), we find } Cz } Y ě δ Γ ÿ p ` q “ m ` n q p ` q | z p,q | ě δ Γ λ m,n } z } X , which achieves the proof. Here δ Γ : “ π ˜ δ α,β only depends on Γ. (cid:3) Proof of Theorem 5.3.
The result of the theorem is a direct consequence of Lemma 4.1, Theorem 4.1, andLemma 5.1. We finally obtain that C is spectrally coercive with r ε p λ q “ ´ Mψ p λ q ` ¯ ´ and r ψ p λ q “ ψ p λ q ,which finishes the proof. (cid:3) Remark 5.1.
We observe that the result of Theorem 5.2 based on clever analysis of Fourier series derivedin [8] , is indeed a particular case of Theorem 5.3 ( α “ and β “ π ) obtained from Ingham type inequalities. Appendix
Let χ P C p IR q be a cut off function with a compact support in p´ , q given by χ p s q “ p ´ | s |q e ´ | s | p´ , q . Then we have the following result.
Proposition 5.3.
The function χ p s q satisfies χ P H p´ , q , κ ` τ ď | p χ p τ q| ď κ ` τ , τ P IR , where κ ą κ are two fixed constants.Proof. Since | p χ p τ q| is even we shall prove the inequality only for τ P IR ` .A forward computation gives p χ p τ q “ ` τ ` ℜ ˆ ´ e ´p ` iτ q p ` iτ q ˙ . Then | p χ p τ q| ď ` τ . On the other hand, we have p χ p τ q “ ż IR ` p s ´ τ q sinc ´ s ¯ ds. EAK OBSERVABILITY 17
Using the estimate sinc(s) ě π for s P p , π q , we get p χ p τ q ě π ż ´ ` p s ´ τ q ds ě π p arctan p ´ τ q ` arctan p ` τ qq “ π arctan ˆ ` τ ˙ ě π « ` τ ´ ˆ ` τ ˙ ff ě π ` τ ě π ` τ , which finishes the proof. (cid:3) Acknowledgements
FT was supported by the grant ANR-17-CE40-0029 of the French National Research Agency ANR (projectMultiOnde).
References [1]
S. A. Avdonin and S. A. Ivanov,
Families of exponentials,
Cambridge University Press, Cambridge, (1995).[2]
K. Ammari and M. Choulli , Logarithmic stability in determining two coefficients in a dissipative wave equation. Exten-sions to clamped Euler-Bernoulli beam and heat equations , J. Diff. Equat., (2015), 3344–3365.[3]
K. Ammari and M. Choulli , Logarithmic stability in determining a boundary coefficient in an ibvp for the wave equation ,Dynamics of PDE., (2017) 33–45.[4] K. Ammari, M. Choulli and F. Triki , Determining the potential in a wave equation without a geometric condition.Extension to the heat equation , Proc. Amer. Math. Soc., (2016), 4381–4392.[5]
K. Ammari, M. Choulli and F. Triki,
H¨older stability in determining the potential and the damping coefficient in awave equation,
Journal of Evolution Equations, in press (2019).[6]
K. Ammari, M. Choulli and F. Triki , How to use observability inequalities to solve some inverse problems for evolutionequations? An unified approach. arXiv:1711.01779.[7]
G. Bao and Y. Kihyun,
On the stability of an inverse problem for the wave equation.
Inverse Problems, (2009), 045003.[8] G. Bao and Y. Kihyun,
Stability for the electromagnetic scattering from large cavities.
Archive for Rational Mechanicsand Analysis, (2016), 1003–1044.[9]
J.-M. Ball,
Strongly continuous semigroups, weak solutions, and the variation of constants formula, Proc. Amer. Math.Soc., (1977), 370–373.[10] C. Bardos, G. Lebeau and J. Rauch,
Sharp sufficient conditions for the observation, control, and stabilization of wavesfrom the boundary,
SIAM J. Control Optim., (1992), 1024–1065.[11] N. Burq and M. Zworski,
Control in the presence of a black box,
J. Amer. Math. Soc., (2004), 443–471.[12] H.-O. Fattorini,
Some remarks on complete controllability.
SIAM Journal on Control, (1966), 686–694.[13] M. Hautus,
Controllability and observability conditions of linear autonomous systems,
Nederl. Akad. Wetensch. Proc.Ser. A., (1969) 443–448.[14] U. Hetmaniuk,
Stability estimates for a class of Helmholtz problems,
Communications in Mathematical Sciences, (2007), 665–78.[15] V. Isakov , Inverse Problems for Partial Differential Equations,
Springer-Verlag, Berlin, 2006.[16]
V. Komornik and P. Loreti,
Fourier Series in Control Theory.
Springer Monographs in Mathematics, Springer, NewYork, 2005.[17]
C. Laurent and M. L´eautaud , Uniform observability estimates for linear waves,
ESAIM: COCV., (2016), 1097–1136.[18] D. Jerison and G. Lebeau,
Nodal sets of sums of eigenfunctions, Harmonic analysis and partial differential equations(Chicago, IL, 1996), 22–239, Chicago Lectures in Math., Univ. Chicago Press, Chicago, IL, 1999.[19]
G. Lebeau,
Contrˆole de l’´equation de Schr¨odinger,
J. Math. Pures Appl, (1992) 267–291.[20] J.-L. Lions,
Contrˆolabilit´e exacte, perturbations et stabilisation de syst`emes distribu´es.
Tome 1, Recherches enMath´ematiques Appliqu´ees, Vol. 8, Masson, Paris, 1988.[21]
J.-L. Lions,
Exact controllability, stabilization and perturbations for distributed systems.
SIAM review, (1988): 1–68. † AND FAOUZI TRIKI ‡ [22] K. Liu,
Locally distributed control and damping for the conservative systems,
SIAM J. Control Optim., (1997), 1574–1590.[23] K. Liu, Z. Liu and B. Rao,
Exponential stability of an abstract nondissipative linear system,
SIAM J. Control Optim., (2001) 149–165.[24] L. Miller,
Controllability cost of conservative systems: resolvent condition and transmutation,
Journal of FunctionalAnalysis, (2005), 425–444.[25]
K. Ramdani, T. Takahashi, G.Tenenbaum and M. Tucsnak,
A spectral approach for the exact observability of infinite-dimensional systems with skew-adjoint generator, Journal of Functional Analysis, (2005), 193–229.[26]
K. Ren and F. Triki , A Global stability estimate for the photo-acoustic inverse problem in layered media , EuropeanJournal of Applied Mathematics, (2019).[27]
W. Rudin,
Real and complex analysis,
McGraw-Hill Book Co., New York, third edition, 1987.[28]
M. Tucsnak and G. Weiss,
Observation and control for operator semigroups,
Springer Science & Business Media, 2009.[29]
D.L. Russell and G. Weiss,
A general necessary condition for exact observability,
SIAM J. Control Optim., (1994),1–23.[30] M. Yamamoto , Stability, reconstruction formula and regularization for an inverse source hyperbolic problem by a controlmethod , Inverse Probl., (1995), 481–496.[31] Q. Zhou and M. Yamamoto,
Hautus condition on the exact controllability of conservative systems,
Internat. J. Control., (1997), 371–379.[32] E. Zuazua . Controllability and observability of partial differential equations: some results and open problems.
Handbookof differential equations: evolutionary equations. Vol. 3. North-Holland, 2007. 527–621. † Universit´e de Monastir, Facult´e des Sciences de Monastir, D´epartement de Math´ematiques, UR Analyse etContrˆole des EDP, UR 13ES64, 5019, Monastir, Tunisia
E-mail address : [email protected] ‡ Laboratoire Jean Kuntzmann, UMR CNRS 5224, Universit´e Grenoble-Alpes, 700 Avenue Centrale, 38401 Saint-Martin-d’H`eres, France
E-mail address ::