Abstract
In this paper, we prove a Carleman estimate for fully discrete approximations of one-dimensional parabolic operators in which the discrete parameters h and △t are connected to the large Carleman parameter. We use this estimate to obtain relaxed observability inequalities which yield, by duality, controllability results for fully discrete linear and semilinear parabolic equations.
Similar content being viewed by others
References
Lebeau, G., Robbiano, L.: Contrôle exact de l’équation de la chaleur. Comm. Partial Differential Equations 20(1-2), 335–356 (1995). https://doi.org/10.1080/03605309508821097
Fursikov, A.V., Imanuvilov, O.Y.: Controllability of evolution equations. Lecture Notes Series. Seoul National University, Research Institute of Mathematics, Global Analysis Research Center, Seoul, vol 34 (1996)
Zuazua, E.: Propagation, observation, and control of waves approximated by finite difference methods. SIAM Rev. 47(2), 197–243 (2005). https://doi.org/10.1137/S0036144503432862
Strikwerda, J.C.: Finite difference schemes and partial differential equations, Second. Society for Industrial and Applied Mathematics (SIAM), PA, Philadelphia (2004). https://doi.org/10.1137/1.9780898717938
Morton, K.W., Mayers, D.F.: Numerical solution of partial differential equations, Second. Cambridge University Press, Cambridge, Cambridge (2005). https://doi.org/10.1017/CBO9780511812248
LeVeque, R.J.: Finite difference methods for ordinary and partial differential equations. Society for Industrial and Applied Mathematics (SIAM), PA, Philadelphia (2007). https://doi.org/10.1137/1.9780898717839, Steady-state and time-dependent problems
Isaacson, E., Keller, H.B.: Analysis of numerical methods. John Wiley & Sons, Inc., New York-London-Sydney (1966)
Ames, W.F.: Numerical methods for partial differential equations, Second. Academic Press [Harcourt Brace Jovanovich, Publishers], Thomas Nelson & Sons, London-Lagos-Melbourne, New York (1977). Computer Science and Applied Mathematics, Applications of Mathematics Series
Lopez, A., Zuazua, E.: Some new results related to the null controllability of the 1-d heat equation. In: Séminaire sur les Équations aux Dérivées Partielles, 1997–1998, p Exp. No. VIII, 22. École Polytech., Palaiseau (1998)
Labbé, S, Trélat, E: Uniform controllability of semidiscrete approximations of parabolic control systems. Systems Control Lett. 55(7), 597–609 (2006). https://doi.org/10.1016/j.sysconle.2006.01.004
Boyer, F., Hubert, F., Le Rousseau, J.: Discrete Carleman estimates for elliptic operators and uniform controllability of semi-discretized parabolic equations. J. Math. Pures Appl. (9) 93(3), 240–276 (2010). https://doi.org/10.1016/j.matpur.2009.11.003
Boyer, F., Hubert, F., Le Rousseau, J.: Discrete Carleman estimates for elliptic operators in arbitrary dimension and applications. SIAM J. Control Optim. 48(8), 5357–5397 (2010). https://doi.org/10.1137/100784278
Boyer, F., Le Rousseau, J.: Carleman estimates for semi-discrete parabolic operators and application to the controllability of semi-linear semi-discrete parabolic equations. Ann. Inst. H. Poincaré Anal. Non Linéaire 31 (5), 1035–1078 (2014). https://doi.org/10.1016/j.anihpc.2013.07.011
Nguyen, T.N.T.: Carleman estimates for semi-discrete parabolic operators with a discontinuous diffusion coefficient and applications to controllability. Math. Control Relat. Fields 4(2), 203–259 (2014). https://doi.org/10.3934/mcrf.2014.4.203
Allonsius, D., Boyer, F., Morancey, M.: Spectral analysis of discrete elliptic operators and applications in control theory. Numer. Math. 140(4), 857–911 (2018). https://doi.org/10.1007/s00211-018-0983-1
Boyer, F., Hernández-Santamaría, V, de Teresa, L.: Insensitizing controls for a semilinear parabolic equation: a numerical approach. Math. Control Relat. Fields 9(1), 117–158 (2019). https://doi.org/10.3934/mcrf.2019007
Allonsius, D., Boyer, F.: Boundary null-controllability of semi-discrete coupled parabolic systems in some multi-dimensional geometries. Math. Control Relat. Fields 10(2), 217–256 (2020). https://doi.org/10.3934/mcrf.2019037
Boyer, F.: On the penalised HUM approach and its applications to the numerical approximation of null-controls for parabolic problems. In: CANUM 2012, 41e Congrès National d’Analyse Numérique, ESAIM Proc., vol 41, EDP Sci., Les Ulis, pp. 15–58. https://doi.org/10.1051/proc/201341002 (2013)
Fattorini, H.O., Russell, D.L.: Uniform bounds on biorthogonal functions for real exponentials with an application to the control theory of parabolic equations. Quart. Appl. Math. 32, 45–69 (1974/75). https://doi.org/10.1090/qam/510972
Zheng, C.: Controllability of the time discrete heat equation. Asymptot. Anal. 59(3-4), 139–177 (2008)
Ervedoza, S., Valein, J.: On the observability of abstract time-discrete linear parabolic equations. Rev. Mat. Complut. 23(1), 163–190 (2010). https://doi.org/10.1007/s13163-009-0014-y
Boyer, F., Hernández-Santamaría, V: Carleman estimates for time-discrete parabolic equations and applications to controllability. ESAIM Control Optim. Calc. Var., 26, Paper No. 12, 43, https://doi.org/10.1051/cocv/2019072 (2020)
Hernández-Santamaría, V: Controllability of a simplified time-discrete stabilized Kuramoto-Sivashinsky system. arXiv preprint arXiv:2103.12238(2021)
Kalman, R.E., Ho, Y-C, Narendra, K.S.: Controllability of linear dynamical systems. Contributions to Differential Equations 1, 189–213 (1963)
Sontag, E.D.: Mathematical control theory. Texts in Applied Mathematics, Second, vol. 6. Springer-Verlag, New York (1998). https://doi.org/10.1007/978-1-4612-0577-7, Deterministic finite-dimensional systems
Boyer, F., Hubert, F., Le Rousseau, J.: Uniform controllability properties for space/time-discretized parabolic equations. Numer. Math. 118(4), 601–661 (2011). https://doi.org/10.1007/s00211-011-0368-1
Fernández-Cara, E, Guerrero, S.: Global Carleman inequalities for parabolic systems and applications to controllability. SIAM J. Control Optim. 45 (4), 1399–1446 (2006). https://doi.org/10.1137/S0363012904439696
Glowinski, R., Lions, J-L, He, J.: Exact and approximate controllability for distributed parameter systems. Encyclopedia of Mathematics and its Applications, vol. 117. Cambridge University Press, Cambridge (2008). https://doi.org/10.1017/CBO9780511721595, A numerical approach
Fabre, C., Puel, J-P, Zuazua, E.: Approximate controllability of the semilinear heat equation. Proc. Roy. Soc. Edinburgh Sect. A 125(1), 31–61 (1995). https://doi.org/10.1017/S0308210500030742
Fernández-Cara, E, Zuazua, E.: Null and approximate controllability for weakly blowing up semilinear heat equations. Ann. Inst. H. Poincaré Anal. Non Linéaire 17(5), 583–616 (2000). https://doi.org/10.1016/S0294-1449(00)00117-7
Lions, J-L: Contrôlabilité exacte, perturbations et stabilisation de systèmes distribués. Tome 1. Recherches en Mathématiques Appliquées [Research in Applied Mathematics], Masson, Paris, vol 8 (1988)
Münch, A, Zuazua, E.: Numerical approximation of null controls for the heat equation: ill-posedness and remedies. Inverse Problems 26(8), 085018, 39 (2010). https://doi.org/10.1088/0266-5611/26/8/085018
Fernández-Cara, E, Münch, A: Numerical null controllability of semi-linear 1-D heat equations: fixed point, least squares and Newton methods. Math. Control Relat. Fields 2(3), 217–246 (2012). https://doi.org/10.3934/mcrf.2012.2.217
Fernández-Cara, E, Münch, A: Strong convergent approximations of null controls for the 1D heat equation. SeMA J. 61, 49–78 (2013). https://doi.org/10.1007/s40324-013-0001-6
Fernández-Cara, E, Münch, A: Numerical exact controllability of the 1D heat equation: duality and Carleman weights. J. Optim. Theory Appl. 163(1), 253–285 (2014). https://doi.org/10.1007/s10957-013-0517-z
Lemoine, J., Marín-Gayte, I, Münch, A: Approximation of null controls for semilinear heat equations using a least-squares approach. arXiv preprint arXiv:2008.12656 (2020)
Lemoine, J., Münch, A: Constructive exact control of semilinear 1d heat equations. arXiv preprint arXiv:2103.09640 (2021)
Lions, J-L: Quelques notions dans l’analyse et le contrôle de systèmes à données incomplètes. In: Proceedings of the XIth Congress on Differential Equations and Applications/First Congress on Applied Mathematics (Spanish) (Málaga, 1989), Univ. Málaga, Málaga, pp. 43–54 (1990)
de Teresa, L.: Insensitizing controls for a semilinear heat equation. Comm. Partial Differential Equations 25(1-2), 39–72 (2000). https://doi.org/10.1080/03605300008821507
Lions, J-L: Some remarks on Stackelberg’s optimization. Math. Models Methods Appl. Sci. 4 (4), 477–487 (1994). https://doi.org/10.1142/S0218202594000273
Lions, J-L: Optimal control of systems governed by partial differential equations. Translated from the French by S. K. Mitter. Die Grundlehren der mathematischen Wissenschaften, Band 170. Springer-Verlag, New York-Berlin (1971)
Araruna, F.D., Fernández-Cara, E, Santos, M.C.: Stackelberg-Nash exact controllability for linear and semilinear parabolic equations. ESAIM Control Optim. Calc. Var. 21(3), 835–856 (2015). https://doi.org/10.1051/cocv/2014052
de Carvalho, P.P., Fernández-Cara, E: Numerical Stackelberg-Nash control for the heat equation. SIAM J. Sci. Comput. 42(5), A2678–A2700 (2020). https://doi.org/10.1137/19M1253320
Acknowledgements
The authors would like to thank Prof. Franck Boyer (Institut de Mathématiques de Toulouse) for some clarifying discussions about the works [11, 26]. They also would like to thank the anonymous referees for their careful reading and valuable comments and suggestions that helped to improve the contents and presentation of this paper.
Funding
This work has been supported by the project A1-S-17475 of CONACyT, Mexico, and by the National Autonomous University of Mexico, grant PAPIIT, IN100919. The work of the first author has been partially supported by the program “Estancias posdoctorales por México” of CONACyT, Mexico.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by: Enrique Zuazua
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Discrete calculus results
1.1 A.1 Results for space-discrete variables
The goal of this first section is to provide a self-contained summary of calculus rules for space-discrete operators like ∂h, \(\overline {\partial _h}\), and to provide estimates for the successive applications of such operators on the weight functions. We present the results without a proof and refer the reader to [11] (see also [13]) for a complete discussion.
To avoid introducing cumbersome notation, we introduce the following continuous difference and averaging operators
With this notation, the results given below will be naturally translated to discrete versions. More precisely, for a function f continuously defined on \({\mathbb {R}}\), the discrete function ∂hf is in fact dhf sampled on the dual mesh \(\overline {\mathfrak {M}}\), and \(\overline {\partial _h} f\) is dhf sampled on the primal mesh \(\mathfrak {M}\). Similar meanings will be used for the averaging symbols \(\tilde {f}\) and \(\overline {f}\) (see Eqs. ?? and ??, respectively) and for more intricate combinations: for instance, \(\widetilde {{\Delta }_h f}=\widetilde {\overline {\partial _h}\partial _h f}\) is the function \(\widehat {\textsf {d}_{h}\textsf {d}_{h} f}\) sampled on \(\overline {\mathfrak {M}}\).
1.1.1 A.1.1 Discrete calculus formulas
Lemma A.1
Let the functions f1,f2 be continuously defined over \({\mathbb {R}}\). We have
The translation of the result to discrete functions \(f_{1},f_{2}\in {\mathbb {R}}^{\mathfrak {M}}\) (resp. \(g_{1},g_{2}\in {\mathbb {R}}^{\overline {\mathfrak {M}}}\)) is
Lemma A.2
Let the functions f1,f2 be continuously defined over \({\mathbb {R}}\). We have
The translation of the result to discrete functions \(f_{1},f_{2}\in {\mathbb {R}}^{\mathfrak {M}}\) (resp. \(g_{1},g_{2}\in {\mathbb {R}}^{\overline {\mathfrak {M}}}\)) is
Lemma A.3
Let the function f be continuously defined over \({\mathbb {R}}\). We have
The following result provides a discrete integration-by-parts formula and a related identity for averaged functions.
Proposition A.4
Let \(f\in {\mathbb {R}}^{\mathfrak {M}\cup \partial \mathfrak {M}}\) and \(g\in {\mathbb {R}}^{\overline {\mathfrak {M}}}\). Then
Lemma A.5
Let f be a sufficiently smooth function defined in a neighborhood of \(\overline {\Omega }\). We have
-
(i)
\(\displaystyle \textsf {s}_{h}^{\pm } f = f \pm \frac {h}{2}{{\int \limits }_{0}^{1}} \partial _{x} f(\cdot \pm \sigma h/2)\text {d}{\sigma }\),
-
(ii)
\(\displaystyle \textsf {m}_{h}^{j} f = f + \frac {h^{2}}{4^{2-j}} {\int \limits }_{-1}^{1} (1-|\sigma |) {\partial _{x}^{2}} f(\cdot + l_{j}\sigma h)\text {d}{\sigma }\),
-
(iii)
\(\displaystyle \textsf {d}_{h}^{j}f ={\partial _{x}^{j}} f + \frac {h^{2}}{8^{2-j}(j+1)!}{\int \limits }_{-1}^{1}(1-|\sigma |)^{j+1}\partial _{x}^{j+2}f(\cdot +l_{j} \sigma h)\text {d}{\sigma }\), j = 1, 2, \(l_{1}=\frac {1}{2}\), l2 = 1.
Proof
The results follow from Taylor formula
at order n = 1 for item (i), n = 2 for item (ii) with j = 1, 2, n = 3 for item (iii) with j = 1 and n = 4 for item (iii) with j = 2. □
1.1.2 A.1.2 Space-discrete computations related to Carleman weights
We present here a summary of results related to space-discrete operations performed on the Carleman weigh functions. The estimates presented below are at the heart of our fully discrete Carleman estimate. The proof of such results can be found on [11].
We set r = esφ and ρ = r− 1. The positive parameters s and h will be large and small, respectively. We highlight the dependence on s,h and λ in the following results. We always assume that s ≥ 1 and λ ≥ 1.
Lemma A.6
Let \(\alpha ,\beta \in \mathbb N\). We have
Let σ ∈ [− 1, 1]. We have
Provided \(\frac {\tau h}{\delta T^{2}}\leq 1\), we have \(\partial _{x}^{\beta }\left (r(t,\cdot )(\partial _{x}^{\alpha } \rho )(t,\cdot +\sigma h)\right )=s^{\alpha } \mathcal O_{\lambda }(1)\). The same expressions hold with r and ρ interchanged if we replace s by − s everywhere.
Proposition A.7
Let \(\alpha \in \mathbb N\). Provided \(\frac {\tau h}{\delta T^{2}}\leq 1\), we have
-
(i)
\(r\textsf {m}_{h}^{j} \partial _{x}^{\alpha } \rho =r\partial _{x}^{\alpha } \rho +s^{\alpha }(sh)^{2}\mathcal O_{\lambda }(1)=s^{\alpha }\mathcal O_{\lambda }(1)\), j = 1, 2,
-
(ii)
\(r \textsf {m}_{h}^{j} \textsf {d}_{h} \rho =r\partial _{x} \rho +s(sh)^{2}\mathcal O_{\lambda }(1)\), j = 0, 1,
-
(iii)
\(r\textsf {d}_{h}^{2}\rho =r{\partial _{x}^{2}}\rho +s^{2}(sh)^{2}\mathcal O_{\lambda }(1)=s^{2}\mathcal O_{\lambda }(1)\).
The same estimates hold with ρ and r interchanged.
Proposition A.8
Let \(\alpha \in \mathbb N\). For k = 0, 1, 2, and provided \(\frac {\tau h}{\delta T^{2}}\leq 1\) we have
-
(i)
\(\textsf {d~}_{h}^{k}(r\textsf {d}_{h}^{2} \rho )={\partial _{x}^{k}}(r{\partial _{x}^{2}}\rho )+s^{2}(sh)^{2}\mathcal O_{\lambda }(1)=s^{2}\mathcal O_{\lambda }(1)\),
-
(ii)
\(\textsf {d~}_{h}^{k}(r\textsf {m}_{h}^{2} \rho )=(sh)^{2}\mathcal O_{\lambda }(1)\).
The same estimates hold with ρ and r interchanged.
Proposition A.9
Let \(\alpha ,\beta \in \mathbb N\) and k,j = 0, 1, 2. Provided \(\frac {\tau h}{\delta T^{2}}\leq 1\), we have
-
(i)
\(\textsf {m}_{h}^{j}\textsf {d~}_{h}^{k} \partial _{x}^{\alpha }\left (r^{2}\widehat {\textsf {d}_{h}\rho } \textsf {d}_{ h}^{2}\rho \right )={\partial _{x}^{k}}\partial _{x}^{\alpha } (r^{2}(\partial _{x}\rho ){\partial _{x}^{2}}\rho )+s^{3}(sh)^{2}\mathcal O_{\lambda }(1)=s^{3}\mathcal O_{\lambda }(1)\),
-
(ii)
\(\textsf {m}_{h}^{j} \textsf {d~}_{h}^{k} \partial _{x}^{\alpha }\left (r^{2}\widehat {\textsf {d}_{h} \rho } \textsf {m}_{ h}^{2}\rho \right )={\partial _{x}^{k}}\partial _{x}^{\alpha }(r\partial _{x} \rho )+s(sh)^{2}\mathcal O_{\lambda }(1)=s\mathcal O_{\lambda }(1)\).
The same estimates hold with ρ and r interchanged.
Remark A.10
We set \(\textsf {d}_{h2}:=((\textsf {s}_{h}^{+})^{2}-(\textsf {s}_{h}^{-})^{2})/2h=\textsf {m}_{h}\textsf {d}_{h}\) and \(\textsf {m}_{h2}:=((\textsf {s}_{h}^{+})^{2}+(\textsf {s}_{ h}^{-})^{2})/2\). The estimates presented in the previous results are then preserved when we replace some of the dh by dh2 and some of the mh by mh2.
1.2 A.2 Results for time-discrete variables
Following the spirit of the previous section, here, we devote to provide a summary of calculus rules manipulating the time-discrete operators Dt and \(\overline D_t\), and also to provide estimates for the application of such operators on the weight functions.
As we did before, to avoid introducing additional notation, we present the following continuous difference operator. For a function f defined on \({\mathbb {R}}\), we set
As for the the space variable, we can obtain discrete versions of the results presented below quite naturally. Indeed, for a function f continuously defined on \({\mathbb {R}}\), the discrete function Dtf amounts to evaluate Dtf at the mesh points \(\mathcal D\) and \(\overline D_t f\) is Dtf sampled at the mesh points \(\mathcal P\).
1.2.1 A.2.1 Time-discrete calculus formulas
Lemma A.11
Let the functions f1 and f2 be continuously defined over \(\mathbb R\). We have
The same holds for
From the above formulas, if f1 = f2 = f, we have the useful identities
The translation of the result to discrete functions \(f,g_{1},g_{2}\in H^{\overline {\mathcal D}}\) is
and
Of course, the above identities also hold for functions \(f,g_{1},g_{2}\in H^{\overline {\mathcal P}}\) and their respective translation operators and difference operator t± and Dt.
The following result covers discrete integration by parts and some useful related formulas.
Proposition A.12
Let {H, (⋅,⋅)H} be a real Hilbert space and consider \(u\in H^{\overline {\mathcal P}}\) and \(v\in H^{\overline {\mathcal D}}\). We have the following:
Moreover, combining the above identities, we have the following discrete integration by parts formula
Remark A.13
If we consider two functions \(f,g\in H^{\overline {\mathcal D}}\), we can combine Eqs. 8 and 10 to obtain the formula
Analogously, for \(f,g\in H^{\overline {\mathcal P}}\), the following holds
Observe that in these formulas, the integrals are taken over the same discrete points. These will be particularly useful during the derivation of the Carleman estimates Eqs. ?? and ??.
1.2.2 A.2.2 Time-discrete computations related to Carleman weights
We present some lemmas related to time-discrete operations applied to the Carleman weights. The proof of these results can be found in [22, Appendix B]. We recall that r = esφ and ρ = r− 1. We highlight the dependence on τ, δ, △t and λ in the following estimates.
Lemma A.14 (Time-discrete derivative of the Carleman weight)
Provided △tτ(T3δ2)− 1 ≤ 1, we have
Lemma A.15 (Discrete operations on the weight 𝜃)
There exists a universal constant C > 0 uniform with respect to △t, δ and T such that
-
(i)
\( |\textsf {D}_t (\theta ^{\ell })| \leq \ell T{\textsf t}^-(\theta ^{\ell +1}) + C \frac {\triangle t}{\delta ^{\ell +2}T^{2\ell +2}} , \quad \ell =1,2,\ldots \)
-
(ii)
\(\textsf {D}_t(\theta ^{\prime }) \leq C T^{2}{\textsf t}^-(\theta ^{3})+C\frac {\triangle t}{\delta ^{4}T^{5}}\).
In addition to the results presented above, since we are dealing with a fully discrete case, we need to give an additional lemma concerning the effect of the time-discrete operator Dt over some discrete operations in the space variable. This is a new result as compared to [13] and [22] but follows the arguments there. The result reads as follows.
Lemma A.16 (Mixed derivatives)
Provided τh(δT2) ≤ 1 and σ is bounded, we have
-
(i)
\(\partial _{t}^{\beta }\left (r(t,x)\partial _{x}^{\alpha }\rho (t,x+\sigma h)\right )=T^{\beta } s^{\alpha }(t) \theta ^{\beta }(t)\mathcal O_{\lambda }(1)\), β = 1, 2 and \(\alpha \in \mathbb N\).
If in addition \(\frac {\triangle t \tau }{T^{3}\delta ^{2}}\leq \frac {1}{2}\), the following estimates hold
-
(ii)
\(\textsf {D}_t(r \textsf {m}_{h}^{2} \rho )= T \textsf {t}^{-}(\theta [sh]^{2})\mathcal O_{\lambda }(1)+\left (\frac {\tau \triangle t}{\delta ^{3} T^{4}}\right )\left (\frac {\tau h}{\delta ^{} T^{2}}\right )\mathcal O_{\lambda }(1)\).
-
(iii)
\(\textsf {D}_t(r \textsf {d}_{h}^{2} \rho )= T \textsf {t}^{-}(s^{2}\theta )\mathcal O_{\lambda }(1)+ \left (\frac {\tau ^{2} \triangle t}{\delta ^{4} T^{6}}\right )\mathcal O_{\lambda }(1)+\left (\frac {\tau \triangle t}{\delta ^{3} T^{4}}\right )\left (\frac {\tau h}{\delta ^{} T^{2}}\right )^{3}\mathcal O_{\lambda }(1)\).
Proof
For β = 1, the proof of item (i) can be found on [13, Proof of Proposition 2.14]. For β = 2, the proof follows exactly as in that work just by noting that \({\partial _{t}^{2}} (r\partial _{x}^{\alpha } \rho )= T^{2} s^{\alpha } \theta ^{2} \mathcal O_{\lambda }(1)\).
To prove item (ii), we shall exploit that the weights ρ and r can be written in separated variables. Indeed, by Lemma A.5(ii), we have
hence
by a first order Taylor formula in the time variable (see formula Eq. 4). Differentiating with respect to t in Eq. 13 and item (i) yield
For estimating A2, we use the change of variable t↦t + γ△t for γ ∈ [0, 1] in item (i) and observe that provided \(\frac {\triangle t \tau }{T^{3}\delta ^{2}} \leq \frac {1}{2}\), we have \(\max \limits _{t\in [0,T+\triangle t]} \theta (t)\leq \frac {2}{\delta T^{2}}\). Therefore,
where we have used that h ≪ 1 to remove one power of h. Putting together Eqs. 14–16 and performing the change of variable \(t\mapsto t-\frac {\triangle t}{2}\) yield the desired result.
Finally, to prove item (iii), we have from Lemma A.5(iii) that
Therefore, arguing as above
Differentiating with respect to t in Eq. 17 and using item (i) yield
Using that \(\frac {\tau h}{\delta T^{2}}\leq 1\) and h ≪ 1, we can adjust some powers in the above expression and obtain
For the term B2, we argue as for A2. We use the change of variable t↦t + γ△t, γ ∈ [0, 1], in item (i) and observe that provided \(\frac {\triangle t \tau }{T^{3}\delta ^{2}}\leq \frac {1}{2}\), we have \(\max \limits _{t\in [0,t+\triangle t]}\theta (t)\leq \frac {2}{\delta T^{2}}\). In this way, we get
Collecting the estimates Eq. 19–Eq. 20 in Eq. 18 and setting the change of variable \(t\mapsto t-\frac {\triangle t}{2}\) gives the desired result. □
Appendix B: Estimates of the cross product in the Carleman estimate
1.1 B.1 Estimates that only require space discrete computations
1.1.1 B.1.1 Proof of Lemma 2.1
The proof is straightforward. First, using Eq. 8, we can relax a little bit the notation by hiding the operator \(\bar {\mathtt {t}}^{-}\). More precisely we have
Now, we can focus on the space variable. Noting that \({\Delta }_h=\overline {\partial _h}\partial _h\) and \(\overline {\partial _h}([\partial _h z]^{2})=2\overline {\partial _h}(\partial _h z)\overline {\partial _h z}\) thanks to Lemma A.1, we can integrate by parts (see formula Eq. 2) and obtain
Using that \(\partial _h(r^{2} \overline {\tilde {\rho }} \overline {\partial _h\rho })=-s\lambda ^{2}(\partial _{x}\psi )^{2}\phi +s\lambda \phi \mathcal {O}(1)+s(sh)^{2}\mathcal O_{\lambda }(1)\) (obtained from Lemma A.9(ii) and A.6) and \(r\overline {\tilde {\rho }}=1+(sh)^{2}\mathcal O_{\lambda }(1)\) (see Lemma A.7(i)), we get
The result follows by shifting back the time integral with formula Eq. 8.
1.1.2 B.1.2 Proof of Lemma 2.2
We proceed as in the previous proof. First, we shift the time integral and then using integration by parts in space, we obtain
Here, we have used that \((z_{|\partial {\Omega }})^{n-\frac {1}{2}}=0\) for n ∈⟦1,M⟧ so no boundary conditions appear.
Using Lemma A.5(iii), we see that \(\widetilde {\partial _{xx}\phi }=\partial _{xx}\phi +h\mathcal O_{\lambda }(1)\); thus, from Proposition A.7(i), we get
On the other hand, by Propositions A.7(i) and A.8, we get
Using estimates Eq. 22–Eq. 23 in Eq. 21, we see that I12 can be written as
where
Using that
in expression Eq. 24, the result follows by Cauchy-Schwarz and Young inequalities and shifting the time integral. Notice that we have adjusted some powers of the product (sh) by using that s ≥ 1 and (sh) ≤ τh(δT2)− 1 ≤ 1.
1.1.3 B.1.3 Proof of Lemma 2.3
Define \(p:=r^{2}{\Delta }_h \rho \overline {\partial _h \rho }\). From Eq. 8 and noting that \(\overline {\partial _h z}=\overline {\partial _h}(\tilde {z})\), we see that
where we have used Lemma A.1. Integrating by parts in space, we get
We observe that \(\tilde {z}_{\frac {1}{2}}=\frac {h}{2}(\partial _h z)_{\frac {1}{2}}\) and \(\tilde {z}_{N+\frac {1}{2}}=-\frac {h}{2}(\partial _h z)_{N+\frac {1}{2}}\). Thanks to A.6 and Proposition A.9(i) notice that \(p=\left [s^{2}\mathcal O_{\lambda }(1)+s^{2}(sh)^{2}\mathcal O_{\lambda }(1)\right ]r\overline {\partial _h\rho }\). Thus,
by using that sh ≤ τh(δT2)− 1 ≤ 1.
From Lemma A.2, formula Eq. 3 and recalling that \((z_{|\partial {\Omega }})^{n-\frac {1}{2}}\) for n ∈⟦1,M⟧, we have that
We claim the following.
Claim B.1
Provided τh(δT2)− 1 ≤ 1, we have
Proof
From Proposition A.9(i), we have \(\partial _h p=\partial _{x}(r^{2}(\partial _{x} \rho )\partial _{xx}\rho )+s^{3}(sh)^{2}\mathcal O_{\lambda }(1)\). On the other hand, a straightforward computation gives
Thus, the first result follows immediately.
For the second one, we note that \(\overline {\partial _h p}={\partial _h}_{\scriptscriptstyle 2} p\) (see Remark A.10), whence \(\overline {\partial _h p}= {\partial _h}_{\scriptscriptstyle 2}(r^{2}{\Delta }_h \rho \overline {\partial _h \rho })\) and using Proposition A.9(i) the results follows from Eq. 28. □
Using Claim B.1 to estimate in Eq. 27 and recalling Eq. 26, we readily get
where
As before, we have adjusted some powers of (sh) in ν21 by recalling that (sh) ≤ τh(δT2)− 1 ≤ 1 and h ≪ 1. The desired result then follows by shifting the integral in time.
1.1.4 B.1.3 Proof of Lemma 2.4
By Lemma A.3, we have that \(\overline {\tilde {z}}=z+\frac {h^{2}}{4}{\Delta }_h z\). Thus
From Proposition A.7(iii) and Lemma A.6, we readily have
whence, combining with Eq. 25, we obtain that
where \(\mu =s^{3}\lambda ^{3}\phi ^{3}\mathcal O(1)+s^{2}\mathcal O_{\lambda }(1)+s^{3}(sh)^{2}\mathcal O_{\lambda }(1)\).
For the term \(I_{22}^{(2)}\), we proceed as follows. We set pϕ := rΔhρ∂xxϕ and by using that \((z_{|\partial {\Omega }})^{n-\frac {1}{2}}=0\) for n ∈⟦1,M⟧, we get after integration by parts in space
Noting that \(\partial _h(z^{2})=2\tilde {z}\partial _h z\), we can integrate once again in the last term of the above to obtain
Let us estimate J1 and J2. For the first term, we have from Eq. 25 that \(\partial _{xx}\phi =\mathcal O_{\lambda }(1)\) and from estimate Eq. 30, we have \(p_{\phi }=s^{2}\mathcal O_{\lambda }(1)+s^{2}(sh)^{2}\O _{\lambda }(1)\). The same estimate holds for \(\widetilde {p_{\phi }}\). From this fact and adjusting some powers of (sh) yields
For estimating J2, we claim the following
Claim B.2
Provided τh(δT2)− 1 ≤ 1 we have \(h^{2}{\Delta }_h p_{\phi }=s^{2}(h+h^{2})\mathcal O_{\lambda }(1)+(sh)^{4}\mathcal O_{\lambda }(1)\).
Proof
From Eq. 25, notice that
On the other hand, a direct computation gives
From Propositions A.7(i) and A.8, estimates Eq. 34, and the fact that \((r\partial _{xx}\phi )=\partial _{x}(r\partial _{xx}\phi )=\partial _{xx}(r\partial _{xx}\phi )=s^{2} \mathcal O_{\lambda }(1)\), we obtain the desired result. □
Using Claim B.2, we readily see that
Putting together Eqs. 29, 31–32, Eqs. 33 and 35, the desired result follows by adjusting some powers of (sh).
1.2 B.2 Estimates involving time-discrete computations
1.2.1 B.2.1 Proof of Lemma 2.5
We begin by shifting the integral in time; hence,
where we have used formula Eq. 3 in the second equality. Observe that no boundary conditions appear since \(z_{|\partial {\Omega }}^{n-\frac {1}{2}}=0\) for n ∈⟦1,M⟧. Noting that
thanks to Lemma A.2, we rewrite Eq. 36 as
Integrating by parts in space the first term in the above expression and using that \(\overline {\partial _h} \tilde {p}=\overline {\partial _h p}\) for \(p\in {\mathbb {R}}^{\mathfrak {M}}\), we have
Using that \(\|\partial _h \varphi \|_{\infty }=\mathcal O_{\lambda }(1)\) and \(\partial _{x}(r\partial _{x} \rho )=s\mathcal O_{\lambda }(1)\), we can prove as in Claim B.1 that
Notice that since (sh) ≤ τh(δT2)− 1 ≤ 1, we can further simplify the above estimate and obtain that both derivatives are of order \(s\mathcal O_{\lambda }(1)\). Thus, from this remark, we have
Using that \(\theta ^{\prime }=(2t-T)\theta ^{2}\) for t ∈ [0,T], we obtain
This ends the proof.
1.2.2 B.2.2 Proof of Lemma 2.7
The proof of this term can be carried out exactly as in [22], since the space variable does not play any major role. For completeness, we sketch it briefly.
Using formula Eq. 7, we write
and integrating by parts in time using Eq. 11, we get
By definition, we have that
thus \((\theta ^{\prime })^{\frac {1}{2}}<0\) and \((\theta ^{\prime })^{M+\frac {1}{2}}>0\). Therefore, recalling that φ < 0 for all x ∈Ω, we see that the first two terms in Eq. 37 are positive and therefore can be dropped. A further computation using Eq. 38 and Lemma A.15(ii) and yields
with \(\mu _{33}=(\tau T^{2}\theta ^{3}+\frac {\tau \triangle t}{\delta ^{4}T^{5}})\mathcal O_{\lambda }(1)\) and \(\gamma _{33}=\triangle t \tau T \theta ^{2}\mathcal O_{\lambda }(1)\), where we have used that \(\varphi =\mathcal O_{\lambda }(1)\).
1.3 B.3 A new estimate
1.3.1 B.3.1 Proof of Lemma 2.8
This is the most delicate and cumbersome estimate since it combines the action of both space and time discrete results. For clarity, we have divided the proof in three steps.
Step 1. An estimate for I13. Using integration by parts in space, we get
where we have used that \(\overline D_t\) commutes with ∂h and mh in the first and second terms, respectively. Using formula Eq. 7, we have that the \(I_{13}^{(1)}\) can be written as
and integrating by parts in time in the first integral, we get
From Proposition A.7(i), we observe that for \(0<(sh)\leq \frac {\tau h}{\delta T^{2}}<\epsilon _{1}(\lambda )\) with 𝜖1(λ) small enough, we have that \(\left (r\overline {\tilde {\rho }}\right )\geq c_{\lambda }>0\); thus, the last three terms of the above equation have prescribed signs (we remark that the extra average does not affect the form of that estimate). That is,
where \(K:=\int \limits \!\!\!{\int \limits }_{Q}|\partial _h(\bar {\mathtt t}^+z)|^{2} \overline D_t\widetilde {(r\overline {\tilde {\rho }})}\). We claim the following.
Claim B.3
Provided \(\frac {\triangle t \tau }{\delta ^{2} T^{3}}\leq \frac {1}{2}\), we have
Proof
Notice that we can write \(2 \overline D_t\widetilde {(r\overline {\tilde {\rho }})}=(\mathtt s^{+}+\mathtt s^{-}) \overline D_t (r\overline {\tilde {\rho }})\). The result follows by applying Lemma A.16(ii). □
Observe that the condition on △t of Claim B.3 is in agreement with our initial hypothesis. Actually, the assumption of Lemma 2.8 is stronger than this condition. Using that \(|\partial _h(\bar {\mathtt t}^+z)|^{2}\leq C |\partial _h(\bar {\mathtt t}^-z)|^{2}+C(\triangle t)^{2}(\overline D_t(\partial _h z))^{2}\) and Eq. 41 we get
for some Cλ > 0 uniform with respect to △t and where \(\nu _{K}:=T \theta (sh)^{2} \mathcal O_{\lambda }(1)+\left (\frac {\tau \triangle t}{\delta ^{3} T^{4}}\right )\left (\frac {\tau h}{\delta ^{} T^{2}}\right )\mathcal O_{\lambda }(1)\). Notice that taking 𝜖1(λ) small enough in our initial hypothesis the last term of Eq. 40 controls the last term of Eq. 42. So, overall, \(I_{13}^{(1)}\) can be bounded as
for a constant \(\tilde c_{\lambda }>0\) uniform with respect to △t.
Let us comeback to the term \(I_{13}^{(2)}\). From Proposition A.8(ii), we have that \(\partial _h[r\overline {\tilde {\rho }}]=(sh)^{2}\mathcal O_{\lambda }(1)\) and using Cauchy-Schwarz and Young inequalities, we get
for some C > 0 only depending on λ. A direct computation shows that \(|\overline D_t(\widetilde {z})|^{2}=|\widetilde {(\overline D_t z)}|^{2}\leq \widetilde {(\overline D_t z)^{2}}\) by convexity. This, together the fact that \(z_{|\partial {\Omega }}^{n-\frac {1}{2}}=0\) for n ∈⟦1,M⟧, we can use Eq. 3 to deduce
Combining Eqs. 43 and 44 will provide a lower bound for I13. This concludes Step 1.
Step 2. An estimate for I23. Using that \(\overline {\tilde {z}}=z+\frac {h^{2}}{4}\overline {\partial _h}\partial _h z\), we can rewrite I23 as
Let us estimate \(I_{23}^{(1)}\). Using identity Eq. 7 and the integration-by parts formula in Eq. 11, we see that
To estimate the last three terms of the above expression, we can use that
For the first one, we can use directly Lemma A.16(iii) sampled on the primal mesh \(\mathcal P\). Thus,
where \(\mu _{P}=T s^{2}\theta \mathcal O_{\lambda }(1)+ \left (\frac {\tau ^{2} \triangle t}{\delta ^{4} T^{6}}\right )\mathcal O_{\lambda }(1)+\left (\frac {\tau \triangle t}{\delta ^{3} T^{4}}\right )\left (\frac {\tau h}{\delta ^{} T^{2}}\right )^{3}\mathcal O_{\lambda }(1)\).
We focus now on the term \(I_{23}^{(2)}\). Using that \({\Delta }_h z=\overline {\partial _h}(\partial _h z)\), we can integrate by parts in space and get
where we have used once again that \(\overline D_t\) commutes with the space-discrete operations ∂h and mh.
Arguing as we did for \(I_{12}^{(1)}\), we see that
We have the following.
Claim B.4
Provided \(\frac {\triangle t \tau }{\delta ^{2} T^{3}}\leq \frac {1}{2}\), we have
The proof is exactly as in Claim B.3 but taking into account the estimate in Lemma A.16(iii).
Using Eq. 45 (and noting that the estimate does not change with the extra average in space), Claim B.4 and the fact that \(|\partial _h(\bar {\mathtt t}^+z)|^{2}\leq C |\partial _h(\bar {\mathtt t}^-z)|^{2}+C(\triangle t)^{2}(\overline D_t(\partial _h z))^{2}\), we compute after a long but straightforward computation that
for some positive C only depending on λ. Above we used that h ≪ 1 to adjust some powers of h. We also notice that in the above equation, all of the terms containing \(|\overline D_t(\partial _h z)|^{2}\) can be made small enough by the initial hypothesis of the lemma. This will be important in the next step since those terms will be absorbed by the corresponding one in Eq. 43.
We turn our attention to the term \(\mathcal J_{2}\). Estimate Eq. 45 and a quick computation yield
where the second inequality is obtained by arguing exactly as for the term \(I_{13}^{(2)}\).
Step 3. Conclusion. Now, we are in position to find a lower bound for I13 + I23. First, we shall notice that the last three terms of Eq. 49 can be controlled by the last term of Eq. 43 by decreasing (if necessary) the parameter 𝜖1(λ) in our initial hypothesis. Then, we just need to combine Eq. 43, Eq. 44, Eq. 46, Eq. 47, Eq. 49, and Eq. 50 to obtain
for some \(c_{\lambda },c^{\prime }_{\lambda }>0\) uniform with respect to h and △t and where
This ends the proof.
Appendix C: Some intermediate lemmas
1.1 C.1 Proof of Lemma 2.12
By shifting the integral in time (see Eq. 8) and then using Eq. 3, we have
where we recall that s = τ𝜃. Since ϕ is a positive function, notice that the last two terms of the above expression are positive.
Let us focus on the term \(\mathcal D\). From Lemma A.2, we get
For the term \(\mathcal D_{1}\), using once again Lemma A.2 gives
where we have used that \(\overline {\partial _h}\partial _h = {\Delta }_h \). On the other hand, integrating by parts in the space variable, from \(\mathcal D_{2}\), we get
Notice that \(\overline {\partial _h} (\phi )=\partial _{x}\phi +h^{2}\mathcal O_{\lambda }(1)=\mathcal O_{\lambda }(1)\) thanks to Lemma A.5(iii). Thus, once the parameter λ is fixed, we can choose h sufficiently small such that the last two terms of Eq. 51 control the last two terms of Eq. 52. Moreover, using items (ii) and (iii) of Lemma A.5, we see that \(\partial _h(\overline {\partial _h} \phi )={\partial _{x}^{2}}\phi +h^{2}\mathcal O_{\lambda }(1)=\O _{\lambda }(1)\) and \(\overline {\phi }=\phi +h\mathcal O_{\lambda }(1)\). From these and putting together Eq. 51–Eq. 52, we obtain
Shifting the integrals in time in the right-hand side of the above expression yields the desired result.
1.2 C.2 Proof of Lemma 2.14
We begin by increasing, if necessary, the parameter τ1 such that τ1 ≥ 1 and τ ≥ 1. Notice that
We repeat the definition of \(\underline {X_{2}}\) for convenience. We have
We remark that at this point, we have the condition \(\frac {\tau h}{\delta T^{2}}\leq \epsilon _{3}\) for some 𝜖3 = 𝜖3(λ) small enough (see Eq. ??). Recalling that δ ≤ 1/2, 0 < T < 1 and Eq. 53 allows us to see that provided
for some κ1 > 0 small enough, the term \(\underline {X_{2}}\) can be bounded as
Notice that in the first inequality we have the product of the small parameters κ1 and 𝜖3. Since 𝜖3 has been already chosen small we can bound it uniformly by 1.
On the other hand, from definition Eq. ?? and rewriting it as \(\underline {W}={\sum }_{i=1}^{3}W^{(i)}\), we proceed to bound each of the terms. For the first one, using that \(\max \limits _{t\in [0,T]}\theta \leq (\delta T^{2})^{-1}\) we get
For W(2), we have
where the condition
holds for some κ2 > 0 small enough. Finally, using Eq. 53, we see that
Since δ ≤ 1/2, τ ≥ 1 and 0 < T < 1, we can combine conditions Eqs. 55 and 59 into a single one verifying
for some 𝜖5 = 𝜖5(λ) small enough. Collecting estimates Eq. 57, Eqs. 58, and 60 gives the desired result.
1.3 C.3 Proof of Lemma 2.16
We adapt the procedure used in the continuous setting. In particular, we follow [27, pp. 1409]. Let us consider \(\eta \in C_{c}^{\infty }({\Omega })\) such that
By the properties of discretization, we can ensure the additional property
uniformly with respect to h. Obviously, the above function is supported within \(\mathcal B\).
By shifting the time integral and the definition of the function η, we have
For the first term in the above expression, we have by using Eq. 62 and Cauchy-Schwarz and Young inequalities
where we have used that s(t) ≥ 1 for all t to adjust the power of s in the last term of the above equation.
For the term \(\mathcal L_{2}\), using that \(\overline {\eta }=\eta +h\mathcal O(1)\) together with Cauchy-Schwarz and Young inequalities
for any γ > 0.
Using estimates Eq. 64–Eq. 65 in Eq. 63 and recalling the properties of η in Eq. 61 gives the desired result after shifting the integral in time.
Rights and permissions
About this article
Cite this article
Casanova, P.G., Hernández-Santamaría, V. Carleman estimates and controllability results for fully discrete approximations of 1D parabolic equations. Adv Comput Math 47, 72 (2021). https://doi.org/10.1007/s10444-021-09885-4
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10444-021-09885-4