Abstract
In this paper, we prove the null controllability of a one-dimensional degenerate parabolic equation with a weighted Robin boundary condition at the left endpoint, where the potential has a singularity. We use some results from the singular Sturm–Liouville theory to show the well-posedness of our system. We obtain a spectral decomposition of a degenerate parabolic operator with Robin conditions at the endpoints, we use Fourier–Dini expansions and the moment method introduced by Fattorini and Russell to prove the null controllability and to obtain an upper estimate of the cost of controllability. We also get a lower estimate of the cost of controllability by using a representation theorem for analytic functions of exponential type.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and main results
Let \(T > 0\) and set \(Q_T:= (0, 1) \times (0, T )\). For \(\alpha , \beta \in {\mathbb {R}}\) with \(0\le \alpha < 2\), consider the equation
provided that \(\mu \in {\mathbb {R}}\) satisfies
In this work, we consider a weighted Robin boundary condition at the left endpoint of the form
and a usual Robin boundary condition at the right endpoint of the form
where
The goal of this work is to prove the null controllability of the following system, with a control \(f(t)\in L^2(0,T)\) acting at the left endpoint,
where our Lagrange form \([\cdot ,\cdot ]\) is given by
Consider the weighted Lebesgue space \(L^2_\beta (0,1):=L^2((0,1);x^\beta dx)\), \(\beta \in {\mathbb {R}}\), endowed with the inner product
and its corresponding norm is denoted by \(\Vert \cdot \Vert _{\beta }\).
Here, we use some results from the singular Sturm–Liouville theory to see the well-posedness of the system (4) with initial data in \(L^2_\beta (0,1)\), although the solution u(t) lives in an interpolation space \({\mathcal {H}}^{-s}\). We say the system (4) is null controllable in \(L^2_\beta (0,1)\) at time \(T>0\) with controls in \(L^2(0,T)\), if for any \(u_0\in L^2_\beta (0,1)\) there exists \(f\in L^2(0,T)\) such that the corresponding solution satisfies \(u(\cdot ,T)\equiv 0\).
We are also interested in the behavior of the cost of the controllability. Consider the set of admissible controls given by
If X is a subspace in \(L^2_\beta (0,1)\), we define the cost of controllability for initial data in X as follows:
The main result of this work is the following.
Theorem 1
Let \(T>0\), \(0\le \alpha <2\), \(\beta \in {\mathbb {R}}\), and \(\mu \) satisfying (2). The next statements hold.
-
1.
Existence of a control. For any \(u_0\in L^2_\beta (0,1)\) there exists a control \(f \in L^2(0, T )\) such that the solution u to (4) satisfies \(u(\cdot ,T ) \equiv 0\).
-
2.
Upper bound of the cost. There exists a constant \(c>0\) such that for every \(\delta \in (0,1)\), we have
$$\begin{aligned}{\mathcal {K}}_{\Phi _0^\perp }(T,\alpha ,\beta ,\mu )\le \frac{c M(T,\alpha ,\nu ,\delta )T^{1/2}}{(\nu +1)\kappa _\alpha ^{5/2}} \exp \left( -\frac{T}{2}\kappa _\alpha ^2 j_{\nu +1,1}^2\right) ,\end{aligned}$$where
$$\begin{aligned}{} & {} \kappa _\alpha :=\frac{2-\alpha }{2},\nonumber \\{} & {} \nu =\nu (\alpha ,\beta ,\mu ):=\sqrt{\mu (\alpha +\beta )-\mu }/\kappa _\alpha ,\nonumber \\{} & {} \Phi _0(x)=\sqrt{2(\nu +1)\kappa _\alpha }\,x^{-a}, \end{aligned}$$(5)\(j_{\nu +1,1}\) is the first positive zero of the Bessel function \(J_{\nu +1}\) (defined in the Appendix), and
$$\begin{aligned} M(T,\alpha ,\nu ,\delta )= & {} \left( 1+\frac{1}{(1-\delta )\kappa _\alpha ^2 T}\right) \left[ \exp \left( \frac{1}{\sqrt{2}\kappa _\alpha }\right) +\frac{1}{\delta ^5}\exp \left( \frac{3}{(1-\delta )\kappa _\alpha ^2 T}\right) \right] \\{} & {} \times \exp \left( -\frac{(1-\delta )^{3/2}T^{3/2}}{8(1+T)^{1/2}}\kappa _\alpha ^3 j_{\nu +1,1}^2\right) .\end{aligned}$$ -
3.
Lower bound of the cost. There exists a constant \(c>0\) such that
$$\begin{aligned}{} & {} c\left( 1+\frac{j_{\nu +1,2}^2}{j_{\nu +1,1}^2}\right) \frac{2^{\nu }|J_{\nu }(j_{\nu +1,1})|\exp {\left( \left( \frac{1}{2}-\frac{\log 2}{\pi }\right) j_{\nu +1,2}\right) }}{\Gamma (\nu +1)^{-1}\left( 2T\kappa _\alpha \right) ^{1/2}(j_{\nu +1,1})^\nu }\\{} & {} \quad \times \exp \left( -\left( j_{\nu +1,1}^2+\frac{j_{\nu +1,2}^2}{2}\right) \kappa _\alpha ^2T \right) \\{} & {} \qquad \le {\mathcal {K}}_{L^2_\beta }(T,\alpha ,\beta ,\mu ),\end{aligned}$$where \(j_{\nu +1,2}\) is the second positive zero of the Bessel function \(J_{\nu +1}\).
We also analyze the null controllability of a similar system but the control acting at the right endpoint,
Consider the corresponding set of admissible controls
and the cost of the controllability given by
where X is a subspace in \(L^2_{\beta }(0,1)\).
Theorem 2
Let \(T>0\), \(\beta \in {\mathbb {R}}\), \(0\le \alpha <2\), and \(\mu \) satisfying (2). The next statements hold.
-
1.
Existence of a control. For any \(u_0\in L^2_\beta (0,1)\) there exists a control \(f \in L^2(0, T )\) such that the solution u to (6) satisfies \(u(\cdot ,T ) \equiv 0\).
-
2.
Upper bound of the cost. There exists a constant \(c>0\) such that for every \(\delta \in (0,1)\), we have
$$\begin{aligned}\widetilde{{\mathcal {K}}}_{\Phi _0^\perp }(T,\alpha ,\beta ,\mu )\le \frac{c M(T,\alpha ,\nu ,\delta )T^{1/2}}{\kappa _{\alpha }^{\nu +1}\Gamma (v+2)} \left( \dfrac{2\nu +1}{4T\textrm{e}}\right) ^{(2\nu +1)/4}\! \exp \left( -\frac{T}{4}\kappa _\alpha ^2 j_{\nu +1,1}^2\right) .\end{aligned}$$ -
3.
Lower bound of the cost. There exists a constant \(c>0\) such that
$$\begin{aligned}{} & {} c\left( 1+\frac{j_{\nu +1,2}^2}{j_{\nu +1,1}^2}\right) \frac{\exp {\left( \left( \frac{1}{2}-\frac{\log 2}{\pi }\right) j_{\nu +1,2}\right) }}{\left( 2T\kappa _\alpha \right) ^{1/2}}\exp \left( -\left( j_{\nu +1,1}^2+\frac{j_{\nu ,2}^2}{2}\right) \kappa _\alpha ^2 T\right) \\{} & {} \le \widetilde{{\mathcal {K}}}_{L^2_\beta }(T,\alpha ,\beta ,\mu ).\end{aligned}$$
2 Previous work
In the last twenty years, there has been extensive research activity on the controllability of degenerate/singular parabolic equations with appropriate boundary conditions, due to both theoretical interest and their interesting applications in engineering, physics, biology, and economics. Currently, there are well-known methods to solve this kind of problems: the use of global Carleman inequalities, the flatness approach, the moment method, the transmutation method. We refer to [7, 9], whose authors obtain Carleman inequalities for degenerate/singular parabolic equations on the unit interval or on a non-empty subset in \({\mathbb {R}}^2\), and as application they prove null controllability by means of controls acting at the boundary or at an interior point in the domain.
Throughout this section consider the differential operator
on the unit interval, where \(a,b\ge 0\) can degenerate somewhere. If \(a = 0\) somewhere in [0, 1], the problem becomes degenerate, while if \(b = 0\), it is singular. We also assume that \(\omega \) is a non-empty subinterval in (0, 1).
Consider the (weighted) boundary operator
provided the limit exists. Notice that \(B_0\) is a Dirichlet boundary operator at \(x=0\), and \(B_1\) is a weighted Neumann boundary operator at \(x=0.\)
In [5, 6], the authors first demonstrated the null controllability, at the time \(T>0\), of the following system,
where \({\mathbb {A}}_0\) is the operator given in (7) in divergence form with \(a(x):=x^\alpha \), \(f\in L^2(Q_T)\), \(u_0\in L^2(0,1)\), \(i=0\) in the weak degenerate case \(0\le \alpha <1\), \(i=1\) in the strong degenerate case \(1\le \alpha <2\).
In [5, 6], the authors build weights related to the degeneracy of the diffusion coefficient a to get Carleman estimates. The authors combine these estimates with Hardy-type inequalities to prove observability for the adjoint system. It can be proved that their Carleman estimates [6, Theorem 2.2] imply a boundary null controllability result with a control acting at \(x = 1\). In this case, our differential operator \({\mathcal {A}}\), given in (10) and considering \(\beta =\mu =0\), generalizes the operator \({\mathbb {A}}_0\) in the divergence form. In [13], the author solves the weak degenerate case (in homogeneous divergence form) by using a Dirichlet boundary control at \(x=0\). There the author uses the transmutation method: First, it proves an observability inequality for the degenerate wave equation \(v_{tt}-(x^\alpha v_x)_x\)=0 considering the usual boundary conditions, uses a transmutation to pass from heat processes to waves; thus, it gets an observability inequality for the heat equation which implies the null controllability.
The next step was to consider coefficients with degeneracy at an interior point or non-smooth coefficients. In [8], the authors analyze the null controllability of the system (8) with homogenous Dirichlet boundary conditions at the endpoints, where \({\mathbb {A}}_0\) is the operator given in the both forms in (7), the initial data \(u_0\) is in X (where \(X=L^2(0,1)\) in the divergence case and \(X=L^2_{1/a}(0,1)\) in the non-divergence case), and the control \(f\in L^2(0,T;X)\) is supported in \(\omega \subset (0,1)\), which can contain the degenerate point \(x_0\). In this work, the diffusion coefficient a is a non-smooth function. When a degenerates at an interior point \(x_0\), the authors distinguish between the so-called weakly degenerate case and the strong degenerate case.
Then, the authors give two versions of Carleman estimates for the adjoint system. In the first one, a is globally non-smooth and does not degenerate; in the second one, a is non-smooth and degenerates at \(x_0\). They prove a weighted Hardy–Poincare inequality for functions which may not be globally absolutely continuous in the domain, but whose irregularity point is compensated by the fact that the weight degenerates exactly there. Then, observability inequalities are obtained from the Carleman estimates, thus they get the null controllability. In the divergence case, the degeneracy point \(x_0\) can be outside as well as inside \(\omega \). In the non-divergence case, only the case in which the degeneracy point lies outside the control region is considered.
An open problem is to obtain a Carleman estimate for the adjoint system (with homogeneous weighted Robin boundary conditions) of the system (4), and try to get a distributed control on \(\omega \) (which could include the degeneracy point) for the system (4).
Another useful tool to prove boundary null controllability of degenerate systems is the so-called flatness method. In [18], the author considers the system (8) with the homogeneous PDE in divergence form, boundary operator \(B_1\), \(a(x)=x^\alpha \), \(\alpha \in [1,2)\), \(u_0\in L^2(0,1)\), and a control h acting at the right endpoint, i.e., \(u(1,t)=h(t)\).
In [18], the author uses the flatness approach to construct explicit (smooth) controls h in some Gevrey classes. To do this, the author uses that \({\mathbb {A}}_0\) is a diagonalizable self-adjoint positive operator, whose corresponding orthogonal basis can be written as a composition of powers of the variable x with a Bessel function of the first kind (and involving its positive zeros), to construct a flat output in a Gevrey class. We think the flatness method could be adapted to prove the boundary null controllability of our system (4), by using Proposition A.1 to construct the corresponding flat output.
In [19], the authors also use the flatness approach to prove the boundary null controllability of the following system:
where \(r_{0}, s_{0},r_{1}, s_{1}\in {\mathbb {R}}\), \(r_j^2+s_j^2>0\), \(u_{0}\in L^2(0,1)\) y \(h\in L^2(0,T)\).
They assume that \(a(x)>0\) and \(\rho (x) > 0\) for a.e \(x\in (0,1)\), \(1/a, b/a, c, \rho \in L^1(0,1)\),
If we multiply the PDE in (4) by \(x^\beta \), we obtain the PDE in (9) with \(a(x)=x^{\alpha +\beta }\), \(b\equiv 0\), \(c(x)=\mu /x^{2-\alpha -\beta }\), \(\rho (x)=x^\beta \). Thus, \(1/a\in L^1(0,1)\) iff \(\alpha +\beta <1\), and \(c\in L^1(0,1)\) iff \(\alpha +\beta >1\). Therefore, our problem does not fit in the scheme of [19]. Moreover, we consider a suitable weighted Robin boundary condition at \(x=0\), where the degeneracy/singularity arises, and the control acts at this point.
The condition \(1/a \in L^1(0,1)\) in [19] implies that the PDE in (9) is a weakly degenerate parabolic equation. In [2], the authors use the flatness approach to show the null controllability of the degenerate parabolic equation without drift (\(b\equiv 0\)) in (9), with the boundary conditions corresponding to \(r_0=0, s_0=1\). The main assumption is that the function x/a(x) is in \(L^p (0, 1)\) for some \(p > 1\), which implies that \(1/a \notin L^1(0,1)\). Thus, a may vanish strongly at \(x = 0\), and the potential c may be singular at the same point, but in [2] the control acts at \(x=1\); by contrast, our control acts at \(x=0\), and we have a drift, provided that \(\beta \ne 0\).
In [21], the author proves some global Carleman estimates for the degenerate/singular parabolic operator \(w_t-{\mathbb {A}}_\lambda w\) with \(a(x)=x^\alpha \), \(b(x)=x^{{\widetilde{\beta }}}\), and boundary conditions (depending on \(\alpha \)) as in (8). The author gets an improved Hardy–Poincaré inequality and obtains an observability result that implies the null controllability of the system (8), with \({\mathbb {A}}_\lambda \) (instead of \({\mathbb {A}}_0\)) in divergence form, by means of a distributed control f. In the case \({\widetilde{\beta }}=2-\alpha \), \(\lambda <\mu (\alpha )\), the corresponding PDE coincides with the PDE in (4) with \(\beta =0\), \(\mu <\mu (\alpha )\).
In [4, 11, 12, 14], the authors use the moment method to prove the boundary null controllability of systems like (9). In [14], the authors consider \(a(x)=\varepsilon x^{\alpha +1}\), \(b(x)=-x^\alpha \), \(\varepsilon ,\alpha \in (0,1)\). They consider \(r_0=r_1=1, s_0=s_1=0\), so their control acts at the left endpoint. This is a strongly degenerate parabolic problem, but at present, we know this kind of degeneracy is related to a Neumann weighted boundary condition, see [12].
In [11], the authors prove the null controllability of the equation (1) with a weighted Dirichlet boundary condition at the left endpoint, provided that \(\alpha +\beta <1\). In the case \(\alpha +\beta >1\), in [12], they get the null controllability of the equation (1) with a weighted Neumann boundary condition at the left endpoint. They consider initial data in \(L^2_\beta (0,1)\) in both cases. In these works, the authors prove suitable versions of a Hardy inequality to assure the well-posedness of their systems, but in the case \(\alpha +\beta =1\) is necessary to consider some results from the singular Sturm–Liouville theory, see [12]. Here, we use that approach to show the well-posedness of our system.
Unfortunately, for this paper, we could not prove a suitable weighted Hardy–Poincaré considering the (weighted) homogeneous Robin boundary conditions in (4). This fact motivate us to use the singular Sturm–Liouville theory, which shows that the operator \(({\mathcal {A}}, D({\mathcal {A}}))\) given in (10) is self-adjoint.
This paper is organized as follows. Section 3 uses some results from the singular Sturm–Liouville theory to show that the operator \({\mathcal {A}}\) given in (10) is self-adjoint. There, we also use Fourier–Dini expansions to show that \({\mathcal {A}}\) is diagonalizable, this allows us to consider initial data in some interpolation spaces. Next, we introduce a notion of a weak solution for both systems and then show the well-posedness of these systems.
In Sect. 4, we prove Theorem 1 by using the moment method introduced by Fattorini & Russell. Here, the idea is to construct a biorthogonal sequence to a family of exponentials involving the eigenvalues of \({\mathcal {A}}\). To do this, we use some results from complex analysis to construct a suitable complex multiplier. As a consequence, we get an upper estimate of the cost of the controllability. Finally, we use a representation theorem, Theorem 13, to obtain a lower estimate of the cost of the controllability.
In Sect. 5, we proceed as before to solve the case when the control acts at the right endpoint.
3 Functional setting and well-posedness
Consider the differential expression M defined by
where \(\displaystyle p(x) = x^{\alpha +\beta }, q(x) = -\mu x^{-2+\alpha +\beta }, w(x) = x^{\beta }\).
Clearly,
thus Mu is defined a.e. for functions u such that \(u, pu_x\in AC_{\text {loc}}(0,1)\), where \(AC_{\text {loc}}(0,1)\) is the space of all locally absolutely continuous functions in (0, 1).
Now, we introduce the operator \({\mathcal {A}}\) given by
From the theory developed in [23], we can build a self-adjoint domain \(D({\mathcal {A}})\) for the operator \({\mathcal {A}}\).
For \(\mu \) satisfying (2), \(0\le \alpha <2\), and \(\beta \in {\mathbb {R}}\), we set
Recall that the Lagrange form associated with M is defined as follows:
The next result shows that \({\mathcal {A}}\) is a diagonalizable operator whose Hilbert basis of eigenfunctions can be written in terms of the function \(x^{1/2+\nu }\), the Bessel function of the first kind \(J_{\nu }\) and the corresponding positive zeros \(j_{\nu +1,k}\), \(k\ge 1\), of the Bessel function \(J_{\nu +1}\), see the proof of Proposition A.1. In the appendix, we give some properties of Bessel functions of the first kind and their zeros.
Proposition 3
Let \(0\le \alpha <2\), \(\beta \in {\mathbb {R}}\), \(\mu < \mu (\alpha +\beta )\), and \(\kappa _\alpha ,\nu \) given in (5). Then, \({\mathcal {A}}:D({\mathcal {A}})\subset L^2_{\beta }(0,1)\rightarrow L^2_{\beta }(0,1)\) is a self-adjoint operator. Furthermore, the family \(\{\Phi _k\}_ {k\ge 0}\) given by
is an orthonormal basis for \(L^2_{\beta }(0,1)\) such that
where \(\lambda _{0}:=0\) and \(\lambda _{k}:=\kappa ^{2}_{\alpha }(j_{\nu +1,k})^2,\, k\ge 1\).
Proof
Since \(1/p,q,w\in L^1(1/2,1)\) we have that \(x=1\) is a regular point.
Case i) Assume \(\sqrt{\mu (\alpha +\beta )-\mu }< \kappa _\alpha \).
First, we will build a (BC) basis \(\{y_{0},z_{0}\}\) at \(x=0\) and a (BC) basis \(\{y_{1}, z_{1}\}\) at \(x=1\), see [23, Definition 10.4.3].
Consider the functions given by
Notice the assumption implies that \(y_{0}, z_{0}\in D_{\max }\). Clearly, \([z_{0}, y_{0}](0)=1\), thus \(\{y_{0}, z_{0}\}\) is a (BC) basis at \(x=0\).
Since \(y_{0}, z_{0}\in L^2_\beta (0,1)\) are linearly independent solutions of \(Mu=0u\) it follows that \(x=0\) is limit-circle (LC), see [23, Definition 7.3.1, Theorem 7.2.2].
Consider also the functions given by
Since \(y_{1}, z_{1}\in D_{\max }\) and \([z_{1}, y_{1}](1)=1\), it follows that \(\{y_{1}, z_{1}\}\) is a (BC) basis at \(x=1\).
Now, we fix \(c,d\in (0,1)\) with \(c<d\). From the Patching Lemma, Lemma 10.4.1 in [23], there exist functions \(g_1, g_2\in D_{\max }\) such that
Thus, the pair \(\{y_+,y_-\}\) is a (BC) basis on (0, 1), see [23, Definition 10.4.3], where
The matrices
satisfy the hypothesis in [23, Proposition 10.4.2], then
is a self-adjoint domain, therefore the operator \({\mathcal {A}}:D({\mathcal {A}})\subset L^2_\beta (0,1)\rightarrow L^2_\beta (0,1)\) is self-adjoint.
Finally, we have that
because \([z_0,y_0](0)=1\), \([u,z_0](0)\) is finite (see [23, Lemma 10.2.3]), and \(\lim _{x\rightarrow 0^+}y_0/z_0(x)=0\). Hence, the result follows.
Case ii) Assume \(\sqrt{\mu (\alpha +\beta )-\mu }\ge \kappa _\alpha \).
The assumption implies that \(z_{0}\notin L^2_{\beta }(0,1)\), then \(x=0\) is limit point (LP). Theorem 10.4.4 in [23] with \(A_1=a, A_2=1\) implies that \(D({\mathcal {A}})=\{u\in D_{\max }\,| (au+u_{x})(1)=0\}\) is a self-adjoint domain.
This concludes the first part of the proof.
Clearly, \(\Phi _k\in C^\infty (0,1)\) and (61) implies that \(\Phi _k\in L^2_\beta (0,1)\) for all \(k\ge 0\). Moreover,
By using (63), we obtain
therefore \((a\Phi _k+\Phi '_k)(1)=0\) for all \(k\ge 1\). Clearly, \((a\Phi _0+\Phi '_0)(1)=0\). Therefore, \(\Phi _k\in D({\mathcal {A}})\) for all \(k\ge 0\).
We set \(v(x)=x^bJ_\nu (cx^r)\) with \(r,c>0\) and \(b\in {\mathbb {R}}\). The proof of Proposition 11 in [12] was shown that
By taking \(r=\kappa _\alpha , b=(1-\alpha -\beta )/2\), and \(c=j_{\nu +1,k}\), we get \({\mathcal {A}}\Phi _k=\lambda _k \Phi _k\) for all \(k\ge 1\). Clearly, \({\mathcal {A}}\Phi _0=0\). The result follows by Proposition A.1. \(\square \)
Remark 4
If \(\sqrt{\mu (\alpha +\beta )-\mu }\ge \kappa _\alpha \), from Lemma 10.4.1(b) in [23], we have that \([u,y_0](0)=0\) for all \(u \in D({\mathcal {A}})\). When \(\sqrt{\mu (\alpha +\beta )-\mu }<\kappa _\alpha \), in the proof of the last proposition was shown that \([u,y_0](0)=0\) for all \(u \in D({\mathcal {A}})\), where \(y_0\) is given in (12).
Remark 5
The family \(\{\Theta _k\}_{k\ge 0}\) given in (67) is the so-called Fourier–Dini basis for \(L^2(0,1)\).
Then, \(({\mathcal {A}},D({\mathcal {A}}))\) is the infinitesimal generator of a diagonalizable self-adjoint semigroup in \(L^2_{\beta }(0,1)\). Thus, we can consider interpolation spaces for the initial data. For any \(s\ge 0\), we define
and we also consider the corresponding dual spaces
It is well known that \({\mathcal {H}}^{-s}\) is the dual space of \({\mathcal {H}}^{s}\) with respect to the pivot space \(L^2_\beta (0,1)\), i.e.,
Equivalently, \({\mathcal {H}}^{-s}\) is the completion of \(L^2_\beta (0,1)\) with respect to the norm
It is well known that the linear mapping given by
defines a self-adjoint semigroup \(\{S(t)\}_{t\ge 0}\) in \({\mathcal {H}}^s \) for all \(s\in {\mathbb {R}}\).
For \(\delta \in {\mathbb {R}}\) and a function \(h:(0,1)\rightarrow {\mathbb {R}}\), we introduce the notion of \(\delta \)-generalized limit of h at \(x=0\) as follows:
Notation: Let \(t>0\) fixed. If \(z\in {\mathcal {H}}^s\) then \(S(t)z\in {\mathcal {H}}^s\), so we write \(\lim _{x \rightarrow 1^-} S(t)z\) instead of \(\lim _{x \rightarrow 1^-} (S(t)z)(x)\).
3.1 Notion of weak solutions for both systems
Now, we consider a convenient definition of a weak solution for the system (4). Let \(\tau >0\) be fixed. We multiply the equation in (4) by \(x^\beta \varphi (x,t)=x^\beta S(\tau -t)z^{\tau }\), \(0\le t\le \tau \), integrate by parts (formally), and by using the boundary conditions for \(u,\varphi \), see Remark 4, we get
Definition 6
Let \(T>0\), \(0\le \alpha <2\), \(\beta \in {\mathbb {R}}\), \(\mu <\mu (\alpha +\beta )\), and a given by (3). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) for some \(s > 0\). A weak solution of (4) is a function \(u \in C^0([0,T];{\mathcal {H}}^{-s})\) such that for every \(\tau \in (0,T]\) and for every \(z^\tau \in {\mathcal {H}}^s\), we have
The next result shows the existence of weak solutions for the system (4) under suitable conditions on the parameters \(\alpha ,\beta ,\mu ,\) and s, and its proof is similar to the proof of Proposition 2.9 in [11].
Proposition 7
Let \(T>0\), \(0\le \alpha <2\), \(\beta \in {\mathbb {R}}\), \(\mu <\mu (\alpha +\beta )\), a given in (3). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) such that \(s>\nu \), with \(\nu \) given in (5). Then, formula (13) defines for each \(\tau \in [0, T ]\) a unique element \(u(\tau ) \in {\mathcal {H}}^{-s}\) that can be written as
where \(B(\tau )\) is the strongly continuous family of bounded operators \(B(\tau ): L^{2}(0,T) \rightarrow {\mathcal {H}}^{-s}\) given by
Furthermore, the unique weak solution u on [0, T] to (4) (in the sense of (13)) belongs to \({C}^{0}\left( [0, T]; {\mathcal {H}}^{-s}\right) \) and fulfills
Proof
Fix \(\tau >0\). Let \(u(\tau )\in H^{-s}\) be determined by the condition (13), hence
where
We claim that \(\zeta (\tau )\) is a bounded operator from \(L^{2}(0, T)\) into \({\mathcal {H}}^{-s}\): consider \(z^{\tau } \in {\mathcal {H}}^{s}\) given by
therefore
By using Lemma A.3 and (70), we obtain that there exists a constant \(C=C(\alpha ,\beta ,\mu )>0\) such that
hence (69) implies that there exists a constant \(C=C(\alpha ,\beta ,\mu ,\tau )>0\) such that
Therefore, \(\Vert \zeta (\tau ) f\Vert _{{\mathcal {H}}^{-s}}\le C\Vert f\Vert _{L^2(0,T)}\) for all \(f\in L^2(0,T)\), \(\tau \in (0,T]\).
Finally, we fix \(f\in L^2(0,T)\) and show that the mapping \(\tau \mapsto \zeta (\tau ) f\) is right-continuous on [0, T). Let \(h>0\) small enough and \(z\in {\mathcal {H}}^s\) given as in (14). Thus, proceeding as in the last inequalities, we have
where
Since \(0\le I(\tau ,k,h)\le 1/2\) uniformly for \(\tau , h>0\), \(k\ge 1\), the result follows by the dominated convergence theorem. \(\square \)
Remark 8
In the following section, we will consider initial conditions in \(L^2_\beta (0,1)\). Notice that \(L^2_\beta (0,1)\subset \mathcal {H}^{-\nu -\delta }\) for all \(\delta >0\), and we can apply Proposition 7 with \(s=\nu +\delta \), \(\delta >0\), then the corresponding solutions will be in \(C^0([0, T ], \mathcal {H}^{-\nu -\delta })\).
As before, we introduce a suitable definition of a weak solution for the system (6).
Definition 9
Let \(T>0\), \(\beta \in {\mathbb {R}}\), \(0\le \alpha <2\), \(\mu <\mu (\alpha +\beta )\) and a given in (3). Let \(f \in L^2(0,T)\) and \(u_0\in L_{\beta }^2(0,1)\). A weak solution of (6) is a function \(u \in C^0\left( [0,T];L_{\beta }^2(0,1)\right) \) such that for every \(\tau \in (0,T]\) and for every \(z^\tau \in L^2_{\beta }(0,1)\), we have
The next result shows the existence of weak solutions for the system (6) under certain conditions on the parameters \(\alpha ,\beta ,\mu \) and a, and its proof is similar to the proof of Proposition 18 in [12].
Proposition 10
Let \(T>0\), \(\beta \in {\mathbb {R}}\), \(0\le \alpha <2\), \(\mu <\mu (\alpha +\beta )\) and a given in (3). Let \(f \in L^2(0,T)\) and \(u_0\in L_{\beta }^2(0,1)\). Then, formula (16) defines for each \(\tau \in [0, T ]\) a unique element \(u(\tau ) \in L_{\beta }^2(0,1)\) that can be written as
where \(B(\tau )\) is the strongly continuous family of bounded operators \({\mathcal {B}}(\tau ): L^{2}(0,T) \rightarrow L_{\beta }^2(0,1)\) given by
Furthermore, the unique weak solution u on [0, T] to (6) (in the sense of (16)) belongs to \({C}^{0}\left( [0, T]; L_{\beta }^2(0,1)\right) \) and fulfills
Proof
Fix \(\tau >0\). Let \(u(\tau )\in L_{\beta }^2(0,1)\) be determined by the condition (16), hence
where
Let \(z^{\tau } \in L_{\beta }^2(0,1)\) written as
therefore
By (11), we get
hence there exists a constant \(C=C(\alpha ,\beta ,\mu ,\tau )>0\) such that
Therefore, \(\Vert \zeta (\tau ) f\Vert _{\beta }\le C\Vert f\Vert _{L^2(0,T)}\) for all \(f\in L^2(0,T)\), \(\tau \in (0,T]\).
Finally, we fix \(f\in L^2(0,T)\) and show that the mapping \(\tau \mapsto \zeta (\tau ) f\) is right-continuous on [0, T). Let \(h>0\) small enough and \(z\in L^2_\beta (0,1)\) given as in (17). Then, we have
where \(I(\tau ,k,h)\rightarrow 0\) as \(h\rightarrow 0^+\), see (15). \(\square \)
4 Control at the left endpoint
4.1 Upper estimate of the cost of the null controllability
Here, we use the moment method, introduced by Fattorini & Russell in [10], to prove the null controllability of the system (4). The first step is to construct a biorthogonal family \(\displaystyle \left\{ \psi _k\right\} _{k\ge 0}\subset L^2(0,T)\) to the family of exponential functions \(\left\{ \textrm{e}^{-\lambda _{k}(T-t)}\right\} _{k\ge 0}\) on [0, T], i.e., that satisfies
This construction will help us to get an upper bound for the cost of the null controllability of the system (4).
Assume that for each \(k\ge 0\) there exists an entire function \(F_k\) of exponential type T/2 such that \(F_k(x)\in L^2({\mathbb {R}})\), and
The \(L^2\)-version of the Paley-Wiener theorem implies that there exists \(\eta _k\in L^2({\mathbb {R}})\) with support in \([-T/2,T/2]\) such that \(F_k(z)\) is the analytic extension of the Fourier transform of \(\eta _k\). Then, we have that
is the family we are looking for.
Now, we proceed to construct the family \(F_k\), \(k\ge 0\). Consider the Weierstrass infinite product
From (68), we have that \(j_{\nu +1, k}=O(k)\) for k large, thus the infinite product converges absolutely in \({\mathbb {C}}\). Hence, \(\Lambda (z)\) is an entire function with simple zeros at \(i\lambda _k\), \(k\ge 0\).
From [22, Chap. XV, p. 498, eq. (3)], we have for \(\nu >-1\) that
[11] proved that
Therefore,
In particular,
It follows that
is a family of entire functions that satisfy (19). Since \(\Psi _{k}(x)\) is not in \(L^2({\mathbb {R}})\), we need to fix this by using a suitable “complex multiplier", thus we follow the approach introduced in [20].
For \(\theta ,\omega >0\), we define
and extended by 0 outside of \((-1, 1)\). Clearly \(\sigma _{\theta }\) is analytic on \((-1,1)\). Set \(C_{\theta }^{-1}:=\int _{-1}^{1}\sigma _{\theta }(t)\textrm{d}t\) and define
\(H_{\omega ,\theta }(z)\) is an entire function, and the next result provides additional properties of \(H_{\omega ,\theta }(z)\).
Lemma 11
The function \(H_{\omega ,\theta }\) fulfills the following inequalities:
where \(c>0\) does not depend on \(\omega \) and \(\theta \).
We refer to [20, pp. 85–86] for the details.
For \(k\ge 0\), consider the entire function \(F_{k}\) given as
For \(\delta \in (0,1)\), we set
Lemma 12
The function \(F_{k}(z)\), \(k\ge 0\), has the following properties:
(i) \(F_{k}\) is of exponential type T/2.
(ii) \(F_{k}\in L^1({\mathbb {R}})\cap L^2({\mathbb {R}})\).
(iii) \(F_k\) satisfies (19).
(iv) Furthermore, there exists a constant \(c>0\), independent of \(T,\alpha \) and \(\delta \), such that
where
Proof
By using (23), (27), (29) and (30), we get that \(F_{k}\) is of exponential type T/2 for all \(k\ge 0\). Moreover, by using (24) and (29), we can see that \(F_{k}\) fulfills (19).
Now, we use (23), (26), (28), (29), and (30) to get
for all \(k\ge 1\). Since the function on the right-hand side is rapidly decreasing in \({\mathbb {R}}\), we have \(F_{k}\in L^1({\mathbb {R}})\cap L^2({\mathbb {R}})\). Finally, the change of variable \(y=(\kappa _\alpha )^{-1}\delta |x|^{1/2}/\sqrt{2}\) implies (32).
When \(k=0\), we have
then we integrate on \({\mathbb {R}}\) and the result follows. \(\square \)
Since \(\eta _k, F_k\in L^1({\mathbb {R}})\), the inverse Fourier theorem yields
hence (20) implies that \(\psi _k\in C([0,T])\). From (31) and (32), we have \(\Vert \psi _0\Vert _{\infty }\le C(T, \alpha ,\delta )\) and
Now, we are ready to prove the null controllability of the system (4). Let \(u_{0}\in L^{2}_\beta (0,1)\). Then, consider its (generalized) Fourier–Dini series with respect to the orthonormal basis \(\{\Phi _{k}\}_{k\ge 0}\),
We set
Since \(\{\psi _k\}_{k\ge 0}\) is biorthogonal to \(\{\textrm{e}^{-\lambda _k(T-t)}\}_{k\ge 0}\), we have
Let \(u\in C([0,T];H^{-s})\) that satisfies (13) for all \(\tau \in (0,T]\), \(z^\tau \in H^s\). In particular, for \(\tau =T\), we take \(z^T=\Phi _k\), \(k\ge 0\), then the last equality implies that
hence \(u(\cdot , T)=0\).
It just remains to estimate the norm of the control f. From (34) and (36), we get
From (21), (22), and (64) (with \(\nu +1\) instead of \(\nu \)), we get that
and by using (70), we get
From (37), (69), and using that \(\lambda _k\ge \lambda _1\), it follows that
Using the expression of \(\omega ,\theta \) given in (30) and the facts \(\theta >0\), \(\delta \in (0,1)\), and \(0<\kappa _\alpha \le 1\), we get that
therefore
By using the definition of \(\lambda _{1}\), and setting \(b_0=0\), we get the estimate for \({\mathcal {K}}_{\Phi _0^\perp }\).
4.2 Lower estimate of the cost of the null controllability
In this section, we get a lower estimate of the cost \({\mathcal {K}}={\mathcal {K}}_{L^2_\beta }(T,\alpha ,\beta ,\mu )\).
We set
For \(\varepsilon >0\) small enough, there exists \(f\in U(\alpha ,\beta ,\mu ,T,u_0)\) such that
Then, in (13), we set \(\tau =T\) and take \(z^\tau =\Phi _k\), \(k\ge 0\), to obtain
from (40) and (70), it follows that
Now, consider the function \(v: {\mathbb {C}} \rightarrow {\mathbb {C}}\) given by
Fubini and Morera’s theorems imply that v(s) is an entire function. Moreover, (42) implies that
We also have that
Consider the entire function F(z) given by
for some \(\delta >0\) that will be chosen later on. Clearly,
From (40), (43) and (44), we obtain
We recall the following representation theorem, see [17, p. 56].
Theorem 13
Let g(z) be an entire function of exponential type and assume that
Let \(\left\{ d_{\ell }\right\} _{\ell \ge 1}\) be the set of zeros of g(z) in the upper half plane \(\Im (z)>0\) (each zero being repeated as many times as its multiplicity). Then,
where
We apply the last result to the function F(z) given in (44). In this case, (43) implies that \(A\le T/2\). Also notice that \(\Im \left( b_{k}\right) >0\), \(k\ge 0\), to get
By using the definition of the constants \(b_k\)’s, we have
where we have used Lemma A.2 and made the change of variables
From (46), we get the estimate
From (45), (47), (48), and (49), we have
The result follows by taking \(\delta =\kappa _\alpha ^2 \left( j_{\nu +1,2}\right) ^2/2\) and then letting \(\varepsilon \rightarrow 0^+\).
5 Control at the right endpoint
5.1 Upper estimate of the cost of the null controllability
Now we show the null controllability of the system (6). Let \(u_{0}\in L^{2}_\beta (0,1)\) given as in (35). We set
Since the sequence \(\{\psi _k\}_{k\ge 0}\) is biorthogonal to \(\{\textrm{e}^{-\lambda _k(T-t)}\}_{k\ge 0}\), we have
Let \(u\in C\left( [0,T];L^{2}_{\beta }(0,1)\right) \) be the weak solution of system (6). In particular, for \(\tau =T\), we take \(z^T=\Phi _k\), \(k\ge 0\), then (16) and (51) imply that \(\left\langle u(\cdot , T), \Phi _{k}\right\rangle _{\beta }=0\) for all \(k \ge 0\), therefore \(u(\cdot , T) \equiv 0\).
Finally, we estimate the norm of the control f. From (18), (34), (38) and (50), we get
By using that \(\textrm{e}^{-x}\le \textrm{e}^{-r}r^rx^{-r}\) for all \(x,r>0,\) the Cauchy–Schwarz inequality, Lemma A.3 and the fact that \(j_{\nu ,k}\ge (k-1/4)\pi \) (by (69)), (35) and \(\lambda _1 \le \lambda _k\), \(k\ge 1\), we obtain that
and the result follows by (39).
5.2 Lower estimate of the cost of the null controllability
Once again, we get a lower estimate of the cost \(\widetilde{{\mathcal {K}}}=\widetilde{{\mathcal {K}}}_{L^2_\beta }(T,\alpha ,\beta ,\mu )\). We set
For \(\varepsilon >0\) small enough, there exists \(f\in {\widetilde{U}}(\alpha ,\beta ,\mu ,T,u_0)\) such that
Then, in (16), we set \(\tau =T\) and take \(z^\tau =\Phi _k\), \(k\ge 0\), to obtain
From (18) and (52), it follows that
Consider the entire function \(v: {\mathbb {C}} \rightarrow {\mathbb {C}}\) given by
Therefore,
Moreover, (53) implies that
Consider the entire function F(z) given by
Clearly,
From (52), (54) and (55) we obtain
We apply Theorem 13 to the function F(z) given in (55). Then, (54) implies that \(A\le T/2\), hence
From (57), we get the estimate
From (48), (56), (58), and (59), we have
the result follows by letting \(\varepsilon \rightarrow 0^+\).
Data availability
This declaration is not applicable.
References
Abramowitz M, Stegun IA (1964) Handbook of mathematical functions with formulas, graphs and mathematical tables, National Bureau of Standards. App. Math. series, Vol. 55
Benoit A, Loyer R, Rosier L (2023) Null controllability of strongly degenerate parabolic equations, ESAIM: COCV 29 48
Bowman F (1958) Introduction to Bessel Functions. Dover Publications Inc, New York
Biccari U, Hernáindez-Santamaría V, Vancostenoble J (2022) Existence and cost of boundary controls for a degenerate/singular parabolic equation. Math Control Relat Fields 12:495–530
Cannarsa P, Martinez P, Vancostenoble J (2005) Null controllability of degenerate heat equations. Adv Differ Equ 10:153–190
Cannarsa P, Martinez P, Vancostenoble J (2008) Carleman estimates for a class of degenerate parabolic operators. SIAM J Control Optim 47:1–19
Cannarsa P, Martinez P, Vancostenoble J (2016) Global Carleman estimates for degenerate parabolic operators with applications. Mem Amer Math Soc 239(1133):209
Fragnelli G, Mugnai D (2016) Carleman estimates, observability inequalities and null controllability for interior degenerate nonsmooth parabolic equations. Mem Amer Math Soc 342(1146):84
Fragnelli G, Mugnai D (2021) Control of degenerate and singular equations - Carleman estimates and observability. SpringerBriefs in Mathematics. BCAM SpringerBriefs, Springer, Cham
Fattorini HO, Russell DL (1971) Exact controllability theorems for linear parabolic equations in one space dimension. Arch Ration Mech Anal 43:272–292. https://doi.org/10.1007/BF00250466
Galo-Mendoza L, López-García M Boundary controllability for a 1D degenerate parabolic equation with drift and a singular potential. Math Control and Related Fields. https://doi.org/10.3934/mcrf.2023027
Galo L, López-García M Boundary controllability for a 1D degenerate parabolic equation with drift and a singular potential and a Neumann boundary condition. arXiv:2304.00178
Gueye M (2014) Exact boundary controllability of 1-D parabolic and hyperbolic degenerate equations. SIAM J Control Optim 52(4):2037–2054
Gueye M, Lissy P (2016) Singular optimal control of a \(1\)-D parabolic-hyperbolic degenerate equation. ESAIM Control Optim Calc Var 22(4):1184–1203
Hochstadt H (1967) The mean convergence of Fourier-Bessel series. SIAM Rev 9:211–218. https://doi.org/10.1137/1009034
Komornik V, Loreti P (2005) Fourier series in control theory. Springer, Cham
Koosis P (1992) The logarithmic integral I & II, Cambridge Studies in Advanced Mathematics 12 (1988) & Cambridge Studies in Advanced Mathematics 21, Cambridge University Press, Cambridge
Moyano I (2016) Flatness for a strongly degenerate 1-D degenerate equation. Math Control Signals Syst 28(4):2822
Martin P, Rosier L, Rouchon P (2016) Null controllability of a one-dimensional parabolic equations by the flatness approach. SIAM J Control Optim 54(1):198–220
Tenenbaum G, Tucsnak M (2007) New blow-up rates for fast controls of Schrodinger and heat equations. J Differ Equ 243:70–100. https://doi.org/10.1016/j.jde.2007.06.019
Vancostenoble J (2011) Improved Hardy-Poincaré inequalities and sharp Carleman estimates for degenerate/singular parabolic problems. Discret Contin Dyn Syst Ser S 4:761–790
Watson GN (1958) A treatise on the theory of Bessel functions. Cambridge University Press, Cambridge, England
Zettl A (2005) Sturm-Liouville Theory, mathematical surveys and monographs, vol 121. Am. Math. Soc, Providence
Acknowledgements
The authors thank the referee for its valuable comments, which have considerably improved this work.
Funding
The second author was partially supported by DGAPA-UNAM [PAPIIT IN109522], and CONACYT-México [A1-S-17475]. The first author was supported by a grant from CONACyT-México.
Author information
Authors and Affiliations
Contributions
Galo-Mendoza worked on Sections 2 and 4. López-García worked on Sections 1, 3, and Appendix, and wrote the manuscript text. Both authors reviewed the manuscript
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Ethical approval
This declaration is not applicable.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A: Bessel functions
Appendix A: Bessel functions
We introduce the Bessel function of the first kind \(J_{\nu }\) as follows:
where \(\Gamma (\cdot )\) is the Gamma function. In particular, for \(\nu >-1\) and \(0<x \le \sqrt{\nu +1}\), from (60), we have (see [1, 9.1.7, p. 360])
A Bessel function \(J_\nu \) of the first kind solves the differential equation
Bessel functions of the first kind satisfy the recurrence formulas (see [1, 9.1.27]):
Recall the asymptotic behavior of the Bessel function \(J_{\nu }\) for large x, see [16, Lem. 7.2, p. 129].
Lemma A.1
For any \(\nu \in {\mathbb {R}}\)
For \(\nu >-1\), \(\ell ,\ell '\in {\mathbb {R}}\), we have (see [3, p. 101])
For \(\nu >-1\), the Bessel function \(J_{\nu }\) has an infinite number of real zeros \(0<j_{\nu , 1}<j_{\nu , 2}<\ldots \), all of which are simple, with the possible exception of \(x=0\). In [16, Proposition 7.8], we can find the next information about the location of the zeros of the Bessel functions \(J_{\nu }\):
Lemma A.2
Let \(\nu \ge 0\).
1. The difference sequence \(\left( j_{\nu , k+1}-j_{\nu , k}\right) _{k}\) converges to \(\pi \) as \(k \rightarrow \infty \).
2. The sequence \(\left( j_{\nu , k+1}-j_{\nu , k}\right) _{k}\) is strictly decreasing if \(|\nu |>\frac{1}{2}\), strictly increasing if \(|\nu |<\frac{1}{2}\), and constant if \(|\nu |=\frac{1}{2}\).
Proposition A.1
Let \(\nu >-1\), \(0\le \alpha <2\) and \(\beta \in {\mathbb {R}}\). The family
is an orthonormal basis for \(L^2_\beta (0,1)\).
Proof
By using (63) and (65) with \(\ell '=j_{\nu +1,k}\), we get
By taking the limit as \(\ell \) goes to \(j_{\nu +1,k}\), and by using (64) (with \(\nu +1\) instead of \(\nu \)), we obtain
Next, we introduce the following family
In [15] was proved that \(\{\Theta _k\}_{k\ge 0}\) is a complete system in \(L^2(0,1)\).
Then, (63), (65) and (66) imply that \(\langle \Theta _k, \Theta _\ell \rangle =\delta _{k,\ell }\) for all \(k,\ell \ge 1\). On the other hand, from (64) with \(\nu +1\) instead of \(\nu \), we obtain that
Therefore \(\langle \Theta _k, \Theta _0\rangle =0\) for all \(k\ge 1\). In conclusion, \(\{\Theta _k\}_{k\ge 0}\) is an orthonormal basis for \(L^2(0,1)\).
Let \({\mathcal {U}}\) be the unitary operator \({\mathcal {U}}:L^2(0,1)\rightarrow L^2_\beta (0,1)\) given by
Notice that \({\mathcal {U}}\Theta _k=\Phi _k\), \(k\ge 0\), therefore \(\Phi _k\), \(k\ge 0\), is an orthonormal basis for \(L^2_\beta (0,1)\). \(\square \)
For \(\nu \ge 0\) fixed, we consider the next asymptotic expansion of the zeros of the Bessel function \(J_{\nu }\), see [22, Section 15.53],
In particular, we have
Lemma A.3
For any \(\nu > -1\) and any \(k\ge 1\), we have
The proof of this result follows by using (A.1).
Lemma A.4
Let \(0\le \alpha < 2\), \(\beta \in {\mathbb {R}}\), a and \(\nu =\nu (\alpha ,\beta ,\mu )\) given in (3) and (5), respectively, then the following limits are finite
Proof
This result follows from (60). \(\square \)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Galo-Mendoza, L., López-García, M. Boundary controllability for a 1D degenerate parabolic equation with a Robin boundary condition. Math. Control Signals Syst. 36, 675–705 (2024). https://doi.org/10.1007/s00498-024-00383-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00498-024-00383-8
Keywords
- Degenerate parabolic equation
- Robin boundary condition
- Sturm–Liouville theory
- Boundary controllability
- Moment method