Abstract
We prove the null controllability of a one-dimensional degenerate parabolic equation with drift and a singular potential. Here, we consider a weighted Neumann boundary control at the left endpoint, where the potential arises. We use a spectral decomposition of a suitable operator, defined in a weighted Sobolev space, and the moment method by Fattorini and Russell to obtain an upper estimate of the cost of controllability. We also obtain a lower estimate of the cost of controllability using a representation theorem for analytic functions of exponential type.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
This work analyzes the null controllability for 1D degenerate/singular parabolic equations when the control acts at the endpoint where the degeneracy/singularity arises. In this paper, we consider a weighted Neumann boundary control.
Let \(T > 0\) and set \(Q:= (0, 1) \times (0, T )\). For \(\alpha ,\beta \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha +\beta >1\), consider the system
provided that \(\mu \in {\mathbb {R}}\) satisfies
where
We notice the constant \(\mu (\alpha +\beta )\) appears in the Hardy inequality (7); thus, the condition (2) corresponds to a so-called sub-critical potential, see [1, p. 765]. We chose the constant \(\gamma \) in such a way that the limit in (A10) is finite, see also (20).
1.1 Dirichlet boundary conditions
Most published articles about degenerate/singular parabolic equations, like system (1), consider a weighted Dirichlet boundary condition \(\lim _{x\rightarrow 0^+}x^{\sigma }u(x,t)=f(t)\) at the point of controllability \(x=0\), with a suitable constant \(\sigma \). In this context, the first result was obtained in [2] in the case of a weak degeneracy \(0 \le \alpha < 1\), without singularity and drift, i.e., \(\beta =\mu =0\). In that work, the null controllability is achieved using \(L^2\) controls. The last result was improved in [3] where the authors prove that the controllability cost behaves as \((1-\alpha )^{-1}\textrm{e}^{1/T}\), both upper and lower estimate when \(\alpha <1\).
In [4], the author analyzes the controllability of the heat equation with a singular inverse-square potential, i.e., with \(\alpha =\beta =0\) and \(\mu <\mu (0)=1/4\). In [5], the authors also study the weakly degenerate case without drift (\(\beta =0\)) and \(\mu \) satisfying (2). All these results give the null controllability using \(H^1\) controls. Unfortunately, the proofs of [4, Theorem 2.1] and [5, Theorem 2.2] are not complete, see [6, Remark 4.2] for details.
Concerning the strongly degenerate case (\(1<\alpha <2\)) with no singularity (\(\mu =0\)), in [7], the authors analyze the null controllability of a degenerate parabolic equation with a degenerate one-order transport term, but they consider a unweighted Dirichlet boundary control at the left endpoint. In [8,9,10], the authors consider the null controllability of 1D degenerate parabolic equations with first-order terms, but they use interior controls.
Concerning the case \(\alpha +\beta <1\), in [6], the authors prove the null controllability of the equation in the system (1) with initial conditions in \(L^2_\beta (0,1)\) but considering a weighted Dirichlet boundary condition at the left endpoint. There and here, suitable versions of a Hardy inequality are proved to assure the well-posedness of the corresponding systems, see Proposition 3.3.
In the case \(\alpha +\beta =1\), the Hardy inequality does not hold anymore, although there exist some generalized Hardy-type inequalities that can be used to obtain controllability results, see [11]. In this work, we choose to utilize certain results from singular Sturm–Liouville theory to establish the well-posedness of the system (6).
1.2 Neumann boundary conditions
In this context, in [12], the author considers the system (1) with a strong degeneracy \(1\le \alpha <2\), without singularity and drift, i.e., \(\beta =\mu =0\), and a homogeneous weighted Neumann condition at \(x=0\). In that work, the author uses the flatness approach to prove the null controllability by means of controls (in the Gevrey class) acting in the right endpoint. The last result was improved in [13], where the authors use \(L^2\) controls and prove that the controllability cost behaves as \(\textrm{e}^{(2-\alpha )^{-2}/T}\).
Recently, in [14], the authors use the flatness approach to prove the null controllability of the following system,
where the nonnegative function a may vanish strongly at \(x = 0\), the potential q may be singular at \(x = 0\), and \(\rho (x)>0\). In this case, the control h is in a Gevrey class and acts at the right endpoint. Only weakly degenerate parabolic equations were analyzed in [15].
As far as we know, this is the first time the null controllability of the system (1) has been proved with a weighted Neumann boundary control at the left endpoint. We also get lower and upper estimates of the cost of controllability.
2 Main results
The first goal of this work is to provide a notion of a weak solution for the system (1) and show the well-posedness of this problem in suitable interpolation spaces. Here, we consider a weighted Neumann boundary condition at the left endpoint to compensate for the singularity of the potential at this point. Then we use the moment method introduced by Fattorini and Russell in [16] to prove the null controllability and show an upper bound estimate of the cost of controllability. Next, we use a representation theorem for analytic functions of exponential type to get a lower bound estimate of the cost of controllability.
Consider the weighted Lebesgue space \(L^2_\beta (0,1):=L^2((0,1);x^\beta \textrm{d}x)\), \(\beta \in {\mathbb {R}}\), endowed with the inner product
and its corresponding norm is denoted by \(\Vert \cdot \Vert _{\beta }\).
In Sect. 3, we show the well-posedness of the system (1) with initial data in \(L^2_\beta (0,1)\), although the solution u(t) lives in an interpolation space \({\mathcal {H}}^{-s}\). Once we know the system (1) admits a unique solution for initial conditions in \(L^2_\beta (0,1)\), it is said the system (1) is null controllable 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 null controllability. Consider the set of admissible controls
Then the cost of the controllability is defined as
The main result of this work solves the case \(\alpha +\beta >1\). First, we introduce the constants
The last constants are chosen so that the equation in (19) yields an eigenfunction of the operator \({\mathcal {A}}\) defined in (16), see (18).
Theorem 2.1
Let \(T>0\) and \(\alpha ,\beta ,\mu ,\gamma \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha +\beta >1\), \(\mu \) and \(\gamma \) satisfying (2) and (3) respectively. 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 (1) 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
$$\begin{aligned} {\mathcal {K}}(T,\alpha ,\beta ,\mu )\le c M(T,\alpha ,\nu ,\delta )T^{1/2}\kappa _\alpha ^{-1/2} \exp \left( -\frac{T}{2}\kappa _\alpha ^2 j_{\nu ,1}^2\right) , \end{aligned}$$where \(j_{\nu ,1}\) is the first positive zero of the Bessel function \(J_\nu \) (defined in the Appendix), and
$$\begin{aligned} \begin{array}{rcl} M(T,\alpha ,\nu ,\delta ) &{} = &{}\displaystyle \left( 1+\frac{1}{(1-\delta )\kappa _\alpha ^2 T}\right) \left[ \exp \left( \frac{1}{\sqrt{2}\kappa _\alpha }\right) +\frac{1}{\delta ^3}\exp \left( \frac{3}{(1-\delta )\kappa _\alpha ^2 T}\right) \right] \\ &{}&{}\\ &{} &{}\displaystyle \times \exp \left( -\frac{(1-\delta )^{3/2}T^{3/2}}{8(1+T)^{1/2}}\kappa _\alpha ^3 j_{\nu ,1}^2\right) . \end{array} \end{aligned}$$ -
3.
Lower bound of the cost. There exists a constant \(c>0\) such that
$$\begin{aligned} \displaystyle \frac{c2^{\nu } \Gamma (\nu +1) \left| J_{\nu }^{\prime }\left( j_{\nu , 1}\right) \right| \exp {\left( \left( \frac{1}{2}-\frac{\log 2}{\pi }\right) j_{\nu ,2}\right) }}{\left( {2T \kappa _\alpha }\right) ^{1/2}\left( j_{\nu , 1}\right) ^{\nu }}&\exp&\left( -\left( j_{\nu ,1}^2+\frac{j_{\nu ,2}^2}{2}\right) \kappa _\alpha ^2 T\right) \\\le & {} {\mathcal {K}}(T,\alpha ,\beta ,\mu ), \end{aligned}$$
where \(j_{\nu ,2}\) is the second positive zero of the Bessel function \(J_\nu \).
To prove this result, we use the moment method by Fattorini and Russell, in particular, we use the biorthogonal family \((\psi _k)_k\) defined in (33) and constructed in [6].
We also exploit this approach to show the null controllability of the system when the control is located at the right endpoint. Hence, consider the following system
the corresponding set of admissible controls
and the cost of the controllability given by
Theorem 2.2
Let \(T>0\) and \(\alpha ,\beta ,\mu ,\gamma \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha +\beta >1\), \(\mu \) and \(\gamma \) satisfying (2) and (3), respectively. 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 (5) 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
$$\begin{aligned} \widetilde{{\mathcal {K}}}(T,\alpha ,\beta ,\mu )\le \frac{c M(T,\alpha ,\nu ,\delta )T^{1/2}}{(2\kappa _\alpha )^{\nu }\Gamma (\nu +1)} \left( \dfrac{2\nu +1}{T\textrm{e}}\right) ^{(2\nu +1)/4} \exp \left( -\frac{T}{4}\kappa _\alpha ^2 j_{\nu ,1}^2\right) . \end{aligned}$$ -
3.
Lower bound of the cost. There exists a constant \(c>0\) such that
$$\begin{aligned} \frac{c\exp {\left( \left( \frac{1}{2}-\frac{\log 2}{\pi }\right) j_{\nu ,2}\right) }}{T^{1/2} \kappa _\alpha ^{3/2}j_{\nu , 1}}\exp \left( -\left( j_{\nu ,1}^2+\frac{j_{\nu ,2}^2}{2}\right) \kappa _\alpha ^2 T\right) \le \widetilde{{\mathcal {K}}}(T,\alpha ,\beta ,\mu ). \end{aligned}$$
Remark 2.3
In [17], the authors prove the null controllability for a parabolic equation with inverse-square potential (i.e., \(\alpha =\beta =0\)) using Dirichlet \(L^2\)-controls acting at the right endpoint and considering a homogenous Dirichlet condition at the left endpoint. They also show that the cost of controllability behaves as \(\textrm{e}^{1/T}\). In [5], the authors improve the last result, they also prove that the cost of controllability behaves as \(\textrm{e}^{1/T}\) for \(0\le \alpha <1\), and the lower estimate is uniform with respect to \(\mu \le \mu (\alpha )\). Thus, with respect to the lower estimate, our technique is worst than the one used in [5, 17].
Finally, we also analyze the null controllability of the system when the parameters satisfy \(0\le \alpha <2\), \(\beta =1-\alpha \), and \(\mu <0\). Thus, consider the next system
The corresponding set of admissible controls is given by
and the cost of the controllability is given by
Theorem 2.4
Let \(T>0\) and \(\alpha ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\mu <0\). The next statements hold.
-
1.
Existence of a control. For any \(u_0\in L^2_{1-\alpha }(0,1)\), 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
$$\begin{aligned} \widehat{{\mathcal {K}}}(T,\alpha ,\mu )\le \frac{c M(T,\alpha ,\nu ,\delta )T^{1/2}}{\kappa _\alpha ^{1/2}\sqrt{-\mu }} \exp \left( -\frac{T}{2}\kappa _\alpha ^2 j_{\nu ,1}^2\right) , \end{aligned}$$where \(\nu =\nu (\alpha ,\mu ):=\sqrt{-\mu }/\kappa _\alpha \).
-
3.
Lower bound of the cost. There exists a constant \(c>0\) such that
$$\begin{aligned} \displaystyle \frac{c2^{\nu } \Gamma (\nu +1) \left| J_{\nu }^{\prime }\left( j_{\nu , 1}\right) \right| \exp {\left( \left( \frac{1}{2}-\frac{\log 2}{\pi }\right) j_{\nu ,2}\right) }}{({2T \kappa _\alpha })^{1/2}\sqrt{-\mu }\left( j_{\nu , 1}\right) ^{\nu }}&\exp&\left( -\left( j_{\nu ,1}^2+\frac{j_{\nu ,2}^2}{2}\right) \kappa _\alpha ^2 T\right) \\{} & {} \le \widehat{{\mathcal {K}}}(T,\alpha ,\mu ). \end{aligned}$$
Remark 2.5
Notice that the lower bound blows up as \(\mu \rightarrow 0^-=\mu (1)\). If \(\alpha =1, \mu =0\), the system (6) becomes the degenerate parabolic equation with a pure Dirichlet boundary condition.
This paper is organized as follows. In Sect. 3, we introduce suitable weighted Sobolev spaces and prove some results about the trace (at the endpoints) of functions in these spaces, as well as on the behavior of these functions at the endpoints, we also show an integration by parts formula. In that section, we prove that the operator given in (16) is diagonalizable, which allows the introduction of interpolation spaces for the initial data. Then we prove the system (1) is well-posed in this setting.
In Sect. 4, we prove Theorem 2.1 using the moment method, as a consequence, we get an upper estimate of \({\mathcal {K}}(T,\alpha ,\beta ,\mu )\). Then we use the representation theorem in Theorem A.5 to obtain a lower estimate of \({\mathcal {K}}(T,\alpha ,\beta ,\mu )\). In Sect. 5, we proceed as before to prove Theorem 2.2. In Sect. 6, we sketch the proof of Theorem 2.4. Finally in Sect. 7, we give some conclusions and open problems.
3 Functional setting and well-posedness
In the context of degenerate/singular parabolic equations, in [5, 6, 14, 15, 18], the authors noted that a suitable state space is a weighted Lebesgue space (\(L^2_\beta (0,1)\) in this work) for this kind of problem, together with weighted Sobolev spaces. The idea of using \(L^2_\beta (0,1)\) is that the expression \(x^\beta {\mathcal {A}}u(x)\), where \({\mathcal {A}}\) is the stationary operator given in (16), can be written in divergence form with a singular potential. This issue motivates us to use the inner product \(\langle \cdot , \cdot \rangle _{\alpha , \beta }\) provided below.
Thus, for \(\alpha ,\beta \in {\mathbb {R}}\), here we consider the weighted Sobolev space
endowed with the inner product
and its corresponding norm denoted by \(\Vert \cdot \Vert _{\alpha ,\beta }\).
3.1 Trace and integration by parts formula
The next result implies that we can talk about the trace at \(x=1\) of functions in \(H_{\alpha , \beta }^{1}(0,1)\), so in our systems, we can impose the usual homogeneous Dirichlet condition at the right endpoint.
Proposition 3.1
Let \(\alpha ,\beta \in {\mathbb {R}}\). Then \(H_{\alpha , \beta }^{1} \subset W^{1,1}(\varepsilon ,1)\) for all \(\varepsilon \in (0,1)\). In particular, \(H_{\alpha ,\beta }^{1}(0,1)\subset C((0,1])\), and \(|u|^2\in W^{1,1}(\varepsilon ,1)\) for all \(u\in H_{\alpha , \beta }^{1}(0,1),\varepsilon \in (0,1)\).
Proof
Let \(u\in H_{\alpha , \beta }^{1}(0,1)\). For \(\varepsilon \in (0,1),\delta \in {\mathbb {R}}\) fixed, there exists a constant \(c(\varepsilon ,\delta )>0\) such that \(x^\delta \le c(\varepsilon ,\delta )\), \(x\in (\varepsilon ,1]\), thus
Hence, we get the existence of the limit \(u(1):=\lim _{x\rightarrow 1^-}u(x)\), and \(u\in C([\varepsilon ,1])\). \(\square \)
Definition 3.2
For \(\alpha ,\beta \in {\mathbb {R}}\), consider the space
The classic Hardy inequality is typically proven for test functions defined on a finite interval. In our context, the following result shows that the Hardy inequality can be extended to functions in the weighted Sobolev space \(H_{\alpha , \beta ,N}^{1}\). As an application, this generalized Hardy inequality implies the weighted Poincaré inequality (10) for functions in \(H_{\alpha , \beta ,N}^{1}\).
Proposition 3.3
For \(\alpha ,\beta \in {\mathbb {R}}\) with \(\alpha +\beta >1\), the Hardy inequality
holds for any \(u\in H_{\alpha , \beta ,N}^{1}\). In particular, \(H_{\alpha , \beta ,N}^{1}\hookrightarrow L^2_{\alpha +\beta -2}(0,1)\).
Proof
Let \(u\in H_{\alpha , \beta ,N}^{1}\) and \(\varepsilon \in (0,1)\). Set \(\delta =\alpha +\beta \). Since \(|u|^2\in W^{1,1}(\varepsilon ,1)\) we have
since \(\delta > 1\) we get
for all \(\varepsilon \in (0,1)\). The result follows by the dominated convergence theorem. \(\square \)
The next result will allow us analyze the behavior at \(x=0\) of functions in \(H_{\alpha , \beta , N}^{1}\), see (9).
Proposition 3.4
Let \(\alpha ,\beta \in {\mathbb {R}}\) with \(\alpha +\beta >1\). Then \(x^{\delta }u\in W^{1,1}(0,1)\) for all \(u\in H_{\alpha , \beta , N}^{1}\) provided that \(\delta > (\alpha +\beta -1)/2\).
Proof
Let \(u\in H_{\alpha , \beta , N}^{1}\) and assume \(2\delta > \alpha +\beta -1\).
Since \((x^{\delta }u)_{x} = x^{\delta }u_{x}+\delta x^{\delta -1}u\), we compute
and Proposition 3.3 implies
Hence, \((x^{\delta }u)_{x}\in L^1(0,1)\). Notice that \(x^\delta \le x^{\delta -1}\) on (0, 1), thus \(x^{\delta }u\in L^1(0,1)\) and the result follows. \(\square \)
Remark 3.5
For \(\alpha ,\beta \in {\mathbb {R}}\) with \(\alpha +\beta >1\), the last result implies the existence of \(L_{\delta }:=\lim _{x\rightarrow 0^{+}}x^{\delta }u(x)\) provided that \(\delta > (\alpha + \beta - 1)/2\), in fact, \(L_{\delta } = 0\). Now choose any \(\delta > (\alpha + \beta - 1)/2\) so
therefore
The last equality will be useful to prove the integration by parts formula given in (15).
From now on, we assume \(\alpha < 2\) and \(\alpha + \beta >1\). For any \(u\in H^1_{\alpha ,\beta ,N}\), we obtain the weighted Poincaré inequality from Proposition 3.3:
therefore
is an equivalent norm to \(\Vert u\Vert _{\alpha , \beta }\) in \(H_{\alpha , \beta , N}^{1}\).
For \(\mu <\mu (\alpha +\beta )\), Proposition 3.3 also implies that
is an equivalent norm to \(\Vert u\Vert _{\alpha ,\beta ,N}\) in \(H_{\alpha , \beta , N}^{1}\). In fact, we have
Since \(C_c^\infty (0,1)\subset H_{\alpha , \beta , N}^{1}\subset L^2_\beta (0,1),\) and (10) implies that the inclusion \((H_{\alpha , \beta ,N}^1,\Vert \cdot \Vert _*)\hookrightarrow L^2_\beta (0,1)\) is continuous, the following definition makes sense.
Definition 3.6
For \(\alpha ,\beta \in {\mathbb {R}}\) with \(\alpha <2\) and \(\alpha +\beta >1\), consider the Gelfand triple \(\left( (H_{\alpha , \beta ,N}^1,\Vert \cdot \Vert _*), L^2_\beta (0,1), H_{\alpha , \beta ,N}^{-1}\right) \), i.e \(H_{\alpha , \beta ,N}^{-1}\) stands for the dual space of \((H_{\alpha , \beta ,N}^1,\Vert \cdot \Vert _*)\) with respect to the pivot space \(L^2_\beta (0,1)\):
The inner product \(\langle \cdot ,\cdot \rangle _*\) induces an isomorphism \({\mathcal {A}}:H_{\alpha , \beta ,N}^1\rightarrow H_{\alpha , \beta ,N}^{-1}\) given by
Let \(D({\mathcal {A}}):={\mathcal {A}}^{-1}(L^2_\beta (0,1))=\{u\in H_{\alpha , \beta ,N}^{1}: {\mathcal {A}} u\in L^2_\beta (0,1)\}=\{u\in H_{\alpha , \beta ,N}^{1}: \exists f\in L^2_\beta (0,1) \text { such that }\langle u,v \rangle _*=\langle f,v\rangle _\beta , \text { for all } v\in H_{\alpha , \beta ,N}^{1} \}\).
The next result gives a handy characterization of \(D({\mathcal {A}})\). It shows the behavior of the derivative of functions in \(D({\mathcal {A}})\) at the endpoints, see (11) and (12), and also provides an integration by parts formula, see (15).
Proposition 3.7
For \(\alpha ,\beta ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha +\beta >1\) and \(\mu <\mu (\alpha +\beta )\), we have
Proof
Let H be the set on the right-hand side, we will show that \(D({\mathcal {A}})=H\).
Pick \(u \in D({\mathcal {A}})\), then there exists \(f \in L^{2}_\beta (0,1)\) such that
In particular,
hence
which implies
therefore \(u\in H\).
Now let \(u \in H\). We claim that \(x^{\delta } u_{x} \in W^{1,1}(0,1)\) for all \(\delta > (\alpha +\beta +1)/2\). Just apply (8) with \(\delta -1\) instead of \(\delta \) to get that \(x^{\delta -1}u_x\in L^1(0,1)\), in particular \(x^{\delta }u_x\in L^1(0,1)\). On the other hand, we have
Notice the last quantity is finite by Proposition 3.3.
Thus, we get the existence of the limit
and we also have that \(\lim _{x\rightarrow 0^+}x^{\delta }u_{x}(x)=0\) provided that \(\delta > (\alpha +\beta +1)/2\), see Remark 3.5. As in the proof of (9), we can see that
Now consider any \(v\in H^{1}_{\alpha ,\beta ,N}\). We claim that \(x^{\alpha +\beta } u_{x} v \in W^{1,1}(0,1)\):
and
On the other hand, (9), (11), and (12) imply that
Thus, from (13), we get
for all \(u \in H, v \in H_{\alpha , \beta , N}^{1}\). Therefore, \(u \in D({\mathcal {A}})\). \(\square \)
3.2 Semigroup theory
For \(\alpha ,\beta ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha +\beta >1\), \(\mu <\mu (\alpha +\beta )\), we consider the unbounded operator \({\mathcal {A}}:D({\mathcal {A}})\subset L^2_\beta (0,1)\rightarrow L^2_\beta (0,1)\) given by
From Proposition 9 in [19, p. 370], we have that \({\mathcal {A}}\) is a closed operator with \(D({\mathcal {A}})\) dense in \(L^2_\beta (0,1)\). We also have that \({\mathcal {A}}:(D({\mathcal {A}}),\Vert \cdot \Vert _{D({\mathcal {A}})})\rightarrow L^2_\beta (0,1)\) is an isomorphism, where
The next result shows that \({\mathcal {A}}\) is a diagonalizable self-adjoint operator whose Hilbert basis of eigenfunctions can be written in terms of a Bessel function of the first kind \(J_{\nu }\) and its corresponding zeros \(j_{\nu ,k}\), \(k\ge 1\), located in the positive half line. In the appendix, we give some properties of Bessel functions and their zeros.
Proposition 3.8
\(-{\mathcal {A}}\) is a negative self-adjoint operator. Furthermore, the family
is an orthonormal basis for \(L^2_\beta (0,1)\) such that
where \(\nu \) is defined in (4).
Proof
From (15), we get that \({\mathcal {A}}\) is a symmetric operator. Letting \(u=v\in D({\mathcal {A}})\) in (15) and using Proposition 3.3, we obtain that \(-{\mathcal {A}}\le 0\).
We claim that \(\text {Ran}(I+{\mathcal {A}})=L^2_\beta (0,1)\): Let \(f \in L_{\beta }^{2}(0,1)\) be given. Since the inner product \(\langle \cdot ,\cdot \rangle _\beta +\langle \cdot ,\cdot \rangle _*\) is equivalent to \(\langle \cdot ,\cdot \rangle _{\alpha ,\beta }\) in \(H_{\alpha , \beta , N}^{1}\) and \(f\in (H_{\alpha , \beta , N}^{1},\Vert \cdot \Vert _{\alpha ,\beta })'\), the Riesz representation theorem implies that there exists a unique \(u\in H_{\alpha , \beta , N}^{1}\) such that
for all \(v\in H_{\alpha , \beta , N}^{1}\). Therefore,
thus \(u\in D({\mathcal {A}})\) and \(u+{\mathcal {A}} u=f\).
It follows that \(-{\mathcal {A}}\) is m-dissipative in \(L^2_\beta (0,1)\) and Corollary 2.4.10 in [20, p. 24] implies that \(-{\mathcal {A}}\) is self-adjoint.
In [21], it was proved that the family
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 \({\mathcal {U}}\Theta _k=\Phi _k\), \(k\ge 1\), therefore \(\Phi _k\), \(k\ge 1\), is an orthonormal basis for \(L^2_\beta (0,1)\). We also can see that \(\Phi _k\in H^1_{\alpha ,\beta ,N}\) using that \(\nu >0,\) (A1) and (A2).
Now we set \(w(x)=y(z)\) with \(z=cx^a\), \(a,c>0\). Assume that \(y=J_\nu \). Therefore, y satisfies the differential Eq. (A3), i.e.,
which implies that
Then we set \(v(x)=x^bw(x)\), \(b\in {\mathbb {R}}\). Hence,
Finally, we take \(a=\kappa _\alpha , b=(1-\alpha -\beta )/2\), and \(c=j_{\nu ,k}\), \(k\ge 1\), to get \(\Phi _k(1)=0\) and \({\mathcal {A}}\Phi _k=\lambda _k\Phi _k\) for all \(k \ge 1\). \(\square \)
Then \(({\mathcal {A}},D({\mathcal {A}}))\) is the infinitesimal generator of a diagonalizable self-adjoint analytic semigroup of contractions in \(L^2_{\beta }(0,1)\). Thus, we 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 \(s\in {\mathbb {R}}\).
3.3 Existence and uniqueness of solutions to system (1)
For \(\delta \in {\mathbb {R}}\) and a function \(z:(0,1)\rightarrow {\mathbb {R}}\), we introduce the notion of \(\delta \)-generalized limit of z at \(x=0\) as follows
Now we consider a convenient definition of a weak solution for system (1), we multiply the equation in (1) by \(x^\beta \varphi (\tau )=x^\beta S(\tau -t)z^{\tau }\), integrate by parts (formally), and take the expression obtained.
Definition 3.9
Let \(T>0\) and \(\alpha ,\beta ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha + \beta >1\), \(\mu <\mu (\alpha +\beta )\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) for some \(s > 0\). A weak solution of (1) 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
where \(\gamma =\gamma (\alpha ,\beta ,\mu )\) is given in (3).
The next result shows the existence of weak solutions for the system (1) under suitable conditions on the parameters \(\alpha ,\beta ,\mu ,\gamma \) and s. The proof is similar to the proof of Proposition 2.9 in [6].
Proposition 3.10
Let \(T>0\) and \(\alpha ,\beta \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha + \beta >1\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) such that \(s>\nu \), where \(\nu \) is given in (4). Then formula (20) 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 (1) (in the sense of (20)) belongs to \({C}^{0}\left( [0, T]; {\mathcal {H}}^{-s}\right) \) and fulfills
Proof
Fix \(\tau >0\). Let \(u(\tau )\in {\mathcal {H}}^{-s}\) be determined by the condition (20), 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
Using Lemma A.3 and (A10), we obtain that there exists a constant \(C=C(\alpha ,\beta ,\mu )>0\) such that
hence, (A9) implies that there exists a constant \(C=C(\alpha ,\beta ,\mu )>0\) 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 \(h>0\) be small enough and \(z\in {\mathcal {H}}^s\) given as in (21). Thus, proceeding as in the last inequalities, we have
where
as \(h\rightarrow 0^+\).
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 3.11
In the following section, we will consider initial conditions in \(L^2_\beta (0,1)\). Notice that \(L^2_\beta (0,1)\subset H^{-\nu -\delta }\) for all \(\delta >0\), and we can apply Proposition 3.10 with \(s=\nu +\delta \), \(\delta >0\), then the corresponding solutions will be in \(C^0([0, T ], H^{-\nu -\delta })\).
4 Proof of Theorem 2.1: control at the left endpoint
4.1 Upper estimate of the cost of the null controllability
In this section, we use the method moment, introduced by Fattorini and Russell in [16], to prove the null controllability of the system (1). In [6, Sect. 3], the authors construct a biorthogonal family \(\displaystyle \{\psi _k\}_{k\ge 1}\subset L^2(0,T)\) to the family of exponential functions \(\{\textrm{e}^{-\lambda _{k}(T-t)}\}_{k\ge 1}\) on [0, T], i.e., that satisfies
That construction will help us to get an upper bound for the cost of the null controllability of the system (1). Here, we sketch the process to get the biorthogonal family \(\displaystyle \{\psi _k\}_{k\ge 1}\), see [6, Sect. 3] for details.
4.1.1 Construction of the biorthogonal family
Consider the Weierstrass infinite product
From (A8), we have that \(j_{\nu , k}=O(k)\) for k large; thus, the infinite product is well-defined and converges absolutely in \({\mathbb {C}}\). Hence, \(\Lambda (z)\) is an entire function with simple zeros at \(i(\kappa _\alpha j_{\nu , k})^2=i\lambda _k\), \(k\ge 1\). It follows that
is an entire function that satisfies \(\Psi _{k}(i\lambda _{l})=\delta _{kl}\) for all \(k,l\ge 1\). Since \(\Psi _{k}(x)\) is not in \(L^2({\mathbb {R}})\), we need to fix this by using a suitable “complex multiplier", to do this we follow the approach introduced in [22].
For \(\theta ,a>0\), 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
Clearly \(H_{a,\theta }(z)\) is an entire function. The following result gives additional information about \(H_{a,\theta }(z)\).
Lemma 4.1
The function \(H_{a,\theta }\) fulfills the following inequalities
where \(c>0\) does not depend on a and \(\theta \).
We refer to [22, pp. 85–86] for the details.
For \(k\ge 1\), consider the entire function \(F_{k}\) given as
For \(\delta \in (0,1)\), we set
Lemma 4.2
For each \(k\ge 1\), the function \(F_{k}(z)\) satisfies 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 \(F_{k}(i\lambda _{l})=\delta _{kl}\) for all \(k,l\ge 1\).
-
iv)
Furthermore, there exists a constant \(c>0\), independent of \(T,\alpha \) and \(\delta \), such that
$$\begin{aligned} \left\| F_{k}\right\| _{L^{1}({\mathbb {R}})} \le \frac{C(T, \alpha ,\delta )}{\lambda _{k}\left| \Lambda ^{\prime }\left( i \lambda _{k}\right) \right| } \exp \left( -\frac{a\lambda _k}{2\sqrt{\theta +1}}\right) , \end{aligned}$$(31)where
$$\begin{aligned} C(T, \alpha ,\delta )=c\sqrt{\theta +1}\left( \exp \left( {\frac{1}{\sqrt{2}\kappa _\alpha }}\right) +\sqrt{\theta +1}\frac{\kappa _\alpha ^2}{\delta ^3}\exp \left( \frac{3 \theta }{4}\right) \right) . \end{aligned}$$(32)
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 \(\eta _k\). Hence,
is the family we are looking for.
Since \(\eta _k, F_k\in L^1({\mathbb {R}})\), the inverse Fourier theorem yields
hence (33) implies that \(\psi _k\in C([0,T])\), and using (31), we have
4.1.2 Proof of Theorem 2.1
Now, we are ready to prove the null controllability of the system (1). Let \(u_{0}\in L^{2}_\beta (0,1)\). Then consider its Fourier series with respect to the orthonormal basis \(\{\Phi _{k}\}_{k\ge 1}\),
We set
Since \(\{\psi _k\}\) is biorthogonal to \(\{\textrm{e}^{-\lambda _k(T-t)}\}_{k\ge 1}\), we have
Let \(u\in C([0,T];H^{-s})\) that satisfies (20) for all \(\tau \in (0,T]\), \(z^\tau \in H^s\). In particular, for \(\tau =T\), we take \(z^T=\Phi _k\), \(k\ge 1\), then the last equality implies that
hence \(u(\cdot , T)\equiv 0\).
It just remains to estimate the norm of the control f. From (34) and (36), we get
Using [23, Chap. XV, p. 498, eq. (3)], we can write
therefore
and by using (18) and (A10) we get
From (37), the last two equalities and using that \(\lambda _k\ge \lambda _1\), it follows that
Using the Cauchy–Schwarz inequality, the fact that \(j_{\nu ,k}\ge (k-1/4)\pi \) (by (A9)) and (35), we obtain that
Using the definition of \(a,\theta \) in (30) and the facts \(\theta >0\), \(\delta \in (0,1)\), \(0<\kappa _\alpha \le 1\), we get that
therefore
and using the definition of \(\lambda _{1}\) the result follows.
4.2 Lower estimate of the cost of the null controllability
In this section, we get a lower estimate of the cost \({\mathcal {K}}={\mathcal {K}}(T,\alpha ,\beta ,\mu )\). We set
For \(\varepsilon >0\) small enough, there exists \(f\in U(\alpha ,\beta ,\mu ,T,u_0)\) such that
Then in (20), we set \(\tau =T\) and take \(z^\tau =\Phi _k\), \(k\ge 1\), to obtain
from (40) and (A10), 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 apply Theorem A.5 to the function F(z) given in (44). In this case, (43) implies that \(A\le T/2\). Also notice that \(\Im \left( a_{k}\right) >0\), \(k\ge 1\), to get
Using the definition of the constants \(a_k\)’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
where
The result follows by taking
5 Proof of Theorem 2.2: control at the right endpoint
5.1 Existence and uniqueness of solutions to system (5)
Here, we analyze the null controllability of the system (5) where \(\alpha +\beta >1\), \(0\le \alpha < 2\), \(\mu \) and \(\gamma \) satisfy (2) and (3) respectively. As in Sect. 4, we give a suitable definition of a weak solution for the system (5).
Definition 5.1
Let \(T>0\) and \(\alpha ,\beta ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha + \beta >1\), \(\mu <\mu (\alpha +\beta )\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) for some \(s > 0\). A weak solution of (5) 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
where \(\gamma =\gamma (\alpha ,\beta ,\mu )\) is given by (3).
The next result shows the existence of weak solutions for the system (5) under certain conditions on the parameters \(\alpha ,\beta ,\mu ,\gamma \) and s.
Proposition 5.2
Let \(T>0\) and \(\alpha ,\beta \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\alpha + \beta >1\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) such that \(s>1/2\). Then formula (51) defines for each \(\tau \in [0, T ]\) a unique element \(u(\tau ) \in {\mathcal {H}}^{-s}\) that can be written as
where \({\widetilde{B}}(\tau )\) is the strongly continuous family of bounded operators \({\widetilde{B}}(\tau ): L^{2}(0,T) \rightarrow {\mathcal {H}}^{-s}\) given by
Furthermore, the unique weak solution u on [0, T] to (5) (in the sense of (51)) 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 (51); hence,
where
Let \(z^{\tau } \in {\mathcal {H}}^{s}\) given by
therefore
By (17), we get
hence (A9) implies that there exists \(C=C(\alpha ,\beta ,\mu )>0\) 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 \(h>0\) be small enough and \(z\in {\mathcal {H}}^s\) given as in (52). Thus, proceeding as in the last inequalities, we have
where \(I(\tau ,k,h)\rightarrow 0\) as \(h\rightarrow 0^+\), see (22). \(\square \)
Remark 5.3
In the following subsections, we will consider initial conditions in \(L^2_\beta (0,1)\). We can apply Proposition 5.2 with \(s=1/2+\delta \), \(\delta >0\), then the corresponding solutions will be in \(C^0([0, T ], H^{-1/2-\delta })\).
5.2 Upper estimate of the cost of the null controllability
5.2.1 Proof of Theorem 2.2
We are ready to prove the null controllability of the system (5). Let \(u_{0}\in L^{2}_\beta (0,1)\) given as follows
We set
Since the sequence \(\{\psi _k\}\) is biorthogonal to \(\{\textrm{e}^{-\lambda _k(T-t)}\}_{k\ge 1}\), we have
Let \(u\in C([0,T];H^{-s})\) be the weak solution of system (5). In particular, for \(\tau =T\), we take \(z^T=\Phi _k\), \(k\ge 1\), then (51) and (56) imply that
therefore \(u(\cdot , T) = 0\).
It just remains to estimate the norm of the control f. From (34), (38), (53), and (55), we get
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 (A9)) and (35), we obtain that
and the result follows by (39).
5.3 Lower estimate of the cost of the null controllability at the right endpoint
Here, we just give a sketch of the proof of a lower estimate for the cost \(\widetilde{{\mathcal {K}}}=\widetilde{{\mathcal {K}}}(T,\alpha ,\beta ,\mu )\). Consider \(u_0\in L^2_\beta (0,1)\) given in (40).
For \(\varepsilon >0\) small enough, there exists \(f\in {\widetilde{U}}(\alpha ,\beta ,\mu ,T,u_0)\) such that
Then in (51), we set \(\tau =T\) and take \(z^T=\Phi _k\), \(k\ge 1\), to obtain
from (40) and (53), it follows that
Next, we proceed as in (4.2)–(50). But in this case, the corresponding functions v and F satisfy
Hence, we can see that
where \({\widetilde{h}}(\alpha ,\beta ,\mu , T)=T^{1/2}\kappa ^{3/2}_{\alpha }j_{\nu ,1}/\sqrt{2}\). The result follows by taking \(\delta =\kappa _\alpha ^2 (j_{\nu ,2})^2/2\) and then letting \(\varepsilon \rightarrow 0^+\).
6 Proof of Theorem 2.4: the critical case \(\alpha +\beta =1\)
Concerning the case \(\alpha +\beta <1\), in [6], we show the system (1) is well-posed when considering suitable weighted Dirichlet condition at the left endpoint and prove the null controllability of the corresponding system. In both cases (\(\alpha +\beta <1\) and \(\alpha +\beta >1\)), our approach is based on the validity of the Hardy inequality, see Proposition 3.3 and [6, Proposition 2.3]. If \(\alpha +\beta =1\), then \(\mu (\alpha +\beta )=0\), and the corresponding Hardy inequality does not provide any information. Thus, to solve the case \(\alpha +\beta =1\), we use the singular Sturm–Liouville theory, see [24] for the definitions used here.
6.1 Singular Sturm–Liouville theory
Assume that \(0\le \alpha <2\), \(\beta \in {\mathbb {R}}\) such that \(\beta = 1-\alpha \), and \(\mu < 0\). Consider the differential expression M defined by
where \(\displaystyle p(x) = x, q(x) = -\mu x^{-1}, w(x) = x^{1-\alpha }\), \('=\frac{\textrm{d}}{\textrm{d}x}\).
Clearly,
thus, Mu is defined a.e. for functions u such that \(u, pu'\in AC_{\text {loc}}(0,1)\), where \(AC_{\text {loc}}(0,1)\) is the space of all locally absolutely continuous functions in (0, 1).
Notice that the operator \({\mathcal {A}}\) given in (16), with \(\beta = 1-\alpha \), can be written as \({\mathcal {A}} =w^{-1}M\). Now, consider
Recall that the Lagrange form associated with M is given as follows
According to [24, Definition 10.4.3], a pair of real-valued functions \(y_{0}, z_{0}\in D_{\max }\) is called a (BC) basis at \(x=0\) if they satisfy \([z_{0},y_{0}](0)=1\). Similarly, a pair of real-valued functions \(y_{1}, z_{1}\in D_{\max }\) is called a (BC) basis at \(x=1\) if they satisfy \([z_{1}, y_{1}](1)=1\). Moreover, a pair of real-valued functions \(y,z\in D_{\max }\) is called a (BC) basis on (0, 1) if they form a (BC) basis at \(x=0\) and at \(x=1\).
Proposition 6.1
Let \(0\le \alpha <2\), \(\mu < 0\), and \(\nu =\sqrt{-\mu }/\kappa _\alpha \). Then \({\mathcal {A}}:D({\mathcal {A}})\subset L^2_{1-\alpha }(0,1)\rightarrow L^2_{1-\alpha }(0,1)\) is a self-adjoint operator. Furthermore, when \(\beta = 1-\alpha \) the family given in (17) is an orthonormal basis for \(L^2_{1-\alpha }(0,1)\) such that
Proof
First, we refer to [24, Definition 7.3.1] to see that \(x=1\) is a regular point because \(1/p,q,w\in L^1(1/2,1)\).
Case i) Suppose \(\sqrt{-\mu }<\kappa _\alpha \).
First, we will build a (BC) basis \(\{y_{0},z_{0}\}\) at \(x=0\) and a (BC) basis \(\{y_{1}, z_{1}\}\) at \(x=1\).
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 \(y_{0}, z_{0}\in L^2_{1-\alpha }(0,1)\) are linearly independent solutions of \(Mu=0u\), it follows that \(x=0\) is limit-circle (LC), see [24, Definition 7.3.1, Theorem 7.2.2].
In the same way, consider also the functions
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 [24], 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 [24, Definition 10.4.3], where
The matrices
satisfy the hypothesis in [24, 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 [24, Lemma 10.2.3]), and \(\lim _{x\rightarrow 0^+}y_0/z_0(x)=0\). Hence, the result follows.
Case ii) \(\sqrt{-\mu }\ge \kappa _\alpha \). Since \(z_0\notin L^2_{1-\alpha }(0,1)\), then \(x=0\) is limit point (LP). The result follows using Theorem 10.4.4 in [24] with \(A_1=1, A_2=0\).
When \(\alpha +\beta =1\), notice that \(\Phi _k\in D({\mathcal {A}})\), \(k\ge 1\), thus the second part of the proof follows using the computations in the proof of Proposition 3.8. \(\square \)
6.2 Existence and uniqueness of solutions to system (6)
As in Sect. 3, consider the corresponding interpolation spaces \({\mathcal {H}}^s\), \(s\in {\mathbb {R}}\).
Definition 6.2
Let \(T>0\) and \(\alpha ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\mu <0\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) for some \(s > 0\). A weak solution of (6) 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 last definition coincides with the one given in [6, Definition 2.8], provided that \(\beta =1-\alpha \). Thus, the next result about the existence of weak solutions for the system (6) can be proved in the same way, under suitable conditions on the parameters \(\alpha ,\beta ,\mu ,\) and s.
Proposition 6.3
Let \(T>0\) and \(\alpha ,\mu \in {\mathbb {R}}\) with \(0\le \alpha <2\), \(\mu <0\). Let \(f \in L^2(0,T)\) and \(u_0\in {\mathcal {H}}^{-s}\) such that \(s>\sqrt{-\mu }/\kappa _\alpha \). Then formula (61) defines for each \(\tau \in [0, T ]\) a unique element \(u(\tau ) \in {\mathcal {H}}^{-s}\) that can be written as
where \({\widehat{B}}(\tau )\) is the strongly continuous family of bounded operators \({\widehat{B}}(\tau ): L^{2}(0,T) \rightarrow {\mathcal {H}}^{-s}\) given by
Furthermore, the unique weak solution u on [0, T] to (6) (in the sense of (20)) belongs to \({C}^{0}\left( [0, T]; {\mathcal {H}}^{-s}\right) \) and fulfills
Thus, we can follow exactly the same steps as in [6, Sects. 2,3], just take \(\beta =1-\alpha \), to get Theorem 2.4.
7 Conclusion, applications, and open problems
As a consequence of Theorems 2.1, 2.2, and 2.4, we obtain the following qualitative result.
Remark 7.1
Using (A4), (A5), (A6), and (A7), we obtain that
thus, the control costs \({\mathcal {K}}(T,\alpha , \beta ,\mu )\) and \(\widehat{{\mathcal {K}}}(T,\alpha ,\mu )\) blow up when \(\alpha \rightarrow 2^{-}\) and/or \(T\rightarrow 0^{+}\). Since \(j_{v,1}<j_{v,2}\), clearly \(\widetilde{{\mathcal {K}}}(T,\alpha , \beta ,\mu )\rightarrow +\infty \) when \(\alpha \rightarrow 2^{-}\) and/or \(T\rightarrow 0^{+}\).
When \(T\rightarrow 0^{+}\), notice that the upper bounds of the control costs for the systems (1) and (5) explode much faster than the lower bounds.
In the critical case \(\alpha + \beta = 1\), we did not have Hardy’s inequality at our disposal, so we could not proceed as in Sect. 3. But we found out that the theory of singular Sturm–Liouville operators is a powerful tool to solve degenerate/singular evolution equations, for instance, in [25], we show the null controllability of the system (1) with a weighted Robin control acting at the left endpoint.
Next, we would like to highlight that the issues addressed in this work are closely connected to various applications. These include the Crocco-type equation, derived from the analysis of the velocity field in laminar flow over a flat plate [13], the Black–Scholes model stemming from option pricing concerns [26], and the Budyko–Sellers model originating from climate-related problems [27], among others.
In [28], the authors analyzed the rapid stabilization of the following system
where f denotes the control, the initial condition \(u_{0}\) is chosen in \(L^2(0,1)\) and \(0\le \alpha <1\).
Using spectral tools associated to the operator \({\mathcal {A}} u:= \left( x^{\alpha }u_{x}\right) _{x}\), they obtain
Theorem 7.2
For any \(\lambda >0\), there exists \(C(\lambda ) > 0\) and a feedback law \(f (t) = K(u(t))\), where \(K\in L^{2}(0,1)'\), such that for any \(u_{0}\in L^2(0,1)\), there exists a unique solution u of (62) that verifies, for any \(t \ge 0\):
An open problem is to analyze the rapid stabilization of the system (1), corresponding to the strongly degenerate case \(1\le \alpha < 2\).
Another open problem would be to improve the given lower bounds on the control cost for each system we have considered.
Regarding the Sturm–Liouville theory, we believe that Theorem 10.6.1 in [24] would be useful for solving and controlling more general degenerate/singular evolution equations.
Data availibility
This declaration is not applicable.
References
Vancostenoble, J.: Improved Hardy–Poincaré inequalities and sharp Carleman estimates for degenerate/singular parabolic problems. Discrete Contin. Dyn. Syst. Ser. S 4(3), 761–790 (2011). https://doi.org/10.3934/dcdss.2011.4.761
Gueye, M.: Exact boundary controllability of 1-D parabolic and hyperbolic degenerate equations. SIAM J. Control Optim. 52(4), 2037–2054 (2014). https://doi.org/10.1137/120901374
Cannarsa, P., Martinez, P., Vancostenoble, J.: The cost of controlling weakly degenerate parabolic equations by boundary controls. Math. Control Relat. Fields 7(2), 171–211 (2017). https://doi.org/10.3934/mcrf.2017006
Biccari, U.: Boundary controllability for a one-dimensional heat equation with a singular inverse-square potential. Math. Control Relat. Fields 9(1), 191–219 (2019). https://doi.org/10.3934/mcrf.2019011
Biccari, U., Hernández-Santamaría, V., Vancostenoble, J.: Existence and cost of boundary controls for a degenerate/singular parabolic equation. Math. Control Relat. Fields 12(2), 495–530 (2022). https://doi.org/10.3934/mcrf.2021032
Galo, L., López-García, M.: Boundary controllability for a 1d degenerate parabolic equation with drift and a singular potential. Math. Control Relat. Fields (2023). https://doi.org/10.3934/mcrf.2023027
Gueye, M., Lissy, P.: Singular optimal control of a 1-D parabolic-hyperbolic degenerate equation. ESAIM Control Optim. Calc. Var. 22(4), 1184–1203 (2016). https://doi.org/10.1051/cocv/2016036
Du, R.: Null controllability for a class of degenerate parabolic equations with the gradient terms. J. Evol. Equ. 19(2), 585–613 (2019). https://doi.org/10.1007/s00028-019-00487-8
Flores, C., Teresa, L.: Carleman estimates for degenerate parabolic equations with first order terms and applications. C. R. Math. Acad. Sci. Paris 348(7–8), 391–396 (2010). https://doi.org/10.1016/j.crma.2010.01.007
Flores, C., Teresa, L.: Null controllability of one dimensional degenerate parabolic equations with first order terms. Discrete Contin. Dyn. Syst. Ser. B 25(10), 3963–3981 (2020). https://doi.org/10.3934/dcdsb.2020136
Vancostenoble, J.: Improved Hardy–Poincaré inequalities and sharp Carleman estimates for degenerate/singular parabolic problems. Discrete Contin. Dyn. Syst. Ser. S 4(3), 761–790 (2011). https://doi.org/10.3934/dcdss.2011.4.761
Moyano, I.: Flatness for a strongly degenerate 1-D parabolic equation. Math. Control Signals Systems 28(4), 28–22 (2016)
Cannarsa, P., Martinez, P., Vancostenoble, J.: The cost of controlling strongly degenerate parabolic equations. ESAIM Control Optim. Calc. Var. 26, 2–50 (2020). https://doi.org/10.1051/cocv/2018007
Benoit, A., Loyer, R., Rosier, L.: Null controllability of strongly degenerate parabolic equations. ESAIM Control Optim. Calc. Var. 29, 48–36 (2023). https://doi.org/10.1051/cocv/2023016
Martin, P., Rosier, L., Rouchon, P.: Null controllability of one-dimensional parabolic equations by the flatness approach. SIAM J. Control Optim. 54(1), 198–220 (2016). https://doi.org/10.1137/14099245X
Fattorini, H.O., Russell, D.L.: Exact controllability theorems for linear parabolic equations in one space dimension. Arch. Rational Mech. Anal. 43, 272–292 (1971). https://doi.org/10.1007/BF00250466
Martinez, P., Vancostenoble, J.: The cost of boundary controllability for a parabolic equation with inverse square potential. Evol. Equ. Control Theory 8(2), 397–422 (2019). https://doi.org/10.3934/eect.2019020
Gueye, M., Lissy, P.: Singular optimal control of a 1-D parabolic-hyperbolic degenerate equation. ESAIM Control Optim. Calc. Var. 22(4), 1184–1203 (2016). https://doi.org/10.1051/cocv/2016036
Dautray, R., Lions, J.-L.: Mathematical Analysis and Numerical Methods for Science and Technology, vol. 2. Springer, Berlin (1988)
Cazenave, T., Haraux, A.: An Introduction to Semilinear Evolution Equations, vol. 13. The Clarendon Press, Oxford University Press, New York (1998)
Hochstadt, H.: The mean convergence of Fourier-Bessel series. SIAM Rev. 9, 211–218 (1967). https://doi.org/10.1137/1009034
Tenenbaum, G., Tucsnak, M.: New blow-up rates for fast controls of Schrödinger and heat equations. J. Differ. Equ. 243(1), 70–100 (2007). https://doi.org/10.1016/j.jde.2007.06.019
Watson, G.N.: A Treatise on the Theory of Bessel Functions. Cambridge University Press, Cambridge (1995)
Zettl, A.: Sturm–Liouville Theory. Mathematical Surveys and Monographs, vol. 121. American Mathematical Society, Providence (2005)
Galo-Mendoza, L., López-García, M.: Boundary controllability for a 1D degenerate parabolic equation with a Robin boundary condition. Math. Control Signal Syst. (2024). https://doi.org/10.1007/s00498-024-00383-8
Black, F., Scholes, M.: The pricing of options and corporate liabilities. J. Polit. Econ. 81(3), 637–659 (1973)
North, G.R., Howard, L., Pollard, D., Wielicki, B.: Variational formulation of budyko-sellers climate models. J Atm Sci 36(2), 255–259 (1979)
Lissy, P., Moreno, C.: Rapid stabilization of a degenerate parabolic equation using a backstepping approach: The case of a boundary control acting at the degeneracy. Math. Control Relat. Fields (2023). https://doi.org/10.3934/mcrf.2023032
Abramowitz, M., Stegun, I.A.: Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables vol. 55, p. 1046. U. S. Government Printing Office, Washington. (1964)
Komornik, V., Loreti, P.: Fourier Series in Control Theory. Springer Monographs in Mathematics. Springer, New York (2005)
Koosis, P.: The Logarithmic Integral I & II. Cambridge Studies in Advanced Mathematics, vol. 12, p. 21. Cambridge University Press, Cambridge (1988, 1992)
Acknowledgements
The authors thank the referees for their 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
Leandro Galo-Mendoza and Marcos López-García have contributed equally to this work.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Bessel functions
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 (A1) we have (see [29, 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 formula ([29], 9.1.27):
Recall the asymptotic behavior of the Bessel function \(J_{\nu }\) for large x, see [30, Lem. 7.2, p. 129].
Lemma A.1
For any \(\nu \in {\mathbb {R}}\)
For \(\nu >0\) the Bessel function \(J_{\nu }\) has an infinite number of real zeros \((j_{\nu ,k})\) all of which are simple, with the possible exception of \(x=0\), see [29, 9.5.2, p. 370], satisfying
We recall Stirling’s formula, see [29, 6.1.39, p. 257],
In [29, 9.3.1, p. 365], we have
In [30, 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 \(|\nu |=\frac{1}{2}\).
For \(\nu \ge 0\) fixed, we consider the next asymptotic expansion of the zeros of the Bessel function \(J_{\nu }\), see [23, Sect. 15.53],
In particular, we have
Lemma A.3
For any \(\nu \ge 0\) and any \(k\ge 1\) we have
The proof of this result follows by using (A2) and the recurrence formula (A4).
Lemma A.4
Let \(\gamma =\gamma (\alpha ,\beta ,\mu )\) and \(\nu =\nu (\alpha ,\beta ,\mu )\) given in (3) and (4) respectively, then the following limit is finite for all \(k\ge 1\), and
Proof
This result follows from (A2). \(\square \)
We recall the following representation theorem, see [31, p. 56].
Theorem A.5
Let g(z) be an entire function of exponential type and assume that
Let \(\left\{ b_{\ell }\right\} _{\ell \ge 1}\) 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
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 drift, a singular potential, and a Neumann boundary condition. Bol. Soc. Mat. Mex. 30, 63 (2024). https://doi.org/10.1007/s40590-024-00638-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40590-024-00638-z
Keywords
- Degenerate parabolic equation
- Drift
- Singular potential
- Boundary controllability
- Neumann condition
- Moment method