Sampling theories of boundary value problems with several internal points of discontinuity

In this paper, our boundary value problem is a Dirac system with transmission conditions at several points of discontinuity. The main purpose of this paper is to derive the sampling theorems of this boundary value problem. To derive the sampling theorems including the construction of the Green’s matrix as well as the vector-valued eigenfunction expansion theorem, we briefly study the spectral analysis of the problem as in Levitan and Sargsjan (Introduction to Spectral Theory: Selfadjoint Ordinary Differential Operators, Translations of Mathematical Monographs, vol. 39, 1975; Sturm-Liouville and Dirac Operators, 1991) in a way similar to that of Fulton (Proc. R. Soc. Edinb., Sect. A 77:293-308, 1977). We derive sampling representations for transforms whose kernels are either solutions or the Green’s matrices of the problem. In the special case when our problem has one point of discontinuity, the obtained results coincide with the corresponding results in Tharwat et al. (Numer. Funct. Anal. Optim. 34:323-348, 2013).

Let \(\mathfrak{H}(D)\) be a class of complex-valued functions defined on D, where D is a subset of \(\mathbb{C}\), which may coincide with \(\mathbb{C}\). We say that a sampling theorem holds for the class \(\mathfrak{H}(D)\) if there are two sequences \(\{\lambda_{k}\} _{k=1}^{\infty}\subset D\) and \(\{S_{k}(\lambda)\}_{k=1}^{\infty}\subset\mathfrak{H}(D)\), such that

where series (1.1) satisfies some sort of convergence. The points \(\{\lambda_{k}\}_{k=1}^{\infty}\), which are for practical reasons preferred to be real, will be called the sampling points and the functions \(\{S_{k}(\lambda)\}_{k=1}^{\infty}\) are called the sampling (reconstructing) functions. The name sampling comes about since the elements of \(\{S_{k}(\lambda)\} _{k=1}^{\infty}\), when a sampling theorem holds, can be recovered from their values at \(\{\lambda_{k}\}_{k=1}^{\infty}\). When a sampling theorem holds for a class \(\{S_{k}(\lambda)\}_{k=1}^{\infty}\), this may have many applications when the discrete (digital) state is preferable, like the role played by the Whittaker-Shannon-Kotel’nikov (WSK) sampling theorem [5–9] in signal processing.

The point now concerns the classes \(\mathfrak{H}(D)\), for which sampling theorems are guaranteed. In most of the known cases in sampling theory the classes of the sampled functions are integral transforms defined on a \(L^{2}\)-space. Now, in the following, we give an example.

Example 1.1

Let \(\mathfrak{H}(\mathbb{C})\) be the space of all bandlimited functions, i.e., functions of the form of finite Fourier transforms,

Thus, \(\mathfrak{H}(\mathbb{C})\) is the Paley-Wiener space \(\mathit{PW}_{\pi }^{2}\) of entire functions of exponential type at most π, which belong to \(L^{2}(\mathbb{R})\) when restricted to \(\mathbb{R}\), cf. [5, 10, 11]. Elements of \(\mathfrak{H}(\mathbb{C})\), in this case, have sampling expansions of the form (1.1) with

where the sampling expansions converge absolutely on \(\mathbb{C}\) and uniformly on both \(\mathbb{R}\) and compact subsets of \(\mathbb{C}\), [12], p.262. This result is the well-known WSK sampling theorem [5–9] which has many applications in signal processing.

The above example is a special case of a generalized sampling theorem for generalized integral transforms. It is known as Kramer’s sampling theorem [5, 13–16] and reads as follows.

Theorem 1.1

Let \(\{\lambda_{k}\}_{k=-\infty}^{\infty}\) be a sequence of real numbers. Let \(I\subseteq\mathbb{R}\) and \(\mathcal{K}(\cdot,\lambda ):I\times\mathbb{C}\to\mathbb{C}\) be a function such that

1. 1.

   \(\mathcal{K}(\cdot,\lambda)\in L^{2}(I)\) for all \(\lambda\in \mathbb{C}\),

2. 2.

   the sequence \(\{\mathcal{K}(\cdot,\lambda_{k})\}_{k=-\infty }^{\infty}\) forms a complete orthogonal set in \(L^{2}(I)\).

Let \(\mathfrak{H}(\mathbb{C})\) be the set of integral transforms

Then elements of \(\mathfrak{H}(\mathbb{C})\) admit the sampling expansions

Series (1.4) converges uniformly wherever \(\Vert \mathcal{K}(\cdot, \lambda)\Vert_{L^{2}(I)}\) is bounded.

We can see that Example 1.1 above is a special case of Theorem 1.1. There are many problems motivated by Kramer’s theorem. The first concerns the situation for which we can obtain the kernel and the sampling points such that the theorem holds. The second concerns the relationship between Kramer’s sampling expansions and Lagrange-type interpolation ones. Notice that the sampling series in Example 1.1 is in the form of a Lagrange-type interpolation expansion. A partial answer to the first question was given by Weiss [17], before the existence of Kramer’s theorem. In this short note [17], the kernel is extracted from the second order boundary value problem

where \(q(\cdot)\) is continuous and real valued on \([a, b]\) and \(\alpha ,\beta\in[0,\pi)\). If we take \(u(\cdot,\lambda)\) to be the solution of (1.6) determined by the initial conditions

then \(u(\cdot,\lambda)\) satisfies Kramer’s theorem with respect to the sequence of the eigenvalues, i.e., the sequence of the eigenvalues, \(\{\lambda_{k}\} _{k=0}^{\infty}\), will be taken as the sampling points. For the second question, it is shown in many articles that Kramer’s expansions are nothing more than Lagrange-type interpolation ones when the kernels of the sampled integral transforms are solutions of certain self adjoint eigenvalue problems or are to be expressed in terms of the Green’s functions of these problems; see [18–20]. Using the connection between Kramer’s theorem and the eigenvalue problems most of the known sampling expansions can easily be obtained.

In [21, 22], the authors discussed sampling theorems of regular one-dimensional continuous Dirac systems; see also [23]. In the case of boundary value problems with one point of discontinuity, Tharwat et al. [4], discussed sampling theorems of Dirac systems; see also [24, 25]. In the present paper we derive sampling theorems associated with Dirac systems with transmission conditions at several points of discontinuity. That is, we define classes of transforms associated with these problems and derive sampling expansions for these classes. We use both solutions and the Green’s matrices to define the sampled transforms. In all cases the obtained sampling expansions are written in the form of a Lagrange interpolation series. The uniform convergence of the obtained sampling expansion as well as the analytic and growth properties of the sampled transforms are given. However, sampling theories associated with Dirac systems, which have several points of discontinuity, do not exist as far as we know. An illustrative example is given in detail in the last section.

In this section we introduce some of the notations and relations that will be used in the sequel; then we prove some useful lemmas and theorems. We consider the Dirac system

and the transmission conditions

where \(\lambda\in\mathbb{C}\);

\(r_{k}>0\) (\(k=1,2,\ldots,m+1\)) are real numbers; the real-valued functions \(q_{1}(\cdot)\) and \(q_{2}(\cdot)\) are continuous in \([a,c_{1})\), \((c_{k-1},c_{k})\), and \((c_{m},b]\) (\(k=2,3,\ldots,m\)), and have finite limits \(q_{1}(c_{k}^{\pm}):=\lim _{x\rightarrow c_{k}^{\pm}}q_{1}(x)\), \(q_{2} (c_{k}^{\pm}):=\lim_{x\rightarrow c_{k}^{\pm}}q_{2}(x)\) (\(k=1,2,\ldots,m\)); \(\gamma_{k}, \delta_{k}, \gamma_{k}^{\prime}, \delta _{k}^{\prime}\in \mathbb{R}\), \(\gamma_{k}, \delta_{k}, \gamma _{k}^{\prime}, \delta_{k}^{\prime}\neq0\) and \(\alpha,\beta\in [0,\pi)\); see [26, 27].

Let \(\mathcal{H}\) be the Hilbert space

The inner product of \(\mathcal{H}\) is defined by

where \(D_{i}=\frac{\delta_{i}\delta_{i}^{\prime}}{\gamma _{i}\gamma_{i}^{\prime}}\), \(D_{i}>0\), \(i=1,2,\ldots,m\), and ⊤ denotes the matrix transpose,

For vector-valued functions \(u(x)\), which are defined on \([a,c_{1})\cup (c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots\cup(c_{m},b]\) and have a finite limit \(u(c_{k}^{\pm}):=\lim_{x\rightarrow c_{k}^{\pm}}u(x)\) (\(k=1,2,\ldots,m\)), by \(u_{(k)}(x)\) (\(k=1,2,\ldots,m+1\)) we denote the functions

which are defined on \(\mathcal{I}_{1}:=[a,c_{1}]\), \(\mathcal {I}_{k}:=[c_{k-1},c_{k}]\) (\(k=2,3,\ldots,m\)) and \(\mathcal {I}_{m+1}:=[c_{m},b]\), respectively.

In the following lemma, we will prove that the eigenvalues of the problem (2.1)-(2.4) are real.

Lemma 2.1

The eigenvalues of the problem (2.1)-(2.4) are real.

Proof

Assume to the contrary that \(\lambda_{0}\) is a non-real eigenvalue of problem (2.1)-(2.4). Let \(\bigl ({\scriptsize\begin{matrix}{} u_{1}(x) \cr u_{2}(x)\end{matrix}} \bigr )\) be a corresponding (non-trivial) vector-valued eigenfunction. By (2.1), we have

Integrating the above equation through \([a, c_{1})\), \((c_{k-1}, c_{k})\) (\(k=2,3,\ldots,m\)), and \((c_{m}, b]\), we obtain

for \(k=2,3,\ldots,m\), and

Then from (2.2), (2.3), and the transmission conditions, we have, respectively,

and

Since \(\lambda_{0}\neq\overline{\lambda}_{0}\), it follows from the last three equations and (2.8), (2.9), (2.10) that

Then \(u_{i}(x)=0\), \(i=1,2\), and this is a contradiction. Consequently, \(\lambda_{0}\) must be real. □

Lemma 2.2

Let \(\lambda_{1}\) and \(\lambda_{2}\) be two different eigenvalues of the problem (2.1)-(2.4). Then the corresponding vector-valued eigenfunctions \(u(x,\lambda_{1})\) and \(v(x,\lambda_{2})\) of this problem satisfy the following equality:

Proof

By (2.1) we obtain

Integrating the above equation through \([a, c_{1})\), \((c_{k-1}, c_{k})\) (\(k=2,3,\ldots,m\)), and \((c_{m}, b]\), and taking into account that \(u(x,\lambda_{1})\) and \(v(x,\lambda_{2})\) satisfy (2.2)-(2.4), we obtain (2.12), where \(\lambda_{1}\neq\lambda_{2}\). □

Now we will construct a special fundamental system of solutions of (2.1) for λ not being an eigenvalue. By virtue of Theorem 1.1 in [1], we will define the two solutions of (2.1)

where

as follows, where \(c_{0}=a\) and \(c_{m+1}=b\). By virtue of Theorem 1.1 in [1] the initial-value problem

has a unique solution \(u=\bigl ({\scriptsize\begin{matrix}{} \varphi_{1,1}(x,\lambda)\cr \varphi_{2,1}(x,\lambda)\end{matrix}} \bigr )\), which is an entire function of \(\lambda\in\mathbb{C}\) for each fixed \(x\in[a,c_{1}]\). Similarly, employing the same method as in proof of Theorem 1.1 in [1], we see that the problem

has a unique solution \(u=\bigl ( {\scriptsize\begin{matrix}{} \chi_{1,m+1}(x,\lambda)\cr \chi_{2,m+1}(x,\lambda) \end{matrix}} \bigr )\), which is an entire function of the parameter λ for each fixed \(x\in[c_{m},b]\).

Now the functions \(\varphi_{i,k+1}(x,\lambda)\) and \(\chi _{i,k}(x,\lambda)\) are defined in terms of \(\varphi_{i,k}(x,\lambda)\) and \(\chi_{i,k+1}(x,\lambda)\), \(i=1,2\), \(k=1,2,\ldots,m\), respectively, as follows. The initial-value problem

has a unique solution \(u=\bigl ( {\scriptsize\begin{matrix}{} \varphi_{1,k+1}(x,\lambda)\cr \varphi_{2,k+1}(x,\lambda) \end{matrix}} \bigr )\) for each \(\lambda\in \mathbb{C}\).

Similarly, the following problem also has a unique solution \(u=\bigl ( {\scriptsize\begin{matrix}{} \chi_{1,k}(x,\lambda)\cr \chi_{2,k}(x,\lambda)\end{matrix}} \bigr )\):

By virtue of equations (2.20) and (2.22) these solutions satisfy both transmission conditions (2.4). These functions are entire in λ for all \(x\in[a,c_{1})\cup (c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots\cup(c_{m},b]\).

Let \(W(\varphi,\chi)({\cdot,\lambda})\) denote the Wronskian of \(\varphi(\cdot,\lambda)\) and \(\chi(\cdot,\lambda)\) defined in [2], p.194, i.e.,

It is obvious that the Wronskians

are independent of \(x\in\mathcal{I}_{k}\) and are entire functions. Taking into account (2.20) and (2.22), a short calculation gives

for each \(\lambda\in\mathbb{C}\).

Corollary 2.3

The zeros of the functions \(\omega_{k}(\lambda)\) (\(k=1,2,\ldots,m+1\)) coincide.

Then we may introduce into the consideration the characteristic function \(\omega(\lambda)\) as

Lemma 2.4

All eigenvalues of problem (2.1)-(2.4) are just zeros of the function \(\omega(\lambda)\).

Proof

Since the functions \(\varphi_{1}(x,\lambda)\) and \(\varphi _{2}(x,\lambda)\) satisfy the boundary condition (2.2) and the transmission conditions (2.4), to find the eigenvalues of the (2.1)-(2.4) we have to insert the functions \(\varphi _{1}(x,\lambda)\) and \(\varphi_{2}(x,\lambda)\) in the boundary condition (2.3) and find the roots of this equation. □

In the following lemma, we show that all eigenvalues of the problem (2.1)-(2.4) are simple.

Lemma 2.5

The eigenvalues of the boundary value problem (2.1)-(2.4) form an at most countable set without finite limit points. All eigenvalues of the boundary value problem (2.1)-(2.4) (of \(\omega(\lambda)\)) are simple.

Proof

The eigenvalues are the zeros of the entire function occurring on the left-hand side, see equation (2.24), in

We have shown (see Lemma 2.1) that this function does not vanish for non-real λ. In particular, it does not vanish identically. Therefore, its zeros form an at most countable set without finite limit points.

By (2.1) we obtain for \(\lambda,\mu\in\mathbb{C}\), \(\lambda\neq\mu\),

Integrating the above equation through \([a, c_{1})\), \((c_{k-1}, c_{k})\) (\(k=2,3,\ldots,m\)) and \((c_{m}, b]\), we obtain

for \(k=2,3,\ldots,m\), and

Then from (2.16) and the transmission conditions (2.20), we have

Dividing both sides of (2.28) by \((\lambda-\mu)\) and by letting \(\mu \rightarrow \lambda\), we arrive at the relation

We show that the equation

has only simple roots. Assume the converse, i.e., equation (2.30) has a double root \(\lambda^{\ast}\), say. Then the following two equations hold:

Equations (2.31) and (2.32) imply that

Combining (2.33) and (2.29), with \(\lambda=\lambda^{\ast}\), we obtain

It follows that \(\varphi_{1}(x,\lambda^{\ast})=\varphi_{2}(x,\lambda^{\ast})=0\), which is impossible. This proves the lemma. □

Here \(\{\varphi(\cdot,\lambda_{n})\}_{n=-\infty}^{\infty}\) will be a sequence of vector-valued eigenfunctions of (2.1)-(2.4) corresponding to the eigenvalues \(\{\lambda_{n}\}_{n=-\infty}^{\infty}\). Since \(\chi(\cdot,\lambda)\) satisfies (2.3) and (2.4), then the eigenvalues are also determined via

Therefore \(\{\chi(\cdot,\lambda_{n})\}_{n=-\infty}^{\infty}\) is another set of vector-valued eigenfunctions which is related by \(\{\varphi(\cdot,\lambda_{n})\}_{n=-\infty}^{\infty}\) with

where \(\tau_{n}\neq0\) are non-zero constants, since all eigenvalues are simple. Since the eigenvalues are all real, we can take the vector-valued eigenfunctions to be real valued.

In this section, we derive the asymptotic formulas of the eigenvalues \(\{\lambda_{n}\}_{n=-\infty}^{\infty}\) and the vector-valued eigenfunctions \(\{\varphi(\cdot,\lambda_{n})\}_{n=-\infty}^{\infty}\). We shall transform equations (2.1), (2.13), (2.16), and (2.20) into the integral equations, see [2],

where \(k=1,2,\ldots,m\).

For \(|\lambda|\rightarrow \infty\) the following estimates hold uniformly with respect to x, \(x\in[a,c_{1})\cup(c_{1},c_{2})\cup (c_{2},c_{3})\cup\cdots\cup(c_{m},b]\), cf. [1], p.55:

Now we will find an asymptotic formula for the eigenvalues. Let \(\gamma _{k}\delta_{k}^{\prime}-\gamma_{k}^{\prime}\delta_{k}=0\), \(k=1,2,\ldots,m\). Since the eigenvalues of the boundary value problem (2.1)-(2.4) coincide with the roots of the equation

then from the estimates (3.7) and (3.8) (with \(k=m\)) in (3.9) we get

which can be written as

Then, from (3.5), (3.6), (3.7), (3.8), and \(\gamma_{k}\delta_{k}^{\prime}-\gamma_{k}^{\prime }\delta_{k}=0\), equation (3.10) has the form

which can be rewritten as

For large \(|\lambda|\) equation (3.12) obviously has solutions which, as is not hard to see, have the form, cf. [1], p.57,

Inserting these values in (3.12), we find that \(\sin\delta_{n}=\mathcal{O} (\frac{1}{n} )\), i.e., \(\delta_{n}=\mathcal{O} (\frac{1}{n} )\). Thus we obtain the following asymptotic formula for the eigenvalues:

Using equations (3.14), we obtain the following asymptotic formulas for the vector-valued eigenfunctions \(\boldsymbol{\varphi}(\cdot,\lambda_{n} )\), where \(\gamma _{k}\delta_{k}^{\prime}-\gamma_{k}^{\prime}\delta_{k}=0\):

where \(k=1,2,\ldots,m\).

Let \(f(\cdot)=\bigl ( {\scriptsize\begin{matrix}{} f_{1}(\cdot)\cr f_{2}(\cdot) \end{matrix}} \bigr )\) be a continuous vector-valued function. Now, we derive the Green’s matrix of problem (2.1)-(2.4). Consider the inhomogeneous eigenvalue problem consisting of the differential system

and the boundary conditions (2.2)-(2.4) with λ is not an eigenvalue of problem (2.1)-(2.4).

Now, we can represent the general solution of (4.1) in the following form:

We applied the standard method of variation of the constants to (4.2), thus, the functions \(\mathcal{A}_{k}(x,\lambda)\) and \(\mathcal{B}_{k}(x,\lambda)\) (\(k=1,2,\ldots,m+1\)) satisfy the linear system of equations

and

Since λ is not an eigenvalue and \(\omega(\lambda)\neq0\), each of the linear system in (4.3), (4.4), and (4.5) has a unique solution which leads to

where \(\mathcal{A}_{k}\) and \(\mathcal{B}_{k}\), \(k=1,2,\ldots,m+1\), are arbitrary constants, and

Substituting equations (4.6), (4.7), and (4.8) into (4.2), we obtain the solution of (4.1),

Then from (2.2), (2.3), and (2.4), we get

Equations (4.10) and (4.11) can be rewritten as follows:

and

respectively. Then, from (4.12) and (4.13), (4.9) can be written as

which can be written as

where

Expanding (4.16) we obtain the concrete form

The matrix \(G(x,\xi,\lambda)\) is called the Green’s matrix of problem (2.1)-(2.4). Obviously \(G(x,\xi,\lambda)\) is a meromorphic function of λ, for every \((x,\xi)\in ([a,c_{1})\cup (c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots\cup(c_{m},b] )^{2}\), which has simple poles only at the eigenvalues. Although the Green’s function looks as simple as that of Dirac systems, cf. e.g. [1, 2], it is rather complicated because of the transmission conditions; see the example at the end of this paper.

The next theorem is a vector-valued eigenfunction expansion theorem. The proof is exactly similar to that of Levitan and Sargsjan derived in [1], pp.67-77; see also [2].

Theorem 4.1

1. (i)

   For \(u(\cdot)\in\mathcal{H}\),

   $$ \bigl\| u(\cdot)\bigr\| _{\mathcal{H}}^{2}=\sum _{n=-\infty}^{\infty}\bigl\vert \bigl\langle u(\cdot), \psi_{n}(\cdot)\bigr\rangle _{\mathcal{H}}\bigr\vert ^{2}. $$

   (4.18)
2. (ii)

   For \(u(\cdot)\in\mathcal{H}\),

   $$ u(x)=\sum_{n=-\infty}^{\infty}\bigl\langle u(\cdot),\psi_{n}(\cdot) \bigr\rangle _{\mathcal{H}} \psi_{n}(x), $$

   (4.19)

   where \(\psi_{n}(\cdot)=\frac{\varphi(\cdot,\lambda_{n})}{\|\varphi (\cdot,\lambda_{n})\|_{\mathcal{H}}}\) is the normalized vector-valued eigenfunction of the problem (2.1)-(2.4). The series is absolutely and uniformly convergent on \([a,c_{1})\cup(c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots\cup (c_{m},b]\).

In this section we introduce two sampling theorems associated with problem (2.1)-(2.4) above. In the first one we define an integral transform whose kernel is \(\varphi(\cdot,\lambda)\), while in the second one the sampled transform will be defined in terms of the Green’s matrix. The first sampling theorem of this section associated with the boundary value problem (2.1)-(2.4) is the following theorem.

Theorem 5.1

Let \(f(x)= \bigl ({\scriptsize\begin{matrix}{} f_{1}(x)\cr f_{2}(x)\end{matrix}}\bigr ) \in\mathcal{H}\). For \(\lambda\in\mathbb{C}\) let

where \(\varphi(\cdot,\lambda)\) is the solution defined in (2.13). Then \(\mathcal{F}(\lambda)\) is an entire function of exponential type that can be reconstructed from its values at the points \(\{\lambda_{n}\}_{n=-\infty}^{\infty}\) via the sampling formula

The series (5.2) converges absolutely on \(\mathbb{C}\) and uniformly on any compact subset of \(\mathbb{C}\). Here \(\omega(\lambda)\) is the entire function defined in (2.24).

Proof

Equation (5.1) can be rewritten in the form

Since both \(f(\cdot)\) and \(\varphi(\cdot,\lambda)\) are in \(\mathcal{H}\), they have the Fourier expansions

where \(\lambda\in\mathbb{C}\) and \(\hat{{f}}(n)\) are the Fourier coefficients

Applying Parseval’s identity to (5.3), we obtain

Now we calculate \(\langle \varphi(\cdot,\lambda),\varphi (\cdot,\lambda_{n})\rangle _{\mathcal{H}}\) and \(\|\varphi(\cdot,\lambda_{n})\|_{\mathcal{H}}\) of \(\lambda\in\mathbb{C}\), \(n\in\mathbb{Z}\). To prove expansion (5.2) we need to show that

Indeed, let \(\lambda\in\mathbb{C}\) and \(n\in\mathbb{Z}\) be fixed. By the definition of the inner product of \(\mathcal{H}\), we have

From the Green’s identity, see [1], p.51, we have

Then from the initial conditions (2.16) and (2.20), equation (5.9) becomes

From (2.36) and (2.18), we have

Substituting from (5.10), (5.11) into (5.8), we get

Letting \(\lambda\rightarrow\lambda_{n}\) in (5.12) and since the zeros of \(\omega(\lambda)\) are simple, we get

Since \(\lambda\in\mathbb{C}\) and \(n\in\mathbb{Z}\) are arbitrary, (5.12) and (5.13) hold for all \(\lambda\in \mathbb{C}\) and all \(n\in\mathbb{Z}\). Therefore from (5.12) and (5.13) we get (5.7). Hence (5.2) is proved with a pointwise convergence on \(\mathbb{C}\). Now we investigate the convergence of (5.2). First we prove that it is absolutely convergent on \(\mathbb{C}\). Using Cauchy-Schwarz’ inequality for \(\lambda\in\mathbb{C}\),

Since \(f(\cdot), \varphi(\cdot,\lambda)\in \mathcal{H}\), the two series in the right-hand side of (5.14) converge. Thus series (5.2) converges absolutely on \(\mathbb{C}\). As for uniform convergence, let \(M\subset\mathbb{C}\) be compact. Let \(\lambda\in M\) and \(N>0\). Define \(\nu_{N}(\lambda)\) to be

Using the same method as developed above

Therefore

Since \([a,b]\times M\) is compact, cf. e.g. [28], p.225, we can find a positive constant \(C_{M}\) such that

Then

uniformly on M. In view of Parseval’s equality,

Thus \(\nu_{N}(\lambda)\rightarrow0\) uniformly on M. Hence (5.2) converges uniformly on M. Thus \(\mathcal{F}(\lambda)\) is an entire function. From the relation

and the fact that \(\varphi_{1,k}(\cdot,\lambda)\) and \(\varphi _{2,k}(\cdot,\lambda)\), \(k=1,2,\ldots,m+1\), are entire functions of exponential type, see Lemma 7.2.1 in [2], we conclude that \(\mathcal{F}(\lambda)\) is of exponential type. □

Remark 5.2

If we replace \(\omega(\lambda)\) by the canonical product

then expansion (5.2) is a Lagrange-type interpolation. Indeed, from Hadamard’s factorization theorem, see [10], there is an entire function with no zeros, \(h(\lambda)\), such that

Thus,

and (5.1), (5.2) remain valid for the function \(\mathcal{F}(\lambda)/h(\lambda)\). Hence

We may redefine (5.1) by taking the kernel \(\frac{\varphi(\cdot,\lambda)}{h(\lambda)}=\widetilde{\varphi }(\cdot,\lambda)\) to get

For more details as regards the argument of Remark 5.2, see [9], p.110.

The next theorem is devoted to vector-type interpolation sampling expansions associated with problem (2.1)-(2.4) for vector-valued integral transforms whose kernels are defined in terms of the Green’s matrix. As we see in (4.16), the Green’s matrix \(G(x,\xi,\lambda)\) of problem (2.1)-(2.4) has simple poles at \(\{\lambda_{k}\}_{k=-\infty}^{\infty}\). Define the function \(G(x,\lambda)\) to be \(G(x,\lambda):=\omega(\lambda )G(x,\xi_{0},\lambda)\), where \(\xi_{0}\in[a,c_{1})\cup(c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots \cup(c_{m},b]\) is a fixed point and \(\omega(\lambda)\) is the function defined in (2.24) or it is the canonical product (5.20).

Theorem 5.3

Let \(f(x)=\bigl ( {\scriptsize\begin{matrix}{} f_{1}(x) \cr f_{2}(x)\end{matrix}} \bigr ) \in\mathcal{H}\). Let \(\frak{F}(\lambda)=\bigl ( {\scriptsize\begin{matrix}{} \mathfrak{F}_{1}(\lambda)\cr \mathfrak{F}_{2}(\lambda)\end{matrix}} \bigr )\) be the vector-valued transform

Then \(\frak{F}(\lambda)\) is a vector-valued entire function of exponential type that admits the vector-valued sampling expansion

The vector-valued series (5.24) converges absolutely on \(\mathbb{C}\) and uniformly on compact subsets of \(\mathbb{C}\). Here (5.24) means

where both series converge absolutely on \(\mathbb{C}\) and uniformly on compact sets of \(\mathbb{C}\).

Proof

Let \(\lambda\in\mathbb{C}\) such that \(\lambda\neq\lambda_{n}\) for \(n\in\mathbb{Z}\). Since \(\frak{F}(\lambda)\) is nothing but the unique solution (multiplied by \(\omega(\lambda)\)) of the inhomogeneous Dirac system (4.1), (2.2)-(2.4) when \(\xi=\xi_{0}\), see [1], p.77, \(\frak{F}(\lambda)\) has the expansion

and \(\{\psi_{n}(\cdot) \}_{n=-\infty}^{\infty}\) is a complete orthonormal set of vector-valued eigenfunctions. Thus

Now, we calculate \(\|\varphi(\cdot,\lambda_{n})\|^{2}_{\mathcal{H}}\) using the Green’s identity [1], p. 51. Let \(\lambda,\mu\in \mathbb{C}\) be different and \(\mu\neq\lambda_{n}\neq\lambda\); then, from (2.20) and (2.22),

where \(\varphi(x,\lambda)\) and \(\chi(x,\mu)\) are the solutions defined above. By the initial conditions (2.16) and (2.18), we have

Letting \(\mu=\lambda_{n}\), for some n, and noting that \(\chi(x,\lambda_{n})\) is a real-valued eigenfunction, from (2.16) and (2.36), equation (5.29) becomes

Then from (2.24), we obtain

Taking the limit in (5.31) when \(\lambda\rightarrow\lambda_{n}\),

Substituting in (5.27), we get the following series, which holds if λ is not an eigenvalue:

To prove (5.24), it suffices to prove that \(\frak{F}(\lambda_{n})=a_{n}\tau_{n}\varphi(\xi_{0}, \lambda_{n})\). Indeed, from (4.16) and (5.23), we have

where

and

Now we prove that \(\mathfrak{F}_{1}(\lambda_{n})=a_{n}\tau_{n}\varphi_{1}(\xi_{0}, \lambda_{n})\) and \(\mathfrak{F}_{2}(\lambda_{n})=a_{n}\tau_{n}\varphi_{2}(\xi_{0}, \lambda_{n})\). From (2.36), (5.35), and (5.36), we get

for \(x,\xi_{0}\in[a,c_{1})\cup(c_{1},c_{2})\cup(c_{2},c_{3})\cup\cdots \cup(c_{m},b]\). The same for \(\mathfrak{F}_{2}(\lambda_{n})\) and the proof of (5.24) is complete. The convergence properties as well as the analytic and growth properties can be established as in Theorem 5.1 above. □

Example 6.1

The boundary value problem

is a special case of problem (2.1)-(2.4) when \(m=2\), \(r_{1}=r_{2}=r_{3}=1\), \(\gamma_{1}=\gamma_{1}^{\prime}=\delta_{2}=\delta_{2}^{\prime }=1\), \(\gamma_{2}=\gamma_{2}^{\prime}=\delta_{1}=\delta _{1}^{\prime}=2\), \(\alpha=\beta=\frac{\pi}{2}\) and \(q_{1}(x)=q_{2}(x)=q(x)\),

Then \(\gamma_{k}\delta_{k}^{\prime}=\gamma_{k}^{\prime}\delta _{k}\), \(k=1,2\). In the notations of Section 2, the solutions \(\varphi(\cdot,\lambda)\) and \(\chi(\cdot,\lambda)\) are

The eigenvalues are the solutions of the equation

Hence the eigenvalues are

which is illustrated in Figure 1. The Green’s matrix of problem (6.1)-(6.4) is given by

where

By Theorem 5.1, the transform

has the following expansion:

where \(\omega' (\frac{1-2n\pi}{4} )=2(-1)^{n+1}\).

The eigenvalues of \(\pmb{\omega(\lambda)}\) .

Full size image

In view of Theorem 5.3, the vector-valued transform

where

The vector-valued transform (6.12) has the following vector-valued expansion:

The author is very grateful to the referees for carefully reading the paper and for their comments and suggestions, which have improved the paper.

Affiliations

1. Department of Mathematics, Faculty of Science, University of Jeddah, Jeddah, Saudi Arabia

   • Mohammed M Tharwat

2. Department of Mathematics, Faculty of Science, Beni-Suef University, Beni-Suef, Egypt

   • Mohammed M Tharwat

Corresponding author

Correspondence to Mohammed M Tharwat.

Competing interests

The author declares that he has no competing interests.

Reprints and Permissions