- Research
- Open access
- Published:
An inverse boundary value problem for transverse vibrations of a bar
Boundary Value Problems volume 2022, Article number: 96 (2022)
Abstract
In this article, we study an inverse problem (IP) for a fourth-order hyperbolic equation with nonlocal boundary conditions. This IP is reduced to the not self-adjoint boundary value problem (BVP) with corresponding boundary condition. Then, we use the separation of variables method, to reduce the not self-adjoint BVP to an integral equation. The existence and uniqueness of the integral equation are established by the contraction mappings principle and it is concluded that this solution is unique for a not-adjoint BVP. The existence and uniqueness of a nonlocal BVP with integral condition is proved. In addition, the fourth-order hyperbolic PDE is discretized using a collocation technique based on the quintic B-spline (QnB-spline) functions and reformed by the Tikhonov regularization function. The noise and analytical data are considered. The numerical outcome for a standard numerical example is discussed. Furthermore, the stability of the discretized system is also analyzed. The rate of convergence (ROC) of the method is also obtained.
1 Introduction
In modern technology, it is necessary to regulate vibration processes in one-dimensional distributed systems, and the relevance of these problems is increasing. In aircraft, such elements are formed simultaneously by bending and torsional vibrations. One of the objectives of the project is to prevent the use of shaft vibrations with an adjustable speed [4, 15]. For such problems, mathematical models of transverse vibrations of rods are built on the basis of a refined theory and such problems are called inverse problems of mathematical physics. Inverse problems for hyperbolic equations of fourth order receive great attention due to the necessity of the generalization for the classical problems [1]. Inverse problems for PDEs in numerous settings have been examined by various authors, e.g., Tikhonov [26], Lavrentiev [18], and Ivanov [16]. This type of problem has various applications such as in biology, medicine, mineral investigation, geophysics, computer tomography, filtration theory, etc. [8, 21, 25]. During computational modeling of specific processes, a condition may occur when the region’s boundary of the real process is challenging for measurements, but it is probable to obtain some further knowledge regarding the phenomena under investigation at the region’s interior points. From the mathematics viewpoint, this condition leads to a new nonlocal problem with integral conditions. In [1–3, 5, 7], the authors studied third-order PDE with integral conditions for unique solvability.
Very few investigations were found in the literature for the numerical computations of the time and/or spacewise coefficients for the IP of the fourth-order equations. For example, Huntul et al. [12, 13] reconstructed the potential coefficient in pseudoparabolic and pseudohyperbolic equations of order four, respectively, from additional measurements. The authors of [11, 20] identified the unknown potential coefficient in Boussinesq and Boussinesq-type equations of order four.
Recently, Huntul et al. [14, 22] obtained results about the numerical solutions of the inverse problem for a higher-order pseudoparabolic equation. The existence and uniqueness of the solution of an inverse boundary value problem for a third order in time pseudoparabolic equation were proved by using analytical and operatortheoretic means, the Fourier method, and the contraction principle. In [10], the authors numerically identify the time-dependent potential coefficient in a fourth-order pseudoparabolic equation with nonlocal initial and boundary conditions supplemented by nonlocal integral observations by applying the quintic B-spline collocation, finite-difference method and the Tikhonov regularization method.
In the domain \(D_{T}=\{(x,t): 0\leq x \leq 1, 0\leq t \leq T \}\), we consider an IP of the hyperbolic equation in fourth order
with ICs
the BCs
the nonlocal integral condition
and the additional conditions
where a, δ are given numbers, functions \(\varphi (x)\), \(\psi (x)\), \(h_{i}(t)\), \(i=1,2\), \(f(x,t)\), \(g(x,t)\) are given, while \(u(x,t)\), \(p(t)\), and \(q(t)\) are the desired functions.
The paper is divided into two parts. Part I discusses theory and proofs that contains Sects. 1, 2, and 3, while Part II investigates numerical experiments that contains Sects. 4, 5, 6, and 7. In Sect. 2, the IP reduces to an equivalent auxiliary IP. Section 3 proves the existence and uniqueness. In Sect. 4, the discretization of the forward problem is solved by using the QnB-spline collocation method. The stability has been analyzed in Sect. 5. Section 6 describes the numerical process of the nonlinear Tikhonov regularization functional. The outcomes for an example are discussed in Sect. 7. Finally, Sect. 8 reveals some concluding remarks.
2 Preliminary results and reduction of problem to an auxiliary IP
We propose the following definition and lemma:
Definition 2.1
We call the triplet \(\{u(x,t), p(t), q(t)\}\) the classic solution of IP, if the subsequent conditions are met:
-
1)
the functions \(u(x,t)\), \(u_{x}(x,t)\), \(u_{xx} (x,t)\), \(u_{xxx}(x,t)\), \(u_{xxxx}(x,t)\), \(u_{t}(x,t)\), \(u_{tt}(x,t)\) are continuous in \(D_{T}\);
-
2)
the functions \(p(t)\), \(q(t)\) are continuous on \([0,T]\);
- 3)
Alongside (1.1)–(1.5), consider the subsequent ODEs:
where δ is a given number, function \(p(t)\in C[0,T]\) is given, \(y = y(t)\) is a desired function, if \(y(t)\) is the solution of (2.1) and (2.2), then \(y(t)\) and all its derivatives are continuous in \([0,T]\).
It is not difficult to determine that the problem
has a trivial solution, if \(\delta \geq 0\). Then, it is known [24] that (2.3) has only one Green function \(G(t,\tau )\), given as
The subsequent Lemma is proved.
Lemma 2.2
Let \(a(t)\in C[0,T]\), and
further \(\delta \geq 0\) and
Then, BVP (2.1) and (2.2) has only a trivial solution.
Proof
It is evident from [24] that the BVP (2.1) and (2.2) is equivalent to the integral equation
Having denoted
we can write equation (2.5) as:
We will examine equation (2.7) in \(C[0,T]\). It can be easily seen that the operator A is continuous in \(C[0,T]\). Let us prove that A is a contraction mapping in \(C[0,T]\). Surely, for any \(y_{1}(t)\), \(y_{2}(t)\) from \(C[0,T]\)
Then, using (2.4) in (2.8), we find A is a contraction mapping in C[0,T]. Thus, in \(C[0,T]\), A has a single fixed point \(y=\{y_{1}, y_{2}\}\), which is a solution of (2.7). Thus, (2.6) has a unique solution in \(C[0,T]\), and so, the boundary value problem (2.1) and (2.2) also has a unique solution in \(C[0,T]\). As \(y(t) = 0\) the boundary value problem (2.1) and (2.2) has only a trivial solution. The lemma is proved. Along with (1.1)–(1.5), we choose the subsequent auxiliary IP. It is needed to find the triple \(\{ u(x,t), p(t), q(t)\}\) of \(u(x,t)\), \(p(t)\), and \(q(t)\) with properties 1) and 2) of the definition of the classical solution of BVP (1.1)–(1.5), from (1.1)–(1.3)
and
 □
The following theorem is valid.
Theorem 2.3
Let \(\varphi (x),\psi (x)\in C[0,1]\), \(h_{i}(t)\in C^{2}[0,T]\), \(i=1,2\), \(h(t)\equiv h_{1}(t)g (\frac{1}{2},t )-h_{2}(t)g(0,t)\neq 0\), \(t\in [0,T]\), \(f(x,t)\), \(g(x,t)\in C(D_{T})\), \(\int _{0}^{1}f(x,t)\,dx=0\), \(\int _{0}^{1}g(x,t)\,dx=0\), \(t\in [0,T]\), and the consistency conditions
be satisfied. Then, we have
-
1.
Every classical solution \(\{ u(x,t), p(t), q(t)\}\) of (1.1)–(1.5) is the solution of (1.1)–(1.3) and (2.9)–(2.11);
-
2.
Every solution \(\{ u(x,t), p(t), q(t)\}\) of (1.1)–(1.3) and (2.9)–(2.11) is a classical solution of (1.1)–(1.5), if
$$\begin{aligned}& \bigl\Vert \rho (t) \bigr\Vert _{C[0,T]} \biggl( \frac{1}{2}+\frac{1}{1+\delta T}+ \frac{\delta T}{2(1+\delta T)} \biggr)T^{2}< 1. \end{aligned}$$(2.16)
Proof
Let \(\{ u(x,t), p(t), q(t)\}\) be a solution of (1.1)–(1.5). Now, integrating (1.1) over x from 0 to 1, we obtain
Assuming that
in view of (1.4), we arrive at the fulfillment of (2.9).
Using \(x=0\) and \(x=\frac{1}{2}\) in (1.1), respectively, we obtain
Under the assumption \(h_{i}(t)\in C^{2}[0,T]\), \(i=1,2\) and differentiating (1.5) twice, we obtain
Considering these relations, from (2.18) and (2.19), taking into account (1.5), the fulfillment of (2.10) and (2.11) follows, respectively.
Now, let \(\{ u(x,t), p(t), q(t)\}\) be a solution to (1.1)–(1.3) and (2.9)–(2.11), and (2.16) is fulfilled. Then from (2.17), in view of (2.9), we obtain
From (1.2) and (2.13), we have
Since, by Theorem 2.3, problem (2.22) and (2.24) has only a trivial solution, \(\int _{0}^{1}u(x,t)\,dx=0\), \(t\in [0,T]\), i.e., conditions (1.4) are satisfied. Further, from (2.10), (2.18), (2.11), and (2.19), we obtain
From (1.2) and (2.15), we have
From (2.25), (2.30), and Lemma 2.2 the condition (1.5) is obtained. The theorem is proved. □
2.1 Auxiliary facts
It is understood that the sequence of the functions
form in \(L_{2}(0,1)\) a biorthogonal system and system (2.31) forms a basis in \(L_{2}(0,1)\), where \(\lambda _{k}=2k\pi \), \(k=1,2,\ldots \) , [17, 23]. They are also Riesz bases in \(L_{2}(0,1)\), see [24]. Then, any function \(g(x)\in L_{2}(0,1)\) can be expanded as a biorthogonal series
where the coefficients \(g_{0}\), \(g_{2k-1}\), \(g_{2k}\), are calculated by the following formulas
From (2.32), we have
and from above equation, we obtain
Under assumptions
and using integration by parts and taking into account (2.37) and (2.38), we obtain
Equation (2.36) implies
Then, from (2.39) and (2.41), we find
Thus, we have
Under the assumptions \(g(x)\in C^{2i}[0,1]\), \(g^{2i+1}(x)\in L_{2}(0,1)\), \(g^{2s}(1)=0\), \(g^{2s-1}(0)=g^{2s-1}(1) \), \(i\geq 1\), \(s=\overline{0,i}\), the following are valid
From the above equations, we find
Now, consider the subsequent spaces.
1. \(B_{2,T}^{5}\) [21] can be illustrated as consisting of all functions of the form
on \(D_{T}\), where \(u_{k}(T)\in C[0,T]\), \(k=0,1,\ldots \) , and
The norm is given by the formula
2. \(E_{T}^{5}\) can be illustrated as consisting of a vector by the formula
The norm of \(z=\{u,p,q\}\) is obtained
It is obvious that \(B_{2,T}^{5}\), \(E_{T}^{5}\) are Banach spaces.
3 Existence and uniqueness of a classical solution of the IP
Since the system (2.31) forms the Riesz basis in \(L_{2}(0,1)\), then each solution of (1.1)–(1.3) and (2.9)–(2.11) can be written as
where
where \(X_{k}(x)\) and \(Y_{k}(x)\) are described by (2.31) and (2.32), respectively.
Using the variable-separation method to find the desired functions \(u_{k}(t)\), \(k=0,1,\ldots \) , from (1.1), (1.2), we obtain
where
Solving (3.3)–(3.6), we obtain
where
After substituting expressions \(u_{0}(t)\), \(u_{2k-1}(t)\), \(u_{2k}(t)\), respectively, from (3.7), (3.8), and (3.9) into (3.1), to find \(u(x ,t)\) of the solution of (1.1)–(1.3) and (2.9)–(2.11), we obtain
Now, to obtain an equation for \(p(t)\), \(q(t)\) of the solution \(\{u(x,t), p(t),q(t)\}\) of (1.1)–(1.3) and (2.9)–(2.11), from (2.10) and (2.11), taking into account (3.1), we have
Assume that
Then, from (3.11) and (3.12), we obtain
Further, after substituting the expression \(u_{2k -1}(t)\) from (3.8) into (3.14) and (3.15), respectively, we have
Thus, the solution of (1.1)–(1.3) and (2.9)–(2.11) was reduced to the solution of (3.10), (3.16), and (3.17) with respect to the unknowns \(u(x,t)\), \(p(t)\), and \(q(t)\).
To study the question of the uniqueness of the solution of (1.1)–(1.3) and (2.9)–(2.11), the following plays an essential role.
Lemma 3.1
If \(\{u(x,t), p(t),q(t)\}\) is any solution of (1.1)–(1.3) and (2.9)–(2.11), then \(u_{k} (t)\), \(k = 0,1,\ldots \) , defined by (3.2), satisfy the system (3.7), (3.8), and (3.9) on \([0,T]\).
Proof
Let \(\{u(x,t), p(t),q(t)\}\) be any solution of (1.1)–(1.3) and (2.9)–(2.11). Then, multiplying in equation (1.1) by \(Y_{k}(x)\), \(k=0,1,2,\ldots \) , and then integrating from 0 to 1 using:
we obtain that (3.3)–(3.5) are satisfied.
Similarly, from (1.2), we find that condition (3.6) is satisfied. Thus, \(u_{k} (t)\), \(k = 0,1,2,\ldots\) , is a solution of (3.3)–(3.6). Hence, it immediately follows that \(u_{k}(t)\), \(k=0,1,2,\ldots \) , satisfy the system (3.7)–(3.9) on \([0,T]\). The lemma is proved. □
Obviously, if
is a solution to (3.7)–(3.9), then the triple \(\{u(x,t), p(t),q(t)\}\) of \(u(x,t)=\sum^{\infty}_{k=0}u_{k}(t)X_{k}(x)\), \(p(t)\) and \(q(t)\) is a solution to (3.7)–(3.9). From Lemma 2.2 it follows that.
Corollary 3.1
Let systems (3.10), (3.16), and (3.17) have a unique solution. Then (1.1)–(1.3) and (2.9)–(2.11) have at most one solution, i.e., if (1.1)–(1.3) and (2.9)–(2.11) has a solution, then it is unique.
Consider the operator
in \(E^{5}_{T}\), where
and \(\tilde{u}_{0}(t)\), \(\widetilde{u}_{2k}(t)\), \(\tilde{u}_{2k-1}(t)\), \(\tilde{p}(t)\), and \(\tilde{q}(t)\) are equal, respectively, to the right sides of (3.7)–(3.9), (3.16), and (3.17). Let \(0\leq \delta < 2\pi a\). It is not difficult to see that
Considering these relations, we have
Let the data of (1.1)–(1.3) and (2.9)–(2.11) fulfill the subsequent conditions:
- \((Q_{1})\):
-
\(\varphi (x)\in C^{5}[0,1]\), \(\varphi ^{6}(x)\in L_{2}(0,1)\), \(\varphi (1)=\varphi ^{\prime \prime}(1)=\varphi ^{(4)}(1)=0\), \(\varphi ^{\prime}(0)=\varphi ^{\prime}(1)\), \(\varphi ^{\prime \prime \prime}(0)=\varphi ^{\prime \prime \prime}(1)\), \(\varphi ^{(5)}(0)=\varphi ^{(5)}(1)\);
- \((Q_{2})\):
-
\(\psi (x)\in C^{3}[0,1]\), \(\varphi ^{4}(x)\in L_{2}(0,1)\), \(\psi (1)=\psi ^{\prime \prime}(1)=0\), \(\psi ^{\prime}(0)=\psi ^{\prime}(1)\), \(\psi ^{\prime \prime \prime}(0)=\psi ^{\prime \prime \prime}(1)\);
- \((Q_{3})\):
-
\(f(x,t),f_{x}(x,t),f_{xx}(x,t),f_{xxx}(x,t)\in C(D_{T})\), \(f_{xxxx}(x,t)\in L_{2}(D_{T})\), \(f(1,t)= =f_{xx}(1,t)=0\), \(f_{x}(0,t)=f_{x}(1,t)\), \(f_{xxx}(0,t)=f_{xxx}(1,t)\), \(t\in [0,T]\);
- \((Q_{4})\):
-
\(g(x,t),g_{x}(x,t),g_{xx}(x,t),g_{xxx}(x,t)\in C(D_{T})\), \(g_{xxxx}(x,t)\in L_{2}(D_{T})\), \(g(1,t)= =g_{xx}(1,t)=0\), \(g_{x}(0,t)=g_{x}(1,t)\), \(g_{xxx}(0,t)=g_{xxx}(1,t)\), \(t\in [0,T]\);
- \((Q_{5})\):
-
\(h_{i}\in C^{2}[0,T]\), \(i=1,2\), \(h(t)\equiv h_{1}(t) (\frac{1}{2},t )-h_{2}(t)g(0,t)\neq 0\), \(t\in [0,T]\), \(0\leq \delta \leq 2\pi \).
Then, from (3.19)–(3.23), taking into account (2.44)–(2.49), we obtain
where
From (3.24)–(3.28), we conclude that
where
Hence, we can prove the subsequent theorem:
Theorem 3.2
Let conditions \((Q_{1})\)–\((Q_{5})\) be fulfilled, and
Then, (1.1)–(1.3) and (2.9)–(2.11) has a unique solution in the sphere \(K=K_{R}(\|z\|_{E^{5}_{T}}\leq R=A(T)+2)\) of \(E^{5}_{T}\).
Proof
In \(E^{5}_{T}\), we consider
where \(z=\{u,p,q\}\) the components \(\Phi _{i}(u, p,q)\), \(i =1,2,3\), of the operator \(\Phi (u, p,q) \) are determined by the RHS of equations (3.10), (3.16), and (3.17). Consider \(\Phi (u, p,q)\) in \(K_{R} = K\) from \(E^{5}_{T}\). Similiar to (3.17), we obtain that for any \(z,z_{1},z_{2}\in K_{R}\) the following estimates are valid.
Then, using (3.31) and (3.33), it follows from (3.30) that Φ acts in \(K_{R} =K\) and it is a contraction mapping. Therefore, in \(K_{R} =K\), the operator has a unique fixed point \(\{u,p,q\}\) that is a solution of equation (3.31).
The \(u(x,t)\), as the element of \(B^{5}_{2,T}\), has continuous derivatives \(u_{x} (x,t)\), \(u_{xx}(x,t)\), \(u_{xxx}(x,t)\), \(u_{xxxx}(x,t)\) in \(D_{T} \). Now, from (3.33)–(3.35), we obtain
Hence, it follows that \(u_{tt} (x,t)\) is continuous in \(D_{T}\).
It is easy to validate that (1.1)–(1.3) and (2.9)–(2.11) are fulfilled in the ordinary sense. Therefore, \(\{u(x,t), p(t),q(t)\}\) is a solution of (1.1)–(1.3) and (2.9)–(2.11), and by Lemma 3.1, it is unique in the ball \(K_{R} =K\). The theorem is proved. □
The subsequent theorem is proved by Lemma 2.2.
Theorem 3.3
Let all the conditions of Theorem 3.2be fulfilled:
Then, in \(K=K_{R}(\|z\|_{E^{5}_{T}}\leq R=A(T)+2)\) of \(E^{5}_{T}\), (1.1)–(1.5) has a unique classical solution.
4 Discretization of the direct problem
We consider the IBVP (1.1)–(1.4), when a, \(q(t)\), \(p(t)\), \(g(x,t)\), and \(f(x,t)\) are given. First, we divide \([0,l]\) into a mesh of equal size \(h=x_{i+1}-x_{i}\), \(i=0,1,\ldots,M\). The discrete form of the direct problem is as follows. We denote \(u(x_{i},t_{j})=u_{i}^{j}\), \(p(t_{j})=p^{j}\), \(q(t_{j})=q^{j}\), \(g(x_{i},t_{j})=g_{i}^{j}\) and \(f(x_{i},t_{j})=f_{i}^{j}\), where \(x_{i}=ih\), \(t_{j}=jk\), \(h=\Delta x=\frac{l}{M}\) and \(k=\Delta t=\frac{T}{N}\) for \(i=0,1,\ldots,M\) and \(j=0,1,\ldots,N\). The QnB-spline \(QB_{i}(x)\), \(i=-2,-1,\ldots,M+1,M+2\) are given by [6, 9]:
The values of \(QB_{i}(x)\), \(QB^{\prime}_{i}(x)\), \(QB^{\prime\prime}_{i}(x)\), \(QB^{\prime\prime\prime}_{i}(x)\), and \(QB^{iv}_{i}(x)\) are given in Table 1, where \(\xi _{1}=\frac{5}{h}\), \(\xi _{2}=\frac{20}{h^{2}}\), \(\xi _{3}=\frac{60}{h^{3}}\), and \(\xi _{4}=\frac{120}{h^{4}}\).
We suppose that \(u(x,t)\) at the point \((x, t_{j})\) is expressed as:
The variation of the \(U_{M}(x,t)\) is expressed as
Using (4.3), we obtain u, \(u_{x}\), \(u_{xx}\), \(u_{xxx}\), \(u_{xxxx}\) as:
where \(C_{i}^{j}=C_{i}(t_{j})\). Now, we discretize equation (1.1) as:
which implies
where
Now, using (4.4)–(4.8) and simplifying the terms, we obtain
The above equation can be written as
where
Now, we discretize the initial conditions (1.2) as
For \(j=0\), using the IC (4.14) in (4.10), we obtain
where
Using the approximated values of u and \(u_{xxxx}\) in (4.15) and simplifying the terms, we obtain
which can be written as
where
The system (4.17) has unknowns \((C_{-2}^{0},C_{-1}^{0}, C_{0}^{0},\ldots, C_{M+1}^{0}, C_{M+2}^{0})\), where \(C_{-2}^{0}\), \(C_{-1}^{0}\), \(C_{M+1}^{0}\), and \(C_{M+2}^{0}\) are outside the domain. For a unique solution, we need to remove these quantities. For this purpose, we use the BCs \(u(1,t)=0\), \(u_{x}(0,t)=u_{x}(1,t)\), \(u_{xx}(1,t)=0\), and \(\int _{0}^{1} u(x,t)\,dx=0\), which give us following equations:
and
where, \(j=0,1,\dots ,N\).
Solving the above equations, we obtain
For \(i=0\), using (4.22) and (4.23) in (4.17), we obtain
where
For \(i=1\), using (4.22) in (4.17), we obtain
where
For \(i=M-1\), using (4.24) in (4.17), we obtain
For \(i=M\), using (4.25) in (4.17), we obtain
At time \(t=0\), the equations (4.26), (4.27), (4.17), (4.28), and (4.29) form a system:
where
Finally, for \(i=0\), using (4.22) and (4.23) in (4.12), we obtain
where
For \(i=1\), using (4.22) in (4.12), we obtain
where
For \(i=M-1\), using (4.24) in (4.11), we obtain
For \(i=M\), using (4.25) in (4.12), we obtain
At the time \(t_{j}\), \(j=1,2,\dots , N\), the equations (4.31), (4.32), (4.12), (4.33), and (4.34) form a system:
where
In order to solve the systems (4.30) and (4.31), we need to determine the initial vector \((C_{0}^{0}, C_{1}^{0}, \dots , C_{M}^{0})\) from the ICs. To remove the \(C_{-2}^{0}\), \(C_{-1}^{0}\), \(C_{M+1}^{0}\), and \(C_{M+2}^{0}\), we use
Using (4.5) and (4.6) in (4.35) and (4.36) and eliminating the unknowns \(C_{-2}^{0}\), \(C_{-1}^{0}\), \(C_{M+1}^{0}\), and \(C_{M+2}^{0}\), we obtain the \((M+1) \times (M+1)\) system:
5 Stability analysis
The von Neumann stability [13, 27] is analyzed in this section. For stability, we choose \(f(x,t)=0\), \(g(x,t)=0\) and assume \(p(t)=\bar{p}\) is a local constant. We discretize the problem as follows:
where
Using equations (4.5) and (4.6) in the above equation, we obtain
where
Now, we consider the trial solution \(C_{i}^{j}=\delta ^{j} e^{ki \Theta}\) at a given point \(x_{i}\), where \(\Theta =\theta h\), where \(k=\sqrt{-1}\). Substituting \(C_{i}^{j}\) in (5.2), we obtain
which can be written as
where
Now, employing the Routh–Hurwitz criterion under \(\delta =\frac{1+\varsigma}{1-\varsigma}\) in the above equation, we obtain
The necessary and sufficient conditions for \(|\delta | \le 1\) are
Substituting the values of \(\Lambda _{1}\), \(\Lambda _{2}\), and \(\Lambda _{3}\) into equation (5.6), we have
It is obvious from equations (5.7)–(5.9) that \(\Lambda _{1} + \Lambda _{2}+\Lambda _{3} \ge 0\), \(\Lambda _{1}- \Lambda _{3} \ge 0\), and \(\Lambda _{1}-\Lambda _{2}+\Lambda _{3} \ge 0\). Hence, the technique is unconditionally stable for the discretized problem.
6 Numerical algorithm for IP
We intend to obtain stable and accurate solutions of \(p(t)\), \(q(t)\) and \(u(x,t)\) that assure (1.1)–(1.5). The considered problem is solved approximately by minimizing the subsequent regularized cost function
where u fulfills (1.1)–(1.4) with known \(p(t)\), \(q(t)\), and \(\gamma >0\) is a regularization parameter initiated for stabilizing the approximate solutions. The discretized form of (6.1) is
Equation (6.2) is minimized by the MATLAB lsqnonlin tool [19].
7 Results and discussion
An example is considered in this section to examine the accuracy and stability. To validate the efficiency, we use the RMSE as:
We take \(T=1\). The lower and upper bounds for \(p(t)\) and \(q(t)\) are considered to be −102 and 102, respectively.
The BVP (1.1)–(1.5) is solved for both exact and perturbed data. The perturbed data is managed as
where \(\epsilon 1_{j}\) and \(\epsilon 2_{j}\) indicate the r.v.s and the subsequent S.D.s
For the perturbed data (7.3), (7.4), \(h_{1}(t_{j})\), and \(h_{2}(t_{j})\) are replaced by \(h_{1}^{\epsilon 1}(t_{j})\) and \(h_{2}^{\epsilon 2}(t_{j})\) in (6.2).
Let us examine the BVP (1.1)–(1.5) with unknowns \(p(t)\) and \(q(t)\), from
the ICs
the BCs
with
and
We consider the exact \(p(t)\), \(q(t)\) and \(u(x,t)\) as:
Theorem 3.3 is fulfilled, which indicates that a unique solution is assured. First, when \(p(t)\) and \(q(t)\) are given by (7.11) and (7.12), the accuracy of (1.1)–(1.5) is validated using (7.7), (7.8), and (7.10). Figure 1 shows the exact (7.13) and numerical \(u(x,t)\), as well as absolute errors, for \(\Delta x=\Delta t\in \{\frac{1}{20},\frac{1}{40},\frac{1}{80} \}\). The approximate additional measurements \(h_{1}(t)\) and \(h_{2}(t)\) in (1.5) are compared to the exact solution (7.9) derived using the QnB-spline method with \(M=N \in \{20,40,80\}\) in Fig. 2. The rate of convergence of the method is checked with \(\Delta t=0.025\), which shows that the method is second-order convergent in space, see Table 2.
In the IP (1.1)–(1.5), the initial guesses for \(\underline{p}\) and \(\underline{q}\) are taken as:
When \(p\%=0\) in (7.5), we use \(\Delta x=\Delta t=\frac{1}{40}\) to start analyzing to recover \(p(t)\), \(q(t)\) and \(u(x,t)\). The \(\mathbb{F}\) in (6.2) is shown in Fig. 3(a), where a monotonic decrease in convergence is realized in 7 itrs for a given tolerance of \(O(10^{-25})\). Figures 3(b) and 3(c) depict the exact ((7.11) and (7.12)) and approximate \(p(t)\), \(q(t)\) without regularization. An accepted and stable accurate \(p(t)\) and \(q(t)\), producing \(\operatorname{RMSE}(p)=1.0827\text{E}{-}3\) and \(\operatorname{RMSE}(q)=6.6000\text{E}{-}3\) can be seen.
Now, as in equation (7.5), we add \(p\%\in \{0.1\%,1\%\}\) to \(h_{1}(t)\) and \(h_{2}(t)\) through (1.5). In Figs. 4 and 5, \(p(t)\) and \(q(t)\) are depicted. As p% increases, the solutions begin oscillations with \(\operatorname{RMSE}(p)\in \{9.1265,59.5488 \}\), as seen in Figs. 4(a) and 5(a), and \(\operatorname{RMSE}(q)\in \{9.0045,56.7323 \}\), as seen in Figs. 4(c) and 5(c). Figures 4(b), 4(d), 5(b), and 5(d) show the recovered \(p(t)\) and \(q(t)\) for various γ, and the most accurate solution is attained for \(\gamma \in \{10^{-7},10^{-6}\}\), producing \(\operatorname{RMSE}(p)\in \{0.0692,0.0519 \}\), and \(\operatorname{RMSE}(q)\in \{0.1403,0.1051 \}\) for \(p\%=0.1\%\), and for \(\gamma \in \{10^{-5},10^{-4}\}\), producing \(\operatorname{RMSE}(p)\in \{0.1297,0.0865 \}\), and \(\operatorname{RMSE}(q)\in \{0.2633,0.1754 \}\) for \(p\%=1\%\), see Table 3. The abs. errors between the exact (7.13) and approximate \(u(x,t)\) are shown in Fig. 6, where the impact of \(\gamma >0\) in minimizing the unstable behavior of the recovered u can be observed.
8 Conclusions
This article discusses the existence and uniqueness of an IP for a fourth-order PDE with nonlocal integral conditions. The spectral analysis technique is used to reduce the problem to an operator equation in a certain Banach space. Then, the principle of contraction maps is used to prove the existence and uniqueness. This work is novel and has never been investigated theoretically and/or numerically before. A collocation method based on QnB-splines is applied for the direct problem. The stability analysis is also discussed for the discretized system. The MATLAB subroutine lsqnonlin is used to solve the resulting nonlinear optimization problem. To deal with stability and accuracy, Tikhonov regularization is employed. The numerical analysis revealed that accurate solutions are attained for \(\gamma \in \{10^{-7}, 10^{-6}\}\) when \(p\%=0.1\%\) and for \(\gamma \in \{10^{-5}, 10^{-4}\}\) when \(p\%=1\%\).
Availability of data and materials
Not applicable.
References
Abdrakhmanov, A.M.: Solvability of a boundary-value problem with an integral boundary condition of the second kind for equations of odd order. Math. Notes 88, 151–159 (2010)
Abdrakhmanov, A.M., Kozhanov, A.I.: A problem with a nonlocal boundary condition for one class of odd-order equations. Russ. Math. 51, 1–10 (2007)
Bouziani, A., Benouar, N.E.: Mixed problem with integral conditions for a third order parabolic equation. Kobe J. Math. 15, 47–78 (1998)
Cannon, J.R.: The solution of the heat equation subject to the specification of energy. Q. Appl. Math. 5, 155–160 (1963)
Denche, M., Memou, A.: Boundary value problem with integral conditions for a linear third-order equation. J. Appl. Math. 2003, 553–567 (2003)
Dhiman, N., Tamsir, M.: Re-modified quintic B-spline collocation method for the solution of Kuramoto-Sivashinsky type equations. In: Multidiscipline Modeling in Materials and Structures (2018). https://doi.org/10.1108/MMMS-06-2018-0111
Eskin, G.: Inverse problems for general second order hyperbolic equations with time-dependent coefficients. Bull. Math. Sci. 7, 247–307 (2017)
Hazanee, A., Lesnic, D., Ismailov, M.I., Kerimov, N.B.: An inverse timedependent source problem for the heat equation with a non-classical boundary condition. Appl. Math. Model. 39, 6258–6272 (2015)
Huntul, M.J.: Recovering a source term in the higher-order pseudo-parabolic equation via cubic spline functions. Phys. Scr. 97, 035004 (2022)
Huntul, M.J., Abbas, M.: An inverse problem of fourth-order partial differential equation with nonlocal integral condition. Adv. Cont. Discr. Mod. 2022, 1–27 (2022)
Huntul, M.J., Tamsir, M.: Identifying an unknown potential term in the fourth-order Boussinesq–Love equation from mass measurement. Eng. Comput. 38, 3944–3968 (2021)
Huntul, M.J., Tamsir, M., Dhiman, N.: An inverse problem of identifying the time-dependent potential in a fourth-order pseudoparabolic equation from additional condition, Numer. Methods Partial Differ. Equ. (2021). https://doi.org/10.1002/num.22778
Huntul, M.J., Tamsir, M., Dhiman, N.: Identification of time-dependent potential in a fourth-order pseudo-hyperbolic equation from additional measurement. Math. Methods Appl. Sci. 45, 5249–5266 (2022)
Huntul, M.J., Tekin, I.: On an inverse problem for a nonlinear third order in time partial differential equation. Results Appl. Math. 15, 100314 (2022)
Ionkin, N.I.: The solution of a boundary value problem of the theory of heat conduction with a nonclassical boundary condition. Differ. Equ. 13, 294–304 (1977)
Ivanov, V.K., Vasin, V.V., Tanina, V.P.: The Theory of Linear Ill-Posed Problems and Its Applications p. 206. Nauka Publishing House, Moscow (1978)
Kaliev, I.A., Sabitova, M.M.: Problems of determining the temperature and density of heat sources by initial and final temperatures. J. Appl. Ind. Math. 12, 332–339 (2010)
Lavrentyev, M.M.: On an inverse problem for the wave equation. Dokl. Akad. Nauk SSSR Reports of the Academy of Sciences of the USSR 157, 520–521 (1964)
Mathworks: Documentation Optimization Toolbox-Least Squares (Model Fitting) Algorithms (2019). Available at www.mathworks.com
Megraliev, Y.T., Alizade, F.K.: Inverse boundary value problem for a Boussinesq type equation of fourth order with nonlocal time integral conditions of the second kind. Vestnik Udmurtskogo Universiteta. Matematika. Mekhanika. Komp’yuternye Nauki 26, 503–514 (2016)
Mehraliev, Y.T.: On a certain inverse boundary value problem for an elliptic equation of second order with an integral condition of the first kind. Trudy Instituta Matematiki i Mekhaniki UrO RAN 19, 226–235 (2013)
Mehraliyev, Y.T., Ramazanova, A.T., Huntul, M.J.: An inverse boundary value problem for a two-dimensional pseudo-parabolic equation of third order. Results Appl. Math. 14, 100274 (2022)
Moiseev, E.I.: On the solution of a single nonlocal boundary value problem by the spectral method. Differ. Equ. 35, 1094–1100 (1999)
Naimark, M.A.: Linear Differential Operators. Nauka, Moscow (1969)
Ramazanova, A.T.: On determining initial conditions of equations flexural torsional vibrations of a bar. Eur. J. Pure Appl. Math. 12, 25–38 (2019)
Tikhonov, A.N.: On the stability of inverse problems. Dokl. Akad. Nauk SSSR Reports of the Academy of Sciences of the USSR 39, 195–198 (1943)
Vichnevetsky, R.: Stability charts in the numerical approximation of partial differential equations: a review. Math. Comput. Simul. 21, 170–177 (1979)
Acknowledgements
The authors are grateful to the anonymous reviewers for their helpful, valuable comments and suggestions for the improvement of this manuscript.
Funding
Not applicable.
Author information
Authors and Affiliations
Contributions
All authors participated in the preparation of all stages of the article and have the same contribution and read the final file and agree to publish it
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
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
Mehraliyev, Y.T., Huntul, M.J., Ramazanova, A.T. et al. An inverse boundary value problem for transverse vibrations of a bar. Bound Value Probl 2022, 96 (2022). https://doi.org/10.1186/s13661-022-01679-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s13661-022-01679-x