Abstract
This work provides the Legendre spectral projection (Galerkin and collocation), iterated Legendre spectral projection, Legendre spectral multi-projection and iterated Legendre spectral multi-projection methods to approximate the solution of weakly singular Hammerstein integral equations of mixed type. The convergence rates of approximate solutions to the exact solutions are obtained for all the above four methods in both \(L^{2}\) and infinity norm. The comparison of convergence rates for all these methods have been discussed. We also have shown that iterated Galerkin improves over Galerkin, multi-Galerkin improves over iterated Galerkin and iterated multi-Galerkin improves over multi-Galerkin in \(L^2\) norm using Legendre polynomial bases.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
We consider the following weakly singular Hammerstein integral equation of mixed type
where the source function f, the kernels \(k_i(.,.)\) and the nonlinear functions \(\psi _i(.,.)\) for \(i=1,2,\ldots ,m\) are known and u is the unknown function to be determined in a Banach space \(\mathbb {X}\). We consider the kernel \(k_i(., .)\) as weakly singular type which is of the form
\(m_{i}(s, t)\in \mathcal{C}([-1, 1]\times [-1, 1])\) and
As a reformulation of boundary value problem, this type of problem (1) arises in nonlinear physical phenomenon such as electromagnetic fluid dynamics [3].
Several authors have used numerical methods such as projection methods (Galerkin and collocation), Petrov-Galerkin method, degenerate kernel method and Nyström method ([6,7,8, 11, 12, 15]) to solve the various linear and nonlinear integral equations because these integral equations can’t solve explicitly. Integral equations of type (1) with smooth and weakly singular kernel were solved numerically in ([2, 5, 9, 10, 13, 14, 18]) using piecewise polynomials as bases. In piecewise polynomial based projection methods the number of partitions should be increased to obtain more accurate approximate solution. So, one has to solve a large system of nonlinear equations, which take lots of time to compute. Therefore, many spectral methods have been developed by using global polynomials in last some years. In the global polynomial based projection methods, if \( \mathcal {P}_{n} \) denotes either orthogonal or interpolatory projection operator, then \( \Vert \mathcal {P}_{n}\Vert _{\infty } \) is unbounded.
We are interested to solve numerically the Hammerstein integral equations of mixed type with weakly singular kernel using Legendre spectral projection, iterated Legendre spectral projection methods. To improve convergence rates further, Legendre multi-projection and iterated Legendre multi-projection method have been used. We evaluate the convergence rates in all the above four methods in both \(L^2\) and infinity norm, even if \(\Vert \mathcal {P}_{n}\Vert _{\infty } \) is unbounded. We have given a comparison of error bounds in all the methods.
We have organized this paper in the following way. We have discussed the abstract framework for the Legendre spectral projection methods for Hammerstein integral equations of mixed type with the weakly singular kernels in Sect. 2. The convergence rates of approximated solution with exact solution have been discussed using spectral projection, iterated spectral projection, spectral multi-projection and iterated spectral multi-projection methods in Sects. 3, 4, 5 and 6, respectively, in both \(L^2\) and infinity norm using Legendre polynomial bases. However, in the end, we have added a remark through which, we have given the comparison of error bounds in all the methods.
Throughout this paper, we assume c is a generic constant which may differ and is independent of n.
2 Hammerstein integral equation of mixed type with weakly singular kernel
In this section, we set up an abstract framework for the Hammerstein integral equation of type (1) with weakly singular kernel of type (2) on the Banach space \(\mathbb {X}=\mathcal {C}[-1,1]\) or \(L^2[-1,1]\). Throughout the paper, the following assumptions are made on \( f,~k_{i}(.,.) \) and \( \psi _{i}(.,u(.)) \):
- (i)
\( f \in \mathcal {C}[-1,1], \)
- (ii)
\( s_{i}= \displaystyle \sup _{s,t \in [-1,1]} |m_{i}(s,t)| < \infty \) for \(i=1,2,\ldots ,m\) and \( M=\sum _{i=1}^{m}s_{i}, \)
- (iii)
\( M_{2}=\displaystyle \sup _{s \in [-1,1]}\int _{-1}^{1} \left| g_{\alpha }|s-t|\right| ^{2}\mathrm{d}t < \infty , \) for \( \frac{1}{2}<\alpha \le 1.\)
- (iv)
The nonlinear functions \( \psi _{i}(t,u) \) are continuous on \( [-1,1]\times \mathbb {R}\) for \(i=1,2,\ldots ,m\). \(\psi _{i}(t,u) \) are Lipschitz continuous in u, i.e., for any \( u_{1},~u_{2} \in \mathbb {R},~\exists \) constants \( c_{i}>0,~i=1,2,\dots ,m \) such that
$$\begin{aligned} \left| \psi _{i}(t,u_{1})-\psi _{i}(t,u_{2}) \right| \le c_{i} |u_{1}-u_{2}|,~\forall ~t\in [-1,1], \end{aligned}$$and \( l_{1}=\sup _{i=1,2,\dots ,m}c_{i}. \)
- (v)
The functions \(\psi _{i}^{(0,1)}(t,u(t))\) exist and are Lipschitz continuous in u, i.e., for any \( u_{1},~u_{2} \in \mathbb {R}, ~ \exists \) constants \( q_{i}>0,~i=1,2,\dots ,m \) such that
$$\begin{aligned} \left| \psi _{i}^{(0,1)}(t,u_{1})-\psi _{i}^{(0,1)}(t,u_{2}) \right| \le q_{i} |u_{1}-u_{2} |,~\forall ~t\in [-1,1], \end{aligned}$$and \( l_{2}=\sup _{i=1,2,\dots ,m}q_{i}\). This implies \( \psi _{i}^{(0,1)}(.,.) \in \mathcal {C}([-1,1]\times \mathbb {R}). \)
Define the integral operator \( \mathcal {K}_{i}:\mathbb {X}\rightarrow \mathbb {X},~i=1,2,\dots ,m \) by
Then the equation (1) can be written in the following operator equation
Next, we define the operator \( \mathcal {T} \) on \( \mathbb {X} \) by
then the equation (3) can be written as
Define the Fréchet derivatives of \( \sum _{i=1}^{m}\mathcal {K}_{i}\psi _{i}(u) \) at \(u_0\) by
Lemma 2.1
([16]) Let\(m_i(s,t)\in \mathcal {C}([-1,1]\times [-1,1]) \)and\(g_{\alpha }|s-t| \)be the weakly singular part of the kernel\(k_i(s,t)\)for\(1\le i \le m\). Then for any\(s_1,s_2\in [-1,1]\), we have the followings:
- (i)
\(\displaystyle \lim _{s_1\rightarrow s_2}\int _{-1}^{1}|m_i(s_1,t)-m_i(s_2,t)|^2\mathrm{d}t\rightarrow 0\), for \(1\le i\le m\),
- (ii)
\(\displaystyle \lim _{s_1\rightarrow s_2}\int _{-1}^{1}|g_{\alpha }|s_1-t|-g_{\alpha }|s_2-t||^2\mathrm{d}t\rightarrow 0,\) for \(\frac{1}{2}<\alpha \le 1\).
The next theorem shows the existence and uniqueness of the solution of equation (5).
Theorem 2.2
Let\( \mathbb {X}=\mathcal {C}[-1,1] \), \( f\in \mathbb {X} \)and\( g_{\alpha }|s-t| \)satisfies the assumption (iii) with\( m_{i}(.,.) \in \mathcal {C}([-1,1]\times [-1,1]) \)and\( s_{i}=\sup _{s,t\in [-1,1]}|m_{i}(s,t)| < \infty . \)Let\( \psi _{i}(t,u(t)) \in \mathcal {C}([-1,1]\times \mathbb {R}) \)satisfy the assumption (iv) with\( \sqrt{2M_{2}}Ml_{1} < 1, \)where\( M=\sum _{i=1}^{m}s_{i} \)and\( l_{1}=\sup _{i=1,2,\dots ,m}c_{i}. \)Then the operator equation\( \mathcal {T}u=u \)has a unique isolated solution\( u_{0}\in \mathbb {X} \), i.e.,\( \mathcal {T}u_{0}=u_{0}. \)
Proof
The proof follows exactly by using similar technique given in Theorem-2.4 of [12]. \(\square \)
We will first approximate the space \( \mathbb {X}\) by a finite dimensional space \(\mathbb {X}_{n}.\) We consider \( \mathbb {X}_{n}=\mathrm{span}\left\{ \phi _{0},\phi _{1},\dots , \phi _{n} \right\} \) as the sequence of Legendre polynomial subspaces of \( \mathbb {X} \) of degree \(\le n\). Define \( L_{i}(s)=\sqrt{\dfrac{2i+1}{2}}\phi _{i}(s),\,\,i=0,1,\dots ,n. \) Since \( L_{i} \) and \( L_{j} \)’s are polynomials
for \(i,j=0,1,\dots ,n. \)
Then the Legendre polynomials \( \left\{ L_{0},L_{1},\dots ,L_{n} \right\} \) be the orthonormal bases for the subspaces \( \mathbb {X}_{n} \) of \( \mathbb {X} \) of degree \( \le n. \) Now we need to introduce the Legendre orthogonal and Legendre interpolatory projection operator.
Let \( \mathcal {P}_{n}^{G}: \mathbb {X}\rightarrow \mathbb {X}_{n} \) be the orthogonal projection defined by
where \( \left\langle u, L_{j} \right\rangle =\int _{-1}^{1}u(t)L_{j}(t)\mathrm{d}t. \)
Let \( \left\{ \tau _{0},\tau _{1}, \dots , \tau _{n} \right\} \) be the zeros of Legendre polynomial of degree \( n+1 \) and define the interpolatory projection \( \mathcal {P}_{n}^{C}:\mathbb {X}\rightarrow \mathbb {X}_{n} \) by
Now onwards, we assume that the projection operator \( \mathcal {P}_n:\mathbb {X}\rightarrow \mathbb {X}_n \) is either orthogonal projection \(\mathcal {P}_{n}^{G}\) or interpolatory projection \(\mathcal {P}_n^{C}\) for notational convenience.
Lemma 2.3
([4]) Consider\(\mathcal {P}_{n}=\mathcal {P}_{n}^{G}\)or\(\mathcal {P}_{n}^{C}\)as the projection operator is defined to be Legendre orthogonal projection or Legendre interpolatory projection operator. Then the following conditions hold:
- (i)
For\(u\in \mathcal {C}[-1,1]\), \(\Vert \mathcal {P}_{n}u\Vert _{L^{2}}\le p\Vert u\Vert _{\infty }\), wherepis a constant independent ofn.
- (ii)
There exists a constant\(c>0\)such that for any\(u\in \mathbb {X}\)and any\(n\in \mathbb {N}\),
$$\begin{aligned} \Vert u-\mathcal {P}_{n}u\Vert _{L^{2}}\le c\inf _{\phi \in \mathbb {X}_{n}}\Vert u-\phi \Vert _{L^{2}} \rightarrow 0~as~n\rightarrow \infty . \end{aligned}$$ - (iii)
For any\(u\in \mathcal {C}^{r}[-1,1],\)there exists a constantcindependent ofnsuch that
$$\begin{aligned}&\Vert \mathcal {P}_{n}u-u\Vert _{L^{2}}\le cn^{-r}\Vert u^{(r)}\Vert _{L^{2}},\\&\Vert \mathcal {P}_{n}^{G}u-u\Vert _{\infty }\le cn^{\frac{3}{4}-r}\Vert u^{(r)}\Vert _{\infty },\\&\Vert \mathcal {P}_{n}^{C}u-u\Vert _{\infty }\le cn^{\frac{1}{2}-r}\Vert u^{(r)}\Vert _{\infty }. \end{aligned}$$ - (iv)
For any\(u\in \mathcal {C}^{r}[-1,1],\)there exists a constantcindependent ofnsuch that
$$\begin{aligned}&\Vert u-\mathcal {P}_{n}^{G}u\Vert _{\infty }\le cn^{\frac{1}{2}-r}V(u^{(r)}), \end{aligned}$$where\(V(u^{(r)})\)denotes the total variation of\(u^{(r)}.\)
Note that \( \Vert u-\mathcal {P}_{n}u\Vert _{\infty }\nrightarrow 0\) as \( n\rightarrow \infty \) for any \( u\in \mathcal {C}[-1,1]. \)
Lemma 2.4
([1]) Let\(\mathbb {X}\)be a Banach space and\(\mathcal{T}\), \(\mathcal{T}_n\in \mathbb {BL(X)}\). If\(\mathcal{T}_n\)is norm convergent to\(\mathcal{T}\)or\(\mathcal{T}_n\)is\(\nu \)-convergent to\(\mathcal{T}\)and\((\mathcal{I}-\mathcal{T})^{-1}\)exists and bounded on\(\mathbb {X}\), then\((\mathcal{I}-\mathcal{T}_n)^{-1}\)exists and uniformly bounded on\(\mathbb {X}\) for sufficiently largen.
3 Legendre spectral projection methods
In this section, Legendre spectral projection methods for weakly singular Hammerstein integral equation of mixed type are being discussed. The convergence rates for approximated solution with exact solution have been evaluated in both \(L^2\) and infinity norm.
The Legendre spectral projection methods for the equation (3) is to find an approximate solution \( u_{n}\in \mathbb {X}_{n} \) such that
If \( \mathcal {P}_{n}=\mathcal {P}_n^G \), then the above scheme leads to Legendre Galerkin method, whereas if \( \mathcal {P}_{n}=\mathcal {P}_n^C \), we get the Legendre collocation method.
Let \( \mathcal {T}_{n} \) be the operator defined by
Then equation (9) can be written as
We need the following lemma and theorem for the convergence rates of the approximate solution \( u_{n} \) to the exact solution \( u_{0} \).
Lemma 3.1
For any\( u,~v\in L^{2}[-1,1] \)or\( \mathcal {C}[-1,1] \), the followings hold
Proof
Using Lipschitz continuity of \(\psi _{i}(.,u(.))\) and Cauchy-Schwarz inequality, we obtain
which completes the proof of first inequality. Similarly, using Lipschitz continuity and Cauchy-Schwarz inequality, we obtain
which completes the proof of (11). From estimate (13), we get
This completes the proof of the lemma. \(\square \)
Theorem 3.2
Let \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) be the Fréchet derivative of \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})(u) \) at \( u_{0}. \) Then \( \left\| (\mathcal {I}-\mathcal {P}_{n})\sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0})\right\| _{L^{2}} \rightarrow 0 \) as \( n\rightarrow \infty . \)
Proof
To prove \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) is a compact operator, we have to show that \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) is uniformly bounded and equicontinuous.
Now using Cauchy-Schwarz inequality, we obtain
where \( d_{i}=\sup _{t\in [-1,1]}|\psi _{i}^{(0,1)}(t,u_{0}(t))| \) and \( d=\displaystyle \sup _{i=1,2,\dots ,m}d_{i} \). Hence, \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) is uniformly bounded. Next to show equicontinuity, consider
Now using Cauchy-Schwarz inequality and Lemma 2.1 in the above estimate, we obtain
This proves that \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) is equicontinuous. Hence by Arzelá-Ascoli’s theorem \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \) is a compact operator.
Let B be a closed unit ball in \( \mathbb {X}. \) Thus, \( S=\left\{ \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0})u :u\in B \right\} \) is relatively compact set in \( \mathbb {X}. \) By using Lemma 2.3, we get
Thus, the proof is completed. \(\square \)
For rest of the paper, we assume that 1 is not an eigenvalue of the operator \( \sum _{i=1}^{m}(\mathcal {K}_{i}\psi _{i})^{\prime }(u_{0}) \).
Theorem 3.3
Let\(u_{0}\in \mathcal {C}^{r}[-1,1],~r\ge 1 \). Then for sufficiently largen, the operator\((\mathcal {I}- \mathcal {T}_{n}^{\prime }(u_{0}))\)is invertible, i.e., there exists a constant\( A_{1}>0 \)such that\( \Vert (\mathcal {I}- \mathcal {T}_{n}^{\prime }(u_{0}))^{-1}\Vert _{L^{2}}\le A_{1} <\infty \). Also the equation (9) has a unique solution\( u_{n}\in B(u_{0},\delta )=\lbrace u:\Vert u-u_{0}\Vert _{L^{2}}<\delta \rbrace \)for some\( \delta >0 \). Moreover, there exists a constant\( 0<q<1 \), independent ofnsuch that
where\( \alpha _{n}=\Vert (\mathcal {I}-\mathcal {T}_{n}^{\prime }(u_{0}))^{-1}(\mathcal {T}_{n}(u_{0})-\mathcal {T}(u_{0}))\Vert _{L^{2}}\).
Proof
Using Theorem 3.2, we get
This shows that \( \mathcal {T}_{n}^{\prime }(u_{0}) \) is norm convergent to \( \mathcal {T}^{\prime }(u_{0})\) in \( L^{2}\)-norm. Since 1 is not an eigenvalue of \( \mathcal {T}^{\prime }(u_{0})\), \( (\mathcal {I}- \mathcal {T}^{\prime }(u_{0})) \) is invertible. Then by the Lemma 2.4, \( (\mathcal {I}- \mathcal {T}_{n}^{\prime }(u_{0}))^{-1} \) exists and is uniformly bounded on \( \mathbb {X} \) for sufficiently large n, i.e., there exists a constant \( A_{1}>0 \) such that \( \Vert (\mathcal {I}- \mathcal {T}_{n}^{\prime }(u_{0}))^{-1}\Vert _{L^{2}}\le A_{1} <\infty . \)
Using \(\Vert \mathcal {P}_{n}u\Vert _{L^{2}}\le p\Vert u\Vert _{\infty } \) and estimate (11) for any \( u\in B(u_{0},\delta ) \), we get
Now using the estimate (16) and \( \Vert (\mathcal {I}- \mathcal {T}_{n}^{\prime }(u_{0}))^{-1}\Vert _{L^{2}}\le A_{1} \), we obtain
Choosing \( \delta \) in such a way that \( q\in (0,1) \), this proves the equation (4.4) of Theorem-2 of [19].
Now using Lemma 2.3, we get
Choose n large enough such that \( \alpha _{n}\le \delta (1-q) \), then equation (4.5) of Theorem-2 of [19] is satisfied. Then by applying Theorem-2 of [19], we get
where \( \alpha _{n}=\Vert (\mathcal {I}-\mathcal {T}_{n}^{\prime }(u_{0}))^{-1}(\mathcal {T}_{n}(u_{0})-\mathcal {T}(u_{0}))\Vert _{L^{2}}\). This completes the proof. \(\square \)
Theorem 3.4
Let\(u_{0}\in \mathcal {C}^{r}[-1,1] \). Let\(u_n^G\)be the Legendre Galerkin and\(u_n^C\)be the Legendre collocation approximation of the equation (9). Then the following hold
Proof
The proof of estimates in \(L^2\) norm follows directly from Theorem-3.3. Using equations (3) and (9), and \(\Vert \mathcal {P}_n\Vert _{\infty }\le c\log n\) (cf., Page-147, [3]), we have
Using estimate (10) in the first term of the right hand side of estimate (18), we get
Using \(u_n=u_n^G\) and \(u_n=u_n^C\), then using Lemma 2.3, we obtain the desired results. \(\square \)
4 Iterated Legendre spectral projection methods
In this section, the iterated Legendre spectral projection methods for Hammerstein integral equations of mixed type with weakly singular kernels have been discussed. The rate of convergence for iterated approximate solution with exact solution have been evaluated for both \(L^2\) and infinity norm.
The iterated approximate solution \(\widetilde{u}_n\) corresponding to the approximate solution \(u_n\) given by equation (9) is defined as
If \(u_n=u_n^{G}\) in equation (20), we get iterated Legendre Galerkin solution and if \(u_n=u_n^{C}\), then we get iterated Legendre collocation solution. To discuss the convergence of the iterated approximate solution \( \widetilde{u}_{n} \) to the exact solution \( u_{0} \), we need the following theorem.
Theorem 4.1
Let\(u_{0}\in \mathcal {C}^{r}[-1,1],~r\ge 1 \)and\( \widetilde{u}_{n} \)be the iterated Legendre Galerkin approximation\(\tilde{u}_n^{G}\)or iterated Legendre collocation approximation\(u_n^{C}\)of\( u_{0} \). Then the following holds
where\(h(s,t)=\sum _{i=1}^{m}k_{i}(s,t)\psi _{i}^{(0,1)}(t,u_0(t))\).
Proof
The steps of the proof follows similarly as in Theorem-4.1 of [16]. So, we omit it. \(\square \)
Theorem 4.2
Let\(u_{0}\in \mathcal {C}^{r}[-1,1],~r\ge 1 \)and\( \widetilde{u}_{n}^{G} \)be the iterated Legendre Galerkin approximation of\( u_{0} \). Then the following holds
Proof
From Theorem-4.1, we have
Since \(\mathcal {P}_n^G\) be the orthogonal projection from the space \(\mathbb {X}\) into \(\mathbb {X}_n\), then we have
By using Hölder’s inequality, Lemma 2.3, and Theorems 2 and 3 of [17], we obtain
where \(\phi _s\in \mathbb {X}_n\). Now using Theorem-3.4 and estimate (22) in estimate (21), we obtain
Hence,
Thus, the proof is completed. \(\square \)
Theorem 4.3
Let\(u_{0}\in \mathcal {C}^{r}[-1,1], ~r\ge 1 \)and\( \widetilde{u}_{n}^{C} \)be the iterated Legendre collocation approximation of\( u_{0} \). Then the following holds
Proof
From Theorem-4.1, we have
Using Cauchy-Schwarz inequality and Lemma 2.3, we get
Substituting estimate (24) and Theorem-3.4 in equation (23), we obtain
Hence,
This completes the proof. \(\square \)
5 Legendre spectral multi-projection methods
To improve the results further, we use now the Legendre spectral multi-projection methods for weakly singular Hammerstein integral equations of mixed type. Define the multi-projection operator \((\mathcal {K}_{n,i}^{M}\psi _{i}):\mathbb {X}\rightarrow \mathbb {X}\) for \(i=1,2,\ldots ,m\) by
The multi-projection method for equation (3) is to find an approximate solution \(u_n^M\in \mathbb {X}\) such that
If \(\mathcal {P}_n=\mathcal {P}_n^G\), then equation (26) leads to multi-Galerkin method and if \(\mathcal {P}_n=\mathcal {P}_n^C\), then equation (26) leads to multi-collocation method. Let
then equation (26) can be written as
The Fréchet derivative of \(\mathcal {T}_n^Mu\) at \(u_0\) is given by
To discuss the convergence rates of \(u_n^M\) to \(u_0\), we need the following lemma and theorem.
Lemma 5.1
For any\(x,y\in \mathbb {X}\), we have
Proof
Using estimate (11), we obtain
The proof is completed. \(\square \)
Theorem 5.2
Let\( u_{0} \in \mathcal {C}^{r}[-1,1], ~r\ge 1 \). Then the operator\( (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0})) \)is invertible on\( \mathcal {C}[-1,1] \)for sufficiently largen, and there exists a constant\( A_{2}>0 \)independent ofnsuch that\( \Vert (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1} \Vert _{L^{2}} \le A_{2} < \infty . \)
Proof
For any \(u\in \mathbb {X}\), we have
The first term of right hand side of estimate (28) becomes
Using equation (12) and Lemma 2.3 in the second term of right hand side of estimate (28), we obtain
Substituting estimates (29) and (30) in estimate (28), and then using Theorem-3.2, we obtain
Hence, \( \mathcal {T}_{n}^{M^{\prime }}(u_{0}) \) is norm convergent to \( \mathcal {T}^{\prime }(u_{0}) \) in \( L^{2} \)-norm for \(r>\beta \). Since 1 is not an eigenvalue of \((\mathcal {K}\psi )^{\prime }(u_0)\), \((\mathcal {I}-\mathcal {T}^{\prime }(u_{0})) \) is invertible on \( \mathbb {X} \). Then by Lemma 2.4, \( (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1} \) exists and is uniformly bounded on \( \mathbb {X} \), for some sufficiently large n, i.e., there exists some \( A_{2}>0 \) such that \( \Vert (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1} \Vert _{L^{2}} \le A_{2} < \infty . \) Thus, the proof is completed. \(\square \)
Theorem 5.3
Let\( u_{0}\in \mathcal {C}^{r}[-1,1],~r\ge 1 \). Then for sufficiently largen, the equation (26) has a unique solution\( {u}_{n}^{M} \in B(u_{0},\delta )=\lbrace u:\Vert u-u_{0}\Vert _{L^{2}}\le \delta \rbrace \)for some\( \delta >0 \). Further, there exists a constant\( 0<q<1 \), independent ofnsuch that
where\( \beta _{n}=\Vert (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1}(\mathcal {T}_{n}^{M}(u_{0})-\mathcal {T}(u_{0}))\Vert _{L^{2}}. \)
Proof
Using Lemma 5.1, we obtain
Using \( \Vert (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1}\Vert _{L^2}\le A_2 \) and estimate (31), we obtain
Choosing \(\delta \) in such a way that \(0<q<1\). This proves equation (4.4) of Theorem-2 of [19].
Now using Theorem-5.2 and Lemma 2.3, we obtain
Choosing n sufficiently large such that \(\beta _n\le \delta (1-q)\). Then equation (4.5) of Theorem-2 of [19] is satisfied. Hence, by applying Theorem-2 of [19], we obtain
where \(\beta _n=\Vert (\mathcal {I}-\mathcal {T}_{n}^{{M}^{\prime }}(u_{0}))^{-1}(\mathcal {T}_{n}^{M}(u_{0})-\mathcal {T}(u_{0}))\Vert _{L^{2}}. \) Thus, the proof of the theorem is completed. \(\square \)
Lemma 5.4
Let\( u_{0}\in \mathcal {C}[-1,1] \). Then the following holds
where\(h(s,t)=\sum _{i=1}^{m}k_{i}(s,t)\psi _i^{(0,1)}(t,u_0(t)).\)
Proof
Using Mean value Theorem, we obtain
where \(0<\theta _3<1\). Using Lipschitz continuity of \(\psi _{i}^{(0,1)}(t,u_0(.))\) and Cauchy-Schwarz inequality, we have
where \(h(s,t)=\sum _{i=1}^{m}k_{i}(s,t)\psi _i^{(0,1)}(t,u_0(t))\). Thus, we obtain the desired result.\(\square \)
Lemma 5.5
Let\( u_{0}\in \mathcal {C}^{r}[-1,1], ~r\ge 1 \). Let\( u_{n}^{M} \)be the approximation of\(u_{0}\). Then the following holds
Proof
We have
Using \(\Vert \mathcal {P}_n\Vert _{\infty }\le c\log n\) (cf., Page-147, [3]) and estimate (10) with Lemma 2.3 in the above estimate, we obtain
This completes the proof. \(\square \)
Theorem 5.6
Let\( u_{0} \in \mathcal {C}^{r}[-1,1],~r\ge 1 \). Then the Legendre multi-Galerkin approximation\( {u}_{n}^{M,G} \)of\( u_{0} \)satisfies the followings
Proof
Using Theorem-5.3, and proceeding similarly as in estimate (32), we obtain
Since \(\mathcal {P}_n^G\) be the orthogonal projection from the space \(\mathbb {X}\) into \(\mathbb {X}_n\), then we have
Using Lemma 5.4 with Hölder’s inequality and Lemma 2.3, we get
where \(\phi _s\in \mathbb {X}_n\). Using Theorems 2 and 3 of [17] in estimate (34), we get
Combining estimates (33) and (35), we obtain
Again, from Lemma 5.5 and using estimate (36), we get
Hence, the proof is completed. \(\square \)
Theorem 5.7
Let\( u_{0} \in \mathcal {C}^{r}[-1,1],~r\ge 1 \). Then the Legendre multi-collocation approximation\( {u}_{n}^{M,C} \)of\( u_{0} \)satisfies the following
Proof
As proceeding similarly as in equation (33), we obtain
Using Lemmas 5.4 and 2.3 with Cauchy-Schwarz inequality, we get
Now combining estimates (37) and (38), we obtain
Substituting estimate (39) in Lemma 5.5, we get
This completes the proof. \(\square \)
6 Iterated Legendre multi-projection methods
In this section, we discuss on the iterated Legendre spectral multi-projection methods for weakly singular Hammerstein integral equations of mixed type in both \(L^2\) and infinity norm.
The iterated approximate solution \(\widetilde{u}_{n}^{M}\) corresponding to the approximate solution \(u^M_n\) given by equation (26) is defined as follows:
To obtain the supercovergence results for the iterated approximate solution \( \widetilde{u}_{n}^{M} \) to the exact solution \( u_{0} \), we need the following lemmas.
Lemma 6.1
For any\(x,y,z\in \mathbb {X}\), the following holds
Proof
Using \(\Vert \mathcal {P}_n\Vert _{\infty }\le c\log n\) (cf., Page-147, [3]) and estimate (11), we obtain
The proof is completed. \(\square \)
Theorem 6.2
Let\( u_{0}\in \mathcal {C}[-1,1] \). Let\( \widetilde{u}_{n}^{M} \)be the iterated approximation of\( u_{0}. \)Then the following holds
Proof
The steps of the proof follows similarly as in Theorem 4.1 of [16]. So, we omit it. \(\square \)
Theorem 6.3
Let\( u_{0}\in \mathcal {C}^{r}[-1,1],~r\ge 1 \). Then the iterated Legendre multi-Galerkin approximation\( \widetilde{u}_{n}^{M,G} \)of\( u_{0} \)satisfies the followings
Proof
From Theorem 6.2, we have
Since \(\mathcal {P}_n^G\) be the orthogonal projection from the space \(\mathbb {X}\) into \(\mathbb {X}_n\), then we have
By applying Hölder’s inequality, estimate (35), and Theorems 2 and 3 of [17], we obtain
where \(\phi _s \in \mathbb {X}_n\). Substituting estimate (43) and Theorem 5.6 in equation (42), we obtain
This completes the proof. \(\square \)
Theorem 6.4
Let\( u_{0}\in \mathcal {C}^{r}[-1,1],~ r\ge 1 \). Then the iterated Legendre multi-collocation approximation\( \widetilde{u}_{n}^{M,C} \) of \( u_{0} \)satisfies the following
Proof
From Theorem 6.2, we have
Using Cauchy-Schwarz inequality, Lemmas 2.3 and 5.4, we obtain
Substituting estimate (45) and Theorem 5.7 in equation (44), we obtain
Hence, the result. \(\square \)
Remark 6.5
Let \(u_n^{G}\), \(u_n^{C}\), \(\widetilde{u}_n^{G}\), \(\widetilde{u}_n^{C}\) be the Legendre Galerkin, Legendre collocation, iterated Legendre Galerkin and iterated Legendre collocation approximations of u, respectively. Let \(u_n^{M,G}\), \(u_n^{M,C}\), \(\widetilde{u}_n^{M,G}\), \(\widetilde{u}_n^{M,C}\) be the Legendre multi-Galerkin, Legendre multi-collocation, iterated Legendre multi-Galerkin, iterated Legendre multi-collocation approximations of u.
From Theorems-3.4, 4.2, 4.3, 5.6, 5.7, 6.3 and 6.4, we observe the following convergence rates in the respective methods.
Legendre projection methods:
Iterated Legendre projection methods:
Legendre multi-projection methods:
Iterated Legendre multi-projection methods:
We observe that
- 1.
Iterated Galerkin improves over Galerkin, multi-Galerkin improves over Iterated Galerkin and Iterated multi-Galerkin improves over multi-Galerkin in \(L^2\) norm using Legendre polynomial bases.
- 2.
However, in infinity norm, multi-Galerkin improves over Galerkin and iterated multi-Galerkin improves over iterated Galerkin using Legendre polynomial bases.
- 3.
In collocation method, no improvement recorded from collocation to multi-collocation and iterated collocation to iterated multi-collocation in \(L^2\) norm.
- 4.
In infinity norm, multi-collocation improves over collocation method. However, there is no improvement from iterated collocation to iterated multi-collocation method.
References
Ahues, M., A. Largillier, and B.V. Limaye. 2001. Spectral Computations for Bounded Operators. New York: Chapman and Hall/CRC.
Allouch, C., D. Sbibih, and M. Tahrichi. 2018. Numerical solutions of weakly singular Hammerstein integral equations. Applied Mathematics and Computation 329: 118–128.
Atkinson, K.E. 1997. The numerical solution of integral equations of the second kind. Cambridge: Cambridge University.
Canuto, C., M.Y. Hussaini, A. Quarteroni, and T.A. Zang. 2006. Spectral Methods: Fundamentals in Single Domains. Berlin: Springer.
Das, P., and G. Nelakanti. 2015. Convergence analysis of Legendre spectral projection methods for Hammerstein integral equations of mixed type. Journal of Applied Mathematics and Computer Science 49: 529–555.
Ganesh, M., and M.C. Joshi. 1989. Discrete numerical solvability of Hammerstein integral equation of mixed type. Journal of Integral Equations and Applications 2 (1): 107–124.
Ganesh, M., and M.C. Joshi. 1991. Numerical solvability of Hammerstein integral equation of mixed type. IMA Journal of Numerical Analysis 11 (1): 21–31.
Hashemizadeh, E., M. Khorramizadeh, and M. Shahbazi. 2018. Numerical solution for system of nonlinear Fredholm-Hammerstein integral equations based on hybrid Bernstein Block-Pulse functions with the Gauss quadrature rulec. Asian-European Journal of Mathematics 11 (6):
Hashemizadeh, E., and M. Rostami. 2015. Numerical solution of Hammerstein integral equations of mixed type using sinc-collocation method. Journal of Computational and Applied Mathematics 279: 31–39.
Heydari, M., Z. Avazzadeh, H.R. Navabpour, and G.B. Longhmani. 2013. Numerical solution of Fredholm integral equations of the second kind by using integral mean value theorem II. Applied Mathematical Modelling 37: 432–442.
Kaneko, H., R.D. Noren, and P.A. Padilla. 1997. Superconvergence of the iterated collocation methods for Hammerstein equations. Journal of Computational and Applied Mathematics 80 (2): 335–349.
Kaneko, H., R.D. Noren, and Y. Xu. 1990. Regularity of the solution of Hammerstein equations with weakly singular kernels. Integral Equations and Operator Theory 13 (5): 660–670.
Maleknejad, K., and E. Hashemizadeh. 2011. Numerical solution of the dynamic model of a chemical reactor by Hybrid functions. Procedia Computer Science 3: 908–912.
Maleknejad, K., and E. Hashemizadeh. 2011. A numerical approach for Hammerstein integral equations of mixed type using operational matrices of Hybrid functions. UPB Scientific Bulletin, Series A 73 (3): 95–104.
Panigrahi, B.L. 2019. Error analysis of Jacobi spectral collocation methods for Fredholm-Hammerstein integral equations with weakly singular kernel. International Journal of Computer Mathematics 96 (6): 1230–1253.
Panigrahi, B.L. 2018. Legendre spectral projection methods for Hammerstein integral equations with weakly singular kernel. International Journal of Applied and Computational Mathematics 4 (143): 1–15.
Panigrahi, B.L., and G. Nelakanti. 2013. Legendre Galerkin method for weakly singular Fredholm integral equations and the corresponding eigenvalue problem. Journal of Applied Mathematics and Computing 43: 175–197.
Sohrabi, S., H. Ranjbar, and M. Saei. 2017. Convergence analysis of the Jacobi-collocation method for nonlinear weakly singular Volterra integral equations. Applied Mathematics and Computation 299: 141–152.
Vainikko, G.M. 1967. A perturbed Galerkin method and the general theory of approximate methods for non-linear equations. USSR Computational Mathematics and Mathematical Physics 7 (4): 1–41.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Patel, S., Panigrahi, B.L. Legendre spectral projection methods for weakly singular Hammerstein integral equations of mixed type. J Anal 28, 387–413 (2020). https://doi.org/10.1007/s41478-019-00175-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41478-019-00175-3
Keywords
- Hammerstein integral equations of mixed type
- Weakly singular kernels
- Legendre spectral projection methods
- Multi-projection methods