Abstract
Spectral methods for solving differential/integral equations are characterized by the representation of the solution by a truncated series of smooth functions. The unknowns to be determined are the expansion coefficients in such a representation. The goal of this article is to give an overview of numerical problems encountered when determining these coefficients and the rich variety of techniques proposed to solve these problems. Therefore, a series of explicit formulae expressing the derivatives, integrals and moments of a class of orthogonal polynomials of any degree and for any order in terms of the same polynomials are addressed. We restrict the current study to the orthogonal polynomials of the Hermite, generalized Laguerre, Bessel, and Jacobi (including Legendre, Chebyshev, and ultraspherical) families. Moreover, formulae expressing the coefficients of an expansion of these polynomials which have been differentiated or integrated an arbitrary number of times in terms of the coefficients of the original expansion are given. In addition, formulae for the polynomial coefficients of the moments of a general-order derivative of an infinitely differentiable function in terms of its original expanded coefficients are also presented. A simple approach to build and solve recursively for the connection coefficients between different orthogonal polynomials is established. The essential results are summarized in tables which could serve as a useful reference to numerical analysts and practitioners. Finally, applications of these results in solving differential and integral equations with varying polynomial coefficients, by reducing them to recurrence relations in the expansion coefficients of the solution, are implemented.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Spectral methods have been used extensively in numerical approximation of differential (integral) equations [1, 2]. These methods use formulae relating the expansion coefficients of derivatives (integrals) appearing in the differential (integral) equation to those of the function itself. In fact, the coefficients of successive derivatives (integrals) of a function are related by a recurrence relation which greatly facilitates the setting up of an algebraic system to determine these unknown coefficients. The advantage of these recurrence formulae is that they possess good stability in the numerical treatment.
The traditional way to introduce a spectral method starts by approximating a solution f(x) by a finite sum
where \(\left\{ {\phi _k } \right\} _{k = 0}^\infty \) is the set of basis functions. The main question which arises is how to choose the basis functions? Once the choice of the basis functions is made, the second question appears: how to determine the expansion coefficients \(a_k\)? A successful expansion basis meets the following requirements:
-
(1)
The approximations \(f_n(x)\) should converge rapidly to f(x) as \(n\rightarrow \infty \).
-
(2)
Given coefficients \(\left\{ {a_k } \right\} _{k = 0}^n\), it should be easy to determine another set of coefficientsFootnote 1\(\left\{ {a^{(p)}_k } \right\} _{k = 0}^n\), \(\left\{ {b^{p,m}_k } \right\} _{k = 0}^n\) and \(\left\{ {b^{(p)}_k } \right\} _{k = 0}^n\) such that
$$\begin{aligned} \begin{aligned} \bullet \qquad&\frac{{\mathrm{d}^p f(x)}}{{{\text {d}}x^p }}=f^{(p)}(x)=\sum \limits _{k=0}^{n}a_k \frac{{\mathrm{d}^p \phi _k(x)}}{{{\text {d}}x^p }} \rightsquigarrow \sum \limits _{k=0}^{n}a^{(p)}_k \phi _k(x),\\ \bullet \qquad&x^m \frac{{\mathrm{d}^p f(x)}}{{{\text {d}}x^p }}=\sum \limits _{k=0}^{n}a^{(p)}_k x^m \phi _k(x) \rightsquigarrow \sum \limits _{k=0}^{n}b^{p,m}_k \phi _k(x),\ m \ge 0,\\ \bullet \qquad&\mathcal {I}^{p}f(x)=f^{(-p)}(x)=\sum \limits _{k=0}^{n}a_k \mathcal {I}^{p} \phi _k(x) \rightsquigarrow \sum \limits _{k=0}^{n}b^{(p)}_k \phi _k(x), \end{aligned} \end{aligned}$$(2)where \(\mathcal {I}^{p}\) is the pth integral operator.
-
(3)
The computation of expansion coefficients \(\left\{ {a_k } \right\} _{k = 0}^n\) from function values \(\left\{ {f(x_i) } \right\} _{i = 0}^n\) and the reconstruction of solution values in nodes from the set of coefficients \(\left\{ {a_k } \right\} _{k = 0}^n\) should be easy, i.e., the conversion between two data sets is algorithmically efficient
$$\begin{aligned} \left\{ {f(x_i) } \right\} _{i = 0}^n\ \leftrightarrows \ \left\{ {a_k } \right\} _{k = 0}^n. \end{aligned}$$
The common denominator of spectral methods is to rely on high-order polynomial expansions, notably trigonometric polynomials for periodic problems, and orthogonal polynomials for nonperiodic boundary value problems. The computation of the expansion coefficients in (2) is the dominant part of the spectral methods. It is also the most time-consuming part of spectral tau and Galerkin methods.
The key computational task in constructing these polynomial approximations and solving differential (integral) equations with polynomial coefficients in spectral methods is the evaluation of the expansion coefficients of the derivatives (integrals) and moments of high-order derivatives of infinitely differentiable functions. This is the main issue that we address in this study, where we restrict our attention to the orthogonal polynomial expansions of the Hermite, generalized Laguerre, Bessel, and Jacobi (including Legendre, Chebyshev, and ultraspherical) families. Formulae for the expansion coefficients of a general-order derivative of an infinitely differentiable function in terms of those of the function itself are constructed for expansions in Chebyshev [3, 5], Legendre [6], ultraspherical [7, 8], Jacobi [9, 10], generalized Laguerre [11], Hermite [12], Bessel [13] and Bernstein [14, 15] polynomials. Many different algorithms for finding the recurrence relations for connection and linearization coefficients for these families are discussed and developed by many authors, see, for instance, [16,17,18,19,20]. It was found that the use of integral operations for constructing spectral approximations improves their rate of convergence, and allows the multiple-boundary conditions to be incorporated more efficiently [21, 22]. The application of integral operators for the treatment of differential equations by orthogonal polynomials dates back to Clenshaw [23] in the late 1950’s. The spectral approximation of the integration form of differential equations was put forward later in the 1960’s in [24] in the spectral space and in [25] in the physical space. The reason for the success of the spectral integration approaches is basically because differentiation is inherently sensitive, as small perturbations in data can cause large changes in result, while integration is inherently stable. Phillips and Karageorghis [26] proved formulae relating the expansion coefficients of an infinitely differentiable function that has been integrated an arbitrary number of times in terms of the expansion coefficients of the function when the expansion functions are the ultraspherical polynomials. They also described how they can be used to solve two-point boundary value problems. Doha [27] proved the same formula but in a simpler way than the formula suggested by Phillips and Karageorghis. Doha proved more general formulae for Jacobi [28], Laguerre [29], Hermite [29] and Bessel [13] polynomials.
Our principal aims in this paper are:
-
(i)
To derive explicit formula for classical orthogonal polynomial expansion coefficients of the derivatives of an arbitrary differentiable function in terms of its original expansion coefficients.
-
(ii)
To present explicit expression for the derivatives of classical orthogonal polynomials of any degree and for any order in terms of the classical orthogonal polynomials themselves.
-
(iii)
To derive explicit formulae for classical orthogonal polynomials coefficients of the moments of a general-order derivative of an infinitely differentiable function in terms of its classical orthogonal polynomials coefficients.
-
(iv)
To obtain explicit expression for classical orthogonal polynomials of any degree that has been integrated an arbitrary number of times in terms of the classical orthogonal polynomials themselves.
-
(v)
To describe a simple algorithmic procedure to compute recursively the expansion coefficients in the connection problem and the expansion coefficients of associated classical orthogonal polynomials.
-
(vi)
To show how to use these formulae for solving ordinary differential equations with polynomial coefficients by reducing them to recurrence relations in the expansion coefficients of their solutions.
It should be mentioned that one of our aims here is to emphasize the systematic character and simplicity of our algorithm, which allows one to implement it in any computer algebra (here the Mathematica symbolic language has been used).
2 Properties of Classical Orthogonal Polynomials
A family \( y(x)=\phi _n (x) = k_n x^n + \ldots \left( n \in \left\{ {0,1, \ldots } \right\} ,\ k_n \ne 0\right) \) of polynomials of degree exactly n is a family of classical orthogonal polynomials if it is the solution of a differential equation of the type (see, [30,31,32,33])
where \(\sigma (x)=a x^2+b x+c\) is a polynomial of at most second degree and \(\tau (x)=d x+e\) is a polynomial of first degree. Since one demands that \(\phi _n(x)\) has exact degree n, then by equating the highest coefficients of \(x^n\) in (3) one gets
The solutions, \(\phi _n(x)\), of Eq. (3) usually called hypergeometric-type polynomials. These polynomials satisfy the orthogonality relation
where \(\rho (x)\) is a function satisfying the so-called Pearson equation,
provided that the following condition
is satisfied. The constant \(h_n\) can be computed from the relation
where \(B_n\) is the normalization constant appearing in the Rodrigues formula
and \(\rho _n(x) = \sigma ^n(x)\rho (x)\). The constants \(k_n\) and \(B_n\) are related by
An important property of classical orthogonal polynomials is that their derivatives, \(\phi ^{(m)}_n(x)\), form orthogonal systems. These systems are orthogonal in the interval \([a', b']\) with respect to the weight function \(\rho _m(x)\), i.e.,
where \(h^{(m)}_n\) can be expressed in terms of \(h_n\) as
and the constant \(A_{mn}\) appearing in the generalization of Rodrigues’ formula (6),
has the form
Koepf and Schmersau [34] showed that any solution \(\phi _n(x)\) of (3) satisfies a recurrence relation of the type
where the coefficients \(\alpha _n\), \(\beta _n\) and \(\gamma _n\) are given by the explicit formulae
and also satisfies a structure relation of the type
where the coefficients \({{\bar{\alpha }}} _n, \ {\bar{\beta }} _n\) and \(\bar{\gamma }_n\) are given by the explicit formulae
moreover, they proved that the power series coefficients \(C_m(n)\) given by
satisfy the recurrence relation
which carries the complete information about the hypergeometric representation of \(\phi _n (x)\).
For the sake of completeness, an appendix A has been included at the end of this paper giving the expressions of \(\sigma (x),\ \tau (x),\ \rho (x),\ \lambda _n, \ h_n,\ \alpha _n,\ \beta _n,\ \gamma _n,\ {\bar{\alpha }}_n,\ {\bar{\beta }}_n, and \ {\bar{\gamma }}_n\) and the hypergeometric series representation for each one of the referred classical orthogonal families (see, Tables 5, 6 in “Appendix”).
3 Expansion Coefficients of the Derivatives/Integrals of \(\phi _n (x)\)
The main objective of this section is to give explicit formulae for the derivatives and integrals of \(\phi _n (x)\), based on the method of Sánchez-Ruiz and Dehesa [35].
Theorem 1
In the expansion
the coefficients \(C^+_{p,i}(n)\) are given by
and satisfy the recurrence relation
Proof
Multiplying both sides of Eq. (13) by \(\rho (x)\phi _k(x)\), and integrating between \(a'\) and \(b'\), orthogonality relation (4) immediately gives
Using the Rodrigues representation (6) for \(\phi _n(x)\), this integral can be written
Integrating by parts k times, and taking into account the orthogonality condition (5), we get
Using the Rodrigues representation (8) for \(\phi _n^{(p+k)}(x)\), this integral can be written
Integrating by parts \((n-p-k)\) times, and taking into account the orthogonality condition (5), we get
which proves (14). Let us write
then using identity (11) leads to the recurrence relation (15), which completes the proof of the theorem. \(\square \)
Note 1
The integral (14) for the case of Bessel polynomials is evaluated by means of the residual theorem.
The following theorem can be proved in a similar way like that of Theorem 1.
Theorem 2
In the expansion
the coefficients \(C^-_{p,i}(n)\) are given by
and satisfy the recurrence relation
Corollary 1
It is easy to show that
where \(\pi _{p-1,n}(x)\) and \({\tilde{\pi }}_{p-1,n}(x)\) are polynomials of degree at most \((p -1)\) and the coefficients \(b_{p,i}(n)\) and \(C^-_{p,i}(n)\) are related by
and moreover, the coefficients \(b_{p,i}(n)\) satisfy a recurrence relation of the type
with \(b_{p-1,-\ell }(j)=0,\qquad \forall \ell >0,\qquad b_{0,0}=1,\qquad b_{p-1,r}(j)=0, \quad r=2p-1,\ 2p.\)
Proof
Integration of (16) p times with respect to x gives immediately (19), which in turn may be written in the form
Making use of relation (17) and knowing that deg \(\sigma ^p(x)\le 2p,\) enables one to show that
and accordingly, formula (19) may be written as formula (20). The recurrence relation (22) is a direct consequence of (18), and this completes the proof of the corollary. \(\square \)
Remark 1
It is to be noted here that the expansion coefficients \(C^+_{p,i}(n)\) and \(C^-_{p,i}(n)\) are related by
Corollary 2
If the classical orthogonal polynomials \(\phi _n(x)\) satisfy a recurrence relation of the type
then
Proof
Making use of (22), noting that \(b_{11}(n) = 0\), enables one to show that
and accordingly formula (20) takes the form (25). \(\square \)
Theorem 3
-
(i)
For Hermite polynomials [12, 29]
$$\begin{aligned} \begin{aligned} D^p H_n (x) = 2^p \frac{{n!}}{{(n - p)!}}H_{n - p} (x),\quad n,\ p \ge 0, \end{aligned} \end{aligned}$$(26)and
$$\begin{aligned} \begin{aligned} \mathcal {I}^{p}H_n (x) = 2^{-p} \frac{{n!}}{{(n + p)!}}H_{n + p} (x),\quad n,\ p \ge 0, \end{aligned} \end{aligned}$$(27) -
(ii)
For generalized Laguerre polynomials [11, 29]
$$\begin{aligned} \begin{aligned} D^p L_n^\alpha (x) = ( - 1)^p \sum \limits _{i = 0}^{n - p} {\left( {\begin{array}{*{20}c} {n - i - 1} \\ {p - 1} \\ \end{array}} \right) } L_i^\alpha (x),\quad n,\ p \ge 1, \end{aligned} \end{aligned}$$(28)and
$$\begin{aligned} \begin{aligned} \mathcal {I}^{p} L_n^\alpha (x) = \sum \limits _{i = 0}^p {( - 1)^{i + p} \left( {\begin{array}{*{20}c} p \\ i \\ \end{array}} \right) } L_{n + p - i}^\alpha (x),\quad n,\ p \ge 1. \end{aligned} \end{aligned}$$(29) -
(iii)
For Jacobi polynomials [9, 28]
$$\begin{aligned} \begin{aligned} D^p P_n^{(\alpha ,\beta )} (x)&= 2^{ - p} (n + \lambda )_p\\&\quad \times \sum \limits _{i = 0}^{n - p} {C_{n - p,i} (\alpha + p,\beta + p,\alpha ,\beta )} P_i^{(\alpha ,\beta )} (x),\quad \lambda = \alpha + \beta + 1, \end{aligned} \end{aligned}$$(30)and
$$\begin{aligned} \mathcal {I}^{p} P_n^{(\alpha ,\beta )} (x) = \frac{{2^p }}{{(n - p + \lambda )_p }}\sum \limits _{i = 0}^{2p} {C_{n + p,n + p - i} (\alpha - p,\beta - p,\alpha ,\beta )} P_{n + p - i}^{(\alpha ,\beta )} (x), \end{aligned}$$(31)where
$$\begin{aligned}\begin{aligned} C_{n,i} (\gamma ,\delta ,\alpha ,\beta )&= \frac{{(n + \eta )_i (i + \gamma + 1)_{n - i} }}{{(n - i)!(\lambda + i)_i }}\\&\quad \times {}_3F_2 \left[ {\begin{array}{*{20}c} \begin{array}{l} - (n - i),n + i + \eta ,i + \alpha + 1 \\ i+\gamma + 1,2i + \lambda + 1 \\ \end{array} &{} {;1} \\ \end{array}} \right] , \\&\qquad \eta = \gamma + \delta + 1. \end{aligned} \end{aligned}$$ -
(iv)
For Bessel polynomials [13]
$$\begin{aligned} \begin{aligned} D^p Y_n^{(\alpha )} (x)&= 2^{ - p} (n - p + 1)_p (n + \alpha + 1)_p \\&\sum \limits _{k = 0}^{n - p} {M_k (\alpha + 2p,\alpha ,n - p)Y_k^{(\alpha )} (x)}, \end{aligned} \end{aligned}$$(32)and
$$\begin{aligned} \begin{aligned} \mathcal {I}^{p} Y_n^{(\alpha )} (x) = \sum \limits _{i = 0}^{2p} {b_{p,i} (n)Y_{n + p - i}^{(\alpha )} (x)},\ n,\ p \ge 0, \end{aligned} \end{aligned}$$(33)with \(Y_{-r}^{(\alpha )}(x)=0,\ r\ge 1,\) where
$$\begin{aligned} M_i (\alpha ,\beta ,n) = ( - 1)^n (2i + \beta + 1)\frac{{(\alpha - \beta )_{n - i} ( - n)_i (\alpha + n + 1)_i }}{{i!(\beta + i + 1)_{n + 1} }}, \end{aligned}$$and
$$\begin{aligned} b_{p,i} =\left( {\begin{array}{*{20}c} {n + p} \\ i \\ \end{array}} \right) \frac{{2^p (2n + 2p - 2i + \alpha + 1)(n + \alpha - p + 1)_ {n + p - i} (2p - i + 1)_i }}{{(n + 1)_p (n + \alpha - p + 1)_p (n + p + \alpha - i + 1)_{n + p + 1} }}. \end{aligned}$$
Remark 2
For Bernstein polynomials [15]
4 The Coefficients of Differentiated/Integrated Expansions of \(\phi _n(x)\)
The main results of this section are two explicit formulae which express the expansion coefficients of a general-order derivative (integral) of an infinitely differentiable function in terms of its original expansion coefficients.
Theorem 4
Suppose we are given a regular function f(x) which is formally expanded in the infinite series
and for the pth derivatives of f(x),
then
where the coefficients \(C^+_{p,i}(n)\) are given by (14). Moreover, the coefficients \(a^{(p)}_n\) satisfy the recurrence relation
Proof
By differentiating Eq. (36) p times and using (13), we obtain
Expanding (40) and collecting similar terms, we obtain
Identifying (37) with (41) gives immediately
Now, let us write
then use of identity (11) leads to the recurrence relation (39), which completes the proof of the theorem \(\square \)
Theorem 5
Suppose that \(f^{(-m)}(x) = D^{-m}f(x)\) for some \(m \ge 1\) is an infinitely differentiable function, and f(x) is formally expanded as in (36). Let \(b^{(p)}_n,\ m \ge p \ge 1,\) denote the expansion coefficients of \(f^{(-p)}(x)\) in the expansion
where \(\pi _{p-1}(x)\) is a polynomial of degree at most \((p - 1)\), then the coefficients \(b^{(p)}_n\) are related to \(a_n\) by
and they satisfy a recurrence relation of the type
Proof
By integrating Eq.(36) p times and using (20), we obtain
Expanding (45), collecting similar terms and noting that \(\phi _{-r}(x) = 0\) for \(r > 0\), we obtain
where \(\pi _{p-1}(x)\) is a polynomial of degree at most \((p - 1)\). Identifying (46) with (42) gives immediately
Now, let us write
then making use of identity (11) leads to the recurrence relation (44), which completes the proof of the theorem. \(\square \)
In view of Corollary 3.2, we obtain the following result:
Corollary 3
If the classical orthogonal polynomials \(\phi _n(x)\) satisfy a recurrence relation of the type (24), then
In the following, we give explicit expressions that relate the expansion coefficients \(a^{(p)}_n\) and \(b^{(p)}_n\) with \(a_n\) for different expansion basis (Hermite, generalized Laguerre, Jacobi, Bessel and Bernstein):
Theorem 6
-
(i)
For Hermite polynomials [12, 29]
$$\begin{aligned} \begin{aligned} a^{(p)}_{n}=2^p\frac{(n+p)!}{n!}a_{n+p}, \end{aligned} \end{aligned}$$(48)and
$$\begin{aligned} \begin{aligned} b^{(p)}_{n}=2^{-p}\frac{(n-p)!}{n!}a_{n-p},\quad n\ge p. \end{aligned} \end{aligned}$$(49) -
(ii)
For generalized Laguerre polynomials [11, 29]
$$\begin{aligned} \begin{aligned} a^{(p)}_{n}=(-1)^p\sum \limits _{i=0}^{\infty }\left( {\begin{array}{c}p+i-1\\ p-1\end{array}}\right) a_{n+i+p}, \end{aligned} \end{aligned}$$(50)and
$$\begin{aligned} \begin{aligned} b^{(p)}_{n}=\sum \limits _{i=0}^{\infty }(-1)^{p+i}\left( {\begin{array}{c}p\\ i\end{array}}\right) a_{n+i-p},\quad n\ge p. \end{aligned} \end{aligned}$$(51) -
(iii)
For Jacobi polynomials [28]
$$\begin{aligned} \begin{aligned} a^{(p)}_{n}&= 2^{-p}\sum \limits _{i=0}^{\infty }(n+i+p+\lambda )_pC_{n+1,n}(\alpha +p,\beta +p,\alpha ,\beta )a_{n+i+p}, \\&n,p\ge 0, \end{aligned} \end{aligned}$$(52)and
$$\begin{aligned} \begin{aligned} b^{(p)}_{n}=\sum \limits _{i=0}^{2p}\frac{2^{p}}{(i+n-2p+\lambda )_p}C_{n+i,n}(\alpha -p,\beta -p,\alpha ,\beta )a_{n+i-p},\quad n\ge p. \end{aligned} \end{aligned}$$(53) -
(iv)
For Bessel polynomials [13]
$$\begin{aligned} \begin{aligned} a^{(p)}_{n}&= 2^{-p}\sum \limits _{i=0}^{\infty }(n+i+1)_p(n+p+i+\alpha +1)_p M_n(\alpha +2p, \alpha , n+i),\\&\quad n\ge 0,\ p\ge 1, \end{aligned} \end{aligned}$$(54)and
$$\begin{aligned} \begin{aligned} b^{(p)}_{n}=\sum \limits _{i=0}^{2p}b_{p,i}(n - p + i)a_{n-p+i},\quad n\ge p. \end{aligned} \end{aligned}$$(55)
Remark 3
For Bernstein polynomials [14]
where
5 Products of Powers and Orthogonal Polynomials
For the evaluation of the expansion coefficients of \(x^{m}f^{(p)}(x)\) in a series of \(\phi _{n}(x)\), the following theorem is needed.
Theorem 7
([35], p. 163)
where
\(k_{-}=\max (0,n-m),k_{+}=\min (n,j).\)
The explicit expressions of (56) when \(\{\phi _{n}(x)\}\) is one of the classical families of Hermite, Laguerre, Jacobi and Bessel polynomials are given in [35] formulae (3.10), (3.17), (3.22) and (3.31), pp. 164–168.
Lemma 1
The expansion (56) may be written in the form
Proof
The expansion (56) can be written as
then multiplying both sides by \(\rho (x)\phi _{j+m-k}(x)\), integrating between \(a'\) and \(b'\) and making use of relation (4), yield
Now, applying the orthogonality properties of \(\phi _{n}\) for \(n =2m+1,\ \ldots ,\ j + m\), we obtain
and by noting that \(\phi _{-r}(x)=0\) for \(r>0\), we get
and accordingly, Eq. (59) takes the form of (58), which completes the proof of Lemma 1. \(\square \)
Corollary 4
If the classical orthogonal polynomials \(\phi _{j}(x)\) satisfy a recurrence relation of the type
then
Proof
Using recurrence relation (59) and putting \(a_{11}(j) = 0\), then it is not difficult to show that \(a_{m,n}(j)=0\), for n odd, and accordingly formula (58) takes the form (62). \(\square \)
The explicit expressions of (58) when \(\{\phi _{n}(x)\}\) is one of the classical orthogonal polynomials (Hermite, generalized Laguerre, Jacobi, Bessel) are given in the following theorem.
Theorem 8
-
(i)
For Hermite polynomials [12]
$$\begin{aligned} x^{m}H_{j}(x)=\sum _{n=0}^{m}a_{m,2n}(j)H_{j+m-2n}(x),\quad m,\ n\ge 0, \end{aligned}$$(63)where
$$\begin{aligned} a_{m,2n}(j)&= \frac{2^{j-m}m!j!}{(j+m-2n)!}\sum \limits _{k=\max (0,j-2n)}^{\min (j+m-2n,j)}\left( \begin{array}{*{20}c} j+m-2n \\ k\end{array}\right) \nonumber \\&\frac{1}{2^{k}(j-k)!(n+k-j)!}. \end{aligned}$$(64) -
(ii)
For generalized Laguerre polynomials [11]
$$\begin{aligned} x^{m}L_{j}^{\alpha }(x)=\sum _{n=0}^{2m}a_{m,n}(j)L_{j+m-n}^{\alpha }(x),\quad m,\ n\ge 0, \end{aligned}$$(65)where
$$\begin{aligned} \begin{aligned} a_{m,n}(j)&=\frac{(-1)^{m-n}(m!)^{2}}{\Gamma (j+m-n+\alpha +1)}\\&\quad \times \sum _{k=\max (0,j-n)}^{\min (j+m-n,j)}\left( \begin{array}{*{20}c} j+m-n \\ k\end{array}\right) \frac{\Gamma (m+k+\alpha +1)}{(j-k)!(n-j+k)!(m-j+k)!}. \end{aligned} \end{aligned}$$(66) -
(iii)
For Jacobi polynomials [10, 28]
$$\begin{aligned} x^{m}P_{j}^{\alpha ,\beta }(x)=\sum _{n=0}^{2m}a_{m,n}(j)P_{j+m-n}^{\alpha ,\beta }(x), \end{aligned}$$(67)where
$$\begin{aligned}&a_{m,n}(j)=\frac{(-1)^{n}2^{j+m-n}m!(2j+2m-2n+\lambda )(j+\lambda )_{m-n}}{(j+\alpha +1)_{m-n}(j+\beta +1)_{m-n}}\nonumber \\&\quad \times \sum _{k=\max (0,j-n)}^{\min (j+m-n,j)}\frac{\left( {\begin{array}{c}j+m-n\\ k\end{array}}\right) \Gamma (j+k+\lambda )}{2^{k}(n-j+k)!\Gamma (3j+2m-2n-k+\lambda +1)}\nonumber \\&\quad \times \sum _{\ell =0}^{j-k}\frac{(-1)^{\ell }\Gamma (2j+m-n-k-\ell +\alpha +1)\Gamma (j+m+\ell -n+\beta +1)}{\ell !(j-k-\ell )!\Gamma (j-\ell +\alpha +1)\Gamma (k+\ell +\beta +1)}\nonumber \\&\quad \times _{2}F_{1}\left[ \left. { \begin{array}{c} j-n-k,j+m-n+\beta +\ell +1\\ 3j+2m-2n-k+\lambda +1 \\ \end{array} } \right| 2 \right] ,\quad \lambda =\alpha +\beta +1. \end{aligned}$$(68) -
(iv)
For Bessel polynomials [13]
$$\begin{aligned} x^{m}Y_{j}^{(\alpha )}(x)=\sum _{n=0}^{2m}a_{m,n}(j)Y_{j+m-n}^{(\alpha )}(x),\quad m\ge 0, \ j\ge 0, \end{aligned}$$(69)where
$$\begin{aligned} \begin{aligned} a_{m,n}(j)&=\frac{(-1)^{j-n}2^{m}m!j!(2j+2m-2n+\alpha +1)\Gamma (j+m-n+\alpha +1)}{(j+m-n)!(2m-n)!\Gamma (j+\alpha +1)\Gamma (2j+2m-n+\alpha +2)}\\&\quad \times \sum _{k=\max (0,j-n)}^{\min (j+m-n,j)}\left( \begin{array}{*{20}c} j+m-n \\ k\end{array}\right) \\&\quad \times \frac{(-1)^{k}\Gamma (j+k+\alpha +1)\Gamma (j+2m-n-k+1)}{(j-k)!(n+k-j)!}.\\ \end{aligned} \end{aligned}$$(70)
Note 2
Doha [11, 12, 28] and Doha and Ahmed [13] proved that the explicit expressions of (64), (66), (68) and (70), when \(\{\phi (x)\}\) is one of the classical orthogonal polynomials (Hermite, generalized Laguerre, Jacobi, Bessel), respectively, satisfy the recurrence relation
with \(a_{m-1,-\ell }(j)=0,\) \(\forall \ell >0,\) \(a_{0,0}(j)=1\), \(a_{m-1,r}(j)=0,\) \(r=2m-1,\ 2m.\)
6 The Expansion Coefficients of the Moments of a General-Order Derivative of an Infinitely Differentiable Function
In this section, we state and prove a theorem which relates the expansion coefficients of \(x^{\ell }f^{(p)}\) in terms of \(a_{i}^{(p)}\).
Theorem 9
Assume that f(x), \(f^{(p)}(x)\) and \(x^\ell \phi _{j}(x)\) have the expansions (36), (37) and (58), respectively, and assume also that
then the expansion coefficients \(b_{i}^{p,\ell }\) are given by
Proof
By letting \(i=k+\ell -j\), then (74) may be written in the form
where
By noting that \(\phi _{-i}(x)=0\), for \(i\ge 1\), then it can be easily shown that
hence
If when considering \(\sum \limits _{2}\), one takes \(k+\ell \) instead of k, then it is not difficult to show that
Substitution of (76) and (77) into (75) gives the required results of (73) and completes the proof of the theorem. \(\square \)
7 Connection Coefficients Between Different Classical Orthogonal Polynomial Systems
In this section, we consider the problem of determining the connection coefficients between different orthogonal polynomial systems. An interesting question is how to transform the Fourier coefficients of a given polynomial corresponding to an assigned orthogonal basis, into the coefficients of another basis orthogonal with respect to a different weight function. The aim is to determine the so-called connection coefficients of the expansion of any element of the first basis in terms of the elements of the second basis.
Suppose V is a vector space of all polynomials over the real or complex numbers and \(V_m\) is the subspace of polynomials of degree less or equal to m. Suppose \(p_0(x), p_1(x), p_2(x),\ldots \) is a sequence of polynomials such that \(p_n(x)\) is of exact degree n; let \(q_0(x), q_1(x), q_2(x),\ldots \) be another such sequence. Clearly, these sequences form a basis for V. It is also evident that \(p_0(x), p_1(x), \ldots , p_m(x)\) and \(q_0(x), q_1(x),\ldots , q_m(x)\) give two bases for \(V_m\). While working with finite-dimensional vector spaces, it is often necessary to find the matrix that transforms a basis of a given space to another basis. This means that one is interested in the connection coefficients \(a_i(n)\) that satisfy
where \({\bar{a}}\) and \({\bar{b}}\) are constants.
The connection coefficients between many of the classical orthogonal polynomial systems have been determined by different kinds of methods, see e.g., [36,37,38]. The aim of this section is to describe a simple procedure (based on the results of Theorem 9) to find recurrence relations between the coefficients \(a_i(n)\) when \(p_i(x)\) and \(q_i(x)\) belong to the class of classical orthogonal polynomials. This gives an alternative and different way to be compared to the approaches of Askey and Gasper [39], Ronveaux et al. [40, 41], Area et al. [16], Koepf and Schmersau [34], Lewanowicz [19, 20, 42], Lewanowicz et al. [17], and S\(\acute{a}\)nchez-Ruiz and Dehesa [35].
The differential equation satisfied by \(\psi _n({\bar{a}}x+{\bar{b}})\) for the cases of Hermite, generalized Laguerre, Jacobi and Bessel polynomials has the form
which may be written in the form
where \(b_2 = a\ {\bar{a}}^2,\ b_1 = 2a {\bar{a}}{\bar{b}}+{\bar{a}}b,\ b_0 = a{\bar{b}}^2+b {\bar{b}}+c,\ c_1 = d{\bar{a}},\ c_0 ={\bar{a}}(d{\bar{b}}+e)\) and \(\mu = {\bar{a}}^2\lambda _n\).
By substituting (78) and by virtue of formulae (72), Eq. (79) takes the form
and by making use of (71) and (73), we obtain
which is of order 4, where
Following the same procedure as in Example 1 of Sect. 8.1, we get recurrence relations satisfied by \(a_i(n)\) when \(\{\phi _i(x)\}\) is one of the classical families of Hermite, generalized Laguerre, Jacobi and Bessel.
7.1 The \(\psi _n({\bar{a}}x+{\bar{b}})\)-Hermite Connection Problem
In this problem
the coefficients \(a_i(n)\) satisfy the recurrence relation
which is of order 4, where
It is to be noted here that the fourth-order recurrence relation (83) generates the coefficients \(a_i(n)\) by recurring backwards with the initial conditions given by \(a_{n+s}(n)=0,\ s = 1, 2, 3,\) and \(a_n(n) = 2^{-n} \times \) Leading coefficient of \(\psi _n({\bar{a}}x+{\bar{b}})\). Table 1 summarizes the \(\psi _n({\bar{a}}x+{\bar{b}})\)-Hermite connection coefficients.
7.2 The \(\psi _n({\bar{a}}x+{\bar{b}})\)-Generalized Laguerre Connection Problem
In this problem
the coefficients \(a_i(n)\) satisfy the recurrence relation
where
with \(a_{n+s}(n)=0,\ s = 1, 2, 3,\) and \(a_n(n) = (-1)^nn! \times \) Leading coefficient of \(\psi _n({\bar{a}}x+{\bar{b}})\). Table 2 summarizes the \(\psi _n({\bar{a}}x+{\bar{b}})\)-generalized Laguerre connection coefficients.
7.3 The \(\psi _n({\bar{a}}x+{\bar{b}})\)-Jacobi Connection Problem
In this problem
the coefficients \(a_i(n)\) satisfy the recurrence relation
where
with \(a_{n+s}(n)=0,\ s = 1, 2, 3,\) and \(a_n(n) = \frac{2^nn!}{(n+\lambda )_n} \times \) Leading coefficient of \(\psi _n({\bar{a}}x+{\bar{b}})\). Table 3 summarizes the \(\psi _n({\bar{a}}x+{\bar{b}})\)-Jacobi connection coefficients.
7.4 The \(\psi _n({\bar{a}}x+{\bar{b}})\)-Bessel Connection Problem
In this problem
the coefficients \(a_i(n)\) satisfy the recurrence relation
where
with \(a_{n+s}(n)=0,\ s = 1, 2, 3,\) and \(a_n(n) = \frac{2^n}{(n+\alpha +1)_n} \times \) Leading coefficient of \(\psi _n({\bar{a}}x+{\bar{b}})\). Table 4 summarizes the \(\psi _n({\bar{a}}x+{\bar{b}})\)-Bessel connection coefficients.
8 Applications
8.1 Ordinary Differential Equations with Varying Coefficients
Let f(x) has the expansion (36), and assume that it satisfies the linear nonhomogeneous differential equation of order m
where \(p_0,p_1,\ldots ,(p_m \ne 0)\) are polynomials in x, and the expansion coefficients of the function p(x) in terms of \(\phi _n(x)\) are known, then formulae (38), (58) and (73), enable us to construct in view of Eq. (90) the linear recurrence relation of order r,
where \(\alpha _0,\alpha _1,\ldots ,\alpha _r\ (\alpha _0 \ne 0, \alpha _r \ne 0)\) are polynomials of the variable k.
An example dealing with nonhomogeneous differential equation is considered to clarify application of the results obtained.
Example 1
Consider the nonhomogeneous differential equation
If f(x) and \(e^{-x}\) are expanded in terms of Hermite polynomials, \(H_i(x)\), in the forms
and
then by virtue of formulae (72), Eq. (91) takes the form
By making use of (63) and (73) for the Hermite case, we obtain
Using formula (48) with (92) yields
The complete solution for Example 1 may be obtained by solving the recurrence relation (93). What is worthy noting that the analytical solution for this recurrence relation is given explicitly by
Analytical solution like (94), is not generally easy to obtain. The alternative approach for solving (93) can be obtained using the modification of Miller’s recurrence algorithm, see [43, 44].
8.2 The Integrated System of Ordinary Differential Equations with Polynomial Coefficients
Let f(x) has the expansion (36), and assume that it satisfies the linear nonhomogeneous differential equation (90). The integration of Eq. (90) m times with respect to x, gives
where \(e_0, \ e_1,\ \ldots , \ e_{m-1}\) are constants of integration. It can be easily shown that
Substitution of (96) into (95) and collecting similar terms containing the same number of repeated integrations yield
Equation (97) may be written in the form
where
If the expansion (36) is substituted into (98), and a linearization of \(Q_i(x)\ \phi _n(x)\) as a linear combination of suitable \(\phi _n(x)\) is made, and if \(D^{-m}p(x)\) is expanded into a series of \(\phi _n(x)\), then making use of (20) enables us to obtain a recurrence relation for the expansion coefficients \(a_n\) of the form
where \({\tilde{\alpha }}_0,{\tilde{\alpha }}_1,\ldots ,{\tilde{\alpha }}_r \ ({\tilde{\alpha }}_0 \ne 0, {\tilde{\alpha }}_s \ne 0)\) are polynomials of the variable k.
Notes
Here the superscript does not mean a derivative!
References
Canuto, C., Hussaini, M.Y., Quarteroni, A., Zang, T.A.: Spectral Methods in Fluid Dynamics. Springer, Berlin (1988)
Gottlieb, D., Orszag, S.A.: Numerical Analysis of Spectral Methods. Theory and Applications. SIAM, Philadeliphia (1977)
Doha, E.H., Abd-Elhameed, W.M.: On the coefficients of integrated expansions and integrals of Chebyshev polynomials of third and fourth kinds. Bull. Malays. Math. Sci. Soc. 37(2), 383–398 (2014)
Fox, L., Parker, I.B.: Chebyshev Polynomials in Numerical Analysis. Oxford University Press, London (1972)
Karageorghis, A.: A note on the Chebyshev coefficients of the general-order derivative of an infinitely differentiable function. J. Comput. Appl. Math. 21, 129–132 (1988)
Phillips, T.N.: On the Legendre coefficients of a general-order derivative of an infinitely differentiable function. IMA. J. Numer. Anal. 8, 455–459 (1988)
Karageorghis, A., Phillips, T.N.: On the coefficients of differentiated expansions of ultraspherical polynomials, ICASE Report No. 89–65 (1989), NASA Langley Research Center, Hampton. VA Appl. Num. Math. 9(1992), 133–141
Doha, E.H.: The coefficients of differentiated expansions and derivatives of ultraspherical polynomials. J. Comput. Math. Appl. 21, 115–122 (1991)
Doha, E.H.: On the coefficients of differentiated expansions and derivatives of Jacobi polynomials. J. Phys. A: Math. Gen. 35, 3467–3478 (2002)
Eslahchi, M.R., Dehghan, M., Ahmadi\_Asl, S.: The general Jacobi matrix method for solving some nonlinear ordinary differential equations. Appl. Math. Model. 36(8), 3387–3398 (2012)
Doha, E.H.: On the connection coefficients and recurrence relations arising from expansions in series of Laguerre polynomials. J. Phys. A: Math. Gen. 36, 5449–5462 (2003)
Doha, E.H.: On the connection coefficients and recurrence relations arising from expansions in series of Hermite polynomials. Integral Transforms Spec. Funct. 15(1), 13–29 (2004)
Doha, E.H., Ahmed, H.M.: Recurrences and explicit formulae for the expansion and connection coefficients in series of Bessel polynomials. J. Phys. A: Math. Gen. 37, 8045–8063 (2004)
Doha, E.H., Bhrawy, A.H., Saker, M.A.: Integrals of Bernstein polynomials: an application for the solution of high even-order differential equations. Appl. Math. Lett. 24(4), 559–565 (2011)
Doha, E.H., Bhrawy, A.H., Saker, M.A.: On the derivatives of Bernstein polynomials: an application for the solution of high even-order differential equations. Bound. Value Probl. 2011(1), 829543 (2011)
Area, I., Godoy, E., Ronveaux, A., Zarzo, A.: Minimal recurrence relations for connection coefficients between classical orthogonal polynomials: discrete case. J. Comput. Appl. Math. 89, 309–325 (1998)
Lewanowicz, S., Godoy, E., Area, I., Ronveaux, A., Zarzo, A.: Recurrence relations for coefficients of Fourier series expansions with respect to \(q\)-classical orthogonal polynomials. Numer. Algorithms 23, 31–50 (2000)
Esmaili, S., Eslahchi, M.R.: A modified spectral method for solving operator equations. J. Comput. Appl. Math. 292, 105–135 (2016)
Lewanowicz, S.: Quick construction of recurrence relations for the Jacobi coefficients. J. Comput. Appl. Math. 43, 355–372 (1992)
Lewanowicz, S.: Recurrences of coefficients of series expansions with respect to classical orthogonal polynomials. Appl. Math. 29, 97–116 (2002)
Elgindy, K.T., Smith-Miles, K.A.: Solving boundary value problems, integral, and integro-differential equations using Gegenbauer integration matrices. J. Comput. Appl. Math. 237, 307–325 (2013)
Zaky, M.A.: An improved tau method for the multi-dimensional fractional Rayleigh-Stokes problem for a heated generalized second grade fluid. Comput. Math. Appl. 75, 2243–2258 (2018)
Clenshaw, C.W.: The numerical solution of linear differential equations in Chebyshev series. Math. Proc. Camb. Phil. Soc. 53, 134–149 (1957)
Clenshaw, C.W., Curtis, A.R.: A method for numerical integration on an automatic computer. Numer. Math. 2, 197–205 (1960)
El-Gendi, S.E.: Chebyshev solution of differential, integral, and integro-differential equations. Comput. J. 12, 282–287 (1969)
Phillips, T.N., Karageorghis, A.: On the coefficients of integrated expansions of ultrspherical polynomials. SIAM J. Numer. Anal. 27, 823–830 (1990)
Doha, E.H.: On the coefficients of integrated expansions and integrals of ultraspherical polynomials and their applications for solving differential equations. J. Comput. Appl. Math. 139, 275–298 (2002)
Doha, E.H.: On the construction of recurrence relations for the expansion and connection coefficients in series of Jacobi polynomials. J. Phys. A: Math. Gen. 37, 657–675 (2004)
Doha, E.H., Ahmed, H.M., El-Soubhy, S.I.: Explicit formulae for the coefficients of integrated expansions of Laguerre and Hermite polynomials and their integrals. Integral Transforms Spec. Funct. 20(7), 491–503 (2009)
Nikiforov, A.F., Uvarov, V.B.: Special Functions of Mathematical Physics. Birkhäuser, Basel (1988)
Masjedjamei, M.: Three finite classes of hypergeometric orthogonal polynomials and their application in functions approximation. Integral Transforms Spec. Funct. 13(2), 169–190 (2002)
Masjedjamei, M.: Classical orthogonal polynomials with weight function \(((a x+ b)^2 +(c x+ d)^2)^{- p} \exp (q \arctan (a x+b)/ (c x+d)); x\in (-\infty,\infty )\) and a generalization of T and F distributions. Integral Transforms Spec. Funct. 15, 137–153 (2004)
Masjedjamei, M.: A basic class of symmetric orthogonal polynomials using the extended Sturm–Liouville theorem for symmetric functions. J. Math. Anal. Appl. 325, 753–775 (2007)
Koepf, W., Schmersau, D.: Representations of orthogonal polynomials. J. Comput. Appl. Math. 90, 57–94 (1998)
Sánchez-Ruiz, J., Dehesa, J.S.: Expansions in series of orthogonal polynomials. J. Comput. Appl. Math. 89, 155–170 (1998)
Szegö, G.: Orthogonal Polynomials, Amer. Math. Soc. Collq. Pub. 23 (1985)
Ismail, M.E.H.: Classical and Quantum Orthogonal Polynomials in One Variable. Cambridge University Press, Cambridge (2005)
Andrews, G.E., Askey, R., Roy, R.: Special Functions. Cambridge University Press, Cambridge (1999)
Askey, R., Gasper, G.: Jacobi polynomials expansions of Jacobi Polynomials with non-negative coefficients. Proc. Camb. Phil. Soc. 70, 243–255 (1971)
Ronveaux, A., Belmehdi, S., Godoy, E., Zarzo, A.: Recurrence relation approach for connection coefficients. Applications to classical discrete orthogonal polynomials, Centre de Researches Mathématiques, CRM proceedings and Lecture notes, vol. 9, pp. 319–335 (1996)
Ronveaux, A., Zarzo, A., Godoy, E.: Recurrence relations for connection coefficients between two families of orthogonal polynomials. J. Comput. Appl. Math. 62, 67–73 (1995)
Lewanowicz, S.: Recurrence relations for the coefficients in Jacobi series solutions of linear differential equations. SIAM J. Math. Anal. 17, 1037–1052 (1986)
Scraton, R.E.: A modification of Miller’s recurrence algorithm. BIT Numer. Math. 12, 242–251 (1972)
Weixlbaumer, C.: Solutions of difference equations with polynomial coefficients, Australia: RISC Linz Johannes Kepler Universitat (2001)
Acknowledgements
The authors would like to thank the anonymous referees for their helpful comments and suggestions which have improved and shortened the original manuscript to its present form.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Davod Khojasteh Salkuyeh.
Rights and permissions
About this article
Cite this article
Doha, E.H., Youssri, Y.H. & Zaky, M.A. Spectral Solutions for Differential and Integral Equations with Varying Coefficients Using Classical Orthogonal Polynomials. Bull. Iran. Math. Soc. 45, 527–555 (2019). https://doi.org/10.1007/s41980-018-0147-1
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41980-018-0147-1
Keywords
- Orthogonal polynomials
- Recurrence relations
- Linear differential equations
- Integral equations
- Connection formulae