Abstract
New sequences of orthogonal polynomials with ultra-exponential weight functions are discovered. In particular, we give an explicit solution to the Ditkin–Prudnikov problem (1966). The 3-term recurrence relations, explicit representations, generating functions and Rodrigues-type formulae are derived. The method is based on differential properties of the involved special functions and their representations in terms of the Mellin–Barnes and Laplace integrals. A notion of the composition polynomial orthogonality is introduced. The corresponding advantages of this orthogonality to discover new sequences of polynomials and their relations to the corresponding multiple orthogonal polynomial ensembles are shown.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction and preliminary results
Throughout the text, \(\mathbb {N}\) will denote the set of all positive integers, \(\mathbb {N}_{0}=\mathbb {N}\cup \{0\}\), and \(\mathbb {R}\) and \(\mathbb {C}\) the field of the real and complex numbers, respectively. The notation \(\mathbb {R}_{+}\) corresponds to the set of all positive real numbers. The present investigation is primarily targeted at analysis of sequences of orthogonal polynomials with respect to the weight functions related to the modified Bessel functions of the second kind or Macdonald functions \(K_\nu (x)\) [5, Vol. II]. The problem was posed by Ditkin and Prudnikov in the seminal work of 1966 [4] to find a new sequence of orthogonal polynomials \(\left( P_n\right) _{n \in \mathbb {N}_{0}}\), satisfying the orthogonality conditions
where \(\delta _{n,m}\) represents the Kronecker symbol, and related to the weight \(2K_0(2\sqrt{x})\) which can be defined in terms of the Mellin–Barnes integral (see [10, relation (8.4.23.1), Vol. III]).
where \(\Gamma (z)\) is the Euler gamma function [5, Vol. I]. The first four polynomials are
Later in 1993 [9] Prudnikov formulated the problem in terms of more general ultra-exponential weight functions \(\rho _{0,k-1}, \ k \in \mathbb {N}\) (see Definition 1 below), and in [13] it was announced in terms of the scaled Macdonald function
This function has the Mellin–Barnes integral representation in the form
and more general ultra-exponential weight functions can be represented, in turn, in terms of Meijer G-functions [15]. Namely, the problem is to find a sequence of orthogonal polynomials \(\left( P_n^\nu \right) _{n\in \mathbb {N}_0} \ (P_n^0 \equiv P_n)\), satisfying the following orthogonality conditions
As was shown in [13] and [2] it is more natural to investigate multiple orthogonal polynomials for two Macdonald weights \(\rho _\nu \) and \(\rho _{\nu +1}\) since it gives explicit formulas, differential properties, recurrence relations and Rodrigues formulas. Nevertheless, there is still an attractive original problem: to understand the nature of such polynomial sequences and their relation to classical systems of orthogonal polynomials and associated multiple orthogonal polynomial ensembles.
On the other hand, the operational calculus associated with the differential operator \(\frac{\mathrm{d}}{\mathrm{d}t}\) gives rise to the Laplace transform
having the exponential function as a kernel, which is the weight function for classical Laguerre polynomials [1], being represented in terms of the Mellin–Barnes integral [10, relation (8.4..3.1), Vol. III]
Meanwhile, the operator \(\frac{\mathrm{d}}{\mathrm{d}t}t\frac{\mathrm{d}}{\mathrm{d}t}\) which is also called the Laguerre derivative [3], leads to the Meijer transform [15], involving the weight \(2K_{0}(2\sqrt{x})\) which is given by (1.2), namely,
This transform is an important example of the so-called Mellin type convolution transforms, which are extensively investigated in [15]. Moreover, we will employ the Mellin transform technique developed in [15] in order to investigate various properties of the scaled Macdonald functions and more general ultra-exponential weights. Specifically, the Mellin transform is defined, for instance, in \(L_{\mu , p}(\mathbb {R}_+),\ 1 \le p \le 2\) (see details in [12]) by the integral
which is convergent in mean with respect to the norm in \(L_q(\mu - i\infty , \mu + i\infty ),\ \mu \in \mathbb {R}, \ q=p/(p-1)\). Moreover, the Parseval equality holds for \(f \in L_{\mu , p}(\mathbb {R}_+),\ g \in L_{1-\mu , q}(\mathbb {R}_+)\)
The inverse Mellin transform is given accordingly
where the integral converges in mean with respect to the norm in \(L_{\mu , p}(\mathbb {R}_+)\)
In particular, letting \(\mu = 1/p\) we get the usual space \(L_p(\mathbb {R}_+; \ \mathrm{d}x)\). Recalling the Meijer transform (1.8) one can treat it as an analog of the Laplace transform (1.6) in the operational calculus associated with the Laguerre derivative. Consequently, the corresponding analog of the classical Laguerre polynomials would be important to investigate, discovering the mentioned Ditkin–Prudnikov polynomial sequence. Finally, we note in this section that in [8] some non-orthogonal polynomial systems were investigated which share the same canonical regular form with Ditkin–Prudnikov polynomial sequence \(\left( P_n\right) _{n\in \mathbb {N}_0}\). An analogous relation occurs, for instance, between the Bernoulli polynomials, which also happen to be non-orthogonal, and the (orthogonal) Legendre polynomials.
2 Properties of the Scaled Macdonald Functions
We begin with
Definition 1
Let \(x,\gamma \in \mathbb {R}_+, \ \nu \ge 0, \ k \in \mathbb {N}_0\). The function \(\rho _{\nu ,k}(x)\) is called the ultra-exponential weight function and it is expressed in terms of the following Mellin–Barnes integral
It is easily seen from the reciprocal formulas (1.9), (1.11) for the Mellin transform that the case \(k=0\) corresponds to the weight function \(\rho _{\nu ,0}(x)= x^\nu e^{-x}\), which is related to the associated classical Laguerre polynomials \(L_n^\nu (x)\) [1]
and \(k=1\) gives the function \(\rho _{\nu ,1}\equiv \rho _\nu \), which is associated with the Prudnikov polynomials \(P_n^\nu \) under orthogonality conditions (1.5). As mentioned above the weights \(\rho _{\nu ,k}\) can be expressed in terms of the Meijer G-functions (see [7]). Concerning the scaled Macdonald function \(\rho _\nu \), we employ the Parseval equality (1.10) to the integral (1.4) to derive the Laplace integral representation for this weight function which will be used later. In fact, we obtain
The direct Mellin transform (1.9) gives the moments of \(\rho _\nu \). Specifically, we obtain
Moreover, the asymptotic behavior of the modified Bessel function at infinity and near the origin [5, Vol. II] gives the corresponding values for the scaled Macdonald function \(\rho _\nu ,\ \nu \in \mathbb {R}\). To be Precise we have
Returning to the Mellin–Barnes integral (1.4), we multiply both sides of this equality by \(x^{-\nu }\) and then differentiate with respect to x under the integral sign. This is possible via the absolute and uniform convergence by \(x \ge x_0 >0\), which can be established using the Stirling asymptotic formula for the gamma function [5, Vol. I]. Therefore we deduce
where the reduction formula \(\Gamma (z+1)= z\Gamma (z)\) for the gamma function is applied. Multiplying the latter equality by \(x^{\nu +1}\) and differentiating again, we involve a simple change of variables and the analyticity on the right half-plane \(\mathrm{Re} s > 0\) of the integrand to end up with the second order differential equation for \(\rho _\nu \)
Further, denoting the operator of the Laguerre derivative by \(\beta = DxD\) and its companion \(\theta =xDx\) (see [11]), where D is the differential operator \(D= {\mathrm{d}\over \mathrm{d}x}\), we calculate the nth power, employing amazing Viskov-type identities [14]
Equalities (2.6) can be proved by the method of mathematical induction. We show how to establish (2.6), using the Mellin transform technique for a class of functions f whose Mellin transforms (1.9) \(f^*(s),\ s=\gamma +i\tau \) belong to the Schwartz space as a function of \(\tau \). As is known, this space is a topological vector space of functions \(\varphi \) such that \( \varphi \in C^\infty (\mathbb {R})\) and \(x^m \varphi ^{(n)} (x) \rightarrow 0, \ |x| \rightarrow \infty ,\ m,n \in \mathbb {N}_0\). This means that one can differentiate under the integral sign in (1.11) infinitely many times. Hence
where
is the Pochhammer symbol [5]. On the other hand,
which proves the first identity in (2.6). Analogously,
This proves the second identity in (2.6). In particular, we easily find the values
The quotient of the scaled Macdonald functions \(\rho _\nu , \rho _{\nu +1}\) is given by the important Ismail integral representation [6]
where \(J_\nu (z), Y_\nu (z)\) are Bessel functions of the first and second kind, respectively [5]. Another interesting integral representation for the scaled Macdonald function \(\rho _\nu \) is given via [10, relation (2.19.4.13), Vol. II] in terms of the associated Laguerre polynomials. Namely, we have
Meanwhile, an important property for the scaled Macdonald functions can be obtained in terms of the Riemann–Liouville fractional integral [15]
In fact, appealing to [5, relation (2.16.3.8), Vol. II]
making simple changes of variables and letting \(\nu =0\), we derive the formula
Moreover, the index law for fractional integrals immediately implies
The corresponding definition of the fractional derivative presumes the relation \( D^\mu _{-}= - D I_{-}^{1-\mu }\). Hence for the ordinary nth derivative of \(\rho _\nu \) we get
Another way to get this formula is to differentiate n-times the integral (1.4), to use the definition of the Pochhammer symbol (2.7) and to make a simple change of variables.
In the meantime, the Mellin–Barnes integral (1.4) and reduction formula for the gamma function yield
Hence we deduce the following recurrence relation for the scaled Macdonald functions
In the operator form it can be written as follows
and more generally
Further, recalling the definition of the operator \(\theta \), identities (2.6), and the Rodrigues formula for Laguerre polynomials, we obtain
This formula permits us to derive an integral representation for the product \(\rho _\nu f_n\), where \(f_n\) is an arbitrary polynomial of degree n
In fact, considering the operator equality and using (2.20), we write
where
will be called the associated polynomial of degree 2n. Then, integrating by parts in the following integral and eliminating the integrated terms, we find
Meanwhile,
Hence, appealing to (2.3) and (2.22), we establish the following integral representation of an arbitrary polynomial \(f_n\) in terms of its associated polynomial \(q^\nu _{2n}\)
The following lemma gives the so-called linear polynomial independence of the scaled Macdonald functions.
Lemma 1
Let \(n, m \in \mathbb {N}_0,\nu \ge 0,\ f_n,\ g_{m}\) be polynomials of degree at most \(n,\ m\), respectively. Let
for all \(x >0\). Then \(f_n \equiv 0, \ g_{m} \equiv 0.\)
Proof
The proof will be based on the Ismail integral representation (2.10) of the quotient \(\rho _\nu / \rho _{\nu +1}\). In fact, let \(r \ge \max \{ n, m+1\}. \) Since \(\rho _{\nu +1} > 0\), we divide (2.25) by \(\rho _{\nu +1} \) and then differentiate r times the obtained equality. Thus we arrive at the relation
Meanwhile, integral representation (2.10) says
where the interchange of the order of integration is allowed by Fubini theorem, taking into account the asymptotic behavior of Bessel functions at infinity and near zero [5, Vol. II]. Further, assuming \(f_n\) by formula (2.21), we substitute it in the left-hand side of (2.26) together with the right-hand side of the latter equality in (2.27). Then, differentiating under the integral sign, which is possible via the absolute and uniform convergence, we deduce
Now, integrating k times by parts in the outer integral with respect to y on the right-hand side of the latter equality, we eliminate integrated terms and then differentiate under the integral sign in the inner integral with respect to s owing to the same arguments as above. Hence we get, combining with (2.26),
Consequently, cancelling twice the Laplace transform (1.6) via its injectivity for integrable continuous functions [12], and taking into account the positivity of the function
on \(\mathbb {R}_+\), we conclude that
Hence \(f_{n,k} =0,\ k=0,\dots , n\) and therefore \(f_n \equiv 0.\) Returning to the original equality (2.25), we find immediately that \(g_{m} \equiv 0.\) Lemma 1 is proved. \(\square \)
Remark 1
An alternative proof of Lemma 1 would follow from the existence of a multiple orthogonal polynomial sequence with respect to the vector of weight functions \((\rho _\nu , \rho _{\nu +1})\) (see [13]).
Let \(\alpha \in \mathbb {R}\) and
According to [13], the sequence of functions \(\left( S^{\nu ,\alpha }_n\right) _{n\in \mathbb {N}_0}\) generates multiple orthogonal polynomials related to the scaled Macdonald functions \(\rho _\nu ,\ \rho _{\nu +1}\). In order to obtain an integral representation for functions \( S^{\nu ,\alpha }_n\), we again employ (1.4), Parseval equality (1.10) for the Mellin transform, and the Mellin–Barnes integral representation for the Laguerre polynomials (see [10, relation (8.4.33.3), Vol. III]). Then, motivating the differentiation under the integral sign by the absolute and uniform convergence and using the reflection formula for the gamma function, we obtain the following chain of equalities
Thus, combining with (2.28), we established the following integral representation for \(S^{\nu ,\alpha }_n(x)\)
Now, employing recurrence relations and differential properties for the Laguerre polynomials [1], in particular, the identity \({d\over \mathrm{d}t} L_n^{\alpha } (t)= - L_{n-1}^{\alpha +1} (t)\), we integrate by parts in (2.29) and differentiate with respect to x under the integral sign by virtue of the absolute and uniform convergence by \( x \ge x_0 >0\) to deduce the corresponding relations for the sequence \(S^{\nu ,\alpha }_n\). Indeed, we have, for instance, for \(\nu >0, \alpha \in \mathbb {R}\)
Hence we obtain the identity
Differentiating (2.29) by x, we get
or,
On the other hand, integrating again by parts in (2.29) under the same conditions, we find
or,
Combining with (2.30) gives the following identity
Meanwhile, from (2.28) and (2.17) we have
Therefore from (2.32) we have
and from (2.33) we find
Moreover, recalling again (2.17), we deduce
Finally, employing the 3-term recurrence relation for Laguerre polynomials
we return to (2.29) to obtain the following identity
3 Prudnikov’s Orthogonal Polynomials
Our goal in this section is to find an explicit expression for Prudnikov’s orthogonal polynomial sequence \(\left( P_n^\nu \right) _{ n\in \mathbb {N}_0},\ \nu \ge 0.\) We will do even more, defining the Prudnikov orthogonality (1.5) in a more general setting for the sequence \(\left( P_n^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0},\ \alpha > -1,\)
Here \(P_n^\nu \equiv P_n^{\nu ,0}.\) Writing it in terms of coefficients
we know that it is of degree exactly n because this sequence is regular; i.e., its leading coefficient \(a_{n,n}\equiv a_n \ne 0\) (see [8]). Furthermore, as follows from the general theory of orthogonal polynomials [1], up to a normalization factor the orthogonality (3.1) is equivalent to the following n conditions
Moreover, the sequence \( \left( P_n^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}\) satisfies the 3-term recurrence relation in the form
where \(P_{-1}^{\nu ,\alpha } (x)\equiv 0\) and
The associated polynomial sequence with \(\left( P_n^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}\) (see (2.22)), which will be used below, has the form
It follows from the orthogonality (3.1)
However, using properties of the scaled Macdonald functions from the previous section one can calculate the following values
In fact, appealing to (3.1), (3.2), (3.6), (2.16), (2.17) and integrating by parts, we derive
since
Therefore we find the formula
In the meantime, taking the corresponding integral representation (2.11) for the product \(x^m \rho _\nu (x)\), we substitute its right-hand side in (3.3) and change the order of integration by Fubini’s theorem. Thus we obtain
But the inner integral with respect to x can be treated, involving the differential operator \(\theta \) (see (2.6)). Indeed, using (3.2) and (2.23), we have
where the interchange of the differential operator \(\theta ^k\) and integration is guaranteed due to the uniform convergence by \(t \in [1/M, M],\ M >0\) of the integral
Moreover, the Rodrigues formula for Laguerre polynomials and Viskov-type identity (2.6) for the operator \(\theta \) imply
Substituting these values in (3.9), it becomes
After m times integration by parts in the latter integral, we end up with the following orthogonality conditions
Analogously, the orthogonality (3.1) is equivalent to the equality
Definition 2
The orthogonality (3.11) is called the composition orthogonality of the sequence \(\left( P_n^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \) in the sense of Laguerre.
Thus we proved the following theorem.
Theorem 1
The Prudnikov orthogonality (3.1) is equivalent to the composition orthogonality (3.11) in the sense of Laguerre; i.e., Prudnikov’s orthogonal polynomials are Laguerre polynomials in the sense of composition orthogonality (3.11).
Meanwhile, in terms of the associated polynomial (3.6) the orthogonality conditions (3.10) can be rewritten, using the commutativity property
and the Rodrigues formula for Laguerre polynomials. Then, integrating by parts an appropriate number of times and taking into account (2.9), we get
or, finally,
On the other hand, developing the polynomial \( Q_{2n} (t)\) in terms of the Laguerre polynomials \(L_n^{\nu +\alpha }(x)\), we find
where
and orthogonality conditions (3.12) immediately imply that
Therefore, the expansion (3.13) becomes
In the meantime, expanding \((-1)^m m! x^m L_m^\nu (x)\) via the Laguerre polynomials \(L_k^{\nu +\alpha } (x)\) as well, we obtain
where coefficients \(d_{m,k}\) are calculated accordingly by the formula (see [10, relation (2.19.14.8), Vol. II])
where \({}_3F_2(a,b,c; d,e;z)\) is the generalized hypergeometric function [10, Vol. III]. It is easily seen from the orthogonality of the Laguerre polynomials \(L_k^{\nu +\alpha } (x)\) that
Moreover, the associated polynomial (3.6) \(Q_{2n}\) has the representation
Lemma 2
Coefficients \(d_{m,k},\ m,k \in \mathbb {N}_0,\) satisfy the following recurrence relation
Proof
In fact, recalling the 3-term recurrence relation (2.37) for Laguerre polynomials and, as its direct consequence, the following equality
we derive from (3.18) via integration by parts
Hence after simplification we get (3.20). \(\square \)
On by other hand, taking into account orthogonality conditions (3.15), we have
and by the uniqueness of the expansion of the associated polynomial \(Q_{2n}\) by Laguerre polynomials we find from (3.19)
We observe via (3.18) that \(d_{0,0}=1,\ d_{m, 2j} \ne 0, \ m= j,\dots , n, \ d_{m, 2j+1} \ne 0, m=j+1,\dots , n.\) But from (3.15) we get for \(n \in \mathbb {N}\)
Consequently, equalities (3.22) represent for the polynomial sequence \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}\ \left( \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \right) \) linear homogeneous systems of \(2n\ (2n+1)\) equations with \(2n+1\ (2(n+1) )\) unknowns. However, if we assume that the free coefficient \(a_{2n,0} \ ( a_{2n+1,0} ) \) is known, we come out with linear non-homogeneous systems of \(2n\ (2n+1)\) equations with \(2n\ (2n+1 )\) unknowns. It can be solved uniquely by Cramer’s rule with nonzero determinant. In fact, we have the following non-homogeneous systems of \(2n,\ 2n+1\) linear equations to determine the sequences \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0},\ \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \), respectively,
Denoting by \(D_{2n},\ D_{2n+1}\) the corresponding nonzero determinants of the systems (3.23), (3.24)
we apply Cramer’s rule to get the expressions for the coefficients of the sequences \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0},\ \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \) in terms of the related free coefficients. Precisely, denoting by
we obtain the values for coefficients of the sequences \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}\ \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \), respectively,
Moreover, returning to (3.5), we immediately obtain the values of the coefficients for the 3-term recurrence relation (3.4). Indeed, we have
In order to find free coefficients of the even and odd Prudnikov’s sequences, we appeal to the identity (3.7) and values (2.4) of the moments for \(\rho _\nu \). Thus using (3.33), we derive from (3.7) for the sequence \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}\)
Hence, taking into account the positive sign of the leading coefficient \(a_{2n}\), we get the value of \(a_{2n,0}\) in the form
Analogously, we obtain the value \(a_{2n+1,0}\) for the odd sequence \(\left( P_{2n+1}^{\nu ,\alpha }\right) _{n\in \mathbb {N}_0}\), namely,
where \( D_{2n+1,0} \equiv D_{2n+1}. \) Leading coefficients for the Prudnikov sequences have the values, accordingly,
Thus we proved the following theorem.
Theorem 2
Let \(\nu \ge 0,\ \alpha > -1,\ n \in \mathbb {N}_0.\) Prudnikov’s sequences of orthogonal polynomials \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}, \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \) have explicit values with coefficients calculated by formulas (3.33), (3.34), respectively, where the determinants \( D_{2n}, D_{2n+1}, D_{2n,k}, D_{2n+1,k}\) are defined by (3.25)- (3.32) and free coefficients \(a_{2n,0}, a_{2n+1,0}\) by (3.37), (3.38). Moreover, the 3-term recurrence relation (3.4) holds with coefficients (3.35)-(3.38).
Remark 2
It would be an interesting problem to study algebraic properties of the determinants (3.25)- (3.32) whose entries satisfy the recurrence relation (3.20).
Corollary 1
Coefficients (3.14) are calculated by formulas
where values \( D_{2n}, D_{2n+1}, D_{2n,k}, D_{2n+1,k}\) are defined by (3.25)-(3.32) and free coefficients \(a_{2n,0}, a_{2n+1,0}\) by (3.39), (3.40).
Our goal now is to find an analog of the Rodrigues formula for Prudnikov’s polynomials. To do this, we recall the representation (2.24) of an arbitrary polynomial in terms of its associated polynomial and representations (2.28), (2.29), (3.15), (3.16) to write the following equalities for the sequence \(\left( P_n^{\nu ,\alpha }\right) _{n\in \mathbb {N}_0}\)
Therefore for sequences \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0},\ \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \) we have, correspondingly,
In the meantime, the sums in (3.47), (3.48) can be treated as follows
Hence, employing the theory of multiple orthogonal polynomials associated with the scaled Macdonald functions and the related Rodrigues formulas (see details in [13, 2]), we find the following expressions
where A-polynomials in front of \(\rho _\nu \) are of degree \(j+n\) as well as B-polynomial in (3.50), while B-polynomial in (3.49) is of degree \(j+n-1\). These polynomials are explicitly calculated in [2]. Therefore formulas (3.47), (3.48) become, respectively,
But Lemma 1 presumes immediately the following identities from (3.51), (3.52)
giving explicit expressions of Prudnikov’s polynomials in terms of the multiple orthogonal polynomials for the scaled Macdonald functions, and two more relations between multiple B-polynomials
On the other hand,
Hence, recalling integral representations (2.3), (2.29), and the explicit formula for Laguerre polynomials [1], we obtain
The problem now is to express \(\rho _{\nu + k},\ k \in \mathbb {N}_0\) in terms of \(\rho _{\nu }\) and \(\rho _{\nu +1}\). To do this, we use the Mellin–Barnes representation (1.4) and the definition (2.7) of the Pochhammer symbol to derive
Then, employing the Leibniz formula and (2.16), we find
Meanwhile, employing the identity from [2] for the scaled Macdonald functions, specifically,
where \(r_{-1}(z;\nu )=0\),
formula (3.57) takes the final expression
Substituting the right-hand side of the equality (3.59) into (3.56), we get finally
Thus, returning to (3.55), we end up with the so-called Rodrigues type formula for the Prudnikov orthogonal polynomials \(P_n^{\nu ,\alpha }\)
Theorem 3
Prudnikov’s orthogonal polynomials \(P_n^{\nu ,\alpha } \) can be obtained from the Rodrigues type formula (3.60), where connection coefficients \(c_{n, j+n} \) are calculated in Corollary 1. Moreover, Prudnikov’s sequences \(\left( P_{2n}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0}, \ \left( P_{2n+1}^{\nu ,\alpha }\right) _{ n\in \mathbb {N}_0} \) are expressed in terms of multiple orthogonal polynomials related to the scaled Macdonald functions by equalities (3.53), (3.54), respectively, where the polynomials \(A_{j,j-1}^\alpha , A_{j,j}^\alpha \) are calculated explicitly in [2] by formulas
Further, the generating function for polynomials \(P_n^{\nu ,\alpha }\) can be defined as usual by the equality
where \(|z| < h_x\) and \(h_x >0\) is a convergence radius of the power series. Then returning to (3.55) and employing (2.28), we have from (3.61)
Hence substituting the value of \(x^k\rho _{\nu -k}(x)\) by formula (3.58), we get, finally, the expression for the generating function for the Prudnikov sequence \(\left( P_n^{\nu ,\alpha } \right) _{n\in \mathbb {N}_0}\), namely,
where \(c_{n,j}\) are defined in Corollary 1.
4 Orthogonal Polynomials with Ultra-Exponential Weights
In this section we will consider a sequence of polynomials \(\left( Q_n^{\nu ,k}\right) _{n\in \mathbb {N}_0}\), which is orthogonal with respect to the weight function (2.1) \(x^\alpha \rho _{\nu ,k}(x)\)
The function \(\rho _{\nu ,k}\) satisfies some interesting properties. In fact, recalling the Mellin–Barnes integral representation (2.1), we write
Hence, as in (2.18)
Further,
Thus we derive the following \(k+1\)th order differential equation for the function \(\rho _{\nu ,k}\), generalizing equation (2.5) for \(\rho _{\nu ,1} \equiv \rho _{\nu }\)
The integral recurrence relation for functions \(\rho _{\nu ,k}\) follows from the Parseval equality (1.10). To be precise, we obtain
An analog of the integral representation (2.11) for \(\rho _{\nu ,k}\) can be deduced in the following manner. In fact, the Mellin–Barnes integral for Laguerre polynomials (see [10, relation (8.4.33.3), Vol. III])
integral (2.1) with the Parseval identity (1.10), and the reflection formula for the gamma function imply the equality for \(k \in \mathbb {N}\)
where
Therefore we obtain the integral representation
Differentiating (4.5) n times by x, where the differentiation under the integral sign is possible due to the absolute and uniform convergence, we take into account the reduction formula for the gamma function and (2.1) to obtain
Consequently, after differentiating both sides of (4.6) n times we find an analog of the representation (2.29), namely,
Now, returning to (4.1), we substitute the function \(\rho _{\nu ,k}\) with the integral (4.4) and interchange the order of integration by Fubini’s theorem. Then, employing again the Viskov-type identities (2.6) for the differential operator \(\theta \), we derive for \(k \in \mathbb {N}\)
Hence it leads to
Theorem 4
Let \(k \in \mathbb {N},\ \nu \ge 0, \alpha > -1\). The orthogonality (4.1) for the sequence of polynomials \(\left( Q_n^{\nu ,\alpha }\right) _{n\in \mathbb {N}_0} \) with the weight \(x^\alpha \rho _{\nu ,k}(x)\) is the composition orthogonality of the same sequence with respect to the weight \( \rho _{\nu ,k-1} \), namely
In particular, for \(k=2\) this sequence is compositionally orthogonal in the sense of Prudnikov.
Further, up to a normalization constant equality, (4.1) is equivalent to the following n conditions
Hence the composition orthogonality (4.9) implies, with the integration by parts and properties of the operator \(\theta \),
Writing \(Q_n^{\nu ,\alpha }\) in the explicit form
we have
where
On the other hand, employing (4.4) for \(k\ge 2\) and observing that owing to the Viskov-type identities (2.6) \(( \theta _t\equiv tDt,\ \beta _y\equiv DyD)\)
we deduce, integrating by parts,
where the differentiation under integral sign is allowed via the absolute and uniform convergence. Thus, returning to (4.11), we plug in the latter expressions and change the order of integration by Fubini’s theorem to write it in the form
where
Meanwhile, recalling (2.1), we get
Therefore we find from (4.14)
where
But it is easily seen from the properties of the Mellin transform [15] and (2.1) that
Now, recalling (4.4), we have
Hence, modifying the formula (4.13), we obtain
Therefore, integrating by parts, we get from (4.18), (4.19), (4.20)
Moreover, in a similar manner as above we derive
So, substituting the right-hand side of the last equality in (4.22) into (4.21) and the obtained expression into (4.16), we find after the interchange of the order of integration and simple change of variables the following orthogonality conditions
where
and
Continuing this process by virtue of the same technique, involving the Mellin and Laplace transforms and the Mellin–Barnes integrals, after the kth step we end up with the equalities
where
and
On the other hand,
Consequently, the orthogonality (4.10) is equivalent to the following conditions
Moreover, we see that \( \left\{ x^m D^m \right\} ^k \left( x^{\nu +m} e^{-x} \right) = x^{\nu } e^{-x} p_{m(k+1)}(x),\) where \(p_{m(k+1)}\) is a polynomial of degree \(m(k+1)\) whose coefficients can be calculated explicitly via properties of the Pochhammer symbol and the Laguerre polynomials. Thus it can be reduced to the orthogonality with respect to the measure \(x^{\nu +\alpha } e^{-x} \mathrm{d}x\) and ideas of the previous section can be applied. We leave all details to the interested reader. Besides, further developments, an analog of Lemma 1 and relations with the multiple orthogonal polynomial ensemble from [7] will be a promising investigation.
References
Chihara, T.S.: An Introduction to Orthogonal Polynomials. Gordon and Breach, New York, London (1978)
Coussement, E., Van Assche, W.: Some properties of multiple orthogonal polynomials associated with Macdonald functions. J. Comput. Appl. Math. 133, 253–261 (2001)
Dattoli, G., Ricci, P.E., Khomasuridze, I.: Operational methods, special polynomial and functions and solution of partial differential equations. Integral Transf. Spec. Funct. 15(4), 309–321 (2004)
Ditkin, V.A., Prudnikov, A.P.: Integral transforms. Mathematical analysis, 1966, 7-82. Akad. Nauk SSSR Inst. Nauchn. Informacii, Moscow, 1967 (in Russian)
Erdélyi, A., Magnus, W., Oberhettinger, F., Tricomi, F.G.: Higher Transcendental Functions, Vols. I and II. McGraw-Hill, New York, London, Toronto (1953)
Ismail, M.E.H.: Bessel functions and the infinite divisibility of the student \(t\)-distribution. Ann. Prob. 5, 582–585 (1977)
Kuijlaars, A.B.J., Zhang, L.: Singular values of products of Ginibre random matrices, multiple orthogonal polynomials and hard eddge scaling limits. Commun. Math. Phys. 332, 759–781 (2014)
Loureiro, A.F., Yakubovich, S.: Central factorials under the Kontorovich–Lebedev transform of polynomials. Integral Transf. Spec. Funct. 24(3), 217–238 (2013)
Prudnikov, A.P.: Orthogonal polynomials with ultra-exponential weight functions, in: W. Van Assche (Ed.), Open Problems, J. Comput. Appl. Math. 48, 239-241, (1993)
Prudnikov, A.P., Brychkov, Yu.A., Marichev, O.I.: Vol. I: Elementary Functions, Gordon and Breach, New York, London, 1986; Vol. II: Special Functions, Gordon and Breach, New York, London, 1986; Vol. III: More Special Functions, Gordon and Breach, New York, London (1990)
Riordan, J.: Combinatorial Identities. Wiley, New York (1968)
Titchmarsh, E.C.: An Introduction to the Theory of Fourier Integrals. Clarendon Press, Oxford (1937)
Van Assche, W., Yakubovich, S.: Multiple orthogonal polynomials associated with Macdonald functions. Integral Transf. Spec. Funct. 9(3), 229–244 (2000)
Viskov, O.V., Srivastava, H.M.: New approaches to certain identities involving differential operators. J. Math. Anal. Appl. 186, 1–10 (1994)
Yakubovich, S., Luchko, Y.: The Hypergeometric Approach to Integral Transforms and Convolutions, Kluwer Academic Publishers. Math. Appl. 287, (1994)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Erik Koelink.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The work was partially supported by CMUP, which is financed by national funds through FCT (Portugal) under the project with reference UIDB/00144/2020. The author thanks Marco Martins Afonso for necessary numerical calculations and verifications of some formulas. Finally, the author is sincerely indebted to referees for useful comments and suggestions which rather improved the presentation of the paper.
Rights and permissions
About this article
Cite this article
Yakubovich, S. Orthogonal Polynomials with Ultra-Exponential Weight Functions: An Explicit Solution to the Ditkin–Prudnikov Problem. Constr Approx 53, 1–38 (2021). https://doi.org/10.1007/s00365-020-09523-0
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00365-020-09523-0
Keywords
- Orthogonal polynomials
- Modified Bessel functions
- Meijer G-function
- Mellin transform
- Associated Laguerre polynomials
- Multiple orthogonal polynomials