On a family of multivariate two-point rational approximants (Q1192514)

From MaRDI portal
scientific article
Language Label Description Also known as
English
On a family of multivariate two-point rational approximants
scientific article

    Statements

    On a family of multivariate two-point rational approximants (English)
    0 references
    0 references
    0 references
    27 September 1992
    0 references
    The usefulness of Padé approximation in a great number of applications has tempted many authors to extending this topic to functions of several variables. So, the multivariate Padé problem can be stated as follows. Given a function \(f\) holomorphic on a neighbourhood of the origin \(U_ 0\) and \(m(m_ 1,\dots,m_ n)\), \(\nu(\nu_ 1,\dots,\nu_ n)\in\mathbb{N}^ n\), the question consists of finding a rational function. (1.1) \(R_{m,\nu}(z)=Q_ m(z)/P_ \nu(z)\), such that \(f(z)- R_{m,\nu}(z)=\sum_{\lambda\in\mathbb{N}^ n}e_ \lambda z^ \lambda\), with \(e_ \lambda=0\), for any \(\lambda\) belonging to a certain lattice \(S\subset\mathbb{N}^ n\). Since the number of parameters to determine is (1.2) \(N_ p=\prod^ n_{j=1}(m_ j+1)+\prod^ n_{j=1}(\nu_ j+1)\), we need \(N_ e=N_ p-1\) linear homogeneous equations and, hence, the lattice \(S\) must have \(N_ p-1\) points. Definition and construction: Let \(f(z_ 1,z_ 2)\) be a function holomorphic to \(U_ 0\) and \(U_ \infty\), neighbourhoods of the origin (0,0) and infinity \((\infty,\infty)\), respectively. Let us suppose that \(f\) admits the following expansions: \[ f_ 0(z_ 1,z_ 2)=\sum^ \infty_{i=0}\sum^ \infty_{j=0}c_{ij}z^ i_ 1z^ j_ 2\quad\text{for } (z_ 1,z_ 2)\in U_ 0 \tag{1.3} \] \[ f_ \infty(z_ 1,z_ 2)=\sum^ \infty_{i=1}\sum^ \infty_{j=1}c^*_{ij}z_ 1^{-i}z_ 2^{-j}, \quad\text{for }(z_ 1,z_ 2)\in U_ \infty. \tag{1.4} \] Thus, given a polynomial \(\overline P_ m\) of degrees \(m_ 1,m_ 2\) in \(z_ 1,z_ 2\) respectively, and two pairs of nonnegative integers \((k_ 1,k_ 2)\) and \((h_ 1,h_ 2)\), such that \(k_ 1+h_ 1=m_ 1\) and \(k_ 2+h_ 2=m_ 2\) we want to find a polynomial \(\overline Q_{m-1}\) of degrees \((m_ 1-1,m_ 2-2)\), satisfying: \[ (f_ 0-(\overline Q/\overline p))(z_ 1,z_ 2)=O_ +(z_ 1^{k_ 1}\circ z^ i_ 2)+O_ +(z_ 1^ i\circ z_ 2^{k_ 2})\;(i\geq 0), \tag{1.5} \] \[ (f_ \infty-(\overline Q/\overline p))(z_ 1,z_ 2)=O_ -(z_ 1^{-h_ 1-1}\circ z_ 2^{-i})+O_ -(z_ 1^{- i}\circ z_ 2^{-h_ 2-1})\;(i\geq 1). \tag{1.6} \] The notation appearing in (1.5) means that \(\overline Q/\overline p\) is a ratonal function whose series expansion in ascending powers of \(z_ 1\) and \(z_ 2\) coincides with \(f\) up to the terms in \(z_ 1^{k_ 1-1}\circ z_ 2^{k_ 2-1}\). The same holds for the notation in (1.6), but now replacing ascending powers by descending powers. The authors define the `` bivariate Laurent polynomials'' (\(L^ 2\)-polynomials), as follows: Definition 1. Let \(m_ 1,n_ 1,m_ 2,n_ 2\) be integers, with \(m_ i\leq n_ i(i=1,2)\) and \(m(m_ 1,m_ 2)\) and \(n(n_ 1,n_ 2)\). The bivariate polynomial \[ P_{(m,n)}(z_ 1,z_ 2)=\sum^{n_ 1}_{i=m_ 1}\sum^{n_ 2}_{j=m_ 2}\lambda_{ij}z^ i_ 1z^ j_ 2 \tag{1.7} \] is said to be a \(L^ 2\)-polynomial of degree \((m,n)\). Denote by \(P(m,n)\) the space of the \(L^ 2\)-polynomials of degree \((p,q)\) with \(m\leq p\) (i.e., \(m_ i\leq p_ i\), \(i=1,2)\) and \(q\leq n\), and by \(P_{(-\infty,+\infty)}\) the union of all the \(P_{(m,n)}\), with \(m\leq n\). We have: \[ Q_{m-1}(z_ 1,z_ 2)=\sum^{h_ 1- 1}_{r=0}\sum^{h_ 2-1}_{s=0}\left(\sum^{h_ 1-1}_{i=0} \sum^{h_ 2-1}_{j=0}b_{ij}C^*_{r-1+1,s-j+1}\right)z_ 1^ rz_ 2^ s+ \tag{1.8} \] \[ +\sum^{m_ 1-1}_{r=h_ 1}\sum^{m_ 2-1}_{s=h_ 2}\left(\sum^{m_ 1}_{i=h_ 1+1}\sum^{m_ 2} _{j=h_ 2+1}b_{ij}C_{i-r-1,j-s-1}\right)z^ r_ 1z^ s_ 2. \] Thus, \(Q_{m-1}\) is a polynomial of degrees \(m_ 1-1\), \(m_ 2-1\) in \(z_ 1,z_ 2\) respectively. Now, consider the ``reciprocal polynomials'': \[ \overline Q_{m-1}(z_ 1,z_ 2)=z_ 1^{m_ 1- 1}z_ 2^{m_ 2-1}Q_{m-1}(z_ 1^{-1},z_ 2^{-1}, \overline P_ m(z_ 1,z_ 2)=z_ 1^{m_ 1}z_ 2^{m_ 2}P_ m(z_ 1^{-1},z_ 2^{-1}). \tag{1.9} \] From (1.8) and (1.9), we can write: \[ \overline P_ m(z_ 1,z_ 2)=\sum^{m_ 1}_{i=0}\sum^{m_ 2}_{j=0}b_{ij}z_ 1^{m_ 1-i}z_ 2^{m_ 2-j}, \tag{1.10} \] \[ \overline Q_{m-1}(z_ 1,z_ 2)=\sum^{k_ 1- 1}_{r=0}\sum^{k_ 2-1}_{s=0}\left(\sum^{m_ 1}_{i=h_ 1+1} \sum^{m_ 2}_{j=h_ 2+1}b_{ij}C_{i+r-m_ 1,j+s-m_ 2}\right)z^ r_ 1z_ 2^ s+ \tag{1.11} \] \[ +\sum^{m_ 1- 1}_{r=k_ 1}\sum^{m_ 2-1}_{s=k_ 2}\left(\sum^{h_ 1- 1}_{i=0}\sum^{h_ 2-1} _{j=0}b_{ij}C^*_{m_ 1-r-i,m_ 2-s- j}\right)z_ 1^ rz_ 2^ s. \] The authors prove the following theorem: Theorem 1: If \(\overline P_ m\) and \(\overline Q_{m-1}\) are given by (1.10) and (1.11), the rational function \(\overline Q_{m- 1}/\overline P_ m\) verifies the conditions (1.5) and (1.6). In this paper the authors extend the concept of two-point Padé-type approximants to functions of two variable, the rectangular way, by means of certain linear functionals acting on the space of bivariate Laurent polynomials, and define the two point Padé-approximants as a particular case, by increasing the order of approximation. This paper contains useful remarks and examples.
    0 references
    0 references
    \(L^ 2\)-polynomials
    0 references
    reciprocal polynomials
    0 references
    Padé approximation
    0 references
    multivariate Padé problem
    0 references
    bivariate Laurent polynomials
    0 references
    two point Padé-approximants
    0 references
    examples
    0 references
    0 references