On the convergence of multivariate Padé approximants (Q2365230)

The paper gives a survey of convergence results for univariate and multivariate Padé approximants. No proofs are included. Let \([m/n]^f= q/p\) denote a univariate Padé approximant for \(f\). This means that \(p\) and \(q\) are polynomials with \(\deg q\leq m\), \(\deg p\leq n\) and \(\text{ord}(fp-q)\geq n+m+1\). Several convergence theorems about the convergence of the approximant to the function \(f\) do exist. The oldest one is the de Montessus-de Ballore theorem (1905) which gives the local uniform convergence as \(m\to\infty\) and \(n\) is fixed. Zinn-Justin (1971) gave a convergence in measure when both \(m\) and \(n\) go to \(\infty\), and \(m/n\to\infty\). From theorems of Karlsson and Wallin (1977) and of Nuttall (1970) and Pommerenke (1973), it follows that if the degree of the denominator is equal to the number of poles of \(f\) in a disk around the origin, then one can expect uniform convergence of the Padé approximants in that disk. If the degree is slightly too large, then one gets only convergence in measure. This kind of convergence theorems and their generalizations to the multivariate case are summarized in this paper. The generalizations that are obtainable depend on the definition of the multivariate approximant. In the bivariate homogeneous case, \(q\) and \(p\) are polynomials containing terms in \(x^iy^j\) where \(i+j = nm, \dots, nm+m\) and \(i+j = nm, \dots, nm+n\) resepectively. The residual \(fp-q\) contains only terms in \(i+j\geq nm+n+m+1\). Most of the convergence results are due to Cuyt and Lubinski. For other definitions of multivariate approximants, similar convergence results are formulated. Where appropriate, hints for convergence results in capacity are also indicated.