Singular rules for a multivariate quotient-difference algorithm (Q1316074)

\(\tau\) being a positive integer, let \({\mathcal I}\) be a collection of \(n+m+1\) distinct \(\tau\)-tuples \(i\equiv (i(1),\dots, i(\tau))\) of nonnegative integers. Let \({\mathcal N}, {\mathcal D}\) be subcollections of \({\mathcal I}\) containing \(n+1\) and \(m+1\) \(\tau\)-tuples respectively and satisfying the condition that if \(i\) is in \({\mathcal N}\), all \((j(1),\dots,j(\tau))\) for which \(0 \leq j(\mu) \leq i (\mu)\) \((1 \leq \mu \leq \tau)\) are present, \({\mathcal D}\) being similarly constrained. \({\mathfrak K}\) being a suitable field, let \(\Delta ({\mathcal I})\) be an interpolation basis \(x(i) \equiv (x(i(1)), \dots, x(i(\tau))) \in {\mathfrak K}^ \tau\), where \(i\) runs through all \(\tau\)-tuples in \({\mathcal I}\), and let \(f(z) \in {\mathfrak K}\) be defined for all \(z \equiv (z(1),\dots,z(\tau)) \equiv x(i)\) in \(\Delta ({\mathcal I})\). Taking as power functions \(\pi (i | z) = \prod (z(1)-x(j(1)))\dots \prod (z(\tau) - x(j(\tau)))\) \(\{0 \leq j(\mu) \leq i(\mu),\;1 \leq \mu \leq \tau\}\), set \(p(z) = \sum a(i) \pi (i | z)\) \(\{i \in {\mathcal N}\}\) and \(q(z)=\sum b(i) \pi (i | z)\) \(\{i \in {\mathcal D}\}\); set \(r({\mathcal D}, {\mathcal N}, {\mathcal I} | z) = p(z)/q(z)\). Subject to suitable conditions, the coefficients \(a(i)\), \(b(i)\) in \(r({\mathcal D}, {\mathcal N}, {\mathcal I})\), conveniently normalised, are determined by the constraints \(qf=p\) at \(z=x(i)\) for all \(i\) in \({\mathcal I}\). \(r({\mathcal D}, {\mathcal N}, {\mathcal I})\) then offers an interpolatory rational approximation to \(f\). Functions of the form \(r({\mathcal D}, {\mathcal N}, {\mathcal I})\) feature as successive links in a chain, in the first term of which either \({\mathcal D}\) or \({\mathcal N}\) is constrained to be the zero \(\tau\)-tuple \(0 \equiv (0,\dots,0)\); subsequently \({\mathcal D}''\), \({\mathcal N}''\), \({\mathcal I}''\) follows \({\mathcal D}'\), \({\mathcal N}'\), \({\mathcal I}'\) where (a) \({\mathcal I}''\) is \({\mathcal I}'\) augmented by a \(\tau\)-tuple not in \({\mathcal I}'\) and (b) either \({\mathcal D}''\) is \({\mathcal D}'\) augmented by the new member of \({\mathcal I}'\) and \({\mathcal N}''\) is \({\mathcal N}'\), or the converse relationship obtains, these possibilities holding in alternation. Subject to further conditions, the members \(r({\mathcal D}, {\mathcal N}, {\mathcal I})\) of such a chain may be exhibited as successive convergents of a continued fraction. Such chains may be concatenated, the first member of each being obtained from that of its predecessor by augmentation of the subcollection \({\mathcal D}\) or \({\mathcal N}\) not constrained to be 0. Subject to still further conditions, the coefficients in the continued fractions associated with each chain may be computed recursively by use of an algorithm similar to the \(q-d\) algorithm. In the presence of breakdown of the last mentioned existence conditions, interpolatory rational functions may be determined by use of a singular variant of the algorithm employed. This variant is the subject of the paper.