Cayley's theorem and its application in the theory of vector Padé approximants (Q1919375)

Cayleys's theorem (1857) says that \(\text{det} A=P fR \cdot P fC\) where \(A\) is a matrix of even dimension which is skew-symmetric after delection of its \(r\)th row and column. \(R\) is the matrix obtained from \(A\) by replacing row \(r\) by the transpose of column \(r\) and \(C\) is obtained similarly by replacing column \(r\) in \(A\) by the transpose of row \(r\). \(Pf M\) is the Pfaffian of the matrix \(M\). This is a scalar value that can be associated with a matrix of even dimension that has a construction somewhat similar to a determinant. A first contribution of this paper is a modern proof of this old theorem. The main interest of the authors lies in vector Padé approximation. If \(f(z)= \sum_{i\geq 0} c_iz^i\) is a vector series in \(C^d \), then \(p/q\) is a \([n/2k]\) vector Padé approximant for \(f\) if \(p \in C^d [z]\), \(q \in C[z]\) with \(\deg p \leq n\), \(\deg q= 2k\) and \(fq-p\) has an order at least \(n+1\). As in the scalar case, the denominator can be written as a determinant and the form of the matrix allows the application of Cayley's theorem. The Pfaffians can then be expressed as multiple integrals which is a consequence of the fact that the coefficients \(c_i\) can be expressed as integrals by the Taylor theorem. It turns out that \(q(z)= q^{P f} (0) q^{P f} (z)\). Inspired by the previous results, a new vector inequality is derived. It implies that \(q^{P f} (x) > 0\) for all \(x \in R\). It can be proved for \(k= 0,1,2\). It is then conjectured that it holds for all \(k\).