WSEAS Transactions on Systems and Control
Print ISSN: 1991-8763, E-ISSN: 2224-2856
Volume 21, 2026
Padé Approximants via Matrix Elementary Row Operations: Algorithm, Error Bounds, and Optimality
Authors: ,
Search Articles
Abstract: We develop a matrix elementary row operations (ERO) framework for computing Padé approximants and analyzing their accuracy, extending an integer augmented-matrix device to the whole Padé table. The $$[m/n]$$ approximant of a formal power series is obtained by row-reducing an augmented Padé–Sylvester matrix, and a determinant criterion certifies nondegeneracy. For the exponential learning function $$L(t) = 1 - e^{-at}$$, a transfer lemma proves that the lower triangle $$m \geq n$$ of the Padé table is nondegenerate, and an exact Beta-integral remainder gives a uniform error bound whose diagonal constant $$K_{n,n} = 1/\binom{2n}{n}$$ is independent of $$a$$. The Padé-to-Taylor error ratio has an exact quotient-of-integrals representation; it decreases strictly, stays below $$K_{n,n}$$, and behaves like $$(2n)!/(at)^{2n}$$ as $$t \to \infty$$. We also give a shape analysis of the Padé curve, an order-selection theorem, and a numerical stability study. For optimal time allocation, the Padé surrogate recovers the exact optimum within a certified bound.
Keywords:
Padé approximants, matrix elementary operations, continued fractions, exponential learning functions, error bounds, Sylvester matrix, Taylor polynomial, numerical stability
Pages: 523-532
DOI: 10.37394/23203.2026.21.45