Title: High-redshift cosmography: auxiliary variables versus Padé polynomials

Abstract: Cosmography becomes non-predictive when cosmic data span beyond the red shift limit $z\simeq1 $. This leads to a \emph{strong convergence issue} that jeopardizes its viability. In this work, we critically compare the two main solutions of the convergence problem, i.e. the $y$-parametrizations of the redshift and the alternatives to Taylor expansions based on Pad\'e series. In particular, among several possibilities, we consider two widely adopted parametrizations, namely $y_1=1-a$ and $y_2=\arctan(a^{-1}-1)$, being $a$ the scale factor of the Universe. We find that the $y_2$-parametrization performs relatively better than the $y_1$-parametrization over the whole redshift domain. Even though $y_2$ overcomes the issues of $y_1$, we get that the most viable approximations of the luminosity distance $d_L(z)$ are given in terms of Pad\'e approximations. In order to check this result by means of cosmic data, we analyze the Pad\'e approximations up to the fifth order, and compare these series with the corresponding $y$-variables of the same orders. We investigate two distinct domains involving Monte Carlo analysis on the Pantheon Superovae Ia data, $H(z)$ and shift parameter measurements. We conclude that the (2,1) Pad\'e approximation is statistically the optimal approach to explain low and high-redshift data, together with the fifth-order $y_2$-parametrization. At high redshifts, the (3,2) Pad\'e approximation cannot be fully excluded, while the (2,2) Pad\'e one is essentially ruled out.