Quantifying biases in orbit-superposition modelling with (non-)parametric kinematics

Department of Astrophysics, University of Vienna, Türkenschanzstrasse 17, A-1180 Vienna, Austria

^⋆ Corresponding author: stefanie.reiter@univie.ac.at

Received: 6 March 2025
Accepted: 12 July 2025

Abstract

Context. Dynamical models enable the recovery of galaxy mass profiles, intrinsic shapes, and, in the case of Schwarzschild orbit-superposition methods, orbital structure. An accurate dynamical inference relies on the precise recovery of the line-of-sight velocity distribution from an integrated spectrum.

Aims. The main goal of this work is to evaluate the effect of different methods for recovering stellar kinematics from an integrated spectrum on the resulting orbit distribution and subsequent implications on the recovered mass and velocity anisotropy profiles.

Methods. We applied the commonly used penalised pixel fitting method pPXF on archival SAURON datacubes of NGC 4550 and NGC 2768, a well-studied counter-rotating and a regular-rotating elliptical galaxy, for a parametrised description of the stellar line-of-sight velocity distribution based on Gauss Hermite expansion. We also used Bayes-LOSVD, a method based on Bayesian framework, to recover a non-parametric representation of the stellar kinematics. Both datasets are used as input for the orbit-superposition code DYNAMITE.

Results. We find that the inferred dynamical properties of NGC 4550 are strongly affected by the difference in the method used to derive the stellar kinematics. Using non-parametric kinematics, we recover two distinct, dynamically cold counter-rotating disks; the orbit solution resulting from the Gauss Hermite stellar kinematics indicates dynamically warmer disks. In addition, the parametrised approach results in notably higher predicted total mass of the galaxy and more radial velocity anisotropy. For NGC 2768, these differences are less significant, but still noticeable, especially in the orbit distribution. We also find that the non-parametric kinematics results are more robust against changes to spatial binning and in the uncertainty computation.

Conclusions. We argue for the advantages of using a non-parametric description of the stellar kinematics for the dynamical modelling of galaxies.