Many dynamical models of the Milky Way halo require assumptions that the distribution function of a tracer population should be independent of time (i.e., a steady state distribution function) and that the underlying potential is spherical. We study the limitations of such modelling by applying a general dynamical model with minimal assumptions to a large sample of galactic haloes from cosmological $N$-body and hydrodynamical simulations. Using dark matter particles as dynamical tracers, we find that the systematic uncertainties in the measured mass and concentration parameters typically have an amplitude of 25% to 40%. When stars are used as tracers, however, the systematic uncertainties can be as large as a factor of $2-3$. The systematic uncertainties are not reduced by increasing the tracer sample size and vary stochastically from halo to halo. These systematic uncertainties are mostly driven by underestimated statistical noise caused by correlated phase-space structures that violate the steady state assumption. The number of independent phase-space structures inferred from the uncertainty level sets a limiting sample size beyond which a further increase no longer significantly improves the accuracy of dynamical inferences. The systematic uncertainty level is determined by the halo merger history, the shape and environment of the halo. Our conclusions apply generally to any spherical steady-state model.
Download