On the optimality of Stein factors

Adrian Röllin

Home/Archive/arXiv

arXiv · 0706.0879

On the optimality of Stein factors

Adrian Röllin

View original source ↗Read source document ↗Export metadata ↓

Abstract

The application of Stein's method for distributional approximation often involves so called Stein factors (also called 'magic factors') in the bound of the solutions to Stein equations. However, in some cases these factors contain additional (undesirable) logarithmic terms. It has been shown for many Stein factors that the known bounds are sharp and thus that these additional logarithmic terms cannot be avoided in general. However, no probabilistic examples have appeared in the literature that would show that these terms in the Stein factors are not just unavoidable artefacts, but that they are there for a good reason. In this article we close this gap by constructing such examples. This also leads to a new interpretation of the solutions to Stein equations.

Explore related subjects

math.PR

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Adrian Röllin. 2013-06-12. On the optimality of Stein factors. https://arxiv.org/abs/0706.0879

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Dark Subspaces and Invariant Measures of Quantum Trajectories

Quantum trajectories are Markov processes describing the evolution of a quantum system subject to indirect measurements. They can be viewed as place dependent iterated function systems or the result of products of dependent and non identically distributed random matrices. In this article, we establish a complete classification of their invariant measures. The classification is done in two steps. First, we prove a Markov process on some linear subspaces called dark subspaces, defined in (Maassen, Kümmerer 2006), admits a unique invariant measure. Second, we study the process inside the dark subspaces. Using a notion of minimal family of isometries from a reference space to dark subspaces, we prove a set of measures indexed by orbits of a unitary group is the set of ergodic measures of quantum trajectories.

math.PR↗

arXiv2026

Expected Signature Kernels for Rough Paths

The expected signature kernel is a similarity measure of probability laws on path space and enters directly into signature maximum mean discrepancies. Building on [P. K. Friz, P. P. Hager, and N. Tapia, Forum Math. Sigma, 10 (2022), e42] and [M. Lemercier, T. Lyons, and C. Salvi, SIAM J. Numer. Anal., 64 (2026), pp. 1095-1125], we characterize the expected signature kernel of group-valued inhomogeneous Lévy processes, aka Lévy rough paths, as the scalar component of the solution to an extended Goursat-type system. We give an original description, first in the continuous case where the finite-dimensional system is exact. In the presence of jumps, this system is inherently infinite-dimensional; we prove its well-posedness and derive finite-dimensional truncations with explicit error control. This permits direct evaluation of the reference-dependent terms in one-sample signature-MMD tests without an auxiliary reference sample. Finally, for conditionally Gaussian martingales, conditioning the usual unbiased empirical squared signature-MMD on the sampled quadratic variations yields an unbiased statistic for the same quantity with no larger variance. We derive computable finite-grid versions and illustrate the resulting variance reduction and gain in power.

math.PR↗

arXiv2026

A central limit theorem for the range of random walk indexed by an infinite tree

We are interested in the second-order fluctuations of the size of the range of a random walk indexed by an infinite tree in ${\mathbb Z}^d$. Consider the Kesten tree with geometric offspring, we view the vertices on the left of its spine (the unique infinite path) as the past, and those on the right as the future. We show that the number of new sites in the future given its past has linear variance when $d>8$, and satisfies a central limit theorem (CLT) with Gaussian limiting distribution when $d>16$. The proof relies on the stationarity of the model under depth-first exploration, a general CLT by Dedecker and Merlevède [7], a truncation technique exploiting the local independence of tree structures, and a recursion argument for moment bounds.

math.PR↗