Biasing with an independent increment: Gaussian approximations and proximity of Poisson mixtures
Fraser Daly · 2025
Read Biasing with an independent increment: Gaussian approximations and proximity of Poisson mixtures by Fraser Daly (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-05586.txt.
Introduction
Fraser Daly proved three sets of math results in a 2025 paper on how close one spread of chance values sits to another. A known trick shifts a chance value by adding a fresh step of like size, which mimics a bias by size or by a zero point. Daly joins that trick with a classic proof method for near-normal sums.
The first result bounds how far blended Poisson spreads sit from each other. The second gives error bounds for sums of linked values, with no cap needed on their size. The third covers sums whose third moment fades away. Each bound is stated in exact terms, so users can judge how much trust a shortcut earns.
Catalog record
| CREATOR | Fraser Daly |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-05586
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Biasing with an independent increment: Gaussian approximations and proximity of Poisson mixtures Fraser Daly Abstract Abstract By exploiting the well-known observation that size-biasing or zero-biasing an infinitely divisible random variable may be achieved by adding an independent increment, combined with tools from Stein’s method for compound Poisson and Gaussian approximations, we establish three sets of approximation results: (a) bounds on the proximity of Poisson mixtures with infinitely divisible mixing distributions, (b) central limit theorems with explicit error bounds for sums of associated or negatively associated random variables which do not require boundedness of the underlying distributions, and (c) a Gaussian approximation theorem under a vanishing third moment condition. These exploit biasing by an independent increment directly, via an intermediate compound Poisson approximation, and through a convex ordering argument, respectively. Applications include a Dickman-type limit theorem, simple random sampling and urn models with overflow. Key words and phrases: size-biasing; zero-biasing; compound Poisson; infinite divisibility; central limit theorem; (negative) association; Stein’s method MSC 2020 subject classification: 60F05; 60E07; 60E15; 62E17 1 Introduction There are well-known connections between infinite divisibility and several of the coupling techniques that are employed frequently in conjunction with Stein’s method for probability approximation, particularly in Poisson, compound Poisson and Gaussian approximation results. Two such couplings are the size- and zero-biased couplings; see ( 1 ) and ( 6 ) below for definitions of these concepts. In Gaussian approximation results, the proximity of a real-valued random variable W to Gaussian can be measured by the proximity of W and its zero-biased version: [ 14 ] gives an overview and many applications of this approach. Similarly, explicit error bounds in Poisson and compound Poisson approximations can be derived in many applications via construction of the size-biased version of the underlying random variable. See [ 5 ] for a detailed discussion of the Poisson case, and [ 4 ] for an introduction to Stein’s method for compound Poisson approximation and [ 16 ] for a discussion of connections with size-biasing. We will also make use of non-zero-biasing, defined in ( 7 ) below, an extension of zero-biasing beyond the case of mean-zero random variables. It is well-known that either size-biasing or non-zero-biasing a compound Poisson random variable can be achieved by adding an independent increment (see [ 32 ] and [ 18 ] , respectively), and that a mean-zero, infinitely divisible random variable may be zero-biased by adding an independent increment [ 25 ] . While these observations on biasing with an independent increment are familiar, they remain under-exploited in applications of Stein’s method. Our overall aim in this note is to demonstrate how these observations can be employed in conjunction with Stein’s method through three applications. Although these applications are on the surface rather different, they each employ similar constructions and illustrate a way of exploiting biasing by the addition of an independent increment. The three applications we will consider are the following: (a) Bounds on the proximity of two mixed Poisson distributions with infinitely divisible mixing distributions, which will be applied to give total variation error bounds in the approximation of certain Bernoulli-weighted sums of independent Poisson random variables by a mixed Poisson with Dickman mixing distribution. Here we use Stein’s method for compound Poisson approximation. (b) Explicit error bounds in Gaussian approximation for sums of associated or negatively associated random variables without the boundedness assumptions required by earlier results, such as the univariate approximation results established by Goldstein and Wiroonsri [ 24 ] and Wiroonsri [ 33 ] . We give applications to simple random sampling and urn models with overflow. This part of our work uses both compound Poisson and Gaussian approximation results from Stein’s method. (c) A Gaussian approximation bound for a random variable whose first three moments match those of the approximating Gaussian distribution, illustrated by a straightforward application to Student’s t distribution. This uses Stein’s method for Gaussian approximation. A natural way to exploit ideas of biasing with an independent increment is in the compound Poisson approximation of another compound Poisson distribution (where, as we will see below, proximity can be measured by the distance between the increments obtained when size-biasing each) or in the Gaussian approximation of a mean-zero random variable (where proximity can be measured by the size of the increment obtained when zero-biasing). This is the idea underlying both (a) and (c) above. In (a) we make use of the fact that a mixed Poisson random variable with an infinitely divisible mixing distribution is compound Poisson. In (c) we use this zero-biasing approach to establish a convex ordering that yields a Gaussian approximation bound for infinitely divisible random variables whose first three moments match those of the approximating Gaussian distribution. Our approach to exploiting biasing with an independent increment in (b) above is less direct. Here we use the fact that our random variable W of interest has a distribution close to compound Poisson; we use results from [ 16 ] to bound the distance of W from a suitably chosen compound Poisson distribution, and then use an approach based on non-zero-biasing with an independent increment to quantify the distance of this compound Poisson distribution from Gaussian. As we will see, compared to a more direct approach in this setting, our method bypasses some technical difficulties associated with compound Poisson approximation via Stein’s method. The remainder of the paper is structured as follows. We will use Section 2 to give brief introductions to Stein’s method for compound Poisson approximation and the notions of size- and zero-biasing in order to provide the necessary background for the remainder of the paper. In Section 3 we establish bounds on the proximity of mixed Poisson distributions with infinitely divisible mixing distributions. Central limit theorems with explicit error bounds for sums of associated or negatively associated random variables are derived in Section 4 . Finally, in Section 5 we present a Gaussian approximation result under the assumption of three matching moments, i.e., a vanishing third moment assumption in the standardised case. 2 Technical preliminaries: Stein’s method for compound Poisson approximation In this section we give an introduction to size- and zero-biasing, and to Stein’s method for compound Poisson approximation. These techniques will be employed in each of the three application areas we discuss in Sections 3 – 5 . In the work that follows we will also need to use Stein’s method for Gaussian approximation; this will be introduced as needed in Sections 4 and 5 below. Given a non-negative random variable X with finite, non-zero mean, its size-biased version, which we denote here by X^{\text{s}} , is defined by \mathbb{E}f(X^{\text{s}})=\frac{\mathbb{E}[Xf(X)]}{\mathbb{E}X}\,, (1) for any function f:\mathbb{R}^{+}\to\mathbb{R} for which the expectation on the right-hand side exists. We recall that X (or, equivalently, its distribution) is said to be infinitely divisible if, for each n=1,2,\ldots , there are independent and identically distributed (i.i.d.) random variables X_{1}^{(n)},\ldots,X_{n}^{(n)} such that X is equal in distribution to X_{1}^{(n)}+\cdots+X_{n}^{(n)} . The observation that X is infinitely divisible if and only if X^{\text{s}} is equal in distribution to X+Y for some random variable Y which is independent of X is due to Steutel [ 32 ] , though he neither uses the language of size-biasing nor requires that X have finite mean. We