Stopping models closed under pgf composition, and the stability of randomly stopped model extensions

arXiv:2609.30106 · math.ST, math.PR, stat.TH · Submitted 2026-09-24 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: We're looking at the paper: "Stopping models closed under pgf composition, and the stability of randomly stopped model extensions".

Emmy: Here's the gist of the paper. Stopping models closed under pgf composition, and the stability of randomly stopped model extensions characterize statistical stability as requiring the underlying stopping model to be closed under pgf composition Characterization of Statistical Stability The statistical stability of randomly stopped sum and extreme transformations is fundamentally tied to the algebraic closure of the stopping model, meaning it requires that the underlying stopping model be closed under pgf composition.

Ada: The transformations based on randomly stopped sums and extremes are statistically stable model extensions if and only if they are an extension, which requires the identity variable to be in N, and N must be closed under pgf composition.

Emmy: Furthermore, for the composite transformations in Definition 4, closure under pgf composition is the sole necessary and sufficient condition to be a statistically stable model extension.

Ada: Structural Properties of Commuting Models Finite-dimensional stopping models closed under pgf composition, connected and with a non-empty interior must necessarily consist of a family of random variables that commute, which restricts the parameter space to be one-dimensional.

Paper summary: Ada: To recap where we are: we're talking about "Stopping models closed under pgf composition, and the stability of randomly stopped model extensions." The main thesis is that statistical stability for these transformations hinges entirely on the stopping model being closed under pgf composition.

Emmy: They characterize the complete set of stopping models for which randomly stopped sum and extreme model transformations function as statistically stable idempotent model extensions. This means they find the exact conditions needed for idempotency to happen.

Ada: And because they have this closure property, any finite-dimensional, connected stopping model closed under pgf composition is necessarily a family of random variables whose pgfs commute, which is a major structural finding.

Emmy: They use the Koenigs function to show that these models form a statistical manifold with a one-dimensional parameterization theta, where theta is defined as the probability that the stopping model equals one, and the probability mass at any integer i is always some polynomial in this parameter.

Kolya: I'm curious about why this matters beyond just proving stability. What does this structural understanding give us when we apply these random stopping transformations to real-world data models?

Ada: It gives us a clear framework for parameter identification and likelihood based inference because of that polynomial relationship with theta. The structure is simple enough for practical use in fitting data.

Emmy: The paper also establishes this duality between the stopping models that include the identity variable, which are the ones yielding stable extensions, and the set of probability distributions supported on positive integers. That's a powerful connection.

Kolya: So if you want to build a flexible lifetime model or an extreme value distribution using these methods, you need your initial setup to adhere to this closure rule for stability. It sets the ground rules for what kind of extensions are even possible in this framework.

Conclusion: Ada: So to wrap up this discussion on "Stopping models closed under pgf composition, and the stability of randomly stopped model extensions," the authors are showing that statistical stability isn't a property you can just assume for any stopping model transformation.

Emmy: No, it requires that the underlying stopping model has that closure under pgf composition. This forces the variables to commute, which simplifies everything down to a one-dimensional parameterization governed by that Koenigs function and its index m.

Ada: What this means for us is that we have a much tighter understanding of when these powerful transformations are actually going to produce reliable, stable statistical extensions. It’s not just about the transformation; it's about the foundational structure of the model itself.

Emmy: It connects abstract algebraic geometry—the commuting functions—to concrete probability theory on positive integers, which is a bridge that helps us identify and analyze these models more rigorously than before.

Kolya: So for those of us building these complex models, this paper tells us exactly what structural prerequisites we need to meet if we want the resulting model extension to be statistically sound in this context. It's about knowing when the algebraic machinery actually works as intended.

Jordi Valero, Josep Ginebra

Department of Statistics, Universitat Politecnica de Catalunya

math.ST, math.PR, stat.TH

Submitted: 2026-09-24

Updated: 2026-11-13

Comments: 45 pages

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 77/100

The gist: Stopping models closed under pgf composition, and the stability of randomly stopped model extensions characterize statistical stability as requiring the underlying stopping model to be closed under…

Key concepts

pgf composition
This is an algebraic operation on functions related to the stopping model. The paper establishes that for statistical stability of extensions, the stopping model must be closed under this composition rule. This closure is a fundamental requirement for the transformations to behave predictably in a statistical sense.
Statistical Stability
This refers to whether applying a transformation (like randomly stopped sums or extremes) results in an idempotent extension of the original statistical model. The paper proves that this stability is not dependent on the starting model but is determined entirely by the algebraic properties of the underlying stopping model.
Commuting Models
When a finite-dimensional stopping model is closed under pgf composition, it must be a family of random variables whose probability generating functions (pgfs) commute. This structural property severely restricts the parameter space, forcing it to be one-dimensional and related to distributions supported on positive integers.

Terminology

Summary

Stopping models closed under pgf composition, and the stability of randomly stopped model extensions characterize statistical stability as requiring the underlying stopping model to be closed under pgf composition

Characterization of Statistical Stability

The statistical stability of randomly stopped sum and extreme transformations is fundamentally tied to the algebraic closure of the stopping model, meaning it requires that the underlying stopping model be closed under pgf composition. The transformations based on randomly stopped sums and extremes are statistically stable model extensions if and only if they are an extension, which requires the identity variable to be in N, and N must be closed under pgf composition. p. 7 Furthermore, for the composite transformations in Definition 4, closure under pgf composition is the sole necessary and sufficient condition to be a statistically stable model extension. p. 7

Structural Properties of Commuting Models

Finite-dimensional stopping models closed under pgf composition, connected and with a non-empty interior must necessarily consist of a family of random variables that commute, which restricts the parameter space to be one-dimensional. These commuting families are intimately related to the set of probability distributions supported on the positive integers. p. 0 The structure of these models is characterized by a Koenigs function KN (t), which determines the model up to its parameter space limit, θN < 1. p. 0

Parameterization and Identification

The set of all non-constant analytic functions on [0, 1) that commute with hN0 is uniquely determined by the derivative at the origin, θ = g′(0), which serves as a global identifiable parameter. When 1 is in the support of N0, this derivative spans the entire interval (0, ∞), leading to a global one-dimensional parametrization for NN0. p. 0 The probability mass at i is a polynomial in θ of degree at most i. p. 0

Classification of Model Spaces

The maximal parameter space DN under θ of the associated stopping model N falls into four categories based on the properties of KN (t). These categories include DN = (0, 1] admitting the identity variable NI, DN = (0, θKN] for some θKN < 1, DN ⊂ N+ resulting in a discrete parameter space, or DN = [1], the degenerate case where N consists solely of NI. p. 2

Examples and Implications

Concrete examples show that the space of statistically stable model extensions is richer than previously conjectured by Marshall and Olkin (1997). Models in Section 6. p. 21, which include NI, serve as counter-examples to the conjecture by Mar24. p. 3 The paper concludes that statistical stability necessitates commutativity and closure under pgf composition. p. 7 The analysis is restricted to stopping models supported on the positive integers, as zero in the support leads to model contractions rather than extensions. p. 2 The index m of the Koenigs function must satisfy m ∈ [−1, 0] for models admitting a maximal parameter space (0, 1]. p. 0

Appendix Summary

The appendix provides proofs for propositions regarding the closure of NN0 under composition, the properties of commuting families, and the characterization of stopping models that admit the parameter space (0, 1] including NI. The proof for Proposition 5 establishes a bijective correspondence between these models and probability distributions supported on the positive integers. p. 2 The proof for Proposition 12 shows that NN0 has an empty interior if N0 has finite support. p. 2 The proof of Theorem 4 demonstrates that the presence of non-commuting random variables forces the model to be infinite-dimensional. p. 2 The analysis confirms that every element in HN must commute with hN0, implying N ⊂ NN0. p. 8 The proof of Proposition 17 establishes that the index m must satisfy m = lim t→1− log KN (t) / log(1 − t) ∈ [−1, 0] for models admitting the maximal parameter space. p. 2 The final result links the geometric model to the case where m = −1. p. 0 The paper notes that zero in the support leads to model contractions, which are excluded from this study. p. 2

Bibliography

Cowen, C.C. (1984). Commuting analytic functions

Feller, W. (1943). On a general class of “contagious” distributions

Gurland, J. (1957). Some interrelations among compound and generalized distributions

Gurland, J. (1958). A general class of contagious distributions

Johnson, N.L., Kemp, A.W., Kotz, S. (2005). Univariate Discrete Distributions

Karlin, S., McGregor, J. (1968a). Embeddability of discrete time simple branching processes into continuous time branching processes

Karlin, S., McGregor, J. (1968b). Embedding iterates of analytic functions with two fixed points into continuous groups

Kuczma, M. (1968). Functional equations in a single variable

Marshall, A.W., Olkin, I. p. 0 (1997). A new method for adding a parameter to a family of distributions with application to the exponential and Weibull families

Neyman, J. (1939). On a new class of “contagious” distributions applicable in entomology and bacteriology

Pranger, W. (1970). Iterations of functions analytic on a disk

Shaked, M. (1975). On the distribution of the minimum and of the maximum of a random number of i. p. 5i.d. random variables

Shaked, M., Wong, T. (1997). Stochastic comparisons of random minima and maxima

Valero, J., Ginebra, J. p. 0 (2025). On statistical model extensions based on randomly stopped extremes

https://doi.org/10.57645/20.8080.02.22

--- Page 1 ---

Stopping models closed under pgf composition, and the stability of randomly stopped model extensions Jordi Valero, Josep Ginebra1 Statistical model transformations based on randomly stopped sums, maxima and minima are widely used to extend statistical models We characterize the complete set of stopping models for which randomly stopped sum and extreme model transformations function as statistically stable (idempotent) model extensions Stability requires the underlying stopping model to be closed under pgf composition We prove that any finite-dimensional, connected stopping model closed under pgf composition is necessarily a family of random variables whose pgfs commute Using the corresponding Koenigs function, we establish that these models form a statistical manifold admitting a global, one-dimensional parametrization θ = Pr(N = 1) ∈ (0, θ∗], where the probability mass at i is a polynomial in θ of degree at most i Finally, we establish a duality between stopping models closed and containing the identity variable (the ones yielding stable extensions) and the set of probability distributions supported on the positive integers

--- Page 2 ---

The transformation of parametric statistical models through randomly stopped sums, introduced by Neyman (1939) and Feller (1943), is a cornerstone of mathematical statistics and stochastic modeling The compound models obtained this way include some of the most versatile models in applied statistics, and are used across disciplines like engineering, natural science, actuarial science, mathematical finance, human-computer interaction and queueing theory Similarly, transformations based on randomly stopped maxima or minima have become essential for constructing flexible lifetime and heavytailed models used in reliability engineering, extreme value analysis, insurance, and epidemiology The existing literature extensively details the distributional properties of specific models resulting from a single application of these transformations using a given stopping and stopped model Conversely, the structural consequences of the repeated application of these transformations remain less understood

--- Page 3 ---

The central challenge tackled here is the characterization of the conditions under which these statistical model transformations, T (·), work as statistically stable extensions. Formally, given an initial parametric statistical model, X, we seek the requirements ensuring that T acts as an extension, in the sense that X ⊂ T (X) for all X, and the conditions under which this extension is statistically stable, in the sense that T is idempotent. p. 1 p. 1 We establish that the statistical stability of these model transformations is not a property of initial models X, but it instead requires that the underlying stopping model N be closed under pgf composition. p. 2 For the model transformation to function as an extension, the stopping model N must include the degenerate random variable at one NI, which we call the identity variable. p. 2

--- Page 4 ---

Our primary contribution, Theorem 4, establishes that any finite-dimensional stopping model closed under pgf composition, connected and with a non-empty interior is necessarily a family of random variables that commute. Consequently, stopping models closed under pgf composition inherit the properties of these commuting families. p. 2 In particular, by leveraging the Schr¨oder functional equation and the corresponding Koenigs function, we are able to treat these models as differentiable statistical manifolds such that they admit a global identifiable one-dimensional parametrization through θ = Pr(N = 1) ∈ (0, θ∗], and for any i > 0, the probability mass at i is a polynomial in θ of degree at most i. p. 2

--- Page 5 ---

The transformation based on randomly stopped sums and extremes are statistical model extensions if and only if NI ∈ N. The composite transformations in Definition 4 always function as statistical model extensions regardless of the stopping model. p. 6

--- Page 7 ---

The following result establishes that for the transformations considered here, statistical stability is tied to the algebraic closure of the stopping model. The first part follows directly from the definitions, while the second part relies on results in Section 4. p. 7 p. 7

Pruning redundant reasoning tokens

The following result establishes that for the transformations considered here, statistical stability is tied to the algebraic closure of the stopping model. The first part follows directly from the definitions, while the second part relies on results in Section 4.

How the training signal is built

The central challenge tackled here is the characterization of the conditions under which these statistical model transformations, T (·), work as statistically stable extensions. Formally, given an initial parametric statistical model, X, we seek the requirements ensuring that T acts as an extension, in the sense that X ⊂ T (X) for all X, and the conditions under which this extension is statistically stable, in the sense that T is idempotent.

Where the gains come from

The statistical stability of these model transformations is not a property of initial models X, but it instead requires that the underlying stopping model N be closed under pgf composition. For the model transformation to function as an extension, the stopping model N must include the degenerate random variable at one NI. p. 2

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Model extension stability analysis: The AI can determine if a given statistical model transformation acts as a statistically stable extension by checking if its underlying stopping model is closed under pgf composition, as stated in Proposition 2.

  2. Model characterization for stable extensions: The system can characterize the conditions under which transformations work as extensions by verifying that the stopping model N includes the identity variable NI, which is required for an extension to function.

  3. Commutativity verification: The AI can identify if a finite-dimensional, connected stopping model is necessarily a family of random variables whose pgfs commute, based on Theorem 4.

  4. Parameter space determination: For models closed under pgf composition and connected with a non-empty interior, the system can determine the global one-dimensional parametrization theta = Pr(N = 1) in (0, theta∗].

  5. Distribution identification: The AI can use the Koenigs function KN to characterize stopping models up to their parameter space limit, allowing for unique identification of these models.

  6. Subclass discrimination: The system can distinguish between different classes of stopping models by analyzing the properties of their Koenigs functions, such as whether they admit the maximal parameter space (0, 1].

Related papers