Stochastic Ordering of Dependent Systems under Transformation Models and Archimedean Copulas
Idir Arab, Milto Hadjikyriakou, Paulo Eduardo Oliveira · 2026
Read Stochastic Ordering of Dependent Systems under Transformation Models and Archimedean Copulas by Idir Arab, Milto Hadjikyriakou, Paulo Eduardo Oliveira (2026) 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-2604-26026.txt.
Introduction
Arab, Hadjikyriakou, and Oliveira compare life spans of systems whose parts are dependent and varied, with each part's distribution built by transforming a shared base. They model dependence with Archimedean copulas, a family that lets them treat hazard, reversed-hazard, and odds models in one framework. For parallel, series, and k-out-of-n style systems, they give conditions for one system to dominate another in a stochastic sense. The conditions rest on monotonicity of the transformation and on traits of the copula generators.
Catalog record
| CREATOR | Idir Arab, Milto Hadjikyriakou, Paulo Eduardo Oliveira |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69911 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.26026
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Stochastic Ordering of Dependent Systems under Transformation Models and Archimedean Copulas Idir Arab, Milto Hadjikyriakou, Paulo Eduardo Oliveira Abstract We study stochastic ordering of system lifetimes with dependent and heterogeneous components whose marginal distributions are obtained through transformations of a common baseline. The dependence structure is modeled via Archimedean copulas, allowing for a unified treatment of several transformation-based models, including proportional hazard, proportional reversed hazard rate and proportional odds families. For parallel, series and ( n − k ) (n-k) -out-of- n n systems, we derive conditions for stochastic dominance based on monotonicity of the transformation and structural properties of the copula generators, formulated through super-additivity and Schur-type arguments. The results provide tractable criteria that extend existing comparisons beyond independence and illustrate the combined effect of dependence and parameter heterogeneity on system reliability. Keywords : Stochastic dominance, Archimedean copula, Majorization, Complex system MSC 2020 Classification : 60E15, 60E05, 62N05 1 Introduction Comparing the behaviour of lifetime distributions between complex systems has attracted considerable interest due to its applications in reliability and the overall performance of devices constructed from multiple components. It is clear that the lifetime of each component reflects on the collective behaviour of the system. Moreover, the way components interact with each other—either due to the system’s physical properties or statistical dependence between components—is crucial for characterising the system itself. Since lifetimes are random, any comparison must capture the stochastic nature of the problem and be based on an appropriate version of a stochastic order between the random variables representing the lifetimes of interest. The basic notions and properties of stochastic orders may be found in the monograph by 16 . Depending on the structure of a particular complex system, its lifetime may be described using order statistics derived from the component lifetimes. Consequently, one is often interested in the stochastic comparison of order statistics. A review of relevant results can be found in 10 and 9 , with the latter studying the effect of dependence between components on the overall lifetime and how to exert some control over that dependence. The physical properties of a system are typically reflected in how the lifetimes of its individual components determine the overall system lifetime. Among the most widely studied configurations are parallel and series systems, where the system operates if at least one component is functioning (parallel) or only if all components are functioning (series). Intermediate configurations lead to the class of (n-k) -out-of- n systems, in which the system continues to operate as long as at least n-k of the n components remain alive. There is an extensive literature on (n-k) -out-of- n systems, which is closely tied to the properties of (n-k) -th order statistics. Most of this work assumes independent component lifetimes, due to the mathematical challenges associated with deriving order statistic distributions in the presence of dependence. Focusing on parallel or series systems, it is not surprising that the first results in the literature compared such systems with respect to several stochastic orderings, assuming independent components with specific lifetime distributions, as in 7 or 8 for exponentially distributed components. More recently, 18 and 13 considered other distribution families for the component lifetimes. A different type of result, characterising non-comparability for a specific criterion, may be found in 3 , which provides results for the proportional hazard rates (PHR) and proportional reversed hazard rates (PRHR) models that we will consider later. The stochastic comparison of systems with dependent, possibly heterogeneous, components has been addressed using characterisations of joint distributions described through specific copula functions, in particular Archimedean copulas. In this direction, 11 proved stochastic comparison results when component lifetimes follow a PHR model; 15 considered stochastic domination with components lifetime distributions from a proportional odds model; and the PRHR model has been studied in 6 , among other models for the individual distributions, with emphasis in comparing systems with respect to dispersive or hazard rate orders. Further related results for distorted distributions and coherent systems may be found in 14 , while 2 and 1 provide results for more specific families of component lifetimes. In this paper, we develop a general framework for stochastic comparison of system lifetimes with dependent and heterogeneous components, whose marginal distributions arise from transformations of a common baseline and whose dependence structure is described by Archimedean copulas. This setting includes several widely used models, such as the PHR and PRHR families, within a unified formulation. For parallel and series systems, we establish stochastic ordering results based on monotonicity properties of the transformation and structural conditions on the copula generators. We further consider (n-k) -out-of- n systems, for which a different approach is required, deriving explicit representations of the system lifetime distribution under Archimedean dependence and obtaining stochastic ordering results when the joint distribution is described through the same copula. The applicability of the proposed framework is illustrated through several transformation models, highlighting the combined effect of dependence and parameter heterogeneity on system reliability. 2 Notations and some preliminary results We now introduce some general concepts and terminology to be used in the sequel. The set of all cumulative distribution functions in \mathbb{R} is denoted as \mathcal{P}(\mathbb{R}) . Given a random variable X we represent with F_{X}(\in\mathcal{P}(\mathbb{R})) and \overline{F}_{X} its cumulative distribution function and survival function, respectively, and may denote this as X\sim F_{X} . Analogously, given a n -dimensional random vector \mathbf{X}=(X_{1},\ldots,X_{n}) the distribution function is represented by F_{(X_{1},\ldots,X_{n})} , while the marginals are denoted as F_{X_{i}} , i=1,\ldots,n . Further, we shall represent as \overline{F}_{(X_{1},\ldots,X_{n})}(x_{1},\ldots,x_{n})={\rm P}(X_{1}>x_{1},\ldots,X_{n}>x_{n}) . Recall that a real valued function g is super-additive if, for every x,\,y\geq 0 , g(x+y)\geq g(x)+g(y) . It is well known that if g(0)\leq 0 , then the convexity of g implies its super-additivity. We recall the definition of stochastic dominance, the key comparison notion between random variables that will be studied. Definition 2.1 Consider two random variables X_{1}\sim F_{X_{1}} and X_{2}\sim F_{X_{2}} . We say that X_{1} is smaller than X_{2} in the usual stochastic order, denoted by X_{1}\leq_{st}X_{2} , if \overline{F}_{X_{1}}(x)\leq\overline{F}_{X_{2}}(x) for every x\in\mathbb{R} . 2.1 Joint distributions and copula representations To study systems with dependent components, we introduce a copula-based representation of joint distribution functions. It is well known, from the Sklar’s Theorem (see 17 ), that, given a random vector \mathbf{X}=(X_{1},\ldots,X_{n}) , there exist n -dimensional distribution functions, C and \widehat{C} , with uniform marginals such that \begin{array}[]{l}\displaystyle F_{(X_{1},\ldots,X_{n})}\left(x_{1},\ldots,x_{n}\right)=C\left(F_{X_{1}}(x_{1}),\ldots,F_{X_{n}}(x_{n})\right),\\[8.61108pt] \displaystyle\overline{F}_{(X_{1},\ldots,X_{n})}\left(x_{1},\ldots,x_{n}\right)=\widehat{C}\left(\overline{F}_{X_{1}}(x_{1}),\ldots,\overline{F}_{X_{n}}(x_{n})\right).\end{array} (1) The functions C and \widehat{C} are the copula and survival copula , respectively, of the vector \mathbf{X} . A particular class of copulas, introduced next, is of special interest. We need to recall a preparatory notion, before the actual definition of the class of interest. Definition 2.2 A real function f is called d -monotone in (a,b) , where a,b\in\mathbb{R} and d\geq 2 , if it is differentiable there up to the order d-2 and the derivatives satisfy (-1)^{i}f^{(i)}(x)\geq 0 , i=0,1,\ldots,d-2 and (-1)^{d-2}f^{(d-2)}(x) is nonincreasing and convex in (a,b) . Definition 2.3 Suppose \psi:[0,\infty)\rightarrow[0,1] is a continuous d -monotone function with \psi(0)=1 and \lim_{x\to\infty}\psi(x)=0 . Then, a copula C_{\psi} is called an Archimedean copula with generator \psi if it can be written as C_{\psi}\left(v_{1},\ldots,v_{n}\right)=\psi\left(\phi\left(v_{1}\right)+\cdots+\phi\left(v_{n}\right)\right) for all v_{i}\in[0,1] , i=1,\ldots,n , where \phi= \psi^{-1}=\sup\{x\in\mathbb{R}:\psi(x)>v\} is the right