A flexible bounded stochastic framework for uncertainty and reliability in physical systems Authors: Tmader Alballa, Abdus Saboor, Muhammad Danish, Hamiden Abd El-Wahed Khalifa Published: 2026 ABSTRACT AbstractBounded random variables arise naturally in physical, engineering, and reliability systems when measurements represent proportions, efficiencies, normalized intensities, or constrained state variables. In this paper, a flexible bounded stochastic framework generated through a beta transformation of the Kumaraswamy (Kw) baseline is introduced, yielding a four-parameter family capable of capturing diverse boundary behaviors and hazard rate (HR) structures. Rigorous theoretical properties of the proposed model are developed, including structural identifiability, limiting behavior at the boundaries, shape characteristics of the probability density function (PDF) and HR functions, and explicit stochastic representations. Closed-form expressions for moments, probability-weighted moments are derived under mild regularity conditions, together with comprehensive information-theoretic characterizations based on Shannon, Rényi, and Tsallis entropies, as well as Kullback–Leibler divergence relative to baseline models. Likelihood-based inference is studied in detail, with explicit score functions, Fisher information, and asymptotic properties of the maximum likelihood estimators (MLE) established. An illustrative application to bounded measurements from an engineered system demonstrates the practical relevance of the theoretical results. The proposed framework provides a mathematically rigorous and interpretable tool for uncertainty quantification and reliability analysis of bounded physical quantities.Keywords: Bounded-response modeling, Unit-interval data, Statistical model, Maximum likelihood estimation, Engineering measurements, Environmental dataSubject terms: Engineering, Mathematics and computing, Physics Introduction Bounded stochastic variables arise naturally throughout the physical sciences, where many quantities of interest are intrinsically confined to a finite interval and are frequently normalized to lie in (0, 1). Representative examples include fractional energy dissipation, material porosity and diffusion ratios, normalized strain measurements, reliability indicators of physical components, and effectiveness measures in transport and diffusion processes. Rigorous probabilistic modeling of such quantities is essential for understanding uncertainty, degradation, and failure dynamics in physical systems. This requires statistical models that are sufficiently flexible to capture skewness, boundary concentration, and diverse hazard-rate (HR) behaviors, while remaining analytically tractable for inference and interpretation2,3,22. Among classical bounded distributions, the beta model occupies a central role because of its interpretability, flexible shape, and foundational importance in modeling proportions and rates22–24. The beta model has been extensively used to model proportions, probabilities, and rates in reliability analysis, Bayesian modeling, project planning and control systems such as the program evaluation and review technique (PERT), stochastic simulation frameworks, cost assessment in industrial and civil engineering projects, and population genomic studies. Despite its broad applicability, the beta model may be restrictive in physical applications where empirical data exhibit persistent asymmetry, pronounced boundary behavior, or non-monotone hazard rate patterns. An alternative bounded model with notable computational advantages is the Kumaraswamy (Kw) model, introduced by Kumaraswamy1. While closely related to the beta model, the Kw model admits closed-form expressions for the cumulative distribution function (CDF) and quantile function (QF), facilitating numerical implementation. The model has been applied to a range of bounded natural phenomena, including individual height measurements, test scores, air temperature indices, and hydrological variables such as daily streamflow and rainfall. Nevertheless, like the beta model, the Kw model may lack sufficient flexibility to accommodate complex boundary dynamics and heterogeneous tail behavior observed in physical reliability and degradation data. Motivated by these limitations, a substantial body of recent research has focused on extending beta- and Kw-type models to enhance model flexibility while retaining analytical tractability. These developments include Kw-based extensions with improved HR modeling2,3, exponentiated and inverted forms designed to accommodate alternative boundary behavior4,5, and beta-generated constructions aimed at enriching HR dynamics6,7. Additional progress on the unit interval has been achieved through transformation-based and higher-parameter models, such as Johnson-type unit models32, modified beta linear exponential models8, higher-parameter beta families9, and unified bounded models10. More recent contributions continue to emphasize interpretability and flexibility for bounded data, including the unit log–log distribution11, and parsimonious two-parameter unit models13. Collectively, these studies highlight the persistent challenge of developing bounded stochastic models capable of capturing the complex empirical behavior encountered in physical reliability, degradation, and transport processes. Carrasco and Cordeiro15 proposed a systematic framework for extending the Kw model, providing a structured mechanism for constructing flexible bounded models with enhanced shape adaptability while preserving analytical tractability. This framework is particularly attractive for applications involving physical quantities constrained to the unit interval. In physical and engineering applications, bounded random variable (r.v.) often arise when measurements represent normalized efficiencies, constrained state variables, degradation indicators, fractional compositions, or sensor-derived process indices. Examples include material porosity, normalized strain or stress measures, reliability scores, bounded quality-control indices, diffusion and transport ratios, and monitored industrial process variables. In such settings, the observed data may exhibit asymmetry, interior concentration, or substantial accumulation near the boundaries, depending on the operating regime and the governing physical constraints. This is particularly relevant in industrial monitoring systems, where bounded measurements may combine routine operating behavior with occasional near-boundary excursions associated with abnormal, depleted, saturated, or transition-like states. These considerations create a practical need for bounded stochastic models that remain analytically tractable while allowing flexible boundary behavior, varied interior shapes, and interpretable reliability characteristics. It is from this applied perspective, rather than from formal generalization alone, that the present beta–Kw (BK) formulation is motivated2,3,8,11,13,22. Building on these developments, we consider in this paper a four-parameter BK model defined on (0, 1), obtained by combining a Kw-type transformation with a beta-generated structure. Rather than treating the formulation as a purely statistical generalization, we emphasize its role as a flexible stochastic representation for bounded physical quantities. The proposed model accommodates a wide range of PDF and HR shapes, including increasing, decreasing, unimodal, and bathtub-type forms commonly observed in physical reliability and degradation contexts. Beyond model’s shape characteristics, information-theoretic measures provide additional insight into uncertainty and variability in physical systems. Entropy-based quantities, such as Rényi and Tsallis entropies, are widely used in statistical physics and related fields to characterize disorder, non-equilibrium behavior, and departures from classical assumptions25–28. Motivated by this perspective, we derive entropy measures for the BK model and examine their dependence on the governing parameters, thereby offering an information-theoretic interpretation of model flexibility. Likelihood-based inference is developed to enable practical estimation of model parameters from observed bounded data. Standard asymptotic properties of the maximum likelihood estimator (MLE) are established, allowing uncertainty quantification through the Fisher information matrix, Wald-type confidence intervals, and likelihood ratio testing where appropriate. These tools provide a principled and reproducible inferential framework; finite-sample performance is additionally assessed via a Monte Carlo simulation (MCS) study. The main contributions of this work are threefold. First, we introduce a flexible bounded stochastic model capable of reproducing a broad range of probability density function (PDF) and HR behaviors encountered in physical reliability and degradation processes. Second, we provide an information-theoretic characterization of uncertainty for the proposed model via Rényi and Tsallis entropy measures. Third, we develop a likelihood-based inferential framework and illustrate its use with bounded data from an engineered system. To better distinguish the standard background material from the original contribution of this work, we organize the manuscript into a preliminary section followed by sections containing the new BK-model development. Section 2 collects the mathematical ingredients and notation used throughout the paper, including the beta model, the Kw baseline model, and the regularized incomplete beta function. Section 3 then introduces the proposed BK model as the main methodological contribution. The subsequent sections develop its structural properties, moment results, entropy-based measures, likelihood inference, asymptotic theory, numerical evaluation, and real-data application. This revised organization is intended to make explicit which parts of the manuscript are preparatory and which parts constitute the new contribution of the present study. Preliminary concepts and notation This section summarizes the standard concepts and notation that are used in the construction and analysis of the proposed BK model. Its purpose is to separate the necessary background material from the main contribution developed in the subsequent sections. Beta model Let V be a beta random variable (r.v.) with shape parameters and . Its PDF is where denotes the beta function. The Kw model The Kw model with parameters and has CDF and PDF Regularized incomplete beta function For , the regularized incomplete beta function is defined by This function is used in beta-generated constructions and plays a central role in defining the BK model. The BK Model This section introduces the proposed BK model and establishes its basic stochastic representation. We define the model through a beta-generated mechanism applied to the Kumaraswamy baseline, derive its CDF and probability density function (PDF), and discuss the role of each parameter in controlling interior shape and boundary behavior. Let X be a bounded r.v. taking values in the unit interval (0, 1). We construct the BK model by applying a beta-generated mechanism to the Kw baseline model. The Kw model with shape parameters and has CDF and corresponding PDF Let V be a beta r.v. with parameters and . The BK model is defined through the transformation where denotes the regularized incomplete beta function. Let and the PDF of the beta model is Define the strictly increasing transformation with inverse By the change-of-variables formula, the PDF of X is Differentiating Eq. (5) and substituting in Eq. (7), one gets where is the beta function. The parameter vector of the BK model is The parameters and primarily govern the baseline shape inherited from the Kw model, while and regulate tail behavior and boundary concentration through the beta-generation mechanism. This separation allows independent control of interior shape and boundary dynamics, which is particularly relevant for modeling bounded physical quantities subject to heterogeneous operational variability. An alternative stochastic representation follows directly from (3). If , then has CDF (3). This representation is useful for analytical derivations and numerical generation. The survival function (SF) and HR function (HRF) follow immediately from (3) and (8). Owing to the four-parameter structure, the BK model accommodates a wide variety of HR shapes, including increasing, decreasing, unimodal, and bathtub-type forms commonly encountered in physical reliability and degradation processes. The BK model reduces to several well-known bounded models as special cases. In particular, setting recovers the Kw model, while additional parameter restrictions yield simpler bounded families. These nesting properties facilitate likelihood-based comparison via standard hypothesis testing procedures. HR shape: boundary behaviour and sufficient conditions We now examine the endpoint behavior of the HR function and state sufficient conditions for qualitatively distinct reliability shapes generated by the BK model. Proposition 1 Let . Then the HR h(x) satisfies: Proof As , we use the expansions Hence the BK PDF (8) satisfies Moreover as . Therefore , which yields (10). As , set . Using the standard small-z asymptotic of the regularized incomplete beta function, Also, since . Consequently, which proves (11). Theorem 1 Let h(x) be the BK HR. (i)If , then as and as . If, in addition, h(x) is continuous on (0, 1) (which holds whenever f(x) is continuous and ), then h(x) attains at least one interior minimum; in particular, the BK HR is bathtub-type (U-shaped) in the sense of possessing an interior minimum. If , then as and as . If, in addition, h(x) is continuous on (0, 1) (which holds whenever f(x) is continuous and ), then h(x) attains at least one interior minimum; in particular, the BK HR is bathtub-type (U-shaped) in the sense of possessing an interior minimum. (ii)If , then as and as . If, moreover, h(x) is nondecreasing on (0, 1), then h(x) is increasing (IFR). If , then as and as . If, moreover, h(x) is nondecreasing on (0, 1), then h(x) is increasing (IFR). (iii)The BK HR cannot be strictly decreasing on the full interval (0, 1), since . The BK HR cannot be strictly decreasing on the full interval (0, 1), since . Proof Part (i) follows from Proposition 1. If h(x) is continuous on (0, 1) and diverges at both boundaries, then h(x) must attain a finite minimum at some interior point, which yields a bathtub-type HR. Part (ii) follows directly from Proposition 1; the IFR statement is immediate under the stated monotonicity condition. Part (iii) follows from (11). Lemma 1 Let with and PDF (8) Then: Consequently, f is unbounded at 0 iff (left J-type behaviour), and f is unbounded at 1 iff (right boundary / reverse J-type behaviour). Proof As , we have and , while so . Since is fixed and the leading boundary order is governed by , this yields (12). As , use , hence and . Substituting these limits in the PDF (8) gives (13). The divergence criteria follow immediately from the powers and . Figure 1 illustrates representative density profiles generated by the BK family under different parameter configurations. These examples help visualize the boundary-dominated, interior-unimodal, and accumulation-type behaviors described by the endpoint asymptotics. Representative PDF profiles illustrating distinct bounded physical regimes, including boundary-dominated decay, interior-dominated behavior, and boundary accumulation. Depending on the parameter values , the BK model generates a wide range of PDF shapes on (0, 1), including monotone decreasing/increasing, unimodal, U-shaped, and boundary-dominated (J-type) profiles. Figure 1 displays representative BK density profiles corresponding to boundary-dominated decay, interior unimodality, and right-boundary accumulation. These examples illustrate how the four-parameter structure can generate qualitatively different bounded regimes through changes in endpoint behavior and interior concentration. The boundary behavior follows directly from the local asymptotics of the PDF. As so the PDF is unbounded at the left boundary (a J-shaped profile with mass concentrated near 0) when and vanishes at 0 when . As so the PDF is unbounded at the right boundary (a reverse J-shaped profile with mass concentrated near 1) when and vanishes at 1 when . In addition, the factor controls interior versus boundary emphasis, with increasing concentration near 0 and producing the opposite effect. Model properties This section groups the main analytical properties of the BK model into three parts: moment results, information-theoretic measures, and divergence-based comparisons. This organization is intended to make the theoretical development easier to navigate. Lemma 2 Let with . Then, for and the series converges absolutely for and converges at . Moreover, for any nonnegative measurable r.v. provided Proof Equation (14) is the generalized binomial expansion. Absolute convergence holds for and convergence at holds when . Under the stated absolute summability condition, Tonelli’s (or Fubini’s) theorem justifies interchanging expectation and summation; see, e.g.,20. Remark 1 Although the BK PDF is given by (8), many derivations are more tractable via the transport representation with , which is equivalent to (8). Moment results We now derive raw-moment expressions and summarize the resulting numerical characteristics of the BK model. These results help quantify how the parameters affect location, dispersion, skewness, and tail behavior on the unit interval. Theorem 2 Let with and PDF (8). Equivalently, let with PDF (4) and using Eq. (6), if , then the raw moment of BK model is derived as: Proof Let and using (4, 5), we have Using the generalized binomial expansion (14), Thus, for , Under the moment-existence condition , the right-hand side of (16) is integrable and the interchange of expectation and summation is justified by Lemma 2 together with Tonelli’s theorem; hence Since , Substituting in Eq. (17) yields (15). Table 1 reports numerical values of the first four raw moments, together with variance, standard deviation, coefficient of variation, skewness, and kurtosis for selected parameter configurations. Table 1Some numerical values of the first four raw moments, variance, standard deviation, coefficient of variation, skewness, and kurtosis of the BK model for selected parameter values. S.D.C.V.CSCK(1.5, 2.5, 1.1, 1.2)0.63270.44090.32680.25310.04050.20110.0542-0.07722.4459(1.1, 2.7, 1.7, 1.9)0.71330.53400.41440.33070.02520.15880.0420-0.09752.8809(2.1, 3.1, 1.9, 2.1)0.63870.42620.29470.63730.01820.13510.0358-0.05072.7568(2.3, 3.3, 2.1, 2.4)0.63730.42120.28700.20070.01510.12270.0324-0.04662.8021(2.5, 3.7, 2.6, 2.7)0.66950.45900.32130.22910.01070.10360.0272-0.04862.8991(2.7, 4.1, 2.9, 3.1)0.68190.47330.33390.23900.00840.09140.0240-0.04612.9449(2.7, 4.1, 3.2, 3.5)0.67990.46980.32940.23400.00750.08660.0227-0.04332.9429(2.7, 4.1, 3.5, 3.7)0.68440.47530.33450.23830.00690.08290.0217-0.04202.9453(2.9, 4.3, 3.5, 3.7)0.68640.47750.33640.23970.00640.08010.0210-0.04182.9566(3.2, 4.7, 3.5, 3.7)0.69540.48930.34800.25000.00570.07530.0197-0.04262.9818 Some numerical values of the first four raw moments, variance, standard deviation, coefficient of variation, skewness, and kurtosis of the BK model for selected parameter values. Table 1 summarizes numerical characteristics of the BK model for selected parameter configurations. The raw moments – indicate increasing mean levels as the shape parameters increase, while the variance and standard deviation gradually decrease, reflecting increased concentration around the mean. The coefficient of variation consistently declines, confirming reduced relative dispersion. The skewness values are negative but small in magnitude, indicating mild left-skewed behavior, whereas the kurtosis values remain close to a mesokurtic benchmark. These results illustrate the flexibility of the BK model for bounded data with controllable dispersion, skewness, and tail behavior. Entropy measures In this subsection, we derive Shannon, Rényi and Tsallis entropies of the BK model. We begin with Shannon entropy. Lemma 3 Let be a sequence of measurable functions on (0, 1) such that where is integrable on (0, 1). Then In particular, if is of Beta type and integrable under the stated parameter conditions, then Tonelli’s theorem (and dominated convergence, where needed) applies. Theorem 3 Let with . Then Shannon entropy of the BK model is where denotes the digamma function. Proof Since and applying log on Eq. (7), one gets Taking expectation on Eq. (19) and multiplying by , we obtain Now, Therefore, Eq. (20) becomes From Eq. (5), we have Since hence Similarly, Let . Using the logarithmic series and Lemma 3, For Thus Substitute (22) and (23) into (21). This yields (18). We next extend the information-theoretic analysis to Rényi and Tsallis entropies, which provide complementary measures of uncertainty under alternative weighting schemes. Theorem 4 For , , the Rényi entropy of order q for PDF (8) is provided and assume the integrability conditions Proof Using (8) in the definition of Rényi entropy Let , so and . Then hence Eq. (27) becomes. Set and . Under the condition (third condition in (25)), Lemma 2 yields the expansion The series in Eq. (29) is termwise nonnegative in absolute value and, under (25), each term produces a beta-integrable kernel on (0, 1). Dominated convergence (equivalently, Fubini’s theorem under the stated integrability conditions) yields Combining the Eqs. (30), (28), (27) and (26) to obtain (24). Corollary 1 For , , and , under the integrability conditions (25), the Tsallis entropy of order q associated with the PDF (8) is Proof The Tsallis entropy of order q is defined by By Theorem 4, under (25) the q-power integral admits the series–beta representation Substituting (33) into (32) yields (31). Kullback–Leibler divergence The next result gives a closed-form Kullback–Leibler (KL) divergence for the BK family in the important case where the transport map is the same, i.e. are fixed and only vary. This result is sufficient for the present study and yields a rigorous closed-form information-theoretic comparison between two BK laws sharing the same transport map. Theorem 5 Let and let with . Denote their PDFs by and , respectively. Then exists (is finite) and admits the closed form where is the digamma function. Equivalently, Proof Using Eq. (5) with is a bijection with derivative . For , let where and . By the BK construction, with PDF Moreover, using Eq. (7), one has Hence, for every Substituting into (34) and using (37), Now set ; then and the integral becomes which proves (36). Since , the beta PDFs are integrable and , hence . Finally, compute explicitly: Taking expectation under yields The standard beta expectations are Substituting these identities gives (35). Likelihood inference and asymptotics This section develops the inferential framework for the BK model. We derive the score equations and observed information matrix, discuss identifiability and regularity conditions, and then state the large-sample properties of the MLE. Throughout this section, let be an i.i.d. sample from the BK model (8) with parameter vector and support . Write The log-likelihood is Score vector and likelihood equations Theorem 6 For , the score vector exists and its components are Proof The derivatives of the normalizing terms are standard: and . The remaining terms follow by direct differentiation using the identities and, for -derivatives, Substituting above into Eq. (38) yields equations (39–42). Remark 2 The MLE is defined as any solution in to the system , equivalently (39–42). No closed-form solution is available; numerical maximization is required. Observed information and Fisher information Define the observed information matrix Theorem 7 For , the -block of is where is the trigamma function. Proof Differentiate equations (39) and (40) using and linearity. The data-dependent terms and do not depend on . Remark 3 Closed expressions for the entries involving are obtainable by differentiating equations (41, 42); however, they are lengthy and offer limited insight. In implementation we evaluate numerically and use for standard errors (SEs) and Wald-type inference. The (expected) Fisher information matrix is defined by whenever the derivatives can be interchanged with expectation. Existence, identifiability and consistency Theorem 8 If for all i and , then in Eq. (38) is finite and in Proof For and we have and hence . If also , then and thus and are finite. The remaining terms in Eq. (38) are evidently finite. Differentiability follows since all compositions involved are on (0, 1) and . Assumption 1 The parametric family is identifiable, i.e., almost everywhere implies . Proposition 2 Let denote the BK PDF on (0, 1), If for two parameter vectors and then . Hence the BK PDF is identifiable. Fix and write . As we have , hence and therefore Thus the left-endpoint exponent is . Equality almost everywhere implies equality of the endpoint exponents, hence Next, as we have , so . Hence so the right-endpoint exponent is . Therefore To separate the remaining degrees of freedom, consider the log-density difference , which equals 0 for almost everywhere x. Using (47, 48) to cancel the endpoint power terms, we obtain where R(x) collects the remaining terms involving and . Since the functions , , and are not affinely dependent on any nontrivial subinterval of (0, 1) when , the identity forces and . Then (47) and (48) yield and . Theorem 9 Assume 1. Let be the true parameter and assume that lies in the interior of a compact set over which is uniquely maximized at . Then any sequence of MLEs satisfies in probability as . Proof Write . By Proposition 2 and continuity of on (0, 1) for each , the uniform law of large numbers applies on K: The stated uniqueness of the maximizer of the limit criterion on K, together with the argmax continuous mapping theorem, yields in probability. Asymptotic normality and inference Assumption 2 Assume that in a neighborhood of the map is twice continuously differentiable for almost everywhere , and there exists an integrable envelope dominating the first and second derivatives so that differentiation may be interchanged with expectation. Assume also that exists and is nonsingular. Theorem 10 Under Assumptions 1 and 2, the MLE satisfies Moreover, is a consistent estimator of Proof By the likelihood equations and a mean-value expansion, for some intermediate point between and . Under 2, is an average of i.i.d. mean-zero scores with covariance , hence by the multivariate CLT. Further, by a law of large numbers and consistency of (Theorem 9), so Slutsky’s theorem yields the stated limit model and the variance consistency. Wald-type confidence intervals Let . Approximate Wald intervals are where is the standard normal quantile. Likelihood ratio tests For testing against , the likelihood ratio statistic is where and denote the unconstrained and constrained MLEs. Under and regularity, , where is the number of constraints. The Monte Carlo simulation study This section examines the finite-sample performance of the MLEs through a Monte Carlo Simulation (MCS) study. The objective is to assess the empirical behavior of bias and mean squared error (MSE) under different parameter settings and sample sizes, thereby complementing the large-sample theoretical results established earlier. A MCS study was conducted to examine the finite-sample performance of the MLEs of the parameters of the BK model. The primary objective of the simulation was to assess the empirical bias and MSE of the estimators under different sample sizes, thereby complementing the asymptotic results established in Section 4.4. Random samples were generated from the BK distribution using the inverse transform sampling method based on its QF. Specifically, if , then a BK random variate was generated as which guarantees that all simulated observations lie in the unit interval (0, 1). The simulation experiment was performed for the following true parameter configurations: representing different shape and boundary behaviors of the BK model. For each configuration, random samples of sizes were generated, and each scenario was replicated times to ensure MCS stability. For each generated sample, the parameter vector was estimated by maximizing the log-likelihood function given in Eq. (38) using the BFGS quasi-Newton optimization algorithm. Let denote the MLE of a generic parameter obtained in the ith replication. The empirical bias and MSE were computed as The numerical results of the simulation study are reported in Table 2. The results show that, as the sample size increases, the absolute biases and MSEs of all parameter estimators decrease steadily and approach zero. This behavior confirms the consistency of the MLEs and supports the asymptotic normality results derived in Theorem 10. Moreover, the estimators of the shape parameters and , which play a central role in governing the boundary and hazard rate behavior of the BK model, exhibit satisfactory stability even for moderate sample sizes. Overall, the simulation study demonstrates that the MLE method provides reliable and accurate parameter estimation for the BK model in finite samples. Table 2Empirical bias and MSE of the MLEs of the BK model parameters based on 8000 Monte Carlo replications for different sample sizes N. N MSEBiasMSEBiasMSEBiasMSEBias (α)0.80.50.50.7500.7713– 0.68170.2522– 0.38170.1522– 0.38170.3449– 0.5817-- - - 1000.7571– 0.67170.2441– 0.37170.1441– 0.37170.3328– 0.5717- - - - 1500.6795– 0.75720.2152– 0.45720.13521– 0.45720.31– 0.6572- -- - 2000.6039– 0.77500.2089– 0.47500.1289– 0.47500.3089– 0.6750- ---3000.4838– 0.69060.1940– 0.39060.1194– 0.39000.3057– 0.5906--- -4000.3937– 0.76780.1730– 0.46780.1130– 0.46780.2901– 0.6678----5000.3362– 0.79730.1478– 0.49730.1078– 0.49730.2867– 0.69731.00.80.71.2500.4569– 0.63460.2431– 0.43460.1662– 0.33460.7508– 0.8346-- --1000.4300– 0.68410.2747– 0.48410.1879– 0.38410.8219– 0.8841-- --1500.4202– 0.64520.2722– 0.49520.1631– 0.39520.7583– 0.8952-- --2000.4175– 0.74540.2703– 0.49540.1402– 0.44540.7330– 0.9454- - - - 3000.4175– 0.75830.2642– 0.58300.1325– 0.45830.7148– 0.9583---- 4000.4081– 0.77400.2331– 0.59000.1256– 0.47400.7130– 0.9740----5000.4062– 0.79200.2241– 0.59200.1631– 0.48520.6830– 0.9952 Empirical bias and MSE of the MLEs of the BK model parameters based on 8000 Monte Carlo replications for different sample sizes N. The non-monotone behavior of the empirical bias in the configuration should be interpreted as a finite-sample effect rather than evidence against consistency of the MLEs. In this case, the endpoint products satisfy and , indicating pronounced boundary influence in the fitted PDF. Such boundary-dominated samples tend to generate a relatively irregular likelihood surface and stronger dependence among the four parameters, especially in moderate sample sizes. As a result, the empirical bias may fluctuate across intermediate values of n even when the overall estimation performance improves. For this reason, the simulation results are more appropriately interpreted through the joint pattern of decreasing mean squared error and gradual stabilization of the estimates, rather than through strict monotonicity of the empirical bias at every sample size. Applications This section evaluates the empirical usefulness of the BK model for bounded process data arising from an engineered system. Our aim is not to claim universal superiority over all competing bounded models, but to assess whether the structural flexibility of the BK family yields a meaningful and interpretable representation of normalized physical measurements. In particular, we focus on boundary concentration, interior operating regimes, and inferential limitations that arise in the fitted analysis. The BK model’s modelling effectiveness is carefully compared against several established bounded models, including the one-parameter Unit Lindley16, the Kumaraswamy (Kw)2, the Topp–Leone (TL)17, beta, the three parameter bounded (3PB)29, the Log Lindley (LL)18 and the unit Lindley (UL)12. The standard statistical methods to evaluate and compare the models have been used , namely the CVM (Cramér-Von Mises), AD (Anderson-Darling), KS (Kolmogorov-Smirnov) statistics along with their corresponding p-values. Additionally, model selection criteria such as the Akaike Information Criterion (AIC), Bayesian information criterion (BIC), corrected Akaike Information Criterion (CAIC) and Hannan-Quinn Information Criterion (HQIC) are also estimated, providing an in-depth assessment of the model’s efficacy. The MLEs and their corresponding SEs for the BK model as well as competing models are provided in Table 3. The goodness-of-fit statistics and summaries are shown in Table 4. Beyond numerical comparison, we also examine whether the fitted BK shape provides a plausible interpretation of the bounded process behavior represented by the data. Table 3MLEs and SEs (in parentheses) for SWaT dataset.ModelMLEs and SEs (in parentheses)BK0.6755 (1.8165)13.4095 (35.5546)21.8180 (52.9681)7.8043 (22.5442)3 PB (,,) –Kw (,) ––Beta (,) ––LL (,) ––TL () –––UL () ––– MLEs and SEs (in parentheses) for SWaT dataset. Table 4Log-likelihood, information criteria, and goodness-of-fit test statistics (CVM, AD, KS) for models fitted to the SWaT dataset.Model AICCAICBICHQICCVMADKSp-valueBK14.352– 18.949– 16.282– 14.966– 18.1720.25671.59870.22080.28333PB7.810– 9.620– 8.120– 6.632– 9.0370.48212.54640.36750.0090Kw13.474– 22.949– 22.243– 20.958– 22.5600.25671.59870.22080.2833Beta12.562– 21.124– 20.418– 19.132– 20.7350.29311.68300.23590.2155LL11.863– 19.726– 19.020– 17.735– 19.3370.26231.52870.23350.2256TL11.587– 21.175– 20.953– 20.179– 20.9810.34001.86240.25480.1487UL13.827– 25.654– 25.432– 24.658– 25.4600.29211.56210.24210.1917 Log-likelihood, information criteria, and goodness-of-fit test statistics (CVM, AD, KS) for models fitted to the SWaT dataset. The SWaT dataset is taken from the Secure Water Treatment testbed31, which records sensor readings gathered from a scaled-down industrial water-treatment testbed operating under both normal and attack circumstances. The plant reproduced real-time operation for approximately 11 days, supplying 5 gallons per minute, and the dataset contains key physical measurements related to the treatment process. In the present study, the SWaT data are used as an example of bounded process variables arising in an engineered monitoring environment, where observations may reflect routine operation together with occasional excursions toward lower or upper normalized ranges. The SWaT dataset was analyzed without separating normal and attack regimes, reflecting realistic operational variability in industrial control systems. The combined dataset reflects realistic operational variability, including both nominal behavior and rare extreme deviations, and therefore provides a stringent test for bounded statistical models. All observations were normalized to the unit interval prior to analysis. For a normalized process variable from the SWaT system, the fitted BK PDF has a direct operational interpretation. A left-heavy PDF indicates that the process spends relatively more time near lower normalized states, whereas a right-heavy PDF indicates persistence near the upper operating range. An interior unimodal PDF suggests concentration around a preferred or stable operating regime. The HRF complements this picture by describing how the conditional tendency to move beyond a given normalized level changes across the unit interval. In particular, an increasing HRF indicates that this tendency becomes stronger as the process variable moves toward higher levels, while a bathtub-type HRF suggests comparatively lower conditional instability in the middle range and greater instability near the extremes. In this sense, the BK model is useful not only because it improves shape flexibility, but also because it permits a physically interpretable description of boundary concentration, interior operating behavior, and reliability-related variation in bounded process data. To assess empirical performance, we compare the BK model with several established bounded models using likelihood-based and goodness-of-fit criteria. However, beyond numerical comparison, we also emphasize whether the fitted BK shape yields a plausible interpretation of the bounded process behavior represented by the data. The SWaT analysis should be interpreted as an application to normalized process measurements from a monitored industrial water-treatment system, rather than as evidence of uniform dominance of the BK model over all competitors. The practical value of the BK family in this setting lies in its structural decomposition: the parameters govern the Kw-type baseline shape, while regulate boundary emphasis and asymmetry through the beta-generated layer. This separation permits distinct adjustment of interior shape and near-boundary concentration, which is useful when process data combine regular operating behavior with occasional excursions toward lower or upper normalized limits. From an engineering perspective, left-boundary concentration may reflect frequent low-level states, right-boundary concentration may indicate persistent operation near the upper normalized range, and intermediate unimodal shapes may represent concentration around a preferred operating regime. For a normalized process variable from the SWaT system, the fitted PDF therefore admits a direct operational interpretation: a left-heavy shape indicates greater residence near lower states, a right-heavy shape indicates persistence near upper states, and an interior unimodal shape indicates concentration around a preferred regime. The corresponding HRF complements this interpretation by describing how the conditional tendency to move beyond a given normalized level changes across the unit interval. In particular, an increasing HRF suggests a growing tendency toward higher-range observations, whereas a bathtub-type HRF indicates comparatively lower conditional instability in the middle range and greater instability near the extremes. The BK model is therefore best viewed here as a flexible descriptive tool for bounded process variables with interpretable shape adaptation, rather than as a universally dominant alternative under every fit criterion. At the same time, for the SWaT fit, some BK parameter estimates are accompanied by SEs that are large relative to the corresponding maximum likelihood estimates. This indicates that the likelihood surface may be relatively flat in some parameter directions, so the individual parameter estimates should be interpreted with caution. Accordingly, in this application the BK model is viewed primarily as a flexible and physically interpretable bounded representation, while parameter-level inference is treated more cautiously than the overall shape-based description of the process behavior. Implications for physical sciences and concluding remarks This section summarizes the main theoretical and applied findings of the paper, discusses their relevance for modeling bounded physical quantities, and highlights the principal limitations and possible directions for future work. Bounded stochastic variables play a fundamental role in the physical sciences, where normalized quantities develop naturally in the narrative of material properties, transport processes, reliability of physical factors, and energy-related measures. In such contexts, probabilistic models must accommodate boundary effects, asymmetric behavior, and diverse failure-rate dynamics while remaining interpretable and computationally tractable. In this paper, a flexible bounded stochastic framework based on a BK construction is examined. The proposed formulation develops a deep class of PDF and HRF shapes, including increasing, decreasing, unimodal, and bathtub-type behaviors, which are commonly confronted in physical degradation and reliability processes. The methodical structure of the model allows explicit derivation of model properties, moments, and entropy-based measures, facilitating both theoretical interpretation and practical implementation. From an information-theoretical perspective, the Rényi and Tsallis entropies extend insight into the uncertainty and condition associated with bounded physical systems under varying parameter systems. Such measures are principally relevant in statistical physics and related fields, where departures from classical equilibrium behavior and non-extensive effects are of interest. The dependence of entropy on model parameters focuses on how changes in system characteristics may influence uncertainty and variability in observed physical quantities. Likelihood-based inference was developed to support parameter estimation from empirical data. A MCS study further supported the inferential framework by showing that the MLEs achieve improved accuracy with increasing sample size, as evidenced by decreasing empirical bias and mean squared error. Overall, the BK framework investigated here provides a mathematically transparent and physically interpretable tool for modeling bounded stochastic variables. Its flexibility, combined with tractable inference and information-theoretic characterization, makes it potentially useful for applications in reliability analysis, materials science, environmental modeling, and other areas where bounded physical quantities and uncertainty play a fundamental role. Future work may explore its integration into specific physical models or its extension to multivariate and spatio-temporal settings. Limitations. The empirical findings should be interpreted with two limitations in mind. First, the present application is based on one bounded industrial dataset, so broader generalization to other physical and engineering settings requires additional case studies. Second, for the SWaT fit, some BK parameter estimates are accompanied by relatively large SEs, indicating that the likelihood surface may be flat in certain directions and that individual parameter estimates should therefore not be over-interpreted. For this reason, the empirical contribution of the present study is best understood as demonstrating feasibility and structural flexibility, rather than uniformly sharp parameter-level inference in every application. Acknowledgements This research is supported by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R404), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. Author contributions A.S. and T.A. conceptualized the study and developed the methodology. A.S. and M.D. implemented the numerical procedures, and carried out the data analyses. They also contributed to the theoretical derivations and verified the analytical results. All authors interpreted the findings, contributed to writing and revising the manuscript, and approved the final version of the manuscript. Funding This research is supported by Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2026R404), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. Data availability The dataset used and analysed during the current study available from the corresponding author on reasonable request. Code availability All scripts used to generate random variates via the BK QF, compute MLEs using BFGS, and reproduce the MCS results and fitted-model tables will be made publicly available upon acceptance (or uploaded as Supplementary Code at submission). Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The dataset used and analysed during the current study available from the corresponding author on reasonable request. All scripts used to generate random variates via the BK QF, compute MLEs using BFGS, and reproduce the MCS results and fitted-model tables will be made publicly available upon acceptance (or uploaded as Supplementary Code at submission).