Random Permutation Set Theory

• Random Permutation Set Theory (RPST) is a mathematical framework that models uncertainty using ordered element permutations, thereby extending classical Dempster-Shafer Theory.
• It establishes invertible transformations between permutation-based and set-based belief representations to refine order-sensitive evidence evaluation.
• RPST integrates combinatorics, entropy, and fusion techniques to improve classification accuracy and support applications in statistical testing and model theory.

Random Permutation Set Theory (RPST) is a mathematical framework for modeling, reasoning, and measuring uncertainty in information where the internal order of elements is essential. RPST generalizes classical frameworks such as Dempster-Shafer Theory (DST), transfering the representation power from unordered sets to sets equipped with the full combinatorics of permutations, and thereby enables refined modeling of ordered evidence, source reliability, fusion, and statistical comparison.

1. Formal Foundations and Structures

RPST extends DST by replacing focal subsets with ordered tuples, encapsulating both the composition and sequence of hypotheses. Given a frame of discernment $\Theta = \{x_1,\ldots,x_n\}$, the Permutation Event Space (PES) consists of all ordered $k$-tuples ($k=0,\ldots,n$), with $P(n, k) = n!/(n-k)!$ length-$k$ permutations. The PES is

$$\text{PES}(\Theta) = \{\emptyset\} \cup \bigcup_{k=1}^n \{\text{all ordered } k\text{-tuples drawn without repetition from }\Theta\}$$

An RPS on $\Theta$ is a probability mass function (PMF) $\mu: \text{PES}(\Theta) \to [0,1]$ satisfying $\mu(\emptyset)=0$ and $\sum_{A\in\text{PES}(\Theta)}\mu(A)=1$. When order is ignored, PES collapses to $2^\Theta$ and $\mu$ is a classical BPA (Xu et al., 2024, Deng et al., 2021, Cheng et al., 12 Oct 2025, Yang et al., 2022).

Relationship to Existing Uncertainty Theories

• DST: Encodes belief only in the presence/absence (unordered) of elements.
• Deng entropy: Reduces to RPST entropy when order is ignored.
• Shannon entropy: Further reduction when only singleton permutations are considered.

RPST thereby subsumes DST, Deng, and Shannon as special cases (Deng et al., 2021).

2. Transformations and Probability Assignment

RPST formalizes systematic and invertible transformations between permutation-based and set-based belief representations:

RPS $\to$ DST (Marginalization)

Order is forgotten by summing the PMF over all permutations whose element sets coincide: $$m(A) = \sum_{\pi\in\text{PES}(\Theta):\,\text{set}(\pi)=A} \mu(\pi)$$

DST $\to$ RPS (Refinement)

Given a BPA $m$ and auxiliary singleton ranking, each focal $A$ with $m(A)$ is split into $|A|!$ permutations. The split weight $\mu(\pi)$ is proportional to $m(\text{Element}(\pi))$ times a support degree $S_\text{ord}(\pi)$: $$S_\text{ord}(\pi) = \prod_{i=1}^k \left[\frac{\text{BetP}(\beta_i)}{\sum_{j=i}^k \text{BetP}(\beta_j)}\right],\quad \mu(\pi) = m(\text{Element}(\pi))\, S_\text{ord}(\pi)$$ where $(\beta_1,\ldots,\beta_k)$ is the ordered tuple of $\pi$ (Xu et al., 2024).

Sequence-Based Probability Transformation

Ordered beliefs are transformed for decision via the Ranked Probability Transformation (RPT): $$\text{Rpt}(x_i) = \sum_{\pi\ni x_i} w(x_i \mid \pi) \mu(\pi)$$ with a rank-dependent weight parameterized by $\lambda\in[0,1]$, interpolating between classical pignistic transformation ($\lambda=0$) and order-centric focus on top hypotheses ($\lambda\to 1$) (Xu et al., 2024).

3. Entropy and Information-Theoretic Measures

RPST introduces a family of entropy measures parameterized by the combinatorics of permutation events: $$H_\text{RPS}(\mathscr{M}) = -\sum_{k=1}^n\sum_{|\pi|=k} \mathscr{M}(\pi)\log\left(\frac{\mathscr{M}(\pi)}{F(k)-1}\right),\quad F(k)=\sum_{\ell=0}^k P(k,\ell)$$ The maximum entropy is achieved by a PMF proportional to $F(|\pi|)-1$, with normalization constant

$$\mathscr{M}^*(\pi) = \frac{F(|\pi|)-1}{\sum_{k=1}^nP(n,k)(F(k)-1)}$$

The maximum RPS entropy grows asymptotically as $\log(e \cdot (n!)^2)$ as $n\to\infty$, providing a conceptual bridge to classical entropy and factorial growth, and significantly reducing computational complexity in practical applications (Deng et al., 2021, Zhou et al., 2024).

4. Fusion, Distance, and Reliability in RPST

Fusion

Information fusion incorporates the order structure of evidence. The Left-orthogonal sum fuses RPS sources in reliability order—the fusion is order-sensitive, generalizing Dempster’s rule: $$\mu'(\pi) = \begin{cases} R_k\cdot \mu(\pi), & |\pi|=1 \ R_k\cdot \mu(\pi) + \frac{1-R_k}{\text{Perm}(n) - n - 1}, & |\pi|>1 \end{cases}$$ where $R_k$ is a source-specific reliability weight computed by aggregating samplewise decision contributions and normalizing (Xu et al., 2024).

Distance and Comparison

The cumulative Jaccard index quantifies the similarity of two RPSs via their ordered focal sets. The distance between two RPSs $\text{Perm}_1$, $\text{Perm}_2$ is: $$d_\text{RPS}^{[Orn]}(\text{Perm}_1,\text{Perm}_2) = \sqrt{\frac12\, (\mathbf{Perm}_1-\mathbf{Perm}_2)^\top\, \underline{CD}^{[Orn]}\, (\mathbf{Perm}_1-\mathbf{Perm}_2)}$$ with tunable emphasis on top ranks and truncation depth, generalizing the Jousselme distance by encoding ordered similarity (Cheng et al., 12 Oct 2025).

Repeatable RPS and Combination

The repeatable RPS ($\mathrm{R}^2\mathrm{PS}$) accommodates permutations with repetitions, introducing left and right junctional sum operators. Associativity, pseudo-Matthew effect, and consistency with DS orthogonal sum are established, with explicit combinatorial formulas for fusion semantics (Yang et al., 2022).

5. Applications and Integration with Random Walks, Quasirandomness, and Statistical Inference

Stochastic Processes and RPST

RPST supports the construction of random walks where each step is interpreted as a function of a randomly drawn permutation from the maximal-entropy PMF. Upon suitable normalization (double limit and rescaling), the process converges to a Wiener process, providing a rigorous link between permutation-based uncertainty reasoning and stochastic process theory (Zhou et al., 2024).

Statistical Testing and Quasirandomness

Permuton convergence and permutation pattern densities provide a bridge between combinatorial randomness and statistical independence. Specifically, a sequence of permutations is quasirandom iff the densities of six key patterns approach the uniform value $1/|\sigma|!$. This finite-obstruction principle enables efficient independence tests in rank statistics, directly connecting to permutation set structures (Crudele et al., 2023).

Shattering and Covering Arrays

In models of random systems, the threshold for a collection of random permutations to shatter all $t$-tuples (i.e., realize all $t$-length patterns in every position subset) exhibits sharp phase transitions of order $\Theta(\log n)$, supporting RPST’s utility for quantifying combinatorial coverage and VC-dimension analogues (Godbole et al., 2013).

6. Model-Theoretic and Structural Perspectives

Advanced research situates RPST in the context of homogeneous structures equipped with multiple linear orders. The random permutation, as the Fraïssé limit of finite double-order structures, displays rich automorphism groups (precisely 39 closed supergroups) and combinatorial symmetries. These form a model-theoretic substrate for permutation set theory: definable sets correspond precisely to those invariant under these groups, and an axiomatization using subsets obtained by first-order formulas in two orders is possible. This perspective unifies combinatorial, probabilistic, and model-theoretic views of RPST (Linman et al., 2014).

7. Practical Impact and Experimental Findings

Experimental studies consistently demonstrate advantages of RPST-based methods in pattern classification, where the representation of ordered hypotheses sharpens decision-making and source fusion. Average classification accuracy significantly exceeds classical DST and standard machine learning models, with order-focused transformations reducing ambiguity and improving robustness (Xu et al., 2024).

---

RPST thus serves as a mathematically principled, extendable, and computationally tractable framework for representing and reasoning with uncertainty in contexts where the order of evidence is paramount. It unites combinatorics, entropy, reliability, and statistical randomness, offering a spectrum of applications across pattern recognition, statistical testing, information fusion, combinatorial design, and model theory.