Separator Enumerator: Concepts & Applications

• Separator enumerator is a mapping from finite words to real numbers that produces a dense, countable subset of [0,1), generalizing the notion of normal numbers.
• It plays a key role in effective dimension theory by underpinning f-normality and finite state dimensions through computable approximations and equidistribution challenges.
• Applications span graph separator algorithms, permutation statistics, and document/table analysis, offering efficient enumeration methods and robust structural detections.

A separator enumerator is a map that assigns real numbers in the unit interval to finite words over a finite alphabet such that the set of assigned values forms a dense, countable subset of $[0,1)$. Separator enumerators originated as a representation-theoretic generalization of normal numbers in the study of effective dimension and randomness. Beyond this, the enumeration of separators occurs across distinct contexts, including permutation statistics, graph theory (vertex separators), and document or table analysis (visual or logical line separators).

1. Formal Definition and Foundational Properties

Let $\Sigma$ be a finite alphabet of cardinality $k\geq 2$. A separator is any countable dense subset $S\subseteq [0,1)$. A separator enumerator (SE) is a map $f:\Sigma^* \to [0,1)$ such that $\mathrm{Im}(f)$ is a separator; that is, $f$ names a countable dense subset of $[0,1)$ indexed by finite words. In foundational studies, including Mayordomo’s work, $f$ is required to be total computable and, in practice, rational-valued (that is, there is an algorithm producing a rational within $2^{-t}$ of $\Sigma$0 given $\Sigma$1) (Pulari, 1 Feb 2026).

Given $\Sigma$2 and $\Sigma$3, define the best-from-below approximation sequence: $\Sigma$4 and the scaled sequence: $\Sigma$5 For the canonical base-$\Sigma$6 system ($\Sigma$7), $\Sigma$8 and $\Sigma$9.

2. Separator Enumerators in Effective Dimension Theory

Separator enumerators are central in the generalization of classical normality and effective dimension. Mayordomo introduced $k\geq 2$0-normality—a point $k\geq 2$1 is $k\geq 2$2-normal if its relativized finite-state dimension, $k\geq 2$3, equals 1, where: $k\geq 2$4 with $k\geq 2$5 the minimal FST encoding length needed for approximate $k\geq 2$6-names within error $k\geq 2$7 (Pulari, 1 Feb 2026). This construction subsumes classical base-$k\geq 2$8 normality and enables a point-to-set principle for finite-state dimension.

The equidistribution problem—whether $k\geq 2$9-normality can be characterized by the distribution of the sequence $S\subseteq [0,1)$0—was answered negatively via explicit SEs $S\subseteq [0,1)$1 and $S\subseteq [0,1)$2 with $S\subseteq [0,1)$3 for all $S\subseteq [0,1)$4, but $S\subseteq [0,1)$5 and $S\subseteq [0,1)$6. No property of the sequence $S\subseteq [0,1)$7 (including equidistribution) can uniformly characterize $S\subseteq [0,1)$8-normality across all SEs (Pulari, 1 Feb 2026).

However, for the subclass of finite-state coherent enumerators (those defined modulo invertible synchronous Mealy machines acting on base-$S\subseteq [0,1)$9 names), $f:\Sigma^* \to [0,1)$0-normality is equivalent to $f:\Sigma^* \to [0,1)$1-adic equidistribution of $f:\Sigma^* \to [0,1)$2.

3. Enumeration Algorithms in Graph Theory

In graph-theoretic contexts, a separator enumerator refers not to a function on words, but to algorithms that enumerate minimal separators (vertex sets whose removal disconnects specified terminals). For an undirected graph $f:\Sigma^* \to [0,1)$3 and distinct $f:\Sigma^* \to [0,1)$4, a minimal $f:\Sigma^* \to [0,1)$5-separator is a vertex set $f:\Sigma^* \to [0,1)$6 that separates $f:\Sigma^* \to [0,1)$7 from $f:\Sigma^* \to [0,1)$8, and is minimal under inclusion (Korhonen, 2020).

The central result is an algorithm that, given $f:\Sigma^* \to [0,1)$9, enumerates all minimal $\mathrm{Im}(f)$0-separators $\mathrm{Im}(f)$1 with $\mathrm{Im}(f)$2 in time

$\mathrm{Im}(f)$3

for any $\mathrm{Im}(f)$4 outputs. This is both fixed-parameter-delay in $\mathrm{Im}(f)$5 and incremental-polynomial time. The algorithm combines Takata’s recursive search tree (branching on candidate additions/exclusions to a building set $\mathrm{Im}(f)$6 connected to $\mathrm{Im}(f)$7) with emptiness checks using the enumeration of important separators (at most $\mathrm{Im}(f)$8 for size $\mathrm{Im}(f)$9), achieving near-optimality barring NP-hardness of the local emptiness check (Korhonen, 2020).

4. Separator Enumeration in Permutation Statistics

Bagno et al. introduced and enumerated separators as a new statistic in permutations, $f$0 (Bagno et al., 2019). An entry $f$1 of a permutation $f$2 is a separator if its omission produces a new 2-block (bond) in the standardized $f$3-permutation. Two types are distinguished: vertical and horizontal separators, corresponding to different local patterns.

The principal enumerative results include extremal cases (permutations with zero or maximal separators), a bivariate generating function for the number of permutations with $f$4 vertical separators,

$f$5

with an explicit formula via Hadamard products and compositions, and asymptotic behavior: $f$6 No simple linear recurrence exists; all results use inclusion–exclusion and generating function techniques (Bagno et al., 2019).

5. Separator Enumeration in Document and Table Analysis

Separator enumeration plays a primary role in visual document analysis and table structure recognition, where separators are lines or boundaries demarcating structural units. For handwritten document images, algorithms directly enumerate separator rows in compressed (RLE) representations, exploiting margin run depths to identify bands corresponding to separators while handling under-/over-segmentation via adaptive splitting/insertion/deletion (R et al., 2017). Experimentally, this yields detection rates exceeding 94% on benchmarks including ICDAR13 and multiple Indic/Persian scripts.

In table structure recognition, advanced architectures such as SepFormer represent a table as a set of vertical and horizontal separators and perform direct one-shot regression of separator parameters (lines or strips) using a DETR-style coarse-to-fine transformer pipeline (Nguyen et al., 27 Jun 2025). The separator enumeration here involves filtering predicted lines by confidence, sorting, and pairing them to define the table cell grid, supporting direct adjacency matrix recovery with high computational efficiency. SepFormer achieves F1 scores up to 98.6% and processes at 25.6 FPS on common TSR benchmarks (Nguyen et al., 27 Jun 2025).

6. Comparative Table: Separator Enumeration Across Domains

Domain	Enumerated Item	Complexity/Method
Effective Dimension	Names in $f$7 via SE	Computable, often rational
Graphs	Minimal $f$8-separators	$f$9
Permutations	Separator positions	Generating function; asymptotics
Handwritten Documents	Separator rows (RLE)	$[0,1)$0 time; thresholding
Table Recognition	Separator lines/strips	DETR-style regression; grid postproc

7. Limitations, Open Questions, and Future Directions

In dimension theory, no universal equidistribution or sequence-based test for $[0,1)$1-normality exists outside finite-state coherent SEs, although $[0,1)$2-adic equidistribution suffices within that subclass (Pulari, 1 Feb 2026). In graph enumeration, NP-completeness of the separator existence test precludes straightforward polynomial-delay enumeration barring P=NP (Korhonen, 2020). For document analysis, global skew and extreme touching/indentation remain obstacles for robust separator detection (R et al., 2017). In the geometric and statistical permutations context, the lack of tractable linear recurrences suggests further analytic or bijective frameworks may be fruitful (Bagno et al., 2019). In TSR, separator regression frameworks such as SepFormer invite further extensions, e.g., to spanning non-rectilinear separators or learning domain-invariant representations (Nguyen et al., 27 Jun 2025).