Prefix Normal Words in Binary Strings

• Prefix normal words are binary words whose every prefix exhibits the maximal count of 1s relative to all factors of the same length.
• They provide unique canonical representations that streamline binary jumbled pattern matching by aligning factor and prefix Parikh profiles.
• Their study leads to efficient algorithms, including Gray-code generation and fixed-length word chains, with extensions to weighted and infinite cases.

Searching arXiv for recent and foundational work on prefix normal words to ground the article in cited papers. Prefix normal words are binary words whose prefixes are extremal with respect to the number of $1$s among all factors of the same length. Introduced by Fici and Lipták in 2011 in the context of binary jumbled pattern matching, they supply canonical representatives of factor-Parikh-vector profiles and have become a meeting point for combinatorics on words, formal-language theory, enumeration, Gray-code generation, and several generalizations, including infinite and weighted variants (Adamson et al., 27 Aug 2025, Burcsi et al., 2016).

1. Definition and basic formalism

Let $w=w_1w_2\cdots w_n\in\{0,1\}^n$. The standard prefix profile is

$P_w(k)=\sum_{i=1}^k w_i,$

the number of $1$s in the prefix of length $k$. The extremal factor profile is written in the literature as $F_w(k)$, $M_w(k)$, or $f_w(k)$, and records the maximum number of $1$s in any factor of length $k$: $w=w_1w_2\cdots w_n\in\{0,1\}^n$0 A binary word is prefix normal, or more precisely $w=w_1w_2\cdots w_n\in\{0,1\}^n$1-prefix normal, if and only if

$w=w_1w_2\cdots w_n\in\{0,1\}^n$2

Equivalently, no factor of length $w=w_1w_2\cdots w_n\in\{0,1\}^n$3 contains more $w=w_1w_2\cdots w_n\in\{0,1\}^n$4s than the prefix of length $w=w_1w_2\cdots w_n\in\{0,1\}^n$5. A dual notion of $w=w_1w_2\cdots w_n\in\{0,1\}^n$6-prefix normality is defined analogously by replacing counts of $w=w_1w_2\cdots w_n\in\{0,1\}^n$7s with counts of $w=w_1w_2\cdots w_n\in\{0,1\}^n$8s (Burcsi et al., 2016, Adamson et al., 27 Aug 2025).

This definition has several immediate consequences. Since the prefix itself is a factor, the inequality form $w=w_1w_2\cdots w_n\in\{0,1\}^n$9 is equivalent to equality in the finite case. The language is prefix-closed: every prefix of a prefix normal word is itself prefix normal. It is also closed under appending $P_w(k)=\sum_{i=1}^k w_i,$0, and more generally, if $P_w(k)=\sum_{i=1}^k w_i,$1 is prefix normal then $P_w(k)=\sum_{i=1}^k w_i,$2 and $P_w(k)=\sum_{i=1}^k w_i,$3 are prefix normal for every $P_w(k)=\sum_{i=1}^k w_i,$4 (Burcsi et al., 2016).

Concrete examples are standard. The word $P_w(k)=\sum_{i=1}^k w_i,$5 is prefix normal because its prefix counts $P_w(k)=\sum_{i=1}^k w_i,$6 match the maximal factor counts at every length. By contrast, $P_w(k)=\sum_{i=1}^k w_i,$7 is not prefix normal: the factor $P_w(k)=\sum_{i=1}^k w_i,$8 has two $P_w(k)=\sum_{i=1}^k w_i,$9s, whereas the prefix $1$0 of the same length has only one (Adamson et al., 27 Aug 2025).

2. Prefix normal forms and Parikh-vector structure

A central theorem is that every binary word has a unique prefix normal form. For any $1$1, there exists a unique $1$2-prefix normal word $1$3 such that its prefix profile equals the maximal factor profile of $1$4: $1$5 The bits of $1$6 are determined by first differences: $1$7 An analogous form $1$8 is defined from maximal $1$9-counts. A word is prefix normal exactly when it coincides with its own prefix normal form (Burcsi et al., 2016).

The significance of these forms is most visible in binary jumbled pattern matching. If $k$0 denotes the set of Parikh vectors of all factors of $k$1, then at each fixed length $k$2 the attainable numbers of $k$3s form an interval. More precisely, if $k$4 and $k$5 are the minimum and maximum numbers of $k$6s in a factor of length $k$7, then every intermediate value occurs. Consequently, two words have the same factor-Parikh set if and only if both of their prefix normal forms agree: $k$8 This is the basic canonical-representation theorem for binary Parikh profiles (Burcsi et al., 2016).

The indexed binary jumbled pattern matching problem reduces directly to these profiles. Given a query Parikh vector $k$9 with $F_w(k)$0, the answer is positive exactly when

$F_w(k)$1

Equivalently, with the two prefix normal forms precomputed,

$F_w(k)$2

This yields a linear-size index with $F_w(k)$3-time queries once the extremal profiles are known; the best known preprocessing time cited in this line of work is $F_w(k)$4 (Burcsi et al., 2016).

3. Local obstructions and structural characterizations

Much of the fine structure of prefix normality is carried by minimal violating factors. If $F_w(k)$5 is a factor of minimum length such that $F_w(k)$6 exceeds the number of $F_w(k)$7s in the prefix of the same length, then $F_w(k)$8 begins and ends with $F_w(k)$9, and it exceeds the corresponding prefix by exactly one $M_w(k)$0: $M_w(k)$1 This makes minimal witnesses highly rigid: they are not arbitrary dense factors but tight, boundary-aligned obstructions (Adamson et al., 27 Aug 2025).

The first run of $M_w(k)$2s plays a special role. If $M_w(k)$3, then no factor starting inside that first $M_w(k)$4-run can witness non-prefix-normality. More generally, any non-prefix-normal word has a witness factor that starts and ends with blocks of $M_w(k)$5s, which localizes the search space to run boundaries rather than arbitrary substrings. This sharply refines the naive view that every factor is equally relevant (Adamson et al., 27 Aug 2025).

These observations produce algorithmic constraints. The 2025 analysis states that only factors starting and ending with $M_w(k)$6 need to be inspected, and that factors overlapping the first $M_w(k)$7-run may be excluded. The total number of such candidate factors depends only on $M_w(k)$8 and is stated as $M_w(k)$9, yielding a substantial reduction when the density of $f_w(k)$0s is small. A practical membership test therefore enumerates only these factors, skips those overlapping the first $f_w(k)$1-run, and stops at the first length for which a factor exceeds the prefix by exactly one $f_w(k)$2 (Adamson et al., 27 Aug 2025).

Equivalent global formulations also exist. In position-of-$f_w(k)$3 notation, with $f_w(k)$4 denoting the position of the $f_w(k)$5-th $f_w(k)$6, prefix normality is equivalent to inequalities such as

$f_w(k)$7

This places the theory simultaneously in a factor-count language, a positional language, and a run-length language (Burcsi et al., 2016).

4. Equivalence classes, palindromes, and asymptotic enumeration

Prefix normality induces an equivalence relation by extremal factor profiles: two words are prefix normal equivalent when their maximum-ones functions coincide. Each equivalence class contains exactly one prefix normal representative, and the prefix normal form is the lexicographically largest element of its class, while the corresponding suffix-normal form is the lexicographically smallest least representative (Fleischmann et al., 2019).

Palindromes occupy a distinguished position inside this quotient structure. A word has a singleton equivalence class if and only if it is a prefix normal palindrome. The same work gives a functional characterization: for a nonzero binary word $f_w(k)$8, being a prefix normal palindrome is equivalent to

$f_w(k)$9

where $1$0 is the reverse-progress transform of the maximum-ones function. This links symmetry of the word to symmetry of its extremal factor profile rather than to direct symbolwise constraints (Fleischmann et al., 2019).

The language-theoretic picture is unusually rich. The language of prefix normal words is not context-free, and it is strictly contained in the language of pre-necklaces. This situates prefix normality between tractable combinatorial constructions and genuinely non-context-free formal-language behavior (Burcsi et al., 2016).

Enumeration remains one of the main open fronts. Balister and Gerke proved that the number $1$1 of prefix normal words of length $1$2 satisfies

$1$3

establishing that prefix normal words are exponentially many but asymptotically sparse among all binary words. Earlier bounds already showed, for sufficiently large $1$4,

$1$5

For fixed density $1$6, however, the ordinary generating function is rational, which contrasts with the absence of a closed form or generating function for the total count $1$7 (Balister et al., 2019, Burcsi et al., 2016, Adamson et al., 27 Aug 2025).

5. Testing, Gray codes, and word-chain generators

The most direct recognition algorithm computes all values $1$8 and compares them to the prefix counts, which is $1$9 by naive sliding-window evaluation. Much of the algorithmic literature consists of avoiding this full quadratic scan through structural constraints. The factor-centric reduction described above is one such refinement; another major development is the realization that prefix normal words form a bubble language (Adamson et al., 27 Aug 2025, Burcsi et al., 2020).

A language is a first-$k$0 bubble language if swapping the first occurrence of $k$1 to $k$2 preserves membership. Prefix normal words satisfy exactly this property. The consequence is a Gray-code generation scheme: all prefix normal words of length $k$3 can be listed so that successive outputs differ by at most two swaps, or by one swap and one bit flip, and this Gray code can be generated in amortized $k$4 time per word. Earlier generation by Bubble-Flip gave $k$5 time per word and lexicographic output via a recursive tree based on the two operations “bubble” and “flip” (Burcsi et al., 2020, Cicalese et al., 2017).

The 2025 paper introduces a different viewpoint: fixed-length word chains generated by permutations. A generator is a permutation $k$6 of $k$7 prescribing a sequence of $k$8 flips,

$k$9

which defines a path from $w=w_1w_2\cdots w_n\in\{0,1\}^n$00 to $w=w_1w_2\cdots w_n\in\{0,1\}^n$01 in the Hamming cube. The permutation is a prefix normal generator if every intermediate word in the chain is prefix normal. This framework yields strong necessary conditions, including

$w=w_1w_2\cdots w_n\in\{0,1\}^n$02

together with a local theory of adjacent swaps. If $w=w_1w_2\cdots w_n\in\{0,1\}^n$03, the adjacent swap is always safe; if $w=w_1w_2\cdots w_n\in\{0,1\}^n$04, safety is governed by a precise local factor criterion in the intermediate chain word. The worst-case number of candidate factors to inspect is $w=w_1w_2\cdots w_n\in\{0,1\}^n$05, bounded above by $w=w_1w_2\cdots w_n\in\{0,1\}^n$06, and the minimal-witness restrictions prune this further (Adamson et al., 27 Aug 2025).

These chain methods also interact with extension by length. Appending $w=w_1w_2\cdots w_n\in\{0,1\}^n$07 preserves prefix normality, whereas appending $w=w_1w_2\cdots w_n\in\{0,1\}^n$08 is structurally delicate. In particular, a prefix normal palindrome with at least one $w=w_1w_2\cdots w_n\in\{0,1\}^n$09 becomes non-prefix-normal upon appending $w=w_1w_2\cdots w_n\in\{0,1\}^n$10. The same work proves that successful append-$w=w_1w_2\cdots w_n\in\{0,1\}^n$11 operations are not isolated: if $w=w_1w_2\cdots w_n\in\{0,1\}^n$12 is prefix normal, then some prefix normal parent $w=w_1w_2\cdots w_n\in\{0,1\}^n$13 satisfies that $w=w_1w_2\cdots w_n\in\{0,1\}^n$14 is also prefix normal. This supports recursive enumeration strategies from length $w=w_1w_2\cdots w_n\in\{0,1\}^n$15 to length $w=w_1w_2\cdots w_n\in\{0,1\}^n$16 (Adamson et al., 27 Aug 2025).

6. Infinite, weighted, and cross-disciplinary extensions

The finite theory extends nontrivially to infinite words. An infinite binary word is prefix normal if and only if every finite prefix is prefix normal. Within Sturmian words, the characterization is exact: a Sturmian word of slope $w=w_1w_2\cdots w_n\in\{0,1\}^n$17 is prefix normal if and only if it is the upper mechanical word with intercept $w=w_1w_2\cdots w_n\in\{0,1\}^n$18, namely

$w=w_1w_2\cdots w_n\in\{0,1\}^n$19

For infinite words, prefix normal forms $w=w_1w_2\cdots w_n\in\{0,1\}^n$20 and $w=w_1w_2\cdots w_n\in\{0,1\}^n$21 still encode the extremal factor counts, and the abelian complexity satisfies

$w=w_1w_2\cdots w_n\in\{0,1\}^n$22

This makes abelian complexity the width between the two infinite prefix normal forms (Cicalese et al., 2018).

A different generalization replaces binary counting with weights over arbitrary finite alphabets. In weighted prefix normality, letters carry values in a strictly totally ordered commutative monoid, and one compares maximal factor weights with prefix weights of the same length. This framework is strictly more expressive than the binary one. Its distinctive obstruction is the presence of gaps: a weight measure may fail to realize the increment from $w=w_1w_2\cdots w_n\in\{0,1\}^n$23 to $w=w_1w_2\cdots w_n\in\{0,1\}^n$24 by a single letter weight. The existence and uniqueness of a weighted prefix normal form for every word hold exactly when the measure is gapfree and injective. Among non-binary injective alphabetically ordered measures, all gapfree measures are equivalent to the standard measure $w=w_1w_2\cdots w_n\in\{0,1\}^n$25, which provides a canonical generalized prefix normal form (Eikmeier et al., 2020).

Prefix normal words also arise outside string algorithms. For caterpillar graphs, the discrete derivative of the leaf function,

$w=w_1w_2\cdots w_n\in\{0,1\}^n$26

is binary, and a sequence $w=w_1w_2\cdots w_n\in\{0,1\}^n$27 with $w=w_1w_2\cdots w_n\in\{0,1\}^n$28, $w=w_1w_2\cdots w_n\in\{0,1\}^n$29, $w=w_1w_2\cdots w_n\in\{0,1\}^n$30, and $w=w_1w_2\cdots w_n\in\{0,1\}^n$31 is realizable by a caterpillar if and only if $w=w_1w_2\cdots w_n\in\{0,1\}^n$32 is a prefix normal word. The reading-caterpillar map $w=w_1w_2\cdots w_n\in\{0,1\}^n$33 furnishes an explicit inverse construction from a prefix normal word to a caterpillar realizing it (Massé et al., 2017).

Several problems remain open. These include exact enumeration and finer asymptotics beyond $w=w_1w_2\cdots w_n\in\{0,1\}^n$34, the number and size distribution of prefix-normal-equivalence classes, the precise structure and connectivity of the generator graph on word-chain generators, and faster-than-quadratic computation of prefix normal forms, which would immediately affect indexed binary jumbled pattern matching. The 2025 word-chain framework suggests additional directions, including weighted analogues of generators and a more group-theoretic understanding of admissible flip permutations (Adamson et al., 27 Aug 2025).