text
stringlengths
128
2.05k
[MATH] is then the set of [MATH] for which the upper endpoint of [MATH] equals the lower endpoint of [MATH] . By Lemma 6.4 [MATH] contains one of [MATH] [MATH] . Without loss of generality assume [MATH] . Let
[MATH] and observe that for each [MATH] there is an [MATH] such that [MATH] and [MATH] , and thus [EQUATION] If [MATH] is an irreducible representation with [MATH] then
[MATH] and so the range of [MATH] is an invariant subspace of [MATH] , whence, one of [MATH] . If [MATH] then [EQUATION] while if [MATH] then
[EQUATION] Theorem 6.6 Let [MATH] in [MATH] . Then there is a free ultrafilter [MATH] and a sequence of pairwise orthogonal finite-rank intervals [MATH] such that
[MATH] and [MATH] contains [EQUATION] Moreover, given any decomposition of the [MATH] as the sums of intervals [MATH] we can replace [MATH] with one of [MATH] or [MATH]
Proof. The existence of the intervals follows from Proposition 6.3 Let [MATH] be the ultrafilter obtained in Lemma 6.4 If [MATH] then, given [MATH] , there is a [MATH]
such that [MATH] for all [MATH] . Thus taking [MATH] , we see that [MATH] and that [MATH] , whence [MATH] . Thus [MATH] is a limit point of [MATH]
and since [MATH] is norm closed, [MATH] Given a decomposition [MATH] , we know from Proposition 6.5 that one of [MATH] [MATH] is in [MATH] . Without loss suppose [MATH] . Again by Lemma 6.4
[MATH] is an ultrafilter. Now let [MATH] . Since [MATH] is an ultrafilter, one of [MATH] But if [MATH] then [EQUATION] which is impossible. Thus [MATH] and so, since [MATH] was arbitrary,
[MATH] . But [MATH] is also an ultrafilter, so in fact [MATH] . Thus we may replace [MATH] with [MATH] Now it follows that the limit of the ranks of the intervals must be [MATH] , for otherwise after finitely many decompositions we could conclude that [MATH] and so
[MATH] . Similarly if [MATH] were not free then [MATH] would contain [MATH] for some [MATH] and, after finitely many decompositions if necessary, we would see that [MATH] for some [MATH] again contrary to hypothesis.
Acknowledgements The author gratefully acknowledges the hospitality of Professor Tony Carbery and the University of Edinburgh Mathematics Department.
# Source: arxiv 1806.04645 # Title: State Complexity of Pattern Matching in Regular Languages # Sections: all # Downloaded: 2026-03-03T02:40:44.409496+00:00
State Complexity of Pattern Matching in Regular Languages Abstract In a simple pattern matching problem one has a pattern [MATH] and a text [MATH] , which are words over a finite alphabet [MATH] One may ask whether [MATH] occurs in [MATH] , and if so, where? More generally, we may have a set [MATH] of patterns and a se...
[MATH] [MATH] , and [MATH] , where [MATH] is the shuffle operation. The state complexity [MATH] of a regular language [MATH] is the number of states in the minimal deterministic finite automaton recognizing [MATH] We derive the following upper bounds on the state complexities of our pattern-matching languages, where
[MATH] , and [MATH] [MATH] [MATH] [MATH] and [MATH] We prove that these bounds are tight, and that to meet them, the alphabet must have at least two letters in the first three cases, and at least [MATH] letters in the last case. We also consider the special case where [MATH] is a single word [MATH] , and obtain the fol...
[MATH] [MATH] [MATH] and [MATH] For unary languages, we have a tight upper bound of [MATH] in all eight of the aforementioned cases.
keywords: all-sided ideal, combined operation, factor, finite automaton, left ideal, pattern matching, prefix, regular language, right ideal, state complexity, subsequence, suffix, two-sided ideal
Introduction Given a regularity-preserving operation on regular languages, we may ask the following natural question: in the worst case, how many states are necessary and sufficient for a deterministic finite automaton (DFA) to accept the language resulting from the operation, in terms of the number of states of the in...
that an [MATH] -state DFA is also necessary in the worst case; for all [MATH] , there exist a language accepted by an [MATH] -state DFA and a language accepted by an [MATH] -state DFA whose intersection is accepted by a minimal DFA with [MATH] states.
This worst-case value is called the state complexity of the operation. The state complexity of a regular language [MATH] , denoted by [MATH] , is the number of states in the minimal DFA accepting [MATH] . Thus [MATH] means the minimal DFA for [MATH] has exactly [MATH] states, and [MATH] means [MATH] can be recognized b...
[MATH] and [MATH] , then [MATH] and this bound is tight for all [MATH] Aside from “basic” operations like union, intersection, concatenation and star, the state complexity of combined operations
such as “star of intersection” and “star of union” has also been studied. We investigate the state complexity of new combined operations inspired by pattern matching problems.
For a comprehensive treatment of pattern matching, see In a pattern matching problem we have a text and a pattern In its simplest form, the pattern [MATH] and the text [MATH] are both words over an alphabet [MATH] Some natural questions about patterns in texts include the following: Does [MATH]
occur in [MATH] and if so, where? Pattern matching has many applications. Aho and Corasick developed an algorithm to determine all occurrences of words from a finite pattern in a given text; this algorithm leads to significant improvements in the speed of bibliographic searches. Pattern matching is used in bioinformati...
in this context the text [MATH] is often a DNA sequence, and the pattern [MATH] is a sequence of nucleotides searched for in the text.
More generally, we can have a set [MATH] of patterns and a set [MATH] of texts. These could be finite sets, or they could be arbitrary regular languages, specified by a finite automaton or a regular expression. For example, many text editors and text processing utilities have a regular expression search feature, which ...
In this paper, we ask whether a pattern from the set [MATH] occurs as a prefix suffix factor or subsequence of a text from the set [MATH] If [MATH] and [MATH] , then [MATH] is a prefix of [MATH] and [MATH] is a suffix of [MATH] If [MATH] for some [MATH] , then [MATH] is a factor of [MATH] If [MATH] , where [MATH] , and...
of [MATH] If [MATH] is any language, then [MATH] is the right ideal [MATH] [MATH] is the left ideal [MATH] , and [MATH] is the two-sided ideal
[MATH] The shuffle [MATH] of words [MATH] is defined as follows: [EQUATION] The shuffle of two languages [MATH] and [MATH] over [MATH] is defined by
[EQUATION] The language [MATH] is an all-sided ideal The language [MATH] consists of all words that contain [MATH] as a subsequence. Such a language could be used, for example, to determine whether a report has all the required sections and that they are in the correct order.
The combined operations we consider are of the form “the intersection of [MATH] with the right (left, two-sided, all-sided) ideal [MATH] ”. We study four problems with pattern sets [MATH] and text sets [MATH]
1. Find [MATH] , the set of all the words in [MATH] each of which begins with a word in [MATH] 2. Find [MATH] , the set of all the words in [MATH] each of which ends with a word in [MATH]
3. Find [MATH] the set of all the words in [MATH] each of which has a word in [MATH] as a factor. 4. Find [MATH] , the set of all the words in [MATH] each of which has a word of [MATH] as a subsequence.
We then repeat these four problems for the case where the pattern is a single word [MATH] In all eight cases we find the state complexity of these operations. We show that for languages [MATH] [MATH] and [MATH] such that [MATH] [MATH] [MATH] the following upper bounds hold:
1. General case: (a) Prefix: [MATH] (b) Suffix: [MATH] (c) Factor: [MATH] (d) Subsequence: [MATH] 2. Single-word case: (a) Prefix:
[MATH] (b) Suffix: [MATH] (c) Factor: [MATH] (d) Subsequence: [MATH] Moreover, in each case there exist languages [MATH] [MATH] [MATH] that meet the upper bounds.
In the general prefix, suffix and factor cases, there exist binary witnesses meeting the bounds. For the general subsequence case, an alphabet of at least [MATH] letters is needed to reach this bound. For the single-word cases we use binary witnesses to reach each of the bounds.
In Section 7.1 , we consider prefix matching in the case where [MATH] is a single word. In addition to considering arbitrary alphabets, in that section we also look at the case where [MATH] and [MATH] are languages over a unary alphabet We prove a tight upper bound of [MATH] on the state complexity of [MATH] in the una...
Terminology and Notation deterministic finite automaton (DFA) is a 5-tuple [MATH] , where [MATH] is a finite non-empty set of states
[MATH] is a finite non-empty alphabet [MATH] is the transition function [MATH] is the initial state, and [MATH] is the set of final states. We extend [MATH] to a function [MATH] inductively as follows: for [MATH] , define [MATH] , and for [MATH] and [MATH] , define [MATH] We extend it further to [MATH] by setting [MATH...
accepts a word [MATH] if [MATH] . The language accepted by [MATH] is the set of all words that [MATH] accepts, and is denoted by [MATH] . If [MATH] is a state of [MATH] , then the language [MATH] of [MATH] is the language accepted by the DFA [MATH] A state is empty (or dead or a sink state ) if its language is empty. T...
indistinguishable if [MATH] A state [MATH] is reachable if there exists [MATH] such that [MATH] A DFA [MATH] is minimal if it has the smallest number of states and the smallest alphabet among all DFAs accepting [MATH] It is well known that a DFA is minimal if it uses the smallest alphabet, all of its states are reachab...
nondeterministic finite automaton (NFA) is a 5-tuple [MATH] , where [MATH] is now a function [MATH] , and all other components are as in a DFA. Extending [MATH] to a function [MATH] , the NFA [MATH] accepts a word [MATH] if [MATH] . As with DFAs, the language accepted by the NFA [MATH] is the set of all accepted words.
Let [MATH] be a language over [MATH] . The quotient of [MATH] by a word [MATH] is the set [MATH] . In a DFA [MATH] , if [MATH] , then [MATH]
transformation of a set [MATH] is a function [MATH] The image of [MATH] under the transformation [MATH] is denoted by [MATH] If [MATH] are transformations of [MATH] , their composition is denoted by [MATH] and defined by
[MATH] ; that is, composition is performed from left to right The preimage of [MATH] under the transformation [MATH] is denoted by [MATH] , and is defined to be the set [MATH] This notation extends to sets: for [MATH] , we have [MATH] and [MATH]
For [MATH] , a transformation [MATH] of a set [MATH] is a [MATH] -cycle if [MATH] [MATH] , …, [MATH] [MATH] and [MATH] for all [MATH] This [MATH] -cycle is denoted by [MATH] A 2-cycle [MATH] is a transposition The identity transformation of [MATH] is denoted by [MATH] ; while this notation omits the set [MATH] , it can...
In a DFA [MATH] , each letter [MATH] induces a transformation of the set of states [MATH] , defined by [MATH] for [MATH] We denote this transformation by [MATH] Specifying the transformation [MATH] induced by each letter [MATH] completely specifies the transition function [MATH] , so we often define [MATH] in this way....
[MATH] means the transformation induced by [MATH] in the DFA [MATH] is the cycle [MATH] We extend the [MATH] notation from letters to words: if [MATH] for [MATH] , then [MATH]
dialect of a regular language [MATH] is a language obtained from [MATH] by replacing or deleting letters of [MATH] in the words of [MATH] In this paper we use only dialects obtained by permuting the letters of [MATH] Thus, for example, if [MATH] , then [MATH] The notion of a dialect is also extended to DFAs.
Henceforth we sometimes refer to state complexity as simply complexity , since we do not discuss other measures of complexity in this paper.
Prefix Matching Let [MATH] and [MATH] be regular languages over an alphabet [MATH] We compute the set [MATH] of all the words of [MATH] that are prefixed by words in [MATH] ; that is, the language [MATH] We want to find the worst-case state complexity of [MATH]
Theorem 1 For [MATH] if [MATH] and [MATH] then [MATH] , and this bound is tight if the cardinality of [MATH] is at least 2. Proof 1
The language [MATH] is the right ideal [MATH] . It is known that the state complexity of [MATH] is at most [MATH] Furthermore, it was shown in
that the complexity of intersection is at most [MATH] if the first input has complexity at most [MATH] and the second has complexity at most [MATH] Hence [MATH] is an upper bound on the complexity of [MATH]
Next we find witnesses that meet this bound. Let [MATH] be accepted by the DFA [MATH] , where [MATH] [MATH] and [MATH] is defined by the transformations [MATH] , and
[MATH] see Figure This DFA is minimal because the shortest word in [MATH] accepted by state [MATH] is [MATH] ; this shortest word distinguishes [MATH] from any other state.
Now let [MATH] be the dialect of [MATH] with the roles of [MATH] and [MATH] interchanged. Thus the DFA [MATH] , where [MATH] and [MATH] is defined by
[MATH] [MATH] , is the minimal DFA of [MATH] This DFA is minimal because any state is distinguished from any other state by the shortest word in
[MATH] that it accepts. To find [MATH] , we concatenate the language [MATH] with the language [MATH] Note that once state [MATH] is reached in the DFA [MATH] recognizing [MATH] , every word is accepted. Thus the transition from [MATH] to 0 is not needed, because it is replaced by a self-loop on state [MATH] under [MATH...
[MATH] of Figure , where [MATH] is defined by [MATH] [MATH] Our last task is to find a DFA accepting [MATH] and prove that it is minimal and has [MATH] states. To achieve this we find the direct product [MATH] of [MATH] and [MATH] ; an example of this product for
[MATH] is given in Figure Let [MATH] , where [MATH] Since DFA [MATH] has [MATH] states, it remains to prove that every state if [MATH] is reachable and every two states are distinguishable.
We observe that [MATH] , for all [MATH] and [MATH] . Therefore, every state is reachable. We also observe that the minimal word in [MATH]
accepted by a state [MATH] is [MATH] , where [MATH] and [MATH] . Therefore, each state in [MATH] has a unique minimal word in [MATH] this makes all states pairwise distinguishable. Hence, [MATH] is minimal, and has state complexity [MATH]
\qed Suffix Matching Let [MATH] and [MATH] be regular languages over an alphabet [MATH] We are now interested in the worst-case state complexity of the set [MATH] of all the words of [MATH] that end with words in [MATH] . More formally, [MATH]
Proposition 2 For [MATH] if [MATH] and [MATH] then [MATH] Proof 2 The language [MATH] is the left ideal [MATH] . It is known that the state complexity of this ideal is at most [MATH]
Furthermore, the complexity of intersection is at most the product of the state complexities of the two operands. Hence [MATH] is an upper bound on the complexity of [MATH]
\qed Our next goal is to prove that this upper bound is tight. We describe witnesses [MATH] and [MATH] that meet the upper bound. Let [MATH] be accepted by the DFA [MATH] , where [MATH] [MATH] and [MATH] is defined by the transformations [MATH] [MATH] See Figure This DFA is minimal because the shortest word in [MATH] a...
It turns out that [MATH] , and its dialect [MATH] shown in Figure , act as witnesses in the case of suffix matching. We denote the language of the DFA [MATH]
by [MATH] Let [MATH] Then [MATH] can be described by the regular expression [MATH] Now we have [EQUATION] The new generator of the left ideal [MATH] is
[MATH] It consists of any word of length [MATH] beginning with [MATH] , possibly followed by any number of words of length [MATH] beginning with [MATH] An NFA accepting the left ideal is shown in Figure
Before we prove that the bound of Proposition is tight, we need a different characterization of the language [MATH] To describe a DFA for this language, we will use binary [MATH] -tuples which we denote by [MATH]
Definition 3 Define the following DFA: [EQUATION] where [EQUATION] In other words, the input [MATH] shifts the tuple [MATH] one position to the left cyclically, while [MATH] shifts the tuple to the left, losing the first component and replacing [MATH] by 1. DFA [MATH] is shown in Figure
Proposition 4 All the states of [MATH] are reachable and pairwise distinguishable. Proof 3 Consider a state [MATH] , and view it as the binary representation of a number [MATH] State [MATH] is reachable by [MATH] and [MATH] by [MATH] If [MATH] is even, it is reachable from [MATH] by [MATH] , and [MATH] is reachable fro...
We note that if a state [MATH] has [MATH] , then [MATH] accepts the word [MATH] For each state [MATH] , define [MATH] Since each state has a unique binary representation, each state has a unique [MATH] , which is a subset of all words accepted by [MATH] Therefore, if [MATH] and [MATH] are distinct states, they are pair...
\qed In the example of Figure , we have [MATH] [MATH] [MATH] [MATH] [MATH] [MATH] [MATH] Recall that the left ideal [MATH] is [MATH] , that is, [MATH]
Lemma 5 DFA [MATH] is isomorphic to the minimal DFA of [MATH] Proof 4 First we prove that each state [MATH] of [MATH] accepts [MATH] , and thus [MATH] accepts a superset of
[MATH] Let [MATH] be an arbitary word from [MATH] Since [MATH] begins with [MATH] , this letter “loads” a 1 into position [MATH] Then this [MATH] is followed by [MATH] arbitrary letters, which shift the 1 into position [MATH] If there is no more input, the word [MATH] is accepted. Otherwise, the next letter is an [MATH...
Next we prove that [MATH] If [MATH] , then [MATH] has length at least [MATH] Let [MATH] , where [MATH] Consider the prefix [MATH] of [MATH] If [MATH] , then [MATH]
is in [MATH] and we are done. If [MATH] , then [MATH] is in [MATH] Now our proof strategy is as follows: jump back [MATH] letters and look at [MATH] If this letter is a [MATH] , then [MATH]
is in [MATH] and we are done. If it’s an [MATH] , then [MATH] is in [MATH] , and we can keep jumping back [MATH] letters at a time until we find a [MATH]
More formally, we claim there exists [MATH] such that [MATH] , and for [MATH] we have [MATH] ; thus [MATH] is in [MATH] , and we are done.
To see this, suppose the above claim is false. We can write [MATH] , where [MATH] is the quotient upon dividing [MATH] by [MATH] , and [MATH] is the remainder with [MATH] Since the claim is false, we have [MATH] In fact, we have [MATH] for [MATH] It follows that [MATH]
is in [MATH] Since [MATH] , the prefix [MATH] cannot lead to an accepting state. Now, if we are in a non-accepting state, and we apply a word from the language [MATH] , we will remain in a non-accepting state. Thus [MATH] is not accepted, which is a contradiction. So the claim must be true, and this completes the proof...
\qed To finally prove that [MATH] meets the bound [MATH] , we construct the direct product of the DFAs [MATH] and [MATH] We show that all [MATH] states in the direct product are reachable and pairwise distinguishable.
We will use the following lemma in the proof of reachability: Lemma 6 If (a) DFAs [MATH] and [MATH] are minimal DFAs, (b) [MATH]
is bijective on [MATH] for all [MATH] , and (c) every state in [MATH] is reachable in the direct product of the DFAs [MATH] , then every state in [MATH] is reachable in [MATH]
Proof 5 Suppose every state in [MATH] is reachable. We will show that [MATH] is reachable for all [MATH] and [MATH] Let [MATH] be a word over [MATH] that such that
[MATH] such a word exists since [MATH] is minimal. Since [MATH] is bijective for all [MATH] the transformation [MATH] is bijective and hence has an inverse. So we may reach [MATH] by first reaching
[MATH] and then applying [MATH] \qed We can now prove the following theorem: Theorem 7 For [MATH] if [MATH] and [MATH] then [MATH] , and this bound is tight if the cardinality of [MATH] is at least 2.
Proof 6 The upper bound follows from Proposition To prove that the upper bound is tight, we show that all states in the direct product [MATH] are reachable and pairwise distinguishable.
Reachability. Let [MATH] denote the state set of [MATH] and let [MATH] denote the initial state of [MATH] The initial state of the direct product is [MATH] Every state of the form [MATH] , where [MATH] , is reachable by [MATH] We observe that [MATH] and [MATH] (where [MATH] is the transition function of [MATH] ) are bo...
Distinguishability. In this part of the proof, to simplify the notation, we simply write [MATH] for the transformation induced by [MATH] in the appropriate DFA. For example, if [MATH] , then [MATH] is equivalent to [MATH] or [MATH]
First note the following facts about [MATH] 1. The word [MATH] sends all states to the final state [MATH] 2. The final state [MATH] is fixed by all words in [MATH]
3. The letter [MATH] permutes the states. Thus if [MATH] and [MATH] are distinct, then [MATH] and [MATH] are distinct. 4. Suppose [MATH] and [MATH] are states, and define [MATH] to be the largest integer [MATH] such that [MATH] , or [MATH] if the states are equal. If [MATH] , then [MATH] and [MATH] are distinguishable ...
5. If [MATH] , then [MATH] sends [MATH] to a state [MATH] and [MATH] to a state [MATH] such that [MATH] Now, let [MATH] and [MATH] be distinct states, where [MATH]
Case 1. [MATH] Without loss of generality, we can assume [MATH] ; otherwise apply [MATH] . Choose a word [MATH] that distinguishes [MATH] and [MATH] in [MATH] ; then [MATH] distinguishes [MATH] and [MATH]
Case 2. [MATH] (and thus [MATH] ). We may assume without loss of generality that [MATH] and [MATH] differ in exactly one component, and that [MATH] Otherwise, first apply [MATH] to reach [MATH] and [MATH] such that [MATH] , and note that this implies [MATH] and [MATH] differ in exactly one component. Then apply [MATH] ...
Suppose now that [MATH] and [MATH] differ in exactly one component and [MATH] Then [MATH] is the index of the component where [MATH] and [MATH] differ. Furthermore, if we apply a word [MATH] , then either [MATH] , or [MATH] and [MATH] differ in exactly one component and [MATH] is the index of this component. So as long...