Büchi Automata

- Nabarun Deka(UG Maths)

Deepak Poonia(Mtech CSA)
Overview :
• Buchi’s Motivation.
• Buchi Automata (Definition)
• Deterministic Buchi Automata and it’s characterization
• Non-deterministic Buchi automata are more powerful than Deterministic Buchi
Automata.
• Characterization of Buchi Recognizable languages.
• Closure Properties (Union, Intersection, Projection)
• MSO Theory of one successor
• Equivalence of S1S formula and regular languages.
• Complementation of Buchi Automata
Motivation :
• Bϋchi’s motivation: Decision procedure for deciding truth of first-
order logic statements about natural numbers and their ordering.
• He was interested in showing that the Monadic second order(MSO)
logic of infinite sequences was decidable.
 Bϋchi Automata :
Definition :
A Bϋchi Automaton is a finite automaton A = (Q, Σ, δ, S, F), where
Q : Finite set of states.
Σ : Finite set of input alphabet.
S : Set of Start states.
F : Set of final states.
δ ⊆ (Q × Σ × Q ) is the transition relation.
Acceptance Condition for Bϋchi
Automata :
• In any run ρ of a word α, some states of Q are visited only finite
number of times and some others are visited infinitely often. Let us
call these sets fin(ρ) and inf(ρ).
• language accepted by such an automaton is
L(A) = {α | there is a run ρ over α such that inf(ρ) ∩ F ≠ ∅}.

• A language L ⊆ is said to be buchi recognizable if it is accepted by

some Bϋchi automaton.
Example 1 :
Accepts all words over {a,b} that have infinitely many a’s.
Example 1 :
Accepts all words over {a,b} that have infinitely many a’s.
Example 1 :
Accepts all words over {a,b} that have infinitely many a’s.

Observe that this is a DBA.

Example 2 :
Accepts all words over {a,b} that have Only finitely many a’s.
Example 2 :
Accepts all words over {a,b} that have Only finitely many a’s.
Example 2 :
Accepts all words over {a,b} that have Only finitely many a’s.

Observe that this is NBA.

 Deterministic Bϋchi Automata :
• Limit languages :
Let U ⊆ Σ* be a language of finite strings. The limit of U, lim(U) is the
set
lim(U) = {α ∈ | n ∈ : α[0..n] ∈ U}
So, a word belongs to lim(U) iff it has infinitely many prefixes in U.
 Deterministic Bϋchi Automata :
• Limit languages :
Let U ⊆ Σ* be a language of finite strings. The limit of U, lim(U) is the
set
lim(U) = {α ∈ | n ∈ : α[0..n] ∈ U}

Theorem 1: A language L ⊆ is recognizable by a deterministic Bϋchi

automaton iff L is of the form lim(U) for some regular language U ⊆ Σ*.

Proof :
Proof :
(Only if part) : Treat the DBA as a DFA over finite words.
(If part) : Treat the DFA as DBA.
 NBA Vs. DBA :
• NBA has strictly more power than DBA.
• There exists languages that can be recognized by NBAs But No DBA
can recognize them.
• Example : Can we design a deterministic Buchi automaton that
accepts the set of words over {a , b} with only finite number of a’s?
 NBA Vs. DBA :
• NBA has strictly more power than DBA.
• There exists languages that can be recognized by NBAs But No DBA
can recognize them.
• Example : Can we design a deterministic Buchi automaton that
accepts the set of words over {a , b} with only finite number of a’s?

• No DBA can recognize this language.

Why?
Proof :
• Proof by contradiction.
• Assume some DBA recognizes this language L.
• So, by Theorem 1, L is of the form lim(U) for some regular language U.
• L  n0 , U.
• a L  n1, a U
• aa L  n2, aa U
• Similarly, aa…. U
• So, aa…. …. Lim(U) = L (by Theorem 1)
• Hence, Contradiction.
 Characterizing Bϋchi-recognizable
languages:
• Let U ⊆ Σ*. Then define
= {α ∈ | α = u0u1u2 ··· where ui ∈ U for all i ∈ }.
• Let U ⊆ Σ* and L ⊆ then define
UL = {α | ∃u ∈ U : ∃β ∈ L : α = uβ}.
 Characterizing Bϋchi-recognizable
languages:
• - regular expression :
G = U1 V1+ U2 V2 + …. + Un Vn
Is called regular expression, Where U1, U2 , V1, V2… are regular
expressions and L(Vi), for all 1 ≤ i ≤ n
• ω-regular languages : A language L ⊆ is said to be ω-regular if it is of
the form i∈[1..n] UiViω , where each Ui and Vi is a regular language of
finite words.
i.e. - regular language is the language described by regular expression.
Characterizing Bϋchi-recognizable
languages:
Theorem 2 : A language is Bϋchi-recognizable iff it is ω-regular.
Proof :
Theorem 2 : A language is Bϋchi-recognizable iff it is ω-regular.

(Only If part) : Let L be recognized by a Bϋchi automaton

A = (Q, Σ, δ, S, F).
• For s, s’ ∈ Q, let ss’ = {w ∈ Σ* | s w→+ s’ }
• It is easy to see that ss’ is regular.
• it follows that we can write L(A) as s∈S,f ∈ F sf .
Proof :
(If part) :
the set of Bϋchi-recognizable languages satisfies the following closure
properties:
(i) If U is regular, then Uω is Bϋchi-recognizable.
(ii) If U is regular and L is Bϋchi-recognizable then UL is Buchi
recognizable.
(iii) If L1, L2,...,Ln are Bϋchi-recognizable, so is i∈[1..n] Li .
From this, it follows that every language of the form i∈[1..n] i , where
each Ui and Vi is regular, is Bϋchi-recognizable.
• If U is regular, then Uω is Bϋchi-recognizable.
• If U is regular, then Uω is Bϋchi-recognizable.
• If U is regular and L is Bϋchi-recognizable then UL is Buchi
recognizable.
• If U is regular and L is Bϋchi-recognizable then UL is Buchi
recognizable.
• If U is regular and L is Bϋchi-recognizable then UL is Buchi
recognizable.
Closure Properties :
Buchi recognizable languages are closed under the following
operations:
• Finite Union.
• Finite Intersection.
• Projection
• Complementation
Closure under Finite Union :
Let L1 and L2 be languages over ∑⍵ , that are recognised by Büchi automata
B1 = (Q1, Σ, δ1, S1, F1) and B2 = (Q2, Σ, δ2, S2 , F2 ) respectively, then,

Recogniser for L1 U L2 is :

B = (Q1∪ Q2 , δ1 ∪ δ2 , S1∪ S2 , F1 ∪ F2 ).

It is easy to see that for every word w ∈ ∑⍵, the NBA B has an accepting
run on w iff at least one of the NBA’s B1 and B2 has an accepting run on w.
Closure under finite intersection :
• Naïve approach : We can try Product construction like we did for
Automata over finite words.
• Issue is in defining the accept states.
• Example : Let
L1 = {w | w[2n] = a, n }
L2 = {w | w[2n+1] = b, n }
Then if we do the usual product construction, it will fail.
 Construction for Intersection :

• G = S1 × G2 × {2}.

Image source : Survey report of Buchi Automata by

Madhvan Mukund, 2011.
 Construction for Intersection :
• We can verify that L(A) = L1 L2.
 Closure under Projection :
• Let 1 and 2 be two alphabets with a projection map
:12
Let L be a language over recognized by the automaton
A1 = (Q1, ,,S1,F1)
Define
(L) = {w| w[i] = (v[i]) for some v L}
Then, we can construct an automaton for (L) as
 Closure under Projection :
• A = (Q1, ,,S1,F1)
• where
(q,a,q’) iff (q,b,q’) for some b such that (b) = a
It is easy to see that L(A) = (L)
 Closure under complementation :
• The construction for complementation is highly non-trivial, so we will
visit it at the end.
 MSO theory of one successor
• Buchi’s original motivation to study automata on infinite words was to
solve a decision problem from logic.
• The logic he looked at is called the Monadic Second Order logic of
One successor (abbreviated as S1S).
• The logic is interpreted over the natural numbers (N_0). Its called a
logic of one successor because we are dealing with the natural
numbers, where each number has one unique successor.
MSO theory of one successor
Formally, the logic is defined as follows:
Terms : 0, x, y, z, succ(x), succ(succ(y)), succ(0)
Atomic Formulas: Let t,t’ be terms. The atomic formulas are
t = t’
t \in X , where X is a subset variable
Formulas: These are built from atomic formulas using Boolean operators
AND, OR and NOT, and the existential quantifier \exists. (since it’s a second
order logic, we can quantify over subsets also)

We have the usual notion of interpretation of formulas.

Examples
S1S formulas as – regular languages
Theorem:

Proof Strategy :
Step 1 : Simplify the syntax (This is purely for convenience)
Step 2 : Induct on the structure of the simplified formula –
i) Construct automatons that recognize atomic formulas
ii) Use the closure properties of ω – regular languages to
construct automatons for general formulas
Simplification of Syntax
• We remove all individual variables from the formula. Our simplified
formula has only set variables.
• Atomic formulas:
Simplification of Syntax
• Eliminate nested occurrences of successor function.
• Example : for the atomic formula

we replace it by
Simplification of Syntax
• We then eliminate the atomic formula

by replacing it with
Simplification of Syntax
• Finally, we eliminate singleton variables using the formula

For example

will be written as
Automaton for
Automaton for
Inductive step
• The inductive step follows from the closure properties of regular
languages.
Issue with sentences
• The construction works for formulas.
• If the formula is a sentence, when we project out the last free
variable, we will end up with an unlabeled directed graph. We just
check if there is an unlabelled path that starts from a start state and
visits a final state infinitely often.
• This is equivalent to checking if the graph has a strongly connected
component X which contains a final state f, and is reachable from
some start state.
ω- regular languages are S1S definable
ω- regular languages are S1S definable
• Theorem
Complementing deterministic Buchi
automata
Complementing non deterministic Buchi
automata
https://www.react.uni-saarland.de/teaching/automata-games-verificati
on-08/notes/notes3.pdf

https://www.react.uni-saarland.de/teaching/automata-games-verificati
on-08/notes/notes4.pdf
References :
• https://www.cmi.ac.in/~sri/Courses/MCSV/Slides/lecture6.pdf
• https://www.cmi.ac.in/~kumar/words/lecture07.pdf
• https://www.cmi.ac.in/~madhavan/papers/pdf/iisc2011-buchi.pdf
• https://nptel.ac.in/courses/106/106/106106136/