📝

1 Previously, in Transducers

1.1 Aperiodicity and Counters

Let \(L\) be a regular language. Prove the equivalence between the following properties.

  1. The minimal DFA of \(L\) is counter-free.
  2. The syntactic monoid of \(L\) is aperiodic.

Assume that \(L\) is recognised by a counter-free automaton (that may not be minimal), is \(L\) aperiodic? What about a non-deterministic counter-free automaton?

Hint: Use the transition monoid

Prove that the transition monoid of the minimal DFA of \(L\) is the syntactic monoid of \(L\).

Hint: Non-deterministic counter-free automaton

Use the transition monoid to define what a counter should be.

Solution: From aperiodicity to counter-freeness

Let us assume that \(A = (Q, q_0, \delta, F)\) is the minimal DFA of \(L\) and that the syntactic monoid of \(L\) is aperiodic. Using the hint, we know that \(\delta_w^n\) is eventually constant for all \(w \in \Sigma^*\). As a consequence, if \(q \in Q\) is such that \(\delta(q, w^n) = q\), then \((\delta_w)^{kn}(q) = q\) for all \(k \in \mathbb{N}\). If \(k\) is large enough, then \(\delta_w^{kn} = \delta_w^{kn+1}\), and therefore \(\delta_w(q) = \delta_w (\delta_w^{kn})(q) = \delta_w^{kn} (q) = q\). We have proven that \(A\) has no counters.

Solution: From counter-freeness to aperiodicity

Assume that the minimal DFA \(A = (Q, q_0, \delta, F)\) recognising \(L\) is counter-free. Let \(w \in \Sigma^*\). We will prove that the sequence \(\delta_w^n\) is eventually constant. Let \(q \in Q\), there exists \(i < j\) such that \(\delta_w^i(q) = \delta_w^j(q)\). Let \(q' :=\delta_w^i(q)\), then \(\delta_w^{j-i}(q') = q'\). Since \(A\) is counter-free, we conclude that \(\delta_w(q') = q'\). In particular, the sequence \(\delta_w^n(q)\) is eventually constant. Now, because \(Q\) is finite, the sequence \(\delta_w^n\) is itself eventually constant. And because there are finitely many functions \(\delta_w\) there exists a uniform bound \(N_0\) such that \(\delta_w^n = \delta_w^m\) for all \(n,m \geq N_0\) and all \(w \in \Sigma^*\).

Solution: Non-minimal counter-free automaton

If \(A\) is a counter-free automaton that recognises \(L\), then the minimal DFA recognising \(L\) is also counter-free.

1.2 Fixed points

A fixed point of a function \(f\) is a value \(x\) such that \(f(x) = x\). For the following models of computation, can we decide if \(f\) has a fixed point?

  • Mealy Machines?
  • Rational Transductions?
  • Two-way Deterministic Transducers with outputs?
Hint: What do you want to prove
  • Mealy Machines: Yes, because the collection of fixed points is a regular language.
  • Rational Transductions: no.
Solution: Solution

For Mealy Machines, the output is letter-to-letter, so if a fixed point exists, it must start with a transition that produces exactly the letter that is read. This means that it has a fixed point if and only if it has a fixed point of length \(1\).

For rational transductions, the problem is undecidable because it is equivalent to the halting problem for Turing Machines. Let \(M\) be a Turing Machine, such that a configuration of \(M\) terminates.

Consider the function \(s_M \colon \Sigma^* \to \Sigma^*\) that maps an encoding of a configuration of \(M\) to the encoding of the successor configuration. Let \(f_M \colon \Sigma^* \to \Sigma^*\) be the rational function that maps a sequence of configurations to the sequence of successor configurations, prepending to the result the initial configuration of \(M\).

A terminating run of \(M\) is a fixed point of \(M\). Conversely, if \(f_M\) has a fixed point, then it must be a valid run of \(M\) (successor configurations are correctly computed), and this run cannot be continued (otherwise it would not be a fixed point). Therefore, the problem of deciding whether \(f_M\) has a fixed point is equivalent to the halting problem for \(M\).

2 Logic

2.1 Kleene Star Stability

Are languages definable in first-order logic closed under kleene star?

2.2 Examples of aperiodic languages

Write a star-free expression that defines the language \((ab)^*\).

2.3 A single existential quantifier is enough

Show that regular languages are definable by \(\mathbb{MSO}\) formulas using a single existential monadic second order quantifier.

Hint: Encode the states with padding

If the automaton has \(n\) states, then represent the state of the automaton for positions that are multiple of \(n\) using a unary encoding of the state plus a separator. How can you then recover the intermediate transitions?

References