📝

1 Previous Models

1.1 Canonical Bimachines

Let us recall that the production function of a rational function \(f\) can be written \(\pi_f \colon \Sigma^* \times \Sigma \times \Sigma^* \to \Gamma^*\). Given a rational function \(f \colon \Sigma^* \to \Gamma^*\), we can define two congruences \(\simeq_l\) and \(\simeq_r\) over \(\Sigma^*\) as follows:

\[ u \simeq_ l v \iff \forall x,y \in \Sigma^*, \forall a \in \Sigma, \forall w \in \Sigma^*, \pi_f(xuy, a, w) = \pi_f(xvy, a, w) \]

And similarly for \(\simeq_r\).

  1. Prove that \(\simeq_l\) and \(\simeq_r\) have finite index.
  2. Construct a canonical bimachine computing \(f\).
  3. What is the complexity of the construction?
  4. Can you refine the construction by first minimising the left congruence, and then the right congruence?

This construction was used in Filiot, Gauwin, and Lhote (2016) to prove the decidability of the following problem: given a rational function \(f\), is it decidable whether \(f\) can be computed by a star-free bimachine?

1.2 The Great Simplification

Given a rational function \(f\), is it decidable whether there exists a Mealy Machine that computes \(f\)?

Hint: Use Monoids Bimachines

Prove that a rational function \(f\) that satisfies \(f(\varepsilon) = \varepsilon\) can be transformed into a monoid-bimachine defined by a finite monoid \(M\), a surjective morphism \(\mu \colon \Sigma^* \twoheadrightarrow M\), and a production map \(\pi \colon M \times \Sigma \times M \to \Gamma^*\), whose semantics is defined as follows for all words \(w \in \Sigma^*\):

\[ f(w) :=\prod_{u a v = w} \pi(\mu(u), a, \mu(v)) \quad . \]

The production function can be generalised to subwords as follows:

\[ \pi(m_l, w, m_r) :=\prod_{uav = w} \pi(m_l \mu(u), a, \mu(v) m_r) \quad . \]

Using this notation \(f(w) = \pi(1_M, w, 1_M)\).

Hint: Decompose the problem

Can you decide if a letter-to-letter unambiguous NFA with outputs is computed by a Mealy Machine? Can you decide if a rational function is computed by a letter-to-letter unambiguous NFA with output?

Hint: What about idempotents

Let \(w \in \Sigma^*\) be such that \(\mu(w)^2 = \mu(w)\) (\(\mu(w)\) is idempotent), and \((m_l, m_r) \in M^2\). What can you say about \(\pi(m_l \mu(w), w, \mu(w)m_r)\)?

Hint: Construct Idempotents

Prove using Ramsey’s theorem that for every finite monoid \(M\) there exists (a computable) \(N \in \mathbb{N}\) such that for all \(w \in M^*\), one can compute \(w = u_1 u_2 u_3\) such that \(\mu(u_2)\) is idempotent\(\mu(u_2)^2 = \mu(u_2)\) –, \(|u_1| \leq N\) and \(|u_3| \leq N\).

Hint: Use Quantitative Pumping Arguments

Assume that \(f\) is computed by a letter-to-letter unambigous NFA with outputs, then \(|f(w)| = |w|\) for all \(w \in \Sigma^*\). Prove that this necessary condition is also sufficient.

To that end, notice that the map \(X \mapsto \pi(m_l, w^X, m_r)\) is a function from \(\mathbb{N}\) to \(\Gamma^*\) that must be size preserving, and therefore that \(|\pi(m_l\mu(w), w,\mu(w) m_r)| = |w|\). Indeed, because \(\mu\) is surjective, there exist words \((x,y) \in \Sigma^*\) such that \(\mu(x) = m_l\) and \(\mu(y) = m_r\). Therefore, for \(X \geq 3\),

\[ f(x w^X y) = \underbrace{\pi(1_M, xw, \mu(wy))}_{\alpha} \pi(\mu(xw), w, \mu(wy))^{X-2} \underbrace{\pi(\mu(xw), y, 1_M)}_{\beta} \quad . \]

Use the above equation to conclude.

2 Logic

2.1 Word representations

Consider two ways of representing a finite word as a model: we either have the order relation \(x < y\), or we have the successor relation \(x = y + 1\). Show that for both ways, \(\mathbb{MSO}\) gives the same expressive power. Is it true for \(\mathbb{FO}\)?

2.2 Short Formulas

Prove that there exists a family of languages \(L_n\) that are defined by a formula of size \(O(n)\) but such that the minimal deterministic automaton for \(L_n\) has size \(\Omega(2^n)\). What about the size of an NFA$?

Hint: Good languages

Consider the language \(L_n\) of words of length exactly \(2^n\).

Hint: The usual trick

Let \(\varphi(x,y)\) be a first order formula. Prove the equivalence between the two following formulas:

  1. \(\psi(x,y) :=\varphi(x,z) \wedge \varphi(z,y)\).
  2. \(\theta(x,y) :=\forall s,t. (s = x \wedge t = z) \vee (s = z \wedge t = y) \Rightarrow \varphi(s,t)\).
Hint: Minimal Automaton

How would you prove that the minimal automaton has at least \(2^n\) states? Using the Myhill-Nerode theorem for instance?

2.3 Logic and Monoids

Let \(q \in \mathbb{N}\) be a fixed quantifier rank.

  1. Prove that the \(\mathbb{MSO}^q\) theory of a word \(uw\) is uniquely determined by the \(\mathbb{MSO}\) theory of \(u\) and \(w\).
  2. What about the \(\mathbb{FO}^q\) theory?
  3. Define the map \(\iota \colon \Sigma^* \to \mathcal{P}(\mathbb{MSO}^q)\) by
Hint: Colored Logic

Define a translation of usual formulas in a coloured logic, where variables are either guaranteed to be taken in \(u\) or guaranteed to be taken in \(w\). This can be seen as an extra type system, or a sorted logic.

Prove that formulas in this typed logic are equivalent to boolean combinations of formulas that have a single type (i.e., monochromatic formulas), taking care of counting the quantifier rank of the resulting sentences.

What have you proven?

Hint: Aperiodicity

To prove that the monoid is aperiodic in the case of \(\mathbb{FO}^q\), it suffices to prove that given a first order sentence \(\varphi\), and a word \(w\), there exists \(n \in \mathbb{N}\) such that \(w^n \models\varphi \iff w^{n+1} \models \varphi\). We will prove the stronger statement by induction: for sentences of quantifier rank \(q\), \(w^{2^q}\) and \(w^{2^q + 1}\) have the same \(q\)-first order types.

3 Two Way Deterministic

3.1 Examples and non-examples

For the following functions, provide the simplest model of computation that can express them.

  • The reverse function
  • The sort function
  • The cycle function, that performs a circular permutation such, for instance mapping \(abcd\) to \(dabc\)
  • The swap function, that swaps the first two letters of a word
Hint: Proof for the reverse using Monoids

Consider a bimachine defined in terms of monoids, i.e., defined by a morphism \(\mu \colon \Sigma^* \to M\), and a production function \(\pi \colon M \times \Sigma \times M \to \Gamma^*\). Let \(e_a\) be the unique idempotent in the image \(\left\{\mu(a^k) \mid k \geq 1\right\}\) and \(e_b\) be the unique idempotent in the image \(\left\{\mu(b^l) \mid l \geq 1\right\}\).

Consider the (generalised) outputs \(\alpha :=\pi(e_a, a^k, e_a e_b)\) and \(\beta :=\pi(e_a e_b, b^l, e_b)\). It is clear that \(\mathsf{reverse}(a^{Xk} b^{Yl}) = b^{Yl} a^{Xk}\), but it is also equal to \(u_0 \alpha^X u_1 \beta^Y u_2\), where \(u_0, u_1, u_2 \in \Gamma^*\). By considering \(Y\) large enough, we conclude that \(\alpha\) is \(b^k\). Similarly, we conclude that \(\beta = a^l\). However, this is absurd, since the number of \(a\)’s and \(b\)’s are not preserved when \(X \neq Y\).

3.2 2DFTs for Languages

Prove that the class of languages recognised by deterministic two-way transducers coincides with the class of languages recognised by deterministic finite automata using monoids.

3.3 Forward Images?

Let \(f\) be computed by a two-way deterministic transducer with outputs, and \(L\) be a regular language. Is it true that \(f(L)\) is a regular language?

Hint: The answer is no

What about \(f(L) = \left\{ a^n b^n \mid n \in \mathbb{N}\right\}\)?

3.4 Expressiveness

Prove that 2DFT are more expressive than rational functions. What about sweeping DFTs that can only change direction at the endpoints of the input?

Hint: Reverse

The reverse function is not rational, but can be performed using a sweeping 2DFT.

Hint: Reverse Map

The reverse map function is not doable by a sweeping 2DFT, but can be done by a 2DFT.

3.5 Languages and Functions

Provide a direct proof of the following inclusion of classes:

\[ \mathsf{2DFA} \cdot \mathsf{Rat} \subseteq \mathsf{Rat} \cdot \mathsf{2DFA} \]

Hint: Use a general decomposition theorem

Every deterministic two-way transducer can be decomposed into a first rational function that computes the state information about the run, followed by a unfold function, that utilizes this information together with the input word to produce the input.

3.6 Class inclusions

Prove that given a function \(f\) computed by a two-way deterministic transducer with outputs, it is decidable whether \(f\) is rational. This is an extremely hard exercise.

References

Filiot, Emmanuel, Olivier Gauwin, and Nathan Lhote. 2016. First-Order Definability of Rational Transductions: An Algebraic Approach. In 2016 31st Annual ACM/IEEE Symposium on Logic in Computer Science (LICS), 1–10.