A function $f: A^* \to A^*$ is regularity-preserving if, for each regular language $L$ of $A^*$, the language $f^{-1}(L)$ is regular. I think I have a proof, as a consequence of more general results, that the function defined by $$ f(a_1 \dotsm a_n) = a_1(a_1a_2)(a_1a_2a_3)\ \dotsm\ (a_1 \dotsm a_n) $$ is regularity-preserving. If this result is correct, could someone provide an elementary proof?
1 Answer
Here is a proposition for an elementary proof:
Let $\mathcal A=(A,Q,q_0,F,\delta)$ be a DFA for $L$, we want to build a DFA $\mathcal A'=(A,Q',q_0',F',\delta')$ for $f^{-1}(L)$. Intuitively, when reading a word $u$, $\mathcal A'$ will remember the state reached in $\mathcal A$ by $f(u)$, together with the action of $u$ on all states of $\mathcal A$. More formally, we take:
- $Q'=Q\times Q^Q$
- $q_0'=(q_0,\mathit{id})$
- $F'=F\times Q^Q$
- $\delta_a'(p,g)=(\delta_a(g(p)),\delta_a\circ g)$
Where $\delta_a:Q\to Q$ is the transition function associated with a letter $a$.
This ensures that after reading a word $u$, the first component of the state of $\mathcal A'$ gives the state reached by $\mathcal A$ on $f(u)$.
-
$\begingroup$ @Denis how is the function g defined? $\endgroup$ Commented Apr 16, 2021 at 19:19
-
1$\begingroup$ @Hermann: $g$ is the second component of the current state: from any state $(p,g)$, reading letter $a$, we go to the state $(\delta_a(g(p)),\delta_a\circ g)$. $\endgroup$– DenisCommented Apr 16, 2021 at 19:26
-
$\begingroup$ I'd add that there is a simple and systematic way to show that a copyful streaming string transducer Σ* → Γ* is regularity-preserving, which takes a monoid morphism with codomain $M$ recognizing a language over Γ and yields a DFA with states $M^k$ recognizing a language over Σ ($k$ is the number of registers). The function considered here can be computed by a copyful SST with 2 registers, so if we compose this systematic construction with taking the transition monoid of a DFA, we get the set of states $(Q^Q)^2$ which is very close to Denis's solution! $\endgroup$ Commented Nov 3, 2022 at 1:07