This is a complete implementation of the DFA minimization algorithm found in

Timo Knuutila,
Re-describing an algorithm by Hopcroft,
Theoretical Computer Science,
https://doi.org/10.1016/S0304-3975(99)00150-4.

This article uses random-access arrays for most of its main data structures. We, instead, use red-black trees, which increases the worst-case runtime. Otherwise, our implementation follows the article closely.

To run the algorithm on an automaton, call

run_Equivalence A X final M trans

with SML/NJ, where

• A is the list of states (ints) of the automaton,
• X the list of alphabet symbols (ints),
• final the list of final states,
• M the measure used for selecting which classes to insert in the list of candidates, and
• trans the transition function encoded as a list of triples (x,s,t) of ints where s steps to t on symbol x.

This will print the states of the minimal DFA.

Here, the measure refers to one of the three values

class | x_class | inv_class

where

• class corresponds to |B|,
• x_class to |x^-1 ∩ B|, and
• inv_class to |x⁻¹(B)|.

Note that in Section 4.5, the author should say that x^-1 ∩ B stands for x(A) ∩ B, not x^-1(A) ∩ B.

We have also corrected a few small mistakes in the article's pseudocode. These corrections are marked in our source code.

Finally, this code has been tested on just a handful of examples. Please let me know of any bugs you find.