Genetic algorithms are very good solved sudoku generators

Abstract

I present a simple and yet effective GA-based approach to content generation in the Sudoku domain. The GA finds multiple full boards which can act as solutions for Sudoku and Killer Sudoku puzzles. In this work I use a binning-based diversity maintenance approach in order to encourage GA to find more solution boards, resluts prove that though both approaches routinely manage to find multiple solution boards it is in fact the simple GA without any diversity maintenance that finds more such boards. Using a simpler approach to manipulate the fitness function to penalize previously found solutions improves the algorithm further.