The Art, Science, and Engineering of Programming, 2022, Vol. 6, Issue 3, Article 12
Submission date: 2021-10-01
Publication date: 2022-02-15
Full text: PDF
Graphs and various graph-like combinatorial structures, such as preorders and hypergraphs, are ubiquitous in programming. This paper focuses on representing graphs in a purely functional programming language like Haskell. There are several existing approaches; one of the most recently developed ones is the “algebraic graphs” approach (2017). It uses an algebraic data type to represent graphs and has attracted users, including from industry, due to its emphasis on equational reasoning and making a common class of bugs impossible by eliminating internal invariants.
The previous formulation of algebraic graphs did not support edge labels, which was a serious practical limitation. In this paper, we redesign the main algebraic data type and remove this limitation. We follow a fairly standard approach of parameterising a data structure with a semiring of edge labels. The new formulation is both more general and simpler: the two operations for composing graphs used in the previous work can now be obtained from a single operation by fixing the semiring parameter to zero and one, respectively.
By instantiating the new data type with different semirings, and working out laws for interpreting the resulting expression trees, we discover an unusual algebraic structure, which we call “united monoids”, that is, a pair of monoids whose unit elements coincide. We believe that it is worth studying united monoids in their full generality, going beyond the graphs which prompted their discovery. To that end, we characterise united monoids with a minimal set of axioms, prove a few basic theorems, and discuss several notable examples.
We validate the presented approach by implementing it in the open-source algebraic-graphs library. Our theoretical contributions are supported by proofs that are included in the paper and have also been machine-checked in Agda. By extending algebraic graphs with support for edge labels, we make them suitable for a much larger class of possible applications. By studying united monoids, we provide a theoretical foundation for further research in this area.
firstname.lastname@example.org, Jane Street, UK / Newcastle University, UK