Dynamic trees, originally described by Sleator and Tarjan, have been studied in detail for non persistent structures providing O(log n) time for update and lookup operations as shown in theory and practice by Werneck.
However, there are two gaps in current theory. First, how the most common dynamic tree operations (link and cut) are computed over a purely functional data structure has not been studied in detail. Second, even in the imperative case, when checking whether two vertices u and v are connected (i.e. in the same component), it is taken for granted that the corresponding location indices (i.e. pointers, which are not allowed in purely functional programming) are known a priori and do not need to be computed, yet this is rarely the case in practice.
In this thesis we address these omissions by formally introducing two new data structures, Full and Top, which we use to represent trees in a functionally efficient manner. Based on a primitive version of finger trees – the de facto sequence data structure for the purely lazy-evaluation programming language Haskell – they are augmented with collection (i.e. set-based) data structures in order to manage efficiently k-ary trees for the so-called linearisation case of the dynamic trees problem. Different implementations are discussed, and their performance is measured.
Our results suggest that relative timings for our proposed structures perform sublinear time per operation once the forest is generated. Furthermore, Full and Top implementations show simplicity and preserve purity under a common interface.
