Page 36 - Tequio 11
P. 36

A phylogenetic tree is an -tree where only the leaves are labelled and all internal vertices have a degree of
                at least 3. As in the case of -trees, a phylogenetic tree is binary if all internal vertices have degree three. As a
                more concrete example of a binary phylogenetic tree consider

                Figure 6.
                A binary phylogenetic tree.

                For  = || ≥ 3 denote by  /  the set of all binary -trees with  leaves. If the context is clear we will also simply
                say binary tree for binary phylogenetic -trees.
                     Given an -tree, there is an associated system of splits on  obtained by considering the two connected
                components obtained by the removal of  in  for each edge  ∈ (). Denote the so-obtained set of splits by
                ∑(). For an example we refer to example [ex_dist]. On the other hand, by Theorem [tree], we get that for ∑ a
                system of splits, there is an -tree  such that ∑ = ∑() if and only if the system of splits is compatible.
                     More general split systems are employed for generalizations of unrooted phylogenetic trees, where graphs
                in such split networks are not necessarily trees, and one or more edges in the graph are used to represent a
                split (Dress et al., 2011; Huson et al., 2010).

                3. Polytopes in phylogenetics
                Both the Tight span and the Lipschitz polytope are associated to a (finite) metric space and relate to a distance-
                preserving embedding in a bigger space. The minimum evolution polytope on the other hand is associated to
                natural numbers  ∈ ℕ nA . In the following, we aim to introduce and motivate the main objects. However, the
                topics are mature research directions and we restrict to a non-exhaustive treatment.

                3.1 Tight span
                Isbell studied the tight span in his investigation of injectivity for metric spaces (Isbell, 1964). In phylogenetics,
                it appeared in relation to reconstruction of phylogenetic trees from finite metric spaces  (Dress, 1984).
                Representations of distances of phylogenetic trees can be seen as a connected one-dimensional polytope.
                Distances between vertices correspond to the sum of the edge lengths of the shortest paths. Hence it is natural
                to ask whether we can embed a given finite metric space distance-preserving into a low-dimensional compact
          34     On some polytopes in phylogenetics/Hoessly/27-40
                polytope. One such possibility is the so-called Tight span.
                     The Tight span often helped to establish properties of classes of metrics, particularly in relation to
                decompositions that are of interest in phylogenetics. Furthermore, the 1-skeleton  of the Tight span is a graph
                                                                                    9
                realisations of the metric (Dress, 1984). For more on the motivation and connection of the study of Tight span
                to phylogenetics we refer to, e.g., (Dress et al., 2011) or (Huson et al., 2010), and for a concrete algorithmic
                application    to,     e.g.,    First    we      consider     an      unbounded      polytope
                 (ô,j) : = ö ∈ ℝ ∣  P +  Ä ≥ (, )  ∀,  ∈ ú.
                             ô

                Definition 3.1 The Tight span of (, ) is given by the minimal points of  (ô,j) , which are defined as  (ô,j) : =
                ö ∈  (ô,j) ∣   ∈  (ô,j)  and  ≤  implies  = ú.

                Note that the Tight span  (ô,j)  corresponds to the bounded faces of the polyhedron  (ô,j) . The Tight span is a
                polytopal complex that is associated to any finite metric space (, ), whose structure often catches features
                of (, ). The Kuratowski embedding is a map  (ô,j) :  → ℝ  that sends elements of  = { - , ⋯ ,  / } to its Tight
                                                                 ô
                span, while preserving their pairwise distance. It is defined as

                                               (ô,j) :   → ℝ ô

                                                       P ↦  (ô,j) ( P ): = (( P ,  Ä )) Ä∈ô
                We have the following.

                Lemma 3.2 The function  (ô,j) : (, ) → ( (ô,j) , || ⋅ || û ) (where  (ô,j) ⊆ ℝ ) is an isometric map into the tight
                                                                             ô
                span  (ô,j) , where for  ∈ ℝ , |||| û : = max î 2 ∈ô {| P |}.
                                       ô

                Example 3.3 Consider the metric on  = { - ,  * } with ( - ,  * ) = 1. As a tree-like metric, it can be illustrated
                as

                                                                                            1
                                                                                    0
                The upper map sends   -   to                       † ( - ,  - ) ° = ¢ £,  *  to ¢ £, hence
                                                                        ( - ,  * )  1  0

                                                                    0
                                                               1
                                      || (ô,j) ( - ) −  (ô,j) ( * )|| û = || ¢ £ − ¢ £ || û = 1 = ( - ,  * ),
                                                               0    1

                and the Tight span looks as follows.

                Figure 7.
                The Tight-span as the blue line.










                Example 3.3 generalises as follows to tree-like metric spaces, which can be classified via their Tight span




                                                 Tequio, enero-abril 2021, vol. 4, no. 11
   31   32   33   34   35   36   37   38   39   40   41