The ultrametric constraint was rst proposed in [GPSW03] within the
context of supertreeconstruction [BE04], and was implemented using toolkit primitives. We review this encodingand show that in most constraint toolkits this is inecient in terms of both space and time.This motivates the creation of a specialised ultrametric propagator over three variables, thatmaintains the ultrametric property over the bounds of those variables. This is presented interms of its propagation methods. This is then extended to a specialised propagator thatmaintains the ultrametric property on a symmetric matrix of variables. Previous work on the ultrametric constraint: First, we start with a denition of the ultrametric constraint. Denition : An ultrametric constraint on three variables (henceforth, Um-3) x, y and z constrains them such that: (x > y = z) (y > x = z) (z > x = y) (x = y = z) (1) This constraint ensures that there is a tie for the least element of the three, i.e., eitherall three are the same, or two are the same and the other is greater. The constraint wasproposed by [GPSW03, Pro06] and was implemented as a literal translation of equation 1using toolkit primitives. Evidence obtained from the JChoco, ECLiPSe and ILog constraint programming toolkits shows that no propagation is done to lower bounds by thiscombination of primitive constraints. This is a due to the disjunctive constraints. In manyconstraint programming toolkits propagation on disjunctive constraints is delayed until allbut one of the disjuncts is disentailed, i.e., delayed-disjunction consistency is used [HSD98].Consequently, in the above encoding of the ultrametric constraint a non-ultrametric instantiation can occur. Consider the case for three variables: x {1, 2, 3}, y {2, 3} andz {3}. The domains of the variables are already at a xed point with respect to delayeddisjunction consistency and a search process can instantiate these variables and fail (byassigning x the value 1). As we shall see later, nding a solution to the supertree problem using toolkit constraints can result in a backtracking search. Of course, higher levelsof consistency would overcome this, such as constructive-disjunction consistency [HSD98],singleton arc-consistency [DB97] or the ltering algorithm of [Lho03]. However, the cost ofthese is greater in the average case than delayed-disjunction, preventing their use in
toolkits. In fact for the Um-3 constraint it is especially unfortunate that the lower bounds maynot be trimmed properly.
Minimum Ultrametric Trees:
An edge-weighted tree is called ultrametric if the distances from the root to all the leaves in the tree are equal. For an n by n distance matrix M, the minimum ultrametric tree for M is an ultrametric tree T = (V, E, w) with leaf set {1,..., n} such that dT(i, j) M[i, j] for all i, j and eEw(e) is minimum, where dT(i, j) is the distance between i and j on T. Constructing minimum ultrametric trees from distance matrices is an important problem in computational biology. In this paper, we examine its computational complexity and approximability. When the distances satisfy the triangle inequality, we show that the minimum ultrametric tree problem can be approximated in polynomial time with error ratio 1.5(1 + log n), where n is the number of species. We also develop an efficient branchand-bound algorithm for constructing the minimum ultrametric tree for both metric and non-metric inputs. The experimental results show that it can find an optimal solution for 25 species within reasonable time, while, to the best of our knowledge, there is no report of algorithms solving the problem even for 12 species. Additive trees:
A generalization of ultrametric trees are additive trees. Remind that in
an ultrametric tree, the number of mutations was assumed to be proportional to the temporal distance of a node to the ancestor and it was also assumed that the mutations took place with the same rate in all paths. Thus an ultramaetric tree is assigned a root and the distance from the root to a leave is constant. But it's a fact, that the evolutionary clock is running differently for different species and even for different regions i.e. in a protein sequence. An unrooted phylogenetic tree is a reflection of our ignorance as to where the common ancestor lies. All nodes of an additive tree except for the leaves have degree three, an additive tree is therefore an unrooted binary tree.
Definition: The additional requirement for an additive metric is:
An additive tree also is characterized by the four point condition:
Any 4 points can be renamed such that
The tree construction from an additive metric works by
successive insertion. There is exactly one tree topology that allows for realization of an additive metric.