Temporal Tree-Child Hybrid Number

• Temporal Tree-Child Hybrid Number is an optimization measure that minimizes the reticulation count in temporal and tree-child networks displaying a set of phylogenetic trees.
• It leverages cherry-picking and tree-child sequences alongside semi-temporal labeling to enforce strict temporal constraints in phylogenetic inferences.
• Recent algorithmic advances offer fixed-parameter tractable methods that highlight how temporal restrictions refine feasibility and optimality in network modeling.

Temporal Tree-Child Hybrid Number is the optimization quantity obtained by minimizing reticulation number over phylogenetic networks that are simultaneously temporal and tree-child and that display a prescribed set of rooted phylogenetic trees. In the formalism developed for arbitrary sets of binary and nonbinary trees, it is written

$$h_{t,TC}(T)=\min\{r(N):N\text{ displays }T,\ N\text{ is tree-child, and }d(N)=0\},$$

where $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$ is the hybridization number, and $d(N)=0$ expresses exact temporality via a semi-temporal labeling framework (Borst et al., 2020). The topic sits at the intersection of phylogenetic network inference, cherry-picking characterizations, parameterized complexity, and the combinatorics of ranked tree-child networks. A central theme in the recent literature is that temporal constraints substantially restrict the admissible solution space relative to unconstrained tree-child networks, often changing feasibility, optimal reticulation count, and algorithmic structure (Borst et al., 2020, Bulteau et al., 2023).

1. Formal setting and core definitions

A rooted phylogenetic network on a leaf set $X$ is a rooted DAG whose leaves are exactly $X$, with internal vertices partitioned into tree vertices and hybridization vertices. In the binary setting, tree vertices have indegree $1$ and outdegree $2$, while reticulation vertices have indegree $2$ and outdegree $1$; more generally, reticulation indegree may exceed $2$ (Borst et al., 2020, Bulteau et al., 2023). A network displays a tree if the tree can be obtained from a subgraph by deleting arcs and vertices and suppressing degree-$r(N)=\sum_{v\neq \rho}(d^-(v)-1)$0 vertices (Borst et al., 2020). A tree-child network is one in which every tree vertex has at least one outgoing tree arc, or equivalently every non-leaf node has at least one child that is a tree vertex or a leaf (Borst et al., 2020, Bulteau et al., 2023).

In the framework of temporal hybridization, a temporal network is a tree-child network equipped with a labeling $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$1 such that tree arcs satisfy $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$2 and reticulation arcs satisfy $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$3 (Borst et al., 2020). The same equal-time versus strict-increase dichotomy appears in the ranking literature, where a ranking is a discrete temporal labeling compatible with tree arcs and reticulation events (Zhang et al., 6 Jun 2025).

Quantity	Definition	Scope
$r(N)=\sum_{v\neq \rho}(d^-(v)-1)$4	$r(N)=\sum_{v\neq \rho}(d^-(v)-1)$5	Hybridization number of a network
$r(N)=\sum_{v\neq \rho}(d^-(v)-1)$6	Minimum $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$7 over temporal networks displaying $r(N)=\sum_{v\neq \rho}(d^-(v)-1)$8	Temporal hybridization number
$r(N)=\sum_{v\neq \rho}(d^-(v)-1)$9	Minimum $d(N)=0$0 over temporal tree-child networks displaying $d(N)=0$1	Temporal tree-child hybrid number
$d(N)=0$2	Minimum number of reticulation arcs violating equality in a semi-temporal labeling	Temporal distance

The introduction of temporal distance refines the strict temporal notion. A semi-temporal labeling requires strict increase on tree arcs and sets each hybridization vertex time to the minimum of its parents’ times; $d(N)=0$3 then counts reticulation arcs whose endpoints do not receive equal times. Temporal networks are exactly those with $d(N)=0$4 (Borst et al., 2020). This places the Temporal Tree-Child Hybrid Number as the zero-distance case of a broader optimization problem over tree-child networks.

2. Sequence characterizations

For binary input trees, the principal structural characterization is via cherry-picking sequences. If $d(N)=0$5 is a set of binary trees on the same taxon set $d(N)=0$6, a sequence $d(N)=0$7 is a cherry-picking sequence if each $d(N)=0$8 is chosen from the set of leaves that lie in a cherry in every currently reduced tree, and the weight is

$d(N)=0$9

where $X$0 and $X$1 is the union of the cherry-neighbor sets of $X$2 across the input trees (Borst et al., 2020). The key theorem states that there exists a temporal network displaying the input trees with reticulation number $X$3 if and only if there exists a cherry-picking sequence of weight $X$4. Consequently, the minimum weight of a cherry-picking sequence equals the minimum temporal hybridization number (Borst et al., 2020).

For tree-child networks more generally, the relevant object is a generalized cherry-picking sequence, also called a tree-child sequence. Its weight is $X$5, and Linz and Semple’s characterization, as quoted and used algorithmically, states that a set of trees admits a tree-child network with reticulation number $X$6 if and only if there is a tree-child sequence of weight $X$7 (Iersel et al., 2019, Borst et al., 2020). The temporal extension in (Borst et al., 2020) identifies a class of non-temporal elements in such sequences, and proves equivalence between bounded temporal distance in the network and bounded numbers of non-temporal elements in the corresponding sequence.

For two trees under the weaker rigid-display model, fork-picking sequences play an analogous role. The rigid hybrid number

$X$8

coincides with the minimum weight of a fork-picking sequence, denoted $X$9 (Huber et al., 2020). The same work proves that, for two trees, the following are equivalent: rigid display by a temporal tree-child network, exact display by a temporal tree-child network, existence of a cherry-picking sequence, and existence of a fork-picking sequence (Huber et al., 2020). This equivalence concerns existence rather than optimal count, and the distinction becomes significant in later comparisons.

3. Algorithmic results and parameterized complexity

The main fixed-parameter tractability results are due to the development of sequence-based search algorithms. For an arbitrary set of $X$0 rooted binary trees with $X$1 leaves each, the minimum temporal hybridization number can be computed in

$X$2

time, where $X$3 is the optimum temporal hybridization number (Borst et al., 2020). The algorithm searches for a minimum-weight cherry-picking sequence using a recursive constraint-based branching scheme. Its analysis uses the potential

$X$4

which yields the $X$5 bound on the search tree (Borst et al., 2020).

The same paper introduces a broader FPT problem: deciding whether there exists a tree-child network displaying the input trees with at most $X$6 reticulations and temporal distance at most $X$7. This semi-temporal problem is solvable in

$X$8

and the case $X$9 recovers temporal tree-child networks and hence $1$0 (Borst et al., 2020). The factor $1$1 comes from branching on at most $1$2 actionable ordered cherry-pairs for each non-temporal step.

For two rooted nonbinary trees, the first FPT algorithm for minimum temporal hybridization runs in

$1$3

again parameterized by the optimum $1$4 (Borst et al., 2020). The factorial reflects branching over permutations within minimal clusters and over bounded sets of terminals. The same paper reports an implementation and experimental analysis; empirically, the $1$5 temporal algorithm often followed a curve closer to $1$6 on the tested instances, and about $1$7 of instances did not admit any temporal network, which motivated the semi-temporal formulation (Borst et al., 2020).

For comparison, the non-temporal tree-child optimization problem was previously shown FPT for arbitrary numbers of binary input trees using tree-child sequences, with running time

$1$8

and a parallel implementation that could deal with up to $1$9 input trees on a standard desktop computer (Iersel et al., 2019). That algorithm does not enforce temporality; this suggests that temporal constraints improve structure for some tasks but also impose an additional feasibility barrier.

4. Relation to unconstrained tree-child hybridization

The strongest contrast with the temporal setting comes from universal and hardness results for unconstrained tree-child networks. For line trees, the tree-child network inference problem remains NP-hard via a reduction from the Shortest Common Supersequence problem on permutations, and the same paper proves that the parsimonious tree-child networks displaying all line trees on a taxon set $2$0 are identical to those displaying all binary trees on $2$1 (Bulteau et al., 2023). The resulting universal hybridization number satisfies

$2$2

with explicit bounds

$2$3

for $2$4 (Bulteau et al., 2023).

These cubic bounds are explicitly non-temporal. The constructions used in the reduction and in the universal-network argument, especially the one-component networks $2$5, generally violate standard temporal constraints because a reticulation receives incoming arcs from vertices lying at different positions on a chain, while temporal labeling would require equal times on all parents of the reticulation (Bulteau et al., 2023). The paper therefore states that its results pertain to unconstrained tree-child networks, and that determining the corresponding temporal quantity $2$6, if finite, remains open (Bulteau et al., 2023).

This separation is conceptually important. In unconstrained tree-child optimization, one may first minimize over all tree-child networks and only then ask whether the optimum happens to be temporal. The available results show that this procedure is not sound in general. A tree-child optimum can fail to admit any temporal labeling, and imposing temporality can increase the minimum reticulation count or even destroy feasibility (Iersel et al., 2019, Bulteau et al., 2023).

5. Ranking, normality, and combinatorial structure

Temporality is tightly connected to ranking. In the combinatorial theory of ranked tree-child networks, a tree-child network is temporal if and only if it is the underlying network of a ranked tree-child network (Bienvenu et al., 2020). Every temporal tree-child network is normal, and ranked semi-binary tree-child networks are normal as well (Bienvenu et al., 2020, Zhang et al., 6 Jun 2025). This normality constraint excludes shortcut arcs and is one reason the temporal subclass is combinatorially more rigid than the full tree-child class.

For binary ranked tree-child networks with $2$7 leaves, if $2$8 denotes the number of branching events and $2$9 the number of reticulations, then

$2$0

The exact number of ranked tree-child networks with $2$1 leaves and reticulation number $2$2 is

$2$3

where the Stirling number is unsigned of the first kind (Bienvenu et al., 2020). Under the uniform model on ranked tree-child networks,

$2$4

and

$2$5

so a uniformly random ranked tree-child network is typically highly reticulated (Bienvenu et al., 2020).

A complementary problem is to count the number of valid temporal rankings of a fixed tree-child network. For separated binary or semi-binary tree-child networks, this can be done in linear time after constructing a rooted tree $2$6 by contracting reticulation events, deleting short-cuts and parallel arcs, and removing leaves. If $2$7 is ranked, the number of rankings is

$2$8

where $2$9 is the number of descendants of $1$0 in $1$1, including $1$2 itself (Zhang et al., 6 Jun 2025). A network has exactly one ranking if and only if $1$3 is a directed path, and a binary tree-child network with $1$4 leaves and $1$5 reticulations has at most one ranking (Zhang et al., 6 Jun 2025). For a uniformly random binary tree-child network with $1$6 reticulations,

$1$7

for fixed $1$8 as $1$9, showing that additional reticulation tends to reduce the number of compatible temporal orders by a factor asymptotically equal to $2$0 (Zhang et al., 6 Jun 2025).

6. Variants, misconceptions, and open directions

A first nomenclatural issue is that “Temporal Tree-Child Hybrid Number” is not introduced uniformly across the literature. In (Borst et al., 2020) it is an explicit optimization quantity over temporal tree-child networks, represented by the $2$1 special case of the temporal-distance framework. In (Zhang et al., 6 Jun 2025), by contrast, there is no separate term formally defined by the authors; the consistent interpretation adopted there is the reticulation count $2$2 of a ranked tree-child network, that is, of a tree-child network admitting a temporal labeling.

A second common misconception is to identify temporal and tree-child constraints. The literature does not support this identification. The 2019 FPT algorithm for tree-child network construction does not ensure temporality by construction, and the universal constructions used to establish $2$3 non-temporal hybridization for all binary trees are generally not temporal (Iersel et al., 2019, Bulteau et al., 2023). This suggests that temporality is not a mild regularity condition but an additional structural restriction with its own optimization problem.

A third distinction concerns exact display versus weaker notions. For two trees, the rigid hybrid number $2$4 satisfies

$2$5

and the gap can be large: for infinitely many leaf-set sizes,

$2$6

(Huber et al., 2020). The same work shows that Rigidly Displaying is NP-complete (Huber et al., 2020). Thus, even within temporal tree-child networks, different notions of how trees are represented lead to genuinely different reticulation minima.

Several open directions are explicit in the cited works. The temporal-distance paper asks whether more biologically meaningful temporal-distance measures admit FPT algorithms, and whether temporal hybridization remains FPT for more than two nonbinary trees with reasonable running time (Borst et al., 2020). The universal-network paper leaves open the determination of temporal analogues of its cubic non-temporal bounds (Bulteau et al., 2023). The ranking literature points to limiting laws, variance, concentration, and extremal structures for the number of rankings at fixed $2$7 as further problems (Zhang et al., 6 Jun 2025). Taken together, these results position the Temporal Tree-Child Hybrid Number as a precise but still evolving concept: formally characterized, algorithmically accessible in several important regimes, and sharply differentiated from the broader non-temporal tree-child hybridization framework.