Events–Trees–Histories: A Unified Framework

• Events–Trees–Histories is a framework that defines how discrete events are structured within trees to represent and preserve causal histories.
• It employs rigorous methodologies such as homogeneous tree indexing, reconciliation maps, and graph neural networks to compute event correlations and ancestral relationships.
• The framework applies across diverse fields—including evolutionary biology, cosmology, and distributed computing—to link local events with comprehensive temporal narratives.

“Events–Trees–Histories” denotes a recurrent formal pattern in which discrete events are organized by tree-structured objects and then interpreted as histories. In probability, measurable events are indexed by homogeneous trees and controlled on strong subtrees (Dodos et al., 2011). In evolutionary modeling, individual births, mutations, speciations, extinctions, and transfers are recorded as genealogical or phylogenetic trees (Costa et al., 2017, Nøjgaard et al., 2017). In cosmology, halo mergers are encoded as merger trees whose structure reflects assembly history and underlying physics (Poulton et al., 2018, Leisher et al., 7 Nov 2025). In distributed computing and verification, histories are made explicit either as history trees of agent indistinguishability or as event collections carried by transition-system configurations (Viglietta, 2024, Abdulla et al., 2015). This suggests a general schema in which trees are not merely static combinatorial objects: they are structured records of how events unfold over time.

1. Conceptual organization of events, trees, and histories

Several of the cited works state the triad explicitly. In individual-based speciation models, “events,” “trees,” and “histories” are described as three interconnected layers of description of the same evolving system: births, deaths, mutations, and species-level changes generate genealogical and phylogenetic trees, while algorithms such as MRCAT and SSEE make the resulting history a concrete, queryable object (Costa et al., 2017). In cosmological structure formation, a merger tree is the standard representation of a sequence of halo collapses, mergers, and accretion events; nodes are subhalos at snapshots and edges are hereditary links across snapshots (Leisher et al., 7 Nov 2025). In anonymous dynamic networks, history trees model how agents become distinguishable as they receive different multisets of messages over time (Viglietta, 2024). In well-structured transition systems with history, configurations are extended by a history component so that events generated by transitions become part of the system state (Abdulla et al., 2015).

Across these settings, the tree serves two related purposes. First, it preserves ancestry, refinement, or causality: descendants in a genealogy, progenitors in a merger tree, refined equivalence classes in a history tree, or extensions of a partial computation in a transition system. Second, it supports a history semantics: a path, a finite subtree, or a finite event set is interpreted as a partial history, while the whole tree represents a space of possible or realized histories. A plausible implication is that “Events–Trees–Histories” is less a single doctrine than a reusable formal interface between local events and global temporal structure.

2. Tree-indexed events and combinatorial regularity

A particularly abstract formulation appears in the study of measurable events indexed by homogeneous trees. A homogeneous tree is uniquely rooted and every node has exactly $b$ immediate successors; its levels are $T(n)=\{t\in T:\ell_T(t)=n\}$, and strong subtrees preserve the branching geometry in a rigid, balanced, level-respecting way (Dodos et al., 2011). If $\{A_t:t\in T\}$ is a family of measurable events in a probability space with $\mu(A_t)\ge \varepsilon>0$ for all $t$, then for every $0<\theta<\varepsilon$ there exists a strong subtree $S$ of infinite height such that for every strong subtree $R\subseteq S$ of height $k$,

$$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$

A corollary gives, for every $T(n)=\{t\in T:\ell_T(t)=n\}$0 and $T(n)=\{t\in T:\ell_T(t)=n\}$1, an exponent

$T(n)=\{t\in T:\ell_T(t)=n\}$2

such that after passing to a strong subtree $T(n)=\{t\in T:\ell_T(t)=n\}$3 of infinite height, every non-empty finite $T(n)=\{t\in T:\ell_T(t)=n\}$4 with $T(n)=\{t\in T:\ell_T(t)=n\}$5 satisfies

$T(n)=\{t\in T:\ell_T(t)=n\}$6

The same work isolates a large subclass of subsets, the free sets, for which the stronger bound $T(n)=\{t\in T:\ell_T(t)=n\}$7 holds (Dodos et al., 2011).

The historical reading is built into the notation. A node can be interpreted as a partial history, a path as a full history, and a finite strong subtree as a finite tree-shaped history. The theorem then states that inside any homogeneous tree of uniformly positive events there exists an infinite regular history space in which no finite local history is too negatively correlated. The proof uses Milliken’s tree theorem, generalized Shelah lines, and the fact that any finite subset of size $T(n)=\{t\in T:\ell_T(t)=n\}$8 lies in a strong subtree of height at most $T(n)=\{t\in T:\ell_T(t)=n\}$9 (Dodos et al., 2011). In this setting, “history” is not inferred retrospectively from leaves; it is already present in the indexing geometry.

3. Evolutionary genealogies, phylogenies, and reconciled histories

In evolutionary modeling, the distinction between genealogical history and phylogenetic history is central. The MRCAT algorithm records, at each generation $\{A_t:t\in T\}$0, a matrix $\{A_t:t\in T\}$1 whose entries are the number of generations back to the most recent common ancestor of individuals $\{A_t:t\in T\}$2 and $\{A_t:t\in T\}$3; for asexual reproduction,

$\{A_t:t\in T\}$4

while sexual models admit maternal, paternal, and general genealogical variants (Costa et al., 2017). The SSEE algorithm instead records species-level history directly through a branching-time matrix $\{A_t:t\in T\}$5 and an extinction vector $\{A_t:t\in T\}$6, preserving all extant and extinct species. In the formulation of that work, MRCAT yields an exact record of ancestry among extant individuals, whereas SSEE yields the “true phylogeny” at the species level, because it records speciation and extinction events directly (Costa et al., 2017).

A more axiomatized evolutionary use of the triad appears in event-labeled gene trees. A gene tree $\{A_t:t\in T\}$7 has internal vertices labeled as speciation $\{A_t:t\in T\}$8, duplication $\{A_t:t\in T\}$9, or horizontal transfer $\mu(A_t)\ge \varepsilon>0$0, leaves labeled $\mu(A_t)\ge \varepsilon>0$1, and edges labeled as transfer or non-transfer (Nøjgaard et al., 2017). A reconciliation map

$\mu(A_t)\ge \varepsilon>0$2

embeds gene-tree vertices into vertices or edges of a species tree $\mu(A_t)\ge \varepsilon>0$3, with leaves mapped to species, speciation events mapped to species-tree lca nodes, and duplication or HGT events mapped to species-tree edges. Time consistency is defined by time maps $\mu(A_t)\ge \varepsilon>0$4 and $\mu(A_t)\ge \varepsilon>0$5: if $\mu(A_t)\ge \varepsilon>0$6, then $\mu(A_t)\ge \varepsilon>0$7, ხოლო if $\mu(A_t)\ge \varepsilon>0$8 and $\mu(A_t)\ge \varepsilon>0$9, then

$t$0

The existence of a time-consistent reconciliation can be decided in $t$1 time by checking acyclicity of a small auxiliary graph rather than constructing explicit timing maps (Nøjgaard et al., 2017). When the species tree is unknown, a cubic-time algorithm decides whether a time-consistent binary species tree exists for a given event-labeled gene tree and constructs one when it does (Lafond et al., 2019).

Deep coalescence complicates the relation between networks and trees. In phylogenetic networks, the set of displayed trees $t$2 is obtained by deleting one incoming edge at each reticulation and contracting degree-$t$3 nodes, but this set is not sufficient in the presence of coalescence effects (Zhu et al., 2016). The paper introduces parental trees $t$4 by first unrolling the network into a MUL-tree and then pruning according to the sampling scheme. The inclusion $t$5 is strict in general, and anomaly zones are defined relative to $t$6: a gene tree is anomalous if its probability exceeds that of every parental tree (Zhu et al., 2016). This reframes the history question. The relevant historical object is neither a single species tree nor merely the set of displayed trees, but a reticulate history whose parental trees form a mixture model for gene genealogies.

4. Cosmological merger trees and assembly histories

In cosmology, merger trees play the same mediating role between local events and macroscopic history. A merger tree records the sequence by which a present-day dark matter halo assembled through progenitor halos, mergers, and accretion (Leisher et al., 7 Nov 2025). In the DREAMS warm-dark-matter simulations, the tree is built from 91 snapshots from $t$7 to $t$8; SUBFIND identifies halos and subhalos, and SubLink connects them across snapshots by particle sharing and binding-energy ranking (Leisher et al., 7 Nov 2025). Each node is a subhalo at a specific snapshot, and each edge is a progenitor–descendant link. In this representation, branch depth, branching multiplicity, and node ordering in snapshot index $t$9 jointly encode the assembly history.

The same literature emphasizes that merger trees are not only data structures for semi-analytic models but also diagnostic objects. “Observing Merger Trees in a New Light” introduces dendograms, which visualize the full merger history of a main branch together with subhalo orbits, halo merger events, and the evolution of halo properties (Poulton et al., 2018). The paper defines branch, main-branch, merged-branch, and interacting-branch using progenitor, descendant, Start/Leaf-Progenitor, and End/Root-Descendant links, and uses dendograms to expose over-merging, truncation, branch swapping, mass-definition artifacts, and subhalo tracking failures in VELOCIraptor, Rockstar, and AHF pipelines (Poulton et al., 2018). Here, history is read off from time–radius–mass–state plots rather than from abstract topology alone.

Deep-learning work on merger trees extends this historical semantics into parameter inference. Merger trees are represented as directed graphs $0<\theta<\varepsilon$0, with node features such as snapshot number $0<\theta<\varepsilon$1, dark-matter mass $0<\theta<\varepsilon$2, stellar mass $0<\theta<\varepsilon$3, gas mass $0<\theta<\varepsilon$4, and star formation rate, and a message-passing GNN predicts warm dark matter mass and feedback parameters (Leisher et al., 7 Nov 2025). The reported $0<\theta<\varepsilon$5 for warm dark matter mass ranges from approximately $0<\theta<\varepsilon$6 to $0<\theta<\varepsilon$7, depending on graph complexity and node features; even with no meaningful node features, the model reaches $0<\theta<\varepsilon$8, and with only snapshot number it reaches $0<\theta<\varepsilon$9 (Leisher et al., 7 Nov 2025). Sensitivity tests show that using only the main branch fails ($S$0), whereas pruned trees lose crucial information and flattened trees retain surprisingly high performance, indicating that the number and temporal distribution of low-mass progenitors carry much of the cosmological signal (Leisher et al., 7 Nov 2025). The paper’s central claim is that the structure of merger trees alone inherits information about the cosmological parameters of the simulations from which they form.

5. Distributed histories and history-aware transition systems

In anonymous communication networks, a history tree $S$1 is an infinite rooted tree whose nodes at level $S$2 are equivalence classes of agents that are indistinguishable at time $S$3 (Viglietta, 2024). Black edges connect a class at level $S$4 to the finer classes into which it splits at level $S$5, and red edges encode message receptions: a red edge $S$6 with multiplicity $S$7, where $S$8 and $S$9, means that each agent represented by $R\subseteq S$0 receives exactly $R\subseteq S$1 identical messages from agents represented by $R\subseteq S$2 at step $R\subseteq S$3. If $R\subseteq S$4 are the children of $R\subseteq S$5, then their anonymities satisfy

$R\subseteq S$6

For an agent $R\subseteq S$7, the view $R\subseteq S$8 is the portion of $R\subseteq S$9 spanned by all directed paths from the root to the node representing $k$0 at time $k$1; the paper states that this view contains all the information that $k$2 can possibly extract from the network after $k$3 communication steps (Viglietta, 2024).

The same structure supports exact algorithmic bounds. In connected dynamic networks, if $k$4 is a non-branching level and there are red edges $k$5 and $k$6 in opposite directions, then

$k$7

This mass-conservation relation determines anonymity ratios across equivalence classes and leads to optimal deterministic algorithms for Average Consensus and Counting, with stabilization in $k$8 steps; matching lower bounds follow from indistinguishability of views (Viglietta, 2024). The framework generalizes to directed, semi-synchronous, asynchronous, and congested models by enriching edge labels or redefining rounds, while preserving the principle that deterministic computation is a function of histories represented as trees (Viglietta, 2024).

A related but more abstract use of historical data appears in well-structured transition systems with history. Configurations are pairs $k$9, where $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$0 is a base state and $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$1 is a history component built from event tokens, for example as a word $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$2 or a multiset $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$3 (Abdulla et al., 2015). Transitions update both state and history, typically in the form

$$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$4

When the base system is a WSTS, the history domain is equipped with a well-quasi-order, and the history update is monotone, the product order on $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$5 again yields a WSTS (Abdulla et al., 2015). This makes ordered and unordered historical properties amenable to coverability methods. In this setting, history ceases to be auxiliary trace metadata and becomes part of the formal semantics.

6. Alternative histories, history counts, and maximally probable tree topologies

A complementary line of work studies not how trees encode histories, but how many histories are compatible with a given tree and which tree topologies maximize that number. For recursively grown trees, a history is a sequence of attachment events

$$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$6

and the history degeneracy $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$7 is the number of distinct such sequences producing a labeled tree $$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$8 (Timár et al., 2020). For equiprobable-sequence models, exact linear-time message passing on the nonbacktracking matrix computes root probabilities, and the paper derives a stepwise most probable history rule: at each step, the probability that a frontier node is next is proportional to its downstream branch size, so the exact stepwise maximum-likelihood reconstruction chooses the candidate with the largest remaining subtree (Timár et al., 2020). The mean logarithmic number of alternative histories obeys

$$\mu\Big(\bigcap_{t\in R} A_t\Big)\ge \theta^{\,p(b,k)}, \qquad p(b,k)=\frac{(2^b-1)^k-1}{2^b-2}.$$9

and the paper reports an “uncertainty principle”: the inferrability of the root and that of the complete history trade off against one another (Timár et al., 2020).

A distinct but related combinatorial notion appears in $T(n)=\{t\in T:\ell_T(t)=n\}$00-furcating trees. A labeled history is a bijection from internal nodes to $T(n)=\{t\in T:\ell_T(t)=n\}$01 that is decreasing along descendant relations, so labeled histories are exactly the linear extensions of the ancestor partial order (Dickey et al., 7 Feb 2026). For a rooted strictly $T(n)=\{t\in T:\ell_T(t)=n\}$02-furcating labeled topology $T(n)=\{t\in T:\ell_T(t)=n\}$03, the number of labeled histories is

$T(n)=\{t\in T:\ell_T(t)=n\}$04

where $T(n)=\{t\in T:\ell_T(t)=n\}$05 is the number of descendant leaves of internal node $T(n)=\{t\in T:\ell_T(t)=n\}$06 (Dickey et al., 7 Feb 2026). Using a connection with Huffman trees, the paper identifies a unique maximally probable unlabeled topology $T(n)=\{t\in T:\ell_T(t)=n\}$07 for every $T(n)=\{t\in T:\ell_T(t)=n\}$08, generalizing the Harding–Hammersley–Grimmett result for bifurcating trees (Dickey et al., 7 Feb 2026). It also formulates Conjecture 6.1 for tie-permitting labeled histories, where simultaneous branching events are allowed across incomparable nodes (Dickey et al., 7 Feb 2026). In this line of work, “history” is a ranking of branching events, and the tree topology is evaluated by the size of its admissible history set.

Taken together, these results show that history ambiguity is not an incidental nuisance but a measurable combinatorial property. In some settings, as in homogeneous trees of measurable events, one seeks structured subtrees where all finite local histories have controlled probability (Dodos et al., 2011). In others, as in recursive growth and $T(n)=\{t\in T:\ell_T(t)=n\}$09-furcating branching, one counts the number of admissible histories and asks which topologies maximize or suppress that count (Timár et al., 2020, Dickey et al., 7 Feb 2026). A plausible implication is that “Events–Trees–Histories” names a common research program: to understand how discrete event systems leave recoverable, regular, or ambiguous traces when their temporal development is compressed into tree structure.