Chronotome: Dynamic Temporal Framework

• Chronotome is an emergent concept defined by its use in real-time topic modeling for streaming embedding spaces, revealing dynamic semantic evolution.
• It extends to describe time-dependent dynamical systems with moving attractors, intrinsic clocks in artificial systems, and communication tempos across species.
• Furthermore, Chronotome operationalizes temporal state through measures in subjective time perception and decentralized clock synchronization in multi-agent networks.

Chronotome is an emergent, polysemous term in recent research. In its most explicit usage, it denotes a real-time topic-modeling and visualization system for streaming embedding spaces that represents semantic similarity in the $X$–$Y$ plane and time on the $Z$-axis (Lim et al., 1 Sep 2025). In a broader interpretive sense, the term has been used to illuminate temporal organization itself: a stable moving attractor in nonautonomous oscillatory dynamics, a minimal intrinsic clocking capacity in artificial living systems, a cross-species communication tempo regime, and operational estimates of subjective or distributed time in human and multi-agent settings (Suprunenko et al., 2014, Robu et al., 2017, Amichay et al., 29 Aug 2025, Aust et al., 2024, Zegers et al., 6 Apr 2025). The literature does not provide a single formal cross-domain definition; instead, it converges on a family of problems concerned with how temporal structure is represented, stabilized, inferred, and acted upon.

1. Chronotome as a real-time spatial-temporal map of embeddings

In "Chronotome: Real-Time Topic Modeling for Streaming Embedding Spaces" (Lim et al., 1 Sep 2025), Chronotome is a real-time topic modeling and visualization system for datasets in which each item has an embedding vector and a timestamp. The system takes time-stamped objects $x_i=(t_i,o_i,e_i)$, where $t_i$ is the timestamp, $o_i$ the object itself, and $e_i$ the vector embedding, and produces a dynamic graph $G$ whose nodes occupy a 3D spatial-temporal layout. Its stated purpose is to make semantic change over time visible and interpretable rather than to display a static similarity structure.

The core construction combines force-based dimensionality reduction in the spatial plane with streaming clustering over time. Semantic similarity is encoded in the $X$–$Y$ layout by a spring model in which cosine similarity $Y$0 defines both the ideal pairwise distance and the spring constant: $Y$1 with Hooke-style force

$Y$2

To suppress weak relations, the method applies a threshold

$Y$3

where $Y$4 is mean edge similarity, $Y$5 the standard deviation of edge similarities, $Y$6 the number of nodes, and $Y$7 is set empirically to $Y$8. Pairs above $Y$9 create attractive spring forces; otherwise they receive a high repulsive force. Time is then projected directly onto the $Z$0-axis by batching data into user-defined timesteps $Z$1 and assigning a shared $Z$2-coordinate to all points in the same batch (Lim et al., 1 Sep 2025).

The streaming update mechanism is designed to preserve temporal continuity in the viewer’s mental map. New batches insert nodes and edges incrementally, while older nodes become progressively more inert through a mass schedule

$Z$3

with $Z$4. At each timestep, the 2D embedding is clustered with HDBSCAN, and cluster labels are generated either by TF-IDF or by an LLM summarizing cluster contents; for images, an LLM first generates textual descriptions. The interface exposes three modes—Latest, Across, and Playback—and three preset camera views—Front View, Iso View, and Side View—together with convex hulls, delta cones or tunnels, legends, and a Data Gallery for direct inspection of cluster members (Lim et al., 1 Sep 2025).

The paper demonstrates the system on three datasets: 669 Bill Gates tweets, 761 Picasso paintings, and 1000 frames from The Grand Budapest Hotel. These case studies are presented as evidence that Chronotome can reveal recurrent themes, topic lineage, stylistic evolution, and visual narrative structure in text and image corpora. Its stated limitations are equally specific: the fully connected graph scales as $Z$5, current performance is limited to about 1000 points, and the system supports cluster splits but not merges (Lim et al., 1 Sep 2025).

2. Chronotome as time-dependent dynamical organization

A mathematically rigorous analogue of a chronotome appears in the theory of chronotaxic systems. These are deterministic nonautonomous oscillatory systems that resist continuous external perturbations while exhibiting complex time-dependent dynamics (Suprunenko et al., 2014). The defining structure is a time-dependent point attractor, or driven steady state, in a drive–response system

$Z$6

for which the response has an attracting and invariant trajectory $Z$7. The relevant limits are that trajectories forget initial conditions and converge toward the moving target state, while initialization on the attractor remains on the attractor (Suprunenko et al., 2014).

The paper generalizes earlier chronotaxic theory by removing the requirement that amplitude and phase dynamics be decoupled. Instead, stabilization is formulated through contraction theory. Infinitesimal deviations satisfy

$Z$8

with a sufficient condition given by uniform negativity of the maximal eigenvalue: $Z$9 An equivalent operational criterion is the existence of a contraction region $x_i=(t_i,o_i,e_i)$0 inside which a finite region $x_i=(t_i,o_i,e_i)$1 cannot be left by trajectories and within which the time-dependent point attractor exists. This matters because it allows chronotaxicity to be identified even when the attractor is not known in advance (Suprunenko et al., 2014).

The nonautonomous Poincaré oscillator is the paper’s canonical example. In rotating polar coordinates, the driven system is written as

$x_i=(t_i,o_i,e_i)$2

$x_i=(t_i,o_i,e_i)$3

with $x_i=(t_i,o_i,e_i)$4. The parameter-space geometry supports a three-way classification. Type I contains a time-dependent point attractor, a closed $x_i=(t_i,o_i,e_i)$5 trajectory, and a non-contraction region; Type II retains the point attractor but not the $x_i=(t_i,o_i,e_i)$6 structure, while still containing a non-contraction region; Type III is fully contracting throughout phase space. These types differ in transient geometry and in response to perturbations. In Type I and Type II, some distances between trajectories may temporarily increase because of non-contraction regions even though the system remains chronotaxic overall (Suprunenko et al., 2014).

This framework is directly relevant to broader chronotome interpretations because it supplies a precise dynamical-systems language for a stable, moving temporal backbone. The paper also discusses temporal chronotaxicity and interacting chronotaxic systems, suggesting that temporal organization may be interval-specific, scale-specific, or distributed across coupled subsystems rather than globally fixed (Suprunenko et al., 2014).

3. Chronotome as intrinsic time measurement in minimal systems

A second formalization arises from minimal abstract clocks in artificial living systems. "Time as it could Be measured in Artificial Living Systems" analyzes how small Markovian systems measure time intrinsically and evaluates them by the mutual information $x_i=(t_i,o_i,e_i)$7 between clock state $x_i=(t_i,o_i,e_i)$8 and true time $x_i=(t_i,o_i,e_i)$9 (Robu et al., 2017). The paper assumes a discrete global tick, a known starting time, and limited memory, so time measurement becomes probabilistic counting rather than symbolic chronology.

The most elementary cases are 2-state clocks. The alternator clock is a symmetric Markov chain with states $t_i$0 and $t_i$1, switching with probability $t_i$2: $t_i$3

$t_i$4

Its probability evolution is

$t_i$5

and the paper identifies five regimes: stuck, overdamped, critically damped, underdamped, and undamped. This clock is characterized as suitable for local time measurement, meaning phase within a cycle, such as odd versus even step (Robu et al., 2017).

The drop clock provides the complementary primitive. It begins in $t_i$6, makes a one-way transition $t_i$7 with probability $t_i$8, and remains in $t_i$9 thereafter. Its survival probability is

$o_i$0

This mechanism measures global time relative to a horizon—early versus late in an interval—rather than phase. The paper stresses that oscillatory and decay-based clocks are not interchangeable approximations of the same task; they instantiate different representational geometries of time (Robu et al., 2017).

The analysis then scales up to cascades of independent clocks, described as a "clock bag," and to composite clocks with constrained information flow. In the bag construction, the first optimized clock is an alternator, the second a drop clock, and almost all subsequent clocks are also drop clocks with similar parameters. In the composite case, the upper clock is fixed as an alternator while the lower clock depends on the joint state, and the optimization

$o_i$1

is implemented by a Lagrangian of the form

$o_i$2

The reported regimes include a perfect clock at high information flow, a soft or fuzzy clock at low flow, and intermediate Pareto-sensitive configurations (Robu et al., 2017).

This line of work supports a chronotome-like interpretation in which temporal organization is not a single clock signal but a structured ensemble of complementary primitives. A plausible implication is that chronotomes, in this sense, are modular, scale-sensitive, and information-efficient rather than unitary.

4. Chronotome as a cross-species communication tempo regime

A biologically grounded use of the concept appears in "A universal animal communication tempo resonates with the receiver's brain" (Amichay et al., 29 Aug 2025). The paper reports a field observation in Thailand in which Pteroptyx malaccae fireflies and nearby crickets produced nearly identical but not synchronized tempos, both near 2.4 Hz and differing by about 10%. This observation motivated a meta-analysis of peer-reviewed literature on isochronous animal communication.

The meta-analysis imposed explicit inclusion criteria because isochrony is inconsistently defined: at least 5 signal repetitions and at least 25% consistency in inter-onset intervals. The resulting survey spans visual, acoustic, and gestural communication across insects, crustaceans, amphibians, birds, fish, mammals including apes, humans, and sea lions, and examples from air, land, and sea across roughly 8 orders of magnitude in body weight. The paper’s main empirical claim is an abundance of evolutionarily distinct species communicating isochronously in a band of approximately 0.5–4 Hz, centered roughly around 2 Hz, which it describes as a frequency hotspot (Amichay et al., 29 Aug 2025).

The proposed mechanism is receiver-side rather than sender-side. The authors argue that many signalers can produce faster signals and that the diversity of signaling machinery across taxa makes a single production constraint unlikely. They therefore hypothesize that the relevant commonality lies in neural biophysics: neurons have intrinsic integration times of a few hundred milliseconds, and circuits built from such neurons should be most responsive to periodic inputs in the same 0.5–4 Hz range (Amichay et al., 29 Aug 2025).

To test that hypothesis, the paper uses Kuramoto oscillator networks as stylized receiver circuits: $G$0 Entrainment is measured by the order parameter $G$1 For $o_i$3 oscillators and all 1665 unique non-isomorphic directed graphs with 10 directed edges, topology has limited effect on entrainment at 2 Hz. For fully connected circuits, forcing-frequency sweeps from 0 to 4 Hz produce a clear resonance peak near the mean intrinsic frequency, about 2 Hz. Supplementary Arnold-tongue analysis reports only one tongue, centered at about 2 Hz, with no additional tongues at harmonics or subharmonics, and similar resonance persists in $o_i$4 networks with 20%, 50%, and 80% input neurons (Amichay et al., 29 Aug 2025).

Within the paper’s own framing, this shared temporal organization is a candidate chronotome: not a rigid universal law, but a hotspot, a probabilistic tendency, and a conserved temporal regime in which communication is especially effective.

5. Chronotome as operational temporal state in humans and networks

Two additional research directions operationalize temporal state in explicitly inferential and control-theoretic terms. In "Automatic Classification of Subjective Time Perception Using Multi-modal Physiological Data of Air Traffic Controllers" (Aust et al., 2024), the ChronoPilot project is described as aiming to develop a device that modulates human subjective time perception. The study’s purpose is to provide the feedback component required for that closed loop by estimating whether time was perceived as slow or fast from wearable physiology.

The experimental setting uses twelve professional air traffic controllers in flight controller helicopter training sequences at a Greek military airport, each completing four sessions over two consecutive days. Physiological modalities are photoplethysmogram, electrodermal activity, and temperature from a thermopile. Sensors come from an EmotiBit wearable device, with 3-wavelength PPG at 25 Hz, EDA at 15 Hz, and temperature at 7.5 Hz. The preprocessing pipeline includes manual artifact cleanup, filtering, segmentation for background subtraction, and feature extraction. The paper states that 24 features were extracted from PPG, EDA, and temperature. Labels were derived from verbal duration estimates and a 1-to-5 slow/fast rating, min-max-scaled per controller and thresholded at medium, where $o_i$5 was labeled fast, yielding 26 slow and 22 fast sequences over 48 total sequences with average length 182 s (Aust et al., 2024).

The machine-learning comparison evaluates eleven classifiers under leave-one-subject-out cross-validation. The strongest reported result is the support vector classifier with all features, reaching mean accuracy $o_i$6, summarized in the abstract as 79%; the majority-class baseline was 54%. LDA also reached $o_i$7 with PPG+EDA, and the paper’s overall interpretation is that EDA provides the most descriptive biomarker. The work is explicitly framed as a step toward closing the feedback loop of a device that would operate online, continuously, and individually for each user (Aust et al., 2024).

A distributed-systems analogue appears in "ChronoSync: A Decentralized Chronometer Synchronization Protocol for Multi-Agent Systems" (Zegers et al., 6 Apr 2025). Each agent has a disturbed hardware clock $o_i$8 with unknown drift $o_i$9, a steerable software clock $e_i$0, an event timer $e_i$1, and estimator states. The objective is practical synchronization of software clocks within user-specified tolerance $e_i$2, together with convergence of software clock drifts to a common user-defined drift $e_i$3. The control law is

$e_i$4

and communication is intermittent, directed in practice, and asynchronous despite an underlying static, connected, undirected graph. The closed loop is modeled as a hybrid system, and the target synchronization-estimation set is shown to be globally practically exponentially stable under the stated Lyapunov and matrix negativity conditions (Zegers et al., 6 Apr 2025).

Taken together, these two papers suggest two operational chronotome-like constructions: one in which temporal state is inferred from physiology to regulate subjective time, and another in which temporal state is synchronized across agents by hybrid consensus and drift estimation. In both cases, time is treated not merely as an external parameter but as a controllable or inferable internal variable.

6. Terminological boundaries and adjacent chrono-prefixed systems

Chronotome should be distinguished from adjacent chrono-prefixed nomenclature that addresses different technical problems. A prominent example is CHRONOS, the Cryogenic sub-Hz cROss torsion-bar detector with quantum NOn-demolition Speed meter, a proposed next-generation ground-based gravitational-wave detector for the unexplored 0.1–10 Hz band (Inoue et al., 27 Sep 2025, Inoue et al., 7 Apr 2026). CHRONOS combines cryogenic torsion-bar test masses with a triangular or ring-cavity Sagnac interferometer and a speed-meter readout designed to suppress quantum radiation-pressure noise and implement a QND measurement in the rotational degree of freedom.

The 2025 proposal estimates strain sensitivities of $e_i$5 at 1 Hz for detectors with arm lengths of 2.5 m, 40 m, and 300 m, together with reach to intermediate-mass black hole binaries out to 380 Mpc at $e_i$6 and stochastic-background sensitivity down to $e_i$7 at 0.2 Hz with 5 yr accumulation (Inoue et al., 27 Sep 2025). The 2026 science case presents the same detector family as targeting strain sensitivity $e_i$8 around 2 Hz and stochastic gravitational-wave background sensitivity $e_i$9 at 2 Hz, with the broader motivation of bridging the observational gap between LISA and terrestrial interferometers (Inoue et al., 7 Apr 2026).

The relevance of such adjacent systems is terminological rather than definitional. CHRONOS is not a chronotome in the sense used by the visualization, dynamical-systems, or artificial-life literature. Nevertheless, its inclusion underscores a broader pattern in which chrono-prefixed concepts are attached to questions of temporal access, temporal measurement, or time-structured observables. A plausible implication is that "chronotome" currently functions less as a settled technical term than as a convergent label for temporal organization across heterogeneous domains: semantic evolution, driven attractor dynamics, intrinsic clocks, communication tempo, subjective-time feedback, and distributed synchronization.