A Symmetry-Based Orthogonal Decomposition of Polytope Occupancy Vectors for Multi-Component Detection of Temporal Structure in Earthquake Magnitude Sequences

We introduce a detector that separates the temporal structure of an earthquake catalog's magnitude sequence intoa small number of interpretable, mutually orthogonal physical dimensions. The prior framework (WCM-PL)embeds a sequence in a 4-dimensional Takens delay space, assigns each point to a vertex of the 24-cell (theregular 4-polytope with 24 vertices), and collapses the resulting occupancy vector to a single scalar - the cosinedistance from the uniform distribution. We show, on both synthetic and real data, that this single-scalar collapseconflates several qualitatively distinct temporal structures (correlation sign, time scale, hyperuniformity, andtemporal asymmetry) into one number, making physically unrelated earthquakes look alike and same-typeearthquakes look different. To undo the conflation we decompose the 24-dimensional occupancy space into theirreducible representations of the signed-permutation group B4 (order 384), which acts transitively on the 24vertices. The space splits cleanly into six orthogonal blocks - uniform(1), sign(6), axispair_a(3), axispair_b(2),resid8(8), resid4(4) - and standardizing each block's energy against a value-multiset-fixed (full-permutation) nullyields a signed multi-component fingerprint. Applied to over one hundred real earthquakes (a global candidatepool, major Japanese JMA events, and USGS global M7.8+ events), the five informative blocks behave nearlyorthogonally (max inter-component correlation 0.51 raw, 0.36 robust), and earthquake affinities that a singlescalar cannot express appear differently in different components. In particular the time-scale component providesan identification axis statistically independent of the conventional index, placing a strongly-structured and anegatively-structured earthquake at the same coordinate. A hypothesis-driven split of one residual block, backedby an explicit null test, adds a seventh "arrow-of-time" component. Our claim is that the new detector separatesseismic temporal structure into orthogonal physical dimensions; we do not claim an exhaustive classification of allearthquakes, nor precursory prediction. --- Data ---Two earthquake catalogs are used and are NOT redistributed in this record:- Global catalog (USGS-derived), 1986-2026, public domain (earthquake.usgs.gov).- JMA unified hypocenter catalog, 1998-2023, obtained per JMA terms of use.Extraction query (identical for every event): all catalog events within 180 daysafter and 100 km (haversine) of the epicenter, sorted by time; the magnitudecolumn of that window is the analyzed sequence. --- Licenses ---Code (code/ in the repro package): AGPL-3.0-or-later.Paper, figures, README, derived data: CC BY 4.0.The WCM-PL core module is published separately atgithub.com/watabe-masanori/cds-polytope (AGPL-3.0-or-later).

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-08-14
DOI
https://doi.org/10.5281/zenodo.21929240
Primary Topic
Earthquake Detection and Analysis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
preprint

A Symmetry-Based Orthogonal Decomposition of Polytope Occupancy Vectors for Multi-Component Detection of Temporal Structure in Earthquake Magnitude Sequences

Masanori Watabe, Claude Opus 4.8 Kurado
Zenodo (CERN European Organization for Nuclear Research)
Earthquake Detection and Analysis
preprint

A Symmetry-Based Orthogonal Decomposition of Polytope Occupancy Vectors for Multi-Component Detection of Temporal Structure in Earthquake Magnitude Sequences

Masanori Watabe, Claude Opus 4.8 Kurado
preprint en

Abstract

We introduce a detector that separates the temporal structure of an earthquake catalog's magnitude sequence intoa small number of interpretable, mutually orthogonal physical dimensions. The prior framework (WCM-PL)embeds a sequence in a 4-dimensional Takens delay space, assigns each point to a vertex of the 24-cell (theregular 4-polytope with 24 vertices), and collapses the resulting occupancy vector to a single scalar - the cosinedistance from the uniform distribution. We show, on both synthetic and real data, that this single-scalar collapseconflates several qualitatively distinct temporal structures (correlation sign, time scale, hyperuniformity, andtemporal asymmetry) into one number, making physically unrelated earthquakes look alike and same-typeearthquakes look different. To undo the conflation we decompose the 24-dimensional occupancy space into theirreducible representations of the signed-permutation group B4 (order 384), which acts transitively on the 24vertices. The space splits cleanly into six orthogonal blocks - uniform(1), sign(6), axispair_a(3), axispair_b(2),resid8(8), resid4(4) - and standardizing each block's energy against a value-multiset-fixed (full-permutation) nullyields a signed multi-component fingerprint. Applied to over one hundred real earthquakes (a global candidatepool, major Japanese JMA events, and USGS global M7.8+ events), the five informative blocks behave nearlyorthogonally (max inter-component correlation 0.51 raw, 0.36 robust), and earthquake affinities that a singlescalar cannot express appear differently in different components. In particular the time-scale component providesan identification axis statistically independent of the conventional index, placing a strongly-structured and anegatively-structured earthquake at the same coordinate. A hypothesis-driven split of one residual block, backedby an explicit null test, adds a seventh "arrow-of-time" component. Our claim is that the new detector separatesseismic temporal structure into orthogonal physical dimensions; we do not claim an exhaustive classification of allearthquakes, nor precursory prediction. --- Data ---Two earthquake catalogs are used and are NOT redistributed in this record:- Global catalog (USGS-derived), 1986-2026, public domain (earthquake.usgs.gov).- JMA unified hypocenter catalog, 1998-2023, obtained per JMA terms of use.Extraction query (identical for every event): all catalog events within 180 daysafter and 100 km (haversine) of the epicenter, sorted by time; the magnitudecolumn of that window is the analyzed sequence. --- Licenses ---Code (code/ in the repro package): AGPL-3.0-or-later.Paper, figures, README, derived data: CC BY 4.0.The WCM-PL core module is published separately atgithub.com/watabe-masanori/cds-polytope (AGPL-3.0-or-later).

Zenodo (CERN European Organization for Nuclear Research)
Sustainable cities and communities
Earthquake Detection and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.