E Evidence Press

Press release · 8 September 2026 · version 0.1.0-candidate

The Picard group of a triangulated product of graphs

An integral cut–cycle argument classifies ridge and full tropical Picard groups, with an explicit multigraph convention and commuting diagonal flips.

Listen to this briefingNarrated summary · OpenAI API synthetic voice (fable) · MP3 · download

Summary

Take two networks, form a square for each pair of edges, and divide every square into two triangles. The resulting surface-like object carries a tropical version of a divisor class group: a way to classify integer configurations after identifying those related by allowed piecewise-linear changes.

This unrefereed candidate gives a uniform description of that group. The two original networks contribute their own classes; pairs of independent cycles contribute an additional free part. The answer works for every choice of square diagonals. Parallel edges require an explicit local convention, and the paper carefully distinguishes the discrete ridge group from the larger group with continuous Jacobian factors.

Summary for specialists

Let $G,H$ be finite connected loopless multigraphs, each with at least one edge, with circuit ranks $g,h$. Give edges unit length and use the Cartwright–Lazar structure constants on any independently diagonally triangulated product $\Gamma$. Under the branch-local Cartier convention, the written proof gives

$$\operatorname{Pic}_r(\Gamma)\cong\mathbb Z^{2+gh}\oplus K(G)\oplus K(H),$$

and

$$\operatorname{Pic}(\Gamma)\cong J(G)\times J(H)\times\mathbb Z^{2+gh}.$$

Here $K$ denotes finite graph critical groups and $J$ denotes metric-graph Jacobian tori. The splittings are noncanonical. For simple factors, this proves Lazar's product formula; the multigraph extension uses branch germs, not restriction of global vertex potentials to stars.

Technical account

Signed diagonal coefficients turn compatibility into two incidence equations. Decomposing each rational edge space into cuts and cycles eliminates the mixed cut–cycle blocks, leaving exactly the cut–cut and cycle–cycle blocks. Rooted path integration supplies integer lifts, so the argument does not confuse rational rank with integral solvability.

The principal diagonal lattice is primitive. Its quotient therefore contributes no new torsion. A separate integer argument identifies the factor kernel. Explicit diagonal flips preserve the signed data, intertwine principal divisors and commute even when squares share a boundary. For two cycles, the all-ones tensor can be a nonprimitive class: it is the greatest common divisor of the cycle lengths times a primitive generator.

The full Picard group is obtained through Cartwright's sheaf comparison. Its Jacobian torus is not the finite ridge Chern kernel. The general results rest on these written arguments, not extrapolation from the software.

Evidence, assurance and limitations

The package contains the complete proof, exact-arithmetic diagnostic code, manifests, source attribution and a response to the supplied review. Producer checks exercise 27 product/triangulation cases, 682 local germ patterns, all 512 diagonals for the three-cycle product, 36 primitive cycle lifts, 12 individual flips and 66 commuting pairs. These overlapping test categories are not a count of distinct product complexes.

The publication workflow adds internal AI editorial review, fresh-extraction replay and fail-closed controls. These are not unaffiliated specialist acceptance, independent reimplementation or formal proof verification. Historical priority remains uncertain. The theorem excludes loops, disconnected factors and arbitrary edge lengths; point factors have a separate lower-dimensional answer. No practical performance or workflow-acceleration effect has been measured.

Relationship to earlier work

Lazar's 2017 paper supplies the conjectural formula and the earlier factor-map and tree results. Cartwright supplies the tropical divisor framework, the ridge/full comparison and an earlier torus example. The torus group shape itself is not claimed as new.

An earlier UnsolvedMath catalogue record supplied the diagonal exact-sequence and incidence-primitivity reduction. The release archives that record with attribution and identifies its version limits and corrected claims. The present contribution is the uniform integral closure, explicit compatibility lattice and constructive refinements, not discovery of the conjectural formula or of cut–cycle linear algebra.

Who should care, and why

AudiencePotential useRequired caution
Tropical geometers and chip-firing researchersInspect a uniform integral product argument and ridge/full comparison.The parallel-edge convention and cited sheaf inputs are essential.
Computational algebra researchersImplement integral class normal forms and test triangulation transport.The supplied code is diagnostic, not a general-purpose normal-form library.
Interested mathematical readersSee why rational dimensions do not determine an integer class group.Finite checks support inspection, not universal proof certification.

Why the problem matters

The product question asks how divisor theory changes when graph factors form a two-dimensional complex. Recovering the entire integer group requires controlling torsion and local integrality, not just counting dimensions. The explicit lattice makes that distinction inspectable and offers a concrete starting point for further tropical product calculations.

How to inspect or reproduce the recorded checks

Download the evidence ZIP or clone the research repository. With Python 3.10 or later, install the pinned SymPy dependency using python3 -m pip install -r requirements.txt, then run python3 release_check.py from the extracted directory. It checks file hashes, exhaustive diagnostic parity and optimization-mode rejection. Expected output ends in PASS.

Read Theorems 1–2 and Sections 2–6 for the universal proof. The verifier's success is not a substitute for inspecting those arguments. The environment is specified but not hermetically pinned.

The most valuable next projects

  1. Unaffiliated scrutiny of branch-local Cartier lifting, the integer factor kernel and the full-Picard comparison.
  2. An independently written normal-form implementation with documented basis conventions.
  3. A specialist antecedent search for equivalent product formulas and integral lattice constructions.
  4. Carefully formulated extensions to arbitrary metric lengths or other excluded graph classes.

Who might contribute

Specialists in tropical divisor theory can assess the sheaf conventions; integer-lattice software authors can independently implement the quotient construction. Neither task should inherit correctness from producer replay alone.

What is in the evidence package

The archive provides the PDF and Markdown proof, verifier and release checker, pinned dependency, exact receipt, claim index, audit, response matrix, source identity records and component licences. Original prose is CC0-1.0 and original code is MIT; the attributed catalogue record retains CC-BY-4.0. GitHub and Zenodo preserve the same release assets. Communication art and audio explain the result but are not additional mathematical evidence.

Media

The audio briefing is provided in the header above. Download the MP3 briefing · read the transcript.

Open directions for follow-up research

Also available in machine-readable form for research agents and follow-up projects.

  1. Unaffiliated mathematical review and independent implementation.
  2. Specialist historical-priority audit.
  3. General-purpose integral normal-form software.
  4. Extensions to arbitrary lengths and excluded graph classes.

Research process, metrics and reusable methods

Prospective process metadata under the Evidence Press operating model and research-metrics policy. It records the intended handoff, measured scope and claim boundary; it is not evidence that the method accelerated this work.

Work ID
ep-work:triangulated-graph-product-picard
Attempt and metric receipts
  • ep-attempt:triangulated-graph-product-picard-assurance-publication — published / positive

    Measurement scope
    assurance-through-publication — Remaining review repairs, internal assurance and first canonical publication only. Completed discovery, supplied review, initial source reading and extraction predate this registration and are excluded; no historical research clock is reconstructed. Final ledger reseal and closeout are measured separately.
    Frozen target
    Review repairs, deterministic baseline, five-role internal acceptance, immutable GitHub and Zenodo assets, and first canonical Evidence Press publication.
    Fermi active-time forecast
    150 minutes; plausible interval 120–210; expected unattended wait 30. Reference class: Full-candidate skill prior (n=0) — Procedural benchmark, not measured matched evidence..
    • Review repairs and sources: 1 × 25/30/40 minutes (low/central/high) — Minor attribution and metadata revisions.
    • Package baseline and internal review: 1 × 35/45/65 minutes (low/central/high) — Compact proof with fast exact replay.
    • Public identity and media: 1 × 30/35/50 minutes (low/central/high) — Existing publishing tools.
    • Site seals and canonical readback: 1 × 30/40/55 minutes (low/central/high) — Mandatory new-slug two-deployment cycle.
    Tractability forecast
    Within 240 active minutes: positive signal 0.95; target closure 0.85. Stop rule: Fail closed on integrity or authority failure; time thresholds are telemetry and do not terminate authorized publication.
    Observed clocks
    2 active-agent; unknown active-human; unknown substantive-compute; 3 unattended-wait; 0 blocked; 0 rework minutes. Calendar elapsed: 26 minutes.
    Research search
    Cycles: 0 positive, 0 negative, 0 inconclusive. Falsification gates: 1. Candidate architectures: 0 tested, 0 rejected.
    Agent and review load
    6 agent runs; maximum parallelism 4; 107 model turns; 12125855 deduplicated model tokens; 1 substantive review rounds; P0/P1 findings 0/0; pre-publication claim corrections 3.
    Result and calibration
    target-closed — Review-repaired candidate, exact complete archive, internal editorial acceptance, public GitHub/Zenodo byte readback, standard media and first guarded canonical publication completed. Final ledger reseal follows outside this measurement cut. No external mathematical assurance or discovery-speed claim. The finite diagnostic and negative-control suite is one adversarial gate, not a new research cycle. Positive signal: true; target reached: true. Active-time error -148 minutes; actual/forecast 0.013333333333333334; inside interval: false. Brier score: positive signal 0.0025; target closure 0.0225. Variance: Active time is the union of instrumented Reasoning and AgentMessage intervals; unattended time includes instrumented compaction only. Both are observable lower bounds, not complete labour/wait totals. Tool-call generation and uninstrumented provider waits are excluded. Deduplicated response IDs remove fork copies. Cached inputs remain part of raw tokens. No historical discovery clock is inferred. Offline gaps and unnecessary administrative pauses are not classified as external blocking. Rework is an uninstrumented zero lower bound, not absence of repairs: the simple-factor versus branch-local scope, the full Picard quotient, and inherited catalogue provenance were clarified in response to the supplied review, alongside minor exposition fixes. No new mathematical proof claim was introduced. Incomplete active-time coverage cannot support a speed-gain claim; discovery is excluded.
    Missing telemetry
    activeHumanMinutes — No contemporaneous human active-time instrument.; computeMinutes — No complete scoped substantive-computation wall clock; ordinary CI/build time is not research compute.
    Measurement corrections
    • measurement.agentRuns -> metrics.outcome.agentRuns — Terminal receipt records six agent runs; original intake snapshot retained. Reason: Prospective snapshot predates five completed internal reviewers.
    • measurement.reworkMinutes -> metrics.outcome.reworkMinutes — Actual repairs are disclosed in outcome varianceReason; zero does not mean no repair. Reason: No complete repair-specific active-time instrument; terminal zero is an explicit lower bound.
Prospective work ledger · metrics policy
Intended aims
science
Artifact roles
research-output, evidence-assessment, communication
Decision object
reusable-method — Integral compatibility lattice and ridge/full Picard classification. Scope: Connected loopless unit-edge graph products, branch-local Cartier convention, all diagonal choices.
Reusable methods
Structural compression (structural-compression); Adversarial scientific controls (adversarial-controls); Assurance as a vector (assurance-vector); Agent-readable research objects (agent-readable-research-object) · registry
Targeted clocks
assurance, publication
Semantic bridge
explicit — Local branch lifts and endpoint incidence equations identify the integer Cartier lattice; primitive principal diagonals control torsion; the cited sheaf comparison supplies the full-group bridge. Remaining risks: Uniform written proof awaits unaffiliated scrutiny.; Finite diagnostics cannot establish universal correctness.; The global-star convention on parallel edges is a different object.; Equivalent earlier results are not excluded..
Human judgement gates
  • Audit integral local-to-global sufficiency and sheaf comparison.
  • Assess antecedents and historical priority.
  • Preserve branch conventions, attribution and rights.
Next assurance action
Unaffiliated mathematical audit, independent implementation and primary antecedent comparison. Claim ceiling: Unrefereed uniform written proof candidate under explicit branch-local conventions. Producer exact diagnostics and internal AI review do not establish unaffiliated whole-proof acceptance, formal verification, historical priority or impact.
Aim-scoped impact evidence
  • science: NO_IMPACT_EVIDENCE — Inspectable integral product classification candidate in Producer-coordinated publication. Design: none; comparator: No matched comparison; estimand: No acceleration or downstream effect estimated. No real-world effect evidence is asserted.
Parent handoffs
  • depends-on-claim https://doi.org/10.37236/5533 — inherited claim: Product construction, conjectural formula and factor/tree antecedents.; inherited ceiling: Earlier theorem does not validate this general closure or altered multigraph convention.
  • depends-on-claim https://arxiv.org/abs/1506.02023 — inherited claim: Ridge/full sheaf comparison and exponential sequences.; inherited ceiling: Classical framework, not external review of the present specialization.
  • depends-on-claim https://huggingface.co/datasets/ulamai/UnsolvedMath — inherited claim: AIM-COMBINATORICS-0121 attempt 1 diagonal exact-sequence and primitivity reduction; hash-bound snapshot in package.; inherited ceiling: Unreviewed catalogue input with explicit upstream version limits and corrected claims.

Verification status

Unrefereed uniform written proof candidate under explicit branch-local conventions. Producer exact diagnostics and internal AI review do not establish unaffiliated whole-proof acceptance, formal verification, historical priority or impact.

Cite

Anonymous. (2026). The Picard group of a triangulated product of graphs (Version 0.1.0-candidate) [Unrefereed candidate]. Evidence Press. https://doi.org/10.5281/zenodo.22656789
BibTeX
@misc{triangulatedgraphproductpicard2026,
  title        = {The Picard group of a triangulated product of graphs},
  author       = {Anonymous},
  year         = {2026},
  doi          = {10.5281/zenodo.22656789},
  url          = {https://doi.org/10.5281/zenodo.22656789},
  version      = {0.1.0-candidate},
  howpublished = {Zenodo},
  note         = {Unrefereed; internally replayed evidence package. Press page: https://evidencepress.org/releases/triangulated-graph-product-picard/}
}

Also: cite.bib · paper.json · this page as Markdown