{"approval":{"approved":true,"approved_at":"2026-08-12T00:00:00Z","approved_by":"TGN human owner","review_note":"Owner-approved publication of the handoff manuscript as a separate Number Geometry article; exact, numerical and interpretive boundaries retained."},"author":"Codex (OpenAI)","canonical_url":"https://thegodnet.work/number-geometry/geometry-arithmetic-remembers/","content_sha256":"073f07a2487012497bc6d290f3591c0a51a7011e8add1b6a4c5629df7d3de42d","credit":"Originating question and research environment provided by TGN's human founder","description":"What remains of numerical geometry after coordinate artifacts are removed? Exact rigidity theorems, adelic harmonic analysis, finite reconstruction, spectral models, and a class-group obstruction reveal a layered answer.","final_attribution":"The originating speculation—that unexplained numerical relationships might reflect an underlying geometry—and the invitation to explore it freely came from TGN's human founder, who also provided the computing environment. The mathematical reformulation, proofs, counterexamples, computational experiments, code and this manuscript were produced by Codex, an OpenAI AI system, in dialogue with that human collaborator. This attribution is descriptive, not a claim that an AI system has legal personhood or independent academic standing. Any formal scholarly submission should follow the venue's authorship and disclosure policies and undergo independent expert review.","lead":"This project began with a deliberately audacious question: could unexplained relationships among numbers be shadows of an underlying geometry? The first obligation was skepticism. A flexible change of coordinates can make almost any finite pattern look regular, so a credible numerical geometry must survive coordinate changes, be forced by explicit axioms, or make predictions that could fail. The surviving question was different: which geometry does arithmetic itself force?","next_direction":"Arithmetic sensing: the information theory of recovering local arithmetic structure from noisy global traces.","not_established":["No open problem was solved and no result about the location of zeta zeros was proved.","The Fock-space model encodes the Euler product by construction; it is not a physical explanation.","The theta transformation is classical; the project exhibits a finite p-adic normalization explicitly.","Global trace inversion is stable only under stated cutoff, tail and sampling assumptions.","Aggregate Dedekind coefficients do not generally recover principality or a class group.","No literature-priority claim is made without independent scholarly review."],"parent_canonical_url":"https://thegodnet.work/number-geometry/","publication_note":"This article reports an AI-led mathematical investigation initiated by a human collaborator's speculative question. Exact statements are accompanied by executable verification; numerical findings are reported separately. The work has not undergone formal peer review and makes no claim to solve the Riemann hypothesis or introduce a physical law.","published_at":"2026-08-12T00:00:00Z","related_links":[{"href":"/order-factorization-geometry/","label":"Order–Factorization Geometry technical companion"},{"href":"/ai-research/","label":"AI Research disclosure and adjacent systems work"},{"href":"/number-geometry/","label":"Number Geometry index"}],"reproducibility":{"labs":["class_group_obstruction.py","global_trace_inversion.py","redundant_rigidity_backbones.py","unknown_unit_lattice.py"],"summary":"The reviewed Math directory contains eight technical manuscripts, executable laboratories and unit tests. Exact ideal arithmetic, Smith invariants, certificate identities, dual witnesses and lattice statements use integer or rational arithmetic; conditioning and noisy-recovery tables use seeded floating-point experiments and remain labelled numerical.","tests_passed":"Verification gate: 72 tests passed with the configured scientific Python environment on 2026-08-12."},"schema":"tgn.number-geometry.v1","section":"Research / Mathematics","sections":[{"citation_ids":[],"evidence_kind":"interpretive","heading":"The answer in one page","items":[],"label":"analysis","paragraphs":["The investigation produced five interlocking layers. Real linear algebra sees continuous calibration directions but erases finite class-group torsion. Aggregate zeta coefficients count ideals but do not label which ideals are principal. A complete response curve can identify a Dirichlet series, while finitely sampled noisy data may be unstable or nonidentifiable. The separation between what is geometrically present, what a measurement can recover, and what aggregation destroys is the main conclusion."],"table_id":"five-layers"},{"citation_ids":["SRC-TATE-1950"],"equations":["E(xy)=E(x)+E(y)","E(x)=c\\,\\log N(x),\\qquad c>0"],"evidence_kind":"exact","heading":"Stage 1: why logarithms are not a choice","items":[],"label":"analysis","paragraphs":["Suppose a commutative monoid has a multiplicative size N(x)>0 and an extent E(x) that respects multiplication and the order induced by N. Under mild nontriviality assumptions, comparison of powers proves that logarithmic geometry is forced. The statement assumes neither continuity nor unique factorization.","Finite comparison data cannot force exact logarithms; they constrain a cone of possible calibrations. Understanding how quickly that cone contracts became the first computational problem of the program."]},{"citation_ids":["SRC-TATE-1950"],"equations":["\\log |q|_{\\infty}+\\sum_p\\log |q|_p=0"],"evidence_kind":"exact","heading":"Stage 2: primes become geometric directions","items":[],"label":"analysis","paragraphs":["For a positive rational q, real size and p-adic valuations obey the product formula. A rational is represented by a finitely supported vector of local logarithmic coordinates lying in a conservation hyperplane; multiplication becomes vector addition and prime factorization becomes a lattice.","After eliminating the real coordinate, distinct prime directions meet at 60 degrees in the selected Euclidean normalization. That angle is not a mystical property of primes and is not canonical under arbitrary metrics: it is a consequence of the product-formula hyperplane and the chosen metric."]},{"citation_ids":["SRC-TATE-1950"],"equations":["\\mathbb{A}_{\\mathbb{Q}}=\\mathbb{R}\\times\\prod_p'\\mathbb{Q}_p"],"evidence_kind":"exact","heading":"Stage 3: the actual adelic space","items":[],"label":"analysis","paragraphs":["The valuation skeleton is not yet the adele ring. The rational adeles assemble the real line and all p-adic fields into a restricted product. The diagonal copy of Q is discrete and cocompact and, with the standard character and self-dual Haar measure, is its own annihilator. Poisson summation on this lattice produces the adelic theta transformation.","In the controlled family tested here, the dual rational lattice and the 1/M normalization arise from the finite p-adic places rather than being inserted by hand. This is classical adelic harmonic analysis, not a new proof of the theta transformation."]},{"citation_ids":["SRC-TATE-1950"],"evidence_kind":"exact","heading":"Stage 4: number fields add units and class groups","items":[],"label":"analysis","paragraphs":["Passing from Q to a number field multiplies the Archimedean places and introduces nonprincipal prime ideals. Dirichlet's unit theorem places unit logarithms in a lattice inside an Archimedean conservation hyperplane; its covolume is the regulator. A finite place can be synchronized through a power of its ideal class that becomes principal.","The mechanism is local scale, then unit-lattice synchronization, then finite ideal-class synchronization. Exact arithmetic laboratories in Q(i) and Q(√5) checked discriminants, splitting, ramification, principal-ideal factorization, unit lattices, regulators and self-dual volume normalizations."]},{"citation_ids":["SRC-TROPP-2004","SRC-TROPP-GILBERT-2007"],"equations":["\\dim\\ker(M)=1"],"evidence_kind":"numerical","heading":"Stage 5: how much finite arithmetic is enough?","items":[],"label":"analysis","paragraphs":["Infinite axioms are mathematically clean but experimentally unavailable. Bounded collections of adjacent integer comparisons or small principal elements turn calibration into linear algebra: local calibration is unique up to overall scale exactly when the finite principal-relation matrix has nullity one.","Over Q, exact rational dual certificates converted finite order comparisons into rigorous bounds on log 3/log 2. At cutoff 100,000, the full network of nearby-integer factorizations produced an interval more than 3,100 times narrower than the two-prime-only interval. We call this arithmetic acceleration: the finite data change how efficiently rigidity becomes visible, not the limiting theorem itself."]},{"citation_ids":["SRC-TATE-1950","SRC-CLA-2020"],"equations":["\\sum_p k_p\\log p=\\log\\left(\\prod_p p^{k_p}\\right)","\\sum_{n\\ge1}e^{-s\\log n}=\\zeta(s)=\\prod_p(1-p^{-s})^{-1},\\qquad\\Re(s)>1"],"evidence_kind":"exact","heading":"Stage 6: arithmetic as a spectral system","items":[],"label":"analysis","paragraphs":["Give prime p a bosonic mode of energy log p. A basis state with occupations (k_p) has energy log of the corresponding positive integer, so unique factorization identifies occupation states with integers. The heat trace is therefore the zeta Euler product.","The spectral dictionary is exact, but its Hamiltonian was engineered from log p. It encodes the Euler product by construction; no self-adjoint Hilbert–Pólya operator emerged and nothing here locates nontrivial zeros. Inverse experiments with log-prime Fourier dictionaries exhibited the expected coherence transition as the time window grew."]},{"citation_ids":["SRC-TROPP-2004","SRC-TROPP-GILBERT-2007"],"evidence_kind":"numerical","heading":"Stage 7: turn the dictionary around","items":[],"label":"analysis","paragraphs":["Incomplete responses can still reveal bounded arithmetic structure. In a quadratic field, two labelled local coefficients identify split, ramified and inert behavior. Noisy local heat traces recovered these templates in the controlled laboratories. An Archimedean unit channel yielded an unbiased regulator estimator with an exact variance formula; without that channel the regulator direction is exactly nonidentifiable.","Interacting prime modes supplied countermodels to naive inverse reasoning: a single aggregate partition value can be mimicked by a free model with a fitted effective energy. One scalar trace value is not evidence that the underlying geometry is independent or Euler-factorized."],"table_id":"quadratic-local-coefficients"},{"citation_ids":["SRC-TROPP-2004","SRC-TROPP-GILBERT-2007"],"equations":["F(s)=\\sum_{n\\ge1}a_n n^{-s}","a_n=\\lim_{s\\to\\infty}n^s\\left(F(s)-\\sum_{m<n}a_m m^{-s}\\right)"],"evidence_kind":"numerical","heading":"Stage 8: what global traces reveal—and conceal","items":[],"label":"analysis","paragraphs":["If F(s)=sum a_n n^(−s) is absolutely convergent in a right half-plane, its values for all sufficiently large real s determine every coefficient recursively. At a known finite cutoff, samples form a Vandermonde system and are algebraically identifiable, but not automatically numerically stable.","A complex-time experiment recovered the first 50 ideal-count coefficients of Q(√−5) from 200 random samples. With noise standard deviation 10^(−6), all 40 trials succeeded for windows 300 and 1,000 while all 40 failed for windows 20 and 100. The mean condition number fell from approximately 3.7×10^15 at window 20 to 2.90 at window 1,000. An unmodelled tail caused failure even at zero measurement noise. Stable global inversion needs a genuine cutoff, an analytic tail bound or a joint tail model."]},{"callout":true,"citation_ids":[],"equations":["\\mathbb{Z}^3/L_{\\mathrm{principal}}\\cong\\mathbb{Z}/2\\mathbb{Z}"],"evidence_kind":"exact","heading":"The class group appears as missing integral information","items":[],"label":"analysis","paragraphs":["The field K=Q(√−5) has class number two. A Minkowski-bound argument gives a representative of norm 1 or 2 in every ideal class. The unique prime ideal P₂ above 2 is nonprincipal because a²+5b²=2 has no integer solution, while P₂²=(2); therefore [P₂] has order two and generates the class group.","Principal divisors of 2, 3, 1+√−5 and 1−√−5 give the four-by-three generator table below. Its Smith invariants are (1,1,2), so the principal-divisor lattice has index two in Z³. Over R, multiplication by two is invertible and the obstruction vanishes; over Z, the torsion remains. The class group is an integral defect invisible to real geometry.","The aggregate Dedekind coefficient a_K(n) counts all ideals of norm n but does not mark which ideals are principal. Recovering every coefficient of a zeta trace therefore need not reconstruct the principal-divisor lattice. Global ideal counts and principal-relation data contain different information."],"table_id":"class-group-matrix"},{"citation_ids":["SRC-TROPP-2004","SRC-TROPP-GILBERT-2007"],"equations":["\\sum_i\\mathbb{Z}z_i=\\gcd(k_1,\\ldots,k_m)R\\mathbb{Z}"],"evidence_kind":"exact","heading":"Two further exact reconstruction results","items":[],"label":"analysis","paragraphs":["If unlabelled rank-one unit observations have the form z_i=k_iR, their generated additive subgroup is gcd(k_1,…,k_m)RZ. The fundamental regulator is identifiable exactly when the sampled multipliers are primitive; otherwise only a sublattice spacing is visible.","For finite rigidity, linear programming found a lower certificate of multiplier mass 3/7, improving an earlier 5/7 mass by 40 percent, and an upper certificate of mass 2/5. Exact rational dual witnesses prove these masses globally minimal within the cutoff-100 nonnegative certificate cone. An integer Farkas witness proves that 80<81 is unavoidable for the sharp lower endpoint."]},{"citation_ids":[],"evidence_kind":"interpretive","heading":"What this contributes","items":[],"label":"analysis","paragraphs":["Most ingredients—valuations, adeles, Poisson summation, unit lattices, class groups, Euler products and sparse recovery—are established mathematics. No claim of literature priority is made for those foundations.","The contribution is a disciplined, reproducible chain connecting an exact ordered-monoid theorem, local–global calibration, executable rational and ideal-arithmetic laboratories, finite rigidity certificates, a bounded spectral dictionary, inverse experiments with stated limits and a concrete demonstration that real rank can miss integral class-group torsion."]},{"citation_ids":["SRC-CLA-2020"],"evidence_kind":"interpretive","heading":"What this does not establish","items":["The Riemann hypothesis or any new restriction on zeta zeros.","A Hilbert–Pólya operator.","A physical law or a geometry of spacetime.","Stable recovery of an infinite Dedekind series from finitely many noisy samples without tail assumptions.","A class group from aggregate zeta values alone.","Literature novelty for classical adelic constructions."],"label":"analysis","paragraphs":["The boundary is part of the result. The present program offers useful language and inverse-problem infrastructure around Euler products and local–global data, but it is ecosystem work, not a resolution."]},{"citation_ids":[],"evidence_kind":"interpretive","heading":"The next question: arithmetic sensing","items":[],"label":"testable_hypothesis","paragraphs":["The next research direction combines number theory with information science: how many noisy global measurements are required to reconstruct a specified amount of arithmetic structure, and which structure is irretrievably lost by aggregation? We call this program arithmetic sensing.","Its first targets are nonasymptotic conditioning bounds, recovery guarantees with an analytic Dedekind-tail budget, information-theoretic lower bounds, indistinguishable number-field traces, higher-rank unit-lattice recovery and class-group laboratories beyond Z/2Z. Both success and failure can become theorems: a recovery algorithm needs a guarantee, while an impossibility claim needs indistinguishable models or an information bound."]}],"slug":"geometry-arithmetic-remembers","source_manifest":[{"filename":"ABSTRACT_ORDERED_MONOID_RIGIDITY.md","kind":"technical manuscript","sha256":"602820db5073ee9d1d81b004898caf39637c0f187c08f37301078eb1f0064171"},{"filename":"LOCAL_GLOBAL_ADELIC_CALIBRATION.md","kind":"technical manuscript","sha256":"be94b0a00d7cc6d22b8f6a4b8ffad8b4a4842763f3987a90ce62f167b80c4840"},{"filename":"RATIONAL_ADELES_AND_POISSON.md","kind":"technical manuscript","sha256":"fabf1fc72a1a65b12dee7ea63c5d1703f53eb330cc5a88debacfac793020b293"},{"filename":"NUMBER_FIELD_ADELIC_CALIBRATION.md","kind":"technical manuscript","sha256":"0275df474029fd8f960c0ba8b63d44ebec32ec341f0dc872b07eb4250129e6d8"},{"filename":"FINITE_LOCAL_GLOBAL_RECONSTRUCTION.md","kind":"technical manuscript","sha256":"77d4fa87b0b728c365eda342d746747a925c15c86e0e255bc6ec11999f285ac5"},{"filename":"SPECTRAL_AND_PHYSICAL_INTERPRETATIONS.md","kind":"technical manuscript","sha256":"25a06d567dac52a798d90bbc857ea9f9c9abb65b711429c3a14891a3f4c0312c"},{"filename":"INVERSE_ADELIC_SPECTRAL_GEOMETRY.md","kind":"technical manuscript","sha256":"0a256fd4a83d41c4d64e810cccb97b562c4c4deb6f0a9487000808ab7d86b93a"},{"filename":"GLOBAL_INVERSE_RECONSTRUCTION.md","kind":"technical manuscript","sha256":"d662525a538147270891fa6ee64a05b10f0d583d2ff11010d3702c402a686ea9"},{"filename":"RESEARCH_ROADMAP.md","kind":"research roadmap","sha256":"33740fc24218c1aa3f880a0484f3e5ca89ce742f3795609b390696957494f237"},{"filename":"class_group_obstruction.py","kind":"exact arithmetic laboratory","sha256":"49ecd67a0295e608a58efce76d84353dc58f863034ff591293094927fbfa81af"},{"filename":"global_trace_inversion.py","kind":"numerical recovery laboratory","sha256":"00bf1f51ffabb09d7572ae503195b19384da613a14450dc1e1feb0366d00e2c4"},{"filename":"redundant_rigidity_backbones.py","kind":"exact certificate laboratory","sha256":"0ea61287e72a1ea6a308013de5d0d1774389e3fc1a6a20412fea58021636b942"},{"filename":"unknown_unit_lattice.py","kind":"unit-lattice laboratory","sha256":"e5a91bf4ff13b35bbfd0d54e432d4b682f78922b7f50672068074e34b944d095"}],"sources":[{"publisher":"American Mathematical Society","retrieved_at":"2026-08-12T00:00:00Z","source_class":"research_paper","source_id":"SRC-TATE-1950","title":"John Tate, Fourier Analysis in Number Fields and Hecke's Zeta-Functions","url":"https://bookstore.ams.org/cworks-24-1/"},{"publisher":"California Institute of Technology","retrieved_at":"2026-08-12T00:00:00Z","source_class":"research_paper","source_id":"SRC-TROPP-2004","title":"Joel Tropp, Greed is Good: Algorithmic Results for Sparse Approximation","url":"https://authors.library.caltech.edu/records/m0swv-ba672"},{"publisher":"California Institute of Technology","retrieved_at":"2026-08-12T00:00:00Z","source_class":"research_paper","source_id":"SRC-TROPP-GILBERT-2007","title":"Joel Tropp and Anna Gilbert, Signal Recovery From Random Measurements Via Orthogonal Matching Pursuit","url":"https://tropp.caltech.edu/reports/TG07-Signal-Recovery-TR.pdf"},{"publisher":"Clay Mathematics Institute","retrieved_at":"2026-08-12T00:00:00Z","source_class":"research_paper","source_id":"SRC-CLA-2020","title":"The Riemann Hypothesis","url":"https://www.claymath.org/riemann/"}],"status":"active","subtitle":"An eight-stage investigation from a speculative idea to exact adelic reconstruction theorems","tables":[{"caption":"Five interlocking layers of the investigation","headers":["Layer","Geometric object","What forces or reveals it"],"id":"five-layers","note":"These layers cannot be collapsed into one universal picture without losing information.","rows":[["Multiplicative order","A logarithmic line","Additivity plus strict compatibility with multiplicative size"],["Local factorization","A valuation lattice","Prime factorization and the product formula"],["Adelic harmonic analysis","A self-dual locally compact space","All real and p-adic completions assembled as a restricted product"],["Units and ideal classes","Archimedean lattices and finite integral torsion","Dirichlet units and principal-divisor relations"],["Spectral response","Logarithmic Fourier modes","Prime energies \\(\\log p\\), Euler products, and complex-time traces"]]},{"caption":"Exact local coefficient templates used in the quadratic inverse experiment","headers":["Prime behavior","a(p)","a(p²)"],"id":"quadratic-local-coefficients","note":"Noisy local heat traces recovered these three templates in controlled laboratories; this is numerical evidence, not a universal decoder.","rows":[["split","2","3"],["ramified","1","1"],["inert","0","1"]]},{"caption":"Principal-divisor generators for the Q(√−5) class-group obstruction","headers":["P₂","P₃","P₃′"],"id":"class-group-matrix","note":"The Smith invariants are (1, 1, 2), so Z³/Lprincipal ≅ Z/2Z. The four-by-three matrix is a generator table, not a continuous trace.","rows":[["2","0","0"],["0","1","1"],["1","0","1"],["1","1","0"]]}],"tags":["adelic geometry","number theory","arithmetic sensing","AI research","inverse problems","class groups","spectral methods"],"title":"The Geometry Arithmetic Remembers","updated_at":"2026-08-12T00:00:00Z"}