Quantization of Ricci Curvature in Information Geometry

Carlos C. Rodríguez · 2026

Read Quantization of Ricci Curvature in Information Geometry by Carlos C. Rodríguez (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-10054.txt.

Introduction

Carlos C. Rodriguez settles a conjecture he posed twenty years earlier about curvature in information geometry. Studying binary Bayesian networks, he had guessed that the volume averaged Ricci scalar under the Fisher metric always lands on positive half integers. He proves the claim for tree structured and complete graph networks through a Beta function cancellation. Loop counterexamples disprove it in general, while Gaussian networks show opposite, negative curvature.

Catalog record

CREATORCarlos C. Rodríguez
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36494
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.10054

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Quantization of Ricci Curvature in Information Geometry Carlos C. Rodríguez Abstract In 2004, while studying the information geometry of binary Bayesian networks (bitnets), the author conjectured that the volume-averaged Ricci scalar ⟨ R ⟩ \langle R\rangle computed with respect to the Fisher information metric is universally quantized to positive half-integers: ⟨ R ⟩ ∈ 1 2 ​ ℤ + \langle R\rangle\in\tfrac{1}{2}\mathbb{Z}^{+} . This paper resolves the conjecture after 20 years. We prove it for tree-structured and complete-graph bitnets via a universal Beta function cancellation mechanism, and we disprove it in general by exhibiting explicit loop counterexamples. A key correction to the original 2004 paper is established: the equal-volume theorem Vol ⁡ ( L ~ n ) = Vol ⁡ ( E ~ n ) = π 1 + 2 ​ n \Vol(\tilde{L}_{n})=\Vol(\tilde{E}_{n})=\pi^{1+2n} (proved via a Markov chain transition-matrix argument) implies the corrected average curvatures ⟨ R ⟩ ​ ( L ~ n ) = ⟨ R ⟩ ​ ( E ~ n ) = ( 2 ​ n − 1 ) / 2 \langle R\rangle(\tilde{L}_{n})=\langle R\rangle(\tilde{E}_{n})=(2n-1)/2 , replacing the erroneous formula n / 2 n/2 of the earlier work. New SymPy computations extend the collapsing-star ( C ~ n \tilde{C}_{n} ) table to n = 5 n=5 parents, revealing a geometric capacity inversion at n = 4 → 5 n=4\to 5 where the average Ricci scalar changes sign. Double-collider loop bitnets yield ⟨ R ⟩ = 36 / 5 ∉ 1 2 ​ ℤ \langle R\rangle=36/5\notin\tfrac{1}{2}\mathbb{Z} , confirming that loops destroy quantization. We extend the program to Gaussian DAG networks (zero-mean, variance-kernel parameterization), where we find a strict sign dichotomy: tree-structured discrete networks have positive curvature, while Gaussian networks form solvable Lie groups with negative constant curvature, yielding the universal formula R Gauss = − ( d + 5 ) ( d − 1 ) / 8 R_{\mathrm{Gauss}}=-(d+5)(d-1)/8 for simple trees, where d d is the parameter dimension. 1 Introduction 1.1 The 20-year conjecture and the 2004 paper In 2004, the author showed [ 1 ] that the information-geometric parameter manifolds of binary Bayesian networks (“bitnets”) are Riemannian manifolds of finite volume under the Fisher metric, and computed exact volumes for several canonical topologies: the complete DAG \tilde{K}_{n} , the directed line \tilde{L}_{n} , the exploding star \tilde{E}_{n+1} (Naïve Bayes), and the collapsing star \tilde{C}_{n+1} . Computer experiments using vTool suggested that the volume-averaged Ricci scalar is always a positive half-integer : Conjecture 1.1 (Rodríguez, 2004) . For any bitnet with Fisher information metric, \langle R\rangle\in\tfrac{1}{2}\mathbb{Z}^{+} . 1.2 Corrections to the 2004 results Before stating our new results we record a correction. The 2004 paper reported [ 1 ] that the volume-averaged Ricci scalar of the exploding star satisfies \langle R\rangle(\tilde{E}_{n+1})=n/2 . This formula is incorrect . Theorem 1.2 (Equal volumes and equal curvatures) . The directed line \tilde{L}_{n} and the exploding star \tilde{E}_{n} (one parent, n-1 children) have identical Fisher information volumes and identical volume-averaged Ricci scalars: \Vol(\tilde{L}_{n})\;=\;\Vol(\tilde{E}_{n})\;=\;\pi^{1+2n}, \langle R\rangle(\tilde{L}_{n})\;=\;\langle R\rangle(\tilde{E}_{n})\;=\;\frac{2n-1}{2}. Proof. Both topologies share exactly 1+2n parameters: one root marginal \rho_{0}\in(0,1) and n conditional pairs (r_{k,0},r_{k,1}) . The Fisher metric is diagonal and block-independent in both cases (each CPT block depends only on \rho_{0} ), so the volume element factorises as a product of 1+2n independent Beta (1/2,1/2) integrals, each equal to \pi : \Vol(\tilde{L}_{n})=\Vol(\tilde{E}_{n})=\pi^{1+2n}. For the curvature, the chain \tilde{L}_{2} has a constant Ricci scalar R=3/2 (verified by direct SymPy computation), so \langle R\rangle(\tilde{L}_{2})=3/2=(2\cdot 2-1)/2 . A Markov-chain induction on the transition-matrix structure establishes the general formula (2n-1)/2 ; the result is confirmed numerically for n=2,3,4 by Monte Carlo volume integration. ∎ Remark 1.3 (Error in 2004) . The 2004 paper computed \langle R\rangle(\tilde{E}_{n+1})=n/2 using an incorrect integration of the pointwise Ricci formula. The correct values are: n=1:\tfrac{1}{2},\quad n=2:\tfrac{3}{2},\quad n=3:\tfrac{5}{2},\quad n=4:\tfrac{7}{2},\quad\ldots These are all positive half-integers, consistent with Conjecture 1.1 , but shifted by 1/2 from the 2004 report. 1.3 Summary of new results (1) Trees and complete DAGs: Conjecture 1.1 holds. The key mechanism is a “Beta cancellation” identity (Section 4 ). (2) Collapsing stars extended: New SymPy computations (Section 3.3 ) extend the \tilde{C}_{n} table to n=5 parents and reveal a curvature phase transition : \langle R\rangle changes sign between n=4 ( \langle R\rangle=16 ) and n=5 ( \langle R\rangle=-272 ). (3) Loop topology: The double-collider bitnet D_{4} (Section 5 ) gives \langle R\rangle=36/5\notin\tfrac{1}{2}\mathbb{Z} , disproving Conjecture 1.1 in general. (4) Gaussian DAG networks: For zero-mean Gaussian DAGs (Section 6 ), the Fisher metric has negative constant curvature R=-(d+5)(d-1)/8 for simple-parent trees, exhibiting a sign dichotomy with the discrete case. (5) Quantum analogy: Section 7 establishes the Bures–Fisher correspondence ( g_{\text{Bures}}=\tfrac{1}{4}g_{\text{Fisher}} ) and the spin-geometry interpretation of the half-integer spectrum. 2 Preliminaries: Bitnets and Their Geometry 2.1 Bitnets A bitnet is a directed acyclic graph (DAG) G=(V,E) with |V|=n binary nodes X_{1},\ldots,X_{n} . Each node X_{k} has m_{k}=|{\rm pa}(k)| parents and 2^{m_{k}} associated parameters \{\theta_{k,s}\}_{s=0}^{2^{m_{k}}-1}\subset(0,1) , where \theta_{k,s}=P(X_{k}=1\mid X_{{\rm pa}(k)}=s) . The total dimension is: d=\sum_{k=1}^{n}2^{m_{k}}. The Fisher information matrix is diagonal for any bitnet [ 1 ] : g_{jj}(\theta)=\pi_{j}(\theta)/[\theta_{j}(1-\theta_{j})] , where \pi_{j}=P(X_{{\rm pa}(v(j))}=j-t(v(j))) is the marginal probability of the parent configuration selecting parameter \theta_{j} . 2.2 Volume element and Ricci scalar The Fisher information volume is: \Vol(\text{bitnet})=\int_{(0,1)^{d}}\prod_{k=1}^{n}W_{k}^{1/2}\,d\theta,\qquad W_{k}=\frac{\prod_{s}P(X_{{\rm pa}(k)}=s)}{\prod_{s}\theta_{k,s}(1-\theta_{k,s})}. (1) For any bitnet \Vol\leq\pi^{d} , with equality iff all nodes are independent. The Ricci scalar R(\theta) is computed from the diagonal metric via the standard Riemann tensor formula. Because the metric is diagonal, the Christoffel symbols and Riemann components involve only diagonal derivatives of the g_{jj} . 3 Volumes and Curvatures of Classical Topologies 3.1 Complete DAGs \tilde{K}_{n} The complete bitnet \tilde{K}_{n} has all available edges; its parameter space is the multinomial model for 2^{n} outcomes (all binary sequences of length n ), with d=2^{n}-1 . By Amari’s sphere isometry [ 5 ] , \tilde{K}_{n} is isometric to the positive orthant of S^{d} of radius 2: Theorem 3.1 (Complete DAGs) . \displaystyle\Vol(\tilde{K}_{n}) \displaystyle=\frac{\pi^{2^{n-1}}}{(2^{n-1}-1)!}, (2) \displaystyle R \displaystyle=\frac{d(d-1)}{4}=\frac{(2^{n}-1)(2^{n}-2)}{4}\quad(\text{constant}). (3) The volumes decrease rapidly: \Vol(\tilde{K}_{1},\ldots,\tilde{K}_{5})=\pi,\pi^{2},\pi^{4}/6,\pi^{8}/5040,\ldots 3.2 Directed lines \tilde{L}_{n} and exploding stars \tilde{E}_{n} Theorem 3.2 (Equal volumes and curvatures) . \Vol(\tilde{L}_{n})=\Vol(\tilde{E}_{n})=\pi^{1+2n},\qquad\langle R\rangle(\tilde{L}_{n})=\langle R\rangle(\tilde{E}_{n})=\frac{2n-1}{2}. For the chain \tilde{L}_{2} (equivalently \tilde{K}_{2} ), SymPy computation confirms R=3/2 everywhere ; no averaging is needed. For longer chains, R is non-constant but the volume-weighted average is (2n-1)/2 . Selected values: \langle R\rangle=\tfrac{1}{2},\tfrac{3}{2},\tfrac{5}{2},\tfrac{7}{2},\tfrac{9}{2},\ldots for n=1,2,3,4,5 . All are positive half-integers. Remark 3.3 (Interpretation) . The exploding star \tilde{E}_{n} (Naïve Bayes) and the directed line \tilde{L}_{n} are geometrically indistinguishable by both volume and average curvature, despite their different graphical structure. This adds to the earlier observation that these topologies have the same MDL complexity [ 1 ] . 3.3 Collapsing stars \tilde{C}_{n+1} (new computations) The collapsing star \tilde{C}_{n+1} has n independent root nodes converging to a single child, with d=n+2^{n} parameters. From the 2004 paper and new SymPy calculations: Theorem 3.4 (Ricci scalar of \tilde{C}_{n+1} ) . The pointwise Ricci scalar has the form: R(\rho_{1},\ldots,\rho_{n})=a_{n}-b_{n}\sum_{i=1}^{n}\frac{1}{\rho_{i}(1-\rho_{i})}, (4) where (\rho_{1},\ldots,\rho_{n}) are the parent node marginals. The coefficient b_{n} satisfies the exact formula: b_{n}=\frac{2^{n-1}(2^{n-1}-1)}{4}. (5) The volume-averaged Ricci scalar is: \langle R\rangle=a_{n}-b_{n}\cdot n\cdot\frac{2^{n+1}}{2^{n-1}-1}. (6) Table 1: Collapsing star \tilde{C}_{n+1} : volumes, coefficients and averaged Ricci scalars. Values for n=1,2,3 were obtained with vTool [ 1 ] ; n=4,5 are new SymPy computations. n parents \tilde{C}_{n+1} d

Read the complete text