Skip to content

The Information Geometry of Prime Factorization

Transition Problem TCU-001: a uniform saddle-point correlation law

Research state: active mathematical target, not proved.
Version: 0.1, 2026-10-11.
Owner: GCL mathematical research, MATHFORGE → MATHSOLVE; certification reserved to MATHCERT.
Companion: Synthesis manuscript.

Research status and authoritative records

Status: Open mathematical target, not a proved transition theorem.
Live proof obligations: MATHSOLVE #1075.
Protected research source: TCU-001 Markdown.
Programme tracker: PINFO-001.
Mathematical typesetting does not confer mathematical certification.

1. The central question

Let \(2\le y\le X\) and let \(N\) be uniformly distributed on

\[ S(X,y)=\{n\le X:P^+(n)\le y\},\qquad \Psi(X,y)=|S(X,y)|. \]

Write \(V_p=v_p(N)\) and \(k=\pi(y)\). The total correlation is

\[ T(X,y)=\sum_{p\le y}H(V_p)-\log\Psi(X,y). \]

It measures the dependence imposed on the prime exponents by the single global inequality \(\sum V_p\log p\le\log X\).

Put \(t=\log X\), \(u=t/\log y\), and define the finite Euler product

\[ Z_y(s)=\prod_{p\le y}(1-p^{-s})^{-1},\quad M_y(s)=\sum_{p\le y}\frac{\log p}{p^s-1},\quad \sigma_y^2(s)=\sum_{p\le y}\frac{(\log p)^2p^s}{(p^s-1)^2}. \]

There is a unique positive counting saddle point \(\alpha=\alpha(X,y)\) satisfying \(M_y(\alpha)=t\).

Target TCU-001 (conjectural, not a proved theorem). Establish an explicit and genuinely uniform error function \(\varepsilon(X,y)\to0\) as both \(y\to\infty\) and \(u=\log X/\log y\to\infty\), with

\[ \boxed{T(X,y)=\log\!\bigl(\alpha\sqrt{2\pi\sigma_y^2(\alpha)}\bigr)-1+\varepsilon(X,y).} \]

If this is false without additional growth conditions, isolate a counterregime and identify its correction term. The target is a two-sided result, not a one-sided bound, and it must be compatible with fixed-prime, sparse-prime, Dickman and saddle-point descriptions wherever those apply.

2. Exact master decomposition (proved algebraically)

Define the product reference measure \(Q_s\) on all positive integers whose prime factors do not exceed \(y\):

\[ Q_s(V_p=j)=(1-p^{-s})p^{-sj}. \]

For the actual marginal \(P_p=\mathcal L(V_p)\), define

\[ \Delta=t-\mathbb E_P[\log N]\ge0,\qquad K_s=\sum_{p\le y}D(P_p\Vert Q_{p,s})\ge0. \]

Then for every \(s>0\),

\[ \begin{split} T(X,y)&=D(P\Vert Q_s)-K_s,\\ D(P\Vert Q_s)&=\log Z_y(s)+s(t-\Delta)-\log\Psi(X,y). \end{split} \]

This identity is finite and exact. It is a KL chain rule, not an approximation.

Hildebrand--Tenenbaum's saddle-point counting theorem [HT93, Theorem 1.6, equation (1.22)] gives uniformly for \(X\ge y\ge2\),

\[ \Psi(X,y)=\frac{X^\alpha Z_y(\alpha)} {\alpha\sqrt{2\pi\sigma_y^2(\alpha)}} \left(1+O(1/u+\log y/y)\right). \]

Consequently in ranges where \(u,y\to\infty\), taking logarithms yields

\[ \boxed{ T(X,y)=\log\!\bigl(\alpha\sqrt{2\pi\sigma_y^2(\alpha)}\bigr) -\alpha\Delta-K_\alpha+O(1/u+\log y/y). }\tag{M} \]

This is the master transition formula. It reduces TCU-001 to two separately testable obligations:

\[ \boxed{\alpha\Delta=1+o(1),\qquad K_\alpha=o(1)} \]

uniformly as \(u,y\to\infty\). The two obligations are sufficient, not asserted necessary separately from the targeted combination. Never infer their truth merely from the counting theorem.

3. An exact local influence representation (proved algebraically)

Let \(W\sim\mathop{\mathrm{Exp}}(s)\) independent of \(Q_s\) and

\[ E=\sum_{p\le y}V_p\log p+W. \]

Its density is

\[ f_E(t)=\frac{s e^{-st}}{Z_y(s)}\Psi(e^t,y). \]

The conditional law \(\mathcal L((V_p)_p\mid E=t)\) is exactly the uniform distribution on integers not exceeding \(e^t\) whose prime factors are at most \(y\). For each prime \(p\), let \(f_{-p}\) denote the density of

\[ E_{-p}=\sum_{\substack{q\le y\\q\ne p}}V_q\log q+W. \]

Then

\[ \boxed{ P_p(j)=Q_{p,s}(j)\,\frac{f_{-p}(t-j\log p)}{f_E(t)} \quad(j\ge0). }\tag{L} \]

The ratio is interpreted as \(0\) for negative arguments. This follows directly from the density factorization of \((V_p,E)\). It is a useful route to the local limit theorem because removing one prime is a controlled perturbation of the full energy.

Set

\[ R_{p,j}(t)=f_{-p}(t-j\log p)/f_E(t). \]

Then

\[ K_s=\sum_{p\le y}\sum_{j\ge0}Q_{p,s}(j)\,R_{p,j}\log R_{p,j} \]

exactly, and the elementary chi-squared inequality gives the finite bound

\[ \boxed{ 0\le K_s\le \sum_{p\le y}\sum_{j\ge0}Q_{p,s}(j)(R_{p,j}-1)^2. }\tag{I} \]

This upper bound is nonasymptotic. A summable local-energy influence estimate making the right-hand side \(o(1)\) would close the second obligation in TCU-001. It would also avoid the uncontrolled step of summing pointwise error terms over a growing set of primes.

4. Boundary tests that any theorem must pass

Fixed \(y\), \(X\to\infty\). Let \(k=\pi(y)\). The continuum simplex calculation gives

\[ T(X,y)\longrightarrow C_k=\log(k!)-k\log k+k-1. \]

At the counting saddle \(\alpha\), \(\alpha\Delta\to k/(k+1)\) and \(K_\alpha\to1/(k+1)\). In particular neither term can be discarded when \(k\) stays fixed, even if \(u\to\infty\). This is why the target explicitly requires \(y\to\infty\).

Sparse-prime joint limit. The internal working Note 007 derives a finite-dimensional simplex inequality. If \(k\to\infty\) and

\[ k^3\sum_{p\le y}\log p=o(\log X), \]

it gives

\[ T(X,y)=\tfrac12\log(2\pi k)-1+o(1),\quad \alpha\Delta\to1,\quad K_\alpha\to0. \]

Status: author-derived / internal proof; external certification and novelty not established. Its original manuscript and replay identifiers are recorded in source provenance record.

Fixed \(u>1\), \(y\to\infty\). The internally derived Dickman expansion has the form

\[ T(X,y)=\log\log y+\gamma-1-\log\rho(u)-J(u)+o(1), \]

where

\[ J(u)=\int_0^1 \frac{a_u(v)\log a_u(v)-a_u(v)+1}{v}\,dv, \qquad a_u(v)=\rho(u-v)/\rho(u). \]

The fixed \(u\) law and the growing \(u\) law are not interchangeable without uniform remainder estimates. Earlier internal Notes 003--005 propose such refinements; their full proof details must be re-audited before promotion.

Unrestricted cutoff. At \(y=X\), the known total-correlation asymptotic is \(\log\log X+\gamma-1+o(1)\). But here \(u=1\) and the TCU-001 hypothesis \(u\to\infty\) fails: this is a compatibility example, not a test within the conjectured domain.

5. Work packages

  • TCU-WP1, slack local law: Prove \(\alpha\Delta=1+o(1)\) uniformly for \(y,u\to\infty\), preferably with an explicit bound in terms of \(\alpha\sigma_y(\alpha)\) and a verified local limit error for the exponential-smoothed energy.
  • TCU-WP2, single-prime leave-one-out local law: Compare \(f_{-p}(t-j\log p)\) to \(f_E(t)\) in the weighted \(L^2\) sense under \(Q_{p,\alpha}\) of (I), uniformly over all \(p\le y\) and \(j\ge0\). Track prime-power tails separately.
  • TCU-WP3, matching transition: Combine (M), WP1 and WP2; obtain a uniform explicit remainder and reconcile it with both the sparse simplex and Dickman regimes. If the target fails, construct a rigorously sourced counterregime.
  • TCU-WP4, reproducible falsification: Independently enumerate accessible \((X,y)\), compute exact \(T,\Delta,K_\alpha,\sigma\), and compare predictions; include \(k=1\), fixed small \(k\), \(u\) close to \(1\), moderate \(u\), and growing \(k\) controls. Numerical agreement is not proof.
  • TCU-WP5, priority and source audit: Compare against HT93, Tenenbaum 2023, literature on probabilistic smooth numbers, conditional local limit theorems, and entropy of logarithmic combinatorial structures. Record theorem-level novelty only after review.

6. Research hygiene

  • A theorem proved under imported analytic estimates must display the exact estimates and their valid ranges.
  • Author self-review, a separate script, and GitHub CI are not independent mathematical certification.
  • Do not infer factoring speedups or an RH result from this information-theoretic observable.
  • A failed TCU-001 conjecture is a legitimate result if the obstruction is rigorous.
  • Repository admission documents the investigation, not its mathematical certification.

Primary references

[HT93] A. Hildebrand and G. Tenenbaum, Integers without large prime factors, Journal de Théorie des Nombres de Bordeaux 5 (1993), 411–484, https://doi.org/10.5802/jtnb.101 . Theorem 1.6 and equation (1.22) provide the counting saddle formula.

[T23] G. Tenenbaum, Sur le biais d'une loi de probabilité relative aux entiers friables, JTNB 35 (2023), 481–493, https://doi.org/10.5802/jtnb.1253 . Relevant caution concerning the biased canonical smooth-number measure.

[Tao25] T. Tao, Smooth numbers and max-entropy (2025), https://terrytao.wordpress.com/2025/09/15/smooth-numbers-and-max-entropy/ .

[DG93] P. Donnelly and G. Grimmett, On the asymptotic distribution of large prime factors, JLMS 47 (1993), 395–404, https://doi.org/10.1112/jlms/s2-47.3.395 .