Skip to content

32. The single rank-parity node: twins and Riemann (an epilogue)

← 31. Riemann via Liouville · Table of contents · 33. First cause and the main theorem →

Lean: Engine/RiemannLiouville.lean (liouville_eq_neg_one_pow_rank, liouville_flip_of_mul_prime, riemann_of_liouville_bound), Engine/MkNode.lean, Engine/CarrierBridge.lean, Engine/ProductCore.lean. Numbers: tools/RESULTS_final_gap.md, prose/51_NumericalEvidence.md. This is the final chapter of the prose.

In 27. ProductCore we carried the twin reduction all the way to a finite product-rank descent, and in 30. RiemannBranch we built an independent branch towards the Riemann hypothesis through the same engine — the forbidden infinite descending chain (see the glossary); the audit of both left one irreducible node each.

Here we take the last step of the entire prose — not closing the nodes, but showing that they are one and the same node, written in the common language of rank parity. The chapter's goal is honest and limited: to exhibit a mechanism shared by both hypotheses, to separate within it the proven part from the hypothetical one, and to leave a precise plan — without passing the unity off as a proof.

Where we are: two nodes, one shape

Let us gather what we have arrived at, one line each. The twins (after 24, 30, 31) have been reduced to the claim that an infinite carrier of pure centres yields an engine — provided the descent flow generates infinitely many factorable starts of fixed rank (GlobalOldAbsorption, the sole remainder of RESULTS_final_gap.md). Riemann (after 28 and its arithmetic reformulation) has been reduced to a bound on the Liouville summatory function

\[L(x) \;=\; \sum_{n\le x} \lambda(n), \qquad \lambda(n) = (-1)^{\Omega(n)},\]

where LiouvilleBound demands \(|L(x)| = O\!\bigl(x^{1/2+\varepsilon}\bigr)\). At first sight these are two unrelated difficulties — the combinatorial fan-in wall and the analytic \(\sqrt{x}\) wall. The thesis of this chapter: both are control of the rank-parity balance at a growing scale, and our rank apparatus (RankNode, cardFactors, deleteFactor) is their common language.

The apparatus: rank as the number of prime factors

Let us introduce exactly the objects on which the generalisation rests. A descent state at rank \(r\) is

Definition 32.1 (RankNode). RankNode r is a structure with a field sign : Sign (the carrier of two \(\sigma\in\{+1,-1\}\), the side \(6m+\sigma\)) and a field factors : Fin r → ℕ (an ordered tuple of \(r\) prime factors of the active side, all \(> A\) on the legal layer). The type is extensional in the pair (sign, factors) (RankNode.ext in ProductCore.lean): two nodes are equal if and only if their sign and all their factors coincide.

The rank of a node is exactly the number of prime factors with multiplicity, that is, the arithmetic function \(\Omega\). In mathlib it is called cardFactors, and this identity is not an analogy but an equality of functions:

\[\mathrm{rank}(6m+\sigma) \;=\; \Omega(6m+\sigma) \;=\; \texttt{cardFactors}\,(6m+\sigma).\]

From 28. MkNode we know that on the legal layer the rank is bounded: factor_rank_le_four gives \(r\le 4\) (factors \(>A\), product \(\le 6X_A+1 < A^5\)), and composite_rank_ge_two — that an impure composite side has \(r\ge 2\). So on the carrier the rank lives in the finite range \(1\le r\le 4\) — which is exactly FactorizationData.hr from 29. CarrierBridge.

The second object is rank lowering. Removing one factor is

Definition 32.2 (deleteFactor). deleteFactor : RankNode (r+1) → Fin (r+1) → RankNode r, \(X \mapsto \langle X.\mathrm{sign},\ i\mapsto X.\mathrm{factors}(k.\mathrm{succAbove}\,i)\rangle\) — it discards the \(k\)-th factor, preserving the sign and the order of the rest. This is precisely the product-rank descent step \(r\to r-1\) from 27.

The key observation that closes the apparatus onto Liouville: removing a factor flips the sign of the rank. If \(n = p\cdot m\) with \(p\) prime, then \(\Omega(n) = \Omega(m)+1\), hence

\[\lambda(n) \;=\; (-1)^{\Omega(n)} \;=\; -(-1)^{\Omega(m)} \;=\; -\lambda(m).\]

In Lean these are two theorems.

Theorem 32.3 (liouville_eq_neg_one_pow_rank). For every \(n\ne 0\)

\[ \lambda(n) \;=\; (-1)^{\texttt{cardFactors}\,n}. \tag{32.1} \]

Proven with no inputs of ours, as a consequence of liouville_apply from mathlib. 🟢

Theorem 32.4 (liouville_flip_of_mul_prime). For a prime \(p\) and every \(m\)

\[ \lambda(p\cdot m) \;=\; -\,\lambda(m). \tag{32.2} \]

Proven via cardFactors_mul + cardFactors_apply_prime. 🟢

Note. The meaning of these two lines is larger than their proof. They say that the Liouville sign is the rank parity of our RankNode, and deleteFactor is the operator that flips the Liouville sign. That is, the product-rank descent we built for the twins as a purely structural height dynamics turns out to be exactly the dynamics of the sign of \(\lambda\). Two apparatuses, invented for different problems, are one and the same operator on one and the same carrier.

The twin wall as a balance of side ranks

Let us return to the twins and write their wall in rank terms. The block statistics of chapter 51 classify the centres by the pair \((\mathrm{rank}_{-},\mathrm{rank}_{+})\) of the ranks of the two sides \(6m-1,\ 6m+1\), coarsened to \(\{0,3\}\) (prime side / "rich" composite). The cells:

\[N_{00}\ (\text{both prime} = \text{twin}),\quad N_{33}\ (\text{both richly composite}),\quad N_{03},\ N_{30}\ (\text{mixed}).\]

The programme's target inequality is B₅ = N₀₀ − N₃₃ > 0 (twins no fewer than "double composites"), and the four-corner form

\[R_{\mathrm{fc}} \;=\; \frac{N_{00}\,N_{33}}{N_{03}\,N_{30}} \;<\; 1,\]

that is, a negative correlation between the ranks of the sides: primality of one side and primality of the other "attract" each other. Numerically (table [A]) \(R_{\mathrm{fc}}\in[0.86,0.98]\) and it grows towards \(1\) with the block — this is exactly the wall of the margin going to zero: the reserve \(1 - R_{\mathrm{fc}}\to 0^+\) as the scale grows. The twin wall is a statement about the balance of the distribution of the two sides' ranks: that the cell \(N_{00}\) does not empty out relative to \(N_{33}\) as \(N\) grows.

Note. It is precisely here that [12–16] ran into the "parity wall": the model four-corner \(20\binom{n}{6} < \binom{n}{3}^2\) is proven (ModelFourCorner), but the transfer model\(\to\)reality requires control of the joint distribution of the ranks \((\mathrm{rank}_-,\mathrm{rank}_+)\), not of a single marginal. This is a distributional, not a pointwise fact — exactly the same type of statement as the bound on \(L(x)\).

The Riemann wall as a balance of Liouville signs

Now Riemann in the same terms. By the identity \(\lambda = (-1)^{\mathrm{rank}}\) the sum

\[L(x) \;=\; \sum_{n\le x} (-1)^{\mathrm{rank}(n)} \;=\; \bigl\lvert\{n\le x:\ \mathrm{rank}(n)\ \text{even}\}\bigr\rvert \;-\; \bigl\lvert\{n\le x:\ \mathrm{rank}(n)\ \text{odd}\}\bigr\rvert\]

is exactly the rank-parity imbalance on the interval \([1,x]\): the difference between the number of integers with an even and an odd number of prime factors. LiouvilleBound — \(|L(x)| = O(\sqrt{x}\cdot x^{\varepsilon})\) — says that this imbalance grows no faster than \(\sqrt{x}\): the rank parities are almost balanced, the deviation of "square-root" size, as in a random \(\pm1\) walk.

And this is not a heuristic: LiouvilleRHBridge is the classical equivalence \(\texttt{LiouvilleBound}\iff\mathrm{RH}\), and riemann_of_liouville_bound proves (conditionally on the bridge) \(\mathrm{RH}\) from the bound. All the analytic content of RH is isolated in the sign-balance estimate.

Let us set the two walls side by side, literally:

twin wall Riemann wall
object \(R_{\mathrm{fc}} = N_{00}N_{33}/(N_{03}N_{30})\) \(L(x) = \sum_{n\le x}(-1)^{\mathrm{rank}(n)}\)
controlled quantity balance of the side ranks \((\mathrm{rank}_-,\mathrm{rank}_+)\) balance of the rank parity \(\mathrm{rank}(n)\bmod 2\)
threshold margin \(1 - R_{\mathrm{fc}} \to 0^+\) $
loss of control \(N_{00}\) empties out \(\Rightarrow\) twins are finite \(L(x)\) grows \(\gg\sqrt{x}\) \(\Rightarrow\) a zero off \(1/2\)
common language RankNode, cardFactors RankNode, cardFactors

Section takeaway. Both lines are about how \(\mathrm{rank}(n) = \texttt{cardFactors}\,n\) is distributed as \(n\to\infty\), and both demand that this distribution not "collapse" into a single parity. This is the chapter's thesis.

Note. There is exactly one difference, and it does not obstruct the unity. The twins look at the joint distribution of the ranks of two shifted sides \(6m\pm1\) (a 2D balance of the cells \(N_{ij}\)); Riemann — at the marginal parity balance of a single \(n\) (a 1D balance of the signs of \(\lambda\)). The marginal is a shadow of the joint: control of the 2D distribution of side ranks dominates control of the 1D parity. It is therefore natural to expect that the mechanism yielding the first also yields the second, and not the other way round.

What is already shared in Lean (proven)

So as not to pass the wished-for off as done, let us separate the proven part. Shared by both hypotheses, already proven are precisely the connecting links, not the bounds themselves:

  • liouville_eq_neg_one_pow_rank, liouville_flip_of_mul_prime — the identity "Liouville sign = rank parity" and the flip under deleteFactor (from mathlib, with no inputs of ours);
  • RankNode.ext, deleteFactor (ProductCore.lean) — the common carrier and the rank-lowering operator, used both in the twins' product-rank descent and as the sign operator of \(\lambda\);
  • factor_rank_le_four, composite_rank_ge_two, mkNode_of_composite (MkNode.lean) — the finiteness of the rank range \(1\le r\le 4\) on the legal layer and the construction of a RankNode from a composite side — pure arithmetic;
  • engine_of_carrier_and_factorize, exists_infinite_fiber (CarrierBridge.lean) — the infinite pigeonhole over rank (pigeonhole — the boxes principle, see the glossary): an infinite carrier splits over \(\mathrm{Fin}\,4\) into rank classes, one of which is infinite; the whole pump machine (descent, the rank-1 base, pigeonhole) is assembled;
  • riemann_of_liouville_bound (RiemannLiouville.lean) — RH from LiouvilleBound through the classical bridge, conditionally.

Section takeaway. All of this is language and assembly, not bounds. It is proven that if one controls the rank distribution of the carrier (for the twins) or the parity balance of \(\lambda\) (for Riemann), then both hypotheses close through the already-proven machines (the EPMI engine / the classical bridge). What is not proven is the control itself.

The unity conjecture (honestly: a conjecture, not a theorem)

Now — the chapter's central statement, presented as what it is.

Conjecture 32.5 (Rank-Parity Unity). There exists a single mechanism controlling the distribution of the rank \(\mathrm{rank}(n) = \texttt{cardFactors}\,n\) at a growing scale — the same one required for GlobalOldAbsorption from 29/24 (the fan-in \(570\to 1\) yields an engine) — from which both bounds follow: $\(\underbrace{N_{00} - N_{33} > 0\ \text{uniformly}}_{\text{twins}} \qquad\text{and}\qquad \underbrace{|L(x)| = O(\sqrt{x}\cdot x^{\varepsilon})}_{\text{Riemann}}.\)$ The mechanism: control of the descent-forest (the forest of descent pedigrees) yields the distribution of rank over the absorber fibers, and this distribution simultaneously yields the twins' carrier balance (\(N_{00}\) does not empty out) and the balance of Liouville signs (\(L(x)\) is small).

Let us underline the honesty boundary in three points.

  1. Both bounds are open. Neither GlobalOldAbsorption (twins, RESULTS_final_gap.md: "irreducible node") nor LiouvilleBound (Riemann, RiemannLiouville.lean: "arithmetic input") is proven. The conjecture asserts not their truth but their commonality — that one and the same control closes them both.
  2. The unity is a conjecture, not a reduction. We have not proven the implication GlobalOldAbsorption ⟹ LiouvilleBound or its converse; had it been proven, RH would follow from the twins' node, and we explicitly rejected that back in 30 (the statements are independent). The unity is about the nature of the difficulty (both are rank-parity control), not about the logical derivability of one from the other.
  3. The final node is comparable in difficulty to the hypotheses themselves. As noted in 30 and 24, the last analytic/combinatorial input of each branch is comparable in difficulty to the hypothesis itself. The common language does not make the node easier — it makes it one: a single wall instead of two.

The closing plan (descent-forest ⟹ rank distribution ⟹ both balances)

The plan is inherited from [24, §Pigeonhole] and 27 and is now written down as a single programme for both walls.

Step 1 — descent-forest. Define the forest of descent pedigrees over the carrier: the vertices are states RankNode r, the edges are deleteFactor/old-peel steps (19, 24), the roots are absorbers \(\le M_0\). The fan-in \(570\to 1\) (24) is the branching of the forest: up to 570 pedigrees into one root. Control of the forest is control of how \(r\) decreases along the branches from the leaves to the root.

Step 2 — rank distribution over the fiber. From the structure of the forest, extract the distribution of \(\mathrm{rank}(n)\bmod 2\) over the fiber of each absorber. Here the already-proven exists_infinite_fiber does the work: an infinite fiber splits over the \(\mathrm{Fin}\,4\) rank classes. The key that must be exhibited: how deleteFactor along a branch changes the parity (the proven flip \(\lambda\mapsto-\lambda\), Theorem 32.4 (liouville_flip_of_mul_prime)) induces an almost-uniform distribution of parities on the fiber — this is exactly the shared rank-parity control.

Step 3a — twins. From the rank distribution over the fiber, derive the carrier balance: that the fraction of centres with both sides prime (\(N_{00}\)) is not suppressed by the fraction of double composites (\(N_{33}\)), that is, \(R_{\mathrm{fc}}<1\) uniformly. This is the transfer of the model four-corner to reality through the distribution of side ranks, not through a single marginal (16).

Step 3b — Riemann. From the same rank distribution, derive \(|L(x)| = O(\sqrt{x})\): the almost-uniformity of the rank parities is exactly the "random-walk" size of the partial sum of the signs of \(\lambda\). Formally — plug the balance control into riemann_of_liouville_bound via LiouvilleRHBridge.

Note (a single point of failure). Both steps 3 feed off the one step 2 — the distribution of rank parity over the descent-forest fiber. If this control is exhibited, both hypotheses close; if not — both are open. That is the operational meaning of the unity: not two independent proofs, but one rank-parity control with two consequences. The critical fork is the same normal-form trilemma from 24/29: the fiber signature must distinguish pedigrees (otherwise the balance is trivially false) and glue them into a finite key (otherwise the pigeonhole is empty). Until this fork is resolved, the unity remains a conjecture.

Epilogue of the prose

Here the map of the path closes. We began (00–11) with Euclid's engine and its laws — the conservation of two, irreversibility, the impossibility of a perpetual engine (EPMI, proven) — and reduced the twins to the block core.

We walked (12–21) the attack lines on the bound, each of which honestly ran into the one parity wall, and decomposed the exit from the pure graph (24) down to a single global node. We closed by arithmetic what could be closed: ProductHall via the separating scale (26), the rank-descent logic (27), mkNode and carrier infinitude (28–29) — lifting all the earlier walls (parity, the trilemma, steering, payment).

We built an independent branch to Riemann (30) and translated it into Liouville arithmetic (31). And in this final chapter we saw that the remaining twins node and the remaining Riemann node are one and the same, written as control of the rank-parity balance: RankNode, cardFactors, deleteFactor — the common language of both.

The honest summary of the whole prose: the machine is proven, the node is one, and it is open. The twins and Riemann are not derived and not passed off as derived; nowhere does the reduction substitute for a proof.

But two questions that stood apart for centuries stand here side by side — on one carrier, with one wall, with one plan.

From the intuition that guided the entire path, it is natural to expect that the wall to be taken is the rank-parity wall, and that taking it once means closing both. Proving this is work beyond this prose. Here we have named it precisely, localised it in a single node, and left it open.

← 31. Riemann via Liouville · Table of contents · 33. First cause and the main theorem →