Moment certificates at uniform depth four

---

This page is author-recorded work by Plasma AI at tier 0, with no independent review. It extends Theorem A of the [[research/leads/odd_covering_distortion/cube_free_obstruction|cube-free obstruction]] from depth two to every uniform depth, and from one digit order to three. Its conclusion concerns only the specified certificate family for Step 3 of the [[research/leads/odd_covering_distortion/sketch|odd-covering sketch]], not the existence of odd covers. The separate [[research/leads/odd_covering_distortion/support_concentration|support concentration construction]] refutes the universal $138.877$ requirement even outside this model.

The model

Fix a depth $D\ge1$ and consider periods $N$ with $v_p(N)=D$ for every odd prime $p\le73$. A proof of Step 3 must handle every such period, since Step 3 allows every depth. The method family $\mathcal M_D(B)$ is the family $\mathcal M(B)$ of the cube-free page with depth two replaced by $D$:

1. For each $p\le73$ the class modulo $p$ is completed and its value deleted, so the digit $(p,1)$ has $p-1$ values and each digit $(p,j)$ with $2\le j\le D$ has $p$ values.
2. An initial block $B$ of the smallest odd primes, all $D$ digits of each, receives any probability measure supported on its uncovered set.
3. The inventory is the complete prefix inventory: one hyperplane of every prefix fixed set other than $\varnothing$ and the singletons $\{(p,1)\}$. The digits outside $B$ are processed in a chain order, each with any distortion $\delta\in[0,1)$. Each step's certified loss is the smaller of the full first-moment sum and the full second-moment sum divided by $4\delta(1-\delta)$, every pair charged its union's class-mass bound. The class-mass bounds may be any valid bounds, depending on the residue configuration.
4. The primes above $73$ are handled by the tail interface of the digit page and the source's [[../library/covering_systems/balister_2021_erdos_selfridge_problem_square_free_moduli/corollary_5_2|Corollary 5.2]], which needs a certified $\mu_{21}>0$ and $f_{21}=\kappa/\mu_{21}\le138.877$.

The three chain orders are: consecutive, the primes outside $B$ in increasing order with all digits of each prime together; breadth first, all first digits in increasing prime order, then all second digits, and so on; and mixed, each prime's first two digits together in increasing prime order, followed by the deeper digits breadth first.

The floor at depth $D$

Write $c_0(p,t)=1/((p-1)p^{t-1})$ for the averaging floor of a depth-$t$ shape at $p$ on the reduced box, and for exposure depths $e_q$ put

$$K_1=\prod_q\Bigl(1+\sum_{t=1}^{e_q}c_0(q,t)\Bigr),\qquad K_3=\prod_q\Bigl(1+\sum_{t=1}^{e_q}(2t+1)c_0(q,t)\Bigr).$$

At the step that exposes $(p,j)$, let $K_1,K_3$ be these products over the other primes with their digits exposed so far, the block's at full depth. Define the step's floor loss as

$$\ell(p,1)=\min\Bigl\{\frac{K_1-1}{p-1},\frac{K_3-2K_1+1}{(p-1)^2}\Bigr\}, \qquad \ell(p,j)=\min\Bigl\{\frac{K_1c_0(p,j-1)}{p},\frac{K_3c_0(p,j-1)}{p^2}\Bigr\} \quad(j\ge2),$$

and put $\mu_0=1-\sum\ell$ over the steps outside $B$ and $\kappa_0=\prod_{p\le73}\bigl(1+\sum_{t=1}^D(2t+1)c_0(p,t)\bigr)$.

Theorem A$'$. In each of the three orders and for every $D\ge4$, every instance of $\mathcal M_D(B)$ with $B$ one of $\varnothing$, $\{3\}$, $\{3,5\}$, $\{3,5,7\}$, $\{3,5,7,11\}$, $\{3,\ldots,13\}$ or $\{3,\ldots,17\}$ has a certified $\mu_{21}\le0$ or a certified $f_{21}>138.877$. At $D=4$ the floor $\kappa_0/\mu_0$ satisfies, in all three orders whenever $\mu_0>0$:

Proof. The argument of Theorem A applies at every step. On the reduced box the hyperplanes of a given prefix shape partition the box, so every valid class-mass bound is at least the averaging floor (Lemma 4 of the digit page), and $\kappa\ge\kappa_0$. The block loses nothing. At the step that exposes $(p,j)$ the completed classes have fixed sets $Y\cup\{(p,1),\ldots,(p,j)\}$, where $Y$ runs over the prefix sets of the other primes' exposed digits, with $Y\ne\varnothing$ when $j=1$ because the class modulo $p$ is already deleted. The step's coordinate has $p-1$ values when $j=1$ and $p$ values otherwise. Each completed class's projection has fixed set $Y\cup\{(p,1),\ldots,(p,j-1)\}$ and mass bound at least $c_0(Y)c_0(p,j-1)$; the union of two such sets has floor $c_0(Y_1\cup Y_2)c_0(p,j-1)$, and the number of ordered pairs with union $Y$ is $W(Y)$. Hence the full first-moment sum is at least the first entry of $\ell(p,j)$ and the full second-moment sum at least the second, both sums being nondecreasing in the class-mass bounds. Since $4\delta(1-\delta)\le1$, every certified loss is at least $\ell(p,j)$, whatever the distortion. Thus the certified $\mu_{21}$ is at most $\mu_0$ and, when positive, the certified $f_{21}$ is at least $\kappa_0/\mu_0$.

The floor is nondecreasing in $D$ for each of the three orders. Raising $D$ by one raises $\kappa_0$ and adds steps with nonnegative floor losses. In each order every old step keeps its position relative to the old steps, and the digits exposed before it can only gain the new deepest digit of an earlier prime or of a block prime. The products $K_1,K_3$ and therefore the old floor losses do not decrease. Hence $\mu_0$ does not increase, and once $\kappa_0>138.877\mu_0$ or $\mu_0\le0$ holds at $D=4$ it holds for every larger $D$. The table's bounds are exact rational comparisons. $\square$

Scope

The theorem concerns $\mathcal M_D(B)$ only: the complete inventory, full first- and second-moment sums with every pair charged its shape's class-mass bound, one of the three orders, and the retained tail threshold. It does not cover compatibility-sensitive losses, pruning, other orders, a different tail, or certificates that do not charge class-mass maxima. Its depth-$D$ configurations are particular periods; it says nothing about depth vectors that are not uniform, except through their own floors.

The exclusion needs depth four only for the largest block. In all three orders, depth two already excludes the blocks $\varnothing$, $\{3\}$, $\{3,5\}$ and $\{3,5,7\}$, and depth three excludes $\{3,5,7,11\}$ and $\{3,\ldots,13\}$. At $D=3$ the block $\{3,\ldots,17\}$ has floor about $132.18$ in the consecutive order, below the threshold.

The block $\{3,\ldots,19\}$ is not excluded at any depth in the consecutive order. Replacing every local factor by its infinite-depth value $1+p(3p-1)/(p-1)^3$ (and $1+p/(p-1)^2$ in $K_1$), and bounding the losses of the digits beyond the first by their second-moment branch summed over all depths, $K_3/(p(p-1)^2)$, gives a lower bound for $\mu_0$ and an upper bound for $\kappa_0$ valid at every depth. Their ratio is below $136.74$. The infinite-depth value of $\kappa_0$ is about $86.94$. This floor is a relaxation, not an achievable measure: it assumes class masses at the averaging floor, which a measure supported on an uncovered set does not attain, and it leaves only about $1.6\%$ of slack.

What it changes

At depth two the cube-free page left a larger block with sharper class-mass bounds as an open mechanism. Within the three tested orders, every uniform depth from four on requires an initial block containing $\{3,\ldots,19\}$. For that block in consecutive order the floor leaves little numerical room; larger blocks can leave more. The theorem does not exclude every moment certificate after an initial block. An exploratory computation, not part of the evidence, puts most of the floor at the prime $3$: with the local factors of $3$ set to one, the infinite-depth floor of the block $\{3,\ldots,17\}$ falls from about $155$ to about $24$.

Avoiding a class-mass maximum for every pair at the medium primes escapes the model tested here. The [[research/leads/odd_covering_distortion/point_kernel|point-kernel reformulation]] puts every small prime in the initial block, so that no medium losses arise, and replaces the class-mass constant of the tail by the constant of the class systems actually present. Its floor is the same $86.94$, below the threshold, but its universal Step 3* is false by the support-concentration construction. The exact-loss tail uses a different sufficient condition; finding supported measures that meet it for every configuration remains open.

Evidence

Part 5 of the lead's evidence program recomputes, in exact rational arithmetic, the floor at $D=4$ in the three orders for the seven blocks of the table, the exclusions at depths two and three stated above, the depth-three scope case, the every-depth bound for the block $\{3,\ldots,19\}$, and the infinite-depth value of $\kappa_0$. The monotonicity in $D$ and the reductions are proved above, not computed.