Introduction
The problem
For a prime q ∤ r q\nmid r q ∤ r write d q = ord q ( r ) d_q=\operatorname{ord}_q(r) d q = ord q ( r ) for the multiplicative order and s q = v q ( r q − 1 − 1 ) = v q ( r d q − 1 ) ≥ 1 s_q=v_q(r^{q-1}-1)=v_q(r^{d_q}-1)\ge1 s q = v q ( r q − 1 − 1 ) = v q ( r d q − 1 ) ≥ 1 for the Wieferich depth ; the Fermat quotient is q q ( u ) = ( u q − 1 − 1 ) / q q_q(u)=(u^{q-1}-1)/q q q ( u ) = ( u q − 1 − 1 ) / q , and q ∈ W r : = { q : s q ≥ 2 } q\in W_r:=\{q:s_q\ge2\} q ∈ W r := { q : s q ≥ 2 } are the Wieferich primes to base r r r . The set W r W_r W r is the basic obstruction in several classical problems: the first case of Fermat’s Last Theorem (Wieferich), the infinitude of non-Wieferich primes (Silverman), and—most recently—the remaining hypothesis ( ⋆ ) (\star) ( ⋆ ) in Zeng’s lattice-theoretic approach [1 ] to the a b c abc ab c conjecture, which is the motivating application of this note.
On the “fixed-modulus” side (fixed p p p , varying base) the Fermat quotient is well understood: Heath–Brown’s Burgess-based estimates, the GRH coverage results of Alexandru–Cobeli–Văjăitu-Zaharescu, and Shparlinski’s unconditional improvements of the Ihara sum. On the “varying-modulus” side (fixed r r r , q q q varying) not even the zero-density of W r W_r W r is known unconditionally. The present note shows that the application layer of ( ⋆ ) (\star) ( ⋆ ) can be reduced to a single summability condition on this varying-modulus side, and isolates the remaining analytic input as a precisely located, numerically supported hypothesis.
Main results
Throughout, q q q denotes an odd prime; the prime q = 2 q=2 q = 2 is handled by the LTE branch of Theorem 1.1 (so D 3 ( 2 ) ∈ { 1 , 2 } D_3(2)\in\{1,2\} D 3 ( 2 ) ∈ { 1 , 2 } ) and contributes O r , a ′ ( E ) O_{r,a'}(E) O r , a ′ ( E ) to S ( E ) S(E) S ( E ) , absorbed into the constant C C C of a first-moment bound S ( E ) ≤ C E S(E)\le CE S ( E ) ≤ C E . Also, r ≥ 2 r\ge2 r ≥ 2 is a fixed integer (not a perfect power, without loss of generality), A ≥ 1 A\ge1 A ≥ 1 is fixed, a ′ ∈ { 1 , … , A } a'\in\{1,\ldots,A\} a ′ ∈ { 1 , … , A } , gcd ( a ′ , r ) = 1 \gcd(a',r)=1 g cd( a ′ , r ) = 1 , and b e = r e − a ′ b_e=r^e-a' b e = r e − a ′ . Set V q ( e ) = v q ( b e ) V_q(e)=v_q(b_e) V q ( e ) = v q ( b e ) , L e = e log r L_e=e\log r L e = e log r , and define the exceptional mass
T s ( e ) = ∑ q ∣ b e V q ( e ) ≥ 3 s q ≥ 2 s q log q , S ( E ) = ∑ e ≤ E T s ( e ) . T_s(e)=\sum_{\substack{q\mid b_e\\ V_q(e)\ge3\\ s_q\ge2}}s_q\log q,\qquad S(E)=\sum_{e\le E}T_s(e). T s ( e ) = q ∣ b e V q ( e ) ≥ 3 s q ≥ 2 ∑ s q log q , S ( E ) = e ≤ E ∑ T s ( e ) .
Call e e e W \mathcal{W} W -exceptional if T s ( e ) ≥ 1 20 L e T_s(e)\ge\frac{1}{20}L_e T s ( e ) ≥ 20 1 L e .
Theorem 1.1 (Order-lifting formula). For odd q ∤ r q\nmid r q ∤ r and every V ≥ 1 V\ge1 V ≥ 1 ,
D V ( q ) : = ord q V ( r ) = d q q max ( 0 , V − s q ) . D_V(q):=\operatorname{ord}_{q^V}(r)=d_q q^{\max(0,V-s_q)}. D V ( q ) := ord q V ( r ) = d q q m a x ( 0 , V − s q ) . For q = 2 q=2 q = 2 : if s 2 = v 2 ( r − 1 ) ≥ 2 s_2=v_2(r-1)\ge2 s 2 = v 2 ( r − 1 ) ≥ 2 the same formula holds; if s 2 = 1 s_2=1 s 2 = 1 (i.e. r ≡ 3 ( m o d 4 ) r\equiv3\pmod4 r ≡ 3 ( mod 4 ) ) then
D V ( 2 ) = 2 max ( 1 , V − v 2 ( r + 1 ) ) . D_V(2)=2^{\max(1,V-v_2(r+1))}. D V ( 2 ) = 2 m a x ( 1 , V − v 2 ( r + 1 )) . The classical layer s q = 2 s_q=2 s q = 2 carries D 3 = d q q D_3=d_q q D 3 = d q q (a “Hensel wall” of one extra factor q q q ), while the super-Wieferich layer s q ≥ 3 s_q\ge3 s q ≥ 3 has D 3 = d q D_3=d_q D 3 = d q (wall-free). The equivalent form s q ≥ 2 ⟺ ord q 2 ( r ) = ord q ( r ) s_q\ge2\Longleftrightarrow\operatorname{ord}_{q^2}(r)=\operatorname{ord}_q(r) s q ≥ 2 ⟺ ord q 2 ( r ) = ord q ( r ) identifies W r \mathcal{W}_r W r with the order-lift locus of Dillon [12 ] .
Theorem 1.2 (Two-base necessary condition). Let q q q be odd, s q ≥ 2 s_q\ge2 s q ≥ 2 . If q 3 ∣ r e − a ′ q^3\mid r^e-a' q 3 ∣ r e − a ′ for some e ≥ 1 e\ge1 e ≥ 1 , then q ∤ a ′ q\nmid a' q ∤ a ′ and q 2 ∣ a ′ q − 1 − 1 q^2\mid a'^{q-1}-1 q 2 ∣ a ′ q − 1 − 1 (i.e. q ∈ W a ′ q\in\mathcal{W}_{a'} q ∈ W a ′ ); if moreover s q ≥ 3 s_q\ge3 s q ≥ 3 then q 3 ∣ a ′ q − 1 − 1 q^3\mid a'^{q-1}-1 q 3 ∣ a ′ q − 1 − 1 . Writing w q = ( r q − 1 − 1 ) / q 2 m o d q w_q=(r^{q-1}-1)/q^2\bmod q w q = ( r q − 1 − 1 ) / q 2 mod q , β q = ( a ′ q − 1 − 1 ) / q 2 m o d q \beta_q=(a'^{q-1}-1)/q^2\bmod q β q = ( a ′ q − 1 − 1 ) / q 2 mod q (defined when s q = 2 s_q=2 s q = 2 ), the least solution e 0 ( q ) e_0(q) e 0 ( q ) satisfies e 0 ( q ) ≡ β q w q − 1 ( m o d q ) e_0(q)\equiv\beta_qw_q^{-1}\pmod q e 0 ( q ) ≡ β q w q − 1 ( mod q ) , and β q = 0 \beta_q=0 β q = 0 forces e 0 ( q ) ≥ q e_0(q)\ge q e 0 ( q ) ≥ q . Finally
e 0 ( q ) ≥ 3 log q log r − O r , a ′ ( 1 ) e_0(q)\ge\frac{3\log q}{\log r}-O_{r,a'}(1) e 0 ( q ) ≥ log r 3 log q − O r , a ′ ( 1 ) unconditionally.
Theorem 1.2 compresses the candidate enumeration from a scan over e e e to a three-filter sieve per prime: q ∈ W r q\in\mathcal{W}_r q ∈ W r , q ∈ W a ′ q\in\mathcal{W}_{a'} q ∈ W a ′ , and a ′ ∈ ⟨ r ⟩ ⊆ ( Z / q 3 Z ) × a'\in\langle r\rangle\subseteq(\mathbb{Z}/q^3\mathbb{Z})^\times a ′ ∈ ⟨ r ⟩ ⊆ ( Z / q 3 Z ) × ; the last filter is necessary but not sufficient [17 ] , numerics.
Theorem 1.3 (One-condition closure). For q q q in the compatible set C r , a ′ \mathcal{C}_{r,a'} C r , a ′ (the three-filter set above), let e 0 ( q ) e_0(q) e 0 ( q ) be as in Theorem 1.2 and define
K r , a ′ = ∑ q ∈ C r , a ′ s q log q ( 1 D 3 ( q ) + 1 e 0 ( q ) ) . K_{r,a'}=\sum_{q\in\mathcal{C}_{r,a'}}s_q\log q\left(\frac{1}{D_3(q)}+\frac{1}{e_0(q)}\right). K r , a ′ = q ∈ C r , a ′ ∑ s q log q ( D 3 ( q ) 1 + e 0 ( q ) 1 ) . If K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ , then S ( E ) ≤ E K r , a ′ S(E)\le EK_{r,a'} S ( E ) ≤ E K r , a ′ , and consequently
1 E # { e ≤ E : T s ( e ) ≥ e log r 20 } ≤ 20 K r , a ′ log r ⋅ 1 + ln E E ⟶ 0. \frac{1}{E}\#\left\{e\le E:T_s(e)\ge\frac{e\log r}{20}\right\}\le\frac{20K_{r,a'}}{\log r}\cdot\frac{1+\ln E}{E}\longrightarrow0. E 1 # { e ≤ E : T s ( e ) ≥ 20 e log r } ≤ log r 20 K r , a ′ ⋅ E 1 + ln E ⟶ 0.
Theorem 1.3 uses no GRH, no sieve, and no information on the distribution of ord q ( r ) \operatorname{ord}_q(r) ord q ( r ) : the hypothesis V q ( e ) ≥ 3 V_q(e)\ge3 V q ( e ) ≥ 3 has already converted the weight 1 / ord q ( r ) 1/\operatorname{ord}_q(r) 1/ ord q ( r ) into 1 / ord q 3 ( r ) 1/\operatorname{ord}_{q^3}(r) 1/ ord q 3 ( r ) .
Theorem 1.4 (Bad-set structure). Write e 0 ( q ) = b q + j q q e_0(q)=b_q+j_qq e 0 ( q ) = b q + j q q with b q = ( β q w q − 1 m o d q ) ∈ [ 0 , q ) b_q=(\beta_qw_q^{-1}\bmod q)\in[0,q) b q = ( β q w q − 1 mod q ) ∈ [ 0 , q ) and j q ∈ [ 0 , d q ) j_q\in[0,d_q) j q ∈ [ 0 , d q ) (layer s q = 2 s_q=2 s q = 2 ). If j q ≥ 1 j_q\ge1 j q ≥ 1 then e 0 ( q ) ≥ q e_0(q)\ge q e 0 ( q ) ≥ q and s q log q / e 0 ( q ) ≤ s q log q / q s_q\log q/e_0(q)\le s_q\log q/q s q log q / e 0 ( q ) ≤ s q log q / q , so the s q = 2 s_q=2 s q = 2 part of this layer is controlled directly by P 4 P_4 P 4 :
∑ q ∈ W r log q q < ∞ , \sum_{q\in\mathcal{W}_r}\frac{\log q}{q}<\infty, q ∈ W r ∑ q log q < ∞ , *and the s q ≥ 3 s_q\ge3 s q ≥ 3 part by P 6 a P_{6a} P 6 a (finiteness of W r ( 3 ) \mathcal{W}_r^{(3)} W r ( 3 ) ). The unique remaining layer is the bad set B r , a ′ = { q ∈ C r , a ′ : s q = 2 , j q = 0 } = { q ∈ C r , a ′ : s q = 2 , e 0 ( q ) < q } \mathcal{B}_{r,a'}=\{q\in\mathcal{C}_{r,a'}:s_q=2,\ j_q=0\}=\{q\in\mathcal{C}_{r,a'}:s_q=2,\ e_0(q)<q\} B r , a ′ = { q ∈ C r , a ′ : s q = 2 , j q = 0 } = { q ∈ C r , a ′ : s q = 2 , e 0 ( q ) < q } (the super-Wieferich layer s q ≥ 3 s_q\ge3 s q ≥ 3 , for which $e_0\le d_q
:::`
may be < q <q < q , is controlled by P 6 a P_{6a} P 6 a and is not part of B r , a ′ \mathcal{B}_{r,a'} B r , a ′ ; see § 12 , and the remaining hypothesis is
∑ q ∈ B r , a ′ s q log q e 0 ( q ) < ∞ . (1) \sum_{q\in\mathcal{B}_{r,a'}} \frac{s_q\log q}{e_0(q)}<\infty. \tag*{(1)} q ∈ B r , a ′ ∑ e 0 ( q ) s q log q < ∞. (1)
The set B r , a ′ \mathcal{B}_{r,a'} B r , a ′ is nonempty already at ( r , a ′ , q ) = ( 17 , 19 , 3 ) (r,a',q)=(17,19,3) ( r , a ′ , q ) = ( 17 , 19 , 3 ) . Layerwise, { q : e 0 ( q ) = b } ⊆ { q : q 3 ∣ r b − a ′ } \{q:e_0(q)=b\}\subseteq\{q:q^3\mid r^b-a'\} { q : e 0 ( q ) = b } ⊆ { q : q 3 ∣ r b − a ′ } is finite for each fixed b b b [17 Layer census] , so divergence of (H-j) can only come from a bad sequence with e 0 ( q n ) → ∞ e_0(q_n)\to\infty e 0 ( q n ) → ∞ .
The subtraction chain organizing this note:
( A P ) ⟶ K r , a ′ < ∞ ⟶ Ξ r < ∞ ⟶ Ξ r ( 2 ) < ∞ ⏟ Theorem 7.5, unconditional + Ξ r ( 3 ) < ∞ ⏟ sufficient, not necessary (Thm. 9.4) . (\mathrm{AP})\longrightarrow K_{r,a'}<\infty\longrightarrow\Xi_r<\infty\longrightarrow\underbrace{\Xi_r^{(2)}<\infty}_{\text{Theorem 7.5, unconditional}}+\underbrace{\Xi_r^{(3)}<\infty}_{\text{sufficient, not necessary (Thm. 9.4)}}. ( AP ) ⟶ K r , a ′ < ∞ ⟶ Ξ r < ∞ ⟶ Theorem 7.5, unconditional Ξ r ( 2 ) < ∞ + sufficient, not necessary (Thm. 9.4) Ξ r ( 3 ) < ∞ .
The exact threshold of the Markov route is Φ r ( E ) = o ( E ) \Phi_r(E)=o(E) Φ r ( E ) = o ( E ) , equivalently A r ( t ) = o ( t ) A_r(t)=o(t) A r ( t ) = o ( t ) (Theorem 9.6 ); Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ and its weakenings Ξ r ( 3 ) ( E ) = o ( E / log E ) \Xi_r^{(3)}(E)=o(E/\log E) Ξ r ( 3 ) ( E ) = o ( E / log E ) , K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ are sufficient rungs of the ladder of § 9 , none of them necessary.
Contributions. (i) Theorem 7.5 (main, unconditional): Ξ r ( 2 ) < ∞ \Xi_r^{(2)}<\infty Ξ r ( 2 ) < ∞ for every base r r r — block decomposition by d q d_q d q plus Stieltjes partial summation, with explicit bound O ( 1 + log log r ) O(1+\log\log r) O ( 1 + log log r ) . (ii) Theorems 8.1–8.4 and Proposition 8.5: the remaining gap is squeezed strictly between P 6 a P_{6a} P 6 a and P 4 ( 3 ) P_4^{(3)} P 4 ( 3 ) , explained structurally (the Hensel wall is the only unconditional decay source), and shown unreachable by block/partial-summation methods along three independent barriers, with a fourth (radical) withdrawn as a bookkeeping artifact and replaced by an identity (Theorem 8.6 ). (iii) Theorems 8.10 and 10.3 : under the Katz double-zero hypothesis (D4), Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ and P 6 a P_{6a} P 6 a hold almost surely (Borel–Cantelli), and (K+) unifies the main line and the bad-set line into one conjecture family; the counting-density reading of (D4) is equivalent to P 6 a P_{6a} P 6 a (Theorem 14.4 ). (iv) § 11 : the unconditional ∑ p log p / ( d p ( p − 1 ) ) < ∞ \sum_p\log p/(d_p(p-1))<\infty ∑ p log p / ( d p ( p − 1 )) < ∞ (Erdős–Murty), the exact identity C r = S W + S o r d + L T E 2 C_r=SW+S_{\rm ord}+LT\,E_2 C r = S W + S ord + L T E 2 , and the strict one-way S W < ∞ ⇒ Ξ r < ∞ SW<\infty\Rightarrow\Xi_r<\infty S W < ∞ ⇒ Ξ r < ∞ ; GRH is redundant throughout the chain. (v) § 9 : the exact threshold of the unconditional question — an elementary pointwise treatment of Case A, the unconditional linear cap Φ r ( E ) ≤ 6 π 2 ( 1 + γ ) log ( r + 1 ) E + o ( E ) \Phi_r(E)\le\frac{6}{\pi^2}(1+\gamma)\log(r+1)E+o(E) Φ r ( E ) ≤ π 2 6 ( 1 + γ ) log ( r + 1 ) E + o ( E ) , the exact threshold Φ r ( E ) = o ( E ) ⟺ A r ( t ) = o ( t ) \Phi_r(E)=o(E)\Longleftrightarrow A_r(t)=o(t) Φ r ( E ) = o ( E ) ⟺ A r ( t ) = o ( t ) (Theorem 9.6 ), the weakest sufficient rung of the ladder Ξ r ( 3 ) ( E ) = o ( E / log E ) \Xi_r^{(3)}(E)=o(E/\log E) Ξ r ( 3 ) ( E ) = o ( E / log E ) (strictly above the threshold, Theorem 9.4 ), and the information-theoretic barrier showing the deep layer is unreachable by any finite-order congruence expansion.
Locating (H-j)
Katz’s Wieferich quotient W u ( q ) = ( u q − 1 − 1 ) / q 2 W_u(q)=(u^{q-1}-1)/q^2 W u ( q ) = ( u q − 1 − 1 ) / q 2 takes values in 1 q 2 Z / Z ⊂ R / Z \frac{1}{q^2}\mathbb{Z}/\mathbb{Z}\subset\mathbb{R}/\mathbb{Z} q 2 1 Z / Z ⊂ R / Z , and Katz conjectured Haar equidistribution for Zariski-dense points on group schemes [2 ] . Writing q q ( u ) = k u + q m u q_q(u)=k_u+q\,m_u q q ( u ) = k u + q m u (k u , m u ∈ [ 0 , q ) k_u,m_u\in[0,q) k u , m u ∈ [ 0 , q ) ), one has q ∣ W u ⟺ k u = 0 q\mid W_u\Longleftrightarrow k_u=0 q ∣ W u ⟺ k u = 0 , and then w q = m q / β m q w_q=m_q/\beta_{m_q} w q = m q / β m q ; the pair ( w q , β q ) (w_q,\beta_q) ( w q , β q ) is the mod-q q q higher layer of the Katz quotient at its Haar-null zeros. The hypothesis (H-j) is controlled by the three-layer discretized joint refinement (K)=(D1)+(D2)+(D3) of § 10 , and we show (Proposition 10.1 ) that (K) implies both (H-j) and K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ . Moreover:
Theorem 1.5 (not a GRH statement). No known theorem gives, under GRH, a nontrivial bound on N r ( x ) N_r(x) N r ( x ) , the density of W r \mathcal{W}_r W r , or P 4 P_4 P 4 ; the natural Chebotarev fields L q = Q ( ζ q 2 , r 1 / q 2 ) L_q=\mathbb{Q}(\zeta_{q^2},r^{1/q^2}) L q = Q ( ζ q 2 , r 1/ q 2 ) have discriminants of size q O ( q 2 ) q^{O(q^2)} q O ( q 2 ) , so quantitative GRH–Chebotarev error terms are uncontrolled; and GRH controls only the ord q ( r ) \operatorname{ord}_q(r) ord q ( r ) (Artin) side, which the Hensel wall 1 / D 3 1/D_3 1/ D 3 has already absorbed. Hence “GRH ⇒ \Rightarrow ⇒ (H-j)” would mispackage an open problem as a conditional theorem.
Instead (H-j) sits strictly between Katz’s conjecture and Murata’s 1991 conjectures [3 ] : it is stronger than Katz (pointwise discretization plus mod-q 2 q^2 q 2 two-base joint equidistribution plus discrete-logarithm uniformity) and weaker than Murata (ii) (finiteness of W r ∩ W a ′ W_r \cap W_{a'} W r ∩ W a ′ ); Murata (i) is exactly P 6 a P_{6a} P 6 a .
Relation to (⋆′)
In Zeng’s framework [1 ] the remaining gap is stated in Section 7 (“The remaining gap, exactly”) as Working hypothesis 6.4: ∣ log x ∣ = o ( log c ) |\log x|=o(\log c) ∣ log x ∣ = o ( log c ) , equivalently O ( log log c ) O(\log\log c) O ( log log c ) , with x = det ( Λ ) ( k − 1 ) / k / Δ x=\det(\Lambda)^{(k-1)/k}/\Delta x = det ( Λ ) ( k − 1 ) / k /Δ and Δ = N ∣ a ′ ∣ 2 \Delta=N|a'|_2 Δ = N ∣ a ′ ∣ 2 ; Theorem 6.5 there gives η ≤ 1 k + O ( log log c log c ) \eta\le\frac{1}{k}+O\left(\frac{\log\log c}{\log c}\right) η ≤ k 1 + O ( l o g c l o g l o g c ) and Theorem 7.1 the equivalent form N ∣ a ′ ∣ 2 = ( c θ c / g ) ( k − 1 ) / k c o ( 1 ) N|a'|_2=(c\theta_c/g)^{(k-1)/k}c^{o(1)} N ∣ a ′ ∣ 2 = ( c θ c / g ) ( k − 1 ) / k c o ( 1 ) (cf. §13). The expression k − 1 k log c − log rad ( a b ) \frac{k-1}{k}\log c-\log\operatorname{rad}(ab) k k − 1 log c − log rad ( ab ) is a heuristic transcription, not a formula of [1 ] (see Remark 13.1); and “degeneration to a b c abc ab c strength” should be read as the structural ceiling of Theorem 10.1 there (Π ≤ 1 / k + o ( 1 ) \Pi\le1/k+o(1) Π ≤ 1/ k + o ( 1 ) , attained at c = r e c=r^e c = r e ; full a b c abc ab c is structurally unreachable inside the framework).
On the line a ′ = O ( 1 ) a'=O(1) a ′ = O ( 1 ) , b e = r e − a ′ b_e=r^e-a' b e = r e − a ′ , one has k e → ∞ k_e\to\infty k e → ∞ , so the requirement degenerates to a b c abc ab c -strength already for bounded a ′ a' a ′ , and Theorem 1.3 governs the k = 2 k=2 k = 2 slice. Theorem 1.2 pins the Layer-III Pillai candidates to W r ∩ W a ′ W_r\cap W_{a'} W r ∩ W a ′ with an explicit class e m o d q e\bmod q e mod q ; Theorem 1.3 pins the Layer-III mass to J r , a ′ J_{r,a'} J r , a ′ , the Pillai-single-solution layer; and (H-j) shows the application-layer gap of Layer III does not require Baker-type bounds—Baker belongs to the deeper k ≥ 3 k\ge3 k ≥ 3 , V q = 2 V_q=2 V q = 2 layer, untouched here. The defect/order-core framework of [15 ] independently corroborates this picture on the Mersenne line.
Preliminaries and notation
For an odd prime q ∤ r q\nmid r q ∤ r , LTE in the form v q ( r d q m − 1 ) = s q + v q ( m ) v_q(r^{d_qm}-1)=s_q+v_q(m) v q ( r d q m − 1 ) = s q + v q ( m ) (valid since v q ( r d q − 1 ) = s q ≥ 1 v_q(r^{d_q}-1)=s_q\ge1 v q ( r d q − 1 ) = s q ≥ 1 ) is used repeatedly. The unconditional bounds
d q ≥ log q log r , s q log q < d q log r (2) d_q\ge\frac{\log q}{\log r},\qquad s_q\log q<d_q\log r \tag*{(2)} d q ≥ log r log q , s q log q < d q log r (2)
follow from q ≤ r d q − 1 q\le r^{d_q}-1 q ≤ r d q − 1 and q s q ≤ r d q − 1 q^{s_q}\le r^{d_q}-1 q s q ≤ r d q − 1 . We write Σ r = ∑ q ∈ W r log q q \Sigma_r=\sum_{q\in W_r}\frac{\log q}{q} Σ r = ∑ q ∈ W r q l o g q (P 4 P_4 P 4 is Σ r < ∞ \Sigma_r<\infty Σ r < ∞ ), Θ r [ s ] ( ∞ ) = ∑ q ∈ W r s q log q q \Theta_r^{[s]}(\infty)=\sum_{q\in W_r}\frac{s_q\log q}{q} Θ r [ s ] ( ∞ ) = ∑ q ∈ W r q s q l o g q (P 6 b P_{6b} P 6 b ), and N r ( x ) = # { q ≤ x : q ∈ W r } N_r(x)=\#\{q\le x:q\in W_r\} N r ( x ) = # { q ≤ x : q ∈ W r } . The chain (H-Θ) ⇒ P 6 b ⇒ ∑ q ∈ W r ( 3 ) log q / q < ∞ \Rightarrow P_{6b}\Rightarrow\sum_{q\in W_r^{(3)}}\log q/q<\infty ⇒ P 6 b ⇒ ∑ q ∈ W r ( 3 ) log q / q < ∞ holds by the Fubini identity s q = ∑ k ≥ 2 1 W r ( k ) ( q ) s_q=\sum_{k\ge2}1_{W_r^{(k)}}(q) s q = ∑ k ≥ 2 1 W r ( k ) ( q ) (nonnegativity of the summands); P 6 a P_{6a} P 6 a (finiteness of W r ( 3 ) W_r^{(3)} W r ( 3 ) ) is strictly stronger in general—an infinite W r ( 3 ) W_r^{(3)} W r ( 3 ) can be sparse enough for the weighted sum to converge. Both reverse implications fail [17 ] . We use ∼ \sim ∼ , ≪ \ll ≪ , O ( ⋅ ) O(\cdot) O ( ⋅ ) with their standard meanings; all implied constants may depend on r r r and a ′ a' a ′ unless stated.
Proof of Theorem 1.1
Write r d q = 1 + q s q u r^{d_q}=1+q^{s_q}u r d q = 1 + q s q u with q ∤ u q\nmid u q ∤ u . For odd q q q and m ≥ 1 m\ge1 m ≥ 1 ,
r d q m − 1 = ( r d q ) m − 1 = q s q u ∑ i = 0 m − 1 r d q i , v q ( r d q m − 1 ) = s q + v q ( m ) , r^{d_qm}-1=(r^{d_q})^m-1=q^{s_q}u\sum_{i=0}^{m-1}r^{d_qi},\qquad v_q(r^{d_qm}-1)=s_q+v_q(m), r d q m − 1 = ( r d q ) m − 1 = q s q u i = 0 ∑ m − 1 r d q i , v q ( r d q m − 1 ) = s q + v q ( m ) ,
because the sum ∑ i = 0 m − 1 r d q i ≡ m ( m o d q ) \sum_{i=0}^{m-1}r^{d_qi}\equiv m\pmod q ∑ i = 0 m − 1 r d q i ≡ m ( mod q ) and induction on v q ( m ) v_q(m) v q ( m ) gives v q ( ∑ i = 0 m − 1 r d q i ) = v q ( m ) v_q\left(\sum_{i=0}^{m-1}r^{d_qi}\right)=v_q(m) v q ( ∑ i = 0 m − 1 r d q i ) = v q ( m ) (the standard LTE step). Moreover ord q V ( r ) \operatorname{ord}_{q^V}(r) ord q V ( r ) is a multiple of d q = ord q ( r ) d_q=\operatorname{ord}_q(r) d q = ord q ( r ) . Hence the least m m m with q V ∣ r d q m − 1 q^V\mid r^{d_qm}-1 q V ∣ r d q m − 1 is m = q max ( 0 , V − s q ) m=q^{\max(0,V-s_q)} m = q m a x ( 0 , V − s q ) , giving D V ( q ) = d q q max ( 0 , V − s q ) D_V(q)=d_qq^{\max(0,V-s_q)} D V ( q ) = d q q m a x ( 0 , V − s q ) . For q = 2 q=2 q = 2 with s 2 ≥ 2 s_2\ge2 s 2 ≥ 2 the same LTE applies; if s 2 = 1 s_2=1 s 2 = 1 then w : = v 2 ( r 2 − 1 ) = 1 + v 2 ( r + 1 ) ≥ 3 w:=v_2(r^2-1)=1+v_2(r+1)\ge3 w := v 2 ( r 2 − 1 ) = 1 + v 2 ( r + 1 ) ≥ 3 , so r 2 ≡ 1 ( m o d 8 ) r^2\equiv1\pmod8 r 2 ≡ 1 ( mod 8 ) , v 2 ( r 2 m − 1 ) = w + v 2 ( m ) v_2(r^{2m}-1)=w+v_2(m) v 2 ( r 2 m − 1 ) = w + v 2 ( m ) , and the least e e e with 2 V ∣ r e − 1 2^V\mid r^e-1 2 V ∣ r e − 1 is 2 max ( 1 , V − w + 1 ) 2^{\max(1,V-w+1)} 2 m a x ( 1 , V − w + 1 ) , i.e. D V ( 2 ) = 2 max ( 1 , V − v 2 ( r + 1 ) ) D_V(2)=2^{\max(1,V-v_2(r+1))} D V ( 2 ) = 2 m a x ( 1 , V − v 2 ( r + 1 )) . □
Remark 3.1. A wall-illustration: ( a ′ , r , q ) = ( 2 , 23 , 3 ) (a',r,q)=(2,23,3) ( a ′ , r , q ) = ( 2 , 23 , 3 ) gives v 3 ( 23 e − 2 ) ≥ 10 ⟺ e ≡ 5 ( m o d 39366 ) v_3(23^e-2)\ge10 \Longleftrightarrow e\equiv5\pmod{39366} v 3 ( 2 3 e − 2 ) ≥ 10 ⟺ e ≡ 5 ( mod 39366 ) ; the unique hit below r e ≤ 10 14 r^e\le10^{14} r e ≤ 1 0 14 is e = 5 e=5 e = 5 (the Reyssat triple), and the next candidate has c ≈ 10 53600 c\approx10^{53600} c ≈ 1 0 53600 .
Proof of Theorem 1.2
From r e ≡ a ′ ( m o d q 3 ) r^e\equiv a'\pmod{q^3} r e ≡ a ′ ( mod q 3 ) raise to the ( q − 1 ) (q-1) ( q − 1 ) -st power: a ′ q − 1 ≡ r e ( q − 1 ) ( m o d q 3 ) a'^{q-1}\equiv r^{e(q-1)}\pmod{q^3} a ′ q − 1 ≡ r e ( q − 1 ) ( mod q 3 ) . Write t = min ( s q , 3 ) t=\min(s_q,3) t = min ( s q , 3 ) and r q − 1 = 1 + q t w r^{q-1}=1+q^tw r q − 1 = 1 + q t w , q ∤ w q\nmid w q ∤ w (for s q = 2 s_q=2 s q = 2 ). The binomial expansion gives, since 2 t ≥ 4 > 3 2t\ge4>3 2 t ≥ 4 > 3 ,
r e ( q − 1 ) = ( 1 + q t w ) e ≡ 1 + e q t w ( m o d q 3 ) . r^{e(q-1)}=(1+q^tw)^e\equiv1+eq^tw\pmod{q^3}. r e ( q − 1 ) = ( 1 + q t w ) e ≡ 1 + e q t w ( mod q 3 ) .
If s q ≥ 3 s_q\ge3 s q ≥ 3 then t = 3 t=3 t = 3 and r q − 1 ≡ 1 ( m o d q 3 ) r^{q-1}\equiv1\pmod{q^3} r q − 1 ≡ 1 ( mod q 3 ) , so a ′ q − 1 ≡ 1 ( m o d q 3 ) a'^{q-1}\equiv1\pmod{q^3} a ′ q − 1 ≡ 1 ( mod q 3 ) . If s q = 2 s_q=2 s q = 2 then a ′ q − 1 ≡ 1 + e q 2 w ( m o d q 3 ) a'^{q-1}\equiv1+eq^2w\pmod{q^3} a ′ q − 1 ≡ 1 + e q 2 w ( mod q 3 ) , giving q 2 ∣ a ′ q − 1 − 1 q^2\mid a'^{q-1}-1 q 2 ∣ a ′ q − 1 − 1 and ( a ′ q − 1 − 1 ) / q 2 ≡ e w ( m o d q ) (a'^{q-1}-1)/q^2\equiv ew\pmod q ( a ′ q − 1 − 1 ) / q 2 ≡ e w ( mod q ) ; solving for the least representative gives e 0 ≡ β q w q − 1 ( m o d q ) e_0\equiv\beta_qw_q^{-1}\pmod q e 0 ≡ β q w q − 1 ( mod q ) , and β q = 0 \beta_q=0 β q = 0 forces e 0 ≡ 0 ( m o d q ) e_0\equiv0\pmod q e 0 ≡ 0 ( mod q ) , i.e. e 0 ≥ q e_0\ge q e 0 ≥ q . Finally q 3 ≤ ∣ r e 0 − a ′ ∣ ≤ r e 0 + ∣ a ′ ∣ q^3\le|r^{e_0}-a'|\le r^{e_0}+|a'| q 3 ≤ ∣ r e 0 − a ′ ∣ ≤ r e 0 + ∣ a ′ ∣ (using b e 0 ≠ 0 b_{e_0}\ne0 b e 0 = 0 ) yields e 0 ≥ 3 log q log r − O r , a ′ ( 1 ) e_0\ge\frac{3\log q}{\log r}-O_{r,a'}(1) e 0 ≥ l o g r 3 l o g q − O r , a ′ ( 1 ) . □
Proof of Theorem 1.3
We need two lemmas.
Proof. Partition [ 1 , E ] [1,E] [ 1 , E ] into dyadic blocks ( E / 2 j + 1 , E / 2 j ] (E/2^{j+1},E/2^j] ( E / 2 j + 1 , E / 2 j ] , 0 ≤ j < ⌈ log 2 E ⌉ 0\le j<\lceil\log_2 E\rceil 0 ≤ j < ⌈ log 2 E ⌉ . Inside block j j j one has e > E j : = E / 2 j + 1 e>E_j:=E/2^{j+1} e > E j := E / 2 j + 1 , hence the exceptional condition T s ( e ) ≥ c e > c E j T_s(e)\ge ce>cE_j T s ( e ) ≥ ce > c E j ; Markov with the fixed threshold c E j cE_j c E j gives at most S ( E / 2 j ) / ( c E j ) ≤ ( C E / 2 j ) / ( c E / 2 j + 1 ) = 2 C / c S(E/2^j)/(cE_j)\le(CE/2^j)/(cE/2^{j+1})=2C/c S ( E / 2 j ) / ( c E j ) ≤ ( C E / 2 j ) / ( c E / 2 j + 1 ) = 2 C / c exceptions per block. □
Lemma 5.1 (Correct counting). For q ∈ C r , a ′ q\in C_{r,a'} q ∈ C r , a ′ and E ≥ 1 E\ge1 E ≥ 1 ,
N q ( E ) : = # { e ≤ E : q 3 ∣ b e } ≤ E D 3 ( q ) + 1 { e 0 ( q ) ≤ E } ≤ E ( 1 D 3 ( q ) + 1 e 0 ( q ) ) . N_q(E):=\#\{e\le E:q^3\mid b_e\}\le\frac{E}{D_3(q)}+\mathbf{1}_{\{e_0(q)\le E\}}\le E\left(\frac{1}{D_3(q)}+\frac{1}{e_0(q)}\right). N q ( E ) := # { e ≤ E : q 3 ∣ b e } ≤ D 3 ( q ) E + 1 { e 0 ( q ) ≤ E } ≤ E ( D 3 ( q ) 1 + e 0 ( q ) 1 ) .
Proof. The solution set of q 3 ∣ r e − a ′ q^3\mid r^e-a' q 3 ∣ r e − a ′ is the single class e ≡ e 0 ( m o d D 3 ) e\equiv e_0\pmod{D_3} e ≡ e 0 ( mod D 3 ) with e 0 ≤ D 3 e_0\le D_3 e 0 ≤ D 3 , so N q ( E ) = ⌊ ( E − e 0 ) / D 3 ⌋ + 1 N_q(E)=\lfloor(E-e_0)/D_3\rfloor+1 N q ( E ) = ⌊( E − e 0 ) / D 3 ⌋ + 1 if e 0 ≤ E e_0\le E e 0 ≤ E and 0 0 0 otherwise; the second inequality uses 1 { e 0 ≤ E } ≤ E / e 0 \mathbf{1}_{\{e_0\le E\}}\le E/e_0 1 { e 0 ≤ E } ≤ E / e 0 . □
Proof of Theorem 1.3. By Theorem 1.2, N q ( E ) > 0 N_q(E)>0 N q ( E ) > 0 forces q ∈ W r ∩ W a ′ q\in W_r\cap W_{a'} q ∈ W r ∩ W a ′ with a ′ ∈ ⟨ r ⟩ ( m o d q 3 ) a'\in\langle r\rangle\pmod{q^3} a ′ ∈ ⟨ r ⟩ ( mod q 3 ) , i.e. q ∈ C r , a ′ q\in C_{r,a'} q ∈ C r , a ′ ; hence
S ( E ) = ∑ q ∈ C r , a ′ s q log q N q ( E ) ≤ E ∑ q ∈ C r , a ′ s q log q ( 1 D 3 ( q ) + 1 e 0 ( q ) ) = E K r , a ′ . S(E)=\sum_{q\in C_{r,a'}}s_q\log q\,N_q(E)\le E\sum_{q\in C_{r,a'}}s_q\log q\left(\frac{1}{D_3(q)}+\frac{1}{e_0(q)}\right)=E K_{r,a'}. S ( E ) = q ∈ C r , a ′ ∑ s q log q N q ( E ) ≤ E q ∈ C r , a ′ ∑ s q log q ( D 3 ( q ) 1 + e 0 ( q ) 1 ) = E K r , a ′ .
For the density bound, the moving threshold is handled directly: with c = log r / 20 c=\log r/20 c = log r /20 ,
# { e ≤ E : T s ( e ) ≥ c e } ≤ 1 c ∑ e ≤ E T s ( e ) e , ∑ e ≤ E T s ( e ) e = ∑ q ∈ C r , a ′ s q log q ∑ e ≤ E q 3 ∣ b e 1 e ≤ ( 1 + ln E ) K r , a ′ , \#\{e\le E:T_s(e)\ge ce\}\le\frac{1}{c}\sum_{e\le E}\frac{T_s(e)}{e},\qquad
\sum_{e\le E}\frac{T_s(e)}{e}=\sum_{q\in C_{r,a'}}s_q\log q\sum_{\substack{e\le E\\q^3\mid b_e}}\frac{1}{e}\le(1+\ln E)K_{r,a'}, # { e ≤ E : T s ( e ) ≥ ce } ≤ c 1 e ≤ E ∑ e T s ( e ) , e ≤ E ∑ e T s ( e ) = q ∈ C r , a ′ ∑ s q log q e ≤ E q 3 ∣ b e ∑ e 1 ≤ ( 1 + ln E ) K r , a ′ ,
the inner sum over the class e ≡ e 0 ( m o d D 3 ) e\equiv e_0\pmod{D_3} e ≡ e 0 ( mod D 3 ) being at most ( 1 + ln E ) ( 1 / D 3 ( q ) + 1 / e 0 ( q ) ) (1+\ln E)(1/D_3(q)+1/e_0(q)) ( 1 + ln E ) ( 1/ D 3 ( q ) + 1/ e 0 ( q )) ; dividing by c E cE c E gives the claim. The argument never uses D 3 ≤ E D_3\le E D 3 ≤ E : the “+1” term is absorbed by 1 / e 0 1/e_0 1/ e 0 . □
Remark 5.2 (K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ : sufficient, not necessary). The condition K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ is the most quotable member of a family of sufficient conditions, not the exact requirement: the threshold of the first-moment route is the strictly weaker (M) ∑ e ≤ E T s ( e ) / e = o ( E ) \sum_{e\le E}T_s(e)/e=o(E) ∑ e ≤ E T s ( e ) / e = o ( E ) (§9), which allows Ξ r ( 3 ) = ∞ \Xi_r^{(3)}=\infty Ξ r ( 3 ) = ∞ .
Proposition 5.3 (Splitting and sufficient conditions). Write Ξ r = ∑ q ∈ W r s q log q D 3 ( q ) = Ξ r ( 2 ) + Ξ r ( 3 ) \Xi_r=\sum_{q\in W_r}\frac{s_q\log q}{D_3(q)}=\Xi_r^{(2)}+\Xi_r^{(3)} Ξ r = ∑ q ∈ W r D 3 ( q ) s q l o g q = Ξ r ( 2 ) + Ξ r ( 3 ) . Then Ξ r ( 2 ) \Xi_r^{(2)} Ξ r ( 2 ) is finite unconditionally (Theorem 7.5 ), and Ξ r ( 3 ) = ∑ q ∈ W r ( 3 ) s q log q d q ≤ ( log r ) # W r ( 3 ) \Xi_r^{(3)}=\sum_{q\in W_r^{(3)}}\frac{s_q\log q}{d_q}\le(\log r)\#W_r^{(3)} Ξ r ( 3 ) = ∑ q ∈ W r ( 3 ) d q s q l o g q ≤ ( log r ) # W r ( 3 ) by (2.1). Hence P 6 a ⇒ Ξ r < ∞ P_{6a}\Rightarrow\Xi_r<\infty P 6 a ⇒ Ξ r < ∞ ; the precise squeeze around Ξ r ( 3 ) \Xi_r^{(3)} Ξ r ( 3 ) and the barriers to its unconditional proof are given in §8. If a ′ = r k a'=r^k a ′ = r k then b e = r k ( r e − k − 1 ) b_e=r^k(r^{e-k}-1) b e = r k ( r e − k − 1 ) and T s ( a ′ ) ( e ) = T s ( 1 ) ( e − k ) T_s^{(a')}(e)=T_s^{(1)}(e-k) T s ( a ′ ) ( e ) = T s ( 1 ) ( e − k ) ; compatibility a ′ ∈ ⟨ r ⟩ ( m o d q 3 ) a'\in\langle r\rangle\pmod{q^3} a ′ ∈ ⟨ r ⟩ ( mod q 3 ) is automatic, and q ∈ W r ⇒ q ∈ W r k q\in W_r\Rightarrow q\in W_{r^k} q ∈ W r ⇒ q ∈ W r k by q q ( r k ) ≡ k q q ( r ) ( m o d q ) q_q(r^k)\equiv kq_q(r)\pmod q q q ( r k ) ≡ k q q ( r ) ( mod q ) , so C r , r k = W r C_{r,r^k}=W_r C r , r k = W r . Moreover e 0 ( q ) = k + D 3 ( q ) ≥ D 3 ( q ) e_0(q)=k+D_3(q)\ge D_3(q) e 0 ( q ) = k + D 3 ( q ) ≥ D 3 ( q ) , so K r , r k ≤ 2 Ξ r K_{r,r^k}\le2\Xi_r K r , r k ≤ 2 Ξ r : Case A is closed unconditionally on (H-j). If a ′ ∉ r Z a'\notin r^{\mathbb Z} a ′ ∈ / r Z then e 0 ≤ D 3 e_0\le D_3 e 0 ≤ D 3 and K r , a ′ ≤ 2 J r , a ′ ′ K_{r,a'}\le2J'_{r,a'} K r , a ′ ≤ 2 J r , a ′ ′ where J r , a ′ ′ = ∑ q ∈ C r , a ′ s q log q e 0 ( q ) J'_{r,a'}=\sum_{q\in C_{r,a'}}\frac{s_q\log q}{e_0(q)} J r , a ′ ′ = ∑ q ∈ C r , a ′ e 0 ( q ) s q l o g q ; in particular Ξ r , a ′ : = ∑ q ∈ C r , a ′ s q log q D 3 ( q ) ≤ J r , a ′ ′ \Xi_{r,a'}:=\sum_{q\in C_{r,a'}}\frac{s_q\log q}{D_3(q)}\le J'_{r,a'} Ξ r , a ′ := ∑ q ∈ C r , a ′ D 3 ( q ) s q l o g q ≤ J r , a ′ ′ is a consequence, not an independent hypothesis.
For a ′ = 1 a'=1 a ′ = 1 the count is exact with no boundary term: V q ( e ) = s q + v q ( e / d q ) V_q(e)=s_q+v_q(e/d_q) V q ( e ) = s q + v q ( e / d q ) for d q ∣ e d_q\mid e d q ∣ e , so { e : V q ( e ) ≥ 3 } = D 3 ( q ) Z \{e:V_q(e)\ge3\}=D_3(q)\mathbb Z { e : V q ( e ) ≥ 3 } = D 3 ( q ) Z and S ( E ) = ∑ q ∈ W r s q log q ⌊ E / D 3 ( q ) ⌋ ≤ E Ξ r S(E)=\sum_{q\in W_r}s_q\log q\lfloor E/D_3(q)\rfloor\le E\Xi_r S ( E ) = ∑ q ∈ W r s q log q ⌊ E / D 3 ( q )⌋ ≤ E Ξ r .
The bad set: proof of Theorem 1.4
For the layer s q = 2 s_q=2 s q = 2 we have D 3 = d q q D_3=d_qq D 3 = d q q . By Theorem 1.2 , e 0 ≡ b q ( m o d q ) e_0\equiv b_q\pmod q e 0 ≡ b q ( mod q ) with b q = β q q − 1 ( m o d q ) ∈ [ 0 , q ) b_q=\beta_qq^{-1}\pmod q\in[0,q) b q = β q q − 1 ( mod q ) ∈ [ 0 , q ) ; writing e 0 = b q + j q q e_0=b_q+j_qq e 0 = b q + j q q gives j q ∈ [ 0 , d q ) j_q\in[0,d_q) j q ∈ [ 0 , d q ) (since e 0 ≤ D 3 = d q q e_0\le D_3=d_qq e 0 ≤ D 3 = d q q ). If j q ≥ 1 j_q\ge1 j q ≥ 1 then e 0 ≥ q e_0\ge q e 0 ≥ q and s q log q e 0 ≤ s q log q q \frac{s_q\log q}{e_0}\le\frac{s_q\log q}{q} e 0 s q l o g q ≤ q s q l o g q ; summing over the layer s q = 2 s_q=2 s q = 2 gives at most 2 Σ r 2\Sigma_r 2 Σ r ; the layer s q ≥ 3 s_q\ge3 s q ≥ 3 is controlled separately by P 6 a P_{6a} P 6 a , via s q log q e 0 ≤ s q log q ≤ d q log r \frac{s_q\log q}{e_0}\le s_q\log q\le d_q\log r e 0 s q l o g q ≤ s q log q ≤ d q log r and the finiteness of W r ( 3 ) W_r^{(3)} W r ( 3 ) . If j q = 0 j_q=0 j q = 0 then e 0 = b q < q e_0=b_q<q e 0 = b q < q with b q ≥ 3 log q log r − O ( 1 ) b_q\ge\frac{3\log q}{\log r}-O(1) b q ≥ l o g r 3 l o g q − O ( 1 ) (Theorem 1.2 ). The witness ( r , a ′ , q ) = ( 17 , 19 , 3 ) (r,a',q)=(17,19,3) ( r , a ′ , q ) = ( 17 , 19 , 3 ) : 17 2 ≡ 19 ( m o d 27 ) 17^2\equiv19\pmod{27} 1 7 2 ≡ 19 ( mod 27 ) gives e 0 = 2 < 3 e_0=2<3 e 0 = 2 < 3 , j q = 0 j_q=0 j q = 0 , contributing 2 log 3 > 2 log 3 3 2\log3>\frac{2\log3}{3} 2 log 3 > 3 2 l o g 3 . For the layerwise finiteness, e 0 ( q ) = b e_0(q)=b e 0 ( q ) = b forces q 3 ∣ r b − a ′ q^3\mid r^b-a' q 3 ∣ r b − a ′ with q > b q>b q > b , and q q q is a cube-divisor prime of the fixed integer r b − a ′ r^b-a' r b − a ′ ; each layer is finite and enumerable. □
Proposition 6.1 (Closing conditions). (H-j) follows from the pair
(SC1) # B r , a ′ ( x ) ≪ ( log x ) A \#B_{r,a'}(x)\ll(\log x)^A # B r , a ′ ( x ) ≪ ( log x ) A for some A < 1 + ε A<1+\varepsilon A < 1 + ε , where ε > 0 \varepsilon>0 ε > 0 is as in (SC2);
(SC2) e 0 ( q ) ≥ ( log q ) 1 + ε e_0(q)\ge(\log q)^{1+\varepsilon} e 0 ( q ) ≥ ( log q ) 1 + ε for all but finitely many q ∈ B r , a ′ q\in B_{r,a'} q ∈ B r , a ′ ;
with A A A and ε \varepsilon ε linked: Abel summation gives ∑ q ∈ B ( x ) ( log q ) − 1 − ε ≪ ∫ log x u A − 2 − ε d u \sum_{q\in B(x)}(\log q)^{-1-\varepsilon}\ll\int^{\log x}u^{A-2-\varepsilon}\,du ∑ q ∈ B ( x ) ( log q ) − 1 − ε ≪ ∫ l o g x u A − 2 − ε d u , which converges if and only if A < 1 + ε A<1+\varepsilon A < 1 + ε . In particular the cleanest sufficient form is # B r , a ′ ( x ) ≪ log x \#B_{r,a'}(x)\ll\log x # B r , a ′ ( x ) ≪ log x with any ε > 0 \varepsilon>0 ε > 0 . Note (SC1) implies a P 2 P_2 P 2 -type bound for N r ( x ) N_r(x) N r ( x ) (the bad set is a subset of W r W_r W r ); it is an intersection-strengthened version of P 2 P_2 P 2 , not a weakening.
Remark 6.2 (Gallagher skeleton). One has # B r , a ′ ( x ) = ∣ M ( x ) ∣ \#B_{r,a'}(x)=|\mathcal M(x)| # B r , a ′ ( x ) = ∣ M ( x ) ∣ in the layer s q = 2 s_q=2 s q = 2 (each q q q contributes at most one b ∈ [ 1 , q ) b\in[1,q) b ∈ [ 1 , q ) since D 3 = d q q > q D_3=d_qq>q D 3 = d q q > q ). Expanding the Wieferich indicator as 1 W r ( q ) = 1 q ∑ t m o d q e ( t q r ( q ) q ) \mathbf1_{W_r}(q)=\frac1q\sum_{t\bmod q}e\left(\frac{tq_r(q)}q\right) 1 W r ( q ) = q 1 ∑ t mod q e ( q t q r ( q ) ) splits ∑ b # { Q b ∩ W r } \sum_b\#\{Q_b\cap W_r\} ∑ b # { Q b ∩ W r } into a main term ∑ b 1 q ∣ Q b ∣ \sum_b\frac1q|Q_b| ∑ b q 1 ∣ Q b ∣ (whose random expectation is ≍ ∑ q q − 4 \asymp\sum_q q^{-4} ≍ ∑ q q − 4 ) and a remainder of varying-modulus Fermat-quotient character sums. A mean-value hypothesis
∑ q ≤ x ∣ ∑ q ′ ∈ Q b e ( t q r ( q ′ ) q ′ ) ∣ ≪ x θ ( θ < 1 ) , \sum_{q\le x}\left|\sum_{q'\in Q_b}e\left(\frac{tq_r(q')}{q'}\right)\right|\ll x^\theta\qquad(\theta<1), q ≤ x ∑ q ′ ∈ Q b ∑ e ( q ′ t q r ( q ′ ) ) ≪ x θ ( θ < 1 ) ,
uniform in the truncated range b ≤ B 0 = 3 log x / log r b\le B_0=3\log x/\log r b ≤ B 0 = 3 log x / log r . The main term is unconditionally O ( log x ⋅ log log x ) O(\log x\cdot\log\log x) O ( log x ⋅ log log x ) : ∑ b ≤ B 0 ∑ q ∈ Q b q − 1 ≤ ∑ b ≤ B 0 ∣ Q b ∣ / b ≤ B 0 log r log 8 ∑ b ≤ B 0 1 b ≪ log x ⋅ log log x \sum_{b\le B_0}\sum_{q\in Q_b}q^{-1}\le\sum_{b\le B_0}|Q_b|/b\le\frac{B_0\log r}{\log8}\sum_{b\le B_0}\frac1b\ll\log x\cdot\log\log x ∑ b ≤ B 0 ∑ q ∈ Q b q − 1 ≤ ∑ b ≤ B 0 ∣ Q b ∣/ b ≤ l o g 8 B 0 l o g r ∑ b ≤ B 0 b 1 ≪ log x ⋅ log log x . Hence (FQM-avg) yields the weighted bound ∑ q ∈ B ( x ) q − 1 = O ( log x ) \sum_{q\in B(x)}q^{-1}=O(\log x) ∑ q ∈ B ( x ) q − 1 = O ( log x ) (not the unweighted (SC1)); and the weighted bound combined with (SC2) closes (H-j), since ∑ B ( x ) ( log q ) − 1 − ε ≪ B 1 ( x ) ( log x ) − 1 − ε + ∫ x B 1 ( t ) d t t ( log t ) 2 + ε \sum_{B(x)}(\log q)^{-1-\varepsilon}\ll B_1(x)(\log x)^{-1-\varepsilon}+\int^x B_1(t)\frac{dt}{t(\log t)^{2+\varepsilon}} ∑ B ( x ) ( log q ) − 1 − ε ≪ B 1 ( x ) ( log x ) − 1 − ε + ∫ x B 1 ( t ) t ( l o g t ) 2 + ε d t with B 1 ( t ) = ∑ q ∈ B ( t ) q − 1 = O ( log t ) B_1(t)=\sum_{q\in B(t)}q^{-1}=O(\log t) B 1 ( t ) = ∑ q ∈ B ( t ) q − 1 = O ( log t ) converges for every ε > 0 \varepsilon> 0 ε > 0 . The small-b b b range b ≤ B 0 b \le B_0 b ≤ B 0 is covered by the trivial O ( ( log x ) 2 ) O((\log x)^2) O (( log x ) 2 ) bound. Fixed-modulus Fermat-quotient sums are understood (Heath–Brown, Shparlinski); (FQM-avg) is the varying-modulus input, at the same depth as Katz’s conjecture.
The unconditional layer: Ξ r ( 2 ) < ∞ \Xi_r^{(2)} < \infty Ξ r ( 2 ) < ∞
Split the governing weight by depth:
Ξ r = Ξ r ( 2 ) + Ξ r ( 3 ) , Ξ r ( 2 ) = ∑ q ∈ W r s q = 2 2 log q d q , Ξ r ( 3 ) = ∑ q ∈ W r s q ≥ 3 s q log q d q . \Xi_r=\Xi_r^{(2)}+\Xi_r^{(3)},\qquad\Xi_r^{(2)}=\sum_{\substack{q\in W_r\\s_q=2}}\frac{2\log q}{d_q},\qquad\Xi_r^{(3)}=\sum_{\substack{q\in W_r\\s_q\ge3}}\frac{s_q\log q}{d_q}. Ξ r = Ξ r ( 2 ) + Ξ r ( 3 ) , Ξ r ( 2 ) = q ∈ W r s q = 2 ∑ d q 2 log q , Ξ r ( 3 ) = q ∈ W r s q ≥ 3 ∑ d q s q log q .
Block decomposition
7.1. Block decomposition. For each integer d ≥ 1 d\ge1 d ≥ 1 set S d = { q ∈ W r : d q = d , s q = 2 , q odd } S_d=\{q\in W_r:d_q=d,\ s_q=2,\ q\text{ odd}\} S d = { q ∈ W r : d q = d , s q = 2 , q odd } ; the prime q = 2 q=2 q = 2 is listed separately and contributes a finite O ( log r ) O(\log r) O ( log r ) term.
Lemma 7.1 (LTE identification). For q ∈ S d q\in S_d q ∈ S d we have d ∣ q − 1 d\mid q-1 d ∣ q − 1 and
v q ( r q − 1 − 1 ) = v q ( r d − 1 ) + v q ( q − 1 d ) = v q ( r d − 1 ) , v_q(r^{q-1}-1)=v_q(r^d-1)+v_q\left(\frac{q-1}{d}\right)=v_q(r^d-1), v q ( r q − 1 − 1 ) = v q ( r d − 1 ) + v q ( d q − 1 ) = v q ( r d − 1 ) , since 0 < ( q − 1 ) / d < q 0<(q-1)/d<q 0 < ( q − 1 ) / d < q . Hence s q = 2 ⟺ q 2 ∥ r d − 1 s_q=2\Longleftrightarrow q^2\parallel r^d-1 s q = 2 ⟺ q 2 ∥ r d − 1 .
Lemma 7.2 (Unconditional mass constraint). The primes q ∈ S d q\in S_d q ∈ S d are pairwise coprime divisors with q 2 ∣ r d − 1 q^2\mid r^d-1 q 2 ∣ r d − 1 , so
( ∏ q ∈ S d q ) 2 ∣ r d − 1 ⟹ ∑ q ∈ S d log q < d log r 2 . \left(\prod_{q\in S_d}q\right)^2\mid r^d-1\quad\Longrightarrow\quad\sum_{q\in S_d}\log q<\frac{d\log r}{2}. q ∈ S d ∏ q 2 ∣ r d − 1 ⟹ q ∈ S d ∑ log q < 2 d log r . Also d ∣ q − 1 d\mid q-1 d ∣ q − 1 gives q i ≥ i d + 1 q_i\ge id+1 q i ≥ i d + 1 for the i i i -th element, hence
∣ S d ∣ log ( d + 1 ) ≤ ∑ q ∈ S d log q < d log r 2 ⟹ ∣ S d ∣ < K d : = d log r 2 log ( d + 1 ) . |S_d|\log(d+1)\le\sum_{q\in S_d}\log q<\frac{d\log r}{2}\quad\Longrightarrow\quad|S_d|<K_d:=\frac{d\log r}{2\log(d+1)}. ∣ S d ∣ log ( d + 1 ) ≤ q ∈ S d ∑ log q < 2 d log r ⟹ ∣ S d ∣ < K d := 2 log ( d + 1 ) d log r . With A ( t ) = # { q ∈ S d : q ≤ t } A(t)=\#\{q\in S_d:q\le t\} A ( t ) = # { q ∈ S d : q ≤ t } : A ( t ) ≤ min ( K d , t / d ) A(t)\le\min(K_d,t/d) A ( t ) ≤ min ( K d , t / d ) .
Lemma 7.3 (Stieltjes partial summation). For f ( t ) = log t / t f(t)=\log t/t f ( t ) = log t / t (t ≥ 3 t\ge3 t ≥ 3 , positive, strictly decreasing, − f ′ ( t ) = log t − 1 t 2 ≤ log t t 2 -f'(t)=\frac{\log t-1}{t^2}\le\frac{\log t}{t^2} − f ′ ( t ) = t 2 l o g t − 1 ≤ t 2 l o g t ), the boundary terms vanish (A ( t ) ≤ K d A(t)\le K_d A ( t ) ≤ K d , f ( t ) → 0 f(t)\to0 f ( t ) → 0 ) and
∑ q ∈ S d log q q = ∫ d + 1 ∞ f d A = − ∫ d + 1 ∞ A ( t ) f ′ ( t ) d t ≤ ∫ d ∞ min ( K d , t d ) log t t 2 d t = : I d . \sum_{q\in S_d}\frac{\log q}{q}=\int_{d+1}^{\infty}f\,dA=-\int_{d+1}^{\infty}A(t)f'(t)\,dt\le\int_d^\infty\min\left(K_d,\frac{t}{d}\right)\frac{\log t}{t^2}\,dt=:I_d. q ∈ S d ∑ q log q = ∫ d + 1 ∞ f d A = − ∫ d + 1 ∞ A ( t ) f ′ ( t ) d t ≤ ∫ d ∞ min ( K d , d t ) t 2 log t d t =: I d .
Lemma 7.4 (Explicit I d I_d I d ). With t ∗ = K d d = d 2 log r 2 log ( d + 1 ) t_\ast=K_dd=\frac{d^2\log r}{2\log(d+1)} t ∗ = K d d = 2 l o g ( d + 1 ) d 2 l o g r : if t ∗ ≤ d t_\ast\le d t ∗ ≤ d then I d ≤ K d log d + 1 d I_d\le K_d\frac{\log d+1}{d} I d ≤ K d d l o g d + 1 ; otherwise
I d ≤ 1 d ∫ d t ∗ log t t d t + K d ∫ t ∗ ∞ log t t 2 d t = ( log t ∗ ) 2 − ( log d ) 2 2 d + log t ∗ + 1 d . I_d\le\frac{1}{d}\int_d^{t_\ast}\frac{\log t}{t}\,dt+K_d\int_{t_\ast}^{\infty}\frac{\log t}{t^2}\,dt=\frac{(\log t_\ast)^2-(\log d)^2}{2d}+\frac{\log t_\ast+1}{d}. I d ≤ d 1 ∫ d t ∗ t log t d t + K d ∫ t ∗ ∞ t 2 log t d t = 2 d ( log t ∗ ) 2 − ( log d ) 2 + d log t ∗ + 1 . Since log t ∗ = 2 log d + O r ( 1 ) \log t_\ast=2\log d+O_r(1) log t ∗ = 2 log d + O r ( 1 ) , in all cases
I d ≤ 3 ( log d ) 2 2 d + O r ( log d d ) , I_d\le\frac{3(\log d)^2}{2d}+O_r\left(\frac{\log d}{d}\right), I d ≤ 2 d 3 ( log d ) 2 + O r ( d log d ) , and d = 1 d=1 d = 1 contributes a finite O r ( 1 ) O_r(1) O r ( 1 ) constant.
Theorem 7.5 (Main theorem, unconditional). For every fixed r ≥ 2 r\ge2 r ≥ 2 ,
Ξ r ( 2 ) = ∑ d ≥ 1 2 d ∑ q ∈ S d log q q ≤ ∑ d ≥ 1 2 I d d ≤ ∑ d ≥ 1 ( 3 ( log d ) 2 d 2 + O r ( log d d 2 ) ) < ∞ \Xi_r^{(2)}=\sum_{d\ge1}\frac{2}{d}\sum_{q\in S_d}\frac{\log q}{q}\le\sum_{d\ge1}\frac{2I_d}{d}\le\sum_{d\ge1}\left(\frac{3(\log d)^2}{d^2}+O_r\left(\frac{\log d}{d^2}\right)\right)<\infty Ξ r ( 2 ) = d ≥ 1 ∑ d 2 q ∈ S d ∑ q log q ≤ d ≥ 1 ∑ d 2 I d ≤ d ≥ 1 ∑ ( d 2 3 ( log d ) 2 + O r ( d 2 log d ) ) < ∞
since ∑ d ≥ 2 ( log d ) 2 / d 2 = ∣ ζ ′ ′ ( 2 ) ∣ + O ( 1 ) < ∞ \sum_{d\ge2}(\log d)^2/d^2=|\zeta''(2)|+O(1)<\infty ∑ d ≥ 2 ( log d ) 2 / d 2 = ∣ ζ ′′ ( 2 ) ∣ + O ( 1 ) < ∞ . No abc, no GRH, no Artin, no Katz, and no count, density, or sparsity of W r \mathcal{W}_r W r : the set W r \mathcal{W}_r W r may even be infinite or of positive density.
Remark 7.6 (Ancestry and priority). The mass constraint of Lemma 7.2 is a Wieferich-depth refinement of the classical engine of Erdős–Murty [18 ] (reproduced in Moree’s survey [27 ] ): primes with ord q ( r ) = d \operatorname{ord}_q(r)=d ord q ( r ) = d divide r d − 1 r^d-1 r d − 1 , so their product divides r d − 1 r^d-1 r d − 1 ; this is how Erdős–Murty bound the density of small-order primes. Here the depth condition s q = 2 s_q=2 s q = 2 upgrades the divisibility to q 2 ∣ r d − 1 q^2\mid r^d-1 q 2 ∣ r d − 1 (Lemma 7.1), and that single extra factor of q q q is what converts an open density question into a convergent sum (cf. Theorem 8.3). We are not aware of an unconditional convergence result for any ord-weighted reciprocal sum over the Wieferich primes in the literature; the closest object is ∑ p 1 / ord p 2 ( 2 ) \sum_p 1/\operatorname{ord}_{p^2}(2) ∑ p 1/ ord p 2 ( 2 ) of Granville–Soundararajan [26 ] , whose convergence is open even under their weaker conjectures — the weight 1 / ord 1/\operatorname{ord} 1/ ord there lacks the saving factor log p / p \log p/p log p / p that the Hensel wall supplies here.
Corollary 7.7 (Explicit bound and numerics). Ξ r ( 2 ) ≤ ∑ d ≥ 1 2 I d d = O ( 1 + log log r ) \Xi_r^{(2)}\le\sum_{d\ge1}\frac{2I_d}{d}=O(1+\log\log r) Ξ r ( 2 ) ≤ ∑ d ≥ 1 d 2 I d = O ( 1 + log log r ) ; summing the block bounds to d ≤ 2 × 10 5 d\le2\times10^5 d ≤ 2 × 1 0 5 gives 5.02, 7.13, 9.37, 10.65 for r = 2 , 3 , 5 , 7 r=2,3,5,7 r = 2 , 3 , 5 , 7 , while the true values (q < 10 6 q<10^6 q < 1 0 6 ) are 0.00003782, 0.08719619, 0.00000011, 0.16094379. A softer Cauchy proof: min ( K d , t / d ) ≤ K d t / d \min(K_d,t/d)\le\sqrt{K_dt/d} min ( K d , t / d ) ≤ K d t / d gives Ξ r ( 2 ) ≤ 2 2 log r log d / d 3 / 2 \Xi_r^{(2)}\le2\sqrt{2\log r}\sqrt{\log d}/d^{3/2} Ξ r ( 2 ) ≤ 2 2 log r log d / d 3/2 and ∑ d d − 3 / 2 log d < ∞ \sum_d d^{-3/2}\sqrt{\log d}<\infty ∑ d d − 3/2 log d < ∞ .
Self-consistency: why this does not contradict the openness of P 4 P_4 P 4
The same block controls three weights with three fates:
Table 2.
Numerically, at D = 10 2 / 10 3 / 10 4 / 10 5 D=10^2/10^3/10^4/10^5 D = 1 0 2 /1 0 3 /1 0 4 /1 0 5 the ∑ Θ d \sum\Theta_d ∑ Θ d upper bounds are 37.4/120.5/288.1/573.3, while the ∑ d Ξ d ( 2 ) \sum_d\Xi_d^{(2)} ∑ d Ξ d ( 2 ) bounds plateau at 4.85/5.40/5.52/5.54. Since Ξ q ( 2 ) = 2 log q d q q ≤ log q q \Xi_q^{(2)}=\frac{2\log q}{d_q q}\le\frac{\log q}{q} Ξ q ( 2 ) = d q q 2 l o g q ≤ q l o g q , the theorem proves a smaller quantity than P 4 P_4 P 4 — but precisely the one the application layer needs. A second consistency point: ∣ S d ∣ ≤ K d ≍ d / log d |S_d|\le K_d\asymp d/\log d ∣ S d ∣ ≤ K d ≍ d / log d may grow linearly in d d d , so ∑ d ∣ S d ∣ \sum_d|S_d| ∑ d ∣ S d ∣ may diverge (W r \mathcal{W}_r W r infinite or even of positive density) while ∑ d Ξ d ( 2 ) \sum_d\Xi_d^{(2)} ∑ d Ξ d ( 2 ) converges. Open counting and convergent weighting do not contradict each other.
The remaining layer: Ξ r ( 3 ) \Xi_ r^{(3)} Ξ r ( 3 )
Write P 4 ( 3 ) P_4^{(3)} P 4 ( 3 ) for the assertion ∑ q ∈ W r ( 3 ) log q q < ∞ \sum_{q\in\mathcal{W}_r^{(3)}}\frac{\log q}{q}<\infty ∑ q ∈ W r ( 3 ) q l o g q < ∞ (P 4 P_4 P 4 restricted to the deep layer; a weakening of P 4 P_4 P 4 ).
Theorem 8.1 (Three-layer squeeze, both ends strict). Pointwise 3 log q q − 1 ≤ Ξ q ( 3 ) : = s q log q d q ≤ log r q \frac{3\log q}{q-1}\le\Xi_q^{(3)}:=\frac{s_q\log q}{d_q}\le\frac{\log r}{q} q − 1 3 l o g q ≤ Ξ q ( 3 ) := d q s q l o g q ≤ q l o g r (U2), and
P 6 a ⟹ Ξ r ( 3 ) < ∞ ⟹ P 4 ( 3 ) , P_{6a}\Longrightarrow\Xi_r^{(3)}<\infty\Longrightarrow P_4^{(3)}, P 6 a ⟹ Ξ r ( 3 ) < ∞ ⟹ P 4 ( 3 ) ,
with both implications strict.
Proof. The left implication is Ξ r ( 3 ) ≤ ( log r ) # W r ( 3 ) \Xi_r^{(3)}\le(\log r)\#\mathcal{W}_r^{(3)} Ξ r ( 3 ) ≤ ( log r ) # W r ( 3 ) by (2.1); the right one is d q ≤ q − 1 d_q\le q-1 d q ≤ q − 1 with s q ≥ 3 s_q\ge3 s q ≥ 3 . Left strict: take W ( 3 ) \mathcal{W}^{(3)} W ( 3 ) infinite, d q = q − 1 d_q=q-1 d q = q − 1 , s = 3 s=3 s = 3 , and q j q_j q j so sparse that ∑ log q j / q j < ∞ \sum\log q_j/q_j<\infty ∑ log q j / q j < ∞ ; then Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ while P 6 a P_{6a} P 6 a fails. Right strict: take log q j = 2 j log 2 \log q_j=2^j\log2 log q j = 2 j log 2 (the P 4 ( 3 ) P_4^{(3)} P 4 ( 3 ) partial sums converge to 0.54169...), d j = 3 log q j log r d_j = \frac{3\log q_j}{\log r} d j = l o g r 3 l o g q j (the minimal order allowed by q 3 ≤ r d q^3 \le r^d q 3 ≤ r d , cf. (U1)), s = 3 s = 3 s = 3 ; then Ξ q j ( 3 ) = log r \Xi^{(3)}_{q_j} = \log r Ξ q j ( 3 ) = log r constantly and Ξ r ( 3 ) ∼ j log r → ∞ \Xi^{(3)}_r \sim j\log r \to\infty Ξ r ( 3 ) ∼ j log r → ∞ (j = 24 j = 24 j = 24 already gives 26.37). □
Thus Ξ r ( 3 ) < ∞ \Xi^{(3)}_r < \infty Ξ r ( 3 ) < ∞ sits strictly between P 6 a P_{6a} P 6 a and P 4 ( 3 ) P^{(3)}_4 P 4 ( 3 ) : it requires both sparsity of W r ( 3 ) \mathcal{W}^{(3)}_r W r ( 3 ) and a lower bound on d q d_q d q .
Remark 8.2 (The counter-models sit on the equality case). The structural lower bound q s q ∣ Φ d q ( r ) ∣ ≤ ( r + 1 ) d q q^{s_q}\lvert\Phi_{d_q}(r)\rvert\le(r+1)^{d_q} q s q ∣ Φ d q ( r )∣ ≤ ( r + 1 ) d q gives d q ≥ s q log q log ( r + 1 ) ≥ 3 log q log ( r + 1 ) d_q \ge\frac{s_q\log q}{\log(r+1)} \ge\frac{3\log q}{\log(r+1)} d q ≥ l o g ( r + 1 ) s q l o g q ≥ l o g ( r + 1 ) 3 l o g q on W r ( 3 ) \mathcal{W}^{(3)}_r W r ( 3 ) ; the right-strictness model of Theorem 8.1 takes d j = 3 log q j log r d_j = \frac{3\log q_j}{\log r} d j = l o g r 3 l o g q j , i.e. the minimal order allowed by q 3 ≤ r d q^3 \le r^d q 3 ≤ r d (cf. (U1)). The counter-models are therefore the pointwise equality cases of the mass constraint, not arbitrary configurations.
Proposition 8.3 (Structural reading: the Hensel dividend is spent once). The exact and unique dividing line between Theorem 7.5 and the unconditionality barrier for Ξ r ( 3 ) \Xi^{(3)}_r Ξ r ( 3 ) is the power of q q q in the order-lifting formula (Theorem 1.1):
The extra factor q q q from the Hensel wall is the only object supplying unconditional decay in the whole chain; the s ≥ 3 s \ge3 s ≥ 3 layer has spent it (q max ( 0 , 3 − s ) = 1 q^{\max(0,3-s)} = 1 q m a x ( 0 , 3 − s ) = 1 ). Any further progress must import counting/distribution information about W r ( 3 ) \mathcal{W}^{(3)}_r W r ( 3 ) itself, not a cleverer block decomposition.
Theorem 8.4 (A weaker sufficient condition (S)). If there exist a slowly growing f f f and a constant S S S with
d q ≥ q f ( q ) , s q ≤ S , ∑ q ∈ W r ( 3 ) f ( q ) log q q < ∞ on W r ( 3 ) , d_q \ge\frac{q}{f(q)}, \qquad s_q \le S, \qquad\sum_{q\in\mathcal{W}^{(3)}_r}\frac{f(q)\log q}{q}<\infty\quad\text{on }\mathcal{W}^{(3)}_r, d q ≥ f ( q ) q , s q ≤ S , q ∈ W r ( 3 ) ∑ q f ( q ) log q < ∞ on W r ( 3 ) , then Ξ q ( 3 ) = s q q d q ⋅ log q q ≤ S f ( q ) log q q \Xi^{(3)}_q = \frac{s_q q}{d_q}\cdot\frac{\log q}{q} \le S f(q)\frac{\log q}{q} Ξ q ( 3 ) = d q s q q ⋅ q l o g q ≤ S f ( q ) q l o g q , so Ξ r ( 3 ) < ∞ \Xi^{(3)}_r < \infty Ξ r ( 3 ) < ∞ . Numerically f = 1 , log q , ( log q ) 2 f = 1,\log q,(\log q)^2 f = 1 , log q , ( log q ) 2 give the P 4 ( 3 ) P^{(3)}_4 P 4 ( 3 ) -type sums 0.542, 1.083, 2.685 (all convergent). But d q ≥ q / f ( q ) d_q \ge q/f(q) d q ≥ q / f ( q ) is not automatic for a sparse W r ( 3 ) \mathcal{W}^{(3)}_r W r ( 3 ) : the exceptional set in Erdős–Murty [18 ] is o ( π ( x ) ) o(\pi(x)) o ( π ( x )) and may contain W r ( 3 ) \mathcal{W}^{(3)}_r W r ( 3 ) entirely. Hence (S) hangs on an “order × Wieferich” correlation, and is not unconditional progress.
Proposition 8.5 (Three barriers and an identity). Three independent routes to an unconditional Ξ r ( 3 ) < ∞ \Xi^{(3)}_r < \infty Ξ r ( 3 ) < ∞ are closed; a fourth (the radical channel) is not a barrier at all — it collapses into the threshold itself (Theorem 8.6).
(1) Per-d d d constraints are insufficient. The block constraints ∑ q : d q = d s q log q ≤ d log r \sum_{q:d_q=d}s_q\log q \le d\log r ∑ q : d q = d s q log q ≤ d log r (from q s q ∣ r d − 1 q^{s_q}\mid r^d-1 q s q ∣ r d − 1 , Lemma 7.1), q ≡ 1 ( m o d d ) q\equiv1\pmod d q ≡ 1 ( mod d ) , and partial summation give only Ξ d ( 3 ) ≤ log r \Xi^{(3)}_d\le\log r Ξ d ( 3 ) ≤ log r per block. There exist pseudo-configurations meeting all constraints with Ξ ( 3 ) ( d ≤ 2 × 10 5 ) = 219549 ∼ log r log D \Xi^{(3)}(d\le2\times10^5)=219549\sim\log r\log D Ξ ( 3 ) ( d ≤ 2 × 1 0 5 ) = 219549 ∼ log r log D divergent (each block still ≤ log 3 = 1.0986 \le\log3=1.0986 ≤ log 3 = 1.0986 ). This is a method-level impossibility, not a claim about the true factor sets of r d − 1 r^d-1 r d − 1 .
(2) The radical channel is not an independent barrier (erratum). With Δ d = log ( r d − 1 ) − log rad ( r d − 1 ) \Delta_d=\log(r^d-1)-\log\operatorname{rad}(r^d-1) Δ d = log ( r d − 1 ) − log rad ( r d − 1 ) : since log \log log is additive but log rad \log\operatorname{rad} log rad is not,
Δ d = ∑ c ∣ d Δ c Φ + ∑ p : d p ∣ d v p ( d / d p ) log p , Δ c Φ = ∑ p : d p = c ( s p − 1 ) log p (3) \Delta_d=\sum_{c\mid d}\Delta^\Phi_c+\sum_{p:d_p\mid d}v_p(d/d_p)\log p,\qquad\Delta^\Phi_c=\sum_{p:d_p=c}(s_p-1)\log p \tag*{(3)} Δ d = c ∣ d ∑ Δ c Φ + p : d p ∣ d ∑ v p ( d / d p ) log p , Δ c Φ = p : d p = c ∑ ( s p − 1 ) log p (3)
for odd p p p (the prime 2 contributes a bounded LTE correction). The primes contributing to BOOK d \operatorname{BOOK}_d BOOK d divide d d d , whereas q ∈ S d ( 3 ) q \in S_d^{(3)} q ∈ S d ( 3 ) forces q ≡ 1 ( m o d d ) q \equiv1 \pmod{d} q ≡ 1 ( mod d ) , hence q ≥ d + 1 q \ge d+1 q ≥ d + 1 : the two sets are disjoint, and the “unconditional divergence” ∑ d ≤ D Δ d / d ≥ c log D \sum_{d\le D}\Delta_d/d \ge c\log D ∑ d ≤ D Δ d / d ≥ c log D claimed in an earlier version is produced entirely by BOOK d \operatorname{BOOK}_d BOOK d — in fact ∑ d ≤ D BOOK d / d = ( S o r d + o ( 1 ) ) log D \sum_{d\le D}\operatorname{BOOK}_d/d = (S_{\rm ord}+o(1))\log D ∑ d ≤ D BOOK d / d = ( S ord + o ( 1 )) log D (Theorem 8.6) — carrying no information about W r ( 3 ) W_r^{(3)} W r ( 3 ) . The correct layer bound uses only the s p ≥ 3 s_p\ge3 s p ≥ 3 part of PRIM d \operatorname{PRIM}_d PRIM d :
Ξ d ( 3 ) ≤ 3 2 d ∑ c ∣ d Δ c Φ , ( 3 ) , Δ c Φ , ( 3 ) = ∑ p : d p = c s p ≥ 3 ( s p − 1 ) log p , \Xi_d^{(3)} \le\frac{3}{2d}\sum_{c\mid d}\Delta_c^{\Phi,(3)}, \qquad\Delta_c^{\Phi,(3)}=\sum_{\substack{p:\,dp=c\\s_p\ge3}}(s_p-1)\log p, Ξ d ( 3 ) ≤ 2 d 3 c ∣ d ∑ Δ c Φ , ( 3 ) , Δ c Φ , ( 3 ) = p : d p = c s p ≥ 3 ∑ ( s p − 1 ) log p ,
since ( s p − 1 ) ≥ 2 3 s p (s_p-1)\ge\frac{2}{3}s_p ( s p − 1 ) ≥ 3 2 s p for s p ≥ 3 s_p\ge3 s p ≥ 3 ; charging the full PRIM d \operatorname{PRIM}_d PRIM d would re-introduce the s q = 2 s_q=2 s q = 2 layer, whose harmonic weight R 2 ( D ) = ∑ q ∈ W r ( 2 ) log q d q H ⌊ D / d q ⌋ R_2(D)=\sum_{q\in W_r^{(2)}}\frac{\log q}{d_q}H_{\lfloor D/d_q\rfloor} R 2 ( D ) = ∑ q ∈ W r ( 2 ) d q l o g q H ⌊ D / d q ⌋ is O ( D ) O(D) O ( D ) , not O ( log D ) O(\log D) O ( log D ) (numerically 3.345 3.345 3.345 against BOOK = 3.091 \operatorname{BOOK}=3.091 BOOK = 3.091 at r = 3 r=3 r = 3 , D = 3000 D=3000 D = 3000 ). Summed, the radical route is the identity of Theorem 8.6: the radical channel is the threshold Φ r ( D ) = o ( D ) \Phi_r(D)=o(D) Φ r ( D ) = o ( D ) of §9, up to the s q = 2 s_q=2 s q = 2 remainder R 2 ( D ) R_2(D) R 2 ( D ) — neither closed nor independently open.
(3) External radical inputs are all insufficient. The abc-type input [22 ] gives Δ d / d ≤ ( 1 − φ ( d ) d + ε ) log r = O ( 1 ) \Delta_d/d\le\left(1-\frac{\varphi(d)}{d}+\varepsilon\right)\log r=O(1) Δ d / d ≤ ( 1 − d φ ( d ) + ε ) log r = O ( 1 ) ; Stewart–Yu [23 ] gives unconditionally log c ≤ κ R 1 / 3 ( log R ) 3 \log c\le\kappa R^{1/3}(\log R)^3 log c ≤ κ R 1/3 ( log R ) 3 , i.e. only log rad = O ( log d ) \log\operatorname{rad}=O(\log d) log rad = O ( log d ) ; Stewart [24 ] gives P ( a n − b n ) > n exp ( log n 10 4 log log n ) P(a^n-b^n)>n\exp\left(\frac{\log n}{10^4\log\log n}\right) P ( a n − b n ) > n exp ( 1 0 4 l o g l o g n l o g n ) — the same; Balaji–Luca [25 ] , Theorem 1.2, give the closest unconditional sparsity, # A K , c , u ∩ [ 1 , N ] ≪ N exp ( − log N 156 log log N ) \#\mathcal{A}_{K,c,u}\cap[1,N]\ll N\exp\left(-\frac{\log N}{156\log\log N}\right) # A K , c , u ∩ [ 1 , N ] ≪ N exp ( − 156 l o g l o g N l o g N ) . All four inputs control Δ d \Delta_d Δ d (including BOOK d \operatorname{BOOK}_d BOOK d ) and yield only Δ d / d = O ( 1 ) \Delta_d/d=O(1) Δ d / d = O ( 1 ) , i.e. Φ r ( D ) = O ( D ) \Phi_r(D)=O(D) Φ r ( D ) = O ( D ) ; the Chen–Ding/Graves–Murty bound C n ′ ≫ a φ ( n ) − ε n C'_n\gg a^{\varphi(n)-\varepsilon n} C n ′ ≫ a φ ( n ) − ε n controls the squarefree part of Φ n \Phi_n Φ n and gives PRIM d ≤ ε d log r \operatorname{PRIM}_d\le\varepsilon d\log r PRIM d ≤ ε d log r , still merely Φ r ( D ) = O ( ε D ) \Phi_r(D)=O(\varepsilon D) Φ r ( D ) = O ( ε D ) . None reaches o ( D ) o(D) o ( D ) .
(4) The DeKoninck–Doyon dichotomy points the other way [21 ] : its other branch is “small radical ⇒ W \Rightarrow W ⇒ W infinite”.
Conclusion: a family of sufficient conditions remains, of which the most easily quoted is
Ξ r ( 2 ) < ∞ + Ξ r ( 3 ) < ∞ , \Xi_r^{(2)}<\infty\qquad+ \qquad\Xi_r^{(3)}<\infty, Ξ r ( 2 ) < ∞ + Ξ r ( 3 ) < ∞ ,
with Theorem 7.5, unconditional under the first condition, and P 6 a P_{6a} P 6 a sufficient; not necessary under the second condition, and the exact threshold of the unconditional question, together with strictly weaker rungs Ξ r ( 3 ) ( E ) = o ( E / log E ) \Xi_r^{(3)}(E)=o(E/\log E) Ξ r ( 3 ) ( E ) = o ( E / log E ) , the block-counting condition (C)), is determined in §9.
Theorem 8.6 (Radical identity). For every r ≥ 2 r\ge2 r ≥ 2 and D ≥ 2 D\ge2 D ≥ 2 , with Δ d = log ( r d − 1 ) − log rad ( r d − 1 ) \Delta_d=\log(r^d-1)-\log\operatorname{rad}(r^d-1) Δ d = log ( r d − 1 ) − log rad ( r d − 1 ) ,
∑ d ≤ D Δ d d = ∑ q ∈ W r ( 3 ) ( s q − 1 ) log q d q H ⌊ D / d q ⌋ + R 2 ( D ) + ( S o r d + o ( 1 ) ) log D , (4) \sum_{d\le D}\frac{\Delta_d}{d}
=
\sum_{q\in W_r^{(3)}}\frac{(s_q-1)\log q}{d_q}H_{\lfloor D/d_q\rfloor}
+R_2(D)+(S_{\rm ord}+o(1))\log D, \tag*{(4)} d ≤ D ∑ d Δ d = q ∈ W r ( 3 ) ∑ d q ( s q − 1 ) log q H ⌊ D / d q ⌋ + R 2 ( D ) + ( S ord + o ( 1 )) log D , (4) where R 2 ( D ) : = ∑ q ∈ W r ( 2 ) log q d q H ⌊ D / d q ⌋ ≤ log r 2 ∑ c ≤ D H ⌊ D / c ⌋ = O ( D ) R_2(D):=\sum_{q\in W_r^{(2)}}\frac{\log q}{d_q}H_{\lfloor D/d_q\rfloor}\le\frac{\log r}{2}\sum_{c\le D}H_{\lfloor D/c\rfloor}=O(D) R 2 ( D ) := ∑ q ∈ W r ( 2 ) d q l o g q H ⌊ D / d q ⌋ ≤ 2 l o g r ∑ c ≤ D H ⌊ D / c ⌋ = O ( D ) by Lemma 7.2, and R 2 ( D ) = O ( log D ) R_2(D)=O(\log D) R 2 ( D ) = O ( log D ) whenever W r ( 2 ) W_r^{(2)} W r ( 2 ) is finite (in particular for every base examined). Hence
2 3 Φ r ( D ) ≤ ∑ d ≤ D Δ d d ≤ Φ r ( D ) + R 2 ( D ) + O ( log D ) , \frac{2}{3}\Phi_r(D)\le\sum_{d\le D}\frac{\Delta_d}{d}\le\Phi_r(D)+R_2(D)+O(\log D), 3 2 Φ r ( D ) ≤ d ≤ D ∑ d Δ d ≤ Φ r ( D ) + R 2 ( D ) + O ( log D ) , so ∑ d ≤ D Δ d / d = o ( D ) ⇒ Φ r ( D ) = o ( D ) \sum_{d\le D}\Delta_d/d=o(D)\Rightarrow\Phi_r(D)=o(D) ∑ d ≤ D Δ d / d = o ( D ) ⇒ Φ r ( D ) = o ( D ) unconditionally, and if R 2 ( D ) = o ( D ) R_2(D)=o(D) R 2 ( D ) = o ( D ) — in particular if W r ( 2 ) W_r^{(2)} W r ( 2 ) is finite — then Φ r ( D ) = o ( D ) ⟺ ∑ d ≤ D Δ d / d = o ( D ) \Phi_r(D)=o(D)\Longleftrightarrow\sum_{d\le D}\Delta_d/d=o(D) Φ r ( D ) = o ( D ) ⟺ ∑ d ≤ D Δ d / d = o ( D ) .
Proof. By (3) , swapping the order of summation,
∑ d ≤ D PRIM d d = ∑ c ≤ D Δ c Φ c H ⌊ D / c ⌋ = ∑ p ∈ W r ( s p − 1 ) log p d p H ⌊ D / d p ⌋ \sum_{d\le D}\frac{\operatorname{PRIM}_{d}}{d}=\sum_{c\le D}\frac{\Delta_{c}^{\Phi}}{c}H_{\lfloor D/c\rfloor}=\sum_{p\in\mathcal{W}_{r}}\frac{(s_{p}-1)\log p}{d_{p}}H_{\lfloor D/d_{p}\rfloor} d ≤ D ∑ d PRIM d = c ≤ D ∑ c Δ c Φ H ⌊ D / c ⌋ = p ∈ W r ∑ d p ( s p − 1 ) log p H ⌊ D / d p ⌋
(primes with s p = 1 s_p=1 s p = 1 contribute 0), and, since only primes p ≤ D p\le D p ≤ D can divide a d ≤ D d\le D d ≤ D with d p ∣ d d_p\mid d d p ∣ d ,
∑ d ≤ D BOOK d d = ∑ p ≤ D log p d p ∑ m ≤ D / d p v p ( m ) m = ( S o r d + o ( 1 ) ) log D , \sum_{d\le D}\frac{\operatorname{BOOK}_{d}}{d}=\sum_{p\le D}\frac{\log p}{d_{p}}\sum_{m\le D/d_{p}}\frac{v_{p}(m)}{m}=(S_{\mathrm{ord}}+o(1))\log D, d ≤ D ∑ d BOOK d = p ≤ D ∑ d p log p m ≤ D / d p ∑ m v p ( m ) = ( S ord + o ( 1 )) log D ,
using ∑ m ≤ M v p ( m ) / m = log M / ( p − 1 ) + O p ( 1 ) \sum_{m\le M}v_p(m)/m=\log M/(p-1)+O_p(1) ∑ m ≤ M v p ( m ) / m = log M / ( p − 1 ) + O p ( 1 ) and the convergence of S o r d = ∑ p log p d p ( p − 1 ) S_{\mathrm{ord}}=\sum_p\frac{\log p}{d_p(p-1)} S ord = ∑ p d p ( p − 1 ) l o g p (Theorem 11.1 ). Splitting the primitive sum by depth: the coefficients ( s q − 1 ) / s q ∈ [ 2 / 3 , 1 ) (s_q-1)/s_q\in[2/3,1) ( s q − 1 ) / s q ∈ [ 2/3 , 1 ) on W r ( 3 ) \mathcal{W}_{r}^{(3)} W r ( 3 ) give the [ 2 / 3 , 1 ) ⋅ Φ r ( D ) [2/3,1)\cdot\Phi_r(D) [ 2/3 , 1 ) ⋅ Φ r ( D ) part, and the remaining s q = 2 s_q=2 s q = 2 part is R 2 ( D ) R_2(D) R 2 ( D ) , bounded by O ( D ) O(D) O ( D ) via Lemma 7.2 exactly as in Theorem 9.2 .
Remark 8.7 (Numerics). For r = 3 r=3 r = 3 , d = 20 d=20 d = 20 : Δ 20 = 6.0868 = PRIM 3.0910 + BOOK 2.9957 \Delta_{20}=6.0868=\operatorname{PRIM}\ 3.0910+\operatorname{BOOK}\ 2.9957 Δ 20 = 6.0868 = PRIM 3.0910 + BOOK 2.9957 (3 20 − 1 = 2 4 ⋅ 5 2 ⋅ 11 2 ⋅ 61 ⋅ 1181 3^{20}-1=2^4\cdot5^2\cdot11^2\cdot61\cdot1181 3 20 − 1 = 2 4 ⋅ 5 2 ⋅ 1 1 2 ⋅ 61 ⋅ 1181 ; the only Wieferich contribution is Δ 5 Φ = log 11 \Delta_{5}^{\Phi}=\log11 Δ 5 Φ = log 11 ). The bookkeeping part grows like S o r d log D S_{\mathrm{ord}}\log D S ord log D (BOOK = 3.058 / 4.113 / 5.340 / 6.491 =3.058/4.113/5.340/6.491 = 3.058/4.113/5.340/6.491 at D = 10 3 / 3 ⋅ 10 3 / 10 4 / 3 ⋅ 10 4 D=10^3/3\cdot10^3/10^4/3\cdot10^4 D = 1 0 3 /3 ⋅ 1 0 3 /1 0 4 /3 ⋅ 1 0 4 ), while the layer-specific bound Ξ d ( 3 ) ≤ 3 2 d ∑ c ∣ d Δ c Φ , ( 3 ) \Xi_{d}^{(3)}\le\frac{3}{2d}\sum_{c\mid d}\Delta_{c}^{\Phi,(3)} Ξ d ( 3 ) ≤ 2 d 3 ∑ c ∣ d Δ c Φ , ( 3 ) keeps the s q = 2 s_q=2 s q = 2 layer out: charging PRIM d \operatorname{PRIM}_d PRIM d in full would add log 11 5 H ⌊ D / 5 ⌋ = 3.345 \frac{\log11}{5}H_{\lfloor D/5\rfloor}=3.345 5 l o g 11 H ⌊ D /5 ⌋ = 3.345 at D = 3000 D=3000 D = 3000 — an O ( log D ) O(\log D) O ( log D ) , not O ( D ) O(D) O ( D ) , term, but one that must not be charged to W r ( 3 ) \mathcal{W}_{r}^{(3)} W r ( 3 ) .
An almost-sure route: the double-zero hypothesis (D4)
Definition 8.8 ((D4)). There is a constant C r > 0 C_r>0 C r > 0 such that for every odd prime q ∤ r q\nmid r q ∤ r ,
P ( s q ≥ 3 ) = P ( q 3 ∣ r q − 1 − 1 ) = P ( Q q ( 1 ) = 0 , Q q ( 2 ) = 0 ) ≤ C r q 2 . \mathbb{P}(s_q\ge3)=\mathbb{P}(q^3\mid r^{q-1}-1)=\mathbb{P}(Q_q^{(1)}=0,Q_q^{(2)}=0)\le\frac{C_r}{q^2}. P ( s q ≥ 3 ) = P ( q 3 ∣ r q − 1 − 1 ) = P ( Q q ( 1 ) = 0 , Q q ( 2 ) = 0 ) ≤ q 2 C r .
A pointwise bound for the Katz quotient at the double-zero scale. (D4) is not a consequence of Katz’s 2015 conjecture (continuous R / Z \mathbb{R}/\mathbb{Z} R / Z equidistribution): attempting to place ( Q ( 1 ) , Q ( 2 ) ) (Q^{(1)},Q^{(2)}) ( Q ( 1 ) , Q ( 2 ) ) in the group-scheme framework G = G m 2 G=\mathbb{G}_m^2 G = G m 2 , the cyclic subgroup generated by a single base lies in a proper subtorus and is likely excluded by the Zariski-density hypothesis of Katz’s Lemma 4.2 (isomorphic to the a ′ = r k a'=r^k a ′ = r k degeneration of ( W ‾ r , W ‾ a ′ ) (\overline{W}_r,\overline{W}_{a'}) ( W r , W a ′ ) ). Hence (D4) must be listed as an independent hypothesis, and it must be distinguished from (D2): (D2) is “two bases × \times × first quotient” modulo q 2 q^2 q 2 joint, (D4) is “one base × \times × first and second quotients”; both give 1 / q 2 1/q^2 1/ q 2 , but (D2) does not imply (D4).
Ensemble separation (why (D4) is not free). Let P = ∏ q ≥ 3 ( Z / q 3 Z ) × \mathcal{P}=\prod_{q\ge3}(\mathbb{Z}/q^3\mathbb{Z})^\times P = ∏ q ≥ 3 ( Z / q 3 Z ) × carry its product Haar measure ν = ⨂ q ν q \nu=\bigotimes_q\nu_q ν = ⨂ q ν q , and let T q = μ q − 1 ( Z / q 3 Z ) T_q=\mu_{q-1}(\mathbb{Z}/q^3\mathbb{Z}) T q = μ q − 1 ( Z / q 3 Z ) . Then ν q ( T q ) = ∣ μ q − 1 ∣ / ∣ ( Z / q 3 ) × ∣ = ( q − 1 ) / ( q 2 ( q − 1 ) ) = 1 / q 2 \nu_q(T_q)=|\mu_{q-1}|/|(\mathbb{Z}/q^3)^\times|=(q-1)/(q^2(q-1))=1/q^2 ν q ( T q ) = ∣ μ q − 1 ∣/∣ ( Z / q 3 ) × ∣ = ( q − 1 ) / ( q 2 ( q − 1 )) = 1/ q 2 exactly , and with the independent indicators 1 T q \mathbf{1}_{T_q} 1 T q
λ = ∑ q ≥ 3 q − 2 = P ( 2 ) − 1 4 = 0.2022475 … , ν ( K = 0 ) = ∏ q ≥ 3 ( 1 − q − 2 ) = 8 π 2 = 0.8105695 … , \lambda=\sum_{q\ge3}q^{-2}=P(2)-\frac14=0.2022475\ldots,\qquad\nu(K=0)=\prod_{q\ge3}(1-q^{-2})=\frac{8}{\pi^2}=0.8105695\ldots, λ = q ≥ 3 ∑ q − 2 = P ( 2 ) − 4 1 = 0.2022475 … , ν ( K = 0 ) = q ≥ 3 ∏ ( 1 − q − 2 ) = π 2 8 = 0.8105695 … ,
where K = # { q : x q ∈ T q } K=\#\{q:x_q\in T_q\} K = # { q : x q ∈ T q } ; the exact law of K K K is Poisson-binomial (the Poisson value e − λ = 0.8168927 e^{-\lambda}=0.8168927 e − λ = 0.8168927 is only an approximation, off by 0.78 % 0.78\% 0.78% ), and Borel–Cantelli gives K < ∞ K<\infty K < ∞ ν \nu ν -a.s. Thus in the Haar ensemble the pointwise bound of Definition 8.8 holds with C r = 1 C_r=1 C r = 1 : (D4) is a free theorem of product measure, and Theorem 8.10 becomes an a.s. statement about a ν \nu ν -random element of P \mathcal{P} P .
The primal question, however, concerns the single point
ι ( r ) = ( r m o d q 3 ) q ≥ 3 ∈ P , ι : Z ↪ P , \iota(r)=(r\bmod q^3)_{q\ge3}\in\mathcal{P},\qquad\iota:\mathbb{Z}\hookrightarrow\mathcal{P}, ι ( r ) = ( r mod q 3 ) q ≥ 3 ∈ P , ι : Z ↪ P ,
the diagonal embedding of the integers. Its image is countable in an uncountable compact group, hence ν ( ι ( Z ) ) = 0 \nu(\iota(\mathbb{Z}))=0 ν ( ι ( Z )) = 0 . Every ν \nu ν -a.s. statement is therefore void on ι ( r ) \iota(r) ι ( r ) , and no product-measure argument can yield a statement about a fixed integer r r r . This resolves the apparent tension between the counting-density reading of (D4) (equivalent to P 6 a P_{6a} P 6 a , Theorem 14.4 ) and the probabilistic reading (Definition 8.8 , finite mean, strictly stronger): the two readings live in different ensembles, and
( D 4 ) H a a r ⇏ ( D 4 ) d i a g . (\mathrm{D4})_{\mathrm{Haar}}\not\Rightarrow(\mathrm{D4})_{\mathrm{diag}}. ( D4 ) Haar ⇒ ( D4 ) diag .
What is missing is precisely a pointwise rigidity of ι ( r ) \iota(r) ι ( r ) relative to ⨂ q ν q \bigotimes_q\nu_q ⨂ q ν q : the assertion that a fixed integer has no systematic preference among the index-q 2 q^2 q 2 Teichmüller subgroups T q T_q T q as q q q varies. This is the exact content of specification S3 (and of the dual→ \to → primal transfer of Proposition 14.5 ); it is neither a consequence of Katz’s continuous R / Z \mathbb{R}/\mathbb{Z} R / Z equidistribution nor of the free Haar statement above.
Proposition 8.9 (Teichmüller anchor). For odd q ∤ r q\nmid r q ∤ r ,
q ∈ W r ( 3 ) ⟺ q 3 ∣ r q 2 − r ⟺ r m o d q 3 ∈ μ q − 1 ( ( Z / q 3 Z ) × ) , q\in\mathcal{W}^{(3)}_r\Longleftrightarrow q^3\mid r^{q^2}-r\Longleftrightarrow r\bmod q^3\in\mu_{q-1}\left((\mathbb{Z}/q^3\mathbb{Z})^\times\right), q ∈ W r ( 3 ) ⟺ q 3 ∣ r q 2 − r ⟺ r mod q 3 ∈ μ q − 1 ( ( Z / q 3 Z ) × ) , where ( Z / q 3 Z ) × ≅ C q − 1 × C q 2 (\mathbb{Z}/q^3\mathbb{Z})^\times\cong C_{q-1}\times C_{q^2} ( Z / q 3 Z ) × ≅ C q − 1 × C q 2 and ∣ μ q − 1 ∣ = q − 1 |\mu_{q-1}|=q-1 ∣ μ q − 1 ∣ = q − 1 : the subgroup has relative size exactly 1 / q 2 1/q^2 1/ q 2 .
Proof. Write r = ω u r=\omega u r = ω u with ω q − 1 = 1 \omega^{q-1}=1 ω q − 1 = 1 (the Teichmüller part) and u ∈ 1 + q Z q u\in1+q\mathbb{Z}_q u ∈ 1 + q Z q (the principal unit). Then r q 2 = ω q 2 u q 2 ≡ ω ( m o d q 3 ) r^{q^2}=\omega^{q^2}u^{q^2}\equiv\omega\pmod{q^3} r q 2 = ω q 2 u q 2 ≡ ω ( mod q 3 ) , since q 2 ≡ 1 ( m o d q − 1 ) q^2\equiv1\pmod{q-1} q 2 ≡ 1 ( mod q − 1 ) and ( 1 + q x ) q 2 ≡ 1 ( m o d q 3 ) (1+qx)^{q^2}\equiv1\pmod{q^3} ( 1 + q x ) q 2 ≡ 1 ( mod q 3 ) . Hence r q 2 ≡ r = ω u ( m o d q 3 ) r^{q^2}\equiv r=\omega u\pmod{q^3} r q 2 ≡ r = ω u ( mod q 3 ) iff u ≡ 1 ( m o d q 3 ) u\equiv1\pmod{q^3} u ≡ 1 ( mod q 3 ) , i.e. r ≡ ω ( m o d q 3 ) r\equiv\omega\pmod{q^3} r ≡ ω ( mod q 3 ) , i.e. ord q 3 ( r ) ∣ q − 1 \operatorname{ord}_{q^3}(r)\mid q-1 ord q 3 ( r ) ∣ q − 1 ; by Theorem 1.1 , ord q 3 ( r ) = d q q max ( 0 , 3 − s q ) \operatorname{ord}_{q^3}(r)=d_q q^{\max(0,3-s_q)} ord q 3 ( r ) = d q q m a x ( 0 , 3 − s q ) divides q − 1 q-1 q − 1 iff s q ≥ 3 s_q\ge3 s q ≥ 3 . □
Layer distinction (mod q 2 q^2 q 2 vs. mod q 3 q^3 q 3 ). The subgroup governing (D4) and Proposition 8.9 is μ q − 1 ( Z / q 3 Z ) \mu_{q-1}(\mathbb{Z}/q^3\mathbb{Z}) μ q − 1 ( Z / q 3 Z ) , corresponding to s q ≥ 3 s_q\ge3 s q ≥ 3 . The cluster { 19 , 22 , 23 } \{19,22,23\} { 19 , 22 , 23 } at q = 13 q=13 q = 13 of § 12.3 lies in μ 12 ( Z / 169 Z ) \mu_{12}(\mathbb{Z}/169\mathbb{Z}) μ 12 ( Z /169 Z ) (mod q 2 q^2 q 2 , the s q ≥ 2 s_q\ge2 s q ≥ 2 Katz/(D2) layer): 19 ∈ μ ( Z / 169 ) 19\in\mu(\mathbb{Z}/169) 19 ∈ μ ( Z /169 ) but 19 ∉ μ ( Z / 2197 ) 19\notin\mu(\mathbb{Z}/2197) 19 ∈ / μ ( Z /2197 ) . These are different subgroups of different groups; there is no contradiction, and the census of § 12.3 tests (D2), not (D4). In the dual ensemble (fixed q q q , r r r uniform in ( Z / q 3 ) × (\mathbb{Z}/q^3)^\times ( Z / q 3 ) × ) one has the exact identity
P ( r q − 1 ≡ 1 ( m o d q 3 ) ) = ∣ μ q − 1 ∣ ∣ ( Z / q 3 ) × ∣ = q − 1 q 2 ( q − 1 ) = 1 q 2 ; \mathbb{P}(r^{q-1}\equiv1\pmod{q^3})=\frac{|\mu_{q-1}|}{|(\mathbb{Z}/q^3)^\times|}=\frac{q-1}{q^2(q-1)}=\frac{1}{q^2}; P ( r q − 1 ≡ 1 ( mod q 3 )) = ∣ ( Z / q 3 ) × ∣ ∣ μ q − 1 ∣ = q 2 ( q − 1 ) q − 1 = q 2 1 ;
in the primal ensemble (fixed r r r , q q q varying) there is no probability space, so the missing tool is precisely the transfer dual→ \to → primal (Proposition 14.5 ).
Consequently (D4) is the purely algebraic assertion that a fixed integer r r r has no systematic preference among the index-q 2 q^2 q 2 Teichmüller subgroups as q q q varies; the constant ∑ q q − 2 = P ( 2 ) = 0.452247514 … \sum_q q^{-2}=P(2)=0.452247514\ldots ∑ q q − 2 = P ( 2 ) = 0.452247514 … is the reciprocal sum of the subgroup indices, not a heuristic input. The Chebotarev obstruction is unchanged: the group varies with q q q . The running example ⟨ 19 ⟩ m o d 343 = { ± 1 , ± 18 , ± 19 } \langle19\rangle\bmod343=\{\pm1,\pm18,\pm19\} ⟨ 19 ⟩ mod 343 = { ± 1 , ± 18 , ± 19 } of § 12 is the image of μ 6 \mu_6 μ 6 . Numerically: zero counterexamples among q < 10 5 × 12 q<10^5\times12 q < 1 0 5 × 12 bases, and ∣ μ q − 1 m o d q ∣ = q − 1 |\mu_{q-1}\bmod q|=q-1 ∣ μ q − 1 mod q ∣ = q − 1 for every q ≤ 97 q\le97 q ≤ 97 enumerated (see the ensemble separation above).
Set X q = s q log q d q 1 s q ≥ 3 ≥ 0 X_q=\frac{s_q\log q}{d_q}\mathbf{1}_{s_q\ge3}\ge0 X q = d q s q l o g q 1 s q ≥ 3 ≥ 0 , so Ξ r ( 3 ) = ∑ q X q \Xi_r^{(3)}=\sum_qX_q Ξ r ( 3 ) = ∑ q X q .
Theorem 8.10 ((D4) ⇒ Ξ r ( 3 ) < ∞ \Rightarrow\Xi_r^{(3)}<\infty ⇒ Ξ r ( 3 ) < ∞ and P 6 a P_{6a} P 6 a , a.s.). Under (D4),
E [ Ξ r ( 3 ) ] ≤ C r ( log r ) P ( 2 ) = 0.452247514 … ⋅ C r log r < ∞ . \mathbb{E}\left[\Xi_r^{(3)}\right]\le C_r(\log r)P(2)=0.452247514\ldots\cdot C_r\log r<\infty. E [ Ξ r ( 3 ) ] ≤ C r ( log r ) P ( 2 ) = 0.452247514 … ⋅ C r log r < ∞. where P ( 2 ) = ∑ q q − 2 P(2)=\sum_q q^{-2} P ( 2 ) = ∑ q q − 2 (over all primes). Hence Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ a.s.; and by the first Borel–Cantelli lemma (no independence needed), # W r ( 3 ) < ∞ \#\mathcal{W}_r^{(3)}<\infty # W r ( 3 ) < ∞ a.s., i.e. P 6 a P_{6a} P 6 a /Murata’s conjecture (i) holds almost surely under (D4).
Proof. By (U2) ((2) ) the pointwise deterministic bound X q ≤ ( log r ) 1 s q ≥ 3 X_q \le(\log r)1_{s_q\geq3} X q ≤ ( log r ) 1 s q ≥ 3 holds, independent of d q d_q d q ; hence E [ X q ] ≤ ( log r ) P ( s q ≥ 3 ) ≤ C r log r / q 2 \mathbb{E}[X_q] \le(\log r)\mathbb{P}(s_q\geq3) \le C_r\log r/q^2 E [ X q ] ≤ ( log r ) P ( s q ≥ 3 ) ≤ C r log r / q 2 . Since X q ≥ 0 X_q\geq0 X q ≥ 0 , Tonelli permits term-by-term expectation, and ∑ q q − 2 < ∞ \sum_q q^{-2}<\infty ∑ q q − 2 < ∞ . Finite expectation implies a.s. finiteness; Borel–Cantelli I needs only ∑ q P ( s q ≥ 3 ) ≤ C r P ( 2 ) < ∞ \sum_q\mathbb{P}(s_q\geq3)\le C_rP(2)<\infty ∑ q P ( s q ≥ 3 ) ≤ C r P ( 2 ) < ∞ and yields # W r ( 3 ) < ∞ \#\mathcal{W}_r^{(3)}<\infty # W r ( 3 ) < ∞ a.s. □ \square □
Remark 8.11. The proof needs no conditioning on d q d_q d q , no k ≥ 4 k\geq4 k ≥ 4 layer hypothesis, no lower bound on orders, and no GRH: (U2) absorbs the ord side entirely, bypassing the “order × \times × Wieferich” correlation of the (D2)/(D3) side. If (D4) is stated in conditional form one gets E [ Ξ r ( 3 ) ] ≤ C r ∑ q log q d q q 2 \mathbb{E}[\Xi_r^{(3)}]\leq C_r\sum_q\frac{\log q}{d_qq^2} E [ Ξ r ( 3 ) ] ≤ C r ∑ q d q q 2 l o g q , whose true values are 0.0346 0.0346 0.0346 –0.1470 0.1470 0.1470 for r = 2 , 3 , 5 , 7 r=2,3,5,7 r = 2 , 3 , 5 , 7 ; the proof itself does not use this refinement.
Scale. (D4) yields E [ # W r ( 3 ) ( x ) ] ≤ C r P ( 2 ) = O ( 1 ) \mathbb{E}[\#\mathcal{W}_r^{(3)}(x)]\leq C_rP(2)=O(1) E [ # W r ( 3 ) ( x )] ≤ C r P ( 2 ) = O ( 1 ) , not ≪ log log x \ll\log\log x ≪ log log x (the latter is the ∑ q 1 / q \sum_q1/q ∑ q 1/ q scale of (D1)). Read as a counting density , # W r ( 3 ) ( x ) ≤ C r ∑ q ≤ x q − 2 = O ( 1 ) \#\mathcal{W}_r^{(3)}(x)\leq C_r\sum_{q\leq x}q^{-2}=O(1) # W r ( 3 ) ( x ) ≤ C r ∑ q ≤ x q − 2 = O ( 1 ) , (D4) is equivalent to P6a (Theorem 14.4 ); it is the probabilistic form of Definition 8.8, with the finite mean ∑ q C r / q 2 \sum_q C_r/q^2 ∑ q C r / q 2 , that is strictly stronger.
(K) alone is insufficient. (D1) gives only O ( 1 / q ) O(1/q) O ( 1/ q ) , and ∑ q < Q 1 / q = 2.198 / 2.483 / 2.705 / 2.887 \sum_{q<Q}1/q=2.198/2.483/2.705/2.887 ∑ q < Q 1/ q = 2.198/2.483/2.705/2.887 at Q = 10 3 , 10 4 , 10 5 , 10 6 Q=10^3,10^4,10^5,10^6 Q = 1 0 3 , 1 0 4 , 1 0 5 , 1 0 6 diverges like log log Q \log\log Q log log Q ; hence “(K) ⇒ Ξ r ( 3 ) < ∞ \Rightarrow\Xi_r^{(3)}<\infty ⇒ Ξ r ( 3 ) < ∞ ” fails, and one must pass to ( K + ) = ( K ) ∪ { ( D 4 ) } (K^+)=(K)\cup\{(D4)\} ( K + ) = ( K ) ∪ {( D 4 )} .
The exact threshold: what an unconditional proof requires
Write δ ( E ) = 1 E # { e ≤ E : e exceptional } \delta(E)=\frac{1}{E}\#\{e\leq E:e\text{ exceptional}\} δ ( E ) = E 1 # { e ≤ E : e exceptional } . Markov with the moving threshold c = log r / 20 c=\log r/20 c = log r /20 gives, unconditionally,
δ ( E ) ≤ 20 E log r ∑ e ≤ E T s ( e ) e , (5) \delta(E)\leq\frac{20}{E\log r}\sum_{e\leq E}\frac{T_s(e)}{e}, \tag*{(5)} δ ( E ) ≤ E log r 20 e ≤ E ∑ e T s ( e ) , (5)
so the exact threshold of the first-moment route is
(M) : ∑ e ≤ E T s ( e ) e = o ( E ) . (6) \text{(M)}:\quad\sum_{e\leq E}\frac{T_s(e)}{e}=o(E). \tag*{(6)} (M) : e ≤ E ∑ e T s ( e ) = o ( E ) . (6)
Condition (M) is sufficient but not necessary for δ ( E ) → 0 \delta(E)\to0 δ ( E ) → 0 (the Markov upper bound may be non-tight: the mass of T s T_s T s can concentrate on few e e e ). We claim only that (M) is the weakest operational sufficient condition of the first-moment route, and it is strictly weaker than every condition used above (Theorem 9.4 ).
Case A: the s q = 2 s_q=2 s q = 2 layer is elementary
Let a ′ = r k a'=r^k a ′ = r k (including a ′ = 1 a'=1 a ′ = 1 ), so b e = r k ( r e − k − 1 ) b_e=r^k(r^{e-k}-1) b e = r k ( r e − k − 1 ) . For q ∤ r q\nmid r q ∤ r , LTE gives, when d q ∣ e − k d_q\mid e-k d q ∣ e − k ,
V q ( e ) = v q ( r e − k − 1 ) = s q + v q ( e − k d q ) ≥ s q , V_q(e)=v_q(r^{e-k}-1)=s_q+v_q\left(\frac{e-k}{d_q}\right)\geq s_q, V q ( e ) = v q ( r e − k − 1 ) = s q + v q ( d q e − k ) ≥ s q ,
and V q ( e ) = 0 V_q(e)=0 V q ( e ) = 0 otherwise; q ∣ e − k ⟺ d q ∣ e − k q\mid e-k\Longleftrightarrow d_q\mid e-k q ∣ e − k ⟺ d q ∣ e − k since d q ∣ q − 1 d_q\mid q-1 d q ∣ q − 1 gives gcd ( d q , q ) = 1 \gcd(d_q,q)=1 g cd( d q , q ) = 1 .
Proposition 9.1 (Pointwise bounds, Case A, unconditional). For e > k e>k e > k ,
T s ( 2 ) ( e ) = ∑ q ∣ e − k q ∈ W r ( 2 ) 2 log q ≤ 2 ∑ q ∣ e − k log q ≤ 2 log ( e − k ) ≤ 2 log e , T_s^{(2)}(e)=\sum_{\substack{q\mid e-k\\q\in\mathcal{W}_r^{(2)}}}2\log q\leq2\sum_{q\mid e-k}\log q\leq2\log(e-k)\leq2\log e, T s ( 2 ) ( e ) = q ∣ e − k q ∈ W r ( 2 ) ∑ 2 log q ≤ 2 q ∣ e − k ∑ log q ≤ 2 log ( e − k ) ≤ 2 log e ,
hence $\sum_{e\leq E}T_s^{(2)}(e)/e\leq2\sum_{e\leq E}\frac{\log e}{e}=O((\log E)^2)=o(E)), unconditionally. Moreover
T s ( e ) ≤ ∑ q 3 ∣ b e V q ( e ) log q ≤ log ( r e − k − 1 ) < e log r , T_s(e)\leq\sum_{q^3\mid b_e}V_q(e)\log q\leq\log(r^{e-k}-1)<e\log r, T s ( e ) ≤ q 3 ∣ b e ∑ V q ( e ) log q ≤ log ( r e − k − 1 ) < e log r ,
so the exception threshold e log r / 20 e\log r/20 e log r /20 sits a factor 20 below the unconditional pointwise bound: density zero needs the average to be far below the bound. Numerically (r = 2 , 3 , 5 , 7 r=2,3,5,7 r = 2 , 3 , 5 , 7 ; W r ( 2 ) \mathcal{W}^{(2)}_r W r ( 2 ) below 10 6 10^6 1 0 6 is { 5 , 11 , 1093 , 3511 , 20771 , 40487 } \{5,11,1093,3511,20771,40487\} { 5 , 11 , 1093 , 3511 , 20771 , 40487 } ): max e ≤ E T s ( 2 ) ( e ) = 8.0147 / 8.0147 / 17.2122 / 22.0080 \max_{e\le E} T_s^{(2)}(e)=8.0147/8.0147/17.2122/22.0080 max e ≤ E T s ( 2 ) ( e ) = 8.0147/8.0147/17.2122/22.0080 at E = 10 2 / 10 3 / 10 4 / 10 5 E=10^2/10^3/10^4/10^5 E = 1 0 2 /1 0 3 /1 0 4 /1 0 5 , against 2 log E = 9.2103 / 13.8155 / 18.4207 / 23.0259 2\log E=9.2103/13.8155/18.4207/23.0259 2 log E = 9.2103/13.8155/18.4207/23.0259 — zero violations.
Thus in Case A the density-zero conclusion for the s q = 2 s_q=2 s q = 2 layer needs neither Theorem 7.5 nor any hypothesis: Proposition 9.1 suffices. Theorem 7.5 remains the stronger result and the required input for Case B (where V q ( e ) ≥ s q V_q(e)\ge s_q V q ( e ) ≥ s q fails and no pointwise bound holds). Note C r , r k = W r C_{r,r_k}=W_r C r , r k = W r (Proposition 5.3 ), so Case A never meets the bad set at all: (H-j) enters only through Case B.
The harmonic threshold
9.2. The harmonic threshold Φ r ( E ) \Phi_r(E) Φ r ( E ) . For q ∈ W r ( 3 ) q\in\mathcal{W}_r^{(3)} q ∈ W r ( 3 ) the solution class of q 3 ∣ b e q^3\mid b_e q 3 ∣ b e is e ≡ e 0 ( m o d d q ) e\equiv e_0\pmod{d_q} e ≡ e 0 ( mod d q ) (wall-free, D 3 = d q D_3=d_q D 3 = d q ), and for any a ′ a' a ′ the class sum satisfies
∑ e ≤ E q 3 ∣ b e 1 e ≤ 1 e 0 ( q ) + H ⌊ E / d q ⌋ d q . \sum_{\substack{e\le E\\q^3\mid b_e}}\frac{1}{e}\le\frac{1}{e_0(q)}+\frac{H_{\lfloor E/d_q\rfloor}}{d_q}. e ≤ E q 3 ∣ b e ∑ e 1 ≤ e 0 ( q ) 1 + d q H ⌊ E / d q ⌋ .
Define
Φ r ( E ) : = ∑ q ∈ W r ( 3 ) d q ≤ E s q log q d q H ⌊ E / d q ⌋ . (7) \Phi_r(E):=\sum_{\substack{q\in\mathcal{W}_r^{(3)}\\d_q\le E}}\frac{s_q\log q}{d_q}H_{\lfloor E/d_q\rfloor}. \tag*{(7)} Φ r ( E ) := q ∈ W r ( 3 ) d q ≤ E ∑ d q s q log q H ⌊ E / d q ⌋ . (7)
Then ∑ e ≤ E T s ( 3 ) ( e ) e ≤ Φ r ( E ) + O ( 1 ) \sum_{e\le E}\frac{T_s^{(3)}(e)}{e}\le\Phi_r(E)+O(1) ∑ e ≤ E e T s ( 3 ) ( e ) ≤ Φ r ( E ) + O ( 1 ) , and together with Proposition 9.1 (Case A; the same bound holds for general a ′ a' a ′ ):
∑ e ≤ E T s ( e ) e ≤ ( log E ) 2 + Φ r ( E ) + O ( log E ) . \sum_{e\le E}\frac{T_s(e)}{e}\le(\log E)^2+\Phi_r(E)+O(\log E). e ≤ E ∑ e T s ( e ) ≤ ( log E ) 2 + Φ r ( E ) + O ( log E ) .
Theorem 9.2 (Unconditional linear cap). Let S d ( 3 ) = { q ∈ W r ( 3 ) : d q = d } S_d^{(3)}=\{q\in\mathcal{W}_r^{(3)}:d_q=d\} S d ( 3 ) = { q ∈ W r ( 3 ) : d q = d } . Since d q = d d_q=d d q = d forces d ∣ q − 1 d\mid q-1 d ∣ q − 1 (hence q ∤ d q\nmid d q ∤ d ) and v q ( r d − 1 ) = s q v_q(r^d-1)=s_q v q ( r d − 1 ) = s q , we have ∏ q ∈ S d ( 3 ) q s q ∣ Φ d ( r ) \prod_{q\in S_d^{(3)}}q^{s_q}\mid\Phi_d(r) ∏ q ∈ S d ( 3 ) q s q ∣ Φ d ( r ) (the d d d -th cyclotomic value), and Φ d ( r ) ≤ ( r + 1 ) φ ( d ) \Phi_d(r)\le(r+1)^{\varphi(d)} Φ d ( r ) ≤ ( r + 1 ) φ ( d ) ; therefore
mass d : = ∑ q ∈ S d ( 3 ) s q log q ≤ φ ( d ) log ( r + 1 ) , \operatorname{mass}_d:=\sum_{q\in S_d^{(3)}}s_q\log q\le\varphi(d)\log(r+1), mass d := q ∈ S d ( 3 ) ∑ s q log q ≤ φ ( d ) log ( r + 1 ) , and, with H m ≤ log m + γ + o ( 1 ) H_m\le\log m+\gamma+o(1) H m ≤ log m + γ + o ( 1 ) uniformly,
Φ r ( E ) ≤ log ( r + 1 ) ∑ d ≤ E φ ( d ) d ( log E d + γ + o ( 1 ) ) = ( 6 π 2 + o ( 1 ) ) ( 1 + γ ) log ( r + 1 ) E . \Phi_r(E)\le\log(r+1)\sum_{d\le E}\frac{\varphi(d)}{d}\left(\log\frac{E}{d}+\gamma+o(1)\right)=\left(\frac{6}{\pi^2}+o(1)\right)(1+\gamma)\log(r+1)E. Φ r ( E ) ≤ log ( r + 1 ) d ≤ E ∑ d φ ( d ) ( log d E + γ + o ( 1 ) ) = ( π 2 6 + o ( 1 ) ) ( 1 + γ ) log ( r + 1 ) E . The constant is C = 6 π 2 ( 1 + γ ) = 0.958832 … C=\frac{6}{\pi^2}(1+\gamma)=0.958832\ldots C = π 2 6 ( 1 + γ ) = 0.958832 … ; the resulting cap 20 C log ( r + 1 ) log r \frac{20C\log(r+1)}{\log r} l o g r 20 C l o g ( r + 1 ) equals 30.39 , 24.20 , 21.35 , 20.49 , 19.87 30.39,24.20,21.35,20.49,19.87 30.39 , 24.20 , 21.35 , 20.49 , 19.87 for r = 2 , 3 , 5 , 7 , 11 r=2,3,5,7,11 r = 2 , 3 , 5 , 7 , 11 (limit 19.18 19.18 19.18 as r → ∞ r\to\infty r → ∞ ) — all trivial, since δ ( E ) ≤ 1 \delta(E)\le1 δ ( E ) ≤ 1 . The same mass bound gives the linear estimate Ξ r ( 3 ) ( E ) : = ∑ d q ≤ E s q log q d q ≤ 6 π 2 log ( r + 1 ) E \Xi_r^{(3)}(E):=\sum_{d_q\le E}\frac{s_q\log q}{d_q}\le\frac{6}{\pi^2}\log(r+1)E Ξ r ( 3 ) ( E ) := ∑ d q ≤ E d q s q l o g q ≤ π 2 6 log ( r + 1 ) E ( 0.668 / 0.843 / 1.089 / 1.264 (0.668/0.843/1.089/1.264 ( 0.668/0.843/1.089/1.264 for r = 2 , 3 , 5 , 7 r=2,3,5,7 r = 2 , 3 , 5 , 7 ).
So the unconditional bound is linear, while the target is o ( E ) o(E) o ( E ) : the required saving is a factor Ω ( log E ) \Omega(\log E) Ω ( log E ) — not a constant factor and not an infinite methodological gulf. It lands precisely on “P1/P2-level counting of W r ( 3 ) \mathcal{W}_r^{(3)} W r ( 3 ) ”, which is the varying-modulus obstruction itself.
Tighter budget and its saturation. For each d d d , ∏ q : d q = d q s q ∣ Φ d ( r ) ≤ ( r + 1 ) φ ( d ) \prod_{q:d_q=d}q^{s_q}\mid\Phi_d(r)\leq(r+1)^{\varphi(d)} ∏ q : d q = d q s q ∣ Φ d ( r ) ≤ ( r + 1 ) φ ( d ) , so m a s s d ≤ φ ( d ) log ( r + 1 ) \mathrm{mass}_d\leq\varphi(d)\log(r+1) mass d ≤ φ ( d ) log ( r + 1 ) , a factor 1.25–2.95 tighter than log ( r d − 1 ) < d log r \log(r^d-1)<d\log r log ( r d − 1 ) < d log r (e.g. r = 19 r=19 r = 19 , d = 6 d=6 d = 6 : φ ( 6 ) log 20 = 5.9915 \varphi(6)\log20=5.9915 φ ( 6 ) log 20 = 5.9915 versus 6 log 19 = 17.67 6\log19=17.67 6 log 19 = 17.67 ). The bound is attained at the observed deep hits: Φ 2 ( 26 ) = 27 = 3 3 \Phi_2(26)=27=3^3 Φ 2 ( 26 ) = 27 = 3 3 , Φ 2 ( 124 ) = 125 = 5 3 \Phi_2(124)=125=5^3 Φ 2 ( 124 ) = 125 = 5 3 and Φ 2 ( 80 ) = 81 = 3 4 \Phi_2(80)=81=3^4 Φ 2 ( 80 ) = 81 = 3 4 give ρ d Φ : = m a s s d / ( φ ( d ) log ( r + 1 ) ) = 1 \rho_d^\Phi:=\mathrm{mass}_d/(\varphi(d)\log(r+1))=1 ρ d Φ := mass d / ( φ ( d ) log ( r + 1 )) = 1 exactly, and Φ 6 ( 19 ) = 343 = 7 3 \Phi_6(19)=343=7^3 Φ 6 ( 19 ) = 343 = 7 3 gives 0.974, Φ 4 ( 239 ) = 2 ⋅ 13 4 \Phi_4(239)=2\cdot13^4 Φ 4 ( 239 ) = 2 ⋅ 1 3 4 gives 0.936. Hence the volume bound carries no cross-q q q slack.
A stratification trap. Collecting primes from q 3 ∣ r d − 1 q^3\mid r^d-1 q 3 ∣ r d − 1 under the weaker condition d q ∣ d d_q\mid d d q ∣ d is unsafe: LTE gives v q ( r d − 1 ) = s q + v q ( d / d q ) v_q(r^d-1)=s_q+v_q(d/d_q) v q ( r d − 1 ) = s q + v q ( d / d q ) , so a classical prime with q ∣ d / d q q\mid d/d_q q ∣ d / d q reaches exponent 3 for free. Concrete false positives: ( r , q , d q , d ) = ( 3 , 11 , 5 , 55 ) (r,q,d_q,d)=(3,11,5,55) ( r , q , d q , d ) = ( 3 , 11 , 5 , 55 ) has v 11 ( 3 55 − 1 ) = 3 v_{11}(3^{55}-1)=3 v 11 ( 3 55 − 1 ) = 3 but s 11 = 2 s_{11}=2 s 11 = 2 ; likewise ( 7 , 5 , 4 , 20 ) (7,5,4,20) ( 7 , 5 , 4 , 20 ) and ( 2 , 1093 , 364 , 397852 ) (2,1093,364,397852) ( 2 , 1093 , 364 , 397852 ) . One must stratify by d q = d d_q=d d q = d exactly and exclude q ∣ d / d q q\mid d/d_q q ∣ d / d q . Equivalently, since d q ∣ q − 1 d_q\mid q-1 d q ∣ q − 1 forces q > d q q>d_q q > d q and q ∤ d q q\nmid d_q q ∤ d q , the bookkeeping contamination requires d ≥ q d q > q 2 d\geq qd_q>q^2 d ≥ q d q > q 2 ; hence for d ≤ E d\leq E d ≤ E and q > E q>\sqrt{E} q > E one has q 3 ∣ r d − 1 ⟺ s q ≥ 3 q^3\mid r^d-1\Longleftrightarrow s_q\geq3 q 3 ∣ r d − 1 ⟺ s q ≥ 3 and d q ∣ d d_q\mid d d q ∣ d , with no contamination.
Theorem 9.4 (The weakest sufficient condition; none of the quoted conditions is necessary). Since H ⌊ E / d ⌋ ≤ log E + γ H_{\lfloor E/d\rfloor}\leq\log E+\gamma H ⌊ E / d ⌋ ≤ log E + γ ,
Ξ r ( 3 ) ( E ) = o ( E log E ) ⟹ Φ r ( E ) = o ( E ) ⟹ δ ( E ) → 0. \Xi_r^{(3)}(E)=o\left(\frac{E}{\log E}\right)\Longrightarrow\Phi_r(E)=o(E)\Longrightarrow\delta(E)\to0. Ξ r ( 3 ) ( E ) = o ( log E E ) ⟹ Φ r ( E ) = o ( E ) ⟹ δ ( E ) → 0. Neither K r , α ′ < ∞ K_{r,\alpha'}<\infty K r , α ′ < ∞ nor Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ is necessary for density zero. Indeed there are configurations meeting every per-block structural constraint with
Ξ d = 1 ( log d ) 2 ( d ≥ 5 ) , Ξ r ( 3 ) ( E ) ∼ E ( log E ) 2 → ∞ , Φ ( E ) = ∑ d ≤ E Ξ d H ⌊ E / d ⌋ = o ( E ) : \Xi_d=\frac{1}{(\log d)^2}\quad(d\geq5),\qquad\Xi_r^{(3)}(E)\sim\frac{E}{(\log E)^2}\to\infty,\qquad\Phi(E)=\sum_{d\leq E}\Xi_dH_{\lfloor E/d\rfloor}=o(E): Ξ d = ( log d ) 2 1 ( d ≥ 5 ) , Ξ r ( 3 ) ( E ) ∼ ( log E ) 2 E → ∞ , Φ ( E ) = d ≤ E ∑ Ξ d H ⌊ E / d ⌋ = o ( E ) : divergent Ξ r ( 3 ) \Xi_r^{(3)} Ξ r ( 3 ) with δ ( E ) → 0 \delta(E)\to0 δ ( E ) → 0 . These are logical models (satisfying all per-block constraints), not claims about the true factor sets of $r^d-1.
The ladder of sufficient conditions, weakest first, all unproved unconditionally:
Table 1.
For orientation: N 3 ( E ) ≤ ∑ d ≤ E φ ( d ) log ( r + 1 ) 3 log ( d + 1 ) ≍ ( log r ) E 2 6 log E N_3(E)\le\sum_{d\le E}\frac{\varphi(d)\log(r+1)}{3\log(d+1)}\asymp\frac{(\log r)E^2}{6\log E} N 3 ( E ) ≤ ∑ d ≤ E 3 l o g ( d + 1 ) φ ( d ) l o g ( r + 1 ) ≍ 6 l o g E ( l o g r ) E 2 is quadratic — useless; a genuine sub-quadratic bound on N 3 ( E ) N_3(E) N 3 ( E ) is itself open. In one sentence: an unconditional proof requires a sublinear bound on the deep-Wieferich mass weighted by 1 / d q 1/d_q 1/ d q — a P1/P2-level statement restricted to W r ( 3 ) W_r^{(3)} W r ( 3 ) and weighted by ord; it is neither manifestly easier nor manifestly harder than P1, but it lives in the same no-fixed-modulus no-man’s-land. Silverman’s a b c abc ab c route gives a non-Wieferich lower bound — the opposite direction; the other branch of [21 ] gives W W W infinite. No unconditional tool currently produces upper bounds of this type.
Theorem 9.5 (Information-theoretic barrier). Let q ∈ W r q\in W_r q ∈ W r , r q − 1 = 1 + q s q u r^{q-1}=1+q^{s_q}u r q − 1 = 1 + q s q u , q ∤ u q\nmid u q ∤ u , and suppose q 3 ∣ r − a ′ q^3\mid r-a' q 3 ∣ r − a ′ . Raising to the ( q − 1 ) (q-1) ( q − 1 ) -st power,
a ′ q − 1 ≡ ( 1 + q s q u ) e ≡ 1 + e q s q u ( m o d q 3 ) , a'^{q-1}\equiv(1+q^{s_q}u)^e\equiv1+e q^{s_q}u\pmod{q^3}, a ′ q − 1 ≡ ( 1 + q s q u ) e ≡ 1 + e q s q u ( mod q 3 ) , the k ≥ 2 k\ge2 k ≥ 2 binomial terms carrying q 2 s q ≥ q 4 q^{2s_q}\ge q^4 q 2 s q ≥ q 4 . For s q = 2 s_q=2 s q = 2 the q 2 q^2 q 2 term survives and e ≡ β q w q − 1 ( m o d q ) e\equiv\beta_qw_q^{-1}\pmod q e ≡ β q w q − 1 ( mod q ) is solvable — this is the second-Fermat-quotient formula of Theorem 1.2 , the entire information content of (H-j)/(K). For s q ≥ 3 s_q\ge3 s q ≥ 3 the term e q s q u ≡ 0 ( m o d q 3 ) e q^{s_q}u\equiv0\pmod{q^3} e q s q u ≡ 0 ( mod q 3 ) : a ′ q − 1 ≡ 1 ( m o d q 3 ) a'^{q-1}\equiv1\pmod{q^3} a ′ q − 1 ≡ 1 ( mod q 3 ) independent of e e e , and no congruence expansion of any finite order recovers e e e ; the least solution e 0 ( q ) e_0(q) e 0 ( q ) is determined only by the discrete-logarithm structure of ⟨ r ⟩ ⊆ ( Z / q 3 Z ) × \langle r\rangle\subseteq(\mathbb{Z}/q^3\mathbb{Z})^\times ⟨ r ⟩ ⊆ ( Z / q 3 Z ) × — a varying-modulus discrete logarithm.
Numerically: four artificial bases r ≡ 1 , 57 , 68 , 124 ( m o d 125 ) r\equiv1,57,68,124\pmod{125} r ≡ 1 , 57 , 68 , 124 ( mod 125 ) with r 4 ≡ 1 ( m o d 125 ) r^4\equiv1\pmod{125} r 4 ≡ 1 ( mod 125 ) (forced s ≥ 3 s\ge3 s ≥ 3 at q = 5 q=5 q = 5 ) give a ′ 4 ≡ 1 ( m o d 125 ) a'^4\equiv1\pmod{125} a ′4 ≡ 1 ( mod 125 ) identically in e = 1 , … , 29 e=1,\ldots,29 e = 1 , … , 29 — no e e e -information; while for the genuine pair ( r , q ) = ( 3 , 11 ) (r,q)=(3,11) ( r , q ) = ( 3 , 11 ) , w q = 4 w_q=4 w q = 4 and a ′ 10 ≡ 1 + 100 e ⋅ 4 ( m o d 1331 ) a'^{10}\equiv1+100e\cdot4\pmod{1331} a ′ 10 ≡ 1 + 100 e ⋅ 4 ( mod 1331 ) predicts e ( m o d 11 ) e\pmod{11} e ( mod 11 ) exactly for e = 1 , 2 , 3 , 5 , 7 e=1,2,3,5,7 e = 1 , 2 , 3 , 5 , 7 . Conclusion: all information about the s ≥ 3 s\ge3 s ≥ 3 layer lives in the ord and counting structure of W r ( 3 ) W_r^{(3)} W r ( 3 ) , not in any congruence expansion — the same obstruction, which now has seven guises: (T1) GRH insufficiency (Theorem 1.5 ); (T2) the radical channel — withdrawn as a barrier and replaced by the identity of Theorem 8.6 ; (T3) the expansion barrier above (Theorem 9.5 ); (T4) no cross-d d d repulsion from divisibility identities — a fixed q q q has a unique d q d_q d q with q ≡ 1 ( m o d d q ) q\equiv1\pmod{d_q} q ≡ 1 ( mod d q ) , so distinct d d d ’s carry disjoint q q q -sets, all multiples e = m d q e=md_q e = m d q are consistent by LTE, and ∏ d ∣ n Φ d ( r ) = r n − 1 \prod_{d\mid n}\Phi_d(r)=r^n-1 ∏ d ∣ n Φ d ( r ) = r n − 1 with n = lcm ( 1.. E ) ∼ e E n=\operatorname{lcm}(1..E)\sim e^E n = lcm ( 1.. E ) ∼ e E is too loose by ∼ 33 \sim33 ∼ 33 orders of magnitude; (T5) no cited theorem controls the cube-full part of Φ n ( a ) \Phi_n(a) Φ n ( a ) (Theorem 14.2 ); (T6) all structural constraints together give zero asymptotic improvement on A r ( t ) A_r(t) A r ( t ) (Theorem 14.2 ); and (T7) the missing dual→ \to → primal transfer of (D4) (Proposition 14.5 ).
Theorem 9.6 (Exact threshold, dyadic form). Put M d = ∑ q : d q = d , s q ≥ 3 s q log q M_d=\sum_{q:dq=d,\ s_q\ge3}s_q\log q M d = ∑ q : d q = d , s q ≥ 3 s q log q and A r ( t ) = ∑ d ≤ t M d d A_r(t)=\sum_{d\le t}\frac{M_d}{d} A r ( t ) = ∑ d ≤ t d M d . Then Φ r ( E ) = o ( E ) \Phi_r(E)=o(E) Φ r ( E ) = o ( E ) if and only if A r ( t ) = o ( t ) A_r(t)=o(t) A r ( t ) = o ( t ) ; precisely
3 2 A r ( E / 2 ) ≤ Φ r ( E ) ≤ ( 1 + log 2 ) A r ( E ) + log 2 ∑ j ≥ 1 A r ( E / 2 j ) . \frac{3}{2}A_r(E/2)\le\Phi_r(E)\le(1+\log2)A_r(E)+\log2\sum_{j\ge1}A_r(E/2^j). 2 3 A r ( E /2 ) ≤ Φ r ( E ) ≤ ( 1 + log 2 ) A r ( E ) + log 2 j ≥ 1 ∑ A r ( E / 2 j ) . so A r ( t ) ≤ ε t A_r(t)\le\varepsilon t A r ( t ) ≤ εt gives Φ r ( E ) ≤ ( 1 + 2 log 2 ) ε E = 2.386294 ε E \Phi_r(E)\le(1+2\log2)\varepsilon E=2.386294\,\varepsilon E Φ r ( E ) ≤ ( 1 + 2 log 2 ) εE = 2.386294 εE . A saving M d = d / log d M_d=d/\log d M d = d / log d is not enough (Φ ( E ) / E ≈ 0.17 \Phi(E)/E\approx0.17 Φ ( E ) / E ≈ 0.17 at E = 10 5 E=10^5 E = 1 0 5 ); one needs M d / d → 0 M_d/d\to0 M d / d → 0 in Cesàro mean. In particular Ξ r ( 3 ) ( E ) = o ( E / log E ) \Xi_r^{(3)}(E)=o(E/\log E) Ξ r ( 3 ) ( E ) = o ( E / log E ) of Theorem 9.4 is a sufficient rung, not the threshold itself.
Proof. Group Φ r ( E ) = ∑ d ≤ E M d d H ⌊ E / d ⌋ \Phi_r(E)=\sum_{d\le E}\frac{M_d}{d}H_{\lfloor E/d\rfloor} Φ r ( E ) = ∑ d ≤ E d M d H ⌊ E / d ⌋ (Theorem 9.2). For d ≤ E / 2 d\le E/2 d ≤ E /2 the floor is ≥ 2 \ge2 ≥ 2 , so H ⌊ E / d ⌋ ≥ H 2 = 3 2 H_{\lfloor E/d\rfloor}\ge H_2=\frac{3}{2} H ⌊ E / d ⌋ ≥ H 2 = 2 3 , giving the lower bound. For the upper bound split [ 1 , E ] [1,E] [ 1 , E ] dyadically: on ( E / 2 j + 1 , E / 2 j ] (E/2^{j+1},E/2^j] ( E / 2 j + 1 , E / 2 j ] one has H ⌊ E / d ⌋ ≤ ( j + 1 ) log 2 + 1 H_{\lfloor E/d\rfloor}\le(j+1)\log2+1 H ⌊ E / d ⌋ ≤ ( j + 1 ) log 2 + 1 , and Abel reordering gives the stated bound. If A r ( t ) ≤ ε t A_r(t)\le\varepsilon t A r ( t ) ≤ εt for all t ≤ E t\le E t ≤ E , then ∑ j ≥ 1 A r ( E / 2 j ) ≤ ε E ∑ j ≥ 1 2 − j = ε E \sum_{j\ge1}A_r(E/2^j)\le\varepsilon E\sum_{j\ge1}2^{-j}=\varepsilon E ∑ j ≥ 1 A r ( E / 2 j ) ≤ εE ∑ j ≥ 1 2 − j = εE . □
Theorem 9.7 (Container; the gap is exactly one o ( 1 ) o(1) o ( 1 ) factor). Let C E = ∏ d ≤ E Φ d ( r ) C_E=\prod_{d\le E}\Phi_d(r) C E = ∏ d ≤ E Φ d ( r ) . Then ∏ q ∈ W r ( 3 ) , d q ≤ E q s q ∣ C E \prod_{q\in\mathcal W_r^{(3)},\ d_q\le E}q^{s_q}\mid C_E ∏ q ∈ W r ( 3 ) , d q ≤ E q s q ∣ C E and
log C E = ∑ d ≤ E ( φ ( d ) log r + O ( 2 ω ( d ) log r ) ) = 3 π 2 E 2 log r ( 1 + o ( 1 ) ) . \log C_E=\sum_{d\le E}\left(\varphi(d)\log r+O(2^{\omega(d)}\log r)\right)=\frac{3}{\pi^2}E^2\log r\left(1+o(1)\right). log C E = d ≤ E ∑ ( φ ( d ) log r + O ( 2 ω ( d ) log r ) ) = π 2 3 E 2 log r ( 1 + o ( 1 ) ) . since ∑ d ≤ E φ ( d ) = 3 π 2 E 2 + O ( E log E ) \sum_{d\le E}\varphi(d)=\frac{3}{\pi^2}E^2+O(E\log E) ∑ d ≤ E φ ( d ) = π 2 3 E 2 + O ( E log E ) and ∑ d ≤ E 2 ω ( d ) = O ( E log E ) \sum_{d\le E}2^{\omega(d)}=O(E\log E) ∑ d ≤ E 2 ω ( d ) = O ( E log E ) . The volume bound gives
A r ( t ) ≤ log ( r + 1 ) ∑ d ≤ t φ ( d ) d = 6 π 2 log ( r + 1 ) t ( 1 + o ( 1 ) ) , A_r(t)\le\log(r+1)\sum_{d\le t}\frac{\varphi(d)}{d}=\frac{6}{\pi^2}\log(r+1)t\left(1+o(1)\right), A r ( t ) ≤ log ( r + 1 ) d ≤ t ∑ d φ ( d ) = π 2 6 log ( r + 1 ) t ( 1 + o ( 1 ) ) , with ∑ d ≤ t φ ( d ) d / ( 6 π 2 t ) = 1.000658 , 1.000761 , 1.000375 , 1.000095 \sum_{d\le t}\frac{\varphi(d)}{d}/(\frac{6}{\pi^2}t)=1.000658,1.000761,1.000375,1.000095 ∑ d ≤ t d φ ( d ) / ( π 2 6 t ) = 1.000658 , 1.000761 , 1.000375 , 1.000095 at t = 10 2 , 3 ⋅ 10 2 , 10 3 , 3 ⋅ 10 3 t=10^2,3\cdot10^2,10^3,3\cdot10^3 t = 1 0 2 , 3 ⋅ 1 0 2 , 1 0 3 , 3 ⋅ 1 0 3 . Thus the container budget is quadratic, the first moment is linear, and what is known differs from what is needed by exactly one factor o ( 1 ) o(1) o ( 1 ) per d d d — the Ω ( log E ) \Omega(\log E) Ω ( log E ) saving of §9 in closed form. In particular the earlier lcm ( 1.. E ) ∼ e E \operatorname{lcm}(1..E)\sim e^E lcm ( 1.. E ) ∼ e E argument is unnecessary: C E C_E C E itself is a fixed integer of size exp ( Θ ( E 2 ) ) \exp(\Theta(E^2)) exp ( Θ ( E 2 )) .
Locating (H-j): Katz, Murata, and the GRH ruling
Write q q ( u ) = k u + q m u q_q(u)=k_u+qm_u q q ( u ) = k u + q m u with k u , m u ∈ [ 0 , q ) k_u,m_u\in[0,q) k u , m u ∈ [ 0 , q ) . Then W ‾ u ( q ) : = W u ( q ) m o d Z = q q ( u ) q m o d Z \overline{W}_u(q):=W_u(q)\bmod\mathbb{Z}=\frac{q_q(u)}{q}\bmod\mathbb{Z} W u ( q ) := W u ( q ) mod Z = q q q ( u ) mod Z has fractional part k u / q k_u/q k u / q , so q ∈ W u ⟺ k u = 0 ⟺ W ‾ u ( q ) = 0 q\in W_u\Longleftrightarrow k_u=0\Longleftrightarrow\overline{W}_u(q)=0 q ∈ W u ⟺ k u = 0 ⟺ W u ( q ) = 0 ; and when q ∈ W u q\in W_u q ∈ W u one has m u ( q ) ≡ ( u q − 1 − 1 ) / q 2 ( m o d q ) m_u(q)\equiv(u^{q-1}-1)/q^2\pmod q m u ( q ) ≡ ( u q − 1 − 1 ) / q 2 ( mod q ) , i.e. m r = w q m_r=w_q m r = w q , m a ′ = β q m_{a'}=\beta_q m a ′ = β q . Thus ( w q , β q ) (w_q,\beta_q) ( w q , β q ) is the mod-q q q higher layer of the Katz quotient at its Haar-null zeros: one order higher than W ‾ \overline{W} W (the mod-q 2 q^2 q 2 information, equivalently u q − 1 − 1 m o d q 3 u^{q-1}-1\bmod q^3 u q − 1 − 1 mod q 3 ). If a ′ = r k a'=r^k a ′ = r k then q q ( r k ) ≡ k q q ( r ) ( m o d q ) q_q(r^k)\equiv kq_q(r)\pmod q q q ( r k ) ≡ k q q ( r ) ( mod q ) and the pair ( W ‾ r , W ‾ a ′ ) (\overline{W}_r,\overline{W}_{a'}) ( W r , W a ′ ) lies on the one-dimensional subtorus { ( x , k x ) } \{(x,kx)\} {( x , k x )} : exactly the degenerate case excluded by the Zariski-density hypothesis of the group-scheme version, matching the Case A/B dichotomy. (Numerically, the joint χ 2 \chi^2 χ 2 for ( 3 , 9 ) (3,9) ( 3 , 9 ) and ( 3 , 27 ) (3,27) ( 3 , 27 ) explodes to 7736 and 3454 against a critical value 36.4, while independent bases pass; see § 12 .)
The continuous equidistribution of Katz does not control point masses at scale 1 / q 1/q 1/ q . (H-j) requires the three-layer discretized joint refinement:
(D1) Pointwise discretization: # { q ≤ x : W ‾ u ( q ) = 0 } ∼ ∑ q ≤ x 1 / q ∼ log log x \#\{q\le x:\overline{W}_u(q)=0\}\sim\sum_{q\le x}1/q\sim\log\log x # { q ≤ x : W u ( q ) = 0 } ∼ ∑ q ≤ x 1/ q ∼ log log x (Lang’s fixed-residue statement [4 ] and the Crandall–Dilcher–Pomerance model [5 ] );
(D2) Mod-q 2 q^2 q 2 two-base joint: for multiplicatively independent ⟨ r , a ′ ⟩ \langle r,a'\rangle ⟨ r , a ′ ⟩ , the pair ( w q , β q ) (w_q,\beta_q) ( w q , β q ) is jointly uniform on F q 2 \mathbb{F}_q^2 F q 2 and P ( q ∈ W r ∩ W a ′ ) ≍ 1 / q 2 P(q\in W_r\cap W_{a'})\asymp1/q^2 P ( q ∈ W r ∩ W a ′ ) ≍ 1/ q 2 ; (D3) Discrete-logarithm layer : conditional on compatibility, j q j_q j q is uniform on [ 0 , d q ) [0,d_q) [ 0 , d q ) , so P ( j q = 0 ) = 1 / d q P(j_q=0)=1/d_q P ( j q = 0 ) = 1/ d q .
Set (K)=(D1)+(D2)+(D3). None of (D1)–(D3) is a GRH theorem or a direct consequence of Katz’s conjecture.
Proposition 10.1 (conditional on the random model (K)). Assume (K) and that r r r , a ′ a' a ′ are multiplicatively independent (Case B). Then
E [ ∑ q ∈ B r , a ′ s q log q b q ] ≪ log r ∑ q log q q 2 < ∞ , \mathbb{E}\left[\sum_{q\in B_{r,a'}}\frac{s_q\log q}{b_q}\right]\ll\log r\sum_q\frac{\log q}{q^2}<\infty, E q ∈ B r , a ′ ∑ b q s q log q ≪ log r q ∑ q 2 log q < ∞ , so (H-j) holds almost surely; together with P 4 P_4 P 4 and Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ one obtains K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ and the density bound of Theorem 1.3 .
Proof. Here (K) is used with its built-in independence: (D2) and (D3) are joint uniformities, so ( w q , β q ) (w_q,\beta_q) ( w q , β q ) and j q j_q j q are independent, and s q log q s_q\log q s q log q (which depends only on r r r ) is independent of b q b_q b q . By (D2)–(D3) P ( q ∈ B ) ≤ P ( q ∈ W r ∩ W a ′ ) ⋅ P ( j q = 0 ) ≤ q − 2 ⋅ d q − 1 ≤ q − 2 P(q\in B)\le P(q\in W_r\cap W_{a'})\cdot P(j_q=0)\le q^{-2}\cdot d_q^{-1}\le q^{-2} P ( q ∈ B ) ≤ P ( q ∈ W r ∩ W a ′ ) ⋅ P ( j q = 0 ) ≤ q − 2 ⋅ d q − 1 ≤ q − 2 ; by (D2) conditional uniformity b q b_q b q is uniform on F q × \mathbb{F}_q^\times F q × (note w q ≠ 0 w_q\ne0 w q = 0 on the layer s q = 2 s_q=2 s q = 2 ), so E [ 1 / b q ] = H q − 1 / q ≍ ( log q ) / q \mathbb{E}[1/b_q]=H_{q-1}/q\asymp(\log q)/q E [ 1/ b q ] = H q − 1 / q ≍ ( log q ) / q ; and s q log q < d q log r ≤ q log r s_q\log q<d_q\log r\le q\log r s q log q < d q log r ≤ q log r by (2) . Multiplying, E [ contribution of q ] ≪ q − 2 ⋅ q log r ⋅ ( log q ) / q = log r ⋅ ( log q ) / q 2 \mathbb{E}[\text{contribution of }q]\ll q^{-2}\cdot q\log r\cdot(\log q)/q=\log r\cdot(\log q)/q^2 E [ contribution of q ] ≪ q − 2 ⋅ q log r ⋅ ( log q ) / q = log r ⋅ ( log q ) / q 2 , summable. Nonnegativity plus Markov give a.s. convergence. The same estimate controls Ξ r , a ′ ≪ ∑ q ( log q ) / q 2 \Xi_{r,a'}\ll\sum_q(\log q)/q^2 Ξ r , a ′ ≪ ∑ q ( log q ) / q 2 directly. □
Sketch of the GRH ruling (Theorem 1.5). (i) Literature: no theorem of the form GRH ⇒ N r ( x ) ≪ x 1 / 2 log A x \Rightarrow N_r(x)\ll x^{1/2}\log^A x ⇒ N r ( x ) ≪ x 1/2 log A x , GRH ⇒ W r \Rightarrow W_r ⇒ W r has density 0, or GRH ⇒ Σ r < ∞ \Rightarrow\Sigma_r<\infty ⇒ Σ r < ∞ is known; the varying-modulus side is precisely where the classical literature is silent. (ii) Algebra: the congruence q 2 ∣ r q − 1 − 1 q^2\mid r^{q-1}-1 q 2 ∣ r q − 1 − 1 has modulus q q q itself; the natural Chebotarev fields L q = Q ( ζ q 2 , r 1 / q 2 ) L_q=\mathbb{Q}(\zeta_{q^2},r^{1/q^2}) L q = Q ( ζ q 2 , r 1/ q 2 ) vary with q q q , are wildly ramified, and satisfy log d L q ≍ q 2 log q \log d_{L_q}\asymp q^2\log q log d L q ≍ q 2 log q , so the termwise error in quantitative GRH–Chebotarev, ∑ q ≤ x x 1 / 2 log ( d L q x n L q ) / q 1 / 2 \sum_{q\le x}x^{1/2}\log(d_{L_q}x^{n_{L_q}})/q^{1/2} ∑ q ≤ x x 1/2 log ( d L q x n L q ) / q 1/2 , is uncontrolled; the averaged versions require a family with bounded discriminant growth, which this family is not; Hooley’s GRH Artin method handles only fixed d d d in Q ( ζ d , r 1 / d ) \mathbb{Q}(\zeta_d,r^{1/d}) Q ( ζ d , r 1/ d ) . (iii) Direction: GRH controls only the ord q ( r ) \operatorname{ord}_q(r) ord q ( r ) (Artin-type) distribution, and the Hensel wall has already converted the weight 1 / ord q ( r ) 1/\operatorname{ord}_q(r) 1/ ord q ( r ) into 1 / D 3 = 1 / ( d q q max ( 0 , 3 − s q ) ) 1/D_3=1/(d_qq^{\max(0,3-s_q)}) 1/ D 3 = 1/ ( d q q m a x ( 0 , 3 − s q ) ) ; hence in any composite “GRH + W-side assumption” package, GRH is redundant. □
Remark 10.2 (Classical coordinates). Murata’s 1991 conjectures [3 ] : (i) p 3 ∣ a p − a p^3\mid a^p-a p 3 ∣ a p − a for only finitely many p p p — exactly P 6 a P_{6a} P 6 a ; (ii) for coprime b , c > 1 b,c>1 b , c > 1 , p 2 ∣ b p − b p^2\mid b^p-b p 2 ∣ b p − b and p 2 ∣ c p − c p^2\mid c^p-c p 2 ∣ c p − c simultaneously for only finitely many p p p — exactly # ( W r ∩ W a ′ ) < ∞ \#(W_r\cap W_{a'})<\infty # ( W r ∩ W a ′ ) < ∞ , the strong form of (HW). Our (H-j) is weaker than Murata (ii) (the intersection is allowed to be infinite; only the bad subset must be 1 / b q 1/b_q 1/ b q -summable) and stronger than Katz’s conjecture (which is continuous equidistribution and silent on all three discretized layers). The correct description of (H-j) is: a weighted refinement interpolating between Katz 2015 and Murata 1991.
The unified family (K+) and the main line
10.1. The unified family (K+) and the main line. Recall ( K + ) = ( K ) ∪ { ( D 4 ) } (K^+)=(K)\cup\{(D4)\} ( K + ) = ( K ) ∪ {( D 4 )} with (D4) from §8.1.
Theorem 10.3 (Unified conclusion). Under ( K + ) (K^+) ( K + ) , for every fixed r ≥ 2 r\ge2 r ≥ 2 and every a ′ a' a ′ with gcd ( a ′ , r ) = 1 \gcd(a',r)=1 g cd( a ′ , r ) = 1 : K r , a ′ < ∞ K_{r,a'}<\infty K r , a ′ < ∞ almost surely; consequently the density bound of Theorem 1.3 holds almost surely for every a ′ a' a ′ .
Proof. Case A: by Proposition 5.3 , K r , r k ≤ 2 Ξ r = 2 ( Ξ r ( 2 ) + Ξ r ( 3 ) ) K_{r,r^k}\le2\Xi_r=2(\Xi_r^{(2)}+\Xi_r^{(3)}) K r , r k ≤ 2 Ξ r = 2 ( Ξ r ( 2 ) + Ξ r ( 3 ) ) ; Ξ r ( 2 ) < ∞ \Xi_r^{(2)}<\infty Ξ r ( 2 ) < ∞ is the unconditional Theorem 7.5 , and Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ a.s. is Theorem 8.10 . Case B: K r , a ′ = Ξ r , a ′ + J ′ ≤ Ξ r + J ′ K_{r,a'}=\Xi_{r,a'}+J'\le\Xi_r+J' K r , a ′ = Ξ r , a ′ + J ′ ≤ Ξ r + J ′ ; the Ξ r \Xi_r Ξ r part is controlled as before (Theorems 7.5 and 8.10 ); J ′ J^{\prime} J ′ is controlled by (D1)–(D3): on the bad set P ( q ∈ B ) ≪ q − 2 d q − 1 ≤ q − 2 \mathbb{P}(q \in\mathcal{B}) \ll q^{-2d_q-1} \leq q^{-2} P ( q ∈ B ) ≪ q − 2 d q − 1 ≤ q − 2 , conditionally E [ 1 / b q ] ≍ ( log q ) / q \mathbb{E}[1/b_q] \asymp(\log q)/q E [ 1/ b q ] ≍ ( log q ) / q , and with (U2),
E [ ∑ q ∈ B r , a ′ s q log q b q ] ≪ log r ∑ q log q q 2 < ∞ , \mathbb{E}\left[\sum_{q\in\mathcal{B}_{r,a^{\prime}}}\frac{s_q\log q}{b_q}\right]\ll\log r\sum_q\frac{\log q}{q^2}<\infty, E q ∈ B r , a ′ ∑ b q s q log q ≪ log r q ∑ q 2 log q < ∞ ,
i.e. (H-j) holds a.s., so J ′ < ∞ J^{\prime}<\infty J ′ < ∞ a.s. Finally apply Theorem 1.3 . □
Erratum note. An earlier draft claimed that “(D4) does not participate in Case B”; this is inaccurate. In Case B, Ξ r , a ′ ≤ Ξ r \Xi_{r,a^{\prime}}\leq\Xi_r Ξ r , a ′ ≤ Ξ r still contains the s ≥ 3 s\geq3 s ≥ 3 layer, which must be controlled by (D4); (K) alone controls only J ′ J^{\prime} J ′ .
Honest strength ordering.
( K + ) ⇒ P 6 a a . s . ⇒ Ξ r ( 3 ) < ∞ ⇒ P 4 ( 3 ) , (\mathrm{K}^{+})\Rightarrow P_{6a}\ \mathrm{a.s.}\Rightarrow\Xi_r^{(3)}<\infty\Rightarrow P_4^{(3)}, ( K + ) ⇒ P 6 a a.s. ⇒ Ξ r ( 3 ) < ∞ ⇒ P 4 ( 3 ) ,
the last three arrows by Theorem 8.1. Read as a counting density, (D4) is equivalent to P 6 a P_{6a} P 6 a (Theorem 14.4 ); its probabilistic form (Definition 8.8 , finite mean) is strictly stronger, yielding Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ a.s. Its value lies in being same-origin with (K) and in covering the main line and the bad-set line uniformly. The final condition set of the chain may be stated either as the classical anchor P 6 a P_{6a} P 6 a or as the unified conjecture family (K+).
Ancillary results and the redundancy of GRH
Three results decide which hypotheses are redundant in the chain; none of them fills the application-layer gap, but together they show GRH is nowhere needed.
Theorem 11.1 (S o r d < ∞ S_{\rm ord}<\infty S ord < ∞ , unconditional). S o r d = ∑ p ∤ r log p d p ( p − 1 ) < ∞ S_{\rm ord}=\sum_{p\nmid r}\frac{\log p}{d_p(p-1)}<\infty S ord = ∑ p ∤ r d p ( p − 1 ) l o g p < ∞ .
Proof. Erdős–Murty [18 ] prove unconditionally ∑ p ≤ x 1 / d p ≪ x / ( log x ) 1 + δ \sum_{p\leq x}1/d_p\ll\sqrt{x}/(\log x)^{1+\delta} ∑ p ≤ x 1/ d p ≪ x / ( log x ) 1 + δ (Murty–Srinivasan [19 ] already gave ≪ x \ll\sqrt{x} ≪ x ). With the decreasing weight w ( t ) = log t t − 1 w(t)=\frac{\log t}{t-1} w ( t ) = t − 1 l o g t (t ≥ 3 t\geq3 t ≥ 3 ), Stieltjes partial summation gives
∑ 3 ≤ p ≤ x w ( p ) d p = A ( x ) w ( x ) − ∫ 3 x A ( t ) w ′ ( t ) d t , \sum_{3\leq p\leq x}\frac{w(p)}{d_p}=A(x)w(x)-\int_3^x A(t)w^{\prime}(t)\,dt, 3 ≤ p ≤ x ∑ d p w ( p ) = A ( x ) w ( x ) − ∫ 3 x A ( t ) w ′ ( t ) d t , with A ( x ) = ∑ p ≤ x 1 / d p A(x)=\sum_{p\leq x}1/d_p A ( x ) = ∑ p ≤ x 1/ d p ; the boundary term is ≪ x − 1 / 2 ( log x ) − δ → 0 \ll x^{-1/2}(\log x)^{-\delta}\to0 ≪ x − 1/2 ( log x ) − δ → 0 , and after t = e u t=e^u t = e u the integral is ≪ ∫ ∞ e − u / 2 u − δ d u < ∞ \ll\int^\infty e^{-u/2}u^{-\delta}\,du<\infty ≪ ∫ ∞ e − u /2 u − δ d u < ∞ . □
The GRH version of the order bound (Kurlberg–Pomerance [16 ] , d p ≥ p / ε ( p ) d_p\geq p/\varepsilon(p) d p ≥ p / ε ( p ) on average) is not needed anywhere: GRH is redundant throughout the chain.
Theorem 11.2 (C r C_r C r exact identity). With X n = log n rad n X_n=\frac{\log n}{\operatorname{rad} n} X n = rad n l o g n , for a ′ = 1 a^{\prime}=1 a ′ = 1 ,
∑ e ≤ E X r e − 1 = ∑ p ∤ r log p ∑ k ≥ 2 ∣ E D k ( p ) ∣ + LTE 2 ( E ) , \sum_{e\leq E}X_{r^e-1}=\sum_{p\nmid r}\log p\sum_{k\geq2}\left\lvert\frac{E}{D_k(p)}\right\rvert+\operatorname{LTE}_2(E), e ≤ E ∑ X r e − 1 = p ∤ r ∑ log p k ≥ 2 ∑ D k ( p ) E + LTE 2 ( E ) , so 1 E ∑ e ≤ E X r e − 1 ↗ C r = S W + S o r d + LTE 2 ( r ) \frac{1}{E}\sum_{e\leq E}X_{r^e-1}\nearrow C_r=S_W+S_{\rm ord}+\operatorname{LTE}_2(r) E 1 ∑ e ≤ E X r e − 1 ↗ C r = S W + S ord + LTE 2 ( r ) , where S W = ∑ q ∈ W r ( s q − 1 ) log q d q S_W=\sum_{q\in\mathcal{W}_r}\frac{(s_q-1)\log q}{d_q} S W = ∑ q ∈ W r d q ( s q − 1 ) l o g q . Hence (H-X) ⟺ S W < ∞ \Longleftrightarrow S_W<\infty ⟺ S W < ∞ . Numerically C 2 = 0.6938 C_2=0.6938 C 2 = 0.6938 , C 3 = 1.9112 C_3=1.9112 C 3 = 1.9112 , C 5 = 1.9745 C_5=1.9745 C 5 = 1.9745 ; for r = 3 r=3 r = 3 the decomposition is S W = 0.479579 S_W=0.479579 S W = 0.479579 , S o r d = 0.391941 S_{\rm ord}=0.391941 S ord = 0.391941 , LTE 2 = 1.039721 \operatorname{LTE}_2=1.039721 LTE 2 = 1.039721 , summing to C 3 = 1.911241 C_3=1.911241 C 3 = 1.911241 .
Indeed X n = ∑ p log p ∑ k ≥ 2 1 v p ( n ) ≥ k X_n=\sum_p\log p\sum_{k\geq2}1_{v_p(n)\geq k} X n = ∑ p log p ∑ k ≥ 2 1 v p ( n ) ≥ k , and v p ( r e − 1 ) ≥ k ⟺ D k ( p ) ∣ e v_p(r^e-1)\geq k\Longleftrightarrow D_k(p)\mid e v p ( r e − 1 ) ≥ k ⟺ D k ( p ) ∣ e (the LTE tower of Theorem 1.1 ); the prime p = 2 p=2 p = 2 contributes the finite correction LTE 2 ( E ) \operatorname{LTE}_2(E) LTE 2 ( E ) . For p ∈ W r p\in\mathcal{W}_r p ∈ W r (s p ≥ 2 s_p\geq2 s p ≥ 2 ) the limiting contribution is log p ∑ k = 2 s p 1 / D k ( p ) = ( s p − 1 ) log p d p \log p\sum_{k=2}^{s_p}1/D_k(p)=\frac{(s_p-1)\log p}{d_p} log p ∑ k = 2 s p 1/ D k ( p ) = d p ( s p − 1 ) l o g p ; for p ∉ W r p\notin\mathcal{W}_r p ∈ / W r it is log p d p ( p − 1 ) \frac{\log p}{d_p(p-1)} d p ( p − 1 ) l o g p .
Theorem 11.3 (Strict one-way). Termwise Ξ r ≤ 3 2 S W + O ( 1 ) \Xi_r \le\frac{3}{2}S_W+O(1) Ξ r ≤ 2 3 S W + O ( 1 ) (ratios Ξ q / S W , q \Xi_q/S_{W,q} Ξ q / S W , q : s = 2 s=2 s = 2 gives 2 / q 2/q 2/ q , s = 3 s=3 s = 3 gives 3 / 2 3/2 3/2 , s ≥ 4 s\ge4 s ≥ 4 gives s / ( s − 1 ) s/(s-1) s / ( s − 1 ) ), so S W < ∞ ⇒ Ξ r < ∞ S_W<\infty\Rightarrow\Xi_r<\infty S W < ∞ ⇒ Ξ r < ∞ . The converse fails: the exact identity S W ( 2 ) = ∑ q q 2 Ξ q ( 2 ) S_W^{(2)}=\sum_q\frac{q}{2}\Xi_q^{(2)} S W ( 2 ) = ∑ q 2 q Ξ q ( 2 ) shows S W S_W S W is the q q q -weighted first moment of Ξ r \Xi_r Ξ r ; the universal bound Ξ q ( 2 ) ≤ 2 log 2 q q 2 \Xi_q^{(2)}\le\frac{2\log^2q}{q^2} Ξ q ( 2 ) ≤ q 2 2 l o g 2 q is summable over all primes (1.2429), while the corresponding S W S_W S W terms ∼ log 2 q q \sim\frac{\log^2q}{q} ∼ q l o g 2 q diverge (92.65). A configuration with d q = q − 1 d_q=q-1 d q = q − 1 and W r \mathcal{W}_r W r of positive density makes Ξ r ≈ 0.818 \Xi_r\approx0.818 Ξ r ≈ 0.818 convergent while S W ∼ 1 m log X S_W\sim\frac{1}{m}\log X S W ∼ m 1 log X diverges; since P1 is open, no theorem excludes it.
Theorem 11.4 (Evolution of the condition set).
P 4 + Ξ r ( 3 ) ⟶ P 4 ′ + Ξ r ( 3 ) ⟶ Ξ r ( 3 ) alone , P_4+\Xi_r^{(3)}\longrightarrow P'_4+\Xi_r^{(3)}\longrightarrow\Xi_r^{(3)}\ \text{alone}, P 4 + Ξ r ( 3 ) ⟶ P 4 ′ + Ξ r ( 3 ) ⟶ Ξ r ( 3 ) alone , where P 4 ′ = ∑ q ∈ W r 1 / q < ∞ P'_4=\sum_{q\in\mathcal{W}_r}1/q<\infty P 4 ′ = ∑ q ∈ W r 1/ q < ∞ was once supported by d q ≥ log q / log r d_q\ge\log q/\log r d q ≥ log q / log r ((2) ), giving Ξ q ( 2 ) ≪ 1 / q \Xi_q^{(2)}\ll1/q Ξ q ( 2 ) ≪ 1/ q . Theorem 7.5 removes P 4 P_4 P 4 , P 4 ′ P'_4 P 4 ′ and P1 from the condition set entirely (they remain open as statements about W r \mathcal{W}_r W r ). The (H-X)/S W S_W S W route is a stronger-than-necessary fallback: strictly one-way (Theorem 11.3 ), not equivalent to the main line.
Table 20.
(K+) collects both routes into one family: (D4) governs the deep layer of A, (D1)–(D3) govern the bad set of B.
Computational evidence
Variable direction. Every statistic in this section is a fixed-r r r , varying-q q q scan, and the hypotheses they test — (D1), (D2), (D4), (K) — are likewise fixed-r r r , varying-q q q pointwise statements. A fixed-q q q census over bases r r r is a different, easier object: by Proposition 8.9 the solutions of u q − 1 ≡ 1 ( m o d q 2 ) u^{q-1}\equiv1\pmod{q^2} u q − 1 ≡ 1 ( mod q 2 ) are exactly the Teichmüller subgroup μ q − 1 \mu_{q-1} μ q − 1 , so a scan over r r r hits a residue class with rate ( q − 1 ) / q 2 (q-1)/q^2 ( q − 1 ) / q 2 — even the trivial family r ≡ 1 ( m o d q 3 ) r\equiv1\pmod{q^3} r ≡ 1 ( mod q 3 ) (d q = 1 d_q=1 d q = 1 ) is included. Such a census cannot test (D1)/(D2)/(D4); see §12.3 for the correction this distinction forces.
All computations were performed with exact integer arithmetic (pow with modulus q 3 q^3 q 3 ); scripts and data are available with this note.
The bad-set scan
For q < 10 6 q<10^6 q < 1 0 6 , twelve bases r ≤ 37 r\le37 r ≤ 37 and a ′ < 30 a'<30 a ′ < 30 : the set W r \mathcal{W}_r W r has 23 elements; the bad set has five elements, all with b q ≤ 4 b_q\le4 b q ≤ 4 :
( 7 , 26 , 5 , 4 ) , ( 17 , 19 , 3 , 2 ) , ( 19 , 10 , 3 , 2 ) , ( 37 , 10 , 3 , 1 ) , ( 37 , 19 , 3 , 2 ) , (7,26,5,4),\quad(17,19,3,2),\quad(19,10,3,2),\quad(37,10,3,1),\quad(37,19,3,2), ( 7 , 26 , 5 , 4 ) , ( 17 , 19 , 3 , 2 ) , ( 19 , 10 , 3 , 2 ) , ( 37 , 10 , 3 , 1 ) , ( 37 , 19 , 3 , 2 ) ,
and the partial sum in (H-j) stabilizes at 6.30. The bad-layer primes are only 3 and 5. A layer census for b ≤ 15 b\le15 b ≤ 15 finds no cube-divisor candidates at all for ( r , a ′ ) (r,a') ( r , a ′ ) equal to ( 2 , 1 ) (2,1) ( 2 , 1 ) , ( 2 , 3 ) (2,3) ( 2 , 3 ) , ( 3 , 1 ) (3,1) ( 3 , 1 ) , ( 3 , 2 ) (3,2) ( 3 , 2 ) or ( 5 , 2 ) (5,2) ( 5 , 2 ) , matching the random expectation ∑ q q − 3 ≈ 0.17 \sum_q q^{-3}\approx0.17 ∑ q q − 3 ≈ 0.17 per integer.
First super-Wieferich sample
The scan exhibits the first observed super-Wieferich prime: s 7 ( 19 ) = 3 s_7(19) = 3 s 7 ( 19 ) = 3 exactly, since 19 6 − 1 = 2 3 ⋅ 3 3 ⋅ 5 ⋅ 7 3 ⋅ 127 19^6 - 1 = 2^3 \cdot3^3 \cdot5 \cdot7^3 \cdot127 1 9 6 − 1 = 2 3 ⋅ 3 3 ⋅ 5 ⋅ 7 3 ⋅ 127 , and 19 is a primitive root modulo 7. Note ⟨ 19 ⟩ m o d 343 = { ± 1 , ± 18 , ± 19 } \langle19\rangle\bmod343 = \{\pm1, \pm18, \pm19\} ⟨ 19 ⟩ mod 343 = { ± 1 , ± 18 , ± 19 } with D 3 = d q = 6 D_3 = d_q = 6 D 3 = d q = 6 : on the super-Wieferich layer e 0 ∈ [ 1 , d q ] e_0 \in[1,d_q] e 0 ∈ [ 1 , d q ] is automatically < q < q < q whenever compatible, but this layer is not the (H-j) bad layer—it is controlled separately by P 6 a P_{6a} P 6 a , via the weight s q log q / d q = 3 log 7 / 6 ≈ 0.973 s_q \log q/d_q = 3\log7/6 \approx0.973 s q log q / d q = 3 log 7/6 ≈ 0.973 .
Systematic super-Wieferich sample. A full scan over 2 ≤ r ≤ 500 2 \le r \le500 2 ≤ r ≤ 500 , q < 2 × 10 5 q < 2 \times10^5 q < 2 × 1 0 5 yields 71 genuine pairs with s q ≥ 3 s_q \ge3 s q ≥ 3 (15 with s q ≥ 4 s_q \ge4 s q ≥ 4 ), on 12 distinct primes q ≤ 113 q \le113 q ≤ 113 and 16 distinct orders d q ≤ 112 d_q \le112 d q ≤ 112 . All satisfy q ∣ Φ d q ( r ) q \mid\Phi_{d_q}(r) q ∣ Φ d q ( r ) , the structural bound 3 log q min ( d q ) ≤ φ ( d q ) log ( r + 1 ) 3\log q_{\min}(d_q) \le\varphi(d_q)\log(r+1) 3 log q m i n ( d q ) ≤ φ ( d q ) log ( r + 1 ) (Remark 9.3 ) and d q ≥ 3 log q log ( r + 1 ) d_q \ge\frac{3\log q}{\log(r+1)} d q ≥ l o g ( r + 1 ) 3 l o g q with zero violations; the accumulated Ξ ( 3 ) \Xi^{(3)} Ξ ( 3 ) mass is 159.947 (a lower bound), and N 3 ( 112 ) ≥ 16 N_3(112) \ge16 N 3 ( 112 ) ≥ 16 . The pair ( 19 , 7 ) (19,7) ( 19 , 7 ) above is the equality case: Φ 6 ( 19 ) = 343 = 7 3 \Phi_6(19) = 343 = 7^3 Φ 6 ( 19 ) = 343 = 7 3 exactly.
Effective twin-Wieferich
Erratum (null hypothesis). The 27 effective pairs are reproduced exactly by counting distinct ordered pairs ( r , a ′ ) (r,a') ( r , a ′ ) with a ′ > 1 a' > 1 a ′ > 1 , gcd ( r , a ′ ) = 1 \gcd(r,a') = 1 g cd( r , a ′ ) = 1 , q < 10 6 q < 10^6 q < 1 0 6 ; a scan over all triples gives 78 (23 trivial, 55 effective), reducing to 27 distinct ordered pairs. (The earlier “49 = 22 + 27” under-counted both layers; the trivial-pair count 22 differs by one from the number of W r \mathcal{W}_r W r hits and awaits the author’s reconciliation.) The expectation 27 ∑ q − 2 ≈ 12.2 27\sum q^{-2} \approx12.2 27 ∑ q − 2 ≈ 12.2 of an earlier version uses the observed number of pairs as the number of candidates and is circular. More importantly, at fixed q q q the congruence u q − 1 ≡ 1 ( m o d q 2 ) u^{q-1} \equiv1 \pmod{q^2} u q − 1 ≡ 1 ( mod q 2 ) has exactly q − 1 q - 1 q − 1 solutions in ( Z / q 2 Z ) × (\mathbb{Z}/q^2\mathbb{Z})^\times ( Z / q 2 Z ) × , namely the Teichmüller subgroup μ q − 1 \mu_{q-1} μ q − 1 (Proposition 8.9 ); the correct null density for a scan over bases is ( q − 1 ) / q 2 (q - 1)/q^2 ( q − 1 ) / q 2 , not 1 / q 2 1/q^2 1/ q 2 . Measured densities for q = 3 , 5 , 7 , 11 , 13 , 17 , 19 q = 3, 5, 7, 11, 13, 17, 19 q = 3 , 5 , 7 , 11 , 13 , 17 , 19 are 0.222, 0.160, 0.122, 0.082, 0.0695, 0.055, 0.049, agreeing with ( q − 1 ) / q 2 (q - 1)/q^2 ( q − 1 ) / q 2 to three decimals; the “factor 4–6 excess” reported in an earlier version is this missing factor q − 1 ∈ [ 2 , 18 ] q - 1 \in[2,18] q − 1 ∈ [ 2 , 18 ] , not positive correlation. Likewise the four pairs at q = 13 q = 13 q = 13 are four of the six ordered pairs inside μ 12 ( Z / 169 Z ) ∩ [ 2 , 30 ] = { 19 , 22 , 23 } \mu_{12}(\mathbb{Z}/169\mathbb{Z}) \cap[2,30] = \{19,22,23\} μ 12 ( Z /169 Z ) ∩ [ 2 , 30 ] = { 19 , 22 , 23 } , and the q = 3 q = 3 q = 3 layer is the structural condition u ≡ ± 1 ( m o d 9 ) = μ 2 ( Z / 9 Z ) u \equiv\pm1 \pmod9 = \mu_2(\mathbb{Z}/9\mathbb{Z}) u ≡ ± 1 ( mod 9 ) = μ 2 ( Z /9 Z ) . All of these are consequences of the Teichmüller anchor, not statistical anomalies.
What remains open. This does not restore numerical support for (D2): its conditional-on-k = 0 k = 0 k = 0 content remains untested, since below q < 3 × 10 5 q < 3 \times10^5 q < 3 × 1 0 5 only 10 primes carry a second hit; a meaningful test needs q ∼ 10 9 q \sim10^9 q ∼ 1 0 9 –10 10 10^{10} 1 0 10 , conditioned on q ∈ W r q \in\mathcal{W}_r q ∈ W r , and is beyond present reach. The bad-set conclusions of this section (five hits, partial sum 6.30) are unaffected.
Katz layer tests
On the full odd-prime population q < 5 ⋅ 10 4 q < 5 \cdot10^4 q < 5 ⋅ 1 0 4 the four-dimensional statistic ( k u , m u ) (k_u,m_u) ( k u , m u ) for independent bases passes marginal uniformity and joint independence (KS ≤ 0.0145 \le0.0145 ≤ 0.0145 against critical 0.0190; joint χ 2 ≤ 29.9 \chi^2 \le29.9 χ 2 ≤ 29.9 against 36.4), while Case A pairs ( 3 , 9 ) (3,9) ( 3 , 9 ) , ( 3 , 27 ) (3,27) ( 3 , 27 ) explode (χ 2 = 7736 , 3454 \chi^2 = 7736,3454 χ 2 = 7736 , 3454 ), exactly reproducing the Zariski-density restriction. The twin-Wieferich numerics of §12 cut against (D2) over the full observable range, so (K) is a heuristic whose second layer is currently unsupported and whose use in Proposition 10.1 must be read conditionally.
Limitation: below 5 ⋅ 10 4 5 \cdot10^4 5 ⋅ 1 0 4 the expected number of Wieferich primes is ≈ 2.4 \approx2.4 ≈ 2.4 and of twin-Wieferich ≈ 0.45 \approx0.45 ≈ 0.45 , so the conditional-on-k = 0 k = 0 k = 0 content of (D1), (D2) has no numerical evidence; the s q ≥ 3 s_q \ge3 s q ≥ 3 branch has the single sample ( 19 , 7 ) (19,7) ( 19 , 7 ) .
Observable summary. All deep hits ever observed have q ≤ 13 q \le13 q ≤ 13 and d q ≤ 6 d_q \le6 d q ≤ 6 ; A o b s ( t ) A_{\mathrm{obs}}(t) A obs ( t ) is a truncated lower bound (saturating for t ≥ 6 t \ge6 t ≥ 6 on the scanned range), not a proof of A ( t ) = o ( t ) A(t) = o(t) A ( t ) = o ( t ) . The statistics of this section and of §14.4–14.5 are the first fixed-census data for W r ( 3 ) \mathcal{W}_r^{(3)} W r ( 3 ) : the m ( q ) m(q) m ( q ) table tests P 6 a P_{6a} P 6 a for all r ≤ X r \le X r ≤ X simultaneously at fixed X X X , while the pair scans test (D1)/(D2) pointwise at fixed r r r —a strictly stronger requirement (§12.3, Proposition 14.5 ).
Relation to ( ⋆ ′ ) (\star') ( ⋆ ′ )
In the framework of [1 ] the remaining gap is stated in Section 7 (“The remaining gap, exactly”) as Working hypothesis 6.4: ∣ log x ∣ = o ( log c ) |\log x|=o(\log c) ∣ log x ∣ = o ( log c ) , equivalently O ( log log c ) O(\log\log c) O ( log log c ) , with x = det ( Λ ) ( k − 1 ) / k / Δ x=\det(\Lambda)^{(k-1)/k}/\Delta x = det ( Λ ) ( k − 1 ) / k /Δ and Δ = N ∣ a ′ ∣ 2 \Delta=N|a'|^2 Δ = N ∣ a ′ ∣ 2 ; Theorem 6.5 there gives η ≤ 1 k + O ( log log c log c ) \eta\le\frac{1}{k}+O\left(\frac{\log\log c}{\log c}\right) η ≤ k 1 + O ( l o g c l o g l o g c ) and Theorem 7.1 the equivalent form N ∣ a ′ ∣ 2 = ( c b c / g ) ( k − 1 ) / k c o ( 1 ) N|a'|^2=(cb_c/g)^{(k-1)/k}c^{o(1)} N ∣ a ′ ∣ 2 = ( c b c / g ) ( k − 1 ) / k c o ( 1 ) . The expression k − 1 k log c − log rad ( a b ) \frac{k-1}{k}\log c-\log\operatorname{rad}(ab) k k − 1 log c − log rad ( ab ) used in an earlier version of this note is a heuristic transcription, not a formula of [1 ] ; and the “degeneration to a b c abc ab c strength” should be read as the structural ceiling of Theorem 10.1 there (Π ≤ 1 / k + o ( 1 ) \Pi\le1/k+o(1) Π ≤ 1/ k + o ( 1 ) , attained at c = r e c=r^e c = r e ; full a b c abc ab c is structurally unreachable inside the framework). Three regimes: 0.3 0.3 0.3 (the present results, 1 2 − 1 5 \frac{1}{2}-\frac{1}{5} 2 1 − 5 1 ), 0.5 0.5 0.5 (k = 2 k=2 k = 2 ), and ( k − 1 ) / k → 1 (k-1)/k\to1 ( k − 1 ) / k → 1 (k → ∞ k\to\infty k → ∞ , i.e. a b c abc ab c ). Along the line a ′ = O ( 1 ) a'=O(1) a ′ = O ( 1 ) , b e = r e − a ′ b_e=r^e-a' b e = r e − a ′ , one has k e = ω ( a ′ b e ) → ∞ k_e=\omega(a'b_e)\to\infty k e = ω ( a ′ b e ) → ∞ , so the hypothesis degenerates to a b c abc ab c -strength already for bounded a ′ a' a ′ ; the present note therefore governs the k = 2 k=2 k = 2 slice. Theorem 1.2 pins the Layer-III Pillai candidates r e − ∏ i q i V i = a ′ r^e-\prod_i q_i^{V_i}=a' r e − ∏ i q i V i = a ′ to W r ∩ W a ′ W_r\cap W_{a'} W r ∩ W a ′ with an explicit class e e e mod q q q ; Theorem 1.3 pins the Layer-III mass to J r , a ′ J_{r,a'} J r , a ′ , the Pillai-single-solution layer (D 3 > E ≥ e 0 D_3>E\ge e_0 D 3 > E ≥ e 0 means “q q q hits exactly once in [ 1 , E ] [1,E] [ 1 , E ] ”); and (H-j) shows the application-layer residual gap does not require Baker-type bounds—the Baker layer belongs to k ≥ 3 k\ge3 k ≥ 3 and the V q = 2 V_q=2 V q = 2 squarefull layer, where the first moment of the union bound diverges (t q = V q − 1 − s q ≤ − 1 t_q=V_q-1-s_q\le-1 t q = V q − 1 − s q ≤ − 1 , no Hensel wall), and which is untouched here. The defect/order-core analysis of [15 ] on the Mersenne line independently corroborates this trichotomy.
Remark 13.1 (Notation and quantifiers in the ( ⋆ ′ ) (\star') ( ⋆ ′ ) interface). The symbol a ′ a' a ′ denotes three different objects in the sources combined here, and no identification between them is asserted: (i) in [1 ] , Definition 3.4, a ′ = h / g h a'=h/g_h a ′ = h / g h is a primitive normal vector in Z k \mathbb{Z}^k Z k , with N = [ ker H : Λ 0 ] N=[\ker H:\Lambda_0] N = [ ker H : Λ 0 ] and Δ = det ( Λ 0 ) = N ∣ a ′ ∣ 2 \Delta=\det(\Lambda_0)=N|a'|^2 Δ = det ( Λ 0 ) = N ∣ a ′ ∣ 2 —this is the N ∣ a ′ ∣ 2 N|a'|^2 N ∣ a ′ ∣ 2 occurring in Working hypothesis 6.4 and Theorem 7.1; (ii) in §9 of [1 ] (Theorem 9.8), a ′ a' a ′ is a scalar Pillai remainder with ∣ a ′ ∣ ≤ c o ( 1 ) |a'|\le c^{o(1)} ∣ a ′ ∣ ≤ c o ( 1 ) ; (iii) throughout the present note a ′ ≤ A a'\le A a ′ ≤ A is a fixed integer and b e = r e − a ′ b_e=r^e-a' b e = r e − a ′ . In particular the line a ′ = O ( 1 ) a'=O(1) a ′ = O ( 1 ) of the present note is a proper subfamily of Layer III in (ii), which is why the first-moment Markov argument can produce an e e e -density-zero conclusion here without resolving that layer in general.
Two uses of k k k must likewise be separated: k = ω ( a b ) k=\omega(ab) k = ω ( ab ) is a global quantity attached to a triple ( a , b , c ) (a,b,c) ( a , b , c ) (so k e = ω ( a ′ b e ) → ∞ k_e=\omega(a'b_e)\to\infty k e = ω ( a ′ b e ) → ∞ along b e = r e − a ′ b_e=r^e-a' b e = r e − a ′ ), whereas the “k = 2 k=2 k = 2 slice” governed by Theorem 1.3 is the local per-prime valuation layer V q ≥ 3 V_q\ge3 V q ≥ 3 . Working hypothesis 6.4 is a pointwise hypothesis ∣ log x ∣ = o ( log c ) |\log x|=o(\log c) ∣ log x ∣ = o ( log c ) with no “almost all” quantifier; Theorem 6.5 is a conditional statement along a sequence of triples with c → ∞ c\to\infty c → ∞ ; and the only density statement in [1 ] , Theorem 11.7(i), is on the c c c -axis and asserts that every coprime decomposition of a density-one set of c c c has q ≤ 1 + o ( 1 ) q\le1+o(1) q ≤ 1 + o ( 1 ) , proved by one-variable counting. There is no quantifier position in [1 ] for averaging over a ′ ∈ [ 1 , A ] a'\in[1,A] a ′ ∈ [ 1 , A ] or discarding a zero-density set of a ′ a' a ′ ; nor would such an average transfer to the pointwise statement of Theorem 6.5. Accordingly no averaging step is used or claimed below.
Acknowledgments
The reductions of §§4–10 were found and machine-verified in the course of an extended collaboration on the Wieferich tail-estimation problem; the numerical protocols follow the standards of [17 ] .
The condition set has now evolved to its minimal form. The s q = 2 s_q=2 s q = 2 layer is unconditionally closed (Theorem 7.5 ). The exact threshold of the first-moment route is
Φ r ( E ) = o ( E ) ⟺ A r ( t ) : = ∑ d ≤ t M d d = o ( t ) , M d = ∑ q : d q = d s q ≥ 3 s q log q \Phi_r(E)=o(E)\quad\Longleftrightarrow\quad A_r(t):=\sum_{d\le t}\frac{M_d}{d}=o(t),\qquad M_d=\sum_{\substack{q:d_q=d\\s_q\ge3}}s_q\log q Φ r ( E ) = o ( E ) ⟺ A r ( t ) := d ≤ t ∑ d M d = o ( t ) , M d = q : d q = d s q ≥ 3 ∑ s q log q
(Theorem 9.6 ); this is equivalent to the statement that the cube-full part of Φ d ( r ) \Phi_d(r) Φ d ( r ) has logarithmic mass o ( d ) o(d) o ( d ) in Cesàro mean. Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ is a convenient sufficient rung but is not necessary (Theorem 9.4 ): a configuration with M d = d / ( log d ) 2 M_d=d/(\log d)^2 M d = d / ( log d ) 2 has Ξ r ( 3 ) ( E ) ∼ E / ( log E ) 2 → ∞ \Xi_r^{(3)}(E)\sim E/(\log E)^2\to\infty Ξ r ( 3 ) ( E ) ∼ E / ( log E ) 2 → ∞ while Φ r ( E ) = o ( E ) \Phi_r(E)=o(E) Φ r ( E ) = o ( E ) . Classically the remaining input is anchored at P 6 a P_{6a} P 6 a (Murata 1991 (i)), probabilistically at (D4); GRH is redundant throughout (Theorem 11.1 and the Hensel-wall absorption of the ord side, Theorem 8.10 ).
Postscript
This section records the outcome of the fifteenth–eighteenth steps of the working chain: what a new tool would have to look like, and the single-variable reformulation of the deep-layer finiteness.
Literature verdict: no existing tool applies
A systematic check of the six natural candidates finds no unconditional bound log rad Φ n ( a ) ≥ φ ( n ) log a − o ( n ) \log\operatorname{rad}\Phi_n(a)\ge\varphi(n)\log a-o(n) log rad Φ n ( a ) ≥ φ ( n ) log a − o ( n ) — i.e. no theorem giving M d = o ( d ) M_d=o(d) M d = o ( d ) :
Stewart [24 , 23 ] controls the greatest prime factor , P ( Φ n ( α , β ) ) > n exp ( log n 104 log log n ) P(\Phi_n(\alpha,\beta))>n\exp\left(\frac{\log n}{104\log\log n}\right) P ( Φ n ( α , β )) > n exp ( 104 l o g l o g n l o g n ) ; a large single-power prime factor is compatible with the remaining mass being concentrated in high powers of small primes.
Bugeaud–Corvaja–Zannier [28 ] gives log gcd ( a n − 1 , b n − 1 ) = o ( n ) \log\gcd(a^n-1,b^n-1)=o(n) log g cd( a n − 1 , b n − 1 ) = o ( n ) : the common factor of two sequences, not the repeated-prime mass of one.
Balaji–Luca [25 ] gives # A K , c , u ∩ [ 1 , N ] ≪ N exp ( − log N 156 log log N ) \#A_{K,c,u}\cap[1,N]\ll N\exp\left(-\frac{\log N}{156\log\log N}\right) # A K , c , u ∩ [ 1 , N ] ≪ N exp ( − 156 l o g l o g N l o g N ) for the K n K n K n -smooth part, for most n n n ; smooth parts include first powers and do not exclude high powers.
Zsigmondy / Bilu–Hanrot–Voutier [29 ] guarantees one primitive divisor, not that all primitive divisors occur once. The paraphrase “p ∤ n p\nmid n p ∤ n implies p p p occurs at most once in Φ n ( r ) \Phi_n(r) Φ n ( r ) ” is false : v 1093 ( Φ 364 ( 2 ) ) = 2 v_{1093}(\Phi_{364}(2))=2 v 1093 ( Φ 364 ( 2 )) = 2 with 1093 ∤ 364 1093\nmid364 1093 ∤ 364 , and Φ 5 ( 3 ) = 121 = 11 2 \Phi_5(3)=121=11^2 Φ 5 ( 3 ) = 121 = 1 1 2 . Indeed all deep Wieferich primes satisfy q ∤ d q q\nmid d_q q ∤ d q (since d q ∣ q − 1 d_q\mid q-1 d q ∣ q − 1 ), so high exponents occur precisely at primitive divisors.
Murty–Séguin [8 ] gives conditional greatest-prime-factor bounds and “super-Wieferich finite ⇒ \Rightarrow ⇒ non-Wieferich infinite”; no upper bound on W r ( 3 ) ( x ) W_r^{(3)}(x) W r ( 3 ) ( x ) better than O ( x 2 ) O(x^2) O ( x 2 ) .
Fellini–Murty [20 ] , Silverman [6 ] , Graves–Murty [22 ] and Ding [22 ] give lower bounds for non-Wieferich primes or a b c abc ab c -conditional results; Ding’s C n ≫ a φ ( n ) − ε n C_n\gg a^{\varphi(n)-\varepsilon n} C n ≫ a φ ( n ) − ε n yields log rad Φ n ( a ) ≥ ( 1 − ε ) φ ( n ) log a \log\operatorname{rad}\Phi_n(a)\ge(1-\varepsilon)\varphi(n)\log a log rad Φ n ( a ) ≥ ( 1 − ε ) φ ( n ) log a , which controls M d M_d M d only with ε → 0 \varepsilon\to0 ε → 0 and a b c abc ab c .
W. Xie, Primary components and large prime factors of cyclotomic values , arXiv:2611.00075v1 (math.RA, 36 pp., preprint, not peer-reviewed). This is the closest same-object reference: Theorem 3.2 gives an unconditional lower bound for the primary part, but Theorem 3.6 takes log rad Φ r ( a ) ≥ q 1 − γ \log\operatorname{rad}\Phi_r(a)\ge q^{1-\gamma} log rad Φ r ( a ) ≥ q 1 − γ as a hypothesis and its counting input is ϑ ( Y ; r , 1 ) \vartheta(Y;r,1) ϑ ( Y ; r , 1 ) with a fixed conductor and an average over bases. It therefore supplies no unconditional log rad ≥ φ ( n ) log a − o ( n ) \log\operatorname{rad}\ge\varphi(n)\log a-o(n) log rad ≥ φ ( n ) log a − o ( n ) and does not transfer to the fixed-base, varying-modulus side; the verdict above is unchanged. Citation caveat: the GenPDF header prints “arXiv:2611.00075v1” followed by “[math.RA] 16 Oct 2026”; the “2611” identifier and the October date are incompatible at the metadata level. We cite the v1 PDF as deposited and do not assert a submission date.
The unconditional side has no lower bound of strength log rad Φ n ( a ) ≥ φ ( n ) log a − o ( n ) \log\operatorname{rad} \Phi_n(a) \ge\varphi(n)\log a - o(n) log rad Φ n ( a ) ≥ φ ( n ) log a − o ( n ) ; the strongest unconditional statements are log rad = O ( log n ) \log\operatorname{rad} = O(\log n) log rad = O ( log n ) type. (D4) has no standard name: Katz [2 ] is a continuous R / Z \mathbb{R}/\mathbb{Z} R / Z equidistribution conjecture, Gras [30 ] is a 1 / p 1/p 1/ p heuristic, and Berger [31 ] ’s formal-group rigidity treats fixed formal groups. If (D4) is read as a counting density it is already P 6 a P_{6a} P 6 a ; a probabilistic reading must specify its probability space (Proposition 14.5 ).
Remark 14.1 (Transcription check). Three paraphrases in the commissioned verification report were wrong and are corrected here; the body of this note is unaffected: (i) v q ( r q − 1 − 1 ) = s q v_q(r^{q-1}-1)=s_q v q ( r q − 1 − 1 ) = s q , not s q + v q ( q − 1 ) s_q+v_q(q-1) s q + v q ( q − 1 ) , since q ∤ ( q − 1 ) / d q q\nmid(q-1)/d_q q ∤ ( q − 1 ) / d q (Lemma 7.1 ); (ii) the translation r m o d q 3 ∈ μ q − 1 r\bmod q^3\in\mu_{q-1} r mod q 3 ∈ μ q − 1 of Proposition 8.9 stands (q ≤ 97 q\le97 q ≤ 97 enumeration, zero counterexamples); (iii) the “primitive divisors occur at most once” paraphrase is false (counterexamples above).
The power of structural constraints: zero
Theorem 14.2 (Zero saving). The small-q q q layer q ≤ t q\le\sqrt{t} q ≤ t is unconditionally o ( t ) o(t) o ( t ) , and for q > t q>\sqrt{t} q > t one has d q > c r log t d_q>c_r\log t d q > c r log t with c r = 3 / ( 2 log r ) c_r=3/(2\log r) c r = 3/ ( 2 log r ) and q ∤ d q q\nmid d_q q ∤ d q (no BOOK contamination, cf. the stratification trap above). Nevertheless every volume-type upper bound obtained from this split is strictly worse than the trivial volume bound:
C ( t ) − V ( t ) = t [ log r − 6 π 2 log ( r + 1 ) ] + O ( log t ) > 0 for every r ≥ 2 C(t)-V(t)=\sqrt{t}\left[\log r-\frac{6}{\pi^2}\log(r+1)\right]+O(\log t)>0\qquad\text{for every }r\ge2 C ( t ) − V ( t ) = t [ log r − π 2 6 log ( r + 1 ) ] + O ( log t ) > 0 for every r ≥ 2 (coefficient 0.0253 0.0253 0.0253 at r = 2 r=2 r = 2 , 0.2558 0.2558 0.2558 at r = 3 r=3 r = 3 , 1.7995 1.7995 1.7995 at r = 100 r=100 r = 100 , since log r > 6 π 2 log ( r + 1 ) \log r>\frac{6}{\pi^2}\log(r+1) log r > π 2 6 log ( r + 1 ) for all r ≥ 2 r\ge2 r ≥ 2 ). Hence all known structural constraints together save zero linear mass: the small-d d d structural emptiness of Remark 9.3 excludes only finitely many d d d and has no asymptotic force. Moreover, a saturated pseudo-configuration satisfying all per-d d d constraints 3 log q ≤ min ( φ ( d ) log ( r + 1 ) , d log r ) 3\log q\le\min(\varphi(d)\log(r+1),d\log r) 3 log q ≤ min ( φ ( d ) log ( r + 1 ) , d log r ) , q > t q>\sqrt{t} q > t , q ≡ 1 ( m o d d ) q\equiv1\pmod d q ≡ 1 ( mod d ) still gives A ( t ) / t = 0.552 / 0.775 / 1.051 / 1.234 A(t)/t=0.552/0.775/1.051/1.234 A ( t ) / t = 0.552/0.775/1.051/1.234 for r = 2 , 3 , 5 , 7 r=2,3,5,7 r = 2 , 3 , 5 , 7 at t = 10 3 t=10^3 t = 1 0 3 , stabilising at 0.561 / 0.783 / 1.058 / 1.243 0.561/0.783/1.058/1.243 0.561/0.783/1.058/1.243 at t = 10 4 t=10^4 t = 1 0 4 . Thus block/volume/partial-summation methods are exhausted (logical pseudo-configurations, not claims about real factor sets).
The ladder has two independent weakenings
Theorem 14.3 (A r ( t ) = o ( t ) A_r(t)=o(t) A r ( t ) = o ( t ) and P 4 ( 3 ) P^{(3)}_4 P 4 ( 3 ) are incomparable). Model M 1 M1 M 1 (d q = q − 1 d_q=q-1 d q = q − 1 , s q = 3 s_q=3 s q = 3 , all primes) gives, by Mertens’ first theorem ∑ p ≤ x log p p = log x + O ( 1 ) \sum_{p\le x}\frac{\log p}{p}=\log x+O(1) ∑ p ≤ x p l o g p = log x + O ( 1 ) , A ( t ) ≈ 3 log t = o ( t ) A(t)\approx3\log t=o(t) A ( t ) ≈ 3 log t = o ( t ) while P 4 ( 3 ) ≈ log t → ∞ P^{(3)}_4\approx\log t\to\infty P 4 ( 3 ) ≈ log t → ∞ ; model M 2 M2 M 2 (one saturated deep prime per d d d with φ ( d ) / d ≥ log r / log ( r + 1 ) \varphi(d)/d\ge\log r/\log(r+1) φ ( d ) / d ≥ log r / log ( r + 1 ) ) gives P 4 ( 3 ) < ∞ P^{(3)}_4<\infty P 4 ( 3 ) < ∞ while A ( t ) = δ r t A(t)=\delta_rt A ( t ) = δ r t with δ r = 0.417 / 0.289 / 0.214 / 0.184 / 0.129 / 0.095 \delta_r=0.417/0.289/0.214/0.184/0.129/0.095 δ r = 0.417/0.289/0.214/0.184/0.129/0.095 for r = 2 , 3 , 5 , 7 , 19 , 100 r=2,3,5,7,19,100 r = 2 , 3 , 5 , 7 , 19 , 100 . Both satisfy the tight structural necessary condition. Hence the ladder
P 6 a ⇒ ( D 4 ) c o u n t ⇒ Ξ r ( 3 ) < ∞ ⇒ { A = o ( t ) } ∖ { P 4 ( 3 ) } P_{6a}\Rightarrow(\mathrm{D4})_{\mathrm{count}}\Rightarrow\Xi_r^{(3)}<\infty\Rightarrow\{A=o(t)\}\setminus\{P_4^{(3)}\} P 6 a ⇒ ( D4 ) count ⇒ Ξ r ( 3 ) < ∞ ⇒ { A = o ( t )} ∖ { P 4 ( 3 ) } has two independent weakenings, and Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ is strictly stronger than either.
The single-variable translation of P6a
Theorem 14.4 (P 6 a ⟺ m ( q ) → ∞ P_{6a}\Longleftrightarrow m(q)\to\infty P 6 a ⟺ m ( q ) → ∞ ). Let m ( q ) = min ( μ q − 1 ( Z / q 3 Z ) ∖ { 1 } ) m(q)=\min\left(\mu_{q-1}(\mathbb{Z}/q^3\mathbb{Z})\setminus\{1\}\right) m ( q ) = min ( μ q − 1 ( Z / q 3 Z ) ∖ { 1 } ) be the least nontrivial Teichmüller representative modulo q 3 q^3 q 3 . Then
P 6 a ⟺ m ( q ) → ∞ . P_{6a}\Longleftrightarrow m(q)\to\infty. P 6 a ⟺ m ( q ) → ∞.
Proof. If q > r 1 / 3 q>r^{1/3} q > r 1/3 then r < q 3 r<q^3 r < q 3 , so by Proposition 8.9 q ∈ W r ( 3 ) ⟺ r ∈ μ q − 1 q\in W_r^{(3)}\Longleftrightarrow r\in\mu_{q-1} q ∈ W r ( 3 ) ⟺ r ∈ μ q − 1 , which forces r ≥ m ( q ) r\ge m(q) r ≥ m ( q ) . Hence { q ∈ W r ( 3 ) : q > r 1 / 3 } ⊆ { q : m ( q ) ≤ r } \{q\in W_r^{(3)}:q>r^{1/3}\}\subseteq\{q:m(q)\le r\} { q ∈ W r ( 3 ) : q > r 1/3 } ⊆ { q : m ( q ) ≤ r } ; if m ( q ) → ∞ m(q)\to\infty m ( q ) → ∞ the right side is finite, and the finitely many primes q ≤ r 1 / 3 q\le r^{1/3} q ≤ r 1/3 add nothing: W r ( 3 ) W_r^{(3)} W r ( 3 ) is finite. Conversely, if m ( q ) ↛ ∞ m(q)\not\to\infty m ( q ) → ∞ , there is X X X with infinitely many q q q having m ( q ) ≤ X m(q)\le X m ( q ) ≤ X ; choosing for each such q q q an element of μ q − 1 ∖ { 1 } \mu_{q-1} \setminus\{1\} μ q − 1 ∖ { 1 } in [ 2 , X ] [2,X] [ 2 , X ] , the pigeonhole principle gives a fixed r ∈ { 2 , … , X } r \in\{2,\ldots,X\} r ∈ { 2 , … , X } lying in μ q − 1 \mu_{q-1} μ q − 1 mod q 3 q^3 q 3 for infinitely many q q q , all but finitely many with q > r 1 / 3 q > r^{1/3} q > r 1/3 ; by Proposition 8.9 again, W r ( 3 ) W_r^{(3)} W r ( 3 ) is infinite. □
Numerically, among the 1228 odd primes q < 10 4 q < 10^4 q < 1 0 4 the median of m ( q ) / q 2 m(q)/q^2 m ( q ) / q 2 is 0.7222 (so m ( q ) ≍ q 2 m(q) \asymp q^2 m ( q ) ≍ q 2 , the exact source of the 1 / q 2 1/q^2 1/ q 2 heuristic, since q − 1 q-1 q − 1 elements are spread over [ 0 , q 3 ) [0,q^3) [ 0 , q 3 ) ). The ten smallest ( q , m ( q ) ) (q,m(q)) ( q , m ( q )) are ( 7 , 18 ) , ( 3 , 26 ) , ( 23 , 42 ) , ( 5 , 57 ) , ( 113 , 68 ) , ( 11 , 124 ) , ( 17 , 158 ) , ( 13 , 239 ) , ( 47 , 295 ) , ( 19 , 333 ) (7,18),(3,26),(23,42),(5,57),(113,68),(11,124),(17,158),(13,239),(47,295),(19,333) ( 7 , 18 ) , ( 3 , 26 ) , ( 23 , 42 ) , ( 5 , 57 ) , ( 113 , 68 ) , ( 11 , 124 ) , ( 17 , 158 ) , ( 13 , 239 ) , ( 47 , 295 ) , ( 19 , 333 ) . For each fixed X ≤ 10 3 X \le10^3 X ≤ 1 0 3 the set { q : m ( q ) ≤ X } \{q:m(q)\le X\} { q : m ( q ) ≤ X } saturates already by Q = 300 Q=300 Q = 300 (3/5/12/15 primes for X = 50 / 100 / 500 / 1000 X=50/100/500/1000 X = 50/100/500/1000 , largest such $q=113); and for X = 10 4 X=10^4 X = 1 0 4 by Q = 3000 Q=3000 Q = 3000 (40 primes, largest $q=1483). This is the first fixed-X X X , all-r ≤ X r \le X r ≤ X numerical evidence for P 6 a P_{6a} P 6 a ; all previous scans were fixed-r r r , varying-q q q . Consequently the counting-density reading of (D4) is equivalent to P 6 a P_{6a} P 6 a , while the probabilistic form with finite mean C r / q 2 C_r/q^2 C r / q 2 is strictly stronger (yielding Ξ r ( 3 ) < ∞ \Xi_r^{(3)}<\infty Ξ r ( 3 ) < ∞ a.s.).
Dual exactness and the trivial family
Proposition 14.5 (Dual ensemble). In the dual ensemble (fixed q q q , r r r uniform in ( Z / q 3 ) × (\mathbb{Z}/q^3)^\times ( Z / q 3 ) × ) one has the exact identity P ( s q ≥ 3 ) = 1 / q 2 \mathbb{P}(s_q \ge3)=1/q^2 P ( s q ≥ 3 ) = 1/ q 2 , verified for q = 3 , 5 , 7 , 11 , 13 , 17 , 19 q=3,5,7,11,13,17,19 q = 3 , 5 , 7 , 11 , 13 , 17 , 19 *(∣ μ q − 1 ∣ = q − 1 |\mu_{q-1}|=q-1 ∣ μ q − 1 ∣ = q − 1 throughout). In short intervals r ≪ q 3 r \ll q^3 r ≪ q 3 the expectation is correct but the variance is large (at R = 100 R=100 R = 100 the counts for q = 11 , 13 , 17 , 19 q=11,13,17,19 q = 11 , 13 , 17 , 19 are all 0 against expectations 0.82, 0.59, 0.34, 0.27). In the primal ensemble (fixed r r r , q q q varying) there is no probability space; the missing tool is exactly the transfer dual→ \to → primal (cf. the ensemble separation of § 8.1 ). (D4) is therefore not a Katz/Gras theorem and has no standard name.
Theorem 14.6 (The trivial family is thin). d q = d d_q=d d q = d implies q 3 ∣ r d − 1 ≤ r d q^3 \mid r^d-1 \le r^d q 3 ∣ r d − 1 ≤ r d , hence q ≤ r d / 3 q \le r^{d/3} q ≤ r d /3 ; in particular d q = 1 d_q=1 d q = 1 forces q ≤ ( r − 1 ) 1 / 3 q \le(r-1)^{1/3} q ≤ ( r − 1 ) 1/3 , so the trivial family r ≡ 1 ( m o d q 3 ) r \equiv1 \pmod{q^3} r ≡ 1 ( mod q 3 ) contributes only finitely many q q q for each fixed r r r . A sweep over q < 5 ⋅ 10 6 q<5\cdot10^6 q < 5 ⋅ 1 0 6 (348512 odd primes) and 24 fixed bases (including the r = 1 + q 3 r=1+q^3 r = 1 + q 3 families 28, 126, 344, 1332, 2198 and large bases 10 4 10^4 1 0 4 , 10 5 10^5 1 0 5 ), 8.36·10 6 10^6 1 0 6 congruence tests, yields 15 deep hits, all with q ∈ { 3 , 5 , 7 , 11 , 13 } q\in\{3,5,7,11,13\} q ∈ { 3 , 5 , 7 , 11 , 13 } , d q ≤ 6 d_q\le6 d q ≤ 6 , s q ∈ { 3 , 4 , 5 } s_q\in\{3,4,5\} s q ∈ { 3 , 4 , 5 } ; the bound q ≤ r d / 3 q\le r^{d/3} q ≤ r d /3 has zero violations and the r = 1 + q 3 r=1+q^3 r = 1 + q 3 families give d q = 1 d_q=1 d q = 1 exactly. The observed A o b s ( t ) A_{\mathrm{obs}}(t) A obs ( t ) saturates for t ≥ 6 t\ge6 t ≥ 6 and is a lower bound truncated at q < Q q<Q q < Q : it is not a proof of A ( t ) = o ( t ) A(t)=o(t) A ( t ) = o ( t ) , since primes with d q ≤ t d_q\le t d q ≤ t may reach q = r t / 3 q=r^{t/3} q = r t /3 .
What the new tool must look like
The missing object must cash out a codimension-two double zero on a varying-modulus family : (D4) is not more codimension — the s q ≥ 3 s_q \ge3 s q ≥ 3 condition ( Q ( 1 ) , Q ( 2 ) ) = ( 0 , 0 ) (Q^{(1)}, Q^{(2)}) = (0, 0) ( Q ( 1 ) , Q ( 2 ) ) = ( 0 , 0 ) already has codimension 2, and the s q = 2 s_q = 2 s q = 2 layer (codimension 1) is closed only because the Hensel wall D 3 = d q q D_3 = d_q q D 3 = d q q donates the extra 1 / q 1/q 1/ q . The Hensel dividend can be spent once . Seven specifications:
S1 uniform in the varying modulus q q q , with pointwise error < q − 2 − ε < q^{-2-\varepsilon} < q − 2 − ε independent of disc L q \operatorname{disc} L_q disc L q (GRH–Chebotarev is unavailable: q q q ramifies in L q = Q ( ζ q 2 , r 1 / q 2 ) L_q = \mathbb{Q}(\zeta_{q^2}, r^{1/q^2}) L q = Q ( ζ q 2 , r 1/ q 2 ) , disc ∼ q O ( q 2 ) \operatorname{disc} \sim q^{O(q^2)} disc ∼ q O ( q 2 ) );
S2 produces A r ( t ) = o ( t ) A_r(t) = o(t) A r ( t ) = o ( t ) , equivalently Φ r ( E ) = o ( E ) \Phi_r(E) = o(E) Φ r ( E ) = o ( E ) (Theorem 9.6 ), not merely Ξ r ( 3 ) < ∞ \Xi_r^{(3)} < \infty Ξ r ( 3 ) < ∞ ;
S3 discrete, pointwise mod q 3 q^3 q 3 ; continuous R / Z \mathbb{R}/\mathbb{Z} R / Z equidistribution does not control 1 / q 1/q 1/ q point masses — equivalently, pointwise rigidity of ι ( Z ) \iota(\mathbb{Z}) ι ( Z ) in P \mathcal{P} P ;
S4 penetrates the index-q 2 q^2 q 2 Teichmüller subgroup μ q − 1 \mu_{q-1} μ q − 1 (Theorem 14.4 );
S5 layer-specific: isolate both the s q = 2 s_q = 2 s q = 2 PRIM layer and the BOOK term q ∣ d / d q q \mid d/d_q q ∣ d / d q (equivalently cut at q > E q > \sqrt{E} q > E );
S6 bypass finite-order congruence expansions (Theorem 9.5 ) or handle the varying-modulus discrete logarithm.
S7 No r r r -uniform output. By the CRT family of Remark 14.7 , any bound uniform in r r r is false; the constant C r C_r C r of (D4) must depend on r r r , and the tool must produce a statement for each fixed base.
Self-checks: the tool must not imply P 4 P_4 P 4 ; must give Φ = o ( E ) \Phi= o(E) Φ = o ( E ) rather than Ξ ( 3 ) < ∞ \Xi^{(3)} < \infty Ξ ( 3 ) < ∞ ; must control the 1 / d q 1/d_q 1/ d q -weighted mass, not N 3 ( E ) N_3(E) N 3 ( E ) alone. T1–T6: GRH-only, gcd/divisibility-identity-only, greatest-prime-factor/single-power/smooth-part/gcd-only, and structure-constraint/block/volume-only mechanisms all fail (see § 9 and §§§ 15.1 –§ 15.2 ). The minimal tool is a single deterministic statement, m ( q ) → ∞ m(q) \to\infty m ( q ) → ∞ (Theorem 14.4 ), or its quantitative form # { q ≤ x : m ( q ) ≤ X } = O X ( 1 ) \#\{q \le x : m(q) \le X\} = O_X(1) # { q ≤ x : m ( q ) ≤ X } = O X ( 1 ) .