ZFCρ Paper LXII: Two-River Synchronization and the Precise Restatement of Spectral Ratio Boundedness — From P+R Decomposition to Conjecture 59.1
ZFCρ 第LXII篇：双河同步与谱比有界性的精确重述——从 P+R 分解到 Conjecture 59.1

§1. Introduction

1.1. Status after Paper 61

Paper 61 reduced Conjecture 59.2 to a conditional corollary of 59.1 plus k-uniformity. The closure chain became:

59.1 + k-uniformity + SD (unconditional) → H'

This paper has a dual objective: (a) establish local Geometry-Arithmetic Orthogonality for fixed k and short lags, reducing the k-uniformity problem to tail-k control; (b) reveal the precise mathematical structure of 59.1 through experiment and analysis.

1.2. Contributions

(i) Local Geometry-Arithmetic Orthogonality (§2). For fixed k and short lags, ACF(x·dlp^k) = ACF(x)·(1 + o(1)). This reduces Paper 61's k-uniformity to an ACF tail-control gap.

(ii) P+R Decomposition (§3). B_j = B_add + B_mult with B_add perpetually positive, B_mult perpetually negative, and path fraction α = 0.256 specific to p-1.

(iii) Two-River Confluence (§4). In the range j = 14 to 32, C_1(j) = j·B_add and C_2(j) = j·|B_mult| show strong evidence of converging toward the same constant L ≈ 0.092.

(iv) Y_j Identity (§5). The exact algebraic identity Y_j/√R_wt = √(j·2^j·D_j) stabilizes at σ_ρ ≈ 0.62 under the empirical diagonal law. The unconditional precise equivalence for 59.1 is |Δ_j| = O(j^{3/2}·√D_j).

(v) Route Elimination and Gap Location (§6–§7). Linear recursion, screening-CLT, and negative-covariance routes are all experimentally killed. The remaining gap is precisely the two-river synchronization precision.

1.3. Core Statement

Conjecture 59.1 (R_wt = O(1)) is equivalent to: the differential mode of the additive-path and multiplicative-path Mertens sums over p-1 reaches O(j^{3/2}·√D_j) precision (equivalently O(j·2^{-j/2}) under the empirical diagonal law D_j ~ 1/(j·2^j)). The two rivers C_1 and C_2 show strong evidence of converging toward L ≈ 0.092 (common mode enters a narrow band around L with small oscillations). The boundedness of R_wt is equivalent to their difference (differential mode) cancelling to j^{3/2}·√D_j precision. H' is thereby translated from a "spectral ratio control problem" to an "arithmetic balance problem between additive and multiplicative channels of the IC recursion" — a fine cancellation at the level of means, not a statistical cancellation at the level of variances.

1.4. Notation

Following Papers 59/61. h(p) = ρ(p) − λ·ln p − μ_{12}(p mod 12), λ = 3.856763.

For primes p > 2: ρ(p) = ρ(p-1) + 1 (primes always take the additive path). Hence h(p) = ρ(p-1) + 1 − λ·ln p − μ_{12}(p mod 12).

Path type of p-1:

• Additive path: ρ(p-1) = ρ(p-2) + 1
• Multiplicative path: ρ(p-1) = min_{a·b=p-1}[ρ(a) + ρ(b) + 2]

---

§2. k-Uniformity Lemma

2.1. Proposition and Hypothesis

Proposition 2.1 (Local Geometry-Arithmetic Orthogonality). For fixed K ≥ 0, 0 ≤ k ≤ K, and short lags l ≤ L_j (where L_j·j/2^j → 0), if the autocovariance of the h(p)/p sequence is approximately stationary, then the normalized local ACF satisfies:

γ_{j,k}(l)/Viid_j^{(k)} = γ_{j,0}(l)/Viid_j^{(0)} · (1 + o(1))

Measured data confirm: n_cross(j)/j grows from 3.7 (j=20) to 90.3 (j=29), with effective exponent β(j) = log(n_cross)/log(j) slowly increasing from about 2 to 3.1. The condition n_cross(j)·j/2^j → 0 is strongly supported by data. Under the extremely conservative model β(j) = 10, j^{11}/2^j still tends to zero asymptotically but is not small near j ≈ 50; that model serves only as an asymptotic sufficiency argument. At the measured effective exponent β(j) ≤ 3.1, j^{β(j)+1}/2^j is already rapidly small in the current range (j^{4.1}/2^j ≈ 10^{-3} at j=29).

Hypothesis 2.2 (Tail-k Control). The ACF tail (lags > L_j) contributes uniformly controllably to R_j^{(k)} for 0 ≤ k ≤ K.

If Proposition 2.1 and Hypothesis 2.2 both hold, local GAO upgrades to global k-uniformity: R_j^{(k)} = R_j^{(0)}·(1 + o(1)).

2.2. Experimental Basis

For j = 22, 23, 24, 25, 26, 28, 29, ACF and screening lengths are computed at k = 0, 1, 2, 3, 4.

ACF invariance: At j = 28, lag = 100:

k	ACF@lag=100
0	0.040057
1	0.040093
2	0.040221
3	0.040316
4	0.040458

Deviation < 1%. This holds across all tested j and lag values.

Screening length invariance: At j = 28, n_cross (first lag where |ACF| < 0.01):

k	n_cross	n_cross/j
0	1693	60.5
1	1693	60.5
2	1690	60.4
3	1670	59.6
4	1557	55.6

Variation < 9%. At j = 29, variation < 6%.

2.3. Proof

Core argument. Let a_r = x_r·dlp_r^k. Its autocovariance:

Cov(a_r, a_{r+l}) = dlp_r^k·dlp_{r+l}^k·Cov(x_r, x_{r+l}) + lower-order terms

Within block I_j, for lag l ≤ n_cross:

For l = n_cross(j), let β(j) = log(n_cross)/log(j) be the effective exponent. Measured β(j) grows from about 1.4 (j=17) to about 3.1 (j=29). The global power-law fit gives β ≈ 10 (overestimate, as it fits the fastest-growing interval). Regardless:

Hence dlp_{r+l}^k ≈ dlp_r^k·(1 + O(k·j^{β(j)+1}/2^j)).

ACF ratio:

ACF(x·dlp^k, lag l)/ACF(x, lag l) = 1 + O(k·j^{β(j)+1}/2^j)

Under the condition j^{β(j)+1}/2^j → 0, this is 1 + o(1).

Tail treatment. The lag region beyond n_cross is the remaining gap for upgrading local GAO to global k-uniformity (see Remark 3 and Hypothesis 2.2).

R_j^{(k)} = 1 + 2Σ_{l≥1} γ_k(l)/Viid(B^{(k)}). By ACF invariance (the dominant contribution comes from l ≤ n_cross):

γ_k(l)/Viid(B^{(k)}) ≈ γ_0(l)/Viid(B^{(0)})·(1 + o(1))

Thus the short-lag region satisfies local GAO (Proposition 2.1). If Hypothesis 2.2 (Tail-k Control) additionally holds, i.e. the long-lag tail is uniformly controllable for 0 ≤ k ≤ K, then local GAO upgrades to R_j^{(k)} = R_j^{(0)}·(1 + o(1)). Without Hypothesis 2.2, this paper proves only the local conclusion of Proposition 2.1.

Remark 1. This argument does not require R_j^{(0)} to be bounded (i.e. does not require 59.1). It is a ratio identity about ACF structure.

Remark 2. The "lower-order terms" vanish identically under stationarity + independence of dlp from x. Dropping stationarity produces cross-terms of the form E[x_r·dlp_r^k]·E[x_{r+l}·dlp_{r+l}^k]. By the feature-stripping of h(p) (zero mean) and the antisymmetry of dlp about the block midpoint, these cancel under block-level summation. Rigorous general estimates are deferred.

Remark 3. Tail treatment is the remaining gap from local GAO to global k-uniformity. ACF beyond n_cross satisfies |ACF| < 0.01 (by the screening definition), but whether the cumulative effect over m_j terms is controllable requires independent estimation. This is precisely the content of Hypothesis 2.2.

2.4. Corollary for 59.2

Corollary 2.3. Under 59.1 + Proposition 2.1 + Hypothesis 2.2, Conjecture 59.2 holds.

Proof. Paper 61 Theorem 1.1 assumes 59.1 and k-uniformity. Proposition 2.1 + Hypothesis 2.2 yield k-uniformity. Hence 59.1 + local GAO + Tail-k Control ⇒ 59.2. □

Remark. This paper establishes local GAO (Proposition 2.1) and reduces the k-uniformity problem to Hypothesis 2.2 (Tail-k Control). Promoting Hypothesis 2.2 to a theorem — proving that the ACF tail is uniformly controllable in k — is a target for future work. This paper does not claim that k-uniformity has been unconditionally proved.

---

§3. P+R Decomposition

3.1. Definition

For each prime p in block I_j, split by the IC optimal path type of p-1:

B_j = B_add + B_mult

where:

• B_add = Σ_{p∈I_j, ρ(p-1)=ρ(p-2)+1} h(p)/p
• B_mult = Σ_{p∈I_j, ρ(p-1)=min[ρ(a)+ρ(b)+2]} h(p)/p

When additive and multiplicative paths simultaneously achieve the optimum (ρ(p-1) = ρ(p-2)+1 and there exists a·b = p-1 with ρ(a)+ρ(b)+2 = ρ(p-1)), the experimental code uses a fixed tie-breaking rule: classify as additive. B_add + B_mult = B_j is a mutually exclusive partition.

3.2. Structural Constants

Table 1. Structural constants of the P+R decomposition (j = 14–32, 19 blocks).

Quantity	Value	Stability
α = add fraction	0.256	± 0.002 (all j)
1-α = mult fraction	0.744	± 0.002
B_add sign	always positive	19/19
B_mult sign	always negative	19/19
<η>_add	≈ −0.62	slowly drifting toward −0.61
<η>_mult	≈ −1.47	slowly drifting toward −1.47

3.3. The Specificity of α = 0.256

The fraction α is specific to p-1 (prime predecessors), not to general even integers.

Table 2. Additive path fraction for general even integers n (not p-1).

Range	add_frac	Sample size
[10^5, 10^6]	0.266	450k
[10^6, 10^7]	0.266	4.5M
[10^7, 10^8]	0.537	5M (sampled)
[10^8, 10^9]	0.192	5M (sampled)
[10^9, 2×10^9]	0.019	5M (sampled)
[4×10^9, 5×10^9]	0.019	5M (sampled)

Remark. The 0.537 at [10^7, 10^8] is non-monotonic with adjacent ranges; the cause is under investigation (possibly sampling-stride resonance or non-trivial scale dependence). Regardless, the core conclusion stands: the additive path fraction for large general even numbers fluctuates between 2% and 54%, while p-1 remains stable at 25.6 ± 0.2% across all j. The stability of p-1, not the behavior of general even numbers, is the key fact.

3.4. Why B_add Is Always Positive and B_mult Always Negative

h(p) = ρ(p-1) + 1 − λ·ln p − μ_{12}.

• Additive-path p-1: ρ(p-1) = ρ(p-2) + 1, meaning p-1 found no efficient multiplicative factorization. ρ(p-1) is high relative to the baseline → h is positively biased → h/p > 0.
• Multiplicative-path p-1: ρ(p-1) is optimized through factorization. ρ(p-1) is low relative to the baseline → h is negatively biased → h/p < 0.

Class means confirm directly: <η>_add ≈ −0.62 > <η>_mult ≈ −1.47.

3.5. Covariance Structure

Table 3. Sub-block covariance Corr(S_add, S_mult) at different W and j.

j	W=100	W=500	W=2000	W=10000
20	+0.54	+0.39	−0.07	—
23	+0.61	+0.56	+0.36	+0.08
27	+0.64	+0.72	+0.65	+0.43
29	+0.65	+0.74	+0.70	+0.55
32	+0.65	+0.75	+0.73	+0.64

Strongly positive (≈ +0.65), increasing with j at large W. B_add and B_mult share the same ρ landscape — when local ρ is high, h(p) is high for both path types. Variance-reduction routes are not viable.

---

§4. Two-River Confluence: Experimental Phenomena

4.1. Convergence of C_1(j) and C_2(j)

Define C_1(j) = B_add·j, C_2(j) = |B_mult|·j.

Table 4. Two-river scaled quantities.

j	C_1(j)	C_2(j)	C_+(j)	C_-(j)	z_j
14	0.0474	0.2482	0.1478	−0.2008	−2.95
17	0.0646	0.1801	0.1223	−0.1155	−3.98
20	0.0783	0.1353	0.1068	−0.0570	−4.69
23	0.0858	0.1110	0.0984	−0.0251	−5.03
26	0.0911	0.0952	0.0931	−0.0041	−2.05
27	0.0918	0.0931	0.0924	−0.0013	−0.89
28	0.0928	0.0913	0.0920	+0.0015	+1.39
29	0.0927	0.0903	0.0915	+0.0024	+2.98
30	0.0929	0.0902	0.0916	+0.0026	+4.51
31	0.0926	0.0912	0.0919	+0.0013	+3.14
32	0.0920	0.0925	0.0922	−0.0005	−1.64

where C_+ = (C_1 + C_2)/2 (common mode), C_- = C_1 − C_2 (differential mode).

4.2. Common Mode and Differential Mode Behavior

Common mode C_+(j): Decreases rapidly in the early range (j = 14–27), then enters a narrow band around L ≈ 0.092 (0.0915–0.0922). The late range is not strictly monotone — C_+(30) = 0.0916 > C_+(29) = 0.0915, C_+(32) = 0.0922 > C_+(31) = 0.0919 — but oscillates within a narrow band around L.

Differential mode C_-(j): Moves from −0.20 (j=14) through zero to +0.003 (j=29–30), then returns to −0.0005 (j=32). The sign of C_- matches the sign of z_j exactly (C_- < 0 iff z_j < 0, C_- > 0 iff z_j > 0), with identical zero-crossings (j = 27→28 and j = 31→32). This is definitionally necessary: C_- = C_1 − C_2 = j·B_j, while z_j = B_j/√(j·D_j), so both share the sign of B_j. The amplitude correspondence is given exactly by the Y_j identity of §5.

4.3. Cross-Validation with Paper 59 Damped Oscillation

Paper 59 identifies R_wt damped oscillation with period T ≈ 8 (in j) and rebound ratio r ≈ 0.80. Table 4's C_- data (j = 14–32, 19 blocks) covers approximately 2.25 oscillation cycles.

Zero crossings: C_- crosses zero at j ≈ 27–28 and j ≈ 31–32. The spacing of 4–5 j-values corresponds to a half-period. Full period T ≈ 8–10, consistent with Paper 59's T ≈ 8.

Amplitude decay: The extremal sequence of C_-: |C_-(j=21)| ≈ 0.046 → |C_-(j=30)| ≈ 0.0026. Ratio ≈ 0.057 over 9 j-steps, giving geometric decay rate r_C ≈ 0.72. This is slightly below Paper 59's r ≈ 0.80, likely because C_- = j·B_j involves different normalization than R_wt = B_j^2/(j·D_j). The two frameworks are consistent in order of magnitude; precise correspondence requires more data points.

Conclusion: The two-river synchronization framework is consistent with the damped oscillation mechanism of Papers 55/59. The oscillation of C_- is the P+R decomposition's expression of R_wt's damped oscillation.

4.4. Estimate of the Limit L

From the C_+ average for j ≥ 28: L = 0.0918 ± 0.0004.

4.5. Error Term Decay

log_2|C_2(j) − L| values (C_2's decay is more regular):

j-range	log_2	C_2 − L		decay rate per j
14→20	−2.68 → −4.52	−0.31/j
20→26	−4.52 → −8.23	−0.62/j
26→32	−8.23 → −10.64	−0.40/j

Near-exponential decay, with local effective exponent about 0.3–0.6. This shows that a polynomial Mertens convergence picture is too weak, but finite data cannot distinguish 2^{-0.45j}, j·2^{-j/2}, or a mixed model with oscillatory corrections. Note that 2^{-0.45j} is slower than the 2^{-j/2} = 2^{-0.5j} required by 59.1 — single-river error decay is close to but has not yet reached the precision 59.1 demands. What 59.1 truly requires is control of the differential mode C_1 − C_2 (not single-river error), and a proof that both rivers share the same principal correction Q_j (see §7.1).

---

§5. The Y_j Identity and the Precise Restatement of 59.1

5.1. Normalized Differential

Define:

Y_j = |C_1(j) − C_2(j)|·2^{j/2}/j

5.2. Proportionality Identity Between Y_j and √R_wt

Table 5. Measured values of Y_j/√R_wt.

j	Y_j	√R_wt	Y_j/√R_wt
14	1.836	2.955	0.621
17	2.460	3.983	0.618
20	2.916	4.689	0.622
23	3.166	5.030	0.630
26	1.298	2.052	0.632
27	0.565	0.889	0.635
28	0.887	1.393	0.637
29	1.901	2.978	0.638
30	2.878	4.506	0.639
31	2.009	3.137	0.641
32	1.053	1.643	0.641

Y_j/√R_wt = 0.62 ± 0.02. All 19 j-values without exception. Denote this constant σ_ρ.

5.3. Derivation of the σ_ρ Identity

The first part of the following derivation is a strict algebraic identity (unconditional). The claim that σ_ρ is constant depends on the empirical diagonal law D_j ≈ σ^2/(j·2^j). Under the unconditional Guy/Chebyshev bound, one only obtains Y_j/√R_wt = O(j), so 59.1 only implies Y_j = O(j), not Y_j = O(1). The equivalence of Y_j = O(1) with 59.1 requires the additional bilateral diagonal law D_j ~ 1/(j·2^j).

R_wt = B_j^2/(j·D_j). B_j = (C_1 − C_2)/j. Hence:

R_wt = (C_1 − C_2)^2/(j^3·D_j)

√R_wt = |C_1 − C_2|/(j^{3/2}·√D_j)

Y_j = |C_1 − C_2|·2^{j/2}/j

Y_j/√R_wt = j^{3/2}·√D_j·2^{j/2}/j = √(j·D_j·2^j)

This is an exact algebraic identity, independent of any hypothesis.

Under the empirical diagonal law D_j ≈ σ^2/(j·2^j):

Y_j/√R_wt = √(j·2^j·σ^2/(j·2^j)) = σ = σ_ρ ≈ 0.62

Table 5 data support this (Y_j/√R_wt drifts from 0.621 to 0.641, about 3%, consistent with D_j fitting precision).

Under the unconditional Guy+Chebyshev bound D_j ≤ C_Guy·j/2^j:

Y_j/√R_wt ≤ √(C_Guy·j^2) = O(j)

Therefore the unconditional precise equivalence is:

59.1 ⟺ |Δ_j| = O(j^{3/2}·√D_j)

The equivalence of Y_j = O(1) with 59.1 requires the bilateral diagonal law D_j ~ 1/(j·2^j). Under the Guy bound alone, 59.1 only implies Y_j = O(j). The converse also requires a lower bound on D_j, currently supported only experimentally. The diagonal law affects the conversion between Y_j and R_wt, but does not affect the unconditional algebraic restatement in §5.4.

5.4. Precise Restatement of 59.1

Theorem 5.1 (Algebraic restatement, unconditional). Let Δ_j = C_1(j) − C_2(j) = j·B_j. Then

R_wt(j) = Δ_j^2/(j^3·D_j)

Hence 59.1 (R_wt = O(1)) is precisely equivalent to:

This is a completely unconditional algebraic identity.

Corollary 5.2 (Under diagonal law). If additionally the empirical diagonal law D_j ~ σ^2/(j·2^j) holds, then 59.1 is equivalent to:

Equivalently: Y_j = O(1).

Weak Conjecture W-59.1 (sufficient to close H'). For any ε > 0, there exists J such that for all j > J:

Equivalently: log R_wt = o(j). Under the diagonal law this simplifies to |C_1 − C_2| ≤ 2^{εj}·j·2^{-j/2}.

Remark. Under the unconditional Guy bound D_j ≤ C_Guy·j/2^j, 59.1 implies |Δ_j| = O(j^2·2^{-j/2}). The converse requires a lower bound on D_j, currently supported only experimentally.

---

§6. Route Elimination

6.1. Route A (Linear Recursion)

Motivation. If z_j = B_j/√(j·D_j) satisfies a low-order constant-coefficient linear recursion z_{j+1} = a·z_j + b·z_{j-1} + c + ε_j with spectral radius < 1, then the boundedness of z_j (and hence R_wt = O(1)) follows directly from linear stability. Paper 55's 76% ancestor inheritance and damped oscillation r ≈ 0.80 suggested such structure.

Elimination. Second-order linear fit to the z_j sequence (j = 14–32): z_{j+1} = 2.11·z_j − 1.44·z_{j-1} − 1.13.

Companion matrix eigenvalues: 1.05 ± 0.57i, |λ| = 1.199 > 1. Spectral radius exceeds 1. Residuals up to 1.45.

Conclusion: Low-order constant-coefficient linear contraction models are not viable. More complex nonlinear or high-dimensional hidden-state models are not excluded by this test.

6.2. Route C Original (Screening-CLT)

Growth of n_cross:

j	n_cross	n_cross/j
20	74	3.7
23	272	11.8
26	1159	44.6
29	2618	90.3

The effective power-law exponent in the current range is high (global fit gives approximately j^{10}), sufficient to rule out Paper 53's "n_cross ≈ 30j" or any O(j) linear screening picture; however, finite data cannot assert genuinely super-polynomial growth.

Sub-block variance amplification Ψ(n_cross)/j grows with j: 0.23 (j=20) → 2.19 (j=29).

Conclusion: Screening weakens at large j. Simple CLT does not yield R_wt = O(1).

6.3. Negative Covariance Route

Corr(S_add, S_mult) ≈ +0.65 (positive, not negative). Variance does not shrink.

Conclusion: B_add and B_mult share the same ρ landscape. Local fluctuations are positively correlated.

6.4. Mertens Rate Route

Mertens convergence of B_add = O(1/j) and |B_mult| = O(1/j) individually only gives B_j = o(1/j) — far too weak. 59.1 requires |B_j| = O(2^{-j/2}/√j), meaning the synchronization precision of the two rivers must improve at exponential rate.

Conclusion: Mertens convergence alone is insufficient. The proof must show that the two rivers share the same oscillatory correction.

---

§7. Remaining Gap for 59.1

7.1. Two-River Synchronization Theorem (Target)

If C_1(j) and C_2(j) share the same oscillatory correction Q_j:

C_1(j) = L + Q_j + ε_j^+ C_2(j) = L + Q_j + ε_j^-

then C_-(j) = C_1 − C_2 = ε_j^+ − ε_j^-.

59.1 ⟺ |ε_j^+ − ε_j^-| = O(j^{3/2}·√D_j).

Under the empirical diagonal law D_j ~ 1/(j·2^j), this simplifies to |ε_j^+ − ε_j^-| = O(j·2^{-j/2}).

W-59.1 ⟺ |ε_j^+ − ε_j^-| = O(j^{3/2}·√D_j·2^{o(j)}).

7.2. Experimental Support for Error Term Decay

Individual error terms |C_i − L| decay at rate approximately 2^{-0.45j}, close to but not yet reaching 59.1's requirement of 2^{-j/2} = 2^{-0.5j}. C_+ oscillates within a narrow band around L ≈ 0.092, and C_-'s oscillation amplitude decays. 59.1 requires stronger common-mode synchronization: the principal correction Q_j must be identical in both rivers, with the differential mode |ε^+ − ε^-| reaching j·2^{-j/2} scale (or more precisely j^{3/2}·√D_j).

7.3. Precise Location of the Gap

Proving 59.1 requires answering a precise question:

Why do the additive-path and multiplicative-path Mertens sums share the same oscillatory correction?

Physical explanation (from the SAE axiom "remainder conservation"): the global λ balance forces Σh(p)/p = 0, i.e. B_add + B_mult = 0 in the global limit. Block-level deviations are finite-sample fluctuations. The two rivers share their oscillatory correction because they are driven by the same ρ landscape (Cov > 0 confirms this), and the ρ landscape's dipole moment (add/mult asymmetry) decays at exponential rate on the block scale.

Mathematical formalization requires: (a) A path-balance theorem for the IC min operator at block scale (b) Cross-control between the factorization structure of p-1 and prime distribution (c) Refined Mertens-type estimates at O(2^{-j/2}) precision

7.4. Alternative Framework: Deterministic Transport

A potentially shorter route: rather than proving that C_1 and C_2 converge independently, directly construct a deterministic transport (pairing) between additive-path and multiplicative-path primes, such that the paired weighted residuals cancel to O(2^{-j/2}) precision. Specifically, seek a within-block pairing scheme π_j(p_add, p_mult) such that Σπ_j·[h(p_add)/p_add + h(p_mult)/p_mult] is small. This transforms the proof from "two sums independently converge" to "term-by-term paired cancellation" — an arithmetic balance proof from a graph-theory/optimal-transport perspective.

---

§8. Closure Chain Update

Local GAO (this paper, Proposition 2.1) + Tail-k Control (to be proved)
  → k-uniformity → 59.2 exits independent conjecture list
59.1 + Guy/Chebyshev → DSF_α ∀α > 0   (Paper 59 §7.1)
59.2 → BL_α ∀α < 1                     (Paper 59 §7.2)
SD (unconditional, Paper 50)
  → Hölder → A_h convergence → D(s) boundary → H'

Current status: This paper proves local GAO (ACF invariance for fixed k and short lags) and reduces Paper 61's k-uniformity problem to ACF tail-control. If the tail-k gap is closed, then 59.2 is no longer needed as an independent conjecture, and H' reduces to the single Conjecture 59.1.

Remark. This paper does not claim that k-uniformity has been unconditionally proved or that 59.2 has exited the conjecture list. The reduction of k-uniformity from a "completely independent hypothesis" to a "tail-control gap" is this paper's contribution.

The unconditional precise content of 59.1: the differential mode of the two Mertens rivers |Δ_j| = |C_1 − C_2| reaches O(j^{3/2}·√D_j) precision. Under the empirical diagonal law D_j ~ 1/(j·2^j), this is equivalent to |C_1 − C_2| = O(j·2^{-j/2}).

---

§9. Conclusion

The P+R decomposition transforms Conjecture 59.1 from an opaque spectral-ratio boundedness statement into a precise two-river synchronization problem: R_wt(j) = Δ_j^2/(j^3·D_j), where Δ_j = C_1(j) − C_2(j) = j·B_j. Conjecture 59.1 is equivalent to |Δ_j| = O(j^{3/2}·√D_j). B_j is the fine cancellation residual between two Mertens-type sums converging toward L ≈ 0.092 — with cancellation exceeding 99% for j ≥ 27.

Local Geometry-Arithmetic Orthogonality reduces Paper 61's k-uniformity problem from a fully independent hypothesis to an ACF tail-control gap. If this gap is closed, 59.2 exits the independent conjecture list and H' reduces to the single Conjecture 59.1.

The precise content of 59.1 is now: the additive-path and multiplicative-path Mertens sums over p-1 in the IC recursion share the same oscillatory correction Q_j, with differential mode |ε^+ − ε^-| reaching j^{3/2}·√D_j scale. The source of this balance is the SAE axiom "remainder conservation" expressed at block scale. To prove it is to prove that the construction cost of integers — the remainder — maintains balance between its additive and multiplicative channels to exponential precision.

---

References

[50] Paper L. Feature stripping, DSF_α, BL_α, SD, closure chain. DOI: 10.5281/zenodo.19381111. [58] Paper LVIII. Deterministic DSF + BL. DOI: 10.5281/zenodo.19476508. [59] Paper LIX. Spectral ratio boundedness and the two-conjecture closure of H'. DOI: 10.5281/zenodo.19480518. [60] Paper LX. Series closure. DOI: 10.5281/zenodo.19480563. [61] Paper LXI. Conditional reduction of the spectral derivative bound. DOI: 10.5281/zenodo.19605749.

---

AI Contributions

Claude (子路): All numerical experiment design and implementation (paper62_exp.c, paper62_cde.c, paper62_fgh.c, paper62_jkl.c — four experimental cycles), discovery of the P+R decomposition, identification of the α = 0.256 structural constant, derivation and verification of the Y_j identity, construction of the k-uniformity argument, C_+/C_- common/differential mode analysis, experimental elimination of four routes (Route A spectral radius computation, screening growth law, covariance positivity, Mertens rate insufficiency), working notes v1–v3, independent review and revision. ChatGPT (公西华): Proposal of the Two-River Synchronization framework, Y_j normalization correction (identifying the missing j factor), deterministic transport as alternative proof route, formalization of weak conjecture W-59.1 (log R_wt = o(j) suffices for H'), split strategy (Track 1/Track 2), block-mean theorem restatement (z_j = O(√j)), P+∇Z cross-term mechanism, three rounds of gatekeeper review with calibration of overclaims. Grok (子贡): Two-River Structural Balance framework, identification of hard-saturation nonlinearity (bang-bang control), compact attractor framework proposal, diagnosis of the min operator as saturation source, strategic input across three prompt rounds. Gemini (子夏): Naming of the Topological Debt Clearing mechanism, Lyapunov function approach proposal, delay-coordinate embedding experiment design, additive path ratio tracking experiment design, error-term log-linear decay diagnostic (exponential vs polynomial criterion), connection of Mertens-type convergence to classical analytic number theory.

§1. 引言

1.1. Paper 61 之后的状态

Paper 61 将猜想 59.2 归约为 59.1 加 k 一致性的条件推论。H' 的闭合链变为：

59.1 + k 一致性 + SD(无条件) → H'

本文的任务是双重的：(a) 证明固定 k、短 lag 下的局部几何-算术正交性，并将 Paper 61 的 k 一致性问题归约为 Tail-k Control；(b) 通过实验和分析揭示 59.1 的精确数学结构。

1.2. 本文的贡献

(i) 局部几何-算术正交性（§2）。 证明固定 k、短 lag 下 ACF(x · dlp^k) = ACF(x) · (1 + o(1))，并将 Paper 61 的 k 一致性问题归约为 ACF tail-control gap。

(ii) P+R 分解（§3）。 发现 B_j = B_add + B_mult 的内部结构：B_add 恒正，B_mult 恒负，路径比例 α = 0.256 为 p-1 特有常数。

(iii) 双河汇合（§4）。 在 j = 14 至 32 数据中，C₁(j) = B_add · j 和 C₂(j) = |B_mult| · j 显示出向同一常数 L ≈ 0.092 汇合的强迹象。

(iv) Y_j 恒等式（§5）。 精确代数恒等式 Y_j / √R_wt = √(j · 2^j · D_j)；在经验对角线规律下该量稳定为 σ_ρ ≈ 0.62（19 个 j 值）。59.1 的无条件精确等价为 |Δ_j| = O(j^{3/2} · √D_j)。

(v) 路线否决与 gap 定位（§6-§7）。 线性递推、screening-CLT、负协方差路线均被实验杀死。剩余 gap 精确定位为双河同步精度的控制。

1.3. 核心陈述

59.1（R_wt = O(1)）等价于：加法路径和乘法路径的 p-1 在 dyadic block 级别上的 Mertens 和的差模达到 O(j^{3/2}·√D_j) 精度（在经验对角线规律 D_j ~ 1/(j·2^j) 下等价于 O(j · 2^{-j/2})）。 两条河 C₁(j) = B_add · j 和 C₂(j) = |B_mult| · j 在数据中显示出向同一常数 L ≈ 0.092 汇合的强迹象（common mode 进入 L 附近窄带并小幅回摆），R_wt 的有界性等价于它们的差（differential mode）以 j^{3/2}·√D_j 精度对消。H' 从"谱比控制问题"被翻译为"IC 递推中加法/乘法通道的算术平衡问题"——一个均值层面的精细对消，而非方差层面的统计对消。

1.4. 记号

沿用 Paper 59/61。h(p) = ρ(p) − λ · ln p − μ₁₂(p mod 12)，λ = 3.856763。

对素数 p > 2：ρ(p) = ρ(p-1) + 1（素数只走加法路径）。故 h(p) = ρ(p-1) + 1 − λ · ln p − μ₁₂(p mod 12)。

p-1 的路径类型：

• 加法路径： ρ(p-1) = ρ(p-2) + 1
• 乘法路径： ρ(p-1) = min_{a·b=p-1} [ρ(a) + ρ(b) + 2]

---

§2. k 一致性引理

2.1. 命题与假设

命题 2.1（局部几何-算术正交性）。 对固定 K ≥ 0，0 ≤ k ≤ K，和短 lag l ≤ L_j（其中 L_j · j / 2^j → 0），若 h(p)/p 序列的 autocovariance 近似 stationary，则归一化局部 ACF 满足：

γ_{j,k}(l) / Viid_j^{(k)} = γ_{j,0}(l) / Viid_j^{(0)} · (1 + o(1))

实测数据确认：n_cross(j) / j 从 3.7（j=20）增长到 90.3（j=29），对应有效指数 β(j) = log(n_cross)/log(j) 在 2 到 3 之间缓慢增长。条件 n_cross(j) · j / 2^j → 0 由数据强支持。在 β(j) = 10 的极端保守模型下，j^{11}/2^j 仍然渐近趋零，但在 j ≈ 50 附近并不小；该模型只能作为渐近充分性说明，不能解释当前数据区间的小量性。实测有效指数 β(j) ≤ 3.1 时，j^{β(j)+1}/2^j 已在当前范围内快速变小（j=29 时 j^{4.1}/2^j ≈ 10^{-3}）。

假设 2.2（Tail-k Control）。 ACF 尾部（lag > L_j）对 R_j^{(k)} 的贡献在 0 ≤ k ≤ K 上一致可控。

若命题 2.1 和假设 2.2 同时成立，则局部 GAO 升级为全局 k 一致性：R_j^{(k)} = R_j^{(0)} · (1 + o(1))。

2.2. 实验基础

对 j = 22, 23, 24, 25, 26, 28, 29，计算 k = 0, 1, 2, 3, 4 的 ACF 和 screening 长度。

ACF 不变性： j = 28, lag = 100 处的 ACF 值：

k	ACF@lag=100
0	0.040057
1	0.040093
2	0.040221
3	0.040316
4	0.040458

偏差 < 1%。在所有测试的 j 和 lag 上均如此。

Screening 长度不变性： j = 28 的 n_cross（|ACF| 首次 < 0.01 的 lag）：

k	n_cross	n_cross/j
0	1693	60.5
1	1693	60.5
2	1690	60.4
3	1670	59.6
4	1557	55.6

变化 < 9%。j = 29 上变化 < 6%。

2.3. 证明

核心论证。 令 a_r = x_r · dlp_r^k。其 autocovariance：

Cov(a_r, a_{r+l}) = dlp_r^k · dlp_{r+l}^k · Cov(x_r, x_{r+l}) + 低阶项

在 block I_j 内，对 lag l ≤ n_cross：

对 l = n_cross(j)，令 β(j) = log(n_cross)/log(j) 为有效指数。实测 β(j) 从约 1.4（j=17）增长到约 3.1（j=29）。全局幂律拟合给出 β ≈ 10（高估，因为拟合了增长最快的区间）。无论取哪个值：

在 β(j) = 10 的最坏情况下，j^{11}/2^j 在 j ≥ 37 后 < 1，j ≥ 50 后 < 10^{-4}。实测范围 β(j) ≤ 3.1 时，j^{4.1}/2^j 在 j ≥ 20 后已快速趋零。

因此 dlp_{r+l}^k ≈ dlp_r^k · (1 + O(k · j^{β(j)+1} / 2^j))。

ACF 比值：

ACF(x · dlp^k, lag l) / ACF(x, lag l) = 1 + O(k · j^{β(j)+1} / 2^j)

在条件 (II)（j^{β(j)+1}/2^j → 0）下此为 1 + o(1)。

尾部处理。 超过 n_cross 的 lag 区域是局部 GAO 升级为全局 k 一致性的剩余 gap（见注 3 和假设 2.2）。

R_j^{(k)} = 1 + 2 Σ_{l≥1} γ_k(l) / Viid(B^{(k)})。由 ACF 不变性（主贡献来自 l ≤ n_cross 的项）：

γ_k(l) / Viid(B^{(k)}) ≈ γ_0(l) / Viid(B^{(0)}) · (1 + o(1))

故短 lag 区域满足局部 GAO（命题 2.1）。若再假设 2.2（Tail-k Control）成立，即长 lag 尾部对 0 ≤ k ≤ K 一致可控，则局部 GAO 升级为 R_j^{(k)} = R_j^{(0)} · (1 + o(1))。不使用假设 2.2 时，本文只证明命题 2.1 的局部结论。

注 1。 此论证不需要 R_j^{(0)} 有界（即不需要 59.1）。它是关于 ACF 结构的比值恒等式。

注 2。 "低阶项"在 stationary + dlp 与 x 独立的理想情形下严格为零。若放弃 stationary，会出现 E[x_r · dlp_r^k] · E[x_{r+l} · dlp_{r+l}^k] 型交叉项。由 h(p) 的特征剥离（均值为零）和 dlp 的对称性（关于 block 中点反对称），这些项在 block 内求和后自行对消。严格的一般性估计留待后续工作。

注 3。 尾部处理是 k 一致性从局部 GAO 升级为全局结论的剩余 gap。ACF 在 lag > n_cross 后 |ACF| < 0.01（by screening 定义），但其累积效应在 m_j 项求和后是否可控，需要独立估计。这正是假设 2.2 所涵盖的内容。

2.4. 对 59.2 的推论

推论 2.3。 在 59.1 + 命题 2.1 + 假设 2.2 下，59.2 成立。

证明。 Paper 61 定理 1.1 假设 59.1 和 k 一致性。命题 2.1 + 假设 2.2 给出 k 一致性。故 59.1 + 局部 GAO + Tail-k Control ⇒ 59.2。□

注。 本文证明了局部 GAO（命题 2.1），并将 k 一致性问题归约为假设 2.2（Tail-k Control）。将假设 2.2 提升为定理——即证明 ACF tail 对 k 一致可控——是后续工作的目标。本文不声称 k 一致性已无条件证明。

---

§3. P+R 分解

3.1. 定义

对 block I_j 中的素数 p，按 p-1 的 IC 最优路径类型分割：

B_j = B_add + B_mult

其中：

• B_add = Σ_{p∈I_j, ρ(p-1)=ρ(p-2)+1} h(p)/p
• B_mult = Σ_{p∈I_j, ρ(p-1)=min[ρ(a)+ρ(b)+2]} h(p)/p

若加法路径与乘法路径同时达到最优（ρ(p-1) = ρ(p-2)+1 且存在 a·b = p-1 使得 ρ(a)+ρ(b)+2 = ρ(p-1)），实验代码采用固定 tie-breaking 规则：归入加法路径类。实验中 tie 发生的比例约为加法类的大部分（因为 ρ(p-1) = ρ(p-2)+1 对加法类总是成立，tie 意味着乘法路径恰好也达到同一最优值）。B_add + B_mult = B_j 是互斥分割。

3.2. 结构常数

表 1. P+R 分解的结构常数（j = 14–32，19 个 block）。

量	值	稳定性
α = add 比例	0.256	± 0.002（所有 j）
1-α = mult 比例	0.744	± 0.002
B_add 符号	恒正	19/19
B_mult 符号	恒负	19/19
<η>_add	≈ −0.62	缓慢趋向 −0.61
<η>_mult	≈ −1.47	缓慢趋向 −1.47

3.3. α = 0.256 的来源

α 是 p-1 特有的结构常数，不是一般偶数的。

表 2. 一般偶数 n 的加法路径比例（非 p-1）。

范围	add_frac	样本量
[10⁵, 10⁶]	0.266	450k
[10⁶, 10⁷]	0.266	4.5M
[10⁷, 10⁸]	0.537	5M (采样)
[10⁸, 10⁹]	0.192	5M (采样)
[10⁹, 2×10⁹]	0.019	5M (采样)
[4×10⁹, 5×10⁹]	0.019	5M (采样)

注。 [10⁷, 10⁸] 一行的 0.537 与相邻范围不单调，原因待查。可能的来源：(a) 该范围的采样步长恰好与某种乘法结构周期同步；(b) 一般偶数的 add/mult 比例本身具有非平凡的尺度依赖性。无论原因如何，核心结论不受影响：一般大偶数的加法路径比例在 2% 到 54% 之间大幅波动，而 p-1 在所有 j 上稳定在 25.6 ± 0.2%。 p-1 的稳定性而非一般偶数的行为是本文的关键事实。

3.4. 为什么 B_add 恒正，B_mult 恒负

h(p) = ρ(p-1) + 1 − λ · ln p − μ₁₂。

• 加法路径的 p-1： ρ(p-1) = ρ(p-2) + 1，即 p-1 没有找到好的乘法分解。ρ(p-1) 相对于基线偏高 → h 偏正 → h/p > 0。
• 乘法路径的 p-1： ρ(p-1) 通过乘法分解被优化。ρ(p-1) 相对于基线偏低 → h 偏负 → h/p < 0。

类均值直接确认：<η>_add ≈ −0.62 > <η>_mult ≈ −1.47。

3.5. 协方差结构

表 3. Sub-block 协方差 Corr(S_add, S_mult) 在不同 W 和 j 下的值。

j	W=100	W=500	W=2000	W=10000
20	+0.54	+0.39	−0.07	—
23	+0.61	+0.56	+0.36	+0.08
27	+0.64	+0.72	+0.65	+0.43
29	+0.65	+0.74	+0.70	+0.55
32	+0.65	+0.75	+0.73	+0.64

强正相关（≈ +0.65），随 j 在大 W 上增强。B_add 和 B_mult 共享同一个 ρ 景观——局部 ρ 偏高时两类素数的 h(p) 都偏高。方差减小路线不可行。

---

§4. 双河汇合的实验现象

4.1. C₁(j) 和 C₂(j) 的收敛

定义 C₁(j) = B_add · j，C₂(j) = |B_mult| · j。

表 4. 双河标度量的收敛。

j	C₁(j)	C₂(j)	C₊(j)	C₋(j)	z_j
14	0.0474	0.2482	0.1478	−0.2008	−2.95
17	0.0646	0.1801	0.1223	−0.1155	−3.98
20	0.0783	0.1353	0.1068	−0.0570	−4.69
23	0.0858	0.1110	0.0984	−0.0251	−5.03
26	0.0911	0.0952	0.0931	−0.0041	−2.05
27	0.0918	0.0931	0.0924	−0.0013	−0.89
28	0.0928	0.0913	0.0920	+0.0015	+1.39
29	0.0927	0.0903	0.0915	+0.0024	+2.98
30	0.0929	0.0902	0.0916	+0.0026	+4.51
31	0.0926	0.0912	0.0919	+0.0013	+3.14
32	0.0920	0.0925	0.0922	−0.0005	−1.64

其中 C₊ = (C₁ + C₂)/2（common mode），C₋ = C₁ − C₂（differential mode）。

4.2. Common mode 和 differential mode 的行为

Common mode C₊(j)： 在早期（j = 14-27）快速下降，随后进入 L ≈ 0.092 附近的窄带（0.0915-0.0922）。后期不是严格单调——C₊(30) = 0.0916 > C₊(29) = 0.0915，C₊(32) = 0.0922 > C₊(31) = 0.0919——而是在 L 附近小幅回摆。

Differential mode C₋(j)： 从 −0.20（j=14）穿零到 +0.003（j=29-30），再回到 −0.0005（j=32）。C₋ 的符号与 z_j 的符号完全一致（C₋ < 0 ⟺ z_j < 0，C₋ > 0 ⟺ z_j > 0），零交叉位置相同（j = 27→28 和 j = 31→32）。这是定义层面的必然：C₋ = C₁ − C₂ = B_j · j，而 z_j = B_j / √(j·D_j)，两者符号相同。振幅对应关系由 §5 的 Y_j 恒等式精确给出。

4.3. 与 Paper 59 阻尼振荡的交叉验证

Paper 59 识别 R_wt 的阻尼振荡周期 T ≈ 8（in j），回弹比 r ≈ 0.80。表 4 的 C₋ 数据（j = 14-32，19 个 block）覆盖约 2.25 个振荡周期。

零交叉： C₋ 的零交叉出现在 j ≈ 27-28 和 j ≈ 31-32。间隔约 4-5 个 j 值，对应半周期。全周期 T ≈ 8-10，与 Paper 59 的 T ≈ 8 一致（考虑到 C₋ 和 R_wt 的零交叉位置略有偏移——C₋ 穿零对应 B_j 穿零，而 R_wt 的谷对应 |z_j| 最小值，两者位置一致）。

振幅衰减： C₋ 的极值序列：|C₋(j=21)| ≈ 0.046 → |C₋(j=30)| ≈ 0.0026。比值约 0.057，对应 9 个 j 步的几何衰减率 r_C ≈ 0.057^{1/9} ≈ 0.72。这略小于 Paper 59 报告的 r ≈ 0.80，可能因为 C₋ = B_j · j 比 R_wt = B_j²/(j·D_j) 多一个 j 归一化和平方操作。考虑到 R_wt ∝ C₋²，r_R ≈ r_C² ≈ 0.52，而 Paper 59 的 r 是 R_wt 序列本身的回弹比而非幅度比。两者在量级上一致，精确对应关系需要更多数据点确认。

结论： 双河同步框架与 Paper 55/59 的阻尼振荡机制在周期和衰减趋势上一致。C₋ 的振荡就是 R_wt 阻尼振荡在 P+R 分解语言中的表达。

4.4. 极限 L 的估计

从 j ≥ 28 的 C₊ 均值：L = 0.0918 ± 0.0004。

4.5. 误差项衰减

j范围	log₂	C₂ − L		每 j 的衰减率
14→20	−2.68 → −4.52	−0.31/j
20→26	−4.52 → −8.23	−0.62/j
26→32	−8.23 → −10.64	−0.40/j

近指数衰减，局部有效指数约 0.3-0.6。这说明多项式 Mertens 收敛图像太弱，但有限数据尚不能区分 2^{−0.45j}、j·2^{−j/2}、或带振荡修正的混合模型。注意 2^{-0.45j} 比 59.1 所需的 2^{-j/2} = 2^{-0.5j} 慢——单河误差的衰减速率接近但尚未达到 59.1 的要求。59.1 真正需要控制的是差模 C₁ − C₂（而非单河误差），且需要证明两河共享同一个主修正 Q_j（见 §7.1）。

---

§5. Y_j 恒等式与 59.1 的精确重述

5.1. 归一化差分

定义：

Y_j = |C₁(j) − C₂(j)| · 2^{j/2} / j

5.2. Y_j 与 √R_wt 的比例恒等式

表 5. Y_j / √R_wt 的实测值。

j	Y_j	√R_wt	Y_j / √R_wt
14	1.836	2.955	0.621
17	2.460	3.983	0.618
20	2.916	4.689	0.622
23	3.166	5.030	0.630
26	1.298	2.052	0.632
27	0.565	0.889	0.635
28	0.887	1.393	0.637
29	1.901	2.978	0.638
30	2.878	4.506	0.639
31	2.009	3.137	0.641
32	1.053	1.643	0.641

Y_j / √R_wt = 0.62 ± 0.02。 19 个 j 值无例外。标记此常数为 σ_ρ。

5.3. σ_ρ 的恒等式推导

以下推导的前半部分是严格代数恒等式（无条件）。σ_ρ 为常数的声明依赖于经验对角线规律 D_j ≈ σ²/(j·2^j)。在无条件 Guy/Chebyshev 粗界下，只能得到 Y_j/√R_wt = O(j)，因而 59.1 只推出 Y_j = O(j)，不能推出 Y_j = O(1)。Y_j = O(1) 与 59.1 的等价性需要额外的对角线双边规律 D_j ~ 1/(j·2^j)。

R_wt = B_j² / (j · D_j)。B_j = (C₁ − C₂)/j。故：

R_wt = (C₁ − C₂)² / (j³ · D_j)

√R_wt = |C₁ − C₂| / (j^{3/2} · √D_j)

Y_j = |C₁ − C₂| · 2^{j/2} / j

Y_j / √R_wt = j^{3/2} · √D_j · 2^{j/2} / j = √(j · D_j · 2^j) = √(j · 2^j · D_j)

这是精确代数恒等式，不依赖任何假设。

在经验对角线规律 D_j ≈ σ² / (j · 2^j) 下：

Y_j / √R_wt = √(j · 2^j · σ² / (j · 2^j)) = σ = σ_ρ ≈ 0.62

表 5 数据支持此规律（Y_j/√R_wt 从 0.621 到 0.641，缓慢漂移约 3%，与 D_j 拟合精度一致）。

在无条件 Guy+Chebyshev 粗界 D_j ≤ C_Guy · j / 2^j 下：

Y_j / √R_wt ≤ √(C_Guy · j²) = O(j)

因此，无条件的精确等价是：

59.1 ⟺ |Δ_j| = O(j^{3/2} · √D_j)

Y_j = O(1) 与 59.1 的等价性需要额外的对角线双边规律 D_j ~ 1/(j·2^j)。在仅有 Guy 粗上界时，59.1 只能推出 Y_j = O(j)，而不能推出 Y_j = O(1)。反向蕴涵也需要 D_j 的下界。对角线规律影响的是 Y_j 与 R_wt 之间的精确换算，不影响 §5.4 的无条件代数重述。

5.4. 59.1 的精确重述

定理 5.1（代数重述，无条件）。 令 Δ_j = C₁(j) − C₂(j) = j · B_j。则

R_wt(j) = Δ_j² / (j³ · D_j)

因此 59.1（R_wt = O(1)）精确等价于：

这是完全无条件的代数恒等式。

推论 5.2（对角线规律下）。 若另有经验对角线规律 D_j ~ σ²/(j·2^j)，则 59.1 等价于：

等价地：Y_j = O(1)。

弱猜想 W-59.1（足以关闭 H'）。 对任意 ε > 0，存在 J 使得对所有 j > J：

等价地：log R_wt = o(j)。在对角线规律下简化为 |C₁ − C₂| ≤ 2^{εj} · j · 2^{-j/2}。

注。 在无条件 Guy 粗界 D_j ≤ C_Guy · j/2^j 下，59.1 蕴涵 |Δ_j| = O(j² · 2^{-j/2})。反向蕴涵需要 D_j 的下界，这目前只有实验支持。

---

§6. 路线否决

6.1. Route A（线性递推）

动机。 若 z_j = B_j/√(j·D_j) 满足低阶常系数线性递推 z_{j+1} = a·z_j + b·z_{j-1} + c + ε_j 且递推算子的谱半径 < 1，则 z_j 的有界性（从而 R_wt = O(1)）可直接从线性稳定性证明。Paper 55 的 76% ancestor inheritance 和阻尼振荡 r ≈ 0.80 提示了这种递推结构的可能性。

否决。 z_j 序列（j = 14-32）的二阶线性拟合：z_{j+1} = 2.11 · z_j − 1.44 · z_{j-1} − 1.13。

Companion matrix 特征值：1.05 ± 0.57i，|λ| = 1.199 > 1。谱半径大于 1。残差最大 1.45。

结论： 低阶常系数线性收缩模型不可行。更复杂的非线性或高维隐藏状态模型不被此检验排除。

6.2. Route C 原版（screening-CLT）

n_cross 的增长：

j	n_cross	n_cross/j
20	74	3.7
23	272	11.8
26	1159	44.6
29	2618	90.3

当前区间的有效幂指数很高（全局幂律拟合给出约 j^{10}），足以否决 Paper 53 的 "n_cross ≈ 30j" 或任何 O(j) 线性 screening 图像；但有限数据不能断言真正超多项式增长。

Sub-block 方差放大 Ψ(n_cross)/j 随 j 增长：0.23（j=20）→ 2.19（j=29）。

结论： Screening 在大 j 上越来越弱。简单 CLT 不给 R_wt = O(1)。

6.3. 负协方差路线

Cov(S_add, S_mult) ≈ +0.65（正，不是负）。方差不缩减。

结论： B_add 和 B_mult 共享同一个 ρ 景观。局部波动正相关。

6.4. Mertens 速率路线

B_add = O(1/j) 和 |B_mult| = O(1/j) 各自的 Mertens 收敛只给 B_j = o(1/j)——太弱。59.1 需要 |B_j| = O(2^{-j/2}/√j)，即两条河的同步精度必须以指数速率提高。

结论： 仅证 Mertens 收敛不够。需要证两条河共享同一个振荡修正。

---

§7. 59.1 的剩余 gap

7.1. 双河同步定理（目标）

如果 C₁(j) 和 C₂(j) 共享同一个振荡修正 Q_j：

C₁(j) = L + Q_j + ε_j⁺ C₂(j) = L + Q_j + ε_j⁻

则 C₋(j) = C₁ − C₂ = ε_j⁺ − ε_j⁻。

59.1 ⟺ |ε_j⁺ − ε_j⁻| = O(j^{3/2} · √D_j)。

在经验对角线规律 D_j ~ 1/(j·2^j) 下，简化为 |ε_j⁺ − ε_j⁻| = O(j · 2^{-j/2})。

W-59.1 ⟺ |ε_j⁺ − ε_j⁻| = O(j^{3/2} · √D_j · 2^{o(j)})。

7.2. 误差项衰减的实验支持

误差项 |C_i − L| 的衰减速率约 2^{-0.45j}，接近但尚未达到 59.1 所需的 2^{-j/2} = 2^{-0.5j}。C₊ 在 L ≈ 0.092 附近的窄带内回摆，C₋ 的振荡幅度在衰减。59.1 需要更强的共模同步：主修正 Q_j 在两河中相同，差模 |ε⁺ − ε⁻| 达到 j · 2^{-j/2}（或更精确地 j^{3/2}·√D_j）尺度。

7.3. gap 的精确定位

59.1 的证明需要回答一个精确的问题：

为什么 add-path 和 mult-path 的 Mertens 和共享同一个振荡修正？

物理解释（来自 SAE 公理"余项守恒"）：全局 λ 平衡强制 Σ h(p)/p = 0，即 B_add + B_mult = 0 在全局极限。block 级别的偏差是有限样本波动。两条河共享振荡修正是因为它们被同一个 ρ 景观驱动（Cov > 0 确认），而 ρ 景观本身在 block 尺度上的偶极矩（add/mult 不对称）以指数速率衰减。

数学形式化需要： (a) IC min 操作在 block 尺度上的路径平衡定理 (b) p-1 的因子化结构与素数分布的交叉控制 (c) 精细 Mertens 型估计，精度达到 O(2^{-j/2})

7.4. ChatGPT 的替代框架

一个可能更短的路线：不分别证 C₁ 和 C₂ 的收敛，而是直接构造 add-path 素数和 mult-path 素数之间的确定性 transport（配对），使得配对后的带权残差以 O(2^{-j/2}) 的精度对消。具体地，寻找 block 内的配对方案 π_j(p_add, p_mult)，使得 Σ π_j · [h(p_add)/p_add + h(p_mult)/p_mult] 为小量。这把证明从"两个和独立收敛"转化为"项之间的配对对消"——一种图论/最优传输视角的算术平衡证明。

---

§8. 闭合链更新

局部 GAO（本文命题 2.1）+ Tail-k Control（待证）
  → k 一致性 → 59.2 退出独立猜想
59.1 + Guy/Chebyshev → DSF_α ∀α > 0   （Paper 59 §7.1）
59.2 → BL_α ∀α < 1                     （Paper 59 §7.2）
SD（无条件，Paper 50）
  → Hölder → A_h 收敛 → D(s) 边界 → H'

当前状态： 本文证明了局部 GAO（固定 k、短 lag 的 ACF 不变性），并将 Paper 61 的 k 一致性问题归约为 ACF tail-control。若 tail-k gap 被关闭，则 59.2 不再需要作为独立猜想，H' 的剩余距离为单一猜想 59.1。

注。 本文不声称已无条件证明 k 一致性或 59.2 已退出猜想列表。k 一致性从"完全独立假设"缩小为"tail-control gap"是本文的贡献。

59.1 的无条件精确内容：两条 Mertens 河的差模 |Δ_j| = |C₁ − C₂| 达到 O(j^{3/2} · √D_j) 精度。在经验对角线规律 D_j ~ 1/(j·2^j) 下，这等价于 |C₁ − C₂| = O(j · 2^{-j/2})。

---

§9. 结论

P+R 分解将猜想 59.1 从一个不透明的谱比有界性声明，转化为一个精确的双河同步问题：R_wt(j) = Δ_j²/(j³·D_j)，其中 Δ_j = C₁(j) − C₂(j) = j·B_j。59.1 等价于 |Δ_j| = O(j^{3/2}·√D_j)。B_j 是两条各自向 L ≈ 0.092 汇合的 Mertens 型和之间的精细残差——在 j ≥ 27 上对消度达 99% 以上。

局部几何-算术正交性将 Paper 61 的 k 一致性问题从完全独立假设缩小为 ACF tail-control gap。若此 gap 被关闭，59.2 退出独立猜想列表，H' 的剩余距离为单一猜想 59.1。

59.1 的精确含义现在是：IC 递推中加法路径和乘法路径在 p-1 上的 Mertens 和共享同一个振荡修正 Q_j，差模 |ε⁺ − ε⁻| 达到 j^{3/2}·√D_j 尺度。这个平衡的来源是 SAE 公理"余项守恒"在 block 尺度上的表达。证明它，就是证明整数的构造成本——余项——在加法和乘法两条通道之间，以指数精度维持平衡。

---

参考文献

[50] 第 L 篇。特征剥离, DSF_α, BL_α, SD, 闭合链。DOI: 10.5281/zenodo.19381111。 [58] 第 LVIII 篇。确定性 DSF + BL。DOI: 10.5281/zenodo.19476508。 [59] 第 LIX 篇。谱比有界性与 H' 的双猜想闭合。DOI: 10.5281/zenodo.19480518。 [60] 第 LX 篇。系列闭合。DOI: 10.5281/zenodo.19480563。 [61] 第 LXI 篇。谱导数界的条件归约。DOI: 10.5281/zenodo.19605749。

---

AI 贡献

Claude（子路）：全部数值实验设计与实现（paper62_exp.c, paper62_cde.c, paper62_fgh.c, paper62_jkl.c 四轮实验），P+R 分解的发现，α = 0.256 结构常数的识别，Y_j 恒等式的推导与验证，k 一致性引理的构建，C₊/C₋ common/differential mode 分析，四条路线的实验否决（Route A 谱半径计算, screening 增长律, Cov 正性, Mertens 速率不足），working notes v1-v3，独立评审与修正。ChatGPT（公西华）：双河同步框架（Two-River Synchronization）的提出，Y_j 归一化修正（识别漏掉的 j 因子），确定性 transport 作为替代证明路线，弱猜想 W-59.1 的形式化（log R_wt = o(j) 足以关闭 H'），分兵策略（Track 1 / Track 2），block-mean 定理重述（z_j = O(√j)），P+∇Z cross-term 机制，对 Paper 62 三轮 prompt 的战略评估。Grok（子贡）：Two-River Structural Balance 框架，硬饱和非线性（bang-bang 控制）识别，紧吸引子框架建议，min 操作作为饱和源的诊断，Paper 62 prompt v1–v3 的战略输入。Gemini（子夏）：Topological Debt Clearing 机制命名，Lyapunov 函数方法建议，delay-coordinate embedding 实验设计，additive path ratio tracking 实验设计，误差项 log-linear 衰减诊断（识别指数 vs 多项式衰减的判据），Mertens 型收敛与经典解析数论的连接。