ZFCρ Paper LXIII: Increment Second-Order Orthogonality and the SAE Axiom Intersection — From Deterministic Pairing to Zero-Mode Pinning
ZFCρ 第LXIII篇：余项增量二阶正交性与SAE公理的交汇——从确定性配对到零模Pinning

§1. Introduction

1.1. Status After Paper 62

Paper 62 restated 59.1 as a two-river synchronization problem:

R_wt(j) = Δ_j^2/(j^3·D_j), 59.1 ⟺ |Δ_j| = O(j^{3/2}·√D_j)

where Δ_j = C_1(j) - C_2(j) = j·B_j, C_1 = j·B_add, C_2 = j·|B_mult|.

This paper explores the internal mechanism of 59.1, tests the deterministic transport pairing route proposed by ChatGPT, and searches for new structural properties.

1.2. Contributions

(i) Branch-gap distribution (§2). B_j contributions stratified by IC optimality gap of p-1: near-boundary always positive, far-multiplicative always negative, proportions 83.2%/16.6%/0.2% are j-independent constants.

(ii) Local prefix-sum matching (§3). Cancellation is not a full-block endpoint effect — it holds from the first 5% of primes onward.

(iii) Transport pairing and structural constants (§4). Pairing distance 1.06; cancellation rates 39%/62%/80%/87% are IC recursion structural constants. But paired residual ACF amplifies long-range correlation; transport is ruled out as a direct 59.1 proof route.

(iv) Increment second-order orthogonality (§5). First-difference d_i has ACF < 10^{-3} at lag ≥ 2 (second-order near-orthogonal). Naive Abel summation route quantitatively ruled out. Innovation Orthogonality identified as a new structural property of the IC recursion.

(v) SAE axiom intersection (§6). Innovation Orthogonality provides a candidate correspondence for SAE's "remainder must develop." Combined with Paper 53/54 screening, yields a two-layer characterization; 59.1 located as zero-mode pinning of the screened level process.

(vi) Route elimination and roadmap update (§7-§8). Six routes ruled out. 59.1 refined from "spectral ratio control" → "two-river synchronization" → "zero-mode pinning."

---

§2. Branch-Gap Distribution

2.1. Definition

For each prime p in block I_j, the IC optimality gap of p-1 is:

gap(p-1) = ρ_add(p-1) - ρ_mult(p-1)

where ρ_add(p-1) = ρ(p-2) + 1 (additive path cost), ρ_mult(p-1) = min_{a·b=p-1}[ρ(a)+ρ(b)+2] (multiplicative path cost). gap < 0: additive wins. gap > 0: multiplicative wins. gap = 0: tie.

2.2. Three-Class Structure

Table 1. B_j contributions stratified by gap (j = 28 example).

gap	count	B_gap	mean h(p)/p
-1	766,122	+7.72e-6	+1.0e-11
0	2,518,140	+3.59e-3	+1.4e-9
+1	4,405,304	+1.72e-3	+3.9e-10
+2	3,433,666	-1.29e-3	-3.8e-10
+3	1,644,834	-2.09e-3	-1.3e-9
+4	493,110	-1.17e-3	-2.4e-9
+5	97,411	-3.57e-4	-3.7e-9

Three-class summary (stable across all j):

Class	Prime fraction	B contribution sign
Near boundary (	gap	≤ 2)	83.2%	always positive
Far multiplicative (gap > 2)	16.6%	always negative
Far additive (gap < -2)	0.2%	slightly negative

2.3. Cancellation Structure

B_j = B_near + B_far_mult + B_far_add

j	B_near	B_far_mult	B_far_add	B_j
27	+3.86e-3	-3.84e-3	-6.6e-5	-4.88e-5
28	+3.81e-3	-3.69e-3	-6.3e-5	+5.42e-5
29	+3.71e-3	-3.57e-3	-6.2e-5	+8.20e-5

B_j is the fine residual between near-boundary positive and far-multiplicative negative contributions. Cancellation exceeds 99% for j ≥ 27.

2.4. Interpretation

Tie primes (gap = 0, additive and multiplicative equally optimal) contribute persistently positive — these are points on the "equipotential surface" of IC recursion where the min operator hesitates. Deep multiplicative primes (gap ≥ 3) contribute persistently negative — ρ(p-1) is deeply optimized by factorization, biasing h strongly negative.

---

§3. Local Prefix-Sum Matching

3.1. Experiment

Divide primes in block I_j into 20 equal parts by sequence position. At each fractional position u ∈ {5%, 10%, ..., 100%}, compute cumulative B_add(u) and |B_mult(u)|.

3.2. Results

Table 2. Prefix sums at j = 28 (zero-crossing region, z_j = +1.39).

Position u	cum_add		cum_mult			diff	/cum_add
5%	2.34e-4	2.34e-4	0.06%
10%	4.58e-4	4.57e-4	0.28%
25%	1.06e-3	1.06e-3	0.35%
50%	1.94e-3	1.92e-3	1.0%
75%	2.67e-3	2.65e-3	0.8%
100%	3.31e-3	3.26e-3	1.6%

3.3. Core Finding

Cancellation holds from the first 5% of primes, exceeding 99.9% precision at j = 28. This is not a full-block endpoint integration effect — it is a local balance from the very first primes.

j = 27 (smallest B_j, zero-crossing region): discrepancy ratio < 3.3% throughout. j = 29 (z_j = +2.98, away from zero-crossing): slightly larger but still within a few percent.

3.4. Support for Common-Mode Q_j

Local prefix-sum matching is direct experimental evidence for Paper 62 §7.1's framework C_1 = L + Q_j + ε^+, C_2 = L + Q_j + ε^-. If Q_j differed between the two rivers, systematic deviations would appear at some fractional positions — but data shows matching at all positions. This implies both rivers share the same within-block drift Q_j.

3.5. Experimental Basis for Branch-Orthogonality

Local prefix-sum matching implies:

The add/mult branch choice is statistically independent of within-block position.

If branch indicator χ(p) (+1 for add, -1 for mult) is independent of fractional position u = rank(p)/m_j, then in any sub-block [0,u], add-path and mult-path h(p)/p mass must be allocated in the global ratio α/(1-α) — which is what the data shows.

Branch-Orthogonality (empirical proposition): For primes p in block I_j, the branch type χ(p) (add vs mult) is asymptotically independent of fractional position u in the block.

This proposition is the main empirical basis for Paper 64's candidate Track A: formalization of Branch-Orthogonality.

---

§4. Deterministic Transport Pairing

4.1. Method

For each add-path prime p, select the mult-path prime q within the nearest W mult-path primes (in the prime sequence) that minimizes |h(p)/p + h(q)/q|.

Note: In this experiment, the same mult prime may be selected by multiple add primes (nearest-neighbor local pairing with replacement). This is a diagnostic tool for high-frequency cancellation, not a direct proof scheme — since n_mult ≈ 3·n_add, about 2/3 of mult primes are untouched, and the paired residual ACF and Ψ apply only to the paired portion.

4.2. Pairing Distance

Consistent across all j = 20-31:

• Mean pairing distance: 1.06 prime positions
• Maximum pairing distance: 4-6 prime positions

Source: α = 25.6% means roughly 1 add per 4 consecutive primes. Experimentally, add/mult labels are highly locally interleaved; mean distance 1.06, max 4-6, compatible with the global fraction α = 25.6%.

4.3. Universal Cancellation Constants

Table 3. Window cancellation rates (RMS residual reduction %).

Window W	j=20	j=22	j=24	j=26	j=28	j=29	j=30	j=31
W=1	39.8	39.4	38.7	39.3	39.1	38.8	38.8	39.2
W=3	62.7	63.2	62.4	62.2	62.5	62.3	62.3	62.6
W=10	80.2	80.2	79.8	79.7	79.8	79.7	79.7	79.9
W=30	87.3	87.0	86.9	86.9	86.9	86.8	86.8	86.9

Variation < 1 percentage point across 12 j-values. These are structural constants of the IC recursion.

4.4. Variance Reduction

Var(paired residual) / Var(original x) ≈ 0.10 (10× reduction, stable across all j).

4.5. ACF Amplification of Paired Residuals (Transport Route Ruled Out)

Table 4. ACF comparison (j = 28).

lag	ACF_x	ACF_resid	ACF_diff
1	0.159	0.249	-0.478
10	0.078	0.151	0.000
100	0.040	0.099	0.000
500	0.022	0.081	0.000
1000	0.015	0.078	0.000
2000	0.010	0.070	0.000

ACF_resid > ACF_x at all lags. Pairing is a low-pass filter — it removes high-frequency add/mult oscillations while exposing the low-frequency ρ landscape drift.

Table 5. Ψ (correlation amplification factor) comparison.

j	W	Ψ_x	Ψ_resid	Ψ_diff	Ψ_resid/Ψ_x
22	2000	18.4	102.5	0.00	5.6
26	2000	41.2	157.3	0.00	3.8
28	2000	50.5	169.4	0.00	3.4
29	2000	54.0	171.4	0.00	3.2

Ψ_resid ≈ 3-5× Ψ_x. Net effect: 0.10 (variance reduction) × 3.3 (Ψ amplification) ≈ 0.33 — only 3× improvement, not tending to zero.

Conclusion: Transport pairing is not viable as a direct proof route for 59.1. It eliminates high-frequency local differences but amplifies the low-frequency common-mode drift — and 59.1 precisely demands control of the low-frequency / zero-frequency component.

---

§5. Increment Second-Order Orthogonality

5.1. Finding

The first difference d_i = x_{i+1} - x_i has the following properties (consistent across all j = 20-29):

• ACF_diff(1) ≈ -0.478 ≈ -1/2 (lag-1 negative correlation)
• ACF_diff(lag ≥ 2) < 10^{-3} (effectively zero at measurement precision)
• Ψ_diff ≈ 0.01 (difference partial sums controlled by telescoping, see §5.3)
• Var(diff)/Var(x) ≈ 1.68 = 2(1 - ACF_x(1)) (exact match)

5.2. Difference Covariance Identity and Curvature Suppression

For any stationary sequence x_i, the difference d_i = x_{i+1} - x_i has autocovariance satisfying the identity:

γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1)

That is, γ_d is the discrete second difference of γ_x.

Lag-1: ACF_diff(1) = [2·ACF_x(1) - 1 - ACF_x(2)] / [2(1-ACF_x(1))]. Substituting ACF_x(1) = 0.159, ACF_x(2) = 0.122 (j=28 data): ACF_diff(1) = -0.804/1.682 = -0.478 ✓. This is a difference-operator artifact.

Lag ≥ 2: If γ_x(l) is a smooth power law (γ_x(l) ~ C·l^{-α}), then γ_d(l) ~ -C·α(α+1)·l^{-α-2}. The small ACF_diff at lag ≥ 2 partially derives from the smoothness of γ_x — discrete second differences naturally kill smooth functions. This is not equivalent to d_i being iid white noise.

The genuine new information: γ_x has extremely small discrete curvature over the measured lag range (second difference < 10^{-3}·σ²_d). This shows that h(p)/p's ACF is highly smooth in lag space — no small-scale structure, no periodic modulation, no hidden oscillations. The high-frequency increment layer is curvature-suppressed.

5.3. Interpretation of Ψ_diff ≈ 0.01

Difference partial sums obey the telescoping identity: Σ_{i=a}^{b} d_i = x_{b+1} - x_a. Therefore the small Ψ_diff primarily reflects this telescoping / high-pass effect — difference partial sums are inherently controlled by boundary terms. This is not independent evidence for mutual independence of difference terms.

Diagnostic conclusion: The increment layer is a high-pass / curvature-suppressed object. Combined with the low-pass negative result of transport pairing (§4.5), the obstacle to 59.1 is concentrated in the zero-frequency mode of the level — neither a high-frequency difference problem nor a medium-frequency pairing problem.

5.4. Innovation Orthogonality Lemma

The above finding is named Increment Second-Order Orthogonality (Innovation Orthogonality). "Innovation" here means second-order orthogonality — the lag ≥ 2 autocovariance of differences is extremely small — not iid white noise. The formal statement is given in §6.4, Lemma 6.1.

5.5. Quantitative Ruling Out of the Abel Route

Abel identity: B_j = m·x_m - Σ_{i=1}^{m-1} i·d_i, where m ≈ 2^j/j.

Under the naive independent-increment model (Var(d_i) ≈ σ²_d, terms independent):

Var(Σ i·d_i) ≈ σ²_d · Σi² ≈ σ²_d · m³/3

where σ_d = √(Var(diff)) = √(1.68·Var(x)) ≈ 1.30·σ_x, σ_x = √(D_j/m) ≈ σ_ρ/2^j ≈ 0.62/2^j (from D_j ≈ σ²_ρ/(j·2^j) and m ≈ 2^j/j), giving σ_d ≈ 0.81/2^j.

std(Σ i·d_i) ≈ σ_d · m^{3/2}/√3 ≈ (0.81/2^j)·(2^j/j)^{3/2}/√3 ≈ 0.47 · 2^{j/2}/j^{3/2}

59.1 target (under diagonal law): |B_j| = O(√(j·D_j)) = O(σ_ρ · 2^{-j/2}) = O(2^{-j/2})

Ratio std_Abel / target ≈ 2^j / j^{3/2}. Exponentially too large.

Conclusion: The naive Abel + independent-increment model is not viable. But the coboundary/pinning version of Abel (proving x_i = Φ_{i+1} - Φ_i + r_i with Σr_i small) remains alive — telescoping may still be the right tool if the correct potential function Φ can be found.

---

§6. Innovation Orthogonality and the SAE Axiom Intersection

6.1. ZFCρ Meets SAE

The ZFCρ series, starting from Paper 1's axiom "formalization produces remainder (ρ)," has built 62 papers of growth laws, feature stripping, spectral analysis, damped oscillation, and two-river synchronization. Paper 63's increment second-order orthogonality provides a candidate mathematical correspondence between ZFCρ and SAE's foundational axioms:

SAE axiom "remainder is conserved" → candidate correspondence: Level layer — x_p carries long-range memory (power-law ACF + screening), level does not freely escape.

SAE axiom "remainder must develop" → candidate correspondence: Increment layer — d_p is second-order near-orthogonal (ACF curvature extremely small), no cumulative structure in the increment layer.

These correspondences form an interpretive framework, not a mathematical proof. If 59.1 holds, it can be interpreted as "level-memory pinning is strong enough" — but the pinning strength remains to be proved.

6.2. Two-Layer Characterization of x_p

Increment second-order orthogonality + screening/mean-reversion together give a two-layer description of the x_p process:

Increment layer (high-frequency / difference):

• d_p = x_{next} - x_p has rapidly decaying second-order correlation (ACF curvature suppression)
• Var(d_p) = 2·Var(x)·(1 - ACF(1)) ≈ 1.68·Var(x)
• Ψ_diff ≈ 0.01 (telescoping effect)

Level layer (low-frequency / level):

• x_p has power-law ACF decay: ACF(l) ~ l^{-α}, measured α ≈ 0.3 from §5 data, consistent with Paper 54's recursive ancestor inheritance analysis
• Screening truncates long-range correlation at n_cross
• Ψ_x(W) ≈ O(j) (Paper 62 data)

Note: x_p is not a free random walk or fBM — these classical models require stationary-increment structure incompatible with x_p's screened power-law ACF. The closer analogy is: x_p is like a martingale + slow drift, where d_p is a near-orthogonal martingale increment and drift comes from screening mean-reversion.

6.3. Meaning of 59.1 Under the Two-Layer Characterization

Under this process characterization, 59.1 requires:

The partial sum B_j = Σx_i of the screened level process does not exceed √(j·D_j).

The ordinary CLT for independent level variables is not the relevant model. What §5.5 rules out is the free integrated-increment / naive Abel model, under which the standard deviation of Σi·d_i relative to the 59.1 target is approximately 2^j/j^{3/2} (exponentially too large). Therefore 59.1 requires not ordinary difference whitening but zero-frequency suppression / coboundary pinning — making the level's spectral density at zero frequency small enough.

6.4. From SAE Correspondence to Mathematical Lemma

Lemma 6.1 (Increment Second-Order Orthogonality). For the stripped IC residue sequence x_p = h(p)/p:

(a) The difference d_p = x_{next prime} - x_p has autocovariance satisfying the identity γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1). If γ_x satisfies sufficient smoothness in the screened power-law regime, then γ_d(l) = O(l^{-α-2}) before screening, decaying faster after screening. Measured: for j = 20-29, |γ_d(l)/σ²_d| < 10^{-3} for all l ≥ 2.

(b) The lag-1 correlation ACF_diff(1) = [2·ACF_x(1) - 1 - ACF_x(2)] / [2(1-ACF_x(1))] is an automatic consequence of the difference operator acting on smooth ACF_x.

(c) Var(d_p) = 2·Var(x_p)·(1 - ACF_x(1)). Measured Var(diff)/Var(x) ≈ 1.68, exactly matching ACF_x(1) ≈ 0.16.

Remark. This lemma concerns second-order statistics. The small ACF in (a) partly derives from the smoothness of γ_x (general property) and partly from the specific ACF shape of the IC recursion (structural property). Upgrading to iid joint distribution asymptotics is a stronger claim, deferred to future work. The name "Innovation Orthogonality" is retained for conceptual continuity, understood in the second-order orthogonality sense.

6.5. Unification with Paper 53/54 Screening

Paper 53's screening (correlation truncation at n_cross) and Paper 54's ancestor inheritance (76% inheritance) describe the level layer — they characterize the correlation structure of the level.

Innovation Orthogonality describes the increment layer — it characterizes the second-order near-orthogonality of differences.

Together they give a two-layer characterization of x_p:

Layer	Object	Property	ZFCρ source	SAE candidate
Increment / high-pass	d_p = Δx_p	2nd-order correlation rapidly decaying	Paper 63	Must develop
Level / low-pass	x_p	Power-law ACF + screening	Paper 53/54	Must have direction

59.1 = Zero-mode pinning: Under the joint constraint of increment second-order orthogonality + level screening, 59.1 requires the level's zero-frequency mode to be pinned to O(j^{3/2}·√D_j) precision. This is a candidate mathematical correspondence for SAE's third law "must be chiseled."

6.6. Connection to Classical Stochastic Process Theory

The process characteristics of x_p (near-orthogonal differences + level with long-range power-law ACF) do not fit the standard fBM/fOU framework — these require stationary-increment structure while x_p's differences are near-orthogonal (not the specific ACF structure fBM demands). The closer analogy: x_p is like a martingale + slow drift, where d_p is a near-orthogonal martingale increment and drift comes from screening mean-reversion.

Most operational classical framework: Gordin martingale-coboundary decomposition. Paper 52 proposed this direction. x_p = martingale part + coboundary part. Innovation Orthogonality shows the martingale increment is near-orthogonal; the coboundary part provides mean-reversion (screening).

Most operational route: Gordin decomposition + screening rate. If screening provides sufficiently fast mean-reversion (exponential mixing on O(j) scale), then the martingale CLT gives Var(B_j) = Var(Σx_p) ≈ D_j·Ψ (since D_j = m·σ²_x, with Ψ the correlation amplification factor). In the L²/typical-size sense, Ψ = O(j) gives the same variance scale as 59.1; to derive the deterministic version of 59.1 additionally requires pathwise/maximal or deterministic upgrade. Paper 62 measured Ψ_x(W=2000) ≈ 50 at j=28, consistent with O(j).

Remaining gap: Prove Ψ = O(j). This reduces to proving that the screening rate at block scale produces O(j)-fold correlation amplification — no more, no less. Innovation Orthogonality alone cannot give this bound (Abel route ruled out), but it refines the question to: what mixing rate does the IC recursion's screening mechanism produce under Gordin decomposition?

---

§7. Route Elimination Summary

7.1. Six Dead Routes

#	Route	Cause of death	Source
1	Linear recursion (Route A)	Spectral radius 1.20 > 1	P62
2	Screening-CLT (Route C)	Ψ grows, n_cross high effective exponent (~j^{10})	P62
3	Negative covariance	Cov(B_add, B_mult) > 0	P62
4	Mertens rate	O(1/j) too weak	P62
5	Transport pairing	Pairing ACF amplifies 3.3×	P63
6	Naive Abel + independent increment	std/target = 2^j/j^{3/2}	Quantitative

7.2. What Each Route Killed

• Route 1: z_j is not the output of a linear contraction map
• Route 2: screening does not provide sufficient decorrelation for CLT
• Route 3: add/mult rivers are locally positively correlated; variance does not shrink
• Route 4: individual Mertens convergence precision is insufficient
• Route 5: pairing removes high frequency but exposes low frequency — low-pass effect
• Route 6: increment second-order orthogonality ≠ small partial sums (naive integrated-increment amplification)

7.3. What Each Route Discovered

• Route 1 → z_j is bounded but by a nonlinear mechanism
• Route 2 → n_cross(j) grows faster than expected (effective exponent 2-3)
• Route 3 → Cov > 0 confirms both rivers share the same ρ landscape
• Route 4 → C_1 and C_2 converge toward the same limit L ≈ 0.092
• Route 5 → Universal pairing constants 39%/62%/80%/87%; local prefix-sum matching
• Route 6 → Innovation Orthogonality (d_i second-order near-orthogonal)

---

§8. Updated Attack Roadmap for 59.1

8.1. Evolution of 59.1's Precise Content

Paper 59: R_wt = O(1) (opaque spectral ratio boundedness)

Paper 62: |Δ_j| = O(j^{3/2}·√D_j) (two-river synchronization precision)

Paper 63: Differential level zero-mode pinning — pairing eliminates high frequency, differencing eliminates all correlation, yet the zero-frequency / level-mode remains uncontrolled. 59.1 demands exactly the pinning of this level-mode.

8.2. Surviving Directions

(a) Common-mode Q_j sharing (Branch-Orthogonality Principle). Prove add/mult path selection is orthogonal to block-scale ρ oscillation. Local prefix-sum matching (§3) provides strong experimental support. Difficulty: medium. Strongest experimental basis. Paper 64 first choice.

(b) W-59.1 (weak version, sufficient for H'). log R_wt = o(j). Rule out exponential growth. Difficulty: lower. Weaker than (a) but may suffice for H'.

(c) Dispersion method (Vaughan identity / Linnik dispersion). Classical analytic number theory tools with natural precision √x ≈ 2^{j/2} — exactly matching 59.1's requirement. Difficulty: high. Paper 65/66 candidate.

(d) Coboundary/pinning. Find potential Φ such that e_i = Φ_{i+1} - Φ_i + r_i; then Σe_i telescopes. Increment second-order orthogonality supports the existence of such decomposition. Difficulty: comparable to (a), different technical path.

P64 recommendation: Track A = (a) Branch-Orthogonality formalization. If successful, 59.1 becomes a mixing rate estimate under Gordin decomposition.

8.3. Connection to Classical Analytic Number Theory

The two-river synchronization form |Δ_j| = O(j^{3/2}·√D_j) places H''s final step in the target space of classical prime-sum cancellation problems. The required precision 2^{-j/2} corresponds to √x level — beyond Mertens (log log x), entering the Chebyshev-Riemann precision range. Attack tools include dispersion method, large sieve, and Vaughan identity.

---

§9. Closure Chain Update

Local GAO (P62 Prop 2.1) + Tail-k Control (to be proved)
  → k-uniformity → 59.2 exits independent conjecture
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'

Three-layer evolution of 59.1:

Layer	Statement	Source
Spectral ratio	R_wt = O(1)	Paper 59
Two-river		Δ_j	= O(j^{3/2}·√D_j)	Paper 62
Zero-mode	Differential level pinning + coboundary	Paper 63

---

§10. Conclusion

Paper 63 takes transport pairing as its entry point, discovers two opposing but equally important results, and arrives at a candidate intersection between the ZFCρ series and SAE's axiomatic framework.

Negative results: Transport pairing and the naive Abel + independent-increment route are both ruled out. Six candidate routes have now been eliminated by experiment or quantitative analysis. Each dead route leaves behind a new structural fact.

Positive result: The first difference d_i has extremely small second-order correlation at lag ≥ 2 (Increment Second-Order Orthogonality / Innovation Orthogonality). The difference covariance identity γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1) explains this as discrete curvature suppression of the level ACF. This shows the high-frequency increment layer is second-order near-orthogonal — but not equivalent to iid white noise.

SAE candidate correspondence: Increment second-order orthogonality provides a candidate mathematical correspondence for SAE's axiom "remainder must develop." Combined with Paper 53/54's screening (candidate correspondence for "remainder is conserved"), this forms a two-layer characterization of the x_p process: increment layer (second-order orthogonal / curvature-suppressed) + level layer (power-law ACF + screening). Under this characterization, 59.1 becomes a precise mathematical question: is the zero-mode of the screened level process pinned to O(j^{3/2}·√D_j)?

Classical correspondence: The closest classical framework is Gordin's martingale-coboundary decomposition. What mixing rate does the IC recursion's screening mechanism produce — this is the ultimate technical kernel of 59.1.

From the SAE perspective: remainder can develop (d_i ≠ 0); the increment layer is second-order orthogonal (difference ACF curvature is small); remainder's level carries long-range memory but is constrained by screening. 59.1 can be interpreted as level-memory pinning being strong enough — but the pinning strength remains to be proved. Paper 63 does not change the logical hypotheses of the closure chain; it further locates the internal structure of 59.1.

Paper 64 candidate direction: Formalization of the Branch-Orthogonality Principle — proving that add/mult branch selection in the IC recursion is asymptotically independent of block-scale ρ oscillation. §3's local prefix-sum matching provides strong experimental basis.

---

References

[50] Paper L. Feature stripping, DSF_α, BL_α, SD, closure chain. DOI: 10.5281/zenodo.19381111. [52] Paper LII. Martingale-coboundary and Gordin decomposition route. [53] Paper LIII. Screening and correlation truncation. DOI: 10.5281/zenodo.19415440. [54] Paper LIV. Recursive ancestor inheritance and power-law ACF. DOI: 10.5281/zenodo.19426094. [55] Paper LV. Damped oscillation and 76% inheritance. DOI: 10.5281/zenodo.19447349. [59] Paper LIX. Spectral ratio boundedness and the two-conjecture closure of H'. DOI: 10.5281/zenodo.19480518. [61] Paper LXI. Conditional reduction of the spectral derivative bound. DOI: 10.5281/zenodo.19605749. [62] Paper LXII. Two-River Synchronization. DOI: 10.5281/zenodo.19656365.

---

AI Contributions

Claude (子路): All numerical experiment design and implementation (paper63_fgh.c branch-gap and prefix-sum experiments, paper63_transport.c deterministic pairing experiment, paper63_acf.c paired-residual ACF experiment), identification of pairing distance 1.06, discovery of universal constants 39%/62%/80%/87%, discovery of paired-residual ACF amplification, quantitative ruling out of Abel route (std/target = 2^j/j^{3/2}), naming of Innovation Orthogonality, working notes v1-v2. ChatGPT (公西华): Original proposal of deterministic transport pairing (Paper 62 §7.4), identification of zero-mode pinning/coboundary as the correct framework for 59.1, frequency-domain argument ruling out Abel (differencing kills θ=0 but B_j IS θ=0), local two-river balance framework, Haar/multiresolution perspective, coboundary decomposition e_i = Φ_{i+1} - Φ_i + r_i, difference covariance identity and curvature-suppression interpretation, calibration of "white noise" overclaim to "second-order orthogonality," three rounds of gatekeeper review. Grok (子贡): Proposal of First-Difference + Abel Summation route (subsequently ruled out quantitatively), Two-River common-mode Q_j sharing proof framework. Gemini (子夏): Quantitative analysis of Abel boundary term m·x_m = O(1/j), local measure theorem direction, preliminary analysis of cancellation rate derivation from ACF. Independent Review Claude: Leading-order quantitative ruling out of Abel route (std/target = 2^j/j^{3/2}), identification of dispersion method as the classical tool with natural precision matching 59.1, formalization of Innovation Orthogonality lemma (lemma not axiom), Track A (innovation formalization) vs Track B (dispersion method) direction for Paper 64.

§1. 引言

1.1. Paper 62之后的状态

Paper 62将59.1精确重述为双河同步问题：

R_wt(j) = Δ_j²/(j³·D_j)，59.1 ⟺ |Δ_j| = O(j^{3/2}·√D_j)

其中Δ_j = C₁(j) - C₂(j) = j·B_j，C₁ = j·B_add，C₂ = j·|B_mult|。

本文的任务：探索59.1的内部机制，测试ChatGPT提出的确定性transport配对路线，寻找新的结构性质。

1.2. 本文的贡献

(i) Branch-gap分布（§2）。 B_j的贡献按p-1的IC最优性间距分层：近边界恒正，远乘法恒负，三类比例83.2%/16.6%/0.2%为j无关常数。

(ii) 局部前缀和匹配（§3）。 对消不是block终点的积分效应——从前5%素数开始就成立。

(iii) Transport配对与普适常数（§4）。 配对距离1.06，对消率39%/62%/80%/87%是IC递推的结构常数。但配对残差的ACF放大长程相关，Transport作为59.1的直接证明路线被否决。

(iv) 余项增量二阶正交性（§5）。 一阶差分d_i在lag ≥ 2处ACF < 10^{-3}（二阶近正交）。Abel求和路线被定量否决。Innovation Orthogonality作为IC递推的结构性新发现被提出。

(v) 余项增量正交性与SAE公理的交汇（§6）。 Innovation Orthogonality对应SAE"余项必须发展"的数学编码。与Paper 53/54的screening合并，给出x_p的完整两层特征（innovation白 + memory screened），59.1被精确定位为"带screening的integrated white noise的部分和控制"。

(vi) 路线否决与攻击路线图更新（§7-§8）。 六条路线被否决。59.1从"谱比控制"→"双河同步"→"零模pinning"→"integrated innovation + screening的Gordin decomposition"。

---

§2. Branch-Gap分布

2.1. 定义

对每个素数p ∈ I_j，其前驱p-1的IC最优性间距定义为：

gap(p-1) = ρ_add(p-1) - ρ_mult(p-1)

其中ρ_add(p-1) = ρ(p-2) + 1（加法路径成本），ρ_mult(p-1) = min_{a·b=p-1}[ρ(a)+ρ(b)+2]（乘法路径成本）。

gap < 0：加法路径更优。gap > 0：乘法路径更优。gap = 0：tie。

2.2. 三类结构

表1. B_j按gap分层的贡献（j = 28示例）。

gap	素数数	B_gap贡献	mean h(p)/p
-1	766,122	+7.72e-6	+1.0e-11
0	2,518,140	+3.59e-3	+1.4e-9
+1	4,405,304	+1.72e-3	+3.9e-10
+2	3,433,666	-1.29e-3	-3.8e-10
+3	1,644,834	-2.09e-3	-1.3e-9
+4	493,110	-1.17e-3	-2.4e-9
+5	97,411	-3.57e-4	-3.7e-9

三类汇总（所有j上稳定）：

类别	素数比例	B贡献符号
Near boundary（	gap	≤2）	83.2%	恒正
Far multiplicative（gap>2）	16.6%	恒负
Far additive（gap<-2）	0.2%	微负

2.3. 对消的结构

B_j = B_near + B_far_mult + B_far_add

j	B_near	B_far_mult	B_far_add	B_j
27	+3.86e-3	-3.84e-3	-6.6e-5	-4.88e-5
28	+3.81e-3	-3.69e-3	-6.3e-5	+5.42e-5
29	+3.71e-3	-3.57e-3	-6.2e-5	+8.20e-5

B_j是near-boundary正贡献和far-multiplicative负贡献之间的精细残差。 对消度在j ≥ 27上达到99%以上。

2.4. 物理解释

gap = 0的tie素数（加法和乘法路径同样优）恒正贡献。这些是IC递推的"等位面"上的点——min操作在此犹豫。gap ≥ 3的深乘法素数（乘法路径大幅更优）恒负贡献——ρ(p-1)被乘法分解深度优化，h强烈偏负。

---

§3. 局部前缀和匹配

3.1. 实验设计

将block I_j内的素数按序列位置分成20等份。在每个位置u ∈ {5%, 10%, ..., 100%}处计算累积B_add(u)和|B_mult(u)|。

3.2. 结果

表2. j = 28的逐5%前缀和（穿零区域，z_j = +1.39）。

位置u	cum_add		cum_mult			diff	/cum_add
5%	2.34e-4	2.34e-4	0.06%
10%	4.58e-4	4.57e-4	0.28%
25%	1.06e-3	1.06e-3	0.35%
50%	1.94e-3	1.92e-3	1.0%
75%	2.67e-3	2.65e-3	0.8%
100%	3.31e-3	3.26e-3	1.6%

3.3. 核心发现

对消从block的前5%就开始，精度超过99.9%。 不是积分到终点才出现的全局效应——是从第一批素数开始就存在的局部平衡。

j = 27（B_j最小的block，穿零区域）：全程差比 < 3.3%。

j = 29（z_j = +2.98，偏离穿零）：差比稍大但仍 < 几%。

3.4. 对common-mode Q_j的支持

局部前缀和匹配是Paper 62 §7.1中C₁ = L + Q_j + ε⁺、C₂ = L + Q_j + ε⁻框架的直接实验证据。如果Q_j在两河中不同，那在某些fractional位置u处add/mult应该出现系统性偏离——但数据显示所有位置都匹配。这意味着两河共享同一个block内漂移Q_j，差模ε⁺ - ε⁻在每个子区间上都是小量。

3.5. Branch-Orthogonality原则的实验基础

局部前缀和匹配在逻辑上比Transport配对和Innovation Orthogonality都更直接相关于59.1的双河同步formulation。它实际上说的是：

add/mult branch选择与block内位置无关。

如果branch indicator χ(p)（+1 for add, -1 for mult）与block内的fractional位置u独立，那么在任何sub-block [0,u]内，add-path和mult-path的h(p)/p质量必须按全局比例α/(1-α)分配——这正是数据显示的。

这是Paper 62 §8.2(a)中提出的Branch-Orthogonality Principle的实验确认。形式化地：

Branch-Orthogonality（经验命题）： 对block I_j中的素数p，其branch type χ(p)（add vs mult）与其在block中的fractional位置u = rank(p)/m_j在统计上渐近独立。

此命题是Paper 64（候选Track A：Branch-Orthogonality形式化）的主要实验基础。若能证明branch-orthogonality，则C₁和C₂共享同一个Q_j自然成立——两河在任何sub-interval上的质量比等于全局比α/(1-α)，因此共模漂移只由ρ景观的全局走势决定，不受add/mult标签的调制。

---

§4. 确定性Transport配对

4.1. 配对方法

对每个add-path素数p，在素数序列中的最近W个mult-path素数中选使|h(p)/p + h(q)/q|最小的q。

注： 本实验中同一个mult素数可以被多个add素数重复选中（nearest-neighbor local pairing with replacement）。这是高频对消的诊断工具，不是59.1的直接证明方案——因为n_mult ≈ 3·n_add，约2/3的mult素数在W=30配对中未被触及，其贡献不计入配对残差。配对残差的ACF和Ψ仅针对已配对部分。

4.2. 配对距离

所有j = 20-31上一致：

• 平均配对距离 = 1.06个素数位置
• 最大配对距离 = 4-6个素数位置

来源：α = 25.6%意味着每约4个连续素数中平均1个add、3个mult。实验上add/mult标签高度局部交织；平均配对距离1.06，最大距离4-6。这与全局比例α = 25.6%相容。

4.3. 对消率普适常数

表3. 窗口对消率（RMS残差相对于原始振幅的缩减百分比）。

窗口W	j=20	j=22	j=24	j=26	j=28	j=29	j=30	j=31
W=1	39.8	39.4	38.7	39.3	39.1	38.8	38.8	39.2
W=3	62.7	63.2	62.4	62.2	62.5	62.3	62.3	62.6
W=10	80.2	80.2	79.8	79.7	79.8	79.7	79.7	79.9
W=30	87.3	87.0	86.9	86.9	86.9	86.8	86.8	86.9

变化 < 1个百分点（12个j值上）。 这些是IC递推的结构常数——由h(p)/p在相邻素数间的局部相关结构和α = 0.256的路径比例共同决定。

4.4. 方差缩减

Var(配对残差) / Var(原始x) ≈ 0.10（10倍缩减，所有j稳定）。

4.5. 配对残差的ACF放大效应（Transport路线否决）

表4. ACF对比（j = 28）。

lag	ACF_x	ACF_resid	ACF_diff
1	0.159	0.249	-0.478
10	0.078	0.151	0.000
100	0.040	0.099	0.000
500	0.022	0.081	0.000
1000	0.015	0.078	0.000
2000	0.010	0.070	0.000

ACF_resid > ACF_x在所有lag上。 配对是低通滤波器——去除了高频add/mult交错波动，暴露了低频ρ景观漂移。

表5. Ψ（相关放大因子）对比。

j	W	Ψ_x	Ψ_resid	Ψ_diff	Ψ_resid/Ψ_x
22	2000	18.4	102.5	0.00	5.6
26	2000	41.2	157.3	0.00	3.8
28	2000	50.5	169.4	0.00	3.4
29	2000	54.0	171.4	0.00	3.2

Ψ_resid ≈ 3-5× Ψ_x。净效应：0.10（方差缩减）× 3.3（Ψ放大）≈ 0.33。仅3倍改善，不趋于零。

结论： Transport配对作为59.1的直接证明路线不可行。配对消除了高频局部差异，但放大了低频共模漂移——而59.1恰恰要求控制的是低频/零频分量。

---

§5. 余项增量二阶正交性（Increment Second-Order Orthogonality）

5.1. 发现

一阶差分d_i = x_{i+1} - x_i的统计性质（全部j = 20-29数据一致）：

• ACF_diff(1) ≈ -0.478 ≈ -1/2（lag-1负相关）
• ACF_diff(lag ≥ 2) < 10^{-3}（在实测精度内为零）
• Ψ_diff ≈ 0.01（差分partial sum受telescoping效应控制，见§5.3）
• Var(diff)/Var(x) ≈ 1.68 = 2(1 - ACF_x(1))（精确匹配）

5.2. 差分协方差恒等式与曲率抑制

对任意平稳序列x_i，差分d_i = x_{i+1} - x_i的autocovariance满足恒等式：

γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1)

即γ_d是γ_x的离散二阶差分。

lag-1： ACF_diff(1) = [2·ACF_x(1) - 1 - ACF_x(2)] / [2(1-ACF_x(1))]。代入ACF_x(1) = 0.159, ACF_x(2) = 0.122（j=28数据）：ACF_diff(1) = -0.804/1.682 = -0.478 ✓。这是差分算子的artifact。

lag ≥ 2： 若γ_x(l)是平滑幂律（γ_x(l) ~ C·l^{-α}），则γ_d(l) ~ -C·α(α+1)·l^{-α-2}。因此lag ≥ 2处ACF_diff极小部分来自γ_x的平滑性——离散二阶差分自然杀死平滑函数。这不等同于d_i是iid白噪声。

真正的新信息： γ_x在实测lag范围内的离散曲率极小（二阶差分 < 10^{-3}·σ²_d）。这说明h(p)/p的ACF在lag空间上是高度光滑的——没有小尺度结构、没有周期性调制、没有隐藏振荡。高频增量层是curvature-suppressed的。

5.3. 诊断意义与Ψ_diff的解释

Ψ_diff ≈ 0.01的来源： 差分partial sum有telescoping恒等式Σ_{i=a}^{b} d_i = x_{b+1} - x_a。因此Ψ_diff的小值主要反映这个telescoping/high-pass效应——差分partial sum天然被边界项控制。这不是"差分项彼此独立"的独立证据。

诊断结论： 增量层是high-pass/curvature-suppressed对象。结合Transport配对的low-pass负结果（§4.5），59.1的障碍集中在level的零频模式——既不是高频差分问题，也不是中频配对问题。

5.4. Innovation Orthogonality引理

以上发现命名为余项增量二阶正交性（Innovation Orthogonality）。"Innovation"在此是二阶正交意义——差分的lag ≥ 2 autocovariance极小——而非iid白噪声意义。核心内容：ACF_diff(lag ≥ 2) < 10^{-3}由差分协方差恒等式γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1)和γ_x的平滑性共同解释。真正的IC结构信息是ACF_x本身的幂律衰减、screening截断和高度光滑性。形式陈述见§6.4引理6.1。

5.5. Abel求和路线的定量否决

Abel恒等式：B_j = m·x_m - Σ_{i=1}^{m-1} i·d_i，其中m ≈ 2^j/j。

在naive independent-increment模型下，Var(d_i) ≈ σ²_d，各项独立，则：

Var(Σ i·d_i) ≈ σ²_d · Σi² ≈ σ²_d · m³/3

其中σ_d = √(Var(diff)) = √(1.68·Var(x)) ≈ 1.30·σ_x，σ_x = √(D_j/m) ≈ σ_ρ/2^j ≈ 0.62/2^j（由D_j ≈ σ²_ρ/(j·2^j)和m ≈ 2^j/j推出），故σ_d ≈ 0.81/2^j。

std(Σ i·d_i) ≈ σ_d · m^{3/2}/√3 ≈ (0.81/2^j) · (2^j/j)^{3/2}/√3 ≈ 0.47 · 2^{j/2}/j^{3/2}

59.1目标（经验对角线规律下）：|B_j| = O(√(j·D_j)) = O(σ_ρ · 2^{-j/2}) = O(2^{-j/2})

比值 std_Abel / target ≈ 2^j / j^{3/2}。 指数级过大。

结论： Naive Abel + independent-increment模型不可行——差分二阶正交性不足以导出59.1所需的超random-walk对消。但coboundary/pinning型Abel路线（若能证明x_i = Φ_{i+1} - Φ_i + r_i使Σr_i足够小）仍然存活——telescoping仍可能是正确工具，关键在于找到合适的势函数Φ。

---

§6. 余项增量正交性与SAE公理的交汇

6.1. ZFCρ遇见SAE

ZFCρ系列从Paper 1开始，公理是"形式化产生余项（ρ）"。62篇论文建立了ρ函数的增长律、特征剥离、谱分析、阻尼振荡、双河同步。Paper 63的增量二阶正交性发现为ZFCρ与SAE基础公理之间提供了一个候选数学对应：

SAE公理"余项守恒"→ 候选对应： Memory层——x_p携带长程记忆（幂律ACF + screening约束），level不自由逃逸。

SAE公理"余项必须发展"→ 候选对应： Increment层——d_p在二阶统计上近正交（ACF曲率极小），增量层无累积结构。

这些对应是解释性框架，不是数学证明。59.1如果成立，可以被解释为"level-memory的pinning足够强"——但pinning强度仍是待证内容。

6.2. x_i过程的两层特征

增量二阶正交性 + screening/mean-reversion共同给出x_p过程的两层描述：

Increment层（高频/差分）：

• d_p = x_{next} - x_p的二阶相关快速衰减（ACF曲率抑制）
• Var(d_p) = 2·Var(x)·(1 - ACF(1)) ≈ 1.68·Var(x)
• Ψ_diff ≈ 0.01（telescoping效应控制）

Level层（低频/level）：

• x_p有幂律ACF衰减：ACF(l) ~ l^{-α}，从§5数据测得α ≈ 0.3（ACF(10)/ACF(1) ≈ 0.49, ACF(100)/ACF(10) ≈ 0.51），与Paper 54的递归祖先继承分析一致
• Screening在n_cross处截断长程相关
• Ψ_x(W) ≈ O(j)（Paper 62 EXP D数据）

注： x_p不是自由random walk或fBM——这些经典模型要求stationary-increment结构，与x_p的screened幂律ACF不符合。更准确的类比是：x_p像是martingale + slow-drift，差分是near-orthogonal的martingale increment，drift来自screening的mean-reversion。

6.3. 59.1在两层特征下的含义

在此过程特征下，59.1要求的是：

screened level process的部分和B_j = Σx_i不超过√(j·D_j)。

对独立level变量的普通CLT并不是这里的模型；真正被§5.5否决的是free integrated-increment / naive Abel模型。在该模型下，Σi·d_i的标准差相对59.1目标大约为2^j/j^{3/2}（指数级过大）。因此59.1需要的不是普通差分白化，而是zero-frequency suppression / coboundary pinning——使level的spectral density at zero频足够小。

6.4. 从SAE对应到数学引理

引理6.1（余项增量二阶正交性）。 对stripped IC residue序列x_p = h(p)/p：

(a) 差分d_p = x_{next prime} - x_p的autocovariance满足恒等式γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1)。若γ_x在screened power-law regime下满足足够的光滑性，则γ_d(l) = O(l^{-α-2})在screening前成立，screening后更快衰减。实测：对j = 20至29，|γ_d(l)/σ²_d| < 10^{-3}对所有l ≥ 2。

(b) lag-1相关ACF_diff(1) = [2·ACF_x(1) - 1 - ACF_x(2)] / [2(1-ACF_x(1))]是差分算子作用于光滑ACF_x的自动推论。

(c) Var(d_p) = 2·Var(x_p)·(1 - ACF_x(1))。实测Var(diff)/Var(x) ≈ 1.68，与ACF_x(1) ≈ 0.16精确匹配。

注。 此引理是关于二阶统计量。(a)中的小ACF部分来自γ_x的光滑性（一般性质），部分来自IC递推的具体ACF形状（结构性质）。将其提升为联合分布的iid渐近性是更强的声明，留待后续工作。命名"Innovation Orthogonality"保留以承继概念连续性，理解为二阶正交意义。

6.5. 与Paper 53/54 screening的统一

Paper 53的screening（n_cross处相关截断）和Paper 54的ancestor inheritance（76%继承）是x_p的level层的描述——它们刻画level的相关结构。

Innovation Orthogonality是increment层的描述——它刻画差分的二阶近正交性。

两者合在一起给出x_p的两层特征：

层	对象	性质	ZFCρ来源	SAE候选对应
Increment / high-pass	d_p = Δx_p	二阶相关快速衰减 / ACF曲率小	Paper 63	不得不发展
Level / low-pass	x_p	幂律ACF + screening	Paper 53/54	不得不有方向（level编码记忆）

59.1 = Zero-mode pinning： 在increment二阶正交 + level screened的联合特征下，59.1要求level的零频模式被pin到O(j^{3/2}·√D_j)精度。这是SAE"不得不被凿"（第三律）的候选数学对应。

6.6. 与经典随机过程理论的连接

x_p的过程特征（差分near-white + level有长程幂律ACF）不符合fBM/fOU的标准框架——这些要求stationary-increment结构，而x_p的差分是白的（fBM的increment应有特定的ACF结构）。更准确的对应是：x_p像martingale + slow-drift，其中d_p是near-white的martingale increment，drift来自screening的mean-reversion。

最有操作性的经典框架：Gordin martingale-coboundary decomposition。 Paper 52已经提出这个方向。x_p = martingale part + coboundary part。Innovation Orthogonality说明martingale increment近白，而coboundary部分提供mean-reversion（screening）。

最有操作性的路线： Gordin decomposition + screening rate。如果screening给出的mean-reversion足够快（exponential mixing on O(j) scale），则martingale CLT给出Var(B_j) = Var(Σx_p) ≈ D_j · Ψ（因为D_j = m·σ²_x，而Ψ是相关放大因子）。在L²/typical-size意义下，Ψ = O(j)给出与59.1相同的方差尺度；要推出确定性的59.1，还需要pathwise/maximal或deterministic upgrade。Paper 62实测Ψ_x(W=2000) ≈ 50（j=28），量级与O(j)一致。

剩余gap： 证明Ψ = O(j)。这归结为证明screening rate在block尺度上给出O(j)倍的相关放大——不多不少。Innovation Orthogonality本身不能给这个bound（Abel路线已被否决），但它把问题精确化为：IC递推的screening mechanism在Gordin decomposition下给出什么样的mixing rate？

---

§7. 路线否决汇总

7.1. 六条死路线

#	路线	死因	实验来源
1	线性递推（Route A）	谱半径1.20 > 1	P62 EXP C/D/E
2	Screening-CLT（Route C）	Ψ增长，n_cross高有效幂指数（全局拟合约j^{10}）	P62 EXP C/D/E
3	负协方差	Cov(B_add,B_mult) > 0	P62 EXP J/K/L
4	Mertens速率	O(1/j)太弱	P62 EXP F/G/H
5	Transport配对	配对ACF放大3.3×	P63 EXP I
6	Naive Abel + independent-increment	std/target = 2^j/j^{3/2}	定量分析

7.2. 每条路线杀死了什么假说

• Route 1：z_j不是线性收缩映射的输出
• Route 2：screening不提供足够的去相关来支撑CLT
• Route 3：add/mult两河局部正相关，方差不缩减
• Route 4：两河各自的Mertens收敛精度不够
• Route 5：配对去高频但暴露低频——低通效应
• Route 6：差分二阶正交 ≠ 部分和小（naive integrated-increment放大）

7.3. 每条路线发现了什么新结构

• Route 1 → z_j有界但机制是非线性的
• Route 2 → n_cross(j)增长比预期快（有效幂指数2-3）
• Route 3 → Cov > 0确认两河共享同一ρ景观
• Route 4 → C₁和C₂收敛到同一极限L ≈ 0.092
• Route 5 → 普适配对常数39%/62%/80%/87%，局部前缀和匹配
• Route 6 → Innovation Orthogonality（d_i二阶近正交）

---

§8. 59.1的攻击路线图更新

8.1. 59.1的精确数学内容（进化）

Paper 59： R_wt = O(1)（不透明的谱比有界性）

Paper 62： |Δ_j| = O(j^{3/2}·√D_j)（双河同步精度）

Paper 63： differential level的零模pinning问题——配对消除高频，差分消除所有相关，唯独零频/level-mode不受控制。59.1要求的恰好是level-mode的pinning。

8.2. 存活方向

(a) Common-mode Q_j共享证明（Branch-Orthogonality原则）。 证明add/mult路径选择与block尺度ρ振荡之间正交——类似P61的几何-算术正交性（dlp权重 ⊥ h(p)相关结构），这里是branch indicator ⊥ block-scale drift。局部前缀和匹配（§3）是这个orthogonality的强实验支持。难度：中等。实验基础最强。Paper 64首选方向。

(b) W-59.1（弱版，足以关闭H'）。 log R_wt = o(j)。排除differential level的指数增长。可能比(a)更短。难度：较低。比(a)弱但可能够H'。

(c) Dispersion method（Vaughan identity/Linnik dispersion）。 经典解析数论工具的natural精度是√x ≈ 2^{j/2}——恰好是59.1需要的精度。如果h(p)能拆成dispersion-compatible form（major arc + minor arc），可以engage经典machinery。这是所有候选路线中精度天然匹配59.1的唯一经典工具。难度：高。需要大量经典解析数论machinery。Paper 65/66候选。

(d) Coboundary/pinning。 寻找势函数Φ使differential residual e_i = Φ_{i+1} - Φ_i + r_i，则Σe_i telescopes到边界项+小残差。差分二阶正交性支持这种decomposition存在。难度：与(a)相当，技术路径不同。

P64方向建议： Track A = (a) Branch-Orthogonality形式化。这是余项守恒公理在block尺度的精确数学表达，实验基础（§3局部前缀和匹配）已经很强。若(a)成功，59.1变成Gordin decomposition下的mixing rate估计问题。

8.3. 与经典解析数论的连接

59.1的双河同步形式|Δ_j| = O(j^{3/2}·√D_j)将H'的最后一步放入经典prime-sum cancellation类问题的target空间。所需精度2^{-j/2}对应√x级别——已超过Mertens（log log x级），进入Chebyshev-Riemann精度层。攻击工具包括dispersion method、large sieve、Vaughan identity等。

---

§9. 闭合链更新

局部 GAO（P62 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'

59.1的精确内容（三层进化）：

层级	表述	来源
谱比层	R_wt = O(1)	Paper 59
双河层		Δ_j	= O(j^{3/2}·√D_j)	Paper 62
零模层	differential level pinning + coboundary	Paper 63

---

§10. 结论

Paper 63以Transport配对为切入点，发现了两个方向相反但同样重要的结果，并由此到达了ZFCρ系列与SAE公理框架的候选交汇点。

负结果： Transport配对和naive Abel + independent-increment路线均被否决。至此六条候选路线全部被实验或定量分析否决。每条死路线都留下了一个新的结构事实。

正结果： 一阶差分d_i的二阶相关在lag ≥ 2处极小（余项增量二阶正交性 / Innovation Orthogonality）。由差分协方差恒等式γ_d(l) = 2γ_x(l) - γ_x(l-1) - γ_x(l+1)解释为level ACF的离散曲率抑制。这说明高频增量层二阶近正交——但不等同于iid白噪声。

SAE候选对应： 增量二阶正交性为SAE公理"余项必须发展"提供了候选数学对应。与Paper 53/54的screening（"余项守恒"的候选对应）合并，形成x_p过程的两层特征：increment层（二阶正交/curvature-suppressed）+ level层（幂律ACF + screening）。59.1在此特征下变成一个精确的数学问题：screened level process的zero-mode是否被pin在O(j^{3/2}·√D_j)范围内？

经典对应： 最接近的经典框架是Gordin martingale-coboundary decomposition。IC递推的screening mechanism给出什么样的mixing rate——这是59.1的最终技术内核。

从SAE视角：余项可以发展（d_i ≠ 0），增量层二阶正交（差分ACF曲率小）；余项的level携带长程记忆但被screening约束。59.1可以被解释为level-memory的pinning足够强——但pinning强度仍是待证内容。Paper 63不改变闭合链的逻辑假设；它进一步定位59.1的内部结构。

Paper 64的候选主攻方向： Branch-Orthogonality Principle的形式化——证明IC递推中add/mult branch选择与block-scale ρ oscillation的渐近独立性。§3的局部前缀和匹配已经提供了强实验基础。

---

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。 [52] 第LII篇。Martingale-coboundary与Gordin分解路线。 [53] 第LIII篇。Screening与相关截断。DOI: 10.5281/zenodo.19415440。 [54] 第LIV篇。递归祖先继承与幂律ACF。DOI: 10.5281/zenodo.19426094。 [55] 第LV篇。阻尼振荡与76%继承。DOI: 10.5281/zenodo.19447349。 [59] 第LIX篇。谱比有界性与H'的双猜想闭合。DOI: 10.5281/zenodo.19480518。 [61] 第LXI篇。谱导数界的条件归约。DOI: 10.5281/zenodo.19605749。 [62] 第LXII篇。双河同步。DOI: 10.5281/zenodo.19656365。

---

AI贡献

Claude（子路）：全部数值实验设计与实现（paper63_fgh.c branch-gap与前缀和实验、paper63_transport.c确定性配对实验、paper63_acf.c配对残差ACF实验），配对距离1.06的识别，普适常数39%/62%/80%/87%的发现，配对ACF放大效应的发现，Abel路线的定量否决（std/target = 2^j/j^{3/2}），Innovation Orthogonality的命名，working notes v1-v2。ChatGPT（公西华）：确定性transport配对方向的原始提出（Paper 62 §7.4），零模pinning/coboundary作为59.1正确框架的识别，Abel求和的频域否决论证（差分杀θ=0但B_j就是θ=0），local two-river balance框架，Haar/multiresolution视角，coboundary decomposition e_i = Φ_{i+1} - Φ_i + r_i的提出，六个后续实验建议。Grok（子贡）：First-Difference + Abel Summation路线的提出（后被定量否决），Two-River common-mode Q_j共享证明的框架建议。Gemini（子夏）：Abel边界项m·x_m = O(1/j)的定量分析，局部测度定理方向的建议，对消率从ACF导出的初步分析。独立评审Claude：Abel路线的leading-order定量否决（std/target = 2^j/j^{3/2}），dispersion method作为精度天然匹配59.1的经典工具的识别，Innovation Orthogonality引理的形式化建议（lemma而非axiom），Track A（innovation formalization）vs Track B（dispersion method）的Paper 64方向建议。