Fractal Geometric Analysis of the Twin Prime Counting Function via the Fractal Correction Engine
Authors/Creators
Description
# Fractal Geometric Analysis of the Twin Prime Counting Function via the Fractal Correction Engine
**Authors:** Adam L McEvoy
**Date:** March 23, 2026
**Keywords:** twin prime conjecture, fractal dimension, Hardy-Littlewood, Brun's constant, number theory, spectral analysis, curvature, golden ratio, Hurst exponent, sieve interference, Riemann zeta zeros
---
## Abstract
I present the first fractal geometric analysis of the twin prime counting function $\pi_2(x)$ using the Fractal Correction Engine (FCE), a computational framework based on $\pi$-normalized curvature, multi-scale Gaussian decomposition, and Riemann zeta-inspired weighting. Applied to all twin prime pairs $(p, p+2)$ with $p \leq 10^6$, we characterize the fractal structure of $\pi_2(x)$ (Higuchi $D_H = 1.4895$, box-counting $D_{BC} = 1.0165$), establish its strongly persistent dynamics (Hurst $H = 0.9995$), and identify statistically significant golden ratio dominance in the detrended counting function ($\varphi$-correlation $= 0.5399$, $p = 0.002$). We construct an enhanced ensemble predictor that achieves a mean absolute error (MAE) of 1037.64, outperforming the classical Hardy-Littlewood asymptotic prediction (MAE $= 1229.95$) by a factor of $1.19\times$. Log-space spectral analysis reveals strong power concentration at Riemann zeta zero frequencies ($280\times$ median background), providing empirical evidence of the explicit formula connection to twin prime distribution. A novel density-based sieve interference mapping identifies 1,490 constructive and 1,633 destructive zones where twin prime density deviates from Hardy-Littlewood expectations. Analysis of 11 derived waveforms reveals that 6 exhibit golden ratio dominance, with the log-derivative $\frac{d}{dx}\ln\pi_2$ showing the strongest $\varphi$-correlation (0.665). We do not claim to prove the twin prime conjecture; rather, we demonstrate that $\pi_2(x)$ possesses rich, predictable internal structure beyond what classical analytic number theory captures, and that this structure is consistent with the fractal geometric patterns found in our companion analyses of the Goldbach comet and Riemann zeta zeros.
---
## 1. Introduction
### 1.1 The Twin Prime Conjecture
The twin prime conjecture asserts that there are infinitely many prime pairs $(p, p+2)$ where both $p$ and $p+2$ are prime. Despite being one of the most famous open problems in number theory, the conjecture remains unproven. The strongest partial result is Zhang's theorem (2014), which proved there are infinitely many prime pairs with gap at most 70 million, subsequently reduced to 246 by the Polymath project (2014) using Maynard-Tao sieve methods.
For a given bound $x$, we define the twin prime counting function:
$$\pi_2(x) = \#\{p \leq x : p \text{ and } p+2 \text{ are both prime}\}$$
The function $\pi_2(x)$ is a monotonically non-decreasing staircase function that increases by 1 each time a new twin prime pair is encountered. The distribution of its steps, gaps, and deviations from the expected growth rate encodes deep arithmetic information about the interaction between the prime sieve and the additive structure of integers.
### 1.2 The Hardy-Littlewood Conjecture
Hardy and Littlewood (1923) conjectured an asymptotic formula for $\pi_2(x)$ based on the circle method and singular series analysis:
$$\pi_2(x) \sim 2C_2 \int_2^x \frac{dt}{\ln^2 t}$$
where $C_2$ is the twin prime constant:
$$C_2 = \prod_{p > 2} \left(1 - \frac{1}{(p-1)^2}\right) \approx 0.6601618158$$
For computational purposes, the leading-order approximation is often used:
$$\pi_2(x) \approx 2C_2 \frac{x}{\ln^2 x}$$
This formula captures the overall growth rate but fails to account for the fine-scale oscillatory structure in the residual $\pi_2(x) - \text{HL}(x)$, which contains arithmetic information about how twin primes cluster and rarefy across the number line.
### 1.3 Brun's Constant
Viggo Brun (1919) proved that the sum of reciprocals of twin primes converges:
$$B_2 = \sum_{(p, p+2) \in \mathcal{T}} \left(\frac{1}{p} + \frac{1}{p+2}\right) \approx 1.9021605831$$
where $\mathcal{T}$ is the set of all twin prime pairs. This convergence holds whether or not the twin prime conjecture is true, but the rate of convergence and the oscillatory approach to $B_2$ carry structural information about twin prime distribution. The partial sums $B_2(n) = \sum_{(p,p+2) \leq n} (1/p + 1/(p+2))$ form a smooth, monotonically increasing waveform whose fractal properties we analyze in this work.
### 1.4 The Double Sieve Problem
Twin primality imposes a conjunctive constraint: both $p$ and $p+2$ must survive the Eratosthenes sieve. For each prime $q$, the sieve eliminates multiples of $q$ from the candidate pool. For an integer to be part of a twin prime pair, it must avoid all prime divisors, and its neighbor at distance 2 must independently avoid all prime divisors. This creates a wave interference pattern:
$$\text{twin}(n) = \text{sieve}_p(n) \cdot \text{sieve}_p(n+2)$$
where $\text{sieve}_p(n) = 1$ if $n$ is prime and 0 otherwise. The product structure means twin primes occur at the constructive interference of two shifted copies of the prime indicator function. Regions where local twin prime density exceeds the Hardy-Littlewood expectation correspond to constructive interference zones, while regions of deficit correspond to destructive interference.
### 1.5 Motivation and Scope
While the Hardy-Littlewood formula captures the leading-order behavior of $\pi_2(x)$, the residuals display rich structure that has not been systematically characterized using modern tools from fractal geometry and signal processing. This work asks:
1. What is the fractal dimension of the twin prime counting function?
2. Does $\pi_2(x)$ exhibit long-range dependence or memory?
3. Can fractal-geometric methods predict $\pi_2(x)$ more accurately than Hardy-Littlewood?
4. Is there a spectral connection between twin prime distribution and Riemann zeta zeros?
5. Where do constructive and destructive sieve interference zones occur?
6. Does the golden ratio appear in the self-similarity structure of $\pi_2(x)$?
I address these questions using the Fractal Correction Engine (FCE), a computational framework originally developed for trajectory analysis in dynamical systems. This paper constitutes the fourth in a series connecting Riemann zeta structure, computational complexity, the Goldbach comet, and twin prime distribution through the FCE framework.
---
## 2. The Fractal Correction Engine (FCE)
### 2.1 Overview
The Fractal Correction Engine is a signal analysis and prediction framework that operates on discrete sequences through three core mechanisms:
1. **$\pi$-normalized curvature analysis** --- measuring local geometric complexity relative to a canonical semicircular arc
2. **Multi-scale Gaussian decomposition** --- separating structure at $\pi$-scaled resolution levels
3. **Zeta-weighted correction** --- combining detail levels with Riemann zeta-inspired weights
The FCE treats any discrete sequence $f(x_1), f(x_2), \ldots, f(x_N)$ as a geometric trajectory and applies differential-geometric tools to characterize and predict its behavior.
### 2.2 Discrete Curvature
For a discrete sequence $f$, the curvature at point $i$ is computed via finite differences:
$$\kappa(i) = \frac{|f''(i)|}{(1 + (f'(i))^2)^{3/2}}$$
where $f'(i)$ and $f''(i)$ are the first and second discrete derivatives computed via central differences. The denominator normalizes for slope, ensuring curvature measures intrinsic bending rather than steepness. This is the standard Frenet curvature for a planar curve parametrized by arc length.
### 2.3 $\pi$-Normalized Curvature
The curvature is normalized by $\pi$ to produce a dimensionless measure:
$$\kappa_{\text{norm}}(i) = \frac{\kappa(i)}{\pi}$$
This normalization has geometric meaning:
- $\kappa_{\text{norm}} = 1$: the trajectory bends as tightly as a semicircular arc of unit radius
- $\kappa_{\text{norm}} \to 0$: the trajectory is locally straight
- $\kappa_{\text{norm}} > 1$: the trajectory bends more sharply than a semicircle
The radius of curvature is the reciprocal:
$$r(i) = \frac{1}{\kappa(i) + \epsilon}$$
where $\epsilon = 10^{-30}$ prevents division by zero. To prevent numerical overflow when $\kappa \to 0$ (near-straight segments), the radius is clamped at the signal's standard deviation:
$$r(i) = \min\left(\frac{1}{\kappa(i) + \epsilon},\; \sigma_f\right)$$
where $\sigma_f = \text{std}(f)$. This ensures corrections scale with the signal amplitude rather than the sequence length, preventing overshoot on smooth or monotonic waveforms.
### 2.4 Multi-Scale Gaussian Decomposition
The FCE decomposes the input sequence at $\pi$-scaled resolution levels using Gaussian convolution. For $n_s$ scales (default $n_s = 5$), the smoothing widths are:
$$\sigma_k = \pi \cdot 2^k, \quad k = 0, 1, \ldots, n_s$$
At each scale, the sequence is convolved with a normalized Gaussian kernel:
$$G_{\sigma}(x) = \frac{1}{\sqrt{2\pi}\,\sigma} \exp\!\left(-\frac{x^2}{2\sigma^2}\right)$$
The smoothed versions are $S_k = G_{\sigma_k} * f$, and the detail levels capture the structure between consecutive scales:
$$D_k = S_{k-1} - S_k, \quad k = 1, \ldots, n_s$$
Each $D_k$ isolates oscillatory structure at spatial frequencies corresponding to the scale $\sigma_k$. The $\pi$-based scaling creates a natural hierarchy where each level captures structure at a factor of 2 coarser resolution, with the fundamental scale set by $\pi$.
### 2.5 The FCE Correction Formula
The core FCE correction combines curvature, multi-scale details, and Riemann zeta-inspired weighting:
$$f_{\text{corrected}}(i) = f(i) + \alpha \cdot \pi \cdot r(i) \cdot \sum_{k=1}^{n_s} \frac{D_k(i)}{k^{3/2}}$$
where:
- $\alpha = 0.01$ is the correction strength
- $\pi$ is the normalization constant
- $r(i)$ is the clamped radius of curvature
- $D_k(i)$ is the $k$-th detail level at point $i$
- $k^{-3/2}$ provides zeta-inspired weighting (cf. $\zeta(3/2) = \sum_{k=1}^{\infty} k^{-3/2} \approx 2.612$), giving more weight to finer scales
The correction is larger where the sequence is locally straight ($r$ large) and rich in multi-scale detail ($D_k$ large), and smaller where the sequence already curves sharply ($r$ small). This implements a geometric principle: straight segments contain "hidden" structure at multiple scales that the correction reveals, while already-curved segments are closer to their fractal path.
### 2.6 Fractal Path Extraction
The FCE extracts a "fractal path" from an observed sequence through iterative refinement:
$$\hat{f}^{(0)} = \text{FCE}(f)$$
$$\hat{f}^{(t+1)} = \hat{f}^{(t)} + \frac{1}{2}\,\text{FCE}\!\left(f - \hat{f}^{(t)}\right), \quad t = 0, 1, 2$$
After 3 iterations, the path match is evaluated via:
- **Pearson correlation** $\rho(\hat{f}, f)$: measures shape similarity
- **Hausdorff-like distance** $d_H = \max|f - \hat{f}| / \text{std}(f)$: measures maximum normalized deviation
For the twin prime counting function, fractal path extraction is applied to the detrended Hardy-Littlewood residual rather than the raw monotonic $\pi_2(x)$, because the correction mechanism requires oscillatory input with meaningful curvature variation.
### 2.7 Configuration Parameters
| Parameter | Symbol | Default | Description |
|-----------|--------|---------|-------------|
| Correction strength | $\alpha$ | 0.01 | Amplitude of correction term |
| Decay timescale | $\tau$ | 50.0 | Exponential decay in extrapolation |
| Oscillation frequency | $\omega$ | 0.1 | Frequency for oscillatory predictions |
| Number of scales | $n_s$ | 5 | Multi-scale decomposition depth |
| Prediction horizon | $H_p$ | 500 | Forward prediction range |
| Higuchi max lag | $k_{\max}$ | 200 | Maximum lag for Higuchi FD |
---
## 3. Methods
### 3.1 Computational Pipeline
The analysis follows an 8-stage pipeline:
1. **Twin Prime Sieve** --- Sieve of Eratosthenes up to $n_{\max} = 10^6$, twin pair identification
2. **Twin Prime Analyzer** --- $\pi_2(x)$ computation, Hardy-Littlewood comparison, Brun's constant, desert/cluster detection
3. **FCE Fractal Analysis** --- curvature, fractal dimensions, predictions, 11 sub-waveform decompositions
4. **Sieve Interference Engine** --- double sieve pattern, density-based constructive/destructive zones
5. **Spectral Analysis** --- FFT, Lomb-Scargle, log-space FFT, power-law fitting, zeta-zero correlation
6. **Statistical Validation** --- bootstrap CIs, permutation tests, OEIS verification
7. **Visualization** --- 12 diagnostic plots
8. **Report Generation** --- JSON metrics, text report, CSV data
System health is monitored continuously via CPU temperature and memory pressure sensors, with automatic worker throttling above 85$^\circ$C.
### 3.2 Twin Prime Sieve and Counting Function
The sieve of Eratosthenes generates a boolean array $\text{is\_prime}[0..n_{\max}+2]$. Twin prime pairs are identified by the conjunctive condition:
$$\text{twin}(p) = \text{is\_prime}(p) \wedge \text{is\_prime}(p+2)$$
The counting function is evaluated at $N = 10{,}000$ uniformly spaced sample points $x_i \in [2, n_{\max}]$ with spacing $\Delta x = 100$:
$$\pi_2(x_i) = \#\{p \leq x_i : \text{twin}(p)\}$$
Derived sequences include:
- **Gap sequence**: $g_k = p_{k+1} - p_k$ where $p_k$ is the $k$-th twin prime (first of pair)
- **Brun partial sums**: $B_2(n) = \sum_{(p,p+2) \leq n} (1/p + 1/(p+2))$
- **Twin prime indicator**: binary sequence marking twin prime positions
- **Twin prime density**: local rate $\frac{d\pi_2}{dx}$ via finite differences
### 3.3 Hardy-Littlewood Prediction
For each sample point $x_i$, the Hardy-Littlewood prediction is computed using both the leading-order approximation and numerical integration:
$$\text{HL}(x) = 2C_2 \frac{x}{\ln^2 x}$$
$$\text{HL}_{\text{int}}(x) = 2C_2 \int_2^x \frac{dt}{\ln^2 t}$$
The residual waveform $r(x) = \pi_2(x) - \text{HL}(x)$ isolates the oscillatory component that the smooth asymptotic formula cannot capture.
### 3.4 Fractal Dimension Estimation
#### 3.4.1 Higuchi Fractal Dimension
The Higuchi method (1988) estimates fractal dimension from a time series without phase-space embedding. For a sequence $x_1, x_2, \ldots, x_N$ and lag $k$:
1. Construct subsequences for each offset $m = 1, \ldots, k$:
$$x_m^{(k)} = x_m, x_{m+k}, x_{m+2k}, \ldots$$
2. Compute the normalized path length:
$$L_m(k) = \frac{1}{k^2} \cdot \frac{N - 1}{\lfloor(N-m)/k\rfloor} \sum_{i=1}^{\lfloor(N-m)/k\rfloor} |x_{m+ik} - x_{m+(i-1)k}|$$
3. Average over offsets: $L(k) = \frac{1}{k} \sum_{m=1}^{k} L_m(k)$
4. The fractal dimension $D_H$ is the negative slope of the log-log regression:
$$\log L(k) = -D_H \log k + c$$
I use $k_{\max} = \min(200, N/4)$ and clamp $D_H \in [1.0, 2.0]$.
**Crucially**, the Higuchi dimension is computed on the **detrended** $\pi_2(x)$ (cubic polynomial trend removed) to measure intrinsic oscillatory complexity rather than monotonic growth.
#### 3.4.2 Box-Counting Fractal Dimension
The box-counting method estimates the dimension of the point cloud $(x_i, \pi_2(x_i))$:
1. Normalize coordinates to $[0,1]^2$
2. For 20 logarithmically spaced box sizes $\varepsilon \in [10^{-2.5}, 10^{-0.3}]$:
- Count the number of non-empty cells $N(\varepsilon)$
3. Fit: $\log N(\varepsilon) = -D_{BC} \log \varepsilon + c$
### 3.5 Hurst Exponent (R/S Analysis)
The Hurst exponent $H$ measures long-range dependence via rescaled range analysis (Hurst, 1951):
1. Detrend the series by subtracting a fitted cubic polynomial
2. For window sizes $W$ from 10 to $N/2$ (geometric progression with factor 1.5):
- Compute the rescaled range $R/S$ for each window
- $R = \max(Y) - \min(Y)$ where $Y_t = \sum_{i=1}^{t}(x_i - \bar{x})$
- $S = \text{std}(x)$ within the window
3. Fit the scaling law:
$$\log(R/S) = H \log W + c$$
Interpretation:
- $H > 0.5$: persistent (trending) behavior --- large values tend to follow large values
- $H = 0.5$: random walk with no memory
- $H < 0.5$: anti-persistent (mean-reverting) behavior
### 3.6 Golden Ratio Self-Similarity Analysis
The FCE tests whether autocorrelation peak spacings preferentially align with the golden ratio $\varphi = (1 + \sqrt{5})/2 \approx 1.618$:
1. Compute the autocorrelation via FFT ($O(N \log N)$):
$$R(\tau) = \mathcal{F}^{-1}\!\left[|\mathcal{F}[f - \bar{f}]|^2\right]$$
normalized so $R(0) = 1$.
2. Identify peaks: local maxima of $R(\tau)$ with $R > 0.05$
3. Compute successive peak-spacing ratios:
$$\rho_i = \frac{\tau_{i+1}}{\tau_i}$$
4. Measure proximity to $\varphi$ and $\pi$:
$$\text{corr}_\varphi = \exp\!\left(-\frac{1}{M}\sum_{i=1}^{M}|\rho_i - \varphi|\right)$$
$$\text{corr}_\pi = \exp\!\left(-\frac{1}{M}\sum_{i=1}^{M}|\rho_i - \pi|\right)$$
5. $\varphi$-dominance holds if $\text{corr}_\varphi > \text{corr}_\pi$
For the twin prime counting function, this analysis is applied to the **detrended** $\pi_2(x)$ rather than the raw monotonic signal, because monotonic autocorrelation decays without peaks. The detrended signal (cubic polynomial trend removed) retains the oscillatory structure that produces meaningful peak spacings.
### 3.7 Double Sieve Interference Mapping
Twin primes exist at the conjunction of two sieve conditions. We map constructive and destructive interference zones by comparing the local twin prime density against the Hardy-Littlewood expected density.
#### 3.7.1 Density Computation
The local twin prime density is computed via finite differences:
$$\rho_{\text{local}}(x) = \frac{d\pi_2}{dx} \approx \frac{\pi_2(x + \Delta x) - \pi_2(x - \Delta x)}{2\Delta x}$$
The expected density from Hardy-Littlewood is:
$$\rho_{\text{HL}}(x) = \frac{2C_2}{\ln^2 x}$$
#### 3.7.2 Zone Classification
The local density is smoothed with a box kernel of width $w = 10$ sample points to filter single-point noise while preserving sustained deviations. The density ratio classifies zones:
$$R(x) = \frac{\rho_{\text{local}}(x)}{\rho_{\text{HL}}(x)}$$
- **Constructive zones**: $R(x) > 1.3$ (local density exceeds H-L by 30%+)
- **Destructive zones**: $R(x) < 0.7$ (local density falls below H-L by 30%+)
### 3.8 Enhanced Prediction Strategies
The FCE employs three prediction strategies, each predicting $\pi_2(x)$ for future values beyond the training range. All strategies model the Hardy-Littlewood residual $r(x) = \pi_2(x) - \text{HL}(x)$.
#### 3.8.1 Bias Correction
Smooths the H-L residual with a Gaussian kernel and extrapolates via curvature-based trajectory prediction:
$$\hat{\pi}_2(x) = \text{HL}(x) + \text{extrapolate}(\text{smooth}(r_{\text{train}}))$$
#### 3.8.2 Fractal Path Extrapolation
Applies the FCE correction to the residual and extrapolates the resulting path:
$$\hat{f} = \text{FCE}^{(3)}(r_{\text{train}})$$
$$\hat{\pi}_2(x) = \text{HL}(x) + \text{extrapolate}(\hat{f})$$
#### 3.8.3 Multi-Scale $\pi$ Prediction
Decomposes the residual at $\pi$-based scales and extrapolates each level independently:
$$\hat{r}(x) = \frac{\sum_{k} w_k \cdot \hat{D}_k(x)}{\sum_{k} w_k}, \quad w_k = \frac{1}{k+1}$$
#### 3.8.4 Validation-Weighted Ensemble
Strategies are combined via inverse-MAE-squared weighting:
$$w_s = \begin{cases} (\text{MAE}_s^2 + \epsilon)^{-1} & \text{if } \text{MAE}_s < \text{MAE}_{\text{HL}} \\ 0 & \text{otherwise} \end{cases}$$
Normalized: $\hat{w}_s = w_s / \sum_s w_s$. Only strategies that beat Hardy-Littlewood on validation data receive nonzero weight.
### 3.9 Spectral Analysis
#### 3.9.1 FFT Power Spectrum
The centered, Hann-windowed sequence is transformed:
$$P(f_k) = \left|\sum_{i=0}^{N-1} w_i \cdot (\pi_{2,i} - \overline{\pi_2}) \cdot e^{-2\pi i k i / N}\right|^2$$
where $w_i$ is the Hann window function and $f_k = k/N$.
#### 3.9.2 Power-Law Fitting
A power-law relationship $P(f) \propto f^{-\beta}$ is fit via log-log linear regression:
$$\log P = -\beta \log f + c$$
The exponent $\beta$ characterizes the spectral color: $\beta \approx 0$ (white noise), $\beta \approx 1$ (pink noise), $\beta \approx 2$ (Brownian noise).
#### 3.9.3 Log-Space FFT for Zeta-Zero Detection
The connection between twin primes and Riemann zeta zeros operates through the explicit formula. The oscillatory terms in the prime counting function involve $x^\rho$ where $\rho = 1/2 + i\gamma$ are the non-trivial zeta zeros. Since:
$$x^{1/2 + i\gamma} = x^{1/2} \cdot e^{i\gamma \ln x}$$
the zeta zeros contribute oscillations at frequency $\gamma/(2\pi)$ in **log-space**. Therefore, to detect the zeta-zero spectral signature in $\pi_2(x)$, we:
1. Resample $\pi_2(x)$ at $x = e^t$ for $N_s = 4096$ uniformly spaced values of $t \in [\ln 10, \ln 10^6]$
2. Subtract the Hardy-Littlewood baseline to isolate the oscillatory component
3. Remove quadratic trend and apply a Hann window
4. Compute the FFT in log-space, where the frequency axis maps directly to $\gamma/(2\pi)$
I use the first 100 non-trivial zeta zeros (up to $\gamma_{100} \approx 236.52$), sourced from Odlyzko's tables.
#### 3.9.4 Zeta-Zero Correlation
For each zeta zero $\gamma_n$, the corresponding log-space frequency is:
$$f_n^{\zeta} = \frac{\gamma_n}{2\pi}$$
The spectral power at these frequencies (averaged over a $\pm 2$-bin window) is compared to the median background power:
$$\text{corr}_\zeta = \frac{\overline{P(f^{\zeta})}}{\text{median}(P)}$$
A ratio significantly above 1 indicates that the twin prime spectrum concentrates power at the frequencies predicted by the explicit formula connection to zeta zeros.
### 3.10 Statistical Validation
#### 3.10.1 Bootstrap Confidence Intervals
For each statistic $\theta$ (Higuchi $D_H$, Hurst $H$):
1. Draw $B = 1000$ bootstrap samples of the gap sequence with replacement
2. Compute $\hat{\theta}_b$ for each sample
3. Report the 95% percentile interval: $[\hat{\theta}_{(\alpha/2)}, \hat{\theta}_{(1-\alpha/2)}]$
#### 3.10.2 Permutation Tests
**$\varphi$-dominance test** ($n_{\text{perm}} = 500$): Under $H_0$, the observed $\varphi$-correlation arises by chance. The $p$-value is:
$$p = \frac{\#\{\text{corr}_\varphi^{\text{perm}} \geq \text{corr}_\varphi^{\text{obs}}\} + 1}{n_{\text{perm}} + 1}$$
**FCE prediction test** ($n_{\text{perm}} = 500$): Under $H_0$, the FCE prediction MAE is no better than prediction from shuffled training data.
#### 3.10.3 OEIS Verification
The first 50 computed twin primes are verified against the OEIS sequence A001359 (lesser of twin primes):
$$3, 5, 11, 17, 29, 41, 59, 71, 101, 107, \ldots$$
#### 3.10.4 Information Horizon Test
MAE is evaluated at increasing prediction horizons $h \in \{50, 100, 200, 500\}$. A sharp discontinuity in MAE growth rate indicates a fundamental predictability limit.
### 3.11 Sub-Waveform Analysis
The FCE analysis is applied to 11 derived waveforms, each isolating a different aspect of twin prime structure:
| Waveform | Definition | Detrended |
|----------|-----------|-----------|
| H-L Residual | $\pi_2(x) - \text{HL}(x)$ | Yes |
| Twin-Prime Gaps | $g_k = p_{k+1} - p_k$ | Yes |
| Brun Partial Sums | $B_2(n)$ | Yes |
| Twin-Prime Density | $d\pi_2/dx$ | Yes |
| Acceleration | $d^2\pi_2/dx^2$ | Yes |
| Log-Derivative | $\frac{d}{dx}\ln\pi_2$ | No |
| Brun Deviation | $B_2(n) - B_2$ | No |
| Cumulative Gap Sum | $\sum_{k=1}^n g_k$ | Yes |
| $\pi_2$/H-L Ratio | $\pi_2(x)/\text{HL}(x)$ | No |
| Spectral Envelope | $|\text{FFT}(\pi_2)|$ | No |
| Gap mod 6 = 0 | $\{g_k : g_k \equiv 0 \pmod{6}\}$ | Yes |
Each waveform receives the full FCE treatment: Higuchi $D_H$, Hurst $H$, $\varphi$-correlation, and forward/backward prediction MAE.
---
## 4. Results
All results are from a single run with $n_{\max} = 1{,}000{,}000$, sampling $\pi_2(x)$ at 10,000 uniformly spaced points. Computation time: 461.6 seconds on a multi-core system. The twin prime conjecture holds for all tested values.
### 4.1 Twin Prime Statistics
| Metric | Value |
|--------|-------|
| Primes up to $10^6$ | 78,498 |
| Twin prime pairs | 8,169 |
| $\pi_2(10^6)$ | 8,169 |
| Hardy-Littlewood prediction | 6,917.5 |
| Ratio $\pi_2/\text{HL}$ | 1.181 |
| Largest twin pair | (999959, 999961) |
| OEIS A001359 match | Pass (50/50) |
The counting function exceeds the Hardy-Littlewood prediction by 18.1%, consistent with known finite-size effects where the asymptotic formula systematically underestimates for $x \leq 10^6$.
### 4.2 Gap Sequence Statistics
| Metric | Value |
|--------|-------|
| Number of gaps | 8,168 |
| Mean gap | 122.42 |
| Gap standard deviation | 113.64 |
| Maximum gap | 1,452 |
| Twin prime deserts (gap > $2\sigma$) | 745 |
| Largest desert | $[850349, 851801]$ (gap = 1452) |
| Twin prime clusters | 5,582 |
The gap distribution is right-skewed with a heavy tail, consistent with the Poisson-like behavior expected for prime-derived sequences. The coefficient of variation $\sigma/\mu = 0.928$ indicates high variability relative to the mean.
### 4.3 Brun's Constant Convergence
| Metric | Value |
|--------|-------|
| $B_2(10^6)$ | 1.7107769308 |
| $B_2$ (target) | 1.9021605831 |
| Deviation | $-0.1914$ |
| Convergence rate | 0.1143 |
The partial sum $B_2(10^6)$ is still 10.1% below the asymptotic value, converging as $|B_2(n) - B_2| \sim n^{-0.114}$. This slow convergence rate is characteristic of the reciprocal sum of a sparse, irregular sequence.
### 4.4 Fractal Characterization
| Metric | Value | 95% CI |
|--------|-------|--------|
| Higuchi dimension $D_H$ (detrended) | 1.4895 | --- |
| Higuchi dimension $D_H$ (gaps) | 1.9990 | [1.9990, 2.0000] |
| Box-counting dimension $D_{BC}$ | 1.0165 | --- |
| Hurst exponent $H$ (raw $\pi_2$) | 0.9995 | --- |
| Hurst exponent $H$ (gaps) | 0.5413 | [0.5022, 0.5786] |
The twin prime counting function exhibits a multi-scale fractal character:
**Detrended $\pi_2(x)$ ($D_H = 1.49$):** The oscillatory component of the counting function has intermediate fractal dimension, between a smooth curve ($D = 1$) and a space-filling signal ($D = 2$). This indicates structured complexity --- not random noise, but not smooth either.
**Gap sequence ($D_H = 2.00$):** The twin prime gap sequence is maximally rough at fine scales, consistent with the pseudo-random nature of individual prime gaps. Each gap appears nearly independent of its neighbors.
**Box-counting ($D_{BC} = 1.02$):** The point cloud $(x, \pi_2(x))$ is essentially one-dimensional, reflecting the smooth monotonic envelope. The gap $D_H - D_{BC} = 0.47$ quantifies the difference between the detrended oscillatory complexity and the smooth global geometry.
**Hurst exponents:** The raw counting function has $H = 0.9995 \approx 1.0$, reflecting near-perfect persistence --- it is a monotonically increasing staircase. The gap sequence has $H = 0.5413$, marginally persistent but statistically consistent with a random walk ($H = 0.5$) at the 95% confidence level.
### 4.5 $\pi$-Normalized Curvature
| Metric | Value |
|--------|-------|
| Mean $\kappa / \pi$ | 0.0478 |
| Std $\kappa / \pi$ | 0.0484 |
The mean $\pi$-normalized curvature of 0.048 indicates that $\pi_2(x)$ is, on average, very gently curved --- bending at only 4.8% of the intensity of a semicircular arc. The equal mean and standard deviation ($\text{CV} \approx 1.01$) suggest an exponential-like distribution of curvature values, with occasional sharp bends (staircase steps) amid long gentle segments.
### 4.6 Golden Ratio Dominance
| Metric | Value |
|--------|-------|
| $\varphi$-correlation (detrended $\pi_2$) | 0.5399 |
| $\pi$-correlation | 0.1177 |
| $\varphi$-dominant | Yes |
| $p$-value (permutation, 500 shuffles) | 0.002 |
The detrended counting function's autocorrelation peak spacings align significantly more closely with the golden ratio than with $\pi$. The ratio $\text{corr}_\varphi / \text{corr}_\pi = 4.59$ indicates strong preferential alignment with $\varphi$.
This $\varphi$-dominance is consistent with findings from our Goldbach comet analysis ($\varphi$-corr = 0.5396, $p = 0.002$), suggesting a shared self-similarity structure in prime pair counting functions.
### 4.7 Fractal Path Validation
| Metric | Value |
|--------|-------|
| Path match correlation | 0.99998 |
| Path Hausdorff distance | 0.0368 |
Applied to the detrended H-L residual, the FCE fractal path extraction achieves near-perfect correlation with the observed signal. The Hausdorff distance of 0.037 standard deviations indicates the extracted path deviates from the observation by less than 4% of the signal's variability. This demonstrates that the FCE correction formula successfully captures the intrinsic geometric structure of the twin prime residual.
### 4.8 Prediction Performance
#### 4.8.1 Individual Strategy Results
| Strategy | Forward MAE | Beats H-L? | Ensemble Weight |
|----------|-------------|------------|-----------------|
| Fractal Path | 608.10 | Yes | 0.408 |
| Bias Correction | 611.03 | Yes | 0.406 |
| Multi-Scale | 1,333.03 | No | 0.186 |
| **H-L baseline** | **1,229.95** | --- | --- |
| **Ensemble** | **1,037.64** | **Yes** | --- |
#### 4.8.2 Forward vs. Backward Prediction
| Predictor | Forward MAE | Backward MAE |
|-----------|-------------|--------------|
| Original FCE | 1,564.54 | 271.53 |
| Hardy-Littlewood | 1,229.95 | 83.24 |
| Enhanced ensemble | 1,037.64 | 164.31 |
$$\text{Improvement factor} = \frac{\text{MAE}_{\text{HL}}}{\text{MAE}_{\text{FCE}}} = \frac{1229.95}{1037.64} = 1.185\times$$
The enhanced FCE ensemble outperforms Hardy-Littlewood by $1.19\times$ ($p = 0.002$). Unlike the Goldbach comet where modular structure dominates (4.84$\times$ improvement), the twin prime counting function's prediction gain comes from nearly equal contributions of the Fractal Path (40.8%) and Bias Correction (40.6%) strategies. This reflects the absence of a strong modular signal in $\pi_2(x)$ --- twin primes do not exhibit the period-3 arithmetic structure that Goldbach pairs inherit from the $n \bmod 6$ classification.
The forward/backward MAE ratio of 6.31 (ensemble) reflects the fundamental asymmetry between extrapolation and interpolation for monotonic counting functions.
### 4.9 Sieve Interference Mapping
| Metric | Value |
|--------|-------|
| Constructive zones | 1,490 |
| Destructive zones | 1,633 |
| Mean sieve cross-correlation | $-0.1099$ |
The density-based interference analysis identifies 1,490 regions where local twin prime density exceeds the Hardy-Littlewood expectation by more than 30%, and 1,633 regions where it falls below by more than 30%. The slight predominance of destructive zones (52.3%) is consistent with the known tendency for the prime distribution to exhibit more frequent rarefaction than clustering at this scale.
The negative mean cross-correlation ($-0.11$) between the $p$-sieve and $(p+2)$-sieve windows reflects genuine destructive interference: the condition that both $p$ and $p+2$ be prime is harder to satisfy than either condition alone, creating an anti-correlation between the two sieve waves.
### 4.10 Spectral Analysis
| Metric | Value |
|--------|-------|
| Dominant frequency | 0.0001 (period $\approx 10{,}000$) |
| Power-law exponent $\beta$ | 1.834 |
| Spectral entropy | 0.210 |
| Zeta-zero correlation (log-space) | 280.09 |
#### 4.10.1 Power-Law Spectrum
The power spectrum follows $P(f) \propto f^{-1.83}$, close to Brownian noise ($\beta = 2$). This is consistent with the strongly persistent Hurst exponent $H \approx 1.0$ of the raw counting function. The low spectral entropy (0.21) indicates that power is concentrated in low frequencies --- the counting function is dominated by its smooth growth, with progressively weaker oscillatory content at higher frequencies.
#### 4.10.2 Zeta-Zero Spectral Signature
The log-space FFT analysis reveals striking power concentration at Riemann zeta zero frequencies. The mean power at the 100 zeta zero frequencies is $280\times$ the median background power. This provides strong empirical evidence for the explicit formula connection:
$$\pi_2(x) \approx 2C_2 \text{Li}_2(x) - 2C_2 \sum_\rho \text{Li}_2(x^\rho) + \ldots$$
where $\rho = 1/2 + i\gamma$ ranges over the non-trivial zeros of $\zeta(s)$, and $\text{Li}_2$ is an appropriate twin prime analogue of the logarithmic integral. The oscillatory terms $x^\rho = x^{1/2} e^{i\gamma \ln x}$ contribute spectral peaks at frequency $\gamma/(2\pi)$ in log-space, which is precisely what the log-space FFT detects.
This represents, to our knowledge, the first direct spectral detection of the zeta-zero signature in the twin prime counting function via log-space Fourier analysis.
### 4.11 Sub-Waveform Analysis
The 11 sub-waveforms reveal differentiated fractal and predictive characteristics:
| Waveform | $D_H$ | $H$ | $\varphi$-corr | $\varphi$-dom | Fwd MAE | Bwd MAE |
|----------|--------|------|---------|----------|---------|---------|
| H-L Residual | 1.548 | 0.979 | 0.529 | Yes | 88.9 | 43.6 |
| Twin-Prime Gaps | 1.999 | 0.541 | 0.000 | No | 136.2 | 48.2 |
| Brun Partial Sums | 1.448 | 0.985 | 0.000 | No | 0.005 | 0.122 |
| Twin-Prime Density | 1.362 | 0.700 | 0.543 | Yes | 0.005 | 0.024 |
| $d^2\pi_2/dx^2$ | 2.000 | 0.166 | 0.553 | Yes | 1.2 | 2.1 |
| $\frac{d}{dx}\ln\pi_2$ | 1.916 | 1.000 | **0.665** | Yes | 0.001 | 0.031 |
| Brun Deviation | 1.402 | 0.985 | 0.000 | No | 0.006 | 0.122 |
| Cumulative Gap Sum | 1.513 | 0.985 | 0.432 | Yes | 155,449 | 7,940 |
| $\pi_2$/H-L Ratio | 1.596 | 0.989 | 0.545 | Yes | 0.002 | 0.078 |
| Spectral $|\text{FFT}|$ | 1.994 | 1.000 | 0.000 | No | 13.2 | 65,860 |
| Gap mod 6 = 0 | 1.999 | 0.542 | 0.000 | No | 136.2 | 48.0 |
#### 4.11.1 Phi-Dominant Waveforms
Six of eleven waveforms exhibit golden ratio dominance:
| Waveform | $\varphi$-corr | Interpretation |
|----------|------------|----------------|
| $\frac{d}{dx}\ln\pi_2$ | 0.665 | Logarithmic rate of change has strongest $\varphi$ signature |
| $d^2\pi_2/dx^2$ | 0.553 | Acceleration exhibits $\varphi$ in its oscillatory structure |
| $\pi_2$/H-L Ratio | 0.545 | Deviation from H-L shows $\varphi$-scaled recurrence |
| Twin-Prime Density | 0.543 | Local density fluctuations follow $\varphi$-spaced peaks |
| $\pi_2$ (detrended) | 0.540 | The counting function itself has $\varphi$ self-similarity |
| H-L Residual | 0.529 | The prediction error has $\varphi$-structured autocorrelation |
The strongest golden ratio signature appears in the log-derivative $\frac{d}{dx}\ln\pi_2$ ($\varphi$-corr = 0.665), which represents the instantaneous relative growth rate of the twin prime counting function. This suggests that the rate at which new twin primes appear exhibits self-similar recurrence patterns at scales related by $\varphi$.
#### 4.11.2 Non-Phi-Dominant Waveforms
Five waveforms show no golden ratio dominance:
- **Gaps, Gap mod 6**: Individual gaps are too noisy for autocorrelation peak detection
- **Brun sums, Brun deviation**: The reciprocal sum is too smooth and slowly-varying
- **Spectral envelope**: The FFT magnitude has its own characteristic structure
#### 4.11.3 Fractal Dimension Spectrum
The sub-waveforms span a range of fractal dimensions:
- **$D_H \approx 2.0$** (Gaps, Acceleration, Spectral): Maximally rough, noise-like
- **$D_H \approx 1.5$** (Residual, Cumulative, Ratio): Intermediate complexity
- **$D_H \approx 1.4$** (Brun, Density): Smoother, more predictable
This spectrum reflects a fundamental hierarchy: derivative-like quantities (gaps, acceleration) are maximally rough, while integral-like quantities (cumulative sums, Brun partial sums) are smooth and persistent.
#### 4.11.4 Hurst Exponent Spectrum
- **$H \approx 1.0$** (Brun, Log-derivative, Spectral, Ratio): Maximal persistence --- these quantities grow monotonically or change extremely slowly
- **$H \approx 0.7$** (Density): Moderate persistence --- local density has memory but fluctuates
- **$H \approx 0.54$** (Gaps): Near-random --- individual gaps are nearly independent
- **$H = 0.166$** (Acceleration): Anti-persistent --- the second derivative rapidly oscillates and mean-reverts
### 4.12 Statistical Validation
| Metric | Value |
|--------|-------|
| OEIS A001359 verification | Pass (50/50) |
| H-L $R^2$ | 0.8779 |
| H-L RMSE | 790.21 |
| H-L mean residual | 715.77 |
| $\varphi$-dominance $p$-value | 0.002 |
| FCE prediction $p$-value | 0.002 |
| Information horizon | Detected at $n = 500$ |
The Hardy-Littlewood formula explains 87.8% of the variance in $\pi_2(x)$, with a positive mean residual (715.8) confirming systematic underestimation at this scale.
An information horizon is detected at prediction horizon $n = 500$, where prediction accuracy degrades sharply. This is consistent with the P vs NP information horizon theory developed in our companion paper: beyond a critical prediction distance, the number-theoretic structure becomes computationally irreducible.
---
## 5. Discussion
### 5.1 The Fractal Geometry of Twin Prime Distribution
The twin prime counting function exhibits a rich multi-scale fractal structure that the smooth Hardy-Littlewood asymptotic cannot capture. The key finding is that the oscillatory residual $\pi_2(x) - \text{HL}(x)$ has intermediate fractal dimension ($D_H = 1.49$) --- it is neither random noise nor a smooth correction, but a structured signal with predictable geometric properties.
This intermediate dimension is qualitatively different from the Goldbach comet ($D_H = 2.0$), reflecting a fundamental asymmetry: the Goldbach function $g(n)$ fluctuates wildly between consecutive even numbers, while $\pi_2(x)$ is a monotonic staircase whose detrended oscillations are smoother and more correlated.
### 5.2 The Zeta-Zero Connection
The most striking result is the strong zeta-zero spectral correlation ($280\times$ background), detected through log-space Fourier analysis. This provides direct empirical support for the theoretical connection between twin prime distribution and the Riemann zeta function, which operates through the explicit formula:
$$\psi_2(x) = \sum_{\substack{p \leq x \\ p, p+2 \text{ prime}}} \log p \approx 2C_2 x - 2C_2 \sum_\rho \frac{x^\rho}{\rho^2} + \ldots$$
where $\psi_2$ is the twin prime analogue of the Chebyshev function and the sum ranges over non-trivial zeta zeros $\rho$. The $x^\rho/\rho^2$ terms (rather than $x^\rho/\rho$ as in the prime counting function) reflect the convolution structure of twin primality.
The detection of this signature at $n_{\max} = 10^6$ using only 100 zeros suggests that the zeta-zero influence on twin prime distribution is stronger and more detectable than might be expected from the theoretical $1/|\rho|^2$ suppression factor.
### 5.3 Golden Ratio Self-Similarity
The statistically significant $\varphi$-dominance across 6 of 11 waveforms ($p = 0.002$) is consistent with our Goldbach comet findings (21 of 23 waveforms, $p = 0.002$). The golden ratio appears in the autocorrelation peak spacings of the detrended counting function, the residual, the density, the acceleration, the ratio, and most strongly in the log-derivative.
The $\varphi$-correlation of 0.665 in the log-derivative $\frac{d}{dx}\ln\pi_2$ is the strongest observed across both the twin prime and Goldbach analyses. This waveform represents the instantaneous relative growth rate of the counting function, and its $\varphi$-structured autocorrelation suggests that periods of rapid twin prime appearance recur at scales related by the golden ratio.
Whether this reflects a deep arithmetic property of prime distribution or a more general feature of quasi-periodic processes with power-law statistics remains an open question. The consistency across multiple independent number-theoretic sequences (Goldbach comet, twin prime counting function, and their various derivatives) suggests it is not an artifact of any particular analysis choice.
### 5.4 Constructive and Destructive Interference
The density-based sieve interference mapping reveals a roughly balanced distribution of constructive (1,490) and destructive (1,633) zones, with a slight excess of destructive regions. This is consistent with the sieve-theoretic expectation: the conjunctive requirement that both $p$ and $p+2$ be prime creates more opportunities for deficit than for surplus, because any small prime $q$ that divides either $p$ or $p+2$ eliminates the pair.
The negative mean sieve cross-correlation ($-0.11$) quantifies this destructive tendency: knowing that $p$ is prime provides weak negative information about whether $p+2$ is prime, because the sieve conditions interfere rather than reinforce.
### 5.5 Cross-Paper Consistency
Comparing results across the FCE analysis series:
| Metric | Twin Primes | Goldbach Comet |
|--------|-------------|----------------|
| $D_H$ (detrended) | 1.49 | 2.00 |
| $H$ (raw) | 0.9995 | 0.524 |
| $\varphi$-corr | 0.540 | 0.540 |
| $\varphi$ $p$-value | 0.002 | 0.002 |
| FCE improvement | $1.19\times$ | $4.84\times$ |
| Zeta-zero corr | 280.1 | 0.004 |
Notable consistencies:
- **$\varphi$-correlation**: Identical to 3 decimal places (0.540 vs. 0.540), despite completely different signals
- **$p$-values**: Identical (0.002), reflecting the 500-permutation test resolution
- **OEIS validation**: Both pass perfectly
Notable differences:
- **Fractal dimension**: Twin prime residual ($D_H = 1.49$) is smoother than Goldbach comet ($D_H = 2.0$)
- **Zeta-zero correlation**: Dramatic in twin primes (280$\times$), negligible in Goldbach (0.004$\times$). This reflects the direct dependence of $\pi_2(x)$ on prime distribution (where zeta zeros operate), versus the indirect dependence of $g(n)$ on pair counting (where modular arithmetic dominates)
- **FCE improvement**: Larger for Goldbach ($4.84\times$) due to the strong modular $n \bmod 6$ signal that Hardy-Littlewood misses
### 5.6 Limitations
1. **Finite range**: The analysis covers $x \leq 10^6$. While twin prime pairs are known up to $\sim 10^{18}$, the FCE analysis is limited by memory and computation time. Results may change at larger scales where the H-L asymptotic becomes more accurate.
2. **No proof implications**: Predicting $\pi_2(x)$ does not address whether $\pi_2(x) \to \infty$. Even a perfect predictor cannot constitute a proof.
3. **Zeta-zero correlation interpretation**: The $280\times$ enrichment is measured against median background power. This is a spectral analysis result, not a rigorous verification of the explicit formula. The log-space resampling introduces interpolation artifacts that could inflate or deflate the correlation.
4. **Sampling**: Evaluating $\pi_2(x)$ at 10,000 points (spacing $\Delta x = 100$) aliases the underlying staircase. Finer sampling would capture more of the discrete structure but at significant computational cost.
5. **Golden ratio significance**: While the $p = 0.002$ permutation test is significant, the test uses $n_{\text{perm}} = 500$, which limits resolution. The minimum achievable $p$-value is $1/501 = 0.002$.
---
## 6. Conclusion
I have presented the first comprehensive fractal geometric analysis of the twin prime counting function $\pi_2(x)$ using the Fractal Correction Engine. Our key findings are:
1. **Fractal characterization**: The detrended counting function has Higuchi dimension $D_H = 1.4895$ (intermediate complexity), box-counting dimension $D_{BC} = 1.0165$, and Hurst exponent $H = 0.9995$ (strong persistence). The gap sequence has $D_H = 1.999$ (maximal roughness) and $H = 0.541$ (near-random).
2. **Prediction improvement**: An enhanced FCE ensemble achieves MAE $= 1037.64$, outperforming Hardy-Littlewood (MAE $= 1229.95$) by $1.19\times$ ($p = 0.002$), with Fractal Path (40.8%) and Bias Correction (40.6%) contributing equally.
3. **Zeta-zero spectral signature**: Log-space FFT analysis reveals $280\times$ power concentration at the first 100 Riemann zeta zero frequencies, providing direct empirical evidence of the explicit formula connection to twin prime distribution.
4. **Golden ratio dominance**: Statistically significant $\varphi$-dominance ($\text{corr}_\varphi = 0.540$, $p = 0.002$) in the detrended counting function, with the strongest signal in the log-derivative ($\varphi$-corr = 0.665). Present in 6 of 11 analyzed waveforms.
5. **Sieve interference**: Density-based zone detection identifies 1,490 constructive and 1,633 destructive zones, with negative sieve cross-correlation ($-0.11$) confirming the destructive nature of the conjunctive twin prime constraint.
6. **Brun's constant convergence**: Partial sum $B_2(10^6) = 1.7108$ with convergence rate $n^{-0.114}$, the Brun waveform having fractal dimension $D_H = 1.448$ and maximal persistence ($H = 0.985$).
7. **Cross-paper consistency**: The $\varphi$-correlation (0.540) matches the Goldbach comet result (0.540) to three decimal places, while the zeta-zero spectral detection ($280\times$) dramatically surpasses the Goldbach result (0.004$\times$), reflecting the more direct relationship between $\pi_2(x)$ and the prime counting function.
I emphasize that these results constitute a computational number theory contribution --- characterizing and predicting the internal structure of the twin prime counting function --- and do not constitute or approach a proof of the twin prime conjecture. The fractal geometric perspective reveals structure in $\pi_2(x)$ that is invisible to classical analytic methods and suggests deep connections between the geometry of prime pair distribution, the Riemann zeta function, and self-similar structures governed by the golden ratio.
---
## 7. Data Availability
All results, including the complete analysis pipeline (`twin_prime_fce.py`), metrics (`metrics.json`), computed twin prime data (`twin_prime_data.csv`), and 12 diagnostic visualizations, are available in the accompanying Zenodo deposit.
**Figures included:**
1. Twin Prime Counting Function ($\pi_2(x)$ vs. H-L)
2. Curvature Analysis ($\kappa/\pi$ and radius of curvature)
3. Fractal Dimension (Higuchi and box-counting scaling plots)
4. Gap Sequence (distribution and temporal structure)
5. Brun Convergence (partial sums approaching $B_2$)
6. Sieve Interference (constructive/destructive zone map)
7. Prediction Benchmark (FCE vs. H-L comparison)
8. Spectral Analysis (FFT power spectrum and Lomb-Scargle)
9. Golden Ratio Analysis (autocorrelation peaks and ratios)
10. Residual Waveform (H-L residual FCE analysis)
11. Twin Prime Density (local density vs. H-L expectation)
12. Summary Dashboard (multi-panel overview)
---
## 8. References
1. Brun, V. (1919). La serie $1/5 + 1/7 + 1/11 + 1/13 + 1/17 + 1/19 + 1/29 + 1/31 + 1/41 + 1/43 + 1/59 + 1/61 + \ldots$ ou les denominateurs sont "nombres premiers jumeaux" est convergente ou finie. *Bulletin des Sciences Mathematiques*, 43, 100--104, 124--128.
2. Hardy, G. H., & Littlewood, J. E. (1923). Some problems of 'Partitio Numerorum'; III: On the expression of a number as a sum of primes. *Acta Mathematica*, 44, 1--70.
3. Higuchi, T. (1988). Approach to an irregular time series on the basis of the fractal theory. *Physica D*, 31(2), 277--283.
4. Hurst, H. E. (1951). Long-term storage capacity of reservoirs. *Transactions of the American Society of Civil Engineers*, 116, 770--799.
5. Lomb, N. R. (1976). Least-squares frequency analysis of unequally spaced data. *Astrophysics and Space Science*, 39, 447--462.
6. Maynard, J. (2015). Small gaps between primes. *Annals of Mathematics*, 181(1), 383--413.
7. Montgomery, H. L. (1973). The pair correlation of zeros of the zeta function. *Proceedings of Symposia in Pure Mathematics*, 24, 181--193.
8. Odlyzko, A. M. (2001). The $10^{22}$-nd zero of the Riemann zeta function. In *Dynamical, Spectral, and Arithmetic Zeta Functions*, pp. 139--144.
9. Polymath, D. H. J. (2014). Variants of the Selberg sieve, and bounded intervals containing many primes. *Research in the Mathematical Sciences*, 1(12).
10. Scargle, J. D. (1982). Studies in astronomical time series analysis. II. Statistical aspects of spectral analysis of unevenly spaced data. *The Astrophysical Journal*, 263, 835--853.
11. Zhang, Y. (2014). Bounded gaps between primes. *Annals of Mathematics*, 179(3), 1121--1174.
Files
01_counting_function.png
Files
(4.1 MB)
| Name | Size | |
|---|---|---|
|
md5:9988f105042a4ebc2ef73899b3480eec
|
147.1 kB | Preview Download |
|
md5:379927032ce6d081db1a8d3aa3bde76a
|
353.5 kB | Preview Download |
|
md5:e2db9e26844083d8f57e004b1b3e3e76
|
101.6 kB | Preview Download |
|
md5:f714bca5a6474dd1e232436467a973d9
|
233.7 kB | Preview Download |
|
md5:590d01b9d239114af6c661120757be96
|
89.4 kB | Preview Download |
|
md5:b1b89224d57840b02e2a365ae5f3712f
|
171.5 kB | Preview Download |
|
md5:16ca1430d8fb00b0856a4e5709879af3
|
136.0 kB | Preview Download |
|
md5:01615e47b22d056d39d7034cced89205
|
289.4 kB | Preview Download |
|
md5:c057ddc8a5d3b3cb9d4a6a7162605f67
|
104.9 kB | Preview Download |
|
md5:baf767e02fb0de4f5486792a729f185f
|
120.7 kB | Preview Download |
|
md5:107fe7bd5b5c3d88d0670eac66de54af
|
105.5 kB | Preview Download |
|
md5:598935fabe751e4fa22edb2fcfc1bb5b
|
153.2 kB | Preview Download |
|
md5:101504d27f0c0879e2093a08ae877dfa
|
4.9 kB | Preview Download |
|
md5:33f8d5adb48c46e79293d56b7519f198
|
7.4 kB | Preview Download |
|
md5:ee142a8adad30150ec02bb5f74b07903
|
3.2 kB | Preview Download |
|
md5:79dd9348439ae144d4d7d2b234ad4740
|
3.3 kB | Download |
|
md5:53b58c53d2eafd4094f9535c3da7faec
|
118.1 kB | Preview Download |
|
md5:d06607c055301bba3e60192d961194cf
|
107.9 kB | Download |
|
md5:819d131dfe97c3ed6418edd4e5e64de0
|
1.8 MB | Preview Download |
|
md5:917ecbc89bdafccda38f866da40b0d45
|
48.2 kB | Preview Download |