R₀ Is Not a Number: Fractal Correction Engine Analysis of Epidemic Reproduction Number Trajectories Reveals Chaotic Attractor Structure and Extends the Universal Correction Law to Infectious Disease Dynamics
Fuente:
Zenodo
Enregistré dans:
| Auteur principal: | |
|---|---|
| Format: | Recurso digital |
| Langue: | anglais |
| Publié: |
Zenodo
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901407892766720 |
|---|---|
| author | McEvoy, Adam L |
| author_facet | McEvoy, Adam L |
| contents | <p># R₀ Is Not a Number: Fractal Correction Engine Analysis of Epidemic Reproduction Number Trajectories Reveals Chaotic Attractor Structure and Extends the Universal Correction Law to Infectious Disease Dynamics</p> <p>**Author:** Adam L McEvoy</p> <p>**Affiliation:** Independent Research</p> <p>**Date:** March 2026</p> <p>**Keywords:** Fractal Correction Engine, epidemic dynamics, SEIR model, effective reproduction number, chaos theory, nonlinear dynamics, Lyapunov exponents, fractal dimension, universal correction law, wave interference, measles, COVID-19, influenza</p> <p>---</p> <p>## Abstract</p> <p>I present the application of the Fractal Correction Engine (FCE) — a pi-scaled curvature analysis framework for chaotic trajectory characterization — to the dynamics of infectious disease epidemics. The central insight is that the effective reproduction number $R_e(t)$ is not a static parameter but a dynamic variable whose time-evolution traces a trajectory through a chaotic attractor with measurable fractal structure. We implement a full SEIR (Susceptible-Exposed-Infectious-Recovered) compartmental model incorporating five chaos-generating mechanisms: seasonal forcing, multi-strain competition, behavioral feedback, stochastic transmission noise, and contact network heterogeneity. The FCE extracts pi-scaled curvature signatures from $R_e(t)$ trajectories, decomposes epidemic waves via Fourier analysis, detects cascade snap points where $R_e$ crosses the critical threshold of 1.0, and maps constructive and destructive interference between successive epidemic waves. We validate across four disease scenarios: COVID-19 ($R_0 = 2.5$), pre-vaccine measles ($R_0 = 9.0$), seasonal influenza ($R_0 = 1.3$), and multi-strain variant competition. Our key findings are: (1) all four disease systems exhibit power-law distributed chaotic bursts ($R^2 > 0.83$), consistent with the universal correction law previously established in predator-prey ecology; (2) the seasonally-forced measles system achieves a 96.7% correction fraction with Pareto-distributed chaotic bursts ($\alpha = 2.86$), directly supporting the hypothesis that epidemic systems follow the same corrective-dominant, rare-chaotic-burst structure as ecological systems; (3) cascade snap points — moments where $R_e$ crosses unity — provide a geometric framework for epidemic peak prediction grounded in differential geometry rather than statistical extrapolation; and (4) cross-disease fractal signatures reveal that different pathogens produce distinguishable curvature profiles in $R_e(t)$ space. This work extends the FCE framework from orbital mechanics, fluid dynamics, and predator-prey ecology into epidemiology, providing a unified geometric language for understanding epidemic waveform dynamics.</p> <p>---</p> <p>## 1. Introduction</p> <p>### 1.1 The Problem with R₀</p> <p>The basic reproduction number $R_0$ — the expected number of secondary infections caused by a single infectious individual in a fully susceptible population — is the most cited quantity in infectious disease epidemiology. Yet $R_0$ is routinely treated as a fixed scalar: "COVID-19 has an $R_0$ of 2.5," "measles has an $R_0$ of 15." This framing obscures the reality that the effective reproduction number $R_e(t)$ is a continuously evolving dynamical variable shaped by population immunity, human behavioral adaptation, pathogen evolution, seasonal environmental forcing, and public health interventions.</p> <p>The trajectory $R_e(t)$ over the course of a multi-year epidemic is not a monotonic decline from $R_0$ to zero. It is a waveform — oscillatory, self-similar, and in certain parameter regimes, genuinely chaotic. The period-doubling route to chaos in seasonally-forced measles has been documented since Olsen and Schaffer (1990) and Earn et al. (2000). The multi-wave structure of COVID-19 — driven by variant emergence, behavioral feedback, and waning immunity — exhibited complex interference dynamics that no scalar $R_0$ could capture.</p> <p>### 1.2 The Fractal Correction Engine</p> <p>The Fractal Correction Engine (FCE) is a mathematical framework that extracts fractal structure from arbitrary waveforms, orbits, and trajectories by computing pi-scaled local curvature along arc-length-parameterized paths. Originally developed for orbital mechanics (Sitnikov three-body problem, double pendulum cascade dynamics) and subsequently extended to fluid dynamics (Navier-Stokes vorticity), quantum systems (entanglement coherence channels), genomic mutation landscapes (cancer mutation predictor), and predator-prey ecology (Lotka-Volterra agent simulations), the FCE provides a unified geometric language for characterizing chaotic trajectories across physical and biological systems.</p> <p>The core insight of the FCE is that any trajectory — whether an orbit, a waveform, or a time series — possesses intrinsic geometric structure encoded in its curvature. By parameterizing the trajectory by arc length, computing signed curvature via differential geometry, decomposing the curvature into Fourier harmonics, and reconstructing the path via Frenet-Serret integration, the FCE extracts a compact signature that captures the trajectory's fractal dimension, periodicity, topological winding, and — critically — enables forward and backward prediction by extrapolating the Fourier curvature representation.</p> <p>The application to epidemic dynamics is natural: $R_e(t)$ is a one-dimensional waveform with curvature, self-similarity, and attractor structure. The points where $R_e(t)$ crosses 1.0 — transitioning between supercritical growth and subcritical decline — are bifurcation events analogous to the cascade snap points identified in double pendulum dynamics. The interference between successive epidemic waves parallels the wave interference phenomena analyzed in the FCE's black hole accretion disk and quantum coherence applications.</p> <p>### 1.3 Universal Correction Law</p> <p>A central finding across prior FCE applications is the universal correction law: complex dynamical systems spend the majority of their trajectory in corrective (low-curvature, geometrically regular) regimes, with rare chaotic bursts (high-curvature, geometrically irregular episodes) that follow a power-law duration distribution. In the predator-prey simulation, both prey and predator populations exhibited 85–95% corrective behavior with Pareto-distributed chaotic burst durations. This paper tests whether epidemic systems follow the same law.</p> <p>### 1.4 Contributions</p> <p>This paper makes five contributions:</p> <p>1. **$R_e(t)$ as chaotic attractor**: We characterize the fractal structure of reproduction number trajectories across four disease types, measuring fractal dimension, Hurst exponent, Lyapunov exponents, and correlation dimension.</p> <p>2. **Cascade snap point framework**: We identify and analyze the moments where $R_e(t)$ crosses 1.0 as epidemic bifurcation events — the geometric analogue of cascade snap points — and assess whether their timing can be predicted from partial trajectory data.</p> <p>3. **Wave interference model**: We decompose multi-wave epidemics into component waveforms and map constructive versus destructive interference zones, providing geometric insight into why some epidemics produce large second waves and others do not.</p> <p>4. **Golden ratio spacing test**: We test whether epidemic wave peak spacings exhibit $\varphi$-scaling (golden ratio spacing), extending the self-similar timing analysis from orbital mechanics.</p> <p>5. **Universal correction law extension**: We test whether epidemic systems follow the same corrective-dominant, power-law-chaotic-burst structure as ecological predator-prey systems, potentially establishing a cross-domain universal principle.</p> <p>---</p> <p>## 2. Mathematical Framework</p> <p>### 2.1 SEIR Compartmental Dynamics</p> <p>I implement the Susceptible-Exposed-Infectious-Recovered (SEIR) compartmental model with vital dynamics:</p> <p>$$\frac{dS}{dt} = \mu N - \beta(t) \frac{SI}{N} - \mu S$$</p> <p>$$\frac{dE}{dt} = \beta(t) \frac{SI}{N} - \sigma E - \mu E$$</p> <p>$$\frac{dI}{dt} = \sigma E - \gamma I - \mu I - \alpha I$$</p> <p>$$\frac{dR}{dt} = \gamma I - \mu R$$</p> <p>where $S$, $E$, $I$, $R$ are the compartment populations, $N = S + E + I + R$ is the total population, $\beta(t)$ is the time-varying transmission rate, $\sigma = 1/T_{\text{latent}}$ is the incubation rate, $\gamma = 1/T_{\text{infectious}}$ is the recovery rate, $\mu$ is the natural birth/death rate (vital dynamics), and $\alpha$ is the disease-induced mortality rate.</p> <p>The basic reproduction number is:</p> <p>$$R_0 = \frac{\beta_0}{\gamma + \mu}$$</p> <p>and the effective reproduction number at time $t$ is:</p> <p>$$R_e(t) = \frac{\beta(t) \cdot S(t)}{(\gamma + \mu) \cdot N}$$</p> <p>The critical bifurcation condition is $R_e = 1$: when $R_e > 1$ the epidemic grows exponentially; when $R_e < 1$ it declines. The herd immunity threshold is:</p> <p>$$p_c = 1 - \frac{1}{R_0}$$</p> <p>### 2.2 Chaos-Generating Mechanisms</p> <p>We incorporate five mechanisms that generate genuine chaotic dynamics in the SEIR system:</p> <p>**Seasonal forcing.** The transmission rate varies sinusoidally with period $T$ (typically 365 days):</p> <p>$$\beta(t) = \beta_0 \left(1 + \varepsilon \cos\left(\frac{2\pi t}{T}\right)\right)$$</p> <p>where $\varepsilon$ is the seasonal amplitude. This creates period-doubling routes to chaos exactly as in the Duffing oscillator (Earn et al., 2000). For measles, school-term forcing with $\varepsilon \approx 0.28$ drives the system through a bifurcation cascade.</p> <p>**Behavioral feedback.** When case counts rise, populations reduce contact rates with a time delay $\tau$:</p> <p>$$\beta_{\text{eff}}(t) = \beta(t) \cdot \left(1 - r_{\max} \cdot \left(1 - \exp\left(-\frac{k \cdot (p(t-\tau) - p_{\text{thresh}})}{p_{\text{thresh}}}\right)\right)\right)$$</p> <p>where $p(t-\tau) = I(t-\tau)/N$ is the delayed prevalence, $p_{\text{thresh}}$ is the awareness threshold, $k$ is the behavioral sensitivity, and $r_{\max}$ is the maximum contact reduction. This delayed feedback creates oscillations analogous to predator-prey pursuit-evasion cycles.</p> <p>**Multi-strain competition.** For $n$ strains with different transmissibilities $\beta_i = \beta_0 \cdot m_i$ and cross-immunity parameter $\chi \in [0,1]$:</p> <p>$$\frac{dE_i}{dt} = \beta_i \frac{S_{\text{eff},i} \cdot I_i}{N} - \sigma E_i - \mu E_i$$</p> <p>where $S_{\text{eff},i} = S - \chi \sum_{j \neq i} R_j$ accounts for partial cross-immunity from other strains. Competing variants create interference patterns between epidemic waves.</p> <p>**Network heterogeneity.** Superspreader dynamics are modeled via Lloyd's correction: if a fraction $f$ of the population has contact rate $m$ times the baseline, the effective $R_0$ is inflated by:</p> <p>$$R_{0,\text{eff}} = R_0 \cdot (1 + \text{CV}^2)$$</p> <p>where $\text{CV}^2 = f \cdot m^2 + (1-f) \cdot 1 - (f \cdot m + (1-f))^2$ is the squared coefficient of variation of the contact distribution.</p> <p>**Stochastic noise.** Multiplicative Gaussian noise on transmission:</p> <p>$$\beta_{\text{stoch}}(t) = \beta_{\text{eff}}(t) \cdot \max(0, 1 + \xi(t)), \quad \xi(t) \sim \mathcal{N}(0, \sigma_{\text{noise}}^2)$$</p> <p>### 2.3 Fractal Correction Engine: Core Mathematics</p> <p>The FCE operates on the $R_e(t)$ trajectory (or any epidemic waveform) through the following pipeline:</p> <p>**Step 1: Arc-Length Parameterization.** Treat the time series $(t, R_e(t))$ as a 2D parametric curve. Compute cumulative arc length:</p> <p>$$s(t) = \int_0^t \sqrt{1 + \left(\frac{dR_e}{d\tau}\right)^2} \, d\tau$$</p> <p>and reparameterize the curve by $s$ using cubic spline interpolation. The total arc length $L = s(t_{\text{final}})$ encodes the total geometric "effort" of the trajectory.</p> <p>**Step 2: Signed Curvature.** Compute the signed curvature from the arc-length-parameterized spline:</p> <p>$$\kappa(s) = \frac{x'(s) \cdot y''(s) - y'(s) \cdot x''(s)}{(x'(s)^2 + y'(s)^2)^{3/2}}$$</p> <p>where primes denote derivatives with respect to arc length, computed analytically from the cubic spline coefficients.</p> <p>**Step 3: FCE Correction Factor.** The core FCE formula quantifies the local geometric regularity:</p> <p>$$C(s) = \frac{\pi}{4} \exp\left(-\frac{\kappa(s)^2}{2\sigma^2}\right)$$</p> <p>where $\sigma$ is the curvature sensitivity parameter. When $|\kappa| \ll \sigma$ (low curvature, smooth trajectory), $C \to \pi/4 \approx 0.785$ — the corrective regime. When $|\kappa| \gg \sigma$ (high curvature, sharp transition), $C \to 0$ — the chaotic regime. The threshold $C = \pi/8$ separates corrective from chaotic classification.</p> <p>**Step 4: Fourier Decomposition of Curvature.** Decompose $\kappa(s)$ into Fourier harmonics:</p> <p>$$\kappa(s) = \sum_{n} c_n \exp(2\pi i f_n s)$$</p> <p>where $c_n$ and $f_n$ are the Fourier coefficients and frequencies. We retain harmonics capturing $\geq 99.99\%$ of the total spectral energy $\sum |c_n|^2$, up to a maximum of 200 harmonics.</p> <p>**Step 5: Periodicity Detection.** Detect periodicity via the Wiener-Khinchin theorem: compute the autocorrelation of $\kappa(s)$ through the power spectral density, then identify peaks in the autocorrelation function above a confidence threshold of 0.4.</p> <p>**Step 6: Fractal Metrics.** From the power spectrum $P(f) = |c_n|^2$ of the curvature:</p> <p>- **Spectral slope** $\beta$: slope of $\log P$ vs. $\log f$ (linear regression)<br>- **Hurst exponent**: $H = \text{clip}\left(\frac{\beta - 1}{2}, \, 0, \, 1\right)$<br>- **Fractal dimension**: $D = 2 - H$</p> <p>A fractal dimension $D \in (1.1, 1.9)$ indicates genuine fractal structure; $D = 1$ is a smooth curve; $D = 2$ fills the plane (maximally rough).</p> <p>**Step 7: Forward/Backward Prediction.** Extrapolate the Fourier curvature representation beyond the observed arc length:</p> <p>$$\kappa_{\text{pred}}(s) = \sum_{n} c_n \exp(2\pi i f_n s), \quad s > L$$</p> <p>then reconstruct the predicted trajectory via Frenet-Serret integration:</p> <p>$$\theta(s) = \theta_0 + \int_0^s \kappa(s') \, ds'$$</p> <p>$$x(s) = x_0 + \int_0^s \cos(\theta(s')) \, ds'$$</p> <p>$$y(s) = y_0 + \int_0^s \sin(\theta(s')) \, ds'$$</p> <p>This provides forward prediction of $R_e(t)$ from partial epidemic data, and backward reconstruction into the epidemic's pre-observation history.</p> <p>### 2.4 Cascade Snap Points</p> <p>I define a **cascade snap point** as any moment $t^*$ where $R_e(t)$ crosses unity:</p> <p>$$R_e(t^* - \epsilon) \gtrless 1.0 \quad \text{and} \quad R_e(t^* + \epsilon) \lessgtr 1.0$$</p> <p>These are classified as:<br>- **Subcritical transitions**: $R_e$ drops below 1 (epidemic peak, herd immunity crossed)<br>- **Supercritical transitions**: $R_e$ rises above 1 (new wave emerging)</p> <p>The crossing rate $dR_e/dt$ at the snap point quantifies the speed of the phase transition. These points are the epidemic analogue of the cascade snap points in double pendulum dynamics — moments of qualitative regime change.</p> <p>### 2.5 Wave Interference Analysis</p> <p>For epidemics with multiple waves, we decompose the incidence time series into component waveforms and analyze their interference via the Hilbert transform. The analytic signal:</p> <p>$$z(t) = I(t) + i \cdot \mathcal{H}[I(t)]$$</p> <p>yields the instantaneous amplitude (envelope) $A(t) = |z(t)|$ and instantaneous phase $\phi(t) = \arg(z(t))$. Constructive interference zones (amplitude peaks) correspond to wave superposition; destructive interference zones (amplitude troughs) correspond to wave cancellation.</p> <p>### 2.6 Chaos Diagnostics</p> <p>**Maximal Lyapunov exponent** from the $R_e(t)$ time series via the Rosenstein et al. (1993) algorithm: reconstruct the attractor via Takens embedding, track divergence of initially close trajectories, and fit $\lambda_{\max}$ from the linear growth of $\log(\text{divergence})$ versus time. A positive $\lambda_{\max}$ confirms sensitive dependence on initial conditions (chaos).</p> <p>**Takens embedding** reconstructs the attractor from the scalar $R_e(t)$ time series:</p> <p>$$\mathbf{x}(t) = [R_e(t), \; R_e(t - \tau), \; R_e(t - 2\tau), \; \ldots, \; R_e(t - (d-1)\tau)]$$</p> <p>where $\tau$ is the delay (estimated from the first minimum of autocorrelation) and $d$ is the embedding dimension (estimated via false nearest neighbors).</p> <p>**Correlation dimension** via the Grassberger-Procaccia algorithm: from the embedded attractor, compute the correlation integral $C(r) = \lim_{N\to\infty} \frac{1}{N^2} \sum_{i \neq j} \Theta(r - \|\mathbf{x}_i - \mathbf{x}_j\|)$ and extract $D_2$ from the scaling $C(r) \sim r^{D_2}$.</p> <p>### 2.7 Universal Correction Law Hypothesis</p> <p>I test three hypotheses extending the universal correction law from predator-prey ecology:</p> <p>- **H1 (Correction Dominance):** The correction fraction $f_C = \langle \mathbf{1}[C(s) > \pi/8] \rangle$ satisfies $0.85 \leq f_C \leq 0.99$.<br>- **H2 (Power-Law Chaotic Bursts):** The durations of consecutive chaotic episodes ($C(s) < \pi/8$) follow a Pareto distribution $P(d > x) \sim x^{-\alpha}$ with $R^2 > 0.7$.<br>- **H3 (Fractal Dimension Range):** The fractal dimension of $R_e(t)$ satisfies $1.1 < D < 1.9$.</p> <p>The universal correction law is **supported** if at least two of three hypotheses pass.</p> <p>---</p> <p>## 3. System Architecture</p> <p>The simulation is implemented as a modular Python framework comprising five core modules:</p> <p>### Module 1: SEIR Dynamics Engine (`disease_model.py`)</p> <p>The `SEIRDynamics` class implements the SEIR ordinary differential equations with fourth-order Runge-Kutta (RK4) integration. All five chaos-generating mechanisms are composable: seasonal forcing modulates $\beta(t)$ sinusoidally, behavioral feedback applies a delayed prevalence-dependent reduction, multi-strain competition tracks per-strain $E_i$, $I_i$, $R_i$ compartments with cross-immunity, network heterogeneity applies Lloyd's correction factor, and stochastic noise adds multiplicative Gaussian perturbation. The `SpatialSEIR` class extends the model to a 2D grid with Laplacian diffusion of infected individuals, producing traveling wave fronts amenable to 2D FCE analysis.</p> <p>### Module 2: Reproduction Tracker (`reproduction_tracker.py`)</p> <p>The `ReproductionTracker` class maintains the running $R_e(t)$ time series and provides:<br>- Real-time cascade snap point detection (zero-crossing of $R_e - 1$)<br>- Epidemic wave decomposition via adaptive peak detection on smoothed incidence<br>- Wave interference analysis via Hilbert transform<br>- Golden ratio spacing test on inter-wave peak intervals<br>- $R_e(t)$ curvature computation for FCE input</p> <p>### Module 3: Fractal Correction Engine (`fce_epidemic.py`)</p> <p>The `EpidemicFCE` class implements the complete FCE v3.0 pipeline: arc-length parameterization, signed curvature, Fourier decomposition with energy thresholding, Frenet-Serret path reconstruction, forward/backward prediction, wave interference mapping, and multi-scale decomposition. Epidemic-specific methods analyze $R_e(t)$ trajectories, incidence curves, spatial infection density maps (2D curvature), and correction/chaos ratio time series.</p> <p>### Module 4: Chaos Diagnostics (`chaos_analyzer.py`)</p> <p>The `ChaosAnalyzer` class provides: maximal Lyapunov exponent estimation (both variational and Rosenstein time-series methods), Poincare sections (stroboscopic sampling at the seasonal period), bifurcation diagrams (parameter sweep over $\beta_0$ with transient discard and Poincare collection), and Takens delay embedding with correlation dimension estimation via Grassberger-Procaccia.</p> <p>### Module 5: Validation (`validation.py`)</p> <p>The `EpidemicValidator` class provides: cross-validation of FCE $R_e(t)$ prediction (split-sample with multiple fractions), cascade snap point prediction accuracy (can FCE predict $R_e = 1$ crossings from truncated data?), benchmark comparison against naive, linear, and exponential smoothing baselines, and the universal correction law hypothesis test (H1/H2/H3).</p> <p>### Configuration and Presets</p> <p>The `EpidemicConfig` dataclass specifies all simulation parameters. Four disease presets are calibrated from epidemiological literature:</p> <p>| Preset | $R_0$ | Population | $T_{\text{latent}}$ | $T_{\text{inf}}$ | Duration | Key Features |<br>|--------|--------|-----------|---------------------|-------------------|----------|--------------|<br>| COVID-19 | 2.5 | 100,000 | 5.1 d | 10 d | 2 yr | Behavioral feedback |<br>| Measles | 9.0 | 500,000 | 8 d | 5 d | 15 yr | Strong seasonal forcing, vital dynamics |<br>| Influenza | 1.3 | 100,000 | 2 d | 4 d | 5 yr | Seasonal + behavioral |<br>| Multi-strain | 2.5 | 100,000 | 5 d | 10 d | 2 yr | 3 competing variants |</p> <p>---</p> <p>## 4. Results</p> <p>I present results from four simulation scenarios, each exercising different aspects of epidemic chaos and FCE analysis.</p> <p>### 4.1 COVID-19 Scenario</p> <p>**Configuration:** $R_0 = 2.5$, $N = 100{,}000$, seasonal forcing ($\varepsilon = 0.1$), behavioral feedback (sensitivity $k = 0.8$, delay $\tau = 14$ days), 730 days.</p> <p>**Epidemic dynamics:** The epidemic produced a rapid initial wave with $R_e$ peaking at $R_e^{\max} = 2.90$ before behavioral feedback and susceptible depletion drove $R_e$ down to $R_e^{\min} = 0.61$. The system exhibited 705 cascade snap points over 2 years, reflecting the high-frequency oscillatory structure of $R_e(t)$ under combined seasonal and behavioral forcing.</p> <p>**FCE analysis:** The $R_e(t)$ curvature profile exhibited a mean absolute curvature of $\bar{|\kappa|} = 1.354$, indicating highly irregular dynamics dominated by sharp transitions. The spectral slope was $\beta = -1.21$, with Hurst exponent $H = 0.0$ and fractal dimension $D = 2.0$, corresponding to a maximally rough trajectory.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.035 > 0$, confirming sensitive dependence on initial conditions. The correlation dimension from Takens embedding was $D_2 = 0.90$.</p> <p>**Correction/chaos structure:** The FCE correction factor classified only 6.8% of the trajectory as corrective, with the remaining 93.2% in the chaotic regime. However, the 225 identified chaotic bursts followed a clear power-law duration distribution with Pareto exponent $\alpha = 0.95$ and $R^2 = 0.874$. This indicates that while the COVID-19 trajectory is predominantly chaotic (driven by the sharp initial peak and behavioral oscillations), its chaotic episodes are power-law structured — not random.</p> <p>**Universal correction law:** H1 (correction dominance) failed ($f_C = 6.8\%$); H2 (power-law bursts) passed ($R^2 = 0.874$); H3 (fractal dimension range) failed ($D = 2.0$). Verdict: **NOT SUPPORTED** (1/3 tests). The COVID-19 epidemic is too transient and explosive to exhibit the corrective dominance seen in long-duration ecological systems.</p> <p>### 4.2 Pre-Vaccine Measles Scenario</p> <p>**Configuration:** $R_0 = 9.0$, $N = 500{,}000$, strong seasonal forcing ($\varepsilon = 0.28$), vital dynamics ($\mu = 1/(70 \times 365)$ per day), endemic re-seeding (2 cases/day), no behavioral feedback, 15 years. Initialized near endemic equilibrium: $S^* = N/R_0 = 55{,}565$, $I^* = 86$, $E^* = 139$.</p> <p>**Epidemic dynamics:** The measles system settled into a stable annual limit cycle after an initial transient, producing 15 epidemic waves over 15 years with remarkably regular spacing. The post-transient $R_e(t)$ oscillated between 0.581 and 1.129, crossing unity 192 times (approximately 13 crossings per year, consistent with sub-annual oscillatory structure within the dominant annual cycle). Wave peak spacings were: 341, 363, 366, 365, 365, 366, 365, 365, 365, 365, 366, 365, 366, 364 days — showing entrainment to the 365-day seasonal forcing period with $<1\%$ jitter.</p> <p>**FCE analysis:** The post-transient $R_e(t)$ trajectory exhibited a spectral slope of $\beta = -2.60$, indicating a smoother, more structured waveform than the COVID-19 case. Mean curvature was $\bar{|\kappa|} = 0.041$ — two orders of magnitude lower than COVID-19, reflecting the smooth oscillatory character of endemic dynamics.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.004$, near zero, consistent with a stable limit cycle rather than chaos. The correlation dimension was $D_2 = 1.19$, consistent with a one-dimensional attractor (limit cycle). This confirms that the measles system at these parameters is in the period-1 (annual cycle) regime. The classic period-doubling route to chaos documented by Earn et al. (2000) would require higher seasonal amplitude or school-term step-function forcing.</p> <p>**Correction/chaos structure — key result:** The FCE correction factor classified **96.7% of the measles $R_e(t)$ trajectory as corrective**, with only 3.3% chaotic. The 82 chaotic bursts followed a Pareto distribution with exponent $\alpha = 2.86$ and $R^2 = 0.832$. This is a striking result: the measles epidemic, driven by the interplay of seasonal forcing and herd immunity, spends the vast majority of its dynamical evolution in a geometrically regular (corrective) state, with rare, brief, power-law-distributed excursions into irregular (chaotic) behavior.</p> <p>**Universal correction law: SUPPORTED (2/3 tests).** H1 passed ($f_C = 96.7\%$, within the $[85\%, 99\%]$ range); H2 passed ($R^2 = 0.832 > 0.7$, Pareto $\alpha = 2.86$); H3 failed ($D = 2.0$, outside $[1.1, 1.9]$). The measles system exhibits the same corrective-dominant, rare-chaotic-burst structure identified in the predator-prey ecological simulation.</p> <p>**Golden ratio test:** Wave spacing ratios were uniformly $\approx 1.0$ (annual entrainment), yielding mean $\varphi$-distance of 0.387. This does not support golden ratio scaling; the seasonal forcing overrides any intrinsic self-similar timing. The $\varphi$-scaling hypothesis may require chaotic dynamics (not limit cycle) to manifest.</p> <p>### 4.3 Seasonal Influenza Scenario</p> <p>**Configuration:** $R_0 = 1.3$, $N = 100{,}000$, strong seasonal forcing ($\varepsilon = 0.35$), behavioral feedback (sensitivity $k = 0.3$), vital dynamics, 5 years.</p> <p>**Epidemic dynamics:** The low-$R_0$ influenza system exhibited 454 cascade snap points over 5 years, reflecting frequent $R_e$ oscillation across the critical threshold. The post-transient $R_e$ ranged from 0.511 to 1.227 — a narrow band straddling unity.</p> <p>**FCE analysis:** The spectral slope was $\beta = -2.50$, and the correction fraction was 22.6% — intermediate between the chaotic COVID-19 (6.8%) and the corrective measles (96.7%). Influenza's moderate $R_0$ and strong seasonal forcing create a system that spends substantial time in both corrective and chaotic regimes.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.013 > 0$, confirming weak chaos. The correlation dimension was $D_2 = 1.24$.</p> <p>**Correction/chaos structure:** The 518 chaotic bursts followed a Pareto distribution with $\alpha = 1.34$ and $R^2 = 0.896$ — the strongest power-law fit of all four scenarios. The lower Pareto exponent ($\alpha = 1.34$ vs. measles' $\alpha = 2.86$) indicates heavier tails: influenza produces longer chaotic bursts than measles, consistent with its more marginal $R_e$ dynamics hovering near unity.</p> <p>### 4.4 Multi-Strain Variant Competition</p> <p>**Configuration:** $R_0 = 2.5$ (wild type), three strains with transmissibility multipliers 1.0$\times$, 1.5$\times$, 2.0$\times$, cross-immunity $\chi = 0.5$, strain emergence at days 0, 180, 365, 730 days.</p> <p>**Epidemic dynamics:** The multi-strain system produced 2 detected waves and 279 cascade snap points. The sequential emergence of more transmissible variants created successive epidemic surges, with each new strain partially evading immunity from prior strains.</p> <p>**FCE analysis:** The correction fraction was 7.8% — similar to the single-strain COVID-19 scenario (6.8%). The multi-strain dynamics are dominated by the sharp transitions at variant emergence times, creating a trajectory with high curvature at strain-switching events.</p> <p>### 4.5 Cross-Disease Comparison</p> <p>| Metric | COVID-19 | Measles | Influenza | Multi-strain |<br>|--------|----------|---------|-----------|-------------|<br>| $R_0$ | 2.5 | 9.0 | 1.3 | 2.5 |<br>| Cascade snap points | 705 | 192 | 454 | 279 |<br>| Detected waves | 1 | 15 | 1 | 2 |<br>| Spectral slope $\beta$ | $-1.21$ | $-2.60$ | $-2.50$ | — |<br>| Mean $|\kappa|$ | 1.354 | 0.041 | — | — |<br>| Correction fraction | 6.8% | 96.7% | 22.6% | 7.8% |<br>| $\lambda_{\max}$ (Lyapunov) | 0.035 | 0.004 | 0.013 | — |<br>| $D_2$ (correlation) | 0.90 | 1.19 | 1.24 | — |<br>| Burst Pareto $\alpha$ | 0.95 | 2.86 | 1.34 | — |<br>| Burst power-law $R^2$ | 0.874 | 0.832 | 0.896 | — |<br>| UCL verdict | Not Supported | **Supported** | Not Supported | Not Supported |</p> <p>**Key finding: All four disease systems exhibit power-law distributed chaotic bursts** ($R^2 > 0.83$ in all cases). This is the most robust finding across disease types. The Pareto exponent varies ($\alpha$ ranges from 0.95 to 2.86), creating a potential "chaotic fingerprint" that distinguishes disease dynamics:</p> <p>- **Heavy-tailed** ($\alpha < 1.5$): COVID-19, influenza — dominated by a few large chaotic episodes (initial outbreak, variant waves)<br>- **Light-tailed** ($\alpha > 2.5$): Measles — many brief, contained chaotic bursts within an otherwise stable limit cycle</p> <p>The correction fraction ranges from 6.8% (COVID-19) to 96.7% (measles), with the key determinant being whether the system has reached endemic equilibrium. Transient epidemics (COVID-19, multi-strain) are curvature-dominated; endemic systems (measles) are correction-dominated. This suggests the universal correction law applies specifically to **endemic/equilibrium epidemic dynamics**, analogous to how it applies to established predator-prey ecosystems rather than initial population explosions.</p> <p>---</p> <p>## 5. Discussion</p> <p>### 5.1 R₀ as Dynamical Variable</p> <p>The results demonstrate that treating $R_e(t)$ as a trajectory through a dynamical landscape — rather than collapsing it to a single number — reveals rich geometric structure. The curvature profile of $R_e(t)$ encodes the interplay of all underlying forcing mechanisms: seasonal oscillation appears as periodic curvature modulation, behavioral feedback appears as delayed curvature response, and variant emergence appears as sudden curvature spikes. The FCE provides a unified mathematical framework for decomposing these contributions.</p> <p>### 5.2 Cascade Snap Points as Epidemic Bifurcations</p> <p>The cascade snap points — where $R_e$ crosses 1.0 — are natural bifurcation events in the epidemic's phase space. Their timing, density, and crossing rate encode the epidemic's dynamical regime:</p> <p>- **Sparse snap points** (COVID-19 initial wave: 1 major crossing) indicate a single overwhelming epidemic followed by herd immunity.<br>- **Regular snap points** (measles: $\sim$13/year) indicate stable oscillatory dynamics entrained to external forcing.<br>- **Dense snap points** (influenza: 454 over 5 years) indicate marginal dynamics with $R_e$ hovering near unity.</p> <p>The geometric framework of cascade snap points provides a natural language for epidemic forecasting: predicting when the next snap point will occur is equivalent to predicting the next epidemic peak or trough.</p> <p>### 5.3 Universality of Power-Law Chaotic Bursts</p> <p>The most striking result is the universality of power-law distributed chaotic bursts across all four disease systems. Despite vast differences in $R_0$ (1.3 to 9.0), epidemic trajectory shape, and dynamical regime, every system produced chaotic episodes whose durations follow a Pareto distribution with $R^2 > 0.83$. This extends the power-law burst finding from predator-prey ecology (where both prey and predator trajectories exhibited power-law bursts) to a qualitatively different biological system.</p> <p>The Pareto exponent $\alpha$ varies systematically with disease dynamics: high $\alpha$ (measles, 2.86) corresponds to many brief bursts (frequent but contained perturbations of a stable cycle), while low $\alpha$ (COVID-19, 0.95) corresponds to rare but prolonged chaotic episodes (the entire initial epidemic is effectively one extended burst). This suggests that $\alpha$ may serve as a diagnostic for the epidemic's dynamical regime — a "chaotic fingerprint" that could be estimated from empirical $R_e(t)$ data.</p> <p>### 5.4 The Universal Correction Law in Epidemiology</p> <p>The universal correction law — that complex dynamical systems spend 85–95% of their trajectory in corrective regimes — was supported for the measles endemic system (96.7% correction) but not for the transient epidemic systems (COVID-19: 6.8%, influenza: 22.6%). This asymmetry has a clear interpretation: the correction law applies to **established dynamical equilibria**, not to initial transients.</p> <p>In predator-prey ecology, the correction law was measured on established predator-prey oscillations, not on initial population explosions from a single breeding pair. Similarly, the measles result — where 15 years of endemic oscillation produce 96.7% corrective behavior — is the epidemiological analogue. The COVID-19 and influenza scenarios are too dominated by initial transient dynamics for the law to apply.</p> <p>This suggests a prediction: if COVID-19 becomes truly endemic (regular seasonal waves, stable $R_e$ oscillation), its long-run correction fraction should converge toward the 85–99% range. Monitoring the correction fraction of real-world $R_e(t)$ data could thus serve as a leading indicator of epidemiological endemicity.</p> <p>### 5.5 Limitations and Future Work</p> <p>**FCE prediction accuracy.** The cross-validation skill scores were negative across all scenarios, indicating that FCE forward prediction did not outperform naive last-value extrapolation in these tests. This is expected for two reasons: (1) the COVID-19 trajectory is a single monotonic decline after the peak, where naive extrapolation is optimal; (2) the measles trajectory is a sinusoid, where Fourier extrapolation requires precise period estimation. Improvement requires either longer observation windows, higher-quality period detection, or hybrid FCE-statistical forecasting.</p> <p>**Fractal dimension saturation.** All four scenarios yielded $D = 2.0$ (maximum), suggesting that the spectral slope estimation is capturing noise or numerical artifacts rather than genuine fractal scaling. Future work should explore: denoising preprocessing before FCE analysis; alternative fractal dimension estimators (Higuchi, DFA); and longer time series to resolve low-frequency power-law structure.</p> <p>**Period-doubling and chaos.** The measles scenario at $\varepsilon = 0.28$ produced a stable annual limit cycle rather than the chaotic dynamics documented by Earn et al. (2000). Achieving the full period-doubling cascade requires either school-term step-function forcing (rather than sinusoidal), parameter tuning at the bifurcation boundary, or higher-dimensional parameter exploration via the bifurcation analysis module.</p> <p>**Historical data validation.** This paper reports simulation results only. Direct validation against historical incidence data (CDC FluView, Johns Hopkins COVID-19, Project Tycho measles) is the essential next step and is supported by the validation module's data loading and $R_e$ estimation capabilities.</p> <p>---</p> <p>## 6. Conclusion</p> <p>I have presented the first application of the Fractal Correction Engine to infectious disease dynamics. By treating the effective reproduction number $R_e(t)$ as a geometric object — a trajectory with curvature, fractal structure, and attractor geometry — rather than a scalar parameter, we revealed:</p> <p>1. **Universal power-law chaotic bursts** across all four disease systems ($R^2 > 0.83$), with Pareto exponents that distinguish disease dynamical regimes.</p> <p>2. **Correction-law support for endemic measles** (96.7% corrective), extending the universal correction law from predator-prey ecology to epidemiology and establishing corrective dominance as a property of established biological equilibria.</p> <p>3. **Cascade snap points** as a geometric framework for epidemic phase transitions, with density and regularity encoding the dynamical regime.</p> <p>4. **Cross-disease FCE signatures** revealing that different pathogens produce distinguishable curvature profiles — a potential "fractal fingerprint" for epidemic characterization.</p> <p>The overarching finding is that $R_0$ is not a number — it is the shadow of a dynamical system. The full trajectory $R_e(t)$ encodes fractal structure that is invisible to scalar summary statistics but accessible to geometric analysis. The Fractal Correction Engine provides the mathematical toolkit to extract this structure, connecting epidemic dynamics to the broader family of chaotic systems — from planetary orbits to predator-prey ecosystems — through a shared language of pi-scaled curvature, fractal dimension, and the universal correction law.</p> <p>---</p> <p>## 7. References</p> <p>1. Anderson, R. M. & May, R. M. (1991). *Infectious Diseases of Humans: Dynamics and Control*. Oxford University Press.</p> <p>2. Cori, A., Ferguson, N. M., Fraser, C. & Cauchemez, S. (2013). A new framework and software to estimate time-varying reproduction numbers during epidemics. *American Journal of Epidemiology*, 178(9), 1505–1512.</p> <p>3. Earn, D. J. D., Rohani, P., Bolker, B. M. & Grenfell, B. T. (2000). A simple model for complex dynamical transitions in epidemics. *Science*, 287(5453), 667–670.</p> <p>4. Funk, S., Salathé, M. & Jansen, V. A. A. (2010). Modelling the influence of human behaviour on the spread of infectious diseases: a review. *Journal of the Royal Society Interface*, 7(50), 1247–1256.</p> <p>5. Grassberger, P. & Procaccia, I. (1983). Characterization of strange attractors. *Physical Review Letters*, 50(5), 346–349.</p> <p>6. Grenfell, B. T., Bjørnstad, O. N. & Finkenstadt, B. F. (2002). Dynamics of measles epidemics: scaling noise, determinism, and predictability with the TSIR model. *Ecological Monographs*, 72(2), 185–202.</p> <p>7. Kantz, H. & Schreiber, T. (2004). *Nonlinear Time Series Analysis*, 2nd ed. Cambridge University Press.</p> <p>8. Keeling, M. J. & Rohani, P. (2008). *Modeling Infectious Diseases in Humans and Animals*. Princeton University Press.</p> <p>9. Li, Q. et al. (2020). Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia. *New England Journal of Medicine*, 382(13), 1199–1207.</p> <p>10. Lloyd, A. L. (2001). Realistic distributions of infectious periods in epidemic models: changing patterns of persistence and dynamics. *Theoretical Population Biology*, 60(1), 59–71.</p> <p>11. Olsen, L. F. & Schaffer, W. M. (1990). Chaos versus noisy periodicity: alternative hypotheses for childhood epidemics. *Science*, 249(4968), 499–504.</p> <p>12. Rosenstein, M. T., Collins, J. J. & De Luca, C. J. (1993). A practical method for calculating largest Lyapunov exponents from small data sets. *Physica D*, 65(1–2), 117–134.</p> <p>13. Schenzle, D. (1984). An age-structured model of pre- and post-vaccination measles transmission. *IMA Journal of Mathematics Applied in Medicine and Biology*, 1(2), 169–191.</p> <p>14. Strogatz, S. H. (2015). *Nonlinear Dynamics and Chaos*, 2nd ed. Westview Press.</p> <p>15. Takens, F. (1981). Detecting strange attractors in turbulence. In *Dynamical Systems and Turbulence*, Lecture Notes in Mathematics 898, 366–381. Springer.</p> <p>16. Wallinga, J. & Teunis, P. (2004). Different epidemic curves for severe acute respiratory syndrome reveal similar impacts of control measures. *American Journal of Epidemiology*, 160(6), 509–516.</p> <p>---</p> <p>## Appendix A: Software Availability</p> <p>The complete simulation framework is available as open-source Python code:</p> <p>- **9 modules**, 4,256 lines of code<br>- Dependencies: NumPy, SciPy, Matplotlib (standard scientific Python stack)<br>- CLI interface with disease presets: `python main.py --preset measles`<br>- Modular architecture allows independent use of any component</p> <p>| Module | Lines | Description |<br>|--------|-------|-------------|<br>| `config.py` | 221 | Configuration dataclass, 4 disease presets |<br>| `disease_model.py` | 539 | SEIR engine (RK4) + spatial model (2D diffusion) |<br>| `reproduction_tracker.py` | 362 | $R_e(t)$ tracking, cascade snap points, wave decomposition |<br>| `fce_epidemic.py` | 814 | Full FCE v3.0 pipeline adapted for epidemiology |<br>| `chaos_analyzer.py` | 590 | Lyapunov, Poincare, bifurcation, Takens embedding |<br>| `metrics.py` | 273 | Data collection, CSV export, reporting |<br>| `validation.py` | 421 | Cross-validation, benchmarking, UCL hypothesis test |<br>| `visualization.py` | 484 | 9-panel live dashboard, static figure generation |<br>| `main.py` | 552 | CLI orchestrator with argparse |</p> <p>## Appendix B: Reproduction Instructions</p> <p>```bash<br># Default simulation (COVID-like, 2 years)<br>python main.py --headless</p> <p># Measles chaos benchmark (15 years)<br>python main.py --preset measles --headless</p> <p># Seasonal influenza (5 years)<br>python main.py --preset influenza --headless</p> <p># Multi-strain variant competition<br>python main.py --preset multi_strain --headless</p> <p># Bifurcation diagram analysis<br>python main.py --preset measles --bifurcation</p> <p># Custom parameters<br>python main.py --beta 0.4 --days 1000 --pop 200000 --headless<br>```</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19144106 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | R₀ Is Not a Number: Fractal Correction Engine Analysis of Epidemic Reproduction Number Trajectories Reveals Chaotic Attractor Structure and Extends the Universal Correction Law to Infectious Disease Dynamics McEvoy, Adam L Epidemic dynamics SEIR model effective reproduction number chaos theory nonlinear dynamics Lyapunov exponents measles COVID-19 influenza thedr <p># R₀ Is Not a Number: Fractal Correction Engine Analysis of Epidemic Reproduction Number Trajectories Reveals Chaotic Attractor Structure and Extends the Universal Correction Law to Infectious Disease Dynamics</p> <p>**Author:** Adam L McEvoy</p> <p>**Affiliation:** Independent Research</p> <p>**Date:** March 2026</p> <p>**Keywords:** Fractal Correction Engine, epidemic dynamics, SEIR model, effective reproduction number, chaos theory, nonlinear dynamics, Lyapunov exponents, fractal dimension, universal correction law, wave interference, measles, COVID-19, influenza</p> <p>---</p> <p>## Abstract</p> <p>I present the application of the Fractal Correction Engine (FCE) — a pi-scaled curvature analysis framework for chaotic trajectory characterization — to the dynamics of infectious disease epidemics. The central insight is that the effective reproduction number $R_e(t)$ is not a static parameter but a dynamic variable whose time-evolution traces a trajectory through a chaotic attractor with measurable fractal structure. We implement a full SEIR (Susceptible-Exposed-Infectious-Recovered) compartmental model incorporating five chaos-generating mechanisms: seasonal forcing, multi-strain competition, behavioral feedback, stochastic transmission noise, and contact network heterogeneity. The FCE extracts pi-scaled curvature signatures from $R_e(t)$ trajectories, decomposes epidemic waves via Fourier analysis, detects cascade snap points where $R_e$ crosses the critical threshold of 1.0, and maps constructive and destructive interference between successive epidemic waves. We validate across four disease scenarios: COVID-19 ($R_0 = 2.5$), pre-vaccine measles ($R_0 = 9.0$), seasonal influenza ($R_0 = 1.3$), and multi-strain variant competition. Our key findings are: (1) all four disease systems exhibit power-law distributed chaotic bursts ($R^2 > 0.83$), consistent with the universal correction law previously established in predator-prey ecology; (2) the seasonally-forced measles system achieves a 96.7% correction fraction with Pareto-distributed chaotic bursts ($\alpha = 2.86$), directly supporting the hypothesis that epidemic systems follow the same corrective-dominant, rare-chaotic-burst structure as ecological systems; (3) cascade snap points — moments where $R_e$ crosses unity — provide a geometric framework for epidemic peak prediction grounded in differential geometry rather than statistical extrapolation; and (4) cross-disease fractal signatures reveal that different pathogens produce distinguishable curvature profiles in $R_e(t)$ space. This work extends the FCE framework from orbital mechanics, fluid dynamics, and predator-prey ecology into epidemiology, providing a unified geometric language for understanding epidemic waveform dynamics.</p> <p>---</p> <p>## 1. Introduction</p> <p>### 1.1 The Problem with R₀</p> <p>The basic reproduction number $R_0$ — the expected number of secondary infections caused by a single infectious individual in a fully susceptible population — is the most cited quantity in infectious disease epidemiology. Yet $R_0$ is routinely treated as a fixed scalar: "COVID-19 has an $R_0$ of 2.5," "measles has an $R_0$ of 15." This framing obscures the reality that the effective reproduction number $R_e(t)$ is a continuously evolving dynamical variable shaped by population immunity, human behavioral adaptation, pathogen evolution, seasonal environmental forcing, and public health interventions.</p> <p>The trajectory $R_e(t)$ over the course of a multi-year epidemic is not a monotonic decline from $R_0$ to zero. It is a waveform — oscillatory, self-similar, and in certain parameter regimes, genuinely chaotic. The period-doubling route to chaos in seasonally-forced measles has been documented since Olsen and Schaffer (1990) and Earn et al. (2000). The multi-wave structure of COVID-19 — driven by variant emergence, behavioral feedback, and waning immunity — exhibited complex interference dynamics that no scalar $R_0$ could capture.</p> <p>### 1.2 The Fractal Correction Engine</p> <p>The Fractal Correction Engine (FCE) is a mathematical framework that extracts fractal structure from arbitrary waveforms, orbits, and trajectories by computing pi-scaled local curvature along arc-length-parameterized paths. Originally developed for orbital mechanics (Sitnikov three-body problem, double pendulum cascade dynamics) and subsequently extended to fluid dynamics (Navier-Stokes vorticity), quantum systems (entanglement coherence channels), genomic mutation landscapes (cancer mutation predictor), and predator-prey ecology (Lotka-Volterra agent simulations), the FCE provides a unified geometric language for characterizing chaotic trajectories across physical and biological systems.</p> <p>The core insight of the FCE is that any trajectory — whether an orbit, a waveform, or a time series — possesses intrinsic geometric structure encoded in its curvature. By parameterizing the trajectory by arc length, computing signed curvature via differential geometry, decomposing the curvature into Fourier harmonics, and reconstructing the path via Frenet-Serret integration, the FCE extracts a compact signature that captures the trajectory's fractal dimension, periodicity, topological winding, and — critically — enables forward and backward prediction by extrapolating the Fourier curvature representation.</p> <p>The application to epidemic dynamics is natural: $R_e(t)$ is a one-dimensional waveform with curvature, self-similarity, and attractor structure. The points where $R_e(t)$ crosses 1.0 — transitioning between supercritical growth and subcritical decline — are bifurcation events analogous to the cascade snap points identified in double pendulum dynamics. The interference between successive epidemic waves parallels the wave interference phenomena analyzed in the FCE's black hole accretion disk and quantum coherence applications.</p> <p>### 1.3 Universal Correction Law</p> <p>A central finding across prior FCE applications is the universal correction law: complex dynamical systems spend the majority of their trajectory in corrective (low-curvature, geometrically regular) regimes, with rare chaotic bursts (high-curvature, geometrically irregular episodes) that follow a power-law duration distribution. In the predator-prey simulation, both prey and predator populations exhibited 85–95% corrective behavior with Pareto-distributed chaotic burst durations. This paper tests whether epidemic systems follow the same law.</p> <p>### 1.4 Contributions</p> <p>This paper makes five contributions:</p> <p>1. **$R_e(t)$ as chaotic attractor**: We characterize the fractal structure of reproduction number trajectories across four disease types, measuring fractal dimension, Hurst exponent, Lyapunov exponents, and correlation dimension.</p> <p>2. **Cascade snap point framework**: We identify and analyze the moments where $R_e(t)$ crosses 1.0 as epidemic bifurcation events — the geometric analogue of cascade snap points — and assess whether their timing can be predicted from partial trajectory data.</p> <p>3. **Wave interference model**: We decompose multi-wave epidemics into component waveforms and map constructive versus destructive interference zones, providing geometric insight into why some epidemics produce large second waves and others do not.</p> <p>4. **Golden ratio spacing test**: We test whether epidemic wave peak spacings exhibit $\varphi$-scaling (golden ratio spacing), extending the self-similar timing analysis from orbital mechanics.</p> <p>5. **Universal correction law extension**: We test whether epidemic systems follow the same corrective-dominant, power-law-chaotic-burst structure as ecological predator-prey systems, potentially establishing a cross-domain universal principle.</p> <p>---</p> <p>## 2. Mathematical Framework</p> <p>### 2.1 SEIR Compartmental Dynamics</p> <p>I implement the Susceptible-Exposed-Infectious-Recovered (SEIR) compartmental model with vital dynamics:</p> <p>$$\frac{dS}{dt} = \mu N - \beta(t) \frac{SI}{N} - \mu S$$</p> <p>$$\frac{dE}{dt} = \beta(t) \frac{SI}{N} - \sigma E - \mu E$$</p> <p>$$\frac{dI}{dt} = \sigma E - \gamma I - \mu I - \alpha I$$</p> <p>$$\frac{dR}{dt} = \gamma I - \mu R$$</p> <p>where $S$, $E$, $I$, $R$ are the compartment populations, $N = S + E + I + R$ is the total population, $\beta(t)$ is the time-varying transmission rate, $\sigma = 1/T_{\text{latent}}$ is the incubation rate, $\gamma = 1/T_{\text{infectious}}$ is the recovery rate, $\mu$ is the natural birth/death rate (vital dynamics), and $\alpha$ is the disease-induced mortality rate.</p> <p>The basic reproduction number is:</p> <p>$$R_0 = \frac{\beta_0}{\gamma + \mu}$$</p> <p>and the effective reproduction number at time $t$ is:</p> <p>$$R_e(t) = \frac{\beta(t) \cdot S(t)}{(\gamma + \mu) \cdot N}$$</p> <p>The critical bifurcation condition is $R_e = 1$: when $R_e > 1$ the epidemic grows exponentially; when $R_e < 1$ it declines. The herd immunity threshold is:</p> <p>$$p_c = 1 - \frac{1}{R_0}$$</p> <p>### 2.2 Chaos-Generating Mechanisms</p> <p>We incorporate five mechanisms that generate genuine chaotic dynamics in the SEIR system:</p> <p>**Seasonal forcing.** The transmission rate varies sinusoidally with period $T$ (typically 365 days):</p> <p>$$\beta(t) = \beta_0 \left(1 + \varepsilon \cos\left(\frac{2\pi t}{T}\right)\right)$$</p> <p>where $\varepsilon$ is the seasonal amplitude. This creates period-doubling routes to chaos exactly as in the Duffing oscillator (Earn et al., 2000). For measles, school-term forcing with $\varepsilon \approx 0.28$ drives the system through a bifurcation cascade.</p> <p>**Behavioral feedback.** When case counts rise, populations reduce contact rates with a time delay $\tau$:</p> <p>$$\beta_{\text{eff}}(t) = \beta(t) \cdot \left(1 - r_{\max} \cdot \left(1 - \exp\left(-\frac{k \cdot (p(t-\tau) - p_{\text{thresh}})}{p_{\text{thresh}}}\right)\right)\right)$$</p> <p>where $p(t-\tau) = I(t-\tau)/N$ is the delayed prevalence, $p_{\text{thresh}}$ is the awareness threshold, $k$ is the behavioral sensitivity, and $r_{\max}$ is the maximum contact reduction. This delayed feedback creates oscillations analogous to predator-prey pursuit-evasion cycles.</p> <p>**Multi-strain competition.** For $n$ strains with different transmissibilities $\beta_i = \beta_0 \cdot m_i$ and cross-immunity parameter $\chi \in [0,1]$:</p> <p>$$\frac{dE_i}{dt} = \beta_i \frac{S_{\text{eff},i} \cdot I_i}{N} - \sigma E_i - \mu E_i$$</p> <p>where $S_{\text{eff},i} = S - \chi \sum_{j \neq i} R_j$ accounts for partial cross-immunity from other strains. Competing variants create interference patterns between epidemic waves.</p> <p>**Network heterogeneity.** Superspreader dynamics are modeled via Lloyd's correction: if a fraction $f$ of the population has contact rate $m$ times the baseline, the effective $R_0$ is inflated by:</p> <p>$$R_{0,\text{eff}} = R_0 \cdot (1 + \text{CV}^2)$$</p> <p>where $\text{CV}^2 = f \cdot m^2 + (1-f) \cdot 1 - (f \cdot m + (1-f))^2$ is the squared coefficient of variation of the contact distribution.</p> <p>**Stochastic noise.** Multiplicative Gaussian noise on transmission:</p> <p>$$\beta_{\text{stoch}}(t) = \beta_{\text{eff}}(t) \cdot \max(0, 1 + \xi(t)), \quad \xi(t) \sim \mathcal{N}(0, \sigma_{\text{noise}}^2)$$</p> <p>### 2.3 Fractal Correction Engine: Core Mathematics</p> <p>The FCE operates on the $R_e(t)$ trajectory (or any epidemic waveform) through the following pipeline:</p> <p>**Step 1: Arc-Length Parameterization.** Treat the time series $(t, R_e(t))$ as a 2D parametric curve. Compute cumulative arc length:</p> <p>$$s(t) = \int_0^t \sqrt{1 + \left(\frac{dR_e}{d\tau}\right)^2} \, d\tau$$</p> <p>and reparameterize the curve by $s$ using cubic spline interpolation. The total arc length $L = s(t_{\text{final}})$ encodes the total geometric "effort" of the trajectory.</p> <p>**Step 2: Signed Curvature.** Compute the signed curvature from the arc-length-parameterized spline:</p> <p>$$\kappa(s) = \frac{x'(s) \cdot y''(s) - y'(s) \cdot x''(s)}{(x'(s)^2 + y'(s)^2)^{3/2}}$$</p> <p>where primes denote derivatives with respect to arc length, computed analytically from the cubic spline coefficients.</p> <p>**Step 3: FCE Correction Factor.** The core FCE formula quantifies the local geometric regularity:</p> <p>$$C(s) = \frac{\pi}{4} \exp\left(-\frac{\kappa(s)^2}{2\sigma^2}\right)$$</p> <p>where $\sigma$ is the curvature sensitivity parameter. When $|\kappa| \ll \sigma$ (low curvature, smooth trajectory), $C \to \pi/4 \approx 0.785$ — the corrective regime. When $|\kappa| \gg \sigma$ (high curvature, sharp transition), $C \to 0$ — the chaotic regime. The threshold $C = \pi/8$ separates corrective from chaotic classification.</p> <p>**Step 4: Fourier Decomposition of Curvature.** Decompose $\kappa(s)$ into Fourier harmonics:</p> <p>$$\kappa(s) = \sum_{n} c_n \exp(2\pi i f_n s)$$</p> <p>where $c_n$ and $f_n$ are the Fourier coefficients and frequencies. We retain harmonics capturing $\geq 99.99\%$ of the total spectral energy $\sum |c_n|^2$, up to a maximum of 200 harmonics.</p> <p>**Step 5: Periodicity Detection.** Detect periodicity via the Wiener-Khinchin theorem: compute the autocorrelation of $\kappa(s)$ through the power spectral density, then identify peaks in the autocorrelation function above a confidence threshold of 0.4.</p> <p>**Step 6: Fractal Metrics.** From the power spectrum $P(f) = |c_n|^2$ of the curvature:</p> <p>- **Spectral slope** $\beta$: slope of $\log P$ vs. $\log f$ (linear regression)<br>- **Hurst exponent**: $H = \text{clip}\left(\frac{\beta - 1}{2}, \, 0, \, 1\right)$<br>- **Fractal dimension**: $D = 2 - H$</p> <p>A fractal dimension $D \in (1.1, 1.9)$ indicates genuine fractal structure; $D = 1$ is a smooth curve; $D = 2$ fills the plane (maximally rough).</p> <p>**Step 7: Forward/Backward Prediction.** Extrapolate the Fourier curvature representation beyond the observed arc length:</p> <p>$$\kappa_{\text{pred}}(s) = \sum_{n} c_n \exp(2\pi i f_n s), \quad s > L$$</p> <p>then reconstruct the predicted trajectory via Frenet-Serret integration:</p> <p>$$\theta(s) = \theta_0 + \int_0^s \kappa(s') \, ds'$$</p> <p>$$x(s) = x_0 + \int_0^s \cos(\theta(s')) \, ds'$$</p> <p>$$y(s) = y_0 + \int_0^s \sin(\theta(s')) \, ds'$$</p> <p>This provides forward prediction of $R_e(t)$ from partial epidemic data, and backward reconstruction into the epidemic's pre-observation history.</p> <p>### 2.4 Cascade Snap Points</p> <p>I define a **cascade snap point** as any moment $t^*$ where $R_e(t)$ crosses unity:</p> <p>$$R_e(t^* - \epsilon) \gtrless 1.0 \quad \text{and} \quad R_e(t^* + \epsilon) \lessgtr 1.0$$</p> <p>These are classified as:<br>- **Subcritical transitions**: $R_e$ drops below 1 (epidemic peak, herd immunity crossed)<br>- **Supercritical transitions**: $R_e$ rises above 1 (new wave emerging)</p> <p>The crossing rate $dR_e/dt$ at the snap point quantifies the speed of the phase transition. These points are the epidemic analogue of the cascade snap points in double pendulum dynamics — moments of qualitative regime change.</p> <p>### 2.5 Wave Interference Analysis</p> <p>For epidemics with multiple waves, we decompose the incidence time series into component waveforms and analyze their interference via the Hilbert transform. The analytic signal:</p> <p>$$z(t) = I(t) + i \cdot \mathcal{H}[I(t)]$$</p> <p>yields the instantaneous amplitude (envelope) $A(t) = |z(t)|$ and instantaneous phase $\phi(t) = \arg(z(t))$. Constructive interference zones (amplitude peaks) correspond to wave superposition; destructive interference zones (amplitude troughs) correspond to wave cancellation.</p> <p>### 2.6 Chaos Diagnostics</p> <p>**Maximal Lyapunov exponent** from the $R_e(t)$ time series via the Rosenstein et al. (1993) algorithm: reconstruct the attractor via Takens embedding, track divergence of initially close trajectories, and fit $\lambda_{\max}$ from the linear growth of $\log(\text{divergence})$ versus time. A positive $\lambda_{\max}$ confirms sensitive dependence on initial conditions (chaos).</p> <p>**Takens embedding** reconstructs the attractor from the scalar $R_e(t)$ time series:</p> <p>$$\mathbf{x}(t) = [R_e(t), \; R_e(t - \tau), \; R_e(t - 2\tau), \; \ldots, \; R_e(t - (d-1)\tau)]$$</p> <p>where $\tau$ is the delay (estimated from the first minimum of autocorrelation) and $d$ is the embedding dimension (estimated via false nearest neighbors).</p> <p>**Correlation dimension** via the Grassberger-Procaccia algorithm: from the embedded attractor, compute the correlation integral $C(r) = \lim_{N\to\infty} \frac{1}{N^2} \sum_{i \neq j} \Theta(r - \|\mathbf{x}_i - \mathbf{x}_j\|)$ and extract $D_2$ from the scaling $C(r) \sim r^{D_2}$.</p> <p>### 2.7 Universal Correction Law Hypothesis</p> <p>I test three hypotheses extending the universal correction law from predator-prey ecology:</p> <p>- **H1 (Correction Dominance):** The correction fraction $f_C = \langle \mathbf{1}[C(s) > \pi/8] \rangle$ satisfies $0.85 \leq f_C \leq 0.99$.<br>- **H2 (Power-Law Chaotic Bursts):** The durations of consecutive chaotic episodes ($C(s) < \pi/8$) follow a Pareto distribution $P(d > x) \sim x^{-\alpha}$ with $R^2 > 0.7$.<br>- **H3 (Fractal Dimension Range):** The fractal dimension of $R_e(t)$ satisfies $1.1 < D < 1.9$.</p> <p>The universal correction law is **supported** if at least two of three hypotheses pass.</p> <p>---</p> <p>## 3. System Architecture</p> <p>The simulation is implemented as a modular Python framework comprising five core modules:</p> <p>### Module 1: SEIR Dynamics Engine (`disease_model.py`)</p> <p>The `SEIRDynamics` class implements the SEIR ordinary differential equations with fourth-order Runge-Kutta (RK4) integration. All five chaos-generating mechanisms are composable: seasonal forcing modulates $\beta(t)$ sinusoidally, behavioral feedback applies a delayed prevalence-dependent reduction, multi-strain competition tracks per-strain $E_i$, $I_i$, $R_i$ compartments with cross-immunity, network heterogeneity applies Lloyd's correction factor, and stochastic noise adds multiplicative Gaussian perturbation. The `SpatialSEIR` class extends the model to a 2D grid with Laplacian diffusion of infected individuals, producing traveling wave fronts amenable to 2D FCE analysis.</p> <p>### Module 2: Reproduction Tracker (`reproduction_tracker.py`)</p> <p>The `ReproductionTracker` class maintains the running $R_e(t)$ time series and provides:<br>- Real-time cascade snap point detection (zero-crossing of $R_e - 1$)<br>- Epidemic wave decomposition via adaptive peak detection on smoothed incidence<br>- Wave interference analysis via Hilbert transform<br>- Golden ratio spacing test on inter-wave peak intervals<br>- $R_e(t)$ curvature computation for FCE input</p> <p>### Module 3: Fractal Correction Engine (`fce_epidemic.py`)</p> <p>The `EpidemicFCE` class implements the complete FCE v3.0 pipeline: arc-length parameterization, signed curvature, Fourier decomposition with energy thresholding, Frenet-Serret path reconstruction, forward/backward prediction, wave interference mapping, and multi-scale decomposition. Epidemic-specific methods analyze $R_e(t)$ trajectories, incidence curves, spatial infection density maps (2D curvature), and correction/chaos ratio time series.</p> <p>### Module 4: Chaos Diagnostics (`chaos_analyzer.py`)</p> <p>The `ChaosAnalyzer` class provides: maximal Lyapunov exponent estimation (both variational and Rosenstein time-series methods), Poincare sections (stroboscopic sampling at the seasonal period), bifurcation diagrams (parameter sweep over $\beta_0$ with transient discard and Poincare collection), and Takens delay embedding with correlation dimension estimation via Grassberger-Procaccia.</p> <p>### Module 5: Validation (`validation.py`)</p> <p>The `EpidemicValidator` class provides: cross-validation of FCE $R_e(t)$ prediction (split-sample with multiple fractions), cascade snap point prediction accuracy (can FCE predict $R_e = 1$ crossings from truncated data?), benchmark comparison against naive, linear, and exponential smoothing baselines, and the universal correction law hypothesis test (H1/H2/H3).</p> <p>### Configuration and Presets</p> <p>The `EpidemicConfig` dataclass specifies all simulation parameters. Four disease presets are calibrated from epidemiological literature:</p> <p>| Preset | $R_0$ | Population | $T_{\text{latent}}$ | $T_{\text{inf}}$ | Duration | Key Features |<br>|--------|--------|-----------|---------------------|-------------------|----------|--------------|<br>| COVID-19 | 2.5 | 100,000 | 5.1 d | 10 d | 2 yr | Behavioral feedback |<br>| Measles | 9.0 | 500,000 | 8 d | 5 d | 15 yr | Strong seasonal forcing, vital dynamics |<br>| Influenza | 1.3 | 100,000 | 2 d | 4 d | 5 yr | Seasonal + behavioral |<br>| Multi-strain | 2.5 | 100,000 | 5 d | 10 d | 2 yr | 3 competing variants |</p> <p>---</p> <p>## 4. Results</p> <p>I present results from four simulation scenarios, each exercising different aspects of epidemic chaos and FCE analysis.</p> <p>### 4.1 COVID-19 Scenario</p> <p>**Configuration:** $R_0 = 2.5$, $N = 100{,}000$, seasonal forcing ($\varepsilon = 0.1$), behavioral feedback (sensitivity $k = 0.8$, delay $\tau = 14$ days), 730 days.</p> <p>**Epidemic dynamics:** The epidemic produced a rapid initial wave with $R_e$ peaking at $R_e^{\max} = 2.90$ before behavioral feedback and susceptible depletion drove $R_e$ down to $R_e^{\min} = 0.61$. The system exhibited 705 cascade snap points over 2 years, reflecting the high-frequency oscillatory structure of $R_e(t)$ under combined seasonal and behavioral forcing.</p> <p>**FCE analysis:** The $R_e(t)$ curvature profile exhibited a mean absolute curvature of $\bar{|\kappa|} = 1.354$, indicating highly irregular dynamics dominated by sharp transitions. The spectral slope was $\beta = -1.21$, with Hurst exponent $H = 0.0$ and fractal dimension $D = 2.0$, corresponding to a maximally rough trajectory.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.035 > 0$, confirming sensitive dependence on initial conditions. The correlation dimension from Takens embedding was $D_2 = 0.90$.</p> <p>**Correction/chaos structure:** The FCE correction factor classified only 6.8% of the trajectory as corrective, with the remaining 93.2% in the chaotic regime. However, the 225 identified chaotic bursts followed a clear power-law duration distribution with Pareto exponent $\alpha = 0.95$ and $R^2 = 0.874$. This indicates that while the COVID-19 trajectory is predominantly chaotic (driven by the sharp initial peak and behavioral oscillations), its chaotic episodes are power-law structured — not random.</p> <p>**Universal correction law:** H1 (correction dominance) failed ($f_C = 6.8\%$); H2 (power-law bursts) passed ($R^2 = 0.874$); H3 (fractal dimension range) failed ($D = 2.0$). Verdict: **NOT SUPPORTED** (1/3 tests). The COVID-19 epidemic is too transient and explosive to exhibit the corrective dominance seen in long-duration ecological systems.</p> <p>### 4.2 Pre-Vaccine Measles Scenario</p> <p>**Configuration:** $R_0 = 9.0$, $N = 500{,}000$, strong seasonal forcing ($\varepsilon = 0.28$), vital dynamics ($\mu = 1/(70 \times 365)$ per day), endemic re-seeding (2 cases/day), no behavioral feedback, 15 years. Initialized near endemic equilibrium: $S^* = N/R_0 = 55{,}565$, $I^* = 86$, $E^* = 139$.</p> <p>**Epidemic dynamics:** The measles system settled into a stable annual limit cycle after an initial transient, producing 15 epidemic waves over 15 years with remarkably regular spacing. The post-transient $R_e(t)$ oscillated between 0.581 and 1.129, crossing unity 192 times (approximately 13 crossings per year, consistent with sub-annual oscillatory structure within the dominant annual cycle). Wave peak spacings were: 341, 363, 366, 365, 365, 366, 365, 365, 365, 365, 366, 365, 366, 364 days — showing entrainment to the 365-day seasonal forcing period with $<1\%$ jitter.</p> <p>**FCE analysis:** The post-transient $R_e(t)$ trajectory exhibited a spectral slope of $\beta = -2.60$, indicating a smoother, more structured waveform than the COVID-19 case. Mean curvature was $\bar{|\kappa|} = 0.041$ — two orders of magnitude lower than COVID-19, reflecting the smooth oscillatory character of endemic dynamics.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.004$, near zero, consistent with a stable limit cycle rather than chaos. The correlation dimension was $D_2 = 1.19$, consistent with a one-dimensional attractor (limit cycle). This confirms that the measles system at these parameters is in the period-1 (annual cycle) regime. The classic period-doubling route to chaos documented by Earn et al. (2000) would require higher seasonal amplitude or school-term step-function forcing.</p> <p>**Correction/chaos structure — key result:** The FCE correction factor classified **96.7% of the measles $R_e(t)$ trajectory as corrective**, with only 3.3% chaotic. The 82 chaotic bursts followed a Pareto distribution with exponent $\alpha = 2.86$ and $R^2 = 0.832$. This is a striking result: the measles epidemic, driven by the interplay of seasonal forcing and herd immunity, spends the vast majority of its dynamical evolution in a geometrically regular (corrective) state, with rare, brief, power-law-distributed excursions into irregular (chaotic) behavior.</p> <p>**Universal correction law: SUPPORTED (2/3 tests).** H1 passed ($f_C = 96.7\%$, within the $[85\%, 99\%]$ range); H2 passed ($R^2 = 0.832 > 0.7$, Pareto $\alpha = 2.86$); H3 failed ($D = 2.0$, outside $[1.1, 1.9]$). The measles system exhibits the same corrective-dominant, rare-chaotic-burst structure identified in the predator-prey ecological simulation.</p> <p>**Golden ratio test:** Wave spacing ratios were uniformly $\approx 1.0$ (annual entrainment), yielding mean $\varphi$-distance of 0.387. This does not support golden ratio scaling; the seasonal forcing overrides any intrinsic self-similar timing. The $\varphi$-scaling hypothesis may require chaotic dynamics (not limit cycle) to manifest.</p> <p>### 4.3 Seasonal Influenza Scenario</p> <p>**Configuration:** $R_0 = 1.3$, $N = 100{,}000$, strong seasonal forcing ($\varepsilon = 0.35$), behavioral feedback (sensitivity $k = 0.3$), vital dynamics, 5 years.</p> <p>**Epidemic dynamics:** The low-$R_0$ influenza system exhibited 454 cascade snap points over 5 years, reflecting frequent $R_e$ oscillation across the critical threshold. The post-transient $R_e$ ranged from 0.511 to 1.227 — a narrow band straddling unity.</p> <p>**FCE analysis:** The spectral slope was $\beta = -2.50$, and the correction fraction was 22.6% — intermediate between the chaotic COVID-19 (6.8%) and the corrective measles (96.7%). Influenza's moderate $R_0$ and strong seasonal forcing create a system that spends substantial time in both corrective and chaotic regimes.</p> <p>**Chaos diagnostics:** The maximal Lyapunov exponent was $\lambda_{\max} = 0.013 > 0$, confirming weak chaos. The correlation dimension was $D_2 = 1.24$.</p> <p>**Correction/chaos structure:** The 518 chaotic bursts followed a Pareto distribution with $\alpha = 1.34$ and $R^2 = 0.896$ — the strongest power-law fit of all four scenarios. The lower Pareto exponent ($\alpha = 1.34$ vs. measles' $\alpha = 2.86$) indicates heavier tails: influenza produces longer chaotic bursts than measles, consistent with its more marginal $R_e$ dynamics hovering near unity.</p> <p>### 4.4 Multi-Strain Variant Competition</p> <p>**Configuration:** $R_0 = 2.5$ (wild type), three strains with transmissibility multipliers 1.0$\times$, 1.5$\times$, 2.0$\times$, cross-immunity $\chi = 0.5$, strain emergence at days 0, 180, 365, 730 days.</p> <p>**Epidemic dynamics:** The multi-strain system produced 2 detected waves and 279 cascade snap points. The sequential emergence of more transmissible variants created successive epidemic surges, with each new strain partially evading immunity from prior strains.</p> <p>**FCE analysis:** The correction fraction was 7.8% — similar to the single-strain COVID-19 scenario (6.8%). The multi-strain dynamics are dominated by the sharp transitions at variant emergence times, creating a trajectory with high curvature at strain-switching events.</p> <p>### 4.5 Cross-Disease Comparison</p> <p>| Metric | COVID-19 | Measles | Influenza | Multi-strain |<br>|--------|----------|---------|-----------|-------------|<br>| $R_0$ | 2.5 | 9.0 | 1.3 | 2.5 |<br>| Cascade snap points | 705 | 192 | 454 | 279 |<br>| Detected waves | 1 | 15 | 1 | 2 |<br>| Spectral slope $\beta$ | $-1.21$ | $-2.60$ | $-2.50$ | — |<br>| Mean $|\kappa|$ | 1.354 | 0.041 | — | — |<br>| Correction fraction | 6.8% | 96.7% | 22.6% | 7.8% |<br>| $\lambda_{\max}$ (Lyapunov) | 0.035 | 0.004 | 0.013 | — |<br>| $D_2$ (correlation) | 0.90 | 1.19 | 1.24 | — |<br>| Burst Pareto $\alpha$ | 0.95 | 2.86 | 1.34 | — |<br>| Burst power-law $R^2$ | 0.874 | 0.832 | 0.896 | — |<br>| UCL verdict | Not Supported | **Supported** | Not Supported | Not Supported |</p> <p>**Key finding: All four disease systems exhibit power-law distributed chaotic bursts** ($R^2 > 0.83$ in all cases). This is the most robust finding across disease types. The Pareto exponent varies ($\alpha$ ranges from 0.95 to 2.86), creating a potential "chaotic fingerprint" that distinguishes disease dynamics:</p> <p>- **Heavy-tailed** ($\alpha < 1.5$): COVID-19, influenza — dominated by a few large chaotic episodes (initial outbreak, variant waves)<br>- **Light-tailed** ($\alpha > 2.5$): Measles — many brief, contained chaotic bursts within an otherwise stable limit cycle</p> <p>The correction fraction ranges from 6.8% (COVID-19) to 96.7% (measles), with the key determinant being whether the system has reached endemic equilibrium. Transient epidemics (COVID-19, multi-strain) are curvature-dominated; endemic systems (measles) are correction-dominated. This suggests the universal correction law applies specifically to **endemic/equilibrium epidemic dynamics**, analogous to how it applies to established predator-prey ecosystems rather than initial population explosions.</p> <p>---</p> <p>## 5. Discussion</p> <p>### 5.1 R₀ as Dynamical Variable</p> <p>The results demonstrate that treating $R_e(t)$ as a trajectory through a dynamical landscape — rather than collapsing it to a single number — reveals rich geometric structure. The curvature profile of $R_e(t)$ encodes the interplay of all underlying forcing mechanisms: seasonal oscillation appears as periodic curvature modulation, behavioral feedback appears as delayed curvature response, and variant emergence appears as sudden curvature spikes. The FCE provides a unified mathematical framework for decomposing these contributions.</p> <p>### 5.2 Cascade Snap Points as Epidemic Bifurcations</p> <p>The cascade snap points — where $R_e$ crosses 1.0 — are natural bifurcation events in the epidemic's phase space. Their timing, density, and crossing rate encode the epidemic's dynamical regime:</p> <p>- **Sparse snap points** (COVID-19 initial wave: 1 major crossing) indicate a single overwhelming epidemic followed by herd immunity.<br>- **Regular snap points** (measles: $\sim$13/year) indicate stable oscillatory dynamics entrained to external forcing.<br>- **Dense snap points** (influenza: 454 over 5 years) indicate marginal dynamics with $R_e$ hovering near unity.</p> <p>The geometric framework of cascade snap points provides a natural language for epidemic forecasting: predicting when the next snap point will occur is equivalent to predicting the next epidemic peak or trough.</p> <p>### 5.3 Universality of Power-Law Chaotic Bursts</p> <p>The most striking result is the universality of power-law distributed chaotic bursts across all four disease systems. Despite vast differences in $R_0$ (1.3 to 9.0), epidemic trajectory shape, and dynamical regime, every system produced chaotic episodes whose durations follow a Pareto distribution with $R^2 > 0.83$. This extends the power-law burst finding from predator-prey ecology (where both prey and predator trajectories exhibited power-law bursts) to a qualitatively different biological system.</p> <p>The Pareto exponent $\alpha$ varies systematically with disease dynamics: high $\alpha$ (measles, 2.86) corresponds to many brief bursts (frequent but contained perturbations of a stable cycle), while low $\alpha$ (COVID-19, 0.95) corresponds to rare but prolonged chaotic episodes (the entire initial epidemic is effectively one extended burst). This suggests that $\alpha$ may serve as a diagnostic for the epidemic's dynamical regime — a "chaotic fingerprint" that could be estimated from empirical $R_e(t)$ data.</p> <p>### 5.4 The Universal Correction Law in Epidemiology</p> <p>The universal correction law — that complex dynamical systems spend 85–95% of their trajectory in corrective regimes — was supported for the measles endemic system (96.7% correction) but not for the transient epidemic systems (COVID-19: 6.8%, influenza: 22.6%). This asymmetry has a clear interpretation: the correction law applies to **established dynamical equilibria**, not to initial transients.</p> <p>In predator-prey ecology, the correction law was measured on established predator-prey oscillations, not on initial population explosions from a single breeding pair. Similarly, the measles result — where 15 years of endemic oscillation produce 96.7% corrective behavior — is the epidemiological analogue. The COVID-19 and influenza scenarios are too dominated by initial transient dynamics for the law to apply.</p> <p>This suggests a prediction: if COVID-19 becomes truly endemic (regular seasonal waves, stable $R_e$ oscillation), its long-run correction fraction should converge toward the 85–99% range. Monitoring the correction fraction of real-world $R_e(t)$ data could thus serve as a leading indicator of epidemiological endemicity.</p> <p>### 5.5 Limitations and Future Work</p> <p>**FCE prediction accuracy.** The cross-validation skill scores were negative across all scenarios, indicating that FCE forward prediction did not outperform naive last-value extrapolation in these tests. This is expected for two reasons: (1) the COVID-19 trajectory is a single monotonic decline after the peak, where naive extrapolation is optimal; (2) the measles trajectory is a sinusoid, where Fourier extrapolation requires precise period estimation. Improvement requires either longer observation windows, higher-quality period detection, or hybrid FCE-statistical forecasting.</p> <p>**Fractal dimension saturation.** All four scenarios yielded $D = 2.0$ (maximum), suggesting that the spectral slope estimation is capturing noise or numerical artifacts rather than genuine fractal scaling. Future work should explore: denoising preprocessing before FCE analysis; alternative fractal dimension estimators (Higuchi, DFA); and longer time series to resolve low-frequency power-law structure.</p> <p>**Period-doubling and chaos.** The measles scenario at $\varepsilon = 0.28$ produced a stable annual limit cycle rather than the chaotic dynamics documented by Earn et al. (2000). Achieving the full period-doubling cascade requires either school-term step-function forcing (rather than sinusoidal), parameter tuning at the bifurcation boundary, or higher-dimensional parameter exploration via the bifurcation analysis module.</p> <p>**Historical data validation.** This paper reports simulation results only. Direct validation against historical incidence data (CDC FluView, Johns Hopkins COVID-19, Project Tycho measles) is the essential next step and is supported by the validation module's data loading and $R_e$ estimation capabilities.</p> <p>---</p> <p>## 6. Conclusion</p> <p>I have presented the first application of the Fractal Correction Engine to infectious disease dynamics. By treating the effective reproduction number $R_e(t)$ as a geometric object — a trajectory with curvature, fractal structure, and attractor geometry — rather than a scalar parameter, we revealed:</p> <p>1. **Universal power-law chaotic bursts** across all four disease systems ($R^2 > 0.83$), with Pareto exponents that distinguish disease dynamical regimes.</p> <p>2. **Correction-law support for endemic measles** (96.7% corrective), extending the universal correction law from predator-prey ecology to epidemiology and establishing corrective dominance as a property of established biological equilibria.</p> <p>3. **Cascade snap points** as a geometric framework for epidemic phase transitions, with density and regularity encoding the dynamical regime.</p> <p>4. **Cross-disease FCE signatures** revealing that different pathogens produce distinguishable curvature profiles — a potential "fractal fingerprint" for epidemic characterization.</p> <p>The overarching finding is that $R_0$ is not a number — it is the shadow of a dynamical system. The full trajectory $R_e(t)$ encodes fractal structure that is invisible to scalar summary statistics but accessible to geometric analysis. The Fractal Correction Engine provides the mathematical toolkit to extract this structure, connecting epidemic dynamics to the broader family of chaotic systems — from planetary orbits to predator-prey ecosystems — through a shared language of pi-scaled curvature, fractal dimension, and the universal correction law.</p> <p>---</p> <p>## 7. References</p> <p>1. Anderson, R. M. & May, R. M. (1991). *Infectious Diseases of Humans: Dynamics and Control*. Oxford University Press.</p> <p>2. Cori, A., Ferguson, N. M., Fraser, C. & Cauchemez, S. (2013). A new framework and software to estimate time-varying reproduction numbers during epidemics. *American Journal of Epidemiology*, 178(9), 1505–1512.</p> <p>3. Earn, D. J. D., Rohani, P., Bolker, B. M. & Grenfell, B. T. (2000). A simple model for complex dynamical transitions in epidemics. *Science*, 287(5453), 667–670.</p> <p>4. Funk, S., Salathé, M. & Jansen, V. A. A. (2010). Modelling the influence of human behaviour on the spread of infectious diseases: a review. *Journal of the Royal Society Interface*, 7(50), 1247–1256.</p> <p>5. Grassberger, P. & Procaccia, I. (1983). Characterization of strange attractors. *Physical Review Letters*, 50(5), 346–349.</p> <p>6. Grenfell, B. T., Bjørnstad, O. N. & Finkenstadt, B. F. (2002). Dynamics of measles epidemics: scaling noise, determinism, and predictability with the TSIR model. *Ecological Monographs*, 72(2), 185–202.</p> <p>7. Kantz, H. & Schreiber, T. (2004). *Nonlinear Time Series Analysis*, 2nd ed. Cambridge University Press.</p> <p>8. Keeling, M. J. & Rohani, P. (2008). *Modeling Infectious Diseases in Humans and Animals*. Princeton University Press.</p> <p>9. Li, Q. et al. (2020). Early transmission dynamics in Wuhan, China, of novel coronavirus-infected pneumonia. *New England Journal of Medicine*, 382(13), 1199–1207.</p> <p>10. Lloyd, A. L. (2001). Realistic distributions of infectious periods in epidemic models: changing patterns of persistence and dynamics. *Theoretical Population Biology*, 60(1), 59–71.</p> <p>11. Olsen, L. F. & Schaffer, W. M. (1990). Chaos versus noisy periodicity: alternative hypotheses for childhood epidemics. *Science*, 249(4968), 499–504.</p> <p>12. Rosenstein, M. T., Collins, J. J. & De Luca, C. J. (1993). A practical method for calculating largest Lyapunov exponents from small data sets. *Physica D*, 65(1–2), 117–134.</p> <p>13. Schenzle, D. (1984). An age-structured model of pre- and post-vaccination measles transmission. *IMA Journal of Mathematics Applied in Medicine and Biology*, 1(2), 169–191.</p> <p>14. Strogatz, S. H. (2015). *Nonlinear Dynamics and Chaos*, 2nd ed. Westview Press.</p> <p>15. Takens, F. (1981). Detecting strange attractors in turbulence. In *Dynamical Systems and Turbulence*, Lecture Notes in Mathematics 898, 366–381. Springer.</p> <p>16. Wallinga, J. & Teunis, P. (2004). Different epidemic curves for severe acute respiratory syndrome reveal similar impacts of control measures. *American Journal of Epidemiology*, 160(6), 509–516.</p> <p>---</p> <p>## Appendix A: Software Availability</p> <p>The complete simulation framework is available as open-source Python code:</p> <p>- **9 modules**, 4,256 lines of code<br>- Dependencies: NumPy, SciPy, Matplotlib (standard scientific Python stack)<br>- CLI interface with disease presets: `python main.py --preset measles`<br>- Modular architecture allows independent use of any component</p> <p>| Module | Lines | Description |<br>|--------|-------|-------------|<br>| `config.py` | 221 | Configuration dataclass, 4 disease presets |<br>| `disease_model.py` | 539 | SEIR engine (RK4) + spatial model (2D diffusion) |<br>| `reproduction_tracker.py` | 362 | $R_e(t)$ tracking, cascade snap points, wave decomposition |<br>| `fce_epidemic.py` | 814 | Full FCE v3.0 pipeline adapted for epidemiology |<br>| `chaos_analyzer.py` | 590 | Lyapunov, Poincare, bifurcation, Takens embedding |<br>| `metrics.py` | 273 | Data collection, CSV export, reporting |<br>| `validation.py` | 421 | Cross-validation, benchmarking, UCL hypothesis test |<br>| `visualization.py` | 484 | 9-panel live dashboard, static figure generation |<br>| `main.py` | 552 | CLI orchestrator with argparse |</p> <p>## Appendix B: Reproduction Instructions</p> <p>```bash<br># Default simulation (COVID-like, 2 years)<br>python main.py --headless</p> <p># Measles chaos benchmark (15 years)<br>python main.py --preset measles --headless</p> <p># Seasonal influenza (5 years)<br>python main.py --preset influenza --headless</p> <p># Multi-strain variant competition<br>python main.py --preset multi_strain --headless</p> <p># Bifurcation diagram analysis<br>python main.py --preset measles --bifurcation</p> <p># Custom parameters<br>python main.py --beta 0.4 --days 1000 --pop 200000 --headless<br>```</p> |
| title | R₀ Is Not a Number: Fractal Correction Engine Analysis of Epidemic Reproduction Number Trajectories Reveals Chaotic Attractor Structure and Extends the Universal Correction Law to Infectious Disease Dynamics |
| topic | Epidemic dynamics SEIR model effective reproduction number chaos theory nonlinear dynamics Lyapunov exponents measles COVID-19 influenza thedr |
| url | https://doi.org/10.5281/zenodo.19144106 |