Abstract
# Fractal Correction Engine Applied to Turbulent Combustion: Pi-Scaled Curvature Analysis, Cascade Snap Point Detection, and Regime-Dependent Flame Diagnostics **Authors:** Adam L McEvoy **Date:** March 2026 **Keywords:** Fractal Correction Engine, turbulent combustion, pi-scaled curvature, cascade snap points, Borghi diagram, level-set flame tracking, thermoacoustic instability, Kolmogorov cascade, fractal dimension --- ## Abstract I present the application of the Fractal Correction Engine (FCE) to turbulent premixed combustion, demonstrating that pi-scaled curvature analysis, cascade snap point detection, and fractal path extraction provide physically meaningful diagnostics across combustion regimes classified by the Borghi-Peters diagram. A fully coupled computational framework was developed integrating a 2D pseudo-spectral Navier-Stokes solver, G-equation level-set flame front tracking, single-step Arrhenius chemistry, multi-species transport, and thermoacoustic instability analysis. The FCE was applied to six diagnostic surfaces derived from the simulation: kinetic energy, velocity probe, vorticity probe, flame area, wrinkling ratio, and mean curvature. Two combustion regimes were investigated: laminar flames ($\text{Ka} = 0.01$, $\text{Da} = 116.57$) and corrugated flamelets ($\text{Ka} = 2.84$, $\text{Da} = 5.55$). Results demonstrate that (i) flame-derived FCE surfaces exhibit higher fractal dimensions ($D = 1.68$--$1.78$) than flow-derived surfaces ($D = 1.22$--$1.39$), revealing the flame front as a fractal amplifier of turbulent structure; (ii) cascade snap points in flame area and wrinkling ratio are temporally coincident, confirming they capture the same flame-turbulence interaction events; (iii) the universal correction law yields 99.3\% corrective behavior in the corrugated flamelet regime; and (iv) thermoacoustic analysis via FCE wave interference mapping produces a negative Rayleigh index ($\text{RI} = -0.427$), confirming acoustic stability. Eighteen of twenty physics validation benchmarks pass across both regimes. These results establish the FCE as a viable diagnostic framework for characterizing multi-scale structure in turbulent reacting flows. --- ## 1. Introduction ### 1.1 Turbulent Combustion as a Multi-Scale Problem Turbulent premixed combustion is among the most challenging problems in computational physics, involving the nonlinear coupling of fluid dynamics, chemical kinetics, heat transfer, and acoustic phenomena across scales spanning several orders of magnitude. The turbulent flame front is a fractal object whose geometric complexity arises from the interaction between the Kolmogorov energy cascade and the reaction-diffusion dynamics of combustion chemistry. The fractal nature of turbulent flame surfaces has been established experimentally by Peters [1], Gouldin [2], and North and Santavicca [3], who measured surface fractal dimensions $D \approx 2.1$--$2.4$ for premixed flames across a range of turbulence intensities. These fractal properties are intimately connected to the turbulent burning velocity enhancement, flame surface area amplification, and the mechanisms of local extinction and reignition. ### 1.2 The Fractal Correction Engine The Fractal Correction Engine (FCE) is a mathematical analysis framework that extracts geometric and topological invariants from arbitrary signals, waveforms, and surfaces using pi-scaled curvature as the fundamental diagnostic quantity. Originally developed for analyzing orbits, waves, and nonlinear dynamical systems, the FCE operates on the principle that any observable trajectory can be decomposed into a fractal path characterized by its curvature spectrum, cascade snap points (moments of abrupt curvature change), and fractal dimension. The central insight of the FCE is that the local curvature of a signal, when scaled by $\pi$, reveals resonance structures and transition points that correspond to physically meaningful events in the underlying dynamical system. For turbulent combustion, these events include flame-vortex interactions, local extinction, reignition, and thermoacoustic coupling. ### 1.3 Objectives This work applies the FCE to turbulent premixed combustion with three objectives: 1. **Demonstrate regime-dependent FCE diagnostics** by comparing FCE analysis across the laminar flames and corrugated flamelets regimes of the Borghi-Peters diagram.2. **Establish the flame front as a fractal amplifier** by showing that flame-derived FCE surfaces exhibit higher fractal dimensions than the underlying flow field.3. **Validate FCE cascade snap point detection** as a physically meaningful diagnostic for flame-turbulence interaction events. --- ## 2. The Fractal Correction Engine: Theory and Methods ### 2.1 Pi-Scaled Curvature Analysis The foundational operation of the FCE is the computation of pi-scaled curvature from a one-dimensional signal $f(t)$. The local curvature is defined as: $$\kappa(t) = \frac{|f''(t)|}{(1 + f'(t)^2)^{3/2}}$$ where $f'(t)$ and $f''(t)$ are the first and second derivatives of the signal, computed via central finite differences on a Savitzky-Golay smoothed version of the input. The pi-scaled curvature is then: $$\kappa_\pi(t) = \frac{\kappa(t)}{\pi}$$ The scaling by $\pi$ normalizes the curvature relative to the curvature of a unit semicircle, establishing a universal reference scale. Points where $\kappa_\pi$ takes integer or half-integer values are identified as **pi-resonance points**, marking moments where the signal's geometric structure aligns with fundamental circular harmonics. The **curvature energy** of the signal is defined as: $$E_\kappa = \int_0^T \kappa(t)^2 \, dt$$ which provides a scalar measure of total geometric complexity. For two-dimensional fields such as the flame front level set $G(\mathbf{x}, t)$, the FCE computes the mean curvature of the iso-surfaces: $$\kappa_{2D} = -\nabla \cdot \left(\frac{\nabla G}{|\nabla G|}\right) = -\frac{G_{xx}G_y^2 - 2G_{xy}G_xG_y + G_{yy}G_x^2}{|\nabla G|^3}$$ This curvature field characterizes the local geometry of the flame front and its pi-scaled version $\kappa_\pi^{2D} = \kappa_{2D}/\pi$ identifies resonance structures in the flame surface geometry. ### 2.2 Fractal Dimension via Box Counting The FCE computes the box-counting fractal dimension of one-dimensional signals and two-dimensional contours. For a 1D signal, the algorithm: 1. Normalizes the signal to the unit interval $[0, 1]$.2. Overlays a grid of boxes with side length $\epsilon = 1/N_{\text{box}}$ for $N_{\text{box}} \in \{4, 8, 16, 32, 64, 128, 256\}$.3. Counts the number of occupied boxes $N(\epsilon)$.4. Fits the fractal dimension from the log-log relationship: $$D = \lim_{\epsilon \to 0} \frac{\log N(\epsilon)}{\log(1/\epsilon)}$$ For 1D signals, $D = 1.0$ indicates a smooth curve, $D \approx 1.5$ a fractal signal, and $D \to 2.0$ a space-filling trajectory. For 2D contours (flame fronts), the box-counting dimension of the contour $D_{2D}$ maps to a 3D surface dimension via $D_{3D} = D_{2D} + 1$, enabling comparison with experimental measurements. ### 2.3 Hurst Exponent via Rescaled Range Analysis The FCE characterizes signal persistence through the Hurst exponent $H$, computed by rescaled range (R/S) analysis: 1. Divide the signal into sub-series of length $n$.2. For each sub-series, compute the cumulative deviation from the mean, the range $R$ of the cumulative deviations, and the standard deviation $S$.3. The ratio $R/S$ scales as $n^H$: $$\frac{R}{S} \sim n^H$$ where:- $H < 0.5$: anti-persistent (mean-reverting) dynamics- $H = 0.5$: uncorrelated random walk- $H > 0.5$: persistent (trending) dynamics The Hurst exponent is related to the fractal dimension by $D = 2 - H$ for self-affine signals, providing a consistency check on the box-counting dimension. ### 2.4 Cascade Snap Point Detection Cascade snap points are moments of abrupt curvature change in the signal trajectory, corresponding to transitions in the underlying dynamical system. The detection algorithm: 1. Compute the pi-scaled curvature $\kappa_\pi(t)$.2. Compute the curvature acceleration: $$\dot{\kappa}(t) = \frac{d\kappa}{dt}$$ 3. Apply a threshold criterion: $$|\dot{\kappa}(t)| > \overline{|\dot{\kappa}|} + n_\sigma \cdot \sigma_{|\dot{\kappa}|}$$ where $n_\sigma = 3.0$ (default) is the threshold in standard deviations. 4. Within each contiguous above-threshold region, identify the peak of $|\dot{\kappa}|$ as the snap point location. 5. Classify severity by $\kappa_\pi$ magnitude at the snap point: | Classification | $\kappa_\pi$ range ||---|---|| Catastrophic | $> 100$ || Major | $10$--$100$ || Moderate | $1$--$10$ || Minor | $0.1$--$1$ || Oscillatory | $< 0.1$ | For turbulent combustion, catastrophic snap points correspond to extinction or reignition events, major snap points to large-scale flame-vortex interactions, and oscillatory snap points to the background turbulent wrinkling of the flame surface. ### 2.5 Forward and Backward Prediction The FCE predicts future and past signal evolution using delay-coordinate embedding combined with nearest-neighbor projection. Given a signal $f(t)$, the delay embedding constructs vectors in an $m$-dimensional phase space: $$\mathbf{x}(t) = [f(t), f(t - \tau), f(t - 2\tau), \ldots, f(t - (m-1)\tau)]$$ where $m = 10$ is the embedding dimension and $\tau = 5$ is the time delay (default values). Forward prediction proceeds by: 1. Constructing the delay embedding of the observed signal.2. Building a k-d tree for nearest-neighbor lookup.3. For each prediction step, finding $k = 5$ nearest neighbors of the current embedding vector.4. Computing inverse-distance-weighted average of successor values: $$\hat{f}(t + \Delta t) = \frac{\sum_i w_i f_{\text{successor}}(i)}{\sum_i w_i}, \quad w_i = \frac{1}{d_i + \epsilon}$$ where $d_i$ is the Euclidean distance to the $i$-th neighbor in embedding space. Backward prediction applies the same algo