Seismic Wave Tomography via the Fractal Correction Engine: Pi-Weighted Curvature Correction for Ray Path Fidelity in Layered Earth Models

Read full text Summarise with VIRAT
Authors: Adam L McEvoy
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

# An Event-Driven 2D Layered-Earth Seismic Ray Tracer Benchmarked Against Exact $\tau(p)$ Travel Times: Checkerboard-Verified Synthetic Tomography, Energy-Tested Interface Physics, and a Rigorous Negative Evaluation of the Fractal Correction Engine **Author:** Adam L McEvoy **Date:** 2026-07-12 **Version:** 2.0 --- ## Abstract I present a reproducible 2D layered-Earth seismic simulation system with exact mathematical ground truth, and use it to deliver both a validated forward-modeling instrument and a statistically defensible negative result. For a spherically symmetric Earth model with piecewise-linear velocity $v(r)$, travel times follow from the classical $\tau(p)$ integrals, which I evaluate to near machine precision ($\sim 10^{-8}$ s quadrature self-check). Against this reference, an event-driven ray tracer — adaptive Runge–Kutta integration with terminal event functions at layer boundaries and an analytic change of variable through turning points — achieves a mean absolute travel-time error of $1.28\times10^{-5}$ s across a 50-ray fan spanning direct P to inner-core phases, at roughly half the computational cost of the fixed-step integrator it replaces (whose error is 16–42 s). Against these baselines I evaluate the Fractal Correction Engine (FCE), a $\pi$-weighted curvature-correction scheme applied at layer-boundary crossings. The previously reported $5.1\times$ improvement in ray-parameter conservation is withdrawn: it decomposes exactly into a step-size confound (matched-step FCE effect $1.00\times$; pure step-size effect $5.10\times$), and the conservation metric itself is tautological because the ray parameter is a fixed parameter of the integrated system. With correct numerics the FCE degrades accuracy by five orders of magnitude, and a paired-seed comparison of its modulation constants (20 seeds, 95% confidence intervals, error measured against the exact reference) inverts the $\pi$-optimality claim: $\pi$ is nominally the worst constant tested. The FCE's own box-counting diagnostic reports fractal dimension $D = 1.000$ for every ray path — correctly finding no fractal structure to correct — and I show that the framework's closed-loop residual methodology, repurposed as a detect-only monitor coasting on the exact baseline, works precisely as specified: silent on healthy rays ($1.4\times10^{-3}$ s max residual), detecting velocity anomalies as class-1 signals, and flagging the corrector's own state jumps as structural defects. The system further includes energy-tested solid–liquid interface coefficients (energy closure to $6.7\times10^{-16}$) enabling SKS mode conversion and true PcP/ScS/PKiKP reflections whose absolute times land within seconds of real-Earth observations; PREM-like anelastic attenuation with canonical operators ($t^*_P = 1.09$ s, $t^*_S = 5.46$ s); and a synthetic travel-time tomography pipeline validated by checkerboard resolution tests, L-curve regularization, and noise-robustness curves (multi-source recovery $r = 0.93$; the inverse crime is broken). I report the negative FCE result with the same prominence the positive claim once had, because the experiments that produced it — matched controls, exact references, and pre-registered statistics — are the substance of the upgrade. **Keywords:** seismic ray tracing, travel-time tomography, $\tau(p)$ method, event detection, checkerboard resolution test, negative result, numerical methods, fractal correction engine --- ## 1. Introduction ### 1.1 What this system is This work provides a self-contained, tested synthetic seismology sandbox consisting of: 1. a six-layer spherically symmetric Earth model with two embedded velocity anomalies, for which exact travel times are computable to machine precision (Section 2);2. two numerical ray tracers — a fixed-step RK4 integrator with step-lagged event handling, and an event-driven replacement — with absolute accuracy quantified against the exact reference rather than against another simulation (Section 3);3. the Fractal Correction Engine, a domain-agnostic path-correction and path-monitoring framework, mapped onto this system and evaluated as a hypothesis rather than assumed as a feature (Section 4);4. energy-tested interface physics (mode conversion, true reflections) and anelastic attenuation (Section 5);5. a linearized travel-time tomography pipeline with the validation apparatus the field expects: checkerboard tests, L-curve damping selection, noise injection, and multi-source coverage (Section 6);6. a complete experimental record (Section 7) and a test suite of 22 automated checks that enforce consistency between the code, the result files, and this document. ### 1.2 What this paper is not Earlier versions of this work claimed a $5.1\times$ improvement from the FCE, $\pi$-optimality of its modulation constant, and framed the result as one link in a chain of cross-domain validations. All three claims are withdrawn in this version. The decisive experiments — matched-step controls, an exact external reference, and paired-seed statistics — are exactly the ones the original study skipped, and running them inverts its conclusions. I consider the resulting negative result more useful than the claims it replaces, and this paper leads with the numbers that survive. --- ## 2. Physical Model and Exact Reference ### 2.1 Earth model The model is PREM-like with six layers (crust, upper mantle, transition zone, lower mantle, liquid outer core with $v_S = 0$, solid inner core). Velocities are linear in depth within each layer and interpolated on a 2000-point radial grid, making $v(r)$ globally continuous: first-order discontinuities are smeared into ramps of one grid interval ($\approx 3.2$ km). Density $\rho(r)$ and PREM-like quality factors $Q_P(r)$, $Q_S(r)$ are carried on the same grid. Two Gaussian velocity anomalies provide tomography targets: a slow plume ($\delta v/v = -3\%$, radius 300 km) at 1500 km depth and a fast slab ($+3\%$, radius 250 km) at 800 km depth, $$v(r,\theta) = v_0(r)\left[1 + \sum_k \frac{\delta v_k}{v}\, \exp\!\left(-\frac{d_k(r,\theta)^2}{2\sigma_k^2}\right)\right],$$ where $d_k$ is the distance to the $k$-th anomaly center in $(r, r\theta)$ coordinates. ### 2.2 Ray theory A seismic ray in a spherically symmetric medium conserves the ray parameter (Bouguer's/Snell's law in spherical geometry): $$p = \frac{r \sin i}{v(r)} = \text{const},$$ where $i$ is the angle between the ray and the local radial direction. The kinematic ray equations, integrated with respect to arc length $s$, are $$\frac{dr}{ds} = \pm\cos i, \qquad\frac{d\theta}{ds} = \frac{\sin i}{r} = \frac{p\,v(r)}{r^2}, \qquad\frac{dT}{ds} = \frac{1}{v(r)},$$ with $\sin i = p\,v(r)/r$. The local ray curvature is $\kappa = \frac{1}{v}\frac{dv}{dr}$ — velocity gradients bend rays toward slower material. A ray turns where $i = \pi/2$, i.e. where $$\eta(r) \equiv \frac{r}{v(r)} = p .$$ ### 2.3 Exact $\tau(p)$ travel times For the spherically symmetric model, angular distance and travel time follow from the classical integrals (the $\tau(p)$ formulation of Buland & Chapman): $$\Delta(p) = \int_{\text{legs}} \frac{p}{r\sqrt{\eta^2(r) - p^2}}\, dr,\qquadT(p) = \int_{\text{legs}} \frac{\eta^2(r)}{r\sqrt{\eta^2(r) - p^2}}\, dr,$$ taken over the down-leg from the source radius $r_{\mathrm{src}}$ to the turning radius $r_t$ and the up-leg from $r_t$ to the surface $R$. Because $v(r)$ is piecewise linear, $v = a + br$ within each grid interval, the turning radius solves $\eta(r) = p$ in closed form: $$r_t = \frac{p\,a}{1 - p\,b}.$$ The integrable square-root singularity at $r_t$ is removed by the substitution $x = \sqrt{r - r_t}$, under which the integrand becomes analytic within each grid panel; per-panel Gauss–Legendre quadrature then converges spectrally. Two internal checks anchor this reference: - **Quadrature self-check:** order-24 vs order-48 Gauss–Legendre agree to $\max|\delta T| = 1.5\times10^{-8}$ s.- **Earth-likeness:** the near-vertical limit ($p = 0.05$ s/rad) gives $\Delta = 179.98°$, $T = 1215.1$ s, against published antipodal PKIKP times of $\approx 1212$ s for ak135/PREM — the simplified model is Earth-like at the $\sim 0.3\%$ level on this benchmark. Every accuracy claim in this paper is an absolute error in seconds against this reference. No claim rests on comparing two approximate methods to each other. --- ## 3. Numerical Ray Tracing ### 3.1 The legacy fixed-step tracer and its measured errors The original integrator advances $[r, \theta, T]$ by fixed-step RK4. Crucially, $p$ is a **fixed parameter** of the right-hand side, not a dynamical variable — a fact that matters for Section 4.3. Events are handled one step late: layer crossings by a sign change between steps, turning points by a $\sin i \geq 0.999$ trigger evaluated after a full step, and surface arrival truncated at $r \geq R - 1$ km. Measured against the exact reference (50-ray fan, anomalies off), this tracer carries a mean absolute travel-time error of 16–42 s depending on configuration, with one wandering ray ($p = 658.6$ s/rad) accumulating a 1263 s error. The dominant contribution is the early-turn trigger, which costs roughly $3°$ of angular distance per deep ray: expanding $\eta^2 - p^2 \approx 2p|\eta'|(r - r_t)$ near the turn, the missing near-turn contribution to $\Delta$ scales as $\sqrt{\delta r}$ and is nearly step-size independent — which explains the initially puzzling observation that the error slightly *inverts* with step size (coarser steps overshoot the trigger deeper, accidentally closer to the true turning point). ### 3.2 The event-driven tracer The replacement integrates each smooth leg with adaptive RK45 and locates every event with terminal event functions: - **layer boundaries** (all five) — the Poincaré sections of the system, located to solver precision;- **surface arrival** at exactly $r = R$ (no truncation);- **turning-point approach** at $\cos i = 0.05$, from which the r

Fields of Study & Subjects

Seismic Imaging and Inversion Techniques Seismic Waves and Analysis High-pressure geophysics and materials

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
25 March 2026
Language
EN