Curvelet-Regularized SPDE Inversion on Piecewise-Planar Fractures with Trace-Graph Coupling
J. J. Segura · 2026
Read Curvelet-Regularized SPDE Inversion on Piecewise-Planar Fractures with Trace-Graph Coupling by J. J. Segura (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-17485.txt.
Introduction
J. J. Segura builds a method to map fractures in rock from sparse measures. Fractured media hide their cracks below ground. Field teams get only sparse point data and traces of cracks in three dimensions. Segura maps those points onto flat fracture planes inferred from trace lines. Each plane gets its own local grid.
A convex objective blends the data with a smooth prior. It adds a sparsity penalty on curvelet terms to keep sharp edges. The result is a sparse-to-dense rebuild of the fracture field. Such maps guide water, oil, and geothermal work. They also help judge where cracks may leak. The method respects the flat, sharp nature of real fractures.
Catalog record
| CREATOR | J. J. Segura |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-17485
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Curvelet-Regularized SPDE Inversion on Piecewise-Planar Fractures with Trace-Graph Coupling J. J. Segura Abstract We formulate a sparse-to-dense reconstruction layer for fractured media in which sparse point measurements are mapped onto piecewise-planar fracture supports inferred from 3D trace polylines. Each plane is discretized in local coordinates and estimated via a convex objective that combines a grid SPDE/GMRF quadratic prior with an ℓ 1 \ell_{1} penalty on undecimated discrete curvelet coefficients, targeting anisotropic, fracture-aligned structure that is poorly represented by isotropic smoothness alone. We further define an along-fracture distance through trace-network geodesics and express connectivity-driven regularization as a quadratic form z ⊤ P ⊤ L G P z z^{\top}P^{\top}L_{G}Pz , where L G L_{G} is a graph Laplacian on the trace network and P P maps plane grids to graph nodes; plane intersections are handled by linear consistency constraints sampled along intersection lines. The resulting optimization admits efficient splitting: sparse linear solves for the quadratic block and coefficient-wise shrinkage for the curvelet block, with standard ADMM convergence under convexity. We specify reproducible synthetic benchmarks, baselines, ablations, and sensitivity studies that isolate directional sparsity and connectivity effects, and provide reference code to generate the figures and quantitative tables. Keywords: fractured media; geostatistics; SPDE; GMRF; curvelets; ADMM; graph Laplacian; geodesic distance. 1 Introduction and Motivation Classical geostatistics often models correlation as a function of Euclidean distance \left\|\bm{x}-\bm{x}^{\prime}\right\| in \mathbb{R}^{d} , leading to variogram-based kriging and Gaussian random field (GRF) models [ 9 , 23 , 7 ] . Fractured media violate this assumption: transport, flow, and even geochemical alteration frequently propagate preferentially along connected fractures and their intersections [ 1 , 19 , 18 ] . Two points close in Euclidean space may be weakly coupled if separated by intact rock, while distant points may be strongly coupled if connected by a fracture corridor. This motivates two complementary principles: (P1) Along-fracture distance: define smoothness and correlation on a fracture network (polyline graph) and treat Euclidean distance as secondary. (P2) Geometric sparsity: fracture-controlled fields often consist of curvilinear, multiscale, directional features (ridges, fronts, channels) that are sparse in directional multiscale frames such as curvelets (and related systems like ridgelets and shearlets) [ 6 , 5 , 3 , 4 , 14 ] . Scope statement. This paper is a methods framework with a working Version 1 architecture: per-plane inference from 3D trace polylines and sparse samples, solved with a Gaussian baseline and an ADMM curvelet-regularized model. We also formalize the Version 2 extension: explicit along-fracture regularization via graph Laplacians and coupling across plane intersections. We do not overclaim physics completeness: the framework is a statistically controlled interpolation/inversion layer intended to be coupled later to flow/transport models. 2 Related Work and Positioning Fractured media, DFNs, and connectivity-driven transport. Connectivity is a dominant control on fractured-rock transport and effective properties [ 1 , 19 , 10 ] . DFN frameworks and rock-joint system models provide geometric/statistical representations of fracture sets [ 11 , 15 , 18 ] . Our focus is different: given observed traces/planes and sparse property samples, reconstruct a fracture-conditioned property field. SPDE/GMRF baselines. Sparse precision formulations and the SPDE approach link GRFs (e.g., Matérn) to GMRFs, enabling scalable inference on grids [ 20 , 16 ] . We use this as a baseline and diagnostic reference. Directional multiscale transforms in inverse problems and geoscience. Curvelets are near-optimal for representing piecewise smooth functions with C^{2} edges [ 6 ] and are standard in sparsity-promoting inverse problems [ 22 , 17 , 2 ] . In geophysics, curvelet frames are known to sparsify wavefront-like seismic events and support data recovery/interpolation [ 13 , 12 ] . Graphs and network priors. Graph Laplacians provide canonical smoothness functionals on networks and support diffusion/spectral methods [ 8 , 21 ] . This aligns naturally with “along-fracture distance”. 3 Data Model 3.1 Inputs 3D trace polylines. We assume fracture traces are provided as polylines in \mathbb{R}^{3} : \mathcal{T}=\{\gamma_{k}\}_{k=1}^{K},\qquad\gamma_{k}=(\bm{x}_{k,1},\bm{x}_{k,2},\ldots,\bm{x}_{k,N_{k}}),\quad\bm{x}_{k,i}\in\mathbb{R}^{3}. Each \gamma_{k} is interpreted as a trace of a fracture plane (or a segment of it) or as geometry lying near a plane. Sparse samples. We also have sparse measurements of a scalar property: \mathcal{S}=\{(\bm{x}_{j},y_{j},\sigma_{j})\}_{j=1}^{N}, with \bm{x}_{j}\in\mathbb{R}^{3} a location, y_{j}\in\mathbb{R} the observed value, and \sigma_{j}>0 an optional noise standard deviation. 3.2 Physical meaning of the latent field z_{p}(u,v) The quantity estimated by our framework is a latent scalar field on a fracture surface . In Version 1 we work plane-by-plane: for each inferred plane p we define a local chart (u,v) by an orthonormal basis \{\bm{e}_{p,1},\bm{e}_{p,2}\} and an origin \bm{x}_{p,0} on the plane, so that any 3D point \bm{x} assigned to plane p is mapped by projection to u(\bm{x})=\bm{e}_{p,1}^{\top}(\bm{x}-\bm{x}_{p,0}),\qquad v(\bm{x})=\bm{e}_{p,2}^{\top}(\bm{x}-\bm{x}_{p,0}). (1) The unknown is then z_{p}(u,v) , discretized on a 2D grid. When we write z(\bm{x}) informally, we mean the pullback z(\bm{x}):=z_{p(\bm{x})}(u(\bm{x}),v(\bm{x})) under the current plane assignment p(\bm{x}) . What does z_{p} represent physically? Its meaning is determined entirely by the measurement process encoded in the data term (Section 3 ): we assume we observe sparse noisy samples y_{j}\approx z(\bm{x}_{j}) (or a known transform thereof), and we regularize the completion of this field. The method is therefore an interpolation/inversion layer that can be coupled to physics, but does not replace a flow/transport solver. Concrete examples include: (i) z_{p}=\log T (log-transmissivity) or z_{p}=\log b (log-aperture) inferred from hydraulic tests on fractures; (ii) geochemical grade, alteration intensity, or damage indices measured at sparse locations on mapped fractures; (iii) a time- fixed tracer concentration (snapshot) or a scalar summary of a transient (e.g. peak concentration or arrival time). For genuinely time-dependent processes, one would estimate a sequence z_{p}(u,v,t) (multiple independent inversions) or embed the prior into a PDE-constrained inversion; this is beyond Version 1 and is part of the motivation for the “physics coupling” roadmap in Section 14 . Why along-fracture metrics matter. The graph-geodesic/Laplacian constructions (Section 9 ) are appropriate when the effective correlation of the target quantity is governed by connectivity along fractures (e.g. hydraulic/thermal transport corridors), rather than by Euclidean proximity in 3D. If the target quantity is controlled by different physics (e.g. isotropic measurement noise on a planar laboratory proxy), Euclidean priors may suffice. 3.3 Notation summary In per-plane derivations (Sections 6–7) we drop the plane subscript p for readability; all operations are per-plane unless stated otherwise. Table 1 : Core symbols and dimensions used throughout. Symbol Meaning p plane index; each fracture plane (or plane cluster) is solved on its own grid in Version 1 (u,v) local coordinates on a plane chart; u,v\in\mathbb{R} z_{p}\in\mathbb{R}^{M} unknown latent field on plane p discretized on a n_{u}\times n_{v} grid; M=n_{u}n_{v} H_{p}\in\mathbb{R}^{N_{p}\times M} observation operator (bilinear sampling) from grid to the N_{p} samples on plane p y_{p}\in\mathbb{R}^{N_{p}} observed data on plane p (projected from 3D points) L_{uv} 2D grid Laplacian on the (u,v) grid (with explicit boundary conditions) Q_{uv} grid precision (SPDE/GMRF), e.g. Q_{uv}=(\kappa^{2}I+L_{uv})^{2} G=(V,E) trace graph on a plane (nodes are polyline vertices, edges follow traces) L_{G} graph Laplacian on G (encodes along-fracture connectivity) C curvelet transform operator (UDCT/ curvelets ); Cz are curvelet coefficients \lambda,\rho curvelet sparsity weight and ADMM penalty parameter 4 Geometry: From 3D Traces to Plane Charts 4.1 Plane fitting for one polyline Given a polyline \gamma_{k} , we fit a plane via PCA/SVD on