OffsetCrust: Variable-Radius Offset Surfaces

• OffsetCrust is a framework that approximates variable-radius offset surfaces by recasting computation as a power-diagram separation problem, enhancing geometrical accuracy.
• It employs a dual-site strategy that samples base and displaced points to capture non-normal displacement directions while refining vertices via local least-squares optimization.
• The method demonstrates advantages over constant-radius techniques and provides insights applicable to neutron-star studies where crust-induced offsets are analyzed.

Searching arXiv for the exact term to ground the article in the relevant paper and adjacent uses. OffsetCrust is a framework for approximating variable-radius offset surfaces by recasting offset computation as a power-diagram separation problem. In its explicit, formal usage, the term refers to the geometry-processing method introduced in "OffsetCrust: Variable-Radius Offset Approximation with Power Diagrams" (Zhao et al., 15 Jul 2025), where a base surface is sampled, paired with off-surface weighted sites, and the desired offset is extracted from power-diagram facets separating the two site classes. In a distinct and nonstandard secondary usage, closely related neutron-star papers use “offset crust” only as an interpretive description for crust-induced shifts, matching effects, or low-density systematic biases, rather than as a named framework (Pan et al., 2020, Gamba et al., 2019, Davis et al., 2024).

1. Formal definition in geometry processing

In the geometry-processing literature, OffsetCrust addresses the problem of computing variable-radius offset surfaces. Given a base surface $\mathcal S$ and a positive radius function $\mathcal R:\mathcal S\to \mathbb R_{>0}$, the offset is formulated through the generalized distance field

$$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$

with the variable-radius offset surface defined as the zero level set

$\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$

For constant radius $r$, this reduces to the usual Minkowski-sum view $\mathcal S \oplus B_r$; for variable radius, each point $p\in\mathcal S$ contributes its own ball $B_{\mathcal R(p)}(p)$, and the offset is the envelope of these balls (Zhao et al., 15 Jul 2025).

The radius field is defined directly on the base surface,

$\mathcal R:\mathcal S\to \mathbb R_{>0},$

and for mesh input it is typically given at vertices and assumed linear over each triangle. If sparse values are given, missing values are interpolated by solving the biharmonic equation

$\Delta^2 \mathcal R = 0.$

A key assumption used by the method is

$\mathcal R:\mathcal S\to \mathbb R_{>0}$0

which ensures that the displacement-direction construction is valid and is also consistent with medial axis transforms, where balls are maximal and non-nested (Zhao et al., 15 Jul 2025).

This formulation distinguishes OffsetCrust from constant-radius offsetting methods. The central difficulty is that, in the variable-radius setting, the envelope geometry is no longer captured by a uniform normal displacement. OffsetCrust therefore replaces the standard normal-offset construction with a weighted arrangement in which the target surface appears as a bisector structure in power distance (Zhao et al., 15 Jul 2025).

2. Power-diagram construction and geometric principle

OffsetCrust uses a power diagram of weighted sites. For weighted sites $\mathcal R:\mathcal S\to \mathbb R_{>0}$1, the power distance is

$\mathcal R:\mathcal S\to \mathbb R_{>0}$2

The method constructs two classes of sites: base sites on $\mathcal R:\mathcal S\to \mathbb R_{>0}$3, weighted by $\mathcal R:\mathcal S\to \mathbb R_{>0}$4, and displaced sites off the surface, weighted by $\mathcal R:\mathcal S\to \mathbb R_{>0}$5. The offset surface is then approximated by the power-diagram facets separating these two groups (Zhao et al., 15 Jul 2025).

The continuous geometric argument is based on a contributing point $\mathcal R:\mathcal S\to \mathbb R_{>0}$6 and its offset point $\mathcal R:\mathcal S\to \mathbb R_{>0}$7. For small $\mathcal R:\mathcal S\to \mathbb R_{>0}$8, the paper defines

$\mathcal R:\mathcal S\to \mathbb R_{>0}$9

with $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$0, $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$1, and $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$2 colinear. The offset point satisfies

$$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$3

and in squared form

$$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$4

This is the reason the envelope can be represented through weighted bisectors, and hence through a power-diagram construction (Zhao et al., 15 Jul 2025).

The site construction follows this derivation directly. For each contributing base point $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$5, OffsetCrust creates the base site $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$6 and the displaced site $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$7. The extracted surface is the subset of the power diagram that separates base from displaced sites (Zhao et al., 15 Jul 2025).

The framework also explains why non-contributing points do not affect the envelope. If an offset point $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$8 lies inside the envelope, then some other site $$\phi(x)\coloneqq \min_{p\in\mathcal S}\big(\|x-p\|-\mathcal R(p)\big),$$9 satisfies

$\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$0

equivalently

$\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$1

Since $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$2, it follows that $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$3 belongs to another power cell. This means only contributors to the envelope matter in the final separation structure (Zhao et al., 15 Jul 2025).

3. Displacement directions and the variable-radius deviation from normal offsets

A defining feature of OffsetCrust is that, in the variable-radius case, the displacement direction is generally not the surface normal. For constant $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$4, the offset direction aligns with the surface normal; this is the classical normal-offset regime. For spatially varying radii, the paper proves that if $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$5, the offset direction $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$6 is obtained by rotating the unit surface normal $\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$7 by the angle

$\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$8

around the axis

$\mathcal S_{\mathcal R}^{\text{off}=\{x\mid \phi(x)=0\}.$9

The geometric interpretation given is that spatial variation in $r$0 tilts the tangent of the envelope away from the base normal; only when $r$1 do the directions coincide with the normals (Zhao et al., 15 Jul 2025).

This displacement-direction theorem is the principal reason a direct transplant of constant-radius crust methods is insufficient. The method must account for a direction field determined jointly by the base geometry and the radius gradient. In the language of the paper, the off-surface points are displaced along $r$2-dependent directions, and only in the constant-radius case do these directions align exactly with the surface normals of $r$3 (Zhao et al., 15 Jul 2025).

A practical implication is that variable-radius offsetting is not merely a weighted version of standard offsetting. The site construction, power weights, and later refinement stages are all designed to manage the mismatch between base normals and true envelope directions. This suggests that the principal technical novelty of OffsetCrust lies not only in adopting power diagrams, but in making them compatible with non-normal displacement geometry (Zhao et al., 15 Jul 2025).

4. Misalignment, fine-tuning, and sampling pipeline

OffsetCrust inherits a known weakness of crust-based approaches: misaligned facets. A facet in the separating structure may be formed either by a base point $r$4 and its own displaced point or by a base point $r$5 and the displaced point of another site. These “cross-pair” facets generate misalignment, and the paper notes that the problem is more likely in the variable-radius setting because displacement directions differ from the normal and vary spatially (Zhao et al., 15 Jul 2025).

To reduce this artifact, the method uses a local least-squares vertex refinement. If $r$6 are the base points contributing to a power-diagram vertex $r$7, with

$r$8

then the ideal constraint is

$r$9

OffsetCrust therefore minimizes

$\mathcal S \oplus B_r$0

and stabilizes it with a small regularizer,

$\mathcal S \oplus B_r$1

where $\mathcal S \oplus B_r$2 is the original vertex and $\mathcal S \oplus B_r$3 is small, typically $\mathcal S \oplus B_r$4. The closed-form solution is

$\mathcal S \oplus B_r$5

with

$\mathcal S \oplus B_r$6

$\mathcal S \oplus B_r$7

The paper characterizes this step as “lightweight” because it is per-vertex, local, quadratic, and solved by a small linear system; no global optimization or explicit self-intersection repair is needed (Zhao et al., 15 Jul 2025).

The sampling pipeline is correspondingly specialized. Triangle-interior points are sampled by blue noise, each generating one displaced point. Edge-type points and vertex-type points use a one base point, multiple displaced points (1vN) strategy. For a manifold edge shared by faces $\mathcal S \oplus B_r$8, multiple directions are generated by spherical linear interpolation between face-based directions. Around each vertex $\mathcal S \oplus B_r$9, the method places a small sphere of radius $p\in\mathcal S$0, intersects it with the surface, samples the resulting closed trajectory, and again uses spherical linear interpolation to generate displaced points. Interior triangle points inside this protected sphere are removed so that vertex-generated displaced points have priority. An optional dihedral-angle threshold can pre-detect sharp feature lines so that 1vN is only applied where needed (Zhao et al., 15 Jul 2025).

These design choices indicate that OffsetCrust is not only a theoretical reformulation. It is also a sampling and reconstruction pipeline adapted to sharp features, anisotropic offset directions, and local ambiguity in the crust-style separating structure (Zhao et al., 15 Jul 2025).

5. Implementation, evaluation, and application domain

The implementation reported for OffsetCrust is in C++, using CGAL exact predicates/exact constructions, TBB for parallelization, Eigen LDLT for the refinement solve, and AABB trees, PQP, and libigl for distance and inside/outside queries. Experiments were run on a machine with Intel i9-13900K and 64 GB RAM (Zhao et al., 15 Jul 2025).

Typical settings reported are 70K blue-noise samples, $p\in\mathcal S$1, $p\in\mathcal S$2, $p\in\mathcal S$3, and a discrete spherical surface with 642 vertices. For evaluation, the paper uses CD (Chamfer Distance), HD (Hausdorff Distance), and NC (Normal Consistency). For constant-radius accuracy it also uses a one-sided normalized distance metric,

$p\in\mathcal S$4

The reported average runtime is about 100 seconds with roughly $p\in\mathcal S$5 samples. Power-diagram computation is the main bottleneck, facet and adjacency extraction is also costly, and the refinement step is relatively cheap (Zhao et al., 15 Jul 2025).

The empirical comparison is made against dual contouring at $p\in\mathcal S$6. OffsetCrust is reported to achieve comparable or better CD/HD stability, fewer outliers, and strongly improved NC after refinement. For inward offsets it often performs better than dual contouring in NC, while outward offsets are competitive though not always superior. The paper also reports good performance on 2K Thingi10K models, strong robustness for $p\in\mathcal S$7, and a roughly order-of-magnitude improvement in normal consistency after refinement in many cases. Ablation studies indicate that more blue-noise samples improve accuracy but increase runtime, smaller $p\in\mathcal S$8 helps small offsets preserve sharpness, finer spherical interpolation improves rounded regions and accuracy, and overly aggressive sharp-feature filtering hurts quality (Zhao et al., 15 Jul 2025).

A prominent application is medial axis transform (MAT) reconstruction. In this setting, the MAT surface is the base surface, radii are already defined at MAT elements, and the recovery of the original boundary becomes a variable-radius offset problem. The method is particularly natural here because MAT balls satisfy the condition

$p\in\mathcal S$9

The paper reports faithful reconstructions of original surfaces with good CD, HD, and NC, while preserving geometry and topology without requiring explicit envelope handling (Zhao et al., 15 Jul 2025).

6. Secondary usages in neutron-star research

Outside geometry processing, “offset crust” is not a standardized term but appears as an interpretive label for several distinct crust-related effects in neutron-star research. These usages should be distinguished from the formal framework named OffsetCrust (Pan et al., 2020, Gamba et al., 2019, Davis et al., 2024).

One usage concerns a radius posterior shift caused by the low-density crust equation of state. In "The impact of the crust equation of state on the analysis of GW170817" (Gamba et al., 2019), the central result is that different crust models do not strongly impact the mass or tidal deformability of a neutron star, but they do affect the inferred radius. Earlier GW170817 analyses fixed densities below $B_{\mathcal R(p)}(p)$0 to the SLy description, yielding

$B_{\mathcal R(p)}(p)$1

Reanalysis with a changed crust gave

$B_{\mathcal R(p)}(p)$2

and the paper estimates a crust-induced systematic radius shift of about $B_{\mathcal R(p)}(p)$3, approximately $B_{\mathcal R(p)}(p)$4 of the neutron-star radius. In that context, “offset crust” refers to the fact that the credible region in $B_{\mathcal R(p)}(p)$5 is nearly unchanged while the radius posterior is shifted (Gamba et al., 2019).

A related but distinct usage concerns non-unified crust-core matching and inference bias. "Inference of neutron-star properties with unified crust-core equations of state for parameter estimation" (Davis et al., 2024) develops CUTER to consistently match a nuclear-physics-informed crust to an arbitrary high-density EoS. The paper argues that a fixed, realistic-but-inconsistent crust causes small but avoidable errors in the estimation of global neutron-star properties and leads to an underestimation of uncertainties. Quantitatively, for tested models the relative error in radius after CUTER reconstruction is $B_{\mathcal R(p)}(p)$6, while fixed-crust treatments can slightly shift medians and shrink posterior widths. For example, under LD+HD filters only with $B_{\mathcal R(p)}(p)$7 fm$B_{\mathcal R(p)}(p)$8, unified treatment gives

$B_{\mathcal R(p)}(p)$9

whereas a unique SLy4 crust gives

$\mathcal R:\mathcal S\to \mathbb R_{>0},$0

Here the “offset” is an inference-level bias induced by inconsistent low-density modeling (Davis et al., 2024).

A third usage is found in "Probing Crust Meltdown in Inspiraling Binary Neutron Stars" (Pan et al., 2020), where the relevant effect is a crust-meltdown-induced phase offset in the gravitational-wave signal. There the inspiraling companion excites crust-core interface modes, the crust yields plastically once the local elastic strain exceeds a breaking strain of order

$\mathcal R:\mathcal S\to \mathbb R_{>0},$1

and dissipative heating eventually melts the crust. The resulting waveform modification is written as

$\mathcal R:\mathcal S\to \mathbb R_{>0},$2

with the paper quoting a $\mathcal R:\mathcal S\to \mathbb R_{>0},$3 phase shift and noting that, for an equal-mass $\mathcal R:\mathcal S\to \mathbb R_{>0},$4 binary, $\mathcal R:\mathcal S\to \mathbb R_{>0},$5 varies from about $\mathcal R:\mathcal S\to \mathbb R_{>0},$6 to $\mathcal R:\mathcal S\to \mathbb R_{>0},$7 depending on the EoS and $\mathcal R:\mathcal S\to \mathbb R_{>0},$8. In that setting, “offset crust” denotes a localized, resonance-triggered phase and time jump in the inspiral waveform, not a surface-reconstruction method (Pan et al., 2020).

A further neutron-star usage concerns crust-core interpolation. "Consistent crust-core interpolation and its effect on non-radial neutron star oscillations" (Canullan-Pascual et al., 4 Feb 2025) studies a thermodynamically and causally consistent interpolation in the pressure–chemical-potential plane, contrasting it with classical concatenation. The interpolation satisfies

$\mathcal R:\mathcal S\to \mathbb R_{>0},$9

introduces an energy-density gap corresponding to a first-order phase transition, and affects low-mass-star radii, crust thickness, and especially the $\Delta^2 \mathcal R = 0.$0-mode frequency, while leaving the $\Delta^2 \mathcal R = 0.$1-mode essentially unchanged. Here again the phrase is descriptive rather than terminologically fixed (Canullan-Pascual et al., 4 Feb 2025).

7. Conceptual scope and distinctions

The term OffsetCrust therefore has a sharply defined meaning in one domain and only a loose descriptive role in another. In geometry processing, it denotes a named framework whose central ingredients are a variable-radius offset formulation, weighted base and displaced sites, power-diagram extraction, and lightweight local refinement (Zhao et al., 15 Jul 2025). In neutron-star studies, by contrast, “offset crust” has been used only to describe crust-induced shifts in radii, waveform phase, or crust-core matching, and the underlying phenomena are physically unrelated to the geometric method (Gamba et al., 2019, Pan et al., 2020, Davis et al., 2024).

This distinction matters because the two literatures use similar words for entirely different technical objects. In geometry processing, the offset is a surface envelope generated by balls of spatially varying radius, and the “crust” refers to a class of reconstruction methods inspired by crust-based surface extraction (Zhao et al., 15 Jul 2025). In neutron-star astrophysics, the offset is a shift in an inferred or observable quantity produced by crust microphysics or EOS treatment, while “crust” refers to the star’s low-density solid outer layers (Gamba et al., 2019, Pan et al., 2020).

A plausible implication is that the geometry-processing usage is likely to dominate the exact proper noun OffsetCrust, because it appears as the title of a dedicated framework (Zhao et al., 15 Jul 2025). The neutron-star usages remain valuable for query interpretation, but they are better understood as context-dependent descriptive phrases rather than as a unified concept.