Arc Ridge: Multidisciplinary Insights

• Arc Ridge is a multidisciplinary concept describing a narrow locus where physical, spectral, molecular, elastic, or variational quantities concentrate.
• In cuprate superconductors, it appears as a ridge of zero-energy spectral weight formed by phase fluctuations, distinguishable from a conventional Fermi arc.
• Across astronomy, molecular studies, elasticity, and neural networks, Arc Ridge denotes both physical structures and algorithmic regularizers, highlighting its diverse applications.

“Arc Ridge” is not a single standardized object in the arXiv corpus. In the cited literature it denotes several distinct structures: a ridge-like enhancement of zero-energy spectral weight in a phase-disordered $d$-wave superconductor; large-scale radio or molecular arcs in the interstellar medium and Galactic center; one-dimensional ridges connecting disclinations in thin elastic sheets; and a path-length-based regulariser for wide feature-learning neural networks (Li et al., 2010, Bracco et al., 2023, Hsieh et al., 2015, Gladbach et al., 2024, Yusef-Zadeh, 2024, Whittle et al., 18 May 2026). This suggests a recurring descriptive logic: an “arc ridge” is typically a narrow locus along which some physically or algorithmically relevant quantity concentrates, but the quantity itself may be spectral, radiative, molecular, elastic, or variational.

1. Terminological scope across research domains

The explicit uses in the cited works are heterogeneous rather than taxonomically unified.

Domain	Object called “arc ridge”	Defining sense
Underdoped cuprates	Ridge in $A(\mathbf{k},\omega=0)$	Zero-energy spectral-weight accumulation along the underlying Fermi surface
Local ISM radio astronomy	Orion–Taurus ridge	Arc-shaped synchrotron loop on the northern wall of the Orion–Eridanus superbubble
Galactic-center molecular gas	Connecting ridge + polar arc	Dense-gas ridge linking disk gas to an extraplanar molecular arc
Thin-sheet elasticity	Ridges between disclinations	One-dimensional singular structures joining point defects
Wide neural networks	Arc ridge	Squared total path length used as a regulariser

In condensed-matter usage, the ridge is an object in momentum-resolved spectroscopy rather than in real space. In radio astronomy and Galactic-center molecular studies, it is a spatial structure observed in synchrotron or molecular-line maps. In elasticity, it is a singular set of the out-of-plane displacement in a convex Föppl–von Kármán setting. In machine learning, it is a regularisation functional defined on parameter trajectories rather than on a spatial manifold.

A recurrent misconception is that the phrase should refer to a curved arc in ordinary geometry. Several cited works contradict that expectation. In the cuprate study, the ridge is a spectral accumulation that only looks like a Fermi arc once finite experimental resolution and thresholding are imposed. In the elastic-sheet study, the relevant connecting structures are straight line segments rather than curved arcs. In the neural-network study, “arc” refers to path geometry in parameter space, not to a visible geometric curve in physical space.

2. Spectral “arc ridge” in underdoped cuprates

In "On the origin of the Fermi arc phenomena in the underdoped cuprates: signature of KT-type superconducting transition" (Li et al., 2010), the term refers to a ridge-like enhancement of the zero-energy spectral weight $A(\mathbf{k},\omega=0)$ running along the underlying Fermi surface. The spectral function is defined by

$$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$

and is computed by solving a Bogoliubov–de Gennes problem for each Monte Carlo phase configuration of a bond-centered XY model with built-in $d$-wave character. The electrons live on a $64\times 64$ square lattice with $t'/t=-0.3$, $\Delta/t=0.1$, and temperatures $k_BT/J=0.5,1.0,1.1,1.2,1.3,1.4$.

The low-temperature reference behavior is the coherent $d$-wave state: the gap vanishes only at the four nodal points, so $A(\mathbf{k},\omega=0)$0 consists of four sharp nodal peaks and no arc exists. Above the Kosterlitz–Thouless scale of the XY model, a qualitatively new structure appears: a continuous ridge of $A(\mathbf{k},\omega=0)$1 emerges along the normal-state Fermi surface $A(\mathbf{k},\omega=0)$2, with strongest weight near the nodes and suppressed but nonzero weight toward the antinodes. The ridge appears abruptly upon crossing the finite-size KT remnant near $A(\mathbf{k},\omega=0)$3, grows rapidly just above that scale, and then increases more slowly at higher temperatures.

The key interpretive point is negative: this is not a Landau Fermi-surface segment. The paper defines the ridge as a pile-up of zero-energy spectral weight caused by phase fluctuations in a $A(\mathbf{k},\omega=0)$4-wave paired state with fixed pairing amplitude and fluctuating bond phase. Once experimental resolution and thresholding are imposed, the visible portion of the ridge becomes what angle-resolved photoemission would identify as a Fermi arc. The authors therefore distinguish the simulated “ridge” from the experimentally observed “arc”: the former is the full continuous enhancement in $A(\mathbf{k},\omega=0)$5, while the latter is the subset exceeding experimental detectability.

The physical mechanism is the proliferation of vortex-like phase fluctuations above the KT transition. Below $A(\mathbf{k},\omega=0)$6, vortices are bound and the effect on $A(\mathbf{k},\omega=0)$7 is negligible. Above $A(\mathbf{k},\omega=0)$8, free vortices disorder the phase field, redistribute spectral weight to $A(\mathbf{k},\omega=0)$9, and generate the ridge. The paper explicitly places this in dialogue with earlier semiclassical phase-fluctuation treatments by Franz and Millis and by Berg and Altman, but goes beyond them by Monte Carlo sampling of full vortex-containing phase configurations and exact diagonalisation of the BdG problem for each sample.

3. Radio and synchrotron arc ridges in the interstellar medium

In local-Galactic radio astronomy, "The Orion-Taurus ridge: a synchrotron radio loop at the edge of the Orion-Eridanus superbubble" (Bracco et al., 2023) defines the Orion–Taurus ridge as a $A(\mathbf{k},\omega=0)$0-wide radio arc in Orion, approximately centered at $A(\mathbf{k},\omega=0)$1. Using Long Wavelength Array total-intensity maps at 50–80 MHz, Planck 30 GHz polarization, 3D dust maps, and molecular-gas column densities, the authors place the ridge at about $A(\mathbf{k},\omega=0)$2 pc and infer a plane-of-the-sky extent of about $A(\mathbf{k},\omega=0)$3 pc. The ridge lies on the northern wall of the Orion–Eridanus superbubble and is traced by dust integrated between 300 and 500 pc, with a radial-profile coincidence between dust and radio peaks quantified by a median offset $A(\mathbf{k},\omega=0)$4 deg.

Its gas and synchrotron properties are unusually specific. The median molecular column is

$A(\mathbf{k},\omega=0)$5

and the corrected low-frequency synchrotron spectral index has median

$A(\mathbf{k},\omega=0)$6

The authors interpret the flatness of $A(\mathbf{k},\omega=0)$7 as evidence for depletion of low-energy $A(\mathbf{k},\omega=0)$8 cosmic-ray electrons and infer plane-of-the-sky magnetic fields larger than a few tens of $A(\mathbf{k},\omega=0)$9G, specifically $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$0–$$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$1G. They describe this as the first detection of diffuse synchrotron emission from cold-neutral, partly molecular gas in the surroundings of the Orion–Eridanus superbubble. In this usage, the ridge is a feedback-compressed, magnetised shell segment rather than a generic radio loop.

A different astronomical usage appears in "Radio study of the Arc and the Sgr A complex near the Galactic center" (Yusef-Zadeh, 2024). There the “Arc” is a $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$2 pc-scale network of narrow linear filaments at $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$3, oriented nearly perpendicular to the Galactic plane, strongly polarized at 6, 3, and 2 cm, and characterised by a flat spectrum. Distinct from these ordered nonthermal filaments are thermal arched filamentary structures identified through radio recombination lines. The same thesis also reports “a very steep-spectrum ridge of emission” emerging from Sgr A and extending roughly perpendicular to the plane, interpreted as a possible low-energy jet from the Galactic nucleus. The combined interpretation is magnetic: the Arc traces a coherent poloidal field over a large central-Galactic volume, while the steep-spectrum ridge suggests outflow from Sgr A along related field geometry.

These two radio-astronomical meanings are structurally similar but physically distinct. The Orion–Taurus ridge is a shell wall in the Solar neighborhood associated with recent superbubble feedback, cold partly molecular gas, and low-frequency synchrotron emissivity. The Galactic-center Arc and Sgr A ridge are components of a magnetically dominated radio continuum complex, involving flat-spectrum polarized filaments, thermal arches, and a possible jet-like steep-spectrum feature.

4. Molecular arc–ridge systems in the Galactic center

In "The connecting molecular ridge in the Galactic center" (Hsieh et al., 2015), “arc ridge” is best understood as a coupled system comprising the polar arc (PA), the connecting ridge (CR), and the disk ridge (DR). The polar arc is an extraplanar molecular ridge north of Sgr A, extending from approximately $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$4 to $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$5, rising at a projected angle of about $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$6 above the Galactic plane and exhibiting a large velocity gradient perpendicular to the plane. The newly identified connecting ridge is a continuous, elongated band of dense gas, most clearly seen in CS(J=4–3), with width $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$7 ($$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$8 pc), running roughly perpendicular to the plane and linking the DR in the disk to the base of the PA in both spatial and position–velocity space.

The observational basis is multi-transition CS mapping with the Caltech Submillimeter Observatory combined with Nobeyama data, all compared at $$A(\mathbf{k},\omega)=-\frac{1}{\pi}\,\mathrm{Im}\,G^R(\mathbf{k},\omega),$$9 resolution ($d$0 pc at 8.5 kpc). The CR is remarkable because it is prominent in CS(J=4–3) but much less obvious in lower-J maps, a behavior attributed to a combination of high excitation and low surface brightness temperature. The paper reports CS(J=4–3)/CS(J=2–1) ratios $d$1 in the relevant structure and argues that the CR, DR, and PA are one coherent system based on spatial continuity, PV continuity, enhanced line widths, and comparable excitation conditions.

The inferred physical conditions from RADEX are closely matched across the three components. For the CR, the best-fit values are $d$2, $d$3 K, and $d$4; the DR and base of the PA differ only modestly from this, with $d$5 and $d$6, respectively. This strengthens the authors’ claim that the DR–CR–PA sequence is a single lifted molecular structure rather than a chance superposition.

The dynamical interpretation is feedback- or field-driven vertical transport. Using a molecular mass $d$7 and expansion speed $d$8, the paper estimates a kinetic energy $d$9 erg for the PA, consistent with multiple supernovae and/or stellar winds, while also discussing magnetic buoyancy or Parker instability as an alternative. In this Galactic-center molecular context, the “ridge” is literally the missing link between disk gas and an extraplanar arc.

5. Ridges between disclinations in thin elastic sheets

In "Connecting disclinations by ridges" (Gladbach et al., 2024), arc-ridge language is realized in a variational elasticity setting rather than in observational astronomy or spectroscopy. The paper studies a thin sheet with finitely many disclinations encoded by the discrete Monge–Ampère measure

$64\times 64$0

within a Föppl–von Kármán approximation and under the non-physical assumption that the out-of-plane displacement $64\times 64$1 is convex. The reference singular displacement $64\times 64$2 solves $64\times 64$3 in $64\times 64$4 with $64\times 64$5 on $64\times 64$6, and the elastic energy is

$64\times 64$7

The central structural theorem states that the singular set of $64\times 64$8 is a finite union of straight line segments $64\times 64$9 connecting pairs of disclinations. Around each such segment there is a rhombus $t'/t=-0.3$0 on whose two triangular halves $t'/t=-0.3$1 is affine with two distinct gradient values. The paper therefore uses “ridge” in a precise geometric sense: a one-dimensional jump set of $t'/t=-0.3$2 that connects point defects. A crucial caution follows immediately: although the broader phrase “arc ridge” might suggest curved connectors, the proven convex FvK geometry produces straight segments, not curved arcs.

The energy scaling is the main quantitative result. For convex admissible $t'/t=-0.3$3, the minimum satisfies

$t'/t=-0.3$4

The lower and upper bounds match up to logarithmic factors. By contrast, if convexity is removed, the paper proves an upper bound $t'/t=-0.3$5. It follows that for sufficiently small $t'/t=-0.3$6, true unconstrained minimizers cannot remain convex. The convexity hypothesis is therefore analytically useful but physically restrictive: it selects a higher-energy ridge network than the one expected for realistic crumpled sheets.

A notable technical ingredient is the generalized monotonicity of the Monge–Ampère measure: if $t'/t=-0.3$7 are convex, coincide on $t'/t=-0.3$8, and $t'/t=-0.3$9 is nonnegative and concave, then

$\Delta/t=0.1$0

This weighted comparison is used to control how curvature can concentrate near ridge cells and to derive the lower bound.

6. Arc ridge in wide feature-learning neural networks

In "Canonical Regularisation of Wide Feature-Learning Neural Networks" (Whittle et al., 18 May 2026), arc ridge is a regulariser rather than a physical structure. The paper distinguishes the kernel regime, where the Jacobian is effectively constant and gradient flow selects the anchored-ridge solution, from the feature-learning regime, where the Jacobian evolves and ordinary ridge biases the training dynamics even in the vanishing-regularisation limit. The canonical regulariser in the kernel regime is anchored ridge, but the paper proves that in the feature-learning regime both anchored ridge and standard ridge distort the inductive bias by pushing parameters along output fibres $\Delta/t=0.1$1, including for pretrained models where the initial implicit prior is informative.

The paper’s canonical alternative is geodesic ridge, defined through the Riemannian geometry of the flow manifold $\Delta/t=0.1$2 and the output fibres $\Delta/t=0.1$3. Geodesic ridge is theoretically canonical but computationally intractable. Arc ridge is introduced as a scalable surrogate based on the squared total path length of the training trajectory,

$\Delta/t=0.1$4

Under exact gradient flow, it upper-bounds geodesic ridge and sits in a “sandwich”

$\Delta/t=0.1$5

The paper argues that, given only this interval for the true canonical penalty, the upper bound is the minimax-robust choice.

The practical attraction is geometric alignment. The endpoint gradient of arc ridge is parallel or anti-parallel to the realised training velocity, so it remains tangent to the flow manifold and does not introduce a fibre component. This is the core sense in which arc ridge is geometry-respecting. The paper also proves that, under gradient flow, arc-ridge-regularised training follows the same path as unregularised gradient flow up to a stopping time determined by

$\Delta/t=0.1$6

where $\Delta/t=0.1$7. Arc ridge is therefore path-equivalent to early stopping in continuous time.

Empirically, the paper studies image and NLP transfer-learning problems, specifically ResNet-18 on UTKFace age regression and DistilBERT on Yelp review regression. Standard and anchored ridge degrade sharply at high regularisation strength, whereas arc ridge degrades more gradually and behaves like controlled undertraining rather than prior destruction. In this literature, “arc ridge” does not mean weight decay; it means path-length regularisation derived as a practical surrogate to geodesic ridge in feature-learning geometry.

7. Comparative interpretation and terminological cautions

Across these literatures, “arc ridge” is a family resemblance term rather than a universal concept. The commonality is concentration along a lower-dimensional locus. In the cuprate paper, the locus lies on the underlying Fermi surface in momentum space. In the Orion–Taurus study and the Galactic-center radio thesis, it is a synchrotron-emitting structure embedded in a magnetised interstellar environment. In the Galactic-center molecular study, it is a dense-gas bridge between planar and extraplanar components. In the elastic-sheet paper, it is a singular line carrying curvature and strain. In the neural-network paper, it is the accumulated length of a parameter-space trajectory.

Several terminological cautions follow. First, an “arc ridge” need not be a true arc: the convex FvK analysis yields straight ridge segments between disclinations. Second, an “arc ridge” need not be a real-space object: the cuprate usage is spectral, and the neural-network usage is variational. Third, an “arc ridge” need not represent a genuine underlying manifold or surface. The cuprate paper explicitly states that the ridge is not a true piece of Fermi surface in the Landau sense, while the neural-network paper explicitly distinguishes arc ridge from classical ridge regularisation.

A related but terminologically distinct extension appears in "Polynomial Ridge Flowfield Estimation" (Scillitoe et al., 2021). That paper does not use the phrase “Arc Ridge,” but it studies local polynomial ridge functions $\Delta/t=0.1$8, sufficient summary plots, and low-dimensional ridge coordinates that the accompanying explanation characterizes as arc-like design-space manifolds. This suggests that the descriptive pull of “arc ridge” can extend beyond literal nomenclature to any setting where dominant variation is organised along a low-dimensional curve or subspace.

Taken together, these usages show that “Arc Ridge” is best treated encyclopedically as a cross-disciplinary label for ridge-dominated organization rather than as the name of a single phenomenon. Its meaning is fixed not by the phrase alone but by the governing observable: $\Delta/t=0.1$9, synchrotron emissivity, molecular-line kinematics, Monge–Ampère/FvK elasticity, or the geometry of gradient-flow trajectories.