Correspondence Curve in Geometry

• Correspondence Curve is a conceptual structure that defines the mapping and alignment between curves in algebraic, computational, and deep learning contexts.
• It utilizes techniques like parametric reduction and group-equivalence to establish invariant signatures across different geometric and algebraic settings.
• Applications span object-image projections, multi-view curve matching in deep learning, moduli theory, and even connections to enumerative invariants in mathematical physics.

A correspondence curve is a generic conceptual structure and computational object in algebraic geometry, computer vision, tropical geometry, and geometric deep learning. It encodes the matching, alignment, or mapping between two (possibly high-dimensional or multiview) curves or curve segments, capturing global or local equivalence, projection, or functional association. Rigorous treatments appear in algebraic approaches to object-image relations, invariant-theoretic classifications, nonabelian Hodge theory on curves, and curvature-constrained correspondence in multi-view matching.

1. Object-Image Correspondence and Projections

The geometric correspondence curve problem is classically motivated by the question: given a space curve $C \subset \mathbb{R}^3$ and a plane curve $C' \subset \mathbb{R}^2$, does there exist a central or parallel projection $\Pi$ (with unknown parameters) such that $C' = \overline{\Pi(C)}$? In the algebraic setting, this question is inverted to the following: does there exist a projective (or affine) equivalence between $C'$ and a canonical image of $C$ under such a projection?

The solution reduces to two main steps:

1. Parametric Reduction: The naive quantifier-elimination problem, over all $3\times4$ projection matrices and curve parameters, is first reduced to the search for a real (or algebraic) parameter tuple (3 or 2 parameters for central or parallel projections, respectively) describing possible shifts and flattenings of $C$ (Burdis et al., 2013, Burdis et al., 2010).
2. Group-Equivalence of Planar Curves: The projection problem is then transformed into the equivalence problem of planar curves under affine or projective group actions. This is solved using Cartan's method of moving frames and the construction of rational differential signatures—a mapping from the curve to an invariant signature in function space (typically a subvariety of $\mathbb{R}^2$ parameterized by jet coordinates).

The correspondence curve in this context is the graph, or the image, under these signature mappings, uniquely encoding the geometric class of a given curve up to the group action (Burdis et al., 2013).

2. Algebraic Correspondence of Curves and Invariant Forms

In the context of algebraic geometry, a correspondence between two smooth projective curves $C$ and $C' \subset \mathbb{R}^2$0 is defined as a finite subgroupoid (or more concretely, a curve $C' \subset \mathbb{R}^2$1 with two finite morphisms $C' \subset \mathbb{R}^2$2, $C' \subset \mathbb{R}^2$3), whose graph is a 1-dimensional subvariety $C' \subset \mathbb{R}^2$4 (Saha, 2012). The properties of $C' \subset \mathbb{R}^2$5, and of invariant/semi-invariant forms it may admit, are central to understanding the interplay between the curves under the correspondence.

A key aspect is the Abelian group $C' \subset \mathbb{R}^2$6 of semi-invariant differentials (modulo scaling), whose possible structure depends tightly on the degrees of the correspondences. For $C' \subset \mathbb{R}^2$7 and high degree separation, the only correspondences that admit invariant differentials arise from "flat" forms or particular polynomial or Chebyshev/Lattès correspondences, reflecting the rigid structure imposed by the projection (Saha, 2012). The correspondence curve in this sense is the geometric realization of this mapping in the product of curve spaces, equipped with invariant forms.

3. Curve-to-Curve Correspondence in Deep Learning and Application Contexts

In multiview geometric deep learning, the correspondence curve problem is abstracted as the task of matching small curve segments (ordered sets of points) across views, for instance in multiview coronary angiography (Wu et al., 2023). The core innovation is to perform direct curve-to-curve matching, with the following steps:

• Curve Parameterization: The query segment $C' \subset \mathbb{R}^2$8 is an ordered list $C' \subset \mathbb{R}^2$9. The correspondence in the target view is sought not as an unordered set, but as a continuous curve (e.g., cubic Bézier), parameterized by control points.
• Transformer-based Mapping: A network $\Pi$0 takes $\Pi$1 and predicts the control points $\Pi$2 defining the correspondence curve $\Pi$3. This introduces smoothness, order, and topological consistency into the mapping.
• Loss Formulation: The training objective combines Chamfer loss between the predicted and ground-truth curves, control-point supervision, and a cycle-consistency term that enforces invertibility via bidirectional mapping.

Quantitative results show that curve-level correspondences yield reduced error in challenging conditions, relative to point-level methods, due to increased topological awareness and anatomical context encoding (Wu et al., 2023).

4. Correspondence in Moduli and Hodge Theoretic Contexts

A distinct but central notion of correspondence curves appears in the study of moduli spaces of bundles, Higgs fields, and nonabelian Hodge theory. The tame parahoric nonabelian Hodge correspondence on curves provides equivalences of categories:

• Polystable parahoric Higgs bundles on $\Pi$4;
• Polystable parahoric flat connections ("de Rham side");
• Filtered local systems ("Betti side"),

all equipped with compatible local data (weights, residues, monodromy) and subject to matching degree and stability conditions (Huang et al., 2022).

The correspondence curve, in this context, is abstract: it is the identification (bijection of categories) between objects parametrized by moduli spaces, with cohomological and representation-theoretic equivalence. The underlying geometric curve $\Pi$5 is the base of all structures, while the Hodge-theoretic or nonabelian analogs (moduli spaces, local systems) are fibered over this base.

5. Tropical, DT/PT, and Feynman-Integral Correspondences

In enumerative geometry and mathematical physics, correspondence curves are realized in formulaic or categorical relationships between geometric counts and analytic invariants:

• Tropical Correspondence Formula: Tropical curves in $\Pi$6 correspond to holistic Gromov-Witten/Donaldson-Thomas invariants via precise generating functions; each tropical curve’s contribution is encoded in a correspondence formula (e.g., trivalent-vertex quantum weights replacing classical counts) (Parker, 2016).
• DT/PT Correspondence: For smooth curves $\Pi$7 in a Calabi-Yau threefold $\Pi$8, the local DT/PT correspondence relates generating functions of local curve-counting invariants for ideal sheaves (DT) and stable pairs (PT), via equality: $\Pi$9. The curve $C' = \overline{\Pi(C)}$0 itself mediates this local correspondence (Ricolfi, 2017).
• Calabi-Yau-to-Curve Correspondence for Feynman Integrals: Certain periods of families of Calabi-Yau threefolds (e.g., the four-loop banana graph) are locally isomorphic to periods of associated families of genus-2 curves. This holomorphic correspondence identifies the integral structures and periods, with Torelli’s theorem providing the link between abelian varieties and Jacobians of curves (Jockers et al., 2024).

6. Summary Table: Contexts for Correspondence Curve

Context	Mathematical Object/Action	Key Reference/arXiv ID
Spatial-to-planar images	Planar/projective/affine curve equivalence	(Burdis et al., 2013, Burdis et al., 2010)
Algebraic correspondences	Graph/subvariety $C' = \overline{\Pi(C)}$1	(Saha, 2012)
Deep multiview matching	Curve segment mapping (transformer-based)	(Wu et al., 2023)
Hodge moduli theory	Category equivalence over base curve $C' = \overline{\Pi(C)}$2	(Huang et al., 2022)
Enumerative invariance	Tropical→GW/DT/PT mapping, local invariants	(Parker, 2016, Ricolfi, 2017)
Feynman periods	Period map: Calabi-Yau $C' = \overline{\Pi(C)}$3 genus-2 curves	(Jockers et al., 2024)

In all of these, the term "correspondence curve" refers to an underlying or induced curve (in parametric, modular, combinatorial, or categorical sense) which functions as a linkage between two geometric, analytic, or algebraic structures. The operationalization of correspondences, and the alignment of invariants, are central to their identification and computational realization.