Geodesically Convex Subspace

Updated 7 February 2026

Geodesically convex subspace is defined as a set in a Riemannian manifold where any two points are connected by a unique minimizing geodesic that remains entirely within the set.
It generalizes Euclidean convexity by substituting straight lines with geodesics, reflecting intrinsic curvature and structure in complex spaces.
This concept underpins modern optimization and analysis on non-Euclidean manifolds, with applications ranging from robotics to computational geometry.

A geodesically convex subspace—also called a geodesically convex set or g-convex set—is a subset $S$ of a Riemannian manifold (or, more generally, a metric space with a geodesic structure) such that for every pair of points $x, y \in S$ , there exists a unique (minimizing) geodesic joining $x$ and $y$ that lies entirely within $S$ . This property generalizes the familiar notion of convexity from Euclidean spaces—where straight lines joining pairs of points in a convex set remain inside the set—by replacing straight lines with geodesics, which are the natural notion of “straightness” in curved or abstract spaces. Geodesic convexity is foundational in the analysis, optimization, and geometry of non-Euclidean spaces, and it underpins modern approaches to optimization, analysis, and algorithm design in manifold settings.

1. Foundations and Definitions

Let $(M, g)$ be a connected Riemannian manifold with associated metric $g$ and distance function $d_g$ . A smooth curve $\gamma: [0,1] \to M$ has length

$L(\gamma) = \int_0^1 \sqrt{\langle \dot\gamma(t), \dot\gamma(t)\rangle_{\gamma(t)}} \, dt,$

and the (Riemannian) geodesic distance between points $x, y \in M$ is

$d_g(x, y) = \inf\{ L(\gamma) \mid \gamma(0)=x, \gamma(1)=y \}.$

A geodesic is a curve $\gamma$ with $\nabla_{\dot\gamma}\dot\gamma=0$ , and a minimizing geodesic is one which realizes this infimum.

Definition: A subset $S \subseteq M$ is geodesically convex if for every $x, y \in S$ , there is a unique minimizing geodesic $\gamma: [0,1] \to M$ with $\gamma(0) = x$ , $\gamma(1) = y$ , and $\gamma([0,1]) \subset S$ (Cohn et al., 2023).

This concept extends naturally to metric spaces that admit geodesics, such as Hadamard (CAT(0)) spaces, Wasserstein spaces, and quotient manifolds.

2. Characterization, Properties, and Comparison

A geodesically convex subspace possesses several key properties, many of which mirror Euclidean convexity in a localized or global sense:

Closure under Geodesic Segmentation: If $x, y \in S$ , then the entire minimizing geodesic connecting $x$ and $y$ remains in $S$ (Vishnoi, 2018, Goyal et al., 2019).
Intersection Stability: The intersection of any collection of geodesically convex sets is again geodesically convex (Goyal et al., 2019).
Geodesic Convex Hull: The smallest geodesically convex set containing a given subset $A$ is its geodesic convex hull, constructed as the intersection of all geodesically convex sets containing $A$ (Goyal et al., 2019, Kiliç et al., 2015).
Locally Convex Neighborhoods: In Riemannian manifolds, sufficiently small geodesic balls (radius below the convexity/injectivity radius) are always geodesically convex (Khajehpour et al., 2018, Goyal et al., 2019).
Total Geodesic Convexity: A “totally geodesically convex” set is one where every geodesic (not just the minimizing one) between points in the set remains inside the set (Vishnoi, 2018). This is a stronger property than ordinary geodesic convexity and corresponds to totally geodesic submanifolds.
Equivalence to Euclidean Convexity: In $\mathbb{R}^n$ with its usual metric, geodesic convexity coincides with usual convexity. On flat manifolds, local coordinate charts map g-convex sets to Euclidean convex sets (Cohn et al., 2023, Vishnoi, 2018).

3. Geodesic Convexity in Metric Spaces and Generalizations

The notion of geodesically convex subspaces applies to a broad array of metric spaces with geodesic structure, including spaces of nonpositive or nonnegative curvature and infinite-dimensional contexts.

Hadamard Spaces (CAT(0)): In a Hadamard space, geodesic convexity is especially robust due to unique geodesics between any pair of points. Closed, geodesically convex subsets inherit strong geometric and analytic properties, including contractibility and a well-behaved recession function at infinity (Hirai, 2022).
Wasserstein Spaces: In Wasserstein space $(\mathcal{P}_2(H), W_2)$ , a subset is geodesically convex if every pair of measures can be joined by a constant-speed Wasserstein geodesic lying within the set (Naldi et al., 2021).
Quotient Manifolds and Matrix Manifolds: For certain matrix manifolds, such as the quotient space arising in fixed-rank positive semidefinite matrix factorization, geodesic balls of explicit radius are geodesically convex, with radii determined by spectral or singular value properties (Luo et al., 2022). This underlies geodesic convexity-based optimization in matrix settings.

4. Functional, Differential, and Optimization Structure

Geodesically convex subspaces support an extensive functional and variational structure:

Convex Functions: A function $f: S \to \mathbb{R}$ is called geodesically convex if its restriction to every minimizing geodesic is convex in the classical sense, i.e.,

$f(\gamma(t)) \le (1-t)f(\gamma(0)) + t f(\gamma(1)), \quad t\in [0,1]$

(Vishnoi, 2018, Hirai, 2022). The Riemannian Hessian criterion characterizes geodesic convexity for $f$ .

Proximal Normal and Separation: Geodesically convex subsets in Riemannian manifolds admit a well-defined proximal normal cone, and local versions of supporting “hyperplanes” (hypersurfaces) can be constructed, providing manifold generalizations of the Euclidean separation theorem (Khajehpour et al., 2018).
Distance Functions and Nonsmooth Analysis: Under sufficient regularity of $S$ , the distance-to-set function $d_S(x)$ may itself be geodesically convex in a neighborhood, which can be established via Hessian conditions, second-order superjets, or support principles (Khajehpour et al., 2018).
Convex Optimization: Many classical optimization and algorithmic techniques generalize when restricted to geodesically convex subspaces, with global optimality guarantees preserved for geodesically convex objectives on geodesically convex sets (Goyal et al., 2019, Vishnoi, 2018).

5. Algorithmic and Methodological Implications

Geodesically convex subspaces serve as feasible regions for optimization and sampling algorithms in manifold and non-Euclidean settings:

Graph of Geodesically Convex Sets (GGCS): Motion planning in robot configuration spaces modeled as Riemannian manifolds relies on covering admissible regions with overlapping g-convex charts, forming a graph where discrete paths correspond to piecewise-geodesic global optimal trajectories (Cohn et al., 2023).
Optimization Reductions: On flat manifolds, geodesically convex charts can be mapped isometrically to Euclidean convex sets, allowing reduction of problems to mixed-integer convex programming (Cohn et al., 2023).
Sampling and Annealing: Algorithms such as the geodesic ball-walk, hit-and-run, and simulated annealing utilize geodesic convexity to ensure validity and tractability of Markov chains and convergence to global optima on manifolds with non-negative curvature (Goyal et al., 2019).

6. Specialized Settings and Structural Theorems

6.1 Metric Spaces and Injectivity

In the $\ell_\infty$ plane $(\mathbb{R}^2, d_\infty)$ , closed geodesically convex subsets are hyperconvex and coincide (up to isometry) with tight spans, the minimal injective hulls containing a given set (Kiliç et al., 2015).

6.2 Wasserstein Submanifolds

In Wasserstein spaces, submanifolds defined by linear or convex constraints (e.g., mean-variance spheres, coupling subspaces) may not be geodesically convex in the ambient space but inherit geodesic convexity for restricted functionals if evolution variational inequality (EVI) gradient flows of these functionals are invariant in the subspace (Chaintron et al., 19 Aug 2025). This principle provides a method for analyzing the convexity of entropy or energy functionals in complex submanifolds and leads to strengthened functional inequalities and improved concentration bounds.

6.3 Infinite-dimensional Function Spaces

In the space of Kähler potentials equipped with the Mabuchi-Semmes-Donaldson metric, the so-called $(S, \omega_0)$ -convexity is preserved along Mabuchi geodesics if the endpoints are strictly convex, guaranteeing non-degeneracy of the evolving Kähler metrics (Hu, 2022).

7. Structural and Geometric Insights

The structural behavior of geodesically convex subspaces is intimately tied to manifold curvature, topology, and geometry:

Curvature Implications: On manifolds of nonpositive curvature (Hadamard spaces), geodesic convexity is globally well-behaved due to the existence and uniqueness of minimizing geodesics (Hirai, 2022). In non-negative curvature, convexity of “inner parallel sets” and the behavior of the distance function retain favorable properties (Goyal et al., 2019).
Totally Geodesic Submanifolds: These are always geodesically convex and retain the property that geodesics in the submanifold are geodesics in the ambient manifold (Vishnoi, 2018).

Summary Table: Fundamental Properties of Geodesically Convex Subspaces

Aspect/Result	Setting	Reference
Unique minimizing geodesic	Riemannian manifold, CAT(0)	(Cohn et al., 2023, Hirai, 2022)
Intersection closure	General geodesic spaces	(Goyal et al., 2019)
Convex hull construction	General geodesic spaces	(Goyal et al., 2019)
Nonsmooth convexity criteria	Riemannian manifold	(Khajehpour et al., 2018)
GGCS planner methodology	Robotics/config manifolds	(Cohn et al., 2023)
EVI-invariance principle	Wasserstein, metric submanifolds	(Chaintron et al., 19 Aug 2025)

Geodesically convex subspaces provide the geometric substrate for transferring the extensive theoretical and algorithmic apparatus of convexity to a broad class of nonlinear and infinite-dimensional contexts. Their study is foundational for modern geometric optimization, analysis on manifolds, computational geometry, and numerous applications in robotics, data analysis, and mathematical physics.