Lipschitz Conic Structure

• Lipschitz Conic Structure is a framework where conic geometry is encoded by Lipschitz continuous functions, ensuring quantitative stability across optimization and metric analysis.
• It unifies diverse applications from convex conic reformulations and local sensitivity of KKT systems to Lipschitz normal embedding in singular metric geometry.
• The framework leverages cone-depth scalar functions and bilipschitz cone models to deliver practical error bounds and regularity conditions in complex variational settings.

Searching arXiv for recent and foundational papers on Lipschitz conic structure across optimization and metric geometry. In the papers considered here, “Lipschitz conic structure” appears in several distinct but related senses. In convex conic optimization, it denotes a reformulation in which a conic constraint is replaced by a concave globally Lipschitz scalar objective defined relative to a strictly feasible interior direction (Renegar, 2015). In metric geometry, it refers to conic or asymptotically conic singular behavior strong enough to force equivalence of inner and outer metrics, i.e. Lipschitz normal embedding (LNE), for singular sub-manifolds and their compactifications (Costa et al., 2024, Costa et al., 2023). In conic sensitivity analysis and nonpolyhedral variational analysis, it appears as local Lipschitz continuity of optimal value functions and locally upper Lipschitz stability of perturbed KKT systems under strict regularity assumptions (Luan et al., 2020, Liu et al., 2015). A plausible umbrella description is that conic geometry is encoded by Lipschitz data: a scalar depth function, a bilipschitz cone model, or a Lipschitzian solution map.

1. Principal meanings and recurrent structures

Across these works, the conic ingredient is always explicit: a proper closed convex cone in optimization, a conic or asymptotically conic metric in singular geometry, or a nonpolyhedral matrix cone in KKT analysis. The Lipschitz ingredient is likewise explicit: global Lipschitz continuity of a cone-defined objective, local Lipschitz continuity of a value function, locally upper Lipschitz behavior of a multifunction, or bilipschitz comparability between inner and outer metrics.

Domain	Conic object	Lipschitz manifestation
Convex conic optimization	proper closed convex cone with distinguished direction $e$	$\lambda_{\min}$ is concave and globally Lipschitz
Parametric and KKT stability	closed convex cone, dual cone, Ky Fan $k$-norm matrix cone	value functions are locally Lipschitz; KKT maps are locally upper Lipschitz
Singular metric geometry	conic metrics, asymptotically conic metrics, conic singular sub-manifolds	inner and outer metrics are equivalent; sets are LNE

The common structural pattern is that the cone is not discarded. Rather, it is compressed into a quantitatively controlled object. In Renegar’s conic reformulation, all cone geometry is absorbed into a single scalar function $x \mapsto \lambda_{\min}(x)$. In globally conic singular geometry, the local or asymptotic cone model yields explicit distance comparison estimates. In Ky Fan $k$-norm optimization, projection derivatives, normal cones, and second-order tangent geometry of the cone determine local error bounds and isolated calmness. This suggests that “Lipschitz conic structure” is best understood as a family of mechanisms by which conic geometry becomes quantitatively stable.

2. Cone geometry encoded by a globally Lipschitz objective

The most explicit use of the phrase occurs in the conic optimization framework of “A Framework for Applying Subgradient Methods to Conic Optimization Problems” (Renegar, 2015). The starting point is the conic program

$$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$

where $\mathcal K \subset \mathcal E$ is a proper, closed, convex cone with nonempty interior, $\mathrm{Affine}$ is an affine subspace, and $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$ is a fixed strictly feasible point, called the distinguished direction. The central construction is

$$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$

Geometrically, $\lambda_{\min}$0 is the amount one can move from $\lambda_{\min}$1 in the $\lambda_{\min}$2 direction before hitting the boundary of the cone, so that $\lambda_{\min}$3. The function behaves like a generalized minimum eigenvalue: for $\lambda_{\min}$4 and $\lambda_{\min}$5, it is the smallest eigenvalue; for $\lambda_{\min}$6 and $\lambda_{\min}$7 coordinatewise,

$\lambda_{\min}$8

This scalar function exactly represents the cone: $\lambda_{\min}$9 It also satisfies the affine-equivariance identity

$k$0

To express Lipschitz continuity, the paper defines

$k$1

and the associated seminorm

$k$2

The basic structural proposition is that $k$3 is concave and globally Lipschitz: $k$4 This is the paper’s precise “Lipschitz conic structure”: the cone can be represented by a globally concave Lipschitz function.

The reformulation fixes $k$5 and considers the affine slice

$k$6

The original conic problem becomes

$k$7

The conic constraint is no longer imposed explicitly. Instead, one uses the radial boundary-hitting map

$k$8

defined whenever $k$9. This map is not Euclidean projection; it is a radial map from the distinguished interior point $x \mapsto \lambda_{\min}(x)$0 to $x \mapsto \lambda_{\min}(x)$1. The central theorem states that if $x \mapsto \lambda_{\min}(x)$2 solves the Lipschitz reformulation, then $x \mapsto \lambda_{\min}(x)$3 is optimal for $x \mapsto \lambda_{\min}(x)$4, and conversely every optimal conic solution yields an optimal point of the reformulated problem. The same paper then shows that “virtually any subgradient method” can be applied to the equality-constrained Lipschitz problem, with no projection onto $x \mapsto \lambda_{\min}(x)$5, only orthogonal projection onto a fixed affine subspace, and with feasible conic points recoverable at every iteration through $x \mapsto \lambda_{\min}(x)$6. A notable feature is the relative-accuracy guarantee

$x \mapsto \lambda_{\min}(x)$7

which is converted exactly into an additive error bound for maximizing $x \mapsto \lambda_{\min}(x)$8.

3. Sensitivity and local Lipschitzian stability in conic optimization

A second optimization meaning of Lipschitz conic structure concerns perturbation theory. In “Two Optimal Value Functions in Parametric Conic Linear Programming,” the conic linear program

$x \mapsto \lambda_{\min}(x)$9

is studied through two marginal functions: $k$0 If the primal problem is strictly feasible and $k$1, then there exists $k$2 such that $k$3 is finite for every $k$4, and $k$5 is Lipschitz on $k$6. If the primal problem is strictly feasible and has a solution, then

$k$7

For objective perturbations, if both primal and dual are strictly feasible, then

$k$8

The paper emphasizes that the right-hand-side perturbation problem is better behaved from the local Lipschitz viewpoint than objective perturbation in the full ambient space, and that polyhedral and nonpolyhedral cones behave differently in finite-$k$9 sensitivity formulas (Luan et al., 2020).

A more local, nonpolyhedral stability result appears in “Locally upper Lipschitz of the perturbed KKT system of Ky Fan $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$0-norm matrix conic optimization problems” (Liu et al., 2015). There the constraint cone is

$$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$1

with $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$2. Under the second-order sufficient condition and strict Robinson CQ, two perturbed KKT solution mappings are locally upper Lipschitz at the origin for a reference KKT point. The residual map is

$$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$3

and local upper Lipschitz continuity of the corresponding solution map implies the local error bound

$$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$4

in a neighborhood of the reference KKT point. The proof is driven by projection derivatives, the graphical derivative of the normal cone mapping, critical cone geometry, second-order tangent curvature, and explicit SVD-based formulas for the Ky Fan $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$5-norm cone. This suggests that, in optimization, Lipschitz conic structure can refer either to a globally Lipschitz cone-depth objective or to local Lipschitzian regularity of solution mappings generated by conic variational geometry.

4. Metric conic singularities, asymptotically conic ends, and LNE

In metric geometry, the phrase is used differently. “Remarks on Lipschitz geometry on globally conic singular manifolds” develops a framework for globally conic singular manifolds and shows how conic singularities and asymptotically conic ends force Lipschitz normal embedding (Costa et al., 2024). Given a subset $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$6 of a pseudo-metric space $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$7, the outer metric is $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$8, while the inner metric $$\left. \begin{array}{rl} \inf & c \cdot x \ \textrm{s.t.} & x \in \mathrm{Affine} \ & x \in \mathcal K \end{array} \right\} CP$$9 is the infimum of lengths of arcs in $\mathcal K \subset \mathcal E$0 joining two points. The subset is LNE when

$\mathcal K \subset \mathcal E$1

The local models are singular Riemannian metrics of conic type. In a collar neighborhood of the boundary, a conic metric can be written in model form

$\mathcal K \subset \mathcal E$2

while an asymptotically conic metric has the form

$\mathcal K \subset \mathcal E$3

A conic singular manifold is a complete length space with finitely many isolated singular points, each modeled by collapsing the boundary of a manifold with conic metric. A globally conic singular manifold has, in addition, noncompact ends covered by charts isometric to manifolds with asymptotically conic metrics. For a closed subset $\mathcal K \subset \mathcal E$4, the relevant transversality condition is that its strict transform is a $\mathcal K \subset \mathcal E$5-submanifold in conic or asymptotic charts; such an $\mathcal K \subset \mathcal E$6 is then a conic singular sub-manifold or a globally conic singular sub-manifold.

The metric control is explicit. For a conic metric,

$\mathcal K \subset \mathcal E$7

and for an asymptotically conic metric,

$\mathcal K \subset \mathcal E$8

where

$\mathcal K \subset \mathcal E$9

The conic inversion $\mathrm{Affine}$0 then yields

$\mathrm{Affine}$1

and globally, for a conic completion $\mathrm{Affine}$2,

$\mathrm{Affine}$3

These estimates drive the main theorems. Theorem 4.5 states that a closed conic singular sub-manifold germ in a conic singular manifold is locally LNE at the singular point. Theorem 4.10 states that if $\mathrm{Affine}$4 is a globally conic singular sub-manifold of a globally conic singular manifold $\mathrm{Affine}$5, then each connected component of $\mathrm{Affine}$6 is LNE in $\mathrm{Affine}$7. The argument uses collar product structure for the strict transform, local cylinder models over links, and a completion/inversion procedure that turns asymptotically conic behavior at infinity into ordinary conic behavior on a compact completion.

5. Global conic singular sub-manifolds and algebraic consequences

A closely related but globally oriented formulation appears in “Global Lipschitz geometry of conic singular sub-manifolds with applications to algebraic sets” (Costa et al., 2023). Here the ambient space is a smooth Riemannian manifold $\mathrm{Affine}$8, with special emphasis on closed subsets of $\mathrm{Affine}$9. A point $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$0 of a closed subset $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$1 is a conic point if, for some $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$2, the strict transform

$$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$3

under spherical blow-up is a closed subset and a $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$4-sub-manifold of the blow-up space. A conic singular sub-manifold is a closed subset all of whose points are conic, with finite singular locus. The paper emphasizes that any conic singular point is isolated.

The local geometric engine is the collar diffeomorphism of the strict transform. If $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$5 is a conic point of $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$6, then there exist $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$7 and a link-preserving diffeomorphism

$$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$8

whose restriction to the front face is the identity. After blow-down, this yields an outer bi-Lipschitz homeomorphism from a truncated tangent cone $$e \in \mathrm{Affine} \cap \mathrm{int}(\mathcal K)$$9 onto $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$0. This is the paper’s most direct bilipschitz cone model.

The main global conclusions are Theorem 4.2 and Theorem 4.13. Theorem 4.2 states that any connected, compact, conic singular sub-manifold of a smooth Riemannian manifold is LNE. Theorem 4.13 states that a connected conic singular sub-manifold $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$1 is LNE. In the Euclidean case, however, one needs control at infinity. The paper therefore defines “conic at $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$2” using spherical compactification and proves Proposition 4.12: a closed subset $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$3 is a conic singular sub-manifold if and only if its closure in the one point compactification $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$4 is a conic singular sub-manifold. The inversion map

$$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$5

is used to identify conic behavior at infinity with ordinary conic behavior near the origin. The necessity of the asymptotic hypothesis is illustrated by the parabola $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$6, which is not LNE.

The same framework yields algebraic applications. The paper states that connected components of generic affine real and complex algebraic sets are conic at infinity, hence LNE. For smooth projective subvarieties, generic affine traces $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$7 are nonsingular and conic at infinity; for generic affine hypersurfaces and generic complete intersections, the zero set is nonsingular and conic at $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$8, so each connected component is LNE. This gives a global mechanism: $$\lambda_{\min}(x) := \inf \{ \lambda : x - \lambda e \notin \mathcal K \}.$$9

6. Obstructions, abnormality, and distinguished conic models

Not every Lipschitz-geometric singularity admits a simple conic description. “Lipschitz geometry and combinatorics of abnormal surface germs” studies definable surface germs through outer metric geometry and shows that the natural building blocks are often Hölder triangles rather than literal cones (Gabrielov et al., 2021). For $\lambda_{\min}$00, the standard $\lambda_{\min}$01-Hölder triangle is

$\lambda_{\min}$02

A germ is LNE when its inner and outer metrics are equivalent, and a Hölder triangle is LNE iff

$\lambda_{\min}$03

The paper’s central conclusion is that any definable Hölder triangle is either Lipschitz normally embedded or contains some abnormal arcs. Abnormal arcs are organized into finitely many abnormal zones, and Theorem 6.10 states that $\lambda_{\min}$04 is the union of finitely many maximal normal zones and finitely many maximal abnormal zones, each maximal abnormal zone being closed perfect and of one of three types: a circular snake, the set of generic arcs in a $\lambda_{\min}$05-snake, or a $\lambda_{\min}$06-complete zone contained in a non-snake $\lambda_{\min}$07-bubble for arbitrarily close $\lambda_{\min}$08. The paper therefore does not give a universal outer conic structure theorem; instead, it shows that finite Hölder-LNE decomposition plus finitely many abnormal zones is the correct structural replacement in this definable surface setting.

A different sort of distinguished conic model appears in “Minimal Piecewise Linear Cones in $\lambda_{\min}$09” (Valfells, 2022). There the question is not bilipschitz equivalence of singularities but which piecewise linear three-dimensional cones in $\lambda_{\min}$10 are mass minimizing with respect to Lipschitz maps in the Almgren–Taylor sense. The main theorem states that there are precisely five such cones, up to rotation: $\lambda_{\min}$11 the cone over the $\lambda_{\min}$12-skeleton of the regular $\lambda_{\min}$13-simplex, and the cone over the $\lambda_{\min}$14-skeleton of the hypercube. This is a classification of Lipschitz-minimizing conic models in a variational sense. It shows that not every combinatorially admissible Plateau-balanced polyhedral cone is minimal once one allows Lipschitz competitors.

Taken together, these results delimit the scope of the subject. In some settings, conic geometry produces exact Lipschitz objectives, bilipschitz cone charts, or globally LNE spaces. In others, such as definable surface germs with abnormal zones, a naive global conic picture fails and must be replaced by a finer Hölder-combinatorial structure. A plausible overall conclusion is that Lipschitz conic structure is not a single theorem but a collection of rigidity principles, each expressing how conic geometry can impose quantitative Lipschitz control on optimization problems, singular metrics, or variational singular models.