Directional Second-Order Tangent Sets

• Directional second-order tangent sets are asymptotic descriptors that capture curvature and second-order behavior in analytic and optimization problems.
• They distinguish between a geometric set defined via analytic arcs and an algebraic set defined through jet conditions on power series.
• These constructions underpin sharp second-order optimality conditions and inform methods for nonconvex, matrix, and cone-constrained optimization.

A directional second-order tangent set describes the asymptotic second-order structure of a set $X$ at a reference point in a prescribed direction. This construction is pivotal in variational analysis, optimization, and geometric study of real or complex analytic sets, matrix varieties, and nonconvex constraint sets. For a closed set $X\subseteq \mathbb{K}^n$ (with $\mathbb{K}=\mathbb{R}$ or $\mathbb{C}$), and $u\in T_0X$ a nonzero tangent direction, one distinguishes between the geometric second-order tangent set $T^2_{0,u}X$, defined by the velocities of analytic arcs with prescribed first-order direction, and the algebraic second-order tangent set $T^{2,a}_{0,u}X$, defined via jet conditions on the initial forms of the defining equations. These tangent sets encode curvature information crucial for characterizing fine optimality in set-constrained nonconvex optimization problems, second-order stationarity in rank and cone-constrained problems, and regularity properties for tangent intersection and chain rules. The analysis leverages Taylor expansions, projection operator formulas, and subregularity-based calculus to produce explicit geometric and algebraic conditions.

1. Definitions and Fundamental Constructions

Let $X\subseteq\mathbb{K}^n$ be a closed analytic set germ at $0$, i.e., defined near $0$ as the zero locus of finitely many convergent power series. The contingent (geometric) tangent cone is

$T_0X := \liminf_{t\to 0} (X - 0)/t$

and the algebraic tangent cone (initial-form) is

$$C_0 X = \{ u \in \mathbb{K}^n \mid f_m(u) = 0\ \text{for all}\ f\in\mathcal{I}(X,0)\}$$

where $f(x) = f_m(x) + f_{m+1}(x) + \dots$ is the Taylor expansion with $\operatorname{ord} f = m$.

The geometric second-order tangent set in direction $u \in T_0X$ is

$$T^2_{0,u} X = \{ w \in \mathbb{K}^n\mid \exists\ \gamma : (0,\epsilon) \to X,\ \gamma(t)=t u + \tfrac{t^2}{2} w + o(t^2)\}$$

The algebraic second-order tangent set is

$$T^{2,a}_{0,u} X = \{ w \in \mathbb{K}^n \mid \text{for all}\ f \in \mathcal{I}(X,0),\ f_m(u)=0,\ \tfrac{1}{2} \langle\nabla f_m(u), w\rangle + f_{m+1}(u)=0\}$$

Similar constructions apply to general closed sets with contingent, Clarke, and Bouligand cones, and in matrix/tensor varieties, to semidefinite and determinantal structures (Trinh, 15 Jan 2026, Yang et al., 27 Nov 2025, Liu et al., 2019, Chen et al., 2019).

2. Comparison of Geometric and Algebraic Second-Order Tangent Sets

For analytic sets, the inclusion

$T^2_{0,u} X \subseteq T^{2,a}_{0,u} X$

holds always, with equality precisely when every algebraic jet arises from an analytic arc (i.e., surjectivity of the second-jet map $\Phi_u: J^2(X)_{0,u} \to J^2(C_0 X)_u$) (Trinh, 15 Jan 2026). Explicit examples (e.g. $X = \{z^2 - x^3 y^3 = 0\} \subseteq \mathbb{K}^3$, $u \in \mathbb{K}^2 \times \{0\}$) demonstrate strict inclusion unless specific regularity holds.

Surjectivity of $\Phi_u$—and hence $T^2_{0,u} X = T^{2,a}_{0,u} X$—occurs in:

• Smooth analytic germs: $X = C_0 X = T_0 X$, affine second-order expansions cover all cases.
• Homogeneous analytic cones: all jets lift.
• Hypersurfaces with $\nabla f_m(u) \ne 0$: implicit function theorem applies.
• Nondegenerate complete intersections: multi-equation implicit function theorem yields lifting.

For matrix varieties, e.g. $M_r = \{ X \mid \operatorname{rank}(X) \le r \}$, $T^2_{M_r}(A;\eta)$ can be written explicitly in terms of singular value derivatives and curvature correction terms (Yang et al., 27 Nov 2025).

3. Calculus Rules and Variational Properties

Under metric subregularity, explicit chain and intersection rules for directional tangent sets are available (Durea et al., 2011, Gfrerer et al., 2019): $$T^2_F(\bar{x};d) = \left\{ w \mid g'(\bar{x})w + \tfrac{1}{2} g''(\bar{x})(d,d) \in T^2_A(g(\bar{x}); g'(\bar{x}) d) \right\}$$ where $F = g^{-1}(A)$ and $g$ is $C^2$; similar formulas hold for intersections, sums, and chain compositions, with suitable adjustment for regularity and convexity. Directional regular (Clarke) tangent cones and normal cones play a central role in handling nonconvex settings (Gfrerer et al., 2019).

4. Applications in Optimization: Necessary and Sufficient Conditions

Directional second-order tangent sets underpin sharp second-order optimality conditions. For $C^2$ optimization on analytic sets (Trinh, 15 Jan 2026):

• First-order: $\langle\nabla f(0), u\rangle \ge 0$ for $u \in T_0 X$.
• Second-order: For $u$ critical ($\langle\nabla f(0),u\rangle=0$),

$$\inf_{w \in T^2_{0,u} X} [\langle u, \nabla^2 f(0) u\rangle + \langle \nabla f(0), w\rangle] \ge 0$$

Sufficiency is characterized under parabolic regularity and strict positivity over $T^2_{0,u} X$.

In semidefinite and second-order cone complementarity problems, the outer second-order tangent set is defined via second-order directional derivatives of the projection operator (Liu et al., 2019, Chen et al., 2019): $$T^2_{\mathcal{Q}}((X,Y);(F,G)) = \{ (S,T) \mid \Pi^{[2]}_{\mathcal{Q}}(X+Y;F+G,S+T) = S \}$$ and necessary/sufficient conditions involve support functionals of these sets.

For non-convex set-constrained problems, necessary conditions take the form

$$\nabla^2_{xx} L(\bar{x}, \lambda)(d,d) - \sigma_{T^2_A(g(\bar{x}); g'(\bar{x})d)}(\lambda) \ge 0$$

with $\sigma$ a generalized support function adapted to nonconvex $T^2$ sets (Gfrerer et al., 2019, Ouyang et al., 2024), and sufficient conditions involve strict inequalities for all nonzero critical directions.

5. Special Structures: Matrix Varieties and Complementarity Sets

Directional second-order tangent sets have been precisely described for determinantal varieties, tensor varieties, and semidefinite sets (Yang et al., 27 Nov 2025). For $M_r = \{ X : \operatorname{rank}(X)\le r\}$, with compact SVD $A = U\Sigma V^T$, and tangent direction $\eta$: $$T^2_{M_r}(A;\eta) = \left\{ \zeta = 2\eta A^\dagger \eta + [U^+ U_{\eta\perp}] \begin{pmatrix} W_1 & W_2 \ W_3 & J_2 \end{pmatrix} [V^+ V_{\eta\perp}]^T \mid \operatorname{rank}(J_2) \le r - \ell \right\}$$ where $A^\dagger$ is the Moore–Penrose pseudoinverse and $J_2$ encodes further rank adjustments. This refinement captures curvature in low-rank optimization, and verification of second-order optimality is shown to be NP-hard.

For SOC complementarity sets $\Omega = \{(u,v)\in K\times K : u^Tv=0\}$, the second-order tangent is

$$T^2_{\Omega}((x,y);(d,w)) = \{ (p,q) \mid \Pi_{K}''(x-y;d-w, p-q) = p \}$$

and explicit block-wise formulas are established for all cases (Chen et al., 2019).

6. Impact on No-Gap and Quadratic Growth Conditions

Directional second-order tangent sets enable "no-gap" second-order optimality criteria in Banach and metric spaces (Christof et al., 2017). The directional curvature functional

$$Q_C^{x, \varphi}(d) = \inf_{r \in T^2_C(x,d)} \langle \varphi, r \rangle$$

yields full equivalence between positivity of $Q_C + J''(\bar{x})h^2$ on the critical cone and quadratic growth at minimizers. In bang-bang control, explicit representations of $Q_C$ settle optimality in singular settings previously not covered by classical Legendre conditions.

In general (including nonconvex and infinite-dimensional problems), sufficient conditions for quadratic growth and necessary conditions for local optimality rely on evaluation of the support functions of directional second-order tangent sets (classical or generalized), and may utilize directional versions of Robinson’s constraint qualification or metric subregularity (Gfrerer et al., 2019, Ouyang et al., 2024, Durea et al., 2011).

7. Key Properties and Limitations

• In general, $T^2_{0,u} X$ may be strictly smaller than $T^{2,a}_{0,u} X$ unless special regularity holds; in nonconvex settings, $T^2$ may not be a cone or even nonempty.
• The outer second-order tangent set need not be convex for nonconvex $X$ (Ouyang et al., 2024, Gfrerer et al., 2019).
• Support functions for nonconvex $T^2$ sets must be lower generalized, not the usual convex-analytic form—this is crucial for sharpness of optimality criteria.
• Under metric subregularity, explicit calculus rules are available for intersections, sums, and implicit maps, circumventing compactness assumptions (Durea et al., 2011).
• In matrix and cone-constrained settings, curvature terms appear explicitly in second-order conditions, distinguishing them fundamentally from polyhedral theory (Yang et al., 27 Nov 2025, Chen et al., 2019).

In summary, directional second-order tangent sets deliver a unified framework for rigorous second-order analysis in constrained optimization, real and complex analytic geometry, and matrix theory, capturing subtle curvature effects and enabling the formulation of necessary and sufficient conditions that extend beyond convex and polyhedral paradigms.