Papers
Topics
Authors
Recent
2000 character limit reached

Tangent Cone in Geometry and Optimization

Updated 16 November 2025
  • Tangent Cone is a mathematical construct that models the limit of secant directions at a point, capturing its first-order infinitesimal structure.
  • It is applied across algebraic geometry, analysis, and optimization to analyze singularities, stratifications, and regularity through varied frameworks like Bouligand and Gromov–Hausdorff limits.
  • Modern research employs tangent cone techniques, including Gröbner bases and block decompositions, to derive precise local and asymptotic characterizations in both bounded and unbounded settings.

A tangent cone, in its classical and modern guises, is the principal local invariant describing the first-order infinitesimal geometry of a subset, singularity, current, variety, or solution set at a point or at infinity. Across algebraic geometry, analysis, geometric measure theory, and optimization, tangent cones generalize the notion of the tangent space, capturing all limiting secant directions and the structure of singular loci. Tangent cones arise as Bouligand or contingent cones, algebraic tangent cones, Gromov–Hausdorff limits, and are foundational in the paper of singularities, stratifications, and regularity theories.

1. Bouligand (Contingent) Tangent Cone: Definitions and Properties

The Bouligand (contingent) tangent cone TAMT_{A}M to an algebraic variety MRNM \subset \mathbb{R}^N at a point AMA \in M is defined by

T(A)M:={XRN: sequences AnM, anR, AnA, an(AnA)X}.T_{(A)}M := \{ X \in \mathbb{R}^N : \exists\ \textrm{sequences}\ A_n \in M,\ a_n \in \mathbb{R},\ A_n \to A,\ a_n(A_n - A)\to X \}.

Equivalently, XT(A)MX \in T_{(A)}M iff there exists an analytic arc y(t)My(t) \in M with y(0)=Ay(0)=A and y(0)=Xy'(0)=X (Kutschan, 2017). This construction is central for real and complex analytic sets, providing a precise language for tangent directions at possibly singular points.

In algebraic geometry, the tangent cone at a point is frequently also described by the zero locus of the lowest-degree (for local tangent cones) or highest-degree (for tangent cones at infinity) homogeneous parts of the defining polynomials. For instance, for the real generic determinantal variety Dr={XRm×n  rankXr}\mathcal{D}_r = \{ X \in \mathbb{R}^{m \times n}\ |\ \mathrm{rank} X \leq r \}, the tangent cone at a rank-kk point admits multiple equivalent characterizations: as limit directions of secants, in terms of block-wise rank constraints, or through algebraic tangent cone equations (Olikier et al., 15 Apr 2025).

2. Tangent Cones to Tensor Train (TT) and Hierarchical Formats

For the TT variety—the set of tensors admitting tensor train factorizations of rank (k1,,kd1)\leq (k_1,\ldots,k_{d-1})—the tangent cone at any point admits a block–TT decomposition generalizing the matrix case: X=(A1,U1,X1)(Z2,A2,U2,X2)(Zd1,Ad1,Ud1,Xd1)(Ad,Vd),X = (A_1, U_1, X_1) \circ (Z_2, A_2, U_2, X_2) \circ \cdots \circ (Z_{d-1}, A_{d-1}, U_{d-1}, X_{d-1}) \circ (A_d,V_d), where the blocks Ui,Xi,Zi,ViU_i,X_i,Z_i,V_i satisfy orthogonality constraints and encode perturbations at each core. Every tangent direction arises this way, yielding both a polynomial parametrization and an implicit description as the intersection of minor equations for all (A[1i]×[i+1d])(A_{[1\dots i]\times[i+1\dots d]}) matricizations via rank constraints: rank[(IP1i)X[1i]×[i+1d](IPi+1d)]sifor all i.\mathrm{rank}\,[ (I - P_{1\dots i}) X_{[1\dots i]\times[i+1\dots d]} (I - P_{i+1\dots d}) ] \leq s_i\quad \text{for all}\ i. This framework generalizes seamlessly to hierarchical Tucker (HT) and other binary tree formats by inductively splitting tangents along the tree (Kutschan, 2017).

3. Tangent Cones at Infinity: Algebraic and Geometric Perspectives

For unbounded pure kk-dimensional algebraic sets XCmX \subset \mathbb{C}^m, several tangent cone definitions at infinity (Whitney's C3,,C4,,C5,C_{3,\infty}, C_{4,\infty}, C_{5,\infty}) capture infinitesimal, vector-tangent, and chordal secant directions, respectively:

  • C3,(X):C_{3,\infty}(X): limit directions obtained by scaling points pjXp_j \in X, pj|p_j|\to\infty.
  • C4,(X):C_{4,\infty}(X): limit directions of tangent vectors at regular points tending to infinity.
  • C5,(X):C_{5,\infty}(X): limits of scaled chords between divergent sequences in XX.

These cones satisfy C3,C4,C5,C_{3,\infty} \subset C_{4,\infty} \subset C_{5,\infty}, with precise dimension bounds. The dimension theorem asserts that if C5,(X)C_{5,\infty}(X) has pure dimension kk, then XX is affine linear (Dias et al., 29 Apr 2024). The algebraic tangent cone at infinity is computed by taking the top-degree homogeneous parts of each generator of the ideal I(X)I(X), and geometric tangent cones at infinity coincide with algebraic ones for complex varieties (Lê et al., 2016). Gröbner basis algorithms explicitly construct these tangent cones.

4. Analytic and Metric Tangent Cone Constructions

In metric geometry and geometric analysis, tangent cones are captured via rescaling procedures:

  • For complete Riemannian manifolds (Mn,g)(M^n,g), the Gromov-Hausdorff tangent cone at pp arises as the pointed limit (M,ri2g,p)(Y,y)(M, r_i^{-2}g, p) \to (Y, y) as ri0r_i\to0; cones at infinity use RiR_i\to\infty.
  • For Ricci-flat manifolds with Euclidean volume growth, uniqueness of the tangent cone at infinity holds when the limiting cross-section is a smooth Einstein manifold (Colding et al., 2012). Quantitative convergence rates are derived via monotonic analytic functionals and Lojasiewicz–Simon inequalities.

For area-minimizing currents with boundary and arbitrary multiplicity, the tangent cone at minimum-density boundary points is unique and consists of a finite sum of half-planes ("open-book" cone), with power-rate convergence (Fleschler, 7 Oct 2024). These results underpin boundary regularity theories generalizing Allard’s classical framework.

5. Stratifications, Singularities, and Algebraic Current Cones

In real closed valued fields, tangent cones Cp(X)\mathcal{C}_p(X) for definable sets XX at points pp are defined via valuation parameters. They support a robust theory mimicking classical secant/ray definitions, and crucially t-stratifications (definable partitions equipped with “translatability”) induce t-stratifications on the tangent cones. By ultrapower transfer, archimedean counterparts yield Whitney stratifications of tangent cones for semi-algebraic sets (Ramírez, 2015). This ensures geometric control is maintained under passage to tangent directions.

For complex analytic and plurisubharmonic currents, tangent cones are built via either weak limits of rescaled currents or via blow-ups to the exceptional divisor. The existence of the tangent cone is characterized via growth rates of Lelong functions and associated integrability conditions (Ghiloufi et al., 2011). For positive-(1,1) De Rham currents in almost complex manifolds, tangent cones are unique under density non-jump hypotheses, with proofs leveraging pseudo-holomorphic blow-up techniques that generalize classical Lelong–Siu approaches (Bellettini, 2011).

6. Singular Inclusions and Optimization: Higher-Order Tangent Cone Theory

In variational analysis and nonlinear programming, tangent cones to solution sets of generalized equations F(x)=f(x)+NC(x)F(x) = f(x) + N_C(x) are characterized in both regular and singular settings. When f(x0)f'(x_0) fails to be surjective (the singular case), higher order derivatives are invoked. Specifically, under ppth-order degeneracy, the tangent cone directions hh are those for which the ppth derivative, via an auxiliary mapping LhL_h, balances the normal cone at x0+hx_0 + h and satisfies a strong metric regularity condition: 0(p1)!f(p)(x0)[h]p+NC(x0+h),0 \in (p-1)!f^{(p)}(x_0)[h]^p + N_C(x_0 + h), with associated continuity assumptions in multivalued mappings. This broadens classical Lyusternik–Robinson results and applies to singular complementarity and programming systems where standard tangent cone descriptions would fail (Prusinska et al., 2018).

7. Tangent Cones in Local and Algebraic Ring Settings

In local commutative algebra, for (S,n)(S, \mathfrak{n}) a regular local ring and In2I \subset \mathfrak{n}^2, the tangent cone of a local ring R=S/IR = S/I at its maximal ideal is defined through the leading ideal II^*—the homogeneous part generated by initial forms of II in the associated graded ring. Bounds on the minimal number of generators of II^* are determined sharply via discrete difference operators on the Hilbert function of RR. In complete intersection cases, the Hilbert function uniquely determines the Betti numbers of II^*, while for three generators sharp lower and upper bounds are deduced via cancellation techniques (Mandal et al., 2014). The Hilbert–Burch framework controls these resolutions in codimension-two, and extensions to higher Cohen-Macaulay tangent cones proceed via regular sequences.


In all domains, tangent cones serve as the bridge between infinitesimal and global structures, facilitate regularity theories, and encode singularity types, degree-theoretic information, and stratification properties. Modern research has extended their computation, uniqueness, and correspondence with algebraic and optimization problems, and they are a central object of current paper across pure and applied mathematics.

Forward Email Streamline Icon: https://streamlinehq.com

Follow Topic

Get notified by email when new papers are published related to Tangent Cone.