🔢 mathematics

On directional second-order tangent sets of analytic sets and applications in optimization

This paper investigates the relationship between geometric and algebraic directional second-order tangent sets for analytic sets, establishing conditions under which they coincide and demonstrating how this equality enables the derivation of explicitly computable second-order optimality conditions for optimization problems on such sets.

Original authors: Le Cong Trinh

Published 2026-04-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Le Cong Trinh

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to navigate a very strange, bumpy landscape. Your goal is to find the lowest point (the "optimal" spot) in this terrain. To do this, you need to understand the shape of the ground around you.

This paper is about a specific type of terrain: Analytic Sets. In math terms, these are shapes defined by very smooth, predictable equations (like polynomials). But unlike a simple hill or a flat plane, these shapes can have sharp corners, self-intersections, or weird "singularities" where the rules of smooth geometry break down.

Here is the breakdown of the paper's story, using a simple analogy.

1. The Two Maps: The "Real" Map vs. The "Algebraic" Map

When you stand at a specific point on this bumpy terrain (let's call it the origin, 0) and look in a specific direction (let's call it direction $u$ ), you want to know: "If I take a tiny step forward, how does the ground curve?"

To answer this, mathematicians use two different "maps" or tools:

The Geometric Map (The Real Deal): This map is built by actually walking the path. It asks: "Can I draw a smooth, continuous curve starting from here, going in direction $u$ , that stays exactly on the terrain?" If I can, the curve tells me exactly how the ground bends (the second-order information). This is the Geometric Tangent Set. It represents what is physically possible to do on the terrain.
The Algebraic Map (The Calculator): This map is built by looking at the blueprint (the equations) of the terrain. It uses a calculator to predict how the ground should bend based on the numbers in the equations, without actually checking if a path exists. This is the Algebraic Tangent Set. It represents what the equations say is possible.

The Problem: Usually, these two maps agree. But sometimes, the Algebraic Map lies. It might say, "Hey, you can curve this way!" but when you try to walk that path, you fall off the cliff because the terrain doesn't actually exist there. The Algebraic Map includes "ghost paths" that look valid on paper but are impossible to walk.

2. The Main Discovery: The "Ghost Path" Gap

The author, Le Cong Trinh, proves two big things:

The Inclusion: The "Real" paths (Geometric) are always a subset of the "Predicted" paths (Algebraic). You can never walk a path that the equations say is impossible.
The Gap: Sometimes, the Algebraic Map is too generous. It includes paths that are mathematically allowed by the equations but geometrically impossible to walk.
- Analogy: Imagine a maze drawn on paper. The Algebraic Map says, "You can turn left here." But the Geometric Map says, "No, there's a wall there." The paper shows specific examples where the "wall" exists, proving the two maps are different.

3. The Solution: When Can We Trust the Calculator?

If the Algebraic Map is wrong, we can't use it to solve optimization problems (finding the lowest point) because we might think a path is safe when it's not.

The paper asks: "When do the two maps match?"

The author finds that for many important types of terrain, the maps do match. If the terrain is:

Smooth (no sharp corners),
A cone (shapes that look the same at any scale),
Or a "nice" intersection of surfaces (where the surfaces cross cleanly),

...then the "Ghost Paths" disappear. The Algebraic Map becomes 100% accurate. In these cases, you don't need to walk the path to know how it curves; you can just trust the equations.

4. The Application: Finding the Best Spot

Why does this matter? Because in the real world, we use math to optimize everything: designing bridges, training AI, or managing finances. We want to find the "best" solution.

To find the best solution, we need to check if a point is a local minimum (the bottom of a valley).

The Hard Way: We have to check the "Real" paths (Geometric). This is computationally expensive and hard to calculate.
The Easy Way: We use the "Calculator" (Algebraic). This is fast and easy.

The Paper's Conclusion:
If your problem involves one of the "nice" terrains mentioned above (smooth, cones, clean intersections), you can safely use the Easy Way. You can replace the difficult geometric check with the simple algebraic formula.

However, the paper also gives a Warning Example. If you are dealing with a weird, singular terrain where the two maps don't match, and you blindly use the Algebraic Map, you might conclude that a point is not a minimum (because the "Ghost Paths" make the math look bad), when in reality, it is the minimum. You would throw away the best solution because you trusted the wrong map.

Summary in One Sentence

This paper tells us when we can trust a mathematical shortcut (the algebraic equations) to tell us how a complex shape curves, and warns us that if we use that shortcut on the wrong kind of shape, we might miss the best solution entirely.

1. Problem Statement

The paper addresses a fundamental gap in variational analysis and optimization concerning directional second-order tangent sets for analytic sets (both real and complex).

Context: In optimization on singular sets, first-order tangent cones ( $T_0X$ ) describe feasible directions but fail to capture second-order behavior necessary for local optimality conditions.
The Conflict: Two distinct notions of second-order tangent sets exist:
1. Geometric ( $T^2_{0,u}X$ ): Defined via the existence of analytic curves $\gamma(t) \subset X$ with a specific second-order expansion $\gamma(t) = tu + \frac{1}{2}t^2w + o(t^2)$ . This set represents coefficients $w$ that are geometrically realizable.
2. Algebraic ( $T^{2,a}_{0,u}X$ ): Defined by the initial forms (lowest degree homogeneous parts) and the next homogeneous terms of the defining equations of $X$ . This set represents coefficients that are algebraically admissible and explicitly computable.
Core Question: Under what conditions do these two sets coincide ( $T^2_{0,u}X = T^{2,a}_{0,u}X$ )? If they differ, the algebraic set may contain "spurious" directions that cannot be realized by actual curves in the set, leading to incorrect optimality conditions if used blindly.

2. Methodology

The author employs a combination of real and complex analytic geometry and variational analysis:

Analytic-Curve Characterization: The paper relies heavily on the characterization that $w \in T^2_{0,u}X$ if and only if there exists an analytic arc in $X$ starting at 0 with tangent $u$ and second-order coefficient $w$ .
Taylor Expansion Analysis: By expanding defining functions $f \in I(X,0)$ along candidate curves, the author derives necessary algebraic conditions for $w$ .
Implicit Function Theorem: To prove equality ( $T^2 = T^{2,a}$ ), the author uses the analytic implicit function theorem to construct specific analytic curves that realize algebraically admissible coefficients under non-degeneracy assumptions.
Counter-Examples: Explicit constructions of real and complex analytic sets where the inclusion is strict, demonstrating that algebraic admissibility does not always imply geometric realizability.

3. Key Contributions and Results

A. General Inclusion and Strictness

Theorem 3.8: Proves the general inclusion $T^2_{0,u}X \subseteq T^{2,a}_{0,u}X$ for any analytic set $X$ and nonzero tangent direction $u$ .
Strictness Examples (Section 3.3):
- Real Case: For $X = \{z^2 - x^3y^3 = 0\} \subset \mathbb{R}^3$ , the algebraic set is the entire space $\mathbb{R}^3$ , while the geometric set is a proper subset (or empty) depending on the signs of the components of $u$ . This shows that algebraic conditions can be satisfied by coefficients that violate the sign constraints required for real analytic curves.
- Complex Case: For the same set in $\mathbb{C}^3$ , $T^{2,a}_{0,u}X = \mathbb{C}^3$ , but $T^2_{0,u}X = \mathbb{C}^2 \times \{0\}$ . The geometric set is strictly smaller because the complex square root imposes constraints on the third coordinate that the algebraic definition misses.

B. The Realizability Condition

Definition 4.1: Introduces the concept of second-order realizability: $X$ is second-order realizable along $u$ if every $w \in T^{2,a}_{0,u}X$ can be realized by an analytic curve in $X$ .
Theorem 4.2: Establishes that $T^2_{0,u}X = T^{2,a}_{0,u}X$ if and only if $X$ is second-order realizable along $u$ . This reframes the equality problem as a realizability problem.

C. Sufficient Conditions for Equality

The paper proves that equality holds for several important classes of analytic sets (Theorem 4.7):

Smooth Analytic Germs: Trivially true as the set is locally a linear subspace.
Homogeneous Analytic Cones: The homogeneity of defining equations ensures that algebraic conditions are sufficient for realizability.
Hypersurfaces with Nondegenerate Directions: If the gradient of the initial form $\nabla f^{[*]}(u) \neq 0$ , the implicit function theorem allows the construction of the required curve.
Nondegenerate Analytic Complete Intersections: If the matrix of gradients of initial forms has full rank, equality holds.

D. Application to Optimization

Optimality Conditions: The paper derives second-order necessary and sufficient conditions for $C^2$ $C^{2}$ optimization problems on closed sets (Theorems 5.1 and 5.3).
- Necessary: $\inf_{w \in T^2_{0,u}X} (\langle u, \nabla^2 f(0)u \rangle + \langle \nabla f(0), w \rangle) \geq 0$ .
- Sufficient: Requires a "parabolic regularity" condition ensuring near-minimizing sequences admit second-order expansions.
Algebraic Reformulation (Corollary 5.4): For the classes identified in Section 4, the geometric set $T^2_{0,u}X$ in the optimality conditions can be replaced by the explicitly computable algebraic set $T^{2,a}_{0,u}X$ .
Warning Example (Example 5.7): Demonstrates that if $T^2_{0,u}X \neq T^{2,a}_{0,u}X$ , using the algebraic set can lead to false negatives (falsely concluding a local minimizer is not optimal). In the example, the algebraic set yields an infimum of $-\infty$ , while the geometric set is empty (making the condition vacuously true), and the point is indeed a local minimizer.

4. Significance

Bridging Geometry and Algebra: The paper provides a rigorous framework for when geometric properties of singular sets can be fully captured by algebraic equations. This is crucial for computational optimization, where algebraic conditions are tractable but geometric ones are not.
Computational Tractability: By identifying specific classes (smooth, cones, nondegenerate hypersurfaces/intersections) where the algebraic and geometric sets coincide, the paper enables the use of explicit algebraic formulas for second-order optimality tests in these cases.
Clarification of Singularities: It highlights the subtle differences between real and complex analytic sets regarding second-order behavior, showing that real analytic sets can exhibit more restrictive geometric realizability than their algebraic definitions suggest.
Robustness of Optimality Conditions: The work warns against the uncritical substitution of algebraic tangent sets for geometric ones in optimization algorithms, providing a clear criterion (realizability) to ensure the validity of such substitutions.

In summary, Le Cong Trinh's work establishes that while algebraic second-order tangent sets are always supersets of geometric ones, they coincide under specific non-degeneracy conditions. This coincidence is the key to making second-order optimality conditions for singular analytic sets both theoretically sound and computationally feasible.