Abstract
The notion of polar degrees is fundamental for assessing the algebraic complexity of polynomial optimization problems of a metric origin. We already recognized this point for Euclidean distance optimization, and it is central for polyhedral norms, with a focus on the Wasserstein metric from optimal transport theory. The punchline is that polar degrees govern linear programming over real algebraic varieties.
Introduction
In metric algebraic geometry, optimization problems often involve finding points on algebraic varieties—shapes defined by polynomial equations—that satisfy specific criteria, such as being the "closest point" to a given data set. The complexity of these problems is not merely a function of the number of variables but is determined by the "algebraic complexity," which counts the number of potential solutions or critical points. This paper establishes a unified framework for "polar degrees," which serve as the primary measure for this complexity.
Geometric and Algebraic Definitions
The authors present three equivalent perspectives on polar degrees, bridging intuitive geometry with abstract topology.
1. Polar Varieties and the Contour Intuition
Geometrically, polar degrees can be understood through the concept of a "contour." If one views a surface in 3-space from a specific point $v$, the outline or contour observed is a polar variety. The degree of this contour curve represents the first polar degree ($\mu_1$).
2. The Gauss Map and Conormal Varieties
More formally, the Gauss map associates each point on a variety with its tangent space. By analyzing how these spaces intersect Schubert varieties—which represent geometric constraints—one can define polar degrees algebraically. Alternatively, one can consider the conormal variety, an object in a higher-dimensional space that combines the variety with its tangent hyperplanes. The multidegrees of the conormal variety are shown to be identical to the polar degrees.
Projective Duality
A significant insight of this work is the role of projective duality. There is a fundamental symmetry between a variety $X$ and its dual variety $X^\vee$ (the set of all its tangent planes). The sequence of polar degrees for $X$ is the reverse of the sequence for $X^\vee$.
Key Results and Formulas
The paper provides several critical theoretical bridges:
- Equivalence Theorem (4.16): Proves the identity between the multidegrees of conormal varieties and the degrees of polar varieties.
- Chern Class Formula (4.20): Provides an explicit summation to convert intrinsic topological invariants (Chern classes) into extrinsic polar degrees.
- ED Degree Formula (4.24): Derives the Euclidean Distance (ED) degree—the number of critical points in a "closest point" problem—entirely in terms of Chern classes.
Conclusion
Polar degrees provide the governing framework for the complexity of linear programming and metric optimization over real algebraic varieties. While challenges remain in computing these values for singular varieties or high-dimensional neural landscapes, the connection between a shape's intrinsic topology and its extrinsic optimization complexity offers a powerful tool for modern algebraic geometry.
Summary updated