Internal Angle Representation

• Internal angle representation is a formalization of angles at nodes and faces, offering a unified framework for geometric, combinatorial, and behavioral analysis.
• It uses metric–combinatorial techniques such as internal angle vectors, flag-angle vectors, and invariant theory to capture essential properties of polytopes and zonotopes.
• Computational approaches leverage sparse linear systems and angle-based embeddings to efficiently solve geometric constraints, benefiting applications from trajectory prediction to motor learning.

Internal angle representation refers to the formalization and quantification of angles formed at nodes, intersections, or faces within geometric, combinatorial, behavioral, and machine learning contexts. It encompasses both the mathematical abstraction of angle data—for example, as vectors or sequences—and the algorithms and combinatorial structures that encode angle-based relationships, constraints, or invariants. These representations are central to fields such as polytope theory, computational geometry, symbolic geometry, trajectory modeling, and human motor learning, where internal angles become the primary carriers of geometric, structural, or behavioral information.

1. Internal Angle Vectors in Polytope Theory

Central to the literature is the characterization of convex polytopes by their internal angles at each face. For a $d$-polytope $P \subset \mathbb{R}^d$, the tangent cone at a $k$-dimensional face $F$ is $T_F P = \operatorname{cone}(-F + P)$. Using a (normalized) cone-valuation $\varphi$, such as solid angle, the $k$-th entry of the interior-angle vector is

$$\alpha_k(P) = \sum_{\dim F = k} \varphi(T_F P), \quad k = 0, 1, \dots, d-1.$$

Thus, the internal angle vector $\alpha(P) = (\alpha_0(P), \ldots, \alpha_{d-1}(P))$ aggregates these values across all faces of each dimension (Backman et al., 2018).

A more specialized version for simplicial polytopes is the $\widehat{\alpha}$-vector, in which each entry $\widehat{\alpha}_i(P)$ sums the cone-angles over all $i$-faces. This encoding enables direct metric–combinatorial synergy, analogous to the face $f$-vector but weighted by intrinsic angle measure (Manecke, 2020).

2. Algebraic and Combinatorial Structure

Angle vectors satisfy deep linear relations generalizing Euler-type formulae. Gram’s relation is uniquely satisfied by the interior-angle vector: $$\alpha_0(P) - \alpha_1(P) + \alpha_2(P) - \cdots + (-1)^{d-1}\alpha_{d-1}(P) = (-1)^{d+1}$$ for all $d$-polytopes (Backman et al., 2018). For zonotopes, interior and exterior angle vectors realize the Whitney numbers of the first and second kind, respectively, associated with the geometric lattice of flats of the zonotope. The theory extends to flag-angle vectors: multilinear invariants indexed by chains of faces, shown to be geometric counterparts to flag-$f$-vectors, and subject to Dehn-Sommerville-type linear relations that echo those for classical face and $h$-vectors.

The $\widehat{\gamma}$-vector, a linear transformation of the $\widehat{\alpha}$-vector patterned after the algebraic $h$-vector, is defined via

$$A_P(t) = \sum_{i=0}^d \widehat{\alpha}_{i-1}(P)(t-1)^{d-i} = \sum_{k=0}^d \widehat{\gamma}_k(P) t^{d-k}$$

for the $\widehat{\alpha}$-polynomial $A_P(t)$ (Manecke, 2020). These vectors underlie a rich system of symmetries (palindromy), nonnegativity, growth, and “flawlessness,” reflecting both metric and combinatorial structure.

3. Computational Approaches to Internal Angle Representation

In algorithmic and symbolic geometry, internal angle representation serves as the foundation for constraint-based geometric reasoning. The Naive Angle Method encodes geometric diagrams as sparse linear systems, with each unknown variable $d_i$ representing the direction (modulo $2\pi$) of a line or ray. Internal angles between two lines $i$ and $j$ are given by $d_i - d_j \bmod 2\pi$ (Todd, 2022).

Geometric constraints—such as specified angles, bisections, or triangle sum properties—become linear equations on the $d_i$’s. The resulting system

$A d = b \pmod{2\pi}$

where $A$ has at most three nonzero entries per row, is solved symbolically or numerically to recover all internal angles. Abstracting the sparsity pattern of $A$ leads to a correspondence with graphs, enabling exhaustive classification and discovery of angle theorems through combinatorial enumeration.

4. Applications in Human Behavior and Trajectory Prediction

Angle-based representations have been adapted in modeling interaction and social context in machine learning. For pedestrian trajectory prediction, angle-based conditioned interaction representations structurally encode agents' spatial relationships and environmental conditions as cyclic angle sequences (Wong et al., 2024). Separate “social” and “conditional” branches embed these sequences, which are then adaptively fused to produce a context-sensitive interaction feature for downstream prediction backbones. While explicit mathematical formulas are not disclosed, the key conceptual advance is using internal angle data not purely for geometric invariants, but as context embeddings for socially conditioned sequence models.

In motor learning, internal angle representation underlies the modeling of adaptation and generalization curves in force-field experiments. Here, the neural estimate of a learned field (directional tuning function $\hat{a}(\theta)$) is modeled as a symmetric Gaussian centered on the trained reach direction $\theta_{\text{train}}$: $$\hat{a}(\theta) = A \exp\left( -\frac{(\theta - \theta_\text{train})^2}{2\sigma^2} \right)$$ This angle-parametric form enables direct testing of internal representations against behavioral generalization data and, when corrected for mechanical impedance, removes artifactual asymmetries (Rezazadeh et al., 2019).

5. Generalizations, Invariant Theory, and Incidence Algebras

The angle representation framework extends to generalized angle vectors, flag-angles, and their interpretations in the algebra of combinatorial incidence. For a polytope $P$, fixing a cone angle $\varphi$ and a multi-index $S = \{0 \le s_1 < \cdots < s_k \le d-1\}$, the flag-angle

$${}_S(P) = \sum_{F_1 \subset \cdots \subset F_k \subset P,\ \dim F_i = s_i} \prod_{j=1}^k \varphi( T_{F_j} F_{j+1} )$$

realizes combinatorial invariants of associated graded posets, such as the flag-Whitney numbers of the geometric lattice of flats for zonotopes (Backman et al., 2018). These representations enable the identification of all linear relations satisfied by angle vectors as the image of algebraic convolutions (e.g., $\zeta$-convolution, $\mu$-convolution) in the incidence algebra $I(L)$. Thus, geometric and combinatorial invariants are unified under the internal angle representation formalism.

6. Limits, Extensions, and Universal Properties

While Dehn-Sommerville-style relations constrain internal angle vectors for simplicial polytopes (palindromic $\widehat{\gamma}$-vectors, nonnegativity, and first-half growth), unimodality can fail in higher dimensions. For example, a six-dimensional polytope with standard solid angle measure exhibits nonnegative, palindromic, but non-unimodal $\widehat{\gamma}$-vector, indicating that angle representation sequences may possess local minima and maxima even under strict metric invariance (Manecke, 2020).

The cone-valuation framework is universal: all key structural results hold for any simple, normalized, nonnegative cone valuation $\alpha \colon C^d \to \mathbb{R}_{\ge 0}$, not only for solid angles. Thus, internal angle representation is fundamentally a valuation-theoretic, combinatorial, and geometric modality, supporting a spectrum of invariants depending on the underlying evaluation algebra.

7. Summary Table: Principal Contexts of Internal Angle Representation

Context	Representation	Principal Reference
Convex/matroid polytopes	Angle vector, flag-angle vector	(Backman et al., 2018, Manecke, 2020)
Symbolic/synthetic geometry	Sparse linear system of angle variables	(Todd, 2022)
Pedestrian trajectory modeling	Angle-based cyclic sequence embedding	(Wong et al., 2024)
Motor learning/generalization	Angle-parametric Gaussian tuning	(Rezazadeh et al., 2019)

Internal angle representation provides a unifying abstraction across geometry, combinatorics, learning theory, and behavioral modeling, encoding angle information at the fundamental level of both structure and computation. Its mathematical theory is grounded in valuation, incidence algebras, and combinatorial invariants, while its algorithmic and application domains extend to sparse systems, graph enumeration, trajectory modeling, and neuro-computational motor control.