Last updated: 2026-09-26

M
Masters level
FDN
Foundational — Knowledge that endures for decades — core principles

The Geometry of Doubt: Blades as Spatial Boxes

1.1 Classical Interval Arithmetic

An interval \([a, b] \subset \mathbb{R}\) with \(a \leq b\) represents a quantity whose precise value is unknown but known to lie within those bounds. The arithmetic of intervals was systematically developed by Moore[1], and extended to rigorous error analysis by Alefeld and Mayer[2]. The four basic operations are:

\[ [a,b] + [c,d] = [a+c,\ b+d] \] \[ [a,b] \cdot [c,d] = [\min(ac,ad,bc,bd),\ \max(ac,ad,bc,bd)] \]

These rules ensure that if \(x \in [a,b]\) and \(y \in [c,d]\), then \(x \circ y \in [a,b] \circ [c,d]\) for any elementary operation \(\circ\). This is the containment property, which guarantees rigorous enclosures. true range never escapes the box

Definition 1.1 — Interval containment

An interval computation \(\mathbf{F}([a,b])\) is a valid enclosure of \(f\) if \(f(x) \in \mathbf{F}([a,b])\) for every \(x \in [a,b]\). A method is inclusion-monotone if narrowing the input interval never widens the output.[1]

The limitation of classical interval arithmetic is that it operates on scalars. When the uncertain quantity is a direction, a rotation, or an area element, scalar intervals cannot represent the uncertainty faithfully. a box can't bound a spin

1.2 Geometric Algebra: A Brief Foundation

Geometric algebra, introduced in its modern form by Hestenes[3] and comprehensively treated by Dorst, Fontijne, and Mann[4], is built on a single product — the geometric product — that unifies the dot product and cross product, and extends naturally to all dimensions.

Given an \(n\)-dimensional real vector space with basis \(\{\mathbf{e}_1, \ldots, \mathbf{e}_n\}\), the geometric algebra \(\mathcal{G}(\mathbb{R}^n)\) is the associative algebra generated by these basis vectors subject to:

\[ \mathbf{e}_i^2 = +1, \quad \mathbf{e}_i \mathbf{e}_j = -\mathbf{e}_j \mathbf{e}_i \quad (i \neq j) \]

The general element of \(\mathcal{G}(\mathbb{R}^n)\) is a multivector: a sum of components of different grades. think polynomial in basis blades

Definition 1.2 — Grade and blade

A k-blade is a product of \(k\) linearly independent 1-vectors. Its grade is \(k\). The grade-\(k\) part of a multivector \(A\) is written \(\langle A \rangle_k\). A multivector is a sum of blades of (possibly different) grades.[4]

Grade 0 Scalar magnitude only Grade 1 Vector directed line Grade 2 Bivector oriented area Grade 3 Trivector oriented volume ∧ ∧ ∧ ∧ ··· Grade increases by 1 with each outer product
Fig. 1.1. The grade hierarchy of blades in \(\mathcal{G}(\mathbb{R}^3)\). Each application of the outer product \(\wedge\) raises the grade by one, building from scalars through vectors, bivectors, and trivectors. After Dorst et al. [4].

1.3 The Outer Product and Oriented Subspaces

The outer product (wedge product) \(\mathbf{a} \wedge \mathbf{b}\) of two vectors represents the oriented parallelogram swept by \(\mathbf{a}\) and \(\mathbf{b}\). It is antisymmetric: \(\mathbf{a} \wedge \mathbf{b} = -\mathbf{b} \wedge \mathbf{a}\). The magnitude \(|\mathbf{a} \wedge \mathbf{b}|\) equals the area of the parallelogram; the sign encodes orientation. sign is the orientation, not an error

\[ \mathbf{a} \wedge \mathbf{b} = \sum_{i < j} (a_i b_j - a_j b_i)\,\mathbf{e}_i \wedge \mathbf{e}_j \]

1.4 Interval Blades: Spatial Uncertainty

We now combine the two ideas. An interval blade is a bounded set of blades representing uncertainty in an oriented subspace. For a 1-blade (vector), this is a cone of possible directions and magnitudes. For a 2-blade (bivector), it is a bounded set of possible oriented area elements.

Definition 1.3 — Interval blade

An interval k-blade \([B_1, B_2]\) is a connected, bounded subset of the space of grade-\(k\) blades in \(\mathcal{G}(\mathbb{R}^n)\), such that \(B_1\) and \(B_2\) are the "extremal" blades with respect to some ordering or bounding criterion. The set satisfies the containment property under the geometric product.[5]

a b Scalar interval [a, b] One dimension of uncertainty e₁ e₂ Δe₁ Δe₂ [B₁, B₂] Interval 2-blade Two dimensions of uncertainty Interval 3-blade Three dimensions of uncertainty
Fig. 1.2. The progression from scalar interval to interval blade. A scalar interval is a segment on the real line; a 2-blade interval is a bounded region in the space of oriented area elements; a 3-blade interval is a bounded region in the space of oriented volumes.

1.5 The Geometric Product

The geometric product of two vectors \(\mathbf{a}\) and \(\mathbf{b}\) decomposes as:

\[ \mathbf{ab} = \mathbf{a} \cdot \mathbf{b} + \mathbf{a} \wedge \mathbf{b} \]

where the inner product (scalar part) captures the symmetric relationship and the outer product (bivector part) captures the antisymmetric, oriented area relationship. This unification is the source of GA's expressive power: a single product encodes both metric (distance/angle) and topological (orientation/incidence) structure. cf. why one product beats two

// Geometric product decomposition (1-vectors a, b) ab = a·b + a∧b scalar bivector // For orthonormal basis: e_i·e_j = δᵢⱼ eᵢeⱼ = δᵢⱼ + eᵢ∧eⱼ (i≠j: eᵢeⱼ = eᵢ∧eⱼ, a pure bivector) // Grade-selecting operator ⟨AB⟩_k = grade-k part of the geometric product AB

1.6 Arithmetic of Interval Blades

The geometric product extends to interval blades by the same containment principle as classical interval arithmetic. For interval blades \([A_1, A_2]\) and \([B_1, B_2]\):

\[ [A_1, A_2] \cdot [B_1, B_2] \supseteq \{ AB \mid A \in [A_1,A_2],\ B \in [B_1,B_2] \} \]
!

Wrapping effect. Like scalar interval arithmetic, blade interval arithmetic can overestimate the true range — the so-called dependency problem. When the same uncertain blade appears multiple times in an expression, naive interval evaluation can produce intervals wider than necessary. Affine arithmetic extensions[5] mitigate this by tracking the source of uncertainty.

Scalar intervals bound a number; interval blades bound a direction, an area, or a volume. The same containment property carries over, along with the same overestimation risk, and the chapters that follow put that combination to work.

References

  1. Moore, R. E., Kearfott, R. B., & Cloud, M. J. (2009). Introduction to Interval Analysis. SIAM.
  2. Alefeld, G., & Mayer, G. (2000). Interval analysis: Theory and applications. Journal of Computational and Applied Mathematics, 121(1–2), 421–464.
  3. Hestenes, D. (1999). New Foundations for Classical Mechanics (2nd ed.). Kluwer.
  4. Dorst, L., Fontijne, D., & Mann, S. (2007). Geometric Algebra for Computer Science. Morgan Kaufmann.
  5. de Figueiredo, L. H., & Stolfi, J. (2004). Affine arithmetic: concepts and applications. Numerical Algorithms, 37(1–4), 147–158.