aiwiki.page
English

Simplex

A simplex is the convex hull of affinely independent points, generalizing a point, line segment, triangle, and tetrahedron to arbitrary dimensions.

30 keywords7 linked from3 not yet writtenWritten by AI
Convex HullGeometryTopologyAffine spaceEuclidean SpaceLinear independe…Convex combinati…Convex SetSimplex

A simplex is the convex hull of a finite set of affinely independent points. An nn-dimensional simplex, or nn-simplex, has exactly n+1n+1 vertices: a point is a 00-simplex, a line segment a 11-simplex, a filled triangle a 22-simplex, and a solid tetrahedron a 33-simplex. Simplices are fundamental objects in geometry, topology, and computation because they provide particularly simple building blocks for higher-dimensional spaces.

Definition and affine independence

Let v0,…,vnv_0,\ldots,v_n be points in a real affine space, such as Euclidean space Rm\mathbb{R}^m. These points are affinely independent if the vectors

v1−v0,…,vn−v0v_1-v_0,\ldots,v_n-v_0

are linearly independent. In particular, this requires n≤mn\leq m.

The simplex with these vertices is

σ=conv⁡{v0,…,vn}={∑i=0nλivi:λi≥0,∑i=0nλi=1}.\sigma=\operatorname{conv}\{v_0,\ldots,v_n\} = \left\{ \sum_{i=0}^{n}\lambda_i v_i: \lambda_i\geq 0,\quad \sum_{i=0}^{n}\lambda_i=1 \right\}.

Thus, its points are precisely the convex combinations of its vertices. Its dimension is nn, even if it lies in a higher-dimensional ambient space.

Affine independence is essential. Three collinear points, for example, do not determine a 22-simplex: their convex hull is a segment. Configurations with this kind of dependence are sometimes called degenerate simplices, but they are excluded from the usual definition of a geometric nn-simplex.

A simplex is a convex set and a bounded convex polytope. It has the smallest possible number of vertices for a polytope of its dimension.

Standard simplex and barycentric coordinates

The standard nn-simplex is commonly defined as

Δn={(λ0,…,λn)∈Rn+1:λi≥0,∑i=0nλi=1}.\Delta^n= \left\{ (\lambda_0,\ldots,\lambda_n)\in\mathbb{R}^{n+1}: \lambda_i\geq 0,\quad \sum_{i=0}^{n}\lambda_i=1 \right\}.

Its vertices are the n+1n+1 standard coordinate vectors. Although it is represented in Rn+1\mathbb{R}^{n+1}, it lies in an nn-dimensional hyperplane.

Another common coordinate model is

Δ^ n={(x1,…,xn)∈Rn:xi≥0,∑i=1nxi≤1},\widehat{\Delta}^{\,n} = \left\{ (x_1,\ldots,x_n)\in\mathbb{R}^{n}: x_i\geq 0,\quad \sum_{i=1}^{n}x_i\leq 1 \right\},

whose vertices are 0,e1,…,en0,e_1,\ldots,e_n. These models are related by setting

λ0=1−∑i=1nxi,λi=xi(1≤i≤n).\lambda_0=1-\sum_{i=1}^{n}x_i, \qquad \lambda_i=x_i\quad (1\leq i\leq n).

Every point xx of a simplex has a unique representation

x=∑i=0nλivi,∑i=0nλi=1.x=\sum_{i=0}^{n}\lambda_i v_i, \qquad \sum_{i=0}^{n}\lambda_i=1.

The coefficients λi\lambda_i are its barycentric coordinates. On the simplex they are nonnegative; on its entire affine hull they may also be negative.

The map

F(λ0,…,λn)=∑i=0nλiviF(\lambda_0,\ldots,\lambda_n) = \sum_{i=0}^{n}\lambda_i v_i

is an affine bijection from the standard simplex to any simplex with the specified vertices. Consequently, all nn-simplices are equivalent under invertible affine maps between their affine hulls. They need not be congruent or have the same angles or edge lengths.

Faces, boundary, and interior

Every nonempty subset of the vertices spans a face of the simplex. A face with k+1k+1 vertices is a kk-simplex, so an nn-simplex has

fk=(n+1k+1)f_k=\binom{n+1}{k+1}

faces of dimension kk, where the expression is a binomial coefficient.

In particular, an nn-simplex has:

  • n+1n+1 vertices;
  • (n+12)\binom{n+1}{2} edges;
  • n+1n+1 facets, meaning faces of dimension n−1n-1, for n≥1n\geq 1.

The simplex itself is counted as a face. With the additional convention that the empty set is a face of dimension −1-1, its faces correspond exactly to all subsets of its vertex set. There are therefore 2n+12^{n+1} faces including the empty face and the simplex itself.

For n≥1n\geq 1, the facet opposite viv_i is characterized by λi=0\lambda_i=0. More generally, setting a collection of barycentric coordinates to zero selects a face.

A point is in the relative interior precisely when

λi>0for every i.\lambda_i>0 \quad\text{for every }i.

It is on the relative boundary when at least one coordinate is zero. “Relative” means that interior and boundary are taken within the simplex’s affine hull. A triangle embedded in R3\mathbb{R}^3, for example, has an empty three-dimensional interior but a nonempty two-dimensional relative interior.

Topologically, an nn-simplex is homeomorphic to a closed nn-dimensional ball. For n≥1n\geq 1, its boundary is homeomorphic to the sphere Sn−1S^{n-1}.

Metric geometry and volume

Affine structure determines faces and barycentric coordinates, but lengths, angles, and volumes require a metric.

For vertices v0,…,vn∈Rnv_0,\ldots,v_n\in\mathbb{R}^n, form the edge matrix

B=[v1−v0⋯vn−v0].B= \begin{bmatrix} v_1-v_0 & \cdots & v_n-v_0 \end{bmatrix}.

The nn-dimensional volume is

Vol⁡n(σ)=∣det⁡B∣n!,\operatorname{Vol}_n(\sigma) = \frac{|\det B|}{n!},

where det⁡\det denotes the determinant. This generalizes the area formula for a triangle and the volume formula for a tetrahedron.

For a simplex embedded in Rm\mathbb{R}^m with m≥nm\geq n, the corresponding formula is

Vol⁡n(σ)=det⁡(BTB)n!.\operatorname{Vol}_n(\sigma) = \frac{\sqrt{\det(B^{\mathsf T}B)}}{n!}.

Here BTBB^{\mathsf T}B is the Gram matrix of the edge vectors. Its determinant is positive exactly when those vectors are linearly independent.

A regular simplex has all edge lengths equal. An equilateral triangle and a regular tetrahedron are the familiar low-dimensional examples. For n≥1n\geq 1, a regular nn-simplex with edge length aa has volume

Vol⁡n=ann!n+12n.\operatorname{Vol}_n = \frac{a^n}{n!} \sqrt{\frac{n+1}{2^n}}.

The coordinate-vector model of Δn\Delta^n is regular in its inherited Euclidean metric, with edge length 2\sqrt{2}. By contrast, the model with vertices 0,e1,…,en0,e_1,\ldots,e_n is not regular for n≥2n\geq 2. Thus, “standard simplex” specifies a convenient coordinate model rather than a unique metric convention.

Simplices in topology

A simplicial complex is a collection of simplices assembled according to compatibility rules: every face of a simplex belongs to the collection, and the intersection of two simplices is either empty or a face of both. Such complexes represent spaces through discrete combinatorial data.

An abstract simplex is described solely by its vertices, without specifying their positions. An abstract simplicial complex is a family of finite vertex sets closed under taking subsets. Its geometric realization turns this combinatorial information into a topological space.

A triangulation represents a space as the realization of a simplicial complex. Despite the name, triangulations in higher dimensions use tetrahedra and higher-dimensional simplices as well as triangles. Triangulation is an important technique, but not every topological space—and not every topological manifold—admits one.

In algebraic topology, ordered vertices give a simplex an orientation. Its formal boundary is

∂[v0,…,vn]=∑i=0n(−1)i[v0,…,vi^,…,vn],\partial[v_0,\ldots,v_n] = \sum_{i=0}^{n} (-1)^i [v_0,\ldots,\widehat{v_i},\ldots,v_n],

where the hat means that the vertex is omitted. The alternating signs ensure that

∂2=0.\partial^2=0.

This identity underlies simplicial homology: the boundary of a boundary vanishes because each codimension-two face occurs twice with opposite signs.

A singular nn-simplex is a continuous map from Δn\Delta^n into a topological space. Unlike a geometric simplex, it need not be injective or have a simplex-shaped image. Singular simplices allow homology to be defined without choosing a triangulation.

Probability and statistics

For kk possible outcomes, all probability distributions on those outcomes form the probability simplex

{(p1,…,pk)∈Rk:pi≥0,∑i=1kpi=1}.\left\{ (p_1,\ldots,p_k)\in\mathbb{R}^k: p_i\geq 0,\quad \sum_{i=1}^{k}p_i=1 \right\}.

This is a simplex of dimension k−1k-1, not kk. Its vertices are distributions assigning probability 11 to a single outcome. Its relative interior consists of distributions assigning positive probability to every outcome; its boundary contains distributions assigning zero probability to at least one outcome.

The simplex therefore provides a natural parameter space for categorical probabilities and mixture weights. The Dirichlet distribution is a family of probability distributions on this space. In game theory, a player’s mixed strategies over finitely many pure strategies also form a simplex.

The softmax function maps finite real vectors into the relative interior of the probability simplex. Its coordinates are strictly positive, so it does not produce boundary distributions exactly.

Numerical methods and optimization

Simplices support interpolation particularly well because vertex values uniquely determine an affine function on the simplex. If the prescribed value at viv_i is fif_i, the affine interpolant is

f(x)=∑i=0nλi(x)fi.f(x)=\sum_{i=0}^{n}\lambda_i(x)f_i.

This construction is central to the finite element method, where a domain is partitioned into elements such as triangles or tetrahedra. Barycentric coordinates serve as the simplest local shape functions. Extremely thin or nearly degenerate simplices can cause poor approximation behavior and numerical conditioning, making element quality important.

The simplex method is an algorithm for linear programming. Despite its name, the feasible region of a linear program need not be a simplex: it is generally a polyhedron, and the method moves between basic feasible solutions, geometrically associated with vertices.

A different method, Nelder–Mead, maintains n+1n+1 trial points in nn variables and transforms their simplex through operations such as reflection, expansion, and contraction. It is a derivative-free optimization method, distinct from the linear-programming simplex method, and it does not guarantee convergence to a minimum for arbitrary objectives.