Gale diagram

Given a ${\displaystyle d}$-dimensional polytope, with ${\displaystyle n}$ vertices, adjoin 1 to the Cartesian coordinates of each vertex, to obtain a ${\displaystyle (d+1)}$-dimensional column vector. The matrix ${\displaystyle A}$ of these ${\displaystyle n}$ column vectors has dimensions ${\displaystyle (d+1)\times n}$, defining a linear mapping from ${\displaystyle n}$-space to ${\displaystyle (d+1)}$-space, surjective with rank ${\displaystyle d+1}$. The kernel of ${\displaystyle A}$ describes linear dependencies among the ${\displaystyle n}$ original vertices with coefficients summing to zero; this kernel has dimension ${\displaystyle n-d-1}$. The Gale transform of ${\displaystyle A}$ is a matrix ${\displaystyle B}$ of dimension ${\displaystyle n\times (n-d-1)}$, whose column vectors are a chosen basis for the kernel of ${\displaystyle A}$. Then ${\displaystyle B}$ has ${\displaystyle n}$ row vectors of dimension ${\displaystyle n-d-1}$. These row vectors form the Gale diagram of the polytope. A different choice of basis for the kernel changes the result only by a linear transformation.[2]

Note that the vectors in the Gale diagram are in natural bijection with the ${\displaystyle n}$ vertices of the original ${\displaystyle d}$-dimensional polytope, but the dimension of the Gale diagram is smaller whenever ${\displaystyle n\leq 2d}$.

A proper subset of the vertices of a polytope forms the vertex set of a face of the polytope, if and only if the complementary set of vectors of the Gale transform has a convex hull that contains the origin in its relative interior. Equivalently, the subset of vertices forms a face if and only if its affine span does not intersect the convex hull of the complementary vectors.[3]

Because the Gale transform is defined only up to a linear transformation, its nonzero vectors can be normalized to all be ${\displaystyle (n-d-1)}$-dimensional unit vectors. The linear Gale diagram is a normalized version of the Gale transform, in which all the vectors are zero or unit vectors.[4]

Given a Gale diagram of a polytope, that is, a set of ${\displaystyle n}$ unit vectors in an ${\displaystyle (n-d-1)}$-dimensional space, one can choose a ${\displaystyle (n-d-2)}$-dimensional subspace ${\displaystyle S}$ through the origin that avoids all of the vectors, and a parallel subspace ${\displaystyle S'}$ that does not pass through the origin. Then, a central projection from the origin to ${\displaystyle S'}$ will produce a set of ${\displaystyle (n-d-2)}$-dimensional points. This projection loses the information about which vectors lie above ${\displaystyle S}$ and which lie below it, but this information can be represented by assigning a sign (positive, negative, or zero) or equivalently a color (black, white, or gray) to each point. The resulting set of signed or colored points is the affine Gale diagram of the given polytope. This construction has the advantage, over the Gale transform, of using one less dimension to represent the structure of the given polytope.[5]

Gale transforms and linear and affine Gale diagrams can also be described through the duality of oriented matroids.[6] As with the linear diagram, a subset of vertices forms a face if and only if there is no affine function (a linear function with a possibly nonzero constant term) that assigns a non-negative value to each positive vector in the complementary set and a non-positive value to each negative vector in the complementary set.[7]

A ${\displaystyle d}$-dimensional polytope with ${\displaystyle n=d+1}$ vertices, the minimum possible, is a simplex. In this case, the linear Gale diagram is 0-dimensional, consisting only of zero vectors. The affine diagram has ${\displaystyle n}$ gray points.[8]

In a ${\displaystyle d}$-dimensional polytope with ${\displaystyle n=d+2}$ vertices, the linear Gale diagram is one-dimensional, with the vector representing each point being one of the three numbers ${\displaystyle -1}$, ${\displaystyle 0}$, or ${\displaystyle +1}$. In the affine diagram, the points are zero-dimensional, so they can be represented only by their signs or colors without any location value. In order to represent a polytope, the diagram must have at least two points with each nonzero sign. Two diagrams represent the same combinatorial equivalence class of polytopes when they have the same numbers of points of each sign, or when they can be obtained from each other by negating all of the signs.[8]

For ${\displaystyle d=2}$, the only possibility is two points of each nonzero sign, representing a convex quadrilateral. For ${\displaystyle d=3}$, there are two possible Gale diagrams: the diagram with two points of each nonzero sign and one zero point represents a square pyramid, while the diagram with two points of one nonzero sign and three points with the other sign represents the triangular bipyramid.[8]

In general, the number of distinct Gale diagrams with ${\displaystyle n=d+2}$, and the number of combinatorial equivalence classes of ${\displaystyle d}$-dimensional polytopes with ${\displaystyle n}$ vertices, is ${\displaystyle \lfloor d^{2}/4\rfloor }$.[8]

In a ${\displaystyle d}$-dimensional polytope with ${\displaystyle n=d+3}$ vertices, the linear Gale diagram consists of points on the unit circle (unit vectors) and at its center. The affine Gale diagram consists of labeled points or clusters of points on a line. Unlike for the case of ${\displaystyle n=d+2}$ vertices, it is not completely trivial to determine when two Gale diagrams represent the same polytope.[8]

Three-dimensional polyhedra with six vertices provide natural examples where the original polyhedron is of a low enough dimension to visualize, but where the Gale diagram still provides a dimension-reducing effect.

• A regular octahedron has linear Gale diagram comprising three pairs of equal points on the unit circle (representing pairs of opposite vertices of the octahedron), dividing the circle into arcs of angle less than ${\displaystyle \pi }$. Its affine Gale diagram consists of three pairs of equal signed points on the line, with the middle pair having the opposite sign to the outer two pairs.[9]
• A triangular prism has linear Gale diagram comprising six points on the circle, in three diametrically opposed pairs, with each pair representing vertices of the prism that are adjacent on two square faces of the prism. The corresponding affine Gale diagram has three pairs of points on a line, like the regular octahedron, but with one point of each sign in each pair.[10]