---
title: Combinatorial Prisms in Graph Theory
url: https://www.emergentmind.com/topics/combinatorial-prisms
type: topic
---

# Combinatorial Prisms in Graph Theory

Combinatorial prisms are product-type or matching-type constructions that create a two-layered combinatorial object from a graph, simplex, or rectangular box. In the literature surveyed here, the term includes generalized prisms \(H\Box K_2\) inside Bubble-sort graphs, ordinary prism graphs \(C_n\times P_2\), complementary prisms \(\Gamma\overline{\Gamma}\), simplicial prisms \(\Delta_n\times\Delta_1\), and slices or inscribed substructures of rectangular prisms. Across these settings, prisms serve as organizing devices for Hamiltonicity, threshold colorings, automorphism groups, representation-theoretic bijections, Ehrhart theory, and enumerative generating functions [1901.03917] [1608.02332] [2110.10416] [2208.12957] [2202.11808] [1009.4859].

## 1. Generalized prisms in the Bubble-sort graph

In the Bubble-sort graph,
\[
BS_n=\mathrm{Cay}\bigl(\mathrm{Sym}_n,\{b_1,b_2,\dots,b_{n-1}\}\bigr),\qquad b_i=(i\;i+1),
\]
a generalized prism is the Cartesian product \(2\text{-}H=H\Box K_2\). The construction used in \(BS_n\) takes \(H=BS_{n-2}\) and defines
\[
GP_n:=2\text{-}BS_{n-2}\subset BS_n.
\]
For fixed distinct symbols \(i,j\in\{1,\dots,n\}\), the subgraph \(GP_n(i,j)\cong 2\text{-}BS_{n-2}\) is the induced subgraph whose vertices are precisely the permutations whose last two positions are \(\{i,j\}\):
\[
V\bigl(GP_n(i,j)\bigr)=
\{[\pi_1\cdots \pi_{n-2}\,i\,j]\cup[\pi_1\cdots \pi_{n-2}\,j\,i]:
\{\pi_1,\dots,\pi_{n-2}\}=\{1,\dots,n\}\setminus\{i,j\}\}.
\]
It has two layers, one ending in \(\dots i\,j\) and one ending in \(\dots j\,i\), each a copy of \(BS_{n-2}\), while the vertical edges are exactly the edges induced by \(b_{n-1}\), which swap the last two entries [1901.03917].

As \(\{i,j\}\) ranges over the \(\binom n2\) unordered pairs, the subgraphs \(GP_n(i,j)\) are vertex-disjoint and together cover every vertex of \(BS_n\) exactly once; the paper describes this as a maximal cover. The factor graph whose vertices are these prisms and whose edges record adjacency between distinct prisms is precisely the Johnson graph \(J(n,2)\). Since \(J(n,2)\) is Hamiltonian, indeed Hamilton-connected, a Hamiltonian cycle in the factor graph can be lifted to a Hamiltonian cycle in \(BS_n\): one traverses prisms in the order given by the factor-graph cycle and, inside each prism, joins the incoming and outgoing vertices by a Hamiltonian path. The key technical input is Proposition 3.2: in each \(GP_n=2\text{-}BS_{n-2}\subset BS_n\), for any two vertices of opposite parity, there is a Hamiltonian path joining them. Because \(BS_n\) is bipartite and the connecting \(b_{n-2}\)-edges toggle parity, the required inside-prism paths always exist. In \(BS_5\), this yields \(\binom52=10\) prisms, each isomorphic to \(2\text{-}BS_3\cong 2\text{-}C_6\). This suggests that generalized prisms function as local Hamiltonian modules inside a larger Cayley graph [1901.03917].

## 2. Prism graphs and threshold colorings

For Fijavž and Kriesell, the prism graph is \(G_n=C_n\times P_2\), with vertex set
\[
V(G_n)=\{v_0,\dots,v_{n-1}\}\cup\{u_0,\dots,u_{n-1}\},
\]
two peripheral \(n\)-cycles, and spokes \(v_i-u_i\). Given integers \(r\ge t\) and a set \(N\subseteq E(G)\) of near edges, an \((r,t)\)-threshold-coloring is a map \(c:V(G)\to\{0,\dots,r-1\}\) such that
\[
|c(u)-c(v)|\le t\quad\forall\,uv\in N,\qquad
|c(u)-c(v)|> t\quad\forall\,uv\in E(G)\setminus N.
\]
A graph is total threshold colorable if there exist integers \(r,t\) such that for every \(N\subseteq E(G)\), the graph admits such a coloring. The main theorem is that for every \(n\ge3\), the prism \(G_n=C_n\times P_2\) is total threshold colorable, and in fact one can take \(r=31\) and \(t=4\) [1608.02332].

The proof proceeds by a case analysis on the arrangement of far edges around the \(n\) squares of the prism. If there is a useful cut of far edges splitting the prism into two ladder-like pieces, each piece can be \((5,1)\)-colored, producing an \((11,1)\)-coloring after reassembly. Otherwise, local structure in balanced or unbalanced 4-cycles is exploited, together with Lemma 2.2, to build a \((13,4)\)-coloring; an augmentation step then yields a uniform \((31,4)\)-coloring. Three auxiliary facts structure the argument: Lemma 2.1 is a 4-cycle sign lemma controlling relative inequalities in a mixed near/far square, Lemma 2.2 is the “A–B decomposition” that patches separate colorings through a small induced matching, and Lemma 3.1 shows that every ladder \(P_n\times K_2\) is \((5,1)\)-total-threshold colorable [1608.02332].

The contrast with Möbius ladders is explicit: Möbius ladders are not total threshold colorable, and from this it follows that there is no characterization of being total threshold colorable in terms of a finite set of forbidden subgraphs. For particular near/far labelings, much smaller parameters can suffice. When all spokes of the triangular prism are near and all peripheral edges are far, a \((6,1)\)-coloring works; for the square prism with the same labeling pattern, the paper gives a concrete \((8,1)\)-coloring. The prism family therefore occupies a specific position in threshold-coloring theory: it is robust under arbitrary near/far edge partitions, but closely related cubic graphs need not share that robustness [1608.02332].

## 3. Complementary prisms, symmetry, and cores

Given a finite simple graph \(\Gamma\) on \(n\) vertices, its complementary prism \(\Gamma\overline{\Gamma}\) is obtained from \(\Gamma\) and its complement \(\overline{\Gamma}\) by adding a perfect matching between corresponding vertices in the two layers. Writing
\[
W_1=\{(v_i,1):i=1,\dots,n\},\qquad W_2=\{(v_i,2):i=1,\dots,n\},
\]
one has \(V(\Gamma\overline{\Gamma})=W_1\cup W_2\), and
\[
|V(\Gamma\overline{\Gamma})|=2n,\qquad
|E(\Gamma\overline{\Gamma})|=\binom n2+n=\frac{n(n+1)}2.
\]
Orel gives a general description of \(\mathrm{Aut}(\Gamma\overline{\Gamma})\). Outside the exceptional Petersen-type families \(C_5(\Lambda)\) and \(A(\Lambda)\), every automorphism either preserves the two layers and acts there by a single \(\phi\in\mathrm{Aut}(\Gamma)\), or swaps the layers and is induced by an antimorphism \(\sigma:\Gamma\to\overline{\Gamma}\). A corollary is that the ratio
\[
r=\frac{|\mathrm{Aut}(\Gamma\overline{\Gamma})|}{|\mathrm{Aut}(\Gamma)|}
\]
can attain only the values \(1\), \(2\), \(4\), and \(12\) [2110.10416].

The same paper determines the Cheeger number:
\[
h(\Gamma\overline{\Gamma})=
\begin{cases}
1-\frac1n,&\text{if \(\Gamma\) or \(\overline{\Gamma}\) contains an edge \(uv\) with \(\deg(u)=1,\deg(v)=n-1\),}\\
1,&\text{otherwise.}
\end{cases}
\]
It also proves that \(\Gamma\overline{\Gamma}\) is vertex-transitive if and only if \(\Gamma\) is vertex-transitive and self-complementary. In this case, if \(n>5\), the complementary prism is Hamiltonian-connected. At the same time, if \(|V(\Gamma)|>1\), then \(\Gamma\overline{\Gamma}\) is never a Cayley graph [2110.10416].

The endomorphism theory is equally rigid. If \(\Gamma\) is \((n-1)/2\)-regular or vertex-transitive self-complementary with \(n>1\), then \(\mathrm{core}(\Gamma\overline{\Gamma})\) is either the full graph, or isomorphic to \(\mathrm{core}(\Gamma)\) sitting in one layer, or isomorphic to \(\mathrm{core}(\overline{\Gamma})\) sitting in the other. From this, Orel proves that if \(\Gamma\) is strongly regular and self-complementary, then \(\Gamma\overline{\Gamma}\) is a core; likewise, if \(\Gamma\) is vertex-transitive self-complementary and either \(\Gamma\) is already a core or \(\mathrm{core}(\Gamma)\) is complete, then \(\Gamma\overline{\Gamma}\) is a core. The model example is \(\Gamma=C_5\): then \(\Gamma\overline{\Gamma}\) is the Petersen graph, with \(10\) vertices, \(15\) edges, automorphism-group ratio \(12\), Cheeger number \(1\), vertex-transitivity, non-Cayley status, Hamiltonian-connectedness, and the core property [2110.10416].

## 4. Simplicial prisms and triangulations

The polytope prism \(\Delta_n\times\Delta_1\) is the product of the standard \(n\)-simplex
\[
\Delta_n=\mathrm{conv}\{e_0,e_1,\dots,e_n\}
\]
with the 1-simplex \(\Delta_1=\mathrm{conv}\{f_0,f_1\}\). Its \(2(n+1)\) vertices are
\[
a_i=(e_i,f_0),\qquad b_i=(e_i,f_1),\qquad i=0,\dots,n.
\]
Any choice of one vertex from each pair \(\{a_i,b_i\}\) yields an \(n\)-simplex. If \(X=x_0x_1\cdots x_n\in\{a,b\}^{n+1}\), then
\[
\Delta_X=\mathrm{conv}\{x_{00},x_{11},\dots,x_{nn}\},
\]
where \(x_{ii}=a_i\) if \(x_i=a\) and \(x_{ii}=b_i\) if \(x_i=b\). Such a simplex lies on the boundary exactly when \(X=a^{n+1}\) or \(X=b^{n+1}\); the internal \(n\)-simplices are therefore in bijection with the words in \(\{a,b\}^{n+1}\) using both letters. In particular, there are \(2^{n+1}-2\) internal \(n\)-simplices [2208.12957].

The same paper identifies these internal simplices with indecomposable two-term presilting complexes over the preprojective algebra \(\Pi_n\) of type \(A_n\). If \(X=x_0\cdots x_n\), then the \(k\)th coordinate of the \(g\)-vector is
\[
g_k=
\begin{cases}
+1,&(x_{k-1},x_k)=(a,b),\\
-1,&(x_{k-1},x_k)=(b,a),\\
0,&\text{otherwise,}
\end{cases}
\]
and this determines the indecomposable presilting complex \(P_X\). Thus
\[
\mathrm{ind}\,2\,\Pi_n\longleftrightarrow \{\text{internal \(n\)-simplices }\Delta_X\}.
\]
A triangulation \(\mathcal T\) of \(\Delta_n\times\Delta_1\) is a maximal collection of non-intersecting internal \(n\)-simplices, equivalently a basic two-term silting complex with \(n\) indecomposable summands. The paper proves that triangulations are in bijection with permutations in \(S_{n+1}\), hence
\[
\#\{\text{triangulations of }\Delta_n\times\Delta_1\}=(n+1)!.
\]
For \(w=i_0i_1\cdots i_n\in S_{n+1}\), the corresponding triangulation consists of the \(n+1\) simplices
\[
\bigl\{\{a_{i_0},\dots,a_{i_j},b_{i_j},\dots,b_{i_n}\}:0\le j\le n\bigr\},
\]
of which the internal ones are those with \(j=0,\dots,n-1\) [2208.12957].

Bistellar flips correspond exactly to silting mutations. If \(P=E\oplus X\) and \(P'=E\oplus Y\), then replacing the indecomposable summand \(X\) by the unique different indecomposable presilting summand \(Y\) matches the replacement of one internal simplex by the unique other internal simplex in a minimal circuit. The symmetric group \(S_{n+1}\) acts simply transitively on the set of triangulations by permuting the labels \(\{0,\dots,n\}\), and the resulting graph of flips is the classical permutohedron of type \(A\) of dimension \(n\). The prism \(\Delta_n\times\Delta_1\) thus mediates a precise correspondence among triangulations, presilting objects, silting objects, mutations, and the symmetric group [2208.12957].

## 5. Slices of rectangular prisms and Ehrhart theory

A different use of prisms arises from rectangular boxes and their hyperplane sections. Fix positive integers \(c_1,\dots,c_n\) and an integer \(k\), and define
\[
\mathscr R_{\mathbf c}=[0,c_1]\times\cdots\times[0,c_n]\subset\mathbb R^n,
\]
together with the thin slice
\[
\mathscr R_{k,\mathbf c}=\{x\in\mathscr R_{\mathbf c}:x_1+\cdots+x_n=k\}
=\{x\in\mathbb R^n:0\le x_i\le c_i,\ \sum_{i=1}^n x_i=k\}.
\]
These polytopes are described as generalizations of the hypersimplex; they are contained in the larger class of polypositroids introduced by Lam and Postnikov, and they coincide with polymatroids satisfying the strong exchange property up to an affinity. Fat slices \(\{a\le \sum x_i\le b\}\) are integrally equivalent to a thin slice in one higher dimension [2202.11808].

For the \((n-1)\)-dimensional lattice polytope \(\mathscr R_{k,\mathbf c}\), the Ehrhart polynomial
\[
L_{\mathscr R_{k,\mathbf c}}(t)=|t\mathscr R_{k,\mathbf c}\cap\mathbb Z^n|
\]
has nonnegative coefficients, and the paper gives a combinatorial formula for all of them in terms of \(\mathbf c\)-compatible weighted permutations. If \(W(\ell,n,m+1,\mathbf c)\) counts compatible weighted permutations with exactly \(m+1\) cycles and total weight \(\ell\), and \(A(m,r)\) is the Eulerian number, then
\[
L_{\mathscr R_{k,\mathbf c}}(t)
=\sum_{m=0}^{n-1}
\left[
\frac{1}{(n-1)!}
\sum_{\ell=0}^{k-1}
W(\ell,n,m+1,\mathbf c)\,
A(m,k-\ell-1)
\right]t^m.
\]
This formula proves Ehrhart positivity and gives a direct combinatorial interpretation of each coefficient [2202.11808].

The \(h^*\)-polynomial also has a combinatorial interpretation. Its coefficient \(h_j^*\) counts \(\mathbf c\)-compatible decorated ordered set partitions of type \((k,n)\) with winding number \(j\). The leading Ehrhart coefficient, equivalently the normalized volume, admits the formula
\[
\mathrm{Vol}(\mathscr R_{k,\mathbf c})
=\frac1{(n-1)!}\sum_{\ell=0}^{k-1}B(\ell,\mathbf c)\,A(n-1,k-\ell-1),
\]
where
\[
B(\ell,\mathbf c)=
\#\{\text{ways to place \(\ell\) indistinguishable balls into \(n\) boxes of capacities \(c_1-1,\dots,c_n-1\)}\}.
\]
The classical Laplace–Stanley identity \(\mathrm{Vol}(\Delta_{k,n})=A(n-1,k-1)\) appears as the special case \(c_i=1\). The same theory identifies the graded algebra
\[
\mathcal V(\mathbf c,k)=
\mathbb F[x_1,\dots,x_n]
\Big\langle
x_1^{\alpha_1}\cdots x_n^{\alpha_n}:
\sum \alpha_i=k,\ \alpha_i\le c_i
\Bigr\rangle
\]
with the Ehrhart ring of \(\mathscr R_{k,\mathbf c}\), so its Hilbert function equals \(L_{\mathscr R_{k,\mathbf c}}(t)\) and its \(h\)-vector coincides with the prism-slice \(h^*\)-polynomial [2202.11808].

## 6. Inscribed polyominoes in a rectangular prism

Goupil and Cloutier study minimal 3D polyominoes inscribed in an \(a\times b\times c\) rectangular prism. Such a polyomino is a finite set of unit cubes in \(\{1,\dots,a\}\times\{1,\dots,b\}\times\{1,\dots,c\}\) that is face-connected, meets all six faces of the ambient prism, and has minimal volume under these conditions. Every such minimal polyomino has exactly
\[
|P|=a+b+c-2
\]
cubes. Equivalently, each of its three orthogonal projections onto the faces of the prism is a minimal 2D polyomino of area \(a+b-1\), \(b+c-1\), or \(c+a-1\) [1009.4859].

Every minimal 3D inscribed polyomino falls into one of three disjoint families: diagonal polyominoes, formed from two hooks joined by a 3D stair along a prism diagonal; \(2D\times2D\) polyominoes, obtained by juxtaposing two perpendicular minimal 2D polyominoes with a short skew-hook; and skew-cross polyominoes, consisting of three mutually perpendicular minimal 2D corner-polyominoes meeting at a central cell of degree \(3\). If
\[
P_{3D}(x,y,z)=\sum_{a,b,c\ge1}p_{3D}(a,b,c)x^ay^bz^c,
\]
then
\[
P_{3D}(x,y,z)=\Diag(x,y,z)+P_{2D\times2D}(x,y,z)+\Skew(x,y,z),
\]
and each summand is a rational function. The 3D stair count is
\[
\Stair(x,y,z)=\sum_{a,b,c\ge1}\binom{a+b+c-3}{a-1,b-1,c-1}x^ay^bz^c
=\frac{xyz}{1-(x+y+z)},
\]
while the skew-cross family has generating function
\[
\Skew(x,y,z)=
\frac{64\,x^3y^3z^3}
{(1-x-y)(1-x-z)(1-y-z)(1-x)^2(1-y)^2(1-z)^2}.
\]
The theory also gives a recurrence for the number \(p_c(a,b,c)\) of corner-polyominoes and closed forms for subfamilies such as stairs, tripods, and corner-polyominoes [1009.4859].

Concrete enumerations illustrate the theory: \(p_{3D}(2,2,2)=32\), \(p_{3D}(2,3,1)=8\), and \(p_{3D}(3,3,3)=2401\). On the diagonal specialization \(x=y=z=t\), the dominant singularity occurs at \(t=1/3\), so
\[
p_{3D}(n,n,n)\sim \kappa\,3^n\,n^\alpha
\qquad\text{for some constants }\kappa,\alpha.
\]
Taken together with prism slices, simplicial prisms, and graph prisms, these results suggest that prism constructions repeatedly convert geometric layering into explicit combinatorics: rational generating functions, cycle decompositions, Hamiltonian assemblies, and representation-theoretic parameterizations all emerge from the same two-level or sectioned architecture.

Source: https://www.emergentmind.com/topics/combinatorial-prisms