Chain Polytope: Theory and Applications

• Chain polytope is a lattice polytope defined for a finite poset, with vertices corresponding to antichains and constraints set by maximal chains.
• It shares Ehrhart equivalence with the order polytope while differing in unimodular equivalence under specific poset conditions, such as the X-free criterion.
• Extensions of the theory include marked, enriched, signed, and double chain polytopes, with applications in toric geometry, commutative algebra, and representation theory.

The chain polytope of a finite poset $P$ is the lattice polytope

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$

Because all coordinates are nonnegative, it suffices to impose the inequalities for maximal chains. Introduced by Stanley as one of the two basic poset polytopes, alongside the order polytope, $\mathcal{C}(P)$ has dimension $|P|$, vertices indexed by antichains, a piecewise-linear correspondence with $\mathcal{O}(P)$, and a rich theory spanning face enumeration, toric rings, triangulations, and several marked, enriched, and signed extensions (Freij-Hollanti et al., 22 Sep 2025, Clarke et al., 2022, Hibi et al., 2012).

1. Definition, vertices, and facets

For a finite poset $P=\{x_1,\dots,x_d\}$, the defining inequalities of $\mathcal{C}(P)$ may be written as

$$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$

for every maximal chain $x_{i_1}<\cdots<x_{i_k}$ in $P$. The maximal-chain formulation is equivalent to the formulation using all chains, since every nonmaximal chain is contained in a maximal chain and all coordinates are nonnegative (Hibi et al., 2012).

The vertices of $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$0 are precisely the indicator vectors of antichains. Writing $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$1, one has

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$2

Moreover, the integer points of $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$3 are exactly these vertices: if $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$4, then each coordinate is in $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$5 and the support is an antichain (Hibi et al., 2012). In the formulation used for order–chain comparisons, the facet inequalities are $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$6 for all $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$7 together with $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$8 for each maximal chain $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$9 (Freij-Hollanti et al., 22 Sep 2025).

Two extremal examples fix the geometry. If $\mathcal{C}(P)$0 is a chain of size $\mathcal{C}(P)$1, then

$\mathcal{C}(P)$2

so $\mathcal{C}(P)$3 is the standard $\mathcal{C}(P)$4-simplex. If $\mathcal{C}(P)$5 is an antichain of size $\mathcal{C}(P)$6, then every chain is a singleton, so $\mathcal{C}(P)$7 (Freij-Hollanti et al., 22 Sep 2025). These two cases already exhibit the basic duality of the theory: highly ordered posets produce simplices, while maximally incomparable posets produce cubes.

2. Relation to the order polytope

The order polytope is

$\mathcal{C}(P)$8

Stanley’s transfer map gives a piecewise-linear bijection between $\mathcal{C}(P)$9 and $|P|$0. In the formulation used in later work,

$|P|$1

and

$|P|$2

This transfer map specializes from the marked setting and makes the relation between order and chain inequalities explicit (Fujita et al., 23 Mar 2026).

A fundamental consequence is Ehrhart equivalence: $|P|$3 and $|P|$4 have the same Ehrhart polynomial. Their common normalized volume is

$|P|$5

where $|P|$6 is the number of linear extensions and $|P|$7 (Hibi et al., 2012, Fujita, 2021). This equality persists in marked and chain-order settings, but unimodular equivalence is substantially more restrictive.

Unimodular equivalence of $|P|$8 and $|P|$9 does not hold in general (Fujita, 2021). In the comparison developed for two-dimensional faces, it is characterized by the absence of the 5-element poset $\mathcal{O}(P)$0 consisting of two incomparable minimal elements below one middle element below two incomparable maximal elements. The order and chain polytopes are unimodularly equivalent if and only if $\mathcal{O}(P)$1 is $\mathcal{O}(P)$2-free; equivalently, in that setting, if and only if their numbers of two-dimensional faces coincide (Freij-Hollanti et al., 22 Sep 2025). A common misconception is that Ehrhart equivalence should force a stronger geometric equivalence; the $\mathcal{O}(P)$3-free criterion shows that it does not.

3. Face structure and low-dimensional geometry

The edge structure of $\mathcal{O}(P)$4 is controlled by connectivity of symmetric differences. If $\mathcal{O}(P)$5 and $\mathcal{O}(P)$6 are antichains, then $\mathcal{O}(P)$7 is an edge of $\mathcal{O}(P)$8 if and only if $\mathcal{O}(P)$9 is connected in $P=\{x_1,\dots,x_d\}$0. Likewise, $P=\{x_1,\dots,x_d\}$1 is a triangular face if and only if each of $P=\{x_1,\dots,x_d\}$2, $P=\{x_1,\dots,x_d\}$3, and $P=\{x_1,\dots,x_d\}$4 is connected (Freij-Hollanti et al., 22 Sep 2025).

A general structural fact for both order and chain polytopes is that every two-dimensional face of a $P=\{x_1,\dots,x_d\}$5-polytope is either a triangle or a square. Hence every $P=\{x_1,\dots,x_d\}$6-face of $P=\{x_1,\dots,x_d\}$7 is triangular or square (Freij-Hollanti et al., 22 Sep 2025). For triangles in $P=\{x_1,\dots,x_d\}$8, there is an explicit parametrization by triples $P=\{x_1,\dots,x_d\}$9, where $\mathcal{C}(P)$0 is a connected, order-convex subposet of height at least $\mathcal{C}(P)$1, $\mathcal{C}(P)$2 is an antichain with $\mathcal{C}(P)$3, and $\mathcal{C}(P)$4 is an antichain in $\mathcal{C}(P)$5 such that $\mathcal{C}(P)$6 and $\mathcal{C}(P)$7 are connected. The corresponding triangle is

$\mathcal{C}(P)$8

with $\mathcal{C}(P)$9, $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$0, and $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$1 (Freij-Hollanti et al., 22 Sep 2025).

Counting triangular faces leads to invariants $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$2, which enumerate biconnected antichains in connected order-convex subposets. Together with companion invariants $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$3 for order polytopes and recursions on deletions of non-extremal elements, they yield the inequality

$$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$4

with equality if and only if $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$5 is $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$6-free. For square faces, the counts are exactly equal: $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$7 Consequently,

$$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$8

with equality if and only if $$x_i\ge 0 \quad (1\le i\le d), \qquad x_{i_1}+\cdots+x_{i_k}\le 1$$9 is $x_{i_1}<\cdots<x_{i_k}$0-free (Freij-Hollanti et al., 22 Sep 2025).

Beyond dimension $x_{i_1}<\cdots<x_{i_k}$1, simplex faces are governed by cliques in the $x_{i_1}<\cdots<x_{i_k}$2-skeleton. A family $x_{i_1}<\cdots<x_{i_k}$3 of antichains gives a clique in the $x_{i_1}<\cdots<x_{i_k}$4-skeleton of $x_{i_1}<\cdots<x_{i_k}$5 if and only if every pairwise symmetric difference $x_{i_1}<\cdots<x_{i_k}$6 is connected in $x_{i_1}<\cdots<x_{i_k}$7. Whenever this holds, $x_{i_1}<\cdots<x_{i_k}$8 is a $x_{i_1}<\cdots<x_{i_k}$9-simplex face (Mori, 2023). This extends the edge and triangle criteria to all clique-induced simplex faces.

4. Toric rings, triangulations, and commutative algebra

Let $P$0, and for $P$1 write $P$2. The toric ring of the chain polytope is

$P$3

graded by $P$4. Using the distributive lattice $P$5 of order ideals and the bijection $P$6, this ring is indexed by

$P$7

Hibi and Li showed that $P$8 is an algebra with straightening laws on $P$9 (Hibi et al., 2012).

For incomparable $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$00, the straightening relation is

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$01

where $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$02 is the order ideal generated by

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$03

This ASL structure implies that the toric ideal of $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$04 has a squarefree quadratic initial ideal. Consequently, every chain polytope possesses a regular unimodular triangulation arising from a flag complex (Hibi et al., 2012). The same framework yields normality, Cohen–Macaulayness, and Koszulness in the standard toric-algebraic sense recorded in that work.

A deeper layer of commutative-algebraic structure concerns the canonical module of $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$05. In the semigroup description used in the study of levelness, the canonical ideal is

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$06

and the anticanonical ideal is its inverse divisorial ideal. The level and anticanonical level properties are characterized by the absence of nonempty $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$07-reduced sequences satisfying condition $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$08. One consequence is that if $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$09 is level, or anticanonical level, then the same is true for $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$10; the converse fails, and explicit counterexamples are given (Miyazaki, 2019). The same work proves that symbolic and ordinary powers of the canonical ideal coincide,

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$11

and that the degrees of generators of the canonical and anticanonical ideals form consecutive integers (Miyazaki, 2019).

5. Marked, chain-order, and representation-theoretic generalizations

For a marked poset $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$12, the marked chain polytope is

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$13

Marked chain-order polytopes $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$14 are obtained by partitioning the unmarked elements into chain and order parts. They interpolate between the marked order polytope and the marked chain polytope, and the piecewise-affine transfer maps $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$15 generalize Stanley’s transfer map while preserving lattice points and Ehrhart polynomials (Fujita, 2021).

In type $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$16, for the Gelfand–Tsetlin poset, the marked chain end is the Feigin–Fourier–Littelmann–Vinberg polytope and the marked order end is the Gelfand–Tsetlin polytope. Every marked chain-order polytope in this family is realized as a Newton–Okounkov body of the flag variety, and the flag variety degenerates to the corresponding irreducible normal projective toric variety (Fujita, 2021). A parallel construction gives an explicit SAGBI degeneration of the flag variety to the toric variety of every marked chain-order polytope of the Gelfand–Tsetlin poset, together with standard monomial theories and PBW-type monomial bases parametrized by lattice points of these polytopes (Makhlin, 2022).

For Young diagrams, restricted chain-order polytopes are obtained by intersecting chain-order polytopes with diagonal-sum hyperplanes. For fixed $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$17, all such restricted chain-order polytopes are related by sequences of combinatorial mutations, hence have the same Ehrhart series. In particular, the paper establishes large families exhibiting period collapse and supplies explicit Ehrhart polynomials and $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$18-vectors for several non-rectangular examples (Clarke et al., 2022). This suggests that the chain-polytope framework is robust under substantial representation-theoretic and polyhedral enrichment.

6. Enriched, signed, double, and alternative chain-polytope constructions

Several later constructions preserve the antichain flavor of Stanley’s chain polytope while changing either the ambient combinatorics or the inequality system. The enriched chain polytope is

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$19

It is centrally symmetric, has the origin as unique interior lattice point, is reflexive, and admits a flag regular unimodular triangulation. Its Ehrhart polynomial is the left enriched order polynomial, and its $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$20-polynomial is $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$21-positive; more strongly, its $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$22-polynomial is the $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$23-polynomial of a flag simplicial complex (Ohsugi et al., 2018). An enriched analogue of Stanley’s transfer map gives a continuous bijection between the enriched order and enriched chain polytopes and bijectively proves Ehrhart equivalence (Okada et al., 2020).

For signed posets, the signed chain polytope $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$24 is defined by alternating signed chain inequalities

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$25

for all chains $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$26 in the signed-poset sense. The integer points of $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$27 are exactly the signed antichains, and $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$28 is reflexive. In contrast with the classical setting, the paper explicitly notes that there is no nice analogue of the order–chain equivalence in general (Beck et al., 2023).

For double posets $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$29, the double chain polytope is

$$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$30

Its facets are given explicitly by lifted chain inequalities from $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$31 and $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$32, and it admits canonical regular unimodular flag triangulations via the anti-blocking and Cayley-sum formalism (Chappell et al., 2016). This places classical chain polytopes inside a broader theory of double poset polytopes, perfect-graph anti-blocking geometry, and affine semigroup rings with quadratic Gröbner bases.

A terminological caution is necessary. In the study of pure posets, the phrase “chain polytope” is also used for the convex hull $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$33 of the $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$34-vectors of maximal chains. Its vertices are maximal chains rather than antichains, it satisfies layer-sum equalities and covering inequalities, and its Ehrhart ring is the “chain algebra” of the pure poset. This object is explicitly distinguished from Stanley’s chain polytope in that work (Lu, 2024). The coexistence of these usages has made the Stanley polytope increasingly identifiable by its antichain-vertex characterization.

The modern theory therefore treats the chain polytope as both a canonical poset polytope and a template for a broad class of variations. Stanley’s original $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$35 remains the central object: a full-dimensional $$\mathcal{C}(P)=\left\{x\in\mathbb{R}^{P}_{\ge 0} : \sum_{p\in C} x_p \le 1 \text{ for every chain } C\subseteq P\right\}.$$36-polytope with antichain vertices, order-equivalent Ehrhart theory, explicitly described low-dimensional faces, toric rings with straightening laws, and a wide extension theory reaching marked, enriched, signed, double, and representation-theoretic settings (Freij-Hollanti et al., 22 Sep 2025, Hibi et al., 2012).