PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND

ALEXANDER POSTNIKOV

Abstract. The volume and the number of lattice points of the permutohedron
Pn are given by certain multivariate polynomials that have remarkable com-
binatorial properties. We give 3 different formulas for these polynomials. We
also study a more general class of polytopes that includes the permutohedron,
the associahedron, the cyclohedron, the Stanley-Pitman polytope, and various
generalized associahedra related to wonderful compactifications of De Concini-
Procesi. These polytopes are constructed as Minkowsky sums of simplices. We
calculate their volumes and describe their combinatorial structure. The coef-
ficients of monomials in Vol Pn are certain positive integer numbers, which
we call the mixed Eulerian numbers. These numbers are equal to the mixed
volumes of hypersimplices. Various specializations of these numbers give the
usual Eulerian numbers, the Catalan numbers, the numbers (n+1)n−1 of trees
(or parking functions), the binomial coefficients, etc. We calculate the mixed
Eulerian numbers using certain binary trees. Many results are extended to an
arbitrary Weyl group.

1. Permutohedron
Let x1 , . . . , xn+1 be real numbers. The permutohedron Pn (x1 , . . . , xn+1 ) is the
convex polytope in Rn+1 defined as the convex hull of all permutations of the vector
(x1 , . . . , xn+1 ):

Pn (x1 , . . . , xn+1 ) := ConvexHull((xw(1) , . . . , xw(n+1) ) | w ∈ Sn+1 ),

where Sn+1 is the symmetric group. Actually, this is an n-dimensional polytope

that belongs to some hyperplane H ⊂ Rn+1 . More generally, for a Weyl group W ,
we can define the weight polytope as a convex hull of a Weyl group orbit:

PW (x) := ConvexHull(w(x) | w ∈ W ),

where x is a point in the weight space on which the Weyl group acts.
For example, for n = 2 and distinct x1 , x2 , x3 , the permutohedron P2 (x1 , x2 , x3 )
(type A2 weight polytope) is the hexagon shown below. If some of the numbers
x1 , x2 , x3 are equal to each other then the permutohedron degenerates into a trian-
gle, or even a single point.

Date: November 21, 2004; based on transparencies dated July 26, 2004.
Key words and phrases. Permutohedron, associahedron, Minkowsky sum, mixed volume,
Weyl’s character formula, Hall’s marriage theorem, wonderful compactification, nested families,
generalized Catalan numbers, Stanley-Pitman polytope, parking functions, hypersimplices, mixed
Eulerian numbers, binary trees, shifted tableaux, Gelfand-Zetlin patterns.
1
2 ALEXANDER POSTNIKOV

(x1 , x2 , x3 )

P2 (x1 , x2 , x3 ) =

Question: What is the volume of the permutohedron Vn := Vol Pn ?

Since Pn does not have the full dimension in Rn+1 , one needs to be careful with
definition of the volume. We assume that the volume form Vol is normalized so
that the volume of a unit parallelepiped formed by generators of the integer lattice
Zn+1 ∩ H is 1. More generally, we can ask the following question.
Question: What is the number of lattice points Nn := Pn ∩ Zn+1 ?
We will see that both Vn and Nn are polynomials of degree n in the variables
x1 , . . . , xn+1 . The polynomial Vn is the top homogeneous part of Nn . The Ehrhart
polynomial of Pn is E(t) = Nn (tx1 , . . . , txn ), and Vn is its top degree coefficient.
We will give 3 totally different formulas for these polynomials.
Let us first mention the special case (x1 , . . . , xn+1 ) = (n+1, . . . , 1). The polytope
Pn (n + 1, n, . . . , 1) = ConvexHull((w(1), ..., w(n + 1)) | w ∈ Sn+1 )
is the most symmetric permutohedron. It is invariant under the action of the
symmetric group Sn+1 . For example, for n = 2, it is the regular hexagon:

regular hexagon

subdivided into 3 rhombi

The polytope Pn (n + 1, . . . , 1) is a zonotope, i.e., Minkowsky sum of line intervals.

It is well known that
• Vn (n + 1, . . . , 1) = (n + 1)n−1 is the number of trees on n + 1 labelled
vertices. Indeed, Pn (n + 1, . . . , 1) can be subdivided into parallelepipeds of
unit volume associated with trees. This follows from a general result about
zonotopes.
• Nn (n + 1, . . . , 1) is the number of forests on n + 1 labelled vertices.
In general, for arbitrary x1 , . . . , xn+1 , the permutohedron Pn (x1 , . . . , xn+1 ) is
not a zonotope. We cannot easily calculate its volume by subdividing it into par-
allelepipeds.
 PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND 3

2. First Formula
Theorem 2.1. Fix any distinct numbers λ1 , . . . , λn+1 such that λ1 +· · ·+λn+1 = 0.
We have
1 X (λw(1) x1 + · · · + λw(n+1) xn+1 )n
Vn (x1 , . . . , xn+1 ) =
n! (λw(1) − λw(2) )(λw(2) − λw(3) ) · · · (λw(n) − λw(n+1) ).
w∈Sn+1

Notice that all λi ’s in the right-hand side are canceled after the symmetrization.
More generally, let W be the Weyl group associated with a rank n root system,
and let α1 , . . . , αn be a choice of simple roots.

Theorem 2.2. Let λ be any regular weight. The volume of the weight polytope is

f X (λ, x)n
Vol PW (x) = ,
|W | (λ, w(α1 )) · · · (λ, w(αn ))
w∈W

where the volume is normalized so that the volume of the parallelepiped generated by
the simple roots αi is 1, and f is the index of the root lattice in the weight lattice.

Idea of Proof. Let us use Khovansky-Pukhlikov’s method [PK]. Instead of counting

the number of lattice points in P , calculate the sum Σ[P ] of formal exponents ea over
lattice points a ∈ P ∩ Zn . We can work with unbounded polytopes. For example,
for a simplicial cone C, the sum Σ[C] is given by a simple rational expression. Any
polytope P can be explicitly presented as an alternating sum of simplicial cones
Σ[P ] = Σ[C1 ] ± Σ[C2 ] ± · · · over vertices of P .
Applying this method to the weight polytope, we obtain the following claim.

Theorem 2.3. For a dominant weight µ, the sum of exponents over lattice points
of the weight polytope PW (µ) equals

X X ew(µ)
Σ[PW (µ)] := ea = ,
w∈W
(1 − e−w(α1 ) ) · · · (1 − e−w(αn ) )
a∈PW (µ)∩(L+µ)

where L be the root lattice.

Compare this claim with Weyl’s character formula! If we replace the product
over simple roots αi in the right-hand side of Theorem 2.3 by a similar product
over all positive roots, we obtain exactly Weyl’s character formula.
Theorem 2.2 and its special case Theorem 2.1 and be deduced from Theorem 2.3
in the same way as Weyl’s dimension formula can be deduced from Weyl’s character
formula. 

Remark 2.4. The sum of exponents Σ[PW (µ)] over lattice points of the weight
polytope is obtained from the character ch Vµ of the irreducible representation Vµ
of the associated Lie group by replacing all nonzero coefficients in ch Vµ with 1.
For example, in type A, ch Vµ is the Schur polynomial sµ and Σ[Pn (µ)] is obtained
from the Schur polynomial sµ by erasing the coefficients of all monomials.
4 ALEXANDER POSTNIKOV

3. Second Formula
Let us use the coordinates y1 , . . . , yn+1 related x1 , . . . , xn+1 by


 y1 = −x1
 y2 = −x2 + x1


y3 = −x3 + 2x2 − x1
 ···


yn+1 = − n0 xn + n1 xn−1 − · · · ± nn x1



Write Vn = Vol Pn (x1 , . . . , xn+1 ) as a polynomial in the variables y1 , . . . , yn+1 .

Theorem 3.1. We have
1 X
Vn = y|S1 | · · · y|Sn| ,
n!
(S1 ,...,Sn )

where the sum is over ordered collections of subsets S1 , . . . , Sn ⊂ [n + 1] such that,

for any distinct i1 , . . . , ik , we have |Si1 ∪ · · · ∪ Sik | ≥ k + 1.
This theorem implies that n! Vn is a polynomial in y2 , . . . , yn+1 with positive
integer coefficients. For example, V1 = Vol ([(x1 , x2 ), (x2 , x1 )]) = x1 − x2 = y2 and
2V2 = 6 y22 + 6 y2 y3 + y32 .
Remark 3.2. The condition on subsets S1 , . . . , Sn in Theorem 3.1 is very similar to
the condition in Hall’s marriage theorem. One just needs to replace the inequality
≥ k + 1 with ≥ k to obtain Hall’s marriage condition.
It is not hard to prove the following analog of Hall’s theorem.
Dragon marriage problem: There are n + 1 brides and n grooms living in a
medieval village. A dragon comes to the village and takes one of the brides. We
are given a collection G of pairs of brides and grooms that can marry each other.
Find the condition on G that ensures that, no matter which bride the dragon takes,
it will be possible to match the remaining brides and grooms.
Proposition 3.3. (Dragon marriage theorem) Let S1 , . . . , Sn ⊂ [n + 1]. The fol-
lowing three conditions are equivalent:
(1) For any distinct i1 , . . . , ik , we have |Si1 ∪ · · · ∪ Sik | ≥ k + 1.
(2) For any j ∈ [n+1], there is a system of distinct representatives in S1 , . . . , Sn
that avoids j. (This is a reformulation of the dragon marriage problem.)
(3) One can remove some elements from Si ’s to get the edge set of a spanning
tree in Kn+1 .

Theorem 3.1 can be extended to a larger class of polytopes discussed below.

4. Generalized permutohedra
Let ∆[n+1] = ConvexHull(e1 , . . . , en+1 ) be the standard coordinate simplex in
Rn+1 . For a subset I ⊂ [n + 1], let ∆I = ConvexHull(ei | i ∈ I) denotes the face of
the coordinate simplex ∆[n+1] . Let Y = {YI } be a collection of parameters YI ≥ 0
for all nonempty subsets I ⊂ [n + 1]. Let us define the generalized permutohedron
Pn (Y ) as the Minkowsky sum of the simplices ∆I scaled by factors YI :
X
Pn (Y) := YI · ∆I (Minkowsky sum)
I⊂[n+1]
 PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND 5

Generalized permutohedra are obtained from usual permutohedra by moving

their faces while preserving all angles. So, instead of n degrees of freedom in usual
permutohedra, we have 2n+1 − 2 degrees of freedom in generalized permutohedra.

a generalized permutohedron

this is also

a generalized permutohedron

The combinatorial type of Pn (Y) depends only on the set of B ⊂ 2[n+1] of I’s
for which YI 6= 0. Here are some interesting examples of generalized permutohedra.
• If YI = y|I| , i.e., the variables YI are equal to each other for all subsets of
the same cardinality, then Pn (Y) is the usual permutohedron Pn .
• If B = {{i, i + 1, . . . , j} | 1 ≤ i ≤ j ≤ n} is the set of consecutive intervals,
then Pn (Y) is the associahedron, also known as the Stasheff polytope. The
polytope Pn (Y) can be equivalently defined as the Newton polytope of
Q
1≤i≤j≤n+1 (xi + xi+1 + · · · + xj ). This is exactly Loday’s realization of the
associahedron, see [L].
• If B is the set of cyclic intervals, then Pn (Y) is a cyclohedron.
• If B is the set of connected subsets of nodes of a Dynkin diagram, then
Pn (Y) the polytope related to De Concini-Procesi’s wonderful compactifi-
cation, see [DP], [DJS].
• If B = {[i] | i = 1, . . . , n + 1} is the complete flag of initial intervals, then
Pn (Y) is the Stanley-Pitman polytope from [SP].
Theorem 3.1 can be extended to generalized permutohedra, as follows.

Theorem 4.1. The volume of the generalized permutohedron Pn (Y) is given by

1 X
Vol Pn (Y) = YS 1 · · · Y S n ,
n!
(S1 ,...,Sn )

where the sum is over ordered collections of subsets S1 , . . . , Sn ⊂ [n + 1] such that,

for any distinct i1 , . . . , ik , we have |Si1 ∪ · · · ∪ Sik | ≥ k + 1.
6 ALEXANDER POSTNIKOV

Theorem 4.2. The number of lattice points in the generalized permutohedron

Pn (Y) is
1 X
Pn (Y) ∩ Zn+1 = {YS1 · · · YSn },
n!
(S1 ,...,Sn )
where the summation is over the same collections (S1 , . . . , Sn ) as before, and
( )
{a }
Y Y
YIaI
:= (Y[n+1] +1){a[n+1] } YI I , where Y {a} = Y (Y +1) · · · (Y +a−1).
I I6=[n+1]

In other words, the formula for the number of lattice points in Pn (Y) is obtained
from the formula for the volume by replacing usual powers in all terms by raising
powers.
These formulas generalize formulas from [SP] for the volume and the number of
lattice points in the Stanley-Pitman polytope. In this case, collections (S1 , . . . , Sn )
of initial intervals Si = [si ] that satisfy the dragon marriage condition, see Proposi-
tion 3.3, are in one-to-one correspondence with parking functions (s1 , . . . , sn ). The
volume Pn (Y) is given by the sum over parking functions.

5. Nested families and generalized Catalan numbers

In this section, we describe the combinatorial structure of the class of generalized
permutohedra Pn (Y) such that the set B ⊂ 2[n+1] of subsets I with nonzero YI
satisfies the following connectivity condition:
(B1) If I, J ∈ B and I ∩ J 6= ∅, then I ∪ J ∈ B.
All examples of generalized permutohedra mentioned in the previous section satisfy
this additional condition. Without loss of generality we will also assume that
(B2) B contains all singletons {i}, for i ∈ [n + 1].
Indeed, the Minkowsky sum of a polytope with ∆{i} , which is a single point, is just
a parallel translation of the polytope.
Sets B ⊂ 2[n+1] that satisfy conditions (B1) and (B2) are called building sets.
Note that condition (B1) implies that all maximal by inclusion elements in B are
pairwise disjoint.
Let us say that a subset N in a building set B is a nested family if it satisfies
the following conditions:
(N1) For any I, J ∈ N , we have either I ⊆ J, or J ⊆ I, or I and J are disjoint.
(N2) For any collection of k ≥ 2 disjoint subsets I1 , . . . , Ik ∈ N , their union
I1 ∪ · · · ∪ Ik is not in B.
(N3) N contains all maximal elements of B.
Let N (B) be the poset of all nested families in B ordered by reverse inclusion.
Theorem 5.1. The poset of faces of the generalized permutohedron P n (Y) ordered
by inclusion is isomorphic to N (B).
This claim was independently discovered by E. M. Feichtner and B. Sturm-
fels [FS].
For a graph G on the set of vertices [n + 1], let BG be the collection of nonempty
subsets I ⊂ [n + 1] such that the induced graph G|I is connected. Then BG satisfies
conditions (B1) and (B2) of a building set. The generalized permutohedron associ-
ated with BG is combinatorially equivalent to the graph associahedron constructed
 PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND 7

by Carr and Devadoss [CD] using blow-ups. In this case, it is enough to require
condition (N2) only for pairs of subsets, in the definition of a nested family.
Remark 5.2. Since our generalized permutohedra include the associahedron, one
can also call them generalized associahedra. However this name is already reserved
for a different generalization of the associahedron studied by Fomin, Chapoton, and
Zelevinsky [FCZ].
Let fB (q) be the f -polynomial of the generalized permutohedron:
n
X X
fB (q) = fi q i = q n+1−|N | ,
i=0 N ∈N (B)

where fi is the number of i-dimensional faces of Pn (Y).

Let us say that a building set B is connected if it has a unique maximal element.
Each building set B is a union of pairwise disjoint connected building sets, called
the connected components of B. For a subset S ⊂ [n + 1], the induced building set is
defined as B|S = {I ∈ B | I ⊂ S}. In the case of building sets BG associated with
graphs G, notions of connected components and induced building sets correspond
to similar notions for graphs.
Theorem 5.3. The f -polynomial fB (q) is determined by the following recurrence
relations:
(1) If B consists of a single singleton, then fB (q) = 1.
(2) If B has connected components B1 , . . . , Bk , then
fB (q) = fB1 (q) · · · fBk (q).
(3) If B ⊂ 2[n+1] is a connected building and n ≥ 1, then
X
fB (q) = q n−|S| fB|S (q).
S([n+1]

Define the generalized Catalan number, for a building set B, as the number
C(B) = fB (0) of vertices of the corresponding generalized permutohedron Pn (Y).
These numbers are given by the recurrence relations similar to the relations in
Theorem 5.3. (But in (3) we sum only over subsets S ⊂ [n + 1] of cardinality n).
For a graph G, let C(G) = C(BG ). Let Tpqr be the graph that has a central
node with 3 attached chains of with p, q and r vertices. For example, T111 is the
Dynkin diagram of type D4 . The above recurrence relations implies the following
expression for the generating function for generalized Catalan numbers:
X C(x) C(y) C(z)
C(Tpqr ) xp y q z r = ,
1 − x C(x) − y C(y) − z C(z)
p,q,r≥0
√
P n 1− 1−4x
where C(x) = n≥0 Cn x = 2x is the generating function for the usual
Catalan numbers.
Proposition 5.4. For the Dynkin diagram of type An , we have C(An ) = Cn =
1 2n


n+1 n is the usual Catalan number. For the extended Dynkin diagram of type
Ân , we have C(Ân ) = 2n


n . For the Dynkin diagram of type Dn , the corresponding
Catalan number is
C(Dn ) = 2 Cn − 2 Cn−1 − Cn−2 .
8 ALEXANDER POSTNIKOV

Remark 5.5. One can define the generalized Catalan number for any Lie type. How-
ever this number does not depend on multiplicity of edges in the Dynkin diagram.
Thus Lie types Bn and Cn give the same (usual) Catalan number as type An .

6. Mixed Eulerian numbers

Let us return to the usual permutohedron Pn (x1 , . . . , xn+1 ). Let us use the
coordinates z1 , . . . , zn related to x1 , . . . , xn+1 by
z1 = x1 − x2 , z2 = x2 − x3 , · · · , zn = xn − xn+1
This coordinate system is canonically defined for an arbitrary Weyl group as the
coordinate system in the weight space given by the fundamental weights.
The permutohedron Pn can be written as the Minkowsky sum
Pn = z1 ∆1n + z2 ∆2n + · · · + zn ∆nn
of the hypersimplices ∆kn = Pn (1, . . . , 1, 0, . . . , 0) (with k 1’s). For example, the
hexagon can be express as the Minkowsky sum of the hypersimplices ∆12 and ∆22 ,
which are two triangles with opposite orientations:

+ =

This implies that the volume of Pn can be written as

X z c1 z cn
Vol Pn = Ac1 ,...,cn 1 · · · n ,
c ,...,c
c1 ! cn !
1 n

where the sum is over c1 , . . . , cn ≥ 0, c1 + · · · + cn = n, and

Ac1 ,...,cn = MixedVolume(∆c1n
1
, . . . , ∆cnn
n
) ∈ Z>0
is the mixed volume of hypersimplices. In particular, n! Vn is a positive integer
polynomial in z1 , . . . , zn . Let us call the integers Ac1 ,...,cn the Mixed Eulerian
numbers.
Examples: The mixed Eulerian numbers are marked in bold.
V1 = 1 z 1 ;
z2 z2
V2 = 1 21 + 2 z1 z2 + 1 22 ;
z3 z2 z2 z3 z2
V3 = 1 3!1 + 2 21 z2 + 4 z1 22 + 4 3!2 + 3 21 z3 + 6 z1 z2 z3 +
z2 z2 z2 z3
+4 22 z3 + 3 z1 23 + 2 z2 23 + 1 3!3 .
Theorem 6.1. Mixed Eulerian numbers have the following properties:
(1) A
Pc1 ,...,c n are positive integers defined for c 1 , . . . , cn ≥ 0, c1 + · · · + cn = n.
1 n−1
(2) c1 !···cn ! Ac1 ,...,cn = (n + 1) .
(3) A0,...,0,n,0,...,0 (n in the k-th position) is the usual Eulerian number Akn ,
i.e., the number of permutations in Sn with k descents.
 PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND 9

(4) A0,...,0,i,n−i,0,...,0 (with k 0’s in front of i, n − i) is equal to the number of

permutations w ∈ Sn+1 with k descents and w(n + 1) = i + 1.
(5) A1,...,1 = n!
(6) Ak,0,...,0,n−k = nk

(7) Ac1 ,...,cn = 1c1 2c2 · · · ncn if c1 +
· · · + ci ≥ i, for i = 1, . . . , n.

1 2n
There are exactly Cn = n+1 n such sequences (c1 , . . . , cn ).
P
(8) Ac1 ,...,cn = n! Cn .
Property (3) follows from the well-know fact that Akn = n! Vol ∆kn ; and prop-
erty (4) follows from the result of [ERS] about the mixed volume of two adjacent
hypersimplices. Property (8) was numerically noticed by Richard Stanley. More-
over, he conjectured the following claim.
Theorem 6.2. Let us write (c1 , . . . , cn ) ∼ (c01 , . . . , c0n ) whenever (c1 , . . . , cn , 0) is a
cyclic shift of (c01 , . . . , c0n , 0). Then, for fixed (c1 , . . . , cn ), we have
X
Ac01 ,...,c0n = n!
(c01 ,...,c0n )∼(c1 ,...,cn )

In other words, the sum of mixed Eulerian numbers

in each equivalence class is n!.
1 2n
There are exactly the Catalan number n+1 n equivalence classes of sequences.

P + A0,0,n,...,0 + · · · +
For example, we have A1,...,1 = n! and An,0,...,0 + A0,n,0,...,0
A0,...,0,n = n!, because the sum of usual Eulerian numbers k Akn is n!.
Remark 6.3. Every equivalence class contains exactly one sequence (c 1 , . . . , cn ) such
that c1 + · · · + ci ≥ i, for i = 1, . . . , n. For this special sequence, the mixed Eulerian
number is given by the simple product Ac1 ,...,cn = 1c1 · · · ncn .
Theorem 6.2 follows from the following claim.
Theorem 6.4. Let Ûn (z1 , . . . , zn+1 ) = Vol Pn . (It does not depend on zn+1 .)
Ûn (z1 , . . . , zn+1 ) + Ûn (zn+1 , z1 , . . . , zn ) + · · · + Ûn (z2 , . . . , zn+1 , z1 ) =
= (z1 + · · · + zn+1 )n
This claim extends to any Weyl group. It has has a simple geometric proof using
alcoves of the associated affine Weyl group. Cyclic shifts come from symmetries of
type A extended Dynkin diagram.
Idea of Proof. The volume of the fundamental alcove times |W | equals the sum of
volumes of n + 1 adjacent permutohedra. For example, the 6 areas of the blue
triangle on the following picture is the sum of the areas of three hexagons.
10 ALEXANDER POSTNIKOV


Corollary 6.5. Fix z1 , . . . , zn+1 , λ1 , . . . , λn+1 such that λ1 + · · · + λn+1 = 0.
Symmetrizing the expression
1 (λ1 z1 + (λ1 + λ2 )z2 + · · · (λ1 + · · · + λn+1 )zn+1 )n
n! (λ1 − λ2 ) · · · (λn − λn+1 )
with respect to (n+1)! permutations of λ1 , . . . , λn+1 and (n+1) cyclic permutations
of z1 , . . . zn+1 , we obtain
(z1 + · · · + zn+1 )n .
It would be interesting to find a direct proof of this claim.

7. Third formula
Let us give a combinatorial interpretation for the mixed Eulerian numbers based
on plane binary trees.
Let T be an increasing plane binary tree with n nodes labelled 1, . . . , n. It is
well-known that the number of such trees is n!. Let vi be the node of T labelled
i, for i = 1, . . . , n. In particular, v1 is the top node of T . Let us define a different
labelling of the nodes v1 , . . . , vn of T by numbers d1 , . . . , dn ∈ [n] based on the
deep-first search algorithm. This labelling is uniquely characterized by the following
condition: For any node vi in T and any vj in the left (respectively, right) branch
of vi , we have dj < di (respectively, di < dj ). In particular, for the left-most node
vl in T , we have dl = 1 and, for the right-most node vr , we have dr = n. Then,
for any node vi , the numbers dj , for all descendants vj of vi (including vi ), form a
consecutive interval [li , ri ] of integers. (In particular, li ≤ di ≤ ri .)
Remark 7.1. For a plane binary tree T , the collection N of intervals [li , ri ], i =
1, . . . , n, is a maximal nested family for the building set B formed by all consecutive
intervals in [n], i.e., the building set for the usual associahedron, see Section 5.
The map T 7→ N , is a bijection between plane binary trees and vertices of the
associahedron in Loday’s realization [L].
Let i = (i1 , . . . , in ) ∈ [n]n be a sequence of integers. Let us say that an increasing
plane binary tree T is i-compatible if, ik ∈ [lk , rk ], for k = 1, . . . , n. For a node vk
in such a tree, define its weight as
( ik −lk +1
dk −lk +1 if ik ≤ dk
wt(vk ) =
rk −ik +1
rk −dk +1 if ik ≥ dk
Define the i-weight of an i-compatible tree T as
n
Y
wt(i, T ) = wt(vk ).
k=1

It is not hard to see that n! wt(i, T ) is always a positive integer.

Here is an example of an i-compatible tree T , for i = (3, 4, 8, 7, 1, 7, 4, 3). The
labels dk of the nodes vk are 5, 2, 6, 8, 1, 7, 4, 3. (They are shown on picture in blue
color.) The intervals [lk , rk ] are [1, 8], [1, 4], [6, 8], [7, 8], [1, 1], [7, 7], [3, 4], [3, 3].
We also marked each node vk by the variable zik . The i-weight of this tree is
wt(i, T ) = 35 · 31 · 31 · 12 · 11 · 11 · 22 · 11 .
 PERMUTOHEDRA, ASSOCIAHEDRA, AND BEYOND 11

z3
v1 5
an i-compartible increasing plane binary tree
z4 z8
v2 6 v3
2
T = z1 z4 z7
v4
v5 1 4 v7 8
i = (3, 4, 8, 7, 1, 7, 4, 3) z3
v8 z7
3 7 v6

Theorem 7.2. The volume of the permutohedron is equal

X X
Vn = z i1 · · · z in wt(i, T ).
i∈[n]n T is i-compartible
n
where the first sum is over n sequences i = (i1 , . . . , in ) and second sum is over
i-compatible increasing plane binary trees with n nodes.
Let us give a combinatorial interpretation for the mixed Eulerian numbers.
Theorem 7.3. Let (i1 , . . . , in ) be any sequence such that zi1 · · · zin = z1c1 · · · zncn .
Then X
Ac1 ,...,cn = n! wt(i, T ),
T is i-compartible
where the sum is over i-compatible increasing plane binary trees with n nodes.
Note that all terms n! wt(T ) in this formula are positive integers. Actually, this
n


theorem gives not just one but c1 ,...,c n
different combinatorial interpretations of
the mixed Eulerian numbers Ac1 ,...,cn for each way to write z1c1 · · · zncn as zi1 · · · zin .
The proof of this theorem is based on the following recurrence relation for the
volume of the permutohedron. Let us write Vol Pn as a polynomial Un (z1 , . . . , zn ).
Proposition 7.4. For any i = 1, . . . , n, we have
n  
∂ X n+1
Un (z1 , . . . , zn ) = wti,k,n Uk−1 (z1 , . . . , zk−1 ) Un−k−1 (zk+1 , . . . , zn ),
∂zi k
k=1
where (
1 i(n − k) if 1 ≤ i ≤ k;
wti,k,n =
n k(n − k) if k ≤ i ≤ n.
Idea of Proof. The derivative ∂Un /∂zi is the rate of change of the volume Vol Pn
of the permutohedron as we move its generating vertex in the direction of the
coordinate zk . (More generally, in the direction of the k-th fundamental weight.)
It can be written as a sum of areas of facets of Pn scaled by some factors. Each
facet of Pn is
a product Pk−1 × Pn−k−1 of two smaller permutohedra. There are
exactly n+1
k facets like this. The corresponding factor wti,k,n tells how fast the
facet moves as we shift the generating vertex of Pn .
This formula can be extended the weight polytope for an arbitrary Weyl group
W . In general, the coefficient wti,k,n equals the inner product of two fundamental
weights (ωi , ωk ). 
Comparing different formulas for Vn , we obtain many interesting combinatorial
identities. For example, we have the following claim.
12 ALEXANDER POSTNIKOV

Corollary 7.5. We have

X n! Y  1

(n + 1)n−1 = 1 + ,
2n h(v)
T v∈T

where is sum is over unlabeled plane binary trees T on n nodes, and h(v) denotes
the “hook-length” of a node v in T , i.e., the number of descendants of the node v
(including v).
Example: n = 3
3 3 3 3 3

2 2 2 2
1 1
1 1 1 1

hook-lengths of binary trees

The identity says that
(3 + 1)2 = 3 + 3 + 3 + 3 + 4.
This identity was combinatorially proved by Seo [S].

References
[CD] M. Carr, S. Devadoss: Coxeter complexes and graph associahedra, math.QA/0407229.
[DJS] M. Davis, T. Januszkiewicz, R. Scott: Nonpositive curvature of blow-ups. Selecta Math.
(N.S.) 4 (1998), no. 4, 491–547.
[DP] C. De Concini, C. Procesi: Wonderful models for subspace arrangements, Selecta Math.
(N.S.) 1 (1995), 459–494.
[ERS] R. Ehrenborg, M. Readdy, E. Steingrı́msson: Mixed volumes and slices of the cube, J.
Combin. Theory Ser. A 81 (1998), no. 1, 121–126.
[FS] E. M. Feichtner, B. Sturmfels: Matroid polytopes, nested sets and Bergman fans,
math.CO/0411260.
[FCZ] S. Fomin, F. Chapoton, A. Zelevinsky: Polytopal realizations of generalized associahedra,
math.CO/0202004.
[L] J.-L. Loday: Realization of the Stasheff polytope, math.AT/0212126.
[PK] A. V. Pukhlikov, A. G. Khovanskii: Finitely additive measures of virtual polyhedra, St. Pe-
tersburg Math. J. 4 (1993), no. 2, 337–356.
[S] S. Seo: A combinatorial proof of Postnikov’s identity and a generalized enumeration of
labeled trees, math.CO/0409323.
[SP] R. P. Stanley, J. Pitman: A polytope related to empirical distributions, plane trees, parking
functions, and the associahedron, Discrete Comput. Geom. 27 (2002), no. 4, 603–634.

Department of Mathematics, M.I.T., Cambridge, MA 02139

E-mail address: apost (at) math (dot) mit (dot) edu
URL: http://www.math-mit.edu/~apost/