Polytopes, Hopf algebras and Quasi-symmetric functions

Victor M. Buchstaber Nickolai Erokhovets

Abstract
In this paper we use the technique of Hopf algebras and quasi-symmetric functions to study
arXiv:1011.1536v1 [math.CO] 6 Nov 2010

the combinatorial polytopes. Consider the free abelian group P generated by all combinatorial
polytopes. There are two natural bilinear operations on this group defined by a direct product ×
and a join > of polytopes. It turns out that (P, ×) is a commutative associative bigraded ring of
polynomials with graduations (2n, 2(m − n)), where n is the dimension, and m is the number of
facets of a polytope. (P, >) is a commutative associative ring without a unit. If we add the empty set
∅ as a (−1)-dimensional polytope, then it gives a unit, and we obtain the commutative associative
threegraded ring of polynomials RP with graduations (2(n + 1), 2(v − n − 1), 2(m − n − 1)), where
v is the number of vertices of a polytope. The ring RP has the structure of a graded Hopf algebra
and the natural correspondence P → L(P ) that sends a polytope to its face lattice embeds RP into
the Rota-Hopf algebra R of graded posets as a Hopf subalgebra. It turns out that P has a natural
Hopf comodule structure over the Hopf algebra RP. Faces operators dk that send a polytope to
the sum of all its (n − k)-dimensional faces define on both rings the Hopf module structures over
the Hopf algebra U that is universal in the category of Leibnitz-Hopf algebras with the antipode
Uk → (−1)k Uk . This structure gives a ring homomorphism R → Qsym ⊗ R, where R is P or
RP. Composing this homomorphism with the characters P n → αn of P, P n → αn+1 of RP, and
with the counit we obtain the ring homomorphisms f : P → Qsym[α], fRP : RP → Qsym[α], and
F∗ : RP → Qsym, where F is the Ehrenborg transformation. We describe the images of these
homomorphisms in terms of functional equations, prove that these images are rings of polynomials
over Q, and find the relations between the images, the homomorphisms and the Hopf comodule
structures. For each homomorphism f, fRP , and F the images of two polytopes coincide if and only
if they have equal flag f -vectors. Therefore algebraic structures on the images give the information
about flag f -vectors of polytopes. The homomorphism f is an isomorphism on the graded group
BB generated by the polytopes introduced by M. Bayer and L. Billera to find the linear span of flag
f -vectors of convex polytopes. This gives the group BB a structure of the ring isomorphic to f (P).
Developing this approach one can build new combinatorial invariants of convex polytopes that can
not be expresses in terms of flag numbers.

1 Introduction
Convex polytopes is a classical object of convex geometry. In recent times the solution of problems
of convex geometry involves results of algebraic geometry and topology, commutative and homological
algebra. There are remarkable results lying on the crossroads of the polytope theory, theory of complex
manifolds, equivariant topology and singularity theory (see the survey [BR]). The example of successful
collaboration of the polytope theory and differential equations can be found in [Buch].
To study the combinatorics of convex polytopes we consider the free abelian group P generated
by all combinatorial convex polytopes. It turns out that this object has a rich algebraic nature. The
direct product × defines on P the structure of a bigraded associative commutative ring (P, ×) with
graduations (2n, 2(m − n)), where n is the dimension, and m is the number of facets of a polytope.
The bigraded structure gives the decomposition of P into a direct sum of finitely generated free abelian
groups
X m−1X
P = P0 + P 2n, 2(m−n) .
m>2 n=1

The duality operator ∗ defines on P another multiplication P ◦ Q = (P ∗ × Q∗ )∗ .

The join of two polytopes P > Q is a bilinear operation of degree +2. Then (P, >) is a commutative
associative ring without a unit. Set RP = Z∅ ⊕ P. The empty set ∅ may be considered as a (−1)-

1
dimensional polytope such that ∅ > P = P . Then the free abelian group RP has the structure of a
threegraded commutative associative ring with graduations (2(n + 1), 2(v − n − 1), 2(m − n − 1)), where
v is the number of vertices of a polytope. It turns out that RP is a graded Hopf algebra with graduation
2(n + 1) and the comultiplication
X
∆(P ) = F ⊗ (P/F )
∅⊆F ⊆P

The correspondence P → L(P ), where L(P ) is a face lattice of a polytope P defines an embedding
of RP as a Hopf subalgebra into the Hopf algebra R of graded posets introduces by Joni and Rota in
[JR].
In the focus of our interest we put the ring of linear operators L(R) = HomZ (R, R), where R = P
or RP, and its subring D generated by the operators dk , k > 1 that map an n-dimensional polytope to
the sum of all its (n − k)-dimensional faces. d = d1 is a derivation, so R is a differential ring.
The ring D has a canonical structure of a Leibnitz-Hopf algebra (see Section 3.2). We prove that D
is isomorphic to the universal Hopf algebra in the category of Leibnitz-Hopf algebras with the antipode
χ(dk ) = (−1)k dk .
D ≃ Z/JD
where Z = ZhZ1 , Z2 , . . . i is the universal Leibnitz-Hopf algebra – the free associative Hopf algebra with
the comultiplication X
∆Zk = Zi ⊗ Zj , Z0 = 1,
i+j=k
P
and JD is the two-sided Hopf ideal generated by the relations (−1)i Zi Zj = 0, n > 1.
i+j=n
This algebra appears in various application of theory of Hopf algebras in combinatorics: over the
rationals it is isomorphic to the graded dual of the odd subalgebra S− (Qsym[t1 , t2 , . . . ], ζQ ) [ABS,
Remark 6.7], to the algebra of forms on chain operators [BLiu, Proposition 3.2], to the factor algebra of
the algebra of Piery operators on a Eulerian poset by the ideal generated by the Euler relations [BMSW,
Example 5.3].
The action of D on R satisfies the property Dω (P Q) = µ ◦ (∆Dω )(P ⊗ Q), so the ring R has the
structure of a Milnor module over D.
The graded dual Hopf algebra to Z is the algebra of quasi-symmetric functions Qsym[t1 , t2 , . . . ]. The
character ξα : P → Z[α], P n → αn induces the homomorphism P → D∗ [α], and the composition with
the embedding D∗ ⊂ Qsym[t1 , t2 , . . . ] gives the ring homomorphism

f : P → Qsym[t1 , t2 , . . . ][α]

We call f (α, t1 , t2 , . . . ) the generalized f -polynomial. In can be shown that f (α, t1 , 0, 0, . . . ) = f1 (α, t)
is a homogeneous f -polynomial in two variables introduced in [Buch]. For simple polytopes we have
f (α, t1 , t2 , . . . ) = f1 (α, t1 + t2 + . . . ).
The homomorphism f can be considered in a more general context. The structure of a right
Milnor module on the ring R over the universal Hopf algebra Z defines a left Hopf comodule struc-
ture R → Qsym ⊗ R. Then any ring homomorphism R → R1 defines the ring homomorphism
R → Qsym⊗R1 . In particular, the character P n → αn defines the ring homomorphism f : P → Qsym[α],
the character εα (P n ) = αn+1 defines a ring homomorphism fRP : RP → Qsym[α], the counit ε defines
a ring homomorphism F∗ : RP → Qsym such that F(P ) is the Ehrenborg F -quasi-symmetric function of
a graded poset L(P ). The image of two polytopes under any of homomorphisms f, fRP , or F∗ coincide
if and only if the polytopes have equal flag f -numbers.
It turns out that there is a nice algebraic structure including the rings P and RP. Namely, the
formula X
∆RP P = F ⊗ (P/F )
F ⊆P

defines on the ring P the structure of a graded Hopf comodule over the Hopf algebra RP with respect to
graduations defined by dimension of a polytope. This approach clarifies the structures mentioned above
and allows us to build new ring homomorphisms from the ring of polytopes P to rings of polynomials
such that the image of a polytope is not determined by its flag f -vector. On this way one can build new

2
combinatorial invariants of polytopes that can not be expressed in terms of flag numbers. This will be
the topic of our next article.
The paper has the following structure:
Section 2 describes the main definitions and constructions involving convex polytopes.
- part 2.1 contains the main definitions and considers the basic constructions involving polytopes;
- in part 2.2 the ring of polytopes P is defined.
- in part 2.3 we establish the structure of a graded Hopf algebra on the ring RP and consider it as
a Hopf subalgebra of the Rota-Hopf algebra R.
- part 2.4 is devoted to linear operators defined on the rings P and RP. In particular, the faces
operators dk are defined.
- part 2.5 recalls important facts about flag f -vectors and emphasizes some peculiarities we need in
this paper.
Section 3 is devoted to necessary definitions and important facts about quasi-symmetric functions
and Hopf algebras.
- part 3.1 contains the definition of quasi-symmetric functions;
- part 3.2 includes the definition of Leibnitz-Hopf algebras;
- part 3.3 – of Lie-Hopf algebras;
- part 3.4 describes the idea of the proof of the shuffle-algebra theorem.
Section 4 contains topological realizations of the Hopf algebras we study.
In Section 5 the structure theorem for the ring D is proved.
In Section 6 we define and study the generalized f -polynomial.
- part 6.1 contains the construction of f ;
- in part 6.2 we find the functional equations that describe the image of the ring P in Qsym[t1 , t2 , . . . ][α].
These equations are equivalent to the generalized Dehn-Sommerville equations discovered by
M. Bayer and L. Billera in [BB].
- part 6.3 describes the properties that define the generalized f -polynomial in a unique way.
In Section 7 we study the Hopf algebra D∗ as a Hopf subalgebra in Qsym.
- in part 7.1 the main theorem is proved;
- part 7.2 contains the applications of the main theorem. In particular, the image of the ring of
polytopes P in D∗ [α] under the natural mapping ϕα is found.
In Section 8 the multiplicative structure of the ring f (P ⊗ Q) is described. We prove that f (P ⊗ Q)
is a free polynomial algebra with dimension of the 2n-th graded component equal to the n-th Fibonacci
number cn (c0 = c1 = 1, cn+1 = cn + cn−1 , n > 1). This gives the decomposition of the Fibonacci series
into the infinite product
X∞ Y∞
1 n 1
2
= c n t = ,
1−t−t n=0 i=1
(1 − ti )ki
where ki is the√ number of generators of degree 2i. The infinite product converges absolutely in the
interval |t| < 5−12 . The numbers kn satisfy the inequalities kn+1 > kn > Nn − 2, where Nn is the
number of decompositions of n into the sum of odd numbers. Moreover, standard methods allow one to
find the explicit formula for kn , which in the case of prime p has the form
p
[2] p−j


X j
kp =
j=1
p−j

3
 In Section 9 we introduce the multiplicative structure on the graded group BB generated by the
Bayer-Billera polytopes arising from the isomorphism with the ring f (P).
Nowadays Hopf algebras is one of the central tools in combinatorics. There is a well-known Hopf
algebra R of posets introduces by Joni and Rota in [JR]. Various aspects of this algebra were studied
in [Ehr, ABS, Sch1, Sch2]. The generalization of the Rota-Hopf algebra was proposed in [RS]. In [Ehr]
R. Ehrenborg introduced the F -quasi-symmetric function, which gives a Hopf algebra homomorphism
from the Rota-Hopf algebra to Qsym[t1 , t2 , . . . ]. Section 10 contains the following results:
- in part 10.1 we consider the Hopf comodule structures arising from the Hopf module structures
on the rings P and RP over the Hopf algebras Z and D;
- in part 10.2 we describe the ring homomorphisms that arise from this structures. As an example,
we show how the generalized f -polynomial and the Ehrenborg F -quasi-symmetric function appear
in this context;
- part 10.3 contains the definition and the main properties of the natural Hopf comodule structure
on the ring P over the Hopf algebra RP ⊂ R;
- part 10.4 describes the interrelations between the Hopf comodule structures and the homomor-
phisms arising in this section. In particular, it is shown that the Hopf comodule structures from
parts 10.1 and 10.3 are related by the Ehrenborg transformation.
Section 11 is devoted to the cone and the bipyramid operators C and B used by M. Bayer and
L. Billera [BB] to find the linear span of flag f -vectors of polytopes. We show that the operations
C and B can be defined on the ring Qsym[t1 , t2 , . . . ][α] in such a way that f (CP ) = Cf (P ) and
f (BP ) = Bf (P ). The same idea works for the ring RP and the homomorphisms fRP , and F∗ .
- in part 11.1 we consider the case of P;
- in part 11.2 – the case of RP.
Section 12 deals with the problem of the description of flag f -vectors of polytopes. For simple
polytope g-theorem gives the full description of the set of flag f -vectors. As it is mentioned in [Z1] even
for 4-dimensional non-simple polytopes the corresponding problem is extremely hard. We show how
this problem can be stated in terms of the ring of polytopes.
This article contains the results and corrections of some inaccuracies made in the work [BuchE].

2 Polytopes
2.1 Definitions and constructions
This section contains main definitions and constructions from the polytope theory. For details see the
books [Gb], [Z2], [BP].
There are two algorithmically different ways to define a convex polytope.
Definition 2.1. A V-polytope is a convex hull of a finite set of points in some Rd .
Definition 2.2. An H-polyhedron is an intersection of finitely many closed halfspaces in some Rd . An
H-polytope is a bounded H-polyhedron.
The classical fact that these two definitions are equivalent is proved in different books, for example
in [Z2, Theorem 1.1].
The dimension of a convex polytope is the dimension of its affine hull. Without loss of generality in
most cases we will assume that an n-dimensional polytope P lies in the space Rn .
It is not difficult to prove the following lemma:
Proposition 2.3. Any n-dimensional convex polytope P n up to a translation can be represented in the
form:
P n = {x ∈ Rn : hai , xi + 1 > 0, i = 1, . . . , m}, (1)
where the system of inequalities is irredundant, that is the absence of any one of them changes the set
P n.

4
 In this case the polytope P n has exactly m facets

Fi = P n ∩ {x ∈ Rn : hai , xi + 1 = 0}, i = 1, . . . , m.

Proof. It is enough to make by a translation the origin 0 the inner point of P .

Two polytopes P n and Qn are said to be combinatorially equivalent if there exists a bijection between
their face lattices L(P ) and L(Q) that preserves an inclusion relation. A combinatorial polytope is an
equivalence class of combinatorially equivalent convex polytopes.
The notion of the general position from the point of view of two different definitions of a convex
polytope give two special classes of polytopes.
Definition 2.4. A polytope P n is called simple if any vertex of P belongs to exactly n facets.
A polytope P n is called simplicial if any facet of P is an (n − 1)-dimensional simplex (i.e. contains
exactly n vertices).
Definition 2.5 (dual polytope). Let P n be an n-dimensional polytope (1). Then a dual (or polar)
polytope is defined as
(P n )∗ = {y ∈ Rn : hy, xi + 1 > 0, ∀x ∈ P n }.
It can be shown (see, e.g. [Z2]) that

(P n )∗ = conv{ai , i = 1, . . . , m},

and (P ∗ )∗ = P .
There is a bijection F ←→ F ♦ between the i-faces of P and the (n − 1 − i)-faces of P ∗ such that

F ⊂ G ⇔ G♦ ⊂ F ♦ .

For any simple polytope P its polar polytope P ∗ is simplicial and vice versa.
Definition 2.6 (face polytope). For the face F of the polytope P let us define a face polytope P/F as

P/F = (F ♦ )∗

It is easy to see that the face polytope has dimension dim P − dim F − 1 and the face lattice

L(P/F ) = [F, P ] = {G ∈ L(P ) : F ⊆ G ⊆ P }.

Example 2.7. Let P n be a simple polytope. Then for any proper k-dimensional face F of P we have
P/F = ∆n−k−1 .
Definition 2.8 (direct product). For two polytopes P n1 ⊂ Rn1 and Qn2 ⊂ Rn2 with m1 and m2 facets
respectively the direct product

P n1 × Qn2 = {(x, y) ⊂ Rn1 × Rn2 : x ∈ P n1 , y ∈ Rn2 }

is an (n1 + n2 )-dimensional convex polytope. Its faces are direct products of faces of P n1 and Qn2 . In
particular, it has m1 + m2 facets.
Definition 2.9 (operation ◦). For two polytopes

P n1 = {x ∈ Rn1 : hai , xi + 1 > 0, i = 1, . . . , m1 },

Qn2 = {y ∈ Rn2 : hbj , yi + 1 > 0, j = 1, . . . , m2 }

let us define the polytope

P n1 ◦ Qn2 = conv(P n1 × {0} ∪ {0} × Qn2 ) ⊂ Rn1 × Rn2

Proposition 2.10. For two polytopes P n1 and Qn2 represented in the form (1) we have

(P n1 × Qn2 )∗ = (P n1 )∗ ◦ (Qn2 )∗

5
Proof. The polytope P n1 × Qn2 is defined by the inequalities
hai , xi + 1 > 0, i = 1, . . . , m1 , hbj , yi + 1 > 0, j = 1, . . . , m2 ,
therefore,
(P n1 × Qn2 )∗ = conv{(ai , 0), (0, bj ), i = 1, . . . , m1 , j = 1, . . . , m2 } = (P n1 )∗ ◦ (Qn2 )∗

Since the product of two simple polytopes is again a simple polytope, we see that for any two
simplicial polytopes P and Q the polytope P ◦ Q is again simplicial.
Example 2.11 (bipyramid). Let us define a bipyramid as BP = I ◦ P . Then we have (BP )∗ = I × P ∗ .
Definition 2.12 (join). For the space Rn let ∆n−1 be the regular simplex conv(e1 , . . . , en ), where
ei = (0, . . . , 0, 1, 0, . . . , 0) is the i-th basis vector. Let P n1 ⊂ Rn1 +1 , and Qn2 ⊂ Rn2 +1 be polytopes
lying in the regular simplices ∆n1 and ∆n2 . A join P > Q is defined as
P > Q = conv(P × {0} ∪ {0} × Q) ⊂ Rn1 +1 × Rn2 +1
It is easy to see that the polytope P > Q lies in the regular simplex ∆n1 +n2 +1 .
From this construction is evident that the join is an associative operation.
Remark 2.13. In the definition of the join we used the polytopes that lie in the regular simplices. In
fact, the affine type of P > Q does not depend on the positions of P and Q in the regular simplices,
moreover, it depends only on the affine types of P and Q.
Indeed, let {v1 , . . . , vn1 +1 } be the set of vertices of P in general position, and {w1 , . . . , wn2 +1 } be the
set of vertices of Q in general position. Since the origin O = 0 ∈ Rn1 +1 ⊂ Rn1 +1 ×Rn2 +1 does not belong
−−→ −−−−−→ −−→ −−−−−→
to aff P , the vectors Ov1 , . . . , Ovn1 +1 are linearly independent. Similarly the vectors Ow1 , . . . , Own2 +1
−−→ −−− − −→ −−→ −− −−−→
form a basis in Rn2 +1 . Then the vectors Ov1 , . . . , Ovn1 +1 , Ow1 , . . . , Own2 +1 are linearly independent in
Rn1 +1 ×Rn2 +1 , and the set of points v1 , . . . , vn1 +1 , w1 , . . . , wn2 +1 is in general position. This implies that
dim P n1 >Qn2 = n1 +n2 +1. Now for any two realizations (P ∈ ∆n1 , Q ∈ ∆n2 ) and (P ′ ∈ ∆n1 , Q′ ∈ ∆n2 )
the correspondence vi → v′i , wj → w′j defines an affine isomorphism aff P > Q ≃ aff P ′ > Q′ such that
P → P ′ , Q → Q′ . Then conv(P ∪ Q) → conv(P ′ ∪ Q′ ), so P > Q ≃ P ′ > Q′ . Since the affine types of P
and Q are uniquely defined by the sets of points {vi } and {wj }, we see, that the affine type of P > Q
depends only on the affine types of P and Q.
Proposition 2.14. P n1 > Qn2 is an (n1 + n2 + 1)-dimensional polytope.
Faces of P > Q up to an affine equivalence are exactly F > G, where ∅ ⊆ F ⊆ P, ∅ ⊆ G ⊆ Q are
faces of P and Q respectively.
Proof. For the proof see Appendix A.
Corollary 2.15. The face lattice L(P > Q) is the direct product of the face lattices L(P ) and L(Q).
Therefore the join operation is well defined on combinatorial polytopes.
Remark 2.16. The join P > Q can be defined in more invariant terms – as the convex hull of P and Q
provided they are placed into affine subspaces of some RN such that their affine hulls aff(P ) and aff(Q)
are skew.
Example 2.17. For the simplices ∆k and ∆l we have ∆k > ∆l = ∆k+l+1 .
Proposition 2.18. For two combinatorial polytopes P and Q
(P > Q)∗ = P ∗ > Q∗
Proof. Indeed, the faces of the polytope (P > Q)∗ have the form (F > G)♦ , and (F > G)♦ ⊆ (F ′ > G′ )♦
if and only if F ′ > G′ ⊆ F > G, i.e. F ′ ⊆ F, G′ ⊆ G, which is equivalent to F ♦ ⊆ (F ′ )♦ , G♦ ⊆ (G′ )♦ .
Therefore the correspondence (F > G)♦ ←→ F ♦ > G♦ is a combinatorial equivalence of the polytopes
(P > Q)∗ and P ∗ > Q∗ .
Example 2.19 (cone). Let us define a cone CP as pt ∗P , where pt is a point. Then (CP )∗ = C(P ∗ ).

6
2.2 Ring of polytopes
Definition 2.20. Denote by P 2n the free abelian group generated by all n-dimensional combinatorial
polytopes. For n > 1 we have the direct sum
X
P 2n = P 2n, 2(m−n) ,
m>n+1

where P n ∈ P 2n, 2(m−n) if it is a polytope with m facets, and rank P 2n, 2(m−n) < ∞ for any fixed n and
m such that n < m. The product of polytopes P × Q turns the direct sum

X X m−1
X
P= P 2n = P 0 + P 2n, 2(m−n)
n>0 m>2 n=1

into a bigraded commutative associative ring, a ring of polytopes. The unit is P 0 = pt, a point.
Let us denote P [2n] = P 0 ⊕ P 2 ⊕ · · · ⊕ P 2n .
The direct product P × Q of simple polytopes P and Q is a simple polytope as well. Thus the abelian
subgroup Ps ⊂ P generated by all simple polytopes is a subring in P.
Definition 2.21. A polytope P n is called indecomposable if it can not be decomposed into a direct
product P1 × P2 of two polytopes of positive dimensions.
Proposition 2.22. P is a polynomial ring generated by indecomposable combinatorial polytopes.
For the proof see Appendix B.
Since the polytope P = P1 × P2 is simple if and only if the polytopes P1 and P2 are simple, Ps is a
polynomial ring generated by all indecomposable simple polytopes.
Since (P ∗ )∗ = P , the correspondence P → P ∗ defines an involutory linear isomorphism of the ring
P.
Any isomorphism of graded abelian groups A : P → P defines a new multiplication

P ×A Q = A−1 (AP × AQ)

such that A is a graded ring isomorphism (P, ×A ) → (P, ×).

In particular, ∗ defines the multiplication

P ◦ Q = (P ∗ × Q∗ )∗

Definition 2.23. Let us define an alpha-character ξα : P → Z[α]:

ξα (P n ) = αn

2.3 The Rota-Hopf algebra

Definition 2.24. Let P be a finite poset with a minimal element 0̂ and a maximal element 1̂.
An element y in P covers another element x in P if x < y and there is no z in P such that x < z < y.
A poset P is called graded if there exists a rank function ρ : P → Z such that ρ(0̂) = 0 and
ρ(y) = ρ(x) + 1 if y covers x. It is easy to see that if a rank function exists, then it is determined
in a unique way.
Set ρ(P) = ρ(1̂), and deg(P) = 2ρ(P).
Two finite graded posets are isomorphic if there exists an order preserving bijection between them.
Let us denote by R the graded free abelian group with basis the set of all isomorphism classes of
finite graded posets.
R has the structure of a graded connected Hopf algebra:
• The multiplication P · Q is a cartesian product P × Q of posets P and Q: let x, u ∈ P, and y, v ∈ Q.
Then
(x, y) 6 (u, v) if and only if x 6 u and y 6 v

7
 • The unit element in R is the poset with one element 0̂ = 1̂
• The comultiplication is X
∆(P) = [0̂, z] ⊗ [z, 1̂],
0̂6z61̂

where [x, y] is the subposet {z ∈ P |x 6 z 6 y}.

• The counit ε is (
1, if 0̂ = 1̂;
ε(P) =
0, else

• The antipode χ is XX
χ(P) = (−1)k [x0 , x1 ] · [x1 , x2 ] . . . [xk−1 , xk ],
k>0 Ck

where Ck = (0̂ = x0 < x1 < · · · < xk = 1̂)

The Hopf algebra of graded posets was originated in the work by Joni and Rota [JR]. Variations of this
construction were studied in [Ehr, ABS, Sch1, Sch2]. The generalization of this algebra can be found in
[RS].
Example 2.25. The simplest Boolean algebra B1 = {0̂, 1̂} is the face lattice of the point pt.
∆(B1 ) = 1 ⊗ B1 + B1 ⊗ 1
χ(B1 ) = −B1
Definition 2.26. There is a natural involutory graded ring isomorphism ∗ of the ring R. For a graded
poset P let P∗ be a graded poset consisting of the same elements but with the reverse inclusion relation,
i.e. x 6P y if and only is x >P∗ y. Then ρ(P∗ ) = ρ(P ), and ∆(P∗ ) = ∗ ⊗ ∗(τR ∆ P), where τR is a ring
homomorphism R ⊗ R → R ⊗ R that interchanges the tensor factors τR (x ⊗ y) = y ⊗ x.
Definition 2.27. A join of polytopes defines a bilinear operation of degree +2 on the ring P. It is
associative and commutative, so (P, >) is a commutative associative ring without a unit.
Let us add a formal unit ∅ of degree −2 corresponding to the empty set. Thus ∅ > P = P > ∅ = P .
Let us denote the ring (P ⊕ Z∅, >) by RP. Then RP is a commutative associative ring.
Given two polytopes: P n1 with v1 vertices and m1 facets, and Qn2 with v2 vertices and m2 facets,
the polytope P n1 > Qn2 has dimension n1 + n2 + 1, while its number of vertices is v1 + v2 , and its number
of facets is m1 + m2 .
Therefore RP is a threegraded ring with graduations 2(n + 1), 2(v − n − 1), and 2(m − n − 1),
where n = dim P , v is the number of vertices, and m is the number of facets. The duality operator ∗
interchanges the graduations 2(v − n − 1) and 2(m − n − 1).
Remark 2.28. Let us recall that in the ring P the polytope P has two graduations 2n and 2(m − n).
We have v(P × Q) = v(P ) · v(Q), so the number of vertices does not give the third graduation in this
case.
Definition 2.29. Let us denote by SP the graded abelian subgroup consisting of all self-dual elements
of the ring RP. According to Proposition 2.18 it is a subring in RP.
Example 2.30. SP contains the subring generated by simplices and all polygons Mn2 .
The correspondence P → L(P ) defines the homomorphism of abelian groups L : P ⊕ Z∅ → R. It is
an injection. Since L(P > Q) = L(P ) · L(Q), it is a homomorphism of graded rings RP → R. Moreover,
L(P ∗ ) = L(P )∗ .
Definition 2.31. Since
X X
∆(L(P )) = [∅, F ] ⊗ [F, P ] = L(F ) ⊗ L(P/F ),
∅⊆F ⊆P ∅⊆F ⊆P

we see that L(RP) is a Hopf subalgebra in R, therefore it induces a Hopf algebra structure on the ring
RP:

8
 • The comultiplication in RP is X
∆(P ) = F ⊗ P/F ;
∅⊆F ⊆P

• The counit is (
1, if P = ∅,
ε(P ) =
0, else;

• The antipode χ is defined as

X X
χ(P ) = (−1)k (F1 /F0 ) > (F2 /F1 ) > · · · > (Fk /Fk−1 ).
k>0 ∅=F0 ⊂F1 ⊂···⊂Fk =P

Remark 2.32. Let τRP : RP ⊗ RP → RP ⊗ RP be the ring homomorphism that interchanges the
tensor factors: τRP (x ⊗ y) = y ⊗ x. Then ∆(P ∗ ) = ∗ ⊗ ∗(τRP ∆P ).
Proposition 2.33. RP is a ring of polynomials generated by all join-indecomposable polytopes.
See Appendix B for the proof.
Definition 2.34. For the ring RP let us define an α-character εα by the formula

εα (P n ) = αn+1 = αρ(P )

Then ε0 is the counit ε.

Definition 2.35. Let B be the abelian subgroup in RP generated by ∅, and all the simplices ∆n , n > 0.
Then L(∆n−1 ) = Bn is a Boolean algebra {0̂, 1̂}n . We have ∆k > ∆l = ∆k+l+1 , so it is a subring in
RP.
Let us denote x = ∆0 = pt. Then ∆n−1 = xn . We have ∆x = 1 ⊗ x + x ⊗ 1, χ(x) = −x, so
B is a Hopf subalgebra isomorphic to the Hopf algebra of polynomials Z[x] with the comultiplication
∆x = 1 ⊗ x + x ⊗ 1.
Definition 2.36. We will denote by R the ring P or the ring RP, and by R[2n] the sum

R0 ⊕ R2 ⊕ R4 ⊕ · · · ⊕ R2n

2.4 Operators
Definition 2.37. Let us define the ring of linear operators L(R) = HomZ (R, R) with the multiplication
given by a composition of operators.
Definition 2.38. For k > 0 let us define faces operators dk that sent a polytope P n to the sum of all
its (n − k)-dimensional faces. X
dk P n = F n−k
F n−k ⊆P n
(
∅ in RP,
In particular, dn+1 P n = , dk ∅ = 0, k > 1.
0 in P

Example 2.39. We have d1 pt = ∅ in RP, and 0 in P.

Definition 2.40. Let us define by D(R) ⊂ L(R) the ring of operators generated by faces operators dk ,
k > 0, and by D(R2n , R2(n−k) ) the set of all operators in D(R) of degree −2k acting on the abelian
group R2n .
Proposition 2.41. For any two polytopes P and Q, and d = d1 we have:

d(P × Q) = (dP ) × Q + P × (dQ) in P, and d(P > Q) = (dP ) > Q + P > (dQ) in RP.

Therefore P and RP are differential rings.

9
Proof. Each facet of P × Q has the form F × Q or P × G, where F, G are facets of P and Q. Therefore,
X X
d(P × Q) = F ×Q+ P × G = (dP ) × Q + P × (dQ)
F ⊂P G⊂Q

Each facet of P > Q has the form F > Q or P > G, where F, G are facets of P and Q. Therefore,

d(P > Q) = (dP ) > Q + P > (dQ)

Example 2.42. xP = pt >P = CP is a cone. For a polytope P of positive dimension we have

d(CP ) = d(pt >P ) = (d pt) > P + pt >(dP ) = ∅ > P + C(dP ) = P + C(dP )

Thus, the multiplication by x = pt in the ring RP defines a cone operator C ∈ L(RP) of degree +2
such that
[d, C] = id
On the other hand, in the ring P the same relation holds for all the polytopes except for P = pt. In
this case we have d(C pt) = dI = 2 pt, while C(d pt) + pt = pt. Therefore, in the ring P we have

[d, C] = id +ξ0

Example 2.43. We have

d(∆k > ∆l ) = (k + 1)∆k−1 > ∆l + (l + 1)∆k > ∆l−1 = (k + l + 2)∆k+l = d∆k+l+1

Example 2.44. Since ∆k = xk+1 , we see that dxk+1 = (k + 1)xk . Therefore Z[x] is a differential
d
subring in (RP, d) with d = dx .
Definition 2.45. The correspondence P → BP defines a linear bipyramid operator of degree +2 on
the rings P and RP. By definition B∅ = pt.
Remark 2.46. It is easy to see that if P is simple and self-dual, i.e. P ∗ is combinatorially equivalent
to P , then P is a simplex. There are many examples of self-dual non-simple polytopes, with simplest
infinite family given by k-gonal pyramids for k ≥ 4. The next example gives a more interesting regular
self-dual polytope.
Example 2.47 (24-cell). Let Q be the 4-polytope obtained by taking the convex hull of the following
24 points in R4 : endpoints of 8 vectors ±ei , 1 ≤ i ≤ 4, and 16 points of the form (± 12 , ± 21 , ± 12 , ± 21 ). It
can be shown that all the facets of Q are octahedra.
The polar polytope Q∗ is specified by the following 24 inequalities:
1
± xi + 1 ≥ 0 for 1 ≤ i ≤ 4, and 2( ± x1 ± x2 ± x3 ± x4 ) + 1 ≥ 0. (2)

Each of these inequalities turns into equality in exactly one of the specified 24 points, so it defines a
supporting hyperplane whose intersection with Q is only one point. This implies that Q has exactly 24
vertices. The vertices of Q∗ may be determined using the “elimination process” to (2) (see [Z2, Section
1.2]), and as the result we obtain 24 points of the form ± ei ± ej for 1 ≤ i < j ≤ 4. Each supporting
hyperplane defined by (2) contains exactly 6 vertices of Q∗ , which form an octahedron. So both Q and
Q∗ have 24-vertices and 24 octahedral facets. In fact, both Q and Q∗ provide examples of a regular
4-polytope called a 24-cell. It is the only regular self-dual polytope different from a simplex. For more
details on 24-cell and other regular polytopes see [C].
In this example we have dQ = 24BI 2 , and (dQ)∗ = 24I 3 . Therefore for a self-dual element P ∈ SP,
dP can be non self-dual. So SP is not a differential subring in RP.
However there is another derivation defined on SP. Since the duality ∗ is a ring homomorphism,
δ = ∗d∗ is a derivation of the ring RP, as well as d.
Proposition 2.48. d + δ is a derivation of the ring SP

10
Proof. Since for any P ∈ SP, P ∗ = P , we have

((d + δ)P )∗ = ∗(d + ∗d∗)P = ∗dP + ∗ ∗ dP = ∗d ∗ P + dP = (δ + d)P.

Proposition 2.49.
X
dk (P × Q) = (di P ) × (dj Q) in P
i+j=k
X
dk (P > Q) = (di P ) > (dj Q) in RP.
i+j=k

Proof. We have
 
X X X X
dk (P n × Qr ) = F n−i × Gr−j = F n−i ×  Gr−j  =
F n−i ⊆P n , Gr−j ⊆Qr , i+j=k i+j=k F n−i ⊆P n Gr−j ⊆Qr
   
X X X X
=  F n−i  ×  Gr−j  = (di P ) × (dj Q)
i+j=k F n−i ⊆P n Gr−j ⊆Qr i+j=k

The same argument works also for P n > Qr . The only difference is that faces of the form ∅ > Gr−j ,
i = n + 1, i + j = k, and F n−i > ∅, j = r + 1, i + j = k are allowed. Here ∅ = dn+1 P n = dr+1 Qr .
Definition 2.50. Let us define an operator Φ : R → R[t], Φ(t) ∈ D(R)[[t]] by the formula

Φ(t) = 1 + dt + d2 t2 + · · · + dk tk + . . .

Definition 2.51. Let fi (P n ) be the number of i-dimensional faces of P n .

Proposition 2.52. We have Φ(−t)Φ(t) = 1 on the ring R, that is for any n > 1

dn − ddn−1 + · · · + (−1)n−1 dn−1 d + (−1)n dn = 0.

Proof. The relation is clear for ∅ ∈ RP and for pt ∈ P. For pt ∈ RP we have

Φ(−t)Φ(t) pt = (1 − dt + . . . )(pt +∅t) = pt −∅t + ∅t = pt

Let us consider the polytope P n of positive dimension.

∞ k
!
X X
i
Φ(−t)Φ(t) = (−1) di dk−i tk .
k=0 i=0

If k > n + 1 then the coefficient of tk in the series Φ(−t)Φ(t)P n is equal to 0. If k = 0, then it is equal
to P n . Let 1 6 k 6 n. Then


ddk−1 − d2 dk−2 + · · · + (−1)k−2 dk−1 d P n =
 
X X X X
=  1− 1 + · · · + (−1)k−2 1 F n−k =
F n−k ⊂P n F n−k+1 ⊃F n−k F n−k+2 ⊃F n−k F n−1 ⊃F n−k
X

= f0 (P n /F n−k ) − f1 (P n /F n−k ) + · · · + (−1)k−2 fk−2 (P n /F n−k ) F n−k
F n−k ⊂P n

Since P n /F n−k is a (k − 1)-dimensional polytope, the Euler formula gives the relation

f0 (P n /F n−k ) − f1 (P n /F n−k ) + · · · + (−1)k−2 fk−2 (P n /F n−k ) = 1 + (−1)k .

11
Thus we obtain
k
X


ddk−1 − d2 dk−2 + · · · + (−1)k−2 dk−1 d P n = (1 + (−1)k )dk P n ⇔ (−1)i di dk−i P n = 0
i=0

The coefficient of tn+1 is equal to 0 in P, and to (1 − f0 + f1 + · · · + (−1)n fn−1 + (−1)n+1 )∅ in RP.

Then the Euler formula implies that it is equal to 0.
Proposition 2.53.

ξ−α Φ(α) = ξα in the ring P

ε−α Φ(α) = ε0 in the ring RP.

Proof. Let us note that ξα dk P n = fn−k (P n )αn−k . Then for any polytope P n we have
n
!
X
n n n−1 n
ξ−α Φ(α)P = (−α) + (−α) fn−1 α + · · · + f0 α = (−1) fi αn = αn = ξα P n ,
i

i=0

since the Euler formula gives the relation:

f0 − f1 + · · · + (−1)n−1 fn−1 + (−1)n = 1

On the other hand, in the ring RP we have εα dk P n = fn−k (P n )αn−k+1 , so

ε−α Φ(α)P n = (−α)n+1 + (−α)n fn−1 α + · · · + (−α)f0 αn + αn+1

According to the Euler formula this expression is equal to 0 for all the polytopes but ∅. Thus
ε−α Φ(α) = ε0 .
Example 2.54. On the ring of simple polytopes Ps ⊂ P we have the relations k!dk |Ps = dk Ps ,


therefore D(Ps ) is isomorphic to the divided power ring dk dl = k+l dt
k dk+l , and Φ(t)|Ps = e .
Let us note that this equality is not valid for Ps ⊂ RP. The problem is that for a simple polytope
k
P n we have dk = dk! for all 1 6 k 6 n. But dn+1 P n = ∅, while
 
dn+1 d dn n d d f0 (P n )
Pn = P = (dn P n ) = f0 (P n ) pt = ∅
(n + 1)! n+1 n! n+1 n+1 n+1

Another example of operators in L(R) is given by the multiplication by elements of R:

[P ](Q) = P Q

Proposition 2.55. The following relation holds

k
X
dk [P ] = [di P ]dk−i
i=0

Proof.
k
X
dk (P Q) = (di P )dk−i Q
i=0

Thus any operator in the ring R D(R) generated by D(R) and {[P ], P ∈ R} can be expressed as a
sum of operators X
[P ]Dω , P ∈ R, Dω ∈ D(R).
P, ω

Let us remind that [pt] = x is a cone operator in L(RP).

12
2.5 Flag f -vectors
Definition 2.56. Let P n be an n-dimensional polytope and S = {a1 , . . . , ak } ⊂ {0, 1, . . . , n − 1}.
A flag number fS = fa1 , ..., ak is the number of increasing sequences of faces

F a1 ⊂ F a2 ⊂ · · · ⊂ F ak , dim F ai = ai .

It is easy to see that for S = {i} the number f{i} = fi is just the number of i-dimensional faces. We
have already defined this number above. Let us denote l(S) = k. Then 0 6 k 6 n.
The collection {fS } of all the flag numbers is called a flag f -vector (or extended f -vector) of the
polytope P n . By the definition f∅ = 1.
Remark 2.57. For the set {a1 , . . . , ak } ⊂ {0, 1, . . . , n − 1} let us denote a0 = −1, and ak+1 = n. Then

f−1, a1 , ..., ak = f−1, a1 , ..., ak , n = fa1 , ..., ak , n = fa1 , ..., ak

This notation corresponds to the fact that ∅ is the only face of P of dimension −1, and any other face
contains ∅, while the polytope P itself is the only face of dimension n, and it contains any other face.
Flag f -vectors have been extensively studied by M. Bayer and L. Billera in [BB], where the generalized
Dehn-Sommerville relations were proved:
Theorem ([BB], Theorem 2.1). Let P n be an n-dimensional polytope, and S ⊂ {0, . . . , n − 1}. If
{i, k} ⊆ S ∪ {−1, n} such that i < k − 1 and S ∩ {i + 1, . . . , k − 1} = ∅, then
k−1
X
(−1)j−i−1 fS∪{j} = (1 − (−1)k−i−1 )fS . (3)
j=i+1

Definition 2.58. For n > 1 let Ψn be the set of subsets S ⊂ {0, 1, . . . , n − 2} such that S contains no
two consecutive integers.
It easy to show by induction that the cardinality of Ψn is equal to the n-th Fibonacci number cn
(cn = cn−1 + cn−2 , c0 = 1, c1 = 1).
Let us remind that for any polytope P we have defined a cone (or a pyramid) CP and a bipyramid (or
a suspension) BP . These two operations are defined on combinatorial polytopes and can be extended
to linear operators on the rings P and RP. By definition B∅ = pt = C∅.
The face lattice of the polytope CP n is:

{∅}; {Fi0 , C∅}; {Fi1 , CFj0 }; . . . ; {Fin−1 , CFjn−2 }; {P n , CFin−1 }; {CP n },

where Fik are k-dimensional faces of P n , and

F a1 ⊂ F a2 in CP ⇔ F a1 ⊂ F a2 in P ,
F a1 ⊂ CF a2 ⇔ F a1 ⊂ F a2 ,
CF a1 ⊂ CF a2 ⇔ F a1 ⊂ F a2 .

The polytope BP n has the face lattice:

{∅}, {C−1∅, C1 ∅, Fi0 }, {Fi1 , C−1 Fj0 , C1 Fk0 }, . . . , {Fkn−1 , C−1 Fjn−2 , C1 Fkn−2 }, {C−1 Fin−1 , C1 Fjn−1 }, {BP n }.

where C−1 and C1 are the lower and the upper cones, and

F a1 ⊂ F a2 in BP ⇔ F a1 ⊂ F a2 in P ,
F a1 ⊂ Cs F a2 ⇔ F a1 ⊂ F a2 ;
Cs F a1 ⊂ Ct F a2 ⇔ s = t and F a1 ⊂ F a2 ;

Definition 2.59. For n > 1 let Ωn be the set of n-dimensional polytopes that arises when we apply
words in B and C that end in C 2 and contain no adjacent B’s to the empty set ∅.

13
 Each word of length n+ 1 from the set Ωn either has the form CQ, Q ∈ Ωn−1 , or BCQ, Q ∈ Ωn−2 , so
cardinality of the set Ωn satisfies the Fibonacci relation |Ωn | = |Ωn−1 |+|Ωn−2 |. Since |Ω1 | = |{C 2 }| = 1,
and |Ω2 | = |{C 3 , BC 2 }| = 2, we see that |Ωn | = cn = |Ψn |.
M. Bayer and L. Billera proved the following fact:
Theorem. Let n > 1. Then
1. For all T ⊆ {0, 1, . . . , n − 1} there is a nontrivial linear relation expressing fT (P ) in terms of
fS (P ), S ∈ Ψn , which holds for all n-dimensional polytopes (see [BB, Proposition 2.2]).
2. The extended f -vectors of the cn elements of Ωn are affinely independent [BB, Proposition 2.3].
Thus the flag f -vectors

{fS (P n )}S∈Ψn : P n – an n-dimensional polytope

span an (cn − 1)-dimensional affine hyperplane defined by the equation f∅ = 1 (see [BB, Theorem
2.6])
Remark 2.60. The first part of the theorem follows from the generalized Dehn-Sommerville relations
(3). In fact, we have fS, n−2, n−1 = 2fS, n−2 , and
n−2
X
fS, i, n−1 = (1 + (−1)n−i )fS, i + (−1)n−i−1 (−1)j−i−1 fS, i, j , i < n − 2.
j=i+1

Therefore, we can get rid of the index n − 1. Then fS1 , i−1, i, i+1, S2 = 2fS1 , i−1, i+1, S2 . Therefore, we can
get rid of triples of consecutive indices.
Now for each set S in the sum we obtain let us take the last pair of consecutive indices {i, i + 1}.
Then fS = fS1 , j, i, i+1, S2 .
i−1
X
fS1 , j, i, i+1, S2 = (1 + (−1)i+1−j )fS1 , j, i+1, S2 + (−1)i−j (−1)k−j−1 fS1 , j, k, i+1, S2
k=j+1

Then the last pair of consecutive indices for each set in the sum consists of smaller numbers. Then after
the sequence if such steps and removing the triples of consecutive indices on each step we will obtain
flag numbers of the form fi, i+1, S2 , where the set {i + 1} ∪ S2 contains no consecutive indices. Then
i−1
X
fi, i+1, S2 = (1 + (−1)i )fi+1, S2 − (−1)i (−1)j fj, i+1, S2
j=0

where all the sets {j, i + 1} ∪ S2 belong to Ψn . This finishes the proof.
In fact, a little bit more stronger statement, which can be easily extracted from the original Bayer-
Billera’s proof, is true.
Let us identify the words in Ωn with the sets in Ψn in such a way that the word
C n+1−ak
BC ak −ak−1 −1 B . . . BC a1 −1 corresponds to the set {a1 −3, . . . , ak −3}. Let us set C < B and or-
der the words lexicographically. Consider a matrix K n of sizes cn ×cn , kQ, S = fS (Q), Q ∈ Ωn , S ∈ Ψn ,.
Proposition 2.61.
det(K n ) = 1.
Proof. For n = 1 the matrix K 1 = (f∅ (I)) = (1).
For n = 2    
2 1 f0 (∆2 ) 1 3
K = = ; det K 2 = 1
1 f0 (I 2 ) 1 4
Let us prove the statement by induction. Consider the matrix K n :
 
K11 K12
Kn =
K21 K22

14
where the block K11 corresponds to the words of the form CQ, Q ∈ Ωn−1 and the sets S, that do not
contain n − 2. Similarly, the block K22 corresponds to the words of the form BCQ, Q ∈ Ωn−2 and the
sets containing n − 2.
Each increasing sequence of faces of the polytope CP n−1 without two faces of adjacent dimensions
has the form
F l1 ⊂ · · · ⊂ F li ⊂ CF li+1 ⊂ · · · ⊂ CF lk ,
so for {a1 , . . . , ak } ∈ Ψn

fa1 , ..., ak (CP n−1 ) = fa1 −1, a2 −1, ..., ak −1 (P n )+· · ·+fa1 , ..., ai , ai+1 −1, ..., ak −1 (P n )+· · ·+fa1 , ..., ak (P n ), (4)

where f−1, a2 , ..., ak = fa2 , ..., ak .

The sets of the form {a1 , . . . , al , al+1 − 1, . . . , ak − 1} may not belong to Ψn−1 . But we can express
the corresponding flag numbers in terms of {fS , S ∈ Ψn−1 } using the generalized Dehn-Sommerville
relations. Let ak < n − 2. Each relation has the form:
 
k−2
X


fS1 , i, k−1, k, S2 = (−1)k−i  1 − (−1)k−i−1 fS1 , i, k, S2 − (−1)j−i−1 fS1 , i, j, k, S2  ,
j=i+1

where all the sets on the right side are lower than the set on the left side. Thus for ak < n − 2

fa1 ..., ak (CP n ) = fa1 , ..., ak (P n ) + lower summands,

so the matrix K11 can be represented as K n−1 T , where T is an upper unitriangular matrix. In particular,
det K11 = det K n−1 · det T = 1 by induction.
Lemma 2.62. Let P be an (n − 2)-dimensional polytope, and S ∈ Ψn . Then for the polytopes BCP
and CBP (
fS (CBP ), if n − 2 ∈
/ S,
fS (BCP ) =
fS (CBP ) + fS\{n−2} (P ), if n − 2 ∈ S,

Proof. The polytope BCP has faces

{F, ∅ ⊆ F ⊆ P }; {CF, ∅ ⊆ F ( P }; {Ca F, ∅ ⊆ F ⊆ P }; {Ca CF, ∅ ⊆ F ( P }; BCP,

where a = ±1
Then each increasing sequence of faces corresponding to the set S has one of the forms
• F l1 ⊂ · · · ⊂ F li
• F l1 ⊂ · · · ⊂ F li ⊂ CF li+1 ⊂ · · · ⊂ CF lj ,
• F l1 ⊂ · · · ⊂ F li ⊂ Ca F li+1 ⊂ · · · ⊂ Ca F lj
• F l1 ⊂ · · · ⊂ F li ⊂ CF li+1 ⊂ · · · ⊂ CF lj ⊂ Ca CF lj+1 ⊂ · · · ⊂ Ca CF lk
• F l1 ⊂ · · · ⊂ F li ⊂ Ca F li+1 ⊂ · · · ⊂ Ca F lj ⊂ Ca CF lj+1 ⊂ · · · ⊂ Ca CF lk
• F l1 ⊂ · · · ⊂ F lj ⊂ Ca CF lj+1 ⊂ · · · ⊂ Ca CF lk
The polytope CBP has faces

{F, ∅ ⊆ F ( P }; {Ca F, ∅ ⊆ F ( P }; BP ; {CF, ∅ ⊆ F ( P }; {CCa F, ∅ ⊆ F ( P }; CBP,

where a = ±1
Then each increasing sequence of faces corresponding to the set S has one of the forms

• F l1 ⊂ · · · ⊂ F li , F li 6= P ,
• F l1 ⊂ · · · ⊂ F li ⊂ CF li+1 ⊂ · · · ⊂ CF lj ,
• F l1 ⊂ · · · ⊂ F li ⊂ Ca F li+1 ⊂ · · · ⊂ Ca F lj

15
 • F l1 ⊂ · · · ⊂ F li ⊂ CF li+1 ⊂ · · · ⊂ CF lj ⊂ CCa F lj+1 ⊂ · · · ⊂ Ca CF lk
• F l1 ⊂ · · · ⊂ F li ⊂ Ca F li+1 ⊂ · · · ⊂ Ca F lj ⊂ CCa F lj+1 ⊂ · · · ⊂ CCa F lk
• F l1 ⊂ · · · ⊂ F lj ⊂ CCa F lj+1 ⊂ · · · ⊂ CCa F lk

We can interchange CCa and Ca C to obtain a one-to-one correspondence between the increasing
sequences containing both C and Ca . There is a natural bijection between sequences containing ex-
actly one of these operations, as well as between sequences containing neither operations nor P . So
fS (BCP ) = fS (CBP ), if n − 2 ∈
/ S.
The only difference appears in the case when the sequence of faces in the polytope BCP has the
form
F a1 ⊂ · · · ⊂ F ak−1 ⊂ P
The sequences of this type give exactly fa1 , ..., ak−1 (P ), so fS (BCP ) = fS (CBP ) + fS\{n−2} (P ), if
n − 2 ∈ S.
Let us consider the polytope BCQ corresponding to one of the lower rows of K n . If Q starts with
C, then there is a row CBQ in the upper part of the matrix. Let us subtract the row CBQ from the
row BCQ. Then Lemma 2.62 implies that the resulting row is
(
′ 0, if n − 2 ∈
/ S;
kBCQ, S = (5)
fS\{n−2} (Q), if n − 2 ∈ S.

If Q starts with B, then there is no row CBQ in the matrix. But since det K n−1 = 1 and all flag
f -numbers of (n − 1)-dimensional polytopes can be expressed in terms of {fS , S ∈ Ψn−1 }, the flag
f -vector of any (n − 1)-dimensional polytope is an integer combination of flag f -vectors of the polytopes
from Ωn−1 . So the flag f -vector of the polytope BQ can be expressed as an integer combination of flag
f -vectors of the polytopes Q′ ∈ Ωn−1 :
X
f (BQ) = nQ′ f (Q′ )
Q′

Using Formula (4) we obtain: X

f (CBQ) = nQ′ f (CQ′ )
Q′

If we subtract the corresponding integer combination of the rows of the upper part of the matrix from
the row BCQ, we obtain the row (5).
Thus we see, that using elementary transformations of rows the matrix K n−1 can be transformed to
the matrix  n−1 
K T K12
0 K n−2
By the inductive assumption det K n−1 = det K n−2 = 1, so det K n = 1.
Remark 2.63. In the proof we follow the original Bayer and Billera’s idea, except for the fact that
they use the additional matrix of face numbers of the polytopes Q ∈ Ψn , and our proof is direct.
Corollary 2.64. The flag f -vector of any n-dimensional polytope P n is an integer combination of the
flag f -vectors of the polytopes Q ∈ Ωn .
Proof. Indeed, any flag f -number is a linear combination of fS , S ∈ Ψn . Since det K n = 1, the vector
{fS (P ), S ∈ Ψn } is an integer combination of the vectors {fS (Q), S ∈ Ψn }, Q ∈ Ωn .
This implies that the whole vector {fS (P )} is an integer combination of the vectors {fS (Q)}, Q ∈ Ωn
with the same coefficients.
This fact is important, when we try to describe the image of the generalized f -polynomial in the
ring of quasi-symmetric functions (see below).
Generalized f -vectors are also connected with a very interesting construction of a cd-index invented
by J. Fine (see [BK, Prop. 2], see also the papers by R. Stanley [St3] and M. Bayer, A. Klapper [BK]).

16
3 Hopf algebras
In this part we follow mainly the notations of [H]. See also [BR] and [CFL].

3.1 Quasi-symmetric functions

Definition 3.1. A composition ω of a number n is an ordered set ω = (j1 , . . . , jk ), ji > 1, such that
n = j1 + · · · + jk . Let us denote |ω| = n, l(ω) = k.
Let us denote by () the empty composition of the number 0. Then |()| = 0, l (()) = 0.
Definition 3.2. Let t1 , t2 , . . . be a finite or an infinite set of variables, deg ti = 2. For a composition
ω = (j1 , . . . , jk ) consider a quasi-symmetric monomial
X j
Mω = tl11 . . . tjlkk , M() = 1.
l1 <···<lk

Degree of the monomial Mω is equal to 2|ω| = 2(j1 + · · · + jk ).

For any two monomials Mω′ and Mω′′ their product in the ring of polynomials Z[t1 , t2 , . . . ] is equal
to !
X X
Mω′ Mω′′ = 1 Mω ,
ω Ω′ +Ω′′ =ω

where for the compositions ω = (j1 , . . . , jk ), ω ′ = (j1′ , . . . , jl′′ ), ω ′′ = (j1′′ , . . . , jl′′′′ ), Ω′ and Ω′′ are all the
k-tuples such that

Ω′ = (0, . . . , j1′ , . . . , 0, . . . , jl′′ . . . , 0), Ω′′ = (0, . . . , j1′′ , . . . , 0, . . . , jl′′′′ . . . , 0),

This multiplication rule of compositions is called the overlapping shuffle multiplication.

For example,
1. ! 
X X X X
M(1) M(1) = ti  tj  = t2i + 2 ti tj = M(2) + 2M(1, 1) .
i j i i<j

This corresponds to the decompositions

(2) = (1) + (1), (1, 1) = (1, 0) + (0, 1) = (0, 1) + (1, 0).

2.
! 
X X X X X
M(1) M(1, 1) = ti  tj tk  = t2i tj + ti t2j + 3 ti tj tk =
i j<k i<j i<j i<j<k

= M(2, 1) + M(1, 2) + 3M(1, 1, 1) .

This corresponds to the decompositions

(2, 1) = (1, 0) + (1, 1), (1, 2) = (1, 0) + (1, 1),

(1, 1, 1) = (1, 0, 0) + (0, 1, 1) = (0, 1, 0) + (1, 0, 1) = (0, 0, 1) + (1, 1, 0).

3.
  !
X X X X
M(1, 1) M(1, 1) =  ti tj  tk tl = t2i t2j + 2 t2i tj tk +
i<j k<l i<j i<j<k
X X X
+2 ti t2j tk + 2 ti tj t2k + 6 ti tj tk tl =
i<j<k i<j<k i<j<k<l

= M(2, 2) + 2M(2, 1, 1) + 2M(1, 2, 1) + 2M(1, 1, 2) + 6M(1, 1, 1, 1) .

17
 This corresponds to the decompositions

(2, 2) = (1, 1) + (1, 1), (2, 1, 1) = (1, 1, 0) + (1, 0, 1) = (1, 0, 1) + (1, 1, 0),
(1, 2, 1) = (1, 1, 0) + (0, 1, 1) = (0, 1, 1) + (1, 1, 0), (1, 1, 2) = (1, 0, 1) + (0, 1, 1) = (0, 1, 1) + (1, 0, 1),
(1, 1, 1, 1) = (1, 1, 0, 0) + (0, 0, 1, 1) = (1, 0, 1, 0) + (0, 1, 0, 1) = (1, 0, 0, 1) + (0, 1, 1, 0) =
= (0, 1, 1, 0) + (1, 0, 0, 1) = (0, 1, 0, 1) + (1, 0, 1, 0) = (0, 0, 1, 1) + (1, 1, 0, 0).

Thus finite integer combinations of quasi-symmetric monomials form a ring. This ring is called a ring
of quasi-symmetric functions and is denoted by Qsym[t1 , . . . , tn ], where n is the number of variables. In
the case of an infinite number of variables it is denoted by Qsym[t1 , t2 , . . . ] or Qsym.
The diagonal mapping ∆ : Qsym → Qsym ⊗ Qsym
k
X
∆M(a1 , ..., ak ) = M(a1 , ..., ai ) ⊗ M(ai+1 , ..., ak )
i=0

defines on Qsym the structure of a graded Hopf algebra.

Proposition 3.3. A polynomial g ∈ Z[t1 , . . . , tr ] is a finite linear combination of quasi-symmetric
monomials if and only if

g(0, t1 , t2 , . . . , tr−1 ) = g(t1 , 0, t2 , . . . , tr−1 ) = · · · = g(t1 , . . . , tr−1 , 0)

Proof. For the quasi-symmetric monomial Mω we have

(
0, ω = (j1 , . . . , jr );
Mω (t1 , . . . , ti , 0, ti+1 , . . . , tr−1 ) =
Mω (t1 , . . . , ti , ti+1 , . . . , tr−1 ), ω = (j1 , . . . , jk ), k < r.

So this property is true for all quasi-symmetric functions.

On the other hand, let the condition of the proposition be true. Let us prove that for a fixed com-
position ω any two monomials tjl11 . . . tjlkk and tjl′1 . . . tjl′k have the equal coefficients glj11,,..., ..., jk j1 , ..., jk
lk and gl′1 , ..., l′l .
1 k
Let li + 1 < li+1 or i = k and li < r. Consider the corresponding coefficients in the polynomial
equation:
g(t1 , . . . , tli −1 , 0, tli , tli +1 , . . . , tr−1 ) = g(t1 , . . . , tli −1 , tli , 0, tli +1 , . . . , tr−1 ).
j
On the left the monomial tjl11 . . . tjlii tli+1
i+1 jk j1 , ..., jk
−1 . . . tlk −1 has the coefficient gl1 , ..., li +1, li+1 ,..., lk , and on the right
glj11,,...,j
..., li , li+1 , ..., lk , so they are equal. Now we can move the index li to the right to li + 1 and in the same
k

manner we can move li to the left, if li−1 < li − 1, or i = 1 and li > 1.

Now let us move step by step the index l1 to 1, then the index l2 to 2, and so on. At last we obtain
that glj11,,..., ..., jk j1 , ..., jk
lk = g1, ..., k .

Remark 3.4. Similar argument shows that the same is true in the case of an infinite number of variables,
if we consider all the expressions
∞
X X X
a+ ajl11,, ...,
..., jk j1 jk
lk t l1 . . . t lk .
k=1 16l1 <···<lk j1 , ..., jk >1

of bounded degree: 2|ω| = 2(j1 + · · · + jk ) < N for all ω. Then the infinite series g of bounded degree
belongs to Qsym if and only if

g(t1 , . . . , ti−1 , 0, ti+1 , . . . ) = g(t1 , . . . , ti−1 , ti+1 , . . . )

for all i > 1. As a corollary we obtain another proof of the fact that quasi-symmetric functions form a
ring.

18
 In [H] M. Hazewinkel proved the Ditters conjecture that Qsym[t1 , t2 , . . . ] is a free commutative
algebra of polynomials over the integers.
The (2n)-th graded component Qsym2n [t1 , t2 , . . . ] has rank 2n−1 .
The numbers βi of the multiplicative generators of degree 2i can be found by a recursive relation:
Y ∞
1−t 1
=
1 − 2t i=1 (1 − ti )βi

There are group homomorphisms:

Vr+1 : Qsym[t1 , . . . , tr ] → Qsym[t1 , . . . , tr+1 ] : Vr+1 Mω (t1 , . . . , tr ) = Mω (t1 , . . . , tr+1 )

Er : Qsym[t1 , . . . , tr+1 ] → Qsym[t1 , . . . , tr ] : tr+1 → 0;

The mapping Er sends all the monomials Mω corresponding to the compositions (j1 , . . . , jr+1 ) of length
r + 1 to zero, and for the compositions ω of smaller length Er Mω (t1 , . . . , tr+1 ) = Mω (t1 , . . . , tr ). It is a
ring homomorphism.
It is easy to see that Er Vr+1 is the identity map of the ring Qsym[t1 , . . . , tr ].
Given r > 0 there is a projection ΠQsym : Z[t1 , . . . , tr ] → Qsym[t1 , . . . , tr ] ⊗ Q:
for s1 < s2 < · · · < sk !
1 X j jk
j1 jk
ΠQsym ts1 . . . tsk = r
1
t l1 . . . t lk ,
k l1 <···<lk

which gives the average value over all the monomials of each type.
Then for the quasi-symmetric monomial Mω , ω = (j1 , . . . , jk ) we have:
X  
1 r 1
ΠQsym Mω =

r Mω =

Mω = Mω ,
k
k kr
16j1 <···<jk 6r

So ΠQsym is indeed a projection.

Remark 3.5. We see that in the theory of symmetric and quasi-symmetric functions there is an
important additional graduation – the number of variables in the polynomial.
Definition 3.6. For a composition ω = (j1 , . . . , jk ) let us define the composition ω ∗ = (jk , . . . , j1 ).
The correspondence Mω → (Mω )∗ = Mω∗ defines an involutory ring homomorphism

∗ : Qsym[t1 , t2 , . . . ] → Qsym[t1 , t2 , . . . ].

Remark 3.7. Let τQsym : Qsym⊗Qsym → Qsym⊗Qsym be the ring homomorphism that interchanges
the tensor factors τQsym (x ⊗ y) = y ⊗ x. Then ∆(Mω∗ ) = ∗ ⊗ ∗(τQsym ∆Mω ).

3.2 Leibnitz-Hopf Algebras

Let R be a commutative associative ring with unity.
Definition 3.8. A Leibnitz-Hopf algebra over the ring R is an associative Hopf algebra H over the ring
R with a fixed sequence of a finite or countable number of multiplicative generators Hi , i = 1, 2 . . .
satisfying the comultiplication formula
X
∆Hn = Hi ⊗ Hj , H0 = 1.
i+j=n

A universal Leibnitz-Hopf algebra A over the ring R is a Leibnitz-Hopf algebra with the universal
property: for any Leibnitz-Hopf algebra H over the ring R the correspondence Ai → Hi defines a Hopf
algebra homomorphism.
Consider the free associative Leibnitz-Hopf algebra over the integers Z = ZhZ1 , Z2 , . . . i in countably
many generators Zi .

19
Proposition 3.9. Z ⊗ R is a universal Leibnitz-Hopf algebra over the ring R.
Set deg Zi = 2i. Let us denote by M the graded dual Hopf algebra over the integers.
It is not difficult to see that M is precisely the algebra of quasi-symmetric functions over the integers.
Indeed, for any composition ω = (j1 , . . . , jk ) we can define mω by the dual basis formula

hmω , Zσ i = δω, σ ,

where Zσ = Za1 . . . Zal for a composition σ = (a1 , . . . , al ). Then the elements mω are multiplied exactly
as the quasi-symmetric monomials Mω .
Let us denote ∞
X
Φ(t) = 1 + Z1 t + Z2 t2 + · · · = Z k tk
k=0

Then the comultiplication formula is equivalent to

∆Φ(t) = Φ(t) ⊗ Φ(t)

Let χ : Z → Z be the antipode, that is a linear operator, satisfying the property

1 ⋆ χ = µ ◦ (1 ⊗ χ) ◦ ∆ = η ◦ ε = µ ◦ (χ ⊗ 1) ◦ ∆ = χ ⋆ 1,

where ε : Z → Z, ε(1) = 1, ε(Zσ ) = 0, σ 6= ∅, is a counit, η : Z → Z, η(a) = a · 1 is a unite map, and µ

is a multiplication in Z.
Then Φ(t)χ(Φ(t)) = 1 = χ(Φ(t))Φ(t) and {χ(Zn )} satisfy the recurrent formulas

χ(Z1 ) = −Z1 ; χ(Zn+1 ) + χ(Zn )Z1 + · · · + χ(Z1 )Zn + Zn+1 = 0, n > 1. (6)

In fact, since Φ(t) = 1 + Z1 t + . . . , we can use the previous formula to obtain

X ∞
1 i
χ(Φ(t)) = = (−1)i (Φ(t) − 1) ;
Φ(t) i=0
n
X X
χ(Zn ) = (−1)k Zj1 . . . Zjk ji > 1, n > 1
k=1 j1 +···+jk =n

This formula defines χ on generators, and therefore on the whole algebra.

Definition 3.10. Let Z op be the Hopf algebra opposite to Z, i.e. Z = Z op as coalgebras, and the
multiplication in Z op is given by the rule a ⋄ b := b · a. Then the antiisomorphism

̺ : Z → Z, Z a1 . . . Z ak → Z ak . . . Z a1

satisfies the property ∆ ◦ ̺ = (̺ ⊗ ̺) ◦ ∆, and defines the Hopf algebra isomorphism Z → Z op .

Proposition 3.11. We have ̺∗ = ∗
Proof. Indeed,
h̺∗ Mω , Zσ i = hMω , ̺Zσ i = hMω , Zσ∗ i = δω, σ∗ .
Therefore, ̺∗ Mω = Mω∗ = Mω∗
Definition 3.12 ([N]). A (left) Milnor module M over the Hopf algebra X is an algebra with unit
1 ∈ R which is also a (left) module over X satisfying
X X
x(uv) = x′n (u)x′′n (v), x ∈ X, u, v ∈ M, ∆x = x′n ⊗ x′′n .

Proposition 3.13. The homomorphism LR : Z → D(R) : Zk → dk defines on the rings P and RP

the structures of left Milnor modules over the Leibnitz-Hopf algebra Z.
The homomorphism RR : Z op → D(R) : RR = LR ◦ ̺ defines on the rings P and RP the structures
of right Milnor modules over the Leibnitz-Hopf algebra Z.

20
Proof. This follows from Proposition 2.49
Let us note, that the word Zω = Zj1 . . . Zjk corresponds under the homomorphism LR to the operator
Dω = dj1 . . . djk , and under the homomorphism RR to the operator Dω∗ = djk . . . dj1 .
Definition 3.14. Denote by U the universal Hopf algebra in the category of Leibnitz-Hopf algebras
with an antipode χ(Hi ) = (−1)i Hi . It is is easy to see that U = Z/JU , where the two-sided Hopf ideal
JU is generated by the relations

Zn − Z1 Zn−1 + · · · + (−1)n−1 Zn−1 Z1 + (−1)n Zn = 0, n > 2 (7)

These relations can be written in the short form as Φ(−t)Φ(t) = 1.

id ⊗1
It can be shown that U ⊗ Q ≃ QhZ1 , Z3 , Z5 , . . . i, and U −−−→ U ⊗ Q is an embedding.
Corollary 3.15. The mappings LR and RR define on the rings P and RP the structures of left and
right Milnor modules over the Leibnitz-Hopf algebra U.

Proof. This follows from Proposition 2.52.

Since ̺(JU ) = JU , we see that ̺ induces a correctly defined homomorphism ̺ : U → U, which gives
the isomorphism U ≃ U op .
Since the factor map Z → Z/JU is an epimorphism, the dual map U ∗ → Z ∗ = Qsym is an
embedding. Then the subring U ∗ ⊂ Qsym is invariant under the involution ̺∗ = ∗.
Definition 3.16. A universal commutative Leibnitz-Hopf algebra C = Z[C1 , C2 , . . . ] is a free commu-
tative polynomial Leibnitz-Hopf algebra in generators Ci of degree 2i. We have C = Z/JC , where the
ideal JC is generated by the relations Zi Zj − Zj Zi .
It is a self-dual Hopf algebra and the graded dual Hopf algebra is naturally isomorphic to the algebra
of symmetric functions Z[σ1 , σ2 , . . . ] = Sym[t1 , t2 , . . . ] ⊂ Qsym[t1 , t2 , . . . ] generated by the symmetric
monomials X
σi = Mωi = t l1 . . . t li ,
l1 <···<li

where ωi = (1, . . . , 1).

| {z }
i
The isomorphism C ≃ C ∗ is given by the correspondence Ci → σi .
Definition 3.17. Denote by US the universal Hopf algebra in the category of commutative Leibnitz-
Hopf algebras with an antipode χ(Hi ) = (−1)i Hi . It is easy to see that US = Z/JU S , where the
two-sided ideal JU S is generated by the relations (7), and the commutators Zi Zj − Zj Zi . Then the
relations (7) for odd n are trivial, while for even n we have
k−1
X
2Z2k = 2 (−1)i−1 Zi Z2k−i + (−1)k−1 Zk2
i=1

It can be shown that US ⊗ Q ≃ Q[Z1 , Z3 , Z5 , . . . ].

3.3 Lie-Hopf Algebras

Definition 3.18. A Lie-Hopf algebra over the ring R is an associative Hopf algebra L with a fixed
sequence of a finite or countable number of multiplicative generators Li , i = 1, 2 . . . satisfying the
comultiplication formula
∆Li = 1 ⊗ Li + Li ⊗ 1, L0 = 1.
A universal Lie-Hopf algebra over the ring R is a Lie-Hopf algebra A satisfying the universal prop-
erty: for any Lie-Hopf algebra L over the ring R the correspondence Ai → Li defines a Hopf algebra
homomorphism A → L.
Consider the free associative Lie-Hopf algebra W over the integers ZhW1 , W2 , . . . i in countably many
variables Wi .

21
Proposition 3.19. W ⊗ R is a universal Lie-Hopf algebra over the ring R.
Let us set deg Wi = 2i. Then the graded dual Hopf algebra is denoted by N . This is the so-called
shuffle algebra.
The antipode χ in W has a very simple form χ(Wi ) = −Wi .

3.4 Lyndon Words

A well-known theorem in the theory of free Lie algebras (see, for example, [Re]) states that the algebra
N ⊗ Q is a commutative free polynomial algebra in the so-called Lyndon words.
Definition 3.20. Let us denote by [a1 , . . . , an ] an element of N∗ , that is the word over N, consisting
of symbols a1 , . . . , an , ai ∈ N. Let us order the words from N∗ lexicographically, where any sym-
bol is larger than nothing, that is [a1 , . . . , an ] > [b1 , . . . , bm ] if and only if there is an i such that
a1 = b1 , . . . , ai−1 = bi−1 , ai > bi , 1 6 i 6 min{m, n}, or n > m and a1 = b1 , . . . , am = bm .
A proper tail of a word [a1 , . . . , an ] is a word of the form [ai , . . . , an ] with 1 < i 6 n. (The empty
word and one-symbol word have no proper tails.)
A word is Lyndon if all its proper tails are larger than the word itself. For example, the words
[1, 1, 2], [1, 2, 1, 2, 2], [1, 3, 1, 5] are Lyndon and the words [1, 1, 1, 1], [1, 2, 1, 2], [2, 1] are not Lyndon. The
set of Lyndon words is denoted by LYN.
The same definitions make sense for any totally ordered set, for example, for the set {1, 2} or for the
set of all odd positive integers.
The role of Lyndon words is described by the following theorem:
Theorem (Chen-Fox-Lyndon Factorization [CFL]). Every word w in N∗ factors uniquely into a de-
creasing concatenation product of Lyndon words

w = u1 ∗ u2 ∗ · · · ∗ uk , ui ∈ LYN, u1 > u2 > · · · > uk

For example, [1, 1, 1, 1] = [1] ∗ [1] ∗ [1] ∗ [1], [1, 2, 1, 2] = [1, 2] ∗ [1, 2], [2, 1] = [2] ∗ [1].
The algebra N is additively generated by the words in N∗ . The word w = [a1 , . . . , an ] corresponds
to the function
h[a1 , . . . , an ], Wσ i = δw, σ
where σ is a composition σ = (b1 , . . . , bl ), Wσ = Wb1 . . . Wbl , and for the word [a1 , . . . , an ]
(
1, σ = (a1 , . . . , an );
δw, σ =
0, else

The multiplication in the algebra N is the so-called shuffle multiplication:

X
[a1 , . . . , an ] ×sh [an+1 , . . . , am+n ] = [aσ(1) , . . . , aσ(n) , aσ(n+1) , . . . , aσ(n+m) ],
σ

where σ runs over all the substitutions σ ∈ Sm+n such that

σ −1 (1) < · · · < σ −1 (n) and σ −1 (n + 1) < · · · < σ −1 (n + m).

For example,

[1] ×sh [1] = [1, 1] + [1, 1] = 2[1, 1];

[1] ×sh [2, 3] = [1, 2, 3] + [2, 1, 3] + [2, 3, 1];
[1, 2] ×sh [1, 2] = [1, 2, 1, 2] + [1, 1, 2, 2] + [1, 1, 2, 2] + [1, 1, 2, 2] + [1, 1, 2, 2] + [1, 2, 1, 2] = 2[1, 2, 1, 2] + 4[1, 1, 2, 2].

There is a well-known shuffle algebra structure theorem:

Theorem. N ⊗ Q = Q[LYN], the free commutative algebra over Q in the symbols from LYN.

22
 The proof follows from the following theorem concerning shuffle products in connection with Chen-
Fox-Lyndon factorization.
Theorem. Let w be a word on the natural numbers and let w = u1 ∗ u2 ∗ · · · ∗ um be its Chen-
Fox-Lyndon factorization. Then all words that occur with nonzero coefficient in the shuffle product
u1 ×sh u2 ×sh · · · ×sh um are lexicographically smaller or equal to w, and w occurs with a nonzero
coefficient in this product.
Given this result it is easy to prove the shuffle algebra structure theorem in the case of arbitrary
totally ordered subset M = {m1 , m2 , . . . } ⊂ N. Let us denote by WM the free associative Lie-Hopf
algebra ZhWm1 , Wm2 , . . . i, let NM be its graded dual algebra, and LYNM be the corresponding set of
all Lyndon words. We need to prove that NM ⊗ Q = Q[LYNM ].
Let m1 be the minimal number in M . The smallest word [m1 ] is Lyndon.
Given a word w we can assume by induction that all the words lexicographically smaller than w
have been written as polynomials in the elements of LYNM . Take the Chen-Fox-Lyndon factorization
w = u1 ∗ u2 ∗ · · · ∗ um of w and consider, using the preceding theorem,

u1 ×sh u2 ×sh · · · ×sh um = aw + (reminder).

By the theorem the coefficient a is nonzero and all the words in (reminder) are lexicographically smaller
than w, hence they belong to Q[LYNM ]. Therefore w ∈ Q[LYNM ]. This proves generation. Since each
monomial in (2n)-th graded component of N ⊗ Q has a unique Chen-Fox-Lyndon decomposition, the
number of monomials in Q[LYNM ] of graduation 2n is equal to the number of monomials of the same
graduation in NM ⊗ Q, monomials in Lyndon words are linearly independent. This proves that Lyndon
words are algebraically independent.
Corollary 3.21. The graded dual algebras to the free associative Lie-Hopf algebras W12 = ZhW1 , W2 i
and Wodd = ZhW1 , W3 , W5 , . . . i are free polynomial algebras in Lyndon words LYN12 and LYNodd
respectively.
The correspondence

1 + Z1 t + Z2 t2 + · · · + = exp(W1 t + W2 t2 + . . . );

defines an isomorphism of Hopf algebras Z ⊗ Q ≃ W ⊗ Q.

However it is not true that N is a free polynomial commutative algebra over the integers.

4 Topological realization of Hopf algebras

In [BR] A. Baker and B. Richter showed that the ring of quasi-symmetric functions has a very nice
topological interpretation.
We will consider CW -complexes X and their homology H∗ (X) and cohomology H∗ (X) with integer
coefficients.
Let us assume that homology groups H∗ (X) have no torsion.
The diagonal map X → X × X defines in H∗ (X) the structure of a graded coalgebra with the
comultiplication ∆ : H∗ (X) → H∗ (X) ⊗ H∗ (X) and the dual structure of a graded algebra in H∗ (X)
with the multiplication ∆∗ : H∗ (X) ⊗ H∗ (X) → H∗ (X).
In the case when X is an H-space with the multiplication µ : X × X → X we obtain the Pontryagin
product µ∗ : H∗ (X) ⊗ H∗ (X) → H∗ (X) in the coalgebra H∗ (X) and the corresponding structure of a
graded Hopf algebra on H∗ (X). The cohomology ring H∗ (X) obtains the structure of a graded dual
Hopf algebra with the diagonal mapping µ∗ : H∗ (X) → H∗ (X) ⊗ H∗ (X).
For any space Y the loop space X = ΩY is an H-space. A continuous mapping f : Y1 → Y2 induces a
mapping of H-spaces Ωf : X1 → X2 , where Xi = ΩYi . Thus for any space Y such that X = ΩY has no
torsion in integral homology we obtain the Hopf algebra H∗ (X). This correspondence is functorial, that
is any continuous mapping f : Y1 → Y2 induces a Hopf algebra homomorphism f∗ : H∗ (X1 ) → H∗ (X2 )
By the Bott-Samelson theorem [BS], H∗ (ΩΣX) is the free associative algebra T (H e ∗ (X)) generated by
e
H∗ (X). This construction is functorial, that is a continuous mapping f : X1 → X2 induces a ring homo-
morphism of the corresponding tensor algebras arising from the mapping f∗ : He ∗ (X1 ) → H e ∗ (X2 ). Denote

23
 e ∗ (X). Since the diagonal mapping ΣX → ΣX × ΣX
elements of H∗ (ΩΣX) by (a1 | . . . |an ), where ai ∈ H
gives the H-map ∆ : ΩΣX → ΩΣX × ΩΣX, it follows from the Eilenberg-Zilber theorem that

∆∗ (a1 | . . . |an ) = (∆∗ a1 | . . . |∆∗ an )

where (a1 ⊗ b1 |a2 ⊗ b2 ) = (a1 |a2 ) ⊗ (b1 |b2 ).

There is a nice combinatorial model for any topological space of the form ΩΣX with X connected,
namely the James construction JX on X. After one suspension this gives rise to a splitting
_
ΣΩΣX ∼ ΣJX ∼ ΣX (n) ,
n>1

where X (n) denotes the n-fold smash power of X.

Example 4.1. There exists a homotopy equivalence

_ _ 
ΣΩΣS 2 → Σ(S 2 )(n) ≃ Σ S 2n .
n>1 n>1
 W 
Therefore there exists an H-map ΩΣ(ΩΣS 2 ) → ΩΣ S 2n that is a homotopy equivalence.
n>1

Using these classical topological results let us describe topological realizations of the Hopf algebras
we study in this work.
We have the following Hopf algebras:
I. 1. H∗ (CP ∞ ) is a divided

power algebra Z[u1 , u2 , . . . ]/I, where the ideal I is generated by the
relations ui uj − i+j
i ui+j , with the comultiplication

n
X
∆un = uk ⊗ un−k ; (8)
k=0

2. H∗ (CP ∞ ) = Z[u] – a polynomial ring with the comultiplication

∆u = 1 ⊗ u + u ⊗ 1;

II. 1. H∗ (ΩΣCP ∞ ) ≃ T (H̃∗ (CP ∞ )) = Zhu1 , u2 , . . . i with ui being non-commuting variables of

degree 2i. Thus there is an isomorphism of rings

H∗ (ΩΣCP ∞ ) ≃ Z

under which un corresponds to Zn .

The coproduct ∆ on H∗ (ΩΣCP ∞ ) induced by the diagonal in ΩΣCP ∞ is compatible with
the one in Z: X
∆un = ui ⊗ uj
i+j=n

So there is an isomorphism of graded Hopf algebras. This gives a geometric interpretation

for the antipode χ in Z : in H∗ (ΩΣCP ∞ ) it arises from the time-inversion of loops.
2. H∗ (ΩΣCP ∞ ) is the graded dual Hopf algebra to H∗ (ΩΣCP ∞ ).

Theorem. [BR] There is an isomorphism of graded Hopf algebras:

H∗ (ΩΣCP ∞ ) ≃ Z = ZhZ1 , Z2 , . . . i, H∗ (ΩΣCP ∞ ) ≃ M = Qsym[t1 , t2 , . . . ].

III. H∗ (BU ) ≃ H∗ (BU ) ≃ Z[σ1 , σ2 , . . . ] ≃ C. It is a self-dual Hopf algebra of symmetric functions. In

the cohomology σi are represented by Chern classes.

24
IV. 1. H∗ (ΩΣS 2 ) = H∗ (ΩS 3 ) = Z[w] – a polynomial ring with deg w = 2 and the comultiplication

∆w = 1 ⊗ w + w ⊗ 1;

2. H∗ (ΩΣS 2 ) = Z[un ]/I is a divided power algebra. Thus H∗ (ΩΣS 2 ) ≃ H∗ (CP ∞ ).

V. H∗ (ΩΣ(ΩΣS 2 )) ≃ Zhw1 , w2 , . . . i. It is a free associative Hopf algebra with the comultiplication
n  
X n
∆wn = wk ⊗ wn−k (9)
k
k=0

  W 
VI. H∗ ΩΣ S 2n ≃ Zhξ1 , ξ2 , . . .i. It is a free associative algebra and has the structure of a
n>1
graded Hopf algebra with the comultiplication

∆ξn = 1 ⊗ ξn + ξn ⊗ 1. (10)
  W 
Therefore, H∗ ΩΣ S 2n gives a topological realization of the universal Lie-Hopf algebra W.
n>1

The homotopy equivalence _ 

a : ΩΣ S 2n −→ ΩΣ(ΩΣS 2 )
n>1

induces an isomorphism of graded Hopf algebras

a∗ : Zhξ1 , ξ2 , . . .i −→ Zhw1 , w2 , . . .i,

and its algebraic form is determined by the conditions:

∆a∗ ξn = (a∗ ⊗ a∗ )(∆ξn ).

For example, a∗ ξ1 = w1 , a∗ ξ2 = w2 − w1 |w1 , a∗ ξ3 = w3 − 3w2 |w1 + 2w1 |w1 |w1 .

Thus using topological results we have obtained that two Hopf algebra structures on the free asso-
ciative algebra with the comultiplications (9) and (10) are isomorphic over Z.
This result is interesting from the topological point of view, since the elements (wn − a∗ ξn ) for n > 2
are obstructions to the desuspension of the homotopy equivalence
_ 
Σ(ΩΣS 2 ) → Σ S 2n .
n>1

We have the commutative diagram:

ΩΣk
ΩΣ(ΩΣS 2 ) / ΩΣCP ∞
O I
O
Ij
I
I
I$
2 k / CP ∞ i / BU det / CP ∞
ΩΣS

Here
• i is the inclusion CP ∞ = BU (1) ⊂ BU ,
• j arises from the universal property of ΩΣCP ∞ as a free H-space;
• the mapping CP ∞ → ΩΣCP ∞ is induced by the identity map ΣCP ∞ → ΣCP ∞ ,
• the mapping k : ΩΣS 2 → CP ∞ corresponds to the generator of H2 (ΩΣS 2 ) = Z.
• det is the mapping of the classifying spaces BU → BU (1) induced by the mappings
det : U (n) → U (1), n = 1, 2, . . . .

25
 As we have mentioned, H∗ (BU ) as a Hopf algebra is isomorphic to the algebra of symmetric functions
(in the generators – Chern classes) and is self-dual.

H∗ (BU ) ≃ H∗ (BU ) ≃ Sym[t1 , t2 , . . . ] = Z[σ1 , σ2 , . . . ].

Since CP ∞ gives rise to the algebra generators in H∗ (BU ), j∗ is an epimorphism on homology, and j ∗
is a monomorphism on cohomology.
• j∗ corresponds to the factorization of the non-commutative polynomial algebra Z by the commu-
tation relations Zi Zj − Zj Zi and sends Zi to σi .
• j ∗ corresponds to the inclusion Sym[t1 , t2 , . . . ] ⊂ Qsym[t1 , t2 , . . . ] and sends σi to Mωi , where
ωi = (1, 1, . . . , 1).
| {z }
i
∗
• k sends u to u1 and defines an embedding of the polynomial ring Z[u] into the divided power
algebra Z[un ]/I.
• the composition det∗ ◦j∗ maps the algebra Z to the divided power algebra H∗ (CP ∞ ). This
corresponds to the mapping Zi → dk |Ps of Z to D(Ps ).

5 Structure of D
Theorem 5.1. The homomorphism LR : Z → D(R) : Zk → dk induces a ring isomorphism:

D(R) ≃ U = Z/JU ,

where U is a universal Leibnitz-Hopf algebra with the antipode χ(Ui ) = (−1)i Ui .

Proof. Let us prove the following lemma:
Lemma 5.2. Let D ∈ D(R) be an operator of graduation 2k. Then for each space R2n , n > k there is
a unique representation
D = u(d2 , d3 , . . . , dk ) + dw(d2 , d3 , . . . , dk−1 ), (11)
and there is a unique representation

D = u′ (d2 , d3 , . . . , dk ) + w′ (d2 , . . . , dk−1 )d, (12)

where u, u′ , w, w′ are polynomials.

Proof. For k = 0 and 1 it is evident.
Since 2d2 = d2 , this is true for k = 2. Let k > 3.
Using the relations (2.52) we obtain


ddi = (−1)i di d + d2 di−1 − d3 di−2 + · · · + (−1)i−1 di−1 d2 + (1 + (−1)i+1 )di+1

So the expressions (11) and (12) exist. P P

Let u(d2 , d3 , . . . , dk ) + dw(d2 , . . . , dk−1 ) = 0, where u = aω Dω , and w = bω Dω .
|ω|=k |ω|=k−1
Then for any n-dimensional polytope P n , n > k we have

ξ1 (u + dw)P n = 0, ε1 (u + dw)P n = 0

Since u + dw has degree 2k, this equality can be written as

X X
aω fn−k, n−k+j1 , n−k+j1 +j2 , ..., n−jl (P n ) + bω fn−k, n−k+1, n−k+1+j1 , ..., n−jl (P n ). (13)
|ω|=k |ω|=k−1

26
Using the generalized Dehn-Sommerville equations we obtain
 
n−k−1
X
fn−k, n−k+1, n−k+1+j1 , ..., n−jl = (−1)n−k−1  (−1)j fj, n−k+1, n−k+1+j1 , ..., n−jl  +
j=0

+ (1 + (−1)n−k )fn−k+1, n−k+1+j1 , ..., n−jl

All the sets {n − k, n − k + j1 , n − k + j1 + j2 , . . . , n − jl }, {j, n − k + 1, n − k + j1 , . . . , n − jl } and

{n − k + 1, n − k + 1 + j1 , . . . , n − jl } for all ω are different and belong to Ψn .
In the case of RP and P k−1 we obtain that n = k − 1, therefore n − k = −1 in the expression (13)
vanishes and we obtain
X X
aω fj1 −1, j1 +j2 −1, ..., n−jl (P n ) + bω f0, j1 , ..., n−jl (P n ).
|ω|=k |ω|=k−1

Again all he sets {j1 − 1, j1 + j2 − 1, . . . , n − jl }, {0, j1, . . . , n − jl } for all ω are different and belong to
Ψn .
Since the vectors {fS (Q), S ∈ Ψn }, Q ∈ Ωn are linearly independent, we obtain that all aω and bω
are equal to 0, so the representation (11) is unique.
We obtain that the monomials Dω = dj1 . . . djl , |ω| = k, ji > 2 and dDω = ddj1 . . . djl |ω| = k−1, ji > 2
form a basis of the abelian group D(R2n , R2(n−k) ).
But each monomial dDω , |ω| = k, ji > 2 can be expressed as an integer combination of the
monomials Dω′ and Dω′ d. So the monomials Dω and Dω d also form a basis. This proves the second
part of the lemma.
Now let us prove the theorem. The mapping LR : Z → D(R) : Zi → di is an epimorphism. Let
z ∈ Z such that deg z = 2k and LR z = 0.
X
z= aω Z ω .
|ω|=k

We know that
k−2
X k
X
Z1 Zk−1 = (−1)k−1 Zk−1 Z1 + (−1)i Zi Zk−i + (1 + (−1)k )Zk − (−1)i Zi Zk−i .
i=2 i=0

So X X
z= a′ω Zω + bω Zω Z1 + z ′ ,
|ω|=k, ji >2 |ω|=k−1, ji >2
′ ′
where z ∈ JU . Since LR z = LR z = 0, we obtain
X X
a′ω Dω + bω Dω d = 0,
|ω|=k, ji >2 |ω|=k−1, ji >2

Lemma 5.2 implies that all the coefficients a′ω and bω are equal to 0, so z = z ′ ∈ JU . This proves that
D(R) ≃ Z/JU .
Remark 5.3. Now we see that the correspondence dk → dk defines a natural isomorphism between
D(RP) and D(P). Therefore from this moment we will often use the notation D for both these algebras,
and the notations L, R for the homomorphisms LR and RR .
P
Corollary 5.4. The comultiplication dk → di ⊗ dj defines on D the structure of a Leibnitz-Hopf
i+j=k
algebra over the ring Z. This Hopf algebra D is isomorphic to the Hopf algebra U, universal in the
category of Leibnitz-Hopf algebras with the antipode χ(Ui ) = (−1)i Ui .
Both P and RP have natural left Milnor module structures over D
It is evident that Dω |R[2n] = 0 for |ω| > n. The following corollary says that the other relations
between operators in D on the abelian group R[2n] are the same as on the whole ring R.

27
Corollary 5.5. We have
D(R[2n] ) = D(R)/Jn
where the ideal Jn is generated by all the operators Dω , |ω| > n.
Proof. As in the proof of Theorem 5.1 using Lemma 5.2 it is easy to show that D(R[2n] ) ≃ Z/JD(R[2n] ) ,
where the ideal JD(R[2n] ) is generated by the the relations Φ(−t)Φ(t) = 1 and Zω = 0 for |ω| > n.
Corollary 5.6. The operators d2 , d3 , d4 , . . . are algebraically independent.
Corollary 5.7. Rank of the (2n)-th graded component of the ring D is equal to the (n − 1)-th Fibonacci
number cn−1 .
Proof. At first, let us calculate rank rkn of the (2n)-th graded component of the ring Zhd2 , d3 , d4 , . . . i.

rk0 = 1, rk1 = 0, rk2 = 1, rk3 = 1, rk4 = 2, rk5 = 3, . . .

It is easy to see that there is a recursive relation

rkn+1 = rkn−1 + rkn−2 + · · · + rk2 +1, n > 3

Then rkn+1 = rkn−1 + rkn . Since rk2 = rk3 = 1, we obtain rkn = cn−2 , n > 2. At last, rank of the
(2n)-th graded component of the ring D is equal to rkn−1 + rkn = cn−3 + cn−2 = cn−1 for n > 3. It is
easy to see that for n = 1 and 2 it is also true.
dk
Let us remind that on the ring of simple polytopes dk |Ps = k! , so the ring D(Ps ) is isomorphic

Ps
to the divided power ring dk dl = k+l
k dk+l .

Proposition 5.8.
D ⊗ Q = Qhd1 , d3 , d5 , . . . i,
The inclusion D ⊂ D ⊗ Q is an embedding, and the operators d2k are expressed in terms of the operators
d1 , d3 , . . . , d2k−1 by the formulas
k 2i−2

 
X X
d2k = (−1)i−1 i−1  d2j1 −1 . . . d2j2i −1  . (14)
i22i−1
i=1 j1 +j2 +···+j2i =i+k, jl >1

Proof. Let us denote X X

a(t) = d2k t2k , b(t) = d2k+1 t2k+1 .
k>0 k>0

Then

Φ(t) = a(t) + b(t) Φ(−t) = a(t) − b(t);

Φ(t) + Φ(−t) Φ(t) − Φ(−t) Φ(t)2 − Φ(−t)2
a(t)b(t) = · = = b(t)a(t).
2 2 4
Then the relation Φ(−t)Φ(t) = 1 is equivalent to the relations

a(t)2 − b(t)2 = 1;
a(t)b(t) = b(t)a(t).
p
Therefore a(t) = 1 + b(t)2 and the formula (14) is valid. Consequently all the operators d2k are
expressed as polynomials in d1 , d3 , . . . with rational coefficients. For example,
d2 dd3 + d3 d d4
d2 = , d4 = − .
2 2 8
This means that the algebra D ⊗ Q is generated by d1 , d3 , d5 , . . . . On the other hand, let us calculate
the number of the monomials d2j1 −1 . . . d2jk −1 with (2j1 − 1) + · · · + (2jk − 1) = n. Denote this number
by ln . Then
l0 = 1, l1 = 1, l2 = 1, l3 = 2, l4 = 3, . . . ,

28
and there is a recursive formula ln+1 = ln + ln−2 + ln−4 + . . . for n > 2. Hence ln+1 = ln + ln−1 for
n > 2, and ln = cn−1 for n > 1.
We see that the number of monomials of degree 2n is equal to dimension of the (2n)-th graded
component according to Corollary 5.7. This implies that they are linearly independent over the rationals.
Therefore d1 , d3 , d5 , . . . are algebraically independent.
At last, the inclusion D ⊂ D ⊗ Q is an embedding, since D is torsion-free.
Definition 5.9. Let us define operators sk by the formula
s(t) = s1 t + s2 t2 + s3 t3 + · · · = log Φ(t).
3
Then s1 = d1 , s2 = 0, s3 = d3 − d6 , and so on.
The relation Φ(−t)Φ(t) = 1 turns into the relation s(−t) + s(t) = 0. Thus s2k = 0 for all k.
Also we have ∆s(t) = 1 ⊗ s(t) + s(t) ⊗ 1, so each operator s2k−1 is a derivation.
Proposition 5.10. There is an isomorphism of Hopf algebras
D ⊗ Q = Qhs1 , s3 , s5 , . . . i,
where Qhs1 , s3 , s5 , . . . i is a free Lie-Hopf algebra in the generators of degree deg s2k−1 = 2(2k − 1),
k > 1, the comultiplication ∆s2k−1 = 1 ⊗ s2k−1 + s2k−1 ⊗ 1, and the antipode χ(s2k−1 ) = −s2k−1

6 Generalized f -polynomial
6.1 Definition
The mapping Φ : P → P[t] is a ring homomorphism. If we set Φ(t2 )t1 = t1 , then the operator Φ(t2 )
defines a ring homomorphism P → P[t1 , t2 ]: P → Φ(t2 )Φ(t1 )P ,
Thus for each n > 0 we can define the ring homomorphism Φn : P → P[[t1 , . . . , tn ]] as a composition:
Φn (t1 , . . . , tn )P = Φ(tn ) . . . Φ(t1 )P.
Definition 6.1. Let P n be an n-dimensional polytope. Let us define a quasi-symmetric function
min{r,n}
X X
fr (α, t1 , . . . , tr )(P n ) = ξα Φ(tr ) . . . Φ(t1 )P n = αn + fa1 , ..., ak αa1 M(n−ak , ..., a2 −a1 ) .
k=1 06a1 <···<ak 6n−1

For n = 1 we obtain f1 (α, t1 )(P n ) = αn + fn−1 αn−1 t1 + · · · + f0 tn1 is a homogeneous f -polynomial

in two variables ([Buch]). So the polynomial fn is a generalization of the f -polynomial.
Consider the increasing sequence of rings
Z[α] ⊂ Z[α, t1 ] ⊂ Z[α, t1 , t2 ] ⊂ . . .
with the restriction maps Er : Z[α, t1 , . . . , tr+1 ] → Z[α, t1 , . . . , tr ]
(Er g)(α, t1 , . . . , tr ) = g(α, t1 , . . . , tr , 0).
Since fr+1 (α, t1 , . . . , tr , 0)(P n ) = fr (α, t1 , . . . , tr )(P n ), we obtain the ring homomorphism
f : P → Qsym[t1 , t2 , . . . ][α] ⊂ lim Z[α, t1 , . . . , tr ] :
←−
r
n
X X
f (α, t1 , t2 , . . . )(P n ) = αn + fa1 , ..., ak αa1 M(n−ak , ..., a2 −a1 ) .
k=1 06a1 <···<ak 6n−1

Definition 6.2. For a set S = {a1 , . . . , ak } ⊂ {0, 1, . . . , n − 1} let us denote by ω(S) the composition
(n − ak , ak − ak−1 , . . . , a2 − a1 ) of the number n − a1 .
Then we can write X X
f (P n ) = fS αa1 Mω(S)
k>0 S: l(S)=k

29
6.2 Image
It follows from the formula, that the restriction

f (α, t1 , . . . , tr , tr+1 , . . . ) → f (α, t1 , . . . , tr , 0, 0, . . . ) = fr (α, t1 , . . . , tr )

is injective on the space of all n-dimensional polytopes, n 6 r.

Theorem 6.3. The image of the space P 2n generated by all n-dimensional polytopes in the ring
Qsym[t1 , . . . , tr ][α], r > n under the mapping fr consists of all the homogeneous polynomials g of
degree 2n satisfying the equations
1.

g(α, t1 , −t1 , t3 , . . . , tr ) = g(α, 0, 0, t3 , . . . , tr );

g(α, t1 , t2 , −t2 , t4 , . . . , tr ) = g(α, t1 , 0, 0, t4 . . . , tr );
...
g(α, t1 , . . . , tr−2 , tr−1 , −tr−1 ) = g(α, t1 , . . . , tr−2 , 0, 0);

2.
g(−α, t1 , . . . , tr−1 , α) = g(α, t1 , . . . , tr−1 , 0);

These equations are equivalent to the Bayer-Billera (generalized Dehn-Sommerville) relations.

Proof. Φ(−t)Φ(t) = 1 = Φ(0)Φ(0), therefore

ξα Φ(tr ) . . . Φ(t3 )Φ(−t1 )Φ(t1 )P n = ξα Φ(tr ) . . . Φ(t3 )Φ(0)Φ(0)P n ;

ξα Φ(tr ) . . . Φ(t4 )Φ(−t2 )Φ(t2 )Φ(t1 )P n = ξα Φ(tr ) . . . Φ(t4 )Φ(0)Φ(0)Φ(t1 )P n ;
...
ξα Φ(−tr−1 )Φ(tr−1 )Φ(tn−2 ) . . . Φ(t1 )P n = ξα Φ(0)Φ(0)Φ(tr−2 ) . . . Φ(t1 )P n .

On the other hand, Proposition 2.53 gives the last relation

ξ−α Φ(α)Φ(tr−1 ) . . . Φ(t1 )P n = ξα Φ(0)Φ(tr−1 ) . . . Φ(t1 )P n .

Now let us proof the opposite inclusion, that is if the homogeneous polynomial g of degree 2n satisfies
the conditions of the theorem, then g = fr (pn ) for some pn ∈ P 2n .
Lemma 6.4. Let P n be an n-dimensional polytope. Then
1. The equation
fr (α, t1 , . . . , tq , −tq , . . . , tr ) = fr (α, t1 , . . . , 0, 0, . . . , tr )
is equivalent to the generalized Dehn-Sommerville relations
at+1 −1
X

(−1)j−at −1 fa1 , ..., at , j, at+1 , ..., ak = 1 + (−1)at+1 −at fa1 , ..., at , at+1 , ..., ak
j=at +1

for 1 6 k 6 min{r − 1, n − 1}, k + 1 − q 6 t 6 r − q.

2. The equation
fr (−α, t1 , . . . , tr−1 , α) = fr (α, t1 , . . . , tr−1 , 0)
is equivalent to the generalized Dehn-Sommerville relations
1 −1
aX
(−1)j fj, a1 , ..., ak = (1 + (−1)a1 −1 )fa1 , ..., ak
j=0

for 0 6 k 6 min{r − 1, n − 1}.

30
Proof.

fr (α, t1 , . . . , tq , −tq , . . . , tr ) =
 
min{r, n}
X X X
αn + fa1 , ..., ak αa1  tln−a
1
k
. . . talk2 −a1  =
k=1 06a1 <···<ak 6n−1 16l1 <···<lk 6r
tq+1 =−tq
 
min{r−2, n}
X X X
= αn + fa1 , ..., ak αa1  tln−a
1
k
. . . talk2 −a1  +
k=1 06a1 <···<ak 6n−1 16l1 <···<lk 6r, lj 6=q, q+1
 
min{r−1, n}
X X X a −a
+ fa1 , ..., ak αa1  tln−a
1
k
. . . tq k+2−j k+1−j . . . talk2 −a1  +
k=1 06a1 <···<ak 6n−1 16l1 <···<lj =q<···<lk 6r, lj+1 6=q+1
 
min{r−1, n}
X X X
+ fa1 , ..., ak αa1  tln−a
1
k
. . . (−tq )ak+1−j −ak−j . . . talk2 −a1  +
k=1 06a1 <···<ak 6n−1 16l1 <···<lj+1 =q+1<···<lk 6r, lj 6=q
 
min{r, n}
X X X a −a
fa1 , ..., ak αa1  tln−a
1
k
. . . (−1)ak+1−j −ak−j tq k+2−j k−j . . . talk2 −a1 
k=1 06a1 <···<ak 6n−1 16l1 <···<lj =q<lj+1 =q+1<···<lk 6r

The first and the second summands form exactly f (α, t1 , . . . , 0, 0, . . . , tr ). Therefore all the coefficients
of the polynomial consisting of the last three summands should be equal to 0.
a −a
Consider the monomial αa1 tln−a
1
k
. . . tq t+1 t . . . talk2 −a1 . Here q = lk+1−t . The existence of a monomial
of this form in the sum is equivalent to the conditions

k 6 min{r − 1, n − 1}, k + 1 − t 6 q, t − 1 6 r − q − 1.

The coefficient of the monomial should be equal to 0. This is equivalent to the relation:
at+1 −1

X
1 + (−1)at+1 −at fa1 , ..., at , at+1 , ..., ak + (−1)j−at fa1 , ..., at , j, at+1 , ak = 0
j=at +1

Now let us consider the remaining relation fr (−α, t1 , . . . , tr−1 , α) = fr (α, t1 , . . . , tr−1 , 0):
 
min{r−1, n}
X X X
= (−α)n + fa1 , ..., ak (−α)a1  tln−a
1
k
. . . talk2 −a1  +
k=1 06a1 <···<ak 6n−1 16l1 <···<lk 6r−1
 
min{r, n}
X X X
3 −a2 a2 −a1 
+ fa1 , ..., ak (−α)a1  tln−a
1
k
. . . talk−1 α =
k=1 06a1 <···<ak 6n−1 16l1 <···<lk =r
 
min{r−1, n}
X X X
αn + fa1 , ..., ak αa1  tln−a
1
k
. . . talk2 −a1 
k=1 06a1 <···<ak 6n−1 16l1 <···<lk 6r−1

This is equivalent to the relations

1 −1
aX
(−1)a1 fa1 , ..., ak + (−1)j fj, a1 , ..., ak = fa1 , ..., ak
j=0

for 0 6 k 6 min{r − 1, n − 1}. If k = 0, then the corresponding relation is exactly the Euler formula

(−1)n + (−1)n−1 fn−1 + · · · + f2 − f1 + f0 = 1.

31
Corollary 6.5. For r > n relations 1. and 2. of the theorem are equivalent to the generalized Dehn-
Sommerville relations for the polytope P n : For S ⊂ {0, . . . , n − 1}, and {i, k} ⊆ S ∪ {−1, n} such that
i < k − 1 and S ∩ {i + 1, . . . , k − 1} = ∅:
k−1
X
(−1)j−i−1 fS∪{j} = (1 − (−1)k−i−1 )fS .
j=i+1

Proof. We see that relations 1. and 2. follow from the generalized Dehn-Sommerville relations.
On the other hand, if i = −1 then the corresponding relation follows from equation 2.
If S = {a1 , . . . , at , at+1 , . . . , as }, i = at > 0, and k = at+1 ; or S = {a1 , . . . , at }, i = at , and k = n,
then we can take q such that s + 1 − t 6 q 6 r − t.
Remark 6.6. For n-dimensional polytopes for different r > n not all the equations are independent.
For i = −1 relation (3) follows from equation 2. and all the equations of type 1. do not contain the
case i = −1.
On the other hand, let S = {a1 , . . . , at , at+1 , . . . , as }, i = at > 0, and k = at+1 ; or S = {a1 , . . . , at },
i = at , and k = n. Lemma 6.4 implies that relation (3) follows from the equation with s+1−t 6 q 6 r−t.
Let us denote a = s + 1 − t, b = r − t. There are two conditions for s, t, namely
1 6 s 6 n − 1 and 1 6 t 6 s.
Let us rewrite these conditions in terms of a and b: t = r − b; s = a + t − 1 = a − b + r − 1, so
r − n 6 b − a 6 r − 2; b 6 r − 1; 1 6 a
This gives us a triangle on the plane (a, b). Each point (a, b) of this triangle corresponds to the conditions
a 6 q 6 b, that is there should be the equation for q in the segment [a, b]. We can imagine that this
segment is the segment [(a, a), (a, b)] on the plane.
Thus for r = n all the equations for t1 , . . . , tn−1 are necessary, for r = n + 1 it is enough to take the
equations for q = 2, 4, 6, . . . , 2[ n2 ]. If r > 2n − 2 one equation
f (α, t1 , . . . , tn−1 , −tn−1 , tn+1 , . . . , tr ) = f (α, t1 , . . . , 0, 0, tn+1 , . . . , tr )
gives all the relations of type 1.
Now let us finish the proof of the theorem. If the homogeneous polynomial g ∈ Qsym[t1 , . . . , tr ][α]
of degree 2n satisfies all the relations of the theorem, then its coefficients satisfy the generalized Dehn-
Sommerville relations. Therefore all the coefficients are linear combinations of the coefficients ga1 , ..., ak ,
where S = {a1 , . . . , ak } ∈ Ψn . As we know, the vectors {fS (Q), S ∈ Ψn }, Q ∈ Ωn form a basis of the
abelian group of all the vectors {fS , S ∈ Ψn , fS ∈ Z} = Zcn . So the vector {gS , S ∈ Ψn } is an integer
combination of the vectors {fS (Q), S ∈ Ψn }, Q ∈ Ωn . This implies that the polynomial g is an integer
combination of the polynomials fr (Q), Q ∈ Ωn with the same coefficients.
Let us remind that rank of the space fr (P 2n ), r > n is equal to cn .
Proposition 6.7. Let r > 2, and let P n be an n-dimensional polytope. Then
fr (α, t1 , . . . , tr )(P n ) = f1 (α, t1 + · · · + tr )(P n )
if and only if P n is simple. Here f1 (α, t) is a usual homogeneous f -polynomial in two variables.
dk
Proof. On the ring of simple polytopes dk |Ps = k! , so Φ(t) = etd .
Ps
Then Φ(tr )Φ(tr−1 ) . . . Φ(t1 ) = Φ(t1 +· · ·+tr ), therefore fr (α, t1 , . . . , tr )(P n ) = f1 (α, t1 +· · ·+tr )(P n ).
On the other hand, let fr (α, t1 , . . . , tr )(P n ) = f1 (α, t1 + · · · + tr )(P n ).
Then f2 (α, t1 , t2 )(P n ) = f1 (α, t1 + t2 )(P n ). So
n−1
X X n−1
X
αn + fi αi (t1n−i + t2n−i ) + fij αi t1n−j tj−i
2 = αn + fi αi (t1 + t2 )n−i
i=0 06i<j6n−1 i=0

In particular, f01 = nf0 , since the coefficients of the monomial t1n−1 t2 on the left and on the right are
n
equal. Hence 2f1 = f01 = nf0 . This implies that the polytope P is simple.

32
Remark 6.8. Letting r tend to infinity we obtain that f (α, t1 , t2 , . . . )(P n ) = f1 (α, t1 + t2 + . . . )(P n )
if and only if P n is simple.
In the case of simple polytopes the equations of the first type are trivial, but the equation of the
second type has the form:

f1 (−α, t1 + · · · + tr−1 + α) = f1 (α, t1 + · · · + tr−1 )

If we denote t = t1 + · · · + tr−1 , then f1 (−α, α + t) = f1 (α, t). This equation is equivalent to the
Dehn-Sommerville relations (after the change of variables h(α, t) = f1 (α − t, t) it is equivalent to the
fact that h(α, t) = h(t, α))

6.3 Characterization of the generalized f -polynomial.

In this part we find the condition that uniquely determines the generalized f -polynomial.
At first let us find the relation between f (P n ) and f (dk P n ).
Proposition 6.9. For any polytope P n ∈ P we have

1 ∂k
f (α, t1 , t2 , . . . )(dk P n ) = f (α, t, t1 , t2 , . . . )(P n )
k! ∂tk t=0

Proof. Indeed, both expressions belong to Qsym[α], and they are uniquely defined by their restrictions
to n variables. Then we have
 
1 ∂k
fn+1 (α, t, t1 , t2 , . . . )(P n ) =
k! ∂tk t=0 tn+1 =tn+2 =···=0

1 ∂k 1 ∂k
fn+1 (α, t, t1 , t2 , . . . , tn )(P n ) = ξα Φ(tn ) . . . Φ(t1 )Φ(t)P n = ξα Φ(tn ) . . . Φ(t1 )dk P n =
k! ∂tk t=0 k! ∂tk t=0
fn (α, t1 , t2 , . . . , tn )(dk P n ) = f (α, t1 , t2 , . . . )(dk P n )|tn+1 =tn+2 =···=0

Corollary 6.10.

f (α, t, t1 , t2 , . . . )(P n ) = f (α, t1 , t2 , . . . )(P n ) + f (α, t1 , t2 , . . . )(dP n )t+

+f (α, t1 , t2 , . . . )(d2 P n )t2 + · · · + f (α, t1 , t2 , . . . )(dn P n )tn
fr+1 (α, t, t1 , . . . , tr )(P n ) = fr (α, t1 , . . . , tr )(P n ) + fr (α, t1 , . . . , tr )(dP n )t+
+fr (α, t1 , . . . , tr )(d2 P n )t2 + · · · + fr (α, t1 , . . . , tr )(dn P n )tn , r > 0;

Proof. The first equality follows from Proposition 6.9.

In fact, both equalities can be proved directly:
n
X
fr+1 (α, t, t1 , . . . , tr )(P n ) = ξα Φ(tr ) . . . Φ(t1 )Φ(t)P n = (ξα Φ(tr ) . . . Φ(t1 )dk P n ) tk =
k=0
n
X
= fr (α, t1 , . . . , tr )(dk P n )tk .
k=0

Letting r tend to infinity we obtain the first equality.

The first equality of Corollary 6.10 is equivalent to the condition f (α, t1 , t2 , . . . )(Φ(t)P n ) = f (α, t, t1 , t2 , . . . )(P n ).

33
 Thus we see that the following diagram commutes
f
P −−−−→ Qsym[t1 , t2 , . . . ][α]
 
 
Φy Ty

f
P[t] −−−−→ Qsym[t, t1 , t2 , . . . ][α]

where f (t) = t, and T : Qsym[t1 , t2 . . . ][α] → Qsym[t, t1 , t2 , . . . ][α] is a ring homomorphism:

T g(α, t1 , t2 . . . ) = g(α, t, t1 , t2 , . . . ), g ∈ Qsym[t1 , t2 , . . . ][α].

Remark 6.11. Note that fr (α, t1 , . . . , tr )(Φ(t)P n ) = fr+1 (α, t, t1 , . . . , tr )(P n ) for all r > 0.
Consider the ring homomorphism Tr+1 : Qsym[t1 , . . . , tr ][α] → Qsym[t, t1 , . . . , tr ][α] defined as

Tr+1 (α) = α, Tr+1 Mω (t1 , . . . , tr ) = Mω (t, t1 , . . . , tr )

Then the corresponding diagram fr (α, t1 , . . . , tr )(Φ(t)P n ) = (Tr+1 fr )(α, t, t1 , . . . , tr ) commutes only for
r > n.
Theorem 6.12. Let ψ : P → Qsym[t1 , t2 , . . . ][α] be a linear map such that
1. ψ(α, 0, 0, . . . )(P n ) = αn ;
2. The following diagram commutes:

ψ
P −−−−→ Qsym[t1 , t2 , . . . ][α]
 
 
Φy Ty

ψ
P[t] −−−−→ Qsym[t, t1 , t2 , . . . ][α]
Then ψ = f .

P implies that ψ(α, 0, 0, . . . )(p) = ξα p for any p ∈ P.

Proof. The first condition
Let ψ(P n ) = αn + ψω (α)Mω , and let ω = (j1 , . . . , jk ).
ω
Since ψ(α, t, t1 , t2 , . . . )(P n ) = ψ(α, t1 , t2 , . . . )(Φ(t)P n ), we have

ψ(α, y1 , y2 , . . . , yk , t1 , t2 , . . . )(P n ) = ψ(α, t1 , t2 , . . . )(Φ(yk ) . . . Φ(y1 )P n )

Therefore,

ψ(α, y1 , . . . , yk , 0, 0, . . . ) = ψ(α, 0, 0, . . . )(Φ(yk ) . . . Φ(y1 )P n ) = ξα Φ(yk ) . . . Φ(y1 )P n = f (α, y1 , . . . , yk , 0, 0, . . . )

Hence ψω (α) = fS(ω) αn−|ω| , where S(ω) = {n − |ω|, n − |ω| + jk , . . . , n − j1 }. This is valid for all ω, so
ψ = f.
Remark 6.13. Let us mention that the mapping T is an isomorphism Qsym[t1 , t2 , . . . ][α] → Qsym[t, t1 , t2 , . . . ][α],
while Φ is an injection and its image is described by the condition: p(t) ∈ Φ(P) is and only if
Φ(−t)p(t) ∈ P.
On the ring of simple polytopes we have f (α, t1 , t2 , . . . )(p) = f1 (α, t1 + t2 + . . . )(p) = f1 (α, σ1 )(p),
so the image of Ps belongs to Z[α, σ1 ]. On the other hand, Φ(t) = etd , so we have the condition
f1 (α, t1 )(etd p) = f1 (α, t + t1 )(p).
Proposition 6.14. Let ψ : Ps → Z[α, t] be a linear mapping such that
1. ψ(α, 0)(P n ) = αn ;
2. One of the following equivalent conditions holds:
(a) ψ(α, t1 )(etd p) = ψ(α, t + t1 )(p);

34
 ∂
(b) ψ(α, t)(dp) = ∂t ψ(α, t)(p).

Then ψ = f1 .
Proof. Let conditions 1 and 2a hold. Then

ψ(α, t + t1 )(p) = ψ(α, t1 )(etd p) = ψ(α, 0)(e(t+t1 )d p) = ξα Φ(t + t1 )p = f1 (α, t1 + t)(p).

It remains to prove that conditions 2a and 2b are equivalent.

Indeed, if ψ(α, t1 )(etd p) = ψ(α, t + t1 )(p), then ψ(α, t)(p) = ψ(α, 0)(etd p), therefore

∂ ∂
ψ(α, t)(p) = ψ(α, 0)(etd p) = ψ(α, 0)(etd dp) = ψ(α, t)(dp).
∂t ∂t
∂
Now let ψ(α, t)(dp) = ∂t ψ(α, t)(p). Then

∂ ∂
ψ(α, t1 )(etd p) = ψ(α, t1 )(d(etd p)) = ψ(α, t1 )(etd p)
∂t ∂t1
 
∂ ∂
Hence ∂t − ∂t1 ψ(α, t1 )(etd p) = 0, so ψ(α, t1 )(etd p) depends on α and t + t1 .
Thus ψ(α, t1 )(etd p) = ψ(α, t + t1 )(e0d p) = ψ(α, t + t1 )(p).
Proposition 6.14 in the form 1, 2b was first proved in [Buch].

7 Structure of D∗
7.1 Main theorem
Definition 7.1. Let us denote by D∗ the graded dual Hopf algebra to D.
Let us identify the Hopf algebra M, which is the graded dual Z ∗ to Z, with Qsym[t1 , t2 , . . . ].
The mappings L : Z → D, R : Z op → D, Zi → di are surjections, so the dual mappings are injections.
So the ring homomorphisms L∗ , and R∗ = ̺∗ ◦ L∗ define two embeddings D∗ ⊂ Qsym, which are related
by the involution ̺∗ = ∗.
In fact, L∗ defines an embedding D∗ ⊂ Qsym as a Hopf subalgebra, while for R∗ we have

∆ ◦ R∗ = τQsym ◦ (R∗ ⊗ R∗ ) ◦ ∆,

where τQsym is a homomorphism Qsym ⊗ Qsym → Qsym ⊗ Qsym that interchanges the tensor factors:
τQsym (x ⊗ y) = y ⊗ x.

Definition 7.2. For each r let us define the ring homomorphism R∗r : D∗ → Qsym[t1 , . . . , tr ] by the
formula
R∗r (ψ) = R∗ (ψ)(t1 , . . . , tr , 0, 0, . . . )
Since the restriction Qsym → Qsym[t1 , . . . , tr ] is injective for any graduation (2n), n 6 r, we see
that R∗r is injective for graduations (2n), n 6 r, of D∗ .
Theorem 7.3. Let r > n. Then the image of the (2n)-th, n > 1, graded component of the ring D∗ in
the ring Qsym[t1 , . . . , tr ] under the map R∗r consists of all the homogeneous polynomials of degree 2n
satisfying the relations:

g(t1 , −t1 , t3 , . . . , tr ) = g(0, 0, t3 , . . . , tr );

g(t1 , t2 , −t2 , t4 , . . . , tr ) = g(t1 , 0, 0, t4 , . . . , tr );
...
g(t1 , . . . , tr−2 , tr−1 , −tr−1 ) = g(t1 , . . . , tr−2 , 0, 0).

35
Proof. The ring R∗ (D∗ ) ⊂ Z ∗ consists of all the linear functions ψ ∈ Z ∗ satisfying the property:

hψ, z1 Φ(t)Φ(−t)z2 i = hψ, z1 z2 i (15)

for all z1 , z2 ∈ Z, that is the coefficients of all tk , k > 1, on the left are equal to 0.
We have X X
R∗ (ψ) = hR∗ (ψ), Zω i = hψ, RZω iMω
ω ω

Therefore,
r
X X
R∗r (ψ) = hψ, djk . . . dj1 iM(j1 , ..., jk ) (t1 , . . . , tr ) =
k=0 (j1 , ..., jk )
r
X X X
= hψ, djk . . . dj1 i tjl11 . . . tjlkk =
k=0 (j1 , ..., jk ) 16l1 <···<lk 6r
r
X X X
= hψ, djk tjlkk . . . dj1 tjl11 i = hψ, Φ(tr ) . . . Φ(t1 )i
k=0 (j1 , ..., jk ) 16l1 <···<lk 6r

Then

R∗r (ψ)(t1 , . . . , ti , −ti , . . . , tr ) = hψ, Φ(tr ) . . . Φ(−ti )Φ(ti ) . . . Φ(t1 )i =

hψ, Φ(tr ) . . . Φ(0)Φ(0) . . . Φ(t1 )i = R∗r (ψ)(t1 , . . . , 0, 0, . . . , tr )

On the other hand, let g ∈ Qsym[t1 , . . . , tr ] be a homogeneous polynomial of degree 2n satisfying all
the relations of the theorem. Consider a unique homogeneous function ψ ∈ M such that deg ψ = 2n,
and ψ(t1 , . . . , tr , 0, 0, . . . ) = g. Let us prove that ψ belongs to the image of D∗ under the embedding R∗ .
It is sufficient to prove the relation (15) in the case when z1 , z2 are monomials. Since

Φ(t)Φ(−t) = 1 + (Z1 − Z1 )t + (2Z2 − Z12 )t2 + · · · = 1 + (2Z2 − Z12 )t2 + . . . , and deg ψ = 2n,

the cases deg z1 + deg z2 = 2n and 2(n − 1) are trivial. Let

z1 = Zω , z2 = Zω′ , ω = (j1 , . . . , jl ), ω ′ = (j1′ , . . . , jl′′ ), |ω| + |ω ′ | = n − k, k > 2.

Then the only equality we need to prove is

k
!
X
i
hψ, Zω (−1) Zk−i Zi Zω′ i = 0
i=0

Let us consider the equality

g(t1 , . . . , tl , tl+1 , −tl+1 , tl+3 , . . . , tl+2+l′ , . . . , tr ) = g(t1 , . . . , tl , 0, 0, tl+3 , . . . , tl+2+l′ , . . . , tr ).

j′ j′
The coefficient of the monomial tj11 . . . tjl l tkl+1 tl+3
1 l′
. . . tl+2+l ′ on the left is exactly

k k
!
X X
(−1)i hψ, Zω Zk−i Zi Zω′ i = hψ, Zω (−1)i Zk−i Zi Zω′ i,
i=0 i=0

and on the right it is equal to 0. So hψ, Zω Φ(t)Φ(−t)Zω′ i = 0 for all ω, ω ′ , therefore, ψ ∈ R∗ (D∗ ).
Definition 7.4. Let us define the operations Θk : Z[t1 , t2 , . . . ] → Z[t, t1 , t2 , . . . ] as

Θk g(t1 , t2 , . . . ) = g(t1 , . . . , tk−1 , t, −t, tk , tk+1 , . . . )

Corollary 7.5. The ring R∗ (D∗ ) ⊂ Qsym is defined by the equations

g ∈ R∗ (D∗ ) if and only if Θk g = g for all k > 1.

36
Proof. Let g be a homogenous quasi-symmetric function of degree 2n representing the function ψ ∈ D∗ .
Since g has bounded degree, each monomial of g(t1 , . . . , tk−1 , t, −t, tk , tk+1 , . . . ) appears with the same
coefficient in the polynomial gr+2 (t1 , . . . , tk−1 , t, −t, tk , . . . , tr ) = g(t1 , . . . , tk−1 , t, −t, tk , . . . , tr , 0, 0, . . . ),
when r is large enough. Then Theorem 7.3 implies that

gr+2 (t1 , . . . , tk−1 , t, −t, tk , . . . , tr ) = gr+2 (t1 , . . . , tk−1 , 0, 0, tk , . . . , tr ).

Therefore each monomial containing t has coefficient 0, so g(t1 , . . . , tk−1 , t, −t, tk , . . . ) does not depend
on t. So we can set t = 0 to obtain

g(t1 , . . . , tk−1 , t, −t, tk , . . . ) = g(t1 , . . . , tk−1 , 0, 0, tk , . . . ) = g(t1 , . . . , tk−1 , tk , . . . )

On the other hand, let g ∈ Qsym2n be a quasi-symmetric function such that Θk g = g for all k > 1.
Then for gn (t1 , . . . , tn ) = g(t1 , . . . , tn , 0, 0, . . . ) we have

gn (t1 , . . . , ti−1 , ti , −ti , ti+2 , . . . , tn ) = g(t1 , . . . , ti−1 , ti , −ti , ti+2 , . . . , tn , 0, 0, . . . ) =

= g(t1 , . . . , ti−1 , ti+2 , . . . , tn , 0, 0, . . . ) = gn (t1 , . . . , ti−1 , ti+2 , . . . , tn , 0, 0) = gn (t1 , . . . , ti−1 , 0, 0, ti+2 , . . . , tn )

for 1 6 i 6 n − 1. Then Theorem 7.3 implies that gn = R∗n (ψ) for some function ψ ∈ D∗ of degree
2n. Since the restriction g(t1 , t2 , . . . ) → g(t1 , . . . , tn , 0, 0, . . . ) is injective for graduation 2n, we have
g = R∗ ψ.
Remark 7.6. Since R∗ = ̺∗ ◦ L∗ , and the subring R∗ (D∗ ) ⊂ Qsym is invariant under the involution
̺∗ = ∗, Theorem 7.3 and Corollary 7.5 remain valid for the mapping L∗ .
Proposition 7.7. D∗ ⊗ Q is a free polynomial algebra

D∗ ⊗ Q ≃ Q[LYNodd ],

where LYNodd are Lyndon words consisting of odd positive integers. Rank of the (2n)-th graded compo-
nent of D∗ ⊗ Q is equal to the (n − 1)-th Fibonacci number cn−1 .
Proof. This follows from Proposition 5.10, Corollary 3.21 to the shuffle algebra structure theorem, and
Corollary 5.7.

7.2 Applications
∗
Let D [α] be a graded ring of polynomials in variable α, where deg α = 2, and for the element ψi of
graduation 2i the element ψi αj has graduation 2(i + j).
Let us consider the mapping ϕα : P → D∗ [α], defined by the formula

hϕα (P ), Di = ξα (DP )

for D ∈ D. Then
!
X X
hϕα (P ×Q), Di = ξα (D(P × Q)) = ξα Di′ P × Di′′ Q = (ξα Di′ P ) (ξα Di′′ Q) = hϕα (P ) · ϕα (Q), Di,
i i
P
where ∆(D) = Di′ ⊗ Di′′ . So ϕα is a ring homomorphism.
i

Proposition 7.8. We have R∗ ◦ ϕα = f , where f is the generalized f -polynomial.

Proof. Indeed,
X X
R∗ ◦ ϕα (P ) = hR∗ ◦ ϕα (P ), Zω i Mω = hϕα (P ), RZω i Mω =
ω ω
X X
= hϕα (P ), Dω∗ i Mω = ξα (Dω P ) Mω∗ = f (P )
ω ω

37
Proposition 7.9. The image of the space P 2n under the mapping ϕα : P → D∗ [α] consists of all
homogeneous functions ψ(α) of degree 2n such that
hψ(−α), Φ(α)Di = hψ(α), Di (16)
for all D ∈ D.
Proof. According to Proposition 2.53 we have ξ−α Φ(α) = ξα , so if ψ(α) = ϕα (p), then
hψ(−α), Φ(α)Di = ξ−α Φ(α)Dp = ξα Dp = hψ(α), Di.
On the other hand, let condition (16) hold. Then the polynomial g(α, t1 , . . . , tn ) = R∗n ψ(α) satisfies the
relation

g(−α, t1 , . . . , tn−1 , α) = hψ(−α), Φ(α)Φ(tn−1 ) . . . Φ(t1 )i =

= hψ(α), Φ(0)Φ(tn−1 ) . . . Φ(t1 )i = g(α, t1 , . . . , tn−1 , 0).
Theorem 7.3 implies that the polynomial g satisfies all the relations of Theorem 6.3, so g = fn (pn ) for
some pn ∈ P 2n . We have R∗n ψ(α) = g = fn (pn ) = R∗n ϕα (pn ). Since the restriction R∗ → R∗n is injective
on the (2j)-th graded component of the ring D∗ for j 6 n, and R∗ is an embedding, we obtain that
ψ(α) = ϕα (pn ).
There is a right action of the ring D on its graded dual D∗ :
(ψ, D) → ψD : hψD, D′ i = hψ, DD′ i.
If deg ψ = 2n and deg D = 2k, then deg ψD = 2(n − k). Similarly, there is an action of D[[α]] on
D∗ [[α]]. Then condition (16) means that ψ(−α)Φ(α) = ψ(α). Since for any homogeneous function ψ(α)
of degree 2n the function ψ(α)Φ(α) still has degree 2n, we obtain the corollary.
Corollary 7.10. Let ψ(α) ∈ D∗ [α]. Then
ψ(α) ∈ ϕα (P) ⇔ ψ(−α)Φ(α) = ψ(α). (17)

Let ψ(α) = ψ0 + ψ1 α + · · · + ψn αn , u(α) = ψ0 + ψ2 α2 + ψ4 α4 + . . . , w(α) = ψ1 α + ψ3 α3 + . . . . Then

u(α)Φ(α) − w(α)Φ(α) = u(α) + w(α). Therefore,
u(α) (Φ(α) − 1) = w(α) (Φ(α) + 1) . (18)
For example,
ψ0 d = 2ψ1 ; ψ0 d3 + ψ2 d = ψ1 d2 + 2ψ3 .
Consider the Hopf algebra D ⊗ Then for the graded dual over Z[ 12 ] Hopf algebra we have:
Z[ 21 ].
(D ⊗ Z[ 2 ]) ≃ D ⊗ Z[ 2 ]. The condition that describes the image of the ring P ⊗ Z[ 12 ] in D∗ ⊗ Z[ 12 ][α]
1 ∗ ∗ 1

is the same, namely relation (18).

Proposition 7.11. In the ring D∗ ⊗ Z[ 21 ][α] relation (18) is equivalent to the relation:
Φ(α)−1 ∞
X  k
Φ(α) − 1 2 Φ(α) − 1
w(α) = u(α) = u(α) Φ(α)−1
= u(α) (−1)k−1 . (19)
Φ(α) + 1 1+ 2
2 k=1

Remark 7.12. In fact, equation (19) implies that the function w(α) is odd, if the function u(α) is even.
Indeed,
1
Φ(−α) − 1 Φ(α) − 1
w(−α) = u(−α) = u(α) 1 = u(α)(1 − Φ(α))Φ(α)−1 (1 + Φ(α))−1 Φ(α) =
Φ(−α) + 1 Φ(α) + 1
Φ(α) − 1
= −u(α) = −w(α).
Φ(α) + 1

Therefore for any even function u(α) the function ψ(α) = u(α) + u(α) Φ(α)−1 2Φ(α)
Φ(α)+1 = u(α) Φ(α)+1 belongs to
the image of ϕα

38
Remark 7.13. The same relation is valid in D∗ ⊗ Q[α].
Let us take ϕα (p) with α = 0. Then we obtain the classical map

ϕ0 : P → D∗ : p → ϕ0 (p) hϕ0 (p), Di = ξ0 Dp, ∀p ∈ P, D ∈ D

Proposition 7.14. The mapping ϕ0 ⊗ 1 : P ⊗ Z[ 12 ] → D∗ ⊗ Z[ 21 ] is a surjection.

Φ(α)−1
Proof. Let ψ0 ∈ D∗ ⊗ Z[ 12 ]. Consider the element ψ(α) = u(α) + w(α), u(α) = ψ0 , w(α) = ψ0 Φ(α)+1 .
According to Z[ 12 ]-version of Corollary 7.10 we obtain that ψ(α) = ϕα (p) ∈ ϕα ⊗ 1(P ⊗ Z[ 12 ]). Then
ψ0 = ψ(0) = ϕ0 (p).
Question 7.15. What is the image of the map ϕ0 over the integers?
Remark 7.16. Let us note, that the space P/ Ker ϕα consists of the equivalence classes of integer
combinations of polytopes under the equivalence relation: p ∼ q if and only if p and q have equal flag
f -vectors. Rank of the (2n)-th graded component of the group P/ Ker ϕα is equal to cn ([BB], see also
Section 2.5).
On the other hand, by Corollary 5.7 rank of the (2n)-th graded component of the ring D∗ is equal
to cn−1 , so the mapping ϕ0 : P/ Ker ϕα → D∗ is not injective.
Example 7.17. Let us consider small dimensions.
• n = 1. P/ Ker ϕα is generated by CC = I. The ring D in this graduation is generated by d. Then
ϕ0 (I) = 2d∗ .

• n = 2. P/ Ker ϕα is generated by CCC = ∆2 and BCC = I 2 . D has one generator d2 . Then

ϕ0 (∆2 ) = 3d∗2 , ϕ0 (I 2 ) = 4d∗2 . Therefore, d∗2 = ϕ0 (I 2 − ∆2 ), Ker ϕ0 is generated by 3I 2 − 4∆2
• n = 3. P/ Ker ϕα is generated by BCCC, CBCC = CI 2 , CCCC = ∆3 , while D is generated by
d3 , d2 d. Then

ϕ0 (BC 3 ) = 5d∗3 + 18(d2 d)∗ , ϕ0 (CBC 2 ) = 5d∗3 + 16(d2 d)∗ , ϕ0 (C 4 ) = 4d∗3 + 12(d2 d)∗

So Im ϕ0 has the basis d∗3 , 2(d2 d)∗ and Ker ϕ0 is generated by 2BC 3 − 6CBC 2 + 5C 4 .
Here by Dω∗ we denote the element of the dual basis: hDω∗ , Dσ i = δω, σ

8 Multiplicative Structure of f (P) ⊗ Q

Theorem 8.1. The ring f (P) ⊗ Q is a free polynomial algebra.
Proof. According to Proposition 7.7 we have:

D∗ ⊗ Q ≃ Q[LYNodd ],

where LYNodd are Lyndon words consisting of odd positive integers. This is a free polynomial algebra.
Therefore D∗ ⊗ Q[α] is a free polynomial algebra in the generators LYNodd and α.
Let {fλ } be the functions in D∗ ⊗ Q corresponding to the Lyndon words. Consider the polynomials
fλ (α) defined as

Φ(α) − 1 es(α) − 1
fλ (α) = uλ (α) + wλ (α); uλ (α) = fλ , wλ (α) = fλ = fλ s(α)
Φ(α) + 1 e +1

Each polynomial fλ (α) is a homogeneous element of degree deg fλ . Let us identify D∗ with its image in
Qsym[t1 , t2 , . . . ] under the embedding R∗ .
Then according to Remark 7.13 we have: fλ (α) ∈ f (P ⊗ Q) = f (P) ⊗ Q.
Lemma 8.2. f (P) ⊗ Q = Q[fλ (α), α2 ].

39
Proof. At first let us proof that the polynomials {fλ (α)}, α2 are algebraically independent.
Let g ∈ Q[x, y1 , y2 , . . . , ys ] be a polynomial such that g(α2 , f1 (α), . . . , fs (α)) = 0. We can write this
equality as

g0 (f1 (α), . . . , fs (α)) + g2 (f1 (α), . . . , fs (α))α2 + · · · + g2t (f1 (α), . . . , fs (α))α2t = 0

for some integers s, t. Let us set α = 0. Then we obtain

g0 (f1 , . . . , fs ) = 0

Since {fλ } are algebraically independent, g0 ≡ 0. Since D∗ [α] is a polynomial ring, we can divide the
equality by α2 to obtain

g2 (f1 (α), . . . , fs (α)) + g4 (f1 (α), . . . , fs (α))α2 + · · · + g2t (f1 (α), . . . , fs (α))α2t−2 = 0

Iterating this step in the end we obtain that all the polynomials g2i ≡ 0, i = 0, 1, . . . , 2t, so g ≡ 0.
Now let us prove that the polynomials {fλ (α)}, α2 generate f (P) ⊗ Q.
Let ψ(α) = ψ0 + ψ1 α+ · · ·+ ψn αn ∈ f (P)⊗ Q be a homogeneous element of degree 2n. The functions
{fλ } generate D∗ ⊗ Q, so we can find a homogeneous polynomial g0 ∈ Q[y1 , . . . , ys ] such that

g0 (f1 , . . . , fs ) = ψ0 , deg fi 6 2n

Let us consider the element ψ(α) − g0 (f1 (α), . . . , fs (α)) ∈ f (P) ⊗ Q. Since θ1 = θ0 d2 for any

θ(α) = θ0 + θ1 α + θ2 α2 + · · · + θr αr ∈ f (P) ⊗ Q,

we obtain that
ψ(α) − g0 (f1 (α), . . . , fs (α)) = ψ̂(α)α2 .
is a homogeneous element of degree 2n. ψ̂(α)α2 ∈ f (P) ⊗ Q, so ψ̂(−α)α2 Φ(α) = ψ̂(α)α2 . Since
D∗ ⊗ Q[α] is a polynomial ring, ψ̂(−α)Φ(α) = ψ̂(α), so ψ̂(α) ∈ f (P) ⊗ Q is a homogeneous element of
degree 2(n − 2).
Iterating this argument in the end we obtain the expression of ψ(α) as a polynomial in {fλ (α)} and
α2 .
This lemma proves the theorem.
Let us note that dimension of the (2n)-th graded component of the ring f (P) ⊗ Q is equal to the
n-th Fibonacci number cn .
Corollary 8.3. There is an isomorphism of free polynomial algebras

f (P) ⊗ Q ≃ Q[LYNodd , α2 ] = Nodd ⊗ Q[α2 ] ≃ Q[LYN12 ] = N12 ⊗ Q.

Proof. Indeed, these three algebras are free polynomial algebras with the same dimensions of the cor-
responding graded components: dimension of the (2n)-th graded component of each algebra is equal to
cn
Let us denote by kn the number of multiplicative generators of degree 2n. According to Corollary
8.3 for n > 3 the number of generators kn is equal to the number of Lyndon words of degree 2n in
LYNodd or in LYN12 .
For example, for small n we have
n LYN2n 12 LYN2nodd
3 [1,2] [3]
4 [1,1,2] [1,3]
5 [1,2,2], [1,1,1,2] [5], [1,1,3]
6 [1,1,2,2], [1,1,1,1,2] [1,5], [1,1,1,3]
7 [1,2,2,2], [1,1,2,1,2], [1,1,1,2,2], [1,1,1,1,1,2] [7], [1,1,5], [1,3,3], [1,1,1,1,3],

40
Corollary 8.4. 1. kn+1 > kn
2. kn > Nn −2, where Nn is the number of different decompositions of n into the sum of odd numbers.
Remark 8.5. It is a well-known fact, that the number of the decompositions of n into the sum of
odd numbers is equal to the number of the decompositions of n into the sum of different numbers. For
example,
2 = 1+1 = 2
3 = 1 + 1 + 1, 3 = 1 + 2, 3
4 = 1 + 1 + 1 + 1, 1 + 3 = 1 + 3, 4
5 = 1 + 1 + 1 + 1 + 1, 1 + 1 + 3, 5 = 1 + 4, 2 + 3, 5
Proof. 1. k1 = k2 = k3 = 1.
2(n+1)
Let n > 3 and let w = [a1 , . . . , ak ] ∈ LYN2n 12 . Then 1w = [1, a1 , . . . , ak ] ∈ LYN12 . Indeed,
6 k and
for any proper tail [ai , . . . , ak ], i > 1 if ai > 1, then [ai , . . . , ak ] > 1w. If ai = 1 then i =
[ai+1 , . . . , ak ] > [a1 , . . . , ak ], since w is Lyndon. Thus kn+1 > kn
2. Any decomposition n = d1 + · · · + dk into the sum of odd numbers d1 6 d2 6 · · · 6 dk gives a
Lyndon word [d1 , . . . , dk ] ∈ LYNodd , except for the case, when k > 1, d1 = · · · = dk = d 6= n.
If d > 5, then the word [1, d − 2, 1, d, . . . , d] is Lyndon. When d = 1 or d = 3 there can be no
Lyndon word corresponding to the decomposition n = 1 + · · · + 1 or n = 3 + · · · + 3, as it happens
for n = 6:
6=1+5=3+3=1+1+1+3=1+1+1+1+1+1
We have k6 = 2 and N6 = 4. Thus kn > Nn − 2.

Then we obtain the corollary:

√
Corollary 8.6. For any t, |t| < 5−12 , there is the following decomposition of the generating series of
Fibonacci numbers into the absolutely convergent infinite product:
X∞ Y∞
1 2 3 4 n 1
2
= 1 + t + t + 2t + 3t + · · · = c n t = ,
1−t−t n=0 i=1
(1 − ti )ki

The numbers kn satisfy the properties kn+1 > kn > Nn − 2, where Nn is the number of decompositions
of n into the sum of odd summands.
√ ∞
P
5−1 1
Proof. Let |t| < 2 . Then the series cn tn converges absolutely to 1−t−t2 . Then
n=0

∞
X
RN = cn |tn | −−−−→ 0.
N →∞
n=N

But we have:
∞
X n−1
Y 1
Sn = c k tk − = an tn + an+1 tn+1 + an+2 tn+2 + . . . ,
i=1
(1 − ti )ki
k=0

where 0 6 ai 6 ci . Then |Sn | 6 Rn , therefore Sn −−−−→ 0, and the formula is valid.

n→∞
Q
n
1
P
∞
The infinite product lim (1−ti )ki
converges absolutely if and only if the series |−kn ln(1−tn )|
n→∞ i=1 n=1
√
5−1 n n
converges. For 0 6 t < 2is true. In this case 0 6 kn t 6 −kn ln(1 − t ), therefore the power
this
∞
P √ √
series kn t converges and its radius of convergence it at least 5−1
n
2 . Thus for |t| <
5−1
2 we have:
n=1
∞
P
kn |tn | < ∞, and the equivalence | − kn ln(1 − tn )| ∼ kn |tn | implies the absolute convergence of the
n=1
infinite product.

41
 Here are the examples in small dimensions:

n generators kn additive basis cn

1 [1] 1 [1] 1
2 [2] 1 [2], [1]*[1] 2
3 [1,2] 1 [1,2], [2]*[1], [1]*[1]*[1] 3
4 [1,1,2] 1 [1,1,2], [1,2]*[1], [2]*[2], [2]*[1]*[1], [1]*[1]*[1]*[1] 5
5 [1,2,2], [1,1,1,2] 2 [1,2,2], [1,1,1,2], [1,1,2]*[1], [1,2]*[2], [1,2]*[1]*[1] 8
[2]*[2]*[1], [2]*[1]*[1]*[1], [1]*[1]*[1]*[1]*[1]
6 [1,1,2,2], [1,1,1,1,2] 2 13
7 [1,2,2,2], [1,1,2,1,2], [1,1,1,2,2], [1,1,1,1,1,2] 4 21
The corresponding products are:

(1 − t)(1 − t2 ) = 1 − t − t2 + t3 ;
(1 − t)(1 − t2 )(1 − t3 ) = 1 − t − t2 + t4 + t5 − t6 ;
(1 − t)(1 − t2 )(1 − t3 )(1 − t4 ) = 1 − t − t2 + 2t5 − t8 − t9 + t10 ;
(1 − t)(1 − t2 )(1 − t3 )(1 − t4 )(1 − t5 )2 = 1 − t − t2 + 2t6 + 2t7 − t8 − t9 − 2t10 −
−t11 − t12 + 2t13 + 2t14 − t18 − t19 + t20 .

The numbers ki can be found in the following way:

∞
X
− log(1 − t − t2 ) = − ki log(1 − ti );
i=1
X∞ n   ∞ ∞
1 X n n+j X X tri
t = ki
n=1
n j=0 j i=1 r=1
r

Thus for any N

N
[2] N −j


X j 1 X
= iki .
j=0
N −j N
i|N

According to the Möbius inversion formula we obtain

 d 
[ 2 ] d−j
 
X X
N kN = d j ·µ N ,
j=0
d−j d
d|N

where µ(n) is the Möbius function, that is



1, n = 1;
r
µ(n) = (−1) , n = p1 . . . pr , {pi } – distinct prime numbers;


0, n is not square-free.

For example, if N is a prime number p, then

p
[2] p−j

p
[2] p−j


X j
X j
pkp = −1 + p ; kp = .
j=0
p−j j=1
p−j

(p−j−1)(p−j)
Each summand is an integer, since p−j
j = j−1 j is an integer, and (p − j) and j are relatively
prime numbers.
Thus k5 = 2, k7 = 4, k11 = 18, and so on.

42
9 Bayer-Billera Ring
Let us consider the free graded abelian group BB ⊂ P, generated by 1 and all the polytopes Q ∈ Ωn , n > 1.
Rank of the (2n)-th graded component of this group is equal to cn .
Since the determinant of the matrix K n is equal to 1, the generalized f -polynomials {f (Q), Q ∈ Ωn }
form a basis of the (2n)-th graded component of the ring f (P).
So the composition of the inclusion i : BB ⊂ P and the mapping f : P → Qsym[t1 , t2 , . . . ][α] is an
isomorphism of the abelian groups BB and f (P). This gives a projection π : P → BB

π(p) = x ∈ BB : f (p) = f (x).

It follows from the definition, that π ◦ i = 1 on the space BB.

Theorem 9.1. The projection π : P → BB defined by the relation f (πp) = f (p) gives the graded group
BB the structure of a graded commutative associative ring with the multiplication x ×BB y = π(x × y)
such that f (x×BB y) = f (x)f (y), and BB ⊗Q is a free polynomial algebra in a countable set of variables.
Question 9.2. To describe the projection π : P → BB in an efficient way.
Question 9.3. To find the multiplicative generators of BB ⊗ Q in an efficient way.
Question 9.4. To describe the multiplication in BB in an efficient way.

10 Hopf comodule structures

Let R be a field or the ring Z.
Definition 10.1. By a (left) Hopf comodule (or Milnor comodule) over a Hopf algebra X we mean an
R-algebra M with a unit provided M is a comodule over X with a coaction b : M → X ⊗ M such that
b(uv) = b(u)b(v), i.e. such that b is a homomorphism of rings.

10.1 Hopf comodule structure arising from the Hopf module structure
Lemma 10.2. Let H be a graded connected torsionfree Hopf algebra over Z with finite rank of each
graded component. If a ring A has the right graded Milnor module structure over H, then A has a left
graded Hopf comodule structure over the graded dual Hopf algebra H∗ defined by the formula
X
∆H∗ (A) = Hω∗ ⊗ AHω
Hω ∈H

where {Hω } is some basis in the graded group H, and {Hω∗ } is the graded dual basis: hHω∗ , Hσ i = δω, σ .
Proof.
! !
X X X X
(1 ⊗ ∆H∗ ) ◦ ∆H∗ A = Hω∗ ⊗ Hσ∗ ⊗ (AHω )Hσ = Hω∗ ⊗ Hσ∗ ⊗A hHτ∗ , Hω Hσ iHτ =
ω σ ω, σ τ
! !
X X X X
= Hω∗ ⊗ Hσ∗ hHτ∗ , Hω Hσ i ⊗ AHτ = Hω∗ ⊗ Hσ∗ h∆Hτ∗ , Hω ⊗ Hσ i ⊗ AHτ =
τ ω, σ τ ω, σ
X
= ∆Hτ∗ ⊗ AHτ = (∆ ⊗ 1) ◦ ∆H∗ A
τ

Since the augmentation in the ring H∗ has the form ε(ξ) = ξ(1), we see that
X X
(ε ⊗ 1)∆H∗ A = (ε ⊗ 1) Hω∗ ⊗ AHω = Hω∗ (1) ⊗ AHω = Hω∗0 (1) ⊗ AHω0 = 1 ⊗ A,
ω ω

where Hω0 is a basis in H0 .

43
 Therefore ∆H∗ is a comodule structure.
Now
!
X X X
∆H∗ (AB) = Hω∗ ⊗ (AB)Hω = Hω∗ ⊗ ′
(AHω, ′′
i )(BHω, i ) =
ω ω i
! !!
X X X X
= Hω∗ ⊗ A hHσ∗ , Hω,
′
i iHσ B hHτ∗ , Hω,
′′
i iHτ =
ω i σ τ
! !
X X X
= hHσ∗ , Hω,
′ ∗ ′′
i ihHτ , Hω, i i Hω∗ ⊗ (AHσ )(BHτ ) =
σ, τ ω i
!
X X X
hHσ∗ · Hτ∗ , Hω i Hω∗ ⊗ (AHσ )(BHτ ) = Hσ∗ · Hτ∗ ⊗ (AHσ )(BHτ ) = (∆H∗ A) · (∆H∗ B) ,
σ, τ ω σ, τ
P ′ ′′
where ∆Hω = Hω, i ⊗ Hω, i . This equality finishes the proof.
i

Remark 10.3. This Hopf comodule structure does not depend on the choice of a graded basis in the
graded group H, since we can write it as
!
X
∗
∆H∗ A = (1 ⊗ A) Hω ⊗ Hω ,
ω
P
and Hω∗ ⊗Hω defines in each graded component Hn the identity operator in (Hn )∗ ⊗Hn ≃ HomZ (Hn , Hn ).
ω

Remark 10.4. The similar argument proves that if a ring A has the left graded Milnor module structure
over H, then A has a right graded Hopf comodule structure over the graded dual Hopf algebra H∗ defined
by the formula X
∆H∗ (A) = Hω A ⊗ Hω∗ .
Hω ∈H

Corollary 10.5. There ia a natural left Hopf comodule structure on the ring R over the Hopf algebra
Qsym induced by the right Hopf module structure over the Hopf algebra Z:
X X X X X
∆R (P ) = Mω ⊗P Zω = M(j1 , ..., jk ) ⊗(P Zj1 . . . Zjk ) = M(jk , ..., j1 ) ⊗(dj1 . . . djk P )
ω k>0 (j1 , ..., jk ) k>0 (j1 , ..., jk )

There are natural right Hopf comodule structures ∆L and ∆D∗ on the ring R over the Hopf algebras
Qsym and D∗ induced by the left Hopf module structures over the Hopf algebras Z and D.
These structures are compatible in the following sense:

∆L = τ ◦ (∗ ⊗ 1) ◦ ∆R (20)
(1 ⊗ L∗ ) ◦ ∆D∗ = ∆L (21)

where τ : Qsym ⊗ R → R ⊗Qsym is a homomorphism that interchanges the tensor factors

τ (Mω ⊗ P ) = P ⊗ Mω .

We have X X
∆L P = (dj1 . . . djk P ) ⊗ M(j1 , ..., jk )
k>0 (j1 , ..., jk )

Proof. The existence of these structures and the explicit formulas follow immediately from Lemma 10.2
and Corollary 10.5. The first relation follows easily from the formulas for ∆L and ∆R .
Since D ≃ Z/JU is a free abelian group, we have Z = Z ′ ⊕ JU . Let us choose graded bases in both
subgroups: {Hβ } in Z ′ , and {Hγ } in JU . Then {Hβ , Hγ } form a graded basis in Z.
Let us note that L∗ (D∗ ) = {ψ ∈ Qsym : ψ|JU = 0}.

44
 Then the elements of the dual basis {Hβ∗ }, {Hγ∗ } form a basis in Qsym, while the functions {Hβ∗ }
form a basis in L∗ (D∗ ). Let Hβ∗ = L∗ Dβ∗ . Then the operators {Dβ = LHβ } form a graded basis in D,
dual to the basis {Dβ∗ }.
Now for any P ∈ R we have
X X
∆L P = (Hβ P ) ⊗ Hβ∗ + (Hγ P ) ⊗ Hγ∗
β γ

But Hγ P = 0 for all γ, therefore,

X X X
∆L P = Hβ P ⊗ Hβ∗ = (LHβ ) P ⊗ Hβ∗ = Dβ P ⊗ L∗ Dβ∗ = (1 ⊗ L∗ ) ◦ ∆D∗ P
β β β

We have the series of mappings

Φ(t1 ) : R → R[t1 ], Φ2 (t1 , t2 ) = Φ(t2 )Φ(t1 ) : R → Qsym[t1 , t2 ] ⊗ R,
...
Φn (t1 , . . . , tn ) = Φ(tn )Φ(tn−1 ) . . . Φ(t1 ) : R → Qsym[t1 , . . . , tn ] ⊗ R
Taking the inverse limit we obtain the mapping
Φ∞ : R → Qsym ⊗ R,
such that for any element P ∈ R of graduation 2n
n
X X
Φ∞ P = M(jk , ..., j1 ) ⊗ (dj1 . . . djk P )
k=0 (j1 , ..., jk )

Thus we have proved the following fact.

Proposition 10.6.
Φ∞ = ∆R

10.2 Ring homomorphisms

Definition 10.7. Any ring homomorphism h : R → A induces the ring homomorphism
Φh : R → Qsym ⊗ A, Φh = (1 ⊗ h) ◦ ∆R
Let us denote by Φh, r the homomorphism R → Qsym[t1 , . . . , tr ] ⊗ A.
Φh, r (P ) = Φh (P )(t1 , . . . , tr , 0, 0, . . . )
Example 10.8. Let us consider several examples.
1. ξα : P → Z[α] induces the homomorphism Φξα : P → Qsym[α]. This homomorphism coincides
with the generalized f -polynomial.
2. εα : RP → Z[α] induces the ring homomorphism Φεα : RP → Qsym[α]. Let us denote this ring
homomorphism by fRP .
3. ε0 : RP → Z induces the homomorphism Φε0 : RP → Qsym. Then Φε0 = F∗ , where F is the
Ehrenborg F-quasi-symmetric function introduced in [Ehr]:
X X
F(P ) = M(ρ(x0 , x1 ), ρ(x1 , x2 ), ..., ρ(xk , xk+1 )) = fa1 , ..., ak (P )M(a1 +1, a2 −a1 , ..., n−ak )
0̂=x0 <x1 <···<xk+1 =1̂ 06a1 <···<ak 6n−1

where P is an n-dimensional polytope, the sum ranges over all chains from 0̂ = ∅ to 1̂ = P in the
face lattice L(P ), and fa1 , ..., ak are flag numbers.
In fact, F is defined on the Rota-Hopf algebra R of graded posets, and it was proved in [Ehr] that
it is a Hopf algebra homomorphism R → Qsym such that F(P∗ ) = F(P)∗ .

45
Proposition 10.9. For any polytope P ∈ P we have

fRP (P ) = F∗ (P ) + αf (P )

Proof. Indeed, let P be an n-dimensional polytope. Then

n
X X

fRP (P ) = fa1 , ..., ak M(n−ak , ..., a1 +1) + αa1 +1 M(n−ak , ..., a2 −a1 )
k=0 06a1 <···<ak 6n−1

Corollary 10.10. For any two polytopes P, Q ∈ P we have

f (P > Q) = f (P ) · F∗ (Q) + F∗ (P ) · f (Q) + αf (P ) · f (Q)

Proof. Since

fRP (P > Q) = fRP (P ) · fRP (Q), and F∗ (P > Q) = F∗ (P ) · F∗ (Q),

we obtain
F∗ (P ) · F∗ (Q) + αf (P > Q) = (F∗ (P ) + αf (P )) · (F∗ (Q) + αf (Q))
Removing the brackets, canceling the equal summands, and dividing by α, we obtain the required
formula.
Corollary 10.11.
f (CP ) = F∗ (P ) + (α + σ1 )f (P )
Theorem 10.12. • For r > n the image fr (P 2n ) ⊂ Qsym[t1 , . . . , tr ][α] consists of homogeneous
polynomials g of degree 2n satisfying the equations
1.
g(α, t1 , −t1 , t3 , . . . , tr ) = g(α, 0, 0, t3 , . . . , tr )
g(α, t1 , t2 , −t2 , . . . , tr ) = g(α, t1 , 0, 0, . . . , tr )
(22)
...
g(α, t1 , . . . , tr−1 , −tr−1 ) = g(α, t1 , . . . , 0, 0)
2.
g(−α, t1 , t2 , . . . , tr−1 , α) = g(α, t1 , t2 , . . . , tr−1 , 0) (23)
• For r > n the image fRP, r (RP 2n ) ⊂ Qsym[t1 , . . . , tr ][α] consists of homogeneous polynomials g
of degree 2n satisfying the following conditions:
1. The equations of type (22)
2. The equation
g(−α, t1 , t2 , . . . , tr−1 , α) = g(0, t1 , t2 , . . . , tr−1 , 0)
• For r > n the image Fr (RP 2n ) ⊂ Qsym[t1 , . . . , tr ] consists of homogeneous polynomials g of degree
2n satisfying the equations of type (22).
In all the cases the equations are equivalent to the Bayer-Billera (generalized Dehn-Sommerville)
relations [BB]
Proof. The first statement of the theorem is exactly Theorem 6.3.
For the second and the third statements we need a lemma corresponding to Lemma 6.4
Lemma 10.13. Let P be an n-dimensional polytope. Then

46
 1. The equation
fRP, r (α, t1 , . . . , tq , −tq , . . . , tr ) = fRP, r (α, t1 , . . . , 0, 0, . . . , tr )
is equivalent to the generalized Dehn-Sommerville relations
at+1 −1
X

(−1)j−at −1 fa1 , ..., at , j, at+1 , ..., ak = 1 + (−1)at+1 −at fa1 , ..., at , at+1 , ..., ak
j=at +1

for −1 6 a1 < a2 < · · · < ak < n, and 1 6 k 6 min{r − 1, n}, k + 1 − q 6 t 6 r − q.

2. The equation
fRP, r (−α, t1 , . . . , tr−1 , α) = fRP, r (0, t1 , . . . , tr−1 , 0)
is equivalent to the generalized Dehn-Sommerville relations
1 −1
aX
(−1)j fj, a1 , ..., ak = (1 + (−1)a1 −1 )fa1 , ..., ak
j=0

for 0 6 k 6 min{r − 1, n − 1}.

3. The equation
F∗r (t1 , . . . , tq , −tq , . . . , tr ) = F∗r (t1 , . . . , 0, 0, . . . , tr )
is equivalent to the generalized Dehn-Sommerville relations
at+1 −1
X

(−1)j−at −1 fa1 , ..., at , j, at+1 , ..., ak = 1 + (−1)at+1 −at fa1 , ..., at , at+1 , ..., ak
j=at +1

for 0 6 a1 < · · · < ak , and 0 6 k 6 min{r − 2, n − 1}, k + 2 − q 6 t 6 r − q.

Proof. Items 1. and 3. are proved in the same manner as Lemma 6.4. Let us consider the relation of
item 2.: fRP, r (−α, t1 , . . . , tr−1 , α) = fRP, r (0, t1 , . . . , tr−1 , 0):
 
min{r−1, n+1}
X X X
= (−α)n+1 + fa1 , ..., ak (−α)a1 +1  tln−a
1
k
. . . talk2 −a1  +
k=1 −16a1 <···<ak 6n−1 16l1 <···<lk 6r−1
 
min{r, n+1}
X X X
3 −a2 a2 −a1 
+ fa1 , ..., ak (−α)a1 +1  tln−a
1
k
. . . talk−1 α =
k=1 −16a1 <···<ak 6n−1 16l1 <···<lk =r
 
min{r−1, n+1}
X X X
fa1 , ..., ak  tln−a
1
k
. . . talk2 −a1 
k=1 −1=a1 <···<ak 6n−1 16l1 <···<lk 6r−1

This is equivalent to the relations

1 −1
aX
fa1 , ..., ak + (−1)j+1 fj, a1 , ..., ak + (−1)a1 +1 fa1 , ..., ak = 0
j=0

for 0 6 a1 < · · · < ak , and 0 6 k 6 min{r − 1, n − 1}. The case k = 0 again corresponds to the Euler
formula
1 − f0 + f1 + · · · + (−1)n fn−1 + (−1)n+1 = 0.

Remark 10.14. Let us mention that in the case of fRP, r and F∗r the equations of type (22) contain
all the generalized Dehn-Sommerville relations, and these equations for fRP, r imply the equation (23).
This follows from the fact that a1 can be equal to −1, which is not possible in the case of the generalized
f -polynomial.

47
 Now the end of the proof of the theorem is the same as for Theorem 6.3.
Proposition 10.15. The correspondence fRP → fRP |α=0 = F∗ is an isomorphism of the images
fRP (RP) and F∗ (RP).
Proof. Indeed, it is clear that this mapping is an epimorphism. On the other hand, let F∗ (P ) = 0. Then
all the flag numbers of the element P ∈ RP are equal to 0. Therefore, fRP (P ) = 0, so the mapping is
a monomorphism.
Corollary 10.16. 1. The ring homomorphism F∗ : RP → R∗ (D∗ ) ⊂ Qsym is an epimorphism,
2. F∗ (RP) ⊗ Q is a free polynomial algebra with dimension of the (2n)-th graded component equal to
the (n − 1)-th Fibonacci number cn−1 (c0 = c1 = 1, cn+1 = cn + cn−1 , n > 1).
Proof. 1. Let g ∈ R∗ (D∗ ) be a quasi-symmetric function of degree 2n. Then according to The-
orem 7.3 gn (t1 , . . . , tn ) = g(t1 , . . . , tn , 0, 0, . . . ) satisfies the relations of type (22). Therefore
gn (t1 , . . . , tn ) = F∗n (P ) for some P ∈ RP 2n . Since the mapping Qsym → Qsym[t1 , . . . , tn ], tj → 0, j > n
is injective on the graduation (2n), we see that g = F∗ (P ).
2. We see that F∗ (RP) = R∗ (D∗ ). Therefore,

F∗ (RP) ⊗ Q ≃ D∗ ⊗ Q ≃ Q[LYNodd ],

and dim F∗ (RP 2n ) ⊗ Q = cn−1 according to Proposition 7.7.

Remark 10.17. Let us mention that deg L(P n ) = 2(n+1), therefore, rank F∗ (RP 2(n+1) ) = cn = rank f (P 2n )

10.3 Natural Hopf comodule structure

P is not a Hopf algebra, but it turns out that there is a natural Hopf comodule structure on the ring P
over the Hopf algebra RP.
Proposition 10.18. Formula X
∆RP P = F ⊗ (P/F )
F ⊆P

defines a natural graded right Hopf comodule structure ∆RP : P → P ⊗ RP on the ring P over the Hopf
algebra RP.
Proof. We should proof that the following two diagrams commute

∆RP
P / P ⊗ RP P LL
LLL
∆RP id ⊗∆ ∆RP
Lid
LL⊗1
LLL
 ∆RP ⊗id   id ⊗ε %
P ⊗ RP / P ⊗ RP ⊗ RP P ⊗ RP / P ⊗Z

and that ∆RP is a ring homomorphism. Indeed, we have

   
X X X
(∆RP ⊗ id) ◦ ∆RP P = (∆RP ⊗ id)  F ⊗ P/F  =  G ⊗ F/G ⊗ P/F =
F ⊆P F ⊆P G⊆F
 
X X X
= G⊗ F/G ⊗ P/F  = G ⊗ ∆(P/G) = (id ⊗∆) ◦ ∆RP P,
G⊆P G⊆F ⊆P G⊆P

and X
(id ⊗ε) ◦ ∆RP (P ) = F ⊗ ε(P/F ) = P ⊗ 1,
F ⊆P

48
 Let us check that ∆RP is a homomorphism of rings P → P ⊗ RP:
X X
∆RP (P × Q) = (F × G) ⊗ (P × Q/F × G) = (F × G) ⊗ (P/F > Q/G) =
F ×G⊆P ×Q F ⊆P, G⊆Q
   
X X
= F ⊗ P/F  ·  G ⊗ Q/G = ∆RP (P ) · ∆RP (Q)
F ⊆P G⊆Q

∆RP (pt) = pt ⊗(pt / pt) = pt ⊗∅ = 1 ⊗ 1.

Corollary 10.19. • Any ring homomorphism P → A, where A is a ring, induces the ring homo-
morphism P → A ⊗ RP
• Any linear homomorphism ψ : RP → Z induces the linear operator P → P, which is a ring
homomorphism, if ψ is a ring homomorphism.
Example 10.20. The homomorphism ξα : P → Z[α] defines the ring homomorphism lα : P → RP[α]:
X
lα (P n ) = αdim F P/F.
F ⊆P

Let us recall that in the case of a simple n-polytope a face polytope of an i-dimensional face is a simplex
∆n−i−1 = xn−i . Therefore,
n
X
lα (P n ) = fi αi xn−i = f1 (α, x).
i=0

This is a homogeneous f -polynomial in two variables defined in [Buch].

Proposition 10.21.

∆RP (P > Q) = (1 ⊗ P ) · ∆RP (Q) + ∆RP (P ) · (1 ⊗ Q) + ∆RP (P )(> ⊗ >)∆RP (Q) (24)

Proof.
X X
∆RP (P > Q) = (∅ > G) ⊗ (P > Q/∅ > G) + (F > ∅) ⊗ (P > Q/F > ∅) +
G⊆Q F ⊆P
X X X
+ (F > G) ⊗ (P > Q/F > G) = G ⊗ (P > (Q/G)) + F ⊗ ((P/F ) > Q) +
F ⊆P, G⊆Q G⊆Q F ⊆P
X
+ (F > G) ⊗ (P/F > Q/G) = (1 ⊗ P )∆RP (Q) + ∆RP (P )(1 ⊗ Q) + ∆RP (P )(> ⊗ >)∆RP (Q)
F ⊆P, G⊆Q

Now if we apply ξα ⊗ 1 to the formula (24), then we obtain the corollary:

Corollary 10.22.
lα (P > Q) = P > lα (Q) + lα (P ) > Q + α lα (P ) · lα (Q)

10.4 Interrelation
Let us remind that the left action of the Leibnitz-Hopf algebra Z in the ring R induces the right Hopf
comodule structure
Xn X
n
∆L P = (dj1 . . . djk P ) ⊗ M(j1 , ..., jk ) ,
k=0 (j1 , ..., jk )

49
Proposition 10.23. The following diagram commutes:

∆RP
P JJ / P ⊗ RP
JJ
JJ
J 1⊗F
∆L JJJ
$ 
P ⊗ Qsym

Proof. Let P be an n-dimensional polytope. Then

n
X X
∆L (P ) = (dj1 . . . djk P ) ⊗ M(j1 , ..., jk ) =
k=0 (j1 , ..., jk )
 
n
X X X X
F n−(j1 +···+jk ) 1 ⊗ M(j1 , ..., jk ) =
k=0 (j1 , ..., jk ) F n−(j1 +···+jk ) F n−(j1 +···+jk ) ⊂F n−(j2 +···+jk ) ⊂···⊂F n−jk ⊂P
 
n X
X n−r
X X X
= Fr ⊗  M(j1 , ..., jk )  =
r=0 F r ⊆P k=0 (j1 , ..., jk ):j1 +···+jk =n−r F r ⊂F r+j1 ⊂···⊂F n−jk ⊂P
X
= F ⊗ F(P/F ) = (1 ⊗ F) ◦ ∆RP (P )
F ⊆P

Corollary 10.24. For any polytope P ∈ P we have

F∗ (lα P ) = f (P )

Proof. We have (ξα ⊗ 1) ◦ ∆L P = f (P )∗ . So if we apply the homomorphism ξα ⊗ 1 to the diagram

above, we obtain F ◦ lα = (ξα ⊗ 1) ◦ ∆L = f ∗ . Therefore, F∗ ◦ lα = f .

Corollary 10.25. The following diagram commutes:

f∗
P −−−−→ Qsym[α]
 

∆RP y

∆y

f ∗ ⊗F
P ⊗ RP −−−−→ Qsym[α] ⊗ Qsym

Proof. 1. Consider the commutative diagram

∆RP ∆L ⊗1
P II / P ⊗ RP / P ⊗ Qsym ⊗ RP
II
II
I 1⊗F 1⊗1⊗F
∆L III
$  ∆L ⊗1 
P ⊗ Qsym / P ⊗ Qsym ⊗ Qsym
SSS
SSS
ξα ⊗1⊗1
SSS
SSS
SS) 
Z[α] ⊗ Qsym ⊗ Qsym

Then
(ξα ⊗ 1 ⊗ 1) ◦ (∆L ⊗ 1) ◦ ∆L = (ξα ⊗ 1 ⊗ 1) ◦ (1 ⊗ ∆) ◦ ∆L = ∆ ◦ f ∗
while
(ξα ⊗ 1 ⊗ 1) ◦ (1 ⊗ 1 ⊗ F) ◦ (∆L ⊗ 1) ◦ ∆RP = (f ∗ ⊗ F) ◦ ∆RP
Since these two compositions correspond to two pathes in the diagram, they are equal.

50
 2. Let us consider also a direct proof. We know that

∆(M(b1 , ..., bk ) ) = 1 ⊗ M(b1 , ..., bk ) + M(b1 ) ⊗ M(b2 , ..., bk ) + · · · + M(b1 , ..., bk−1 ) ⊗ M(bk ) + M(b1 , ..., bk ) ⊗ 1

and
X X
f (α, t1 , t2 , . . . )(P n )∗ = αn + αa1 M(a2 −a1 , ..., n−ak ) = αa1 M(a2 −a1 , ..., n−ak )
F a1 ⊂···⊂F ak F a1 ⊂···⊂F ak ⊂P n

Then
X
∆(f (P n )∗ ) = αa1 M(a2 −a1 , ..., as −as−1 ) ⊗ M(as+1 −as , ..., n−ak ) =
F a1 ⊂···⊂Gas ⊂···⊂F ak ⊂P n
! !
X X X
a1
= α M(a2 −a1 , ..., as −as−1 ) ⊗ M(as+1 −as , ..., n−ak ) =
Gas ⊆P n F a1 ⊂···⊂F as−1 ⊂Gas Gas ⊂F as+1 ⊂···⊂F ak ⊂P n
X
= f ∗ (G) ⊗ F(P n /G) = (f ∗ ⊗ F) ◦ ∆RP (P n ).
G⊆P n

11 Operators B and C
We know that the cone and the bipyramid operations can be considered as linear operators on the rings
P and RP. It turns out that there are corresponding operations on the rings Qsym[α] and Qsym,
which commute with the homomorphisms f, fRP , and F∗ .

11.1 The case of P

At first let us consider the ring P. We have

[C, d] = 1 + ξ0

For k > 2 we have the similar situation.

(
n Cdk P n + dk−1 P n , k < n + 1,
dk CP = n n
1 + dk−1 P = C∅ + dk−1 P , k = n + 1

1 ∂k
So we have [dk+1 , C] = dk + k! ∂αk α=0 ξα , k > 0. This can be reformulated in terms of Φ(t).

Proposition 11.1. We have

Φ(t)C = (C + t)Φ(t) + tξt
Proof. Let P be an n-polytope. Then

Φ(t)CP = CP + (CdP + P )t + · · · + (Cdn P + dn−1 P )tn + (1 + dn P )tn+1 = CΦ(t)P + tΦ(t)P + tξt P

Now let us consider the bipyramid. dBP = 2CdP for any polytope of positive dimension, while
dB pt = dI = 2 pt, and 2Cd pt = 0. So dBP = 2CdP + 2ξ0 . For k > 2 we have
(
n 2Cdk P n + dk−1 P n , k < n + 1;
dk BP = n n
2 + dk−1 P = 2C∅ + dk−1 P , k = n + 1;

2 ∂k
So dk+1 B = 2Cdk+1 + dk + k! ∂αk ξα , k > 1.
α=0

51
Proposition 11.2.
Φ(t)B = (B − 2C − t) + (2C + t)Φ(t) + 2tξt
Proof. For a point we have: Φ(t)B pt = Φ(t)I = I + (2 pt)t, and
((B − 2C − t) + (2C + t)Φ(t) + 2tξt ) pt = I − 2I − (pt)t + 2I + (pt)t + 2(pt)t = I + (2 pt)t
Let P be an n-polytope, n > 0. Then

Φ(t)BP = BP + (2CdP )t + (2Cd2 P + dP )t2 + · · · + (2Cdn P + dn−1 P )tn + (2 + dn P )tn+1 =

= (B − 2C − t)P + 2CΦ(t)P + tΦ(t)P + 2tξt P

Let us denote A = 2C − B. Then from Propositions 11.1 and 11.2 we have

Φ(t)A = A + t + tΦ(t)
At last, Propositions 11.1 and 11.2 imply the following formula.
Proposition 11.3.
Φ(t)[B, C] = [B, C] + At + t2
Proof.

Φ(t)[B, C] = Φ(t)(BC − CB) = (B − 2C − t + (2C + t)Φ(t) + 2tξt ) C − ((C + t)Φ(t) + tξt ) B =

= (B−2C −t)C +(2C +t) ((C + t)Φ(t) + tξt )+2t2 ξt −(C +t) (B − 2C − t + (2C + t)Φ(t) + 2tξt )−t2 ξt =
= (BC − CB) − t(B − 2C − t) + 2Ctξt + t2 ξt + 2t2 ξt − 2Ctξt − 2t2 ξt − t2 ξt = [B, C] + At + t2

Now let us apply this formulas to the generalized f -polynomial.

Proposition 11.4. Mappings C, A : Qsym[α] → Qsym[α] defined by the formulas
∞
X
Cg = (α + σ1 )g + tn g(tn , t1 , . . . , tn−1 , 0, 0, . . . )
n=1
∞
X
Ag = αg(α, 0, 0, . . . ) + t1 g(α, t1 , t2 , . . . ) + (tn + tn−1 )g(α, tn , tn+1 , . . . )
n=2

satisfy the relations

f (CP ) = Cf (P ), f (AP ) = Af (P )
for all P ∈ P.
Proof. At first let us prove that this mappings are correctly defined, that is for any g ∈ Qsym[α] its
images Cg, and Ag belong to Qsym[α]. We will use the criterion obtained in Proposition 3.3 and
Remark 3.4. It is easy to see that if g has bounded degree, then Cg and Ag have bounded degree.
Let i > 1. Then

X i−1
X
(Cg)(α, t1 , . . . , ti−1 , 0, ti+1 , . . . ) = (α+ tj )g(α, t1 , . . . , ti−1 , 0, ti+1 , . . . )+ tn g(tn , t1 , . . . , tn−1 , 0, 0, . . . )+
j6=i n=1
∞
X
+ tn g(tn , t1 , . . . , ti−1 , 0, ti+1 , . . . , tn−1 , 0, 0, . . . ) =
n=i+1
i−1
X
= (α + σ1 (t1 , . . . , ti−1 , ti+1 , . . . ))g(α, t1 , . . . , ti−1 , ti+1 , . . . ) + tn g(tn , t1 , . . . , tn−1 , 0, 0, . . . )+
n=1
∞
X
+ti+1 g(ti+1 , t1 , . . . , ti−1 , 0, 0, . . . )+ tn g(tn , t1 , . . . , ti−1 , ti+1 , . . . , tn−1 , 0, 0, . . . ) = (Cg)(α, t1 , . . . , ti−1 , ti+1 , . . . )
n=i+2

52
Now consider A. For i = 1 the relation is clear. Let i > 1. Then
i−1
X
(Ag)(α, t1 , . . . , ti−1 , 0, ti+1 , . . . ) = αg(α, 0, 0, . . . ) + (tn + tn−1 )g(α, tn , . . . , ti−1 , 0, ti+1 , . . . )+
n=1
∞
X
+ (0 + ti−1 )g(α, 0, ti+1 , . . . ) + (ti+1 + 0)g(α, ti+1 , ti+2 , . . . ) + (tn + tn−1 )g(α, tn , tn+1 , . . . ) =
n=i+2
i−1
X
= αg(α, 0, 0, . . . ) + (tn + tn−1 )g(α, tn , . . . , ti−1 , ti+1 , . . . ) + (ti+1 + ti−1 )g(α, ti+1 , ti+2 , . . . )+
n=1
∞
X
+ (tn + tn−1 )g(α, tn , tn+1 , . . . ) = (Ag)(α, t1 , . . . , ti−1 , ti+1 , . . . )
n=i+2

On the other hand, we have

Φ(t1 )C = (C + t1 )Φ(t1 ) + t1 ξt1

Φ(t2 )Φ(t1 )C = (C + t2 + t1 )Φ(t2 )Φ(t1 ) + t2 ξt2 Φ(t1 ) + t1 ξt1
...
Φ(tn )Φ(tn−1 ) . . . Φ(t1 )C = (C + tn + · · · + t1 )Φ(tn )Φ(tn−1 ) . . . Φ(t1 ) + tn ξtn Φ(tn−1 ) . . . Φ(t1 ) + · · · + t1 ξt1
...
∞
X
Φ∞ C = (C + σ1 )Φ∞ + tn f (tn , t1 , . . . , tn−1 , 0, . . . )
n=1
∞
X
f (CP ) = ξα Φ∞ CP = (α + σ1 )ξα Φ∞ P + tn f (tn , t1 , . . . , tn−1 , 0, . . . ) = Cf (P )
n=1

Φ(t1 )A = A + t1 + t1 Φ(t1 )
Φ(t2 )Φ(t1 )A = A + t2 + (t2 + t1 )Φ(t2 ) + t1 Φ(t2 )Φ(t1 )
...
Φ(tn )Φ(tn−1 ) . . . Φ(t1 )A = A + tn + (tn + tn−1 )Φ(tn ) + (tn−1 + tn−2 )Φ(tn )Φ(tn−1 ) + · · · + t1 Φ(tn ) . . . Φ(t1 )
...
∞
X
Φ∞ A = A + t1 Φ∞ (t1 , t2 , . . . ) + (tn + tn−1 )Φ∞ (tn , tn+1 , . . . )
n=2
∞
X
f (AP ) = ξα Φ∞ AP = αf (α, 0, 0, . . . ) + t1 f (α, t1 , t2 , . . . ) + (tn + tn−1 )f (α, tn , tn+1 , . . . ) = Af (P )
n=2

Now the mapping B : Qsym[α] → Qsym[α] can be defined as B = 2C − A. Then f (BP ) = Bf (P ).

11.2 The case of RP

Now let us consider the ring RP. In this ring CP is just the multiplication by the element x = pt. So

Φ(t)CP = Φ(t)(xP ) = (Φ(t)x) > (Φ(t)P ) = (x + t)Φ(t)P = (C + t)Φ(t)P

Φ∞ (CP ) = Φ∞ (xP ) = Φ∞ (x) > Φ∞ (P ) = (x + σ1 )Φ∞ (P ) = (C + σ1 )Φ∞ (P )
fRP (CP ) = fRP (xP ) = fRP (x)fRP (P ) = (α + σ1 )fRP (P )
F∗ (CP ) = F∗ (xP ) = F∗ (x) F∗ (P ) = σ1 F∗ (P )

53
In fact, the existence of the empty polytope ∅ such that dn+1 P n = ∅ for any n-polytope P n makes the
formulas connecting C, B, and dk simpler. We have:

[C, d] = 1, [C, dk+1 ] = dk , k > 1

dB = 2Cd + ε0 , dk+1 B = 2Cdk+1 + dk , k > 1

Proposition 11.5.
Φ(t)B = (B − 2C − t) + (2C + t)Φ(t) + ε0 t
Proof. Indeed,
X
Φ(t)BP = BP + (2CdP + ε0 (P ))t + (2Cd2 P + dP )t2 + (2Cdk P + dk−1 P )tk =
k>3

= (B − 2C − t)P + 2CΦ(t)P + tΦ(t)P + ε0 (P )t

Then
Φ(t)A = A + t + tΦ(t) − ε0 t
Proposition 11.6. The mappings ARP : Qsym[α] → Qsym[α], A0 : Qsym → Qsym defined by the
formulas
∞
X
ARP g = αg(α, 0, 0, . . . ) + t1 g(α, t1 , t2 , . . . ) + (tn + tn−1 )g(α, tn , tn+1 , . . . ) − σ1 g(0, 0, . . . )
n=2
∞
X
A0 g = t1 g(t1 , t2 , . . . ) + (tn + tn−1 )g(tn , tn+1 , . . . ) − σ1 g(0, 0, . . . )
n=2

satisfy the relations

fRP (AP ) = ARP fRP (P ), F∗ (AP ) = A0 F∗ (P )
for all P ∈ RP.
Proof. From the proof of Proposition 11.4 we see that both mappings ARP and A0 are correctly defined.
Then

Φ(t1 )A = A + t1 + t1 Φ(t1 ) − ε0 t1
Φ(t2 )Φ(t1 )A = A + t2 + (t2 + t1 )Φ(t2 ) + t1 Φ(t2 )Φ(t1 ) − ε0 (t2 + t1 )
...
Φ(tn )Φ(tn−1 ) . . . Φ(t1 )A = A + tn + (tn + tn−1 )Φ(tn ) + (tn−1 + tn−2 )Φ(tn )Φ(tn−1 ) + . . .
+t1 Φ(tn ) . . . Φ(t1 ) − ε0 (tn + · · · + t1 )
...
∞
X
Φ∞ A = A + t1 Φ∞ (t1 , t2 , . . . ) + (tn + tn−1 )Φ∞ (tn , tn+1 , . . . ) − ε0 σ1
n=2
∞
X
fRP (AP ) = εα Φ∞ A = εα (AP ) + t1 εα Φ∞ (t1 , t2 , . . . ) + (tn + tn−1 )εα Φ∞ (tn , tn+1 , . . . ) − ε0 (P )σ1 =
n=2
∞
X
= αfRP (α, 0, 0, . . . ) + t1 fRP (α, t1 , t2 , . . . ) + (tn + tn−1 )fRP (α, tn , tn+1 , . . . ) − σ1 fRP (0, 0, . . . ) = ARP fRP (P )
n=2
∞
X
F∗ (AP ) = ε0 Φ∞ A = ε0 (AP ) + t1 ε0 Φ∞ (t1 , t2 , . . . ) + (tn + tn−1 )ε0 Φ∞ (tn , tn+1 , . . . ) − ε0 (P )σ1 =
n=2
∞
X
= t1 F∗ (t1 , t2 , . . . ) + (tn + tn−1 ) F∗ (tn , tn+1 , . . . ) − σ1 F∗ (0, 0, . . . ) = A0 F∗ (P )
n=2

54
 As before BRP and B0 are defined as BRP = 2CRP −ARP , B0 = 2C0 −A0 , where CRP g = (α+σ1 )g,
C0 g = σ1 g, and
fRP (BP ) = BRP fRP (P ), F∗ (BP ) = B0 F∗ (B)
At last

Φ(t)(BC − CB) = (B − 2C − t + (2C + t)Φ(t) + ε0 t) C − (C + t)Φ(t)B =

= BC − (2C + t)C + (2C + t)(C + t)Φ(t) − (C + t) (B − 2C − t + (2C + t)Φ(t) + ε0 t) =
= BC − CB + (2C − B)t + t2 − (C + t)ε0 t = BC − CB + At + t2 − (x + t)tε0

So we obtain the following proposition.

Proposition 11.7. In the ring RP

Φ(t)[B, C] = [B, C] + At + t2 − (x + t)tε0

12 Problem of the Description of Flag Vectors of Polytopes

We have mentioned that on the space of simple polytopes

f (α, t1 , t2 , . . . )(P n ) = f1 (α, t1 + t2 + . . . )(P n ) = f1 (α, σ1 )(P n ).

The only linear relation on the polynomial f1 is f1 (−α, α + t) = f1 (α, t), and it is equivalent to the
Dehn-Sommerville relations. In fact, for the polynomial g = g(α, σ1 ) ∈ Qsym[t1 , t2 , . . . ][α] this condition
is necessary and sufficient to be an image of an integer combination of simple polytopes.
One of the outstanding results in the polytope theory is the so-called g-theorem, which was formulated
as a conjecture by P. McMullen [Mc1] in 1970 and proved by R. Stanley [St1] (the necessity) and L. Billera
and C. Lee [BL] (the sufficiency) in 1980.
For an n-dimensional simple polytope P n let us define an h-polynomial:

h(α, t)(P n ) = h0 αn + h1 αn−1 t + · · · + hn−1 αtn−1 + hn tn = f1 (α − t, t)(P n )

Since f1 (−α, α + t) = f1 (α, t), the h-polynomial is symmetric: h(α, t) = h(t, α). So hi = hn−i .
Definition 12.1. A g-vector of a simple polytope P n is the set of numbers g0 = 1, gi = hi −hi−1 , 1 6 i 6 [ n2 ].
For any positive integers a and i there exists a unique ”binomial i-decomposition” of a:
     
ai ai−1 aj
a= + + ···+ ,
i i−1 j
where ai > ai−1 > · · · > aj > j > 1.
Let us define      
ai + 1 ai−1 + 1 aj + 1
ahii = + + ···+ , 0hii = 0.
i+1 i j+1






For example, a = a1 , so ah1i = a+1 2 = (a+1)a
2 ; for i ≥ a we have: a = ii + i−1
i−1 + · · · +
i−a+1
i−a+1 , so
ahii = a.
Theorem (g-theorem, [St1], [BL]). Integer numbers (g0 , g1 , . . . , g[ n2 ] ) form the g-vector of some simple
n-dimensional polytope if and only if they satisfy the following conditions:
hni
hii
g0 = 1, 0 6 g1 , 0 6 gi+1 6 gi , i = 1, 2, . . . , − 1.
2
As it is mentioned in [Z1] even for 4-dimensional non-simple polytopes the corresponding problem
involving flag f -vectors is extremely hard.
Modern tools used to obtain linear and nonlinear inequalities satisfied by flag f -numbers are based
on the notion of a cd-index [BK, BE1, BE2, EF, ER, L, St3], a toric h-vector [BE, BL, K, St2, Sts] and
its generalizations [F1, F2, L], a ring of chain operators [BH, BLiu, Kalai].

55
 In [St1] R. Stanley constructed for a simple polytope P a projective toric variety XP so that {hi }
are the even Betti numbers of the (singular) cohomology of XP . Then the Poincare duality and the
Hard Lefschetz theorem for XP imply the McMullen conditions for P .
In [Mc2] P. McMullen gave the purely geometric proof of these conditions using the notion of a
polytope algebra.
In [St2] Stanley generalized the definition of an h-vector to an arbitrary polytope P in such a way
that if a polytope is rational the generalized ĥi are the intersection cohomology Betti numbers of the
associated toric variety (see the validation of the claim in [Fi]).
The set of numbers (ĥ0 , . . . , ĥn ) is called a toric h-vector. It is nonnegative and symmetric: ĥi = ĥn−i ,
and in the case of rational polytope the Hard Lefschetz theorem in the intersection cohomology of the
associated toric variety proves the unimodality

1 = ĥ0 6 ĥ1 6 · · · 6 ĥ[ n2 ]

Generalizing the geometric method introduced by P. McMullen [Mc2] Kalle Karu [K] proved the uni-
modality for an arbitrary polytope.
The toric h-vector consists of linear combinations of the flag f -numbers. In general case it does not
contain the full information about the flag f -vector. In [L] C. Lee introduced an ”extended toric” h-
vector that carries all the information about the flag f -numbers and consists of a collection of nonnegative
symmetric unimodal vectors. In [F1] and [F2] J. Fine introduced the generalized h- and g-vectors that
contain the toric h- and g-vectors as subsets, but can have negative entries.
Let us look on the problem of the description of flag f -vectors from the point of view of the ring of
polytopes.
g-theorem describes all the polynomials ψ(α, t1 , t2 , . . . ) ∈ Qsym[t1 , t2 , . . . ][α] that are images of
simple polytopes under the ring homomorphism f :

ψ(α, t1 , t2 , . . . ) = f (α, t1 , t2 , . . . )(P n ), P n – a simple n-dimensional polytope.

In the general case we know the criterion when the polynomial ψ ∈ Qsym[t1 , t2 , . . . ][α] is the image
of an integer combination of polytopes.

Question 12.2. Find a criterion when the polynomial ψ ∈ Qsym[t1 , t2 , . . . ][α] is the image of an
n-dimensional polytope.

A Join
Proposition A.1. P n1 > Qn2 is an (n1 + n2 + 1)-dimensional polytope.
Faces of P n1 >Qn2 up to an affine equivalence are exactly F >G, where ∅ ⊆ F ⊆ P n1 , ∅ ⊆ G ⊆ Qn2
are faces of P n1 and Qn2 respectively.
Proof. The formula for dimension of P n1 > Qn2 follows from Remark 2.13.
The nonempty faces of a polytope are exactly the subsets of the polytope that minimize some linear
function on it. Let c = (c1 , c2 ) be a linear function in (Rn1 +1 × Rn2 +1 )∗ = (Rn1 +1 )∗ × (Rn2 +1 )∗ . Then
c1 has its minimum c1 on the face F of P n1 , and c2 has its minimum c2 on the face G of Qn2 . Since
any point of P n1 > Qn2 has the form t1 p + t2 q, where p ∈ P n1 , q ∈ Qn2 , t1 , t2 > 0, t1 + t2 = 1, we have
• if c1 < c2 , then c has its minimum on the face F > ∅ of P n1 > Qn2 ;
• if c1 > c2 , then c has its minimum on ∅ > G;
• if c1 = c2 , then c has its minimum on the set

X = {t1 p + t2 q, p ∈ F, q ∈ G, t1 , t2 > 0, t1 + t2 = 1}.

Let F lie in the k-simplex conv{p1 , . . . , pk+1 } ⊂ aff(F ), and G lie in the l-simplex
−−→
conv{q1 , . . . , ql+1 } ⊂ aff(G). Now let {Opi , i = 1, . . . , k+1} be a basis in Rk+1 = aff{O, pi , i = 1, . . . , k+1}
−−→
and {Oqj , j = 1, . . . , l + 1} be a basis in Rl+1 = aff{O, qj , j = 1, . . . , l + 1}. Then X = F > G.

56
 Note that c1 can have the constant value on P n1 , or c2 can have the constant value on Qn2 . In these
cases F = P n1 , or G = Qn2 .
Let λP be the linear function equal to the sum of all the coordinates in Rn1 +1 , and λQ be the linear
function equal to the sum of all the coordinates in Rn2 +1 . Then

λP (p) = 1, λQ (p) = 0 for any p ∈ P n1 = P n1 > ∅

λP (q) = 0, λQ (q) = 1 for any q ∈ Qn2 = ∅ > Qn2

Now let F 6= ∅, F ⊆ P n1 be a face of P n1 , and let c1 ∈ (Rn1 +1 )∗ be a linear function that has
its minimum c1 exactly on F .
Similarly, let (G, c2 , c2 ) be the corresponding data for Qn2 . Then let
∗
c = (c1 , c2 ) ∈ Rn1 +1 × Rn2 +1 . By an addition or a subtraction of λP and λQ with coefficients we
can obtain any of the cases c1 < c2 , c1 = c2 , or c1 > c2 , so F > ∅, F > G, ∅ > G are faces of P n1 > Qn2 .
Thus faces of the polytope P n1 > Qn2 up to an affine equivalence have the form

F > G, ∅ ⊆ F ⊆ P n1 , ∅ ⊆ G ⊆ Qn2 .

B Structure theorems
Proposition B.1. P is a polynomial ring generated by indecomposable combinatorial polytopes.
Proof. We will show that any polytope P n of positive dimension can be represented in a unique up to the
order of factors way as a product of indecomposable polytopes of positive dimensions P = P1 × · · · × Pk .
It is clear that such a decomposition exists. Now let

P1 × · · · × Pk = Q1 × · · · × Qs ,

where
1 6 dim P1 6 dim P2 6 · · · 6 dim Pk , and 1 6 dim Q1 6 dim Q2 6 · · · 6 dim Qs .
Let us consider a face F of the form v1 × · · · × vk−1 × Pk ≃ Pk , where vi are vertices. Then we have
F = G1 × · · · × Gs , where Gi is a face of Qi . Since Pk is indecomposable, we see that all Gj but one
are points, and Pk is combinatorially equivalent to a face of some Qr . Similarly Qr is combinatorially
equivalent to a face of some Pt . Then dim Pk 6 dim Qr 6 dim Pt 6 dim Pk , therefore, Pk = Qr .
Similarly Qs ≃ Pl for some l. So dim Pk = dim Qs and without loss of generality we can set Pk = Qs .
If Pk = I then Pi = Qj = I for all i, j, so the statement is true. Let Pk 6= I. We know that

v1 × · · · × vk−1 × Pk = w1 × · · · × ws−1 × Pk

Let us consider a face G of the form v1 × · · · × vi−1 × I × vi+1 × · · · × vk−1 × Pk , where I is the edge that
joins the vertices vi and v′i of the polytope Pi . Then G = w1 × · · · × wj−1 × J × wj+1 × · · · × ws−1 × Pk ,
where J is the edge that joins the vertices wj and w′j of some polytope Qj .
Since v1 × · · · × v′i × · · · × Pk ⊂ G, Pk is indecomposable, and Pk 6= J, we have

v1 × · · · × v′i × · · · × Pk = w1 × · · · × w′j × · · · × Pk .

Changing the edges we can achieve any vertex v′1 × · · · × v′k−1 of P1 × · · · × Pk−1 , so for any such a
vertex we have v′1 × · · · × v′k−1 × Pk = w′1 × · · · × w′s−1 × Pk . Therefore for any face F1 × · · · × Fk−1 × Pk

F1 × · · · × Fk−1 × Pk = G1 × · · · × Gs−1 × Pk

Thus for any face F of the polytope P1 × · · · × Pk−1 there is a face G of the polytope Q1 × · · · × Qs−1
such that F × Pk = G × Pk , and vice versa. Moreover, F ⊂ F ′ if and only if G ⊂ G′ . Thus

P1 × · · · × Pk−1 = Q1 × · · · × Qs−1

Iterating this step in the end we obtain that k = s, and up to the order of the polytopes Pi = Qi ,
i = 1, . . . , k.

57
Proposition B.2. RP is a ring of polynomials generated by all join-indecomposable polytopes.
Proof. Let P1 > P2 · · · > Pk = Q1 > Q2 > · · · > Qs , where Pi , Qj are polytopes,

0 6 dim P1 6 dim P2 6 · · · 6 dim Pk , 0 6 dim Q1 6 dim Q2 6 · · · 6 dim Qs

and each polytope Pi and Qj can not be decomposed as a join of polytopes of lower dimensions. Let
dim Pk > dim Qs . Consider the face ∅>∅>· · ·>Pk . It can be represented in the form G1 >G2 >· · ·>Gs ,
where Gj is a face of Qj . Since Pk is join-indecomposable, all Gj but one should be equal to ∅, and
one of them should be equal to Pk . Let Pk = Gr . Then Pk is a face of Qr , so Pk = Qr , since
dim Pk > dim Qs > dim Qr . Without loss of generality we can take r = s.
We have P ′ > Pk = Q′ > Pk , and the combinatorial equivalence identifies the face ∅ > Pk of the first
polytope with the same face ∅ > Pk of the second. Consider the face P ′ > ∅. It should have the form
G1 > G2 where G1 is a face of Q′ and G2 is a face of Pk . Then ∅ > G2 should be a face of ∅ > Pk and
P ′ > ∅, therefore G2 = ∅ and G1 = P ′ . Since dim P ′ = dim Q′ , we see that P ′ = Q′ .
So P1 > P2 > · · · > Pk−1 = Q1 > Q2 > . . . Qs−1 . Let s > k. Iterating this step in the end we obtain
P1 = Q1 > Q2 > · · · > Qs−k+1 . Therefore k = s and P1 = Q1 .
Thus each polytope can be decomposed in a unique up to permutations way as P1 > · · · > Pk , where
Pj are join-indecomposable. Since RP additively is a free abelian group generated by combinatorial
polytopes and ∅, it is a ring of polynomials generated by all join-indecomposable polytopes.

C Examples
Proposition C.1. Any monomial Mω such that Mω ∈ F(RP) has the form M(2k+1) , k > 0.
Proof. Indeed, for the monomial M(a1 , ..., ak ) we have

M(a1 , ..., ak ) (t, −t, t1 , t2 , . . . ) = M(a1 , ..., ak ) (t1 , t2 , . . . ) + (ta1 + (−t)a1 ) M(a2 , ..., ak ) (t1 , t2 , . . . )+
+ (−1)a2 ta1 +a2 M(a3 , ..., ak )

If k > 2 the last summand does not cancel, so k = 1. For M(n) with n = 2l + 1 we have

M(2l+1) (t1 , . . . , ti−1 , t, −t, ti , . . . ) = t2l+1

1 +· · ·+t2l+1
i−1 +t
2l+1
−t2l+1 +t2l+1
i +· · · = M(2l+1) (t1 , . . . , ti−1 , ti , . . . ),

while for n = 2l the term containing t2l + (−t)2l does not cancel.
Proposition C.2. M(2k+1, 2l+1) − M(2l+1, 2k+1) ∈ F(RP)
Proof. Indeed,

M(2k+1, 2l+1) (t1 , . . . , ti−1 , t, −t, ti , . . . ) = M(2k+1, 2l+1) (t1 , . . . , ti−1 , ti , . . . )+

X X
+ t2k+1
j (t2l+1 + (−t)2l+1 ) + (t2k+1 + (−t)2k+1 )t2l+1 j + (−1)2l+1 t2k+2l+2
j<i j>i

Then


M(2k+1, 2l+1) − M(2l+1, 2k+1) (t1 , . . . , ti−1 , t, −t, ti , . . . ) =


M(2k+1, 2l+1) − M(2l+1, 2k+1) (t1 , . . . , ti−1 , ti , . . . )

This proposition can be generalized.

Proposition C.3. Let (j1 , . . . , jk ) be the set of odd positive integer numbers. Then the quasi-symmetric
polynomial X
(−1)σ M(jσ(1) , ..., jσ(k) )
σ∈Sk

belongs to the image of F.

58
Proof. Indeed, since jl are odd, we have

M(j1 , ..., jk ) (t1 , . . . , ti−1 , t, −t, ti , . . . ) = M(j1 , ..., jk ) (t1 , . . . , ti−1 , ti , . . . )−

X js+3
− tjl11 . . . tjlss tjs+1 +js+2 tls+3 . . . tjlkk
l1 <···<ls <i6ls+3 <···<lk

For any pair (js+1 , js+2 ) there exists a unique monomial M(j1 , ..., js , js+2 , js+1 , js+3 , ..., jk ) , and this mono-
j j j j
mial has the opposite sign. Therefore all the monomials of the form tl1σ(1) . . . tlsσ(s) tjσ(s+1) +jσ(s+2) tls+3
σ(s+3)
. . . tlkσ(k)
vanish.
Let us calculate the image of the Ehrenborg transformation F in small dimensions.
n = 0 The image of the point x = pt is M(1) = σ1 . So

F(RP 2 ) = R∗ ((D∗ )2 ) = Zhσ1 i.

n = 1 The image of the segment I = pt ∗ pt is M(2) + 2M(1, 1) = σ12 . So

F(RP 4 ) = R∗ ((D∗ )4 ) = Zhσ12 i.

2
n = 2 The image of the m-gon Mm is

M(3) + mM(2, 1) + mM(1, 2) + 2mM(1, 1, 1) = M(3) + m(σ1 σ2 − σ3 )

Therefore,
F(RP 6 ) = R∗ ((D∗ )6 ) = ZhM(3) , σ1 σ2 − σ3 i,
where
M(3) = F(4∆2 − 3I 2 ), σ1 σ2 − σ3 = F(I 2 − ∆2 ),

n = 3 The image of a 3-dimensional polytope is

M(4) + f0 M(1, 3) + f1 M(2, 2) + f2 M(3, 1) + f01 M(1, 1, 2) + f02 M(1, 2, 1) + f12 M(2, 1, 1) + f012 M(1, 1, 1, 1)

Since f0 − f1 + f2 = 2, f01 = f02 = f12 = 2f1 , and f012 = 4f1 , we obtain

M(4) + 2M(3, 1) + f0 M(1, 3) − M(3, 1) + f1 M(2, 2) + M(3, 1) + 2 (σ1 σ3 − 2σ4 ) ,

Then

F(RP 8 ) = R∗ ((D∗ )8 ) = ZhM(4) + 2M(3, 1) , M(1, 3) − M(3, 1) , M(2, 2) + M(3, 1) + 2 (σ1 σ3 − 2σ4 )i,

where

M(4) + 2M(3, 1) = F(5∆3 − 6CI 2 + 2B∆2 ), M(1, 3) − M(3, 1) = F(−∆3 + 3CI 2 − 2B∆2 )
M(2, 2) + M(3, 1) + 2 (σ1 σ3 − 2σ4 ) = F(−CI 2 + B∆2 )

References
[ABS] M. Aguiar, N. Bergeron, F. Sottile, Combinatorial Hopf Algebras and Generalized Dehn-
Sommerville relations. Compositio Mathematica 142(1) 2006, 1–30, arXiv: math/0310016v1
[math.CO].
[BR] A. Baker, B. Richter, Quasisymmetric functions from a topological point of view. Math. Scand. 103
(2008), 208–242.
[BB] M. M. Bayer, L. J. Billera, Generalized Dehn-Sommerville relations for polytopes, spheres and
Eulerian partially ordered sets. Invent. Math. 79: 143–157, 1985.

59
[BE] M. M. Bayer and R. Ehrenborg, The toric h-vectors of partially ordered sets. Trans. Amer. Math.
Soc. 352 (2000), 4515–4531 (electronic).
[BK] M. M. Bayer, A. Klapper, A new index for polytopes. Discrete Comput. Geom. 6, 33–47 (1991).
[BL] M. M. Bayer and C. W. Lee, Combinatorial aspects of convex polytopes. in Handbook of convex
geometry, Vol. A, B, North-Holland, Amsterdam, 1993, pp. 485–534.
[BMSW] N. Bergeron, S. Mykytiuk, F. Sottile, and S. van Willigenburg, Pieri Operations on Posets.
J. of Comb. Theory Series A 91(2000) 84-110. arXiv: math/0002073v2 [math.CO] 21 Jul 2000.
[BE1] L. Billera, R. Ehrenborg, Monotonicity of the cd-index for polytopes. Math. Z. 233(200) 421-441.
[BE2] L. Billera, R. Ehrenborg and M. Readdy, The cd-index of zonotopes and arrangements. Mathe-
matical Essays in Honor of Gian-Carlo Rota, Birkhauser, Boston, 1998.

[BH] L. Billera, G. Hettey, Linear inequalities for flags in graded posets. J. of Combinatorial Theory
Series A, 2000, V89, N1, 77-104 , arXiv: math/9706220v1 [math.CO]
[BL] L. Billera, C. Lee, A proof of sufficiency of McMullen’s conditions for f -vectors of simplicial poly-
topes. J. Combin. Theory, Ser. A, 1981, V31, N3, 237-255.
[BLiu] L. Billera, N. Liu, Non-commutative enumeration in graded posets. Journal of Algebraic Combi-
natorics 12 (2000), 7-24.

[BS] R. Bott, H. Samelson, On the Pontryagin product in spaces of paths. Comment. Math. Helv., 27,
1953, 320–337
[BP] V. M. Buchstaber, T. E. Panov, Torus actions and their applications in topology and combina-
torics, Providence, R.I.: American Mathematical Society, 2002. (University Lecture Series; V.24).
[BR] V. M. Buchstaber, N. Ray, An invitation to toric topology: Vertex four of a remarkable tetrahedron.
Toric topology. Contemporary Mathematics (AMS), v. 460, Providence, R.I., 2008, 1–27; MIMS,
2008.31: http://www.manchester.ac.uk/mims/eprints.
[Buch] V. M. Buchstaber, Ring of Simple Polytopes and Differential Equations. Proceedings of the
Steklov Institute of Mathematics, v. 263, 2008, 1–25.
[BuchE] Victor M. Buchstaber, Nickolai Erokhovets, Ring of Polytopes, Quasi-symmetric functions and
Fibonacci numbers, arXiv:1002.0810v1 [math.CO].

[BuchE1] V. M .Buchstaber, N. Erokhovets Algebra of operators on the ring of polytopes and quasi-
symmetric functions, Russian Mathematical Surveys (2010),65(2):381.
[C] H. S. M. Coxeter, Regular Polytopes. 3rd edition, Dover Publications, 1973.
[CFL] K. T. Chen, R. H. Fox, and R. C. Lyndon, Free differetional calculus. IV, Ann. Math. 68 (1958),
81–95.
[Ehr] R. Ehrenborg, On Posets and Hopf Algebras. Advances in Mathematics, 119, 1–25 (1996).
[EF] R. Ehrenborg, H. Fox, Inequalities for cd-indices of joins and products of polytopes. Combinatorica
23 (2003), 427-452.
[EKZ] D. Eppstein, G. Kuperberg, G. M. Ziegler, Fat 4-polytopes and fatter 3-spheres. Discrete Ge-
ometry: in honour of W. Kuperberg’s 60th birthday, Pure and Applied Mathematics. A series of
Monographs and Textbooks, vol. 253, Marcel Dekker, Inc., 2003, pp.239-265.
[ER] R. Ehrenborg, M, Readdy, Coproducts and the cd-index. J. Algebraic Combin. 8 (1998), 273-299.
[Fi] K. -H. Fieseler, Rational Intersection Cohomology of Projective Toric Varieties. Journal f.d. reine
u. angew. Mathematik 413 (1991), 88-98.

60
[F1] J. Fine, A complete h-vector for convex polytopes. Preprint, arXiv: 0911.5722v1 [math.CO]
[F2] J. Fine, A complete g-vector for convex polytopes. Preprint, arXiv: 1001.1562v1 [math.CO]
[Fr] A. Frohmader, Face vectors of flag complexes. Israel Journal of Mathematics, 164, 2008, 153–164.
[Gb] B. Grunbaum, Convex Polytopes. Vol. 221 of Graduate Texts in Mathematics, Springer-Verlag,
New York, Second ed., 2003.
[H] M. Hazewinkel, The Algebra of Quasi-Symmetric Functions is free over integers. Advances in Math-
emetics 164, 283–300 (2001).
[JR] S. Joni and G.-C. Rota, Coalgebras and bialgebras in combinatorics. Stud. Appl. Math. 61(1979),
93-139.
[Kalai] G. Kalai, A new basis for polytopes. J. Combinatorial Theory, Ser A 49 (1988), 191–208.
[K] Kalle Karu, Hard Lefschetz Theorem for Nonrational Polytopes. Invent. Math. 157 (2004), 419–447,
arXiv: math/0112087v4 [math.AG] 17 Dec 2002.
[L] Carl W. Lee, Sweeping the cd-Index and the Toric h-Vector. 2009, preprint:
http://www.ms.uky.edu/∼lee/cd.pdf
[Mc1] P. McMullen, The numbers of faces of simplicial polytopes. Israel J. Math. 1971, V9, 559-570.
[Mc2] P. McMullen, On Simple Polytopes. Invent. math. 113, 419-444 (1993).
[N] S. P. Novikov, Various doublings of Hopf algebras. Operator algebras on quantum groups, complex
cobordisms. Russian Mathematical Surveys, 47:5, 1992, 198–199.
[RS] N. Ray, W. Schmitt, Combinatorial models for coalgebraic structures. Advances in Mathematics
1998, Vol 138, Iss 2, 211-262.
[Re] C. Reutenauer, Free Lie algebras. Oxford Univ. Press, Oxford, UK, 1993.
[Sch1] William Schmitt, Antipodes and incidence coalgebras. Journal of Combinatorial Theory A, 46:264-
290, 1987.
[Sch2] William Shcmitt, Incidence Hopf Algebras. J. Pure Appl. Algebra 96(1994), 299-330.
[St1] R. P. Stanley, The number of faces of simplicial convex polytope. Advances in Math. 1980. V. 35,
N3, 236-238.
[St2] R. P. Stanley, Generalized h-vectors, intersection cohomology of toric varieties, and related results.
Commutative Algebra and Combinatorics, Advanced Studies in Pure Mathematics 11, Kinokuniya,
Tokyo, and North-Holland, Amsterdam/New York, 1987, pp. 187-213.
[St3] R. P. Stanley, Flag f-vectors and the cd-index. Math. Z. 216, 483–499 (1994).
[Sts] C. Stenson, Relationships among flag f-vector inequalities for polytopes. Discrete and Computa-
tional Geometry 31 (2004), pp. 257-273
[Z1] G. M. Ziegler, Face numbers of 4-polytopes and 3-spheres. Proceedings of the ICM 2002 Beijing,
Volume III, 2004, 625–634; arXiv: math/0208073v2 [math.MG] 20 Jun 2003.
[Z2] G. M. Ziegler, Lectures on Polytopes. Springer-Verlag, New York, 2007.

61