FREE QUASI-SYMMETRIC FUNCTIONS AND DESCENT

ALGEBRAS FOR WREATH PRODUCTS, AND

NONCOMMUTATIVE MULTI-SYMMETRIC FUNCTIONS
arXiv:0806.3682v1 [math.CO] 23 Jun 2008

JEAN-CHRISTOPHE NOVELLI AND JEAN-YVES THIBON

Abstract. We introduce analogs of the Hopf algebra of Free quasi-symmetric

functions with bases labelled by colored permutations. When the color set is a
semigroup, an internal product can be introduced. This leads to the construction
of generalized descent algebras associated with wreath products Γ ≀ Sn and to the
corresponding generalizations of quasi-symmetric functions. The associated Hopf
algebras appear as natural analogs of McMahon’s multisymmetric functions. As a
consequence, we obtain an internal product on ordinary multi-symmetric functions.
We extend these constructions to Hopf algebras of colored parking functions, colored
non-crossing partitions and parking functions of type B.

Contents
1. Introduction 2
2. Background and notations 3
2.1. Colored alphabets 4
2.2. Colored standardization 4
2.3. Colored shifted operations 4
3. Free quasi-symmetric functions of level ℓ 5
3.1. FQSym(ℓ) and FQSym(Γ) 5
3.2. Duality in FQSym(ℓ) 6
3.3. Algebraic structure of FQSym(ℓ) 8
3.4. Primitive elements of FQSym(ℓ) 8
3.5. Dendriform structure of FQSym(ℓ) 9
3.6. Internal product of FQSym(ℓ) 9
4. Noncommutative symmetric functions of level ℓ 10
4.1. ℓ-partite numbers 10
4.2. The Hopf algebra Sym(ℓ) 10
4.3. Algebraic structure of Sym(ℓ) 11
4.4. Primitive elements of Sym(ℓ) 12
4.5. Duality in Sym(ℓ) 12
5. Quasi-symmetric functions of level ℓ 13
5.1. Cauchy formula of level ℓ 13
5.2. Poirier’s Quasi-symmetric functions 13
5.3. Coproducts and alphabets 14
Date: November 14, 2021.
1
2 J.-C. NOVELLI AND J.-Y. THIBON

5.4. Generalized descent algebras 17

5.5. Ordinary multi-symmetric functions 18
6. The Mantaci-Reutenauer algebra 19
6.1. Monochromatic complete functions 19
6.2. Primitive elements of MR(ℓ) 20
6.3. Duality 21
7. Level ℓ parking quasi-symmetric functions 21
7.1. Usual parking functions 21
7.2. Colored parking functions 22
7.3. Duality 23
7.4. Algebraic structure of PQSym(ℓ) 23
7.5. Primitive elements of PQSym(ℓ) 24
7.6. Bidendriform and tridendriform structure 25
8. Type B algebras 26
8.1. Parking functions of type B 26
8.2. Non-crossing partitions of type B 26
9. Colored analogs of planar binary trees: PBT(ℓ) 27
9.1. Definition of PBT(ℓ) 27
9.2. Algebraic structure of PBT(ℓ) 28
10. Examples 29
10.1. Multigraded coinvariants and colored Klyachko idempotents 29
10.2. A formula of Raney 32
References 33

1. Introduction
The Hopf algebra of Free Quasi-Symmetric Functions FQSym [4] is an algebra of
noncommutative polynomials associated with the sequence (Sn )n≥0 of all symmetric
groups. It is connected by Hopf homomorphisms to several other important algebras
associated with the same sequence of groups : Free symmetric functions (or coplactic
algebra) FSym [36, 4], Non-commutative symmetric functions (or descent algebras)
Sym [7], Quasi-Symmetric functions QSym [9], Symmetric functions Sym, and also,
Planar binary trees PBT [22, 12], Matrix quasi-symmetric functions MQSym [4, 11],
Parking functions PQSym [19, 29], and so on.
Most of these Hopf algebras are endowed with an internal product, generalizing the
one of ordinary symmetric functions. The basic example is provided by noncommu-
tative symmetric functions, whose homogeneous components can be identified with
the Solomon descent algebras of symmetric groups [7].
Symmetric groups are the Coxeter groups of type A, and there are descent algebras
for other types as well. However, the direct sums of the descent algebras of types B or
D are not Hopf algebras in any natural way. But there are Hopf algebras associated
with wreath products Zℓ ≀Sn , the Mantaci-Reutenauer algebras [27], which admit
internal products, and contain the Solomon algebras of type B for ℓ = 2.
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 3

From the point of view of symmetric functions, MR(ℓ) , the Mantaci-Reutenauer

algebra of level ℓ is the free product of ℓ copies of Sym. It is therefore the natural
noncommutative analog of (Sym)⊗ℓ ≃ Sym(X0 ; . . . ; Xℓ−1 ), the algebra of symmetric
functions in ℓ independent alphabets, which is also the Grothendieck ring of the
tower of algebras (C[Zℓ ≀Sn ]). And indeed, it has been shown in [14] that MR(ℓ) was
the Grothedieck ring of projective modules over the 0-Ariki-Koike-Shoji algebras, a
degeneracy of the Hecke algebras associated with Zℓ ≀Sn .
However, with ℓ independent alphabets, one can build a larger Hopf algebras. In
the commutative case, it is the algebra of multi-symmetric functions, first introduced
by McMahon [26], and briefly investigated by Gessel from a modern point of view in
[10]. It is defined as follows. Setting Xi = {xi,j |j = 1, . . . , n}, the multi-symmetric
polynomials are the invariants of Sn in C[X0 , . . . , Xℓ−1] for the diagonal action (by
the automorphisms σ(xi,j ) = xi,σ(j) ). This is an algebra, which, as usual, acquires a
Hopf algebra structure in the limit n → ∞.
In the following, we will start with a level ℓ analogue of FQSym, whose bases are
labelled by ℓ-colored permutations. Imitating the embedding of Sym in FQSym, we
obtain a Hopf subalgebra of level ℓ called Sym(ℓ) , which is a natural noncommutative
analog of McMahon’s algebra of multi-symmetric functions, and turns out to be dual
to Poirier’s quasi-symmetric functions. Its homogenous components can be endowed
with an internal product, thus providing an analog of Solomon’s descent algebras for
wreath products, bigger than the Mantaci-Reutenauer algebras, and in which most
useful properties such as the splitting formula remain valid. By commutative image,
this yields an internal product on multi-symmetric functions.
The Mantaci-Reutenauer descent algebra MR(ℓ) arises as a natural Hopf subalge-
bra of Sym(ℓ) and its dual is computed in a straightforward way by means of an
appropriate Cauchy formula.
Finally, we introduce a Hopf algebra of colored parking functions, and use it to
define Hopf algebras structures on parking functions and non-crossing partitions of
type B.
The main results of this paper have been announced in the draft [32]. Since then,
some of these results, in particular the construction of Sym(ℓ) , have been used by
Baumann and Hohlweg [2], whose paper provide detailed proofs. Hence, we shall only
include the proofs which cannot be found in their paper. In particular, we propose
an alternative approach to the internal product, which is introduced by a duality
argument, and derive its main properties from those of the dual coproduct.
Acknowledgements. This project has been partially supported by the grant ANR-06-BLAN-0380.
The authors would also like to thank the contributors of the MuPAD project, and especially of
the combinat part, for providing the development environment for their research (see [15] for an
introduction to MuPAD-Combinat).

2. Background and notations

We first explain how to adapt the classical definitions and operations to the ℓ-
colored case.
4 J.-C. NOVELLI AND J.-Y. THIBON

2.1. Colored alphabets. We shall start with an ℓ-colored alphabet

(1) A = A0 ⊔ A1 ⊔ · · · ⊔ Aℓ−1 ,
such that all Ai are of the same cardinality N, which will be assumed to be infinite
in the sequel. Let C be the alphabet {c0 , . . . , cℓ−1 } and A be the auxiliary ordered
alphabet {1, 2, . . .} (the letter C stands for colors and A for alphabet) so that A can
be identified with the cartesian product A × C:
(2) A ≃ A × C = {(a, c), a ∈ A, c ∈ C}.
A colored letter (i, c) will be denoted in bold type i. Given two colored words, their
concatenation is obtained by concatenating separately the elements coming from A
and from C. We will sometimes allow ℓ = ∞.
2.2. Colored standardization. Let w be a word in A, represented as (v, u) with
v ∈ A∗ and u ∈ C ∗ . Then the colored standardized word Std(w) of w is
(3) Std(w) := (Std(v), u),
where Std(v) is the usual standardization on words.
Recall that the standardization process sends a word v of length n to a permutation
Std(v) ∈ Sn , called the standardized of v, defined as the permutation obtained
by iteratively scanning v from left to right, and labelling 1, 2, . . . the occurrences
of its smallest letter, then numbering the occurrences of the next one, and so on.
Alternatively, Std(v) is the permutation having the same inversions as v.
For example, Std(abcadbdaa) = 157286934:
a b c a d b d a a
(4) a1 b5 c7 a2 d8 b6 d9 a3 a4
1 5 7 2 8 6 9 3 4
so that
(5) Std(abcadbdaa, 144120100) = (157286934, 144120100).
2.3. Colored shifted operations. For an element v = (v1 , v2 , . . .) of A and an
integer k, denote by v[k] the shifted word (v1 + k) · · · (vn + k), e.g., 312[4] = 756.
Given a colored word α = (α, u), we set α[k] = (α[k], u).
The shifted concatenation of two words v and v ′ is defined by
(6) v • v ′ := v · v ′ [k]
where k is the length of v.
The shifted concatenation of two colored words (v, c), (v ′ , c′ ) is defined by
(7) (v, c) • (v ′ , c′ ) := (v · v ′ [k], c · c′ ).
For example,
(8) (13241, 00322) • (12, 23) = (1324167, 0032223).
Finally, recall that the shuffle product of two words au and bv is defined by
(9) au bv = a(u bv) + b(au v),
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 5

where a and b are letters and u and v are words, with the initial conditions
(10) u ǫ=ǫ u = u, ǫ being the empty word.
This extends to the colored case, considering colored words as the concatenation of
biletters.
The shifted shuffle product is
(11) u ⋒ v := u v[k],
where k is the size of u.

3. Free quasi-symmetric functions of level ℓ

3.1. FQSym(ℓ) and FQSym(Γ) . A colored permutation is a pair (σ, u), with σ ∈ Sn
and u ∈ C n , the integer n being the size of this permutation.
Definition 3.1. The dual free colored quasi-ribbon Gσ,u labelled by a colored per-
mutation (σ, u) of size n is the noncommutative polynomial
X
(12) Gσ,u := w ∈ ZhAi.
w∈An ;Std(w)=(σ,u)

Recall that the convolution of two permutations σ and µ is the set σ∗µ (identified
with the formal sum of its elements) of permutations τ such that the standardized
word of the |σ| first letters of τ is σ and the standardized word of the remaining
letters of τ is µ (see [39]). We then have:
Theorem 3.2. Let (σ ′ , u′) and (σ ′′ , u′′) be colored permutations. Then
X
(13) Gσ′ ,u′ Gσ′′ ,u′′ = Gσ,u′ ·u′′ .
σ∈σ′ ∗σ′′

Therefore, the dual free colored quasi-ribbons span a Z-subalgebra FQSym(ℓ) of the
free associative algebra.
Proof – This is immediate from the product of the usual free quasi-symmetric func-
tions Gσ :
X
(14) Gσ′ Gσ′′ = Gσ .
σ∈σ′ ∗σ′′

Note that all colored permutations indexing a product of G have given colors at
the same places. For example,
G(21,41) G(12,31) = G(2134,4131) + G(3124,4131) + G(4123,4131)
(15)
+ G(3214,4131) + G(4213,4131) + G(4312,4131) .
One can define a coproduct by the usual trick of sums of alphabets: observe that
we only need a total order on A to define the colored standardisation, so that taking
two isomorphic copies A′ and A′′ of A, we define A′ ⊕ A′′ as (A′ ⊕ A′′ ) × C, where
6 J.-C. NOVELLI AND J.-Y. THIBON

A′ ⊕ A′′ denotes the ordered sum. Assuming furthermore that A′ and A′′ commute,
we identify f (A′ )g(A′′) with f ⊗ g and define a coproduct by:
(16) ∆f (A) = f (A′ ⊕ A′′ ).
By construction, this is an algebra morphism from FQSym(ℓ) to FQSym(ℓ) ⊗
FQSym(ℓ) , so that
Theorem 3.3. FQSym(ℓ) is a graded connected bialgebra. Hence, it is a Hopf alge-
bra. The coproduct is given by
X
(17) ∆Gσ,u := G(σ′ ,u′ ) ⊗ G(σ′′ ,u′′ ) .
(σ ′ ,σ ′′ ,u′ ,u′′ )
(σ,u)∈(σ ′ ,u′ )⋒(σ ′′ ,u′′ )

For example,
∆G3142,2412 = 1 ⊗ G3142,2412 + G1,4 ⊗ G231,212 + G12,42 ⊗ G12,21
(18)
+ G312,242 ⊗ G1,1 + G3142,2412 ⊗ 1.
Proof – This is again immediate from the coproduct of the usual free quasi-symmetric
functions Gσ :
X
(19) ∆Gσ = Gσ′ ⊗ Gσ′′ .
σ∈σ′ ⋒ σ′′

3.2. Duality in FQSym(ℓ) .

Definition 3.4. The free ℓ-quasi-ribbon Fσ,u labelled by a colored permutation (σ, u)
is the noncommutative polynomial
(20) Fσ,u := Gσ−1 ,u·σ−1 ,
where the action of a permutation on the right of a word permutes the positions of
the letters of the word.
For example,
(21) F3142,2142 = G2413,1422 .
The product and coproduct of the Fσ,u can be easily described in terms of shifted
shuffle and deconcatenation of colored permutations.
Theorem 3.5. Let σ ′ and σ ′′ be two colored permutations. Then
X
(22) Fσ′ Fσ′′ = Fσ ,
σ∈σ′ ⋒σ′′

and
X
(23) ∆Fσ = FStd(w′ ) ⊗ FStd(w′′ ) .
w′ ,w′′
σ=w′ .w′′
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 7

Proof – Without colors, these formulas are the usual product of coproduct formulas
of the F in FQSym. With colors, one just has to observe that colors follow the letter
to which they are attached.

Note that all colored permutations indexing a product of F have given colors
associated with the same values, which is consistent with the corresponding remark on
the G since places and values are exchanged when taking the inverse of a permutation.
For example, compare the following with Equation (15):
F(21,14) F(12,31) = F(2134,1431) + F(2314,1341) + F(2341,1314)
(24)
+ F(3214,3141) + F(3241,3114) + F(3421,3114) .
Here is an example of coproduct on the F basis:
∆F(23514,14212) = 1 ⊗ F(23514,14212) + F(1,1) ⊗ F(2413,4212) + F(12,14) ⊗ F(312,212)
(25)
+ F(123,142) ⊗ F(12,12) + F(2341,1421) ⊗ F(1,12) + F(23514,14212) ⊗ 1.

Let us define a scalar product on FQSym(ℓ) by

(26) hFσ,u , Gσ′ ,u′ i := δσ,σ′ δu,u′ ,
where δ is the Kronecker symbol.
Theorem 3.6. For any U, V, W ∈ FQSym(ℓ) ,
(27) h∆U, V ⊗ W i = hU, V W i,
so that, FQSym(ℓ) is self-dual: the map Fσ,u 7→ Gσ,u ∗ is an isomorphism from
FQSym(ℓ) to its graded dual.
Proof – Straightforward from Theorem 3.5.

Note 3.7. Let φ be any bijection from C to C, extended to words by concatenation.

Then if one defines the free ℓ-quasi-ribbon as
(28) Fσ,u := Gσ−1 ,φ(u)·σ−1 ,
the previous theorems remain valid since one only permutes the labels of the basis
(Fσ,u ). Moreover, if C has a semigroup structure, the colored permutations (σ, u) ∈
Sn × C n can be interpreted as elements of the semi-direct product Hn := Sn ⋉ C n
with multiplication rule
(29) (σ; c1 , . . . , cn ) · (τ ; d1 , . . . , dn ) := (στ ; cτ (1) d1 , . . . , cτ (n) dn ).
In furthermore C is a group, one can choose φ(γ) := γ −1 and define the scalar product
as before, so that the adjoint basis of the (Gh ) becomes Fh := Gh−1 . In the sequel,
we will be mainly interested in the cases C := Z /ℓ Z, and we will indeed make that
choice for φ whenever C is a group.
8 J.-C. NOVELLI AND J.-Y. THIBON

3.3. Algebraic structure of FQSym(ℓ) . Recall that a permutation σ of size n is

connected [27, 4] if, for any i < n, the set {σ1 , . . . , σi } is different from {1, . . . , i}.
We denote by C the set of connected permutations, and by cn := |Cn | the number
of such permutations in Sn . For later reference, we recall that the generating series
of cn is Sequence A003319 of [41]:
!−1
X X
n n
(30) c(t) := cn t = 1− n!t = t+t2 +3 t3 +13 t4 +71 t5 +461 t6 +O(t7 ) .
n≥1 n≥0

Let the connected colored permutations be the (σ, u) with σ connected and u arbi-
trary. Their generating series is given by c(ℓt).
From [4], we immediately get
Proposition 3.8. FQSym(ℓ) is free over the set Fσ,u (or Gσ,u ), where (σ, u) is
connected.

For example, the generating series of the algebraic generators of FQSym(2) is

(31) 2 t + 4 t2 + 24 t3 + 208 t4 + 2272 t5 + 29504 t6 + 441216 t7 + . . .
3.4. Primitive elements of FQSym(ℓ) . Let L(ℓ) be the primitive Lie algebra of
FQSym(ℓ) . Since ∆ is not cocommutative, FQSym(ℓ) cannot be the universal en-
veloping algebra of L(ℓ) . But since it is cofree, it is, according to [23], the universal
enveloping dipterous algebra of its primitive part L(ℓ) .
Let Gσ,u be the multiplicative basis defined by Gσ,u = Gσ1 ,u1 · · · Gσr ,ur where
(σ, u) = (σ1 , u1 ) • · · ·• (σr , ur ) is the unique maximal factorization of (σ, u) ∈ Sn ×C n
into connected colored permutations.
Proposition 3.9. Let Vσ,u be the adjoint basis of Gσ,u . Then, the family (Vα,u )α∈C
(ℓ)
is a basis of L(ℓ) . In particular, we have dim Ln = ℓn cn . Moreover, L(ℓ) is free.
Proof – The first part of the statement follows from [4]. The second part comes from
the fact that FQSym(ℓ) is bidendriform (Theorem 3.10 below).

For example, since L(ℓ) is free, the generating series by degree of its generators is
(with ℓ = 2):
Y n
1− (1 − tn )ℓ cn = 1 − (1 − t)2 (1 − t2 )4 (1 − t3 )24 · · ·
n≥1
(32)
= 2 t + 3 t2 + 16 t3 + 158 t4 + 1 796 t5
+ 24 250 t6 + 372 656 t7 + . . .
and the Hilbert series of the universal enveloping algebra of FQSym(ℓ) (its domain
of cocommutativity) is, again with ℓ = 2,
Y n
(1 − tn )−ℓ cn = 1 + 2 t + 7 t2 + 36 t3 + 283 t4 + 2 898 t5 + 36 169 t6
(33) n≥1

+ 524 976 t7 + . . .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 9

3.5. Dendriform structure of FQSym(ℓ) . Foissy introduced the notion of biden-

driform bialgebras [6], generalizing the notion of dendrifom algebras (cf. [24]) and
proved some conjectures about FQSym, presented in [4]. We shall adapt this tech-
nology to the colored case. We shall not recall all the theory, since complete details
can be found in [6].
Recall that the generators of FQSym(ℓ) as a dendrifom algebra are called totally
primitive elements [6] and that their generating series is given by
PQ − 1
(34) TP := ,
PQ2
where PQ is the generating series of FQSym(ℓ) .
Theorem 3.10. The algebra FQSym(ℓ) has a structure of bidendriform bialgebra [6],
hence is free as a Hopf algebra and as a dendriform algebra, cofree, self-dual, and its
primitive Lie algebra is free.
Moreover, the totally primitive elements of FQSym(ℓ) are the totally primitive
elements of FQSym with any coloring.
Proof – It has been done by Foissy in [6] in the case of FQSym. But since the
dendriform and codendriform structure do not involve the color alphabet C, the
property is true in this case as well.
Similarly, colors do not play any role in determining if a given element is (totally)
primitive or not.

For example, the dendriform generators of FQSym have as degree generating series
X
(35) dgi ti = t + t3 + 6 t4 + 39 t5 + 284 t6 + 2305 t7 + . . .
i

so that the dendriform generators of FQSym(2) have as degree generating series 2i dgi :
(36) 2 t + 8 t3 + 96 t4 + 1248 t5 + 18176 t6 + 295040 t7 + . . . ,
Note that there cannot be any relation, even dendriform relations, among the
elements F1,c where c ∈ C, so that FQSym(l) contains the free dendriform algebra
PBT(ℓ) on ℓ generators.

3.6. Internal product of FQSym(ℓ) . When C is a semigroup, an internal product

can be defined on FQSym(ℓ) by
(37) Fσ,u ∗ Fτ,v = Fµ,w
where (µ, w) is the product (σ, u) · (τ, v) in the wreath product, defined by formula
(29), that is

(38) F(σ,u) ∗ F(τ,v) = F(στ,(uτ )·v) ,

where uτ is the word uτ1 . . . uτn and u · v is the componentwise product defined by
(u1 v1 , . . . , un vn ).
10 J.-C. NOVELLI AND J.-Y. THIBON

For example, if the color group is Z

(39) F(1324,1011) ∗ F(2413,3200) = F(3412,3311) .

(40) F(165324,102011) ∗ F(625413,322011) = F(462315,423023) .

This can be reduced to any Z /ℓ Z, e.g., with ℓ = 3,
(41) F(165324,102011) ∗ F(625413,022011) = F(462315,120020) .
In the G basis, one has
(42) G(σ,u) ∗ G(τ,v) = G(τ σ,u.(vσ)) ,

4. Noncommutative symmetric functions of level ℓ

4.1. ℓ-partite numbers. Following McMahon [26], we define an ℓ-partite number
n = (n1 , . . . , nℓ ) as a column vector in Nℓ , and a vector composition of n of weight
P
|n| := ℓ1 ni and length m as a ℓ × m matrix I of nonnegative integers, with row
sums vector n and no zero column.
For example,
 
1 0 2 1
(43) I = 0 3 1 1
4 2 1 3
0 1
4
is a vector composition (or a 3-composition, for short) of the 3-partite number @5A
10
of weight 19 and length 4.
For each n ∈ Nℓ of weight |n| = n, we define a level ℓ complete homogeneous
noncommutative symmetric function as
X
(44) Sn := G1···n,u .
u;|u|i =ni

It is the sum of all possible colorings of the identity permutation with ni occurrences
of color i for each i.

4.2. The Hopf algebra Sym(ℓ) . Let Sym(ℓ) be the subalgebra of FQSym(ℓ) gen-
erated by the Sn (with the convention S0 = 1). The Hilbert series of Sym(ℓ) is easily
found to be
X (1 − t)ℓ
(ℓ) n
(45) Sℓ (t) := dim Symn t = .
n
2(1 − t)ℓ − 1
For example, with ℓ = 2, one has
(46) S2 (t) := 1 + 2 t + 7 t2 + 24 t3 + 82 t4 + 280 t5 + 956 t6 + 3264 t7 + . . .
which is Sequence A003480 of [41].
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 11

For general ℓ, it is well-known in the combinatorial folklore (and easy to prove by

means of generating series expansions) that, for all n ≥ 1,
n
X
(47) ncsn (ℓ) := n! dim(Symn(ℓ) ) = S(n, k) pk ℓk
k=1

where S(n, k) is the sequence of absolute values of Stirling numbers of the first kind
(sequence A130534 of [41]) and pk is the sequence of ordered Bell numbers (also
known as packed words or preferential arrangements, Sequence A000670 of [41]).
Theorem 4.1. Sym(ℓ) is free over the set {Sn , |n| > 0}, so that a linear basis is
given by
(48) S I = Si 1 · · · Si m ,
where i1 , · · · , im are the columns of I.
Moreover, Sym(ℓ) is a Hopf subalgebra of FQSym(ℓ) and the coproduct of the
generators is given by
X
(49) ∆Sn = Si ⊗ Sj ,
i+j=n

where the sum i + j is taken in the space Nl . In particular, Sym(ℓ) is cocommutative.

Proof – Consider a linear relation between the S I . Since without colors the algebra is
free (it is Sym), the linear relation splits into many linear relations involving terms
of the form S i1 ...ir with |i1 |, . . . , |ir | fixed. But there are no relations of this form,
thanks to the definition of the Gσ as a sum of colored words.
Given the coproduct of the Gσ of FQSym(ℓ) and since the Sn are sums of G, the
coproduct of an Sn amounts to unshuffling the color words, whence their coproduct
formula.

For example,
0 1 0 1 0 1 0 1 0 1 0 1 0 1
1 1 0 0 1 1 0
@0A @0A @0A @0A @0A @0A @0A
2 2 0 2 0 1 1
∆S = S ⊗S +S ⊗S +S ⊗S
(50) 0 1 0 1 0 1 0 1 0 1 0 1
0 1 1 0 0 1
@0A @0A @0A @0A @0A @0A
1 1 0 2 0 2
+S ⊗S +S ⊗S +S ⊗S .

4.3. Algebraic structure of Sym(ℓ) . The number of generators of Sym(ℓ) of degree

n is given by the number of ℓ-partite numbers of total sum n. So its generating series
is
X ℓ + n − 1
−ℓ
(51) (1 − t) − 1 = tn .
n≥1
n

We shall denote by G(ℓ) the set of nonzero ℓ-partite numbers.

12 J.-C. NOVELLI AND J.-Y. THIBON

4.4. Primitive elements of Sym(ℓ) . Sym(ℓ) being a graded connected cocommu-

tative Hopf algebra, it follows from the Cartier-Milnor-Moore theorem that it is the
universal enveloping algebra of L(ℓ) :
(52) Sym(ℓ) = U(L(ℓ) ),
where L(ℓ) is the Lie algebra of its primitive elements. Let us now prove
Theorem 4.2. As a graded Lie algebra, the primitive Lie algebra L(ℓ) of Sym(ℓ) is
free over a set indexed by G(ℓ).
Proof – If L(ℓ) is free, by standard arguments on generating series, the number of
generators of L(ℓ) in degree n must be the number of algebraic generators of Sym(ℓ)
in degree n, parametrized for example by G(ℓ). We will now show that L(ℓ) has at least
this number of generators and that those generators are algebraically independent,
(ℓ)
determining completely thee dimensions of the homogeneous components Ln of L(ℓ)
whose generating series begins by
     
ℓ+1 2 ℓ+1 ℓ+2
(53) t+ t + + t3 + . . .
2 2 3
Following Reutenauer [39] p. 58, denote by π1 the Eulerian idempotent, that is, the
endomorphism of Sym(ℓ) defined by π1 = log∗ (Id). It is obvious, thanks to the
definition of Sp that
(54) π1 (Sp ) = Sp + · · · ,
where the dots stand for terms SI such that I are vector compositions with strictly
more than one column. Since the Sp where p is a ℓ-partite number are algebraically
(ℓ)
independent, the dimension of Ln is at least equal to the cardinality of the elements
(ℓ)
of G(l) of size n. So L is indeed free over a set of primitive elements parametrized
by G(ℓ).

For example, with ℓ = 2, the generating series of the dimensions of L(ℓ) is

(55) 1 + 2 t + 4 t2 + 12 t3 + 31 t4 + 92 t5 + 256 t6 + 772 t7 + . . .
With ℓ = 3, one finds
(56) 1 + 3 t + 9 t2 + 36 t3 + 132 t4 + 534 t5 + 2140 t6 + 8982 t7 + . . .

4.5. Duality in Sym(ℓ) . Recall that the underlying colored alphabet A can be seen
(i)
as A0 ⊔ · · · ⊔ Aℓ−1 , with Ai = {aj , j ≥ 1}. Let x = (x(0) , . . . , x(ℓ−1) ), where the x(i)
are ℓ commuting variables. In terms of A, the generating function of the complete
functions can be written as
→
!−1
Y X X
(j) (j)
(57) σx (A) = 1− x ai = Sn (A)xn ,
i≥0 0≤j≤ℓ−1 n

where xn = (x(0) )n0 · · · (x(ℓ−1) )nℓ−1 .

 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 13

This realization gives rise to a Cauchy formula (see [20] for the l = 1 case), which
in turn allows one to identify the dual of Sym(ℓ) with an algebra introduced by S.
Poirier in [35]. It is detailed in the following section.
Note that Sym(ℓ) is the natural noncommutative analog of McMahon’s algebra of
multisymmetric functions [26, 10].

5. Quasi-symmetric functions of level ℓ

(i)
5.1. Cauchy formula of level ℓ. Let X = X 0 ⊔· · ·⊔X ℓ−1 , where X i = {xj , j ≥ 1},
be an ℓ-colored alphabet of commutative variables, also commuting with A. Imitating
the level 1 case (see [4]), we define the Cauchy kernel

→
Y
(58) K(X, A) = σ“x(0) ,...,x(ℓ−1) ” (A).
j j
j≥1

Expanding on the basis S I of Sym(ℓ) , we get as coefficients what can be called the
level ℓ monomial quasi-symmetric functions MI (X)

X
(59) K(X, A) = MI (X)S I (A),
I

defined by
X
(60) MI (X) = xij11 · · · xijmm ,
j1 <···<jm

with I = (i1 , . . . , im ).
These functions form a basis of a subalgebra QSym (ℓ) of K[X], which we shall call
the algebra of quasi-symmetric functions of level ℓ.

5.2. Poirier’s Quasi-symmetric functions. The functions MI (X) can be recog-

nized as a basis of one of the algebras introduced by Poirier: the MI coincide with
the M(C,v) defined in [35], p. 324, formula (1), up to indexation.
Following Poirier, we introduce the level ℓ quasi-ribbon functions by summing over
an order on ℓ-compositions: an ℓ-composition C is finer than C ′ , and we write C ≤ C ′ ,
if C ′ can be obtained by repeatedly summing up two consecutive columns of C such
that no nonzero element of the left one is strictly below a nonzero element of the
right one.
This order can be described in a much easier and natural way if one recodes an ℓ-
composition I as a pair of words, the first one d(I) being the set of sums of the elements
of the first k columns of I, the second one c(I) being obtained by concatenating the
words iIi,j while reading of I by columns, from top to bottom and from left to right.
For example, the 3-composition of Equation (43) satisfies
(61) d(I) = {5, 10, 14, 19} and c(I) = 13333 22233 1123 12333 .
14 J.-C. NOVELLI AND J.-Y. THIBON

Moreover, this recoding is a bijection if the two words d(I) and c(I) are such that the
descent set of c(I) is a subset of d(I). The order previously defined on ℓ-compositions
is in this context the inclusion order on sets d: (d′, c) ≤ (d, c) iff d′ ⊆ d.
It allows us to define the level ℓ quasi-ribbon functions FI by
X
(62) FI = MI′ .
I′ ≤I

Notice that this last description of the order ≤ is reminiscent of the order ≤′ on
descent sets used in the context of quasi-symmetric functions and non-commutative
symmetric functions: more precisely, since it does not modify the word c(I), the order
≤ restricted to ℓ-compositions of weight n amounts for ℓn copies of the order ≤′ . The
computation of its Möbius function is therefore straightforward.
Moreover, one can obtain the FI as the commutative image of certain Fσ,u : any
pair (σ, u) such that σ has descent set d(I) and u = c(I) will do.
5.3. Coproducts and alphabets.
5.3.1. Recall that to define the Gσ,u (A) of an ℓ-colored alphabet A = A × C, we
only need a total order on A. Hence, if B is another copy of A commuting with A,
we can define A + B as (A + B) × C where A + B is the ordinal sum, and thus make
sense of Gσ,u (A + B).
As usual, we identify F (A)G(B) with F ⊗ G.
Lemma 5.1. For any F ∈ FQSym(ℓ) , F(A + B) ∈ FQSym(ℓ) ⊗ FQSym(ℓ) , and
(63) F(A + B) = ∆F ,
where ∆ is the coproduct defined by (17).
Proof – It is sufficient to show this for ℓ = 1, which is done in [4].

Let us observe that on this picture, it is clear that the restriction of ∆ to Sym(ℓ)
is dual to the product of QSym(ℓ) . By definition of the Cauchy kernel (58), we have
(64) K(X, A + B) = K(X, A)K(X, B) ,
and by (59), this implies that ∆ is dual of the multiplication of QSym(ℓ) .
5.3.2. The same description can be applied to the quasi-symmetric side. Let X =
X × C and Y = Y × C, where Y is a copy of X. Again, define X + Y as the ordinal
sum of X and Y , and X + Y = (X + Y ) × C.
Lemma 5.2. The map ∇ : F 7→ F (X + Y) defines a coproduct on QSym(ℓ) , which
is dual to the product of Sym(ℓ) .
Proof – This follows from the identity
(65) K(X + Y, A) = K(X, A)K(Y, A) .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 15

5.3.3. The internal coproduct. From now on, we assume that the color set C is an
additive semigroup, such that every element γ ∈ C has a finite number of decompo-
sitions γ = α + β.
We define the C-product T = X ×C Y of two C-colored alphabets by
X
(66) T(γ) = {t(γ)
rs = x(α) (β)
r ys } ,
α+β=γ

with the pairs (r, s) ordered lexicographically.

Proposition 5.3. The map δ : F 7→ F (X ×C Y) defines a coassociative coproduct
on QSym(ℓ) .
We define the internal product ∗ of Sym(ℓ) as the dual product of the map δ.
Proof – The coassociativity condition
(67) (δ ⊗ Id) ◦ δ = (Id ⊗ δ) ◦ δ
translates as the associativity of the C-product
(68) (X ×C Y) ×C Z = X ×C (Y ×C Z)
which is clear since both sides are by definition
( )
X
(69) tµpqr = x(α) (β) (γ)
p yq zr
α+β+γ=µ

with the lexicographic order on triples (p, q, r).

Example 5.4. With ℓ = 2 and C = Z /2 Z,

„ « „ « 20 „ « „ «1 „ «3
1 0 0 2 1 0 0 2
S 1 1 ∗S 1 0 = µ 6BS 1 ⊗ S 1 C ∗ ∆S 1
2 4@ A 2
0 7
5

0 „ « „ «1 0 „ « „ «1 0 „ « „ «1 0 „ « „ «1
1 2 0 0 1 0 1 0 1
B 1 0 CB 1 1 C B 1
= @S ∗S A @S ∗S A + @S ∗S 1 0 C BS 1 ∗ S 0 C
A@ A

„ « „ « „ «
1 1 1 1 0 0 0 0
=S 1 0 +S 0 0 1 +S 1 1 1 .

5.3.4. The splitting formula. The definition of ∗ by duality with the C-product im-
plies that it satisfies the splitting formula. We just need to check a trivial property:
Lemma 5.5. The C-product is distributive over the colored ordinal sum:
(70) (X′ + X′′ ) ×C Y = X′ ×C +X′′ ×C Y .
16 J.-C. NOVELLI AND J.-Y. THIBON

Proposition 5.6. Let µr : (Sym(ℓ) )⊗r → Sym(ℓ) be the product map. Let ∆(r) :
(Sym(ℓ) ) → (Sym(ℓ) )⊗r be the r-fold coproduct, and ∗r be the extension of the internal
product to (Sym(ℓ) )⊗r . Then, for F1 , . . . , Fr , and G ∈ Sym(ℓ) ,

(71) (F1 · · · Fr ) ∗ G = µr [(F1 ⊗ · · · ⊗ Fr ) ∗r ∆(r) G].

Proof – It is enough to consider the case r = 2. Let F ∈ QSym(ℓ) and U, W, W ∈
Sym(ℓ) . We have, writing XY = X ×C Y for short, and assuming duality between X
and A, Y and B and so on,
hF, (UV ) ∗ W i = hF (X′ X′′ ), (UV )(A′ )W (A′′ )i
= hF (X′ + Y′ )X′′ ), U(A′ )V (B ′ )W (A′′)i
= hF (X′ X′′ + Y′ Y′′ ), U(A′ )V (B ′ )W (A′′ + B ′′ )i
= hF (X′ + Y′ ), U(A′ )V (B ′ )W (A′ + B ′ )i
= h∇F, (U ⊗ V ) ∗2 ∆W i = hF, µ[(U ⊗ V ) ∗2 ∆W ]i .

5.3.5. Evaluation of internal products. Let us start with the simplest case, Si ∗ Sj .
The coefficient of S K in this product is the coefficient of Mi ⊗ Mj in δMK , which
is also the coefficient of xi yj in MK (x ×C y). This is zero if K has more than one
column, so that
X
(72) Si ∗ Sj = dnij Sn
n

contains only one-part vector compositions.

Lemma 5.7. If the color semigroup is C = N, then the coefficient dnij of Sn in Si ∗ Sj
is equal to the coefficient of the monomial symmetric function mµ in the product
mα mβ , where µ is the partition (0n0 1n1 . . . ), α = (0i0 1i1 . . . ), β = (0j0 1j1 . . . ), the
monomial functions being taken over an alphabet of n letters, where n = |i|.
Proof – From (72), we only need to compute the coproducts δMn . For this, it is
sufficient to use alphabets of the form x = {x} ⊗ C, y = {y} ⊗ C. Then,
!nk
Y X
(i) (j)
(73) Mn (x ×C y) = x y ,
k≥0 i+j=k

and we see that the coefficient of Mi (x)Mj (y) = xi yj is the same as the coefficient of
hα ⊗ hβ in
!nk
Y X
(74) hi ⊗ hj = ∆ (hn0 0 hn1 1 . . . ) = ∆hµ .
k≥0 i+j=k
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 17

For example,
0 1
„ « „ « „ « 2
2 3 1 @1A
2 1 3 1
(75) S ∗S = 3S +S .
This result is compatible with the fact that
(76) m11 m1 = 3 m111 + m21 .
As another example,
0 1
0 0 1 0 1
B1C 0 0
0 1 0 1 B C
0 1 B1C
B C
B1C
B C
B0C
B C
@2A @1A @0A @0A @2A
1 1 1 2 1
(77) S ∗S =S +2S +2S .
One can then check that the previous result amounts to selecting the partitions of
size at most 3 in
(78) m211 m21 = m421 + 2 m331 + 2 m322 + . . .
Together with the splitting formula (71), this result determines all the products
S I ∗ S J , since one also has
(79) Si ∗ SJ = SJ ∗ Si
thanks to the isomorphism of ordered colored alphabets
(80) x ×C Y ≃ Y ×C x,
where x = {x} ⊗ C.
When the color group is Z /ℓ Z, the result is obtained by reduction modulo ℓ, e.g.,
with l = 3, we get from example (77)
0 1 0 1 0 1 0 1 0 1
0 1 0 2 1
@2A @1A @2A @1A @0A
1 1 1 0 2
(81) S ∗S = 2S +2S +2S .
Note that the coefficient of an S I can change when computing modulo ℓ: for all pairs
′

of parts k and k ′ added together, a factor k+kk
appears.
5.4. Generalized descent algebras. In a preliminary draft of this work [32], we
introduced the internal product in a different way. Assuming that C has a semigroup
structure, we regard colored permutations as elements of the wreath product H =
C ≀ Sn , and for h′ , h′′ ∈ H, we set
(82) Gh′ ∗′ Gh′′ = Gh′′ h′
(opposite law, as in the classical case of Sym).
Theorem 5.8. Let C be a commutative semigroup.
(1) Sym(C) n is a subalgebra of FQSym(C) , for the operation ∗′ defined previously.
(2) The restriction of ∗′ to Sym(C) n satisfies the splitting formula (71).
(3) S i ∗′ S j is given by Lemma 5.7.
(4) S i ∗′ S J = S J ∗′ S i .
18 J.-C. NOVELLI AND J.-Y. THIBON

Proof – The proofs of (1) and (2) can be found in [2]. (3) and (4) follow from the
definition in (44) and the internal product on FQSym(C) given by (38).

This provides an analogue of Solomon’s descent algebra for the wreath product
C ≀ Sn . Note that the definition remains valid for C = Z, so that we get a descent
b n = Z ≀ Sn .
algebra for the (extended) affine Weyl groups of type A, S

5.5. Ordinary multi-symmetric functions. A consequence of the above results is

that the algebra Sym(ℓ) of ordinary (commutative) multi-symmetric functions admits
an internal product. If we denote by F ∈ Sym(ℓ) the commutative image of F ∈
Sym(ℓ) , we have

(83) F ∗ G = F ′ ∗ G′ as soon as F = F ′ and G = G′ .

For example,

0 1 0 1 0 1 0 1
„ « „ « 0 0 2 0 0 2 0 1 2 0 0 2
0 3 1 2 @1 0 1A @0 1 1A @0 0 0A @1 1 0A
1 1 1 1 0 1 0 1 0 0 1 0 1 0 0 1
S ∗S = S +S +S +S
0 1 0 1
0 1 1 0 0 1
@0 0 2A @1 1 2A
1 0 0 0 0 0
+ 2S + 2S
(84) 0 1 0 1 0 1 0 1
1 0 1 1 0 1 1 0 1 1 0 2
@0 1 1A @1 1 0A @1 0 1A @0 0 0A
1 0 0 0 0 1 0 1 0 1 1 0
+S +S +S +S
0 1 0 1 0 1
1 0 0 0 0 1 0 0 2
@1 1 2A @2 1 1A @2 0 0A
0 0 0 0 0 0 0 1 0
+ 2S + 2S + 2S .

0 1 0 1 0 1 0 1
„ « „ « 2 0 0 2 0 0 2 0 1 2 0 0
0 3 2 1 @1 0 1A @1 1 0A @0 0 0A @0 1 1A
1 1 1 1 0 1 0 0 0 1 1 1 0 1 0 0
S ∗S = S +S +S +S
0 1 0 1
1 0 1 1 0 0
@2 0 0A @2 1 1A
0 1 0 0 0 0
+ 2S + 2S
(85) 0 1 0 1 0 1 0 1
0 1 1 0 1 1 0 1 1 0 2 1
@1 0 1A @1 1 0A @0 1 1A @0 0 0A
0 1 0 0 0 1 1 0 0 1 0 1
+S +S +S +S
0 1 0 1 0 1
0 0 1 0 1 0 0 2 0
@1 2 1A @1 1 2A @0 0 2A
0 0 0 0 0 0 1 0 0
+ 2S + 2S + 2S .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 19

0 1 0 1 0 1 0 1
„ « „ « 0 0 2 0 0 2 1 0 2 0 0 2
3 0 1 2 @1 0 1A @0 1 1A @0 0 0A @1 1 0A
1 1 1 1 0 1 0 1 0 0 0 1 1 0 0 1
S ∗S = S +S +S +S
0 1 0 1
1 0 1 0 0 1
@0 0 2A @1 1 2A
0 1 0 0 0 0
+ 2S + 2S
(86) 0 1 0 1 0 1 0 1
1 1 0 1 1 0 1 1 0 1 2 0
@0 1 1A @1 0 1A @1 1 0A @0 0 0A
1 0 0 0 1 0 0 0 1 1 0 1
+S +S +S +S
0 1 0 1 0 1
1 0 0 0 1 0 0 2 0
@1 2 1A @2 1 1A @2 0 0A
0 0 0 0 0 0 0 0 1
+ 2S + 2S + 2S .
0 1 0 1 0 1 0 1
„ « „ « 2 0 0 2 0 0 2 1 0 2 0 0
3 0 2 1 @1 0 1A @1 1 0A @0 0 0A @0 1 1A
1 1 1 1 0 1 0 0 0 1 1 0 1 1 0 0
S ∗S = S +S +S +S
0 1 0 1
1 1 0 1 0 0
@2 0 0A @2 1 1A
0 0 1 0 0 0
+ 2S + 2S
(87) 0 1 0 1 0 1 0 1
1 0 1 1 0 1 1 0 1 2 0 1
@0 1 1A @1 1 0A @1 0 1A @0 0 0A
1 0 0 0 0 1 0 1 0 0 1 1
+S +S +S +S
0 1 0 1 0 1
0 0 1 1 0 0 2 0 0
@2 1 1A @1 1 2A @0 0 2A
0 0 0 0 0 0 0 1 0
+ 2S + 2S + 2S .
If one denotes by h the commutative image of S, one easily checks that

(88) 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
„ « „ « „ « „ « 2 0 0 2 1 0 2 0 0
3 0 2 1 @1A @0A @1A @0A @0A @0A @0A @1A @1A
1 1 1 1 0 1 0 1 0 1 1 0 0
h h ∗h h = 2h h h +h h h +h h h
0 1 0 1 0 1 0 1 0 1 0 1
1 1 0 1 0 0
@2A @0A @0A @2A @1A @1A
0 0 1 0 0 0
+ 2h h h + 2h h h
0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1
1 1 0 1 1 0 2 1 0
@1A @0A @1A @1A @1A @0A @0A @0A @0A
0 1 0 0 0 1 0 1 1
+ 2h h h +h h h +h h h
0 1 0 1 0 1 0 1 0 1 0 1
1 0 0 2 0 0
@1A @2A @1A @0A @2A @0A
0 0 0 0 0 1
+ 4h h h + 2h h h .

6. The Mantaci-Reutenauer algebra

6.1. Monochromatic complete functions. Let ei be the canonical basis of Nℓ .
For n ≥ 1, let
(89) Sn(i) = Sn·ei ∈ Sym(ℓ) ,
20 J.-C. NOVELLI AND J.-Y. THIBON

be the monochromatic complete symmetric functions.

(i)
Proposition 6.1. The Sn generate a Hopf subalgebra MR(ℓ) of Sym(ℓ) , which is
(ℓ)
isomorphic to the Mantaci-Reutenauer descent algebra of the wreath products Sn =
(Z /ℓ Z) ≀ Sn if C = Z /ℓ Z.
Proof – MR(ℓ) is obviously stable by the product and coproduct coming from Sym(ℓ) ,
hence is a Hopf subalgebra of Sym(ℓ) .
As a Hopf algebra, it is clearly isomorphic to the Mantaci-Reutenauer algebra,
having the same number of independent generators in each degree, and behaving in
the same way (divided powers) under the coproduct. The isomorphism for the internal
product comes from the splitting formula, which gives explicitly S (I,u) ∗ S (J,v) .

Since MR(ℓ) has ℓ generators in each degree, its dimensions are given by
1 X
(90) P = 1 + ℓ (ℓ + 1)n−1 tn .
1 − k≥1 ℓtk n≥1

The bases of MR(ℓ) are labelled by colored compositions, that is, pairs formed by a
composition and a color vector of the same length:
(91) (I, u) = ((i1 , . . . , im ), (u1, . . . , um )).

6.2. Primitive elements of MR(ℓ) . MR(ℓ) being a subalgebra of a graded connected

cocommutative Hopf algebra, is itself a graded connected cocommutative Hopf alge-
bra, so that, thanks to the Cartier-Milnor-Moore theorem, it is the universal envelop-
ing algebra of L(ℓ) :

(92) MR(ℓ) = U(L(ℓ) ),

where L(ℓ) is the Lie algebra of its primitive elements. The following property is
obtained in the same way as Theorem 4.2.
Theorem 6.2. As a graded Lie algebra, the primitive Lie algebra L(ℓ) of MR(ℓ) is
free over a set indexed by colored compositions.
For example, with ℓ = 2, the generating series of the dimensions of L(ℓ) is
(93) 1 + 2 t + 3 t2 + 8 t3 + 18 t4 + 48 t5 + 116 t6 + 312 t7 + . . .
With ℓ = 3, one finds
(94) 1 + 3 t + 6 t2 + 20 t3 + 60 t4 + 204 t5 + 670 t6 + 2340 t7 + . . .
(ℓ)
More generally, the dimension of Ln is given by the Witt polynomials

ℓP if n = 1,
(95) qn (ℓ) := 1 n/d
n d | n µ(d)(l + 1) if n > 1,
(ℓ)
so that, for n ≥ 2, the dimension of Ln coincide with those of a free Lie algebra on
ℓ + 1 generators of degree 1.
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 21

6.3. Duality. The duality is easily worked out by means of the appropriate Cauchy
kernel. The generating function of the complete functions is
ℓ−1 X
X
(96) σxMR (A) := 1 + Sn(j) · (x(j) )n ,
j=0 n≥1

and the Cauchy kernel is as usual

→
Y X
MR
(97) K (X, A) = σxMR
i
(A) = M(I,u) (X)S (I,u) (A),
i≥1 (I,u)

where (I, u) runs over colored compositions (I, u) = ((i1 , . . . , im ), (u1, . . . , um )). The
MI,u are called the monochromatic monomial quasi-symmetric functions and satisfy
X (u ) (u )
(98) M(I,u) (X) = (xj1 1 )i1 · · · (xjmm )im .
j1 <···<jm

Proposition 6.3. The M(I,u) span a subalgebra of C[X] which can be identified with
the graded dual of MR(ℓ) through the pairing

(99) hM(I,u) , S (J,v) i = δI,J δu,v ,

where δ is the Kronecker symbol.
Note that this algebra can also be obtained by imposing the relations
(p) (q)
(100) xi xi = 0, for p 6= q

on the variables of QSym (l) .

Baumann and Hohlweg [2] have another construction of the dual of MR(ℓ) (implic-
itly defined in [35], Lemma 11).

7. Level ℓ parking quasi-symmetric functions

7.1. Usual parking functions.

7.1.1. The combinatorial objects. Recall that a parking function on [n] = {1, 2, . . . , n}
is a word a = a1 a2 · · · an of length n on [n] whose nondecreasing rearrangement
a↑ = a′1 a′2 · · · a′n satisfies a′i ≤ i for all i. Let PFn be the set of such words. It is
well-known that |PFn | = (n + 1)n−1 .
One says that a has a breakpoint at b if |{ai ≤ b}| = b. The set of all breakpoints of
a is denoted by BP (a). Then, a ∈ PFn is said to be prime if BP (a) = {n} (see [42]).
Let PPFn ⊂ PFn be the set of prime parking functions on [n]. It can easily be
shown that |PPFn | = (n − 1)n−1 (see [43]).
Finally, one says that a has a match at b if |{ai < b}| = b − 1 and |{ai ≤ b}| ≥ b.
The set of all matches of a is denoted by Ma(a).
22 J.-C. NOVELLI AND J.-Y. THIBON

7.1.2. Algebraic structure on parking functions. The algebra PQSym of parking

functions [29, 31] is very similar to the algebra FQSym of permutations.
Since parking functions are closed under the shifted shuffle, one defines a product
on the vector space with basis (Fa ) by
X
(101) Fa′ Fa′′ = Fa .
a∈a′ a′′ [k]

The coproduct on this basis is given by the parkization algorithm [29]: for w =
w1 w2 · · · wn on {1, 2, . . .}, let us define
(102) d(w) := min{i|#{wj ≤ i} < i} .
If d(w) = n + 1, then w is a parking function and the algorithm terminates, return-
ing w. Otherwise, let w ′ be the word obtained by decrementing all the elements of
w greater than d(w). Then Park(w) := Park(w ′ ). Since w ′ is smaller than w in the
lexicographic order, the algorithm terminates and always returns a parking function.
For example, let w = (3, 5, 1, 1, 11, 8, 8, 2). Then d(w) = 6 and the word w ′ =
(3, 5, 1, 1, 10, 7, 7, 2). Then d(w ′) = 6 and w ′′ = (3, 5, 1, 1, 9, 6, 6, 2). Finally, d(w ′′ ) =
8 and w ′′′ = (3, 5, 1, 1, 8, 6, 6, 2), that is a parking function. Thus, Park(w) =
(3, 5, 1, 1, 8, 6, 6, 2).
We then have
X
(103) ∆Fa = FPark(u) ⊗ FPark(v) .
u,v
u.v=a

7.1.3. Duality. Let Ga = F∗a ∈ PQSym∗ be the dual basis of (Fa ). If h , i denotes
the duality bracket, the product on PQSym∗ is given by
X X
(104) Ga′ Ga′′ = h Ga′ ⊗ Ga′′ , ∆Fa i Ga = Ga ,
a a∈a′ ∗a′′

where the convolution a′ ∗a′′ of two parking functions is defined as

X
(105) a′ ∗a′′ = a.
u,v;a=u·v,
Park(u)=a′ ,Park(v)=a′′

By duality, one easily gets the formula for the coproduct of Ga as

X
(106) ∆Ga := Gu ⊗ Gv .
u,v;a∈u⋒v

7.2. Colored parking functions. Let ℓ be an integer, representing the number of

allowed colors. A colored parking function of level ℓ and size n is a pair composed of
a parking function of length n and a word on [ℓ] of length n.
Since there is no restriction on the coloring, it is obvious that there are ℓn (n+1)n−1
colored parking functions of level ℓ and size n.
With two colors, one finds the sequence ai = 2(2i+2)i−1 , known as A097629 in [41]:
(107) 1 + 2 t + 12 t2 + 128 t3 + 2000 t4 + 41472 t5 + 1075648 t6 + . . .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 23

Since the convolution of two parking functions contains only parking functions, one
easily builds as in [29] an algebra PQSym(ℓ) on colored parking functions:

X
(108) G(a′ ,u′ ) G(a′′ ,u′′ ) = G(a,u′ ·u′′ ) .
a∈a′ ∗a′′

We can define a coproduct using sums of alphabets: again, we only need a total
order on A to define the colored parkization, so that taking two isomorphic copies A′
and A′′ of A, we define A′ ⊕ A′′ as (A′ ⊕ A′′ ) × C, where A′ ⊕ A′′ denotes the ordered
sum. Assuming furthermore that A′ and A′′ commute, we identify f (A′)g(A′′ ) with
f ⊗ g and define a coproduct by:
(109) ∆f (A) = f (A′ ⊕ A′′ ).
By construction, this is an algebra morphism from PQSym(ℓ) to PQSym(ℓ) ⊗
PQSym(ℓ) , so that
Theorem 7.1. PQSym(ℓ) is a graded connected bialgebra, hence a Hopf algebra.
More precisely, the coproduct can be computed in the following way:
X
(110) ∆G(a,u) = G(a′ ,u′ ) ⊗ G(a′′ ,u′′ ) .
(a′ ,a′′ ,u′ ,u′′ )
(a,u)∈(a′ ,u′ )⋒(a′′ ,u′′ )

Proof – Straightforward from the previous definitions.

For example,
(111) ∆G(41142,22115) = 1 ⊗ G(41142,22115) + G(112,215) ⊗ G(11,21) + G(41142,22115) ⊗ 1.
∗
7.3. Duality. Let F(a,u) = G∗(a,u) ∈ PQSym(ℓ) be the dual basis of (Ga ). If h , i
∗
denotes the duality bracket, the product on PQSym(ℓ) is given by
X
(112) F(a′ ,u′ ) F(a′′ ,u′′ ) = F(a,u) ,
(a,u)∈(a′ ,u′ )⋒(a′′ ,u′′ )

where the shifted shuffle of two colored parking functions is such that colors follow
their letters.
Using the duality bracket once more, one easily gets the formula for the coproduct
of F(a,u) as
X
(113) ∆F(a,u) = F(Park(p′ ),u′ ) ⊗ F(Park(p′′ ),u′′ ) .
(p′ ,u′ ),(p′′ ,u′′ )
p′ p′′ =a;u′ u′′ =u

7.4. Algebraic structure of PQSym(ℓ) . Recall that a word w over N∗ is connected

if it cannot be written as a shifted concatenation w = u • v, and anti-connected if its
mirror image w is connected.
24 J.-C. NOVELLI AND J.-Y. THIBON

We denote by CP the set of connected parking functions, and by pn := |CP n | the

number of such parking functions of size n. For later reference, we recall that the
generating series of pn is Sequence A122708 of [41]:
!−1
X X
p(t) := pn tn = 1 − (n + 1)n−1tn
n≥1 n≥0

= t + 2 t2 + 11 t3 + 92 t4 + 1 014 t5 + 13 795 t6 + 223 061 t7 + . . .

Let the connected colored parking functions be the (a, u) with a connected and u
arbitrary. Their generating series is given by p(ℓt).
From [31], we immediately get
Proposition 7.2. PQSym(ℓ) is free over the set Fσ,u (or Gσ,u ), where (σ, u) is
connected.
For example, one gets the following generating series for the algebraic generators
(connected parking functions with ℓ = 2):
(114) 2 t + 8 t2 + 88 t3 + 1 472 t4 + 32 448 t5 + 882 880 t6 + 28 551 808 t7 . . .
7.5. Primitive elements of PQSym(ℓ) . Let L′ (ℓ) be the primitive Lie algebra of
PQSym(ℓ) . Since ∆ is not cocommutative, PQSym(ℓ) cannot be the universal en-
veloping algebra of L′ (ℓ) . But since it is cofree, it is, according to [23], the universal
enveloping dipterous algebra of its primitive part L′ (ℓ) .
Let Ga,u be the multiplicative basis defined by Ga,u = Ga1 ,u1 · · · Gar ,ur where
(a, u) = (a1 , u1 ) • · · ·• (ar , ur ) is the unique maximal factorization of (a, u) ∈ Sn ×C n
into connected colored parking functions.
Proposition 7.3. Let Va,u be the adjoint basis of Ga,u . Then, the family (Vα,u )α∈CP
is a basis of L′ (ℓ) . In particular, we have dim L′ (ℓ) n
n = ℓ pn . Moreover, L
′ (ℓ)
is free.
Proof – The first part of the statement follows from [4]. The second part comes from
the fact that PQSym(ℓ) is bidendriform (Theorem 7.4 below).

For example, since L′ (ℓ) is free, the generating series of the degree of its generators
is (with ℓ = 2):
Y n
1− (1 − tn )ℓ pn = 1 − (1 − t)2 (1 − t2 )8 (1 − t3 )88 · · ·
n≥1
(115)
= 2 t + 7 t2 + 72 t3 + 1 276 t4 + 28 944 t5 + 805 288 t6
+ 26 462 232 t7 + . . .
and the Hilbert series of the universal enveloping algebra of PQSym(ℓ) (its domain
of cocommutativity) is, again with ℓ = 2:
Y n
(1 − tn )−ℓ cn = 1 + 2 t + 11 t2 + 108 t3 + 1 713 t4 + 36 470 t5 + 969 919 t6
(116) n≥1
+ 30 847 464 t7 + . . .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 25

7.6. Bidendriform and tridendriform structure.

Theorem 7.4. The algebra PQSym(ℓ) has a structure of bidendriform bialgebra,
hence is free as a Hopf algebra and as a dendriform algebra, cofree, self-dual, and its
primitive Lie algebra is free.
Moreover, the totally primitive elements of PQSym(ℓ) are the totally primitive
elements of PQSym with any coloring.
Proof – It has been done in [31] in the case of FQSym. But since the dendriform
and codendriform structure do not involve the color alphabet C, the property is true
in this case as well.
Also, colors do not play any role in determining whether a given element is (totally)
primitive or not, so the last statement holds.

For example, the dendriform generators of PQSym have as degree generating

series
X
(117) dgpi ti = t + t2 + 7 t3 + 66 t4 + 786 t5 + 11 378 t6 + 189 391 t7 + . . .
i

so that the dendriform generators of PQSym(2) have as degree generating series

2i dgpi :
(118) 2 t + 4 t2 + 56 t3 + 1 056 t4 + 25 152 t5 + 721 792 t6 + 24 242 048 t7 + . . .

7.6.1. Tridendriform structure.

Conjecture 7.5. As in the case of PQSym, we conjecture that PQSym(ℓ) is a free
dendriform trialgebra.
Note that there cannot be any relations, even tridendriform relations, among the
elements F1,c where c ∈ C, so that PQSym(l) contains the free tridendriform algebra
TD (ℓ) on ℓ generators.
Recall that, if PQSym(ℓ) is free as a tridendriform algebra, its generators have as
generating series
PQ − 1
(119) TD := ,
2PQ2 − PQ
where PQ is the generating series of PQSym(ℓ) .
For example, the tridendriform generators of PQSym have as degree generating
series
X
(120) tgpi ti = t + 5 t3 + 50 t4 + 634 t5 + 9 475 t6 + 163 843 t7 + . . .
i

so that the tridendriform generators of PQSym(2) have as degree generating series

2i tgpi :
(121) 2 t + 40 t3 + 800 t4 + 20 288 t5 + 606 400 t6 + 20 971 904 t7 + . . .
26 J.-C. NOVELLI AND J.-Y. THIBON

8. Type B algebras
8.1. Parking functions of type B. In [38], Reiner defined non-crossing partitions
of type B by analogy to the classical case. In our context, he defined the level 2
case. This allowed him to derive, by analogy with a simple representation theoretical
result, a definition of parking functions of type B as the words on [n] of size n.
We shall build another set of words, also enumerated by nn that sheds light on the
relation between type-A and type-B parking functions and provides a natural Hopf
algebra structure on the latter.
First, fix two colors 0 < 1. We say that a pair of words (a, u) composed of a
parking function and a binary colored word is a level 2 parking function if
• the only elements of a that can have color 1 are the matches of a.
• for all element of a of color 1, all letters equal to it and to its left also have
color 1,
• all elements of a have at least once the color 0.
For example, there are 27 level 2 parking functions of size 3: there are the 16 usual
ones all with full color 0, and the eleven new elements
(111, 100), (111, 110), (112, 100), (121, 100), (211, 010),
(122)
(113, 100), (131, 100), (311, 010), (122, 010), (212, 100), (221, 100).
The first time the first rule applies is with n = 4, where one has to discard the
words (1122, 0010) and (1122, 1010) since 2 is not a match of 1122. On the other
hand, both words (1133, 0010) and (1133, 1010) are B4 -parking functions since 1 and
3 are matches of 1133.
Theorem 8.1. The restriction of PQSym(2) to the F elements indexed by level 2
parking functions is a subalgebra of PQSym(2) . The restriction of PQSym(2) to the
G elements indexed by level 2 parking functions is a subcoalgebra of PQSym(2) .
Proof – The shifted shuffle of two F elements indexed by level 2 parking functions
only consists in level 2 parking functions since the definition only involves matches
(preserved by shifted shuffle) and positions of colors 0 and 1 on a given letter inside
a word, also preserved by shifted shuffle. The same property holds for the coproduct
on the G side: a match on either side of the tensor product comes from a match of
the original word and all equal letters go to the same side of the tensor product in
the same order.

8.2. Non-crossing partitions of type B. Remark that in the level 1 case, non-
crossing partitions are in bijection with non-decreasing parking functions. To extend
this correspondence to type B, let us start with a non-decreasing parking function b
(with no color). We factor it into the maximal shifted concatenation of prime non-
decreasing parking functions,

and we choose a color, here 0 or 1, for each factor. We
2n
obtain in this way n words π, which can be identified with type B non-crossing
partitions.
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 27

Let
X
(123) Pπ = Fa
a↑ =π

where w ↑ denotes the nondecreasing rearrangement of the letters of w. Then,

Theorem 8.2. The Pπ , where π runs over the above set of non-decreasing signed
parking functions, form the basis of a cocommutative Hopf subalgebra NCPQSym(2)
of PQSym(2) .
Proof – The subalgebra part comes from the fact that the shifted shuffle does not mix
prime factors. The coalgebra part selects pieces of each factor, hence satisfies that
each letter of a (new) factor has identical color. The cocommutativity part comes
from the fact that all rearrangements of a given word are considered at the same
time.
All this can be extended to higher levels in a straightforward way: allow each prime
non-decreasing parking function to choose any color among ℓ and use the factorization
as above. Since non-decreasing parking functions are in bijection with Dyck words,
the choice can be described as: each block of a Dyck word with no return-to-zero,
chooses one color among ℓ. In this version, the generating series is obviously given
by
1
(124) √
1− 1−4t
.
1−ℓ 2
For ℓ = 3, we obtain Sequence A007854 of [41].

9. Colored analogs of planar binary trees: PBT(ℓ)

9.1. Definition of PBT(ℓ) . In the case with one color, the Hopf algebra PBT of
Planar binary trees initially defined by Loday and Ronco [22] can be embedded in
FQSym in the following way [12, 13]:
X
(125) PT = Fσ ,
σ;shape (P(σ))=T

where P is a simple algorithm: it is the well-known binary search tree insertion, such
as presented, for example, by Knuth in [18].
In the colored case, the definition is almost the same:
X
(126) PT,u = F(σ,u) ,
(σ,u);shape (P(σ))=T

where u is a color word whose length is equal to the number of leaves of T . Note
that this algebra (without its realization on words) has been previously studied by
Maria Ronco [40].
Given the definitions, the generating series of the dimensions of PBT(ℓ) is
(127) 1 + ℓ t + 2ℓ2 t2 + 5ℓ3 t3 + 14ℓ4 t4 + . . .
that is, the generating series of Catalan numbers multiplied by ℓn .
28 J.-C. NOVELLI AND J.-Y. THIBON

9.2. Algebraic structure of PBT(ℓ) . Since PBT is generated by the trees with
no right branch (starting from the root), the same holds in PBT(ℓ) :

Proposition 9.1. PBT(ℓ) is free over the set PT,u , where T is a tree with no right
branch.
For example, the generating series of the algebraic generators of PBT(ℓ) is
(128) ℓ t + ℓ2 t2 + 2ℓ3 t3 + 5ℓ4 t4 + 14ℓ5 t5 + . . .
that is, the generating series of shifted Catalan numbers Cn−1 multiplied by ℓn .

9.2.1. Primitive elements of PBT(ℓ) . Let L(ℓ) be the primitive Lie algebra of the
algebra PBT(ℓ) . Since ∆ is not cocommutative, PBT(ℓ) cannot be the universal
enveloping algebra of L(ℓ) . But since it is cofree, it is, according to [23], the universal
enveloping dipterous algebra of its primitive part L(ℓ) .
Using the same arguments as in the case of FQSym(ℓ) , one then proves
Proposition 9.2. The Lie algebra L(ℓ) is free. Moreover
(129) dim L(ℓ) n
n = ℓ Cn−1 ,

For example, since L(ℓ) is free, the generating series of the degree of its generators
is (with ℓ = 2):
Y n
1− (1 − tn )ℓ Cn−1 = 1 − (1 − t)2 (1 − t2 )4 (1 − t3 )16 (1 − t4 )80 · · ·
n≥1
(130)
= 2 t + 3 t2 + 8 t3 + 46 t4 + 252 t5
+ 1 558 t6 + 9 800 t7 + . . .

and the Hilbert series of the universal enveloping algebra of PBT(ℓ) (its domain of
cocommutativity) is, again with ℓ = 2,
Y n
(1 − tn )−ℓ Cn−1 = 1 + 2 t + 7 t2 + 28 t3 + 139 t4 + 762 t5 + 4 549 t6
(131) n≥1

+ 28 464 t7 + . . .

9.2.2. Dendriform structure of PBT(ℓ) . Recall that PBT is the free dendriform al-
gebra on one generator. Since colors do not play any role in determining if a given
element is (totally) primitive or not, the same holds for PBT(ℓ) :

Proposition 9.3. The algebra PBT(ℓ) is the free dendriform algebra on ℓ generators.
It has also the structure of bidendriform bialgebra.
Note 9.4. It is also possible to define a colored analog of CQSym the Catalan
Quasi-symmetric algebra, but the natural definition leads to a non-cocommutative
algebra, hence not sharing the basic property of CQSym itself.
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 29

10. Examples
10.1. Multigraded coinvariants and colored Klyachko idempotents. One of
the very first applications of the theory of noncommutative symmetric functions was
to provide an explanation for the following coincidence. On the one hand, consider
the representation of Sn in the coinvariant algebra
(132) Hn = C[x1 , . . . , xn ]Sn = C[x1 , . . . , xn ]/J ,
where J is the ideal generated by symmetric polynomials without constant term. It
is known [25] that the graded Frobenius characteristic of the action of Sn on Hn is
X
chq (Hn ) = q k ch(Hn )(k)
k
(133)   X
X
= (q)n hn = q maj(I) rI (X),
1−q In

where rI (X) are the ribbon Schur functions.

On the other hand, Klyachko [17] introduced a remarkable Lie idempotent in C Sn
X
(134) κn = ζ maj(σ) σ
σ∈Sn

where ζ is a primitive n-th root of unity.

In terms of noncommutative symmetric functions, both expressions can be inter-
preted as specializations of
  X
A
(135) Kn (q) := (q)n Sn = q maj(I) RI .
1−q In

This is the graded noncommutative characteristic of an action of Hn (0) on coin-

variants. This is a projective module, hence also an Sn -module, and taking the
commutative image, (A = X), one obtains the characteristic of Sn . But Kn (q) can
also be interpreted as an element of the descent algebra of Sn . A simple computation
shows that for q = ζ, a primitive n-th root of unity, Kn (ζ) is a primitive element
of Sym of commutative image pn /n, hence is a Lie idempotent (see [20]). Actually,
Kn (q) is the noncommutative Hall-Littlewood function H̃1n (A; q), and this special-
ization property is a special case of a noncommutative version [11] of the classical
property [11, 21, 25].
There is a similar phenomenon here. Let q1 , . . . , qn be independent variables, and
consider the noncommutative symmetric function
X
(136) Kn (A; q1 , . . . , qn ) := qMAJ(I) RI ,
In

where
(137) qMAJ(i1 ,...,ir ) := (q1 . . . qi1 )r (qi1 +1 . . . qi1 +i2 )r−1 . . . (qi1 +···+ir−2 . . . qi1 +···+ir−1 ).
For example,
(138) K3 = R3 + q1 q2 R21 + q1 R12 + q12 q2 R111 .
30 J.-C. NOVELLI AND J.-Y. THIBON

K4 = R4 + q1 q2 q3 R31 + q1 q2 R22 + q12 q22 q3 R211

(139)
+ q1 R13 + q12 q2 q3 R121 + q12 q2 R112 + q13 q22 q3 R1111 .
Its commutative image is the multigraded characteristic of Hn with respect to the
partition degree (cf. [3]).
One may also regard Kn as an element of
(140) Sym(Z) (Z)
n ⊂ FQSymn

by means of the identification

X
(141) RI (A)q1α1 . . . qnαn = Gσ,α1 ...αn
Des(σ)=I

and writing it as
X
(142) Kn (A; q1 , . . . , qn ) = RI (A)qMAJ(I)
In

Note that this element lives in the (descent) algebra of the extended affine Weyl
group of type A
(143) b n = Zn ⋉Sn = P ⋉ Sn ,
S
where P is the weight lattice. One can also interpret it as an element of the usual
affine Weyl group
(144) e n = Q ⋉ Sn
S
where Q is the root lattice
n
X
(145) Q = {α ∈ P | αi = 0}.
i=1

This amounts to imposing the relation

(146) q1 . . . qn = 1,
which replaces the root of unity condition q n = 1 in the one-parameter case.
It has been proved by McNamara and Reutenauer [28] that under condition (146),
e n . Within the formalism of the present
Kn (A; q1 , . . . , qn ) was a Lie idempotent in C S
paper, this can be seen as follows: the authors of [28] introduce a twisted product on
An = C(x1 , . . . , xn )[Sn ] by the formula
(147) f (x)σ · g(x)τ = f (x)σ[g(x)]στ ,
where permutations act on functions as automorphisms, i.e., σ(xi ) = xσ(i) , and in
−1
particular, on monomials by σ[xc ] = xcσ . Hence,
(148) (σxc ) · (τ xd ) = στ xcτ +d ,
which is the same as Formula (29) so that An can be identified with the homogeneous
component of degree n of FQSym(Z) , by setting
(149) σxc ≡ Gσ xc ≡ Gσ,c .
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 31

McNamara and Reutenauer then introduce the formal series in FQSym(Z)

Q
XX xσ(1) · · · xσ(j)
(150) Θ(x) = Q j∈Des(σ) σ
n≥0 σ∈S i=1n (1 − xσ(1) · · · xσ(i) )
n

which, applying (148), and under the identification (149), can be rewritten as
Y← X ←
Y
(151) Θ(x) = Gidn ,l =
n σ1 (A(l) ) .
l≥0 n≥0 l≥0

Indeed, introducing the new variables

(152) yj = x1 x2 · · · xj ,
and applying (148) we can write
Q
X X d∈Des(σ) yd
Θ(x) = Gσ
n≥0 σ∈Sn
(1 − y1 )(1 − y2 ) · · · (1 − yn )
(153) X 1
= Kn (y1 , . . . , yn−1 ) ,
n≥0
((y))n
where ((y))n = (1 − y1 )(1 − y2 ) · · · (1 − yn ) and
X Y
(154) Kn (y1 , . . . , yn−1) = Gσ yd
σ∈Sn d∈Des(σ)

is the twisted version of the multiparameter Klyachko element introduced in [20,

(11)]. By Moebius inversion over the lattice of compositions of n, we have
X Y
(155) Kn (y1 , . . . , yn−1) = RI · yd
In d∈Des(I)

1 X 1 Y yd
(156) Kn (y1 , . . . , yn−1 ) = SJ · ,
((y))n Jn
1 − yn 1 − yd
d∈Des(J)

which implies the expression

←
!
Y X
(157) Θ(x) = Sn ynl .
l≥0 n≥0

Each factor of the right-hand side is grouplike (for the coproduct of FQSym(l) ):
X X X X X
(158) ∆ Sn ynl = Si yil ⊗ Sj yjl = Si y i ⊗ Sj yjl
n≥0 n≥0 i+j=n i≥0 j≥0

so that also
(159) ∆Θ(x) = Θ(x) ⊗ Θ(x) .
Extracting the term of degree n, we find
X  1 1

(160) ∆Kn = Ki ⊗ Kj ((y))n
i+j=n
((y)) i ((y)) j
32 J.-C. NOVELLI AND J.-Y. THIBON

so that if we send yn to 1, all terms vanish except the extreme ones, and we get a
primitive element. This is the main result of [28].
10.2. A formula of Raney. Raney [37] gave a combinatorial interpretation of the
coefficients of the unique solution g(t) ∈ Q[Y, Z][[t]] of the functional equation
ℓ
X
(161) g(t) = t yk ezk g(t) ,
k=1
P gn n+1
with g(t) = n≥0 n! t . This defining equation is of the form
(162) g(t) = tφ(g(t)),
Pℓ
with φ(u) = k=1 yk ezk u . Hence, the coefficient gn of tn+1 in g(t) is
(163)
1
gn = [un ]φ(u)n+1)
n+1   
1 X n+1 n
= y1n1 . . . yℓnℓ (n1 z1 )q1 . . . (nℓ zℓ )qℓ .
n + 1 n +···+n =n+1 n1 , . . . , nℓ q1 , . . . , qℓ
1 ℓ
q1 +···+qℓ =n

Thus, gn ∈ N[Y, Z]. Its combinatorial interpretation can be mechanically derived

by means of a colored version of the noncommutative Lagrange inversion formula as
formulated in [33, 8]. Consider the functional equation
X
(164) g= bk Sn(k) g n ,
k,n
(k)
where Sn = Sn (A(k) ) is a colored complete function and bk are noncommuting letters.
We can set
X
(165) Sn = bk Sn(k) ,
k

so that (164) can be rewritten as

X
(166) g= Sn g n ,
n≥0

and the solution of [33] reads

g = S 0 + S 10 + (S 200 + S 110 ) + . . .
X
(167) = S Ev(π).0 ,
π∈NDPF
P
where NDPF is the set of nondecreasing parking functions. Note that S 0 = bk is a
priori different from 1, and does not commute with the other S i . Each term S Ev(π).0
represents an ordered tree T in Polish notation, so that, for example
◦@
~~ @
(168) ◦ ◦
◦
 FREE QUASI-SYMMETRIC FUNCTIONS FOR WREATH PRODUCTS 33

is S 2010 . P (k)
Replacing each S i by ℓk=1 bk Si , the expression S Ev(π).0 becomes a sum over all ℓ-
colorings of the tree T , so that, one recovers the combinatorial interpretation of Raney
(proved in a different way by Foata [5]): let n = (n1 , . . . , nℓ ) and q = (q1 , . . . , qℓ ) ∈ Nℓ
and let B(n, q) be the set of ℓ-colored trees on n = |n| vertices with nk vertices of
color k and such that the sum of the arities of vertices of color k is qk . Then
X 1 X
(169) g= |B(n, q)|yn zq .
n≥0
n!

References
[1] P. Biane, Parking functions of types A and B, Electronic J. Combin. 9 (2002), # 7.
[2] P. Baumann and C. Hohlweg, A Solomon-type epimorphism for Mantaci-Reutenauer’s al-
gebra of a wreath product G ≀ Sn , Trans. Amer. Math. Soc. 360 (2008), 1475–1538.
[3] F. Caselli, Diagonal invariants and the refined multimahonian distribution, preprint,
arXiv:math.CO/0805.2860.
[4] G. Duchamp, F. Hivert, and J.-Y. Thibon, Noncommutative symmetric functions VI: free
quasi-symmetric functions and related algebras, Internat. J. Alg. Comput. 12 (2002), 671–717.
[5] D. Foata, La série génératrice exponentielle dans les problèmes d’énumération, Montréal,
Presses de l’Univ. de Montréal, 1974.
[6] L. Foissy, Bidendriform bialgebras, trees, and free quasi-symmetric functions, J. Pure Appl.
Algebra 209 (2007), no. 2, 439–459.
[7] I.M. Gelfand, D. Krob, A. Lascoux, B. Leclerc, V. S. Retakh, and J.-Y. Thibon,
Noncommutative symmetric functions, Adv. in Math. 112 (1995), 218–348.
[8] I. Gessel, Noncommutative Generalization and q-analog of the Lagrange Inversion Formula,
Trans. Amer. Math. Soc. 257 (1980), no. 2, 455–482.
[9] I. Gessel, Multipartite P-partitions and inner products of skew Schur functions, [in “Combi-
natorics and algebra”, C. Greene, Ed.], Contemporary Mathematics, 34 (1984), 289–301.
[10] I. Gessel, Enumerative Applications of Symmetric Functions, Sém. Lothar. Combin., 17a
(1987), 17p. (electronic).
[11] F. Hivert, Hecke algebras, difference operators, and quasi-symmetric functions, Adv. Math.,
155 (2000), no. 2, 181–238.
[12] F. Hivert, J.-C. Novelli, and J.-Y. Thibon, Un analogue du monoı̈de plaxique pour les
arbres binaires de recherche, C. R. Acad. Sci. Paris Sér I Math. 332 (2002), 577–580.
[13] F. Hivert, J.-C. Novelli, and J.-Y. Thibon, The algebra of binary search trees, Theo.
Comp. Sci. 339 (1) (2005), 129–165.
[14] F. Hivert, J.-C. Novelli, and J.-Y. Thibon, Yang-Baxter bases of 0-Hecke algebras and
representation theory of 0-Ariki-Koike-Shoji algebras, Adv. in Math. 205 (2006), 504–548.
[15] F. Hivert and N. Thiéry, MuPAD-Combinat, an open-source package for research in alge-
braic combinatorics, Sém. Lothar. Combin. 51 (2004), 70p. (electronic).
[16] V. G. Kac, Infinite-dimensional Lie algebras, Cambridge Univ. Press, 1991.
[17] A.A. Klyachko, Lie elements in the tensor algebra, Siberian Math. J. 15 (1974), 1296–1304.
[18] D. E. Knuth, The art of computer programming, vol.3: Sorting and searching, (Addison-
Wesley, 1973).
[19] A. G. Konheim and B. Weiss, An ocupancy discipline and applications, SIAM J. Applied
Math. 14 (1966), 1266–1274.
[20] D. Krob, B. Leclerc, and J.-Y. Thibon, Noncommutative symmetric functions II: Trans-
formations of alphabets, Internal J. Alg. Comput. 7 (1997), 181–264.
[21] A. Lascoux, B. Leclerc and J.-Y. Thibon, Green polynomials and Hall-Littlewood func-
tions at roots of unity, Europ. Jour. Comb. 15 (1994), 173–180.
34 J.-C. NOVELLI AND J.-Y. THIBON

[22] J.-L. Loday and M.O. Ronco, Hopf algebra of the planar binary trees, Adv. Math. 139
(1998) n. 2, 293–309.
[23] J.-L. Loday and M.O. Ronco, Algèbres de Hopf colibres, C. R. Acad. Sci. Paris Sér I Math.
337 (2003), 153–158.
[24] J.-L. Loday, Scindement d’associativité et algèbres de Hopf, Actes des Journées Mathé-
matiques à la Mémoire de Jean Leray, Sémin. Congr. Soc. Math. France 9 (2004), 155–172.
[25] I.G. Macdonald, Symmetric functions and Hall polynomials, 2nd ed., Oxford University
Press, 1995.
[26] P.A. McMahon, Combinatory analysis, Cambridge University Press, 1915, 1916; Chelsea
reprint, 1960.
[27] R. Mantaci and C. Reutenauer, A generalization of Solomon’s descent algebra for hyper-
octahedral groups and wreath products, Comm. Algebra 23 (1995), 27–56.
[28] P. McNamara and C. Reutenauer, P-Partitions and a Multi-Parameter Klyachko Idempo-
tent, Elec. Jour. Comb. 11(2) (2005), #R21.
[29] J.-C. Novelli and J.-Y. Thibon, A Hopf Algebra of Parking Functions, Proc. FPSAC’04
Conf., Vancouver.
[30] J.-C. Novelli and J.-Y. Thibon, Construction of dendriform trialgebras, C.-R. Acad. Sci.,
Paris Sr. I 342 (2006), 365–369.
[31] J.-C. Novelli and J.-Y. Thibon, Hopf algebras and dendriform structures arising from park-
ing functions, Funda. Math. 193 (2007), 189–241.
[32] J.-C. Novelli and J.-Y. Thibon, Free quasi-symmetric functions of arbitrary level, preprint,
math.CO/0405597.
[33] J.-C. Novelli and J.-Y. Thibon, Noncommutative Symmetric Functions and Lagrange In-
version, Intern. Jour. Alg. and Comput, to appear.
[34] F. Patras, L’algèbre des descentes d’une bigèbre graduée, J. Algebra 170 ( 1994), 547–566.
[35] S. Poirier, Cycle type and descent set in wreath products, Disc. Math., 180 (1998), 315–343.
[36] S. Poirier and C. Reutenauer, Algèbre de Hopf des tableaux, Ann. Sci. Math. Québec 19
(1995), 79–90. P∞
[37] G. Raney, A formal solution of i=1 Ai eBi X = X, Can. J. Math. 16 (1964), 755–762.
[38] V. Reiner, Non-crossing partitions for classical reflection groups, Preprint available at
ftp://s6.math.umn.edu/pub/papers/reiner/
[39] C. Reutenauer, Free Lie algebras, Oxford University Press, 1993.
[40] M.O. Ronco, Primitive elements in a free dendriform Hopf algebra, Contemp. Maths. 267
(2000) 245–264.
[41] N.J.A. Sloane, The On-Line Encyclopedia of Integer Sequences,
http://www.research.att.com/~njas/sequences/
[42] R. P. Stanley, Parking functions and noncrossing partitions, Electronic J. Combin. 4 (1997),
# 2.
[43] R. P. Stanley, Enumerative combinatorics, vol. 2, Cambridge University Press, 1999.

Université Paris-Est, Institut Gaspard Monge, 5 Boulevard Descartes, Champs-

sur-Marne, 77454 Marne-la-Vallée cedex 2, FRANCE
E-mail address, Jean-Christophe Novelli: [email protected]
E-mail address, Jean-Yves Thibon: [email protected]