ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH

DERIVATION
arXiv:0906.4730v2 [math.RA] 22 Feb 2010

JEAN-LOUIS LODAY

Dedicated to Tornike Kadeishvili in honor of his sixtieth birthday

Abstract. We study the operad of associative algebras equipped with a

derivation. We show that it is determined by polynomials in several vari-
ables and substitution. Replacing polynomials by rational functions gives an
operad which is isomorphic to the operad of “moulds”. It provides an efficient
environment for doing integro-differential calculus. Interesting variations are
obtained by using formal group laws. The preceding case corresponds to the
additive formal group law. We unravel the notion of homotopy associative
algebra with derivation in the spirit of Kadeishvili’s work.

Introduction
To any one-dimensional formal group law we show how to associate a type of
algebras, that is an operad, whose space of n-ary operations is the algebra of ratio-
nal functions in n variables. If the formal group law is polynomial, then the spaces
of polynomials in n variables form a suboperad. We show that, for the additive
formal group law, the polynomial operad is simply the operad AsDer of associative
algebras equipped with a derivation, and the rational functions operad RatF ct is
isomorphic to the operads of “moulds”. Since the operad RatF ct contains both
the derivation operator and its inverse, that is the integral operator, the whole
integro-differential calculus can be written in terms of operadic calculus. The op-
erad RatF ct is related to the dendriform and tridendriform operads, and several
others.
In order to deal with formal group laws in noncommutative variables, one has to
modify the notion of operad and work with “pre-shuffle algebras” introduced by M.
Ronco in [17]. The idea is essentially to forget parallel composition in the definition
of an operad by means of the partial compositions.
In the appendix we show that the operad AsDer, which is quadratic, is a Koszul
operad. We describe explicitly the notion of “homotopy associative algebra with
derivation”, that is AsDer∞ -algebra.

Notation. In this paper K is a commutative unital algebra and all modules over K
are supposed to be free. Its unit is denoted by 1. Since most of the time we think
of K as being a field we will say vector space or space for a free module over K.
The tensor product of vector spaces over K is denoted by ⊗. The tensor product

2000 Mathematics Subject Classification. 17A30, 18D50, 18G55, 16S99.

Key words and phrases. Derivation, operad, dendriform, tridendriform, formal group law, Ito
integral, mould.
1
2 J.-L. LODAY

of n copies of the space V is denoted by V ⊗n . The symmetric group Sn is acting

on V ⊗n by place-permutation.
The polynomial algebra, resp. the algebra of rational functions, on the set of
variables {x1 , . . . , xn } is denoted by K[x1 , . . . , xn ], resp. K(x1 , . . . , xn ). The unit is
denoted by 1n or 1 if there is no confusion.

1. Prerequisites on operads
This is a very brief introduction on algebraic operads, whose purpose is essen-
tially to set up notations. One can consult for instance [14], or [13], for more
details.

1.1. Nonsymmetric operad. A nonsymmetric operad, or ns operad for short, is

a graded vector space {Pn }n≥0 equipped with a particular element of P1 denoted
id and called the identity operation, and a family of linear maps

γi1 ,...,ik : Pk ⊗ Pi1 ⊗ · · · ⊗ Pik → Pn , n = i1 + · · · + ik ,

which satisfy the following associative

L and unital properties. On the functor P :
⊗n
Vect → Vect defined as P(V ) := n Pn ⊗ V the operations γi1 ,...,ik induce a
transformation of functors γ : P ◦ P → P. Then, γ is supposed to be associative.
The element id can be interpreted as a morphism from the identity functor I to P
that we denote by ι : I → P. Then, ι is supposed to be a unit for γ.
There is an alternative definition of a ns operad which uses the so-called partial
compositions. For any i ≤ m the partial composition

◦i : Pm ⊗ Pn → Pm+n−1

evaluated on (µ, ν) is the operation γ1,...,1,n,1,...,1 evaluated on (µ; id, . . . id, ν, id, . . . , id).
Pictorially it is represented by the following grafting of trees, where the root of ν
is grafted onto the ith leaf of µ:

PPPGGG
PPPG wwnwnwnnn
NNNJJ νn z t
NNJNJJ ztzttt
NNJNJ z
NNJNJJ i ztztztt
NNJ tzzt
µ

A ns operad can be defined as a graded vector space equipped with an identity

operation and partial compositions which satisfy the unital axioms and the following
two axioms (which replace associativity). For any λ ∈ Pl , µ ∈ Pm and ν ∈ Pn :

Axiom I (sequential composition):

(λ ◦i µ) ◦i−1+j ν = λ ◦i (µ ◦j ν), 1 ≤ i ≤ l, 1 ≤ j ≤ m.
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 3

It corresponds to the two possibilities of composing in the following diagram:

QQQHHH v mm
QQQH vmvmvmm
MMMII ν
MMIMII {{ uu
MMIMII {u{uuu
MMIMI j {{uu {
MIMI {u{u{uu
µ
OOOKK
OOKOKK xx ss
OOKOK i xxsxsss
OKOOKK xs
OOK sxxsxss
λ

Axiom II (parallel composition):

(λ ◦i µ) ◦k+m−1 ν = (λ ◦k ν) ◦i µ, 1 ≤ i < k ≤ l.

It corresponds to the two possibilities of composing in the following diagram:

LL QQQEEE EEE
LL QQ ykykykkkk yyyxxx
LL µ
LL GG · · · ··· ν xxx
LL G }x
LL GG }}xx
LL GGi
LL GG k }x}x}xx
LL G }
LLGG
LLGG }
x}x}xx
L }x}x
λ

1.2. Symmetric operads. A symmetric operad, or operad for short, is a family of

right Sn -modules {P(n)}n≥0 equipped with transformation of functors γ : P ◦ P →
P and ι : I → P which are associative and unital. Here P stands for the so-called
Schur functor defined as
M
P(V ) := P(n) ⊗Sn V ⊗n .
n

Any operad and any ns operad gives rise to a notion of algebras over this operad.
There is a definition of operad using the partial compositions. See loc. cit. for
details.

2. Associative algebras with a derivation

2.1. Definition. Let A be nonunital associative algebra over K. A derivation of
A is a linear map DA : A → A which satisfies the Leibniz relation
DA (ab) = DA (a)b + aDA (b)
′
for any a, b ∈ A. Let (A′ , DA ) be another associative algebra with derivation. A
′
morphism f : (A, DA ) → (A′ , DA ) is a linear map f : A → A′ which is a morphism
of associative algebras and which commutes with the derivations:
′
f ◦ DA = DA ◦ f.
4 J.-L. LODAY

The operad governing the category of associative algebras with derivation admits
the following presentation. There are two generating operations, one of arity 1, that
we denote by D, and one of arity 2 that we denote by µ. The relations are:

µ ◦ (µ, id) = µ ◦ (id, µ),
D ◦ µ = µ ◦ (D, id) + µ ◦ (id, D).
They account for the associativity of the product and for the Leibniz relation. Since,
in the relations, the variables stay in the same order, the category of associative
algebras with derivation can be encoded by a nonsymmetric operad that we denote
by AsDer. So it is determined by a certain family of vector spaces AsDern , n ≥ 1,
and composition maps
γi1 ,...,ik : AsDerk ⊗ AsDeri1 ⊗ · · · ⊗ AsDerik → AsDern
where n = i1 + · · · + ik .
2.2. Theorem. As a vector space AsDern is isomorphic to the space of polynomials
in n variables:
AsDern = K[x1 , . . . , xn ].
The composition map γ = γi1 ,...,ik is given by
γ(P ; Q1 , . . . , Qk )(x1 , . . . , xn ) =
P (x1 + · · · + xi1 , xi1 +1 + · · · + xi1 +i2 , xi1 +i2 +1 + · · · , . . .)Q1 (x1 , . . . , xi1 )Q2 (xi1 +1 , . . .) · · · .
Under this identification the operations id, D, µ correspond to 11 , x1 ∈ K[x1 ] and to
12 ∈ K[x1 , x2 ] respectively. More generally the operation
(a1 , . . . , an ) 7→ Dj1 (a1 )Dj2 (a2 ) · · · Djn (an )
corresponds to the monomial xj11 xj22 · · · xjnn .
Graphically the operation xj11 xj22 · · · xjnn is pictured as a planar decorated tree as
follows:

···

Dj1 Dj2 ··· Djn

RRR
RRR
EE ··· xx
RRR EEEE xxx
RRR EE x
RRR EE xx
RRR xx
µn

Proof. Since µ is an associative operation the space of n-ary operations generated

by µ is one-dimensional. Let us denote by µn the composition of n − 1 copies of µ.
Using the Leibniz relation we see that any composite of copies of µ and D can be
written uniquely as composites of copies of D first and then composites of copies
of µ. Hence any n-ary operation is a linear combination of operations of the form
µn ◦ (Dj1 , . . . , Djn )
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 5

for some sequence of nonnegative integers {j1 , . . . , jn }. We denote this operation

by xj11 xj22 · · · xjnn . We obtain: AsDern = K[x1 , . . . , xn ].
In order to prove the formula for the composition map γ, it is sufficient to prove
that
(P ◦i Q)(x1 , . . . , xn+m−1 ) =
P (x1 , . . . , xi−1 , xi +· · · xi+m−1 , xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ),
where − ◦i − is the ith partial composition, P ∈ AsDern , Q ∈ AsDerm . It is
sufficient to prove this formula when P and Q are monomials:

Dk1 H · · · Dkm
HHH uu
uu
µm

Di1 TTTT· · · Dji · · · jj Djn

TTTT j jjjj
TT jjjj
µn

By direct inspection we see that it is sufficient to treat the case Q = µm , and

in fact the case P = Dℓ , Q = µm . In other words we need to compute the element
Dℓ (a1 · · · am ). By the Leibniz relation, for ℓ = 1, this is exactly the action of the
operation x1 + · · · + xm . Recursively we get that the operation (a1 , . . . , am ) 7→
Dℓ (a1 · · · am ) is (x1 + · · · + xm )ℓ as expected. So we are done. 
2.3. Lemma. The two binary operations a ⊢ b and a ⊣ b of AsDer2 given by x1
and x2 respectively satisfy the following relations
(a ⊢ b) ⊣ c = a ⊢ (b ⊣ c),
(a ⊣ b) ⊢ c + (a ⊣ b) ⊣ c = a ⊣ (b ⊢ c) + a ⊢ (b ⊢ c).
Proof. By a
straightforward operadic calculus we get
x1 ◦1 x1 = (x1 + x2 )x1 , x1 ◦1 x2 = (x1 + x2 )x2 , x2 ◦1 x1 = x3 x1 , x2 ◦1 x2 = x3 x2 ,

x1 ◦2 x1 = x1 x2 , x1 ◦2 x2 = x1 x3 , x2 ◦2 x1 = (x2 + x3 )x2 , x2 ◦2 x2 = (x2 + x3 )x3 .

It follows immediately that
x2 ◦1 x1 = x1 ◦2 x2 ,
x1 ◦1 x2 + x2 ◦1 x2 = x1 ◦2 x1 + x2 ◦2 x1 .
If we put a ⊢ b := x1 (a, b) and a ⊣ b := x2 (a, b), then we get the expected formulas.


2.4. Remarks. The relations of Lemma 2.3 give rise to a new type of ns operad
generated by two operations. Other examples include magmatic, dendriform [9],
cubical [11], duplicial [10], compatible-two-associative [4].
Since AsDer is a ns operad, it is completely determined by its free algebra over
one generator. This free algebra has also been computed in [7] by Guo and Keigher.
6 J.-L. LODAY

2.5. An elementary example. Let K = k[y] be the polynomial algebra in one

variable over the field k. We consider the associative algebra with derivation
∂
(A, DA ) = (k[y][x], ∂x ). In AsDer(k[y])∧ := Πn AsDern we consider the opera-
tion
X yn
exp(yD) := ( Dn ).
n!
n≥0
Then for any polynomial p(x) ∈ k[x] the following formula holds in k[y][x] = k[x, y]:
∂
exp(y
)(p(x)) = p(x + y).
∂x
This is a key formula for studying vertex algebras, see for example [16].

3. First variation and dendriform algebras

Since the composition γ in the operad AsDer needs only the sum, the prod-
uct and the substitution of variables to be defined, it can be extended to many
generalizations of the polynomial algebras: C ∞ -functions, rational functions, etc,
provided that they are commutative. The rational function case is interesting since
it permits us to treat integration, represented by the rational function x111 , which is
the inverse of the operation x1 coding for derivation.

3.1. The rational functions operad. We define a new ns operad RatF ct by

RatF ctn := K(x1 , . . . , xn )

(rational functions in the variables x1 , . . . , xn ). The composition γ is given by the

formula of Theorem 2.2, or, equivalently by the partial composition formula:
(P ◦i Q)(x1 , . . . , xn+m−1 ) :=
P (x1 , . . . , xi−1 , xi +· · ·+xi+m−1 , xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).

3.2. Proposition. The partial compositions − ◦i − as defined above make RatF ct

into a ns operad.

Proof. There are two axioms to check, cf. 1.1. The first one follows from the fact
that addition of variables is a formal group law, that is F (x, y) := x + y satisfies
F (F (x, y), z) = F (x, F (y, z)). The second axiom follows from the fact that the
algebra of rational functions is commutative. 
We observe that the structure of associative algebra of RatF ct1 is precisely the
algebra structure of the rational functions in one variable K(x1 ).

3.3. Integro-differential calculus. In the operad RatF ct the derivation opera-

tion D, represented by x1 ∈ K(x1 ) = RatF ct1 , admits an inverse for composition,
R
that is 1/x1 , which is the integration operation . So we can write the integro-
differential calculus within the operad RatF ct. Here is an example.
For integrable functions f and g on R define
Z x Z x

(f ≺ g)(x) := f (x) g(t)dt and (f ≻ g)(x) := f (t)dt g(x).
0 0
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 7

Then, it is shown in [9] that these two operations satisfy the dendriform axioms
(see below) as a consequence of integration by parts:
Z x Z x Z x Z t  Z x Z t 
g(t)dt h(t)dt = g(t) h(s)ds dt + g(s)ds h(t)dt .
0 0 0 0 0 0

We should be able to recover this property by computing in the operad RatF ct

since f ≻ g = (1/x1 )(f, g) and f ≺ g = (1/x2 )(f, g), for 1/x1 , 1/x2 ∈ RatF ct2 .
This is the object of Proposition 3.5.
More analogous formulas can be found in [3].

3.4. Dendriform algebras. Let us recall from [8, 9] that a dendriform algebra is
a vector space equipped with two binary operations a ≺ b and a ≻ b satisfying the
relations

 (a ≺ b) ≺ c = a ≺ (b ≺ c + b ≻ c),
(a ≻ b) ≺ c = a ≻ (b ≺ c),

(a ≺ b + a ≻ b) ≻ c = a ≻ (b ≻ c).
It is proved in loc. cit. that the ns operad Dend, which encodes the dendriform
algebras, is spanned in arity n by the planar binary trees with n + 1 leaves, a set
L
that we denote by P BTn+1 . Let Dend(K) = n K[P BTn+1 ] be the free dendriform
algebra on one generator Y. The following relation holds in Dend(K):

s∨t=s≻Y≺t ,

where s ∨ t stands for the grafting of the two binary trees s and t.

3.5. Proposition. There is an inclusion ϕ of the ns operad Dend to the ns operad

RatF ct of rational functions induced by
 1
≻ 7→ x1 ,
1
≺ 7→ x2 .

It sends a planar binary tree t to a rational function ϕ(t) according to the following
rules: ϕ(|) = 1 and

ϕ(s) ϕ(t)
ϕ(s ∨ t) = ∈ RatF ctp+q+1 ,
(x1 + · · · + xp ) (xp+2 + · · · + xp+q+1 )

when s, resp. t, is a tree with p + 1 leaves, resp. q + 1 leaves.

Proof. The fact that the two operations satisfy the dendriform axioms is a conse-
quence of a straightforward operadic calculus:
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
x1 ◦1 x1 = x1 +x2 x1 , x1 ◦1 x2 = x1 +x2 x2 , x2 ◦1 x1 = x3 x1 , x2 ◦1 x2 = x3 x2 ,

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
x1 ◦2 x1 = x1 x2 , x1 ◦2 x2 = x1 x3 , x2 ◦2 x1 = x2 +x3 x2 , x2 ◦2 x2 = x2 +x3 x3 .
8 J.-L. LODAY

It follows immediately that

1 1 1 1 1 1 1 1 1 1 1 1
x1 ◦1 x1 + x1 ◦1 x2 = x1 +x2 x1 + x1 +x2 x2 = x1 x2 = x1 ◦2 x1 ,

1 1 1 1 1 1
x2 ◦1 x1 = x3 x1 = x1 ◦2 x2 ,

1 1 1 1 1 1 1 1 1 1 1 1
x2 ◦1 x2 = x3 x2 == x2 +x3 x2 + x2 +x3 x3 = x2 ◦2 x1 + x2 ◦2 x2 .

1 1
If we put a ≻ b := x1 (a, b) and a ≺ b := x2 (a, b), then we get the expected formulas.
As for the second assertion, we first prove the formula for the grafting of trees.
The proof is a straightforward operadic calculus, whose steps are the following.
In Dend(K) we have s ∨ t = s ≻ Y ≺ t = (s ≻ Y) ≺ t. We first compute
s ≻ Y = γ(≻; s, Y) = (≻ ◦1 s) ◦p+1 Y. Applying ϕ we get
1 
2
ϕ(s ≻ Y) = ◦1 s ◦p+1 Y
x1
1
= ϕ(s)(x1 , . . . , xp ) ◦n Y
x1 + · · · + xp
1
= ϕ(s)(x1 , . . . , xp ).
x1 + · · · + xp

A similar computation for the left product leads to the expected formula.
The rational functions ϕ(t) for t ∈ P BTn are linearly independent (proof by
induction). Hence ϕ : Dendn = K[P BTn+1 ] → RatF ctn is injective for all n.


3.6. Examples. Here is the image of the pb trees under ϕ in low dimension:

1 1
x1 x2

• /•

1 1
x1 x1 +x2

•F
ooo FFF
1 1 ooo
w FF
x1 +x2 x2 • FF
FF
F"
1 1
• x1 x3
xx
 x
1 1 xx
x2 x2 +x3 • OOO xxx
OOO x x
' |x
•
1 1
x2 +x3 x3
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 9

1
3.7. Associativity of ∗. Let us adopt the notation Φ(u, v) = u + v1 . For any
variables x1 , x2 , x3 the formula

Φ(x1 , x2 )Φ(x1 + x2 , x3 ) = Φ(x2 , x3 )Φ(x1 , x2 + x3 )

is immediate to satisfy (cf. the proof of Proposition 3.5):

1 1 1 1 x1 + x2 + x3 1 1 1 1
( + )( + )= = ( + )( + ).
x1 x2 x1 + x2 x3 x1 x2 x3 x2 x3 x1 x2 + x3
Viewed as an equality in the space RatF ct3 of ternary operations, it simply says
that the operation ∗ =≻ + ≺ is associative. Indeed the left part of the equality is
Φ ◦1 Φ and the right part is Φ ◦2 Φ.
This formula is reminiscent of the cocycle condition of a 3-cochain φ in group
cohomology:
φ(x, y) + φ(xy, z) = x φ(y, z) + φ(x, yz),
and to the construction of the McLane invariant of a crossed module.

3.8. Comparison with the operad of moulds. The work of J. Ecalle led F.
Chapoton to introduce the operad of moulds in [2, 3]. It is a nonsymmetric operad
denoted by M ould, which is determined by

M ouldn := K(x1 , . . . , xn ),

and by the partial compositions given by the formula

(P ◦i Q)(x1 , . . . , xn+m−1 ) :=
(xi +· · ·+xi+m−1 )P (x1 , . . . , xi−1 , xi +· · ·+xi+m−1 , xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).

3.9. Proposition. There is an isomorphism of ns operads RatF ract ∼

= M ould.

Proof. By direct inspection we verify that the map RatF ractn → M ouldn given by

P (x1 , . . . , xn ) 7→ (x1 + · · · + xn )P (x1 , . . . , xn )

is compatible with the operadic compositions. Since we are working with rational
functions, the element x1 + · · · + xn is invertible and so this map is an isomorphism.


3.10. Remarks. (a) Under the isomorphism of Proposition 3.9, Proposition 3.5
can be found in [2].
(b) Observe that the polynomials do not form a suboperad of M ould.
(c) If we think of the operation 12 ∈ K(x1 , x2 ) = RatF ct2 as a third binary
1 1
operation, then it is easy to check that the three operations ≻= x1 , ≺= x2 , · = 12
satisfy the 7 axioms of a graded tridendriform algebra (cf. [1, 12]). We come back
to this point in the next section.
10 J.-L. LODAY

4. Second variation of AsDer

In probability theory there is a variation of the integration by parts called the

Ito integral. Its counterpart in the derivation framework, that could be called Ito
derivation, is a linear map DA : A → A which satisfies the following relation:

DA (ab) = DA (a)b + aDA (b) + DA (a)DA (b).

One can treat both the derivation case, the Ito derivation case (and even more)
by introducing a parameter λ ∈ K as follows (we could also work with a formal
parameter q, that is take K = k[q]).

4.1. Parametrized derivation. By definition a λ-derivation is a linear map DA :

A → A, where A is an associative algebra, such that

DA (ab) = DA (a)b + aDA (b) + λDA (a)DA (b)

for any a, b ∈ A. For λ = 0 we get the derivation, for λ = 1 we get the Ito derivation.
By convention λ = ∞ stands for the case where DA is an algebra homomorphim:
DA (ab) = DA (a)DA (b).
We denote by λ-AsDer the operad of associative algebras equipped with a λ-
derivation.
We introduce the notation

θλ (x1 , . . . , xn ) := (x1 + · · · + xn ) + · · · + λk−1 θk (x1 , . . . , xn ) + · · · + λn−1 (x1 . . . xn )

where θk (x1 , . . . , xn ) is the kth symmetric function of the variables x1 , . . . , xn .

4.2. Theorem. As a vector space λ-AsDern is isomorphic to the space of polyno-

mials in n variables:
λ-AsDern = K[x1 , . . . , xn ].

The composition map γ is given by

γ(P ; Q1 , . . . , Qk )(x1 , . . . , xn ) =
P (θλ (x1 , . . . , xi1 ), θλ (xi1 +1 , . . . , xi1 +i2 ), · · · )Q1 (x1 , . . . xi1 )Q2 (xi1 +i2 , . . .) · · · .

Under this identification the operations id, D, µ correspond to 11 , x1 ∈ K[x1 ] and to

12 ∈ K[x1 , x2 ] respectively. More generally the operation

(a1 , . . . , an ) 7→ Dj1 (a1 )Dj2 (a2 ) · · · Djn (an )

corresponds to the monomial xj11 xj22 · · · xjnn .

Proof. The proof is the same as in the case λ = 0 performed in the first section.
See also the proof of Proposition 4.8. 
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 11

4.3. Remark. The associativity property of the composition in the operad λ-AsDer
implies that

θλ (x1 , . . . , xi , θλ (xi+1 , . . . , xi+j ), xi+j+1 , . . . , xn ) = θλ (x1 , . . . , xn ).

This formula can also be proved directly by observing that

1 + λθλ (x1 , . . . , xm ) = Πm
i=1 (1 + λxi ).

4.4. The parametrized operad λ-RatF ct. As in section 2 we can put an operad
structure on the rational functions by using the formulas of Theorem 4.2. It gives
a new operad, denoted λ-RatF ct, for which

λ-RatF ctn = K(x1 , . . . , xn ),

and the partial composition is given by

(P ◦i Q)(x1 , . . . , xn+m−1 ) :=
P (x1 , . . . , xi−1 , θλ (xi , · · · , xi+m−1 ), xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).

4.5. λ-TriDendriform algebras. In [12] we introduced the notion of tridendri-

form algebra, which is an algebra with 3 binary operations satisfying 7 relations
(one for each of the cells of a triangle). The graded version was studied by Chapo-
ton in [1]. There exists a parametrized version which handles both versions (and
more) as follows. By definition a λ-tridendriform algebra has 3 binary operations
denoted by a ≺ b, a ≻ b, a · b and 7 relations (one for each cell of the triangle):

 (x ≺ y) ≺ z = x ≺ (y ∗ z) ,
(x ≻ y) ≺ z = x ≻ (y ≺ z) ,

(x ∗ y) ≻ z = x ≻ (y ≻ z) ,

 (x ≻ y) · z = x ≻ (y · z) ,
(x ≺ y) · z = x · (y ≻ z) ,

(x · y) ≺ z = x · (y ≺ z) ,

(x · y) · z = x · (y · z) ,
where x ∗ y := x ≺ y + x ≻ y + λ x · y.
For λ = 1 we get Loday-Ronco’s tridendriform algebra [12], for λ = 0 we get
Chapoton’s graded tridendriform algebra [1].

4.6. Proposition. In the operad λ-RatF ct the binary operations

12 12
≻:= , ≺:= , and · := 12
x1 x2
satisfy the λ-tridendriform axioms.

Proof. The proof is an easy operadic computation analogous to the one performed
in the proof of Proposition 3.5. We leave it to the reader. 
12 J.-L. LODAY

4.7. Formal group laws. Let F (x, y) be a formal group law. This is a power
series in the commutative variables x and y with coefficients in K which satisfies
the relations

F (x, 0) = x = F (0, x) , F (F (x, y), z) = F (x, F (y, z)).

For instance, for any λ ∈ K, θλ (x, y) := x + y + λxy is a polynomial formal group

law. For λ = 0 it is the additive formal group law, for λ = 1 it is the multiplicative
formal group law. Here is another example which is not polynomial. Let ǫ ad δ be
elements in the commutative ring K and let Q(x) := 1 − 2δx2 + ǫx4 . We suppose
p
that 2 is invertible in K so that Q(x) exists as a formal power series in x. The
theory of elliptic curves gives rise to the Jacobi formal group law defined by
p p
x Q(y) + y Q(x)
F (x, y) = .
1 − ǫx2 y 2
Let F be any formal group law. By induction we define

F (x1 , . . . , xn ) := F (F (x1 , . . . , xn−1 ), xn ).

4.8. Proposition. For any formal group law F there is a well-defined nonsymmetric
operad RatF ctF given by:

RatF ctF
n := K(x1 , . . . , xn ),

(P ◦i Q)(x1 , . . . , xn+m−1 ) :=
P (x1 , . . . , xi−1 , F (xi , · · · , xi+m−1 ), xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).

Proof. Axiom I is an immediate consequence of the equality

F (x1 , . . . , xi , F (xi+1 , . . . , xi+j ), xi+j+1 , . . . , xn ) = F (x1 , . . . , xn ),

which follows from the associativity of F . Axiom II is a consequence of the com-

mutativity of the algebra of rational functions. 

5. P-algebras with derivation

Let P be an algebraic operad, whose space of n-ary operations is denoted by

P(n). Recall that P(n) is a right Sn -module. A derivation on a P-algebra A is
a linear map DA : A → A such that, for any operation µ ∈ P(n) the following
formula holds:
n
X
DA (µ(a1 , . . . , an )) = µ(a1 , . . . , DA (ai ), . . . , an ).
i=1

A similar computation as in the previous sections shows that the operad governing
P-algebras with derivation, denoted by PDer, is such that

PDer(n) = K[x1 , . . . , xn ] ⊗ P(n)

 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 13

the action of the symmetric group Sn being the diagonal action (recall that Sn is
acting on K[x1 , . . . , xn ] by permuting the variables). The composition γ is obtained
by combining the composition in P and the formula in Theorem 2.2.
For instance ComDer(n) = K[x1 , . . . , xn ] is an Sn -module and the composition
map γ of the operad ComDer is given by the same formula as for the operad
AsDer. It means that these formulas are compatible with the symmetric group
action. Another way of phrasing this result is the following. Consider the forgetful
functor which associates to a symmetric operad P the nonsymmetric operad P e such
^ = AsDer.
that Pen = P(n). Then we have ComDer
If we start with a ns operad P, then PDer is a ns operad, where PDern =
K[x1 , . . . , xn ] ⊗ Pn . In terms of standard constructions in the operad framework, it
is the Hadamard product of P with AsDer.

6. The “pre-shuffle algebra” of associative algebras with a

derivation

If we replace the space of polynomials (or formal power series) by the space of
noncommutative polynomials (or noncommutative formal power series), then we do
not get an operad anymore. However there is a variation of the notion of operads
which permits us to provide a similar treatment in this noncommutative framework,
it is the notion of “pre-shuffle algebra” due to M. Ronco [17] .

6.1. Pre-shuffle algebra [17]. A pre-shuffle algebra is a family of vector spaces

Pn equipped with composition maps

•i : Pm ⊗ Pn → Pm+n−1 , 1 ≤ i ≤ m,

defined for n ≥ 1, m ≥ 1, which satisfy the following relations:

(λ •i µ) •i−1+j ν = λ •i (µ •j ν) ,

for 1 ≤ i ≤ l, 1 ≤ j ≤ m. In other words, the difference with ns operads is that we

keep only axiom I and we disregard axiom II (see 1.1), therefore any algebraic ns
operad is a pre-shuffle algebra.
We denote by P erAsDer the pre-shuffle algebra which is generated by a unary
operation D and a binary operation µ, which satisfy the following relations:


 µ •1 µ = µ •2 µ ,

D •1 µ = µ •1 D + µ •2 D ,

 (α •i D) •j µ = (α •j µ) •i D ,

(α •i µ) •j+1 D = (α •j D) •i µ ,

for any operation α and i < j.

14 J.-L. LODAY

Observe that the first relation is the associativity of µ, the second relation is
saying that D is a derivation, the third and fourth relations say that the operations
D and µ commute for parallel composition.

6.2. Theorem. As a vector space P erAsDern is isomorphic to the space of non-

commutative polynomials in n variables:
P erAsDern = Khx1 , . . . , xn i.
The composition map •i is given by
(P •i Q)(x1 , . . . , xn+m−1 ) =
P (x1 , . . . , xi−1 , xi +· · ·+xi+m−1 , xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).
Under this identification the operations id, D, µ correspond to 11 , x1 ∈ Khx1 i and
to 12 ∈ Khx1 , x2 i respectively. More generally the operation


· ((µ •jk D) •jk−1 D) · · · •j1 D
corresponds to the noncommutative monomial xjk xjk−1 · · · xj1 .

Graphically the operation xjk xjk−1 · · · xj1 is pictured as a planar decorated tree
with levels as follows (example: x1 x2 xn x2 ) :
···
D ···
··· D
D ···
D TTTJTJJJ· · · ttt
TTTJtt

Proof. By [17] we know that the free preshuffle algebra on a certain set of generating
operations is spanned by some leveled planar trees whose vertices are labeled by the
operations. Because of the relations entwining D and µ we can move the operations
D up in a leveled tree composition of operations. Because of the associativity of µ
the trees involving only µ give rise to corollas (as in the AsDer case). Hence the
operad P erAsDer is spanned by the trees of the form indicated above. Note that
the levels indicating the order in which the copies of the operation D are performed
is necessary in the preshuffle algebra framework. 

6.3. Remark. One can also check that P erAsDer is a shuffle algebra in the sense
of [17].

6.4. Formal group laws in noncommutative variables. Let F (x, y) be a for-

mal group law in noncommutative variables. This is a series in K((x, y)) which
satisfies the relation
F (F (x, y), z) = F (x, F (y, z)).
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 15

For instance, the Baker-Campbell-Hausdorff series is the formal power series defined
as
BCH(x, y) := log(exp(x) exp(y)).
We recall that the first terms are
1 1
BCH(x, y) = x + y + [x, y] + ([[x, y], y] + [x, [x, y]]) + · · · .
2 12
By induction we define
F (x1 , . . . , xn ) = F (F (x1 , . . . , xn−1 ), xn ).

6.5. Proposition. For any formal group law F in noncommutative variables there
is a well-defined preshuffle algebra P erRatF ctF given by:
P erRatF ctF
n := K((x1 , . . . , xn )),

(P ◦i Q)(x1 , . . . , xn+m−1 ) :=
P (x1 , . . . , xi−1 , F (xi , · · · , xi+m−1 ), xi+m , . . . , xn+m−1 )Q(xi , . . . , xi+m−1 ).
If F is polynomial, then the restriction to Khx1 , . . . , xn i is still a preshuffle algebra.

Proof. Axiom I is an immediate consequence of the equality

F (x1 , . . . , xi , F (xi+1 , . . . , xi+j ), xi+j+1 , . . . , xn ) = F (x1 , . . . , xn ).
The last assertion is immediate. 

6.6. Corollary. Let F be the additive formal group law F (x, y) = x + y. Then the
associated polynomial preshuffle algebra is P erAsDer.

7. Appendix: Homotopy associative algebras with derivation

We know that homotopy associative algebras are A∞ -algebras as defined by Jim

Stasheff in [18]. Our purpose is to describe the notion of homotopy associative
algebras with a derivation, that is to unravel the operad AsDer∞ . The solution
is given by the Koszul duality theory of quadratic operads, see for instance [13]
where the Ginzburg-Kapranov theory is extended to operads generated by binary
and unary operations. A quadratic operad P admits a Koszul dual cooperad P ¡ .
The cobar construction over P ¡ is the operad of P-algebras up to homotopy (i.e.
the minimal model P∞ := Ω P ¡ of the operad P) whenever the Koszul complex of
the operad P is acyclic. In this appendix we compute the cooperad AsDer¡ and
its linear dual AsDer! , we prove that the Koszul complex (AsDer¡ ◦ AsDer, δ) is
acyclic and we unravel the co! bar construction AsDer∞ := Ω AsDer¡ . So we
get a precise description of the notion of associative algebra with derivation up to
homotopy.
If the parameter λ ∈ K is different from 0, then the operad λ-AsDer is not a
quadratic operad since the term D(a)D(b) needs three generating operations to be
16 J.-L. LODAY

defined. So one needs new techniques to extend Koszul duality to this case, see
[15].

7.1. The AsDer! -algebras. The relations defining the operad AsDer are qua-
dratic since each monomial involves only the composition of two operations. Hence
AsDer is suitable for applying the Koszul duality theory. We use the notations and
results of [13].

7.2. Proposition. The Koszul dual operad of AsDer is the operad AsDer! gener-
ated by the unary operation d and the binary operation µ which satisfy the following
relations:

d ◦ d = 0, d ◦ µ = µ ◦ (d, id) = µ ◦ (id, d), µ ◦ (µ, id) = µ ◦ (id, µ).

In other words an AsDer! -algebra is an associative algebra A equipped with a

linear map d such that d2 = 0 and d(ab) = d(a)b = ad(b).
Proof. The operad AsDer is generated by the graded vector space

E = (0, K D, K µ, 0, . . .).

The weight 2 subspace of the free operad T (E), denoted T (E)(2) , is spanned by
the operations which are composite of two of the generating operations. It is of
dimension 6 with basis

µ ◦1 µ, µ ◦2 µ, µ ◦1 D, µ ◦2 D, D ◦1 µ, D ◦1 D.

The subspace of relations R is of dimension 2 spanned by D ◦1 µ − µ ◦1 D − µ ◦2 D

and µ ◦1 µ − µ ◦2 µ. The Koszul dual cooperad AsDer¡ is cogenerated by sD and
sµ (s is the shift of degree), with s2 R as corelations:

AsDer¡ = Id ⊕ sE ⊕ s2 R ⊕ · · · .

By definition the Koszul dual operad AsDer! of AsDer is, essentially, the linear
dual of AsDer¡ . Let us denote by d the linear dual of sD (put in degree 0) and µ
the linear dual of sµ (put in degree 0). Then the space T (Kd ⊕ Kµ)(2) is also of
dimension 6 and the quotient AsDer2! = R∨ = T (Kd⊕ Kµ)/R⊥ is two dimensional.
So R⊥ is the 4-dimensional space spanned by the elements

d ◦1 d, d ◦1 µ − µ ◦1 d, d ◦1 µ − µ ◦2 d, µ ◦1 µ − µ ◦2 µ.

7.3. Proposition. The space of n-ary operations of the operad AsDer! is 2-dimensional:

AsDern! = K µn ⊕ K dµn , n ≥ 2, and AsDer1! = K id ⊕ K d.

 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 17

The partial composition ◦i is given by

µm ◦ i µn = µm+n−1 ,
dµm ◦i µn = dµm+n−1 ,
µm ◦i dµn = dµm+n−1 ,
dµm ◦i dµn = 0,

where by convention µ1 = id (so d ◦1 µn = dµn ).

Proof. The generating binary operation µ generates the operation µn in arity n.

The relations entwining µ and d imply that the only other possibility to create an
operation in arity n is to compose with a copy of d. The formula for the partial
composition is obtained by direct inspection. 

7.4. Proposition. The operad AsDer is a Koszul operad.

Proof. Most of the methods for proving Koszul duality would work in this simple
case. We choose to write down explicitly the “rewriting system method”, see [13].
The operad AsDer is presented by the generators D and µ and the rewriting
relations

µ ◦1 µ 7→ µ ◦2 µ,
D ◦1 µ 7→ µ ◦1 D + µ ◦2 D.

The critical monomials are µ ◦1 (µ ◦1 µ) and D ◦1 (µ ◦1 µ). The first critical

monomial is known to be confluent (Koszulity of the operad As), but let us recall
the proof. One one hand, one has

((xy)z)t 7→ (x(yz))t 7→ x((yz)t) 7→ x(y(zt)).

On the other hand, one has

((xy)z)t 7→ (xy)(zt) 7→ x(y(zt)).

Since one ends up with the same element, we have shown that the first critical
monomial is confluent.
Let us show confluence for the second critical monomial. One one hand, one has
D((xy)z) 7→ D(x(yz)) 7→ (Dx)(yz) 7→ x(D(yz)) 7→ (Dx)(yz) + x((Dy)z + y(Dz)) 7→
(Dx)(yz) + x((Dy)z) + x(y(Dz)).
On the other hand, one has
D((xy)z) 7→ (D(xy))z) + (xy)(Dz) 7→ ((Dx)y)z + (x(Dy))z + (xy)(Dz) 7→
(Dx)(yz) + x((Dy)z) + x(y(Dz)).
Since one ends up with the same element, we are done.
Since the critical monomials are confluent the ns operad AsDer is Koszul. 
18 J.-L. LODAY

7.5. Homotopy associative algebra with derivation. Since the operad AsDer
is a Koszul operad, its minimal model is given by the operad ΩAsDer¡ . The aim
of this section is to describe this operad explicitly.
By definition a homotopy associative algebra with derivation is an algebra over
the differential graded ns operad AsDer∞ constructed as follows. In arity n the
space (AsDer∞ )n is spanned by the planar trees with n leaves whose nodes with
k inputs are labelled by mk or Dmk when k ≥ 2 and by D when k = 1. Observe
that the symbol Dmk is to be taken as a whole and not as mk followed by D. The
homological degree of mk and of Dmk is k − 1 for k ≥ 2, the homological degree of
D is 0.
Example:
EE
E yyy }}
}
m2 P D
PPP
vvv
Dm3

D
D

The differential map ∂ : (AsDer∞ )n → (AsDer∞ )n of homological degree −1 is

induced by
∂(D) = 0, P
∂(mn ) = − n=p+q+r (−1)p+qr mk ◦ (id⊗p ⊗ mq ⊗ id⊗r ),
k=p+1+r
P k>1,q>1
∂(Dmn ) = − n=p+q+r (−1)
p+qr
Dmk ◦ (id⊗p ⊗ mq ⊗ id⊗r )
k=p+1+r
P k≥1,q>1
− n=p+q+r (−1)p+qr mk ◦ (id⊗p ⊗ Dmq ⊗ id⊗r )
k=p+1+r
k>1,q≥1

for n ≥ 2.
For instance, in low arity we get:
∂(m2 ) = 0,
∂(m3 ) = m2 ◦ (id, m2 ) − m2 ◦ (m2 , id),
∂(Dm2 ) = −D ◦ m2 + m2 ◦ (id, D) + m2 ◦ (D, id).

7.6. Proposition. The operad AsDer∞ is isomorphic to the minimal model ΩAsDer¡
of the operad AsDer. Hence a homotopy associative algebra with derivation is an
algebra over the operad AsDer∞ .

Proof. By definition the cobar construction over the cooperad AsDer∞ ¡ is the free
operad T (sAsDer¡ ). Since, in arity n ≥ 2, the space (AsDer¡ )n is spanned by two
elements and the space AsDer¡ 1 by one element (the notation overline means that
we get rid of the identity operation), the free operad T (sAsDer¡ ) is spanned by the
 ON THE OPERAD OF ASSOCIATIVE ALGEBRAS WITH DERIVATION 19

planar rooted trees with labelled nodes as in the description of AsDer∞ . In the
free operad the operadic composition is given by grafting.
The operad ΩAsDer¡ is a differential graded operad, so we need to describe the
differential map. It is sufficient to describe it on the operadic generators, that is on
the corollas labelled by either mk or Dmk . This boundary map is deduced from the
cooperad structure of AsDer¡ , that is, from the operad structure of AsDer! . From
the formulas in Proposition 7.3 we deduce the formulas given in the construction
of the operad AsDer∞ . 

7.7. Transfer theorem. Let us recall that the interest of the notion of “algebra
up to homotopy” lies, in part, in the following transfer theorem. Let (A, δ) be
a differential graded associative algebra with derivation DA of degree 0, that is
δ(ab) = δ(a)b + (−1)|a| aδ(b) and δ(DA (a)) = DA (δ(a)), and let (V, δ) be a retract
by deformation of the chain complex (A, δ) (e.g. (H(A), 0) when K is a field).
Then (V, δ) can be equipped with a AsDer∞ -algebra structure transferred from the
AsDer-algebra structure of (A, δ). In particular there exist analogues of the Massey
products on H(A). Since AsDer is Koszul, all these results are particular examples
of [13], Chapter 9, which extend the results obtained by Tornike Kadeishvili [6] on
differential graded associative algebras.

References
[1] F. Chapoton, Algèbres de Hopf des permutoèdres, associaèdres et hypercubes, Advances in
Math. (150), no 2 (2000), 264–275.
[2] F. Chapoton, The anticyclic operad of moulds, International Math. Research Notices (2007)
no 20.
[3] F. Chapoton, F. Hivert, J.-C. Novelli, J.-Y. Thibon, An operational calculus for the Mould
operad, Int. Math. Res. Not. IMRN 2008, no. 9, Art. ID rnn018, 22 pp.
[4] V. Dotsenko, Compatible associative products and trees, J. Algebra & Number Theory, 3
(2009), no. 5, 567-586.
[5] V. Dotsenko, A. Khoroshkin, Gröbner basis for operads, Duke Math. Journal, to appear.
[6] T. Kadeishvili, The algebraic structure in the homology of an A(∞)-algebra. Soobshch. Akad.
Nauk Gruzin. SSR, 108:249–252, 1982.
[7] Li Guo; W. Keigher, On differential Rota-Baxter algebras. J. Pure Appl. Algebra 212 (2008),
no. 3, 522–540.
[8] J.-L. Loday, Algèbres ayant deux opérations associatives (digèbres). C. R. Acad. Sci. Paris
Sér. I Math. 321 (1995), no. 2, 141–146.
[9] J.-L. Loday, Dialgebras, in “Dialgebras and related operads”, Springer Lecture Notes in
Math. 1763 (2001), 7-66.
[10] J.-L. Loday, Generalized bialgebras and triples of operads, Astérisque 320 (2008), x+116 pp.
[11] J.-L. Loday, M. Ronco, Hopf algebra of the planar binary trees. Adv. in Maths 139 (1998),
293–309.
[12] J.-L. Loday, M. Ronco, Trialgebras and families of polytopes, in ”Homotopy Theory: Rela-
tions with Algebraic Geometry, Group Cohomology, and Algebraic K-theory” Contemporary
Mathematics 346 (2004), 369–398.
[13] J.-L. Loday, B. Vallette, Algebraic operads, in preparation.
20 J.-L. LODAY

[14] M. Markl, J. Stasheff, S. Shnider, Operads in algebra, topology and physics. Mathematical
Surveys and Monographs, 96. American Mathematical Society, Providence, RI, 2002. x+349
pp.
[15] S. Merkulov, B. Vallette, Deformation theory of representations of prop(erad)s. I. J. Reine
Angew. Math. 634 (2009), 51–106.
[16] T. Robinson, New perspectives on exponentiated derivations, the formal Taylor theorem, and
Faà di Bruno’s formula, arXiv math:0903.3991.
[17] M. Ronco, Shuffle bialgebras, Ann. Inst. Fourier (2010), to appear.
[18] J.D. Stasheff, Homotopy associativity of H-spaces. I, II. Trans. Amer. Math. Soc. 108 (1963),
275-292; ibid. 293–312.

Institut de Recherche Mathématique Avancée, CNRS et Université de Strasbourg,

7 rue R. Descartes, 67084 Strasbourg Cedex, France
E-mail address: [email protected]