HAL Id: hal-00092521
https://hal.archives-ouvertes.fr/hal-00092521
Submitted on 18 Sep 2006
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
CHEVALLEY COHOMOLOGY FOR KONTSEVICH’S GRAPHS
Didier Arnal, Angela Gammella, Mohsen Masmoudi
To cite this version:
Didier Arnal, Angela Gammella, Mohsen Masmoudi. CHEVALLEY COHOMOLOGY FOR KONT-
SEVICH’S GRAPHS. Pacific Journal of Mathematics, 2005, 218, pp.201-239. �hal-00092521�
CHEVALLEY COHOMOLOGY FOR KONTSEVICH’S GRAPHS
DIDIER ARNAL, ANGELA GAMMELLA, AND MOHSEN MASMOUDI
Abstract. We introduce the Chevalley cohomology for the graded Lie algebra Tpoly(Rd) of polyvector fields on Rd. This cohomology occurs naturally in the problem of construction and classification of formalities on the space Rd. Consi- dering only graph formalities,i.e. formalities defined with the help of graphs like in the original construction of Kontsevich in [K1, K2], we define, as in [AM] for the Hochschild cohomology, the Chevalley cohomology directly on spaces of graphs.
More precisely, observing first a noteworthy property for the Kontsevich’s explicit formality onRd, we restrict ourselves to graph formalities with that property. With this restriction, we obtain some simple expressions of the Chevalley coboundary operator, especially, we can write this cohomology directly on the space of purely aerial, non-oriented graphs. We give finally examples and applications.
1. Introduction
In this article, we study formalities on the space R
d. A formality is a formal non linear mapping F between two formal graded manifolds T
poly( R
d)[1] and D
poly( R
d)[1], intertwinning their natural vector fields Q and Q
0.
Here, T
poly( R
d)[1] (resp. D
poly( R
d)[1]) denotes the space of polyvector fields (resp.
polydifferential operators) on R
dgraded by |α| = degree(α) = k − 2 if α is a k-vector field (resp. |D| = m −2 if D is an m-differential operator) -[1] stands for this choice of translation on degrees- and viewed as a formal manifold (see [K1, K2]). The monomial functions α
1.α
2. . . . .α
non T
poly( R
d) are elements of the space S
n(T
poly( R
d)[1]) of symmetric n-polyvector fields on T
poly( R
d)[1] (that means α
2.α
1= (−1)
|α1||α2|α
1.α
2).
The manifold T
poly( R
d)[1] is equipped with the formal bilinear vector field Q = Q
2, defined with the help of the Schouten bracket [ , ]
S:
Q
2(α
1.α
2) = (−1)
(|α1|−1)|α2|[α
1, α
2]
S. Similarly, D
poly( R
d)[1] is equipped with the formal vector field
Q
0= Q
01+ Q
02, defined by
Q
01(D
1) = −d
HD
1, Q
02(D
1.D
2) = (−1)
(|D1|−1)|D2|[D
1, D
2]
G.
Date: 08/02/04.
1
Here, d
Hdenotes the usual Hochschild coboundary operator: if D is a m-differential operator,
d
HD(f
1, . . . , f
m+1) = f
1D(f
2, . . . , f
m+1) − D(f
1f
2, . . . , f
m+1) + · · · + + (−1)
mD(f
1, . . . , f
m)f
m+1and [ , ]
Gis the Gerstenhaber bracket.
A formality F is then given by a sequence of mappings F
n: F
n: S
nT
poly( R
d)[1]
→ D
poly( R
d)[1], homogeneous with degree 0 and such that the ‘formality equation’:
d
H(F
n)(α
1. . . . .α
n) = 1 2
X
ItJ={1,...,n},|I|6=0,|J|6=0
ε
α(I, J)Q
02F
|I|(α
I).F
|J|(α
J)
− 1 2
X
k6=`
ε
α(k`, 1 . . . k` . . . n)F b
n−1Q
2(α
k.α
`).α
1. . . . . α [
kα
`. . . . .α
nholds. Here, if I = {i
1< · · · < i
`}, the notation α
Imeans α
i1. . . . .α
i`.
We shall impose moreover that F
1is the canonical mapping F
1(0)from T
poly( R
d) to D
poly( R
d) defined by
F
1(0)(ξ
1∧ . . . ∧ ξ
n)(f
1, . . . , f
n) = 1 n!
X
σ∈Sn
ε(σ)
n
Y
i=1
ξ
σ(i)(f
i), for any vector fields ξ
kand any functions f
i.
Let us now choose a coordinate system (x
t) on R
d. In [K2], M. Kontsevich built explicitly a formality U for R
d, using families of graphs drawn on configuration spaces.
A graph Γ has aerial and terrestrial vertices. The aerial vertices are labelled p
1, . . . , p
nand elements of the Poincar´ e half plane
H = {z ∈ C , =(z) > 0}.
The terrestrial vertices q
1< · · · < q
mare on the real line. The edges of Γ are arrows starting from an aerial vertex, ending to a terrestrial or an aerial vertex; there are no arrow of the form p ~
ip
iand no multiple arrow. If we fix a total ordering O on the edges of Γ, we get an oriented graph (Γ, O). We say that O is compatible if, for all i, the arrows starting from p
iare before those starting from p
i+1and denote by GO
n,mthe set of oriented graphs (Γ, O) with n labelled aerial vertices, m labelled terrestrial vertices and O compatible.
Let us consider such an oriented graph (Γ, O) in GO
n,m. Let us also suppose there
are k
iedges starting from the vertex p
i(1 ≤ i ≤ n). In [K2], Kontsevich defines a
natural operator B
(Γ,O)assigning a m-differential operator B
(Γ,O)(α
1⊗ · · · ⊗ α
n) to a
n-uple (α
1, . . . , α
n) of polyvector fields α
i. This operator vanishes except if, for each
i, α
ibelongs to T
polyki−1( R
d) (α
iis a k
i-polyvector field). Let us first consider all the
multi-indexes (t
1, . . . , t
|k|) with |k| = P
k
iand 1 ≤ t
r≤ d for all 1 ≤ r ≤ |k|. We denote by end(a) the set of edges arriving on the vertex a and, if these edges are e
i1, . . . , e
i`, by ∂
end(a)the operator
∂
end(a)= ∂
l∂x
ti1
. . . ∂x
ti`. Then, we denote by star(p
i) the ordered set e
ij1< · · · < e
ijki
of edges starting from p
iand, if α
iis a k
i-vector field, by α
star(pi i)the following component of α
i:
α
star(pi i)= α
tij1...tjki. Finally, if, for each i, α
iis a k
i-vector field,
B
(Γ,O)(α
1⊗ · · · ⊗ α
n)(f
1, . . . , f
m) = X
1≤t1,...,t|k|≤d n
Y
i=1
∂
end(pi)α
star(pi i)m
Y
j=1
∂
end(qj)f
j.
B
(Γ,O)will be called the graph operator associated with (Γ, O).
The explicit formality U of Kontsevich can now be written as a sum U = X
n
U
nwith
U
n= X
m≥0
X
(Γ,O)∈GOn,m
w
(Γ,O)B
(Γ,O),
where the coefficient w
(Γ,O), the weight of (Γ, O) is an integral on a compactified configuration space. To be precise, for 2n + m − 2 ≥ 0, let Conf (n, m) be the space of n distinct points p
iin H and m distinct points q
jon the real line ∂H. Let us act on Conf (n, m) by the group G of transformations z 7→ az + b (a > 0, b real). Consider the quotient space
C
n,m= Conf (n, m)/G.
In [K2], Kontsevich associates with each oriented graph (Γ, O) the following form ω
(Γ,O)on C
n,m:
ω
(Γ,O)= 1 k!
n
^
i=1
dΦ
ei1
∧ . . . ∧ dΦ
eiki
here {e
i1< e
i2< · · · < e
iki} denotes the ordered set star(p
i) formed by the k
iedges starting from p
i, k! := k
1! . . . k
n! and, if e
i`= p ~
ia,
Φ
ei`
= Φ
p~ia= 1 2π Arg
a − p
ia − p ¯
i.
The weight w
(Γ,O)is then defined as the value of the integral ω
(Γ,O)on the connected
component C
n,m+of C
n,mfor which q
1< · · · < q
m.
In this work, we are looking for graph formalities, i.e. formalities on the space R
dof the form F = P
n
F
nwhere the F
nare homogeneous mapping (with degree 0) of the form
F
n= X
m≥0
X
(Γ,O)∈GOn,m
c
(Γ,O)B
(Γ,O),
with real coefficients c
(Γ,O). We shall use the notation F
n= B
γnwhere γ
nis the linear combination
γ
n= X
m≥0
X
(Γ,O)∈GOn,m
c
(Γ,O)(Γ, O).
Let us assume now we found F
1, . . . , F
n−1(with F
1= F
1(0)= U
1) such that the formality equation holds up to order n − 1. Then, the next term F
n, if it exists, is a solution of an equation:
d
H◦ F
n= E
n, that is
d
H(F
n(α
1. . . . .α
n)) = E
n(α
1. . . . .α
n) = E
n(α
{1,...,n}),
where E
n(α
{1,...,n}) is a Hochschild cocycle. It is well-known that the Hochschild cohomology is localized in T
poly( R
d)[1]. More precisely, the total skewsymmetrization
a
◦ E
n(α
{1,...,n}) of E
n(α
{1,...,n}) is a polydifferential operator of order 1, . . . , 1, that is the image by F
1(0)of a polyvector field. Moreover, there exists an operator A
nsuch that
E
n(α
{1,...,n}) = (
a◦ E
n+ d
H◦ A
n)(α
{1,...,n}).
Put now
ϕ
n= F
1−1◦
a◦ E
n, that is
ϕ
n(α
{1,...,n}) = F
1−1 aE
n(α
{1,...,n})
! ,
then ϕ
n: S
n(T
poly( R
d)[1] → T
poly( R
d)[1] is homogeneous with degree |ϕ
n| = 1.
In section 2, we define the Chevalley coboundary operator ∂ on T
poly( R
d). Then, we show that the mapping ϕ
ndescribed above is a Chevalley cocycle, and, if it is a coboundary i.e. ϕ
n= ∂φ
n−1, we can add to F
n−1a Hochschild coboundary so that
a
(E
n) vanishes and thus find a F
nfor which the formality equation holds up to order n.
In Section 3, we etablish a remarkable property for the Kontsevich’s weights. For any graph Γ (with k
iedges starting from p
i), denote by ∆ the purely aerial graph obtained by cutting the legs p ~
iq
jand the feet q
jof Γ and by `
ithe number of aerial edges starting from p
i. We prove that:
a
X
(Γ,O)∈GO(1)n,m
w
(Γ,O)B
(Γ,O)= X
(∆,O∆)∈GO(0)n
w
(∆,O∆)1 m!
X
GO(1)n,m3(Γ,O)⊃(∆,O∆)
`!
k! ε(Γ)B
(Γ,O).
Here GO
n,m(1)denotes the subspace of GO
n,mformed by the oriented graphs having exactly one leg for each foot, GO
(0)nis the set of purely aerial oriented graphs (∆, O
∆) with n aerial vertices and O
∆compatible and ε(Γ) is an explicit sign depending only on Γ.
This property suggests us to study what we call K-graph formalities. A K-graph formality up to order n is a graph formality F at order n−1 such that ϕ
n= F
1−1◦
a◦E
nhas the form ϕ
n= X
(∆,O∆)∈GO(0)n
c
(∆,O∆)C
(∆,O∆)with real coefficients c
(∆,O∆)and where
C
(∆,O∆)= X
m≥0
1 m!
X
GOn,m(1) 3(Γ,O)⊃(∆,O∆)
`!
k! ε(Γ)B
(Γ,O).
In Section 4, we first give some simple expressions of our Chevalley coboundary operator. Then, we restrict ourselves to K-graph formalities and study the Chevalley cohomology related to the question of building such formalities.
In Section 5, we show that the coboundary operator ∂ can be written directly on the aerial part of the graphs.
We devote the last section to explicit computations and applications. In particular, we prove the triviality of the cohomology for small value of n and give the restriction of the cohomology for linear formalities.
2. Chevalley cohomology and formalities
In this section, we first define a graded Chevalley cohomology in a general algebraic setting i.e. for cochains C : S
n(
g[1]) →
M[1] where
gis a graded Lie algebra and
M
a graded
g-module. In fact two Chevalley coboundary operators are naturally associated with the formality equation for R
d. The first one, say ∂
0, is obtained by endowing D
poly( R
d) with a T
poly( R
d)-graded module structure, cochains are mappings C : S
nT
poly( R
d)[1]
→ D
poly( R
d)[1]. The other one, ∂, is obtained by considering T
poly( R
d) as a graded module over itself, cochains are mapping C : S
nT
poly( R
d)[1]
→ T
poly( R
d)[1]. Using both ∂ and ∂
0, we show that the obstructions to formalities can be interpreted as cocycles for ∂.
2.1. Chevalley cohomology.
Let (
g, [ , ]) be a graded Lie algebra and
Ma graded module over
g. For reasons
of homogeneity, we prefer to work with
g[1] and
M[1]. Thus, we replace [ , ] and the
action of
gon
Mrespectively by [ , ]
0and [ , ]
M, defined for homogeneous α, β in
g[1]
and m in
M[1] with degree |α| (resp. |β|, |m|) by
[α, β]
0= (−1)
(|α|+1)|β|[α, β]
[α, m]
M= (−1)
(|α|+1)|m|α.m.
The space C
n(
g,
M) of n-cochains consists of mappings C from S
n(
g[1]) to
M[1]. The Chevalley coboundary ∂C of a n-cochain C, homogeneous with degree |C|, is the n + 1-cochain defined as
∂C (α
1. . . . .α
n+1) =
n+1
X
i=1
(−1)
|C||αi|ε
α(i, 1 . . .ˆ ı . . . n + 1)[α
i, C(α
1. . . . α ˆ
i. . . .α
n+1)]
M− 1 2
X
i6=j
ε
α(ij, 1.. ı . . . , n b + 1)(−1)
|C|C([α
i, α
j]
0.α
1. . . α d
iα
j. . . .α
n+1).
Here the α
iare homogeneous elements of
g, |α
i| denotes the degree of α
iin
g[1] and for any permutation σ of {1, . . . , n}, ε
α(σ) is the sign of σ in the graded sense. We shall denote by C
[q]n(
g,
M) the subspace of C
n(
g,
M) formed by the n-cochains of degree q and by H
[q]n(
g,
M) the corresponding cohomology group. Note that ∂ sends C
[q]n(
g,
M) into C
[q+1]n+1(
g,
M).
Extending usual techniques to the graded case, it is possible to prove Lemma 2.1. (See [Ga] for an explicit computation)
The operator ∂ is a cohomology operator i.e. ∂
2= ∂ ◦ ∂ = 0.
Let us now return to the graded Lie algebras (T
poly( R
d), [ , ]
S) and (D
poly( R
d), [ , ]
G) where [ , ]
Sis the Schouten bracket and [ , ]
Gthe Gerstenhaber bracket. Let us precise first our conventions for these spaces and brackets.
Let α be a k-vector field and {e
i} the canonical basis of R
d. We put:
α = X
j1,...,jk
α
j1...jke
j1⊗ · · · ⊗ e
jk= X
j1<j2<···<jk
α
j1...jke
j1∧ . . . ∧ e
jk= 1 k!
X
j1...jk
α
j1...jke
j1∧ . . . ∧ e
jk.
For any k
1-vector field α
1and k
2-vector field α
2(the degree of α
iis k
i− 1 in T
poly( R
d)), we define first a polyvector field α
1• α
2with its components
α
1• α
i21···k1+k2−1= 1 k
1!k
2!
X
σ∈Sk
1+k2−1
ε(σ)
k1
X
`=1
(−1)
`−1d
X
s=1
α
i1σ(1)...iσ(`−1)siσ(`)...iσ(k1−1)∂
sα
i2σ(k1)...iσ(k1+k2−1).
Now, [α
1, α
2]
Scan be written as
[α
1, α
2]
S= (−1)
k2(k1−1)α
1• α
2− (−1)
k2−1α
2• α
1.
Note that this choice for the Schouten bracket is denoted [ , ]
0Sin [AMM] and [MT].
On the other hand, for any m
1-differential operator D
1and any m
2-differential operator D
2(the degree of D
iis m
i− 1 in D
poly( R
d)), we may write [D
1, D
2]
Gin the form
[D
1, D
2]
G= D
1◦ D
2− (−1)
(m1−1)(m2−1)D
2◦ D
1where
D
1◦ D
2(f
1, . . . , f
m1+m2−1) =
m1
X
j=1
(−1)
(m2−1)(j−1)D
1(f
1, . . . , f
j−1, D
2(f
j, . . . , f
j+m2−1), f
j+m2, . . . , f
m1+m2−1).
Recall the canonical mapping F
1(0)from T
poly( R
d) into D
poly( R
d): each k-vector field α can be viewed as a k-differential operator F
1(0)(α) of order 1,. . . ,1:
F
1(0)(α)
(f
1, . . . , f
k) =< α, df
1∧ . . . ∧ df
k>= 1
k! α
i1···k∂
i1f
1. . . ∂
ikf
k. Now we consider the following action of T
poly( R
d):
α.D =
a◦ [F
1(0)(α), D]
G∀α ∈ T
poly( R
d), ∀D ∈ D
poly( R
d)
where
adenotes the usual skewsymmetrization of differential operators and [ , ]
Gthe Gerstenhaber bracket. This action defines a T
poly( R
d)-graded module structure on D
poly( R
d). Indeed, one can prove
Proposition 2.2.
The following equalities
i)
a◦ [D
1, D
2]
G=
a◦ [D
1,
a◦ D
2]
Gii) F
1(0)([α
1, α
2]
S) =
a◦ [F
1(0)(α
1), F
1(0)(α
2)]
Giii)
a◦ [F
1(0)([α
1, α
2]
S), D]
G=
a◦ [F
1(0)(α
1),
a◦ [F
1(0)(α
2), D]
G]
G−
− (−1)
(k1−1)(k2−1)a◦ [F
1(0)(α
2),
a◦ [F
1(0)(α
1), D]
G]
Ghold for any D
1, D
2, D in D
poly( R
d), any k
1-vector field α
1and k
2-vector field α
2in T
poly( R
d). Especially, item (iii) means
[α
1, α
2]
S.D = α
1.(α
2.D) − (−1)
(k1−1)(k2−1)α
2.(α
1.D)
and thus D
poly( R
d) is a T
poly( R
d)-module.
Let us now endow D
poly( R
d) with the T
poly( R
d)-graded structure described above.
If C : V
nT
poly( R
d)
= S
nT
poly( R
d)[1]
→ D
poly( R
d)[1] is a mapping homogeneous with degree |C|, we can define its Chevalley coboundary ∂
0C. The latter can be written using the vector fields Q and Q
0, associated respectively with T
poly( R
d) and D
poly( R
d):
∂
0C(α
1. . . . .α
n+1) =
n+1
X
i=1
(−1)
|C||αi|ε
α(i, 1 . . . ˆ ı . . . n + 1)
a◦ Q
02F
1(0)(α
i).C(α
1. . . . α ˆ
i. . . .α
n+1)
− 1 2
X
i6=j
ε
α(ij, 1 . . . ı . . . n b + 1)(−1)
|C|C
Q
2(α
i.α
j).α
1. . . α d
iα
j. . . .α
n+1.
To simplify the writing, we will sometimes write α
iinstead of F
1(0)(α
i).
On the other hand, considering T
poly( R
d) as a graded module over itself, one can define the Chevalley cohomology for T
poly( R
d) . If C : S
nT
poly( R
d)[1]
→ T
poly( R
d)[1]
is an n-cochain, homogeneous with degree |C|, its coboundary ∂C is:
∂C(α
1. . . . .α
n+1) =
n+1
X
i=1
(−1)
|C||αi|ε
α(i, 1 . . .ˆ ı . . . n + 1)Q
2α
i.C(α
1. . . . α ˆ
i. . . .α
n+1)
− 1 2
X
i6=j
ε
α(ij, 1 . . . ı . . . n b + 1)(−1)
|C|C
Q
2(α
i.α
j).α
1. . . α d
iα
j. . . .α
n+1.
Remark 2.1.
For any ϕ : S
nT
poly( R
d)[1]
→ T
poly( R
d)[1], we have:
∂
0(F
1(0)◦ ϕ) = F
1(0)◦ ∂ϕ.
2.2. Obstruction to formalities.
The two Chevalley coboundary operators ∂ and ∂
0enable us to reformulate the for- mality equation. Indeed, suppose we want to construct a formality F from T
poly( R
d) to D
poly( R
d). We need thus to solve recursively the formality equation (see [K, AMM]
for notations)
d
H(F
n)(α
1. . . . .α
n) = 1 2
X
ItJ={1,...,n},|I|≥1,|J|≥1
ε
α(I, J )Q
02F
|I|(α
I).F
|J|(α
J)
− 1 2
X
k6=`
ε
α(k`, 1 . . . k` . . . n)F b
n−1Q
2(α
k.α
`).α
1. . . . . α [
kα
`∂ . . . .α
n,
where d
His the Hochschild coboundary operator.
Let us impose the first component F
1to be F
1(0). Assume there exist mappings F
2, . . . , F
n−1, homogeneous with degree 0, and satisfying the formality equation up to order n − 1. Denote by E
nthe right hand side of the equation at the order n.
Then E
nis a Hochschild cocycle: d
HE
n= 0 (see [AM] for instance). Thus E
n=
a◦ E
n+ d
HC,
where
a◦ E
nis a differential operator of order 1,. . . , 1 and E
nis a coboundary if and only if
a◦ E
n= 0. But:
a
◦ E
n(α
1. . . . .α
n) = ∂
0aF
n−1(α
1. . . . .α
n) +
aR
n(α
1. . . . .α
n), where
R
n(α
1. . . . .α
n) = 1 2
X
ItJ={1,...,n},|I|≥2,|J|≥2
ε
α(I, J )Q
02F
|I|(α
I).F
|J|(α
J) .
It follows directly from this expression that the degree of R
nand
a◦ R
nare both 1, |R
n| = |
a◦ R
n| = 1. Moreover,
Theorem 2.3.
The skewsymmetrization
a◦ E
nof E
ncan be identified through the inverse mapping of F
1with a ∂-cocycle. If this cocycle is exact, we can find F
n−10and F
n0, homogeneous with degree 0, such that F
2, . . . , F
n−2, F
n−10, F
n0satisfy the formality equation up to order n.
Proof: The proof proceed in three steps.
1. Let us first see that
a◦ R
nis a cocycle for ∂
0:
∂
0aR
n(α
1. . . . .α
n+1) =
=
n+1
X
i=1
(−1)
|αi|ε
α(i, 1 . . . ˆ ı . . . n + 1)
aQ
02α
i.
aR
n(α
1. . . . . α b
i. . . . .α
n+1) + 1
2 X
i6=j
ε
α(ij, 1 . . . ı . . . n b + 1)
aR
nQ
2(α
i.α
j).α
1. . . . . α d
iα
j. . . . .α
n+1= 1 2
n+1
X
i=1
(−1)
|αi|ε
α(i, 1 . . . ˆ ı . . . n + 1) X
ItJ={1...ˆı...n+1},|I|≥2,|J|≥2
ε
α0(I, J )
a
Q
02α
i.
aQ
02(F
|I|(α
I).F
|J|(α
J))
+ 1 4
X
i6=j
ε
α(ij, 1 . . . ı . . . n b + 1) X
ItJ={0,1...bı...n+1},|I|≥2,|J|≥2
ε
α00(I, J)
a
Q
02F
|I|(α
I).F
|J|(α
J)
= 1
2 (I) + 1 4 (II),
here α
0:= Q
2(α
i.α
j), ε
α0:= ε
α\{αi}and ε
α00:= ε
(α∪{αo})\{αi,αj}. The first term is:
(I) =
n+1
X
i=1
X
ItJ={1...ˆı...n+1}
|I|≥2,|J|≥2
(−1)
(|αI|+|αJ|)|αi|ε
α(i, I, J)
aQ
02a
Q
02(F
|I|(α
I).F
|J|(α
J)).α
i.
By Proposition 2.2,
aQ
02satisfies the graded Jacobi identity, thus:
(I) = −
n+1
X
i=1
X
ItJ={1...ˆı...n+1}
|I|≥2,|J|≥2
(−1)
|αJ|(|αI|+|αi|)ε
α(i, I, J )
aQ
02a
Q
02(F
|J|(α
J).α
i).F
|I|(α
I)
− X
ItJ={1...ˆı...n+1}
|I|≥2,|J|≥2
ε
α(i, I, J)
aQ
02a
Q
02(α
i.F
|I|(α
I)).F
|J|(α
J)
= − 2 X
ItJ={1...ˆı...n+1}
|I|≥2,|J|≥2
ε
α(i, I, J)
aQ
02a
Q
02(α
i.F
|I|(α
I)).F
|J|(α
J) .
Similarly, the second term is:
(II) = X
i6=j
ε
α(ij, 1 . . . ı . . . n b + 1) X
ItJ={0,1...ı...n+1}b
|I|≥2,|J|≥2
ε
α00(I, J)
aQ
02F
|I|(α
I).F
|J|(α
J)
= X
i6=j
X
I=I1t{0}
I1tJ={1...ı...n+1}b
ε
α(ij, I
1, J)
aQ
02F
|I|(Q
2(α
i.α
j).α
I1).F
|J|(α
J)
+ X
i6=j
X
J=J1t{0}
ItJ1={1...ı...n+1}b
ε
α(ij, 1 . . . ı . . . n b + 1)ε
α00(I, {0}, J
1)
a
Q
02F
|I|(α
I).F
|J|(Q
2(α
i.α
j).α
J1)
= X
i6=j
h X
I=I1t{0}
I1tJ={1...bı...n+1}
ε
α(ij, I
1, J)
aQ
02F
|I|(Q
2(α
i.α
j).α
I1).F
|J|(α
J) +
+ X
J=J1t{0}
ItJ1={1...bı...n+1}
ε
α(ij, I, J
1)(−1)
(|αi|+|αj|+1)|αJ|aQ
02F
|I|(α
I).F
|J|(Q
2(α
i.α
j).α
J1) i
= 2 X
i6=j
X
I=I1t{0}
I1tJ={1...bı...n+1}
ε
α(ij, I
1, J)
aQ
02F
|I|(Q
2(α
i.α
j).α
I1).F
|J|(α
J) .
Putting (I) and (II) together, we get:
∂
0(
aR
n)(α
1. . . . .α
n+1) = 1
2 (I) + 1 4 (II)
= X
I0tJ={1...n+1}
|J|≥2,|I0|≥3
ε
α(I
0, J) h
− X
i∈I0(I0=It{i})
ε
α{i}tI(i, I)
aQ
02a
Q
02(α
i.F
|I|(α
I)).F
|J|(α
J)
+ 1 2
X
i6=j∈I0,I0=I1t{ij}
I=I1t{0}
ε
α{ij}∪I1
(ij, I
1)
aQ
02F
|I|(Q
2(α
i.α
j).α
I1).F
|J|(α
J) i
.
Now, Proposition 2.2 and the definition of ∂
0yield:
(∗) ∂
0(
aR
n)(α
1. . . . .α
n+1) = − X
I0tJ={1...n+1}
|J|≥2, |I0|≥3
ε
α(I, J )
aQ
02∂
0aF
|I0|−1(α
I0).F
|J|(α
J) .
On the other hand, since the formality equation holds up to order n − 1, we have:
∂
0aF
p−1+
aR
p=
a(E
p) =
a(d
H(F
p)) = 0 ∀p ≤ n − 1.
But |I
0| ≤ n − 1 for all I
0in the expression (∗), thus:
−∂
0aF
|I0|−1(α
I0) =
aR
|I0|(α
I0) = 1 2
X
StT=I0,|S|≥2,|T|≥2
ε
αStT(S, T )
aQ
02(F
|S|(α
S).F
|T|(α
T)).
Finally, (∗) becomes:
∂
0(
aR
n)(α
1. . . . .α
n+1) =
= 1 2
X
StTtJ={1...n+1}
|S|≥2,|T|≥2,|J|≥2
ε
α(S ∪ T, J )ε
αStT(S, T )
aQ
02a
Q
02(F
|S|(α
S).F
|T|(α
T)).F
|J|(α
J)
= 1 2
X
StTtJ={1...n+1}
|S|≥2,|T|≥2,|J|≥2
ε
α(S, T, J )
aQ
02a
Q
02(F
|S|(α
S).F
|T|(α
T)).F
|J|(α
J)
:= S
0.
But thanks to the Jacobi identity, S
0vanishes. Hence ∂
0(
aR
n) = 0 and ∂
0(
aE
n) = 0.
2. Let us put
ϕ
n= F
1−1◦
a◦ E
n. Since
∂
0(
a◦ E
n) = ∂
0F
1(ϕ
n) = F
1(∂ϕ
n) = 0,
ϕ
nis a cocycle for ∂ .
3. Let us assume that ϕ
n= ∂φ
n−1where φ
n−1: S
n−1T
poly( R
d)[1]
→ T
poly( R
d)[1].
Of course, d
HF
1(φ
n−1) = 0. Therefore, the mappings F
20= F
2, . . . , F
n−20= F
n−2, F
n−10= F
n−1− F
1◦ φ
n−1satisfy the formality equation up to order n − 1. Moreover, the Hochschild cocycle E
n0corresponding to these F
p0satisfies
a
◦ E
n0=
a◦ E
n− ∂
0(F
1◦ φ
n−1) =
a◦ E
n− F
1(∂φ
n−1) = 0.
We are now able to find F
n0such that E
n0= d
HF
n0. This ends the proof.
3. Skewsymmetrization
The aim of this section is the proof of a noteworthy property of the Kontsevich’s weights and the definition of K-graph formalities.
3.1. Skewsymmetrization and 1-graphs.
Let us first consider a m-differential operator D on R
d, vanishing on constants. We can decompose D as
D = D
(1)+ D
(>1),
where D
(1)has order 1 in each of its arguments and D
(>1)has order larger than 1 for at least one of its arguments. The skewsymmetrization
a(D) of D i.e.
a
(D)(f
1, . . . , f
m) = 1 m!
X
σ∈Sm
ε(σ)D(f
σ−1(1), . . . , f
σ−1(m)) satisfies
a
(D) =
a(D
(1)) +
a(D
(>1)) and therefore
a
(D)
(1)=
a(D
(1)).
We assume D to be defined with the help of graphs:
D
(α1,...,αn)= X
Γ
c
ΓB
Γ(α
1⊗ · · · ⊗ α
n)
where the c
Γare real. To compute a(D)
(1), we need only to consider D
(α(1)1,...,αn)
= X
Γ∈G(1)