A representation formula for maps on supermanifolds
Frédéric Héleina兲
Institut de Mathématiques de Jussieu, UMR 7586, Université Denis Diderot-Paris 7, Case 7012, 2 place Jussieu, 75251 Paris Cedex 5, France
共Received 17 September 2007; accepted 11 January 2008;
published online 12 February 2008
兲
We analyze the notion of morphisms of rings of superfunctions which is the basic concept underlying the definition of supermanifolds as ringed spaces
共i.e., follow-
ing Berezin, Leites, Manin, etc.兲. We establish a representation formula for all共pull-back兲
morphisms from the algebra of functions on an ordinary manifolds to the superalgebra of functions on an open subset of a superspace. We then derive two consequences of this result. The first one is that we can integrate the data associated with a morphism in order to get a共
nonunique兲
map defined on an ordinary space共
and uniqueness can be achieved by restriction to a scheme兲
. The second one is a simple and intuitive recipe to compute pull-back images of a function on a manifoldMby a map from a superspace toM. ©2008 American Institute of Physics.关
DOI:10.1063/1.2840464兴
I. INTRODUCTION
The theory of supermanifolds, first proposed by Salam and Strathdee16 as a geometrical framework for understanding the supersymmetry, is now well understood mathematically and can be formulated in roughly two different ways: either by defining a notion of superdifferential structure with “supernumbers” which generalizes the differential structure of Rp and by gluing together these local models to build a supermanifold. This is the approach proposed by Dewitt5 and Rogers.14,10 Alternatively, one can define supermanifolds as ringed spaces, i.e., objects on which the algebra
共
or the sheaf兲
of functions is actually a superalgebra共
or a sheaf of superalge- bras兲
. This point of view was adopted by Berezin,4Letes,11and Manin12and was recently further developed13 by Deligne and Morgan,7Freed,9and Varadarajan.17The first approach is influenced by differential geometry, whereas the second one is inspired by algebraic geometry. Of course, all these points of view are strongly related, but they may lead to some subtle differences共see
Batchelor,3 Bartocci et al.,2 and Bahraini1兲. For a synthetic overview and a comparison of the
various existing theories, see Ref.15.The starting point of this paper was to understand some implications of the theory of super- manifolds according to the second point of view,4,11,12,8,9,17
i.e., the one inspired by algebraic geometry. The basic question is to understand Rp兩q, the space with p ordinary
共bosonic兲
coordi- nates andqodd共fermionic兲
coordinates. There is no direct definition nor picture of such a space beside the fact that the algebra of functions onRp兩qshould be isomorphic to C⬁共R
p兲关
1, . . . ,q兴,
i.e., the algebra overC⬁共R
p兲
spanned byqgenerators1, . . . ,qwhich satisfy the anticommutation relationsij+ji= 0. Hence,C⬁共R
p兲关
1, . . . ,q兴
is isomorphic to the set of sections of the flat vector bundle overRpwhose fiber is the exterior algebra⌳
*Rq. To experiment further Rp兩q, we define what should be maps from open subsets of Rp兩q to ordinary manifolds. We adopt the provisional definition of an open subset of⍀
of Rp兩q to be a space on which the algebra of functions is isomorphic toC⬁共兩⍀兩兲关
1, . . . ,q兴, where 兩⍀兩
is an open subset ofRp. So let us choose such an open set⍀
and a smooth ordinary manifoldNand analyze what should be mapsfroma兲Electronic mail: helein@math.jussieu.fr.
49, 023506-1
0022-2488/2008/49共2兲/023506/19/$23.00 © 2008 American Institute of Physics
⍀
toN. Again, there is no direct definition of such an object except that by the chain rule it should define a ring morphism*from the ring C⬁共N兲
of smooth functions on N to the ring C⬁共兩⍀兩兲
⫻关
1, . . . ,q兴. The morphism property means that
*
共f
+g兲=
*f+*g, ∀,
苸
R, ∀f,g苸
C⬁共N兲 共1兲
and*
共
fg兲=共
*f兲共*g兲, ∀f,g苸
C⬁共N兲. 共2兲
We restrict ourself toevenmorphisms, which means here that we impose to*fto be in the even partC⬁共兩⍀兩兲关
1, . . . ,q兴
0ofC⬁共兩⍀兩兲关
1, . . . ,q兴.
In the Sec. II, we prove our main result
共Theorem 1.1兲
which shows that, for any even morphism*, there exists a smooth mapfrom兩⍀兩
toNand a family of vector fields共⌶
x兲
x苸兩⍀兩depending on x苸
兩⍀兩
and tangent to N and with coefficients in the commutative subalgebra R关1, . . . ,q兴
0such that*f=
共1 ⫻
兲
*共e
⌶f兲, ∀f苸
C⬁共N兲. 共3兲
One may interpret the term e⌶ as an analog with odd variables of the standard Taylor series representation,g共x兲=
兺
k=0
⬁ kg
共
x兲
k共x
0兲 共x
−x0兲
kk! =
共e
兺i=1n 共xi−x0i兲共/xi兲g兲共x0兲
for a functiongwhich is analytic in a neighbourhood ofx0. We also show that the vector field
⌶ 共which is not unique兲
can be build as a combination of commuting vector fieds. Then, the rest of this paper is devoted to the consequences of this result.The second section explores in details the structure behind relation
共3兲. First, we exploit the
fact that one can assume that the vector fields which compose⌶
commute, so that one can integrate them locally. This gives us an alternative description of morphisms. Eventually, this study leads us to a factorization result for all even morphisms as follows. First, let us denote by⌳
+2*Rq the subspace of all even elements of the exterior algebra⌳
*Rq of positive degree共
i.e.,⌳
+2*Rq⯝
R2q−1−1兲
. We construct an idealIq共兩⍀兩兲
of the algebraC⬁共兩⍀兩 ⫻⌳
+2*Rq兲
in such a way that if we consider the quotient algebraAq共兩⍀兩兲
ªC⬁共兩⍀兩 ⫻⌳
+2*Rq兲
/Iq共兩⍀兩兲, then there exists a canoni-
cal isomorphism T⍀*:Aq共兩⍀兩兲
→C⬁共⍀兲
. By following the theory of scheme of Grothendieck, we associate to Aq共兩⍀兩兲
its spectrum SpecAq共兩⍀兩兲
, a kind of geometric object embedded in兩⍀兩
⫻⌳
+2*Rq. Then, for any even morphism * from C⬁共N兲
to C⬁共兩⍀兩兲关
1, . . . ,q兴, there exists a
smooth map⌽
from兩⍀兩 ⫻⌳
+2*Rq toN, such that*=T⍀*ⴰ
⌽
兩쐓*,where ∀f
苸
C⬁共N兲, ⌽
兩쐓*f=fⴰ⌽modIq共兩⍀兩兲. So by dualizing we can think of the map
⌽
兩쐓: SpecAq共兩⍀兩兲→N
as the restriction of⌽
to SpecAq共兩⍀兩兲. Hence, we obtain an interpretation
of a map on a supermanifold as a function defined on an共almost兲
ordinary space. This reminds somehow the theory developped by Vladimirov and Volovich18who represent a map on a super- space as a function depending on many auxiliary ordinary variables satisfying a system of so- called “Cauchy–Riemann-type equations.” However, their description in terms of ordinary func- tions satisfying first order equations differs from our point of view.The last section is devoted to applications of our results for understanding the use of super- manifolds by physicists. First, we explain briefly how one can reduced the study of maps between two supermanifolds to the study of maps from a supermanifold to an ordinary one by using charts.
Second, we recall why it is necessary to incorporate the notion of the functor of point
共
as illus- trated in this framework in Refs.7,9, and19兲
in the definition of a mapbetween supermanifoldsin terms of ring morphisms. Then, we address the simple question of computing the pull-back image*f of a map f on an ordinary manifoldNby a mapfrom an open subset ofRp兩qtoN. For instance, consider a superfield =+11+22+12F from R3兩2
关
with coordinates共
x1,x2,t,1,2兲兴
toRand look at the Berezin integral,Iª
冕
R3兩2d3xd2*f,where f:R→R is a smooth function. Such a quantity arises for instance in the action
兰
R3兩2d3xd2共
41⑀abDaDb+*f兲
and then fplays the role of a superpotential. Following Berezin’s rules, the integralIis equal to the integral overR3of the coefficient of12in the development of*f, which is actually
*f=fⴰ+1
共f
⬘ⴰ兲
1+2共f
⬘ⴰ兲
2+12关共
f⬘ⴰ兲F
−共f
⬙ⴰ兲
12兴, 共4兲
so thatI=兰
R3d3x关共
f⬘ⴰ兲
F−共
f⬙ⴰ兲
12兴
. The development共
4兲
is well known and can be obtained by several approaches. For instance in Ref. 6 or 9, one computes the coefficient of 12 in the development of*f by the rule*共
−12共D
1D2−D2D1兲
*f兲
, whereD1 andD2 are derivatives with respect to1and2, respectively, andis the canonical embeddingR3R3兩2. Here, we propose a recipe which, I find, is simple, intuitive, but mathematically safe for performing this computation共this recipe is of course equivalent to the already existing rules!兲. It consists roughly in the
following: we reinterpret the relation=+11+22+12F as*=*e11+22+12F=*
共1 +
11兲共1 +
22兲共1 +
12F兲,共5兲
where• 1,2, and Fare first order differential operators which acts on the right, i.e., for instance,
∀f
苸
C⬁共R兲,
af=df共a兲
=f⬘a, and so*af=共f
⬘ⴰ兲
a;• 1,2, andF areZ2-graded in such a way that*is even, i.e., since1 and2are odd, 1
and2are odd andF is even; and
• all the symbols1,2,1,2, andF supercommute.
Let us use the supercommutation rules to develop
共5兲, we obtain
*f=*f+1*1f+2*2f+12*Ff−12*12f, ∀f
苸
C⬁共R兲.
Then, we let the first order differential operators act and this gives us exactly
共4兲.
All these rules are expounded in details in the Sec. IV of this paper. Their justification is precisely based on the results of Sec. II.
II. EVEN MAPS FROMRp円qTO A MANIFOLDN
Our first task will be to study even morphisms * from C⬁
共N兲
to C⬁共兩⍀兩兲关
1, . . . ,q兴, i.e.,
maps between these two superalgebras which satisfy共1兲
and共2兲. Let us first precise the sense of
even. If A=A0丣A1 and B=B0丣B1 are two Z2-graded rings with unity, a ring morphism :B→Ais said to beevenif it respects the grading, i.e.,∀b
苸
B␣,共b兲 苸
A␣ for␣= 0 , 1. In the case at hand,B=C⬁共N兲
is purely even, i.e.,B1=兵0其, and so
*is even if and only if it mapsC⬁共N兲
to C⬁共兩⍀兩兲关
1, . . . ,q兴
0, the even part ofC⬁共兩⍀兩兲关
1, . . . ,q兴. We then say that
is anevenmap from⍀
toN. In the following, we shall denote the rings C⬁共兩⍀兩兲关
1, . . . ,q兴
andC⬁共兩⍀兩兲关
1, . . . ,q兴
0by, respectively,C⬁
共⍀兲
andC⬁共⍀兲
0 and we shall denote by Mor共C
⬁共N兲
,C⬁共⍀兲
0兲
the set of even morphisms fromC⬁共N兲
toC⬁共⍀兲
.We observe that because of hypothesis
共1兲, any such morphism is given by a finite family
共a
i1. . .i2k兲
of linear functionals on C⬁共N兲
with values inC⬁共兩⍀兩兲, where 共i
1¯i2k兲苸 冀1 ,q冁
2kand 0艋
k艋 关
q/2兴 共关
q/2兴
is the integer part ofq/2兲
, by the relation*f=
兺
k=0 关q/2兴
1艋i1⬍¯⬍
兺
ik艋qai1¯i2k
共f兲
i1¯i2k=a쏗共f兲
+兺
1艋i1⬍i2艋q
ai1i2
共f兲
i1i2+ ¯.Here, we will assume that theai1¯i2k’s are skew symmetric in
共i
1¯i2k兲. At this point, it is useful
to introduce the following notations: For any positive integer k, we let Iq共
k兲ª 兵共
i1¯ik兲 苸 冀1 ,
q冁k兩
i1⬍
¯⬍
ik其, we denote by
I=共i
1¯ik兲
an element of Iq共k兲
and we then write I ªi1. . .ik. It will be also useful to use the convention Iq共0兲
=兵쏗其. We let
Iqª艛
k=0q Iq共k兲,
I0q ª艛
k=0关q/2兴Iq共
2k兲
,I1qª艛
k=0关共q−1兲/2兴Iq共
2k+ 1兲
, andI2qª艛
k=1关q/2兴Iq共
2k兲
. Hence, the preceding relation can be written as*f=
兺
k=0 关q/2兴
I苸I
兺
q共2k兲aI
共f 兲
I=兺
I苸I0 q
aI
共f兲
I共6兲
or
共
*f兲共x兲
=兺
I苸I0 q
aI
共f 兲共x兲
I, ∀x苸 兩⍀兩.
A. Construction of morphisms
We start by providing a construction of morphisms satisfying
共
1兲
and共
2兲
. We note :兩⍀兩
⫻
N→Nthe canonical projection map and consider the vector bundle*TN: the fiber over each point共x,
q兲苸 兩⍀兩 ⫻
Nis the tangent spaceTqN. For anyI苸I2q, we choose a smooth section Iof*TNover
兩⍀兩 ⫻
N and we consider theR关1, . . . ,q兴
0-valued vector field,⌶
ª兺
I苸I2q
II.
Alternatively,
⌶
can be seen as a smooth family共⌶
x兲
x苸兩⍀兩 of smooth tangent vector fields onN with coefficients inR关1, . . . ,q兴
0. So each⌶
xdefines a first order differential operator which acts on the algebra C⬁共N兲
丢R关1, . . . ,q兴
0, i.e., the set of smooth functions on N with values in R关1, . . . ,q兴
0, by the relation⌶
xf=兺
I苸I2q
共共
I兲
x·f兲
I, ∀f苸
C⬁共N兲
丢R关1, . . . ,q兴
0.Here, we do not need to worry about the position of I since it is an even monomial. We now define
共letting ⌶
0= 1兲e⌶ª
兺
n=0
⬁
⌶
n n! =兺
n=0 关q/2兴
⌶
n n!,which can be considered again as a smooth family parametrized byx
苸 兩⍀兩
of differential operators of order at most关q
/2兴acting on C⬁共N兲
丢R关1, . . . ,q兴
0. Now, we choose a smooth map :兩⍀兩
→Nand we consider the map
1
⫻
:兩⍀兩→兩⍀兩 ⫻
N x哫 共x,
共x兲兲,
which parametrizes the graph of . Lastly, we construct the following linear operator on C⬁
共N兲 傺
C⬁共N兲
丢R关1, . . . ,q兴
0:C⬁
共N兲 苹
f哫 共1 ⫻
兲
*共e
⌶f兲苸
C⬁共⍀兲,
where共1 ⫻
兲
*共e
⌶f兲共x兲ª共e
⌶xf兲共
共x兲兲
=兺
n=0 关q/2兴
冉 共⌶兲
n!xnf冊 共
共x兲兲,
∀x苸 兩⍀兩.
We observe that actually, for anyx苸
兩⍀兩, we only need to define ⌶
xon a neighborhood of共x兲
in N, i.e., it suffices to define the section⌶
on a neighborhood of the graph ofin兩⍀兩 ⫻
N关
or even on their Taylor expansion inq at order关
q/2兴
around共
x兲兴
.Lemma 1.1: The map f
哫 共
1⫻
兲
*共
e⌶f兲
is a morphism fromC⬁共N兲
toC⬁共⍀兲
0, i.e., satisfies assumptions(1)and (2).Proof:Property
共
1兲
is obvious, so we just need to prove共
2兲
. We first remark that for anyx苸 兩⍀兩
,⌶
xsatisfies the Leibniz rule,⌶
x共fg兲
=共⌶
xf兲g+f共⌶xg兲, ∀f,g苸
C⬁共N兲
丢R关1, . . . ,q兴
0, which immediately implies by recursion that⌶
xn
共fg兲
=兺
j=1
n n!
共
n−j兲
!j!共⌶
x n−jf兲共⌶
xjg兲, ∀f,g
苸
C⬁共N兲
丢R关1, . . . ,q兴
0, ∀n苸
N.共7兲
We deduce easily thate⌶x
共fg兲
=共e
⌶xf兲共e⌶xg兲, ∀x苸 兩⍀兩,
∀f,g苸
C⬁共N兲
丢R关1, . . . ,q兴
0,共8兲
by developping both sides and using共7兲. Now, relation 共8兲
is true, in particular, for functions f,g苸
C⬁共N兲
and if we evaluate this identity at the point共x兲 苸
Nwe immediately conclude thatf
哫共1 ⫻
兲
*共e
⌶f兲satisfies共2兲.
䊏The following result says that actually all morphisms are of the previous type.
Theorem 1.1: Let *:C⬁
共N兲
→C⬁共⍀兲
0 be a morphism. Then there exists a smooth map:兩⍀兩→N and a smooth family
共
I兲
I苸I2q of sections of*TN defined on a neighborhood of the graph of in兩⍀兩 ⫻
N,such that if⌶ª 兺
I苸I2qII,then*f=
共1 ⫻
兲
*共e
⌶f兲, ∀f苸
C⬁共N兲. 共9兲
Proof:Let *:C⬁共N兲
→C⬁共⍀兲
0 which satisfies共1兲
and共2兲. We denote by
aI the functionals involved in identity共6兲. We also introduce the following notation: for any
N苸N, O共共N兲兲
will represent a quantity of the formO共共N兲
兲
=兺
n=N
⬁
兺
I苸Iq共n兲
cII,
where the coefficientscI’s may be real constants or functions. The result will follow by proving by recursion onn
苸
N*the following property:共
Pn兲
: There exists a smooth map :兩⍀兩
→N and there exists a family of vector fields共
I兲
I, where I苸
Iq共
2k兲
and1艋
k艋
n, defined on a neighborhood of the graph ofin兩⍀兩 ⫻
N, such that if⌶
nª兺
k=1
n
兺
I苸Iq共2k兲
II, then
*f=
共
1⫻
兲
*共
e⌶nf兲
+O共共2n+1兲兲
, ∀f苸
C⬁共N兲
.Proof of
共P
1兲:
We start from relation共2兲
and we expand both sides by using共6兲: we first obtain
by identifying the terms of degree 0 in thei’s,a쏗
共
fg兲共
x兲
=共
a쏗共
f兲共
x兲兲共
a쏗共
g兲共
x兲兲
, ∀x苸 兩⍀兩
, ∀f,g苸
C⬁共N兲
, which implies that for anyx苸 兩⍀兩
, there exists some value共
x兲 苸
Nsuch thata쏗
共
f兲共x兲=f共
共x兲兲,
∀x苸 兩⍀兩,
∀f苸
C⬁共N兲.
In other words, there exists a function:兩⍀兩→Nsuch thata쏗
共f兲
=fⴰ. Sincea쏗共f 兲
must beC⬁ for any smoothf, this implies that苸
C⬁共兩⍀兩
,N兲. The relations between the terms of degree 2 in共2兲
areaI
共fg兲共x兲
=共a
I共f兲共x兲兲共a
쏗共g兲共x兲兲
+共a
쏗共f 兲共x兲兲共a
I共g兲共x兲兲
=
共a
I共f兲共x兲兲g共
共x兲兲
+f共
共x兲兲共a
I共g兲共x兲兲,
∀x苸 兩⍀兩,
∀f,g苸
C⬁共N兲,
∀I苸
Iq共2兲,
which implies that for any x苸兩⍀兩, each
aI共·兲共x兲
is a derivation acting on C⬁共N兲, with support 兵
共x兲其, i.e.,
∀I苸
Iq共2兲
there exist tangent vectors共
I兲
x苸
T共x兲N such thataI
共
f兲共x兲=共共
I兲
x·f兲共共x兲兲,
∀f苸
C⬁共N兲
.SinceaI
共
f兲
must be smooth for anyf苸
C⬁共N兲
, the vectors共
I兲
xshould depend smoothly onx, i.e., x哫共
I兲
xis a smooth section of*TN. It is then possible共see the Proposition 1.1 below兲
to extend it to a smooth section of *TN on a neighborhood of the graph of . If we now set共⌶
1兲
xª
兺
I苸Iq共2兲共
I兲
xI, we have on the one hand, e共⌶1兲xf=f+兺
I苸Iq共2兲
共共
I兲
x·f兲I+O共共3兲兲,
∀x苸 兩⍀兩,
∀f苸
N,and on the other hand,
共
*f兲共x兲
=f共
共x兲兲
+兺
I苸Iq共2兲
共共
I兲
x·f兲共
共x兲兲
I+O共共3兲兲,
∀x苸 兩⍀兩,
from which共
P1兲
follows.Proof of
共P
n兲
⇒共P
n+1兲: We assume 共P
n兲
so that a map苸
C⬁共兩⍀兩
,N兲and a vector field⌶
nhave been constructed. Let us denote bybI the linear forms onC⬁
共N兲
such that共1 ⫻
兲
*共e
⌶nf兲=兺
k=0 关q/2兴
I苸I
兺
q共2k兲bI
共
f兲I.共10兲
Then, property
共P
n兲
is equivalent toaI=bI, ∀k
苸 冀
0,n冁
, ∀I苸
Iq共
2k兲
.共
11兲
We use Lemma 1.1: it says us that f哫共1 ⫻
兲
*共e
⌶nf兲
is a morphism; hence,共1 ⫻
兲
*共e
⌶n共fg兲兲
=
关共1 ⫻
兲
*共e
⌶nf兲兴关共1 ⫻
兲
*共e
⌶ng兲兴, so by using共10兲,
兺
k=0n+1
兺
I苸Iq共2k兲
bI
共fg兲
I=兺
k=0
n+1
兺
j=0
k
兺
J苸Iq共2k−2j兲,K苸Iq共2j兲
bJ
共f 兲b
K共g兲
JK+O共共2n+3兲兲. 共12兲
However, the morphism property
共
2兲
for*implies also兺
k=0n+1
兺
I苸Iq共2k兲
aI
共fg兲
I=兺
k=0
n+1
兺
j=0
k
兺
J苸Iq共2k−2j兲,K苸Iq共2j兲
aJ
共f 兲a
K共g兲
JK+O共共2n+3兲兲. 共13兲
We now substract共12兲
to共13兲
and use共11兲: it gives us
兺
I苸Iq共2n+2兲
共a
I共fg兲
−bI共fg兲兲
I=兺
I苸Iq共2n+2兲
关共a
I共f兲
−bI共f兲兲a
쏗共g兲
+a쏗共f兲共a
I共g兲
−bI共g兲兲兴
I. Hence, if we denote␦aIªaI−bI, we obtain that␦aI
共fg兲
=␦aI共f兲共g
ⴰ兲
+共f
ⴰ兲
␦aI共g兲,
∀I苸
Iq共2n
+ 2兲.By the same reasoning as in the proof of
共P
1兲, we conclude that
∀I苸
Iq共2n
+ 2兲, there exist smooth sectionsI of*TNdefined on a neighborhood of the graph of, such that␦aI
共
f兲共x兲=共共
I兲
x·f兲共共x兲兲,
∀x苸 兩⍀兩,
∀I苸
Iq共2n
+ 2兲.Now let us define
⌶
n+1ª⌶
n+兺
I苸Iq共2n+2兲
II. Then, it turns out that
e⌶n+1f=
兺
k=0
n+1
冉 ⌶
n+I苸Iq兺
共2k+2兲II冊
kk! f+O共共2n+3兲
兲
=兺
k=0 n+1
⌶
nk
k!f+
兺
I苸Iq共2n+2兲
I·fI+O共共2n+3兲
兲
=e⌶nf+
兺
I苸IM共2n+2兲
I·fI+O共共2n+3兲
兲,
so that共1 ⫻
兲
*共e
⌶n+1f兲=*f+O共共2n+3兲兲.
Hence, we deduce
共
Pn+1兲
. 䊏Proposition 1.1: In the preceding result, it is possible to construct smoothly the vector fields
I’s in such a way that
关共
I兲
x,共J兲
x兴
= 0, ∀x苸 兩⍀兩,
∀I,J苸
I2q.Proof:Recall that in the previous proof, in order to build
⌶
n+1 out of⌶
n, we introduced, for each I苸
Iq共
2n+ 2兲
, an unique smooth section x哫共
I兲
x of *TN. We will explain here how to extend each such vector fields defined along the graph ofto a neighborhood of the graph of in兩⍀兩 ⫻
N in order to achieve the claim in the proposition. For that purpose, we prove that for some setVª
兵共
x,,q兲 苸
*TN⫻
N兩x苸 兩⍀兩
,苸
T共x兲N,q苸
V共x兲其
, where eachV共x兲 is a neighborhood of共x兲
inN, there exists a smooth mapV →TN,
共x,
,q兲哫 共q,V共x,
,q兲兲,such that ∀共x,
兲 苸
*TN, V共x,,共x兲兲
= and ∀x苸兩⍀兩
fixed, ∀,苸
T共x兲N,关V共
x,, ·兲
,V共x,, ·兲兴
= 0, i.e., the vector fieldsq哫V共
x,,q兲
andq哫V共
x,,q兲
commute onV共x兲.Then, the proposition will follows by extending each vector
共
I兲
x苸
T共x兲N on V共x兲 by q哫
V共x,共
I兲
x,q兲
.The construction is the following. Let
共
Ua兲
a苸Abe a covering ofNby open subsets, let共
a兲
a苸Abe a partition of unity, and let
共y
a兲
a苸Abe a family of charts associated with this covering. For any x苸兩⍀兩, let
Axª兵a 苸
A兩
共x兲苸
Ua其. For any
a苸
Ax and for any linear isomorphismᐉ
:T共x兲N→Rn, where n= dimN, letRx,ᐉ,a be the unique linear automorphism ofRn such that Rx,ᐉ,aⴰdya兩共x兲=
ᐉ
.We then set
yx,ᐉ
共
q兲
ªa兺
苸Ax
a
共
q兲
Rx,ᐉ,aⴰya共
q兲
, ∀q苸
N.We observe that dyx,ᐉ兩共x兲=
ᐉ
and hence, by the inverse mapping theorem, there exists an open neighborhoodV共x兲 of共x兲
inNsuch that the restriction ofyx,ᐉtoV共x兲 is a diffeomorphism. We then defineV共x,,q兲ª
共dy
x,ᐉ兩q兲
−1共ᐉ共
兲兲,
∀q苸
V共x兲.Because of the obvious relationyx,uⴰᐉ=uⴰyx,ᐉ for all linear automorphismu ofRn, it is clear that the definition ofV共x,,q兲does not depend on
ᐉ 共for the same reason
V共x兲is also independant ofᐉ兲. Moreover,
q哫
V共x,,q兲 is simply a vector field which is a linear combination with constant coefficients of the vector fields共
/yx,iᐉ兲
i=1,. . .,n so that the property关V共x,
, ·兲
,V共x,, ·兲兴
= 0 fol- lows. Note also that these vector fields are of course not canonical since they obviously depend onthe charts. 䊏
Remark 1.1: If we assume furthermore that the image ofis contained in an open subset U ofNsuch that there exists a local chart y=
共y
1, . . . ,yn兲:
U→Rn, then it is possible to choose all the vector fieldsI such that共
I兲
x·共
J兲
x·y= 0, ∀x苸 兩⍀兩,
∀I,J苸
I2q.共14兲
Indeed in this case the proof of Proposition 1.1 is much simpler since we do not need to use a partition of unity in order to build V. We just set Vª兵共x,
,q兲苸
*TN⫻
N兩
x苸兩⍀兩
,苸
T共x兲N,q苸
U其 and define V by V共x,,q兲ª共dy
兩q兩兲
−1ⴰdy兩共x兲兩共
兲. Then, for each 共x,
兲苸
*TN fixed, the vector field q哫V共x,
,q兲 has constant coordinates in the variables y␣. Hence, (14) follows.Remark 1.2: We can write an alternative formula for e⌶ by developping this exponential: in each term of the form
共兺
III兲
n, we can see that each monomial which appears contains at most one time any operatorI, so we obtaine⌶=
兺
I苸I0q
I
冉
n兺
艌0n!1I1,. . .,I兺
n苸I0q⑀I
I1¯InI1¯In
冊
,共
15兲
with the convention that the I0q
共0兲= 쏗
contribution is the identity. Here, we have introduced the notation ⑀II1. . .In
: first all the ⑀쏗I1. . .In’s vanish except for ⑀쏗쏗= 1, so that e⌶= 1 mod关1, . . . ,q
兴.
Second, for k
艌
1, if I1=共i
1,1, . . . ,i1,2k1
兲, . . .,
In=共i
n,1, . . . ,in,2kn
兲,
and I=共i
1, . . . ,i2k兲
we write that I1¯In=I if and only if k1+¯+kn=k,兵i
1,1, . . . ,i1,2k1, . . . ,in,1, . . . ,in,2k
n
其= 兵i
1, . . . ,i2k其
and∀j, Ij⫽쏗
(i.e., kj⬎
0). Then,• if I1¯In⫽I,⑀I I1¯In
= 0; and
• if I1¯In=I, ⑀I I1¯In
is the signature of the permutation
共
i1,1, . . . ,i1,2k1, . . . ,in,1, . . . ,in,k
2n
兲哫 共
i1, . . . ,i2k兲
.The preceding expression of e⌶can be recovered by another way: since all the operatorsII
commute, we have1
e⌶=e兺I苸I2qII=
兿
I苸I2 q
eII, which gives also the same result by a straightforward development.
III. A FACTORIZATION OF THE MORPHISM
* A. Integrating the vector fields
I’sIn the same spirit as a tangent vector at a pointq to a manifoldN can be seen as the time derivative of a smooth curve which reachesq, we can describe the I-components of the mor- phism*as higher order approximations of a smooth map from some vector space with values in N. Indeed, let *
苸
Mor共C⬁共N兲
,C⬁共⍀兲
0兲: then by the preceding result
* is characterized by a map苸
C⬁共兩⍀兩
,N兲and 2q−1− 1 vector fields1I tangent toN defined on a neighborhood of the graph ofin兩⍀兩 ⫻
N. By Proposition 1.1, these vector fields can moreover be chosen so that they pairwise commute whenx苸 兩⍀兩
is fixed. So, for anyx苸 兩⍀兩
, we can integrate simultaneously all vector fields共
I兲
xin order to construct a map,⌽共x, · 兲:U
x共⌳
+2*Rq兲
→ N,where
⌳
+2*Rq⯝
R2q−1−1is the subspace of even elements of positive degree of the exterior algebra⌳
*Rq andUx共⌳
+2*Rq兲
is a neighborhood of 0 in⌳
+2*Rq, such that⌽共
x,0兲
=共
x兲 共
16兲
and denoting by共s
I兲
I苸I2q the linear coordinates on⌳
+2*Rq,
⌽
sI
共x,s兲
=I共⌽共x,s兲兲,
∀s苸
Ux共⌳
+2*Rq兲,
∀I苸
I2q.共17兲
We hence obtain a map⌽
from a neighborhood of兩⍀兩 ⫻ 兵0其
in兩⍀兩 ⫻⌳
+2*Rq to N. By using a cutoff function argument, we can extend this map to an application⌽: 兩⍀兩 ⫻⌳
+2*Rq哫
N. Lastly, we introduce theR关1, . . . ,q兴-valued vector field on 兩⍀兩 ⫻⌳
+2*Rq,ª
兺
I苸I2 q
I
si, so that by
共
17兲 ⌽
*=⌶
=兺
I苸I2qII. Then, relation共
9兲
implies*f共x兲=
共e
共f
ⴰ⌽兲兲共x,0兲,
∀f苸
C⬁共N兲,
∀x苸 兩⍀兩,
or by letting:兩⍀兩→ 兩⍀兩 ⫻⌳
+2*Rq,x哫共x, 0兲
to be the canonical injection,*f=*
共e
共f
ⴰ⌽兲兲. 共18兲
Alternatively by using共
15兲
, we have*f共x兲=
兺
I苸I0 q
I
冉
k兺
艌0k!1I1,. . .,I兺
k苸I0 q⑀I
I1¯Ik k
共f
ⴰ⌽兲
sI1¯sIk
共x,0兲 冊, ∀x苸 兩⍀兩
. 共19兲
It is useful to introduce the differential operatorsD쏗ª1 and
1Note that cardIq共2k兲=q!/共q− 2k兲!共2k兲! and兺k=0
关q/2兴q!/共q− 2k兲!共2k兲! = 2q−1.