A GENERAL CONSTRUCTION OF WEIL FUNCTORS

WOLFGANG BERTRAM, ARNAUD SOUVAY

Abstract. We construct the Weil functor T A corresponding to a general Weil

algebra A = K ⊕ N : this is a functor from the category of manifolds over a
general topological base field or ring K (of arbitrary characteristic) to the category
of manifolds over A. This result simultaneously generalizes results known for
ordinary, real manifolds (cf. [KMS93]), and results obtained in [Be08] for the case
of the higher order tangent functors (A = T k K) and in [Be10] for the case of
jet rings (A = K[X]/(X k+1 )). We investigate some algebraic aspects of these
general Weil functors (“K-theory of Weil functors”, action of the “Galois group”
AutK (A)), which will be of importance for subsequent applications to general
differential geometry.

1. Introduction
The topic of the present work is the construction and investigation of general
Weil functors, where the term “general” means: in arbitrary (finite or infinite)
dimension, and over general topological base fields or rings. Compared to the (quite
vast) existing literature on Weil functors (see, e.g., [KMS93, K08, K00, KM04]),
this adds two novel viewpoints: on the one hand, extension of the theory to a
very general context, including, for instance, base fields of positive characteristic,
and on the other hand, introduction of the point of view of scalar extension, well-
known in algebraic geometry, into the context of differential geometry. This aspect
is new, even in the context of usual, finite-dimensional real manifolds. We start by
explaining this item.
A quite elementary approach to differential calculus and -geometry over general
base fields or -rings K has been defined and studied in [BGN04, Be08]; see [Be11]
for an elementary exposition. The term “smooth” always refers to the concept
explained there, and which is called “cubic smooth” in [Be10]. The base ring K is
a commutative unital topological ring such that K× , the unit group, is open dense
in K, and the inversion map K× → K, t 7→ t−1 is continuous. For convenience,
the reader may assume that K is a topological k-algebra over some topological field
k, where k is his or her favorite field, for instance k = R, and one may think of
K as R ⊕ jR with, for instance, j 2 = −1 (K = C), or j 2 = 1 (the “para-complex
numbers”) or j 2 = 0 (“dual numbers”). In our setting, the analog of the “classical”
Weil algebras, as defined, e.g., in [KMS93], is as follows:

2010 Mathematics Subject Classification. 13A02, 13B02, 15A69, 18F15, 58A05, 58A20, 58A32,
58B10, 58B99, 58C05.
Key words and phrases. Weil functor, Taylor expansion, scalar extension, polynomial bundle,
jet, differential calculus.
1
2 WOLFGANG BERTRAM, ARNAUD SOUVAY

Definition 1.1. A Weil K-algebra is a commutative and associative K-algebra A,

with unit 1, of the form A = K ⊕ NA , where N = NA is a nilpotent ideal. We
assume, moreover, N to be free and finite-dimensional over K. We equip A with
the product topology on N ∼
= Kn with respect to some (and hence any) K-basis.
As is easily seen (Lemma 3.2), A is then again of the same kind as K, hence it is
selectable as a new base ring. Since an interesting Weil algebra A is never a field,
this explains why we work with base rings, instead of fields. Our main results may
now be summarized as follows (see Theorems 3.6 and 4.4 for details):
Theorem 1.2. Assume A = K ⊕ N is a Weil K-algebra. Then, to any smooth
K-manifold M, one can associate a smooth manifold T A M such that:
(1) the construction is functorial and compatible with cartesian products,
(2) T A M is a smooth manifold over A (hence also over K), and for any K-
smooth map f : M → N, the corresponding map T A f : T A M → T A N is
smooth over A,
(3) the manifold T A M is a bundle over the base M, and the bundle chart changes
in M are polynomial in fibers (we call this a polynomial bundle, cf. Defini-
tion 4.1),
(4) if M is an open submanifold U of a topological K-module V , then T A U can
be identified with the inverse image of U under the canonical map VA → V ,
where VA = V ⊗K A is the usual scalar extension of V ; if, in this context,
f : U → W is a polynomial map, then T A f coincides with the algebraic
scalar extension fA : VA → WA of f .
The Weil functor T A is uniquely determined by these properties.
The Weil bundles T A M are far-reaching generalizations of the tangent bundle
T M, which arises in the special case of the “dual numbers over K”, A = T K =
K ⊕ εK (ε2 = 0). The theorem shows that the structure of the Weil bundle T A M is
encoded in the ring structure of A in a much stronger form than in the “classical”
theory (as developed, e.g., in [KMS93]): the manifold T A M plays in all respects
the rôle of a scalar extension of M, and hence T A can be interpreted as a functor
of scalar extension, and we could write MA := T A M – an interpretation that is
certainly very common for mathematicians used to algebraic geometry, but rather
unusual for someone used to classical differential geometry; in this respect, our
results are certainly closer to the original ideas of André Weil ([Weil]) than much
of the existing literature. In subsequent work we will exploit this link between
the “algebraic” and the “geometric” viewpoint to investigate features of differential
geometry, most notably, bundles, connections, and notions of curvature.
The algebraic point of view naturally leads to emphasize in differential geometry
certain aspects well-known from the algebraic theory. First, Weil algebras and
-bundles form a sort of “K-theory” with respect to the operations
(1) tensor product: A ⊗K B ∼
= K ⊕ (NA ⊕ NB ⊕ NA ⊗ NB ),
(2) Whitney sum: A ⊕K B := (A ⊗K B)/(NA ⊗K NB ) ∼ = K ⊕ NA ⊕ NB .
Whereas (2) corresponds exactly to the Whitney sum of the corresponding bundles
over M, one has to be a little bit careful with the bundle interpretation of (1)
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 3

(Theorem 4.5): Weil bundles are in general not vector bundles, hence there is no
plain notion of “tensor product”; in fact, (1) rather corresponds to composition of
Weil functors:
T A⊗B M ∼= T B (T A M).
Second, following the model of Galois theory, for understanding the structure of
the Weil bundle T A M over M, it is important to study the automorphism group
AutK (A). Indeed, by functoriality, any automorphism Φ of A induces canonically a
diffeomorphism ΦM : T A M → T A M which commutes with all A-tangent maps T A f ,
for any f belonging to the K-diffeomorphism group of M, Diff K (M). Thus we have
two commuting group actions on T A M: one of AutK (A) and the other of Diff K (M).
As often in group theory, we get a better understanding of one group action by
knowing another group action commuting with it. Here, an important special case
is the one of a graded Weil algebra (Chapter 5; in [KM04] the term homogeneous
Weil algebra is used): in this case, there exist “one-parameter subgroups” of au-
tomorphisms, and the Weil algebra carries as new structure a “composition like
product”, turning it into a (in general non-commutative) near-ring, similar to the
near-ring of formal power series (Theorem 5.7).
In [Be08] and [Be10], two prominent cases of Weil functors have been studied,
and the present work generalizes these results: the higher order tangent functors T k ,
corresponding to the iterated tangent rings, T k+1K := T (T k K), and the jet functors
Jk , corresponding to the (holonomic) jet rings Jk K := K[X]/(X k+1). Since both
cases play a key rôle for the proof of our general result, we recall (and slightly extend)
the results for these cases (see Appendices A and B). In particular, we describe
in some detail the canonical K× -action, which appears already in the framework
of difference calculus, and which gives rise to the natural grading of these Weil
algebras.
The core part for the proof of Theorem 1.2 is a careful investigation of the relation
between two foundational concepts, namely those of jet, and of Taylor expansion. As
is well-known in the classical setting (see, e.g., [Re83]), both concepts are essentially
equivalent, but the jet-concept is of an “invariant” or “geometric” nature, hence
makes sense independently of a chart, whereas Taylor expansions can be written
only with respect to a chart, hence are not of “invariant” nature. This is reflected
by a slight difference in their behaviour with respect to composition of maps: for
jets we have the “plainly functorial” composition rule
(1.1) Jkx (g ◦ f ) = Jkf (x) g ◦ Jkx f ,
whereas for Taylor polynomials (which we take here without constant term) we have
truncated composition of polynomials:


(1.2) Taykx (g ◦ f ) = Taykf (x) g ◦ Taykx f mod (deg > k) .
This lack of functoriality is compensated by the advantage that, like every poly-
nomial, the Taylor polynomial always admits algebraic scalar extensions, so that
the k-th order Taylor polynomial P = Taykx f : V → W of a K-smooth function
f : V ⊃ U → W at x ∈ U extends naturally to a polynomial PA : V ⊗K A → W ⊗K A.
Our general construction of Weil functors combines both extension procedures: in
4 WOLFGANG BERTRAM, ARNAUD SOUVAY

a first step, based on differential calculus reviewed in Appendices A and B, we con-

struct the jet functor Jk , which associates to each smooth map f its (“simplicial”)
k-jet Jk f . Using this invariant object, we define in a second step the Taylor poly-
nomial Taykx f at x by a “chart dependent” construction (Section 2.2), and then we
prove the transformation rule (1.2) (Theorem 2.13). Note that, in the classical case,
our Taylor polynomial Taykx f coincides of course with the usual one, but we do not
define it in the usual way in terms of higher order differentials (since we then would
have to divide by j!, which is impossible in arbitrary characteristic). In a third step,
we consider the algebraic scalar extension of the Taylor polynomial from K to N :
if the degree k is higher than the order of nilpotency of N , then
(1.3) TxA f := (Taykx f )N : VN := V ⊗ N → WN
satisfies a plainly functorial transformation rule, so that finally the Weil
functor
T A can be defined by T A f : U × VN → W × WN , (x, y) 7→ f (x), TxA f (y) . This
strategy requires to develop some general tools on continuity and smoothness of
polynomials, which we relegate to Appendix C.
From the point of view of analysis, the interplay between jets and Taylor poly-
nomials is reflected by the interplay between (generalized) difference quotient maps
and (various) remainder conditions for the remainder term in a “limited expansion”
of a function f around a point x. The first aspect is functorial and well-behaved in
arbitrary dimension, hence is suitable for defining our general differential calculus;
it implies certain radial limited expansions, together with their multivariable ver-
sions, which represent the second aspect, and which are the starting point for our
definition of Taylor polynomials (Chapter 2).
This work is organized in four main chapters and three appendices, as follows:
2. Taylor polynomials, and their relation with jets
3. Construction of Weil functors
4. Weil functors as bundle functors on manifolds
5. Canonical automorphisms, and graded Weil algebras
Appendix A: Difference quotient maps and K× -action
Appendix B: Differential calculi
Appendix C: Continuous polynomial maps, and scalar extensions
The results of the present work will allow a more conceptual and more general ap-
proach to most of the topics treated in [Be08]; this will be investigated in subsequent
work.

Notation. As already mentioned, K is a commutative unital topological base

ring, with dense unit group K× , and A is a Weil K-algebra. Concerning “cubic” and
“simplicial” differential calculus, specific notation is explained in Appendices A and
B. In particular, superscripts of the form [·] refer to “cubic”, and superscripts of the
form h·i refer to “simplicial” differential calculus. In general, the “cubic” notions
are stronger than the “simplicial” ones (“cubic implies simplicial”, see Theorem
B.6). As a rule, boldface variables are used for k-tuples or n-tuples, such as, e.g.,
v = (v1 , . . . , vk ), 0 = (0, . . . , 0).
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 5

2. Taylor polynomials, and their relation with jets

2.1. Limited expansions. Let V, W be topological K-modules and f : U → W a
map defined on an open set U ⊂ V . Notation concerning extended domains, like
U [1] and U hki , is explained in Appendix A.
Definition 2.1. We say that f admits radial limited expansions if there exist con-
tinuous maps ai : U × V → W and Rk : U [1] → W such that, for (x, v, t) ∈ U [1] ,
k
X
f (x + tv) = f (x) + ti ai (x, v) + tk Rk (x, v, t)
i=1

and the remainder condition Rk (x, v, 0) = 0 hold. We say that f admits multi-
i
variable radial limited expansions if there exist continuous maps bP i : U ×V → W
and Sk : U hki → W , such that, for v = (v1 , . . . , vk ) ∈ V k with x + ki=1 vi ∈ U,
k
X
2 k
f (x + tv1 + t v2 + . . . + t vk ) = f (x) + ti bi (x, v1 , . . . , vi ) + tk Sk (x, v; 0, . . . , 0, t)
i=1

and the remainder condition Sk (x, v; 0, . . . , 0, 0) = 0 hold.

Taking 0 = v2 = v3 = . . . = vk , the multi-variable radial condition implies the
radial condition. For the next result, cf. Appendix B for definition of the class C hki .
Theorem 2.2 (Existence and uniqueness of limited expansions). Assume f : U →
W is of class C hki . Then f admits limited expansions of both types from the preceding
definition, and such expansions are unique, given by
bi (x, v1 , . . . , vi ) = f hii (x, v1 , . . . , vi ; 0),
ai (x, h) = f hii (x, h, 0; 0).
Proof. Existence follows from Theorem B.5, by letting there s0 = s1 = . . . = sk−1 =
0 and sk = t. Uniqueness is proved as in [BGN04], Lemma 5.2. 
Recall from Appendix B the definition of the class C [k] , and the fact that C [k]
implies C hki (Theorem B.6).
Corollary 2.3. Assume f is of class C [k] . Then, for i = 1, . . . , k, the normalized
differential
ai (x, ·) : V → W, h 7→ Dhi f (x) := ai (x, h)
is continuous and polynomial (in the sense of Definition C.2), homogeneous of de-
gree i, hence smooth.
Proof. See [Be10], Cor. 1.8. The last statement follows from Theorem C.3. 
2.2. Taylor polynomials. We can now define Taylor polynomials, which are the
core of the construction of Weil functors.
Definition 2.4. Let f : U → W be of class C [k] and fix x ∈ U. The polynomial
k
X k
X
Taykx f : V → W, h 7→ Dhi f (x) = ai (x, h)
i=1 i=1
is called the k-th order Taylor polynomial of f at the point x.
6 WOLFGANG BERTRAM, ARNAUD SOUVAY

Theorem 2.5. If f : U → W is of class C [ℓ] , then for all k ≤ ℓ, Taykx f is a smooth

polynomial map of degree at most k, without constant term. If, moreover, f is
itself a polynomial map of degree at most k, then Tayk0 f coincides with f , up to the
constant term:
f = f (0) + Tayk0 f ,
and its homogeneous part fi of f of degree i is equal to ai (x, ·) = f hii (x, ·, 0; 0).
Proof. The fact that Taykx f is smooth and polynomial follows from Corollary 2.3.
Next, assume that f is a homogeneous polynomial, say, of degree j with j ≤ k.
Uniqueness of the radial Taylor expansion (Theorem 2.2) implies that, for every
homogeneous map of degree j and of class C [k] , we have f (h) = f hji (x, h, 0; 0) (see
[Be08], Cor. 1.12 for the proof in case j = 2, which generalizes without changes),
whence the claim. 
The radial limited expansion can now be written as
(2.1) f (x + th) = f (x) + Taykx f (th) + tk Rk (x, h, t) .
If the integers are invertible in K, then, by uniqueness of the expansion, it coincides
with the usual Taylor expansion: Dvj f (x) = j!1 dj f (x)(v, . . . , v), see [BGN04].

2.3. Normalized differential and polynomiality.

Definition 2.6. Let f : U → W be of class C [k] . Recall from above the definition
of the normalized differential Dvi f (x)
P := f hii (x, v, 0; 0). We define, for all multi-
indices α ∈ Nk such that |α| := k
i αi ≤ k, and for all v ∈ V , x ∈ U, the
normalized polynomial differential (which is well-defined, by Corollary B.7)


Dvα f (x) := Dvαkk ◦ . . . ◦ Dvα11 f (x).
The following is a generalization of Schwarz’s Lemma (Th. B.2 (iv)):
Lemma 2.7. Let f : U ⊂PV → W a a map of class C [k] . Then, for all multi-indices
α ∈ Nk such that |α| := i αi ≤ k, and for all v ∈ V k , the map Dvα f (x) does not
depend on the order in wich we compose the normalized differentials Dvαii .
Proof. Dvα f (x) is obtained, by restriction to s = 0, from the map
 
α
Dv,s f (x) := Dvαk,s(k) ◦ . . . ◦ Dvα11,s(1) f (x),
k

where for all 1 ≤ i ≤ k, we let Dvαii,s(i)

:= f hαi i (x, vi , 0; s(i) ). From the definition of
the simplicial different quotient map, we get, for non-singular s(1) :
α1 (1) (1)

X f x + (s j − s 0 )v1
Dvα11,s(1) f (x) = Q
1
(1) (1)
.
j1 =0 i=0,...,ĵ1 ,...,α1 (sj1 − si )

In a second step, for non-singular s(2) , we get:

α2 X
α1 (1) (1) (2) (2)

  X f x + (sj1 − s0 )v1 + (sj2 − s0 )v2
Dvα22,s(2) ◦ Dvα11,s(1) f (x) = Q (1) (1) Q (2) (2)
,
j2 =0 j1 =0 (s
i=0,...,ĵ1 ,...,α1 j1 − s i ) (s
i=0,...,ĵ2 ,...,α2 j2 − s i )
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 7

and so on: by induction, we finally get, for non singular s(i) :

α1 αk Pk (ℓ) (ℓ)

  X X f x + ℓ=1 (s j − s 0 )vℓ
Dvαk,s(k) ◦ . . . ◦ Dvα11,s(1) f (x) = ··· Qk Q
ℓ
(ℓ) (ℓ)
.
k
j1 =0 jk =0 ℓ=1 i=0,...,ĵℓ ,...,αℓ (s j ℓ
− s i )
Obviously, the right-hand side term does not change if we apply the operators Dvαii,s(i)
in another order. Hence, by continuity and density of K× in K, this remains true
for singular values of s(i) , and in particular for s(i) = 0. 
Theorem 2.8. Let f : U → W be a map of class C [k] . Then:
(1) For all x ∈ U and for all multi-indices α ∈ Nk such that |α| ≤ k, the map
V k → W, v 7→ Dvα f (x) is polynomial multi-homogeneous of multidegree α.
(2) The map v 7→ f hji (x, v; 0) is polynomial. More precisely, for all 1 ≤ j ≤ k,
X
(2.2) f hji (x, v; 0) = Dvα f (x) .
Pk
α∈Nk , i=1 iαi =j


Proof. (1) Note that Dvα f (x) = Dvαkk ◦ . . . ◦ Dvα11 f (x) is polynomial homogeneous
of degree αk in vk , by Corollary 2.3. By the previous lemma, the value of Dvα f (x) is
independent of the order in which we compose the normalized differentials. There-
fore Dvα f (x) is also polynomial homogeneous of degree αi in vi for all 1 ≤ i ≤ k,
i.e., v 7→ Dvα f (x) is a polynomial multi-homogeneous map of multidegree α.
(2) On the one hand, we use the k-th order radial limited expansion, successively
for each variable vi , 1 ≤ i ≤ k. This is well-defined (see Corollary B.7).
 P   P 
f x + ki=1 ti vi = f x + ki=2 ti vi + t1 v1
k k
!
X X
= tα1 Dvα11 f x + ti vi + tk Rk1 (x, v1 , t)
α1 =0 i=2
k k−α k
!
X X1 X
= tα1 t2α2 (Dvα22 ◦ Dvα11 )f x+ ti
α1 =0 α2 =0 i=3

+t k
Rk1 (x, v1 , t) +t k
Rk2 (x, v1 , v2 , t)
X

= tα1 . . . tkαk Dvαkk ◦ . . . ◦ Dvα11 f (x)
P
0≤α1 ,...,αk ≤k, i αi ≤k
k
X
+ tk Rki (x, v1 , . . . , vi , t)
i=1
X P
= t i iαi Dvα f (x) + tk Rk (x, v, t),
α∈Nk ,|α|≤k
k
X X
= f (x) + tj Dvα f (x) + tk Rk (x, v, t),
P
j=1 α∈Nk , i iαi =j
Pk P P
where Rk (x, v, t) := i=1 Rki (x, v1 , . . . , vi , t) + P tj−k Dvα f (x) sat-
j>k α∈Nk , i iαi =j
isfies the remainder condition Rk (x, v, 0) = 0.
8 WOLFGANG BERTRAM, ARNAUD SOUVAY

On the other hand, we use the k-th order multi-variable radial limited expansion:
k
X k
X
f (x + i
t vi ) = f (x) + tj f hji (x, v1 , . . . , vj ; 0) + tk Sk (x, v; 0, . . . , 0, t)
i=1 j=1

Relation (2.2) now follows by uniqueness of this expansion (Theorem 2.2). 

Remark 2.9. If the integers are invertible in K, then relation (2.2) reads
X d|α| f (x, v α )
(2.3) f hji (x, v; 0) = ,
P α!
α∈Nk , i iαi =j

where α! := α1 ! . . . αk ! and v α := (v1 , . . . , v1 , . . . , vk , . . . , vk ) (vi appearing αi times).

| {z } | {z }
α1 × αk ×
dα1 f (x)(v1 ,...,v1 )
Indeed, Dvα11 f (x) = α1 !
;
iterating this, and using Theorem B.2, we get
 α 
α2 d 1 f (·)(v1 ,...,v1 )

d α1 !
(x)(v2 , . . . , v2 )
Dvα11,v,α22 f (x) = Dvα22 ◦ Dvα11 f (x) =
α2 !
α1 +α2
d f (x)(v1 , . . . , v1 , v2 , . . . , v2 )
= ,
α1 !α2 !
d|α| f (x, v α )
and so on: by induction, we have f α (x, v) = , whence (2.3). Note that
α!
these formulae are in keeping with the formulae given in [Be08], Chapter 8; however,
the methods used there are less well adapted to the case of arbitrary characteristic.
2.4. The simplicial K-jet as a scalar extension. Recall from Appendix B the
hki
definition of the (simplicial) k-jet of f , Jk f , and that J(s) and in particular Jk are
functors (Theorem B.4). They commute with direct products, and hence, applied
to the ring (K, +, m) with ring multiplication m and addition a, they yield new
hki
rings, denoted by J(s) K, resp. Jk K. These rings have been determined explicitly in
[Be10]:


J(s) K ∼ Jk K ∼
hki
(2.4) = K[X]/ X(X − s1 ) · · · (X − sk ) , = K[X]/(X k+1).

We denote the class of the polynomial X in Jk K by δ, so that 1, δ, . . . , δ k is a K-basis

of Jk K. The following facts can be proved in a conceptual way, without using the
explicit isomorphism:
Lemma 2.10. The K-algebra Jk K is N-graded, i.e., of the form
Jk K = E0 ⊕ E1 ⊕ . . . ⊕ Ek with Ej · Ei ⊂ Ei+j .
In particular, E1 ⊕ . . . ⊕ Ek is a nilpotent subalgebra.
Proof. This is a direct consequence of Theorem B.4: Jk m commutes with the canon-
ical K× -action; hence this action is by ring automorphisms. Thus the eigenspaces
Ej = {x ∈ Jk K | ∀r ∈ K× : r.x = r j x} define a grading of Jk K. 
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 9

The canonical projection

π k : Jk K → K, [P (X)] 7→ P (0)

is a ring homomorphism having a section σ k : K → Jk K, t 7→ t·1 (classes of constant

polynomials), called the canonical injection or canonical zero section. We denote
by Jk0 K = hδ, . . . , δ k i the kernel of π k (the fiber of π k : Jk K → K over 0). Note that
Jk K is again a topological ring having a dense unit group; hence we can speak of
maps that are smooth (or of class C [k] ) over this ring.

Theorem 2.11 (Simplicial Scalar Extension Theorem). If f : U → W is smooth

over K, then f admits a unique extension to Jk K-smooth map F : Jk U → Jk W :
there exists a unique map F : Jk U → Jk W (namely, F = Jk f ) such that
(1) F is smooth over the ring Jk K, and
(2) F (x) = f (x) for all x ∈ U, that is, F ◦ σU = σW ◦ f , where σU : U → Jk U
and σW : W → Jk W are the canonical injections.
More precisely, any Jk K-smooth map F : Jk U → Jk W is uniquely determined by its
restriction to the base σU (U) ⊂ Jk U.

Proof. Existence has been proved in [Be10], Theorem 2.7. Uniqueness is a conse-
quence of Theorem 2.8: for F as in the claim, we will establish an “explicit formula”
in terms of its values on the base σU (U). Let z = (v0 , . . . , vk ) = v0 +δv1 +. . .+δ k vk ∈
Jk U, with x ∈ U and vi ∈ V . Since F is smooth over the ring Jk K, we may take
t = δ and use the radial expansion of F (cf. proof of Theorem 2.8) at order k + 1:
we get
k
!
X X P
F x+ δ i vi = F (x) + δ i iαi Dvα F (x),
i=1 06=α∈Nk

where, in contrast to Theorem 2.8, no remainder term appears here, since δ k+1 = 0.
Now, it follows directly from the proof of Lemma 2.7 that Dvα F (x) is determined
by its values on the base (i.e., if F (x) = 0 for all x ∈ U, then Dvα F (x) = 0): for
non-singular values of s(i) , 1 ≤ i ≤ k, this is seen for the map Dv,s
α
F (x) where we
(i) αi
can take all s ∈ K , since the simplicial difference quotients are in terms of the
P (i) (i)
values of F at the points x + ki=1 (sji − s0 )vi ∈ U. By density, uniqueness then
also follows for singular values of s. 

Corollary 2.12. Assume P, Q : V → W are K-smooth polynomial maps. Then:

i ) Jk P : Jk V → Jk W is a Jk K-smooth Jk K-polynomial map, and coincides with
the algebraic scalar extension (cf. Appendix A, Definition C.6) PJk K of P from
K to Jk K.
ii ) The restriction Jk0 P of Jk P to the fiber Jk0 V = V ⊗K Jk0 K over 0 is again a
polynomial map, and it coincides with the algebraic scalar extension of P from
K to the (non-unital) ring Jk0 K.
iii ) Assume P (0) = Q(0). Then Jk0 P = Jk0 Q if, and only if, P ≡ Q mod (deg > k)
(i.e., P and Q coincide up to terms of degree > k).
10 WOLFGANG BERTRAM, ARNAUD SOUVAY

Proof. i) The extension PJk K of P from K to Jk K is Jk K-smooth, Jk K-polynomial

and satisfies the extension condition:
PJk K ◦ σU = σW ◦ P,
(see Theorem C.7, with A = J K). By the preceding theorem, the Jk K-smooth map
k

Jk P coincides with PJk K , and thus is Jk K-polynomial.

Item ii) is proved by the same argument. Finally, iii) follows from ii) since the
algebraic scalar extension of a polynomial whithout constant term P from K to Jk0 K
vanishes if and only if P contains no homogeneous terms of degree j = 1, . . . , k. 
2.5. Link between Taylor polynomials and simplicial jets. It follows from
Theorem 2.8 that Jk f (x, v) is polynomial in v. We are going to show that this
polynomial can be interpreted as a scalar extension of the Taylor polynomial Taykx f :
Theorem 2.13 (Scalar extension of the Taylor polynomial). Assume f, g : U → W
and h : U ′ → W ′ are of class C [k] such that f (x) = g(x) and h(U ′ ) ⊂ U. Then:
i ) Taykx f = Taykx g ⇐⇒ Jkx f = Jkx g.
ii ) The polynomial map Jkx f is the scalar extension of the Taylor polynomial Taykx f
from K to the nilpotent part Jk0 K = δK ⊕ . . . ⊕ δ k K of the ring Jk K:
Jkx f = (Taykx f )Jk0 K .
iii ) We have the following “chain rule for Taylor polynomials”:
 
Taykx (g ◦ h) = Taykh(x) g ◦ Taykx h mod (deg > k)
(where mod (deg > k) denotes the truncated composition of polynomials).
Proof. i), “⇐”: assume Jkx f = Jkx g, then
k
X
Taykx f (v) = f hii (x, v, 0; 0) = α ◦ Jkx f (v, 0) = α ◦ Jkx f ◦ κ(v)
i=1
where
α : Jk0 W = W k → W, (w1 , . . . , wk ) 7→ w1 + . . . + wk
and
κ : V → Jk0 V, v 7→ (v, 0) .
It follows that Jkx f = Jkx g implies Taykx f = Taykx g.
i), “⇒”: assume that Taykx f = Taykx g. Then for φ := f − g we have φ(x) = 0 and
Taykx φ = 0, i.e.,
∀j = 1, . . . , k, ∀v ∈ V : φhji (x, v, 0; 0) = 0.
In order to prove that Jkx φ = 0, we have to show that φhji (v0 , . . . , vj ; 0) = 0, for all
v ∈ Jk U. This is done by computing φ(x + tv1 + t2 v2 + . . . + tk vk ) in two different
ways, using first the radial limited expansion, and then the multi-variable radial
limited expansion: let w := v1 + tv2 + . . . + tk−1 vk ; since φ(x) = 0 and Tayjx φ = 0
for j = 1, . . . , k, we get
k
X
φ(x + tw) = tj φhji (x, w, 0; 0) + tk (φhki (x, w, 0; 0, t) − φhki (x, w, 0; 0))
j=0
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 11

= tk (φhki (x, w, 0; 0, t) − φhki (x, w, 0; 0)) .

On the other hand, the multi-variable limited expansion gives, with x =: v0 ,
k
X
k
φ(x + tv1 + . . . + t vk ) = tj φhji (v0 , . . . , vj ; 0) + tk (φhki (v; 0, t) − φhki (v; 0)).
j=0

By uniqueness of the radial limited expansion, φhji (v0 , . . . , vj ; 0) = 0 for j = 1, . . . , k.

(ii) Choose the origin in V such that x = 0. Now let f : U → W be of class C [k]
and let P := Tayk0 f . Since P coincides (up to the additive constant P (0) = 0) with
its own Taylor polynomial (Theorem 2.5), it follows that Tayk0 f = Tayk0 P , whence,
by (i), Jk0 f = Jk0 P , and the latter is Jk0 K-polynomial, and coincides with its algebraic
scalar extension from K to Jk0 K (Corollary 2.12). Note that the homogeneous parts
of degree > k vanish, hence Jkx f is of degree at most k.
(iii) Let R := Tayk0 (g ◦ h), P := Taykh(0) g, Q := Tayk0 h. By (i), Jk0 h = Jk0 Q and
Jkh(0) g = JkQ(0) P . Using this, and functoriality of Jk , we get
Jk0 R = Jk0 (g ◦ h) = Jkh(0) g ◦ Jk0 h = JkQ(0) P ◦ Jk0 Q = Jk0 (P ◦ Q),
whence, by Corollary 2.12, R ≡ P ◦ Q mod (deg > k). 

3. Construction of Weil functors

3.1. Weil algebras. The notion of Weil algebra has been defined in the introduc-
tion (Definition 1.1). Weil algebras form a category:
Definition 3.1. A morphism of Weil K-algebras is a continuous K-algebra homo-
morphism φ : A → B preserving the nilpotent ideals: φ(NA ) ⊂ NB . The automor-
phism group of A is denoted by AutK (A).
Lemma 3.2. The canonical projection π A : A → K of a Weil algebra is continuous,
and so is its section σ A : K → A, x 7→ x · 1. The unit group A× is open and dense
in A, and inversion A× → A is continuous.
Proof. Recall first that every element of the group Gl(n + 1, K) acts by homeo-
morphisms on Kn+1 (with product topology), hence the topology on A = K ⊕ N is
indeed independent of the chosen K-basis. The continuity of π A and of σ A is then
clear.
An element x + y ∈ A = K ⊕ N is invertible if, and only if, x is invertible in K:
indeed, its inverse is given by
Xr
−1 −1
(x + y) = x (−1)j (x−1 y)j ,
j=0

where r ∈ N is such that N r = 0. Hence A× = K× × N is open and dense in A,

and inversion is seen to be continuous since inversion in K is continuous. 
Example 3.3. The iterated tangent rings and the jet rings,
M
T kK ∼
= K[X1 , . . . , Xk ]/(X12 , . . . , Xk2 ) ∼
=K⊕ εα K ,
α∈{0,1}k ,α6=0
12 WOLFGANG BERTRAM, ARNAUD SOUVAY

k
M
Jk K ∼
= K[X]/(X k+1 ) ∼
=K⊕ δj K ,
j=1
(cf. (2.4)) are Weil K-algebras. Note that the permutations of the elements εα ,
induced by the permutation group Sk , define natural Weil algebra automorphisms
of T k K. The canonical K× -action (Appendix A) on T k K and on Jk K is also by
automorphisms. More generally, the following truncated polynomial algebras in
several variables are Weil algebras:
Wrk (K) := K[X1 , . . . , Xk ]/Ir
where I := I0 := hX1 , . . . , Xk i is the ideal generated by all linear forms and Ir :=
I r+1 is the ideal of all polynomials of degree greater than r. This is indeed a Weil
algebra: as a K-module, this quotient is the space of polynomials of degree at most
r in k variables, which is free. For k = 1, we have Wr1 (K) = Jr K, in particular,
W11 = T K. If A is any Weil algebra, and a1 , . . . , an a K-basis of N , then, if N r+1 = 0,


Wrn (K) → A = K ⊕ N , P 7→ P (0), P (a1 , . . . , an )
is well-defined and defines a surjective algebra homomorphism. Thus every Weil
algebra is a certain quotient of an algebra Wrn (K). If K is a field, then such a repre-
sentation with minimal r and n is in a certain sense unique, with n = dim(N /N 2)
and r the smallest integer with N r+1 = 0 (see [K08], Sections 1.5 – 1.7 for the real
case; the arguments carry over to a general field), but if K is not a field, this will no
longer hold (for instance, K itself may then be a Weil algebra over some other field
or ring). It goes without saying that a “classification” of Weil algebras is completely
out of reach.
Lemma 3.4. Let A = K ⊕ N and B = K ⊕ M be two Weil algebras over K.
(1) The tensor product A ⊗ B (where ⊗ = ⊗K ) is a Weil algebra over K, with
decomposition
A ⊗ B = K ⊕ (N ⊕ M ⊕ N ⊗ M) .
(2) The “Whitney sum” A ⊕K B := A ⊗ B/N ⊗ M is a Weil algebra over K,
with decomposition
A ⊕K B ∼
= K ⊕ (N ⊕ M)
(3) Both constructions are related by the following “distributive law”




A ⊗K B ⊕ B′ ∼ = A ⊗K B ⊕A A ⊗K B′ .
Proof. The tensor product of two commutative algebras is again a commutative
algebra, and we have a chain of ideals N ⊗ M ⊂ (N ⊕ M ⊕ N ⊗ M) ⊂ A ⊗ B. Since
A⊗B is again free and finite-dimensional over K, the product topology is canonically
defined on A ⊗ B and on the respective quotients. The given decompositions are
standard isomorphisms on the algebraic level, and by the preceding remarks they
are also homeomorphisms. 
Example 3.5. The tensor product T k K ⊗ T ℓ K is naturally isomorphic to T k+ℓ K.
The direct sum T K ⊕K . . . ⊕K T K (n factors) is naturally isomorphic to W1n (K) (the
Weil algebra of “n-velocities”). The Weil algebra T k K is a quotient of W1k (K).
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 13

3.2. Extended domains. As a first step towards the definition of Weil functors,
we have to define the extended domains of open sets U in a topological K-module
V . The algebraic scalar extension T A V := VA := V ⊗ A decomposes as
VA = V ⊗ (K ⊕ N ) = V ⊕ (V ⊗ N ) = V ⊕ VN ,
and, if N is homeomorphic to Kn with respect to a K-basis of N , then VN is
isomorphic, as topological K-module, to V n with product topology. The canonical
projection and its section,
πV := πVA : VA = V ⊕ VN → V, σV := σVA : V → V ⊕ VN
are continuous. More generally, for any non-empty subset U ⊂ V we define the
A-extended domain to be
T A U := (πV )−1 (U) = U × VN ⊂ VA .
For any x ∈ U, the set TxA U := T A ({x}) ∼= VN ⊂ T A U is called the fiber over x.
Let P : V → W be a K-polynomial map, of degree at most k. Let PA : VA → WA
be its scalar extension from KPto A and PN : VN → WN be its scalar extension
from K to N . That is, if P = ki=0 Pi with Pi homogeneous of degree i, then
k
X k
X
PA (v ⊗ a) = (Pi )A (v ⊗ a) = Pi (v) ⊗ ai
i=0 i=0

(cf. Appendix C). Then PA extends P in the sense that PA (v ⊗ 1) = P (v) ⊗ 1, i.e.,
PA ◦ σV = σW ◦ P.
Note that we have also P ◦ πV = πW ◦ PA . In the same way, we define PN ; mind that
there is no commutative diagram of sections, as there is no natural section K → N .
3.3. Construction of Weil functors. The following main result generalizes The-
orem 2.11 from Jk K to the case of an arbitrary Weil algebra A:
Theorem 3.6. (Existence und uniqueness of Weil functors) Let f : U → W be of
class C [∞] over K and A a Weil K-algebra. Then f extends to an A-smooth map:
there exists a map T A f : T A U → T A W such that:
[∞]
(1) T A f is of class CA ,
(2) T A f ◦ σU = σW ◦ f , i.e., T A f (x, 0) = (f (x), 0) for all x ∈ U,
(3) πW ◦ T A f = f ◦ πU .
The map T A f is uniquely determined by properties (1) and (2): if F : T A U → T A W
[∞]
is of class CA and such that F ◦ σU = σW ◦ f , then F = T A f . More generally, any
A-smooth map F : T A U → T A W is entirely determined by its values on the base
σU (U).
Proof. Let f : U → W be of class C [k] . Assume A = K ⊕ N is a Weil algebra with
N nilpotent of order k + 1. For all x ∈ U, define
TxA f := (Taykx f )N : VN → WN
to be the scalar extension from K to N of the Taylor polynomial Taykx f , and let


T A f : T A U → T A W, (x, z) 7→ f (x), TxA f (z) .
14 WOLFGANG BERTRAM, ARNAUD SOUVAY

It satisfies property (3). Since TxA f is polynomial without constant term, property
(2) of the theorem is fulfilled. In order to prove property (1), we prove first that
T A f is continuous: first of all, according to Theorem C.7 (Appendix C), since
P := Taykx f : V → W is a continuous polynomial, its scalar extension PN : VN →
WN , z 7→ (Taykx f )N (z) is continuous. By direct inspection of the proof of Theorem
C.7, one sees that the dependence on x is also continuous, i.e., (x, z) 7→ (Taykx f )N (z)
is again continuous, and hence (x, y) 7→ T A f (x, y) is continuous (cf. Remark C.9).
Next we prove the functoriality rule T A (f ◦ g) = T A f ◦ T A g. For this, we use
the “chain rule for Taylor polynomials” (Theorem 2.13, Part (iii)), together with
nilpotency of N and the fact that, if P is a polynomial containing only terms of
degree > k, then PN = 0 by nilpotency. From this we get
TxA (g ◦ f ) = (Taykx (g ◦ f ))N

= Taykf (x) g ◦ Taykx f mod (deg > k) N


= Taykf (x) g ◦ Taykx f N



= Taykf (x) g N ◦ Taykx f N
= TfA(x) g ◦ TxA f .

Thus T A is a functor. It is obviously product preserving in the sense that T A (f ×g) =

T A f × T A g.
Now we can prove that T A f is smooth over A. In fact, all arguments used
for the proof of [Be08], Theorem 6.2 (see also [Be10], Theorems 3.6, 3.7) apply:
T A (K) = K ⊗ A is again a ring, and this ring is canonically isomorphic to A
itself; the conditions defining the class C [1] over K can be defined by a commutative
diagram invoking direct products, hence, applying a product preserving functor
yields the same kind of diagram over the ring A = T A K. One gets that
(T A f )]1[,A = T A (f ]1[,K).
Since f [1] is smooth, T A (f [1],K) is continuous by the preceding steps of the proof,
hence (T A f )]1[,A admits a continuous extension (T A f )[1],A = T A (f [1],K), proving that
T A f is C [1] over A. By induction, f is then actually C [∞] over A. This proves the
existence statement.
Uniqueness is proved by adapting the method used in the proof of Theorem 2.11:
fix aPK-basis (1 = a0 , a1 , . . . , an ) of A and write an element of T A U in the form
x + ni=1 ai vi with x ∈ U and vi ∈ V . For F = T A f , we develop in a similar way as
in the proof of Theorem 2.8, replacing the scalar ti by ai (i = 1, . . . , n) and taking
k + 1-th order radial expansions:
n
X X


F x+ ai vi = F (x) + aα Dvα F (x),
i=1 06=α∈Nn

where, as in the proof of Theorem 2.11, no remainder term appears, because of the
nilpotency of a1 , . . . , an . Since x ∈ U, we have by assumption F (x) = f (x), and
since all vi ∈ V , as in the proof of Theorem 2.11, it follows that Dvα F (x) = Dvα f (x),
hence T A f is determined by its values on the base. In the same way, we can develop
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 15

any T A K-smooth map F , thus proving that F is determined by its values on the
base U. 

4. Weil functors as bundle functors on manifolds

4.1. Manifolds. Next we state the manifold-version of the preceding result. In
order to fix terminology, let us recall the definition of smooth manifolds:
Definition 4.1. Let V be a topological K-module, called the model space of the
manifold. A (smooth) K-manifold (with atlas, and modelled on V ) is a pair (M, A),
where M is a topological space and A is a K-atlas of M, i.e., an open covering (Ui )i∈I
of M, together with bijections φi : Ui → Vi := φi (Ui ) onto open subsets Vi ⊂ V ,
[∞]
such that the chart changes φij := φi ◦ φ−1
j |φj (Vji ) : Vji → Vij are of class CK , where
Vij := φi (Ui ∩ Uj ). Then K-smooth maps between manifolds are defined in the usual
way.
For our purposes, it will be useful to assume always that a manifold is given with
atlas (maximal or not). The category of K-manifolds will be denoted by ManK . If A
is a Weil K-algebra, then the category ManA of smooth A-manifolds is well-defined.
Theorem 4.2. (Weil functors on manifolds)
(1) There is a unique functor T A : ManK → ManA , which coincides on open
subsets U of topological K-modules with the assignment U 7→ T A U, f 7→ T A f
described in Theorem 3.6. Moreover, this functor is product preserving.
(2) The construction from (1) is functorial in A: if Φ : A → B is a morphism
of Weil K-algebras, then this defines canonically and in a functorial way for
all K-manifolds M a smooth map ΦM : T A M → T B M such that, for all
K-smooth maps f : M → N, we have
T B f ◦ ΦM = ΦN ◦ T A f.
Proof. Recall that a manifold is equivalently given by the following data:
• a topological K-module V (the model space),
• open sets (Vij )i,j∈I ⊂ V , where I is a discrete index set,
• K-smooth maps (φij )i,j∈I (“chart changes”) satisfying the cocycle relations:
φii = id and φij φjk = φik (where defined).
We may then define the K-manifold M to be the set of equivalence classes M :=
S/ ∼, where S := {(i, x)|x ∈ Vii } ⊂ I × V and (i, x) ∼ (j, y) if and only if
φij (y) = x, equipped with the quotient topology. Conversely, we put Vi := Vii ,
Ui := {[(i, x)], x ∈ Vi } ⊂ M and φi : Ui → Vi , [(i, x)] 7→ x to recover the previous
data.
Now we prove the existence statement in (1). The functor T A associates to the
topological K-module V , to the open sets Vij ⊂ V and to the K-smooth maps φij ,
the topological A-module T A V , the open sets T A Vij ⊂ T A V and the A-smooth maps
T A φij . If M is a K-manifold with model V and atlas (Vij , φij ), then T A V is a model
and (T A Vij , T A φij ) is an atlas of A-manifold. With those data, we have seen that
we can construct an A-manifold, denoted by T A M.
16 WOLFGANG BERTRAM, ARNAUD SOUVAY

The proof of the uniqueness statement in (1) is obvious, since a manifold is

entirely given by its model, charts domains and chart changes.
(2) For open U ⊂ V we define ΦU : V ⊗A ⊃ T A U → V ⊗B, v⊗a 7→ v⊗Φ(a). This
is a K-linear and continuous map, hence K-smooth. In particular, the collection of
maps ΦU : T A U → T B U for chart domains U defines a well-defined smooth map
ΦM : T A M → T B M.
Since ΦV commutes with scalar extension of polynomials (PB ◦ ΦV = ΦW ◦ PA ),
it also commutes with extended maps, i.e., T B f ◦ ΦM = ΦN ◦ T A f . 
Remark 4.3. If Φ : A → B is as in (2), then any B-module becomes an A-module by
r.v := Φ(r).v. In this way, the target manifold T B M can also be seen as a smooth
manifold over A, in such a way that ΦM becomes A-smooth. This remark will be
important for further developments in differential geometry (subsequent work).
4.2. Weil functors: bundle point of view. Next we state the bundle version
of the main theorem, and we give the formulation of certain operations on Weil
bundles in terms of their Weil algebras. The precise definitions of notions related
to bundles are explained after the statement of the results.
Theorem 4.4 (Weil functors as bundle functors). Let M ∈ ManK modelled on V
and A = K ⊕ N a Weil algebra such that N is nilpotent of order k + 1. Then
(1) T A M is a (A, K)-smooth polynomial bundle of degree k with section over
M and with fiber modelled on VN = V ⊗K N . More precisely, the chart
changes are polynomial in fibers of degree k and without constant term. In
particular, if N 2 = 0, then T A M is a vector bundle over M.
(2) T A : ManK → SBunAK is the unique functor into bundles with section which
coincides on open subsets U in topological K-modules with the assignment
U 7→ T A U.
(3) If Φ : A → B is a morphism of Weil algebras, then (ΦM , idM ) is a K-smooth
and intrinsically linear bundle morphism between T A M and T B M.
Theorem 4.5 (The “K-theory of Weil bundles”). Let A = K ⊕ N and B = K ⊕ M
be K-Weil algebras.
(1) The Weil functor defined by the Whitney sum A ⊕K B of two Weil algebras
(cf. Lemma 3.4) is naturally isomorphic to the Whitney sum of T A and T B
over the base manifold, i.e., for all M ∈ ManK ,
T A⊕K B M ∼
= T A M ×M T B M,
where ×M denotes the bundle product over M. By transport of structure,
this defines a structure of A ⊕K B-manifold on T A M ×M T B M.
(2) The Weil functor defined by the tensor product A ⊗K B is isomorphic to
the composition T B ◦ T A , and the typical fiber of T A⊗K B M over M is K-
diffeomorphic to
VN ⊕ VM ⊕ VN ⊗M .
(3) There is a natural bundle isomorphism T A⊗B M ∼ = T B⊗A M called the gener-
alized flip.
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 17

(4) There is a natural “distributivity isomorphism” of bundles over T A M

T A (T B M ×M T B M) ∼
′ ′
= T A T B M ×T A M T A T B M.
We stress once again that the bundles T A M and T B M are in general not vector
bundles, so that there is no natural “fiberwise notion of tensor product”. Never-
theless, there exists some relation between the bundle T A⊗B M and what one might
expect to be a “fiberwise tensor product”; this question is closely related to the
topic of connections and will be left to subsequent work. – Before turning to the
(easy) proofs of both theorems, let us give the relevant definitions:
Definition 4.6. An (A, K)-smooth fiber bundle (with atlas) is given by:
(1) a surjective K-smooth map π : E → M from an A-smooth manifold E (the
total space), onto a K-smooth manifold M (the base),
(2) a type: an operation on the left µ : G × F → F , (g, y) 7→ ρ(g)y of a group
G (the structural group) on a K-smooth manifold F (the typical fiber),
(3) a bundle atlas, which induces the condition of local triviality: there are
• a K-manifold atlas (Ui , φi )i of M, and
• A-diffeomorphisms αi : π −1 (Ui ) → Vi × F , called bundle charts, such
that the following diagram commutes:
αi
E ⊃ π −1 (Ui ) / V ×F
i
NNN
NNNπ
φ−1 ◦prVi
NNN  i
NNN
'
Ui
(4) We require the bundle charts to be compatible, that is, for all bundle chart
changes
αij := αi ◦ αj−1 : Vij × F → Vij × F,
there exist maps γij : Vij → G ( transition functions) satisfying αij (x, y) =
(φij (x), ρ(γij (x))y).
A bundle morphism between two (A, K)-smooth bundles π : E → M and π ′ : E ′ →
M ′ is, as usual, a pair of maps (Φ, φ), where Φ : E → E ′ is an A-smooth map
and φ : M → M ′ is a K-smooth map such that π ′ ◦ Φ = φ ◦ π. Finally, a bundle
with section is a fiber bundle together with a K-smooth section σ : M → E of
π : E → M, and a morphism of bundles with section is a morphism of fiber bundles
commuting with sections: Φ ◦ σ = σ ′ ◦ φ.
Note that if we fix x ∈ Vij , then the maps y 7→ ρ(γij (x))y are K-diffeomorphisms,
so that we can see G (in fact ρ(G)) as a subgroup of Diff K (F ). We do not require µ
and γij to be smooth. If it is the case (in particular, if G is a K-Lie group), then the
bundle is said to be strongly differentiable. Obviously, (A, K)-smooth bundles form
a category, denoted by BunAK . Bundles with section also form a category, denoted
by SBun.
Definition 4.7 (Polynomial bundle). Let V, W be topological K-modules.
(1) A fiber bundle with atlas is called a K-polynomial bundle of degree k if the
typical fiber is a K-module and if the structural group G acts polynomially of
18 WOLFGANG BERTRAM, ARNAUD SOUVAY

degree k on the typical fiber F (thus ρ(G) is then a subgroup of the polynomial
group GPk (F ), see Definition C.2), i.e., if the bundle chart changes αij
are K-polynomial of degree k in fibers. In particular, an affine bundle is a
polynomial bundle of degree 1. If, moreover, the bundle chart changes are
without constant term, then the bundle is a vector bundle.
(2) A map f : E → E ′ between fiber bundles with atlas is called intrinsically
K-linear (resp. K-polynomial) if the typical fibers are K-modules, if it maps
fibers to fibers and if, with respect to all charts from the given atlasses, the
chart representation of f : Ex → Ef′ (x) is K-linear (resp. K-polynomial).

Proof. (of Theorem 4.4) (1) We have seen that T A M is an A-manifold. Moreover,
the Weil algebras morphisms π A : A → K and its section σ A : K → A, t 7→ t · 1
A
induce K-smooth morphisms πM : T A M → M and its section σM A
: M → T AM
A
by Theorem 4.2. Locally, over a chart domain U, π is given by the linear map
T A U = U × VN → U. The section σ A is locally given by U → U × VN , x 7→ (x, 0).
Let us show that this bundle is indeed locally trivial, with typical fiber VN , and
that the chart changes are polynomial in fibers. The bundle atlas is given by the
K-manifold atlas (Ui , φi ) of M and by the A-diffeomorphisms αi : (π A )−1 (Ui ) =
Ui × VN → Vi × VN , (x, y) 7→ T A f (x, y) = (f (x), (Taykx f )N (y)). The maps y 7→
(Taykx f )N (y) are K-smooth polynomial, of degree at most k and without constant
term, hence define an (A, K)-smooth polynomial bundle over M. In particular, if N
is nilpotent of order 2, then T A M → M is a polynomial bundle of degree 1 without
constant term, that is, a vector bundle.
(2) follows directly from the uniqueness statement in 3.6, and (3) from 4.2. 

Proof. (of Theorem 4.5). (1) Let f : V ⊃ U → W smooth over K. The result
follows from
T A U ×U T B U = U × VN × VM = T A⊕K B U
and applying part (2) of the preceding theorem.
(2) Let f : V ⊃ U → W smooth over K. We have

T B (T A U)) = T B (U × VN ) = (U × VN ) × (V × VN )M
= U × VN × VM × VN ⊗M
= U × VN ⊕M⊕N ⊗M
= T A⊗B U
Applying twice Theorem 3.6 and noting that σTAA U ◦ σUA = σUA⊗A , it follows that

T B (T A f ) : T B (T A U) → (W ⊗K A) ⊗K B
is an extension of f that is smooth over the ring T B (T A K) = T B A = A⊗K B. Hence,
by the uniqueness statement in Theorem 4.2, T B (T A f ) = T A⊗B f .
(3) follows from the Weil algebra isomorphism A ⊗ B ∼ = B ⊗ A.
(4) follows from (1), (2), (3) and Lemma 3.4 (3). 
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 19

5. Canonical automorphisms, and graded Weil algebras

5.1. Canonical automorphisms. Concerning the action of the “Galois group”
AutK (A), Theorem 4.4 implies immediately that it acts canonically by certain intrin-
sically K-linear bundle automorphisms of T A M, called canonical automorphisms:
AutK (A) × T A M → T A M, (Φ, u) 7→ ΦM (u) .
This action commutes with the natural action of the diffeomorphism group Diff K (M):
Diff K (M) × T A M → T A M, (f, u) 7→ T A f (u) .
Here are some important examples of canonical automorphisms:
Example 5.1. For each r ∈ K× , there is a canonical automorphism T K → T K,
x + εy 7→ x + εry. The corresponding canonical map T M → T M is multiplication
by the scalar r in each tangent space. This example generalizes in two directions:
Example 5.2. By induction, the preceding example yields an action of (K× )k by
automorphisms on the iterated tangent manifold T k K. The action of the diagonal
subgroup K× then gives the canonical K× -action ρ[·] described in Appendix B.
Example 5.3. For each r ∈ K× , there is a canonical automorphism Jk K → Jk K,
P (X) 7→ P (rX). The corresponding action of K× by automorphisms is the action
ρh·i described in Appendix B. It is remarkable that, in these cases, the canonical au-
tomorphisms can be traced back to isomorphisms on the level of difference calculus
(Appendix A):
Theorem 5.4. Let r ∈ K× , s ∈ Kk and M a K-manifold with atlas modelled on
V . There is a canonical bundle isomorphism
Jk(s1 ,...,sk ) M → Jk(r−1 s1 ,...,r−1 sk ) M
given in all bundle charts by v = (v0 , . . . , vk ) 7→ r.v = (v0 , rv1 , . . . , r k vk ).
Proof. This is a restatement of the homogeneity property Theorem B.4 (ii) in terms
of the invariant language of manifolds (cf. [Be10] for notation). 
There is a similar result for the K× -action on the bundles T(t) k
M. For general
k k
automorphisms Φ of J K or T K, it seems to be difficult or even impossible to
realize them as limit cases of a families of isomorphisms in difference calculus.
Example 5.5. The map T T K → T T K, x + ε1 y1 + ε2 y2 + ε1 ε2 y12 7→ x + ε1 y2 +
ε2 y1 + ε1 ε2 y12 is an automorphism, called the flip. The corresponding canonical
diffeomorphism T T M → T T M is also called the flip (see [KMS93]). By induction,
we get an action of the symmetric group Sk on T k M (see [Be08]). Recall ([BGN04])
that the flip comes from Schwarz’s Lemma, and that the proof of Schwarz’s Lemma
in loc. cit. uses a symmetry of difference calculus in t = (t1 , t2 , t12 ) when t12 = 0. It
is not clear whether such a symmetry extends to difference calculus for all t.
In a similar way, for any Weil algebra A, the map A ⊗ A → A ⊗ A, a ⊗ a′ 7→
′
a ⊗ a is an automorphism, called the generalized flip. The corresponding canonical
diffeomorphism T A⊗A M → T A⊗A M is also called the generalized flip.
20 WOLFGANG BERTRAM, ARNAUD SOUVAY

5.2. Graded Weil algebras and their automorphisms.

Definition 5.6. A Weil algebra A = K ⊕ N is called N-graded (of length k) if it
is of the form A = A0 ⊕ . . . ⊕ Ak with free submodules Ai such that Ai · Aj ⊂ Ai+j
and A0 = K.
In [KM04], graded Weil algebras are called homogeneous Weil algebras. All ex-
amples of Weil algebras considered so far are graded – in fact, it is not so easy to
construct a Weil algebra that does not admit an N-grading (see [KM04]) – and in
Lemma 2.10 we have seen that such gradings arise naturally in differential calcu-
lus. We are interested in graded Weil algebras becauseLthey admit a “big” group
k
of automorphisms: if we denote an element a of A = i=0 Ai by (ai )0≤i≤k , where
ai ∈ Ai for all i, then obviously, for r ∈ K× ,
A → A, (ai )0≤i≤k 7→ (r i ai )0≤i≤k
defines a “one-parameter family” of automorphisms, corresponding to the derivation
A → A, (ai )0≤i≤k 7→ (iai )0≤i≤k .
This generalizes the canonical K× -action on T k K and on Jk K from Example 5.3.
But there are other canonical maps: recall that usual composition of formal power
series without constant
P term, Q, P ∈ K[[X]]0P, is given by the following explicit
formula, for Q(X) = ∞ b
n=1 n X n
and P (X) = ∞ n
n=1 an X :
∞
X n
X X
n
(5.1) Q ◦ P (X) = cn X with cn = bj aα1 · · · aαj .
n=1 j=1 α∈Nj ,|α|=n

Via the “shift” S : K[[X]] → K[[X]]0 , P (X) 7→ XP (X), we transfer this law to
K[[X]] and define a product
∞
X ∞
X ∞
X
n n −1
(5.2) Q ⋆ P (X) := ( bn X ) ⋆ ( an X ) := (S (SQ ◦ SP ))(X) = un X n
n=0 n=0 n=0
n
P P
with un = bj aα0 · · · aαj (for n = 0, this has to be interpreted as
j=0 α0 +...+αj =n−j
u0 = b0 a0 ). Formally, Q ⋆ P = X1 ((XQ) ◦ (XP )). Now, the formula for un is
compatible with graded algebras, leading to the following result:
Theorem 5.7. Let A be a graded Weil algebra of length k. Then there is a well-
defined “product” on A, given by
i
X X
b ⋆ a = (bi ) ⋆ (ai ) := (ui ), ui = bj aα0 · · · aαj .
j=0 α0 +...+αj =i−j

This product is associative and right-distributive (i.e., (b + b′ ) ⋆ a = b ⋆ a + b′ ⋆ a; in

other words, (A, +, ⋆) is a near-ring). The right-multiplication operators
Ra : A → A, b 7→ Ra (b) := b ⋆ a,
with a ∈ A, are algebra endomorphisms of the Weil algebra (A, +, ·). They are
isomorphisms if, and only if, a ∈ A× , i.e., iff a0 ∈ K× .
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 21

Proof. The product is indeed well-defined: with notation Pkfrom the theorem, we have
i
indeed ui ∈ Ai . The map Ψ : A → A[[X]], (ai )i 7→ i=0 ai X is a K-linear map
onto its image A′ := A0 ⊕ A1 X ⊕ · · · ⊕ Ak X k . It intertwines all algebraic structures
considered so far: addition +, algebra product · (here we use nilpotency of A) and
⋆ (as defined in the theorem, resp. by equation (5.2)). Therefore all claims now
follow immediately from known properties of the near-ring of formal power series
A[[X]]. 
Corollary 5.8. With assumptions and notation as in the theorem, the group (A× , ⋆)
acts by canonical and intrinsically linear automorphisms on each Weil bundle T A M,
and the monoid (A, ⋆) acts by intrinsically linear bundle endomorphisms.
Example 5.9. If a = a0 = r ∈ K× , the operators Ra give us again the canonical
K× -action.
L
Example 5.10. Let A = T k K = K ⊕ α εα K (cf. [Be08], Chapter 7), and let a such
that ai = 0 for i 6= 1 and ai = εj for a fixed j ∈ {1, . . . , k}. Then Ra is the shift
operator denoted by S0j in [Be08], Chapter 20.
Example 5.11. Similarly, for A = Jk K = K[X]/(X k+1), with a = a1 = δ, we get a
“shift” [P (X)] 7→ [P (X 2 )].
Appendix A. Difference quotient maps and K× -action
In this appendix we recall some basic definitions concerning difference calculus
from [BGN04] and [Be10], and we emphasize the fact that the group K× acts, in a
natural way, on all objects. In this appendix, K may be any commutative unital
ring and V any K-module (no topology will be used).
A.1. Domains and K× -action. Let U ⊂ V be a non-empty set, called “domain”.
We define two kinds of “extended domains”, the cubic one, denoted by U [k] and the
simplicial one, denoted by U hki for k ∈ N, which will later be used as domains of
definition of generalized kinds of tangent maps, for a given map defined on U. By
convention, U [0] := U =: U h0i .
Definition A.1 (“cubic domains”). The first extended domain of U is defined as
U [1] := {(x, v, t) ∈ V × V × K| x ∈ U, x + tv ∈ U} .
We say that U is the base of U [1] , and the maps
π [1] : U [1] → U, (x, v, t) 7→ x, and its section σ [1] : U → U [1] , x 7→ (x, 0, 0)
are called the canonical projection, resp. injection. We call
U ]1[ := {(x, v, t) ∈ U [1] | t ∈ K× }
the set of non-singular elements in the extended domain. Letting t = 0, we define
the most singular set or tangent domain
T U := {(x, v, 0) ∈ U [1] } ∼
= U × V.
By induction, we define the higher order extended cubic domains (resp., the set of
their non-singular elements) for k ∈ N⋆ by
U [k+1] := (U [k] )[1] , U ]k+1[ := (U ]k[ )]1[ ,
22 WOLFGANG BERTRAM, ARNAUD SOUVAY

[k]
and we let T k+1U := T (T k U). There are canonical projections π[j] : U [k] → U [j] ,
[k]
and their sections σ[j] : U [j] → U [k] called canonical injections, for all j ≤ k. Note
that
U [2] ⊂ (V 2 × K) × (V 2 × K) × K ∼ = V 4 × K3 ,
k k
and similarly we will consider U [k] as a subset of V 2 × K2 −1 and identify T k U with
k
U × V 2 −1 . Elements of V will be called “space variables”, and elements of K will
be called “time variables”. We separate space variables and time variables. Cor-

respondingly, we denote elements of U [k] by (v, t) = (vα )α⊂{1,...,k} , (tα
)∅6=α⊂{1,...,k} .
With this notation, we have in particular, v∅ ∈ U, and T k U = { v, 0 ∈ U [k] }.
An explicit description of the extended domains for k > 1 by conditions in terms
of sets is fairly complicated, as the number of variables growths exponentially. In
order to get a rough understanding of their structure, it is useful to note a sort of
“homogeneity property”.
Definition A.2. The zero order action of K× on U is the trivial action K× ×U → U,
(r, x) 7→ x. The canonical K× -action on the first extended domain is given by
ρ[1] : K× × U [1] → U [1] , (r, (x, v, t)) 7→ ρ[1] (r).(x, v, t) := (x, rv, r −1t).
This is indeed well-defined: x + tv ∈ U if and only if x + r −1 trv ∈ U. Moreover,
the sets U ]1[ and T U are stable under this action. Next, the canonical action of K×
on U [2] is defined as follows: write (x, u, t) = ((v∅ , v1 , t1 ), (v2 , v12 , t12 ), t2 ) ∈ U [2] with
x ∈ U [1] , u ∈ V [1] and t ∈ K such that x + tu ∈ U [1] . For r ∈ K× , let



ρ[2] (r). (v∅ , v1 , t1 ), (v2 , v12 , t12 ), t2 := ρ[1] (r).x, rρ[1] (r).u, r −1t

= (v∅ , rv1 , r −1t1 ), (rv2 , r 2 v12 , t12 ), r −1 t2
By induction, we define the canonical action ρ[k+1] : K× × U [k+1] → U [k+1] via



ρ[k+1] (r). x, u, t := ρ[k] (r).x, rρ[k] (r).u, r −1t ,
which can also be written as



ρ[k+1] (r). (vα )α , (tα )α6=∅ := (r |α| vα )α , (r |α|−2tα )α6=∅ ,
where |α| is the cardinality of the set α ⊂ {1, . . . , k + 1}. The sets U ]k+1[ and T k+1 U
are stable under this action.
The canonical projections and injections are equivariant with respect to this
action. Note that the operator ρ[k] (r) is K-linear. An element of U [k] will be called
homogeneous of degree ℓ if it is an eigenvector for all ρ[k] (r), where r ∈ K× , with
eigenvalue r ℓ . For instance, elements x in the base U are homogeneous of degree
zero. Now we define another kind of extended domain and explain its relation to
the ones considered above:
Definition A.3 (“simplicial domains”). Let U ⊂ V be non-empty, and let (in all
the following) s0 := 0. For k ∈ N⋆ , we define the extended simplicial domains
U h1i := U [1]
 , 

h2i 3 2 v0 ∈ U, v0 + (s1 − s0 )v1 ∈ U,
U := (v0 , v1 , v2 ; s1 , s2 ) ∈ V × K 
v0 + (s2 − s0 )v1 + (s2 − s1 )(s2 − s0 )v2 ∈ U
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 23

j−1
i Y
 X
U hki := (v; s) ∈ V k+1 × Kk | v0 ∈ U, ∀i = 1, . . . , k : v0 + (si − sm )vj ∈ U .
j=1 m=0

Its set of non-singular elements is

U ikh := {(v; s) ∈ U hki | ∀i 6= m : si − sm ∈ K× }.
Its set of most singular elements is
Jk U := {(v; 0) ∈ U hki } ∼
= U × V k.
For j < k, there are obvious canonical projections and injections
hki hki
πhji : U hki → U hji , and its section σhji : U hji → U hki .
For the subset of most singular elements, there are also the canonical projection
π k : J k U → U and its section σ k : U → Jk U.
Compared to the cubic case, this definition has two advantages: it is “explicit”,
and the number of variables grows linearly instead of exponentially; its drawback is
that it is not inductive. This will be overcome by imbedding the simplicial domains
into the cubic ones (Lemma A.5 below). Note that in [Be10], s0 has been considered
as a variable. Since all “simplicial formulas” invoke only differences si − sj , this
variable may be frozen to the value s0 = 0, as done here. There is an obvious
K× -action:
Definition A.4. We define the canonical action ρhki of K× on U hki by



ρhki (r). v0 , v1 , . . . , vk ; s1 , . . . , sk := v0 , rv1 , r 2v2 , . . . , r k vk ; r −1s1 , . . . , r −1sk .
It is immediately seen that ρhki (r)(v; s) ∈ U hki if and only if (v; s) ∈ U hki , and that
U ikh and Jk U are stable under this action.
Like in the cubic case, projections and injections are K× -equivariant, and we may
speak of homogeneous elements (of degree ℓ). Again, elements from the base U are
homogeneous of degree zero. An important difference with the cubic case is that
here, in contrast to the cubic case, scalars are always homogeneous of the same
degree −1. Next, we define an equivariant imbedding into the cubic domains:
Lemma A.5. The map gk : U hki → U [k] defined by gk (v; s) := (u, t) with
( (
v0 if α=∅ 1 if α = {i, i + 1}
uα = vi if α = {1, . . . , i} tα = si − si−1 if α = {i}
0 else 0 else
is a well-defined, K-affine and K× -equivariant imbedding of U hki into U [k] . More-
over, gk (U ikh ) ⊂ U ]k[ .
Proof. The fact that gk (U hki ) ⊂ U [k] is directly checked (and follows also from
the recursion procedure used in [Be10], Lemma 1.5). In order to check the K× -
equivariance gk ◦ ρhki (r) = ρ[k] (r) ◦ gk , note that on the level of space variables v,
homogeneous elements vi of degree i are sent again to homogeneous elements of
degree i (since |{1, . . . , i}| = i). On the level of time variables s, homogeneous
elements si of degree −1 are sent again to homogeneous elements of degree −1
24 WOLFGANG BERTRAM, ARNAUD SOUVAY

(since |{i}| = 1) and, for |α| = 2, the K× -action on homogeneous scalars is trivial.
Altogether, this implies the equivariance. The map gk is clearly injective: the
inverse (of its corestriction to gk (U hki ) is given by: gk −1
|gk (U hki )
(u, t) := (v; s) with
( (
v0 = u∅ s0 = 0
and Pi
vi = u{1,...,i} for all i > 1 si = j=0 t{j}

Note that this inverse is also K-affine. 

The proof shows that the scalar components tα with |α| = 2 play a special rôle
since they are the only ones that are homogeneous of degree zero; one may say that
they are a sort of “pivots”. Related to this, note that gk does not map Jk U to
T k U. The imbedding gk has been used in [Be10], Theorem 1.6. For k = 1, g1 is the
identity, and for k = 2, 3 , we have explicitly


g2 (v0 , v1 , v2 ; s1 , s2 ) = (v0 , v1 , s1 ), (0, v2 , 1), s2 − s1 ,



g3 (v0 , v1 , v2 , v3 ; s1 , s2 , s3 ) = (v0 , v1 , s1 ), (0, v2 , 1), s2 −s1 , (0, 0, 0), (0, v3, 0), 1 , s3 −s2 .

A.2. Difference calculus. Let V, W be K-modules, U ⊂ V a non-empty set and

f : U → W a map. We first define “cubic” difference quotients and then “simplicial”
ones, also called generalized divided differences. The map gk will imbed them into
the cubic calculus.
Definition A.6 (“cubic difference quotients”). The first order difference quotient
of f is the map
f (x + tv) − f (x)
f ]1[ : U ]1[ → W, (x, v, t) 7→ ,
t
and the extended tangent map is the map
T ]1[ f : U ]1[ → W ]1[ , (x, v, t) 7→ (f (x), f ]1[ (x, v, t), t).
The higher order cubic difference quotients and higher order extended tangent maps
are defined by induction

]1[
f ]k+1[ := f ]k[ : U ]k+1[ → W

]1[
T ]k+1[f := T ]k[ : U ]k+1[ → W ]k+1[.
In [Be10], explicit formulae for f ]2[ and T ]2[ f are given; they are quite complicated
and not very useful. Here are two main properties of this construction:
Theorem A.7. Let f : U → W and g : U ′ → W ′ with f (U) ⊂ U ′ . Then
i ) Functoriality: T ]k[(g ◦ f ) = T ]k[ g ◦ T ]k[ f and T ]k[idU = idT ]k[ U .
ii ) Homogeneity: for all r ∈ K× , T ]k[f ◦ ρ[k] (r) = ρ[k] (r) ◦ T ]k[f .
Proof. (i) For k = 1, this is an easy computation, and for k > 1, it follows immedi-
ately by induction (see [BGN04]). (ii) For k = 1:
 
−1 f (x + r −1 trv) − f (x) −1

]1[
T f (x, rv, r t) = f (x), −1
, r t = f (x), rf ]1[(x, v, t), r −1 t
r t
and for k > 1, the result follows by induction. 
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 25

Now we come to simplicial difference calculus and to its imbedding into cubic
calculus. In the following, recall that, by definition, s0 = 0.
Definition A.8 (“simplicial difference quotients”). For a map f : U → W we
define (generalized) divided differences f ikh : U ikh → W by
f (v0 ) f (v0 + (s1 − s0 )v1 )
f i1h (v0 , v1 ; s1 ) := f ]1[ (v0 , v1 , s1 ) = +
s0 − s1 s1 − s0
f (v 0 ) f (v0 + (s 1 − s0 )v1 )
f i2h (v0 , v1 , v2 ; s1 , s2 ) := + +
(s0 − s1 )(s0 − s2 ) (s1 − s0 )(s1 − s2 )
f (v0 + (s2 − s0 )v1 + (s2 − s1 )(s2 − s0 )v2 )
,
(s2 − s0 )(s2 − s1 )
P Qj−1

ikh f (v0 ) Xk
f v0 + ij=1 m=0 (si − sm )vj
f (v; s) := Q + Q .
j=1,...,k (s0 − sj ) i=1 j=0,...,î,...,k (si − sj )

Define the extended k-jet by Jikh f : U ikh → W ikh ,

(v; s) 7→ Jikh f (v; s) := f (v0 ), f i1h (v0 , v1 ; s1 ), . . . , f ikh (v0 , . . . , vk ; s1 , . . . , sk ); s).
Theorem A.9. The map gk : U hki → U [k] defines an imbedding of simplicial into
cubic difference calculus in the sense that, for all f : U → W , we have
Jikh f
U ikh −→ W ikh
T ]k[f ◦ gk = (−1)k gk ◦ Jikh f : gk ↓ (−1)k gk ↓
T ]k[ f
U ]k[ −→ W ]k[
Proof. See [Be10], Lemma 1.5 and Theorem 1.6. 
Theorem A.10. Let f : U → W and g : U ′ → W ′ with f (U) ⊂ U ′ . Then
i ) Functoriality: Jikh (g ◦ f ) = Jikh g ◦ Jikh f and Jikh idU = idJikh U .
ii ) Homogeneity: for all r ∈ K× , Jikh f ◦ ρhki (r) = ρhki (r) ◦ Jikh f .
Proof. Both statements can be seen as a consequence of Theorems A.7 and A.9
above. An independent proof of (i) is given in [Be10], Theorem 2.10, and (ii) can
also be proved by an easy direct computation. 

Appendix B. Differential calculi

Differential calculus is the extension of difference calculus to singular values. In
order to do this, we need additional structure, such as, e.g., topology. We therefore
assume that K is a topological ring such that its unit group K× is open dense
in K, and we assume that V, W are topological K-modules, U ⊂ V is open and
f : U → W is a continuous map. The class of continuous maps will be denoted by
C 0 (see [BGN04] for other “C 0 -concepts”). There are two concepts of differential
calculus, which we call “cubic” and “simplicial”.
[1]
Definition B.1 (“cubic differentiability”). We say that f : U → W is of class CK
(or just C [1] if the base ring K is clear from the context) if there exists a continuous
map f [1] : U [1] → W extending the first order difference quotient map: for all
26 WOLFGANG BERTRAM, ARNAUD SOUVAY

(x, v, t) ∈ U ]1[ , we have f [1] (x, v, t) = f (x+tv)−f

t
(x)
. By density of K× in K, the map
f [1] is unique if it exists, and so is the value
df (x)v := f [1] (x, v, 0) =: ∂v f (x).
The extended tangent map is then defined by
[1]

T [1] f : U [1] → W [1] , (x, v, t) 7→ T [1] f (x, v, t) =: T(t) f (x, v) := f (x), f [1] (x, v, t), t .
[k]
The classes CK (or shorter: just C [k] ) are defined by induction: we say that f is
of class C [k+1] if it is of class C [k] and if f [k] : U [k] → W is again of class C [1] ,
where f [k] := (f [k−1])[1] . The higher order extended tangent maps are defined by
T [k+1] f := T [1] (T [k] f ), and the k-th order cubic differentials, at x ∈ U, are defined
by dk f (x) : V k → W, (v1 , . . . , vk ) 7→ ∂v1 . . . ∂vk f (x).
Theorem B.2. Let f : U → W and g : U ′ → W ′ of class C [k] with f (U) ⊂ U ′ .
i ) Functoriality: T [k](g ◦ f ) = T [k] g ◦ T [k] f and T [k]idU = idT [k] U .
ii ) Homogeneity: for all r ∈ K× , T [k]f ◦ ρ[k] (r) = ρ[k] (r) ◦ T [k]f .
iii ) Linearity: the differential df (x) : V → W is continuous and linear.
iv ) Symmetry: the k-th order cubic differential map dk f (x) : V k → W is contin-
uous, k times multilinear and symmetric.
Proof. (i) and (ii) follow “by density” from Theorem A.7. Homogeneity of the
differential is a special case of (ii), and additivity is proved in a similar way (see
[BGN04]). (iv) is a direct consequence of (iii) and of Schwarz’ Lemma ([BGN04]).

[1]
Functoriality is equivalent to saying that, for t ∈ K fixed, T(t) is a functor, and
[1]
for t = 0 this gives the usual chain rule. Moreover, for each t, the functor T(t)
[1] [1] [1]
commutes with direct products: T(t) (f × g) is naturally identified with T(t) f × T(t) g.
[k]
Analogously, for fixed time variables t, T(t) are functors commuting with direct
[k]
products. In particular, for t = 0, T(0) is a functor, called iterated tangent functor
and denoted by T k . Note that T 1 =: T is the usual tangent functor. From this,
[k]
we deduce that T [k] K, T(t) K and T k K are again rings, with product and addition
obtained by applying the functor to product and addition in K. See [Be10] for more
information on these rings.
hki
Definition B.3 (“simplicial differentiability”). We say that f is of class CK , or
just of class C hki , if, for all 1 ≤ ℓ ≤ k, there are continuous maps f hℓi : U hℓi → W
extending f iℓh . Note that, by density of K× in K, the extension f hℓi is unique (if it
exists), and hence in particular the value f hℓi (v; 0), called the ℓ-th order simplicial
differential is uniquely determined. We define also
Jhki f : U hki → W hki , (v; s) 7→ (f (v0 ), f h1i (v0 , v1 ; s1 ), . . . , f hki (v; s); s),
and, for any fixed element s ∈ Kk , we define the simplicial s-extension of f by
hki hki
J(s) f : J(s) U → W k+1, v 7→ Jhki f (v; s) ,
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 27

hki 
where J(s) U := v ∈ V k+1 | (v; s) ∈ U hki . The simplicial k-jet of f is
hki

Jk f := J(0) f : Jk U → Jk W, v 7→ Jk f (v) = f hℓi (v0 , . . . , vℓ ; 0) ℓ=0,...,k .
where f h0i := f by convention.
Theorem B.4. Let f : U → W and g : U ′ → W ′ of class C hki with f (U) ⊂ U ′ .
i ) Functoriality: Jhki (g ◦ f ) = Jhki g ◦ Jhki f and Jhki idU = idJhki U .
ii ) Homogeneity: for all r ∈ K× , Jhki f ◦ ρhki (r) = ρhki (r) ◦ Jhki f .
This follows “by density” from Theorem A.10. As in the cubic case, for fixed time
hki
variables s, J(s) is a functor commuting with direct products. From this, it follows
hki
as above that Jhki K, J(t) K and Jk K are again rings. Again, we refer to [Be10] for
hki
more information on these rings. In particular, for s = 0, J(0) is a functor called
the k-th order jet functor, denoted by Jk . Note that J1 = T 1 is the usual tangent
functor, and that for s = 0, functoriality gives equation (1.1). According to [Be10],
Theorem 1.7 and Corollary 1.11, there is an equivalent characterization of the class
C hki in terms of “limited expansions”, having the advantage that no denominator
terms appear:
Theorem B.5. A map f : U → W is of class C hki if, and only if the following
simplicial limited expansions hold: for all 1 ≤ ℓ ≤ k, there exist continuous maps
f hℓi : U hℓi → W such that, whenever (v, s) ∈ U hℓi ,


f v0 + (s1 − s0 )v1 ) = f (v0 ) + (s1 − s0 )f h1i (v0 , v1 ; s0 , s1 )


f v0 + (s2 − s0 )(v1 + (s2 − s1 )v2 ) = f (v0 ) + (s2 − s0 )f h1i (v0 , v1 ; s0 , s1 )+
(s2 − s1 )(s2 − s0 )f h2i (v0 , v1 , v2 ; s0 , s1 , s2 )
..
! .
j−1
k Y j−1
k Y
X X
f v0 + (sk − sℓ )vj = f (v0 ) + (sk − sℓ )f hji (v0 , . . . , vj ; s0 , . . . , sj )
j=1 ℓ=0 j=1 ℓ=0

The maps f hℓi defined by this condition coincide with those defined in the definition
above.
Theorem B.6 (“cubic implies simplicial”). If f is of class C [k] , then f is of class
C hki , and the map gk : U hki → gk (U hki ) ⊂ U [k] defines a smooth imbedding of
simplicial into cubic differential calculus in the sense of Theorem A.9.
Proof. This follows “by density” from Theorem A.9 (see [Be10], Theorem 1.6). 
We conjecture that also “simplicial implies cubic”, i.e., if f is C h∞i , then it is
also of class C [∞] , but at present this conjecture is not settled. Therefore we will
work throughout with a C [k] -assumption.
Corollary B.7. If f is of class C [k] , then f hji is of class C [k−j] , for all j ≤ k.
Proof. Via the imbedding gj : U hji → U [j] , we can consider f hji : U hji → W as a
partial map of f [j] : U [j] → W . The latter is of class C [k−j], hence the former is also
of class C [k−j] , and by the preceding theorem thus also of class C hk−ji . 
28 WOLFGANG BERTRAM, ARNAUD SOUVAY

Appendix C. Continuous polynomial maps, and scalar extensions

The purpose of this appendix is to show that continuous K-polynomial mappings
are K-smooth, and that their scalar extensions by Weil K-algebras A are again
continuous, hence smooth over A.
C.1. Continuous (multi-)homogeneous maps. As in the main text, V, W are
topological modules over a topological ring K.
P
Theorem C.1 (separation of homogeneous parts). Assume P = ki=0 Pi is a sum
of K-homogeneous maps Pi : V → W of degree i (i.e., for all r ∈ K, Pi (rx) =
r i Pi (x)). Then the following are equivalent:
(1) The map P : V → W is continuous.
(2) For i = 0, . . . , k, the homogeneous part Pi : V → W is continuous.
P
Assume P = α∈Nn Pα : V n → W is a finite sum of multi-homogeneous maps
Pα : V n → W , with α = (α1 , . . . , αn ), i.e., for all r ∈ K and 1 ≤ i ≤ n,
Pα (x1 , . . . , rxi , . . . , xn ) = r αi Pα (x1 , . . . , xn ). Then the following are equivalent:
(1) The map P : V n → W is continuous.
(2) For all α ∈ Nn , the multi-homogeneous part Pα : V n → W is continuous.
Proof. We prove the first equivalence. Obviously, (2) implies (1). Let us prove
the converse. Assume that P is continuous. Since P0 is constant, it is continuous.
Without loss of generality, we may assume that P0 = 0. Fix scalars r1 , . . . , rk ∈ K
and define continuous maps (depending on these scalars)
k
X
Q1 (x) := r1 P (x) − P (r1 x) = (r1 − r1i )Pi (x),
i=2

Qi+1 (x) := (ri+1 )i+1 Qi (x) − Qi (ri+1 x).

Then Qi is a linear combination of Pi+1 , . . . , Pk , and in particular, we find that
Qk−1 (x) = λPk (x) with
λ = (r1 − r1k )(r22 − r2k ) · · · (rk−1
k−1 k
− rk−1 ).
In order to prove that Pk is continuous, it suffices to show that we may choose
r1 , . . . , rk−1 ∈ K such that the scalar λ is invertible, because then we have Pk (x) =
λ−1 Qk−1 (x). Since Qk−1 is continuous by construction, it then follows that Pk is
continuous.
To prove our claim, write rii − rik = rii (1 − rim ) with m = k − i. Since the map
f : K → K, r 7→ 1 − r m is continuous and K× is open, U := f −1 (K× ) is open, and
U is non-empty since 0 ∈ U. Since K× is open and dense, U ′ := U ∩ K× is open
and non-empty, and we may choose ri ∈ U ′ . Doing so for all i, we get an invertible
scalar λ.
Having proved that Pk is continuous, we replace P by P − Pk and show as above
that Pk−1 is continuous, and so on for all homogeneous parts.
The second equivalence is proved similarly: proceed as above with respect to the
first variable x1 in order to separate terms according to their degree in x1 , then use
the same procedure with respect to the second variable x2 , and so on. 
 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 29

C.2. Continuous polynomial maps. The following general definition of a K-

polynomial map, given in [Bou], ch. 4, par. 5, has been used in [BGN04] (loc. cit.,
Appendix A, Def. A.5):
Definition C.2. A map p : V → W between K-modules V and W is called homo-
geneous polynomial of degree k if, for any system (ei )i∈I of generators of
P V , there
exist coefficients aα ∈ W (where α : I → N has finite support and |α| := i αi = k)
such that !
X X Y
p ti ei = tα aα where tα := tαi i .
i∈I α i
If V is free (in particular, if K is a field), then this is equivalent to saying that there
exists a k times multilinear map m : V k → W such that p(x) = m(x, . . . , x). In
any case, a polynomial map is a finite sum of homogeneous polynomial maps.
The set of polynomial maps p : V → W is denoted by Pol(V, W ), and we let
Pol(V, W )0 := {p : V → W polynomial | p(0) = 0}. If V = W , the set of
polynomial maps p : V → V having an inverse polynomial map q : V → V forms a
group denoted by GP(V ), called the general polynomial group of V . By GP(V )0 we
denote the stabilizer subgroup of 0.
Note that, if p(x) = m(x, . . . , x), then continuity of p does not always imply
continuity of m (if the integers are invertible in K, then, by polarization, we may
find symmetric and continuous m).
Theorem C.3. Every continuous K-polynomial map p : V → W is K-smooth.
Proof. Assume p : V → W is continuous polynomial. By theorem C.1, we may
assume without loss of generality that p is homogeneous of degree k. Assume first
that p(x) = m(x, . . . , x) with multilinear m : V k → W . Since f is continuous,
F : V × V → W, (x, v) 7→ p(x + v) = m(x + v, . . . , x + v) .
is continuous and polynomial. Using multilinearity, we expand
k
X
(C.1) p(x + v) = m(x + v, . . . , x + v) = M k−i,i (x, v),
i=0
k−i,i
where M (x, v) is the sum of all terms in the expansion of m containing i times
the argument v and k − i times the argument x. The i-th term in (C.1) is the
homogeneous part of bi-degree (k − i, i) of the continuous map F , and hence, by
theorem C.1, is again a continuous function of (x, v). Now, for t ∈ K× , we get by a
similar expansion as above
k
[1] f (x + tv) − f (x) X i−1 k−i,i
p (x, v, t) = = t M (x, v),
t i=1

and since M k−i,i (x, v) is continuous, the right hand side defines a continuous func-
tion of (x, v, t), proving that a continuous extension of the difference quotient func-
tion exists, and hence p is C [1] . Moreover, p[1] is again continuous polynomial (of
total degree at most 2k − 1), hence by induction it follows that p is of class C [∞] .
30 WOLFGANG BERTRAM, ARNAUD SOUVAY

If p(x) is not directly given by a multilinear map m, then fix a system of generators
(ei ) of V and consider the free module E spanned by the ei , together with its
canonical surjection E → V . The map F lifts to map F̃ : E 2 × K → W , which
we may decompose as above, giving rise to a map m̃ and to maps M̃ k−i,i . Passing
again to the quotient, we see that the bi-homogeneous components M k−i,i are still
continuous maps, and p[1] can be extended continuously as above. 
Definition C.4. Assume p : V → W is a polynomial map of the form p(x) =
k
Pn m : V → W is K-multilinear. For
m(x, . . . , x) where n ∈ N, α = (α1 , . . . , αn ) ∈
N with |α| := i=1 αi = k and v = (v1 , . . . , vn ) ∈ V n , let
n

M α (v) := m(v1 : α1 ; . . . ; vn : αn )
be the sum of all terms m(w1 , . . . , wk ) with exactly αi among w1 , . . . , wk equal to vi .
Lemma C.5. Assume p : V → W is a polynomial map of the form p(x) =
m(x, . . . , x) where m : V k → W is K-multilinear. Then, if p is continuous, so
is the map
M α : V n → W, v = (v1 , . . . , vn ) 7→ M α (v) ,
and it does not depend on the choice of m, so that we may write pα (v) := M α (v).
Proof. This follows as above, by considering the continuous map
n
! n n
!
X X X
F : V n → W, v 7→ p vi = m vi , . . . , vi
i=1 i=1 i=1
α
whose α-homogeneous component is precisely M . 
C.3. Scalar extensions. A scalar extension of K is, by definition, a commutative
and associative K-algebra A. If A is unital, then there is a natural map K → A,
t 7→ t · 1.
Definition C.6. Let p : V → W be a K-polynomial map, homogeneous of degree
k. If p(x) = m(x, . . . , x) with multilinear m : V k → W , then the scalar extension
from K to A of p is the map
pA : VA = V ⊗K A → WA , v ⊗ a 7→ P (v) ⊗ ak = mA (av, . . . , av),
where mA : (VA )k → WA is the multilinear map defined by the universal propery of
the tensor product. If V is not free, let as above E → V be the surjection defined
by a system of generators, define P̃A : EA → WA ; then this map passes to V as a
map PA : VA → WA .
Theorem C.7. Assume K is a topological ring with dense unit group, and A a
scalar extension of K which is a topological K-algebra, homeomorphic to Kn with
respect to some K-basis a1 , . . . , an . Assume p : V → W is continuous polynomial
over K. We equip VA with the product topology with respect to the decomposition
Mn n
M
VA = V ⊗K ( Kai ) = (V ⊗ ai ),
i=1 i=1

and similarly for WA . Then pA : VA → WA is a continuous A-polynomial map.

 A GENERAL CONSTRUCTION OF WEIL FUNCTORS 31

Proof. Note that the topology on VA does not depend on the choice of the K-basis
of A since the group GLn (K) acts by homeomorphisms. Assume first that p is
homogeneous of degree k and of the form p(x) = m(x, . . . , x). Then we have:
n
! n n
!
X X X
pA vi ⊗ ai = mA vi ⊗ ai , . . . , vi ⊗ ai
i=1 i=1 i=1
n
X
= m(vj1 , . . . , vjk ) ⊗ aj1 · · · ajk
j1 ,...,jk =1
X
= M α1 ,...,αn (v1 , . . . , vn ) ⊗ aα1 1 · · · aαnn ,
Pn
(α1 ...,αn )∈Nn , i=1 αi =k
X
= M α (v) ⊗ aα
α∈Nn ,|α|=k

According to Lemma C.5, the map M α is continuous, and hence pA is continuous.

If p is not of the form p(x) = m(x, . . . , x), then we use similar arguments as at the
end of the proof of Theorem C.3. 
Remark C.8. If A is unital, then we have a natural injection σ A : V → V A , v 7→
v ⊗K 1A , and pA “extends” p in the sense that pA ◦ σ A = σ A ◦ p.
Remark C.9. If p is as in the theorem, depending moreover continuously on a
parameter y (say, p(x) = py (x), jointly continuous in (x, y)), then, since M α does
not depend on the choice of m (see lemma C.5), the proof of the theorem shows
that (y, z) 7→ (py )A (z) is again jointly continuous in (y, z).

References
[Be08] Bertram, W., Differential Geometry, Lie Groups and Symmetric Spaces over General
Base Fields and Rings, Memoirs of the AMS, volume 192, number 900 (2008). arXiv:
math.DG/0502168
[Be10] Bertram, W., “Simplicial differential calculus, divided differences, and construction of
Weil functors”, to appear in Forum Math., arxiv : math.DG/1009.2354
[Be11] Bertram, W., Calcul différentiel topologique élémentaire, Calvage et Mounet, Paris 2011;
voir http://www.iecn.u-nancy.fr/˜bertram/livre.pdf
[BGN04] Bertram, W., H. Gloeckner and K.-H. Neeb, “Differential Calculus over general base
fields and rings”, Expo. Math. 22 (2004), 213 –282.
[Bou] Bourbaki, N., Algèbre. Chapitres 4-5, Hermann, Paris 1971
[Ko81] Kock, A., Synthetic Differential Geometry, London Math. Soc. Lecture Notes 51, Cam-
bridge, 1981.
[KMS93] Kolar, I., P. Michor and J. Slovak, Natural Operations in Differential Geometry,
Springer, Berlin, 1993.
[K08] Kolar, I., “Weil bundles as generalized jet spaces”, pp. 625-664 in: D. Krupka and D.
Saunders (eds.): Handbook of Global Analysis, Elsevier, Amsterdam, 2008.
[K00] Kureš, M., “On the simplicial structure of some Weil bundles”, Rend. del Circ. Mat.
Palermo 63 (2000), 131 – 140.
[KM04] Kureš, M., and W.M. Mikulski, “Natural operators lifting vector fields to bundles of
Weil contact elements”, Czechoslowak Math. J. 54 (129) (2004), 855-867
[Re83] Reinhart, B.L., Differential Geometry of Foliations – The Fundamental Integrability
Problem, Springer, Berlin 1983
32 WOLFGANG BERTRAM, ARNAUD SOUVAY

[Weil] Weil, A., “Théorie des points proches sur les variétés différentiables”, Colloque de
Géom. Diff., Strasbourg 1953, pp. 111–117 (dans : A. Weil, Œuvres scientifiques, Vol.
2, Springer, New York 1980, pp. 103–109)
[White] White, J.E., The Method of Iterated Tangents with Applications in Local Riemannian
Geometry, Pitman, Boston 1982.

Institut Élie Cartan Nancy, Nancy-Université, CNRS, INRIA, Boulevard des

Aiguillettes, B.P. 239, F-54506 Vandœuvre-lès-Nancy, France
E-mail address: [email protected], [email protected]