Ega 4

LOCAL STUDY OF SCHEMES AND THEIR MORPHISMS (EGA IV)
build hack [CC]
C ONTENTS
Summary 2
1. Relative finiteness conditions. Constructible sets in preschemes 3
1.1 Quasi-compact morphisms 3
2. Base change and flatness 4
3. Associated prime cycles and primary decomposition 4
4. Change of base field for algebraic preschemes 4
5. Dimension, depth, and regularity in locally Noetherian preschemes 4
6. Flat morphisms of locally Noetherian preschemes 4
7. Relations between a local Notherian ring and its completion. Excellent rings 4
8. Projective limits of preschemes 4
9. Constructible properties 4
10. Jacobson preschemes 4
11. Topological properties of finitely presented flat morphisms. Flatness criteria 4
12. Study of fibres of finitely presented flat morphisms 4
13. Equidimensional morphisms 4
14. Universally open morphisms 4
15. Study of fibres of a universally open morphism 4
16. Differential invariants. Differentially smooth morphisms 4
16.1 Normal invariants of an immersion 4
16.2 Functorial properties of the normal invariants of an immersion 7
16.3 Fundamental differential invariants of morphisms of preschemes 11
16.4 Functorial properties of differential invariants 13
16.5 Relative tangent sheaves and bundles; derivations. 21
16.6 Sheaf of p-differentials and exterior differentials. 26
16.7 The PX/Sn (F ). 28
16.8 Differential operators. 30
16.9 Regular and quasi-regular immersions 35
16.10 Differentially smooth morphisms. 39
16.11 Differential operators on a differentially smooth S-prescheme 40
16.12 Case of characteristic zero: Jacobian criterion for differentially smooth morphisms. 42
17. Smooth morphisms, unramified (or net) morphisms, and étale morphisms. 42
17.1 Formally smooth morphisms, formally unramified morphisms, formally étale
morphisms. 42
17.2 General properties of differentials 45
17.3 Smooth morphisms, unramified morphisms, étale morphisms 46
17.4 Characterizations of unramified morphisms. 48
18. Supplement on étale morphisms. Henselian local rings and strictly local rings 48
19. Regular immersions and normal flatness 48
20. Meromorphic functions and pseudo-morphisms 48
20.0 Introduction 48
20.1 Meromorphic functions 48
20.2 Pseudo-morphisms and pseudo-functions 52
21. Divisors 52
References 52
1
LOCAL STUDY OF SCHEMES AND THEIR MORPHISMS (EGA IV) 2
S UMMARY
IV-1 | 222
§1. Relative finiteness conditions. Constructible sets in preschemes.
§2. Base change and flatness.
§3. Associated prime cycles and primary decomposition.
§4. Change of base field for algebraic preschemes.
§5. Dimension, depth, and regularity for locally Noetherian preschemes.
§6. Flat morphisms of locally Noetherian preschemes.
§7. Relations between a local Noetherian ring and its completion. Excellent rings.
§8. Projective limits of preschemes.
§9. Constructible properties.
§10. Jacobson preschemes.
§11.1 Topological properties of finitely presented flat morphisms. Flatness criteria.
§12. Study of fibres of finitely presented flat morphisms.
§13. Equidimensional morphisms.
§14. Universally open morphisms.
§15. Study of fibres of a universally open morphism.
§16. Differential invariants. Differentially smooth morphisms.
§17. Smooth morphisms, unramified (or net) morphisms, and étale morphisms.
§18. Supplement on étale morphisms. Henselian local rings and strictly local rings.
§19. Regular immersions and normal flatness.
§20. Meromorphic functions and pseudo-morphisms
§21. Divisors.
The subjects discussed in the chapter call for the following remarks. IV-1 | 223
(a) The common property of all the subjects discussed is that they all related to local properties
of preschemes or morphisms, i.e. considered at a point, or the points of a fibre, or on a
(non-specified) neighbourhood of a point or of a fibre. These properties are generally of a
topological, differential, or dimensional nature (i.e. bringing the ideas of dimension and depth into
play), and are linked to the properties of the local rings at the points considered. One type of
problem is the relating, for a given morphisms f : X → Y and point x ∈ X, of the properties
of X at x with those of Y at y = f ( x ) and those of the fibre Xy = f −1 (y) at x. Another is the
determining of the topological nature (for example, the constructibility, or the fact of being
open or closed) of the set of points x ∈ X at which X has a certain property, or for which
the fibre X f ( x) passing through x has a certain property at x. Similarly, we are interested in
the topological nature of the set of points y ∈ Y such that X has a certain property at all the
points of the fibre Xy , or those such that this fibre itself has a certain property.
(b) The most important idea for the following chapters is that of flat morphisms of finite presentation,
as well as the particular cases of smooth morphisms and étale morphisms. Their detailed study
(as well as that of connected questions) really starts in §11.
(c) Sections §§1–10 can be considered as being preliminary in nature, and as developing three
types of techniques, used, not only in the other sections of the chapter, but also, of course, in
the follow chapters:
(c1) Sections §§1–4 are envisaged as treating the diverse aspects of the idea of change of base,
above all in relation with the conditions of finiteness or flatness; we there initiate the
technique of descent, with its most elementary aspects (the questions of “effectiveness”
linked to this technique will be studied in Chapter V).
(c2) Sections §§5–7 are focused on what we may call Noetherian techniques, since the
preschemes considered are always locally Noetherian, whereas, on the contrary, there is
generally no finiteness condition imposed on the morphisms; this is essentially due to
1The order and content of §§11–21 are given only as an indication of what the titles will be, and will possibly be modified
before their publication. [Trans.] This was indeed the case: many of §§11–21 ended up having entirely different titles or content. See
here.
the fact that the ideas of dimension and depth are hardly manageable except in the case
of Noetherian local rings. Recall that §7 constitutes a “delicate (?)” theory of Noetherian
local rings, not much used in what follows in the chapter.
(c3) Sections §§8–10 describe, amongst other things, the means of eliminating the Noetherian
hypotheses on the preschemes considered, by substituting such hypotheses for suitable
ones of finiteness (“finite presentation”) on the morphisms considered: the advantage of
this substitution is that the latter such hypotheses (those of finiteness on the morphisms)
are stable under base change, which is not the case for the Noetherian hypotheses on
the preschemes. The technique permitting this substitution relies, in some part, on the
use of the idea of the projective limit of preschemes, thanks to which we can reduce a
question to the same question with Noetherian hypotheses; on the other hand, it relies
on the systematic use of constructible sets, which have the double interest of being
preserved under taking inverse images (of arbitrary morphisms) and by direct images IV-1 | 224
(of morphisms of finite presentation), and having manageable topological properties
in locally Noetherian preschemes. The same techniques often even allow to restrict to
the case of more specific Noetherian rings, for example the Z-algebras of finite type, and
it is here that the properties of “excellent” rings (studied in §7) intervene in a decisive
manner. Independently of the question of elimination of Noetherian hypotheses, the
techniques of §§8–10, elementary in nature, find constant use in nearly all applications.
§1. R ELATIVE FINITENESS CONDITIONS . C ONSTRUCTIBLE SETS IN PRESCHEMES

In this section, we will resume the exposé of “finiteness conditions” for a morphism of preschemes
f : X → Y given in (I, 6.3 and 6.6). There are essentially two notions of “finiteness” of a global nature
on X, that of quasi-compact morphism (defined in (I, 6.6.1)) and that of a quasi-separated morphism; on
the other hand, there are two notions of “finiteness” of a local nature on X, that of a morphism locally
of finite type (defined in (I, 6.6.2)) and that of a morphism locally of finite presentation. By combining
these local notions with the preceding global notions, we obtain the notion of a morphism of finite
type (defined in (I, 6.3.1)) and of a morphism of finite presentation. For the convenience of the reader,
we will give again in this section the properties stated in (I, 6.3 and 6.6), referring to their labels in
Chapter I for their proofs.
In nos 1.8 and 1.9, we complete, in the context of preschemes, and making use of the previous
notions of finiteness, the results on constructible sets given in (0III , §9).
1.1. Quasi-compact morphisms
Definition (1.1.1). — We say that a morphism of preschemes f : X → Y is quasi-compact if the

continuous map f from the topological space X to the topological space Y is quasi-compact (0, 9.1.1),
in other words, if the inverse image f −1 (U ) of every quasi-compact open subset U of Y is quasi-
compact (cf. (I, 6.6.1)).
If B is a basis for the topology of Y consisting of affine open sets, then for f to be quasi-compact,
it is necessary and sufficient that for all V ∈ B, f −1 (V ) is a finite union of affine open sets. For example,
if Y is affine and X is quasi-compact, every morphism f : X → Y is quasi-compact (I, 6.6.1).
If f : X → Y is a quasi-compact morphism, then it is clear that for every open subset V of Y, the
restriction of f to f −1 (V ) is a quasi-compact morphism f −1 (V ) → V. Conversely, if (Uα ) is an open
cover of Y and f : X → Y is a morphism such that the restrictions f −1 (Uα ) → Uα are quasi-compact,
then f is quasi-compact. As a result, if f : X → Y is an S-morphism of S-preschemes, and if there IV-1 | 225
exists an open cover (Sλ ) of S such that the restrictions g−1 (Sλ ) → h−1 (Sλ ) of f (where g and h are
the structure morphisms) are quasi-compact, then f is quasi-compact.
§2. B ASE CHANGE AND FLATNESS

§3. A SSOCIATED PRIME CYCLES AND PRIMARY DECOMPOSITION
§4. C HANGE OF BASE FIELD FOR ALGEBRAIC PRESCHEMES
§5. D IMENSION , DEPTH , AND REGULARITY IN LOCALLY N OETHERIAN
PRESCHEMES
§6. F LAT MORPHISMS OF LOCALLY N OETHERIAN PRESCHEMES

§7. R ELATIONS BETWEEN A LOCAL N OTHERIAN RING AND ITS COMPLETION .
E XCELLENT RINGS
§8. P ROJECTIVE LIMITS OF PRESCHEMES
§9. C ONSTRUCTIBLE PROPERTIES
§10. J ACOBSON PRESCHEMES
§11. T OPOLOGICAL PROPERTIES OF FINITELY PRESENTED FLAT MORPHISMS .
F LATNESS CRITERIA
§12. S TUDY OF FIBRES OF FINITELY PRESENTED FLAT MORPHISMS
§13. E QUIDIMENSIONAL MORPHISMS
§14. U NIVERSALLY OPEN MORPHISMS
§15. S TUDY OF FIBRES OF A UNIVERSALLY OPEN MORPHISM
§16. D IFFERENTIAL INVARIANTS . D IFFERENTIALLY SMOOTH MORPHISMS
In this paragraph we will present, in global form, some notions of differential calculus particularly IV-4 | 5
useful in algebraic geometry. We will ignore many classic developments in differential geometry
(connections, infinitesimal transformations associated to vector fields, jets, etc.), although these
notions are translated in a particularly natural way for schemes. We will similarly ignore phenomena
exclusive to characteristic p > 0 (some of which are seen, in the affine case, in (0, 21). For certain
complements to the differential formalism for preschemes the reader may consult Exposés II and VII
of [eAG64] as well as subsequent chapters of this treatise.
16.1. Normal invariants of an immersion
(16.1.1). Let ( X, OX ), (Y, OY ) be two ringed spaces and f = (ψ, θ ) : Y → X a morphism of ringed
spaces (0, 4.1.1) such that the homomorphism
θ # : ψ∗ (OX ) −→ OY
is surjective, so that OY is identified with a sheaf of quotient rings ψ∗ (OX )/I f . We can then endow
ψ∗ (OX ) with the I f -preadic filtration.
Definition (16.1.2). — The OY -augmented sheaf of rings ψ∗ (OX )/I fn+1 is called the n’th normal
invariant of f ; the ringed space (Y, ψ∗ (OX )/I fn+1 ) is called the n’th infinitesimal neighbourhood of Y
(n)
along f and is denoted by Y f or simply Y (n) . The sheaf of graded rings associated to the sheaf of
filtered rings ψ∗ (OX )
Gr• ( f ) = (I fn /I fn+1 )
M
(16.1.2.1)
n⩾0
is called the sheaf of graded rings associated to f . The sheaf Gr1 ( f ) = I f /I f2 is called the conormal
sheaf of f (that will be denoted by NY/X when there is no risk of confusion).
It is clear that the OY (n) = ψ∗ (OX )/I fn+1 (that we also denote O (n) ) form a projective system of IV-4 | 6
Yf
sheaves of rings on Y, the transition homomorphism ϕnm : OY (m) → OY (n) for n ⩽ m identifies OY (n)
with the quotient of OY (m) by the power (I f /I fn+1 )m of the augmentation ideal of OY (n) , kernel of
ϕ0n : OY (n) → OY . The Y (n) therefore form a inductive system of ringed spaces, all having underlying
space Y, and we have canonical morphisms of ringed spaces hn : Y (n) → X equal to (ψ, θn ), where
θn# is the canonical morphism ψ∗ (OX ) → ψ∗ (OX )/I fn+1 . It is clear that the sheaf Gr• ( f ) is a sheaf of
graded algebras over the sheaf of rings OY = Gr0 ( f ) and the Grk ( f ) of OY -modules.
As with every sheaf of filtered rings, we have a canonical surjective homomorphism of graded
OY -algebras
•
(16.1.2.2) SO Y
(Gr1 ( f )) −→ Gr• ( f )
which coincide in degrees 0 and 1 with the identities.
Examples (16.1.3). —
(i) Suppose that X is a locally ringed space, Y is reduced to a single point y (endowed with a
ring Oy ) and that, if x = ψ(y), θ # : Ox → Oy is a surjective homomorphism of rings having as
kernel the maximal ideal mx of Ox . So the OY (n) are identified with the rings Ox /mnx +1 and
Gr• ( f ) with the graded ring associated with the local ring Ox endowed with the mx -preadic
filtration.
(ii) Suppose that Y is a closed subset of an open subspace U of X and that the OY is induced on
Y by a quotient sheaf OU /I , where I is an ideal of OU such that Ix = Ox for every x ̸∈ Y;
if X is a locally ringed space we also suppose that Ix ̸= Ox for y ∈ Y so that (Y, OY ) is a
locally ringed space.
Let ψ0 : Y → U be the canonical injection and denote by θ0 : OU → (ψ0 )∗ (OY ) the ho-
momorphism such that θ0# is the canonical homomorphism ψ0∗ (OU ) = OU |Y → (OU /I )|Y,
so that j0 = (ψ0 , θ0 ) : Y → U is a morphism of ringed spaces (and of locally ringed spaces
if X is a locally ringed space); if i : U → X is the canonical injection (morphism of ringed
spaces), j = i ◦ j0 is the morphism (ψ, θ ) of Y to X where ψ : Y → X is the canonical
injection and θ : OX → ψ∗ (OY ) is the homomorphism such that θ # = θ0# . Since θ # is surjec-
tive we can apply the previous definitions; OY (n) is equal to ψ0∗ (OU /I n+1 ), and we have
(ψ0 )∗ (OY(n) ) = OU /I n+1 , and Grn ( j) = Grn ( j0 ) = ψ0∗ (I n /I n+1 ) = j0∗ (I n /I n+1 ).
(16.1.4). The example (16.1.3, (ii)) shows that in general the OY (n) are not canonically endowed with a
structure of an OY -module, or a fortiori with a structure of an OY -algebra. The data of such structure
is equivalent to the data of a homomorphism of sheaves of rings λn : OY → OY (n) , right inverse to
the augmentation morphism ϕ0n ; it is also equivalent to the data of a morphism of ringed spaces
(1Y , λn ) : Y (n) → Y left inverse to the canonical morphism (1Y , ϕ0n ) : Y → Y (n) .
Proposition (16.1.5). — Let f = (ψ, θ ) : Y → X be an immersion of preschemes. Then:
(i) Gr• ( f ) is a quasi-coherent graded OY -algebra. IV-4 | 7
(ii) The Y (n) are preschemes, canonically isomorphic to subpreschemes of X.
(iii) Every homomorphism of sheaves of rings λn : OY → OY (n) , right inverse to the augmentation
homomorphism ϕ0n , makes the OY (n) and OY (k) for k ⩽ n quasi-coherent OY -algebras; the OY -module
structures induced from the above structures on the Grk ( f ) for k ⩽ n coincide with the ones defined
in (16.1.2).
Proof. (i) Since the question is local on X and Y, we can reduce to the case where Y is a closed
subpreschemes of X defined by an quasi-coherent ideal I of OX ; since OY is the restriction to
Y of OX /I the assertion (i) is evident, and Y (n) is the closed subprescheme of X defined by the
quasi-coherent ideal I n+1 of OX . Finally, to prove (iii) we notice that the data of λn makes the
ideal I /I n of the augmentation ϕ0n and their quotients I /I k+1 (1 ⩽ k ⩽ n) OY -modules, and
it suffices to prove by induction on k that the I /I k+1 are quasi-coherent OY -modules and the
structure of quotient OY -module induced on I k /I k+1 is the same as defined on (16.1.2). The
second assertion is immediate, I k /I k+1 being killed by I /I n+1 ; the first result, by induction on
k, is trivial for k = 1 and for I /I k+1 being an extension of I /I k by I k /I k+1 (1.4.17). □
Corollary (16.1.6). — Under the general hypotheses of (16.1.5), if the immersion f is locally of finite
presentation then the Grn ( f ) are quasi-coherent OY -modules of finite type.
Proof. Indeed, with the notation from the proof of (16.1.5), I is an ideal of finite type of OX (1.4.7),
therefore the I n /I n+1 are OY -modules of finite type, hence the conclusion. □
Corollary (16.1.7). — Under the general hypotheses of (16.1.5), let g : X → Y be a morphism of preschemes,
hn g
left inverse to f . Therefore, for every n, the composite morphism (1, λn ) : Y (n) −→ X −→ Y defines a
homomorphism of sheaves of rings λn : OY → OY (n) right inverse to the augmentation ϕ0n , making OY (n) a
quasi-coherent OY -algebra; via these homomorphisms, the transition homomorphism ϕnm : OY (m) → OY (n)
(n ⩽ m) are homomorphisms of OY -algebras. Also, if g is locally of finite type, then the OY (n) are quasi-coherent
OY -modules of finite type.
Proof. The first assertion is an immediate result from the definitions and (16.1.5). On the other hand,
if g is locally of finite type, then f is locally of finite presentation (1.4.3, (v)); the Grn ( f ) being then
quasi-coherent OY -modules of finite type by (16.1.6), the same goes for the OY -modules I /I n+1 ,
being extensions of a finite number of the Grk ( f ) (III, 1.4.17). □
Proposition (16.1.8). — Let X be a locally Noetherian prescheme, j : Y → X an immersion; Then the Y (n)
are locally Noetherian preschemes, the Grn ( j) are coherent OY -modules and the Gr• ( j) is a coherent sheaf of
rings over the space Y.
Proof. Everything is local on X and Y, so we reduce to the case where X is affine and j is a closed
immersion and therefore all the assertions are evident except for the last, which follows from the
fact that if A is a Noetherian ring and I is an ideal of A, then grI• ( A) is a Noetherian ring, taking
into account the exactness of the functor ψ∗ and (0, 5.3.7). □
Proposition (16.1.9). — Let X be a prescheme, j : Y → X an immersion locally of finite presentation, y a IV-4 | 8

point of Y. The following conditions are equivalent:
(a) There exists an open neighbourhood U of y in Y such that j|U is a homeomorphism of U onto an open
set of X.
(b) There is an integer n > 0 such that the canonical homomorphism
(ϕn−1,n )y : OY(n) ,y −→ OY(n−1) ,y
is bijective.
(c) There is an integer n > 0 such that (Grn ( j))y = 0.
In addition, if the integer n satisfies (b) or (c), then there is a neighbourhood V of y in Y such that
Grm ( j)|V = 0 for m ⩾ n and that ϕnm |V : OY (m) |V → OY (n) |V is bijective for m ⩾ n.
Proof. Since the questions is local on Y, we can restrict ourselves to the case where j is a closed
immersion, Y being defined by a quasi-coherent ideal of finite type I of OX . The equivalence of (b) and
(c), for a given n, is immediate; also, since I n /I n+1 is an OX -module of finite type, there is an open
neighbourhood U of y in X such that I n |U = I n+1 |U (0, 5.2.2), so we also have I n |U = I m |U for
m ⩾ n proving the last assertions. To prove that (a) implies (b), we can restrict ourselves to the cases
where the underlying space of Y is equal to the underlying space of X and where I is generated by
a finite number of sections over X: since I is contained in the nilradical N of OX (I, 5.1.2), it is now
nilpotent which proves b). Finally, to prove that (b) implies (a), we can restrict ourselves to the case
where I n = I m ; therefore, for every y ∈ Y, since Iy ⊂ my , maximal ideal of OX,y , we must have
Iyn = 0 because of Nakayama’s lemma, since Iy is an ideal of finite type. The set of x ∈ X such that
Ixn = 0 is an open U of X contained in Y (0, 5.2.2); since on the other hand Ix ̸= 0 for x ∈ / Y, we
must have U = Y. □
Corollary (16.1.10). — For a restriction of the immersion j to an open neighbourhood of y in Y to be an open

immersion (in other words, for j to be a local isomorphism on the point y), it is necessary and sufficient that
(Gr1 ( j))y = (NY/X )y = 0.
Proof. The condition is clearly necessary, and the previous reasoning applied to n = 1 proves that it
is sufficient. □
Remark (16.1.11). —
(i) Under the conditions of the definition (16.1.1), the projective limit of the projective system
(OY(n) , ϕnm ) of sheaves of rings over Y is called the normal invariant of infinite order of f , and
sometimes denoted by OY (∞) . When X is a locally noetherian prescheme, j : Y → X a closed
immersion, Y then is a closed subprescheme of X defined by a coherent ideal I and OY (∞) is
exactly the formal completion of OX along Y (I, 10.8.4), and Y (∞) = (Y, OY (∞) ) is the formal
prescheme that is the completion of X along Y (I, 10.8.5). In all cases, we could say that Y (∞)
is the formal neighbourhood of Y in X (via the morphism f ). In the particular case we have
just considered, it is the formal prescheme that is the inductive limit of the infinitesimal
neighbourhoods of order n.
(ii) Note that for a morphism of preschemes f = (ψ, θ ) : Y → X, it can happen that the
homomorphism θ # : ψ∗ (OX ) → OY is surjective without f being a local immersion and IV-4 | 9
without f being injective. We have an example by taking Y to be a sum of preschemes Yλ all
isomorphic to Spec(Ox ), where x ∈ X, ad taking f to be the morphism equal to the canonical
morphism in each of the Yλ .
16.2. Functorial properties of the normal invariants of an immersion
(16.2.1). Let f = (ψ, θ ) : Y → X and f ′ = (ψ′ , θ ′ ) : Y ′ → X ′ by two morphisms of ringed spaces such
that θ # and θ ′# are surjective; consider a commutative diagram of morphisms of ringed spaces
f
(16.2.1.1) YO /X
O
u v
Y′ / X′
f′
Let u = (ρ, λ), v = (σ, µ). We have ρ∗ (ψ∗ (OX )) = ψ′∗ (σ∗ (OX )) and as a result a commutative
diagram of homomorphisms of sheaves of rings over Y ′
ψ′∗ (µ# )
ρ∗ (ψ∗ (OX )) = ψ′∗ (σ∗ (OX )) / ψ′∗ (OX ′ )
ρ∗ (θ # ) θ ′#

ρ∗ (OY ) / OY ′
λ#
from which we conclude, if I and I ′ are the kernels of θ # and θ ′# , that we have ψ′∗ (µ# )(ρ∗ (I )) ⊂
I ′ , having in mind the exactness of the functor ρ∗ . We deduce that, for every integer n, ψ′∗ (µ# )(ρ∗ (I n )) ⊂
n
I ′ , which shows that ψ′∗ (µ# ) defines, passing to the quotients, a homomorphism of sheaves of
rings
n +1
(16.2.1.2) νn : ρ∗ (ψ∗ (OX )/I n+1 ) −→ ψ′∗ (OX ′ )/I ′
and therefore a morphism of ringed spaces wn = (ρ, νn ) : Y ′(n) → Y (n) (which, for n = 0, is none
other than u). It follows immediately from this definition that the diagrams
hmn hm
Y (O n) / Y (m) /X
O O
wn wm v (n ⩽ m)
Y ′(n) / Y ′(m) / X′
h′mn h′m
(where the horizontal arrows are the canonical morphisms (16.1.2)) are commutative.
By passage to the quotients via the morphisms (16.2.1.2), and taking into account the exactness IV-4 | 10
of the functor ρ∗ , we obtain a di-homomorphism of graded algebras (relative to the morphism
λ# : ρ∗ (OY ) → OY ′ )
(16.2.1.3) gr(u) : ρ∗ (Gr• ( f )) −→ Gr• ( f ′ )
(or, if you like, a ρ-morphism (0, 3.5.1) Gr• ( f ) → Gr• ( f ′ )), and in particular a di-homomorphism of
conormal sheaves
gr1 (u) : ρ∗ (Gr1 ( f )) −→ Gr1 ( f ′ ).
It is also immediate that these homomorpisms give rise to a commutative diagram
(16.2.1.4) ρ ∗ ( SO
• (Gr ( f )))
1
/ ρ∗ (Gr• ( f ))
Y
S(gr1 (u)) gr(u)

• (Gr ( f ′ ))
SO 1
/ Gr• ( f ′ )
Y
where the horizontal arrow are the canonical morphisms (16.1.2.2).

Finally, if we have a commutative diagram of morphisms of ringed spaces
f
YO /X
O
u v
YO ′ / X′
f′ O
u′ v′
Y ′′ / X ′′
f ′′
where f ′′ = (ψ′′ , θ ′′ ) is such that θ ′′# is surjective, and if wn′ and wn′′ are defined from u′ , v′ for one
and u′′ = u ◦ u′ , v′′ = v ◦ v′ for the other, we have wn′′ = wn ◦ wn′ , which follows immediately
from the definitions and from (0, 3.5.5); we have also gr(u′′ ) = gr(u′ ) ◦ ρ′∗ (gr(u)) if u′ = (ρ′ , λ′ ).
Therefore we can say that Y (n) and Gr• ( f ) depend functorially on f .
Proposition (16.2.2). — With the notation and hypotheses of (16.2.1), suppose also that f , f ′ , u, and v are
morphisms of preschemes. We have:
(i) The morphisms wn : Y ′(n) → Y (n) are morphisms of preschemes.
(ii) If Y ′ = Y × X X ′ , u and f ′ the canonical projections, and if f is an immersion or if v is flat, we have
Y ′(n) = Y (n) × X X ′ .
(iii) If Y ′ = Y × X X ′ and if v is flat (resp. if f is an immersion), the homomorphism
Gr(u) = gr(u) ⊗ I : Gr• ( f ) ⊗OY OY ′ −→ Gr• ( f ′ )
is bijective (resp. surjective).
Proof.
(i) The hypotheses immediately imply that, for every y′ ∈ Y ′ , ρ∗y′ (θψ# ′ (y′ ) ) is a local homomor-
phism (I, 1.6.2), so wn is a morphism of preschemes (I, 2.2.1). IV-4 | 11
(ii) and (iii) If f is an immersion, we can restrict ourselves to the case where f is a closed immersion, Y
being defined by a quasi-coherent ideal I of OX and Y (n) by the ideal I n+1 ; the assertions
follows from (I, 4.4.5).
Second, suppose that v is flat; we can restrict ourselves to the case where X = Spec( A),
Y = Spec( B), X ′ = Spec( A′ ) are affines, A′ being a flat A-module; so Y ′ = Spec( B′ ) where
B′ = B ⊗ A A′ ; in addition, if I is the kernel of the homomorphism A → B, the kernel I′ of
n +1
A′ → B′ is identified with I ⊗ A A′ by flatness, and I ′n /I ′ is equal to
ψ′∗ (σ∗ ((In /In+1 )e) ⊗σ∗ (OX ) OX ′ ) =
ψ′∗ (σ∗ ((In /In+1 ))e)⊗ψ′∗ (σ∗ (OX )) ψ′∗ (OX ′ ) = ρ∗ (I n /I n+1 ) ⊗ρ∗ (ψ∗ (OX )) ψ′∗ (OX ′ )
and in particular for n = 0, we have
OY ′ = ρ∗ (OY ) ⊗ρ∗ (ψ∗ (OX )) ψ′∗ (OX ′ )
n +1
from which we have canonical isomorphism of I ′n /I ′ with
ρ∗ (I n /I n+1 ) ⊗ρ∗ (OY ) OY ′ = (I n /I n+1 ) ⊗OY OY ′
which proves (iii). Let now Cn = Γ(Y, OY (n) ), Cn′ = Γ(Y ′ , OY ′(n) ). As Y (n) and Y ′(n) are
affine schemes (16.1.5), the kernel Kn (resp. K′n ) of the homomorphism Cn → Cn−1 (resp.
n +1
Cn′ → Cn′ −1 ) is Γ(Y, I n /I n+1 ) (resp. Γ(Y, I ′n /I ′ )); therefore we can deduce from the
above results that K′n = Kn ⊗ A A′ . Now, we have a commutative diagram
0 / Kn ⊗ A A ′ / Cn ⊗ A A′ / Cn−1 ⊗ A A′ /0
r sn s n −1

0 / K′n / Cn′ / C′ /0
n −1
where the vertical arrow of the left is bijective and the two lines are exact (A′ being a flat
A-module). We deduce by induction that sn is bijective for every n, because it is true by
hypothesis for n = 0, and is deduced by application of the five lemma for all n. That proves
the second assertion of (ii).
□
Corollary (16.2.3). — Let g : X → Y, u : Y ′ → Y be two morphisms of preschemes, X ′ = X ×Y Y ′ ,

g′ : X ′ → Y ′ and v : X ′ → X by the canonical projections. Let f : Y → X by a Y-section of X (and therefore
an immersion), f ′ = f (Y ′ ) : Y ′ → X ′ the Y ′ -section of X ′ deduced from f by the base change u. We have:
(n) (n)
(i) The morphism wn : Y f′ ′ → Yf corresponding to f , f ′ , u, v (16.2.1) and the canonical morphism
(n) (n) (n)
h′n : Y f′ ′ → X ′ identifies Y f′ ′ with the product Yf × X X ′ .
(ii) If we endow O (n) (resp. O ′ (n) ) with the structure of an OY -algebra defined by g (resp. with the
Yf Y′
f
structure of an OY ′ -algebra defined by g′ ) (16.1.5, (iii)), then the homomorphism of OY ′ -algebras
(16.2.3.1) ρ ∗ (O (n) ) ⊗OY OY′ −→ OY′ (n)
Yf f′
induced by the homomorphism νn (16.2.1.2) is bijective. Also, the homomorphism of OY ′ -modules IV-4 | 12
(16.2.3.2) Gr1 (u) : Gr1 ( f ) ⊗OY OY ′ −→ Gr1 ( f ′ )
is bijective.
Proof.
(i) Let us first note that f ′ : Y ′ → X ′ and u : Y ′ → Y identifies Y ′ with the product Y × X X ′ (via
the structure morphisms f : Y → X and v : X ′ → X) (14.5.12.1). The conclusion of (i) now
follows from (16.2.2, (ii)), the morphism g being an immersion.
(ii) The commutative diagram
Yf
(n) o Y f′ ′
(n)
wn
hn h′n

Xo v
X′
g g′

Yo u
Y′
(n) (n)
identifies Y f′ ′ with the product Yf × X X ′ , so (I, 3.3.9) it identifies (via the morphisms
(n) (n) (n) (n)
g′ ◦ h′n and wn ) Y f′ ′ to the product Y f ×Y Y ′ . Since Y f (resp. Y f′ ′ ) is the affine
prescheme over Y (resp. over Y ′ ) associated with the OY -algebran O (n) (resp. to the OY ′ -
Yf
algebran O ), the fact that the canonical homomorphism (16.2.3.1) is bijective follows
Y ′ ′ (n)
f
from (II, 1.5.2). Finally, the canonical homomorphism (16.2.3.1) is compatible with the aug-
mentations O (n) → OY and O ′ (n) → OY ′ ; since O (n) is a direct sum (as an OY -module) of
Yf Y′ Yf
f
OY and the augmentation ideal I /I n+1 , we can therefore see that the canonical homomor-
phism (16.2.3.1), restricted to I /I n+1 ⊗OY OY ′ , is a bijection of the latter onto I ′ /I ′n+1 .
For n = 1 this shows that Gr1 (u) is bijective.
□
We note that, under the hypotheses of (16.2.3), the homomorphisms Grn (u) are surjective in view
of the above, but are not bijective in general for n ⩾ 2. However:
Corollary (16.2.4). — Under the hypotheses of (16.2.3), suppose that u : Y ′ → Y is a flat morphism (resp.
that the Grn ( f ) are flat OY -modules for n ⩽ m). Then the homomorphism
Grn (u) : Grn ( f ) ⊗OY OY ′ −→ Grn ( f ′ )
is bijective for all n (resp. for n ⩽ m).
Proof. If u is flat, then we deduce by base change that the same is true for v : X ′ → X, and we already
know in this case that Gr(u) is bijective (16.2.2, (iii)). If the Grn ( f ) are flat for n ⩽ m, then we first
see by induction on n that the same holds for I /I n+1 for n ⩽ m, because of the exact sequences
0 / I n /I n+1 / I /I n+1 / I /I n /0
(0, 6.1.2); in addition, we have the commutative diagram IV-4 | 13
0 / (I n /I n+1 ) ⊗O OY ′ / (I /I n+1 ) ⊗O OY ′ / (I /I n ) ⊗O OY ′ /0
Y Y Y

0 / I ′ n /I ′ n+1 / I ′ /I ′ n+1 / I ′ /I ′ n /0
in which the lines are exact (the first by flatness (0, 6.1.2)) and the two last vertical arrows are bijective
by virtue of (16.2.2, (ii)); hence the conclusion. □
Remarks (16.2.5). —
(i) The reasoning of (16.2.2, (i)) still applies to (16.2.1.1) when these are morphisms of locally
ringed spaces (I, 1.8.2).
(ii) In (16.2.2, (ii)), the conclusion is no longer necessarily valid if we only suppose that v and
f are morphisms of preschemes ( f satisfying the condition of (16.1.1)). For example (with
the notation of the proof of (16.2.2, (ii))), it can happen that I = 0 but the kernel I′ of
A′ → B′ = B ⊗ A A′ is not zero and that B′ ̸= 0, in which case we have Y (n) = Y for all n,
but Y ′
(n)
̸= Y ′ . We have an example of this by taking A = Z, B = Q, A′ = ∏∞ h
h=1 ( Z/m Z )
where m > 1.
(16.2.6). Consider the particular case of the diagram (16.2.1.1) where X ′ = X, v is the identity, X
a prescheme, Y a subprescheme of X, Y ′ a subprescheme of Y, f , u, and f ′ = f ◦ u the canon-
ical injections; the di-homomorphism (16.2.1.3) gives us, by tensoring with OY ′ over ρ∗ (OY ), a
homomorphism of graded OY ′ -algebras
(16.2.6.1) u∗ (Gr• ( f )) −→ Gr• ( f ′ ).
On the other hand, we identify OY to ψ∗ (OX )/I f and OY ′ to ρ∗ (OY )/Iu ; since ρ∗ is an exact functor,
we have ρ∗ (OY ) = ρ∗ (ψ∗ (OX ))/ρ∗ (I f ) = ψ′∗ (OX )/ρ∗ (I f ), and since OY ′ is moreover identified
with ψ′∗ OX /I f ′ , we see that Iu = I f ′ /ρ∗ (I f ). We deduce that for every integer n there is a
canonical homomorphism I fn′ /I fn′ +1 → Iun /Iun+1 , from which we have a canonical morphism of
graded OY ′ -algebras
(16.2.6.2) Gr• ( f ′ ) −→ Gr• (u).
Proposition (16.2.7). — Let X be a prescheme, Y a subprescheme of X, Y ′ a subprescheme of Y, j : Y ′ → Y
the canonical injection. We then have an exact sequence of conormal sheaves (OY ′ -modules)
(16.2.7.1) j∗ (NY/X ) / NY ′ /X / NY ′ /Y /0
where the arrows are the degree 1 components of the canonical homomorphisms (16.2.6.1) and (16.2.6.2).
Proof. The problem being local, we can restrict to the case where X = Spec( A), Y = Spec( A/I)
and Y ′ = Spec( A/K), I and K being ideals of A such that I ⊂ K; everything reduces to seeing
that the sequence of canonical morphisms I/KI → K/K2 → (K/I)/(K/I)2 → 0 is exact, which is IV-4 | 14
immediate given that the image of I/KI in K/K2 is (I + K2 )/K2 and that (K/I)/(K/I)2 is identified
with K/(I + K2 ). □
It is easy to give examples where the sequence (16.2.7.1) extended on the left by 0 is not exact; with
the above notation, it suffices to take A = k[ T ], I = AT 2 , K = AT, because then (I + K2 )/K2 = 0
and I/KI ̸= 0. See however (16.9.13) and (19.1.5) for some cases where the extended sequence is
indeed exact.
16.3. Fundamental differential invariants of morphisms of preschemes
Definition (16.3.1). — Let f : X → S be a morphism of preschemes, ∆ f : X → X ×S X the

corresponding diagonal morphism, which is an immersion (I, 5.3.9). We denote by P nf or PX/S n ,
and call the sheaf of principal parts of order n of the S-prescheme X, the OX -augmented sheaf of rings,
n-th normal invariant of ∆ f (16.1.2). We will also write P ∞ ∞
f = PX/S = lim P n , Grn (P f ) =
←−n X/S
Grn (PX/S ) = Grn (∆ f ) (16.1.2); the OX -module Gr1 (∆ f ), augmentation sheaf of ideals of PX/S 1 , is
denoted by Ω f or Ω X/S , and is called the OX -module of 1-differentials of f , or of X with respect to S,

1 1
or of the S-prescheme X.
It follows from this definition that PX/S

0 is canonically identified with OX (16.1.2).
We have (16.1.2.2) a canonical surjective morphism of graded OX -algebras
•
(16.3.1.1) SO X
(Ω1X/S ) −→ Gr• (PX/S ).
And it follows from Definition (16.3.1) that for every open U of X we have P nf|U = P nf |U, P ∞
f |U
=
P∞f |U, Grn (P f |U ) = Grn (P f )|U, Ω f |U = Ω f |U (in other words, the notions introduced are local on
1 1
X).
(16.3.2). Denote by p1 , p2 the two canonical projections of the product X ×S X; since ∆ f is an X-

section of X ×S X for both p1 and p2 , each of these morphisms define, for all n, a homomorphism
of sheaves of rings OX → PX/S n , right inverse of the augmentation P n
X/S → OX (16.1.7); we
can also say that we thus define on PX/Sn two quasi-coherent augmented OX -algebra structures; the
corresponding OX -module structures on on Grn (PX/S n ) are the same. We also have, by passing to
∞ .
the limit, two OX -algebra structures on PX/S
(16.3.3). The morphism s = ( p2 , p1 )S : X ×S X → X ×S X is an involutive automorphism of X ×S X,

called the canonical symmetry, such that
(16.3.3.1) p1 ◦ s = p2 , p2 ◦ s = p1 , s ◦ ∆f = ∆f .
If we put s = (ρ, λ), pi = (πi , µi ) (i = 1, 2), ∆ f = (δ, ν), λ# is then an isomorphism of

ρ∗ (π1∗ (OX )) onto π2∗ (OX ), and δ∗ (λ# ) fixes δ∗ (OX ×S X ) and the kernel I of the homomorphism
ν# : δ∗ (OX ×S X ) → OX . Therefore:
Proposition (16.3.4). — The homomorphism σ = δ∗ (λ# ) induced from s (and also called the canonical
symmetry) is an involutive automorphism of the projective system (PX/S n ) of O -augmented sheaves of
X IV-4 | 15
∞
rings, and as a result also of the projective limit PX/S . This automorphism permutes the OX -algebra structure
on PX/S
n and on PX/S ∞ .
(16.3.5). In what follows, the two OX -algebra structures defined on the PX/S n ∞ will play
and on PX/S
very different roles: we will now agree, unless said otherwise, that when PX/S
n ∞ is considered as an
or PX/S
OX -algebra, it is the algebra structure induced by p1 .
For every open U of X and every section t ∈ Γ(U, OX ), we will simply denote by t.1 or even t the
image of t under the structure morphism Γ(U, OX ) → Γ(U, PX/S n ) (resp. Γ (U, O ) → Γ (U, P ∞ ))
X X/S
(that is to say, the homomorphism corresponding to p1 ).
Definition (16.3.6). — We denote by dnf , or dnX/S (resp. d∞ ∞ n ∞

f , or d X/S ), or simply d (resp. d ), the
∞
homomorphism of sheaves of rings OX → P f = PX/S (resp. OX → P f = PX/S ) induced by p2 .
n n ∞
For every open U of X, and every t ∈ Γ(U, OX ), dn t (resp. d∞ t) is called the principal part of order n
(resp. principal part of infinite order) of t. We set dt = d1 t − t, and we say that dt is the differential of t
(an element of Γ(U, Ω1X/S ), also denoted d X/S (t)).
It follows immediately 2 from this definition that we have

(16.3.6.1) d(t1 t2 ) = t1 dt2 + t2 dt1
for every t1 , t2 in Γ(UOX ), that is, d is a derivation of the ring Γ(U, OX ) in the Γ(U, OX )-module
Γ(U, Ω1X/S ).
In all notation introduced in (16.3.1) and (16.3.6), we will sometimes replace S by A when
S = Spec( A).
(16.3.7). Suppose in particular that S = Spec( A) and X = Spec( B) are affine schemes, B then being
an A-algebra. Then ∆ f corresponds to the canonical surjective homomorphism π : B ⊗ A B → B such
that π (b ⊗ b′ ) = bb′ , with kernel I = IB/A (0, 20.4.1); P nf is the structure sheaf of the prescheme
n ), where
Spec( PB/A
n
PB/A = ( B ⊗ A B)/In+1 ;
Gr• (P f ) is the quasi-coherent OX -module corresponding to the graded B-module
grI• ( B ⊗ A B) = (In /In+1 );
M
n⩾0
in particular Ω1f
= Ω1X/S
is the quasi-coherent OX -module corresponding to the B-module of 1-
differentials of B over A, Ω1B/A (0, 20.4.3). The projection morphisms p1 : X ×S X → X, p2 :
X ×S X → X corresponding to the two homomorphisms of rings j1 : B → B ⊗ A B, j2 : B → B ⊗ A B
n
such that j1 (b) = b ⊗ 1, j2 (b) = 1 ⊗ b, so that (by the convention of (16.3.5)), PB/A is always
n ; the ring j1
considered as a B-algebra via the composite homomorphism B −
→ B ⊗ A B → PB/A
j2 n
homomorphism B − → B ⊗ A B → PB/A is denoted by dnB/A and corresponds to dnX/S acting on
Γ( X, OX ); for every t ∈ B, dt is equal to d B/A t, defined in (0, 20.4.6).
If πn : B ⊗ A B → PB/An is the canonical homomorphism, so we have, in light of the preceding
definitions,
(16.3.7.1) πn (b ⊗ b′ ) = b · πn (1 ⊗ b′ ) = b · dnB/A (b′ ) for b ∈ B, b′ ∈ B.
Proposition (16.3.8). — The image of the canonical homomorphism dnX/S : OX → PX/S
n generates the IV-4 | 16
OX -module PX/S .
n
Proof. We immediately reduce to the case where X = Spec( B) and S = Spec( A) are affine and the
proposition follows from (16.3.7.1) since πn is surjective. We note that in general dnX/S is not surjective
(even for n = 1). □
Proposition (16.3.9). — Suppose that f : X → S is a morphism locally of finite type. Then the P nf and the
Grn (P f ) are quasi-coherent OX -modules of finite type.
Proof. This follows from (16.1.6) and from the fact that ∆ f is locally of finite presentation (I, 4.3.1). □
2[Trans.] This is, locally we have (0, 20.1.1).
16.4. Functorial properties of differential invariants
(16.4.1). Consider a commutative diagram of morphisms of preschemes
Xo
u
(16.4.1.1) X′
f f′

So w
S′
We deduce a commutative diagram
Xo
u
X′
∆f ∆f′

X ×S X o v
X ′ ×S′ X ′
where v is the composite homomorphism (I, 5.3.5) and (I, 5.3.15).

( p′ ,p2′ )S u×S u
(16.4.1.2) X ′ ×S′ X ′ −−−−
1
→ X ′ ×S X ′ −−−→ X ×S X.
So we induce from u and v, as explained in (16.2.1), homomorphisms of augmented sheaves of
rings
(16.4.1.3) νn : ρ∗ (PX/S
n
) −→ PXn ′ /S′
(where we put u = (ρ, λ)); these homomorphisms form a projective system, and therefore give at
the limit a homomorphism of sheaves of graded rings
∞
(16.4.1.4) ν∞ : ρ∗ (PX/S ) −→ PX∞′ /S′ ;
on the other hand, by passing to the quotient, the homomorphisms νn give rise to a di-homomorphism
of graded algebras (relative to λ# ):
(16.4.1.5) gr(u) : ρ∗ (Gr• (PX/S )) −→ Gr• (PX ′ /S′ ).
(16.4.2). If we have a commutative diagram IV-4 | 17
u′
Xo X′ o
u
X ′′
f f′ f ′′

So w
S′ o S′′
w′
we deduce a commutative diagram
u′
Xo X′ o
u
X ′′
∆f ∆f′ ∆ f ′′

X ×S X o v
X ′ ×S′ X ′ o X ′′ ×S′′ X ′′
v′
where v′ is defined from u′ , w′ , f ′ , f ′′ as v is from u, w, f , f ′ . We verify immediately that if u′′ = u ◦ u′ ,

w′′ = w ◦ w′ , then the composite homomorphism v ◦ v′ is equal to the homomorphism v′′ deduced
from u′′ , v′′ , f , f ′′ as v is from u, w, f , f ′ . If we put u′ = (ρ′ , λ′ ), u′′ = (ρ′′ , λ′′ ) it follows (16.2.1) that
∗
the homomorphism νn′′ : ρ′′ (PX/S n ) → Pn
X ′′ /S′′ is equal to the composite
∗
∗ ρ′ (νn# ) ∗ ν′
ρ′ (ρ∗ (PX/S
n
)) −−−−→ ρ′ (PXn ′ /S′ ) −
→n
PXn ′′ /S′′ ,
and we have analogous transitivity properties for the homomorphisms (16.4.1.4) and (16.4.1.5),
n , P ∞ and Gr (P
which lets us say that the PX/S X/S • X/S ) depend functorially on f .
(16.4.3). We verify immediately (for example, by restricting ourselves to the affine case with help of
(16.3.7)) that with the notation of (16.4.1), the diagram
λ# / OX ′
(16.4.3.1) ρ ∗ (O X )

ρ∗ (PX/S
n ) / Pn′ ′
νn X /S
where the vertical arrows are the ones defining the algebra structure chosen in (16.3.5) (that is to say,
the ones coming from the first projections) is commutative; the same goes for the diagram
λ# / OX ′
(16.4.3.2) ρ ∗ (O X )
ρ∗ (dnX/S ) dnX ′ /S′

ρ∗ (PX/S
n ) / Pn′ ′
νn X /S
the vertical arrows defining here the algebra structure from the second projection; besides, if σ and IV-4 | 18
σ′ are the canonical symmetries corresponding to f and f ′ (16.3.4), we have
νn ◦ ρ∗ (σ ) = σ′ ◦ νn
which switches one diagram with the other. We deduce from (16.4.3.1) a canonical homomorphism
of augmented OX ′ -algebras
(16.4.3.3) Pn (u) : u∗ (PX/S
n n
) = PX/S ⊗OX OX ′ −→ PXn ′ /S′
and it follows from (16.4.3.2) that the diagram
(16.4.3.4) OX ′
id / OX ′
u∗ (dnX/S ) dnX ′ /S′

u∗ (PX/S
n ) / Pn′ ′
Pn (u) X /S
is commutative. We deduce a homomorphism of graded OX ′ -algebras

(16.4.3.5) Gr• (u) : u∗ (Gr• (PX/S )) −→ Gr• (PX ′ /S′ )
and in particular a homomorphism of OX ′ -modules
(16.4.3.6) Gr1 (u) : Ω1X/S ⊗OX OX ′ −→ Ω1X ′ /S′
giving rise to a commutative diagram
(16.4.3.7) OX ′
id / OX ′
d X/S ⊗1 d X ′ /S′

Ω1X/S ⊗OX OX ′ / Ω1 ′ ′
X /S
(16.4.4). When S = Spec( A), S′ = Spec( A′ ), X = Spec( B), X ′ = Spec( B′ ) are affine, so that we have
a commutative diagram of ring homomorphisms
BO / B′
O
A / A′
the image of IB/A in B′ ⊗ A′ B′ is contained in IB′ /A′ , and the homomorphism νn corresponds to the
homomorphism of rings PB/A n → PBn′ /A′ induced from the homomorphism B ⊗ A B → B′ ⊗ A′ B′ by
passing to quotients. The homomorphism (16.4.3.6) corresponds to the homomorphism defined in
(0, 20.5.4.1), and the commutative diagram (16.4.3.7) to the diagram (0, 20.5.4.2).
Proposition (16.4.5). — Suppose that X ′ = X ×S S′ , f ′ and u the canonical projections. Then the canonical IV-4 | 19
homomorphisms Pn (u) (16.4.3.3) and Gr1 (u) (16.4.3.6) are bijective.
Proof. We have X ′ ×S′ X ′ = ( X ×S X ) ×S S′ , and it suffices to apply (16.2.3, (ii)) replacing g by the
first p1 : X ×S X → X and f by the diagonal ∆ f . □
We note that under the hypotheses of (16.4.5) the homomorphism Gr• (u) (16.4.3.5) is surjective,
but not bijective in general. However (16.2.4):
Corollary (16.4.6). — Under the hypotheses of (16.4.5), suppose in addition that w : S → S′ is flat (resp.
that Grn (PX/S
n ) are flat O -modules for n ⩽ m); then the homomorphism
X
Grn (u) : u∗ (Gr(PX/S

n
)) −→ Gr(PXn ′ /S′ )
is bijective for each n (resp. for n ⩽ m).
Proof. Indeed, if w is flat, then so is v : X ′ ×S′ X ′ → X ×S X, so the conclusion follows from
(16.2.4). □
(16.4.7). Let S be a prescheme, E a quasi-coherent OS -Module, and set X = V(E ) (II, 1.7.8), the
vector bundle associated to E , equal to Spec(SOS (E )). Let f : X → S be the structure morphism.
For every open U of S and every section t ∈ Γ(U, E ), t is identified with a section of SOS (E ) over U;
let t′ be its image in Γ( f −1 (U ), OX ) = Γ(U, f ∗ (OX )) = Γ(U, SOS (E )), and set
(16.4.7.1) δ(t) = dnX/S (t′ ) − t′ ∈ Γ( f −1 (U ), PX/S

n
);
it is clear that δ is a di-homomorphism of modules (corresponding to the homomorphism of rings
Γ(U, OS ) → Γ( f −1 (U ), OX )) of Γ(U, E ) into Γ( f −1 (U ), PX/S
n ), and therefore the image belongs to
the augmentation ideal of Γ( f −1 (U ), PX/S

n ). We deduce (by varying U) a canonical homomorphism
of OX -algebras
(16.4.7.2) f ∗ (SOS (E )) −→ PX/S
n
and in view of the above remark, if K is the ideal kernel of augmentation SOS (E ) → OS , the
image of K n+1 by (16.4.7.2) is zero, so that by factoring by K n+1 , we finally have a canonical
homomorphism
(16.4.7.3) δn : f ∗ (SOS (E )/K n +1 n

) −→ PX/S .
Proposition (16.4.8). — Under the conditions of (16.4.7), the homomorphisms δn are bijective and form a
projective system of isomorphisms; we deduce an isomorphism of graded OS -algebras
(16.4.8.1) f ∗ ( SO
•
S
(E )) −→ Gr• (PX/S ).
Proof. The fact that homomorphisms (16.4.7.3) form a projective system follows immediately from
their definition. To prove they are isomorphisms, it suffices to prove that (16.4.8.1) is an isomorphism, IV-4 | 20
since both filtrations involved in (16.4.7.3) are finite (Bourbaki, Alg. comm., chap. III, §2, no 8, cor. 3
of th. 1). To do this, consider the split exact sequence of OS -modules
(16.4.8.2) 0 /E u / E ⊕E v /E /0
where, for every pair of sections s, t of E over an open U of S, we take u(s) = (−s, s) and v(s, t) =
s + t. We have
X ×S X = Spec(SOS (E ) ⊗OS SOS (E )) = Spec(SOS (E ⊕ E ))
((II, 1.4.6) and (II, 1.7.11)), and the diagonal morphism X → X ×S X corresponds (II, 1.2.7) to the
homomorphism of OX -algebras S(v) : SOS (E ⊕ E ) → SOS (E ) (II, 1.7.4), such that if I is the kernel
of this homomorphism, then we have
n
PX/S = f ∗ (SOS (E ⊕ E )/I n+1 ).
The proposition now will be a consequence of the following lemma: □
u v
Lemma (16.4.8.3). — Let Y be a ringed space, 0 → F ′ − → F ′′ → 0 an exact sequence of OY
→ F −
modules such that each point y ∈ Y has an open neighbourhood V such that the sequence 0 → F ′ |V →
F |V → F ′′ |V → 0 is split. Let I be the kernel ideal of S(v):
SOY (F ) −→ SOY (F ′′ ),
• ( S (F )) be the graded O -algebra associated to the O -algebra S (F ) endowed with the
and let grI OY Y Y OY
I -preadic filtration. Then the homomorphism of graded OY -algebras
•
(16.4.8.4) SO Y
(F ′ ) ⊗OY SO
•
Y
(F ′′ ) −→ grI
•
(SOY (F ))
(where the first member is the graded tensor product of symmetric OY -algebras endowed with the canonical
gradation (II, 1.7.4) and (II, 2.1.2)), induced by the canonical injection
F ′ −→ I = gr1I (SOY (F )),
is bijective.
Proof. The injection F ′ → I indeed canonically gives a homomorphism of graded OY -algebras

SO• (F ′ ) → gr• ( S• (F )), and since the second member is by definition a graded S• (F ′′ )-
Y I OY OY
algebra, we induce the canonical homomorphism (16.4.8.4) by tensoring the above with SO • (F ′′ ).
Y
′
To prove the lemma we can, being a local problem, restrict to the case where F = F ⊕ F , u and v′′
the canonical homomorphisms. Then the graded algebra SO • (F ) is canonically identified with the
Y
• (F ′ ) ⊗ • ′′
graded tensor product SO
Y
OY SOY (F ) (II, 1.7.4), and it is immediate that I is therefore the
ideal I ′ ⊗OY SO• (F ′′ ), where I ′ is the augmentation ideal of S• (F ′ ), that is to say the (direct)
OY
Y
m (F ′ ) for m ⩾ 1. We conclude that I n = I ′n ⊗ • ′′ ′n
sum of the SO
Y
OY SO (F ), where this time I is
Y
m (F ′ ) for m ⩾ n; we have therefore I n /I n+1 = Sn (F ) ⊗ • ′′
the direct sum of the SO
Y OY OY SOY (F ),
which proves that (16.4.8.4) is bijective. □
Having proved the lemma, it remains to see that the homomorphism (16.4.8.1) is the image by IV-4 | 21
f ∗ of the homomorphism (16.4.8.4) corresponding to the exact sequence (16.4.8.2); we can easily
see that it follows from the definition of u (16.4.8.2) and of δ (16.4.7.1), given the definition of the
OX -algebra structures of PX/S
n and of the dnX/S (16.3.5) and (16.4.3.6).
In particular:
Corollary (16.4.9). — Under the conditions of (16.4.7), we have a canonical isomorphism
(16.4.9.1) gr1 (δ) : f ∗ (E ) ≃ Ω1X/S .
Corollary (16.4.10). — If S = Spec( A), E = OSm , so that
X = Spec( A[ T1 , . . . , Tm ]),
then PX/S n is canonically identified with the OX -algebra corresponding to the quotient A[ T1 , . . . , Tm ]-algebra
A[ T1 , . . . , Tm , U1 , . . . , Um ]/Kn+1 , where the Ui (1 ⩽ i ⩽ m) are m new indeterminates and K is the ideal
generated by U1 , . . . , Um .
We thus recover in particular the structure of Ω1X/S in this case (0, 20.5.13).
In addition, note that the dnX/S then corresponds to a polynomial F ( T1 , . . . , Tm ), the class modulo
n +
K 1 of F ( T1 + U1 , . . . , Tm + Um ), which follows from the definition (16.4.7.1).
Proposition (16.4.11). — Let f : X → S be a morphism, g : S → X a S-section of X, S(n) the n-th

infinitesimal neighbourhood of S by the immersion g (16.1.2). Then there exists a unique isomorphism of
OS -algebras
(16.4.11.1) ϖn : g∗ (PX/S
n
) −→ O (n)
Sg
(via the OS -algebra structure on O (n) defined by f (16.1.7)), making the diagram
Sg
(16.4.11.2) OS = g ∗ (O X )
λn
/O (n)
: Sg
g∗ (dnX/S ) ϖn
&
g∗ (PX/S
n )
commutative (where λn is the structure morphism).
Proof. In light of (I, 5.3.7), where we replace X, Y, S by S, X, S respectively and f by g, the diagrams
g g
(16.4.11.3) S /X S /X
g ∆f g ∆f

X / X ×S X X / X ×S X
( g◦ f ,1X )S (1X ,g◦ f )S
identifies S with the product of the ( X ×S X )-preschemes X and X by the morphisms ∆ f and IV-4 | 22
( g ◦ f , 1X )S (resp. (1X , g ◦ f )S ). On the other hand, the diagrams
( g◦ f ,1X )S (1X ,g◦ f )S
(16.4.11.4) X /X X /X
f p1 f p2

S /X S /X
g g
identify X to the product of X-preschemes S and X ×S X via the morphisms g and p1 (resp. p2 )
(particular case of the associativity formula (I, 3.3.9.1)). We can say that ∆ f , considered as an X-
section of X ×S X (relative to p1 or p2 ) plays the role of a universal section for the S-sections of X: each
of these sections g in fact are deduced by base change ( g ◦ f , 1X )S : X → X ×S X. The definition of the
homomorphism ω̄n and the fact that it is bijective follows from the remarks of (16.2.3, (ii)) applied to
the first diagram (16.4.11.4). The commutativity of the first diagram (16.4.11.4) follows also from
(16.2.3, (ii)) this time applied to the second diagram (16.11.4). To explain ϖn , we can restrict ourselves
to the case where g is a closed immersion: Indeed, for every s ∈ S, there is an open neighbourhood
W of s in S such that g(W ) is closed in an open set U of X, and it is clear that g|W is a W-section of
the morphism U ∩ f −1 (W ). We can then suppose that S is a closed subprescheme of X defined by a
quasi-coherent ideal K . Then the preceding definitions show that if W is an open of S, t is a section
of OX over f −1 (W ), ϖn (dn t|W ) is equal to the canonical image of t in Γ(W, (OX /K n+1 )|W ). The
uniqueness of ϖn then follows since the image of OX under dnX/S generates the OX -module PX/S n
(16.3.8). □
Corollary (16.4.12). — Let k be a field, X a k-prescheme, x a point of X rational over k. Then (PX/S
n ) ⊗
x Ox
k ( x ) is canonically isomorphic (as an augmented k ( x )-algebra) to Ox /mx .
n + 1
Proof. It suffices to consider the unique k-section g of X such that g(Spec(k)) = { x }. □
Corollary (16.4.13). — Let f : X → S be a morphism, s a point of S, Xs = X ×S Spec(k (s)) the fibre of f

in s. If x ∈ Xs is rational over k (s), (PX/S ′ n+1 , is the
x Os k ( s ) is canonically isomorphic to OXs ,x /mx
n ) ⊗
maximal ideal of OXs ,x ; more precisely, this isomorphism sends (dn t) x ⊗ 1 (where t is a section of OX over
n +1
an open neighbourhood of x in X) to the class of t x ⊗ 1 modulo m′x .
Proof. This follows from (16.4.5) and (16.4.12). □
The preceding corollaries justify the terminology “sheaf of principal parts of order n”.
Proposition (16.4.14). — Let ρ : A → B be a morphism of rings, S a multiplicative subset of B. Then the

canonical homomorphisms
(16.4.14.1) S−1 PB/A

n
−→ PSn−1 B/A
n
deduced from the canonical homomorphisms PB/A → PSn−1 B/A (16.4.4), form a projective system and are IV-4 | 23
bijective.
Proof. It suffices to remark that S−1 (( B ⊗ A B)/In+1 ) = S−1 ( B ⊗ A B)/(S−1 I)n+1 by flatness, and
that S−1 ( B ⊗ A B) = (S−1 B) ⊗ A (S−1 B) (I, 1.3.4). □
Corollary (16.4.15). — The notation being that of (16.4.14), let R be a multiplicative subset of A such that
ρ( R) ⊂ S. Then we have canonical isomorphisms
(16.4.15.1) S−1 PB/A

n
≃ PSn−1 B/R−1 A
forming a projective system.
Proof. It evidently suffices to define canonical isomorphisms
(16.4.15.2) PSn−1 B/A ≃ PSn−1 B/R−1 A
that is to say that we reduce to the case there ρ( R) is consists of invertible elements of B. But then the
isomorphism (16.4.15.2) is simply induced by the canonical isomorphism B ⊗ A B → B ⊗ R−1 A B by
passing to quotients (I, 1.5.3). □
Corollary (16.4.16). — Let f : X → S be a morphism of preschemes, x a point of X, s = f ( x ). Then we

have canonical isomorphisms
n
(14.4.16.1) (PX/S ) x ≃ POn x /Os
forming a projective system.
We deduce from these isomorphisms of the associated graded rings, and in particular a canonical
isomorphism
(16.4.16.2) (Ω1X/S ) x ≃ Ω1Ox /Os .
Corollary (16.4.17). — Let k be a field, K the field of rational functions k( T1 , . . . , Tr ). Then, for every
n which sends U to dn T − T .1 is
integer n, the homomorphism of K [U1 , . . . , Ur ] (Ui indeterminates) into PK/k i i i
surjective and defines an isomorphism from the quotient K [U1 , . . . , Un ]/mn+1 (where m is the ideal generated
n .
by the Ui ) to PK/k
Proof. This follows from (16.4.8), (16.4.10) and (16.4.14), where we take A = k, B = k [ T1 , . . . , Tr ] and
S = B − {0}. □
We thus recover the fact that the dTi form a basis of the K-vector space Ω1K/k (0, 20.5.10).
(16.4.18). Let f : X → Y, g : Y → Z be two morphisms of preschemes, and consider the canonical

homomorphism of augmented OX -algebras (16.4.3.3)
n n
(16.4.18.1) gX/Y/Z : PX/Z −→ PX/Y
(16.4.18.2) f X/Y/Z : f ∗ (PY/Z

n n
) −→ PX/Z .
Then gX/Y/Z is surjective, and its kernel is the sheaf of ideals generated by the image under f X/Y/Z
of the augmentation ideal of f ∗ (PX/Z
n ).
Proof. First note that gX/Y/Z corresponds to the case in (16.4.3.3) where X ′ = X, S′ = Y and S = Z, IV-4 | 24
u = 1X , and f X/Y/Z to the case where we replace X ′ , X, S, S′ by X, Y, Z, Z respectively and u, f by f ,
g respectively.
We have a commutative diagram (I, 3.5)

∆f j
(16.4.18.3) X / X ×Y X / X ×Z X
p f ×z f
#
f
Y / Y ×Z Y
∆g
where j = (1X , 1X ) Z is an immersion, j ◦ ∆ f = ∆ g◦ f , and p is the structure morphism. Since we

can restrict ourselves to the case where X, Y and Z are affine, we can suppose that the immersions
∆ f , ∆ g and j are closed, so that OX and OX ×Y X are identified respectively with OX ×Z X /I and
OX ×Z X /L , where L ⊃ I are two quasi-coherent ideals corresponding respectively to the immer-
sions ∆ g◦ f and j. The OX -algebra PX/Z n is identified with OX ×Z X /I n+1 , and PX/Yn is identified
with OX ×Y X /(I /L ) n + 1 , which is to say with OX ×Z X /(I n + 1 + L ), and therefore with the quotient
of PX/Zn by (I n+1 + L )/I n+1 . But we know (loc. cit) that if p and j make X ×Y X the product of
the (Y × Z Y )-preschemes Y and X × Z X, so if OY is identified to OY ×Z Y /K , where K is the ideal
corresponding to ∆ g , L is equal to ( f × Z f )∗ (K ).OX ×Z X (I, 4.4.5). Since (I n+1 + L )/I n+1 is the
ideal of PX/Z
n generated by the image of L , we deduce the proposition. □
Corollary (16.4.19). — With the notation of (16.4.18), we have an exact sequence of quasi-coherent
OX -modules
f X/Y/Z gX/Y/Z
(16.4.19.1) f ∗ (ΩY/Z
1 ) / Ω1 / Ω1 / 0.
X/Z X/Y
When X, Y, Z are affine, we recover the exact sequence (0, 5.7.1).

Proposition (16.4.20). — Let f : Y → Z be a morphism, j : X → Y a closed immersion, K the quasi-
coherent sheaf of ideals of OY corresponding to j. It follows that PX/Y n = OX = OY /K , the canonical
homomorphism jX/Y/Z : j∗ (PY/Z n ) → Pn
X/Z is surjective, and its kernel is the ideal of j∗ (PY/Z
n ) generated
∗
by j (OY · dY/Z (K )) (it should be noted that dY/Z (K ) is a subsheaf of abelian groups of PX/Z , but not an
n n n
OY -module in general).
Proof. We know (I, 5.3.8) that the diagonal ∆ j : X → X ×Y X is an isomorphism, from which the
first assertion follows. If ϖ1 and ϖ2 are the two canonical homomorphisms of algebras OY → PY/Zn
corresponding respectively to the two canonical projections p1 , p2 of Y × Z Y → Y, recall that by

definition ((16.3.5) and (16.3.6)) ϖ1 is the structure homomorphism of the OY -algebra PY/Z
n and
∗
ϖ2 = dY/Z . The OX -algebra j (PY/Z ) is therefore identified with PY/Z /ϖ1 (K )PY/Z and its
n n n n
quotient by the ideal generated by j∗ (dY/Z

n (K )) to PY/Z
n / ( ϖ (K ) + ϖ (K ))P n . Now note that
1 2 Y/Z
we have a commutative diagram IV-4 | 25
j
Yo X
∆f ∆ f ◦j

Y ×Z Y o X ×Z X
j× Z j
identifying X with the product of the (Y × Z Y )-preschemes Y and X × Z X (I, 5.3.7). Since j × Z j is
an immersion, we therefore deduce from this remark and from (16.2.2) that if ∆Y/Z
n and ∆nX/Z denote
the infinitesimal neighbourhoods of order n of Y and X by the canonical immersions ∆ f and ∆ f ◦ j
respectively, then we have a diagram
∆Y/Z
n o ∆nX/Z

Y ×Z Y o X ×Z X
j× Z j
making ∆nX/Z
the product of the (Y × Z Y )-preschemes ∆Y/Z
n and X × Z X. We can also say that
PX/Z is identified with the sheaf of rings PY/Z ⊗OY× Y OX ×Z X . But we see immediately that
n n
Z
(for example, by restricting to the affine case) that OX ×Z X = OY ×Z Y /( p1∗ (K ) + p2∗ (K ))OY ×Z Y .
Therefore PX/Z
n is identified with the quotient of PY/Zn by the ideal generated by the image in
∗ ∗
PY/Z of p1 (K ) + p2 (K ). But by definition this ideal is generated by ϖ1 (K ) + ϖ2 (K ).
n □
Corollary (16.4.21). — Let f : Y → Z be a morphism, j : X → Y an immersion. We have an exact sequence

of quasi-coherent OX -modules
(16.4.21.1) NX/Y / j ∗ ( Ω1 ) / Ω1 / 0.
Y/Z X/Z
When X, Y, Z are affine, we recover the exact sequence (0, 20.5.12.1).

Corollary (16.4.22). — If f : X → S is a morphism locally of finite presentation, PX/S
n and Ω1X/S are
quasi-coherent OX -modules of finite presentation.
Proof. We immediately reduce to the case where S = Spec( A) is affine, X = Spec( B), where
B = A[ T1 , . . . , Tr ]/K, K being an ideal of finite type of C = A[ T1 , . . . , Tr ]. Applying (16.4.20) where
Z = S, Y = Spec(C ) and K = K. e Then j∗ (P n ) is a free OX -module of finite rank (16.4.10) and
Y/Z
the hypothesis on K implies that j∗ (OY .dY/Z n (K )) generates a quasi-coherent OX -module of finite
type; hence the conclusion. □
Proposition (16.4.23). — Let X, Y be two S-preschemes, Z = X ×S Y their product, p : X ×S Y → X and
q : X ×S Y → Y the canonical projections. Then the canonical homomorphism
(16.4.23.1) p Z/X/S ⊕ q Z/Y/S : p∗ (Ω1X/S ) ⊕ q∗ (ΩY/S
1
) −→ Ω1(X × S Y ) /S
is bijective.
Proof. The commutative diagram IV-4 | 26
q
Yo X ×S Y o
id
X ×S Y
g h p

So So X
id f
gives us a factorization of the canonical isomorphism Pn ( p) (16.4.5)

p∗ (PX/S
n n
) −→ PZ/S n
−→ PZ/Y
and similarly, switching X with Y, we have a factorization of the isomorphism Pn (q)
q∗ (PY/S
n n
) −→ PZ/S n
−→ PZ/Y .
This proves that the canonical homomorphism (16.4.18.1)
p Z/X/S : p∗ (PX/S
n n
) −→ PZ/S (resp. q Z/X/S : q∗ (PY/S
n n
) −→ PZ/S )
is injective, and that the kernel of the canonical surjective homomorphism (16.4.18.2)
n n n n
PZ/S −→ PZ/Y (resp. PZ/S −→ PZ/X )
is direct summand of the image p Z/X/S (resp. q Z/Y/S ). On the other hand, this kernel is, by virtue of
(16.4.18), generated by the image by q Z/Y/S (resp. p Z/X/S ) of the augmentation ideal of q∗ (PY/S n )
(resp. p∗ (PX/S
n )). We conclude the proposition by considering the case n = 1. □
We immediately generalize (16.4.23) to the case of a product of any finite number of S-preschemes.
Remarks (16.4.24). —
(i) We will see (17.2.3) that when the morphism f : X → Y in (16.4.18) is smooth, the homo-
morphism f X/Y/Z in (16.4.19.1) is locally left invertible and in particular injective. Similarly,
when the morphism f ◦ j : X → Z of (16.4.20) is smooth, the homomorphism on the left
in (16.4.21.1) is locally left invertible and a fortiori injective (17.2.5). In Chapter V, we will
also give a variant, in the case of modules over a prescheme, of the “imperfection modules”
studied in (0, 20.6), and the exact sequences where they occur.
(ii) Let X be a topological space, A a sheaf of rings over X and B a A -algebra over X. Then it
is clear that
U 7−→ PΓn(U,B )/Γ(U,A ) (U open in X)
is a presheaf of augmented Γ(U, B )-algebras, and therefore the associated sheaf PB n
/A is
an augmented B-algebra. In the particular case where X is a prescheme, f = (ψ, θ ) : X → S
a morphism of preschemes, it follows easily from (16.4.16) and from the exactness of the
functor lim that PX/Sn is canonically isomorphic to PO n . It follows that the formalism
X /ψ (OS )
∗
−→
developed in the present paragraph could be considered as a particular case of a differential IV-4 | 27
formalism for ringed spaces endowed with a sheaf of algebras over the structure sheaf.
However, we did not start with this point of view, which is less intuitive and less convenient
for applications. It also seems that, for various kinds of “varieties”, the “global” constructions
of the P n analogous to those we have used here are also better suited for applications.
16.5. Relative tangent sheaves and bundles; derivations.
(16.5.1). Let f = (ψ, θ ) : X → S be a morphism of ringed spaces. For every OX -module F , we say
S-derivation (or ( X/S)-derivation, or f -derivation) of OX to F for every homomorphism of sheaves of
additive groups D : OX → F satisfying the following conditions:
(a) for every open V of X, and all pair of sections (t1 , t2 ) of OX over V, we have
(16.5.1.1) D ( t1 t2 ) = t1 D ( t2 ) + D ( t1 ) t2 ;
(b) for every open V of X, every section t of OX over V, and every section s of OS over an open
U of S such that V ⊂ f −1 (U ), we have
(16.5.1.2) D ((s|V )t) = (s|V ) D (t).
It is clear that this amounts to saying that, for all x ∈ X, the homomorphism of additive groups
Dx : Ox → Fx is an O f ( x) -derivation.
Another interpretation consists of considering the OX -algebra DOX (F ) as equal to OX ⊕ F ,
the algebra structure being defined by the condition that for every open V of X, the product of
two sections of OX (resp. of a section of OX and a section of F ) over V is defined by the ring
structure of Γ(V, OX ) (resp. the Γ(V, OX )-module structure on Γ(V, F )), and the product of two
sections of F over V is chosen to be zero; then F is an ideal of DOX (F ), the kernel of the canonical
augmentation DOX (F ) → OX , and to say that D is an S-derivation of OX to F means that 1OX + D
is an OS -homomorphism of algebras from OX to DOX (F ), which, composed with the augmentation,
gives 1OX .
The S-derivations of OX to F clearly form a Γ( X, OX )-module DerOS (OX , F ).
When F = OX , an S-derivation of OX to itself is simply called an S-derivation of OX .
Proposition (16.5.2). — Let A be a ring, B an A-algebra, L a B-module; let S = Spec( A), X = Spec( B),
F =e L. Then the map D 7→ Γ( D ) which sends every S-derivation D of OX to F to the map Γ( D ) : t 7→ D (t)
of B to L, is an isomorphism of B-modules from DerS (OX , F ) to Der A ( B, L) (cf. (0, 20.1.2)).
Proof. This follows immediately from the given interpretation of S-derivations in terms of homo- IV-4 | 28
morphisms of algebras, analogous to the interpretation given in (0, 20.1.6), and from the canonical
correspondence between homomorphisms of OX -algebras and homomorphisms of B-algebras
((I, 1.3.13) and (I, 1.3.8)). □
Proposition (16.5.3). — Let f = (ψ, θ ) : X → S be a morphism of preschemes.
(i) The differential d X/S : OX → Ω1X/S (16.3.6) is an S-derivation.
(ii) For every OX -module F , the map u 7→ u ◦ d X/S is an isomorphism of Γ( X, OX )-modules
(16.5.3.1) HomOX (Ω1X/S , F ) ≃ DerS (OX , F ).
Proof. The assertion (i) has already been written (16.3.6). On the other hand, it is immediate (in
light of (0, 20.4.8)) that u 7→ u ◦ d X/S is injective, considering the restrictions to a fibre Ox of the
two sides and using (16.4.16.2). To see that the homomorphism (16.5.3.1) is surjective, consider an
S-derivation D : OX → F ; for every affine open V = Spec( B) of X, such that f (V ) is contained in
an affine open U = Spec( A) of S, DV : B → Γ(V, F ) is an A-derivation, and therefore there exists a
unique B-homomorphism uV : Ω1B/A → Γ(V, F ) such that DV = uV ◦ d B/A (0, 20.4.8); in addition,
the uniqueness of uV shows immediately that for an affine open W ⊂ V we have uW = uV |W, and
therefore the uV define a homomorphism of OX -modules u : OX → F answering the question. □
(16.5.4). With the notation of (16.5.1), for every open U of X, DerS (OU , F |U ) is a Γ(U, OX )-module
and it is clear that the map U 7→ DerS (OU , F |U ) is a presheaf; in fact, it is even a sheaf (and therefore
an OX -module), in light of the pointwise characterization of S-derivations, seen in (16.5.1). This
OX -module is denoted by D erS (OX , F ) and is called the sheaf of S-derivations of OX in F , and what
we have seen is further expressed in the following corollary:
Corollary (16.5.5). — For every OX -module F , the homomorphism of OX -modules induced by u 7→
u ◦ d X/S
(16.5.5.1) H omOX (Ω1X/S , F ) −→ D erS (OX , F )
is bijective.
Corollary (16.5.6). — (i) If the morphism f : X → S is locally of finite presentation and if F is a
quasi-coherent OX -module, then D erS (OX , F ) is a quasi-coherent OX -module.
(ii) If in addition S is locally Noetherian and if F is coherent, then D erS (OX , F ) is a coherent OX -
module.
Proof. The assertion (i) follows from the isomorphism (16.5.5.1), from (16.4.22), and (I, 1.3.12); the
assertion (ii) follows from (0, 5.3.5). □
(16.5.7). We set
(16.5.7.1) GX/S = H omOX (Ω1X/S , OX ) = D erS (OX , OX ),
and say that it is the sheaf of S-derivations of OX , or even the tangent sheaf of X relative to S: it is therefore
the dual of the OX -module Ω1X/S . If f is locally of finite presentation, GX/S is a quasi-coherent OX - IV-4 | 29
module; if in addition S is locally Noetherian, then GX/S is coherent (16.5.6).
(16.5.8). Suppose in particular that Ω1X/S is a locally free OX -module (of finite rank) (which will be
the case then f is smooth (17.2.3)); then GX/S is locally free OX -module of the same rank as Ω1X/S at
each point. More specifically, suppose that Ω1X/S is of rank n at a point x; then there are n sections si
(1 ⩽ i ⩽ n) of OX over an affine neighbourhood U of x such that the canonical images of the dsi in
Ω1X/S ⊗OX k ( x ) form a basis if this k( x )-vector space; by virtue of Nakayama’s lemma, the germs
(dsi ) x = d(si ) x of the dsi at the point x form a basis of the Ox -module (Ω1X/S ) x , and therefore, by
restricting U, we can suppose that the dsi form a basis of the Γ(U, OX )-module Γ(U, Ω1X/S ). So the

Γ(U, OX )-module Γ(U, GX/S ) is dual to the above; we denote by ( Di )1⩽i⩽n or ∂s∂ the dual
i 1⩽i⩽n
basis of (dsi )1⩽i⩽n , so that, by (16.5.3), we have

∂
(16.5.8.1) Di s j = ⟨ Di , ds j ⟩ = , ds j = δij (Kronecker’s symbol).
∂si
Every Γ(S, OS )-derivation of the Γ(S, OS )-algebra Γ(U, OX ) is therefore written in an unique way as
n n
∂
D= ∑ ai Di = ∑ ai ∂si ,
i =1 i =1
where the ai (1 ⩽ i ⩽ n) are sections of OX over U. For every section g ∈ Γ(U, OX ), if we put
dg = ∑in=1 ci dsi , then we have ci = ⟨ Di , dg⟩ = Di g by virtue of (16.5.8.1), in other words,
n n
∂g
(16.5.8.2) dg = ∑ ( Di g)dsi = ∑ ∂si dsi .
i =1 i =1
(16.5.9). Let D1 , D2 be two S-derivations of OX . For every open U of X, if D1U , D2U are the corre-
sponding derivations of the ring Γ(U, OX ), the bracket
[ D1U , D2U ] = D1U ◦ D2U − D2U ◦ D1U
is also a derivation in this ring, and therefore the ψ∗ (OS )-endomorphism of OX

(16.5.9.1) [ D1 , D2 ] = D1 ◦ D2 − D2 ◦ D1
is also an S-derivation; as we immediately check that this bracket satisfies the Jacobi identity, we have
thus defined on DerS (OX , OX ) a Γ(S, OS )-Lie algebra structure. Since the definition of this structure
commutes with the restriction to an open of X, we thus see that GX/S is canonically equipped with
a ψ∗ (OS )-Lie algebra structure. Note that the mapping ( D1 , D2 ) 7→ [ D1 , D2 ] is not Γ( X, OX )-bilinear.
(16.5.10). For every base change g : S′ → S, if we set X ′ = X ×S S′ , then we see (16.4.5) that we have
a canonical isomorphism
(16.5.10.1) Ω1X/S ⊗S S′ ≃ Ω1X ′ /S′ ,
from which we deduce, by (16.5.10.1), a canonical homomorphism (Bourbaki, Alg., chap. II, 3rd ed., IV-4 | 30
§5, no 3)
(16.5.10.2) GX/S ⊗OS OS′ −→ GX ′ /S′ ,
which is neither injective nor surjective in general. However:
Proposition (16.5.11). — (i) If g : S′ → S is a flat morphism and if f is locally of finite type (resp.
locally of finite presentation), then the homomorphism (16.5.10.2) is injective (resp. bijective).
(ii) If Ω1X/S is a locally free OX -module of finite type, then the homomorphism (16.5.10.2) is bijective.
Proof. The assertion (ii) follows from Bourbaki, Alg., chap. II, 3rd ed., §5, no 3, prop. 7. The assertion
(i) follows similarly from Bourbaki, Alg. Comm., chap. I, §2, no 10, prop. 11 and from the fact that if f
is locally of finite type (resp. locally of finite presentation), then Ω1X/S is an OX -module of finite type
(resp. of finite presentation) ((16.3.9) (16.4.22)). □
(16.5.12). Since Ω1X/S is a quasi-coherent OX -module, we can consider the vector bundle over X
defined by Ω1X/S (II, 1.7.8)
(16.5.12.1) TX/S = V(Ω1X/S )

which is called the tangent bundle of X relative to S. We have therefore a canonical bijection (II, 1.7.9)
Γ( TX/S /S) ≃ HomOX (Ω1X/S , OX ) = Γ( X, GX/S )
by definition of GX/S , and we can replace X by an open set U of X in this isomorphism; so we can
say that the tangent sheaf of X relative to S is isomorphic to the sheaf of germs of S-sections of the
tangent bundle of X relative to S. If f : X → Y is an S-morphism, we saw (16.4.19) that we have a
canonical homomorphism f X/Y/S : f ∗ (ΩY/S 1 ) → Ω1
X/S , which, having in mind that
V( f ∗ (ΩY/S
1 1
)) = V(ΩY/S ) ×Y X (II, 1.7.11) ,
gives us an X-morphism TX/S ( f ) : TX/S → TY/S ×Y X. If g : Y → Z is a second S-morphism, we
have TX/S ( g ◦ f ) = ( TY/S ( g) × 1X ) ◦ TX/S ( f ) (0, 20.5.4.1).
It follows from (16.5.10.1) and from (II, 1.7.11) that for every base change g : S′ → S we have a
canonical isomorphism
(16.5.12.2) TX ′ /S′ ≃ TX/S ×S S′ = TX/S × X X ′ .
(16.5.13). For every point x ∈ X, we define the tangent space of X at the point x (relative to S) to be the
set of points in the fibre TX/S × X Spec(k( x )) that are rational over k ( x ), that is, the set
(16.5.13.1) TX/S ( x ) = Homk( x) (Ω1X/S ⊗Ox k( x ), k( x )),
which is the dual of the k( x )-vector space Ω1Ox /Os /mx · Ω1Ox /Os . When Ω1X/S is an OX -module of finite
type, then TX/S ( x ) is a vector space of finite rank over k ( x ), and for every base change g : S → S′ , IV-4 | 31
and every point x ′ ∈ X ′ = X ×S S′ over x, we have a canonical isomorphism
(16.5.13.2) TX ′ /S′ ( x ′ ) ≃ TX/S ⊗k( x) k ( x ′ ).
If x is rational over k(s), where s = f ( x ) (so that k(s) → k( x ) is an isomorphism), it follows from
(16.4.13) that we have a canonical isomorphism
′
(16.5.13.3) TX/S ( x ) = TXs /k(s) ( x ) = Homk(s) (m′x /mx2 , k( x )),
where m′x is the maximal ideal of OXs ,x = OX,x /ms OX,x . In the case where S is the spectrum of a
field k, we recover the definition of the Zariski tangent space of a point x ∈ X rational over k, as the
dual of mx /m2x .
Let Y be a second S-prescheme and let g : Y → X be an S-morphism; then we have a canonical
homomorphism of OY -modules (16.4.19)
(16.5.13.4) gY/X/S : g∗ (Ω1X/S ) −→ ΩY/S
1
.
Now note that if y ∈ Y and x = g(y), then we have
g∗ (Ω1X/S ) ⊗OY k (y) = (Ω1X/S ⊗OX k( x )) ⊗k( x) k(y)
and consequently, if Ω1X/S is an OX -module of finite type, then we can identify
Homk(y) ( g∗ (Ω1X/S ) ⊗OY k(y), k (y))

with TX/S ( x ) ⊗k( x) k(y). We therefore deduce from the homomorphism (16.5.13.4) a homomorphism
of k (y)-vector spaces
(16.5.13.5) Ty ( g) : TY/S (y) −→ TX/S ( x ) ⊗k( x) k(y)
called the linear map tangent to g at the point y. When y is rational over k(s), we can identify k(s),
k(y), and k( x ), and Ty ( g) is then a homomorphism of k(s)-vector spaces TY/S (y) → TX/S ( x );
also note that in this case, g∗ (Ω1X/S ) ⊗OY k(y) is identified with Ω1X/S ⊗OX k( x ), and the above
homomorphism is therefore defined without any finiteness conditions on Ω1X/S and it is none other
than the homomorphism TY/S ( g) (16.5.12) restricted to the fibre of TY/S at the point y.
(16.5.14). The interpretation of derivations of an A-algebra B to a B-module L, given in (0, 20.1.1),
translates to the language of preschemes in the following way.
Consider two morphisms of preschemes f : X → S, g : Y → S, and a closed subprescheme Y0
of Y defined by a zero-square ideal J of OY (so that Y and Y0 have the same underlying subspace).
Suppose we are given an S-morphism u0 : Y0 → X, so that we have a commutative diagram
u0
(16.5.14.1) X _o Y0
u
f j

So g
Y
and we suggest looking for an S-morphism u : Y → X such that u0 = u ◦ j (in other words, if it is IV-4 | 32
possible to complete the diagram above by the dotted arrow u, keeping it commutative).
For that, consider an affine open U = Spec(C ) of Y; its inverse image j−1 (U ) is the affine open
U0 = Spec(C/L), where L = Γ(U, J ), a zero-square ideal in C; suppose that U is small enough so
that u0 (U0 ) is contained in an affine open V = Spec( B) of X and that g(U ) = f (u0 (U0 )) is contained
in an affine open W = Spec( A) of S, so that B and C are A-algebras and u0 |U0 corresponds to
an A-homomorphism ψ from B to C/L; Let P(U0 ) be the set of restrictions u|U of the sought
homomorphisms, which corresponds canonically to A-homomorphisms of algebras ϕ : B → C
ϕ
such that the composite B − → C → C/L is equal to ψ. So we know (0, 20.1.1) that the set of such
homomorphisms is either empty or of the form ϕ1 + Der A ( B, L); when P(U0 ) is not empty, the
additive group Der A ( B, L) acts by addition on P(U0 ), which is therefore an affine space for the
additive group Der A ( B, L) (or even a principal homogeneous space (or torsor) under Der A ( B, L)).
Now notice that, since L is equipped with a B-module structure via ψ, we have an isomor-
phism v 7→ v ◦ d B/A of HomB (Ω1B/A , L) onto Der A ( B, L) (0, 20.4.8). Besides, as L is square-zero,
therefore a (C/L)-module, every B-homomorphism v : Ω1B/A → L can be considered as a (C/L)-
homomorphism Ω1B/A ⊗ B (C/L) → L. As I is square-zero, it can be considered as a quasi-coherent
OY0 -module; let’s introduce the OY0 -module

(16.5.14.2) G = H om(u0∗ (Ω1X/S ), I );
it follows from the fact that Ω1B/A = Γ(V, Ω1X/S ) (16.3.7) that we can write Der A ( B, L) = Γ(U0 , G ).
As P(U0 ) is defined as a set of S-morphisms U → X, it is clear that U0 7→ P(U0 ) is a sheaf of sets
P on Y0 . We can use this fact to prove that the map h : Γ(U0 , G ) × P(U0 ) → P(U0 ) defining the
torsor structure on P(U0 ) is independent of choice of V and W and also that, if U ′ ⊂ U is a second
affine open of Y, U0′ its inverse image in Y0 , then the diagram
(16.5.14.3) Γ(U0 , G ) × P(U0 )

h / P(U0 )

Γ(U0′ , G ) × P(U0′ ) / P (U ′ )
h′ 0
is commutative (the vertical arrows being the restrictions). In light of the above remark, we reduce
to proving the commutativity of the above diagram when h is defined as such from affine opens V,
W and h′ from affine opens V ′ ⊂ V and W ′ ⊂ W. But because of the preceding description of h, this IV-4 | 33
follows from the commutativity of the diagram (0, 20.5.4.2).
The mapping Γ(U0 , G ) × P(U0 ) → P(U0 ) therefore define a homomorphism of sheaf of sets m :
G × P × P such that, for all open sets U0 for which Γ(U0 , P ) ̸= ∅, mU0 : Γ(U0 , G ) × Γ(U0 , P ) →
Γ(U0 , P ) is an external law defining in Γ(U0 , P ) a torsor structure for the group Γ(U0 , G ).
(16.5.15). In general, when we are given a sheaf of sets P over a topological space Z, a sheaf of
groups G (not necessarily commutative), and a homomorphism of sheaves of sets m : G × P → P
such that, for every open U ⊂ Z such that Γ(U, P ) ̸= ∅, mU : Γ(U, G ) × Γ(U, P ) → Γ(U, P )
makes Γ(U, P ) a torsor under the group Γ(U, G ), then we say that P is a pseudo-torsor (or formally
principal homogeneous sheaf ) under the sheaf of groups G . We say that P is a torsor (or principal
homogeneous sheaf ) under G 3 if in addition Γ(U, P ) ̸= ∅ for every open U ̸= ∅ in a suitable basis
for the topology of Z.
For the general theory of torsors, we refer to [eAG64]; we will limit ourselves to recalling the
canonical correspondence between isomorphism classes of torsors (for a given G ) and elements
from the cohomology set H1 ( Z, G ). Consider a torsor P under G and an open cover (Uλ ) of
Z such that Γ(Uλ , P ) ̸= ∅ for every λ; denote by pλ an element of Γ(Uλ , P ). For every pair
of indices λ, µ such that Uλ ∩ Uµ ̸= ∅, there then exists a unique element γλµ of Γ(Uλ ∩ Uµ , G )
such that γλµ · ( pµ |Uλ ∩ Uµ ) = pλ |Uλ ∩ Uµ ; in addition, if λ, µ, ν are three indices such that
Uλ ∩ Uµ ∩ Uν ̸= ∅, then the restrictions γλµ ′ , γ′ , γ′ of γ , γ , γ
µν λν λµ µν λν to Uλ ∩ Uµ ∩ Uν satisfy
′ ′ ′
the condition γλν = γλµ γµν ; in other words, (λ, µ) 7→ γλµ is a 1-cocycle of the cover (Uλ ) with
values in G . If, for every λ, p′λ is a second element of Γ(Uλ , P ), then there exists a unique element
β λ ∈ Γ(Uλ , G ) such that p′λ = β λ · pλ , and the 1-cocycle (γλµ
′ ) corresponding to the family ( p′ ) is
λ
given by γλµ′ = β γ β−1 , that is, it is cohomologous to γ . Conversely, the data of a 1-cocycle ( γ )
λ λµ µ λµ λµ
defines, for every pair (λ, µ), an automorphism θλµ of the sheaf of sets G |Uλ ∩ Uµ , namely the right
translation by γλµ , and the fact that it is a cocycle shows that we can glue the sheaves of sets G |Uλ
via the automorphisms θλµ (0, 3.3.1); we thus obtain a torsor under G , denoted P, and if we take for
pλ the unit section over Uλ , then the corresponding 1-cocycle is none other than the given 1-cocycle
′ = β γ β−1 cohomologous to it, then we
(γλµ ); in addition, if we replace (γλµ ) by a 1-cocycle γλµ λ λµ µ
check immediately that the torsor obtained is isomorphic to P.
In particular, if (γλµ ) is a 1-coboundary, in other words of the form γλµ = β λ β− 1
µ , then the torsor
P obtained is isomorphic to G (considered as a torsor under itself by left translations); we say in this
case that P is trivial, and the converse is evident.
In particular, it follows from (III, 1.3.1) that we have:
Proposition (16.5.16). — Let Z be an affine scheme, G a quasi-coherent OZ -module; then every torsor over
G is trivial.
3[Trans.] This is nowadays more commonly called a G -torsor rather then a torsor under G .
Returnning to the problem considered in (16.5.13), we thus obtain: IV-4 | 34

Proposition (16.5.17). — Let X, Y be two S-preschemes, Y0 a closed subprescheme of Y defined by a
quasi-coherent ideal I of OY such that I 2 = 0, j : Y0 → Y the canonical injection. Let u0 : Y0 → X
be an S-morphism, and P the sheaf of sets on Y such that, for every open U of Y, Γ(U, P ) is the set of
S-morphisms u : U → X such that u0 |U0 = u ◦ ( j|U0 ), where U0 = j−1 (U ). Then there exists on P the
structure of a pseudo-torsor over the OY0 -module G = H omOY (u0∗ (Ω1X/S ), I ).
0
In particular:
Corollary (16.5.18). — With the notation of (16.5.16), suppose that Y is affine and Ω1X/S is of finite
presentation; if there is a open cover (Uα ) of Y, and, for every index α, an S-morphism vα : Uα → X such
that, if Uα0 = j−1 (Uα ), we have vα ◦ ( j|Uα0 ) = u0 |Uα0 , then there is an S-morphism u : Y → X such that
u ◦ j = u0 .
Proof. Indeed, G is a quasi-coherent OY0 -module (I, 1.3.12); by (16.5.16) and the fact that Y0 is then
affine, the sheaf P, which is by hypothesis a torsor over G , and not only a pseudo-torsor, is trivial;
but if w is an isomorphism from G to P (as it is a torsor over G ), the image under w of the zero
section of G is the S-morphism we want. □
16.6. Sheaf of p-differentials and exterior differentials.

(16.6.1). Let f : X → S be a morphism of preschemes. We define the sheaf of p-differentials of X relative
to S (p integer) to be the p’th exterior power (0, 4.1.5) of the OX -module Ω1X/S , denoted by
p
p
Ω X/S (Ω1X/S ).
^
(16.6.1.1) =
p p
So we have Ω0X/S = OX , and Ω X/S = 0 for p < 0; the Ω X/S are the homogeneous components of
the exterior algebra of Ω1X/S
p
Ω•X/S (Ω1X/S ) (Ω1X/S ),
^ M^
(16.6.1.2) = =
p∈Z
which is therefore a quasi-coherent graded anti-commutative OX -algebra whose elements of degree 1

are square-zero. For every affine U of X, we have Γ(U, Ω•X/S ) = (Γ(U, Ω1X/S )), where Γ(U, Ω1X/S )
V
is considered as a Γ(U, OX )-module.

When S = Spec( A) and X = Spec( B) are affines, B being then an A-algebra, we have (0, 4.1.5)
p p p
Ω X/S = (Ω B/A )∼ , by putting Ω B/A = p Ω1B/A .
V
Theorem (16.6.2). — There is one and only one endomorphism d of the sheaf of additive groups Ω•X/S with
the following properties:
(i) d ◦ d = 0.
(ii) For every open set U of X and every section f ∈ Γ(U, OX ) we have d f = d X/S f . IV-4 | 35
′ p
(iii) For every open set U of X, every pair of integers p, q and every pair of sections ω p ∈ Γ(U, Ω X/S ),
q
ωq′′ ∈ Γ(U, Ω X/S ), we have
(16.6.2.1) d(ω ′p ∧ ωq′′ ) = (dω ′p ) ∧ ωq′′ + (−1) p ω ′p ∧ dωq′′ .
Also, d is an endomorphism of graded ψ∗ (OX )-modules of degree +1.
Proof. Suppose that we have proved the existence of an endomorphism d. For every affine open
p
U of X, every section of Ω X/S over U is (because of (ii)) a linear combination if a finite number of
elements of the form g(d f 1 ∧ d f 2 ∧ · · · ∧ d f p ), where g and the f i are sections of OX over U (0, 20.4.7).
The conditions (i) and (iii) then show, by induction on p, that we necessarily have
(16.6.2.2) d( g(d f 1 ∧ d f 2 ∧ · · · ∧ d f p )) = dg ∧ d f 1 ∧ d f 2 ∧ · · · ∧ d f p .
This therefore proves the uniqueness of d and the last claim of the theorem. By virtue of this
uniqueness property, to show the existence of d, we can restrict ourselves to the case where S =
Spec( A) and X = Spec( B) are affines. Now (Bourbaki, Alg., chap. III, 3rd ed., §10) to define an
V
A-derivation D of degree +1 of an exterior algebra ( M ) (where M is a B-module and an A-
algebra), such derivation taking its values in a graded anti-commutative A-algebra C = ∞
L
n=0 Cn ,
whose elements of degree 1 are square-zero, it suffices to give arbitrarily an A-derivation D0 of B in
C1 and an A-homomorphism D1 of M in C2 ; then it exists one and only one A-anti-derivation D of
Λ( M) in C coinciding with D0 in B and D1 in M.
In the present case, D0 is necessarily equal to d B/A by (ii); we reduce to seeing, having (16.6.2.2)
in mind, that there is an A-homomorphism u of Ω1B/A in Ω2B/A such that
(16.6.2.3) u( g.d f ) = dg ∧ d f
for whichever f , g in A; it suffices to show that there is an A-homomorphism v : B ⊗ A Ω1B/A → Ω2B/A
such that
(16.6.2.4) v( g.ω ) = dg ∧ ω
for g ∈ B and ω ∈ Ω1B/A .Finally, since Ω1B/A
= I/I2 (where I = IB/A is the kernel of the canonical
homomorphism B × A B → B) and that Ω1B/A
is generated by elements of the form g.d f , it is enough
to define an A-homomorphism w : B ⊗ A ( B ⊗ A B) → Ω2B/A such that
(16.6.2.5) w( g′ ⊗ g ⊗ f ) = dg′ ∧ ( g.d f )
and such that w is zero on the image of B ⊗ AI2 . Or, since the second member of (16.6.2.5) is
A-trilinear in g′ , g and f , the existence of w verifying (16.6.2.5) is immediate. Since, on the other
hand, I is generated by elements of the form 1 ⊗ x − x ⊗ 1 (x ∈ B), we reduce to checking that when
z = (1 ⊗ x − x ⊗ 1)(1 ⊗ y − y ⊗ 1) we have w( g′ ⊗ z) = 0. Or, since z = 1 ⊗ ( xy) + ( xy) ⊗ 1 − x ⊗
y − y ⊗ x, the formula (16.6.2.4) shows that it is enough to see that we have d( xy) − x.dy − y.dx = 0, IV-4 | 36
which is to say that d is a derivation.
It remains to be shown that d verifies the condition (i). Now, the square of an anti-derivation is a
derivation (Bourbaki, loc. cit.), and since Ω•B/A is generated by Ω1B/A as a B-algebra, it is enough
to verify that d(dz) = 0 for z ∈ B and x ∈ Ω1B/A ; in the first case, this follows from the formula
(16.6.2.3) when g = 1; for the second, we can restrict ourselves to the case where z = g.d f with f , g
in B, and then we have, because of (16.6.2.1) and (16.6.2.3),
d(d( g.d f )) = d(dg ∧ d f ) = (d(dg)) ∧ (d f ) − (dg) ∧ (d(d f )) = 0.
□
Definition (16.6.3). — The anti-derivation d defined in (16.6.2) (also denoted by d X/S ) is called the
exterior differential on X (relative to S).
Proposition (16.6.4). — For every base change g : S′ → S, if we put X ′ = X ×S S′ , the canonical morphism
(16.6.4.1) Ω•X/S ⊗S S′ −→ Ω•X ′ /S′
deduced from the isomorphism (16.5.10.1) is bijective. Also, if s is a section of Ω•X/S over an open set U of
X, s ⊗ 1 its inverse image, section of Ω•X ′ /S′ over the inverse image U ′ of U in X ′ , we have d X ′ /S′ (s ⊗ 1) =
d X/S (s) ⊗ 1.
Proof. The first claim is immediate, the formation of the exterior algebra of a module commutes with
extending the scalar ring. To prove the second, we can, because of (16.6.2.2), restrict ourselves to the
case where s ∈ Γ(U, OX ), and in this case the claim has already been proven (16.4.3.7). □
(16.6.5). Suppose that Ω1X/S is an locally free OX -module of rank n in a point x, so that we have n
sections si ∈ Γ(U, OX ) such that the dsi form a basis for the Γ(U, OX )-module Γ(U, Ω1X/S ) (16.5.8).
Then, for every integer p ⩾ 1, the p-differentials dsi1 ∧ dsi2 ∧ · · · ∧ dsi p (for i1 ⩽ i2 ⩽ . . . ⩽ i p elements
p
of [1, n]) form a basis of (np) elements of Γ(U, Ω X/S ) over Γ(U, OX ). Also the formula (16.6.2.2) shows
that for every section g ∈ Γ(U, OX ), we have
∂g
(16.6.5.1) d( g.dsi1 ∧ dsi2 ∧ · · · ∧ dsi p ) = ∑(−1)r ∂sk dsi1 ∧ · · · ∧ dsir ∧ dsk ∧ dsir+1 ∧ · · · ∧ dsi p
k
where, in the second member, k varies in the set of the n − p indexes different from the ih , ir being
the biggest index < k.
We note that the relation d(dg) = 0 for every section g ∈ Γ(U, OX ) expresses itself in the form
Di ( D j g ) = D j ( Di g ) for i ̸= j;
in other words, the derivations Di defined in (16.5.7) commute with each other.
16.7. The PX/S
n (F ).
(n)
(16.7.1). Let f : X → S be a morphism of preschemes, F an OX -module. We denote by X∆ the n’th
f
(n)
infinitesimal neighbourhood of X via the diagonal morphism ∆ f : X → X ×S X, by hn : X∆ → X ×S X IV-4 | 37
f
the canonical morphism (16.1.2), and consider the two composite morphisms
(n) (n) h
n p1 (n) n (n) h p2
p1 : X∆ −→ X ×S X −→ X, p2 : X∆ −→ X ×S X −→ X
f f
(n)
so that, by definition, p1 corresponds to the homomorphism of sheaves of rings OX → PX/S
n which
(n)
we have chosen to define the OX -algebra structure on PX/S
n (16.3.5), and p2 to the homomorphism
(n)
of sheaves of rings dnX/S : OX → PX/S
n (16.3.6). Since X∆ and X have the same underlying subspace,
f
we can write
(n) (n)
(16.7.1.1) n
PX/S = ( p1 )∗ (( p2 )∗ (OX )).
More generally, we define
(n) (n)
(16.7.1.2) n
PX/S (F ) = ( p1 )∗ (( p2 )∗ (F )).
so that PX/S
n = PX/S
n (O ); by definition, P n (F ) is an O -module.
X X/S X
(16.7.2). If we come back to the definition of the inverse image of modules on ringed spaces (0, 4.3.1)
(n)
and having in mind that X∆ and X have the same underlying space, we see that we can write the
f
definition (16.7.1.2) in the form
n n
(16.7.2.1) PX/S (F ) = PX/S ⊗O X F ,
but where you have to be careful that, in the interpretation of the symbol ⊗, PX/S
n is endowed with
the structure of OX -module defined by the homomorphism of sheaves of rings d X/S : OX → PX/S
n n . It
follows immediately from such formula (or directly from (16.7.1.2)) that PX/S n (F ) is canonically
equipped with a PX/S -module structure.

n
Proposition (16.7.3). — (i) The functor F 7→ PX/S n (F ) from the category of O -modules to the
X
category of PX/S
n -modules is right exact, and commutes with arbitrary inductive limits; it is exact
when PX/Sn is flat.

(ii) If F is a quasi-coherent OX -module (resp. of finite type, resp. of finite presentation), then PX/S
n (F )
is quasi-coherent (resp. of finite type, resp. of finite presentation).

Proof. The claims from (i) follow from (16.7.2.1) and the consideration of the symmetry of PX/S
n
(16.3.4). The claims from (ii) follow from the right exactness of the functor F 7→ PX/S (F ).
n □
(16.7.4). The two structures of OX -module on PX/S n define in PX/S
n (F ) two structures of O -
X
modules, which happen to be permutable, and therefore an OX -bimodule structure. It is convenient
to denote the structure coming from the structure homomorphism OX → PX/S n (chosen in (16.3.5))
on the left and the one coming from the homomorphism d X/S : OX → PX/S on the right. On other
n n
words, for every open U of X, and every triplet a ∈ Γ(U, OX ), b ∈ Γ(U, PX/S
n ), t ∈ Γ (U, F ), we
have by definition
(16.7.4.1) a(b ⊗ t) = ( ab) ⊗ t, (b ⊗ t) a = (b.dn a) ⊗ t = b ⊗ ( at) = (dn a).(b ⊗ t).
The OX -module structure coming from the definition (16.7.1.2) is therefore, under these conventions,
the left OX -module structure. If F is a quasi-coherent OX -module, then the same is true for PX/S
n (F ) IV-4 | 38
for any one of its OX -module structures. If also F is of finite type (resp. of finite presentation) and
f : X → S is locally of finite type (resp. locally of finite presentation), PX/S
n (F ) is (for any one of its
OX -module structures) of finite type (resp. of finite presentation), which is a consequence of (16.3.9)
and (16.4.22).
(16.7.5). The definition (16.7.2.1) entails the existence of a homomorphism of sheaves of commutative
groups
(16.7.5.1) dnX/S,F : F −→ PX/S
n
(F ) (also denoted dnX/S )
such that, in the notations of (16.7.4), we have
(16.7.5.2) dnX/S,F (t) = 1 ⊗ t
and consequently, because of (16.7.4.1)
(16.7.5.3) dnX/S,F ( at) = (1 ⊗ t) a = (dnX/S,F (t)).a
(IV.16.7.5.4) dnX/S,F ( at) = (dnX/S,F ( a)).(1 ⊗ t) = (dnX/S,F ( a))(dnX/S,F (t)).

Therefore it is OX -linear for the structure of right OX -module on PX/S
n (F ), and semilinear (relative
to the homomorphism σ (16.3.4)) for the left OX -module structure.

Proposition (16.7.6). — The right OX -module PX/S
n (F ) is generated by the image of F by the homomor-
phism d X/S,F .
Proof. This is an immediate consequence of (16.7.5.3) and of the particular case F = OX (16.3.8). □
(16.7.7). The canonical homomorphisms of sheaves of rings
m n
ϕnm : PX/S −→ PX/S
for n ⩽ m (16.1.2) define, because of (16.7.2.1), canonical homomorphisms
m n
PX/S (F ) −→ PX/S (F ) (n ⩽ m)
which are homomorphisms of OX -bimodules in light of (16.1.6) and (7.4.1); also we have commuta-
tive diagrams
PX/S
m (F ) / P n (F )
d : X/S
dnX/S,F dnX/S,F
F
We have therefore a projective system of OX -bimodules (PX/S
n (F )), and we define
∞ n
(16.7.7.1) PX/S (F ) = lim PX/S (F ).
←−
Also, this shows that the homomorphisms (16.7.5.1) form a projective system of homomorphisms,
and therefore define a canonical homomorphism
(16.7.7.2) d∞ ∞
X/S,F : F → PX/S (F ).
IV-4 | 39
(16.7.8). Let F , G be two OX -modules; it follows immediately from the definition (16.7.2.1) that we
have a canonical isomorphism of PX/Sn -modules
n n n
(16.7.8.1) PX/S (F ⊗OX G ) ≃ PX/S (F ) ⊗P n PX/S (G )
X/S
(Bourbaki, Alg., chap. II, 3rd ed., §5, n.1, prop. 3).
We conclude in particular (or we see directly from the definition (16.7.2.1)) that if F has an OX -
algebra structure (not necessarily associative), PX/S n (F ) has a canonical O -algebra structure; the
X
latter is associative (resp. commutative, res. unital, resp. a Lie algebra) if F is so. Also the canonical
homomorphisms PX/S m (F ) → P n (F ) for n ⩽ m (16.7.7) are then algebra di-homomorphisms;
X/S
similarly, (16.7.5.1) is then an OX -algebra homomorphisms when PX/S n (F ) is equipped with the
OX -algebra structure from its structure of right OX -module.

With the same notations, we equally have a canonical homomorphisms of PX/S n -modules
n n n
(16.7.8.2) PX/S (H omOX (F , G )) −→ H omP n (PX/S (F ), PX/S (G ))
X/S
(Bourbaki, Alg., chap. II, 3rd ed., §5, n.3), which is bijective when PX/S
n is locally free of finite type (loc.
cit., prop. 7).
(16.7.9). Suppose we are in the situation described in (16.4.1); then from the canonical homomorphism
Pn (u) (16.4.3.3) we deduce immediately a canonical homomorphism of OX ′ -bimodules
(16.7.9.1) u∗ (PX/S
n
(F )) −→ PXn ′ /S′ (u∗ (F )).
We leave it to the reader to extend the properties seen in (16.4) in the case F = OX .
Remark (16.7.10). — The definition of PX/S
n (F ) in the form (16.7.1.2) still makes sense when F is
(n)
a sheaf of sets (the inverse image of a sheaf of sets by p2 being defined in (0, 3.7.1)); a variant of
this definition allows us to define the “jet schemes” (relatively to S) for any prescheme X.
4
16.8. Differential operators.
Definition (16.8.1). — Let f = (ψ, θ ) : X → S be a morphism of preschemes, F , G two OX -modules,
n an integer ⩾ 0. We say that a morphism D : F → G of sheaves of additive groups is a differential
operator of order ⩽ n (relative to S) if there is a homomorphism of OX -modules u : PX/S
n (F ) → G
(where PX/S (F ) is equipped with the structure of left OX -modules (16.7.4)) such that we have
n
D = u ◦ dnX/S,F .
It is clear, because of the existence of canonical morphisms
m n
PX/S (F ) −→ PX/S (F )
for n ⩽ m (16.7.7), that a differential operator of order ⩽ n is a differential operator of order ⩽ m IV-4 | 40
for n ⩽ m. If D : F → G is a differential operator of order ⩽ n, then, for every open set U of X,
D |U : F |U → G |U is a differential operator of order ⩽ n.
We say that a homomorphism D : F → G is a differential operator (relative to S) if, for every
x ∈ X, there is an open neighbourhood U of x and an integer n ⩾ 0 such that D |U : F |U → G |U
is a differential operator of order ⩽ n. The order of a differential operator is the least upper bound
of all integers n so that D is a differential operator of order ⩽ n (and therefore +∞ if there is no
such integer); such order is always finite if X is quasi-compact. The differential operator of order
0 are exactly the homomorphisms of OX -modules F → G ; the operators of order < 0 are zero
by convention. For n ⩾ 0 a differential operator of order n is not in general a homomorphism of
OX -modules but always a homomorphism of ψ∗ (OS )-modules.
When F = OX , a differential operator of order ⩽ 1 of OX to G can be put in the form of v + D,
where v : OX → G is an OX -homomorphism, and D is an S-derivation (16.5.1) of OX to G : this results
from the structure of PB/A (0, 20.4.8).
(16.8.2). To describe in a more precise manner a differential operator of order ⩽ n, D : F → G ,
it suffices, for every open set U of X whose image in S is contained in an affine open set V,
to characterize the homomorphism D = DU : Γ(U, F ) → Γ(U, G ). If we put Γ(V, OS ) = A,
Γ(U, OX ) = B, so that B is an A-algebra, we have Γ(U, PX/Sn (F )) = ( B ⊗ B ) /In+1 , where we
A
abbreviate I = IB/A . Also put M = Γ(U, F ), N = Γ(U, G ); then the definition of D means that for
every pair (U, V ) satisfying the above, the A-homomorphism D : M → N factors through
v
M −→ (( B ⊗ A B)/In+1 ) ⊗ B M −
→N
where the first arrow is the canonical morphism t 7→ 1 ⊗ t, and v is a B-homomorphism, the structure
of B-module coming from the first factor (whereas we recall that in the formation of the tensor
product over B, the structure of B-module on ( B ⊗ A B)/In+1 comes from the second factor B). Note
also that the B-module (( B ⊗ A B)/In+1 ) ⊗ B M is isomorphic to ( B ⊗ A M )/In+1 ( B ⊗ A M), where
( B ⊗ A M) is considered as a ( B ⊗ A B) module and its structure of B-module comes from t 7→ 1 ⊗ t
of B in B ⊗ A B. Let then D ′ be the B-homomorphism of B ⊗ A M to N such that D ′ (b ⊗ t) = bD (t);
then condition of factorization of D is to say that D ′ must be zero on the B-module In+1 ( B ⊗ A M ).
(16.8.3). It is clear that the set of differential operators of order ⩽ n from F to G forms an ad-
ditive group, denoted by DiffnX/S (F , G ); when F = G = OX , we also write DiffnX/S instead of
DiffnX/S (OX , OX ).
4For a more general formalism, see the exposé VII of [eAG64] (due to P. Gabriel).
We have seen (16.8.1), that given two open sets U ⊃ V of X, we have a restriction homomorphism
DiffnX/S (F |U, G |U ) −→ DiffnX/S (F |V, G |V )
n
from which we deduce that U 7→ DiffU/S (F |U, G |U ) is a presheaf of additive groups; in fact, it is IV-4 | 41
actually a sheaf, since for an open set U varying in X, the homomorphisms u 7→ u ◦ dU/S, n
F |U are
isomorphisms of sheaves of additive groups
n n
(16.8.3.1) HomOU (PU/S (F |U ), G |U ) ≃ DiffU/S (F |U, G |U ),
because of the fact that the image of F by dnX/S,F generates PX/S
n (F ) (16.7.6). We denote this sheaf
by D iff X/S (F , G ), and we have:

n
Proposition (16.8.4). — The isomorphisms (16.8.3.1) define an isomorphism of sheaves of additive groups
n
(18.4.1) H omOX (PX/S (F ), G ) ≃ D iff nX/S (F , G ).
When F = G = OX , we also write D iff nX/S instead of D iff nX/S (OX , OX ); it results from (16.6.4)
that D iff nX/S is the dual of PX/S
n ; so we also write ⟨ t, D ⟩ instead of u ( t ) if t is a section of P n
X/S over
an open set and if u is a homomorphism from PX/S n to OX corresponding to D.
(16.8.5). When PX/S
n (F ) has an O -bimodule structure (16.7.4), we deduce canonicallly an O -
X X
bimodule structure on H omOX (PX/S n (F ), G ), and therefore on D iff n
X/S (F , G ) because of (16.8.4.1).
More precisely, to the left OX -module structure on PX/Sn (F ) corresponds, because of (16.8.1), the
left OX -module structure on D iff X/S (F , G ) explained as follows: for every open set U of X, every
n
section a ∈ Γ(U, OX ) and every differential operator D : F |U → G |U, aD is the differential operator
which, for every section t ∈ Γ(U, F ), makes correspond the section
(16.8.5.1) ( aD )(t) = a( D (t))
of Γ(U, G ). Similarly, the right OX -module structure on D iff nX/S (F , G ) is made explicit as fol-
lows: under the same notations as above, Da is the operator which, to every t ∈ Γ(U, F ), makes
correspond the section
(16.8.5.2) ( Da)(t) = D ( at).
Proposition (16.8.6). — If f : X → S is a morphism locally of finite presentation, F a quasi-coherent OX -
module of finite presentation and G a quasi-coherent OX -module, then D iff nX/S (F , G ) is a quasi-coherent
OX -module for any of the structures defined in (16.8.5).
Proof. The proposition follows from the fact that, under these hypothesis, PX/S
n is a quasi-coherent
OX -module of finite presentation (16.7.4) and of (I, 1.3.12) □
(16.8.7). The set of differential operators (of unspecified order (16.8.1)) is denoted by DiffX/S (F , G );
we also see as in (16.8.3) that U 7→ DiffU/S (F |U, G |U ) is a sheaf of additive groups, which we
will denote by D iff X/S (F , G ). It is immediate that D iff X/S (F , G ) is the reunion of the increasing
filtered family of its subsheaves D iff nX/S (F , G ); if X is quasi-compact, DiffX/S (F , G ) is similarly IV-4 | 42
the union of its subgroups DiffnX/S (F , G ) (16.8.1). The OX -bimodule structure on the D iff nX/S (F , G )
induce therefore an OX -bimodule structure on D iff X/S (F , G ), further explained in (16.8.5.1) and
(16.8.5.2).
Note that, for n ⩽ m, we have a commutative diagram
(16.8.7.1) H omOX (PX/S

n (F ), G ) ∼ / D iff n (F , G )
X/S

H omOX (PX/S
m (F ), G ) ∼ / D iff m (F , G )
X/S
where the horizontal arrows are the isomorphisms (16.8.4.1) and the horizontal arrow on the left
comes from the canonical morphism PX/S m (F ) → P n (F ) (16.7.7). For every open set U of X, we
X/S
∞
then endow Γ(U, PX/S (F )) = lim Γ(U, PX/S n (F )) of the projective limit topology of the discrete
←−
n (F )), which defines on Γ (U, P ∞ (F )) a topological Γ (U, O )-bimodule
topologies on Γ(U, PX/S X/S X
∞ (F ) shows itself as a sheaf valued in the category of topological commutative
structure, so that PX/S
groups (0, 3.2.6). So [God58, II, 1.11], the limit of the inductive system of sheaves of commutative
groups (H omOX (PX/S n (F ), G )) is precisely the sheaf of continuous germs of homomorphisms from
∞
PX/S (F ) to G (the latter equipped with the discrete topology): the continuous homomorphisms
Γ(U, PX/S∞ (F )) into the discrete group G indeed correspond bijectively to the inductive systems of
group homomorphisms Γ(U, PX/S n (F )) → Γ (U, G ). We can furthermore express (16.8.4) by saying
there is a canonical isomorphism

∞
H om. contOX (PX/S (F ), G ) ≃ D iff X/S (F , G )
∞ (F )
where the first member denotes the sheaf of germs of continuous homomorphisms from PX/S
to G .
Proposition (16.8.8). — Let F , G be two OX -modules, D : F → G a homomorphism of ψ∗ (OS )-modules,
n an integer ⩾ 0. The following conditions are equivalent:
(a) D is a differential operator of order ⩽ n.
(b) For all sections a of OX over an open set U, the homomorphism Da : F |U → G |U such that, for
every section t of F over an open set V ⊂ U, we have
(16.8.8.1) Da (t) = D ( at) − aD (t)
is a differential operator of order ⩽ n − 1.
(c) For every open set U of X, every family ( ai )1⩽i⩽n+1 of n + 1 sections of OX over U and every section
t of F over U, we have the identity
(16.8.8.2) ∑ (−1)Card( H ) ( ∏ ai ) D (( ∏ ai )t) = 0

H ⊂ In+1 i∈ H i∈
/H
(where In+1 is the interval 1 ⩽ i ⩽ n + 1 of N).

IV-4 | 43
Proof. Lets prove first the equivalence of (a) and (b). By definition, to prove that D is a differential
operator of order ⩽ n, it suffices to prove so for the restriction D |U : F |U → G |U to any affine open
set U of X, and on the other hand the property (c) is valid for every open set U if it is so on every
affine open set. We can therefore restrict ourselves to the case where S = Spec( A) and X = Spec( B)
are affines. Because of (16.8.2) (where we use the same notations), the condition (a) means that the
A-homomorphism D ′ : B ⊗ A M → N such that D ′ (b ⊗ t) = bD (t) is zero on In+1 ( B ⊗ A M), which
is, because of (0, 20.4.4), equivalent to saying that D ′ is annihilates every element of the form
n +1
( ∏ ( ai ⊗ 1 − 1 ⊗ ai )).(1 ⊗ t)
i =1
where ai ∈ B and t ∈ M. Now, this element can be written as ∑ H ⊂ In+1 (∏i∈ H ai ) ⊗ ((∏i∈/ H ai )t), and
the image under D ′ of this element is the first member of (16.8.8.2), which proves the equivalence of
(a) and (c).
Let us prove now the equivalence of (b) and (c). Let us reason by induction on n, the statement
being trivial for n = 0. Writing an+1 instead of a in the condition (b), we see, by the induction
hypothesis, that condition (b) means that for every family ( ai )1⩽i⩽n of n sections of OX over U and
every section t of F over U, that
′
′
∑ (−1)Card( H ) ( ∏ ai ) Dan+1 (( ∏ ai )t) = 0.
H ⊂ In+1 i∈ H ′ / H′
i∈
But if we replace on this relation Dan+1 by the definition (16.8.8.1), we check immediately that we
have, up to sign, the first member of (16.8.8.2); from which we conclude. □
Proposition (16.8.9). — If D : F → G is a differential operator of order ⩽ n, and D ′ : G → H a

differential operator of order ⩽ n′ , then D ′ ◦ D : F → H is a differential operator of order ⩽ n + n′ .
Proof. By hypothesis, we can write D = u ◦ dnX/S,F and D ′ = v ◦ dnX/S,G , where u : PX/S
n ⊗O X F →
′
G and v : PX/S
n ⊗OX G → H are OX -homomorphisms. It all comes down to showing that the
composite homomorphism of sheaves of additive groups

′
dnX/S,F dnX/S,G
F / P n ⊗O F u /G / P n ′ ⊗O G
X/S X X/S X
factors as
′
+n
dnX/S,F
F / P n + n ′ ⊗O F w / P n ′ ⊗O G
X/S X X/S X
where w is an OX -homomorphism. It suffices to prove the

Lemma (16.8.9.1). — There is one and only one OX -homomorphism
n+n n ′n n n ′ ′
(16.8.9.2) δ : PX/S −→ PX/S (PX/S ) = PX/S ⊗OX PX/S
making the following diagram commute IV-4 | 44
′
+n
dnX/S
(16.8.9.3) OX / P n+n′
X/S
dnX/S δ

PX/S
n / P n ′ (P n )
′ X/S X/S
dnX/S,P n
X/S
We will then have, indeed, a commutative diagram deduced from (16.8.9.3) by tensorization with
F
′
+n
dnX/S,F
F / P n + n ′ (F )
X/S
dnX/S,F δ ⊗1

PX/S
n (F ) / P n′ (P n ((F )))
′ X/S X/S
dnX/S,P n (F )
X/S
and on the other hand, we verify immediately the from definition (16.7.5) that the diagram
PX/S
n (F ) u /G
′ ′
dnX/S,P n dnX/S,G
(F )
X/S

n′ (P n ((F )))
PX/S / P n ′ (G )
X/S 1⊗ u X/S
is commutative. We finish the proof by taking w to be the composite OX -homomorphism

δ ⊗1 1⊗ u
n+n
PX/S (F )
′
/ P n′ (P n (F )) / P n ′ (G ).
X/S X/S X/S
□
Proof (16.8.9.3). It remains to prove the lemma (16.8.9.3). Considering (16.7.6), which proves the
uniqueness of δ, we are brought back to the case where S = Spec( A) and X = Spec( B) are affines;
letting I = IB/A , it suffices to define a canonical homomorphism of B-modules
′ ′
ϕ : ( B ⊗ A B)/In+n +1 −→ (( B ⊗ A B)/In +1 ) ⊗ B (( B ⊗ A B)/In+1 )
the B-module structure of the two members coming from the first B factor; recall that on tensor
′
product of the second member, ( B ⊗ A B)/In +1 must be considered as a right B-module by its IV-4 | 45
second B factor , and ( B ⊗ A B)/In+1 as a left B-module by its first B factor (16.7.2). It is the same to
define a homomorphism of B-modules
′
ϕ0 : B ⊗ A B −→ (( B ⊗ A B)/In +1 ) ⊗ B (( B ⊗ A B)/In+1 )
′
and prove it is zero on In+n +1 . Now, we immediately define a homomorphism by the condition
that
ϕ0 (b ⊗ b′ ) = πn′ (b ⊗ 1) ⊗ πn (1 ⊗ b′ ) for b, b′ in B
under the notations of (16.3.7). Also, it is immediate that ϕ0 is a homomorphism of rings. Now, we
can write
ϕ0 (b ⊗ 1 − 1 ⊗ b) = πn′ (b ⊗ 1 − 1 ⊗ b) ⊗ πn (1 ⊗ 1) + πn′ (1 ⊗ b) ⊗ πn (1 ⊗ 1) − πn′ (1 ⊗ 1) ⊗ πn (1 ⊗ b)
and we have
πn′ (1 ⊗ b) ⊗ πn (1 ⊗ 1) = πn′ (1 ⊗ 1)b ⊗ πn (1 ⊗ 1) = πn′ (1 ⊗ 1) ⊗ bπn (1 ⊗ 1) = πn′ (1 ⊗ 1) ⊗ πn (b ⊗ 1)
from which, finally
(16.8.9.4) ϕ0 (b ⊗ 1 − 1 ⊗ b) = πn′ (b ⊗ 1 − 1 ⊗ b) ⊗ πn (1 ⊗ 1) + πn′ (1 ⊗ 1) ⊗ πn (b ⊗ 1 − 1 ⊗ b).
A product of n + n′ + 1 of terms of the form (16.8.9.4) is therefore necessarily zero, because the same
is true for the product of n + 1 terms of the form πn (b ⊗ 1 − 1 ⊗ b) and of n′ + 1 terms of the form
πn′ (b ⊗ 1 − 1 ⊗ b). The conclusion therefore results from (0, 20.4.4). □
Corollary (16.8.10). — The sheaf D iff X/S (OX , OX ) (also denoted D iff X/S ) is canonically endowed with
the structure of sheaf of rings, and the D iff nX/S form an increasing filtration compatible with such structure.
In particular, D iff 0X/S is a sheaf of subrings of D iff X/S , which is canonically identified with OX
(16.8.1). The formulas (16.8.5.1) and (16.8.5.2) show that the structure of OX -bimodule of D iff X/S
comes from the multiplication on the left and on the right by sections of OX considered as a sheaf of
subrings of D iff X/S .
Remarks (16.8.11). — (i) Suppose that F = ⊕λ∈ L Fλ ; then it is clear (16.7.2.1) that PX/S
n (F ) =
⊕λ∈ L PX/S
n (F ); since the functor U 7 → Γ (U, F ) commutes with the formation of arbitrary
λ
direct sums, dnX/S,F is the homomorphism whose restriction to each Fλ is dnX/S,F : Fλ →
λ
PX/S
n (F ); then we conclude immediately that we have
λ
DiffnX/S (F , G ) = ∏ DiffnX/S (Fλ , G ),

λ∈ L
and therefore also (0, 3.2.6)
D iff nX/S (F , G ) = ∏ Diff nX/S (Fλ , G ).
λ∈ L
Moreover, if G = ∏µ∈ M Gµ (0, 3.2.6), we have
n
HomOX (PX/S (F ), G ) = ∏ n
HomOX (PX/S (F ), Gµ ),
µ∈ M
every homomorphism u from PX/S

n (F ) to G corresponds bijectively to the family of its IV-4 | 46
composites uµ : PX/S
n (F ) → G → G . We have therefore
µ
DiffnX/S (F , G ) = ∏ DiffnX/S (F , Gµ ),
µ∈ M
and consequently also

D iff nX/S (F , G ) = ∏ Diff nX/S (F , Gµ ).
µ∈ M
(ii) So far, we have hardly encountered differential operators F → G where F and G are not
locally free of finite rank, in which case the structure is reduced locally, because of (i), to the
case of the sheaf D iff X/S ; the latter will be studied later (16.11) in a particular case.
16.9. Regular and quasi-regular immersions

Definition (16.9.1). — Let X be a ringed space. We say that an ideal I of OX is regular (resp.
quasi-regular) if, for every point x ∈ Supp(OX /I ), there is an open neighbourhood of x in X and a
regular sequence (0, 15.2.2) (resp. quasi-regular (0, 15.2.2)) of elements of Γ(U, OX ) which generates
I |U.
We say that a regular (resp. quasi-regular) sequence of sections of OX over U which generates
I |U is called a regular system (resp. quasi-regular system) of generators of I |U.
Definition (16.9.2). — Let j : Y → X be an immersion of preschemes and let U be an open set such
that j(Y ) ⊂ U and that j is a closed immersion of Y in U. We say that j is regular (resp. quasi-regular)
if the closed subprescheme j(Y ) of U associated to j is defined by a regular (resp. quasi-regular)
ideal of OU (condition independent of the choice of U).
We say that a subprescheme Y of a prescheme X is regularly immersed (resp. quasi-regularly
immersed) if the canonical injection j : Y → X is a regular immersion (resp. quasi-regular immersion).
If Y is a closed subprescheme and I is the ideal of OX that defines Y, it is the same as asking I to
be regular (resp. quasi-regular).
For example, if A is an integral ring, f and element ̸= 0 of A, the closed subprescheme V ( f ) of
Spec( A) (isomorphic to Spec( A/( f ))) is regularly immersed in Spec( A).
Every regular ideal is quasi-regular (0, 15.2.2); every regular immersion is quasi-regular (cf.
(16.9.11) for a converse).
Proposition (16.9.3). — Let X be a ringed space, I an ideal of OX , ( f i )1⩽i⩽m a finite sequence of sections
of OX over X generating I . For the ( f i ) to be a quasi-regular sequence (0, 15.2.2), it is necessary and
sufficient that the following conditions are verified:
(i) The canonical images of f i in I /I 2 form a basis of this OX /I -module.
(ii) The canonical surjective homomorphism (16.1.2.2)
• •
SO X /I
(I /I 2 ) −→ GrI (O X )
is bijective.
Also, if this is true, then every sequence ( f i′ )1⩽i⩽n of n sections of I over X which generates I is
quasi-regular.
IV-4 | 47
Proof. The two conditions stated above only translate the ones in (0, 15.2.2), given the definition of
the canonical homomorphisms (16.2.1.1). The last claim follows from the fact that, if a module M
over a commutative ring A admits a basis of n elements, every system of n generators of M is a basis
(Bourbaki, Alg. comm., chap. II, §3, cor. 5 of th. 1). □
Corollary (16.9.4). — Let X be a locally ringed space, I an ideal of OX . For I to be quasi-regular, it is
necessary and sufficient that it verifies the following conditions:
(i) I is of finite type.
(ii) I /I 2 is a locally free OX /I -module.
(iii) The canonical homomorphism
• •
(16.9.4.1) SO X /I
(I /I 2 ) −→ GrI (O X )
is bijective.
Proof. The necessity of the conditions follows immediately from (16.9.3). To see that they are
sufficient, it is enough to show, because of (16.9.3), that if, in a point x ∈ Supp(OX /I ), there is
a neighbourhood U of x in X and n-sections f i (1 ⩽ i ⩽ n) of I over U such that the image in
I /I 2 form a basis of (I /I 2 )|U over (OX /I )|U, so there will be a neighbourhood V ⊂ U of x
such that the f i |V generate I |V. Now, by hypothesis, we have Ix ̸= Ox , so that Ix is contained in
the maximal ideal of OX ; since Ix is an Ox module of finite type and the classes of ( f i ) x in Ix /Ix2
generate this (Ox /Ix )-module, Nakayama’s lemma shows that the ( f i ) x generate Ix . Since I is of
finite type, we conclude by (0, 5.2.2). □
Corollary (16.9.5). — Let X be a locally ringed space, I a quasi-regular ideal of OX , ( f i )1⩽i⩽n a sequence
of sections of I over X, x a point of Supp(OX /I ). The following conditions are equivalent:
(a) There is a neighbourhood of x in X such that f i |U form a quasi-regular sequence of elements Γ(U, OX )
generating I |U.
(b) The ( f i ) x form a system of generators of Ix whose size is as small as possible.
(b’) The ( f i ) x form a minimal set of generators of Ix .
(c) If f¯i is the canonical image of f i in Γ( X, I /I 2 ), the ( f¯i ) x form a basis for the (Ox /Ix )-module
Ix /Ix2 .
Proof. By hypothesis, Ox is a local ring, Ix an ideal of finite type of Ox contained in the maximal
ideal of Ox ; the equivalence of (b), (b’) and (c) results from Nakayama’s lemma (Bourbaki, Alg. comm.,
chap. II, §3, no 2, prop. 5). It is clear that (a) implies (c) because of (16.9.3); on the other hand, from
(0, 5.2.2) it follows that if condition (c) is verified (and therefore so is (b)), there is a neighbourhood
U of x in X such that (I /I 2 )|U has constant rank equal to n, and that the f i |U generate I |U; it
suffices now to apply the last assertion of (16.9.3) to U. □
Remarks (16.9.6). — (i) Under the general hypothesis of (16.9.5), for the sequence ( f i ) to gen-
erate I , it is not enough that the ( f¯i )y form a basis of the (Oy /Iy )-module (Iy /Iy2 ) for
all y ∈ X. We have an example by taking X = Spec( A), where A is a Dedekind ring, and IV-4 | 48
I = e I, where I is a non-principal ideal of A; then indeed Iy /Iy2 = 0 in every point y
different from x ∈ X corresponding to I, and Ix /Ix2 has rank 1 over the field Ox /Ix ; also,
I is evidently a regular ideal.
(ii) In (16.9.5), one cannot replace “quasi-regular” by “regular”, even when X is a prescheme (cf.
(16.9.12)). Indeed, denote by B the ring of germs of infinitely differentiable functions on the
point 0 on R; it has a maximal ideal m generated by the germ of t of the identity mapping to
0, and the intersection n of the mk for k > 0 is not reduced to 0. Now let A be the quotient
ring B[ T ]/nTB[ T ], and let f 1 , f 2 be the canonical images in A of the elements t and T of B[ T ].
The sequence ( f 1 , f 2 ) is regular in A: indeed, f 1 is not a zero divisor in A, because the relation
tP[ T ] ∈ nTB[ T ], for a polynomial P ∈ B[ T ], implies that the products of t by the coefficients
of P belongs to the ideal n, and it results immediately that the coefficients are the same in
n, so P[ T ] ∈ nTB[ T ]. Since B/tB is isomorphic to R, A/ f 1 A is isomorphic to the ring of
polynomials R[ T ], therefore integral, and the image of f 2 in A/ f 1 A, being equal to T, is not
a zero divisor, so that our claim is true. However, f 2 is a zero divisor in A, since for every
non-zero element x ∈ n, the image of x in A is ̸= 0, but the image of xT is zero. We conclude
that the sequence ( f 2 , f 1 ) is not regular in A; on the other hand, the ideal I = f 1 A + f 2 A is
distinct from A, so the conditions (b), (b’) and (c) of (16.9.5) do not imply the condition (a)
when we replace “quasi-regular” by “regular”.
(16.9.7). If X = Spec( A) is an affine scheme, we’ll say that the ideal I of A is regular (resp. quasi-
regular) if the ideal J = e
I of OX is regular (resp. quasi-regular); we note that this notion is local and
does not imply the existence of a system of generators of I forming in A a regular (resp. quasi-regular)
sequence as the example (16.9.5) shows; however this is true if A is local (16.9.5).
The proposition (16.9.4) can be translated in terms of quasi-regular immersions in the following
manner:
Proposition (16.9.8). — Let j : Y → X be a morphism of preschemes; for j to be a quasi-regular immersion,
it is necessary and sufficient that j satisfies the following conditions:
(i) j is an immersion locally of finite presentation.
(ii) The conormal sheaf Gr1 ( j) = NY/X (16.1.2) is a locally free OY -module.
(iii) The canonical homomorphism
•
SO Y
(Gr1 ( j)) −→ Gr• ( j)
(16.1.2.2) is bijective.
Proof. The problem being local on Y, we can restrict ourselves to the case where j is the canonical
injection of a closed subprescheme Y of X, so the translation of (16.9.4) into (16.9.8) results from
the description of Gr1 ( j) and Gr• ( j) in terms of the ideal I of OX defining the subprescheme Y
(16.1.3, (ii)). □
Corollary (16.9.9). — Let Y be a prescheme, X an Y-prescheme, j : Y → X a Y-section of X, so that the n’th

normal invariant A (n) of j (16.1.2) is an augmented OY -algebra (16.1.7); take A (∞) = lim A (n) . For j to be
←−
a quasi-regular immersion it is necessary and sufficient that j is locally of finite presentation, and that every
y ∈ Y admits an affine open neighbourhood U of ring C such that A (∞) |U is isomorphic as an augmented
OU -algebra to OU [[ T1 , . . . , Tn ]].
Proof. We can reduce to the case where j is a closed immersion by restricting to a small neighbour-
hood of y (see the argument on (16.4.11)), and therefore OY is identified with the quotient algebran
OX /I and the canonical surjective homomorphism OX → OY admits a right inverse (16.1.7). We IV-4 | 49
can therefore suppose X = Spec( B), Y = Spec( A) are affines, B being then an augmented A-algebra,
and the augmentation ideal I being of finite type. Since A (n) is identified with ( B/In+1 )∼ , the
corollary follows from the equivalence of (b) and (c) in (0, 19.5.4) when B/I = A. □
We note that, on the affine case considered, the fact that j is an quasi-regular immersion is further
equivalent, because of (0, 19.5.4), to saying that B is a formally smooth A-algebra for the I-preadic
topology.
We also note that the condition that j is locally of finite presentation is always satisfied when
X → Y is locally of finite type (1.4.3, (v)).
Proposition (16.9.10). — Let X be a locally Noetherian prescheme, Y a subprescheme of X, j : Y → X the
canonical injection, y a point of Y.
(i) For there to exist an open neighbourhood U of y in X such that the restriction Y ∩ U → U of j be a
regular immersion, it is necessary and sufficient that the kernel Iy of the surjective homomorphism
OX,y → OY,y is generated by a regular sequence of elements of OX,y .
(ii) For the immersion j to be regular, it is necessary and sufficient that it is quasi-regular.
Proof. (i) We can reduce to the case where Y is a subprescheme defined by a coherent ideal
I of OX . The condition is clearly necessary. Conversely, if Iy is generated by a regular
sequence (si )y , where the si are sections of I over an open neighbourhood U of y in X, we
can suppose that the si generate I |U (0, 5.5.2) and form a regular sequence (0, 15.2.4), from
which the claim follows.
(ii) The fact that a quasi-regular immersion is regular follows from (i) and the identification of
quasi-regular and regular sequences of OX,y , formed from elements of the maximal ideal
(0, 15.1.11)
□
Corollary (16.9.11). — Let X be a locally Noetherian prescheme; then every quasi-regular ideal of OX is
regular.
Remarks (16.9.12). — (i) We note that a regular immersion is not in general a flat morphism,
and therefore a fortiori neither are quasi-regular morphisms in the sense of (6.8.1).
(ii) Let A be a local Noetherian ring; it follows immediately from (16.9.4) and from (0, 17.1.1)
that for A to be regular, it is necessary and sufficient that its maximal ideal m is quasi-regular
(or regular, which amounts to the same thing given that A is Noetherian). For an affine
Noetherian scheme X to be regular, it is necessary and sufficient that for every closed point
x ∈ X, the canonical injection Spec(k ( x )) → X to be a regular immersion.
Proposition (16.9.13). — Let X be a locally Noetherian prescheme, Y a subprescheme of X, Y ′ a subprescheme
of Y, such that the canonical injection j : Y ′ → Y is regular. Then the sequence of OY ′ -modules
(16.9.13.1) 0 / j∗ (NY/X ) / NY ′ /X / NY ′ /Y /0
is exact; furthermore, for every x ∈ X, there is an open neighbourhood U of x such that the restrictions to U IV-4 | 50
of the homomorphisms of (16.9.13.1) form a split exact sequence.
Let us first prove the following lemma:
Lemma (16.9.13.2). — Let A be a ring, I an ideal of A, A′ = A/I, ( f i )1⩽i⩽r a sequence of elements of A

which is A′ -regular, K = ∑i f i A, L = I + K, K′ = ∑i f i A′ , so that C = A/L is isomorphic to A′ /K′ . For
every integer n > 0, and every integer N ⩾ n, we have the relation
(16.9.13.3) I ∩ Kn = IKn + IK N .
Proof. It is clearly sufficient to prove that every element of the first is contained in the second, and,
by induction on n, we reduce to the case N = n + 1. An element of the first member of (16.9.13.3),
being in Kn , is written as P( f 1 , . . . , f r ), where P ∈ A[ T1 , . . . , Tr ] is homogeneous of degree n. If f i′ is
the canonical image of f i in A′ , the hypothesis P( f 1 , . . . , f n ) ∈ I means that P( f 1′ , . . . , f n′ ) = 0. But
P( f 1′ , . . . , f n′ ) ∈ K′n , so the canonical image of P( f 1′ , . . . , f n′ ) in K′n /K′n+1 is zero. Now the hypothesis
that ( f i ) is A′ -regular implies that the canonical homomorphism SC n (K′ /K′2 ) → K′n /K′n+1 is bijective
(0, 15.1.9); we conclude that the coefficients of P are in L = I + K. It follows immediately that
P( f 1 , . . . , f n ) ∈ IKn + Kn+1 , and since P( f 1 , . . . , f n ) ∈ I, we finally have P( f 1 , . . . , f n ) ∈ IKn + IKn+1
which proves the lemma. □
By taking the quotient of both side of (16.9.13.3) by IKn , we see that the relations (16.9.13.3) for
N ⩾ n entail
(I ∩ Kn )/IKn ⊂ K N .( A/(IKn )).
\
(16.9.13.4)
N⩾n
We deduce the
Corollary (16.9.13.5). — Suppose that the hypothesis of (16.9.13.2) are verified and also that the ring A is
Noetherian and K is contained in the radical of A. Then for every integer n > 0,
(16.9.13.6) I ∩ Kn = IKn .
Proof. Indeed, the second member of (16.9.13.4) is then zero, given that A/IKn is an A-module of
finite type (Bourbaki, Alg. comm., chap. III, §3, no 3, prop. 6). □
Taking in particular n = 2 in (16.9.13.6), and remarking that we have L2 = I2 + IK + K2 =
IL + K2 ; since IL ⊂ L2 , we deduce that
I ∩ L2 = IL + (I ∩ K2 ) = IL + IK2 = IL,
in other words,
(16.9.13.7) I ∩ L2 = IL,
which we can also express in saying that the canonical homomorphism
I/IL −→ (I + J2 )/J2
is bijective.
Proof (16.9.13). Having demonstrated the lemmas, let us prove the first claim of (16.9.13): It is clearly
enough to prove that the sequence of stalks of the sheaves appearing in (16.9.13.1), in a point x ∈ Y ′ , IV-4 | 51
is exact. Now, if we take A = OX,x , we can write OY,x = A′ = A/I, where I is an ideal contained
in the maximal ideal of A, then OY ′ ,x = A′ /K′ , where K′ is generated by an A′ -regular sequence of
elements of A′ , which themselves are images of elements from a A′ -regular sequence of elements of
A belonging to the maximal ideal of A. If K is the ideal generated by the latter and L = I + K, we
have OY ′ ,x = A/L, and since we are in the situation of (16.9.13.5), the canonical homomorphism
I/IL → (I + J2 )/J2 is bijective. But this shows that the sequence
0 / I/IL / L/L2 / (I/L)/(I/L)2 /0
is exact (see the demonstration of (16.2.7)), and the modules making up this sequence are precisely
the stalks in x of the sheaves of (16.9.13.1). The second claim follows from the fact that NX/Y is a
locally free OY ′ -module (16.9.8) and Bourbaki, Alg., chap. II, 3rd ed., §1, no 11, prop. 21. □
16.10. Differentially smooth morphisms.

Definition (16.10.1). — We say that a morphism of preschemes f : X → S is differentially smooth
(or that X is differentially smooth over S) if it satisfies the following conditions:
(i) Ω1X/S is a locally projective OX -module, which is to say that every point x ∈ X admits an
affine open neighbourhood U such that Γ(U, Ω1X/S ) is a projective Γ(U, OX )-module (not
necessarily of finite type).
(ii) The canonical morphism (16.3.1.1)
•
SO X
(Ω1X/S ) −→ Gr• (PX/S )
is bijective.
In particular, if Ω1X/S is locally free of finite rank, the PX/S
n are locally free of finite rank OX -
modules (being extensions of such modules).
We say that f is differentially smooth at a point x ∈ X if there is an open neighbourhood U of x in X
such that f |U is differentially smooth.
We will see later (17.12.4) that a smooth morphism is differentially smooth, which justifies the
terminology; but the converse is not true; indeed, a monomorphism f : X → S is differentially smooth,
since Ω1X/S = 0 because of (I, 5.3.8), and consequently the surjective homomorphism (16.3.1.1) is
clearly bijective; however a monomorphism is not even necessarily flat, neither a fortiori smooth.
Lets limit ourselves to proving the following proposition:
Proposition (16.10.2). — Let A be a ring, B a formally smooth A-algebra for the discrete topology (0, 19.3.1).
Then Spec( B) is differentially smooth over Spec( A).
Proof. Indeed, B ⊗ A B is then (with the discrete topology) a formally smooth B-algebra (by on or
the other canonical homomorphisms b 7→ b ⊗ 1, b 7→ 1 ⊗ b of B to B ⊗ A B ) (0, 19.3.5, (iii)); therefore IV-4 | 52
B ⊗ A B is a formally smooth A-algebra with the discrete topology (0, 19.3.5, (ii)). Taking I = IB/A ,
it follows that B ⊗ A B is a formally smooth A-algebra for the I-preadic topology (0, 19.3.8); since by
hypothesis B = ( B ⊗ A B)/I is a formally smooth A-algebra for the discrete topology, the proposition
follows from the equivalence of (a) and (b) in (0, 19.5.4). □
Proposition (16.10.3). — For a morphism f : X → S to be differentially smooth, it is necessary and
∞ ) is
sufficient that for every x ∈ X, there is an affine open neighbourhood of x, of ring A, such that Γ(U, PX/S
an augmented topological A-algebra isomorphic to the completion B, b where B = S A (V ), V being a projective
A-module and B being endowed with the B -preadic topology (where B+ is the augmentation ideal). If Ω1X/S
+
b with the ring of formal series A[[ T1 , . . . , Tn ]].

is locally free of finite rank, we can replace B
Proof. The notion of a differentially smooth morphism is clearly local on X, so we reduce to the case
where S = Spec( B), X = Spec(C ). Consider C ⊗ B C as a C-algebra (by the first factor); take I = IC/B
and endow C with the I-preadic topology; we can apply to the topological C algebra C ⊗ B C and to
the ideal I of C ⊗ B C the equivalence of (b) and (c) of (0, 19.5.4), given that (C ⊗ B C )/I = C is clearly
a formally smooth C-algebra for the discrete topologies. The topology of Γ(U, PX/S ∞ ) is clearly the
projective limit topology of this ring (16.1.11). □

We note that the integer n of the proposition (16.10.3) is the rank of Ω1X/S in the point x. We
shall see (17.13.5) that when f is differentially smooth and locally of finite type, n is equal to the
dimension of the fiber f −1 ( f ( x )).
Proposition (16.10.4). — Let f : X → S, g : S′ → S be two morphisms, and take X ′ = X ×S S′ ,
f ′ = f (S′ ) : X ′ → S′ .
(i) If f is differentially smooth, the same is true for f ′ .
(ii) Conversely, if g is faithfully flat and quasi-compact, and if f ′ is differentially smooth and Ω1X ′ /S′ is a
finite type OX ′ -module, f is differentially smooth and Ω1X/S is a finite type OX -module.
Proof. Indeed, if f is differentially smooth, the Grn (PX/S
n ) are flat O -modules; therefore by (16.4.6),
X
the homomorphism Grn (PX/S n )⊗
OX OX ′ → Grn (PX ′ /S′ ) are bijective for every n, because of the
n
commutative of the diagram (16.2.1.3), it follows from the definition (16.10.1) that f ′ is differentially
smooth. On the other hand, if g is faithfully flat and quasi-compact, it follows also from (16.4.6) that
Grn (PX/S ′
OX OX ′ → Grn (PX ′ /S′ ) is bijective for every n. Suppose also that f is differentially
n )⊗ n
′
smooth and Ω X ′ /S′ is of finite rank. Since the canonical projection X → X is a faithfully flat and
1
quasi-compact morphism, it results first from (2.5.2) that Ω1X/S is an OX -module locally free of finite
rank, then from (2.2.7) that the canonical homomorphism (16.3.1.1) is bijective, and therefore f is
differentially smooth. □
Proposition (16.10.5). — For a morphism locally of finite type f : X → S to be differentially smooth, it is
necessary and sufficient that the diagonal immersion ∆ f : X → X ×S X to be quasi-regular.
IV-4 | 53
Proof. Being a local problem, we can reduce to the case where S and X are affines, and therefore the
diagonal subprescheme of X ×S X is closed. The hypothesis that f is locally of finite type implies that
∆ f is locally of finite presentation (I, 4.3.1), therefore the diagonal prescheme of X ×S X is defined
by an ideal I of finite type, and Ω1X/S = I /I 2 is an OX -module of finite type. The proposition is
now immediate from the comparison of the conditions of (16.10.1) and (16.9.4). □
Remark (16.10.6). — Let f : X → S be a morphism such that the OX -module Ω1X/S is locally free
of finite rank. It results from (0, 20.4.7) that every x ∈ X has an open neighbourhood such that
there is a finite family (zλ )λ∈ L of sections of OX over U for which (dzλ )λ∈ L forms a basis of the
Γ(U, OX )-module Γ(U, Ω1X/S ).
16.11. Differential operators on a differentially smooth S-prescheme

(16.11.1). Let f : X → S be a morphism, U an open set of X, (zλ )λ∈ L a family of sections of OX
over U such that the dzλ form system of generators of Ω1X/S |U = ΩU/S
1 . Let m be an integer or the
symbol ∞, and take, for every λ
(16.11.1.1) ζ λ = δzλ = dm zλ − zλ ∈ Γ(U, PX/S
m
).
We will also use the usual notations from analysis; for every p = ( pλ ) ∈ N( L) (so that pλ = 0
except for a finite number of indices), we put
(16.11.1.2) |p| = ∑ pλ , p= ∏( pλ !)
λ λ

p
(16.11.1.3) = p!/(q!(p − q)!), for p, q in N( L) , q ⩽ p
q
and we adopt the convention that (pq) = 0 when q ̸⩽ p,
(16.11.1.4) zp = ∏ ( zλ ) pλ , ζp = ∏ ( ζ λ ) pλ .
λ λ
We therefore have, with this notation

p
(16.11.1.5) m m
d (z ) = (d (z)) = (ζ + z)p =
p p
∑ q zp−q ζ q
q⩽p

p p−q m q
(16.11.1.6) ζ p = (dm z − z)p = ∑ (−1)|p−q| q
z d ( z ).
q⩽p
Since the dzλ generate Ω1X/S , and are the images of δzλ , and as the canonical morphism (16.3.1.1)
is surjective, we conclude that for finite m, the δzλ generate the OU -algebra PU/S n (Bourbaki, Alg.
comm., chap. III, §2, n 8, cor. 2 du th. 1). Therefore the ζ (for |p| ⩽ m) generate the OU -module
o p
n . A differential operator D ∈ Diffm
PU/S U/S is consequently entirely determined by the values of
p
⟨ζ , D ⟩for |p| ⩽ m, or, which amounts to the same by (16.11.1.5) and (16.11.1.6), by the values of the IV-4 | 54
⟨dm (z), D ⟩ = D (zp ) for |p| ⩽ m; more precisely, it follows from (16.11.5) that we have

p
(16.11.1.7) D ( z ) = ⟨ d ( z ), D ⟩ = ∑
p m p
⟨ζ q , D ⟩zp−q .
q⩽p
q
Theorem (16.11.2). — Let f : X → S be a morphism, U an open set of X, (zλ )λ∈ L a family of sections
of OX over U such that the family (dzλ )λ∈ L generates Ω1X/S |U = ΩU/S 1 . The following conditions are
equivalent:
(a) f |U is differentially smooth and (dzλ ) is a basis of the OU -module ΩU/S 1 .
(b) There is a family ( Dp )p∈N( L) of differential operators of OU to itself verifying the conditions

q q−p
(16.11.2.1) Dp ( z q ) = z , (p, q in N( L) ).
p
Also, when these conditions are verified, the family ( Dp ) is uniquely determined by the conditions
(16.11.2.1) and satisfies the relations
(p + q)!
(16.11.2.2) Dq ◦ Dp = Dp ◦ Dq = Dp + q (p, q in N( L) ).
p!q!
Finally, if L is finite, for every integer m, the Dp such that |p| ⩽ m form a basis of the OU -module
D iff U/S
m , in other words, every differential operator of order ⩽ m on U can be written uniquely as
D= ∑ a p Dp
|p|⩽m
where the ap are sections of OX over U.
Proof. Note first that because of (16.11.1.6) and (16.11.1.5), we verify immediately that the conditions
(16.11.2.1) are equivalent to
(16.11.2.3) ⟨ζ p , Dq ⟩ = δpq (Kronecker’s symbol).
The existence of the family ( Dp ) verifying these conditions implies first (by taking |p| = 1) that
the dzλ are linearly independent, and therefore form a basis of the OU -module ΩU/S 1 . Then, for
p
every integer m ⩾ 1, we deduce similarly from (16.11.2.3) that the ζ such that |p| ⩽ m are linearly
independent; it follows that the canonical homomorphism (16.3.1.1) is injective, and therefore
bijective, which proves that (b) implies (a). The converse follows immediately from the definition
(16.10.1), the fact that ζ p form a basis of PU/S
n for |p| ⩽ m implies the existence and uniqueness of a
family of homomorphisms uq,m : PU/S → OU (|q| ⩽ m) such that ⟨ζ p , uq,m ⟩ = δp, q for |p| ⩽ m,
n
|q| ⩽ m. For a given value of q, the differential operators corresponding to uq,m for m ⩾ |q| are
identified with the same operator Dq . This proves that (a) implies (b), and also that the family ( Dq )
is uniquely determined, and that, if L is finite, for |p| ⩽ m, the Dp form a basis of the dual D iff U/S
m
of PU/S . Finally, the relations (16.11.2.2) follows immediately from the expression of the values of
n
the three operators considered on the zr , and of the fact that the ζ r for |r| ⩽ m generate PU/S
n . □
IV-4 | 55
Remarks (16.11.3). — (i) The fact that, because of (16.11.2.2), the Dp are pairwise permutable
naturally do not imply that the OU -algebra D iff U/S is commutative, the Dp do not commute
with the sections of OU unless n = 0.
(ii) The indices p such that |p| = 1 are the ϵ λ = (ϵλµ )µ∈ L where ϵλµ = 0 if µ ̸= λ and ϵλλ = 1;
when L is finite, the operators Dϵ λ are exactly the S-derivations Di introduced in (16.5.7).
We note that in general (contrary to what happens in classical analysis), it is not true that a
differential operator of any order can be written as a linear combination of powers of Di (cf.
(16.12)).
(iii) For every integer r ⩾ 1, we can define the notion of differentially smooth up to order r by
replacing in (16.10.1) the condition (ii) by the condition that the homomorphisms
m n
SO X
(Ω1X/S ) −→ Grm (PX/S )
are bijective for every m ⩽ r. The argument of (16.11.2) proves also that if, in the condition (a),
we replace “differentially smooth” by “differentially smooth up to order r”, this condition is
equivalent to (b) by restricting ourselves to p ∈ N( L) , q ∈ N( L) such that |p| ⩽ r, |q| ⩽ r.
16.12. Case of characteristic zero: Jacobian criterion for differentially smooth morphisms.
(16.12.1). We say that a prescheme X is of characteristic p (p equal to zero or a prime number) if,
for every affine open set U of X, the ring Γ(U, OX ) is of characteristic p (0, 21.1.1). It follows from
(0, 21.1.3) that for X to be of characteristic 0, it is necessary and sufficient that for every closed
point x of X, the residue field k ( x ) is of characteristic 0, or even that X can be given a structure of
Q-prescheme (necessarily unique).
Theorem (16.12.2). — Let X be a scheme of characteristic 0, f : X → S a morphism. If Ω1X/S is a locally
free OX -module (not necessarily of finite type), f is differentially smooth.
Proof. The problem being local on X, we can suppose there is a family (zλ ) of sections of OX over X
such that the (dzλ ) is a basis for the OX -module Ω1X/S . Applying the criterion (16.11.2), it is enough
for the operators
Dp = (p!)−1 ∏ Dλλ
p
λ
(where the Dλ are the coordinate forms corresponding to the basis (dzλ )) to verify the relations
(16.11.2.1), which is a consequence of the fact that the Dλ are derivations. □
(16.12.3). The theorem above is not true if we discard the hypothesis that X is of characteristic 0. For
example, if S = Spec(k), where k is a field of characteristic p > 0, X = Spec(K ) where K = k(α)
where α ∈ / k, α p ∈ k, we verify immediately that Ω1X/S has rank 1, and that the morphism X → S IV-4 | 56
has rank 1, and that the morphism X → S is differentially smooth up to order p − 1 (16.11.3, (iii)),
but not of order p. However, the proof of (16.12.2) proves that if Ω1X/S is locally free, and if n!1OX is
inversible in Γ( X, OX ), then X is differentially smooth over S up to order n.
§17. S MOOTH MORPHISMS , UNRAMIFIED ( OR NET ) MORPHISMS , AND ÉTALE

MORPHISMS .
In this paragraph, we revisit the concepts studied in (0III , 9), expressed in the geometric language
of schemes from a global point of view, for preschemes locally of finite presentation over a given
base.
Most of the results (except 17.7, 17.8, 17.9, 17.13, and 17.16) are reduced to various properties
already encountered in (0III , 9).
For more specific results on étale morphisms, the reader should consult §18.
17.1. Formally smooth morphisms, formally unramified morphisms, formally étale morphisms.
Definition (17.1.1). — Let f : X → Y be a morphism of preschemes. We say that f is formally smooth
(resp. formally unramified, resp. formally étale) if, for all affine schemes Y ′ , all closed subschemes Y0′ of
Y ′ defined by a nilpotent ideal J of OY ′ , and every morphism Y ′ → Y, the map
(17.1.1.1) HomY (Y ′ , X ) −→ HomY (Y0′ , X )
induced by the canonical map Y0′ → Y ′ , is surjective (resp. injective, resp. bijective).
One also says that X is formally smooth (resp. formally unramified, resp. formally étale) over Y.
It is clear that for f to be formally étale, it is necessary and sufficient for f to be formally smooth
and formally unramified.
Remark (17.1.2). —
(i) Suppose that Y = Spec( A) and X = Spec( B) are affine, so that f comes from a homomor-
phism of rings ϕ : A → B. According to (0, 19.3.1) and (0, 19.10.1), saying that f is formally
smooth (resp. formally unramified, resp. formally étale) means that, via ϕ, B is a formally
smooth (resp. formally unramified, resp. formally étale) A-algebra, for the discrete topologies on
A and B.
(ii) To verify that f is formally smooth (resp. formally unramified, resp. formally étale), we
can, in Definition (17.1.1), restrict to the case where J 2 = 0. To see this, if f satisfies the
corresponding condition of Definition (17.1.1) in the particular case J 2 = 0, and if we have
J n = 0, then we consider the closed subscheme Yj′ of Y ′ defined by the sheaf of ideals
J j+1 for 0 ⩽ j ⩽ n − 1, so that Yj′ is a closed subscheme of Yj′+1 defined by a square-zero

sheaf of ideals; the hypotheses imply that each of the maps
HomY (Yj′+1 , X ) −→ HomY (Yj′ , X ) (0 ⩽ j ⩽ n − 1)
is surjective (resp. injective, resp. bijective); by composition, we conclude that the same IV | 57
holds for (17.1.1.1).
(iii) Note that the properties of the morphism f defined in (17.1.1) are properties of the repre-
sentable functor (0III , 8.1.8)
Y ′ 7−→ HomY (Y ′ , X )
from the category of Y-preschemes to the category of sets; they keep a meaning for any
contravariant functor with the same domain and codomain, representable or not.
(iv) Assume that the morphism f is formally unramified (resp. formally étale); consider an
arbitrary Y-prescheme Z and a closed subprescheme Z0 of Z defined by a locally nilpotent
sheaf of ideals J of OZ . Then the map
(17.1.2.1) HomY ( Z, X ) −→ HomY ( Z0 , X )
induced by the canonical injection Z0 → Z, is still injective (resp. bijective). To see this, let
(Uα ) be an affine open covering of Z such that the sheaves of ideals J |Uα are nilpotent,
and for each α, let Uα0 be the inverse image of Uα in Z0 , which is the closed subprescheme of
Uα defined by J |Uα . Let f 0 : Z0 → X by a Y-morphism; by hypothesis, for each α, there
is at most one (resp. one and only one) Y-morphism f α : Uα → X whose restriction to Z0
coincides with f 0 |Uα . We immediately conclude that if f α and f β are defined, then, for each
affine open V ⊂ Uα ∩ Uβ , we have f α |V = f β |V, as the restrictions of these morphisms to
the inverse image V0 of V in Z0 coincide. There is therefore at most one (resp. one and only
one) Y-morphism f : Z → X whose restriction to Z0 coincides with f 0 .
Proposition (17.1.3). —
(i) A monomorphism of preschemes is formally unramified; an open immersion is formally étale.
(ii) The composition of two formally smooth (resp. formally unramified, resp. formally étale) morphisms
is formally smooth (resp. formally unramified, resp. formally étale).
(iii) If f : X → Y is a formally smooth (resp. formally unramified, resp. formally étale) S-morphism, then
so is f (S′ ) : X(S′ ) → Y(S′ ) for any base extension S′ → S.
(iv) If f : X → X ′ and g : Y → Y ′ are two formally smooth (resp. formally unramified, resp. formally
étale) S-morphisms, then so is f ×S g : X ×S Y → X ′ ×S Y ′ .
(v) Let f : X → Y and g : Y → Z be two morphisms; if g ◦ f is formally unramified, then so is f .
(vi) If f : X → Y is a formally unramified morphism, then so is f red : Xred → Yred .
Proof. According to (I, 5.5.12), it suffices to prove (i), (ii), and (iii). The assertions in (i) are both
trivial. To prove (ii), consider two morphisms f : X → Y, g : Y → Z, an affine scheme Z ′ , a closed
subscheme Z0′ of Z defined by a nilpotent ideal and a morphism Z ′ → Z. Suppose that f and g
formally smooth, and consider a Z-morphism u0 : Z0′ → X; the hypothesis on g implies that there IV | 58
exists a Z-morphism v : Z ′ → Y such that f ◦ u0 = v ◦ j (where j : Z0′ → Z is the canonical injection);
the hypothesis on f then implies that there exists a morphism u : Z ′ → X such that f ◦ u = v and
u ◦ j = u0 , therefore ( g ◦ f ) ◦ u is equal to the given morphism Z ′ → Z and u ◦ j = u0 , which proves
that g ◦ f is formally smooth; we argue the same way when we suppose that f and g are formally
unramified.
Finally, to prove (iii), let X ′ = XS′ , Y ′ = YS′ , f ′ = f S′ ; consider an affine scheme Y ′′ , a closed
subscheme Y0′′ defined by a nilpotent sheaf of ideals, and a morphism g : Y ′′ → Y ′ making
Y ′′ a Y ′ -prescheme; we then know by (I, 3.3.8) that HomY ′ (Y ′′ , X ′ ) is canonically identified with
HomY (Y ′′ , X ), and HomY ′ (Y0′′ , X ′ ) with HomY (Y0′′ , X ), and the conclusion follows immediately
from Definition (17.1.1). □
We note that a closed immersion is not necessarily formally smooth.
Proposition (17.1.4). — Let f : X → Y and g : Y → Z be two morphisms, and suppose that g is formally
unramified. Then, if g ◦ f is formally smooth (resp. formally étale), so is f .
Proof. Let Y ′ be an affine scheme, Y0′ a closed subscheme of Y ′ defined by a nilpotent sheaf of ideals,
h : Y ′ → Y a morphism, j : Y0′ → Y ′ the canonical injection, u0 : Y0′ → Y a Y-morphism, such that
f ◦ u0 = h ◦ j. Suppose that g ◦ f is formally smooth; then there exists a morphism u : Y ′ → X
such that u ◦ j = u0 and ( g ◦ f ) ◦ u = g ◦ h. But these two relations imply that f ◦ u and h are
Z-morphisms from Y ′ to Y such that ( f ◦ u) ◦ j = h ◦ j; by virtue of the hypothesis that g is formally
unramified, we get that f ◦ u = h, in other words that u is a Y-morphism; thus f is formally smooth.
Taking into account (17.1.3, (v)), this proves the proposition. □
Corollary (17.1.5). — Suppose that g is formally étale; then, for g ◦ f to be formally smooth (resp. formally
unramified, resp. formally étale), it is necessary and sufficient that f is.
Proof. This follows from (17.1.4) and (17.1.3, (ii) and (iv)). □
Proposition (17.1.6). — Let f : X → Y be a morphism of preschemes.
(i) Let (Uα ) be an open covering of X and, for each α, let iα : Uα → X be the canonical injection. For f
to be formally smooth (resp. formally unramified, resp. formally étale), it is necessary and sufficient
that each f ◦ iα is.
(ii) Let (Vλ ) be an open covering of Y. For f to be formally smooth (resp. formally unramified, resp.
formally étale), it is necessary and sufficient that each of the restrictions f −1 (Vλ ) → Vλ of f is.
Proof. First note that (ii) is a consequence of (i): if jλ : Vλ → Y and iλ : f −1 (Vλ ) → X are the
canonical injections, then the restriction f λ : f −1 (Vλ ) → Vλ of f is such that jλ ◦ f λ = f ◦ iλ ; if f
is formally smooth (resp. formally unramified), then so is f ◦ iλ since iλ is formally étale (17.1.3);
but since jλ is formally étale, this means that f λ is formally smooth (resp. formally unramified), by
virtue of (17.1.5). Conversely, if all the f λ are formally smooth (resp. formally unramified), the same
applies to jλ ◦ f λ (17.1.3), so also to f in virtue of (i).
If we take into account that the iα are formally étale, everything comes down to proving that if IV | 59
the f ◦ iα are formally smooth (resp. formally unramified), then the same applies to f .
Therefore let Y ′ be an affine scheme, Y0′ a closed subscheme of Y ′ defined by a nilpotent ideal
J , which we may assume to satisfy J 2 = 0 (17.1.2, (ii)), and finally let g : Y ′ → Y be a morphism.
Suppose we are given a Y-morphism u0 : Y0′ → X; denote by Wα (resp. Wα0 ) the prescheme induced
by Y ′ (resp. Y0′ ) on the open subset u0−1 (Uα ) (we recall that Y ′ and Y0′ share the same underlying
topological space). Let us first suppose that the f ◦ iα are formally unramified, and show that, if u′ and u′′
are two Y-morphisms from Y ′ to X whose restrictions to Y0′ coincide, then we have u′ = u′′ . Indeed,
taking into account (17.1.2, (iv)), the hypothesis that the f ◦ iα are formally unramified implies that
for all α, we have u′ |Wα = u′′ |Wα , since the restrictions of both Y-morphisms to Wα0 coincide. Hence
the conclusion follows.
Now suppose that the f ◦ iα are formally smooth and prove the existence of a Y-morphism u :
Y ′ → X whose restriction to Y0′ is u0 . Now, since Y ′ is an affine scheme, we can apply (16.5.17), the
hypotheses of which are satisfied, and the conclusion of which precisely proves the existence of
u. □
We can therefore say that the notions introduced in (17.1.1) are local on X and Y, which always
allows, in virtue of (17.1.2, (i)), to be reduced to the study of formally smooth (resp. formally
unramified, resp. formally étale) algebras.
17.2. General properties of differentials

Proposition (17.2.1). — For a morphism f : X → Y to be formally unramified, it is necessary and sufficient
that Ω1f = 0 (what we still write Ω1X/Y = 0 (16.3.1)).
Proof. Taking into account (17.1.6), we reduce to the case where Y = Spec( A) and X = Spec( B) are
affine, and the conclusion then follows from (0, 20.7.4) and the interpretation of Ω1X/Y in this case
(16.3.7). □
Corollary (17.2.2). — Let f : X → Y and g : Y → Z be two morphisms. For f being formally unramified,
it is necessary and sufficient that the canonical morphism (16.4.19)
f ∗ (ΩY/Z
1
) −→ Ω1X/Z
is surjective.
Proof. This is an immediate consequence of (17.2.1) and the exact sequence (16.4.19.1). □
Proposition (17.2.3). — Let f : X → Y be a formally smooth morphism.
(i) The OX -module Ω1X/Y is locally projective (16.10.1). If f is locally of finite type, then Ω1X/Y is locally
free and of finite type.
(ii) For all morphisms g : Y → Z, the sequence (16.4.19) of OX -modules
(17.2.3.1) 0 −→ f ∗ (ΩY/Z
1
) −→ Ω1X/Z −→ Ω1X/Y −→ 0
is exact; moreover, for each x ∈ X, there exists an open neighbourhood U of x such that the restrictions
to U of the homomorphisms in (17.2.3.1) form a split exact sequence.
IV | 60
Proof.
(i) We know (16.3.9) that if f is locally of finite type, then Ω1f is an OX -module of finite type. To
prove that, in all cases, it is locally projective, we can reduce, by virtue of (17.1.6), to the case
where Y = Spec( A) and X = Spec( B) are affine, and the result follows from the hypothesis
on f and from (0, 20.4.9) and (0, 19.2.1).
(ii) Again, we can restrict to the case where X, Y, and Z are affine (17.1.6), and the conclusion in
this case follows from the interpretation of the sheaves of modules in the sequence (17.2.3.1)
and from (0, 20.5.7).
□
Corollary (17.2.4). — If f : X → Y is formally étale, then, for all morphisms g : Y → Z, the canonical
homomorphism of OX -modules
f ∗ (ΩY/Z
1
) −→ Ω1X/Z
is bijective.
Proof. This follows from the exactness of the sequence (17.2.3.1) and from the fact that we then have
Ω1X/Y = 0 (17.2.1). □
Proposition (17.2.5). — Let f : X → Y be a morphism, X ′ a subprescheme of X such that the composite
j f
morphism X ′ −→ X − → Y (where j is the canonical injection) is formally smooth. Then the sequence of
OX -modules (16.4.21)
(17.2.5.1) 0 −→ NX ′ /X −→ Ω1X/Y ⊗OX OX ′ −→ Ω1X ′ /Y −→ 0
is exact; moreover, for each x ∈ X, there exists an open neighbourhood U of x such that the restrictions to U
of the homomorphisms in (17.2.5.1) form a split exact sequence.
Proof. By virtue of (17.1.6), we reduce to the case where Y = Spec( A) and X = Spec( B) are affine,
and X ′ = Spec( B/J), where J is an ideal of B. The conormal sheaf NX ′ /X then corresponds to the
B-module J/J2 (16.1.3), and the conclusion follows from (0, 20.5.14). □
Proposition (17.2.6). — Let X and Y be two preschemes, f : X → Y a morphism locally of finite type. The
following conditions are equivalent:
(a) f is a monomorphism.
(b) f is radicial and formally unramified.
(c) For each y ∈ Y, the fibre f −1 (y) is empty or k(y)-isomorphic to Spec(k(y)) (in other words, it is
reduced to a single point z such that k (y) → Oz /my Oz is an isomorphism).
Proof. The fact that (a) implies (c) follows from (8.11.5.1). It is clear that (c) implies that f is radicial;
let us prove that it also follows from (c) that Ω1X/Y = 0, which will prove that (c) implies (b)
(17.2.1). Note that the OX -module Ω1X/Y is quasi-coherent of finite type (16.3.9). It follows from
(I, 9.1.13.1) that, for (Ω1X/Y ) x = 0, it is necessary and sufficient that if we set Y1 = Spec(k(y)),
X1 = f −1 (y) = X ×Y Y1 , then we have (Ω1X /Y ) x = 0; but as the morphism f 1 : X1 → Y1
1 1
induced by f is formally unramified by virtue of the hypothesis (c) (17.1.3), the conclusion follows
from (17.2.1). Finally, let us prove that (b) implies (a); for this, consider the diagonal morphism
g = ∆ f : X → X ×Y X; since f is radicial, g is surjective (1.8.7.1); on the other hand, Ω1X/Y is
by definition the conormal sheaf Gr1 ( g) of the immersion g (16.3.1), and to say that f is formally
unramified therefore means that Gr1 ( g) = 0 (17.2.1). In addition, g is locally of finite presentation IV | 61
(1.4.3.1); therefore the hypothesis Gr1 ( g) = 0 implies that g is an open immersion (16.1.10); being
surjective, this immersion is an isomorphism, hence f is a monomorphism (I, 5.3.8). □
17.3. Smooth morphisms, unramified morphisms, étale morphisms
Definition (17.3.1). — We say that a morphism f : X → Y is smooth (resp. unramified, or net 5 resp.
étale) if it is locally of finite presentation and formally smooth (resp. formally unramified, resp.
formally étale).
We then also say that X is smooth (resp. unramified, resp. étale) over Y.
We will see later (17.5.2) that this definition of a smooth morphism coincides with the definition
already given in (6.8.1); until then, we will exclusively use definition (17.3.1).
It is clear that saying that f is étale means that it is both smooth and unramified.
Remark (17.3.2). —
(i) Note that definition (17.3.1) can be phrased using only the functor
Y ′ 7−→ HomY (Y ′ , X )
considered in (17.1.2, (iii)) because to say that f is locally of finite presentation is equivalent
to saying that the preceding functor commutes with projective limits of affine schemes (8.14.2).
(ii) Let A be a ring and B an A-algebra. We say that B is a smooth (resp. unramified, resp. étale)
A-algebra if the corresponding morphism Spec( B) → Spec( A) is smooth (resp. unramified,
resp. étale). It is equivalent to say that B is an A-algebra of finite presentation (1.4.6) that
is furthermore formally smooth (resp. formally unramified, resp. formally étale) for the
discrete topologies.
(iii) It follows from (17.1.6) and the definition of a morphism locally of finite presentation (1.4.2)
that the notion of a smooth (resp. unramified, resp. étale) morphism is local on X and on Y.
Proposition (17.3.3). —
(i) An open immersion is étale. For an immersion to be unramified, it is necessary and sufficient to it be
locally of finite presentation.
(ii) The composition of two smooth (resp. unramified, resp. étale) morphisms is smooth (resp. unramified,
resp. étale).
(iii) If f : X → Y is a smooth (resp. unramified, resp. étale) S-morphism, then so is f (S′ ) : X(S′ ) → Y(S′ )
for any base extension S′ → S.
(iv) If f : X → X ′ and g : Y → Y ′ are smooth (resp. unramified, resp. étale) S-morphisms, then so is
f ×S g : X ×S Y → X ′ ×S Y ′ . IV | 62
5The words “net” and “formally net” seem more preferable to the terminology used in “unramified” (resp. formally
unramified”) and will be used almost exclusively in Chapter V. In this chapter, we have kept the old terminology so as not to
conflict with 0, 19.10.
(v) Let f : X → Y and g : Y → Z be two morphisms; if g is locally of finite type and if g ◦ f is

unramified, then f is unramified.
Proof. This follows from (1.4.3) and (17.1.3). □
Proposition (17.3.4). — Let f : X → Y and g : Y → Z be two morphisms, and suppose that g is
unramified. Then, if g ◦ f is smooth (resp. unramified, resp. étale), so is f .
Proof. As g and g ◦ f are locally of finite presentation, so is f (1.4.3, (v)); the conclusion thus follows
from (17.1.4) and (17.1.3, (v)). □
Corollary (17.3.5). — Suppose that g is étale; then, for f to be smooth (resp. unramified, resp. étale) it is
necessary and sufficient that g ◦ f is.
Proof. This follows from (17.3.4) and (17.3.3, (ii)). □
Proposition (17.3.6). — Let g : Y → S and h : X → S be two morphisms locally of finite presentation. For
an S-morphism f : X → Y to be unramified, it is necessary and sufficient that the canonical homomorphism
(16.4.19)
f ∗ (ΩY/S
1
) −→ Ω1X/S
is surjective.
Proof. As f is locally of finite presentation (1.4.3, (v)), the proposition follows from (17.2.2). □
Definition (17.3.7). — Let f : X → Y be a morphism. We say that f is smooth (resp. unramifed, resp.
étale) at a point x ∈ X, if there exists an open neighbourhood U of x in X such that the restriction
f |U is a smooth (resp. unramified, resp. étale) morphism from U to Y.
We then also say that X is smooth (resp. unramified, resp. étale) over Y at the point x.
Taking into account remark (17.3.2, (iii)), it is equivalent to say that f is smooth (resp. unramified,
resp. étale) and to say that f is smooth (resp. unramified, resp. étale) at all points of X.
It is clear that the set of points of X at which the morphism f : X → Y is smooth (resp. unramified,
resp. étale) is open in X.
Proposition (17.3.8). — For all preschemes Y and all locally free OY -modules E of finite type, the vector
bundle prescheme V(E ) (II, 1.7.8) associated to E is a smooth Y-prescheme.
Proof. Indeed (17.3.2, (iii)), we can restrict ourselves to the case where Y = Spec( A) is affine
and V(E ) = Spec( A[ T1 , . . . , Tr ]); as A[ T1 , . . . , Tr ] is a formally smooth A-algebra for the discrete
topologies (0, 19.3.2), and of finite presentation, this proves the proposition (17.3.2, (ii)) □
Corollary (17.3.9). — Under the hypotheses of (17.3.8), the projective prescheme P(E ) (II, 4.1.1) is a smooth
Y-prescheme.
Proof. We can still restrict to the case where Y = Spec( A) is affine and P(E ) = PYr . We then know
(II, 2.3.14) that we have a finite open cover of PrA by the D+ ( Ti ) (0 ⩽ i ⩽ r) respectively equal
to the spectrum of the ring S( f ) , where we wrote S for A[ T1 , . . . , Tr ] and f for Ti ; but it follows
immediately from the definition of S( f ) (II, 2.2.1), that this ring, in this case, is isomorphic to
A[ T0 , . . . , Ti−1 , Ti+1 , . . . , Tr ]; hence the corollary follows by (17.3.8). □
17.4. Characterizations of unramified morphisms. IV | 63
§18. S UPPLEMENT ON ÉTALE MORPHISMS . H ENSELIAN LOCAL RINGS AND

STRICTLY LOCAL RINGS
§19. R EGULAR IMMERSIONS AND NORMAL FLATNESS

§20. M EROMORPHIC FUNCTIONS AND PSEUDO - MORPHISMS
20.0. Introduction Most of the concepts and results of §§20 and 21 directly relate to Chapter I, and
hardly depend on Chapters I and IV, except for the occasional usage of the notion of depth and of a
local regular ring (in (20.6), (21.11), (21.13), and (21.15)), of Zariski’s “Main theorem” in (20.4) and
(21.12), and the properties of transversely regular immersions in (20.6) and (21.15).
In §20, we introduce several variants of the concept of a rational map, already studied in (I, 7) from IV-4 | 226
a point of view still fairly close to the classical point of view, and for this reason quite ill-suited to the
case of not necessarily reduced preschemes. The notions and results of §20 are used in §21 (nos (21.1)
and (21.7)) to develop the general notion of a divisor and its most elementary properties. This
notion is especially convenient when the local rings of the preschemes considered are Noetherian
and integrally closed, and especially when they are also factorial ((21.6) and (21.7)), because of their
identification in the latter case with the notion of a 1-codimensional cycle (a linear combination of
irreducible subpreschemes of codimension 1). In (21.9), we determine the divisors on a Noetherian
prescheme of dimension 1 but not necessarily normal, which is useful for various applications.
The (21.11) and (21.12) give two important theorems, due respectively to Auslander–Buchsbaum
and Van der Waerden, and relate the notion of a factorial ring (the nos (21.9), (21.11), and (21.12)
are independent of each other). In the nos (21.13) and (21.14), also independent of the previous
three, we study a useful variant of the notion of a local factorial ring, that of a local parafactorial
ring, which is introduce in particular [?] in the development of comparison theorem of the Picard
group of a projective prescheme X over a field k and a “hyperplane section”. We will see in (21.14.1)
(Ramanujam–Samuel theorem) that the local parafactorial rings are much more numerous than one
would have expected a priori.
In (20.5), (20.6), and (21.15), we review the previous notions but from a point of view “relative” to
a fixed base prescheme. For the moment these notions are used only relatively rarely; in particular,
the concept of a relative divisor is hardly used except when it comes to positive divisors, and in
this case it is explained advantageously without the help of the notion of a relative meromorphic
functions, using the notion of a transversely regular immersion of codimension 1. It will therefore
be advantageous to omit these sections on a first reading.
20.1. Meromorphic functions

(20.1.1). Let ( X, OX ) be a ringed space, and let S be a subsheaf of sets of OX . For every open U of
X, consider the ring of fractions Γ(U, OX )[Γ(U, S )−1 ] (Bourbaki, Alg. comm., chap. II, §2, no 1). It is
immediate that the map U 7→ Γ(U, OX )[Γ(U, S )−1 ] is a presheaf of rings ((0, 1.5.1) and (0, 1.5.7)). We
denote by OX [S −1 ] the sheaf of rings associated to this presheaf and we say that this is the sheaf of
rings of fractions of OX with denominators in S ; this is a flat OX -module. It is immediate that for every
x ∈ X, we have a canonical isomorphism
(20.1.1.1) (OX [S −1 ]) x ≃ Ox [Sx−1 ],

since the reasoning of (0, 1.4.5) generalizes immediately in the case where we have an inductive
system ( Aα , ϕβα ) of rings, and for each index α a subset Sα of Aα such that ϕβα (Sα ) ⊂ S β for α ⩽ β; IV-4 | 227
we then take for S the inductive limit in A = lim Aα of the inductive limit of the subsets (Sα ).
−→
(20.1.2). Now let F be an OX -module. We then set
(20.1.2.1) F [ S −1 ] = F ⊗ O X O X [ S −1 ] ,
and we say that this is the sheaf of modules of fractions of F with denominators in S ; it is immediate
that it is associated to the presheaf of modules U 7→ Γ(U, F )[Γ(U, S )−1 ], and that for every x ∈ X,
we have a canonical isomorphism

(20.1.2.2) (F [S −1 ]) x ≃ Fx [Sx−1 ].
(20.1.3). We will focus here on the case where S is the subsheaf S (OX ) of OX such that for every
open U, Γ(U, S ) is the set of regular elements of the ring Γ(U, OX ); it is immediate that it is a sheaf
(and not only a presheaf), the regularity of a section of OX over U being verified “fibre by fibre” (i.e.
meaning that the germ of the section in x is regular in OX for all x ∈ U), in other words S (OX ) x is
none other then the set of regular elements of OX,x . The corresponding sheaf of rings
M X = O X [ S −1 ]
is called the sheaf of germs of meromorphic functions on X, and the sections of MX over X are called the
meromorphic functions on X; they form a ring which we denote by M( X ). For every OX -module F ,
F ⊗ O X M X = F [ S −1 ]
is also denoted MX (F ) and called the sheaf of germs of meromorphic sections of F ; its sections over
X form an M ( X )-module denoted by M ( X, F ), whose elements are called meromorphic sections of
F over X. These definitions imply that for every open U of X, we have a canonical isomorphism
MX (F )|U ≃ MU (F |U ), in particular MX |U ≃ MU .
(20.1.3.1). If X is a reduced prescheme, then we note that if an element s ∈ Γ(U, OX ) is such that sξ ̸= 0
for every maximal point ξ of U, then s is regular. Indeed, if st = 0 for a t ∈ Γ(U, OX ), then we have
sξ tξ = 0, so tξ = 0 since OX,ξ is a field, and say that tξ = 0 for every maximal point ξ of X means that
t = 0: we are immediately reduced to the case where U is affine, and an element of a reduced ring
which belongs to all the minimal prime ideals is zero by definition. The converse is true if the set of
irreducible components of X is locally finite. We immediately reduce to the case where X = Spec( A)
is affine; if pi (1 ⩽ i ⩽ n) are the minimal prime ideals of A and if s ∈ pi for an index i, then there
exists t ∈ A such that t ∈ p j for j ̸= i and t ̸∈ pi (Bourbaki, Alg. comm., chap. II, §1, no 1, Prop. 1); so
we have st ∈ pi for all i, and as a result st = 0 since A is reduced; s is therefore nonregular.
(20.1.4). For every open U of X, the homomorphism t 7→ t/1 from Γ(U, OX ) to Γ(U, OX )[Γ(U, S )−1 ]
(which is none other than the total ring of fractions of Γ(U, OX )) is injective; these homomorphisms IV-4 | 228
thus define a canonical injective homomorphism
(20.1.4.1) i : OX −→ MX
which allows us to identify OX with a subsheaf of MX . Given a meromorphic function ϕ ∈ M( X ),
we say that ϕ is defined on an open U of X if ϕ|U is a section of OU over U; the axioms of sheaves
show that there is, for a given section ϕ, a largest open on which ϕ is defined; we call this the domain
of definition of ϕ and denote it by dom(ϕ).
(20.1.5). For every OX -module F , we obtain from (20.1.4.1) a di-homomorphism consisting of i and
the homomorphism of sheaves of additive groups
(20.1.5.1) 1F ⊗ i : F −→ MX (F ) = F ⊗OX MX .
We note that the latter is not injective in general; when it is injective, we say that F is strictly
torsion-free: this means that for every open U of X and every section s ∈ Γ(U, OX ) which is a regular
element in this ring the homothety z 7→ sz of Γ(U, F ) is injective; this condition is evidently satisfied
if F is locally free.
Proposition (20.1.6). — Let X be a locally Noetherian prescheme, F a quasi-coherent OX -module. For F
to be strictly torsion-free, it is necessary and sufficient that Ass(F ) ⊂ Ass(OX ).
Proof. We immediately reduce to the case where X = Spec( A) is affine, F = M, e and we know that
the elements s of A belonging to an ideal of Ass( M ) are exactly those for which the homothety
z 7→ sz is not injective (Bourbaki, Alg. comm., chap. IV, §1, no 1, cor. 2 of prop. 2). □
(20.1.7). If u is a section of MX (F ) over X, then we say that u is defined at a point x ∈ X if there exists
an open neighbourhood V of x in X such that u|V is the image of a section of F over V under the
di-homomorphism (20.1.5.1). We say that u is defined on an open U of X if it is defined at every point
in U; there is still a larger open in which u is defined, called the domain of definition of u and denoted
dom(u). When F is strictly torsion-free, such that F is identified by (20.1.5.1) with a subsheaf of
MX (F ), then saying u is defined on U means that u|V is a section of F over U.
(20.1.8). In accordance with the general notation of (0I , 5.4.7), we denote by MX∗ the sheaf of
multiplicative groups such that Γ(U, MX∗ ) is (for every open U of X) the group of invertible ele-
ments of Γ(U, MX ). This sheaf is none other than the sheaf S (MX ) defined in (20.1.3): indeed, if
s ∈ Γ(U, S (MX )), then for every x ∈ U, there exists an open neighbourhood V ⊂ U of x such that
s|V is a regular element in the total ring of fractions of Γ(V, OX ), and we know that such an element is
necessarily invertible in this ring of fractions. We say that the sections of MX∗ over X are the regular
meromorphic functions (note that we are deviating here from the terminology followed by certain
authors, who call “regular” meromorphic functions those which are sections of OX , identified with a
subsheaf of MX ).
Let L be an invertible OX -module (0I , 5.4.1); then it is clear that MX (L ) = L ⊗OX MX is an IV-4 | 229
invertible MX -module. Let U be an open such that L |U is isomorphic to OU ; as every automorphism
of MU is multiplication by an invertible element of Γ(U, MX ) (0I , 5.4.7), it is equivalent to say that
a section s ∈ Γ(U, MX (L )) has an invertible image in Γ(U, MX ) under an isomorphism or by any
isomorphism on Γ(U, MX ); we say in this case that s is a regular meromorphic section of L over U; a
section s of L over X is called a regular meromorphic section of L if, for every open U such that L |U is
isomorphic to OU , s|U is a regular meromorphic section of L over U. We denote by (MX (L ))∗ the
subsheaf of MX (L ) such that for every open U, Γ(U, (MX (L ))∗ ) is the set of regular meromorphic
sections of L over U. Let s be a meromorphic section of L over X (i.e. a section of MX (L )); it
defines a homomorphism hs : MX → MX (L ) which sends every section t of MX over an open U to
(s|U )t. It follows immediately from the above that for s to be regular, it is necessary and sufficient
for hs to be injective, and in fact hs is then a bijective homomorphism from MX to MX (L ), and
its restriction to MX∗ is a bijection to (MX (L ))∗ . We conclude that the homothety t 7→ ts is an
isomorphism from M ( X ) to M ( X, L ).
(20.1.9). Let s be a regular meromorphic section of an invertible OX -module L over X; then for every
OX -module F , s similarly defines a homomorphism hs ⊗ 1F : MX (F ) → MX (F ⊗OX L ), which
is again bijective.
(20.1.10). Let s be a meromorphic section of an invertible OX -module L over X; for s to be regular,
it is necessary and sufficient for there to exists a meromorphic section s′ of L −1 over X such that
the canonical image of s ⊗ s′ in MX (0I , 5.4.3) is the unit section, and this section s′ is then unique:
indeed, the necessity of the local existence of such a section is evident, and its local uniqueness
implies its global (and unique) existence; moreover, the existence of s′ is trivially sufficient for s to
be regular. We will take s′ = s−1 .
Finally, if L ′ is a second invertible OX -module, s (resp. s′ ) a regular meromorphic section of L
(resp. L ′ ) over X, then s ⊗ s′ is evidently a regular meromorphic section of L ⊗ L ′ over X.
(20.1.11). If f : X ′ → X is a morphism of ringed spaces, then there is in general no natural map
sending a meromorphic function on X to a meromorphic function on X’. For example, if X is the
spectrum of a local integral domain A, X ′ its residue field k, then there is no natural homomorphism
from the field of fractions K of A to k, and we can only send an element of K to an element of k if it is
already in A.
In general, if f = (ψ, θ ), then for every open U of X, denote by S f (U ) the set of regular sections
s ∈ Γ(U, OX ) such that the image of s under
Γ(θ # ) : Γ(U, OX ) −→ Γ( f −1 (U ), OX ′ )
is a regular section. It is immediate that U 7→ S f (U ) is a subsheaf of the sheaf of sets S (OX ), which
we denote by S f . We set M f = OX [S f−1 ]; this is a subsheaf of rings of MX , and we canonically IV-4 | 230
obtain from θ# ψ ∗ (O
: X ) → OX ′ a homomorphism of sheaves of rings : θ ′# ψ ∗ (M
f ) → MX ′
extending θ (Bourbaki, Alg. comm., chap. II, §2, n 1, prop. 2); hence, recalling that f ∗ (M f ) =
# o
ψ∗ (M f ) ⊗ψ∗ (OX ) OX ′ , we get a canonical homomorphism of OX ′ -algebras
(20.1.11.1) f ∗ (M f ) −→ MX ′ .
For every meromorphic function ϕ on X that is a section of M f , Γ(θ ′# )(ϕ) is a meromorphic function
on X ′ , called the inverse image of ϕ under f , and denoted by ϕ ◦ f is there is no cause for confusion.
Similarly, if F is an OX -module, then we set M f (F ) = F ⊗OX M f , and we immediately obtain
from θ ′# a canonical homomorphism (which is also written as u 7→ u ◦ f )
Γ( X, M f (F )) −→ Γ( X ′ , MX ′ ( f ∗ (F ))).
In addition, if u ∈ Γ( X, M f (F )) is defined (20.1.7) at a point x, then u coincides, on a neighbourhood
U of x, with a section of the form ∑i hi ⊗ (ti /si ), where the hi belong to Γ(U, F ), the ti to Γ(U, OX ),
and the si to Γ(U, S f ). As by hypothesis the images of the si in Γ( f −1 (U ), OX ′ ) are regular, we see
that u ◦ f is defined at every point of f −1 (U ); in other words, we have
(20.1.11.2) f −1 (dom(u)) ⊂ dom(u ◦ f ).
We will see later (20.6.5, (i)) examples (with F = OX ) where the two sides of (20.1.11.2) can be
different.
Consider in particular the case where M f = MX ; then, if L is an invertible OX -module, the
image in MX ′ ( f ∗ (L )), under Γ(θ ′# ), of a regular meromorphic section of L over X (20.1.8) is a
regular meromorphic section of f ∗ (L ) over X ′ , as it follows immediately from the definition of
its sections, and from the fact that a homomorphism of rings sends an invertible element to an
invertible element.
Let f ′ : X ′′ → X ′ be a second morphism of ringed spaces, and suppose that M f = MX and
M f ′ = MX ′ ; then, if we set f ′′ = f ◦ f ′ , we also have M f ′′ = MX , and we immediately see that for
every meromorphic section u of F over X, we have u ◦ f ′′ = (u ◦ f ) ◦ f ′ .
Proposition (20.1.12). — If the morphism f : X ′ → X is flat (0I , 6.7.1), then we have M f = MX , and the
homomorphism ϕ 7→ ϕ ◦ f is defined on all of M ( X ). In addition, if f is a (flat) morphism of locally ringed
spaces, then we have dom(ϕ ◦ f ) = f −1 (dom(ϕ)); if in addition f is surjective (thus faithfully flat), then
the homomorphism ϕ 7→ ϕ ◦ f is injective.
Proof. The first assertion follows from the fact that, if B is an A-algebra which is a flat A-module,
then every element of A not a divisor of 0 in A is not a divisor of 0 in B (0I , 6.3.4). To prove the
other assertions, note that, for every x ′ ∈ X ′ , if x = f ( x ′ ), then OX ′ ,x′ is a flat OX,x -module, and as
the homomorpism OX,x → OX ′ ,x′ is local by hypothesis, it is injective ((0I , 6.5.1) and (0I , 6.6.2)); if
we set A = OX,x , B = OX ′ ,x′ , such that A identifies with a subring of B, then ( f ∗ (MX )) x′ is equal to
S−1 A ⊗ A B = S−1 B, where S is the set of regular elements of A, (MX ′ ) x′ is equal to T −1 B, where T is
the set of regular elements of B, and as we have seen that S ⊂ T, the homomorphism S−1 B → T −1 B IV-4 | 231
is injective; in other words, this proves that the homomorphism (20.1.11.1) f ∗ (MX ) → MX ′ is
injective (hence the last assertion of the statement). The quotient f ∗ (MX )/OX ′ identifies with an
OX ′ -submodule of MX ′ /OX ′ , and ( f ∗ (MX )/OX ′ ) x′ identifies with (MX /OX ) x ⊗OX,x OX ′ ,x′ . Then
suppose that x ̸∈ dom(ϕ); the image of ϕx in (MX /OX ) x is therefore ̸= 0; by faithful flatness, we
deduce that that it is the same for the image of (ϕ ◦ f ) x′ , so x ′ ̸∈ dom(ϕ ◦ f ), which finishes the
proof. □
Remark (20.1.13). — Let X be a reduced complex analytic space; then the notion of a meromorphic
function on X defined above coincides with the usual notion. Consider on the other hand a
prescheme Y, locally of finite type over the field C; we then know that we can associate to Y an
analytic space Y an having the same underlying topological space, and the canonical morphism
f : Y an → Y is flat [?]; by virtue of (20.1.12), the canonical homomorphism u 7→ u ◦ f from M (Y ) to
M(Y an ) is therefore always defined and is injective; but it is not surjective in general. For example,
when Y = V0r (ErrIII , 14) is the affine space of dimension r over C, M(Y ) canonically identifies with
the field R(Y ) of rational functions on Y (20.2.13, (i)), while M (Y an ) is the field of usual meromorphic
functions on Cr . Because of this fact, it is often preferable, in algebraic geometry, to abstain from the
terminology introduced in this section, and to use the equivalent terminology of “pseudo-function”
which will be defined below.
20.2. Pseudo-morphisms and pseudo-functions
§21. D IVISORS
R EFERENCES
[CC] H. Cartan and C. Chevalley, Séminaire de l’école normale supérieure, 8th year (1955–56), géométrie algébrique.
[eAG64] M. Demazure et A. Grothendieck, Schémas en groupes, 1963–1964.
[God58] R. Godement, Théorie des faisceaux, Actual. Scient. et Ind., no. 1252, Paris (Hermann), 1958.

Ega 4

Uploaded by

Copyright:

Available Formats

Ega 4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ega 4

Uploaded by

Copyright:

Available Formats

LOCAL STUDY OF SCHEMES AND THEIR MORPHISMS (EGA IV)

build hack [CC]

§1. R ELATIVE FINITENESS CONDITIONS . C ONSTRUCTIBLE SETS IN PRESCHEMES

1.1. Quasi-compact morphisms

Definition (1.1.1). — We say that a morphism of preschemes f : X → Y is quasi-compact if the

§2. B ASE CHANGE AND FLATNESS

§6. F LAT MORPHISMS OF LOCALLY N OETHERIAN PRESCHEMES

16.1. Normal invariants of an immersion

Proposition (16.1.9). — Let X be a prescheme, j : Y → X an immersion locally of finite presentation, y a IV-4 | 8

Corollary (16.1.10). — For a restriction of the immersion j to an open neighbourhood of y in Y to be an open

S(gr1 (u)) gr(u)

where the horizontal arrow are the canonical morphisms (16.1.2.2).

Corollary (16.2.3). — Let g : X → Y, u : Y ′ → Y be two morphisms of preschemes, X ′ = X ×Y Y ′ ,

(0, 6.1.2); in addition, we have the commutative diagram IV-4 | 13

16.3. Fundamental differential invariants of morphisms of preschemes

Definition (16.3.1). — Let f : X → S be a morphism of preschemes, ∆ f : X → X ×S X the

denoted by Ω f or Ω X/S , and is called the OX -module of 1-differentials of f , or of X with respect to S,

It follows from this definition that PX/S

(16.3.2). Denote by p1 , p2 the two canonical projections of the product X ×S X; since ∆ f is an X-

(16.3.3). The morphism s = ( p2 , p1 )S : X ×S X → X ×S X is an involutive automorphism of X ×S X,

If we put s = (ρ, λ), pi = (πi , µi ) (i = 1, 2), ∆ f = (δ, ν), λ# is then an isomorphism of

Definition (16.3.6). — We denote by dnf , or dnX/S (resp. d∞ ∞ n ∞

It follows immediately 2 from this definition that we have

16.4. Functorial properties of differential invariants

(16.4.1). Consider a commutative diagram of morphisms of preschemes

We deduce a commutative diagram

where v is the composite homomorphism (I, 5.3.5) and (I, 5.3.15).

(16.4.2). If we have a commutative diagram IV-4 | 17

we deduce a commutative diagram

where v′ is defined from u′ , w′ , f ′ , f ′′ as v is from u, w, f , f ′ . We verify immediately that if u′′ = u ◦ u′ ,

u∗ (dnX/S ) dnX ′ /S′

is commutative. We deduce a homomorphism of graded OX ′ -algebras

Grn (u) : u∗ (Gr(PX/S

(16.4.7.1) δ(t) = dnX/S (t′ ) − t′ ∈ Γ( f −1 (U ), PX/S

the augmentation ideal of Γ( f −1 (U ), PX/S

(16.4.7.3) δn : f ∗ (SOS (E )/K n +1 n

F ′ −→ I = gr1I (SOY (F )),

Proof. The injection F ′ → I indeed canonically gives a homomorphism of graded OY -algebras

Corollary (16.4.9). — Under the conditions of (16.4.7), we have a canonical isomorphism

(16.4.9.1) gr1 (δ) : f ∗ (E ) ≃ Ω1X/S .

Corollary (16.4.10). — If S = Spec( A), E = OSm , so that

Proposition (16.4.11). — Let f : X → S be a morphism, g : S → X a S-section of X, S(n) the n-th

commutative (where λn is the structure morphism).

Proof. It suffices to consider the unique k-section g of X such that g(Spec(k)) = { x }. □

Corollary (16.4.13). — Let f : X → S be a morphism, s a point of S, Xs = X ×S Spec(k (s)) the fibre of f

Proof. This follows from (16.4.5) and (16.4.12). □

Proposition (16.4.14). — Let ρ : A → B be a morphism of rings, S a multiplicative subset of B. Then the

(16.4.14.1) S−1 PB/A

(16.4.15.1) S−1 PB/A

Proof. It evidently suffices to define canonical isomorphisms

(16.4.15.2) PSn−1 B/A ≃ PSn−1 B/R−1 A

Corollary (16.4.16). — Let f : X → S be a morphism of preschemes, x a point of X, s = f ( x ). Then we

(16.4.16.2) (Ω1X/S ) x ≃ Ω1Ox /Os .

(16.4.18). Let f : X → Y, g : Y → Z be two morphisms of preschemes, and consider the canonical

(16.4.18.2) f X/Y/Z : f ∗ (PY/Z

We have a commutative diagram (I, 3.5)

where j = (1X , 1X ) Z is an immersion, j ◦ ∆ f = ∆ g◦ f , and p is the structure morphism. Since we

When X, Y, Z are affine, we recover the exact sequence (0, 5.7.1).

corresponding respectively to the two canonical projections p1 , p2 of Y × Z Y → Y, recall that by