Algebraic Atiyah-Singer index theorem
Nguyen Le Dang Thi

TL;DR
This paper develops an algebraic weak version of the Atiyah-Singer index theorem and illustrates it through examples involving elliptic differential operators on smooth projective schemes over a field.
Contribution
It introduces an algebraic approach to the Atiyah-Singer index theorem and applies it to specific elliptic operators derived from the Atiyah class.
Findings
Computed examples of elliptic operators on smooth projective schemes
Established an algebraic framework for the index theorem
Connected Atiyah class to index computations
Abstract
The aim of this work is to give an algebraic weak version of the Atiyah-Singer index theorem. We compute then a few small examples with the elliptic differential operator of order coming from the Atiyah class in , where is a smooth projective scheme over a perfect field .
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAlgebraic Geometry and Number Theory · Advanced Topics in Algebra · Advanced Algebra and Geometry
Algebraic Atiyah-Singer index theorem
Nguyen Le Dang Thi
(Date: 06. 02. 2017)
Abstract.
The aim of this work is to give an algebraic weak version of the Atiyah-Singer index theorem. We compute then a few small examples with the elliptic differential operator of order coming from the Atiyah class in , where is a smooth projective scheme over a perfect field .
Key words and phrases:
-theory, motivic cohomology, differential operators
1991 Mathematics Subject Classification:
14F22, 14F42
We follow Grothendieck [EGA4, §16.8] to recall briefly the notion of differential operators. Let be a morphism of schemes. Consider the Cartesian square
[TABLE]
The diagonal is an immersion. One defines the -th normal invariant of as
[TABLE]
It is clear that form a projective system. One defines
[TABLE]
The first projection induces an -algebra structure on and the second projection induces a morphism
[TABLE]
If is an -module, one defines
[TABLE]
Let be two -modules. Let be a morphism of the underlying abelian sheaves. is called a differential operator of order relative , if there exists a morphism of -modules , such that the following diagram commutes:
[TABLE]
The morphism between the underlying abelian sheaves is not -linear, but it is -linear. Let us denote by the set of all differential operator of order between and . One defines
[TABLE]
It follows from the definition [EGA4, 16.8.3.1] that one has an isomorphism of abelian groups
[TABLE]
If is locally of finite type, then is a quasi-coherent -module of finite type (cf. [EGA4, Prop. 16.3.9]). If is locally of finite presentation, then is a quasi-coherent -module of finite presentation (cf. [EGA4, Cor. 16.4.22]). Consequently, if is a quasi-coherent -module of finite type resp. locally of finite presentation and is locally of finite type resp. locally of finite presentation, then is also a quasi-coherent -module of finite type resp. locally of finite presentation. If is locally Noetherian and is proper, then by [EGA3, Thm. 3.2.1] the higher direct image sheaf
[TABLE]
is coherent for any and . Consequently, if is proper over a Noetherian ring , then for all coherent sheaf the Zariski cohomology groups are -modules of finite type . A differential operator induces a homomorphism of abelian groups
[TABLE]
where we write for the category of abelian sheaves on and is the terminal object in . In general, we can not say anything about the kernel or cokernel of this homomorphism. However, induces also a homomorphism of -modules
[TABLE]
This motivates us to give the following definition:
Definition 1**.**
- (1)
Let be a -scheme of finite type, where is a field and be a differential operator between two -modules of order . is called a Fredholm operator, if the -linear morphism between -vector spaces
[TABLE]
has finite-dimensional kernel and cokernel. 2. (2)
Let be a scheme of finite type over a Noetherian ring and be a differential operator between two -modules. is called a Fredholm operator, if the kernel and cokernel of the -module morphism
[TABLE]
are -modules of finite type.
For a proper -scheme , where is a Noetherian ring, any differential operator between coherent -modules is a Fredholm operator.
Definition 2**.**
Let be a proper -scheme and be a differential operator between coherent -modules. We define the index of to be
[TABLE]
Let be any scheme and be the category of quasi-coherent -modules. Following Grothendieck [EGA2, §1.7] we define a vector bundle associated to a quasi-coherent sheaf to be the -scheme
[TABLE]
The structure morphism is an affine morphism. For any morphism , we have
[TABLE]
We call the [math]-section, if it corresponds to the [math]-homomorphism . Now we assume is an -module of finite type. Then is a morphism of finite type [EGA2, Prop. 1.7.11 (ii)]. So if is a Noetherian scheme, then is also a Noetherian scheme. Consider the locally small abelian category of coherent sheaves on . We let
[TABLE]
to be the Grothendieck group of . By [EGA2, Prop. 1.7.15] the morphism is a closed immersion. Let be the complement open immersion. By [SGA6, §IX, Prop. 1.1] one has a localization exact sequence
[TABLE]
If is the structure morphism of over a base scheme , then any differential operator of order between -modules defines a two terms complex of the underlying abelian sheaves
[TABLE]
Or equivalently, a two terms complex of -modules
[TABLE]
We obtain then a two terms complex of -modules
[TABLE]
Remark that if is Noetherian and is of finite type, then is also Noetherian, so , and are coherent -modules, if and are coherent. Moreover, by [EGA4, Prop. 16.4.5] there is a canonical isomorphism
[TABLE]
Definition 3**.**
Let be a morphism of finite type, where is a Noetherian scheme. Let be an -module of finite type and be the associated vector bundle. Let be a differential operator between coherent -modules of order . Then is called an elliptic operator with respect to , if there is an isomorphism
[TABLE]
The isomorphism is called the symbol of .
By the localization exact sequence 1, an elliptic operator of order with respect to an -module of finite type defines a class , which becomes [math] after restricting to the complement of the [math]-section. If a morphism locally of finite type, then the sheaf of relative differentials is an -module of finite type.
Definition 4**.**
Let be a morphism of finite type over a Noetherian scheme . A differential operator between coherent -modules is called elliptic, if it is elliptic with respect to .
Let be a Noetherian scheme. Denote by the locally small exact category of locally free sheaves of finite rank on . We let
[TABLE]
For an arbitrary morphism of Noetherian schemes one has a functorial pullback homomorphism
[TABLE]
For a proper morphism between Noetherian schemes there is a pushforward homomorphism
[TABLE]
which is well-defined as an exact sequence of coherent -modules
[TABLE]
gives rise to an exact sequence of coherent -modules
[TABLE]
The functoriality of the pushforward follows easily from the Leray-Grothendieck spectral sequence
[TABLE]
If is a regular Noetherian scheme, then the canonical homomorphism induced by the obvious exact embedding of categories
[TABLE]
is an isomorphism, since every coherent sheaf on a regular scheme has a finite locally free resolution.
Let now be an arbitrary scheme. We denote by the stable motivic homotopy category together with the formalism of six functors as in [Ay08], [CD12] and [Hoy14, Appendix C]. Let be a motivic ring spectrum parameterized by schemes. We will assume is stable by base change, i.e. for any morphism of schemes there is an isomorphism of spectra in
[TABLE]
Let be a scheme and be a motivic ring spectrum. The unit gives rise to a class
[TABLE]
We have a tower of -schemes given by the obvious embeddings
[TABLE]
Let , which is an object in the pointed unstable motivic homotopy category of Morel-Voevodsky [MV99]. Let be the obvious map in , which gives rise to a map . is called oriented, if there is a class
[TABLE]
such that . We say is oriented, if is oriented for any scheme and for any morphism one has . If is regular, by [MV99, §4, 1.15, 3.7] there is a canonical isomorphism in , which gives to a canonical isomorphism
[TABLE]
For an oriented motivic ring spectrum , one can define the first Chern class as
[TABLE]
Assume is a regular scheme. Let be a smooth -scheme. For a vector bundle associated to a locally free -module of finite rank , there is an isomorphism (cf. [NSO09, Thm. 2.11])
[TABLE]
where is the projective bundle associated to [EGA2, §4]. So is a free module over with the basis
[TABLE]
So for a vector bundle of rank one can define the higher Chern classes
[TABLE]
where one puts and for . Recall that one has a homotopy cofiber sequence [MV99, §3]
[TABLE]
where denotes the Thom space of the vector bundle . The homotopy cofiber sequence induces a long exact sequence
[TABLE]
The projective bundle formula tells us that is a split epimorphism, so is a free module of rank over , which is just . The Thom class is the unique class in
[TABLE]
where is the structure morphism. Let be its section. The Thom isomorphism is given by
[TABLE]
Its inverse is the composition
[TABLE]
which sends
[TABLE]
If is the universal quotient bundle on , i.e. there is an exact sequence
[TABLE]
Then by Whitney sum formula we have
[TABLE]
Now let be the motivic ring spectrum representing the algebraic -theory constructed in [Rio10], [CD12, §13]. If is regular and is a smooth -scheme, then one has a natural isomorphism
[TABLE]
is an oriented motivic ring spectrum. Indeed, by Bott periodicity one has an isomorphism
[TABLE]
The Bott element is the image of under this isomorphism and . The orientation of is given by
[TABLE]
For any line bundle , its first Chern class is
[TABLE]
For an arbitrary morphism of regular schemes one has a commutative diagram [CD12, §13.1]
[TABLE]
Now assume is a locally free sheaf of finite rank on a smooth scheme over a regular base . We have a homotopy cofiber sequence (cf. [MV99, §3])
[TABLE]
where denotes the Thom space of the vector bundle . This homotopy cofiber sequence gives rise to an exact sequence
[TABLE]
The natural isomorphism 2 gives rise to a commutative diagram with exact rows (cf. [CD12, §13.4, (K6a)]), where we abbreviate for :
[TABLE]
The bottom row is the exact sequence 1. For a Noetherian scheme and an -module of finite type the functor
[TABLE]
is exact, since is an affine morphism, which implies that . So we still can define the pushforward
[TABLE]
Hence, one may try to define the topological index by applying the homomorphism
[TABLE]
then composing with . However, is not necessary projective, so we may run into difficulties, when we apply later the Grothendieck-Riemann-Roch theorem. Now we restrict ourselves to the situation where is a smooth proper scheme over a field . is a locally free -module of finite rank. Let be an elliptic operator of order . Let denotes the vector bundles associated to . The symbol defines then an element
[TABLE]
which lies in the image of the homomorphism
[TABLE]
Let us denote by a class, which is mapped to
[TABLE]
such that its image in is trivial. We call this class a symbol class associated to the symbol .
Definition 5**.**
Let be a smooth proper -scheme. Let be an elliptic operator, where . The topological index of is defined as
[TABLE]
where
[TABLE]
is the inverse Thom isomorphism of algebraic -theory and
[TABLE]
If is any scheme, let denote the Beilinson rational motivic cohomology ring spectrum in constructed in [Rio10], [CD12, Defn. 14.1.2]. Let
[TABLE]
be the total Chern character, which is an isomorphism of rational spectra, if , where is a perfect field [Rio10, Defn. 6.2.3.9, Rem. 6.2.3.10]. One has the following result due to Riou:
Theorem 6**.**
[Rio10, Thm. 6.3.1]**(Grothendieck-Riemann-Roch) Let be a perfect field. Let be a smooth projective morphism of smooth -schemes. There is a commutative diagram in
[TABLE]
As explained in [CD12, §13.7] induces the usual pushforward on -theory. Now we can compute
Proposition 7**.**
Let be a smooth projective scheme over a perfect field . Let be an elliptic operator between coherent -modules. Then one has a formula
[TABLE]
where means that we take the pushforward on motivic cohomology
[TABLE]
Proof.
This is a trivial consequence of Thm. 6, where . ∎
Now let be an arbitrary differential operator of order on a projective scheme over a Noetherian ring and . As gives rise to an -module homomorphism
[TABLE]
we may take the twisting
[TABLE]
which in turn gives us a differential operator of order
[TABLE]
We call a twisting of .
Proposition 8**.**
Let be a smooth projective scheme over a perfect field and be a differential operator of order between coherent -modules. Then there exists a number , such that for all one has
[TABLE]
Proof.
This is quite trivial. By the result of Serre (see e.g [EGA3, Thm. 2.2.1, Prop. 2.2.2]), there is a number , such that for all and all
[TABLE]
and a number , such that for all and all
[TABLE]
We take to be the maximal of and . We have trivially that for all
[TABLE]
The proposition follows now easily from the Hirzebruch-Riemann-Roch theorem
[TABLE]
∎
Now we are ready to state the following algebraic weak form of the Atiyah-Singer index theorem
Theorem 9**.**
Let be a smooth projective scheme over a perfect field and be an elliptic differential operator of order , where are coherent -modules. There exists a number , such that for all there is an equality
[TABLE]
Proof.
It remains to prove that
[TABLE]
which means that we have to show there is a commutative diagram
[TABLE]
By construction, we have a commutative diagram
[TABLE]
So it remains to see that we have a commutative diagram
[TABLE]
But this is quite obvious, since the closed immersion is the composition of the [math]-section and (cf. [EGA2, Prop. 8.3.2]). ∎
Finally, we will give in the last part of this paper a few small examples. For this purpose, we always restrict now ourselves to the case of a smooth projective -scheme of dimension , where is a perfect field. Let us begin with the following lemma:
Lemma 10**.**
(de Jong and Starr [dJS06, Lem. 2.1]) Let be a smooth complete intersection of type . Then
[TABLE]
Proof.
The proof is so elementary, so we reproduce the proof. Let denote the closed embedding. Consider the short exact sequence
[TABLE]
Therefore,
[TABLE]
Consider the Euler sequence
[TABLE]
One has
[TABLE]
The lemma follows easily. ∎
Consider the Atiyah class in
[TABLE]
The morphism defines a differential operator of order
[TABLE]
Lemma 11**.**
* is an elliptic operator.*
Proof.
Let be the vector bundle associated to . We pullback the Atiyah class via to obtain an exact sequence
[TABLE]
For each point we have by [EGA4, Cor. 16.4.12]
[TABLE]
The morphism becomes obviously an isomorphism at each point . ∎
From now on we always consider the elliptic differential operator . We write and for a curve
[TABLE]
Proposition 12**.**
Let be a smooth curve of genus , which is a smooth complete intersection of two quadrics. For all integer number one has
[TABLE]
For one has
[TABLE]
Proof.
is smooth complete intersection of two quadrics. By the lemma 10 we have . This implies
[TABLE]
for any number , where we write for the hyperplane section. As , the canonical divisor is trivial. So we have and hence
[TABLE]
This implies easily that for . ∎
Proposition 13**.**
Let be a flat projective morphism over a perfect field , such that the generic fiber is a smooth curve of genus one in . Then there exists a number , such that
[TABLE]
Proof.
We have for :
[TABLE]
We take two embeddings and , such that they are compatible with
[TABLE]
Let be a big integer number. By [EGA3, Cor. 7.9.13, Seconde partie] is a locally free -module of rank . By flat base change for Zariski cohomology of coherent sheaves it is enough to compute , where is the base change of to an algebraic closure . must be an elliptic curve. For an elliptic curve , we know that and is very ample, i.e. . The last isomorphism means simply that we can embed
[TABLE]
By Riemann-Roch theorem for curves we also have . Now we can apply the theorem of Grothendieck for the decomposition of vector bundles on and we have
[TABLE]
Let be another big integer number. We have
[TABLE]
So
[TABLE]
If this dimension is already odd, then there is nothing to prove. So we assume
[TABLE]
Let be another big integer number. After twisting by we have
[TABLE]
Now we take simply , and . We obtain
[TABLE]
By projection formula we have for any number an isomorphism
[TABLE]
This implies
[TABLE]
where the first equality follows easily from the adjunction
[TABLE]
Therefore, we can conclude that there exists a number such that , which finishes the proof. ∎
Proposition 14**.**
Let be a smooth -surface, which is a smooth complete intersection, over a perfect field .
- (1)
If , then for a big number one has
[TABLE] 2. (2)
If , then for a big number one has
[TABLE]
Proof.
We write and for the canonical divisor. For an algebraic surface one has
[TABLE]
For a smooth projective -surface one has and . is the dual bundle of . So we have by lemma 10
[TABLE]
For any number one has
[TABLE]
If then and if then . The result follows now easily, since and . ∎
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[Ay 08] J. Ayoub, Les six opération de Grothendieck et le formalisme des cycles évanescents dans le monde motivique I and II. Astérisque 𝟑𝟏𝟒 314 \bf{314} , 𝟑𝟏𝟓 315 \bf{315} , (2008).
- 2[CD 12] C. D. Cisinski, F. Déglise, Triangulated category of mixed motives, ar Xiv:0912.2110 v 3 [math.AG], Preprint (2012).
- 3[d JS 06] A. J. de Jong, J. Starr, Low degree complete intersections are rationally simply connected, Preprint, 74 pages.
- 4[EGA 2] A. Grothendieck, J. Dieudonné, Éléments de géométrie algébrique. II. Étude globale élémentaire de quelques classes de morphismes, Publ. Math. IHÉS 𝟖 8 \mathbf{8} (1961).
- 5[EGA 3] A. Grothendieck, J. Dieudonné, Éléments de géométrie algébrique. III. Étude cohomologique des faisceaux cohérents, Publ. Math. IHÉS 𝟏𝟏 11 \mathbf{11} (1961) and 𝟏𝟕 17 \mathbf{17} (1963).
- 6[EGA 4] A. Grothendieck, J. Dieudonné, Éléments de géométrie algébrique. IV. Étude locale des schémas et des morphismes de schémas IV, Publ. Math. IHÉS 𝟐𝟎 , 𝟐𝟒 , 𝟐𝟖 , 𝟑𝟐 20 24 28 32 \mathbf{20,24,28,32} (1964-1967).
- 7[Hoy 14] M. Hoyois, A quadratic refinement of the Grothendieck-Lefschetz-Verdier trace formula, Algebr. Geom. Topol. 14, no. 𝟔 6 \mathbf{6} , (2014).
- 8[MV 99] F. Morel, V. Voevodsky, 𝔸 1 superscript 𝔸 1 \mathbb{A}^{1} -homotopy theory of schemes, IHÉS Publ. Math., ( 𝟗𝟎 ) 90 \mathbf{(90)} , 45-143, (1999).
