This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Transfers on commutative group schemes

Junnosuke Koizumi Graduate School of Mathematical Sciences, University of Tokyo, 3-8-1 Komaba, Meguro-ku, Tokyo 153-8914, Japan jkoizumi@ms.u-tokyo.ac.jp
Abstract.

We prove that any commutative group scheme separated over a noetherian normal scheme admits a canonical structure of a presheaf with transfers, which is characterized by a simple condition on radicial transfers.

1991 Mathematics Subject Classification:
Primary 14F42; Secondary 14L15

Introduction

Voevodsky’s theory of presheaves with transfers (see for example [MVW06, Part 1]) plays a central role in his theory of motives over a field. Roughly speaking, they are presheaves on the category of smooth schemes equipped with covariant “transfer maps” for finite surjective morphisms, functorial in an appropriate sense. One important example is commutative group schemes, e.g. 𝔾a\mathbb{G}_{a}, 𝔾m\mathbb{G}_{m}, the Witt ring scheme and abelian varieties. In particular, smooth commutative group schemes over a field equipped with transfers are known to give examples of reciprocity sheaves (see [KSY16], [KSY]), for which a nice motivic theory can be applied.

On the other hand, transfer structures on group schemes over a general base scheme has not been studied much yet. In this paper we prove the following general existence result.

Theorem 0.1 (see Theorem 3.8).

Let SS be a noetherian normal scheme and GG be a separated commutative group scheme over SS. Then there exists a canonical structure of a presheaf with transfers over SS on GG.

One way to prove that a presheaf has a transfer structure is to show that it is a qfh sheaf. In this way, Ancona-Huber-Lehalleur [AHL14] proved that for any smooth commutative group scheme GG over a noetherian excellent scheme SS, the presheaf G=GG_{\mathbb{Q}}=G\otimes_{\mathbb{Z}}\mathbb{Q} admits a unique transfer structure. However, being a qfh sheaf is so strong a condition that we cannot expect in general that GG itself should be so. Also, transfers constructed in this way are comparatively inexplicit and difficult to compute.

Another way is to use the symmetric product and construct transfers geometrically. In this way, Spieß-Szamuely [SS03] and Barbieri-Viale-Kahn [BVK16, Lemma 1.4.4] proved that any commutative group scheme locally of finite type over a field admits a canonical transfer structure. Our proof of Theorem 0.1 is based on this idea. In section 2, we imitate the construction of locally free transfers due to Spieß-Szamuely in a more general setting. In section 3, we define the canonical transfer on group schemes and prove the functoriality to establish Theorem 0.1. In section 4, we characterize the canonical transfer by a simple condition on radicial transfers.

Acknowledgements

I am grateful to Shuji Saito for his support in my studies. I also thank Hiroyasu Miyazaki for many interesting discussions.

1. Review of relative cycles

In this section we recall the theory of relative cycles in the style of Cisinski-Déglise [CD19]; nothing in this section is our original. Let XX be a noetherian scheme. A cycle on XX is a formal \mathbb{Z}-linear combination of integral closed subschemes of XX. A component of a cycle α\alpha is an integral closed subscheme whose coefficient in α\alpha is non-zero. For a closed subscheme WW of XX, we define the associated cycle [W][W] by

[W]=i=1nlength(𝒪X,Wi)[Wi]\textstyle[W]=\sum_{i=1}^{n}\operatorname{length}(\mathcal{O}_{X,W_{i}})[W_{i}]

where W1,,WnW_{1},\dots,W_{n} are the irreducible components of WW. If f:XYf\colon X\to Y is a morphism between noetherian schemes and α\alpha is a cycle on XX, then the cycle fαf_{*}\alpha on YY is defined by linearly extending

[V]{[k(ξ):k(f(ξ))][{f(ξ)}¯]([k(ξ):k(f(ξ))]<)0(otherwise)[V]\mapsto\begin{cases}[k(\xi):k(f(\xi))][\overline{\{f(\xi)\}}]&([k(\xi):k(f(\xi))]<\infty)\\ 0&(\text{otherwise})\end{cases}

where ξ\xi is the generic point of VV.

Until the end of this section, we fix a noetherian base scheme SS and consider only noetherian schemes over SS.

Definition 1.1.

Let XX be a noetherian SS-scheme and α\alpha a cycle on XX.

  • We say that α\alpha is finite over SS if every component of α\alpha is finite over SS.

  • We say that α\alpha is flat over SS if every component of α\alpha is flat over SS.

  • We say that α\alpha is pseudo-dominant over SS if every component of α\alpha is dominant over some irreducible component of SS.

Consider the following diagram of noetherian schemes.

(1.5)

Let α\alpha (resp. β\beta) be a cycle on XX (resp. YY). If α\alpha is \mathbb{Z}-universal over SS in the sense of [CD19], then a cycle αSβ\alpha\otimes_{S}\beta on X×SYX\times_{S}Y called the pullback of α\alpha by β\beta is defined. The operation S{-}\otimes_{S}{-} is bilinear.

Lemma 1.2.

Let XX be a noetherian SS-scheme and α\alpha be a cycle on XX. Then α\alpha is finite and flat over SS \implies \mathbb{Z}-universal over SS \implies pseudo-dominant over SS.

Proof.

This follows directly from the definition of \mathbb{Z}-universal cycles; see [CD19, Definition 8.1.47]. ∎

Lemma 1.3.

Consider the diagram (1.5) of noetherian schemes. Let α\alpha be a cycle on XX finite and flat over SS and write α=i=1nmi[Vi]\alpha=\sum_{i=1}^{n}m_{i}[V_{i}]. Then we have

αS[Y]=i=1nmi[Vi×SY].\textstyle\alpha\otimes_{S}[Y]=\sum_{i=1}^{n}m_{i}[V_{i}\times_{S}Y].
Proof.

See [CD19, 8.1.35 (P3)]. ∎

Lemma 1.4.

Suppose that SS is reduced. Let XX be a noetherian SS-scheme and α\alpha be a cycle on XX \mathbb{Z}-universal over S. Then there exists a dominant blow-up p:SSp\colon S^{\prime}\to S such that αS[S]\alpha\otimes_{S}[S^{\prime}] is flat over SS^{\prime}.

Proof.

See [CD19, Lemma 8.1.18 and 8.1.35 (P5)]. ∎

Definition 1.5.

For a noetherian SS-scheme XX, we define c0(X/S)c_{0}(X/S) to be the abelian group of cycles on XX finite and \mathbb{Z}-universal over SS. An element of c0(X/S)c_{0}(X/S) is called a relative 0-cycle on XX over SS.

Let X,Y,ZX,Y,Z be noetherian SS-schemes. We set cS(X,Y):=c0(X×SY/X)c_{S}(X,Y):=c_{0}(X\times_{S}Y/X) and call its elements finite correspondences from XX to YY over SS. For example, the graph Γf\Gamma_{f} of a morphism f:XYf\colon X\to Y over SS gives an element [Γf]cS(X,Y)[\Gamma_{f}]\in c_{S}(X,Y), for which we simply write ff. For αcS(X,Y)\alpha\in c_{S}(X,Y) and βcS(Y,Z)\beta\in c_{S}(Y,Z), we define a cycle βα\beta\circ\alpha on X×SZX\times_{S}Z by the formula

βα=pr13(βYα)\beta\circ\alpha={{\mathrm{pr}}_{13}}_{*}(\beta\otimes_{Y}\alpha)

where pr13:X×SY×SZX×SZ{{\mathrm{pr}}_{13}}_{*}\colon X\times_{S}Y\times_{S}Z\to X\times_{S}Z is the canonical projection. By the definition of pullback (cf. [CD19, Theorem 8.1.39]) and [CD19, Corollary 8.2.6], we have βYαc0(X×SY×SZ/X)\beta\otimes_{Y}\alpha\in c_{0}(X\times_{S}Y\times_{S}Z/X). If ZZ is separated and of finite type over SS, then it follows from [CD19, Section 9.1.1] that βαcS(X,Z)\beta\circ\alpha\in c_{S}(X,Z). This gives a bilinear pairing

:cS(X,Y)×cS(Y,Z)cS(X,Z).\circ\colon c_{S}(X,Y)\times c_{S}(Y,Z)\to c_{S}(X,Z).
Lemma 1.6.

Let SS be a noetherian scheme and X,Y,Z,WX,Y,Z,W be noetherian SS-schemes.

  1. (1)

    Let αcS(X,Y)\alpha\in c_{S}(X,Y), βcS(Y,Z)\beta\in c_{S}(Y,Z) and γcS(Z,W)\gamma\in c_{S}(Z,W). Suppose that Z,WZ,W are separated and of finite type over SS. Then we have

    (γβ)α=γ(βα).(\gamma\circ\beta)\circ\alpha=\gamma\circ(\beta\circ\alpha).
  2. (2)

    Let f:XYf\colon X\to Y be an SS-morphism and βcS(Y,Z)\beta\in c_{S}(Y,Z). Suppose that ZZ is separated and of finite type over SS. Then we have

    βf=βY[X].\beta\circ f=\beta\otimes_{Y}[X].
Proof.

See [CD19, Proposition 9.1.7]. In loc.cit., it is assumed that X,Y,Z,WX,Y,Z,W are all separated and of finite type over SS, but the proof works verbatim. ∎

Let SmS{\mathrm{Sm}}_{S} denote the category of smooth separated SS-schemes of finite type. Using the pairing above as composition, we can define an additive category CorS{\mathrm{Cor}}_{S} whose objects are the same as SmS{\mathrm{Sm}}_{S} and the morphisms are finite correspondences. The category SmS{\mathrm{Sm}}_{S} can be embedded into CorS{\mathrm{Cor}}_{S} by the graph construction. A presheaf CorSopAb{\mathrm{Cor}}_{S}^{\mathrm{op}}\to{\mathrm{Ab}} is called a presheaf with transfers over SS.

2. Locally free transfers

In this section we construct transfers for finite locally free morphisms.

Let AA be a ring and BB a finite AA-algebra. We write BdB^{\odot d} for the AA-algebra (Bd)Sd(B^{\otimes d})^{S_{d}}, i.e. the subalgebra of BdB^{\otimes d} fixed under the canonical action of the symmetric group SdS_{d}.

Suppose that BB is free of rank dd as an AA-module. Let e1,,ede_{1},\dots,e_{d} be a basis of BB over AA and let I[m,n]dI_{[m,n]}^{d} denote the set of SdS_{d}-orbits in {m,m+1,,n}d\{m,m+1,\dots,n\}^{d}. For an orbit ΓI[1,d]d\Gamma\in I_{[1,d]}^{d}, we set

eΓ:=(i1,,id)Γei1eid.\textstyle e_{\Gamma}:=\sum_{(i_{1},\dots,i_{d})\in\Gamma}e_{i_{1}}\otimes\dots\otimes e_{i_{d}}.

Then {eΓ}ΓI[1,d]d\{e_{\Gamma}\}_{\Gamma\in I_{[1,d]}^{d}} is a basis of BdB^{\odot d} over AA. We define an action of BdB^{\odot d} on dB\wedge^{d}B by

eΓ(b1bd)=(i1,,id)Γei1b1eidbd.\textstyle e_{\Gamma}\cdot(b_{1}\wedge\dots\wedge b_{d})=\sum_{(i_{1},\dots,i_{d})\in\Gamma}e_{i_{1}}b_{1}\wedge\dots\wedge e_{i_{d}}b_{d}.

This is well-defined since we are taking a sum over all possible permutations of (i1,,id)(i_{1},\dots,i_{d}). In this way we obtain a morphism of AA-algebras

u:BdEndA(dB)A.u\colon B^{\odot d}\to\operatorname{End}_{A}(\wedge^{d}B)\simeq A.

One can easily verify that this does not depend on the choice of the basis.

Lemma 2.1.

Let KK be a field and BB be a finite local KK-algebra of dimension dd with residue field KK. Let φ\varphi denote the composition BdBKB^{\otimes d}\xrightarrow{\cdot}B\twoheadrightarrow K. Then the restriction of φ\varphi to BdB^{\odot d} equals u:BdKu\colon B^{\odot d}\to K.

Proof.

Let JJ denote the maximal ideal of BB. First note that JJ is a nilpotent ideal since BB is artinian local. Take a basis e1,,ede_{1},\dots,e_{d} of BB over KK so that ed=1e_{d}=1 and JiJ^{i} is spanned by e1,,edie_{1},\dots,e_{d_{i}} as a KK-vector space, where di=dimKJid_{i}=\dim_{K}J^{i}. For ΓI[1,d]d\Gamma\in I^{d}_{[1,d]} we have

φ(eΓ)={1(Γ={(d,,d)})0(otherwise)\varphi(e_{\Gamma})=\begin{cases}1&(\Gamma=\{(d,\dots,d)\})\\ 0&(\text{otherwise})\end{cases}

since JJ is an ideal. For an element bb of BB, we define its order ord(b)\operatorname{ord}(b) to be the maximum value of ii satisfying bJib\in J^{i} (we set ord(0)=\operatorname{ord}(0)=\infty). Let b1,,bdBb_{1},\dots,b_{d}\in B satisfy ord(bi)ord(bi)\operatorname{ord}(b_{i})\geq\operatorname{ord}(b_{i}) for all ii and ord(bj)>ord(bj)\operatorname{ord}(b_{j})>\operatorname{ord}(b_{j}) for some jj. Then the number of bib_{i}’s contained in JkJ^{k} exceeds the dimension of JkJ^{k} for some kk, so we get b1bd=0dBb_{1}\wedge\dots\wedge b_{d}=0\in\wedge^{d}B. Using this fact, the action of BdB^{\odot d} on dB\wedge^{d}B can be computed as

eΓ(e1ed)={e1ed(Γ={(d,,d)})0(otherwise).e_{\Gamma}\cdot(e_{1}\wedge\dots\wedge e_{d})=\begin{cases}e_{1}\wedge\dots\wedge e_{d}&(\Gamma=\{(d,\dots,d)\})\\ 0&(\text{otherwise}).\end{cases}

This completes the proof. ∎

Lemma 2.2.

Let AA be a ring and Bi(i=1,2)B_{i}\;(i=1,2) be finite free AA-algebras of rank did_{i}. The morphism of AA-algebras

pr1d1pr2d2:(B1×B2)(d1+d2)B1d1AB2d2{\mathrm{pr}}_{1}^{\otimes d_{1}}\otimes{\mathrm{pr}}_{2}^{\otimes d_{2}}\colon(B_{1}\times B_{2})^{\otimes(d_{1}+d_{2})}\to B_{1}^{\otimes d_{1}}\otimes_{A}B_{2}^{\otimes d_{2}}

restricts to a morphism

p:(B1×B2)(d1+d2)B1d1AB2d2p\colon(B_{1}\times B_{2})^{\odot(d_{1}+d_{2})}\to B_{1}^{\odot d_{1}}\otimes_{A}B_{2}^{\odot d_{2}}

and fits into the following commutative diagram.

(B1×B2)(d1+d2)\textstyle{(B_{1}\times B_{2})^{\odot(d_{1}+d_{2})}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}u\scriptstyle{u}p\scriptstyle{p}A\textstyle{A\ignorespaces\ignorespaces\ignorespaces\ignorespaces}B1d1AB2d2\textstyle{B_{1}^{\odot d_{1}}\otimes_{A}B_{2}^{\odot d_{2}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}uu\scriptstyle{u\otimes u}A\textstyle{A}
Proof.

Let e1,,ed1e_{1},\dots,e_{d_{1}} (resp. ed1+1,,ed1+d2e_{d_{1}+1},\dots,e_{d_{1}+d_{2}}) be a basis of B1B_{1} (resp. B2B_{2}) over AA. For Γ1I[1,d1]d1\Gamma_{1}\in I_{[1,d_{1}]}^{d_{1}} and Γ2I[d1+1,d1+d2]d2\Gamma_{2}\in I_{[d_{1}+1,d_{1}+d_{2}]}^{d_{2}} we define Γ1Γ2I[1,d1+d2]d1+d2\Gamma_{1}\ast\Gamma_{2}\in I_{[1,d_{1}+d_{2}]}^{d_{1}+d_{2}} by concatenation. Then for an orbit ΓI[1,d1+d2]d1+d2\Gamma\in I_{[1,d_{1}+d_{2}]}^{d_{1}+d_{2}} we have

(pr1d1pr2d2)(eΓ)={eΓ1eΓ2(Γ=Γ1Γ2 for some Γ1I[1,d1]d1,Γ2I[d1+1,d1+d2]d2)0(otherwise).\displaystyle({\mathrm{pr}}_{1}^{\otimes d_{1}}\otimes{\mathrm{pr}}_{2}^{\otimes d_{2}})(e_{\Gamma})=\begin{cases}e_{\Gamma_{1}}\otimes e_{\Gamma_{2}}&(\Gamma=\Gamma_{1}\ast\Gamma_{2}\text{ for some }\Gamma_{1}\in I_{[1,d_{1}]}^{d_{1}},\;\Gamma_{2}\in I_{[d_{1}+1,d_{1}+d_{2}]}^{d_{2}})\\ 0&(\text{otherwise}).\end{cases}

This proves the first assertion. In the first case we have

eΓ(e1ed1+d2)=\displaystyle e_{\Gamma}\cdot(e_{1}\wedge\dots\wedge e_{d_{1}+d_{2}}){}={} (i1,,id1+d2)Γei1e1eid1+d2ed1+d2\displaystyle\textstyle\sum_{(i_{1},\dots,i_{d_{1}+d_{2}})\in\Gamma}e_{i_{1}}e_{1}\wedge\dots\wedge e_{i_{d_{1}+d_{2}}}e_{d_{1}+d_{2}}
=\displaystyle{}={} (i1,,id1)Γ1(id1+1,,id1+d2)Γ2ei1e1eid1+d2ed1+d2\displaystyle\textstyle\sum_{(i_{1},\dots,i_{d_{1}})\in\Gamma_{1}}\sum_{(i_{d_{1}+1},\dots,i_{d_{1}+d_{2}})\in\Gamma_{2}}e_{i_{1}}e_{1}\wedge\dots\wedge e_{i_{d_{1}+d_{2}}}e_{d_{1}+d_{2}}
=\displaystyle{}={} (eΓ1(e1ed1))(eΓ2(ed1+1ed1+d2))\displaystyle(e_{\Gamma_{1}}\cdot(e_{1}\wedge\dots\wedge e_{d_{1}}))\wedge(e_{\Gamma_{2}}\cdot(e_{d_{1}+1}\wedge\dots\wedge e_{d_{1}+d_{2}}))

In the second case the action of eΓe_{\Gamma} on d1+d2(B1×B2)\wedge^{d_{1}+d_{2}}(B_{1}\times B_{2}) is trivial. These results imply the commutativity of the diagram. ∎

Let SS be a scheme. For a finite morphism f:YXf\colon Y\to X between SS-schemes, the symmetric product SymXd(Y)\operatorname{Sym}^{d}_{X}(Y) is defined to be the quotient of Y×X×XYY\times_{X}\dots\times_{X}Y (dd times) by the canonical action of the symmetric group SdS_{d}, i.e.

SymXd(Y)=Spec¯X(f𝒪Y)d\operatorname{Sym}^{d}_{X}(Y)=\underline{\operatorname{Spec}}_{X}(f_{*}\mathcal{O}_{Y})^{\odot d}

(see [Gro71, Exposé V, 1] for basic facts about quotients of schemes by finite groups).

Suppose that ff is a finite locally free morphism of constant rank dd. The morphism of 𝒪X\mathcal{O}_{X}-algebras u:(f𝒪Y)d𝒪Xu\colon(f_{*}\mathcal{O}_{Y})^{\odot d}\to\mathcal{O}_{X} (see the preamble to Lemma 2.1) gives rise to a morphism f:XSymXd(Y)f^{\sharp}\colon X\to\operatorname{Sym}^{d}_{X}(Y) (note that this morphism is considered also in [SV96, Section 6]). Now let GG be a commutative group scheme over SS. For any gG(Y)g\in G(Y), the morphism

Y×X×XY(g,,g)G×S×SG+GY\times_{X}\dots\times_{X}Y\xrightarrow{(g,\dots,g)}G\times_{S}\dots\times_{S}G\xrightarrow{+}G

is SdS_{d}-invariant, so it descends to a morphism σ(g):SymXd(Y)G\sigma(g)\colon\operatorname{Sym}_{X}^{d}(Y)\to G. We define fgG(X)f_{*}g\in G(X) to be the composition

XfSymXd(Y)σ(g)G.X\xrightarrow{f^{\sharp}}\operatorname{Sym}_{X}^{d}(Y)\xrightarrow{\sigma(g)}G.
Lemma 2.3.

Let SS be a scheme, KK be a field over SS and L/KL/K be a finite extension. Let f:SpecLSpecKf\colon\operatorname{Spec}L\to\operatorname{Spec}K be the corresponding morphism. Then we have

f=\displaystyle f_{*}{}={} NmL/K:𝔾m,S(L)𝔾m,S(K),\displaystyle\operatorname{Nm}_{L/K}\colon\mathbb{G}_{m,S}(L)\to\mathbb{G}_{m,S}(K),
f=\displaystyle f_{*}{}={} TrL/K:𝔾a,S(L)𝔾a,S(K).\displaystyle\operatorname{Tr}_{L/K}\colon\mathbb{G}_{a,S}(L)\to\mathbb{G}_{a,S}(K).
Proof.

Let e1,,ede_{1},\dots,e_{d} be a basis of LL over KK. Let a𝔾m,S(L)=L×a\in\mathbb{G}_{m,S}(L)=L^{\times}. By definition, the action of adLda^{\otimes d}\in L^{\odot d} on dL\wedge^{d}L is equal to the scalar multiplication by faf_{*}a. On the other hand, we have

ad(e1ed)=ae1aed=detK(L𝑎L)(e1ed),a^{\otimes d}\cdot(e_{1}\wedge\dots\wedge e_{d})=ae_{1}\wedge\dots\wedge ae_{d}=\operatorname{det}_{K}(L\xrightarrow{a}L)\cdot(e_{1}\wedge\dots\wedge e_{d}),

so we get fa=detK(L𝑎L)=NmL/K(a)f_{*}a=\operatorname{det}_{K}(L\xrightarrow{a}L)=\operatorname{Nm}_{L/K}(a). The proof of the second equality is similar. ∎

Lemma 2.4.

Let SS be a scheme, f:YXf\colon Y\to X be a finite locally free morphism of constant degree dd between SS-schemes and h:XXh\colon X^{\prime}\to X be a morphism of SS-schemes. Consider the following Cartesian diagram.

Y\textstyle{Y^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}h\scriptstyle{h^{\prime}}f\scriptstyle{f^{\prime}}Y\textstyle{Y\ignorespaces\ignorespaces\ignorespaces\ignorespaces}f\scriptstyle{f}X\textstyle{X^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}h\scriptstyle{h}X\textstyle{X}

For any commutative group scheme GG over SS, we have

hf=fh:G(Y)G(X).h^{*}f_{*}=f^{\prime}_{*}{h^{\prime}}^{*}\colon G(Y)\to G(X^{\prime}).
Proof.

Since YY and SymXd(Y)\operatorname{Sym}^{d}_{X}(Y) are finite locally free over XX, all the constructions we used to define ff_{*} are compatible with base-change, so the claim is obvious. ∎

Lemma 2.5.

Let SS be a scheme, KK a field over SS and f:XSpecKf\colon X\to\operatorname{Spec}K a finite morphism of degree dd. Suppose that XX is connected and has a section s:SpecKXs\colon\operatorname{Spec}K\to X. Then for any commutative group scheme GG over SS, we have

f=ds:G(X)G(K).f_{*}=d\cdot s^{*}\colon G(X)\to G(K).
Proof.

Let gG(X)g\in G(X). Consider the following diagram of SS-schemes.

SpecK\textstyle{\operatorname{Spec}K\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}s\scriptstyle{s}f\scriptstyle{f^{\sharp}}X\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}Δ\scriptstyle{\Delta}X×K×KX\textstyle{X\times_{K}\dots\times_{K}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}g××g\scriptstyle{g\times\dots\times g}G×S×SG\textstyle{G\times_{S}\dots\times_{S}G\ignorespaces\ignorespaces\ignorespaces\ignorespaces}+\scriptstyle{+}SymKd(X)\textstyle{\operatorname{Sym}_{K}^{d}(X)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}σ(g)\scriptstyle{\sigma(g)}G\textstyle{G}

The left triangle is commutative by Lemma 2.1 and the right square is commutative by the definition of σ(g)\sigma(g). Therefore the total trapezoid is commutative, which implies fg=dsgf_{*}g=d\cdot s^{*}g. ∎

Lemma 2.6.

Let SS be a scheme and fi:YiX(i=1,2)f_{i}\colon Y_{i}\to X\;(i=1,2) be finite locally free morphisms of constant degree did_{i} between SS-schemes. Let GG be a commutative group scheme over SS. For any g1G(Y1)g_{1}\in G(Y_{1}) and g2G(Y2)g_{2}\in G(Y_{2}), we have

f1g1+f2g2=(f1,f2)(g1,g2){f_{1}}_{*}g_{1}+{f_{2}}_{*}g_{2}=(f_{1},f_{2})_{*}(g_{1},g_{2})

in G(X)G(X), where (f1,f2):Y1Y2X(f_{1},f_{2})\colon Y_{1}\sqcup Y_{2}\to X and (g1,g2):Y1Y2G(g_{1},g_{2})\colon Y_{1}\sqcup Y_{2}\to G are morphisms induced by the universal property of coproducts.

Proof.

Consider the following diagram of SS-schemes.

X\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(f1,f2)\scriptstyle{(f_{1}^{\sharp},f_{2}^{\sharp})}SymXd1(Y1)×XSymXd2(Y2)\textstyle{\operatorname{Sym}^{d_{1}}_{X}(Y_{1})\times_{X}\operatorname{Sym}^{d_{2}}_{X}(Y_{2})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}σ(g1)×σ(g2)\scriptstyle{\sigma(g_{1})\times\sigma(g_{2})}π\scriptstyle{\pi}G×SG\textstyle{G\times_{S}G\ignorespaces\ignorespaces\ignorespaces\ignorespaces}+\scriptstyle{+}X\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(f1,f2)\scriptstyle{(f_{1},f_{2})^{\sharp}}SymXd1+d2(Y1Y2)\textstyle{\operatorname{Sym}^{d_{1}+d_{2}}_{X}(Y_{1}\sqcup Y_{2})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}σ(g1,g2)\scriptstyle{\sigma(g_{1},g_{2})}G\textstyle{G}

Here, the morphism π\pi is induced by the morphism pp given in Lemma 2.2. The left square is commutative by Lemma 2.2 and the right square is commutative by the definition of σ\sigma. Therefore the total rectangle is commutative, which implies the desired equality. ∎

Corollary 2.7.

Let SS be a scheme, KK an algebraically closed field over SS and f:XSpecKf\colon X\to\operatorname{Spec}K a finite morphism. Write X={x1,,xn}X=\{x_{1},\dots,x_{n}\} and set di=dimK𝒪X,xid_{i}=\dim_{K}\mathcal{O}_{X,x_{i}}. Let pi:{xi}SpecKp_{i}\colon\{x_{i}\}\xrightarrow{\sim}\operatorname{Spec}K be the canonical projection and ri:{xi}Xr_{i}\colon\{x_{i}\}\to X be the canonical closed immersion. Then for any commutative group scheme GG over SS, we have

f=i=1ndipiri:G(X)G(K).\textstyle f_{*}=\sum_{i=1}^{n}d_{i}\cdot{p_{i}}_{*}r_{i}^{*}\colon G(X)\to G(K).
Proof.

This follows immediately from Lemma 2.5 and Lemma 2.6. ∎

3. The canonical transfer

In this section we fix a noetherian base scheme SS and a separated commutative group scheme GG over SS. First we note the following fact.

Lemma 3.1.

Let f:XYf\colon X\to Y be a dominant morphism between SS-schemes. If YY is reduced, then f:G(Y)G(X)f^{*}\colon G(Y)\to G(X) is injective.

Proof.

Suppose that g,h:YGg,h\colon Y\to G are two morphisms over SS and fg=fhf^{*}g=f^{*}h. Then (g,h):YG×SG(g,h)\colon Y\to G\times_{S}G sends the generic points of YY to points inside the diagonal ΔG\Delta_{G}. Since GG is separated over SS, the diagonal ΔG\Delta_{G} is closed in G×SGG\times_{S}G and hence the image of YY lies in ΔG\Delta_{G} set-theoretically. Since YY is reduced, this morphism factors through ΔG\Delta_{G} scheme-theoretically, i.e. g=hg=h. ∎

Let X,YX,Y be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). We construct the canonical transfer α:G(Y)G(X)\alpha^{*}\colon G(Y)\to G(X) in the following two cases:

  1. (1)

    α\alpha is flat over XX.

  2. (2)

    XX is normal.

First suppose that α\alpha is flat over XX. Write α=i=1nmi[Vi]\alpha=\sum_{i=1}^{n}m_{i}[V_{i}]. Then each ViV_{i} is finite locally free of constant rank over XX. Let pi:ViXp_{i}\colon V_{i}\to X and qi:ViYq_{i}\colon V_{i}\to Y be canonical projections. We define

α=i=1nmipiqi:G(Y)G(X).\textstyle\alpha^{*}=\sum_{i=1}^{n}m_{i}{p_{i}}_{*}q_{i}^{*}\colon G(Y)\to G(X).

Note that for an SS-morphism f:XYf\colon X\to Y, this definition of ff^{*} coincides with the usual pullback.

Next suppose that XX is normal. Take a dense open subset UXU\subset X such that α|U\alpha|_{U} is flat over UU. By the next lemma, for any gG(Y)g\in G(Y) the element (α|U)gG(U)(\alpha|_{U})^{*}g\in G(U) lies in the image of G(X)G(U)G(X)\to G(U). We define α:G(Y)G(X)\alpha^{*}\colon G(Y)\to G(X) by setting αg:=(α|U)g\alpha^{*}g:=(\alpha|_{U})^{*}g; this does not depend on the choice of UU.

Lemma 3.2.

Let XX be a noetherian normal integral scheme, UXU\subset X be a dense open subset and f:VXf\colon V\to X a finite morphism. Set VU=V×XUV_{U}=V\times_{X}U and suppose that fU:VUUf_{U}\colon V_{U}\to U is finite locally free of constant rank dd. Then for any gG(V)g\in G(V), the element fU(g|VU)G(U){f_{U}}_{*}(g|_{V_{U}})\in G(U) lies in the image of G(X)G(U)G(X)\to G(U).

Proof.

Recall that fU(g|VU){f_{U}}_{*}(g|_{V_{U}}) is defined to be the composition

UfUSymUd(VU)σ(g|VU)G.U\xrightarrow{f_{U}^{\sharp}}\operatorname{Sym}^{d}_{U}(V_{U})\xrightarrow{\sigma(g|_{V_{U}})}G.

By construction, we have SymUd(VU)SymXd(V)×XU\operatorname{Sym}^{d}_{U}(V_{U})\simeq\operatorname{Sym}^{d}_{X}(V)\times_{X}U. Since XX is noetherian, SymXd(V)\operatorname{Sym}^{d}_{X}(V) is finite over XX. Since XX is normal, fUf^{\sharp}_{U} uniquely extends to a morphism f:XSymXd(V)f^{\sharp}\colon X\to\operatorname{Sym}_{X}^{d}(V). Then fU(g|VU){f_{U}}_{*}(g|_{V_{U}}) is the image of

XfSymXd(V)σ(g)GX\xrightarrow{f^{\sharp}}\operatorname{Sym}^{d}_{X}(V)\xrightarrow{\sigma(g)}G

under G(X)G(U)G(X)\to G(U). ∎

Example 3.3.

Let X,YX,Y be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). Suppose that XX is normal. Write α=i=1nmi[Vi]\alpha=\sum_{i=1}^{n}m_{i}[V_{i}] and let qi:ViYq_{i}\colon V_{i}\to Y be the canonical projection. Then the canonical transfer α:𝔾m,S(Y)𝔾m,S(X)\alpha^{*}\colon\mathbb{G}_{m,S}(Y)\to\mathbb{G}_{m,S}(X) defined above is given by

αg=i=1n(Nmk(Vi)/k(X)(qig))mi.\textstyle\alpha^{*}g=\prod_{i=1}^{n}(\operatorname{Nm}_{k(V_{i})/k(X)}(q_{i}^{*}g))^{m_{i}}.

This follows from Lemma 2.3 and Lemma 2.4. Similarly, the canonical transfer α:𝔾a,S(Y)𝔾a,S(X)\alpha^{*}\colon\mathbb{G}_{a,S}(Y)\to\mathbb{G}_{a,S}(X) is given by

αg=i=1nmiTrk(Vi)/k(X)(qig).\textstyle\alpha^{*}g=\sum_{i=1}^{n}m_{i}\operatorname{Tr}_{k(V_{i})/k(X)}(q_{i}^{*}g).

In the following, we will prove that the canonical transfer defined above is functorial, i.e. αβ=(βα)\alpha^{*}\beta^{*}=(\beta\circ\alpha)^{*}. We start with a special case.

Lemma 3.4.

Let X,YX,Y be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). Suppose that α\alpha is flat over XX and YY is separated and of finite type over SS. Let KK be an algebraically closed field and ρcS(SpecK,X)\rho\in c_{S}(\operatorname{Spec}K,X). Then we have

ρα=(αρ):G(Y)G(K).\rho^{*}\alpha^{*}=(\alpha\circ\rho)^{*}\colon G(Y)\to G(K).
Proof.

Firstly, ρcS(SpecK,X)\rho\in c_{S}(\operatorname{Spec}K,X) is a formal \mathbb{Z}-linear combination of KK-rational points on SpecK×SX\operatorname{Spec}K\times_{S}X. By linearity we may assume that ρ\rho is a morphism from SpecK\operatorname{Spec}K to XX over SS. Write α=i=1nmi[Vi]\alpha=\sum_{i=1}^{n}m_{i}[V_{i}]. Then each ViV_{i} is finite and flat over XX and hence finite locally free of constant rank over XX. Consider the following diagram, where the left square is Cartesian.

Vi×XSpecK\textstyle{V_{i}\times_{X}\operatorname{Spec}K\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}ρi\scriptstyle{\rho^{\prime}_{i}}pi\scriptstyle{p^{\prime}_{i}}Vi\textstyle{V_{i}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}pi\scriptstyle{p_{i}}qi\scriptstyle{q_{i}}Y\textstyle{Y}SpecK\textstyle{\operatorname{Spec}K\ignorespaces\ignorespaces\ignorespaces\ignorespaces}ρ\scriptstyle{\rho}X\textstyle{X}

By Lemma 2.4, we have

ρα=i=1nmiρpiqi=i=1nmipiρiqi.\textstyle\rho^{*}\alpha^{*}=\sum_{i=1}^{n}m_{i}\rho^{*}{p_{i}}_{*}q_{i}^{*}=\sum_{i=1}^{n}m_{i}{p^{\prime}_{i}}_{*}{\rho^{\prime}_{i}}^{*}q_{i}^{*}.

On the other hand, since α\alpha is flat over XX, we have

αρ=αXSpecK=i=1nmi[Vi×XSpecK]\displaystyle\textstyle\alpha\circ\rho=\alpha\otimes_{X}\operatorname{Spec}K=\textstyle\sum_{i=1}^{n}m_{i}[V_{i}\times_{X}\operatorname{Spec}K]

by Lemma 1.3. Write [Vi×XSpecK]=j=1nimij[ξij][V_{i}\times_{X}\operatorname{Spec}K]=\sum_{j=1}^{n_{i}}m_{ij}[\xi_{ij}]. Let pij:{ξij}SpecKp_{ij}\colon\{\xi_{ij}\}\to\operatorname{Spec}K be the canonical projection and rij:{ξij}Vi×XSpecKr_{ij}\colon\{\xi_{ij}\}\to V_{i}\times_{X}\operatorname{Spec}K be the canonical closed immersion. Then we have

(αρ)=i=1nmij=1nimijpijrijρiqi.\textstyle(\alpha\circ\rho)^{*}=\sum_{i=1}^{n}m_{i}\sum_{j=1}^{n_{i}}m_{ij}{p_{ij}}_{*}r_{ij}^{*}{\rho_{i}^{\prime}}^{*}q_{i}^{*}.

Therefore it suffices to show that

pi=j=1nimijpijrij\textstyle{p^{\prime}_{i}}_{*}=\sum_{j=1}^{n_{i}}m_{ij}{p_{ij}}_{*}r_{ij}^{*}

holds for each ii. This follows from Corollary 2.7. ∎

Lemma 3.5.

Let X,YX,Y be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). Suppose that XX is normal and YY is separated and of finite type over SS. Let KK be an algebraically closed field and ρ:SpecKX\rho\colon\operatorname{Spec}K\to X be a morphism onto the generic point of XX. Then we have

ρα=(αρ):G(Y)G(K).\rho^{*}\alpha^{*}=(\alpha\circ\rho)^{*}\colon G(Y)\to G(K).
Proof.

Let UXU\subset X be a dense open subset such that α|U\alpha|_{U} is flat over UU. Let j:UXj\colon U\to X be the canonical open immersion and ρ:SpecKU\rho^{\prime}\colon\operatorname{Spec}K\to U be the restriction of ρ\rho. Consider the following diagram.

G(Y)\textstyle{G(Y)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}α\scriptstyle{\alpha^{*}}(αρ)\scriptstyle{(\alpha\circ\rho)^{*}}G(Y)\textstyle{G(Y)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}(α|U)\scriptstyle{(\alpha|_{U})^{*}}(α|Uρ)\scriptstyle{(\alpha|_{U}\circ\rho^{\prime})^{*}}G(K)\textstyle{G(K)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}G(X)\textstyle{G(X)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}j\scriptstyle{j^{*}}ρ\scriptstyle{\rho^{*}}G(U)\textstyle{G(U)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}ρ\scriptstyle{{\rho^{\prime}}^{*}}G(K)\textstyle{G(K)}

Let us verify that all faces are commutative. Two triangles on the top and the bottom are clearly commutative. The commutativity of the left square follows from the definition of α:G(Y)G(X)\alpha^{*}\colon G(Y)\to G(X). The commutativity of the right square follows from Lemma 3.4. Therefore we get ρα=(αρ)\rho^{*}\alpha^{*}=(\alpha\circ\rho)^{*}. ∎

Lemma 3.6.

Let X,Y,XX,Y,X^{\prime} be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). Suppose that XX is normal and YY is separated and of finite type over SS. Let f:XXf\colon X^{\prime}\to X be a dominant morphism such that αf\alpha\circ f is flat over XX^{\prime}. Then we have

fα=(αf):G(Y)G(X).f^{*}\alpha^{*}=(\alpha\circ f)^{*}\colon G(Y)\to G(X^{\prime}).
Proof.

Let KK be an algebraic closure of the function field of XX^{\prime} and ρ:SpecKX\rho\colon\operatorname{Spec}K\to X^{\prime} be the canonical morphism. Then we have

ρ(αf)\displaystyle\rho^{*}(\alpha\circ f)^{*} =(αfρ)\displaystyle{}={}(\alpha\circ f\circ\rho)^{*} (by Lemma 3.4)
=(fρ)α\displaystyle{}={}(f\circ\rho)^{*}\alpha^{*} (by Lemma 3.5)
=ρfα.\displaystyle{}={}\rho^{*}f^{*}\alpha^{*}.

Since ρ:G(X)G(K)\rho^{*}\colon G(X^{\prime})\to G(K) is injective, this completes the proof. ∎

Lemma 3.7.

Let X,YX,Y be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y). Suppose that XX is normal and YY is separated and of finite type over SS. Let KK be an algebraically closed field and ρcS(SpecK,X)\rho\in c_{S}(\operatorname{Spec}K,X). Then we have

ρα=(αρ):G(Y)G(K).\rho^{*}\alpha^{*}=(\alpha\circ\rho)^{*}\colon G(Y)\to G(K).
Proof.

As in the proof of Lemma 3.4, we may assume that ρ\rho is a morphism SpecKX\operatorname{Spec}K\to X over SS. By Lemma 1.4, there is a dominant blow-up π:XX\pi\colon X^{\prime}\to X such that απ=αX[X]\alpha\circ\pi=\alpha\otimes_{X}[X^{\prime}] is flat over XX^{\prime}. Since KK is algebraically closed, we can lift ρ\rho to ρ:SpecKX\rho^{\prime}\colon\operatorname{Spec}K\to X^{\prime}. Then we have

ρα\displaystyle\rho^{*}\alpha^{*} =ρπα\displaystyle{}={}{\rho^{\prime}}^{*}\pi^{*}\alpha^{*}
=ρ(απ)\displaystyle{}={}{\rho^{\prime}}^{*}(\alpha\circ\pi)^{*} (by Lemma 3.6)
=(απρ)\displaystyle{}={}(\alpha\circ\pi\circ\rho^{\prime})^{*} (by Lemma 3.4)
=(αρ).\displaystyle{}={}(\alpha\circ\rho)^{*}.

This completes the proof. ∎

Theorem 3.8.

Let X,Y,ZX,Y,Z be noetherian integral schemes over SS and αcS(X,Y)\alpha\in c_{S}(X,Y), βcS(Y,Z)\beta\in c_{S}(Y,Z). Suppose that X,YX,Y are normal and Y,ZY,Z are separated and of finite type over SS. Then we have

αβ=(βα):G(Z)G(X).\alpha^{*}\beta^{*}=(\beta\circ\alpha)^{*}\colon G(Z)\to G(X).

In particular, if SS is normal then the canonical transfer defines a structure of a presheaf with transfers over SS on GG.

Proof.

Let KK be an algebraic closure of the function field of XX and let ρ:SpecKX\rho\colon\operatorname{Spec}K\to X be the canonical morphism. Repeatedly applying Lemma 3.7, we get

ραβ\displaystyle\rho^{*}\alpha^{*}\beta^{*} =(αρ)β\displaystyle{}={}(\alpha\circ\rho)^{*}\beta^{*}
=(βαρ)\displaystyle{}={}(\beta\circ\alpha\circ\rho)^{*}
=ρ(βα).\displaystyle{}={}\rho^{*}(\beta\circ\alpha)^{*}.

Since ρ:G(X)G(K)\rho^{*}\colon G(X)\to G(K) is injective, this completes the proof. ∎

4. Characterization

In this section we characterize the canonical transfer by a simple condition on radicial transfers. We continue to fix a noetherian base scheme SS and a separated commutative group scheme GG over SS. First we note the following remarkable property of the canonical transfer.

Lemma 4.1.

Let X,YX,Y be noetherian integral SS-schemes and VX×SYV\subset X\times_{S}Y be an integral closed subscheme finite flat radicial of degree dd over XX. Suppose that YY is separated and of finite type over SS. Let p:VXp\colon V\to X and q:VYq\colon V\to Y be the canonical projections. Then for any gG(Y)g\in G(Y), the element dqgd\cdot q^{*}g can be uniquely written as php^{*}h for some hG(X)h\in G(X). Moreover, this hh coincides with [V]g[V]^{*}g.

We write tV(g)t_{V}(g) for the element hh appearing in the statement.

Proof.

The uniqueness follows from Lemma 3.1. We prove that the element [V]g[V]^{*}g satisfies the condition p[V]g=dqgp^{*}[V]^{*}g=d\cdot q^{*}g. Consider the following diagram, where the left square is Cartesian.

V×XV\textstyle{V\times_{X}V\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}pr2\scriptstyle{{\mathrm{pr}}_{2}}pr1\scriptstyle{{\mathrm{pr}}_{1}}V\textstyle{V\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}q\scriptstyle{q}p\scriptstyle{p}Y\textstyle{Y}V\textstyle{V\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}p\scriptstyle{p}Δ\scriptstyle{\Delta}X\textstyle{X}

We have p[V]g=ppqg=pr1pr2qgp^{*}[V]^{*}g=p^{*}p_{*}q^{*}g={{\mathrm{pr}}_{1}}_{*}{\mathrm{pr}}_{2}^{*}q^{*}g by Lemma 2.4, so it suffices to prove that

pr1pr2g=dg{{\mathrm{pr}}_{1}}_{*}{\mathrm{pr}}_{2}^{*}g^{\prime}=d\cdot g^{\prime}

holds for any gG(V)g^{\prime}\in G(V). We claim that the diagonal morphism Δ:VV×XV\Delta\colon V\to V\times_{X}V satisfies

pr1=dΔ:G(V×XV)G(V).{{\mathrm{pr}}_{1}}_{*}=d\cdot\Delta^{*}\colon G(V\times_{X}V)\to G(V).

If this claim is proved, then evaluating at pr2g{\mathrm{pr}}_{2}^{*}g^{\prime} we get the desired equality.

Let KK be an algebraic closure of the function field of VV and let W=SpecK×XVW=\operatorname{Spec}K\times_{X}V. Then WW is finite radicial of degree dd over SpecK\operatorname{Spec}K, and has a section ΔK:SpecKW\Delta_{K}\colon\operatorname{Spec}K\to W induced by Δ\Delta. By Lemma 2.4 and the injectivity of G(V)G(K)G(V)\to G(K), it suffices to prove

pr1=dΔK:G(W)G(K).{{\mathrm{pr}}_{1}}_{*}=d\cdot\Delta_{K}^{*}\colon G(W)\to G(K).

This follows from Lemma 2.5. ∎

Actually, the above condition characterizes the canonical transfer.

Theorem 4.2.

Suppose that SS is normal and we are given a structure of a presheaf with transfers over SS on GG; write α\alpha^{\dagger} for the map induced by a finite correspondence α\alpha. Assume the following:

For any connected X,YSmSX,Y\in{\mathrm{Sm}}_{S} and for any integral closed subscheme VX×SYV\subset X\times_{S}Y finite flat radicial over XX, we have [V]g=tV(g)[V]^{\dagger}g=t_{V}(g) (see Lemma 4.1).

Then α=α\alpha^{\dagger}=\alpha^{*} holds for any finite correspondence α\alpha.

Proof.

Let X,YSmSX,Y\in{\mathrm{Sm}}_{S} be connected and αcS(X,Y)\alpha\in c_{S}(X,Y). We prove that α=α:G(Y)G(X)\alpha^{\dagger}=\alpha^{*}\colon G(Y)\to G(X) holds. Write α=i=1nmi[Vi]\alpha=\sum_{i=1}^{n}m_{i}[V_{i}]. Let KK be a separable closure of the function field of XX. Then each Vi×XSpecKV_{i}\times_{X}\operatorname{Spec}K is a disjoint union of finite radicial schemes over SpecK\operatorname{Spec}K, so we can write

Vi×XSpecK=j=1niSpecAijV_{i}\times_{X}\operatorname{Spec}K=\textstyle\coprod_{j=1}^{n_{i}}\operatorname{Spec}A_{ij}

where AijA_{ij} is a finite local KK-algebra such that the residue field LijL_{ij} is purely inseparable over KK. We conclude by limit argument that there is some connected USmSU\in{\mathrm{Sm}}_{S} and an étale morphism π:UX\pi\colon U\to X such that

Vi×XU=j=1niWij,V_{i}\times_{X}U=\textstyle\coprod_{j=1}^{n_{i}}W_{ij},

where WijW_{ij} is an irreducible closed subscheme of U×SYU\times_{S}Y whose reduction is finite flat radicial over UU. By Lemma 1.6 (2) and Lemma 1.3 we have απ=i=1nj=1nimij[(Wij)red]\alpha\circ\pi=\sum_{i=1}^{n}\sum_{j=1}^{n_{i}}m_{ij}[(W_{ij})_{\mathrm{red}}] for some mijm_{ij}, so we have (απ)=(απ):G(Y)G(U)(\alpha\circ\pi)^{\dagger}=(\alpha\circ\pi)^{*}\colon G(Y)\to G(U) by our assumption. Since π=π:G(X)G(U)\pi^{\dagger}=\pi^{*}\colon G(X)\to G(U) is injective by Lemma 3.1, we get α=α\alpha^{\dagger}=\alpha^{*}. ∎

References

  • [AHL14] Giuseppe Ancona, Annette Huber and Simon Lehalleur “On the relative motive of a commutative group scheme” In Algebraic Geometry 3, 2014 DOI: 10.14231/AG-2016-008
  • [BVK16] Luca Barbieri-Viale and Bruno Kahn “On the derived category of 1-motives” 381, Asterisque Societe Mathematique de France, 2016
  • [CD19] Denis-Charles Cisinski and Frédéric Déglise “Triangulated Categories of Mixed Motives” Springer International Publishing, 2019 DOI: 10.1007/978-3-030-33242-6
  • [Gro71] Alexander Grothendieck “Revêtements étales et groupe fondamental (SGA 1)” 224, Lecture notes in mathematics Springer-Verlag, 1971
  • [KSY] Bruno Kahn, Shuji Saito and Takao Yamazaki “Reciprocity sheaves, II” arXiv:1707.07398 [math.AG]
  • [KSY16] Bruno Kahn, Shuji Saito and Takao Yamazaki “Reciprocity sheaves (with two appendices by Kay Rülling)” In Compositio Mathematica 152.9 London Mathematical Society, 2016, pp. 1851–1898 DOI: 10.1112/S0010437X16007466
  • [MVW06] Carlo Mazza, Vladimir Voevodsky and Charles Weibel “Lecture Notes on Motivic Cohomology” 2, Clay Mathematics Monographs, 2006
  • [SS03] Michael Spieß and Tamás Szamuely “On the Albanese map for smooth quasi-projective varieties” In Mathematische Annalen 325, 2003, pp. 1–17 DOI: 10.1007/s00208-002-0359-8
  • [SV96] Andrei Suslin and Vladimir Voevodsky “Singular homology of abstract algebraic varieties.” In Inventiones mathematicae 123.1, 1996, pp. 61–94 DOI: 10.1007/BF01232367