betaFeedback

WorksMathematics

Serre’s conjecture on the Artin character for a regular local ring

Abstract

We prove Serre’s conjecture on the Artin character of a finite group G acting on a regular local ring A with isolated fixed points. The case of positive characteristic is due to Abe. In mixed characteristic we use an equivariant deformation to reduce to the case where A has an isolated singularity, in which case we can make use of the results of Beraldo and Pippi on the twisted conductor formula.

Introduction

Let A\mathcal{A} be a regular local ring with maximal ideal m\mathfrak{m} and residue field k=A/mk = \mathcal{A}/\mathfrak{m}, and let GG be a finite group acting on A\mathcal{A}. Write X=Spec⁡A\mathcal{X} = \operatorname{Spec}\mathcal{A}. Motivated by the form of the local terms in the global Lefschetz trace formula, Serre conjectured that the multiplicities of the fixed point for g≠1g \ne1 could be used to define a character of GG, analogous to the classical Artin character for local fields. More precisely, for g∈Gg \in G, put

Ig(A)=(g(a)−a∣a∈A).I_g(\mathcal{A}) = (g(a) - a \mid a \in\mathcal{A}).

We write IgI_g when the ring is understood. We make the following assumptions:

(i) the inclusion AG⊂A\mathcal{A}^{G} \subset\mathcal{A} induces an isomorphism AG/(AG∩m)≃k\mathcal{A}^{G}/(\mathcal{A}^{G} \cap\mathfrak{m}) \simeq k;

(ii) A/Ig(A)\mathcal{A}/I_g(\mathcal{A}) has finite length for g≠1g \ne1.

For g≠1g \ne1, write iA(g)=length⁡A(A/Ig(A))i_A(g)=\operatorname{length}_A(A/I_g(A)). Define

aA,G(g)={−iA(g),g≠1,∑h≠1iA(h),g=1.(1)a_{A,G}(g)= \begin{cases} -i_A(g), & g \ne1,\\ \sum_{h\ne1}i_A(h), & g=1. \end{cases} \tag*{(1)}

This is a central function on GG. For a discrete valuation ring it is the classical Artin character. The following assertion was conjectured by Serre in 1960 [18 §6].

Theorem 1.1. For every prime ℓ\ell invertible in AA, aA,Ga_{A,G} is the character of a finite-dimensional Qℓ\mathbb{Q}_{\ell}-representation of GG.

The case of a ring containing a field of positive characteristic is due to Abe [2]. In residue characteristic zero the action is tame, and the assertion follows from an easy calculation. There is an approach to the conjecture via vanishing cycles which goes back to Kato, Shuji Saito and Takeshi Saito [13 §5], as explained by Abe [2 Introduction and §2.6]. In particular for arithmetic surfaces, a proof along these lines appears in Abbes [1 Lemma 1.2 and Corollary 1.3].

We treat the mixed-characteristic case by reducing to the twisted conductor formula of Beraldo–Pippi [5 §8.5]. After completion and approximation, the action on XX comes from a regular algebra AA, finite free over a smooth algebra RR over a complete discrete valuation ring CC of characteristic zero. The group acts trivially on RR. The residue field kk of CC may be taken algebraically closed. There is a closed kk-point bb of Spec⁡R\operatorname{Spec} R whose fiber in X=Spec⁡AX=\operatorname{Spec} A is supported at a single point xx, and all the nonidentity fixed schemes are supported at xx (Proposition 2.2). We use a deformation to replace XX by a regular flat CC-scheme X′X' whose nonsmooth locus is supported at xx. This allows us to apply the results of Beraldo and Pippi [5].

Let us explain this in more detail, beginning with the deformation argument. We must construct X′X' without changing the algebras OX,x/Ig\mathcal{O}_{X,x}/I_g. Consider the RR-scheme ZZ parametrizing algebra structures with GG-action on a free module of rank rank⁡RA\operatorname{rank}_R A, and let W→ZW\to Z be its universal family. The algebra AA, with a choice of basis, defines a section z:Spec⁡R→Zz:\operatorname{Spec} R\to Z. The fibers of X→Spec⁡RX\to\operatorname{Spec} R are complete intersections, since this morphism is finite flat between regular schemes. Away from bb they also have free GG-action. It follows that Z/RZ/R is smooth along z(Spec⁡R∖{b})z(\operatorname{Spec} R\setminus\{b\}), and W/RW/R is smooth above this locus (Lemma 4.1).

The parameter lemma of §3 gives a smooth affine RR-scheme TT, a section e:Spec⁡R→Te:\operatorname{Spec} R\to T, and a morphism f:T→Zf:T\to Z with f∘e=zf\circ e=z, such that ff is smooth along e(Spec⁡R∖{b})e(\operatorname{Spec} R\setminus\{b\}). Put Y=W×ZTY=W\times_ZT. The projection Y→WY\to W is then smooth along X∖{x}X\setminus\{x\}, so Y/CY/C is smooth there. Let rr be the relative dimension of T/RT/R at e(b)e(b). Near e(b)e(b), the section is cut out by relative parameters t1,…,trt_1,\ldots,t_r. Their pullbacks to YY form a regular sequence with quotient OX,x\mathcal{O}_{X,x}, so YY is regular at xx. After shrinking, we obtain a regular flat affine CC-scheme Y=Spec⁡DY=\operatorname{Spec} D with

X=V(t1,…,tr),ti∈DG.X=V(t_1,\ldots,t_r),\qquad t_i\in D^G.

Let H⊂DH\subset D define the reduced nonsmooth locus of Y/CY/C, and let mx⊂D\mathfrak{m}_x\subset D be the ideal of xx. For g≠1g\ne1, put Jg=(g(a)−a∣a∈D)J_g=(g(a)-a\mid a\in D), and set J=⋂g≠1JgJ=\bigcap_{g\ne1}J_g. Fix an integer N≥2N\ge2, to be chosen sufficiently large. We replace the equations tit_i by

ti′=ti+δi,δi∈(HJmxN)G,X′=V(t1′,…,tr′).t'_i=t_i+\delta_i,\qquad\delta_i\in(HJ\mathfrak{m}_x^N)^G,\qquad X'=V(t'_1,\ldots,t'_r).

The factor JJ ensures that, for every $g\ne 1,

D/(t1′,…,tr′,Jg)=D/(t1,…,tr,Jg).D/(t'_1,\ldots,t'_r,J_g)=D/(t_1,\ldots,t_r,J_g).

Thus the fixed schemes agree, and their multiplicities iA(g)i_A(g) are unchanged. The factor HH similarly gives X′∩V(H)=X∩V(H)X' \cap V(H) = X \cap V(H), so X′X' meets the nonsmooth locus of the ambient scheme only at xx.

It remains to make the new equations transverse on U=Y∖(V(J)∪V(H)∪{x})U = Y \setminus\left(V(J) \cup V(H) \cup\{x\}\right). Here GG acts freely, and U→U/GU \to U/G is finite étale with smooth quotient over CC. The ideal HJmxNHJ\mathfrak{m}_x^N restricts to the unit ideal on UU. Taking products over GG of elements of this ideal nonzero on a prescribed orbit gives finitely many invariant elements with no common zero on UU. Together with their products with invariant coordinate functions, these supply arbitrary values and first derivatives at each closed point of the special fiber (U/G)k(U/G)_k. A Bertini incidence argument therefore allows us to choose the δi\delta_i so that the differentials of the ti′t'_i are independent at every common zero on that special fiber. The special fiber of X′X' is then smooth on UU, and we have seen that every point of X′∖{x}X' \setminus\{x\} lies in UU.

Finally, N≥2N \ge2 leaves the independent linear parts of the tit_i at xx unchanged, so X′X' is regular there. Moreover, OX′,x/mxN≃OX,x/mxN\mathcal{O}_{X',x}/\mathfrak{m}_x^N \simeq\mathcal{O}_{X,x}/\mathfrak{m}_x^N equivariantly. For sufficiently large NN, a uniformizer of CC remains nonzero in this regular local ring, which implies flatness over CC. After shrinking about xx, X′X' is regular and flat; its generic fiber is smooth in characteristic zero. Hence the nonsmooth locus of X′/CX'/C is supported at xx, with all fixed-point multiplicities preserved (Proposition 4.3).

The nonsmooth locus of X′/CX'/C, with its reduced structure, is proper over CC, since it is supported at the closed point xx. This is the properness hypothesis needed for the twisted Bloch conductor formula of Beraldo and Pippi [5 §8.5.6]. For g≠1g \ne1, we apply their formula to the structure sheaves of the diagonal Δ\Delta and the graph Γg\Gamma_g of gg in X′×CX′X' \times_C X'. It compares a localized intersection number with two cohomological traces arising from vanishing cycles.2 Lemma 5.1 identifies the intersection number with

length⁡OX′,x(OX′,x/Ig)=iA(g).\operatorname{length}_{\mathcal{O}_{X',x}}\left(\mathcal{O}_{X',x}/I_g\right) = i_A(g).

Now put K=Frac⁡(C)K = \operatorname{Frac}(C) and choose a sufficiently large finite Galois extension L/KL/K, with group H0=Gal⁡(L/K)H_0 = \operatorname{Gal}(L/K). The two cohomological terms yield finite virtual representations V0V_0 of GG and V1V_1 of G×H0G \times H_0 over Qℓ\mathbb{Q}_\ell. To identify their traces, we use the duality maps in Beraldo and Pippi’s construction: replacing the diagonal by the graph inserts the geometric action of gg in the trace (Lemma 5.2). Let Art⁡H0\operatorname{Art}_{H_0} be the classical Artin representation associated with L/KL/K. It is defined over Qℓ\mathbb{Q}_\ell by Serre’s theorem [19 §19.2, Theorem 44(i)]. The conductor formula becomes

iA(g)=Tr⁡(g,V0)−Tr⁡(g,Hom⁡H0(Art⁡H0,V1))(g≠1).i_A(g) = \operatorname{Tr}(g,V_0) - \operatorname{Tr}\left(g,\operatorname{Hom}_{H_0}(\operatorname{Art}_{H_0},V_1)\right) \qquad(g \ne1).

This interpretation of the Artin-weighted term is what gives integrality: taking Hom⁡H0(Art⁡H0,−)\operatorname{Hom}_{H_0}(\operatorname{Art}_{H_0},-) is an exact functor over Qℓ\mathbb{Q}_\ell. Consequently

V:=−V0+Hom⁡H0(Art⁡H0,V1)V := -V_0 + \operatorname{Hom}_{H_0}(\operatorname{Art}_{H_0},V_1)

is an integral virtual representation over Qℓ\mathbb{Q}_\ell, and Tr⁡(g,V)=−iA(g)\operatorname{Tr}(g,V) = -i_A(g) for every g≠1g \ne1. A straightforward argument which goes back to Serre’s original treatment then shows that

W=V−dim⁡(VG)Qℓ[G]W = V - \dim(V^G)\mathbb{Q}_\ell[G]

is a genuine (not just virtual) representation, and has character aA,Ga_{A,G}.

AI use

Extensive use of GPT Astra was made both to arrive at the proofs of the results presented here, and to write a first draft of the manuscript. Significant human effort was needed to check and refine the arguments, and make the paper more readable. Part of this work was done at The Overhang, held in San Francisco 9-10, September 2026. We thank the organizers and other participants for a very pleasant experience.

Reduction to finite type algebras

For a separated scheme XX with GG-action and g≠1g \ne1, let Ig(X)⊂OXI_g(X) \subset\mathcal{O}_X be the ideal sheaf of the fixed scheme Fix⁡X(g)\operatorname{Fix}_X(g). If X=Spec⁡AX = \operatorname{Spec} A, we also write Ig(A)I_g(A) for the corresponding ideal of AA, generated by the elements g(a)−ag(a)-a for a∈Aa \in A.

We first record a lemma asserting that the fixed point scheme can be detected by a finite length quotient of AA.

Lemma 2.1. Let (A,m)(A,\mathfrak{m}) and (A′,m′)(A',\mathfrak{m}') be Noetherian local rings with GG-action. Fix g∈Gg \in G and a positive integer NN. Suppose there is a GG-equivariant isomorphism α:A/mN≃A′/(m′)N\alpha: A/\mathfrak{m}^{N} \simeq A'/(\mathfrak{m}')^{N}. If L=length⁡A(A/Ig(A))<NL = \operatorname{length}_A(A/I_g(A)) < N, then this isomorphism induces

A/Ig(A)≃A′/Ig(A′).A/I_g(A) \simeq A'/I_g(A').

Proof. If L=0L = 0, equivariance of α\alpha gives Ig(A′)+(m′)N=A′I_g(A') + (\mathfrak{m}')^N = A', whence Ig(A′)=A′I_g(A') = A'. Both quotients are then zero. Assume L>0L > 0. One has mL⊂Ig(A)\mathfrak{m}^L \subset I_g(A), so the existence of α\alpha implies (m′)L⊂Ig(A′)+(m′)N(\mathfrak{m}')^L \subset I_g(A') + (\mathfrak{m}')^N. In B′=A′/Ig(A′)B' = A'/I_g(A'), with maximal ideal n′\mathfrak{n}', this gives (n′)L=(n′)N(\mathfrak{n}')^L = (\mathfrak{n}')^N. Since N>LN > L, Nakayama’s lemma, applied to (n′)L(\mathfrak{n}')^L, gives (n′)L=0(\mathfrak{n}')^L = 0. Hence α\alpha induces the claimed isomorphism. □

Under the hypotheses of Theorem 1.1, completion preserves regularity and the residue field, and the GG-action extends to A^\widehat{A}. Lemma 2.1 shows that completion preserves A/Ig(A)A/I_g(A) for g≠1g \ne1. The residue-field condition also persists, since AG⊂(A^)GA^G \subset(\widehat{A})^G already surjects onto kk. We may therefore replace AA by its completion.

Suppose now that AA is complete, and put B=AGB = A^G. We claim that AA is finite over BB. Choose generators x1,…,xrx_1,\ldots,x_r of m\mathfrak{m}, and let q⊂B\mathfrak{q} \subset B be generated by the nonleading coefficients of the orbit polynomials ∏g∈G(T−g(xi))\prod_{g \in G}(T-g(x_i)). Then xi∣G∣∈qAx_i^{|G|} \in\mathfrak{q}A for every ii, so qA\mathfrak{q}A is m\mathfrak{m}-primary. The equality of residue fields implies that A/qAA/\mathfrak{q}A is a finite BB-module. Choose lifts v1,…,vs∈Av_1,\ldots,v_s \in A of generators. Successive substitution in A=∑jBvj+qAA = \sum_j Bv_j+\mathfrak{q}A expresses any element of AA modulo qnA\mathfrak{q}^nA as ∑jbj,nvj\sum_j b_{j,n}v_j, with bj,n∈Bb_{j,n} \in B and bj,n+1−bj,n∈qnb_{j,n+1}-b_{j,n} \in\mathfrak{q}^n. These coefficients converge in AA, and their limits lie in BB, which is closed since the action is continuous. Thus the vjv_j generate AA over BB. The theorem of Eakin and Nagata [9] now shows that BB is Noetherian. Since m∩BA=m\sqrt{\mathfrak{m} \cap B}A = \mathfrak{m}, Artin–Rees identifies its maximal-ideal topology with the topology induced from AA; hence BB is also complete.

Proposition 2.2. Let AA satisfy the hypotheses of Theorem 1.1, and suppose that AA has mixed characteristic (0,p)(0,p). There exist a complete discrete valuation ring CC with uniformizer pp, with algebraically closed residue field k0k_0, an integral smooth finite-type CC-algebra RR, and a regular finite free RR-algebra AA with RR-linear GG-action, having the following properties. For some closed k0k_0-point bb of Spec⁡R\operatorname{Spec} R, the fiber of X=Spec⁡AX = \operatorname{Spec} A over bb is supported at a single point xx with residue field k0k_0. For 1≠g∈G1 \ne g \in G, the fixed scheme of gg is supported at xx, and

length⁡OX,x(OX,x/Ig)=iA(g).\operatorname{length}_{\mathcal{O}_{X,x}}(\mathcal{O}_{X,x}/I_g) = i_A(g).

Proof. We recall the argument used by Kato and Takeshi Saito [15 proof of Lemma 7.2.10], following the method of Kato, Shuji Saito and Takeshi Saito [13 Lemma 5.3].

By the preceding argument we may assume that A\mathcal{A} is complete and finite over the complete Noetherian local ring B=AG\mathcal{B}=\mathcal{A}^{G}. Put d=dim⁡A=dim⁡Bd=\dim\mathcal{A}=\dim\mathcal{B}.

Choose a Cohen coefficient ring C⊂BC\subset\mathcal{B}. It is a complete discrete valuation ring with uniformizer pp and residue field kk. Choose u1,…,ud−1∈Bu_1,\ldots,u_{d-1}\in\mathcal{B} so that p,u1,…,ud−1p,u_1,\ldots,u_{d-1} is a system of parameters. The induced map

R^=C[[u1,…,ud−1]]⟶B\widehat{R}=C[[u_1,\ldots,u_{d-1}]]\longrightarrow\mathcal{B}

is finite by complete Nakayama and hence injective by dimension. The displayed parameters form a regular sequence on A\mathcal{A}, since A\mathcal{A} is Cohen–Macaulay. The Auslander–Buchsbaum formula therefore shows that A\mathcal{A} is free over R^\widehat{R}.

Next, choose an algebraic closure k‾\overline{k} and an extension C⊂C′C\subset C' of complete discrete valuation rings with common uniformizer pp and residue extension k⊂k‾k\subset\overline{k}. Put R^′=R^⊗CC′\widehat{R}'=\widehat{R}\otimes_C C'. Replacing (C,k,R^,A)(C,k,\widehat{R},\mathcal{A}) by (C′,k‾,R^′,A⊗R^R^′)(C',\overline{k},\widehat{R}',\mathcal{A}\otimes_{\widehat{R}}\widehat{R}'), we may assume kk is algebraically closed.

Write

Rh=(C[u1,…,ud−1](p,u1,…,ud−1))hR^h=\left(C[u_1,\ldots,u_{d-1}]_{(p,u_1,\ldots,u_{d-1})}\right)^h

for the henselisation, and write q=(p,u1,…,ud−1)⊂Rh\mathfrak{q}=(p,u_1,\ldots,u_{d-1})\subset R^h. By Artin’s approximation theorem [3 Theorem 1.10], for every N≥1N\geq1, there exists a finite free RhR^h-algebra Ah\mathcal{A}^h with GG-action and a GG-equivariant isomorphism

Ah/qNAh≃A/qNA.\mathcal{A}^h/\mathfrak{q}^N\mathcal{A}^h\simeq\mathcal{A}/\mathfrak{q}^N\mathcal{A}.

The algebra Ah\mathcal{A}^h is local because its closed fiber is local. Since qNA⊂mN\mathfrak{q}^N\mathcal{A}\subset\mathfrak{m}^N, the displayed isomorphism induces an equivariant isomorphism modulo the NNth powers of the maximal ideals. For N≥2N\geq2 this gives edim⁡Ah=d=dim⁡Ah\operatorname{edim}\mathcal{A}^h=d=\dim\mathcal{A}^h, so Ah\mathcal{A}^h is regular.

Take N≥2N\geq2 larger than the lengths of A/Ig(A)\mathcal{A}/I_g(\mathcal{A}) for g≠1g\ne1. Then, by Lemma 2.1,

Ah/Ig(Ah)≃A/Ig(A)(g≠1).\mathcal{A}^h/I_g(\mathcal{A}^h)\simeq\mathcal{A}/I_g(\mathcal{A})\qquad(g\ne1).

Finally, descending to a suitable affine étale neighborhood of C[u1,…,ud−1]C[u_1,\ldots,u_{d-1}] produces the CC-algebra RR and the RR-algebra AA of the proposition. □

A parameter lemma

The following lemma is a special case of Spivakovsky’s strengthened form of Néron–Popescu desingularization [20 Theorem 1.2]. We give an elementary proof for the hypotheses needed here. Its Jacobian and adjugate calculations follow the method of Elkik [10 §0.2 and I, Lemma 1]; the related implicit-function argument is attributed by Elkik to Tougeron [10 II, Lemma 2].

Lemma 3.1. Let RR be a regular Noetherian domain, ZZ an affine scheme of finite presentation over RR, and z:Spec⁡R→Zz:\operatorname{Spec}R\to Z a section. Write Sm⁡(Z/R)\operatorname{Sm}(Z/R) for the smooth locus of Z/RZ/R and put U=z−1(Sm⁡(Z/R))U=z^{-1}(\operatorname{Sm}(Z/R)). After replacing Spec⁡R\operatorname{Spec}R by a neighborhood of a prescribed point, there are a smooth affine RR-scheme TT, a section e:Spec⁡R→Te:\operatorname{Spec}R\to T, and an RR-morphism f:T→Zf:T\to Z such that

f∘e=z,f is smooth at e(u) for every u∈U.f\circ e=z,\qquad f\text{ is smooth at }e(u)\text{ for every }u\in U.

Proof. For a smooth affine integral RR-scheme TT, a section e:Spec⁡R→Te:\operatorname{Spec}R\to T, and an RR-morphism f:T→Zf:T\to Z with f∘e=zf\circ e=z, put

U(f)={u∈U∣f is smooth at e(u)},B(f)=U∖U(f).U(f)=\{u\in U\mid f\text{ is smooth at }e(u)\},\qquad B(f)=U\setminus U(f).

Thus U(f)U(f) is open and B(f)B(f) is closed in UU. We claim that for every u0∈B(f)u_0 \in B(f) there are a smooth affine morphism p:T1→Tp:T_1 \to T with T1T_1 integral, a section s:T→T1s:T \to T_1 of pp, and an RR-morphism f1:T1→Zf_1:T_1 \to Z such that, with e1=s∘ee_1=s \circ e, one has

f1∘s=f,U(f)∪{u0}⊆U(f1).f_1 \circ s=f,\qquad U(f)\cup\{u_0\}\subseteq U(f_1).

Here U(f1)U(f_1) is defined using the section e1e_1.

The lemma follows from this claim by Noetherian induction. Start with T=Spec⁡RT=\operatorname{Spec} R, e=id⁡e=\operatorname{id} and f=zf=z. Whenever B(f)B(f) is nonempty, the claim replaces it by the proper closed subset B(f1)⊊B(f)B(f_1)\subsetneq B(f). Since UU is Noetherian, this process terminates with B(f)B(f) empty.

We prove the claim, beginning with the preservation of U(f)U(f). Suppose that p:T1→Tp:T_1 \to T is smooth, s:T→T1s:T \to T_1 is a section of pp, and f1:T1→Zf_1:T_1 \to Z is an RR-morphism with f1∘s=ff_1 \circ s=f. For u∈U(f)u \in U(f), write TuT_u, T1,uT_{1,u} and ZuZ_u for the fibers of TT, T1T_1 and ZZ over u∈Spec⁡Ru \in\operatorname{Spec} R. Every tangent vector to ZuZ_u at z(u)z(u) lifts under ff to a tangent vector to TuT_u at e(u)e(u). Applying ss gives a lift under f1f_1 to T1,uT_{1,u} at e1(u)e_1(u). Since T1T_1 and ZZ are smooth over RR at these points, f1f_1 is smooth at e1(u)e_1(u).

It remains to construct pp, ss, f1f_1 with f1∘s=ff_1 \circ s=f and u0∈U(f1)u_0 \in U(f_1). Put A=Γ(T,OT)A=\Gamma(T,\mathcal{O}_T), a regular Noetherian domain, and set

W=Z×RT,σ=(f,id⁡):T⟶W,t0=e(u0).W=Z\times_R T,\qquad\sigma=(f,\operatorname{id}):T\longrightarrow W,\qquad t_0=e(u_0).

The morphism W→TW\to T is smooth at σ(t0)\sigma(t_0), since u0∈Uu_0 \in U. We will construct pp, ss and a TT-morphism h:T1→Wh:T_1\to W with h∘s=σh\circ s=\sigma and hh smooth at s(t0)s(t_0). The projection W→ZW\to Z is smooth since T/RT/R is smooth, so composing hh with it will give the required f1f_1.

Choose a presentation W=Spec⁡(A[Y1,…,Ym]/I)W=\operatorname{Spec}(A[Y_1,\ldots,Y_m]/I) and put c=(σ∗Y1,…,σ∗Ym)∈Amc=(\sigma^*Y_1,\ldots,\sigma^*Y_m)\in A^m. Choose equations F=(F1,…,Fq)⊂IF=(F_1,\ldots,F_q)\subset I which generate II near σ(t0)\sigma(t_0), and reorder the variables so that the Jacobian minor

Δ=det⁡(∂Fi∂Yj)1≤i,j≤q\Delta=\det\left(\frac{\partial F_i}{\partial Y_j}\right)_{1\leq i,j\leq q}

is invertible at σ(t0)\sigma(t_0). Such choices exist by the Jacobian criterion. Clearing denominators gives a∈((F):I)a\in((F):I) with a(c)a(c) invertible at t0t_0; in particular, a(c)Δ(c)≠0a(c)\Delta(c)\ne0.

Put b=a(c)b=a(c) and d=Δ(c)d=\Delta(c), and write the Jacobian matrix of FF at cc in blocks (J∣J1)(J\mid J_1), where

J=(∂Fi∂Yj(c))1≤i,j≤q,J1=(∂Fi∂Yj(c))1≤i≤qq+1≤j≤m.J=\left(\frac{\partial F_i}{\partial Y_j}(c)\right)_{1\leq i,j\leq q},\qquad J_1=\left(\frac{\partial F_i}{\partial Y_j}(c)\right)_{\substack{1\leq i\leq q\\q+1\leq j\leq m}}.

Thus det⁡J=d\det J=d. Let JadJ^{\mathrm{ad}} be the adjugate of JJ. For vectors of variables ww of length qq and vv of length m−qm-q, define the vector δ=(δ1,…,δm)\delta=(\delta_1,\ldots,\delta_m) in two blocks by

δ1,…,q=db2Jadw,δq+1,…,m=d2b2v.(2)\delta_{1,\ldots,q}=db^2J^{\mathrm{ad}}w,\qquad \delta_{q+1,\ldots,m}=d^2b^2v. \tag*{(2)}

Since F(c)=0F(c)=0, expansion at cc gives

F(c+δ)=Jδ1,…,q+J1δq+1,…,m+d2b4Q(w,v)=db2JJadw+d2b2J1v+d2b4Q(w,v)=d2b2(w+J1v+b2Q(w,v)).(3)\begin{aligned} F(c+\delta) &=J\delta_{1,\ldots,q}+J_1\delta_{q+1,\ldots,m}+d^2b^4Q(w,v)\\ &=db^2JJ^{\mathrm{ad}}w+d^2b^2J_1v+d^2b^4Q(w,v)\\ &=d^2b^2\bigl(w+J_1v+b^2Q(w,v)\bigr). \tag*{(3)} \end{aligned}

Here Q∈A[w,v]qQ\in A[w,v]^q has no terms of degree less than two: each term beyond the linear part contains at least two factors δi\delta_i, each divisible by db2db^2. The last equality uses JJad=d id⁡JJ^{\mathrm{ad}}=d\,\operatorname{id}. Consider the closed subscheme of AAm=Spec⁡A[w,v]A_A^m=\operatorname{Spec} A[w,v] defined by

w+J1v+b2Q(w,v)=0.w+J_1v+b^2Q(w,v)=0.

Along the section w=v=0w = v = 0, the Jacobian matrix with respect to ww is id⁡q\operatorname{id}_q. Inverting its determinant gives an affine open neighborhood T1T_1 of the image of this section in the subscheme just defined. The projection (w,v)↦v(w,v) \mapsto v defines an étale morphism T1→AAm−qT_1 \to\mathbb{A}^{m-q}_A, and hence a smooth morphism p:T1→Tp:T_1 \to T. The section s:T→T1s:T \to T_1 is given by w=v=0w = v = 0.

Furthermore,

a(c+δ)=b(1+bQ′)a(c+\delta)=b(1+bQ')

for a polynomial Q′∈A[w,v]Q' \in A[w,v] vanishing at zero. Invert 1+bQ′1+bQ'. Since T1T_1 is flat over the domain AA, multiplication by bb is injective on OT1\mathcal{O}_{T_1}. For P∈IP \in I, the relation aP∈(F)aP \in(F) and (3.1.2) now give P(c+δ)=0P(c+\delta)=0. Thus c+δc+\delta defines a TT-morphism h:T1→Wh:T_1 \to W with h∘s=σh \circ s=\sigma. Since T1T_1 is smooth over the regular ring AA, it is regular. The connected image s(T)s(T) lies in a single connected component of T1T_1, which is open, closed and integral. Replacing T1T_1 by this component keeps it affine and smooth over TT, and makes it integral.

To check smoothness of hh, fix t∈D(bd)⊂Tt \in D(bd) \subset T and write T1,tT_{1,t} and WtW_t for the fibers of p:T1→Tp:T_1 \to T and W→TW \to T over tt. In the following calculation, all matrices and scalars are taken over κ(t)\kappa(t). Since QQ has no linear terms, a tangent vector (w˙,v˙)(\dot{w},\dot{v}) to T1,tT_{1,t} at s(t)s(t) satisfies w˙+J1v˙=0\dot{w}+J_1\dot{v}=0. Thus these tangent vectors have the form (−J1τ,τ)(-J_1\tau,\tau), with τ∈κ(t)m−q\tau\in\kappa(t)^{m-q}. By (3.1.1), hh sends such a tangent vector to

(−db2Jad⁡J1τ, d2b2τ)=(bd)2(−J−1J1τ,τ).\left(-db^2J^{\operatorname{ad}}J_1\tau,\ d^2b^2\tau\right)=(bd)^2\left(-J^{-1}J_1\tau,\tau\right).

Since a(c)=ba(c)=b is invertible at tt, the FiF_i generate II near σ(t)\sigma(t). Hence the tangent vectors (ξ,η)(\xi,\eta) to WtW_t at σ(t)\sigma(t) are exactly the solutions of Jξ+J1η=0J\xi+J_1\eta=0. As det⁡J=d\det J=d is invertible, they are precisely (−J−1J1τ,τ)(-J^{-1}J_1\tau,\tau) for τ∈κ(t)m−q\tau\in\kappa(t)^{m-q}. The scalar (bd)2(bd)^2 is also invertible, so hh induces an isomorphism between the tangent space of T1,tT_{1,t} at s(t)s(t) and that of WtW_t at σ(t)\sigma(t). Both T1T_1 and WW are smooth over TT at these points, so hh is étale at s(t)s(t). Since t0∈D(bd)t_0 \in D(bd) and W→ZW \to Z is smooth, the composite f1:T1→hW→Zf_1:T_1 \xrightarrow{h} W \to Z is smooth at e1(u0)=s(t0)e_1(u_0)=s(t_0). It satisfies f1∘s=ff_1 \circ s=f, and hence U(f)∪{u0}⊆U(f1)U(f)\cup\{u_0\}\subseteq U(f_1), proving the claim.

Equivariant deformation

We now work over the ring CC constructed in Proposition 2.2, and write kk for its algebraically closed residue field k0k_0. All group actions on CC-schemes are over CC.

Finite algebras with group action

For a positive integer nn, let Zn,GZ_{n,G} be the scheme of commutative associative algebra structures with unit e1e_1 on a free module with ordered basis e1,…,ene_1,\ldots,e_n, together with an action of GG by algebra automorphisms. Thus, for a ring BB, a BB-point is an algebra structure on BnB^n with specified unit e1e_1, and invertible matrices ρg\rho_g preserving multiplication and the unit, with ρ1=1\rho_1=1 and ρgh=ρgρh\rho_{gh}=\rho_g\rho_h. This defines an affine scheme of finite type over Z\mathbb{Z}. Without the GG-action, this is the scheme of based algebras with specified first basis vector as unit studied by Poonen [17 Proposition 1.1 and Definition 1.5]. Denote the spectrum of the universal algebra by

Wn,G⟶Zn,G.W_{n,G} \longrightarrow Z_{n,G}.

If a BB-point z∈Zn,G(B)z \in Z_{n,G}(B) corresponds to a BB-algebra DD, then lifting zz to a BB-point of Wn,GW_{n,G} is equivalent to giving an augmentation D→BD \to B. We will use this functor of points description of Wn,GW_{n,G}.

Lemma 4.1. Let zz be a geometric point of Zn,GZ_{n,G} corresponding to a finite algebra which is a local complete intersection and has a free GG-action. Then Zn,GZ_{n,G} is smooth over Z\mathbb{Z} at zz, and Wn,GW_{n,G} is smooth over Z\mathbb{Z} at every point above zz.

Proof. When GG is trivial, these assertions are the smoothness of the finite syntomic parameter space and of its universal family, as proved in The Stacks Project [21 Lemma 49.11.4 and Remark 49.11.6]. For Wn,GW_{n,G}, one needs to check that one can lift the augmentation corresponding to a point of Wn,GW_{n,G}. The complete-intersection presentation can be chosen with its variables in the augmentation ideal; lifting its equations with zero constant term lifts both the algebra and the augmentation.

For general GG, let B′↠BB' \twoheadrightarrow B be a small extension of local Artinian rings, and let D/BD/B be the GG-equivariant BB-algebra corresponding to a point zB∈Zn,G(B)z_B \in Z_{n,G}(B) lifting zz. It is finite syntomic over BB, since it is finite flat and its geometric closed fiber is a complete intersection; see The Stacks Project [21 Lemma 49.11.1]. As GG acts freely on Spec⁡D\operatorname{Spec} D, Spec⁡D→Spec⁡DG\operatorname{Spec} D \to\operatorname{Spec} D^G is a finite étale GG-torsor. Hence the algebra DG/BD^G/B is finite syntomic by faithfully flat descent of flatness and étale descent of the complete-intersection property. By the case when GG is trivial, it lifts to (DG)′/B′(D^G)'/B'. Grothendieck’s invariance theorem under nilpotent thickenings [11 Exposé I, Theorem 8.3] lifts the finite étale torsor Spec⁡D→Spec⁡DG\operatorname{Spec} D \to\operatorname{Spec} D^G and its action uniquely to a finite étale GG-torsor Spec⁡D′→Spec⁡(DG)′\operatorname{Spec} D' \to\operatorname{Spec}(D^G)'.

Finally, if z~B∈Wn,G(B)\widetilde{z}_B \in W_{n,G}(B) lifts zBz_B, then we may also lift the corresponding augmentation D→BD \to B, by first choosing the lift (DG)′/B′(D^G)'/B' together with an augmentation (DG)′→B′(D^G)' \to B' lifting DG→BD^G \to B, and then using formal étaleness to extend it to D′→B′D' \to B' lifting D→BD \to B. Lifting the chosen basis of D/BD/B to D′/B′D'/B', with first vector 11, completes the proof by the infinitesimal criterion for smoothness.

Invariant sections

The transversality argument below is the classical Bertini incidence argument, applied on the free quotient; compare Katz [8 Exposé XVII, §3.1]. The congruences imposed on the sections will also preserve the fixed algebras.

We shall shrink affine GG-schemes about a fixed point xx. Such shrinkings may be taken affine and invariant: replace a principal neighborhood D(a)D(a) by D(∏g∈Gg(a))D\left(\prod_{g\in G}g(a)\right).

For a CC-scheme YY, write Ys=Y×CkY_s=Y\times_C k for its special fiber and Crit⁡(Y/C)\operatorname{Crit}(Y/C) for its reduced nonsmooth locus.

Lemma 4.2. Let Y=Spec⁡DY=\operatorname{Spec} D be regular, flat, and of finite type over CC. Let GG act on YY, fixing a closed kk-point xx. Suppose that invariant functions t1,…,trt_1,\ldots,t_r define a subscheme XX which is regular of codimension rr and flat over CC near xx. Assume that

X∩Crit⁡(Y/C)andFix⁡X(g) (g≠1)X \cap\operatorname{Crit}(Y/C) \quad\text{and}\quad\operatorname{Fix}_X(g)\ (g\ne1)

are supported at xx. For every sufficiently large integer NN, there are invariant functions ti′=ti+δit'_i=t_i+\delta_i, with δi∈mxN\delta_i\in\mathfrak{m}_x^N, whose common zero scheme X′X' has the following properties after an invariant shrinking about xx:

(i) X′X' is regular and flat over CC, and Crit⁡(X′/C)\operatorname{Crit}(X'/C) is supported at xx;

(ii) OX′,x/Ig(X′)x≃OX,x/Ig(X)x\mathcal{O}_{X',x}/I_g(X')_x \simeq\mathcal{O}_{X,x}/I_g(X)_x for every g≠1g\ne1;

(iii) the order-NN truncations at xx of X′X' and XX are GG-equivariantly isomorphic.

Proof. Restrict to the component of YY containing xx, and let n⊂D\mathfrak{n} \subset D be the ideal of xx. The case r=0r=0 is immediate. Let π\pi be a uniformizer of CC. Choose N≥2N \ge2 such that π∉mX,xN\pi\notin\mathfrak{m}_{X,x}^{N}. Let HH define Crit⁡(Y/C)\operatorname{Crit}(Y/C). Set

J=⋂g≠1Ig(D),a=HJN.J=\bigcap_{g\ne1}I_g(D), \qquad\mathfrak{a}=H J^N.

We shall choose δi∈aG\delta_i \in\mathfrak{a}^G.

Put U=Y∖(V(H)∪V(J)∪{x})U=Y\setminus\bigl(V(H)\cup V(J)\cup\{x\}\bigr). The quotient

U⟶V=U/GU \longrightarrow V=U/G

is a finite étale torsor; see The Stacks Project [21 Lemma 66.14.2]. Thus VV is smooth over CC, and is open in Spec⁡DG\operatorname{Spec}D^G. The orbit polynomials show that DD is finite over DGD^G; hence DGD^G is of finite type over CC by Artin–Tate. The ideal a\mathfrak{a} restricts to the unit ideal on UU. For each orbit in UU, prime avoidance gives an element of a\mathfrak{a} nonzero along the orbit. Taking its norm and using quasi-compactness gives finitely many aj∈aGa_j\in\mathfrak{a}^G with no common zero on VV.

We choose the perturbations first on the special fiber VsV_s. The invariant functions tit_i descend to VV; write tˉi\bar{t}_i for their restrictions to VsV_s. We seek perturbations δˉi\bar{\delta}_i for which the common zero scheme of tˉ1+δˉ1,…,tˉr+δˉr\bar{t}_1+\bar{\delta}_1,\ldots,\bar{t}_r+\bar{\delta}_r is smooth of codimension rr in VsV_s. All differentials in this construction are taken on VsV_s over kk. If VsV_s is empty, take δi=0\delta_i=0 and omit this choice. Otherwise, VsV_s is smooth of pure dimension

n=dim⁡OX,x+r−1≥r.n=\dim\mathcal{O}_{X,x}+r-1\ge r.

The equality follows from the dimension formula for Y/CY/C, and the inequality from the flatness of X/CX/C.

Choose generators bhb_h of the CC-algebra DGD^G, and let EE be the finite-dimensional kk-space of functions on VsV_s spanned by the restrictions of the aja_j and ajbha_jb_h. At every closed point v∈Vsv\in V_s, these functions allow us to prescribe both a value and a differential: the map

E⟶k⊕Tv∗Vs,ϵ⟼(ϵ(v),dvϵ)E\longrightarrow k\oplus T_v^*V_s,\qquad\epsilon\longmapsto\bigl(\epsilon(v),d_v\epsilon\bigr)

is surjective. Here Tv∗VsT_v^*V_s is the cotangent space of VsV_s at vv. Indeed, choose jj with aj(v)≠0a_j(v)\ne0. Since the bhb_h generate DGD^G over CC, their differentials dvbhd_vb_h span Tv∗VsT_v^*V_s. The functions

aj(bh−bh(v))=ajbh−bh(v)aja_j\bigl(b_h-b_h(v)\bigr)=a_jb_h-b_h(v)a_j

belong to EE, vanish at vv, and have differentials aj(v)dvbha_j(v)d_vb_h. They therefore give every differential with value zero, while aja_j supplies an arbitrary value.

Now vary a tuple ϵ=(ϵ1,…,ϵr)\epsilon=(\epsilon_1,\ldots,\epsilon_r) in the affine space ErE^r. Let Σ⊂Vs×Er\Sigma\subset V_s\times E^r be the closed locus of pairs (v,ϵ)(v,\epsilon) for which all the functions tˉi+ϵi\bar{t}_i+\epsilon_i vanish at vv and their differentials in Tv∗VsT_v^*V_s are linearly dependent. For a fixed vv, the surjectivity just proved shows that vanishing imposes rr independent conditions on the tuple, and that, subject to these conditions, its rr differentials can be prescribed arbitrarily. In a basis of Tv∗VsT_v^*V_s they form an arbitrary rr by nn matrix. The locus where its rank is less than rr has codimension n−r+1n-r+1. Thus the fiber of Σ→Vs\Sigma\to V_s over vv has codimension n+1n+1 in ErE^r, and

dim⁡Σ≤n+dim⁡Er−(n+1)=dim⁡Er−1.\dim\Sigma\le n+\dim E^r-(n+1)=\dim E^r-1.

The closure of the image of Σ\Sigma in ErE^r is therefore a proper subset. Since kk is infinite, choose a kk-point outside this closure, and write δˉi\bar{\delta}_i for its components. At every common zero of the tˉi+δˉi\bar{t}_i+\bar{\delta}_i, their differentials are independent. The Jacobian criterion shows that their common zero scheme in VsV_s is smooth of codimension rr.

Express each δˉi\bar{\delta}_i as a kk-linear combination of the restrictions of the aja_j and ajbha_jb_h, and lift the coefficients to CC. This gives δi∈aG\delta_i \in\mathfrak{a}^G lifting δˉi\bar{\delta}_i. We set X′⊂YX' \subset Y to be the subscheme defined by ti′=ti+δit'_i=t_i+\delta_i. The special fiber Xs′∩UsX'_s\cap U_s is the pullback of the common zero scheme just constructed along the finite étale map Us→VsU_s\to V_s, so it is smooth over kk.

Since δi∈H∩Ig(D)\delta_i\in H\cap I_g(D) for every g≠1g\ne1, one has

D/(t1′,…,tr′,Ig(D))=D/(t1,…,tr,Ig(D)),D/(t'_1,\ldots,t'_r,I_g(D))=D/(t_1,\ldots,t_r,I_g(D)),
D/(t1′,…,tr′,H)=D/(t1,…,tr,H).(4)D/(t'_1,\ldots,t'_r,H)=D/(t_1,\ldots,t_r,H). \tag*{(4)}

These equalities show that X′∖{x}X'\setminus\{x\} lies in UU. Thus X′X' is smooth away from xx.

Finally we show that Crit⁡(X′/C)\operatorname{Crit}(X'/C) is supported at xx. The images of the tit_i in mY,x/mY,x2\mathfrak{m}_{Y,x}/\mathfrak{m}_{Y,x}^2 are linearly independent, since OX,x\mathcal{O}_{X,x} is regular of codimension rr. As δi∈nN\delta_i\in\mathfrak{n}^N, the ti′t'_i are also linearly independent, so OX′,x\mathcal{O}_{X',x} is regular. Moreover the choice of NN shows that π\pi is nonzero in OX′,x\mathcal{O}_{X',x}. Thus X′X' is flat over CC at xx. After an invariant affine shrinking, X′X' is regular and flat. Its generic fiber is smooth in characteristic zero, and its special fiber is smooth away from xx. Together with (4.2.1) and the equality of the order-NN truncations, this proves the lemma. □

The deformation proposition

Proposition 4.3. Let RR be an integral smooth finite-type CC-algebra, and let AA be a regular finite free RR-algebra with RR-linear GG-action. Suppose that the fiber of X=Spec⁡AX=\operatorname{Spec} A over a closed kk-point bb of Spec⁡R\operatorname{Spec} R is supported at a single point xx, and that for 1≠g∈G1\ne g\in G, the fixed scheme of gg is supported at xx. There is a regular flat affine finite-type GG-scheme X′/CX'/C, with a closed kk-point x′x', such that

Crit⁡(X′/C), Fix⁡X′(g) (g≠1)are supported at x′,\operatorname{Crit}(X'/C),\ \operatorname{Fix}_{X'}(g)\ (g\ne1)\quad\text{are supported at }x',

and

OX′,x′/Ig(X′)x′≃OX,x/Ig(X)x(g≠1).\mathcal{O}_{X',x'}/I_g(X')_{x'}\simeq\mathcal{O}_{X,x}/I_g(X)_x\qquad(g\ne1).

For any N≥1N\ge1, we may also arrange that OX′,x′/mx′N≃OX,x/mxN\mathcal{O}_{X',x'}/\mathfrak{m}_{x'}^N\simeq\mathcal{O}_{X,x}/\mathfrak{m}_x^N.

Proof. We will embed XX, near xx, in a regular flat CC-scheme YY by invariant equations, with Y/CY/C smooth along X∖{x}X\setminus\{x\}. Lemma 4.2 will then give the required deformation.

After shrinking Spec⁡R\operatorname{Spec} R about bb, choose an RR-basis of AA beginning with 11. The point xx is fixed by GG, since it is the unique point of its fiber, and its residue field is kk, since kk is algebraically closed. The multiplication and GG-action on AA, in the chosen basis, define a section

z:Spec⁡R⟶ZR=Zn,G×ZSpec⁡R,n=rank⁡RA.z:\operatorname{Spec} R\longrightarrow Z_R=Z_{n,G}\times_Z\operatorname{Spec} R,\qquad n=\operatorname{rank}_R A.

Write WR=Wn,G×ZSpec⁡RW_R=W_{n,G}\times_Z\operatorname{Spec} R for the universal family over ZRZ_R. Its pullback along zz is XX.

The finite flat morphism of regular schemes X→Spec⁡RX\to\operatorname{Spec} R is a local complete intersection, and away from bb the GG-action on XX is free. Lemma 4.1 therefore shows that ZR/RZ_R/R is smooth along z(Spec⁡R∖{b})z(\operatorname{Spec} R\setminus\{b\}), and WR/RW_R/R is smooth above this locus.

Apply Lemma 3.1 at bb. It gives a smooth affine RR-scheme TT, a section e:Spec⁡R→Te:\operatorname{Spec} R\to T, and an RR-morphism f:T→ZRf:T\to Z_R such that f∘e=zf\circ e=z and ff is smooth along e(Spec⁡R∖{b})e(\operatorname{Spec} R\setminus\{b\}). Put Y=WR×ZRTY=W_R\times_{Z_R}T. Then Y×T,eSpec⁡R=XY\times_{T,e}\operatorname{Spec} R=X. The projection Y→TY\to T is finite flat, and the universal GG-action on YY is trivial on TT. Thus YY is flat over CC. At a point of X∖{x}X\setminus\{x\}, the projection Y→WRY\to W_R is smooth, being the base change of ff, and WR/RW_R/R is smooth at its image. Since R/CR/C is smooth, Y/CY/C is smooth along X∖{x}X\setminus\{x\}.

We next choose the invariant equations defining XX in YY. Let rr be the relative dimension of T/RT/R at e(b)e(b). As TT is smooth over RR, after shrinking Spec⁡R\operatorname{Spec} R about bb and TT about e(b)e(b), there is an étale map T→ARrT \to\mathbb{A}^{r}_{R} for which e(Spec⁡R)e(\operatorname{Spec} R) is the inverse image of the zero section. We denote by t1,…,trt_{1},\ldots,t_{r} the pullbacks to TT of the coordinate functions. Pulling back to YY gives

X=V(t1,…,tr)⊂Y.X = V(t_{1},\ldots,t_{r}) \subset Y.

The tit_i form a regular sequence on TT, and hence on YY by flatness. Their pullbacks are invariant because GG acts trivially on TT. The quotient OY,x/(t1,…,tr)=OX,x\mathcal{O}_{Y,x}/(t_{1},\ldots,t_{r}) = \mathcal{O}_{X,x} is regular, so OY,x\mathcal{O}_{Y,x} is regular as well. Shrink YY to an invariant regular affine neighborhood of xx, and replace XX by its intersection with this neighborhood.

We have now obtained all the hypotheses of Lemma 4.2: XX is regular of codimension rr in YY and flat over CC, while X∩Crit⁡(Y/C)X \cap\operatorname{Crit}(Y/C) and the nonidentity fixed schemes of XX are supported at xx. Applying that lemma gives our result.

The trace formula and the character

Below we consider a pointed CC-scheme (X,x)(X,x) with GG-action, satisfying the conclusion of Proposition 4.3. That is, XX is a regular, CC-flat, affine GG-scheme, which is smooth outside the kk-point xx, and such that Fix⁡X(g)\operatorname{Fix}_{X}(g) is supported on xx for g≠1g \ne1. Below we use only that the singular locus of X/CX/C is proper over CC.

We first express the fixed-point multiplicities as intersection numbers. The results of Beraldo and Pippi then express these numbers as traces on two fixed virtual representations. An elementary character argument will give the prescribed value at 1 and prove positivity.

The local intersection

We refer the reader to Bloch’s conductor conjecture [7] and the localized intersection product of Kato and Takeshi Saito [14 Theorem 3.2.1 and Definition 3.2.2] for background.

Put P=X×CXP = X \times_{C} X, let Δ⊂P\Delta\subset P be the diagonal, and let Γg⊂P\Gamma_{g} \subset P be the image of (1,g):X→P(1,g) : X \to P. For g≠1g \ne1 and integers i≥0i \ge0, the sheaves

Ti=Tor⁡iOP(OΔ,OΓg)\mathcal{T}_{i} = \operatorname{Tor}^{\mathcal{O}_{P}}_{i}(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_{g}})

are supported at (x,x)(x,x), since Δ∩Γg≃Fix⁡X(g)\Delta\cap\Gamma_{g} \simeq\operatorname{Fix}_{X}(g), and their stalks there have finite length over CC. The stable intersection number is defined as

[[OΔ,OΓg]]C=length⁡C(T2q)−length⁡C(T2q+1),q≫0.(5)[[\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_{g}}]]_{C} = \operatorname{length}_{C}(\mathcal{T}_{2q}) - \operatorname{length}_{C}(\mathcal{T}_{2q+1}), \qquad q \gg0. \tag*{(5)}

The result below calculates the two lengths on the right-hand side and shows that the left-hand side is the multiplicity of the fixed point of gg.

Lemma 5.1. For g≠1g \ne1 and n≥0n \ge0, one has

Tn≃{OX,x/Ig(X)x,n even,0,n odd.\mathcal{T}_{n} \simeq \begin{cases} \mathcal{O}_{X,x}/I_{g}(X)_{x}, & n\ \text{even},\\ 0, & n\ \text{odd}. \end{cases}

In particular,

[[OΔ,OΓg]]C=length⁡OX,x(OX,x/Ig(X)x).[[\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_{g}}]]_{C} = \operatorname{length}_{\mathcal{O}_{X,x}}(\mathcal{O}_{X,x}/I_{g}(X)_{x}).

Proof. The equality of intersection numbers follows from Kato and Takeshi Saito’s comparison theorem [14 Proposition 3.2.3], using their local hypersurface embedding [14 Lemma 5.1.1]. We give a direct local proof.

Fix g≠1g \ne1. We will compute the Tor groups Tn\mathcal{T}_n using a hypersurface presentation of the completed local ring. Put B=O^X,x\mathcal{B} = \widehat{\mathcal{O}}_{X,x} and d=dim⁡Bd = \dim\mathcal{B}. A uniformizer π\pi of C\mathbb{C} cannot be part of a regular system of parameters of B\mathcal{B}, as then Ig(B)I_g(\mathcal{B}) would be generated by d−1d-1 elements, contradicting that B/Ig(B)\mathcal{B}/I_g(\mathcal{B}) has finite length. Choose regular parameters x1,…,xdx_1,\ldots,x_d of B\mathcal{B}. The resulting surjection C[[x1,…,xd]]→B\mathbb{C}[[x_1,\ldots,x_d]] \to\mathcal{B} has a height-one prime kernel, which is principal since the source is regular. Thus

B=C[[x1,…,xd]]/(F0).\mathcal{B} = \mathbb{C}[[x_1,\ldots,x_d]]/(F_0).

To resolve the diagonal, use the completed local ring

D=O^P,(x,x)=B[[y1,…,yd]]/(F0(y)),\mathcal{D} = \widehat{\mathcal{O}}_{\mathcal{P},(x,x)} = \mathcal{B}[[y_1,\ldots,y_d]]/(F_0(y)),

where the xix_i and yiy_i are the parameters from the first and second factors. Put Q=B[[y1,…,yd]]Q = \mathcal{B}[[y_1,\ldots,y_d]] and vi=yi−xi∈Qv_i = y_i-x_i \in Q. Since F0(x)=0F_0(x)=0 in B\mathcal{B}, we may write F0(y)=∑iciviF_0(y)=\sum_i c_i v_i with ci∈Qc_i \in Q. We use the same letters for the images of the cic_i in D\mathcal{D}. We have D/(v1,…,vd)=B\mathcal{D}/(v_1,\ldots,v_d)=\mathcal{B}. Let KQK_Q denote the Koszul complex on the viv_i, and KDK_{\mathcal{D}} the Koszul complex on the images of the viv_i in D\mathcal{D}. Thus,

KD=⋀∙(De1⊕⋯⊕Ded),∣ei∣=1,dei=vi.K_{\mathcal{D}}=\bigwedge\nolimits^{\mathbin{\bullet}}(\mathcal{D}e_1\oplus\cdots\oplus\mathcal{D}e_d),\qquad|e_i|=1,\qquad de_i=v_i.

and similarly for KQK_Q. As v1,…,vdv_1,\ldots,v_d is a regular sequence in QQ, KQK_Q is a resolution of Q/(v1,…,vd)≃B≃D/(v1,…,vd)Q/(v_1,\ldots,v_d)\simeq\mathcal{B}\simeq\mathcal{D}/(v_1,\ldots,v_d).

Q/(v1,…,vd)≃B≃D/(v1,…,vd).Q/(v_1,\ldots,v_d)\simeq\mathcal{B}\simeq\mathcal{D}/(v_1,\ldots,v_d).

Since F0(y)F_0(y) is a nonzero divisor in QQ, the exact sequence of Koszul complexes obtained from

0⟶Q→F0(y)Q⟶D⟶00\longrightarrow Q \xrightarrow{F_0(y)} Q \longrightarrow\mathcal{D} \longrightarrow0

shows that H0(KD)=H1(KD)=BH_0(K_{\mathcal{D}})=H_1(K_{\mathcal{D}})=\mathcal{B}, with all other homology zero. The element z=∑iciei∈KQz=\sum_i c_i e_i\in K_Q maps to a cycle in KDK_{\mathcal{D}}, whose class generates the B\mathcal{B}-module H1(KD)H_1(K_{\mathcal{D}}): the connecting map sends this class to 1∈B1\in\mathcal{B}, since dz=F0(y)dz=F_0(y).

To construct a resolution of B\mathcal{B} by free D\mathcal{D}-modules, we use a construction of Tate [22]. Let ΓD(t)=⨁j≥0Dt[j]\Gamma_{\mathcal{D}}(t)=\bigoplus_{j\geq0}\mathcal{D}t^{[j]} be the divided power algebra, graded with t[j]t^{[j]} in degree 2j2j. Consider the complex

KD⊗DΓD(t),∣t∣=2,dt=∑iciei(6)K_{\mathcal{D}}\otimes_{\mathcal{D}}\Gamma_{\mathcal{D}}(t),\qquad|t|=2,\qquad dt=\sum_i c_i e_i \tag*{(6)}

Its term in homological degree ii is the degree-ii homogeneous component of the tensor product. The differential extends the Koszul differential by the graded Leibniz rule and

d(t[j])=(∑iciei)t[j−1](j≥1).d(t^{[j]})=\left(\sum_i c_i e_i\right)t^{[j-1]}\qquad(j\geq1).

This differential squares to zero because zz is a Koszul cycle and z∧z=0z\wedge z=0. We verify that this complex resolves B\mathcal{B} over D\mathcal{D}. Filter (5.1.1) by divided power degree, and consider the associated spectral sequence. The differential on the associated graded is the Koszul differential, since the additional term lowers divided power degree by one. After taking Koszul homology, the differential on the E1E_1-page is induced by dt=zdt=z and gives an isomorphism

H0(KD)t[j]≃Bt[j]→ ∼ B⋅[z]t[j−1]≃H1(KD)t[j−1](j≥1).H_0(K_{\mathcal{D}})t^{[j]}\simeq\mathcal{B}t^{[j]}\xrightarrow{\ \sim\ }\mathcal{B}\cdot[z]t^{[j-1]}\simeq H_1(K_{\mathcal{D}})t^{[j-1]}\qquad(j\geq1).

Thus, on the E2E_2-page only Bt[0]\mathcal{B}t^{[0]} in degree zero remains. The filtration is finite in each total degree, so the spectral sequence converges and shows that (5.1.1) resolves B\mathcal{B} as a D\mathcal{D}-module.

We now tensor this resolution with D/(y1−g(x1),…,yd−g(xd))≃B\mathcal{D}/(y_1-g(x_1),\ldots,y_d-g(x_d))\simeq\mathcal{B} to compute the Tn\mathcal{T}_n. We obtain the complex KB⊗BΓB(t)K_{\mathcal{B}}\otimes_{\mathcal{B}}\Gamma_{\mathcal{B}}(t), where KBK_{\mathcal{B}} is the Koszul complex on g(x1)−x1,…,g(xd)−xd∈Bg(x_1)-x_1,\ldots,g(x_d)-x_d\in\mathcal{B}. Since B/Ig(B)\mathcal{B}/I_g(\mathcal{B}) has finite length, they form a system of parameters of the regular ring B\mathcal{B}, and hence a regular sequence. Thus KBK_{\mathcal{B}} is a resolution of B/Ig(B)\mathcal{B}/I_g(\mathcal{B}). Using the filtration by divided power degree, we again obtain a spectral sequence whose E1E_1-page is

(B/Ig(B))⊗BΓB(t).(\mathcal{B}/I_g(\mathcal{B}))\otimes_{\mathcal{B}}\Gamma_{\mathcal{B}}(t).

a graded module concentrated in even total degrees. Thus the spectral sequence degenerates at the E1E_1-page, the even Tor groups are B/Ig(B)\mathcal{B}/I_g(\mathcal{B}) and the odd Tor groups are zero.

The inputs from Beraldo and Pippi

Blanc, Robalo, Toën and Vezzosi identified the realization of a singularity category with inertia-invariant vanishing cycles in the proper flat case [6 Theorem 4.39]. Toën and Vezzosi used this comparison to obtain a categorical form of Bloch’s formula, which relates the localized self-intersection of the diagonal to the Euler characteristic of vanishing cycles together with its Swan conductor [23 Theorem 5.2.2 and Remark 5.2.3]. We need a version of this result which computes the intersection of the diagonal with the graph of an element g∈Gg\in G, for which we use the results of Beraldo and Pippi [4]. We recall some results from loc. cit., adapted to our setting, and then use this to prove Lemma 5.2 below.

Fix a prime ℓ≠char⁡k\ell\ne\operatorname{char} k. Put K=Frac⁡(C)K=\operatorname{Frac}(C), choose an algebraic closure K‾\overline{K} of KK, and put IK=Gal⁡(K‾/K)I_K=\operatorname{Gal}(\overline{K}/K), which is the inertia group since kk is algebraically closed. We use Deligne’s normalization of the vanishing-cycle complex [8 Exposé XIII, §2.1]

Qℓ,Xs⟶RΨX/CQℓ⟶RΦX/CQℓ⟶Qℓ,Xs[1].\mathbb{Q}_{\ell,X_s}\longrightarrow R\Psi_{X/C}\mathbb{Q}_{\ell}\longrightarrow R\Phi_{X/C}\mathbb{Q}_{\ell}\longrightarrow\mathbb{Q}_{\ell,X_s}[1].

Let β∈Qℓ(1)\beta\in\mathbb{Q}_{\ell}(1) be a basis vector, and set

ν0=RΓ(Xs,RΦX/CQℓ),Λ=Qℓ[β,β−1]=⨁n∈ZQℓ(n)[2n],ν=ν0⊗QℓΛ,\nu_0=R\Gamma(X_s,R\Phi_{X/C}\mathbb{Q}_{\ell}),\qquad\Lambda=\mathbb{Q}_{\ell}[\beta,\beta^{-1}]=\bigoplus_{n\in\mathbb{Z}}\mathbb{Q}_{\ell}(n)[2n],\qquad\nu=\nu_0\otimes_{\mathbb{Q}_{\ell}}\Lambda,

so β\beta has degree −2-2.

Let F=RΓ(IK,Λ)F=R\Gamma(I_K,\Lambda), with its natural augmentation F⟶ΛF\longrightarrow\Lambda, so that the cohomology of FF is two-periodic up to Tate twist. The vanishing-cycle sheaves are constructible and supported on the proper locus Crit⁡(X/C)\operatorname{Crit}(X/C), so ν0\nu_0 is bounded with finite-dimensional cohomology. Its GG-action is g↦(g−1)∗g\mapsto(g^{-1})^* and commutes with inertia. All tensor products below are derived, and superscripts IKI_K and ILI_L denote derived continuous invariants.

For a nontrivial finite Galois extension L/KL/K in K‾\overline{K}, put IL=Gal⁡(K‾/L)I_L=\operatorname{Gal}(\overline{K}/L), H=IK/ILH=I_K/I_L, and

R0=F,R1=F[H](1−e0),e0=1∣H∣∑h∈Hh,R_0=F,\qquad R_1=F[H](1-e_0),\qquad e_0=\frac{1}{|H|}\sum_{h\in H}h,
M0=νIK[−1],M1=cofib⁡(νIK⟶νIL).(7)M_0=\nu^{I_K}[-1],\qquad M_1=\operatorname{cofib}(\nu^{I_K}\longrightarrow\nu^{I_L}). \tag*{(7)}

Restriction also defines the FF-action on νIL\nu^{I_L}. The group HH acts on νIL\nu^{I_L}, and we may consider the projections corresponding to the idempotents e0e_0 and 1−e01-e_0.

νIK≃e0νIL,M1≃(1−e0)νIL.\nu^{I_K}\simeq e_0\nu^{I_L},\qquad M_1\simeq(1-e_0)\nu^{I_L}.

Thus M1M_1 is a module over the reduced group algebra R1R_1. The GG-actions on M0M_0 and M1M_1 commute with these coefficient actions. Both modules, and the extension L/KL/K, will be chosen once for the whole group GG.

Choose L/KL/K sufficiently large that ILI_L acts unipotently on the cohomology sheaves of RΦX/CQℓ\mathcal{R}\Phi_{X/C}\mathbb{Q}_{\ell}, as in the proof of [5 Theorem 8.4.3]. Beraldo and Pippi identify the coefficient algebras and modules in (5.2), and prove that MiM_i is perfect over RiR_i [4 Proposition 5.4.1 and Propositions 5.5.2–5.5.5]. Both modules are also perfect over FF, since R1R_1 is a direct summand of the finite free FF-module F[H]F[H].

Recall also the duality in these results. Verdier duality for inertia-invariant vanishing cycles identifies the right duals of M0M_0 and M1M_1 with M0M_0 and M1(1)M_1(1), respectively [4 Propositions 5.4.1 and 5.5.5]. For M1M_1, the comparison of [4 Proposition 5.5.7 and Remark 5.6.4] identifies M1⊗R1M1(1)M_1 \otimes_{R_1} M_1(1) with M1⊗R1M1M_1 \otimes_{R_1} M_1. They write the diagonal evaluation as

ev⁡iOΔ:Mi⊗FMi⟶Ri,\operatorname{ev}^{\mathcal{O}_{\Delta}}_i:M_i\otimes_F M_i\longrightarrow R_i,

with the first factor carrying the right RiR_i-module structure. Together with the diagonal coevaluation, this makes the first copy of MiM_i a dual of the second [4 Corollaries 5.7.10 and 5.7.13]. Under the resulting identification of the first factor with RHom⁡Ri(Mi,Ri)R\operatorname{Hom}_{R_i}(M_i,R_i), evaluation is the pairing d⊗m↦d(m)d\otimes m\mapsto d(m).

Let CLC_L be the integral closure of CC in LL, and let aH=aCL,Ha_H=a_{C_L,H} be the classical Artin character, defined by (1.1). For a finite group HH, write K0(Qℓ[H])K_0(\mathbb{Q}_{\ell}[H]) for its representation ring over Qℓ\mathbb{Q}_{\ell}.

Lemma 5.2 (The graph trace formula). Under the hypotheses on X/CX/C and GG at the beginning of this section, there exist a finite Galois extension L/KL/K, with group HH, and virtual representations

V0∈K0(Qℓ[G]),V1∈K0(Qℓ[G×H]),V_0\in K_0(\mathbb{Q}_{\ell}[G]),\qquad V_1\in K_0(\mathbb{Q}_{\ell}[G\times H]),

such that, for every g≠1g\ne1,

[[OΔ,OΓg]]C=Tr⁡(g,V0)−1∣H∣∑h∈HaH(h)Tr⁡((g,h−1),V1).(8)[[\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g}]]_C =\operatorname{Tr}(g,V_0)-\frac{1}{|H|}\sum_{h\in H}a_H(h)\operatorname{Tr}((g,h^{-1}),V_1). \tag*{(8)}

Proof. Use the extension L/KL/K chosen above, and put

Ui=Mi⊗FΛ,Vi=[H0(Ui)]−[H1(Ui)](i=0,1).U_i=M_i\otimes_F\Lambda,\qquad V_i=[H^0(U_i)]-[H^1(U_i)]\qquad(i=0,1).

As the MiM_i are perfect over FF the UiU_i are perfect over Λ\Lambda, so the two cohomology groups are finite-dimensional. They carry the GG-action and, for i=1i=1, the commuting HH-action. Thus V0V_0 and V1V_1 belong to the stated representation rings.

Beraldo and Pippi observe in [4 Remark 1.6.4] that their argument extends to endomorphisms. We use the explicit twisted conductor formula of [5 §8.5.6], applied to OΔ\mathcal{O}_{\Delta} and OΓg\mathcal{O}_{\Gamma_g}. Its geometric term is the stable even Tor degree minus the stable odd Tor degree [5 §8.5.2], which is [[OΔ,OΓg]]C[[\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g}]]_C since the Tor sheaves are supported at the kk-point (x,x)(x,x). Using [5 Lemma 8.4.4(1)–(2)], the formula gives

[[OΔ,OΓg]]C=Tr⁡F(OΔ,OΓg)(M0)−αH(Tr⁡R1(OΔ,OΓg)(M1)).[[\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g}]]_C =\operatorname{Tr}^{(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g})}_{F}(M_0) -\alpha_H\left(\operatorname{Tr}^{(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g})}_{R_1}(M_1)\right).

Here Tr⁡Ri(OΔ,OΓg)(Mi)\operatorname{Tr}^{(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g})}_{R_i}(M_i) is the twisted trace defined in [5 §8.5.5]. It takes values in HH⁡0(Ri/F)=H0(HH⁡(Ri/F))\operatorname{HH}_0(R_i/F)=H^0(\operatorname{HH}(R_i/F)), where the relative Hochschild homology complex is

HH⁡(Ri/F)=Ri⊗RieRi,Rie=Ri⊗FRiop.\operatorname{HH}(R_i/F)=R_i\otimes_{R_i^e}R_i,\qquad R_i^e=R_i\otimes_F R_i^{\mathrm{op}}.

For the graph Γg\Gamma_g, Beraldo and Pippi’s construction gives the twisted trace as the degree-zero class of the composite

F→(1⊗g)∘coev⁡i△Mi⊗RiMi→∼(Mi⊗FMi)⊗RieRi→ev⁡i△⊗1Ri⊗RieRi=HH⁡(Ri/F).F \xrightarrow{(1\otimes g)\circ\operatorname{coev}_{i}^{\triangle}} M_i\otimes_{R_i}M_i \xrightarrow{\sim} (M_i\otimes_F M_i)\otimes_{R_i^e}R_i \xrightarrow{\operatorname{ev}_{i}^{\triangle}\otimes1} R_i\otimes_{R_i^e}R_i=\operatorname{HH}(R_i/F).

The first arrow is defined by composing the diagonal coevaluation with 1⊗g1\otimes g, where gg acts as g∗=(g−1)∗g_*=(g^{-1})_*. The last arrow is induced by the diagonal evaluation recalled above, as in [5 §§8.5.3–8.5.4]. Here we use OΓg=(1,g)∗OΔ\mathcal{O}_{\Gamma_g}=(1,g)_*\mathcal{O}_{\Delta} and functoriality of the construction of loc. cit. with respect to pushforward on the second factor [4 Theorem 3.2.8, Corollary 4.3.6 and Proposition 5.6.3]. The diagonal evaluation and coevaluation satisfy the duality identities [4 Corollaries 5.7.10 and 5.7.13].

The Artin functional

αH:HH⁡0(R1/F)⟶HH⁡0(F/F)≃H0(F)\alpha_H:\operatorname{HH}_0(R_1/F)\longrightarrow\operatorname{HH}_0(F/F)\simeq H^0(F)

is the H0(F)H^0(F)-linear map which sends the class of h(1−e0)h(1-e_0) to aH(h)a_H(h). We denote this class by [h][h].

For M0M_0, the displayed composite gives the scalar FF-linear trace:

Tr⁡F(OΔ,OΓg)(M0)=Tr⁡F(g,M0).\operatorname{Tr}_{F}^{(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g})}(M_0)=\operatorname{Tr}_F(g,M_0).

For M1M_1, Hattori’s group-algebra trace calculation [12 §1.2, Example 7], in the form of [4 Lemma A.1.4 and Corollary A.2.2], applies directly to the diagonal evaluation and coevaluation. With gg inserted as above, it gives

Tr⁡R1(OΔ,OΓg)(M1)=1∣H∣∑h∈HTr⁡F(gh−1,M1)[h].\operatorname{Tr}_{R_1}^{(\mathcal{O}_{\Delta},\mathcal{O}_{\Gamma_g})}(M_1)=\frac{1}{|H|}\sum_{h\in H}\operatorname{Tr}_F(gh^{-1},M_1)[h].

Here the traces on the right are FF-linear traces, taking values in H0(F)H^0(F). The group-algebra formulas in [4] are stated for the identity; inserting the commuting endomorphism gg in their evaluation–coevaluation calculation gives the factors gh−1gh^{-1} above. The calculation applies to any evaluation and coevaluation satisfying the duality identities.

The augmentation identifies H0(F)H^0(F) with H0(Λ)=QℓH^0(\Lambda)=\mathbb{Q}_\ell: inertia has ℓ\ell-cohomological dimension one, and its degree-one cohomology contributes only odd degrees after periodization. Extension of scalars preserves traces, so

Tr⁡F(g,Mi)=Tr⁡(g,H0(Ui))−Tr⁡(g,H1(Ui)).\operatorname{Tr}_F(g,M_i)=\operatorname{Tr}(g,H^0(U_i))-\operatorname{Tr}(g,H^1(U_i)).

The same identity holds for gh−1gh^{-1} on M1M_1. Substituting in the preceding trace formulas proves (5.2.1). The extension and the two virtual representations were chosen independently of gg. □\square

The virtual representation

The weighted sum in Lemma 5.2 is itself a virtual character. We record the elementary identity that makes this integrality explicit.

Lemma 5.3. Let HH be a finite group, let EE be a finite-dimensional Qℓ[H]\mathbb{Q}_\ell[H]-module with character aa, and let U∈K0(Qℓ[G×H])U\in K_0(\mathbb{Q}_\ell[G\times H]). Then Hom⁡H(E,U)∈K0(Qℓ[G])\operatorname{Hom}_H(E,U)\in K_0(\mathbb{Q}_\ell[G]) is defined by additivity, and

Tr⁡(g,Hom⁡H(E,U))=1∣H∣∑h∈Ha(h)Tr⁡((g,h−1),U)(g∈G).(9)\operatorname{Tr}\bigl(g,\operatorname{Hom}_H(E,U)\bigr)=\frac{1}{|H|}\sum_{h\in H}a(h)\operatorname{Tr}\bigl((g,h^{-1}),U\bigr)\qquad(g\in G). \tag*{(9)}

Proof. It suffices to treat an actual representation UU. On E∨⊗QℓUE^\vee\otimes\mathbb{Q}_\ell U, let HH act diagonally and let GG act through UU. The averaging projector ∣H∣−1∑h∈Hh|H|^{-1}\sum_{h\in H}h on E∨⊗QℓUE^\vee\otimes\mathbb{Q}_\ell U has image Hom⁡H(E,U)\operatorname{Hom}_H(E,U) and commutes with gg. The trace of gg on Hom⁡H(E,U)\operatorname{Hom}_H(E,U) is therefore the trace of the composite of gg with the projector on E∨⊗QℓUE^\vee\otimes\mathbb{Q}_\ell U, namely

1∣H∣∑h∈Ha(h−1)Tr⁡((g,h),U).\frac{1}{|H|}\sum_{h\in H}a(h^{-1})\operatorname{Tr}((g,h),U).

Replace hh by h−1h^{-1}. Exactness of invariants in characteristic zero extends the identity to virtual representations.

Proposition 5.4. There exists V∈K0(Qℓ[G])V \in K_0(\mathbb{Q}_\ell[G]) such that

Tr⁡(g,V)=−length⁡OX,x(OX,x/Ig(X)x)(g≠1).\operatorname{Tr}(g,V)=-\operatorname{length}_{\mathcal{O}_{X,x}}(\mathcal{O}_{X,x}/I_g(X)_x)\qquad(g\ne1).

Proof. Choose L/KL/K, HH, V0V_0 and V1V_1 as in Lemma 5.2. By Serre’s theorem [19 §19.2, Theorem 44(i)], aHa_H is the character of a Qℓ\mathbb{Q}_\ell-representation Art⁡H\operatorname{Art}_H. Applying Lemma 5.3 to the weighted sum in (5.2.1) gives

[[OΔ,OΓg]]C=Tr⁡(g,V0)−Tr⁡(g,Hom⁡H(Art⁡H,V1))(g≠1).(10)[[\mathcal{O}_\Delta,\mathcal{O}_{\Gamma_g}]]_C=\operatorname{Tr}(g,V_0)-\operatorname{Tr}(g,\operatorname{Hom}_H(\operatorname{Art}_H,V_1))\qquad(g\ne1). \tag*{(10)}

Since Hom⁡H(Art⁡H,−)\operatorname{Hom}_H(\operatorname{Art}_H,-) is exact over Qℓ\mathbb{Q}_\ell, the class

V=−V0+Hom⁡H(Art⁡H,V1)∈K0(Qℓ[G])V=-V_0+\operatorname{Hom}_H(\operatorname{Art}_H,V_1)\in K_0(\mathbb{Q}_\ell[G])

is an integral virtual representation. Its trace is the negative of the intersection number in (5.4.1). Lemma 5.1 identifies that intersection number with the fixed-point length, proving the proposition.

Positivity and conclusion

As in Serre’s treatment of the classical case [19 §19.1], positivity follows from integrality and character orthogonality.

Lemma 5.5. Suppose that V∈K0(Qℓ[G])V\in K_0(\mathbb{Q}_\ell[G]) satisfies Tr⁡(g,V)=−i(g)\operatorname{Tr}(g,V)=-i(g) for g≠1g\ne1, where ii is a nonnegative integer-valued central function on G∖{1}G\setminus\{1\}. Then

W=V−dim⁡(VG)Qℓ[G]W=V-\dim(V^G)\mathbb{Q}_\ell[G]

is the class of a representation, and its character is −i(g)-i(g) for g≠1g\ne1, and ∑g≠1i(g)\sum_{g\ne1}i(g) at 11.

Proof. The regular character vanishes off 11, and dim⁡(WG)=0\dim(W^G)=0. Hence the inner product of WW and the trivial character is 00, so the character of WW at 11 is ∑g≠1i(g)\sum_{g\ne1}i(g). Viewing character values in C\mathbb{C}, the multiplicity of an irreducible character χ\chi in WW is

1∣G∣∑g≠1i(g)(χ(1)−Re⁡χ(g))≥0,\frac{1}{|G|}\sum_{g\ne1}i(g)(\chi(1)-\operatorname{Re}\chi(g))\ge0,

where we use i(g)=i(g−1)∈Zi(g)=i(g^{-1})\in\mathbb{Z} and Re⁡χ(g)≤χ(1)\operatorname{Re}\chi(g)\le\chi(1). These multiplicities are integers since WW is a virtual Qℓ\mathbb{Q}_\ell-representation. Serre’s Proposition 33 [19 §12.1] then shows that WW is the class of an actual Qℓ\mathbb{Q}_\ell-representation.

Proof of Theorem 1.1. If the residue characteristic of AA is zero, then it is easy to see that iA(g)=1i_A(g)=1 for every g≠1g\ne1, and aA,Ga_{A,G} is the augmentation character. If AA has characteristic p>0p>0, apply Abe’s theorem [2].

Suppose AA has mixed characteristic. By Proposition 2.2, we may assume A=OX,xA=\mathcal{O}_{X,x} for a finite type affine GG-scheme XX over a dvr CC, with uniformizer pp, and x∈Xx\in X a closed point with residue field k=C/pCk=C/pC. Then by Proposition 4.3, we may assume that Crit⁡(X/C)\operatorname{Crit}(X/C) is supported at xx. We may now apply Proposition 5.4 to (X,x)(X,x) to obtain V∈K0(Qℓ[G])V\in K_0(\mathbb{Q}_{\ell}[G]) with

Tr⁡(g,V)=−iA(g)(g≠1).\operatorname{Tr}(g,V)=-i_A(g) \qquad(g\ne1).

The function iAi_A is central and nonnegative, and Ig(A)=Ig−1(A)I_g(A)=I_{g^{-1}}(A) gives iA(g)=iA(g−1)i_A(g)=i_A(g^{-1}). Lemma 5.5 therefore gives a representation with character (1), as required.

References

  1. [1]A. Abbes, Cycles on arithmetic surfaces, Compos. Math. 122 (2000), 23–111; published article.DOI
  2. [2]T. Abe, On the Serre conjecture for Artin characters in the geometric case, Invent. Math. 241 (2025), 605–622; arXiv:2405.19601.arxiv.org/abs/2405.19601
  3. [3]M. Artin, Algebraic approximation of structures over complete local rings, Publ. Math. Inst. Hautes Études Sci. 36 (1969), 23–58; published article.DOI
  4. [4]D. Beraldo and M. Pippi, Proof of the Deligne–Milnor conjecture, arXiv:2410.02327v1 (2024).arxiv.org/abs/2410.02327
  5. [5]D. Beraldo and M. Pippi, Categorification of the localized intersection product and Bloch conductor formula, arXiv:2511.10527v1 (2025).arxiv.org/abs/2511.10527
  6. [6]A. Blanc, M. Robalo, B. Toën and G. Vezzosi, Motivic realizations of singularity categories and vanishing cycles, J. Éc. polytech. Math. 5 (2018), 651–747; published article.arxiv.org/abs/1607.03012
  7. [7]S. Bloch, Cycles on arithmetic schemes and Euler characteristics of curves, in Algebraic geometry, Bowdoin, 1985, Proc. Sympos. Pure Math. 46, Part 2, Amer. Math. Soc., 1987, 421–450; published article.DOI
  8. [8]P. Deligne and N. Katz, Groupes de monodromie en géométrie algébrique, II (SGA 7 II), Lecture Notes in Math. 340, Springer, 1973; Exposé XIII by P. Deligne and Exposé XVII by N. M. Katz; published volume.
  9. [9]P. M. Eakin, Jr., The converse to a well known theorem on Noetherian rings, Math. Ann. 177 (1968), 278–282; published article.DOI
  10. [10]R. Elkik, Solutions d’équations à coefficients dans un anneau hensélien, Ann. Sci. École Norm. Sup. (4) 6 (1973), 553–603; published article.
  11. [11]A. Grothendieck et al., Revêtements étales et groupe fondamental (SGA 1), Lecture Notes in Math. 224, Springer, 1971; revised edition.
  12. [12]A. Hattori, Rank element of a projective module, Nagoya Math. J. 25 (1965), 113–120; published article.DOI
  13. [13]K. Kato, S. Saito and T. Saito, Artin characters for algebraic surfaces, Amer. J. Math. 110 (1988), 49–75; published article.DOI
  14. [14]K. Kato and T. Saito, On the conductor formula of Bloch, Publ. Math. Inst. Hautes Études Sci. 100 (2004), 5–151; published article.DOI
  15. [15]K. Kato and T. Saito, Ramification theory for varieties over a local field, Publ. Math. Inst. Hautes Études Sci. 117 (2013), 1–178; published article.
  16. [16]M. Kisin, Endomorphisms of logarithmic schemes, Rend. Sem. Mat. Univ. Padova 109 (2003), 247–281; published article.
  17. [17]B. Poonen, The moduli space of commutative algebras of finite rank, J. Eur. Math. Soc. 10 (2008), 817–836; author’s version.
  18. [18]J.-P. Serre, Sur la rationalité des représentations d’Artin, Ann. of Math. (2) 72 (1960), 405–420; published article.
  19. [19]J.-P. Serre, Linear representations of finite groups, Graduate Texts in Math. 42, Springer, 1977; published volume.DOI
  20. [20]M. Spivakovsky, A new proof of D. Popescu’s theorem on smoothing of ring homomorphisms, J. Amer. Math. Soc. 12 (1999), 381–444; published article.DOI
  21. [21]The Stacks Project, §49.11 and Lemma 66.14.2.
  22. [22]J. Tate, Homology of Noetherian rings and local rings, Illinois J. Math. 1 (1957), 14–27; published article.DOI
  23. [23]B. Toën and G. Vezzosi, Trace and Künneth formulas for singularity categories and applications, Compos. Math. 158 (2022), 483–528; arXiv:1710.05902.

Paper details

vlnc:W_kh32hhy7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 16

downloads · 2

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

GPT Astra

View statement
Looking for related papers…
Contents