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

Euler factors of equivariant LL–functions
of Drinfeld modules and beyond

Cristian D. Popescu  and  Nandagopal Ramachandran Dept. of Mathematics, University of California at San Diego, San Diego, CA 92093, USA [email protected] [email protected]
Abstract.

In [2], the first author and his collaborators proved an equivariant Tamagawa number formula for the special value at s=0s=0 of a Goss–type LL–function, equivariant with respect to a Galois group GG, and associated to a Drinfeld module defined on 𝔽q[t]\mathbb{F}_{q}[t] and over a finite, integral extension of 𝔽q[t]\mathbb{F}_{q}[t]. The formula in question was proved provided that the values at 0 of the Euler factors of the equivariant LL–function in question satisfy certain identities involving Fitting ideals of certain GG–cohomologically trivial, finite 𝔽q[t][G]\mathbb{F}_{q}[t][G]–modules associated to the Drinfeld module. In [2], we prove these identities in the particular case of the Carlitz module. In this paper, we develop general techniques and prove the identities in question for arbitrary Drinfeld modules. Further, we indicate how these techniques can be extended to the more general case of higher dimensional abelian tt–modules, which is relevant in the context of the proof of the equivariant Tamagawa number formula for abelian tt–modules given by N. Green and the first author in [4]. This paper is based on a lecture given by the first author at ICMAT Madrid in May 2023 and builds upon results obtained by the second author in his PhD thesis [12].

Key words and phrases:
Drinfeld modules, tt–motives, local shtukas, étale cohomology, crystalline cohomology, equivariant motivic LL–functions, equivariant Tamagawa number formula, Euler factors
2010 Mathematics Subject Classification:
11G09, 11M38, 11F80

1. Introduction

Let E/F/kE/F/k be a tower of finite field extensions, where k:=𝔽q(t)k:=\mathbb{F}_{q}(t), qq is a power of a prime pp, and E/FE/F is Galois, of abelian Galois group GG. Further, let 𝒪F\mathcal{O}_{F} and 𝒪K\mathcal{O}_{K} denote the integral closure of A:=𝔽q[t]A:=\mathbb{F}_{q}[t] in FF and KK, respectively, and consider a Drinfeld module EE defined on A:=𝔽q[t]A:=\mathbb{F}_{q}[t] and with coefficients in 𝒪F\mathcal{O}_{F}.

In [2], the authors associated to the set of data (K/F/k,E)(K/F/k,E) a GG–equivariant Goss–type LL–function

ΘK/FE:𝕊[G],\Theta^{E}_{K/F}:\mathbb{S}_{\infty}\to\mathbb{C}_{\infty}[G],

where 𝕊\mathbb{S}_{\infty} is a certain Goss space of mixed characteristic containing naturally a copy of 0\mathbb{Z}_{\geq 0} and \mathbb{C}_{\infty} is the completion of the algebraic closure of k:=𝔽q((t1))k_{\infty}:=\mathbb{F}_{q}((t^{-1})) with respect to the unique extension of the valuation vv_{\infty} of kk of uniformizer 1/t.1/t. (See the Introduction in [2] for the precise definitions.)

As a natural GG–equivariant generalization of the Goss ζ\zeta–function ζFE\zeta_{F}^{E} associated to EE and studied by Taelman in [13], the LL–function ΘK/FE\Theta_{K/F}^{E} is defined as an infinite product of Euler factors

ΘK/FE(s):=vPv,G(Nvs)1,\Theta_{K/F}^{E}(s):={\prod_{v}}^{\prime}P_{v}^{\ast,G}(Nv^{-s})^{-1},

where vv runs over all the primes in the maximal spectrum MSpec(𝒪F){\rm MSpec}(\mathcal{O}_{F}) of 𝒪F\mathcal{O}_{F} which are tamely ramified in K/FK/F and of good reduction for EE. For each such prime vv, Pv,G(X)P_{v}^{\ast,G}(X) is a polynomial with coefficients in A[G]A[G], very closely related to the well understood characteristic polynomial

Pv(X):=detAv0(XIrσ(v)Tv0(E))P_{v}(X):={\rm det}_{A_{v_{0}}}(X\cdot I_{r}-\sigma(v)\mid T_{v_{0}}(E))

of the action of a Frobenius morphism σ(v)Gal(F¯/F)\sigma(v)\in{\rm Gal}(\overline{F}/F) associated to vv on the v0v_{0}–adic Tate module Tv0(E)T_{v_{0}}(E) associated to the rank rr Drinfeld module EE at any prime v0MSpec(A){vA}v_{0}\in{\rm MSpec}(A)\setminus\{v\cap A\}, viewed as a free, rank rr module over the completion Av0A_{v_{0}} of AA at v0v_{0}. In fact, one has equalities (see [2], Introduction):

Pv,G(1)=Pv(evσv)Pv(0),ΘK/FE(0)=vPv(0)Pv(evσv),P_{v}^{\ast,G}(1)=\frac{P_{v}(e_{v}\sigma_{v})}{P_{v}(0)},\qquad\Theta_{K/F}^{E}(0)={\prod_{v}}^{\prime}\frac{P_{v}(0)}{P_{v}(e_{v}\sigma_{v})},

where the infinite product converges in 𝔽q((t1))[G]\mathbb{F}_{q}((t^{-1}))[G], σvG\sigma_{v}\in G is a choice of Frobenius for vv in GG, and eve_{v} is the idempotent in A[G]A[G] associated to the trivial character of the inertia group of vv in GG.

The main result of [2] is the proof of a Tamagawa number formula of the type

ΘK/FE,(0)=Vol(E(K/))Vol(K/),\Theta_{K/F}^{E,\mathcal{M}}(0)=\frac{{\rm Vol}(E(K_{\infty}/\mathcal{M}))}{{\rm Vol}(K_{\infty}/\mathcal{M})},

where ΘK/FE,(0)\Theta_{K/F}^{E,\mathcal{M}}(0) is a certain Euler product completion of ΘK/FE(0)\Theta_{K/F}^{E}(0) at primes vMSpec(𝒪F)v\in{\rm MSpec}(\mathcal{O}_{F}) which are either wildly ramified in K/FK/F or of bad reduction for EE, constructed out of some additional arithmetic data \mathcal{M} (called a taming module) for K/FK/F. The numerator and denominator of the right side in the above formula are both volumes (with values in 𝔽q((t1))[G]\mathbb{F}_{q}((t^{-1}))[G] and defined precisely in [2]) of certain compact A[G]A[G]–modules of Arakelov type E(K/)E(K_{\infty}/\mathcal{M}) and K/K_{\infty}/\mathcal{M}.

The above formula, proved in [2] for Drinfeld modules and extended in [4] for higher dimensional abelian tt–modules, generalizes to the GG–equivariant setting Taelman’s celebrated class–number formula [13] for the value ζFE(0)\zeta_{F}^{E}(0) of the Goss zeta–function associated to EE. The interested reader should also consult [1] for a slightly different proof in the abelian tt–module case, using deformation theory, which only gives the desired result under certain restrictive conditions.

However, the proof of the above formula in [2] and [4] (and the same applies to the proofs given in [13] and [1]) hinges upon an essential equality at the level of Euler factors, namely

Pv(evσv)Pv(0)=|E(𝒪K/v)|G|𝒪K/v|G in 𝔽q((t1))[G],\frac{P_{v}(e_{v}\sigma_{v})}{P_{v}(0)}=\frac{|E(\mathcal{O}_{K}/v)|_{G}}{|\mathcal{O}_{K}/v|_{G}}\qquad\text{ in }\mathbb{F}_{q}((t^{-1}))[G],

for all good and tame primes vv as above, where the numerator and denominator of the right side are certain special generators of the Fitting ideals of the finite A[G]A[G]–modules of finite projective dimension E(𝒪K/v)E(\mathcal{O}_{K}/v) and 𝒪K/v\mathcal{O}_{K}/v, respectively. In the Appendix of [2], we prove the above equality only in the simplest case where E:=𝒞E:=\mathcal{C} is the Carlitz module, and Taelman does the same in [13], in the case where GG is trivial.

In this paper, we develop techniques which allow us to prove the above equality for arbitrary Drinfeld modules (see Theorem 2.3 and its proofs in the unramified case, given in §5 and tamely ramified case, given in §6.) In §8, we also indicate how our techniques can be easily extended to prove the above formula for higher dimensional abelian tt–modules, satisfying a certain purity condition. The case of arbitrary abelian tt–modules will be treated in a separate, upcoming paper.

Acknowledgement. We would like to thank Urs Hartl for his expert and generous help with the material on local shtukas in §7. His private e-mail exchanges with the first author [6] were invaluable to us.

2. The statement of the problem

Let pp be a prime number and let qq be a power of pp. In what follows, k:=𝔽q(t)k:=\mathbb{F}_{q}(t) denotes the rational function field in one variable over 𝔽q.\mathbb{F}_{q}. For any commutative 𝔽q\mathbb{F}_{q}-algebra R,R, we denote by τ\tau the qq-power Frobenius endomorphism of RR. We denote by R{τ}R\{\tau\} the twisted polynomial ring in τ\tau, with the property that

τx=xqτxR.\tau\cdot x=x^{q}\cdot\tau\hskip 10.0pt\forall\hskip 5.0ptx\in R.

Let FF be a finite, separable extension of 𝔽q(t)\mathbb{F}_{q}(t) and let KK be a finite abelian extension of FF with Galois group G.G. We also assume that the field of constants in KK is 𝔽q,\mathbb{F}_{q}, i.e.

K𝔽¯q=𝔽q.K\cap\overline{\mathbb{F}}_{q}=\mathbb{F}_{q}.

Let us denote 𝔽q[t]\mathbb{F}_{q}[t] by A.A. Note that if vv denotes an arbitrary normalized valuation on 𝔽q(t)\mathbb{F}_{q}(t) and \infty denotes the normalized valuation of uniformizer 1/t1/t, then

A={a𝔽q(t)|v(a)0, for all v}.A=\{a\in\mathbb{F}_{q}(t)\,\big{|}\,v(a)\geq 0,\text{ for all }\hskip 5.0ptv\neq\infty\}.

Let 𝒪F\mathcal{O}_{F} and 𝒪K\mathcal{O}_{K} denote the integral closures of AA in FF and KK, respectively. In what follows, we abuse notation and use the same letter for normalized valuations and the associated maximal ideals of elements of strictly positive valuation.

Next, we consider a Drinfeld module EE of rank rr\in\mathbb{N} defined on AA with values in 𝒪F{τ}.\mathcal{O}_{F}\{\tau\}. More precisely, EE is given by an 𝔽q\mathbb{F}_{q}-algebra morphism

ϕE:A𝒪F{τ},ttτ0+e1τ++erτr,\phi_{E}:A\to\mathcal{O}_{F}\{\tau\},\hskip 20.0ptt\mapsto t\cdot\tau^{0}+e_{1}\tau+\dots+e_{r}\cdot\tau^{r},

where ai𝒪Fa_{i}\in\mathcal{O}_{F}, for all ii and er0.e_{r}\neq 0. This gives rise to a functor

E:(𝒪F{τ}[G]modules)(A[G]modules).E:(\mathcal{O}_{F}\{\tau\}[G]-\text{modules})\to(A[G]-\text{modules}).

In other words, for any 𝒪F{τ}[G]\mathcal{O}_{F}\{\tau\}[G]-module M,M, we denote by E(M)E(M) the A[G]A[G]-module whose underlying 𝔽q[G]\mathbb{F}_{q}[G]–module is MM and the AA-action is given by

tm=ϕE(t)m=tm+e1τm++erτrmt\star m=\phi_{E}(t)\cdot m=t\cdot m+e_{1}\tau\cdot m+\dots+e_{r}\tau^{r}\cdot m

Let v0MSpec(A)v_{0}\in{\rm MSpec}(A) and let Av0A_{v_{0}} and kv0k_{v_{0}} denote the completions of AA and kk with respect to the valuation v0v_{0}. For all nn\in\mathbb{N}, we denote by E[v0n]E[v_{0}^{n}] the Av0A_{v_{0}}-module of v0nv_{0}^{n}-torsion points of E,E, i.e.

E[v0n]={xE(F¯)|fx=0, for all fv0n}.E[v_{0}^{n}]=\{x\in E(\overline{F})|f\star x=0,\text{ for all }f\in v_{0}^{n}\}.

The v0v_{0}-adic Tate module of EE is defined as

Tv0(E)=HomAv0(kv0/Av0,E[v0]).T_{v_{0}}(E)=\text{Hom}_{A_{v_{0}}}(k_{v_{0}}/A_{v_{0}},E[v_{0}^{\infty}]).

Since AA is a PID, we also have

Tv0(E)=limE[v0n],T_{v_{0}}(E)=\varprojlim E[v_{0}^{n}],

where the transition maps in the projective limit are given by multiplication with a generator of v0v_{0}, while E[v0]=n1E[v0n].E[v_{0}^{\infty}]=\bigcup_{n\geq 1}E[v_{0}^{n}]. Recall that E[v0n]E[v_{0}^{n}] and Tv0(E)T_{v_{0}}(E) are free modules of rank rr over A/v0nA/v_{0}^{n} and Av0A_{v_{0}}, respectively, and are endowed with obvious Av0A_{v_{0}}-linear, continuous GFG_{F}-actions, where GF=Gal(F¯/F).G_{F}=Gal(\overline{F}/F).

Let vv\in MSpec(𝒪F\mathcal{O}_{F}), such that vv0.v\nmid v_{0}. Fix a choice of decomposition group G(v)GF,G(v)\subset G_{F}, and a Frobenius morphism σ(v)G(v).\sigma(v)\in G(v). Then, it is known (see [2] and the references therein) that if EE has good reduction at vv (i.e. verv\nmid e_{r}), the GFG_{F}-representation Tv0(E)T_{v_{0}}(E) is unramified at vv and the polynomial

Pv(X)=detAv0(XIrσ(v)|Tv0(E))P_{v}(X)={\rm det}_{A_{v_{0}}}(X\cdot I_{r}-{\sigma(v)}|T_{v_{0}}(E))

is independent of v0v_{0} and actually lies in A[X].A[X]. Above, IrI_{r} denotes the r×rr\times r identity matrix.

Definition 2.1.

Let MM be an A[G]A[G]-module which is free of rank mm as an 𝔽q[G]\mathbb{F}_{q}[G]-module. Then it is known (see [2] Proposition A.4.1) that the Fitting ideal FittA[G]0(M){\rm Fitt}_{A[G]}^{0}(M) is principal and has a unique tt–monic generator fM(t)A[G]=𝔽q[G][t]f_{M}(t)\in A[G]=\mathbb{F}_{q}[G][t] of degree m.m. We denote this generator by |M|G,|M|_{G}, i.e.

|M|G=fM(t)𝔽q[G][t].|M|_{G}=f_{M}(t)\in\mathbb{F}_{q}[G][t].

The following is Proposition A5.1. from the Appendix in [2]:

Proposition 2.2.

Assume that vv is tamely ramified in K/FK/F and let EE be any Drinfeld module as above. Let w0w_{0} denote the prime in AA sitting below vv and let f(v/w0)=[𝒪F/v:A/w0].f(v/w_{0})=[\mathcal{O}_{F}/v:A/w_{0}]. Then the following hold:

  1. (1)

    The 𝔽q[G]\mathbb{F}_{q}[G]-modules 𝒪K/v\mathcal{O}_{K}/v and E(𝒪K/v)E(\mathcal{O}_{K}/v) are free of rank nv=[𝒪F/v:𝔽q]n_{v}=[\mathcal{O}_{F}/v:\mathbb{F}_{q}] and therefore |𝒪K/v|G|\mathcal{O}_{K}/v|_{G} and |E(𝒪K/v)|G|E(\mathcal{O}_{K}/v)|_{G} are monic polynomials of tt–degree nvn_{v}.

  2. (2)

    We have an equality

    |𝒪K/v|G=Nv|\mathcal{O}_{K}/v|_{G}=Nv

    where NvNv denotes the unique monic generator of w0f(v/w0)w_{0}^{f(v/w_{0})} and f(v/w0):=[𝒪F/v:A/w0]f(v/w_{0}):=[\mathcal{O}_{F}/v:A/w_{0}].

Let IvGvGI_{v}\subset G_{v}\subset G denote the inertia and decomposition groups of vv in GG, respectively. Let σv\sigma_{v} be the image of σ(v){\sigma(v)} via the Galois restriction map G(v)GvG(v)\twoheadrightarrow G_{v}. Our main goal in this paper is the proof of the following.

Theorem 2.3.

Assume that vMSpec(𝒪F)v\in{\rm MSpec}(\mathcal{O}_{F}) is tamely ramified in K/FK/F and that EE has good reduction at vv. Then, we have an equality in 𝔽q[G][[1/t]]\mathbb{F}_{q}[G][[1/t]]

Pv(σvev)Pv(0)=|E(𝒪K/v)|G|𝒪K/v|G,\frac{P_{v}(\sigma_{v}e_{v})}{P_{v}(0)}=\frac{|E(\mathcal{O}_{K}/v)|_{G}}{|\mathcal{O}_{K}/v|_{G}},

where ev=1|Iv|σIvσe_{v}=\frac{1}{|I_{v}|}\sum_{\sigma\in I_{v}}\sigma is the idempotent of the trivial character of IvI_{v} in A[G].A[G].

A proof of the above statement in the case where EE is the Carlitz module CC, defined by ϕC(t)=t+τ\phi_{C}(t)=t+\tau, was given in the Appendix of [2]. Below, we develop techniques which settle the above theorem for a general Drinfeld module EE. Proposition 1.2(2) gives us a good understanding of |𝒪K/v|G|\mathcal{O}_{K}/v|_{G}. Therefore a major portion of our work is directed towards understanding the relation between |E(𝒪K/v)|G|E(\mathcal{O}_{K}/v)|_{G} and Pv(σvev).P_{v}(\sigma_{v}e_{v}).

3. The reduction E¯\overline{E} of EE modulo vv

In this section, we fix a prime vMSpec(𝒪F)v\in\text{MSpec}(\mathcal{O}_{F}) such that EE has good reduction at v.v. We are not assuming that vv is necessarily tamely ramified in K/FK/F. Let us denote by w0w_{0} the prime in AA that lies below v.v. After reduction of EmodvE\mod v (i.e. reduction of the coefficients of EE modulo vv), we obtain the rank rr Drinfeld module E¯,\overline{E}, defined over 𝒪F/v\mathcal{O}_{F}/v, given by the 𝔽q\mathbb{F}_{q}-algebra morphism

ϕE¯:A𝒪F{τ}𝒪F/v{τ}\phi_{\overline{E}}:A\to\mathcal{O}_{F}\{\tau\}\twoheadrightarrow\mathcal{O}_{F}/v\{\tau\}

where ϕE¯(t)=i(t)τ0++i(er)τr\phi_{\overline{E}}(t)=i(t)\cdot\tau^{0}+...+i(e_{r})\cdot\tau^{r} with i:A𝒪F𝒪F/vi:A\xhookrightarrow{}\mathcal{O}_{F}\twoheadrightarrow\mathcal{O}_{F}/v being the obvious map. The Drinfeld module E¯\overline{E} has rank rr, characteristic w0:=ker(i)w_{0}:=\ker(i), and height hh. By definition, hh is the unique integer 0<hr0<h\leq r which, for any generator πw0\pi_{w_{0}} of w0w_{0}, satisfies the equality

ϕE¯(πw0)=bd0hτd0h+bd0h+1τd0h+1++bd0rτrd0,\phi_{\overline{E}}(\pi_{w_{0}})=b_{d_{0}h}\tau^{d_{0}h}+b_{d_{0}h+1}\tau^{d_{0}h+1}+\dots+b_{d_{0}r}\tau^{rd_{0}},

where d0:=[A/w0:𝔽q]d_{0}:=[A/w_{0}:\mathbb{F}_{q}], bi𝒪F/vb_{i}\in\mathcal{O}_{F}/v, and bd0h0b_{d_{0}h}\neq 0. (See [3], Section 4.5 for the existence of hh.)

Recall that, by the notation introduced above, we have a field isomorphism 𝒪F/v𝔽qnv.\mathcal{O}_{F}/v\simeq\mathbb{F}_{q^{n_{v}}}. Consequently, the Frobenius morphism Frobqnv:=τnv{\rm Frob}_{q^{n_{v}}}:=\tau^{n_{v}} is an endomorphism of E¯\overline{E} and it acts naturally and linearly on all the Tate modules associated to E¯\overline{E}.

Next, we fix v0MSpec(A)v_{0}\in\text{MSpec}(A), v0w0v_{0}\neq w_{0} and consider the characteristic polynomial of the action of the qnvq^{n_{v}}-power Frobenius morphism, viewed as an endomorphism of E¯\overline{E}, on the free Av0A_{v_{0}}–module Tv0(E¯)T_{v_{0}}(\overline{E}) of rank rr:

fE¯(X)=detAv0(XIrFrobqnvTv0(E¯)).f_{\overline{E}}(X)={\rm det}_{A_{v_{0}}}(X\cdot I_{r}-{\rm Frob}_{q^{n_{v}}}\mid T_{v_{0}}(\overline{E})).

Then, fE¯(X)f_{\overline{E}}(X) is independent of v0v_{0} and lies in A[X]A[X]. (See §4.12 in [3].) By Theorem 4.12.15 in [3] and the discussion preceding that, we have the following.

Proposition 3.1.

Any root α\alpha of fE¯f_{\overline{E}} satisfies the following properties:

  1. (1)

    w(α)=0w(\alpha)=0 for all finite places ww of 𝔽q(t)(α),\mathbb{F}_{q}(t)(\alpha), except for exactly one place above w0.w_{0}.

  2. (2)

    There is only one place of 𝔽q(t)(α)\mathbb{F}_{q}(t)(\alpha) lying above .\infty.

  3. (3)

    |α|=qnvr|\alpha|_{\infty}=q^{\frac{n_{v}}{r}} where |.||.|_{\infty} denotes the unique extension to 𝔽q(t)(α)\mathbb{F}_{q}(t)(\alpha) of the normalized absolute value of 𝔽q(t)\mathbb{F}_{q}(t) corresponding to .\infty.

  4. (4)

    [𝔽q(t)(α):𝔽q(t)][\mathbb{F}_{q}(t)(\alpha):\mathbb{F}_{q}(t)] divides r.r.

Further, one can consider the characteristic polynomial of Frobqnv{\rm Frob}_{q^{n_{v}}} (viewed as endomorphism of E¯\overline{E}) acting on the free Aw0A_{w_{0}}–module Tw0(E¯)T_{w_{0}}(\overline{E}) of rank (rh)(r-h) (see [3], Section 4.5 for the calculation of the rank)

gE¯(X)=detAw0(XIrhFrobqnvTw0(E¯)).g_{\overline{E}}(X)={\rm det}_{A_{w_{0}}}(X\cdot I_{r-h}-{\rm Frob}_{q^{n_{v}}}\mid T_{w_{0}}(\overline{E})).

This is a monic polynomial in Aw0[X]A_{w_{0}}[X] of degree (rh).(r-h). In §? below we will prove the following.

Proposition 3.2.

The polynomial gE¯(X)g_{\overline{E}}(X) divides the polynomial fE¯(X)f_{\overline{E}}(X) in Aw0[X]A_{w_{0}}[X].

Proof.

See §7 below for the proof of a stronger statement. ∎

Let 𝒪v\mathcal{O}_{v} and FvF_{v} be the completions at vv of 𝒪F\mathcal{O}_{F} and FF, respectively. Our choice of decomposition group G(v)G(v) corresponds to choosing an embedding F¯Fv¯\overline{F}\to\overline{F_{v}} at the level of separable closures of FF and FvF_{v}, such that Galois restriction induces a group isomorphism G(Fv¯/Fv)G(v)G(\overline{F_{v}}/F_{v})\simeq G(v). Since EE has good reduction at vv and the Galois representations E[v0n]E[v_{0}^{n}] are unramified at vv, it is not difficult to see that we have

E[v0n]𝒪vnr, for all n1,E[v_{0}^{n}]\subseteq\mathcal{O}_{v}^{nr},\text{ for all }n\geq 1,

where 𝒪vnr\mathcal{O}_{v}^{nr} is the integral closure of 𝒪v\mathcal{O}_{v} in the maximal unramified extension FvunrF_{v}^{unr} of FvF_{v} in Fv¯\overline{F_{v}}. Moreover, the reduction modv\mod v map induces isomorphisms of Av0[[G(v)¯]]A_{v_{0}}[[\overline{G(v)}]]–modules

(1) E[v0n]E¯[v0n],Tv0(E)Tv0(E¯),E[v_{0}^{n}]\simeq\overline{E}[v_{0}^{n}],\qquad T_{v_{0}}(E)\simeq T_{v_{0}}(\overline{E}),

where

G(v)¯:=G(v)/I(v)G(𝔽qnv¯/𝔽qnv).\overline{G(v)}:=G(v)/I(v)\simeq G(\overline{\mathbb{F}_{q^{n_{v}}}}/\mathbb{F}_{q^{n_{v}}}).

The group isomorphism above sends σ(v)¯\overline{{\sigma(v)}} (the image of our choice of Frobenius σ(v)\sigma(v) in G(v)¯\overline{G(v)}) to Frobqnv{\rm Frob}_{q^{n_{v}}}. Consequently, we have an equality of characteristic polynomials in A[X]A[X]:

(2) fE¯(X)=Pv(X).f_{\overline{E}}(X)=P_{v}(X).

Consequently, Proposition 3.1 gives us information on the roots of the characteristic polynomial Pv(X)P_{v}(X). The following corollary regarding the coefficients of Pv(X)P_{v}(X) will be particularly useful in what follows.

Corollary 3.3.

Let Pv(X)=a0+a1X++ar1Xr1+XrP_{v}(X)=a_{0}+a_{1}X+\cdots+a_{r-1}X^{r-1}+X^{r}, with a0,,ar1Aa_{0},\dots,a_{r-1}\in A. Then, we have

  1. (1)

    degt(a0)=nv{\rm deg}_{t}(a_{0})=n_{v} and 0<degt(ai)<nv0<{\rm deg}_{t}(a_{i})<n_{v}, for all i>0i>0.

  2. (2)

    Pv(X)𝔽q[X][t]P_{v}(X)\in\mathbb{F}_{q}[X][t] is a polynomial of degree nvn_{v} in tt with the same leading coefficient as a0a_{0}.

  3. (3)

    a0=ρNva_{0}=\rho\cdot Nv, for some ρ𝔽q×\rho\in\mathbb{F}_{q}^{\times}, where NvNv is the unique monic generator of w0f(v/w0).w_{0}^{f(v/w_{0})}.

Above, degt(){\rm deg}_{t}(\ast) denotes the degree in tt of a polynomial in A=𝔽q[t].A=\mathbb{F}_{q}[t].

Proof.

Let α1,,αrA¯\alpha_{1},...,\alpha_{r}\in\overline{A} denote the roots of Pv(X)P_{v}(X) in the integral closure of AA. Then

Pv(X)=i=1r(Xαi)\displaystyle P_{v}(X)=\prod_{i=1}^{r}(X-\alpha_{i}) =(1)ri=1rαi+(1)r1(j=1rijαi)X++Xr\displaystyle=(-1)^{r}\prod_{i=1}^{r}\alpha_{i}+(-1)^{r-1}\bigg{(}\sum_{j=1}^{r}\prod_{i\neq j}\alpha_{i}\bigg{)}X+...+X^{r}
=a0+a1X++ar1Xr1+Xr.\displaystyle=a_{0}+a_{1}\cdot X+\dots+a_{r-1}\cdot X^{r-1}+X^{r}.

Let |||\cdot|_{\infty} denote an extension to 𝔽q(t)(α1,,αr)\mathbb{F}_{q}(t)(\alpha_{1},...,\alpha_{r}) of the normalized absolute value of 𝔽q(t)\mathbb{F}_{q}(t) corresponding to \infty (also denoted by |||\cdot|_{\infty} below.) By Proposition 1.4, we have |αi|=qnvr|\alpha_{i}|_{\infty}=q^{\frac{n_{v}}{r}}, for all ii. Therefore, we have

|a0|=|i=1rαi|=qnv,degt(a0)=logq(|a0|)=nv.|a_{0}|_{\infty}=\bigg{\lvert}\prod_{i=1}^{r}\alpha_{i}\bigg{\rvert}_{\infty}=q^{n_{v}},\qquad{\rm deg}_{t}(a_{0})=\log_{q}(|a_{0}|_{\infty})=n_{v}.

Furthermore, since |||\cdot|_{\infty} is non-archimedean, we have

|ai|qnvrir,degt(ai)=logq(|ai|)nvrir<nv, for all i1.|a_{i}|_{\infty}\leq q^{n_{v}\cdot\frac{r-i}{r}},\qquad{\rm deg}_{t}(a_{i})=\log_{q}(|a_{i}|_{\infty})\leq n_{v}\cdot\frac{r-i}{r}<n_{v},\quad\text{ for all }i\geq 1.

This concludes the proof of part (1).

Part (2) is a direct consequence of part (1). Further, since a0=(1)ri=1rαia_{0}=(-1)^{r}\prod_{i=1}^{r}\alpha_{i}, part (3) is a direct consequence of Proposition 3.1(1)–(3). ∎


4. Fitting ideals of Tate modules and consequences

We begin by stating a general commutative algebra result regarding Fitting ideals of modules over certain rings of equivariant Iwasawa algebra type. For a proof of this result, see Proposition 4.1 in [5].

Proposition 4.1 (Greither–Popescu).

Let RR be a semi-local, compact topological ring, and let Γ\Gamma be a pro-cyclic group, topologically generated by γ.\gamma. Suppose that MM is an R[[Γ]]R[[\Gamma]]–module which is free of rank nn as an RR–module. Let MγMn(R)M_{\gamma}\in M_{n}(R) denote the matrix of the action of γ\gamma on some RR-basis of M.M. Then, we have an equality of R[[Γ]]R[[\Gamma]]–ideals

FittR[[Γ]](M)=(detR(XInMγ)|X=γ){\rm Fitt}_{R[[\Gamma]]}(M)=\bigg{(}{\rm det}_{R}(X\cdot I_{n}-M_{\gamma})\bigg{\lvert}_{X=\gamma}\bigg{)}
Proof.

See the proof of Proposition 4.1 in [5]. ∎

An immediate consequence of the above proposition is the following.

Corollary 4.2.

For all v0MSpec(A)v_{0}\in{\rm MSpec}(A), we have the following equalities of Av0[[G(v)¯]]A_{v_{0}}[[\overline{G(v)}]]–ideals.

  1. (1)

    If v0w0v_{0}\neq w_{0}, then

    FittAv0[[G(v)¯]](Tv0(E¯))=FittAv0[[G(v)¯]](Tv0(E))=(fE¯(σ(v)¯))=(Pv(σ(v)¯)).{\rm Fitt}_{A_{v_{0}}[[\overline{G(v)}]]}(T_{v_{0}}(\overline{E}))={\rm Fitt}_{A_{v_{0}}[[\overline{G(v)}]]}(T_{v_{0}}(E))=(f_{\overline{E}}(\overline{\sigma(v)}))=(P_{v}(\overline{\sigma(v)})).
  2. (2)

    If v0=w0v_{0}=w_{0}, then

    FittAw0[[G(v)¯]](Tw0(E¯))=(gE¯(σ(v)¯)){\rm Fitt}_{A_{w_{0}}[[\overline{G(v)}]]}(T_{w_{0}}(\overline{E}))=(g_{\overline{E}}(\overline{\sigma(v)}))
Proof.

Apply the proposition above to R:=Av0R:=A_{v_{0}}, Γ:=G(v)¯\Gamma:=\overline{G(v)}, γ:=σ(v)¯\gamma:=\overline{\sigma(v)} and the module

M:=Tv0(E)Tv0(E¯),M:=T_{v_{0}}(E)\simeq T_{v_{0}}(\overline{E}),

which is Av0A_{v_{0}}–free of rank rr, if v0w0v_{0}\neq w_{0} and, respectively, the module

M:=Tw0(E¯),M:=T_{w_{0}}(\overline{E}),

which is Aw0A_{w_{0}}–free of rank (rh)(r-h), if v0=w0.v_{0}=w_{0}.

Next, we fix a prime wMSpec(𝒪K)w\in{\rm MSpec}(\mathcal{O}_{K}) lying above vv. We let G(w)=G(v)GKG(w)=G(v)\cap G_{K}, I(w)=I(v)GKI(w)=I(v)\cap G_{K} and σ(w):=σ(v)f\sigma(w):=\sigma(v)^{f}, where f:=f(w/v)=[𝒪K/w:𝒪F/v]f:=f(w/v)=[\mathcal{O}_{K}/w:\mathcal{O}_{F}/v]. Then, σ(w)G(w)\sigma(w)\in G(w) is a choice of Frobenius for ww and its image σ(w)¯G(w)¯:=G(w)/I(w)\overline{\sigma(w)}\in\overline{G(w)}:=G(w)/I(w) corresponds via the group isomorphism G(w)¯G𝒪K/w\overline{G(w)}\simeq G_{\mathcal{O}_{K}/w} to Frobqfnv{\rm Frob}_{q^{fn_{v}}}, viewed as an endomorphism of E¯\overline{E}.

Lemma 4.3.

Let v0MSpec(A)v_{0}\in{\rm MSpec}(A). Then, we have the following canonical isomorphisms of Av0[Gv]A_{v_{0}}[G_{v}]–modules:

  1. (1)

    Tv0(E¯)/(1σ(w)¯)Tv0(E¯)E¯(𝒪K/w)v0=E(𝒪K/w)v0.T_{v_{0}}(\overline{E})/(1-\overline{\sigma(w)})T_{v_{0}}(\overline{E})\simeq\overline{E}(\mathcal{O}_{K}/w)_{v_{0}}={E}(\mathcal{O}_{K}/w)_{v_{0}}.

  2. (2)

    Tv0(E)/(1σ(w))Tv0(E)E(𝒪K/w)v0T_{v_{0}}(E)/(1-\sigma(w))T_{v_{0}}(E)\simeq E(\mathcal{O}_{K}/w)_{v_{0}}, assuming that v0w0v_{0}\neq w_{0}.

Above, we let Mv0:=MAAv0M_{v_{0}}:=M\otimes_{A}A_{v_{0}} for any AA–module MM.

Proof.

By the definition of E¯\overline{E}, we have

E(𝒪K/w)=E¯(𝒪K/w).E(\mathcal{O}_{K}/w)=\overline{E}(\mathcal{O}_{K}/w).

Consequently, part (2) follows from part (1) via the second isomorphism in (1) above.

In order to prove part (1), apply the functor HomAv0(,E¯[v0])\ast\to{\rm Hom}_{A_{v_{0}}}(\ast,\overline{E}[v_{0}^{\infty}]) to the exact sequence of Av0A_{v_{0}}–modules

0Av0kv0kv0/Av00.0\longrightarrow A_{v_{0}}\longrightarrow k_{v_{0}}\longrightarrow k_{v_{0}}/A_{v_{0}}\longrightarrow 0.

Since the Av0A_{v_{0}}–module E¯[v0]:=nE¯[v0n]\overline{E}[v_{0}^{\infty}]:=\cup_{n}\overline{E}[v_{0}^{n}] is divisible and therefore injective (as Av0A_{v_{0}} is a PID), the above functor is exact. Therefore, we obtain the following exact sequence of Av0[G(v)¯]A_{v_{0}}[\overline{G(v)}]–modules:

(3) 0Tv0(E¯)HomAv0(kv0,E¯[v0])E¯[v0]0.0\longrightarrow T_{v_{0}}(\overline{E})\longrightarrow{\rm Hom}_{A_{v_{0}}}(k_{v_{0}},\overline{E}[v_{0}^{\infty}])\longrightarrow\overline{E}[v_{0}^{\infty}]\longrightarrow 0.

Now, it is easy to see that one has an isomorphism of kv0[G(v)¯]k_{v_{0}}[\overline{G(v)}]–modules

kv0Av0Tv0(E¯)HomAv0(kv0,E¯[v0]),ξϕ(xϕ(ξx^)),k_{v_{0}}\otimes_{A_{v_{0}}}T_{v_{0}}(\overline{E})\simeq{\rm Hom}_{A_{v_{0}}}(k_{v_{0}},\overline{E}[v_{0}^{\infty}]),\quad\xi\otimes\phi\to(x\to\phi(\widehat{\xi\cdot x})),

for all ξ,xkv0\xi,x\in k_{v_{0}} and all ϕTv0(E¯)=HomAv0(kv0/Av0,E¯[v0])\phi\in T_{v_{0}}(\overline{E})={\rm Hom}_{A_{v_{0}}}(k_{v_{0}}/A_{v_{0}},\overline{E}[v_{0}^{\infty}]), where xξ^\widehat{x\cdot\xi} is the class of xξx\cdot\xi in kv0/Av0k_{v_{0}}/A_{v_{0}}.

Now, Proposition 3.1(3) and Proposition 3.2 show that the eigenvalues of σ(w)¯=σ(v)¯f=(Frobqnv)f\overline{\sigma(w)}=\overline{\sigma(v)}^{f}=({\rm Frob}_{q^{n_{v}}})^{f} acting on the kv0k_{v_{0}}–vector space kv0Av0Tv0(E¯)k_{v_{0}}\otimes_{A_{v_{0}}}T_{v_{0}}(\overline{E}) are all different from 11. Consequently, (σ(w)¯1)(\overline{\sigma(w)}-1) is an automorphism of this kv0k_{v_{0}}–vector space. Consequently, when one takes the σ(w)¯\overline{\sigma(w)}–invariants and coinvariants in the exact sequence (3) above, one obtains an isomorphism of Av0[Gv]A_{v_{0}}[G_{v}]–modules

Tv0(E¯)/(1σ(w)¯)Tv0(E¯)E¯[v0]σ(w)¯=1=E¯(𝒪K/w)v0,T_{v_{0}}(\overline{E})/(1-\overline{\sigma(w)})T_{v_{0}}(\overline{E})\simeq\overline{E}[v_{0}^{\infty}]^{\overline{\sigma(w)}=1}=\overline{E}(\mathcal{O}_{K}/w)_{v_{0}},

which concludes the proof of the Lemma.

Corollary 4.4.

For all v0MSpec(A)v_{0}\in{\rm MSpec}(A), the following equalities of Av0[Gv¯]A_{v_{0}}[\overline{G_{v}}]–ideals hold:

FittAv0[Gv¯]E(𝒪K/w)v0={(Pv(σv¯)=(fE¯(σv¯)), if v0w0(gE¯(σv¯)), if v0=w0.{\rm Fitt}_{A_{v_{0}}[\overline{G_{v}}]}E(\mathcal{O}_{K}/w)_{v_{0}}=\begin{cases}\big{(}P_{v}(\overline{\sigma_{v}}\big{)}=\big{(}f_{\overline{E}}(\overline{\sigma_{v}})\big{)},&\text{ if }v_{0}\neq w_{0}\\ \big{(}g_{\overline{E}}(\overline{\sigma_{v}})\big{)},&\text{ if }v_{0}=w_{0}.\end{cases}

Further, if v0w0v_{0}\neq w_{0}, then we have an equality of A(v0)[Gv¯]A_{(v_{0})}[\overline{G_{v}}]–ideals

FittA(v0)[Gv¯]E(𝒪K/w)v0=(Pv(σv¯)=(fE¯(σv¯)).{\rm Fitt}_{A_{(v_{0})}[\overline{G_{v}}]}E(\mathcal{O}_{K}/w)_{v_{0}}=\big{(}P_{v}(\overline{\sigma_{v}}\big{)}=\big{(}f_{\overline{E}}(\overline{\sigma_{v}})\big{)}.

Here, G¯v:=Gv/Iv\overline{G}_{v}:=G_{v}/I_{v}, σ¯v\overline{\sigma}_{v} is the image of σv\sigma_{v} in Gv¯\overline{G_{v}}, and A(v0)A_{(v_{0})} is the localization of AA at v0v_{0}.

Proof.

First, note that the isomorphism of Av0[Gv]A_{v_{0}}[G_{v}]–modules in Lemma 4.3(1) can be rewritten as an isomorphism of Av0[Gv¯]A_{v_{0}}[\overline{G_{v}}]–modules

Tv0(E¯)Av0[[Gv¯]]Av0[G(v)¯]E(𝒪K/w)v0,T_{v_{0}}(\overline{E})\otimes_{A_{v_{0}}[[\overline{G_{v}}]]}A_{v_{0}}[\overline{G(v)}]\simeq E(\mathcal{O}_{K}/w)_{v_{0}},

where the ring morphism π:Av0[[G(v)¯]]Av0[Gv¯]\pi:A_{v_{0}}[[\overline{G(v)}]]\twoheadrightarrow A_{v_{0}}[\overline{G_{v}}] is the Av0A_{v_{0}}–linear map given by Galois restriction, which maps σ(v)¯σv¯.\overline{\sigma(v)}\to\overline{\sigma_{v}}. The isomorphism above permits us to apply the well known base–change property of Fitting ideals which, combined with Corollary 4.2 above, leads to the equalities of Av0[Gv¯]A_{v_{0}}[\overline{G_{v}}]–ideals

FittAv0[Gv¯]E(𝒪K/w)v0=π(FittAv0[[G(v)¯]](Tv0(E)))={(Pv(σv¯)=(fE¯(σv¯)), if v0w0(gE¯(σv¯)), if v0=w0..{\rm Fitt}_{A_{v_{0}}[\overline{G_{v}}]}E(\mathcal{O}_{K}/w)_{v_{0}}=\pi\big{(}{\rm Fitt}_{A_{v_{0}}[[\overline{G(v)}]]}(T_{v_{0}}(E))\big{)}=\begin{cases}\big{(}P_{v}(\overline{\sigma_{v}}\big{)}=\big{(}f_{\overline{E}}(\overline{\sigma_{v}})\big{)},&\text{ if }v_{0}\neq w_{0}\\ \big{(}g_{\overline{E}}(\overline{\sigma_{v}})\big{)},&\text{ if }v_{0}=w_{0}.\end{cases}.

Next, assume that v0w0v_{0}\neq w_{0} and observe that since E(𝒪K/w)E(\mathcal{O}_{K}/w) is finite (and therefore AA–torsion), we have isomorphisms

E(𝒪K/w)AA(v0)(E(𝒪K/w)AA(v0))A(v0)Av0E(𝒪K/w)v0.E(\mathcal{O}_{K}/w)\otimes_{A}A_{(v_{0})}\simeq\big{(}E(\mathcal{O}_{K}/w)\otimes_{A}A_{(v_{0})}\big{)}\otimes_{A_{(v_{0})}}A_{v_{0}}\simeq E(\mathcal{O}_{K}/w)_{v_{0}}.

Consequently, base–change for Fitting ideals applied to the ring extension A(v0)[G]Av0[G]A_{(v_{0})}[G]\subseteq A_{v_{0}}[G] and the last equality of ideals displayed above gives

Pv(σv¯)Av0[G]=FittA(v0)[Gv¯](E(𝒪K/w)v0)Av0[Gv¯].P_{v}(\overline{\sigma_{v}})A_{v_{0}}[G]={\rm Fitt}_{A_{(v_{0})}[\overline{G_{v}}]}\big{(}E(\mathcal{O}_{K}/w)_{v_{0}}\big{)}A_{v_{0}}[\overline{G_{v}}].

However, since the ring extension A(v0)[G]Av0[G]A_{(v_{0})}[G]\subseteq A_{v_{0}}[G] is faithfully flat (because A(v0)Av0A_{(v_{0})}\subseteq A_{v_{0}} is), we have

FittA(v0)[Gv¯](E(𝒪K/w)v0)\displaystyle{\rm Fitt}_{A_{(v_{0})}[\overline{G_{v}}]}\big{(}E(\mathcal{O}_{K}/w)_{v_{0}}\big{)} =FittA(v0)[Gv¯](E(𝒪K/w)v0)Av0[Gv¯]A(v0)[Gv¯]\displaystyle={\rm Fitt}_{A_{(v_{0})}[\overline{G_{v}}]}\big{(}E(\mathcal{O}_{K}/w)_{v_{0}}\big{)}A_{v_{0}}[\overline{G_{v}}]\cap A_{(v_{0})}[\overline{G_{v}}]
=Pv(σv¯)Av0[Gv¯]A(v0)[Gv¯]=Pv(σv¯)A(v0)[Gv¯].\displaystyle=P_{v}(\overline{\sigma_{v}})A_{v_{0}}[\overline{G_{v}}]\cap A_{(v_{0})}[\overline{G_{v}}]=P_{v}(\overline{\sigma_{v}})A_{(v_{0})}[\overline{G_{v}}].

Above, we used the fact that if RRR\subseteq R^{\prime} is a faithfully flat extension of commutative rings and II is an ideal in RR, then IRR=IIR^{\prime}\cap R=I. (See [11], Chapter 2, Section 4, 4.C(ii).) ∎

In the next two sections, we provide a proof of Theorem 2.3. For technical reasons which will become apparent shortly, we treat first the unramified case.

5. The unramified case

We keep the notations and assumptions of the previous section. In addition, we assume that the prime vv is unramified in K/F.K/F. Consequently, we have Gv¯=Gv\overline{G_{v}}=G_{v} and σv¯=σv\overline{\sigma_{v}}=\sigma_{v} throughout.

Lemma 5.1.

Under the current assumptions, we have equalities of ideals.

  1. (1)

    FittA(v0)[G](E(𝒪K/v)v0)=(Pv(σv))=(fE¯(σv)),{\rm Fitt}_{A_{(v_{0})}[G]}(E(\mathcal{O}_{K}/v)_{v_{0}})=(P_{v}(\sigma_{v}))=(f_{\overline{E}}(\sigma_{v})), for all v0MSpec(A)v_{0}\in{\rm MSpec}(A), with v0w0v_{0}\neq w_{0}.

  2. (2)

    FittAw0[G](E(𝒪K/v)w0)=(gE¯(σv)).{\rm Fitt}_{A_{w_{0}}[G]}(E(\mathcal{O}_{K}/v)_{w_{0}})=(g_{\overline{E}}(\sigma_{v})).

Proof.

In this case, we have an isomorphism of A(v0)[G]A_{(v_{0})}[G]–modules

E(𝒪K/v)v0E(𝒪K/w)v0A(v0)[Gv]A(v0)[G],E(\mathcal{O}_{K}/v)_{v_{0}}\simeq E(\mathcal{O}_{K}/w)_{v_{0}}\otimes_{A_{(v_{0})}[G_{v}]}A_{(v_{0})}[G],

for all v0MSpec(A)v_{0}\in{\rm MSpec}(A). (See the Appendix of [2].) Therefore, the equality in the Lemma follows from Corollary 4.4 and the base–change property of Fitting ideals. ∎

We are ready to prove the following refinement of Theorem 2.3 in the unramified case.

Proposition 5.2.

Assume that vv is unramified in K/FK/F. Let ρ𝔽q×\rho\in\mathbb{F}_{q}^{\times} as defined in Corollary 3.3(3). Then, ρ1Pv(0)\rho^{-1}P_{v}(0) and ρ1Pv(σv)\rho^{-1}P_{v}(\sigma_{v}) are monic polynomials of tt–degree nvn_{v} in A=𝔽q[t]A=\mathbb{F}_{q}[t] and A[G]=𝔽q[G][t]A[G]=\mathbb{F}_{q}[G][t], respectively, and the following hold.

  1. (1)

    ρ1Pv(0)=|𝒪K/v|G.\rho^{-1}P_{v}(0)=|\mathcal{O}_{K}/v|_{G}.

  2. (2)

    ρ1Pv(σv)=|E(𝒪K/v)|G.\rho^{-1}P_{v}(\sigma_{v})=|E(\mathcal{O}_{K}/v)|_{G}.

  3. (3)

    Pv(σv)/Pv(0)=|E(𝒪K/v)|G/|𝒪K/v|GP_{v}(\sigma_{v})/P_{v}(0)=|E(\mathcal{O}_{K}/v)|_{G}/|\mathcal{O}_{K}/v|_{G} in 𝔽q[G][[t]]\mathbb{F}_{q}[G][[t]], i.e. Theorem 2.3 holds.

Proof.

According to Corollary 3.3 (see (1) and (3) in loc.cit.), Pv(0)𝔽q[t]P_{v}(0)\in\mathbb{F}_{q}[t] and Pv(X)𝔽q[X][t]=A[X]P_{v}(X)\in\mathbb{F}_{q}[X][t]=A[X], viewed as polynomials in tt, have degrees equal to nvn_{v} and leading coefficients equal to ρ\rho. Therefore, ρ1Pv(0)\rho^{-1}P_{v}(0) and ρ1Pv(σv)\rho^{-1}P_{v}(\sigma_{v}) are indeed monic polynomials of common tt–degree nvn_{v}. Further, Corollary 3.3(3) shows that ρ1Pv(0)=Nv\rho^{-1}P_{v}(0)=Nv which, if combined to Proposition 2.2(2), proves part (1) of the statement above.

Next, we focus on the proof of equality (2) in the Proposition above. In order to simplify the notation, let

f:=ρ1Pv(σv),g:=|E(𝒪K/v)|G.f:=\rho^{-1}P_{v}(\sigma_{v}),\qquad g:=\big{|}E(\mathcal{O}_{K}/v)\big{|}_{G}.

Then, ff and gg are both monic polynomials in tt, of degrees equal to nvn_{v}. (See Proposition 2.2(2) and Corollary 3.3.) Further, Lemma 5.1(1) and the definition of gg imply that they satisfy the following equalities

fA(v0)[G]=gA(v0)[G]=FittA(v0)[G]E(𝒪K/v)v0, for all v0MSpec(A){w0}.fA_{(v_{0})}[G]=gA_{(v_{0})}[G]={\rm Fitt}_{A_{(v_{0})}[G]}E(\mathcal{O}_{K}/v)_{v_{0}},\quad\text{ for all }v_{0}\in\text{MSpec}(A)\setminus\{w_{0}\}.

Now, it is easy to check that the total ring of fractions of A[G]A[G] is k[G]k[G]. Since gg and ff are monic, they are not zero–divisors in A[G]A[G] and k[G]k[G]. Therefore, the equalities above imply that

fgk[G](v0w0A(v0)[G]×)(v0w0A(v0))[G].\frac{f}{g}\in k[G]\cap\big{(}\bigcap_{v_{0}\neq w_{0}}A_{(v_{0})}[G]^{\times}\big{)}\subseteq\big{(}\bigcap_{v_{0}\neq w_{0}}A_{(v_{0})}\big{)}[G].

This implies that there exists ξA[G]\xi\in A[G] and m0m\in\mathbb{Z}_{\geq 0}, such that

fg=ξπw0m, with πw0ξ in A[G] if m1,\frac{f}{g}=\frac{\xi}{\pi_{w_{0}}^{m}},\qquad\text{ with $\pi_{w_{0}}\nmid\xi$ in $A[G]$ if }m\geq 1,

where πw0A\pi_{w_{0}}\in A is the unique monic generator of the maximal ideal w0w_{0}.

We claim that m=0m=0. In order to prove that, let us note that Lemma 5.1 implies that

gE¯(σv)Aw0[G]=gAw0[G].g_{\overline{E}}(\sigma_{v})A_{w_{0}}[G]=gA_{w_{0}}[G].

On the other hand, Proposition 3.2 implies that

gE¯(σv)fE¯(σv) in Aw0[G].g_{\overline{E}}(\sigma_{v})\mid f_{\overline{E}}(\sigma_{v})\text{ in }A_{w_{0}}[G].

Consequently, since f=ρ1Pv(σv)=ρ1fE¯(σv)f=\rho^{-1}P_{v}(\sigma_{v})=\rho^{-1}f_{\overline{E}}(\sigma_{v}), we have

gf in Aw0[G].g\mid f\text{ in }A_{w_{0}}[G].

Therefore, ξ/πw0mAw0[G]\xi/\pi_{w_{0}}^{m}\in A_{w_{0}}[G], which implies that

πw0mξ in Aw0[G].\pi_{w_{0}}^{m}\mid\xi\text{ in }A_{w_{0}}[G].

However, since πw0,ξA[G]\pi_{w_{0}},\xi\in A[G], this shows that the divisibility above happens in A[G]A[G]. Therefore, we have m=0m=0, as claimed and, consequently

fgA[G]=𝔽q[G][t].\frac{f}{g}\in A[G]=\mathbb{F}_{q}[G][t].

However, since ff and gg are monic polynomials in tt of common degree nvn_{v}, we must have

f=g,f=g,

which concludes the proof of part (2) of the Proposition. Part (3) is a consequence of parts (1) and (2).

Remark 5.3.

Note that since Corollary 3.3 is valid in general, regardless of the ramification status of vv in K/FK/F, the polynomials ρ1Pv(X)\rho^{-1}P_{v}(X), ρ1Pv(evσv)\rho^{-1}P_{v}(e_{v}\sigma_{v}), and ρ1Pv(0)\rho^{-1}P_{v}(0) are monic of degree nvn_{v} in tt in general, even in the tamely ramified case. Also, the proof given to part (1) of the Proposition above is valid in general, so even in the tamely ramified case.

6. The tamely ramified case

Now, suppose that vv is tamely ramified in K/FK/F, i.e. p|Iv|p\nmid|I_{v}|. As above, we denote by

ev:=1IvσIvσe_{v}:=\frac{1}{I_{v}}\sum_{\sigma\in I_{v}}\sigma

the idempotent in A[G]A[G] associated to the trivial character of the inertia group IvI_{v} of vv in GG. The goal of this section is to prove the following analogue of Proposition 5.2 in this case.

Proposition 6.1.

Assume that vv is tamely ramified in K/FK/F. Let ρ𝔽q×\rho\in\mathbb{F}_{q}^{\times} as defined in Corollary 3.3(3). Then, ρ1Pv(0)\rho^{-1}P_{v}(0) and ρ1Pv(evσv)\rho^{-1}P_{v}(e_{v}\sigma_{v}) are monic polynomials of tt–degree nvn_{v} in A=𝔽q[t]A=\mathbb{F}_{q}[t] and A[G]=𝔽q[G][t]A[G]=\mathbb{F}_{q}[G][t], respectively, and the following hold.

  1. (1)

    ρ1Pv(0)=|𝒪K/v|G.\rho^{-1}P_{v}(0)=|\mathcal{O}_{K}/v|_{G}.

  2. (2)

    ρ1Pv(evσv)=|E(𝒪K/v)|G.\rho^{-1}P_{v}(e_{v}\sigma_{v})=|E(\mathcal{O}_{K}/v)|_{G}.

  3. (3)

    Pv(evσv)/Pv(0)=|E(𝒪K/v)|G/|𝒪K/v|GP_{v}(e_{v}\sigma_{v})/P_{v}(0)=|E(\mathcal{O}_{K}/v)|_{G}/|\mathcal{O}_{K}/v|_{G} in 𝔽q[G][[t]]\mathbb{F}_{q}[G][[t]], i.e. Theorem 2.3 holds.

Proof.

Observe that part (3) is a consequence of parts (1) and (2). Since Remark 5.3 settles everything but part (2) in the more general, tamely ramified case, we will focus below on proving part (2).

Note that, if e:=|Iv|e:=|I_{v}| and ww is a fixed prime in KK above vv, as before we have isomorphisms of A[G]A[G]–modules

𝒪K/v𝒪K/weA[Gv]A[G],E(𝒪K/v)E(𝒪K/we)A[Gv]A[G].\mathcal{O}_{K}/v\simeq\mathcal{O}_{K}/w^{e}\otimes_{A[G_{v}]}A[G],\qquad E(\mathcal{O}_{K}/v)\simeq E(\mathcal{O}_{K}/w^{e})\otimes_{A[G_{v}]}A[G].

Consequently, base–change for Fitting ideals and the isomorphism 𝒪K/we𝒪w/we\mathcal{O}_{K}/w^{e}\simeq\mathcal{O}_{w}/w^{e} show that it suffices for us to work locally and prove that we have an equality of A[Gv]A[G_{v}]–ideals

(4) FittA[Gv]E(𝒪w/we)=(Pv(evσv)),{\rm Fitt}_{A[G_{v}]}E(\mathcal{O}_{w}/w^{e})=(P_{v}(e_{v}\sigma_{v})),

where 𝒪w\mathcal{O}_{w} is the ring of integers in the ww–adic completion KwK_{w} of KK and GvG_{v} is now viewed as G(Kw/Fv)G(K_{w}/F_{v}), where FvF_{v}, 𝒪v\mathcal{O}_{v} etc. have the obvious meanings. Also, let us note that the faithful flatness of the ring extension A[Gv]A[G]A[G_{v}]\subseteq A[G] combined with part (1) of the Proposition give us

FittA[Gv](𝒪K/we)=(Pv(0))=(Nv).{\rm Fitt}_{A[G_{v}]}(\mathcal{O}_{K}/w^{e})=(P_{v}(0))=(Nv).

Let K:=KIvK^{\prime}:=K^{I_{v}} denote the maximal sub-extension of K/FK/F, which is unramified at v.v. Let ww^{\prime} denote the prime in KK^{\prime} lying below ww. Since K/FK^{\prime}/F is unramified at vv, Proposition 5.2 leads to an equality of ideals

FittA[Gv/Iv](E(𝒪w/w))=(Pv(σv)){\rm Fitt}_{A[G_{v}/I_{v}]}(E(\mathcal{O}_{w^{\prime}}/w^{\prime}))=(P_{v}({\sigma}^{\prime}_{v}))

where σv{\sigma}^{\prime}_{v} denotes the Frobenius element of vv in K/FK^{\prime}/F. Since 𝒪w/w𝒪w/w\mathcal{O}_{w^{\prime}}/w^{\prime}\simeq\mathcal{O}_{w}/w (as Kw/KwK_{w}/K^{\prime}_{w^{\prime}} is totally ramified at ww^{\prime}), we also have an equality of ideals

(5) FittA[Gv/Iv](E(𝒪w/w))=(Pv(σv)){\rm Fitt}_{A[G_{v}/I_{v}]}(E(\mathcal{O}_{w}/w))=(P_{v}({\sigma}^{\prime}_{v}))

For simplicity, let e=|Iv|.e=|I_{v}|. Then, Proposition A.5.1. in [2] gives isomorphisms of A[Gv]A[G_{v}]–modules

ev(𝒪w/we)𝒪w/w and (1ev)(𝒪w/we)w/we.e_{v}(\mathcal{O}_{w}/w^{e})\simeq\mathcal{O}_{w}/w\hskip 10.0pt\text{ and }\hskip 10.0pt(1-e_{v})(\mathcal{O}_{w}/w^{e})\simeq w/w^{e}.

Since 𝒪w/we=ev(𝒪w/we)(1ev)(𝒪w/we)\mathcal{O}_{w}/w^{e}=e_{v}(\mathcal{O}_{w}/w^{e})\oplus(1-e_{v})(\mathcal{O}_{w}/w^{e}), the isomorphisms above lead to an equality of A[Gv]A[G_{v}]–ideals

(6) FittA[Gv](E(𝒪w/we))=FittA[Gv](E(𝒪w/w))FittA[Gv](E(w/we)).{\rm Fitt}_{A[G_{v}]}(E(\mathcal{O}_{w}/w^{e}))={\rm Fitt}_{A[G_{v}]}(E(\mathcal{O}_{w}/w))\cdot{\rm Fitt}_{A[G_{v}]}(E(w/w^{e})).
Remark 6.2.

The definition of Fitting ideals implies that if MM is a finitely generated A[Gv]A[G_{v}]–module, then

FittA[Gv](evM)=evFittA[Gv](M)(1ev)A[Gv],FittA[Gv]((1ev)M)=evA[Gv](1ev)FittA[Gv](M).{\rm Fitt}_{A[G_{v}]}(e_{v}M)=e_{v}{\rm Fitt}_{A[G_{v}]}(M)\oplus(1-e_{v})A[G_{v}],\qquad{\rm Fitt}_{A[G_{v}]}((1-e_{v})M)=e_{v}A[G_{v}]\oplus(1-e_{v}){\rm Fitt}_{A[G_{v}]}(M).

We will use equality (6) to prove (4). The obvious ring isomorphism evA[Gv]Av[Gv/Iv]e_{v}A[G_{v}]\simeq A_{v}[G_{v}/I_{v}] which sends evσvσve_{v}\sigma_{v}\to\sigma_{v}^{\prime} and equality (5) give

(7) FittA[Gv](E(𝒪w/w))=evP(σv)A[Gv](1ev)A[Gv].{\rm Fitt}_{A[G_{v}]}(E(\mathcal{O}_{w}/w))=e_{v}P(\sigma_{v})A[G_{v}]\oplus(1-e_{v})A[G_{v}].

Next, we will calculate FittA[Gv](E(w/we)).{\rm Fitt}_{A[G_{v}]}(E(w/w^{e})). For that, we need a definition and a couple of lemmas.

Definition 6.3.

Let RR be a commutative ring and MM a finitely presented RR–module. MM is called quadratically presented if there exists an n>0n>0 and an exact sequence of RR–modules

RnRnM0.R^{n}\to R^{n}\to M\to 0.

MM is called locally quadratically presented if M𝔪M_{\mathfrak{m}} is a quadratically presented R𝔪R_{\mathfrak{m}}–module for all 𝔪MSpec(R)\mathfrak{m}\in{\rm MSpec}(R).

The following computationally useful result is due to Johnston and Nickel. (See [8]).

Lemma 6.4 (Johnston–Nickel).

If RR is a commutative ring, CC is a locally quadratically presented RR and

0ABC00\to A\to B\to C\to 0

is an exact sequence of finitely presented RR–modules, then

FittR(B)=FittR(A)FittR(C).{\rm Fitt}_{R}(B)={\rm Fitt}_{R}(A)\cdot{\rm Fitt}_{R}(C).

A supply of locally quadratically presented A[Gv]A[G_{v}]–modules in the context at hand is given by the following Lemma, whose proof uses concepts of group cohomology discussed in detail in the Appendix of [2].

Lemma 6.5.

Under the above assumptions, the following hold.

  1. (1)

    The A[Gv]A[G_{v}]–submodule wiw^{i} of 𝒪w\mathcal{O}_{w} is GvG_{v}–cohomologically trivial (GvG_{v}–c.t.), for all i0i\geq 0.

  2. (2)

    For all j>i0j>i\geq 0, the A[Gv]A[G_{v}]–modules wi/wjw^{i}/w^{j} and E(wi/wj)E(w^{i}/w^{j}) are locally quadratically presented.

Proof.

(1) Fix i0i\geq 0. Since the A[Gv]A[G_{v}]–module wiw^{i} is annihilated by pp and p|Iv|p\nmid|I_{v}|, we have

Hr(Iv,wi)=0, for all r1.{\rm H}^{r}(I_{v},w^{i})=0,\text{ for all }r\geq 1.

Consequently, the inflation–restriction sequences in group–cohomology give us exact sequences of AA–modules

0Hr(Gv/Iv,(wi)Iv)Hr(Gv,wi)Hr(Iv,wi)Gv,0\to{\rm H}^{r}(G_{v}/I_{v},(w^{i})^{I_{v}})\to{\rm H}^{r}(G_{v},w^{i})\to{\rm H}^{r}(I_{v},w^{i})^{G_{v}},

for all r1r\geq 1. However, (wi)Iv=(w)i(w^{i})^{I_{v}}=(w^{\prime})^{i^{\prime}}, for some i0i^{\prime}\geq 0. Since Kw/FvK^{\prime}_{w^{\prime}}/F_{v} is unramified, we have isomorphisms of 𝒪v[Gv/Iv]\mathcal{O}_{v}[G_{v}/I_{v}]–modules

𝒪v[Gv/Iv]𝒪wπviwi.\mathcal{O}_{v}[G_{v}/I_{v}]\simeq\mathcal{O}_{w^{\prime}}\overset{\pi_{v}^{i^{\prime}}}{\longrightarrow}{w^{\prime}}^{i^{\prime}}.

where the second isomorphism is multiplication with πvi\pi_{v}^{i^{\prime}}, where πv\pi_{v} is a generator of vv (as an ideal of 𝒪v\mathcal{O}_{v}). This shows that the 𝒪v[Gv/Iv]\mathcal{O}_{v}[G_{v}/I_{v}]–module (wi)Iv(w^{i})^{I_{v}} is Gv/IvG_{v}/I_{v}–induced and therefore Gv/IvG_{v}/I_{v}–c.t. Now, the exact sequence above implies that

Hr(Gv,wi)=0, for all r1,{\rm H}^{r}(G_{v},w^{i})=0,\qquad\text{ for all }r\geq 1,

which shows that, indeed, wiw^{i} is GvG_{v}–c.t., for all i0i\geq 0.

(2) Part (1) implies that the A[Gv]A[G_{v}] modules w1/wjw^{1}/w^{j} are GvG_{v}–c.t. (since wjw^{j} and wiw^{i} are. Therefore, since AA is a PID, they have projective dimension at most 11 over A[Gv]A[G_{v}]. However, wi/wjw^{i}/w^{j} is finite and since AA is infinite, wi/wjw^{i}/w^{j} must have projective dimension exactly 11 over A[Gv]A[G_{v}]. Therefore, there are α𝔪,β𝔪\alpha_{\mathfrak{m}},\beta_{\mathfrak{m}}\in\mathbb{Z}, with 0α𝔪β𝔪0\leq\alpha_{\mathfrak{m}}\leq\beta_{\mathfrak{m}} and exact sequences of A[Gv]𝔪A[G_{v}]_{\mathfrak{m}}–modules

0A[Gv]α𝔪A[Gv]β𝔪(wi/wj)𝔪0,0\to A[G_{v}]^{\alpha_{\mathfrak{m}}}\to A[G_{v}]^{\beta_{\mathfrak{m}}}\to(w^{i}/w^{j})_{\mathfrak{m}}\to 0,

for all 𝔪MSpec(A[Gv])\mathfrak{m}\in{\rm MSpec}(A[G_{v}]). Again, by the finiteness of (wi/wj)(w^{i}/w^{j}), one concludes that α𝔪=β𝔪\alpha_{\mathfrak{m}}=\beta_{\mathfrak{m}}, which gives the desired local quadratic presentations, concluding the proof. ∎

Corollary 6.6.

Under the above assumptions, the following hold.

  1. (1)

    For all i>0i>0, the identity map induces an isomorphism of A[Gv]A[G_{v}]–modules

    wi/wi+1E(wi/wi+1).w^{i}/w^{i+1}\simeq E(w^{i}/w^{i+1}).
  2. (2)

    We have equalities of A[Gv]A[G_{v}]–ideals

    FittA[Gv]E(w/we)=FittA[Gv](w/we)=evA[Gv](1ev)NvA[Gv].{\rm Fitt}_{A[G_{v}]}E(w/w^{e})={\rm Fitt}_{A[G_{v}]}(w/w^{e})=e_{v}A[G_{v}]\oplus(1-e_{v})Nv\cdot A[G_{v}].
Proof.

(1) The identity map induces an isomorphism of 𝔽q[G]\mathbb{F}_{q}[G]–modules between the two modules in question. However, since τ(wi)wqiwi+1\tau(w^{i})\subseteq w^{qi}\subseteq w^{i+1}, τ\tau acts trivially on wi/wi+1w^{i}/w^{i+1}. Therefore, we have

ϕE(t)x=tx+a1τ(x)++arτr(x)=tx,for all xwi/wi+1,\phi_{E}(t)\cdot x=t\cdot x+a_{1}\tau(x)+...+a_{r}\tau^{r}(x)=t\cdot x,\qquad\text{for all }x\in w^{i}/w^{i+1},

which shows that the identity is also an isomorphism of 𝔽q[Gv][t]=A[Gv]\mathbb{F}_{q}[G_{v}][t]=A[G_{v}]–modules.

(2) Consider the short exact sequences of A[Gv]A[G_{v}]–modules

0wi/wi+1w/wi+1w/wi0,0E(wi/wi+1)E(w/wi+1)E(w/wi)0,0\to w^{i}/w^{i+1}\to w/w^{i+1}\to w/w^{i}\to 0,\qquad 0\to E(w^{i}/w^{i+1})\to E(w/w^{i+1})\to E(w/w^{i})\to 0,

for all i=1,,e1.i=1,\dots,e-1. Apply Lemma 6.5(2) and Lemma 6.4 to conclude that

FittA[Gv](w/wi+1)=FittA[Gv](w/wi)FittA[Gv](wi/wi+1),{\rm Fitt}_{A[G_{v}]}(w/w^{i+1})={\rm Fitt}_{A[G_{v}]}(w/w^{i})\cdot{\rm Fitt}_{A[G_{v}]}(w^{i}/w^{i+1}),
FittA[Gv]E(w/wi+1)=FittA[Gv]E(w/wi)FittA[Gv]E(wi/wi+1),{\rm Fitt}_{A[G_{v}]}E(w/w^{i+1})={\rm Fitt}_{A[G_{v}]}E(w/w^{i})\cdot{\rm Fitt}_{A[G_{v}]}E(w^{i}/w^{i+1}),

for all i=1,,e1.i=1,\dots,e-1.. Consequently, we obtain the following equalities of A[Gv]A[G_{v}]–ideals.

FittA[Gv](w/we)=i=1e1FittA[Gv](wi/wi+1),FittA[Gv]E(w/we)=i=1e1FittA[Gv]E(wi/wi+1).{\rm Fitt}_{A[G_{v}]}(w/w^{e})=\prod_{i=1}^{e-1}{\rm Fitt}_{A[G_{v}]}(w^{i}/w^{i+1}),\qquad{\rm Fitt}_{A[G_{v}]}E(w/w^{e})=\prod_{i=1}^{e-1}{\rm Fitt}_{A[G_{v}]}E(w^{i}/w^{i+1}).

Now, we combine the above equalities with the A[Gv]A[G_{v}]–module isomorphisms in part (1) to obtain

FittA[Gv]E(w/we)=FittA[Gv](w/we).{\rm Fitt}_{A[G_{v}]}E(w/w^{e})={\rm Fitt}_{A[G_{v}]}(w/w^{e}).

Now, since w/we=(1ev)(𝒪w/we)=(1ev)(𝒪K/we)w/w^{e}=(1-e_{v})(\mathcal{O}_{w}/w^{e})=(1-e_{v})(\mathcal{O}_{K}/w^{e}), we have

FittA[Gv]E(w/we)=evA[Gv](1ev)FittA[Gv]𝒪K/we=evA[Gv](1ev)NvA[Gv],{\rm Fitt}_{A[G_{v}]}E(w/w^{e})=e_{v}A[G_{v}]\oplus(1-e_{v}){\rm Fitt}_{A[G_{v}]}\mathcal{O}_{K}/w^{e}=e_{v}A[G_{v}]\oplus(1-e_{v})Nv\cdot A[G_{v}],

which concludes the proof of the Corollary. ∎

Now, we use Corollary 6.6(2) and equalities (6)–(7) to conclude that

FittA[Gv]E(𝒪w/w)=evP(σv)A[Gv](1ev)NvA[Gv].{\rm Fitt}_{A[G_{v}]}E(\mathcal{O}_{w}/w)=e_{v}P(\sigma_{v})\cdot A[G_{v}]\oplus(1-e_{v})Nv\cdot A[G_{v}].

However, note that if Pv(X)=Xr+ar1Xr1++a1Xr+a0P_{v}(X)=X^{r}+a_{r-1}X^{r-1}+\dots+a_{1}X^{r}+a_{0}, then Pv(0)=a0=ρNvP_{v}(0)=a_{0}=\rho\cdot Nv and therefore

Pv(evσv)=evPv(σv)+(1ev)Pv(0)=evPv(σv)+(1ev)ρNv.P_{v}(e_{v}\sigma_{v})=e_{v}P_{v}(\sigma_{v})+(1-e_{v})P_{v}(0)=e_{v}P_{v}(\sigma_{v})+(1-e_{v})\rho N_{v}.

Now, recalling that ρ𝔽q×\rho\in\mathbb{F}_{q}^{\times}, we combine the last two displayed equalities to conclude that

FittA[Gv]E(𝒪w/w)=Pv(evσv)A[Gv].{\rm Fitt}_{A[G_{v}]}E(\mathcal{O}_{w}/w)=P_{v}(e_{v}\sigma_{v})A[G_{v}].

This proves equality (4) and concludes the proof of Proposition 6.1.∎

7. The tt–motive and local 𝔽q\mathbb{F}_{q}–shtukas associated to E¯\overline{E}

This section is dedicated to the proof of a more precise version of Proposition 3.2, which will be stated below. The arguments below ensued from a lengthy and enlightening e-mail exchange [6] between the first author and Urs Hartl, to whom we would like to express our gratitude.

As before, E¯\overline{E} is the reduction modulo vv of the original Drinfeld module EE defined on AA and over 𝒪F\mathcal{O}_{F}. So, E¯\overline{E} is a Drinfeld module of rank rr, defined on AA, over the finite field 𝒪F/v=𝔽qdv\mathcal{O}_{F}/v=\mathbb{F}_{q^{d_{v}}}, and of characteristic w0=vAw_{0}=v\cap A. In §3, we denoted by hh the height of E¯\overline{E}, picked a prime v0MSpec(A){w0}v_{0}\in{\rm MSpec}(A)\setminus\{w_{0}\}, and considered

fE¯(X):=detAv0(XIrτ1Tv0(E¯)),gE¯(X):=detAw0(XIrhτ1Tw0(E¯)),f_{\overline{E}}(X):={\rm det}_{A_{v_{0}}}(X\cdot I_{r}-\tau_{1}\mid T_{v_{0}}(\overline{E})),\qquad g_{\overline{E}}(X):={\rm det}_{A_{w_{0}}}(X\cdot I_{r-h}-\tau_{1}\mid T_{w_{0}}(\overline{E})),

where τ1:=Frobqdv\tau_{1}:={\rm Frob}_{q^{d_{v}}} is the qdvq^{d_{v}}–power Frobenius endomorphism of E¯\overline{E}, acting naturally and linearly on its various Tate modules. In the same section, we stated the well known result that fE¯(X)f_{\overline{E}}(X) is independent on v0w0v_{0}\neq w_{0} and in fact fE¯(X)A[X]f_{\overline{E}}(X)\in A[X] (which will also be proved very briefly below) and claimed without a proof that gE¯(X)fE¯(X)g_{\overline{E}}(X)\mid f_{\overline{E}}(X) in Aw0[X]A_{w_{0}}[X]. The precise goal of this section is to prove the following.

Proposition 7.1.

Let Aw0¯\overline{A_{w_{0}}} be the absolute integral closure of Aw0A_{w_{0}}. Then the following hold.

  1. (1)

    The roots of gE¯(X)=0g_{\overline{E}}(X)=0 in Aw0¯\overline{A_{w_{0}}} are exactly those roots α\alpha of fE¯(X)=0f_{\overline{E}}(X)=0 which satisfy

    |α|w0=1,|\alpha|_{w_{0}}=1,

    where ||w0|\cdot|_{w_{0}} is the unique extension to Aw0¯\overline{A_{w_{0}}} of the normalized absolute value on Aw0A_{w_{0}}.

  2. (2)

    If α\alpha is such a root, then its multiplicities in gE¯(X)g_{\overline{E}}(X) and fE¯(X)f_{\overline{E}}(X) are the same.

  3. (3)

    In particular, gE¯(X)fE¯(X)g_{\overline{E}}(X)\mid f_{\overline{E}}(X) in Aw0[X].A_{w_{0}}[X].

Obviously, part (3) of the above Proposition is a direct consequence of parts (1) and (2). In order to prove parts (1) and (2), we need to introduce first the tt–motive M¯:=ME¯\overline{M}:=M_{\overline{E}} associated to E¯\overline{E}. We follow most of the notations and conventions in [3], Chapter 5. Let L1:=𝒪F/vL_{1}:=\mathcal{O}_{F}/v and let L:=L1¯=𝔽q¯L:=\overline{L_{1}}=\overline{\mathbb{F}_{q}} be a fixed algebraic closure. In what follows, 𝔾a\mathbb{G}_{a} denotes the additive affine line, viewed as a scheme over Spec(𝔽q){\rm Spec}(\mathbb{F}_{q}). We think of E¯\overline{E} as a functor from the category of L1L_{1}–algebras to the category of AA–modules

E¯:[L1–alg][A–mod],L𝔾a(L),\overline{E}:[\text{$L_{1}$--alg}]\longrightarrow[\text{$A$--mod}],\qquad L^{\prime}\to\mathbb{G}_{a}(L^{\prime}),

where 𝔾a(L)\mathbb{G}_{a}(L^{\prime}) is endowed with a natural AA–module structure via the 𝔽q\mathbb{F}_{q}–algebra (injective) morphism

AϕE¯L1{τ}L{τ}=End𝔽qL(𝔾a).A\overset{\phi_{\overline{E}}}{\longrightarrow}L_{1}\{\tau\}\subseteq L^{\prime}\{\tau\}={\rm End}_{\mathbb{F}_{q}}^{L^{\prime}}(\mathbb{G}_{a}).
Definition 7.2.

As LL is a perfect field containing L1L_{1} (the field of definition of E¯\overline{E}), we follow loc.cit. and define the tt–motive over LL associated to EE as the left L{τ}𝔽qA=L{τ}[t]L\{\tau\}\otimes_{\mathbb{F}_{q}}A=L\{\tau\}[t]–module

M¯(L):=Hom𝔽qL(E¯(L),𝔾a(L))=L{τ},\overline{M}(L):={\rm Hom}_{\mathbb{F}_{q}}^{L}(\overline{E}(L),\mathbb{G}_{a}(L))=L\{\tau\},

endowed with the left L{τ}𝔽qAL\{\tau\}\otimes_{\mathbb{F}_{q}}A–module structure given by

(λa)μ:=λμϕE¯(a), for all λL{τ},aA,μM¯(L).(\lambda\otimes a)\ast\mu:=\lambda\circ\mu\circ\phi_{\overline{E}}(a),\qquad\text{ for all }\lambda\in L\{\tau\},a\in A,\mu\in\overline{M}(L).
Remark 7.3.

It is important to note that the L{τ}[t]L\{\tau\}[t]–module M¯(L)\overline{M}(L) has some distinctive properties (see loc. cit. for proofs): First, it is obvious that M¯(L)\overline{M}(L) is a free L{τ}=(L{τ}1)L\{\tau\}=(L\{\tau\}\otimes 1)–module of rank 11 (which is the dimension of the tt–motive M¯(L)\overline{M}(L)) and (less obvious) that it is a free L[t]=(L𝔽qA)L[t]=(L\otimes_{\mathbb{F}_{q}}A)–module of rank rr (which is the rank of the tt–motive M¯(L)\overline{M}(L).) Second, it is important to note that since LL is perfect, τM¯(L)\tau\overline{M}(L) is an L{τ}[t]L\{\tau\}[t]–submodule of M¯(L)\overline{M}(L) and, as a consequence of the definition of ϕE¯\phi_{\overline{E}}, we have

(1tι(t)1)(M¯(L)/τM¯(L))=0,(1\otimes t-\iota(t)\otimes 1)(\overline{M}(L)/\tau\overline{M}(L))=0,

where i:AL1Li:A\to L_{1}\subseteq L is the obvious 𝔽q\mathbb{F}_{q}–algebra map of kernel w0w_{0}.

It is not difficult to check that the evaluation (perfect) pairing

E¯(L)×M¯(L)𝔾a(L),(e,μ)μ(e)\overline{E}(L)\times\overline{M}(L)\to\mathbb{G}_{a}(L),\qquad(e,\mu)\to\mu(e)

gives rise to an isomorphism of AA–modules

ξ:E¯(L)HomL{τ}[t](M¯(L),L((t1))/tL[t]),e[μi0μ(ϕE¯(ti)(e))ti¯],\xi:\overline{E}(L)\simeq{\rm Hom}_{L\{\tau\}[t]}(\overline{M}(L),L((t^{-1}))/tL[t]),\qquad e\to\big{[}\,\mu\to\overline{\sum_{i\geq 0}\mu(\phi_{\overline{E}}(t^{i})(e))\cdot t^{-i}}\,\big{]},

where τ\tau acts on L((t1))/tL[t]L((t^{-1}))/tL[t] by raising the coefficients of the Laurent series in question to the qq–th power and L[t]L[t] acts via multiplication. For every fAf\in A, this leads to a natural isomorphism of A/fA/f–modules

ξ[f]:E¯[f]HomL{τ}𝔽qA/f(M¯(L)/f,L[t]/fL[t]),\xi[f]:\overline{E}[f]\simeq{\rm Hom}_{L\{\tau\}\otimes_{\mathbb{F}_{q}}A/f}\big{(}\overline{M}(L)/f,\,L[t]/fL[t]\big{)},

after identifying L[t]/fL[t](L((t1))/tL[t])[f]L[t]/fL[t]\simeq(L((t^{-1}))/tL[t])[f] via the isomorphism ρ^tρ/f^\widehat{\rho}\to\widehat{t\rho/f}.

Now, we fix an arbitrary v0MSpec(A)v_{0}\in{\rm MSpec}(A) and let πv0A\pi_{v_{0}}\in A denote the monic generator of v0v_{0}. We let

Av0nr:=L^𝔽qAv0:=limn(L𝔽qA/v0n),M¯(L)v0:=M¯(L)^AAv0:=limn(M¯(L)AA/v0n).A^{nr}_{v_{0}}:=L\widehat{\otimes}_{\mathbb{F}_{q}}A_{v_{0}}:={\varprojlim_{n}}\,(L\otimes_{\mathbb{F}_{q}}A/v_{0}^{n}),\qquad\overline{M}(L)_{v_{0}}:=\overline{M}(L)\widehat{\otimes}_{A}A_{v_{0}}:={\varprojlim_{n}}\,(\overline{M}(L)\otimes_{A}A/v_{0}^{n}).

Note that if dv0:=[A/v0:𝔽q]d_{v_{0}}:=[A/v_{0}:\mathbb{F}_{q}], then we have natural isomorphisms of topological rings

Av0𝔽qdv0[[πv0]],Av0nrL[[πv0]]dv0.A_{v_{0}}\simeq{\mathbb{F}}_{q^{d_{v_{0}}}}[[\pi_{v_{0}}]],\qquad A^{nr}_{v_{0}}\simeq L[[\pi_{v_{0}}]]^{d_{v_{0}}}.

Further, note that since M¯(L)\overline{M}(L) is a free (L𝔽qA=L[t])(L\otimes_{\mathbb{F}_{q}}A=L[t])–module of rank rr (see the Remark above), then M¯(L)v0\overline{M}(L)_{v_{0}} is a free Av0nrA_{v_{0}}^{nr}–module of rank rr and, consequently, a free L[[πv0]]L[[\pi_{v_{0}}]]–module of rank rdv0rd_{v_{0}}. In addition, if we view Frobq{\rm Frob}_{q} as the canonical topological generator of Gal(L/𝔽q)=Gal(Av0nr/Av0)Gal(L/\mathbb{F}_{q})=Gal(A_{v_{0}}^{nr}/A_{v_{0}}), then the free Av0nrA_{v_{0}}^{nr}–module M¯(L)v0\overline{M}(L)_{v_{0}} is endowed with a Frobq{\rm Frob}_{q}–semilinear endomorphism, abusively denoted τ\tau, and given by

τ:=τ^1:M¯(L)^AAv0M¯(L)^AAv0.\tau:=\tau\widehat{\otimes}1:\overline{M}(L)\widehat{\otimes}_{A}A_{v_{0}}\longrightarrow\overline{M}(L)\widehat{\otimes}_{A}A_{v_{0}}.
Definition 7.4.

The data (M¯(L)v0,τ)(\overline{M}(L)_{v_{0}},\tau) consisting of the free Av0nrA_{v_{0}}^{nr}–module M¯(L)v0\overline{M}(L)_{v_{0}} of rank rr together with its Frobq{\rm Frob}_{q}–semilinear endomorphism τ\tau defined above is called the local 𝔽q\mathbb{F}_{q}–shtuka over LL associated to E¯\overline{E} at v0v_{0}.

The link between the local shtuka (M¯(L)v0,τ)(\overline{M}(L)_{v_{0}},\tau) and the Tate module Tv0(E¯)T_{v_{0}}(\overline{E}) is obtained by taking the projective limit as nn\to\infty of the isomorphisms ξ[πv0n]\xi[\pi_{v_{0}}^{n}] defined above, to get an isomorphism of Av0A_{v_{0}}–modules

ξv0nr:Tv0(E¯)HomAv0nr{τ}(M¯(L)v0,Av0nr), for all v0MSpec(A).{\xi}_{v_{0}}^{nr}:T_{v_{0}}(\overline{E})\simeq{\rm Hom}_{A_{v_{0}}^{nr}\{\tau\}}(\overline{M}(L)_{v_{0}},A_{v_{0}}^{nr}),\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

The following useful result is due to Hartl–Singh [7], building upon earlier work of Laumon [10].

Proposition 7.5.

For all v0MSpec(A)v_{0}\in{\rm MSpec}(A), the local 𝔽q\mathbb{F}_{q}–shtuka (M¯(L)v0,τ)(\overline{M}(L)_{v_{0}},\tau) over LL splits canonically as a direct sum of local 𝔽q\mathbb{F}_{q}–shtukas over LL

(M¯(L)v0,τ)=(M¯(L)v0et,τ)(M¯(L)v0nil,τ),(\overline{M}(L)_{v_{0}},\tau)=(\overline{M}(L)_{v_{0}}^{et},\tau)\oplus(\overline{M}(L)_{v_{0}}^{nil},\tau),

where M¯(L)v0et\overline{M}(L)_{v_{0}}^{et} is the maximal Av0nr{τ}A_{v_{0}}^{nr}\{\tau\}–submodule of M¯(L)v0\overline{M}(L)_{v_{0}} on which the restriction of τ\tau is bijective and M¯(L)v0nil\overline{M}(L)_{v_{0}}^{nil} is the maximal Av0nr{τ}A_{v_{0}}^{nr}\{\tau\}–submodule of M¯(L)v0\overline{M}(L)_{v_{0}} on which the restriction of τ\tau is topologically nilpotent (i.e. there exists an n>0n>0 such that τn(M¯(L)nil)πv0M¯(L)nil\tau^{n}(\overline{M}(L)^{nil})\subseteq\pi_{v_{0}}\overline{M}(L)^{nil}.)

Proof.

See the proofs of Propositions 2.7 and 2.9 in [7] and keep in mind that the field LL is perfect in our context. Also, note that in loc.cit. the authors work with the L[[πv0]]L[[\pi_{v_{0}}]]–module structures (local) rather than the Av0nrA_{v_{0}}^{nr}—module structures (semi-local), but the transition local/semi–local is seamless. ∎

In light of the above result, since obviously τ\tau acts as an isomorphism of Av0nrA_{v_{0}}^{nr} (i.e. (Av0nr,τ)(A_{v_{0}}^{nr},\tau) is an étale local 𝔽q\mathbb{F}_{q}–shtuka over LL), the isomorphism of Av0A_{v_{0}}–modules ξv0nr\xi_{v_{0}}^{nr} defined above can be rewritten as

ξv0nr:Tv0(E¯)HomAv0nr{τ}(M¯(L)v0et,Av0nr), for all v0MSpec(A).{\xi}_{v_{0}}^{nr}:T_{v_{0}}(\overline{E})\simeq{\rm Hom}_{A_{v_{0}}^{nr}\{\tau\}}(\overline{M}(L)^{et}_{v_{0}},A_{v_{0}}^{nr}),\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

The next crucial step is provided by the following extension to the case of GLn(Av0nr){\rm GL_{n}}(A_{v_{0}}^{nr}) of Lang’s well known theorem on GLn(L){\rm GL}_{n}(L) (see [9]).

Lemma 7.6 (Hartl, [6]).

Under the above assumptions, the following hold.

  1. (1)

    The map GLn(Av0nr)GLn(Av0nr){\rm GL}_{n}(A_{v_{0}}^{nr})\to{\rm GL_{n}}(A_{v_{0}}^{nr}) taking XFrobq(X)1XX\to{\rm Frob}_{q}(X)^{-1}\cdot X is surjective.

  2. (2)

    Any free Av0nrA_{v_{0}}^{nr}–module \mathcal{M} of finite rank nn, endowed with a bijective, Frobq{\rm Frob}_{q}–semilinear endomorphism tt satisfies the property that the standard map

    t=1Av0Av0nr,amam\mathcal{M}^{t=1}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\to\mathcal{M},\qquad a\otimes m\to am

    is an isomorphism of Av0nrA_{v_{0}}^{nr}–modules.

Proof.

Since we have a ring isomorphism Av0nrL[[πv0]]dv0A_{v_{0}}^{nr}\simeq L[[\pi_{v_{0}}]]^{d_{v_{0}}}, it suffices to prove part (1) for GLn(L[[πv0]]){\rm GL}_{n}(L[[\pi_{v_{0}}]]). So, given a matrix AGLn(L[[πv0]])A\in{\rm GL}_{n}(L[[\pi_{v_{0}}]]), i.e.

A=A0+A1πv0+, with A0GLn(L) and AiMn(L), for all i1,A=A_{0}+A_{1}\cdot\pi_{v_{0}}+\dots,\qquad\text{ with }A_{0}\in{\rm GL}_{n}(L)\text{ and }A_{i}\in M_{n}(L),\text{ for all }i\geq 1,

we need to find a matrix XGLn(L[[πv0]]X\in{\rm GL}_{n}(L[[\pi_{v_{0}}]] given by

X=X0+X1πv0+, with X0GLn(L) and XiMn(L), for all i1,X=X_{0}+X_{1}\cdot\pi_{v_{0}}+\dots,\qquad\text{ with }X_{0}\in{\rm GL}_{n}(L)\text{ and }X_{i}\in M_{n}(L),\text{ for all }i\geq 1,

such that the matrices XiX_{i} satisfy the relations

i=0mFrobq(Xmi)Ai=Xm, for all m0.\sum_{i=0}^{m}{\rm Frob}_{q}(X_{m-i})\cdot A_{i}=X_{m},\qquad\text{ for all }m\geq 0.

Lang’s theorem (see loc.cit.) implies that part (1) is true for GLn(L){\rm GL}_{n}(L), so we can find a matrix X0GLn(L)X_{0}\in{\rm GL}_{n}(L) satisfying the 0–th relation above. After multiplying the mm–th relation above to the right by A01Frobq(X0)1=X01A_{0}^{-1}{\rm Frob}_{q}(X_{0})^{-1}=X_{0}^{-1} we obtain the equivalent relation

XmX01=Frobq(XmX01)+i1mFrobq(Xmi)AiX01,X_{m}X_{0}^{-1}={\rm Frob}_{q}(X_{m}X_{0}^{-1})+\sum_{i\geq 1}^{m}{\rm Frob}_{q}(X_{m-i})\cdot A_{i}\cdot X_{0}^{-1},

which consists of one Artin–Schreier equation for each entry of XmX01X_{m}X_{0}^{-1}. Since LL is algebraically closed, these equations have solutions. Therefore, inductively, one can find matrices XmX_{m}, for all m0m\geq 0, as desired.

Part (2) follows immediately from part (1) in a standard way: take a basis e¯\overline{e} of \mathcal{M} over Av0nrA_{v_{0}}^{nr} and let AA be the matrix of tt in that basis. Let XGLn(Av0nr)X\in{\rm GL}_{n}(A_{v_{0}}^{nr}) such that A=Frobq(X)1XA={\rm Frob}_{q}(X)^{-1}\cdot X. Then e¯:=Xe¯\overline{e^{\prime}}:=X\cdot\overline{e} is an Av0nrA_{v_{0}}^{nr}-basis of \mathcal{M} which is contained in t=1\mathcal{M}^{t=1}. This concludes the proof. ∎

By applying the Lemma above to :=M¯(L)v0et\mathcal{M}:=\overline{M}(L)_{v_{0}}^{et} and t=τt=\tau, we conclude that we have the following natural isomorphisms of Av0nr{τ}A_{v_{0}}^{nr}\{\tau\}–modules

M¯(L)v0et(M¯(L)v0et)τ=1Av0Av0nr, for all v0MSpec(A).\overline{M}(L)_{v_{0}}^{et}\simeq(\overline{M}(L)_{v_{0}}^{et})^{\tau=1}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr},\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

The above isomorphism leads to a further isomorphism of Av0A_{v_{0}}–modules

HomAv0nr{τ}(M¯(L)v0et,Av0nr)HomAv0((M¯(L)v0et)τ=1,Av0), for all v0MSpec(A),{\rm Hom}_{A_{v_{0}}^{nr}\{\tau\}}(\overline{M}(L)^{et}_{v_{0}},A_{v_{0}}^{nr})\simeq{\rm Hom}_{A_{v_{0}}}((\overline{M}(L)^{et}_{v_{0}})^{\tau=1},A_{v_{0}}),\qquad\text{ for all }v_{0}\in{\rm MSpec}(A),

which, if composed with the map ξv0nr\xi_{v_{0}}^{nr} gives an isomorphism of Av0A_{v_{0}}–modules

ξv0:Tv0(E¯)HomAv0((M¯(L)v0et)τ=1,Av0), for all v0MSpec(A).\xi_{v_{0}}:T_{v_{0}}(\overline{E})\simeq{\rm Hom}_{A_{v_{0}}}((\overline{M}(L)^{et}_{v_{0}})^{\tau=1},A_{v_{0}}),\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

This prompts the following.

Definition 7.7.

The first v0v_{0}–adic étale cohomolgy group of E¯\overline{E} is defined by

Het1(E¯,Av0):=(M¯(L)v0et)τ=1, for all v0MSpec(A).{\rm H}^{1}_{et}(\overline{E},\,A_{v_{0}}):=(\overline{M}(L)^{et}_{v_{0}})^{\tau=1},\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

Note that the maps ξv0\xi_{v_{0}} lead to the following Av0A_{v_{0}}–module isomorphisms.

Tv0(E¯):=HomAv0(Tv0(E),Av0)Het1(E¯,Av0), for all v0MSpec(A).T_{v_{0}}(\overline{E})^{\ast}:={\rm Hom}_{A_{v_{0}}}(T_{v_{0}}(E),A_{v_{0}})\simeq{\rm H}^{1}_{et}(\overline{E},\,A_{v_{0}}),\qquad\text{ for all }v_{0}\in{\rm MSpec}(A).

Further, the first v0v_{0}–adic crystalline cohomology group of E¯\overline{E} is defined by

Hcris1(E¯,Av0nr):=M¯(L)v0.{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr}):=\overline{M}(L)_{v_{0}}.

Note that for all v0MSpec(A)v_{0}\in{\rm MSpec}(A) we have isomorphisms and inclusions of Av0nrA_{v_{0}}^{nr}–modules

Tv0(E¯)Av0Av0nr=Het1(E¯,Av0)Av0Av0nrHcris1(E¯,Av0nr)etHcris1(E¯,Av0nr).T_{v_{0}}(\overline{E})^{\ast}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}={\rm H}^{1}_{et}(\overline{E},A_{v_{0}})\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\simeq{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr})^{et}\subseteq{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr}).

The following holds at primes v0v_{0} different from the characteristic of E¯\overline{E}. (See [3], Chapter 5 as well.)

Lemma 7.8.

If v0v_{0} is an element in MSpec(A){\rm MSpec}(A) different from the characteristic w0w_{0} of E¯\overline{E}, then

M¯(L)v0et=M¯(L)v0,(M¯(L)v0)τ=1Av0Av0nrM¯(L)v0.\overline{M}(L)_{v_{0}}^{et}=\overline{M}(L)_{v_{0}},\qquad(\overline{M}(L)_{v_{0}})^{\tau=1}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\simeq\overline{M}(L)_{v_{0}}.

In other words, τ\tau is bijective on M¯(L)v0\overline{M}(L)_{v_{0}} and we have canonical isomorphisms of Av0nr[τ1]A_{v_{0}}^{nr}[\tau_{1}]–modules

Tv0(E¯)Av0Av0nrHet1(E¯,Av0)Av0Av0nrHcris1(E¯,Av0nr).T_{v_{0}}(\overline{E})^{\ast}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\simeq{\rm H}^{1}_{et}(\overline{E},A_{v_{0}})\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\simeq{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr}).
Proof.

(sketch) It is easy to show that since τϕE¯(πv0n)\tau\nmid\phi_{\overline{E}}(\pi_{v_{0}}^{n}) in L{τ}L\{\tau\}, τ\tau is injective and therefore bijective on the finite dimensional LL–vector spaces M¯(L)/πv0n\overline{M}(L)/\pi_{v_{0}}^{n}, for all n1n\geq 1. The bijection of τ\tau on M¯(L)v0\overline{M}(L)_{v_{0}} is obtained now by taking the projective limit as nn\to\infty. ∎

Now, we have all the necessary ingredients to prove parts (1) and (2) of Proposition 7.1.

Proof.

First, take v0MSpec(A){w0}v_{0}\in{\rm MSpec}(A)\setminus\{w_{0}\} and observe that, based on the previous Lemma and Definition, we have canonical isomorphisms of Av0nr[τ1]A_{v_{0}}^{nr}[\tau_{1}]–modules

Tv0(E¯)Av0Av0nrHcris1(E¯,Av0nr)=M¯(L)^AAv0M¯(L)L𝔽qA(L^𝔽qAv0)M¯(L)L[t]Av0nr.T_{v_{0}}(\overline{E})^{\ast}\otimes_{A_{v_{0}}}A_{v_{0}}^{nr}\simeq{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr})=\overline{M}(L)\widehat{\otimes}_{A}A_{v_{0}}\simeq\overline{M}(L)\otimes_{L\otimes_{\mathbb{F}_{q}}A}(L\widehat{\otimes}_{\mathbb{F}_{q}}A_{v_{0}})\simeq\overline{M}(L)\otimes_{L[t]}A_{v_{0}}^{nr}.

As a consequence, from the definition of fE¯f_{\overline{E}}, we have

fE¯(X)=detAv0(XIrτ1Tv0(E¯))=detAv0nr(XIrτ1Hcris1(E¯,Av0nr))=detL[t](XIrτ1M¯(L)).f_{\overline{E}}(X)={\rm det}_{A_{v_{0}}}(XI_{r}-\tau_{1}\mid T_{v_{0}}(\overline{E})^{\ast})={\rm det}_{A_{v_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{v_{0}}^{nr}))={\rm det}_{L[t]}(XI_{r}-\tau_{1}\mid\overline{M}(L)).

The last equality proves that fE¯(X)f_{\overline{E}}(X) is independent of v0v_{0} and that it has coefficients in L[t]L[t]. Further, if one applies the analogues of Lemmas 7.6 and 7.8 to the finite, étale 𝔽q\mathbb{F}_{q}–shtukas M¯(L)/v0n\overline{M}(L)/{v_{0}}^{n} over LL (see [7]), one concludes that fE¯(X)f_{\overline{E}}(X) has coefficients in Av0A_{v_{0}}. Since L[t]Av0=AL[t]\cap A_{v_{0}}=A (intersection viewed inside Av0nrA_{v_{0}}^{nr}), fE¯(X)f_{\overline{E}}(X) has coefficients in AA, as stated before.

Now, from the definitions, we also have similar natural isomorphisms of Aw0nr[τ1]A_{w_{0}}^{nr}[\tau_{1}]–modules

Hcris1(E¯,Aw0nr)=M¯(L)^AAw0M¯(L)L𝔽qA(L^𝔽qAw0)M¯(L)L[t]Aw0nr.{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})=\overline{M}(L)\widehat{\otimes}_{A}A_{w_{0}}\simeq\overline{M}(L)\otimes_{L\otimes_{\mathbb{F}_{q}}A}(L\widehat{\otimes}_{\mathbb{F}_{q}}A_{w_{0}})\simeq\overline{M}(L)\otimes_{L[t]}A_{w_{0}}^{nr}.

Therefore, when combining these with the second note in Definition 7.7, we obtain equalities

fE¯(X)\displaystyle f_{\overline{E}}(X) =detL[t](XIrτ1M¯(L))=detAw0nr(XIrτ1M¯(L)L[t]Aw0nr)\displaystyle={\rm det}_{L[t]}(XI_{r}-\tau_{1}\mid\overline{M}(L))={\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid\overline{M}(L)\otimes_{L[t]}A_{w_{0}}^{nr})
=detAw0nr(XIrτ1Hcris1(E¯,Aw0nr))\displaystyle={\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr}))
=detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)et)detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)nil)\displaystyle={\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{et})\cdot{\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{nil})
=detAw0(XIrhτ1Tw0(E¯))detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)nil)\displaystyle={\rm det}_{A_{w_{0}}}(XI_{r-h}-\tau_{1}\mid T_{w_{0}}(\overline{E})^{\ast})\cdot{\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{nil})
=gE¯(X)detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)nil).\displaystyle=g_{\overline{E}}(X)\cdot{\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{nil}).

Firstly, this shows that gE¯(X)g_{\overline{E}}(X) divides fE¯(X)f_{\overline{E}}(X) in Aw0nr[X]A_{w_{0}}^{nr}[X]. However, since gE¯(X),fE¯(X)g_{\overline{E}}(X),f_{\overline{E}}(X) are both in the polynomial ring Aw0[X]=Aw0nr[X]τ=1A_{w_{0}}[X]=A_{w_{0}}^{nr}[X]^{\tau=1}, this divisibility holds in Aw0[X]A_{w_{0}}[X]. Secondly, note that

gE¯(X)=detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)et),fE¯(X)/gE¯(X)=detAw0nr(XIrτ1Hcris1(E¯,Aw0nr)nil).g_{\overline{E}}(X)={\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{et}),\qquad f_{\overline{E}}(X)/g_{\overline{E}}(X)={\rm det}_{A_{w_{0}}^{nr}}(XI_{r}-\tau_{1}\mid{\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr})^{nil}).

Since τ1\tau_{1} (which is a power of τ\tau) acts bijectively on the étale piece and topologically nilpotently on the nilpotent piece of Hcris1(E¯,Aw0nr){\rm H}^{1}_{cris}(\overline{E},A_{w_{0}}^{nr}), the roots of gE¯g_{\overline{E}} are all w0w_{0}–adic units and the roots of fE¯(X)/gE¯(X)f_{\overline{E}}(X)/g_{\overline{E}}(X) are in the maximal ideal of an absolute integral closure Aw0¯\overline{A_{w_{0}}} of Aw0A_{w_{0}}. This concludes the proof. ∎

8. Final thoughts: the more general case of pure, abelian tt-modules

Most of the techniques developed in this paper can be extended to the more general case of pure, abelian tt–modules, as indicated briefly below. The very general case of abelian tt–modules, whose equivariant LL–functions are the main object of study in [4], will be treated in upcoming work.

With notations as in §2, let Mn(𝒪F)M_{n}(\mathcal{O}_{F}) be the ring of n×nn\times n matrices with entries in 𝒪F\mathcal{O}_{F}, for a fixed n1n\geq 1.

Definition 8.1.

A tt–module EE of dimension nn, defined over 𝒪F\mathcal{O}_{F} is an 𝔽q\mathbb{F}_{q}-algebra morphism

ϕE:AMn(𝒪F{τ}),ϕE(t)=M0τ0+M1τ++Mτ,\phi_{E}:A\to M_{n}(\mathcal{O}_{F}\{\tau\}),\qquad\phi_{E}(t)=M_{0}\tau^{0}+M_{1}\tau+...+M_{\ell}\tau^{\ell},

where MiMn(𝒪F)M_{i}\in M_{n}(\mathcal{O}_{F}), 1\ell\geq 1, M0M_{\ell}\neq 0 and (M0tIn)n=0(M_{0}-t\cdot I_{n})^{n}=0.

For any 𝒪F{τ}\mathcal{O}_{F}\{\tau\}–module XX, one denotes by

E(X),LieE(X)E(X),\qquad{\rm Lie}_{E}(X)

the additive group XnX^{\oplus n}, endowed with the AA–module structure given by ϕE\phi_{E} and evτ=0ϕE{\rm ev}_{\tau=0}\circ\phi_{E} (evaluation of ϕE(a)\phi_{E}(a) at τ=0\tau=0, for all aAa\in A), respectively.

Note that Drinfeld modules defined on AA and over 𝒪F\mathcal{O}_{F} are the 11–dimensional tt–modules defined in §2 above. However, unlike the 11–dimensional case, where the rank of the Drinfeld module is simply equal to \ell, in the case of higher dimensional tt–modules, introducing the notion of rank is much more subtle. We indicate briefly how this is done below. This will lead to the more restrictive notion of an abelian tt–module.

Let F¯{\overline{F}} be the separable closure of FF. As in §7, we let E(F¯):=𝔾an(F¯)E(\overline{F}):=\mathbb{G}_{a}^{n}(\overline{F}) (where 𝔾a\mathbb{G}_{a} is viewed as a group scheme over 𝔽q\mathbb{F}_{q}), endowed with the AA–module structure induced by the structural morphism

ϕE:AMn(𝒪F){τ}Mn(F¯){τ}=End𝔽qF¯(E(F¯)).\phi_{E}:A\to M_{n}(\mathcal{O}_{F})\{\tau\}\subseteq M_{n}(\overline{F})\{\tau\}={\rm End}^{\overline{F}}_{\mathbb{F}_{q}}(E(\overline{F})).

As in §7, we define the tt–motive over F¯\overline{F} associated to EE by

M(F¯):=ME(F¯):=Hom𝔽qF¯(E(F¯),𝔾a(F¯)),M(\overline{F}):=M_{E}(\overline{F}):={\rm Hom}_{\mathbb{F}_{q}}^{\overline{F}}(E(\overline{F}),\mathbb{G}_{a}(\overline{F})),

endowed with the (F¯{τ}𝔽qA)=F¯{τ}[t](\overline{F}\{\tau\}\otimes_{\mathbb{F}_{q}}A)=\overline{F}\{\tau\}[t]–module structure given by

(λa)μ:=(λμϕE(a)), for all aA,λF¯{τ},μME(F¯).(\lambda\otimes a)\ast\mu:=(\lambda\circ\mu\circ\phi_{E}(a)),\qquad\text{ for all }a\in A,\lambda\in\overline{F}\{\tau\},\mu\in M_{E}(\overline{F}).

As an F¯{τ}\overline{F}\{\tau\}–module, ME(F¯)M_{E}(\overline{F}) is clearly free of rank nn (which is called the dimension of the motive). However, as an L[t]L[t]–module this is not finitely generated, in general, which brings us to the following.

Definition 8.2.

The tt–module EE and the tt–motive ME(F¯)M_{E}(\overline{F}) are called abelian if ME(F¯)M_{E}(\overline{F}) is a finitely generated and (automotically, see [3] Chapter 5) free F¯[t]\overline{F}[t]–module. In that case, the rank over F¯[t]\overline{F}[t] of ME(F¯)M_{E}(\overline{F}) is called the rank of EE and also called the rank of ME(F¯)M_{E}(\overline{F}).

It is not hard to see that if EE is a Drinfeld module of classical rank rr, then ME(F¯)M_{E}(\overline{F}) is F¯[t]\overline{F}[t]–free of rank rr. So, Drinfeld modules are 11–dimensional abelian tt–modules and the two notions of rank coincide.

Assuming now that EE is an abelian tt–module as above, of dimension nn and rank rr, a prime vMSpec(𝒪F)v\in{\rm MSpec}(\mathcal{O}_{F}) is called a prime of good reduction for EE if the tt–module E¯\overline{E} of dimension nn, defined over 𝒪F/v\mathcal{O}_{F}/v by

ϕE¯:AϕEMn(𝒪F{τ})Mn(𝒪F/v{τ})\mathcal{\phi}_{\overline{E}}:A\overset{\phi_{E}}{\longrightarrow}M_{n}(\mathcal{O}_{F}\{\tau\})\twoheadrightarrow M_{n}(\mathcal{O}_{F}/v\{\tau\})

(composing ϕE\phi_{E} with the usual mod vv reduction map) is abelian of the same rank rr as EE.

Now, assume that EE is an abelian tt–module as above, of dimension nn and rank rr, and let v0MSpec(A)v_{0}\in{\rm MSpec}(A) and vMSpec(𝒪F)v\in{\rm MSpec}(\mathcal{O}_{F}), with vv0v\nmid v_{0} and such that EE has good reduction at vv. One can show without difficulty (see [3] Chapter 5) that one has isomorphisms of Av0A_{v_{0}}–modules, for all v0MSpec(A).v_{0}\in{\rm MSpec}(A).

E[v0m](Av0/v0m)r,Tv0(E):=limmE[v0m]Av0r.E[v_{0}^{m}]\simeq(A_{v_{0}}/v_{0}^{m})^{r},\qquad T_{v_{0}}(E):=\varprojlim_{m}E[v_{0}^{m}]\simeq A_{v_{0}}^{r}.

Also, the G(F¯/F)G(\overline{F}/F)–representation Tv0(E)T_{v_{0}}(E) is unramified at all vMSpec(𝒪F)v\in\text{MSpec}(\mathcal{O}_{F}) with vv0v\nmid v_{0} and the polynomials

Pv(X):=detAv0(XIrσ(v)Tv0(E))P_{v}(X):={\rm det}_{A_{v_{0}}}(X\cdot I_{r}-\sigma(v)\mid T_{v_{0}}(E))

are independent of vv and have coefficients in AA. (See [4] and the references therein.) It is also true (see [4]) that if vv is tamely ramified in K/FK/F, then the finite A[G]A[G]–modules E(𝒪K/v)E(\mathcal{O}_{K}/v) and LieE(𝒪K/v){\rm Lie}_{E}(\mathcal{O}_{K}/v) are free over 𝔽q[G]\mathbb{F}_{q}[G] and therefore the monic polynomials

|E(𝒪K/v)|G,|LieE(𝒪K/v)|G|E(\mathcal{O}_{K}/v)|_{G},\qquad|{\rm Lie}_{E}(\mathcal{O}_{K}/v)|_{G}

are well defined in A[G]=𝔽q[G][t]A[G]=\mathbb{F}_{q}[G][t]. The abelian tt–module analogue of Theorem 2.3 is the equality

Pv(σvev)Pv(0)=|E(𝒪K/v)|G|LieE(𝒪K/v)|G.\frac{P_{v}(\sigma_{v}e_{v})}{P_{v}(0)}=\frac{|E(\mathcal{O}_{K}/v)|_{G}}{|{\rm Lie}_{E}(\mathcal{O}_{K}/v)|_{G}}.

The above equality was used in [4] to prove an equivariant Tamagawa number formula for the value at 0 of the GG–equivariant LL–function associated to the data (E,F/K)(E,F/K) as above.

In order to prove the above displayed equality in the general case of abelian tt–modules EE, one can try to extend to the higher dimensional context the techniques developed in the previous sections of this paper. And, indeed, it is not very difficult to see that the techniques developed in §4, §6 and §7 extend in full to the higher dimensional context. In particular, Propositions 3.2 and 7.1 hold true in general. However, a higher dimensional analogue of Proposition 3.1 does not hold true, in general. A slightly weaker version of Proposition 3.1 (see Theorem 5.6.10 in [3]), which is strong enough for our purposes, holds true for a smaller class of abelian tt–modules EE and primes of good reduction vMSpec(𝒪F)v\in{\rm MSpec}(\mathcal{O}_{F}), namely those for which E¯\overline{E} (the reduction of EE mod vv) is, in addition, a pure tt–module. The notion of purity is technical and will not be described here. Instead, we refer the interested reader to §5.5 in [3]. To summarize, if we assume purity of E¯\overline{E}, then the last displayed equality above can be proved with the techniques developed in this paper. If, for example, EE is a tensor product of Drinfeld modules (in the category of tt–modules), then purity is achieved. The more general case of arbitrary abelian tt–motives will be treated in an upcoming paper.

References

  • [1] T. Beaumont, On equivariant class number formulas for tt–modules, https://arxiv.org/abs/2110.15696, 36 pages.
  • [2] J. Ferrara, N. Green, Z. Higgins, and C.D. Popescu, An equivariant Tamagawa number formula for Drinfeld modules and applications, J. Algebra and Number Theory 16 (2022), no. 9, 2215–2264.
  • [3] D. Goss, Basic structures of function field arithmetic, Springer Berlin, Heidelberg, 1998.
  • [4] N. Green and C.D. Popescu, An equivariant Tamagawa number formula for tt–modules and applications, https://arxiv.org/abs/2206.03541 (June 2022), 27 pages.
  • [5] C. Greither and C.D. Popescu, The Galois module structure of \ell-adic realizations of Picard 1-motives and applications, Int. Math. Res. Not. 2012 (2012), no. 5, 986–1036.
  • [6] U. Hartl, Private email communication with C.D. Popescu, (2024).
  • [7] U. Hartl and R.K. Singh, Local shtukas and divisible local Anderson modules, Canad. J. Math. 71 (2019), no. 5, 1163–1207. MR 4010425
  • [8] H. Johnston and A. Nickel, Noncommutative Fitting invariants and improved annihilation results, J. Lond. Math. Soc. (2) 88 (2013), no. 1, 137–160. MR 3092262
  • [9] S. Lang, Algebraic groups over finite fields, Amer. J. Math. 78 (1956), 555–563. MR 86367
  • [10] G. Laumon, Cohomology of Drinfeld modular varieties. Part I, Cambridge Studies in Advanced Mathematics, vol. 41, Cambridge University Press, Cambridge, 1996, Geometry, counting of points and local harmonic analysis. MR 1381898
  • [11] H. Matsumura, Commutative algebra, The Benjamin/Cummings Publishing Co. Inc., Second Edition, 1980.
  • [12] N. Ramachandran, Some Fitting ideal computations in Iwasawa theory over \mathbb{Q} and in the theory of Drinfeld modules, Ph.D. thesis, University of California San Diego (2024).
  • [13] L. Taelman, Special LL-values of Drinfeld modules, Ann. of Math. (2) 175 (2012), no. 1, 369–391. MR 2874646