Review pending

Proposition 23.7.1 · Conditional place-weight summation

Proposition 23.7.1. (Conditional place-weight summation). Let F⊂EF\subset E be number fields. Fix unsquared absolute values extending those on Qp\mathbb Q_p and the usual real absolute value, and put cv(F)=[Fv:Qv]/[F:Q]c_v(F)=[F_v:\mathbb Q_v]/[F:\mathbb Q]. Suppose the completion-degree identity ∑w∣v[Ew:Fv]=[E:F]\sum_{w\mid v}[E_w:F_v]=[E:F] is supplied. Suppose real numbers ava_v and awa_w satisfy aw=ava_w=a_v for w∣vw\mid v, and their weighted families are finitely supported or absolutely summable. Then ∑wcw(E)aw=∑vcv(F)av.\sum_w c_w(E)a_w=\sum_v c_v(F)a_v.

Argument in the manuscript

Proof. We first establish the completion-degree identity under the stated number-field hypotheses.

Finite places. Let FF be a field with a specified real-valued nonarchimedean absolute value vv. Let H=F^vH=\widehat F_v, let E/FE/F be a finite extension of degree nn, and let w∣vw\mid v mean an absolute value on EE restricting to this very absolute value on FF. Write EwE_w for the resulting completion. Set C=H⊗FEC=H\otimes_F E. There is a canonical indexing of the Artinian local factors CwC_w of CC by w∣vw\mid v, with canonical valued residue field EwE_w. In particular n=∑w∣vℓCw(Cw)[Ew:H].n=\sum_{w\mid v}\ell_{C_w}(C_w)[E_w:H]. For separable E/FE/F all lengths are one.

A basis of E/FE/F gives dim⁡HC=n\dim_H C=n. Its ideals are finite-dimensional HH-subspaces, so descending chains stabilize. By [StacksAlg] (Tag 00JB, Lemma 10.53.6) it is the finite product of its localizations at its maximal ideals. For a maximal ideal PP let D=C/PD=C/P, a finite field extension of HH. NAG-001450 supplies the unique extension of the absolute value of HH, completeness of DD, and equivalence with a finite-dimensional coordinate norm, without separability. The map E→DE\to D, a↦1⊗a mod Pa\mapsto 1\otimes a\bmod P, is a unital field embedding, so defines an extension wPw_P of vv. The image of EE is dense in DD: the images of an FF-basis eie_i of EE span DD over HH, and each ∑ihiei\sum_i h_i e_i is a limit of ∑ifiei\sum_i f_i e_i with fi∈Ff_i\in F. Indeed the error is bounded by max⁡i∣hi−fi∣ ∣ei∣wP\max_i |h_i-f_i|\,|e_i|_{w_P}. Thus DD identifies isometrically with EwPE_{w_P}, fixing EE.

Conversely, for any extending ww, the inclusion of FF into EwE_w extends isometrically to HH. Multiplication gives a map C→EwC\to E_w. Its image is a finite-dimensional HH-domain, hence a field: multiplication by each nonzero element is an injective linear endomorphism and therefore surjective. With its induced norm this field is complete by NAG-001450, hence closed in EwE_w. It contains the dense copy of EE and consequently is all of EwE_w. Its kernel is maximal. Starting from PP, the isometry from the previous paragraph agrees with both structure maps on EE and on HH (on HH by density of FF); thus the resulting kernel is exactly PP. Starting from ww, the norm on the residue field is the induced norm by uniqueness, so this construction recovers ww. This proves the asserted bijection without counting embeddings. The algebra here is formed over the original, possibly incomplete, FF; it is finite-dimensional over HH and already complete for a coordinate norm.

The simple modules in a composition series of CwC_w are its residue field. Each therefore contributes exactly [Ew:H][E_w:H] to the HH-dimension. Adding the contributions of the full local factors gives the formula. To check separability, take a finite tower of simple generators for E/FE/F. At each step the minimal polynomial of the new generator over the intermediate field divides a separable polynomial over FF. It is coprime to its derivative. Starting with HH, if the tensor algebra for the intermediate extension is a product of fields, tensoring with the next simple extension quotients a polynomial ring over each factor by the image of that minimal polynomial. The Bezout identity with its derivative survives in the field. The distinct monic irreducible factors are pairwise coprime, and the Chinese remainder map gives a product of fields again. Induction proves the length-one assertion. In characteristic zero every irreducible nonconstant polynomial is separable, since its derivative is nonzero and of smaller degree.

At a finite place vv of a number field FF over pp, use the absolute value extending the specified one on Qp\mathbb Q_p. The same result for F/QF/\mathbb Q shows that Fv/QpF_v/\mathbb Q_p is finite. For w∣vw\mid v in a finite extension E/FE/F the identification above uses this Fv⊂EwF_v\subset E_w isometrically. Since the number fields have characteristic zero, the length-one case gives ∑w∣v[Ew:Fv]=[E:F]\sum_{w\mid v}[E_w:F_v]=[E:F] at each finite place.

Archimedean places. Let F⊂EF\subset E be number fields, n=[E:F]n=[E:F], and vv an absolute value on FF extending the usual absolute value on Q\mathbb Q. Count each absolute value ww on EE extending this fixed vv once. With the usual unsquared moduli on all archimedean completions, one has Fv⊗FE≃∏w∣vEw,∑w∣v[Ew:Fv]=n.F_v\otimes_F E \simeq \prod_{w\mid v}E_w,\qquad \sum_{w\mid v}[E_w:F_v]=n. If Fv=RF_v=\mathbb R and the factors comprise rr real and ss complex completions, then r+2s=nr+2s=n. If Fv=CF_v=\mathbb C, there are nn complex factors, each of degree one over FvF_v.

Elementary field input. For a nonconstant complex polynomial PP, the leading term implies ∣P(z)∣→∞|P(z)|\to\infty as ∣z∣→∞|z|\to\infty. A sequence approaching its infimum can therefore be taken in a bounded square. Successive subdivision into four closed subsquares, each time retaining one containing infinitely many sequence terms, gives a convergent subsequence by completeness of the real coordinates. Continuity gives a minimizer z0z_0. If P(z0)≠0P(z_0)\ne0, write P(z0+h)/P(z0)=1+chk+O(∣h∣k+1)P(z_0+h)/P(z_0)=1+c h^k+O(|h|^{k+1}) with k≥1k\ge1 and c≠0c\ne0. Choose δ\delta among 1,−1,i,−i1,-1,i,-i with Re⁡(cδ)<0\operatorname{Re}(c\delta)<0. It has a unit-modulus kk-th root uu: writing k=2abk=2^a b with bb odd, the bb-th power permutes those four units, and successive square roots of a unit vv are available since (1+v)/∣1+v∣(1+v)/|1+v| squares to vv for v≠−1v\ne-1, and i2=−1i^2=-1. For small positive tt, ∣P(z0+tu)/P(z0)∣2=1+2Re⁡(cδ)tk+O(tk+1)<1,|P(z_0+tu)/P(z_0)|^2 =1+2\operatorname{Re}(c\delta)t^k+O(t^{k+1})<1, a contradiction. Division by a linear factor and induction show that every complex polynomial splits. Conjugation of roots then shows that each irreducible real polynomial is linear or quadratic with a pair of nonreal conjugate roots. A quotient by such a quadratic is C\mathbb C over R\mathbb R, by evaluation at either root: a nonreal root and 1 already span C\mathbb C over R\mathbb R.

Completion factors. First take any number field DD, with base Q\mathbb Q and H=RH=\mathbb R. More generally let D/BD/B be a finite extension of number fields and fix a completion HH of BB already identified isometrically with R\mathbb R or C\mathbb C. Write A=H⊗BDA=H\otimes_B D. Adjoin a finite generating list for D/BD/B one element at a time. At each step the minimal polynomial over the preceding field is separable: in characteristic zero it is coprime to its derivative. Over each field factor reached so far the Bezout identity with its derivative persists. Factoring the resulting polynomial and applying the Chinese remainder identity for coprime factors gives a product of fields. By the preceding paragraph these fields are real or complex over a real factor, and complex over a complex factor. Starting with HH yields a product decomposition of AA; its HH-dimension is [D:B][D:B] by a basis. In particular all local lengths here are one.

Each factor carries the usual modulus, independent of the choice between conjugate identifications with C\mathbb C. In fact any multiplicative absolute value ν\nu on C\mathbb C extending the usual absolute value on R\mathbb R satisfies ν(i)=1\nu(i)=1, hence ν(z)≤∣Re⁡z∣+∣Im⁡z∣≤2∣z∣\nu(z)\le|\operatorname{Re}z|+|\operatorname{Im}z|\le2|z|. Apply this inequality to zjz^j and let j→∞j\to\infty; applying it also to z−1z^{-1} gives ν(z)=∣z∣\nu(z)=|z|. Thus each factor restricts to the specified absolute value on HH.

The image of DD in a factor spans it over HH; it is dense because the finitely many coefficients of a linear combination can be approximated in BB. The factor is complete, so is the completion of DD for the induced ww. Conversely, for any extending ww, the inclusion of BB in DwD_w extends isometrically to HH. Multiplication defines A→DwA\to D_w. Exactly one primitive factor idempotent maps to 1 in this field, so the image is the corresponding factor. Its norm is its usual one by the preceding uniqueness argument (immediate for a real factor); it is therefore complete and closed. It contains dense DD, hence equals DwD_w. These assignments are inverse: the completion isometry fixing DD fixes the prescribed image of HH by density of BB, hence gives the same kernel in AA. This identifies factors with places themselves, including exactly one factor per complex-conjugate pair over a real place.

Apply the argument first to F/QF/\mathbb Q and R\mathbb R to obtain the claimed alternatives for FvF_v. Apply it again to E/FE/F with H=FvH=F_v. Taking FvF_v-dimensions gives precisely the two counts above.

These two calculations give the asserted degree identity for every place in the proposition. At infinity all moduli are unsquared.

The tower formula gives cw(E)=cv(F)[Ew:Fv]/[E:F]c_w(E)=c_v(F)[E_w:F_v]/[E:F]. Sum first over the finite set of ww above each vv and use the completion-degree identity above. Absolute summability permits regrouping. At the archimedean base place Qv=R\mathbb Q_v=\mathbb R; a complex completion has degree two and its absolute value remains unsquared. ◻

Citation

Non-Archimedean Geometry Foundations, Tag NAG-900470.

Proof in chapter