Continuous, single-column reading. Equations, tables, proofs, references and appendices stay in document order.
Contents · 16 sections
RESEARCH MANUSCRIPT
Prepared for Higher Categorical Semantics
Adeles and the Product Formula
An introductory higher-categorical account
Abstract. The adelic product formula states that the local additive Haar modules of a nonzero global field element multiply to one. This introduction builds adeles and ideles as filtered colimits of products, recovers units by a pullback, and constructs the adelic norm by regrouping the places over each place of the base field. The determinant, viewed as a symmetric monoidal functor to graded lines, gives the local-to-global norm comparison. These constructions reduce the proof for all global fields to prime factorization over Q and degree bookkeeping over Fq(t). A second proof uses the exact sequence of the additive adelic quotient and its Haar measures. The idele-class quotient and the adelic complex of a regular curve exhibit the groupoid and derived aspects of the presentation.
A global fieldK is a finite extension of Q or of Fp(t) for a prime p. A place is an equivalence class of nontrivial absolute values, where equivalence means that they induce the same topology. We write VK for the set of places, VK,∞ for the finite set of archimedean places, and Kv for the completion at v. At a nonarchimedean place, Kv is a complete discretely valued field with finite residue field. Its valuation ring, maximal ideal, and residue field size are Ov={x:ordv(x)≥0},mv,qv=#(Ov/mv), with ordv(πv)=1 for a uniformizer πv. At an archimedean place Kv is R or C. Each pair of conjugate complex embeddings contributes one place. These are the standard place and completion theorems for global fields [1, §13.1].
We use the normalized local module∣⋅∣v on Kv×. It is the factor by which a scalar dilates additive Haar measure. Equivalently, for a=0 it is ∣a∣v=⎩⎨⎧qv−ordv(a),∣a∣usual,∣a∣usual2,v nonarchimedean,Kv=R,Kv=C. Section 3 derives all three cases from Haar measure. The square at a complex place reflects its real dimension.
The proofs use products, filtered colimits, and pullbacks in spaces or in an ∞-topos. Ordinary sets and groups are regarded as discrete objects when we discuss their algebraic structure. When topology matters, we specify it explicitly and give a sheaf model that retains it. The arithmetic input is supplied by the completion and norm theorems and by calculations on the two rational base fields.
Adeles and ideles
Restricted products
Let J be a set of indices, J∞⊂J a finite subset, and Uj→Xj a family of maps in a category with products and filtered colimits. Only the indices outside J∞ require a choice of Uj. Write F(J) for the poset of finite S⊇J∞. Define RJ(X,U)=S∈F(J)limj∈S∏Xj×j∈/S∏Uj. In an ∞-category the products and colimit are the corresponding ∞-categorical ones. The transition from S to T⊇S uses Uj→Xj for j∈T∖S. If XS denotes the stage in parentheses, the universal property is Map(RJ(X,U),Y)≃F(J)oplimMap(XS,Y). Thus a map out of the restricted product is given by maps on the stages with homotopy-coherent compatibility. For inclusions of discrete objects in spaces (2) is the usual set of tuples belonging to Uj outside a finite set. The poset is filtered by finite union, and filtered colimits of spaces preserve 0-truncation and commute with finite limits [4, Thm. 9.1.5.1].
Definition 1. The adeles and ideles of K are AKIK=∏′v∈VK(Kv,Ov),=∏′v∈VK(Kv×,Ov×), with every archimedean place included in each finite stage. We give every stage its product topology and the union the topology for which a subset is open exactly when its intersection with each stage is open.
For example the adele stages over Q have the form R×∏p∈SfinQp×∏p∈/SfinZp. An adele can have infinitely many nonzero coordinates. The finite restriction concerns the places where a coordinate fails to be integral. Similarly, an idele has nonzero coordinates everywhere and is a unit at almost every nonarchimedean place.
At a nonarchimedean place both Ov and Ov× are compact and open in their respective local groups. It follows that the stages are open, the additive adele group and the idele group are locally compact, and the coordinatewise algebra operations are continuous in their respective topologies. In fact two inputs lie in one common stage, and on each stage the operations are continuous products of local operations. The unrestricted archimedean and finitely many nonarchimedean factors are locally compact, and the remaining product is compact. Coordinate projections separate points, so both groups are Hausdorff. See [2, §25.2] and [3, §26.1].
Units as an inverse graph
Proposition 2. The unit group of the underlying ring AK is IK. With topology, IK is the inverse graph {(x,y)∈AK×AK:xy=1} under the identification x↦(x,x−1).
Proof. For a commutative ring object, the object of units is the pullback of multiplication R×R→R along the unit 1→R. Filtered colimits commute with these finite limits in an ∞-topos [5, Prop. 6.4.5.9]. Units also commute with products, since the pullback condition is coordinatewise. Applied to the adele stages this gives the claimed unit group. In tuples, both x and x−1 must be integral at almost all places. In the inverse graph, intersecting with AK,S×AK,S gives the idele stage IK,S with its product topology. These are open and cover the graph. ◻
Over Q, let u(p) be the idele equal to p at p and to 1 everywhere else. Along the primes in increasing order it converges to 1 in the adele topology, because a basic neighborhood restricts only finitely many coordinates within the stage integral at every finite place. It stays outside the open idele stage requiring a unit at every prime. Thus the inverse graph is needed to obtain the idele topology from the adele ring.
A sheaf model for the topology
On the site of profinite sets with finite jointly surjective covering families, a space T defines the set-valued sheaf P↦Cont(P,T), viewed as a 0-truncated object in the ∞-topos of space-valued sheaves. This is the usual condensed-set construction [6, §1]. Products are computed on sections. For either restricted product above, a continuous map from a profinite P into the union has compact image and therefore lands in one stage. Indeed, the open stages cover the image, and a finite subcover can be absorbed by taking the union of its finite indexing sets. The resulting sectionwise filtered colimit is a sheaf, since the set-valued sheaf condition for a finite covering family is a finite equalizer condition. Consequently (2) also presents both restricted products as sheaves in this model. The finite-limit proof of Proposition 2 then applies to these sheaves and recovers the topological inverse graph.
Regrouping places
Lemma 3. Let p:W→V have finite fibers with W∞=p−1(V∞). There is a canonical equivalence RW(X,U)Xv′≃RV(X′,U′),=p(w)=v∏Xw,Uv′=p(w)=v∏Uw.
Proof. The stages indexed by p−1(S), with S finite and containing V∞, are cofinal. For a finite stage T on the W side, the S satisfying T⊆p−1(S) form a poset with least element p(T)∪V∞. At each such stage, the product over W is the iterated product over the fibers and then over V, by the universal property of products. The equivalences respect the transition maps. For topological spaces with open inclusions the same stagewise argument identifies the restricted product topologies, since the cofinal stages give their open sets. ◻
In particular, compatible fiberwise maps of pairs (Xv′,Uv′)→(Yv,Bv) induce a map on restricted products. The regrouping takes place inside each product stage, so it works for infinite sets of places without an interchange law for filtered colimits and infinite products.
Norms and local modules
Determinants and completions
For a finite field extension L/F and a∈L×, the field norm is NL/F(a)=detF(ma), where ma is multiplication by a on L as an F-vector space. This makes NL/F a group homomorphism. The determinant is functorial on the groupoid of finite-dimensional vector spaces and linear isomorphisms. With direct sum on the source and tensor product on the target it is symmetric monoidal with values in the Picard groupoid of Z-graded lines: E↦(⋀dimEE,dimE). A Picard groupoid is a symmetric monoidal groupoid whose objects are invertible under tensor product up to isomorphism. Interchanging lines of degrees m,n contributes the sign (−1)mn. This is the sign obtained when the corresponding blocks of wedge factors are interchanged. The exterior power construction also commutes with extension of scalars [7, 8].
Proposition 4 (Norm under completion). Let L/F be a finite separable extension of global fields. For each place v of F there is a canonical isomorphism L⊗FFv≃w∣v∏Lw. Under F↪Fv, NL/F(a)=w∣v∏NLw/Fv(a)(a∈L×).
Proof. Choose a primitive element for L/F with minimal polynomial h. Over Fv, the distinct irreducible factors of h give a product of field quotients by the Chinese remainder theorem. Each factor carries the unique extending local absolute-value class and the image of L is dense in it. For density, the image spans the factor over Fv and F is dense in Fv. Conversely the map to each completion Lw factors through one of these fields. Its image is complete and contains the dense image of L, so it is all of Lw. Thus the factors are indexed by the extensions of v to L [1, Thm. 13.5]. The map in (3) is induced by the completion maps, giving its independence of h.
After base change, multiplication by a is the direct sum of the local multiplications. Base change takes the determinant line of L to that of L⊗FFv. The direct sum is sent to the tensor product of the local lines. The scalar on their tensor product is the product of the local scalars, giving (4). Any sign in reordering the lines cancels when conjugating a line automorphism by that reordering. ◻
Haar measure and the local norm
Let μ be a nonzero additive Haar measure of a local field M. A scalar a∈M× acts by an additive homeomorphism. Uniqueness of Haar measure gives a positive factor dM(a) defined by μ(aE)=dM(a)μ(E) for measurable E. It is independent of the chosen measure and is multiplicative because dilations compose. In a nonarchimedean field, a unit preserves the valuation ring and a uniformizer sends it to a subgroup of index q. Taking the valuation ring to have measure 1 gives dM(a)=q−ord(a). Real or complex scalar multiplication has real determinant of absolute value ∣a∣usual or ∣a∣usual2, respectively. This proves (1).
Lemma 5 (Local norm compatibility). If M/E is a finite extension of local fields, then dM(a)=dE(NM/E(a)).
Proof. Choose an E-basis of M and use product Haar measure on E[M:E]. For a linear automorphism of En the Haar dilation factor is dE of its determinant. For an elementary shear this follows from translation invariance in one coordinate and Fubini’s theorem. Coordinate permutations preserve product measure and have determinant ±1, of module 1. Diagonal maps scale each coordinate by its own module. Gaussian elimination decomposes an invertible matrix into these transformations. Apply the result to ma and use the independence of Haar normalization. ◻
For nonarchimedean M/E, with residue degree f and ramification index e, qM=qEf. Comparing exponents in Lemma 5 gives ordE(NM/E(a))=fordM(a). Here f weights the norm on arbitrary elements. For a base scalar b∈E× one has ordM(b)=eordE(b) and dM(b)=dE(b)ef. In particular the local norm maps local units to local units. For C/R it sends z to zz, consistent with the squared modulus.
The adelic comparison
First we check that elements of a global field give diagonal ideles. A number field is finite separable over Q. In positive characteristic, let Fq be the algebraic closure of Fp in K. This is finite, since for every finite constant subfield k′⊂K the degree of k′(t)/Fp(t) over Fp(t) is [k′:Fp], bounded by [K:Fp(t)]. A separating transcendence basis over the perfect field Fq gives a parameter s with K/Fq(s) finite separable [9]. The initial parameter t may give an inseparable extension, so we use the separating parameter in the proof.
Lemma 6 (Finite support). For a∈K×, both a and a−1 are integral at all but finitely many nonarchimedean places of K.
Proof. Choose a rational base field F as above and a monic polynomial over F for each of a and a−1. At almost all nonarchimedean places of F their coefficients are integral, by factorization of rational numbers or rational functions. They are then integral at each place of K lying above those base places. If a root of a monic polynomial with integral coefficients had negative valuation, its highest-degree term would have strictly smaller valuation than every other term, which prevents the sum from vanishing. Finitely many places of K lie above the finitely many excluded base places by Proposition 4. ◻
Thus K× embeds by a diagonal homomorphism ΔK:K×↪IK. The diagonal ring map into AK includes 0 as well. For general ideles set μK:IK⟶R>0,μK(x)=v∈VK∏∣xv∣v. Each stage map uses only the finitely many indices of that stage. These maps are continuous homomorphisms and compatible, so the universal property gives a continuous homomorphism on the whole idele group. For a finite separable extension, Lemma 3 and the unit property of the local norms give the idele norm(NL/FI(x))v=w∣v∏NLw/Fv(xw). The norms are continuous as determinants, and a fiber over a place outside the finite stage contributes a unit. Therefore (7) is a continuous homomorphism. The two compatibilities are μLNL/FI∘ΔL=μF∘NL/FI,=ΔF∘NL/F. The first is Lemma 5 multiplied over fibers, using finite support. The second is Proposition 4.
The adelic product formula
Theorem 7. For every global field K and a∈K×, v∈VK∏∣a∣v=1. Equivalently μK∘ΔK is the trivial character.
Proof. By (8) and (9), for any finite separable extension L/F we have μL(ΔLa)=μF(ΔFNL/F(a)). The rational base fields used in Lemma 6 therefore suffice.
Over Q, write a=±∏ppnp with only finitely many np nonzero. The p-factor in the product is p−np and the real factor is ∏ppnp.
Over Fq(t), the finite places correspond to monic irreducibles P∈Fq[t], with residue size qdegP. The remaining place is infinity, with residue size q and ord∞(t)=−1. Write a=c∏PPnP for c∈Fq×. Its finite product is q−∑PnPdegP, and ord∞(a)=−∑PnPdegP. Hence the factor at infinity is q∑PnPdegP. This proves the claim in both cases and (11) completes the proof. ◻
Example 8 (A complex place and a ramified prime). In Q(i) take a=(2+i)/(1+i). Its only nonzero finite valuations are 1 at (2+i) and −1 at (1+i). The residue sizes are 5 and 2, as follows from the indices of the principal ideals in Z[i]. The finite factors are 1/5 and 2. At the single complex place, ∣a∣usual2=5/2. The product is (1/5)⋅2⋅(5/2)=1. In particular, although 2 ramifies with index two, the generator 1+i has valuation one and local module 1/2.
Example 9 (A place of degree two). In F5(t) take a=(t2+2)/(t−1)3. The numerator is irreducible because 3 is a nonsquare in F5. Its place has module 5−2. The factors at t−1 and infinity are 53 and 5−1, respectively. They give 5−2535−1=1.
The idele-class character
The algebraic idele-class group is CK=IK/ΔK(K×). It is the set of components of the action groupoid IK//K×: an arrow labelled a sends x to ΔK(a)x. Multiplication of objects and of arrow labels makes this groupoid a Picard groupoid. The action is free, so its isotropy groups are trivial and it is equivalent to the discrete groupoid on CK. Theorem 7 says that μK is constant on these orbits. Equivalently μK sends every arrow to an identity in the discrete Picard groupoid with group of objects R>0. It thereby gives a symmetric monoidal functor on the action groupoid. With the quotient topology on CK, the induced character is continuous by the universal property of the topological quotient.
An additive proof by Haar measures
The local module is defined through additive Haar measure. We can therefore prove the product formula by constructing a Haar measure on AK preserved by K×. The construction uses the additive quotient AK/K.
Proposition 10. The diagonal copy of K is closed and discrete in AK, and AK/K is compact.
Proof. First let F be Q or Fq(t), with B=Z or Fq[t], respectively. For x∈AF choose a nonzero d∈B such that dxv is integral at every finite polynomial or prime place of B. The Chinese remainder theorem gives b∈B with b≡dxv(moddOv) at all places dividing d. Here B/(πr)≃Ov/πrOv for the corresponding prime π of B. Then x−b/d is integral at every place of B. In the number-field base case subtract an integer to put its real component in [0,1]. In the function-field base case subtract its polynomial part at infinity to put its infinity component in t−1Fq[[t−1]]. These operations yield representatives in a compact product of local sets. For discreteness, all coordinates at places of B may be required to be integral and the infinity coordinate to lie in (−1,1) or t−1Fq[[t−1]], respectively. This open neighborhood intersects the diagonal in 0, since an element integral at every place of B lies in B.
For K/F finite separable of degree n, choose an F-basis ei and its trace-dual basis ei∗, using nondegeneracy of the trace pairing. Away from finitely many v, all these basis elements are integral at each w∣v by Lemma 6. Under (3), the coefficient of ei in x=(xw)w∣v is w∣v∑TrKw/Fv(ei∗xw). This follows by base change of the trace of multiplication. For integral xw the result is integral, since traces of integral elements lie in the integrally closed ring Ov. Conversely, integral coefficients give integral xw by integrality of the ei. The basis thus identifies the local lattices Ovn and ∏w∣vOw outside finitely many places. Regrouping as in Lemma 3, these local linear homeomorphisms give AFn≃AK as additive topological groups and identify Fn with K. This gives discreteness and compactness of the quotient for K. A discrete subgroup of a Hausdorff topological group is closed. ◻
For the short exact sequence of locally compact abelian groups 0⟶K⟶AK⟶AK/K⟶0, use counting measure on K and probability Haar measure on the compact quotient. Quotient integration gives the Haar measure on AK determined by these choices. For a continuous compactly supported function f, its integral is the integral over AK/K of ∑b∈Kf(x+b), where x is any representative of the coset. Multiplication by a∈K× permutes the summation group and preserves probability Haar measure on the quotient. Therefore it preserves the resulting Haar measure on AK.
On a restricted product stage, the restriction of a Haar measure is proportional to the product of local Haar measures, with the integral factors given probability measure. Choose a stage containing every place at which a is a nonunit. Multiplication by a is an automorphism of this stage and has dilation factor ∏v∣a∣v, computed on product sets that are unrestricted within the unit-preserving integral factors. It follows that the product is 1. This is the measure argument behind Tate’s proof [10, Thm. 4.3.1].
Functoriality is explicit in this argument: the Haar measures on a locally compact abelian group form an R>0-torsor, and quotient integration combines the subgroup and quotient torsors into the middle one. The counting and probability choices give a section fixed by the action of K×.
At the level of connective algebraic K-theory, the same sequence gives a nullhomotopy for the map induced by adelic extension of scalars [11, Rem. 0.3]. For a finite-dimensional K-vector space E, tensor the additive sequence with E and give each term the topology of a finite power of K, AK, or AK/K, respectively. Change of basis acts continuously, so this defines a short exact sequence of exact functors to the exact category of second-countable locally compact abelian groups. Its short exact sequences use closed embeddings and their topological quotients. Additivity identifies the middle functor’s map of K-theory spectra with the sum of the endpoint maps. The endpoints factor through countable discrete and second-countable compact groups, respectively. Countable direct sums in the first category and countable products in the second yield exact endofunctors T with T≃id⊕T. By additivity their K-theory spectra are contractible, giving the asserted nullhomotopy.
Curves and adelic totalization
Let X be a regular integral Noetherian scheme of dimension one, with function field K and set of closed points ∣X∣. Write Ox for the completed local ring at a closed point and Kx=Frac(Ox). There is one generic point, and every strict specialization chain goes from it to a closed point. The additive normalized adelic complex of OX is K⊕x∈∣X∣∏Oxd∏′x∈∣X∣(Kx,Ox), in cohomological degrees 0 and 1, with d(a,(bx))=(a−bx)x. The terms come from the generic point, the closed points, and the two-point flags. The complex represents RΓ(X,OX), the derived global sections [12, Ex. 2.2]. Equivalently it is the homotopy fiber of d as a map of degree-zero objects in the derived ∞-category of abelian groups.
For the dimension-one calculation, let K be the sheaf of rational functions. For each closed point let ix:{x}↪X be its inclusion. There is a flasque resolution 0⟶OX⟶K⟶x∈∣X∣⨁ix∗(Kx/Ox)⟶0. At a closed point, DVR approximation identifies K/OX,x with Kx/Ox. Each rational function has only finitely many poles on a Noetherian curve. The last sheaf is the sheaf of finitely supported local principal parts on opens, because all opens are quasi-compact. Its restrictions are surjective by extension by zero. The generic stalk has no principal parts since any finite support can be deleted from a neighborhood. Taking global sections gives the complex [K→⨁xKx/Ox]. Complex (12) maps to this complex by projecting off ∏xOx in degree zero and reducing coordinates modulo Ox in degree one. The kernel is a copy of ∏xOx mapped by minus the identity to itself and is acyclic. This identifies the two complexes.
For a smooth, proper, geometrically integral curve over Fq, the closed points index all the places of its function field. Set Div(X)deg(x∑nx[x])=x∈∣X∣⨁Z[x],=x∑nx[κ(x):Fq]. The homomorphism divI:IK→Div(X) sends u to ∑xordx(ux)[x], and logμK(u)=−(logq)deg(divIu). For diagonal a, this divisor is the principal divisor of a. Thus the product formula says exactly that a principal divisor has degree zero. Geometrically, for nonconstant a the map a:X→PFq1 is finite flat, and the zero and pole divisors are its fibers over 0 and ∞. Their degrees both equal the rank of the finite flat map, with lengths accounting for multiplicity even in the inseparable case [13, §42.18]. Constants have zero divisor.
There is also a groupoid version of the divisor description. The action groupoid Div(X)//K× uses translation by principal divisors. Its components are divisor classes and the automorphisms of each object are the global units Fq×. This records the constant scalars that act trivially on divisors. The functor sending D to OX(D) identifies it with the Picard groupoid of line bundles and their isomorphisms. An arrow D→D+div(a) is sent to multiplication by a−1. Every line bundle has a nonzero rational section defining a divisor, and every isomorphism between these subsheaves of the rational-function sheaf acts at the generic point by a unique rational scalar. The valuation conditions on this scalar identify exactly the arrows in the action groupoid. The identifications respect tensor products. By (13), degree defines a symmetric monoidal functor from this Picard groupoid to discrete Z.
For X=SpecOK of a number field, the closed points give the finite adeles in (12). The arithmetic product formula also uses the archimedean places, with the real and complex modules of (1). More generally, the Beilinson cosimplicial adelic ring AX∙ packages iterated completion along flags on a Noetherian scheme. Groechenig’s adelic descent theorem gives Perf(X)≃TotPerf(AX∙), where Perf denotes the ∞-category of perfect complexes and Tot the limit of the cosimplicial diagram [14, Thm. 3.1 in the arXiv version]. The calculation above identifies the degree-zero and degree-one additive terms in the regular-curve case.
Conclusion
The product formula factors into three operations that respect different structures. Regrouping the restricted product gives the domain for the idele norm. The symmetric monoidal determinant identifies the global norm after completion, and additive Haar dilation fixes the local normalizations. The rational base calculations then prove triviality on the diagonal. In the curve case the same triviality has the geometric form of degree zero, while the adelic complex exhibits how the local rings and the generic field enter derived global sections.
Preparation disclosure. The manuscript was drafted and typeset with AI assistance.
References
A. V. Sutherland. Global fields and the product formula. MIT 18.785, Lecture 13, 2021. Lecture notes.
A. V. Sutherland. The ring of adeles, strong approximation. MIT 18.785, Lecture 25, 2021. Lecture notes.
A. V. Sutherland. The idele group, profinite groups, and infinite Galois theory. MIT 18.785, Lecture 26, 2021. Lecture notes.
J. Lurie. Exactness of filtered colimits. Kerodon, Theorem 9.1.5.1.
J. Lurie. Higher Topos Theory. Annals of Mathematics Studies 170, Princeton University Press, 2009. Author’s copy.
P. Scholze. Lectures on condensed mathematics. Bonn lectures, 2019, updated 2026. Lecture notes.
F. F. Knudsen and D. Mumford. The projectivity of the moduli space of stable curves I. Preliminaries on det and Div. Mathematica Scandinavica, 39(1), 19–55, 1976. Published article.
The Stacks Project Authors. Determinants. Tag 0FJ9.
B. Conrad. Separating transcendence basis. Stanford University. Course handout.
J. T. Tate. Fourier Analysis in Number Fields and Hecke’s Zeta-Functions. PhD thesis, Princeton University, 1950. Thesis scan.
D. Clausen. A K-theoretic approach to Artin maps. arXiv:1703.07842, 2017.
O. Braunling, M. Groechenig, and J. Wolfson. Geometric and analytic structures on the higher adèles. Research in the Mathematical Sciences, 3, article 22, 2016. Open-access article.
The Stacks Project Authors. Principal divisors and pushforward. Section 42.18.
M. Groechenig. Adelic descent theory. Compositio Mathematica, 153(8), 1706–1746, 2017. Published article, open preprint.
Complete flattened LaTeX source
\documentclass[10pt,letterpaper,twocolumn]{article}
\usepackage[T1]{fontenc}
\usepackage{lmodern,microtype}
\usepackage{amsmath,amssymb,amsthm,mathtools}
\usepackage[letterpaper,top=.73in,bottom=.72in,left=.73in,right=.73in,
columnsep=.29in,headsep=.19in,footskip=.31in]{geometry}
\usepackage{xcolor,fancyhdr,balance,needspace}
\usepackage[colorlinks=true,linkcolor=blue!44!black,citecolor=blue!44!black,
urlcolor=blue!48!black,pdftitle={Adeles and the Product Formula},
pdfsubject={An introductory higher-categorical account of the adelic product formula}]{hyperref}
\definecolor{ink}{HTML}{18334A}
\definecolor{softink}{HTML}{536474}
\setlength{\headheight}{12pt}
\setlength{\parindent}{1em}
\setlength{\parskip}{.5pt}
\setlength{\emergencystretch}{1.3em}
\allowdisplaybreaks[1]
\raggedbottom
\pagestyle{fancy}
\fancyhf{}
\fancyhead[L]{\small\sffamily\color{softink}Adeles and the Product Formula}
\fancyhead[R]{\small\sffamily\color{softink}Research manuscript}
\fancyfoot[C]{\small\sffamily\color{softink}\thepage}
\renewcommand{\headrulewidth}{.25pt}
\makeatletter
\renewcommand{\section}{\@startsection{section}{1}{\z@} {-2.6ex plus -.6ex minus -.2ex}{1.2ex plus .2ex} {\large\bfseries\sffamily\color{ink}}}
\renewcommand{\subsection}{\@startsection{subsection}{2}{\z@} {-1.5ex plus -.3ex minus -.1ex}{.65ex plus .1ex} {\normalsize\bfseries\sffamily\color{ink}}}
\makeatother
\newtheorem{theorem}{Theorem}[section]
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newcommand{\A}{\mathbb A}
\newcommand{\Q}{\mathbb Q}
\newcommand{\R}{\mathbb R}
\newcommand{\C}{\mathbb C}
\newcommand{\Z}{\mathbb Z}
\newcommand{\F}{\mathbb F}
\newcommand{\I}{\mathbb I}
\newcommand{\OO}{\mathcal O}
\newcommand{\Rprod}{\mathop{\prod\nolimits'}}
\newcommand{\fin}{\mathrm{fin}}
\newcommand{\vabs}[1]{\lvert #1\rvert}
\newcommand{\um}{\mathfrak m}
\DeclareMathOperator{\Nm}{N}
\DeclareMathOperator{\Hom}{Map}
\DeclareMathOperator{\ord}{ord}
\DeclareMathOperator{\RP}{R}
\DeclareMathOperator{\Frac}{Frac}
\DeclareMathOperator{\Perf}{Perf}
\DeclareMathOperator{\Tot}{Tot}
\DeclareMathOperator{\Div}{Div}
\begin{document}
\twocolumn[
\begin{@twocolumnfalse}
\noindent{\small\sffamily\bfseries\color{ink}RESEARCH MANUSCRIPT}
\hfill{\small\sffamily\color{softink}Prepared for Higher Categorical Semantics\par}
\vspace{.60em}
\noindent{\color{ink}\rule{\textwidth}{.65pt}\par}
\vspace{1.05em}
{\noindent\fontsize{26}{29}\selectfont\bfseries\color{ink}
Adeles and the Product Formula\par}
\vspace{.33em}
{\noindent\large\color{softink}An introductory higher-categorical account\par}
\vspace{1.05em}
\begin{minipage}{.92\textwidth}
\small
\textbf{Abstract.}
The adelic product formula states that the local additive Haar modules of
a nonzero global field element multiply to one. This introduction builds
adeles and ideles as filtered colimits of products, recovers units by
a pullback, and constructs the adelic norm by regrouping the places
over each place of the base field. The determinant, viewed as a
symmetric monoidal functor to graded lines, gives the local-to-global
norm comparison. These constructions reduce the proof for all global
fields to prime factorization over $\Q$ and degree bookkeeping over
$\F_q(t)$. A second proof uses the exact sequence of the additive
adelic quotient and its Haar measures. The idele-class quotient and
the adelic complex of a regular curve exhibit the groupoid and
derived aspects of the presentation.
\smallskip
\textbf{Keywords.} Global fields, adeles, ideles, product formula,
determinant norm, restricted products, adelic descent.
\end{minipage}
\par\vspace{1.45em}
\end{@twocolumnfalse}
]
\thispagestyle{plain}
\section{Global fields and their places}
A \emph{global field} $K$ is a finite extension of $\Q$ or of
$\F_p(t)$ for a prime $p$. A \emph{place} is an equivalence class
of nontrivial absolute values, where equivalence means that they induce
the same topology. We write $V_K$ for the set of places, $V_{K,\infty}$
for the finite set of archimedean places, and $K_v$ for the completion at
$v$. At a nonarchimedean place, $K_v$ is a complete discretely valued
field with finite residue field. Its valuation ring, maximal ideal,
and residue field size are
\[
\OO_v=\{x:\ord_v(x)\geq0\},\quad
\um_v,\quad q_v=\#(\OO_v/\um_v),
\]
with $\ord_v(\pi_v)=1$ for a uniformizer $\pi_v$.
At an archimedean place $K_v$ is $\R$ or $\C$. Each pair of
conjugate complex embeddings contributes one place. These are the
standard place and completion theorems for global fields
\cite[\S13.1]{sutherland}.
We use the \emph{normalized local module} $\vabs{\cdot}_v$ on
$K_v^\times$. It is the factor by which a scalar dilates additive
Haar measure. Equivalently, for $a\ne0$ it is
\begin{equation}\label{eq:normalization}
\vabs a_v=
\begin{cases}
q_v^{-\ord_v(a)}, & v\text{ nonarchimedean},\\
\vabs a_{\rm usual}, & K_v=\R,\\
\vabs a_{\rm usual}^{2}, & K_v=\C.
\end{cases}
\end{equation}
Section~\ref{sec:norms} derives all three cases from Haar measure.
The square at a complex place reflects its real dimension.
The proofs use products, filtered colimits, and pullbacks in spaces or
in an $\infty$-topos. Ordinary sets and groups are regarded as discrete
objects when we discuss their algebraic structure. When topology
matters, we specify it explicitly and give a sheaf model that retains
it. The arithmetic input is supplied by the completion and norm
theorems and by calculations on the two rational base fields.
\section{Adeles and ideles}\label{sec:adeles}
\subsection{Restricted products}
Let $J$ be a set of indices, $J_\infty\subset J$ a finite subset,
and $U_j\to X_j$ a family of maps in a category with products and
filtered colimits. Only the indices outside $J_\infty$ require a
choice of $U_j$. Write $\mathcal F(J)$ for the poset of finite
$S\supseteq J_\infty$. Define
\begin{equation}\label{eq:restricted}
\RP_J(X,U)=\varinjlim_{S\in\mathcal F(J)}
\left(\prod_{j\in S}X_j\times\prod_{j\notin S}U_j\right).
\end{equation}
In an $\infty$-category the products and colimit are the corresponding
$\infty$-categorical ones. The transition from $S$ to $T\supseteq S$
uses $U_j\to X_j$ for $j\in T\setminus S$. If $X_S$ denotes the
stage in parentheses, the universal property is
\[
\Hom(\RP_J(X,U),Y)
\simeq \varprojlim_{\mathcal F(J)^{\rm op}}\Hom(X_S,Y).
\]
Thus a map out of the restricted product is given by maps
on the stages with homotopy-coherent compatibility. For inclusions
of discrete objects in spaces
\eqref{eq:restricted} is the usual set of tuples belonging to $U_j$
outside a finite set. The poset is filtered by finite union, and
filtered colimits of spaces preserve $0$-truncation and commute
with finite limits \cite[Thm.~9.1.5.1]{kerodon}.
\begin{definition}
The \emph{adeles} and \emph{ideles} of $K$ are
\begin{align*}
\A_K&=\Rprod_{v\in V_K}(K_v,\OO_v),\\
\I_K&=\Rprod_{v\in V_K}(K_v^\times,\OO_v^\times),
\end{align*}
with every archimedean place included in each finite stage.
We give every stage its product topology and the union the
topology for which a subset is open exactly when its intersection
with each stage is open.
\end{definition}
For example the adele stages over $\Q$ have the form
$\R\times\prod_{p\in S_{\fin}}\Q_p\times
\prod_{p\notin S_{\fin}}\Z_p$. An adele can have infinitely many
nonzero coordinates. The finite restriction concerns the places
where a coordinate fails to be integral. Similarly, an idele has
nonzero coordinates everywhere and is a unit at almost every
nonarchimedean place.
At a nonarchimedean place both $\OO_v$ and $\OO_v^\times$ are compact
and open in their respective local groups. It follows that the
stages are open, the additive adele group and the idele group are
locally compact, and the coordinatewise algebra operations are
continuous in their respective topologies. In fact two inputs lie
in one common stage, and on each stage the operations are
continuous products of local operations. The unrestricted
archimedean and finitely many nonarchimedean factors are
locally compact, and the remaining product is compact.
Coordinate projections separate points, so both groups are Hausdorff.
See \cite[\S25.2]{sutherland25} and
\cite[\S26.1]{sutherland26}.
\subsection{Units as an inverse graph}
\begin{proposition}\label{prop:units}
The unit group of the underlying ring $\A_K$ is $\I_K$.
With topology, $\I_K$ is the inverse graph
\[
\{(x,y)\in\A_K\times\A_K:xy=1\}
\]
under the identification $x\mapsto(x,x^{-1})$.
\end{proposition}
\begin{proof}
For a commutative ring object, the object of units is the pullback
of multiplication $R\times R\to R$ along the unit $1\to R$.
Filtered colimits commute with these finite limits in an
$\infty$-topos \cite[Prop.~6.4.5.9]{htt}.
Units also commute with products, since the
pullback condition is coordinatewise. Applied to the adele stages
this gives the claimed unit group. In tuples, both $x$ and
$x^{-1}$ must be integral at almost all places. In the
inverse graph, intersecting with $\A_{K,S}\times\A_{K,S}$ gives the
idele stage $\I_{K,S}$ with its product topology.
These are open and cover the graph.
\end{proof}
Over $\Q$, let $u^{(p)}$ be the idele equal to $p$ at $p$ and to
$1$ everywhere else. Along the primes in increasing order it
converges to $1$ in the adele topology, because a basic
neighborhood restricts only finitely many coordinates within the
stage integral at every finite place. It stays outside the open
idele stage requiring a unit at every prime. Thus the inverse
graph is needed to obtain the idele topology from the adele ring.
\subsection{A sheaf model for the topology}
On the site of profinite sets with finite jointly surjective
covering families, a space $T$ defines the set-valued sheaf
$P\mapsto\operatorname{Cont}(P,T)$, viewed as a $0$-truncated
object in the $\infty$-topos of space-valued sheaves. This is
the usual condensed-set construction \cite[\S1]{scholze}.
Products are computed on sections. For either restricted product
above, a continuous map from a profinite $P$ into the union has
compact image and therefore lands in one stage. Indeed, the open
stages cover the image, and a finite subcover can be absorbed
by taking the union of its finite indexing sets.
The resulting sectionwise filtered colimit is a sheaf, since
the set-valued sheaf condition for a finite covering family is
a finite equalizer condition. Consequently \eqref{eq:restricted}
also presents both restricted products as sheaves in this model.
The finite-limit proof of Proposition~\ref{prop:units} then
applies to these sheaves and recovers the topological inverse graph.
\subsection{Regrouping places}
\begin{lemma}\label{lem:regroup}
Let $p:W\to V$ have finite fibers with $W_\infty=p^{-1}(V_\infty)$.
There is a canonical equivalence
\begin{align*}
\RP_W(X,U)&\simeq\RP_V(X',U'),\\
X'_v&=\prod_{p(w)=v}X_w,\qquad
U'_v=\prod_{p(w)=v}U_w .
\end{align*}
\end{lemma}
\begin{proof}
The stages indexed by $p^{-1}(S)$, with $S$ finite and
containing $V_\infty$, are cofinal. For a finite stage $T$
on the $W$ side, the $S$ satisfying $T\subseteq p^{-1}(S)$
form a poset with least element $p(T)\cup V_\infty$.
At each such stage, the product over $W$ is the iterated
product over the fibers and then over $V$, by the universal
property of products. The equivalences respect the transition
maps. For topological spaces with open inclusions the same
stagewise argument identifies the restricted product topologies,
since the cofinal stages give their open sets.
\end{proof}
In particular, compatible fiberwise maps of pairs
$(X'_v,U'_v)\to(Y_v,B_v)$ induce a map on restricted products.
The regrouping takes place inside each product stage, so it works
for infinite sets of places without an interchange law for
filtered colimits and infinite products.
\section{Norms and local modules}\label{sec:norms}
\subsection{Determinants and completions}
For a finite field extension $L/F$ and $a\in L^\times$, the
field norm is $\Nm_{L/F}(a)=\det_F(m_a)$, where $m_a$ is
multiplication by $a$ on $L$ as an $F$-vector space. This
makes $\Nm_{L/F}$ a group homomorphism.
The determinant is functorial on the groupoid of finite-dimensional
vector spaces and linear isomorphisms. With direct sum on the
source and tensor product on the target it is symmetric monoidal
with values in the Picard groupoid of $\Z$-graded lines:
$E\mapsto(\bigwedge^{\dim E} E,\dim E)$. A Picard groupoid
is a symmetric monoidal groupoid whose objects are invertible
under tensor product up to isomorphism. Interchanging lines of
degrees $m,n$ contributes the sign $(-1)^{mn}$. This is the
sign obtained when the corresponding blocks of wedge factors
are interchanged. The exterior power construction also commutes
with extension of scalars \cite{knudsen,stacksdet}.
\begin{proposition}[Norm under completion]\label{prop:completions}
Let $L/F$ be a finite separable extension of global fields.
For each place $v$ of $F$ there is a canonical isomorphism
\begin{equation}\label{eq:completions}
L\otimes_F F_v \simeq \prod_{w\mid v}L_w .
\end{equation}
Under $F\hookrightarrow F_v$,
\begin{equation}\label{eq:normbase}
\Nm_{L/F}(a)=\prod_{w\mid v}\Nm_{L_w/F_v}(a)
\qquad(a\in L^\times).
\end{equation}
\end{proposition}
\begin{proof}
Choose a primitive element for $L/F$ with minimal polynomial $h$.
Over $F_v$, the distinct irreducible factors of $h$ give a
product of field quotients by the Chinese remainder theorem.
Each factor carries the unique extending local absolute-value
class and the image of $L$ is dense in it. For density, the
image spans the factor over $F_v$ and $F$ is dense in $F_v$.
Conversely the map to each completion $L_w$ factors through
one of these fields. Its image is complete and contains the
dense image of $L$, so it is all of $L_w$.
Thus the factors are indexed by the extensions of $v$
to $L$ \cite[Thm.~13.5]{sutherland}.
The map in \eqref{eq:completions} is induced by the
completion maps, giving its independence of $h$.
After base change, multiplication by $a$ is the direct sum
of the local multiplications. Base change takes the
determinant line of $L$ to that of $L\otimes_F F_v$.
The direct sum is sent to the tensor product of the local
lines. The scalar on their tensor product is the product
of the local scalars, giving \eqref{eq:normbase}. Any sign
in reordering the lines cancels when conjugating a line
automorphism by that reordering.
\end{proof}
\subsection{Haar measure and the local norm}
Let $\mu$ be a nonzero additive Haar measure of a local
field $M$. A scalar $a\in M^\times$ acts by an additive
homeomorphism. Uniqueness of Haar measure gives a positive
factor $d_M(a)$ defined by $\mu(aE)=d_M(a)\mu(E)$ for
measurable $E$. It is independent of the chosen measure and
is multiplicative because dilations compose. In a nonarchimedean
field, a unit preserves the valuation ring and a uniformizer
sends it to a subgroup of index $q$. Taking the valuation
ring to have measure $1$ gives $d_M(a)=q^{-\ord(a)}$.
Real or complex scalar multiplication has real determinant
of absolute value $\vabs a_{\rm usual}$ or
$\vabs a_{\rm usual}^{2}$, respectively. This proves
\eqref{eq:normalization}.
\begin{lemma}[Local norm compatibility]\label{lem:local}
If $M/E$ is a finite extension of local fields, then
\[
d_M(a)=d_E\bigl(\Nm_{M/E}(a)\bigr).
\]
\end{lemma}
\begin{proof}
Choose an $E$-basis of $M$ and use product Haar measure
on $E^{[M:E]}$. For a linear automorphism of $E^n$ the
Haar dilation factor is $d_E$ of its determinant.
For an elementary shear this follows from translation
invariance in one coordinate and Fubini's theorem. Coordinate
permutations preserve product measure and have determinant
$\pm1$, of module $1$. Diagonal maps scale each coordinate
by its own module. Gaussian elimination decomposes an
invertible matrix into these transformations. Apply the
result to $m_a$ and use the independence of Haar normalization.
\end{proof}
For nonarchimedean $M/E$, with residue degree $f$ and
ramification index $e$, $q_M=q_E^f$. Comparing
exponents in Lemma~\ref{lem:local} gives
\begin{equation}\label{eq:ordnorm}
\ord_E(\Nm_{M/E}(a))=f\,\ord_M(a).
\end{equation}
Here $f$ weights the norm on arbitrary elements. For a base
scalar $b\in E^\times$ one has $\ord_M(b)=e\ord_E(b)$ and
$d_M(b)=d_E(b)^{ef}$. In particular the local norm maps
local units to local units. For $\C/\R$ it sends
$z$ to $z\overline z$, consistent with the squared modulus.
\subsection{The adelic comparison}
First we check that elements of a global field give diagonal
ideles. A number field is finite separable over $\Q$.
In positive characteristic, let $\F_q$ be the algebraic closure
of $\F_p$ in $K$. This is finite, since for every finite
constant subfield $k'\subset K$ the degree of
$k'(t)/\F_p(t)$ over $\F_p(t)$ is $[k':\F_p]$, bounded by
$[K:\F_p(t)]$. A separating transcendence basis over the
perfect field $\F_q$ gives a parameter $s$ with
$K/\F_q(s)$ finite separable \cite{conrad}. The initial
parameter $t$ may give an inseparable extension, so we
use the separating parameter in the proof.
\begin{lemma}[Finite support]\label{lem:support}
For $a\in K^\times$, both $a$ and $a^{-1}$ are integral
at all but finitely many nonarchimedean places of $K$.
\end{lemma}
\begin{proof}
Choose a rational base field $F$ as above and a monic
polynomial over $F$ for each of $a$ and $a^{-1}$. At
almost all nonarchimedean places of $F$ their coefficients
are integral, by factorization of rational numbers or
rational functions. They are then integral at each place
of $K$ lying above those base places. If a root of a
monic polynomial with integral coefficients had negative
valuation, its highest-degree term would have strictly smaller
valuation than every other term, which prevents the sum from
vanishing. Finitely many places of $K$ lie above the
finitely many excluded base places by
Proposition~\ref{prop:completions}.
\end{proof}
Thus $K^\times$ embeds by a diagonal homomorphism
$\Delta_K:K^\times\hookrightarrow\I_K$.
The diagonal ring map into $\A_K$ includes $0$ as well.
For general ideles set
\begin{equation}\label{eq:globalmodule}
\mu_K:\I_K\longrightarrow\R_{>0},\qquad
\mu_K(x)=\prod_{v\in V_K}\vabs{x_v}_v .
\end{equation}
Each stage map uses only the finitely many indices of that
stage. These maps are continuous homomorphisms and compatible,
so the universal property gives a continuous homomorphism
on the whole idele group. For a finite separable extension,
Lemma~\ref{lem:regroup} and the unit property of the local
norms give the \emph{idele norm}
\begin{equation}\label{eq:idele-norm}
(\Nm^\I_{L/F}(x))_v=\prod_{w\mid v}
\Nm_{L_w/F_v}(x_w).
\end{equation}
The norms are continuous as determinants, and a fiber over
a place outside the finite stage contributes a unit. Therefore
\eqref{eq:idele-norm} is a continuous homomorphism.
The two compatibilities are
\begin{align}
\mu_L&=\mu_F\circ\Nm^\I_{L/F},\label{eq:norm-compact}\\
\Nm^\I_{L/F}\circ\Delta_L
&=\Delta_F\circ\Nm_{L/F}.\label{eq:diag-compact}
\end{align}
The first is Lemma~\ref{lem:local} multiplied over fibers,
using finite support. The second is
Proposition~\ref{prop:completions}.
\Needspace{9\baselineskip}
\section{The adelic product formula}\label{sec:product}
\begin{theorem}\label{thm:product}
For every global field $K$ and $a\in K^\times$,
\begin{equation}\label{eq:product}
\prod_{v\in V_K}\vabs a_v=1 .
\end{equation}
Equivalently $\mu_K\circ\Delta_K$ is the trivial character.
\end{theorem}
\begin{proof}
By \eqref{eq:norm-compact} and \eqref{eq:diag-compact},
for any finite separable extension $L/F$ we have
\begin{equation}\label{eq:reduction}
\mu_L(\Delta_L a)
=\mu_F\bigl(\Delta_F\Nm_{L/F}(a)\bigr).
\end{equation}
The rational base fields used in Lemma~\ref{lem:support}
therefore suffice.
Over $\Q$, write $a=\pm\prod_p p^{n_p}$ with only
finitely many $n_p$ nonzero. The $p$-factor in the product is
$p^{-n_p}$ and the real factor is $\prod_p p^{n_p}$.
Over $\F_q(t)$, the finite places correspond to monic
irreducibles $P\in\F_q[t]$, with residue size $q^{\deg P}$.
The remaining place is infinity, with residue size $q$
and $\ord_\infty(t)=-1$. Write
$a=c\prod_P P^{n_P}$ for $c\in\F_q^\times$. Its finite
product is $q^{-\sum_P n_P\deg P}$, and
$\ord_\infty(a)=-\sum_P n_P\deg P$. Hence the factor
at infinity is $q^{\sum_P n_P\deg P}$. This proves the
claim in both cases and \eqref{eq:reduction} completes the proof.
\end{proof}
\begin{example}[A complex place and a ramified prime]
In $\Q(i)$ take $a=(2+i)/(1+i)$. Its only nonzero
finite valuations are $1$ at $(2+i)$ and $-1$ at $(1+i)$.
The residue sizes are $5$ and $2$, as follows from the
indices of the principal ideals in $\Z[i]$. The finite
factors are $1/5$ and $2$. At the single complex place,
$\vabs{a}_{\rm usual}^2=5/2$. The product is
$(1/5)\cdot2\cdot(5/2)=1$. In particular, although $2$
ramifies with index two, the generator $1+i$ has valuation
one and local module $1/2$.
\end{example}
\begin{example}[A place of degree two]
In $\F_5(t)$ take $a=(t^2+2)/(t-1)^3$.
The numerator is irreducible because $3$ is a nonsquare
in $\F_5$. Its place has module $5^{-2}$.
The factors at $t-1$ and infinity are $5^3$ and $5^{-1}$,
respectively. They give $5^{-2}5^3 5^{-1}=1$.
\end{example}
\subsection{The idele-class character}
The algebraic idele-class group is
$C_K=\I_K/\Delta_K(K^\times)$. It is the set of components
of the action groupoid $\I_K//K^\times$: an arrow labelled $a$
sends $x$ to $\Delta_K(a)x$. Multiplication of objects and
of arrow labels makes this groupoid a Picard groupoid.
The action is free, so its isotropy groups are trivial and
it is equivalent to the discrete groupoid on $C_K$.
Theorem~\ref{thm:product} says that $\mu_K$ is constant
on these orbits. Equivalently $\mu_K$ sends every arrow to
an identity in the discrete Picard groupoid with group
of objects $\R_{>0}$. It thereby gives a symmetric monoidal
functor on the action groupoid. With the quotient topology
on $C_K$, the induced character is continuous by the
universal property of the topological quotient.
\section{An additive proof by Haar measures}\label{sec:haar}
The local module is defined through additive Haar measure.
We can therefore prove the product formula by constructing
a Haar measure on $\A_K$ preserved by $K^\times$. The
construction uses the additive quotient $\A_K/K$.
\begin{proposition}\label{prop:lattice}
The diagonal copy of $K$ is closed and discrete in $\A_K$,
and $\A_K/K$ is compact.
\end{proposition}
\begin{proof}
First let $F$ be $\Q$ or $\F_q(t)$, with $B=\Z$ or
$\F_q[t]$, respectively. For $x\in\A_F$ choose a
nonzero $d\in B$ such that $d x_v$ is integral
at every finite polynomial or prime place of $B$.
The Chinese remainder theorem gives $b\in B$ with
$b\equiv d x_v \pmod {d\OO_v}$ at all places dividing $d$.
Here $B/(\pi^r)\simeq\OO_v/\pi^r\OO_v$ for the
corresponding prime $\pi$ of $B$. Then $x-b/d$ is
integral at every place of $B$. In the number-field
base case subtract an integer to put its real component
in $[0,1]$. In the function-field base case subtract its
polynomial part at infinity to put its infinity component
in $t^{-1}\F_q[[t^{-1}]]$. These operations yield
representatives in a compact product of local sets.
For discreteness, all coordinates at places of $B$ may be
required to be integral and the infinity coordinate to
lie in $(-1,1)$ or $t^{-1}\F_q[[t^{-1}]]$, respectively.
This open neighborhood intersects the diagonal in $0$,
since an element integral at every place of $B$ lies in $B$.
For $K/F$ finite separable of degree $n$, choose an
$F$-basis $e_i$ and its trace-dual basis $e_i^*$, using
nondegeneracy of the trace pairing. Away from finitely
many $v$, all these basis elements are integral at each
$w\mid v$ by Lemma~\ref{lem:support}. Under
\eqref{eq:completions}, the coefficient of $e_i$ in
$x=(x_w)_{w\mid v}$ is
\[
\sum_{w\mid v}\operatorname{Tr}_{K_w/F_v}(e_i^*x_w).
\]
This follows by base change of the trace of multiplication.
For integral $x_w$ the result is integral, since traces of
integral elements lie in the integrally closed ring $\OO_v$.
Conversely, integral coefficients give integral $x_w$ by
integrality of the $e_i$. The basis thus identifies the
local lattices $\OO_v^n$ and $\prod_{w\mid v}\OO_w$
outside finitely many places. Regrouping as in
Lemma~\ref{lem:regroup}, these local linear homeomorphisms
give $\A_F^n\simeq\A_K$ as additive topological groups
and identify $F^n$ with $K$. This gives discreteness and
compactness of the quotient for $K$. A discrete subgroup
of a Hausdorff topological group is closed.
\end{proof}
For the short exact sequence of locally compact abelian groups
\[
0\longrightarrow K\longrightarrow\A_K
\longrightarrow\A_K/K\longrightarrow0,
\]
use counting measure on $K$ and probability Haar measure on
the compact quotient. Quotient integration gives the Haar
measure on $\A_K$ determined by these choices. For a
continuous compactly supported function $f$, its integral
is the integral over $\A_K/K$ of
$\sum_{b\in K}f(x+b)$, where $x$ is any representative of
the coset. Multiplication by $a\in K^\times$ permutes the
summation group and preserves probability Haar measure on
the quotient. Therefore it preserves the resulting Haar
measure on $\A_K$.
On a restricted product stage, the restriction of a Haar
measure is proportional to the product of local Haar
measures, with the integral factors given probability
measure. Choose a stage containing every place at which
$a$ is a nonunit. Multiplication by $a$ is an automorphism
of this stage and has dilation factor $\prod_v\vabs a_v$,
computed on product sets that are unrestricted within the
unit-preserving integral factors. It follows that the
product is $1$. This is the measure argument behind
Tate's proof \cite[Thm.~4.3.1]{tate}.
Functoriality is explicit in this argument: the Haar
measures on a locally compact abelian group form an
$\R_{>0}$-torsor, and quotient integration combines the
subgroup and quotient torsors into the middle one. The
counting and probability choices give a section fixed by
the action of $K^\times$.
At the level of connective algebraic $K$-theory, the same
sequence gives a nullhomotopy for the map induced by
adelic extension of scalars \cite[Rem.~0.3]{clausen}.
For a finite-dimensional $K$-vector space $E$, tensor the
additive sequence with $E$ and give each term the topology
of a finite power of $K$, $\A_K$, or $\A_K/K$,
respectively. Change of basis acts continuously, so this
defines a short exact sequence of exact functors to the exact
category of second-countable locally compact abelian groups.
Its short exact sequences use closed embeddings and their
topological quotients. Additivity identifies the middle
functor's map of $K$-theory spectra with the sum of the
endpoint maps. The endpoints factor through countable
discrete and second-countable compact groups, respectively.
Countable direct sums in the first category and countable
products in the second yield exact endofunctors $T$ with
$T\simeq\mathrm{id}\oplus T$. By additivity their $K$-theory
spectra are contractible, giving the asserted nullhomotopy.
\section{Curves and adelic totalization}\label{sec:curves}
Let $X$ be a regular integral Noetherian scheme of dimension
one, with function field $K$ and set of closed points $|X|$.
Write $\widehat{\OO}_x$ for the completed local ring at a
closed point and $K_x=\Frac(\widehat{\OO}_x)$.
There is one generic point, and every strict specialization
chain goes from it to a closed point. The
additive normalized adelic complex of $\OO_X$ is
\begin{equation}\label{eq:curvecomplex}
\left[
K\oplus\prod_{x\in|X|}\widehat{\OO}_x
\xrightarrow{\ d\ }
\Rprod_{x\in|X|}(K_x,\widehat{\OO}_x)
\right],
\end{equation}
in cohomological degrees $0$ and $1$, with
$d(a,(b_x))=(a-b_x)_x$. The terms come from the generic
point, the closed points, and the two-point flags. The
complex represents $R\Gamma(X,\OO_X)$, the derived global
sections \cite[Ex.~2.2]{bgw}.
Equivalently it is the homotopy fiber of $d$ as a map
of degree-zero objects in the derived $\infty$-category
of abelian groups.
For the dimension-one calculation, let $\mathcal K$ be the
sheaf of rational functions. For each closed point let
$i_x:\{x\}\hookrightarrow X$ be its inclusion. There is a
flasque resolution
\begin{align*}
0\longrightarrow\OO_X\longrightarrow\mathcal K
&\longrightarrow
\bigoplus_{x\in|X|} i_{x*}(K_x/\widehat{\OO}_x)
\longrightarrow0 .
\end{align*}
At a closed point, DVR approximation identifies
$K/\OO_{X,x}$ with $K_x/\widehat{\OO}_x$.
Each rational function has only finitely many poles on a
Noetherian curve. The last sheaf is the sheaf of finitely
supported local principal parts on opens, because all opens
are quasi-compact. Its restrictions are surjective by
extension by zero. The generic stalk has no principal parts
since any finite support can be deleted from a neighborhood.
Taking global sections gives the complex
$[K\to\bigoplus_x K_x/\widehat{\OO}_x]$. Complex
\eqref{eq:curvecomplex} maps to this complex by projecting
off $\prod_x\widehat{\OO}_x$ in degree zero and reducing
coordinates modulo $\widehat{\OO}_x$ in degree one. The
kernel is a copy of $\prod_x\widehat{\OO}_x$ mapped by
minus the identity to itself and is acyclic. This identifies
the two complexes.
For a smooth, proper, geometrically integral curve over
$\F_q$, the closed points index all the places of its
function field. Set
\begin{align*}
\Div(X)&=\bigoplus_{x\in|X|}\Z[x],\\
\deg\Bigl(\sum_x n_x[x]\Bigr)
&=\sum_x n_x[\kappa(x):\F_q].
\end{align*}
The homomorphism $\operatorname{div}_{\I}:\I_K\to\Div(X)$
sends $u$ to $\sum_x\ord_x(u_x)[x]$, and
\begin{equation}\label{eq:divmodule}
\log\mu_K(u)=-(\log q)\deg(\operatorname{div}_{\I}u).
\end{equation}
For diagonal $a$, this divisor is the principal divisor
of $a$. Thus the product formula says exactly that a
principal divisor has degree zero. Geometrically, for
nonconstant $a$ the map $a:X\to\mathbb P^1_{\F_q}$ is
finite flat, and the zero and pole divisors are its fibers
over $0$ and $\infty$. Their degrees both equal the rank
of the finite flat map, with lengths accounting for
multiplicity even in the inseparable case
\cite[\S42.18]{stacksdiv}. Constants have zero divisor.
There is also a groupoid version of the divisor description.
The action groupoid $\Div(X)//K^\times$ uses translation
by principal divisors. Its components are divisor classes
and the automorphisms of each object are the global units
$\F_q^\times$. This records the constant scalars that act
trivially on divisors. The functor sending $D$ to $\OO_X(D)$
identifies it with the Picard groupoid of line bundles and
their isomorphisms. An arrow $D\to D+\operatorname{div}(a)$
is sent to multiplication by $a^{-1}$. Every line bundle has
a nonzero rational section defining a divisor, and every
isomorphism between these subsheaves of the rational-function
sheaf acts at the generic point by a unique rational scalar.
The valuation conditions on this scalar identify exactly the
arrows in the action groupoid. The identifications respect
tensor products. By \eqref{eq:divmodule}, degree defines a
symmetric monoidal functor from this Picard groupoid to
discrete $\Z$.
For $X=\operatorname{Spec}\OO_K$ of a number field, the
closed points give the finite adeles in
\eqref{eq:curvecomplex}. The arithmetic product formula
also uses the archimedean places, with the real and complex
modules of \eqref{eq:normalization}. More generally, the
Beilinson cosimplicial adelic ring $\A_X^\bullet$ packages
iterated completion along flags on a Noetherian scheme.
Groechenig’s adelic descent theorem gives
$\Perf(X)\simeq\Tot\Perf(\A_X^\bullet)$, where $\Perf$
denotes the $\infty$-category of perfect complexes and
$\Tot$ the limit of the cosimplicial diagram
\cite[Thm.~3.1 in the arXiv version]{groechenig}. The calculation above
identifies the degree-zero and degree-one additive terms
in the regular-curve case.
\section{Conclusion}
The product formula factors into three operations that
respect different structures. Regrouping the restricted
product gives the domain for the idele norm. The symmetric
monoidal determinant identifies the global norm after
completion, and additive Haar dilation fixes the local
normalizations. The rational base calculations then prove
triviality on the diagonal. In the curve case the same
triviality has the geometric form of degree zero, while
the adelic complex exhibits how the local rings and the
generic field enter derived global sections.
\smallskip
\noindent\textbf{Preparation disclosure.}
The manuscript was drafted and typeset with AI assistance.
\balance
\begin{thebibliography}{14}
\small
\bibitem{sutherland}
A.~V. Sutherland.
\emph{Global fields and the product formula}.
MIT 18.785, Lecture 13, 2021.
\href{https://ocw.mit.edu/courses/18-785-number-theory-i-fall-2021/mit18_785f21_lec13.pdf}{Lecture notes}.
\bibitem{sutherland25}
A.~V. Sutherland.
\emph{The ring of adeles, strong approximation}.
MIT 18.785, Lecture 25, 2021.
\href{https://ocw.mit.edu/courses/18-785-number-theory-i-fall-2021/mit18_785f21_lec25.pdf}{Lecture notes}.
\bibitem{sutherland26}
A.~V. Sutherland.
\emph{The idele group, profinite groups, and infinite Galois theory}.
MIT 18.785, Lecture 26, 2021.
\href{https://ocw.mit.edu/courses/18-785-number-theory-i-fall-2021/mit18_785f21_lec26.pdf}{Lecture notes}.
\bibitem{kerodon}
J. Lurie.
\emph{Exactness of filtered colimits}.
Kerodon, \href{https://kerodon.net/tag/05XS}{Theorem 9.1.5.1}.
\bibitem{htt}
J. Lurie.
\emph{Higher Topos Theory}.
Annals of Mathematics Studies 170, Princeton University Press, 2009.
\href{https://people.math.harvard.edu/~lurie/papers/HTT.pdf}{Author's copy}.
\bibitem{scholze}
P. Scholze.
\emph{Lectures on condensed mathematics}.
Bonn lectures, 2019, updated 2026.
\href{https://people.mpim-bonn.mpg.de/scholze/Condensed.pdf}{Lecture notes}.
\bibitem{knudsen}
F.~F. Knudsen and D. Mumford.
The projectivity of the moduli space of stable curves I.
Preliminaries on det and Div.
\emph{Mathematica Scandinavica}, 39(1), 19--55, 1976.
\href{https://doi.org/10.7146/math.scand.a-11642}{Published article}.
\bibitem{stacksdet}
The Stacks Project Authors.
\emph{Determinants}.
\href{https://stacks.math.columbia.edu/tag/0FJ9}{Tag 0FJ9}.
\bibitem{conrad}
B. Conrad.
\emph{Separating transcendence basis}.
Stanford University.
\href{https://virtualmath1.stanford.edu/~conrad/210BPage/handouts/septrbasis.pdf}{Course handout}.
\bibitem{tate}
J.~T. Tate.
\emph{Fourier Analysis in Number Fields and Hecke's Zeta-Functions}.
PhD thesis, Princeton University, 1950.
\href{https://sites.math.rutgers.edu/~alexk/2023S572/Tate1950.pdf\#page=44}{Thesis scan}.
\bibitem{clausen}
D. Clausen.
\emph{A K-theoretic approach to Artin maps}.
\href{https://arxiv.org/abs/1703.07842}{arXiv:1703.07842}, 2017.
\bibitem{bgw}
O. Braunling, M. Groechenig, and J. Wolfson.
Geometric and analytic structures on the higher ad\`eles.
\emph{Research in the Mathematical Sciences}, 3, article 22, 2016.
\href{https://escholarship.org/content/qt6223t25q/qt6223t25q.pdf}{Open-access article}.
\bibitem{stacksdiv}
The Stacks Project Authors.
\emph{Principal divisors and pushforward}.
\href{https://stacks.math.columbia.edu/tag/02RS}{Section 42.18}.
\bibitem{groechenig}
M. Groechenig.
Adelic descent theory.
\emph{Compositio Mathematica}, 153(8), 1706--1746, 2017.
\href{https://doi.org/10.1112/S0010437X17007217}{Published article},
\href{https://arxiv.org/abs/1511.06271}{open preprint}.
\end{thebibliography}
\end{document}