01.02.15 · foundations / groups

Galois cohomology, Hilbert's Theorem 90, and the Brauer group of a field

shipped3 tiersLean: none

Anchor (Master): Serre Local Fields Ch. X and Galois Cohomology Ch. I–II; Gille–Szamuely Central Simple Algebras and Galois Cohomology Ch. 2–4; Lang Algebra Ch. XVIII (Brauer group via crossed products)

Intuition Beginner

Take a field together with a bigger field sitting on top of it, related by a tidy symmetry group: the automorphisms of the big field that leave the small field fixed. This symmetry group is the heart of Galois theory. Now ask a measurement question: given a number in the big field, multiply together all the versions of it produced by the symmetries. The result is a number in the small field, and we call it the norm. The norm is the multiplicative shadow a big-field number casts down onto the small field.

A natural puzzle: which big-field numbers have norm equal to one? Norm one means the shadow is as plain as possible. Hilbert's answer, more than a century old, is startlingly clean. When the symmetry group is a single rotation repeated, every norm-one number arises in one fixed way: as a ratio between a number and its rotated copy. Nothing exotic ever appears.

This unit is about that rigidity and the bookkeeping that explains it. The same bookkeeping, pushed one notch further, classifies a whole zoo of algebras built over a field.

Visual Beginner

Picture the symmetry group as a clock with a few positions, each position a way of relabelling the big field. A number in the big field gets copied to every clock position; the norm is the product going all the way around the dial. The norm-one numbers are the ones whose round-the-dial product lands exactly on the value one.

The second panel shows the resolution. Every norm-one number is drawn as a ratio: a chosen element on top, the same element after one click of the clock on the bottom. As you sweep the chosen element through the big field, these ratios sweep out exactly the norm-one set, with nothing left over and nothing missing. That single picture — norm one means a ratio across one rotation — is Hilbert's Theorem 90.

Worked example Beginner

Let the small field be the real numbers and the big field the complex numbers. The symmetry group has two positions: do nothing, and flip the sign of the imaginary part (complex conjugation). The norm of a complex number is the number times its conjugate, which is the squared distance from zero. Norm one therefore means the complex numbers on the unit circle.

Hilbert's picture says every unit-circle number is a ratio of some complex number to its conjugate. Check it. Take any complex number $z$ on the unit circle and write $w = 1 + z$ . Then $w$ divided by its conjugate equals $z$ whenever $w$ is not zero, and the one excluded value, $z = - 1$ , is the ratio of $i$ to its conjugate $- i$ . So sweeping $w$ across the complex numbers and forming the ratio of $w$ to its conjugate sweeps out the entire unit circle.

This is the smallest case of the theorem. The two-position clock is real-over-complex, the norm is squared modulus, and the norm-one set is the circle, captured exactly by ratios across the one flip. The rest of the unit replaces this two-position clock with any cyclic clock, and then with the full machinery that counts how algebras over a field fit together.

Check your understanding Beginner

Formal definition Intermediate+

Let $L / F$ be a finite Galois extension with Galois group $G = Gal (L / F)$ , the inclusion-reversing apparatus of the Galois correspondence being assumed (it is developed as a companion topic in 01.02.13). A (left) $G$ -module is an abelian group $A$ written multiplicatively or additively, equipped with an action of $G$ by automorphisms; the basic example is $A = L^{\times}$ , the multiplicative group of the big field, on which $σ \in G$ acts by $σ \cdot a = σ (a)$ . The subgroup of fixed points is $A^{G}$ ; for $A = L^{\times}$ it is $F^{\times}$ .

Definition (cochains and cohomology in low degree). Write $A$ additively for the formulas. A $1$ -cochain is a function $f : G \to A$ . It is a $1$ -cocycle (or crossed homomorphism) if $$ f(\sigma\tau) = f(\sigma) + \sigma \cdot f(\tau) \qquad \text{for all } \sigma, \tau \in G, $$ and a $1$ -coboundary if $f (σ) = σ \cdot a - a$ for some fixed $a \in A$ . The first cohomology group is $$ H^1(G, A) = \frac{Z^1(G, A)}{B^1(G, A)} = \frac{{1\text{-cocycles}}}{{1\text{-coboundaries}}}. $$ A $2$ -cocycle is a function $c : G \times G \to A$ with $$ \sigma \cdot c(\tau, \rho) - c(\sigma\tau, \rho) + c(\sigma, \tau\rho) - c(\sigma, \tau) = 0, $$ a $2$ -coboundary is $c (σ, τ) = σ \cdot b (τ) - b (σ τ) + b (σ)$ for some $1$ -cochain $b$ , and $H^{2} (G, A) = Z^{2} (G, A) / B^{2} (G, A)$ . When $A = L^{\times}$ is written multiplicatively these read $f (σ τ) = f (σ) σ (f (τ))$ and $c (σ, τ) c (σ τ, ρ)^{- 1} c (σ, τ ρ) c (σ, τ)^{- 1}$ patterns, with coboundaries of the shape $σ (a) / a$ .

For an infinite extension one uses the profinite group $G = Gal (F^{sep} / F)$ with its Krull topology, and cochains are required to be continuous (equivalently, $H^{n} (G, A) = lim_{L / F} H^{n} (Gal (L / F), A^{(L)})$ over finite Galois $L / F$ ). This continuous cohomology is what the Brauer group below uses.

Definition (torsor / principal homogeneous space). For $A = L^{\times}$ , a class in $H^{1} (G, A)$ classifies isomorphism classes of torsors: sets with a free transitive $A$ -action twisted by a cocycle. The vanishing of $H^{1}$ says every such torsor over $L^{\times}$ is the standard one, which is the conceptual content of Theorem 90.

Counterexamples to common slips

$H^{1} (G, A)$ need not vanish for a general module $A$ : only the specific coefficient module $L^{\times}$ (and its additive sibling $L^{+}$ , which gives $H^{1} = 0$ by the normal basis theorem) is so well-behaved. For $A = μ_{n}$ the group $H^{1} (G, μ_{n})$ is the Kummer group $F^{\times} / (F^{\times})^{n}$ , rarely zero.
The norm map $N_{L / F} (a) = \prod_{σ \in G} σ (a)$ is multiplicative but the additive trace $Tr_{L / F}$ is the additive analogue; conflating them breaks the cyclic form of Theorem 90, which is a statement about the multiplicative group.
A $2$ -cocycle valued in $L^{\times}$ is a factor set; two factor sets give the same Brauer class precisely when they differ by a coboundary, not merely when the algebras have the same dimension. Equal dimension is far from sufficient.

Key theorem with proof Intermediate+

Theorem (Hilbert 90). Let $L / F$ be finite Galois with group $G$ . Then $H^{1} (G, L^{\times}) = 0$ . Equivalently, when $G = ⟨ σ ⟩$ is cyclic of order $n$ , an element $a \in L^{\times}$ has norm $N_{L / F} (a) = 1$ if and only if $a = b / σ (b)$ for some $b \in L^{\times}$ .

Proof. Cohomological form. Let $f : G \to L^{\times}$ be a $1$ -cocycle, so $f (σ τ) = f (σ) σ (f (τ))$ . By Dedekind's lemma the distinct automorphisms in $G$ are linearly independent over $L$ , so the $L$ -linear combination $\sum_{τ \in G} f (τ) τ$ is not the zero map; choose $c \in L^{\times}$ with $$ b := \sum_{\tau \in G} f(\tau),\tau(c) \neq 0 . $$ Apply $σ$ and use the cocycle relation $σ (f (τ)) = f (σ τ) / f (σ)$ : $$ \sigma(b) = \sum_{\tau} \sigma(f(\tau)),\sigma\tau(c) = \frac{1}{f(\sigma)}\sum_{\tau} f(\sigma\tau),(\sigma\tau)(c) = \frac{b}{f(\sigma)} . $$ Hence $f (σ) = b / σ (b)$ , a coboundary, so the class of $f$ in $H^{1}$ is the identity element. Since $f$ was arbitrary, $H^{1} (G, L^{\times}) = 0$ .

Cyclic form. With $G = ⟨ σ ⟩$ of order $n$ , suppose $N_{L / F} (a) = a σ (a) \dots σ^{n - 1} (a) = 1$ . Define a candidate cocycle on the generator by $f (σ^{i}) = a σ (a) \dots σ^{i - 1} (a)$ (empty product $1$ for $i = 0$ ); the norm-one hypothesis makes $f (σ^{n}) = N_{L / F} (a) = 1 = f (id)$ consistent, and one checks $f (σ^{i + j}) = f (σ^{i}) σ^{i} (f (σ^{j}))$ , so $f$ is a genuine cocycle with $f (σ) = a$ . By the cohomological form just proved, $f$ is a coboundary: $f (σ) = b / σ (b)$ , that is $a = b / σ (b)$ . Conversely any $a$ of that shape has norm $\prod_{i} σ^{i} (b / σ (b)) = b / σ^{n} (b) = b / b = 1$ by telescoping. $□$

Bridge. Theorem 90 builds toward the entire classification of central simple algebras, and it appears again in 04.03.08 as the étale/Picard-group statement $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) = Pic (X)$ with the field case giving zero. The foundational reason the proof works is Dedekind's linear independence of characters: it is exactly the same averaging device — twist by every group element and sum — that produces invariant vectors throughout representation theory, so Theorem 90 generalises the "average over the group" trick from finite groups acting on vector spaces to a Galois group acting on a field. This is exactly the vanishing that, one cohomological degree higher, fails in an organized way and thereby measures something: the central insight is that $H^{1} (G, L^{\times}) = 0$ is what makes the long exact sequence collapse so that $H^{2} (G, L^{\times})$ becomes a clean invariant. Putting these together, the bridge is that the degree-one vanishing clears the obstruction to identifying degree-two cohomology with algebras, and the same norm map reappears as the local invariant in 21.02.05.

Exercises Intermediate+

Exercise 4 (hard, short-answer).

Prove the additive analogue: $H^{1} (G, L^{+}) = 0$ for $L / F$ finite Galois (the additive Hilbert 90), using the normal basis theorem.

Hint

A normal basis makes $L$ a free rank-one module over the group algebra $F [G]$ ; compute $H^{1}$ of a free module.

Answer

The normal basis theorem says $L ≅ F [G]$ as an $F [G]$ -module, i.e. $L^{+}$ is induced (co-induced, the two agreeing for finite $G$ ) from the one-dimensional fixed $F$ -module along the identity subgroup. By Shapiro's lemma $H^{1} (G, F [G]) ≅ H^{1} ({1}, F) = 0$ , since cohomology of the identity group in positive degree vanishes. Hence $H^{1} (G, L^{+}) = 0$ . Equivalently, a trace-zero additive cocycle is a coboundary $σ \mapsto σ (b) - b$ , the additive mirror of the multiplicative statement. Rubric: full credit for invoking the normal basis theorem to identify $L^{+}$ as free over $F [G]$ and applying Shapiro / the vanishing for induced modules.

Exercise 5 (hard, short-answer).

Let $A = (a, b)_{F}$ be the quaternion algebra over $F$ ( $char F \neq = 2$ ) generated by $i^{2} = a$ , $j^{2} = b$ , $ij = - j i$ . Show $A$ is either a division algebra or a matrix algebra $M_{2} (F)$ , according to whether the norm form is anisotropic.

Hint

A central simple algebra of dimension $4$ has degree $2$ ; its only possible matrix decompositions are $M_{1} (D)$ with $D$ of dimension $4$ or $M_{2} (F)$ .

Answer

$A$ is central simple of dimension $4$ over $F$ , so by Artin–Wedderburn $A ≅ M_{r} (D)$ with $r^{2} dim_{F} D = 4$ . The options are $r = 1$ (so $A = D$ a division algebra of dimension $4$ ) and $r = 2$ , $D = F$ (so $A = M_{2} (F)$ ). The reduced norm form $N (x_{0} + x_{1} i + x_{2} j + x_{3} ij) = x_{0}^{2} - a x_{1}^{2} - b x_{2}^{2} + ab x_{3}^{2}$ is multiplicative; $A$ is a division algebra iff every nonzero element is invertible iff $N$ is anisotropic (no nonzero zero). If $N$ represents zero, the isotropic vector yields a zero divisor, forcing $A ≅ M_{2} (F)$ . Rubric: full credit for the dimension dichotomy and the anisotropy criterion via the reduced norm.

Exercise 6 (hard, short-answer).

Show that the crossed-product construction sends a $2$ -cocycle $c \in Z^{2} (G, L^{\times})$ to an associative $F$ -algebra, and that a coboundary yields an algebra isomorphic to $M_{n} (F)$ ( $n = ∣ G ∣$ ).

Hint

Build $A = ⨁_{σ \in G} L u_{σ}$ with $u_{σ} ℓ = σ (ℓ) u_{σ}$ and $u_{σ} u_{τ} = c (σ, τ) u_{σ τ}$ ; associativity is the $2$ -cocycle identity.

Answer

Define $A = (L, G, c) = ⨁_{σ} L u_{σ}$ with multiplication forced by $u_{σ} ℓ = σ (ℓ) u_{σ}$ and $u_{σ} u_{τ} = c (σ, τ) u_{σ τ}$ . Associativity $(u_{σ} u_{τ}) u_{ρ} = u_{σ} (u_{τ} u_{ρ})$ unwinds to $c (σ, τ) c (σ τ, ρ) = σ (c (τ, ρ)) c (σ, τ ρ)$ , which is exactly the $2$ -cocycle condition; so $c \in Z^{2}$ makes $A$ associative, with $L$ as a maximal subfield and $dim_{F} A = n^{2}$ . If $c$ is a coboundary $c (σ, τ) = σ (b_{τ}) b_{σ τ}^{- 1} b_{σ}$ , rescaling $u_{σ} \mapsto b_{σ}^{- 1} u_{σ}$ reduces the factor set to $c \equiv 1$ , giving the group-algebra-twisted $A ≅ End_{F} (L) ≅ M_{n} (F)$ . Rubric: full credit for deriving associativity from the cocycle identity and the rescaling that splits a coboundary.

Advanced results Master

The degree-two cohomology $H^{2} (G, L^{\times})$ is where the algebra lives. Fix $L / F$ finite Galois with group $G$ of order $n$ . The crossed-product algebra $(L, G, c)$ attached to a factor set $c \in Z^{2} (G, L^{\times})$ — built in Exercise 6 as $⨁_{σ} L u_{σ}$ with $u_{σ} u_{τ} = c (σ, τ) u_{σ τ}$ — is a central simple $F$ -algebra of dimension $n^{2}$ containing $L$ as a maximal subfield. Two factor sets give $F$ -isomorphic crossed products precisely when they differ by a coboundary, so the construction descends to an injection $H^{2} (G, L^{\times}) ↪ Br (L / F)$ , where $Br (L / F)$ is the set of similarity classes of central simple algebras split by $L$ . The hard direction — every $L$ -split central simple algebra is a crossed product for $L$ — uses the Skolem–Noether and double-centraliser theorems and yields the isomorphism $Br (L / F) ≅ H^{2} (G, L^{\times})$ .

The Brauer group $Br (F)$ is the set of similarity (Morita) classes of central simple $F$ -algebras, where $A \sim B$ iff $M_{r} (A) ≅ M_{s} (B)$ for some $r, s$ , equivalently iff their division-algebra cores agree by Artin–Wedderburn (01.02.14); the operation is $\otimes_{F}$ (01.02.10), with identity the class of $F$ itself and inverse the opposite algebra $A^{op}$ (since $A \otimes_{F} A^{op} ≅ End_{F} (A) ≅ M_{d i m A} (F)$ ). Taking the colimit over all finite Galois $L / F$ and passing to the separable closure gives the fundamental isomorphism $$ \mathrm{Br}(F) ;\cong; H^2!\big(\mathrm{Gal}(F^{\mathrm{sep}}/F),, (F^{\mathrm{sep}})^\times\big), $$ continuous cohomology of the absolute Galois group with coefficients in the multiplicative group of the separable closure. This single line packages all finite-dimensional division algebras over $F$ as cohomology classes.

Two computations anchor the theory. First, $Br (R) ≅ Z /2$ : the absolute Galois group is $Gal (C / R) = Z /2$ , and the cohomology $H^{2} (Z /2, C^{\times})$ is $R^{\times} / N_{C / R} (C^{\times}) = R^{\times} / R_{> 0} = {\pm 1}$ , the two classes being $R$ itself and Hamilton's quaternions $H = (- 1, - 1)_{R}$ , the unique nonsplit real division algebra. Second, $Br (F_{q}) = 0$ (Wedderburn's little theorem: every finite division ring is a field): the absolute Galois group $\hat{Z}$ has $H^{2} (\hat{Z}, \overline{F}_{q}^{\times}) = 0$ because the norm maps in finite-field extensions are surjective, so there are no nonsplit central simple algebras over a finite field.

The local theory continues the pattern: for a $p$ -adic field $K$ there is a canonical invariant isomorphism $inv_{K} : Br (K) \sim Q / Z$ , and the global Brauer group of a number field sits in the exact sequence $0 \to Br (k) \to ⨁_{v} Br (k_{v}) \sum inv_{v} Q / Z \to 0$ — the reciprocity backbone used in 21.02.05 and the obstruction of 21.02.09.

Synthesis. The foundational reason a field carries a Brauer group is that degree-one Galois cohomology vanishes (Hilbert 90) while degree-two does not, and this is exactly the asymmetry that turns $H^{2} (Gal, (F^{sep})^{\times})$ into a faithful catalogue of division algebras. Putting these together, the crossed-product construction is dual to the cocycle bookkeeping: a factor set builds an algebra, an algebra recovers a factor set, and the two are inverse up to similarity, so the central insight is that an associativity constraint on $u_{σ} u_{τ} = c (σ, τ) u_{σ τ}$ is the $2$ -cocycle identity. This generalises the quaternion case — where the single nonsplit class over $R$ is the order-two element of $Z /2$ — to every field at once, and it appears again in 21.02.05 and 21.02.09, where the same invariant map organizes local-global reciprocity. The bridge is that the elementary norm-one rigidity proved at Beginner level grows, one cohomological degree at a time, into the classification that controls central simple algebras over every field.

Full proof set Master

The cohomological and cyclic forms of Hilbert 90 are proved in full in the Key theorem section. The structural claims of the Master tier are recorded here.

Proposition (coboundaries split the crossed product). If $c = δ b$ is a $2$ -coboundary, $c (σ, τ) = σ (b_{τ}) b_{σ τ}^{- 1} b_{σ}$ for a $1$ -cochain $b : G \to L^{\times}$ , then the crossed product $(L, G, c) ≅ M_{n} (F)$ , $n = ∣ G ∣$ .

Proof. In $A = (L, G, c) = ⨁_{σ} L u_{σ}$ set $v_{σ} = b_{σ}^{- 1} u_{σ}$ . Then $$ v_\sigma v_\tau = b_\sigma^{-1} u_\sigma b_\tau^{-1} u_\tau = b_\sigma^{-1}\sigma(b_\tau^{-1}) u_\sigma u_\tau = b_\sigma^{-1}\sigma(b_\tau)^{-1} c(\sigma,\tau) u_{\sigma\tau}. $$ Substituting $c (σ, τ) = σ (b_{τ}) b_{σ τ}^{- 1} b_{σ}$ collapses this to $b_{σ τ}^{- 1} u_{σ τ} = v_{σ τ}$ . So in the new basis the factor set is identically $1$ , and $A ≅ (L, G, 1)$ , the algebra with $u_{σ} u_{τ} = u_{σ τ}$ and $u_{σ} ℓ = σ (ℓ) u_{σ}$ . This is the twisted group algebra acting on $L$ by $ℓ \cdot u_{σ} \mapsto ℓ σ$ ; the action map $A \to End_{F} (L)$ is an $F$ -algebra homomorphism between central simple algebras of the same dimension $n^{2}$ , hence (the source being simple) injective, hence an isomorphism. Since $End_{F} (L) ≅ M_{n} (F)$ , we get $A ≅ M_{n} (F)$ . $□$

Proposition (Brauer group of the reals). $Br (R) ≅ Z /2$ , generated by the class of the Hamilton quaternions $H$ .

Proof. The absolute Galois group of $R$ is $G = Gal (C / R) = {1, σ} ≅ Z /2$ , with $σ$ complex conjugation, and $Br (R) = Br (C / R) = H^{2} (G, C^{\times})$ because $C$ is the separable closure. For a cyclic group $G = ⟨ σ ⟩$ of order $n$ acting on a module $M$ , periodicity of cyclic cohomology gives $H^{2} (G, M) ≅ \hat{H}^{0} (G, M) = M^{G} / N (M)$ where $N = \sum_{i} σ^{i}$ . Here $M = C^{\times}$ , $M^{G} = R^{\times}$ , and $N (z) = z \overset{z}{ˉ} = ∣ z ∣^{2}$ , so $N (C^{\times}) = R_{> 0}$ . Therefore $$ H^2(G, \mathbb{C}^\times) \cong \mathbb{R}^\times / \mathbb{R}{>0} \cong {\pm 1} \cong \mathbb{Z}/2 . $$ The order-two class is represented by the factor set with $c (σ, σ) = - 1$ , whose crossed product $\bigoplus \mathbb{C} u\sigma $w i t h$ u_\sigma^2 = -1 $,$ u_\sigma z = \bar z u_\sigma $i se x a c tl y$ \mathbb{H} $. B y F r o b e ni u s^{'} s t h eor e m$ \mathbb{H} $i s t h eo n l y n o n co mm u t a t i v e f ini t e - d im e n s i o na l d i v i s i o na l g e b r a o v er$ \mathbb{R} $, ma t c hin g t h eor d er - tw o g r o u p .$ \square$

Proposition (Brauer group of a finite field). $Br (F_{q}) = 0$ .

Proof. The absolute Galois group of $F_{q}$ is $\hat{Z}$ , topologically generated by Frobenius, and $Br (F_{q}) = H^{2} (\hat{Z}, \overline{F}_{q}^{\times})$ , the colimit over the finite cyclic layers $Gal (F_{q^{m}} / F_{q}) = Z / m$ . For each layer, the cyclic periodicity formula gives $H^{2} (Z / m, F_{q^{m}}^{\times}) ≅ F_{q}^{\times} / N_{F_{q^{m}} / F_{q}} (F_{q^{m}}^{\times})$ . The norm map on finite multiplicative groups is $x \mapsto x^{(q^{m} - 1) / (q - 1)}$ , a surjection of the cyclic group $F_{q^{m}}^{\times}$ onto $F_{q}^{\times}$ (the index equals the kernel size by counting), so the quotient is the identity group. Each layer contributes $0$ , hence the colimit $Br (F_{q}) = 0$ . Equivalently every finite-dimensional division algebra over $F_{q}$ is commutative — Wedderburn's little theorem — so there are no nonsplit central simple algebras. $□$

Connections Master

The Fundamental Theorem of Galois Theory 01.02.13 supplies the correspondence this unit's coefficient module rides on. Group cohomology of $G = Gal (L / F)$ only makes sense once the subgroup–subfield dictionary is in place: fixed fields $L^{H}$ , normal subgroups matching Galois subextensions, and the degree-equals-order count are what let the inflation–restriction sequence and the colimit over finite layers operate. This unit ships in the same wave as 01.02.13 and treats it as a companion rather than a strict prerequisite, but the conceptual debt is total: without the correspondence there is no Galois module.

Semisimple rings and the Artin–Wedderburn structure theorem 01.02.14 is what makes the Brauer group well-defined. The similarity relation $A \sim B$ collapses each central simple algebra to its division-algebra core, and Artin–Wedderburn $A ≅ M_{r} (D)$ is precisely the theorem guaranteeing that core is unique. The crossed-product algebras of this unit are central simple of dimension $n^{2}$ , so they are matrix rings over division rings by that theorem, and the group law $\otimes_{F}$ on Brauer classes is Morita equivalence read through it.

The tensor product of modules and algebras 01.02.10 furnishes the group operation. $Br (F)$ is a group under $A \otimes_{F} B$ , with the splitting $A \otimes_{F} A^{op} ≅ End_{F} (A)$ providing inverses; this is the same universal-property tensor product used to define the algebra structure on $L \otimes_{F} L$ and to express base change $Br (F) \to Br (K)$ , $A \mapsto A \otimes_{F} K$ , under which a class splits exactly when $K$ is a splitting field.

The étale cohomology unit 04.03.08 carries Hilbert 90 into geometry. There the statement $H_{\overset{e}{ˊ} t}^{1} (X, G_{m}) ≅ Pic (X)$ is the sheaf-theoretic descendant of $H^{1} (G, L^{\times}) = 0$ : over a field the Picard group is the identity group, recovering Theorem 90, while the degree-two étale cohomology $H_{\overset{e}{ˊ} t}^{2} (X, G_{m})$ is the cohomological Brauer group of a scheme, globalising the $Br (F) ≅ H^{2} (Gal, (F^{sep})^{\times})$ of this unit. The field case is the punctual fibre of the geometric statement.

The Hilbert symbol and product formula 21.02.05 and the Brauer–Manin obstruction 21.02.09 are the number-theoretic payoff. They assume the field-level machinery built here — central simple algebras as $H^{2}$ classes, the local invariant map $inv_{v} : Br (k_{v}) \to Q / Z$ — and run it through every place of a number field, summing invariants to zero. The quaternion class that generates $Br (R) ≅ Z /2$ here is exactly the archimedean local term in that global sum.

Historical & philosophical context Master

The norm-one theorem is Satz 90 of David Hilbert's 1897 Zahlbericht (Die Theorie der algebraischen Zahlkörper) ^{[Hilbert 1897]}, where it appears as a statement about cyclic extensions of number fields, with no cohomology in sight — Hilbert proves directly that a norm-one element of a cyclic extension is a quotient $b / σ (b)$ . The recasting as the vanishing of a cohomology group came only with the cohomological algebra of the 1940s–50s; Emmy Noether's theory of crossed products and the Brauer group (Richard Brauer's 1929 Über Systeme hyperkomplexer Zahlen) supplied the degree-two side, and the synthesis $Br (F) ≅ H^{2} (Gal, (F^{sep})^{\times})$ is the centrepiece of Serre's Local Fields ^{[Serre 1979]} and Lang's Algebra ^{[Lang 2002]}. Gille and Szamuely's modern treatment ^{[Gille Szamuely 2017]} makes the equivalence of the algebraic and cohomological pictures the organizing principle of the subject.

The philosophical lesson is that cohomology is a measuring device for the failure of a naive expectation, and the value of the measurement depends sharply on the degree. One might hope every norm-one element is a coboundary, every torsor is the standard one, every central simple algebra is a matrix ring; Hilbert 90 confirms the first hope in degree one for the coefficient module $L^{\times}$ , and the surprise is that the same construction in degree two refuses to vanish and thereby becomes informative. The Brauer group is the precise quantity by which fields fail to have only matrix algebras over themselves — zero for finite fields and algebraically closed fields, $Z /2$ for the reals, $Q / Z$ for local fields — and the entire local-global architecture of class field theory is the statement that these failures, summed over all places, cancel. That a single cohomological invariant simultaneously classifies division algebras and underwrites reciprocity is the structural unity Lang's text was written to display.

Bibliography Master

@book{lang2002algebra,
  author    = {Lang, Serge},
  title     = {Algebra},
  edition   = {Revised 3rd},
  series    = {Graduate Texts in Mathematics},
  volume    = {211},
  publisher = {Springer},
  year      = {2002}
}

@book{serre1979localfields,
  author    = {Serre, Jean-Pierre},
  title     = {Local Fields},
  series    = {Graduate Texts in Mathematics},
  volume    = {67},
  publisher = {Springer},
  year      = {1979},
  note      = {Translated by Marvin Jay Greenberg}
}

@book{gilleszamuely2017,
  author    = {Gille, Philippe and Szamuely, Tam\'as},
  title     = {Central Simple Algebras and Galois Cohomology},
  edition   = {2nd},
  series    = {Cambridge Studies in Advanced Mathematics},
  volume    = {165},
  publisher = {Cambridge University Press},
  year      = {2017}
}

@book{hilbert1897zahlbericht,
  author    = {Hilbert, David},
  title     = {Die Theorie der algebraischen Zahlk\"orper (Zahlbericht)},
  journal   = {Jahresbericht der Deutschen Mathematiker-Vereinigung},
  volume    = {4},
  pages     = {175--546},
  year      = {1897}
}

@article{brauer1929,
  author  = {Brauer, Richard},
  title   = {\"Uber Systeme hyperkomplexer Zahlen},
  journal = {Mathematische Zeitschrift},
  volume  = {30},
  pages   = {79--107},
  year    = {1929}
}

Prerequisites

01.02.06
01.02.10
03.01.02

Tier anchors

beginner: Conrad — Math 121 notes on Hilbert 90 (informal cyclic-extension picture); Stewart Galois Theory Ch. 21 (norm and trace, motivating the norm-one problem)
intermediate: Serre Local Fields Ch. X §§1–5 (group cohomology, Hilbert 90); Lang Algebra Ch. VI §10 and Ch. XVIII §4 (the H^1=0 theorem, crossed products)
master: Serre Local Fields Ch. X and Galois Cohomology Ch. I–II; Gille–Szamuely Central Simple Algebras and Galois Cohomology Ch. 2–4; Lang Algebra Ch. XVIII (Brauer group via crossed products)

References

Lang — Algebra (Springer GTM 211, rev. 3rd ed., 2002) · Ch. VI §10 (Hilbert's Theorem 90, cyclic and cohomological forms), Ch. XVIII §4 (central simple algebras, crossed products, the Brauer group)
Serre — Local Fields (Springer GTM 67, 1979) · Ch. X §§1–7 (cohomology of finite and profinite groups, Hilbert 90, the Brauer group of a local field)
Gille and Szamuely — Central Simple Algebras and Galois Cohomology (Cambridge Studies 165, 2nd ed., 2017) · Ch. 2 (central simple algebras, Brauer group), Ch. 4 (Galois cohomology, Hilbert 90, the crossed-product isomorphism Br(k) ≅ H^2)
Hilbert — Die Theorie der algebraischen Zahlkörper (Zahlbericht, 1897) · Satz 90: the norm-one theorem for a cyclic extension, the source of the modern Theorem 90

Estimated time

beginner: 18m
intermediate: 45m
master: 85m