21.02.09 · number-theory / quadratic-forms-local-fields

The Brauer-Manin obstruction

shipped3 tiersLean: none

Anchor (Master): Manin 1971 *Actes ICM Nice* Vol. 1 (originator); Grothendieck 1968 *Le groupe de Brauer I-III* (in *Dix exposés sur la cohomologie des schémas*, North-Holland) (Br(X) = H^2_et(X, G_m)); Colliot-Thélène and Sansuc 1977-1987 (conic bundles, Châtelet surfaces, the conjecture that Brauer-Manin is the only obstruction for rational surfaces); Skorobogatov 2001 *Torsors and Rational Points*; Skorobogatov 1999 *Invent. Math.* 135 (Brauer-Manin is not the only obstruction); Colliot-Thélène and Skorobogatov 2021 *The Brauer-Grothendieck Group*

Intuition Beginner

A Diophantine equation asks for solutions in whole numbers or fractions. There is a tempting strategy for showing solutions exist: check that the equation is solvable "locally" — solvable over the real numbers, and solvable in each $p$ -adic world (which amounts to solvability modulo every power of every prime). If a solution survives every local test, one hopes a genuine rational solution exists. When this hope holds, we say the equation satisfies the local-global principle, also called the Hasse principle.

For quadratic equations the hope is correct, and that is the content of the Hasse-Minkowski theorem. But for cubic and higher-degree equations the hope can fail. There are equations with solutions in every local world and yet no solution in rational numbers at all. The earliest clean example is a genus-one curve found by Reichardt and Lind; the famous cubic example is Selmer's $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ .

Why would an equation pass every local test yet have no global solution? Manin's discovery in 1971 was that a hidden bookkeeping device explains many such failures. Attached to a geometric object is an algebraic gadget called its Brauer group. Each candidate local solution can be fed into a Brauer-group element, producing a small fractional number at each place. Class field theory demands that these fractions add up to zero whenever the local solutions come from one global rational solution. If no consistent family of local solutions can make the sum vanish, no rational point can exist — even though each place individually is happy.

Visual Beginner

Picture a variety $X$ as a single dot at the centre, and around it a ring of local worlds: the real place and one world for each prime $p$ . A candidate "adelic point" is a choice of one local point in each world at once. Now overlay a measuring instrument — a Brauer-group element $A$ . Feeding the local point at place $v$ into $A$ produces a small fraction $inv_{v}$ , a number in $Q / Z$ . The reciprocity law of class field theory says: for a family of local points that comes from a single rational point, these fractions sum to zero around the ring.

The Brauer-Manin set is the collection of adelic points whose fractions sum to zero for every measuring instrument at once. Rational points always land inside it. When the set is empty but local points still exist, the equation has passed every individual local test yet cannot assemble those tests into a global solution. That empty-but-locally-nonempty situation is the obstruction. The picture captures the essential move: a global accounting rule rejects local data that no rational point could ever produce.

Worked example Beginner

Consider the Hasse principle failing for Selmer's cubic $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ . We do not redo Selmer's hard descent; instead we walk through what the obstruction reports.

Step 1. Local solvability. At the real place, the cubic is indefinite — mixed signs let one balance terms — so a real point exists. At each prime $p$ away from $2, 3, 5$ the curve reduces to a smooth cubic over $F_{p}$ , which has points by the Hasse bound, and these lift $p$ -adically. At $p = 2, 3, 5$ small hand computations again find local points. So the adelic point set is non-empty.

Step 2. The measuring instrument. Attached to this curve is a non-constant Brauer-group element $A$ of order three (built from the cube roots governing the equation). Evaluating $A$ at a local point produces a fraction in $\frac{1}{3} Z / Z = {0, \frac{1}{3}, \frac{2}{3}}$ at each place.

Step 3. The accounting. One computes that for every choice of adelic point, the three fractions that are non-zero (concentrated at the primes $3$ and $5$ and one more place) add up to $\frac{1}{3}$ or $\frac{2}{3}$ — never to zero. Reciprocity forbids a rational point from giving a non-zero sum.

Step 4. The verdict. Since no adelic point makes the sum vanish, the Brauer-Manin set is empty while the adelic set is not. The equation is everywhere locally solvable yet has no rational point, and the obstruction has explained the failure.

What this tells us: local solvability is necessary but not sufficient. A single global accounting rule, invisible at any one place, can rule out rational solutions that each local world would have allowed.

Check your understanding Beginner

Formal definition Intermediate+

Let $K$ be a number field with adele ring $A_{K} = \prod_{v}^{'} K_{v}$ (restricted product over the places $v$ ), and let $X$ be a smooth, projective, geometrically integral variety over $K$ . The set of adelic points $X (A_{K})$ carries a natural topology; for $X$ projective it agrees with $\prod_{v} X (K_{v})$ , and it is non-empty exactly when $X$ has a $K_{v}$ -point at every place. The diagonal gives an inclusion of the rational points $X (K) ↪ X (A_{K})$ .

The Brauer group $Br (X)$ is the group of Azumaya algebras over $X$ modulo Morita equivalence, identified by Grothendieck with the cohomological Brauer group, the torsion subgroup of the étale cohomology group $H_{et}^{2} (X, G_{m})$ . For a field $k$ this specialises to $Br (k) = H^{2} (Gal (\overset{ˉ}{k} / k), \overset{ˉ}{k}^{\times})$ , the group of central simple algebras over $k$ . A class $A \in Br (X)$ can be evaluated at a point $P \in X (K_{v})$ by pulling back along $P : Spec K_{v} \to X$ , yielding $A (P) \in Br (K_{v})$ .

Local class field theory provides the local invariant maps, injections $$ \mathrm{inv}_v : \mathrm{Br}(K_v) \hookrightarrow \mathbb{Q}/\mathbb{Z}, $$ with $inv_{v}$ an isomorphism for non-archimedean $v$ , $inv_{R} : Br (R) \sim \frac{1}{2} Z / Z$ , and $Br (C) = 0$ . The reciprocity law of global class field theory is the exactness of $$ 0 \to \mathrm{Br}(K) \to \bigoplus_v \mathrm{Br}(K_v) \xrightarrow{\sum_v \mathrm{inv}_v} \mathbb{Q}/\mathbb{Z} \to 0, $$ the Albert-Brauer-Hasse-Noether sequence; its central content is the sum-zero law $\sum_{v} inv_{v} (A) = 0$ for every $A \in Br (K)$ .

The Brauer-Manin pairing is $$ X(\mathbb{A}_K) \times \mathrm{Br}(X) \to \mathbb{Q}/\mathbb{Z}, \qquad \big((P_v)_v,, A\big) \mapsto \sum_v \mathrm{inv}_v\big(A(P_v)\big). $$ For a fixed $A$ only finitely many local terms are non-zero (a constructibility / good-reduction fact), so the sum is finite. The Brauer-Manin set is the left kernel cut out by the whole Brauer group: $$ X(\mathbb{A}_K)^{\mathrm{Br}} = \big{ (P_v) \in X(\mathbb{A}_K) : \textstyle\sum_v \mathrm{inv}_v(A(P_v)) = 0 \text{ for all } A \in \mathrm{Br}(X) \big}. $$ By the sum-zero law applied to constant classes and functoriality, $X (K) \subseteq X (A_{K})^{Br} \subseteq X (A_{K})$ .

Counterexamples to common slips

Only the quotient $Br (X) / Br (K)$ matters. A constant class pulled back from $Br (K)$ pairs to $\sum_{v} inv_{v}$ of a fixed global element, which is zero by reciprocity for every adelic point; so constant classes impose no condition and the obstruction is governed by $Br (X) / Br (K)$ .
The pairing needs the projective (or smooth proper) hypothesis to make $X (A_{K}) = \prod_{v} X (K_{v})$ with the correct topology; for open varieties one must use the restricted product and the evaluation map need not be locally constant in the naive sense.
An empty Brauer-Manin set with $X (A_{K}) \neq = \emptyset$ proves there is no rational point, but a non-empty Brauer-Manin set does not prove a rational point exists: Skorobogatov 1999 exhibits $X (A_{K})^{Br} \neq = \emptyset$ yet $X (K) = \emptyset$ .

Key theorem with proof Intermediate+

Theorem (Manin 1971). Let $X$ be a smooth projective geometrically integral variety over a number field $K$ . Then $X (K) \subseteq X (A_{K})^{Br}$ . Consequently, if $X (A_{K}) \neq = \emptyset$ while $X (A_{K})^{Br} = \emptyset$ , the variety $X$ has no $K$ -rational point although it is everywhere locally solvable; the emptiness of the Brauer-Manin set is the obstruction to the Hasse principle.

Proof. Fix $A \in Br (X)$ and a rational point $P \in X (K)$ . View $P$ simultaneously as a $K_{v}$ -point at every place via the diagonal $X (K) ↪ \prod_{v} X (K_{v})$ . Evaluation is functorial: the local value $A (P) \in Br (K_{v})$ is the image of the single global class $A (P) \in Br (K)$ under the localisation map $Br (K) \to Br (K_{v})$ . Here $A (P)$ on the left denotes the pullback of $A$ along $P : Spec K \to X$ , a well-defined element of $Br (K)$ because $P$ is a single $K$ -morphism.

Now compute the pairing of the adelic point $(P)_{v}$ (the constant family given by $P$ ) against $A$ : $$ \sum_v \mathrm{inv}_v\big(A(P)_v\big) = \sum_v \mathrm{inv}_v\big(\mathrm{loc}_v(A(P))\big) = \sum_v \mathrm{inv}_v\big(A(P)\big), $$ where the last sum is the reciprocity sum of the fixed global Brauer class $A (P) \in Br (K)$ . By the Albert-Brauer-Hasse-Noether reciprocity law — the exactness of $Br (K) \to ⨁_{v} Br (K_{v}) \sum inv_{v} Q / Z$ — this sum vanishes. The element $A$ was arbitrary, so the adelic point coming from $P$ pairs to zero with all of $Br (X)$ , hence lies in $X (A_{K})^{Br}$ . This proves $X (K) \subseteq X (A_{K})^{Br}$ .

For the consequence, suppose $X (A_{K})^{Br} = \emptyset$ . If $X$ had a rational point $P$ , the inclusion just proved would place $P$ (as an adelic point) inside the empty set, an impossibility. So $X (K) = \emptyset$ . When additionally $X (A_{K}) \neq = \emptyset$ , the variety is everywhere locally solvable yet has no rational point, and the Hasse principle fails for $X$ . The finiteness of the pairing sum — only finitely many $inv_{v} (A (P_{v}))$ are non-zero for fixed $A$ — follows because $A$ extends to an Azumaya algebra over a model of $X$ over the ring of $S$ -integers for a finite set $S$ , and at every place outside $S$ the evaluation $A (P_{v})$ is unramified, hence has $inv_{v} = 0$ . $□$

Bridge. This argument builds toward the entire modern theory of rational points, and the foundational reason it works is exactly the reciprocity sum-zero law $\sum_{v} inv_{v} = 0$ on $Br (K)$ , the same global constraint that powers Hasse-Minkowski. The central insight is that a rational point produces one global Brauer class whose local invariants are forced to cancel, so any adelic family whose invariants refuse to cancel cannot come from a rational point. This is exactly the mechanism that appears again in the Hilbert product formula $\prod_{v} (a, b)_{v} = 1$ of 21.02.05, which is the degree-two, order-two shadow of the same reciprocity sequence; the Brauer-Manin pairing generalises that product formula from a single quaternion symbol to the full Brauer group of a variety. Putting these together, the Hasse principle for quadratic forms in 21.02.08 and its failure for cubics become one statement: the kernel of localisation on $Br$ is captured by the sum-zero relation, and the Brauer-Manin set is dual to the rational points through exactly this pairing. The bridge is the recognition that local solvability data lives in $⨁_{v} Br (K_{v})$ and rationality is detected by where the reciprocity map sends it.

Exercises Intermediate+

Exercise 3 (medium, short-answer).

Explain why, for a fixed $A \in Br (X)$ , only finitely many places contribute to $\sum_{v} inv_{v} (A (P_{v}))$ .

Hint

$A$ spreads out to an Azumaya algebra over a model of $X$ over $O_{K, S}$ for some finite $S$ . What is $inv_{v}$ of an unramified class?

Answer

The class $A$ extends to an Azumaya algebra over an integral model $X$ of $X$ over the ring of $S$ -integers $O_{K, S}$ for a finite set of places $S$ . For $v \in / S$ and any $P_{v} \in X (K_{v})$ , the point reduces to an $O_{v}$ -point of $X$ (by properness), so $A (P_{v}) \in Br (K_{v})$ is unramified, i.e. comes from $Br (O_{v}) = Br (residue field) = 0$ for the residue field finite. Hence $inv_{v} (A (P_{v})) = 0$ for all $v \in / S$ , leaving a finite sum. Rubric: full credit for " $A$ spreads out; unramified evaluations have invariant zero outside a finite $S$ ".

Exercise 4 (medium, short-answer).

Show that the Hilbert product formula $\prod_{v} (a, b)_{v} = 1$ is a special case of the reciprocity sum-zero law $\sum_{v} inv_{v} = 0$ .

Hint

The quaternion algebra $(a, b / K)$ defines a class in $Br (K) [2]$ ; its local invariant is $0$ or $\frac{1}{2}$ according to the Hilbert symbol value.

Answer

The quaternion algebra $(a, b / K)$ is a central simple algebra of order dividing two, hence a class in $Br (K) [2]$ . Its localisation at $v$ has invariant $inv_{v} (a, b / K) = 0$ if $(a, b)_{v} = + 1$ (the algebra splits over $K_{v}$ ) and $inv_{v} (a, b / K) = \frac{1}{2}$ if $(a, b)_{v} = - 1$ . Under the identification $\frac{1}{2} Z / Z ≅ Z /2$ , the symbol $(a, b)_{v} = (- 1)^{2 inv_{v}}$ . The reciprocity law $\sum_{v} inv_{v} ((a, b / K)) = 0$ in $Q / Z$ then says the number of places with $inv_{v} = \frac{1}{2}$ is even, i.e. $\prod_{v} (a, b)_{v} = (- 1)^{# {v : (a, b)_{v} = - 1}} = + 1$ . Rubric: full credit for translating $inv_{v} \in {0, \frac{1}{2}}$ to the Hilbert symbol and reading sum-zero as even-count.

Exercise 5 (medium, short-answer).

A Châtelet surface is given by $y^{2} - a z^{2} = P (x)$ with $a \in K^{\times}$ and $P$ a separable quartic. Describe how one builds a Brauer class on it and what data the obstruction uses.

Hint

Think of the quaternion symbol $(a, P (x))$ as a function on the surface; where it is defined it gives a Brauer element.

Answer

On the function field, the quaternion algebra $(a, P (x))$ defines an element of $Br (K (X))$ . Because $y^{2} - a z^{2} = P (x)$ , on the surface $P (x)$ is a norm from $K (a)$ wherever $(y, z) \neq = 0$ , so the symbol $(a, P (x))$ extends to an unramified class $A \in Br (X)$ (after checking residues vanish along each component, using that $P$ factors and the conic $y^{2} - a z^{2} = P (x)$ controls ramification). The obstruction uses: the factorisation type of $P$ over $K$ and over each $K_{v}$ , the local Hilbert symbols $(a, P (x_{v}))_{v}$ at adelic points, and the demand that $\sum_{v} inv_{v} (A (P_{v})) = 0$ . Colliot-Thélène-Sansuc showed for Châtelet surfaces this single class governs everything: Brauer-Manin is the only obstruction. Rubric: full credit for " $(a, P (x))$ quaternion class, evaluate via local Hilbert symbols, sum-zero condition".

Exercise 6 (hard, short-answer).

State precisely what "Brauer-Manin is the only obstruction to the Hasse principle for $X$ " means, and give one class of varieties where it is a theorem and one example where it is known to fail.

Hint

The statement is an implication about $X (A_{K})^{Br} \neq = \emptyset$ . Rational surfaces (Colliot-Thélène-Sansuc) versus Skorobogatov's bielliptic example.

Answer

"Brauer-Manin is the only obstruction to the Hasse principle for $X$ " means: $X (A_{K})^{Br} \neq = \emptyset ⟹ X (K) \neq = \emptyset$ . Equivalently, the only way local points fail to assemble into a rational point is the Brauer-Manin sum-zero failure. Theorem case: Colliot-Thélène-Sansuc (with Swinnerton-Dyer for the analytic input) proved this for Châtelet surfaces and, conjecturally extended, for all smooth rational surfaces and conic bundles over $P^{1}$ ; the Châtelet case is unconditional. Failure case: Skorobogatov 1999 constructed a smooth projective surface (a quotient of a product of curves, bielliptic-type) with $X (A_{K})^{Br} \neq = \emptyset$ but $X (K) = \emptyset$ , so Brauer-Manin alone leaves a residual obstruction, repaired by the finer étale-Brauer (descent) obstruction. Rubric: full credit for the implication, the Châtelet/rational-surface theorem, and the Skorobogatov failure.

Exercise 7 (hard, short-answer).

The Reichardt-Lind curve $2 y^{2} = z^{4} - 17 w^{4}$ is everywhere locally solvable but has no rational point. Sketch why this is consistent with the genus-one theory and which obstruction detects it.

Hint

A genus-one curve with points everywhere locally is a torsor under its Jacobian elliptic curve, representing a class in the Tate-Shafarevich group.

Answer

The curve $C : 2 y^{2} = z^{4} - 17 w^{4}$ has genus one and is locally solvable at every place (small Hensel computations at $2, 17$ and the real place, smooth reduction elsewhere). A genus-one curve that is everywhere locally solvable is a torsor under its Jacobian elliptic curve $E$ , and its class lies in $\Sha (E / Q)$ , the Tate-Shafarevich group — the group measuring failure of the Hasse principle for $E$ -torsors. The Reichardt-Lind class is a non-zero element of $\Sha (E / Q)$ , so $C$ has no rational point. The Brauer-Manin obstruction detects exactly this: for torsors under abelian varieties, the Brauer-Manin obstruction is equivalent to the Tate-Shafarevich obstruction (Manin's original observation), so $C (A_{Q})^{Br} = \emptyset$ . Rubric: full credit for "genus-one torsor, non-zero $\Sha$ class, Brauer-Manin equals the $\Sha$ obstruction here".

Advanced results Master

Theorem (Manin 1971; the obstruction). For a smooth projective geometrically integral variety $X$ over a number field $K$ , the Brauer-Manin set satisfies $X (K) \subseteq \overline{X (K)} \subseteq X (A_{K})^{Br} \subseteq X (A_{K})$ , where the closure is taken in the adelic topology. The obstruction to the Hasse principle is the failure of $X (A_{K})^{Br} \neq = \emptyset \Rightarrow X (K) \neq = \emptyset$ ; the obstruction to weak approximation is the failure of $X (A_{K})^{Br} = X (A_{K})$ .

The pairing factors through $Br (X) / Br (K)$ , and for many varieties this quotient is finite and computable. The algebraic part $Br_{1} (X) = ker (Br (X) \to Br (\overset{ˉ}{X}))$ fits into the exact sequence from the Hochschild-Serre spectral sequence $0 \to Pic (X) \to Pic (\overset{ˉ}{X})^{G_{K}} \to Br (K) \to Br_{1} (X) \to H^{1} (G_{K}, Pic (\overset{ˉ}{X})) \to H^{3} (G_{K}, \overset{ˉ}{K}^{\times})$ , identifying the algebraic Brauer quotient with the Galois cohomology group $H^{1} (G_{K}, Pic (\overset{ˉ}{X}))$ when $H^{3} (G_{K}, \overset{ˉ}{K}^{\times}) = 0$ (always for number fields with no real place issue). This is the computational engine: the obstruction reduces to a Galois-cohomology calculation on the geometric Picard group.

Theorem (Colliot-Thélène-Sansuc; Châtelet surfaces). For a Châtelet surface $X : y^{2} - a z^{2} = P (x)$ over a number field, with $P$ a separable quartic, the Brauer-Manin obstruction is the only obstruction to the Hasse principle and to weak approximation: $X (A_{K})^{Br} \neq = \emptyset$ implies $X (K)$ is dense in $X (A_{K})^{Br}$ .

The proof (Colliot-Thélène, Sansuc, Swinnerton-Dyer, in two long papers 1987) combines the descent formalism with an analytic fibration argument over the base $P^{1}$ , using a delicate application of additive characters and a result on prime values of binary forms in the spirit of Dirichlet. The result was the first systematic confirmation that the obstruction is sharp for a positive-dimensional family of surfaces, and it underlies the conjecture (Colliot-Thélène) that Brauer-Manin is the only obstruction for all rationally connected varieties.

Theorem (Skorobogatov 1999; insufficiency). There exists a smooth projective surface $X$ over $Q$ , a quotient of a product of two curves by a finite group action, with $X (A_{Q})^{Br} \neq = \emptyset$ yet $X (Q) = \emptyset$ . The Brauer-Manin obstruction is therefore not the only obstruction to the Hasse principle in general.

Skorobogatov's example is repaired by the étale-Brauer obstruction: one applies the Brauer-Manin condition not to $X$ directly but to every twist of $X$ by torsors under finite étale group schemes, intersecting the resulting sets. This descent-refined set $X (A_{K})^{\overset{e}{ˊ} t,Br}$ is empty for his surface. Whether the étale-Brauer obstruction is itself the only obstruction for all varieties was open for a decade until Poonen 2010 produced a variety where even the étale-Brauer set is non-empty with no rational point, ending the search for a single universal cohomological obstruction.

Theorem (relation to Tate-Shafarevich). Let $C$ be a torsor under an abelian variety $A$ over $K$ , everywhere locally solvable, representing a class $[C] \in \Sha (A / K) \subseteq H^{1} (K, A)$ . Then the Brauer-Manin obstruction on $C$ is non-empty if and only if $[C] \neq = 0$ , assuming the Tate-Shafarevich group is finite. For such $C$ the Brauer-Manin obstruction is the only obstruction to the Hasse principle.

This is the abelian-variety case where the picture is cleanest: the obstruction is governed by the Cassels-Tate pairing on $\Sha$ , and the local invariants of the relevant Brauer classes recover the pairing of $[C]$ against itself. The Reichardt-Lind and Selmer examples are both instances: each represents a non-zero element of a Tate-Shafarevich group, and the Brauer-Manin obstruction detects the non-vanishing.

Synthesis. Putting these together, the Brauer-Manin obstruction is the foundational reason the local-global heuristic of 21.02.08 succeeds for quadratic forms and fails for cubics: this is exactly the statement that the reciprocity sum-zero law on the Brauer group, which is dual to the rational points through the evaluation pairing, generalises the Hilbert product formula from a single order-two symbol to the entire Brauer group of a variety. The central insight is that local solvability data is an element of $⨁_{v} Br (K_{v})$ and rationality is detected by where global reciprocity sends it; the obstruction is non-empty precisely when the local data refuses to lie in the image of $Br (K)$ . This is exactly the mechanism that recurs in the Tate-Shafarevich group, where the Cassels-Tate pairing is dual to the Brauer-Manin pairing on torsors, and the bridge is the recognition that $\Sha$ , the Brauer-Manin set, and the étale-Brauer refinement are three faces of one cohomological accounting. The pattern recurs throughout arithmetic geometry: weak approximation, the integral Hasse principle, and the Colliot-Thélène conjecture for rationally connected varieties are all governed by the same pairing, and the bridge from the elementary Hilbert symbol to the modern theory of rational points is the single observation that reciprocity is a sum-to-zero law on $Br$ .

Full proof set Master

Proposition 1 (the pairing is well-defined and finite). For a smooth projective $X / K$ and $A \in Br (X)$ , the function $v \mapsto inv_{v} (A (P_{v}))$ is zero for all but finitely many $v$ , so $\sum_{v} inv_{v} (A (P_{v})) \in Q / Z$ is defined for every adelic point $(P_{v})$ .

Proof. Choose a finite set of places $S$ , containing the archimedean places, such that $A$ extends to an Azumaya algebra $A$ over a smooth proper model $X \to Spec O_{K, S}$ of $X$ ; such an $S$ exists because $A$ is represented by an Azumaya algebra over a Zariski-open of $X$ and the obstruction to extending spreads over only finitely many primes. For $v \in / S$ , properness of $X$ over $O_{v}$ gives that any $P_{v} \in X (K_{v}) = X (O_{v})$ extends to an $O_{v}$ -point, so $A (P_{v})$ lies in the image of $Br (O_{v})$ . The Brauer group of a complete local ring with finite residue field $κ (v)$ satisfies $Br (O_{v}) = Br (κ (v)) = 0$ (Brauer groups of finite fields vanish). Hence $A (P_{v}) = 0$ and $inv_{v} (A (P_{v})) = 0$ for $v \in / S$ . The sum is finite. $□$

Proposition 2 (rational points pair to zero). For every $A \in Br (X)$ and $P \in X (K)$ , $\sum_{v} inv_{v} (A (P)) = 0$ .

Proof. The point $P$ is a $K$ -morphism $Spec K \to X$ , so $A (P) := P^{*} A \in Br (K)$ is a single global class. The diagonal $X (K) \to \prod_{v} X (K_{v})$ sends $P$ to the constant family $(P)_{v}$ , and evaluation commutes with the localisation maps: $A (P)_{v} = loc_{v} (A (P))$ . Therefore $$ \sum_v \mathrm{inv}_v(A(P)_v) = \sum_v \mathrm{inv}_v(\mathrm{loc}_v(A(P))). $$ The right-hand sum is the composite $Br (K) \to ⨁_{v} Br (K_{v}) \to Q / Z$ applied to $A (P)$ , which is zero because the sequence $0 \to Br (K) \to ⨁_{v} Br (K_{v}) \sum inv_{v} Q / Z \to 0$ is a complex (the composite of consecutive maps is zero) — this is the Albert-Brauer-Hasse-Noether reciprocity law. $□$

Proposition 3 (constant classes impose no condition). If $A \in Br (K) \subseteq Br (X)$ is the pullback of a global class along $X \to Spec K$ , then $\sum_{v} inv_{v} (A (P_{v})) = 0$ for every adelic point $(P_{v})$ , not only for rational points.

Proof. For a constant class $A = π^{*} A_{0}$ with $A_{0} \in Br (K)$ and $π : X \to Spec K$ , the evaluation at any $P_{v} \in X (K_{v})$ factors through the structure map: $A (P_{v}) = (π \circ P_{v})^{*} A_{0} = loc_{v} (A_{0})$ , independent of $P_{v}$ (the composite $π \circ P_{v}$ is the structure morphism $Spec K_{v} \to Spec K$ ). Hence $\sum_{v} inv_{v} (A (P_{v})) = \sum_{v} inv_{v} (loc_{v} (A_{0})) = 0$ by reciprocity. So the pairing depends on $A$ only through its image in $Br (X) / Br (K)$ , and only the non-constant part of the Brauer group can obstruct. $□$

Proposition 4 (the obstruction implies no rational point). If $X (A_{K}) \neq = \emptyset$ and $X (A_{K})^{Br} = \emptyset$ , then $X (K) = \emptyset$ while $X$ is everywhere locally solvable.

Proof. Everywhere local solvability is the statement $X (K_{v}) \neq = \emptyset$ for all $v$ , which is $X (A_{K}) \neq = \emptyset$ for $X$ projective. If $X (K)$ contained a point $P$ , then by Proposition 2 the adelic point $(P)_{v}$ would pair to zero with every $A \in Br (X)$ , so $(P)_{v} \in X (A_{K})^{Br}$ , contradicting $X (A_{K})^{Br} = \emptyset$ . Hence $X (K) = \emptyset$ , and the Hasse principle fails for $X$ . $□$

Proposition 5 (Hilbert product formula as the order-two case). Let $C : z^{2} = a x^{2} + b y^{2}$ be a conic over $K$ with $a, b \in K^{\times}$ . Then $C$ has a $K$ -point iff $(a, b)_{v} = + 1$ for all $v$ , and the Brauer-Manin obstruction on $C$ is governed by the single quaternion class $(a, b / K)$ , recovering $\prod_{v} (a, b)_{v} = 1$ .

Proof. A smooth conic over $K$ is a Severi-Brauer variety of dimension one, associated to the quaternion algebra $Q = (a, b / K)$ ; $C ≅ P_{K}^{1}$ iff $Q$ splits iff $C (K) \neq = \emptyset$ . Locally, $C (K_{v}) \neq = \emptyset$ iff $Q \otimes K_{v}$ splits iff $inv_{v} (Q) = 0$ iff $(a, b)_{v} = + 1$ . So $C$ is everywhere locally solvable iff $inv_{v} (Q) = 0$ for all $v$ . The reciprocity law gives $\sum_{v} inv_{v} (Q) = 0$ ; with each term in ${0, \frac{1}{2}}$ , the number of places where $inv_{v} (Q) = \frac{1}{2}$ is even, which is $\prod_{v} (a, b)_{v} = + 1$ . If all local symbols are $+ 1$ then all $inv_{v} (Q) = 0$ , so $Q \in Br (K)$ is locally zero everywhere, hence zero by injectivity of $Br (K) \to ⨁_{v} Br (K_{v})$ , so $C (K) \neq = \emptyset$ . The conic satisfies the Hasse principle, and the Brauer-Manin obstruction (here a single class) vanishes precisely on the rational points. $□$

Connections Master

Hasse-Minkowski theorem 21.02.08. Hasse-Minkowski is the statement that the Hasse principle holds for quadratic forms, equivalently for the conics and quadrics they cut out. The Brauer-Manin obstruction is the precise measure of how and why the Hasse principle fails once one leaves the quadratic world. The two units are dual halves of one story: where Hasse-Minkowski certifies that local solvability assembles into a rational solution, the obstruction quantifies the cohomological reason this assembly can break for cubics like Selmer's curve and genus-one curves like Reichardt-Lind. The same reciprocity sum-zero law underlies both — it forces the global symbol to vanish in the quadratic case and detects the residual class in the higher-degree case.
Hilbert symbol and the product formula 21.02.05. The Hilbert symbol $(a, b)_{v}$ is the simplest local invariant, and the product formula $\prod_{v} (a, b)_{v} = 1$ is the order-two, degree-two shadow of the Brauer-Manin pairing. A quaternion algebra is a Brauer class of order two; its local invariants are $0$ or $\frac{1}{2}$ , encoding the Hilbert symbol, and reciprocity is the product formula. The Brauer-Manin obstruction generalises this from one symbol attached to two elements to the full Brauer group attached to a variety, replacing the product formula by the evaluation pairing $X (A_{K}) \times Br (X) \to Q / Z$ .
$ℓ$ -adic Galois representations 21.05.01. The Brauer group $Br (X) = H_{et}^{2} (X, G_{m})_{tors}$ lives in étale cohomology, the same cohomology theory that carries the $ℓ$ -adic Galois representations of $Gal (\overset{ˉ}{K} / K)$ . The algebraic part of the obstruction is computed from the Galois action on the geometric Picard group $Pic (\overset{ˉ}{X})$ via the group $H^{1} (G_{K}, Pic (\overset{ˉ}{X}))$ , a Galois-cohomology calculation of exactly the kind that the theory of Galois representations systematises. The transcendental part of $Br (X)$ relates to the $H^{2}$ Galois representation and to the image of $ℓ$ -adic cohomology.
Modularity and BSD 21.06.01. For a torsor under an elliptic curve $E / K$ , the Brauer-Manin obstruction is equivalent to the Tate-Shafarevich obstruction: a non-zero class in $\Sha (E / K)$ is exactly an everywhere-locally-solvable torsor with no rational point, detected by the Brauer-Manin pairing through the Cassels-Tate pairing. The conjectural finiteness of $\Sha (E / K)$ is part of the refined Birch-Swinnerton-Dyer conjecture, so the Brauer-Manin obstruction for genus-one curves is the geometric face of the arithmetic of elliptic curves that BSD governs.

Historical & philosophical context Master

The recognition that local solvability does not imply global solvability is older than the cohomological language that explains it. Lind's 1940 Uppsala dissertation and Reichardt's 1942 paper Einige im Kleinen überall lösbare, im Grossen unlösbare diophantische Gleichungen ^{[Reichardt 1942]} produced the genus-one curve $2 y^{2} = z^{4} - 17 w^{4}$ , everywhere locally solvable with no rational point — the first clean counterexample to the Hasse principle, predating by a decade Selmer's 1951 cubic $3 x^{3} + 4 y^{3} + 5 z^{3} = 0$ ^{[Selmer 1951]}. These examples raised a question without a framework: what cohomological invariant detects the failure? Châtelet's 1944 study of Severi-Brauer varieties ^{[Châtelet 1944]} supplied the geometric objects — conic bundles and the surfaces $y^{2} - a z^{2} = P (x)$ — on which the answer would be tested, but the Brauer group of a scheme did not yet exist as a tool.

The framework arrived with Grothendieck's three-part Le groupe de Brauer ^{[Grothendieck 1968]}, identifying the Brauer group of a scheme with the torsion of $H_{et}^{2} (X, G_{m})$ and giving it the functoriality needed for evaluation at points. Manin's 1970 ICM address at Nice, published 1971 as Le groupe de Brauer-Grothendieck en géométrie diophantienne ^{[Manin 1971]}, combined Grothendieck's Brauer group with the reciprocity law of class field theory to define the pairing and the obstruction, and showed it accounts for the Reichardt-Lind and Selmer failures. The decisive vindication came from Colliot-Thélène and Sansuc ^{[Colliot-Thélène and Sansuc 1987]}, who with Swinnerton-Dyer proved that for Châtelet surfaces the Brauer-Manin obstruction is the only obstruction, turning a heuristic into a sharp theorem for a positive-dimensional family and motivating the conjecture that the same holds for all rationally connected varieties.

Philosophically, the Brauer-Manin obstruction reframes the local-global question as a problem about the failure of a localisation map to be surjective. Solutions live globally; tests live locally; the gap between them is a cohomology group. The arc from Reichardt 1942 through Manin 1971 to Skorobogatov 1999 ^{[Skorobogatov 2001]} — who showed the obstruction is itself sometimes insufficient, requiring the finer étale-Brauer refinement — traces a single idea deepening: that the obstructions to rational points are organised by the cohomology of the absolute Galois group acting on the geometry of the variety, and that no one cohomological invariant need be final.

Bibliography Master

@incollection{Manin1971,
  author    = {Manin, Yu. I.},
  title     = {Le groupe de {B}rauer-{G}rothendieck en g{\'e}om{\'e}trie diophantienne},
  booktitle = {Actes du Congr{\`e}s International des Math{\'e}maticiens, Nice 1970},
  volume    = {1},
  publisher = {Gauthier-Villars},
  address   = {Paris},
  year      = {1971},
  pages     = {401--411}
}

@incollection{Grothendieck1968,
  author    = {Grothendieck, Alexander},
  title     = {Le groupe de {B}rauer {I}, {II}, {III}},
  booktitle = {Dix expos{\'e}s sur la cohomologie des sch{\'e}mas},
  publisher = {North-Holland},
  address   = {Amsterdam},
  year      = {1968},
  pages     = {46--188}
}

@article{ReichardtLind1942,
  author  = {Reichardt, Hans},
  title   = {Einige im {K}leinen {\"u}berall l{\"o}sbare, im {G}rossen unl{\"o}sbare diophantische {G}leichungen},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {184},
  year    = {1942},
  pages   = {12--18},
  note    = {together with C.-E. Lind, Uppsala dissertation 1940}
}

@article{Selmer1951,
  author  = {Selmer, Ernst S.},
  title   = {The {D}iophantine equation $ax^3 + by^3 + cz^3 = 0$},
  journal = {Acta Mathematica},
  volume  = {85},
  year    = {1951},
  pages   = {203--362}
}

@article{Chatelet1944,
  author  = {Ch{\^a}telet, Fran{\c{c}}ois},
  title   = {Variations sur un th{\`e}me de {H}. {P}oincar{\'e}},
  journal = {Annales scientifiques de l'{\'E}cole Normale Sup{\'e}rieure},
  volume  = {61},
  year    = {1944},
  pages   = {249--300}
}

@article{ColliotTheleneSansuc1987,
  author  = {Colliot-Th{\'e}l{\`e}ne, Jean-Louis and Sansuc, Jean-Jacques},
  title   = {La descente sur les vari{\'e}t{\'e}s rationnelles, {II}},
  journal = {Duke Mathematical Journal},
  volume  = {54},
  year    = {1987},
  pages   = {375--492}
}

@article{ColliotTheleneSansucSwinnertonDyer1987,
  author  = {Colliot-Th{\'e}l{\`e}ne, Jean-Louis and Sansuc, Jean-Jacques and Swinnerton-Dyer, Peter},
  title   = {Intersections of two quadrics and {C}h{\^a}telet surfaces, {I}--{II}},
  journal = {Journal f{\"u}r die reine und angewandte Mathematik},
  volume  = {373--374},
  year    = {1987},
  pages   = {37--107, 72--168}
}

@article{Skorobogatov1999,
  author  = {Skorobogatov, Alexei N.},
  title   = {Beyond the {M}anin obstruction},
  journal = {Inventiones Mathematicae},
  volume  = {135},
  year    = {1999},
  pages   = {399--424}
}

@book{Skorobogatov2001,
  author    = {Skorobogatov, Alexei N.},
  title     = {Torsors and Rational Points},
  publisher = {Cambridge University Press},
  series    = {Cambridge Tracts in Mathematics},
  volume    = {144},
  year      = {2001}
}

@book{Poonen2017,
  author    = {Poonen, Bjorn},
  title     = {Rational Points on Varieties},
  publisher = {American Mathematical Society},
  series    = {Graduate Studies in Mathematics},
  volume    = {186},
  year      = {2017}
}

@book{ColliotTheleneSkorobogatov2021,
  author    = {Colliot-Th{\'e}l{\`e}ne, Jean-Louis and Skorobogatov, Alexei N.},
  title     = {The {B}rauer-{G}rothendieck Group},
  publisher = {Springer},
  series    = {Ergebnisse der Mathematik und ihrer Grenzgebiete (3)},
  volume    = {71},
  year      = {2021}
}

@article{Poonen2010,
  author  = {Poonen, Bjorn},
  title   = {Insufficiency of the {B}rauer-{M}anin obstruction applied to {\'e}tale covers},
  journal = {Annals of Mathematics},
  volume  = {171},
  year    = {2010},
  pages   = {2157--2169}
}

@book{Serre1973,
  author    = {Serre, Jean-Pierre},
  title     = {A Course in Arithmetic},
  publisher = {Springer-Verlag},
  series    = {Graduate Texts in Mathematics},
  volume    = {7},
  year      = {1973},
  note      = {English translation of \emph{Cours d'arithm{\'e}tique}, Presses Universitaires de France, 1970}
}

Prerequisites

21.02.08
21.02.05
21.02.07

Tier anchors

beginner: Manin 1974 *Cubic Forms* (North-Holland) informal opening; Poonen 2017 *Rational Points on Varieties* (AMS GSM 186) Ch. 8 informal opening; Serre 1973 *A Course in Arithmetic* Ch. IV §3 (Hasse principle and Selmer's example, informal)
intermediate: Skorobogatov 2001 *Torsors and Rational Points* (Cambridge Tracts 144) Ch. 5; Poonen 2017 *Rational Points on Varieties* Ch. 8; Colliot-Thélène and Skorobogatov 2021 *The Brauer-Grothendieck Group* (Springer Ergebnisse 71) Ch. 13
master: Manin 1971 *Actes ICM Nice* Vol. 1 (originator); Grothendieck 1968 *Le groupe de Brauer I-III* (in *Dix exposés sur la cohomologie des schémas*, North-Holland) (Br(X) = H^2_et(X, G_m)); Colliot-Thélène and Sansuc 1977-1987 (conic bundles, Châtelet surfaces, the conjecture that Brauer-Manin is the only obstruction for rational surfaces); Skorobogatov 2001 *Torsors and Rational Points*; Skorobogatov 1999 *Invent. Math.* 135 (Brauer-Manin is not the only obstruction); Colliot-Thélène and Skorobogatov 2021 *The Brauer-Grothendieck Group*

References

Manin, Yu. I. — Le groupe de Brauer-Grothendieck en géométrie diophantienne · *Actes du Congrès International des Mathématiciens* (Nice, 1970), Vol. 1, Gauthier-Villars (Paris, 1971), pp. 401-411. The originator paper: introduces the obstruction to the Hasse principle obtained by pairing adelic points against the Brauer group via the sum of local invariants, and shows it detects the Selmer counterexample and other failures.
Grothendieck, A. — Le groupe de Brauer I, II, III · In *Dix exposés sur la cohomologie des schémas*, North-Holland (Amsterdam, 1968), pp. 46-188. Establishes the Brauer group of a scheme via Azumaya algebras and the cohomological Brauer group $H^2_{\mathrm{et}}(X, \mathbb{G}_m)$, with the comparison between the two and the basic functoriality used in the Brauer-Manin pairing.
Colliot-Thélène, J.-L. and Sansuc, J.-J. — La descente sur les variétés rationnelles, II · *Duke Mathematical Journal* 54 (1987), 375-492 (with the first part *La descente sur les variétés rationnelles*, in *Journées de géométrie algébrique d'Angers* 1979, Sijthoff & Noordhoff 1980, pp. 223-237). Develops the descent theory on rational varieties and frames the conjecture that the Brauer-Manin obstruction is the only obstruction to the Hasse principle and to weak approximation for rational surfaces, including Châtelet surfaces and conic bundles over $\mathbb{P}^1$.
Skorobogatov, A. N. — Torsors and Rational Points · Cambridge Tracts in Mathematics 144, Cambridge University Press (2001). The canonical modern textbook: Ch. 5 develops the Brauer-Manin obstruction in full, Ch. 6 the descent / étale-Brauer refinement; the standard reference for $X(\mathbb{A}_K)^{\mathrm{Br}}$ and its relation to torsors under algebraic groups.
Skorobogatov, A. N. — Beyond the Manin obstruction · *Inventiones Mathematicae* 135 (1999), 399-424. Constructs a smooth projective surface (a bielliptic-type quotient) with $X(\mathbb{A}_K)^{\mathrm{Br}} \neq \emptyset$ but $X(K) = \emptyset$, showing the Brauer-Manin obstruction is not the only obstruction and motivating the étale-Brauer (descent) obstruction.
Colliot-Thélène, J.-L. and Skorobogatov, A. N. — The Brauer-Grothendieck Group · Springer Ergebnisse der Mathematik (3) 71 (2021). The comprehensive modern monograph: $\mathrm{Br}(X)$ for schemes, the algebraic and transcendental parts, the Brauer-Manin pairing (Ch. 13), and the state of the art on the only-obstruction conjectures.
Selmer, E. S. — The Diophantine equation $ax^3 + by^3 + cz^3 = 0$ · *Acta Mathematica* 85 (1951), 203-362. The canonical counterexample $3 x^3 + 4 y^3 + 5 z^3 = 0$: a smooth plane cubic with points over every $\mathbb{Q}_v$ but no non-zero rational point. Manin's obstruction was constructed to explain exactly this failure.
Reichardt, H. — Einige im Kleinen überall lösbare, im Grossen unlösbare diophantische Gleichungen · *Journal für die reine und angewandte Mathematik* 184 (1942), 12-18; together with Lind, C.-E. (Uppsala dissertation 1940). The Reichardt-Lind genus-one curve $2 y^2 = z^4 - 17 w^4$ (equivalently $y^2 = 17 x^4 - 2$ type): everywhere locally solvable, no rational point — the earliest clean Hasse-principle counterexample, predating Selmer's cubic.
Poonen, B. — Rational Points on Varieties · American Mathematical Society Graduate Studies in Mathematics 186 (2017). Ch. 8 gives a textbook account of the Brauer-Manin obstruction with the adelic-point topology, the evaluation map, and worked Châtelet-surface examples; the most accessible modern entry point.
Serre, J.-P. — A Course in Arithmetic · Springer Graduate Texts in Mathematics 7 (1973). Ch. II-III (local fields, the Hilbert symbol $(a,b)_v$ and its values as elements of $\mathrm{Br}(\mathbb{Q}_v)[2]$); Ch. IV §3 (Hasse-Minkowski, the Hasse principle, Selmer's example as a failure for cubics).
Châtelet, F. — Points rationnels et variétés de Severi-Brauer · *Variations sur un thème de H. Poincaré*, Annales scientifiques de l'École Normale Supérieure 61 (1944), 249-300. Introduces the Severi-Brauer varieties and the surfaces $y^2 - a z^2 = P(x)$ now called Châtelet surfaces, the principal worked family for the Brauer-Manin obstruction and the Colliot-Thélène-Sansuc theorem.

Estimated time

beginner: 20m
intermediate: 50m
master: 95m