Past exam of the mathematics course of the University of Cambridge / 2012 / iii / Paper 28 (Local Fields)

Every Non-Archimedean absolute value on $\mathbb{Q}$ is either the trivial absolute value, or $|\cdot {|}_p^c$ for a unique prime $p$ and some $c>0$. Here $|p{|}_p=p^{-1}$.

To prove this classification, the ultrametric inequality gives $|n|\le 1$ for all integers $n$. If every nonzero integer has absolute value one, multiplicativity makes the absolute value on a field trivial on all rational numbers. Otherwise some positive integer has absolute value less than one, and its prime factorization gives a prime $p$ with $0<|p|<1$. There cannot be two such primes: if $p\ne q$ and $ap+bq=1$ with integers $a,b$, then $1\le \max (|p|,|q|)<1$, a contradiction. Every other prime consequently has absolute value one. Unique prime factorization now givesConversely these are Non-Archimedean absolute values, by the valuation inequality $v_p(x+y)\ge \min (v_p(x),v_p(y))$. Thus $\boxed{|\cdot |=|\cdot {|}_p^c(c>0)\text{ or the trivial value}}$. This is the non-Archimedean part of Ostrowski theorem, including the trivial case.

Write $\mathcal{O}=\{x:|x|\le 1\}$. First, $\mathcal{O}$ is compact. A compact neighborhood of zero contains a sufficiently small closed ball $t^m\mathcal{O}$, since $|t{|}^m\to 0$; that ball is closed in the compact neighborhood. Scaling back proves compactness of $\mathcal{O}$. Also $K$ is complete: a Cauchy sequence has a tail in a translate of a compact neighborhood, hence a convergent subsequence, and the Cauchy condition makes the whole sequence converge. This is the completeness of locally compact nontrivially valued fields argument.

The additive subgroup $t\mathcal{O}$ is open: a sufficiently small ball around any of its points stays in $t\mathcal{O}$ by the ultrametric inequality. Thus the compact discrete quotient $\mathcal{O}/t\mathcal{O}$ is finite. Choose a set $A\subset \mathcal{O}$ of representatives, with zero representing the zero coset. For $x\in \mathcal{O}$, recursively choose $a_n\in A$ and $x_{n+1}\in \mathcal{O}$ such that $x_n=a_n+t{x}_{n+1}$, beginning with $x_0=x$. ThenFor arbitrary $x\in K$, multiply first by a large power of $t$ to put it in $\mathcal{O}$, then divide the resulting expansion by that power. This gives the digit expansion with a nonuniformizer.

For uniqueness, take the first exponent $r$ at which two expansions differ. After dividing their difference by $t^r$, all the later terms belong to the closed subgroup $t\mathcal{O}$, whereas $a_r-b_r$ does not, because the representatives are distinct. The difference cannot be zero. Conversely every series with coefficients in $A$ and only finitely many nonzero negative-index terms converges by completeness and $|t|<1$. Every element therefore has a unique expansion $\boxed{x=\sum _{n\ge r}{a}_n{t}^n}$, with the harmless initial zero coefficients understood as part of the same indexed series. No assumption that $t$ is a uniformizer is needed.

The compactness and completeness proved above apply again. The finite ring $\mathcal{O}/t\mathcal{O}$ has a positive integer characteristic $m$, so $m\in t\mathcal{O}$ and $|m|<1$. Prime factorization supplies a prime $p$ with $|p|<1$. In characteristic zero the restriction to $\mathbb{Q}$ is therefore a positive power of the p-adic absolute value, by the preceding classification. Completing $\mathbb{Q}$ embeds the P-adic number field ${\mathbb{Q}}_p$ into the complete field $K$.

Now $p\mathcal{O}$ is open, so $\mathcal{O}/p\mathcal{O}$ is finite. It is an ${\mathbb{F}}_p$-vector space; choose lifts $e_1,\dots ,e_d\in \mathcal{O}$ of a basis. Successively reducing modulo $p\mathcal{O}$ gives, for every $x\in \mathcal{O}$,Convergence in $K$ follows from $|p|<1$. Every element of $K$ becomes integral after multiplication by a power of $p$, so $K={\sum }_j{\mathbb{Q}}_p{e}_j$. Consequently $\boxed{[K:{\mathbb{Q}}_p]<\infty }$: $K$ is a p-adic field. This finite-module proof does not assume discreteness of the original value group in advance.

Take the formal Laurent series field $K=\mathbb{C}((T))$, with $v_T$ the least exponent of a nonzero term and $|x|=2^{-v_T(x)}$. Its value group is $\mathbb{Z}$, so this is a discrete valuation. A Cauchy sequence has an eventually stable coefficient at every exponent, and its tail has a common lower bound on its exponents. The stabilized coefficients therefore give a formal Laurent series field element to which the sequence converges. Hence $K$ is complete.

Its valuation ring is $\mathbb{C}[[T]]$, and its residue field is the infinite field $\mathbb{C}$. If $K$ were locally compact, the previous compact-ball argument would make $\mathbb{C}[[T]]$ compact. Its quotient by the open ideal $T\mathbb{C}[[T]]$ would then be compact and discrete, hence finite, contradicting its identification with $\mathbb{C}$. Thus $\boxed{\mathbb{C}((T))\text{ is complete and discretely valued but not locally compact}}$.

A useful version is the strong form of Hensel lemma. Let $K$ be complete for a Non-Archimedean absolute value, let $f\in {\mathcal{O}}_K[X]$, and suppose $a\in {\mathcal{O}}_K$ satisfiesThen there is a unique root $\alpha$ in the ball $|\alpha -a|<|f^{\prime }(a)|$, and $|\alpha -a|\le |f(a)/f^{\prime }(a)|$.

Put $h=|f^{\prime }(a)|$ and $\eta =|f(a)|/h^2<1$. Starting with $x_0=a$, apply Newton iteration over a valued field $x_{n+1}=x_n-f(x_n)/f^{\prime }(x_n)$. Taylor expansion over the valuation ring givesfor integral $x,u$. Inductively $|f^{\prime }(x_n)|=h$ and $|f(x_n)|/h^2\le {\eta }^{2^n}$. Indeed the Newton increment has size at most $h{\eta }^{2^n}<h$, so the derivative size is unchanged, and the new function value is bounded by the square of the increment. The increments tend to zero, and the ultrametric inequality makes the sequence Cauchy. Completeness supplies a root with the stated distance bound.

If $b,c$ are distinct points of the ball, their divided difference is $f^{\prime }(a)$ plus a term of size at most $\max (|b-a|,|c-a|)<h$, so $(f(b)-f(c))/(b-c)$ is nonzero. Two roots in that ball are impossible. This proves the stated Hensel lemma. In particular a simple root modulo the maximal ideal lifts uniquely to a root in its residue class.

For $p=2$, the even residue class contains exactly one root, by the simple-root form of Hensel lemma at $a=0$: $f(0)=20$ and $f^{\prime }(0)=-5$ is a unit. For the odd classes use the strong form of Hensel lemma:In both cases $v_2(f(a))>2v_2(f^{\prime }(a))=2$. Each gives exactly one root in $a+4{\mathbb{Z}}_2$. These two balls partition the odd integers, so there are no further roots.

For $p=3$, reduction gives $\overline{f}=X^3+X+2$, whose sole root is $2$. The derivative $3X^2-5$ is always a unit modulo $3$, so that root lifts uniquely. For $p=5$, the only possible residue is zero. But if $x\in 5{\mathbb{Z}}_5$, then $x^3-5x\in 25{\mathbb{Z}}_5$ while $20$ has valuation one, so $f(x)$ cannot vanish. The exact root counts are

The polynomial is an Eisenstein polynomial at $5$, so it is irreducible and $[K:\mathbb{Q}]=3$. Its polynomial discriminant isOne must remove the index at $2$ before using this as the field discriminant. Put $\beta =({\alpha }^2+\alpha )/2$. Direct reduction using ${\alpha }^3=5\alpha -20$ givesThus $B=\mathbb{Z}+\mathbb{Z}\alpha +\mathbb{Z}\beta$ is a ring finite over $\mathbb{Z}$, so its elements are integral. Its basis has discriminant of elements of a number field $-2575=-5^2\cdot 103$. The discriminant-index formula for an integral lattice says $\mathrm{disc}(B)=[{\mathcal{O}}_K:B{]}^2{\Delta }_K$, so only $5$ could divide the remaining index.

We use these standard different ideal facts: the norm of the different is the absolute field discriminant; its localization is the local different; an unramified prime has different exponent zero, and a tamely ramified prime of ramification index $e$ has different exponent $e-1$. The Eisenstein polynomial at $5$ gives total ramification of degree three, which is tame because $5\nmid 3$. Its contribution to $v_5({\Delta }_K)$ is therefore two. This excludes a further index factor $5$, proving ${\mathcal{O}}_K=B$ and $\boxed{{\Delta }_K=-2575}$.

Since the index $[{\mathcal{O}}_K:\mathbb{Z}[\alpha ]]=2$ is prime to $5$ and $103$, Dedekind factorization theorem applies at both primes. Modulo $5$, $f=X^3$, giving a single prime $\mathfrak{p}=(5,\alpha )$ with residue degree one and ramification index three. Modulo $103$,Thus $\mathfrak{q}=(103,\alpha -6)$ has residue degree one and ramification index two; the other prime over $103$ is unramified. Both ramified primes are tame. No other prime ramifies because none divides ${\Delta }_K$. Consequently

Choose $a\in K^{\times }$ with $0<|a{|}_1<1$. For any $x\ne 0$, the powers $x^n$ tend to zero precisely when $|x{|}_1<1$. Equality of the two topologies therefore impliesIn particular $0<|a{|}_2<1$. For positive integers $n$ and arbitrary integers $m$, apply this equivalence to $x^n{a}^{-m}$. It says thatThe inequality reverses on division by either negative logarithm. The rational cuts of the two real numbers are equal, so density of $\mathbb{Q}$ makes the numbers equal. Hence the equivalent absolute values satisfyThe formula also holds at zero. Nontriviality is essential to choosing $a$; no Archimedean or non-Archimedean restriction was used in the proof.

The idele group is the restricted productwith componentwise multiplication and the restricted product topology on the idele group relative to the unit groups at finite places. We need weak approximation for number fields only at the finitely many specified coordinates, not density in the full restricted-product topology.

Here is a proof including the Archimedean places. At each finite place in $S$, choose $a_v\in K$ sufficiently close to $x_v$, using the definition of the completion. Choose a nonzero $d\in {\mathcal{O}}_K$ clearing their denominators. Pick powers ${\mathfrak{p}}_v^{m_v}$ so that $|z/d{|}_v<\epsilon$ whenever $z\in {\mathfrak{p}}_v^{m_v}$. Put $I={\prod }_{v\in S,v\text{ finite}}{\mathfrak{p}}_v^{m_v}$. For each positive rational integer $M$ coprime to the rational primes below these places, the Chinese remainder theorem gives $b_M\in {\mathcal{O}}_K$ with $b_M\equiv Md{a}_v(\mathrm{mod}{\mathfrak{p}}_v^{m_v})$. All of $b_M+I$ satisfy the same congruences.

Under the Minkowski embedding of a number field, $I$ is a full lattice in ${\mathbb{R}}^{r_1}\times {\mathbb{C}}^{r_2}$. A fundamental parallelepiped is bounded, so every real vector is within a fixed distance of this lattice, uniformly over its translates. Choose a target vector whose coordinates at the specified Archimedean places are those of $Md{x}_v$, assigning arbitrary values at the other Archimedean places. There is $b\in b_M+I$ within that fixed distance of the target. Dividing by $Md$ makes all specified Archimedean errors tend to zero as $M\to \infty$ through the allowed integers. At finite places $M$ is a unit, so the earlier congruence bounds persist. By taking the initial approximations sufficiently close, the ultrametric inequality gives the requested errors there too. Thus some $\boxed{y=b/(Md)\in K}$ satisfies every required inequality. If there are no finite places, set $I={\mathcal{O}}_K$ and omit the congruences.

For each finite place $v\in S$, choose $d_v\in K_v^{\times }$ such that $K_v(\sqrt{d_v})/K_v$ is the unramified quadratic extension. Such a choice exists even at residue characteristic two. At odd residue characteristic use a unit with nonsquare residue. At residue characteristic two, choose $\overline{c}$ for which $X^2-X-\overline{c}$ is irreducible over the finite residue field, lift $c$ to ${\mathcal{O}}_v$, and set $d_v=1+4c$. The lifted polynomial has unit derivative at every residue point and gives the unramified degree-two extension; its root is $(1+\sqrt{d_v})/2$.

The square classes are open in $K_v^{\times }$. Indeed the strong form of Hensel lemma applied to $X^2-u$ at $1$ shows that $u$ is a square whenever $v(u-1)>2v(2)$. Thus any sufficiently close approximation $a$ to $d_v$ has $a/d_v$ square. By weak approximation for number fields, choose $a\in K$ satisfying this at every finite $v\in S$. Then the local extensions $K_v(\sqrt{a})$ are all unramified of degree two. At least one such condition ensures $a$ is not a square in $K$. If $S$ contains no finite place, impose the same condition at one auxiliary finite place.

Set $L=K(\sqrt{a})$. At every specified finite prime there is one prime of $L$ above it, with ramification index one and residue degree two. Hence $\boxed{\mathfrak{p}{\mathcal{O}}_L\text{ is prime for every finite }\mathfrak{p}\in S}$. The dyadic case is covered by the unramified construction, rather than by a nonexistent nonsquare residue unit in characteristic two.

Normalize $v_K$ to have value group $\mathbb{Z}$, and extend it to an algebraic closure. Suppose $\alpha$ is a root of a monic Eisenstein polynomial of degree $n$. A root cannot have nonpositive valuation: the leading term would then have uniquely smallest valuation. If $w=v_K(\alpha )>0$, the constant term has valuation one, all the other nonleading terms have valuation strictly greater than one, and the leading term has valuation $nw$. Cancellation forces $nw=1$. The value group of $K(\alpha )$ is $(1/e)\mathbb{Z}$, so $n\mid e$. Since $e\le [K(\alpha ):K]\le n$, equality holds throughout. Thus the polynomial is irreducible and the extension is totally ramified.

Conversely, let $L/K$ be totally ramified of degree $n$, and choose a uniformizer ${\pi }_L$. Its extended $v_K$ value is $1/n$. The same denominator argument gives $[K({\pi }_L):K]\ge n$, so $K({\pi }_L)=L$. Every conjugate has valuation $1/n$, by the unique extension of an absolute value to a finite extension of a complete field. The nonleading coefficients of its minimal polynomial are elementary symmetric expressions in these conjugates, so their valuations are positive, hence at least one because the coefficients lie in $K$. The constant coefficient is their product and has valuation exactly one. Its minimal polynomial is an Eisenstein polynomial, and $\boxed{L=K({\pi }_L)}$.

One form of Krasner's lemma is: if $K$ is a complete non-Archimedean field, $\alpha$ is separable over $K$, and $\beta$ is algebraic withwhere ${\alpha }^{\prime }$ ranges over the other $K$-conjugates, then $K(\alpha )\subseteq K(\beta )$. If $\alpha \in K$, the conclusion is immediate.

To prove the nontrivial case, let $\sigma$ be an automorphism of an algebraic closure fixing $K(\beta )$. The unique extension of an absolute value to a finite extension implies that $\sigma$ preserves the extended absolute value. ThereforeThis is strictly less than the distance to every other conjugate, so $\sigma (\alpha )=\alpha$. As $\alpha$ is separable over $K(\beta )$, being fixed by all these automorphisms puts it in $K(\beta )$. This proves $\boxed{K(\alpha )\subseteq K(\beta )}$. The strict inequality is what prevents switching to another conjugate.

We count extensions up to $K$-isomorphism. An extension of degree $n$ has a maximal unramified extension $E/K$ inside it, with $[E:K]=f\mid n$, and $L/E$ is totally ramified of degree $e=n/f$. The unramified extension of each degree is unique, so there are only finitely many possibilities for $E$.

Fix $E$ and $e$. The coefficient space of monic degree-$e$ Eisenstein polynomials is compact: its nonconstant nonleading coefficients lie in the compact maximal ideal, and its constant coefficient lies in the compact set ${\mathfrak{m}}_E\setminus {\mathfrak{m}}_E^2$. Each such polynomial $F$ has a root $\alpha$ generating a degree-$e$ extension. For sufficiently close coefficients of a polynomial $G$, the numbers $G(\alpha )$ and $G^{\prime }(\alpha )-F^{\prime }(\alpha )$ are arbitrarily small, while $F^{\prime }(\alpha )\ne 0$ because the characteristic is zero. The strong form of Hensel lemma, applied in $E(\alpha )$, supplies a root $\beta$ of $G$ as close to $\alpha$ as desired. Choose the neighborhood so that the Krasner's lemma inequality holds. Then $E(\alpha )\subseteq E(\beta )$, and both degrees are $e$, so the extensions are equal. Compactness gives a finite cover by such neighborhoods, proving finiteness of degree-$n$ extensions. If extensions are realized as subfields of a fixed algebraic closure rather than counted up to isomorphism, each has only finitely many embeddings, so that count is finite too.

For odd $p$, write $K^{\times }={\pi }^{\mathbb{Z}}{\mathcal{O}}_K^{\times }$. The square map is bijective on $1+{\mathfrak{m}}_K$ by Hensel lemma, and the finite residue field has cyclic multiplicative group of even order. ThusThe quadratic extensions from square classes correspond to its three nontrivial classes. Representatives are $u,\pi ,u\pi$, where $u$ is a unit with nonsquare residue. There are $\boxed{3}$ quadratic extensions: $K(\sqrt{u})$ is unramified, while $K(\sqrt{\pi })$ and $K(\sqrt{u\pi })$ are totally ramified. For completeness, two nonsquare parameters give the same extension only if their ratio is a square: writing $\sqrt{b}=x+y\sqrt{a}$ and squaring forces $x=0$, hence $b/a=y^2$.

Herbrand quotient and local norm indices. Let $G=⟨\sigma ⟩$ have order $n$. For an additive $G$-module $M$, put $D=\sigma -1$ and $N=1+\sigma +\cdots +{\sigma }^{n-1}$. The relevant Tate cohomology of a cyclic group groups areIf both groups are finite, their size ratio is the Herbrand quotientMultiplicative modules use products for $N$ and $D(m)=\sigma (m)/m$. A finite module has quotient one: $\#\ker N=\#M/\#NM$, while $\#DM=\#M/\#M^G$, and the two cohomology orders are equal. The six-term periodic cohomology sequence shows $h(B)=h(A)h(C)$ for a short exact sequence $0\to A\to B\to C\to 0$ when these groups are finite. Thus the quotient is unchanged by finite-index changes of lattices. For the trivial $G$-module $\mathbb{Z}$, the zeroth group is $\mathbb{Z}/n\mathbb{Z}$ and the negative first group is zero, so $h_G(\mathbb{Z})=n$.

Now let $L/K$ be a cyclic extension of p-adic fields of degree $n$. For a sufficiently deep principal unit subgroup $U_L^r=1+{\mathfrak{m}}_L^r$, the p-adic logarithm is a $G$-equivariant isomorphism with the additive p-adic lattice ${\mathfrak{m}}_L^r$; it suffices to take $r>v_L(p)/(p-1)$. The normal basis theorem makes $L$ a regular $K[G]$-module. Consequently ${\mathfrak{m}}_L^r$, as a ${\mathbb{Z}}_p[G]$-lattice, is commensurable with a direct sum of $[K:{\mathbb{Q}}_p]$ copies of ${\mathbb{Z}}_p[G]$. These regular lattices have zero Tate groups: its invariants are the multiples of the sum of the basis elements, every such element is a norm, and vectors with coefficient sum zero are images of $D$. Commensurability and the finite-module calculation give $h_G(U_L^r)=1$. Since $U_L/U_L^r$ is finite, $h_G(U_L)=1$. These comparisons also establish finiteness of the Tate groups concerned.

The valuation exact sequence $1\to U_L\to L^{\times }\to \mathbb{Z}\to 0$ now gives $h_G(L^{\times })=n$. Hilbert theorem 90 makes ${\hat{H}}^{-1}(G,L^{\times })$ trivial. One can prove the cyclic statement directly: for $a$ of norm one set $A_0=1$, $A_i=a\sigma (a)\cdots {\sigma }^{i-1}(a)$ and choose $c$ for which $b={\sum }_i{A}_i{\sigma }^i(c)\ne 0$. Such a $c$ exists by linear independence of distinct field automorphisms. Then $\sigma (b)=b/a$, so $a=b/\sigma (b)$ is a coboundary. Therefore the local cyclic norm index isThis is the central use of the Herbrand quotient: it calculates a norm index without first constructing the local reciprocity map.

Let $e,f$ be the ramification index and residue degree. With normalized integer valuations, $v_K(Nx)=f{v}_L(x)$. A norm is a unit exactly when its preimage is a unit, and the norm valuations fill $f\mathbb{Z}$. Hence there is an exact sequenceThe unit norm index is $\boxed{[{\mathcal{O}}_K^{\times }:N{\mathcal{O}}_L^{\times }]=e}$. In an unramified extension all units are norms and the obstruction is the valuation modulo $n$; in a totally ramified cyclic extension the entire index comes from units. The norm subgroup is open: on deep principal units, logarithm carries the norm to the field trace, and the trace of a full p-adic lattice contains a sufficiently deep p-adic lattice in $K$.

Hilbert norm residue symbol and the local-to-global principle. More generally, for a local field $F$ containing ${\mu }_m$, fix the local reciprocity map ${\mathrm{rec}}_F$ with uniformizers acting as arithmetic Frobenius on unramified extensions. The Hilbert norm residue symbol isThis is independent of the chosen root, is bilinear, and has value one exactly when $b$ is a norm from $F(a^{1/m})$. These are standard consequences of Local Artin reciprocity. The case $m=2$, for which the values are signs, is the one directly governing quadratic forms. The quadratic Hilbert symbol at a place $v$ is defined for $a,b\in K_v^{\times }$ bywith value one for all $b$ when $a$ is a square. Equivalently it is one precisely when $z^2=a{x}^2+b{y}^2$ has a nonzero solution over $K_v$. For nonsquare $a$, a solution has $y\ne 0$ and gives $b=(z/y{)}^2-a(x/y{)}^2$; the converse follows from the same norm identity. When $a$ is square the conic is already isotropic. Symmetry follows from this conic criterion. The local cyclic norm index gives a norm subgroup of index two; its sign character is multiplicative in $b$, and symmetry gives multiplicativity in $a$. Thus the symbol is a nondegenerate bilinear pairing on the square-class group of a field, since every nonsquare $a$ gives a nontrivial norm character. Also $(a,-a{)}_v=1$, because $N(\sqrt{a})=-a$, and $(a,a{)}_v=(a,-1{)}_v$.

For an odd-residue-characteristic p-adic field with residue size $q$, write $a={\pi }^{r}u$, $b={\pi }^{s}w$, and let $\chi$ be the quadratic character of the residue units. ThenThe unramified quadratic extension has every unit as a norm and only even norm valuations; in a ramified quadratic extension the norm of a unit has square residue. The preceding unit norm index is two, so the square-residue condition in the ramified case is also sufficient. These facts, together with $N(\sqrt{\pi })=-\pi$, determine the formula on the generators $u,\pi$ of the square-class group. At a real place the symbol is negative exactly when both arguments are negative, and at a complex place it is always one. For ${\mathbb{Q}}_2$, with $u,w$ odd units, the dyadic formula isOnly residue classes modulo eight and the parities of $r,s$ enter; other dyadic fields retain the norm definition.

The Hilbert reciprocity law states ${\prod }_v(a,b{)}_v=1$ for $a,b\in K^{\times }$. Only finitely many factors can be nontrivial: outside the places above two, the Archimedean places, and the finite places where $a$ or $b$ is not a unit, the odd-residue formula gives one. For $K=\mathbb{Q}$ this product formula is a formulation of quadratic reciprocity, including its supplementary laws. It forces local norm obstructions to occur with compatible parity. It is a necessary compatibility law, not by itself a substitute for the following local-to-global theorem.

The Hasse-Minkowski theorem states that a nondegenerate quadratic form over a number field has a nonzero isotropic vector if and only if it does over every completion. Equivalently, two nondegenerate quadratic forms are isometric globally if and only if they are isometric at every place. The forward directions are immediate; the reverse directions are the substantive global theorem. Over non-Archimedean completions, local isometry classes are determined by dimension, determinant square class and the Hasse invariant of a quadratic formAt real places one uses signature, and at complex places dimension suffices. Reciprocity gives ${\prod }_v{ϵ}_v(q)=1$ for a globally diagonalized form. These invariants make the theorem practically usable: local square classes and norm characters replace an unrestricted search for rational solutions. In particular every quadratic form of dimension at least five over a p-adic field is isotropic; for such a form over a number field the only isotropy obstructions are definite signatures at real places.

As a concrete application, take nonsquare $a\in K^{\times }$. Then $b$ is a global norm from $K(\sqrt{a})$ if and only if $(a,b{)}_v=1$ at every place. The local conditions make the ternary form $⟨1,-a,-b⟩$ isotropic everywhere. Hasse-Minkowski theorem gives a global solution of $z^2-a{x}^2=b{y}^2$; since $a$ is nonsquare, $y$ cannot be zero, and division by $y$ gives the global norm. For example $-1$ fails to be a norm from $\mathbb{Q}(i)$ already at the real place. Thus the quadratic Hilbert symbol detects local norm solvability, while the Hasse-Minkowski theorem turns solvability at all places into a global quadratic solution. The reciprocity and local-to-global theorems in this essay are stated as standard results; the norm interpretation, bilinearity and application are derived above.