TANT 12 - Extending absolute values

Hello there! These are notes for the twelfth class of the course "Topics in algebra and number theory" held in Block 4 of the academic year 2017/18 at the University of Copenhagen.

In the previous lecture we have seen two useful applications of completeness: Hensel's lemma, which allows us to solve equations in complete fields by solving them in their residue fields, and a characterization of norms on finite dimensional vector spaces over complete fields.

The aim of this lecture is to use the second result as a starting point to study the theory of the extensions of absolute values to an algebraic extension. We have already seen in the second lecture that if we have a field extension KL and an absolute value ψ:LR0 we can restrict it to an absolute value ϕ:KR0.

Suppose now that we have an absolute value ϕ:KR0. Can we extend it to an absolute value ψ:LR0? And if we have two absolute values ψ1,ψ2:LR0 which restrict to the same absolute value on K, how are they related?


Extending complete absolute values

We will start to answer to this question from the complete case. Indeed, we have already seen an example of extension of an absolute value from a complete field in the fourth lecture, where we proved in Lemma 10 that if (K,ϕ) is a complete valued field and L=K(i) is a quadratic extension of K with i2+1=0 then L is complete with respect to the absolute value ψ:LR0 defined by ψ(x+iy)=ϕ(x)2+ϕ(y)2. To be honest, we didn't prove completely that ψ was an absolute value, because we didn't prove that it satisfied the triangle inequality.

It is now time to remedy to this lack and to generalize this to arbitrary finite extensions of complete fields. To do so we recall that if KL is a finite extension of fields we can define the field norm NL/K:LKαdetK(μα) where μα:LL is the K-linear map defined by μα(x):=αx. Observe that the multiplicativity of the determinant immediately implies that the norm is multiplicative, i.e. NL/K(αβ)=NL/K(α)NL/K(β) for all α,βL.

 Theorem 1  Let (K,ϕ) be a complete valued field and let KL be a finite extension of degree d=[L:K]. Then the map ψ:LR0 defined by ψ(x):=dϕ(NL/K(x)) is the unique extension of the absolute value ϕ to L. Moreover, L is complete with respect to ϕ.

 Proof  If d=1 there is nothing to prove. Moreover if d1 and K is Archimedean then Theorem 16 of the fourth lecture tells us that d=2, KR and LC and in this case this theorem is equivalent to Lemma 10 of the fourth lecture.

Suppose hence that (K,ϕ) is non-Archimedean and d2. It is trivial to observe that ψ satisfies the first two axioms of Definition 1 of the second lecture. We want to check now that ψ(x+y)max(ψ(x),ψ(y)) and we know from Exercise 17 of the second lecture that this is equivalent to check that ψ(1+t)1 for all tL such that ψ(t)1. Observe that for every xL we have that NL/K(x)=fx(0)mx where fx(T)K[T] is the minimal polynomial of x and mxN. Observe moreover that f1+x(T)=fx(T1) which implies that ψ(1+t)1 if and only if ϕ(ft(1))1. This would automatically follow if ft(T)Aϕ[T].

Thus let tL be any element with ψ(t)1. We have that ft(0)Aϕ because ψ(t)1 is equivalent to ϕ(ft(0))1 and we want to prove that ftAϕ. This follows from Hensel's lemma (see Theorem 5 of the previous lecture). Indeed if ft(T)Aϕ[T] then we can "clear denominators", i.e. we can find αK× with ϕ(α)<1 such that αft(T)Aϕ[T]. Observe that since f is monic and ϕ(ft(0))1 if we write αft(T)=nj=0cjTj we have that c0,cnmϕ. This is equivalent to say that 0<max{k:Tk¯αftκϕ[T]}<deg(ft) and thus by Hensel's lemma there exists a polynomial g(T)Aϕ[T] such that deg(g)<deg(ft) and gft. But this is absurd because the minimal polynomial ft is irreducible.

To conclude we only have to observe that ψ is a norm on L considered as a vector space over K, in the sense of Definition 8 of the previous lecture. Thus we only have to apply Theorem 14 of the previous lecture to see that L is complete with respect to ψ and that ψ is unique up to equivalence. Since ψ extends ϕ we also see that ψ is unique. Q.E.D.

The proof of the previous theorem tells us that the ring Aψ associated to a non-Archimedean, complete absolute value ψ:LR0 which extends ϕ:KR0 is strongly related to Aϕ. More precisely for every αL we have that αAψ if and only if fα(x)Aϕ[x]. This property is so important that it deserves a name.

 Definition 2  Let AB be an extension of rings. Then the integral closure of A in B is the ring ¯AB of all elements bB such that f(b)=0 for some monic polynomial f(x)=xn+an1xn1++a0A[x].

 Corollary 3  Let (K,ϕ) be a non-Archimedean, complete valued field, let KL be a finite extension and let ψ:LR0 be an absolute value extending ϕ. Then the ring AψL is the integral closure of Aϕ inside L.

Extending absolute values

We deal now with the general problem of extending an absolute value ϕ:KR0 to a finite extension KL.
To proceed we will need to assume that this extension is separable, i.e. that for every αL we have that fα(x)0, where fα(x)K[x] is the minimal polynomial of α.
This assumption is not restrictive at all. Indeed if K has characteristic zero every finite extension KL is separable because K is perfect.
If K has characteristic p>0 then for every finite field extension KL there exists a unique intermediate extension KEL such that KE is separable and EL is purely inseparable, i.e. for every αL there exists nN such that αpnK (for a proof of this statement see Lemma 9.14.6 of the Stacks Project). It is clear now from the definition of a purely inseparable extension EL that every absolute value ψ:ER0 extends uniquely to an absolute value ξ:LR0 simply by setting ξ(α):=(ψ(αpn))pn where nN is such that αpnE.

We can thus assume without further hesitation that our finite extension KL is separable. Recall that in this case the primitive element theorem tells us that L=K(α) for some αL. Thus we can reduce to the case of simple extensions in the following theorem.

 Theorem 4  Let (K,ϕ) be a valued field and let L=K(α) be a finite simple extension of fields, and assume that the minimal polynomial fα(x)K[x] is separable. Then we have a bijection {ψ:LR0ψϕ}{g1(x),,gr(x)} where ψϕ means that ψ extends ϕ and {g1,,gr}Kϕ[x] are the irreducible factors of the polynomial fα(x) over the completion Kϕ of K with respect to ϕ. In particular ϕ can be extended to L in at least one way and at most [L:K]-ways and we have an isomorphism KϕKLψϕLψ as Kϕ-algebras.

 Proof  Observe first of all that since fα(x) is separable it has only simple factors over Kϕ, i.e. fα(x)=g1(x)gr(x)Kϕ[x] for some distinct, monic, irreducible polynomials g1,,grKϕ[x]. Recall now that LK[x]/(fα(x)) (see this answer on Math StackExchange) and thus we have an isomorphism KϕKLKϕKK[x](fα(x))Kϕ[x](fα(x))rj=1Kϕ[x](gj(x))() as one sees applying Corollary 2.24 of these notes by Keith Conrad and Chinese remainder's theorem (see Lemma 10.14.4 of the Stacks Project).

Recall now that Kϕ[x]/(gj(x)) is a field for every j{1,,r} because gj(x)Kϕ[x] is irreducible (see Proposition 5.21 of these notes by Patrick Morandi). Moreover, this is a finite extension of the complete field (Kϕ,Φ) and thus, by Theorem 1 there exists a unique absolute value Ψ:Kϕ[x]/(gj(x))R0 which extends Φ. Now, if we compose the canonical map LKϕKL defined as x1x with the isomorphism () and with the projection ri=1Kϕ[x]/(gi(x)) we see that K_{\phi}[x]/(g_j(x)) is also a (very possibly infinite) extension of L . Thus we can restrict the absolute value \Psi to an absolute value \psi \colon L \to \mathbb{R}_{\geq 0} and it is easy to see that \psi extends \phi because the diagram on the right is commutative. Thus we have constructed a map F \colon \{ g_1(x), \dots, g_r(x) \} \to \{ \psi \, \colon L \to \mathbb{R}_{\geq 0} \, \mid \, \psi \mid \phi \} which associates to every irreducible factor of f_{\alpha}(x) \in K_{\phi}[x] an extension of the absolute value \phi . In particular we have seen that every absolute value can be extended in at least one way.

Suppose now that \psi \colon L \to \mathbb{R}_{\geq 0} is an absolute value extending \phi . Then we know already from the fourth lecture that the completion L_{\psi} is an extension of L (by definition) and of K_{\phi} (by the universal property). Thus we can use the fact that the tensor product of algebras over a ring is a pushout (see page 27 of the same notes by Keith Conrad) to get a map K_{\phi} \otimes_K L \to L_{\psi} . Observe that the image of this map is dense in L_{\psi} because it contains L and it is closed in L_{\psi} because it is a K_{\phi} -subspace of L_{\psi} and L_{\psi} is finite dimensional by Exercise 5, and thus we can apply Exercise 15 of the previous lecture. Thus the map K_{\phi} \otimes_K L \to L_{\psi} is surjective. We can use now Exercise 6 and the isomorphism K_{\phi} \otimes_{K} L \cong \prod_{i = 1}^r K_{\phi}[x]/(g_i(x)) to see that the surjective map K_{\phi} \otimes_K L \to L_{\psi} factors through a projection \prod_{i = 1}^r K_{\phi}[x]/(g_i(x)) \twoheadrightarrow K_{\phi}[x]/(g_j(x)) for some j \in \{ 1,\dots,r\} . In particular the induced map K_{\phi}[x]/(g_j(x)) \to L_{\psi} is an isomorphism of fields. This construction gives us another map G \colon \{ \psi \, \colon L \to \mathbb{R}_{\geq 0} \, \mid \, \psi \mid \phi \} \to \{ g_1(x), \dots, g_r(x) \} and it is not difficult to see that this is the inverse to the map defined in the previous paragraph.

Indeed to prove that G \circ F is the identity we only need to observe that K_{\phi}[x]/(g_i(x)) \cong K_{\phi}[x]/(g_j(x)) if and only if i = j . To prove that F \circ G is the identity let \psi be an absolute value extending \phi and let \psi' := F(G(\psi)) . Then we have that L_{\psi'} \cong K_{\phi}[x]/(g_j(x)) where g_{j}(x) = G(\psi) . But then L_{\psi'} \cong L_{\psi} and this implies that \psi = \psi' . Q.E.D.

 Exercise 5  Let (K,\phi) \hookrightarrow (L,\psi) be a finite extension of valued fields. Prove that (K_{\phi},\Phi) \hookrightarrow (L_{\psi},\Psi) is also a finite extension of degree less or equal to [L \colon K] . (Hint: if \alpha \in L_{\psi} then it is a limit of a sequence of elements \{ \alpha_k \} \subseteq L and each one of them satisfies an equation of degree (at most) [L \colon K] . Prove that \alpha satisfies the equation obtained as a limit of these equations and conclude).

 Exercise 6  Let R \subseteq \prod_{i \in I} A_i be a subring of a product or rings. Prove that R is an integral domain if and only if R \subseteq A_i for some \( i \in I \) and A_i is itself an integral domain. (This is a very easy exercise).

 Example 7  Let K \subseteq L be an extension of number fields, and let \phi \colon K \to \mathbb{R}_{\geq 0} be a non-Archimedean absolute value. We know already that \phi is equivalent to \lvert \cdot \rvert_{\mathfrak{p}} for some prime ideal \mathfrak{p} \subseteq \mathcal{O}_K (see Theorem 7 of the third lecture). Let now \psi \colon L \to \mathbb{R}_{\geq 0} be an absolute value extending \phi , and suppose that L is equivalent to \lvert \cdot \rvert_{\mathfrak{P}} for some \mathfrak{P} \subseteq \mathcal{O}_L . Then for every \alpha \in \mathcal{O}_K we have that \alpha \in \mathfrak{p} if and only if \lvert \alpha \rvert_{\mathfrak{p}} < 1 , and this is true if and only if \lvert \alpha \rvert_{\mathfrak{P}} < 1 , which holds if and only if \alpha \in \mathfrak{P} \cap \mathcal{O}_K . Thus we have proved that \mathfrak{p} = \mathfrak{P} \cap \mathcal{O}_K , i.e. that \mathfrak{p} is the contraction of \mathfrak{P} to \mathcal{O}_K . We see in this way that extensions of absolute values correspond (in the case of number fields) to extensions of prime ideals.

Ramification and inertia

We have just seen that every absolute value \phi \colon K \to \mathbb{R}_{\geq 0} on a field K admits a finite number of extensions to a finite extension L \supseteq K . We would like now to understand the extension of valued fields (K,\phi) \hookrightarrow (L,\psi) , and to do so we start by defining two important invariants of this extension.

 Definition 8  Let (K,\phi) \hookrightarrow (L,\psi) be an extension of valued fields. We define the inertia index e(L \mid K) = e(\psi \mid \phi) as the index of the subgroup \phi(K^{\times}) \leq \psi(L^{\times}) and the ramification index f(L \mid K) = f(\psi \mid \phi) as the degree of the extension of residue fields \kappa_{\phi} \subseteq \kappa_{\psi} (see the diagram on the right).

 Definition 9  We say that an extension of valued fields (K,\phi) \hookrightarrow (L,\psi) is:
  • unramified if  e(L \mid K) = 1 and the extension \kappa_{\phi} \subseteq \kappa_{\psi} is separable;
  • totally ramified if f(L \mid K) = 1 ;
  • tamely ramified if e(L \mid K) is relatively prime to the characteristic of \kappa_{\phi} and the extension \kappa_{\phi} \subseteq \kappa_{\psi} is separable.
It seems from Definition 9 that the two numbers e(L \mid K) and f(L \mid K) are somehow "complementary", i.e. when one is "big" the other one needs to be small and vice versa. This is precisely what happens, as the following theorem shows.

 Theorem 10  Let (K,\phi) be a non-Archimedean, valued field, and let K \subseteq L be a finite, separable extension of fields. Then we have the inequality \sum_{\psi \mid \phi} e(\psi \mid \phi) f(\psi \mid \phi) \leq [L \colon K] and the equality holds if \phi is a discrete absolute value.

 Proof  Observe first of all that the dimension of K_{\phi} \otimes_K L as a K_{\phi} -vector space is equal to the dimension of L as a K -vector space, i.e. to [L \colon K] . Recall from Theorem 4 that we have an isomorphism K_{\phi} \otimes_K L \cong \prod_{\psi \mid \phi} L_{\psi} and observe that the dimension of the vector space on the right is equal to \sum_{\psi \mid \phi} [L_{\psi} \colon K_{\phi}] . Thus we have that [L \colon K] = \sum_{\psi \mid \phi} [L_{\psi} \colon K_{\phi}] and we would have proved the theorem if we knew that [L_{\psi} \colon K_{\phi}] \geq e(\psi \mid \phi) f(\psi \mid \phi) for every \psi \mid \phi . Thanks to Exercise 2 of the ninth lecture we know that e(\psi \mid \phi) = e(\Psi \mid \Phi) and f(\psi \mid \phi) = f(\Psi \mid \Phi) , where \Phi and \Psi are the extensions of \phi and \psi to the completions K_{\phi} and L_{\psi} . To prove the theorem it is hence sufficient to prove it for complete fields.

Thus from here until the end of the proof we will assume that (K,\phi) is complete and that \psi \colon L \to \mathbb{R}_{\geq 0} is the unique extension of \phi to L . In this case we simply have to prove that e(\psi \mid \phi) f(\psi \mid \phi) \leq [L \colon K] , because the extension is unique. Let R \subseteq A_{\psi}^{\times} be a set of elements whose reduction modulo \mathfrak{m}_{\psi} becomes a \kappa_{\phi} -basis for \kappa_{\psi} and let S \subseteq L^{\times} be a set of elements whose image under \psi is a set of coset representatives for the subgroup \phi(K^{\times}) \subseteq \psi(L^{\times}) .

Suppose now that \sum_{r \in R, \, s \in s} a_{r,s} r s = 0 for some a_{r,s} \in K which are all zero except from a finite number of them. Then if \alpha_s := \sum_{r \in R} a_{r,s} r \in L we have that \psi(\alpha_s) \leq \max_r \phi(a_{r,s}) by the ultrametric inequality and the fact that R \subseteq A_{\psi} . Thus we have that a_{r_s,s}^{-1} \alpha_s \in A_{\psi}^{\times} for some r_s \in R (the one such that \phi(a_{r_s,s}) \geq \phi(a_{r,s}) for all r \in R ). Thus if 0 \neq \psi(\alpha_{s_1} s_1) = \psi(\alpha_{s_2} s_2) for some s_1, s_2 \in S we have that s_1 = s_2 . Indeed by the definition of S this is true if and only if \psi(s_1 s_2^{-1}) \in \phi(K^{\times}) and we have that \psi(s_1 s_2^{-1}) = \psi(\alpha_{s_2} \alpha_{s_1}^{-1}) = \psi(a_{r_{s_2},s_2}^{-1} \, \alpha_{s_2}) \psi(a_{r_{s_1},s_1} \alpha_{s_1}^{-1}) \phi(a_{r_{s_2},s_2} a_{r_{s_1},s_1}^{-1}) = \phi(a_{r_{s_2},s_2} a_{r_{s_1},s_1}^{-1}) \in \phi(K^{\times}) which implies that s_1 = s_2 and thus that all the elements \alpha_s s have either zero or different absolute values. This finally implies that a_{r,s} = 0 for all r \in R and s \in S because 0 = \psi\left( \sum_{r \in R, \, s \in s} a_{r,s} r s \right) = \psi\left( \sum_{s \in S} \alpha_s s \right) = \max_{s \in S} \psi(\alpha_s s)
which implies that \alpha_s = 0 for all s \in S and thus that a_{r,s} = 0 for all r \in R and s \in S .
Thus we have shown that the set \{ r s \mid r \in R, \, s \in S \} \subseteq L is a set of elements which are linearly independent over K . This proves that R and S are finite because K \subseteq L is finite and that \# R \cdot \# S = f(\psi \mid \phi) e(\psi \mid \phi) \leq [L \colon K] .

Suppose finally that \phi(K^{\times}) \leq \mathbb{R}_{> 0} is a discrete subgroup. Then \psi(L^{\times}) is also a discrete subgroup (because \phi(K^{\times}) is a subgroup of finite index inside \psi(L^{\times}) ) and we know (see Lemma 2 of the third lecture) that A_{\phi} and A_{\psi} are discrete valuation rings. Let \pi \in A_{\phi} and \Pi \in A_{\psi} be uniformizers, and let T_{\phi} \subseteq A_{\phi} be a set of representatives for the quotitent A_{\phi}/\mathfrak{m}_{\phi} which contains zero. Then the set T_{\psi} := \sum_{r \in R} T \cdot r = \left\{ \sum_{r \in R} t_r \, r \mid t_r \in T_{\phi} \right\} is a set of representatives for the quotient A_{\psi}/\mathfrak{m}_{\psi} which contains zero, because by definition R \subseteq A_{\psi}^{\times} was a set of elements whose reduction modulo \mathfrak{m}_{\psi} gave us a basis for the extension of fields  A_{\phi}/\mathfrak{m}_{\phi} \subseteq A_{\psi}/\mathfrak{m}_{\psi} .
To conclude we only need to observe that \Pi^{e(\psi \mid \phi)} = u \pi for some u \in A_{\phi}^{\times} because \phi(\pi) and \psi(\Pi) generate \phi(K^{\times}) and \psi(L^{\times}) respectively. Thus we have that \mathfrak{m}_{\psi}^n is generated by \Pi^n = \pi^i \Pi^j with 0 \leq j < e(\psi \colon \phi) and we can apply Proposition 8 of the ninth lecture to see that every \alpha \in A_{\psi} can be written as \alpha = \sum_{i,j} t_{i,j} \pi^i \Pi^j for some t_{i,j} \in T_{\psi} . Thus using the definition of T_{\psi} we know that every element \alpha \in A_{\psi} can be written as \alpha = \sum_{r \in R} \sum_{j = 1}^{e(\psi \mid \phi)} \left( \sum_{k = 0}^{+\infty} t_{i,j,k} \pi^k \right) r \Pi^j for some t_{i,j,k} \in T_{\phi} . This proves that A_{\psi} = \bigoplus_{r \in R} \bigoplus_{j = 1}^{e(\psi \mid \phi)} A_{\phi} \, r \, \Pi^j and thus that the set \{ r \, \Pi^j \mid r \in R, \, j \in \{ 1, \dots, e(\psi \mid \phi) \} \, \} is a K-basis for L , which implies that f(\psi \mid \phi) e(\psi \mid \phi) = [L \colon K] as we wanted to prove. Q.E.D.

 Exercise 11  Let K \subseteq L be an extension of number fields, let \mathfrak{P} \subseteq \mathcal{O}_L be a prime ideal and let \mathfrak{p} := \mathfrak{P} \cap \mathcal{O}_K . Prove that e(\lvert \cdot \rvert_{\mathfrak{P}}, \lvert \cdot \rvert_{\mathfrak{p}}) is the maximum of all n \in \mathbb{N} such that \mathfrak{P}^n \mid \mathfrak{p} \mathcal{O}_L . (Hint: recall what we said in the proof of the previous theorem about the uniformizers...).

Conclusions and references 

In this lecture we managed to:
  • prove that every absolute value which makes a field K complete admits a unique extension to every finite extension of K ;
  • prove that every absolute value on a field K has a finite number of extensions to every finite extension of K ;
  • define the ramification and inertia index of an extension of valued fields, and prove a fundamental inequality regarding them.
References for this lecture include:

Comments

Popular posts from this blog

A (not quite) perfect(oid) talk

TANT 1 - An introduction to p-adic numbers (and much more)

TANT 2 - Absolute values