estimates, which we already know to be elements ofZp[ζq−1], are in fact elements ofZp. It would be a good idea for readers to skip this section, and return to it only if they want to understand the use of the Frobenius action in the proofs of Theorems 4.18, 5.12, 6.13, and 7.14. Section 2.8 includes notations we use with multivariable polynomials. There we also include a basic fact about polynomials, which we use to prove that certain of our bounds on p-adic valuations of weights are sharp.
We consider certain algebraic extensions of Qp whose behavior is similar.
Proposition 2.2 (Unramified Extensions of the p-Adics). Qp(ζpn−1) is a degree n Galois extension of Qp, and the Galois group of Qp(ζpn−1) over Qp is the cyclic group of order n generated by the automorphism Fr which takes ζpn−1 to ζppn−1. The elements of Qp(ζpn−1) that are integral over Zp form the ring Zp[ζpn−1]. Zp[ζpn−1] is a discrete valuation ring of characteristic 0 with the unique nonzero prime ideal generated by p, and Qp(ζpn−1) is the field of fractions of Zp[ζpn−1]. Thus each nonzero element of Qp(ζpn−1) can be written uniquely as pmu with m ∈ Z and u a unit in Zp[ζpn−1], where the nonzero elements of Zp[ζpn−1] are precisely such elements with m ≥ 0, and the units in Zp[ζpn−1] are precisely such elements with m = 0. Qp(ζpn−1) and Zp[ζpn−1] are complete in the topology defined by this valuation. The set consisting of zero and the powers of ζpn−1 is a set of representatives of the equivalence classes modulo p in Zp[ζpn−1]. The quotient modulop of Zp[ζpn−1]is the finite field Fpn, whose cyclic group of units is generated by the reduction modulo pof ζpn−1. The automorphismFr onQp[ζpn−1]induces an automorphism (which we shall also call Fr) on Fpn; this induced automorphism takes each element to its pth power and it generates the Galois group of order n of Fpn over Fp. Furthermore, Qp(ζpn1−1)∩Qp(ζpn2−1) =Qp(ζpgcd(n1,n2)−1) andZp[ζpn1−1]∩Zp[ζpn2−1] =Zp[ζpgcd(n1,n2)−1].
Thus, Qp(ζpn1−1) ⊆Qp(ζpn2−1) if and only if n1 |n2, and Zp[ζpn1−1]⊆ Zp[ζpn2−1] if and only ifn1 |n2.
Proof. This follows from Proposition 16 (along with Corollary 1) in Chapter IV of [47], assuming the theory developed in that book up until that point, most particularly Propo- sitions 3 and 8 of Chapter II.
Let us examine in more detail the valuation mentioned in Propositions 2.1 and 2.2 above.
For a nonzero element a∈ Qp(ζpn−1), the unique integer m such that a = pmu for some unit u ∈ Zp[ζpn−1] is called the p-adic valuation of a. If a also lies in Qp(ζpn0
−1), then it is not hard to show that the p-adic valuation ofa in this other field is precisely the same as its p-adic valuation in the Qp(ζpn−1). Thus no reference to the field containing a is necessary, and thep-adic valuation ofais simply denotedvp(a). We definevp(0) =∞, and we consider∞strictly greater than any integer and set anything plus∞to∞. We say that
two elementsaandbarecongruent modulopmto mean thatvp(a−b)≥m. Thus the notion of equivalence of elements modulo powers ofpis independent of which unramified extension of Qp we regard as the ambient field. With these conventions, we have the following easily verified properties of thep-adic valuation:
Lemma 2.3 (Properties of vp). For any a, b∈Qp(ζpn−1), we have the following:
(i) vp(a) =∞ if and only if a= 0.
(ii) vp(ab) =vp(a) +vp(b).
(iii) vp(a+b)≥min{vp(a), vp(b)}, with equality when vp(a)6=vp(b).
For a ∈ Qp(ζpn−1), we also define the p-adic absolute value of a, denoted |a|p, to be p−vp(a), where p−∞ is considered to be 0. The properties of vp translate into properties of
|·|p as follows:
Lemma 2.4 (Properties of |·|p). For any a, b∈Qp(ζpn−1), we have the following:
(i) |a|p = 0 if and only if a= 0.
(ii) |ab|p =|a|p|b|p.
(iii) |a+b|p ≤max{|a|p,|b|p}, with equality when |a|p 6=|b|p.
Thus the p-adic absolute value provides a metric on Qp(ζpn−1), where the distance between a and b is |a−b|p. This metric defines a topology on Qp(ζpn−1) that we call the p-adic topology. It is this topology that is discussed in Propositions 2.1 and 2.2 above.
In Proposition 2.2, we saw that the finite fields of characteristic p can be obtained as quotients modulo p of rings of algebraic integers in unramified extensions of Qp. We shall also be interested in the quotients of these rings modulo powers of p. The Galois ring of characteristic pm and order pmn, denoted GR(pm, n) is the quotient modulo pm of Zp[ζpn−1]. Note that GR(p, n) is the finite fieldFpn of orderpn. Also note that GR(pm,1) is the integer residue ringZ/pmZ. The ring GR(pm, n) contains Z/pmZas a subring and can be thought of as an extension ofZ/pmZobtained by adjoining a root of unity of orderpn−1.
Furthermore, the statements regarding intersections and containments of extensions of Qp
andZp in Proposition 2.2 imply that GR(pm, n1)∩GR(pm, n2) = GR(pm,gcd(n1, n2)), and therefore GR(pm, n1)⊆GR(pm, n2) if and only ifn1|n2. In this case, GR(pm, n2) is a free GR(pm, n1)-module. Note that GR(pm, n) is a principal ideal ring withm+1 ideals, namely pjGR(pm, n2) forj = 0,1, . . . , m. Herep0GR(pm, n2) is the entire ring andpmGR(pm, n2) is the zero ideal. For more information on Galois rings, the reader should consult the book of McDonald [34].
It does not make sense to consider the ring GR(pm1, n) as a subring of GR(pm2, n) when m1 < m2, since the two rings have different characteristics. However, we do have a way of relating elements of the one ring to elements of the other. Since m1 < m2, reduction modulopm1 furnishes an epimorphism from GR(pm2, n) to GR(pm1, n). Thus form1 ≤m2, if a ∈ GR(pm2, n), we define πm1(a) ∈ GR(pm1, n) to be the reduction modulo pm1 of a.
Also, if a∈ Zp[ζpn−1], we define πm1(a) ∈GR(pm1, n) to be the reduction modulo pm1 of a. Since ζpn−1 is a root of unity of order pn−1 over Qp and since π1(ζpn−1) is a root of unity of orderpn−1 in Fpn (see Proposition 2.2 above), we know that πm(ζpn−1) is a root of unity of order pn−1 in GR(pm, n) for every positive integer m. As a convention, we defineπ∞ to be the identity map onZp[ζpn−1].
Now we wish to define a map from GR(pm1, n) to GR(pm2, n) whenm1 < m2. Proposi- tion 2.2 tells us that each element a ∈ Zp[ζpn−1] can be written uniquely as P∞
i=0aipi, where each ai is either zero or a power of ζpn−1. We call this the canonical expan- sion of a ∈ Zp[ζpn−1]. This implies that when m is a positive integer, each element a of GR(pm, n) can be written uniquely as Pm−1
i=0 aipi, where each ai is either zero or a power of πm(ζpn−1). We likewise call this the canonical expansion of a ∈ GR(pm, n).
For m1 ≤ m2 and a ∈ GR(pm1, n), with canonical expansion a = Pm1−1
i=0 aipi, we de- fine τm2(a) to be an element b ∈ GR(pm2, n) with canonical expansion b = Pm1−1
i=0 bipi, where bi = 0 whenever ai = 0 and bi = πm2(ζpn−1)j whenever ai = πm1(ζpn−1)j. We also define τ∞(a) = Pm1−1
i=0 cipi, where ci = 0 whenever ai = 0 and ci = ζpjn−1 when- ever ai = πm1(ζpn−1)j. For each positive integer m, we call πm the Teichm¨uller lift to characteristic pm and call τ∞ theTeichm¨uller lift to characteristic 0.
Form1 ≤m2inZ+∪ {∞}, we haveπm1◦πm2 =πm1 andτm2◦τm1 =τm2. Furthermore,
ifa∈GR(pm1, n), πm1(τm2(a)) =a, but if a∈GR(pm2, n), then it is not always true that τm2(πm1(a)) =a.
To summarize the relationships between the various unramified extensions ofQp and the various Galois rings, we have the following commutative diagrams, where unmarked arrows are inclusion maps:
Qp(ζpn−1) ←−−−− Zp[ζpn−1] −−−−→πm GR(pm, n) −−−−→π1 Fpn
x
x
x
x
Qp ←−−−− Zp
πm
−−−−→ Z/pmZ −−−−→π1 Fp
and
Qp(ζpn−1) ←−−−− Zp[ζpn−1] ←−−−−τ∞ GR(pm, n) ←−−−−τm Fpn
x
x
x
x
Qp ←−−−− Zp
τ∞
←−−−− Z/pmZ ←−−−−τm Fp.
In each diagram, the two rows coincide when n = 1, and the last two columns coincide when m= 1.
We transplant the notion ofp-adic valuation from the unramified extensions ofQpto the Galois rings. The p-adic valuation of a nonzero elementa∈GR(pm, n), denoted vp(a), is defined to be the greatestksuch thata∈pkGR(pm, n). We definevp(0) =∞in GR(pm, n).
Thus we have defined vp: GR(pm, n) → {0,1, . . . , m−1,∞}. Note that for any m1 ∈ Z, m2 ∈Z+∪ {∞} with m1 ≤m2, anda∈GR(pm1, a), we have vp(τm2(a)) =vp(a). On the other hand, ifm1≤m2 are positive integers anda∈GR(pm2, n), thenvp(πm1(a)) =vp(a) if vp(a) < m1 or vp(a) = ∞, but vp(πm1(a)) = ∞ when m1 ≤ vp(a) < ∞. Likewise, if a ∈ Zp[ζpn−1] and m1 is a positive integer, then vp(πm1(a)) = vp(a) if vp(a) < m1 or vp(a) =∞, butvp(πm1(a)) =∞when m1≤vp(a)<∞.
In Proposition 2.2, we used Fr to denote the field automorphism of Qp(ζpn−1) that fixes Qp pointwise and takes ζpn−1 to ζppn−1. Note that Fr restricted to Zp[ζpn−1] is an automorphism of rings. We also used Fr to denote the field automorphism it induces on Fpn through reduction modulo p; this automorphism takes every element to its pth power.
Because the automorphism Fr of the field Qp(ζpn−1) takes pm topm, it permutes the ideal pmZp[ζpn−1] of the ringZp[ζpn−1], and thus induces an automorphism of the ring GR(pm, n)
for each positive m through reduction modulo pm. This automorphism fixes pointwise the subring GR(pm,1) =Z/pmZof GR(pm, n) and maps the elementπm(ζpn−1) toπm(ζpn−1)p. Throughout the thesis we call all of these automorphisms theFrobenius automorphismand denote them all by Fr.
It is not difficult to show that Fr commutes withπmandτm for allm∈Z+∪ {∞}. That is, we have the commutative diagrams
Qp(ζpn−1) ←−−−− Zp[ζpn−1] −−−−→πm GR(pm, n) −−−−→π1 Fpn Fr
y Fr
y Fr
y Fr
y Qp(ζpn−1) ←−−−− Zp[ζpn−1] −−−−→πm GR(pm, n) −−−−→π1 Fpn
and
Qp(ζpn−1) ←−−−− Zp[ζpn−1] ←−−−−τ∞ GR(pm, n) ←−−−−τm Fpn Fr
y Fr
y Fr
y Fr
y Qp(ζpn−1) ←−−−− Zp[ζpn−1] ←−−−−τ∞ GR(pm, n) ←−−−−τm Fpn, where unlabeled arrows are inclusion maps.
Recall thatQp(ζpn1−1)⊆Qp(ζpn2−1) if and only ifn1 |n2, and recall that GR(pn1, m)⊆ GR(pn2, m) if and only if n1 | n2. Suppose n1 | n2. Since Fr generates the Galois group Gal(Qp(ζpn2−1)/Qp) of order n2 and the Galois group Gal(Qp(ζpn1−1)/Qp) of order n1, we see that Frn1 generates the Galois group Gal(Qp(ζpn2−1)/Qp(ζpn1−1)) of ordern2/n1. Thus an element of Qp(ζpn2−1) is in Qp(ζpn1−1) if and only if it is fixed by Frn1. Likewise, for any positive integer m, an element of GR(pm, n2) is in GR(pm, n1) if and only if it is fixed by Frn1. This can be checked by writing the canonical expansion of an arbitrary element of GR(pm, n2) and applying Frn1. If any coefficient is neither zero nor a power ofπm(ζpn1−1), the element will not be fixed by Frn1; otherwise the element will be fixed.
If n1 | n2, we define the trace map Trnn21: Qp(ζpn2−1) → Qp(ζpn1−1) by Trnn21(a) = P(n2/n1)−1
j=0 Frn1j(a). Since Fr commutes with πm for all positive integers m, it induces a trace map on the Galois rings, which we express with the same notation, i.e., we write Trnn21: GR(pm, n2) → GR(pm, n1). That this trace map is a surjective GR(pm, n1)-linear map follows from the fact that Trnn21:Qp(ζpn2−1) → Qp(ζpn1−1) is a surjective Qp(ζpn1−1)- linear map. Of course Trnn21: GR(p, n2) → GR(p, n1) is just the usual trace from Fpn2 to
Fpn1. Since Fr commutes with the maps πm and τm for all m ∈Z+∪ {∞}, we also know that Trnn21 commutes with πm and τm for all n1, n2 ∈Z+ and m ∈Z+∪ {∞}. In terms of commutative diagrams, we have
Qp(ζpn2−1) ←−−−− Zp[ζpn2−1] −−−−→πm GR(pm, n2) −−−−→π1 Fpn2 Trnn21
y Tr
n2 n1
y Tr
n2 n1
y Tr
n2 n1
y Qp(ζpn1−1) ←−−−− Zp[ζpn1−1] −−−−→πm GR(pm, n1) −−−−→π1 Fpn1
and
Qp(ζpn2−1) ←−−−− Zp[ζpn2−1] ←−−−−τ∞ GR(pm, n2) ←−−−−τm Fpn2 Trnn21
y Tr
n2 n1
y Tr
n2 n1
y Tr
n2 n1
y Qp(ζpn1−1) ←−−−− Zp[ζpn1−1] ←−−−−τ∞ GR(pm, n1) ←−−−−τm Fpn1, where unlabeled arrows are inclusion maps.