1: %\documentclass[letterpaper]{birkmult}
2: \documentclass[letterpaper]{amsart}
3: \usepackage{amssymb}
4: \usepackage{amsmath}
5:
6: \newtheorem{theorem}{Theorem}[section]
7: \newtheorem{lemma}[theorem]{Lemma}
8: \newtheorem{proposition}[theorem]{Proposition}
9: \newtheorem{corollary}[theorem]{Corollary}
10: \newtheorem*{theorem*}{Theorem}
11:
12: \theoremstyle{definition}
13: \newtheorem*{definition}{Definition}
14: \newtheorem{remark}{Remark}
15: \newtheorem*{example}{Example}
16: \newtheorem{conjecture}{Conjecture}
17: \newtheorem{guess}{Guess}
18: \newtheorem*{goal}{Goal}
19: \newtheorem*{observation}{Observation}
20: \newtheorem{assumption}{Assumption}
21: \newtheorem{hypothesis}{Hypothesis}
22: \newtheorem*{assertion}{Assertion}
23: \newtheorem{question}{Question}
24:
25: \newcommand{\euO}{\mathfrak O}
26: \newcommand{\euP}{\mathfrak P}
27:
28:
29: \newcommand{\bW}{\mathbf W}
30: \newcommand{\bQ}{\mathbb Q}
31: \newcommand{\bZ}{\mathbb Z}
32: \newcommand{\bF}{\mathbb F}
33:
34: \newcommand{\bU}{\mathbf U}
35:
36: %\renewcommand{\baselinestretch}{1.5}
37:
38: \begin{document}
39: \title[Galois scaffolding]{One-dimensional elementary abelian extensions have Galois scaffolding}
40:
41: \author{G. Griffith Elder}
42: \email{elder@vt.edu}
43: \address{Department of
44: Mathematics \\ University of Nebraska at Omaha\\ Omaha, NE 68182-0243
45: U.S.A.} \curraddr{Department of Mathematics \\ Virginia Tech\\
46: Blacksburg, VA 24061-0123 U.S.A.}
47: \date{May 2, 2007}
48: \subjclass{11S15} \keywords{Ramification, Galois module structure}
49: \thanks{The author was partially supported by National Science Foundation Grant No. 201080.}
50:
51: \bibliographystyle{amsalpha}
52:
53: \begin{abstract}
54: We define a variant of normal basis, called a {\em Galois
55: scaffolding}, that allows for an easy determination of valuation, and
56: has implications for Galois module structure.
57: We identify fully ramified, elementary abelian extensions of local
58: function fields of characteristic $p$, called {\em one-dimensional},
59: that, in a particular sense, are as simple as cyclic degree $p$
60: extensions, and prove the statement in the title above.
61: \end{abstract}
62:
63:
64:
65: \maketitle
66:
67:
68: \section{Introduction}
69:
70:
71: The Normal Basis Theorem states that in a finite, Galois extension
72: $L/K$ with $G=\mbox{Gal}(L/K)$,
73: there are elements $\rho\in L$ whose conjugates
74: $\{\sigma\rho:\sigma\in G\}$ provide a field basis for
75: $L$ over $K$. In the setting of local field extensions, the most
76: important property of an element is its valuation and so we asked in
77: \cite{elder:blms} about the valuation of these elements: Are there are
78: valuations (integer certificates) that guarantee that any element
79: bearing the specified valuation be a normal basis generator? ({\em
80: i.e.} $v\in \mathbb{Z}$ so that $\rho\in L$ and $v_L(\rho)=v$ implies
81: $\{\sigma\rho:\sigma\in G\}$ is a basis for $L$ over
82: $K$.)
83:
84:
85:
86: In this paper, we ask for more. Let $L/K$ be a fully ramified
87: $p$-extension of local fields with finite residue field of
88: characteristic $p$, and let $v_L$ denote the normalized, additive
89: valuation. We ask, in addition to the above property, that there be
90: an explicit basis $\{\Theta_i\}$ of the group algebra $K[G]$ over $K$,
91: which may depend upon the extension $L/K$ but should be independent of
92: the element $\rho$, with the additional property that the valuations
93: associated with this basis, $\{v_L(\Theta_i\rho)\}$, give a complete
94: set of residues modulo $[L:K]$. These two ingredients, an integer
95: certificate and a basis, make up what we call a {\em Galois
96: scaffolding}.
97:
98:
99:
100: \paragraph{\bf Prototype: Cyclic extensions of degree $p$}
101: Let $L/K$ be a ramified, cyclic, degree $p$ extension of local fields with
102: $\mbox{Gal}(L/K)=\langle\sigma\rangle$. Assume that the ramification
103: break number for $L/K$ is $b$ and $\gcd(p,b)=1$. Note that this does
104: not restrict the extension when $K$ has characteristic $p$ and is only
105: a minor restriction when $K$ has characteristic $0$ \cite[III.~Prop
106: 2.3]{fesenko}. Let $\rho\in L$ be any element with $v_L(\rho)\equiv
107: b\bmod p$. Then $v_L((\sigma-1)^i\rho)\equiv(i+1)b\bmod p$ for $0\leq
108: i\leq p-1$. In particular, $v_L((\sigma-1)^i\rho)$ yields a complete
109: set of residues modulo $p$, and so we have a Galois scaffolding: Pick
110: any integer $\equiv b\bmod p$ and the basis, $\{(\sigma-1)^i:0\leq i\leq
111: p-1\}$.
112:
113: Galois scaffolding should be viewed as normal bases with the important
114: advantage that the valuation of any element expressed in terms of the
115: Galois scaffolding can be easily determined. In the example above,
116: since $L/K$ is fully ramified, every element $\alpha\in L$ can be
117: expressed as $\alpha=\sum_{i=0}^{p-1}a_i(\sigma-1)^i\rho$ for certain
118: $a_i\in K$. Then $v_L(\alpha)=\min\{v_L(a_i)+ib+v_L(\rho):0\leq i\leq
119: p-1\}$. We repeat ourselves for emphasis. Normal bases and power
120: bases (polynomial bases) in a prime element are two common bases. The
121: first allows the Galois action to be easily followed. The second
122: allows for an easy determination of valuation. These two properties are
123: usually at tension and so Galois scaffolding are remarkable for the
124: delicate balance that they achieve\footnote{It is easy to see that
125: Galois scaffolding are not universally available. Considering any
126: unramified extension, where there can be no integer certificate.}.
127:
128: Galois scaffolding in ramified, cyclic, degree $p$ extensions have
129: made Galois module structure in these extensions tractable
130: \cite{ferton,borevich,aiba,desmit}, along with Galois module structure
131: in fully ramified, cyclic, degree $p^2$ extensions
132: \cite{elder:annals}. In this paper, we will restrict our attention to
133: fully ramified elementary abelian extensions of local function fields
134: that are, in a particular sense, as simple as a ramified cyclic
135: extension of degree $p$, and give an explicit Galois scaffolding for
136: these extensions. We are motivated by the fact that much about Galois
137: module structure in wildly ramified extensions remains poorly
138: understood despite the topic's venerable age.
139:
140:
141:
142: \subsection{Notation}
143: Let $p$ be a prime integer and let $\mathbb{F}_p$ be the finite field
144: with $p$ elements. Let $K=\mathbb{F}((t))$ be a local function field
145: with residue field $\mathbb{F}$, which is either $\mathbb{F}_q$, a
146: finite field with $q$ elements where $q$ is a power of $p$, or
147: $\overline{\mathbb{F}}_p$, the algebraic closure. Let
148: $\wp:K\rightarrow K$ denote the $\mathbb{F}_p$-linear map
149: $\wp(x)=x^p-x$, and let $\phi$ denote the ring homomorphism
150: $\phi(x)=x^p$. Use subscripts to denote field of reference. So $\pi_K$
151: is a prime element of $K$, and $v_K$ is the valuation normalized so
152: that $v_K(\pi_K^t)=t$. Let $\euO_K=\{x\in K:v_K(x)\geq 0\}$ be the
153: valuation ring, and let $\euP_K=\pi_K\euO_K$ be its maximal ideal.
154: Let $L/K$ denote a fully ramified, Galois $p$-extension, with
155: $G=\mbox{Gal}(L/K)$. Define its ramification filtration by
156: $$G_i=\{\sigma\in G:v_L((\sigma-1)\pi_L)\geq i+1\}.$$
157:
158: \subsection{One-dimensional elementary abelian extensions}
159: It is a basic observation in Artin-Schreier Theory that the elementary
160: abelian extensions of $K$ lie in one-to-one correspondence with the
161: finite subspaces of the $\bF_p$-vector space, $K/K^\wp $, where $K^\wp
162: $ denotes the image of $\wp $.
163:
164: Assume for the moment that the residue field of $K$ is algebraically
165: closed, $\mathbb{F}=\overline{\mathbb{F}}_p$. Define
166: $K_{(n)}=\phi^n(K)=\mathbb{F}((t^{p^n}))$ for $n\geq 1$. Of course,
167: $K/K_{(n)}$ is an inseparable field extension, and so, in particular,
168: $K$ is a vector space over $K_{(n)}$. Since the residue field of $K$
169: is algebraically closed, $K/K^\wp$ is also a vector space over
170: $K_{(n)}$. We define {\em one-dimensional elementary abelian
171: extensions} to be those fully ramified elementary abelian extensions
172: of degree $p^i$ with $i\leq n+1$ that correspond to an $i$-dimensional
173: $\bF_p$-subspace of a one-dimensional $K_{(n)}$-subspace of
174: $K/K^\wp$. Of course, we are principally interested in maximal
175: extensions where $i=n+1$.
176:
177: More generally, we can include the finite residue field case and define
178: one-dimensional elementary abelian extensions of degree $p^{n+1}$ to
179: be those that can be expressed as $L=K(x_0,\ldots ,x_n)$ with
180: $\wp(x_i)=x_i^p-x_i=\phi^n(\Omega_i)\cdot \beta$ for some $\beta\in
181: K$ with $v_K(\beta)=-b$, $b>0$ and $\gcd(b,p)=1$; and some $\Omega_i\in K$
182: that span an $n+1$-dimensional subspace over $\bF_p$, with $\Omega_0=1$ and
183: $$v_K(\Omega_n)\leq \cdots \leq v_K(\Omega_1)\leq v_K(\Omega_0)=0.$$
184: Without any loss of generality, we can assume moreover that whenever
185: $v_K(\Omega_i)=\cdots =v_K(\Omega_j)$ for $i<j$, the projections
186: of $\Omega_i, \ldots \Omega_j$ into
187: $\phi^n(\Omega_i)\beta\euO_K/\phi^n(\Omega_i)\beta\euP_K$ are linearly
188: independent over $\mathbb{F}_p$. It should be clear from this
189: construction that the upper ramification numbers in one-dimensional
190: elementary abelian extensions of degree $p^{n+1}$ are congruent to
191: each other modulo $p^n$. Of course, the converse is not necessarily
192: true.
193:
194: Simple examples of a one-dimensional elementary abelian extensions are
195: \begin{enumerate}
196: \item[(1)] extensions of the form $K(y)$ with $y^q-y=\beta$ (Lemma 5.2).
197: \end{enumerate}
198:
199: It is probably not surprising that we are able to find Galois
200: scaffolding for a slightly broader class of extension, {\em near
201: one-dimensional elementary abelian extensions}, which arise when we
202: allow some error into the equations $\wp(x_i)=\phi^n(\Omega_i)\cdot
203: \beta$ defined above. In particular, we may replace those equations
204: with $\wp(x_i)\equiv \phi^n(\Omega_i)\cdot \beta+\epsilon_i$ for some
205: error terms $\epsilon_i\in K$ that satisfy a technical bound (6) and
206: use the same Galois scaffolding as for one-dimensional elementary
207: abelian extensions.
208:
209: Simple examples of near one-dimensional elementary abelian extensions are
210: \begin{enumerate}
211: \item[(2)] fully ramified biquadratic extensions (Lemma 5.1), and
212: \item[(3)] fully and weakly ramified $p$-extensions (Lemma 5.3).
213: \end{enumerate}
214:
215: Evidently, our Galois
216: scaffolding is not effected by small errors. This last observation
217: can be rephrased in terms of twists by characters of Galois
218: representations, along the lines of \cite[\S2.2.3]{elder:necbreaks}.
219:
220:
221:
222: \subsection{Galois scaffolding}
223: Assume the notation of the previous section and assume that $L/K$ is
224: near one-dimensional elementary abelian.
225:
226: Relabel $\Omega_j^{(0)}=\Omega_j$, and perform the following
227: elementary row operations on the matrix
228: $[\phi^i(\Omega_j^{(0)})]_{0\leq i,j\leq n}$, which in passing we note
229: resembles the square root of a discriminant matrix. The first column
230: is a column of $1$'s. So start with the $i=n$ row and work down to
231: the $i=1$ row, subtracting the $i-1$st row from the $i$th row. The
232: $i=0$ row and $i=0$ column of our matrix now agree with (1) below. If
233: we ignore them, the result is
234: $[\phi^{i-1}(\wp(\Omega_j^{(0)}))]_{1\leq i,j\leq n}$. Divide each
235: entry in a row by the first entry of the row. The result is
236: $[\phi^{i-1}(\wp(\Omega_j^{(0)})/\wp(\Omega_1^{(0)}))]_{1\leq i,j\leq
237: n}$. Define $\Omega_j^{(1)}=\wp(\Omega_j^{(0)})/\wp(\Omega_1^{(0)})$
238: for $1\leq j\leq n$. Observe that $v_K(\Omega_n^{(1)})\leq \cdots \leq
239: v_K(\Omega_1^{(1)})=0$ and that the $\{\Omega_j^{(1)}\}_{1\leq j\leq
240: n}$ span an $n$ dimensional vector space over $\mathbb{F}_p$. Again we
241: have a matrix $[\phi^{i-1}(\wp(\Omega_j^{(1)}))]_{1\leq i,j\leq n}$
242: whose first column is a column of $1$'s. Again, starting with the
243: $i=n$ row and working down to the $i=2$ row, we subtract the $i-1$st
244: row from the $i$th row. If we continue, following the same sequence of
245: steps as above, and repeat as often as necessary, we get
246: \begin{equation}
247: [\Omega]=\begin{bmatrix}
248: 1 & \Omega_1^{(0)} & \Omega_2^{(0)} &\cdots &\Omega_n^{(0)} \\
249: 0 & 1 &\Omega_2^{(1)} &\cdots &\Omega_n^{(1)} \\
250: & & \ddots & & &\\
251: 0 & 0 & \cdots &1& \Omega_n^{(n-1)}\\
252: 0 & 0 & \cdots &0& 1
253: \end{bmatrix}
254: \end{equation}
255: where $\Omega_j^{(0)}=\Omega_j$ and $\Omega_j^{(j)}=1$ for $0\leq
256: j\leq n$; and the $\Omega_j^{(i)}\in K$ for $1\leq i\leq n$ and $j> i$
257: are defined recursively by
258: $\Omega_j^{(i)}=\wp(\Omega_j^{(i-1)})/\wp(\Omega_{i}^{(i-1)})$.
259: Apply $\phi^{n-i-1}$ to row $i$ of $[\Omega]$, and get
260: $$[\Omega^\phi]=[\phi^{n-i-1}(\Omega_j^{(i)})]_{0\leq i,j\leq n}.$$
261:
262:
263: If we define the binomial coefficient $\binom{X}{i}$ by
264: $X\cdot(X-1)\cdots(X-i+1)/i!\in\mathbb{Q}[X]$, then we can define {\em
265: truncated exponentiation} to be the polynomial that results from the
266: truncation of the binomial series at the $p$th term:
267: \begin{equation}
268: (1+X)^{[Y]}:=\sum_{i=0}^{p-1}\binom{Y}{i}X^i\in\mathbb{Z}_{(p)}[X,Y]
269: \end{equation}
270: where $\mathbb{Z}_{(p)}$ denotes the integers localized at $p$.
271:
272: Choose $\sigma_i\in G=\mbox{Gal}(L/K)$ based upon our choice of
273: generators for $L/K$ by asking that
274: $$[(\sigma_i-1)x_j]=[\delta_{ij}]=I$$ ({\em i.e.}
275: $\sigma_ix_i=x_i+1$ and $\sigma_ix_j=x_j$ for $j\neq i$). Define
276: $$[\Delta_{i,j}]=[\Omega^\phi]^{-1}.$$
277: Now
278: for $0\leq i\leq n$ define $\Theta_{(i)}\in
279: K[\sigma_n,\sigma_{n-1},\ldots ,\sigma_{n-i}]$ recursively by
280: \begin{equation}
281: \Theta_{(i)}=\sigma_{n-i}\Theta_{(0)}^{[-\Delta_{n-i,n}]}
282: \Theta_{(1)}^{[-\Delta_{n-i,n-1}]}\cdots
283: \Theta_{(i-1)}^{[-\Delta_{n-i,n-(i-1)}]}.
284: \end{equation}
285:
286:
287:
288: Note that each
289: $\Theta_{(i)}$ is a 1-unit, {\em i.e.} $\Theta_{(i)}\in
290: 1+(\sigma-1:\sigma\in G)\subseteq K[G]$ where
291: $(\sigma-1:\sigma\in G)$ can be viewed as the augmentation ideal, the
292: Jacobson radical, or the nilradical. In particular, $\alpha^p=0$ for all
293: $\alpha\in (\sigma-1:\sigma\in G)$. This means that
294: $(\Theta_{(i)}-1)^p=0$, and so
295: $\Theta_{(i)}^{[\Delta_{j,k}]}\Theta_{(i)}^{[-\Delta_{j,k}]}=1$. As a
296: result, and since $\Delta_{n-r,n-r}=1$, the recursive definition for
297: $\Theta_{(i)}$ can be rewritten as
298: $$\sigma_{n-i}
299: =\Theta_{(0)}^{[\Delta_{n-i,n}]}\Theta_{(1)}^{[\Delta_{n-i,n-1}]}\cdots
300: \Theta_{(i-1)}^{[\Delta_{n-i,n-(i-1)}]}\Theta_{(i)}^{[\Delta_{n-i,n-i}]},$$
301: which
302: suggests
303: the matrix equation:
304: $$\begin{bmatrix}
305: \Delta_{0,0} & \Delta_{0,1} &\cdots & \Delta_{0,n} \\
306: 0 & \Delta_{1,1} &\cdots & \Delta_{1,n} \\
307: & & \ddots & &\\
308: 0 &\cdots &0 & \Delta_{n,n}
309: \end{bmatrix}\cdot \begin{bmatrix}
310: \Theta_{(n)}\\\Theta_{(n-1)}\\\vdots
311: \\\Theta_{(0)}\end{bmatrix}=\begin{bmatrix} \sigma_0\\\sigma_1\\\vdots
312: \\\sigma_n\end{bmatrix} ,$$ where addition is replaced by
313: multiplication and scalar multiplication is replaced truncated
314: exponentiation. Since truncated exponentiation does not distribute,
315: $(\Theta_{(i)}\Theta_{(j)})^{[\Delta]}\neq \Theta_{(i)}^{[\Delta]}
316: \Theta_{(j)}^{[\Delta]}$ (which is easy to check with $p=2$), we have
317: $[\Theta_{(n-j)}]\neq [\Omega^\phi]\cdot[\sigma_i]$,
318: despite the fact that $[\Omega^\phi]=[\Delta_{i,j}]^{-1}$. In other
319: words, this matrix equation is simply a convenient way to express a
320: recursive definition -- no more, no less.
321:
322: We are prepared to state the main result of the paper, which is proven
323: in \S3, \S4.
324:
325:
326: \begin{theorem}
327: Let $L/K$ be a near one-dimensional elementary abelian extension, as defined in \S1.2. Let
328: $\Theta_{(i)}\in K[\mbox{\rm Gal}(L/K)]$ be defined as in {\rm (3)}.
329: For $1\leq i\leq n$, let $m_i=v_K(\Omega_{i-1})-v_K(\Omega_i)$, and
330: choose any $\alpha_j\in K$ with
331: $v_K(\alpha_j)=p^{n-j-1}\sum_{i=j+1}^{n}p^im_i$. Let $b_m$ be the
332: largest {\rm (}lower{\rm )} ramification break number of $L/K$.
333: Given any $\rho\in L$ with $v_L(\rho)\equiv b_m\bmod p^{n+1}$ and
334: $a_s\in\{0,\ldots , p-1\}$,
335: $$v_L\left (\prod_{s=0}^{n}\alpha_{n-s}^{a_s}(\Theta_{(s)}-1)^{a_s}\rho\right
336: )=v_L(\rho)+\sum_{s=0}^{n} a_sp^sb_m.
337: $$
338: \end{theorem}
339:
340: As the integers $\sum_{s=0}^{n} a_sp^s$ run
341: through all possibilities from $0$ to $p^{n+1}-1$, the integers
342: $(\sum_{s=0}^{n} a_sp^s)b_m$ run through all residues modulo $p^{n+1}$.
343: Therefore
344: \begin{corollary}
345: $L$ has a Galois scaffolding.
346: \end{corollary}
347: \begin{corollary}
348: Any element in $L$ of valuation $b_m$ generates a normal field basis.
349: \end{corollary}
350: This last corollary provides evidence for the Conjecture in
351: \cite{elder:blms}.
352:
353: \section{Cyclic extensions of degree $p$}
354:
355: This paper is concerned with Galois, fully and thus wildly ramified
356: $p$-extensions that are, in a certain sense, as simple as cyclic
357: extensions of degree $p$. And so, we should take a moment to consider
358: the prototype: Let $L/K$ be a cyclic, ramified extension of degree
359: $p$. So $L=K(x)$ where $x$ satisfies $\wp(x)=x^p-x=\beta$ for some
360: $\beta\in K$ with $v_K(\beta)=-b$, $b>0$ and $\gcd(b,p)=1$. Let
361: $\langle\sigma\rangle=\mbox{Gal}(L/K)$ with $\sigma x=x+1$. Since
362: $v_L((\sigma-1)x)=0$, it is easy to see that the integer $b$ is the
363: ramification break number for $L/K$. Since $\wp(x)=\beta$ is really a
364: statement about the norm of $x$, namely $N_{L/K}(x)=\beta$, we have
365: $v_L(x)=-b$ as well.
366:
367:
368: We may rewrite $x^p-x=\beta$ as $x\cdot
369: \binom{x-1}{p-1}=-\beta$, where $\binom{x-1}{p-1}$ is a binomial
370: coefficient. Then
371: $$\binom{x-1}{p-1}\in L$$
372: generates $L/K$, satisfies
373: $v_L(\binom{x-1}{p-1})=-(p-1)b\equiv b\bmod p$ and, we contend,
374: is a particularly insightful element
375: to consider. Recall the definition of truncated
376: exponentiation and notice the striking similarity between
377: $$\sigma^{[i]}\binom{x-1}{p-1}=\sigma^i\binom{x-1}{p-1}= \binom{x-1+i}{p-1} \mbox{ for } 0\leq i\leq p-1,$$
378: and the equation in
379: \begin{lemma} Let $L=K(x)$ with $x^p-x=\beta\in K$ be a cyclic extension
380: with $\langle\sigma\rangle=\mbox{\rm Gal}(L/K)$.
381: Given $A\in L$,
382: $$\sigma^{[A]}\binom{x-1}{p-1}=\binom{x-1+A}{p-1}.$$
383: \end{lemma}
384: \begin{proof}
385: Recall Pascal's identity
386: $\binom{X}{i-1}+\binom{X}{i}=\binom{X+1}{i}\in \mathbb{Q}[X]$, which
387: can be rewritten as $\binom{X+1}{i}-\binom{X}{i}=\binom{X}{i-1}$. This
388: leads to the nice observation, used in \cite{desmit}, that
389: $(\sigma-1)\binom{x-1}{i}=\binom{x-1}{i-1}$ for $0\leq
390: i\leq p-1$, and therefore
391: $$(\sigma-1)^i\binom{x-1}{p-1}=\binom{x-1}{p-1-i}\quad\mbox{for }0\leq
392: i\leq p-1.$$
393: Under the substitution $X=\sigma-1$ and
394: $Y=A\in L$, we find
395: $$\sigma^{[A]}\binom{x-1}{p-1}=
396: \sum_{i=0}^{p-1}\binom{A}{i}(\sigma-1)^i\binom{x-1}{p-1}=
397: \sum_{i=0}^{p-1}\binom{A}{i}\binom{x-1}{p-1-i}\in L.$$ Vandermonde's
398: Convolution Identity $\sum_{i=0}^{p-1}\binom{X}{i}\binom{Y}{p-1-i}=
399: \binom{X+Y}{p-1}\in\mathbb{Z}_{(p)}[X,Y]$ results from considering the
400: coefficient of $Z^{p-1}$ in the identity
401: $(1+Z)^X(1+Z)^{Y}=(1+Z)^{X+Y}\in\mathbb{Q}[X,Y][[Z]]$. If we replace
402: $X=A$ and $Y=x$, we find
403: $\sum_{i=0}^{p-1}\binom{A}{i}\binom{x-1}{p-1-i}= \binom{x-1+A}{p-1}\in
404: L$.\end{proof}
405:
406: In \cite{elder:newbreaks} a refined ramification filtration was
407: introduced. It grew out of the possibility that the natural
408: $\mathbb{F}_p$-action on $\sigma$ could be extended to a residue field
409: ``action;'' a possibility that is certainly suggested by this striking
410: similarity.
411:
412: In this paper, we will develop a Galois scaffolding based on this similarity.
413: Specifically, we suppose that $L/K$ sits in a more general Galois
414: extension $M/N$, and we suppose furthermore that $L/N$ is normal and
415: that $\gamma\in\mbox{Gal}(M/N)$. So $\gamma^{-1} x=x+\delta$ for some
416: $\delta\in L$ and
417: $\sigma^{[\delta]}\binom{x-1}{p-1}=\binom{x-1+\delta}{p-1}=\gamma^{-1}\binom{x-1}{p-1}$. But
418: then
419: $$\gamma\sigma^{[\delta]}\binom{x-1}{p-1}=\binom{x-1}{p-1}.$$ If
420: $\delta\neq 0$ and $\gamma\not\in\langle\sigma\rangle$, then neither
421: $\sigma$ nor $\gamma$ individually fix the field generator
422: $\binom{x-1}{p-1}$. Yet together, using truncated exponentiation, they
423: do. As a result, if we suppose that $\delta\in N$ then the stabilizer
424: of $\binom{x-1}{p-1}$ in $N[\mbox{Gal}(M/N)]$ is larger than expected.
425:
426: \section{Galois scaffolding}
427:
428: This section is motivated by the observation of \S2 concerning the
429: stabilizer of $\binom{x-1}{p-1}$ and should be considered ``top-down.''
430: We begin with an generic abelian $p$-extension, which we ``organize''
431: using the ramification filtration. This ``organization'' defines a
432: matrix $[\Delta]$. If the coefficients of $[\Delta]$ lie in our base
433: field $K$, the extension satisfies a strong assumption, which makes it
434: possible for us to construct a Galois scaffolding, but also makes the
435: extension elementary abelian. At the end of the section, one question
436: remains: Are there any elementary abelian extensions that satisfy this
437: strong assumption? In \S4 we construct extensions that do -- from the
438: ``bottom-up.''
439:
440:
441:
442: Let $K_n/K$ be a fully
443: ramified, abelian extension of degree $p^{n+1}$. The case $n=0$
444: was addressed in \S1. So assume $n\geq 1$. Let $G=\mbox{Gal}(K_n/K)$
445: and let $G_i=\{\sigma\in G: v_n((\sigma-1)\pi_n)\geq i+1\}$ denote the
446: Hilbert ramification groups with break numbers $b_1<b_2<\cdots<b_m$
447: such that $G=G_{b_1}$, $G_{b_i}\supsetneq G_{b_i+1}=G_{b_{i+1}}$ and
448: $G_{b_m+1}=\langle e\rangle$. Because $K$ is characteristic $p$,
449: $\gcd(b_1,p)=1$, and by \cite[IV\S2 Prop 11]{serre:local}, $b_i\equiv
450: b_1\bmod p$.
451:
452:
453: Organize the extension by choosing a
454: filtration of $n+1$ subgroups that include the Hilbert ramification
455: groups and satisfy $G_{(i)}/G_{(i+1)}\cong C_p$,
456: $$G=G_{(0)}\supsetneq G_{(1)}\supsetneq \cdots\supsetneq
457: G_{(n)}\supsetneq G_{(n+1)}=\langle e\rangle.$$ Indeed, since each
458: quotient of consecutive Hilbert ramification groups is elementary
459: abelian, this is easy to do. The result is a set $\{\sigma_0,\sigma_1,
460: \ldots ,\sigma_n\}$ that generates $G$ (though probably not a minimal
461: generating set), such that $G_{(i)}=\langle \sigma_i,\sigma_{i+1},
462: \ldots , \sigma_n\rangle$ and the projection of $\sigma_i$ generates
463: $G_{(i)}/G_{(i+1)}\cong C_p$. For $i\geq 0$, let the fixed field
464: of $G_{(i)}$ be $K_{i-1}$, with $K_{-1}=K$ and define
465: $b_{(i)}=v_n((\sigma_i-1)\pi_n)-1$. This means that $b_{(0)}\leq
466: b_{(1)}\leq \ldots \leq b_{(n)}$ is a list of $n+1$ not necessarily
467: distinct integers and $\{b_{(0)},\ldots
468: ,b_{(n)}\}=\{b_1,\ldots , b_m\}$.
469:
470: Since $K_n/K$ is abelian, the Theorem of Hasse-Arf states that the
471: upper ramification numbers are integers \cite[IV\S3]{serre:local},
472: which is equivalent to
473: $b_i\equiv b_m\bmod [G:G_{b_{i+1}}]$ for $1\leq i\leq m$, and also
474: to
475: \begin{equation}
476: b_{(i)}\equiv b_{(n)}\bmod p^{i+1}\mbox{ for }0\leq i\leq n.
477: \end{equation}
478:
479: Since $\{b_{(0)},\ldots ,b_{(n)}\}$ is the set of ramification break
480: numbers for $K_n/K$, the ramification break numbers for $K_i/K$ are
481: $\{b_{(0)},\ldots ,b_{(i)}\}$ \cite[IV \S1 Prop 3 Cor]{serre:local}.
482: Altogether,
483: $\mbox{Gal}(K_i/K_{i-1})=G_{b_{(i)}}/G_{b_{(i+1)}}=\langle\bar{\sigma}_i\rangle\cong
484: C_p$, with $K_i/K_{i-1}$ having ramification break number $b_{(i)}$.
485: As a result, there are $X_i\in K_i$ such that $v_i(X_i)=-b_{(i)}$,
486: $\wp(X_i)=X_i^p-X_i=B_i\in K_{i-1}$ and $\sigma_iX_i=X_i+1$. Define
487: $$\Delta_{i,j}=(\sigma_i-1)X_j.$$ So $\Delta_{i,j}=0$ when $i>j$,
488: and
489: $\Delta_{i,i}=1$. Because
490: $X_j\in K_j$ and
491: $\sigma_i\sigma_j=\sigma_j\sigma_i$, we have
492: $\Delta_{i,j}\in K_{j-1}$ when $i<j$. Furthermore,
493: $v_j(\Delta_{i,j})=v_j((\sigma_i-1)X_j)=b_{(i)}-b_{(j)}\leq 0$.
494: Collect these $\Delta_{i,j}$ into a matrix, whose $j$th column lies in
495: $K_{j-1}$,
496: $$[\Delta]=\begin{bmatrix}
497: \Delta_{0,0} & \Delta_{0,1} &\cdots & \Delta_{0,n} \\
498: 0 & \Delta_{1,1} &\cdots & \Delta_{1,n} \\
499: & & \ddots & &\\
500: 0 &\cdots &0 & \Delta_{n,n}
501: \end{bmatrix}.
502: $$
503: Motivated by the final comment in \S2, and the fact that we want
504: a basis for $K[G]$ over $K$, we
505: impose
506: \begin{assumption}
507: $\Delta_{i,j}\in K\mbox{ for all }0\leq i,j\leq n$.
508: \end{assumption}
509: \begin{lemma}
510: Under Assumption 1,
511: $K_n/K$ is elementary abelian.
512: \end{lemma}
513: \begin{proof}
514: Since $\Delta_{i,j}\in K$, we have
515: $\sigma_i^kX_j=X_j+k\Delta_{i,j}$ for $0\leq k\leq p$. This means that
516: $\sigma_i^pX_j=X_j$ for all $0\leq i,j\leq n$, and in particular,
517: $\sigma_i^pX_n=X_n$ for all $0\leq i\leq n$. Since
518: $v_n(X_n)=-b_{(n)}$, we have $\gcd(v_n(X_n),p)=1$ and thus
519: $K_n=K(X_n)$.\end{proof}
520:
521: We will proceed in three steps towards our Galois scaffolding. First we choose
522: a nice element $\mathbb{X}\in K_n$ with $v_n(\mathbb{X})=b_{(n)}=b_m$.
523: Then we determine a basis for $K[G]$ over $K$ so that the valuations of
524: these basis elements applied to $\mathbb{X}$ yield a complete set of
525: residues mod $p^{n+1}$. Finally we prove in Proposition 3.3 that this second step
526: continues to hold when $\mathbb{X}$ is replaced by any element of
527: valuation $b_m\bmod p^{n+1}$.
528:
529:
530: Define
531: $$\rho=\prod_{j=0}^n\binom{X_j}{p-1}\in K_n.$$
532: Because of (4), we may
533: choose $\alpha_j\in K$ such that $v_j(\alpha_j)=b_{(n)}-b_{(j)}$.
534: Therefore $v_j(\alpha_j^{-(p-1)}\binom{X_j}{p-1})=-(p-1)b_{(n)}$ for
535: $0\leq j\leq n$. Choose $\alpha\in K$ with $v_K(\alpha)=b_{(n)}$
536: Define $\mathcal{A}=\alpha\prod_{j=0}^n\alpha_j^{-(p-1)}\in K$. So
537: $v_n(\mathcal{A})\equiv 0\bmod p^{n+1}$ and
538: $$\mathbb{X}=
539: \mathcal{A}\rho
540: =
541: \alpha\prod_{j=0}^n\alpha_j^{-(p-1)}\binom{X_j}{p-1}$$
542: has valuation $v_n(\mathbb{X})=p^{n+1}b_{(n)}-(p-1)\sum_{j=0}^np^{n-j}b_{(n)}=b_{(n)}=b_m$.
543:
544:
545:
546:
547: Recall (3), namely the recursive definition for $\Theta_{(i)}\in
548: K[G]$ for $0\leq i\leq n$.
549:
550:
551: \begin{lemma} For $0\leq i, j\leq n$,
552: $$
553: \Theta_{(i)}\binom{X_j}{p-1}=\begin{cases}
554: \binom{X_j}{p-1}&\mbox{if }j\neq n-i,\\
555: \binom{X_j+1}{p-1}&\mbox{if }j= n-i.
556: \end{cases}
557: $$
558: \end{lemma}
559: \begin{proof}
560: We proceed by induction. For $i=0$,
561: $\Theta_{(i)}=\Theta_{(0)}=\sigma_n$ and since $\sigma_n$ fixes
562: $K_{n-1}$ while $\binom{X_j}{p-1}\in K_j$, the result is clear. Now
563: assume the result for $0\leq i< k$ and consider
564: $\Theta_{(k)}\binom{X_j}{p-1}$. Because $\Theta_{(k)}$ is a product
565: (3), we need to examine the effect of each factor
566: $\Theta_{(i)}^{[-\Delta_{n-k,n-i}]}$ in that product, namely
567: $\Theta_{(i)}^{[-\Delta_{n-k,n-i}]}\binom{X_j}{p-1}$ for $0\leq i<k$.
568: By induction $(\Theta_{(i)}-1)^r
569: \binom{X_{n-i}}{p-1}=\binom{X_{n-i}}{p-1-r}$ for $0\leq r\leq p-1$,
570: and $(\Theta_{(i)}-1)^r \binom{X_j}{p-1}=0$ for $j\neq n-i$.
571: Therefore using Lemma 2.1, we have
572: $$\Theta_{(i)}^{[-\Delta_{n-k,n-i}]}\binom{X_j}{p-1}=\begin{cases}
573: \binom{X_j}{p-1}&\mbox{for }j\neq n-i,\\
574: \binom{X_j-\Delta_{n-k,j}}{p-1}&\mbox{for }j= n-i.
575: \end{cases}$$
576: If $j<n-k$, then every factor of $\Theta_{(k)}$ and thus
577: $\Theta_{(k)}$ acts trivially on $\binom{X_j}{p-1}$. If $j=n-k$ then
578: the only factor of $\Theta_{(k)}$ to act non-trivially is
579: $\sigma_{n-k}=\sigma_j$. As a result, $\Theta_{(k)}
580: \binom{X_j}{p-1}=\sigma_j\binom{X_j}{p-1}= \binom{X_j+1}{p-1}$. If
581: $j>n-k$, then exactly two factors of $\Theta_{(k)}$ to act
582: non-trivially, namely $\sigma_{n-k}$ and
583: $\Theta_{(n-j)}^{[-\Delta_{n-k,j}]}$. So
584: $$\Theta_{(k)}\binom{X_j}{p-1}=
585: \sigma_{n-k}\Theta_{(n-j)}^{[-\Delta_{n-k,j}]}\binom{X_j}{p-1}
586: =\sigma_{n-k}\binom{X_j-\Delta_{n-k,j}}{p-1}=\binom{X_j}{p-1}.$$\end{proof}
587:
588:
589:
590: Now notice that for $0\leq r\leq p-1$, we have
591: \begin{multline*}(\Theta_{(i)}-1)^r\mathbb{X}=(\Theta_{(i)}-1)^r\mathcal{A}\prod_{j=0}^n\binom{X_j}{p-1}
592: =\mathcal{A}\prod_{j\neq i}\binom{X_j}{p-1}\cdot (\Theta_{(i)}-1)^r
593: \binom{X_{n-i}}{p-1}\\=\mathcal{A}\prod_{j\neq i}\binom{X_j}{p-1}\cdot
594: \binom{X_{n-i}}{p-1-r}.
595: \end{multline*}
596: Therefore
597: $(\Theta_{(i)}-1)^r\mathbb{X}=\mathbb{X}\binom{X_{n-i}}{p-1-r}
598: \binom{X_{n-i}}{p-1}^{-1}$ and so
599: $v_n((\Theta_{(i)}-1)^r\mathbb{X})=b_n+rp^ib_{n-i}$.
600: Moreover given
601: $c_i\in\{0,1\ldots ,p-1\}$, we have
602: $$\prod_{i=0}^n(\Theta_{(i)}-1)^{c_i}\mathbb{X}=\mathcal{A}\prod_{j=0}^n\binom{X_j}{p-1-c_{n-j}},$$
603: and
604: using the $\alpha_j\in K$ with
605: $v_j(\alpha_j)=b_{(n)}-b_{(j)}$,
606: \begin{equation}
607: v_n\left (\prod_{i=0}^n\alpha_{n-i}^{c_i}(\Theta_{(i)}-1)^{c_i}\mathbb{X}\right)=\left(1+\sum_{i=0}^nc_ip^i\right)b_{(n)}.\end{equation}
608: Therefore
609: $$\left \{\prod_{i=0}^n\alpha_{n-i}^{c_i}(\Theta_{(i)}-1)^{c_i}:0\leq c_i\leq p-1\right \}$$
610: is the desired basis.
611:
612: \begin{proposition}
613: Under Assumption 1, we have a Galois scaffolding. Let $\mathbf{X}\in
614: K_n$ be any element with $v_n(\mathbf{X})\equiv b_{(n)}=b_m \bmod p^{n+1}$. Let
615: $\Theta_{(i)}\in K[G]$ be as defined in {\rm (3)}, and let
616: $\alpha_j\in K$ with $v_K(\alpha_j)=(b_{(n)}-b_{(j)})/p^{j+1}\in\mathbb{Z}$, then
617: $$v_n\left (\prod_{i=0}^n\alpha_{n-i}^{c_i}(\Theta_{(i)}-1)^{c_i}\mathbf{X}\right)=v_n(\mathbf{X})+\sum_{i=0}^nc_ip^ib_m.$$
618: \end{proposition}
619: \begin{proof}
620: Using (5), we can express $\mathbf{X}$ as a linear combination of
621: $\prod_{i=0}^n\alpha_{n-i}^{c_i}(\Theta_{(i)}-1)^{c_i}\mathbb{X}$ with
622: coefficients in $K$. It is enough therefore to show that
623: when we apply $\prod_{i=0}^n\alpha_{n-i}^{d_i}(\Theta_{(i)}-1)^{d_i}$
624: with $0\leq
625: d_i\leq
626: p-1$ to any term in this linear combination, we increase valuation by at least
627: $\sum_{i=0}^nd_ip^ib_m$, namely that
628: $$v_n\left(\prod_{i=0}^n\alpha_{n-i}^{c_i+d_i}(\Theta_{(i)}-1)^{c_i+d_i}\mathbb{X}\right)\geq
629: v_n\left(\prod_{i=0}^n\alpha_{n-i}^{c_i}(\Theta_{(i)}-1)^{c_i}\mathbb{X}\right)+\sum_{i=0}^nd_ip^ib_m.$$
630: If any sum $c_i+d_i\geq p$ then $(\Theta_{(i)}-1)^{c_i+d_i}=0$ and the
631: valuation of the left-hand-side is infinite. So we are left with the
632: case where all sums $c_i+d_i<p$. But in this case, we can use (5) to
633: determine that we have equality.
634: \end{proof}
635:
636:
637:
638:
639:
640:
641: \section{Near One-dimensional Elementary Abelian Extensions}
642:
643: In contrast with \S3, this section is ``bottom-up''. Motivated by the
644: idea of maximal refined ramification in \cite{elder:necbreaks}, we
645: follow \S1.2 and define the class of near one-dimensional elementary
646: abelian extensions, by describing how the generators of each extension
647: are related. We organize these generators by size (by valuation) as
648: in \S1.2, and then define the matrix $[\Omega^\phi]$ over $K$ as in
649: \S1.3. Our organization of the
650: generators, ``organizes'' the matrix $[\Omega^\phi]$. The main result
651: of the section is that this also ``organizes'' the extension in
652: essentially the same fashion as in \S3. In particular,
653: $[\Omega^\phi]\cdot[\Delta]=I$, which means that near one-dimensional
654: elementary abelian extension satisfy Assumption 1 and thus possess
655: Galois scaffolding.
656:
657: Recall the notation of \S1.2: Let $L=K(x_0, \ldots , x_n)$ with
658: $\wp(x_i)=\phi^n(\Omega_i)\cdot \beta +\epsilon_i$ for some $\beta\in
659: K$ with $v_K(\beta)=-b$, $b>0$ and $\gcd(b,p)=1$; some $\Omega_i\in K$
660: that span an $n+1$-dimensional subspace over $\bF_p$; and some ``error
661: terms'' $\epsilon_i\in K$, whose size will be controlled by (6)
662: below. Initially, we merely assume
663: $v_K(\epsilon_i)>v_K(\phi^n(\Omega_i)\beta)$, so the ramification
664: break number of $K(x_i)/K$ is $-v_K(\phi^n(\Omega_i)\beta)$.
665:
666: Furthermore recall $\Omega_0=1$ and that the other $\Omega_i$ are
667: ``organized'' (relabelled) so that $v_K(\Omega_n)\leq \cdots \leq
668: v_K(\Omega_1)\leq v_K(\Omega_0)=0$, and if $v_K(\Omega_i)=\cdots
669: =v_K(\Omega_j)$ for $i<j$, the projections of $\Omega_i, \ldots
670: \Omega_j$ into
671: $\phi^n(\Omega_i)\beta\euO_K/\phi^n(\Omega_i)\beta\euP_K$ are linearly
672: independent over $\mathbb{F}_p$. This means that $K(x_i,\ldots ,x_j)$
673: has one break in its ramification filtration at
674: $-v_K(\phi^n(\Omega_i)\beta)$.
675:
676: For $1\leq i\leq n$, define
677: $m_i=v_K(\Omega_{i-1})-v_K(\Omega_i)\geq 0$.
678: We control the size of the error terms with: For
679: $1\leq i\leq n$,
680: \begin{multline}
681: v_K(\epsilon_i)>-\frac{b}{p^n}-\sum_{j=1}^ip^jm_j+
682: \sum_{j=i+1}^n(p^n-p^j)m_j\\
683: =v_K(\phi^n(\Omega_i)\beta)+\frac{(p^n-1)b}{p^n}-(p-1)\sum_{j=1}^{n-1}p^jv_K(\Omega_j),
684: \end{multline}
685: which since $v_K(\Omega_j)\leq 0$ is clearly stronger than our initial
686: assumption, $v_K(\epsilon_i)>v_K(\phi^n(\Omega_i)\beta)$. Notice
687: further that if, for a particular $i$, the right-hand-side of (6) is
688: zero, then (6) is equivalent to ``no error'' ({\em i.e.}
689: $\epsilon_i=0$), since the inequality $v_K(\epsilon_i)>0$ implies
690: $\epsilon_i\in K^\wp$.
691:
692:
693: Choose $\sigma_i\in G=\mbox{Gal}(L/K)$ based upon our generators so
694: that $[(\sigma_i-1)x_j]=[\delta_{ij}]=I$. Define $H_{(i)}=\langle
695: \sigma_i, \ldots , \sigma_n\rangle$, and let $K_{i-1}=K(x_0,\ldots
696: ,x_{i-1})$ be the fixed field of $H_{(i)}$. So $K_{-1}=K$ and
697: $K_n=L$. As noted earlier,
698: $$u_{(i)}=b+p^n\sum_{j=1}^{i}m_j$$ is the ramification number of
699: $K(x_i)/K$, and is therefore an upper ramification number of $L/K$. By
700: considering our assumptions on the $\Omega_i$, one sees that the set
701: of upper ramification numbers is $\{u_{(0)}, \ldots , u_{(n)}\}$. We
702: may pass to the lower ramification numbers using the Herbrand function
703: $\psi(x)$ \cite[IV \S3]{serre:local}. Again considering our assumptions on
704: the $\Omega_i$, one sees that $\{b_{(0)}, \ldots ,b_{(n)}\}$ is the set of lower ramification numbers where
705: $$b_{(i)}=b+p^n\sum_{j=1}^{i}p^jm_j.$$ Moreover, $b_{(i)}$ is the
706: ramification number of $K_i/K_{i-1}$, and it is clear that the groups
707: $H_{(i)}$ are the groups $G_{(i)}$ defined in \S3. We can express the restriction on the error terms in (6) in terms of ramification numbers:
708: $v_K(\epsilon_i)>-b_{(n)}/p^n+u_{(n)}-u_{(i)}$.
709:
710: Our next step is to construct the $X_i\in K_i$ of \S3.
711: Recall
712: the $\Omega_j^{(i)}$ defined in \S1.3.
713: Define $X_j^{(0)}=x_j$. And
714: for $j\geq i$, recursively define
715: \begin{equation}
716: X_j^{(i)}=X_j^{(i-1)}- \phi^{n-i}(\Omega_j^{(i-1)})
717: X_{i-1}^{(i-1)}.
718: \end{equation}
719: If we use this definition to replace $X_j^{(i-1)}$ in (7) with
720: $X_j^{(i-2)}- \phi^{n-i+1}(\Omega_j^{(i-2)}) X_{i-2}^{(i-2)}$, we find
721: that $X_j^{(i)}=X_j^{(i-2)}- \phi^{n-i+1}(\Omega_j^{(i-2)})
722: X_{i-2}^{(i-2)}- \phi^{n-i}(\Omega_j^{(i-1)}) X_{i-1}^{(i-1)}$.
723: If we continue in this way, we eventually find
724: $X_j^{(i)}=X_j^{(0)}-\sum_{k=0}^{i-1} \phi^{n-k-1}(\Omega_j^{(k)})
725: X_{k}^{(k)}$. Consider the case $i=j$. Since $x_j=X_j^{(0)}$ and
726: $\Omega_j^{(j)}=1$, this can be rewritten as
727: $x_j=\sum_{k=0}^j\phi^{n-k-1}\Omega_j^{(k)}X_k^{(k)}$.
728: Recall that
729: $[\Omega^\phi]=[\phi^{n-i-1}(\Omega_j^{(i)})]_{0\leq i,j\leq n}$.
730: Therefore
731: $$[X_0^{(0)},X_1^{(1)},\ldots ,X_n^{(n)}]
732: \cdot[\Omega^\phi]=[x_0,x_1,x_2,\cdots , x_n].$$
733: Since
734: $I=[(\sigma_i-1)x_j]$,
735: we find that
736: $[(\sigma_i-1)X_j^{(j)}]\cdot[\Omega^\phi]=I$. Therefore
737: \begin{equation}
738: [(\sigma_i-1)X_j^{(j)}]=[\Omega^\phi]^{-1}.
739: \end{equation}
740:
741: Clearly $K_j=K(x_0,\ldots ,x_j)=K(X_0^{(0)},\ldots ,X_j^{(j)})$. If
742: we could determine that $v_j(X_j^{(j)})=-b_{(j)}$, then we could choose
743: the $X_j=X_j^{(j)}$ and find that $[\Delta_{i,j}]=[\Omega^\phi]^{-1}$.
744: As a result, our extension would satisfy Assumption 1. The remainder
745: of this section is therefore concerned with the valuation
746: $v_j(X_j^{(j)})$. Since the $\Omega_j^{(i)}$ are an important
747: ingredient in the definition of the $X_j^{(j)}$, given in (7), we need
748:
749: \begin{lemma} For $0\leq i<j\leq n$
750: $$v_K(\Omega_j^{(i)})=-p^i\sum_{k=i+1}^jm_k$$
751: \end{lemma}
752: \begin{proof} We induct on $i$.
753: Since $m_k= v_K(\Omega_{k-1}^{(0)})-v_K(\Omega_k^{(0)})$ for $1\leq
754: k\leq n$, the result holds for $i=0$.
755: For $i>1$, we assume the result. So in particular,
756: $v_K(\Omega_n^{(i-1)})\leq \cdots \leq v_K(\Omega_{i+1}^{(i-1)})\leq
757: v_K(\Omega_i^{(i-1)})=0$. Then
758: $v_K(\wp(\Omega_j^{(i-1)}))= pv_K(\Omega_j^{(i-1)})$
759: and thus using the definition for $\Omega_j^{(i)}$ in \S1.3,
760: we find that
761: $v_K(\Omega_j^{(i)})= pv_K(\Omega_j^{(i-1)})
762: -pv_K(\Omega_i^{(i-1)}) $ and result follows.
763: \end{proof}
764:
765: To assist in our analysis of $v_j(X_j^{(j)})$, define
766: $B_0=\beta$,
767: $E_j^{(0)}=\epsilon_j$ for $j>0$.
768: Then for $i>0$ recursively define
769: \begin{equation}
770: B_i=-\phi^{n-i}(\wp(\Omega_i^{(i-1)}))X_{i-1}^{(i-1)}+E_i^{(i-1)}
771: \end{equation}
772: and $E_j^{(i)}=E_j^{(i-1)}- \phi^{n-i}(\Omega_j^{(i)}) E_i^{(i-1)}$
773: for $j>i$. And $E_i^{(i)}=0$. The significance of these $B_i$ and
774: $E_j^{(i)}$ results from
775: \begin{lemma} For $j\geq i$
776: $$\wp(X_j^{(i)})=\phi^{n-i}(\Omega_j^{(i)})B_i+E_j^{(i)}$$
777: \end{lemma}
778: \begin{proof}
779: The statement is clear for $i=0$. Assume that it holds for $i-1$.
780: Therefore
781: $\wp(X_j^{(i-1)})=\phi^{n-i+1}(\Omega_j^{(i-1)})B_{i-1}+E_j^{(i-1)}$
782: and in particular, $\wp(X_{i-1}^{(i-1)})=B_{i-1}$.
783: Consider $\wp(X_j^{(i)})$.
784: It is easy to see that
785: $\wp(aX)=\phi(a)\wp(X)+\wp(a)X$. Therefore using (7) we find that
786: \begin{multline*}
787: \wp(X_j^{(i)})=\wp(X_j^{(i-1)})-
788: \phi^{n-i+1}(\Omega_j^{(i-1)})\wp(X_{i-1}^{(i-1)})-
789: \phi^{n-i}(\wp(\Omega_j^{(i-1)}))X_{i-1}^{(i-1)}\\
790: =\phi^{n-i+1}(\Omega_j^{(i-1)})B_{i-1}+ E_j^{(i-1)}-
791: \phi^{n-i+1}(\Omega_j^{(i-1)})B_{i-1}-
792: \phi^{n-i}(\wp(\Omega_j^{(i-1)}))X_{i-1}^{(i-1)}\\ = E_j^{(i-1)} -
793: \phi^{n-i}(\wp(\Omega_j^{(i-1)}))X_{i-1}^{(i-1)},
794: \end{multline*}
795: which, using (9), can be seen to agree with the statement for $i$.
796: \end{proof}
797:
798: \begin{lemma} Assume the bounds given in {\rm (6)}. Then for
799: $1\leq i\leq n$, we have $$v_K(E_i^{(i-1)})>-b_{(i)}/p^i.$$
800: \end{lemma}
801: \begin{proof}
802: Use Lemma 4.1 to determine that (6) is equivalent to
803: $$v_K(\phi^{n-i}(\Omega_{n}^{(i)})\epsilon_i)>-b_{(n)}/p^n.$$
804: We are interested in $v_K(E_i^{(i-1)})$. So
805: recall that
806: $E_j^{(i)}=E_j^{(i-1)}- \phi^{n-i}(\Omega_j^{(i)}) E_i^{(i-1)}$
807: for $j>i$, which means that $E_j^{(i)}=E_j^{(0)}- \sum_{k=1}^{i}
808: \phi^{n-k}(\Omega_j^{(k)}) E_k^{(k-1)}$, and
809: in particular,
810: \begin{equation}
811: E_i^{(i-1)}=\epsilon_i- \sum_{k=1}^{i-1}
812: \phi^{n-k}(\Omega_{i}^{(k)}) E_k^{(k-1)}.
813: \end{equation}
814: In order that
815: $v_K(E_i^{(i-1)})>-b_{(i)}/p^i$ for $1\leq i\leq n$, it is sufficient to
816: prove
817: \begin{eqnarray}
818: v_K(\epsilon_i)&>&-b_{(i)}/p^i\mbox{ for } 1\leq i\leq n,\mbox{ and}\\
819: v_K(\phi^{n-k}(\Omega_{i}^{(k)})
820: E_k^{(k-1)})&>&-b_{(i)}/p^i\mbox{ for } 1\leq k\leq i-1\leq n-1.
821: \end{eqnarray}
822:
823: Let $A_i=-b_{(i)}/p^i+v_K(\phi^{n-i}(\Omega_n^{(i)}))$. Using Lemma
824: 4.1, we find that
825: $-b_{(i)}/p^i+v_K(\phi^{n-i}(\Omega_n^{(i)}))=-b_{(i-1)}/p^i+v_K(\phi^{n-i+1}(\Omega_n^{(i-1)}))$. As
826: a result, $A_i>A_{i-1}$, since $-b_{(i-1)}/p^i>-b_{(i-1)}/p^{i-1}$. We
827: are given by (6) that
828: $v_K(\phi^{n-i}(\Omega_{n}^{(i)})\epsilon_i)>-b_{(n)}/p^n=A_n$. So
829: $v_K(\phi^{n-i}(\Omega_{n}^{(i)})\epsilon_i)>A_j$ for all $j$, including
830: $j=i$. Therefore (11) follows from (6).
831:
832: Focus on (12), which is equivalent to
833: $v_K(E_k^{(k-1)})>B_i^k$ where
834: $B_i^k=
835: -b_{(i)}/p^i-v_K(\phi^{n-k}(\Omega_{i}^{(k)}))$. Since
836: $-b_{(i)}/p^i-v_K(\phi^{n-k}(\Omega_{i}^{(k)}))=
837: -b_{(i-1)}/p^i-v_K(\phi^{n-k}(\Omega_{i-1}^{(k)}))$, we have
838: $B_i^k>B_{i-1}^k$. And thus (12) is equivalent to
839: \begin{equation}
840: v_K(\phi^{n-k}(\Omega_{n}^{(k)})
841: E_k^{(k-1)})>-b_{(n)}/p^n\mbox{ for }1\leq k\leq n-1.
842: \end{equation}
843:
844: Switch the roles of $i$ and $k$ in (10) and then apply
845: $\phi^{n-k}(\Omega_{n}^{(k)})$ to both sides:
846: $$\phi^{n-k}(\Omega_{n}^{(k)})E_k^{(k-1)}=
847: \phi^{n-k}(\Omega_{n}^{(k)})\epsilon_k- \sum_{i=1}^{k-1}
848: \phi^{n-k}(\Omega_{n}^{(k)})\phi^{n-i}(\Omega_{k}^{(i)})
849: E_i^{(i-1)}.$$
850:
851: By Lemma 4.1,
852: $v_K(\phi^{n-k}(\Omega_{n}^{(k)})\phi^{n-i}(\Omega_{k}^{(i)}))=
853: v_K(\phi^{n-i}(\Omega_{n}^{(i)}))$. Therefore (13) follows from (6) by
854: induction on $k$.
855: \end{proof}
856:
857: \begin{lemma} Assume the bounds in {\rm (6)}. Then for $0\leq j\leq n$, $v_j(X_j^{(j)})=-b_{(j)}$.
858: \end{lemma}
859: \begin{proof}
860: It is clear that $v_0(X_{0}^{(0)})=-b_{(0)}$. So for $i>0$, assume
861: that
862: $v_{i-1}(X_{i-1}^{(i-1)})=-b_{(i-1)}=-b-p^n\sum_{j=1}^{i-1}p^jm_j$.
863: Using Lemma 4.1, we see that
864: $v_K(\wp(\Omega_i^{(i-1)}))=-p^im_i$. So
865: $v_K(\phi^{n-i}(\wp(\Omega_i^{(i-1)})))=-p^nm_i$ and therefore
866: $v_{i-1}(\phi^{n-i}(\wp(\Omega_i^{(i-1)})))=-p^n\cdot p^im_i$. So
867: $v_{i-1}(\phi^{n-i}(\wp(\Omega_i^{(i-1)})X_{i-1}^{(i-1)})=-b_{(i)}$.
868: By Lemma 4.3, $v_{i-1}(E_i^{(i-1)})>-b_{(i)}$. Therefore
869: $v_{i-1}(B_i)=-b_{(i)}$.
870: Lemma 4.2 implies that in particular the norm
871: $N_{K_{i}/K_{i-1}}(X_i^{(i)})=\wp(X_i^{(i)})=B_i$, which means that
872: $v_{i}(X_i^{(i)})=-b_{(i)}$.
873: \end{proof}
874:
875: As a result, we can put all this together and find
876: \begin{proposition}
877: Near one-dimensional elementary abelian extensions satisfy Assumption 1.
878: \end{proposition}
879:
880:
881: \section{Examples of near one-dimensional elementary abelian extensions}
882:
883: \begin{lemma}
884: Fully ramified biquadratic extensions are near one-dimensional
885: elementary abelian extensions.
886: \end{lemma}
887: \begin{proof}
888: Biquadratic extensions are special in that there is only one
889: nontrivial residue modulo $2$. Let $L/K$ be a fully ramified
890: biquadratic extension. We may assume that $L=K(x_0,x_1)$ with
891: $x_0^2-x_0=\beta$, $x_1^2-x_1=\beta_1$, $v_K(\beta_1)\leq
892: v_K(\beta)<0$ and both of $v_K(\beta_1)$ and $v_K(\beta)$ odd.
893: Because the difference of two odd numbers is even, there is a
894: $\mu_0\in K$ such that $\mu_0^2\beta_1\equiv \beta\bmod
895: \beta\euP_K$. Let $\beta=\mu_0^2\beta_1+\tau_0$ for some
896: $v_K(\tau_0)>v_K(\beta)$. Since we can replace $\beta$ by any element
897: in its coset $\beta+K^{\wp}$, we may assume $v_K(\tau_0)=0$, or
898: $v_K(\tau_0)<0$ with $v_K(\tau_0)$ odd. If $v_K(\tau_0)$ odd, then
899: there is a $\mu_1\in K$ such that $\mu_1^2\beta_1\equiv \tau_0\bmod
900: \tau_0\euP_K$, and thus $\beta=(\mu_0+\mu_1)^2\beta_1+\tau_1$ for
901: $v_K(\tau_1)>v_K(\tau_0)$. Continue in this way until
902: $\beta=\mu^2\beta_1+\tau$ for some $\mu\in K$ and either $\tau=0$ or
903: $v_K(\tau)=0$.
904:
905: If $\tau=0$, then $\beta_1=\mu^{-2}\beta$ and the extension is
906: one-dimensional. If $v_K(\tau)=0$, then
907: $\beta_1=\Omega_1^2\beta+\epsilon_1$ where $\epsilon_1=-\tau\mu^{-2}$
908: and $\Omega_1=\mu^{-1}$. Continuing to translate into the notation of
909: \S4, we note that $b=-v_K(\beta)$ and
910: $m_1=-v_K(\Omega_1)=v_K(\mu)$. So $v_K(\epsilon_1)=-2m_1>-b/2-2m_1$,
911: which is the inequality given by (6). So the extension is near
912: one-dimensional.
913: \end{proof}
914: \begin{lemma}
915: Let $K=\mathbb{F}((t))$ with $\mathbb{F}_q\subseteq \mathbb{F}$, and
916: let $\beta\in K$ with $v_K(\beta)<0$ and $\gcd(v_K(\beta),p)=1$. Then
917: $L=K(y)$ with $y^q-y=\beta$ is a one-dimensional
918: elementary abelian extension of $K$.
919: \end{lemma}
920: \begin{proof}
921: Let $q=p^f$ and let $\{1=\omega_0,\omega_1, \cdots ,\omega_{f-1}\}$ be
922: a basis for $\mathbb{F}_q$ over $\mathbb{F}_p$. Then
923: $x_i=\sum_{r=0}^{f-1}\phi^r(\omega_iy)$ where $y^q-y=\beta$ satisfies
924: $x_i^p-x_i=\omega_i\beta$. Of course $\phi$ is an automorphism of
925: $\mathbb{F}_q$. So we let may set $\Omega_i=\phi^{-f+1}(\omega_i)$.
926: \end{proof}
927:
928: The following class of fully and weakly ramified $p$-extensions ({\em i.e.}
929: with $G=G_1$ and $G_2=\{e\}$) is notable for being wildly ramified while
930: possessing a normal integral basis (for the maximal ideal)
931: \cite{ullom}.
932:
933: \begin{lemma}
934: Let $L/K$ be a noncyclic, fully and weakly ramified $p$-extension,
935: then $L/K$ is a near one-dimensional elementary abelian extension.
936: \end{lemma}
937: \begin{proof}
938: The extension is elementary abelian \cite[IV \S2]{serre:local}, with
939: one break in its ramification filtration at $b=1$. As a result there
940: is only one upper ramification break number, also at $u=1$. Thus
941: $L=K(x_0,x_1,\ldots x_n)$ with $v_K(\wp(x_i))=-1$. Let
942: $\beta=\wp(x_0)$. Then there are units $\omega_i\in\mathbb{F}$ such
943: that $\wp(x_i)=\omega_i\beta\bmod \euO_K$. Since $\phi$ is an
944: automorphism of $\mathbb{F}$, we may let
945: $\Omega_i=\phi^{-n}(\omega_i)$ and find $\epsilon_i\in \euO_K$ such
946: that $\wp(x_i)=\phi^n(\Omega_i)\beta+\epsilon_i$ with either
947: $\epsilon_i=0$ or $v_K(\epsilon_i)=0$. Using the notation of \S4, we
948: find that $m_i=0$ and in all cases $v_K(\epsilon_i)\geq
949: 0>-1/p^n=-b/p^n$, which is (6).
950: \end{proof}
951:
952:
953: \bibliography{bib} \end{document}