1: % Note #1: p-adic numbers were discovered by KUMMER (>=1850) AND Hensel 

2: %          (at least one book (e.g., Bosch, Non-Arch Ana...) 

3: %          cites just Hensel, around 1905) 

4: %

5: 

6: \documentclass[notitlepage]{jams-l} 

7: \usepackage{amssymb,amsmath,latexsym,epsfig,euscript,floatfig,wrapfig} 

8: 

9: \setlength{\topmargin}{.25in} 

10: \setlength{\textwidth}{6in} 

11: \setlength{\oddsidemargin}{.125in} 

12: \setlength{\evensidemargin}{.125in} 

13: \setlength{\textheight}{9.7in} 

14: 

15: \newlength{\jmr}

16: \settowidth{\jmr}{{\footnotesize Rojas, J.\ Maurice}}

17: \newlength{\hwl}

18: \settowidth{\hwl}{{\footnotesize Lenstra, Hendrik W.}}

19: \newlength{\khov}

20: \settowidth{\khov}{{\footnotesize Khovanski, Askold G.}}

21: 

22: \newtheorem{smi}{Smirnov's Theorem} 

23: \renewcommand{\thesmi}{\unskip}

24: \newtheorem{nota}{Notation} 

25: \renewcommand{\thenota}{\unskip}

26: \newtheorem{lip}{Lipshitz's Theorem}

27: \renewcommand{\thelip}{\unskip}

28: \newtheorem{thm}{Theorem}

29: \newtheorem{lemma}{Lemma}

30: \newtheorem{prop}{Proposition}

31: \newtheorem{dfn}{Definition}

32: \newtheorem{cor}{Corollary}

33: \newtheorem{rem}{Remark}	

34: \newtheorem{ex}{Example}

35: 

36: \newcommand{\gf}{\mathrm{Gal}(K/\Q)} 

37: \newcommand{\tf}{{\tilde{f}}}

38: \newcommand{\twF}{{\tilde{F}}} 

39: \newcommand{\blah}{\hspace{1in}} 

40: \renewcommand{\mod}{\mathbf{mod}} 

41: \newcommand{\pspa}{{\mathbf{PSPACE}}} 

42: \newcommand{\pnp}{{$\mathbf{P}^\np$}} 

43: \newcommand{\ph}{{\mathbf{PH}}}  

44: \newcommand{\am}{{\mathbf{AM}}} 

45: \newcommand{\hie}{{\mathbf{coNP}^{\mathbf{NP}}}}

46: \newcommand{\np}{{\mathbf{NP}}}

47: \newcommand{\conp}{{\mathbf{coNP}}}

48: \newcommand{\corp}{{\mathbf{coRP}}}

49: \newcommand{\rp}{{\mathbf{RP}}}

50: \newcommand{\bpp}{{\mathbf{BPP}}}

51: \newcommand{\crap}{\pp^{\np^\np}}

52: \newcommand{\pp}{\mathbf{P}}

53: \newcommand{\hn}{\mathbf{HN}}

54: \newcommand{\nc}{\mathbf{NC}}

55: \newcommand{\expt}{{\mathbf{EXPTIME}}}

56: \newcommand{\eps}{\varepsilon}

57: \newcommand{\cA}{\mathcal{A}}

58: \newcommand{\cD}{\mathcal{D}}

59: \newcommand{\cE}{\mathcal{E}}

60: \newcommand{\od}{{\bar{d}}}

61: \newcommand{\oD}{{\bar{D}}}

62: \newcommand{\oE}{{\bar{E}}}

63: \newcommand{\oF}{{\bar{F}}}

64: \newcommand{\oP}{{\bar{P}}}

65: \newcommand{\cN}{\mathcal{N}}

66: \newcommand{\cO}{\mathcal{O}}

67: \newcommand{\supp}{\mathrm{Supp}}

68: \newcommand{\init}{\mathrm{In}}

69: \newcommand{\conv}{\mathrm{Conv}}

70: \newcommand{\size}{\mathrm{size}}

71: \newcommand{\thth}{{\underline{\mathrm{th}}}}

72: \newcommand{\rd}{ {\underline{ \mathrm{rd} } }  }

73: \newcommand{\st}{ {\underline{ \mathrm{st} } }  }

74: \newcommand{\nd}{{\underline{\mathrm{nd}}}}

75: \newcommand{\Pro}{{\mathbb{P}}}

76: \newcommand{\sres}{\mathrm{sres}}

77: \newcommand{\F}{\mathbb{F}}

78: \newcommand{\Q}{\mathbb{Q}}

79: \newcommand{\R}{\mathbb{R}}

80: \newcommand{\C}{\mathbb{C}}

81: \newcommand{\N}{\mathbb{N}}

82: \newcommand{\Z}{\mathbb{Z}}

83: \newcommand{\fii}{\varphi} 

84: \newcommand{\cp}{\mathfrak{p}} 

85: 

86: \newcommand{\ch}{\mathrm{char}}

87: \newcommand{\Le}{\mathrm{Length}} 

88: \newcommand{\ord}{\mathrm{ord}}

89: \newcommand{\area}{\mathrm{Area}}

90: \newcommand{\spec}{\mathrm{Spec}}

91: \newcommand{\divisor}{\mathrm{Div}}

92: \newcommand{\codim}{\mathrm{codim}}

93: \newcommand{\choo}{\mathrm{Chow}}

94: \newcommand{\pert}{\mathrm{Pert}}

95: \newcommand{\choa}{\choo_{\mathrm{A}}}

96: \newcommand{\fan}{\mathrm{Fan}}

97: \newcommand{\verte}{\mathrm{Vert}}

98: \newcommand{\cone}{\mathrm{Cone}} 

99: \newcommand{\res}{\mathrm{Res}}

100: \newcommand{\affres}{\mathrm{AffRes}}

101: \newcommand{\newt}{\mathrm{Newt}} 

102: \newcommand{\hyp}{\mathrm{Hyper}} 

103: \newcommand{\hilb}{\mathrm{\mathbf{Hilb}}} 

104: 

105: \newcommand{\Sn}{\mathcal{S}^{n-1}}

106: \newcommand{\Non}{(\N\cup\{0\})^n}

107: \newcommand{\Zn}{\Z^n}

108: \newcommand{\Zm}{\Z^m}

109: \newcommand{\Zno}{\Z^n\!\setminus\!\{\bO\}} 

110: \newcommand{\Qn}{\Q^n} 

111: \newcommand{\Rn}{\R^n}

112: \newcommand{\Rm}{\R^m}

113: \newcommand{\Cn}{\C^n} 

114: \newcommand{\Cm}{\C^m} 

115: \newcommand{\Cs}{\C^*}

116: \newcommand{\Rs}{\R^*} 

117: \newcommand{\Rsn}{{(\R^*)}^n}

118: \newcommand{\Csn}{{(\C^*)}^n}

119: \renewcommand{\qed}{$\blacksquare$}

120: \newcommand{\cL}{{\mathcal{L}}}

121: \newcommand{\cM}{{\mathcal{M}}}

122: \newcommand{\cH}{\mathcal{H}}

123: \newcommand{\cI}{\mathcal{I}}

124: \newcommand{\cR}{\mathcal{R}}

125: \newcommand{\cP}{\mathcal{P}} 

126: \newcommand{\cQ}{\mathcal{Q}}

127: \newcommand{\cS}{\mathcal{S}}

128: \newcommand{\cC}{\mathcal{C}}

129: \newcommand{\cU}{\mathcal{U}}

130: \newcommand{\bO}{\mathbf{O}}

131: \newcommand{\vol}{\mathrm{Vol}}

132: \newcommand{\htp}{\mathrm{HTP}}

133: \newcommand{\ratcurve}{\mathrm{RatCurve}}

134: \newcommand{\biggy}{\mathrm{Big}}

135: \newcommand{\card}{\mathrm{ExactCard}}

136: \newcommand{\cd}{\mathfrak{d}}

137: \newcommand{\lie}{\mathrm{Li}}

138: 

139: \newenvironment{mymatrix}{\begin{array}{*{20}{c}}}{\end{array}}

140: 

141: \begin{document}

142: 

143: \title[Arithmetic Fewnomial Systems]{Finiteness for 

144: Arithmetic Fewnomial Systems} 

145: 

146: \subjclass{Primary 

147: 11G25; % Arithmetic Algebraic Geometry; varieties over finite and 

148:        % local ground fields

149: Secondary 

150: 11G35, % ...global ground fields

151: 14D10, % algebraic geometry; families, fibrations; arithmetic ground fields

152: 14G20. % algebraic geometry; arithmetic problems/diophantine geometry; 

153:        % local ground fields 

154: } 

155: 

156: \author{J.\ Maurice Rojas}\thanks{

157: This research was partially supported by Hong Kong UGC Grant \#9040469-730 

158: and a grant from the Texas A\&M Faculty of Sciences.} 

159: 

160: \address{Department of Mathematics, Texas A\&M University, College Station, 

161: Texas 77843-3368, USA. }  

162: \email{rojas@math.tamu.edu } 

163: \urladdr{http://www.math.tamu.edu/\~{}rojas } 

164: 

165: \dedicatory{To my god-daughter, Monica Althoff. }  

166: 

167: \date{\today} 

168: 

169: \mbox{}\\

170: \vspace{-.9in} 

171: \maketitle

172: 

173: \mbox{}\\

174: \vspace{-1.5cm} 

175: \begin{abstract} 

176: Suppose $\cL$ is any finite algebraic extension of either the ordinary 

177: rational numbers or the $p$-adic rational numbers. Also let $g_1,\ldots,g_k$ 

178: be polynomials in $n$ variables, with coefficients in $\cL$, such that the 

179: total number of monomial terms appearing in at least one $g_i$ is exactly $m$. 

180: We prove that the maximum number of isolated roots of 

181: $G\!:=\!(g_1,\ldots,g_k)$ in $\cL^n$ 

182: is finite and depends solely on $(m,n,\cL)$, i.e., is independent of the 

183: degrees of the $g_i$. We thus obtain an arithmetic analogue of Khovanski's 

184: Theorem on Fewnomials, extending earlier work of Denef, Van den Dries, 

185: Lipshitz, and Lenstra. 

186: \end{abstract} 

187: 

188: \section{Introduction} 

189: A consequence of Descartes' Rule (a classic result dating back to 1641) 

190: is that any real univariate polynomial with exactly $m\!\geq\!1$ monomial 

191: terms has at most $2m-1$ real roots. In this paper, we extend this result 

192: in two directions at once: we bound the number of isolated roots of 

193: polynomial {\bf systems} over {\bf $\pmb{\cp}$-adic} fields, independent of 

194: the degrees of the underlying polynomials. As 

195: a consequence, we also obtain analogous results over any number 

196: field. The resulting bounds are, unfortunately, non-explicit. 

197: Nevertheless, the existence of such bounds on the number of roots was 

198: previously unknown in the multivariate non-Archimedean and number field cases. 

199: So let us now detail our results and some important precursors. 

200: 

201: Extending the sharp bound of Descartes' Rule to 

202: polynomial systems has already proven difficult in the special case 

203: of the real numbers $\R$ (perhaps the most familiar metrically complete field): 

204: the best general result in this direction, Khovanski's Theorem on Fewnomials  

205: \cite{kho,few}, gives an explicit bound on the number of non-degenerate 

206: real roots\footnote{One can then extend this to counting isolated roots and 

207: connected components via some algebraic tricks. See, e.g., \cite{real,tri}.} 

208: that is independent of the underlying polynomial degrees,  

209: but exponential in the number of monomial terms. 

210: Whether the number of non-degenerate roots is in fact {\bf polynomial} 

211: in the number of monomial terms for every {\bf fixed} number of 

212: equations and variables is an intriguing open question, and the 

213: answer is still unknown even in the case of two polynomials in two unknowns. 

214: (The polynomial system $(x^2_1-x_1,\ldots,x^2_n-x_n)$ easily shows us 

215: that fixing $n$ is necessary if we would like polynomiality in the number of 

216: monomial terms.) 

217: 

218: As for the $p$-adic univariate case, Hendrik W.\ Lenstra, Jr.\ has shown that 

219: for any degree $d$ algebraic extension $L$ of $\Q_p$, the maximum number of 

220: roots in $L$ of a $g\!\in\!L[x_1]$ with exactly $m\!\geq\!2$ 

221: monomial terms is no more than $1+1.582 \cdot (p^{f_L}-1)(m-1)^2(1+

222: \frac{e_L\log(e_L(m-1)/\log p)}{\log p})$ \cite[prop.\ 7.2]{lenstra}, 

223: where $e_L$ and $f_L$ respectively denote the {\bf ramification degree} and 

224: {\bf residue field degree} of $L$. (Recall that $e_L$ and $f_L$ are integers 

225: satisfying $e_Lf_L\!=\!d$; see section \ref{sec:local} for their definitions.) 

226: In particular, this bound is independent of the degree of $g$. As a 

227: consequence, Lenstra also derived a bound of $1+4.566\cdot 

228: (m-1)^2(d+10)2^d(\log(d(m-1))+0.367)$ \cite[prop.\ 8.1]

229: {lenstra} for the analogous situation where one replaces $L$ by any degree 

230: $d$ algebraic extension $K$ of $\Q$.\footnote{

231: In both cases, a bound of $1$ is 

232: trivial to derive when $m\!\leq\!1$, and the bounds we quoted have the added 

233: benefit of counting the multiplicities of the non-zero roots. Furthermore, the 

234: result over $K$ actually is a stronger statement which in fact bounds the 

235: number of roots of bounded degree over $K$. } 

236: 

237: By recent work of the author \cite{tri} this polynomiality  

238: of the number of roots in the number of monomials can be extended 

239: to certain systems of $n$ polynomials in $n$ unknowns, provided 

240: we fix $n$ and restrict to real algebraic number fields. 

241: (The example from the paragraph before last tells us that 

242: fixing $n$ is necessary in the $\cp$-adic and number field cases as well.) 

243: However, at the expense of less explicit bounds, one can 

244: extend Lenstra's results much farther. 

245: 

246: \begin{nota} 

247: Let $\mathbf{\cL}$ be a field and $\cL^*\!:=\!\cL\setminus\{0\}$. 

248: If $\pmb{G}\!:=\!(g_1,\ldots,g_k)$ where, for all 

249: $i$, \scalebox{.85}[1]{$\pmb{g_i}\!\in\!\cL[x_1,\ldots,x_n]\setminus\{0\}$,} 

250: and the number of monomial terms appearing in at least one $g_i$ is exactly 

251: $m$, then we call $G$ a {\bf ($\pmb{k\times n}$)  

252: $\pmb{m}$-sparse polynomial system (over $\pmb{\cL}$)}. 

253: Also, we say a root $\zeta$ of $G$ is {\bf isolated} 

254: (resp.\ {\bf non-degenerate}) iff $\zeta$ is an irreducible component 

255: of the zero set of $G$ over the algebraic closure of $\cL$ 

256: (resp.\ $k\!=\!n$ and the Jacobian of $G$, evaluated at $\zeta$, is 

257: invertible). 

258: \end{nota} 

259: 

260: \begin{thm}

261: \label{thm:local}

262: For any (rational) prime $p$ and positive integer $d$, let $L$ be any degree 

263: $d$ algebraic extension of $\Q_p$. Also let $G$ be any 

264: $k\!\times\!n$ $m$-sparse polynomial system over $L$. Then there is an 

265: absolute constant $\gamma(n,m)$ such that the number of isolated roots of $G$ 

266: in $L^n$ is no more \mbox{than $p^{dn}(1-\frac{1}{p^{f_L}})^n\gamma(n,m)$.} 

267: \end{thm} 

268: \begin{cor} 

269: \label{cor:global}

270: Let $K$ be any degree $d$ algebraic extension of $\Q$ and let $G$ be 

271: any $k\!\times\!n$ $m$-sparse polynomial system over $K$. Then 

272: the number of isolated roots of $G$ in $K^n$ 

273: is no more \mbox{than $2^{dn}(1-\frac{1}{2^d})^n\gamma(n,m)$.} 

274: \end{cor} 

275: \noindent 

276: Theorem \ref{thm:local} generalizes an analogy over $\Z_p$, 

277: initiated by Jan Denef and Lou Van den Dries in \cite{vandenef}, of 

278: Khovanski's Theorem on Fewnomials. Corollary \ref{cor:global} establishes 

279: a higher-dimensional analogue of Lenstra's aforementioned result for 

280: univariate sparse polynomials over number fields. We can also extend 

281: our finiteness results even further to count isolated roots of bounded degree 

282: over $L$ or $K$ (see corollary \ref{cor:bigger} of section \ref{sec:extend}). 

283: 

284: We prove theorem \ref{thm:local} and corollary \ref{cor:global}

285: in sections \ref{sec:local} and \ref{sec:global} respectively.  

286: The proofs, while short, involve deep non-effective results of Jan 

287: Denef and Lou Van den Dries \cite{vandenef} and Leonard Lipshitz 

288: \cite{lipshitz} on $p$-adic sub-analytic functions,  

289: as well as an elegant extension of the 

290: classical $p$-adic Newton polygon by A.\ L.\ Smirnov \cite{smirnov}. In 

291: particular, aside from the case $n\!=\!1$ (cf.\ remark \ref{rem:heart} of 

292: the next section), there appear to be no explicit bounds on the function 

293: $\gamma(n,m)$ yet. So a more direct and effective approach would be of the 

294: utmost interest. 

295: 

296: \subsection{$p$-adic Analysis and $p$-adic Newton Polytopes} 

297: \label{sec:padic} 

298: \mbox{}\\

299: 

300: We first state the following combined paraphrase of two results of Lipshitz: 

301: \begin{lip} (See \cite[thm.\ 2]{lipshitz}.) 

302: For any (rational) prime $p$, let $\C_p$ denote the completion (with respect 

303: to any $p$-adic metric) of the algebraic closure of $\Q_p$. 

304: Also let $G$ be any $n\!\times\!n$ $m$-sparse polynomial system over $\C_p$. 

305: Then there is an absolute constant $\beta'(n,m)$ (independent of $p$) such 

306: that $G$ has no more than $\beta'(n,m)$ isolated roots 

307: $x\!:=\!(x_1,\ldots,x_n)\!\in\!\C^n_p$ 

308: satisfying $|x_i-1|_p\!\leq\!\frac{1}{p}$ for all $i$, where $|\cdot|_p$ 

309: denotes the unique $p$-adic norm on $\C_p$ with $|p|_p\!=\!\frac{1}{p}$. 

310: \qed

311: \end{lip} 

312: \begin{rem} 

313: In the above $p$-adic context, we also have the following equivalent 

314: definition of isolation for roots: a root 

315: $x$ of $G$ is isolated iff for some $\eps\!>\!0$, 

316: we have $\max_i|x_i-x'_i|_p\!>\!\eps$ for every other root 

317: $(x'_1,\ldots,x'_n)$ of $G$. So in essence, an isolated root of $G$ in 

318: $\C^n_p$ can be contained within a small $p$-adic ``brick,'' away from all 

319: other roots of $G$. Lipshitz's original statement in fact dealt with roots with 

320: {\bf algebraic integer} coordinates in $\C^n_p$, but the statement above is 

321: equivalent since the ultrametric inequality implies 

322: $|x|_p\!\leq\!\max\{|x-1|_p,|1|_p\}\!=\!1$.  

323: \end{rem} 

324: \begin{rem} 

325: \label{rem:heart} 

326: Lenstra has derived an explicit 

327: upper bound on the number of roots $x_1$ of $g_1$ in $\C_p$ with 

328: $|x_1-1|_p\!\leq\!\frac{1}{p^{r}}$ for any given 

329: $r\!>\!0$ \cite[prop.\ 7.1]{lenstra}.\footnote{ The original statement was in 

330: terms of $\ord_p$, counted 

331: multiplicities, and in fact gave a decreasing function of $p$. } 

332: Taking $r\!=\!1$ one then obtains 

333: $\beta'(1,m)\!\leq\!1.582\cdot(m-1)(1+1.443\cdot(\log(m-1)+0.367))$ for 

334: $m\!\geq\!2$. (Note that $\beta'(n,0)\!=\!\beta'(n,1)\!=\!0$ for all $n$.) 

335: Whether Lenstra's explicit bound on the number of roots in $\C_p$ 

336: ``$p$-adically close to the identity'' extends to $n\times n$ sparse 

337: polynomial systems is an open problem, even in the case $n\!=\!2$. 

338: Nevertheless, the proofs of theorem \ref{thm:local} and corollary 

339: \ref{cor:global} are structured so that explicit bounds on $\gamma(n,m)$ can 

340: be easily derived should such a result become available. 

341: \end{rem} 

342: 

343: \noindent 

344: Lipshitz's Theorem is based partially on an earlier result of 

345: Denef and Van den Dries \cite[pg.\ 105]{vandenef} over the subring $\Z_p$ 

346: but also injects model-theoretic techniques (see \cite{lipshitz} for further 

347: details).  

348: 

349: The key to proving theorem \ref{thm:local} is to further limit 

350: the number of roots defined over a subfield of $\C_p$ by seeing which 

351: possible vectors of valuations can occur. In particular, we will 

352: use the following extension of the classical univariate 

353: $p$-adic Newton polygon (see, e.g., \cite[ch.\ IV, sec.\ 3]{koblitz} for the 

354: latter construction). To clarify the statement, let us make the following 

355: definitions: 

356: \begin{dfn} 

357: \label{dfn:poly}

358: For any $a\!=\!(a_1,\ldots,a_n)\!\in\!\Zn$, let 

359: $x^a\!:=\!x^{a_1}_1\cdots x^{a_n}_n$. 

360: Writing any $g\!\in\!\cL[x_1,\ldots,x_n]$ in the form $\sum_{a\in \Zn} 

361: c_ax^a$, we call $\pmb{\supp(g)}\!:=\!\{a \; | \;

362: c_a\!\neq\!0\}$ the {\bf support} of $g$. 

363: Then, for any $n\times n$ polynomial system $G$ over $\C_p$, its {\bf 

364: $\pmb{n}$-tuple of $\pmb{p}$-adic Newton polytopes}, 

365: $\pmb{\Delta_p(G)}\!=\!(\Delta_p(g_1),\ldots,\Delta_p(g_n))$, is 

366: defined as follows: $\pmb{\Delta_p(g_i)}\!:=\!\conv(\{(a,\ord_p c_a) \; | \; 

367: a\!\in\!\supp(g_i) \})\!\subset\!\R^{n+1}$, where $\conv(S)$ denotes the 

368: convex hull of\footnote{i.e., smallest convex set containing...} 

369: a set $S\!\subseteq\!\R^{n+1}$ and $\ord_p : \C_p \longrightarrow 

370: \Q\cup\{+\infty\}$ is the usual discrete valuation\footnote{So, for example, 

371: $\ord_p 0\!=\!+\infty$ and $\ord_p(p^kr)\!=\!k$ whenever $r$ is a unit in 

372: $\Z_p$ and $k\!\in\!\Q$.} of $\C_p$. Finally, for any $w\!\in\!\Rn$ and any 

373: compact subset $B\!\subset\!\Rn$, let the {\bf face of $\pmb{B}$ with inner 

374: normal $\pmb{w}$}, $\pmb{B^w}$, be the set of points $x\!\in\!B$ which 

375: minimize the inner product $w\cdot x$. 

376: \end{dfn} 

377: \begin{ex}

378: Consider the $2\!\times\!2$ $6$-sparse polynomial system 

379: \[F\!:=\!(50x^{18}_1-3125x^9_2-162x_2,49x^{18}_2-35x^9_1-109375x_1)\] 

380: over $\Q_5$. Then the corresponding pair of $5$-adic Newton 

381: polytopes is  

382: \[\Delta_5(F)\!=\!(\conv(\{(18,0,2),(0,9,5),(0,1,0)\}),

383: \conv(\{(0,18,0),(9,0,1),(1,0,6)\})).\] 

384: Note that each polytope is in fact a triangle embedded in $\R^3$. 

385: \end{ex}  

386: 

387: \begin{smi} 

388: \cite[thm.\ 3.4]{smirnov}

389: \label{thm:smirnov}

390: Let $\ord_p x$ be the vector $(\ord_p x_1,\ldots,\ord_p x_n)$ and 

391: let $v\!:=\!(v_1,\ldots,v_n)\!\in\!\Rn$. Then 

392: for any $n\!\times\!n$ polynomial system $G$ over $\C_p$, 

393: the number of isolated roots $x\!:=\!(x_1,\ldots,x_n)$ of $G$ in $(\C^*_p)^n$ 

394: satisfying $\ord_p x\!=\!v$ (counting multiplicities) 

395: is no more than $\cM(\pi(\Delta^{v'}_p(g_1)),\ldots,\pi(\Delta^{v'}_p(g_n)))$, 

396: where $v'\!:=\!(v_1,\ldots,v_n,1)$, $\pi : \R^{n+1}\longrightarrow \Rn$ is the 

397: natural projection forgetting the $v_{n+1}$ coordinate, $\cM$ denotes 

398: mixed volume \cite{buza} (normalized so that 

399: \scalebox{1}[1]{$\cM(\conv(\{\bO,e_1,\ldots,e_n\}),

400: \ldots,\conv(\{\bO,e_1,\ldots,e_n\}))\!=\!1$}), and 

401: $e_i$ is the $i^\thth$ standard basis vector of $\Rn$. \qed 

402: \end{smi} 

403: \begin{rem} 

404: \label{rem:convex} 

405: Note that the number of roots of $G$ in $(\C^*_p)^n$ with given 

406: valuation vector 

407: thus depends strongly on the individual exponents of $G$ --- not just on 

408: the number of monomial terms. However, the number of possible distinct 

409: valuation vectors occuring for any single $G$ can be combinatorially bounded 

410: from above as a function depending solely on $n$ and the number of monomial 

411: terms (cf.\ section \ref{sec:local}). In particular, it is only the 

412: {\bf lower}\footnote{Those with positive 

413: $x_{n+1}$ coordinate for their inner normals...} faces of the Newton polytopes 

414: that matter. 

415: \end{rem} 

416: 

417: \noindent 

418: Note that Smirnov's Theorem provides a non-Archimedean extension of 

419: Bernstein's famous mixed volume bound over $(\C^*)^n$ \cite{bkk,jpaa}.  

420: We also point out that aside from a result of Kamel A.\ Atan and 

421: J.\ H.\ Loxton in the $2\times 2$ case \cite{loxton}, Smirnov's result appears 

422: to be the first higher-dimensional version of the classical univariate  

423: $p$-adic Newton polygon. 

424: 

425: While we will leave the algorithmic issues of $p$-adic Newton polytopes 

426: for another paper, let us at least observe one salient fact before pointing 

427: out references to the computational literature: In searching for 

428: $v'\!\in\!\R^{n+1}$ giving a positive number of roots with valuation vector 

429: $v$, it suffices to restrict one's search to the inner normals of the lower 

430: $n$-dimensional faces of the {\bf Minkowski sum}\footnote{The Minkowski sum 

431: of any finite collection of subsets $A_1,\ldots,A_k\subseteq\!\Rn$ is simply 

432: the set $\{a_1+\cdots+a_k \; | \; a_i\in A_i$ for all $i\}$. } 

433: $\Delta_p(g_1)+\cdots+\Delta_p(g_n)$. The last fact follows from basic convex 

434: geometry (see, e.g., \cite{buza}). While there currently seems to 

435: be no direct software implementation of $p$-adic Newton polytopes, the 

436: underlying algorithms have already been implemented in the related context 

437: of mixed volume computation, and a detailed description including 

438: complexity bounds can be found in \cite{emiphd,sparse}. 

439: \begin{ex}

440: A convenient way to visualize how many roots with valuation vector $v$ 

441: appear for a given $2\times 2$ polynomial system is to draw the projected 

442: lower faces of the underlying Minkowski sum. For instance, in our last 

443: example, we obtain the following representation:\footnote{In the 

444: illustration, the {\bf ordinary} Newton polygons we refer to are 

445: a construction similar to the $p$-adic Newton polygon, embedded in $\R^2$ 

446: instead of $\R^3$, where one essentially uses the {\bf trivial} valuation 

447: ($|a|\!=\!1$ for any $a\!\in\!\C^*_p$) instead of the $p$-adic 

448: valuation. }\\ 

449: \mbox{}\hspace{1in}\epsfig{file=5adic.ps,height=2.9in} 

450: 

451: It is then easily checked\footnote{In this case, a brute-force search 

452: among the cross-products of the pairs of triangle edges suffices to generate 

453: our normal vectors: $2$ out of the resulting $9$ possibilities are true 

454: inner normals of lower $2$-dimensional faces of the underlying Minkowski sum. 

455: Each resulting pair $(\pi(\Delta^{v'}_5(g_1)),\pi(\Delta^{v'}_5(g_2)))$ 

456: turns out to be a pair of line segments. The mixed area of any such pair 

457: (whose Minkowski sum is a lighter-colored cell in our illustration) is simply 

458: the absolute value of the determinant of the underlying vectors. }   

459: that among the $323$ roots $(x_1,x_2)$ of $G$ in $(\C^*_5)^2$, there are 

460: exactly $8$ with $(\ord_5 x_1,\ord_5 x_2)\!=\!(\frac{5}{8},\frac{53}{4})$ 

461: and $315$ with $(\ord_5 x_1,\ord_5 x_2)\!=\!(-\frac{1}{9},0)$. 

462: In particular, we see that there are roots lying in extensions of $\Q_5$ 

463: of degree at least $9$. 

464: \end{ex} 

465: 

466: \section{Roots of Bounded Degree Over $\cp$-adic Fields 

467: and Number Fields } 

468: \label{sec:extend} 

469: Here we use an observation on $p$-adic algebraic extensions to 

470: prove the following combined strengthening of theorem \ref{thm:local} 

471: and corollary \ref{cor:global}. First, let us say that 

472: a vector $x\!:=\!(x_1,\ldots,x_n)$ defined over the algebraic closure of a 

473: field $\cL$ {\bf is of degree $\pmb{\delta}$ over $\pmb{\cL}$} iff each 

474: $x_i$ is of degree $\delta$ over $\cL$. 

475: \begin{cor}

476: \label{cor:bigger}  

477: Suppose $G$ is a $k\!\times\!n$ $m$-sparse polynomial system over $\cL$, 

478: where $\cL$ is a degree $d$ algebraic extension of either $\Q$ or $\Q_p$. 

479: Then for any positive integer $\delta$, there is an absolute constant 

480: $\gamma'(d,\delta,n,m)$ (resp.\ 

481: $\gamma'_p(d,\delta,n,m)$) such that $G$ has no more than 

482: $\gamma'(d,\delta,n,m)$ (resp.\ 

483: $\gamma'_p(d,\delta,n,m)$) isolated roots in 

484: $\C^n$ (resp.\ $\C^n_p$) 

485: of degree $\leq\!\delta$ over $\cL$, 

486: according as $\cL$ is an algebraic extension of $\Q$ or $\Q_p$.   

487: \end{cor} 

488: 

489: \noindent 

490: {\bf Proof:} Focusing first on the case where $\cL$ is an algebraic extension 

491: of $\Q_p$, note that there are only finitely many algebraic extensions 

492: of degree $\leq\!d\delta$ of $\Q_p$ \cite[ch.\ II, prop.\ 14]{lang}. 

493: Letting $L$ be the compositum of all these fields, note that 

494: $L$ is then also a finite algebraic extension of $\Q_p$.  

495: More to the point, any root of $G$ in $\C^n_p$ of degree 

496: $\leq\!\delta$ over $\cL$ must then also lie in $L^n$. 

497: So the $p$-adic case of our corollary follows 

498: immediately from theorem \ref{thm:local}. 

499: 

500: To prove the case where $\cL$ is an algebraic extension of 

501: $\Q$, note that such an $\cL$ embeds naturally as a subfield of the $L$ we 

502: just defined for the $p$-adic case. Taking $p\!=\!2$ to fix ideas, 

503: we then see that the degree of $\cL\Q_2$ over $\Q_2$ is 

504: no more than $d$, and thus any $x_i$ of degree $\leq\!\delta$ over $\cL$ 

505: embeds in an extension of $\Q_2$ of degree $\leq\!d\delta$. Thus any such 

506: $x_i$ can be assumed to lie in $L$, and we again conclude by theorem 

507: \ref{thm:local}. \qed 

508: 

509: \begin{rem} 

510: Note that we immediately obtain from our proof above 

511: that \[\gamma'_p(d,\delta,n,m) \leq p^{nD_p}(1-\frac{1}{p^{f_p}})^n\gamma(n,m) 

512: \text{ \ and \ } 

513: \gamma'(d,\delta,n,m)

514: \leq 2^{nD_p}(1-\frac{1}{2^{f_2}})^n\gamma(n,m),\] 

515: where $D_p$ (resp.\ $f_p$) is the degree (resp.\ residue field degree) over 

516: $\Q_p$ of the compositum of all algebraic extensions of $\Q_p$ of degree 

517: $\leq\!d\delta$.  

518: \end{rem}  

519: 

520: \section{Proving our Main Local Result (Theorem \ref{thm:local})} 

521: \label{sec:local} 

522: 

523: The following lemma will allow us to reduce to the case 

524: $k\!=\!n$. 

525: \begin{lemma}

526: \label{lemma:gh}

527: Following the notation of theorem \ref{thm:local}, there is a matrix 

528: $[a_{ij}]\!\subset\!\Q^{n\times k}_p$ such that

529: the zero set of 

530: $G\!:=\!(a_{11}g_1+\cdots+a_{1k}g_k,\ldots,a_{n1}g_1+\cdots+a_{nk}g_k)$

531: in $\C^n_p$ is the union of the zero set of $G$ in $\C^n_p$ and a finite

532: (possibly empty) set of points. \qed  

533: \end{lemma}  

534: \noindent 

535: A stronger version of the above lemma appears 

536: in \cite[sec.\ 3.4.1]{giustiheintz}, but phrased over $\C$ instead. 

537: However, the proof there carries over to any algebraically closed field with 

538: no difficulty whatsoever. 

539: 

540: Returning to the proof of theorem \ref{thm:local}, we see that lemma 

541: \ref{lemma:gh} allows us to replace $G$ by a new 

542: $n\times n$ polynomial system (clearly still $m$-sparse) which 

543: has at least as many isolated roots as our original $G$. 

544: Abusing notation slightly, let $G$ denote this new $n\times n$ 

545: polynomial system. 

546: 

547: Applying Smirnov's Theorem to $G$, recall that 

548: $\cM(\pi(\Delta^{v'}_p(g_1)),\ldots,\pi(\Delta^{v'}_p(g_n)))\!>\!0 

549: \Longrightarrow \ v'$ is an inner normal of a lower $n$-dimensional face of 

550: the Minkowski sum $\Sigma_G\!:=\!\Delta_p(g_1)+\cdots+\Delta_p(g_n)$ (cf.\ 

551: section \ref{sec:padic}). 

552: It is then easily checked 

553: that $\Sigma_G$ has at most $mn$ vertices and thus, since 

554: any $n$-dimensional face consists of at least $n+1$ vertices, $\Sigma_G$ 

555: has at most $\begin{pmatrix}mn\\ n+1\end{pmatrix}$ 

556: $n$-dimensional faces. In particular, this implies that 

557: the number of distinct values for the vector 

558: $\ord_p x$, where $x\!\in\!(\C^*_p)^n$ is a root of $G$, is no more than 

559: $\begin{pmatrix}mn\\ n+1\end{pmatrix}$. So let us fix $v\!\in\!\Rn$ and 

560: see how many roots of $G$ in $(L^*)^n$ can have valuation vector $v$. 

561: 

562: Let $R_p\!:=\!\{a\!\in\!\C_p \; 

563: | |a|_p\!\geq\!1\}$ be the ring of algebraic integers in $\C_p$, $M_p$ the 

564: unique maximal ideal of $R_p$, $\F_L\!:=\!(R_p\cap L)/(M_p\cap L)$,  

565: and let $\pi$ be any generator of the principal ideal $M_p\cap L$ 

566: of $R_p\cap L$. Also let $e\!:=\!\max_{y\in L^*}\{ 

567: |\ord_p y|^{-1}\}$ and $f\!:=\!\log_p\#\F_L$.  

568: The last two quantities are respectively known as the {\bf ramification 

569: degree} and {\bf residue field degree} of $L$,  

570: and we can in fact pick $\pi$ so that $\pi^e\!=\!p$ as well 

571: \cite[ch.\ III]{koblitz}. Doing this, then 

572: fixing a set $A_L\!\subset\!R_p$ of representatives for $\F_L$ 

573: (i.e., a set of $p^f$ elements of $R_p\cap L$, exactly {\bf one} of 

574: which lies in $M_p$, whose image 

575: mod $M_p\cap L$ is $\F_L$), we can then write any 

576: $x_i\!\in\!L$ uniquely as $\sum^{+\infty}_{j=e\ord_p(x_i)}a^{(i)}_j\pi^j$ for 

577: some sequence of $a^{(i)}_j\!\in\!A_L$ \cite[cor., pg.\ 68, sec.\ 3, ch.\ III]

578: {koblitz}. 

579: 

580: Note in particular that 

581: $\frac{\sum^{+\infty}_{j=t}c_j\pi^j}{\pi^t(c_t+\cdots+c_{e+t-1}\pi^{e-1})}

582: \!\in\!R_p$, and in fact $\left|\frac{\sum^{+\infty}_{j=t}c_j\pi^j}

583: {\pi^t(c_t+\cdots+c_{e+t-1}\pi^{e-1})}-1\right|_p\!\leq\!\frac{1}{p}$,  

584: for any sequence of representatives $(c_t,c_{t+1},\ldots)\!\in\!A^{+\infty}_L$ 

585: with $c_t\!\in\!A_L\setminus M_p$. 

586: 

587: Now consider the polynomial system $H$ where  

588: \[H(z_1,\ldots,z_n)\!:=\!G(\pi^{ev_1}(c^{(1)}_{ev_1}+\cdots+c^{(1)}_{ev_1+e-1}

589: \pi^{e-1})z_1,\ldots,

590: \pi^{ev_n}(c^{(n)}_{ev_n}+\cdots+c^{(n)}_{ev_n+e-1}\pi^{e-1})z_n),\] 

591: for any fixed vectors $(c^{(1)}_{ev_1},\ldots,c^{(1)}_{ev_1+e-1}),\ldots,   

592: (c^{(n)}_{ev_n},\ldots,c^{(n)}_{ev_n+e-1})\!\in\!A^{e}_L$ with 

593: $c^{(1)}_{ev_1},\ldots,c^{(n)}_{ev_n}\!\not\in\!M_p$. 

594: Lipshitz's Theorem then tells us that the number of 

595: isolated roots $z$ of $H$ in $\C^n_p$ satisfying\\ 

596: $|z_1-1|_p,\ldots,|z_n-1|_p\!\leq\!\frac{1}{p}$ is no more than 

597: $\beta'(n,m)$. 

598: 

599: Since there are $(p^f-1)p^{f(e-1)}\!=\!p^d(1-\frac{1}{p^f})$ possibilities for 

600: each $e$-tuple $(c^{(i)}_0,\ldots,c^{(i)}_{e-1})$,  

601: our last observation tells us that the number of isolated roots 

602: $x$ of $G$ in $(L^*)^n$ satisfying 

603: $\ord_p x\!=\!v$ is no more than 

604: $p^{dn}(1-\frac{1}{p^f})^n\beta'(n,m)$. So the total number of isolated roots 

605: of $G$ in $(L^*)^n$ is no more than 

606: $\begin{pmatrix}mn\\ n+1\end{pmatrix}p^{dn}(1-\frac{1}{p^f})^n\beta'(n,m)$. 

607: 

608: To conclude, we simply set all possible subsets of the variables equal 

609: to zero (which of course never increases the number of monomial terms) 

610: and apply our result recursively to the resulting polynomial systems 

611: in fewer variables. 

612: We thus obtain our theorem, along with an obvious bound of 

613: $\gamma(n,m)\!\leq\!1+\sum^n_{i=1} \begin{pmatrix}n \\ 

614: i\end{pmatrix}\begin{pmatrix}mi\\ i+1\end{pmatrix}\beta'(i,m)$ for all 

615: $n\!\in\!\N$. \qed 

616: 

617: \section{Proving Our Main Global Result (Corollary \ref{cor:global})} 

618: \label{sec:global} 

619: Since $\Q$ naturally embeds in $\Q_p$ for any prime $p$, 

620: $K$ embeds in a degree $d$ algebraic extension, $L$, of 

621: $\Q_p$. So let us fix $p\!=\!2$, say. Our corollary then follows  

622: immediately from theorem \ref{thm:local}. 

623: 

624: \begin{rem}  

625: The following improved bound for corollary \ref{cor:global} 

626: follows immediately from our proof above: 

627: $2^{dn}(1-\frac{1}{2^{f_L}})^n\gamma(n,m)$, where $L$ is 

628: as in the proof. 

629: \end{rem} 

630: 

631: \section*{Acknowledgements} 

632: The author thanks Peter B\"urgisser for informing him of 

633: the important paper \cite{vandenef}, Angus McIntyre 

634: for suggesting this paper to Peter B\"urgisser, and 

635: Hendrik W.\ Lenstra, Jr.\ and Leonard Lipshitz for useful e-mail discussions. 

636: Special thanks 

637: go to an anonymous referee for graciously and elegantly pointing out some 

638: earlier errors of the author which were neither graceful nor elegant. 

639: The author is also very grateful to the architects and staff 

640: of {\tt MathSciNet} ({\tt http://ams.rice.edu/mathscinet/search}), 

641: without which, he wouldn't have found the paper \cite{lipshitz} until much 

642: later. 

643: 

644: I dedicate this paper to my god-daughter, Monica Althoff. 

645: 

646: \footnotesize

647: \bibliographystyle{acm}

648: \begin{thebibliography}{A}

649: 

650: \bibitem[AL86]{loxton} Atan, Kamel A.\ and Loxton, J.\ H., {\it

651: ``Newton Polyhedra and Solutions of Congruences,''} 

652: Diophantine Analysis (Kensington, 1985), pp.\ 67--82, 

653: London Math.\ Soc.\ Lecture Note Ser., 109, 

654: Cambridge Univ.\ Press, Cambridge, 1986. 

655: 

656: \bibitem[BS96]{bs} Bach, Eric and Shallit, Jeff, {\it

657: Algorithmic Number Theory, Vol.\ I: Efficient Algorithms,}

658: MIT Press, Cambridge, MA, 1996.

659: 

660: \bibitem[BKK76]{bkk} Bernstein, D.\ N., Kushnirenko, A.\ G., and

661: Khovanski, A.\ G., {\it ``Newton Polyhedra,''} Uspehi Mat.\ Nauk 31 (1976),

662: no.\ 3(189), pp.\ 201--202.

663: 

664: \bibitem[BZ88]{buza} Burago, Yu. D. and Zalgaller, V. A., {\it

665: Geometric Inequalities,} Grundlehren der mathematischen Wissenschaften 285,

666: Springer-Verlag (1988).

667: 

668: \bibitem[DV88]{vandenef} Denef, Jan and van den Dries, Lou, {\it

669: ``$p$-adic and Real Subanalytic Sets,''} Annals of Mathematics (2) 

670: {\bf 128} (1988), no.\ 1, pp.\ 79--138. 

671: 

672: \bibitem[Emi94]{emiphd} Emiris, Ioannis Z., {\it ``Sparse Elimination and

673: Applications in Kinematics,''} Ph.D. dissertation, Computer Science

674: Division, U. C. Berkeley (December, 1994), available on-line at {\tt

675: http://www.inria.fr/saga/emiris}.

676: 

677: \bibitem[ER01]{sparse} Emiris, Ioannis Z.\ and Rojas, J.\ Maurice, {\it 

678: ``Some Sparse Results on Sparse Elimination,''} preprint.  

679: 

680: \bibitem[GH93]{giustiheintz} Giusti, Marc and Heintz, Joos,

681: {\it ``La d\'etermination des points isol\'es et la dimension

682: d'une vari\'et\'e alg\'ebrique peut se faire en temps polynomial,''}

683: Computational Algebraic Geometry and Commutative Algebra (Cortona,

684: 1991), Sympos.\ Math.\ XXXIV, pp.\ 216--256, Cambridge University

685: Press, 1993.

686: 

687: \bibitem[Kho80]{kho} Khovanski, Askold G., {\it ``On a Class of 

688: Systems of Transcendental Equations,''} Dokl.\ 

689: Akad.\ Nauk SSSR {\bf 255} (1980), no.\ 4, pp.\ 804--807; 

690: English transl.\ in Soviet Math.\ Dokl.\ {\bf 22} (1980), 

691: no.\ 3. 

692: 

693: \bibitem[Kho91]{few} \underline{\hspace{\khov}}, {\it Fewnomials,}

694: AMS Press, Providence, Rhode Island, 1991.

695: 

696: \bibitem[Kob84]{koblitz} Koblitz, Neal I., {\it $p$-adic Numbers, $p$-adic 

697: Analysis, and Zeta-Functions,}

698: $2^\nd$ ed., Graduate Texts in Mathematics, 58, Springer-Verlag, 

699: New York-Berlin, 1984.

700: 

701: \bibitem[Lan94]{lang} Lang, Serge, {\it ``Algebraic Number Theory,''} 

702: $2^\nd$ ed., Springer-Verlag, New York, 1994. 

703: 

704: \bibitem[Len99]{lenstra} \underline{\hspace{\hwl}}, {\it ``On the 

705: Factorization of Lacunary Polynomials,''} Number Theory in Progress,

706: Vol.\ 1 (Zakopane-K\'oscielisko, 1997), pp.\ 277--291, de Gruyter,

707: Berlin, 1999.

708: 

709: \bibitem[Lip88]{lipshitz} Lipshitz, Leonard, {\it ``$p$-adic 

710: Zeros of Polynomials,''} J.\ Reine Angew.\ Math.\ {\bf 390} (1988), 

711: pp.\ 208--214. 

712: 

713: \bibitem[Roj99]{jpaa} Rojas, J.\ Maurice, {\it ``Toric

714: Intersection Theory for Affine Root Counting,''} Journal of Pure and

715: Applied Algebra, vol.\ 136, no.\ 1, March, 1999, pp.\ 67--100.

716: 

717: \bibitem[Roj00a]{real} \underline{\hspace{\jmr}}, {\it ``Some Speed-Ups and

718: Speed Limits for Real Algebraic Geometry,''} Journal of Complexity, FoCM 1999

719: special issue, vol.\ 16, no.\ 3 (sept.\ 2000), pp.\ 552--571.

720: 

721: \bibitem[LRW01]{tri} Li, Tien-Yien; Rojas, J.\ Maurice; Wang, Xiaoshen, 

722: {\it ``Descartes' Rule for Trinomials in the Plane and Beyond,''} Math ArXiV 

723: preprint {\tt http://xxx.arXiv.org/abs/math.CO/0008069}, submitted 

724: for publication. 

725: 

726: \bibitem[Smi97]{smirnov} Smirnov, A.\ L., {\it ``Torus Schemes Over a 

727: Discrete Valuation Ring,''}  St.\ Petersburg Math.\ J.\ {\bf 8} (1997), 

728: no.\ 4, pp.\ 651--659.  

729: 

730: \end{thebibliography}

731: 

732: \end{document} 

733: 

734: