1: \documentclass[12pt]{article}

2: \usepackage{amsmath,pifont}

3: %\usepackage{eucal}

4: \usepackage{amssymb}

5: \usepackage{amsthm}

6: \usepackage{psfig}

7: \usepackage[arrow,matrix,tips,2cell]{xy}

8: \usepackage[dvips]{graphics}

9: %\usepackage[notref,notcite]{showkeys}

10: %\usepackage{intlim}

11: \newcommand{\ignore}[1]{\relax}

12: \newcommand{\mnote}{\marginpar}

13: \newcommand{\Aff}{\operatorname{\mathcal Aff}}

14: \newcommand{\Conv}{\operatorname{\mathcal Conv}}

15: \newcommand{\Rat}{\operatorname{\mathcal Rat}}

16: \newcommand{\C}{\mathbb C}

17: \newcommand{\R}{\mathbb R}

18: \newcommand{\Z}{\mathbb Z}

19: \newcommand{\N}{\mathbb N}

20: \newcommand{\NN}{\mathcal N}

21: \newcommand{\Q}{\mathbb Q}

22: \newcommand{\T}{\mathbb T}

23: \newcommand{\U}{\mathcal U}

24: \newcommand{\E}{\mathcal E}

25: \newcommand{\M}{\mathcal M}

26: \newcommand{\bm}{\overline{\mathcal M}}

27: \newcommand{\F}{\mathcal F}

28: \newcommand{\DD}{\mathcal D}

29: \newcommand{\XX}{\mathcal X}

30: \newcommand{\HH}{\mathcal H}

31: \newcommand{\UU}{\mathcal U}

32: \newcommand{\LL}{\mathcal Log}

33: \newcommand{\LLL}{\mathcal L}

34: \newcommand{\s}{\mathcal S}

35: \newcommand{\V}{{V}^{\circ}}

36: \newcommand{\ve}{\stackrel{\to}{v}}

37: \newcommand{\PP}{\mathcal P}

38: \newcommand{\Nn}{\{1,\ldots,n\}}

39: \newcommand{\ind}{\operatorname{ind}}

40: \newcommand{\lk}{\operatorname{lk}}

41: \newcommand{\re}{\operatorname{Re}}

42: \newcommand{\im}{\operatorname{Im}}

43: \newcommand{\const}{\operatorname{const}}

44: \newcommand{\trop}{\operatorname{trop}}

45: \newcommand{\Val}{\operatorname{Val}}

46: \newcommand{\val}{\operatorname{val}}

47: \newcommand{\Arg}{\operatorname{Arg}}

48: \newcommand{\Tau}{\mathcal T}

49: \newcommand{\ntrop}{N^{\operatorname{irr}}_{\operatorname{trop}}}

50: \newcommand{\ntropr}{N_{\operatorname{trop}}}

51: \newcommand{\ntropR}{N^{\operatorname{irr}}_{\operatorname{trop},{\mathbb R}}}

52: \newcommand{\ntropRr}{N_{\operatorname{trop},{\mathbb R}}}

53: \newcommand{\ntropRW}{N^{\operatorname{irr}}_{\operatorname{trop},{\mathbb R},W}}

54: \newcommand{\ntropRWr}{N_{\operatorname{trop},{\mathbb R},W}}

55: \newcommand{\ncl}{N^{\operatorname{irr}}}

56: \newcommand{\nclr}{N}

57: \newcommand{\spec}{\operatorname{Spec}}

58: \newcommand{\ver}{\operatorname{Vert}}

59: \newcommand{\specm}{\operatorname{Spec}_m}

60: \newcommand{\specf}{\operatorname{Spec}_{\mathcal F}}

61: \newcommand{\Alga}{\operatorname{Alga}}

62: 

63: 

64: %\newtheorem{prop}{Proposition}

65: \newtheorem{thm}{Theorem}

66: 

67: 

68: %\newtheorem{thm}{Theorem}

69: \newtheorem{thmquote}{Theorem}

70: \renewcommand{\thethmquote}{\hspace{-8pt}}

71: \newtheorem{lem}{Lemma}[section]

72: \newtheorem{lemmaA}{Lemma A\hspace{-5pt}}

73: \newtheorem{theoremA}{Theorem A\hspace{-5pt}}

74: \newtheorem{claim}{Claim}

75: \renewcommand{\theclaim}{\hspace{-5pt}}

76: \newtheorem{cor}{Corollary}

77: \newtheorem{thma}[lem]{Theorem}

78: %\newtheorem{thm}[lem]{Theorem}

79: \newtheorem{coro}[lem]{Corollary}

80: \newtheorem{prop}[lem]{Proposition}

81: \newtheorem{add}[lem]{Addendum}

82: \theoremstyle{definition}

83: \newtheorem{defn}[lem]{Definition}

84: \newtheorem{defnA}{Definition}

85: \renewcommand{\thedefnA}{\hspace{-5pt}}

86: \newtheorem{exa}[lem]{Example}

87: \newtheorem{que}[lem]{Question}

88: \theoremstyle{remark}

89: \newtheorem{rmk}[lem]{Remark}

90: \newtheorem{rem}[lem]{Remark}

91: 

92: 

93: 

94: %\renewcommand{\theexa}{}

95: \newtheorem{exam}[lem]{Example}

96: \newtheorem{exer}[lem]{Exercise}

97: 

98: %\renewcommand{\thermk}{}

99: \newtheorem{ack}{Acknowledgement}

100: \renewcommand{\theack}{\hspace{-5pt}}

101: 

102: 

103: \newcommand{\Rtr}{{\mathbb R}_{\operatorname{trop}}}

104: \newcommand{\tor}{(\C^\times)^{2}}

105: \newcommand{\tordva}{(\C^*)^{2}}

106: \newcommand{\tortri}{(\C^*)^{3}}

107: \newcommand{\rtordva}{(\R^*)^{2}}

108: \newcommand{\rtortri}{(\R^*)^{3}}

109: \newcommand{\rtor}{(\R^\times)^{2}}

110: \newcommand{\torn}{(\C^*)^{n}}

111: \newcommand{\rtorn}{(\R^*)^{n}}

112: \newcommand{\st}{\operatorname{st}}

113: \newcommand{\ft}{\operatorname{ft}}

114: \newcommand{\ev}{\operatorname{ev}}

115: \newcommand{\sign}{\operatorname{Sign}}

116: \newcommand{\subdiv}{\operatorname{Subdiv}}

117: \newcommand{\irr}{\operatorname{irr}}

118: \newcommand{\mult}{\operatorname{mult}}

119: \newcommand{\degree}{\operatorname{deg}}

120: \newcommand{\conj}{\operatorname{conj}}

121: \newcommand{\perimeter}{\operatorname{Perim}}

122: \newcommand{\dH}{d_{\operatorname{Haus}}}

123: \newcommand{\dd}{\partial}

124: \newcommand{\am}{\mathcal{A}}

125: \newcommand{\z}{\mathcal{Z}}

126: \newcommand{\D}{\mathcal{D}}

127: \newcommand{\cp}{{\mathbb C}{\mathbb P}}

128: \newcommand{\rp}{{\mathbb R}{\mathbb P}}

129: \newcommand{\pp}{{\mathbb P}}

130: \newcommand{\ppp}{{\mathcal P}}

131: \newcommand{\qqq}{{\mathcal R}}

132: \newcommand{\lp}{\Delta\cap{\mathbb Z}^n}

133: \newcommand{\Log}{\operatorname{Log}}

134: \newcommand{\grad}{\operatorname{grad}}

135: \newcommand{\Hess}{\operatorname{Hess}}

136: \newcommand{\Vol}{\operatorname{Vol}}

137: \newcommand{\MA}{\operatorname{M}}

138: %\newcommand{\TR}{\mathbf{T}}

139: \newcommand{\Int}{\operatorname{Int}}

140: \newcommand{\ov}{\operatorname{ov}}

141: \renewcommand{\setminus}{\smallsetminus}

142: \newcommand{\ord}{\operatorname{ord}}

143: \newcommand{\Mgn}{{\mathcal M}_{g,n}}

144: \newcommand{\MM}{{\mathcal M}}

145: \newcommand{\Jac}{\operatorname{Jac}}

146: \newcommand{\Div}{\operatorname{Div}}

147: \newcommand{\Pic}{\operatorname{Pic}according to \cite{Co}}

148: \newcommand{\Sym}{\operatorname{Sym}}

149: 

150: %\renewcommand{\marginpar}[1]{\relax}

151: \newcommand{\Area}{\operatorname{Area}}

152: \newcommand{\area}{\operatorname{Area}}

153: \newcommand{\OO}{\mathcal O}

154: %\newcommand{\red}{\operatorname{red}}

155: \newcommand{\ttor}{(\T^\times)^n}

156: \newcommand{\Funct}{\operatorname{Funct}}

157: 

158: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

159: 

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

161: \newcommand{\cT}{\mathcal T}

162: \newcommand{\sT}{\mathsf T}

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

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

165: \newcommand{\bP}{\overline{P}}

166: \newcommand{\barD}{\overline{\Delta}}

167: \newcommand{\barE}{\overline{E}}

168: \newcommand{\barF}{\overline{F}}

169: 

170: \newcommand{\Pt}{\widetilde{P}}

171: \newcommand{\Rh}{\widehat{R}}

172: \newcommand{\Qt}{\widetilde{Q}}

173: \newcommand{\Rt}{\widetilde{R}}

174: 

175: \newcommand{\cb}{\textup{\ding{114}}}

176: 

177: 

178: 

179: \newcommand{\logf}{\div}

180: \newcommand{\ca}{\circledast}

181: \newcommand{\parl}{\rightrightarrows}

182: \newcommand{\apar}{\rightleftarrows}

183: \newcommand{\pline}{\mathbb{P}^1}

184: \newcommand{\bbP}{\mathbb{P}}

185: \DeclareMathOperator{\Res}{\texttt{resultant}}

186: \DeclareMathOperator{\Env}{\texttt{envelope}}

187: \DeclareMathOperator{\lcm}{lcm}

188: \DeclareMathOperator{\br}{br}

189: 

190: \begin{document}

191: \title

192: {Geometry of planar log-fronts}

193: \author{Grigory Mikhalkin and Andrei Okounkov}

194: \maketitle

195: \section{Introduction}

196: 

197: \subsection{Frozen boundaries}\label{motiv}

198: 

199: Given two polynomials $P(z,w)$ and $Q(z,w)$, one

200: can study solutions of the system

201: %

202: \begin{equation}

203:   \label{PQeq}

204:   \begin{cases}

205:      P(z,w)=0\,, \\

206:   Q(e^{-x} z, e^{-y} w) =0\,.

207:   \end{cases}

208:  \end{equation}

209: %

210: as a function of a point $(x,y)\in \R^2$.  A solution

211: $z(x,y)$ and $w(x,y)$

212: of \eqref{PQeq} solves a first order quasilinear PDE

213: %

214: \begin{equation*}

215:   \label{Burg}

216:   \frac{z_x}{z} + \frac{w_y}{w} = 1\,, \quad P(z,w)=0\,,

217: \end{equation*}

218: %

219: which is closely related to the complex Burgers equation

220: and arises in the theory of random surfaces, see \cite{Burg}.

221: The singularities of $z(x,y)$ and $w(x,y)$ occur when

222: \eqref{PQeq} has a multiple root, that is, when the two

223: curves in \eqref{PQeq} are tangent.

224: In the random surface context,

225: this marks the boundary between order (e.g.\ crystalline

226: facet) and disorder,

227: called \textit{frozen boundary}.

228: 

229: For example, let $P$ and $Q$ define

230: rational curves in $\rp^2$ of degree $2$ and $4$, respectively,

231: positioned with respect to the coordinate axes of $\rp^2$ as illustrated in

232: Figure \ref{PandQ}.

233: %

234: \begin{figure}[!h]

235:   \centering

236:   \scalebox{0.4}{\includegraphics{Pcurve.ps}}

237: \quad \scalebox{0.4}{\includegraphics{Qcurve.ps}}

238:   \caption{Curves $P$ and $Q$ for the frozen boundary in Figure \ref{froz}}

239:   \label{PandQ}

240: \end{figure}

241: %

242: Points where $P$ and $Q$ intersect the axes may be

243: fixed so that the frozen boundary will be inscribed in a hexagon as

244: illustrated in Figure \ref{froz}.

245: %

246: \begin{figure}[!h]

247:   \centering

248:   \rotatebox{90}{\scalebox{0.5}{\includegraphics{2pbpp.ps}}}

249:   \caption{An example of frozen boundary}

250:   \label{froz}

251: \end{figure}

252: %

253: The probabilistic meaning of this curve

254: is the following.

255: 

256: The well-known

257: arctic circle theorem of Cohn, Larsen, and Propp \cite{CLP}

258: states that the limit shape of a typical 3D partition contained

259: in a cube has a frozen boundary which is a

260: circle. More precisely, projected in

261: the $(1,1,1)$-direction, the cube becomes a hexagon and

262: the $(1,1,1)$-projection of the frozen boundary is an inscribed

263: circle. Now suppose we additionally weight each 3D partition $\pi$ by a weight

264: which a product over all boxes

265: $$

266: \cb=(i,j,k)\in \Z^3

267: $$

268: in $\pi$ of some periodic

269: function of $i-j$ and $j-k$. If the period equals $2$, the

270: frozen boundary will look as in Figure \ref{froz}. The coefficients

271: of the curve $P$ are obtained from the periodic weights, while

272: the coefficients of $Q$ are fixed by boundary conditions.

273: See \cite{Burg} for a detailed discussion of this procedure in general.

274: 

275: The geometry and, especially, the

276: singularities of frozen boundaries are of considerable

277: interest. Note that these are curves of some complexity:

278: in our current example frozen boundary has degree $28$

279: and genus $27$.

280: 

281: 

282: 

283: 

284: 

285: \subsection{Log-front}

286: 

287: 

288: This motivates the following definition.

289: Given two curves $P,Q\subset\tor$, consider

290: %

291: \begin{equation} \label{defR}

292: R =\left\{(a,b) \, \big| \,

293: \textup{$\tau_{a,b}\cdot Q$ is tangent to $P$}

294: \right\} \subset \tor

295: \end{equation}

296: %

297: where

298: $$

299: \tau_{a,b}: \tor\to\tor

300: $$

301: it the dilation $(z,w)\mapsto (az,bw)$ and

302: tangency at a singular point  means that for

303: some branches of the two curves in question

304: their unique tangent lines coincide.

305: For simplicity, we assume that

306: tangency occurs only at isolated points, that is,

307: no component of $P$ is a dilate of a component of $Q$.

308: 

309: 

310: We call

311: the curve $R$ the \emph{log-front} of $P$ and $Q$.

312: The reason for such name will be explained below.

313: We will denote the construction \eqref{defR}

314: by

315: $$

316: R=P\logf Q\,.

317: $$

318: The obvious property

319: $$

320: Q\logf P = R(a^{-1},b^{-1})

321: $$

322: and \eqref{double_dual} below justify such notation.

323: 

324: Clearly, $R$ is an algebraic curve.

325: Various ways to compute its equation

326: $R(a,b)=0$ will be discussed in Section \ref{sRhat}.

327: The frozen boundary from Section \ref{motiv} is

328: given by $R(e^x,e^y)=0$.

329: 

330: 

331: 

332: \subsection{Symmetry between $Q$ and $R$}

333: The definition of $R$ can be recast into several

334: equivalent forms.

335: Let $\cP$ denote the hypersurface

336: $$

337: \cP=\left\{P(az,bw)=0\right\} \in \tor \times \tor

338: $$

339: and let $\pi_{ab}$ and $\pi_{zw}$ be the projections

340: from $\tor\times \tor$ to the respective factors.

341: 

342: The tangency in definition \eqref{defR} can be

343: rephrased by saying that $R$ is formed by critical

344: values of the map

345: %

346: \begin{equation}

347:   \label{cS}

348:     \pi_{ab}: \quad\cS=\cP \cap \pi_{xy}^{-1} \left(Q\right) \to \tor

349: \end{equation}

350: %

351: of complex surfaces.

352: In other words, the surface

353: $$

354: \pi_{ab}^{-1} \left(R\right) \cap \pi_{xy}^{-1} \left(Q\right)

355: \cong R \times Q

356: $$

357: is tangent to the hypersurface $\cP$ along their intersection, showing

358: a certain symmetry

359: between the roles of $Q$ and $R$ for fixed $P$. In

360: particular

361: %

362: \begin{equation}

363:   \label{double_dual}

364:  Q \subset P \logf(P \logf Q)\,.

365: \end{equation}

366: %

367: The multiplicity, with which $Q$ occurs in the

368: right-hand side of \eqref{double_dual} equals

369: the degree of the

370: \emph{logarithmic Gau\ss\ map} of $P$, see below.

371: 

372: Yet another way to say the same thing is

373: that $R$ is the envelope of the family

374: of curves

375: %

376: \begin{equation}

377:   \label{envelope}

378:      \big\{P(az,bw)=0\big\}_{Q(z,w)=0}\,,

379: \end{equation}

380: %

381: indexed by points of $Q$

382: 

383: 

384: \subsection{Classical constructions}

385: 

386: Among examples of the operation \eqref{defR}

387:  there are the

388: following two classical constructions.

389: 

390: First, let $P$ be a general line, for example,

391: $$

392: P(z,w) = z + w + 1 \,.

393: $$

394: Then its dilates $\tau_{ab}\cdot P$ form an open

395: set of the dual projective plane and hence $R$ is

396: an open set of the dual curve $Q^\vee$, namely, its

397: intersection with $\tor$. In this

398: case, \eqref{double_dual} becomes an equality.

399: 

400: 

401: Second, the additive analog of \eqref{defR}

402: %

403: \begin{equation} \label{defadd}

404: P - Q =\left\{(a,b) \, \big| \,

405: \textup{$Q(z-a,w-b)=0$ is tangent to $P$}

406: \right\} \subset \C^2

407: \end{equation}

408: %

409: is a limit case of \eqref{defR}. For

410: analytic plane curves, constructions $P\logf Q$ and \eqref{defadd}

411: are, in fact, equivalent by taking the logarithms.

412: When $P$ is

413: a circle of radius $t$

414: $$

415: P(z,w) = z^2 + w^2 - t^2

416: $$

417: the real locus of $P-Q$ contains

418: %by \eqref{envelope}

419: the front at time $t$ of a wave that was emitted at time zero from all

420: real points of $Q$ and is propagating with unit

421: velocity. Such curve is called a \emph{wave-front}.

422: It is this example that motivates

423: the general term \emph{log-front}.

424: 

425: Pl\"ucker formulas relate singularities of a curve $Q$

426: to the singularities of its dual $Q^\vee$. A formula

427: of Klein further constraints the singularities of the

428: real locus of $Q^\vee$. Analogous formulas for wave

429: fronts were obtained by O.~Viro in \cite{V-usp}. The

430: goal of this note is prove an analog of Pl\"ucker

431: and Klein formulas in the general case.

432: 

433: In probabilistic applications, the curve $P$ is

434: a real algebraic curve of a very special kind, namely,

435: it is a \emph{Harnack curves}. See \cite{Mi-obzor} and the Appendix

436: for a discussion of the properties of Harnack curves

437: and \cite{KOS,KO1} for connections with probability. Harnack

438: curves have many remarkable features that general real

439: plane curves lack. As it turns out, the assumption that

440: $P$ is Harnack is also essential for our

441: derivation of Klein-type formulas

442: for $R$.

443: 

444: It may

445: be noted here that both our formulas and their

446: proofs involve nothing but elementary geometry of

447: plane curves and would have been, no doubt,

448: obtained by Pl\"ucker and Klein had their seen

449: a need for them.

450: 

451: 

452: \subsection{Acknowledgments}

453: 

454: We are grateful to C.~Faber, R.~Kenyon and O.~Viro

455: for many useful discussions. Some of our formulas generalize

456: unpublished results obtained jointly

457: with C.~Faber.%, see \cite{FO}.

458: 

459: \section{Preliminaries}

460: 

461: There are many good books on plane algebraic

462: curves, see e.g. \cite{B,W}. With the probability audience

463: in mind, we collected in this section an explanation

464: of some basic

465: notions that will be used later.

466: 

467: \subsection{Newton polygons}\label{torsurf}

468: 

469: Let $P\subset\tor$ be a curve with the equation

470: $$

471: P(z,w)=\sum\limits_{i,j}p_{ij}\, z^i \, w^j=0\,.

472: $$

473: By definition, its \emph{Newton polygon} is

474: $$

475: \Delta_P=\operatorname{Convex\ Hull}

476: \left\{(i,j)\in\Z^2 \,\big|\, p_{ij}\neq 0\right\}\,.

477: $$

478: This is a refinement of the degree of $P$.

479: The polygon $\Delta_P$ is defined only  up to

480: translation by a lattice vector once we treat $P\subset\tor$

481: as a geometric curve rather than a polynomial. Let

482: $|\Delta_P|$ denote the number of lattice points in

483: $\Delta_P$.

484: 

485: The Newton polygon  defines a projective toric surface

486: \begin{equation}\label{cT}

487: \cT=\overline{\Big\{\big(z^i w^j\big)_{(i,j)\in \Delta_P}\Big\}}_{(z,w)\in \tor}

488: \subset \cp^{|\Delta_P|-1}

489: \end{equation}

490: of degree

491: $$

492: \deg \cT = 2 \Area(\Delta_P)\,.

493: $$

494: {}From now on, $P$ will denote the closed curve in

495: $\cT$ defined by $P(z,w)=0$. This is a

496: hyperplane section avoiding the torus

497: fixed points, which are the only possible singularities

498: of $\cT$. We call the components $\mathcal{D}_E$ of

499: $$

500: \partial \cT = \cT \setminus \tor = \bigcup_{\textup{edges $E$ of $\Delta_P$}} \mathcal{D}_E

501: $$

502: the \emph{boundary divisors} and the points of

503: $$

504: \partial P = P \cap \partial \cT

505: $$

506: the \emph{boundary points} of $P$. For simplicity

507: we assume throughout the paper that the boundary

508: points of both $P$ and $Q$ are smooth.

509: 

510: By construction,

511: $P$ intersects $\mathcal{D}_E$  in $|E|$ points

512: counting multiplicity, where $|E|$ is the length

513: of $E$ in lattice units. The multiplicities of these

514: intersection points define a partition $\lambda(E)$

515: of $|E|$ for every edge $E$ of $\Delta_P$. We will

516: call a lattice polygon with such additional

517: partition data a \emph{marked polygon}.

518: 

519: 

520: \subsection{Amoebas}

521: 

522: The image $\Log(P)$ of a curve $P$ under the map

523: $$

524: \Log((z,w)) = (\log|z|,\log|w|)

525: $$

526: is called the \emph{amoeba} of $P$. The boundary

527: points of $P$ give rise to the so-called

528: \emph{tentacles} of the amoeba $\Log(P)$, see

529: for example Figure \ref{fig_tang_amoeb} which shows

530: logarithmic images $\Log(\R P)$ of the real

531: loci of two plane curves.

532: 

533: The directions of

534: the tentacles are the outward normals to the

535: sides of the Newton polygon. For example,

536: Newton polygons of curves from

537: Figure \ref{fig_tang_amoeb} are plotted in

538: Figure \ref{fig_Polygons}.

539: %

540: \begin{figure}[!htbp]

541:   \centering

542:   \scalebox{0.33}{\includegraphics{Newton1.ps}}

543:   \quad

544:   \scalebox{0.33}{\includegraphics{Newton2.ps}}

545:   \caption{Newtons polygons of curves in Figure \ref{fig_tang_amoeb}}

546:   \label{fig_Polygons}

547: \end{figure}

548: %

549: The number of

550: tentacles in a given direction is the number

551: of points of $P$ on the corresponding boundary

552: divisor.

553: 

554: 

555: 

556: 

557: \subsection{Logarithmic Gau\ss\ map}\label{subsectionGauss}

558: 

559: Since $\tor$ is an abelian group, the tangent spaces at

560: all points of $\tor$ are canonically identified with

561: tangent space $T_{(1,1)}$ at the identity $(1,1)\in\tor$. Mapping

562: the tangents to $P$ to their images in $\bbP(T_{(1,1)})$

563: gives the \emph{logarithmic Gau\ss\ map}

564: $$

565: \gamma_P: \Pt \to \pline\,,

566: $$

567: where $\Pt$ is the normalization of $P$. In coordinates,

568: $$

569: \gamma_P((z,w)) = \frac{z \frac{\partial}{\partial z} P}

570: {w \frac{\partial}{\partial w} P}\,.

571: $$

572: Note that the values of $\gamma_P$ at the boundary points of $P$

573: are determined by the slopes of the corresponding edges of $\Delta_P$.

574: 

575: The degree of  $\gamma_P$ can be computed as follows,

576: cf.\ \cite{Ka,Mi-obzor}.

577: Recall that the \emph{multiplicity} $m(p)$ of a point $p\in P$ is

578: the multiplicity with which $P$ intersects a generic

579: line through $p$. Another important characteristic of

580: a singular point is its \emph{Milnor number} $\mu(p)$, see \cite{Milnor}.

581: It may be defined as

582: the local intersection number of $\frac{\partial}{\partial z} P=0$

583: and $\frac{\partial}{\partial w} P=0$ at $p$.

584: 

585: 

586: 

587: \begin{prop}\label{dGauss}

588: We have

589: %

590: \begin{equation}

591:   \label{deg_gam}

592:      \deg \gamma_P = 2 |\Delta^\circ_P | + |\partial P| - 2 -

593: \sum_{p\in P}(\mu(p)+m(p)-1)\,.

594: \end{equation}

595: %

596: where $|\Delta^\circ_P |$ is the number of points in the

597: interior of the Newton polygon $\Delta_P$ and  $|\partial P|$

598: is the number of boundary points of $P$ not counting

599: multiplicity.

600: \end{prop}

601: 

602: Clearly, only

603: singular points of $P$ contribute to the sum over $p\in P$

604: in \eqref{deg_gam} since for a nonsingular point

605: $p$ we have $m(p)=1$ and $\mu(p)=0$.

606: 

607: \begin{proof}

608: If $P$ is smooth and transverse to

609: the boundary

610: then the proposition follows from

611: Kouchnirenko's formula \cite{Ku} (also using

612: Pick's formula for the area of a lattice polygon).

613: 

614: A singular point of $P$ subtracts $\mu(p)+m(p)-1$,

615: which is

616: the same computation as for the ordinary Gau\ss\ map,

617: see e.g.\ Theorem 7.2.2.\ in \cite{W}. The

618: effect of tangency to the boundary is determined similarly,

619: see e.g.\ Section \ref{proofPolygon} below.

620: \end{proof}

621: 

622: 

623: 

624: The critical points of $\gamma_P$ are known as the

625: logarithmic inflection points. Note that, for example,

626: a cusp is not a logarithmic inflection point, but a

627: sharp, or ramphoid, cusp (locally looking like $x^2=y^5$) is.

628: By the Riemann-Hurwitz

629: formula applied to $\gamma_P$, the count of

630: logarithmic inflection points with multiplicity is

631: %

632: \begin{equation}\label{logFlex}

633: \textup{\# logarithmic inflection points}=2\deg \gamma_P - \chi(\Pt).

634: \end{equation}

635: 

636: 

637: \subsection{Geometric genus and adjunction}

638: 

639: The geometric genus of $P$ is defined as the genus

640: $g(\Pt)$ of its normalization $\Pt$. It may

641: be computed as follows, cf. \cite{Kh},

642: %

643: \begin{equation}

644:   \label{adju}

645:   g(\Pt) = |\Delta^\circ_P | - \frac12 \sum_{p\in P} (\mu(p)+ \beta(p) -1)

646: \end{equation}

647: %

648: where $\beta(p)$ is the number of branches of $P$ through $p$.

649: %Thus we have

650: This is known as the adjunction formula.

651: 

652: The number

653: $\frac12 (\mu(p)+ \beta(p) -1)$ is a nonnegative integer which

654: vanishes if $p$ is a smooth point of $P$. It equals $1$

655: for cusps and nodes, so if the curve has only those

656: singularities, we get

657: %

658: \begin{equation}

659: \textup{\# cusps} +

660: \textup{\# nodes}= |\Delta^\circ_P | - g(\Pt) \label{n_nodes}\,.

661: \end{equation}

662: %

663: Combining \eqref{deg_gam} and \eqref{adju} we get

664: %

665: \begin{equation}

666: \sum_{p\in P} (m(p)-\beta(p))=

667: - \deg \gamma_P - \chi(\Pt\setminus \partial P)\label{n_cusps}\,.

668: \end{equation}

669: %

670: 

671: \begin{rmk}

672: A geometrically-minded reader might appreciate the following

673: alternative proof (or rather a rephrasing of the proof)

674: of \ref{n_cusps}.

675: 

676: Suppose that $\Pt\setminus \partial P$ is smooth.

677: We may deduce the equation

678: \begin{equation}\label{eq-smooth}

679: \deg \gamma_P=- \chi(\Pt\setminus \partial P)

680: \end{equation}

681: by the following application of the maximum principle, cf. \cite{Mi}.

682: Let $\lambda:\R^2\to\R$ be a linear map which we may choose

683: with a generic slope of the kernel.

684: The function $\lambda\circ\Log$ is a pluriharmonic function on $\tor$

685: and thus restricts to a harmonic function $h$

686: on $\deg \gamma_P=- \chi(\Pt\setminus \partial P)$.

687: Thus all critical points of $h$ are of index 1.

688: 

689: Furthermore, $h$ exhibits $P$ as a cobordism between $P_-\subset\dd P$

690: and $P_+\subset\dd P$, where the splitting $\dd P=P_+\cup P_-$

691: corresponds to $h^{-1}(+\infty)$ and $h^{-1}(-\infty)$.

692: Thus, the number of critical points of $h$ equals $\chi(P-P_+,P_-)=\chi(P\setminus\dd P)$

693: (the latter equality follows from additivity of Euler characteristic).

694: On the other hand the number equals $\deg\gamma_P$, so we get \eqref{eq-smooth}.

695: Each singularity $p$ of $P_-\subset\dd P$ subtracts $\sum(\mu(P)-\beta(p)-1)$

696: from $-\chi(P)$ and $\sum(\mu(P)-m(p)-1)$ from $\deg\gamma_P$.

697: \end{rmk}

698: 

699: \subsection{Nodal and cuspidal numbers of a plane curve}\label{bcdefn}

700: The right-hand side of \eqref{n_cusps}

701: may be interpreted as the total

702: \emph{cuspidal number} $c(P)$ of $P$. It equals the

703: number of cusps if $P$ has only cusps and

704: ordinary multiple points. An example of

705: formula \eqref{n_cusps} may be seen in Figure

706: \ref{E6}. In the bottom half of Figure \ref{E6},

707: we see two inflection point come together to

708: form a point of the form $y=x^4$. For the

709: dual curves, which have the same genus and the same

710: degree of the

711: logarithmic Gau\ss\ map, we see a merger of

712: cusps into a $E_6$ singularity $y^3=x^4$ which

713: has the cuspidal number equal to 2.

714: \begin{figure}[!htbp]

715:   \centering

716:   \scalebox{0.4}{\includegraphics{E6.eps}}

717:   \caption{Two cusps coalescing in a $E_6$ singularity.}

718:   \label{E6}

719: \end{figure}

720: %

721: 

722: Another useful number is the {\em nodal number} $b(P)$.

723: %Let $s\in \bar{A}$ be a singular point and $B_1,\dots,B_l$

724: %be the branches of $\bar{A}$ that pass through $s$.

725: %Denote with $\#(B_j,B_k)$ the local intersection number

726: %of the branches $B_j$ and $B_k$ at $s$.

727: First we define the {\em local nodal number}

728: {$b_s(P)$} at every singular point $s\in P$

729: as the sum $\sum\limits_{j\neq k}\#(B_j,B_k)$ of the intersection

730: numbers over all distinct pairs of branches $B_j$, $B_k$ via $s$.

731: Clearly, $b_s(P)=0$ if and only if $s$ is a locally irreducible singular

732: point of $P$. If $s$ is an ordinary node of $P$

733: %(i.e. an $A_1$-point)

734: then $b_s(P)=1$.

735: {\em The nodal number $b(P)$ is the sum of the local nodal numbers

736: over all singular points of $P$.}

737: 

738: In many applications the numbers $c(P)$ and $b(P)$ play the r\^ole

739: of the total numbers of cusps and nodes even if $P$ has higher singularities.

740: 

741: 

742: \subsection{Integration w.r.t.\ Euler characteristic}\label{Viro-calculus}

743: 

744: Because of the equation

745: $$

746: \chi(A \cap B) = \chi(A) + \chi(B) - \chi(A\cap B)

747: $$

748: the Euler characteristic may be viewed as a signed

749: finite-additive measure. If $f$ is a function on

750: $X$ taking finite many values, its integral

751: with respect to the Euler characteristic is

752: defined by

753: $$

754: \int_X f\, d\chi  = \sum_{y}  y \, \chi(f^{-1}(y)) \,.

755: $$

756: Calculus of such integrals was developed by O.~Viro

757: \cite{V}. Applications of this calculus to the classical

758: Pl\"ucker and Klein formulas may also be found in \cite{V}.

759: 

760: More generally, given a map

761: $$

762: F: X \to Y

763: $$

764: one may push-forward $F_*f$ of  $f$ by the formula

765: $$

766: \left[F_*f\right](y) = \int_{F^{-1}(y)} f \, d\chi \,.

767: $$

768: Under additional hypotheses on $F$ and $f$, this operation has

769: natural functorial properties like $(FG)_*=F_* G_*$, see e.g.\

770: Section 7.3 in \cite{W}. The case of main importance for

771: us will be when $F$ is a nonconstant map of smooth curves,

772: in which case,

773: $$

774: F_*1 = \deg F - \br F \,,

775: $$

776: where $\br F$ is the \emph{branch divisor} of $F$, that is,

777: the sum of all critical points of $F$ with their

778: multiplicities.

779: 

780: 

781: \section{The equation of $R$}

782: 

783: 

784: \subsection{$R$ as a Resultant}\label{sRhat}

785: 

786: 

787: 

788: 

789: Consider the curve $\cR$ defined by the equations

790: %

791: \begin{equation}

792:   \label{eqRh}

793:      P(az,bw)=Q(z,w)=

794: \det

795: \begin{pmatrix}

796: \frac{\partial}{\partial z} P(az,bw) &  \frac{\partial}{\partial z} Q

797: \vspace{2 mm}\\

798: \frac{\partial}{\partial w} P(az,bw) &  \frac{\partial}{\partial w} Q

799: \end{pmatrix} = 0 \,.

800: \end{equation}

801: %

802: in $\tor\times\tor$. We have $\cR\subset\cS$, where

803: the surface $\cS$ was defined in \eqref{cS}. In general,

804: $\cR$ may have several

805: components. The Wronskian in \eqref{eqRh} vanishes at

806: any singular point of $q$ of $Q$, hence such a point contributes

807: a translate of  $P$ to $\cR$. The multiplicity of

808: this component equals $m(q)+\mu(q)-1$, see the proof of

809: Proposition \ref{dGauss}. Symmetrically, singular points of

810: $P$ contribute copies of $Q$ to $\cR$. The remaining

811: components correspond to actual tangency. The curve $R$

812: is the projection of these components to the $(a,b)$-plane.

813: 

814: 

815: 

816: The equation of $R$ may be found by eliminating

817: $z$ and $w$ from \eqref{eqRh}, followed by factoring off

818: spurious components caused by singularities. Gr\"obner basis

819: algorithms give one way to perform this elimination.

820: We find that in practice it is easier and faster to

821: use resultants for this computation.

822: 

823: Consider the polynomial

824: $$

825: R_1(a,b,z)=\textup{resultant}_w\big(P(az,bw),Q(z,w)\big) \,.

826: $$

827: The equation $R_1(a,b,z)=0$ cuts out the image $\cS_1$ of

828: $\cS$ under the projection along the $z$ direction.

829: This projection may create singularities, specifically

830: $\cS_1$ may intersect itself along curves. Consider

831: $$

832: R_2(a,b)=\textup{resultant}_z\left(R_1,

833: \frac{\partial}{\partial z} R_1\right) \,.

834: $$

835: Its zero locus consists of critical points of the

836: projection of $\cS_1$ along the $z$ direction, together

837: with contributions of singularities. Those can be

838: identified and removed by factoring $R_2$ as they

839: all occur with multiplicity greater than one. For

840: example, the image of double point curves of $\cS_1$

841: will occur in $R_2$ with multiplicity two.

842: 

843: Exact factorization is possible and rather effective

844: for polynomials with rational coefficients. In

845: probabilistic applications, the coefficients may

846: be known only approximately, making exact

847: factorization impossible. This is why it is useful

848: to know the Newton polygon of $R$, which will be

849: determined below. Also note that sometimes, albeit

850: rarely, components of $R$ may occur with

851: multiplicity in $R_2$ as, for example, in \eqref{double_dual}.

852: 

853: 

854: \subsection{Multiplicity of tangency}

855: 

856: Given two curves $C$ and $C'$ and a point $p$

857: define their tangency multiplicity at $p$ as

858: $$

859: (C \curlywedge C')_p = (C \cdot C')_p - m_{C}(p)\,  m_{C'}(p)

860: $$

861: where $(C \cdot C')_p$ is the local intersection

862: multiplicity of $C$ and $C'$ at $p$ and $m_C(p)$ is

863: the multiplicity of the point $p$ on $C$. Note that

864: $(C \curlywedge C')_p =0$ if $p\notin C$ or if $C$ and $C'$

865: intersect transversely at $p$. Also note that the

866: tangency multiplicity is additive over branches of both

867: $C$ and $C'$.

868: 

869: We have the following

870: 

871: \begin{prop} The multiplicity of a point $(a,b)\in R$ is

872: the total tangency multiplicity of $P(az,bw)=0$ and

873: $Q(z,w)=0$, assuming they are not tangent at

874: infinity.

875: \end{prop}

876: 

877: In other words, the local cuspidal number at each branch

878: of a singular point of $R$ is equal to the corresponding tangency

879: multiplicity minus one.

880: 

881: \begin{proof}

882: Since multiplicity is a local quantity, we can work with

883: the additive version \eqref{defadd}.

884: We may work with each branch of the singularity individually.

885: Suppose $P$ and $Q$

886: are tangent at $z=w=0$ and that $w=0$ is their

887: common tangent. In this case the origin will belong to

888: $P-Q$ and $b=0$ will be the corresponding tangent.

889: We need to compute the total multiplicity $m_R$ of all

890: branches of $R$ corresponding to this tangency. It

891: equals the number of

892: roots $b_i$ of $R(\epsilon,b)=0$ with $b_i\approx 0$

893: where $\epsilon$ is a fixed small nonzero number. For $b=b_i$

894: the curves

895: %

896: \begin{equation}

897:   \label{PQeps}

898: P(z+\epsilon,w+b)=0\,, \quad  Q(z,w)=0\,,

899: \end{equation}

900: %

901: are tangent.

902: 

903: Consider the intersection $U$ of the curve defined by

904: \eqref{PQeps} in the $(z,w,b)$-space with a neighborhood

905: of the origin. We will compute the Euler characteristic

906: of $U$ in two different ways. Viewing $U$ as a degree

907: $m_{P}\,  m_{Q}$ branched

908: covering of the $z$ line, we get

909: $$

910: \chi(U) = - m_{P}\,  m_{Q} + m_{P} +  m_{Q} \,,

911: $$

912: where $m_P$ and $m_Q$ is the multiplicity of

913: the origin on $P$ and $Q$, respectively.

914: 

915: On the other hand, we may view $U$ as a branched

916: covering of the $b$-line. It has degree $(P\cdot Q)_0$,

917: where $(P\cdot Q)_0$ is the intersection multiplicity of

918: $P$ and $Q$ at the origin. We have

919: $$

920: \chi(U) = (P\cdot Q)_0 - m_P (m_Q-1) - m_Q (m_P-1) - m_R

921: $$

922: because each tangency in \eqref{PQeps} corresponds to

923: a simple branchpoint. This concludes proof.

924: \end{proof}

925: 

926: 

927: 

928: 

929: 

930: 

931: 

932: 

933: \section{Newton polygon of log-front}

934: 

935: 

936: \subsection{Formula for $\Delta_R$}

937: 

938: Let $\Delta_P$ and $\Delta_Q$ denote the marked Newton

939: polytopes of $P$ and $Q$. We denote by $-\Delta_Q$ the

940: reflection of $\Delta_Q$ about the origin, with the

941: corresponding marking.

942: 

943: 

944: We say that two edges $E$ and $F$ are opposite

945: and

946: write $E\apar F$, if their outward normals point in the opposite

947: direction. As we will see, in the simplest case when

948: no edge of $-\Delta_Q$ is opposite to an edge of $\Delta_P$,

949: the Newton polygon $\Delta_R$

950: is simply the Minkowski sum

951: %

952: \begin{equation}

953:   \label{Minks}

954:     \deg(\gamma_Q) \cdot \Delta_P  + \deg(\gamma_P) \cdot (-\Delta_Q)\,.

955: \end{equation}

956: %

957: Every pair of opposite edges

958: $E\subset \Delta_P$ and $F\subset -\Delta_Q$

959: makes the polygon $\Delta_R$ smaller than \eqref{Minks}.

960: 

961: Given two partitions $\lambda$ and $\mu$, we define

962: %

963: \begin{align}

964: \langle \lambda,\mu \rangle  & = \sum_i \lambda'_i \, \mu'_i

965: \label{lan1} \\

966: & = \sum_{i,j} \min(\lambda_i,\mu_j) \label{lan2} \,,

967: \end{align}

968: %

969: where $\lambda'$ denotes the conjugate partition.

970: The equivalence of \eqref{lan1} and \eqref{lan2} is

971: an elementary combinatorial fact.

972: 

973: Given an edge $E$, let

974: $$

975: \mathbf{n}_E = \frac{E}{|E|}

976: $$

977: is the unit (non-oriented) lattice segment in the direction of $E$.

978: Note that (similarly to Newton polygons) such lattice segments

979: are defined only up to translations by $\Z^2$.

980: 

981: \begin{thm}\label{thmNewton}

982: The Newton polygon of the log-front $R$ is

983: the unique polygon $\Delta_R$ satisfying

984: %

985: \begin{equation}

986:   \label{polRes}

987: \Delta_R + \sum_{E\apar F}

988: \langle \lambda(E),\lambda(F) \rangle \, \mathbf{n}_E=

989: \deg(\gamma_Q) \cdot \Delta_P  + \deg(\gamma_P) \cdot (-\Delta_Q)\,,

990: \end{equation}

991: %

992: where the summation is over all pairs $E\subset \Delta_P$ and

993: $F\subset  -\Delta_Q$

994: of opposite edges and  $\lambda(E)$ are the corresponding partition

995: markings, i.e.\ multiplicities of intersection with the boundary.

996: \end{thm}

997: 

998: 

999: Minkowski ``subtraction'' of segments, implicit in formula \eqref{polRes},

1000: simply reduces the length of all edges in that direction by the

1001: given amount.

1002: 

1003: 

1004: 

1005: \subsection{Proof}\label{proofPolygon}

1006: 

1007: By definition, a boundary point of $R$ is a tangency of

1008: the curves $P$ and $Q$, dilated with respect to one

1009: another by an infinite amount. In logarithmic coordinates,

1010: infinite dilation becomes an infinite shift. Shifted

1011: by an infinite amount, the amoeba $\Log(P)$ of $P$

1012: is either a union of parallel lines (tentacles) or

1013: empty, in which case there will not be any

1014: boundary points of $R$ in this direction. This is

1015: illustrated in Figure \ref{fig_tang_amoeb}.

1016: 

1017: \begin{figure}[!htbp]

1018:   \centering

1019:   \scalebox{0.5}{\includegraphics{tangent_amoebas.ps}}

1020:   \caption{Tangency to a tentacle creates a boundary

1021: point of a log-front}

1022:   \label{fig_tang_amoeb}

1023: \end{figure}

1024: 

1025: Tangency to a tentacle occurs at points mapped by

1026: the logarithmic Gau\ss\ map to the slope of the tentacle.

1027: Therefore, if $\Log(P)$ and $\Log(Q)$ do not have

1028: tentacles in the same direction (that is,

1029: $\Delta_P$ and $\Delta_Q$ have no parallel edges) then

1030: each tentacle of $\Log(P)$ contributes $\deg \gamma_Q$

1031: points to $\partial R$. The multiplicity of the resulting point

1032: in $\partial R$ equals

1033: the multiplicity of the corresponding point of $\partial P$.

1034: 

1035: The effect of a tentacle-to-tentacle tangency may be

1036: studied in a local model, for example,

1037: $$

1038: P=\{w-(z-1)^k\}\,, \quad Q=\{w-(z-1)^l\}\,.

1039: $$

1040: In this case,

1041: $$

1042: \gamma_P(z)\sim\frac{k}{z-1}\,, \quad

1043: \gamma_Q(z)\sim\frac{l}{z-1}\,, \quad z\to 1 \,,

1044: $$

1045: Equating $\gamma_P(az)=\gamma_Q(z)$ gives a parametrization

1046: of $R$ such that

1047: %

1048: \begin{equation}

1049:   \label{ttt}

1050:   a \sim 1, \quad b\propto (z-1)^{k-l}\,, \quad

1051:  z\to 1 \,.

1052: \end{equation}

1053: %

1054: This means that tentacles of $P$ and $Q$ pointing in the

1055: opposite direction (which means that $k$ and $l$ have opposite

1056: sign) produce a tentacle of combined multiplicity

1057: $|k|+|l|$ pointing in the same direction as

1058: the $P$-tentacle. Tentacles pointing in the same direction,

1059: by contrast,  produce a tentacle of multiplicity

1060: $|k-l|$ pointing in one of the two directions.

1061: This rule can be phrased more naturally in terms of the

1062: reflected Newton polygon $-\Delta_Q$ because the

1063: reflection flips the direction of tentacles.

1064: 

1065: The polygons $\Delta_P$ and $-\Delta_Q$ can, in total,

1066: have as many as 4 edges with the same slope.

1067: Let $E,\barE\subset\Delta_P$

1068: and $F,\barF\subset-\Delta_Q$ be such a 4-tuple of

1069: edges and let $\lambda(E),\dots$ denote the

1070: partition marking.

1071: Assume that the outward normals to $E,F$ point

1072: in the same direction. By our computations, the polygon

1073: $\Delta_R$ will have an edge in the same direction as $E$ and $F$

1074: of length

1075: %

1076: \begin{gather}

1077: |E| (\deg \gamma_Q - \ell(\lambda(F)) - \ell(\lambda(\barF))) +

1078: |F| (\deg \gamma_P - \ell(\lambda(E)) - \ell(\lambda(\barE)))

1079: \notag \\

1080: +\sum_{i,j} \left(\lambda(E)_i +\lambda(F)_j\right) +   \label{|G|}\\

1081: \sum_{i,j} \max(\lambda(E)_i-\lambda(\barF)_j,0) +

1082: \sum_{i,j} \max(\lambda(F)_i-\lambda(\barE)_j,0) \notag \,.

1083: \end{gather}

1084: %

1085: Note that the middle line in \eqref{|G|} cancels with a part of

1086: the first line. A further cancellation is obtained from

1087: $$

1088: - |E| \, \ell(\lambda(\barF)) +

1089: \sum_{i,j} \max(\lambda(E)_i-\lambda(\barF)_j,0) =

1090: - \sum_{i,j} \min(\lambda(E)_i,\lambda(\barF)_j)

1091: $$

1092: using formula \eqref{lan2}. This concludes proof.

1093: 

1094: Counting the boundary points of $R$ without respect to the

1095: corresponding direction $\mathbf{n}_E$ gives us the following corollary.

1096: 

1097: \begin{coro}

1098: The cardinality of $|\dd R|$ equals to

1099: $$\deg\gamma_P|\dd\Delta_Q|+\gamma_Q|\dd\Delta_P|-\sum_{E\apar F}

1100: \langle \lambda(E),\lambda(F) \rangle.$$

1101: %Recall that $|\dd\Delta_P|$ is the number of lattice points on the boundary

1102: %of $\Delta_P$.

1103: \end{coro}

1104: 

1105: \subsection{Examples}\label{expl_poly}

1106: 

1107: Let $P$ and $Q$ be generic with Newton polygons

1108: from Figure \ref{fig_Polygons}. In this case,

1109: the Newton polygon $\Delta_R$ is the larger of the

1110: two polygons plotted in Figure \ref{fig_newton_front}.

1111: If $P$ develops a tangency to the boundary then

1112: $\Delta_R$ shrinks and becomes the smaller polygon

1113: in Figure \ref{fig_newton_front}. What happens

1114: in this case, is that $R$ becomes reducible with

1115: one component being a boundary divisor. The boundary divisor

1116: corresponds to tangency with $P$ at the newly

1117: developed point of tangency to infinity.

1118: 

1119: 

1120: \begin{figure}[!htbp]

1121:   \centering

1122:   \scalebox{0.4}{\includegraphics{Newton3.ps}}

1123:   \caption{Possible $\Delta_R$ for $\Delta_P$ and $\Delta_Q$

1124: from Figure \ref{fig_Polygons}.}

1125:   \label{fig_newton_front}

1126: \end{figure}

1127: 

1128: As another example, consider the case when $P$ and $Q$

1129: are generic curves of degree $d_P$ and $d_Q$.

1130: In this case, both $\Delta_P$ and $\Delta_Q$ are triangles

1131: with sides of length $d_P$ and $d_Q$, while

1132: $\Delta_R$ is a hexagon with sides of length

1133: $$

1134:   d_P d_Q (d_Q-1), d_Q d_P (d_P -1), \dots\,,

1135: $$

1136: cyclically repeated. In particular, when $d_P=1$ we get

1137: a triangle with side $d_Q (d_Q-1)$, reproducing

1138: the very classical formula

1139: for the degree of dual curve.

1140: 

1141: In the example in Figure \ref{froz}, the degrees of

1142: logarithmic Gauss\ maps of $P$ and $Q$ are $4$ and $10$

1143: respectively, hence $\Delta_R$ is a hexagon

1144: with sides $8$ and $12$.

1145: 

1146: 

1147: \subsection{Reconstructing the curve $Q$ by the log-front $R$}

1148: It is instructive to have a closer look at the

1149: computation done in \eqref{ttt} in the case

1150: $k=l$. In this case, the point $(a,b)$ does

1151: not escape at infinity, but it is still a pole

1152: of the logarithmic Gauss map of $R$ (see also

1153: Section \ref{sGR} below). In other words,

1154: the curve $R$ is tangent to

1155: $a=1$ at the corresponding point.

1156: 

1157: In probabilistic applications,

1158: the curve $P$ is given and one knows that the frozen

1159: boundary is compact and tangent to given

1160: lines. This allows to fix the real boundary

1161: points of $Q$ and their multiplicities

1162: (which have to match the corresponding

1163: multiplicities for $P$). In particular,

1164: this is how the curve $Q$ in Figure

1165: \ref{PandQ} is determined from the

1166: requirement that the frozen boundary

1167: in Figure \ref{froz} is inscribed in

1168: a hexagon.

1169: 

1170: 

1171: 

1172: 

1173: \section{Pl\"ucker-type formulas for $R$}

1174: 

1175: 

1176: \subsection{Logarithmic Gau\ss\ map of $R$}\label{sGR}

1177: 

1178: 

1179: Let $\Rh$ be the product

1180: of the normalizations $\Pt$ and $\Qt$ over their

1181: logarithmic Gau\ss\ maps:

1182: %

1183: \begin{equation} \label{Gauss_diag}

1184: \xymatrix{

1185: & \Rh \ar[dr]^{\pi_Q} \ar[dl]_{\pi_P}  \ar[dd]^{\gamma_R}\\

1186: \Pt \ar[dr]_{\gamma_P}&& \Qt \ar[dl]^{\gamma_Q}

1187:   \\

1188: & \pline

1189: }

1190: \end{equation}

1191: %

1192: In English, a point of $\Rh$ corresponds to a tangency of

1193: a branch of $P$ to a translate of a branch of $Q$.

1194: Note that $\Rh$ is a partial normalization of the plane curve $R$:

1195: we have the factorization $\Rt\to\Rh\to R$ for the normalization $\Rt\to R$.

1196: We claim that the logarithmic Gau\ss\ map of

1197: $\Rt$ factors through the natural map $\Rt\to \Rh$.

1198: 

1199: \begin{prop}

1200: The logarithmic Gau\ss\ map for $R$ is the composition

1201: $$

1202: \Rt \to \Rh \overset{\gamma_R}\to \pline

1203: $$

1204: \end{prop}

1205: 

1206: \begin{proof}

1207: The essential geometric content of this result is

1208: already implicit in \eqref{double_dual}. In

1209: coordinates, the claim is elementary to check on the open dense set

1210: where tangency is

1211: nondegenerate.  It suffices to analyze the

1212: Gauss\ map of the additive analog $P-Q$ of the

1213: log front $P\div Q$. Let $P$ and $Q$ be given by

1214: $$

1215: y=P(x)\,,\quad y=Q(x) \,.

1216: $$

1217: Then $P-Q$ is parametrized by

1218: $$

1219: (x-s,P(x)-Q(s))

1220: $$

1221: where $s=s(x)$ is a solution of $P'(x)=Q'(s)$. The

1222: Gauss\ map of the above curve is clearly

1223: $P'(x)=Q'(s)$.

1224: \end{proof}

1225: 

1226: 

1227: \begin{cor}\label{T_deg}

1228: \begin{equation}

1229:   \label{gamma_R}

1230: \deg \gamma_R = \deg \gamma_P \, \deg \gamma_Q \,.

1231: \end{equation}

1232: \end{cor}

1233: %

1234: 

1235: 

1236: 

1237: 

1238: \subsection{Geometric genus of $R$}

1239: 

1240: 

1241: 

1242: The geometric genus of $R$, or equivalently, the

1243: Euler characteristic $\chi(\Rt)$ can be computed by

1244: Riemann-Hurwitz formula applied to the

1245: map $\gamma_R$ in \eqref{Gauss_diag}. To do this we

1246: need to be able to compare $\chi(\Rt)$ and $\chi(\Rh)$.

1247: 

1248: The map $\gamma_R$ is ramified over the branchpoints

1249: of either $\gamma_P$ or $\gamma_Q$. Suppose that

1250: at a point $p\in\Pt$ the map $\gamma_P$ is given by

1251: %

1252: \begin{equation}

1253:   \label{coordx}

1254:     x \to x^{\nu}\,, \quad \nu=1,2,\dots\,,

1255: \end{equation}

1256: %

1257: in a suitable local coordinate $x\in \C$ centered at $p$.

1258: In other words, suppose that $p$ is a logarithmic

1259: inflection point of multiplicity $\nu-1$.

1260: Note that such point may be singular or non-singular point of $P$.

1261: This

1262: integer $\nu$ will be denoted by $\nu(p)$.

1263: 

1264: A point of $\Rh$ is a pair $(p,q)\in \Pt \times \Qt$ such that

1265: $$

1266: \gamma_P(p) = \gamma_Q(q) = x \in \pline \,.

1267: $$

1268: Let $x_p$ and $x_q$ be local coordinates at $p$ and $q$

1269: as in \eqref{coordx}. The local equation of $\Rh$

1270: is

1271: $$

1272: x_p^{\nu(p)} = x_q^{\nu(q)} \,, \quad (x_p,x_q)\in \C^2 \,.

1273: $$

1274: This has $\gcd(\nu(p),\nu(q))$ branches and hence

1275: produces $\gcd(\nu(p),\nu(q))$ points

1276: $$

1277: r_1,\dots, r_{\gcd(\nu(p),\nu(q))} \in \Rt

1278: $$

1279: that are mapped to $x$ by the logarithmic Gau\ss\ map.

1280:  Each is a logarithmic inflection point

1281: of multiplicity $\lcm(\nu(p),\nu(q))-1$. Note that while

1282: all $r_i\in \Rt$ are logarithmic

1283: inflection points of the same multiplicity, the

1284: multiplicities of the corresponding branches of $R$ may

1285: be different and are not determined by the numbers $\nu(p)$

1286: and $\nu(q)$.

1287: 

1288: {}From definitions

1289: %

1290: \begin{equation}

1291:   \label{chiRh}

1292:     \chi(\Rh) = \int_{\Rh} 1 \, d\chi  = \int_{\pline} (\gamma_P)_* 1 \,

1293: (\gamma_Q)_* 1 \, d\chi \,.

1294: \end{equation}

1295: %

1296: We conclude

1297: 

1298: \begin{thm}\label{T_gen}

1299: The Euler characteristic $\chi(\Rt)$ is given by

1300: $$

1301: \chi(\Rt) = \chi(\Rh) + \sum_{\gamma_P(p)=\gamma_Q(q)} (\gcd (\nu(p),\nu(q)) -1)

1302: $$

1303: where $ \chi(\Rh) = ((\gamma_P)_* 1 ,

1304: (\gamma_Q)_* 1)_\chi =

1305: - 2 \deg \gamma_P \, \deg \gamma_Q +

1306: \chi(\Pt)\, \deg \gamma_Q + \chi(\Qt) \, \deg \gamma_P$.

1307: \end{thm}

1308: %

1309: The above inner product with respect to the Euler characteristic

1310: is defined as in \eqref{chiRh}. Note that generically the

1311: the branchpoints of $\gamma_P$ are disjoint from branchpoints of $\gamma_Q$.

1312: In this case,

1313: the above formulas may be simplified as follows.

1314: %

1315: \begin{cor}

1316: If tangency does not occur at two logarithmic inflection points then

1317: %

1318: \begin{equation}

1319:   \label{chiRs}

1320:   \chi(\Rt) = \chi(\Rh) = - 2 \deg \gamma_P \, \deg \gamma_Q +

1321: \chi(\Pt)\, \deg \gamma_Q + \chi(\Qt) \, \deg \gamma_P  \,.

1322: \end{equation}

1323: %

1324: \end{cor}

1325: 

1326: 

1327: 

1328: 

1329: 

1330: \subsection{Nodes and cusps of $R$}

1331: 

1332: %\ignore{

1333: The total cuspidal number of $R$ may now be

1334: determined using the formula \eqref{n_cusps}.

1335: \ignore{

1336: %Corollary \ref{T_deg} and Theorem \ref{T_gen}

1337: %once we trace the marking of the Newton polygon

1338: %$\Delta_R$ from Theorem \ref{thmNewton}.

1339: %Instead of patching together these rather complicated formulas

1340: %we take here a more direct path.

1341: %If $P$ and $Q$ are generic, the only singularities

1342: %of $R$ will be cusps and nodes and their

1343: %number may be determined using \eqref{n_nodes}.

1344: %

1345: %Denote the number of

1346: %vertices of the Minkowski sum polygon $\Delta=\Delta_P+\Delta_Q$

1347: %pairs of parallel sides in the Newton polygon $\Delta_R$

1348: %with $n$.

1349: %Let

1350: %\begin{equation}\label{sdp}

1351: %S^{\dd}_P=\tilde{P}\setminus\nu_{\bar{P}}^{-1}(P),\

1352: %S^{\dd}_Q=\tilde{Q}\setminus\nu_{\bar{Q}}^{-1}(Q).

1353: %\end{equation}

1354: Note that each point $s\in \dd P$

1355: is canonically associated with an edge of $\Delta_P$.

1356: %namely, it is the edge corresponding

1357: %to the divisor containing $\nu_{\bar{P}}(s)$

1358: %(resp. $\nu_{\bar{Q}}(s)$).

1359: Let $\zeta(s)\in\Z^2$ be the primitive integer

1360: outward normal vector to this edge of $\Delta_P$.

1361: Let $\delta(P,Q)$ be equal to the number of pairs

1362: $(s_P,s_Q)\in \dd P\times \dd Q$ such that

1363: $\zeta(s_P)$ and $\zeta(s_Q)$ are not parallel

1364: minus the number of pairs

1365: $(s_P,s_Q)\in \dd P\times \dd Q$ such that

1366: $\zeta(s_P)$ and $\zeta(s_Q)$ are parallel.

1367: Let $\bar\delta(P,Q)$ be the number defined in

1368: a similar way as $\delta(P,Q)$ but counting each pair

1369: $(s_P,s_Q)$ with the multiplicity equal to $m_Pm_Q$,

1370: where $m_P$ is the multiplicity of the branch of $\bar{P}$

1371: corresponding to $s_P$ and $m_Q$ is the multiplicity

1372: of the branch of $\bar{Q}$ corresponding to $s_Q$.

1373: Let $\delta_c(P,Q)=\bar\delta(P,Q)-\delta(P,Q)$.

1374: Clearly, if both curves $\bar{P}$ and $\bar{Q}$ are

1375: immersed near $\C T_{\Delta_{PQ}}\setminus\tor$

1376: then all relevant multiplicities are equal to one and

1377: $\delta_c(P,Q)=0$.

1378: %For a lattice polygon $\Delta$ let us denote with

1379: %$l_{\dd\Delta}=\#(\dd\Delta\cap\Z^2)$$

1380: %the number of lattice points on its perimeter.

1381: }

1382: 

1383: \begin{thm}\label{cuspidal}

1384: We have the following expressions for the cuspidal number $c_{{R}}$

1385: of the resultant curve $R=R(P,Q)$.

1386: $$c(R)=-\chi(\Rt)+|\dd R|

1387: -(\chi(\Pt)+c(P)-|\dd P|)(\chi(\Qt)+c(Q)-|\dd Q|).$$

1388: \end{thm}

1389: 

1390: \begin{proof}%[Proof of Theorem \ref{cuspidal}]

1391: This formula can be obtained as a straightforward

1392: combination of \eqref{n_cusps} and Corollary \ref{T_deg}.

1393: %and Theorem \ref{T_gen}.

1394: Nevertheless it is instructive to prove it by the Viro

1395: calculus (see Section \ref{Viro-calculus}) to prepare

1396: a way for the real counterpart in the next section.

1397: %(note that \cite{V} already contains application of this

1398: %calculus to the Pl\"ucker and Klein formulas in the classical case).

1399: 

1400: %Recall that $\Delta_{PQ}=\Delta_Q-\Delta_P$ was defined as the Minkowski

1401: %difference.

1402: Consider the family of translates $\tau_s({P})$ parameterized

1403: by $s\in\tor$.

1404: %$s\in\cT\setminus\ver(\cT)$,

1405: %where $\ver(\cT)$ are the vertices of the toric surface

1406: %$\cT$ corresponding to the polygon $\Delta_{PQ}$ as in \eqref{cT}.

1407: %i.e. the fixed points of the $\tor$-action.

1408: %Clearly, $\chi(\cT\setminus\ver(\cT))=0$.

1409: %

1410: Denote with $X(s)$

1411: %$\#(\tau_s(\bar{P}),\bar{Q})$

1412: the Euler characteristic of

1413: the space of pairs $(t_P,t_Q)\in\tilde{P}\times\tilde{Q}$,

1414: such that $\tau_s(t_P)=t_Q$.

1415: %and either $t_P\notin\dd P$ or $t_Q\notin\dd Q$.

1416: %$\tau_s(\nu_{\bar{P}}(t_P))=\nu_{\bar{Q}}(t_Q)$

1417: %and either $\nu_{\bar{P}}(t_P)\in\tor$ or $\nu_{\bar{Q}}(t_Q)\in\tor$.

1418: We have

1419: \begin{equation}\label{cint}

1420: \int\limits_{\tor} X(s) d\chi(s)=

1421: (\chi(\tilde{P})-|\dd P|)(\chi(\tilde{Q})-|\dd Q|).

1422: %(l_{\dd\Delta_P}-\sigma_P)(l_{\dd\Delta_Q}-\sigma_Q).

1423: \end{equation}

1424: It may be viewed as a corollary of the Fubini theorem

1425: since each pair of points from $\tilde{P}\setminus\dd P$ and

1426: $\tilde{Q}\setminus\dd Q$ will

1427: appear in $X(s)$ once.

1428: %, unless both of them

1429: %are mapped to the boundary of $\cT$.

1430: On the other hand we have

1431: \begin{multline}

1432: \label{cT2}

1433: \int\limits_{\tor} X(s) d\chi(s)=

1434: %l_{\dd\Delta_P}l_{\dd\Delta_P}-\delta(P,Q)

1435: -\chi(\tilde{R})+|\dd R|\\-c(P)(\chi(\tilde{Q})-|\dd Q|)

1436: -c(Q)(\chi(\tilde{P})-|\dd P|)

1437: -c(P)c(Q)-c(R).

1438: \end{multline}

1439: Indeed, we have $\chi(\tor)=0$ while

1440: $X(s)$ is constant for generic $s\in\cT$

1441: (namely, by Bernstein-Kouchnirenko formula it is equal to

1442: $\Vol(\Delta_P,\Delta_Q)$).

1443: However the value $X(s)$ drops

1444: if $s\in R$ and it drops further (by $m-1$) if there is a singular point

1445: of multiplicity $m$ at a branch of $\tilde{R}\setminus\dd R$ through $s$

1446: or if $\tau_s(\Pt)$ and $\Qt$ have a tangency of higher order

1447: (which in turn corresponds to a cusp of $\tilde{R}$).

1448: %As $\int\limits_{\C T_{\Delta}\setminus\ver(\C T_{\Delta})}

1449: %\#(\tau_s(\bar{P}),\bar{Q}) d\chi(s)$

1450: %also compute some pairs $(t_P,t_Q)$ with

1451: %$\nu_{\bar{P}}(t_P),\nu_{\bar{Q}}(t_Q)\notin\tor$

1452: %we have to insert the correction by $l_{\dd\Delta_P}l_{\dd\Delta_P}-\delta(P,Q)$.

1453: We get the theorem as the combination of \eqref{cint} and \eqref{cT2}.

1454: \end{proof}

1455: 

1456: If $P$ and $Q$ are generic, %(in a proper sense),

1457: the only singularities of $R$ will be cusps and nodes and the

1458: nodal number of $R$ may be recovered from the adjunction formula.

1459: 

1460: For example, let $P$ and $Q$ be generic curves

1461: of degree $d_P$ and $d_Q$. In this case

1462: %we have $\bar\delta(P,Q)=3d_Pd_Q$,

1463: %%$l_{\dd\Delta_P}=3d_P$, $l_{\dd\Delta_Q}=3d_Q$,

1464: %$\chi(\tilde{P})=3d_P-d_P^2$, $\chi(\tilde{Q})=3d_P-d_Q^2$

1465: %and

1466: $$

1467: \chi(\Rt) = -d_P d_Q (4 d_P d_Q -3 d_P-3 d_Q)\,.

1468: $$

1469: Note that this grows very fast with $d_P$ and $d_Q$.

1470: Already when $P$ and $Q$ are generic conics, the

1471: geometric genus of $R$ is $9$.

1472: 

1473: The Newton polygon

1474: of $R$ and its boundary  were determined above in Section \ref{expl_poly}.

1475: Generically, $R$ will not  have multiple point

1476: on the boundary. For the number of cusps Theorem \ref{cuspidal} produces

1477: $$

1478: \textup{\# cusps} = 3 d_P^2 d_Q^2-6 d_P d_Q \,.

1479: $$

1480: Accordingly for the number of nodes we

1481: get:

1482: $$

1483: \textup{\# nodes} =

1484: \frac{d_P^4 d_Q^2}{2}+2 d_P^3 d_Q^3+\frac

1485: {d_P^2 d_Q^4}{2}-3 d_P^3 d_Q^2-3 d_P^2 d_Q^3

1486: -2 d_P^2 d_Q^2+9 d_P d_Q\,.

1487: $$

1488: %

1489: For $d_P=1$, these specialize to the classical

1490: Pl\"ucker formulas.

1491: The last two formulas were

1492: obtained in \cite{FO}.

1493: 

1494: 

1495: 

1496: \section{Klein-type formula for the log-front $R$}

1497: \subsection{Refinement of the nodal and cuspidal numbers}

1498: For this section it will be important that both $P$ and $Q$

1499: are defined over the field $\R$ of real numbers. In this case, clearly,

1500: the curve $R$ is defined over $\R$ as well.

1501: 

1502: An algebraic curve $P\subset\cT$ is defined over $\R$

1503: if and only if it is invariant with respect to the involution

1504: of complex conjugation $\conj:\cT\to\cT$.

1505: The fixed point locus of this involution is the real

1506: toric surface $\R\cT\subset\cT$.

1507: The real locus $\R{P}$ coincides with the intersection

1508: $P\cap\R\cT$.

1509: 

1510: For a real curve $P$ we may refine both the cuspidal number $c(P)$

1511: and the nodal number $b(P)$ as follows.

1512: Let $c^{\re}(P)$ be the number of cusps of $P$

1513: (counted with multiplicity as in Section \ref{bcdefn}) that are

1514: %from the real branches of singularities of $P$.

1515: {\em real}, i.e. contained in $\R T_\Delta$.

1516: In other words, to get $c^{\re}(P)$ we add over (real and imaginary)

1517: branches of {\em real} singular points of $P$ the multiplicities of these

1518: branches diminished by 1.

1519: We set $c^{\im}(P)=c(P)-c^{\re}(P).$

1520: 

1521: Let $b^{\re}_+(P)$ be the sum over all singular

1522: points $s$ of $P$ in $\R\cT$ of the number

1523: of the pairs of conjugate imaginary branches of $P$.

1524: %Equivalently, $b^{\re}_+(P)$ is one half of the

1525: %cardinality of $\nu_{\bar{A}}^{-1}(\R T_{\Delta_A})$

1526: %(recall that $\nu_{\bar{A}}:\tilde{A}\to \bar{A}$ is the

1527: %normalization).

1528: If $s\in\R P$ is a singular point

1529: we may introduce $b^{\re}_-(P,s)$ as the sum

1530: of the local intersection numbers over all possible

1531: pairs of the real branches

1532: of $P$ at $s$. Let $b^{\re}_-(P)$

1533: be the sum of $b^{\re}_-(P,s)$ over all

1534: real singular points of $P$ and let

1535: $b^{\im}(P)=b(P)-b^{\re}_+(P)-b^{\re}_-(P)$.

1536: 

1537: Thus we get the refinements $$c(P)=c^{\re}(P)+c^{\im}(P)$$

1538: and $$b(P)=b^{\re}_+(P)+b^{\re}_-(P)+b^{\im}(P).$$

1539: In the case when the only singularities of $P$ are ordinary cusps and

1540: nodes the numbers $c^{\re}(P)$ and $c^{\im}(P)$

1541: are the numbers of real and complex cusps respectively while

1542: the numbers $b^{\re}_-(P),b^{\re}_+(P)$ and $b^{\im}(P)$

1543: are the numbers of real hyperbolic ($A_1^-$), real elliptic ($A_1^+$)

1544: and imaginary nodes.

1545: 

1546: It is convenient to define the {\em boundary nodal number}

1547: $b(\dd P)$ to be equal to the sum of the multiplicities

1548: of the boundary points of $P$ minus the simple cardinality of $\dd P$.

1549: (Note that since $\dd P$ is 0-dimensional this boundary nodal number

1550: also works as the boundary counterpart of the cuspidal number.)

1551: Again we have the refinement $b(\dd P)=b^{\re}(\dd P)+b^{\im}(\dd P)$

1552: that counts real and imaginary boundary points separately.

1553: Thus $b^{\re}(\dd P)$ is the measure of nontransversality of the real

1554: locus $\R P$ to the boundary divisor of the toric surface $\cT$.

1555: 

1556: \subsection{Computations for $R$}

1557: 

1558: In general, the set of refined nodal and cuspidal numbers

1559: $b^{\re}_+,b^{\re}_-,c^{\re},c^{\im}$ for

1560: the log-front $R$ is not determined by

1561: the corresponding sets for $P$ and ${Q}$.

1562: Let us recall a well-known example illustrating this

1563: in the case of classical projective duality, i.e. if

1564: $P$ is given by the polynomial $z+w+1$: two real inflection

1565: points may disappear together with a bitangent real line.

1566: \begin{exa}

1567: Let $\R Q_1,\R Q_2\subset\rp^2$ be real quartic curves

1568: pictured in Figure \ref{quart}. The first curve can be constructed

1569: by perturbation of the union of two ellipses while the second one

1570: can be constructed by perturbation of the union of four lines.

1571: We have $b^{\re}_+({R_1})=0,c^{\re}({R_1})=8,

1572: b^{\re}_+({R_2})=4,c^{\re}({R_1})=0$.

1573: \begin{figure}[h]

1574: \centerline{\psfig{figure=quart.eps,height=1.4in,width=3.5in}}

1575: \caption{\label{quart} Real quartic curves $\R Q_1$ and $\R Q_2$

1576: whose projective duals have distinct $b^{\re}_+$ and $c^{\re}$.}

1577: \end{figure}

1578: \end{exa}

1579: 

1580: The classical Klein's formula \cite{Kl-paper}

1581: allows us to compute

1582: $2b^{\re}_+(\bar{R})+c^{\re}(\bar{R})$

1583: in the case when $P$ is a line.

1584: To generalize this statement for a larger class of real curves $P$

1585: let us look at the

1586: %so-called {\em coamoebas} of the curves

1587: %$P,Q\subset\tor$.

1588: argument map $\Arg:\tor\to S^1\times S^1$

1589: defined by $$\Arg(z,w)=(\arg(z),\arg(w)).$$

1590: The image $\Arg(A)\subset S^1\times S^1$

1591: is called {\em the coamoeba} or {\em the alga} of $A\subset\tor$, cf. \cite{Pa},

1592: \cite{KeVa}.

1593: 

1594: The real 2-torus $S^1\times S^1$ is a group which has $\Z_2\times\Z_2$

1595: as its subgroup.

1596: Let $T=S^1\times S^1/\Z_2\times\Z_2$ be the quotient group and

1597: $\beta:S^1\times S^1\to T=S^1\times S^1/\Z_2\times\Z_2$ be

1598: the projection map. Note that $T$ is itself a group isomorphic

1599: to $S^1\times S^1$ and the zero in this group is $0_T=\beta(1,1)$.

1600: 

1601: We need to compactify the map $\beta\circ\Arg:P\setminus\dd P\to T$.

1602: Note that this map does not extend to $P$.

1603: Let us consider $\hat{P}$ to be {\em the real blow-up} of $\tilde{P}$

1604: at the finite collection of points

1605: $\dd P\subset\tilde{P}$.

1606: Naturally, $\hat{P}$ is a closed non-orientable surface whose

1607: Euler characteristic coincides with that of $\tilde{P}\setminus{\dd P}$

1608: (in the case of non-compact spaces

1609: we use Euler characteristic for homology with closed support)

1610: Furthermore, we have a natural extension

1611: $$\Alga_P:\hat{P}\to T$$

1612: such that $\Alga_P|_{\tilde{P}\setminus\dd P}=

1613: \beta\circ\Arg|_{\tilde{P}\setminus\dd P}$.

1614: 

1615: %Note that the involution of complex conjugation $\conj:\bar{A}\to\bar{A}$

1616: %of a curve $\bar{A}$ lifts to its normalization $\tilde{A}$.

1617: %We may define $\R\tilde{A}$ as the fixed point set of this lift

1618: %so that $\nu_{\bar{A}}(\R\tilde{A})\subset\R\bar{A}$.

1619: %Define the function

1620: %$$\rho:\tilde{Q}\to\Z$$

1621: %to be $\rho(s)=\chi((\hat\Arg_P)^{-1}(\Arg_Q(s)))$,

1622: %if $s\in Q$ and

1623: %$\rho(s)=...$

1624: 

1625: Consider the following integral

1626: $$I_{\R}=\int\limits_{t\in\R\cT\setminus\ver(\R\cT)}

1627: X(t) d\chi_{\R\cT}(t)$$

1628: (note that here we use the family $\tau_t(P)$

1629: only for $t\in\R\cT$).

1630: This integral is the real counterpart of the integral \eqref{cint}

1631: from the proof of Theorem \ref{cuspidal} where we defined the number

1632: $X(t)$, but now we use only the real translations.

1633: %was defined, but is defined

1634: %using only real translations.

1635: 

1636: As in Theorem \ref{cuspidal} the integral $I_{\R}$ can be computed

1637: in two different ways.

1638: 

1639: \ignore{

1640: \begin{prop}\label{generalKlein}

1641: We have the equality

1642: \ignore{

1643: 2b^{\re}_+(\bar{R})+c^{\re}(\bar{R})=

1644: (4-2n)\Vol(\Delta_P,\Delta_Q)\\

1645: -\sum\limits_{s\in\Sigma\tilde{Q}\setminus\nu_{\bar{Q}}^{-1}(\R T_\Delta)}

1646: c(s)\chi(\hat\Arg_P)^{-1}(\hat\Arg_Q(s))\\

1647: -\sum\limits_{s\in\Sigma\tilde{P}\setminus\nu_{\bar{P}}^{-1}(\R T_\Delta)}

1648: c(s)\chi(\hat\Arg_Q)^{-1}(\hat\Arg_P(s))

1649:  \\

1650: -\int\limits_{\alpha\in T}

1651: \chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)\\

1652: -\int\limits_{\alpha\in T}

1653: \chi((\hat\Arg_P)^{-1}(\alpha)\cap\dd\tilde{P})

1654: \chi((\hat\Arg_Q)^{-1}(\alpha)\cap\dd\tilde{Q})

1655: d\chi_T(\alpha).

1656: }

1657: 

1658: \begin{multline}\label{1I}

1659: I=\int\limits_{\alpha\in T}

1660: \chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)\\

1661: -\int\limits_{\alpha\in T}

1662: \chi((\hat\Arg_P)^{-1}(\alpha)\cap\dd\tilde{P})

1663: \chi((\hat\Arg_Q)^{-1}(\alpha)\cap\dd\tilde{Q})

1664: d\chi_T(\alpha).

1665: \end{multline}

1666: 

1667: 

1668: \end{prop}

1669: 

1670: \begin{proof}

1671: %To compute $\int\limits_{T\setminus\{0_T\}}

1672: %\chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)$

1673: %we may use the family $\tau_t(P)$ for $t\in\R T_\Delta$.

1674: 

1675: \ignore{

1676: One one hand

1677: \begin{multline}\label{1I}

1678: %\int\limits_{\R T_\Delta}\chi(\tau_t(\bar{P})\cap \bar{Q})d\chi_{\R T_\Delta}(t)

1679: I=

1680: \int\limits_{\alpha\in T}

1681: \chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)\\

1682: -\int\limits_{\alpha\in T}

1683: \chi((\hat\Arg_P)^{-1}(\alpha)\cap\dd\tilde{P})

1684: \chi((\hat\Arg_Q)^{-1}(\alpha)\cap\dd\tilde{Q})

1685: d\chi_T(\alpha),

1686: \end{multline}

1687: }

1688: Note that if two points $s_P\in\hat{P}$ and $s_Q\in\hat{Q}$ are different

1689: by a translation by $t\in\R T_\Delta$ then

1690: $\hat\Arg_P(s_P)=\hat\Arg_Q(s_Q)$.

1691: Conversely, if $\hat\Arg_P(s_P)=\hat\Arg_Q(s_Q)$ and

1692: either $s_P$ or $s_Q$ is projecting to $\tor$ by the

1693: composition of the blowup map and the normalization

1694: (of $\hat{P}$ or $\hat{Q}$ respectively) then there

1695: exists $t\in\R T_\Delta\setminus \ver(\R T_{\Delta_R})$

1696: such that $\tau_t(\bar{P})$ and $\bar{Q}$ have

1697: $\nu_{\bar{Q}}(s_Q)\in\C T_{\Delta_R}$ as one of their

1698: intersection point.

1699: 

1700: Suppose that a point of $\tau_t(\bar{P})\cap\bar{Q}$,

1701: $t\in\R T_\Delta\setminus \ver(\R T_{\Delta_R})$, corresponds

1702: to a pair $(s_P,s_Q)$ with $s_P\in\hat{P}$ and $s_Q\in\hat{Q}$

1703: such that $s_P$ is a point projecting to $u_P\in S^{\dd}_{P}$

1704: by the blowup $\hat{P}\to\tilde{P}$ and

1705: $s_Q$ is a point projecting to $u_Q\in S^{\dd}_{Q}$

1706: by the blowup $\hat{Q}\to\tilde{Q}$.

1707: Then the side of $\dd\Delta_P$ corresponding to $u_P$

1708: has to be parallel to the side of $\dd\Delta_Q$ corresponding to $u_Q$.

1709: But each such pair $(u_P,u_Q)$ gives a circle of the pairs

1710: $(s_P,s_Q)$ contributing to our intersection and thus contributes

1711: zero to the Euler characteristic.

1712: \end{proof}

1713: 

1714: 

1715: \ignore{

1716: We do not have a correction coming from the pairs of points

1717: corresponding to the exceptional divisors of $\hat{P}$ and $\hat{Q}$

1718: as their total contribution to the Euler characteristic is zero.

1719: %Here we can exclude $0_T$ from $T$

1720: %as $\chi((\hat\Arg_P)^{-1}(0_T))\chi((\hat\Arg_Q)^{-1}(0_T))=

1721: %\chi(\R\tilde{P})\chi(\R\tilde{Q})$ which is equal to zero as the

1722: %Euler characteristic of a closed 1-manifold is zero.

1723: 

1724: On the other hand,

1725: \begin{multline*}\label{otherhand}

1726: %\int\limits_{\R T_\Delta}\chi(\tau_t(\bar{P})\cap \bar{Q})d\chi_{\R T_\Delta}(t)=

1727: I=(4-n)\Vol(\Delta_P,\Delta_Q)

1728: -\sum\limits_{s\in\Sigma\tilde{Q}\setminus\nu_{\bar{Q}}^{-1}(\R T_\Delta)}

1729: c(s)\chi(\hat\Arg_P)^{-1}(\hat\Arg_Q(s))\\

1730: -\sum\limits_{s\in\Sigma\tilde{P}\setminus\nu_{\bar{P}}^{-1}(\R T_\Delta)}

1731: c(s)\chi(\hat\Arg_Q)^{-1}(\hat\Arg_P(s))

1732: -2b^{\re}_+(\bar{R})-c^{\re}(\bar{R}).

1733: %-\int\limits_{\rp^1}\chi(\gamma^{\R}_P^{-1}(t))\chi(\gamma^{\R}_Q^{-1}(t)),

1734: \end{multline*}

1735: %where $\gamma^{\R}_P:\R\tilde{P}\to\rp^1$ and $\gamma^{\R}_Q:\R\tilde{Q}\to\rp^1$

1736: %are the logarithmic Gauss maps for the curves $\tilde{P}$ and $\tilde{Q}$

1737: %(cf. \cite{Ka}) restricted to their real part.

1738: To verify this formula

1739: %\eqref{otherhand}

1740: we note that $\chi(\R T_{\Delta_R}\setminus\ver(\R T_{\Delta_R}))=4-n$,

1741: $\#(\tau_t(\bar{P}),\bar{Q})=

1742: \Vol(\Delta_P,\Delta_Q)$ by Kouchnirenko-Bernstein formula if

1743: $\tau_t(\bar{P})$ and $\bar{Q}$ are transverse (which means,

1744: in particular, that their intersection is disjoint from their

1745: singular points).

1746: We have a reduction of this number if

1747: $\tau_t(\bar{P})$ and $\bar{Q}$ are tangent or if their intersection

1748: contain singularities of either curve.

1749: The terms

1750: $$\sum\limits_{s\in\Sigma\tilde{Q}\setminus\nu_{\bar{Q}}^{-1}(\R T_\Delta)}

1751: c(s)\chi(\hat\Arg_P)^{-1}(\hat\Arg_Q(s))

1752: +\sum\limits_{s\in\Sigma\tilde{P}\setminus\nu_{\bar{P}}^{-1}(\R T_\Delta)}

1753: c(s)\chi(\hat\Arg_Q)^{-1}(\hat\Arg_P(s))$$

1754: take care of the latter case.

1755: The term $2b^{\re}_+(\bar{R})$ takes care of the imaginary tangencies

1756: while the term $c^{\re}(\bar{R})$ takes care of the real tangencies

1757: (there is a 1-dimensional family of such tangencies but the rest contributes

1758: zero to the Euler characteristic).

1759: \end{proof}

1760: }

1761: }

1762: %\subsection{The case when $P$ is a simple Harnack curve}

1763: %Proposition \ref{generalKlein} contains the quantity

1764: %\begin{multline*}

1765: %\int\limits_{\alpha\in T}

1766: %\chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)

1767: %\end{multline*}

1768: For this computation to depend only on visible characteristics

1769: of the curves $P$ and $Q$ we need to assume that $P$ is a

1770: Harnack curve (see Appendix A).

1771: %The right-hand side of

1772: %%Proposition \ref{generalKlein} contains the quantity

1773: %\eqref{1I}

1774: %may not be immediately visible from the geometry of $\R P$ and $\R Q$.

1775: %However, if $P$ is a simple Harnack curve (see eg. \cite{MR})

1776: %then we can compute this quantity explicitly with the help of

1777: %the formulas deduced in Appendix A.

1778: 

1779: 

1780: Let $$\epsilon(\Delta_P,\Delta_Q)=\sum\limits_{\Delta'_P,\Delta'_Q}

1781: \Area(\Delta'_P+\Delta'_Q),$$

1782: where $\Delta'_P$ (resp. $\Delta'_Q$) run over all possible

1783: sides of the polygon $\Delta_P$ (resp. $\Delta_Q$) and

1784: $\Delta'_P+\Delta'_Q$ stands for the parallelogram obtained

1785: as the Minkowski sum of the intervals $\Delta'_P$ and $\Delta'_Q$.

1786: 

1787: 

1788: %Let $\perimeter(\Delta_Q)$ be the number of lattice point on

1789: %the perimeter of $\Delta_Q$, i.e. $\#(\Delta_Q\cap\Z^2)$.

1790: \begin{prop}\label{intHarnack}

1791: If $P$ is a Harnack curve and $Q$ is any curve defined over $\R$

1792: then

1793: %\begin{multline*}

1794: %\int\limits_{\alpha\in T}

1795: %\chi((\hat\Arg_P)^{-1}(\alpha))\chi((\hat\Arg_Q)^{-1}(\alpha))d\chi_T(\alpha)\\

1796: %-\int\limits_{\alpha\in T}

1797: %\chi((\hat\Arg_P)^{-1}(\alpha)\cap\dd\tilde{P})

1798: %\chi((\hat\Arg_Q)^{-1}(\alpha)\cap\dd\tilde{Q})

1799: %\chi_T(\alpha)\\

1800: \begin{multline*}

1801: I_{\R}=2\Area(\Delta_P)(\chi(\tilde{Q})-2b^{\re}_+({Q})-b^{\re}(\dd Q)

1802: -|\dd\Delta_Q|)+\\

1803: (2b^{\re}_+({P})+b^{\re}(\dd P))(2b^{\re}_+({Q})+b^{\re}(\dd Q))-\epsilon(\Delta_P,\Delta_Q),

1804: \end{multline*}

1805: %\end{multline*}

1806: %where $\epsilon(\Delta_P,\Delta_Q)$ is defined to be

1807: \end{prop}

1808: \begin{proof}

1809: Note that if two points $s_P\in\hat{P}$ and $s_Q\in\hat{Q}$ are different

1810: by a translation by $t\in\R\cT$ then

1811: $\Alga_P(s_P)=\Alga_Q(s_Q)$.

1812: Conversely, if $\Alga_P(s_P)=\Alga_Q(s_Q)$ and

1813: either $s_P$ or $s_Q$

1814: is not in the exceptional divisor of the blowup map

1815: %and the normalization

1816: %(of $\hat{P}$ or $\hat{Q}$ respectively)

1817: then there

1818: exists $t\in\R\cT\setminus \ver(\R\cT)$

1819: such that $\tau_t({P})$ and ${Q}$ have

1820: $s_Q$ as one of their

1821: intersection point. The Euler characteristic of

1822: the space formed by the pairs $(s_P,s_Q)$ with

1823: $\Alga_P(s_P)=\Alga_Q(s_Q)$

1824: and such that $s_P$ and $s_Q$ are from the exceptional

1825: divisors of the blowups $\hat{P}\to\tilde{P}$ and $\hat{Q}\to\tilde{Q}$ respectively

1826: is $\epsilon(\Delta_P,\Delta_Q)$.

1827: 

1828: We have $\chi((\Alga_P)^{-1}(\alpha))=2\Area(\Delta_P)$

1829: for any $\alpha\in T\setminus\{0_T\}$ by Lemma A\ref{arg-harnack}.

1830: Note that

1831: %$\chi((\hat\Arg_P)^{-1}(0_T))=2b^{\re}_+(Q))$.

1832: $\chi((\Alga_Q)^{-1}(T\setminus \{0_T\}))=\chi(\hat{Q})-2b^{\re}_+({Q})-b^{\re}(\dd Q)=

1833: \chi(\tilde{Q})-|\dd\Delta_Q|-2b^{\re}_+({Q})-b^{\re}(\dd Q)$.

1834: The contribution of $\alpha=0_T$ to the left-hand side of

1835: Proposition \ref{intHarnack} is

1836: $(2b^{\re}_+({P})+b^{\re}(\dd P))(2b^{\re}_+({Q})+b^{\re}(\dd Q))$ since $\R\hat{P}$

1837: is a closed 1-manifold and thus $\chi(\R\hat{P})=0$.

1838: \end{proof}

1839: 

1840: 

1841: Let $n$ be the number of vertices of the polygon

1842: $\Delta_{PQ}=\Delta_Q-\Delta_P$.

1843: \begin{prop}\label{2prop}

1844: If $P$ is a Harnack curve and $Q$ is any curve defined over $\R$

1845: immersed near the boundary of $\R\cT$ then

1846: \begin{multline*}

1847: I_{\R}=\Vol(\Delta_P,\Delta_Q)(4-2n)\\-2\Area(\Delta_P)c^{\im}({Q})-(2b^{\re}_+(P)+

1848: b^{\re}(\dd P))c^{\re}(Q)-

1849: (2b^{\re}_+({R})+c_{\re}({R})).

1850: \end{multline*}

1851: \end{prop}

1852: \begin{proof}

1853: The Euler characteristic of $\R\cT\setminus\ver(\R\cT)$

1854: is $4-2n$. For a generic $t$ we have $X(t)=

1855: \Vol(\Delta_P,\Delta_Q)$ by the Bernstein-Kouchnirenko formula \cite{Be}, \cite{Ku}.

1856: This number gets decreased if $\tau_t({P})$ and ${Q}$ are tangent

1857: or if one of their intersection point is singular for a branch of $\tilde{Q}$

1858: (note that ${P}$ is an immersed smooth curve since it is Harnack).

1859: The latter case contributes $-2\Area(\Delta_P)c^{\im}({Q})

1860: -(2b^{\re}_+(P)+b^{\re}(\dd P))c^{\re}(Q)$.

1861: 

1862: If $\tau_t({P})$ and ${Q}$ are tangent at a point in

1863: a non-real point then we have a bitangency since both ${P}$

1864: and ${Q}$ are invariant with respect to the involution of

1865: complex conjugation. This contributes $-2b^{\re}_+({R})$.

1866: \ignore{

1867: To compute the Euler characteristic of

1868: the tangencies at real points

1869: we note that for every point $s_q\in Q$ we have exactly $a_P$

1870: points such that there exists $t\in\R T_{\Delta_R}\setminus\ver(\R T_{\Delta_R})$

1871: with $\tau_t(\bar{P})$ tangent to $\bar{Q}$ at $s_Q$ since there are

1872: $a_P$ points in $\R\bar{P}$ with the same image of the Logarithmic Gauss

1873: map as $s_Q$ since $P$ is simple Harnack (see \cite{Mi}).

1874: For a point $s_q\in S^{\dd}_Q$ we have $a_P$ minus the number of points

1875: at $S^{\dd}_Q$ with the same image of the Gauss map as $s_Q$ plus the number

1876: of those points from this set that $\tilde{P}$ has the same local intersection

1877: number with the boundary divisor as $\tilde{Q}$ has at $s_Q$. This contributes

1878: $\sigma(\R P,\R Q)$.

1879: }

1880: The tangencies at real points contribute $\chi(\R\tilde{R})=0$ plus

1881: $c_{\re}({R})$ where the tangencies of $\tau_t({P})$ and ${Q}$

1882: are of higher order.

1883: \end{proof}

1884: 

1885: \subsection{Klein's formula}

1886: Combining Proposition \ref{intHarnack} and \ref{2prop}

1887: we get the following theorem.

1888: \begin{thm}\label{Harthm}

1889: If $P$ is a simple Harnack curve and $Q$ is any curve defined over $\R$

1890: immersed near the boundary of $\R\cT$

1891: then

1892: \begin{multline*}

1893: 2b^{\re}_+({R})+c^{\re}({R})=

1894: (4-2n)\Vol(\Delta_P,\Delta_Q)+\\

1895: 2\Area(\Delta_P)(|\dd\Delta_Q|-\chi(\tilde{Q})

1896: +2b^{\re}_+({Q})+b^{\re}(\dd Q)-c^{\im}({Q}))\\

1897: -(2b^{\re}_+({P})+b^{\re}(\dd P))(2b^{\re}_+({Q})+b^{\re}(\dd Q)+c^{\re}(Q))

1898: +\epsilon(\Delta_P,\Delta_Q).

1899: \end{multline*}

1900: \end{thm}

1901: %\begin{proof}

1902: %Since $\bar{P}$ is smooth we have the set $\Sigma\tilde{P}$ empty.

1903: %Since $\chi(\hat\Arg_P)^{-1}(\alpha)=\Area(\Delta_P)$

1904: %for $\alpha\in T\setminus 0_T$

1905: %we have

1906: %$$\sum\limits_{s\in\Sigma\tilde{Q}\setminus\nu_{\bar{Q}}^{-1}(\R T_\Delta)}

1907: %c(s)\chi(\hat\Arg_P)^{-1}(\hat\Arg_Q(s))=\Area(\Delta_P)c^{\im}(\bar{Q}).$$

1908: %\end{proof}

1909: 

1910: \begin{coro}\label{H-coro}

1911: Let $\R{P},\R{Q}\in\rp^2$ be two curves of degree $d_P$ and $d_Q$

1912: (respectively) not passing via $(0:0:1),(0:1:0),(1:0:0)$.

1913: Suppose that $P$ is a simple Harnack curve and ${Q}$

1914: is a smooth curve. Then

1915: $$

1916: 2b^{\re}_+({R})+c^{\re}({R})=

1917: d^2_Pd^2_Q-2d_Pd_Q.

1918: $$

1919: \end{coro}

1920: \begin{proof}

1921: We have $n=6$, $\Area(\Delta_P)=\frac{d_P^2}{2}$, $\chi(\tilde{Q})=3d_Q-d^2_Q$,

1922: $|\dd\Delta_Q|=3d_Q$ and $\epsilon(\Delta_P,\Delta_Q)=6d_Pd_Q$.

1923: \end{proof}

1924: 

1925: 

1926: Let us deduce the Klein formula \cite{Kl}

1927: in its classical form from Theorem \ref{Harthm}

1928: in the case when $P$ is a line $z+w+1=0$

1929: and $\R Q$ is a curve of degree $d$ in $\rp^2$ not passing

1930: via $(0:0:1),(0:1:0),(1:0:0)$. In this case $\Delta_Q$

1931: is a triangle with vertices $(0,0)$, $(d,0)$ and $(0,d)$.

1932: The line $P$ is a simple Harnack curve of degree 1 with $a_P=1$

1933: and thus

1934: we may apply Theorem \ref{Harthm}. The polygon $\Delta_R$

1935: is a hexagon in this case, so $n=6$, while $\Vol(\Delta_P,\Delta_Q)=d$

1936: and $\epsilon(\Delta_P,\Delta_Q)=6d$.

1937: 

1938: Note that the classical Klein formula computes

1939: $2b^{\re}_+({R})+c^{\re}({R})$

1940: for the dual curve in $\R P^2$ while our formula does it for

1941: ${R}\subset\R\cT$. The toric surface

1942: $\R\cT$ is the result of the blowup of $\R P^2$ at

1943: three points and, in general, such blowup might change the

1944: characteristics $b^{\re}_+$ and $c^{\re}$.

1945: Let us assume that ${Q}$ intersects the boundary

1946: divisor $\cp^2\setminus\tor$ transversely, so that this

1947: blowup is disjoint from ${Q}$.

1948: Then $\chi(\hat{Q})=\chi(\tilde{Q})-3d$.

1949: 

1950: We get

1951: \begin{multline*}

1952: 2b^{\re}_+({R})+c^{\re}({R})=-8d

1953: +2b^{\re}_+({Q})-\chi(\tilde{Q})+3d-c^{\im}({Q})+6d\\

1954: =d-c^{\im}({Q})+2b^{\re}_+({Q})-\chi(\tilde{Q}).

1955: \end{multline*}

1956: Note that $c^{\im}({Q})=c_{{Q}}-c^{\re}({Q})$

1957: while the classical Pl\"ucker formula \cite{Pl}

1958: for ${Q}$ implies that

1959: $\chi(\tilde{Q})+c_{{Q}}=2d-d^*,$

1960: where $d^*$ is the degree of $R\subset\tor\subset\cp^2$

1961: (called {\em the class of $Q$} in Klein's paper \cite{Kl-paper}).

1962: Thus we get

1963: $2b^{\re}_+({R})+c^{\re}({R})=

1964: -d+d^*+2b^{\re}_+({Q})+c^{\re}({Q})$ and, therefore,

1965: \begin{equation}

1966: d-2b^{\re}_+({Q})-c^{\re}({Q})=

1967: d^*-2b^{\re}_+({R})-c^{\re}({R}),

1968: \end{equation}

1969: which is the Klein formula in its original form \cite{Kl-paper}.

1970: 

1971: \subsection{Example}

1972: Let us go back to the log-front from Figure \ref{froz}.

1973: By Theorem \ref{Harthm} the sum of the real cusps of the

1974: log-front and twice the number of real solitary nodes equals 24.

1975: Indeed, we have $n=6$, $\Vol(\Delta_P,\Delta_Q)=8$,

1976: $\Area(\Delta_P)=4$, $|\dd\Delta_Q|=12$, $\chi(\tilde{Q})=2$,

1977: $2b_+^{\re}(Q)=b_+(\dd Q)=c(Q)=0$ while $\epsilon(\Delta_P,\Delta_Q)=48$.

1978: 

1979: There are 6 real cusps visible on Figure \ref{froz}. Furthermore,

1980: Figure \ref{3quad} shows images of the remaining 3 quadrants

1981: under the map $(x,y)\mapsto (\log|x|,\log|y|)$. There are 12 more

1982: real cusps. Thus by Theorem \ref{Harthm} our log-front has 3 real

1983: solitary points.

1984: 

1985: \begin{figure}[!h]

1986:   \centering

1987:   {\scalebox{0.5}{\includegraphics{other_quadr.ps}}}

1988:   \caption{Images in the other three quadrants}

1989:   \label{3quad}

1990: \end{figure}

1991: 

1992: 

1993: 

1994: 

1995: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1996: 

1997: 

1998: \newpage

1999: \section*{Appendix A: Harnack curves and their algae.}

2000: 

2001: In his 1876 paper \cite{Ha} A. Harnack produced

2002: for each $d$ examples of algebraic curves of degree $d$ in $\rp^2$

2003: with $\frac{(d-1)(d-2)}{2}+1$ real (topological) components.

2004: Furthermore, in the same paper he has shown that $\frac{(d-1)(d-2)}{2}+1$

2005: is the upper bound for the number of components

2006: of any curve of degree $d$ in $\rp^2$

2007: 

2008: Among the curves constructed by Harnack there were

2009: some ``canonical" curves whose topological arrangement

2010: in $\rp^2$ is especially easy to describe.

2011: Note that any component of a smooth curve is either

2012: contractible (i.e. bounds a disk in $\rp^2$)

2013: or is isotopic to $\rp^1\subset\rp^2$. A contractible component

2014: is called an {\em oval} while the disk bounded by it is called

2015: the {\em interior} of the oval. The oval whose interior is

2016: disjoint from other ovals is called {\em empty}.

2017: 

2018: For an odd $d=2k+1$

2019: there is a smooth algebraic curve of degree $d$ that

2020: consists of $k(2k-1)$ empty ovals

2021: and a non-contractible component.

2022: For an even $d=2k$ there is a smooth algebraic curve of degree $d$ that

2023: consists of $(k-1)(2k-1)$ empty ovals and

2024: one other oval whose interior contains $\frac{(k-1)(k-2)}{2}$

2025: of the $(k-1)(2k-1)$ empty ovals.

2026: 

2027: %Let $\R A\subset\tor$ be a curve whose Newton polygon is

2028: %a triangle with vertices $(0,0)$, $(d,0)$ and $(0,d)$.

2029: Let $P\subset\cT$ be a curve given by a real polynomial

2030: with the Newton Polygon $\Delta_P$ and $\cT$ be the toric

2031: surface corresponding to $\Delta_P$.

2032: Since $P$ is defined over

2033: $\R$ there is the real locus $\R P=P\cap\rtor$ which coincides

2034: with the fixed point set of the involution of complex conjugation

2035: on $P$.

2036: %Consider the logarithm map $$\Log:\tor\to\R^2,$$

2037: %$\Log(z,w)=(\log|z|,\log|w|)$. The image $\Log(A)\subset\R^2$

2038: %is called the amoeba of $A$, see \cite{GKZ}.

2039: 

2040: \begin{defn}\label{sH}

2041: A curve $P$ is called a {\em Harnack

2042: curve} if for every $(x,y)\in\R^2$ the set $\Log^{-1}(x,y)\cap P$

2043: consists of no more than two points.

2044: %there exist three arcs $\alpha_1,\alpha_2,\alpha_3$ on the

2045: %same topological component of $\R A$ such that

2046: \end{defn}

2047: 

2048: \begin{rmk}

2049: Earlier we called such curves {\em simple Harnack curves}

2050: to distinguish them from other curves in the Harnack construction.

2051: However, by now we have convinced ourselves that these {\em simple} curves

2052: are the most beautiful in the Harnack series of constructions.

2053: We propose to drop ``simple" from their name and call them

2054: {\em Harnack curves}.

2055: \end{rmk}

2056: 

2057: Harnack curves (from Definition \ref{sH})

2058: exist for any convex lattice polygon $\Delta$, cf. \cite{IV}.

2059: In \cite{Mi} it was shown that the topological type of

2060: the triad $(\R\cT;\R P,\dd\R\cT)$

2061: %(and of the compactification $(\R\c T,\R {A})$

2062: %in the real toric surface $\R T_{\Delta}$ corresponding to $\Delta$)

2063: depends only on $\Delta$ if $\R P$ is a smooth Harnack curve

2064: transverse to infinity.

2065: %Furthermore, as usual we may compactify the torus $\rtor$ to the toric surface

2066: %$\R\cT$ corresponding to the Newton polygon $\Delta$ and the topological

2067: %type of $(\R\c T,\overline{\R P})$ will remain determined only by $\Delta$.

2068: %As in the main part of the paper from now on in the Appendix

2069: %we denote with $P$ the corresponding compactification of the curve

2070: %so that $P\setminus\dd P=P\cap\tor$.

2071: 

2072: In the case when $\Delta$ is

2073: a triangle with vertices $(0,0)$, $(d,0)$ and $(0,d)$ we have

2074: $\R\cT=\rp^2$ and

2075: %the compactification $\R\bar{A}$ of

2076: %$\R A$ in $\rp^2$

2077: $\R P\subset\rp^2$ is a curve consisting of $\frac{(d-1)(d-2)}{2}$

2078: empty oval and one other component which is non-contractible if

2079: $d$ is odd and an oval containing $\frac{(d-2)(d-4)}{8}$ empty ovals

2080: if $d$ is even. In \cite{KO1} it was shown that all such curve

2081: form a contractible subspace in the space of all real curves

2082: of degree $d$.

2083: 

2084: Recently it was discovered that

2085: Harnack curves possess many extremal characteristic properties, among them

2086: are the following.

2087: \begin{itemize}

2088: 

2089: \item (\cite{MR}) We have

2090: $$\Area(\Log(P))=\pi^2\Area(\Delta),$$

2091: if $\R P$ is a Harnack curve with the Newton polygon $\Delta$.

2092: In the same time by \cite{PR}

2093: for any curve $A\subset\tor$ given by a (not necessarily

2094: real) polynomial with the Newton polygon $\Delta$ we have

2095: $$\Area(\Log(A))\le\pi^2\Area(\Delta).$$

2096: 

2097: Furthermore, if

2098: $\Area(\Log(A))=\pi^2\Area(\Delta)$ then $A$ can be translated

2099: (by a multiplication  with some $(z_0,w_0)\in\tor$) to a

2100: Harnack curve, see \cite{MR}.

2101: 

2102: \item (\cite{Mi}) The curve $\Log(\R P)\subset\R^2$ is embedded,

2103: does not have inflection points and contains $\#(\Int(\Delta)\cap\Z^2)$

2104: compact components (called {\em ovals}) if $\R P$ is a Harnack curve.

2105: Each such oval comes from one of the four quadrants in $\rtor$.

2106: The number of the ovals coming from the four quadrant

2107: equals to the number of lattice points $(j,k)\in\Z^2$ in the interior

2108: $\Int(\Delta)$ with given residue mod 2: $j\equiv j_0\pmod{2}$,

2109: $k\equiv k_0\pmod{2}$ (clearly there are four possible pairs of residues).

2110: 

2111: Furthermore any curve with that many ovals of $\Log(\R P)\approx P\setminus\dd P$

2112: is Harnack if the remaining component of $\R P$ intersects the infinity $\dd\R\cT$

2113: in a maximal way (see \cite{Mi}).

2114: \end{itemize}

2115: 

2116: The only singularities of a Harnack curve

2117: $\R P$ are isolated double points

2118: (the singularities of type $A_1^+$ according to \cite{AVGZ}) in $\rtor$,

2119: see \cite{MR}.

2120: Note that even though $\R P$ has to be smooth near the boundary

2121: divisor $\dd\R\cT$ it does not have to be transverse

2122: to the boundary divisor. An example of a singular simple Harnack

2123: curve of degree 6 in $\rp^2$ which is not transverse to the boundary

2124: divisor is sketched in Figure \ref{exaHarnack}.

2125: %\end{itemize}

2126: \begin{figure}[h]

2127: \centerline{\psfig{figure=exaHarnack.eps,height=2.5in,width=2.7in}}

2128: \caption{\label{exaHarnack} A singular simple Harnack curve in $\rp^2$ and

2129: its position with respect to the boundary divisor.}

2130: \end{figure}

2131: 

2132: The goal of this appendix is to give yet another characteristic property

2133: of the Harnack curves in terms of their algae.

2134: Suppose that $P$ is a Harnack curve %, see \cite{Mi},\cite{MR},\cite{KO}.

2135: Recall that in the previous section we denoted with $\hat{P}$

2136: the result of the real blowing up of $\tilde{P}$ at the points $\dd P$

2137: and with $$\Alga_P:\hat{P}\to T$$

2138: such that $\Alga_P|_{\tilde{P}\setminus\dd P}=

2139: \beta\circ\Arg|_{\tilde{P}\setminus\dd P}$.

2140: %of $\tilde{P}\setminus\nu_{\bar{P}}^{-1}(\bar{P}\setminus P)$

2141: %and $\R\hat{P}$ is the real part of $\hat{P}$.

2142: %Consider the argument map $\Arg:\tor\to T=S^1\times S^1$,

2143: %$\Arg(z,w)=(\arg(z),\arg(w)$.

2144: %Clearly, $\Arg|_P$ extends

2145: %to the map $$\hat\Arg_P:\hat{P}\to T.$$

2146: Recall that $P_0=\Alga^{-1}_P(0_T)$.

2147: 

2148: \begin{lemmaA}\label{arg-harnack}

2149: If $P$ is a Harnack curve then the restriction of the map $\Alga_P$

2150: to $\hat{P}\setminus P_0$

2151: $$\hat{P}\setminus P_0\to T\setminus\{0_T\}$$ is

2152: an unbranched covering of degree $2Area(\Delta_P)$.

2153: %Furthermore, $P_0\setminus\R\hat{P}$ is a finite set.

2154: \end{lemmaA}

2155: %Recall (see section \ref{logGauss}) that $a_P$ is the degree of

2156: %the logarithmic Gauss map of $P$. If $\bar{P}$ is a smooth curve

2157: %orthogonal to the boundary divisor $\C T_{\Delta_P}\setminus\tor$

2158: %then $a_P=\Area\Delta_P$.

2159: \begin{proof}

2160: %[Proof of Lemma A1]

2161: The critical points of the map $\Arg|_P$ are the points

2162: such that the image of the Logarithmic Gauss map $\gamma_P$ is in $\rp^1$

2163: (see \cite{Mi}).

2164: By \cite{Mi} $\gamma^{-1}(\rp^1)=\R\tilde{P}$.

2165: The map $\Alga_P|_{\hat{P}\setminus P_0}$ is proper since

2166: $\Alga_P^{-1}(0_T)= P_0$.

2167: 

2168: Since $\chi(T\setminus\{0_T\})=-1$ we have the degree of the covering

2169: $\Alga_P|_{\hat{P}\setminus P_0}$ equal to $-\chi(P\setminus P_0)$.

2170: Let us start with a generic $P$ and

2171: study how does $\chi(P\setminus P_0)$ changes when we

2172: deform $P$ in the class of Harnack curve.

2173: Since $P$ is a simple Harnack curve we have $P\setminus P_0$ non-singular.

2174: The only singularities of $P$ are the real isolated double points.

2175: Each such point contributes $2$ to the Euler characteristic of the

2176: normalization of $P$, but also $2$ gets subtracted when we remove

2177: this point from $\hat{P}$.

2178: 

2179: Note that $P$ may also have the ``boundary" singularity.

2180: This means that ${P}$ is not transversal to the boundary divisor.

2181: In this case each point of tangency of order $m$ between ${P}$

2182: and $\dd\cT$ gives $m-1$ points in $\hat{P}\setminus P_0$.

2183: Thus, while $\chi(\hat{P})$ gets increased by $m-1$

2184: in the case of such tangency (in comparison with $\chi(\hat{P})$

2185: in the transversal case) in turn $\chi(\hat{P}\setminus P_0)$ gets

2186: decreased by $m-1$. Thus in both cases the net effect

2187: of possible singularities on $\chi(\hat{P}\setminus P_0)$ is zero,

2188: so for the computation of the degree of our covering

2189: we may assume that $\tilde{P}$ is smooth and transversal to

2190: the boundary divisor of $\cT$.

2191: In this case $$\chi(\hat{P}\setminus P_0)=\chi(\hat{P})=\chi(P)=\Area\Delta.$$

2192: The last equality is a corollary of

2193: Khovanskii's formula \cite{Kh}.

2194: %To compute the degree of the covering $\Arg_P$ we note that

2195: %%by Khovanskii's formula \cite{Kh} the Euler characteristic

2196: %%of $\hat{P}$ (that is equal to that of $P$) is $\Area(\Delta_P)$.

2197: %$\chi(\hat{P})$ is equal to $a_P$ since the logarithmic Gauss map

2198: %has no real critical points (see \cite{Mi}).

2199: \end{proof}

2200: 

2201: We may compactify the set-up of Lemma \ref{arg-harnack}

2202: to get the following Theorem describing the alga of

2203: a Harnack curve.

2204: Denote the blow-up of $T$ centered in $0_T$ with

2205: $B:\hat{T}\to T$.

2206: 

2207: Note that $\hat{T}$ is a surface with a natural involution

2208: induced by $\conj$: $(\alpha,\beta)\mapsto (-\alpha,-\beta)$,

2209: here we think of $\alpha,\beta\in S^1$ as arguments of complex numbers.

2210: Clearly, the 1-dimensional part of the

2211: fixed-point set of this involution is the exceptional

2212: divisor of the blow-up. Denote it with $\R\hat{T}$.

2213: 

2214: For the compactifying theorem we need to blow up $\hat{P}$

2215: even further. Recall that $\hat{P}$ is a smooth (real) surface

2216: equipped with an involution coming from complex conjugation.

2217: Note that $0_T$

2218: is the regular value of a map $P\setminus\R P\to T$

2219: and thus $P_0\subset\hat{P}$

2220: is a disjoint union of embedded circles corresponding to the

2221: ovals of $\R\tilde{P}$ and some isolated points.

2222: Let $\Pi$ be the result of (real) blow-up of $\hat{P}$

2223: at $P_0$. Clearly, a blow-up at a smooth submanifold of codimension 1

2224: does not change the surface $\hat{P}$ thus only blowups at

2225: the isolated points of $P_0$ matter.

2226: Such points come either from isolated double points of $P$

2227: or from tangency of $\tilde{P}$ with $\dd\cT$.

2228: %the boundary divisor of $\C T_{\Delta}$.

2229: Note that

2230: $$\chi(\Pi)=\chi(\hat{P}\setminus P_0).$$

2231: 

2232: \begin{theoremA}

2233: There exists a map $p$ completing the commutative diagram

2234: \begin{center}\mbox{}

2235: \xymatrix{

2236: \Pi \ar[r]^{p} \ar[dr]_{\Alga_P} & \hat{T}\ar[d]^{B}\\

2237:   & T}.

2238: \end{center}

2239: The map $p$ is a covering of degree $2\Area(\Delta_P)$

2240: which is equivariant with respect to the involution of complex conjugation

2241: defined on $\Pi$.

2242: %In particular, $$\R\hat{P}=\hat\Arg_P^{-1}(\R\hat{T}).$$

2243: 

2244: Furthermore, any real curve $P$ such that $\Alga_P$ lifts to

2245: a covering $\Pi\to\hat{T}$ is a Harnack curve.

2246: \end{theoremA}

2247: \begin{proof}

2248: The point of $\R\hat{T}$ is specified by a tangent line to $T$ at $0_T$.

2249: The logarithmic Gauss map $\gamma_P$ takes real values

2250: at $\R\hat{P}$. Thus the tangent line at $s\in\R\hat{P}$

2251: is real and gives (after multiplication by $i$)

2252: a tangent direction at $0_T\in T$. We define the value $p(s)$

2253: to be this direction. At points of $P_0\setminus \R\hat{P}$

2254: the map is defined by the blowup itself (recall that $0_T$

2255: is the regular value of the map $P\setminus\R P\to T$).

2256: 

2257: For the converse we note that by \cite{Mi} the amoeba map

2258: $\Log|_P$ does not have critical points outside of $\R P$

2259: since $\Alga_P$ does not have any. Thus $\gamma_P$ cannot

2260: have any critical points on $\R P$.

2261: 

2262: \end{proof}

2263: 

2264: 

2265: \begin{thebibliography}{99}

2266: 

2267: \bibitem{AVGZ}  Arnold, V. I., Gusein-Zade, S. M.,

2268: Varchenko, A. N., {\em Singularities of differentiable maps.

2269: Vol. I. The classification of critical points, caustics and wave fronts.}

2270: Monographs in Mathematics, 82. Birkha"user Boston, Inc., Boston, MA, 1985.

2271: 

2272: \bibitem{Be} Bernstein, D.N.,

2273: {\em The number of roots of a system of equations}.

2274: Functional Anal. Appl. {\bf 9} (1975), 183-185.

2275: 

2276: \bibitem{B} Brieskorn, E., Kn\"orrer, H.,

2277: {\em Ebene algebraische Curven,} Birkh\"auser Verlag, Basel-Boston, 1981.

2278: 

2279: \bibitem{Co} Coolidge, J.L.,

2280: {\em A treatise on algebraic plane curves,}

2281: Dover Publications, Inc., New York 1959

2282: 

2283: \bibitem{CLP}

2284: Cohn, H., Larsen, M., Propp, J.,

2285: \emph{The shape of a typical boxed plane partition},

2286: New York J.\ Math.\ 4 (1998), 137--165.

2287: 

2288: \bibitem{GKZ}  Gelfand I.M., Kapranov M.M., Zelevinsky A.V.,

2289: {\em Discriminants, resultants, and multidimensional determinants}.

2290: Mathematics: Theory \& Applications.

2291: Birkh\"auser Boston, Inc., Boston, MA, 1994.

2292: 

2293: \bibitem{FO} Faber, C., Okounkov, A., unpublished.

2294: 

2295: \bibitem{KeVa} Feng, B., He Y.-H.,

2296: Kennaway K. D., Vafa C.,

2297: {\em Dimer Models from Mirror Symmetry and Quivering Amoebae.}

2298: \texttt{hep-th/0511287}

2299: 

2300: \bibitem{Ha} Harnack, A.,

2301: {\em \"Uber die Vieltheiligkeit der ebenen algebraischen Curven,}

2302: Math. Ann. 10 (1876), 189--199.

2303: 

2304: \bibitem{IV} Itenberg, I., Viro O., {\em Patchworking

2305: algebraic curves disproves the Ragsdale conjecture.}

2306: Math. Intelligencer 18 (1996), no. 4, 19--28.

2307: 

2308: \bibitem{Ka} Kapranov, M.M., {\em A characterization of $A$-discriminantal

2309: hypersurfaces in terms of the logarithmic Gauss map},

2310: Math. Ann. {\bf 290} (1991), 277-285

2311: 

2312: 

2313: \bibitem{KO1}

2314: Kenyon, R., Okounkov, A.,

2315: Planar dimers and Harnack curves,

2316: \texttt{math.AG/0311062}.

2317: 

2318: 

2319: \bibitem{Burg}

2320: Kenyon, R., Okounkov, A.,

2321: Limit shapes and complex Burgers equation,

2322: \texttt{math-ph/0507007}.

2323: 

2324: \bibitem{KOS}

2325: Kenyon, R., Okounkov, A., Sheffield, S.,

2326: Dimers and amoebae,

2327: \texttt{math-ph/0311005}.

2328: 

2329: \bibitem{Kh} Khovanskii, A.G., {\em Newton polyhedra and toric

2330: varieties.} Funktional. Anal. i Prilozhen., {\bf 11} (1977), no. 4, 56--64.

2331: 

2332: \bibitem{Kl-paper} Klein, F., {\em Eine neue Relation

2333: zwischen den Singularit\"aten einer algebraichen Curve},

2334: Math. Ann., {\bf 10} (1876) 199-210.

2335: 

2336: \bibitem{Kl} Klein, F., {\em Elementarmathematik

2337: vom h\"oheren Standpunkte aus. Zweiter Band: Geometrie},

2338: (German) Dritte Auflage. Springer-Verlag, Berlin 1968.

2339: 

2340: 

2341: \bibitem{Ku} Kouchnirenko, A.G.,

2342: {\em Newton polytopes and the Bezout theorem},

2343: Functional Anal. Appl. {\bf 10} (1976), 233-235.

2344: 

2345: \bibitem{Mi} Mikhalkin, G., {\em Real algebraic curves,

2346: the moment map and amoebas.}  Ann. of Math. (2)  151  (2000),  no. 1, 309--326.

2347: 

2348: \bibitem{Mi-obzor} Mikhalkin, G.,

2349: {\em Amoebas of algebraic varieties and tropical geometry.}

2350: Different faces of geometry,  257--300, Int. Math. Ser. (N. Y.),

2351: Kluwer/Plenum, New York, 2004.

2352: 

2353: \bibitem{MR} Mikhalkin, G., Rullg\aa rd, H.,

2354: {\em Amoebas of maximal area.}  Internat. Math. Res. Notices  2001,  no. 9, 441--451.

2355: 

2356: \bibitem{Milnor}  Milnor, J.,

2357: {\em Singular points of complex hypersurfaces},

2358: Annals of Mathematics Studies, No. 61

2359: Princeton University Press, University of Tokyo Press 1968.

2360: 

2361: \bibitem{Pa} Passare, M.,

2362: {\em Amoebas, convexity and the volume of integer polytopes.}

2363: Complex analysis in several variables---Memorial Conference of

2364: Kiyoshi Oka's Centennial Birthday,  263--268, Adv. Stud. Pure Math.,

2365: 42, Math. Soc. Japan, Tokyo, 2004.

2366: 

2367: \bibitem{PR}  Passare, M., Rullg\aa rd, H.,

2368: {\em Amoebas, Monge-Ampe`re measures, and triangulations of the Newton polytope.}

2369: Duke Math. J.  121  (2004),  no. 3, 481--507.

2370: 

2371: \bibitem{Pl} Pl\"ucker, J., {\em Solution d'une question fondamentale

2372: concernant la th\'eorie generale des courbes}, J. Reine Angew. Math.,

2373: {\bf 12} (1834), 105-108.

2374: 

2375: \bibitem{V-usp} Viro O.Ya., private communications.

2376: %{\em Generalization of Klein's formula

2377: %and wave fronts}, Uspekhi Mat. Nauk.

2378: 

2379: \bibitem{V} Viro O.Ya.,

2380: {\em Some integral calculus based on Euler characteristic},

2381: Lecture Notes in Math., 1346 (1988), Springer, Berlin, 127-138.

2382: 

2383: 

2384: \bibitem{W} Wall, C.T.C,

2385: \emph{Singular points of plane curves},

2386: Cambridge University Press, 2004.

2387: 

2388: 

2389: \end{thebibliography}

2390: 

2391: 

2392: \end{document}

2393: 

2394: % LocalWords:  Grigory Mikhalkin Okounkov PQeq PDE az bw defR Propp zw xy Gau

2395: % LocalWords:  hypersurface nonreduced ucker Viro ij jk toric abelian Milnor dr

2396: % LocalWords:  gam Kouchnirenko's Pick's Hurwitz Minkowski lan polRes dl diag

2397: % LocalWords:  nondegenerate branchpoints chiR quadrics defadd ramphoid FG cS

2398: % LocalWords:  adjunction cuspidal functorial nonconstant eqRh obner PQeps

2399: % LocalWords:  branchpoint coordx chiRh

2400: