1: \documentclass[11pt]{amsart}

2: \setlength{\textheight}{660pt}

3: \setlength{\textwidth}{450pt}

4: \setlength{\topmargin}{-10pt}

5: \setlength{\footskip}{30pt}

6: \setlength{\oddsidemargin}{20pt}

7: \setlength{\evensidemargin}{-15pt}

8: \usepackage{amsmath,amssymb,mathrsfs,amscd}

9: \usepackage{graphicx}

10: \usepackage[all]{xy}

11: 

12: \pagestyle{plain}

13: \title{Double solid twistor spaces: the case of arbitrary signature}

14: 

15: \author{Nobuhiro Honda}

16: \thanks

17: {$^{\dag}$This work was partially supported by

18: Research Fellowships of the 

19: Japan Society for the Promotion

20: of Science for Young Scientists.\\

21: {\it{Mathematics Subject Classifications}} (2000) 32L25, 32G05, 32G07, 53A30,

22: 53C25\\

23: {\it{Keywords}}\ \  twistor space, minitwistor space, Moishezon manifold, conic bundle, double covering, bimeromorphic transformation,  self-dual metric}

24: \date{}

25: 

26: \newcommand{\ol}{\overline}

27: \newcommand{\ra}{\rightarrow}

28: \newcommand{\lra}{\longrightarrow}

29: \newcommand{\da}{\downarrow}

30: \newcommand{\Ra}{\Rightarrow}

31: \newcommand{\set}{\,|\,}

32: \newcommand{\proofend}{\hfill$\square$}

33: \newcommand{\linfty}{\langle l_{\infty}\rangle}

34: \newcommand{\vphi}{\varphi}

35: \newcommand{\bsl}{\backslash}

36: 

37: \newtheorem{prop}{Proposition}[section]

38: \newtheorem{lemma}[prop]{Lemma}

39: \newtheorem{claim}[prop]{Claim}

40: \newtheorem{thm}[prop]{Theorem}

41: \newtheorem{rmk}[prop]{Remark}

42: \newtheorem{cor}[prop]{Corollary}

43: \newtheorem{example}[prop]{Example}

44: \newtheorem{definition}[prop]{Definition}

45: \newtheorem{condition}[prop]{Condition}

46: \newtheorem{question}[prop]{Question}

47: 

48: \begin{document}

49: 

50: \begin{abstract}

51: In a recent paper (\cite{Hon07-3}) we constructed a series of new Moishezon twistor spaces  which are a kind of variant of the famous LeBrun twistor spaces.

52: In this paper we explicitly give  projective models of another series of Moishezon twistor spaces on 

53: $n\mathbf{CP}^2$ for arbitrary $n\ge 3$, which can be regarded as a

54: generalization of the twistor spaces of  `double solid type' on $3\mathbf{CP}^2$ studied by Kreu\ss ler, Kurke, Poon and the author.

55: Similarly to the twistor spaces of `double solid type' on $3\mathbf{CP}^2$, 

56:  projective models of the present twistor spaces have a natural structure of  double covering of a $\mathbf{CP}^2$-bundle over $\mathbf{CP}^1$.

57: We explicitly give a defining polynomial of the branch divisor of the double covering, whose restriction to fibers is degree four.

58: If $n\ge 4$ these are new  twistor spaces, to the best of the author's knowledge.

59: We also compute the dimension of the moduli space of these twistor spaces.

60: Differently from \cite{Hon07-3}, the present investigation is based on analysis of pluri-(half-)anticanonical systems of the twistor spaces.

61: \end{abstract}

62: 

63: \maketitle

64: 

65: \bigskip\noindent

66: \section{ Introduction}

67: 

68: The Weyl curvature tensor of a Riemannian metric on a manifold   is invariant under changes of the metric by multiplying a function.

69: When the manifold is more than three-dimensional, it measures how the metric is distant from a conformally flat metric.

70: If the manifold is four-dimensional and oriented, the Weyl curvature tensor decomposes into two components, called self-dual and anti-self-dual part.

71: A metric on an oriented four-manifold is called \!{\it self-dual}\, if the anti-self-dual part of the Weyl curvature vanishes identically.

72: A significant property of self-dual metrics is that, by so called a Penrose correspondence, there naturally associates a very special kind of a complex threefold, called the {\em twistor space}.

73: As a differential manifold, it is a sphere bundle over the four-manifold.

74: Although its fibers  are complex submanifolds of the twistor space, their normal bundles are non-trivial but the same as that of line in $\mathbf{CP}^3$.

75: 

76: Nowadays thanks to a theorem of C.\,Taubes \cite{T92},  it is known that there exist a plenty of compact manifolds which admit a self-dual metric.

77: More precisely, any oriented compact 4-manifold admits a self-dual metric, after gluing  sufficiently many complex projective planes.

78: Accordingly, there exist a huge number of compact twistor spaces.

79: N.\,Hitchin \cite{Hi81} showed that compact twistor spaces do not admit a K\"ahler metric, except two well-known examples.

80: This result looks  made it difficult to investigate compact twistor spaces by usual methods in complex algebraic geometry.

81: But Y.\,S.\,Poon \cite{P86} and C.\,LeBrun \cite{LB91} changed the scene dramatically by constructing explicit examples of Moishezon twistor spaces.

82: (A compact complex manifold is called Moishezon if it is bimeromorphic to a projective manifold.)

83: Later, F.\,Campana and B.\,Kreu\ss ler \cite{CK98} found a new series of Moishezon twistor spaces and described their structure in a quite detailed form.

84: The base 4-manifolds of all these twistor spaces are $n\mathbf{CP}^2$, the connected sums of $n$ copies of complex projective planes.

85: This is partly because  by a theorem of Campana \cite{C91}, if a compact four-manifold admits a self-dual metric whose twistor space is Moishezon (or of Fujiki's class $\mathscr C$, more strongly), then the four-manifold is (homeomorphic to) the four-sphere or $n\mathbf{CP}^2$.

86: 

87: In \cite{Hon07-3}, the author explicitly constructed a family of Moishezon twistor spaces with $\mathbf C^*$-action on $n\mathbf{CP}^2$ for arbitrary $n\ge 2$, which seem to be new in the case $n\ge 4$.

88: The main tool for the construction was the anticanonical system of the twistor spaces.

89: In fact, the image of its associated meromorphic map is a (singular) rational surface whose structure is independent of the number $n$ of connected sum.

90: Further, the meromorphic map can be regarded as a quotient map by the $\mathbf C^*$-action.

91: Namely the image surface was so called a minitwistor space.

92: By analyzing the structure of the meromorphic quotient map, we constructed the twistor spaces on $n\mathbf{CP}^2$ as explicit bimeromorphic modifications of  (again explicitly constructed) conic bundles over (the minimal resolution of) the minitwistor spaces.

93: 

94: 

95: In this paper we provide yet another family of Moishezon twistor spaces on $n\mathbf{CP}^2$ for arbitrary $n\ge 3$.

96: As those in \cite{Hon07-3} explained above, these twistors have a $\mathbf C^*$-action induced by the $U(1)$-symmetries of the corresponding self-dual metrics.

97: Further, when $n=3$, they coincide with those in \cite{Hon07-3}.

98: (Namely they  coincide with non-LeBrun twistor spaces on $3\mathbf{CP}^2$ with $\mathbf C^*$-action studied in \cite{Hon07-2}.)

99: But when $n\ge 4$, they are new twistor spaces, to the best of the author's knowledge.

100: %A basic reason for this is that the associated minitwistor spaces have different structures depending on $n$, and all of them are new and different from the minitwistor spaces in \cite{Hon07-3}, provided $n\ge 4$.

101: Among other properties, the most characteristic feature of the present twistor spaces is that {\em their projective models have a natural structure of double covering over some $\mathbf{CP}^2$-bundle over $\mathbf{CP}^1$, 

102: as a natural generalization of the double covering structure of the twistor spaces on $3\mathbf{CP}^2$ studied in \cite{Hon07-2, KK92, P92}.}

103: (On the other hand, the twistor spaces we constructed in \cite{Hon07-3} were also generalization of non-LeBrun twistor spaces on $3\mathbf{CP}^2$ with $\mathbf C^*$-action.

104: But they are different generalization respecting the structure of minitwistor spaces.)

105: 

106: 

107: %More precisely, a pluri-(half-)anticanonical system on the twistor space on $n\mathbf{CP}^2$ gives a meromorphic map which is generically 2 to 1 over its image.

108: %The image is explicitly given as follows.

109: %By considering the set of planes in $\mathbf{CP}^n$ containing a (fixed) line, we obtain  a projection from $\mathbf{CP}^n$ to $\mathbf{CP}^{n-2}$ whose fibers are the planes and whose center is the line.

110: %Its inverse image over a rational normal curve in $\mathbf{CP}^{n-2}$ is 

111: %a (rational) threefold which has cyclic quotient singularties along the line.

112: %This is exactly the image of the generically 2 to 1 map from the twistor spaces.

113: %%When $n=3$, the threefold becomes $\mathbf{CP}^3$ itself

114: %Blowing up the image threefold along the line, we obtain the total space of  the $\mathbf{CP}^2$-bundle $\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)$ over $\mathbf{CP}^1$.

115: %Projective models of the twistor spaces have a structure of double covering over this $\mathbf{CP}^2$-bundle, whose branch divisor has degree four along fibers of the $\mathbf{CP}^2$-bundle.

116: %Hence the present twistor spaces can be regarded as a generalization of  non-LeBrun twistor spaces on $3\mathbf{CP}^2$ with $\mathbf C^*$-action, respecting the structure of double covering.

117: %In fact, projective models of the latter twistor spaces (on $3\mathbf{CP}^2$)  have 

118: %a natural structure of  a double covering of $\mathbf{CP}^3$ whose branch divisor is a quartic surface, and  if blow-up $\mathbf{CP}^3$ along a line then we obtain a $\mathbf{CP}^2$-bundle $\mathbf P(\mathscr O(1)^{\oplus 2}\oplus \mathscr O)$ over $\mathbf{CP}^1$.

119: %These are exactly the situation when we put $n=3$ in the above explanation.

120: 

121: %

122: %Unfortunately it is not easy to prove the existence of pluri-half-anticanoncial divisors on the twistor spaces which guarantee that the pluri-half-anticanonical system induces a generically 2 to 1 map over the images.

123: %So here we outline how we prove it.

124: %The pluri-half-anticanonical system we will consider is the system $|(n-2)F|$, where $F:=K^{-1/2}.$ 

125: %The system has an `obvious' subsystem composed by the pencil $|F|$, and we call its 

126: %members as `trivial members'.

127: %The point is to find a `non-trivial' member of the system $|(n-2)F|$.

128: %It seems to be difficult to do this if we consider the system alone.

129: %So we first study a (non-complete) linear system $|(n-1)F|^{\mathbf C^*}$ consisting of divisors defined by $\mathbf C^*$-invariant sections of the line bundle $(n-1)F$.

130: %The meromorphic map associated to this system will turned out to give a quotient map from the twistor space to a minitwistor space (by the $\mathbf C^*$-action.)

131: %Further we will find an irreducible curve in the minitwistor space whose inverse image splits into two irreducible components.

132: %Then adding some divisors to each of the irreducible components, we finally obtain the required member of the system $|(n-2)F|$.

133: 

134: We outline how these results are obtained, by explaining contents of each sections.

135: %In Section 2 we investigate structures of the minitwistor spaces associated to the present twistor spaces with $\mathbf C^*$-action in detail.

136: In \S \ref{ss-S} we precisely explain what kind of twistor spaces we shall consider, by specifying the structure of a divisor $S$ which is a member of the half-anticanonical system of the twistor space.

137: In \S \ref{ss-multanS} we state properties of some pluri-anticanonical systems of the surface $S$.

138: They will be a basis for the following investigation.

139: In \S \ref{ss-lst} we consider pluri-half-anticanonical systems on the twistor spaces and provide basic commutative diagrams of meromorphic maps which are indispensable for analyzing the structure of the present twistor spaces.

140: We also prove the first key result (Lemma \ref{lemma-nt1}) that the sum of some degree-one divisors belongs to a system $|(n-1)F|^{\mathbf C^*}$, the subsystem of $|(n-1)F|$ 

141: consisting of divisors  defined by $\mathbf C^*$-invariant sections of $(n-1)F$, where $F$ denotes the canonical half-anticanonical bundle of the twistor spaces.

142: Next in \S \ref{ss-mt} we determine the image of the meromorphic map associated to this system in explicit form (Theorem \ref{thm-mt1}).

143: In particular, we show that it is a normal rational surface in $\mathbf{CP}^{n+1}$ whose degree is $2(n-1)$.

144: The meromorphic map can be regarded as a (meromorphic) quotient map of the $\mathbf C^*$-action:

145: thus the image surface is a minitwistor space.

146: We also determine its singularities and give their minimal resolutions (Prop.\,\ref{prop-mt2}).

147: In contrast to the situation we obtained in \cite{Hon07-3},

148: the structure of these minitwistor spaces depend on $n$, and they form $(n-2)$-dimensional moduli.

149: When $n=3$, they coincides with the minitwistor space studied in \cite{Hon-p-1} and \cite{Hon07-3}.

150: When $n\ge 4$, they are new minitwistor spaces, as far as the author knows.

151: 

152: In Section 3 by analyzing the meromorphic quotient map in the previous section we realize  projective models of the twistor spaces as  conic bundles over  (the minimal resolution of) the minitwistor spaces.

153: In \S \ref{ss-pr} we concretely give a partial elimination of the indeterminacy locus of the meromorphic map.

154: Then in \S \ref{ss-dl} and \S \ref{ss-nt} we explicitly construct a $\mathbf{CP}^2$-bundle over the resolution of the  minitwistor space and realize a projective model of the twistor space as a conic subbundle of this $\mathbf{CP}^2$-bundle, by explicitly giving its defining equation.

155: The idea used here is similar to those in \cite{Hon07-3}, Section 3.

156: An interesting observation is that, as in \cite{Hon07-3}, a discriminant curve of the conic bundle is a hyperplane section of the minitwistor spaces with respect to its natural realization in a projective space (Lemma \ref{lemma-hps1}).

157: We show that the inverse image of this discriminant curve splits into two irreducible components and correspondingly we obtain a pair of (mutually conjugate) $\mathbf C^*$-invariant divisors $Y$ and $\ol{Y}$ in the twistor space (Prop.\,\ref{prop-Y1}).

158: We next determine the cohomology classes of $Y$ and $\ol{Y}$ (in $H^2(Z,\mathbf Z)$) by showing that  adding some divisor to $Y$ (and $\ol{Y}$) gives  members of the system $|(n-2)F|$ (Prop.\,\ref{prop-nt4}).

159: Existence of these members is a key in obtaining the following presentations of the twistor spaces.

160: 

161: In Section 4 we investigate a complete linear system $|(n-1)F|$.

162: Based on the results of the previous section,  especially the existence of the above divisors in  the system $|(n-2)F|$, we concretely give generators of $|(n-1)F|$.

163: Then using them we explicitly determine the image of the meromorphic map associated to $|(n-1)F|$ (Theorem \ref{thm-bim}).

164: We also show that the map is bimeromorphic onto the image.

165: In particular, we obtain the second projective models of the twistor spaces.

166: They are birational to codimension two subvarieties in certain $\mathbf{CP}^4$-bundle over $\mathbf{CP}^1$, whose restriction to each fibers are quartic surfaces.

167: 

168: In the final section, combining results obtained so far, we investigate the meromorphic map associated to the system $|(n-2)F|$.

169: We show that the map is generically 2 to 1 over its image, and  the image is a rational scroll of planes in $\mathbf{CP}^n$, whose degree is $n-2$.

170: By blowing-up its vertices which form a line, we obtain a $\mathbf{CP}^2$-bundle $\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus \mathscr O)$ over $\mathbf{CP}^1$.

171: Then we explicitly give a meromorphic map from the projective model in Section 4 to this $\mathbf{CP}^2$-bundle, and show that the map is generically 2 to 1.

172: We further  determine the defining equation of the branch locus of this map in an explicit form.

173: These results are summarized as Theorem \ref{thm-dc2}.

174: When $n=3$, the equation of the branch divisor coincides with the one we obtained in \cite{Hon07-2} for a non-LeBrun twistor space on $3\mathbf{CP}^2$ with $\mathbf C^*$-action.

175: Thus it becomes apparent that {\em the present twistor spaces are natural generalization of the twistor spaces on $3\mathbf{CP}^2$ studied in \cite{Hon07-2}, respecting the structure of double covering.}

176: We also show that the branch divisor of the double covering is irreducible, non-normal surface, and birational to ruled surface of genus $[(n-1)/2]$ (Prop.\,\ref{prop-B}).

177: We further obtain some constraint on the defining equation of the branch divisor (Prop.\,\ref{prop-dr}), which is a generalization of the constraint appeared in the case of $n=3$ in \cite{Hon07-2}.

178: Consequently  the moduli space of the present twistor spaces can be computed to be $n$-dimensional.

179: Finally we remark that the present twistor spaces can be obtained as a small deformation of LeBrun metrics with torus action which preserves a particular $U(1)$-action, and compare with the results we obtained in \cite{Hon07-1}.

180: 

181: In summary, we investigate the twistor spaces by using three linear systems:

182: (a) the non-complete system $|(n-1)F|^{\mathbf C^*}$ from which we derive a conic bundle description over the minitwistor spaces (these are studied in Sections 2 and 3),

183: (b) the complete linear system $|(n-1)F|$ which gives projective models that are  birational to the total space of a fiber space over  $\mathbf{CP}^1$ whose fibers are quartic surfaces

184: (these are studied in Section 4),

185: (c) the complete linear system $|(n-2)F|$ which gives a generically 2 to 1 covering over a rational scroll of planes in $\mathbf{CP}^n$ (these are studied in Section 5).

186: Of course, the last presentation is most simple because the double covering is uniquely determined by only specifying the branch divisor which is expressed by a {\em single }polynomial.

187: It seems difficult for the author to obtain the last presentation without studying both (a) and (b) in detail (as we did in this paper).

188: 

189: 

190: 

191: 

192: \bigskip\noindent

193: {\bf Notations and Conventions.}

194: As in \cite{Hon07-3}, to save  notations we adapt the following convention.

195: If $\mu:X\to Y$ is a bimeromorphic morphism of  complex variety and $W$ is a complex subspace in $X$, we write $W$ for the image $\mu(W)$  if the restriction $\mu|_W$ is still bimeromorphic.

196: Similarly, if $V$ is a complex subspace of $Y$, we write $Y$ for the strict transform of $Y$.

197: %If $D$ is a divisor on a variety $X$, $[D]$ means the associated line bundle.

198: If $D$ is a divisor on a variety $X$, the dimension of a complete linear system $|D|$ always means $\dim H^0(X,[D])-1$.

199: The base locus is denoted by ${\rm Bs}\,|D|$.

200: If a Lie group $G$ acts on $X$ by means of biholomorphism and $D$ is $G$-invariant,

201: $G$ naturally acts on the vector space $H^0(X,[D])$.

202: Then $H^0(X,[D])^{G}$ means the subspace of $G$-invariant sections.

203: Further $|D|^G$ means its associated linear system.

204: %(So all members of $|D|^G$ are $G$-invariant.)

205: If $s$ is a non-zero section of a holomorphic line bundle on $X$, $(s)$ denotes the divisor defined as the zero locus of $s$.

206: When discussing cohomology classes represented by  complex curves $C_1$ and $C_2$ on a complex surface $S$,

207: we write $C_1\sim C_2$ to mean that they are cohomologous; namely if  $C_1$ and $C_2$ determine the same element in $H^2(S,\mathbf Z)$.

208: (This is used only in Section 3.)

209: 

210: If $Z$ is a twistor space, $F$ always denotes the canonical square root of the anticanonical line bundle of $Z$ (often called the `fundamental line bundle').

211: The degree of a divisor on $Z$ means its intersection number with twistor lines.

212: 

213: 

214: 

215: \section{Analysis of the structure of minitwistor spaces}

216: \subsection{Construction of the surface $S$ contained in the fundamental system}

217: \label{ss-S}

218: As in \cite{Hon07-3}, we start our investigation by specifying structure of a real irreducible member of the fundamental system on  the twistor spaces.

219: 

220: 

221: So we consider the product surface $\mathbf{CP}^1\times\mathbf{CP}^1$ equipped with a real structure given by 

222: \begin{align}\label{rs1}

223: \text{(complex conjugation)}\times\text{(anti-podal)}.

224: \end{align}

225: We write $\mathscr O(1,0)=p_1^*\mathscr O(1)$ and $\mathscr O(0,1)=p_2^* \mathscr O (1)$, where $p_i$ denotes the projection to the $i$-th factor.

226: We choose a non-real curve $C_1\in|\mathscr O(1,0)|$ and a point $P_1\in C_1$.

227: Next we blow-up $\mathbf{CP}^1\times\mathbf{CP}^1$ at $P_1$ and $\ol{P}_1$ to obtain a surface with $c_1^2=6$, which has distinguished curves $C_1$ and $\ol{C}_1$ satisfying $C_1^2=\ol{C}_1^2=-1.$

228: Subsequently we blow-up the surface at the intersection points of $C_1\cup\ol{C}_1$ and the exceptional curves of the last blow-ups.

229: Then we obtain a surface with $c_1^2=4$ possessing distinguished curves $C_1$ and $\ol{C}_1$ satisfying $C_1^2=\ol{C}_1^2=-2$.

230: By repeating this procedure $(n-1)$-times, we obtain a surface $S'$ with $c_1^2=8-2(n-1)$ possessing distinguished curves $C_1$ and $\ol{C}_1$ satisfying $C_1^2=\ol{C}_1^2=1-n$.

231: This surface $S'$ has an obvious structure of  toric surface.

232: Then as a final operation to obtain the required surface, we choose a conjugate pair of points which are on the exceptional curves of the final blow-ups (in obtaining the surface $S'$) but which are {\em not}  fixed points of the torus action.

233: Then the action of the torus is killed and 

234: consequently we obtain a non-toric surface $S$ with a non-trivial $\mathbf  C^*$-action

235: which satisfies $c_1^2=8-2n$.

236: The surface $S$ has a natural real structure induced from \eqref{rs1}.

237: Let $B_1$ and $\ol{B}_1$ be the exceptional curves of the final blowing-up $S\to S'$.

238: By choosing the pair of blown-up point generically,

239: we can suppose that there exists no $\mathbf C^*$-invariant $(-2)$-curve intersecting both of $B_1$ and $\ol{B}_1$.

240: Namely we suppose that the pair of blown-up points on $S'$ do not belong to the same $\mathbf C^*$-orbit closure.

241: Then there exists a unique pair of $(-1)$-curves $B_2$ and $\ol B_2$ intersecting $\ol B_1$ and $B_1$ transversally respectively.

242: We note that only freedom involved in the construction of the surface $S$ is the choice of the final blown-up points on $S'$.

243: 

244: \begin{figure}

245: \includegraphics{ds_cycle1_small.eps}

246: \caption{Construction of the surface $S$ in the case $n=4$.}

247: \label{fig-cycle1}

248: \end{figure}

249: 

250: 

251: \begin{figure}

252: \includegraphics{ds_cycle2_small.eps}

253: \caption{The anticanonical cycle in the surface $S$.}

254: \label{fig-cycle2}

255: \end{figure}

256: 

257: It is readily seen that if $n\ge 4$ the anticanonical system of the surface $S$ consists of a unique member and it is a cycle of $(2n+2)$ smooth rational curves containing all exceptional curves in obtaining $S'$.

258: We  write $C$ for this anticanonical cycle (analogously to the notation in \cite{Hon07-3}) and let

259: \begin{align}

260: C=\sum_{i=1}^{n+1} C_i+\sum_{i=1}^{n+1}\ol{C}_i

261: \end{align}

262: be its decomposition into  irreducible components orderly named as in Figure \ref{fig-cycle2}.

263: Here, among these, only $C_2$ and $\ol{C}_2$ are  $\mathbf C^*$-fixed.

264: Note that the pair $B_1$ and $\ol{B}_1$ of  the exceptional curves of the final blow-ups $S\to S'$ are not contained in this cycle $C$.

265: These intersect $C_2$ and $\ol C_2$ respectively.

266: By construction

267: it is easy  to see that the self-intersection numbers of the irreducible components are given by

268: \begin{align}

269: C_1^2=1-n,\,\,\, C_2^2=C_3^2=\cdots=C_n^2=-2,\,\,\, C_{n+1}^2=-1.

270: \end{align}

271: Of course, we have $\ol{C}_i^2=C_i^2$ for any $i$.

272: Note that if $n=3$ the construction of the  surface $S$ is the same for that of the surface $S$ treated in \cite{Hon07-3}.

273: If $n\ge 4$ the present surface is different from \cite{Hon07-3}, since there the unique anticanonical curve was the cycle consisting of 8 irreducible components only.

274: %This difference will be reflected to  structural differences of the twistor spaces.

275: 

276: 

277: \subsection{Properties of some pluri-anticanonical systems of $S$}\label{ss-multanS}

278: We wish to study  algebro-geometric structure of twistor spaces which contain the surface $S$ as its real irreducible member of the fundamental system $|F|$.

279: Before doing this, in this subsection, we collect basic properties of some pluri-anticanonical systems of $S$.

280: They will be a basis of our analysis on the multi-systems $|mF|$ on the twistor spaces.

281: 

282: \begin{prop}\label{prop-multan1} Suppose $n\ge 4$ and 

283: let $S$ be the surface constructed above, and $C$ the anticanonical cycle whose irreducible components are named orderly as in Figure \ref{fig-cycle2}. Then we have the following.

284: (i) $\dim H^0(mK^{-1}_S)=1$ for any $1\le m\le n-3$.

285: (ii) $\dim H^0((n-2)K^{-1}_S)=3$ and $\dim H^0((n-2)K^{-1}_S)^{\mathbf C^*}=1$.

286: (iii) The fixed component of the net $|(n-2)K^{-1}_S|$ is given by

287: \begin{align}\label{fixcompo1}

288: (n-3)\sum_{i=1}^4(C_i+\ol{C}_i)+\sum_{i=5}^n (n+1-i)(C_i+\ol{C}_i).

289: \end{align}

290: (iv) The  movable part of the net $|(n-2)K_S^{-1}|$ does not have base points.

291: (v) The associated morphism $S\to\mathbf{CP}^2$ is the composition of the following 2 morphisms: the first one is a succession of blowing-downs of $C_{n+1}, C_n,\cdots, C_6, C_5$ and their conjugates, and the second one is a generically 2 to 1 map from the resulting surface to $\mathbf{CP}^2$. 

292: Further, the branch curve of the second map is a sum of two different $\mathbf C^*$-invariant irreducible conics, where $\mathbf C^*$ acts on $\mathbf{CP}^2$ by 

293: \begin{align}\label{action49}

294: (y_0,y_1,y_2)\longmapsto (y_0,sy_1,s^{-1}y_2), \,\,\,s\in\mathbf C^*

295: \end{align}

296: in a homogeneous coordinate.

297: \end{prop}

298: 

299: Although these are not straightforward to see, 

300: we omit a proof since they can be shown by standard arguments and computations in algebraic geometry.

301: Note that by (v) the morphism $S\to\mathbf{CP}^2$ associated to the movable part of the system $|(n-2)K_S^{-1}|$ is generically 2 to 1, and its discriminant locus, consisting of two $\mathbf C^*$-invariant conics,  has two $A_3$-singularities at the intersection points.

302: Hence the double covering in the usual sense has two $A_3$-singularities over there.

303: The components $C_1,C_2, C_3$ and their conjugates arise as the exceptional curves of their minimal resolutions, and 

304: $C_4$ and $\ol{C}_4$ are mapped biholomorphically to the line $\{y_0=0\}$ in the coordinate of \eqref{action49}.

305: Thus the morphism $S\to\mathbf{CP}^2$ factors as

306: \begin{equation}\label{factor1}

307: S\lra \tilde{S}_0\lra S_0\lra \mathbf{CP}^2,

308: \end{equation}

309: where $S_0\to\mathbf{CP}^2$ is the double covering branched along

310: the sum of two $\mathbf C^*$-invariant conics, $\tilde{S}_0\to S_0$ is the minimal resolution of the two $A_3$-singularities, and 

311: $S\to \tilde{S}_0$ is the contraction of $C_{n+1}, C_n,\cdots, C_6, C_5$ and their conjugates.

312: Thus the structure of $\tilde S_0$ is independent of $n$.

313: Note that by (ii) of Prop.\,\ref{prop-multan1} non-zero section of $H^0((n-2)K_S^{-1})^{\mathbf C^*}$ is unique up to constant.

314: Its zero divisor is exactly the non-reduced curve $(n-2)C$.

315: 

316: Note also that Prop.\,\ref{prop-multan1} holds even when $n=3$.

317: In that case (i)  claims nothing and (ii) means $|K_S^{-1}|$ is a net.

318: Further

319:  every components of \eqref{fixcompo1} disappear (since $C_5,C_6,\cdots$ do not exist) and hence the anticanonical system becomes free.

320: Moreover $S=\tilde{S}_0$ holds in \eqref{factor1}.

321: This is exactly the situation we encountered in \cite{Hon07-2}, where we used the double covering map to detect  twistor lines.

322: 

323: We will also need the following results concerning the system $|(n-1)K^{-1}_S|$.

324: 

325: \begin{prop}\label{prop-multan2}

326: Let $S$ be as in Prop.\,\ref{prop-multan1}. Then we have the following.

327: (i) $\dim H^0((n-1)K^{-1}_S)=5$ and $\dim H^0((n-1)K^{-1}_S)^{\mathbf C^*}=3$.

328: (ii) The fixed component of the system $|(n-1)K^{-1}_S|$ is given by

329: \begin{align}

330: (n-2)\left(C_1+\ol{C}_1+C_2+\ol{C}_2\right)+\sum_{i=3}^n(n-i+1)(C_i+\ol{C}_i).

331: \end{align}

332: (iii) The movable part of the 4-dimensional system $|(n-1)K^{-1}_S|$ does not have base point.

333: (iv) The associated morphism $S\to\mathbf{CP}^4$ is birational onto its image, and the image is a quartic surface.

334: (v) The meromorphic map associated to the net $|(n-1)K_S^{-1}|^{\mathbf C^*}$ is a morphism and its image is a conic in $\mathbf{CP}^2$.

335: Further, its general fibers are smooth rational curves intersecting $C_2$ and $\ol{C}_2$.

336: \end{prop}

337: 

338: Again we do not write a proof of this proposition.

339: In short, these two propositions mean that the system $|(n-2)K_S^{-1}|$ gives a generically 2 to 1 morphism onto $\mathbf{CP}^2$ and $|(n-1)K^{-1}_S|$ gives a birational morphism onto a quartic surface in $\mathbf{CP}^4$.

340: 

341: 

342: %Next let $Z$ be a twistor space on $n\mathbf{CP}^2$ with $n\ge 3$ and suppose that $Z$ contains $S$ as its real member of the system $|F|$, where $F$ denotes the line bundle $(-1/2)K_Z$ as in Convention.

343: %Then by the cohomology exact sequence of 

344: %the exact sequence 

345: %\begin{equation}\label{ses1}

346: %0\lra \mathscr O_Z\lra F\lra K^{-1}_S\lra 0,

347: %\end{equation}

348: %we obtain an exact sequence

349: %\begin{equation}\label{ses2}

350: %0\lra H^0(\mathscr O_Z)\lra H^0(F)\lra H^0(K^{-1}_S)\lra 0,

351: %\end{equation}

352: %since we have  $H^1(\mathscr O_Z)=0$ by the  simply connectedness of the base 4-manifold \cite{LB92}.

353: %By Prop.\,\ref{prop-multan1}, this means that if $n\ge 4$ the system $|F|$ is a pencil and its base locus is the anticanonical cycle $C$.

354: %(If $n=3$, $|F|$ is not a pencil but $|F|^{\mathbf C^*}$ becomes a pencil.)

355: %Then we have the following proposition which describes the set of reducible members of the pencil.

356: 

357: \subsection{Analysis of linear systems on the twistor spaces}\label{ss-lst}

358: With these preliminary results on the structure of the surface $S$, we begin our study on twistor spaces. 

359: The following proposition is about the structure of the system $|F|$ and easy to see.

360: 

361: \begin{prop}\label{prop-fundamental}

362: Suppose $n\ge 4$ and let $Z$ be a twistor space on $n\mathbf{CP}^2$ with $\mathbf C^*$-action which is compatible with the real structure.

363: Suppose that the complex surface $S$ in Prop.\,\ref{prop-multan1}( and \ref{prop-multan2})

364: is equivariantly contained in $Z$ as a real $\mathbf C^*$-invariant member of the system $|F|$.

365: Then we have the following.

366: (i) $\dim |F|=\dim |F|^{\mathbf C^*}=1$ and {\rm Bs}\,$|F|=C$.

367: (ii) The pencil $|F|$ has precisely $(n+1)$ reducible members, and they are of the form $S_i^++S_i^-$ $(1\le i\le n+1$), where $S_i^+$ and $S_i^-$ are mutually conjugate, $\mathbf C^*$-invariant divisors  of  degree one.

368: (iii) $L_i:=S_i^+\cap S_i^-$ $(1\le i\le n+1)$ are $\mathbf C^*$-invariant twistor lines joining conjugate pairs of singular points of the cycle $C$.

369: \end{prop}

370: 

371: %We again omit a proof of this proposition.

372: We name the reducible members  $S_i^++S_i^-$ $(1\le i\le n+1$) by imposing that the invariant twistor line $L_i=S_i^+\cap S_i^-$ intersects $C_{i-1}\cap C_i$, where $C_0=\ol{C}_{n+1}$.

373: Further, we distinguish the 2 irreducible components of $S_i^++S_i^-$ by declaring that $S_i^+$ contains the component $C_{n+1}$.

374: %(Namely, in Figure \ref{}, the left half  `divided by $L_i$' is $S_i^+$.)

375: 

376: The existence of the surface $S$  in the twistor space has the following consequence on the induced $U(1)$-action on the base 4-manifold.

377: 

378: \begin{prop}\label{prop-u1}

379: Let $Z$ and $S$ be as in Prop.\,\ref{prop-fundamental}.

380: Then the induced $U(1)$-action on $n\mathbf{CP}^2$ satisfies the following.

381: (i) $U(1)$-fixed locus consists of one sphere and $n$ isolated points.

382: (ii) There exists a unique isolated $U(1)$-fixed point whose twistor line is fixed by $\mathbf C^*$-action.

383: (iii) There exists a unique isolated $U(1)$-fixed point whose isotropy subgroup of its twistor line is a cyclic subgroup of $U(1)$ (or $\mathbf C^*$) with order $(n-2)$.

384: (iv) The isotropy subgroup of all other invariant twistor lines are identity.

385: \end{prop}

386: 

387: \noindent

388: Proof.

389: The $U(1)$-fixed sphere in (i) is the image of the $\mathbf C^*$-fixed curve $C_2$ by the twistor fibration $Z\to n\mathbf{CP}^2$.

390: The twistor line in (ii) is the one going through a $\mathbf C^*$-fixed point $B_1\cap B_2$, where $B_1$ and $B_2$ are $(-1)$-curves in $S$ obtained in the construction of $S$.

391: The twistor line in (iii) is $L_1$.

392: Since the intersection point $C_i\cap C_{i+1}$ ($3\le i\le n$ ) are isolated $\mathbf C^*$-fixed points on $S$, $L_i$ ($4\le i\le n+1)$ are over the isolated $U(1)$-fixed points on $n\mathbf{CP}^2$.

393: Thus we obtain $n$ isolated $U(1)$-fixed points as in (i).

394: (iv) can also be verified by computing the induced $U(1)$-action on the tangent spaces of $n\mathbf{CP}^2$ at the $U(1)$-fixed points and then computing the induced action on the spin bundles.

395: \proofend

396: 

397: \bigskip

398: The twistor line in (ii) will play an important role in our analysis of the structure of the twistor spaces. So we introduce the following.

399: 

400: \begin{definition} \label{def-ftl}

401: {\em 

402: We will call the unique twistor line in (ii) {\em the fixed twistor line}.

403: We always denote it by $L_0$.

404: }

405: \end{definition}

406: 

407: 

408: 

409: Since $mF|_S\simeq mK_S^{-1}$ by adjunction formula, 

410: Prop.\,\ref{prop-multan1} (i) means that 

411: in order to obtain (bimeromorphic)  projective models of the twistor spaces,

412: we need to consider a linear system $|mF|$ for $m\ge n-2$.

413: To study these systems systematically, we introduce the following.

414: 

415: 

416: \begin{definition} \label{def-trivial}(i) 

417: For each positive $m\in\mathbf Z$, let $V_m\subset H^0(mF)$ be a linear subspace generated by the image of  a natural multi-linear map

418: \begin{equation}

419: H^0(F)\times H^0(F)\times\cdots\times H^0(F)\,\lra\, H^0(mF)

420: \end{equation}

421: given by $(s_1,s_2,\cdots,s_m)\mapsto s_1\otimes s_2\otimes \cdots\otimes s_m$.

422: Prop.\,\ref{prop-multan1} (i) and Prop.\,\ref{prop-fundamental} (i) implies $\dim V_m=m+1$ and $V_m\subset H^0(mF)^{\mathbf C^*}$.

423: (ii) A divisor $D\in|mF|$ is said to be a trivial member if $D\in |V_m|$.

424: In other words, $D$ is said to be a trivial member if $D=S_1+\cdots+S_m$ for some $S_i\in |F|$. ($S_i$ is not necessarily irreducible, of course.)

425: (iii) The meromorphic map from $Z$ to $\mathbf P^{\vee}V_m= \mathbf{CP}^m$ associated to the system $|V_m|$ is denoted by $\Psi_m$. 

426: The image of $\Psi_m$ is denoted by $\Lambda_m$.

427: $\Lambda_m$ is a rational normal curve. In particular, its degree in $\mathbf{CP}^m$ is $m$.

428: \end{definition}

429: 

430: Since we consider the systems $|mF|$ for different values of $m$, we introduce the following.

431: 

432: \begin{definition}\label{def-asso1}

433: For each positive $m\in\mathbf Z$, let 

434: \begin{align}

435: \Phi_m:Z\to\mathbf P^{\vee}H^0(mF)\,\,\,{\text{and}}\,\,\,\,\Phi_m^{\mathbf C^*}:Z\to\mathbf P^{\vee}H^0(mF)^{\mathbf C^*}

436: \end{align}

437:  be the meromorphic maps associated to the system $|mF|$ and $|mF|^{\mathbf C^*}$ respectively.

438: \end{definition}

439: 

440: We note an obvious relation between two meromorphic maps $\Psi_m$ and $\Phi_m$ in the two definitions.

441: Namely for each $m$ there is a commutative diagram of meromorphic maps

442: \begin{equation}\label{cd1}

443:  \CD

444: Z@>{\Phi_m}>>\mathbf P^{\vee}H^0(mF) \\

445:  @V\Psi_m VV @VV{\pi_m}V\\

446: \Lambda_m@>>>\mathbf P^{\vee}V_m\\

447:  \endCD

448:  \end{equation}

449:  where $\pi_m$ is the projection induced from the inclusion $V_m\subset H^0(mF)$.

450:  (The bottom row is an inclusion as a rational normal curve as in Def.\,\ref{def-trivial}.)

451: We note that the diagram \eqref{cd1} is $\mathbf C^*$-equivariant, where $\mathbf C^*$-actions are trivial on $\Lambda_m$ and $\mathbf P^{\vee}V_m$.

452: Taking the $\mathbf C^*$-fixed part, we obtain the diagram

453: \begin{equation}\label{cd2}

454:  \CD

455: Z@>{\Phi_m^{\mathbf C^*}}>>\mathbf P^{\vee}H^0(mF)^{\mathbf C^*} \\

456:  @V\Psi_m VV @VV{\pi_m^{\mathbf C^*}}V\\

457: \Lambda_m@>>>\mathbf P^{\vee}V_m\\

458:  \endCD

459:  \end{equation}

460: We use these diagrams to reveal a structure of the twistor spaces.

461: As an immediate consequence of these diagrams, we obtain that the images  $\Phi_m(Z)$ and $\Phi_m^{\mathbf C^*}(Z)$ are always contained in $\pi_m^{-1}(\Lambda_m)$ and 

462:  $(\pi_m^{\mathbf C^*})^{-1}(\Lambda_m)$ respectively.

463: 

464: 

465: 

466: Prop.\,\ref{prop-fundamental} and Prop.\,\ref{prop-multan1} (i) imply that if $m<n-2$ we have the coincidence $V_m=H^0(mF)=H^0(mF)^{\mathbf C^*}$.

467: Hence the projection $\pi_m$ in the diagram \eqref{cd1} is the identity map and therefore the image $\Phi_m(Z)$ is the rational normal curve $\Lambda_m$.

468: Hence the map $\Phi_m$ gives little information  for $m<n-2$.

469: If $m=n-2$, Prop.\,\ref{prop-multan1} and the diagram \eqref{cd1} give a possibility that $\Phi_{n-2}$  gives a generically 2 to 1 covering onto its image.

470: Also, if $m=n-1$, Prop\,\ref{prop-multan2} and the diagram \eqref{cd1} give a possibility that  the map $\Phi_{n-1}$ is bimeromorphic over its image.

471: (Both of these expectations will turned out to be true.)

472: %But our method is complicated a little bit, we explain 

473: 

474: A basic tool relevant to this kind of problem is the short exact sequence

475: \begin{align}\label{ses1}

476: 0\lra (m-1)F\lra mF\lra mK_S^{-1}\lra 0.

477: \end{align}

478: At least for the case  $n=4$, this sequence is enough for proving that $\Phi_{n-2}$ is generically 2 to 1 onto its image.

479: %Actually, when $n=3$, we set $m=n-2=1$ in the sequence \eqref{ses1}  and consider the associated cohomology exact sequence. 

480: %Then since we have $H^1(\mathscr O_Z)=0$ (by the  simply connectedness of the base 4-manifold \cite{LB92}), the restriction homomorphsimm  $H^0(F)\to H^0(K_S^{-1})$ is surjective.

481: %Since  $|K_S^{-1}|$ is 2-dimensional as in Prop.\,\ref{prop-multan1} (ii) and its associated map gives a generically 2 to 1 map to $\mathbf{CP}^2$, 

482: % the diagram \eqref{cd1} means that $\Phi_1$ gives generically 2 to 1 map onto its image.

483: Actually, when  $n=4$, we put $m=n-2=2$ in  \eqref{ses1}.

484:  In this case, we use  Riemann-Roch formula and Hitchin's vanishing theorem to deduce

485:  $\dim H^0(F)-\dim H^1(F)=10-2n=2$.

486:  It then follows  $H^1(F)=0$ by Prop.\,\ref{prop-fundamental} (i).

487: Hence we obtain the surjectivity of the restriction map $H^0(2F)\to H^0(2K_S^{-1})$.

488: Therefore  by Prop.\,\ref{prop-multan1}, $S\in|F|$ is mapped surjectively to $\mathbf{CP}^2$  by $\Phi_{n-2}=\Phi_2$ and it is generically 2 to 1.

489: Hence by the diagram \eqref{cd1}, the image $\Phi_2(Z)$ is 3-dimensional and $\Phi_2$ is generically 2 to 1 onto its image.

490: 

491: When $n\ge 5$ we still put $m=n-2$ in \eqref{ses1}  and it becomes

492: \begin{equation}\label{ses3}

493: 0\lra (n-3)F\lra (n-2)F\lra (n-2)K^{-1}_S\lra 0.

494: \end{equation}

495: However, a computation similar to the above shows that $H^1((n-3)F)\neq 0$ (or more precisely its dimension is  quite  high) and the cohomology exact sequence of \eqref{ses3} does not mean the existence of a non-trivial member of $|(n-2)F|$. 

496: Namely, if $n\ge 5$ the argument using cohomology exact sequence fails.

497: This difficulty  always happens when we consider twistor spaces on $n\mathbf{CP}^2$, $n\ge 5$.

498: For example, in \cite{Hon07-3}, we needed to show that the system $|2F|=|K^{-1}_Z|$ has a non-trivial member, in order to show that the image of  its associated map is 2-dimensional.

499: This was shown by proving that a sum of some four degree-one divisors gives a non-trivial member.

500: Unfortunately, in the present case, a computation shows that if  a sum of degree-one divisors belongs to $|(n-2)F|$, then it must be a trivial member, already for the case $n=4$.

501: So the method of \cite{Hon07-3} does not work directly.

502: 

503: We overcome this difficulty by considering a system of higher degree, $|(n-1)F|$.

504: For this system we can find a non-trivial member which is a sum of degree-one divisors as follows.

505: 

506: \begin{lemma}\label{lemma-nt1}

507: Let $Z$ be as in Prop.\,\ref{prop-fundamental}.

508: Then the $\mathbf C^*$-invariant divisors

509: \begin{equation}\label{nt1}

510: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^+\,\,\,\,{\text{and}}\,\,\,\,\,

511: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-

512: \end{equation}

513: are non-trivial members of the system $|(n-1)F|^{\mathbf C^*}$.

514: \end{lemma}

515: 

516: \begin{rmk}

517: {\em

518: In general, even if a sum of degree-one divisors belongs to a system $|mF|$ for some $m\in\mathbf Z$, it does necessarily belong to the subsystem $|mF|^{\mathbf C^*}$.

519: }

520: \end{rmk}

521: 

522: \noindent Proof of Lemma \ref{lemma-nt1}.

523: To show that the divisors \eqref{nt1} are members of $|(n-1)F|$, 

524: since the restriction map of the cohomology groups

525: $H^2(Z,\mathbf Z)\to H^2(S,\mathbf Z)$ is always injective,

526: it suffices to show that the restriction of the divisors \eqref{nt1} to $S$ belongs to 

527: $|(n-1)K_S^{-1}|$.

528: Since the restrictions $S_i^+|_S$ and $S_i^-|_S$ are precisely two halves of the anticanonical cycle $C$ divided by the twistor line $L_i$, 

529: we can write these restrictions as  sums of irreducible components of $C$.

530: Once this is done, it is a routine work to show that they belong to $|(n-1)K_S^{-1}|$.

531: (For example, it suffices to verify that the intersection numbers with generators of $H^2(S,\mathbf Z)$ coincide.)

532: We omit the detail of the verification.

533: It can also be verified that the restrictions moreover belong to the subsystem $|(n-1)K_S^{-1}|^{\mathbf C^*}$.

534: This means that the 2 divisors \eqref{nt1} are contained in the subsystem $|(n-1)F|^{\mathbf C^*}$.

535: 

536: Finally we show that the divisors are non-trivial members.

537:  If the former divisor of \eqref{nt1} is a trivial member, there must exist $1\le i,j\le n+1$ such that  $S_i^++S_j^+$ is a member of the pencil $|F|$.

538:  This contradicts Prop.\,\ref{prop-fundamental} (ii).

539:  Hence the divisor is not a trivial member.

540: Then by the reality of the system $|V_m$,  the latter divisor is not a trivial  member as well.

541:   \proofend

542: 

543: \bigskip 

544: By using Lemma \ref{lemma-nt1} we can explicitly give generators of the system $|(n-1)F|^{\mathbf C^*}$ as follows.

545: 

546: \begin{prop}\label{prop-generator1}

547: Let $Z$ be as in Prop.\,\ref{prop-fundamental}.

548: Then we have the following.

549: (i) $\dim H^0((n-1)F)^{\mathbf C^*}=n+2$.

550: (ii) As generators of the system $|(n-1)F|^{\mathbf C^*}$ we can take the following $\mathbf C^*$-invariant divisors:

551: \begin{align}\label{trivial1}

552: (n-1-k)(S_1^++S_1^-)+k(S_2^++S_2^-),\,\,0\le k\le n-1,

553: \end{align}

554: \begin{align}\label{nt2}

555: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^+,\,\,

556: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-.

557: \end{align}

558: (iii) ${\rm Bs}\,|(n-2)F|^{\mathbf C^*}=C-C_{n+1}-\ol{C}_{n+1}.$

559: \end{prop}

560: 

561: Note that the $n$ divisors \eqref{trivial1} are (independent) generators of the $(n-1)$-dimensional linear system $|V_{n-1}|$.

562: 

563: \bigskip

564: \noindent

565: Proof of Prop.\,\ref{prop-generator1}.

566: We first show that $\dim H^0((n-1)F)^{\mathbf C^*}\le n+2$.

567: By the cohomology exact sequence of \eqref{ses1} with $m=n-2$, we obtain an exact sequence

568: \begin{align}

569: 0\lra H^0((n-3)F)\lra H^0((n-2)F)\lra H^0((n-2)K_S^{-1}).

570: \end{align}

571: Since the divisor $S\in |F|$ is $\mathbf C^*$-invariant, this sequence is $\mathbf C^*$-equivariant.

572: Taking $\mathbf C^*$-fixed part, we obtain an exact sequence

573: \begin{align}\label{ses4}

574: 0\lra H^0((n-3)F)^{\mathbf C^*}\lra H^0((n-2)F)^{\mathbf C^*}\lra H^0((n-2)K_S^{-1})^{\mathbf C^*}.

575: \end{align}

576: Now as is already remarked we have $V_m=H^0(mF)=H^0(mF)^{\mathbf C^*}$ for $m\le n-3$ and $\dim V_m=m+1$ for any $m\ge 0$.

577: On the other hand, by Prop.\,\ref{prop-multan1} (ii), we have $\dim H^0((n-2)K_S^{-1})^{\mathbf C^*}=1$.

578: Hence by the exact sequence \eqref{ses4} we obtain $\dim H^0((n-2)F)^{\mathbf C^*}\le (n-1)$.

579: Taking $m=n-1$ in \eqref{ses1}, the same argument implies an exact sequence

580: \begin{align}\label{ses5}

581: 0\lra H^0((n-2)F)^{\mathbf C^*}\lra H^0((n-1)F)^{\mathbf C^*}\lra H^0((n-1)K_S^{-1})^{\mathbf C^*}.

582: \end{align}

583: Further $\dim H^0((n-2)K_S^{-1})^{\mathbf C^*}=3$ by Prop.\,\ref{prop-multan2} (i).

584: Hence we obtain $\dim H^0((n-1)F)^{\mathbf C^*}\le (n-1)+3=n+2$.

585: 

586: 

587: To finish a proof of (i) and (ii), it remains to see that the $(n+2)$ divisors \eqref{trivial1} and \eqref{nt2} are linearly independent.

588: The $n$ divisors \eqref{trivial1} are clearly linearly independent and its base locus is exactly the cycle $C$.

589: On the other hand a ($\mathbf C^*$-invariant) section of the line bundle $(m-1)F$ which defines 

590: the first one of \eqref{nt2} %divisor $(n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^+$ 

591: does not belong to $V_{n-1}$, since the divisor does not contain the component $\ol{C}_{n+1}$.

592: Thus \eqref{trivial1} plus the first divisor of \eqref{nt2} are linearly independent.

593: It is readily seen that the base locus of this $n$-dimensional subsystem (of $|(n-1)F|^{\mathbf C^*}$) is $C-\ol{C}_{n+1}$.

594: In particular, it contains $C_{n+1}$.

595:  On the other hand, the second divisor of \eqref{nt2} does not contain $C_{n+1}$.

596:  Hence this divisor is not a member of the above $n$-dimensional subsystem.

597: Thus we have shown that the $(n+2)$ divisors \eqref{trivial1} and \eqref{nt2} are linearly independent. Hence we obtain (i) and (ii).

598: 

599: (iii) is immediate if we note that Bs\,$|V_{n-1}|=C$ and that we can explicitly write the restrictions $S_i^{\pm}|_S$ $(1\le i\le n+1)$ as a sum of the components $C_j$ and $\ol C_j$.

600: \proofend

601: 

602: \bigskip

603: We note that as we have $\dim H^0((n-1)F)^{\mathbf C^*}=n+1$ the sequence

604: \begin{align}\label{ses55}

605: 0\lra H^0((n-2)F)^{\mathbf C^*}\lra H^0((n-1)F)^{\mathbf C^*}\lra H^0((n-1)K_S^{-1})^{\mathbf C^*}\lra 0

606: \end{align}

607: is exact.

608: 

609: \subsection{Defining equations of the minitwistor spaces and their singularities}\label{ss-mt}

610: Thanks to Prop.\,\ref{prop-generator1}, we can determine the image of the meromorphic map $\Phi_{n-1}^{\mathbf C^*}$ (associated to the system $|(n-1)F|^{\mathbf C^*}$) as follows.

611: (Recall that by the diagram \eqref{cd2}, $\Phi_{n-1}^{\mathbf C^*}(Z)\subset(\pi_{n-1}^{\mathbf C^*})^{-1}(\Lambda_{n-1})$ holds.)

612: 

613: \begin{thm}\label{thm-mt1}

614: Let $Z$ be as in Prop.\,\ref{prop-fundamental}.

615: Then there exist a homogeneous coordinate $(z_1,z_2,\cdots,z_{n})$ on $\mathbf P^{\vee}V_{n-1}=\mathbf{CP}^{n-1}$ and two sections $z_{n+1},z_{n+2}\in H^0((n-1)F)^{\mathbf C^*}$ which satisfy the following.

616: (i) $\{z_i\set 1\le i\le n+2\}$ is a basis of $H^0((n-1)F)^{\mathbf C^*}$.

617: (ii) With respect to the non-homogeneous coordinate $(z_2/z_1,z_3/z_1,\cdots,z_n/z_1)$ on \ $\mathbf P^{\vee}V_{n-1}$, the rational normal curve $\Lambda_{n-1}$ in $\mathbf P^{\vee}V_{n-1}$ is given by

618: \begin{align}\label{rnc1}

619: \left\{\left(\lambda,\lambda^2,\cdots,\lambda^{n-1}\right)\set\lambda\in\mathbf C\right\}.

620: \end{align} 

621: (iii) $z_{n+1}$ and $z_{n+2}$ define the 2 divisors \eqref{nt2} respectively and also satisfy $\ol{z}_{n+2}=z_{n+1}$.

622: (iv) 

623: The image $\mathscr T:=\Phi_{n-1}^{\mathbf C^*}(Z)$ is a surface which satisfies not only  relations in the defining ideal of $\Lambda_{n-1}$, but also the following quadratic equation

624: \begin{align}\label{nt3}

625: z_{n+1}z_{n+2}=z_2\left(

626: z_n-\sigma_1z_{n-1}+\sigma_2z_{n-2}+\cdots+(-1)^{n-1}\sigma_{n-1}z_1

627: \right),

628: \end{align}

629: where $\sigma_i$ is the elementary symmetric polynomial  of degree $i$ of some $(n-1)$ real numbers $\lambda_3,\cdots,\lambda_{n+1}$.

630: (v) The degree of the surface $\mathscr T$ in $\mathbf{CP}^{n+1}$ is $2(n-1)$.

631: \end{thm}

632: 

633: \noindent

634: Proof.

635: Let $e_i$ ($1\le i\le n+1$) be a section of the line bundle $[S_i^+]$ which defines $S_i^+$.

636: We put

637: \begin{align}

638: y_i=e_i\cdot \ol{e}_i\in H^0(F)\hspace{5mm} (1\le i\le n+1).

639: \end{align}

640: (Of course, this defines the divisor $S_i^++S_i^-$.)

641: We choose $\{y_1,y_2\}$ as a basis of $H^0(F)$.

642: Since these are also a basis of the real part $H^0(F)^{\sigma}$, there exist real numbers $\lambda_3,\cdots,\lambda_{n+1}$ such that 

643: \begin{align}

644: y_i=y_2-\lambda_iy_1

645: \end{align}

646: hold.

647: We put $\lambda_2=0$.

648: Then since $S_i^++S_i^-\in |F|^{\sigma}\simeq S^1$ are arranged in a linear order, either $\lambda_2=0<\lambda_3<\lambda_4<\cdots<\lambda_{n+1}$ or $\lambda_2=0>\lambda_3>\lambda_4>\cdots>\lambda_{n+1}$ holds.

649: We put %for $1\le i\le n$,

650: \begin{align}\label{z1}

651: z_1=y_1^{n-1},\,z_2=y_1^{n-2}y_2,\,\cdots\,,z_n=y_2^{n-1}.

652: \end{align}

653: Then it is obvious from the definition of the space $V_{n-1}$ that $\{z_1,z_2,\cdots,z_{n}\}$ is a basis of $V_{n-1}$.

654: Also it is immediate to see that the curve $\Lambda_{n-1}$ is represented as in \eqref{rnc1}.

655: We put 

656: \begin{align}

657: z_{n+1}=e_1^{n-2}e_2e_3\cdots e_{n+1}\,\,\,\text{and}\,\,\,\, z_{n+2}=\ol{e}_1^{n-2}\ol{e}_2\ol{e}_3\cdots \ol{e}_{n+1}.

658: \end{align}

659: Then by Prop.\,\ref{prop-generator1}, $\{z_1,\cdots,z_{n+2}\}$ is a basis of $H^0((n-1)F)^{\mathbf C^*}$. 

660: Also it is obvious that $z_{n+1}$ and $z_{n+2}$ define the divisors \eqref{nt2} respectively.

661: Then we have

662: \begin{align}

663: z_{n+1}z_{n+2}&= (e_1\ol{e}_1)^{n-2}\cdot (e_2\ol{e}_2)\cdot (e_3\ol{e}_3)\cdots(e_{n+1}\ol{e}_{n+1})\\

664: &=y_1^{n-2}y_2y_3\cdots y_{n+1}\\

665: &=y_1^{n-2}y_2\cdot(y_2-\lambda_3y_1)(y_2-\lambda_4y_1)\cdots(y_2-\lambda_{n+1}y_1).

666: \label{dcp1}

667: \end{align} 

668: Expanding the right-hand side and using \eqref{z1}, we obtain the equation \eqref{nt3}.

669: Thus we obtain (iv).

670: For (v) note that the surface $\mathscr T$ is an intersection of a rational scroll of lines whose degree is $(n-1)$ (= the degree of $\Lambda_{n-1}$ in $\mathbf{CP}^{n-1}$) and the hyperquadric \eqref{nt3}.

671: The claim is immediate from this.

672: %

673: %take a general hyperplane in $\mathbf{CP}^{n-1}$ and consider its inverse image by the projection $\pi_{n-1}^{\mathbf C^*}:\mathbf{CP}^{n+1}\to\mathbf{CP}^{n-1}$.

674: %Then the intersection of the hyperplane and $\mathscr T$ consists of $(n-1)$ conics.

675: %Cutting these by general hyperplane, we obtain $2(n-1)$ points.

676: %Hence the degree of $\mathscr T$ is $2(n-1)$.

677: \proofend

678: 

679: \bigskip

680: Since the restriction map in the sequence \eqref{ses55} is surjective, Prop.\,\ref{prop-multan2} (v) implies the following

681: 

682: \begin{prop}\label{prop-quot1}

683: Restriction of the meromorphic map $\Phi_{n-1}^{\mathbf C^*}$ to a general member  $S\in |F|$ is a morphism and its general  fibers are irreducible smooth rational curves in $S$.

684: (Namely the restriction can be regarded as a (holomorphic) quotient map of $\mathbf C^*$-action on $S$.)

685: \end{prop}

686: 

687: 

688: By the diagram \eqref{cd2}, we have the following commutative diagram of meromorphic maps

689: \begin{equation}\label{cd3}

690:  \CD

691: Z@>{\Phi_{n-1}^{\mathbf C^*}}>>\mathscr T \\

692:  @V\Psi_{n-1}VV @VV{\pi_{n-1}^{\mathbf C^*}}V\\

693: \Lambda_{n-1}@>>>\mathbf{CP}^{n-1}\\

694:  \endCD

695:  \end{equation}

696:  By Prop.\,\ref{prop-quot1}, the map $\Phi_{n-1}^{\mathbf C^*}:Z\to\mathscr T$ can be regarded as a quotient map by $\mathbf C^*$-action.

697:  Namely the complex surface $\mathscr T$ can be considered as a minitwistor space of the present twistor space $Z$.

698:  

699:  Next we investigate geometric structure of the surface $\mathscr T$.

700:  The equation \eqref{nt3} means that  by the projection $\pi_{n-1}^{\mathbf C^*}$, $\mathscr T$ is birational to a conic bundle over $\Lambda_{n-1}$ whose discriminant locus is  the intersection of $\Lambda_{n-1}$ and the union of 2 hyperplanes determined as the zero locus of the right-hand side of \eqref{nt3}.

701: Since $z_2=y_1^{n-2}y_2$ as in \eqref{z1}, the rational normal curve $\Lambda_{n-1}$ intersects the hyperplane $\{z_2=0\}$ transversally at $(1,0,0,\cdots,0)$, and  touches the hyperplane at $(0,0,\cdots,0,1)$ with multiplicity $(n-2)$, 

702: where we are using the homogeneous coordinate $(z_1,\cdots,z_n)$ on $\mathbf{CP}^{n-1}$ given in Theorem\,\ref{thm-mt1}.

703:  Also, by \eqref{dcp1}, the intersection of $\Lambda_{n-1}$ with another hyperplane consists of $(n-2)$ points $(\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})$ ($3\le i\le n+1$) with respect to the  coordinate on $\Lambda_{n-1}$ in \eqref{rnc1}.

704: The fibers of these $(n+1)$ discriminant points consist of two irreducible components.

705:  The reducible fiber over the point $(\lambda_i,\lambda_i^2,\cdots,\lambda_{n+1}^2)\in\Lambda_{n-1}$ is precisely $\Phi(S_i^+)+\Phi(S_i^-)$ for $2\le i\le n+1$ and that over the touching point $(0,\cdots,0,1)\in\Lambda_{n-1}$ is $\Phi(S_1^+)+\Phi(S_1^-)$.

706: (See also Figure \ref{fig-mt}.)

707: 

708: Note that the complex surface $\mathscr T$ is uniquely determined by the set $\{\lambda_3,\cdots,\lambda_{n+1}\}$ of different $n-1$ numbers.

709: Further, two surfaces $\mathscr T$ and $\mathscr T'$ determined by $\{\lambda_3,\cdots,\lambda_{n+1}\}$  and $\{\lambda'_3,\cdots,\lambda'_{n+1}\}$ respectively are isomorphic iff there exists a constant $c\in\mathbf R^{\times}$ such that $\{c\lambda_3,\cdots,c\lambda_{n+1}\}=\{\lambda'_3,\cdots,\lambda'_{n+1}\}$.

710: Thus the present minitwistor spaces form $(n-2)$-dimensional family.

711: 

712: It is obvious that with respect to the homogeneous coordinates  $(z_1,\cdots,z_n)$ on $\mathbf P^{\vee}V_{n-1}$ and $(z_1,\cdots,z_{n+2})$ on $\mathbf P^{\vee} H^0((n-1)F)^{\mathbf C^*}$, the projection $\pi_{n-1}^{\mathbf C^*}$ is explicitly given by

713: %\begin{align}

714: $\pi_{n-1}^{\mathbf C^*}(z_1,\cdots,z_{n+2})=(z_1,\cdots,z_n).$

715: %\end{align}

716: Hence the center of $\pi_{n-1}^{\mathbf C^*}$ is the line defined by 

717: \begin{align}

718: l_{\infty}:=\{z_1=z_2=\cdots=z_n=0\}.

719: \end{align}

720: Fibers of $\pi_{n-1}^{\mathbf C^*}$ are planes containing $l_{\infty}$.

721: It can be readily computed by Theorem\,\ref{thm-mt1} that we have

722: \begin{align}\label{inters1}

723: l_{\infty}\cap \mathscr T=\{(0,0,\cdots,0,1,0), (0,0,\cdots,0,0,1)\}

724: \end{align}

725: in the homogeneous coordinate $(z_1,\cdots,z_{n+2})$.

726: Fibers of $\pi_{n-1}^{\mathbf C^*}|_{\mathscr T}$ form a pencil on $\mathscr T$ whose inverse image by $\Phi_{n-1}^{\mathbf C^*}$ are the pencil $|F|$ on $Z$.

727: The base locus of this pencil on $\mathscr T$ is precisely the 2 points \eqref{inters1}.

728: 

729: Singularities of the image surface $\mathscr T$ is described as follows.

730: 

731: \begin{prop}\label{prop-mt2}

732: Let $\mathscr T$ be as in Theorem\,\ref{thm-mt1}.

733: Then the singularities of $\mathscr T$ consist of the following 3 points.

734: (a) the 2 intersection points $l_{\infty}\cap \mathscr T$ in \eqref{inters1},

735: (b) the point $(0,\cdots,0,1,0,0)$ in the homogeneous coordinate $(z_1,\cdots,z_{n+2})$.

736: Further, the 2 points of (a) are isomorphic to the cyclic quotient singularity of $\mathbf Z_{n-1}$-action $(z,w)\mapsto (\zeta z,\zeta w)$ on $\mathbf C^2$ with $\zeta=e^{\frac{2\pi\sqrt{-1}}{n-1}}$.

737: The point (b) is isomorphic to a rational double point of A$_{n-3}$-type.

738: \end{prop}

739: 

740: We omit a proof since the description of $\mathscr T$ obtained in Theorem\,\ref{thm-mt1} is completely explicit and by using it, it is not difficult to derive the conclusion.

741: Instead we only note that the fiber of $\pi_{n-1}^{\mathbf C^*}:\mathscr T\to\mathbf{CP}^{n-1}$ in the diagram \eqref{cd3}  over the (touching) point $(z_1,\cdots,z_n)=(0,\cdots,0,1)$ consists of 2 irreducible component, and that the point in (b) is precisely the intersection point of these components. 

742: 

743: Let $\hat{\mathscr T}\to\mathscr T$ be the blowing-up of the embedded surface $\mathscr T\subset\mathbf{CP}^{n+1}$ along the line $l_{\infty}$.

744: Then the pair of  cyclic quotient singularities ((a) of Prop.\,\ref{prop-mt2}) are resolved.

745: At the same time, the indeterminacy locus of the projection $\mathscr T\to\Lambda_{n-1}$ is resolved so that $\hat{\mathscr T}$ has a structure of a conic bundle over $\Lambda_{n-1}$.

746: Let $\Gamma$ and $\ol{\Gamma}$ be the exceptional curves over the 2 points \eqref{inters1}.

747: These are of course  smooth rational curves in $\hat{\mathscr T}$ whose self-intersection numbers  are $(1-n)$.

748: $\hat{\mathscr T}$ still has $A_{n-3}$-singularity ((b) of Prop.\,\ref{prop-mt2}).

749: Let $\tilde{\mathscr T}\to\hat{\mathscr T}$ be its minimal resolution.

750: The exceptional curve of this resolution is a string of $(-2)$-curves consisting of $(n-3)$ irreducible components.

751: The composition

752: \begin{align}\label{resol1}

753: \nu:\tilde{\mathscr T}\lra\hat{\mathscr T}\lra\mathscr T

754: \end{align}

755:  is the minimal resolution of the surface $\mathscr T$.

756: Since the natural projection $\tilde{\mathscr T}\to\Lambda_{n-1}$ has precisely $(n+1)$ reducible fibers, and since the number of their irreducible components is $2n+(n-1)=3n-1=(2n-2)+(n+1)$, we obtain

757: \begin{align}\label{chern}

758: c_1^2(\tilde{\mathscr T})=8-(2n-2)=10-2n.

759: \end{align}

760: From this we obtain $b_2(\tilde{\mathscr T})=2n$.

761: 

762: 

763: \section{Description of the twistor spaces as conic bundles}

764: \subsection{Partial elimination of the indeterminacy locus of the quotient meromorphic map}\label{ss-pr}

765: We recall that the surface $\mathscr T$ studied in \S \ref{ss-mt} was the image of the meromorphic map $\Phi_{n-1}^{\mathbf C^*}$ associated to the linear system $|(n-1)F|^{\mathbf C^*}$, and it can also be regarded as a quotient space of $\mathbf C^*$-action on the twistor space $Z$.

766:  Since we know generator of this system  in an explicit form as in 

767: Prop.\,\ref{prop-generator1}, we can in principle eliminate the indeterminacy locus of the map $\Phi_{n-1}^{\mathbf C^*}$, by a succession of blowing-ups. % (as we did for the system $|2F|$ in \cite{Hon07-3}).

768: As in Prop.\,\ref{prop-generator1} we have Bs\,$|(n-1)F|^{\mathbf C^*}=C-C_{n+1}-\ol{C}_{n+1}$.

769: Because the generators in the proposition intersect in a quite complicated way,

770:  it looks not easy to give a complete elimination.

771: So from now on we give an elimination only in a neighborhood of the curve $C_2\cup\ol{C}_2$.

772: This is enough for the purpose of obtaining  projective models of the present twistor spaces as conic bundles.

773: We recall that $C_2$ and $\ol{C}_2$ are only  components among the cycle $C$ which are pointwisely fixed by the $\mathbf C^*$-action.

774: In the terminology of \cite{S82}, $C_2$ and $\ol{C}_2$ are `source' and `sink' of the $\mathbf C^*$-action on $Z$.

775: (Namely, by the $\mathbf C^*$-action, general points on $Z$ goes to points on $C_2$ and $\ol{C}_2$ as $s\to 0$ and $s\to\infty$.)

776: In the following we repeat the next usual way of  eliminations of the base locus:

777: \begin{itemize}

778: \item blow-up along base curves,

779: \item compute the total transforms of generators,

780: \item compute their fixed components

781: \item remove the fixed component from the total transforms

782: \item  compute the base curves

783: \end{itemize}

784: 

785: As the first step, let $Z_1\to Z$ be blowing-up along $C_2\cup \ol{C}_2$, and $E_2$ and $\ol{E}_2$ the exceptional divisors over $C_2$ and $\ol{C}_2$ respectively.

786: Since the 2 divisors $S_2^+$ and $S_3^-$ (in $Z$) intersect transversally along $C_2$, and the self-intersection numbers of $C_2$ in  $S_2^+$ and $S_3^-$ are $(-1)$ for both,

787: the normal bundle of $C_2$ in $Z$ is isomorphic to $\mathscr O(-1)^{\oplus 2}$.

788: Hence $E_2$  is biholomorphic to the trivial $\mathbf{CP}^1$-bundle over $C_2$.

789: Similarly, $\ol{E}_2\simeq \ol{C}_2\times\mathbf{CP}^1$.

790: Then since every generators in Prop.\,\ref{prop-generator1} contain the curve $C_2\cup\ol{C}_2$ with the same multiplicity $(n-1)$, 

791: their total transforms in $Z_1$ contain $E_2+\ol{E}_2$ with multiplicity $(n-1)$.

792: Hence subtracting  $(n-1)E_2+(n-1)\ol{E}_2$ from the total transforms of generators,

793: we obtain a linear system on $Z_1$ 

794: whose generators can still be written as \eqref{trivial1} and \eqref{nt2}, where

795:  we are using the same notations to denote the divisors in $Z$ and their strict transforms in $Z_1$ as promised in the convention.

796:  

797:  This $(n+1)$-dimensional linear system on $Z_1$ still has a base locus which intersects $E_2$ or $\ol{E}_2$.

798:  Namely, the curves $C_1$ and $C_3$ (in $Z_1$), intersecting $E_2$ transversally at a unique point for each, are (still)  base curves.

799:  Similarly, $\ol{C}_1$ and $\ol{C}_3$ (in $Z_1$) are base curves intersecting $\ol{E}_2$ transversally at a unique point  for each.

800:  More precisely, all divisors in \eqref{trivial1} (viewed as divisors on $Z_1)$ obviously contain $C_1,\ol{C}_1,C_3$ and $\ol{C}_3$ with multiplicity $(n-1)$.

801:  On the other hand, the former divisor of \eqref{nt2} contains these 4 curves with multiplicity $(n-2)$, $n$, $n$ and $(n-2)$ respectively.

802:  The latter divisor of \eqref{nt2} contains these curves with multiplicities $n,n-2,n-2,n$ respectively.

803:  %Hence $C_2$ and $C_4$ are base curves of the system.

804:  %Similarly, the curve $\ol{C}_1$ and $\ol{C}_3$ are base curves which intersect $\ol{E}_2$.

805:  It can also be  verified (by using Prop.\,\ref{prop-generator1}) that there is no base locus other than these which intersects $E_2$ or $\ol{E}_2$.

806:   Let $Z_2\to Z_1$ be the blowing-up along $C_1\cup \ol{C}_1\cup C_3\cup\ol{C}_3$, and $E_1, \ol{E}_1, E_3$ and $\ol{E}_3$ their exceptional divisors respectively.

807:   Then  all the total transforms of the divisors \eqref{trivial1} and \eqref{nt2} into $Z_1$ contain $E_1, \ol{E}_1, E_3 $ and $\ol{E}_3$  with the above multiplicities respectively.

808:   So removing the maximal common divisor $(n-2)(E_1+E_3+\ol{E}_1+\ol{E}_3)$

809:   (namely, the fixed component) from the total transforms, we obtain a linear system on $Z_2$ (with the same dimension $(n+1)$) whose generators are 

810:   \begin{align}\label{trivial2}

811: (n-1-k)(S_1^++S_1^-)+k(S_2^++S_2^-)+(E_1+E_3+\ol{E}_1+\ol{E}_3),\,\,\,0\le k\le n-1,

812: \end{align}

813: \begin{align}\label{nt4}

814: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^++2E_3+2\ol{E}_1,\,\,\,

815: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-+2E_1+2\ol{E}_3.

816: \end{align}

817: Of course, \eqref{trivial2} and \eqref{nt4}  correspond \eqref{trivial1} and \eqref{nt2} respectively.

818: From these, we can deduce that the base locus  which intersect $E_2$ are the following $(n-1)$ rational curves.

819: \begin{align}\label{base1}

820: S_1^+\cap E_1\,\,\,\,{\text{and}}\,\,\,S_i^-\cap E_3\,\,(4\le i\le n+1).

821: \end{align}

822: %More precisely, the divisors \eqref{trivial2} (in $Z_2)$ contain the curve $S_1^+\cap E_1$ by multiplicity $n-k$, and the curve $S_i^-\cap E_3$ ($4\le i\le n+1)$ by multipicity 1.

823: %On the other hand, the former divisor of \eqref{nt4} contains  $S_1^+\cap E_1$ by multiplicity $n-2$, and the curve $S_i^-\cap E_3$ ($4\le i\le n+1)$ by multipicity 2.

824: %Further, the latter divisor of \eqref{nt4} contains  $S_1^+\cap E_1$ by multiplicity $2$, and the curve $S_i^-\cap E_3$ ($4\le i\le n+1)$ by multipicity 1.

825: %By considering the conjugation of these, we deduce that 

826: Similarly, the base locus of the linear system (on $Z_2$) intersecting $\ol{E}_2$ are the following $(n-1)$ rational curves

827: \begin{align}\label{base2}

828: S_1^-\cap \ol{E}_1\,\,\,\,{\text{and}}\,\,\,S_i^+\cap \ol{E}_3\,\,(4\le i\le n+1).

829: \end{align}

830: Let $Z_3\to Z_2$ be the blowing-up along $2(n-1)$ rational curves \eqref{base1} and \eqref{base2}, $F_1$ and $D_i$ ($4\le i\le n+1$) the exceptional divisors over the curves \eqref{base1} respectively, and $\ol{F}_1$ and $\ol{D}_i$  ($i\le 4\le n+1$) the exceptional divisors over the curves \eqref{base2} respectively.

831: Counting the multiplicity of the generators \eqref{trivial2} and \eqref{nt4} along the base curves \eqref{base1} and \eqref{base2} respectively, we deduce that the fixed component of the $(n+1)$-dimensional linear system on $Z_3$ is precisely

832:  $(F_1+\ol{F}_1)+\sum_{i=4}^{n+1}(D_i+\ol{D}_i)$.

833:  Removing this  from the total transforms of the generators \eqref{trivial2} and \eqref{nt4}, 

834: as  generators of the movable part of  the linear system on $Z_3$, we can choose the following divisors:

835:    \begin{align}\label{trivial3}

836: (n-1-k)(S_1^++S_1^-)+k(S_2^++S_2^-)+(E_1+E_3+\ol{E}_1+\ol{E}_3)+(n-1-k)(F_1+\ol{F}_1),

837: \end{align}

838: with $0\le k\le n-1$, and 

839: \begin{align}\label{nt5}

840: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^++2E_3+2\ol{E}_1+(n-3)F_1+\sum_{i=4}^{n+1}D_i+\ol{F}_1,

841: \end{align}

842: \begin{align}\label{nt6}

843: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-+2\ol{E}_3+2E_1+F_1+(n-3)\ol{F}_1+\sum_{i=4}^{n+1}\ol{D}_i.

844: \end{align}

845: This linear system on $Z_3$ still has base curves intersecting $E_2\cup\ol E_2$. 

846: Namely, the  two rational curves $E_1\cap F_1$ and $\ol{E}_1\cap \ol{F}_1$ are such curves.

847: So let $Z_4\to Z_3$ be the blowing-up along these 2 curves, and $F_2$ and $\ol{F}_2$ the exceptional divisors over there respectively.

848: Pulling back the divisors \eqref{trivial3}--\eqref{nt6} and removing its fixed component $F_2+\ol{F}_2$, we obtain, as  generators of an $(n+1)$-dimensional linear system on $Z_4$, 

849: the following divisors:

850:   \begin{align}\label{trivial4}

851: (n-1-k)(S_1^++S_1^-)+k(S_2^++S_2^-)+(E_1+E_3+\ol{E}_1+\ol{E}_3)+(n-1-k)\sum_{i=1}^2(F_i+\ol{F}_i)

852: \end{align}

853: with $0\le k\le n-1$, and 

854: \begin{align}\label{nt7}

855: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^++2E_3+2\ol{E}_1+(n-3)F_1+(n-4)F_2+\sum_{i=4}^{n+1}D_i+\ol{F}_1,

856: \end{align}

857: \begin{align}\label{nt8}

858: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-+2\ol{E}_3+2E_1+F_1+(n-3)\ol{F}_1+(n-4)\ol{F}_2+\sum_{i=4}^{n+1}\ol{D}_i.

859: \end{align}

860: If $n=4$, since the coefficients of $F_2$ in   \eqref{nt7} and $\ol{F}_2$ in \eqref{nt8} are both zero,  this system on $Z_4$ is free, at least in a neighborhood of $E_2\cup\ol{E}_2$.

861: So we stop the operations here.

862: If $n\ge 5$, the base locus of this system which intersects  $E_2\cup\ol{E}_2$ consists of 2 rational curves $E_1\cap F_2$ and $\ol{E}_1\cap\ol{F}_2$.

863: So let $Z_5\to Z_4$ be the blowing-up along these 2 curves with the exceptional divisor $F_3$ and $\ol{F}_3$ respectively.

864: When $n=5$, the linear system on $Z_5$ obtained by pulling back the divisors  \eqref{trivial4}--\eqref{nt8} and removing the fixed component  $F_3+\ol{F}_3$ becomes free, and we stop the operations here.

865: For $n\ge 6$,  repeating this operation $(n-4)$ times,  we obtain a string of 3-folds $Z_{n}\to Z_{n-1}\to\cdots \to Z_5$ and exceptional  divisors $F_4,\ol{F}_4,F_5,\ol{F}_5,\cdots F_{n-2},\ol{F}_{n-2}$, together with an $(n+1)$-dimensional linear system on $Z_n$ whose generators are given by

866:   \begin{align}\label{trivial5}

867: (n-1-k)(S_1^++S_1^-)+k(S_2^++S_2^-)+(E_1+E_3+\ol{E}_1+\ol{E}_3)+(n-1-k)\sum_{i=1}^{n-2}(F_i+\ol{F}_i)

868: \end{align}

869: with $0\le k\le n-1$, and 

870: \begin{align}\label{nt9}

871: (n-2)S_1^+\,+\sum_{i=2}^{n+1} S_i^++2E_3+2\ol{E}_1+\sum_{i=1}^{n-2}(n-2-i)F_i+\sum_{i=4}^{n+1}D_i+\ol{F}_1,

872: \end{align}

873: \begin{align}\label{nt10}

874: (n-2)S_1^-\,+\sum_{i=2}^{n+1} S_i^-+2\ol{E}_3+2E_1+F_1+\sum_{i=1}^{n-2}(n-2-i)\ol{F}_i+\sum_{i=4}^{n+1}\ol{D}_i.

875: \end{align}

876: Then basically because the coefficients of $F_{n-2}$ in \eqref{nt9} and  $\ol{F}_{n-2}$ in \eqref{nt10} 

877: are both zero, this linear system on $Z_n$ is free, at least  in a neighborhood of the divisor $E_2\cup\ol{E}_2$.

878: Thus we have obtained an explicit sequence of blowing-ups $Z_n\to Z_{n-1}\to\cdots\to Z_1\to Z$ which eliminates the base locus of the system $|(n-1)F|^{\mathbf C^*}$, at least in a neighborhood of $C_2\cup\ol{C}_2$.

879: 

880: Then we have the following

881: 

882: \begin{prop}\label{prop-elim1}

883: Consider the following diagram %of meromorphic maps

884: \begin{equation}\label{cd4}

885:  \CD

886: Z_n@>>> Z \\

887:  @V\tilde{\Phi}_{n-1}^{\mathbf C^*}VV  @VV{\Phi_{n-1}^{\mathbf C^*}}V\\

888: \tilde{\mathscr T}@>{\nu}>>\mathscr T\\

889:  \endCD

890:  \end{equation}

891:  where $\nu$ is the minimal resolution of $\mathscr T$ as in \eqref{resol1}, and 

892: $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ is a  meromorphic map  uniquely determined by the commutativity of the diagram.

893: Then  $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ is holomorphic in  neighborhoods of $E_2$ and $\ol{E}_2$, and its restrictions onto these divisors are both biholomorphic.

894: \end{prop}

895: 

896: \noindent

897: Proof.

898: We consider the pencil $|F|$ on $Z$ whose base locus is the cycle $C$ as in Prop.\,\ref{prop-fundamental} (i).

899: Since the first operation $Z_1\to Z$ blow-ups $C_2\cup\ol{C}_2\subset C$,

900: the strict transform of the pencil $|F|$ into $Z_1$ makes sense.

901: Similarly, since $Z_2\to Z_1$ blow-ups $C_1\cup\ol{C}_1\cup C_3\cup\ol{C}_3$ which are still contained in the base locus of the pencil on $Z_1$,

902: the strict transform of the pencil into $Z_2$ also makes sense.

903: Evidently this pencil on $Z_2$ has no base point, at least in a neighborhood of $E_2\cup\ol{E}_2$.

904: Pulling this back by the sequence of blowing-ups $Z_n\to Z_{n-1}\to\cdots\to Z_2$, we obtain a pencil on $Z_n$ which has no base point in a neighborhood of $E_2\cup\ol{E}_2$.

905: By restricting this pencil to $E_2$ and $\ol{E}_2$ ($\subset Z_n$), we obtain pencils on 

906:  $E_2$ and $\ol{E}_2$ without base points.

907:  

908:  Since the sequence  $Z_n\to Z_{n-1}\to\cdots\to Z$ eliminates the base locus of the system $|(n-1)F|^{\mathbf C^*}$ in a neighborhood of $C_2\cup \ol{C}_2$, the composition map $Z_n\to Z\to\mathscr T$ is holomorphic in a neighborhood of $E_2\cup\ol{E}_2$.

909:  By the commutative diagram \eqref{cd3}, members of the above pencils on $E_2$ and $\ol{E}_2$ are mapped to fibers of the (rational) conic bundle $\pi_{n-1}^{\mathbf C^*}:\mathscr T\to\Lambda_{n-1}$.

910:  General (irreducible) members of the pencils are mapped biholomorphically to smooth fiber of $\mathscr T\to\Lambda_{n-1}$.

911: Further, from our explicit way of the sequence of blow-ups, it is obvious that these pencils (on $E_2$ and $\ol{E}_2$) have precisely $(n+1)$ reducible members, each of which are explicitly given by the restriction of the divisors

912:  \begin{align}

913:  S_1^++\sum_{i=1}^{n-2}F_i,\,\,S_2^++S_2^-,\,\,S_3^++S_3^-,\,\,S_i^-+\ol{D}_i\,\,(4\le i\le n+1)

914:  \end{align}

915:  to $E_2$, and the restrictions of their conjugates to $\ol{E}_2$.

916:  On the other hand, as is already explained in \S\ref{ss-mt},

917:  reducible fibers of the (rational) conic bundle  $\pi_{n-1}^{\mathbf C^*}:\mathscr T\to\Lambda_{n-1}$ are over the points $(\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})\in\Lambda_{n-1}$, $2\le i\le n+1$, and the touching point $(0,\cdots,0,1)\in\Lambda_{n-1}$.

918:  Further the fiber over $(\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})$ is $\Phi_{n-1}^{\mathbf C^*}(S_i^+)+\Phi_{n-1}^{\mathbf C^*}(S_i^-)$ for $2\le i\le n+1$ and that over $(0,\cdots,0,1)$ is 

919:  $\Phi_{n-1}^{\mathbf C^*}(S_1^+)+\Phi_{n-1}^{\mathbf C^*}(S_1^-)$.

920:  Therefore, the map $E_2\to\mathscr T$ (which is the restriction of the composition $Z_n\to Z\to\mathscr T$) is the birational morphism which contracts the following rational curves:

921:  \begin{align}

922:  E_2\cap E_1,\,\,E_2\cap E_3\,\,\,\text{and}\,\,\,F_i\cap E_2\,\,\,\text{for}\,\,\,1\le i\le n-3.

923:   \end{align}

924:  By the configuration of these curves (which can be read off from our explicit way of the sequence $Z_n\to\cdots Z)$, it follows that the morphism $E_2\to\tilde{\mathscr T}$ is precisely the minimal resolution of $\mathscr T$.

925:  Hence we obtain $E_2\simeq\tilde{\mathscr T}$, by the uniqueness of the minimal resolution.

926:  By reality, we also obtain $\ol E_2\simeq\tilde{\mathscr T}$.

927:  

928:  Finally we show that $\tilde{\Phi}_{n-1}^{\mathbf C^*}:Z_n\to\tilde{\mathscr T}$ in the proposition is holomorphic  in neighborhoods of $E_2$ and $\ol{E}_2$.

929:  By our choice of the sequence $Z_n\to Z_{n-1}\to\cdots Z_1\to Z$, only $E_1,E_3,\ol{E}_1$ and $\ol{E}_2$ are mapped to the conjugate pair of singular points of $\mathscr T$ ((a) of Prop.\,\ref{prop-mt2}), and only $F_i$ and $\ol{F}_i$ $(1\le i\le n-3)$ are mapped to the $A_{n-3}$-singularity of $\mathscr T$ ((b) of Prop.\,\ref{prop-mt2}), by the composition $Z_n\to Z\to\mathscr T$.

930:  Further, all intersections of these exceptional divisors with $E_2$ and $\ol{E}_2$ are precisely the exceptional curves of the minimal resolutions $E_2\to\mathscr T$ or $\ol{E}_2\to\mathscr T$.

931:  These imply that the map $Z_n\to Z\to \mathscr T$ is lifted to a map $Z_n\to \tilde{\mathscr T}$ in a way that it is still holomorphic on $E_2$ and $\ol{E}_2$.

932: % 

933: % 

934: % Since $|F|$ has precisely $(n+1)$ reducible members as in Prop.\,\ref{prop-fundamental} (ii), these pencils on  $E_2$ and $\ol{E}_2$ have precisely $(n+1)$ reducible members respectively.

935: % Further, only the reducible member corresponding to $S_1^++S_1^-$ have the special property that it consists of $(n-1)$ irreducible components, which are explicitly given by the intersections with $(n-1)$ divisors $S_1^+, F_1,F_2,\cdots,F_{n-2}$.

936: % All other reducible members (of the pencils on $E_2$ and $\ol{E}_2$) consist of two irreducible components.

937: % Since the members of these pencils are mapped to fibers of $\pi_{n-1}^{\mathbf C^*}:

938: %\mathscr T\to\Lambda_{n-1}$ due to the diagram \eqref{cd3},

939: %the structure of the restrictions of the composition map $Z_n\to Z\to \mathscr T$ onto $E_2$ and $\ol{E}_2$ can be explicitly described.

940: %In particular, it can be seen that these are the (full) minimal resolution of the surface $\mathscr T$.

941: %Moreover, since the restriction of $\Phi_{n-1}^{\mathbf C^*}$ to members $S\in |F|$ is nothing but the $\mathbf C^*$-quotient morphism by Prop.\,\ref{prop-quot1} which is holomorphic on the curve $C_2$ and $\ol{C}_2$, 

942: %the lifted meromorphic map $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ is still holomorphic on $E_2$ and $\ol{E}_2$.

943: %Now it is obvious that its restrictions to $E_2$ and $\ol{E}_2$ are biholomorphic.

944: \proofend

945: 

946: %\begin{prop}

947: %Let $Z$ be as in Prop.\,\ref{prop-fundamental} and $\Phi_{n-1}^{\mathbf C^*}:Z\to\mathscr T\subset\mathbf{CP}^{n+1}$ be the meromorphic map associated to the system $|(n-1)F|^{\mathbf C^*}$ as in Def.\,\ref{def-asso1}.

948: %Then after the blowing-up process $Z_n\to Z_{n-1}\to\cdots \to Z_1\to Z$ explicitly given above, $\Phi_{n-1}^{\mathbf C^*}$ becomes a morphism, at least in a neighborhood of $E_2$ and $\ol{E}_2$, where $E_2$ and $\ol{E}_2$ are the exceptional divisor over $C_2$ and $\ol{C}_2$ respectively.

949: %Further, $\Phi_{n-1}^{\mathbf C^*}:Z\to\mathscr T$  can be lifted to a meromorphic map $Z_n\to\tilde{\mathscr T}$ which is holomorphic on neighborhoods of $E_2$ and $\ol{E}_2$ and its restriction to $E_2$ and $\ol{E}_2$ are isomorphisms.

950: %\end{prop}

951: % 

952: %In this way we have eliminated the base locus of the system $|(n-1)F|^{\mathbf C^*}$, at least in a neighborhood of the curve $C_2\cup\ol{C}_2$.

953: %Hence the composition map $Z_n\to Z\to \mathbf{CP}^{n+1}$ is a morphism in a neighborhood of $E_2$ and $\ol{E}_2$.

954: %We note that all blowing-ups are performed in  $\mathbf C^*$-invariant centers and hence $\mathbf C^*$ acts naturally on $Z_n$.

955: %It can be readily seen that the two divisors $E_2$ and $\ol{E}_2$ are contained in the fixed locus of this $\mathbf C^*$-action.

956: 

957: %Since we blew up $C_1\cup \ol{C}_1\cup C_3\cup\ol{C}_3$ (in $Z_1$) which are contained in the base locus of the system $|\mu_1^*F-E_2-\ol{E}_2|$ ($\mu_1:Z_1\to Z$ is the blowing-up along $C_2\cup\ol{C}_2$),  

958: %the meromorphic map $\Phi_{n-1}^{\mathbf C^*}:Z\to\mathscr T$ naturally lifts to a meromorphic map $Z_n\to \hat{\mathscr T}$ which is holomorphic in a neighborhood of $E_2\cup\ol{E}_2$. (Recall that $\hat{\mathscr T}$ was the blowing-up at the 2 cyclic quotient singularities of $\mathscr T$ of Prop.\,\ref{prop-mt2} (b), which was the base points of a pencil corresponding to the pencil $|F|$ as explained before Prop.\,\ref{prop-mt2}.)

959: %Hence we obtain  holomorphic maps from $E_2$ to $\hat{\mathscr T}$ and $\ol{E}_2$ to   $\hat{\mathscr T}$ as  restrictions.

960: %Then since each member $S\in |F|\simeq\Lambda_{n-1}$ is mapped surjectively to a fiber of $\pi_{n-1}^{\mathbf C^*}$  (see the diagram \eqref{cd3}) and since the intersection $S\cap E_2$ in $Z_1$ (or $Z_n$) clearly covers $E_2$ as $S$ runs through in the pencil $|F|$, 

961: %these 2 morphisms must be surjective.

962: %Further, these morphisms must be generically 1 to 1 since general fibers of the meromorphic map $\Phi_{n-1}^{\mathbf C^*}$ is irreducible as in Prop.\,\ref{prop-quot1}.

963: %Hence the morphisms from $E_2$ and $\ol{E}_2$ to $\hat{\mathscr T}$ are birational.

964: 

965: %We show that these morphisms coincide with the minimal resolution of the unique singularity of $\hat{\mathscr T}$ which is a rational double point of type A$_{n-3}$ (as in Prop.\,\ref{prop-mt2} (b)), and hence $E_2$ and $\ol{E}_2$ (in $Z_n$) are biholomorphic to $\tilde{\mathscr T}$, the minimal resolution of the original quartic surface $\mathscr T$.

966: 

967: % To see this, we first note that $E_2$ and $\ol{E}_1$ in $Z_1$ is biholomorphic to $\mathbf{CP}^1\times\mathbf{CP}^1$.

968: %When we blow-up up $Z_1$ along $C_1\cup C_3\cup\ol{C}_1\cup\ol{C}_3$ to obtain $Z_2$, 

969: %$E_1$ and $\ol{E}_1$ are blown-up at 2 points $\{E_1\cap C_1, E_1\cap C_3\}$ and $\{\ol{E}_1\cap \ol{C}_1, \ol{E}_1\cap \ol{C}_3\}$ respectively.

970: %These 2 points are the base locus of the pencil $|\mu_1^*F-E_2-\ol{E}_2|$ restricted to $E_2$ and $\ol{E}_2$ respectively.

971: %Hence after the blowing-up $Z_2\to Z_1$, these pencils on $E_2$ and $\ol{E}_2$ in $Z_2$ becomes free, and their members are precisely the intersection of the strict transforms of $S\in |F|$ with $E_2$ and $\ol{E}_2$ respectively.

972: %We consider the morphism $E_2\to\mathbf{CP}^1$ associated to this pencil on $E_2$ in $Z_2$.

973: %Obviously this morphism has precisely 2 reducible members.

974: %We see that under the above interpretation of members of the pencil (on $E_2$), these 2 members are the restrictions $(S_2^++S_2^-)|_{E_2}$ and $(S_3^++S_3^-)|_{E_2}$.

975: %In fact, since $S_2^+$ and $S_3^-$ intersect transversally along $C_2$ in $Z$, and the normal bundles of $C_2$ in $S_2^+$ and $S_3^-$ coincide, the restriction $S_2^+|_{E_2}$ and $S_3^-|_{E_2}$ in $Z_1$ are disjoint sections of the projection $E_2\to C_2$ whose self-intersection numbers (in $E_2$) are zero.

976: %On the other hand, since $S_2^-$ in $Z$ intersects $C_2$ transversally at a unique point $C_1\cap C_2$, the restriction $S_2^-|_{E_2}$ in $Z_1$ is a fiber of the projection  $E_2\to C_2$.

977: %Similarly, $S_3^+|_{E_2}$ in $Z_1$ is a different fiber of the projection  $E_2\to C_2$.

978: %Further, we have  $S_2^-\cap S_3^-=C_1$ in $Z_1$ and hence the intersection point $S_2^-\cap S_3^-\cap E_2$ goes through $C_1$ in $Z_1$.

979: %Similarly,  the intersection point $S_2^+\cap S_3^+\cap E_2$ goes through $C_3$ in $Z_1$.

980: %Since $Z_2\to Z_1$ blows-up $C_1$ and $C_3$ in a neighborhood of $E_2$, these imply that $(S_2^++S_2^-)|_{E_2}$ and $(S_3^++S_3^-)|_{E_2}$ are reducible members of the pencil on $E_2$ in $Z_2$.

981: %We also note that $S_1^-$ and $S_i^-$ for $4\le i\le n+1$ is disjoint from $E_2$ in $Z_1$ and $Z_2$, since these are disjoint from $C_2$ in $Z$.

982: %Thus we have understood the structure of the morphism $E_2\to\mathbf{CP}^1$ in $Z_2$.

983: %So next we consider the effect of the blowing-up $Z_3\to Z_2$.

984: %In a neighborhood of $E_2$, this is precisely blowing-up along $(n-1)$ rational curves $S_1^+\cap E_1$ and $S_i^-\cap E_3$ with $4\le i\le n+1$.

985: %All these intersect $E_2$ transversally at a unique point respectively and they lie on mutually different fibers of the morphism $E_2\to\mathbf{CP}^1$. % (since they lie on mutually  different  members of the str

986: %So after blowing up (namely on $Z_3)$, there appear $(n-1)$ new reducible fibers of the morphism $E_2\to\mathbf{CP}^1$.

987: %Thus on $Z_3$, the morphism $E_2\to\mathbf{CP}^1$ has precisely $(n+1)$ reducible fibers.

988: %On the other hand, by the diagram \eqref{cd3}, this morphism corresponds the projection $\hat{\mathscr T}\to\Lambda_{n-1}$.

989: %Further, as in the explanation after the diagram, this projection has precisely $(n+1)$ reducible fibers.

990: %All of them consist of 2 irreducible components since $\hat{\mathscr T}$ is a conic bundle.

991: %Therefore the morphism $E_2\to\hat{\mathscr T}$ in problem is biholomorphic on smooth locus of $\hat{\mathscr T}$.

992: %Further, by the blowing-ups $Z_{n}\to Z_{n-1}\to\cdots \to Z_4$ there arises the exceptional divisors $F_1,F_2,\cdots, F_{n-2}$ (and their conjugates).

993: %Restricted to the divisor $E_2$ in $Z_n$, these exceptional divisors determine a string of $(n-2)$ rational curves.

994: %Among these, the intersection with $F_1,F_2, \cdots, F_{n-3}$ are $(-2)$-curves in $E_2\subset Z_n$.

995: %Hence in $Z_n$, the morphism $E_2\to\mathbf{CP}^1$ has a unique fiber consisting of $(n-1)$ rational curves.

996: %Among these rational curves, precisely one is the intersection $S_1^+\cap E_2$.

997: %On the other hand, the A$_{n-3}$-singularity of the conic bundle $\hat{\mathscr T}\to\Lambda_{n-1}$  is over the point $(0,0,\cdots,0,1)\in\Lambda_{n-1}\subset\mathbf{CP}^{n-1}$, at which the rational normal curve $\Lambda_{n-1}$ touches one of the plane $\{z_2=0\}$ of the right-hand side of \eqref{nt3}.

998: %Since $z_2=y_1^{n-2}y_2$ and since the zero divisor of $y_1\in H^0(F)$ was $S_1^++S_1^-$, the fiber of the conic bundle on which  the A$_{n-3}$-singularity lie is the image of this divisor $S_1^++S_1^-$ (by the map $\Phi_{n-1}^{\mathbf C^*}$).

999: %Further $\Phi_{n-1}^{\mathbf C^*}(S_1^+)\cap\Phi_{n-1}^{\mathbf C^*}(S_1^-)\, (=\Phi_{n-1}^{\mathbf C^*}(L_1))$ is the A$_{n-3}$-singularity.

1000: %Therefore, the image of the above distinguished fiber of  $E_2\to\mathbf{CP}^1$ in $Z_n$ into $\hat{\mathscr T}$ must be the distinguished fiber of $\hat{\mathscr T}\to\Lambda_{n-1}$.

1001: %These mean that $E_2\to \hat{\mathscr T}$ is the minimal resolution of the A$_{n-3}$-singularity, as claimed.

1002: 

1003: \subsection{Description of the discriminant locus of the quotient map}\label{ss-dl}

1004: Next based on Prop.\,\ref{prop-elim1} 

1005: we give projective models of our twistor spaces as conic bundles over the minimal resolution of the minitwistor spaces.

1006: Namely we explicitly construct a $\mathbf{CP}^2$-bundle over the minimal resolution $\tilde{\mathscr T}$, and show that our twistor space $Z$ is bimeromorphically embedded into this bundle as a conic bundle.

1007: We also give a defining equation of the conic bundle.

1008: 

1009: Although the map $\tilde{\Phi}_{n-1}^{\mathbf C^*}:Z_n\to\tilde{\mathscr T}$ in Prop.\,\ref{prop-elim1} is holomorphic on a neighborhood of $E_2\cup\ol{E}_2$, 

1010: it  still has indeterminacy locus because there are still a base locus.

1011: In fact, the curve $C_4\cup\ol{C}_4$ in $Z_n$ is still a base curve and it is not resolved even after blowing-up along $C_4\cup\ol{C}_4$.

1012: (Namely another base locus appears on the exceptional divisors over $C_4$ and $\ol{C}_4$.)

1013: Also, there remain many base curves on the exceptional divisors $E_3$ and $\ol{E}_3$ (in $Z_n$) which have a complicated structure.

1014: To remove these base curves completely, we need a lot of blow-ups and it looks difficult to give them in an explicit form.

1015: We do not persist in them and  take any  sequence of blowing-ups $\tilde Z\to Z_n$ along $\mathbf C^*$-invariant non-singular centers which eliminates the  base locus of the system.

1016: We can suppose that all centers of the blow-ups are disjoint from $E_2\cup\ol{E}_2$ since $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ is already holomorphic on these divisors.

1017: We denote the resulting morphism by $\tilde{\Phi}^{\mathbf C^*}:\tilde{Z}\to\tilde{\mathscr T}$.

1018: General fibers of $\tilde{\Phi}^{\mathbf C^*}$ are $\mathbf C^*$-invariant irreducible rational curves by Prop.\,\ref{prop-quot1}, and  $E_2$ and $\ol{E}_2$ are sections of $\tilde{\Phi}^{\mathbf C^*}$ which are fixed by the $\mathbf C^*$-action on $\tilde Z$.

1019: We consider the direct image sequence of the exact sequence 

1020: \begin{align}

1021: 0\,\lra\, \mathscr O_{\tilde Z}\,\lra\,\mathscr O_{\tilde Z}(E_2+\ol{E}_2)\,\lra\,N_{E_2/\tilde{Z}}\oplus N_{\ol{E}_2/\tilde{Z}}\,\lra\,0.

1022: \end{align}

1023: Since every fiber of $\tilde{\Phi}^{\mathbf C^*}$ is at most a string of rational curves,

1024: we have $R^1\tilde\Phi^{\mathbf C^*}\mathscr O_{\tilde Z}=0$.

1025: Hence a part of the direct image sequence becomes

1026: \begin{align}\label{ses50}

1027: 0\,\lra\, \mathscr O_{\tilde{\mathscr  T}}\,\lra\,(\tilde\Phi^{\mathbf C^*})_*\mathscr O_{\tilde Z}(E_2+\ol{E}_2)\,\lra\,N_{E_2/\tilde{Z}}\oplus N_{\ol{E}_2/\tilde{Z}}\,\lra\,0,

1028: \end{align}

1029: where $N_{E_2/\tilde{Z}}$ and $ N_{\ol{E}_2/\tilde{Z}}$ are considered as line bundles over $\tilde{\mathscr T}$.

1030: On the other hand there are obvious isomorphisms $N_{E_2/\tilde{Z}}\simeq N_{E_2/Z_n}$ and $N_{\ol{E}_2/\tilde{Z}}\simeq N_{\ol{E}_2/Z_n}$.

1031: Further since our sequence of blow-ups $Z_n\to Z_{n-1}\to \cdots\to Z_1\to Z$ are explicit, we can concretely compute the normal bundles $N_{E_2/Z_n}$ and $ N_{\ol{E}_2/Z_n}$.

1032: In particular, basically by the reason that the degree of $N_{E_2/Z_n}$ along general fibers of the natural projection $E_2\to C_2$ is $-1$, we obtain 

1033: \begin{align}

1034: H^1(N_{E_2/Z_n})=H^1(N_{\ol{E}_2/Z_n})=0.

1035: \end{align}

1036: These imply that the exact sequence \eqref{ses50} splits and we obtain an isomorphism

1037: \begin{align}

1038: (\tilde\Phi^{\mathbf C^*})_*\mathscr O(E_2+\ol{E}_2)

1039: \simeq

1040: N_{E_2/Z_n}\oplus N_{\ol{E}_2/Z_n}\oplus\mathscr O_{\tilde{\mathscr T}}.

1041: \end{align}

1042: Then let

1043: \begin{equation}\label{p2bdle}

1044: \mu: \tilde{Z}\lra\mathbf{P}( N_{E_2/Z_n}^{\vee}\oplus N_{\ol{E}_2/Z_n}^{\vee}\oplus\mathscr O)

1045: \end{equation}

1046: be the relative meromorphic map over $\tilde{\mathscr T}$ associated to the pair $\{\mathscr O(E_2+\ol{E}_2),\tilde{\Phi}^{\mathbf C^*}\}$.

1047: Obviously $\mu$ is bimeromorphic over its image and

1048: the image $\mu(\tilde Z)$ is a conic bundle (over $\tilde{\mathscr T}$).

1049: The discriminant locus of the projection $\mu(\tilde{Z})\to \tilde{\mathscr T}$ is a member of the system $|N_{E_2/Z_n}^{\vee}\otimes N_{\ol{E}_2/Z_n}^{\vee}|$.

1050: Note that this system can be explicitly determined.

1051: For the purpose of determining the discriminant curves, we show the following.

1052: 

1053: \begin{lemma}\label{lemma-hps1}

1054: Let $L_0$ be the fixed twistor line in $Z$ (Def.\,\ref{def-ftl}). 

1055: Then we have the following.

1056: (i) $L_0$ is disjoint from the base locus of the system $|(n-1)F|^{\mathbf C^*}$, and the image

1057: $\mathscr C_0:=\Phi_{n-1}^{\mathbf C^*}(L_0)$ is a curve on $\mathscr T$. % which do not goes through singular points of $\mathscr T$.

1058: (ii) $\mathscr C_0$ is a hyperplane section of $\mathscr T$ with respect to the natural embedding $\mathscr T\subset\mathbf{CP}^{n+1}$.

1059: (iii) The virtual genus of the curve $\mathscr C_0$ is $n-2$.

1060: %(iii) We have $(\Phi_{n-1}^{\mathbf C^*})^{-1}(\mathscr C_0)=Y+\ol{Y}$,  where $Y$ and $\ol{Y}$ are $\mathbf C^*$-invariant irreducible divisors intersecting transversally along the $\mathbf C^*$-fixed twistor line $L_0$.

1061: %(iv) The degrees of $Y$ and $\ol{Y}$ are $(n-1)$.

1062: \end{lemma}

1063: \noindent

1064: Since $\mathscr C_0$ is a rational curve, (iii) means that $\mathscr C_0$ is a singular curve.

1065: 

1066: \noindent

1067: Proof of Lemma \ref{lemma-hps1}.

1068: By Prop.\,\ref{prop-generator1}, we have Bs\,$|(n-1)F|^{\mathbf C^*}=C-C_{n+1}-\ol{C}_{n+1}$.

1069: $L_0$ is not contained in this base curve since $L_0$ is itself real.

1070: If $L_0$ intersects this base curve, the intersection must be a $\mathbf C^*$-fixed point.

1071: However, twistor lines going through such a point  on $C$ are not $\mathbf C^*$-fixed, as in Prop.\,\ref{prop-u1}.

1072: Hence $L_0\cap$\,Bs\,$|(n-1)F|^{\mathbf C^*}=\emptyset$.

1073: Then the image $\mathscr C_0$ of $L_0$ cannot be a point since $(n-1)F\cdot L_0=2(n-2)\neq 0$. 

1074: Therefore $\mathscr C_0$ is a curve and we obtain (i).

1075: 

1076: To show (ii) we set $\tilde{\mathscr C}_0:=\nu^{-1}(\mathscr C_0)$ and we determine the cohomology class of $\tilde{\mathscr C}_0$ on $\tilde{\mathscr T}$.

1077: As generators of the cohomology group $H^2(\tilde{\mathscr T},\mathbf Z)\simeq\mathbf Z^{2n}$, 

1078: we choose the following $2n$ curves:

1079: \begin{align}\label{gen1}

1080: \Gamma,\,f,\,s_3^+, d_i\,(4\le i\le n+1),\,s_2^-,\,f_j\,(1\le j\le n-2),

1081: \end{align}

1082: where  $f$ is a fiber of the conic bundle $\tilde{\mathscr T}\to\Lambda_{n-1}$, $s_i^{\pm}$ are the images of the degree-one divisors $S_i^{\pm}\subset \tilde Z$ under $\tilde{\Phi}^{\mathbf C^*}$ or $\tilde{\Phi}_{n-1}^{\mathbf C^*}$, $f_j$ and $d_i$ are the images of the exceptional divisors $F_j$ and $D_i$ respectively (arose  in obtaining $Z_n$) under the same map, and $\Gamma$ is one of the 2 exceptional curves of the partial resolution $\hat{\mathscr T}\to\mathscr T$ specified by the property that it intersects $s_2^-$. (See Figure \ref{fig-mt}.)

1083: Then since $L_0$ intersects $S$ transversally at 2 points for general $S\in |F|$ and 

1084: since $\tilde{\Phi}_{n-1}^{\mathbf C^*}|_S$ is holomorphic quotient map by Prop.\,\ref{prop-quot1}, we have $\tilde{\mathscr C}_0\cdot f=2$.

1085: Similarly, we have $\tilde{\mathscr C}_0\cdot d_i=\tilde{\mathscr C}_0\cdot s_3^+=\tilde{\mathscr C}_0\cdot s_2^-=\tilde{\mathscr C}_0\cdot f_{n-2}=1$ and $\tilde{\mathscr C}_0\cdot\Gamma=\tilde{\mathscr C}_0\cdot f_j=0$ for $1\le j\le n-3$.

1086: From this we can deduce that, as cohomology classes,

1087: \begin{align}\label{cohom1}

1088: \tilde{\mathscr C}_0\sim 2\Gamma+(n-1)f+s_2^-+\sum_{k=1}^{n-2}kf_k-s_3^+-\sum_{j=4}^{n+1}d_j.

1089: \end{align}

1090: Since we have an explicit realization of $\mathscr T$  as an embedded surface in $\mathbf{CP}^{n+1}$ as in Theorem\,\ref{thm-mt1}, we can check that the cohomology class \eqref{cohom1} is precisely the pullback of the hyperplane section class, by the minimal resolution $\nu$.

1091: Hence we obtain (ii).

1092: Finally, by  \eqref{cohom1} we obtain

1093: \begin{align}

1094: \tilde{\mathscr C}_0^2=2n-2,\,\,\,K\cdot\tilde{\mathscr C}_0=-4

1095: \end{align}

1096: on $\tilde{\mathscr T}$.

1097: From this it follows that the virtual genus of $\tilde{\mathscr C}_0$ is $n-2$.

1098: Since $\tilde{\mathscr C}_0$ and $\mathscr C_0$ are biholomorphic by $\nu$, 

1099: the virtual genus are same.

1100: Thus we obtain (iii).

1101: \proofend

1102: %Further, because $\mathscr C_0$ does not meet  singular points of $\mathscr T$, 

1103: 

1104: \begin{figure}

1105: \includegraphics{ds_mt_small.eps}

1106: \caption{Reducible fibers of the projection $\tilde{\mathscr T}\to\Lambda_{n-1}$}

1107: \label{fig-mt}

1108: \end{figure}

1109: 

1110: \begin{prop}\label{prop-discrim1}

1111: The discriminant locus of the conic bundle $\mu(\tilde Z)\to\tilde{\mathscr T}$ consists of the following curves.

1112: (a) The 2 exceptional curves $\Gamma$ and $\ol{\Gamma}$ of the partial resolution $\hat{\mathscr T}\to\mathscr T$.

1113: (b) Reducible fibers of the projection $\tilde{\mathscr T}\to\Lambda_{n-1}$ over the $(n-1)$ points $(0,0\cdots,0,1)$ and $(1,\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})$ for $4\le i\le n+1$

1114: in the homogeneous coordinate $(z_1,\cdots,z_n)$ on $\mathbf{CP}^{n-1}$ as before.

1115: (c) The curve $\tilde{\mathscr C}_0=\nu^{-1}(\mathscr C_0)$, where $\mathscr C_0=\Phi_{n-1}^{\mathbf C^*}(L_0)$ as in Lemma \ref{lemma-hps1}, and $\nu:\tilde{\mathscr T}\to\mathscr T$ is the minimal resolution as before.

1116: \end{prop}

1117: 

1118: Recall that in (b) the fiber over the point $(0,0\cdots,0,1)\in\Lambda_{n-1}$ is a string of $(n-1)$ rational curves. Other reducible fibers consist of 2 irreducible components.

1119: Note that the proposition claims that reducible fibers of $\tilde{\mathscr T}\to\Lambda_{n-1}$ over the 2 points $(1,\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})$ for $i=2,3$ do not contained in the discriminant locus of $\mu(\tilde Z)\to\tilde{\mathscr T}$.

1120: 

1121: \noindent

1122: Proof of Prop.\,\ref{prop-discrim1}. By our explicit way of the elimination for the indeterminacy locus in  neighborhoods of $C_2$ and $\ol{C}_2$, we have

1123: $\{\Gamma,\ol{\Gamma}\}=\{\tilde{\Phi}_{n-1}^{\mathbf C^*}(E_1),\tilde{\Phi}_{n-1}^{\mathbf C^*}(\ol{E}_1)\}$.

1124: We distinguish $\Gamma$ and $\ol\Gamma$ by supposing $\tilde{\Phi}_{n-1}^{\mathbf C^*}(E_1)=\Gamma$.

1125: (This is compatible with the previous distinction we made in \eqref{gen1}).

1126: Then we have $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\Gamma)=E_1+\ol{E}_3$ (in $Z_n$), and both $E_1$ and $\ol{E}_3$ are mapped surjectively to the curve $\Gamma$ by $\tilde{\Phi}_{n-1}^{\mathbf C^*}$.

1127: Hence $\Gamma$ is contained in the discriminant locus of $\tilde{\Phi}_{n-1}^{\mathbf C^*}$.

1128: Since the full elimination $\tilde{\Phi}^{\mathbf C^*}:\tilde{Z}\to\tilde{\mathscr T}$ factors through $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ by our choice, the inverse image $(\tilde{\Phi}^{\mathbf C^*})^{-1}(\Gamma)$ contains $E_1$ and $\ol{E}_3\,(\subset \tilde Z)$ at least.

1129: Hence  $\Gamma$ is contained in the  discriminant locus of $\tilde{\Phi}^{\mathbf C^*}$ as well.

1130: Further, since the blow-ups $\tilde{Z}\to Z_n$ do not touch the divisor $E_2$ by our choice, 

1131: $E_1$ and $\ol{E}_3$ still intersect $E_2$ in $\tilde{Z}$, along  curves which are mapped biholomorphically to $\Gamma$ by $\tilde{\Phi}^{\mathbf C^*}$.

1132: Therefore, $E_1$ and $\ol{E}_3$ are not contracted to  curves by the map $\mu$.

1133: This means that the conic bundle $\mu(\tilde{Z})\to\tilde{\mathscr T}$ contains $\Gamma$ as a discriminant curve.

1134: By reality, the same is true for the conjugate curve $\ol{\Gamma}$.

1135: 

1136: Next, to see that the fibers in (b) are also contained in the discriminant locus, 

1137: we again recall that the fibers of the projection $\tilde{\mathscr T}\to\Lambda_{n-1}$ are the images of the members of the pencil $|F|$.

1138:  In particular, the reducible  fiber over the point $(1,\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})\in\Lambda_{n-1}$ is the image of the member $S_i^++S_i^-$ for $2\le i\le n+1$, and the curves

1139:  $\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^+)$ and $\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^-)$ are precisely the irreducible components of the fiber.

1140:  Then we see again by the explicit way of blowing-ups that we have, for $4\le i\le n+1$,

1141:  \begin{align}

1142: (\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^+))=S_i^++D_i

1143: \,\,\,{\text{and }}\,\,\,\,

1144: (\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^-))=S_i^-+\ol{D}_i.

1145: \end{align}

1146: (Note that for $i=2, 3$ these do not hold because $D_2, D_3,\ol{D}_2$ and $\ol{D}_3$ do not exist.

1147: Instead, $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^+))=S_i^+$ and $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_i^-))=S_i^-$ hold for $i=2,3$.

1148: This is why the fibers over  the points $(1,\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})$ are not contained in the discriminant locus of $\mu(\tilde Z)\to\tilde{\mathscr T}$ for $i=2,3$.)

1149: Hence by the same reasoning for $\Gamma$,

1150: the fibers over $(1,\lambda_i,\lambda_i^2,\cdots,\lambda_i^{n-1})\in\Lambda_{n-1}$, $4\le i\le n+1$, are discriminant curves for $\mu(\tilde{Z})\to\tilde{\mathscr T}$.

1151: On the other hand, the reducible  fiber over the point $(0,0\cdots,0,1)\in\Lambda_{n-1}$ corresponds to the reducible member $S_1^++S_1^-\in|F|$.

1152: The curve $\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_1^+)$ (resp.\,$\tilde{\Phi}_{n-1}^{\mathbf C^*}(F_{n-2})$) is the component of the reducible fiber which intersects $\ol{\Gamma}$ (resp.\,$\Gamma$).

1153: We have

1154:  \begin{align}

1155: (\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_1^+))=S_1^++\ol{F}_{n-2}

1156: \,\,\,{\text{and }}\,\,\,\,

1157: (\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(S_1^-))=S_1^-+F_{n-2},

1158: \end{align}

1159: and $S_1^+$, $\ol{F}_{n-2}$, $S_1^-$ and $F_{n-2}$ are mapped surjectively to the irreducible components.

1160: Similarly, we have 

1161:  \begin{align}

1162: (\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\Phi}_{n-1}^{\mathbf C^*}(F_i))=F_i+\ol{F}_{n-2-i}

1163: \end{align}

1164: for $1\le i\le n-3$, and $F_i$ and $\ol{F}_i$ are mapped surjectively to the components.

1165: These mean that all the irreducible components of the fiber over $(0,0,\cdots,0,1)\in\Lambda_{n-1}$ are discriminant curves of $\tilde{\Phi}_{n-1}^{\mathbf C^*}$.

1166: By the same reasoning for $\Gamma$ and $\ol{\Gamma}$ above, this is still true for the conic bundle  $\mu(\tilde{Z})\to\tilde{\mathscr T}$.

1167: Thus we have seen that all fibers in (b) are actually discriminant curves.

1168: 

1169: 

1170: Next we show that $\tilde{\mathscr C}_0$ is also a discriminant curve.

1171: Since $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ can be viewed as a $\mathbf C^*$-quotient map, the inverse image $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\mathscr C}_0)$ is a $\mathbf C^*$-invariant divisor in $Z_n$, which clearly contains the fixed twistor line $L_0$.

1172: Further, by the explicit form of $U(1)$-action in a neighborhood of the fixed point on $n\mathbf{CP}^2$ under $L_0$,

1173: the $U(1)$-action on the twistor space in a neighborhood of any point of $L_0$ takes the form 

1174: \begin{align}

1175: (u,v,w)\mapsto (su,s^{-1}v, w),\,\,s\in U(1)

1176: \end{align}

1177: where $L_0=\{u=v=0\}$ in the neighborhood.

1178: Hence any $\mathbf C^*$-invariant divisor containing $L_0$ must contain at least one of the (locally defined) divisors $\{u=0\}$ and $\{v=0\}$.

1179: Moreover, since the map $\tilde{\Phi}_{n-1}^{\mathbf C^*}$ is continuous in a neighborhood of $L_0$ (since it is holomorphic there), these two divisors must be mapped to the same curve in $\tilde{\mathscr T}$.

1180: Hence the divisor $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\mathscr C}_0)$ contains both of the two divisors $\{u=0\}$ and $\{v=0\}$.

1181: This means that $(\tilde{\Phi}_{n-1}^{\mathbf C^*})^{-1}(\tilde{\mathscr C}_0)$ has ordinary double point along $L_0$.

1182: (Later on this will turned out to be decomposed into 2 irreducible components in $Z_n$.)

1183: Hence $\tilde{\mathscr C}_0$ is a discriminant curve of $\tilde{\Phi}_{n-1}^{\mathbf C^*}$.

1184: Therefore the same is true for the conic bundle 

1185:  $\mu(\tilde{Z})\to\tilde{\mathscr T}$.

1186:  

1187:  Thus we have seen that all curves in (a), (b) and (c) are actually discriminant curves.

1188:  Finally we show that there is no discriminant curve other than these.

1189:  As is already noted,

1190:  the discriminant curve is a member of the system $|N_{E_2/Z_n}^{\vee}\otimes N_{\ol{E}_2/Z_n}^{\vee}|$. 

1191: Thus it is enough to show that the sum of all curves in (a), (b) and (c) already belongs to this system.

1192: But since the normal bundles and all curves are explicitly given, this can again be verified by routine computations of intersection numbers.

1193: \proofend

1194: 

1195: 

1196: \subsection{Existence of non-trivial members and their Chern classes}\label{ss-nt}

1197: Using Prop.\,\ref{prop-discrim1}, we can readily obtain $\mathbf C^*$-invariant divisors in the twistor spaces which will play an essential role in our analysis of the structure of the twistor spaces as follows.

1198: (In the course of the proof, we obtain a projective model of the twistor space as a conic bundle.)

1199: 

1200: \begin{prop}\label{prop-Y1}

1201: Let $\mathscr C_0$ be as in Lemma\,\ref{lemma-hps1}.

1202: Then the inverse image $(\Phi_{n-1}^{\mathbf C^*})^{-1}(\mathscr C_0)$ splits into 2 irreducible components $Y$ and $\ol{Y}$  intersecting transversally along the fixed twistor line $L_0$.

1203: Further, the degrees of $Y$ and $\ol{Y}$  are $n-1$.

1204: \end{prop}

1205: 

1206: \noindent Proof.

1207: First we recall that the 2 divisors $E_2$ and $\ol{E}_2$ in $\tilde Z$ are contained in the $\mathbf C^*$-fixed locus.

1208: Since the $\mathbf C^*$-action on $\tilde Z$ is effective and non-trivial, we can suppose that  it acts on the normal bundle $N_{E_2/\tilde{Z}}$ by usual scalar multiplication on each fiber.

1209: Hence by using the reality, the natural $\mathbf C^*$-action on the bundle 

1210: $N_{E_2/\tilde{Z}}^{\vee}\oplus N_{\ol{E}_2/\tilde{Z}}^{\vee}\oplus\mathscr O$ is given by

1211: \begin{align}

1212: (x,y,t)\longmapsto (sx,s^{-1}y, t),\,\,\,s\in\mathbf C^*.

1213: \end{align}

1214: where $(x,y,t)$ represents points of the bundle.

1215: Since the conic bundle $\mu(\tilde Z)$ is $\mathbf C^*$-invariant,

1216: its defining equation in $\mathbf P(N_{E_2/\tilde{Z}}^{\vee}\oplus N_{\ol{E}_2/\tilde{Z}}^{\vee}\oplus\mathscr O)$ must be of the form

1217: \begin{align}\label{cbdle2}

1218: xy=P_0P_1t^2,

1219: \end{align}

1220: where $P_0$ is a section of a line bundle over $\tilde{\mathscr T}$ whose zero divisor is $\tilde{\mathscr C}_0$, and $P_1$ is a section of a line bundle whose zero divisor is the sum of all discriminant curves in (a) and (b) of Prop.\,\ref{prop-discrim1}.

1221: The equation \eqref{cbdle2} immediately implies that the inverse image of the curve $\tilde{\mathscr C}_0$ splits into 2 irreducible components

1222: $\{x=P_0=0\}$ and $\{y=P_0=0\}$, which are clearly mutually conjugate.

1223: Hence the corresponding divisor $(\Phi_{n-1}^{\mathbf C^*})^{-1}(\mathscr C_0)$ (in $Z$) splits into two irreducible components, for which we  denote by  $Y$ and $\ol{Y}$.

1224: Then these $Y$ and $\ol{Y}$ (in $Z$)  intersect transversally along $L_0$ since it is true already in $\mu(\tilde Z)$ and since the blow-ups $\tilde Z\to Z_n\to\cdots \to Z_1\to Z$ do not touch $L_0$.

1225: Finally, since $\mathscr C_0$ is a hyperplane section as in Lemma \ref{lemma-hps1}, $Y+\ol{Y}$  belongs to the system $|(n-1)F|^{\mathbf C^*}$.

1226: Hence its degree is $2(n-1)$. Hence by reality the degrees of $Y$ and $\ol{Y}$ are both $(n-1)$.

1227: \proofend

1228: 

1229: 

1230: 

1231: \begin{rmk}{\em

1232: As in the proof, the equation \eqref{cbdle2} gives a projective model of our twistor spaces as a conic bundle.

1233: This is an analogue of the equation of the conic bundles of the twistor spaces, obtained in \cite{Hon07-3}.

1234: A remarkable difference between the two cases is the following.

1235: In \cite{Hon07-3}, the structure of the surface (minitwistor space) $\mathscr T$ or its resolution $\tilde{\mathscr T}$ does not depend on $n$.

1236: Instead, the number of the irreducible components of the discriminant curve increases as $n$ does.

1237: In contrast, for the present twistor spaces, 

1238: the structure of $\mathscr T$ and $\tilde{\mathscr T}$ changes depending on $n$. In fact, the minimal resolution $\tilde{\mathscr T}$ has a conic bundle structure over the rational curve $\Lambda_{n-1}$, which has precisely  $(n+1)$ reducible fibers.

1239: (As is clear from the construction, this reflects the fact that the number of reducible members of the pencil $|F|$ on the twistor spaces is $(n+1)$, which in turn is based on the number of irreducible components of the anticanonical cycle $C$. In \cite{Hon07-3}, the cycle $C$ always consists of 8 components for any $n$.)

1240: }

1241: \end{rmk}

1242: 

1243: Next we would like to determine the Chern classes of the divisors $Y$ and $\ol{Y}$. For this purpose we prove the following.

1244: (Note that we have not made distinction between $Y$ and $\ol{Y}$ yet.)

1245: 

1246: \begin{lemma}\label{lemma-Y2}

1247: Let $Y$ and $\ol{Y}$ be as in Prop.\,\ref{prop-Y1} and $S$ a real  irreducible member of the pencil $|F|$. Then the restriction $Y|_S$ or $\ol{Y}|_S$ coincides with the following curve on $S$:

1248: \begin{align}\label{rest1}

1249: (n-2)C_1+\sum_{i=2}^n(n+1-i)C_i\,\,+B_1+B_2.

1250: \end{align}

1251: \end{lemma}

1252: \noindent

1253: (For $B_1$ and $B_2$, see the initial construction of our surface $S$ given in \S \ref{ss-S}.) 

1254: 

1255: \noindent

1256: Proof of Lemma \ref{lemma-Y2}.

1257: Since $Y+\ol{Y}\in |(n-1)F|^{\mathbf C^*}$ as in the proof of Prop.\,\ref{prop-Y1} and 

1258: Bs\,$|(n-1)F|^{\mathbf C^*}=C-C_{n+1}-\ol{C}_{n+1}$ by Prop.\,\ref{prop-generator1}, 

1259: $Y+\ol{Y}$ contains $C_i$ and $\ol{C}_i$ for $1\le i\le n$.

1260: We first show that the component $C_2$ (and $\ol{C}_2$ also) cannot be contained in both of $Y$ and $\ol{Y}$ simultaneously.

1261: For this, we recall that as in the proof of Prop.\,\ref{prop-Y1}, $Y$ and $\ol Y$ are bimeromorphic images of the divisors $\{x=P_0=0\}$ and $\{y=P_0=0\}$ in the conic bundle $\mu(\tilde Z)$.

1262: In  $\mu(\tilde Z)$, the 2 sections $E_2$ and $\ol E_2$ (of $\mu(\tilde Z)\to\tilde{\mathscr T}$) are defined by $\{x=t=0\}$ and $\{y=t=0\}$, where we have not yet specified whether $E_2=\{x=t=0\}$ or $E_2=\{y=t=0\}$ holds.

1263: So here we suppose that  $E_2=\{x=t=0\}$ holds.

1264: We also suppose that $Y$ and $\ol Y$ are bimeromorphic images of $\{x=P_0=0\}$ and $\{y=P_0=0\}$ respectively.

1265: Then $\{x=P_0=0\}\cap E_2$ in $\mu(\tilde Z)$ is a curve which is biholomorphic to $\tilde{\mathscr C}_0$ by the projection $\mu (\tilde Z)\to\tilde{\mathscr T}$, and evidently $\{y=P_0=0\}\cap {E}_2=\emptyset$ in  $\mu(\tilde Z)$.

1266: Similarly $\{x=P_0=0\}\cap \ol{E}_2=\emptyset$ and $\{y=P_0=0\}\cap \ol E_2\simeq \tilde{\mathscr C}_0$ in  $\mu(\tilde Z)$.

1267: Then since the two bimeromorphic maps $\mu:\tilde Z\to \mu(\tilde Z)$ and $\tilde Z\to Z_n$ are biholomorphic in a neighborhood of $E_2$ and $\ol E_2$, in $Z_n$ also, $Y\cap E_2\simeq\tilde{\mathscr C}_0$ and $\ol Y\cap E_2=\emptyset$ hold.

1268: These mean that in $Z$ also, $Y\supset C_2$ and $\ol Y\cap C_2=\emptyset$ from the above equations.

1269: 

1270: Next we show by using the computations in the proof of Lemma \ref{lemma-hps1} that the restriction $Y|_S$ contains the curve $C_2$ with multiplicity  $(n-1)$. 

1271: For this, we recall that the exceptional divisor $E_2\subset \tilde{Z}$ is biholomorphic to the surface $\tilde{\mathscr T}$ by the map $\tilde{\Phi}^{\mathbf C^*}:\tilde Z\to\tilde{\mathscr T}$.

1272: Let $h\in H^2(E_2,\mathbf Z)\simeq H^2(\tilde{\mathscr T},\mathbf Z)$ be the cohomology class of fibers for the natural projection $E_2\to C_2$.

1273: Then since $Y+\ol Y=(\Phi^{\mathbf C^*}_{n-1})^{-1}(\mathscr C_0)$ and $C_2\not\subset \ol Y$,  in order to see that $Y|_S$ contains $(n-1)C_2$, 

1274: it suffices to show that

1275: \begin{align}\label{kemul}

1276: h\cdot \tilde{\mathscr C}_0=n-1.

1277: \end{align}

1278: Using the $2n$ curves \eqref{gen1} as a basis of $ H^2(\tilde{\mathscr T},\mathbf Z)$, it can be verified that we have

1279: \begin{align}\label{cohom2}

1280: h\sim \Gamma+s_2^-+\sum_{k=1}^{n-2}kf_k.

1281: \end{align}

1282: On the other hand, the cohomology class of $\tilde{\mathscr C}_0$ is given by \eqref{cohom1}.

1283: By \eqref{cohom1} and \eqref{cohom2} we obtain \eqref{kemul}.

1284: Hence 

1285: the restriction $Y|_S$ contains the curve $C_2$ with multiplicity $(n-1)$.

1286: 

1287: We next show that the divisor $(Y+\ol{Y})|_S$ contains the curve

1288: \begin{align}\label{rest3}

1289: (n-2)(C_1+\ol{C}_1)+\sum_{i=2}^n(n+1-i)(C_i+\ol{C}_i),

1290: \end{align}

1291: by using intersection numbers as a main tool.

1292: (The curve \eqref{rest3} is exactly (\eqref{rest1}$-B_1-B_2)$ plus its conjugate curve).

1293: Since we have

1294: $

1295: Y+\ol{Y}\supset C-C_{n+1}-\ol{C}_{n+1}

1296: $ and $(Y+\ol Y)|_S\supset (n-1)(C_1+\ol C_1)$ as before, 

1297:  the curve 

1298: \begin{align}\label{cohom3}

1299: \left(Y+\ol{Y}\right)|_S-\left\{

1300: (C_1+\ol{C}_1)+(n-1)(C_2+\ol{C}_2)+\sum_{i=3}^n(C_i+\ol{C}_i)

1301: \right\}

1302: \end{align}

1303: must be a zero divisor or an effective curve (on $S$).

1304: Further, by using $(Y+\ol{Y})|_S=(n-1)K_S^{-1}$, the intersection number of the curve \eqref{cohom3} and $C_1$ can be computed to be

1305: \begin{align}

1306: (n-1)(2-n) +(n-1)=(n-1)(3-n).

1307: \end{align}

1308: Since this is negative, the curve \eqref{cohom3}  contains $C_1$.

1309: By the same computations, we see that the curve \eqref{cohom3}  contains $C_3$ too.

1310: Hence by reality, it  contains $\ol{C}_1+\ol{C}_3$ also.

1311: So subtracting $(C_1+\ol{C}_1)+(C_3+\ol{C}_3)$ from \eqref{cohom3}, we obtain that 

1312: \begin{align}\label{cohom4}

1313: \left(Y+\ol{Y}\right)|_S-\left\{

1314: 2(C_1+\ol{C}_1)+(n-1)(C_2+\ol{C}_2)+2(C_3+\ol{C}_3)+\sum_{i=4}^n(C_i+\ol{C}_i)

1315: \right\}

1316: \end{align}

1317: is still effective, or a zero divisor.

1318: If $n=4$, this already proves the claim that $(Y+\ol{Y})|_S$ contains the curve \eqref{rest3}.

1319: If $n\ge 5$,  the intersection numbers of the class \eqref{cohom4} with $C_1$ and $C_3$ can be computed to be $(n-1)(4-n)$ and $(4-n)$ respectively, both of which are negative.

1320: Therefore, subtracting $(C_1+\ol{C}_1)+(C_3+\ol{C}_3)$ from \eqref{cohom4}

1321: and then computing its intersection number with $C_4$ which turns out to be $(-2)<0$, it follows that 

1322: the class 

1323: \begin{align}\label{cohom5}

1324: \left(Y+\ol{Y}\right)|_S-\left\{

1325: 3(C_1+\ol{C}_1)+(n-1)(C_2+\ol{C}_2)+3(C_3+\ol{C}_3)+2(C_4+\ol{C}_4)+\sum_{i=5}^n(C_i+\ol{C}_i)

1326: \right\}

1327: \end{align}

1328: is still effective, or a zero divisor.

1329: If $n=5$, this proves the claim that $(Y+\ol{Y})|_S$ contains the curve \eqref{rest3}.

1330: For general $n$, by repeating this argument we obtain that 

1331: $(Y+\ol{Y})|_S$ contains the curve \eqref{rest3}.

1332: 

1333: Next we show that the remaining curves $B_1,B_2,\ol B_1$ and $\ol{B}_2$ are also contained in $(Y+\ol{Y})|_S$.

1334: But this is obvious since the restriction of $\Phi_{n-1}^{\mathbf C^*}:Z\to\mathscr T$ onto $S\in |F|$ is a holomorphic quotient map by Prop.\,\ref{prop-quot1}, and since both $Y$ and $\ol{Y}$ are the unions of the 2 reducible  fibers of these quotient maps consisting of  2 irreducible components.

1335: Thus we have proved that $(Y+\ol Y)|_S$ contains the curve \eqref{rest1} and its conjugate curve.

1336: 

1337: Next we see that this inclusion is moreover an equality.

1338: For this, by computing intersection numbers, we can verify that \eqref{rest1} plus its conjugate curve is a member of the system $|(n-1)K_S^{-1}|$.

1339: On the other hand, we have $Y+\ol Y\in |(n-1)F|$ and $(n-1)F|S\simeq (n-1)K_S^{-1}$.

1340: Further the restriction map $H^2(Z,\mathbf Z)\to H^2(S,\mathbf Z)$ is injective.

1341: These imply the coincidence.

1342: 

1343: Finally we show that the curve $Y|_S$ is exactly \eqref{rest1}.

1344: For this it is enough to see that $\ol Y$ does not contain the components $C_i$ for $4\le i\le n$.

1345: (Recall that we have already seen that $C_1\not\subset\ol Y$ and $C_3\not\subset Y$.)

1346: By our explicit construction of the surface $S$, we see that $\mathbf C^*$ acts on the chain $C_3+C_4+\cdots+C_{n+1}$ in such a way that if $z\in C_i$ is not a fixed point then $sz\in C_i$ goes to the fixed point $C_i\cap C_{i+1}$ as $s\to\infty$ (since $C_2$ and $\ol C_2$ are `source' and `sink'), where $C_{n+2}:=\ol C_1$.

1347: Since $\ol Y$ actually contains $\ol C_1$, this means that if $\ol Y$ contains $C_i$ ($3\le i\le n$), then $\ol Y$ contains $C_j$ for $i\le j\le n+1$.

1348: However, we already know that $Y+\ol Y$, and hence $\ol Y$ does not contain $C_{n+1}$.

1349: \proofend

1350: 

1351: \bigskip

1352: We are now ready to prove the key result, which means the existence of  `non-trivial' members of the system $|(n-2)F|$ (in the sense of Def.\,\ref{def-trivial}):

1353: 

1354: 

1355: \begin{prop}\label{prop-nt4}

1356: Let $Y$ and $\ol{Y}$ be the $\mathbf C^*$-invariant divisors as in Prop.\,\ref{prop-Y1}.

1357: Then both of the two $\mathbf C^*$-invariant divisors

1358: \begin{align}\label{nt11}

1359: Y+(n-3)S_1^-\, \,\,{\text{and}}\,\,\,\,\,

1360: \ol{Y}+(n-3)S_1^+

1361: \end{align}

1362: belong to the linear system $ |(n-2)F|$.

1363: Further, if  $x_1$ and $x_2\in H^0((n-2)F)$ are mutually conjugate  sections defining divisors 

1364: \eqref{nt11} respectively, $\mathbf C^*$  acts on these by $(x_1,x_2)\mapsto(sx_1,s^{-1}x_2)$ or  $(x_1,x_2)\mapsto(s^{-1}x_1,sx_2)$ for $s\in\mathbf C^*$.

1365: \end{prop}

1366: 

1367: \noindent

1368: Proof.

1369: Since $S_1^-|_S=\sum_{i=1}^{n+1}\ol{C}_i$, 

1370:  Lemma \,\ref{lemma-Y2} means 

1371: \begin{align}\label{rest2}

1372: (Y+(n-3)S_1^-)|_S=(n-2)C_1+\sum_{i=2}^n(n+1-i)C_i\,\,+B_1+B_2+(n-3)\sum_{i=1}^{n+1}\ol{C}_i.

1373: \end{align}

1374: It is a routine computation to verify that the right-hand side of \eqref{rest2} belongs to the system $|(n-2)K_S^{-1}|$.

1375: (For example, it is enough to show the coincidence of their intersection numbers with the curves $\{C_i, \ol{C}_i,  B_j, \ol{B}_j\set 1\le i\le n+1,j=1,2\}$, which generate $H^2(S,\mathbf Z)$.)

1376: Then since the restriction map $H^2(Z,\mathbf Z)\to H^2(S,\mathbf Z)$ is injective and $(n-2)F|_S\simeq (n-2)K_S^{-1}$, we obtain $Y+(n-3)S_1^-\in |(n-2)F|$.

1377: Since $(n-2)F$ is a real bundle we also have $\ol Y+(n-3)S_1^+\in |(n-2)F|$.

1378: Furthermore, it can be verified that the curve \eqref{rest2} is the inverse image of a $\mathbf C^*$-invariant but non-real line in $\mathbf{CP}^2$, by the morphism associated to the system $|(n-2)K_S^{-1}|$.

1379: Therefore by Prop.\,\ref{prop-multan1} (v), precisely one of $x_1$ and $x_2$ is acted by a scalar multiplication of $\mathbf C^*$.

1380: With the aid of the reality, this implies the final claim of the proposition.

1381: \proofend

1382: 

1383: 

1384: 

1385: \section{Bimeromorphic images of the twistor spaces}

1386: In Section 2 we considered the system $|(n-1)F|^{\mathbf C^*}$ and showed that the image of its associated map is a complex surface $\mathscr T$ whose defining equation can be explicitly determined.

1387: The map could also be regarded as a quotient map  by $\mathbf C^*$-action

1388: and $\mathscr T$ can be regarded as an orbit space (i.\,e.\,minitwistor space).

1389: In Section 3 we studied this map in detail and  finally found  non-trivial members of  the system $|(n-2)F|$ (Prop.\,\ref{prop-nt4}).

1390: Once we obtain these non-trivial members, it is possible to show that 

1391: the map $\Phi_{n-2}$ (associated to $|(n-2)F|$)  is generically 2 to 1 covering onto its image.

1392: However, it seems difficult yet to derive a detailed form of a defining equation of the discriminant locus of the covering.

1393: %These members  are not contained in the system $|(n-2)F|^{\mathbf C^*}$.

1394: 

1395: To remedy this, in this section, we investigate the complete system $|(n-1)F|$ and show that its associate map $\Phi_{n-1}$ is bimeromorphic onto its image.

1396: We further  give  defining equations of the image  in a projective space.

1397: It provides another projective model of our twistor spaces which is different from the conic bundle description in Section 3.

1398: The equations will be used in order to derive a defining equation of the branch divisor of the (generically) double covering map $\Phi_{n-2}$.

1399: 

1400: %Though these projective models are given in an explicit form, its geometric structure is not so apparent from the description.

1401: %In particular, it is not clear at all why the present twistor spaces can be regarded as a generalization of twistor spaces on $3\mathbf{CP}^2$ of `double solid type' into $n\mathbf{CP}^2$, $n$ arbitrary.

1402: %So we consider meromorphic map associated to the system $|(n-2)F|$ and show that its image is the preimage of rational normal curve $\Lambda_{n-2}\subset\mathbf{CP}^{n-2}$ under the a projection $\mathbf{CP}^n\to\mathbf{CP}^{n-2}$.

1403: %If we blow-up the center of the projection (which is a line in $\mathbf{CP}^{n}$) the image becomes isomorphic to the total space of the $\mathbf{CP}^2$-bundle 

1404: %\begin{equation}

1405: %\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)\lra \Lambda_{n-2}\simeq\mathbf{CP}^1.

1406: %\end{equation}

1407: 

1408: First we give  generators of the system $|(n-2)F|$  explicitly as follows.

1409: \begin{prop}\label{prop-dc1}

1410: Let $Z$ be a twistor space on $n\mathbf{CP}^2$ which has the complex surface $S$ constructed in \S \ref{ss-S} as a real member of the system $|F|$ as before.

1411: Then we have the following.

1412: (i) $\dim |(n-2)F|=n$,

1413: (ii) As generators of the system, we can choose the following divisors.

1414: (a) generators of the system $|V_{n-2}|$,

1415: (b) 

1416: $(x_1)=Y+(n-3)S_1^-$ and $(x_2)=\ol{Y}+(n-3)S_1^+$.

1417: (iii) {\rm Bs}\,$|(n-2)F|=C-C_{n+1}-\ol{C}_{n+1}$.

1418: %(iii) The meromorphic map $\Phi_{n-2}:Z\to\mathbf{CP}^n$ associated to the system 

1419: 

1420: \end{prop}

1421: 

1422: 

1423: 

1424: \noindent

1425: Proof.

1426: By the cohomology sequence of the sequence \eqref{ses3}, we obtain an exact sequence

1427: \begin{align}

1428: 0\,\lra\,H^0((n-3)F)\,\lra H^0((n-2)F)\,\lra H^0((n-2)K_S^{-1}).

1429: \end{align}

1430: By Prop.\,\ref{prop-fundamental} and Prop.\,\ref{prop-multan1} (i) we have $H^0((n-3)F)=V_{n-3}=\mathbf C^{n-2}$.

1431: Further by Prop.\,\ref{prop-multan1} (ii) we have $\dim H^0((n-2)K_S^{-1})=3$.

1432: These imply $\dim H^0((n-2)F)\le n+1$.

1433: 

1434: Since the divisor $Y+(n-3)S_1^-$ is a non-trivial member,  $|V_{n-2}|$ and $Y+(n-3)S_1^-$ generate $(n-1)$-dimensional subsystem of $|(n-2)F|$.

1435: So to prove (i) and (ii) it suffices to show that the remaining divisor $\ol{Y}+(n-3)S_1^+$ is not contained in this $(n-1)$-dimensional subsystem.

1436: But this is obvious if we note that $\mathbf C^*$ acts trivially on $V_{n-2}$ and we have $(x_1,x_2)\mapsto (sx_1,s^{-1}x_2)$ or $(s^{-1}x_1,sx_2)$ by Prop.\,\ref{prop-nt4}, so that $x_2$ cannot be a linear combination of elements of $V_{n-2}$ and $x_1$.

1437: 

1438: For (iii), since  Bs$\,|V_{n-2}|=C$, it follows Bs\,$|(n-2)F|\subset C$.

1439: Hence we have

1440: \begin{equation}

1441: {\rm Bs}\,|(n-2)F|=C\cap (Y+(n-3)S_1^-)\cap (\ol Y+(n-3)S_1^+).

1442: \end{equation}

1443: By Lemma  \ref{lemma-Y2}, the right-hand side is seen to be exactly $C-C_{n+1}-\ol C_{n+1}$.

1444: \proofend

1445: 

1446: \bigskip

1447: By using this proposition we can determine the structure of the system $|(n-1)F|$ as follows.

1448: \begin{prop}\label{prop-bim1}

1449: Let $Z$ be a twistor space on $n\mathbf{CP}^2$ 

1450: as in Prop.\,\ref{prop-dc1}.

1451: Then we have the following.

1452: (i) $\dim |(n-1)F|=n+5$,

1453: (ii) As generators of the system, we can choose the following divisors.

1454: (a) generators of the $(n-1)$-dimensional system $|V_{n-1}|$,

1455: (b) the 2   non-trivial members of $|(n-1)F|^{\mathbf C^*}$ given in Lemma \ref{lemma-nt1}.

1456: (c) the 4 divisors defined by the following sections of $(n-1)F$:

1457: \begin{align}

1458: z_{n+3}:=x_1y_1,\,\,\,z_{n+4}:=x_2y_1,\,\,\,

1459: z_{n+5}:=x_1y_2,\,\,\,z_{n+6}:=x_2y_2.

1460: \end{align}

1461: where $x_1$ and $x_2$ are mutually conjugate sections of $(n-2)F$ determining non-trivial members as in Prop.\,\ref{prop-nt4}.

1462: (iii) 

1463: {\rm Bs}\,$|(n-1)F|=C-C_{n+1}-\ol C_{n+1}$.

1464: %\begin{equation}

1465: %(z_{n+1}):=(n-2)S_1^++\sum_{i=2}^{n+1}S_i^+,\,\,\,

1466: %(z_{n+2}):=(n-2)S_1^-+\sum_{i=2}^{n+1}S_i^-,

1467: %\end{equation}

1468: %\begin{equation}

1469: %(z_{n+3}):=Y+(n-3)S_1^-+(S_1^++S_1^-),\,\,\,(z_{n+4}):=\ol{Y}+(n-3)S_1^++(S_1^++S_1^-)

1470: %\end{equation}

1471: %\begin{equation}

1472: %(z_{n+5}):=Y+(n-3)S_1^-+(S_2^++S_2^-),\,\,\,(z_{n+6}):=\ol{Y}+(n-3)S_1^++(S_2^++S_2^-)

1473: %\end{equation}

1474: \end{prop}

1475: 

1476: \noindent

1477: Proof.

1478: From the sequence \eqref{ses1} with $m=n-1$, we obtain an exact sequence

1479: \begin{align}

1480: 0\,\lra\,H^0((n-2)F)\,\lra H^0((n-1)F)\,\lra H^0((n-1)K_S^{-1}).

1481: \end{align}

1482: We have $\dim H^0((n-2)F)=n+1$ by Prop.\,\ref{prop-dc1} and $\dim H^0((n-1)K_S^{-1})=5$ by Prop.\,\ref{prop-multan2}.

1483: These imply $\dim H^0((n-1)F)\le n+6$.

1484: To prove (i) and (ii), it suffices to show that  $V_{n-1}=\mathbf C^n$ and $\{z_{n+i}\set 1\le i\le 6\}$ are linearly independent, where

1485:  $z_{n+1}$ and $z_{n+2}$ denote mutually conjugate sections of $(n-1)F$ defining the 2 non-trivial members \eqref{nt1} (as in Theorem\,\ref{thm-mt1}).

1486: We first show that $V_{n-1}$ and $z_{n+i}$, $1\le i\le 4$, generate $(n+4)$-dimensional subspace of $H^0((n-1)F)$.

1487: By Prop.\,\ref{prop-generator1} we have $\dim (V_{n-1}+\mathbf C\langle z_{n+1},z_{n+2}\rangle)=n+2$.

1488: From the explicitness we can verify that restriction of any member of the corresponding $(n+1)$-dimensional subsystem to $S\in |F|$ contains the curve $\ol C_2$ with multiplicity $(n-1)$.

1489: On the other hand the restriction of the divisor $(z_{n+3})$ ($=Y+(n-3)S_1^-+S_1^++S_1^-)$ 

1490: contains $\ol C_2$  with multiplicity only $(n-2)$.

1491: These imply $z_{n+3}\not\in V_{n-1}\oplus \mathbf C\langle z_{n+1},z_{n+2}\rangle$.

1492: Similarly, considering multiplicity along the component $C_2$, we obtain $z_{n+4}\not\in V_{n-1}\oplus \mathbf C\langle z_{n+1},z_{n+2},z_{n+3}\rangle$.

1493: Hence $\dim (V_{n-1}+\mathbf C\langle z_{n+i}\set 1\le i\le 4\rangle)=n+4$.

1494: 

1495: We next show that $z_{n+5}\not\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 4\rangle$ and $z_{n+6}\not\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 5\rangle$ using $\mathbf C^*$-action.

1496: By Prop.\,\ref{prop-nt4} we have either $(x_1,x_2)\mapsto (sx_1,s^{-1}x_2)$ or 

1497:  $(x_1,x_2)\mapsto (s^{-1}x_1,sx_2)$ for $s\in\mathbf C^*$.

1498: We may suppose that the former holds.

1499: Then since $y_1$ and $y_2$ belong $H^0(F)=H^0(F)^{\mathbf C^*}$ we have

1500: $(z_{n+3},z_{n+4},z_{n+5},z_{n+6})\mapsto (sz_{n+3},s^{-1}z_{n+4},sz_{n+5},s^{-1}z_{n+6})$ for $s\in\mathbf C^*$.

1501: From this it readily follows  that  $z_{n+5}\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 4\rangle$ means $z_{n+5}\in \mathbf Cz_{n+3}$.

1502:  Since $(z_{n+3})\neq(z_{n+5})$, this is a contradiction and we obtain $z_{n+5}\not\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 4\rangle$.

1503: Similarly, $z_{n+6}\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 5\rangle$ implies $z_{n+6}\in \mathbf Cz_{n+4}$ and this is also a contradiction.

1504: Hence we obtain $z_{n+6}\not\in V_{n-1}\oplus\mathbf C\langle z_{n+i}\set 1\le i\le 5\rangle$.

1505: Thus we obtain (i) and (ii).

1506: 

1507: Finally since we have Bs\,$|(n-1)F|^{\mathbf C^*}=C-C_{n+1}-\ol{C}_{n+1}$ by Prop.\,\ref{prop-generator1}, we have Bs\,$|(n-1)F|\subset C-C_{n+1}-\ol{C}_{n+1}$.

1508: Further, all the 4 divisors $(z_{n+i})$, $3\le i\le 6$, contain the cycle $C$.

1509: These mean Bs\,$|(n-1)F|= C-C_{n+1}-\ol{C}_{n+1}$ and we obtain (iii).

1510: \proofend

1511: 

1512: \begin{thm}\label{thm-bim}

1513: Let $Z$ be a twistor space on $n\mathbf{CP}^2$ as in Prop.\,\ref{prop-dc1} and $\Phi_{n-1}:Z\to\mathbf{CP}^{n+5}$ the meromorphic map associated to the linear system $|(n-1)F|$.

1514: Then $\Phi_{n-1}$ is bimeromorphic onto its image $X:=\Phi_{n-1}(Z)$.

1515: Further, for a homogeneous coordinate $(z_1,\cdots,z_{n+6})$,  defining equations of the image $X$ are given by the following.

1516: (a) equations in the defining ideal of the rational normal curve $\Lambda_{n-1}\subset\mathbf P^{\vee}V_{n-1}=\mathbf{CP}^{n-1}$ whose homogeneous coordinate is $(z_1,\cdots,z_n)$.

1517: (b) the following 5 quadratic equations.

1518: \begin{align}

1519: z_1z_{n+5}&=z_2z_{n+3},\label{nt12}\\

1520: z_1z_{n+6}&=z_2z_{n+4},\label{nt13}\\

1521: z_{n+1}z_{n+2}&=z_1\{z_n-\sigma_1z_{n-1}+\sigma_2z_{n-2}-\cdots+(-1)^{n-1}\sigma_{n-1}z_1\}

1522: ,\label{nt14}\\

1523: z_{n+3}z_{n+4}&=z_1g(z_1,\cdots,z_{n+2})\label{nt15}

1524: \end{align}

1525: where  $g$ is a linear polynomial of $z_1,\cdots,z_{n+2}$, and $\sigma_i$ are the elementary symmetric polynomials of $\lambda_3,\cdots,\lambda_{n+1}$ as in Theorem\,\ref{thm-mt1}.

1526: \end{thm}

1527: 

1528: \noindent

1529: Proof.

1530: By the diagram \eqref{cd1} with $m=n-1$, the image $\Phi_{n-1}(Z)$ satisfies all equations in the defining ideal of $\Lambda_{n-1}\subset\mathbf P^{\vee}V_{n-1}$.

1531: Further, since the restriction map $H^0((n-1)F)\to H^0((n-1)K_S^{-1})$ is surjective as proved in Prop.\,\ref{prop-bim1}, 

1532: the restriction of $\Phi_{n-1}$ onto $S\in |F|$  coincides with the meromorphic map associated to $|(n-1)K_S^{-1}|$.

1533: Since the latter map is actually (holomorphic and) birational onto its image by Prop.\,\ref{prop-multan2}, it follows from the diagram \eqref{cd1} that $\Phi_{n-1}$ is bimeromorphic onto its image.

1534: 

1535: Let $\{y_1,y_2\}$ and $\{z_1,\cdots,z_{n+2}\}$ be the same meaning as in  Theorem\,\ref{thm-mt1} and its proof.

1536:  The latter gives a homogeneous coordinate on $\mathbf P^{\vee}H^0((n-1)F)^{\mathbf C^*}=\mathbf{CP}^{n+1}$. 

1537:  By Theorem\,\ref{thm-mt1} these satisfy \eqref{nt14} (which is the same as \eqref{nt3}).

1538: Further let  $\{x_1,x_2\}$ and  $\{z_{n+3},z_{n+4},z_{n+5},z_{n+6}\}$  be as in Prop.\,\ref{prop-bim1}.

1539: Then we have

1540: \begin{align}

1541: z_1z_{n+5}=y_1^{n-1}\cdot x_1y_2=y_1^{n-2}y_2\cdot x_1y_1=z_2z_{n+3}

1542: \end{align}

1543: and we obtain \eqref{nt12}.

1544: Similarly we have

1545: \begin{align}

1546: z_1z_{n+6}=y_1^{n-1}\cdot x_2y_2=y_1^{n-2}y_2\cdot x_2y_1=z_2z_{n+4}

1547: \end{align}

1548: and we obtain \eqref{nt13}.

1549: Next we have

1550: \begin{align}\label{40}

1551: z_{n+3}z_{n+4}=x_1x_2y_1^2.

1552: \end{align}

1553: Now,  since $Y+\ol{Y}\in |(n-1)F|^{\mathbf C^*}$,  there is a linear polynomial $g(z_1,\cdots,z_{n+2})$ satisfying 

1554: \begin{align}\label{g2}

1555: \left(g\right)=Y+\ol{Y}.

1556: \end{align}

1557: Therefore 

1558: since $(x_1x_2)=Y+\ol{Y}+(n-3)(S_1^++S_1^-)$ and $(y_1)=S_1^++S_1^-$, we obtain

1559: \begin{align}

1560: x_1x_2=g\cdot y_1^{n-3}.

1561: \end{align}

1562: By multiplying $y_1^2$ on both-hand sides, we obtain

1563: \begin{align}

1564: x_1y_1\cdot x_2y_1=gy_1^{n-1}.

1565: \end{align}

1566: Hence by \eqref{40} we obtain \eqref{nt15}.

1567: \proofend

1568: 

1569: \bigskip

1570: The structure of the bimeromorphic image $X$ becomes clearer if we blow-up $\mathbf{CP}^{n+5}$ along the center of the projection $\pi_{n-1}:\mathbf{CP}^{n+5}\to\mathbf{CP}^{n-1}$.

1571: The center is $\mathbf{CP}^5$ which is explicitly given by $\{z_1=\cdots=z_n=0\}$ in the coordinate of Theorem \ref{thm-bim}.

1572: By the blowing-up, $\mathbf{CP}^{n+5}$ becomes biholomorphic to the total space of the $\mathbf{CP}^6$-bundle $\mathbf P(\mathscr O(1)^{\oplus 6}\oplus \mathscr O)\to\mathbf{CP}^{n-1}$.

1573: Restricting over $\Lambda_{n-1}$ whose degree is $(n-1)$, we obtain the bundle $\mathbf P(\mathscr O(n-1)^{\oplus 6}\oplus \mathscr O)\to\Lambda_{n-1}=\mathbf{CP}^1$.

1574: Let $\hat X$ be the strict transform of $X=\Phi_{n-1}(Z)$ into this $\mathbf{CP}^6$-bundle.

1575: Putting $\xi_i=z_{n+i}/z_1$ for $1\le i\le 6$ and using $(\xi_1,\cdots,\xi_6)\in\mathscr O(1)^{\oplus 6}$ as a non-homogeneous fiber coordinate over $\Lambda_{n-1}\backslash\{(0,\cdots,0,1)\}$, the equation of $\hat X$ is given by

1576: \begin{align}

1577: \xi_5&=\lambda\xi_3,\label{nt16}\\

1578: \xi_6&=\lambda\xi_4,\label{nt17}\\

1579: \xi_1\xi_2&=\lambda (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1}),\label{nt18}\\

1580: \xi_3\xi_4&=g(1,\lambda,\cdots,\lambda^{n-1},\xi_1,\xi_2),\label{nt19}

1581: \end{align}

1582: where $\lambda=z_2/z_1=y_2/y_1$ is a non-homogeneous coordinate on $\Lambda_{n-1}\backslash\{(0,\cdots,0,1)\}$ as in Theorem\,\ref{thm-mt1}.

1583: The 2 equations \eqref{nt16} and \eqref{nt17} determine a $\mathbf{CP}^4$-subbundle in the $\mathbf{CP}^6$-bundle.

1584: In each fiber of this $\mathbf{CP}^4$-bundle,  \eqref{nt18} and \eqref{nt19} determine a quartic surface.

1585: If $\lambda\in\mathbf R$ and if $\lambda\neq\lambda_i$ for $2\le i\le n+1$ (recall $\lambda_2=0$), 

1586: this quartic surface is nothing but the birational image of the corresponding real irreducible member $S\in|F|$, by the map associated to $|(n-1)K_S^{-1}|$.

1587: Thus, the bimeromorphic image $X=\Phi_{n-1}(Z)$ is bimeromorphic to a fiber space over $\Lambda_{n-1}=\mathbf{CP}^1$ whose general fibers are irreducible quartic surfaces.

1588: If $\lambda=\lambda_i$ for $2\le i\le n+1$ or $\lambda=\infty$, then the right-hand side of \eqref{nt18} vanishes and consequently the fiber degenerates into 2 irreducible components.

1589: Of course, these are the images of reducible members $S_i^++S_i^-$ for $1\le i\le n+2$.

1590: 

1591: \section{Projective models as generically double coverings}

1592: In this section by using the results in the previous section we investigate  the meromorphic map $\Phi_{n-2}$ (associated to the system $|(n-2)F|$ on the twistor space) and show that it gives a  generically 2 to 1 covering onto its image.

1593: Next we derive a defining equation of the discriminant locus of the double covering by using defining equations of the image $X=\Phi_{n-1}(Z)$ obtained in Theorem \ref{thm-bim}.

1594: 

1595: In order to investigate a relation between two meromorphic maps  $\Phi_{n-1}$ and $\Phi_{n-2}$, we consider an injection $H^0((n-2)F)\ra H^0((n-1)F)$ given by $\zeta\mapsto \zeta\otimes y_1$, where $y_1\in H^0(F)$ satisfies $(y_1)=S_1^++S_1^-$ as before.

1596: Let 

1597: \begin{align}

1598: f:\mathbf P^{\vee}H^0((n-1)F)=\mathbf{CP}^{n+5}\,\lra\,\mathbf P^{\vee}H^0((n-2)F)=\mathbf{CP}^{n}

1599: \end{align}

1600: be the projection induced from the injection.

1601: %Then there is a commutative diagram 

1602: %\begin{equation}\label{cd5}

1603: % \CD

1604: %Z@>{\Phi_{n-1}}>>\mathbf{CP}^{n+5} \\

1605: % @| @VV{f}V\\

1606: %Z@>{\Phi_{n-2}}>>\mathbf{CP}^n\\

1607: % \endCD

1608: % \end{equation}

1609: Recall that by Prop.\,\ref{prop-dc1} we can choose a set of sections

1610: \begin{align}

1611: y_1^{n-2},\,y_1^{n-3}y_2,\cdots,y_2^{n-2},\, x_1,\,x_2

1612: \end{align}

1613: as a basis of $H^0((n-2)F)\simeq\mathbf C^{n+1}$.

1614: By taking (tensor) products with $y_1$ we obtain a set of sections 

1615: \begin{align}

1616: z_1=y_1^{n-1},\,z_2=y_1^{n-2}y_2,\cdots,z_{n-1}=y_1y_2^{n-2},\, z_{n+3}=x_1y_1,\,z_{n+4}=x_2y_1

1617: \end{align}

1618: which is a part of a basis of $H^0((n-1)F)$ given in Prop.\,\ref{prop-bim1}.

1619: Hence the projection $f$ is explicitly given by

1620: \begin{align}\label{proj1}

1621: (z_1,\cdots,z_n, z_{n+1},\cdots,z_{n+6})\longmapsto (z_1,\cdots,z_{n-1}, z_{n+3},z_{n+4}).

1622: \end{align}

1623: Combining various meromorphic maps appeared so far, we obtain the following commutative diagram

1624: \begin{equation}\label{cd-big}

1625: \xymatrix{

1626: \mathbf{CP}^{n+5}\ar[rr]^{\pi_{n-1}}\ar[dd]_f& &\mathbf{CP}^{n-1}\ar[dd]\\

1627: &Z\ar[lu]^{\Phi_{n-1}}\ar[ru]_{\Psi_{n-1}}\ar[ld]_{\Phi_{n-2}}\ar[rd]^{\Psi_{n-2}}&\\

1628: \mathbf{CP}^{n}\ar[rr]_{\pi_{n-2}} & &\mathbf{CP}^{n-2}

1629: }

1630: \end{equation}

1631: where $\Psi_{n-1}(Z)=\Lambda_{n-1}$ and $\Psi_{n-2}(Z)=\Lambda_{n-2}$.

1632: Of course the right projection $\mathbf{CP}^{n-1}\to\mathbf{CP}^{n-2}$ is the one induced by the inclusion $V_{n-2}\to V_{n-1}$ given by $\zeta\mapsto\zeta\otimes y_1$,

1633: and it is explicitly given by $(z_1,\cdots,z_n)\mapsto(z_1,\cdots,z_{n-1})$.

1634: This maps the curve $\Lambda_{n-1}$ isomorphically to $\Lambda_{n-2}$.

1635: By blowing-up along the centers of $\pi_{n-1}$ and $\pi_{n-2}$, and then restricting over $\Lambda_{n-1}$ and $\Lambda_{n-2}$, we obtain the diagram

1636: \begin{equation}\label{cd7}

1637:  \CD

1638: \mathbf{P}(\mathscr O(n-1)^{\oplus 6}\oplus\mathscr O)@>>>\Lambda_{n-1} \\

1639:  @V{\hat f}VV \hspace{-5mm}@VVV\\

1640: \mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)@>>>\Lambda_{n-2}\\

1641:  \endCD

1642:  \end{equation}

1643:  where $\hat f$ is the meromorphic map induced from $f$.

1644: The indeterminacy locus of  $\hat f$ is $\mathbf{CP}^3$-subbundle of the $\mathbf{CP}^6$-bundle.

1645: This locus actually intersects the strict transform $\hat X$ of $X$  (as in the final part of the previous section).

1646: Hence $\hat f$ has indeterminacy locus on $\hat X$.

1647: If we blow-up along the $\mathbf{CP}^3$-subbundle, we obtain a morphism $\tilde f:\tilde X\to\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)$, where $\tilde X$ denotes the strict transform of $\hat X$ into the blown-up space.

1648:  

1649:  

1650:  We put 

1651:  \begin{equation}\label{eta}

1652:  \eta_1=z_{n+3}/z_1,\,\,\,\eta_2=z_{n+4}/z_1

1653:  \end{equation}

1654:   and use $(\eta_1,\eta_2)$ as a non-homogeneous fiber coordinate on the $\mathbf{CP}^2$-bundle in \eqref{cd7}.

1655:  

1656: We have now reached the main result of this paper.

1657:  

1658:  

1659: \begin{thm}\label{thm-dc2}

1660: Let $Z$ be a twistor space on $n\mathbf{CP}^2$ as in Prop.\,\ref{prop-dc1} and $\Phi_{n-2}:Z\to\mathbf{CP}^n$ the meromorphic map associated to the system $|(n-2)F|$ as before.

1661: Then we have the following.

1662:  (i)  $\Phi_{n-2}(Z)=(\pi_{n-2})^{-1}(\Lambda_{n-2})$ holds. 

1663: In other words, if $\hat X$ denotes the strict transform of the bimeromorphic image $X=\Phi_{n-1}(Z)$,

1664: we have  $\hat f(\hat X)=\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)$.

1665: % the strict transform of $\Phi_{n-2}(Z)$ into $\mathbf{P}(\mathscr O(1)^{\oplus 2}\oplus\mathscr O)\to\mathbf{CP}^{n-2}$ is exactly the bundle

1666: %$\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)\to\Lambda_{n-2}$ obtained by the restriction.

1667: (ii) The surjective morphism $\tilde f:\tilde X\to\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)$ (obtained above) is generically 2 to 1.

1668: (iii) In the above non-homogeneous coordinate $(\eta_1,\eta_2)$ on the $\mathbf{CP}^2$-bundle, the defining equation of the branch divisor $B$ of $\tilde f$ is given by the following equation.

1669: \begin{align}\label{branch1}

1670: \left\{

1671: \eta_1\eta_2-\hat{g}(\lambda)

1672: \right\}^2

1673: =\lambda(\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1})

1674: \end{align}

1675: where $\hat{g}(\lambda)$ is a polynomial with real coefficients whose degree is at most $(n-1)$.

1676: \end{thm} 

1677: 

1678: \begin{rmk}\label{rmk-3}

1679: {\em

1680: As we will see below, the polynomial $\hat{g}$ in \eqref{branch1} is obtained from the polynomial $g$ in \eqref{nt15} by putting

1681: \begin{equation}\label{dcp5}

1682: g(1,\lambda,\cdots,\lambda^{n-1},\xi_1,\xi_2)=\hat g(\lambda)+c\xi_1+\ol{c}\xi_2,

1683: \end{equation}

1684: where $c$ is a non-zero constant. %Here, we have used the reality of $g_2$ to mean

1685: }

1686: \end{rmk}

1687: 

1688: \begin{rmk}\label{rmk-4}

1689: {\em

1690: The image $\Phi_{n-2}(Z)=(\pi_{n-2})^{-1}(\Lambda_{n-2})$ is a rational scroll of planes in $\mathbf{CP}^n$ and has a cyclic quotient singularities along the center (= a line) of the projection $\pi_{n-2}$, where the order of the cyclic group is $n-2$.

1691: }

1692: \end{rmk}

1693: 

1694: 

1695: \begin{rmk}

1696: {\em

1697: If $n=3$, the equation \eqref{branch1} becomes 

1698: \begin{align}

1699: \left\{

1700: \eta_1\eta_2-\hat{g}(\lambda)

1701: \right\}^2

1702: =\lambda(\lambda-\lambda_3)(\lambda-\lambda_4)

1703: \end{align}

1704: where $\hat g$ is a quadratic polynomial.

1705: %By applying a linear fractional transformation fixing $\infty$, we may suppose $\lambda_3=1$.

1706: In effect, this is exactly the equation of the branch quartic surface we  obtained in \cite{Hon07-2}.

1707: In this sense, the present twistor spaces can be regarded as a generalization of the twistor spaces on $3\mathbf{CP}^2$ of `double solid type' into $n\mathbf{CP}^2$, $n$ arbitrary.

1708: }

1709: \end{rmk}

1710: 

1711: \noindent Proof of Theorem \ref{thm-dc2}.

1712: %By the diagrams \eqref{cd-big} and \eqref{cd7}, it is obvious that $\hat f(\hat X)$ is contained in the bundle $\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)\to\Lambda_{n-2}$.

1713: Since the restriction map $H^0((n-2)F)\to H^0((n-2)K_S^{-1})$ is surjective as proved in Prop.\,\ref{prop-dc1}, the restriction of $\Phi_{n-2}$ onto $S\in|F|$ coincides with the rational map associated to the system $|(n-2)K_S^{-1}|$.

1714: By Prop.\,\ref{prop-multan1} the latter map gives a generically 2 to 1 covering 

1715: onto $\mathbf{CP}^2$.

1716: Hence by the diagram \eqref{cd1} with $m=n-2$ we obtain that the map $\hat f:\hat X\to\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)$ is surjective and generically 2 to 1.

1717: These mean (i) and (ii).

1718: 

1719: Next to prove (iii) we use the non-homogeneous fiber coordinates $(\xi_1,\cdots,\xi_6)$ and $(\eta_1,\eta_2)$ on 

1720: the bundles $\mathbf P(\mathscr O(n-1)^{\oplus 6}\oplus \mathscr O)$ and 

1721: $\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus \mathscr O)$ respectively.

1722: In these coordinates, by \eqref{proj1}, the projection $\hat f$ is explicitly given by

1723: \begin{align}\label{proj2}

1724: \hat f:(\xi_1,\xi_2,\xi_3,\xi_4,\xi_5,\xi_6)\longmapsto(\eta_1,\eta_2)=(\xi_3,\xi_4).

1725: \end{align}

1726: We recall that  defining equations of $\hat X$ are explicitly given by \eqref{nt16}--\eqref{nt19}.

1727: Let $(\eta_1,\eta_2)$ be a point. Then by \eqref{proj2}, a point $(\xi_1,\cdots,\xi_6)$ belongs to $\hat f^{-1}(\eta_1,\eta_2)$ iff $\xi_3=\eta_1$ and $\xi_4=\eta_2$ hold.

1728: Moreover by \eqref{nt16}--\eqref{nt19} if the point belongs to $\hat X$ further, we have

1729: \begin{align}

1730: \xi_1\xi_2&=\lambda (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1}),\label{nt20}\\

1731: \eta_1\eta_2&=g(1,\lambda,\cdots,\lambda^{n-1},\xi_1,\xi_2).\label{nt21}

1732: \end{align}

1733: Therefore the point $(\eta_1,\eta_2)$ belongs to the branch point of $\tilde f:\tilde X\to\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus \mathscr O)$ iff the equations \eqref{nt20} and \eqref{nt21} have a unique solution, viewed as equations for $(\xi_1,\xi_2)$.

1734: If we write $g$ as in  \eqref{dcp5}, \eqref{nt21} is equivalent to $\xi_2=\ol{c}^{-1}\{\eta_1\eta_2-\hat g(\lambda)-c\xi_1\}.$ 

1735: Substituting this into \eqref{nt20} and multiplying $\ol c$ to its both-hand sides, we obtain

1736: \begin{align}

1737: c\xi_1^2-\{\eta_1\eta_2-\hat g(\lambda)\}\xi_1+\ol{c}\lambda  (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1})=0

1738: \end{align}

1739: This has a unique solution iff the discriminant vanishes:

1740: \begin{align}\label{disc5}

1741: \{\eta_1\eta_2-\hat g(\lambda)\}^2-4|c|^2\lambda  (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1})=0.

1742: \end{align}

1743: Now recalling that $g(z_1,\cdots,z_{n+2})$ was originally a section of the line bundle $(n-1)F$ whose zero divisor is $Y+\ol{Y}$ as in \eqref{g2}, we may suppose that $|c|=1/2$ by multiplying a non-zero constant.

1744: Thus the equation \eqref{disc5} becomes

1745:  \begin{align}\label{disc6}

1746: \{\eta_1\eta_2-\hat g(\lambda)\}^2=\lambda (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1}).

1747: \end{align}

1748: Hence we obtain (iii).

1749: \proofend

1750:  

1751:  \bigskip  

1752:  The structure of the branch divisor $B$ in Theorem \ref{thm-dc2} is described as follows.

1753: 

1754: \begin{prop}\label{prop-B}

1755: %Let $Z$ be a twistor space   as in Theorem \ref{thm-dc2} and 

1756: Let $B$ be the branch divisor  defined by the equation \eqref{branch1}.

1757: Then we have the following.

1758: (i) $B$ is irreducible and birational to a ruled surface of genus $[(n-1)/2]$, where 

1759: $[k]$ denotes the biggest integer not greater than $k$.

1760: (ii) Fibers of the natural projection $B\to\mathbf{CP}^1$ (induced from the projection $\mathbf{P}(\mathscr O(n-2)^{\oplus 2}\oplus\mathscr O)\to\mathbf{CP}^1)$ is 

1761: non-reduced iff $\lambda=\lambda_i$ for $2\le i\le n+1$ or $\lambda=\infty$.

1762: Moreover, the surface $B$ has rational double points of type $A_{3n-9}$ along the fiber over $\lambda=\infty$ which is a sum two lines in $\mathbf{CP}^2$.

1763: \end{prop}

1764:  

1765:  Since our equation \eqref{branch1} of $B$ is explicit, it is not difficult to derive the conclusions. We leave it to the interested reader. Note that (ii) means that if $n\ge 4$,  $B$ is non-normal.

1766: 

1767:  \bigskip

1768:  Next we show that the polynomial $\hat g(\lambda)$ in the defining equation \eqref{branch1} of the branch divisor must satisfy certain constraint.

1769:  It is a generalization of a constraint appeared in \cite[Prop.\,2.3 or Condition (A)]{Hon07-2}, which was expressed in terms of a double root of a polynomial.

1770:  

1771:  \begin{prop}\label{prop-dr}

1772:  Let $Z$ be a twistor space on $n\mathbf{CP}^2$  as in Theorem \ref{thm-dc2}, 

1773:  whose projective model $\hat X$ is realized as a double covering of the bundle $\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus \mathscr O)\to\mathbf{CP}^1$ branched along a surface \eqref{branch1}. 

1774: Consider  the double covering of $\mathbf{CP}^1$ whose branch locus is given by

1775: \begin{align}\label{branch2}

1776: \hat g(\lambda)^2-\lambda (\lambda-\lambda_3)(\lambda-\lambda_4)\cdots(\lambda-\lambda_{n+1})=0.

1777: \end{align}

1778: Then this double covering is a rational curve.

1779:  \end{prop}

1780:  

1781:  \noindent

1782:  Proof.

1783:  Let $L_0\subset Z$ be the fixed line as before.

1784:  Then since Bs\,$|(n-2)F|=C-C_{n+1}-\ol{C}_{n+1}$ by Prop.\,\ref{prop-dc1},

1785: the meromorphic map  $\Phi_{n-2}$ is holomorphic on $L_0$.

1786: Further, since $(n-2)F\cdot L_0=2(n-2)\neq 0$, the image $\Phi_{n-2}(L_0)$ cannot be a point.

1787: Hence $\Phi_{n-2}(L_0)$ is a $\mathbf C^*$-fixed curve in $\mathbf{CP}^n$.

1788: %Let $\mathbf{CP}^1_{\infty}\subset \mathbf{CP}^n$ be the center of the projection $\pi_{n-2}$.

1789: %Then we can use $(x_1,x_2)$ as a homogeneous coordinate on  $\mathbf{CP}^1_{\infty}$.

1790: %Since $\mathbf C^*$ acts as $(x_1,x_2)\mapsto(sx_1,s^{-1}x_2)$ or $(s^{-1}x_1,sx_2)$ for $s\in\mathbf C^*$ and acts trivially on $V_{n-2}$, $\mathbf C^*$-fixed locus of $\mathbf{CP}^n$ is a union of the codimension-2 linear subspace $\{x_1=x_2=0\}\simeq\mathbf{CP}^{n-2}$ and two points $(0,\cdots,0,1,0), (0,\cdots, 0,0,1)$ lying on $\mathbf{CP}^1_{\infty}$.

1791: %These mean the image $\Phi_{n-2}(L_0)$ is contained in the codimension-2 linear subspace.

1792: %In particular, $\Phi_{n-2}(L_0)$ is disjoint from $\mathbf{CP}^1_{\infty}$.

1793: Let $l_0$ be the image curve of $L_0$ and $\Phi_{n-2}(L_0)$ into the bundle $\mathbf P(\mathscr O(n-2)^{\oplus 2}\oplus \mathscr O)\to\mathbf{CP}^1$.

1794: Since $\mathbf C^*$ acts on fibers of this bundle as $(\eta_1,\eta_2)\mapsto (s\eta_1,s^{-1}\eta_2)$ or $(s^{-1}\eta_1,s\eta_2)$ by the choice of $(\eta_1,\eta_2)$ in \eqref{eta},

1795: $\mathbf C^*$-fixed locus of this bundle consists of 3 sections.

1796: But two of them are conjugate pair.

1797: Therefore we obtain $l_0=\{\eta_1=\eta_2=0\}$.

1798: Namely the `zero section' of the $\mathbf{CP}^2$-bundle is exactly the image of the fixed twistor line $L_0$.

1799: Thus we obtain a holomorphic map from $L_0$ to $l_0$, which preserves the real structure.

1800: The real structure on $L_0$ has no real point and that on $l_0$ has real point,

1801: since the real structure acts on the base space $\Lambda_{n-2}=\mathbf{CP}^1$ by complex conjugation.

1802: Therefore the map $L_0\to l_0$ cannot be isomorphic.

1803: Hence, since $\Phi_{n-2}$ is generically 2 to 1, the map $L_0\to l_0$ is 2 to 1.

1804: The branch locus of this map is the intersection of $l_0$ and  the branch locus \eqref{branch1} of $\Phi_{n-2}$.

1805: It is exactly \eqref{branch2}.

1806: Since $L_0$ is of course rational, this implies the claim of the proposition.

1807: \proofend

1808:  

1809:  \bigskip

1810: Of course, Prop.\,\ref{prop-dr} means that the equation \eqref{branch2} has multiple roots.

1811: Next by using Theorem \ref{thm-dc2} and this proposition we compute the dimension of the moduli space of our twistor spaces.

1812: For this we first count the number of parameters contained in the equation \eqref{branch1} of the branch divisor $B$.

1813:  Since  $\deg\hat g=n-1$ in general,  $\hat g$ contains $n$ real parameters.

1814:  (Note that we lost freedom of multiplying non-zero constants when we set $|c|=1/2$ in \eqref{disc5}.)

1815: Moreover, there are $n-1$ real parameters $\lambda_3,\lambda_4,\cdots,\lambda_{n+1}$ contained.

1816: %Summing these, the equation \eqref{branch1} has $2n-1$ real parameters.

1817: On the other hand, for the non-homogeneous coordinate $\lambda$ on $\mathbf{CP}^1$,  the coordinate change $\lambda\mapsto c\lambda$, $c\in\mathbf R^*$ is allowed.

1818:    This drops the dimension by one.

1819:  Further, the constraint obtained in Prop.\,\ref{prop-dr} drops the number of parameters by $n-2$.

1820:  To see this, we note that since the degree of the left-hand side of \eqref{branch2} is $2(n-1)$ in general, the virtual genus of the double covering of $\mathbf{CP}^1$ branched at the roots of \eqref{branch2} is $n-2$.

1821: Hence the rationality of the double cover drops the dimension by $n-2$.

1822: In conclusion, the dimension of the moduli space of our twistor spaces becomes

1823: \begin{equation}\label{moduli1}

1824: \{n+(n-1)\}-\{1+(n-2)\}=n.

1825: \end{equation}

1826: 

1827:  \bigskip

1828:  Finally we make a remark on the existence of our twistor spaces.

1829:  Let $Z_{\rm LB}$ be a LeBrun twistor space on $n\mathbf{CP}^2$, which admits not only $\mathbf C^*$-symmetries but also $(\mathbf C^*)^2$-symmetries.

1830:  Namely the associated LeBrun metric is supposed to admit not only $U(1)$-action but also $U(1)^2$-action.

1831: In the paper \cite{Hon07-1} we studied $U(1)$-equivariant small deformations of the twistor space $Z_{\rm LB}$.

1832: In particular, we have determined which $U(1)$-subgroup of $U(1)^2$ admits equivariant deformation whose resulting twistor spaces are not LeBrun twistor spaces.

1833: The result (Prop.\,2.1 of \cite{Hon07-1}) says that there are precisely $(n-1)$ subgroups $K_i\subset U(1)^2$, $1\le i\le n-1$, which satisfy this property.

1834: (These subgroups are specified in terms of which irreducible component is fixed by the $K_i$-action, among the anticanonical cycle in a smooth toric surface contained in $Z_{\rm LB}$.)

1835: Among these $(n-1)$ subgroups, $K_1$-equivariant deformations of $Z_{\rm LB}$ can yield the present twistor spaces we have studied in this paper.

1836: To see this, since all results in this paper rely on the structure of the surface $S$ contained in the system $|F|$, it suffices to verify that a smooth toric surface $S_{\rm LB}$ contained in $|F|$ of $Z_{\rm LB}$ can be $K_1$-equivariantly deformed into our surface $S$ constructed in \S \ref{ss-S}, and that the divisor $S_{\rm LB}$ survives under $K_1$-equivariant deformations of $Z_{\rm LB}$.

1837: These properties can be proved by the same argument we have given in \cite[\S 5.1]{Hon07-3} in proving the existence of the twistor spaces studied in the paper.

1838: Here we only remark that the dimensions of the moduli space actually coincide:

1839: in Prop.\,2.1 of \cite{Hon07-1} we have shown that the moduli space of  non-LeBrun self-dual metrics on $n\mathbf{CP}^2$ obtained by $K_1$-equivariant deformation (of LeBrun metric with torus action) is $n$-dimensional, by determining $U(1)^2$-action on the cohomology group $H^1(\Theta_{Z_{\rm LB}})$ governing small deformations of $Z_{\rm LB}$.

1840: This is (of course) equal to the dimension we have obtained in \eqref{moduli1}.

1841: (See also \cite[Example 2.4]{Hon07-1}.)

1842:  

1843:  

1844:  

1845: \begin{thebibliography}{99}

1846: 

1847: 

1848: \bibitem{AHS}M. Atiyah, N. Hitchin, I. Singer,

1849:  {\em  Self-duality in four-dimensional

1850: Riemannian geometry},

1851: Proc. Roy. Soc. London, Ser. A {\bf 362} (1978)

1852: 425--461.

1853: 

1854: %\bibitem{C91}F. Campana,

1855: %{\em  The class $\mathscr C$ is not stable by small deformations},

1856: %Math. Ann. {\bf 229} (1991) 19--30.

1857: 

1858: \bibitem{C91}F. Campana

1859: {\em On twistor spaces of the class $\mathscr C$},

1860: J. Diff. Geom. {\bf 33} (1991) 541--549.

1861: 

1862: \bibitem{CK98}F. Campana, B. Kreu\ss ler,

1863: {\em  A conic bundle description of Moishezon twistor spaces 

1864: without effective divisor of degree one},

1865: Math. Z. {\bf 229} (1998) 137--162.

1866: 

1867: %\bibitem{GH78}G.W. Gibbons, S.W. Hawking,

1868: %{\em

1869: %Gravitational multi-instantons},

1870: % Phys. Lett. {\bf 78B} (1978) 430--432.

1871: 

1872: %\bibitem{H79}N. Hitchin, 

1873: % {\em  Polygons and gravitons},

1874: %Math. Proc. Cambridge Philos. Soc.  {\bf 83} (1979)

1875: %465--476.

1876: 

1877: %\bibitem{Hi80}N. Hitchin, 

1878: % {\em  Linear field equations on self-dual spaces},

1879: % Proc. Roy. Soc. London Ser. A {\bf 370} (1980) 173-191.

1880: 

1881: \bibitem{Hi81}N. Hitchin, 

1882:  {\em  K\"aherian twistor spaces},

1883: Proc. London Math. Soc. (3) {\bf 43} (1981) 133-150.

1884: 

1885: 

1886: \bibitem{H82-1}N. Hitchin, 

1887:  {\em  Complex manifolds and Einstein's equations},

1888: Lecture Notes in Math.  {\bf 970} (1982)

1889: 73-99.

1890: 

1891: \bibitem{H82-2}N. Hitchin, 

1892:  {\em Monopoles and geodesics},

1893: Comm. Math. Phys.  {\bf 83} (1982)

1894: 579--602.

1895: 

1896: %\bibitem{Hon99}N. Honda,

1897: %{\em Donaldson-Friedman construction and deformation of a triple of compact complex spaces}, Osaka J. Math. {\bf 36} (1999) 641--672.

1898: 

1899: 

1900: %\bibitem{Hon01}N. Honda,

1901: %{\em Equivariant deformations of meromorphic actions on compact complex

1902: %manifolds}, Math. Ann. {\bf 319} (2001) 469--481.

1903: 

1904: %\bibitem{Hon03}N. Honda,

1905: %{\em Donaldson-Friedman construction and deformation of a triple of compact complex spaces, II}, Math. Nachr. {\bf 256} (2003) 48--57.

1906: 

1907: 

1908: \bibitem{Hon07-1}N. Honda,

1909:  {\em  Equivariant deformations of LeBrun's self-dual metric with torus action},

1910: Proc. Amer. Math. Soc. {\bf 135} (2007) 495-505.

1911: 

1912: \bibitem{Hon07-2}N. Honda,

1913:  {\em  Self-dual metrics and twenty-eight bitangents},

1914: J. Diff. Geom. {\bf 75} (2007) 175--258.

1915: 

1916: \bibitem{Hon07-3}N. Honda,

1917:  {\em  Explicit construction of new Moishezon twistor spaces},

1918: preprint.

1919: math.DG/0701278.

1920: 

1921: \bibitem{Hon-p-1}N. Honda,

1922:  {\em  New examples of minitwistor spaces and their moduli space},

1923: preprint. mathDG/0508088.

1924: 

1925: %\bibitem{Hor76}E. Horikawa,

1926: % {\em Deformations of holomorphic maps III}, Math. Ann. {\bf 222} (1976) 275--282.

1927: 

1928: 

1929: %\bibitem{J95}D. Joyce,

1930: % {\em  Explicit construction of self-dual 4-manifolds},

1931: %Duke Math. J. {\bf 77} (1995) 519-552.

1932: 

1933:  

1934: %\bibitem{Kod}K. Kodaira,

1935: % {\em On stability of compact submanifolds of complex manifolds},

1936: %Amer. J. Math. {\bf 85} (1963) 79--94.

1937: 

1938: %\bibitem{Ku92}

1939: %H. Kurke,

1940: %{\em Classification of twistor spaces with a pencil of surfaces of degree 1, I },

1941: %Math. Nachr. {\bf 158} (1992) 67--85.

1942: 

1943: %\bibitem{Kr97}

1944: %B. Kreu\ss ler, 

1945: %{\em Moishezon twistor spaces without effective divisors of degree one},

1946: %J. Alg. Geom {\bf 6} (1997) 379--390.

1947: 

1948: %\bibitem{Kr98}

1949: %B. Kreu\ss ler, st

1950: %{\em On the algebraic dimension of twistor spaces over the connected sum of four complex projective planes}, Geom. Dedicata {\bf 71} (1998) 263-285.

1951: 

1952: \bibitem{Kr89} B. Kreu\ss ler,  {\em Small resolutions of double solids,

1953: branched over a 13-nodal quartic surfaces}, Ann. Global Anal. Geom. {\bf

1954: 7} (1989), 227--267.

1955: 

1956: 

1957: 

1958: \bibitem{KK92} B. Kreu\ss ler, H. Kurke, {\em Twistor spaces over the

1959: connected

1960:  sum of 3 projective planes}, Compositio Math. {\bf 82} (1992), 25--55.

1961: 

1962: \bibitem{LB91}C. LeBrun, {\em Explicit self-dual metrics on

1963: ${\mathbf{CP}}^2\#\cdots\#{\mathbf{CP}}^2$}, J. Diff. Geom. {\bf 34}

1964: (1991) 223--253.

1965: 

1966: %\bibitem{LB92}C. LeBrun, {\em Twistors, K\"ahler Manifolds, and Bimeromorphic Geometry I}, J. Amer. Math. Soc. {\bf 5} (1992) 289--316.

1967: 

1968: %\bibitem{LB93}C. LeBrun, {\em Self-Dual Manifolds and Hyperbolic Geometry}, in Einstein Manifolds and Yang-Mills Connections, T. Mabuchi and S. Mukai, eds.  99--131, Marcel Dekker, 1993.

1969: %\bibitem{LB92}C. LeBrun,  {\em Twistors, K\"ahler manifolds, and bimeromorphic geometry I}, J. Amer. Math. Soc. {\bf 5} (1992), 289--316.

1970: 

1971: \bibitem{PP94}H. Pedersen, Y. S. Poon, {\em Self-duality and differentiable structures on the connected sum of complex projective planes}, Proc. Amer. Math. Soc. {121} (1994)859-864. 

1972: 

1973: %\bibitem{PP95}H. Pedersen, Y. S. Poon, {\em Equivariant connected sums of compact self-dual manifolds}, Math. Ann. {\bf 301} (1995) 717--749.

1974: 

1975: \bibitem{P86}Y. S. Poon, {\em Compact self-dual manifolds of positive

1976: scalar curvature}, J. Diff. Geom. {\bf 24} (1986) 97--132.

1977: 

1978: \bibitem{P92}Y. S. Poon, {\em On the algebraic structure of twistor spaces}, J. Diff. Geom. {\bf 36} (1992), 451--491.

1979: 

1980: \bibitem{S82} A. J. Sommese, {\em Some examples of $\mathbf C^*$-actions},

1981: %Group actions and vector fields (Vancouver, B.C., 1981), 

1982: Lecture Notes in Math. {\bf 956}, 118--124,  (1982).

1983: 

1984: %\bibitem{U}K. Ueno, {\em Classification theory of algebraic varieties and compact complex spaces}, Lect. Note Math. {\bf 439} (1975)

1985: 

1986: \bibitem{T92}

1987: C. Taubes, {\em The existence of anti-self-dual conformal structures}, J. Diff. Geom. {\bf 36 }(1992) 163--253. 

1988: 

1989: \end{thebibliography}

1990: 

1991: \small

1992: \vspace{13mm}

1993: \hspace{7.5cm}

1994: $\begin{array}{l}

1995: \mbox{Department of Mathematics}\\

1996: \mbox{Graduate School of Science and Engineering}\\

1997: \mbox{Tokyo Institute of Technology}\\

1998: \mbox{2-12-1, O-okayama, Meguro, 152-8551, JAPAN}\\

1999: \mbox{{\tt {honda@math.titech.ac.jp}}}

2000: \end{array}$

2001: 

2002: 

2003: 

2004: 

2005: 

2006: 

2007: 

2008: 

2009: 

2010: 

2011: 

2012: 

2013: 

2014: 

2015: 

2016: 

2017: 

2018: 

2019: 

2020: \end{document}

2021: 

2022: 

2023: 

2024: 

2025:  

2026: