1: \section{A proof of non-birationality}
2: \label{sec:two}
3:
4: In this section we prove, assuming given a space $X$ with the
5: properties required by Theorem~\ref{thm:main}, that there exists a
6: pair of Calabi-Yau threefolds $X'$ and $X''$ which are derived
7: equivalent but not birational.
8:
9: \paragraph
10: The idea of the construction is to consider the various powers
11: $X^{(k)}$ of $X$, for $k$ coprime to $n$, the fiber degree of $X/S$.
12: They are all Calabi-Yau threefolds, admitting a genus one fibration
13: structure, and they are all Fourier-Mukai equivalent to each other.
14: In the general case, we'll argue that there are at least three
15: non-isomorphic such powers of $X$. On the other hand, we will see
16: that in the birational equivalence class of $X=X^{(1)}$ there is at
17: most one other space that admits a genus one fibration. Therefore we
18: conclude that at least two of the spaces $X^{(k)}$ are not birational
19: to each other.
20:
21: There is one special case to consider, namely when all the $X^{(k)}$
22: spaces, admit not just one, but two genus one fibrations. In this
23: case we'll argue that each one of the $X^{(k)}$ spaces is unique in
24: its birational equivalence class, but at least two of these spaces are
25: non-isomorphic. Thus, again, we are able to find non-birational spaces
26: which are Fourier-Mukai equivalent.
27:
28: \begin{Proposition}
29: \label{prop:isom}
30: For $k$ and $k'$ coprime to $n$, $X^{(k)}$ is isomorphic to $X^{(k')}$
31: as fibrations over $S$ (with induced structure from $X/S$) if and only
32: if $k'=\pm k \bmod n$.
33: \end{Proposition}
34:
35: \begin{Proof}
36: Assume that $X^{(k)} iso X^{(k')}$ as fibrations over $S$. The idea
37: of the proof is to argue that there exists an automorphism of $J/S$,
38: the relative Jacobian of $X/S$, that maps the Brauer class
39: representing $X^{(k)}$ to the Brauer class of $X^{(k')}$. We then
40: argue that the only such automorphism possible is negation along the
41: fibers of $J/S$, which implies that $k' = \pm k \bmod n$.
42:
43: Let $J/S = X^{(0)}/S$ be the relative Jacobian of $X/S$. We will only
44: be interested in properties of the generic fiber of $J/S$, so we will
45: not be concerned with compactifying $J$. Let $\eta$ be the generic
46: point of $S$, and consider $J_\eta$, the generic fiber of $J/S$. It
47: is a smooth elliptic curve over the non-algebraically closed field
48: $\eta$.
49:
50: As argued in~\cite{CalEll}, there is a distinguished cohomology class
51: $\alpha$ in $\Br(J_\eta/\eta)$, the Brauer group of $J_\eta$,
52: corresponding to the original fibration $X/S$. The class $\alpha$ can
53: be constructed either by means of Ogg-Shafarevich theory, or as an
54: obstruction class to the existence of a universal sheaf on
55: $X\times_\eta J$. See~\cite{Cal} or~\cite{CalEll} for details.
56:
57: By~\cite[6.5]{CalEll}, we can find an identification of the relative
58: Jacobian of $X^{(k)}/S$ with $J$, such that the Brauer class
59: corresponding to $X^{(k)}$ is $\alpha^k$. This identification of the
60: relative Jacobian of $X^{(k)}/S$ with $J/S$ depends, however, on
61: knowledge of the fact that $X^{(k)}$ was constructed as the moduli
62: space of stable sheaves of rank 1, degree $k$ on the fibers of $X/S$.
63: If $X^{(k)}\iso X^{(k')}$, regarding $X^{(k)}$ as the moduli space of
64: line bundles of degree $k'$ on the fibers of $X/S$ may give rise to a
65: different identification of $J_{X^{(k)}}/S$ with $J/S$. Thus, all we
66: can conclude is that $\alpha^{k'} = \phi(\alpha^{k})$ for some
67: automorphism $\phi$ of $J_\eta/\eta$.
68:
69: The curve $J_\eta$ is an elliptic curve over the non-algebraically
70: closed field $\eta$. An automorphism $\phi$ of $J_\eta/\eta$ could
71: either be fixing the origin of $J_\eta$, or act by translation by a
72: non-zero section, or a combination of both. Consider first the case
73: when the origin is fixed by $\phi$. We can regard $\phi$ as an
74: automorphism of the fibration $J_U/U$ over some Zariski open set $U$
75: in $S$. Possibly restricting $U$ further, we can assume that all the
76: fibers of $J_U/U$ are smooth, and thus we get automorphisms of these
77: smooth elliptic curves over an algebraically closed field, fixing the
78: origin. Since the fibration is not isotrivial,
79: by~\cite[IV.4.7]{HarAG} such an automorphism has got to be negation
80: along the fibers of $J_U/U$. The corresponding action on
81: $\Br(J_\eta/\eta)$ maps $\alpha^k$ to $\alpha^{-k}$. Therefore we
82: conclude that, with regards to automorphisms fixing the origin, the
83: only possibility is for $k' = \pm k \bmod n$.
84:
85: We must rule out now the possibility of the existence of a translation
86: automorphism. Such an automorphism must act as translation by a
87: non-zero section of $J_\eta/\eta$. Such a section must be torsion:
88: otherwise its closure over all of $J/S$ would give a divisor which
89: would be linearly independent from either the zero section of $J$, or
90: from any divisor pulled back from $S$. Thus we'd have $\rk \Cl J/S >
91: 2$, contradicting the well-known fact (see, for
92: example,~\cite{DolGro}) that $\rk \Cl J/S = \rk \Cl X/S$, which is 2.
93:
94: Assume that $J/S$ has a torsion section. Consider a general smooth
95: curve $C$ in $S$, and consider the restriction $J_C\ra C$ of $J$ to
96: $C$. It is a smooth elliptic fibration of dimension 2, and the
97: non-trivial torsion section of $J$ restricts to a non-zero torsion
98: section of $J_C/C$. Its fibers are not multiple or reducible, since
99: the original fibration only had isolated such fibers. It is also not
100: isotrivial. But by~\cite[Proposition 5.3.4 (ii)]{CosDol} it does not
101: have any torsion sections, which is a contradiction. Thus
102: $J/S$ does not have any torsion sections.
103:
104: The reverse implication is easy and is left to the reader.
105: \end{Proof}
106:
107: \begin{Proposition}
108: \label{prop:notbir}
109: If one of the spaces $X^{(k)}$ (which we'll call $X$ without loss of
110: generality) admits two structures of genus one fibration, then $X$ is
111: the unique smooth projective Calabi-Yau variety in its birational
112: equivalence class. Otherwise, there may exist at most one more smooth
113: projective Calabi-Yau variety birational to $X$ admitting a genus one
114: fibration structure.
115: \end{Proposition}
116:
117: \begin{Proof}
118: This proof is based on ideas from the Mori program. Observe that
119: smooth, projective Calabi-Yau varieties are minimal in the sense of
120: this program. Therefore, all other minimal varieties birational to
121: $X$ differ from $X$ by a finite number of flops.
122:
123: Since the rank of the Picard group of $X$ is 2, the nef cone of $X$ is
124: a wedge in $\R^2$ (see Figure~\ref{Fig1}). Faces of this nef cone
125: correspond to extremal contractions of $X$, when they have rational
126: slope. Each face could thus yield one of three cases:
127:
128: \begin{itemize}
129: \item[--] an irrational slope face, which does not yield any geometric
130: transition in the Mori program;
131: \item[--] a $K$-trivial fibration (a fibration with fibers which are
132: either K3 or abelian surfaces, or a genus one fibration);
133: \item[--] a flopping contraction.
134: \end{itemize}
135:
136: Note that there are precisely two faces of the nef cone, because of
137: the topology of $\R^2$. If $X$ admits two structures of genus one
138: fibration, it means that both faces of the Mori cone correspond to
139: $K$-trivial fibrations, and therefore there are no flopping
140: contractions that can be performed on $X$. Thus $X$ is the unique
141: minimal projective variety in its birational class.
142:
143: If $X$ admits a unique structure of genus one fibration, this labels
144: one face of the nef cone. The other face of the nef cone could have
145: irrational slope, correspond to a K3 or abelian surface fibration on
146: $X$, or to a flop. In the first and second cases, we are done by an
147: argument similar to the previous case ($X$ is unique in its birational
148: class). Otherwise, let $X'$ be the result of flopping $X$ along the
149: corresponding side of the nef cone. The nef cone of $X'$ is a new
150: wedge in $\R^2$, adjacent to the nef cone of $X$.
151:
152: \begin{figure}
153: \begin{center}
154: \resizebox{\textwidth}{!}{\includegraphics*{fig1.eps}}
155: \caption{The nef cones of birational models of $X$}
156: \label{Fig1}
157: \end{center}
158: \end{figure}
159:
160: Now consider the new face of the nef cone of $X'$. It could again
161: have irrational slope, or yield a $K$-trivial fibration. If it does,
162: our analysis of minimal models of $X$ is finished ($X$ and $X'$ are
163: the only minimal varieties birational to $X$, because the only
164: flopping contraction of $X$ yields $X'$, and vice versa). The
165: $K$-trivial fiber space structure on $X'$ could be that of a genus one
166: fibration, or a K3 or abelian surface fibration, but in any case this
167: yields at most one more variety birational to $X$ with a genus one
168: fibration structure, and the process has stopped.
169:
170: If, on the other hand, the new face of the nef cone of $X'$ yields a
171: flopping contraction, then $X'$ does not have any genus one fibration
172: structure. Indeed, any such structure must arise from a $K$-trivial
173: fibration-type face of the nef cone, and both faces of
174: $\mathsf{nef}(X)$ yield flops, not fiber spaces. By flopping the new
175: ray gives rise to a new space minimal birational model $X''$ of $X$,
176: but $X'$ itself does not need to be considered for examples of genus
177: one fibrations birational to $X$.
178:
179: This process can now be continued. At each step we consider the new
180: face of the nef cone of the last space constructed. If this face
181: either has irrational slope, or yields a $K$-trivial fibration, we
182: might have arrived at one more genus one fibration on a space
183: birational to $X$, but the generating process of birational models of
184: $X$ has stopped. Or, if this face yields a flop, we generate a new
185: minimal model of $X$, but the current space does not have any genus
186: one fibration structure. Only the ends of the chain of nef cones can
187: yield minimal models of $X$ with genus one structure. One of them is
188: $X$ itself, and there can be at most one other such end.
189: \end{Proof}
190:
191: \begin{Theorem}
192: If a space $X$ satisfying the properties required by
193: Theorem~\ref{thm:main} exists, then there exist smooth, projective
194: Calabi-Yau threefolds $X'$ and $X''$ that are Fourier-Mukai
195: equivalent, but not birational.
196: \end{Theorem}
197:
198: \begin{Proof}
199: Consider the collection $\sC = \{X^{(k)}\}$ for $k$ coprime to $n$,
200: $1\leq k \leq (n-1)/2$.
201:
202: If all the spaces in $\sC$ admit a unique structure of genus one
203: fibration (which must be the one arising from regarding them as
204: relative moduli spaces on $X/S$), then Proposition~\ref{prop:isom}
205: shows that they must all be distinct. By the assumptions made on $n$,
206: there are at least 3 of them. On the other hand, by
207: Proposition~\ref{prop:notbir}, they can not all be birational to $X$,
208: thus the conclusion of the theorem.
209:
210: If one of the $X^{(k)}$'s in the collection admits two genus one
211: fibration structures, then it is unique in its birational equivalence
212: class. If any other space $X^{(k')}$ in $\sC$ is non-isomorphic
213: to it, then $X^{(k)}$ and $X^{(k')}$ is a pair of Fourier-Mukai
214: equivalent, non-birational Calabi-Yau threefolds.
215:
216: The only case left to consider is if all the spaces in $\sC$ are
217: isomorphic to each other, each admitting precisely two structures of
218: genus one fibration. Note that on each space in $\sC$, there is a
219: distinguished such structure, arising from their construction as
220: moduli spaces on $X/S$. Since there are at least 3 different
221: allowable values of $k$, we must be able to find $k$ and $k'$ such
222: that $X^{(k)}/S$ and $X^{(k')}/S$ are isomorphic as fibrations, with
223: the induced structure from $X/S$. This contradicts
224: Proposition~\ref{prop:isom}, and therefore this case is not possible.
225: \end{Proof}
226:
227: