1:
2: \subsection{Main result}
3:
4: Let $X$ be a smooth complex projective surface,
5: and let $D$ be a normal crossing divisor of $X$.
6: Let $(\nbige,\nabla)$ be
7: a flat meromorphic connection
8: on $(X,D)$,
9: i.e.,
10: $\nbige$ denotes a locally free $\nbigo_X(\ast D)$-module,
11: and $\nabla:\nbige\lrarr\nbige\otimes\Omega_{X/\cnum}^1$
12: denotes a flat connection.
13: We discuss a conjecture of C. Sabbah
14: under the algebraicity assumption.
15:
16: \begin{thm}
17: \label{thm;07.4.15.10}
18: There exists a regular birational morphism
19: $\pi:\Xtilde\lrarr X$ such that
20: $\pi^{-1}(\nbige,\nabla)$ has the good formal structure.
21: \end{thm}
22:
23: See Subsection \ref{subsection;07.4.30.1}
24: for good formal structure.
25: For explanation of the meaning of the theorem,
26: let us recall the very classical result in the curve case.
27: (See the introduction of \cite{majima} for more detail,
28: for example.)
29: Let $C$ be a smooth projective curve,
30: and let $Z\subset C$ be a finite subset.
31: Let $(\nbige,\nabla)$ be a meromorphic connection on $(C,Z)$,
32: i.e., $\nbige$ is a locally free $\nbigo(\ast Z)$-module
33: with a connection $\nabla$.
34: Let $P$ be any point of $Z$,
35: and let $(U,t)$ be a holomorphic coordinate neighbourhood
36: around $P$ such that $t(P)=0$.
37: The local structure of $\nbige$ around $P$
38: can be understood by
39: the formal structure at $P$
40: and the Stokes structure around $P$.
41: Namely,
42: take a ramified covering
43: $\varphi_P:(\Utilde,t_d)\lrarr (U,t)$ given by $t=t_d^d$,
44: where $d$ is a large integer
45: divided by $(\rank(\nbige)!)^3$, for example.
46: Let $\Ptilde\in\Utilde$ be the inverse image of $P$.
47: The formal completion of $\varphi_P^{\ast}(\nbige,\nabla)$
48: at $\Ptilde$ is decomposed into the direct sum
49: $\bigoplus_{\gminia\in \cnum(\!(t_d)\!)/\cnum[\![t_d]\!]}
50: (\nbige_{\gminia},\nabla_{\gminia})$,
51: where $\nabla_{\gminia}-d\gminia$ are regular.
52: (In the curve case,
53: we do not have to assume that the base space is
54: algebraic.
55: In fact, the decomposition can be obtained
56: for any connections on formal curves.)
57: Then, the formal decomposition
58: can be lifted to the decomposition
59: on any small sectors by the asymptotic analysis,
60: which leads us the Stokes structure.
61:
62:
63: It is a challenging and foundational problem to obtain the generalization
64: in the higher dimensional case.
65: The systematic study
66: was initiated by H. Majima,
67: who developed the asymptotic analysis in the higher dimensional case.
68: (See \cite{majima}, for example.)
69: Briefly speaking,
70: his result gives the lifting of a formal decomposition
71: to the decomposition on small sectors.
72: Inspired by Majima's work,
73: Sabbah (\cite{sabbah4}) developed the asymptotic analysis
74: in the other framework.
75: He observed the significance of the understanding
76: on the formal structure of the irregular connection.
77: He proposed the conjecture which says
78: that Theorem \ref{thm;07.4.15.10}
79: may hold without the algebraicity assumption,
80: and he established it in the case $\rank(\nbige)\leq 5$.
81: He also reduced the problem to the study
82: of the turning points contained in the smooth part
83: of the divisor $D$,
84: without any assumption on the rank.
85:
86: Sabbah gave some interesting applications of the conjecture,
87: one of which is a conjecture of B. Malgrange
88: on the absence of the confluence phenomena for
89: flat meromorphic connections.
90: Recently, Y. Andr\'e (\cite{andre}) proved Malgrange's conjecture
91: motivated by Sabbah's conjecture.
92:
93: In this paper,
94: we will give a proof of the algebraic version of Sabbah's conjecture.
95: In \cite{mochi6},
96: the author intends to establish the correspondence
97: of semisimple algebraic holonomic $D$-modules
98: and polarizable wild pure twistor $D$-modules
99: through wild harmonic bundles,
100: on smooth projective surfaces and
101: the higher dimensional varieties,
102: which is related with a conjecture of M. Kashiwara \cite{k5}.
103: Theorem \ref{thm;07.4.15.10} has the foundational importance
104: for the study.
105:
106: \subsection{Main ideas}
107:
108: Let $k$ be an algebraically closed field,
109: and let $(\nbige,\nabla)$ be a meromorphic flat connection on
110: $k[\![s]\!](\!(t)\!)$.
111: If the characteristic number $p$ of $k$ is positive,
112: we always assume that $p$ is much larger than
113: $\rank\nbige$ and the Poincar\'e rank of $\nbige$
114: with respect to $t$.
115: Let $\nabla_t$ denote
116: the induced relative connection
117: $\nbige\lrarr\nbige\otimes
118: \Omega^1_{k[\![s]\!](\!(t)\!)/k[\![s]\!]}$.
119: The induced connection
120: $(\nbige,\nabla_t)\otimes k(\!(s)\!)(\!(t)\!)$
121: is denoted by $(\nbige_1,\nabla_1)$.
122: The specialization of $(\nbige,\nabla_t)$
123: at $s=0$ is denoted by $(\nbige_0,\nabla_0)$.
124: We have the set of the irregular values
125: $\Irr(\nbige_0,\nabla_0)
126: \subset k(\!(t_d)\!)/k[\![t_d]\!]$
127: and
128: $\Irr(\nbige_1,\nabla_1)
129: \subset
130: k(\!(s_d)\!)(\!(t_d)\!)/k(\!(s_d)\!)[\![t_d]\!]$,
131: where $t_d$ and $s_d$ denote $d$-th roots of $t$ and $s$,
132: respectively.
133: Briefly and imprecisely speaking,
134: one of the main issues is how to compare
135: $\Irr(\nbige_i,\nabla_i)$ $(i=0,1)$.
136: Ideally, we hope that
137: $\Irr(\nbige_1,\nabla_1)$
138: is contained in $k[\![s]\!](\!(t_d)\!)/k[\![s,t_d]\!]$,
139: and that the specialization at $s=0$ gives
140: $\Irr(\nbige_0,\nabla_0)$.
141: However, they are not true, in general.
142:
143: In the case $p>0$,
144: we have the $p$-curvature $\psi$ (resp. $\psi_i$) of
145: the connection $\nabla$
146: (resp. $\nabla_i$).
147: \[
148: \psi\in \End(\nbige)\otimes F^{\ast}\Omega^1_{k[\![s]\!](\!(t)\!)/k},
149: \quad
150: \psi_1\in\End(\nbige_1)\otimes
151: F^{\ast}\Omega^1_{k(\!(s)\!)(\!(t)\!)/k(\!(s)\!)},
152: \quad
153: \psi_0\in\End(\nbige_0)\otimes
154: F^{\ast}\Omega^1_{k(\!(t)\!)/k}
155: \]
156: Here, $F$ denotes the absolute Frobenius map.
157: In the following,
158: we use the notation $\psi(t\del_t)$
159: to denote $\psi(F^{\ast}t\del_t)$,
160: for simplicity.
161: Let $\Sp(\psi(t\del_t))$ denote the set of the eigenvalues of
162: $\psi(t\del_t)$,
163: which is contained in $\nbiga_t$,
164: where
165: $\nbiga$ denotes a finite extension of $k[\![s,t]\!]$
166: and $\nbiga_t$ denotes a localization of $\nbiga$
167: with respect to $t$.
168: Similarly,
169: let $\Sp(\psi_i(t\del_t))$ denote the set of the eigenvalues
170: of $\psi_i(t\del_t)$ for $i=0,1$, and then
171: $\Sp(\psi_0(t\del_t))\subset
172: k(\!(t_d)\!)$
173: and $\Sp(\psi_1(t\del_t))\subset k(\!(s_d)\!)(\!(t_d)\!)$
174: for some appropriate $d\in\seisuu_{>0}$.
175: We may have the natural inclusion
176: $\kappa_1:\nbiga_t\lrarr k(\!(s_d)\!)(\!(t_d)\!)$
177: and the specialization
178: $\kappa_0:\nbiga_t\lrarr k(\!(t_d)\!)$ at $s=0$.
179: Clearly, $\psi_i(t\del_t)$ $(i=0,1)$ are naturally obtained from
180: $\psi(t\del_t)$ by $\kappa_i$,
181: and hence
182: $\Sp(\psi_i(t\del_t))$ are obtained from
183: $\Sp(\psi(t\del_t))$ by $\kappa_i$.
184: Recall that the irregular value of $\nabla_i$ can be related
185: with the negative part of the eigenvalues
186: of $\psi_i(t\del_t)$ (Lemma \ref{lem;07.4.14.3}),
187: where the negative part of $f=\sum f_j\cdot t_d^j\in R(\!(t_d)\!)$
188: is defined to be $f_-:=\sum_{j<0}f_j\cdot t_d^j$.
189: Hence, we have the following diagram:
190: \[
191: \begin{CD}
192: \Sp\bigl(\psi_1(t\del_t)\bigr)
193: @<{\kappa_1}<<
194: \Sp\bigl(\psi(t\del_t)\bigr)
195: @>{\kappa_0}>>
196: \Sp\bigl(\psi_0(t\del_t)\bigr)\\
197: @VVV @. @VVV\\
198: \Irr(\nbige_1,\nabla_1)
199: @. @.
200: \Irr(\nbige_0,\nabla_0)
201: \end{CD}
202: \]
203: But, we should remark that
204: $\kappa_0(\alpha)_-$ and $\kappa_1(\alpha)_-$
205: cannot be directly related, in general.
206:
207: Let us consider the simplest case where
208: the ramification of $\nbiga$ over $k[\![s,t]\!]$
209: may occur only at the divisor $\{t=0\}$.
210: Then, $\Sp(\psi(t\del_t))$ is contained in
211: $k[\![s]\!](\!(t_d)\!)$,
212: and $\kappa_0(\alpha)_-$ is the specialization of
213: $\kappa_1(\alpha)_-$ at $s=0$
214: for any $\alpha\in\Sp(\psi(t\del_t))$.
215: Thus, we can compare the irregular values
216: of $\nabla_i$ $(i=0,1)$ in this simplest case.
217:
218: Then, we have to consider what happens
219: if the ramification of $\nbiga$ may be non-trivial.
220: As the second simplest case,
221: we assume that
222: the ramification may occur only
223: at the normal crossing divisor
224: $(t)\cup(s')$ of $\Spec^fk[\![s,t]\!]$,
225: where $s'=s+t\cdot h(t)$.
226: Then,
227: $\Sp(\psi(t\del_t))$
228: are contained in $k[\![s_d']\!](\!(t_d)\!)$,
229: where $s_d'$ denotes a $d$-th root of $s'$.
230: We assume, moreover, that
231: $\Sp(\psi(t\del_t))$
232: are contained in $k[\![s]\!](\!(t_d)\!)+k[\![s_d',t_d]\!]$.
233: Then, the negative part of the eigenvalues behave well
234: with respect to the specialization,
235: i.e.,
236: $\kappa_1(\alpha)_-=\kappa_0(\alpha)_-$
237: for any $\alpha\in \Sp(\psi(t\del_t))$.
238: Hence, we can compare the irregular values
239: of $\nabla_i$ $(i=0,1)$
240: in this mildly ramified case
241: (Lemma \ref{lem;07.4.14.10}).
242:
243: We would like to apply such consideration to our problem.
244: Essentially, the problem is the following,
245: although we will discuss it in a different way.
246: Let $(\nbige,\nabla)$ be a meromorphic flat connection
247: with a lattice $E$ on $(X,D)$.
248: For simplicity, we assume
249: everything is defined over $\seisuu$.
250: Then, we have the mod $p$-reductions
251: $(\nbige_p,\nabla_p):=(\nbige,\nabla)\otimes \Fpbar$
252: over $(X_p,D_p):=(X,D)\otimes\Fpbar$
253: with the lattice $E_p=E\otimes\Fpbar$.
254: Let $\psi_p\in\End(E_p)\otimes F^{\ast}\Omega_{X_p}(ND_p)$
255: denote the $p$-curvature.
256: Then, we have the spectral manifold
257: $ \Sigma_p(\psi_p):=
258: \bigl\{(x,\omega)\,\big|\,
259: \mbox{$\omega$ eigenvalues of $\psi_{p|x}$}
260: \bigr\}
261: \subset
262: F^{\ast}\bigl(\Omega^1_{X_p}\otimes\nbigo(ND_p)\bigr)$.
263:
264: For simplicity, we assume that
265: $\psi_p$ has the distinct eigenvalues
266: at the generic point.
267: Then, we hope that
268: the ramification of the projection $\pi_p$
269: of $\Sigma_p(\psi_p)$ to $X_p$
270: may happen at normal crossing divisor,
271: after some blow ups,
272: i.e.,
273: $ R(\pi_p):=
274: \bigl\{x\in X_p\,\big|\,\mbox{\rm $\pi_p$ is not etale at $x$}\bigr\}$
275: is normal crossing.
276: If we fix $p$,
277: it is easy to obtain such birational map
278: because we are considering the surface case.
279: But, for our problem,
280: we would like to control the ramification
281: for almost all $p$ at once.
282: So we need something more.
283:
284: Here, we recall the important observation of
285: J. Bost, Y. Laszlo and C. Pauly \cite{laszlo-pauly}
286: which says that
287: we have $\Sigma_p'$ contained in
288: $\Omega^1_{X_p}\otimes\nbigo(ND)$,
289: such that
290: $\Sigma_p(\psi_p)$ is the pull back of $\Sigma_p'$.
291: So, we have only to control
292: the ramification curves $R(\pi_p')$
293: of the projection $\pi_p'$ of $\Sigma_p'$ to $X_{p}$.
294: Then, it is not difficult to see that
295: the arithmetic genus of $R(\pi_p')$ are dominated,
296: independently of $p$.
297: So, the complexity of the singularities
298: of these ramification curves are bounded,
299: and thus we can control them uniformly.
300: (See Section \ref{section;07.4.28.2}.)
301:
302: \subsection{Acknowledgement}
303:
304: The author thanks C. Sabbah for the discussion
305: and his attractive conjecture.
306: It is a pleasure to express his gratitude to
307: R. Bezrukavnikov,
308: O. Biquard,
309: M. Kontsevich, J. Li,
310: C. Simpson, A. Usnich,
311: D. Wei for some discussions.
312: He is grateful to Y. Tsuchimoto and A. Ishii
313: for their constant encouragement.
314: Special thanks goes to K. Vilonen.
315: He thanks the colleague in Kyoto University
316: for their support.
317: This paper is prepared during his visit at IHES and ICTP.
318: The author is grateful for their excellent hospitality.
319: He also thanks the partial financial supports by
320: Sasakawa Foundation
321: and Ministry of Education, Culture, Sports, Science and Technology.
322:
323: The contents of this paper is
324: a detailed proof of a theorem given
325: in a talk at the conference
326: `Algebraic Analysis and Around'
327: in honour of Professor Masaki Kashiwara's 60's birthday.
328: The author would like to express
329: his gratitude to the organizers.
330:
331: It is an extremely great pleasure
332: for the author to dedicate this paper
333: to Masaki Kashiwara with admiration
334: for his great works and
335: his leading role in the development of
336: current mathematics.
337: