The local Laplace transform of
an elementary irregular meromorphic connection
MARCOHIEN(*) - CLAUDESABBAH(**)
ABSTRACT- We give a definition of the topological local Laplace transformation for a Stokes- filtered local system on the complex affine line and we compute in a topological way the Stokes data of the Laplace transform of a differential system of elementary type.
MATHEMATICSSUBJECTCLASSIFICATION(2010). 14D07, 34M40
KEYWORDS. Laplace transformation, meromorphic connection, Stokes matrix, Stokes fil- tration.
1. Introduction
1.1 ±Riemann-Hilbert correspondence and Laplace transformation
Let M be a holonomic Cth@ti-module and let be its Laplace transform, which is a holonomic C[t0]h@t0i-module through the correspondence t0@t and t @t0, that is, given by the kernel exp ( tt0). According to the Riemann-Hilbert correspondence as stated by Deligne (see [Del07], [Mal91] and [Sab13a, Chap. 5]), the holonomic Cth@ti-moduleM corresponds to a Stokes-perverse sheaf on the projective lineP1t with affine coordinatet, which is a Stokes-filtered local system in the neighbourhood oft 1. Similarly, corresponds to a Stokes-perverse sheaf onP1t0, which is a Stokes-filtered local system in the neighbourhood oft0 1. In the following, we will denote by t the coordinate ofP1t0 centered at c1such that
(*) Indirizzo dell'A.: Lehrstuhl fuÈr Algebra und Zahlentheorie, UniversitaÈtsstraûe 14, 86159 Augsburg, Deutschland.
E-mail: marco.hien@math.uni-augsburg.de
This research was supported by the grant DFG-HI 1475/2-1 of the Deutsche For- schungsgemeinschaft.
(**) Indirizzo dell'A.: UMR 7640 du CNRS, Centre de MatheÂmatiques Laurent Schwartz, EÂcole polytechnique, F-91128 Palaiseau cedex, France.
E-mail: Claude.Sabbah@polytechnique.edu
This research was supported by the grants ANR-08-BLAN-0317-01 and ANR-13-IS01- 0001-01 of the Agence nationale de la recherche.
t1=t0 on P1t0nnf0;c1g, and we continue to set c1 ft0g. The topological Laplace transformation is the corresponding transformation at the level of the category of Stokes-perverse sheaves. We will use results of [Moc14] to make clear its definition. A topological Laplace transformation can also be defined in the set- ting of enhanced ind-sheaves considered in [DK13], which corresponds, through the Riemann-Hilbert correspondence of loc. cit., to the Laplace transformation of ho- lonomicD-modules (see also [KS14]).
1.2 ±Riemann-Hilbert correspondence and local Laplace transformation In this article, we will consider the local Laplace transformationF(0;1)from the category of finite dimensionalC ftg-vector spaces with connection to that of finite dimensionalC ftg-vector spaces with connection. It is defined as follows. A finite dimensionalC ftg-vector space with connectionMcan be extended in a unique way as a holonomicCth@ti-moduleMwith a regular singularity at infinity and no other singularity at finite distance thant0. ThenF(0;1)(M) is by definition the
germ of at c1: ft0 1g. We will regard F(0;1)(M) as a C ftg-vector
space with connection. It is well-known that has at most a regular singularity at t00 and no other singularity at finite distance. Therefore, giving F(0;1)(M) is equivalent to giving the localized moduleC[t0;t0 1]C[t0] .
The Deligne-Riemann-Hilbert correspondence associates to M a Stokes-fil- tered local system (L;L) onS1t0(the circle with coordinate argt). Similarly, we will denote by F(0;1)top (L;L) the Stokes-filtered local system on S1t0 associated withF(0;1)(M). In this setting, we will address the following questions:
To make explicit the topological local Laplace transformation functorF(0;1)top . To use this topological definition to compute, in some examples, the Stokes
structure ofF(0;1)(M) in terms of that ofM.
1.3 ±Statement of the results
The first goal will be achieved in §3. As for the second one, we will restrict to the family of examples consisting of elementary meromorphic connections, denoted by El(r; ';R) in [Sab08]. Namely, r:u7!tup is a ramification of orderpof the variable t,'is a polynomial inu 1 without constant term, andR is a finite dimensional C fug-vector space with a regular connection. Setting E '(C fug;d d'), we define
M El(r; ';R):r(E ' R):
If we extendRas a freeC[u;u 1]-moduleRof finite rank with a connection having a regular singularity at u0 and u 1 and no other singularity, and set
E '(C[u;u 1];d d'), the extensionMofMconsidered above is nothing but the freeC[t;t 1]-module with connection
MEl(r; ';R):r(E 'R):
Letqbe the pole order of'. We also encodeRas a pair (V;T) of a vector space with an automorphism (the formal monodromy). We call'theexponential factorofrM and set ExpExp(rM) f'g.
ASSUMPTION1.3.1. We will assume in this article thatp;qare coprime.
By the stationary phase formula of [Fan09, Sab08] the formal Levelt-Turrittin type of the local Laplace transform F(0;1)(M) is that of El(br; ';b R), whereb
the ramificationbr:h7!thas orderbp:pq, i.e.,br2hpq(c(r; ')hCfhg) for some nonzero constant c(r; '); writing r and ' in the h variable, it is expressed asbr(h) r0(h)='0(h);
the exponential factor'(h)b '(h) '0(h)r(h)=r0(h) has pole orderq;
the regular partRb corresponds to the pair (V;b T)b (V;( 1)qT).
Our goal is then to compute explicitly the Stokes structure of the non-ramified meromorphic connectionbrF(0;1)(M), which is usually non trivial. It is of pure level q. Stokes structures of pure levelq can be represented in various ways by generalized monodromy data. We make use of the following description (see Section 2.2 for details).
To the formal data attached to brF(0;1)(M), which consist of the family Expd :Exp(brF(0;1)(M)) f'b (h)j 2mpqg (mn is the group of n-th roots of unity and 'b '(b h(1o(h))) and the pair ( 2m
pqV; 2m
pq( 1)qT), we add:
(1.3.2) the choice of a generic argument#bo2S1h0, giving rise to a total ordering of the set Exp, hence an order preserving numberingd f0;. . .;pq 1g 'mpq, (1.3.3) a set ofunramified linear Stokes data (L`);(S`1` );(FL`)
`0;...;2q 1of pure
level q, which consists of
a family (L`)`0;...;2q 1 consisting of 2qC-vector spaces, isomorphismsS`1` :L` !L`1,
for each m2 f0;. . .;q 1g, a finite exhaustive (1) increasing filtration FL2m (resp. decreasing filtration FL2m1) indexed by f0;. . .;pq 1g, with the property that the filtrations are mutually opposite with respect
(1) We say that an increasing filtrationFLindexed by a totally ordered finite setOis exhaustiveifFmaxOLL. We then setF<minOL0. For a decreasing filtration, we have FminOLLand we setF>maxOL0.
to the isomorphismsS`1` , i.e., for any m2 f0;. . .;q 1g:
L2mpqM1
k0
FkL2m\S2m2m 1 FkL2m 1
; (1.3.3)
L2m1pqM1
k0
FkL2m1\S2m12m FkL2m
; with the convention that L 1L2q 1.
Note that the opposite filtrations induce uniquely determined splittings, that is, filtered isomorphisms
t2m:L2m !grFL2m:Lpq 1
k0 grFkL2m; t2m1:L2m1 ! grFL2m1 :Lpq 1
k0 grkFL2m1; (
(1:3:4)
where the right hand side carries its natural increasing (resp. decreasing) filtration.
The resulting isomorphismsS`1` :t`1S`1` t`1:grFL`!grFL`1are upper/
lower block triangular - depending on the parity of`- and their blocks (S`1` )j;jon the diagonal are isomorphisms. These matrices are commonly called the Stokes matrices or Stokes multipliers. We call the family
S brF(0;1)(M)
: (L`);(S`1` );(FL`)
`0;...;2q 1
1:3:5
the linear part of the Stokes data ofbrF(0;1)(M). The notion of morphism between such families is obvious.
On the other hand, we will define in § 1.4 standard linear Stokes data that we denote bySstd(V;T;p;q).
The main result regarding the explicit determination of the Stokes data of the local Laplace transform of El(r; ';R) can be stated as follows.
THEOREM 1.3.6. With the previous notation and assumptions, for a suitable choice of #bo, the linear Stokes data S brF(0;1)(M)
, for M El(r; ';(V;T)), are isomorphic to the standard linear Stokes dataSstd(V;T;p;q).
REMARK1.3.7. (1) In order to descend from this result to the linear Stokes data of F(0;1)(M), we would need to identify the mpq-action on the standard linear Stokes data. This will not be achieved in this article.
(2) The arguments in the proof also lead to an explicit computation of the topo- logical monodromy ofF(0;1)(M) up to conjugation (see Proposition 1.4.13).
(3) Similar results were obtained by T. Mochizuki in [Moc10, § 3], by using explicit bases of homology cycles, while we use here cohomological methods.
Also, the results of loc. cit. do not make explicit a standard model for the linear Stokes data, although it should be in principle possible to obtain such a model from these results.
1.4 ±The standard linear Stokes data
We will now define the setSstd(V;T;p;q) attached to a finite dimensional vector space V equipped with an automorphism T, and to a pair of coprime integers (p;q)2(N)2. We refer to Section 6 for a geometric motivation which leads to the definitions below. We start by defining two orderings onmpq.
DEFINITION1.4.1. For ; 02mpq, we set
(1:4:1) <odd 0 if Re e "i p 0p<0 for some"such that 0<"1 and <ev 0, 0<odd . We enumeratempqaccording to the even ordering, i.e., (1:4:1) mpq f k jk0;. . .;pq 1gwith 0<ev <ev pq 1:
Since (p;q)1, we also have (pq;q)1 and there exist a;b2Z with ap1b(pq). Setj:exp
2pi pq
2mpq. Then the even orderingon mpq is expressed by
1j0<evja<evj a<ev <evjka<evj ka<ev <evev max; (1:4:2)
withk2h 1;pq2
, with
ev max exp
api pq
if pq is odd andais even () or odd ( );
exp (api) 1 if pq is even:
8<
:
Theodd orderingis the reverse ordering.
Fork2Zwe set
evin(k): qk pq1
2
;
evout(k):k evin(k) pk pq
1 2
: 8>
><
>>
: (1:4:3)
Both are increasing functions of k. If k2 f0;. . .;pq 1g, we have
evin(k)2 f0;. . .;qgandevout(k)2 f0;. . .;pg. Furthermore, let minevin(k):min k05 pq
2q jevin(j)evin(k) for allj2[k0;k]
(1:4:4) and
(1:4:5) maxevout(k):max k00<pq pq
2p jevout(j)evout(k) for allj2[k;k00]
(see Remark 6.1.3(3) for an explanation). In a similar way we set fork2Z:
oddin(k): qk pq
;
oddout(k):k 1 oddin(k) pk pq 1
2 f 1;. . .;p 1g;
8>
><
>>
(1:4:6) :
and if k2 f0;. . .;pq 1g we have oddin(k)2 f0;. . .;q 1g and oddout(k)2 f 1;. . .;p 1g. We define minoddin (k) and maxoddout(k) literally as in (1.4.4) and (1.4.5) replacing ``ev'' by ``odd''. Note that
h qk pq
1 2; qk
pq
i\N[()evin(k)oddin(k)1:
Let V be a finite dimensional complex vector space together with an auto- morphism T2Aut(V). Let us denote by1k the complex vector space of rank one with chosen basis element 1k. Let us consider direct sum Vpqofpqcopies of V and let us keep track of the indices by writing
Vpq pqM1
k0
VC1k: (1:4:7)
CONVENTION1.4.8. For anyj2Z, we will write
v1j:T s(v)1js(pq) fors2Zsuch thatjs(pq)2[0;pq 1]:
DEFINITION1.4.9. Assumepq53. Then:
(1) The isomorphismsoddev :Lpq 1
k0 V1k!Lpq 1
k0 V1k is defined as (1:4:9)
soddev v1k:
v1k; if h
pqkq 1 2;pqkqi
\N[ v1minodd
in (k) 1
v1maxodd out(minoddin (k))
v1maxodd
out(k)1; if h
pqkq 1 2;pqkqi
\N6[ 8>
>>
>>
><
>>
>>
>>
: (2) The isomorphismsevodd:Lpq 1
k0 V1k!Lpq 1
k0 V1k is defined as (1:4:9)
sevodd v1k:
v1k; if h
pqqk 1 2;pqkqi
\N6[ v1minevin(k) 1
v1maxevout(minevin(k))
v1maxevout(k)1; if h
pqqk 1 2;pqkqi
\N[ 8>
>>
>>
>>
<
>>
>>
>>
>:
Ifpq1, the isomorphismsoddev is defined as
(1:4:10) soddev (v10): v10(IdT 1)v11 and soddev (v11):v11; andsevoddis defined as
(1:4:10) sevodd v10:v10 and sevodd v11: IdT 1v10 v11: We are now ready to define the following particular set of Stokes data.
DEFINITION1.4.11. The standard linear Stokes dataSstd(V;T;p;q) are given by (1) the vector spaces L`:Vpq for`0;. . .;2q 1,
(2) for eachm0;. . .;q 1, the isomorphisms
S2m2m 1:sevodd:L2m 1 !L2m (cf:(1:4:9));
S2m12m :soddev :L2m !L2m1 (cf:(1:4:9)):
(3) the isomorphismS2q2q 1:diag(T;. . .;T)sevodd:L2q 1 ! L0. (4) the filtrationsFkL2m:L
k04kV1k0andFkL2m1:L
k05kV1k0. Note that the opposedness property (1.3.3) for the above data is not ob- vious. We will not give a direct proof since it follows a posteriori from our main result by which these data are the Stokes data attached to some Stokes structure. The arguments in the proof of Theorem 1.3.6 also lead to the fol- lowing explicit computation of the topological monodromy of brF(0;1)(M), where we set
in(k):evin(k) qk pq1
2
; maxout(k):maxevout(k):
(1:4:12)
PROPOSITION 1.4.13. The topological monodromy of brF(0;1)(M) is con- jugate to the automorphism of the vector spaceVpqLpq 1
k0 VC1k (using the notation(1.4.7)and Convention1.4.8) given by
Tbtop(v1k) v1k1 if in(k1)in(k);
v(1k 1maxout (k)1maxout (k)1) if in(k1)in(k)1:
(
REMARK1.4.14. The topological monodromy can of course be deduced from the Stokes data by the formula
Ttop:S02q 1S2q2q 12 S102Aut(L0):
However, we will give a more direct computation in § 6.3. This approach is less involved than the detour over the Stokes data and could perhaps be applied in more general situations without having to understand the full information of the Stokes data.
The article is organized as follows. Section 2 recalls the notion of Stokes-filtered local system and explains the correspondence with that of linear Stokes data - generalized monodromy data - describing Stokes structures of pure levelq. In Section 3 the construction of the topological Laplace transformation is discussed in general, and the main tool (Th. 3.3.1) is [Moc14, Cor. 4.7.5]. We then concentrate on the case of an elementary meromorphic connection. A description of the Stokes data forbrF(0;1)(M) in cohomological terms is proved in the subsequent section - Theorem 4.3.3.
The remaining sections provide the proof that these data are isomorphic to the standard linear Stokes data of Definition 1.4.11. Section 5 identifies the filtered vector spaces (L`;FL`)` obtained by the cohomological computation of Theorem 4.3.3 with the filtered vector spaces entering in the standard linear Stokes data.
This is done by using Morse theory. However, the Morse-theoretic description of the linear Stokes data expressed through Theorem 4.3.3 does not seem suitable to compute the Stokes matricesS`1` . This is why, in Section 6, we construct a simple Leray covering giving rise to a basis of the cohomology space L`for each`, which allows us to identify the matricesS`1` given by Theorem 4.3.3 to those given by the standard linear Stokes data.
It remains to compare the Stokes filtration obtained in Theorem 4.3.3 with that obtained from the standard linear Stokes data. The problem here is that the simple Leray covering constructed in Section 6 may not be adapted to the Morse-theoretic computation. In Section 7 we change the construction of the Leray covering in order both to keep the same combinatorics to compute the matricesS`1` and to identify the filtration of Theorem 4.3.3 with the standard Stokes filtration 1.4.11(4).
This is the contents of Corollary 7.2.5, which concludes the proof of Theorem 1.3.6.
2. Stokes-filtered local systems and Stokes data
2.1 ±Reminder on Stokes-filtered local systems
We refer to [Del07], [Mal91] and [Sab13a, Chap. 2] for more details.
Let $:Af1t !A1t be the real blowing-up of the origin in A1t. Then Af1t A1t[S1t0 is homeomorphic to a semi-closed annulus, with S1t0:$ 1(0)Af1t. Similarly, we consider Af1u, andrextends as thep-fold covering :S1u0!S1t0. For any#2S1u0, the order onu 1C[u 1] at#is the additive order defined by (2:1:1) '4#0()
exp (W(u)) has moderate growth in some open sector centered at#:
We set
' <#0()'4#0 and '60:
(2:1:2)
This is equivalent to exp('(u)) having rapid decay in some open sector centered at#.
DEFINITION2.1.3 (see e.g. [Sab13a, Chap. 2]). A Stokes-filtered local system with ramificationrconsists of a local systemL onS1t0such that 1L is equipped with a family of subsheaves L4' 1L indexed by a finite subset Expu 1C[u 1] with the following properties:
(1) For each#2S1u0, the germsL4';#('2Exp) form an increasing filtration of ( 1L)# L (#).
(2) SetL< ';#P
c<#'L4c;#, where the sum is taken inL (#) (and the sum indexed by the empty set is zero). ThenL<';#is the germ at#of a subsheaf
L<'ofL4'.
(3) The filtration is exhaustive, i.e.,S
'2ExpL4' 1L andT
'2ExpL<'0.
(4) Each quotient gr'L :L4'=L<'is a local system.
(5) P
'2Exprk gr'L rkL.
(6) For each z2mp and '2Exp, setting 'z(u):'(zu), we have L4'z ' ez 1L4' through the equivariant isomorphism 1L 'ez 1 1L, where ez:S1u0 !S1u0 is induced by the multiplication by z.
REMARK2.1.4. It is equivalent to start with a family of subsheaves L<'
1L such that L<';#L<c;# if '4#c, and to define L4' by the formula L4';#T
c;c<#'L<c;#with the convention that the intersection indexed by the
empty set isL#. The familyL<'will be simpler to obtain in our computation of Laplace transform.
REMARK2.1.5 (Extension of the index set). It is useful to extend the indexing set of the filtration and to define the subsheaves L4c 1L for any c2u 1C[u 1]. For such acand for any'2Exp, we denote byS1'4c the open subset ofS1u0defined by
#2S1'4c()'4#c:
We shall abbreviate by (b'4c)!the functor on sheaves composed of the restriction to this open set and the extension by zero from this open set toS1u0. ThenL4c is defined by the formula
L4c X
'2Exp
b'4c!L4': (2:1:5)
The germ ofL4cat any#2S1u0can also be written as L4c;# X
'2Exp
'4#c
L4';#; (2:1:5)
with the convention that the sum indexed by the empty set is zero. Note also that, ifc is not ramified, i.e.c2t 1C[t 1], then L4c is a subsheaf of L. A similar definition holds forL<c, but one checks thatL<c L4cifc2=Exp. Of particular interest will beL40L.
2.2 ±Stokes data of pure level q
In this section, we will define the notion of Stokes data attached to a given Stokes structure (L;L), which - after various choices to be fixed a priori - describes the Stokes filtration in terms of opposite filtrations on a generic stalk.
After choosing appropriate basis, the passage from one filtration to the other could be expressed by matrices. These matrices are usually referred to as the Stokes matrices orStokes multipliers in different approaches to Stokes struc- tures (see e.g. [BJL79]). In level q1, the following description has already been given in [HS11], section 2.b.
We will restrict to the case of an unramified Stokes-filtered local system, i.e., we assume that, in Definition 2.1.3, r is equal to Id, and we will work with the variableu. We will setS1S1u0. We identifyS1R=2pZand will call an element
#2S1an argument.
The description of these data by combinatorial means will highly depend on the various pole orders of the factors'2Exp and their differences' cas well. The situation we have in mind as an application allows to get rid of many of these difficulties, since we will be able to restrict to a single pole order:
DEFINITION2.2.1. We say that a finite subsetEu 1C[u 1] is of pure level qif the pole order of each'2Eas well as
the pole order of each difference' cfor';c2E,'6c,
equalsq. We say that an unramified Stokes structure (L;L) is pure of levelqif the associated set Exp of exponential factors is pure of levelq.
Let Eu 1C[u 1] be a finite subset of pure levelq. In particular, there are exactly 2qStokes directions for any pair'6cinE, where a Stokes direction is an argument #2S1 at which the two factors ';c do not satisfy one of the two in- equalities'4#corc4#'. More precisely, if
(c ')(u)u qg(u)
with a polynomialg(u) withg(0)60, the Stokes directions are St(';c):n
#2S1j q#arg (g(0))p 2 or 3p
2 mod 2po :
At these directions, the asymptotic behaviour of exp(' c) changes from rapid decay to rapid growth and conversely. Let StDir(E) denote the set of all Stokes
directions StDir(E):S
'6c2ESt(';c). An argument#o2S1is said to begeneric
with respect to Eif
n#o`p
q j`2Zo
\StDir(E)[:
(2:2:2)
Let us fix a generic#oand let us set#` :#o`p=q(`0;. . .;2q 1). Then each of the open intervals (#`; #`1) contains exactly one Stokes direction for each pair '6c in E. For " >0 small enough, the same holds for the intervals I`:(#` "; #`1") with which we cover the circleS1S2q 1
`0 I`.
The category of unramified linear Stokes data of pure level q indexed by the finite setf1;. . .;rgover the fieldCis defined by (1.3.3). A morphism of two linear Stokes data of this type is given byC-linear maps between the vector spaces which commute with all theS`1` and respect the filtrations. We denote this category by Stdata(q;r). A similar definition can be given over any fieldk.
We now fix a finite set Eu 1C[u 1] of pure level q. L et Ststruct(E) be the category of unramified Stokes structures (L;L) on S1 with ExpE. After choosing an appropriate argument#o as above, we construct a functor
FE;#o :Ststruct(E)!Stdata(q;#E) (2:2:3)
as follows.
We can arrange the elements inEaccording to the ordering (2.1.1) with respect to the generic direction#o, i.e., we write
E f'1<#o '2<#o'3<#o <#o '#Eg:
We then have the same ordering at any#2mwith even index and the reverse order for#2m1. This follows from the fact that each inequality'i<# 'jbecomes reversed whenever the argument#passes a Stokes direction for the pair'i,'j, and there is exactly one Stokes direction for each pair inside (#`; #`1).
We invoke the fundamental result going back to Balser, Jurkat and Lutz:
PROPOSITION2.2.4. For each open interval I4S1 of width p=q2"for " >0 small enough, there is a unique splitting
t:LjI'M
'2E
gr'LjI compatible with the filtrations.
PROOF. See [Mal83, Lem. 5.1] or [BJL79, Th. A]. p
In particular, writing grL :L
'2Egr'L for the associated graded sheaf endowed with the Stokes filtration naturally induced by using the order (2.1.1), we
have unique filtered isomorphisms
t`:(L;L)jI` ! (grL;(grL))jI` (2:2:5)
over the intervalsI`chosen above.
For each `0;. . .;2q 1, we let L(`) :G(I`;L) be the sections of the local systemL overI`. We have canonical isomorphisms
a`:L(`) !L#` and b`:L(`) ! L#`1
betweenL(`) and the corresponding stalks ofL. Writing L`:L#` for the stalk at#`, we define
S`1` :b`1a`:L` !L`1:
We call this isomorphism the clockwise analytic continuation. Additionally, we de- fine, form0;. . .;q 1,
FjL2m:(L4'j)#2mL2m; FjL2m1:(L4'j)#2m1 L2m1: (
Then the data
(L`)`;(S`1` )`;(FL`)` define a set of linear Stokes data as in (1.3.3).
PROPOSITION2.2.6. Given a finite set E of pure level q and a generic#o2S1, the construction above defines an equivalence of categories
FE;#o :Ststruct(E)!Stdata(q;#E)
between the Stokes structures with exponential factors E and the Stokes data of pure level q indexed byf1;. . .;#Eg.
PROOF. In the caseq1, a similar statement is proved in [HS11] or [Sab13b].
The same proof holds in our situation as well. p
2.3 ±Reminder on the local Riemann-Hilbert correspondence.
Let M be a finite dimensional C ftg-vector space with connection, and let r:u7!uptbe a ramification such thatrMhas a formal Levelt-Turrittin de- composition
C u C fugrM ' M
'2Exp (M)
(E '(u) R');
with Exp:Exp(M)u 1C[u 1] (we use an opposite sign '(u) with respect to
the usual notation as it will be more convenient for the Stokes filtration). Moreover, by the uniqueness of the Levelt-Turrittin decomposition, we have, for eachz2mp (p-th root of the unity)
R'z R'; with'z(h):'(zh):
(2:3:1)
It may happen that'z'for some'2Exp andz2mp.
The Riemann-Hilbert correspondence M 7!(L;L) goes as follows (see [Del07], [BV89], [Mal91], [Sab13a, Chap. 5]). Let us denote byF the local system attached toMon a punctured neighbourhood offt0g, that we regard as well as a local system on the open annulus (A1t)or on the semi-closed annulusAf1t. Similarly, extendingras a covering map (A1u) !(A1t)orAf1u!Af1t, we definer 1F . We also use the notation
L :F jS1
t0; 1L :F jS1 u0:
One can attach toMthe family of subsheavesL4':H0DRmod 0(rM E')jS1
of 1L('2Exp), where DRmod 0denotes the de Rham complex with coefficientsu0
in the sheaf of holomorphic functions onAf1unnS1u0 having moderate growth along S1u0. The equivariance property 2.1.3(6) is obtained through the isomorphisms zrM rM. Similarly, we can consider the family of subsheaves L<':
H0DRrd 0(rM E') of 1L ('2Exp), where DRrd 0 denotes the de Rham complex with coefficients in the sheaf of holomorphic functions onAf1unnS1u0having rapid decay alongS1u0.
REMARK2.3.2. The equivalence of categoriesM !Mconsidered in § 1.2 reads as follows, by introducing the notion of Stokes-filtered sheaves. A ramified Stokes- filtered sheaf on fA1tis defined (in the present setting) as the pair formed by a local system F onAf1t and a Stokes-filtrationL of its restrictionL :F jS1
t0. Since S1t0is a deformation retract ofAf1t, givingF is equivalent to givingL, hence the equivalence.
With this notion of Stokes-filtered sheaf, we can define a subsheafF40 ofF , which coincides withF away fromS1t0, and whose restriction toS1t0isL40(see Remark 2.1.5). In the following, we will also consider the real blow-up spacePe1t ofP1t att0 andt 1, which is homeomorphic to a closed annulus, and we will still denote byF (resp.F40) the push-forward ofF (resp.F40) by the open inclusionAf1t,!Pe1t. 3. The topological Laplace transformationF(0;1)top
3.1 ±Reminder on the local Laplace transformationF(0;1)
We keep the notation as in the introduction, and we will focus on the coordinatet, so that the Laplace kernel is now exp( t=t). Moreover, since we only want to
deal with F(0;1)(M), we only consider the localized Laplace transform C[t0;t0 1]C[t0] , that we will denote by from now on. We consider the fol- lowing diagram:
The localized Laplace transform is defined by the formula
bp(pME t=t):
We can regard the C[t;t;t 1]h@t; @ti-module pME t=t both as a holonomic DP1
tA1t-module and as a meromorphic bundle with connection on P1t A1t with poles along the divisor D:D0[D1[D in P1t A1t (with D0 f0g A1t, D1 f1g A1tandD P1t fc1g). It has irregular singularities alongD. As indicated in §1.2, the germ of att0 only depends on the germMofMatt0, and it is denoted byF(0;1)(M).
Letbr:h7!h tbe a ramification such thatbr is non-ramified at infinity, that is,
C h C fhgbrF(0;1)(M)' M
2
(E (h)Rb );
withExp( )d h 1C[h 1]. ThenbrF(0;1)(M) is obtained through the diagram
by the formula
brF(0;1)(M)C fhg C[h;h 1]bp(pME t= (h)):
Moreover, by the uniqueness of the Levelt-Turrittin decomposition, we have, for each 2m (bp-th root of the unity)
Rb Rb ; with (h): ( 1h):
The stationary phase formula of [Fan09, Sab08] makes explicit the correspondence (p;Exp;(R')'2Exp)7!(bp;Exp;d (Rb ) 2 ).
3.2 ±The monodromy ofF(0;1)M
Let us denote by Lc the local system attached toF(0;1)MonS1t0. The local system Lc can equivalently be regarded as a local system Fc onGanm;t. Let us denote byTtopthe topological monodromy ofMaround the origin, and byTbtopthe topological monodromy of F(0;1)M around t0 (with the notation as in § 1.2).
Since has no singularity except at 0;1,Tbtop1is the topological monodromy of around t00. Therefore, there are various ways of calculating the topological monodromyTbtop:
either by comparing it with the monodromy att 1ofM, which is nothing, up to changing orientation, but the topological monodromy of M att0, sinceMhas no singular point on (A1;ant ), and its singular point att 1is regular; this will be done in Proposition 3.2.1;
or by a direct topological computation att 1; this is done in § 6.3;
lastly, by obtaining it from the Stokes data att 1, once we have computed them; this is however not the most economical way.
Given an automorphismT, we will denote byTl (l2C) the component of T with eigenvaluel. Since has a regular singularity att 1, with monodromy equal toTtop1, classical results give:
PROPOSITION 3.2.1. For l61, we have Tbtop;lTtop;l. Moreover, if M is a successive extension of germs of rank-one meromorphic connections, the Jordan blocks of size k52 of Tbtop;1 are in one-to-one correspondence with the Jordan blocks of size k 1of Ttop;1.
SKETCH OF PROOF. Only for this proof, we denote by the non localized Laplace transform ofM. It has a regular singularity att00. The monodromyTtop1 ofMat infinity is known to be equal to the monodromy of the vanishing cycles of at t00, while Tbtop1 is the monodromy of the nearby cycles of att00 (cf. e.g.
[Sab06, Prop. 4.1(iv)] withz1). The first assertion follows. For the second as- sertion, note that, if M is a rank-one meromorphic connection, thenM is irre- ducible as aCth@ti-module, and thus is also irreducible, so is a minimal exten- sion att00, which implies the second assertion forM. In general, note that in a exact sequence 0! 0! ! 00!0, if the extreme terms are minimal ex- tensions att00, then so is the middle term. This concludes the proof of the second
assertion. p
As in Remark 2.3.2, let us consider the real blowing-up map$:Pe1t !P1t of P1t at t0 and t 1 (so that Pe1t is a closed annulus) and let us denote by eb
:Pe1t Ganm;t!Ganm;t the projection. Then we have
Fc'R1ebDRmod (D0[D1)(pME t=t) (3:2:2)
and all otherRkebvanish. We note that, fort60, twisting withE t=thas no effect near t0. Similarly, sinceMis regular at1,E t=t is the only important factor alongt 1.
Computation. Let us use the notation of Remark 2.3.2. Let us fixto2Cand let us compute the fibreFcto. Recall thatPe1t is a closed annulus with boundary components S1t0 andS1t1. L etF40 be the sheaf H0DRmod 0M on the semi-closed annulus Pe1tnnS1t1. We denote by (Pe1t)4to1 the closed annulusPe1t with the closed interval arg t argto2[ p=2;p=2] mod 2pin S1t1deleted. This is the domain where the function exp(t=to) has rapid decay. We consider the inclusions
Pe1tnnS1t1,!a1to
(Pe1t)4to1,!b1to Pe1t: We then have (similarly to [Sab13a, §7.3]):
PROPOSITION 3.2.3. The fibre Fcto is equal to H1(Pe1t;b1to;!a1to;F40) and the other Hk vanish. It is isomorphic to the vanishing cycle space at t0 of the perverse sheaf$F40onA1t. The monodromy ofFcwith respect totis obtained by rotating the interval (p=2;3p=2)argtoS1t1in the counterclockwise direction
with respect to to. p
REMARK3.2.4. We can similarly compute the fibre of ebr 1Fc at any ho by replacing argto with bpargho in the formula above.
3.3 ±The Stokes structure ofF(0;1)M
Let us denote by Lc the Stokes filtration of Lc, regarded as a family of subsheaves ofebr 1(Lc). Due to the ramificationbr, we now work with the variableh.
Let P
g
1t A1h be the real oriented blowing-up of P1t A1h along the components D0;D1;D of Dregarded now inP1t A1h. We have a similar diagram (see e.g.[Sab13a, Chap. 8]):
We note thatP
g
1t A1his the productPe1t Af1hof the closed annulusPe1t by the semi- closed annulusAf1h. One defines on such a real blown-up space the sheaf of holo- morphic functions with rapid decay near the boundary (i.e., the inverse image by the real blow-up map of the divisorD). There is a corresponding de Rham complex that we denote by DRmodD(see loc. cit.).THEOREM 3.3.1 (Mochizuki, [Moc14, Cor. 4.7.5]). For each 2h 1C[h 1] the natural morphism
DRmod brF(0;1)(M) E (h)
!RebDRmodD(pME (h) t= (h))[1]
is a quasi-isomorphism and the natural morphism
R1ebDRmodD(pME (h) t= (h))!b|R1ebDRmod (D0[D1)(pME t= (h)) is injective.
REMARKS3.3.2 (1) The result of [Moc14, Cor. 4.7.5] refers to the real blow-up P1t Af1halongD , not along all the components ofD. The above statement is obtained by using that, with respect to$:P
g
1t A1h!P1t Af1h, we haveR$DRmodD(pME t= (h))DRmodD (pME t= (h)):
This follows from the identificationR$AmodD AmodP1 D
t (D0[D1),
as in [Sab13a, Prop. 8.9].
(2) Setbr(h)ch (1o(h)) and let us choose abp-th rootc1= (1o(h)) ofbr(h)=h . Then thebp-th roots ofbr(h) are writtenbr1= (h) hc1= (1o(h)) ( 2m ).
Now, for 2h 1C[h 1], we set (h): br1= (h) . Them -equivariance is then induced by the isomorphism
E (h) t= (h)E ; where we still denote by the maph7!br1= (h).
The theorem expresses therefore the Stokes filtration as the push-forward byeb of a family of complexes indexed byExp. The questions stated in the introductiond reduce now to
expressing, for any 2Exp, the complex DRd modD(pME (h) t= (h)), to- gether with the isomorphisms
1DRmodD(pME (h) t= (h))'DRmodD(pME );
in terms of the Stokes-filtered local system attached toM,
computing in a topological way the push-forward byebonce this complex is well understood.
While H0DRmodD is easy to compute from the data (F ;L), it happens in general that the complex DRmodD has higher cohomology, which is not easy to express in terms of (F ;L). The main reason is that the meromorphic connection
pME (h) t= (h)may not be good (in the sense of [Sab00]) at (0;c1), due to in- determinacy 0=0 of the functions
v'; (u;h): (h) r(u)=br(h) '(u) (3:3:3)
foru!0 andh!0 ('2Exp, 2Exp).d
PROPOSITION 3.3.4. There exists a projective modification e:Z!P1t A1h which consists of a succession of point blowing-ups above(0;c1), such that for each 2Exp, the pull-backd pME (h) t= (h) is good along the normal crossing
divisor DZ:e 1(D). p
This result is a particular case of a general result due to Kedlaya [Ked10]
and Mochizuki [Moc09a] (see also [Moc09b]), but can be proved in a much easier way in the present setting, since it essentially reduces to resolving the indeterminacy 0=0 of the rational functions v'; (u;h) for '2Exp(M) and 2Exp( ), and to using [Sab00, Lem. III.1.3.3]. A particular case will be made precise in §3.4. The results below do not depend on the choice of Z satisfying the conclusion of Theorem 3.3.5. We decompose DZ as the union of the strict transforms ofD0;D1;D that we denote by the same letters, and the exceptional divisor E.
The main tool is then the higher dimensional Hukuhara-Turrittin theorem (originally due to Majima [Maj84], see also [Sab93, Proof of Th. 7.2], [Moc11, Chap. 20], [Sab13a, Th. 12.5 & Cor. 12.7]). We consider the real blow-up space Ze of Z along the components of DZ and the natural lift ee:Ze !P
g
1t A1h of e:Z!P1t A1h.THEOREM3.3.5. If Z is as in Proposition3.3.4, then the complex DRmodDZe(pME (h) t= (h))
has cohomology in degree zero only. p
We can then use a particular case [Sab13a, Prop. 8.9] of [Moc14, Cor. 4.7.5], which is easier to prove becauseeis a projective modification, to get:
ReeDRmodDZe(pME (h) t= (h))DRmodD(pME (h) t= (h)):
We will apply Theorem 3.3.1 through its corollary below.
COROLLARY 3.3.6 (of Theorem 3.3.1). Let Z be as in Proposition3.3.4.Then, for each 2h 1C[h 1]the natural morphism
DRmod F(0;1)(M) E (h)
!R1(ebee)H0DRmodDZe(pME (h) t= (h)
is a quasi-isomorphism and the natural morphism
R1(ebee)H0DRmodDZe(pME (h) t= (h))!ebr 1Fb
is injective. p
The pull-back (bpe) 1(fh0g)e 1(D ) is the unionD [E. Its pull-back in Z, denoted bye De [E, is equal to (ebe ee) 1(S1h0). We will denote byeb|the open inclusionZenn(De [Ee),!Z.e
Let r:u7!upt be a ramification such thatrM is non-ramified at t0.
Consider the fibre productZe0:P
g
1uA1hZe as a topological space:
Let us writeDe0Z:es 1(DeZ) andeb|0:Ze0 nnes 1(De [E)e ,!Ze0. Note that the restric- tion ofee0toZe0 nnDe0Zis a homeomorphism.
Let$:Ze !Zbe the oriented real blow-up map. Below we use the convention of Remark 2.3.2, and forez02Ze0we set
eu(ep0ee0) (ez0)2Pe1u; Exp Exp if eu2S1u0; 0 if eu2P1unnS1u0: (
and, as usual, for a pointez02Ze0overz:$es(ez0)2Zand a functiong(u;h):
e0'4 e0g()Re e0' e0g
<0 in some neighb:ofez0
ore0' e0g is bounded in some neighb:of ($ es) 1(z):
Recall that the Stokes filtration (F 0;F 04 ) on the local systemF 0:er 1F satisfies the equivariance conditionz 1F 04'F 04'z insidez 1F 0F 0for eachz2mp.
DEFINITION3.3.7. (1) The sheafG0onZe0is defined aseb|0G0, whereG0 is the subsheaf of (pe0ee0) 1F 0 on Ze0 nn(es 1(De [E)) whose germ ate ez02Ze0 nn(es 1(De [E)) is defined bye
G0 : X
F 04'; F 0:
(2) The sheafG40 is the subsheaf ofG0 which coincides with the latter away fromDeZ0 and whose germG40 ; at eachez02Ze0is defined by the formula