\documentclass[12pt,doublespace]{amsbook}
\usepackage{graphicx}
\usepackage{amssymb,amscd,amsmath,amsthm}

\newtheorem{theorem}{Theorem}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{proposition}[theorem]{Proposition}
\theoremstyle{definition}
\newtheorem{definition}{Definition}
\newtheorem{condition}[theorem]{Condition}
%\newtheorem{remark}[theorem]{Remark}
\theoremstyle{remark}
\newtheorem{remark}{Remark}

\newcommand{\A}{{\Bbb A}}
\newcommand{\C}{{\Bbb C}}
\newcommand{\G}{{\Bbb G}}
\renewcommand{\H}{{\Bbb H}}
\renewcommand{\P}{{\Bbb P}}
\newcommand{\R}{{\Bbb R}}
\newcommand{\Q}{{\Bbb Q}}
\newcommand{\Z}{{\Bbb Z}}


\newcommand{\Res}{\operatorname{Res}}
\newcommand{\rank}{\operatorname{rank}}
\newcommand{\Tr}{\operatorname{Tr}}
\newcommand{\Sol}{\operatorname{Sol}}
\newcommand{\Spec}{\operatorname{Spec}}
\newcommand{\GL}{\operatorname{GL}}
\newcommand{\SL}{\operatorname{SL}}
\newcommand{\Hom}{\operatorname{Hom}}
\newcommand{\Pic}{\operatorname{Pic}}
\newcommand{\Der}{\operatorname{Der}}
\newcommand{\End}{\operatorname{End}}
\newcommand{\ad}{\operatorname{ad}}
\newcommand{\del}{\partial}
\newcommand{\Ext}{\operatorname{Ext}}


\newcommand{\sC}{\mathcal{C}}
\newcommand{\sD}{\mathcal{D}}
\newcommand{\sE}{\mathcal{E}}
\newcommand{\sL}{\mathcal{L}}
\newcommand{\sM}{\mathcal{M}}
\newcommand{\sO}{\mathcal{O}}
\newcommand{\sV}{\mathcal{V}}



%\title{Monodromy, formal structure of irregular connections and an example of
%the period determinant of an irregular connection over an elliptic curve}



\author{Byungheup Jun}
\date{21. July. 2002}
\address{Fb.6 Mathematik, Universit\"at-GH Essen, 45 117 Essen, Germany}
\email{mat9f0@uni-essen.de}

\begin{document}


\thispagestyle{empty}


\begin{center}
\ \\

\LARGE
A study on irregular singular algebraic connections on curves:
	monodromies, canonical decompositions and an example of 
	the determinant of period integrals \\
\ \\
\ \\
\ \\
\ \\

\large Dissertation\\
\normalsize zur Erlangung des Doktorgrades\\
(Dr. rer. nat.)\\
\ \\
\ \\
\ \\
dem Fachbereich 6 - Mathematik und Informatik\\
der Universit\"at Essen\\

vorgelegt von\\
\ \\
\ \\
\large Byungheup Jun\\
\normalsize aus Seoul, Korea\\
\ \\
\ \\
\ \\
\ \\
\ \\
Essen\\
2002\\

\end{center}


\newpage
\noindent 
$$
\begin{array}{lll}
\mbox{{\rm Vorsitzender der Pr\"ufungskomission:}} & &
	\mbox{{\rm Prof. Dr. Dieter Lutz}}\\
\mbox{{\rm Erster Gutachter:}} & & \mbox{{\rm Prof. Dr. H\'el\`ene Esnault}}\\
\mbox{{\rm Zweiter Gutachter:}} & & \mbox{{\rm Prof. Dr. Eckart Viehweg}}\\
\end{array}
$$
\\[2cm]
\noindent\hspace*{0.4mm} Tag der Disputation: 29.10.2002





\begin{abstract}
In this thesis, we will handle three aspects of  irregular connections : 
monodromies,  formal structure at irregular singular point and an example of the
determinant of a connection on the Legendre elliptic curve.

In the first part, we study monodromies of algebraic connections on the trivial vector bundle.
We prove that on an nonsingular affine variety over $\C$, any rank $1$ monodromy arises 
as the underlying local system of an algebraic connection on the trivial bundle. A generalization
of this for higher rank monodromies will be given.

The second part is devoted to a canonical decomposition of a formal connection at
its irregular singular point after its slopes.  We will give another
argument for the proof of the theorem of existence and uniqueness of
the slope decomposition and Turritin-Levelt decomposition of a connection in several variables using the insight of \cite{bbe}. 
Applying a lemma from \cite{bbe} we show that the slope decomposition and the
Turritin-Levelt decomposition of the relative connection in one variable is stabilized by 
the absolute connection.

Finally, in the last part we calculate an example of period determinant of a connection
$\nabla=d + dy$ over the Legendre elliptic curve.
%The de Rham cohomology and the cycles in the homology of the dual connection and
%describe the period matrix. 
Terasoma's work on the period of (regular singular) connections over $\P^1$ is introduced to approximate by a sequence of regular singular connections the direct image 
connection $\pi_*(\nabla)$ 
for $\pi: U\to \A^1=\Spec k[y], (x,y)\mapsto y$  as an intermediate step. 
We will make the comparison of the period of the direct image connection on $\P^1$, equivalently 
the approximation, and that of
$\nabla$ over $U$.
\end{abstract}
%\maketitle
\tableofcontents

\chapter*{Introduction}

In this thesis, we study some properties of irregular connections.

On a nonsingular variety $X$ over a field of characteristic $0$, a connection $\nabla$
 is a differential operator of order one from a vector bundle $E_X$ to $E_X\otimes \Omega_X^1$. It is the generalization of the exterior differentiation $d$ on $\sO_X$, 
which defines the de Rham complex.  
We extend the operator to $E\otimes\Omega^i$ by Leibniz rule. 
If it satisfies the relation $\nabla^2 =0$, we say $\nabla$ is integrable and call $(E\otimes\Omega^*_X, \nabla)$ the de Rham complex of $\nabla$.

Originally, connections were identified with homogeneous linear differential equations over
an analytic space. 
%Similarly to  many other correspondences between complex algebraic
%geometry and analytic geometry,
% like GAGA(\cite{GAGA}), 
%an algebraic connection $\nabla$ over a complex variety $X$ is
%naturally related to an analytic connection, which enables us to define 
%the monodromy. 
%
%one can define a connection in algebraic
%way and relate it with analytic one(homogeneous linear differential equation).
%The
%integrability  condition is analogy of the famous Frobenius theorem for the existence of integral manifold for a distribution from a text of differential geometry(see \cite{wa}). 
An analytic integrable connection yields an involutive distribution on the underlying
vector bundle. Applying the Frobenius theorem for the existence of the integral
manifold of a distribution, we have locally in a neighborhood of $p$ a unique solution of $\nabla$ with the
initial value $v$ in $E_p$.
One can extend 
the local solution of $\nabla$ 
(i.e. the local section of the analytic sheaf 
$\ker \nabla$) along a path. 
This analytic continuation of the solution along a loop based at $p$
in $\pi_1(X,p)$ yields an automorphism of $E_p$. 
This in turn represents $\pi_1(X,p)$ in
$\GL(E_p)$
and gives a monodromy representation of the fundamental group. 
The sheaf of local solutions of $\nabla$ is called the local system 
associated to $\nabla$.  
On the other hand, given an integrable
connection the de Rham cohomology is defined 
as the (hyper)cohomology of the de Rham
complex. 
The de Rham cohomology can be paired with the homology to produce periods.

When the underlying variety $X$ is projective and nonsingular, all the above remains the same
while passing from the algebraic connection to  the analytic one via GAGA.

For non-complete varieties, one cannot relate an analytic connection with the algebraic one directly.
But  with
the regularity of the connection at infinity, which appears in basic text books
of differential equations, 
the theorem of existence and uniqueness of an algebraic connection for 
a given representation
of the fundamental group was  given by Deligne in \cite{de}. 
He also gave the  comparison theorem that there is 
a natural quasi-isomorphism between the analytic 
and the algebraic de Rham complex in this case, so the two de Rham 
cohomologies are  the same. 
The classical Poincar\'e-Verdier duality also holds between the de Rham cohomology
and the de Rham cohomology with compact support (\cite{st}).

Without the regularity condition at infinity, that is when the connection is 
irregular, 
the above properties may not be satisfied in general. 
In this thesis, we are mainly interested in the  monodromies,
canonical decompositions and  periods for irregular connections.
Each topic will be presented in one chapter of this thesis.

In Chapter 1, we  will handle the monodromy of  irregular connections
on nonsingular affine varieties. The result is based on a theorem which  first appeared
in a letter  from Deligne to Esnault 
(\cite{DEL})
showing that  for any rank $1$ representation of the fundamental group of
an affine curve
there exists a connection on the trivial bundle of rank $1$ with the chosen representation as its monodromy. 
%We generalize the theorem in two directions. One is for rank $1$ representation of the fundamental group of a higher dimensional affine variety. 
We will directly extract from the proof the obstruction to realize a rank $1$ representation of the fundamental group as a connection on the trivial
bundle of rank $1$.
We will generalize the theorem for rank $1$ representations on any affine variety and for representations of
arbitrary rank over curves. 
%The other is for higher rank representations of
%the fundamental group of an affine curve. We finally prove that any representation of
%the fundamental group of an affine curve can be realized as an algebraic connection.
This chapter is based on  the result  published as \cite{j1}.

Chapter 2 is devoted to the canonical decompositions
of a formal connection at its irregular singular point after the slopes. 
We present the slope decomposition and the Turritin-Levelt decomposition for
absolute connections.
For relative connections, it is  presented by several authors in different situations
(cf. \cite{bbe}, \cite{le} and \cite{ma1}).
While those proofs are computational, here
we present them in short  by virtue of  lemma 6.4.3 of \cite{bbe}.
All the idea of Chapter 2 is essentially contained in loc. cit. and this presentation
will be published separately in \cite{j2}.

Finally, in Chapter 3, we calculate
the period determinant of an example of an irregular connection $\nabla=d + dy$ on the
affine Legendre elliptic curve $\Spec (k[x,y]/(y^2 - x(x-1)(x-\lambda))$. 
Historically, periods are defined as an integration of   ``algebraic
differential" forms. A general definition appears in \cite{kz}.
Some periods are produced while taking the pairing of homology with cohomology.
Bloch and Esnault developed the homology for irregular connections on curves
and proved that it makes a perfect pairing with de Rham cohomology(\cite{be}).
In the last part of  \cite{kz}, 
the period of the pairing between the de Rham cohomology and the homology for an irregular connection is discussed and named ``exponential period''.
%At the end of loc.cit., 
We are interested in the determinant of the period matrix in this case.
% of pairing of the homology for
%an irregular connection and the de Rham cohomology. 
On $\P^1$, some examples
were calculated by Terasoma in \cite{te2} through an approximation by a sequence of
the periods of regular connections. 
We give an example in the higher genus case, which reduces  to a computation on $\P^1$ via the projection formula. The period determinant will be approximated using the product formula in \cite{te1}.

\chapter*{Acknowledgements}

First of all, I would like to thank my thesis advisor Professor H\'el\`ene Esnault for giving me the greatest chance in my mathematical life as well as for those advises on mathematics 
and on the life. 

I am also grateful to Professor Eckart Viehweg for warm advises and for teaching me the cohomology theory in the year 1997. It does not count in my Ph.D period, nevertheless
I learned how an expert  sees the objects in algebraic geometry. 
I thank Professor Spencer Bloch
for showing interest on this work and the encouragement.
Additionally I thank him for the arrangements
when I visited the University of  Chicago twice.

I owe Prof. Pierre Deligne the first theorem in Chapter 1 in this thesis and am thankful
for reading the chapter and for some suggestions as the referee when I submitted the part. 
I thank Professor Bernard Malgrange for giving useful comments on the materials
in Chapter 2
and for reading and accepting \cite{j2}.
I also thank Professor Tomohide Terasoma for his mathematical work. Even though I haven't
met him in person, his work inspired me to produce the result in the last chapter.

I would like to thank Silke Lekaus, Alexander Schwarzhaupt, Pedro Luis del Angel,
Herbert Kanarek, Andreas Knutsen, Stefan M\"uller-Stach, Jaya Iyer, Tomasz Szemberg, Marco Hien,  Keiji Oguiso, Jong-Won Lee, Wioletta Syszdek, Stefan Kukulies, Jinhyun 
Park
for mathematics, drinking beer together and some other fun in life.
Especially, to Silke Lekaus, I am really grateful for  escorting  me several times
to the
most dreadful
place(you-know-where) in Essen.

My study in Essen was supported by the DFG-Graduiertenkolleg ``Mathematische und
ingenieurwissenschaftliche Methoden f\"ur sichere Datenubertragung und
Informationsvermittlung" at the University of Essen.

Finally, I thank my parents. Especially, my mother taught me the powers of 2 when I was a 
kid before entering primary school. It was the first step of my mathematical life.

 
\chapter[Monodromy of an algebraic connection]{Monodromy of algebraic 
connections on the trivial vector bundle}



Let $X$ be a nonsingular projective algebraic variety over $\C$. $X_{an}$ will denote $X(\C)$ the set of $\C$-valued points in $X$ with strong topology.
Using the existence theorem
of Cauchy-Kovalevska and Serre's GAGA (\cite{GAGA}), 
we have a version of  Riemann-Hilbert
correspondence: there exists a unique algebraic vector bundle with a connection for a given
linear representation of the fundamental group $\pi_1(X_{an})$ and vice versa.

Let $U$ be a nonsingular  affine  open set in $X$ and  $D=X-U$ be normal crossing divisors.
Then the above Riemann-Hilbert correspondence is not true  any more. There may 
exist several algebraic connections with a given monodromy. 
Deligne showed that  there is again the
Riemann-Hilbert correspondence, requiring the regularity condition(\cite{de}).


%So one can think about another restriction to have some good conditions on
%the realization of the monodromy.
In this chapter, allowing irregular singularity, we are interested in a question
how far one can trivialize the underlying vector bundle of the connection 
for a given monodromy. A stronger version of this asks if  the underlying
vector bundle of the connection can be  trivial.
In one of the letters from Deligne to Esnault \cite{DEL}, 
Deligne answered without the regularity condition 
any rank $1$-representation 
of the fundamental group of an affine curve can be obtained
as a connection on a trivial bundle(see \S \ref{r1}).
Esnault interpreted the obstruction of the realization of the rank 
$1$-representation on a trivial bundle as the cokernel of the map
\begin{equation}
H^0(U,\Omega^1_U)\stackrel{j}{\to} H^1(U_{an},\C).
\end{equation}

The second cohomology group is the singular cohomology group of $U_{an}$ with $\C$
coefficient and can be identified with the $\sC^\infty$ de Rham cohomology 
$H^1_{dR,\sC^\infty}(U_{an})$, the analytic de Rham cohomology 
$H^1_{dR, an}(U)=\H^1(U_{an},\Omega^*_{U_{an}})$
and the algebraic de Rham cohomology $H^1_{dR, alg}(U)=\H^1(U,\Omega^*_U)$ 
after the Grothendieck's generalized de Rham theorem. 
Thanks to Grothendieck's de Rham theorem, the above de Rham cohomologies
are isomorphic to each other and 
we may denote the algebraic de Rham cohomology by $H^1_{dR}(U)$ simply.

From the construction of the hypercohomology, $\H^1(U,\Omega^*_U)$ is generated
by $H^0(U,\Omega^1_U)$ and $H^1(U,\sO_U)$. 
When $U$ is affine, the higher coherent sheaf cohomology should be trivial
so that the previous map is surjective.
Then Deligne uses the duality of homology and cohomology group to realize
the rank $1$-representation on the trivial bundle. 
Once we assume the surjectivity of the map $H^1(U,\C)\to H^1(U,\C^*)$,
it is  extended directly to rank $1$ case on higher dimensional 
varieties(Corollary \ref{equiv} and Theorem \ref{ns}).

Theorem \ref{repremon} extends the similar argument for representations
of arbitrary rank.\\
%on any 
%rank of representation.\\

\textsc{Theorem \ref{repremon}.}
Let $U$ be a smooth affine curve over $\C$.
For any given rank $r$  representation  $\rho$ of 
the fundamental group, there exists an algebraic connection $\nabla$
on the trivial vector bundle of rank $r$ with the monodromy $\rho$.\\

In section \ref{chen}, we recall the iterated integral \`a la K.T. Chen to 
describe the monodromy of connections on a trivial vector bundle.

The main theorem will be proved in section \ref{curvemon}.

As an example for the realization problem, we present any  
unipotent representation of rank $2$ can appear as a connection on
the trivial bundle(see Section \ref{uni}).

We concentrate on the dimension $1$ case. But some results are valid even for
higher dimensional varieties.
%treat any 
%rank of representations on  and some concrete examples: Deligne's work on 
%rank $1$-representation(see \S \ref{r1}) and 

\section{Rank 1 representations}
\label{r1}

Let  $U$ be a nonsingular affine curve, $X$ be its unique completion
and $D=X-U$. 
%Let $p$ be a fixed point in $U(\C)$.
\begin{theorem}[Deligne\cite{DEL}]
\label{fir}
For given rank $1$ monodromy representation of $\pi_1(U_{an})$, for any given
rank $1$ monodromy representation $\rho$ of $\pi_1(U_{an})$, there exists an
algebraic connection $\nabla$ on $\sO_U$ with the underlying monodromy $\rho$.
\end{theorem}

\begin{proof}
From the exponential sequence, we have the following exact sequence of
singular cohomology groups:
\begin{equation}
\label{ex}
H^1(U_{an},\C) \stackrel{\exp}\longrightarrow H^1(U_{an}, \C^*) \longrightarrow
H^2(U_{an},\Z) \to H^1(U_{an}, \C).
\end{equation}

We know $H^1(U_{an},\C^*)$ is the group of isomorphism classes of rank $1$
local systems on $U_{an}$ and $H^2(U_{an},\Z)$ vanishes by dimension 
reason.

Via the identification(cf. \cite{gr})
$$
H^1_{dR} (U) := \H^1(U,\Omega^\bullet) = H^1(U_{an}, \C),
$$
the de Rham class of $\omega$ in $H^0(U,\Omega^1)$ yields a class $[\omega]$
in $H^1(U_{an},\C)$, such that
$$
\exp [\omega] \in H^1 (U_{an},\C^*) = \Hom(\pi_1(U_{an}),\C^*)
$$
is the underlying monodromy of $(\sO, d-\omega)$.

On an affine variety, the higher coherent sheaf cohomology vanishes. 
Thus $H^0(U,\Omega^1)$ surjects onto $\Hom (\pi_1(U_{an}), \C^*)$.

This finishes the proof.
\end{proof}

A clever observation of the proof makes us generalize in the following:
\begin{corollary}
\label{equiv}
Let $U$ be an affine variety over $\C$. Then the followings are equivalent.
\begin{enumerate}
\item $H^2(U_{an},\Z)$ is torsion-free.
\item For any given rank $1$ representation $\rho$ of $\pi_1(U_{an})$, 
there exists an integrable algebraic connection $\nabla$ on $\sO_U$ with the monodromy 
$\rho$.
\item If a line bundle $\sL$ has vanishing Chern class in $H^2_{dR}(U)$, then
for any rank $1$ representation $\rho$ of $\pi_1(U_{an})$, there exists
an integrable algebraic connection $\nabla$ on $\sL$ with
the monodromy $\rho$.
\end{enumerate}
\end{corollary}

\begin{proof}
(1) $\Leftrightarrow$ (2)
As in the proof of Theorem \ref{fir}, one has the exact sequence arising from
the exponential sequence (\ref{ex}). 
On an affine variety $U$, $\Gamma(U,\Omega^1_{closed})$ generates
$H^1_{dR}(U) = H^1(U_{an},\C)$. The the map $\exp$ is surjective if and only
if 
$$
\Gamma(U,\Omega^1_{closed}) \to H^1(U_{an}, \C^*)
$$
is surjective of equivalently, $H^2(U_{an},\Z)$ is torsion-free.

(2) $\Rightarrow$ (3) Consider the complex of sheaves on the Zariski 
topology of $U$:
$$
\sC^\bullet : \sO^* \stackrel{d \log}{\rightarrow}
\Omega^1 \stackrel{d}\to \Omega_2 \rightarrow \cdots.
$$
The first hypercohomology of $\sC^\bullet$ is the group of isomorphism classes of 
line bundles with integrable connection, which will be denoted by $\Pic^{\nabla}(U)$.
The filtration of $\sC$ by its degree gives
$$
\begin{CD}
\H^1(U,\sC^\bullet) @>>> H^1(U,\sO^*) @>>> \H^2(U,\Omega^{\ge 1}) \\
 @|  		@|    @| \\
\Pic^\nabla(U) @>>> \Pic(U) @>{c}>> H^2_{dR} (U),
\end{CD}
$$
where $c$ denotes the first Chern class map and the last vertical map is an
isomorphism since the kernel and the cokernel are
$$
H^1(U,\sO) = H^2(U,\sO) = 0.
$$

Since $c(\sL)=0$, from the exact sequence, there exists an integrable
connection $\nabla_0$ on $\sL$.
Let $\rho_0$ be the monodromy of $(\sL, \nabla_0)$. As one can realize
any given monodromy on $\sO$, one has an algebraic connection $\nabla_1$
with the monodromy $\rho_0^{-1}\circ \rho$ on $\sO$. Twisting $(\sL, \nabla_0)$ 
with $(\sO,\nabla_1)$, one has the connection $\nabla=\nabla_0\otimes \nabla_1$
on $\sL$ with the desired monodromy $\rho$.

(3) $\Rightarrow$ (2) Clear. 
\end{proof}

As a special case of the corollary, we have the following:
\begin{theorem}
\label{ns}
Let $X$ be a smooth projective variety over $\C$ and $D$ be a normal crossing 
divisor such that the complement $U$ is affine and such that the components
generate the N\'eron-Severi group $NS(X)$ of $X$. Then for any  rank $1$ monodromy $\rho$
in $\Hom(\pi(U_{an}), \C^*)$, there exits an algebraic connection $\nabla=d-\omega$ on
$\sO_U$ with the underlying monodromy $\rho$.
\end{theorem}

\begin{proof} 
By the equivalence of  Corollary \ref{equiv},
it suffices to show that $H^2(U_{an},\Z)$ is torsion-free.
But in the exact sequence (\ref{ex}), torsion comes from $H^1(U_{am},\C^*)$.
Using the existence theorem in \cite{de}, one has a line bundle
on $X$ with connection $(\sL,\nabla)$ which is regular singular at $D$
with the underlying monodromy $\rho$. The image of $\sL$ maps into 
$NS(X)$, and thus a torsion element in $H^2(U_{an},\Z)$ vanishes.
Therefore $H^2(U_{an}, \Z)$ is torsion-free.
\end{proof}
\begin{remark}
With a commutative diagram, one can summarize the result in a simple
form where all rows and columns are exact:
\begin{equation}
\begin{array}{ccccccccc}
   & &  0 &&  0 && 0 \\
   & & \Big\downarrow & & \Big\downarrow & & \Big\downarrow \\
0& \to & {\frac{H^0(U,\Omega^1)_\Z}{d\log H^0(U,\sO^*)}} & \to & 
{\frac{H^0(U,\Omega^1_{closed})}{d\log H^0 (U,\sO^*)}} & \to & H^1(U_{an}, \C^*) & \to & 0 \\
 & & \Big\downarrow & & \Big\downarrow & & \Big\downarrow \\
0 & \to & \Pic^{\nabla_0}(U) & \to & \Pic^\nabla (U) & \to & \Hom (\pi_1(U_{an}),\C^*) & \to & 0 \\
  & & \Big\downarrow & & \Big\downarrow & & \Big\downarrow \\
0 & \to & \Pic (U) & \to & \Pic(U) & \to & 0 \\
 & & \Big\downarrow & & \Big\downarrow & & \\
 & & H^2_{dR} (U) &  = & H^2_{dR}(U)
\end{array}
\end{equation}
In the above diagram, $\Pic^{\nabla_0}(U)$ is the subgroup of $\Pic^\nabla(U)$ 
with respect to the trivial monodromy and $H^0(U,\Omega^1)_\Z$ is the group of 
 global 
$1$-forms with integral periods. When $\dim U =1$, one has $H^2_{dR}(U)=0$.
Thus, we have $\Pic^\nabla(U)$ and $\Pic^{\nabla_0}(U)$ as extensions of
$\Pic(U)$.
\end{remark}

\section{Chen's iterated integral}
\label{chen}
Suppose $W_1,\ldots,W_k$ are matrices in $M(r\times r, \Gamma(X,\Omega^1))$ 
and $\gamma:[0,1]\rightarrow X$ a piecewise smooth path.
Then the iterated integral of $W_1,\ldots,W_k$ over $\gamma$ is defined
as
\begin{equation}
\int_{\Delta_k \gamma} \negthickspace W_1\ldots W_k
= \int_0^1 \negthickspace \int_0^{t_1} \negthickspace \cdots\int_0^{t_{k-1}}
  \negthickspace  w_1(t_1)w_2(t_2)\cdots w_k(t_{k-1}) dt_k\cdots dt_1,
\end{equation}
where $w_i = \gamma^* W_i$.

Let $\nabla = d-A$ be a connection on a trivial bundle of rank $r$.
Thus $A\in \Gamma(X, \Omega^1\otimes \frak{g}\frak{l}(r)).$

\begin{proposition}[Chen\cite{C1},\cite{C2}]
Suppose $\gamma$ is a loop, then the monodromy of the connection $\nabla=d-A$
along $\gamma$ is given
by the absolute convergent series of the iterated integrals.
\begin{equation}
m(\gamma) = I + \int_\gamma A + \int_{\Delta_2 \gamma} AA
        + \int_{\Delta_3 \gamma} AAA + \cdots  \in \GL(r,\C).
\end{equation}
\end{proposition}
\begin{proof} (compare to \cite{C1}, \cite{C2} and \cite{Si})
Let us prove the absolute convergency of the series first.
The $k$-th term of the series is bounded by
\begin{equation}
\begin{split}
 ||\int_0^1 \negthickspace \int_0^{t_1} \negthickspace \cdots\int_0^{t_{n-1}}
  \negthickspace \underset{\text{$k$-times}}{\underbrace{A \cdots A}} ||
& \leq
\int_0^1 \negthickspace \int_0^{t_1} \negthickspace \cdots\int_0^{t_{n-1}}
  \negthickspace ||A||^k\\
& \leq \frac{1}{k!} |\gamma|^k ||A||^k,
\end{split}
\end{equation}
where $|\gamma|$ is the length of the loop $\gamma$ and $||A||$ is the matrix
norm of $A$.
Thus the summation is bounded by the $\exp| \gamma | \|A\|$.
Hence the series is absolute convergent.

Let us denote the $k$-th term of the summation by $I_k(\gamma(t))$.
Then we have
\begin{equation}
\begin{split}
\nabla_{\gamma'(t)} &I_k(\gamma(t))  = (d/dt - A(t))I_k(\gamma(t))\\
&= \frac{d}{dt}\int_0^t\negthickspace \int_0^{t_1}\cdots \int_0^{t_{k-1}}
        A(t_1)A(t_2)\ldots A(t_k) dt_k\ldots dt_2 dt_1\\
& \quad - A(t)I_k(\gamma(t)) \\
&= \int_0^{t}\int_0^{t_2}\cdots\int_0^{t_{k-1}} A(t)A(t_2)\ldots A(t_k)
        dt_k\ldots dt_2 \\
& \quad - A(t)I_k(\gamma(t)) \\
&= A(t)\int_0^t\int_0^{t_1}\cdots\int_0^{t_{k-2}} A(t_1)A(t_2)\ldots A(t_{k-1})
        dt_{k-1}\ldots dt_2 dt_1 \\
& \quad - A(t)I_k(\gamma(t)) \\
&=  A(t)I_{k-1}(\gamma(t)) - A(t)I_k(\gamma(t)).
\end{split}
\end{equation}
Now we can calculate the whole covariant derivative
of $m(\gamma)$  along $\gamma$:
\begin{equation}
\begin{split}
\nabla_{\gamma(t)}m(\gamma) & = \nabla_{\gamma'(t)} [ I + \int_\gamma A +
        I_2(\gamma(t)) + I_3(\gamma(t)) + \ldots]\\
&= -A(t) + [A(t) - A(t)\int_\gamma A] + [I_1(\gamma(t)) - A(t) I_2(\gamma(t))]\\
& \quad + [A(t) I_3(\gamma(t)) - A(t) I_4(\gamma(t))] + \ldots \\
&= [-A(t) + A(t)] + [-A(t)I_1(\gamma(t)) + A(t)I_1(\gamma(t))]\\
& \quad  + [- A(t) I_2(\gamma(t))+ A(t) I_2(\gamma(t))] +\ldots\\
&= 0.
\end{split}
\end{equation}
Therefore $m(\gamma)$ gives the monodromy around the loop $\gamma$.
\end{proof}

\begin{corollary}[Chen\cite{C1}]
On the structure sheaf $\sO$ with connection $\nabla = d - \omega$ where $\omega$ is an
$1$-form on the affine open set of a smooth curve, the monodromy along $\gamma$
is given by
\begin{equation}
m(\gamma) = \exp \int_\gamma \omega.
\end{equation}
\end{corollary}
\begin{proof} (compare to \cite{C1})
%We use the same convention as in the proof of the previous theorem.
When the bundle is trivial of rank $1$, the iterated integral can be seen as
the Lebesgue integral of
\begin{equation}
\begin{split}
A(t_1,..., t_k) &\overset{def}= W(t_1)W(t_2)\cdots W(t_k) \\
        & = \gamma^*\omega(t_1)\gamma^*\omega(t_2)\cdots
        \gamma^*\omega(t_k)
\end{split}
\end{equation}
over $\Delta_k= \{ (t_1,t_2,\ldots,t_k)\in [0,1]^k| 0\leq t_1 \leq
        t_2 \leq \ldots \leq t_k \leq 1 \}$.
The integration gives a complex measure on $[0,1]^k$.
Let $\mu(M)$ be $\int_M A$ for a measurable set $M$ in $[0,1]^k$.

We have
\begin{equation}
\begin{split}
\mu([0,1]^k) &= \int_0^1\int_0^1\cdots\int_0^1
                W(t_1)W(t_2)\cdots W(t_k) dt_k\ldots dt_1\\
        & = [\int_0^1 W(t) dt]^k = [\int_\gamma \omega]^k
\end{split}
\end{equation}
by Fubini theorem.

On the other hand, we have again
\begin{equation}
\begin{split}
\mu([0,1]^k) &= \sum_{\sigma\in S_k} \mu(\Delta^\sigma_k) \\
        & = k! \mu(\Delta_k)
\end{split}
\end{equation}
where $S_k$ is the permutation group of $k$-elements and $\Delta^\sigma_k$ is
the following $k$-simplex
\begin{equation}
\Delta^\sigma_k
        = \{ (t_1,...,t_k)\in I^k| \sigma(t_1)\leq\ldots\leq \sigma(t_k) \} .
\end{equation}

It follows that
\begin{equation}
\begin{split}
I_k(\gamma) &=  \mu(\Delta_k) \\
        &= \frac{1}{k!} \mu(I^k) = \frac{1}{k!} [\int_\gamma \omega]^k.
\end{split}
\end{equation}


Therefore,
\begin{equation}
\begin{split}
m(\gamma) &= \sum_{k=0}^\infty I_k(\gamma)
        = \sum_{k=0}^\infty \frac{1}{k!}[\int_\gamma \omega]^k \\
        &= \exp\int_\gamma \omega.
\end{split}
\end{equation}
\end{proof}

\begin{remark}
Compared with the local monodromy formula in \cite{de}, 
the previous corollary gives the
same result from another side because $\Res_p(\nabla)$ is $(-1/2\pi i)\int_\gamma \omega$ 
in this case, where $\gamma$ is a simple loop around $p$.
\end{remark}


\section[Monodromy representation of arbitrary rank]{Realization of monodromy representations of arbitrary rank}
\label{curvemon}
%\input{gen}
This section begins with a lemma due to Serre.
We recall its geometric version  due to  Atiyah\cite{At}.
%The lemma is a weak version of more general theorem due to Serre in \cite{Bou}.
\begin{lemma}[Serre]
Let $E$ be a vector bundle of rank $r$ on an affine curve $U$, then
\begin{equation}
E = \sO^{r-1} \oplus \sL,
\end{equation}
where $\sL = \det E = c_1(E) \in \Pic(U)$.
\end{lemma}
\begin{proof} (Atiyah \cite{At})
%Let $X$ be the unique smooth projective completion of $U$.
%And denote the smallest coherent sheaf on $X$ containing $E$ by $E$ again.
%Since the points in $D=X-U$ are very ample, $E(lp)$ is generated by
%global sections on $X$ and still $E(lp)(U)=E(U)$.
%Denote those global sections by $s_1,...,s_k$ where $k=\dim H^0(E(lp))$.
%Then we get $E(lp)$ from the quotient of the trivial bundle of rank $k$.
%\begin{equation}
%\begin{split}
%\bigoplus_{i=1}^k\sO & \overset{p}{\longrightarrow} E(lp)\\
%(f_1,...,f_k) &\longmapsto     \sum_{i=1}^k f_i s_i
%\end{split}
%\end{equation}
%where $f_i$ lives in $\sO$.
%
If $\rank(E)=1$, then $\det(E)=E$. 
So we have nothing to prove.

Suppose that there exists a global nonvanishing section $s$
in the vector bundle $E$ whose rank is greater than $1$.
Then we have a vector bundle $E'=E/\sO$ defined by the exact sequence:
\begin{equation}
\begin{split}
0 \to \sO \to &E \to E' \to 0.\\
       1 \mapsto &s
\end{split}
\end{equation}
%Denote $E/\sO$ by $E'$.
The above exact sequence is splitting, since $\Ext^1(E',\sO)$ is trivial on the
affine curve.
Hence we have $E=\sO\oplus E'$.
If $E'$ is of rank $1$, then there remains nothing to do.
When $E'$ is of rank greater than $1$, then the existence of global
nonvanishing section in the vector bundle on affine curves concludes
the proof.

Let $X$ be the smooth projective completion of $U$ and $E$ be a vector bundle
of rank $r$ on $U$.
And denote a locally free sheaf $F$ on $X$ such that $F|_U=E$.
Since the divisor $D=X-U$ is ample in $X$, $F(\ell\cdot D)$ is globally
generated for a sufficiently large $\ell$ on $X$ and we have still
$F(\ell \cdot D)|_U = E$.
Hence there is an exact sequence of sheaves:
\begin{equation}
0 \to K \stackrel{i}{ \to} \sO^{r+k} \overset{p}{\to} E \to 0,
\end{equation}
where $K$ is the kernel of $p$.


Tensoring the previous exact sequence with 
$\kappa (x)=\sO_x/m_x$ for $x\in U$, we have the
following exact sequence of vector spaces.
\begin{equation}
0\to K_x \overset{i_x}{\to} \C^{r+k} \to E_x \to 0.
\end{equation}
The first inclusion yields the inclusion of projective spaces:
\begin{equation}
\P(K_x) \overset{\P_x(i)}{\longrightarrow} \P^{r+k-1}.
\end{equation}
On the whole $U$, we have the inclusion of the associated projective bundles
\begin{equation}
\P(K) \overset{\P(i)}{\longrightarrow} \P(\sO^{r+k}) = \P^{r+k-1} \times U
\overset{\pi_1}{\longrightarrow} \P^{r+k-1},
\end{equation}
where $\pi_1$ is the first projection.

Then we have, since $r=\rank(E)$ is greater than $1$,
\begin{equation}
\begin{split}
\dim (\pi_1\circ\P(i)(\P(K))) & \le \dim(\P(K_x)) + \dim(U) \\
         &= k < r+k-1 = \dim (\P^{r+k-1}).
\end{split}
\end{equation}

Hence the map cannot be onto and we have a point $p$
outside of the image of the $\pi_1\circ\P(i)$ in $\P^{r+k-1}$.
It corresponds to a global section of $E$ nowhere vanishing on $U$.
\end{proof}

\begin{remark}
The previous theorem presented above holds for
any  projective module $M$ of finite rank $r$ over a Dedekind domain $A$.
We have $M \cong A^{r-1}\oplus L$, where $L\in \Pic(A)$. The above theorem
is a special case when $U=\Spec A$.
\end{remark}

As a direct consequence, a vector bundle $E$ on a nonsingular affine curve
is trivial if and only if $\det E = \sO$.

Finally we present the main theorem of this chapter.

\begin{theorem}
\label{repremon}
Let $U$ be a smooth affine curve over $\C$.
For any given rank $r$  representation  $\rho$ of 
the fundamental group, there exists an algebraic connection $\nabla$
on the trivial vector bundle of rank $r$ with the monodromy $\rho$.
\end{theorem}
\begin{proof}
Let $X=U\cup D$ be the smooth completion of $U$. Using the existence theorem
in \cite{de}, there exists a vector bundle $E$ on $X$ with a connection $\nabla_0$
regular singular at $D$, such that its restriction on $U$ gives the monodromy $\rho$.

By the decomposition theorem of a vector bundle, we have 
$E|_U \cong \sO^{r-1}\oplus\sL$, where $\sL =\det (E|_U)$.
Since $\Pic(U)$, as a quotient of a divisible group $\Pic^0(X)$, is divisible, 
one has a line bundle $\sM \in \Pic(U)$ such that $\sM^{\otimes r} = \sL$.
Moreover, using the equivalences in the Corollary \ref{equiv}, $\sM$ can be equipped
with an algebraic connection $\nabla_1$ with the trivial monodromy as $H^2_{dR}(U)$ vanishes.
Twisting $(E|_U,\nabla_0)$ with $(\sM, \nabla_1)^{\otimes -1}$ trivializes the underlying
vector bundle of the connection $\nabla_0\otimes\nabla_1^{\otimes -1}$ since
$$
\det (E|_U \otimes \sM^{-1}) = \sL \otimes \sM^{\otimes -r} =\sO \quad \text{in $\Pic(U)$.}
$$
\end{proof}

%\begin{remark}
%In the above theorem, one can not  
%$A$ is a matrix of one forms, which might have higher poles at $D$.
%ie. One cannot say that it is regular anymore.)
%\end{remark}


\section{Unipotent rank 2 representations}
\label{uni}
%\input{upot}
Now we know on a curve every monodromy representation can be realized on 
the trivial bundle.
Deligne's construction for the realization on the trivial bundle of rank $1$ 
can be seen as a specific example of the main theorem in the previous section.
We can construct one more example after the idea of Deligne.

Let us recall the result of Deligne in the first section.
It says that for any rank $1$ representation of the fundamental group of the
open affine curve, there exists an algebraic connection on the structure
sheaf with the given underlying monodromy.

Let $U$ be a nonsingular affine variety over $\C$.

Given  a representation $\rho$ of $\pi_1(U_{an})$ paths through $H_1(U_{an})$.
Then there exists a closed differential form $\omega$ in $H^0(U,\Omega)$ such
that
\begin{equation}
\rho(\gamma) = \exp\int_{\gamma} \omega,
\end{equation}
for every $\gamma$ in $\pi_1(U)$.
It is possible because $H^1(U,\C)$ is dual to $H_1(U_{an})$ by the Poincar\'e
duality and $H^0(U,\Omega)$ surjects to the first de Rham cohomology group.

Now the same idea can be applied to the unipotent rank $2$ case.
There is an isomorphism between $\G_a$ and $U(2)$ the unipotent matrix group
of rank $2$.
The isomorphism is given as following:
\begin{equation}
\begin{split}
\G_a & \overset{\simeq}\longrightarrow U(2)\\
 a & \mapsto \begin{pmatrix}1 & a \\ 0 & 1 \end{pmatrix}.
\end{split}
\end{equation}

Then any representation $\rho$ can be written as
\begin{align}
\rho(\gamma) =
        \begin{pmatrix} 1 & \tilde{\rho} \\
                        0 & 1 \end{pmatrix}
\end{align}
where $\tilde{\rho}$ is an additive representation of the fundamental group
to $\G_a$.

The following theorem is, in fact, a consequence of Corollary \ref{equiv}. 
But it shows the shape of a connection realizing unipotent rank $2$ representation.

\begin{theorem}
For any unipotent rank $2$ representation, 
there exists an algebraic connection $\nabla$ on the rank $2$ trivial
bundle whose underlying local system corresponds to the prescribed monodromy
representation $\rho$ if and only if any rank $1$ representation can be
found as the underlying monodromy of a connection on the trivial bundle.
\end{theorem}
\begin{proof} 
From the above paragraph, it suffices to show that
there exists a closed differential form $\omega$ such that
$\tilde{\rho}(\gamma) = \int_{\gamma} \omega$
by the same idea as Deligne on the rank $1$ case.
We can see the representation can be realized on the trivial bundle by the
connection $\nabla = d - A$ where the connection matrix is
\begin{align}
A = \begin{pmatrix} 0 & \omega \\ 0 & 0 \end{pmatrix}.
\end{align}

The monodromy can be calculated by Chen's iterated integral.
The monodromy of $\nabla= d-A$ around $\gamma$ is:
\begin{equation}
\begin{split}
I+ &\int_{\gamma} A + \int_{\Delta_2 \gamma} AA + \int_{\Delta_3 \gamma} AAA
 \ldots \\
 & = I + \int_\gamma A \quad \text{(for $A^2=0$.)} \\
 & = I + \begin{pmatrix} 0 & \int_\gamma \omega \\
                        0  & 0  \end{pmatrix} \\
 & = \begin{pmatrix} 1 & \tilde{\rho}(\gamma) \\
                     0 & 1  \end{pmatrix} \\
 & = \rho(\gamma),
\end{split}
\end{equation}
which fulfills  the desired monodromy.
\end{proof}

%\input{maac.tex}

\chapter[Canonical decomposition of a formal connection]{Canonical decomposition of a formal connection at an irregular singular point}


In this chapter  we recall Malgrange's theorem 3.2.1 in \cite{ma1}
and give a short and simple argument for the proof: 
If $\nabla$ is an absolute integrable connection on 
$R((z)) := \C \{x_1,x_2,\ldots, x_n\}((y))$, then the Levelt-Turritin
decomposition of the $R$-relative connection $\nabla_{rel}$ after
ramification $y^{1/N}$ is stabilized by $\nabla$.
Indeed Malgrange's theorem is reformulated by Andr\'e and Baldassarri
in \cite{ab} algebraically. Applying Galois descent for $R((y^{1/N}))$ 
over $R((y))$, we have a corollary of the result that the slop decomposition
on $R((y))$ of $\nabla_{rel}$ is also stabilized by $\nabla$.

Here, in the sprit of Lemma 6.4.3 of \cite{bbe},
while both proofs are computational, we don't need any explicit computation on the
order of poles. Firstly, we observe the slope decomposition in one variable is 
stabilized under $\nabla$. 
Then starting from the slope decomposition, passing by a standard argument
(splitting lemma), we conclude that the Turritin-Levelt decomposition
after ramification is also absolute.

The material presented in this chapter was known already in \cite{ab}, but 
as in Chap.3 of \cite{ma}, starting from the fact that the slope decomposition
stabilized, we can conclude that the classical (but complicated) Turritin-Levelt
decomposition is stabilized by $\nabla$.

\section{Slope decomposition of a relative connection}

Let $k$ be a field of characteristic $0$ and $K$ be any extension of $k$. 
Throughout this chapter, the term ``absolute'' means relative to $k$.

Denote by $\sD$ the ring of differential operator over $K((z))$.
An absolute connection $\nabla$ on $V$ a finite dimensional  vector space over $K((z))$ is a differential
operator of order $1$ from $V$ with values in $V\otimes_k \Omega_{K((z))/k}$
\begin{equation}
\begin{split}
\nabla : V \to V\otimes_k \Omega_{K((z))/k},
\end{split}
\end{equation}
satisfying the Leibniz rule $\nabla (f\cdot v) =  df \cdot v + f\nabla(v)$ for $f\in K((z))$ and $v\in V$.

It is said to be integrable if $\nabla^2 = 0$.
One associates a unique connection $\nabla_{rel}$ to $\nabla$.
$\nabla_{rel}$ endows  $V$  with a natural $\sD$-module structure.

Let the rank of $V$ be $r$.
A vector $v$ in $V$ is called cyclic vector if $\{v, Dv,\ldots, D^{r-1} v\}$ is a basis
of $V$ for  $D:= d/dz$.
It is well-known that $V$ has a cyclic basis (cf. \cite{de}, Chap.2 and \cite{ma}, Chap.3).

A cyclic basis satisfies a polynomial in $D$:
\begin{equation}
a v := (D^n + a_{n-1} D^{n-1} + \ldots + a_0) v = 0,
\end{equation}
for some $a_0, \ldots, a_{n-1}$ in $K((z))$.

When $v$ is a cyclic basis satisfying the polynomial $a$, 
\begin{equation}
V \cong \sD/\sD_a.
\end{equation}

\begin{definition}
For a nonzero differential operator
$$
a := D^{n} + a_{n-1} D^{n-1} + \ldots + a_1 D + a_0,
$$
the Newton polygon of $a$ is the convex hull of the quadrants
$
\{ u\le k, v\ge k-v(a_k) | v(a_k) < \infty \}
$
in $\R^2$. We call the slopes of the non-vertical edges of the Newton polygon of $a$
in the right half plane, the slopes of $a$. If $V \cong \sD/\sD a$, we call 
the Newton polygon of $a$ the Newton polygon of $V$ and the slopes of $a$ the slopes of
$V$. 
\end{definition}

The definitions of the Newton polygon and the slopes for $V$ are well-defined,
since they are independent of the choice of a cyclic vector in $V$(cf. Chap.3 in \cite{ma}).

Let $0 \le p_1 < p_2 <\ldots < p_s$ be the slopes of $V = \sD/ \sD a$.
Then there exists $b, c$ in $\sD$ such that $a=bc$, where the slopes of
$b$ (resp. $c$) are $p_1, \ldots, p_r$ (resp. $p_{r+1},\ldots, p_s$) for a 
integer between $1, s$. It yields a short exact sequence of $\sD$-modules:
$$
0 \to \frac{\sD}{\sD b} \to \frac{\sD}{ \sD a} \to \frac{\sD}{ \sD c} \to 0.
$$
Moreover, the above short exact sequence splits in a unique manner.
\begin{lemma}[\cite{ma}, 3.1.4]
\label{extl}
Suppose $V$ and $W$ have different slopes, then
$$
\Ext^i_\sD(V,W) = \Ext^i_\sD(V,W) = 0,
$$ 
for $i=0,1$.
\end{lemma}

A direct consequence of the previous lemma is that we have the unique slope decomposition
of $V$:
$$
V = \bigoplus_i V_i,
$$
where $V_i$ has only a single slope.

\section{Slope decomposition of an absolute connection}
In this section, we introduce the lemma 6.4.3 of \cite{bbe}, which plays 
the bridge from the relative to the absolute case. We still keep the
same conventions and notations as before and any connection in
this section is assumed integrable.

\begin{lemma}[\cite{bbe}]
\label{relslope}
Let $(V,\nabla)$ be an absolute connection on $K((z))$.
If 
$$
(V,\nabla_{rel}) = (V_1, \nabla_{{rel},1}) \oplus (V_2, \nabla_{{rel}, 2})
$$
and 
$$
\Hom_\sD (V_1, V_2) = 0,
$$
then the absolute connection $\nabla$ stabilizes both of $V_1, V_2$
(i.e. $\nabla(V_i) \subset V_i \otimes \Omega^1_{K((z))/k}$).
Thus 
$$
(V,\nabla) = (V_1,\nabla_1)\oplus (V_2,\nabla_2),
$$
where $\nabla_i$ is an absolute connection on $V_i$ whose associated 
relative connection
is $\nabla_{rel}$.
\end{lemma}

\begin{proof}
(compare with \cite{bbe}, 6.4.3)
$\nabla$ stabilizes $V_i$ if and only if the $K((z))$-linear map 
\begin{equation}
\begin{split}
\eta_{ij} : V_i &\to V_j \otimes\Omega_{K//k} \\
	v  & \mapsto \nabla (v) \quad \mod{V_i\otimes \Omega_{K/k}}
\end{split}
\end{equation}
is zero for $i\ne j$.

$K((z))\otimes \Omega_{K/k}$ has a natural $\sD$-module structure
$$
\frac{d}{dz} : f\otimes\omega \mapsto \frac{df}{dz} \otimes \omega
$$
as a subspace of $\Omega_{K((z))/k}$.

Hence $V_j \otimes\Omega_{K/k}$ is equipped with a natural $\sD$-module
structure which agrees with the action of $\nabla_j$ on $V_j\otimes\Omega_{K((z))/k}$.
The integrability of $\nabla$ yields 
$$
\eta_{ij}\circ \nabla_{i, d/dz} = \nabla_{j, d/dz} \circ \eta_{ij}.
$$
Therefore we obtain 
$$
\eta_{ij} \in \Hom_\sD (V_i, V_j\otimes \Omega_{K/k}) = 0.
$$
\end{proof}

Together with lemma \ref{relslope}, the previous lemma yields the decomposition 
theorem for absolute connections as in \cite{ma1}:
\begin{theorem}
Let $(V,\nabla)$ be an absolute connection on $K((z))$. 
Then there exists a unique decomposition of $V$
$$
(V,\nabla) = \bigoplus_i (V_i, \nabla_i),
$$
where $V_i$ has a unique slope $p_i$ as a $\sD$-module.
\end{theorem}


\section[Turritin-Levelt decomposition]{Turritin-Levelt decomposition of an absolute connection}

Recall the well-known decomposition theorem of a connection due to Turritin(\cite{tu}), 
Levelt(\cite{le}).
It is a direct sum decomposition of a connection on $K((z))$ with each direct
summand of the form:
$$
(L\otimes M, \nabla_L\otimes \nabla_M),
$$
where

1. $(L,\nabla_L)$ is a connection of rank $1$ over $K((z))$, and\\
\indent 2. $(M,\nabla_M)$ is regular singular at $(z)$.\\

Let $(L_1\otimes M_1, \nabla_{L_1} \otimes \nabla_{M_1})$ and  
$(L_2\otimes M_2, \nabla_{L_2} \otimes \nabla_{M_2})$ be two direct summands of Turritin-Levelt
decomposition such that $L_1, L_2$ are not isomorphic to each other.

Then it follows
$$
\Ext^i_\sD (L_1 \otimes M_1, L_2\otimes M_2) = \Ext^i_\sD (M_1, L^{-1}_1\otimes L_2\otimes M_2)
=0
$$
for $i=0,1$.

Recall the main theorem of \cite{le}:
\begin{theorem}[Levelt \cite{le}]
Let $(V,\nabla)$ be a connection over $K((z))$ relative to $K$.
Then $(V,\nabla)$ has a unique Turritin-Levelt decomposition after finite
extension $\hat{K}$ of $K$ and a ramification $z=t^n$.
(i.e. $(V\otimes \hat{K}((t)),\nabla\otimes d)$ has Turritin-Levelt decomposition.)
\end{theorem}
This is the relative version of the theorem, which we now extend it 
to the absolute case.

Returning to the absolute case, let 
$(V,\nabla)$ be a absolute integrable connection over $K((z))$.
Applying Lemma \ref{extl}, \ref{relslope}, we obtain the following from the above 
theorem.
\begin{theorem}
Let $(V,\nabla)$ be an absolute integrable connection over $K((z))$. 
Then the Turritin-Levelt  decomposition of its relative connection obtained
after finite extension $\hat{K}$ of $K$ and finite ramification $z=t^n$, is stabilized
by $\nabla$.
Hence one has a unique Turritin-Levelt decomposition of an absolute connection
$$
(V,\nabla) = \bigoplus_i (L_i,\nabla_{L_i}) \otimes (M_i, \nabla_{M_i}),
$$
where $L_i \ne L_j$ for $i\ne j$.
\end{theorem}


\chapter[period of an irregular connection]{Period of an irregular connection over an elliptic curve: $\nabla=d+dy$}

In this chapter, we will calculate the period determinant of the
connection $\nabla=d + dy$ over
the Legendre elliptic curve: $y^2 = x(x-1)(x-\lambda)$. $\nabla$ 
has an irregular singularity at
$\infty$. The period for irregular connections is defined to be a 
pairing of de Rham cohomology and
a newly defined homology in \cite{be}. It is shown to be perfect. 
%It appears as a certain type of integration. (cf. \cite{te})



For a regular singular connection, the period is defined 
as a pairing of de Rham cohomology and the homology 
relative to the singularities valued in its dual local system, via a 
singular integration.
From the view  of the theory of characteristic classes, it was discussed 
profoundly by T.~Saito and T.~Terasoma in \cite{st}. 
Especially on $\P^1$, it is described  by Terasoma explicitly as a  product 
formula written in terms of the Gamma factors and the tame symbols in \cite{te1}. 

As for the irregular connections,  very few cases are known. 
On $\A^1- \{ \lambda_1,\ldots,\lambda_n \}$, when the connection
is of the form $d+\sum \frac{s_i}{x-\lambda_i} +df$ for a polynomial $f$ and positive real numbers $s_i$ , the period determinant 
%for its small extension 
%at $\Sigma= \{ \lambda_1, \ldots,\lambda_n\}$ 
is calculated via approximation by series of rank $1$ regular singular period
(\cite{te2}). 
%he  integrates the de Rham forms with compact support in 
%$\P^1 -\{ \text{some points} \}$
%calculating the period for a certain irregular connection of rank $1$
%over $\P^1$ with some additional conditions.

In our case, the period determinant involves some elliptic integrals, 
which are not easily calculated.  
Let $\pi$ be the second projection of $U$ onto $\A^1$, $(x,y)\mapsto y$.
%The main idea is 
% to approximate the period integrations by the period pairing of regular connections on $\P^1$.
%to calculate the period of its direct image connection under 
%the second projection $\pi:U\to \A^1,  (x,y)\mapsto y$ in order 
%to make it a higher rank connection on $\P^1$. 
Then the connection we consider is pulled-back from $\P^1$ via $\pi^*$.
Therefore we apply the projection formula to reduce  the period computation
on the elliptic curve to a computation on $\P^1$ where we then mostly apply
Terasoma's idea.

%Then at least those integrations 
%are on $\P^1$, so that
%it can be approximated by a sequence of period integrations of regular
% singular connections on $\P^1$. 
% Those periods  of regular connections  
%still do not  satisfy some additional conditions for applying  the product 
%formula: specially 
%those on the eigenvalues of the residues at the auxiliary regular singular 
%points obtained while projecting down.

Let $D\subset \P^1$ be those 
ramification points and $d_D$ be the standard exterior differentiation 
of $\sO(-D)$ in $\sO$. 
If necessary,
tensored with $d_D$,  $\pi_*(\nabla)$ will be approximated by  connections 
whose periods are computable using the product formula. 
Returning to the original connection, we take another connection 
$\nabla_\Sigma= \nabla \otimes d_\Sigma$, where $\Sigma := \pi^{-1} D$.
% is the divisor of ramification of
 %$\pi$ and $d_D$ is the standard exterior differentiation  on $\sO(-D)$. 
%Its period  integral is made via an approximation by some periods 
%of a series of  regular singular connections. 
This new connection on $U$ produces the 
same periods as $d_D \otimes \pi_*(\nabla)$ since
they have the same relative homology groups as well as
the same de Rham cohomology groups.
Finally, we will compare the periods of $\nabla$ with that of $\nabla_\Sigma$ 
through the long exact sequence of de Rham cohomology and of the homology to produce
exact value of the  period determinant of $\nabla$ on $U$. 

\section[De Rham cohomology and irregular homology]{De Rham cohomology and homology for irregular singular connection}

In this section, we will calculate the de Rham cohomology and the homology defined in \cite{be}, 
for  the irregular connection $d + dy$ on $X$ a complete Legendre curve in $\P^2$.
This has an irregular singular point of irregularity
$3$ at infinity and its monodromy is trivial. 

\subsection{de Rham cohomology}
In general, for an integrable connection $\nabla: E \to E\otimes \Omega^1_U$ on a nonsingular variety $U$ over a field of characteristic $0$, the de Rham cohomology of $\nabla$, 
$H^i_{dR}(U, \nabla)$ is defined as the hypercohomology of the associated de Rham complex. 

In our case, since $U=X-\{\infty\}$ is an affine curve and $E = \sO$, $H^i_{dR} (U, \nabla)= 0$ 
for $i > 1$ and 
\begin{equation*}
\begin{split}
H^1_{dR}(U, \nabla) :=& \H^1 (U, (\Omega^*_U, \nabla) )= 
\frac{\Gamma(U, \Omega^1)}{\nabla\Gamma(U, \sO)} \\
=& \frac{\Gamma(X, \Omega^1(*\infty)}{\nabla\Gamma(X, \sO(*\infty))}.
\end{split}
\end{equation*}

%Let us take an increasing filtration $W^{\nabla,\infty}_\bullet$ on the meromorphic extension of de Rham complex of $\nabla$ on $X$. We write it $W_\bullet$ in short if there is no danger of confusion.

%Define $W_n ( \sO):= \sO(n\cdot\infty)$ and $W_n (\Omega) := \Omega((n+4)\cdot\infty)$. 
%Let $t$ be the local parameter at $\infty$. Then we have the following short exact sequence of
%complexes:
%\begin{split}
%0 \to W_n  \to W_{n+1} \to gr_{n+1} \to 0,\\
%\begin{equation}
%\begin{CD}
%\sO(n\cdot\infty) @>\nabla>> \Omega((4+n)\cdot\infty) \\
% @VVV				@VVV \\
%\sO((n+1)\cdot\infty) @>\nabla>> \Omega((5+n) \cdot\infty)\\
% @VVV 				@VVV\\
% k <\frac1{t^{n+1}}> @>\nabla \mod{\frac1{t^{n+4}}}>> k<\frac1{t^{n+5}}>
%\end{CD}
%\end{split}
%\end{equation}

%This equals the short exact sequence 
%$$0 \to W_n \to W_{n+1} \to gr_{n+1} \to 0.$$
% which  
%yields the long exact sequence of cohomologies:
%\begin{equation}
%\begin{split}
%0 \to \Gamma(\sO(n\infty)) \stackrel{\alpha}{\to} \Gamma(\sO((n+1)\infty) \to k 
%\to W_n H^1_{dR}\\ \to W_{n+1} H^1_{dR} 
%\to H^1(gr_{n+1}) = 0.
%\end{split}
%\end{equation}

%Since $gr_{n+1}$ is quasi-isomorphic to $0$, we have
%$$
%H^1_{dR}(U, \nabla) = \H^1(X, (W_n, \nabla)),
%$$
%for any $n$.

We have a natural inclusion
\begin{equation}
\begin{CD}
\sO(n\cdot\infty) @>{\nabla}>> \Omega((4+n)\cdot\infty) \\
 @V{i}VV	@V{i}VV \\
\sO(*\infty) @>{\nabla}>> \Omega(*\infty),
\end{CD}
\end{equation}
whose cokernel is
$$
\bigoplus_{l>n} k<\frac1{t^l}> \stackrel{\nabla \mod{\frac1{t^{n+4}}}}{\longrightarrow}
\bigoplus_{l>n+4} k<\frac{dt}{t^l}>.
$$
Since $\nabla \mod{\frac1{t^{n+4}}}$ is an isomorphism of $k$-vector space,
$\operatorname{Coker} i$ is quasi-isomorphic to $0$. 

%Since $gr_{n+1}$ is quasi-isomorphic to $0$, 
Thus we have obtained the de Rham cohomology of $\nabla$ in another way
$$
H^1_{dR}(U, \nabla) = \H^1(X, \sO(n\cdot\infty)\stackrel{\nabla}{\to} \Omega((4+n)\cdot\infty)),
$$
for any $n$.


Taking the  filtration on the complex 
$$
\sO(n\cdot\infty) \stackrel{\nabla}{\to} \Omega((n+4)\cdot\infty)
$$ 
by its degree, we have
the following short exact sequence:
\begin{equation}
\begin{CD}
 0 @>>> \Omega((4+n)\infty)\\
 @VVV 	 @VVV \\
\sO(n\infty) @>{\nabla}>> \Omega((4+n)\infty) \\
 @VVV  @VVV\\
\sO(n\infty) @>>> 0
\end{CD}
\end{equation}
This yields again the long exact sequence of cohomologies:
\begin{equation}
\begin{split}
\ldots \to  \Gamma(X, \sO(n\cdot\infty) ) & \stackrel{\nabla}{\to} 
\Gamma(X, \Omega((4+n)\cdot \infty))  \stackrel{\alpha}{\to} 
 H_{dR}^1(U,\nabla) \\
& \to H^1(X,\sO(n\cdot \infty)) \stackrel{\beta}{\to} 
H^1(X, \Omega((n+4)\cdot \infty))\to \ldots
\end{split}
\end{equation}

We want to know when the map $\alpha$ is surjective. $\alpha$ is surjective
if and only if $\beta$ is injective. 
When $n=0$, since $h^1(\sO(0\cdot\infty)) = h^0(\sO)= 1$ and 
$h^1(\Omega(4\cdot\infty)) = h^0 (\sO(- 4\cdot\infty)) =0$, $\beta$ can not 
be injective. For $n>0$, $h^1(\sO(n\cdot \infty) = h^0( - n\cdot \infty) =0$, so
$\beta$ is always injective.

It follows that 
$$
\Gamma(X,\Omega( (n+4) \cdot\infty)) \stackrel{\alpha}{\to} H^1_{dR}(U,\nabla)
$$
is surjective for $n>0$.

%Since $\operatorname{Coker} (\alpha) = k$ if and only if $n>0$, 
%$W_1 H^1_{dR}(U,\nabla) = H^1_{dR}(U,\nabla)$. Therefore we have the
%theorem.
%This is compatible with $\nabla$, 
%so it yields again an increasing filtration on $H^1_{dR} (U,\nabla)$.
%As the de Rham cohomology is a finite dimensional $k$-vector space, 
%$W_m H^1_{dR}(U, \nabla) = H^1_{dR}(U,\nabla)$ for $m \gg 0$. 
%So  it is generated by $\Gamma(X, \Omega(m D))$ for sufficiently large $m$. 
%Knowing that $\nabla$ has no meromorphic solution, we have 
%$$
%\rank W_m H^1_{dR} (U,\nabla) 
%=  \rank \Gamma(X, \Omega((m+4)\cdot\infty)) - \rank \Gamma(X, \sO(m \cdot\infty)).
%$$ 
%Using the Riemann-Roch formula, one has on an elliptic curve $X$, 
%$$
%\rank\Gamma( X,\sO(m\cdot\infty)) = \rank \Gamma(X,\Omega(m\cdot\infty))
% =  m,\quad \text{ for $m > 0$}.
%$$
 
%Thus we have $\rank H^1_{dR}(U,\nabla) = W_m H^1_{dR}(U,\nabla) = 4$ for $m \gg 0$. 
%
%As $\rank \Gamma(X,\sO) = 1$ exceptionally, $\rank W_0 H^1_{dR}(U,\nabla) = 3$. 
%For $m> 0$, $\rank W_m H^1_{dR}(U,\nabla)$ is always $4$. 
%Especially for $m=1$,
% when $k=1$ the filtration is already maximal in $H^1_{dR}(U,\nabla)$.
%Hence, we have the following result:
As a consequence, when $n=1$, we have
\begin{proposition}
$$
H^1_{dR}(U, \nabla) = %W_1 H^1_{dR}(U,\nabla) = 
\displaystyle\frac{\Gamma(X, \Omega_X
(5\cdot \infty))}{\nabla\Gamma(X,\sO_X(\infty))}.
$$

Therefore, $H^1_{dR}(U,\nabla)$ is generated by $\Gamma(X, \Omega(5\cdot \infty))$ with the relation
\begin{equation*}
\begin{split}
\nabla \Gamma(X,\sO(\infty)) =&\nabla\Gamma(X,\sO) \\
 =& \text{the rank $1$ vector space generated by $dy$}.
\end{split}
\end{equation*}
So 
$$
\frac{dx}y, x\frac{dx}y, dx, xdx.
$$
make a basis of $H^1_{dR}(U,\nabla)$.
\end{proposition}

\subsection{Homology}
We denote by $H^{irreg}_i$ the homology for irregular connections defined by Bloch and Esnault in \cite{be}.  
This homology group is defined for irregular singular connections over curves.
It is defined for $i=0,1, 2$. Note that $H^{irreg}_2(U,\nabla^*)$ vanishes in
this case for dimension reason (cf. \cite{be}).
This makes a perfect pairing with the de Rham cohomology of its dual connection.
It is composed of topological cycles and chains on which the solution 
of the connection decays rapidly.

Since the pairing of de Rham cohomology of a connection and the homology 
of the dual  connection 
is perfect, one concludes from the previous section that 
\begin{eqnarray}
\rank H^{irreg}_1(U,\nabla^*) =4, &\quad\text{and},\\
\rank H^{irreg}_0(U,\nabla^*) = 0.&
\end{eqnarray} 

We will describe the cycles generating $H^{irreg}_1(U,\nabla^*)$ here.

Let $\nabla^*$ be the dual of $\nabla$, $\Delta$ be a small disk around $\infty$ and 
$\Delta^*=\Delta - \{\infty\}$.  
%Let us denote the homology for irregular connections by $H_i^{irreg}$. 

Then
Theorem 1.1 in \cite{be} tells us that there is a homology long exact sequence of the form :
\begin{equation}
\begin{split}
0 \longrightarrow H_1(U, \C) \stackrel{1\otimes \exp y}{\longrightarrow} 
	H_1^{irreg}(U, \nabla^*) \longrightarrow H_1^{irreg}(\Delta^*, \partial \Delta,  \nabla^*) \\
	\stackrel{\delta}{\longrightarrow} H_0 (U,\C) \longrightarrow H^{irreg}_0 (U,\nabla^*) \to 0.
\end{split}
\end{equation}

In the above sequence, $H_1(U,\C) = H_1 (X,\C)$ which is generated by $\gamma_1, \gamma_2$.
Therefore we have still two other cycles that map to the kernel of 
$$
\delta: H_1^{irreg}( \Delta^*,\partial\Delta,\nabla^*) \to H_0 (U,\C)= \C<\exp y>.
$$ 

Let us describe the cycles in $H^{irreg}_1(\Delta^*,\partial\Delta, \nabla^*)$.
Fix a local parameter $t$ at $\infty$, so that we can write the solution of $\nabla^*$ locally
$\exp (-t^{3})$. Then on a small disk around $\infty$, we have sectors along which the solution decays rapidly : 
$$
\frac{2(i-1)\pi}3 - \frac{\pi}6 < \arg t < \frac{2(i-1)\pi}3 +\frac{\pi}6, \quad\quad i=1,2,3.
$$
We call the above the \textit{ $i$-th rapid decay sector}.

Let $\eta_i$ be a chain from a fixed point $p$ at the boundary of $\Delta$ to $\infty$ 
along the $i$-th rapid decay sector. Since there are three such sectors,  
$\eta_i\otimes \exp y$, for $i=1,2,3$ generate 
$H^{irreg}_1(\Delta^*,\partial \Delta, \nabla^*)$. 
Then the map $\delta$ is the augmentation 
$$
\sum_i a_i \eta_i \otimes \exp y \mapsto \sum_i a_i <\exp y>
$$
and the kernel is generated by $(\eta_1 - \eta_2)\otimes \exp y$ and 
$(\eta_2 - \eta_3)\otimes\exp y$. 


\subsection{Period matrix}
We can describe now the period matrix of the connection. 
From now on, $k$ is a fixed subfield of $\C$ containing $\lambda$. (i.e. it is the field of
definition of the elliptic curve.)

We have seen 
$\Gamma(X, \Omega(5\infty))$ generates $H^1_{dR}(U,\nabla)$
and
$$
\frac{dx}y, x \frac{dx}y, dx, x dx
$$
make a basis of the $k$-vector space $H^1_{dR}(U,\nabla)$. 
We will denote  the generators by $\omega_1, \omega_2, \omega_3$ and $\omega_4$,
respectively.
%The  canonical $k$-structure in $H^1_{dR}(U,\nabla)$  given by the basis 
%$\{\omega_i\}_{i=1,2,3,4}$  yields  a unique $\C$-structure in natural way.\\

In the other hand, we have the cycles in $H^{irreg}_1 (U, \nabla^*)$ are
\begin{equation*}
%\begin{split}
\xi_i  = \begin{cases}
	\gamma_i \otimes \exp y & \text{ for $i=1,2,$ } \\
 (\eta_{i-2} - \eta_{i-1})\otimes \exp y & \text{ for $i=3,4$.}
	\end{cases}
%\end{split}
\end{equation*} 

If we denote $\eta_{i-2}-\eta_{i-1}$ by $\gamma_i$ for $i=3,4$,
the pairing of $ \omega_i$ and $\xi_j$ is  the integral
$$
< \xi_j, \omega_i > = \int_{\gamma_j} \exp y \cdot \omega_i.
$$

The integrations are the pairing of $H^1_{dR}(U,\nabla)\otimes\C$ and 
$H^{irreg}_1(U, \nabla)$, which
produces complex numbers. 


We are interested in the determinant of the matrix of the period integrals:
$$
per(U,\nabla) := \det \left(  <\xi_j,\omega_i>\right)_{i,j=1,2,3,4}.
$$

It is not a well-defined number in $\C$ since we can change the basis of $H^1_{dR}(U,\nabla)$ by matrix in 
$\GL(H^1_{dR}(U,\nabla))$ and
we don't have any canonical choice of a basis in $H^{irreg}_1(U,\nabla^*)$.
Fixing a local solution of $\nabla^*$  yields, in this case since the monodromy
is trivial,  its global solution. For a chosen nonzero solution $f$ of $\nabla^*$, one has 
a  $\Q$-vector space generated by $\gamma_{i}\otimes f$ for $i=1,2,3,4$
% and $(\eta_{i-2} - \eta_{i-2})_{i=3,4}\otimes f$ 
gives  a $\Q$-structure on $H_1^{irreg}(U,\nabla^*)$. But we don't know if $c\exp y$ 
for a $c\in \C$ is more canonical than $\exp y$.
Thus we have a well-defined class of numbers in $\C^*/k^*$ depending on the choice of the solution of $\nabla^*$. 

Throughout this article, $\exp y$ is chosen to give a $\Q$-structure on
$H^{irreg}_1(U,\nabla^*)$.


\section{Direct image connection}


We keep the same convention as in the previous section.

% for $\nabla$, $X$ the legendre curve, etc in this section.
We want to apply the product formula of Terasoma, which will be briefly reviewed 
in the next section.  
Unfortunately, it cannot be applied directly because
it is done only for regular singular connections on $\P^1$ with some extra conditions 
(cf. \cite{te1}).
So instead of calculating the determinant of the period matrix directly, we take its direct image under the projection map $\pi:X \to \P^1$, $(x,y)\mapsto y$ and then, 
we will approximate the period integral of the irregular connection $\pi_*(\nabla)$ by  that of regular connections.

\subsection{Splitting and the direct image connection of the ramification}
We still denote by $\nabla$ the connection on $U = \Spec k[x,y]/ (y^2 - x(x-1)(x-\lambda)$:
$$
\nabla : \sO_U \to \sO_U \otimes\Omega^1,\quad\text{with}~~~\nabla(1)= dy.
$$
Let  $\pi$ be the 2nd projection
of $\A^2=\{(x,y)\}$ to $\A^1$, $(\sO_U,\nabla)=\pi^*((\sO_{\A^1},d+dy))$.
One should be careful that here $\A^1$ is parametrized by $y$.
%Note that $\nabla=d+dy$ can be defined on both $U, \A^1$. 
%So if there is no danger of confusion,
%we will keep the same notation.
As $\nabla = \pi^*(d+dy)$,
using the Leibniz formula for tensor product of connections and the projection formula, we have
\begin{equation}
\begin{split}
\pi_*(\sO_U, \nabla) =& \pi_*\pi^*(\sO_{\A^1},d+dy) \\
	=& (\sO_{\A^1}, d+dy) \otimes \pi_*(\sO_U,d).
\end{split}
\end{equation} 


Let $f(x)$ be the polynomial $x(x-1)(x-\lambda)$.
The trace map of $L := k(x,y)/(y^2-f(x))$ in $k(y)$ has a section, which yields a splitting of 
$\pi_* \sO_U$.
$$
0 \to \ker (\Tr_{L/k(y)}) \to \pi_*\sO_U \overset{\Tr_{L/k(y)}}{\to}  \sO_{\A^1} \to 0.
$$

$\pi_*\sO_U$ is generated by $(1, x, x^2)$ over $\sO_{\A^1}$. 
With respect to the basis $(1, x, x^2)$, one can
calculate the $\ker(\Tr)$ as well as the trace map:
\begin{equation}
\Tr(1) = 3,
\end{equation}
\begin{equation}
\Tr(x) = \Tr
\begin{pmatrix}
0 & 0 & y^2 \\
1 &  0 & -\lambda \\
0 & 1 & \lambda+1
\end{pmatrix}
= \lambda + 1,
\end{equation}
and
\begin{equation}
\Tr(x^2) = \Tr
\begin{pmatrix}
0 & y^2 &  (\lambda+1)y^2 \\
0 & - \lambda & y^2- \lambda(\lambda+1) \\
1 & \lambda+1 & (\lambda+1)^2-1
\end {pmatrix} = \lambda^2 + 1.
\end{equation}
Hence for $a, b, c$ in $k(y)$,
\begin{equation}
\Tr ( a + b x + c x^2) = 3 a + (\lambda+1) b + (\lambda^2+1) c,
\end{equation}
and this is in $\ker(\Tr)$ if and only if 
$$a = -\frac{\lambda+1}3 b - \frac{\lambda^2+1}{3} c.$$
Therefore $\ker(\Tr)$ is generated by the followings:
$$
v_1:= -\frac{\lambda+1}3 + x, v_2 :=  -\frac{\lambda^2+1}{3} + x^2.
$$

Since $d$ and $\Tr$ commute, the splitting is stable under $d$. Thus,
$$
\pi_*(\sO_U,d) = (\sO_{\A^1},d)\oplus (\ker (\Tr), \nabla_1),
$$
as connections. %We denote the rank $2$ part in the above by $(\sV, \nabla_\sV)$.
$\nabla_1$ has regular singularity at the roots of $(y^2-f(x_1))(y^2-f(x_2))$ where $x_1, x_2$ are
the roots of $f'(x)= 3x^2 - 2(\lambda+1)+\lambda$.

$f'(x)$ has two distinct roots $x_1 \ne x_2$ if and only if  
$$
\lambda^2 -\lambda +1 \ne 0.
$$


Suppose $\lambda^2 -\lambda+1 \ne 0$. 
Then $ \nabla_1$ has 4 regular singularities $\pm\sqrt{f(x_1)}, \pm\sqrt{f(x_2)}$.
The connection matrix of $\nabla_1$ in the basis $v_1, v_2$ is obtained from the following calculation:
\begin{equation}
\begin{split}
\nabla_1 (v_1) = & d (v_1) = dx \\
	= & \frac{2y}{3x^2 - 2(\lambda+1) x + \lambda} dy \\
	= & \frac{(y^2-f(x_1))(y^2-f(x_2))}{3(x-x_1)(x-x_2)} 
			\frac{2y}{(y^2 - f(x_1))(y^2-f(x_2))} dy \\
	=& \frac13(f_0(y) + f_1(y) x + f_2(y)x^2) \frac{2y}{(y^2 - f(x_1))(y^2-f(x_2))} dy,
\end{split} 
\end{equation}
where $f_0,f_1,f_2$ are polynomials in $y$.

Let $N_i$ denote the $i$-th Newton polynomial of $x_1$, $x_2$ (i.e. $x_1^i + x_2^i$) 
and $M$ denote $x_1 x_2$. Then
\begin{equation}
\begin{split}
&\frac{(y^2-f(x_1))(y^2-f(x_2))}{(x-x_1)(x-x_2)} \\
	 =&	(x^2+ (x_1 - (\lambda+1))x + x_1^2-(\lambda+1)x_1 + \lambda) \\
	&~~~~~~~~~~\cdot(x^2 + (x_2 - (\lambda+1))x + x^2_2 - (\lambda+1)x_2 + \lambda)\\
	=& x^4 + (x_1 + x_2 -2(\lambda+1)) x^3  \\
		&+ (\lambda^2 - 2 (x_1 + x_2-2) \lambda + x_1^2 + x_2^2 + x_1x_2 - 2x_1x_2 + 1)x^2\\
		& + ((x_1 + x_2 -2) \lambda^2 - (x_1^2 + x_2^2 + 2 x_1 x_2 + 3(x_1+x_2))\lambda \\
		& +	x_1 x_2(x_1+x_2 - (x_1^2 +x_2^2) - 2 x_1 x_2 + x_1+ x_2 \\
		& +  ( (x_1 x_2 - x_1 - x_2 + 1)\lambda^2 + \\
		&~~~~(-x_1 x_2 ( x_1+x_2) + x_1^2 + x_2^2 + 
			2 x_1 x_2 - x_1 - x_2)\lambda + \\
			& \hspace{12pt} x_1^2 x_2^2 - x_1 x_2(x_1+x_2) + x_1 x_2 \\
	=&   ((-N_1 + 1)\lambda -N_1 + N_2 + M)  x^2 \\
	   &+ 	((N_1 - 1)\lambda^2 + (2N_1 -N_2 -2M - 1)\lambda + (M + 1)N_1 -N_2 -2M + y^2)
	 x \\
	  &+ (-N_1 + M + 1)\lambda^2 + ((-M - 1)N_1 + N_2 + 2M - y^2)\lambda + (-M + y^2)N_1 \\
	&+ (M^2 + M - y^2)
\end{split}
\end{equation}

Since,
\begin{equation}
N_1 = \displaystyle{\frac{2(\lambda+1)}3},
\end{equation}
\begin{equation}
M  =  \displaystyle{\frac\lambda3},
\end{equation}
{and}
\begin{equation}
N_2  =\displaystyle{\frac{2(2\lambda^2+ \lambda +2)}9},
\end{equation}
we obtain 
\begin{equation}\label{f0}
f_0(y) = - \frac19 \lambda(\lambda -1)^2 - \frac{\lambda+1}{3} y^2,
\end{equation}

\begin{equation}\label{f1}
f_1(y) =  \frac19(\lambda+1)(2\lambda^2 - 3\lambda +2) + y^2
\end{equation}
and
\begin{equation}\label{f2}
f_2(y) = - \frac29 (\lambda^2 - \lambda +1).
\end{equation}

For simplicity, we will write
\begin{equation}
\omega= \frac{2y}{3(y^2 - f(x_1))(y^2 - f(x_2))} dy \in \Gamma(\A^1, \Omega( \log D)),
\end{equation}
for $D = \{ \pm \sqrt{f(x_1)}, \pm\sqrt{f(x_2)}  \}$

Then $\nabla_1 (v_1) = (f_1 v_1+ f_2 v_2 )\cdot \omega$.

Using the Leibniz rule, 
\begin{equation}
\begin{split}
\nabla_1 (v_2) & = d (- \frac{\lambda^2 + 1}{3} + x^2) \\
	&= dx^2 \\
	& = 2 x (f_0 + f_1 x + f_2 x^2 ) \omega \\
	& = (2(f_0 - f_2 \lambda) v_1 + 2(f_1 + (\lambda+1)f_2)v_2)\cdot \omega.
\end{split}
\end{equation}
Hence the connection matrix of $\ker \Tr$ is
\begin{equation}
\begin{split}
&\begin{pmatrix}
f_1 & 2(f_0 - \lambda f_2)\\
f_2 & 2 (f_1 + (\lambda+1) f_2)
\end{pmatrix} \cdot \omega \\
= &
\frac23 \frac{ydy}{(y^2 - f(x_1))(y^2 - f(x_2))} \\
& \cdot
\begin{pmatrix}
\frac19(\lambda +1)(2\lambda^2 - 3\lambda +2) + y^2 & \frac29 \lambda(\lambda+1) - 
		\frac23(\lambda+1)y^2 \\
-\frac29(\lambda^2 -\lambda +1) & -\frac29 \lambda(\lambda+1) + 2y^2
\end{pmatrix} dy.
\end{split}
\end{equation}
with respect to $v_1, v_2$. 

The above connection matrix has only simple poles at $y = \pm\sqrt{f(x_1)}$, 
$\pm\sqrt{f(x_2)}$.\\

Now we will calculate the same for the case $\lambda^2 - \lambda +1=0$.
In this case, $x_1 = x_2$  and $\nabla_1$ has only two regular singularities $\pm \sqrt{f(x_1)}$.
Via a similar calculation, we obtain its connection matrix. %Let $x_3$ be the root of $f(x)-f(x_1)$ different from $x_1$. 
Note that $f(x)-f(x_1)$ has a triple root $\frac{\lambda+1}3$. 
From the coefficient of $f(x)-f(x_1)$, we have
\begin{equation}
x_1 = \frac{\lambda+1}3.
\end{equation}

Then
\begin{equation}
\begin{split}
\nabla_1(v_1) =& dx =  \frac{2y}{3x^2 - 2(\lambda+1)x+\lambda} dy \\
	=& \frac{y^2-f(x_1)}{3(x-x_1)^2} \frac{2y}{y^2-f(x_1)} dy \\
	=& \frac{x-x_1}3 \frac{2y}{y^2 - f(x_1)} dy \\
	=& (x-\frac{\lambda+1}3) \frac{2y}{3(y^2 -f(x_1))} dy\\
	=& \frac{2ydy}{3(y^2-f(x_1))} v_1.
\end{split}
\end{equation}

If we denote $\displaystyle{\frac{2ydy}{3(y^2-f(x_1))}}$  by $\omega'$, we have
\begin{equation}
\nabla_1(v_1) = \omega' \otimes v_1.
\end{equation}

Similarly,
\begin{equation}
\begin{split}
\nabla(v_2) =& 2x dx \\
	=& 2(x^2 - \frac{\lambda+1}3 x) \frac{2y}{3(y^2 -f(x_1)} dy\\
	=&  \omega'  \otimes (2 v_2 - \frac{2(\lambda+1)}3 v_1) 
\end{split}
\end{equation}

Thus in this case, the connection matrix is
\begin{equation}
\begin{split}
\begin{pmatrix}
1 & -\frac23 (\lambda+1) \\
0 & 2 
\end{pmatrix} \omega' 
= \begin{pmatrix}
1 & -\frac23 (\lambda+1) \\
0 & 2 \end{pmatrix}
\frac{2ydy}{3(y^2-f(x_1))}.
\end{split}
\end{equation}

%\input{dy2.tex}
%\chapter{Direct image connection : Part 2}


\subsection{Residues}
For the generic $\lambda$ (i.e. $\lambda^2 - \lambda +1 \ne 0$), the rank 2 part of $\pi_*\nabla$ is 
$$d+ \begin{pmatrix} dy & 0 \\ 0 & dy \end{pmatrix} +
\begin{pmatrix} 
f_1 & 2(f_0 - f_2 \lambda) \\
f_2 & 2(f_1 + f_2(\lambda+1))
 \end{pmatrix}\omega,
$$ 
with $f_0, f_1, f_2$ defined in (\ref{f0}), (\ref{f1}) and (\ref{f2}).

%with respect to the basis: $v_1 := -\frac{\lambda+1}3 + x, v_2 :=  -\frac{\lambda^2+1}{3} + x^2$, where
%\begin{eqnarray*}
%f_0(y)  =& -\frac19 \lambda(\lambda-1)^2 - \frac{\lambda+1}3 y^2, \\
%f_1(y)  =& \frac19(\lambda+1)(2\lambda^2 -3\lambda+2) + y^2,\\
%f_2(y)  =& -\frac29(\lambda^2 - \lambda+1), ~~~\text{and}\\
%\omega =& \frac{2y dy}{3(y^2 -f(x_1))(y^2-f(x_2))}.
%\end{eqnarray*}
We will denote the rank 2 connection by $(\sV, \nabla_\sV)$ for a while.
%Thus, denoting the rank-$1$ connection on $\sO$ with $d+dy$ by $\sL$, and by  $M$
%the rank 2 connection with matrix 
This is the tensor product of $(\sO_{\A^1}, d + dy)$ and $(\ker  \Tr, \nabla_1)$.
$$
(\sV, \nabla_\sV) = (\sO_{\A^1}, d+dy) \otimes (\ker\Tr, d+A),
$$
for 
\begin{equation}
\begin{split}
A =& \begin{pmatrix} 
f_1 & 2(f_0 - \lambda f_2)  \\
f_2 & 2(f_1 + (\lambda+1) f_2)
 \end{pmatrix}\omega \\
=& \begin{pmatrix}
\frac19(\lambda+1)(2\lambda^2 - 3\lambda+2) + y^2 
	& \frac29 \lambda(\lambda^2+1) - \frac23(\lambda+1)y^2 \\
-\frac29 (\lambda^2 - \lambda+1) & -\frac29 \lambda(\lambda+1) + 2y^2
\end{pmatrix} 
\omega.
\end{split}
\end{equation}

%we have $V = \sL \otimes M$.
%
%$$
%A = $$
%{\bf\Huge This part should be fixed!!!!}
%Here, $\sL$ will be approximated by the sequence of regular connections $\sL_m$  of rank-$1$ 
%$$d+ \frac{m}{y+m}dy.$$
%to calculate the period.

As tensoring $(\sO_{\P^1},d+dy)$ does not change the residues at $\pm\sqrt{f(x_1)}$, $\pm\sqrt{f(x_2)}$,
we only have to calculate those for $(\sV, \nabla_\sV)$ 
%$\pm\sqrt{f(x_1)}, \pm\sqrt{f(x_2)}$ of
in order to check if Terasoma's formula is applicable.


Since $x_1, x_2$ are the roots of $f'(x)= 3x^2 - (\lambda+1) x + \lambda$, 
\begin{equation}
f(x_i) = - \frac29(\lambda^2 - \lambda +1) x_i + \frac19\lambda(\lambda+1),
\quad \text{for $i=1,2$}
\end{equation}
and  the connection matrix  of $(\sV,\nabla_\sV)$ has at most simple poles at $\sqrt{f(x_1)}$,
the residue  is 
\begin{equation}
\begin{split}
&\lim_{y \to \sqrt{f(x_i)}} (y- \sqrt{f(x_i)}) A = \\
&\begin{pmatrix}
\frac19(\lambda+1)(2\lambda^2 - 3\lambda+2) + f(x_i) 
	& \frac29 \lambda(\lambda^2+1) - \frac23(\lambda+1)f(x_i) \\
-\frac29 (\lambda^2 - \lambda+1) & -\frac29 \lambda(\lambda+1) + 2f(x_i)
\end{pmatrix} 
\Res_{\sqrt{f(x_i)}} (\omega).
\end{split}
\end{equation}

This is the product of 
\begin{equation*}
\begin{split}
\Res_{\sqrt{f(x_i)}} \omega = & \lim_{y \to \sqrt{f(x_i)}}  (y-\sqrt{f(x_i)})<\omega, \frac{d}{dy}> \\
	=& \frac{1}{3(f(x_i) - f(x_j))}
\end{split}
\end{equation*}
and
$$
A|_{y=\sqrt{f(x_i)}} = 
\frac{2(\lambda^2 - \lambda +1)}9 \begin{pmatrix}
\lambda +1 - x_i & \frac23((\lambda+1) x_i + \lambda) \\
-1 & -2 x_i
\end{pmatrix}.
$$

Whereas, 
$$
f(x_i) - f(x_j) = - \frac29 (x_i - x_j)(\lambda^2 - \lambda + 1).
$$

The trace and the determinant of the above are
\begin{equation*}
\Tr A|_{y=\sqrt{f(x_i)}} = \frac{2(\lambda^2-\lambda+1)}9 (\lambda+1 - 3 x_i),
\end{equation*}
\begin{equation*}
\begin{split}
& \det A|_{y=\sqrt{f(x_i)}} \\=& \frac{4(\lambda^2 - \lambda+1)^2}{81}
		(2x_i^2 - 2(\lambda+1)x_i + \frac23(\lambda+1)x_i + \frac23\lambda) \\
	=& \frac{4(\lambda^2 - \lambda+1)^2}{81}  \\
		&\cdot(2 (\frac23(\lambda +1)x_i - \frac\lambda3) - 2(\lambda+1)x_i + 		\frac23(\lambda+1)x_i + \frac23\lambda) \\
	=& 0.
\end{split}
\end{equation*}

Hence the characteristic polynomial of $\Res_{\sqrt{f(x_i)}}(\nabla_\sV)$ is
$$
P(\Res_{\sqrt{f(x_i)}}(\nabla_\sV),X) = X^2 + 
\frac{(3 x_i  -(\lambda+1))}{3(x_i - x_j)} X.
$$

One fixes a branch of the square root :
$$
x_i = \frac{\lambda+1 + \sqrt{\lambda^2 - \lambda +1}}{3},
$$
then automatically,
$$
x_j = \frac{\lambda+1 -\sqrt{\lambda^2 -\lambda+1}}3.
$$
Then one has the characteristic polynomial
$$
X^2 + \frac12 X.
$$

Therefore, the residue at each ramification point of $\pi$ has eigenvalue $0, -\frac12$.
This means Terasoma's result is not directly applicable in this case.\\

When $\lambda^2 - \lambda + 1=0$, the rank $2$ part of $\pi_*(\nabla)$ is
\begin{equation}
\begin{split}
d + I dy + 
\begin{pmatrix}
1 & - \frac23(\lambda+1) \\ 0  & 2 \end{pmatrix} \omega'.
\end{split}
\end{equation}

The residue matrix at $\sqrt{f(x_1)}$ is simply 
\begin{equation}
\begin{split}
& \begin{pmatrix}
1 & -\frac23(\lambda+1) \\
0 & 2 
\end{pmatrix} \Res_{\sqrt{f(x_1)}} \omega',
\end{split}
\end{equation}
for 
\begin{equation}
\begin{split}
\Res_{y = \sqrt{f(x_1)}} \omega'& = \lim_{y\to \sqrt{f(x_1)}} \frac{2y}{3(y^2 - f(x_1))} (y - \sqrt{f(x_1)}) \\
	&= \frac{2\sqrt{f(x_1)}}{3\cdot (2 \sqrt{f(x_1)})} = \frac13.
\end{split}
\end{equation}

Therefore there are two  eigenvalues of the residue of $\nabla_1$ at the ramification point $\sqrt{f(x_1)}$ : $1/3, 2/3$.
In the next section, we will see the product formula is not directly applied to approximate the period determinant of $\pi_*(\nabla)$.


%\input{dy3.tex}

\section{Terasoma's work}



Let us recall the main theorem in \cite{te1}. The theorem tells the exact value of determinant of
the period matrix of a connection on $\A^1$ with some extra conditions.

\subsection{De Rham cohomology with compact support and relative homology}
Let $D = \{\lambda_1, \lambda_2,\ldots,\lambda_n \}$ be $n$ distinct points in $\A^1$. 
Then a logarithmic connection with poles at $D \cup \{\infty\}$ 
$$
\nabla: \sV \longrightarrow \sV \otimes \Omega_{\P^1}(\log (D+\infty))
$$
on a trivial bundle $\sV$ over 
$\A^1$ can be written as 
$$
\nabla = d + \sum_{i=1}^n \frac{B^{(i)}}{x-\lambda_i}, 
$$
here $B^{(i)}$ in $\operatorname{End}(\C^n)$ is called  the residue matrix of $\nabla$ at $\lambda_i$
 and denoted by
$\Res_{\lambda_i} (\nabla)$.
The residue of at $\infty$ is  $B^{(\infty)} =\sum^n_{i=1} (-B^{(i)})$. 
%If $r= \rank \sV$, 
%then they are $(r,r)$-matrices.

Throughout this section, we will assume the condition below:
\begin{condition}\label{cond}
 Assume that no two eigenvalues of $B^{(i)}$ (resp. $-B^{(\infty)}$)  differ by an integer and
that  they have positive real part.
\end{condition}



Let us recall the notion of small extension and its properties from \cite{st} and \cite{te1}.
\begin{definition}
%Let $r$ be the rank of $\nabla$.
If no eigenvalue of residues is an integer less than or equal to $0$, $\nabla$
is called a small extension of $\nabla|_U$ along $D$.
\end{definition}

When $\nabla$ is small at $D$, then
the de Rham cohomology $H^1_{dR} (\A^1, \nabla)$ 
of the logarithmic de Rham complex is generated by 
$$
(\frac{1}{x-\lambda_i} - \frac1{x-\lambda_{i+1}}) dx \otimes w,\quad\text{for $i=1,2,\ldots,n-1$}
$$
where $w$ is a vector in $\sV$. Thus if $\rank \sV =r$, $\rank H^1_{dR}(\A^1, \nabla) = r(n-1)$.  
This cohomology is canonically isomorphic to the cohomology $H_{dR}^1 (\A^1,  j_! j^*(\nabla))$, 
where $j$ is the open embedding $j: \A^1-D \hookrightarrow \A^1$
% and $\sV^\nabla$ is the local system of 
%$\nabla$ on $\A^1-D$ 
(cf. \cite{st}, \cite{te1}).

One can define the corresponding homology theory to make a pairing with the de Rham cohomology described previously. 
This is the relative homology of the pair $(\A^1, D)$ valued in $\nabla^*$. 
It is defined as the relative homology of 
$(U , \cup_i D_i )$ valued in the dual
local system where $D_i$ is a small disk around $\lambda_i$, which is independent of the choice
of sufficiently small $D_i$.
It can be seen as Borel-Moore homology which has values in  the local system. 
A cycle in this homology is the topological cycle valued in its underlying local system.
So it can be written as linear sum of
$$
\gamma_\alpha \otimes f_\alpha,
$$
for $\gamma_\alpha$ a relative cycle of $H_i(\A^1,D, \Z)$ and $f_\alpha$ a branch of the
solution of $\nabla^*$. 
 We will simply denote the homology by $H_i(\A^1,D, \nabla^*)$. 
For dimension reason,
$H_i(\A^1,D,\nabla^*)$ vanishes for $i>1$.

Note that it depends only on the local system of $\nabla^*$.
The main theorem of \cite{te1} tells us that 
$H_1(\A^1,D, \nabla^* )$ and $H^1_{dR}(\A^1, \nabla )$ are in
perfect pairing via integration. 

Let us define the pairing. 
%when we define the pairing of $<w, v>$ for 
For a vector $v$ in $\sV$ and $f$ a branch of the solution of $\nabla^*$ over $\gamma$,
\begin{equation}
\begin{split}
\omega:=& (\frac1{x-\lambda_i} - \frac1{x-\lambda_{i+1}}) dx \otimes v\quad 
		\text{in $H^1_{dR}(\A^1, \nabla)$} ,\\
\delta:= &\gamma \otimes f\quad\quad \quad\quad\quad\text{in $H_1(\A^1, D,  \nabla^*)$}.
\end{split}
\end{equation} 
 as
$$
<\omega, \delta > = \int_{\gamma} (\frac1{x-\lambda_i} -\frac1{x-\lambda_{i+1}})<v,f>dx.
$$

To state the main theorem, we need some definitions.

\subsection{Tame symbol}
Let $\nabla$ be a rank $1$ connection on the trivial bundle with log poles at $\lambda_1, \lambda_2, ....,\lambda_n$ as above and the residue at $\lambda_i$ is a complex number
$b^i$ whose real part is positive. 
Then the connection can be written in the form :
$$
\nabla = d + \sum^n_{i = 1} \frac{b^i}{x-\lambda_i}.
$$
The solution of its dual connection $\nabla^*$ is 
a multivalued function on $U := \C -\{\lambda_1, \lambda_2,...,\lambda_n\}$
$$
\prod^n_{i=1} \exp( b^i \log (x-\lambda_i)) = \prod^n_{i=1} (x-\lambda_i)^{b^i}.
$$

Let $p$ be a fixed point in $U$ and $\gamma_i$ is a path from $p$ to $\lambda_i$.
Fix a branch of $\log (x-\lambda_i)$ around $\lambda_i$ to have real value
at $\lambda_i+\epsilon$ for $\epsilon \in \R_{>0}$ and continue it 
analytically  along $\gamma_i$. 
Denote $D(x)_{\gamma_i}$ the above solution on $\gamma_i$ of $\nabla^*$ with the chosen 
branch of logarithms.
Then the \textit{tame symbol for the rank $1$ connection $\nabla$} is defined as
\begin{equation}
\begin{split}
(\nabla, (x-\lambda_i))_{\gamma_i}:& = \lim_{x\to \lambda_i} 
\frac{D(x)_{\gamma_i}}{(x - \lambda_i)^{b^i}} \\
	& = \prod_{j \ne i} (\lambda_i - \lambda_j)^{b^j}.
\end{split}
\end{equation}

In the same manner, for a path $\gamma_\infty$ from $p$ to $\infty$, we define
$$
(\nabla, \frac1{x})_{\gamma_\infty} := \lim_{x\to \infty} D(x)_{\gamma_\infty} x^{b^\infty}.
$$
for $x^{b^\infty}$ defined by  $\exp (b^\infty \log x)$ for the principal branch of the logarithm. 

For  a higher rank connection, we define the \textit{tame symbol} as the tame symbol for its determinant connection.
%Assume again the above condition on the eigenvalues of the residue. Then we define the tame 
%symbol to be that of its determinant connection.


\subsection{Gamma factor}

Let us take the same assumption that the residue at $\lambda_i$ has only eigenvalues with
positive real part.
The Gamma factor is a generalization of the Gamma function for the residue: 
\begin{equation}
\Gamma_{x}( \nabla )  = \begin{cases}  
 \det \left(\displaystyle{\int_0^\infty 
x^{\Res_\lambda (\nabla)} e^{ -x} \frac{dx}x} \right)
&\text{for $x$ in $D$,}\\
 \det\left(\displaystyle{\int_0^\infty x^{-\Res_\infty(\nabla)}}
e^{-x}\frac{dx}x \right)
& \text{for $x=\infty$.}
\end{cases}
\end{equation}

Since the value above is invariant under constant  change of basis,
it is equal to the product of the values of the Gamma  function at the eigenvalues with multiplicity. 
Specially, when $\nabla$ is of rank $1$,  
the Gamma factor is same as the value of the classical Gamma function at the residue.


\subsection{The product formula}
Now we are prepared to state the main theorem in \cite{te1} with tame symbols and Gamma factors. 
We will assume the Condition \ref{cond}. imposed on $\nabla$.

Let $\{ e_i \}_{i=1,\ldots,r}$ be the basis for the underlying vector bundle of $\nabla$ and $\{e_j^*\}$ be the dual basis. Let $\delta_j:=\gamma_j - \gamma_{j+1}$ and $\Sol(\nabla^*) (e_q^*)$ be the branch of $\Sol(\nabla^*)$ analytically continued along $\delta$ with the initial value $e_q^*$ at $p$.
Then define 
$$
\omega_i (e_p)  := (\frac1{x-\lambda_i} - \frac1{x-\lambda_{i+1}} )dx \otimes e_p,
$$ 
and 
$$
\delta_j (e_q^*) := \delta_i \otimes \Sol(\nabla^*) (e_q^*).
$$
We define a period matrix for $(i,j)$ as
$$
A_{ij} := 
\left( \int_{\delta_j(e^*_q)} \omega_i (e_p) \right)_{1\le p,q \le r}.
$$

%We have now prepared to state the main theorem in \cite{te1}. 
With the notations introduced above, 
the main theorem in \cite{te1} gives us the exact
value of the  above period determinant in 
$\C^*$ for the basis of $H^1_{dR}(\A^1, \nabla)$ 
and of $H_1(\A^1, D, \nabla^*)$ described above.

\begin{theorem}
\label{teramain}
The determinant of the period matrix is
$$
\det \left( A_{ij} \right)_{1\le i,j \le n-1} = \prod^n_{i=1} (\nabla, x-\lambda)_{\gamma_i} \cdot (\nabla, 1/x)^{-1}_{\gamma_\infty} \prod^n_{i=1} \Gamma_{\lambda_i} (\nabla) 
\cdot \Gamma_\infty( \nabla)^{-1}.
$$
in $\C^*$.
\end{theorem}
\begin{proof} See \cite{te1}. \end{proof}

As a direct consequence,  $H^1_{dR} (\A^1,\nabla)$ and $H_1(\A^1,D,\nabla^*)$ make a perfect pairing, since the values of tame symbol and
the Gamma factor never vanish.\\


In the other hand, knowing that the differential forms
$$
\omega_i (e_p) = (\frac1{x-\lambda_i} - \frac1{x-\lambda_{i+1}}) dx \otimes e_p
$$
for $i=1,\ldots,n-1$, $p =1,\ldots, r$
make a basis of $H^1_{dR}(\A^1, \nabla)$, one 
can take  the other basis :
$$
\eta_i (e_p) = \frac{x^{i-1}dx}{\prod_{k=1}^n (x-\lambda_k)}\otimes e_p,
$$
for $i=1,\ldots,n-1$, $p=1,\ldots,r$.

\begin{proposition}
$\eta_i$ defined above generate $H^1_{dR}(\P^1, \nabla)$.
\end{proposition}
\begin{proof}
Clearly, $\eta_i(e_p)$ defines a de Rham cocycle in $H^1_{dR}(\P^1, \nabla)$. 
Since the pairing
$$
H^1_{dR}(\A^1, \nabla) \times H_1 (\A^1,D, \nabla^*) \stackrel{<,>}{\longrightarrow} \C^*
$$
is perfect by Theorem \ref{teramain},
it suffices to show that the period matrix 
$$
\begin{pmatrix}
\int_{\delta_j(e_q^*)} \eta_i(e_p)
\end{pmatrix}_{\begin{smallmatrix}i,j= 1,\ldots, n-1, \\ p,q= 1,\ldots,r\end{smallmatrix}}
$$
has nonzero determinant.

For simplicity, we will check it only for $r=1$, nevertheless the proof for higher rank cases is essentially the same.
Since $p=q=1$, we write
$$
\omega_i = (\frac1{x-\lambda_i} -\frac1{x-\lambda_{i+1}}) dx
$$
and
$$
\eta_i = \frac{x^{i-1}}{\prod^n_{k=1} (x-\lambda_i)}
$$
without the subscripts $p, q$.

Let 
\begin{equation*}
\begin{split}
\eta'_1 := & \eta_1,\quad \text{and}\\
\eta'_i :=& \prod_{k=1}^{i-1} (x-\lambda_k) \eta_i\\
	=& \frac{dx}{\prod_{k=i}^n (x-\lambda_k)}, \quad \text{for $i=2,\ldots, n-1$}.
\end{split}
\end{equation*}
Then we have the relation
\begin{equation*}
\begin{split}
x \eta'_i =& \frac{(x- \lambda_i + \lambda_i)dx}{(x-\lambda_i) \cdots (x-\lambda_n)} \\
	=& \frac{(x-\lambda_i)dx}{\prod_{k=i}^n (x-\lambda_k)} + 
	\frac{\lambda_i dx}{\prod_{k=i}^n (x-\lambda_k)} \\
	=& \eta'_{i+1} + \lambda_i\eta_i.
\end{split}
\end{equation*}
Thus we have that $\eta_i = \eta'_i + \sum_{k>i} c_k \eta'_k$ for some constants $c_k$, 
so
$$
\det \begin{pmatrix}\int_{\delta_j} \eta_i \end{pmatrix} = \det \begin{pmatrix}\int_{\delta_j} \eta'_i\end{pmatrix}.
$$

To simplify further the above determinant, we now need the following lemma.

\begin{lemma}
Let $k$ be a field of characteristic $0$. For $n$-distinct numbers 
$\lambda_1, \ldots, \lambda_n \in k$ $(n>2)$,
 the rational function in $x$ 
$$
\frac{1}{(x-\lambda_1)(x-\lambda_2) \cdots (x-\lambda_n)}
$$
is uniquely written  in the form
$$
c_1(\frac{1}{x-\lambda_1}- \frac{1}{x-\lambda_2}) + c_2 ( \frac{1}{x-\lambda_2}- \frac{1}{x-\lambda_3})+\ldots+
c_{n-1}(\frac1{x-\lambda_{n-1}} - \frac1{x-\lambda_n}). 
$$
\end{lemma}
\begin{proof}
One knows from elementary algebra
$$
\frac1{(x-\lambda_1)(x-\lambda_2) \cdots (x-\lambda_n)}= \sum_{i=1}^{n} \frac{a_i}{x-\lambda_i}
$$
for some $a_i$ in $k$.
Note that $a_i$ is the residue of 
$\frac1{(x-\lambda_1)(x-\lambda_2)\cdots(x-\lambda_n)}$ at $\lambda_i$.

So the above can be written again as
\begin{equation}
\begin{split}
&a_1 (\frac1{x-\lambda_1} - \frac1{x-\lambda_2}) 
+ (a_1 + a_2) (\frac1{x-\lambda_2} - \frac1{x-\lambda_3}) \\
+& (a_1 + a_2 + a_3) ( \frac1{x-\lambda_2} - \frac1{x-\lambda3}) 
+\cdots\\
+
& (a_1 + a_2 + \ldots + a_{n -1}) (\frac1{x-\lambda_{n-2}}  - \frac1{x-\lambda_{n-1}} )\\
+& (a_1 + a_2 + \ldots + a_{n-1} + a_n) \frac1{x-\lambda_n}.
\end{split}
\end{equation}
Whence the coefficient of the last term is zero because this is the sum of the whole residues
of a rational function. 
Once $a_1$ is determined, the other coefficients are uniquely determined. 
Since the residue is uniquely defined, we have directly the uniqueness. \end{proof}

In the proof of the previous lemma, one has 
$$
a_1 = c_1 = \frac{1}{(\lambda_1-\lambda_2)(\lambda_1 - \lambda_3)\cdots(\lambda_1 -\lambda_n)}.
$$
Thus we have
\begin{equation}
\begin{split}
\eta'_i = &\frac{dx}{(x-\lambda_i) \ldots(x-\lambda_n)} \\
	=& \frac{1}{(\lambda_i - \lambda_{i+1}) 
	(\lambda_i - \lambda_{i+2}) \cdots (\lambda_i - \lambda_n)} \omega_i \\
	& + \sum_{k > i} c_k \omega_k,
\end{split}
\end{equation}
for some (unimportant) constants $c_k$.

Seeing
\begin{equation}
\begin{split}
\eta'_{n-1} =  & \frac{dx}{(x-\lambda_{n-1})(x-\lambda_n)} \\
	= & \frac1{\lambda_{n-1} - \lambda_n} (\frac1{x-\lambda_{n-1}} - \frac1{x-\lambda_n})dx \\
	=& \frac1{\lambda_{n-1}-\lambda_n}\omega_{n-1},
\end{split}
\end{equation}
we know now the determinant is
\begin{equation}
\det \left( \int_{\delta_j} \eta_i \right) =
\frac1{\prod_{i<j}(\lambda_i -\lambda_j)}
\det
\begin{pmatrix}
\int_{\delta_j} \omega_i 
\end{pmatrix}.
\end{equation}

Therefore the determinant doesn't vanish and $\{ \delta_j \}$ and $\{ \omega_i\}$ for 
$i,j = 0,1,\ldots, n-1$ make  a perfect pairing. 
This finishes the proof of the proposition. \end{proof}

\begin{remark}
One can prove the previous proposition only with a bit of calculation showing
$\eta_i$ is linear sum of $\omega_i$ for $i=1,\ldots, n-1$ and vice versa. But here we used the
perfectness of the pairing and calculated the value, which is indispensable to approximate
the period of an irregular connection. 
\end{remark}

\subsection{Example}
Together with Theorem \ref{teramain}, the previous proposition yields the main theorem in
\cite{te0}, which plays key role in \cite{te2}. This example is taken from \cite{te2}. Here we
expose the determinant of a matrix whose elements are period integrals as that arising from the
pairing of the de Rham cohomology and the homology for a connection. This view enables us to apply the product formula to calculate the period determinant. 

Let $\lambda_1, \lambda_2,\ldots, \lambda_n$ be complex numbers and $s_1$, $s_2, \ldots, s_n$
be complex numbers with positive real part. Define the multivalued forms as
$$
\alpha_i := \prod^n_{k=1} (x-\lambda_k)^{s_k-1} x^{i-1} dx,
~~~~ \text{for $i=1,\ldots, n-1$.}
$$
Take a base point $b \in \C-\{\lambda_1, \dots, \lambda_n\}$ and define the path $I_i$ as 
$\gamma_{b,\lambda_{i+1}} - \gamma_{b,\lambda_i}$. Fix a branch of $\alpha_i$ as
an analytic continuation on $I_i$, we can evaluate the integral
$$
\int_{I_j} \alpha_i.
$$

Let $\nabla = d+ \sum_{k =1}^n \frac{dx}{x-\lambda_k}$ be a connection on $\A^1$. 
Then $H^1_{dR}(\A^1,\nabla)$ is generated with the differential forms
$$
\eta_i := \frac{x^{i-1}dx}{\prod_k (x-\lambda_k)}.
$$ 
In the other hand, the dual local system has the multivalued solution
$\prod_k(x-\lambda_k)^{s_k}$ and its homology is generated  by
$$
C_j := I_j\otimes \prod_k(x-\lambda_k)^{s_k}
$$ 
for the appropriate branch of the solution.

The pairing of $\eta_i$ and $C_j$ is by definition
\begin{equation}
\begin{split}
<C_j, \eta_i > = &\int_{I_j} \Sol(\nabla^*)\cdot\eta_i \\
	=& \int_{I_j} \prod^n_{k=1} (x-\lambda_k)^{s_k-1} x^{i-1} dx \\
	= & \int_{I_j} \alpha_i.
\end{split}
\end{equation}

Hence the period determinant is
\begin{equation}
\begin{split}
&\det \left( \int_{I_j} \alpha_j \right) \\
= &\frac1{\prod_{i<j} (\lambda_i - \lambda_j)} 
\det (\int_{I_j} (\frac1{x-\lambda_i} - \frac1{x-\lambda_{i+1}}) \prod_k (x-\lambda_k)^{s_k} dx)\\
 =&  (\frac1{\prod_{i<j} (\lambda_i - \lambda_j)} )
	\frac{\Gamma(s_1)\cdots\Gamma(s_n)}{\Gamma(s_1+\ldots+s_n)}
	\prod_k (\nabla, x- \lambda_k)_{\gamma_k} \cdot (\nabla,\frac1x)_\infty^{-1}
\end{split}
\end{equation}
by Theorem \ref{teramain}.



To get the exact value of the above, we only have to calculate the tame symbols:
\begin{equation}
\begin{split}
(\nabla,x- \lambda_k)_{\gamma_k} =& \prod_{i \ne k} (\lambda_k -\lambda_i)^{s_k} \\
(\nabla, \frac1x)_{\gamma_\infty} = & 1.
\end{split}
\end{equation}


Hence the determinant is equal to
\begin{equation}
\begin{split}
&\frac{\Gamma(s_1)\cdots\Gamma(s_n)}{\Gamma(s_1+\ldots+s_n)} \frac{\prod^n_{i=1}\prod_{j\ne i} (\lambda_j - \lambda_i)^{s_i}}{\prod_{i>j}(\lambda_j
	 - \lambda_i)}\\
 =& \frac{\Gamma(s_1)\cdots\Gamma(s_n)}{\Gamma(s_1+\ldots+s_n)}
\frac{\prod_{i\ne j}(\lambda_j - \lambda_i)}{\prod_{i>j} (\lambda_j - \lambda_i)
\prod^n_{i=1} \prod_{j\ne i}(\lambda_j - \lambda_i)^{s_i -1}}\\
=& \frac{\Gamma(s_1)\cdots\Gamma(s_n)}{\Gamma(s_1+\ldots+s_n)}
	\prod_{i<j}(\lambda_j - \lambda_i) \prod^n_{i=1}(\prod_{j\ne i} (\lambda_j - \lambda_i))^{s_i -1}.
\end{split}
\end{equation}

Then we have the theorem :
\begin{corollary}[\cite{te0}, \cite{va}] 
With the notations above,
%The determinant of the period matrix is
$$
\det (\int_{I_j} \alpha_i) = \frac{\Gamma(s_1)\cdots\Gamma(s_n)}{\Gamma(s_1 + \ldots +s_n)}
\prod_{i<j} (\lambda_j - \lambda_i) \prod_{i=1}^n (\prod_{j\ne i}(\lambda_j -\lambda_i))^{s_i -1}
$$
in $\C^*$.
\end{corollary}


\begin{remark}
In the original statement of the previous theorem in \cite{te0}, one has some assumptions: $\lambda_i$ are real numbers and $s_i$ are positive real numbers.
In view of the determinant as the period matrix arising from the pairing of the de Rham cohomology of a connection and the homology of its dual connection, we could write the determinant in terms
of tame symbol and  Gamma factor, which can be defined in more general situation.
\end{remark}

\begin{remark}
In higher rank case, the forms $(\frac1{x-\lambda_i}-\frac1{x-\lambda_{i+1})}dx\otimes e_i$ can be
changed into $\frac{x^{i-1}dx}{\prod_{i=1}^{n-1} (x- \lambda_i)}\otimes e_i$, too. In this case, the period determinant is the original one divided by $\Delta^r$, where $\Delta$ is the Vandermonde 
determinant of $\lambda_1,\ldots, \lambda_n$.
\end{remark}

%\input{dy4.tex}

%\chapter{Homology for irregular connections}
%
%\input{irreg.tex}

\section{Period integral}



In this section, we evaluate the determinant of the period determinant. 
We will assume $\lambda^2 - \lambda +1 \ne 0 $ throughout this section. We will handle the case
$\lambda^2 -\lambda+ 1 = 0$ in the next section separately.

Firstly, we will calculate the period of the rank $3$ connection 
$\pi_*\pi^*(\sO_{\A^1}, d + dy + \varpi)$, where
$\varpi$ is the one form
\begin{equation}
\varpi:= (\frac1{y+\sqrt{f(x_1)}} + \frac1{y-\sqrt{f(x_1)}} +
	\frac1{y+ \sqrt{f(x_2)}} + \frac1{y-\sqrt{f(x_2)}}) dy. 
\end{equation}

%Here, $d+\varpi$ is canonically identified  with the exterior differentiation on
%$\sO(-D)$. Thus, we will denote by $(\sO(-D), d_D)$ the connection $d+\varpi$.

Using the projection formula, we have canonically
$$
\pi_*\pi^*(\sO_{\A^1}, d + dy + \varpi) = (\sO_{\A^1}, d+dy) \otimes (\sO_{\A^1}, d+\varpi)
\otimes \pi_*(\sO_U, d).
$$

It is decomposed as direct sum of rank $1$ and rank $2$ connections. 
Thus calculating the periods separately, we can multiply them to get the whole period. 

%$\nabla_\Sigma$
%over $\P^1$, whose rank $2$ part yields nontrivial information, is 
%\begin{equation*}
%\begin{split}
%\nabla_{\Sigma}
% = &d + I dy +
%	\begin{pmatrix}
%		f_1 & 2(f_0 - f_2 \lambda) \\
%		f_2 & 2(f_1 + f_2(\lambda+1))
%	\end{pmatrix} \omega \\ 
%	&+ I (\frac1{y+\sqrt{f(x_1)}} + \frac1{y-\sqrt{f(x_1)}} +
%	\frac1{y+ \sqrt{f(x_2)}} + \frac1{y-\sqrt{f(x_2)}}) dy. 
%\end{split}
%\end{equation*}

The period determinant of the above connection will be compared later with 
that of the original one.

\subsection{Small extension}
The general theory of regular singular connections is developed in \cite{de}. 
Deligne's Theorem of existence tells us that, for a  regular singular
connection $(E,\nabla)$ on a nonsingular variety $U$
there exists  a logarithmic extension 
$$
\bar{\nabla} : \bar{E}\to \bar{E}\otimes \Omega^1(\log D)
$$ 
on its completion $X = U \cup D$ for a normal crossing divisor $D$ in $X$.
 
Let $j:U\hookrightarrow X$ be the open embedding.

\begin{definition}
If the eigenvalues of its residue at a  irreducible component $D_i$ of $D$ 
is not an integer smaller than or equal to $0$ (resp. bigger than $0$), $\bar{\nabla}$ is called a small (resp. big) extension of $\nabla$ at $D_i$.  When $\bar{\nabla}$ is a
 small extension of $\nabla$ at each component of $D$, $\bar{\nabla}$ is called a small extension of $\nabla$.
\end{definition}

\begin{remark}
Note that the above definition is the generalization of the notion 
\textit{small extension} discussed in the previous section. In previous section, 
the notion of small extension was only defined for meromorphic connections
on the trivial bundle over $\P^1$. Here, we have extended it for connections on 
an arbitrary curve (cf. \cite{st}).
\end{remark}

The de Rham complex of a small (resp. big) extension $(\bar{E}, \bar{\nabla})$ 
is quasi-isomorphic to 
$j_! (E^\nabla)$ (resp. $Rj_* (E^\nabla)$)  in strong topology (cf. Lemma 1.6 in \cite{st}). 
Thus we have
\begin{equation}
\H^i (X, (\bar{E}, \bar{\nabla}) )  \simeq 
	\begin{cases}
		H^i_c (U, \sE) & \text{if $\bar{\nabla}$ is small,} \\
		H^i (U, \sE) & \text{if $\bar{\nabla}$ is big.}
	\end{cases} 
\end{equation}

Returning to the calculation, from now on, we take 
$U = \Spec k[x,y]/(y^2 -f(x))$ for $f(x)= x(x-1)(x-\lambda)$ and 
assume $f(x_1)\ne f(x_2)$, which is equivalent
to $\lambda^2-\lambda +1 \ne 0$.

The product formula of Terasoma is valid only for small extensions. 
The rank $2$ part of $\pi_* \nabla$ is not small at the ramification points. 
It has the eigenvalues
$0, 1/2$ of the residues at $D = \{\pm \sqrt{f(x_1)}$, $\pm \sqrt{f(x_2)}\}$.

Let $\nabla_D$ be the tensor of $d+dy$ and $d+\varpi$ on $\sO_{\A^1}$:

\begin{equation}
\begin{split}
\nabla_D =&d + dy \\ 
	&+ (\frac1{y - \sqrt{f(x_1)}} + \frac1{y+ \sqrt{f(x_1)}}
		+ \frac1{y-\sqrt{f(x_2)}} + \frac1{y+ \sqrt{f(x_2)}}) dy.
\end{split}
\end{equation}


Hence we will calculate the period for the connection
$$
\pi_* \pi^* ( \sO_{\A^1}, \nabla_D),
$$
%where $D$ is the divisor of ramification points $\pm \sqrt{f(x_1)}$, $\pm\sqrt{f(x_2)}$ and

As before, applying the projection formula, we obtain
\begin{equation*}
\begin{split}
\pi_*\pi^* (\sO_{\A^1}(-D), \nabla_D) =& (\sO(-D), \nabla_D) 
					\otimes \pi_*(\sO_X, d) \\
		=& (\sO(-D), \nabla_D) \otimes ((\sO_{\A^1}, d) \oplus (\sV, \nabla_\sV)). 
\end{split}
\end{equation*}

Note $\pi_* \pi^*(\sO_{\A^1}(-D), \nabla_D)$ is small since the rank $1$ part has only $1$ as the residue and the rank $2$ part has the eigenvalues $1/2, 1$ of the residue
at a point in $D$. 

Since the connection in consideration splits into the direct sum of rank $1$ and rank $2$ connections, the period will be given by the product of the period of the rank $1$ part with that of  the rank $2$ part.


One is now prepared to evaluate the period by approximation. The approximation will be made
with a sequence of regular singular connections to produce the period for the irregular connection.
This was already applied in \cite{te2} without the interpretation in terms of connection, for a special
type of rank $1$ connection. We will imitate the process for a higher rank connection on $\A^1$.

\subsection{Rank $1$ part}
The rank $1$ part has been already estimated in \cite{te2}.  
We recall the Theorem 2.3.3. in op.cit. To state the theorem, we need to reorder
some notations. 

Let $F(x)$ be a polynomial of degree $d$, $\lambda_i, \ldots, \lambda_n$ be distinct
complex numbers and $s_1, s_2, \ldots, s_n$ be positive real numbers.
 
For a rank $1$ connection 
$$
\nabla= d+ dF +\sum_{i=1}^n \frac{s_i}{x-\lambda_i}
$$ 
on $\A^1$, the de Rham cohomology is generated by 
$$
\eta_i = \frac{x^{i-1} dx}{\prod_{k=1}^n (x-\lambda_k)}, \hspace{20pt}
\text{for $i=1,\ldots, n-1$}.
$$ 

And the homology valued in the dual connection is generated by these
cycles 
\begin{equation*}
\begin{split}
I_j \otimes \prod_{k=1}^n (x-\lambda_k)^{s_k-1} \exp(F(x)),~~~&
	\text{for $j= 1,2,\ldots, n-1,$ and  }\\
J_j \otimes \prod_{k=1}^n (x-\lambda_k)^{s_k-1}\exp(F(x)),~~~&
	\text{for $j= 1,2,\ldots, d$.}
\end{split}
\end{equation*}
where $I_j$ is a path from $\lambda_j$ to $\lambda_{j+1}$ and $J_j$ is a path
from $\lambda_n$ to $\infty$ along $j$-th rapid decay sector.

It is a theorem on the determinant of the following matrix of confluent hypergeometric functions:
$$
D = \det \big( 
(\int_{I_j}\prod_{k=1}^n (x-\lambda_k)^{s_k-1} \eta_i )_{\begin{smallmatrix}
	i = 1,\ldots, n+d-1\\
	j = 1,\ldots, n-1
	\end{smallmatrix}}
	(\int_{J_j}\prod_{k=1}^n (x-\lambda_k)^{s_k-1} \eta_i )_{\begin{smallmatrix}
	i=1,\ldots,n+d-1\\
	j=1,\ldots,d
	\end{smallmatrix}}\big)
$$

\begin{theorem}[Terasoma\cite{te2}]
The determinant $D$ of the confluent hypergeometric functions is the number in $\C^*$:
\begin{equation}
\begin{split}
D = &(2\pi)^{(d-1)/2}\Gamma(s_1)\cdots\Gamma(s_n) (da_d)^{-s-(d-1)/2}
		(-1)^{ds+d(d-1)/4}\\
	& \times \prod_{i=1}^n(\prod_{j\ne i} (\lambda_i -\lambda_j))^{s_i-1}
		\prod_{i<j} (\lambda_j - \lambda_i) \\
	& \times \prod_{i=1}^n \exp(F(\lambda_i)) 
		\prod_{F'(u)=0} \exp (F(u)),
\end{split}
\end{equation}
where $s=s_1+\cdots+s_n$.
\end{theorem}
\begin{proof}
See \cite{te2}. 
\end{proof}

In our case, the connection is
$$
\nabla = d + dy + (\frac1{x+\sqrt{f(x_1)}} + \frac1{x- \sqrt{f(x_1)}}
	+\frac1{x+\sqrt{f(x_2)}} + \frac1{x-\sqrt{f(x_2)}})dy.
$$
Hence the period determinant is
\begin{equation}
\begin{split}
D = &(\sqrt{f(x_1)}+\sqrt{f(x_2)})(2\sqrt{f(x_1)}) (\sqrt{f(x_1)} -\sqrt{f(x_2)})\\
& \times (\sqrt{f(x_1)}- \sqrt{f(x_2)}) (-2\sqrt{f(x_2)})(- \sqrt{f(x_2)} - \sqrt{f(x_1)})\\
 = & 2^2 \sqrt{f(x_1)} \sqrt{f(x_2)} (f(x_1) - f(x_2))^2
\end{split}
\end{equation}
in $\C$.

\subsection{Rank $2$ part}
Recall that the rank $2$ part of $\pi_*(\sO_U, d)= (\ker \Tr, \nabla_1)$ has the connection 
$$
\nabla_1 = d+ \begin{pmatrix}
f_1 & 2(f_0 - f_2 \lambda) \\
f_2 & 2(f_1 + f_2(\lambda+1))
\end{pmatrix} \omega.
$$

The rank $2$ part of $\pi_*\pi^*(d+dy)$ has the connection
\begin{equation}
\begin{split}
 \nabla_D \otimes \nabla_1 
=&d + I dy \\
	& + I\big( \frac1{y-\sqrt{f(x_1)}} + \frac1{y+\sqrt{f(x_1)}} 
		+ \frac1{y-\sqrt{f(x_2)}} + \frac1{y+\sqrt{f(x_2)}}\big) dy \\
	& + \begin{pmatrix} f_1  &2(f_0 - f_2 \lambda) \\
		f_2 & 2 (f_1 + f_2(\lambda+1)) \end{pmatrix} \omega.
\end{split}
\end{equation}

Let $e_1, e_2$ be a basis of $\sV$ and $e^*_1, e^*_2$ be its dual basis so 
that $<e_i, e^*_j>$ $=\delta_{ij}$. 
Let $\gamma_i$ be the path from $0$ to $-\sqrt{f(x_1)},
 \sqrt{f(x_1)},  -\sqrt{f(x_2)},$  $\sqrt{f(x_2)}$, respectively for $i=1,2,3,4$. Let $\gamma_\infty$ be the path from $p$ to $\infty$ along the unique rapid decaying sector around $\infty$. 
$I_j$ is the chain $\gamma_{i+1} - \gamma_i$ for $i=1,2,3$.
$I_4$ is $\gamma_\infty - \gamma_3$.
%We will explain the cycles $I_j$ later.

We will take the cocycles 
\begin{equation}
\begin{split}
\eta_{i,a} :=\frac{y^{i-1} dy}{(y-\sqrt{f(x_1)})(y+\sqrt{f(x_1)})(y-\sqrt{f(x_2)})(y+\sqrt{f(x_2)})} \otimes e_a,\\
\intertext{for $i=1,\ldots,4$ and $a=1,2$}
\end{split}
\end{equation}
as a basis for $H^1_{dR} (\A^1, \nabla_D\otimes\nabla_1)$, and the cycles
\begin{equation}
\begin{split}
C_{j,b} := I_j \otimes (\exp y \cdot \Sol(\nabla_\sV^*))(e^*_b),&\quad\text{for $j=1,\ldots, 4$,}
%\intertext{and}
%\quad\quad  I_4\otimes (\exp y\cdot\Sol(\nabla_\sV^*))(e^*_b)
\end{split}
\end{equation}
as a basis for $H^{irreg}_1(\A^1, D, (\nabla_D\otimes \nabla_1)^*)$ 
where the notations are as defined in Section 3.


The  period pairing is the integration 
\begin{equation}
\label{pd}
\begin{split}
&P_{i,j,a,b}:= < \eta_{i,a},C_{j,b}>\\
= &< \frac{y^{i-1}dy}{(y-\sqrt{x_1})(y+\sqrt{x_1})(y-\sqrt{x_2})(y+\sqrt{x_2})}\otimes e_a,
  I_j\otimes \exp y \cdot \Sol(\nabla_\sV^*)(e^*_b) > \\
%\end{split}
%\end{equation}
%which becomes  the period integral  of the following form:
%\begin{equation}
%\begin{split}
= &\int_{I_j} \frac{y^{i-1} \exp y < e_a, \Sol(\nabla_\sV^*)(e^*_b)> dy}{(y-\sqrt{x_1}) (y+\sqrt{x_1}) (y - \sqrt{x_2})(y+\sqrt{x_2})}\\
\intertext{ for $i=1,2,3,4$ and $a, b = 1,2$.}
\end{split}
\end{equation}

Let $I^{(m)}_j$ be $I_j$ for $j=1,2,3$ and $I^{(m)}_4$ be $\gamma_{(-m)} - \gamma_4$, where
$\gamma_{(-m)}$ is a path from $0$ to $-m$.
The above integral can be approximated by
\begin{equation}
\begin{split}
& P_{(m),i,j,a,b}\\
= &\int_{I^{(m)}_j}
\frac{y^{i-1} (1+ \frac{y}m)^m <e_a, \Sol (\nabla_\sV^*)(e^*_b) > dy}{(y-\sqrt{f(x_1)})(y+\sqrt{f(x_1)})
	(y-\sqrt{f(x_2)})(y+\sqrt{f(x_2)})} \\
=&  (\frac1m)^m\int_{I_j} 
\frac{y^{i-1}(y+m)^{m+1} < e_a, \Sol (\nabla_\sV^*)(e^*_b) > dy}{(y-\sqrt{f(x_1)})(y+\sqrt{f(x_1)})
	(y-\sqrt{f(x_2)})(y+\sqrt{f(x_2)})(y+m)},
\end{split}
\end{equation}
as $m$ tends to $\infty$.


The above, in turn, appears as the period integral of the regular singular connection
\begin{equation}
\begin{split}
\nabla^{(m)} =& d + I \frac{(m+1) dy}{y+m} \\
&+ I (\frac1{y-\sqrt{f(x_1)}}+ \frac1{y+\sqrt{f(x_1)}} + \frac1{y-\sqrt{f(x_2)}} + \frac1{y+\sqrt{f(x_2)}})dy\\
& + \begin{pmatrix}
	f_1 & 2(f_0 - f_2 \lambda) \\
	f_2 & 2(f_1 + f_2 (\lambda+1)) 
    \end{pmatrix} \omega.
\end{split}
\end{equation}
Note that the form $\frac{(m+1) dy}{y+m}$ was chosen to approximate $dy$.
Thus we have to  calculate the period for a fixed $m$.

Now we need to calculate its tame symbols and Gamma factors.

\subsubsection{Tame symbols}
Tame symbols depend only on the determinant bundle.
\begin{equation}
\begin{split}
\det \nabla^{(m)} =& d + \frac{2(m+1)}{y+m} dy\\
	+& \frac52(\frac1{y-\sqrt{f(x_1)}} + \frac1{y+\sqrt{f(x_1)}} + \frac1{y-\sqrt{f(x_2)}} +\frac1{y+\sqrt{f(x_2)}})dy
\end{split}
\end{equation}

The multivalued solution of its dual local system is
\begin{equation}
\Sol(\det\nabla^*) = (y^2 - f(x_1))^{5/2} (y^2 - f(x_2))^{5/2} (y+m)^{2(m+1)}.
\end{equation}
%From this we can calculate the tame symbol:
It follows that
\begin{equation}
\begin{split}
(\nabla^{(m)}, y-\sqrt{f(x_1)})_{\gamma_1} &= (2\sqrt{f(x_1)})^{5/2}
				 (f(x_1)-f(x_2))^{5/2} (\sqrt{f(x_1)} +m)^{2(m+1)},\\
(\nabla^{(m)}, y+\sqrt{f(x_1)})_{\gamma_2}  &= ( -2 \sqrt{f(x_1)})^{5/2}
				(f(x_1) -f(x_2))^{5/2} (-\sqrt{f(x_1)}+m)^{2(m+1)},\\
(\nabla^{(m)}, y-\sqrt{f(x_2)})_{\gamma_3} &= (2\sqrt{f(x_2)})^{5/2}
				 (f(x_2)-f(x_1) )^{5/2} (\sqrt{f(x_2)}+m)^{2(m+1)},\\
(\nabla^{(m)}, y+\sqrt{f(x_2)})_{\gamma_4} &= (-2\sqrt{f(x_2)})^{5/2}
				 (f(x_2)-f(x_1))^{5/2} (-\sqrt{f(x_2)}+m)^{2(m+1)},\\
(\nabla^{(m)},y+m)_{\gamma_{-m}} &=  (m^2 - f(x_1))^{5/2}(m^2 -f(x_2)^{5/2}\quad\text{and}\\
(\nabla, \frac1y)_{\gamma_\infty} & = 1.
\end{split}
\end{equation}

\subsubsection{Gamma factors}
Gamma factors depend  on the residue and its eigenvalues.

At $\pm\sqrt{f(x_i)}$ for $i=1,2$, the residue has two eigenvalues $1, 3/2$. Thus the Gamma 
factor at $\pm\sqrt{f(x_i)}$ for $i=1,2$ is
$$
\Gamma_{\pm\sqrt{f(x_i)}}(\nabla) =\Gamma(1)\Gamma(\frac32) = \frac12\Gamma(\frac12) =\frac{\sqrt{\pi}}2.
$$

At $y=-m$, since the residue has twice of $m+1$ as eigenvalue,
$$
	\Gamma_{-m}(\nabla) = \Gamma(m+1)^2.
$$

Let $A_\infty d(1/y)$ be the connection form of $\nabla^{(m)}$.
At $y=\infty$, the connection form of $\nabla$ is
\begin{equation}
\begin{split}
A_\infty d(1/y)&:= - ((m+1)+5) d(1/y)/(1/y)I\\
& - \frac23
\begin{pmatrix}
\frac19(\lambda+1)(2\lambda^2 -3\lambda +2) + y^2 & \frac29\lambda(\lambda+1)-
					\frac23(\lambda+1)y^2 \\
-\frac29 (\lambda^2 - \lambda +1) & -\frac29 \lambda(\lambda+1) + 2y^2
\end{pmatrix}\\
&\cdot
\frac{y^2}{(y^2-f(x_1))(y^2-f(x_2))}\frac{d(1/y)}{(1/y)}.
\end{split}
\end{equation}

The residue at $\infty$ is
$$
\lim_{y\to \infty} A_\infty (1/y)= -\frac23
\begin{pmatrix}
(m+1) + \frac{14}3 &  - \frac23(\lambda+1) \\
0 &  (m+1) + \frac{16}3
\end{pmatrix}.
$$

It follows that the residue has two eigenvalues $- (m+1)+\frac{14}3$ and $-(m+1) +\frac{16}3$.
Hence the Gamma factor at $\infty$ is
$$
\Gamma_\infty (\nabla) = \Gamma((m+1)+\frac{14}3)\Gamma((m+1)+\frac{16}3).
$$

\subsubsection{Period determinant of $\nabla^{(m)}$}
Now we have prepared everything to calculate the period of $\nabla^{(m)}$: 
the tame symbols and the Gamma factors.

Let
$$
C^{(m)}_{j,b} := I_j^{(m)} \otimes (y+m)^{m+1} \Sol(\nabla^*_\sV)(e^*_b) 
$$
and 
 \begin{eqnarray}
\omega^{(m)}_{1a} :=& (\displaystyle{\frac{1}{y+\sqrt{x_1}}- \frac{1}{y-\sqrt{x_1}}})dy\otimes e_a,\\
\omega^{(m)}_{2a} :=& (\displaystyle{\frac{1}{y-\sqrt{x_1}} -\frac{1}{y+\sqrt{x_2}}})dy\otimes e_a,\\
\omega^{(m)}_{3a} :=& (\displaystyle{\frac{1}{y+\sqrt{x_2}} -\frac{1}{y-\sqrt{x_2}}})dy\otimes e_a,\\
\omega^{(m)}_{4a} :=& (\displaystyle{\frac{1}{y-\sqrt{x_1}} -\frac{1}{y+m}})dy\otimes e_a.
\end{eqnarray}
Applying Terasoma's product formula, the period of $\nabla^{(m)}$ is
\begin{equation}
\begin{split}
D_{(m)} :=& \det \left( < \omega^{(m)}_{i,a}, J^{(m)}_{j,b}>\right)_{\begin{smallmatrix}
		1\le i \le 4, 1\le a\le 2,\\
		1\le j\le 4, 1\le b\le 2
		\end{smallmatrix}}
   \\
=& 2^{6}\pi^2 f(x_1)^{5/2} f(x_2)^{5/2} (f(x_1)-f(x_2))\\
& \times (m^2 - f(x_1))^{2(m+1)+5/2}
	(m^2 - f(x_2))^{2(m+1)+5/2} \\
	%(m^2 - f(x_1))^{5/2} (m^2- f(x_2))^{5/2}\\
&\times \frac{\Gamma(m+1)^2}{\Gamma((m+1)+\frac{14}3)\Gamma((m+1)+\frac{16}3)}.
\end{split}
\end{equation}

The determinant of period matrix $P_{(m)}$ is the complex number
converging to the determinant of the period matrix $(P_{i,j,a,b})$ defined in
($\ref{pd}$)
\begin{equation}
\begin{split}
\det P_{(m)} =& \det \left(  P_{(m), i,j,a,b}\right) \\
	=&(\frac1m)^{8m} \frac{D_{(m)}}{\Delta_{(m)}^2} 
\end{split}
\end{equation}
where $\Delta_{(m)}$ is the Vandermonde determinant for
$
\pm\sqrt{f(x_1)}, \pm \sqrt{f(x_2)} , -m:
$
%the period in consideration should be the above number divided by $\Delta^2$ where
\begin{equation}
\label{vander}
\begin{split}
&\Delta_{(m)} = \\
&(\sqrt{f(x_2)}+\sqrt{f(x_1)})2\sqrt{f(x_1)} (-\sqrt{f(x_2)} + \sqrt{f(x_1)}) (-m +\sqrt{f(x_1)})\\
&\times(\sqrt{f(x_1)} - \sqrt{f(x_2)})(-2\sqrt{f(x_2)})(-m- \sqrt{f(x_2)})\\
&\times(-\sqrt{f(x_2)} - \sqrt{f(x_1)})(-m - \sqrt{f(x_1)})\\
&\times (-m + \sqrt{f(x_2)})\\
&=  - 2^2 \sqrt{f(x_1)}\sqrt{f(x_2)} (f(x_2)-f(x_1))^2 (m^2 -f(x_1))(m^2-f(x_2)).
\end{split}
\end{equation}

Thus the period is
\begin{equation}
\begin{split}
&(\frac1m)^{8m} \frac{P_m}{\Delta_{(m)}^2} \\
=& (\frac1m)^{8m} 2^2 \pi^2 f(x_1)^{3/2} f(x_2)^{3/2} (f(x_1)-f(x_2))^{-3}\\
&\times( m^2 - f(x_1))^{2(m+1)+1/2} (m^2 - f(x_2))^{2(m+1)+1/2}\\
	&\times \frac{\Gamma(m+1)^2}{\Gamma((m+1)+\frac{14}3)\Gamma((m+1)+\frac{16}3)}\\
=& \frac{2^2\pi^2 f(x_1)^{3/2} f(x_2)^{3/2}}{(f(x_1) - f(x_2))^3}\\
 &\times (\frac1m)^{8m} (m^2 - f(x_1))^{2m} (m^2-f(x_2))^{2m}\\
 &	\times m^{-10}(m^2 - f(x_1))^{5/2} (m^2 - f(x_2))^{5/2} \\
 &	\times m^{10} \frac{\Gamma(m+1)^2}{\Gamma((m+1)+\frac{14}3) \Gamma((m+1)+\frac{16}3)}.
\end{split}
\end{equation}

As $m$ goes to $\infty$,
\begin{eqnarray}
 (\frac1m)^{8m} (m^2 - f(x_1))^{2m} (m^2-f(x_2))^{2m} & \to 1\\
  m^{-10}(m^2 - f(x_1))^{5/2} (m^2 - f(x_2))^{5/2} &\to 1\\
 m^{10} \frac{\Gamma(m+1)^2}{\Gamma((m+1)+\frac{14}3) \Gamma((m+1)+\frac{16}3)} & \to 1.
\end{eqnarray}

Therefore the period determinant for $(\sV, \nabla_D\otimes\nabla_\sV)$ is  
$$
 \frac{2^2\pi^2 f(x_1)^{3/2} f(x_2)^{3/2}}{(f(x_1) - f(x_2))^3}.
$$


Now we have the period determinant of $\pi_*\pi^*(\nabla_D)$:
% this yields, together with the rank $1$ period,
\begin{equation}
\begin{split}
\label{Fperiod}
%\tag{$*$}
%\tag{$*$}
&(\text{Period of $(\sO_{\A^1},\nabla_D)$} ) \times
(\text{Period of $(\sV, \nabla_D\otimes \nabla_\sV)$})\\
=&\frac{2^4\pi^2 f(x_1)^2 f(x_2)^2}{f(x_1)- f(x_2)}.
\end{split}
\end{equation}

\subsection{Comparison}
Recall $U$ is the affine Legendre elliptic curve ( i.e. the plane curve defined by 
$y^2 =x(x-1)(x-\lambda)$)
%and $U=X - \{ \infty\}$ and 
and $\nabla = d + dy$ is a connection on $\sO_U$ as in the 
beginning. 
In this part, we will compare the period obtained previously and the
period of $\nabla$ on $U$.  
%$\pi$ is the projection map $(x,y)\mapsto y$. 
$D= \{ \pm\sqrt{f(x_1)}, \pm\sqrt{f(x_2)}\}$  is the divisor in $\P^1$ and $\Sigma= \pi^{-1} D$.
% the set of
%ramification points in $\P^1$ and $\pi^{-1} D$, respectively.
Let $\pi_*(\nabla)_D$ be the rank $3$ connection on $\A^1$
$$
\pi_*(\nabla)_D := d_D \otimes \pi_* \nabla,
$$
whose period we have just calculated, where
\begin{equation}
\begin{split}
d_D &:=\\
& d + (\frac1{y-\sqrt{f(x_1)}} + \frac1{y+\sqrt{f(x_1)}} + \frac1{y- \sqrt{f(x_2)}} + \frac1{y+\sqrt{f(x_2}}) dy.
\end{split}
\end{equation}
(i.e. the standard exterior differentiation of $\sO(-D)$ valued in $\sO$.)

In the same manner, we define $\nabla_\Sigma :=d_\Sigma\otimes \nabla$.

Then we have the canonical isomorphism of de Rham cohomologies
$$
\pi^* : H^1_{dR} ( \A^1 , \nabla_D) \to H^1_{dR} (U, \nabla_\Sigma ) .
$$

Let $\omega= \pi^* \eta $ be a de Rham form in $H^1_{dR}(U,\nabla_\Sigma)$ for a form $\eta$ in 
$H^1_{dR} (\A^1,\nabla_D)$ and $\gamma$ be a cycle in
the homology of $U$ with values in the dual local system.
Then, as the pairing is functorial under change of variables,
\begin{equation}
\begin{split}
\int_\gamma \omega  = & \int_\gamma \pi^* \eta \\
	=& \int_{\pi_*\gamma} \eta.
\end{split}
\end{equation}
Hence the period (\ref{Fperiod}) of $\pi_*(\nabla)_D$ on $\A^1$ is the same as that of $\nabla_\Sigma$ on $U$.

The following short exact sequence of de Rham complexes 
\begin{equation}
%\begin{split}
\begin{CD}
\sO_U(-\Sigma) 	@>>> 		\sO_U  	@>>>     \sO_\Sigma \\
@VV{\nabla_\Sigma}V	@VV{\nabla}V 				@VVV \\
\Omega_U(-\Sigma)(\log \Sigma) @=  \Omega_U     @>>>  0
\end{CD} 
%\end{split}
\end{equation}
yields 
\begin{equation}
%\begin{CD}
\begin{split}
H^0_{dR} (U,\nabla) \to H^0 (U, \sO_\Sigma) \stackrel{\nabla}{\to} H^1_{dR} (U, \nabla_\Sigma)
 \to H^1_{dR} (U,\nabla) \to H^1 (U,\sO_\Sigma)\
%\end{CD}
\end{split}
\end{equation}

Whereas $H^0_{dR}(U,\nabla) = 0$, for $\nabla$ has no (single-valued) solution on $U$ and
$H^1(U,\sO_\Sigma) =0$ for dimensional reason. Thus $H^1_{dR} (U,\nabla_\Sigma)$ is an extension of $H^1_{dR}(U,\nabla)$ by $H^0(U,\sO_\Sigma)$.

In the other hand, the relative homology $H^{irreg}_1(U,\Sigma,\nabla^*)$ is isomorphic to
$H^{irreg}_1(\A^1, D, \pi_*(\nabla^*))$. 
Recall the homology for a irregular connection is defined as the homology of
the complex $\sC^{irreg}_*( U,\nabla^*)$ generated by elements $c\otimes \epsilon$
with $c:\Delta^n \to X$ and $\epsilon\in \sL_{c(b)}$ for $b$ the barycenter of $\Delta^n$,
$\sL$ the local system of $\nabla^*$ and $\epsilon$ decays rapidly near $\infty$.

The relative cohomology appears in the short exact sequence
of the singular complexes valued in $\nabla^*$:
$$
0 \to \sC^{irreg}_*(\Sigma, \nabla^*) \to \sC^{irreg}_*(U,\nabla^*) 
\to \sC^{irreg}_*(U,\Sigma,\nabla^*) \to 0,
$$
which yields also the long exact sequence of homologies: 
\begin{equation}
\begin{split}
...\to H^{irreg}_1(\Sigma,\nabla^*) \to H^{irreg}_1 ( U, \nabla^*) \to H^{irreg}_1 (U,\Sigma,\nabla^*)\\
\to
H^{irreg}_0(\Sigma, \nabla^*) \to H^{irreg}_0 (U,\nabla^*) \to ...
\end{split}
\end{equation}
The exact definition of the singular complex for an irregular connection can be found in \cite{be}.

In the above sequence, $H^{irreg}_1(\Sigma,\nabla^*)$ and $H^{irreg}_0 (U,\nabla^*)$ vanish,  respectively 
by dimension reason and by the duality of the de Rham cohomology and the 
homology for irregular connections.

\begin{theorem}
Let $\gamma\otimes \Sol(\nabla^*)$ be a cycle in $H^{irreg}_1(U,\nabla^*)$, thus it is 
also a cycle in $H^{irreg}_1(U,\Sigma,\nabla^*)$.
Suppose $\omega = \nabla f$ for a function $f$ in $H^0(U,\sO_\Sigma)$.
Then the pairing of $\gamma\otimes \Sol(\nabla^*)$ with $\nabla f$ is $0$.
\end{theorem} 
\begin{proof}
The pairing is given as the integration
$$
<\gamma\otimes \Sol(\nabla^*), \nabla> = \int_\gamma <\Sol(\nabla^*), \nabla f> .
$$
This is equal to
\begin{equation}
\begin{split}
& \int_\gamma (d < \Sol(\nabla^*), f > - <\nabla^*(\Sol(\nabla^*)), f> )\\ =&
				\int_\gamma d<\Sol(\nabla^*),f> \\
	= & \int_{\partial \gamma} <\Sol(\nabla^*), f>
\end{split}
\end{equation}
by Stokes' theorem.

If $\gamma$ is a closed cycle, its boundary $\partial \gamma$ is $0$. Otherwise, we can
assume without loss of generality, 
$$
\gamma : [0,1] \to U\cup \{\infty\}
$$
such that $\gamma(0) = \gamma(1) = \infty$, moreover, for some small 
$\epsilon>0$, $\gamma( 0,\epsilon)$, $\gamma(1-\epsilon, 1)$ 
are inside the rapid decay sector around $\infty$. 

Thus
$$
\int_\gamma \nabla f  = \lim_{\epsilon\to0} [<\Sol(\nabla^*),f>  ]^{\gamma(1-\epsilon)}_{\gamma(\epsilon)} = 0,
$$
since $\Sol (\nabla^*)$ decays rapidly along the sector where $\gamma$ goes along.
\end{proof}

Let us denote by $per( \Sigma, \nabla)$ the period determinant of the  pairing of $H^0(U,\sO_\Sigma)$ and
$H^{irreg}_0(\Sigma, \nabla^*)$ for suitable choice of basis. 
If we take a basis of $H^1_{dR}(U,\Sigma,\nabla)$ containing $\{\nabla(f_\alpha)\}_\alpha$ where
$\{f_\alpha\}_\alpha$ is a basis of $H^0(\Sigma,\nabla)$,  it follows 
\begin{corollary}
The determinant of period of $\nabla_\Sigma$ is 
$$
 per(U,\Sigma, \nabla_\Sigma) = per(U, \nabla) \times per(\Sigma, \nabla).
$$
in $\C^*/k^*$.
\end{corollary}

$\Sigma$ is given by the ideal 
$$I = (y^2 -f(x_1))(y^2 -f(x_2))
$$ 
in 
$R:= k[x,y]/(y^2-x(x-1)(x-\lambda))$. 

Hence,
$$
H^0(U,\sO_\Sigma) = R/I.
$$
This is the $k$-vector space of dimension $8$ generated by
$$
1, y, y^2, y^3, x, xy, xy^2, xy^3.
$$

And the homology $H^{irreg}_0 (\Sigma, \nabla^*)$ is generated by
$$
p \otimes \exp y,
$$
for $p_i\in \Sigma$.

Let $x_3$ (resp. $ x_4$ ) be the root of $f(x) - f(x_1)$ (resp. of $f(x) -f(x_2)$), respectively,
such that  $x_3 \ne x_1$ (resp. $x_4\ne x_2$). 
A point $p$ in $\Sigma$ can be written as $(x_i, \pm \sqrt{f(x_i)})$ and
the pairing is given as
$$
< p \otimes \exp y, x^a y^b > 
= x_i^a (\pm \sqrt{f(x_i)})^b \cdot\exp (\pm \sqrt{f(x_i)}),
$$
for $a=0,1$ and $b=1,2,3,4$.

Let $M_i$ be the following matrix:
$$
M_i := \begin{pmatrix}
1 & \sqrt{f(x_i)} & f(x_i)  & f(x_i)\sqrt{f(x_i)}   \\
1 & - \sqrt{f(x_i)}   & f(x_i) & - f(x_i)\sqrt{f(x_i)} \\ 
\end{pmatrix}
$$
for $i=1,2$.

Then the period matrix in consideration is 
$$
Q := L
\begin{pmatrix}
M_1 &  x_1 M_1  \\
M_2 &  x_2 M_2  \\
M_1  & x_3 M_1 \\
M_2 & x_4 M_2 \\
\end{pmatrix},
$$
where $L= (l_{ij})$ is a $(8,8)$-diagonal matrix with entries 
$$
l_{ii}= \begin{cases}
e^{f(x_1)} & \text{for $i\equiv 1 \pmod{4}$}\\
e^{-f(x_1)} &\text{for $i\equiv 2 \pmod{4}$}\\
e^{f(x_2)} & \text{for $i \equiv 3 \pmod{4}$}\\
e^{-f(x_2)} & \text{for $i\equiv 0 \pmod{4}$}
\end{cases}
$$ 

It follows that its determinant is
\begin{equation}
\begin{split}
\det Q =&  \det 
\begin{pmatrix}
M_1 &  x_1 M_1  \\
M_2 &  x_2 M_2  \\
M_1  & x_3 M_1 \\
M_2 & x_4 M_2 \\
\end{pmatrix} \\
	=&
\begin{pmatrix}
M_1 &  x_1 M_1  \\
M_2 &  x_2 M_2  \\
0   & (x_3 - x_1)  M_1 \\
0   & (x_4 - x_2)  M_2 \\
\end{pmatrix}\\
=& (x_3 - x_1)^2 (x_4 - x_2)^2 \Delta_\Sigma^2,
\end{split}
\end{equation}
where $\Delta_\Sigma$ is the Vandermonde determinant for $\sqrt{f(x_1)}, -\sqrt{f(x_1)},$
$\sqrt{f(x_2)}, -\sqrt{f(x_2)}$. 
This is
\begin{equation}
\begin{split}
\Delta_\Sigma =& -2 \sqrt{f(x_1)} (\sqrt{f(x_2)} - \sqrt{f(x_1)})(-\sqrt{f(x_2)} -\sqrt{f(x_1)})\\
	&\times (\sqrt{f(x_2)}+ \sqrt{f(x_1)}) (-\sqrt{f(x_2)} + \sqrt{f(x_1)})\\
	& \times (-2\sqrt{f(x_2)} ) \\
	=& 2^2 \sqrt{f(x_1)} \sqrt{f(x_2)} (f(x_2)-f(x_1))^2.
\end{split}
\end{equation}

$x_1, x_3$ are the two roots of 
$$
\frac{f(x) - f(x_1)}{x-x_1} = 
x^2 + (x_1- (\lambda+1)) x + x_1^2 - (\lambda+1)x_1 + \lambda.
$$
So, 
$$
(x_1 - x_3)^2 = \frac{-3 x_1^2 + 2(\lambda+1) x_1 + (\lambda+1)^2}4.
%\frac{(\lambda+1) - 3 x_1^2}{4}.
$$
And by the same way, we obtain
$$
(x_2 - x_4)^2 = \frac{-3x_2^2 + 2(\lambda+1) x_2 + (\lambda+1)^2}4.
%\frac{(\lambda+1) - 3 x_2^2}{4}
$$
and again
\begin{equation}
\begin{split}
(x_1-x_3)^2 (x_2 - x_4)^2 =  2^{-4} (\lambda^2 - \lambda + 1)^2.
\end{split}
\end{equation}

Altogether, 
\begin{equation}
\begin{split}
\det Q =& (x_3 - x_1)^2 (x_4 - x_2)^2 \Delta_\Sigma^2 \\
	=& (\lambda^2 - \lambda +1)^2 f(x_1) f(x_2) (f(x_2) - f(x_1))^4.
\end{split}
\end{equation}

At last we conclude that the period of $\nabla$ over $U$ is
the value
$$
\frac{2^4 \pi^2 f(x_1) f(x_2) }{(\lambda^2 - \lambda+1)^2(f(x_1)-f(x_2))^5}.
$$

We calculate now the values $f(x_1)f(x_2)$ and $f(x_2) - f(x_1)$:
\begin{equation}
\begin{split}
f(x_1) f(x_2) =& -\frac1{3^3} \lambda^2 (\lambda-1)^2,\\
\intertext{and}
(f(x_1) - f(x_2))^2 =& (f(x_1)+f(x_2))^2 - 4f(x_1)f(x_2) \\
	=& \frac{2^4}{3^6} (\lambda^2 -\lambda+1)^3.
\end{split}
\end{equation}

It follows that 
$$
f(x_1) -f(x_2) = \frac{2^2}{3^3}(\lambda^2 -\lambda+1)\sqrt{\lambda^2 - \lambda+1},
$$
for a suitable choice of the branch of the square root.

Finally, we get the main result of this chapter:
\begin{theorem}
The period determinant of $\nabla=d+dy$ over the affine Legendre elliptic curve $y^2 = x(x-1)(x-\lambda)$ defined for $\lambda \ne 0, 1, \frac{1\pm \sqrt{3}i}2$
 is
$$
-2^{-6}3^{12}\pi^2 \frac{\lambda^2(\lambda-1)^2}{(\lambda^2 - \lambda +1)^9
	\sqrt{\lambda^2 - \lambda+1}}
$$
in $\C^*/k^*$.
\end{theorem}





 
%\input{integral.tex}

\section{Exceptional case: $\lambda^2 -\lambda+1 = 0$}



Finally, we handle the case $\lambda^2 - \lambda + 1 =0$.
Recall $x_1= \frac{\lambda+1}3$ is the double root of $f'(x)$ as well as the triple
root of $f(x) - f(x_1)$.

As we have seen before, 
\begin{equation}
\pi_*(\nabla) \simeq (\sO_{\A^1},d+ dy)  \oplus  (\ker \Tr, \nabla_1 \otimes (d+dy)),
\end{equation}
where
the rank $2$ part has the connection
$$
d + I dy + 
\begin{pmatrix}
1 & - \frac23 (\lambda+1) \\
0 & 2 
\end{pmatrix} \omega'
$$
as calculated in Section 3.
Let us take $\nabla_D$ the above connection $\nabla_1\otimes (d+dy)$ for short.

Since $H^1_{dR} (\A^1, d+dy)$ is trivial,
the period of $\pi_*(\nabla)$ is equal to that of $(\ker\Tr, \nabla_D)$.
$(\ker \Tr,\nabla_D)$ is of rank $2$ and has two regular singularities at 
$y= \pm\sqrt{f(x_1)}$. 

Let $\Sigma := (x_1, \pm\sqrt{f(x_1)})$ in $U$.
Then we have the following short exact sequence:
\begin{equation}
0 \to H^0 (\Sigma, \sO_\Sigma ) \stackrel{\nabla}{\to} 
H_{dR}^1( U, \nabla_\Sigma) \to  H_{dR}^1 ( U, \nabla) \to 0,
\end{equation}
%where $\Sigma = \{ (x_1, \pm\sqrt{f(x_1)})\}$ in $U$.

Since $\rank H^1_{dR} (U,\nabla) = 4$ and $\rank H^0 (\Sigma, \sO_\Sigma) =2$,
the rank of $H^1_{dR} (U, \nabla_\Sigma)$ is $6$. 
Note that $\pi^*$ is the canonical isomorphism between $H^1_{dR}(U, \nabla_\Sigma)$
and  $H^1_{dR}(\A^1, \nabla_D)$.

In the other hand the homologies make the following short exact sequence:
\begin{equation}
0 \to H^{irreg}_1 (U, \nabla^*) \to H^{irreg}_1 (U, \Sigma, \nabla^*) \stackrel{\delta}{\to} 
H^{irreg}_0 (\Sigma, \nabla^*) \to 0.
\end{equation}


Therefore the period we want to calculate satisfies
\begin{equation}
{per} (U, \nabla_\Sigma) = {per} (\Sigma, \nabla) \cdot 
	{per} (U, \nabla).
\end{equation}

Now $per (U, \nabla_\Sigma)$ will be approximated as in the previous section.
For the approximation, we need to modify the cycles. 
Let $\gamma_1, \gamma_2$ a fixed path from
$0$ to $-\sqrt{f(x_1)}$ and to $\sqrt{f(x_1)}$ in $\A^1$, respectively. $\gamma_m$ is a path from 
$0$ to $-m$. Now define $I_1^{(m)} := \gamma_2 -\gamma_1$ and $I_2^{(m)} := \gamma_m - \gamma_2$. Note that $I_1^{(m)}$ is independent of $m$ and $I_2^{(m)}$ goes to infinity along
the rapid decay sector.

In the other hand,
let $\eta_i (e_q) $ be  
$\displaystyle{\frac{y^{i-1}}{y^2 - f(x_1)} dy\otimes e_q}$ for $i=1,2$, which generate
$H^1_{dR}(\A^1, \nabla_D)$. We still denote 
by $\Sol(\nabla, v)$ the solution of a connection $\nabla$ with an initial value $v$ at $0$.

It is the determinant of the matrix whose elements are the pairings
\begin{equation}
< I_j\otimes \exp y \Sol (\nabla_1^*, e^*_p) , \eta_i (e_q) > = \int_{I_j} <\Sol 
(\nabla_1^*, e^*_p), e_q > \exp y  \cdot \eta_i 
\end{equation}
for $p,q= 1,2$ and $i,j =1,2$.

Denote $\nabla_{(m)}$ by the connection $\nabla_1\otimes (d+ \frac{m+1}m dy)$.
Then the above integral is
\begin{equation}
\begin{split}
\lim_{m\to \infty} \int_{I^{(m)}_j}
	\frac{1}{m^m} <\Sol (\nabla_1^*, e_p^*), e_q> \frac{(y+m)^m y^{i-1}}{(y^2 - f(x_1))} dy\\
=  \lim_{m\to \infty} <\eta^{(m)}_i (e_p), I_j^{(m)}\otimes \Sol(\nabla_{(m)}, e^*_q)> 
\end{split}
\end{equation}
where $\eta^{(m)}_i (e_p)$ is the de Rham form $\frac{y^{i-1}dy}{(y^2 -f(x_1))(y+m)}\otimes e_p$.

Let us denote  its determinant  by $P_m$.

Applying the product formula, we obtain

\begin{equation}
\begin{split}
P_m \\
=& 
\frac{1}{m^{4m}\Delta_{(m)}^2}
(\nabla_{(m)}, y-\sqrt{f(x_1)})_{\gamma_1} \cdot (\nabla_{(m)}, y+\sqrt{f(x_1)})_{\gamma_2}\\
& \times
(\nabla_{(m)}, y+m)_{\gamma_m}  \cdot (\nabla_{(m)}, \frac1y)^{-1}_\infty \\
& \times \Gamma(\nabla_{(m)})_{- \sqrt{f(x_1)}} \cdot  \Gamma(\nabla_{(m)})_{\sqrt{f(x_1)}}
 \cdot \Gamma(\nabla_{(m)})_{-m}\cdot \Gamma(\nabla_{(m)})_\infty^{-1}.
\end{split}
\end{equation}

The values of the tame symbols in the above equality depend only on the determinant
connection of $\nabla_m$ and the (multivalued) solution of its dual connection:
\begin{equation}
\begin{split}
\det\nabla_{(m)} =& d + \frac{dy}{y-\sqrt{f(x_1)}} + \frac{dy}{y+\sqrt{f(x_1)}} + \frac{2(m+1)dy}{y+m} \\
\Sol(\det\nabla^*_{(m)}) =& (y^2 -f(x_1)) (y+m)^{2(m+1)}.
\end{split}
\end{equation}
Hence the tame symbols are
%\begin{eqnarray}
\begin{equation}
(\nabla_{(m)}, x - \sqrt{f(x_1)})_{\gamma_1} = 2\sqrt{f(x_1)} 
(\sqrt{f(x_1)} + m)^{2(m+1)},
\end{equation}
\begin{equation}
(\nabla_{(m)}, x + \sqrt{f(x_1)})_{\gamma_2} =  -2\sqrt{f(x_1)} (-\sqrt{f(x_1)}+m)^{2(m+1)},
\end{equation}
\begin{equation}
(\nabla_{(m)}, x + m)_{\gamma_m} = m^2 - f(x_1),
\end{equation}
%\end{eqnarray}
and
\begin{equation}
(\nabla_{(m)}, \frac1x)_\infty = 1.
\end{equation}

And the Gamma factors are
\begin{equation}
\Gamma_{\pm\sqrt{f(_1)}}(\nabla_{(m)}) = \Gamma(\frac13)\Gamma(\frac23),
\end{equation}
\begin{equation}
\Gamma_{-m}(\nabla_{(m)}) =  \Gamma(m+1)^2 ,
\end{equation}
\begin{equation}
\Gamma_\infty(\nabla_{(m)}) = \Gamma(\frac23+(m+1)) \Gamma(\frac43 + (m+1)).
\end{equation}

Altogether, we get the period $P_m$ as the product of the followings:
\begin{eqnarray}
\Gamma(\frac13)^2 \Gamma(\frac23)^2 \\
\frac{(m^2 -f(x_1))^{2m}}{m^{4m}} \to& 1 \\
\frac{m^2 - f(x_1)}{m^2} \to& 1\\
m^2 \frac{\Gamma(m+1)^2}{\Gamma((m+1)+\frac23)\Gamma((m+1)+\frac43)}
\to & 1.
\end{eqnarray}

From the functional equation satisfied by the Gamma function:
$$
\Gamma(z) \Gamma(1-z) = \frac{\pi}{\sin \pi z},
$$
we obtain
\begin{equation}
\begin{split}
\Gamma(\frac13)^2\Gamma(\frac23)^2  =& (\frac\pi{\sin\frac\pi3})^2 \\
	=& \frac{2^2 \pi^2}3.
\end{split}
\end{equation} 

Therefore altogether it follows
\begin{equation}
\begin{split}
per( U, \nabla_\Sigma) =& \Gamma(\frac13)^2 \Gamma(\frac23)^2 \\
	=& \frac{2^2 \pi^2}3.
\end{split}
\end{equation}

Since $1,y$ is a basis of $\Gamma(\sO_\Sigma)$
\begin{equation}
\begin{split}
per(\Sigma,\nabla) =&\det \begin{pmatrix} 1 & -\sqrt{f(x_1)} \\
	1 & \sqrt{f(x_1)} \end{pmatrix}\\
	= &2 \sqrt{f(x_1)}.
\end{split} 
\end{equation}

From $x_1 = \frac{\lambda+1}3$ and $\lambda^2 - \lambda+1=0$, it follows
$$
f(x_1) = \frac{2\lambda -1}9 = \frac{\pm\sqrt{-3}}9.
$$

Finally, we obtain the period
\begin{equation}
\begin{split}
per(U,\nabla) =&  \frac{per(U, \nabla_\Sigma)}{per(\Sigma,\nabla)} \\
	= & \frac{2^2/3\cdot \pi^2}{2/3 (-3)^{1/4}}  \\
	= & \frac{2\pi^2}{(-3)^{1/4}}
\end{split}
\end{equation}
in $\C^*/k^*$.

\begin{thebibliography}{99}
\bibitem{At} Atiyah, M. F.: \textit{Vector bundles over an elliptic curve},
	Proc. London Math. Soc. (3) {\bf 7} (1957) 414-452.
\bibitem{ab}  Andr\'e,~Y.,~Baldassarri,~F. : \textit{De Rham cohomology
	of differential modules on algebraic varieties}, Progress in Math. {\bf 189},
	Brikh\"auser, (2001).
\bibitem{bbe} A. Beilinson, S. Bloch and H. Esnault :
	\textit{$\sE$-factors for Gauss-Manin determinants, preprint, Essen}, 69 pages.
\bibitem{be}
	Bloch,~S., Esnault,~H. : \textit{Homology for irregular connections},
	math.AG/0005137, 16 pages.
\bibitem{bo} Borel et al.: \textit{Algebraic D-module, Perspectives in Math},
	Academic press.
\bibitem{Bou} Bourbaki, N.: \textit{\'El\'ements de math\'ematique, 
Alg\`ebre commutative}, Masson, Paris (1983).
\bibitem{C1} Chen, K.-T.: \textit{Iterated integrals and exponential homomorphisms},
	Proc. London Math. Soc. {\bf 4} (1954), 502-512.
\bibitem{C2} Chen, K.-T.: \textit{Integration of paths, geometric invariants and 
	a generalized Baker-Hausdorff formula}, Ann. of Math. {\bf 65} (1957),
	163-178.
\bibitem{de}
	Deligne,~P. : \textit{Equations diff\'erentielles \`a points singuliers r\'eguliers}, 
	Springer Lecture Notes in Math. {\bf 163} (1970)
\bibitem{DEL} Deligne, P.; Esnault, H.: Letter  to H.~Esnault, Apr. 22, 1999.
\bibitem{gm} Gelfand, S., Manin,~Y. : \textit{Methods of homological algebra},
	Springer. (1996)
\bibitem{gr} Grothendieck,~A. : \textit{On the de Rham cohomology of algebraic
	varieties}, Publ. Math. Inst. Hautes. \'Etudes Sci. {\bf 29} (1966) 95--103
\bibitem{ha} 
	Hartshorne,~R.: \textit{Algebraic Geometry}, Graduate Texts in Mathematics, Springer. {\bf 52} (1977)
\bibitem{hi}
	Hironaka,~H. : \textit{Resolution of singularities of an algebraic variety over a field of
	characteristic zero}, Ann. of Math. {\bf 79} (1964). I: 109--203; II:205--326.
\bibitem{j1} Jun, B. : \textit{Monodromies of algebraic connections on the trivial bundle},
	C. R. Acad. Sci. Paris, t. {\bf 331}, S\'erie I, 809--813 (2000).
\bibitem{j2} Jun, B. : \textit{A remark on the canonical decomposition of a connection 
	at an irregular singular point,} to appear in Abh. Mathe. Sem. Univ. Hamburg. 
\bibitem{ka} Katz, N. : \textit{Nilpotent connections and the monodromy theorem;
	applications of a result of Turrittin}, Publ. Math. IHES. {\bf 39},
	176--232 (1970).
\bibitem{kz} Kontsevich,~M., Zagier,~D. : \textit{Periods}, Mathematics unlimited---2001
	and beyond, Springer 771--808 (2001)
\bibitem{le} Levelt, A. : \textit{Jordan decomposition of a class of singular
differential operators}, Ark. Mat. {\bf 13--1}, 1--27 (1975).
\bibitem{ma}
	Malgrange,~B. : \textit{Equations diff\'erentielles \`a coefficients polynomiaux},
	Progress in Math., Birkh\"auser {\bf 96} (1991)
\bibitem{ma1} Malgrange, B.: \textit{Connexions m\'eromorphes 2 Le r\'eseau
	canonique}, Inv. math. {\bf 124}, 367--387 (1996).
\bibitem{ro} Robba,~P. : \textit{Lemme de Hensel pour les op\'erateurs
	diff\'erentiels}, Ens. Math. {\bf 26}, fasc. 3--4, 279--311 (1980)
\bibitem{st}
	Saito,~T., Terasoma,~T. : \textit{Determinant of period integrals}, J. of AMS, Vol. 10, No. 4,
	865--937 (1997)
\bibitem{GAGA} Serre, J.-P.: \textit{G\'eom\'etrie alg\'ebrique et g\'eom\'etrie 
	analytique}, Ann. Inst. Fourier. {\bf 6} (1955-1956), 1-42.
\bibitem{Si} Simpson, C.: \textit{Transcendental aspects of the Riemann-Hilbert
	correspondence}, Illinois J. of Math. {\bf 34} (1990), 368-391.
\bibitem{te0}
	Terasoma,~T.: \textit{Exponential Kummer coverings and determinants of hypergeometric
		functions}, Tokyo J. Math. Vol. 16, No. 2, 497--508 (1993)
\bibitem{te1}
	Terasoma,~T.: \textit{A product formula for period integrals}, Math. Ann. 
	{\bf 298}, 577--589 (1994)
\bibitem{te2}
	Terasoma,~T.: \textit{Confluent hypergeometric functions and wild ramification},
	J. of Alg. {\bf 195}, 1--18 (1996)
\bibitem{tu} Turrittin, H.~L.: \textit{Convergent solutions of ordinary linear
	homogeneous differential equations in the neighborhood of an irregular
	singular point}, Acta Math. {\bf 93}, 27--66 (1955).
\bibitem{va}
	Varchenko,~A. N.: \textit{The Euler beta function, the Vandermonde determinant,
	Legendre's equation and critical values of linear functions on a configuration of 
	hyperplanes. {\rm I, II}}, Math. USSR Izv., {\bf 35}, 543--571 (1990); ibid., {\bf 36},
	155--167 (1991)
\bibitem{wa}
	Warner,~F.W. : \textit{Foundations of differentiable manifolds and Lie groups}, 
		Scott, Foresman and Co. (1971)
\bibitem{ww}
	Whittaker, E.~T., Watson, G.~N. : \textit{A course of modern analysis}, Cambridge University
	Press (1963)
\end{thebibliography}

\end{document}
