Abstract
A Lamé connection is a logarithmic -connection over an elliptic curve , , having a single pole at infinity. When this connection is irreducible, we show that it is invariant under the standard involution and can be pushed down to a logarithmic -connection on with poles at , , and . Therefore the isomonodromic deformation of an irreducible Lamé connection, when the elliptic curve varies in the Legendre family, is parametrized by a solution of the Painlevé VI differential equation . The variation of the underlying vector bundle along the deformation is computed in terms of the Tu moduli map: it is given by another solution of , which is related to by the Okamoto symmetry (Noumi–Yamada notation). Motivated by the Riemann–Hilbert problem for the classical Lamé equation, we raise the question whether the Painlevé transcendents do have poles. Some of these results were announced in [6].
Export citation and abstract BibTeX RIS
In memory of Andrei Bolibrukh
Motivations
The classical Lamé equation (here in the Legendre form)
with , , determines a projective structure on the elliptic curve
having a Fuchsian singularity at the point at infinity. The projective charts on the affine part of the curve are local determinations of the maps
where and range over independent solutions of (1). Put . Then there is a local coordinate at such that one of the projective charts around this point is given by
- when ,
- or when , ,
- (regular) when .
The monodromy of any projective chart after analytic continuation along any loop is given by , where
is the projective monodromy representation of (1) computed in the basis . The following natural question goes back to Poincaré (see [1]): which topological representations
can occur as the monodromy of a Fuchsian projective structure?
Every Fuchsian projective structure on the once- punctured torus (that is, having moderate growth at the puncture) is of the above form: the parameter stands for the underlying complex structure, (or ) for the Fuchsian type of the puncture, and is an accessory parameter. The number of parameters fits with the dimension of the space of such topological representations up to conjugacy (see [2]). One thus expects that a generic representation should be the monodromy of some Lamé equation. The corresponding question for regular projective structures on complete curves has been answered only recently in [3] by pants decomposition and gluing methods. Our initial aim, from which the present work evolved, was to use the isomonodromy method to answer the Lamé case as a test. As we shall see, we actually reduce the question to the existence of poles for Painlevé VI transcendents, which looks difficult, though of a different nature.
The Lamé equation may be viewed as a logarithmic -connection on the trivial vector bundle over the elliptic curve , having a single pole at : every eigenvector of the residual matrix connection at provides a cyclic vector by going back to the scalar elliptic form. On the other hand, the Riemann–Hilbert correspondence asserts that every representation
is the monodromy of a logarithmic -connection on some vector bundle of rank over , having a single pole at . Our initial question takes the following form: given a topological representation (3), can we choose the complex structure of in such a way that the realizing connection is defined on the trivial bundle? Now the question fits perfectly into the setting of isomonodromic deformations.
Starting with a `Lamé connection' on , we consider its isomonodromic deformation arising when the complex structure of the curve varies (here the deformation parameter must be regarded as an element of the universal covering , which is the Teichmüller space of the once- punctured torus). If the pole of the Lamé connection is not apparent, then this deformation is unique and is characterized by the fact that its monodromy representation is locally constant. Equivalently, the deformation is induced by the unique integrable logarithmic connection on the universal curve with the following properties: this connection has a single pole along the section and its restriction to the initial fibre coincides with . Our initial question takes the following form: for which connections (or representations) can we ensure that the underlying vector bundle becomes trivial for some convenient parameter along the isomonodromic deformation of a Lamé connection?
In this paper we shall compute the variation of the vector bundle along the isomonodromic deformation of the initial Lamé connection. This variation is given by a solution of the Painlevé VI equation with appropriate parameters, and the bundle becomes trivial only when this Painlevé transcendent has a pole (other than , and ). We are finally led to the following question, which seems to be open: do Painlevé transcendents have poles?
Actually, much more interesting is the corresponding question for regular connections over curves of genus . In accordance with [4], regular projective structures correspond there to regular connections on the maximally unstable indecomposable -bundle (an -oper in the sense of [5], §2.7). The main result of [3] can be rephrased as follows: this special bundle occurs along the isomonodromic deformation if and only if the monodromy is irreducible and does not lie in (up to -conjugacy). Can we prove this directly by computing the variation of the bundle?
Another interesting question for is whether a given topological representation can be realized as the monodromy of a connection on the trivial bundle for an appropriate choice of the complex structure of ? In the case when the image of this representation is a discrete subgroup, this provides an embedding : the fundamental matrix of the associated linear system determines an equivariant map (where is the universal covering). The existence of compact curves in quotients of is still an open problem. Thus the isomonodromic approach yields a common geometrical framework for questions arising in various contexts.
Some of our results were announced in [6].
§ 1. The main result
Isomonodromic deformations of meromorphic connections on the Riemann sphere have been the subject of extensive studies (see [7], [8]). In this situation the underlying vector bundle is constant on a Zariski- open subset of the parameter space (see [9]), which enables one to compute the isomonodromy condition explicitly in the form of Schlesinger equations. Painlevé equations arise after further reduction in the simplest case, that of rank with poles. To observe continuous deformations of the underlying bundle, one must switch to connections over curves of genus . The simplest non-trivial case (regular connections of rank over an elliptic curve) was considered in [10], [11]. It was observed that the variation of the underlying line bundle along an isomonodromic deformation is a Painlevé transcendent. In the present paper we study the next most difficult case, that of logarithmic connections of rank with a single pole over an elliptic curve.
Throughout the paper, a Lamé connection is a pair consisting of a locally trivial holomorphic vector bundle of rank over an elliptic curve belonging to the Legendre family
and of a traceless logarithmic connection having (at most) a single pole at the point at infinity:
(here we identify the vector bundles , with the corresponding sheaves of holomorphic sections). From now on, such connections will be considered up to holomorphic bundle isomorphisms. The exponent of is defined (up to a sign) as the difference between the eigenvalues of the residual matrix of the connection at .
The underlying vector bundle of a Lamé connection has trivial determinant bundle because the connection is traceless. By the results of Atiyah [12], almost all vector bundles of rank with trivial determinant over are decomposable, that is, they take the form
The complete list is obtained by adding four extra bundles , . The set of semistable bundles consists of all decomposable bundles with , that is, all bundles of the form
together with the four indecomposable bundles. The corresponding moduli space is (see [13]) and the quotient map is given by
If we denote the -torsion points of by , , then is the unique non-trivial extension
and the moduli map identifies with the trivial extension . In particular, the point corresponds to both the trivial vector bundle and .
The isomonodromic deformation of a Lamé connection is defined as follows. Consider the universal covering (the Teichmüller space of the punctured torus) and the universal Legendre curve over this parameter space: its fibre at any point is precisely the curve (by an abuse of notation, we write for a point of and for its projection on ). The point at infinity determines a section of this fibration. For every Lamé connection on there is a unique flat logarithmic connection on the total space having the section as the polar set and inducing the initial connection on (see [7], [14], [15]). The deformation induced by the family is the isomonodromic deformation of . If the pole of is not an apparent singular point (that is, a point with local monodromy ), then is precisely the unique deformation having a constant monodromy representation. The exponent of the Lamé connection is constant along such a deformation. Finally, one can speak of the variation of the underlying vector bundle along the deformation: just consider the moduli map defined above. Here is our main result.
Theorem 1. Let be the isomonodromic deformation of an irreducible Lamé connection. Then the underlying vector bundle is semistable for a Zariski- open subset of the parameter space (see [15]) and its Tu invariant , which is defined by (7), is a solution of the Painlevé VI differential equation
with parameters , where is the exponent of the Lamé connection.
By `Zariski-open' we mean that the exceptional values of form a discrete subset of the parameter space . This property follows directly from [15].
It is already known that isomonodromic deformations of (generic) Lamé connections are parametrized by the Painlevé VI equation with parameters specified above. Namely, the isomonodromic deformation equations on the elliptic curve were computed directly in [16], [17], and the elliptic form of the Painlevé VI equation (see [18]) was recognized.
Our approach to these results is quite different. We first prove, using the Riemann–Hilbert correspondence, that every irreducible Lamé connection can be pushed down via the -fold covering to a logarithmic connection of rank with four poles on and the poles are the ramification values , , and of the covering. Thus we are back to the classical case of Fuchs: the isomonodromic deformation is parametrized by a solution (to be denoted by ) of the Painlevé VI equation with parameters . This already explains why the Painlevé VI equation arises in the Lamé case. So far, no computation is needed. Elementary birational geometry is used to go back to the initial deformation , and the moduli of the vector bundles can be expressed in terms of : we recognize in the image of under the Okamoto symmetry (see [19]). This automatically implies that is also a solution of the Painlevé VI equation, but with new parameters .
A similar assertion holds for the classical Painlevé VI interpretation in terms of the logarithmic -connections with four poles on , when we consider the parabolic bundle determined by the eigendirections of the residual matrix of the connection (see [20], [21]).
More generally, we can start with the isomonodromic deformation of a logarithmic connection of rank with four poles on , parametrized by any Painlevé VI transcendent . Then one can lift this deformation to the Legendre elliptic curve as a logarithmic connection of rank with poles at the ramification points in such a way that the moduli of the corresponding vector bundles are obtained from by the Okamoto symmetry . This provides a new geometric interpretation of this strange symmetry. We thus obtain in a natural way an isomonodromic deformation problem (a Lax pair) for the general elliptic form of the Painlevé VI equation, just by considering those traceless logarithmic connections of rank on (with poles at the second-order points ) that moreover commute with the elliptic involution . This was also considered in [22].
When we set , all Lamé connections with vanishing exponent are reducible, but regular ones can still be pushed down to . Our result remains valid in this case and we arrive at the following corollary.
Corollary 2. ([10], [11]) Let be the isomonodromic deformation of a regular connection of rank on the Legendre deformation , and let be the underlying line bundle, where . Then the function is a solution of the Painlevé VI equation with parameters .
In this case one can compute explicitly the variation of the line bundle by means of elliptic functions and obtain the following result.
Corollary 3. (Picard [23], see [24]) The general solution of the Painlevé VI equation with parameters is given by
where is the universal covering and the are the half-periods of .
The Painlevé transcendents described in Corollary 3 have poles if and only if either or is non-real.
Structure of the paper
The paper is self-contained in the sense that it contains several background/survey sections, namely:
§8 (appendix) contains basic facts about bundles, connections, moduli spaces, parabolic structures, elementary transformations;
§4 describes how the Painlevé VI equation arises as an isomonodromy condition for linear differential equations;
§5 gives a geometric and modular description of the phase portrait of the Painlevé VI equation, including Okamoto's space of initial conditions.
The original part of the paper is concentrated in the following sections.
§2: construction of Lamé connections by elliptic pullback of some systems on ;
§3: how to compute the Tu invariant of the bundle in terms of the parabolic structure of the downward system;
§6: why most of the Lamé connections are elliptic pullbacks (a Riemann– Hilbert approach);
§7: conclusion of the proof of the main theorem.
§ 2. Our main construction: elliptic pullback
Here we construct Lamé connections by lifting to the elliptic two-fold covering certain -connections having logarithmic poles at the critical values of . Later we will prove that all irreducible Lamé connections can be obtained in this way. This will be used to parametrize their isomonodromic deformations by means of Painlevé VI solutions in an explicit way.
Let us fix exponents and consider a logarithmic - connection on with poles at , , , and with prescribed exponents (that is, the eigenvalues of the residual matrix at are equal to for ). Such a connection will be called a Heun connection.
An important piece of data to be used later is the parabolic structure . It is defined by the eigenlines of the residue of at with respect to the eigenvalue , where . If does indeed have poles at each of the , the parabolic structure is perfectly well defined by the connection and the choice of exponents (they are defined up to a sign). However, it is important to allow non-singular points in our construction if we want to fit with the usual Painlevé VI phase space (see [25]). If and the corresponding point is non-singular, then every line is an eigenline and we have to choose one for our construction. The set of data with the properties above is called a parabolic Heun connection with parameter .
Example 4. If is a trivial bundle, then is defined by a Fuchsian system
The residual matrix at is equal to for , where is defined by
Restrictions on the exponents are given by the equalities , and the parabolic structure is given by .
To motivate the following construction, we point out that for the special exponents
it will provide a Lamé connection with exponent at infinity.
Step 1. We pull back the connection to the elliptic covering
This yields a logarithmic -connection on of the form
with poles at the points , , , of ramification and twice the initial exponents, . The parabolic structure corresponds to eigenlines with respect to the eigenvalues . The point has unipotent monodromy (resp. is non-singular) for if and only if the point is for . We have already noticed that for the exponents , , the corresponding singular points of are projectively apparent, that is, they have local monodromy . They will disappear in the next two steps.
Remark that we could choose an initial connection on with a single pole at so that its lifting will be of Lamé type. But the monodromy would then be trivial and the Lamé connection `very reducible'.
Step 2. We make a convenient birational bundle modification
such that the new connection
is still logarithmic and has poles at with eigenvalues and for . This is done by successively applying elementary transformations with respect to the parabolic structure over each singular point (see §8.10 for the definition and properties of ):
The new connection is not traceless: we have
and the trace is the unique logarithmic connection on having at each a pole with residue and trivial monodromy. Over the affine chart , is defined in a convenient trivialization of the line bundle by
Step 3. We now twist by an appropriate connection of rank 1 in order to restore the property of being traceless. To do this, we choose the unique square root of defined on the line bundle : is given over the affine chart by
The resulting -connection
has exponent at , . For the special parameters
is a Lamé connection (that is, having a single pole at ) with exponent (at ). We shall see that all irreducible Lamé connections can be obtained in this way. When is an odd integer, we note that is a unipotent singular point for if and only if is unipotent for . When is even, is always apparent for the Lamé connection (never unipotent non-trivial).
§ 3. Computing the vector bundle of an elliptic pullback
In the notation of §2 we would like to determine the vector bundle over the elliptic curve in terms of the initial connection . In fact, the construction of depends only on and the parabolic structure . For simplicity we restrict our attention to irreducible connections, although the general case could be handled by similar arguments. The goal of this section is to prove the following theorem.
Theorem 5. Let be an irreducible parabolic connection and let be its elliptic pullback. Then is semistable and one of the following cases holds.
1) The bundle is trivial and no three lines coincide. In particular, the cross-ratio
is well defined and we have
Precisely, is indecomposable if and only if
- either (the diagonal case),
- or and only two of the lines coincide (the other two being mutually distinct).
2) and none of the lines coincides with the line subbundle . Then either is trivial or coincides with the indecomposable bundle depending on whether or not all the lines lie on the line subbundle .
For the applications we have in mind, it should be emphasized that along the isomonodromic deformation of such a connection , the set of parameters for which is non-trivial is always a strict (possibly empty) analytic subset of the parameter space (see [15]).
We start the proof of Theorem 5 by justifying the restrictions on and . They result from the following lemma.
Lemma 6. Let be a curve of genus , a vector bundle, a logarithmic connection with a reduced effective divisor , and a line bundle which is not -invariant. Then the integer
is a bound for the number of those poles of where coincides with an eigenline.
Here we mean the eigenline of the residue of .
determines a homomorphism of line bundles and, therefore, a section
Since is not -invariant, is a non-trivial section vanishing at precisely the following points:
- •the non-singular points of where is stabilized: ;
- •the poles of where coincides with an eigenline of the residue of .
This gives the result.
In our situation, and . We deduce that for every line bundle (since is irreducible, no line bundle can be -invariant). Therefore is either trivial or equal to . In the former case, every embedding passes through at most two eigenlines. In the latter, passes through no eigenlines.
3.1. Projective bundles
To compute the modular invariant for the elliptic pullback, it is convenient to work with the associated projective bundle. Given any vector bundle over a curve , we denote the associated -bundle by . Another vector bundle gives rise to the same -bundle if and only if for some line bundle . When and are both determinant-free, is a -torsion point of the Jacobian. Hence there are at most determinant-free vector bundles giving rise to the same -bundle, where is the genus of . Line bundles are in a one-to-one correspondence with sections . The total space of the bundle is a ruled surface, and every section determines a curve on , to be denoted by for simplicity. The normal bundle of in is identified with the line bundle , where is the corresponding line bundle, so that the self-intersection is given by
Recall that is semistable if and only if for every line bundle . If is any line bundle distinct from , then the composite
is a non-trivial homomorphism of line bundles. It determines a non-trivial holomorphic section of vanishing at those points where and coincide. This yields a formula for the intersection number of the corresponding sections and :
A vector bundle is decomposable (that is, of the form ) if and only if admits two non-intersecting sections and (they correspond to and ) or, equivalently, two sections and having opposite self-intersection numbers. In this case the -bundle may be viewed as the fibrewise compactification of the line bundle obtained by adding a section at infinity. To be precise, is the zero section and is the section at infinity in .
If is an elliptic curve (for example, ) and , then is semistable if and only if has a section with zero self-intersection. The four indecomposable bundles defined in (8) correspond to the same -bundle , which is given by the unique non-trivial extension
The corresponding ruled surface is characterized by the existence of one and only one section with zero self-intersection. By the results of Atiyah [12] (see also [13]), all other semistable determinant-free vector bundle are decomposable: the corresponding -bundles are of the form
where are the two points of the fibre (see the definition (7)). We note that the modular invariant is determined by up to the action of the -torsion points of the elliptic curve : the determinant-free vector bundles with -bundle are the four semistable bundles with modular invariants
3.2. Ruled surfaces and elliptic pullback
Let us now describe the construction of §2 in terms of ruled surfaces. We consider a rational ruled surface equipped with the parabolic structure determined by a point on the fibre for .
Step 1. The elliptic ruled surface is obtained as the two-fold ramified covering branching along the four fibres . It makes the following diagram commutative:
We equip with a parabolic structure by putting for .
Step 2 (and 3). The birational transformation
is obtained by blowing up the four points and then blowing down the strict transforms of the four fibres. Step 3 is irrelevant from the projective point of view since it suffices to multiply by a line bundle in order to obtain . Therefore .
3.3. Elliptic ruled surfaces and elementary transformations
By Lemma 6, the parabolic ruled surface we start with is of the following form:
either and no three points lie on the same horizontal line;
or is the second Hirzebruch surface, that is, the total space of , and none of the points lies on the `negative' section corresponding to .
In either case we write for the corresponding elliptic pullback constructed in §3.2.
Proposition 7. When , choose a coordinate and consider the cross-ratio
where the points are defined by the formula . Then the following three assertions hold.
If , then is a decomposable ruled surface
where are the two points over .
If (all four points lie on the same irreducible curve of bidegree ), then is the indecomposable ruled surface .
If , then at least two points lie on the same horizontal line. Moreover, if the two other points lie on another horizontal line, then is a trivial bundle. Otherwise is the indecomposable ruled surface .
When , the following two assertions hold.
If all four points lie on a section with self-intersection (that is, on a section induced by an arbitrary embedding ), then is a trivial bundle.
Otherwise is the indecomposable ruled surface .
Proof. We first consider the generic case when and . Then one can choose the vertical coordinate in such a way that
One easily checks by computation that there is a unique curve of bidegree intersecting each fibre at the point with multiplicity . The equation of is of the form , where
The discriminant with respect to the variable is given by
Since it does not vanish identically and its roots are simple, the curve is reduced, irreducible and smooth. Its lifting to the trivial bundle splits into the union of two distinct sections, and , which intersect each other at exactly the four points without multiplicity. After elementary transformations with centres we obtain disjoint sections and of . This already means that is the compactification of a line bundle. To determine this line bundle, we consider the horizontal section of passing through . It intersects the -curve at two points ( and the point with coordinate ) and lifts to a section of intersecting over and, say, , and then intersecting over and , where are the two points of over . After elementary transformations with centres we obtain a section of intersecting at and at . Thus . This argument is summarized in Fig. 1.
When , the curve degenerates to twice the -curve passing through all the points , namely, the diagonal section . Its lifting to is the graph of the two-fold covering with self-intersection . After elementary transformations we obtain a section of having zero self-intersection. More precisely, the normal bundle of is trivial. Indeed, the section is induced in , , by the line bundle generated by its meromorphic section whose divisor is . The normal bundle of in is therefore given by
After elementary transformations we obtain that
If were decomposable, it would be a trivial -bundle, and we must now exclude this possibility. Consider the centres of the inverse elementary transformations. If were the trivial -bundle, most of horizontal sections would avoid the four points and would define, back on , a section with self- intersection . This is impossible. Thus is the indecomposable bundle .
We now assume that (the other cases or are similar). We are in one of the following three cases:
1) , and ;
2) , and ;
3) and .
The last case is easy since the three horizontal sections , and (which are defined by , and respectively) are transformed on the elliptic pullback into disjoint sections and and a third section that intersects at and , and at and . We promptly deduce that
We now study the first case, where only and coincide (this is similar to the diagonal case). The section of induces a section of having a trivial normal bundle. If were decomposable, it would be the trivial bundle, and a generic constant section would provide a section of the trivial bundle having self-intersection , a contradiction. Thus is the indecomposable bundle .
We finally consider the case when . As above, we easily see that the exceptional section , which is induced by , yields a section of having a trivial normal bundle. Again, as in the diagonal case, consider the centres of the elementary transformations inverse to ; they are contained in . If is the trivial bundle, then its constant sections give rise to a pencil of sections of having the four points as base points. A special member of this pencil is given by the union of and the four fibres over the points , and the pencil itself consists of all sections of passing through the four points and disjoint from (plus the special one). The elliptic involution permutes these sections, and so acts on the parameter space . This action has at least two fixed points, namely, and another section that can be pushed down as a section of . By construction, passes through all the points and does not intersect ; it is a -curve, as required. Conversely, when all lie on a -curve , we obtain a second section of , which yields a trivialization. We note that the pencil considered above comes from a pencil not of sections of , but of curves that intersect each fibre twice.
§ 4. Isomonodromic deformations and the Painlevé VI equation
In this section we recall how isomonodromic deformations of logarithmic - connections over the 4-punctured sphere are parametrized by Painlevé VI solutions, and how we can use this parametrization to compute the variation of the bundle of the corresponding elliptic pullback. We first recall what an isomonodromic deformation is.
4.1. Isomonodromic deformations and flat connections
Let be a complex projective curve and let be a vector bundle of rank over equipped with a (flat) logarithmic -connection with a reduced effective polar divisor . For every topologically trivial analytic deformation of the punctured curve there is a unique deformation of the vector bundle and the connection such that the monodromy data remain constant. This follows from the Riemann–Hilbert correspondence (Proposition 24 with parameters). The monodromy data consist of the monodromy representation complemented by the `parabolic structure' at apparent singular points. Equivalently, if we denote the total space of the deformation by and the corresponding (smooth) divisor by , then the deformation is induced by the unique flat logarithmic - connection with polar divisor inducing on the slice .
In this paper we consider the case of the 4-punctured sphere and the once- punctured torus. Their deformations are parametrized by the corresponding Teichmüller spaces that are both isomorphic to the Poincaré half-plane . To be precise, we start with the isomonodromic deformation of a logarithmic - connection over the Riemann sphere with poles at , , and , where ranges over the universal covering . Consider the deformation of the corresponding elliptic pullback constructed in §2. We easily see that is still an isomonodromic deformation of some logarithmic connection over the Legendre family of elliptic curves with poles contained in the ramification locus of the elliptic curve. The parameter space is now understood as the Teichmüller space of the torus. For the special parameters we obtain the isomonodromic deformation of a Lamé connection.
4.2. The Painlevé VI equation and Fuchsian equations
Although we do not really need it, it is interesting to recall how the Painlevé VI equation was originally derived as an isomonodromic equation for Fuchsian projective structures on the 4-punctured sphere with one extra branch point. After normalizing the singular points as , , and by a Möbius transformation, the corresponding Fuchsian second-order differential equation takes the form
Here is the branch point, is the local exponent at , and is fixed by the relation
Note that the parameters and are residues of :
The singular point has exponent . This point is apparent (that is, a branch point of the projective chart) if and only if the parameter is given by
Under these assumptions, the local charts , where and range over independent solutions of (13), form a projective atlas on the complement of , , , and in the Riemann sphere. At each of the singular points one of the projective charts takes the form (or, possibly, in the case when ) for a convenient local coordinate at . At the point one of the projective charts takes the form (a simple branch point). Conversely, every projective structure on the Riemann sphere having five singularities with moderate growth, one of which is a simple branch point, is conjugate by a Möbius transformation to an element of the family above. Such projective structures are characterized by the following data:
the positions and of the singular points,
the exponents ,
the monodromy representation of a projective chart up to conjugacy.
A small deformation of an equation (13) with moving singular points and is said to be isomonodromic if the projective charts have constant monodromy representation (up to conjugacy). Such deformations are characterized by the following classical theorem.
Theorem 8. (Fuchs–Malmquist) A deformation of (13) parametrized by the position of the singular point is isomonodromic if and only if the exponents remain fixed and the parameters satisfy the non-autonomous Hamiltonian system
The first Hamiltonian equation (17) is of the form
Substituting (18) in the second equation (17), we obtain the Painlevé VI equation (9) with parameters . From a chronological point of view, the Painlevé VI equation was first derived by Fuchs, and the Hamiltonian form was discovered later by Malmquist.
4.3. The Painlevé VI equation and Fuchsian systems
Let us now recall how Painlevé VI solutions correspond to isomonodromic deformations of logarithmic -connections with singular points , , and over the Riemann sphere. Let be such a deformation. Assume that it is irreducible (this depends only on the monodromy and not on the value of ). By Lemma 6, the underlying bundle is either trivial or equal to . It turns out that must be trivial for all values of outside a discrete subset of : there are no non-trivial irreducible isomonodromic deformations of such connections on the bundle (see [15]). It is thus enough to consider isomonodromic deformations of -Fuchsian systems
The residual matrix at the singular point is given by
We denote the eigenvalues of by . Then
After a change of variables , where , we normalize
We exclude the case when : the connection must be singular at . Then the following theorem holds.
Theorem 9. A small deformation of the system (10), normalized by (22), is isomonodromic if and only if the eigenvalues are constant and the function satisfies the differential equation
and the Painlevé VI equation (9) with parameters
We first deduce Theorem 9 from the Fuchs–Malmquist theorem (Theorem 8).
Proof. The vector is an eigenvector with eigenvalue at . Since the system (10) is irreducible, this vector is not invariant and can be chosen for a cyclic vector to derive a scalar Fuchsian equation. Namely, if is a solution of (10), then the function
satisfies the scalar equation with exponents
and parameters
In fact, the Darboux coordinates have the following interpretation. The point is the unique other point at which is again an eigenvector of the system (10). The corresponding eigenvalue is
It follows from Theorem 8 that a deformation of the system (10) is isomonodromic if and only if the auxiliary variables and defined by (25) and (26) satisfy the Hamiltonian system (17) with the non-autonomous Hamiltonian defined in (16).
Let us now explain how to reconstruct the system (10) uniquely (up to a gauge transformation) from a solution of the Painlevé VI equation. We first introduce an auxiliary variable by the formula (18). This gives us a unique scalar equation (13) from which one can reconstruct a Fuchsian system by the standard method. The resulting system is defined (up to a gauge transformation) by the formula (10) with equations (21) and
where and is defined by (14). The coefficients of this system can immediately be deduced from the equations (21). The standard formulae given by Jimbo and Miwa (see [26], pp. 199, 200) assume that so that the matrix can be further normalized as a diagonal matrix by an additional gauge transformation. The resulting formulae are much more complicated than those above. The way we obtain the formulae (27) is described in §§5.3–5.5.
4.4. The vector bundle of an elliptic pullback
Coming back to our initial problem and wishing to apply Theorem 5, we would like to parametrize the parabolic structure induced by (or, equivalently, by the system (10)) in terms of the Painlevé VI transcendent that parametrizes the deformation. We deduce from (21) and (27) that the eigenline associated with the eigenvalue over the pole is given by
whence
(recall that was normalized by (22)). These expressions for no longer hold for a gauge-equivalent system (10) (for example, through the Jimbo–Miwa normalization), but the cross-ratio
depends only on the system (22) up to gauge transformations. We note that the formula (29) gives an elegant definition of the auxiliary variable in terms of the parabolic structure of the connection, and .
Corollary 10. Let be the isomonodromic deformation defined by the Painlevé VI solution as above, and let be the elliptic pullback of this deformation. Then the bundle is semistable and has invariant
Proof. By construction, the bundle is trivial. By Theorem 5, is semistable and has invariant
All computations above hold only under the generic assumptions that . On the other hand, it is well known that constant solutions , , or correspond to isomonodromic deformations of reducible connections. Thus Corollary 10 is enough for proving Theorem 1. However, we can be more precise and check for special values of and whether is indecomposable or not.
§ 5. The geometry of the Painlevé VI equation
Here we introduce some moduli space of -connections with poles at , , and and eigenvalues . It contains all irreducible connections. This space was originally used in Okamoto's work [27] to construct a good space of initial conditions for the Painlevé VI equation from which the Painlevé property can be read off geometrically. In [25] this space was identified with a moduli space of connections, extending the dictionary established in §4.3. We recall this construction and then use it to determine the vector bundle of the elliptic pullback for special values of and .
5.1. Okamoto's space of initial conditions
The Painlevé property, which characterizes the Painlevé equations among all differential equations of the form , says that all Painlevé VI solutions can be analytically continued as meromorphic solutions along any path avoiding , and . Painlevé VI solutions become meromorphic and global on the universal covering of the 3-punctured sphere.
The (naive) space of initial conditions for the Painlevé VI equation at some point fails to describe all solutions in a neighbourhood of . Painlevé VI solutions are meromorphic and some of them have a pole at ; we have to add them. The good space of initial conditions is
To construct it, Okamoto considers the phase portrait of the Painlevé VI equation in the variables
(introducing an auxiliary equation ): it is defined by a rational vector field that determines a singular holomorphic foliation on any rational compactification. For example, we can start with and observe that the singularities of the foliation are located on the special fibres (these do not concern us) and at the infinity of the -factor. The latter are degenerate and lie along a one-dimensional section of the -projection; we must blow up this section in order to reduce the degeneracy of the singular points. After nine successive blow-ups, Okamoto obtains a fibre bundle
(we just ignore what happens over ) which is not locally trivial as an analytic bundle (but is as a topological bundle). There is a non-vertical divisor consisting of vertical leaves (with respect to the -projection) and singular points. On the complement of this divisor, the Painlevé foliation is transversal to and induces a local analytic trivialization of the bundle
By construction, the fibre over any point may be interpreted as a set of germs of meromorphic -solutions. Actually, for special parameters , there are leaves staying at the infinity of the affine chart . These cannot be regarded as meromorphic solutions. They should rather be viewed as `constant solutions '. The divisor actually coincides with the reduced polar divisor of the closed -form defined in the affine chart as
where is defined by (16). The kernel of this -form determines the Painlevé fibration.
We now describe the parameter space starting with the Hirzebruch ruled surface . Define a reduced divisor as the union of the section having self-intersection and the four fibres over , , and . Next, we fix two points on each vertical component of , none of which lies on the horizontal component. Blowing up these eight points, we obtain the compact space . Let us preserve the notation for the strict transform of this divisor. The complement is the space of initial conditions. It remains to define the position of these eight points as a function of and (see §5.6).
5.2. Projective structures and Riccati foliations
We go back to the approach of Fuchs, where the Painlevé transcendents parametrize isomonodromic deformations of Fuchsian projective structures with singular points (see §4.2). Such a structure can be defined by a Fuchsian second-order differential equation (13). One can also determine it by the data consisting of a logarithmic -connection together with a line subbundle which is not -invariant and plays the role of a cyclic vector (see §4.3). In [5] these data are referred to as `-opers'. A more geometrical picture inspired by the works of Ehresman uses the notion of a `projective oper'. This is a triple , where is the associated -bundle, is the induced projective connection and is the section corresponding to . For example, the system (10) determines the Riccati equation
on a trivial -bundle by the formula , where
We prefer to consider the associated phase portrait, that is, the singular holomorphic foliation induced by this equation on the ruled surface and also called a Riccati foliation (see [28] or §8.8). Its singular points are located at the poles of the Riccati equation. Precisely, in the notation (21), the singular points are
From the point of view of foliations 1 we say that and have exponents and respectively. They correspond to the eigenlines of the system with eigenvalues and (pay attention to the sign). If , then either the singular point of the system is logarithmic and the two singular points of coincide, or the singular point of the system is apparent and the Riccati foliation is non-singular. In the latter case we shall introduce an additional parabolic structure (see below). The section defined by plays the role of a cyclic vector. It has two tangencies with the Riccati foliation, namely, at and (where passes through a singular point of the foliation), see the bottom diagram in Fig. 2.
Download figure:
Standard imageIn this picture the foliation is regular, transversal to the -fibre over a generic point and therefore transversely projective (see [29]). Thus it induces a projective structure on the section that projects to the base . Clearly, the resulting projective structure on (a Zariski- open subset of) is preserved by birational bundle transformations, and it is natural to look for the simplest birational model. In the triple above, the point artificially plays a special role since we require in the normalization (22) that the section passes through the singular point of the foliation. It is more natural to apply an elementary transformation at this point and obtain the following more symmetric picture (see [30]): a Riccati foliation on the Hirzebruch ruled surface having singular points over , , and , and the section now has a simple tangency with the foliation at the point . In fact, is the unique negative section (that is, with self-intersection ). The exponents (eigenvalues) of the foliation are now given by
Over each pole of the Riccati equation, the foliation has one or two singular points depending on the exponent ; when , one of the singular points accidentally lies on (see the middle diagram in Fig. 2).
We now add a parabolic structure which will be convenient in our further application of elementary transformations. It is a necessary tool in order to get a smooth moduli space when one of the exponents vanishes.
The Okamoto space of initial conditions can be viewed as the moduli space of such Riccati foliations . To be precise, we fix the exponents and the parameter and consider the data , where
is the Hirzebruch ruled surface equipped with the ruling structure ,
is a regular Riccati foliation on which is transversal to the ruling structure outside ,
over each , either the fibre is invariant and there is one singular point for each exponent , or and the foliation is regular and transversal to the ruling,
, where either is a singular point with exponent over , or , is regular over and is any point of the fibre.
Such data correspond exactly to projectivizations of semistable parabolic connections, which were used in [25] to construct the moduli space .
5.3. Riccati foliations on
Putting in the scalar equation (13), we obtain the Riccati equation
Recall that and is defined by the formula (16). The equation (34) defines a Riccati foliation , which naturally compactifies on the Hirzebruch surface given by two charts
with transition map . The singular points of the foliation lie on the five fibres:
with corresponding exponents (up to a sign)
To be precise, the singular points in the first chart are given by
Here (resp. ) is the singular point with exponent (resp. ) and . At , the singular points are given in the chart by the equations
The `cyclic vector' is given by the section defined in these charts by and respectively.
5.4. Riccati foliations on
After an elementary transformation centred at the nodal singular point of the section we obtain a Riccati foliation on the Hirzebruch surface with poles , , , and exponents , , , respectively, as in §5.2. The apparent singular point has disappeared.
If we define by the usual charts
with transition map , then the negative section is given by the formulae and respectively and the Riccati foliation is induced by the equation
This equation is obtained from (34) by setting . The singular points in the first chart are now given by
and the singular points at are given in the chart by
(here again has exponent ).
5.5. Riccati foliations on
Finally, by making the last elementary transformation centred at the singular point of , we obtain a Riccati foliation with eigenvalues
The underlying ruled surface depends on the relative position of the singular point of with respect to the negative section : the Riccati foliation is defined on
when (generic case),
when (codimension case).
In particular, in the latter case.
If , then the Riccati equation defining the foliation on can be obtained from (37) by setting . We obtain the equation
This is precisely the projectivization of the Fuchsian system in (27). The singular points with eigenvalues correspond to the parabolic structure defined by the formulae (28):
and
5.6. The moduli space
With each pair we have associated a Riccati foliation on the Hirzebruch surface with poles , , , and exponents .
Conversely, let be such a Riccati foliation on with poles , , and . Then the unique tangency between the negative section and the foliation uniquely determines . Now, if , then we perform an elementary transformation at this tangency point and thus define a Riccati foliation on . There is a unique section with self-intersection passing through the singular points , and of . Choose a chart such that the section and the negative section are given by the equations and respectively. In fact, is uniquely determined if we normalize the -coefficient of the Riccati equation as in (34). We then observe that all singular points and for depend only on the parameters and ; the singular point given by is the only one depending on the particular foliation . Define by the formula . We note that the Riccati foliation is characterized by the position of the nodal singular point of on the Hirzebruch surface (after the normalization above).
Theorem 11. [25] Fix the parameters
and consider Riccati foliations on the Hirzebruch surface having at most simple poles at , , and , with exponents over such that the negative section of is not -invariant (the semistability condition). Add moreover a parabolic structure which consists of the following data over each point :
either of the singular point with exponent ;
or, when and there is actually no singular point, of any point of the fibre (the only case when the parabolic structure is relevant, that is, not determined by the foliation itself).
The moduli space of such pairs up to bundle automorphisms is a quasi- projective rational surface that can be described as follows. Start with the Hirzebruch surface and write for the reduced divisor consisting of the four fibres together with the negative section . Define eight points and by the formulae
in the first chart and
in the chart . Carry out the following operations for each .
If , then blow up the two points and .
If , then successively blow up the point and the point where the exceptional divisor meets the strict transform of the fibre .
Denote the resulting eight-point blow-up of by . Then the moduli space is the complement in of the strict transform of the divisor .
The eight points and in this theorem are nothing but the singular points of the foliation on described in §5.3. When the nodal singular point tends to (resp. ), the singular point (resp. ) of the corresponding foliation on (see §5.4) tends to the negative section . The exact limit depends on the way in which the nodal point tends to or . If , then the fibre of over consists of two disjoint copies and of the affine line which contain and respectively: they stand for the moduli spaces of those foliations whose singular points (resp. ) lie on the negative section . If , then the fibre of over is the union of an affine line and a projective line that intersect each other transversally at one point and are projected to the point . The component consists of those foliations whose point lies on the negative section . The compact component consists of those parabolic foliations which actually have no pole over . If we neglect the parabolic structure, then the rational curve blows down to a quadratic singular point.
Sketch of the proof of Theorem 11. Let be a Riccati foliation on having at most simple poles over , , , such that the negative section is not invariant (semistability). In the standard chart the foliation is given by
where and the coefficients and do not vanish simultaneously. Bundle automorphisms are given by changes of coordinates of the form , , . The single tangency between and the negative section is given by . By making the change of coordinates , one can achieve the following normalization:
where . We first assume that , whence we can assume that . Using a change of coordinates of the form , we can further assume that . Incidentally, assuming that , we obtain a unique normal form
It follows from (37) that
The residue at is found from the formula
and the exponents are determined by the discriminant
We similarly get
Once the parameter is fixed, one can uniquely determine , , and as functions of , and (that is, ) provided that . In a neighbourhood of one can still express , and as functions of , and , and the moduli space of such foliations is locally isomorphic to the surface
We promptly see that the moduli space consists, over , of an affine line parametrized by over each of the points .
If , then the graph
is clearly obtained by blowing up the two points and then deleting the level , that is, the strict transform of . The formula (46) for the residue shows that the following conditions are equivalent over :
- (resp. ),
- (resp. ,
- the singular point (resp. ) of the corresponding foliation lies on the negative section .
If , then the graph
is obtained after two blow-ups, and the affine line parametrized by is the set of all foliations whose singular point lies on the negative section . The surface with equation has a singular point at . This corresponds to the case when the residue (46) vanishes, that is, actually has no singular point at . The parabolic data at provide a desingularization of the surface. Indeed, the moduli space of pairs is locally parametrized in a neighbourhood of by the set
(or it would be better to write , where ). The parabolic data parametrize the exceptional divisor (thus realizing the blow-up). The intersection is given by and describes the foliation without singular points at whose parabolic structure lies on the negative section .
The study of is similar except that one must choose another normalization for :
where is the other chart of .
The deformation is analytically (but not algebraically) trivial and the trivialization is given by the Painlevé flow (see [31], [25]). The good phase space for the Painlevé VI equation is the (locally analytically trivial) fibration
The map is regular (when none of the vanish) and endows (for fixed) with the structure of an affine -bundle with double fibres over . Finally, we note that the formula (37) defines an explicit section (universal Riccati foliation)
over .
5.7. The moduli space of -connections
By §5.5 we have an isomorphism
from the previous moduli space to the moduli space of -connections with exponents
We recall that for every point of this moduli space the underlying vector bundle is either trivial or equal to . It is actually trivial on a Zariski- open subset of . The locus of the non-trivial bundles is Malgrange's theta divisor defined for fixed by the exceptional divisor arising from the blow-up of . Indeed, the points of the last divisor correspond to those foliations for which lies on the negative section ; applying gives a foliation on the Hirzebruch surface . In fact we have in this case: we are back to the foliation of §5.3, where . To be precise, putting
in (13) and letting tend to , we obtain the Heun equation
For our special choice of the parameters we obtain the Lamé equation (1) with and .
Finally, the open set (resp. the closed subset ) may be viewed as the moduli space of Riccati foliations on (resp. ) with simple poles and exponents , excluding those foliations for which the section passing through (resp. the exceptional section) is totally -invariant.
5.8. Okamoto symmetries
There are many birational transformations
that induce biregular diffeomorphisms between moduli spaces
equivariant with respect to the projection . They were studied in [32]. Some of them are the classical Schlesinger transformations arising from geometrical transformations of connections (resp. Riccati foliations); together with a strange extra symmetry, they generate the full group of Okamoto symmetries.
5.8.1. Change of signs
First, we can change the `spin structure', that is, the signs of the parameters . This does not change the Riccati foliation or the coefficients , , of the normal form (45), but the variable is modified as follows:
5.8.2. Permutation of poles
One can permute the four poles of the Riccati foliation. The resulting group of order is generated by the elements
(the change of variable is given by , and respectively; we get , and respectively).
Together with sign changes, we already get a linear group of order acting on our moduli space.
5.8.3. Elementary transformations
Let be a Riccati foliation on representing a point of . An elementary transformation centred at one of the singular points (say, ) of gives a new Riccati foliation with simple poles over , , , and with the shifted parameters . The resulting bundle is either the trivial bundle , or the Hirzebruch surface . However, after two (or, more generally, an even number of) such elementary transformations we are back to . Indeed, since the type of the Hirzebruch surface shifts by under each elementary transformation, we must only exclude the possibility that, for example,
This would mean that each of the two successive elementary transformations has centre on the negative section. In this case, the negative section of is the strict transform of . Since is not -invariant, we obtain that is not -invariant. But then Proposition 26 (see §8.8) gives a negative tangency, a contradiction. Thus we have defined a biregular transformation
Omitting the huge formula, we note that is given by the unique tangency point between and the only section of that has self-intersection and passes through and .
More generally, given any quadruple
we construct a biregular transformation
where
with the convention that when . As above, there is a unique section with self-intersection and tangency of multiplicity with the foliation at each point ; the extra tangency between and is at .
We now get an infinite affine group of transformations. We denote it by .
5.8.4. The Okamoto symmetry
To generate the full group of biregular transformations described in [32], we need an extra symmetry,
(called in [19]) or any of its conjugates. So far there is no geometric interpretation of this symmetry as long as we interpret as the moduli space of rank- connections (or Riccati foliations). To derive the full Okamoto group from natural transformations of connections, one has to deal with isomonodromic deformations of connections of rank (see [26]) or more (see [19]).
The conjugate of the Okamoto symmetry above by the change of signs is equal to in the notation of [19] and is given by
We recognize in the Tu invariant of the underlying vector bundle of the elliptic pullback (see Corollary 10).
5.9. Special configurations
For special values of the parameter the moduli space contains complete rational curves (independently of ). They arise from the curves on avoiding the negative section and passing through or for each value of . They correspond either to the locus of reducible connections or to the locus of connections with an apparent singular point. It turns out that there are no other complete curves in (see [25]). In particular, there are no such curves for generic values of . Here are some examples.
Suppose that and the foliation has an apparent singular point at . Then the pole disappears after one elementary transformation centred at . We thus obtain a Riccati foliation of hypergeometric type on (it cannot be by Proposition 26), that is, a foliation with poles at , , and exponents . Conversely, a foliation as above can be recovered from by an elementary transformation at any point of the fibre . This gives us a rational family of foliations (parametrized by the fibre ). The corresponding rational curve in the moduli space is given by the equation
This is the only curve on with the following properties:
has degree ;
does not intersect the negative section ;
intersects the fibre at both and for ;
intersects the fibre twice at ;
is singular at : it has two smooth branches.
Proof. To compute this family in the moduli space, we start with the Riccati foliation that can be normalized to
(we exclude some reducible cases here) and choose a parabolic structure over . The horizontal section is sent by the elementary transformation to the negative section: the corresponding value corresponds to the unique tangency point between and . We have already noted that the map has degree . We now claim that the foliation determines a point over (resp. ) in the moduli space if and only if the parabolic structure and the singular point (resp. ) of lie on the same horizontal section, namely, on . Indeed, this section is transformed by into the negative section of . Thus the curve passes once through each of the points , . The same holds over and . Finally, when is an apparent singular point for , it cannot lie on the negative section (otherwise an application of would give a hypergeometric foliation on , having by Proposition 26 a tangency of multiplicity with the negative section, a contradiction). We can now compute the equation of the curve . Since has degree and does not intersect the negative section of , the equation of takes the form , where and are polynomials of degrees and respectively. The condition that passes through all the points and , except , determines not only , but a pencil of curves; they are all smooth and vertical at (and thus escape from over this point), except one of them which has a singularity with normal crossing at .
In the case when , the foliations with apparent singular points at are obtained as follows. Take the hypergeometric foliation on with exponents , choose a parabolic structure at and apply twice. We again get a rational curve in the moduli space which projects down to a curve with the following properties:
has degree ;
does not intersect the negative section ;
intersects the fibre at both and for ;
intersects the fibre three times at and once at .
5.10. Bolibrukh–Heu transversality
A remarkable result of Bolibrukh ([9], §5.2, Proposition 5.6) asserts in our context that the isomonodromic deformation of an irreducible -connection is `mostly' defined on the trivial bundle.
Theorem 12. (Bolibrukh) Let be a local isomonodromic deformation. Then one of the following cases holds.
The underlying bundle is trivial outside a discrete subset of the parameter space .
and the destabilizing subsheaf is -invariant.
In particular, the former case holds when is irreducible. Bolibrukh proved a more general result for certain logarithmic connections of arbitrary rank on the Riemann sphere. The rank case was considered by Heu [15] in full generality (for regular or irregular -connections on an arbitrary Riemann surface). We will use the following corollary.
Proposition 13. Let be a local isomonodromic deformation of an -connection. Assume that is trivial and two eigenlines and , , coincide along the whole deformation. Then the connection is reducible: the constant line bundle determined by the lines is -invariant.
Proof. Applying to the deformation, we obtain an isomonodromic deformation on the Hirzebruch surface . By Theorem 12, this is possible only when the connection is reducible.
Remark 14. The following more general result holds. Consider an irreducible Riccati foliation in defined on the trivial bundle . The sections with self-intersection ( even) form a -dimensional family. The smooth curve has exactly tangencies with counting multiplicities. Then the tangency locus can be completely contained in the fibres over only for isolated points of the parameter space . When , this yields the previous proposition. When , we obtain, for example, that all four parabolics cannot lie on a curve of bidegree along the deformation.
§ 6. Lamé connections
The aim of this section is to give a rough description of the moduli space of Lamé connections up to biregular bundle transformations, using the Riemann–Hilbert correspondence. Here we fix an elliptic curve
We will say which Lamé connections are invariant under the elliptic involution
In §7 we will see that -invariant Lamé connections can be pushed down by the double covering
to logarithmic connections with poles at the four ramification points .
Let be a Lamé connection on the elliptic curve , thus having a simple pole at . When the exponent is not an integer, the connection can be reduced to the following matrix form:
where is any local coordinate of at , and is a convenient local holomorphic trivialization of . On the other hand, if (say, ), then the pole is said to be resonant and the matrix form can be reduced (by a local gauge transformation as above) to
The point is said to be an apparent singular point for (actually regular when ) in the former case, and a logarithmic singular point in the latter.
The connection is regular on the affine part of and inherits a monodromy representation
which is well defined by up to -conjugacy. We fix a loop which goes to , turns around once, and comes back to the initial point. Then the matrix is called the local monodromy of around . It is conjugate to
All of this is obviously independent of the choice of the base point for the fundamental group. We note that the singular point is apparent if and only if the local monodromy is equal to , that is, it belongs to the centre of ; this can occur only when .
6.1. The Riemann–Hilbert correspondence
For every exponent we have an analytic map
which sends every Lamé connection (considered up to holomorphic bundle isomorphisms) to its monodromy representation (considered up to -conjugacy). This map is almost one-to-one: it is surjective, and its restriction to the set of connections without apparent singular points (that is, connections with ) is injective.
If has an apparent singular point at , where , then all horizontal sections have meromorphic extension to . Holomorphic sections of this kind are contained in a line subbundle (say, ) of near . In terms of the normal form (47), is the constant line bundle generated by . Except in very special cases, cannot be extended to a line bundle on the whole of : it is only defined in a neighbourhood of . The fibre of at coincides with the eigenline of the residual matrix associated with the positive eigenvalue . By the Riemann– Hilbert correspondence over the punctured curve , any two Lamé connections with the same monodromy representation are conjugate to each other by a gauge transformation over . This conjugacy extends to a global gauge transformation if and only if it conjugates the corresponding local line bundles as defined above. One can restore the injectivity of the Riemann– Hilbert map in the following way. Regard the monodromy representation as an action of the fundamental group on the space of germs of solutions at the base point . Then analytic continuation of local holomorphic solutions at to the base point along (half-) yields a one-dimensional subspace . In other words, is obtained by analytic continuation of (as a -invariant line bundle) along . The Lamé connection (with an apparent singular point) is characterized by the pair
up to conjugacy:
This is a kind of parabolic structure for the space of representations.
6.2. The Fricke moduli space
We recall how to describe the moduli space of representations following Fricke (see [2]). Fix standard generators of the fundamental group in such a way that their commutator represents a small loop turning once around the puncture, as before. We neglect the base point since it plays no role in our discussion. A representation is determined by the images of generators,
The ring of polynomial functions on that are invariant under -conjugacy, is generated by the functions
For example, the trace of the commutator is given by
whence we have
The geometric quotient of by the action of - conjugacy is identified with by means of the composite
To be precise (see [33], [2]), if , then the fibre over consists of the single -conjugacy class of the irreducible representation defined by the matrices
This normal form is obtained in any basis of the form , where is an eigenvector for the product with eigenvalue . It depends only on the choice of the root . The commutator is therefore given by
One can check by a direct computation that the matrices and above share a common eigenvector if and only if .
Corollary 15. A Lamé connection is reducible if and only if .
The elliptic involution , , acts on our moduli space by sending the -class defined by to that defined by . This may be seen by choosing a fixed point of for the base point of the fundamental group. The following lemma is one of the key points in our construction.
Lemma 16. An -conjugacy class is stabilized by the elliptic involution if and only if it consists of either irreducible or Abelian representations. In other words, for every pair generating an irreducible or Abelian subgroup of there is a matrix such that
In the irreducible case is unique up to a sign and .
so that the involution acts trivially on the quotient, that is, on the triples . Since irreducible -classes are characterized by the corresponding triples , they are -invariant. Another way to prove this is to note that the matrix in the statement of the lemma must transpose the two eigenvectors of each of the matrices , . We look for an element of transposing the corresponding points of , that is, sending a quadruple to the quadruple . Since the cross-ratios of these quadruples coincide, such an element exists. Moreover, is an involution since its square fixes the four points and, therefore, . One must study the degenerate cases when and/or separately; we omit this discussion. In the reducible case we have, for example, and exists if and only if , which characterizes the Abelian case. Finally, interchanges the upper-triangular and lower-triangular -representations but stabilizes the diagonal ones.
When the matrices and are in the normal form above, the matrix in the lemma is given, up to a sign, by the formula
We resume our discussion of the non-resonant case.
Corollary 17. Given an elliptic curve and , the Lamé connections on with exponent are in one-to-one correspondence with points of the smooth affine hypersurface
They are irreducible and -invariant.
Proof. The connections are irreducible () and have no apparent singular points. Therefore the Riemann–Hilbert correspondence is injective and the desired assertions follow from Lemma 16.
6.3. Resonant cases
We now complete the picture by studying the case of resonant parameters .
6.3.1.
The monodromy representation is irreducible and is characterized by the corresponding triple . The local monodromy around is unipotent, with repeated eigenvalue , and is given by the commutator
We have and precisely when , the unique singular point of the surface.
Proposition 18. The Lamé connections with exponent having a logarithmic singular point are in one-to-one correspondence with the smooth points of the Markov affine hypersurface
They are irreducible and -invariant.
Proof. The same as for Corollary 17.
When , the image of the monodromy representation is the dihedral group of order (that is, quaternionic):
and the singular point of the connection is apparent: . In this case the Lamé connection is not characterized by its monodromy representation. Indeed, we have the following proposition.
Proposition 19. The Lamé connections over the singular point are in one-to-one correspondence with . They have the same monodromy representation into the dihedral subgroup of order in , and thus are irreducible. All of them are -invariant.
Proof. The Lamé connection is determined by its monodromy representation (acting on the space of solutions ) and the line corresponding to the solutions holomorphic at after analytic continuation along . In other words, the connection is determined by a triple
Another triple represents a connection gauge equivalent to the initial one if and only if
The monodromy representation, being irreducible here, has centralizer acting trivially on : once the monodromy representation is fixed, the gauge equivalence classes of connections are in one-to-one correspondence with .
One can easily check that the action of on Lamé connections induces the following action on the corresponding triples:
It turns out that the matrix in Lemma 16 (see (48)) takes the following form when (and, for example, ):
It conjugates to , thus proving the -invariance of the corresponding connection, a kind of miracle.
Remark 20. In fact, the moduli space of Lamé connections with a fixed exponent may be viewed, over the surface , as the minimal resolution obtained by a single blow-up of the singular point : the exceptional divisor stands for the set of connections with apparent singular point considered in Proposition 19. This will follow immediately from a similar result obtained in [25] for connections over after our descent construction. Let us also give some direct arguments. For every point in the smooth part of the affine surface , we have a one-dimensional subspace of all solutions holomorphic at after analytic continuation along . Using the local model (47) in the logarithmic case, one can check that coincides with an eigenspace of the local monodromy , namely,
The exceptional divisor of is given by the equation in the homogeneous coordinates and can be parametrized by
One can easily verify that the line tends to as the representation tends to the point in the parametrization above.
6.3.2.
In this case, several non-Hausdorff phenomena occur in the moduli spaces of representations and connections. The Hausdorff quotient is given by the Cayley affine hypersurface
The singular points of are
They play the same role in the sense that they are permuted when one changes signs of the generators:
Over a smooth point there are exactly three distinct - conjugacy classes, namely:
In the triangular cases, and can be chosen arbitrarily provided that , that is, .
Each of the two triangular representations corresponds to a unique Lamé connection (with a logarithmic singular point). Being permuted by , they are not -invariant. The moduli space of these triangular connections is a two-fold covering of the smooth part of , the two sheets of which are permuted around each of the four singular points.
When , the diagonal representation corresponds to a unique (regular) Lamé connection which is -invariant.
When , Lamé connections over the diagonal representation have an apparent singular point. There are exactly three equivalence classes of such connections corresponding to the following choices of the line bundle (see the proof of Proposition 19):
(any choice , , is equivalent to ). The involution permutes the first two connections but fixes the `generic' third one: in both situations it suffices to choose the matrix
(again see the proof of Proposition 19). Incidentally, the set of Lamé connections with diagonal monodromy splits into the union of another two-fold covering of , with Galois involution , and a copy of on which acts trivially.
Finally, over each smooth point there are exactly five Lamé connections (or three if ), only one of which is -invariant.
Now consider a singular point, say, (we recall that the four singular points play the same role). There are infinitely many distinct - conjugacy classes over this point in the fibre defined by the unipotent pairs
and the central pair (when ). If , then these representations correspond bijectively to Lamé connections that are -invariant. If , then for every unipotent representation there are exactly two Lamé connections given by and (all choices of are equivalent) and one Lamé connection with trivial monodromy; all of them are -invariant.
Remark 21. When there is an apparent singular point and the direction is fixed by the monodromy, we get a -invariant line bundle (again denoted by ) of positive degree, whence it follows that the bundle is unstable. Indeed, in this case we have , , and the -horizontal sections of are holomorphic and vanish to order at . Then by the Fuchs relation. We do not want to consider this kind of deformations in this paper.
We resume a part of our discussion.
Proposition 22. If , then all semistable and -invariant Lamé connections have an apparent singular point at (or are regular at when ) and their monodromy data belong to the following list:
- , where , and when ;
- , where , and when .
When , we must also add the four connections with monodromy .
6.4. Irreducible Lamé connections are elliptic pullbacks
We now check that -invariant representations actually come from representations of the 4- punctured sphere by means of the elliptic covering. To be precise, let us consider the elliptic pullback construction of §2 from the monodromy representation point of view. Given a connection with exponents
we consider its monodromy representation
It is determined by matrices satisfying the equations
The monodromy of its elliptic pullback is therefore given by the matrices
(see [34], §2, for details), whose commutator
has the following trace:
Clearly, this representation is -invariant since when we obtain from (49) that
Conversely, suppose that determines the monodromy of a -invariant Lamé connection. Then there is a matrix conjugating to . It is clear from the previous sections that we can assume that has zero trace: . Then it is straightforward to check that is the elliptic pullback of the following representation:
If the monodromy is irreducible, then is the unique (up to a sign) quadruple whose elliptic pullback gives the representation .
Corollary 23. Let be an irreducible Lamé connection with exponent . Then there is a unique (up to isomorphism) connection with exponents
such that is the elliptic pullback of .
Proof. Consider the monodromy representation of . There is a unique quadruple lifting to the representation such that (one must choose the sign of and, therefore, of the quadruple properly). Assume that the non-resonance condition holds: . By the Riemann–Hilbert correspondence there is a unique (up to isomorphism) connection with prescribed monodromy and exponents. By construction, the elliptic pullback of must have exponent and monodromy representation , the same as those of . Again by the (uniqueness part of the) Riemann–Hilbert correspondence, the elliptic pullback of must be isomorphic to . This proves the corollary in the non-resonant case. When , the same proof holds provided that the singular point is logarithmic (that is, has infinite monodromy). However, when the pole of becomes apparent, we must use the parabolic structure to restore the injectivity of the Riemann–Hilbert correspondence. We do not give details, but the key point in the proof is given by Proposition 19.
§ 7. Proof of Theorem 1
We now give the proof of Theorem 1 in detail. Let be an isomonodromic deformation of an irreducible Lamé connection with exponent . By Corollary 23, it is the elliptic pullback of an isomonodromic deformation , where . By the Bolibrukh transversality theorem (see §5.10) there is an open set of parameters for which the bundle is trivial and the parabolic directions are pairwise distinct; moreover, they do not lie on a curve of degree . Therefore the cross-ratio
is not special and we can apply Proposition 7 and Corollary 10, which yield an explicit formula for the Tu invariant:
where are the invariants of . In particular, coincides with the Okamoto symmetry of (see §5.8) and, therefore, is also a solution of the Painlevé VI equation with parameters
§ 8. Appendix: flat logarithmic -connections
Here we recall the basic facts about flat logarithmic connections. More details can be found in [14], [35], [30], [15].
A meromorphic connection of rank over a smooth complex manifold is a pair , where is a locally trivial holomorphic vector bundle of rank over and is a -linear morphism of sheaves,
(where is the sheaf of meromorphic sections of , and is the sheaf of meromorphic sections of the canonical bundle ) satisfying the Leibnitz rule
for all sections of the structure sheaf and all sections of the vector bundle . From an analytic point of view, if is given by the charts
glued by the transition maps
then is given in these trivializing charts by a tuple of differential operators of the form
satisfying the compatibility conditions
We adopt this analytic point of view throughout the paper. Meromorphic connections are considered up to holomorphic isomorphisms of vector bundles.
8.1. The polar divisor
We say that has a pole at a point if at least one of the entries of the corresponding matrix has a pole at . The order of the pole is the maximal order over all entries. It is easily seen to be independent of the choices of the chart and the local trivialization . The polar divisor of is a well-defined positive divisor on .
8.2. Flatness and the monodromy representation
A horizontal section (or a solution) of is any section of satisfying . In a chart, horizontal sections are solutions of the Pfaffian system . The connection is said to be flat (or integrable) if the following equality holds in every chart:
(in one chart is actually enough). This is equivalent to the existence of a basis of horizontal holomorphic sections for at every regular point. In other words, a connection is flat if and only if it is locally trivial at every regular point, that is, it is given by () in a convenient local trivialization of . This basis admits analytic continuation along any path in (just by gluing the local trivializations of using transition maps of the bundle). Therefore, fixing a point and a basis as above in a neighbourhood of , we get the monodromy representation of with respect to , that is, a homomorphism
defined in the following way. If is the new basis of horizontal sections around obtained after analytic continuation along , then is given by
If we change the basis of horizontal sections to another basis , , then the new monodromy representation is given by
Therefore the monodromy representation is well defined by up to - conjugacy and we shall simply write for any representative of the conjugacy class.
8.3. Flat logarithmic connections
A flat connection is said to be logarithmic if it has only simple poles (that is, is reduced) and the connection matrix in every chart is such that its differential also has simple poles. The latter condition is equivalent to the fact that the connection has a product structure in the neighbourhood of any smooth point of the polar divisor : there are local coordinates on and a local trivialization such that is given by and the connection matrix depends on only one variable . Therefore the -conjugacy class of the residual matrix is constant along each irreducible component of , and one can speak of the eigenvalues of the connection at each pole, that is, at every component of . A pole is said to be resonant if at least two of the eigenvalues differ by an integer: , . At every smooth point of a non-resonant pole, the connection matrix can be further reduced to its principal part
In the resonant case, for each pair one can reduce the -entry of to a resonant monomial . For each irreducible component of the divisor we fix a path in joining the base point to a smooth point of . Consider a loop in based at , going first very close to along , turning once around , and going back to along . The conjugacy class of is independent of the choices and is called the local monodromy of around ; the eigenvalues are given by . If the local monodromy is diagonalizable, then so is the residual matrix; the converse is not true.
8.4. Trace and twist
The trace of a connection is the meromorphic connection of rank , where is the determinant of , defined in the previous notation by the transition maps , and is given by . The connection is said to be traceless if its trace is the trivial connection (on the trivial bundle). The polar divisor of the trace is bounded by that of the initial connection. The twist of by a connection of rank is a connection of rank given by their tensor product . If is defined in the same open covering by with transition maps , then the twist is given by the matrices with transition maps . We have
The trace of a flat (resp. logarithmic) connection is flat (resp. logarithmic).
8.5. -connections
For convenience of notation we now restrict ourselves to flat logarithmic -connections (that is, those that are traceless of rank ). Their monodromy representation takes values in . For each irreducible component of the polar divisor , the exponent is defined (up to a sign) as the difference between the two eigenvalues of the residual matrix. The corresponding local monodromy matrix has trace . The component is resonant if and only if . In this case (say, when ) the connection matrix can be reduced to one of the following:
in the neighbourhood of any smooth point of . The corresponding local monodromy matrix is of the form
respectively, where . The pole is said to be apparent in the former case (there is no pole when ) and logarithmic in the latter. In both cases we note that the bounded solutions
form a one-dimensional subspace in the space of solutions.
8.6. The Riemann–Hilbert correspondence
We define the Riemann–Hilbert correspondence as a map
which sends every connection (considered up to holomorphic bundle isomorphisms) to its monodromy representation (considered up to conjugacy). This map is surjective provided that has normal crossings [14] or has dimension [30]. Moreover, it is injective provided that none of the exponents is a non-zero integer. In fact, the lack of injectivity comes from apparent singular points. One can restore the injectivity in the resonant case by enriching the monodromy data as follows. For every fix a basis of solutions near and consider the one-dimensional subspace of those solutions that are bounded around after analytic continuation along the path . The full monodromy data, which characterize the connection up to isomorphism, consist of the monodromy transformation and the set of all (), where ranges over the set of all indices such that . Any base change , , yields new monodromy data
(for the standard action of on ).
Proposition 24. Assume that is a reduced divisor with normal crossings and let , , be as above. Then the set of flat logarithmic -connections with polar divisor and exponents around modulo isomorphisms is in one-to-one correspondence with the set of pairs , where
the homomorphism is such that for all ,
the line is -invariant for all ,
modulo the -action defined by (50).
8.7. Reducible -connections
A line subbundle is said to be -invariant if it is generated by -horizontal sections. In this case the connection induces a meromorphic connection on . The connection is said to be reducible if it admits such an invariant line bundle, and irreducible otherwise. When the connection is reducible, its monodromy representation is also reducible: the monodromy group has a common eigenvector. In the logarithmic case with normal crossings of the polar divisor, the converse is true: is reducible if and only if is.
8.8. Projective -connections and Riccati foliations
A meromorphic connection of rank 2 induces a projective -connection on . If the linear connection is given in the trivializing chart by
where , , and are meromorphic -forms on , then the projective connection is defined in the projective trivializing chart by
Another linear connection defines the same projective connection if and only if it is the twist of by some meromorphic connection of rank .
Conversely, if (which always holds when is a curve), then every -bundle is the projectivization of some vector bundle of rank . Moreover, the formulae above show that for every meromorphic projective -connection on and every meromorphic (linear) connection on the line bundle , there is a unique meromorphic linear connection on lifting the projective connection from with prescribed trace on .
When is a curve, there are two topological types of -bundles: the topological type of is determined by the parity of . We note that topological triviality is a condition for the existence of a square root of . In other words, a -bundle is topologically trivial if and only if can be lifted as an -vector bundle: setting , we have with . This -lifting depends on the choice of a square root: it is well defined up to points of order 2 in , and there are possible liftings over a curve of genus . Finally, every meromorphic projective connection on lifts uniquely to a linear -connection on (with the same pole divisor).
When is a curve, the total space of is a ruled surface , and the Riccati equation determines a singular foliation on whose leaves are the graphs of horizontal sections of the projective connection. The pair is called a Riccati foliation (see [28]). This foliation is regular and transversal to the ruling outside the polar locus of . Over the poles of the projective connection, the -fibre is the disjoint union of a vertical fibre of and one or two singular points. To be precise, when is logarithmic (with simple poles), the singular points correspond to eigenlines of the linear connection . Let and be the eigenvalues of at some pole . We denote the corresponding eigenlines by and . The exponent (or the Camacho–Sad index of the vertical leaf) at the singular point of the Riccati foliation is . Let be a section.
Proposition 25. Let be a curve, a Riccati foliation with simple poles over , and an -invariant section. Then
where are the exponents of the singular points passed through by , and ranges over the set of all invariant fibres of .
This is a particular case of the Camacho–Sad formula (see [28], p. 37).
Proof. Viewed as a projective connection, there is a unique lifting of the projective connection such that the -invariant line bundle corresponding to is the trivial bundle, and the connection induced by on is the trivial connection: the eigenvalues of over the pole are given by and . Then Fuchs relations give
and we have
Proposition 26. Let be a curve of genus , a Riccati foliation over with poles (counted with multiplicities), and a section which is not -invariant. Then the number of tangencies between and the fibration (including the singular points lying on ) is given by
(counted with multiplicities).
This is a particular case of Proposition 2 in [28], p. 37.
Proof. Choose any lifting of the projective connection and apply Lemma 6 to the line bundle corresponding to .
8.9. Stability of bundles and connections
A vector bundle of rank over a curve is said to be stable (resp. semistable) if
for all line subbundles . This notion is invariant under projective equivalence: the -bundle is stable (resp. semistable) if
for all sections .
Similarly, we say that a connection is stable (resp. semistable) if
for all -invariant line subbundles . Again, this notion is invariant under projective equivalence: the projective connection is stable (resp. semistable) if
for all -invariant sections . In particular, all irreducible connections are stable even if the bundle is unstable. However, for a semistable connection, the bundle cannot be arbitrarily unstable: by Lemma 6, the stability index is bounded by
where is the polar divisor of .
8.10. Meromorphic and elementary gauge transformations
We recall the definition of an elementary transformation of a vector bundle of rank over a curve . For every point and every line in the fibre over , one usually defines two birational bundle transformations
which are unique up to post-composition with a bundle isomorphism. When restricted to the punctured curve , both transformations induce isomorphisms. In a neighbourhood of they can be described as follows. Choose a local coordinate at and a trivialization such that the line is spanned by . This, in particular, induces a trivialization of on . The elementary transformations can be defined by the commutative diagram
where
All three bundles and are constructed by gluing the local trivial bundle to the same restricted bundle through different bundle isomorphisms (the identity or ) over the punctured neighbourhood . The isomorphisms given by this construction extend as birational bundle transformations. We have
On the other hand, induce the same birational -bundle transformation
since and coincide both in and .
We claim that our construction depends only on the `parabolic structure' , not on the choice of the local trivialization . Indeed, for another choice
we must verify that , where . If , then ; since must be spanned by , we have and is holomorphic with .
A similar calculation shows that the lines defined by in the construction above are independent of our choices. In other words, if is equipped with a parabolic structure over , then the elementary transformations induce a birational transformation
of parabolic bundles which is well defined up to left and right composition with parabolic bundle isomorphisms. It also follows from the computations above that the composites
are parabolic bundle isomorphisms. In this sense, and are inverse to each other. We can also consider a general parabolic bundle of rank over , where is a finite subset and is a section of the projective bundle induced over . The elementary transformations of parabolic bundles over are defined for in the same way as above (note that induces an isomorphism of parabolic bundles over ), and as the identity for . Finally, if are two distinct points, then the elementary transformations and commute (up to parabolic bundle isomorphisms), so that one can define for any subset .
We now describe the action of elementary transformations on parabolic connections , where is a parabolic bundle over in the notation above and is a meromorphic connection on . Let and denote by the pushforward of under the elementary transformations . Then are meromorphic connections on . In the notation above, if is given in the coordinates by the formula
then is defined in the local trivializations of by , where
If is not a pole of , then has a logarithmic pole at . When is a pole of order for , there are two cases:
if is an eigenvector of at (that is, of at ), then has a pole of order or at ;
otherwise has a pole of order at .
In either case, is an eigenvector of at .
When is a logarithmic connection, the connections are logarithmic if and only if either is regular, or is a pole and is an eigenline for . One can then choose the coordinate in such a way that is given by with
(including the regular case ) with the restriction in the middle case. Then are given in the coordinates by , where
respectively. We resume.
If is a regular point of , that is, , then the connections are logarithmic with eigenvalues and .
If is a pole and is an eigenvalue of at , then there is one and only one eigenline associated with except in the diagonal case when . When is traceless, the last case does not occur and (the eigenspace of) each eigenvalue determines a parabolic structure over .
If are the eigenvalues of at and is the eigenline associated with , then (resp. ) has eigenvalues (resp. ). The connections are of diagonal type if and only if is. The parabolic structure over corresponds to the eigenvalue .
The trace of the connection is changed by the formula
where is the unique logarithmic connection on having a single pole at with residue and trivial monodromy. Indeed, the monodromy is not changed by a birational bundle transformation.
Footnotes
- 1
For example, is the Camacho–Sad index of along the fibre at (see [28]).