Table Of Content

6 Estimated transversality and rational maps 0 0 2 Rosa Sena-Dias n a February 2, 2008 J 0 3 Abstract ] G In this paper, we address a question of Donaldson’s on the best S estimate that can be achieved for the transversality of an asymptot- . h ically holomorphic sequence of sections of increasing powers of a line t bundleoveranintegralsymplecticmanifold. Morespecifically,wefind a m an upper bound for the transversality of n such sequences of sections over a 2n-dimensional symplectic manifold. In the simplest case of [ S2, we also relate the problem to a well known question in potential 2 theory (namely, that of finding logarithmic equilibrium points), thus v establishing an experimental lower bound for the transversality. 6 1 7 1 Introduction 1 1 5 InhisworkonsymplecticLefschetzpencils,Donaldsonintroducedthenotion 0 of estimated transversality for a sequence of sections of a bundle tensored / h with increasing powers of a line bundle. This, together with asymptotic t a holomorphicity, is the key ingredient in the construction of symplectic sub- m manifolds. Despite its importance in the area, estimated transversality has : v remained a mysterious property. In this paper, one of our aims is to shed Xi some light into this notion by studying it in the simplest possible case, namely that of S2. We state some new results about high degree rational r a maps on the 2-sphere that can be seen as consequences of Donaldson’s existence theorem for pencils, and explain how one might go about answering a question of Donaldson: what is the best estimate for transversality that can be obtained? We also show how the methods applied to S2 can be further generalized to prove the following: Theorem 1 Let X be a symplectic manifold of dimension 2n with symplectic form ω such that [ω/2π] lies in H2(X,Z), and a compatible almost complex structure. Let L X be a Hermitian line bundle whose Chern → 1 class is [ω/2π]. There exits η < 1 such that, if we have n asymptot- 0 ically holomorphic sequences of sections s , ,s of Lk satisfying η 0 n ··· ≤ s 2+ + s 2 1, then η < η . 0 n 0 || || ··· || || ≤ Donaldson’s results (see [Do1], [Do3]) ensure that such sequences exist for some η and any choice of complex structure. Recall that a sequence of sections s of Lk is said to be asymptotically holomorphic if its ∂¯is bounded k { } independently of k. It is not hard to see that Donaldson’s transversality theorem for the symplectic manifold (S2,ω ) implies the following interesting result: FS Proposition 1 There is 0 < η 1, such that, for each k large enough, ≤ there exists a pair of homogeneous polynomials, (p ,q ), of degree k, in two k k complex variables, defining a function from C2 to C2, that takes S3 C2 ⊂ into an annulus of outer radius 1 and inner radius √η. ThispropositionisaconsequenceofamoregeneralresultforcomplexKa¨hler manifolds that comes from applying Donaldson’s techniques to the complex setting (using the techniques appearing in [Do1] for the Ka¨hler setting; this is an exercise which we carry out for the sake of completeness). The special case of S2 is treated in more detail in section 3. To get a feeling for how strongtheaboveresultis,letustrytotakep (z,w) = zk andq (z,w) = wk. k k Then,theimageofS3 byeachofthemaps(p ,q )iscontainedinanannulus k k of outer radius 1 but whose inner radius is 1/√2k 1. − Infact,asequenceofpairsofhomogeneouspolynomialssatisfyingPropo- sition 1 is very special. We prove: Theorem 2 Let (p ,q ) be a sequence of pairs of homogeneous polynomials k k as above. The map p /q , thought of as a degree k map of CP1 to itself, has k k asymptotically uniformly distributed fibers, in the sense that, if xk denote i the points in one fiber for i = 1, k, counted with multiplicity, and f is a ··· 2 function on S2, then C k 1 1 C f (cid:12)(cid:12)k Xi=1f(xki)− |S2|ZS2f(cid:12)(cid:12)≤ ||△k ||∞, (1) (cid:12) (cid:12) (cid:12) (cid:12) and, in particular,(cid:12)tends to zero. (cid:12) A similar result holds for the branch points of p /q . The problem of dis- k k tributing points on S2 is an old and important problem with many applica- tions. Ithasbeenaddressedbyseveralbranchesofmathematics,forexample 2 in potential theory (see [RSZ]) andin arithmetic numbertheory (see [BSS]). This result is sharper than results found through potential theory or arithmetic number theory methods (although in this last case the bound for the expression in inequality (1) is for functions f in L2 and not simply in 2). C Indeed the bound (1) is optimal in the sense that we cannot expect to get a better asymptotic boundin k for 2 functions usingsecond derivatives. The C statement of the above Proposition 1, has nothing to do with symplectic geometry, it is simply a statement about rational maps. We try to look at it without using the techniques of [Do3] and give an explicit construction of polynomials which are experimentally seen to satisfy the required property. This involves the choice of two sets of k points, the zeroes of p and q on k k S2. We will choose two sets of asymptotically uniformly distributed points which are a slight modification of the so called generalized spiral points. Generalized spiral points come from trying to solve a problem in potential theory, that of distributing a big number of charges on the 2 sphere subject to a logarithmic potential. They are described in [RSZ] as a good approx- imation of the actual solution to this problem. Even though the problem itself remains unsolved, some things are known about the optimal distribu- tion. We will discuss the relations between this problem and our own. One of theupshotsofthisexplicit constructionisthatitallows toexperimentally determine a lower boundfor the constant η appearingabove. Together with the upper bound coming from Theorem 1 (that can be made totally explicit for S2), this gives a partial answer to Donaldson’s question. A brief outline of this paper is the following: In section 2, we give a review of the results in [Do1] and [Do3] and explain how they work in the complexsetting. Insection3,weexplainhowtheseresultsimplyProposition 1 and we also calculate an explicit upper bound for η appearing in that Proposition, this bound turns out to be of the order e 1036! We also prove − Theorem 2. Section 4 gives a construction of a sequence of polynomials which are experimentally shown to satisfy the condition in Proposition 1, as well as some steps towards the proof of the fact that they indeed satisfy the required condition. We also describe some relations between estimated transversality and the logarithm equilibrium problem in potential theory. Section 5 generalizes the method described in section 3 to find an upper boundfor η in the general case, thus proving Theorem 1. We end with some conjectures on how to find polynomials satisfying Proposition 1 via PDE theory. Acknowledgments I am very grateful to my advisor, Peter Kron- heimer, for his guidance and support throughout my Ph.D. and also for the many helpful discussions on the subject of this paper. I would also like 3 to thank Pedro for being an inexhaustible source of joy and peace to me. 2 Background The notion of linear system is one of importance in complex geometry. For example,itisoftenusingalinearsystemthatoneisabletorealizeacomplex manifold(whenitsatisfiescertainconstrains)asasubmanifoldofCPN. This is thecontent of Kodaira’sembeddingtheorem. More trivially, ageneric linear system of dimension 0 (whose existence is easy to establish) gives rise to a divisor, i.e., a complex submanifold. Moving one step up, a generic one dimensional linear system gives rise to a Lefschetz pencil, i.e., a holomorphic map X CP1 (defined away from a codimension 4 subvariety) → with the simplest possible singularities. It is well know, that every symplectic manifold has an almost-complex structure, which is the same as saying, that the symplectic category generalizes the Ka¨hler category. The natural question is then: is it possible to generalize the notion of linear system to this new setting, and use it to study symplectic manifolds, as it was used to study complex manifolds? Even in the simplest case of linear systems of dimension 0, this poses problems. The most important difficulty with which one is faced is the non-existence of holomorphic sections of complex bundles over symplectic manifolds (except in the integrable case where the manifold is actually complex). In his paper [Do1], Donaldson, resolves this issue by substituting the holomorphic condition by what is called asymptotic holomorphicity. One looks for sections of an increasing power of a line bundle which have a bounded ∂¯. In the holomorphic setting, a complex submanifold is obtained as the zero set of a transverse, holomorphic section. The holomorphicity condition becomes asymptotic holomorphicity, how about the transverse condition? Even though this condition could easily be trans- lated into the symplectic picture, it is no longer strong enough. It needs to become ”estimated transversality”, that is, transverse with a good estimate independent of the (increasing) degree of the bundle. Using this notion in a very key manner, Donaldson establishes the existence of sections of bundles whose zero set is symplectic, therefore proving an important existence theorem for symplectic submanifolds. In [Do3], he goes one step further and proves the existence of the analogue of pencils. These symplectic Lefschetz pencils completely characterize symplectic manifolds: very roughly, a manifold is symplectic exactly when it can be seen (after blow up) as a bundle over CP1, with symplectic fibers, some of which have simple singularities. Surprisingly, the techniques used to prove this, and in particular the notion 4 of estimated transversality, can be used to prove new theorems in the complex setting. Even for S2, one can prove new and unexpected results on rational maps of high degree. 2.1 Symplectic Lefschetz pencils and estimated transversality A natural question to ask in symplectic geometry is: Does every symplectic manifold have symplectic submanifolds? There have, so far, been two approaches to this question: 1. The oldest one, by Gromov, is, given an almost-complex structure J on X, to look for 2-dimensional submanifolds which are J invariant. More precisely, Y a dimension 2 submanifold of X is a J holomorphic curve if, for all y Y, J(T Y) = T Y. Every holomorphic curve is y y ∈ symplectic. 2. The more recent one, the one we are going to be concerned with here, by Donaldson, is, assuming that [ω/2π] is an integral class and that L X is a complex line bundle with c (L) = [ω/2π], to look for 1 sec→tions of the bundleLk X with ∂¯s < ∂s (here ∂ and ∂¯are with → | | | | respect to some almost-complex structure). Their zero set will be a codimension 2 symplectic submanifold of X. To be more precise, in his paper [Do1], Donaldson proves the existence of such sections: Theorem 3 (Donaldson, [Do1]) Let X be a manifold with a symplectic form ω, such that [ω/2π] H2(X,Z) and letL X be a complex Hermitian ∈ → line bundle with a connection form, whose curvature is iω. For sufficiently large k, there is a sequence, s of sections of Lk such that: k 1. s is bounded by 1, s C√k and s Ck where C is k k k | | |∇ | ≤ |∇∇ | ≤ independent of k, 2. ∂¯s is bounded by some constant C, independent of k, k | | 3. there is a constant η, independent of k such that s η = ∂s k k | |≤ ⇒ | |≥ η√k. When k is large enough, C < η√k along the zero set of s , therefore k ∂¯s < ∂s which implies that s 1(0) is a submanifold of X and that | k| | k| −k it is symplectic. Condition 3 in Theorem 3 plays an extremely important 5 role in the story. A sequence satisfying it is said to be η transverse to zero (or simply η transverse). In fact, we can generalize this notion further: Definition 1 Let X be a manifold with a metric, L X a Hermitian → complex line bundle, E X a Hermitian complex vector bundle and τ k → { } a sequence of sections of E Lk. Let η be a positive number. Then, we say ⊗ that τ is η transverse to zero if k { } τ η = [∂τ ] v,[∂τ ] v η2k v 2, k k ∗ k ∗ | | ≤ ⇒ h i ≥ | | for all v section of E Lk. ⊗ Note that, if ∂τ is not surjective, this will not be possible, since for some k v = 0, [∂τ ] v = 0. Theabove definition is the same as asking that τ η, k ∗ k 6 | |≤ needstoimplythat∂τ issurjectiveandhasapointwiserightinverse, whose k norm is smaller than η 1k 1/2. Condition 2 in Theorem 3 is referred to as − − asymptotic holomorphicity. Generalizing this result further, Auroux in [Au1], proves Theorem 3 with L replaced by L E, where E is a complex vector bundle of any ⊗ rank. Theorem 3 is an existence theorem for the symplectic analogs of linear systems of dimension 0. What about pencils, i.e., linear systems of dimension 1? The first thing to do is to generalize the notion to this new setting. Definition 2 Let X be a manifold with a symplectic form. Then, a map F, defined on X minus a codimension 4 manifold, is a symplectic Lefschetz pencil if, for every point p X, one of the following conditions is satisfied: ∈ F is a submersion at p, • F is not defined at p, in which case, there are compatible complex • coordinates z , z , centered at p, with F = z /z , 1 n 1 2 ··· F is defined at p, but it is not a submersion at p, there are compatible • complex coordinates z , z , centered at p, with F = z2+ z2. 1 ··· n 1 ··· n Here, ”compatible complex coordinates” simply means a map, defined lo- cally around p to Cn, such that, the pullback of the standard symplectic form on Cn is ω at the origin. Donaldson in [Do3], proves the following: Theorem 4 (Donaldson, [Do3]) Let X be a manifold with a symplectic form ω, such that [ω/2π] H2(X,Z) and letL X be a complex Hermitian ∈ → line bundle with a connection form, whose curvature is iω. For sufficiently large k, there is a sequence of pairs of sections (s ,s ) of Lk such that: 0 1 6 1. s 2+ s 2 1, s C√k and s Ck, i= 0,1, 0 1 i i | | | | ≤ |∇ |≤ |∇∇ | ≤ 2. ∂¯s and ∂¯s are bounded by a constant independent of k, 0 1 | | | | 3. s is η transverse to zero, for some η independent of k, 0 4. (s ,s ) is η transverse to zero, 0 1 5. ∂(s /s ) is η transverse to zero, away from the zero locus of (s ,s ). 1 0 0 1 As explained in [Do3], for large k, after perturbing the sections s slightly, 1 the map s /s will give rise to a Lefschetz pencil. 1 0 2.2 Estimated transversality in the complex setting Since any Ka¨hler manifold is also symplectic, one could apply Theorems 3 and4totheKa¨hlercase. Atfirstsight,thesetheoremsseemnottogeneralize theexistence theoremsforcomplex submanifoldsandholomorphicLefschetz pencils, since they do not produce holomorphic sections of bundles. But Donaldson, in [Do1], proves that in the Ka¨hler case, the asymptotically holomorphic condition in Theorem 3 can be strengthened to holomorphic. The result then becomes a new theorem for Ka¨hler manifolds: Theorem 5 (Donaldson, [Do1]) Let X be a Ka¨hler manifold with integral cohomology. Let ω be the symplectic form on X and let L X be a → complex Hermitian line bundle with a connection form, whose curvature is iω. For sufficiently large k, there is a sequence of holomorphic sections of Lk, s , such that: k { } s is bounded by 1, k • | | there is a constant η, independent of k, such that s η = ∂s k k • | |≤ ⇒ | |≥ η√k. It is then natural to ask: what does Theorem 4 say for a Ka¨hler manifold? Theorem 6 Let X be a Ka¨hler manifold with symplectic form ω, such that [ω/2π] is in H2(X,Z) and a complex Hermitian line bundle L X with a → connection form, whose curvature is iω. For sufficiently large k, there is a sequence of holomorphic sections of Lk, (s ,s ), such that: 0 1 1. s 2+ s 2 1, s C√k and s Ck, i= 0,1, 0 1 i i | | | | ≤ |∇ |≤ |∇∇ | ≤ 2. s is η transverse to zero, for some η independent of k, 0 7 3. (s ,s ) is η transverse to zero, 0 1 4. ∂(s /s ) is η transverse to zero, away from the zero locus of (s ,s ). 1 0 0 1 TheproofofTheorem6isacombinationofelementsintheproofofTheorem 4andelementsintheproofofTheorem5. Thefirststepistogets ,asection 0 of Lk for big k, which is holomorphic and η transverse to zero, for some η independent of k, by using Theorem 5. Next, one needs to build a sequence of pairs of sections of Lk, that are η transverse to zero (note that this pair of sequences can be built close to (s ,0), so that its first term will be η/2 0 transverse). Instead of using the asymptotically holomorphic pair built in the proof of Theorem 4 and perturbing it to make it holomorphic, as in the proof of Theorem 5, one can use the methods of [Do3] for building pencils, together with the following existence lemma proved in [Do1]: Lemma 1 (Donaldson,[Do1]) There are constants a, b and c such that, given any p X and any k large, there is a holomorphic section σ of Lk, p ∈ satisfying the following estimates: e bkd2(p,q) σ (q) e akd2(p,q) if d(p,q) ck 1/3, − p − − • ≤ | | ≤ ≤ σ (q) e ak1/3 if d(p,q) ck 1/3. p − − • | | ≤ ≥ The section σ also satisfies p ∂σ (q) √ke akd2(p,q) if d(p,q) ck 1/3, p − − • | | ≤ ≤ ∂σ (q) √ke ak1/3 if d(p,q) ck 1/3. p − − • | | ≤ ≥ To proceed, the strategy is the usual one. Use a covering lemma to cover X with colored neighborhoods of points where the above lemma applies (two neighborhoods with the same color are well separated). Now start with any pair of holomorphic sections, (s ,0) for example, and modify it, over all 0 the balls of the first color, by adding a section of the form (w σ ,w σ ) 1 pi 2 pi for each ball (of color 1). The vector w = (w ,w ) comes from applying a 1 2 transversality lemma to the representation of the pair with respect to the trivialization σ . The lemma is: pi Lemma 2 (Donaldson,[Do3]) Given a map f : Bm(11/10) Bn(1), → which is holomorphic and any small positive number δ, there is a w Cn ∈ with norm smaller than δ, such that f+w is η = δ/logpδ transverse to zero. 8 In this way, we achieve both holomorphicity and transversality over each ball in the first color. Note that the balls of a given color do not interfere with each other since they are well separated (here we use the inequalities in Lemma 1). Next, we apply the same method to the balls of the second color, thus achieving transversality there. But we need to make sure that we donotspoilthe transversality achieved over the balls withthefirstcolor. This is a consequence of the inequalities in Lemma 1. We can keep up this process until we have gone trough all the colors. Just as in the proof of Theorem 4, (see [Do3]), the method produces an η transverse sequence of sections, for some η independent of k. These sequences are actually holomorphic, simply because they are a sum of holomorphic sections. Starting from the η transverse pair, we slightly modify the second term s , to obtain 1 transversality for ∂(s /s ), justas in [Do3], butusingholomorphic reference 1 0 peak sections. We make use of the following lemma: Lemma 3 Let δ be a positive number smaller than 1/2 and p a point in X. For sequences s and s as above, large k and small r, there is a π Cn 0 1 ∈ (not depending on k) with norm smaller than δ, such that, s +σ 1 p,π ∂ s (cid:18) 0 (cid:19) is δ/logp(1/δ) transverse to zero, over the ball of center p and radius rk 1/2. − Here, σ = πασα p,π p α X and σα comes from Lemma 1. p This can be proved exactly as the corresponding statement in [Do3]. Now we can proceed as in [Do3]. The final result will be holomorphic, because the σ are. Note that a modification of the above construction for the pair p,π can be used to prove the following: Proposition 2 Let X be a Ka¨hler manifold, of complex dimension n, with symplectic form ω, such that [ω/2π] lies in H2(X,Z). Let L X be a → complex Hermitian line bundle L X with a connection form, whose cur- → vature is iω. There exits 0 < η 1 and n sequences of holomorphic sections ≤ s , ,s of Lk satisfying: η s 2+ + s 2 1. 0 n 0 n ··· ≤ || || ··· || || ≤ 9 3 Estimated transversality for rational maps on S2 3.1 The problem on S2 In [Do3], Donaldson asks the question: Given a symplectic manifold, what is the best η for which we can find (s ,s ) Γ(Lk) satisfying the conditions 0 1 ∈ of Theorem 4? (We need to normalize the pairs to have L norm 1 for ∞ this question to make sense.) We will address this question for S2 with the Fubini-Study metric. Now, S2 with the Fubini-Study form is Ka¨hler, so we can apply to it Theorem 6. In fact, the existence of a pair of holomorphic sections of (k), satisfying only conditions 1 and 3 in the theorem, seems O to already give an interesting result. This last condition becomes somewhat simpler in the context of 2-dimensional manifolds, for there can be no surjective maps from C2 to C. Namely, it becomes s 2 + s 2 η so that, 0 1 | | | | ≥ when s 2 η/2, s 2 η/2. Also, holomorphic sections of (k) are easy 0 1 | | ≤ | | ≥ O to characterize, they are simply homogeneous polynomials of degree k in two complex variables. In this way, we prove Proposition 1. A way to find a lower bound for the best η appearing in that proposition is then to explic- itly determine a sequence of pairs of homogeneous polynomials of degree k, (p ,q ), such that the number k k max( p 2 + q 2) k k || || || || min( p 2+ q 2) k k || || || || is bounded independently of k. Where p stands for the norm of p as k k || || a section of (k), i.e., letting [z : w] be the homogeneous coordinates in O S2 = CP1, p (z,w) k p [z : w]= | | . || k|| (z2 + w 2)k/2 | | | | The inverse of this bound will provide the lower bound we are looking for. There is a chart on S2 minus the south pole, obtained by stereographic pro- jection trough the south pole. It is centered at the north pole and identifies S2 minus the south pole with C. In homogenous coordinates on S2, it is simply given by z = z/w. Over S2 minus the south pole, (k) admits a O trivialization, i.e., a global section which we denote by wk. In fact, when we identify sections of (k) with homogeneous polynomials of degree k, this O section corresponds precisely to the polynomial wk. It is actually defined over all of S2, but vanishes at the south pole. Its norm in the z coordinate is simply, 1 . (1+ z 2)k/2 | | 10