Holomorphic and pseudo-holomorphic curves on rationally connected varieties (Q973679)

In his usual friendly but thorough style, Kollár tells us what we ought to know about curves on varieties and more especially about genus \(0\) curves on rationally connected varieties. He starts with a reminder of how satisfactory the theory of codimension~\(2\) classes and divisors on smooth projective varieties is, and contrasts this with our far less complete knowledge about curves and their (co)homology classes. Section~1 of the paper gives and discusses some examples, conjectures and questions about this. For example, if \(X\) is a sufficiently general hypersurface of sufficiently large degree, is the homology class of a curve on \(X\) divisible by \(\deg X\), as conjectured by Griffiths and Harris? If \(X\) is a smooth projective Fano variety, do the homology classes of rational curves generate \(H_2(X,{\mathbb Z})\)? Section~2 is concerned with the spaces of curves on \(X\) or of maps from a given curve to \(X\). The endpoint is a pair of results analogous to what is sometimes called the Riemann part of Riemann-Roch, bounding from below the dimensions of these spaces at some given base point, i.e.\ of the component of the space of morphisms containing some given \(f: C\to X\) or component of the Hilbert scheme containing some given \(C\subset X\), under appropriate conditions. These bounds, Kollár points out, are frequently sharp in the sense that they become precise on small almost-complex deformations of \(X\) but they tend to fall a long way short of reality for \(X\) itself. For example, they predict no curves on hypersurfaces of degree \(6\) in \({\mathbb P}^4\), but of course there are many. In other words, algebraic varieties have more curves than most almost complex manifolds do; and while that is no surprise, it means that from the point of view of algebraic geometry these are the wrong bounds, and one should look for bounds that become sharp under algebraic deformations (but fail entirely for almost complex deformations). Section~3 briefly studies uniruled varieties and free curves. Roughly, a free curve is a rational curve on \(X\) that can be deformed in any direction in \(X\): more precisely, it is a map \(f: {\mathbb P}^1\to X\) such that \(f^*T_X\) is globally generated (i.e.\ \(H^1({\mathbb P}^1,f^*T_X(-1))=0\)). Being uniruled is the same thing as having such a curve. In Section~4 we see a sketch of the basic facts about rationally connected varieties. These are smooth projective varieties having a very free curve: that means a rational curve on \(X\) that can be deformed in any direction in \(X\) while keeping a point fixed, or \(H^1({\mathbb P}^1,f^*T_X(-2))=0\). This seems to be a better behaved condition than being uniruled, and can be characterised in many ways: the paper lists a few, but the simplest to state is that almost any two points can be joined by a rational curve. As an example of the good behaviour (more are given in the paper) we note that a smooth projective variety deformation equivalent to a rationally connected smooth pojective variety is also rationally connected. The final section relates all this to symplectic geometry, dealing with stable curves and Gromov-Witten invariants. The main question raised here (which was asked by the author in the 1990s but still seems to be hard) is whether being rationally connected is a symplectic deformation invariant property: if two smooth projective varieties are deformation equivalent considered as symplectic manifolds and one of them is rationally connected, does the other one have to be? The answer is known to be yes if ``rationally connected'' is replaced by ``uniruled''. In general, few proofs are given, but clear instructions are given for finding them in the literature. This article would be a good starting point for a research student or anyone else wanting to learn about the open problems in this area.