On the \((-1)\)-curve conjecture of Friedman and Morgan (Q1319216)

The \((-1)\)-curve conjecture from \textit{R. Friedman} and \textit{J. W. Morgan} is a conjectured relation between the non-minimality of a surface and its connected sum decompositions [Bull. Am. Math. Soc., New Ser. 18, No. 1, 1-19 (1988; Zbl 0662.57016)]: every \((-1)\)-sphere (i.e. an embedded 2- sphere with self intersection \(-1)\) in a surface of non-negative Kodaira dimension, is homologous to a \((-1)\)-curve up to sign. This is the special case \(N = \overline \mathbb{P}^ 2\) of the generalized \((-1)\)-curve conjecture: for every connected sum decomposition of a surface \(X\) of non negative Kodaira dimension into \(X= X' \# N\) with \(N\) negative definite and simply connected, the sublattice \(H_ 2(N) \subset H_ 2 (X)\) is generated by \((-1)\)-curves. The rational homology version of the \((-1)\)- curve conjecture implies that all plurigenera are diffeomorphism invariants [\textit{R. Friedman} and \textit{J. W. Morgan}, ``Smooth 4- manifolds and complex surfaces'' (1994)], and in particular the invariance of the Kodaira dimension (Van de Ven conjecture). The article addresses the \((-1)\)-curve conjectures for the case \(p_ g>0\), and assuming that the connected sum decompositions are good, i.e. behave like connected sums with \(\overline \mathbb{P}^ 2\) as far as the Donaldson polynomials are concerned (conjecturally all decompositions are good). The simply connectedness hypothesis in the article can be dropped using a paper by \textit{J. W. Morgan} and \textit{T. S. Mrowka} [Int. Math. Res. Not. 1992, No. 10, 223-230 (1992; Zbl 0787.57011)]. The results are proved for good connected sum decompositions but here we give the \((-1)\)- sphere version. Let \(X\) denote a surface with \(p_ g > 0\), \(X_{\min}\) its unique minimial model and \(K_{\min}\) the canonical class of the minimal model or its pull back to \(X\). (1) The homology class \(e\) of a \((-1)\)-sphere is a class of type \((1,1)\). (2) if \((b_ 1(X) = 0\), \(p_ g\) or \(K^ 2_{\min}\) is even, and \(| K_{\min} |\) contains a reduced curve), or \(X_{\min}\) has Picard number 1, then \(e\) is rationally homologous to a \((-1)\)-curve up to sign. The main ingredient is a refinement of an idea of Tyurin: the algebraic definition by O'Grady and Morgan of the Donaldson polynomials is used to prove that the polynomials are of pure Hodge type. In fact the polynomials are shown to define algebraic cycles on products of the surface. See the author's forthcoming paper [``Some remarks on the Kronheimer-Mrowka classes of algebraic surfaces'', J. Differ. Geom. (to appear)] for an applications of pureness to the conjecture that the canonical class of the minimal model is invariant (which implies the \((- 1)\)-curve conjecture) and a proof that there are no conditions on the degree of the polynomial.