Abstract. In this paper, we define and investigate Z 2 -homology cobordism invariants of Z 2 -homology 3-spheres which turn out to be related to classical invariants of knots. As an application, we show that many lens spaces have infinite order in the Z 2 -homology cobordism group and we prove a lower bound for the slice genus of a knot on which integral surgery yields a given Z 2 -homology sphere. We also give some new examples of 3-manifolds which cannot be obtained by integral surgery on a knot.
Mathematics Subject Classification (2000). 57M27, 57M25.Keywords. Homology 3-spheres, homology cobordism, slice genus.In recent years, gauge theoretical tools and new results in 4-dimensional topology have successfully been used to study the structure of the integral homology cobordism group. It is for instance a consequence of Donaldson's famous theorem about the intersection forms of smooth 4-manifolds that the Poincaré homology sphere Σ(2, 3, 5) has infinite order in this group, and M. Furuta found a family of Brieskorn spheres which generates a subgroup of infinite rank [12]. Recently N. Saveliev [22] showed, using the w-invariant introduced in [11], that a Brieskorn sphere with non-trivial Rokhlin invariant has infinite order in the integral homology cobordism group. However, many 3-manifolds arising naturally in knot theory, for instance double coverings of the 3-sphere branched along knots, are not integral homology spheres, but still Z 2 -homology spheres. As in the case of integral homology spheres, the set of Z 2 -homology spheres modulo the Z 2 -homology cobordism relation forms a group, the so called Z 2 -homology cobordism group Θ 3 Z 2 . To study this group, we introduce, based on Furuta's result on the intersection forms of smooth 4-dimensional spin manifolds [13], two invariants of Z 2 -homology spheres which turn out to be in fact invariants of the cobordism class (see Theorem 1). Exploiting that these invariants are closely related to classical knot invariants like signature and slice genus, we prove estimates for them in the case of lens spaces, which enables us to exhibit many examples of lens spaces which have infinite order inThe first author gratefully acknowledges support from the Deutsche Forschungsgemeinschaft.