Abstract. Let π : X −→ S be a flat proper family of smooth connected projective curves parametrized by some smooth scheme of finite type over C. On every such a family, suitable derivatives "along the fibers" (in the sense of Lax) of the relative wronskian, as defined by Laksov and Thorup, are constructed. They are sections of suitable jets extensions of the g(g + 1)/2-th tensor power of the relative canonical bundle of the family itself.The geometrical meaning of such sections is discussed: the zero schemes of the (k − 1)-th derivative (k ≥ 1) of a relative wronskian correspond to families of Weierstrass Points (WP's) having weight at least k.The locus in Mg, the coarse moduli space of smooth projective curves of genus g, of curves possessing a WP of weight at least k, is denoted by wt(k). The fact that wt(2) has the expected dimension for all g ≥ 2 was implicitly known in the literature. The main result of this paper hence consists in showing that wt(3) has the expected dimension for all g ≥ 4. As an application we compute the codimension 2 Chow (Q-)class of wt(3) for all g ≥ 4, the main ingredient being the definition of the k-th derivative of a relative wronskian, which is the crucial tool which the paper is built on. In the concluding remarks we show how this result may be used to get relations among some codimension 2 Chow (Q-)classes in M 4 (g ≥ 4), corresponding to varieties of curves having a point P with a suitable prescribed Weierstrass Gap Sequence, relating to previous work of Lax.0. Introduction 0.1. The main goal that motivated the investigations which led to this paper was to deal with loci of the (coarse) moduli space M g of the smooth projective curves of genus g (defined over an algebraically closed field of characteristic zero), whose points correspond to curves possessing some special Weierstrass points.We are, in fact, especially interested in the computation of the classes in the Chow ring of M g with rational coefficients (as defined by Mumford in [Mu2], see section 1.5) of some of such loci. To achieve this goal we have been in some sense forced to introduce and to study a locus in M g which does not seem to have been considered in the earlier literature. In the moduli space M g define: i.e. the locus of (isomorphism classes of) curves in M g possessing a Weierstrass point with weight at least k. In order to study such a locus, above defined only as a set, one needs to put on it a reasonable structure of scheme and to study how its dimension varies, by varying k. For instance it is quite clear that wt(1) must be all of M g , a fancy way of saying that every curve of genus g ≥ 2 has at least 1 Weierstrass point.
0.2.In order to put on wt(k) a scheme structure, one has to deal with some natural but very useful tools, developed in section 2, whose construction, indeed, forms the conceptual core of the paper. As is well known, the Weierstrass Points (WP's in the following) on a smooth curve of genus g, can be detected as the zero-locus of the wronskian section of the bundle K, K being the ca...