Nouveaux Aspects Combinatoires de Théorie Des Noeuds Et Des Noeuds Virtuels
A knot is an embedding of a circle into a 3-dimensional manifold. When this manifold is the sphere, knots can be described combinatorially using Gauss diagrams. Forgetting about the actual knots, one can study Gauss diagrams independently: this is called virtual knot theory. In the first part we define a general version of virtual knots that depends on a group G endowed with a Z/2-valued homomorphism w. When G and w are suitably chosen, this version generalizes knot theory in a given thickened surface - i.e. a 3-manifold endowed with a line bundle projection onto a surface. Besides encoding knots, Gauss diagrams can also encode Vassiliev's finite-type knot invariants. A complete set of criteria is given to detect these invariants in the present framework. Notably, the criterion for Reidemeister III gives a positive answer to a conjecture of Polyak. Several examples are given, including an improvement of Grishanov and Vassiliev's theorem on planar chain invariants. The third part is a draft investigating a plan to find an algorithm that tells whether a knot in the solid torus is isotopic to a closed braid. The first step is achieved: it consists of a characterization of Gauss diagrams of closed braids. We state and investigate a conjecture which predicts that for diagrams with minimal number of crossings, this first step is enough. The last part is a joint work with T.Fiedler, investigating invariants of non generic loops in the space of all immersions of a circle into the 3-space. This space is infinite dimensional, stratified by the degree of non genericity of an immersion. Vassiliev's theory was based on adding to the usual knots all strata with only ordinary double points as singularities. Here we forbid these double points and regard only some higher codimensional strata with a certain kind of triple points. We show that the resulting space is not simply connected, by exhibiting a non trivial 1-cocycle. Weighting this cocycle gives a new formula for the Casson invariant, using triple unknottings.