Let S be a compact surface properly embedded in a smooth or PL 3-manifold M. A compressing diskD is a disk embedded in M such that and the intersection is transverse. If the curve ∂D does not bound a disk inside of S, then D is called a nontrivial compressing disk. If S has a nontrivial compressing disk, then we call S a compressible surface in M. If S is neither the 2-sphere nor a compressible surface, then we call the surface incompressible. Note that 2-spheres are excluded since they have no nontrivial compressing disks by the Jordan-Schoenflies theorem, and 3-manifolds have abundant embedded 2-spheres. Sometimes one alters the definition so that an incompressible sphere is a 2-sphere embedded in a 3-manifold that does not bound an embedded 3-ball. Such spheres arise exactly when a 3-manifold is not irreducible. Since this notion of incompressibility for a sphere is quite different from the above definition for surfaces, often an incompressible sphere is instead referred to as an essential sphere or a reducing sphere.
Compression
Given a compressible surface S with a compressing disk D that we may assume lies in the interior of M and intersects S transversely, one may perform embedded 1-surgery on S to get a surface that is obtained by compressingSalongD. There is a tubular neighborhood of D whose closure is an embedding of D × with D × 0 being identified with D and with Then is a new properly embedded surface obtained by compressing S along D. A non-negative complexity measure on compact surfaces without 2-sphere components is b0 − χ, where b0 is the zeroth Betti number and χ is the Euler characteristic. When compressing a compressible surface along a nontrivial compressing disk, the Euler characteristic increases by two, while b0 might remain the same or increase by 1. Thus, every properly embedded compact surface without 2-sphere components is related to an incompressible surface through a sequence of compressions. Sometimes we drop the condition that S be compressible. If D were to bound a disk inside S, then compressing S along D would result in a disjoint union of a sphere and a surface homeomorphic to S. The resulting surface with the sphere deleted might or might not be isotopic to S, and it will be if S is incompressible and M is irreducible.
Algebraically incompressible surfaces
There is also an algebraic version of incompressibility. Suppose is a proper embedding of a compact surface in a 3-manifold. Then S is π1-injective if the induced map on fundamental groups is injective. In general, every π1-injective surface is incompressible, but the reverse implication is not always true. For instance, the Lens spaceL contains an incompressible Klein bottle that is not π1-injective. However, if S is two-sided, the loop theorem implies Kneser's lemma, that if S is incompressible, then it is π1-injective.
Seifert surfaces
A Seifert surfaceS for an orientedlinkL is an oriented surface whose boundary is L with the same induced orientation. If S is not π1 injective in S3 − N, where N is a tubular neighborhood of L, then the loop theorem gives a compressing disk that one may use to compress S along, providing another Seifert surface of reduced complexity. Hence, there are incompressible Seifert surfaces. Every Seifert surface of a link is related to one another through compressions in the sense that the equivalence relation generated by compression has one equivalence class. The inverse of a compression is sometimes called embedded arc surgery. The genus of a link is the minimal genus of all Seifert surfaces of a link. A Seifert surface of minimal genus is incompressible. However, it is not in general the case that an incompressible Seifert surface is of minimal genus, so π1 alone cannot certify the genus of a link. Gabai proved in particular that a genus-minimizing Seifert surface is a leaf of some taut, transversely oriented foliation of the knot complement, which can be certified with a taut sutured manifold hierarchy. Given an incompressible Seifert surface S for a knot K, then the fundamental group of S3 − N splits as an HNN extension over π1, which is a free group. The two maps from π1 into π1 given by pushing loops off the surface to the positive or negative side of N are both injections.