Lecture 12: Prime Factorization of Knots Notes by Zach Haney February 25, 2016 1 Introduction In this lecture, we look at prime knots and “factoring" knots into a connected sum of prime knots. Our goal for prime factorization is to see a connection to the integers: unique factorization! We recall the main definition here: Definition 1.1. A nontrivial knot K is prime precisely when K = K1 #K2 implies either K1 or K2 is trivial. 2 Prime Factorization Theorem 2.1. For any two knots K1 and K2 , g(K1 #K2 ) = g(K1 ) + g(K2 ). Proof: By the boundary-connected sum, we know that g(K1 #K2 ) ≤ g(K1 ) + g(K2 ). To the equality, we must show the reverse inequality is true. Let Σtot be a minimal genus Seifert surface for K1 #K2 . The strategy will be to produce from Σtot another Seifert surface, Σ∗tot of minimal genus such that Σ∗tot = Σ1 #Σ2 is a connected sum of Seifert surfaces for K1 and K2 . As seen in the figure: 1 That is, we’ll see that g(K1 ) + g(K2 ) ≤ g(Σ1 ) + g(Σ2 ) = g(Σ∗tot ) = g(Σtot ), and we’ll know that g(Σtot ) = g(K1 #K2 ). So our main goal from above is to construct Σ∗tot . To recognize the connected sum, we need to see the separating sphere, as shown in the figure above, which intersects, without loss of generality, the knot transversally in a single pair of points and the Seifert surface in an arc joining those points. Our main issue to confront is that the minimal genus Seifert surface Σtot for K1 #K2 may NOT represent a connected sum decomposition. We’ll see, of course, that it in fact does using surgery. So consider the PL sphere S 2 ⊂ S 3 which splits K1 #K2 into its summands. As before, without loss, we know that Σtot intersects S 2 transversally, so that Σtot ∩ S 2 is a compact 1-manifold with with exactly two boundary points where K1 #K2 pierces the separating 2-sphere. If Σtot ∩ S 2 is a single arc α as seen in the figure below, then this arc separates Σtot into two components, namely Σ1 #Σ2 , bounding K1 and K2 individually. So we are done. Otherwise, Σtot ∩ S 2 consists of precisely one arc α and a finite number of disjoint, simple closed curves. To remove the closed curves inductively, it suffices to construct ∗∗ 2 2 another Seifert surface Σ∗∗ tot for K1 #K2 of the same genus but with Σtot ∩S ⊂ Σtot ∩S . Let C ⊂ Σtot ∩ S 2 be a closed curve which is “innermost” on S 2 \ α precisely when the interior of the disk D2 ⊂ S 2 \ α, bounded by C, is disjoint from Σtot . Perform surgery along C ⊂ Σtot , removing an annular neighborhood of C in Σtot and capping the boundary circles by disks, a shown in the figure Claim: C separates Σtot (and must bound a disk D2 ⊂ Σtot ), so the new surface F obtained by surgery is disconnected. 2 Why is that true? Well, otherwise, F would b a new Seifert surface for K1 #K2 with a strictly smaller genus as in the figure, and that’s not possible from our assumptions! Let Σ∗∗ tot ⊂ F be the component bounding K1 #K2 . Surgery cannot increase the ∗∗ genus, so g(Σ∗∗ tot ) = g(Σtot ), and Σtot is a Seifert surface for K1 #K2 whose intersection with S 2 has strictly fewer components than Σtot did. Proposition 2.2. If K1 is nontrivial, so is the connected sum K1 #K2 for any other knot K2 . If K1 ' K1 #K2 , then K2 must be trivial. We’ve seen this before! This is the third proof, and in particular doesn’t involve any wild knots such as those used by Mazur. Proof: If K1 #K2 is trivial, then g(K1 #K2 ) = 0 which means g(K1 ) = 0 = g(K2 ). Also, g(K1 ) = g(K1 #K2 ) says that g(K2 ) = 0. Corollary 2.3. A knot of genus 1 is prime. In fact, any nontrivial knot is a finite sum of prime knots. Proof: If K is not prime, K = K1 #K2 for two nontrivial knots each of which have smaller genus. Induct on the genus, and one works down to zero. 3 Theorem 2.4. Suppose a knot K can be decomposed as K = P #Q, with P prime, and also as K = K1 #K2 . Then either a) K1 = P #K1∗ and Q = K1∗ #K2 for some K1∗ or b) K2 = P #K2∗ and Q = K1 #K2∗ for some K2∗ . Proof: Let Σ be a separating S 2 , intersecting K transversally at two points for the decomposition K = K1 #K2 , as seen for example in the figure. Simillarly, the decomposition K = P #Q implies that there is some 3-ball B ⊂ S 3 such that B ∩ K is an arc α, intersecting ∂B transversally at two points, so that the ball-arc pair (B, α) becomes, after gluing the trivial ball-arc pair to its boundary, the pair (S 3 , P ). We recall that a ball-arc pair is a 3-ball with a tangle inside, intersecting its boundary at two points; moreover, the trivial ball-arc pair is the 3-ball plus the unknotted diameter inside. Without loss, Σ intersects ∂B transversally in a union of simple, closed curves disjoint from K. If Σ ∩ ∂B = ∅, then B is contained in one of the components of S 3 \ Σ, and the result follows immediately. The figure here gives these situations. Our goal is to remove curves in Σ ∩ ∂B via surgery on B. As Σ ∩ K is two points, any oriented simple, closed curve in Σ \ K has linking number 0 or ±1 with K. (Why ±1? The curves would be meridians of K!). (i) Eliminate components of linking number zero of Σ ∩ ∂B. Select innermost curve in Σ ∩ ∂B which does not link K. This curve bounds a disk D ⊂ Σ, with D ∩ ∂B = ∂D. The curve ∂D must also bound a disk D0 ⊂ ∂B which does NOT intersect K. That is, D0 ∩ K = ∅, since ∂D andd K are unlinked. Look to this figure 4 below to see visually what is happening in this case By Schoenflies, the sphere D ∪ D0 bounds a 3-ball. We “move” D0 across ∂B to the other side of D by adding or subtracting a 3-ball to/from B, disjoint from K, as seen in the figure above. The new pair (B, α) corresponds to (S 3 , P ) as before, but Σ ∩ ∂B has fewer components. Repeating this, we may assume all curves in Σ∩∂B have linking number ±1 with K, as see in this figure (ii) Eliminate components of linking number 1 of Σ ∩ ∂B. a) If Σ ∩ B has a component which is a disk D00 , then D00 intersects K in one point. Because P is prime, one side of D00 in B is a trivial ball-arc pair. Subtract from B a regular neighborhood of this trivial ball-arc pair to produce a new (B, α) with fewer components than Σ ∩ B. Repeat until all components of Σ ∩ B are annuli. 5 b) Let A be an annular component of Σ ∩ B. Then ∂A also bounds an annulus A in ∂B. Choose A “outermost”, i..e farthest from α, so that A0 ∩ Σ = ∂A0 . , Then A ∪ A0 ' T 2 is a torus which bounds a solid region M ⊂ B. Let δ be the disk which is one of the components of ∂B \ A0 . Then ∂δ is a circle of ∂A0 , and δ intersects K at one point. We “thicken" δ to a regular neighborhood Nδ in B \ M . Then (Nδ , Nδ ∩ α) is a trivial ball-arc pair. The union M ∪ Nδ must be a ball, because its boundary is a sphere. (Why? Schoenflies!) Hence, (M ∪ Nδ , Nδ ∩ α) is a ball-arc pair as well. As P is prime, either (M ∪ Nδ , Nδ ∩ α) is trivial (CASE 1) or (M ∪ Nδ , Nδ ∩ α) is a copy of (B, α) itself (CASE 2). Looking at case 1, we see M is a solid torus. Do surgery in B as before to remove the annulus A from Σ ∩ ∂B. Repeat on the rest of the annuli. In case two, M is the exterior of α in B, with meridian ∂δ. 0 6 Case 1 Case 2 Without loss, Σ bounds a ball U ⊂ S 3 containing both the summand K1 and M, as well as the boundary torus ∂M . This torus contains K1 and is peripheral to P. This means P must be a summand of K1 = P #K1∗ for some K1∗ . The same torus in S 3 contains K1∗ #K2 which is then Q = K1∗ #K2 , as seen in the following figure. 7 8
© Copyright 2026 Paperzz