Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Notes on the knot concordance invariant Upsilon

Notes on the knot concordance invariant Upsilon CHARLES LIVINGSTON Abstract. Ozsv ath, Stipsicz, and Szab o have de ned a knot concordance invariant  taking values in the group of piecewise linear functions on the closed interval [0; 2]. This paper presents a description of one approach to de ning  and proving its basic properties. 1. Introduction In [9], Ozsv ath, Stipsicz, and Szab o used the Heegaard Floer knot complex CFK (K ) of a knot K  S to de ne a piecewise linear function  (t) with domain [0; 2]. The function K ! K K induces a homomorphism from the smooth knot concordance group to the group of functions on the interval [0; 2]. Among its properties,  (t) provides bounds on the four-genus, g (K ), the K 4 three-genus, g (K ), and, consequently, the concordance genus, g (K ). This note describes a simple 3 c approach to de ning  (t) using CFK (K ) and proving its basic properties. Acknowledgments Thanks go to Jen Hom, Slaven Jabuka, Swatee Naik, Peter Ozsv ath, Shida Wang, and C.-M. Michael Wong for their comments. Matt Hedden pointed out the structure theorem for ltered knot complexes presented in the appendix and its usefulness in simplifying a key proof. Suggestions from the referee led to valuable improvements in the exposition. 2. Knot Complexes We begin by describing the algebraic structure of the Heegaard Floer complex of a knot K , denoted CFK (K ), rst de ned in [7]. This is a vector space over the eld F with two elements. To simplify notation, we write CF(K ) for CFK (K ). Here we summarize its basic properties. The chain complex CF(K ) has an integer valued grading and the boundary map @ is of degree 1. The grading is called the Maslov grading. The grading of a homogeneous element is denoted gr(x). The complex CF(K ) has an Alexander ltration consisting of an increasing sequence of subcomplexes. The ltration level of an element x 2 CF(K ) is denoted Alex(x). There is a similar ltration, called the algebraic ltration, and ltration levels of elements are denoted Alg(x). There is an action of the Laurent polynomial ring F[U; U ] on CF(K ). The action of U commutes with @, lowers gradings by 2, and lowers Alexander and algebraic ltration levels by 1. Let  denote F[U; U ]. As a {module, CF(K ) is free on a nite set of generators, fx g . To simplify notation, we suppress the indexing set. The set of elements fU x g i 1ir i k2Z forms a bil ltered graded basis for CF(K ): for any triple of integers, (g; m; n), the subspace of CF(K ) spanned by elements of grading g, Alexander ltration level less than or equal to m, and algebraic ltration level less than or equal to n, has as basis a subset of fU x g. The singly ltered complex (CF(K ); Alg) with {structure is chain homotopy equivalent to complex T  where 1 2  has grading 0 and ltration level 0, and the boundary map is trivial. (The same statement holds for the Alexander grading, but we do not use this fact.) The author was supported by a Simons Foundation grant and by NSF-DMS-1505586. arXiv:1412.0254v3 [math.GT] 2 Oct 2020 2 CHARLES LIVINGSTON The construction of CF(K ) depends on a series of choices. However, there is a natural de nition of chain homotopy equivalence for graded, bi ltered chain complexes with {action. A key result of [7] is that in this sense, the chain homotopy equivalence class of CF(K ) is a well-de ned knot invariant. As an example, Figure 1 presents a schematic diagram of the complex for the torus knot T (3; 7). As a {module it has nine ltered generators, with algebraic and Alexander ltration levels indicated by the rst and second coordinate, respectively. Five of the generators, indicated with black dots, have grading 0; the four white dots represent generators of grading one. The boundary map is indicated by the arrows. The rest of CF(K ) is the direct sum of the U , k 2 Z, translates of this nite complex; for instance, applying U shifts the diagram one down and to the left. 3. Filtrations We now discuss more general ltrations on vector spaces. In our applications, the vector space will be CF(K ): De nition 3.1. A real-valued (discrete) ltration on a vector space C is a collection of subspaces F = fC g indexed by the real numbers. This collection must satisfy the following properties: (1) C  C if s  s . s s 1 2 1 2 (2) C = [ C . s2R s (3) \ C = f0g. s2R s (4) (discreteness) C =C is nite dimensional when s  s . s s 1 2 2 1 Given a discrete ltration F = fC g on C, we can de ne an associated function on C, which we temporarily also denote by F , given by F (x) = minfs 2 R j x 2 C g. Notice that F ((1; s]) = C . Given an arbitrary real-valued function f on C, one can de ne an associated ltration with C = Span(f ((1; s])). The resulting ltration need not be discrete. Notation. In cases in which more than one ltration might be under consideration, we will write (C;F ) rather than C . s s De nition 3.2. A set of vectors fz g in the real ltered vector space C is called a ltered basis if it is linearly independent and every C has some subset of fz g as a basis. If C is also graded, s i C =  G , then we say the basis is a ltered graded basis if each C \ G has a subset of fz g i s k i i=1 UPSILON NOTES 2 as a basis. -1 -1 0 1 2 3 4 5 6 7 Figure 1. CFK (T (3; 7)) Figure 1. The staircase complex for the torus knot T (3, 7). −1 Notation When the filtration F is understood, we write C = F ((−∞ ,s]). An alternative view of filtration consists of describing C as an nondecreasing nested union of subspaces, C = ∪ C ,where C /C is finite dimensional when s∈ R s s s 2 1 s ≤ s . We can define a filtration F(x) = min{s ∈ S | x ∈ C }. 1 2 s Definition 3.1. Aset of vectors {x } in the real filtered vector space C, is called a filtered basis if it is linearly independent and every C has some subset of {x } s i as a basis. 4. The definition of Υ (t). Let F be a filtration on C = CFK (K). Definition 4.1. Let ν(C, F) = 2 min{s | Image ( H (C ) → H (C)) contains a ∗ s ∗ nontrivial element of grading 0}. A simple argument shows that for any t ∈ [0, 2], the convex combination of t t Alexander and algebraic filtrations, F = Alex+(1− )Alg, defines a real-valued 2 2 discrete filtration on C(K). Definition 4.2. Υ (t)= ν(C, F ). K t 4.1. Example. Consider the knot K = T (3, 7) with CFK (K) as illustrated in Figure ??. The portion of the complex shown has homology F, at Maslov grading t ∞ We use the notation C to denote the complex CFK (K) with filtration F . The set C is thus generated by the bifiltered generators with Alexander and algebraic filtration levels satisfying 2 2 Alex ≤ s +(1 − )Alg. t t UPSILON NOTES 3 4. The definition of the filtration F on CF(K ). For any t 2 [0; 2], the convex combination of Alexander and algebraic ltrations, Alex + (1 )Alg, de nes a real-valued function on CF(K ), to which we associate a ltration denoted F . That is, for all s 2 R, (CF(K );F ) is spanned by all vectors x 2 CF(K ) such that Alex(x) + (1 t s )Alg(x)  s. Theorem 4.1. If 0  t  2, the ltration F on CF(K ) is a ltration by subcomplexes and is discrete. The action of U lowers ltration levels by 1. Proof. To see that these are subcomplexes, suppose that x 2 (CF(K );F ) . Write x = x where t s i t t Alex(x ) + (1 )Alg(x )  s for all i. Since @x = @x , we only need to check that for each i i i 2 2 i, @x 2 (CF(K );F ) . Let x have Alex(x ) = a and Alg(x ) = b. Then Alex(@x ) = a  a and i t s i i i i 0 t t t 0 t 0 t t Alg(@x ) = b  b: Since both and (1 ) are nonnegative, a + (1 )b  a + (1 )b  s, 2 2 2 2 2 2 as desired. The discreteness of the ltration depends on two properties of CF(K ). First, letting g denote the three-genus, g (K ), according to [8] one has g  Alex(x) Alg(x)  g for all x. From this it follows that for given s < s , there are k and k in R such that 1 2 1 2 (CF(K ); Alex)  (CF(K );F )  (CF(K );F )  (CF(K ); Alex) : k t s t s k 1 1 2 2 t t (The values of k and k can be chosen to be s (1 )g and s + (1 )g, respectively, but 1 2 1 2 2 2 we do not need this level of detail.) Second, the Alexander ltration is discrete, so the quotient (CF(K ); Alex) =(CF(K ); Alex) is nite dimensional. k k 2 1 Finally, that U lowers ltration levels by one is immediate. 5. The definition of  (t) For each t 2 [0; 2] and for all s 2 R, the set (CF(K );F )  CF(K ) is a subcomplex. Thus, we t s can make the following de nition. De nition 5.1. Let (CF(K );F ) = minfsj Image (H ((CF(K );F ) ) ! H (CF(K )) is surjectiveg. t 0 t s 0 De nition 5.2.  (t) = 2(CF(K );F ). K t 5.1. Example. Consider the knot K = T (3; 7) with CF(K ) as illustrated in Figure 1. The portion of the complex shown has homology F, at grading 0. The subcomplex (CF(K );F ) is generated by the bi ltered generators with Alexander and alge- t s braic ltration levels satisfying 2 2 (5.1) Alex  s + (1 )Alg: t t Observation The lattice points which contain a ltered generator at ltration level t all lie on a line of slope m = 1 ; with lattice points parametrized by the pair (Alg; Alex). Alternatively, if a line of slope m contains distinct lattice points representing bi ltration levels of generators at the same F ltration level, then t = : 1 m In the diagram for T (3; 7) shown in Figure 1, the illustrated line in the plane corresponds to t = and s = 2. Since the lower half-plane bounded by this line contains a generator of H (CF(K )), while no half plane bounded by a parallel line with smaller value of s contains such a generator, we have ( ) = 2(2) = 4. 5 4 CHARLES LIVINGSTON Continuing with K = T (3; 7), it is now clear that for m < 2 (that is, for t < ), the least s for which (CF(K );F ) contains a generator of H (CF(K )) corresponds to the line through (0; 6), t s 0 t t which has ltration level 6 + (1 )0 = 3t. 2 2 For 2 < m < 1 (that is, for < t < 1), the least s for which (CF(K );F ) contains a generator t s t t of H (CF(K )) corresponds to the line through (2; 2), which has ltration level 2 + (1 )2 = 2. 2 2 Multiplying by 2 and checking the value t = yields 6t if 0  t (t) = T (3;7) 4 if  t  1: 6. An alternative definition of  and In the appendix we prove Theorem A.1, which has as an immediate consequence the following result. Theorem 6.1. The ltered graded chain complex (CF(K );F ) is isomorphic to a ltered graded complex of the form T A; where T  A has the structure of a {module and the isomorphism is a {module isomorphism. The summand T has the properties that: (1) it is isomorphic to  as a {module; (2) the element 1 2  = T has grading 0. Furthermore, A is acyclic as an un ltered complex. Notice that since all gradings in T are even, the boundary operator restricted to T is trivial. When placed in this simple form, the computation of ((CF(K );F )) is simple: it is the F t t ltration level of 1 2  = T . Hence, we have the following result. Corollary 6.2.  (t) equals 2 times the F { ltration level of 1 2  T for the decomposition K t (CF(K );F ) T A. 7. Products and additivity According to [7], there is a (graded) chain homotopy equivelance of complexes CF(K ) CF(K ) ' CF(K # K ) 1  2 1 2 that preserves the {structure. Each of CF(K ), CF(K ) and CF(K # K ) has an algebraic ltration. To distinguish these, we 1 2 1 2 1 2 1;2 write Alg , Alg and Alg . Similarly, the Alexander and F ltrations will be distinguish with superscripts. Momentarily we write CF = CF(K ) and CF = CF(K ). For each t 2 [0; 2] the ltrations F 1 1 2 2 2 1 2 and F on CF and CF induce a ltration F F on CF CF , de ned via: 1 2 1  2 t t t 1 2 (CF CF ;F F ) = 1  2 s t t 1 2 1 2 Image( (CF ;F ) (CF ;F ) ! (CF ;F ) (CF ;F )): s +s =s 1 s F 2 s 1  2 1 2 t 1 t 2 t t Notice that the direct sum is in nite and each summand is in nitely generated. Again, according to [7], for the connected sum of knots, the equivalence CF(K ) CF(K ) ' CF(K # K ) 1  2 1 2 is a ltered equivalence for both the Alexander and algebraic ltrations. To state this explicitly, 1 2 1;2 (CF(K ); Alex ) (CF(K ); Alex ) ' (CF(K # K ); Alex ) 1  2 1 2 and 1 2 1;2 (CF(K ); Alg ) (CF(K ); Alg ) ' (CF(K # K ); Alg ): 1  2 1 2 UPSILON NOTES 5 Theorem 7.1. For all t 2 [0; 1], 1;2 1 2 (CF(K );F ) (CF(K );F ) ' (CF(K # K );F ): 1  2 1 2 t t t Proof. Fix bases fx g and fy g for the free {modules CF(K ) and CF(K ) so that the sets of all i i 1 2 k k translates fU x g and fU y g, k 2 Z, form graded bi ltered bases for CF(K ) and CF(K ) (as F{ i i 1 2 vector spaces). The F-vector space CF(K ) CF(K ) is generated by the set of all tensor products, 1  2 k j fU x U x g, but note that these do not form a basis; for instance, Ux y = x Uy. i l When selecting elements from fU x g, we will sometimes refer to them as x; similarly for y. Note 1;2 1 2 1;2 that in particular, for such basis elements, Alg (x y) = Alg (x) + Alg (y) and Alex (x y) = 1 2 Alex (x) + Alex (y). 1;2 1 2 The proof of the theorem consists of showing that the ltrations F F and F on CF(K ) t t t CF(K ) are the same. 1;2 If an element z 2 CF(K ) CF(K ) has F ltration level s, then it can be written as the 1  2 sum of elements x y with t t Alex(x y) + (1 )Alg(x y)  s: 2 2 This is the same as t t t t Alex(x) + (1 )Alg(x) + Alex(y) + (1 )Alg(y)  s: 2 2 2 2 1 2 1 2 This implies that F (x) +F (y)  s. This in turn implies that (F F )(x y)  s. Thus, for all t t t t 1;2 1 2 z 2 CF(K ) CF(K ), (F F )(z)  F (z). 1  2 t t t 1 2 Similarly, suppose that z 2 CF(K ) CF(K ) has F F ltration level s. Then it is the sum 1  2 t t t t of elements x y, each of which satis es F (x) +F (y)  s. This can be expanded and rewritten as 1 2 t t (Alex(x) + Alex(y)) + (1 )(Alg(x) + Alg(y))  s: 2 2 1;2 1;2 In other words, z is the sum of elements x y with F (x y)  s. Hence, F (x y)  s. t t Theorem 7.1, along with Theorem 6.1, o ers a fast proof of the additivity of . Theorem 7.2. For each t 2 [0; 2],  (t) =  (t) +  (t): K # K K K 1 2 1 2 Proof. One only needs to check this for complexes of the form T  A, as given in Theorem 6.1. Acyclic summands do not a ect the value of  (t). Thus, we only need consider the case of complexes T (K ) T (K ), for which the statement is clear. 1  2 Similarly, Theorem 6.1 o ers a fast proof of the following. Theorem 7.3. For an arbitrary knot K ,  (t) =  (t). K K Proof. According to [7], the complexes CF(K ) and CF(K ) are duals: CF(K ) = CF(K ) . More precisely, CF(K ) is isomorphic to the complex Hom (CF(K );F), having underlying vector space the space of F{homomorphisms with nite dimensional (that is, nite) support. If we x a basis fx g of CF(K ) as a {module so that the set fU x g forms a graded bi ltered i i basis of CF(K ), then we can denote the elements of the dual basis by (U x ) . The dual complex is readily understood in terms of these bases. (1) An easy exercise shows that the action of U on the dual basis is of the form U (U x ) = k1 (U x ) . In particular, the set fx g forms a basis for the {module CF(K ) . (2) For any ltration F on CF(K ), we can de ne a ltration F on the dual space as follows: (CF(K ) ;F ) = f 2 CF(K ) j ((CF(K );F ) 0 ) = 0 for all s > sg: s s The choice of signs ensures that the dual ltration is increasing. Thus, F (x ) = F (x ). (3) The boundary operator for the dual space acts in the expected way with respect to basis elements: if x is a component of @y, then y is a component of @x . 6 CHARLES LIVINGSTON These three observations are easily summarized in terms of diagrams such as in Figure 1: the diagram for CF(K ) is obtained from that for CF(K ) by rotating the gure by 180 degrees around the origin and reversing all the arrows. t t There are two ltrations on CF(K ) of interest. The rst is Alex + (1 )Alg ; the second is 2 2 t t F = ( Alex + (1 )Alg) . By using the chosen basis and its dual basis, it is possible to see that 2 2 these two ltrations are the same, as follows. We use coordinates (i; j) for the plane. For a basis t t vector x, its dual vector x is in F if and only if it lies on or above the line j + (1 )i = t. 2 2 If this is the case, then when rotated 180 degrees about the origin it lies on or below the line t t t  t j + (1 )i = t. These are precisely the dual vectors for which Alex + (1 )Alg  t. 2 2 2 2 The proof of the theorem is now reduced to an elementary calculation for the simple complex T (K ) and its dual T (K ) . 8. Basic properties of  (t) and  (t). We now present some basic results concerning  (t) and its derivative. An initial observation is that  (0) = 0 and, since CF(K ) is nitely generated,  (t) is continuous at 0. Thus, we focus on K K t > 0. Theorem 8.1. (1) For every knot K , (K ) is a continuous piecewise linear function. 0 0 (2) At a nonsingular point of  (t), the value of j (t)j is jijj, where (i; j) is the bi ltration K K level of some ltered generator of CF(K ) with homological grading 0. 0 2 (3) Singularities in  (t) can occur only at values of t such that some line of slope 1 K t 0 0 contains at least two lattice points, (i; j) and (i ; j ), each of which represents the algebraic and Alexander gradings of ltered generators of CF(K ) of homological grading 0. 0 0 0 (4) If  (t) has a singularity at t, then the jump in  (t) at t, denoted  (t), satis es K K K 0 2 0 0 0 0 j (t)j = ji i j for some pair (i; i ) for which there are lattice points (i; j) and (i ; j ) K t as in the previous item. Proof. The proof is discussed in terms of the diagram of the complex, as illustrated for the knot T (3; 7) in the previous section. t t Suppose  (t) = 2s and there is precisely one lattice point (i; j) with j + (1 )i = s which 2 2 represents the bi ltration level of a ltered generator of CF(K ). (This will be the case for all but 0 0 0 a nite number of values of t.) For a nearby t, say t , the value of  (t ) = 2s will be such that 0 0 t t 0 the same vertex (at (i; j)) lies on the line j + (1 )i = s . That is, for all nearby values of t, the 2 2 t t value of s is given by j + (1 )i. Written di erently, 2 2 (t) = 2i + (i j)t: In particular, we see that  (t) is piecewise linear o a nite set. Now consider a singular value of t, at which  (t) = 2s and there are two or more pairs (i; j) t t 2 for which j + (1 )i = s. Notice that this line in the (i; j){plane has slope m = 1 . For t 2 2 t close to t and t < t, we have 0 0 (t ) = 2i + (i j)t 0 0 for one of those pairs (i; j). If t is near t and t > t, then 0 0 0 0 0 (t ) = 2i + (i j )t 0 0 for another (or possibly the same) of these pairs, (i ; j ). Notice that these are equal at t, giving the continuity of  (t). 0 0 We now see that a singularity of  (t) occurs if (j i) 6= (j i ). With these observations, the proofs of (1), (2), and (3) are complete. 0 0 For (4), our computations have shown that the change in  (t), denoted  (t), is given by K K 0 0 0 0 0 (t) = (j j ) (i i ) for some appropriate (i; j) and (i ; j ). Since both are assumed to lie on K UPSILON NOTES 7 2 0 2 0 a line of slope 1 , we have j j = (1 )(i i ), so t t 2 2 0 0 0 0 (t) = (1 )(i i ) (i i ) = (i i ): t t This completes the proof of the theorem. Corollary 8.2. For any knot K and for t = with gcd(p; q) = 1, (t) = kp; where k is some integer if p is odd, or half-integer if p even. t 0 0 0 Proof. By Theorem 8.1 (4), j  (t)j = ji i j for some pair of integers i and i , where there are 2 K two lattice points on a line of slope m = 1 : Thus, we want to constrain the possible di erences between the rst coordinates of such lattice points. p 2qp 2qp For t = , m = . Since gcd(p; q) = 1, in reduced terms, this is either m = q p p q(p=2) or m = if p is odd or even, respectively. Two lattice points on such a line have rst (p=2) coordinates di ering by a multiple of p or of , if p is odd or even, respectively. The completes the proof. 9. The three-genus, g (K ). 0 0 Theorem 9.1. For nonsingular points of  (t), j (t)j  g (K ). K K Proof. According to [8], if K is of genus g, then all elements of CF(K ) have ltration level (i; j), where g  i j  g: It follows immediately from the second statement of Theorem 8.1 that j (t)j  g (K ). We also observe that the genus of K constrains the possible points of singularity of  (t). 0 p Theorem 9.2. Suppose that  (t) has a singularity at t = , with gcd(p; q) = 1. Then: K q If p is odd, q  g (K ). If p is even, q  2g (K ). Proof. Suppose that a line of slope m = , where 0 < b < a contains two distinct points of the form (i; j) with ji jj  g (K ). It follows quickly that the genus bound implies a  2g (K ) b: To express this in terms of t, suppose t = with gcd(p; q) = 1. Then 2 2q p m = 1 = : t p If p is odd, then gcd(2q p; p) = 1. If p is even, say p = 2k, then gcd(2q p; p) = gcd(2q; p) = 2 qk and m = , with q and k relatively prime. In the rst case, with p odd, we have 2q p  2g (K ) p, so q  g (K ). 3 3 In the second case, with p even, we have q k  2g (K ) k, so q  2g (K ). 3 3 8 CHARLES LIVINGSTON 10.  (t) as a knot concordance invariant If knots K and K are concordant, then there is an equality among d{invariants: d(S (K ); s ) = 1 2 1 m 3 N1 N1 3 d(S (K ); s ) for all N 2 Z and m 2 Z,  m  . Here S (K ) denotes N surgery on 2 m N 2 2 N K , d is the Heegaard Floer correction term, and s is a Spin structure, with m given by a speci c enumeration of Spin structures; all are described in [5]. (In the case that N is odd, this range of m includes all possible Spin structures.) If N is large, then d(S (K ); s ) = D(K ) + S(N ), where D(K ) is the largest grading of a class 1 0 z in the homology of CF(K ) for which U z is nontrivial for all k > 0, and S(N ) is some fi0;j0g rational function de ned on the integers, independent of K . In the case that K is slice, we see that the maximal grading D(K ) = D(u), where u is the unknot. This implies that for a slice knot K , D(K ) = 0. We have a nesting of complexes CF(K )  (CF(K );F ) : fi0;j0g t 0 Since (0; 0) is at F ltration level 0, it follows that (CF(K );F )  0; thus  (t)  0. t t K However, K is also slice, so  (t)  0. It follows that  (t) = 0. An additive invariant of K K knots that vanishes on slice knots is a concordance invariant. 11. The concordance-genus The concordance-genus g (K ) of a knot K , de ned in [4], is the minimal genus among all knots concordant to K . Since  (t) is a concordance invariant, the genus bounds in Section 9 apply to the concordance genus. 0 0 Theorem 11.1. For all nonsingular points of  (t), j (t)j  g (K ). The jumps in  (t) occur K c K K p q at rational numbers . For p odd, q  g (K ). If p is even,  g (K ). c c q 2 12. Bounds on the four-genus, g (K ). Let CF(K ) denote the bi ltered subcomplex CF(K ) . We let  (K ) denote the 0;m fi0;jmg minimum value of m such that the homology of CF(K ) contains a nontrivial grading 0 element 0;m of the homology of CF(K ), which we recall is isomorphic to  with 1 at grading 0. There is the following result of Hom and Wu [1], built from work of Rasmussen [10]. (In [1] the invariant  is described; the equivalence with  is presented in [9].) Proposition 12.1 (Proposition 2.4, [1]).   g (K ). Based on this, we show that  (t) provides a bound on g (K ). K 4 Theorem 12.2. For all t 2 [0; 2], j (t)j  tg (K ). K 4 Proof. Since (0; m) is at F ltration level tm=2, we have the containment CF(K )  (CF(K );F ) : 0;m t tm=2 Since CF(K ) contains an element of grading 0 in the homology of CF(K ), so does the subcomplex 0; (CF(K );F ) . Thus, (CF(K );F )  t =2. By the previous proposition, (CF(K );F ) t t =2 t t tg (K )=2. ConsideringK , we have (CF(K );F )  tg (K )=2; it follows that(CF(K );F )  tg (K )=2. t 4 t 4 Combining these yields j(CF(K );F )j  tg (K )=2: t 4 Multiplying by 2 yields the desired conclusion.  UPSILON NOTES 9 13. Crossing change bounds Here we sketch a proof of Proposition 1.10 of [9]. The argument is essentially the same as used in [3] to prove the corresponding fact about  (K ). Theorem 13.1. Let K and K be knots with identical diagrams, except at one crossing which is either negative or positive, respectively. Then for t 2 [0; 1], (t)   (t)   (t) + t: K K K + + Proof. First note that K #K can be changed into the slice knot K #K by changing a + + + negative crossing to positive. Thus, g (K #K )  1. It follows that 4 + (13.1) t   (t)  (t)  t: K K Next, note that K #K # T (2; 3) can be changed into the slice knot K #K by changing + + + one negative crossing to positive and one positive crossing to negative. Thus, it too has four-genus at most 1: it bounds a singular disk with two singularities of opposite sign, and these can be tubed together. A simple computation for T (2; 3) yields  (t) = t for 0  t  1. Thus, T (2;3) t   (t)  (t) t  t; K K which we rewrite as (13.2) 0   (t)  (t)  2t: K K Combining Equations 13.1 and 13.2, 0   (t)  (t)  t: K K Adding  (t) to all terms yields the desired conclusion, (t)   (t)   (t) + t: K K K + + Note This argument can be easily modi ed to show that if there is a singular concordance from K to J with a single positive double point, then  (t)   (t)   (t) + t: K J K 14. The Ozsvath-Szabo  -invariant and  (t) for small t For small t,  (t) is determined by the  invariant de ned in [6]. We review the de nition below. Here is the statement of the result. Theorem 14.1. For t small,  (t) =  (K )t. c c The subquotient complex CF(K ) =CF(K ) will be denoted CF(K ). (Usually, CF is fi0g fi<0g written CFK.) It is ltered by the Alexander ltration and has homology F, supported in grad- ing 0. The invariant  (K ) is de ned to be the least integer  such that the map on homology c c H (CF(K ) ) ! H (CF(K )) F is surjective. 0 fjg 0 We wish to relate  (K ) =  to an invariant of CF(K ). The needed technical result is the following. Lemma 14.2. If  (K ) =  , then there is a cycle w 2 CF(K ) representing a nontrivial fi0;jg[fi<0g element in H (CF(K )). Proof. From the de nition of  we see that there is a chain x 2 CF(K ) that in the fi0;jg[fi<0g c c quotient CF(K ) is a cycle that represents a generator of H (CF(K )). Since the chain x represents a cycle in CF(K ), it has the property that @x = y, where y 2 CF(K ) . Note that y is a cycle and gr(y) = 1. Since H (CF(K ) ) = 0, there is a chain z 2 i<0 1 i<0 CF(K ) with @z = y. Thus, x +z is a cycle in CF(K ) . The map H (CF(K ) ) ! i<0 fi0;jg[fi<0g 0 i0 H (CF(K )) is an isomorphism; both groups are isomorphic to F. Thus, x +z represents a generator of H (CF(K ) ). The map H (CF(K ) ) ! H (CF(K )) is an isomorphism, completing the 0 i0 0 i0 0 proof.  10 CHARLES LIVINGSTON Proof, Theorem 14.1. For t small, we consider the ltration F and the ltration level s =  . Then one has CF(K ) = CF(K ) . By Lemma 14.2, this subcomplex contains a cycle that s fi0;jg[fi<0g represents an element of grading 0 in H (CF(K )). Thus, for this F ltration,    . On the other hand, suppose that  <  . Then there would exist a cycle z 2 CF(K ) fi0;j1g[fi<0g representing a generator of H (CF(K )) of grading 0. However, the image of z in CF(K ) would be an c c element in CF(K ) that represents a generator of H (CF(K )). But  is by de nition the lowest 1 0 level at which this can occur. Thus, we see that  =  . To conclude, recall that  (t) = 2, so  (t) =  (K )t, as desired. K K Note. With care, one can check that in this argument, the condition that t be small can be made precise by requiring that t < 1=g (K ). Of course, once the result is established for some set of small t, then Theorem 9.2 provides the bound t < 1=g (K ). 15. Equivalence of definitions of  (t) In this section we explain why  (t) as de ned here agrees with that of [9]. Beginning with CF(K ), a new complex tCF(K ) can be constructed as follows. As an F{vector space, 1=n 1=n tCF(K ) = CF(K ) F[v ; v ]; 1=n 2 1=n 1=n where U acts on F[v ] via multiplication by v . This has the structure of an F[v ; v ]{module. 0 1=n 1=n To simplify notation, we write  = F[v ; v ]. There are (rational) ltrations Alg and Alex on tCF(K ) which are consistent with those on the 1=n 2 {submodule CF(K ). The action of v lowers ltration levels by 1=2n. Thus, U = v lowers ltration levels by 1, as it should. Similarly, the Maslov grading M (x) naturally extends to tCF(K ) 1=n 2 so that the action of v lowers this grading by 1=n, and thus U = v continues to lower the Maslov grading by 2. There is a rational grading on tCF(K ) de ned via the Maslov grading, M , along with the algebraic and Alexander ltrations. If x is an element at ltration level (i; j), then: (15.1) gr (x) = M (x) t(j i): (In [9], only generators at algebraic ltration level 0 are used to de ne gr , so i = 0 and the formula gr (x) = M (x) tAlex(x) is presented.) One checks that U to lowers gr {gradings by 2, so on the t t 1=n extension to tCF(K ), v lowers gradings by 1 and v lowers gradings by 1=n. If x is a ltered generator of CF(K ) with @x = y , then the boundary @ is de ned so that l t @ x = v y 2 tCF(K ), with the values of given explicitly in [9]. This extends naturally to a t l l boundary operator on all of tCF(K ). Given that the operator @ is well-de ned, it is a simple matter to determine its value. Suppose that x is a ltered generator of CF(K ) at ltration level (i; j), Maslov grading g, and suppose also 0 0 that @x = y . Let y denote one of the terms in this sum, at ltration level (i ; j ), necessarily of grading g 1. Then viewed as an element of tCF, x is of grading g t(j i), and y has grading 0 0 g 1t(j i ). In @ x, the term v y appears, and is such that gr. (v y) = gr (x) 1. Rewriting t t 0 0 this, we have (g 1) t(j i ) = g t(j i) 1. That is, 0 0 (15.2) = t((j j ) (i i )): As two examples, Figure 2 illustrates the complexes tCF(K ) for K = T (3; 7), with t = and t = 2. The construction is straightforward using Equation 15.1 and the fact that v shifts along the diagonal a distance of 1=2 down and to the left. The portion of the complex illustrated was chosen because its homology is F in grading 0 and represents the generator of the homology of tCF , the full complex consists of the illustrated complex along with in grading 0. In the case that t = 3 UPSILON NOTES 11 all its translates a distance , k 2 Z, along the diagonal. In the case of t = 2, the translates are those a distance along the diagonal. Figure 2. tCF (T (3; 7)) and tCF (T (3; 7)) t=1=3 t=2 It is apparent from these examples that the Alexander ltration is not a ltration of the chain complex, since some arrows increase the Alexander ltration level. However, as is easily veri ed, the algebraic ltration is a ltration on the chain complex. De nition 15.1. For t = , denote by tCFK (K ) the complex tCF(K ) . i0 Note. In [9], this complex is denoted tCFK(K ). In fact, it is the complex that is explicitly con- structed. Here we rst introduced the in nity complex to be consistent with our earlier constructions. De nition 15.2. For t = ,  (t) is the maximal grading of a class in the homology of tCFK (K ) that maps to a nontrivial element in the homology of tCF(K ). Equivalently, it is the maximal grading of a class in the homology of tCFK (K ) which is not in the kernel of v for all k > 0. Lemma 15.3. The value of  (t) as just de ned is equal to 2s, where s is the least number for which the homology of tCF(K ) contains an element of grading 0 that represents a nontrivial is element of the homology of tCF(K ). Proof. This follows from a simple change of coordinates. 15.1. The two de nitions of  (t) agree. Suppose that using this de nition of  (t), we have K K (t) = 2s. This implies that tCF(K ) contains a cycle z representing a nontrivial generator of K is grading 0 in the homology of tCF(K ). Write z = x , where the x are ltered generators. Some l l x has ltration level (s; j), and none of the x has algebraic ltration level greater than s. l l From the regrading formula given in Equation 15.1, gr (x) = M (x)t(ji), we see that generators of CF(K ) at ltration level (i; j) and grading 0 yield generators of grading 0 in tCF(K ) at ltration t t level (i + (j i); j + (j i)). (Recall that shifting down and to the left by t units decreases the 2 2 grading by 2t.) We are thus led to consider the transformation t t t t (i; j) 7! ((1 )i + j; i + (1 + )j): 2 2 2 2 12 CHARLES LIVINGSTON Its inverse is given by t t t t (i; j) 7! ((1 + )i j; i + (1 )j): 2 2 2 2 Under this transformation, for a xed value of s, the vertical line f(s; z) j z 2 Rg, is carried to the t t t t line (in the CF(K ){plane) f((1 + )s z; s + (1 )z) j z 2 Rg: Relabeling the coordinate system 2 2 2 2 (x; y), this is the line 2 2 y = (1 )x + ( )s: t t Comparing with Equation 5.1, we see that the homology of the ltered complex (CF(K );F ) t s contains a generator of grading 0 that is nontrivial in the homology of CF(K ), and that this is not the case for (CF(K );F ) for any s < s. Thus, the value of  (t) as de ned in Section 5 is 2s, t s K and the de nitions agree. Appendix A. A structure theorem for CF(K ). In [2, Chapter 11], vertical and horizontal reductions of CF(K ) are discussed. That presentation applies to the ltered complex (CF(K );F ), but adjustments in the details would be required be- cause, for instance, the horizontal and vertical ltrations are integer valued rather than being real ltrations. Since the argument in the present case is straightforward, we present it in detail. Viewed as a {module, CF(K ) is freely generated by a nite set fw g . We again simplify i 1im notation by suppressing the indexing set and write fw g. This set can be chosen so that the set fU w g forms a bi ltered graded basis for the F{complex CF(K ). We will refer to any such set i k2Z fw g as a {basis for CF(K ). A {module change of basis among the w that preserves gradings i i and ltration levels induces a change of bi ltered graded basis for the F{complex CF(K ). We will refer to any such change of basis as a {change of basis of CF(K ). Analogous notation will be used when working with the ltered graded complex (CF(K );F ). Theorem A.1. Let t 2 [0; 2]. As a {module, CF(K ) has a basis f ; ; : : : ; g, inducing a 1 k splitting of CF(K ) (as a {module) as the direct sum CF(K ) T A, where T is freely generated by and A is freely generated by f ; : : : ; g. This splitting has the following properties. 1 k (CF(K );F ) T A as a ltered graded F-complex. The complex T has ltered graded basis fU g , the boundary map is trivial on T , and k2Z gr( ) = 0. The complex A has ltered graded basis fU g and has trivial homology: H (A) = 0. i k2Z Proof. We begin with the {generating set of CF(K ), fw g. By replacing generators with their U translates and renaming the generators, we can decompose this into two subsets: fx g, all of grading 0, and fy g, all of grading 1. To simplify notation, we abbreviate the ltered graded F{complex (CF(K );F ) by CF . t t (1) Let A be a cycle in CF having the least ltration level among cycles representing nontrivial classes in H (CF ): After reordering the generators, we can write A = x + + x , with 0 t 1 k the ltration levels nonincreasing. Replacing x with x +  + x as the rst generating 1 1 k element (over ) induces a ltered change of basis for CF . Thus, the rst element of the {basis, which we now denote A , is a cycle of least ltration level representing a nontrivial element of H (CF ). 0 t (2) Consider the set of all generating elements y that have the property that A is a component i 1 of @y . After reordering the basis, we can assume these are fy ; y ; : : : ; y g for some k, and i 1 2 k that the ltrations are in nondecreasing order. Make the {change of basis that replaces each y , 2  i  k, with y + y . This induces a ltered change of basis of CF . Now, the i i 1 t only generator having A as a component of its boundary is y , which we relabel B . 1 1 1 (3) After perhaps reordering the x , we have either @B = A or @B = A + x + + x for i 1 1 1 1 2 k some k  2, with the ltration levels nonincreasing. Since @ = 0, it follows that B is not a component of any element in the image of @. UPSILON NOTES 13 If @B = A , then we see that fA ; B g generates an acyclic summand of CF , and thus 1 1 1 1 t A would not represent a nontrivial element in homology. We have @B = A +x + x for some k  2. Make the {change of basis that replaces 1 1 2 k x with x +  x , now calling this new element A . Then @B = A + A . Note that 2 2 k 2 1 1 2 since A is a cycle and A + A = @B is a cycle, that A is a cycle representing the same 1 1 2 1 2 homology class as A . Hence the ltration level of A is greater than or equal to that of A . 1 2 1 (4) We now repeat the previous argument, making a change of basis so that the only basis elements with boundary that include A as a component are B and perhaps a second 2 1 generator that we denote B . (5) This step-by-step procedure must eventually stop, at which time there is constructed a summand of the F{complex CF D = A B ! A B ! A  ! B ! A : 1 1 2 2 3 k1 k Note that the process must end with an A ; if it stopped with a B , the resulting complex k k would be acyclic and thus not contain a nontrivial element in homology. This complex is a summand of the complex CF . Note that D is a summand of a direct sum decomposition of CF , as a subcomplex and also as a submodule of the {module. (6) Since A has the lowest ltration level among the A , we can replace each A with A + A 1 i i 1 i to form a new basis. The complex then splits in the following way: A  [B ! (A + A ) B ! (A + A )  ! B ! (A + A )] : 1 1 1 2 2 1 3 k1 1 k We let T = A . It satis es the required conditions of the theorem. Since as a {module, H (T ) H (CF ), the complementary summand to T must be acyclic. That complementary summand yields the summand A in the statement of the theorem. References [1] J. Hom and Z. Wu, Four-ball genus bounds and a re nement of the Ozsv ath-Szab o tau-invariant, arxiv.org/ abs/1401.1565. [2] R. Lipshitz, P. Ozsv ath, and D. Thurston, Bordered Heegaard Floer homology, arxiv.org/abs/0810.0687. [3] C. Livingston, Computations of the Ozsv ath-Szab o knot concordance invariant, Geom. Topol. 8 (2004), 735{ [4] C. Livingston, The concordance genus of knots, Alg. and Geom. Top. 4 (2004), 1{22. [5] P. Ozsv ath and Z. Szab o, Absolutely graded Floer homologies and intersection forms for four-manifolds with boundary, Adv. Math. 173 (2003), 179{261. [6] P. Ozsv ath and Z. Szab o, Knot Floer homology and the four-ball genus, Geom. Topol. 7 (2003), 615{639. [7] P. Ozsv ath and Z. Szab o, Holomorphic disks and knot invariants, Adv. Math. 186 (2004), 58{116. [8] P. Ozsv ath and Z. Szab o, Holomorphic disks and genus bounds, Geom. Topol. 8 (2004), 311{334. [9] P. Ozsv ath, A. Stipsicz, and Z. Szab o, Concordance homomorphisms from knot Floer homology, arxiv.org/ abs/1407.1795. [10] J. Rasmussen, Floer homology and knot complements, arxiv.org/abs/math/0306378. Charles Livingston: Department of Mathematics, Indiana University, Bloomington, IN 47405 Email address : livingst@indiana.edu http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Mathematics arXiv (Cornell University)

Notes on the knot concordance invariant Upsilon

Livingston, Charles
Mathematics , Volume 2020 (1412) – Nov 30, 2014
Download PDF
13 pages

Loading next page...
 
/lp/arxiv-cornell-university/notes-on-the-knot-concordance-invariant-upsilon-fyYcc37uR1
ISSN
1472-2739
eISSN
ARCH-3343
DOI
10.2140/agt.2017.17.111
Publisher site
See Article on Publisher Site

Abstract

CHARLES LIVINGSTON Abstract. Ozsv ath, Stipsicz, and Szab o have de ned a knot concordance invariant  taking values in the group of piecewise linear functions on the closed interval [0; 2]. This paper presents a description of one approach to de ning  and proving its basic properties. 1. Introduction In [9], Ozsv ath, Stipsicz, and Szab o used the Heegaard Floer knot complex CFK (K ) of a knot K  S to de ne a piecewise linear function  (t) with domain [0; 2]. The function K ! K K induces a homomorphism from the smooth knot concordance group to the group of functions on the interval [0; 2]. Among its properties,  (t) provides bounds on the four-genus, g (K ), the K 4 three-genus, g (K ), and, consequently, the concordance genus, g (K ). This note describes a simple 3 c approach to de ning  (t) using CFK (K ) and proving its basic properties. Acknowledgments Thanks go to Jen Hom, Slaven Jabuka, Swatee Naik, Peter Ozsv ath, Shida Wang, and C.-M. Michael Wong for their comments. Matt Hedden pointed out the structure theorem for ltered knot complexes presented in the appendix and its usefulness in simplifying a key proof. Suggestions from the referee led to valuable improvements in the exposition. 2. Knot Complexes We begin by describing the algebraic structure of the Heegaard Floer complex of a knot K , denoted CFK (K ), rst de ned in [7]. This is a vector space over the eld F with two elements. To simplify notation, we write CF(K ) for CFK (K ). Here we summarize its basic properties. The chain complex CF(K ) has an integer valued grading and the boundary map @ is of degree 1. The grading is called the Maslov grading. The grading of a homogeneous element is denoted gr(x). The complex CF(K ) has an Alexander ltration consisting of an increasing sequence of subcomplexes. The ltration level of an element x 2 CF(K ) is denoted Alex(x). There is a similar ltration, called the algebraic ltration, and ltration levels of elements are denoted Alg(x). There is an action of the Laurent polynomial ring F[U; U ] on CF(K ). The action of U commutes with @, lowers gradings by 2, and lowers Alexander and algebraic ltration levels by 1. Let  denote F[U; U ]. As a {module, CF(K ) is free on a nite set of generators, fx g . To simplify notation, we suppress the indexing set. The set of elements fU x g i 1ir i k2Z forms a bil ltered graded basis for CF(K ): for any triple of integers, (g; m; n), the subspace of CF(K ) spanned by elements of grading g, Alexander ltration level less than or equal to m, and algebraic ltration level less than or equal to n, has as basis a subset of fU x g. The singly ltered complex (CF(K ); Alg) with {structure is chain homotopy equivalent to complex T  where 1 2  has grading 0 and ltration level 0, and the boundary map is trivial. (The same statement holds for the Alexander grading, but we do not use this fact.) The author was supported by a Simons Foundation grant and by NSF-DMS-1505586. arXiv:1412.0254v3 [math.GT] 2 Oct 2020 2 CHARLES LIVINGSTON The construction of CF(K ) depends on a series of choices. However, there is a natural de nition of chain homotopy equivalence for graded, bi ltered chain complexes with {action. A key result of [7] is that in this sense, the chain homotopy equivalence class of CF(K ) is a well-de ned knot invariant. As an example, Figure 1 presents a schematic diagram of the complex for the torus knot T (3; 7). As a {module it has nine ltered generators, with algebraic and Alexander ltration levels indicated by the rst and second coordinate, respectively. Five of the generators, indicated with black dots, have grading 0; the four white dots represent generators of grading one. The boundary map is indicated by the arrows. The rest of CF(K ) is the direct sum of the U , k 2 Z, translates of this nite complex; for instance, applying U shifts the diagram one down and to the left. 3. Filtrations We now discuss more general ltrations on vector spaces. In our applications, the vector space will be CF(K ): De nition 3.1. A real-valued (discrete) ltration on a vector space C is a collection of subspaces F = fC g indexed by the real numbers. This collection must satisfy the following properties: (1) C  C if s  s . s s 1 2 1 2 (2) C = [ C . s2R s (3) \ C = f0g. s2R s (4) (discreteness) C =C is nite dimensional when s  s . s s 1 2 2 1 Given a discrete ltration F = fC g on C, we can de ne an associated function on C, which we temporarily also denote by F , given by F (x) = minfs 2 R j x 2 C g. Notice that F ((1; s]) = C . Given an arbitrary real-valued function f on C, one can de ne an associated ltration with C = Span(f ((1; s])). The resulting ltration need not be discrete. Notation. In cases in which more than one ltration might be under consideration, we will write (C;F ) rather than C . s s De nition 3.2. A set of vectors fz g in the real ltered vector space C is called a ltered basis if it is linearly independent and every C has some subset of fz g as a basis. If C is also graded, s i C =  G , then we say the basis is a ltered graded basis if each C \ G has a subset of fz g i s k i i=1 UPSILON NOTES 2 as a basis. -1 -1 0 1 2 3 4 5 6 7 Figure 1. CFK (T (3; 7)) Figure 1. The staircase complex for the torus knot T (3, 7). −1 Notation When the filtration F is understood, we write C = F ((−∞ ,s]). An alternative view of filtration consists of describing C as an nondecreasing nested union of subspaces, C = ∪ C ,where C /C is finite dimensional when s∈ R s s s 2 1 s ≤ s . We can define a filtration F(x) = min{s ∈ S | x ∈ C }. 1 2 s Definition 3.1. Aset of vectors {x } in the real filtered vector space C, is called a filtered basis if it is linearly independent and every C has some subset of {x } s i as a basis. 4. The definition of Υ (t). Let F be a filtration on C = CFK (K). Definition 4.1. Let ν(C, F) = 2 min{s | Image ( H (C ) → H (C)) contains a ∗ s ∗ nontrivial element of grading 0}. A simple argument shows that for any t ∈ [0, 2], the convex combination of t t Alexander and algebraic filtrations, F = Alex+(1− )Alg, defines a real-valued 2 2 discrete filtration on C(K). Definition 4.2. Υ (t)= ν(C, F ). K t 4.1. Example. Consider the knot K = T (3, 7) with CFK (K) as illustrated in Figure ??. The portion of the complex shown has homology F, at Maslov grading t ∞ We use the notation C to denote the complex CFK (K) with filtration F . The set C is thus generated by the bifiltered generators with Alexander and algebraic filtration levels satisfying 2 2 Alex ≤ s +(1 − )Alg. t t UPSILON NOTES 3 4. The definition of the filtration F on CF(K ). For any t 2 [0; 2], the convex combination of Alexander and algebraic ltrations, Alex + (1 )Alg, de nes a real-valued function on CF(K ), to which we associate a ltration denoted F . That is, for all s 2 R, (CF(K );F ) is spanned by all vectors x 2 CF(K ) such that Alex(x) + (1 t s )Alg(x)  s. Theorem 4.1. If 0  t  2, the ltration F on CF(K ) is a ltration by subcomplexes and is discrete. The action of U lowers ltration levels by 1. Proof. To see that these are subcomplexes, suppose that x 2 (CF(K );F ) . Write x = x where t s i t t Alex(x ) + (1 )Alg(x )  s for all i. Since @x = @x , we only need to check that for each i i i 2 2 i, @x 2 (CF(K );F ) . Let x have Alex(x ) = a and Alg(x ) = b. Then Alex(@x ) = a  a and i t s i i i i 0 t t t 0 t 0 t t Alg(@x ) = b  b: Since both and (1 ) are nonnegative, a + (1 )b  a + (1 )b  s, 2 2 2 2 2 2 as desired. The discreteness of the ltration depends on two properties of CF(K ). First, letting g denote the three-genus, g (K ), according to [8] one has g  Alex(x) Alg(x)  g for all x. From this it follows that for given s < s , there are k and k in R such that 1 2 1 2 (CF(K ); Alex)  (CF(K );F )  (CF(K );F )  (CF(K ); Alex) : k t s t s k 1 1 2 2 t t (The values of k and k can be chosen to be s (1 )g and s + (1 )g, respectively, but 1 2 1 2 2 2 we do not need this level of detail.) Second, the Alexander ltration is discrete, so the quotient (CF(K ); Alex) =(CF(K ); Alex) is nite dimensional. k k 2 1 Finally, that U lowers ltration levels by one is immediate. 5. The definition of  (t) For each t 2 [0; 2] and for all s 2 R, the set (CF(K );F )  CF(K ) is a subcomplex. Thus, we t s can make the following de nition. De nition 5.1. Let (CF(K );F ) = minfsj Image (H ((CF(K );F ) ) ! H (CF(K )) is surjectiveg. t 0 t s 0 De nition 5.2.  (t) = 2(CF(K );F ). K t 5.1. Example. Consider the knot K = T (3; 7) with CF(K ) as illustrated in Figure 1. The portion of the complex shown has homology F, at grading 0. The subcomplex (CF(K );F ) is generated by the bi ltered generators with Alexander and alge- t s braic ltration levels satisfying 2 2 (5.1) Alex  s + (1 )Alg: t t Observation The lattice points which contain a ltered generator at ltration level t all lie on a line of slope m = 1 ; with lattice points parametrized by the pair (Alg; Alex). Alternatively, if a line of slope m contains distinct lattice points representing bi ltration levels of generators at the same F ltration level, then t = : 1 m In the diagram for T (3; 7) shown in Figure 1, the illustrated line in the plane corresponds to t = and s = 2. Since the lower half-plane bounded by this line contains a generator of H (CF(K )), while no half plane bounded by a parallel line with smaller value of s contains such a generator, we have ( ) = 2(2) = 4. 5 4 CHARLES LIVINGSTON Continuing with K = T (3; 7), it is now clear that for m < 2 (that is, for t < ), the least s for which (CF(K );F ) contains a generator of H (CF(K )) corresponds to the line through (0; 6), t s 0 t t which has ltration level 6 + (1 )0 = 3t. 2 2 For 2 < m < 1 (that is, for < t < 1), the least s for which (CF(K );F ) contains a generator t s t t of H (CF(K )) corresponds to the line through (2; 2), which has ltration level 2 + (1 )2 = 2. 2 2 Multiplying by 2 and checking the value t = yields 6t if 0  t (t) = T (3;7) 4 if  t  1: 6. An alternative definition of  and In the appendix we prove Theorem A.1, which has as an immediate consequence the following result. Theorem 6.1. The ltered graded chain complex (CF(K );F ) is isomorphic to a ltered graded complex of the form T A; where T  A has the structure of a {module and the isomorphism is a {module isomorphism. The summand T has the properties that: (1) it is isomorphic to  as a {module; (2) the element 1 2  = T has grading 0. Furthermore, A is acyclic as an un ltered complex. Notice that since all gradings in T are even, the boundary operator restricted to T is trivial. When placed in this simple form, the computation of ((CF(K );F )) is simple: it is the F t t ltration level of 1 2  = T . Hence, we have the following result. Corollary 6.2.  (t) equals 2 times the F { ltration level of 1 2  T for the decomposition K t (CF(K );F ) T A. 7. Products and additivity According to [7], there is a (graded) chain homotopy equivelance of complexes CF(K ) CF(K ) ' CF(K # K ) 1  2 1 2 that preserves the {structure. Each of CF(K ), CF(K ) and CF(K # K ) has an algebraic ltration. To distinguish these, we 1 2 1 2 1 2 1;2 write Alg , Alg and Alg . Similarly, the Alexander and F ltrations will be distinguish with superscripts. Momentarily we write CF = CF(K ) and CF = CF(K ). For each t 2 [0; 2] the ltrations F 1 1 2 2 2 1 2 and F on CF and CF induce a ltration F F on CF CF , de ned via: 1 2 1  2 t t t 1 2 (CF CF ;F F ) = 1  2 s t t 1 2 1 2 Image( (CF ;F ) (CF ;F ) ! (CF ;F ) (CF ;F )): s +s =s 1 s F 2 s 1  2 1 2 t 1 t 2 t t Notice that the direct sum is in nite and each summand is in nitely generated. Again, according to [7], for the connected sum of knots, the equivalence CF(K ) CF(K ) ' CF(K # K ) 1  2 1 2 is a ltered equivalence for both the Alexander and algebraic ltrations. To state this explicitly, 1 2 1;2 (CF(K ); Alex ) (CF(K ); Alex ) ' (CF(K # K ); Alex ) 1  2 1 2 and 1 2 1;2 (CF(K ); Alg ) (CF(K ); Alg ) ' (CF(K # K ); Alg ): 1  2 1 2 UPSILON NOTES 5 Theorem 7.1. For all t 2 [0; 1], 1;2 1 2 (CF(K );F ) (CF(K );F ) ' (CF(K # K );F ): 1  2 1 2 t t t Proof. Fix bases fx g and fy g for the free {modules CF(K ) and CF(K ) so that the sets of all i i 1 2 k k translates fU x g and fU y g, k 2 Z, form graded bi ltered bases for CF(K ) and CF(K ) (as F{ i i 1 2 vector spaces). The F-vector space CF(K ) CF(K ) is generated by the set of all tensor products, 1  2 k j fU x U x g, but note that these do not form a basis; for instance, Ux y = x Uy. i l When selecting elements from fU x g, we will sometimes refer to them as x; similarly for y. Note 1;2 1 2 1;2 that in particular, for such basis elements, Alg (x y) = Alg (x) + Alg (y) and Alex (x y) = 1 2 Alex (x) + Alex (y). 1;2 1 2 The proof of the theorem consists of showing that the ltrations F F and F on CF(K ) t t t CF(K ) are the same. 1;2 If an element z 2 CF(K ) CF(K ) has F ltration level s, then it can be written as the 1  2 sum of elements x y with t t Alex(x y) + (1 )Alg(x y)  s: 2 2 This is the same as t t t t Alex(x) + (1 )Alg(x) + Alex(y) + (1 )Alg(y)  s: 2 2 2 2 1 2 1 2 This implies that F (x) +F (y)  s. This in turn implies that (F F )(x y)  s. Thus, for all t t t t 1;2 1 2 z 2 CF(K ) CF(K ), (F F )(z)  F (z). 1  2 t t t 1 2 Similarly, suppose that z 2 CF(K ) CF(K ) has F F ltration level s. Then it is the sum 1  2 t t t t of elements x y, each of which satis es F (x) +F (y)  s. This can be expanded and rewritten as 1 2 t t (Alex(x) + Alex(y)) + (1 )(Alg(x) + Alg(y))  s: 2 2 1;2 1;2 In other words, z is the sum of elements x y with F (x y)  s. Hence, F (x y)  s. t t Theorem 7.1, along with Theorem 6.1, o ers a fast proof of the additivity of . Theorem 7.2. For each t 2 [0; 2],  (t) =  (t) +  (t): K # K K K 1 2 1 2 Proof. One only needs to check this for complexes of the form T  A, as given in Theorem 6.1. Acyclic summands do not a ect the value of  (t). Thus, we only need consider the case of complexes T (K ) T (K ), for which the statement is clear. 1  2 Similarly, Theorem 6.1 o ers a fast proof of the following. Theorem 7.3. For an arbitrary knot K ,  (t) =  (t). K K Proof. According to [7], the complexes CF(K ) and CF(K ) are duals: CF(K ) = CF(K ) . More precisely, CF(K ) is isomorphic to the complex Hom (CF(K );F), having underlying vector space the space of F{homomorphisms with nite dimensional (that is, nite) support. If we x a basis fx g of CF(K ) as a {module so that the set fU x g forms a graded bi ltered i i basis of CF(K ), then we can denote the elements of the dual basis by (U x ) . The dual complex is readily understood in terms of these bases. (1) An easy exercise shows that the action of U on the dual basis is of the form U (U x ) = k1 (U x ) . In particular, the set fx g forms a basis for the {module CF(K ) . (2) For any ltration F on CF(K ), we can de ne a ltration F on the dual space as follows: (CF(K ) ;F ) = f 2 CF(K ) j ((CF(K );F ) 0 ) = 0 for all s > sg: s s The choice of signs ensures that the dual ltration is increasing. Thus, F (x ) = F (x ). (3) The boundary operator for the dual space acts in the expected way with respect to basis elements: if x is a component of @y, then y is a component of @x . 6 CHARLES LIVINGSTON These three observations are easily summarized in terms of diagrams such as in Figure 1: the diagram for CF(K ) is obtained from that for CF(K ) by rotating the gure by 180 degrees around the origin and reversing all the arrows. t t There are two ltrations on CF(K ) of interest. The rst is Alex + (1 )Alg ; the second is 2 2 t t F = ( Alex + (1 )Alg) . By using the chosen basis and its dual basis, it is possible to see that 2 2 these two ltrations are the same, as follows. We use coordinates (i; j) for the plane. For a basis t t vector x, its dual vector x is in F if and only if it lies on or above the line j + (1 )i = t. 2 2 If this is the case, then when rotated 180 degrees about the origin it lies on or below the line t t t  t j + (1 )i = t. These are precisely the dual vectors for which Alex + (1 )Alg  t. 2 2 2 2 The proof of the theorem is now reduced to an elementary calculation for the simple complex T (K ) and its dual T (K ) . 8. Basic properties of  (t) and  (t). We now present some basic results concerning  (t) and its derivative. An initial observation is that  (0) = 0 and, since CF(K ) is nitely generated,  (t) is continuous at 0. Thus, we focus on K K t > 0. Theorem 8.1. (1) For every knot K , (K ) is a continuous piecewise linear function. 0 0 (2) At a nonsingular point of  (t), the value of j (t)j is jijj, where (i; j) is the bi ltration K K level of some ltered generator of CF(K ) with homological grading 0. 0 2 (3) Singularities in  (t) can occur only at values of t such that some line of slope 1 K t 0 0 contains at least two lattice points, (i; j) and (i ; j ), each of which represents the algebraic and Alexander gradings of ltered generators of CF(K ) of homological grading 0. 0 0 0 (4) If  (t) has a singularity at t, then the jump in  (t) at t, denoted  (t), satis es K K K 0 2 0 0 0 0 j (t)j = ji i j for some pair (i; i ) for which there are lattice points (i; j) and (i ; j ) K t as in the previous item. Proof. The proof is discussed in terms of the diagram of the complex, as illustrated for the knot T (3; 7) in the previous section. t t Suppose  (t) = 2s and there is precisely one lattice point (i; j) with j + (1 )i = s which 2 2 represents the bi ltration level of a ltered generator of CF(K ). (This will be the case for all but 0 0 0 a nite number of values of t.) For a nearby t, say t , the value of  (t ) = 2s will be such that 0 0 t t 0 the same vertex (at (i; j)) lies on the line j + (1 )i = s . That is, for all nearby values of t, the 2 2 t t value of s is given by j + (1 )i. Written di erently, 2 2 (t) = 2i + (i j)t: In particular, we see that  (t) is piecewise linear o a nite set. Now consider a singular value of t, at which  (t) = 2s and there are two or more pairs (i; j) t t 2 for which j + (1 )i = s. Notice that this line in the (i; j){plane has slope m = 1 . For t 2 2 t close to t and t < t, we have 0 0 (t ) = 2i + (i j)t 0 0 for one of those pairs (i; j). If t is near t and t > t, then 0 0 0 0 0 (t ) = 2i + (i j )t 0 0 for another (or possibly the same) of these pairs, (i ; j ). Notice that these are equal at t, giving the continuity of  (t). 0 0 We now see that a singularity of  (t) occurs if (j i) 6= (j i ). With these observations, the proofs of (1), (2), and (3) are complete. 0 0 For (4), our computations have shown that the change in  (t), denoted  (t), is given by K K 0 0 0 0 0 (t) = (j j ) (i i ) for some appropriate (i; j) and (i ; j ). Since both are assumed to lie on K UPSILON NOTES 7 2 0 2 0 a line of slope 1 , we have j j = (1 )(i i ), so t t 2 2 0 0 0 0 (t) = (1 )(i i ) (i i ) = (i i ): t t This completes the proof of the theorem. Corollary 8.2. For any knot K and for t = with gcd(p; q) = 1, (t) = kp; where k is some integer if p is odd, or half-integer if p even. t 0 0 0 Proof. By Theorem 8.1 (4), j  (t)j = ji i j for some pair of integers i and i , where there are 2 K two lattice points on a line of slope m = 1 : Thus, we want to constrain the possible di erences between the rst coordinates of such lattice points. p 2qp 2qp For t = , m = . Since gcd(p; q) = 1, in reduced terms, this is either m = q p p q(p=2) or m = if p is odd or even, respectively. Two lattice points on such a line have rst (p=2) coordinates di ering by a multiple of p or of , if p is odd or even, respectively. The completes the proof. 9. The three-genus, g (K ). 0 0 Theorem 9.1. For nonsingular points of  (t), j (t)j  g (K ). K K Proof. According to [8], if K is of genus g, then all elements of CF(K ) have ltration level (i; j), where g  i j  g: It follows immediately from the second statement of Theorem 8.1 that j (t)j  g (K ). We also observe that the genus of K constrains the possible points of singularity of  (t). 0 p Theorem 9.2. Suppose that  (t) has a singularity at t = , with gcd(p; q) = 1. Then: K q If p is odd, q  g (K ). If p is even, q  2g (K ). Proof. Suppose that a line of slope m = , where 0 < b < a contains two distinct points of the form (i; j) with ji jj  g (K ). It follows quickly that the genus bound implies a  2g (K ) b: To express this in terms of t, suppose t = with gcd(p; q) = 1. Then 2 2q p m = 1 = : t p If p is odd, then gcd(2q p; p) = 1. If p is even, say p = 2k, then gcd(2q p; p) = gcd(2q; p) = 2 qk and m = , with q and k relatively prime. In the rst case, with p odd, we have 2q p  2g (K ) p, so q  g (K ). 3 3 In the second case, with p even, we have q k  2g (K ) k, so q  2g (K ). 3 3 8 CHARLES LIVINGSTON 10.  (t) as a knot concordance invariant If knots K and K are concordant, then there is an equality among d{invariants: d(S (K ); s ) = 1 2 1 m 3 N1 N1 3 d(S (K ); s ) for all N 2 Z and m 2 Z,  m  . Here S (K ) denotes N surgery on 2 m N 2 2 N K , d is the Heegaard Floer correction term, and s is a Spin structure, with m given by a speci c enumeration of Spin structures; all are described in [5]. (In the case that N is odd, this range of m includes all possible Spin structures.) If N is large, then d(S (K ); s ) = D(K ) + S(N ), where D(K ) is the largest grading of a class 1 0 z in the homology of CF(K ) for which U z is nontrivial for all k > 0, and S(N ) is some fi0;j0g rational function de ned on the integers, independent of K . In the case that K is slice, we see that the maximal grading D(K ) = D(u), where u is the unknot. This implies that for a slice knot K , D(K ) = 0. We have a nesting of complexes CF(K )  (CF(K );F ) : fi0;j0g t 0 Since (0; 0) is at F ltration level 0, it follows that (CF(K );F )  0; thus  (t)  0. t t K However, K is also slice, so  (t)  0. It follows that  (t) = 0. An additive invariant of K K knots that vanishes on slice knots is a concordance invariant. 11. The concordance-genus The concordance-genus g (K ) of a knot K , de ned in [4], is the minimal genus among all knots concordant to K . Since  (t) is a concordance invariant, the genus bounds in Section 9 apply to the concordance genus. 0 0 Theorem 11.1. For all nonsingular points of  (t), j (t)j  g (K ). The jumps in  (t) occur K c K K p q at rational numbers . For p odd, q  g (K ). If p is even,  g (K ). c c q 2 12. Bounds on the four-genus, g (K ). Let CF(K ) denote the bi ltered subcomplex CF(K ) . We let  (K ) denote the 0;m fi0;jmg minimum value of m such that the homology of CF(K ) contains a nontrivial grading 0 element 0;m of the homology of CF(K ), which we recall is isomorphic to  with 1 at grading 0. There is the following result of Hom and Wu [1], built from work of Rasmussen [10]. (In [1] the invariant  is described; the equivalence with  is presented in [9].) Proposition 12.1 (Proposition 2.4, [1]).   g (K ). Based on this, we show that  (t) provides a bound on g (K ). K 4 Theorem 12.2. For all t 2 [0; 2], j (t)j  tg (K ). K 4 Proof. Since (0; m) is at F ltration level tm=2, we have the containment CF(K )  (CF(K );F ) : 0;m t tm=2 Since CF(K ) contains an element of grading 0 in the homology of CF(K ), so does the subcomplex 0; (CF(K );F ) . Thus, (CF(K );F )  t =2. By the previous proposition, (CF(K );F ) t t =2 t t tg (K )=2. ConsideringK , we have (CF(K );F )  tg (K )=2; it follows that(CF(K );F )  tg (K )=2. t 4 t 4 Combining these yields j(CF(K );F )j  tg (K )=2: t 4 Multiplying by 2 yields the desired conclusion.  UPSILON NOTES 9 13. Crossing change bounds Here we sketch a proof of Proposition 1.10 of [9]. The argument is essentially the same as used in [3] to prove the corresponding fact about  (K ). Theorem 13.1. Let K and K be knots with identical diagrams, except at one crossing which is either negative or positive, respectively. Then for t 2 [0; 1], (t)   (t)   (t) + t: K K K + + Proof. First note that K #K can be changed into the slice knot K #K by changing a + + + negative crossing to positive. Thus, g (K #K )  1. It follows that 4 + (13.1) t   (t)  (t)  t: K K Next, note that K #K # T (2; 3) can be changed into the slice knot K #K by changing + + + one negative crossing to positive and one positive crossing to negative. Thus, it too has four-genus at most 1: it bounds a singular disk with two singularities of opposite sign, and these can be tubed together. A simple computation for T (2; 3) yields  (t) = t for 0  t  1. Thus, T (2;3) t   (t)  (t) t  t; K K which we rewrite as (13.2) 0   (t)  (t)  2t: K K Combining Equations 13.1 and 13.2, 0   (t)  (t)  t: K K Adding  (t) to all terms yields the desired conclusion, (t)   (t)   (t) + t: K K K + + Note This argument can be easily modi ed to show that if there is a singular concordance from K to J with a single positive double point, then  (t)   (t)   (t) + t: K J K 14. The Ozsvath-Szabo  -invariant and  (t) for small t For small t,  (t) is determined by the  invariant de ned in [6]. We review the de nition below. Here is the statement of the result. Theorem 14.1. For t small,  (t) =  (K )t. c c The subquotient complex CF(K ) =CF(K ) will be denoted CF(K ). (Usually, CF is fi0g fi<0g written CFK.) It is ltered by the Alexander ltration and has homology F, supported in grad- ing 0. The invariant  (K ) is de ned to be the least integer  such that the map on homology c c H (CF(K ) ) ! H (CF(K )) F is surjective. 0 fjg 0 We wish to relate  (K ) =  to an invariant of CF(K ). The needed technical result is the following. Lemma 14.2. If  (K ) =  , then there is a cycle w 2 CF(K ) representing a nontrivial fi0;jg[fi<0g element in H (CF(K )). Proof. From the de nition of  we see that there is a chain x 2 CF(K ) that in the fi0;jg[fi<0g c c quotient CF(K ) is a cycle that represents a generator of H (CF(K )). Since the chain x represents a cycle in CF(K ), it has the property that @x = y, where y 2 CF(K ) . Note that y is a cycle and gr(y) = 1. Since H (CF(K ) ) = 0, there is a chain z 2 i<0 1 i<0 CF(K ) with @z = y. Thus, x +z is a cycle in CF(K ) . The map H (CF(K ) ) ! i<0 fi0;jg[fi<0g 0 i0 H (CF(K )) is an isomorphism; both groups are isomorphic to F. Thus, x +z represents a generator of H (CF(K ) ). The map H (CF(K ) ) ! H (CF(K )) is an isomorphism, completing the 0 i0 0 i0 0 proof.  10 CHARLES LIVINGSTON Proof, Theorem 14.1. For t small, we consider the ltration F and the ltration level s =  . Then one has CF(K ) = CF(K ) . By Lemma 14.2, this subcomplex contains a cycle that s fi0;jg[fi<0g represents an element of grading 0 in H (CF(K )). Thus, for this F ltration,    . On the other hand, suppose that  <  . Then there would exist a cycle z 2 CF(K ) fi0;j1g[fi<0g representing a generator of H (CF(K )) of grading 0. However, the image of z in CF(K ) would be an c c element in CF(K ) that represents a generator of H (CF(K )). But  is by de nition the lowest 1 0 level at which this can occur. Thus, we see that  =  . To conclude, recall that  (t) = 2, so  (t) =  (K )t, as desired. K K Note. With care, one can check that in this argument, the condition that t be small can be made precise by requiring that t < 1=g (K ). Of course, once the result is established for some set of small t, then Theorem 9.2 provides the bound t < 1=g (K ). 15. Equivalence of definitions of  (t) In this section we explain why  (t) as de ned here agrees with that of [9]. Beginning with CF(K ), a new complex tCF(K ) can be constructed as follows. As an F{vector space, 1=n 1=n tCF(K ) = CF(K ) F[v ; v ]; 1=n 2 1=n 1=n where U acts on F[v ] via multiplication by v . This has the structure of an F[v ; v ]{module. 0 1=n 1=n To simplify notation, we write  = F[v ; v ]. There are (rational) ltrations Alg and Alex on tCF(K ) which are consistent with those on the 1=n 2 {submodule CF(K ). The action of v lowers ltration levels by 1=2n. Thus, U = v lowers ltration levels by 1, as it should. Similarly, the Maslov grading M (x) naturally extends to tCF(K ) 1=n 2 so that the action of v lowers this grading by 1=n, and thus U = v continues to lower the Maslov grading by 2. There is a rational grading on tCF(K ) de ned via the Maslov grading, M , along with the algebraic and Alexander ltrations. If x is an element at ltration level (i; j), then: (15.1) gr (x) = M (x) t(j i): (In [9], only generators at algebraic ltration level 0 are used to de ne gr , so i = 0 and the formula gr (x) = M (x) tAlex(x) is presented.) One checks that U to lowers gr {gradings by 2, so on the t t 1=n extension to tCF(K ), v lowers gradings by 1 and v lowers gradings by 1=n. If x is a ltered generator of CF(K ) with @x = y , then the boundary @ is de ned so that l t @ x = v y 2 tCF(K ), with the values of given explicitly in [9]. This extends naturally to a t l l boundary operator on all of tCF(K ). Given that the operator @ is well-de ned, it is a simple matter to determine its value. Suppose that x is a ltered generator of CF(K ) at ltration level (i; j), Maslov grading g, and suppose also 0 0 that @x = y . Let y denote one of the terms in this sum, at ltration level (i ; j ), necessarily of grading g 1. Then viewed as an element of tCF, x is of grading g t(j i), and y has grading 0 0 g 1t(j i ). In @ x, the term v y appears, and is such that gr. (v y) = gr (x) 1. Rewriting t t 0 0 this, we have (g 1) t(j i ) = g t(j i) 1. That is, 0 0 (15.2) = t((j j ) (i i )): As two examples, Figure 2 illustrates the complexes tCF(K ) for K = T (3; 7), with t = and t = 2. The construction is straightforward using Equation 15.1 and the fact that v shifts along the diagonal a distance of 1=2 down and to the left. The portion of the complex illustrated was chosen because its homology is F in grading 0 and represents the generator of the homology of tCF , the full complex consists of the illustrated complex along with in grading 0. In the case that t = 3 UPSILON NOTES 11 all its translates a distance , k 2 Z, along the diagonal. In the case of t = 2, the translates are those a distance along the diagonal. Figure 2. tCF (T (3; 7)) and tCF (T (3; 7)) t=1=3 t=2 It is apparent from these examples that the Alexander ltration is not a ltration of the chain complex, since some arrows increase the Alexander ltration level. However, as is easily veri ed, the algebraic ltration is a ltration on the chain complex. De nition 15.1. For t = , denote by tCFK (K ) the complex tCF(K ) . i0 Note. In [9], this complex is denoted tCFK(K ). In fact, it is the complex that is explicitly con- structed. Here we rst introduced the in nity complex to be consistent with our earlier constructions. De nition 15.2. For t = ,  (t) is the maximal grading of a class in the homology of tCFK (K ) that maps to a nontrivial element in the homology of tCF(K ). Equivalently, it is the maximal grading of a class in the homology of tCFK (K ) which is not in the kernel of v for all k > 0. Lemma 15.3. The value of  (t) as just de ned is equal to 2s, where s is the least number for which the homology of tCF(K ) contains an element of grading 0 that represents a nontrivial is element of the homology of tCF(K ). Proof. This follows from a simple change of coordinates. 15.1. The two de nitions of  (t) agree. Suppose that using this de nition of  (t), we have K K (t) = 2s. This implies that tCF(K ) contains a cycle z representing a nontrivial generator of K is grading 0 in the homology of tCF(K ). Write z = x , where the x are ltered generators. Some l l x has ltration level (s; j), and none of the x has algebraic ltration level greater than s. l l From the regrading formula given in Equation 15.1, gr (x) = M (x)t(ji), we see that generators of CF(K ) at ltration level (i; j) and grading 0 yield generators of grading 0 in tCF(K ) at ltration t t level (i + (j i); j + (j i)). (Recall that shifting down and to the left by t units decreases the 2 2 grading by 2t.) We are thus led to consider the transformation t t t t (i; j) 7! ((1 )i + j; i + (1 + )j): 2 2 2 2 12 CHARLES LIVINGSTON Its inverse is given by t t t t (i; j) 7! ((1 + )i j; i + (1 )j): 2 2 2 2 Under this transformation, for a xed value of s, the vertical line f(s; z) j z 2 Rg, is carried to the t t t t line (in the CF(K ){plane) f((1 + )s z; s + (1 )z) j z 2 Rg: Relabeling the coordinate system 2 2 2 2 (x; y), this is the line 2 2 y = (1 )x + ( )s: t t Comparing with Equation 5.1, we see that the homology of the ltered complex (CF(K );F ) t s contains a generator of grading 0 that is nontrivial in the homology of CF(K ), and that this is not the case for (CF(K );F ) for any s < s. Thus, the value of  (t) as de ned in Section 5 is 2s, t s K and the de nitions agree. Appendix A. A structure theorem for CF(K ). In [2, Chapter 11], vertical and horizontal reductions of CF(K ) are discussed. That presentation applies to the ltered complex (CF(K );F ), but adjustments in the details would be required be- cause, for instance, the horizontal and vertical ltrations are integer valued rather than being real ltrations. Since the argument in the present case is straightforward, we present it in detail. Viewed as a {module, CF(K ) is freely generated by a nite set fw g . We again simplify i 1im notation by suppressing the indexing set and write fw g. This set can be chosen so that the set fU w g forms a bi ltered graded basis for the F{complex CF(K ). We will refer to any such set i k2Z fw g as a {basis for CF(K ). A {module change of basis among the w that preserves gradings i i and ltration levels induces a change of bi ltered graded basis for the F{complex CF(K ). We will refer to any such change of basis as a {change of basis of CF(K ). Analogous notation will be used when working with the ltered graded complex (CF(K );F ). Theorem A.1. Let t 2 [0; 2]. As a {module, CF(K ) has a basis f ; ; : : : ; g, inducing a 1 k splitting of CF(K ) (as a {module) as the direct sum CF(K ) T A, where T is freely generated by and A is freely generated by f ; : : : ; g. This splitting has the following properties. 1 k (CF(K );F ) T A as a ltered graded F-complex. The complex T has ltered graded basis fU g , the boundary map is trivial on T , and k2Z gr( ) = 0. The complex A has ltered graded basis fU g and has trivial homology: H (A) = 0. i k2Z Proof. We begin with the {generating set of CF(K ), fw g. By replacing generators with their U translates and renaming the generators, we can decompose this into two subsets: fx g, all of grading 0, and fy g, all of grading 1. To simplify notation, we abbreviate the ltered graded F{complex (CF(K );F ) by CF . t t (1) Let A be a cycle in CF having the least ltration level among cycles representing nontrivial classes in H (CF ): After reordering the generators, we can write A = x + + x , with 0 t 1 k the ltration levels nonincreasing. Replacing x with x +  + x as the rst generating 1 1 k element (over ) induces a ltered change of basis for CF . Thus, the rst element of the {basis, which we now denote A , is a cycle of least ltration level representing a nontrivial element of H (CF ). 0 t (2) Consider the set of all generating elements y that have the property that A is a component i 1 of @y . After reordering the basis, we can assume these are fy ; y ; : : : ; y g for some k, and i 1 2 k that the ltrations are in nondecreasing order. Make the {change of basis that replaces each y , 2  i  k, with y + y . This induces a ltered change of basis of CF . Now, the i i 1 t only generator having A as a component of its boundary is y , which we relabel B . 1 1 1 (3) After perhaps reordering the x , we have either @B = A or @B = A + x + + x for i 1 1 1 1 2 k some k  2, with the ltration levels nonincreasing. Since @ = 0, it follows that B is not a component of any element in the image of @. UPSILON NOTES 13 If @B = A , then we see that fA ; B g generates an acyclic summand of CF , and thus 1 1 1 1 t A would not represent a nontrivial element in homology. We have @B = A +x + x for some k  2. Make the {change of basis that replaces 1 1 2 k x with x +  x , now calling this new element A . Then @B = A + A . Note that 2 2 k 2 1 1 2 since A is a cycle and A + A = @B is a cycle, that A is a cycle representing the same 1 1 2 1 2 homology class as A . Hence the ltration level of A is greater than or equal to that of A . 1 2 1 (4) We now repeat the previous argument, making a change of basis so that the only basis elements with boundary that include A as a component are B and perhaps a second 2 1 generator that we denote B . (5) This step-by-step procedure must eventually stop, at which time there is constructed a summand of the F{complex CF D = A B ! A B ! A  ! B ! A : 1 1 2 2 3 k1 k Note that the process must end with an A ; if it stopped with a B , the resulting complex k k would be acyclic and thus not contain a nontrivial element in homology. This complex is a summand of the complex CF . Note that D is a summand of a direct sum decomposition of CF , as a subcomplex and also as a submodule of the {module. (6) Since A has the lowest ltration level among the A , we can replace each A with A + A 1 i i 1 i to form a new basis. The complex then splits in the following way: A  [B ! (A + A ) B ! (A + A )  ! B ! (A + A )] : 1 1 1 2 2 1 3 k1 1 k We let T = A . It satis es the required conditions of the theorem. Since as a {module, H (T ) H (CF ), the complementary summand to T must be acyclic. That complementary summand yields the summand A in the statement of the theorem. References [1] J. Hom and Z. Wu, Four-ball genus bounds and a re nement of the Ozsv ath-Szab o tau-invariant, arxiv.org/ abs/1401.1565. [2] R. Lipshitz, P. Ozsv ath, and D. Thurston, Bordered Heegaard Floer homology, arxiv.org/abs/0810.0687. [3] C. Livingston, Computations of the Ozsv ath-Szab o knot concordance invariant, Geom. Topol. 8 (2004), 735{ [4] C. Livingston, The concordance genus of knots, Alg. and Geom. Top. 4 (2004), 1{22. [5] P. Ozsv ath and Z. Szab o, Absolutely graded Floer homologies and intersection forms for four-manifolds with boundary, Adv. Math. 173 (2003), 179{261. [6] P. Ozsv ath and Z. Szab o, Knot Floer homology and the four-ball genus, Geom. Topol. 7 (2003), 615{639. [7] P. Ozsv ath and Z. Szab o, Holomorphic disks and knot invariants, Adv. Math. 186 (2004), 58{116. [8] P. Ozsv ath and Z. Szab o, Holomorphic disks and genus bounds, Geom. Topol. 8 (2004), 311{334. [9] P. Ozsv ath, A. Stipsicz, and Z. Szab o, Concordance homomorphisms from knot Floer homology, arxiv.org/ abs/1407.1795. [10] J. Rasmussen, Floer homology and knot complements, arxiv.org/abs/math/0306378. Charles Livingston: Department of Mathematics, Indiana University, Bloomington, IN 47405 Email address : livingst@indiana.edu

Journal

Mathematics – arXiv (Cornell University)

Published: Nov 30, 2014

There are no references for this article.