Diagrams of *-Trisections

We use the notation described at the begining of the subsection. Remark 3.2 allows us to only look at the connected components of \widetilde C_0 \cup_{\widetilde F_{all}} \widetilde C_1, which are in correspondance with |\widetilde F_{all}|. We then suppose that \widetilde F_{all} is a connected non-empty surface.

The fact that \widetilde C_{all} is the common sub-compression body between \widetilde C^0 and \widetilde C^1 implies that there are no essential simple closed curves in \widetilde F_{all} bounding disks in both \widetilde C_0 and \widetilde C_1.

In particular, if both collections of curves \widetilde \delta_i are non-empty, then the result of compressing \widetilde F_{all} along \widetilde \delta_0 \cup \widetilde \delta_1 contains no sphere components, and the compression body \widetilde C-int(\widetilde C_{all}), determined by \widetilde\delta_0 \cup \widetilde \delta_1 in \widetilde F_{all}, has non-empty negative boundary. In other words, we do not need 3-dimensional 3-handles to build \widetilde C - int(C_{all}) using the loops \widetilde \delta_0 \cup \widetilde \delta_1.

Suppose that both \widetilde \delta_0 and \widetilde \delta_1 are non-empty. The observation above tells us that no 3-handles are needed to build each \widetilde C_i. Thus the 3-manifold \widetilde C_0\cup \widetilde C_1 is built from \widetilde F_{all}\times [0,1] by attaching 2-handles along \widetilde\delta_i\times\{i\}\subset \widetilde F_{all}\times\{i\} for i=0,1. This implies that \left(\widetilde C_0\cup \widetilde C_1\right)\times [0,1] can be built from \widetilde F_{all}\times D^2 \approx \widetilde F_{all}\times [0,1]^2 by attaching 4-dimensional 2-handles along \widetilde \delta_i \times \{(-1)^i\} \subset \widetilde F_{all}\times \{(-1)^i\} for i=0,1 with framing given by the fiber surface. Since \widetilde\delta_0 and \widetilde \delta_1 are disjoint in \widetilde F_{all}, we can isotope the attaching regions of the 2-handles in \widetilde F_{all}\times S^1 to lie in the same fiber \widetilde F_{all}\times \{1\}. Using the previous paragraph we get that \left(\widetilde C_0\cup \widetilde C_1\right)\times [0,1] is diffeomorphic to \left(\widetilde C-int(\widetilde C_{all})\right)\times [0,1]. Notice that the same conclusion holds if one of \widetilde \delta_i is empty.

To end, recall that \widehat C=\widetilde C-int(\widetilde C_{all}) is a compression body satisfying \partial_+ \widehat C=\widetilde C_{all} and \partial_- \widehat C=\partial_-\widetilde C. If \partial_- \widehat C is empty, then \widehat C is a handlebody and \widetilde Z\approx \widehat C\times [0,1] is a 1-handlebody. On the other hand, if \partial_- \widehat C\neq \emptyset then \widehat C is built from \left(\partial_- \widehat C\right)\times [0,1] by attaching 1-handles along \left(\partial_- \widehat C\right)\times \{1\} and \widetilde Z has the right diffeomorphism class.

The argument in the third paragraph of the proof of Lemma 8 of [bridge_trisections_4M] proves the following statement: “Suppose A_1 and A_2 are 4-manifolds with connected boundary satisfying (1) A_i has unique trivial disks (up to isotopy relative their boundary), and (2) any embedded sphere in \partial A_i bounds a properly embedded 3-ball in A_i. Then A_1 \natural A_2 has unique trivial disks.” Also in Lemma 8 of [bridge_trisections_4M] is proven that 1-handlebodies satisfy properties (1) and (2). Hence, using the decomposition of Lemma 3.4, it is enough to check uniqueness of trivial disks for F \times D^2 where F is a connected closed surface of positive genus.

Let F\neq S^2 be a closed surface and let L be an unlink in F\times S^1. Suppose D and D' are two collections of properly embedded trivial disks in F\times D^2 with boundary equal to L. We want to show that they are isotopic relative to L. There exist a collection of disks D_* (resp. D'_*) in F\times S^1 isotopic to D (resp. D') via an isotopy fixing L. Perturb D_* and D'_* to intersect transversely (away from L) in a finite set of loops. Suppose int(D_*)\cap int(D'_*)\neq \emptyset. Let E\subset D_* be a disk with interior disjoint from D'_* and boundary a loop in int(D_*)\cap int(D'_*). Let E' be the disk in D'_* with same boundary as E. Since F\times S^1 is irreducible, the embedded 2-sphere E\cup E' bounds a 3-ball B\subset F\times S^1. Push the interior of B into int(F\times D^2) to make it disjoint from D_* and D'_*, we can do this since D_*\cup D'_* lies in F\times S^1. Isotope E' through B to remove the intersection loop E\cap E' \subset int(D_*)\cap int(D'_*). We can perform this type of isotopy until int(D_*)\cap int(D'_*)=\emptyset. Again, the sphere D_*\cup D'_* bounds a collection of 3-balls in F\times S^1 which we can use (as before) to isotope D_* onto D'_* as desired.

Suppose first that each component of U has bridge number one. Since U is an unlink, we can find a set of disks D_*\subset Y bounded by U intersecting F_{k,b} in a single arc per disk in D_*. The spheres S_* in Y can be chosen to be \partial \eta (D_*). In this case, the curves determining the compression bodies are given by \delta^i=\widetilde \delta \cup \left( S_* \cap F_{k,b}\right). Using such curves, the right hand side of Equation 3.7 is diffeomorphic to \widetilde Z \natural \left(\natural_{|U|}S^1\times B^3\right)\approx Z, which is what we want. One way to see this is to first note that if c\subset \delta_{all} is a separating curve decomposing F_{k,b} as F_{k_1,b_1}\#_c F_{k_2,b_2}, then the 4-manifold from the right hand side decomposes as a boundary connected sum of 4-manifolds built in the same way from the surfaces F_{k_j,b_j} (j=1,2); and then check that the 4-manifold obtained from an annulus F_{0,2} and \delta^0=\delta^1=\delta_{all}=\emptyset is S^1\times B^3.

We now proceed with the general case. First recall that any bridge position of an unlink of unknots in a 3-manifold is always perturbed. This result was proven for S^3 in [Otal], and for any 3-manifold in [hayashi_unknot]. A perturbation is a local isotopy of a link in bridge position around the Heegaard surface shown in Figure ?. It increases the bridge number by one. Figure ? describes how the complement of the link changes after a perturbation, and shows us how the \delta^i sets increase by one loop. The main observation is that the 4-manifolds built from the right hand side of Equation 3.7 are diffeomorphic because the operation of removing two disks to F_{k,b} and adding two curves (not in \delta_{all}) corresponds to add two 4-dimensional 1-2-cancelling pairs of handles (See Figure ?).

Hence, for any bridge position of U in \widetilde Y, after fixing a set of spheres S_*, we can slide the curves of \delta^i-\delta_{all} over the curves in \delta^i until ‘detecting’ a pair of curves intersecting like in Figure ?. This pattern determines a perturbation of U which we undo to decrease the bridge number, without changing the resulting 4-manifold. We can repeat this procedure until each component of U has bridge number one, in which case the result is known. Therefore, for any bridge position of U, Equation 3.7 holds.

Let X_i = W_i \cup_f W_i'. We will show that X=X_1\cup X_2\cup X_3 is a \star-trisection.

We will first focus on the pairwise intersection X_i\cap X_j. Fix i\neq j and let C=W_i\cap W_j and C'=W_i'\cap W_j'. By definition C and C' are compression bodies with positive boundary the connected \star-trisection surfaces \Sigma and \Sigma', respectively. Let P_M=\partial_-C\cap M and P_{M'}=f(P_M)=\partial_-C'\cap M' be the negative boundary components of C and C' lying inside the gluing regions M and M'. By assumption each component of P_M and P_{M'} is a surface with boundary. This implies that H=C\cup_{f} C' is also a compression body with positive boundary \widetilde \Sigma=\Sigma\cup_f\Sigma'. In particular, the triple intersection X_1\cap X_2\cap X_3 is a copy of \widetilde \Sigma.

Let \varepsilon \subset \Sigma and \varepsilon'\subset \Sigma' be loops determining C and C'. Let a\subset P_M be pairwise disjoint properly embedded arcs which cut P_M into disks and let a'=f(a) be the corresponding arcs in P_{M'}. We can think of the arcs in a (resp. a') as subsets of \Sigma (resp. \Sigma'), disjoint from \varepsilon (resp. \varepsilon'). The curves in \widetilde \Sigma=\Sigma\cup_f \Sigma' determining H are given by \varepsilon\cup \varepsilon' and the glued arcs a\cup_f a'.

We now show that each piece X_i is diffeomorphic to a standard piece. Fix i\in \{1,2,3\}, by definition W_i\approx Z(C^0,C^1, C_{all}) and W'_i\approx Z(C'^0,C'^1,C'_{all}) for some compression bodies. We use the notation of Remark 3.9; i.e., C^0 and C^1 are determined by essential loops \Delta^0,\Delta^1\subset\Sigma, and the common curves \Delta_{all}=\Delta^0\cap \Delta^1 determine the C_{all}. Let F_{all} be the negative boundary of C_{all}. For j=0,1 let C_j=C^j-int(C_{all}) be the compression body with positive boundary the surface F_{all}, and let P_j be the negative boundary of C_j. Recall that, due to Remark 3.10, W_i\cap \partial W contains a copy of C_0\cup_{F_{all}} C_1.

If F_{all} is disconnected, we can pick \Delta_{all}\subset \Sigma having separating simple closed curves. Each such curve decomposes the diagram (\Sigma;\Delta^1,\Delta^0, \Delta_{all}) as a connected sum of standard diagrams in lower genus surfaces. This decomposes W_i into a boundary connected sum of the corresponding standard 4-manifolds. The same argument holds for W'_i. Hence, it is enough to check that W_i\cup_f W'_i is a standard piece when F_{all} and F'_{all} are connected surfaces.

Suppose F_{all} is a connected surface. In particular, the curves in \Delta_{all}\subset \Sigma do not separate the surface and C_{all} is built from F_{all}\times[0,1] by attaching |\Delta_{all}| 1-handles(No 3-handles are needed in this case since we are assuming F_{all} is non empty.) along F_{all}\times\{1\}. Then, using Equation ? we conclude that W_i is diffeomorphic to the result of adding |\Delta_{all}| 1-handles to (C_0\cup_{F_{all}} C_1)\times [0,1]. Since F_{all} is connected, Remark 3.2 explains that the location of the feet of the 1-handles is not necessary. Furthermore, since (\Sigma; \Delta^0, \Delta^1, \Delta_{all}) is a standard pair and F_{all} is connected, it follows that \Delta_0\cup\Delta_1 only has curves of one type (see Remark 3.9). In any case, we get that (C_0\cup_{F_{all}}C_1)\times [0,1]) is a 1-handlebody of genus N. Here N=\left( 1-\chi(F_{all})-|\Delta_0|-|\Delta_1|-2|\delta^0_{stab}|\right) if \Delta_0\cup\Delta_1 is of type 1 or N=1 if type 2. One can see this by building W_i like in Equation 3.8 and noticing that each circle in \Delta_0\cup \Delta_1 cancels a 1-handles from the 1-handlebody F_{all}\times[0,1]^2 (recall F_{all} has boundary). On the other hand, to build X_i we must glue (C_0\cup_{F_{all}}C_1)\times[0,1] and (C'_0\cup_{F'_{all}}C'_1)\times[0,1] using f\times\{1\}, and then add copies of C_{all}\times[0,1] and C'_{all}\times[0,1] as in Equation ? (see right side of Figure ?). Therefore, X_i=W_i\cup_f W'_i is a 1-handlebody of genus K=|\Delta_{all}|+|\Delta'_{all}|+N.

To end, observe that the boundary of X_i=W_i\cup_f W'_i has a decomposition of the form H^0\cup_{\widetilde \Sigma} H^1 where H^j=(C^j\cup_f C'^j). In particular, both H^0 and H^1 are handlebodies and this is a Heegaard splitting for \partial X_i = \#_K S^1\times S^2. Waldhaussen’s Theorem [waldhausen] implies that this splitting is a g(\widetilde\Sigma)-K times stabilization of the standard genus K Heegaard splitting. We can take this standard splitting to be H_{all} and conclude that X_i is diffeomorphic to the standard piece Z(H^1,H^0, H_{all}). This finishes the proof.

Throughout this argument \beta and \gamma will be pushed around as needed. If \gamma or \beta are in the way of sliding a_i over b_i for i=1,2, simply include them in the slide as in the left of Figure ?. Since c is embedded and a_1\cap \alpha=\emptyset, the n intersections of c with \alpha occur on a_2\cup a_3. Isotope c such that a_2\cap\alpha=\emptyset. Since a_1 and a_2 both miss \alpha, we may isotope them such that a_1 is a small segment leaving b_2, a_2 is a small segment leaving b_3 with a_1\cup a_2 being contained in a small disk D disjoint from \lambda\cup\mu. This way, all of the intersections of c with \lambda\cup\mu occur on a_3. Therefore, since c is embedded in \Sigma, a_3-D is a properly embedded arc in \Sigma-D. Thus we may assume from the beginning that the trisection as well as the embedded decomposed curve are isotopic to the model in the right of Figure ? where we have suppressed the \beta and \gamma curves. At this point in the argument, the \beta and \gamma curves cannot be assumed to lie in any kind of special position with respect to c.

Let w be the word in the alphabet \lbrace \mu,\lambda \rbrace which represents [c] and decompose w into three subwords w_1,w_2,w_3 such that w=w_1w_2w_3 and each w_i records the intersections of a_i with \mu and \lambda. Initially, as described above, we have that w_1 and w_2 are empty words and w_3 is a certain permutation of the multiset \lbrace m\mu,n\lambda\rbrace which allows a_3 to be an embedded arc. We claim that a decomposed c representing [m\mu+n\lambda], m>0, in such a way that w_1 and w_2 are empty words, can be slid to be a representative of [(m-1)\mu+n\lambda] with w_1 and w_2 being empty words.

Suppose that c represents [m\mu+n\lambda], m>0 with w_1 and w_2 empty. Then since m>0, let j\geq 0 such that \lambda^j\mu is a prefix of w_3. The endpoint of a_2 connected to a_3 is b_1. We can make b_1 “move past" \alpha and a_1 by performing the local move in Figure ?.

We can therefore move b_1 along a_3 until w_2 reads \lambda^j\mu, Figure ?. Using the fact that this is a genus one trisection, and the fact that \lambda^j\mu is embedded, by sliding a_2 over b_2 when necessary as in Figure ? we can commute \mu past \lambda^j. After an isotopy of a_1 and a_2, reading from left to right in the two leftmost frames in ?, we achieve w_1=\mu and w_2=\lambda^j. From this position, we can slide a_1 against \alpha to yield w_1=\emptyset at the expense of adding a single twist of a_1 around b_3. This introduces a bigon between a_1 and \lambda which is easily resolved. This slide and subsequent isotopy are shown in the last 2 frames.

Now it is possible to commute \mu past \lambda^j by sliding a_2 over b_2 when necessary.

After completing this slide to eliminate \mu, we can shrink a_2, removing \lambda^j from w_2 and appending \lambda^j to the front of w_3. The decomposed curve at this stage is a representative of [(m-1)\mu+n\lambda] with w_1=w_2 empty, so the claim is proved. By repeating this process, we can slide c to the model representative of [n\lambda] below with w_1=w_2 being empty and w_3=\lambda^l.

Decompose D(\frac{a}{b},\frac{c}{d},\frac{p}{q}) into two pieces as suggested by Figure Proof; let D' denote the thrice punctured torus component of this decomposition and let P denote the thrice punctured sphere component. Notice that this \star-trisection D' is a diagram for the complement of an embedded curve in a simply connected 4-manifold. Specifically, D' is the result of taking the complement of the c=\lambda curve in a genus 1 trisection with curves \alpha=a\lambda+b\mu, \beta=c\lambda+d\mu, and \gamma=p\lambda+q\mu where the curve c has been decomposed as suggested by the left side of Figure Proof. By Corollary 6.4, the decomposed curve c is slide equivalent to an immersed decomposed curve c' representing the trivial curve where some of the arcs twist around the boundary points, as in the last frame of Figure ?. In particular, \alpha and \beta are disjoint from c'. Using the fact that c' represents the trivial loop, we can slide \gamma against b_3 until \gamma is disjoint from c' also. Thus there is a curve \delta separating c' from \alpha,\beta,\gamma. By surgering D' along \delta we get two components. Let Q be the component coming from the side of \delta containing c' and let T be the torus component containing \alpha,\beta,\gamma. Notice that S=Q\cup_f P, with the attaching map f given by the a_i, is a genus two trisection of a closed 4-manifold with an intersection form of full rank. By the work of Meier and Zupan [genus_two_std], S is a trisection diagram for S^2\times S^2 or S^2\widetilde\times S^2. The component T is a genus one trisection of a simply connected closed 4-manifold: S^4,\CP2 or \overline{\CP2}.

The condition F\cdot F = F'\cdot F' is so that the gluing map f: \partial \eta F \rightarrow \partial \eta F' exists. To trisect the fiber sum of X and X' one must trisect the complements X-\eta(F) and X'-\eta(F') using the bridge trisections. Since b=b' and c_i=c'_i for i=1,2,3, the handle decompositions in \partial \eta (F) and \partial \eta (F') induced by the \star-trisections agree. Thus we can apply Lemma 5.1 and obtain a trisection for X\natural X'. The \star-trisection surface for X-\eta(F) is connected of genus g and 2b boundary components; similarly for X'-\eta(F'). By Remark 5.2, the trisection surface for X\natural X' is a closed surface of genus G=g+g'+ 2b-1.

To determine the value of each K_i we need to dive into the proof of Lemma 5.1. Fix i\in \{1,2,3\}, the 4-manifold piece W_i-\eta(F) is determined by the tuple (\Sigma; \Delta^1, \Delta^0, \Delta_{all}) as in Remark 3.9; similarly for W'_i-\eta(F'). By definition of bridge trisection, F\cap \partial W_i is a c_i component unlink in bridge position. By construction, \Delta_{all} contains separating curves breaking \Sigma as the connected sum of a genus g closed surface and a c_i spheres, each of them with positive even number of boundary components. Each sphere corresponds to a component of the unlink F\cap \partial W_i. The curves (\Delta^0, \Delta^1) in the genus g surface determine a 1-handlebody with k_i 1-handles. For each planar surface, the curves (\Delta^0,\Delta^1) are only of type 2 which determine 4-dimensional pieces that get glued with other 4-dimensional pieces coming from W'_i-\eta(F'). After gluing each planar surface determines a copy of S^1\times B^3 (see paragraph 4 of the proof of Lemma 5.1). Thus, as explained in paragraph 3 of the proof of Lemma 5.1, the glued 4-manifold piece W_i\cup W'_i is diffeomorphic to \natural_{K_i} S^1\times B^3 where K_i=k_i+k'_i + c_i.

The decomposition of X will be given as follows: Divide the 2-handles of X by h^2=h^2_P\cup h^2_r where h^2_P are the ones used to build P and h^2_r the rest of the 2-handles. Define X_1=B^4[h^1\cup h^2_P]. We can see \Sigma as embedded in \partial X_1 by stabilizing the standard circular decomposition in \partial X_1 as in Subsection ?, say

Define X_2 = \eta_X(C_\beta)[h^2_r] and X_3 = X-int(X_1\cup X_2). The proof that X=X_1\cup X_2\cup X_3 is indeed a relative trisection is the same as in Theorem 1 of [relative_trisections_2].

The proof is the same as in Theorem 5 on [relative_trisections]. The only observation is that the proof of Lemma 13 of [relative_trisections], a key lemma for this result, does not apply when \Sigma_\alpha is closed. This problem can be solved by considering the annulus a\times [-1,1] for any loop in P instead of the disk in the proof of Lemma 13. The proof then works.