Tags: #subjects/geomtop/Floer-theory #projects/reading-groups #projects/my-talks #projects/active #resources/papers #todo/learning/definitions Refs: Floer homology
Sarkar Wang Paper Talk
References
Paper: Sarkar-Wang, An algorithm for computing some Heegaard-Floer homologies
Intro/Overview
-
Goals:
- Given \(Y\) a closed oriented smooth 3-manifold, compute the Heegaard-Floer homology \(\widehat{\operatorname{HF}}(Y; {\mathbb{F}}_2) \in {\mathsf{gr}\,}{\mathsf{Ab}}\).
- Given \(K \hookrightarrow Y\) a knot, compute the knot Floer homology \(\widehat{\operatorname{HFK}}(Y, K; {\mathbb{F}}_2)\)
-
Strategy: turn admissible diagrams into nice diagrams using isotopies and handleslides
- Note: stabilization not used in this paper.
- Main theorem:
Definitions
Use a cylindrical reformulation of \(\widehat{\operatorname{HF}}\) due to Lipshitz.
Given a pointed Heegaard diagram \begin{align*} ( \Sigma, \vec \alpha, \vec \beta, w) ,\end{align*} the generators of \(\operatorname{CF}\) are formal sums of points \(\mathbf{x} = \sum_{i=1}^{g-k-1} x_i\) where each \(\alpha\) curve contains some \(x_i\) and each \(\beta\) curve contains some \(x_j\), and \(k\) is the number of basepoints in \(\vec w\).
A region is a connected component of the complement of the curves, so \begin{align*} R_i \in \pi_0 \Sigma \setminus\left\{{\vec \alpha, \vec \beta}\right\} .\end{align*} A formal sum of regions is a 2-chain.
Given 2 generators \(\mathbf{x}, \mathbf{y}\) define \(\pi_2(\mathbf{x}, \mathbf{y})\) to be the set of all 2-chains \(\phi\) satisfying \({{\partial}}^2 \phi = \mathbf{y} - \mathbf{x}\). Call such 2-chains domains.
Define \(n_p(\phi)\) to be the coefficient of the region \(R_i \ni p\), then \(\phi\) is a positive domain if \(n_p(\geq 0)\) for all \(p \in \Sigma\setminus\left\{{ \vec \alpha, \vec \beta }\right\}\).
Define \(\pi_2^0(\mathbf{x}, \mathbf{y})\) to be all domains \(\phi\) such that \(n_{\vec w} = 0\). A Heegaard diagram \(\mathcal{H}\) is admissible if for every generator \(\mathbf{x} \in \operatorname{CF}\), every positive domain \(\varphi\in \pi_2^0(\mathbf{x}, \mathbf{x})\) is trivial.
A region is good if it is an \(n{{\hbox{-}}\mathrm{gon}}\) with \(n\leq 4\), and bad if \(n\geq 5\).
For a disc region \(D\), define the badness \begin{align*} b(D) \coloneqq\max\left\{{n-2, 0}\right\} .\end{align*}
Note
- \(D\in 2{{\hbox{-}}\mathrm{gon}}\implies b(D) = 0\)
- \(D\in 4{{\hbox{-}}\mathrm{gon}}\implies b(D) = 0\)
- \(D\in 6{{\hbox{-}}\mathrm{gon}}\implies b(D) = 1\)
- \(D\in 8{{\hbox{-}}\mathrm{gon}}\implies b(D) = 2\)
Goal: do moves where
- \(d(\mathcal{H}' ) \leq d( \mathcal{H} )\), so total distance doesn’t increase.
- \(b( \mathcal{H}' ) \leq b( \mathcal{H} ) -1\), so badness decreases.
Algorithm Overview
-
Start from an admissible pointed Heegaard diagram, end up with an admissible nice pointed Heegaard diagram using isotopies and handleslides on the \(\beta\) curves.
-
Overview of strategy:
- Isotope all regions to disks
- Define a complexity for the diagram
- Show it’s minimized iff all regions not containing basepoints are good
- Do an isotopy or handleslide that strictly decreases the complexity.
-
Recipe:
- Kill non-disk regions (easier)
- Make all but one region bigons or squares (harder)
Step 1: Killing Non-disk Regions
Short procedure: ensure every \(\alpha\) curve intersects some \(\beta\) curve and vice-versa
-
Fix \(j\), what is in between \(\alpha_j\) and the nearest \(\beta\) curve?
-
Case 1: \(\alpha, \alpha, \beta\).
- Connect \(p\) to \(q\), do a finger move of \(\beta\) curve to introduce 2 intersections with chosen \(\alpha\)
- Yields 4 new intersections.
-
Case 2: \(\alpha, \beta, \beta\).
- Finger move on \(\beta\) curve, pulling all other \(\beta\) curves with it.
Now every \(\alpha_j\) intersects some \(\beta_k\) All non-disk regions have \(\geq 2\) boundary components, so we’ll try to reduce the number of boundary components to one. Problem: poly-annuli regions
How to fix:
- Creates no new non-disks
- Decreases number of boundary components by 1.
- Repeat until equal to 1.
Step 2: Converting Regions to Bigons or Squares
Now all regions are discs.
\(\alpha\) curves should be red, \(\beta\) should be blue!
For a Heegaard diagram \(\mathcal{H}\) define the distance of \(D\) as \begin{align*} d(D) \coloneqq\min \left\{{ {\sharp}(\gamma\cap\vec\beta) {~\mathrel{\Big\vert}~}w'\in D, w \xrightarrow{\gamma} w', \gamma\in \vec\alpha^c }\right\} \in {\mathbb{Z}}_{\geq 0} .\end{align*}
This is the smallest number of intersection points on any arc connecting the basepoint \(w\) to \(w'\in D\). Define the total distance \begin{align*} d( \mathcal{H} ) \coloneqq\max\left\{{ d(D)}\right\} \in {\mathbb{Z}}_{\geq 0} .\end{align*}
Computing the distance via a path.
Can only cross \(\beta\) curves, need to stay in \(\vec \alpha^c\).
For a fixed distance \(d\) define the distance \(d\) complexity as \begin{align*} c_d \coloneqq{\left[ { \sum b(D_i), -b(D_1), -b(D_2), \cdots} \right]} && b(D_1) \geq b(D_2) \geq \cdots \end{align*} For a fixed \(d\), order various \(c_d( \mathcal{H}' )\) lexicographically.
Main theorem
Setup: fix \(D_0\) to be the region containing the basepoint, \(D_m\) to be the least bad region. Find an adjacent region \(D_*\) with 1 smaller distance:
Idea: we will finger move \(b_*\) through \(D_m\) to reduce the badness of \(D_m\):
Look at what happens locally: we introduce some new regions, usually less bad:
We can push through \(4{{\hbox{-}}\mathrm{gon}}\) regions:
So continue, then do some casework:
Case 1: Reach a Bigon
Case 2: Smaller Distance