return

Celebratio Mathematica

Yakov M. Eliashberg

Legendrian surgery

by Tobias Ekholm

1.  Introduction

Start­ing some time around 2004, Yasha Eli­ash­berg ini­ti­ated the handle ap­proach to com­put­ing holo­morph­ic curve in­vari­ants of Wein­stein man­i­folds and their con­tact bound­ar­ies, fo­cus­ing in par­tic­u­lar on re­lat­ing the most ele­ment­ary part of closed string sym­plect­ic field the­ory (SFT) [1], or­bit con­tact ho­mo­logy, to the simplest part of open string SFT, the Le­gendri­an or Chekan­ov–Eli­ash­berg dg-al­gebra, of the Le­gendri­an at­tach­ing spheres for crit­ic­al handles. Al­though per­haps not the ori­gin­al in­ten­tion, this led to con­nec­tions and iso­morph­isms between all the vari­ous fla­vors of “tree-level” holo­morph­ic curve the­or­ies for Wein­stein man­i­folds. Cor­res­pond­ing res­ults for “high­er loops” con­sti­tute very in­ter­est­ing prob­lems where some ini­tial pro­gress has been made but much re­mains to be un­der­stood.

In this pa­per we will ex­plain the steps in the proof of the most im­port­ant and ba­sic Le­gendri­an sur­gery iso­morph­ism that ex­presses the wrapped Flo­er co­homo­logy of the cocore disks of a Wein­stein man­i­fold in terms of the Chekan­ov–Eli­ash­berg dg-al­gebra of at­tach­ing spheres. We will then dis­cuss ex­ten­sions and rami­fic­a­tions of this res­ult to re­lated the­or­ies: sym­plect­ic co­homo­logy with its product struc­ture, Hoch­schild ho­mo­logy and co­homo­logy of wrapped Flo­er co­homo­logy, par­tially wrapped Fukaya cat­egor­ies and Chekan­ov–Eli­ash­berg dg-al­geb­ras with coef­fi­cients in chains on the based loop space coef­fi­cients, as­so­ci­ated dg-al­geb­ras for sin­gu­lar Le­gendri­ans, and cut-and-paste meth­ods.

2.  Weinstein manifolds

Con­sider an ex­act sym­plect­ic \( 2n \)-man­i­fold \( X \) with sym­plect­ic form \( \omega=d\lambda \). A vec­tor field \( Z \) which is \( \omega \)-dual to \( \lambda \), \( \lambda=\iota_{Z}\omega \) is called a Li­ouville vec­tor field. For such a vec­tor field, if \( L \) de­notes the Lie de­riv­at­ive, \( L_{Z}\omega=d(\iota_{Z}\omega) + \iota_{Z}d\omega=\omega \), which means that \( \omega \) ex­pands along the flow of \( Z \) in the pos­it­ive dir­ec­tion and con­tracts in the neg­at­ive.

The man­i­fold \( X \) is a Wein­stein man­i­fold if \( Z \) is com­plete and if it ad­mits a Morse func­tion \( F\colon X\to\mathbb{R} \) for which \( Z \) is gradi­ent-like. It fol­lows in par­tic­u­lar that out­side a com­pact sub­set, \( X \) is sym­plec­to­morph­ic to \( Y\times [0,\infty) \), where \( Y=F^{-1}(T) \) for some suf­fi­ciently large \( T \) and where the sym­plect­ic form on \( Y\times[0,\infty) \) is \( d (e^{t}\alpha) \), \( \alpha=\lambda|_{Y} \). Here \( (Y,\alpha) \) is a con­tact man­i­fold which is the ideal con­tact bound­ary of \( X \). Com­pact ver­sions, \( \overline X=X\setminus F^{-1}(T,\infty) \) of \( X \) are called Wein­stein do­mains, one can move between Wein­stein do­mains and Wein­stein man­i­folds by adding or re­mov­ing cyl­indric­al ends.

The most ba­sic ex­ample of a Wein­stein man­i­fold is \( \mathbb{C}^{n}\approx\mathbb{R}^{2n} \), the as­so­ci­ated Wein­stein do­main is the \( 2n \)-ball \( B^{2n} \), with the stand­ard sym­plect­ic form \( \sum_{j=1}^{n} dx_{j}\wedge dy_{j}= d (xdy) \), Li­ouville vec­tor field \( \sum_{j} \frac12\left(x_{j}\partial_{x_{j}} + y_{j}\partial_{y_{j}}\right) \), and with the ideal con­tact bound­ary be­ing the stand­ard con­tact sphere \( S^{2n-1} \) with con­tact struc­ture giv­en by com­plex tan­gen­cies.

Oth­er ex­amples are co­tan­gent bundles \( T^{\ast} M \) of com­pact man­i­folds \( M \) with Li­ouville form \( pdq \). Equip \( M \) with a Rieman­ni­an met­ric, take the Li­ouville vec­tor field as \( \sum_{j} \frac12 p_{j}\partial_{p_{j}}+X_{H} \), where \( H=pdq(\nabla f) \) for some small Morse func­tion \( f\colon M\to\mathbb{R} \), \( X_{H} \) is the Hamilto­ni­an vec­tor field, and the ex­haust­ing Morse func­tion is \( F(q,p)=\frac12 p^{2}+f(q) \).

Con­sider now a Wein­stein man­i­fold \( X \). Us­ing the equa­tion \( L_{Z}\omega=\omega \), it is straight­for­ward to check that the stable man­i­folds \( W^{\mathrm{s}} \) of the crit­ic­al points are iso­trop­ic, that is, \( \omega \) van­ishes along \( W^{\mathrm{s}} \). In par­tic­u­lar, crit­ic­al points have in­dices \( \le n \), and or­der­ing the crit­ic­al levels ac­cord­ing to in­dex then gives the fol­low­ing handle de­com­pos­i­tion of \( X \): \[ \overline{X} = \overline{X}^n \supset \overline{X}^{n-1} \supset \dots \supset \overline{X}^{1} \supset \overline{X}^0, \quad \text{ where } \quad \overline{X}^{m} = \overline{X}^{m-1} \cup_{\Lambda^{m}} H^{m}, \quad m=1,\dots,n. \] Here \( \overline{X}^{0} \) is the \( 2n \)-ball and \( \overline{X}^{m} \) is ob­tained from \( \overline{X}^{m-1} \) by at­tach­ing \( m \)-handles \( H^{m}\approx \bigsqcup \overline{T^{\ast} D^{m}}\times B^{2(n-m)} \) along a col­lec­tion of iso­trop­ic \( (m-1) \)-spheres \( \Lambda^{m} \) (the des­cend­ing spheres of the iso­trop­ic stable man­i­folds) in the con­tact bound­ary \( Y^{m-1} \) of \( X^{m-1} \).

Figure 1.  A Lagrangian handle, core disk \( L \) and attaching sphere \( \Lambda \), cocore disk \( C \) with Legendrian boundary \( \Gamma \).

For \( m < n \) there is an \( h \)-prin­ciple for iso­trop­ic \( m \)-spheres in con­tact \( (2n-1) \)-man­i­folds, which means that the sym­plect­ic to­po­logy of \( X^{m} \) is uniquely de­term­ined by the ho­mo­topy class of the tan­gent map of \( \Lambda_{m} \). Con­sequently all the in­ter­est­ing sym­plect­ic to­po­logy of \( X=X^{n} \) is car­ried by the Le­gendri­an iso­topy class of the at­tach­ing spheres \( \Lambda=\Lambda^{n}\subset Y^{n-1}=\partial \overline{X}^{n-1} \). This ob­ser­va­tion can be taken as the start­ing point for the Le­gendri­an sur­gery ap­proach to sym­plect­ic to­po­logy in­vari­ants. It says that all sym­plect­ic in­vari­ants of \( X \) can be ob­tained from \( \Lambda \).

We will call the stable man­i­folds \( L \) of the in­dex \( n \) crit­ic­al points the core disks and the un­stable man­i­folds \( C \) its cocore disks. The first Le­gendri­an sur­gery res­ult we will ex­plain can then be stated as say­ing that the wrapped Flo­er co­homo­logy \( CW^{\ast}(C) \) of \( C \) is quasii­so­morph­ic to the Chekan­ov–Eli­ash­berg dg-al­gebra \( CE^{\ast}(\Lambda) \) of \( \Lambda \).

Be­low we will sim­pli­fy nota­tion and write \( \overline{X}=\overline{X}_{0}\cup_{\Lambda} H \), where \( \overline{X}_{0} \) is the sub­crit­ic­al part of \( X \), that is, the sub­level set of all handles of in­dex \( < n \) and \( H \) is the uni­on of crit­ic­al \( n \)-handles, one copy of \( T^{\ast}D^{n} \) for each handle, at­tached along \( \Lambda\subset Y_{0}=\partial\overline{X}_{0} \). We will also sim­pli­fy nota­tion and write, for ex­ample, \( X=X_{0}\cup_{\Lambda} H \), etc., and not dis­tin­guish in nota­tion between a Wein­stein do­main and the Wein­stein man­i­fold which is its com­ple­tion. We use the nota­tion \( Y=\partial_{\infty} X \) and \( Y_{0}=\partial_{\infty} X_{0} \) to de­note ideal con­tact bound­ar­ies. Fur­ther­more, we will write \( L \) and \( C \) for the stable and un­stable man­i­folds of the in­dex \( n \) crit­ic­al points in \( H \), and call them the Lag­rangi­an core and cocore disks, re­spect­ively. See Fig­ure 1.

3.  The basic surgery isomorphism

3.1.  Reeb orbits and Reeb chords
Con­sider a con­tact man­i­fold \( V \) with con­tact form \( \alpha \), in our case \( V=Y=\partial_{\infty} X \) or \( V=Y_{0}=\partial_{\infty} X_{0} \) and \( \alpha \) is the re­stric­tion of \( \lambda \). The Reeb vec­tor field \( R \) of \( \alpha \) is the vec­tor field that gen­er­ates the ker­nel of \( d\alpha \), \( \iota_{R}d\alpha=0 \), nor­mal­ized by \( \alpha(R)=1 \).

Flow loops of \( R \) are called Reeb or­bits. For gen­er­ic con­tact form, Reeb or­bits are isol­ated and trans­verse (lin­ear­ized re­turn map in gen­er­al po­s­i­tion). For our treat­ment of holo­morph­ic curve the­or­ies we will use para­met­er­ized Reeb or­bits. This means that we fix a point on each Reeb or­bit and con­sider all its mul­tiples para­met­er­ized, start­ing the para­met­er­iz­a­tion at that point. If \( \Lambda\subset V \) is a Le­gendri­an sub­man­i­fold then Reeb chords of \( \Lambda \) are flow lines of \( R \) start­ing and end­ing on \( \Lambda \). Again, for gen­er­ic data, Reeb chords are isol­ated and trans­verse (lin­ear­ized flow from start point to en­d­point in gen­er­al po­s­i­tion).

We next in­tro­duce the chain com­plexes un­der­ly­ing our main holo­morph­ic curve the­or­ies. We start with the the­ory after handle at­tach­ment.

3.1.1.  Generators of the wrapped Floer cohomology complex
Con­sider a Lag­rangi­an \( C\subset X \) with ideal Le­gendri­an bound­ary \( \Gamma\subset Y \). Be­low \( C \) will be the cocore disks of a Lag­rangi­an handle at­tach­ment, but the defin­i­tion makes sense more gen­er­ally. The wrapped Flo­er co­homo­logy com­plex of \( C \) is most of­ten defined by pick­ing a Hamilto­ni­an dis­place­ment of \( C \) with pre­scribed be­ha­vi­or at in­fin­ity and tak­ing gen­er­at­ors Hamilto­ni­an chords of \( C \) and dif­fer­en­tial count­ing per­turbed holo­morph­ic curves asymp­tot­ic to the gen­er­at­ors. For Le­gendri­an sur­gery pur­poses it is ne­ces­sary to use the fol­low­ing equi­val­ent defin­i­tion.

Con­sider a Morse func­tion \( f\colon C\to C \) with gradi­ent equal to the Li­ouville vec­tor field at in­fin­ity. Let \( C_{-} \) be a time \( -\epsilon \) shift along \( f \), viewed as a Hamilto­ni­an in \( T^{\ast} C \). Then the wrapped Flo­er co­homo­logy com­plex \( CW^{\ast}(C) \) of \( C \) is gen­er­ated by Reeb chords start­ing on \( C \) and end­ing on \( C_{-} \) and in­ter­sec­tion points in \( C\cap C_{-} \). Since the shift is small we see that gen­er­at­ors cor­res­pond to Reeb chords of \( \Gamma \) and crit­ic­al points of \( f \); see ([e8], Ap­pendix B.1).

3.1.2.  Generators of the Chekanov–Eliashberg dg-algebra
Be­fore handle at­tach­ment, we use the Chekan­ov–Eli­ash­berg dg-al­gebra \( CE^{\ast}(\Lambda) \) of a Le­gendri­an \( \Lambda\subset Y_{0} \). It is the al­gebra of com­pos­able words of Reeb chords of \( \Lambda \), where a word of chords is com­pos­able if the Reeb chord en­d­point of a chord lies in the same con­nec­ted com­pon­ent of \( \Lambda \) as the start point of the fol­low­ing chord. Mul­ti­plic­a­tion is con­cat­en­a­tion, where the product is zero if the res­ult is non­com­pos­able.
3.2.  Handle attachment and Reeb dynamics
Lemma 3.1: Reeb chords of the cocore bound­ary sphere \( \Gamma=\partial_{\infty} C \) are in nat­ur­al 1-1 cor­res­pond­ence with com­pos­able words of Reeb chords of the Le­gendri­an at­tach­ing link \( \Lambda \). More pre­cisely, giv­en an ac­tion level \( \mathfrak{a} > 0 \) there ex­ists a handle size \( \delta > 0 \) such that for all handles of size smal­ler than \( \delta \), Reeb chords of \( \Gamma \) of ac­tion less than \( \mathfrak{a} \) are in nat­ur­al 1-1 cor­res­pond­ence with com­pos­able words of Reeb chords of \( \Lambda \) of total ac­tion less than \( \mathfrak{a} \).

Proof. Out­side the handle at­tach­ment re­gion, the Reeb flows in \( Y_{0} \) and \( Y \) are nat­ur­ally iden­ti­fied. We take the cocore as the fiber of \( T^{\ast} D \) at the cen­ter of \( D \). In­side the handle the Reeb flow is the lift of the geodes­ic flow. In par­tic­u­lar, the Reeb flow takes fiber hemi­spheres of the at­tach­ing neigh­bor­hood to the bound­ary of the at­tach­ing re­gion by a de­gree one map. A straight­for­ward fixed point ar­gu­ment now es­tab­lishes the 1-1 cor­res­pond­ence; see ([e4], Sec­tion 5) for de­tails. □

3.3.  Wrapped Floer cohomology
Figure 2.  Holomorphic disks for the \( \mu_k \)-operations on \( CW^{\ast}(C) \).
We first con­sider the dif­fer­en­tial \( d \) in the wrapped Flo­er co­homo­logy com­plex \( CW^{\ast}(C) \). This dif­fer­en­tial counts anchored holo­morph­ic strips with a pos­it­ive and neg­at­ive punc­ture at a Reeb chord con­nect­ing \( \Gamma_{-} \) to \( \Gamma \) or an in­ter­sec­tion point between \( C_{-} \) and \( C \). Here “an­chor­ing” means that we con­sider disks with ad­di­tion­al neg­at­ive punc­tures at marked or­bits and pure chords where we fill with ri­gid punc­tured spheres and disks. The ba­sic prop­erty of the curve-count­ing map is the fol­low­ing.
Lemma 3.2: The map \( d\colon CW^{\ast}(C)\to CW^{\ast}(C) \) is a dif­fer­en­tial \( d^2=0 \).

Proof. The square of the dif­fer­en­tial counts the ends of a 1-di­men­sion­al mod­uli space; see ([e8], Ap­pendix B.1). □

In fact, the dif­fer­en­tial is the \( \mu_{1} \)-op­er­a­tion in an \( A_{\infty} \)-al­gebra struc­ture on \( CW^{\ast}(C) \), where the op­er­a­tion \( \mu_{k} \) is defined by choos­ing \( k \) par­al­lel cop­ies of \( C \), where \( C_{-} \) is the first par­al­lel copy which is the time \( -1 \) shift along the func­tion \( \epsilon f_{1} \) for \( f_{1}=f \) and the \( k^\mathrm{ th} \) copy is the time \( -1 \) shift along the func­tion \[ \left(\sum_{r=1}^{k} \epsilon^{k}\right) f_{1} + \left(\sum_{r=2}^{k} \epsilon^{3+r}\right)f_{2} + \dots + \epsilon^{k+1}f_{k}. \]

Then the \( \mu_{k} \)-op­er­a­tion counts disks with pos­it­ive punc­tures con­nect­ing cop­ies in in­creas­ing or­der and one neg­at­ive out­put punc­ture; see Fig­ure 2.

We then have the fol­low­ing res­ult.

Lemma 3.3: The op­er­a­tions \( \mu_{k}\colon CW^{\ast}(C)^{\otimes k}\to CW^{\ast}(C) \) sat­is­fy the \( A_{\infty} \)-re­la­tions.

Proof. Again the com­pos­i­tions count the ends of a 1-di­men­sion­al mod­uli space. The proof re­quires that for suf­fi­ciently close par­al­lel cop­ies, mod­uli spaces with bound­ary com­pon­ents on Lag­rangi­ans cor­res­pond­ing to any in­creas­ing num­ber­ing in the sys­tem of par­al­lel cop­ies be ca­non­ic­ally iden­ti­fied. This fol­lows from the form of the shift­ing func­tions and the fact that trans­versely cut-out solu­tions per­sist un­der suf­fi­ciently small per­turb­a­tion; see ([e8], Ap­pendix B.1).

3.4.  The Chekanov–Eliashberg dg-algebra
Figure 3.  Holomorphic disks for the differential on \( CE^{\ast}(\Lambda) \).
We next con­sider the dif­fer­en­tial \( \partial \) of the Chekan­ov–Eli­ash­berg dg-al­gebra. The dif­fer­en­tial counts holo­morph­ic disks in the sym­plect­iz­a­tion \( Y_{0}\times \mathbb{R} \) anchored at or­bits with one pos­it­ive and sev­er­al neg­at­ive punc­tures; see Fig­ure 3.

We ex­tend it to a dg-al­gebra dif­fer­en­tial by the Leib­n­iz rule.

Lemma 3.4: The map \( \partial\colon CE^{\ast}(\Lambda)\to CE^{\ast}(\Lambda) \) is a dif­fer­en­tial, \( \partial^{2}=0 \).

Proof. The square of the dif­fer­en­tial counts ends of a 1-di­men­sion­al mod­uli space; see, for ex­ample, [e8]. □

3.5.  The chain map
The cobor­d­ism \( W=X\setminus X_{0} \), with the Lag­rangi­ans \( L \) and \( C \) with con­nec­ted com­pon­ents that in­ter­sect at single points in the middle of each handle, gives a nat­ur­al chain map \begin{equation} \tag{3.1} \Phi^{CW}\colon CW^{\ast}(C)\to CE^{\ast}(\Lambda),\quad \Phi^{CW}=\sum_{k=1}^{\infty}\Phi^{CW}_{j}. \end{equation} More pre­cisely, \( \Phi_{k}^{CW} \) counts holo­morph­ic curves with \( k \) pos­it­ive punc­tures at Reeb chords of \( C \), two punc­tures where the map is asymp­tot­ic to in­ter­sec­tions in \( C\cap L \) and sev­er­al neg­at­ive punc­tures at Reeb chords of \( \Lambda \), where mod­uli spaces of many pos­it­ive punc­tures of \( C \) are defined us­ing sys­tems of par­al­lel cop­ies as in the defin­i­tion of the \( A_{\infty} \)-op­er­a­tions in or­der to avoid “bound­ary break­ing”; see Fig­ure 4.
Figure 4.  Holomorphic disks for the surgery map \( CW^{\ast}(C)\to CE^{\ast}(\Lambda) \).

We view \( CE^{\ast}(\Lambda) \) as an \( A_{\infty} \)-al­gebra gen­er­ated by words of Reeb chords with dif­fer­en­tial \( \mu_{1}=\partial \), with \( \mu_{2} \) giv­en by the con­cat­en­a­tion product, and with all high­er \( \mu_k \) equal to zero.

Lemma 3.5: The map \( \Phi^{CW}=\sum_{k}\Phi_{k}^{CW} \) in Equa­tion (3.1) is an \( A_{\infty} \)-map.

Proof. To see this we con­sider 1-di­men­sion­al mod­uli spaces of disks, as in the defin­i­tion of \( \Phi_{k} \), and note that the bound­ary points of such a mod­uli do cor­res­pond to split­tings at Reeb chords, which is pre­com­pos­ing \( \Phi_{k} \) with \( \mu_{j} \) in the pos­it­ive end or post­com­pos­ing \( \Phi_{k} \) with \( \mu_{1} \) at the neg­at­ive end, or split­ting at one of the Lag­rangi­an in­ter­sec­tion points in \( C\cap L \), which cor­res­ponds to post­com­pos­ing \( \Phi_{j} \) with \( \mu_{2} \) in the neg­at­ive end. The lemma fol­lows; see ([e8], Ap­pendix B.2) for more de­tails. □

3.6. Chain isomorphism
We show in this sec­tion that the \( A_{\infty} \)-map \( \Phi^{CW} \) in Equa­tion (3.1) is in fact a chain iso­morph­ism. More pre­cisely, we have the fol­low­ing.
Lemma 3.6: For any ac­tion cut-off \( \mathfrak{a}_{0} > 0 \), the im­age un­der \( \Phi_{1}^{CW} \) of a Reeb chord \( \overline{w} \) of \( \Gamma \) cor­res­pond­ing to a word \( w \) of Reeb chords of \( \Lambda \) of ac­tion \( \mathfrak{a}(\overline w) < \mathfrak{a}_{0} \) is \[ \Phi_{1}^{CW}(\overline{w}) = \pm w + E(\overline{w}), \] where \( E(\overline{w}) \) is a sum of Reeb chord words \( v \) of ac­tion \( \mathfrak{a}(v) < \mathfrak{a}(w) \).
Figure 5.  Constructing isomorphism disks by gluing.

Proof. The proof uses in­duc­tion on ac­tion. For one-let­ter words an ex­pli­cit con­struc­tion gives one trans­versely cut-out strip which is unique by an ac­tion ar­gu­ment. As­sume in­duct­ively that the count of curves in mod­uli space of disks con­nect­ing \( \overline{w} \) to \( w \) for all words of length \( \le m \) equals \( \pm 1 \). Then con­sider the mod­uli space of disks with two pos­it­ive punc­tures at \( \overline{w}_{0} \) and \( \overline{w}_{1} \) and neg­at­ive punc­tures at \( w_{0}w_{1} \). By ac­tion this mod­uli space has only two bound­ary break­ings (see Fig­ure 5): break­ing at a point in \( C\cap L \), by in­duc­tion we know that the count of such con­fig­ur­a­tions equal \( \pm 1 \), and break­ing in­to a disk with two pos­it­ive and one neg­at­ive punc­ture fol­lowed by an iso­morph­ism disk, we con­clude that both of these mod­uli spaces must also con­tain \( \pm 1 \) ele­ments; see ([e4], Sec­tion 7). □

Us­ing a straight­for­ward ac­tion fil­tra­tion ar­gu­ment (see ([e8], Ap­pendix B.2)) we then get the main sur­gery iso­morph­ism the­or­em:

The­or­em 3.7: The nat­ur­al \( A_{\infty} \)-chain map \[ \Phi^{CW}\colon CW^{\ast}(C)\to CE^{\ast}(\Lambda) \] is a quasii­so­morph­ism. □

4.  Further results

The Le­gendri­an sur­gery proof is power­ful and can be used to de­rive many closely re­lated res­ults. In this sec­tion we will dis­cuss some of them.
4.1.  Symplectic homology and linearized contact homology
Chan­ging the sur­gery iso­morph­ism we can in­stead map in­to sym­plect­ic co­homo­logy. The stand­ard con­struc­tion of sym­plect­ic co­homo­logy of \( X \) starts from a time-de­pend­ent Hamilto­ni­an with stand­ard lin­ear be­ha­vi­or at in­fin­ity of slope not in the ac­tion spec­trum of the con­tact form. The com­plex is gen­er­ated by Hamilto­ni­an or­bits and the dif­fer­en­tial counts solu­tions to the Cauchy–Riemann equa­tions per­turbed by the Hamilto­ni­an vec­tor field on cyl­in­ders that in­ter­pol­ate between or­bits. Here we will use a ver­sion of sym­plect­ic co­homo­logy that arises by first mak­ing the Hamilto­ni­an time-in­de­pend­ent, where time-in­de­pend­ent or­bits is viewed as Bott-de­gen­er­ate time-de­pend­ent or­bits, and then turn­ing the Hamilto­ni­an off, where per­turbed curves be­come stand­ard holo­morph­ic curves with asymp­tot­ic mark­ers re­mem­ber­ing the Morse the­ory on the or­bits.

Our com­plex for sym­plect­ic ho­mo­logy is thus Reeb or­bits to­geth­er with a Morse com­plex of the un­der­ly­ing geo­met­ric or­bit and crit­ic­al points of a Morse func­tion on \( X \). The dif­fer­en­tial counts anchored holo­morph­ic cyl­in­ders and spheres with Morse data at or­bits; we call these dec­or­ated or­bits. We de­note the cor­res­pond­ing com­plex \( SC^{\ast}(X) \); see ([3], Sec­tion 3.3).

We next want to mod­el this com­plex be­fore sur­gery. The coun­ter­part of Lemma 3.1 says that Reeb or­bits after sur­gery are in nat­ur­al 1-1 cor­res­pond­ence with Reeb or­bits be­fore sur­gery and cyc­lic words of Reeb chords of the at­tach­ing spheres; see ([e4], Sec­tion 5). In or­der to mod­el this situ­ation we use a two-copy \( L_0\cup L_{1} \) of the des­cend­ing man­i­fold. Here \( L_{0}\cap L_{1}=\{z\} \) is one point near the middle of the handle, and Reeb chords of the bound­ary of \( \Lambda_0\cup\Lambda_1 \) con­sist of two cop­ies of Reeb chords of \( \Lambda \) and two ad­di­tion­al Reeb chords \( x \) and \( y \) for each com­pon­ent of \( \Lambda \) cor­res­pond­ing to the max­im­um and a min­im­um of a Morse func­tion mak­ing the Reeb shift gen­er­ic.

We call a Reeb chord con­nect­ing \( \Lambda_0 \) to \( \Lambda_1 \) mixed and chords con­nect­ing \( \Lambda_{j} \) to it­self pure. We then con­sider the two-copy Chekan­ov–Eli­ash­berg dg-al­gebra com­plex which is gen­er­ated by words of chords in which ex­actly one chord is mixed and oth­ers are pure. The dif­fer­en­tial counts holo­morph­ic curves with mixed pos­it­ive punc­ture, one mixed neg­at­ive punc­ture, and any num­ber of pure neg­at­ive punc­tures. Rather than us­ing this com­plex alone we use the (very small) Lag­rangi­an Flo­er co­homo­logy com­plex of \( L_0 \) and \( L_{1} \) with coef­fi­cients in the two-copy com­plex, where the dif­fer­en­tial also counts curves with one Lag­rangi­an in­ter­sec­tion point and neg­at­ive end with one mixed punc­ture. In this case, this means simply add the gen­er­at­or \( z \) and ob­serve that \( \partial z = y \). We de­note this com­plex \( CF^{\ast}((L_{0},\Lambda_0),(L_{1},\Lambda_1)) \) and write simply \( \operatorname{H} CF^{\ast}(L,\Lambda) \) for the cor­res­pond­ing Hoch­schild com­plex that is ob­tained by identi­fy­ing words up to cyc­lic per­muta­tion.

There is now a nat­ur­al sur­gery cobor­d­ism map \[ \Phi^{SC}=\Phi_{\mathrm{o}}^{SC}\oplus\Phi_{\mathrm{w}}^{SC}\colon SC^{\ast}(X) \to SC^{\ast}(X_0)\oplus \operatorname{H} CF^{\ast}(L,\Lambda), \] where \( \Phi_{\mathrm{o}}^{SC} \) is the stand­ard cobor­d­ism map count­ing cyl­in­ders between Morse dec­or­ated or­bits, and where \( \Phi_{\mathrm{w}}^{SC} \) counts disks with pos­it­ive punc­ture at a dec­or­ated or­bit, mixed dis­tin­guished punc­ture at 1 and a punc­ture map­ping to \( z \) at \( -1 \). We think of the right-hand side as a com­plex with dif­fer­en­tial \[ d = \left(\begin{matrix} d_{\mathrm{oo}} & d_{\mathrm{wo}}\\ d_{\mathrm{ow}} & d_{\mathrm{ww}} \end{matrix}\right), \] where \( d_{\mathrm{oo}} \) is the usu­al cyl­in­der count­ing dif­fer­en­tial on \( SC^{\ast}(X_0) \), \( d_{\mathrm{wo}}=0 \), \( \,d_{\mathrm{ww}} \) is in­duced from the dif­fer­en­tial on \( \operatorname{H} CF^{\ast}(L,\Lambda) \), and where \( d_{\mathrm{ow}} \) counts disks with pos­it­ive punc­ture at a Morse dec­or­ated or­bit and dis­tin­guished neg­at­ive punc­ture at mixed chord, sim­il­ar to \( \Phi_{\mathrm{w}} \).

The­or­em 4.1: The map \( \Phi^{SC} \) is a chain iso­morph­ism.

Proof. We ad­apt the Morse func­tions on the or­bits cor­res­pond­ing to cyc­lic words so that these func­tions have a max­im­um and a min­im­um on each chord on the un­der­ly­ing or­bit. It then fol­lows that \( \Phi^{CW} \) takes an or­bit with max­im­um on a chord \( c \) to the cor­res­pond­ing word of chords with the chord \( c \) mixed, and the or­bit with a min­im­um on \( c \) to the cor­res­pond­ing word with \( c \) re­placed by \( cx \), where \( c \) is the pure chord and \( x \) is the mixed chord at the min­im­um Reeb chord of the shift. The proof is then dir­ectly ana­log­ous to The­or­em 3.7. □

This has the fol­low­ing con­sequence.

Co­rol­lary 4.2: The chain map \( \Phi_{\mathrm{w}}^{SC}\colon SC^{\ast}(X) \to \operatorname{H} CF^{\ast}(L,\Lambda) \) is a quasii­so­morph­ism.

Proof. The com­plex \( SC^{\ast}(X_0) \) is con­tract­ible as the sym­plect­ic ho­mo­logy com­plex of a sub­crit­ic­al man­i­fold. □

Com­bin­ing The­or­em 3.7 and Co­rol­lary 4.2 we learn that the Hoch­schild com­plex \( \operatorname{H} CW^{\ast}(C) \) of \( CW^{\ast}(C) \) is quasii­so­morph­ic to the sym­plect­ic ho­mo­logy com­plex \( SC^{\ast}(X) \).

It is also pos­sible to define the coun­ter­part of \( SC^{\ast}(X) \) without Morse data on the or­bits. The cor­res­pond­ing com­plex is known as cyl­indric­al con­tact ho­mo­logy and is iso­morph­ic to the \( S^{1} \)-equivari­ant ver­sion of \( SC^{\ast}(X) \). The stand­ard ap­proach to de­fin­ing this \( S^{1} \)-equivari­ant ver­sion is to define a BV-op­er­at­or that de­forms the Hamilto­ni­an per­turb­a­tion by ro­tat­ing the do­main. This op­er­a­tion does not square to zero and one has to add (in­fin­itely many) cor­rec­tion terms. However, in the Morse–Bott de­scrip­tion of \( SC^{\ast}(X) \) the BV-op­er­at­or \( \xi \) ad­mits a simple de­scrip­tion that does square to zero: if \( \gamma \) is a Reeb or­bit, and \( \widehat{\gamma} \) and \( \check{\gamma} \) de­note \( \gamma \) dec­or­ated by a max­im­um and a min­im­um, re­spect­ively, then \( \xi(\check{\gamma})=\widehat{\gamma} \) and \( \xi(\widehat{\gamma})=0 \). The cor­res­pond­ing op­er­a­tion on cyc­lic words is \( \underline{x}w\mapsto \sum \underline{w} \), where the sum goes over all ways of choos­ing mixed chord in \( w \). With this ex­pli­cit form of the BV-op­er­at­or it is straight­for­ward to ob­tain the cyl­indric­al con­tact ho­mo­logy us­ing a mod­el for \( S^{1} \)-equivari­ant ho­mo­logy on \( \operatorname{H} CF^{\ast}(L,\Lambda) \) to­geth­er with the \( S^{1} \)-ac­tion giv­en by the BV-op­er­at­or \( \xi \) sat­is­fy­ing \( \xi^{2}=0 \).

4.2.  Upside down surgery

A Lag­rangi­an handle has the form \( T^{\ast}D^{n} \). Here we think of \( \partial D^{n}\times\{0\} \) as the at­tach­ing sphere \( \Lambda \) that we fill by the core disk \( D^{n}\times \{0\} \) which is then \( L \). Note that the dy­nam­ics of the handle are the same if we view it from the oth­er side: start from the bound­ary of the fiber at 0, that is, \( \Gamma \) the bound­ary of the cocore disk, and at­tach the cocore disk \( C \). We call this up­side down Lag­rangi­an handle at­tach­ment or up­side down Le­gendri­an sur­gery. This then means that if we shrink the size of the handle around \( \Gamma \) we have the coun­ter­part of Lemma 3.1:
Lemma 4.3: If \( Y_{0} \) is equipped by the con­tact form in­duced from that on \( Y \) by up­side down Le­gendri­an sur­gery, Reeb chords of \( \Lambda \) are in nat­ur­al 1-1 cor­res­pond­ence with com­pos­able words of Reeb chords of \( \,\Gamma \). □

By ana­logy with \( \Phi^{CW} \), we now have a chain map of in­fin­ity coal­geb­ras \[ \Phi_{CW}\colon \operatorname{B} CW^{\ast}(C)\to LCE^{\ast}(\Lambda). \] Here the left-hand side is the bar com­plex of \( CW^{\ast}(C) \), that is, \( \operatorname{B} CW^{\ast}(C) \) is gen­er­ated by words of Reeb chords of \( \Gamma \) and crit­ic­al points in \( C \) with dif­fer­en­tial in­duced by the \( \mu_{k} \)-op­er­a­tions. This bar com­plex \( \operatorname{B} CW^{\ast}(C) \) is nat­ur­ally a (in­fin­ity) coal­gebra with dif­fer­en­tial \( c_1 \), cop­roduct \( c_2 \) cor­res­pond­ing to split­ting words in two all pos­sible ways and all high­er \( c_{k} \) equal to zero. The right-hand side is the lin­ear­ized Chekan­ov–Eli­ash­berg dg-al­gebra (here we as­sume that \( CE^{\ast}(\Lambda) \) has an aug­ment­a­tion) where the op­er­a­tions \( c_{k} \) cor­res­pond to the part of the (aug­men­ted) dif­fer­en­tial with \( k \) neg­at­ive punc­tures. In dir­ect ana­logy with The­or­em 3.7 we have the fol­low­ing.

The­or­em 4.4: The map \( \Phi_{CW} \) is a quasii­so­morph­ism. □
Figure 6.  Curves counted by the open-closed map.

We con­sider also the cyc­lic ver­sion of this map. We start from the Hoch­schild com­plex \( \operatorname{H} CW^{\ast}(C) \) of \( CW^{\ast}(C) \) gen­er­ated by cyc­lic words of Reeb chords, one of which is dis­tin­guished. In ana­logy with the core disk, we think of this as a two-copy com­plex where the dis­tin­guished punc­ture is mixed and where we also have the in­ter­sec­tion point \( z=C_{0}\cap C_1 \) as a gen­er­at­or with \( \partial z=x \), where is the min­im­um of the Reeb shift at in­fin­ity and where the max­im­um \( y \) of the Reeb shift plays a role ana­log­ous to \( x \) for the two copy of \( L \). We then add to \( \operatorname{H} CW^{\ast}(C) \) the com­plex \( SC^{\ast}(X) \). The dif­fer­en­tial counts, ex­cept for the usu­al curves also disks with sev­er­al pos­it­ive punc­tures and a neg­at­ive punc­ture at the cen­ter, the loc­a­tion of the dis­tin­guished pos­it­ive punc­ture at a fixed bound­ary loc­a­tion de­term­ined by the mark­er at or­bit in the cen­ter and an an­ti­pod­al punc­ture map­ping to \( z \). We de­note this com­plex \[ \operatorname{H} CW^{\ast}(C)\oplus SC^{\ast}(X) \] and in ana­logy with The­or­em 4.1 we have a nat­ur­al chain map \[ \Psi\colon \operatorname{H} CW^{\ast}(C)\oplus SC^{\ast}(X) \to SC^{\ast}(X_{0}) \] which is a chain iso­morph­ism. The con­nect­ing ho­mo­morph­ism in the long ex­act se­quence as­so­ci­ated to the short ex­act se­quence \[ 0 \to SC^{\ast}(X) \to \operatorname{H} CW^{\ast}(C)\oplus SC^{\ast}(X) \to \operatorname{H} CW^{\ast}(C) \to 0 \] is called the open-closed map; see [e1], [e2]. We will de­note it \[ \mathcal{OC}\colon \operatorname{H} CW^{\ast}(C)\to SC^{\ast}(X). \] See Fig­ure 6.

Since \( SC^{\ast}(X_{0}) \) is con­tract­ible we find:

Co­rol­lary 4.5: The open-closed map \( \mathcal{OC}\colon \operatorname{H} CW^{\ast}(C)\to SC^{\ast}(X) \) is a quasii­so­morph­ism. □
4.3.  The closed-open map and isomorphisms of Hochschild homology and cohomology
The open-closed and closed-open maps were ori­gin­ally stud­ied in the set­ting of wrapped Flo­er co­homo­logy from the Hamilto­ni­an view point; see [e2], [e1]. Here we con­tin­ue in­stead with the Le­gendri­an sur­gery per­spect­ive and use wrapped Flo­er co­homo­logy without Hamilto­ni­an; see ([e8], Ap­pendix B.1).

The Hoch­schild chain com­plex \( \operatorname{H} CW^{\ast}(C) \) above can be thought of as gen­er­ated by words of neg­at­ive punc­tures at Reeb chords along the bound­ary of a form­al disk (one of which is dis­tin­guished, we take it to be mixed). The Hoch­schild dif­fer­en­tial then at­taches holo­morph­ic disks with sev­er­al pos­it­ive and one neg­at­ive punc­ture to such words in all pos­sible ways (the neg­at­ive and one of the pos­it­ive punc­tures mixed). Sim­il­arly, we con­sider the Hoch­schild co­chain com­plex \( \operatorname{H}^{\prime} CW^{\ast}(C) \) gen­er­ated by chords along the bound­ary of a form­al disk, all pos­it­ive ex­cept one which is neg­at­ive (as in the dif­fer­en­tial the neg­at­ive and one pos­it­ive punc­ture mixed). The dif­fer­en­tial on \( \operatorname{H}^{\prime} CW^{\ast}(C) \) is ob­tained by at­tach­ing the disks cor­res­pond­ing to the dif­fer­en­tial on \( \operatorname{H} CW^{\ast}(C) \) with one neg­at­ive and sev­er­al pos­it­ive punc­tures in all pos­sible ways, from above and be­low.

The com­plex \( \operatorname{H}^{\prime} CW^{\ast}(C) \) has a nat­ur­al product which acts by glu­ing neg­at­ive and pos­it­ive punc­tures of one to the oth­er. This product has a nat­ur­al unit \( e \) which is the sum of words of two-punc­tured disks with the same chord at the pos­it­ive and neg­at­ive punc­ture. Fur­ther­more, there is a curve-count­ing map, the closed-open map \[ \mathcal{CO}\colon SC^{\ast}(X)\to \operatorname{H}^{\prime} CW^{\ast}(C), \] which counts disks with a pos­it­ive punc­ture with a mark­er at the cen­ter map­ping to an or­bit with Morse dec­or­a­tion, one mixed neg­at­ive bound­ary punc­ture at a fixed po­s­i­tion, and any num­ber of pos­it­ive bound­ary punc­tures with the mixed chord at the point op­pos­ite the fixed po­s­i­tion; see Fig­ure 7.

Figure 7.  Curves counted by the closed-open map.
Lemma 4.6: The map \( \mathcal{CO} \) is a chain map that re­spects the product (on ho­mo­logy).

Proof. The sym­plect­ic co­homo­logy product fol­lowed by the iso­morph­ism gives a disk with two pos­it­ive punc­tures. Look­ing at pos­sible split­tings we find the product on the Hoch­schild co­chains up to ex­act terms. □

Con­sider now the com­plex \( \operatorname{H} CW^{\ast}(C) \). This is nat­ur­ally a \( \operatorname{H}^{\prime}CW^{\ast}(C) \)-mod­ule: if \( a\in \operatorname{H} CW^{\ast}(C) \) is a cyc­lic word and \( r\in\operatorname{H}^{\prime} CW^{\ast}(C) \), then \( r\cdot a \) is the sum of cyc­lic words ob­tained by at­tach­ing the pos­it­ive punc­tures of \( r \) to con­sec­ut­ive (neg­at­ive) punc­tures of \( a \). Pre­com­pos­ing the product by the map \( \mathcal{CO} \) we find that \( \operatorname{H} CW^{\ast}(C) \) is also an \( SC^{\ast}(X) \)-mod­ule. The sym­plect­ic ho­mo­logy com­plex has the pairs-of-pants product and is hence an \( SC^{\ast}(X) \)-mod­ule it­self. We have the fol­low­ing.

Lemma 4.7: The open-closed map \( \mathcal{OC}\colon \operatorname{H} CW^{\ast}(C)\to SC^{\ast}(X) \) is a map of \( SC^{\ast}(X) \)-mod­ules.

Proof. The bound­ary of the 1-di­men­sion­al (re­duced) mod­uli spaces of curves with pos­it­ive chord punc­tures at the bound­ary, one neg­at­ive and one pos­it­ive in­teri­or punc­ture on the dis­tin­guished ray from the cen­ter to the dis­tin­guished punc­ture, cor­res­pond ex­actly to first mul­tiply­ing and then ap­ply­ing the iso­morph­ism or first ap­ply­ing the iso­morph­ism and then mul­tiply­ing. □

To­geth­er with a de­gen­er­a­tion of mod­uli spaces ar­gu­ment, this leads to the fol­low­ing res­ult.

The­or­em 4.8: The map \( \mathcal{CO}\colon SC^{\ast}(X)\to \operatorname{H}^{\prime} CW^{\ast}(C) \) is a quasii­so­morph­ism.

Proof. We use two par­al­lel fibers \( C_0 \) and \( C_1 \) as source of the map \( \mathcal{OC} \) and tar­get of the map \( \mathcal{CO} \), re­spect­ively. We will take them to lie close to­geth­er. Note that for the Le­gendri­ans at in­fin­ity of the par­al­lel fibers, the shift from \( \partial_{\infty}C_0 \) to \( \partial_{\infty} C_{1} \) is in­duced by a Morse func­tion on the sphere that is the re­stric­tion of a lin­ear func­tion that is neg­at­ive at the south pole, van­ishes on the equat­or, and pos­it­ive at the north pole. Note then that Reeb chords \( C_0\to C_1 \) and \( C_1\to C_0 \) cor­res­pond to Reeb chords \( C\to C \) and short Reeb chords near \( C_0 \), and that there­fore, there is a nat­ur­al 1-1 cor­res­pond­ence between Reeb chords \( C_0\to C_{1} \) and Reeb chords \( C_1\to C_0 \).

We re­mark that when we dis­cuss mod­uli spaces of holo­morph­ic curves for the Lag­rangi­ans \( C_0 \) and \( C_{1} \) be­low, we need to use sys­tems of par­al­lel cop­ies as de­scribed above. We use sep­ar­ate sys­tems for \( C_{0} \) and \( C_{1} \) (see ([e8], Ap­pendix B)), but will leave these sys­tems im­pli­cit in the nota­tion.

Let 1 de­note the unit in \( SC^{\ast}(X) \). In the com­plex \( SC^\ast(X) \), 1 is rep­res­en­ted by the min­im­um of the Morse func­tion on \( X \). Since \( \mathcal{CO} \) is an iso­morph­ism, we find \( u\in \operatorname{H} CF^{\ast}(C) \) such that \( \mathcal{OC}(u)=1 \). Take \( s\in SC^{\ast}(X) \). Then, since \( \mathcal{OC} \) is a map of \( SC^{\ast}(X) \)-mod­ules with the mod­ule struc­ture on \( \operatorname{H} CF^{\ast}(C) \) in­duced by \( \mathcal{CO} \) fol­lowed by the nat­ur­al ac­tion of \( \operatorname{H}^{\prime} CF^{\ast}(C) \) on \( \operatorname{H} CF^{\ast}(C) \), we find \[ \mathcal{OC} ( \mathcal{CO} (s) \cdot u )= s\cdot \mathcal{OC}(u)=s\cdot 1=s. \] It fol­lows that \( \mathcal{CO} \) is in­ject­ive.

To show that \( \mathcal{CO} \) is sur­ject­ive, note that \( \mathcal{CO}(1) \) counts curves with a pos­it­ive punc­ture at the min­im­um on \( X \). The only such curves cor­res­pond to flow lines start­ing at the min­im­um and end­ing at a fixed point (the mid­point, say) of any Reeb chords. It fol­lows that \[ \mathcal{CO}(1) = e =\sum_{\text{Reeb chords } c} c_{\mathrm{pos}}\otimes c_{\mathrm{neg}}. \]

Con­sider the mod­uli spaces \( \mathcal{M}(u;e) \) in­volved in the equa­tion \[ \mathcal{CO}\circ\mathcal{OC}(u) = e. \] The ele­ments in these mod­uli spaces are in­fin­ite-length cyl­in­ders with pos­it­ive punc­tures cor­res­pond­ing to \( u \) at one bound­ary com­pon­ent and one pos­it­ive and one neg­at­ive punc­ture along the oth­er bound­ary com­pon­ent cor­res­pond­ing to any chord in \( e \). Glu­ing at the middle or­bit we gain one di­men­sion since the mark­er in the middle dis­ap­pears. This means that the res­ult­ing mod­uli space has di­men­sion two.

Con­sider a cycle \( r\in \operatorname{H}^{\prime} CF^{\ast}(C) \) and a non­trivi­al con­tri­bu­tion to \( r\cdot u \). We use this con­tri­bu­tion to de­gen­er­ate the mod­uli spaces in \( \mathcal{M}(u;e) \). We de­gen­er­ate the an­nu­lus in­to two disks \( D_{\mathrm{up}} \) with pos­it­ive punc­tures ac­cord­ing to the pos­it­ive punc­tures in \( r \) and the pos­it­ive punc­ture in the \( e \)-term cor­res­pond­ing to the neg­at­ive punc­ture of \( r \), and two neg­at­ive punc­tures, and a disk \( D_{\mathrm{dn}} \) in the lower part with pos­it­ive punc­tures at the chords in \( r\cdot u \) that are not the neg­at­ive chord of \( r \). Un­der this de­form­a­tion the mod­uli space un­der­goes ad­di­tion­al split­tings that will only af­fect the res­ult up to ex­act terms, us­ing the as­sump­tions that \( r \) and \( u \) are cycles.

Figure 8.  Degeneration of the composition \( \mathcal{CO}\circ\mathcal{OC}(u) \) according to \( r\in \operatorname{H}^{\prime}CW^{\ast}(C) \) and gluing by the rotation cobordism.

We next glue \( D_{\mathrm{up}} \) and \( D_{\mathrm{dn}} \) via a cobor­d­ism cor­res­pond­ing to an iso­topy that in­ter­changes \( \partial C_0 \) and \( \partial C_1 \). The Lag­rangi­an cobor­d­ism \( T\subset \mathbb{R}\times\partial X \) cor­res­ponds to the Le­gendri­an iso­topy in \( \partial X \) that ro­tates \( \partial C_0 \) to \( \partial C_1 \), and then takes the shift­ing func­tion to its neg­at­ive. It is easy to see that if \( C_0 \) and \( C_1 \) are suf­fi­ciently close and the ro­ta­tion is suf­fi­ciently slow, all ri­gid holo­morph­ic curves in the cobor­d­ism cor­res­ponds to (re­para­met­er­ized) trivi­al Reeb chord strips of \( c \) and strips (cor­res­pond­ing to Morse flow lines) that in­ter­changes the small Reeb chords that fol­lows the ro­ta­tion. Let \( \mathcal{M}(T) \) de­note the mod­uli space of ri­gid holo­morph­ic curves with bound­ary on \( T \).

Not­ing that the curves in \( \mathcal{M}(T) \) are strips that ex­actly give the nat­ur­al 1-1 cor­res­pond­ence between Reeb chords \( C_0\to C_1 \) and \( C_1\to C_0 \), we find that the res­ult of glu­ing \( D_{\mathrm{dn}} \) via \( \mathcal{M}(T) \) to \( D_{\mathrm{up}} \) are new an­nuli of di­men­sion 2 that have pos­it­ive punc­tures ac­cord­ing to \( r\cdot u \) along one bound­ary com­pon­ent and pos­it­ive and one neg­at­ive punc­ture ac­cord­ing to \( r \). Then, de­form­ing the Lag­rangi­ans to the stand­ard \( C_0\cup C_1 \) and fol­low­ing the do­mains un­til it splits at a Reeb or­bit, we find that \[ \mathcal{CO}(\mathcal{OC}(r\cdot u))=r, \] up to ex­act terms; see Fig­ure 8. It fol­lows in par­tic­u­lar that \( \mathcal{CO} \) is sur­ject­ive. The the­or­em fol­lows. □

4.4.  The symplectic homology product and Calabi–Yau structures
In this sec­tion we con­sider the product on \( SC^{\ast}(X) \). The iso­morph­ism \( \Phi^{SC}\colon SC^{\ast}(X)\to \operatorname{H} CE^{\ast}(\Lambda) \) al­lows us to push the pair-of-pants product \( \mu \) on \( SC^{\ast}(X) \) to the Hoch­schild com­plex \( \operatorname{H} CE^{\ast}(\Lambda) \). This is re­lated to a du­al­ity Calabi–Yau struc­ture on \( CE^{\ast}(\Lambda) \).

In or­der to com­pute the product we con­sider the com­pos­i­tion \[ \Phi^{SC}\circ \mu\colon SC^{\ast}(X)\otimes SC^{\ast}(X)\to \operatorname{H} CE^{\ast}(\Lambda). \] The curves con­trib­ut­ing to this com­pos­i­tion are disks with two pos­it­ive punc­tures near the ori­gin and neg­at­ive bound­ary punc­tures and one ex­tra punc­ture op­pos­ite the dis­tin­guished mixed punc­ture map­ping to \( z \). As in the proof of The­or­em 4.8, at the glu­ing we gain one di­men­sion from a dis­ap­pear­ing con­form­al con­straint. We use it to con­trol the con­form­al struc­ture of the do­main. Study­ing split­tings we find the fol­low­ing; see [2].

The­or­em 4.9: The sym­plect­ic co­homo­logy product is rep­res­en­ted on \( \operatorname{H} CE^{\ast}(\Lambda) \) by at­tach­ing disks with two pos­it­ive punc­tures and aux­il­i­ary neg­at­ive punc­tures to pairs of cyc­lic words of chords. Here the disks are of two types: disks with one Lag­rangi­an in­ter­sec­tion punc­ture at \( z \) and one pos­it­ive and one neg­at­ive punc­ture at the same Reeb chord \( c \) for any \( c \); and two level disks: a tri­angle fol­lowed by a disk with two pos­it­ive punc­tures; see ([2], Figure 1). □
4.5.  Partial wrapping and coefficients in chains of the based loop space
Start­ing from the study of at­tach­ments of punc­tured handles along knot conor­mals in [e3], a more sys­tem­at­ic study of Le­gendri­an sur­gery for par­tially wrapped Flo­er co­homo­logy was un­der­taken in [e8], where it was con­jec­tured that the par­tially wrapped Flo­er co­homo­logy of a link­ing disk at a Le­gendri­an stop is iso­morph­ic to the Chekan­ov–Eli­ash­berg dg-al­gebra of the Le­gendri­an stop with coef­fi­cients in chains on its based loop space. This res­ult was later proved ([e6], The­or­em 1.2) by trans­lat­ing everything to the sur­gery lan­guage.

This trans­la­tion shows that any con­struc­tion in ex­act Flo­er the­ory ad­mits a sur­gery de­scrip­tion and gives, for ex­ample, a rather dir­ect and geo­met­ric proof of the cut-and-paste meth­ods of [e9]; see ([e6], The­or­em 1.3) and ([e7], The­or­em 1.1). Here we will il­lus­trate the idea by giv­ing the defin­i­tion of the sur­gery de­scrip­tion of par­tially wrapped Flo­er co­homo­logy and a sample res­ult that can be proved with this tech­no­logy.

Con­sider Wein­stein do­mains \( W \) of di­men­sion \( 2n \) and \( V \) of di­men­sion \( 2n-2 \). We fix a Lag­rangi­an skel­et­on \( L \) with a fixed handle struc­ture for \( V \) and think of \( V \) as a co­tan­gent bundle of \( L \). We take a Le­gendri­an em­bed­ding of \( L \) in­to \( \partial W \) to be a con­tact em­bed­ding of a neigh­bor­hood \( V\times (-\epsilon,\epsilon) \) of the zero sec­tion in the con­tact­iz­a­tion of \( V \). Giv­en such an em­bed­ding, con­sider \( W \) with the fixed em­bed­ding of \( V\times (-\epsilon,\epsilon) \) in \( \partial W \) and also the neg­at­ive half \( [0,-\infty)\times\mathbb{R}\times V \) of the sym­plect­iz­a­tion of \( \mathbb{R}\times V \) with \( V\times (-\epsilon,\epsilon)\times V \) em­bed­ded in \( \{0\}\times \mathbb{R}\times V \). We at­tach a \( V \)-handle, \( T^{\ast} I\times V \), to the uni­on. Note that the handles of \( V \) give at­tach­ing spheres for \( X \). We define the Chekan­ov–Eli­ash­berg dg-al­gebra of \( L \) to be the Chekan­ov–Eli­ash­berg dg-al­gebra of the Le­gendri­an at­tach­ing spheres of \( X \). Note that it is gen­er­ated by Reeb chords of \( L \) to­geth­er with Reeb chords of the Le­gendri­an at­tach­ing spheres in \( V \) in the middle of the handle.

In case \( L \) is smooth, ob­serve that the in­tern­al Reeb chords of \( L \) give a dg-al­gebra quasii­so­morph­ic to the chains on the based loop space of \( L \). Us­ing this ob­ser­va­tion one can show that the dg-al­gebra of \( L \) is then iso­morph­ic to the dg-al­gebra with loop space coef­fi­cients; see ([e6], The­or­em 1.2).

We give one il­lus­tra­tion of how this con­struc­tion can be used. Con­sider a smooth man­i­fold \( M \) and fix a Morse func­tion \( f\colon M\to\mathbb{R} \) with one max­im­um and one min­im­um. It is well known that the wrapped Flo­er co­homo­logy of the co­tan­gent fiber in \( T^{\ast} M \) is quasii­so­morph­ic to chains on the based loop space of \( M \), \[ CW^{\ast}(T_{m}M)\approx C_{\ast}(\Omega M). \] For the pur­poses of this dis­cus­sion, we think of this res­ult as not cut­ting \( M \) at all, or to com­ply with what comes next, as cut­ting \( M \) be­low the min­im­um of the Morse func­tion. We can also cut \( M \) just be­low the max­im­um. If \( M_{\le a} \) is the sub­level set then the co­tan­gent bundle \( T^{\ast}M_{\le a} \), with corners roun­ded, is a sub­crit­ic­al Wein­stein do­main with the level sphere \( \Lambda_{a}=\partial M_{\le a} \) as a Le­gendri­an sub­man­i­fold. By the sur­gery iso­morph­ism, The­or­em 3.7, \( \,CE^{\ast}(\Lambda_{a}) \) is also iso­morph­ic to \( CW^{\ast}(T_{m} M) \). It turns out the same holds true if we cut at any reg­u­lar level \( b \). More pre­cisely, we get a sub­crit­ic­al Wein­stein do­main \( T^{\ast}M_{\le b} \) with a Le­gendri­an level sur­face \( \Lambda_{b}=f^{-1}(b) \). Con­sider now the Chekan­ov–Eli­ash­berg dg-al­gebra of \( \Lambda_{b} \) with coef­fi­cients in chains of the based loop space of the su­per­level set \( M_{\ge b} \), \[ CE^{\ast}(\Lambda_{b};C_{\ast}(\Omega(M_{\ge b}))). \] Then, in fact, \begin{equation} CE^{\ast}(\Lambda_{b};C_{\ast}(\Omega(M_{\ge b}))) \approx CW^{\ast}(T_{m}^{\ast}M), \tag{4.1} \end{equation} for every reg­u­lar value \( b \). The iso­morph­isms in Equa­tion (4.1) il­lus­trate the fact that in the ex­act case, Flo­er co­homo­logy can be ex­pressed either in terms of to­po­logy or Reeb dy­nam­ics, and fur­ther that it is pos­sible to in­ter­pol­ate between these ex­treme cases by hav­ing a part in to­po­logy and an­oth­er in Reeb dy­nam­ics.

The iso­morph­isms in this ex­ample are straight­for­ward to de­rive, work­ing in Le­gendri­an sur­gery present­a­tions. We de­scribe them here as an il­lus­tra­tion of the great gen­er­al­ity and flex­ib­il­ity of Eli­ash­berg’s ori­gin­al sur­gery ap­proach to holo­morph­ic curve the­or­ies in Wein­stein do­mains. It can handle very gen­er­al ob­jects, and provides, for ex­ample, Chekan­ov–Eli­ash­berg dg-al­gebra or Flo­er co­homo­logy for Le­gendri­ans with bound­ary with corners with coef­fi­cients in chains on the based loop space, in terms of the usu­al dg-al­geb­ras of Le­gendri­an at­tach­ing spheres in suit­able Wein­stein man­i­folds. Such sin­gu­lar Lag­rangi­ans and Le­gendri­ans are very use­ful for in­stance as cen­ters of mir­ror sym­metry charts; see [e5] for a simple ex­ample.

To­bi­as Ek­holm is a pro­fess­or of math­em­at­ics at Uppsala Uni­versity and Dir­ect­or of In­sti­tut Mit­tag-Leffler.

Works

[1] Y. Eli­ash­berg, A. Givent­al, and H. Hofer: “In­tro­duc­tion to sym­plect­ic field the­ory,” pp. 560–​673 in Vis­ions in Math­em­at­ics. Edi­ted by N. Alon, J. Bour­gain, A. Connes, M. Gro­mov, and V. Mil­man. 2000. Spe­cial volume, GA­FA2000, of Geo­met­ric and Func­tion­al Ana­lys­is. MR 1826267 Zbl 0989.​81114 incollection

[2] F. Bour­geois, T. Ek­holm, and Y. Eli­ash­berg: “Sym­plect­ic ho­mo­logy product via Le­gendri­an sur­gery,” Proc. Natl. Acad. Sci. USA 108 : 20 (2011), pp. 8114–​8121. MR 2806647 Zbl 1256.​53049 article

[3] F. Bour­geois, T. Ek­holm, and Y. Eli­ash­berg: “Ef­fect of Le­gendri­an sur­gery,” Geom. To­pol. 16 : 1 (2012), pp. 301–​389. With an ap­pendix by Sheel Gan­atra and Mak­sim May­danskiy. MR 2916289 article