return

Celebratio Mathematica

Walter D. Neumann

Splice diagrams and splice-quotient surface singularities

by Jonathan Wahl

It has been my great pleas­ure to col­lab­or­ate on many pa­pers with Wal­ter Neu­mann, a ter­rif­ic math­em­atician and friend. I’ll try to present an over­all ac­count of some of our work, primar­ily on splice dia­grams and splice-quo­tient sin­gu­lar­it­ies. Our goal was to be able to write down ex­pli­cit equa­tions of cer­tain kinds of nor­mal sur­face sin­gu­lar­it­ies, giv­en only the to­po­logy of their links. I men­tion our mo­tiv­a­tions, and why and how we were led to cer­tain kinds of ques­tions, leav­ing de­tailed proofs (and care­ful defin­i­tions) to the ori­gin­al pa­pers. One can also con­sult in­ter­est­ing more re­cent work of oth­ers, such as [e13], which uses a trop­ic­al geo­metry ap­proach.

1. Provocative early work of Walter

While I first be­came friendly with Wal­ter in 1972 at the In­sti­tute for Ad­vanced Study, our col­lab­or­a­tion did not be­gin un­til 1987, well after he had writ­ten two pa­pers about nor­mal sur­face sin­gu­lar­it­ies which I found very pro­voc­at­ive.

Re­call that if \( (X,0) \) is a com­plex nor­mal sur­face sin­gu­lar­ity, its link \( \Sigma \) is the ori­ented 3-man­i­fold which is its neigh­bor­hood bound­ary, i.e., \( X \) in­ter­sec­ted with a small sphere about 0. From the min­im­al good res­ol­u­tion \( \pi:(\tilde{X},E)\rightarrow (X,0) \), where \( \pi^{-1}(0)=E \) is a uni­on of smooth curves with strong nor­mal cross­ings, \( \Sigma \) is the bound­ary of a tu­bu­lar neigh­bor­hood of \( E \). One as­so­ci­ates to E a plumb­ing graph \( \Gamma \), en­cod­ing the gen­era, self-in­ter­sec­tions, and in­ter­sec­tions of the curves. From \( \Gamma \), one can re­con­struct \( \Sigma \) by the plumb­ing con­struc­tion, as made clear in sem­in­al pa­pers [e1] and [e2] by Mum­ford and Hirzebruch. Thus the plumb­ing graph \( \Gamma \) de­term­ines the to­po­logy of \( \Sigma \). Wal­ter’s ob­vi­ously im­port­ant the­or­em proves that not only does \( \Gamma \) de­term­ine the 3-man­i­fold \( \Sigma \), but the con­verse is true as well.

The­or­em 1.1: [1] Sup­pose \( \Sigma \) is a sin­gu­lar­ity link. Then \( \Sigma \) de­term­ines the min­im­al good res­ol­u­tion graph \( \Gamma \).

In fact, with the ex­cep­tion of lens spaces (from cyc­lic quo­tient sin­gu­lar­it­ies) and some two-tor­us bundles over the circle (from cusp sin­gu­lar­it­ies), already \( \pi_1(\Sigma) \) de­term­ines \( \Gamma \).

The the­or­em was proved by a cal­cu­lus for gen­er­al plumb­ing dia­grams (not ne­ces­sar­ily neg­at­ive-def­in­ite), and im­plied oth­er ba­sic res­ults (e.g., \( S^1\times S^2 \) is not a sin­gu­lar­ity link). Thus, any in­vari­ant con­cocted from \( \Gamma \) is to­po­lo­gic­al, and not just a sin­gu­lar­ity in­vari­ant. On the oth­er hand, the geo­met­ric genus and Mil­nor num­ber de­pend in gen­er­al on ana­lyt­ic in­form­a­tion.

Figure 1.

A later res­ult of Wal­ter’s in­volved cer­tain sin­gu­lar­it­ies with a ra­tion­al ho­mo­logy sphere (or \( \mathbb{Q} \)HS) link; that is, \( E \) is a tree of smooth ra­tion­al curves. Then \( H_1(\Sigma; \mathbb{Z})\equiv H_1 \) is the fi­nite dis­crim­in­ant group \( D(\Gamma) \), cal­cu­lated for ex­ample as the coker­nel of the in­ter­sec­tion mat­rix \( (E_i\cdot E_j) \); thus, \( \Sigma \) has a fi­nite uni­ver­sal abeli­an cov­er­ing (or UAC) \( \Sigma^{\prime}\rightarrow \Sigma \). That cov­er­ing is real­ized by a fi­nite map of nor­mal germs \( (X^{\prime},0)\rightarrow (X,0) \), which is a quo­tient by an ac­tion of the group \( H_1 \); this is the UAC of \( (X,0) \).

A weighted-ho­mo­gen­eous sin­gu­lar­ity with \( \mathbb{Q} \)HS link has res­ol­u­tion graph shown in Fig­ure 1.

Figure 2.

Here, the con­tin­ued frac­tion ex­pan­sion \( n/q=b_1-1/b_2-\cdots -1/b_s \) rep­res­ents a string of ra­tion­al curves em­an­at­ing from the cen­ter; see Fig­ure 2.

(We shall as­sume that \( t\geq 3 \).) The ana­lyt­ic type of \( (X,0) \) is uniquely de­term­ined by the graph plus the loc­a­tion of the \( t \) in­ter­sec­tion points on the cent­ral curve; in fact, one can write \( X=\operatorname{Spec}\bigoplus_{i=0}^{\infty} A_i \), where each \( A_i \) is com­put­able as a co­homo­logy group from data on the cent­ral \( \mathbb P^1 \) [e5], The­or­em 5.1. Still, equa­tions de­fin­ing the sin­gu­lar­ity might be quite com­plic­ated, where­as the UAC is simple.

The­or­em 1.2: [2] With nota­tion as above, the UAC of the weighted-ho­mo­gen­eous sin­gu­lar­ity \( (X,0) \) is a Brieskorn–Pham com­plete in­ter­sec­tion \( (X^{\prime},0)\subset (\mathbb{C}^t,0) \), defined by equa­tions \[ \sum_{j=1}^t a_{ij}z_j^{n_j}=0,\quad i=1,\cdots, t-2. \] In ad­di­tion, the dis­crim­in­ant group \( H_1 \) acts on \( X^{\prime} \) di­ag­on­ally on the \( t \) co­ordin­ates \( z_j \).

The mat­rix \( (a_{ij}) \) in the The­or­em sat­is­fies the Hamm con­di­tion [e4], that all max­im­al minors are nonzero.

I found this a very sur­pris­ing res­ult! It is easy to show that \( (X^{\prime},0) \) is Goren­stein; but why is it a com­plete in­ter­sec­tion? And, while its de­fin­ing equa­tions must be weighted-ho­mo­gen­eous, why are they of the very spe­cial Brieskorn–Pham type? Wal­ter’s proof was simply to write down these equa­tions, find an ap­pro­pri­ate rep­res­ent­a­tion of the dis­crim­in­ant group \( H_1 \) on \( \mathbb{C}^t \), and then to prove that (for ju­di­ciously chosen \( a_{ij} \)) the quo­tient is the sin­gu­lar­ity we star­ted with.

At­tempts to find a nat­ur­al gen­er­al state­ment which in­cludes this the­or­em led years later to our defin­i­tion of splice-quo­tient sin­gu­lar­it­ies in [6] (see Sec­tion 5).

2. Casson invariant and splice diagrams

In 1987 I at­ten­ded a lec­ture giv­en by Sir Mi­chael Atiyah at the Weyl Centen­ni­al Con­fer­ence at Duke Uni­versity, en­titled “New in­vari­ants of three and four-man­i­folds” [e6]. An­drew Cas­son had dis­covered a new in­teger in­vari­ant \( \lambda(\Sigma) \) for a com­pact in­teg­ral ho­mo­logy 3-sphere (or \( \mathbb{Z} \)HS); its mod 2 re­duc­tion is the Rokh­lin in­vari­ant (see, e.g., [e7], The­or­em 12.1). Nat­ur­ally I asked about its value for the link \( \Sigma(p,q,r) \) of the Brieskorn sin­gu­lar­ity \( x^p+y^q+z^r=0 \), which is a \( \mathbb{Z} \)HS when \( p, q, r \) are pair­wise re­l­at­ively prime. Atiyah answered that based on de­vel­op­ing work by Fin­tushelStern, \( \lambda \) was a com­plic­ated ex­pres­sion in­volving the num­ber of in­teger points in some three-di­men­sion­al re­gion. I guessed it was re­lated to the sin­gu­lar­ity it­self; a call to Wal­ter com­pletely cla­ri­fied the situ­ation.

First, \( \lambda(\Sigma) \) is minus one-half the “num­ber” of equi­val­ence classes of \( \operatorname{SU}(2) \) rep­res­ent­a­tions of \( \pi_1(\Sigma) \), where num­ber is an al­geb­ra­ic count done us­ing a Hee­gaard de­com­pos­i­tion of \( \Sigma \). Wal­ter ex­plained that L. Green­berg had found an ac­tu­al count of such rep­res­ent­a­tions, giv­en by the num­ber of in­teger points in some re­gion, and Fin­tushel–Stern were prov­ing that \( \lambda(\Sigma(p,q,r)) \) in­volves simply the ac­tu­al count. But then we real­ized that a for­mula of Brieskorn [e3] im­plied that the cor­res­pond­ing ex­pres­sion for \( \lambda \) was equal to \( 1/8 \) the sig­na­ture of the in­ter­sec­tion pair­ing on the second ho­mo­logy of the Mil­nor fiber of the sin­gu­lar­ity. (As the pair­ing is even, nonde­gen­er­ate, and un­im­od­u­lar, a ba­sic res­ult im­plies that the sig­na­ture is di­vis­ible by 8.) Put­ting to­geth­er work of oth­ers, we had real­ized a beau­ti­ful and pro­voc­at­ive ex­pres­sion for the Cas­son in­vari­ant of these par­tic­u­lar links. Our pa­per [4] con­tained the fol­low­ing con­jec­ture, ap­peal­ing yet wildly op­tim­ist­ic.

Cas­son In­vari­ant Con­jec­ture: (CIC) Sup­pose \( (X,0) \) is the link of a hy­per­sur­face (or com­plete in­ter­sec­tion) sin­gu­lar­ity whose link \( \Sigma \) is an in­teg­ral ho­mo­logy sphere. Then the Cas­son in­vari­ant of \( \Sigma \) equals one-eighth the sig­na­ture of the Mil­nor fiber.
The­or­em 2.1: [4] The Cas­son In­vari­ant Con­jec­ture is true in the fol­low­ing cases:
  1. The Brieskorn–Pham com­plete in­ter­sec­tion \( \sum_{j=1}^t a_{ij}z_j^{n_j}=0,\,i=1,\cdots, t-2 \), with pair­wise re­l­at­ively prime \( n_i \).
  2. The sin­gu­lar germ \( z^n+f(x,y)=0 \), where \( f(x,y)=0 \) is an ir­re­du­cible plane curve sin­gu­lar­ity and \( n \) is re­l­at­ively prime to all the num­bers among the Puiseux pairs.
  3. The com­plete in­ter­sec­tion sin­gu­lar­ity in \( \mathbb{C}^4 \) defined by \[ x^n=u^{n+1}+v^ny,\quad y^n=v^{n+1}+u^nx. \]

The meth­od was cor­rect but math­em­at­ic­ally dis­ap­point­ing: we com­pute and com­pare the Cas­son in­vari­ant and sig­na­ture in each case. Wal­ter real­ized that the key was the use of splice dia­grams, in­tro­duced earli­er by Sieben­mann, and ex­tens­ively stud­ied by Eis­en­budNeu­mann [3]. Spli­cing refers to a dif­fer­ent to­po­lo­gic­al con­struc­tion than plumb­ing, al­low­ing one to make a new \( \mathbb{Z} \)HS from two old ones. Spe­cific­ally, if \( (\Sigma_i, K_i) \) is a pair of a ho­mo­logy 3-sphere and a knot, one re­moves the in­teri­ors of closed tu­bu­lar neigh­bor­hoods \( K_1\times D^2 \) of \( K_1 \) and \( D^2\times K_2 \) of \( K_2 \), and pastes the \( \Sigma_i \) to­geth­er along the tor­us bound­ar­ies \( K_1\times S^1 \) and \( S^1\times K_2 \) — in oth­er words, switch­ing the roles of me­ridi­an and lon­git­ude (and pay­ing at­ten­tion to ori­ent­a­tions!). This pro­cess yields a new ho­mo­logy sphere, which in our situ­ation can be rep­res­en­ted by a splice dia­gram.

Figure 3.

A splice dia­gram \( \Delta \) is a fi­nite tree with ver­tices only of valency 1 (leaves) or \( \ge3 \) (nodes) and with a col­lec­tion of in­teger weights at each node, as­so­ci­ated to the edges de­part­ing the node. The Brieskorn link \( \Sigma(p,q,r) \) is rep­res­en­ted by the dia­gram shown in Fig­ure 3.

The next ex­ample is the dia­gram from spli­cing \( \Sigma(2,3,7) \) with \( \Sigma(2,5,11) \) along the knots giv­en by the last co­ordin­ates, shown in Fig­ure 4.

Figure 4.

For an edge con­nect­ing two nodes in a splice dia­gram the edge de­term­in­ant is the product of the two weights on the edge minus the product of the weights ad­ja­cent to the edge. In the above ex­ample, the one edge con­nect­ing two nodes has edge de­term­in­ant \( 77-60=17 \).

The­or­em 2.2: [3] The ho­mo­logy spheres that are sin­gu­lar­ity links are in one-to-one cor­res­pond­ence with splice dia­grams sat­is­fy­ing these con­di­tions:
  • The weights around a node are pos­it­ive and pair­wise coprime.
  • The weight on an edge end­ing in a leaf is great­er than 1.
  • All edge de­term­in­ants are pos­it­ive.

From the res­ol­u­tion dia­gram for a ho­mo­logy sphere sin­gu­lar­ity link, one as­so­ci­ates a splice dia­gram as fol­lows: dis­reg­ard all ver­tices of valency 2; at each node and out­go­ing edge, take the ab­so­lute value of the de­term­in­ant of the out­er dia­gram. For ex­ample, the splice dia­gram above arises from the res­ol­u­tion dia­gram shown in Fig­ure 5.

Figure 5.

There are meth­ods to com­pute the res­ol­u­tion dia­gram from the splice dia­gram; see a gen­er­al dis­cus­sion of plumb­ing versus spli­cing in the Ap­pendix to [6].

One reas­on to con­sider splice dia­grams for the CIC is that the Cas­son in­vari­ant is known to be ad­dit­ive un­der spli­cing. All \( \mathbb{Z} \)HS sin­gu­lar­ity links ad­mit splice dia­grams built on the \( \Sigma(p,q,r) \) and more gen­er­al Brieskorn–Pham com­plete in­ter­sec­tions, so pro­du­cing these dia­grams for the ex­amples in The­or­em 2.1 al­lows a dir­ect cal­cu­la­tion of the Cas­son in­vari­ants. The sig­na­tures in these cases, however, are gen­er­ally ex­tremely dif­fi­cult to com­pute, and we had to use sev­er­al dif­fer­ent meth­ods.

One con­sequence of the CIC would be that for a com­plete in­ter­sec­tion sin­gu­lar­ity with \( \mathbb{Z} \)HS link, the to­po­logy of the link de­term­ines the geo­met­ric genus. (We fool­ishly guessed ori­gin­ally that we could re­place “com­plete in­ter­sec­tion” by “Goren­stein”, but [e8] found a counter­example.) Such a sur­pris­ing res­ult should give one pause be­fore be­ing too con­fid­ent of its cor­rect­ness. On the oth­er hand, the Cas­son in­vari­ant of a \( \mathbb{Z} \)HS \( \Sigma \) is one-eighth the sig­na­ture of some simply con­nec­ted spin-man­i­fold whose bound­ary is \( \Sigma \); the CIC as­serts that the Mil­nor fiber is such a man­i­fold.

3. Singularities of splice type

After the com­ple­tion in 1989 of our pa­per on the Cas­son in­vari­ant, Wal­ter (and I) pur­sued oth­er top­ics in the 1990s, he work­ing primar­ily on hy­per­bol­ic 3-man­i­folds and geo­met­ric group the­ory. But in 1998 Wal­ter spent a semester at Duke Uni­versity, at which time we re­sumed our dis­cus­sion of two big ques­tions (bey­ond the CIC) left open for us from [4].

The first was the dearth of ex­pli­cit ex­amples of hy­per­sur­face and com­plete in­ter­sec­tion sin­gu­lar­it­ies with \( \mathbb{Z} \)HS links (so, the in­ter­sec­tion mat­rix \( (E_i\cdot E_j) \) has de­term­in­ant \( \pm 1 \)). Could one write down equa­tions of such sin­gu­lar­it­ies for at least some splice dia­grams from The­or­em 2.2? The com­plete in­ter­sec­tion ex­amples of The­or­em 2.1(3) gave a hint; we were lucky that Henry Laufer had provided equa­tions for us from the cor­res­pond­ing res­ol­u­tion graph in the simplest case \( n=2 \).

Second, since the Cas­son in­vari­ant adds un­der spli­cing, per­haps the CIC is true be­cause one can “splice” sin­gu­lar­it­ies and their Mil­nor fibers, match­ing spli­cing on the link level, and so that on the Mil­nor fiber level the sig­na­tures add. Both of these is­sues were even­tu­ally dis­cussed in [6] and [7].

A splice dia­gram with a single node and \( t \) leaves arises from a Brieskorn–Pham com­plete in­ter­sec­tion in \( \mathbb{C}^{t} \) as in The­or­em 2.1(1). A vari­able is as­signed to each of the leaves, and one defines a sin­gu­lar­ity by \( t-2 \) gen­er­ic lin­ear com­bin­a­tions of monomi­als from each leaf. Weighted-ho­mo­gen­eity is giv­en by as­sign­ing to each leaf a weight which is the product of all the oth­er weights from the oth­er edges. In ad­di­tion, note that adding to the equa­tions terms of weight great­er than the weight \( n_1n_2\cdots n_t \) gives oth­er sin­gu­lar­it­ies with the same res­ol­u­tion and splice dia­grams.

Figure 6.

Con­sider now the two-node splice dia­gram in Fig­ure 6, where we have already ad­ded a vari­able to each leaf.

Re­call that \( p,q,p^{\prime},q^{\prime} \) are \( \geq 2 \), the triples \( p,q,r \) and \( p^{\prime},q^{\prime},r^{\prime} \) are each pair­wise re­l­at­ively prime, and \( rr^{\prime} > pqp^{\prime}q^{\prime} \).

Defin­i­tion 3.1: The dia­gram sat­is­fies the semig­roup con­di­tions if \( r^{\prime}\in \mathbb{N}(p,q) \) (the semig­roup gen­er­ated by \( p \) and \( q \)) and \( r\in \mathbb{N}(p^{\prime},q^{\prime}) \).

Since \( r^{\prime}\geq (p-1)(q-1) \) im­plies that \( r^{\prime}\in \mathbb{N}(p,q) \), the edge de­term­in­ant con­di­tion does guar­an­tee that at least one of \( r,r^{\prime} \) is in the ap­pro­pri­ate semig­roup.

The­or­em 3.2: For the splice dia­gram above, as­sume the semig­roup con­di­tions are sat­is­fied, and write \[ r^{\prime}=\alpha p+\beta q, \quad r=\gamma p^{\prime}+\delta q^{\prime}. \] Then the isol­ated com­plete in­ter­sec­tion sin­gu­lar­ity giv­en by \begin{equation*} \eqalign{ X^p+Y^q+Z^{\delta}W^{\gamma}&=0,\cr Z^{p^{\prime}}+W^{q^{\prime}}+X^{\beta}Y^{\alpha}&=0} \end{equation*} has \( \mathbb{Z} \)HS link with the giv­en splice dia­gram.

In ana­logy with the one-node case, we have first as­signed a vari­able to each leaf. Next, to each node we as­so­ci­ate a weight which is the product of the weights of the sur­round­ing edges (so, \( pqr \) for the left node). Third, fix­ing a node we as­sign a weight to each leaf, by tak­ing the product of the weights ad­ja­cent to (but not on) the route from the node to that leaf; for the left node, this weight for the up­per left leaf is \( qr \), while the weight for the up­per right leaf is \( pqq^{\prime} \). Cru­cially, the semig­roup con­di­tion in­volving \( r \) al­lows one to write the first equa­tion, the sum of three monomi­als for the three edges of the node, each of which is ho­mo­gen­eous with re­spect to these weights, of total de­gree \( pqr \).

Thus, with the weights from the left node, the first equa­tion is weighted ho­mo­gen­eous, while the second is not. Still, the as­so­ci­ated graded is an in­teg­ral do­main, and its nor­mal­iz­a­tion is the sin­gu­lar­ity \( X^p+Y^q+T^r=0 \). One starts to re­solve the sin­gu­lar­ity with a weighted blow-up of \( \mathbb{C}^4 \). A key point is that set­ting the vari­able \( Z \) or \( W \) equal to 0 gives a monomi­al curve, cor­res­pond­ing to the “\( r \)” knot in \( \Sigma(p,q,r) \). But adding high­er weight terms to the equa­tions would not af­fect the to­po­logy.

One can sim­il­arly write down a sys­tem of equa­tions for an ar­bit­rary splice dia­gram \( \Delta \), as long as ap­pro­pri­ate semig­roup con­di­tions are as­sumed. Start by as­sign­ing a vari­able \( z_w \) to every leaf \( w \). Next, for each node \( v \), take as weight \( d_v \) the product of the weights \( d_{ve} \) on the \( t \) sur­round­ing edges. To every leaf \( w \), define its \( v \)-weight to be \( l_{vw} \), the product of all weights ad­ja­cent to (but not on) the path from \( v \) to \( w \); and define \( l^{\prime}_{vw} \) to be the same product ex­clud­ing the weights around \( v \). The semig­roup con­di­tion means: the weight on any edge \( e \) is in the semig­roup gen­er­ated by the \( l^{\prime}_{vw} \) of all the out­er leaves on the \( e \)-side of \( v \). Writ­ing \( d_{ve}=\Sigma_w\alpha_{vw}l^{\prime}_{vw} \), it fol­lows that the ad­miss­ible monomi­al \( M_{ve}=\prod_w z_w^{\alpha_{vw}} \) has weight \( d_v \), the weight of the node. Now take \( t-2 \) lin­ear com­bin­a­tions of those \( t \) monomi­als so that the coef­fi­cient mat­rix \( (a_{vie}) \) has all max­im­al minors nonzero, and add to each such lin­ear com­bin­a­tion a term \( H_{vi} \) of weight \( > d_v \): \[ \Sigma_ea_{vie}M_{ve}+H_{vi}=0, \quad i=1,\cdots, t-2. \] Fi­nally, choose such equa­tions for every node \( v \) of \( \Delta \). The fol­low­ing is a spe­cial case of The­or­em 5.4 be­low, proved ori­gin­ally in [6].

The­or­em 3.3: For any splice dia­gram \( \Delta \) sat­is­fy­ing the semig­roup con­di­tions, the equa­tions above define an isol­ated com­plete in­ter­sec­tion sin­gu­lar­ity whose link is the \( \mathbb{Z} \)HS cor­res­pond­ing to that splice dia­gram.

One calls these sin­gu­lar­it­ies of splice type. The proof is by in­duc­tion on the num­ber of nodes; the key step starts with an “end node” (all but one of whose em­an­at­ing edges is a leaf), takes a weighted blow-up, and uses in­duct­ively the res­ult for a sub­dia­gram. Again, set­ting a vari­able equal to 0 gives a monomi­al curve cor­res­pond­ing to a knot in the link.

How gen­er­al are sin­gu­lar­it­ies of splice type among sin­gu­lar­it­ies with \( \mathbb{Z} \)HS link? By an im­port­ant gen­er­al res­ult of Popes­cu-Pam­pu [e12], every splice dia­gram as in The­or­em 2.2 is the link of some Goren­stein sin­gu­lar­ity. One should keep in mind a class of ex­amples in the pa­per [e8] of Némethi, Lu­engo-Velasco, and Melle-Hernan­dez.

Ex­amples:
  1. There ex­ists a Goren­stein sin­gu­lar­ity, not a com­plete in­ter­sec­tion, whose link is the Brieskorn sphere \( \Sigma(2,13,31) \).
  2. There ex­ists a Goren­stein sin­gu­lar­ity, not a com­plete in­ter­sec­tion, whose link is a ho­mo­logy sphere but which does not sat­is­fy the semig­roup con­di­tions.

An ana­lyt­ic con­di­tion dis­tin­guishes sin­gu­lar­it­ies of splice type from oth­er sin­gu­lar­it­ies with the same link. Each leaf of the splice (or res­ol­u­tion) dia­gram gives a knot in \( \Sigma \), unique up to iso­topy. A key point in prov­ing that splice dia­gram equa­tions give in­teg­ral ho­mo­logy sphere links is to show that the vari­able \( z_{i} \) as­so­ci­ated to a leaf cuts out the cor­res­pond­ing knot in \( \Sigma \). In oth­er words, the curve \( C_{i} \) giv­en by \( z_{i}=0 \) is ir­re­du­cible, and its prop­er trans­form \( D_{i} \) on the min­im­al good res­ol­u­tion is smooth and in­ter­sects trans­versely the ex­cep­tion­al curve cor­res­pond­ing to the leaf of the splice dia­gram. Then the ex­ist­ence of such “end-curve func­tions” im­plies the semig­roup con­di­tion on the splice dia­gram.

The­or­em 3.4: [7], Section 8.1. Let \( (X,0) \) be a nor­mal sur­face sin­gu­lar­ity whose link \( \Sigma \) is an in­teg­ral ho­mo­logy sphere. As­sume that for each of the \( t \) leaves \( w_i \) of the splice dia­gram \( \Delta \) of \( \Sigma \), there is a func­tion \( z_{i} \) in­du­cing the end knot as above. Then
  1. \( \Delta \) sat­is­fies the semig­roup con­di­tion;
  2. \( X \) is a com­plete in­ter­sec­tion of em­bed­ding di­men­sion less than or equal to t;
  3. \( z_{1},\cdots,z_{t} \) gen­er­ate the max­im­al ideal of the loc­al ring of \( X \) at 0, and \( X \) is a com­plete in­ter­sec­tion of splice type with re­spect to these gen­er­at­ors.

Thus, some nat­ur­al open ques­tions are as fol­lows:

Ques­tions:
  1. Does there ex­ist a com­plete in­ter­sec­tion sin­gu­lar­ity for the two-node splice dia­gram based on \( (2,3,1; 2,3,37) \), where the semig­roup con­di­tion fails?
  2. Is every com­plete in­ter­sec­tion with \( \mathbb{Z} \)HS link of splice type?
  3. As­sum­ing there are com­plete in­ter­sec­tions with \( \mathbb{Z} \)HS link that are not of splice type, do they sat­is­fy the Cas­son In­vari­ant Con­jec­ture?

4. Casson Invariant and Milnor Fiber Conjectures

What about the CIC for these sin­gu­lar­it­ies of splice type? It was real­ized early on that it is easi­er to re­state it in terms of the geo­met­ric genus \( p_g \) of the sin­gu­lar­ity (as op­posed to the sig­na­ture). The­or­em 2 of [7] proves the con­jec­ture for sin­gu­lar­it­ies for which the nodes of the splice dia­gram lie on one line. The gen­er­al case was proved by Némethi and Ok­uma, us­ing a more power­ful in­duc­tion.

The­or­em 4.1: [e10] The Cas­son In­vari­ant Con­jec­ture is true for sin­gu­lar­it­ies of splice type.

As in­dic­ated pre­vi­ously, the valid­ity of the CIC for sin­gu­lar­it­ies of splice type sug­gests that there is a con­struc­tion of “spli­cing Mil­nor fibers” for which the ad­dit­iv­ity of the sig­na­tures is geo­met­ric­ally clear.

Sup­pose the equa­tions \( f_i(z_1,\dots,z_n)=0 \), \( i=1,\dots,n-2 \) define a sin­gu­lar­ity \( X \) of splice type, cor­res­pond­ing to a splice dia­gram \( \Delta \). Then the curve \( z_j=0 \) cuts out in \( \Sigma \) the knot \( K_j \) cor­res­pond­ing to the \( j \)th leaf of \( \Delta \). The Mil­nor fiber of the sin­gu­lar­ity at 0 of the com­plete in­ter­sec­tion curve \( (f_1,\dots,f_{n-2},z_j)^{-1}(0) \) can be con­sidered as a sur­face \( G_j \) in the Mil­nor fiber of \( X \), with \( \partial G_j=G_j\cap \Sigma=K_j \). We pro­posed in [7], Sec­tion 6, a con­jec­tur­al it­er­at­ive de­scrip­tion of a Mil­nor fiber in terms of the Mil­nor fibers of sim­pler com­plete in­ter­sec­tion sur­face sin­gu­lar­it­ies and fibers \( G_j \) as above ly­ing in­side them.

Let \( \Delta \) be a splice dia­gram sat­is­fy­ing the semig­roup con­di­tions, and write \[ \Sigma=\Sigma_1\frac{K_1\quad K_2}{}\Sigma_2 \] to rep­res­ent two ho­mo­logy spheres de­term­ined by cut­ting \( \Delta \) at an edge to form two dia­grams with dis­tin­guished leaves. Then \( \Delta_1 \) and \( \Delta_2 \) also sat­is­fy the semig­roup con­di­tion, so \( \Sigma_1 \) and \( \Sigma_2 \) are both com­plete in­ter­sec­tion sin­gu­lar­ity links giv­en by equa­tions of splice type. They thus have Mil­nor fibers, say \( F_1 \) and \( F_2 \), with \( \partial F_i=\Sigma_i \).

For the knot \( K_1\subset \Sigma_1 \), con­sider as above \( G_1\subset F_1 \), as well as a tu­bu­lar neigh­bor­hood \( G_1\times D^2\subset F_1 \). Sim­il­arly, con­sider \( D^2\times G_2\subset F_2 \).

De­note \[ F_1^o:=F_1-(G_1\times( D^2)^o),\quad F_2^o:=F_2-( (D^2)^o\times G_2), \] so \( \partial F_1^o \) is the uni­on of \( G_1\times S^1 \) and the ex­ter­i­or (com­ple­ment of an open tu­bu­lar neigh­bor­hood) of the knot \( K_1\subset \Sigma_1 \), and sim­il­arly for \( \partial F_2^o \).

Mil­nor Fiber Con­jec­ture: The Mil­nor fiber \( F \) is homeo­morph­ic to the res­ult \( \overline{F} \) of past­ing: \[ \overline{F}:=F_1^o\cup_{G_1\times S^1} (G_1\times G_2)\cup_{S^1\times G_2}F_2^o, \] where we identi­fy \( G_1\times S^1 \) with \( G_1\times\partial G_2 \) and \( S^1\times G_2 \) with \( \partial G_1\times G_2 \).

The con­struc­tion of \( \overline{F} \) ex­tends spli­cing on the bound­ary links. It is not hard to see that sig­na­tures add, spe­cific­ally \[ \operatorname{sign} \overline{F}=\operatorname{sign}\ F_1+\operatorname{sign}\ F_2. \] Thus the Mil­nor Fiber Con­jec­ture im­plies the Cas­son In­vari­ant Con­jec­ture.

In [7], The­or­em 8.2, the Mil­nor Fiber Con­jec­ture is veri­fied for the pre­vi­ously dis­cussed sin­gu­lar­it­ies \( \{z^n+f(x,y)=0\} \) with \( \mathbb{Z} \)HS link. In [e11], the con­jec­ture was proved for it­er­ated sus­pen­sions. Re­cently, M. A. Cueto, P. Popes­cu-Pam­pu, and D. Stepan­ov [e13], [e14] have an­nounced a proof of the gen­er­al con­jec­ture, which com­bines tools from trop­ic­al and log­ar­ithmic geo­metry (in the sense of Fon­taine and Il­lusie).

5. Splice-quotient singularities

Figure 7.

In 2000, as we worked through the ba­sics of sin­gu­lar­it­ies of splice type, we began to look for gen­er­al­iz­a­tions of the un­ex­pec­ted res­ult of The­or­em 1.2, that the UAC of a weighted-ho­mo­gen­eous sin­gu­lar­ity with \( \mathbb{Q} \)HS link is a com­plete in­ter­sec­tion.

From the point of view of clas­si­fic­a­tion of sin­gu­lar­it­ies, the easi­est ex­amples not covered by The­or­em 1.2 are the quo­tient-cusps. These are sin­gu­lar­it­ies whose res­ol­u­tion graphs have the form shown in Fig­ure 7 for \( k \geq 2, e_i\geq 3 \) and some \( e_j > 2 \).

They are ra­tion­al sin­gu­lar­it­ies, log-ca­non­ic­al, and taut, i.e, the to­po­logy of the link de­term­ines the ana­lyt­ic type. The above quo­tient-cusp is double-covered by the cusp sin­gu­lar­ity whose res­ol­u­tion graph is shown in Fig­ure 8.

Figure 8.

It fol­lows that the uni­ver­sal abeli­an cov­er is also a cusp.

It is easy to de­term­ine, giv­en the res­ol­u­tion graph, when a cusp is a com­plete in­ter­sec­tion; and, this was the case for the UAC for the few ex­amples we checked. However, it re­quired a de­tailed and lengthy ana­lys­is of the un­im­od­u­lar mat­rix clas­si­fy­ing the quo­tient-cusp to prove the fol­low­ing.

The­or­em 5.1: [5] The uni­ver­sal abeli­an cov­er of a quo­tient-cusp is a com­plete in­ter­sec­tion of em­bed­ding di­men­sion 4.

Though we wrote down equa­tions and group ac­tion for the UAC, at the time (2001) we’d not yet for­mu­lated a gen­er­al meth­od for do­ing so, and our res­ult seemed ad hoc.

Giv­en the res­ol­u­tion dia­gram \( \Gamma \) of a sin­gu­lar­ity with \( \mathbb{Q} \)HS link \( \Sigma \), the meth­od de­scribed after The­or­em 2.2 al­lows one to pass from \( \Gamma \) to a splice dia­gram \( \Delta \). Only in the \( \mathbb{Z} \)HS case can one in­sure that the weights around a node are re­l­at­ively prime, and only in that case can one re­cov­er \( \Gamma \) from \( \Delta \).

Non­ethe­less, the same ap­proach as above in the \( \mathbb{Z} \)HS case means that as­sum­ing the semig­roup con­di­tions on the as­so­ci­ated splice dia­gram, one can write down a set of splice dia­gram equa­tions [6], Defin­i­tion 2.5, which give isol­ated com­plete in­ter­sec­tion sur­face sin­gu­lar­it­ies in \( \mathbb{C}^t \), where \( t \) is the num­ber of ends of \( \Gamma \) or \( \Delta \).

When \( \Gamma \) is star-shaped, Neu­mann’s The­or­em 1.2 proves that these equa­tions give the UAC of the weighted ho­mo­gen­eous sin­gu­lar­ity of graph \( \Gamma \); also, his res­ult de­scribes the group ac­tion. Neg­lect­ing this group ac­tion in the gen­er­al \( \mathbb{Q} \)HS case, the fol­low­ing sug­gest­ive as­ser­tion ap­pears not to have been proved in full gen­er­al­ity:

Con­jec­ture 5.2: If the graph \( \Gamma \) sat­is­fies the semig­roup con­di­tions, then the link of the com­plete in­ter­sec­tion sin­gu­lar­ity defined by splice dia­gram equa­tions is the UAC of \( \Sigma \).

A dif­fi­culty is that splice dia­gram equa­tions de­pend only on the splice dia­gram \( \Delta \), while the cov­er­ing group of the UAC is the dis­crim­in­ant group \( D(\Gamma) \), not com­put­able from \( \Delta \). Re­call its con­struc­tion: let \( \mathbb E=\bigoplus_{j=1}^n \mathbb{Z} E_j \) be the lat­tice spanned by all the ex­cep­tion­al curves, and \( \mathbb E^* \) the dual lat­tice, with dual basis \( \{e_i\} \) defined by \( e_i(E_j)=\delta_{ij} \). Then \[ D(\Gamma)=\mathbb E^*/\mathbb E\cong H_1(\Sigma); \] view­ing \( \Sigma \) as the bound­ary of a tu­bu­lar neigh­bor­hood of \( E \), \( e_i \) rep­res­ents an ori­ented circle over a point of \( E_i \). The pair­ing of \( \mathbb E^*/\mathbb E \) in­to \( \mathbb{Q}/\mathbb{Z} \) is the to­po­lo­gic­al link­ing pair­ing on \( H_1(\Sigma) \). The key is to con­struct a nat­ur­al rep­res­ent­a­tion of \( D(\Gamma) \).

Pro­pos­i­tion 5.3: ([6], Proposition 5.2) Let \( e_{1},\dots,e_{t} \) be the ele­ments of \/ \( \mathbb{E}^{\star} \) cor­res­pond­ing to the \( t \) leaves of \( \Gamma \).
  1. The ho­mo­morph­ism \( \mathbb{E}^{\star}\rightarrow \mathbb{Q}^{t} \) defined by \[ e\mapsto (e\cdot e_{1},\dots,e\cdot e_{t}) \] in­duces an in­jec­tion \[ D(\Gamma)=\mathbb{E}^{\star}/\mathbb{E} ~\hookrightarrow~(\mathbb{Q} /\mathbb{Z} )^{t}. \]
  2. Ex­po­nen­ti­at­ing each \( \mathbb{Q}/\mathbb{Z}\hookrightarrow \mathbb{C}^* \) via \( r \mapsto \operatorname{exp}(2\pi i r) \) provides a faith­ful di­ag­on­al rep­res­ent­a­tion \( D(\Gamma)\hookrightarrow (\mathbb{C}^*)^t \).

Now, as­sume \( \Gamma \) is a graph with \( t \) ends. We have a di­ag­on­al rep­res­ent­a­tion of the dis­crim­in­ant group in \( \mathbb{C}^t \). If \( \Delta \) sat­is­fies the semig­roup con­di­tions, one has splice type equa­tions in \( \mathbb{C}^t \). What we need is to be able to choose those equa­tions which be­have equivari­antly with re­spect to the group ac­tion.

Defin­i­tion 5.4: (congruence conditions) Let \( \Gamma \) be a res­ol­u­tion dia­gram, yield­ing a splice dia­gram \( \Delta \) sat­is­fy­ing the semig­roup con­di­tions. Then \( \Gamma \) sat­is­fies the con­gru­ence con­di­tions if for each node \( v \), one can choose for every ad­ja­cent edge \( e \) an ad­miss­ible monomi­al \( M_{ve} \) so that \( D(\Gamma) \) trans­forms each of these monomi­als ac­cord­ing to the same char­ac­ter.

Clearly, the semig­roup plus con­gru­ence con­di­tions to­geth­er mean that the dis­crim­in­ant group acts on ap­pro­pri­ate splice dia­gram com­plete in­ter­sec­tion sin­gu­lar­it­ies (high­er or­der terms must also be \( D(\Gamma) \)-equivari­ant). Here is the main res­ult:

The­or­em 5.5: (splice-quotient singularities [6], Theorem 7.2) Sup­pose \( \Gamma \) sat­is­fies the semig­roup and the con­gru­ence con­di­tions. Then:
  1. Splice dia­gram equa­tions for \( \Gamma \) define an isol­ated com­plete in­ter­sec­tion sin­gu­lar­ity \( (X,0) \).
  2. The dis­crim­in­ant group \( D(\Gamma) \) acts freely on a punc­tured neigh­bor­hood of 0 in \( X \).
  3. \( Y=X/D(\Gamma) \) has an isol­ated nor­mal sur­face sin­gu­lar­ity, and a good res­ol­u­tion whose as­so­ci­ated dual graph is \( \Gamma \).
  4. \( X\rightarrow Y \) is the uni­ver­sal abeli­an cov­er­ing.
  5. \( X\rightarrow Y \) maps the curve \( z_{w}=0 \) to an ir­re­du­cible curve, whose prop­er trans­form on the good res­ol­u­tion of \( Y \) is smooth and in­ter­sects the ex­cep­tion­al curve trans­vers­ally, along \( E_{w} \). In fact the func­tion \( z_w^{\operatorname{det}(\Gamma)}\equiv x_w \), which is \( D(\Gamma) \)-in­vari­ant and hence defined on \( Y \), van­ishes to or­der \( \operatorname{det}(\Gamma) \) on this curve.

A sin­gu­lar­ity \( (Y,0) \) which arises in this way is called a splice-quo­tient sin­gu­lar­ity.

Of course, there are many things to be checked! The proof is by in­duc­tion on the num­ber of nodes, and one needs to con­sider cer­tain non­min­im­al res­ol­u­tions.

We con­jec­tured that ra­tion­al sur­face sin­gu­lar­it­ies were splice quo­tients; but the first proof of this fact was due to T. Ok­uma, who com­bined the semig­roup and con­gru­ence con­di­tions in a nov­el way.

The­or­em 5.6: [e9] Ra­tion­al sur­face sin­gu­lar­it­ies, and min­im­ally el­lipt­ic sin­gu­lar­it­ies with \( \mathbb{Q} \)HS links, are splice quo­tient sin­gu­lar­it­ies.

Even­tu­ally, we were able to prove a gen­er­al res­ult, which in­cluded Ok­uma’s, identi­fy­ing the cru­cial ana­lyt­ic prop­erty of a splice-quo­tient \( (Y,0) \): for every end of the graph, there is an “end-curve func­tion”on \( Y \), as in \( (5) \) of The­or­em 5.5. Our res­ult (be­low) gen­er­al­ized The­or­em 3.4, but the proof was far more dif­fi­cult be­cause of the group ac­tion.

End Curve The­or­em: [8] Let \( (Y,0) \) be a nor­mal sur­face sin­gu­lar­ity with \( \mathbb{Q} \)HS link \( \Sigma \). Sup­pose that for each leaf \( w \) of the res­ol­u­tion dia­gram \( \Gamma \) there ex­ists a cor­res­pond­ing end curve func­tion \( x_w:Y\rightarrow \mathbb{C} \) which cuts out (pos­sibly with mul­ti­pli­city) an end knot \( K_w\subset \Sigma \) (or end curve) for that leaf. Then \( (Y,0) \) is a splice-quo­tient sin­gu­lar­ity, and a choice of a suit­able root \( z_w \) of \( x_w \) for each \( w \) gives co­ordin­ates for the splice-quo­tient de­scrip­tion.

We close with a caveat: As in­dic­ated in The­or­em 10.1 of [6], the gen­er­al splice type equa­tion can be found us­ing any choice of ad­miss­ible monomi­als, as long as one al­lows the ad­di­tion of high­er or­der equivari­ant terms. This im­plies that in the \( \mathbb{Z} \)HS situ­ation, what look like “equisin­gu­lar” de­form­a­tions of sin­gu­lar­it­ies of splice type are also of splice type. However, even for a weighted ho­mo­gen­eous \( \mathbb{Q} \)HS sin­gu­lar­ity, a pos­it­ive weight de­form­a­tion of the de­fin­ing equa­tions might no longer be a splice-quo­tient. For ex­ample, \( z^2=x^4+y^9+txy^7 \) ([8], Ex­ample 10.4) is a pos­it­ive weight de­form­a­tion, but not a de­form­a­tion of splice-quo­tients; func­tions on the res­ol­u­tion lift un­der de­form­a­tion, but for \( t\neq 0 \) they are no longer end-curve func­tions.

Jonath­an Wahl is Emer­it­us Pro­fess­or of Math­em­at­ics at the Uni­versity of North Car­o­lina at Chapel Hill. His col­lab­or­a­tion with Wal­ter Neu­mann for over 30 years res­ul­ted in sev­en pub­lished pa­pers.

Works

[1] W. D. Neu­mann: “A cal­cu­lus for plumb­ing ap­plied to the to­po­logy of com­plex sur­face sin­gu­lar­it­ies and de­gen­er­at­ing com­plex curves,” Trans. Am. Math. Soc. 268 : 2 (1981), pp. 299–​344. MR 632532 Zbl 0546.​57002 article

[2] W. D. Neu­mann: “Abeli­an cov­ers of quasi­ho­mo­gen­eous sur­face sin­gu­lar­it­ies,” pp. 233–​243 in Sin­gu­lar­it­ies (Ar­cata, CA, 20 Ju­ly–7 Au­gust 1981), part 2. Edi­ted by P. Or­lik. Pro­ceed­ings of Sym­po­sia in Pure Math­em­at­ics 40. Amer­ic­an Math­em­at­ic­al So­ci­ety (Provid­ence, RI), 1983. MR 713252 Zbl 0519.​32010 incollection

[3] D. Eis­en­bud and W. Neu­mann: Three-di­men­sion­al link the­ory and in­vari­ants of plane curve sin­gu­lar­it­ies. An­nals of Math­em­at­ics Stud­ies 110. Prin­ceton Uni­versity Press, 1985. MR 817982 Zbl 0628.​57002 book

[4] W. Neu­mann and J. Wahl: “Cas­son in­vari­ant of links of sin­gu­lar­it­ies,” Com­ment. Math. Helv. 65 : 1 (1990), pp. 58–​78. MR 1036128 Zbl 0704.​57007 article

[5] W. D. Neu­mann and J. Wahl: “Uni­ver­sal abeli­an cov­ers of quo­tient-cusps,” Math. Ann. 326 : 1 (2003), pp. 75–​93. MR 1981612 Zbl 1032.​14010 ArXiv math/​0101251 article

[6] W. D. Neu­mann and J. Wahl: “Com­plete in­ter­sec­tion sin­gu­lar­it­ies of splice type as uni­ver­sal abeli­an cov­ers,” Geom. To­pol. 9 : 2 (2005), pp. 699–​755. MR 2140991 Zbl 1087.​32017 ArXiv math/​0407287 article

[7] W. D. Neu­mann and J. Wahl: “Com­plex sur­face sin­gu­lar­it­ies with in­teg­ral ho­mo­logy sphere links,” Geom. To­pol. 9 : 2 (2005), pp. 757–​811. MR 2140992 Zbl 1087.​32018 ArXiv math/​0301165 article

[8] W. D. Neu­mann and J. Wahl: “The end curve the­or­em for nor­mal com­plex sur­face sin­gu­lar­it­ies,” J. Eur. Math. Soc. 12 : 2 (2010), pp. 471–​503. MR 2608949 Zbl 1204.​32019 ArXiv 0804.​4644 article