return

Celebratio Mathematica

Walter D. Neumann

Contributions on Lipschitz geometry of complex singularities

by Lev Birbrair and Anne Pichon

Anne

I first met Wal­ter in 2002 in New York at a con­fer­ence on sin­gu­lar­ity the­ory he or­gan­ized with Ágnes Sz­ilárd. He then vis­ited me sev­er­al times in Mar­seille, and we wrote two short pa­pers on ana­lyt­ic real­iz­a­tions of mul­tilinks in 3-di­men­sion­al man­i­folds, one with An­drás Némethi. A couple of years later, in June 2009, Wal­ter vis­ited Mar­seille with his col­lab­or­at­or Lev Birbrair. The first day, they star­ted to ex­plain their sem­in­al res­ults on in­ner Lipschitz geo­metry of com­plex sin­gu­lar germs on the black­board of the cafet­er­ia, and offered me one of the greatest days — if not the greatest — of my math­em­at­ic­al life. It turned out after a couple of hours that our view­points were per­fectly com­ple­ment­ary and would lead to a glob­al and com­plete pic­ture of the in­ner Lipschitz geo­metry of nor­mal sur­face germs. At the end of the day, we had a three-line state­ment that a sur­face sin­gu­lar­ity has a nat­ur­al thick-thin de­com­pos­i­tion, sim­il­ar to the one de­scribed by Gregori Mar­gulis for neg­at­ive curvature spaces. Our col­lab­or­a­tion really took off there, and I am very grate­ful to both of them to have brought me onto the team, and par­tic­u­larly to Wal­ter, who had the idea of this vis­it in Mar­seille.

Lev

I learned who Wal­ter Neu­mann is when I was a stu­dent. His clas­sic the­or­em about plumb­ing cal­cu­lus and res­ol­u­tion of com­plex sur­face sin­gu­lar­it­ies was already fam­ous. We met at sev­er­al con­fer­ences in sin­gu­lar­ity the­ory. His talks were al­ways very in­ter­est­ing, but be­fore the Trieste con­fer­ence in 2005 we nev­er dis­cussed math­em­at­ics. His lec­ture in Trieste stim­u­lated Al­ex­an­dre Fernandes and me to think about Lipschitz geo­metry of com­plex sur­faces. When I vis­ited Columbia Uni­versity that same year, Al­ex­an­dre and I had the first ex­amples of non­con­ic­al sin­gu­lar­it­ies of com­plex sur­faces with re­spect to the in­ner met­ric. Wal­ter said to me: “Ex­cel­lent. The Lipschitz geo­metry is nonempty. We must see what we can do.” This was the point were the col­lab­or­a­tion star­ted. When Anne joined the group, the the­ory be­came much deep­er. All Wal­ter’s ex­per­i­ence and ca­pa­city turned out to be ex­tremely use­ful for the cre­ation of the mod­ern Lipschitz geo­metry. Wal­ter is a won­der­ful col­lab­or­at­or and I am very happy that I had the great priv­ilege of work­ing with him.

We will present an over­view of the res­ults we proved with Wal­ter on the in­ner Lipschitz clas­si­fic­a­tion of com­plex sur­faces, and then, of some of the res­ults he proved with Anne on their out­er Lipschitz geo­metry and on Lipschitz nor­mal em­bed­dings. We refer to the ori­gin­al pa­pers for pre­cise defin­i­tions, state­ments and proofs, as well as to the re­cent lec­ture notes volume on Lipschitz geo­metry of sin­gu­lar­it­ies ed­ited by Wal­ter and Anne [11].

1. Seminal works

Let \( (X,0) \) be a germ of real (or com­plex) ana­lyt­ic space em­bed­ded in some \( (\mathbb{R}^n,0) \) (or \( (\mathbb{C}^n,0) \)). A nat­ur­al ques­tion is:

How does \( X \) look in a neigh­bor­hood of the ori­gin?

There are mul­tiple an­swers to this vague ques­tion, de­pend­ing on the cat­egory we work in. The fam­ous Con­ic­al Struc­ture The­or­em gives a com­plete an­swer in the to­po­lo­gic­al cat­egory: let \( B_{\epsilon} \) be the \( n \)-ball with ra­di­us \( \epsilon \) centered at the ori­gin of \( \mathbb{R}^n \), let \( S_{\epsilon} \) be its bound­ary, and let \( X^{(\epsilon)} = S_{\epsilon} \cap X \) be the link at dis­tance \( \epsilon \) of \( (X,0) \); for \( \epsilon > 0 \) suf­fi­ciently small, we have an homeo­morph­ism of pairs \[ (B_{\epsilon}, X \cap B_{\epsilon}) \cong (B_{\epsilon}, \operatorname{Cone}( X^{(\epsilon)})), \] where \( \operatorname{Cone}(X^{(\epsilon)}) \) de­notes the uni­on of seg­ments join­ing the ori­gin to a point \( x \in X^{(\epsilon)} \). But this state­ment com­pletely ig­nores the geo­met­ric prop­er­ties of the set \( (X,0) \), and a nat­ur­al ques­tion is then:

How does the link \( X^{(\epsilon)} \) evolve met­ric­ally as \( \epsilon \) tends to zero?

In oth­er words, is \( X \cap B_{\epsilon} \) met­ric­ally con­ic­al, i.e., bi-Lipschitz equi­val­ent to the straight cone over \( X^{(\epsilon)} \)? Or are there some parts of \( X^{(\epsilon)} \) which shrink faster than lin­early when \( \epsilon \) tends to zero?

These ques­tions can be ap­proached from two dif­fer­ent view­points de­pend­ing on the choice of the met­ric. Giv­en an ana­lyt­ic germ \( (X,0) \), and an em­bed­ding \( (X,0) \hookrightarrow (\mathbb{R}^n,0) \), there are two nat­ur­al choices for the met­ric on \( (X,0) \): the out­er met­ric \( d_o \) in­duced by the Eu­c­lidean met­ric of the am­bi­ent space \( \mathbb{R}^n \), i.e., \( d_o(x,y) = \|x-y\|_{\mathbb{R}^n} \), and the in­ner met­ric \( d_i \) defined as the in­duced arc-length met­ric on \( (X,0) \). We call out­er (resp. in­ner) Lipschitz geo­metry of \( (X,0) \) its equi­val­ence class up to loc­al out­er (resp. in­ner) bi-Lipschitz homeo­morph­ism.

The pi­on­eer­ing work on the Lipschitz geo­metry of com­plex spaces is the 1969 pa­per [e1] by Frédéric Pham and Bern­ard Teis­si­er (for an Eng­lish trans­la­tion, see the fi­nal chapter of [11]), where they present a com­plete clas­si­fic­a­tion of the Lipschitz geo­metry of com­plex curves for the out­er met­ric. The high­er di­men­sions re­mained un­ex­plored for dec­ades. In par­tic­u­lar, the in­ner Lipschitz geo­metry of com­plex sin­gu­lar­it­ies was com­pletely ig­nored. Ac­tu­ally, it was known for a long time that com­plex curve germs are met­ric­ally con­ic­al, i.e., in­ner bi-Lipschitz equi­val­ent to straight cones over their links, and it was be­lieved by ex­perts that the same was true in high­er di­men­sions. They were wrong! The first ex­ample of a non­con­ic­al com­plex sur­face germ was giv­en by Lev and Al­ex­an­dre Fernandes in [e13]: for \( k \geq 2 \), the sur­face sin­gu­lar­ity \( A_k : x^2+y^2+z^{k+1}=0 \) is not met­ric­ally con­ic­al. Short after this pi­on­eer­ing work, Wal­ter joined the team, and they wrote a series of pa­pers giv­ing a lot of new ex­amples ([2], [3], [4]) sug­gest­ing that fail­ure of met­ric con­ic­alness is a com­mon phe­nomen­on. For ex­ample, they proved that if the two low­est weights of a weighted ho­mo­gen­eous sin­gu­lar­ity \( (X,0) \) are un­equal, then \( (X,0) \) is not met­ric­ally con­ic­al. In par­tic­u­lar, among ADE sin­gu­lar­it­ies, only \( A_1 \) and \( D_4 \) are met­ric­ally con­ic­al. This work brought about the re­sur­gence of the field. Lipschitz geo­metry of germs be­came since then one of the most at­tract­ive and flour­ish­ing fields in sin­gu­lar­ity the­ory.

Be­fore present­ing Wal­ter’s main con­tri­bu­tions to the field, let us say a few words on the reas­ons that make the Lipschitz geo­metry of sin­gu­lar spaces so in­ter­est­ing and pop­u­lar.

First, while the out­er and in­ner met­rics on \( (X,0) \) defined above de­pend on the choice of an em­bed­ding \( (X,0) \hookrightarrow (\mathbb{R}^n,0) \), their out­er and in­ner Lipschitz geo­met­ries do not de­pend on this choice (see, e.g., Pro­pos­i­tion 7.2.13 of [e20]). In oth­er words, the in­ner and out­er Lipschitz geo­met­ries of \( (X,0) \) are de­term­ined by the ana­lyt­ic type of \( (X,0) \). Moreover, the out­er Lipschitz geo­metry of \( (X,0) \) de­term­ines the in­ner Lipschitz geo­metry, and the lat­ter ob­vi­ously de­term­ines its to­po­lo­gic­al type, i.e., its class up to loc­al homeo­morph­ism. In ad­di­tion, through the pi­on­eer­ing ex­amples men­tioned above, Wal­ter, Al­ex­an­dre Fernandes and Lev dis­covered that the in­ner Lipschitz geo­metry of a com­plex ana­lyt­ic germ may not be de­term­ined by its to­po­lo­gic­al type. There­fore, the in­ner and out­er Lipschitz geo­met­ries in­deed give two in­ter­est­ing in­ter­me­di­ate clas­si­fic­a­tions between the ana­lyt­ic type and the to­po­lo­gic­al type.

An­oth­er mo­tiv­a­tion comes from the “tame­ness” of Lipschitz clas­si­fic­a­tion of space germs. Ana­lyt­ic types of sin­gu­lar com­plex space germs con­tain con­tinu­ous mod­uli, and this is why it is dif­fi­cult to de­scribe a com­plete ana­lyt­ic clas­si­fic­a­tion. For ex­ample, con­sider the fam­ily of com­plex plane curves germs \( (X_t,0)_{t \in \mathbb{C}} \) with equa­tions \( xy(x-y)(x-ty)=0 \). For every pair \( (t,t^{\prime}) \) with \( t \neq t^{\prime} \), \( (X_t,0) \) is not ana­lyt­ic­ally equi­val­ent to \( (X_{t^{\prime}},0) \). On the con­trary, in [e3], Laurent Sieben­mann and Den­nis Sul­li­van con­jec­tured that the set of Lipschitz struc­tures is tame, i.e., the set of equi­val­ence classes of com­plex al­geb­ra­ic sets in \( \mathbb{C}^n \), defined by poly­no­mi­als of de­gree less than or equal to \( k \) is fi­nite. One of the most im­port­ant res­ults on Lipschitz geo­metry of com­plex al­geb­ra­ic sets is the proof of this con­jec­ture by Tadeusz Mostowski in [e4], which was gen­er­al­ized to the real set­ting a couple of years later by Adam Parusiński ([e6] and [e8]) and more re­cently to the o-min­im­al set­ting by Nguy­en Nhan and Guil­laume Valette ([e14]). Then a com­plete clas­si­fic­a­tion of the Lipschitz geo­metry of sin­gu­lar spaces seemed to be a more reach­able goal. But so far, there did not ex­ist ex­pli­cit de­scrip­tion of the equi­val­ence classes, ex­cept in the case of com­plex plane curves stud­ied by Frédéric Pham and Bern­ard Teis­si­er.

2. The thick-thin decomposition of surfaces and the complete inner Lipschitz classification

When we star­ted to work to­geth­er with Wal­ter in June 2009 in Mar­seille, it turned out after a couple of hours that our view­points were per­fectly com­ple­ment­ary. They would even­tu­ally lead to a glob­al and com­plete pic­ture of the in­ner Lipschitz geo­metry of nor­mal sur­face germs. At the end of the day, we had a three-line state­ment say­ing that a sur­face sin­gu­lar­ity has a nat­ur­al thick-thin de­com­pos­i­tion, sim­il­ar to the one de­scribed by Mar­gulis for neg­at­ive curvature spaces. From this start­ing point, it took us more than two years to write a full proof of the the­or­em and based on it, to give a com­plete clas­si­fic­a­tion of in­ner geo­metry of nor­mal com­plex sur­face germs, which even­tu­ally ap­peared in [7].

In the first ex­amples of non­con­ic­al nor­mal sur­face germs presen­ted by Lev, Al­ex­an­dre Fernandes and Wal­ter, the ob­struc­tion to met­ric­al con­ic­alness of nor­mal is giv­en by the ex­ist­ence of fast loops. A fast loop in \( (X,0) \) is a con­tinu­ous fam­ily of loops \( \gamma_{\epsilon} \subset X^{(\epsilon)} \) para­met­rized by the ra­di­us \( \epsilon > 0 \) such that \( \gamma_{\epsilon} \) is ho­mo­top­ic­ally non­trivi­al in the link \( X^{(\epsilon)} \) and such that the length of \( \gamma_{\epsilon} \) shrinks faster than lin­early, i.e., there ex­ists a real num­ber \( q > 1 \) such that \[ \lim_{\epsilon \to 0} \frac{\text{length}(\gamma_\epsilon)}{ \epsilon^q}=0. \]

The key res­ult of [7] is the Thick-Thin De­com­pos­i­tion The­or­em (The­or­em 1.6 in that art­icle) which states the ex­ist­ence and uni­city (up to a very nat­ur­al equi­val­ence re­la­tion) of what we called a (min­im­al) thick-thin de­com­pos­i­tion of every nor­mal sur­face germ \( (X,0) \): \[ (X,0) = (X_{\text{thick}},0) \cup (X_{\text{thin}},0). \] Roughly speak­ing, it re­flects the fact that all fast loops of min­im­al length in­side their ho­mo­topy class are con­cen­trated in a min­im­al 4-di­men­sion­al semi­al­geb­ra­ic sub­germ \( (X_{\text{thin}},0) \) of \( (X,0) \), while the 4-di­men­sion­al semi­al­geb­ra­ic sub­germ \( (X_{\text{thick}}, 0) \), defined as the clos­ure of the com­ple­ment of \( (X_{\text{thin}},0) \) in \( (X,0) \) is es­sen­tially met­ric­ally con­ic­al.

As a first ap­plic­a­tion, the thick-thin de­com­pos­i­tion en­abled us to char­ac­ter­ize the nor­mal sur­face germs which are met­ric­ally con­ic­al:

The­or­em 2.1: ([7], Theorem 7.5, Corollary 1.8) A nor­mal sur­face germ \( (X,0) \) is met­ric­ally con­ic­al if and only if its thin part is empty, so that its thick part is the whole \( (X,0) \).

The thick-thin de­com­pos­i­tion is re­mark­ably re­lated to the to­po­logy of the link \( X^{(\epsilon)} \):

The­or­em 2.2: ([7], Theorem 1.7) The thick and thin parts can be de­com­posed as uni­ons of 4-di­men­sion­al semi­al­geb­ra­ic sets pair­wise dis­joint out­side the ori­gin: \[ (X_{\mathrm{thick}}, 0) = \bigcup_{i=1}^r (Y_i,0) \quad \text{ and } \quad (X_{\mathrm{thin}}, 0) = \bigcup_{j=1}^s (Z_j,0)\] in such a way that it in­duces a de­com­pos­i­tion of the link: \[ X^{(\epsilon)} = \bigcup_{i=1}^r Y_i,^{(\epsilon)} \cup \bigcup_{j=1}^s Z_j^{(\epsilon)} \] such that for \( \epsilon > 0 \) suf­fi­ciently small,
  1. each \( Y^{(\epsilon)}_i \) is a Seifert fibered man­i­fold;
  2. each \( Z^{(\epsilon)}_j \) is a graph man­i­fold (uni­on of Seifert man­i­folds glued along bound­ary com­pon­ents) and not a sol­id tor­us;
  3. there ex­ist ra­tion­al num­bers \( q_j > 1 \) and fibra­tions \( \zeta_j^{(\epsilon)}: Z^{(\epsilon)}_j\to S^1 \) de­pend­ing smoothly on \( \epsilon\le \epsilon_0 \) whose fibers have dia­met­er of or­der \( \epsilon^{q_j} \).

One of the re­mark­able prop­er­ties of the thick-thin de­com­pos­i­tion is that it can be ex­pli­citely de­scribed in a very easy way through a suit­able res­ol­u­tion of the sin­gu­lar­ity by us­ing the plumb­ing cal­cu­lus de­veloped by Wal­ter in [1] thirty years be­fore (see Jonath­an Wahl’s con­tri­bu­tion in the present volume). Let us de­scribe this more pre­cisely. Con­sider the min­im­al good res­ol­u­tion \( \pi:(\widetilde X,E)\to (X,0) \) which factors through the nor­mal­ized blowup \( e_0 \) of the ori­gin. Call \( \mathcal L \)-curve any ir­re­du­cible com­pon­ent of the ex­cep­tion­al di­visor \( \pi^{-1}(0) \) which maps sur­ject­ively onto an ir­re­du­cible com­pon­ent of \( e_0^{-1}(0) \). By blow­ing-up once each in­ter­sec­tion point (if any) between \( \mathcal L \)-curves, we can as­sume that no \( \mathcal L \)-curves in­ter­sect. De­note by \( \Gamma \) the dual graph of the ex­cep­tion­al di­visor \( \pi^{-1}(0) \), i.e., the graph whose ver­tices \( \nu \) are in bijec­tion with the ex­cep­tion­al com­pon­ents \( E_{\nu} \) of \( \pi^{-1}(0) \) and the edges between \( \nu \) and \( \nu^{\prime} \) with the points of \( E_{\nu} \cap E_{\nu^{\prime}} \), and let \( V(\Gamma) \) be the set of ver­tices of \( \Gamma \). Call \( \mathcal L \)-node of \( \Gamma \) any \( \nu \in V(\Gamma) \) cor­res­pond­ing to an \( \mathcal L \)-curve. A node of \( \Gamma \) is a ver­tex \( \nu \) which is an \( \mathcal L \)-node or has 3 in­cid­ent edges or such that the cor­res­pond­ing curve \( E_{\nu} \) has genus \( > 0 \). For each ver­tex \( v \in V(\Gamma) \), let \( N(E_v) \) be a small tu­bu­lar neigh­bor­hood of \( E_v \) in \( \widetilde X \). For each \( \mathcal L \)-node \( \nu \), let \( \Gamma_{\nu} \) be the max­im­al con­nec­ted sub­graph of \( \Gamma \) con­tain­ing \( \nu \) and no oth­er node, and set \( N(\Gamma_{\nu})= \bigcup_{v \in\Gamma_{\nu}} N(E_\nu) \). Then \[ X_{\text{thick}} = \bigcup_{\nu \in L(\Gamma)} \pi(N(\Gamma_{\nu})), \] where \( L(\Gamma) \) de­notes the set of \( \mathcal L \)-nodes.

Figure 1. The two thick-thin decompositions in the Briançon–Speder family
\( x^5+z^{15}+y^7z+txy^6=0 \).
Left: \( t\neq 0 \). Right: \( t=0 \).

Many ex­pli­cit ex­amples are de­scribed in [7]. Let us give just one. Con­sider the fam­ily of sur­face sin­gu­lar­it­ies \( (X_t,0)\subset (\mathbb{C}^3,0) \) with equa­tions \( x^5+z^{15}+y^7z+txy^6=0 \). It is the \( \mu \)-con­stant fam­ily in­tro­duced by Bri­ançon and Speder in [e2]. It has con­stant to­po­lo­gic­al type, while the thick-thin de­com­pos­i­tion changes rad­ic­ally when \( t \) be­comes 0. The two graphs of Fig­ure 1 are the res­ol­u­tion graphs of \( \pi \) for \( t=0 \) and for \( t \neq 0 \); the \( \mathcal L \)-nodes are the circled ver­tices, the sub­trees \( \Gamma_{\nu} \) cor­res­pond­ing to the com­pon­ents of the thick part have black ver­tices, while the max­im­al sub­graphs with white ver­tices cor­res­pond to the com­pon­ents of the thin part. The neg­at­ive num­ber weight­ing each ver­tex is the self-in­ter­sec­tion of the cor­res­pond­ing ex­cep­tion­al curve in \( \widetilde{X} \). The weight \( [8] \) means that the cor­res­pond­ing ex­cep­tion­al curve has genus 8, while the oth­ers have genus zero. For \( t \neq 0 \), \( (X_t,0) \) has three thick com­pon­ents and a single thin one. For \( t =0 \), it has one com­pon­ent of each type.

The second main res­ult of [7], built on a re­fine­ment of the thick-thin de­com­pos­i­tion, gives the com­plete in­ner Lipschitz clas­si­fic­a­tion of nor­mal sur­face germs:

The­or­em 2.3: ([7], Theorem 1.9) The bi-Lipschitz geo­metry of \( (X,0) \) de­term­ines and is uniquely de­term­ined by the fol­low­ing data:
  1. the graph de­com­pos­i­tion of \( X^{(\epsilon)} \) defined by the thick-thin de­com­pos­i­tion;
  2. in the nota­tion of The­or­em 2.2, for each thin zone \( Z_j^{(\epsilon)} \), the ho­mo­topy class of the fo­li­ation by fibers of the fibra­tion \( \zeta_j^{(\epsilon)}: Z_j^{(\epsilon)}\to S^1 \), and for each Seifert piece \( S_{\nu}^{(\epsilon)} \) of \( Z_j^{(\epsilon)} \), a ra­tion­al num­ber \( q_{\nu} \geq q_j \) with \( q_\nu=1 \) such that the fibers of the re­stric­tion \( \zeta_j^{(\epsilon)} \) to \( S_{\nu}^{(\epsilon)} \) have dia­met­er of or­der \( \epsilon^{q_{\nu}} \).

The high­er-di­men­sion­al case re­mains up to now al­most un­ex­plored.

3. Outer Lipschitz geometry

As already men­tioned, the com­plete Lipschitz clas­si­fic­a­tion of com­plex curve germs for the out­er met­ric was es­tab­lished by Frédéric Pham and Bern­ard Teis­si­er in [e1] us­ing the concept of Lipschitz sat­ur­a­tions. In fact, any curve germ \( (X,0)\subset (\mathbb{C}^N,0) \) is bi-Lipschitz equi­val­ent to its im­age \( (\ell(C), 0) \) by a gen­er­ic lin­ear plane pro­jec­tion \( \ell : \mathbb{C}^n \to \mathbb{C}^2 \). There­fore, it is suf­fi­cient to clas­si­fy plane curve germs. The main res­ult of [e1] says that the out­er Lipschitz type of a com­plex curve germ \( (C,0) \subset (\mathbb{C}^2,0) \) de­term­ines and is de­term­ined by its em­bed­ded to­po­lo­gic­al type, i.e., by the homeo­morph­ism class of the pair \( (S^3_{\epsilon}, C^{(\epsilon)}) \). This res­ult was re­vis­ited and com­pleted by Wal­ter and Anne with a geo­met­ric view­point:

The­or­em 3.1: ([8], Theorem 1.1) Let \( (C_1,0)\subset (\mathbb{C}^2,0) \) and \( (C_2,0)\subset (\mathbb{C}^2,0) \) be two germs of com­plex curves. The fol­low­ing are equi­val­ent:
  1. \( (C_1,0) \) and \( (C_2,0) \) have same Lipschitz geo­metry, i.e., there is a homeo­morph­ism of germs \[\phi: (C_1,0)\to (C_2,0)\] which is bi-Lipschitz for the out­er met­ric;
  2. there is a homeo­morph­ism of germs \( \phi: (C_1,0)\to (C_2,0) \), holo­morph­ic ex­cept at 0, which is bi-Lipschitz for the out­er met­ric;
  3. \( (C_1,0) \) and \( (C_2,0) \) have the same em­bed­ded to­po­logy, i.e., there is a homeo­morph­ism of germs \( h: (\mathbb{C}^2,0) \to (\mathbb{C}^2,0) \) such that \( h(C_1)=C_2 \);
  4. there is a bi-Lipschitz homeo­morph­ism of germs \( h: (\mathbb{C}^2,0) \to (\mathbb{C}^2,0) \) with \( h(C_1)=C_2 \).

The equi­val­ence of (ii) and (iii) is the res­ult of Pham and Teis­si­er. The equi­val­ence of (i), (iii) and (iv) is the new con­tri­bu­tion. The proof of \( \text{(ii)} \Rightarrow \text{(iii)} \) based on what we call a bubble trick ar­gu­ment, which can be con­sidered as a pro­to­type for ex­plor­ing Lipschitz geo­metry of sin­gu­lar spaces in vari­ous set­tings. An ex­ten­ded ver­sion of the proof is giv­en in [e20]. The first oc­curence of a bubble trick ar­gu­ment we found in the lit­ter­at­ure is in the pa­per of Jean-Pierre Henry and Adam Parusiński [e11]. This im­port­ant tool is also used in re­cent works such has the con­struc­tion of the mod­er­ately dis­con­tinu­ous ho­mo­logy by Javi­er Fernández de Boba­dilla, Sonja Hein­ze, Maria Pe Pereira and José Ed­son Sam­paio in [e18].

The ques­tion of the out­er Lipschitz clas­si­fic­a­tion of high­er-di­men­sion­al germs is still open. Ac­tu­ally, a very small amount of ana­lyt­ic in­vari­ants are de­term­ined by the to­po­lo­gic­al type of an ana­lyt­ic germ (even if one con­siders the em­bed­ded to­po­lo­gic­al type), and a first nat­ur­al ques­tion to ask is wheth­er the Lipschitz clas­si­fic­a­tion is suf­fi­ciently ri­gid to catch ana­lyt­ic in­vari­ants:

Which ana­lyt­ic­al in­vari­ants are in fact Lipschitz in­vari­ants?

Dur­ing the last dec­ade, it was shown that a large amount of ana­lyt­ic in­vari­ants are de­term­ined by the out­er Lipschitz geo­metry in the case of a sur­face germ. The main res­ult of [8] is the pi­on­eer­ing res­ult in this dir­ec­tion. It shows that the out­er Lipschitz class con­tains a lot of in­form­a­tion on the sin­gu­lar­ity and con­firms that out­er Lipschitz geo­metry of sin­gu­lar­it­ies is a very prom­ising area to ex­plore. It can be seen as a first key­stone to a com­plete clas­si­fic­a­tion of the out­er Lipschitz geo­metry of com­plex sur­faces. The high­er-di­men­sion­al set­ting re­mains al­most un­ex­plored.

The­or­em 3.2: ([6], Theorem 1.2) If \( (X,0) \) is a nor­mal com­plex sur­face sin­gu­lar­ity then the out­er Lipschitz geo­metry on \( X \) de­term­ines:
  1. the dec­or­ated res­ol­u­tion graph of the min­im­al good res­ol­u­tion of \( (X,0) \) which re­solves the basepoints of a gen­er­al lin­ear sys­tem of hy­per­plane sec­tions, i.e., the min­im­al good res­ol­u­tion which factors through the blow-up of the max­im­al ideal;
  2. the mul­ti­pli­city of \( (X,0) \);
  3. the max­im­al ideal cycle in its res­ol­u­tion;
  4. for a gen­er­ic hy­per­plane \( H \), the out­er Lipschitz geo­metry of the curve \( (X\cap H,0) \);
  5. the dec­or­ated res­ol­u­tion graph of the min­im­al good res­ol­u­tion of \( (X,0) \) which re­solves the basepoints of the fam­ily of po­lar curves of plane pro­jec­tions, i.e., the min­im­al good res­ol­u­tion which factors through the Nash modi­fic­a­tion;
  6. the to­po­logy of the dis­crim­in­ant curve of a gen­er­ic plane pro­jec­tion;
  7. the out­er Lipschitz geo­metry of the po­lar curve of a gen­er­ic plane pro­jec­tion.

“Dec­or­ated res­ol­u­tion graph” means the res­ol­u­tion graph dec­or­ated with ar­rows cor­res­pond­ing to the com­pon­ents of the strict trans­form of the re­solved curve.

The proof of this res­ult is based on a soph­ist­ic­ated ver­sion of the bubble trick which leads to the graph de­com­pos­i­tion of the link \( X^{(\epsilon)} \) de­scribed by the res­ol­u­tion graph of point (i).

The Lipschitz in­vari­ance of the mul­ti­pli­city (point (ii) in the above the­or­em) was then ex­ten­ded bey­ond the nor­mal case by Al­ex­an­dre Fernandes and José Ed­son Sam­paio ([e15]) for a hy­per­sur­face in \( \mathbb{C}^3 \) and by Javi­er Fernández de Boba­dilla, Al­ex­an­dre Fernandes and José Ed­son Sam­paio ([e17]) for the gen­er­al sur­face case. However it is now known that the mul­ti­pli­city is not a Lipschitz in­vari­ant in high­er di­men­sions ([e19]).

4. Lipschitz normal embeddings

An­oth­er im­port­ant con­tri­bu­tion of Wal­ter in the area of Lipschitz geo­metry of sin­gu­lar­it­ies is his series of works on Lipschitz nor­mal em­bed­ding. A sub­set (or a germ) in \( \mathbb{R}^n \) is Lipschitz Nor­mally Em­bed­ded (LNE for short) if the iden­tity map is a bi-Lipschitz homeo­morph­ism (loc­al in case of a germ) between the in­ner and out­er met­rics. Of course, every com­pact smooth sub­man­i­fold in \( \mathbb{R}^n \) is LNE. But this prop­erty turns out to be fairly rare among sin­gu­lar spaces, and the res­ults ob­tained by Wal­ter with his coau­thors sug­gest that it plays an im­port­ant (quasi un­ex­plored) role in sev­er­al areas of sin­gu­lar­ity the­ory such as res­ol­u­tion of sin­gu­lar­it­ies or the struc­ture of arc and jet spaces.

The pi­on­eer­ing res­ult on LNE is the fol­low­ing the­or­em due to Lev and Tadeusz Mostowski. Its ori­gin­al ver­sion is proved in the semi­al­geb­ra­ic set­ting, but this res­ult re­mains true with the same proof in the more gen­er­al sub­ana­lyt­ic cat­egory and even, in the o-min­im­al cat­egory. We state it here in the sub­ana­lyt­ic set­ting.

The­or­em 4.1: [e9] Let \( X\subset\mathbb{R}^m \) be a con­nec­ted sub­ana­lyt­ic set. Then there ex­ist a LNE sub­ana­lyt­ic set \( \widetilde{X}\subset\mathbb{R}^q \) and a sub­ana­lyt­ic in­ner bi-Lipschitz homeo­morph­ism \( p : X \rightarrow \widetilde{X} \).

The map \( p : X \rightarrow \widetilde{X} \) is called a nor­mal em­bed­ding of \( X \).

As a con­sequence of this the­or­em, one ob­tains the fol­low­ing al­tern­at­ive defin­i­tion of the no­tion of Lipschitz nor­mal em­bed­ding. The proof was nev­er writ­ten be­fore. That is why we in­clude it in this manuscript.

Pro­pos­i­tion 4.2: A con­nec­ted sub­ana­lyt­ic sub­space \( X \) of \( \mathbb{R}^n \) is LNE if and only if there ex­ist a sub­ana­lyt­ic bi-Lipschitz homeo­morph­ism \( f : (X,d_i) \to (X,d_o) \), that is, there ex­ists and a real num­ber \( K\geq 1 \) such that for all \( x,y \) in \( X \), \[ \frac{1}{K} d_i(f(x),f(y)) \leq d_o(x,y) \leq {K} d_i(f(x),f(y)). \]

Proof. As­sume there ex­ist a sub­ana­lyt­ic bi-Lipschitz homeo­morph­ism \( f : (X,d_i) \to (X,d_o) \) and a real num­ber \( K\geq 1 \) as in the state­ment. Let \( p : (X,d_i) \to ( \widetilde{X},d_i) \) be a nor­mal em­bed­ding of \( X \) as in The­or­em 4.1. We de­duce that \( p \circ f^{-1} \) is bi-Lipschitz with re­spect to the out­er met­rics. There­fore \( p \circ f^{-1} \) is also bi-Lipschitz with re­spect to the in­ner met­rics. This im­plies that \( \mathrm{Id}_X = f \circ f^{-1} = f \circ p^{-1} \circ p \circ f^{-1} \) is bi-Lipschitz from \( (X,d_i) \) to \( (X,d_o) \). ☐

Com­plex and real al­geb­ra­ic sets of \( \mathbb{C}^n \) or \( \mathbb{R}^n \) are sub­ana­lyt­ic sets. By The­or­em 4.1, they ad­mit a glob­al sub­ana­lyt­ic Lipschitz nor­mal em­bed­ding. Tadeusz Mostowski asked if there ex­ists a com­plex al­geb­ra­ic Lipschitz nor­mal em­bed­ding when \( X \) is a com­plex al­geb­ra­ic set, i.e., a (sub­ana­lyt­ic) Lipschitz nor­mal em­bed­ding for which the im­age set \( \tilde{X}\subset\mathbb{C}^n \) is a com­plex al­geb­ra­ic set.

The an­swer is pos­it­ive in com­plex di­men­sion 1: every com­plex al­geb­ra­ic curve ad­mits a com­plex al­geb­ra­ic Lipschitz nor­mal em­bed­ding. This fol­lows from the fact that an ir­re­du­cible germ of com­plex al­geb­ra­ic curve is sub­ana­lyt­ic­ally bi-Lipschitz homeo­morph­ic with re­spect to the in­ner met­ric to the germ of \( \mathbb{C} \) at the ori­gin (see for ex­ample Pro­pos­i­tion 2.1 of [8]). Then a nor­mal em­bed­ding can be con­struc­ted by us­ing the so-called tent pro­ced­ure in­tro­duced in [e9].

Wal­ter, Lev and Al­ex­an­dre Fernandes answered neg­at­ively to Mostowski’s ques­tion in high­er di­men­sion by prov­ing the fol­low­ing res­ult:

The­or­em 4.3: ([5], Theorem 2.3) If \( 1 < a < b \) and \( a \) is not a di­visor of \( b \) then no neigh­bor­hood of 0 in the Brieskorn sur­face in \( \mathbb{C}^3 \) \[ X= \{(x,y,z)\in\mathbb{C}^3 \mid x^b+y^b+z^a=0\} \] ad­mits a com­plex al­geb­ra­ic nor­mal em­bed­ding.

The proof of this res­ult is based on a beau­ti­ful to­po­lo­gic­al ar­gu­ment due to Wal­ter that we re­pro­duce here:

Proof. As­sume there ex­ists a com­plex al­geb­ra­ic Lipschitz nor­mal em­bed­ding \( p : X \to\widetilde{X} \). By The­or­em 1.3 of [2], the germ \( (X,0) \) is met­ric­ally con­ic­al for the in­ner met­ric by a sub­ana­lyt­ic homeo­morph­ism, and so is \( (\widetilde{X},0) \). Since \( (\widetilde{X},0) \) is LNE, it is then sub­ana­lyt­ic­ally met­ric­ally con­ic­al for the out­er met­ric. By a res­ult of Bernig and Lytchak (see Re­mark 2, The­or­em 1.2 of [e12]), this im­plies that \( (\widetilde{X}, 0) \) is sub­ana­lyt­ic­ally bi-Lipschitz homeo­morph­ic to its tan­gent cone \( C_0 \widetilde{X} \) at 0 for the out­er met­ric, which is the com­plex cone over a pro­ject­ive al­geb­ra­ic curve \( C \). Since \( (X,0) \) has isol­ated sin­gu­lar­ity, then its link is a to­po­lo­gic­al man­i­fold, which im­plies that \( C \) is ir­re­du­cible and loc­ally ir­re­du­cible. Now, if \( C \) had a sin­gu­lar point, this would con­tra­dict the fact that \( C_0 \widetilde{X} \) is LNE. There­fore, \( C_0\widetilde{X} \) is a com­plex al­geb­ra­ic cone over a smooth ir­re­du­cible pro­ject­ive curve, thus its link is a \( S^1 \)-bundle. On the oth­er hand, the link of \( X \) at 0 is a Seifert fibered man­i­fold with \( b \) sin­gu­lar fibers of de­gree \( a/\gcd(a,b) \). This is a con­tra­dic­tion be­cause the Seifert fibra­tion of a Seifert fibered man­i­fold (oth­er than a lens space) is unique up to dif­feo­morph­ism. ☐

An­oth­er im­port­ant con­tri­bu­tion of Wal­ter is a char­ac­ter­iz­a­tion of nor­mal sur­face sin­gu­lar­it­ies which are LNE. In [e16], Lev proved with Rodrigo Mendes that a semi­al­geb­ra­ic germ \( (X,0) \) is LNE if and only if if all pairs of real arcs \( p_1 \) and \( p_2 \) in \( (X,0) \) have equal in­ner and out­er con­tacts. This cri­terion is dif­fi­cult to use ef­fect­ively in prac­tice to prove LNE-ness of a germ since it re­quires to com­pute the in­ner and out­er con­tact or­ders of an im­mense amount of pairs of arcs. Wal­ter, with Anne and Helge Ped­er­sen, es­tab­lished in [9] an ana­log­ous cri­terion for com­plex sur­face germs where the num­ber of pairs of arcs to be tested is re­duced drastic­ally to just fi­nitely many types of pairs.

This ef­fi­cient LNE cri­terion led to the dis­cov­ery of sev­er­al non­trivi­al in­fin­ite fam­ily of LNE nor­mal sur­face germs. For ex­ample:

The­or­em 4.4: [10] Let \( (X,0) \) be a nor­mal com­plex sur­face germ as­sume that it is ra­tion­al. Then \( (X,0) \) is LNE if and only if it is a min­im­al sin­gu­lar­ity.

Min­im­al sin­gu­lar­it­ies were in­tro­duced by Janos Kollár in [e5]. In di­men­sion 2, they are ra­tion­al sur­face sin­gu­lar­it­ies which play a ma­jor role in res­ol­u­tion the­ory (see [e7], [e10]).

Sev­er­al oth­er fam­il­ies were dis­covered later us­ing the cri­terion of [9], for ex­ample among su­per-isol­ated sin­gu­lar­it­ies [e21]. But LNE-ness of com­plex sur­faces is far from be­ing com­pletely un­der­stood, and the high­er-di­men­sion­al case is still in its in­fancy.

5. Dissemination

In June 2018, Wal­ter or­gan­ized with Anne an in­ter­na­tion­al school on Lipschitz geo­metry of sin­gu­lar­it­ies in Cuerna­vaca, Mex­ico. About fifty PhD stu­dents and young re­search­ers at­ten­ded the event, and the the­ory and prom­ising ex­ten­sions were in­tro­duced a huge vari­ety of view­points, by Wal­ter him­self and also Patrick Popes­cu-Pam­pu, Maria Apare­cida Soares Ru­as, Bern­ard Teis­si­er, Dav­id Trot­man, among oth­ers.

The lec­ture notes were com­pleted and col­lec­ted in a Lec­ture Notes in Math­em­at­ics volume [11], ed­ited by Wal­ter and Anne, which is the first in­tro­duct­ory book on Lipschitz geo­metry of sin­gu­lar­it­ies, and which is now in the hands of many PhD stu­dents and young re­search­ers. The volume also con­tains an eng­lish ver­sion of the his­tor­ic­al pa­per of Pham and Teis­si­er [e1], which was a pre­print in French of the École Poly­tech­nique dur­ing 50 years.

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] L. Birbrair, A. Fernandes, and W. D. Neu­mann: “Bi-Lipschitz geo­metry of weighted ho­mo­gen­eous sur­face sin­gu­lar­it­ies,” Math. Ann. 342 : 1 (2008), pp. 139–​144. MR 2415318 Zbl 1153.​14003 ArXiv 0704.​2041 article

[3] L. Birbrair, A. Fernandes, and W. D. Neu­mann: “Bi-Lipschitz geo­metry of com­plex sur­face sin­gu­lar­it­ies,” Geom. Ded­icata 139 (2009), pp. 259–​267. MR 2481850 Zbl 1164.​32005 ArXiv 0804.​0194 article

[4] L. Birbrair, A. Fernandes, and W. D. Neu­mann: “Sep­ar­at­ing sets, met­ric tan­gent cone and ap­plic­a­tions for com­plex al­geb­ra­ic germs,” Se­lecta Math. (N.S.) 16 : 3 (2010), pp. 377–​391. MR 2734336 Zbl 1200.​14010 ArXiv 0905.​4312 article

[5] L. Birbrair, A. Fernandes, and W. D. Neu­mann: “On nor­mal em­bed­ding of com­plex al­geb­ra­ic sur­faces,” pp. 17–​22 in Real and com­plex sin­gu­lar­it­ies (São Car­los, Brazil, 27 Ju­ly–2 Au­gust 2008). Edi­ted by M. Manoel, M. C. Romero Fuster, and C. T. C. Wall. Lon­don Math­em­at­ic­al So­ci­ety Lec­ture Note Series 380. Cam­bridge Uni­versity Press, 2010. Ded­ic­ated to our friends Maria (Cid­inha) Ru­as and Terry Gaffney in con­nec­tion to their 60th birth­days. MR 2759086 Zbl 1215.​14057 ArXiv 0901.​0030 incollection

[6] W. D. Neu­mann and A. Pichon: Lipschitz geo­metry of com­plex sur­faces: Ana­lyt­ic in­vari­ants and equisin­gu­lar­ity. Technical report, November 2012. ArXiv 1211.​4897 techreport

[7] L. Birbrair, W. D. Neu­mann, and A. Pichon: “The thick-thin de­com­pos­i­tion and the bilipschitz clas­si­fic­a­tion of nor­mal sur­face sin­gu­lar­it­ies,” Acta Math. 212 : 2 (2014), pp. 199–​256. MR 3207758 Zbl 1303.​14016 ArXiv 1105.​3327 article

[8] W. D. Neu­mann and A. Pichon: “Lipschitz geo­metry of com­plex curves,” J. Sin­gul. 10 (2014), pp. 225–​234. MR 3300297 Zbl 1323.​14003 ArXiv 1302.​1138 article

[9] W. D. Neu­mann, H. M. Ped­er­sen, and A. Pichon: “A char­ac­ter­iz­a­tion of Lipschitz nor­mally em­bed­ded sur­face sin­gu­lar­it­ies,” J. Lond. Math. Soc. (2) 101 : 2 (2020), pp. 612–​640. MR 4093968 Zbl 1441.​14015 ArXiv 1806.​11240 article

[10] W. D. Neu­mann, H. M. Ped­er­sen, and A. Pichon: “Min­im­al sur­face sin­gu­lar­it­ies are Lipschitz nor­mally em­bed­ded,” J. Lond. Math. Soc. (2) 101 : 2 (2020), pp. 641–​658. MR 4093969 Zbl 1441.​14016 ArXiv 1503.​03301 article

[11] In­tro­duc­tion to Lipschitz geo­metry of sin­gu­lar­it­ies (Cuerna­vaca, Mex­ico, 11–22 June 2018). Edi­ted by W. Neu­mann and A. Pichon. Lec­ture Notes in Math­em­at­ics 2280. Spring­er (Cham, Switzer­land), 2020. MR 4200092 Zbl 1456.​58002 book