REVISTA MATEMATICA COMPLUTENSE ISSN 1139-1138 (2000) vol. XIII, num. 2, 369-382 METRIC THEORY OF SEMIALGEBRAIC CURVES Lev BIRBRAIR* and Alexandre C.G. FERNANDES Abstract We present a complete bi-Lipschitz classification of germs of semialgebraic curves (semialgebraic sets of the dimension one). For this purpose we introduce the so-called Hélder Semicomplex, a bi- Lipschitz invariant. Hélder Semicomplex is the collection of all first exponents of Newton-Puiseux expansions, for all pairs of branches of a curve. We prove that two germs of curves are bi-Lipschitz equivalent if and only if the corresponding Hélder Semicomplexes are isomorphic. We also prove that any Hélder Semicomplex can be realized as a germ of some plane semialgebraic curve. Finally, we compare these Hilder Semicomplexes with Holder Complexes- complete bi-Lipschitz invariant of two-dimensional semialgebraic sets. 1 Introduction In this paper, we are going to study singular semialgebraic curves (one- dimensional semialgebraic sets) from the metric viewpoint. Some recent papers on the metric theory of semialgebraic sets were devoted to intrin- sic metric properties of semialgebraic surfaces (see [2],[1],[6]). One of the main questions of this theory is the question of bi-Lipschitz classifica- tion. This question was investigated in [1], for germs of semialgebraic surfaces, with respect to the intrinsic (length) metric. The problem of bi-Lipschitz classification with respect to the induced (euclidian) metric is more complicated. Namely, if two semialgebraic sets are bi-Lipschitz “Partially supported by CMP by grant N300985/93-2 1991 Mathematics Subject Classification: 14P25. Servicio de Publicaciones. Universidad Complutense. Madrid, 2000 369LEV BIRBRAIR AND ALEXANDRE FERNANDES METRIC THEORY OF ... equivalent with respect to the euclidian metric they are bi-Lipschitz equivalent with respect to the length metric, but the converse is not true. The main result of the paper is a bi-Lipschitz classification of germs of semialgebraic curves with respect to the euclidian metric. We con- struct an invariant - the so-called Hélder Semicompler. This invariant is obtained in the following way. Take two branches X; and Xp of a semialgebraic curve X at the point zo € X. Let 24(r) € X; and 2(r) € X2 be two points on these branches such that ding(z1(r); x0) = dina(22(r),20) =r. The function dina(z1(r),22(r)) (the induced met- ric) is semialgebraic and, thus, admits a Newton-Puiseux expansion at 0. Hélder Semicomplex is the collection of all first exponents of Newton- Puiseux expansions, for all pairs of branches. Note, that this procedure is closely related to the calculation of Lojasiewicz exponent [4]. We prove that Hélder Semicomplex is a complete invariant. It means that germs of two semialgebraic curves are bi-Lipschitz equivalent if their Hélder Semicomplexes are combinatorially equivalent (isomorphic). We prove (Theorem 4.1), moreover, that, for every Hélder Semicomplex, there exists a germ of a plane semialgebraic curve realizing it (i.e. the Hélder Semicomplex of this curve coincides with the given one) (Theo- rem 5.1). We study in section 6 some relations between Hélder Complexes [1] and Hélder Semicomplexes. Hélder Complexes defined in {1} are complete invariants of germs of semialgebraic surfaces. We prove that if a two-dimensional set X is normally embedded at zo (i.e. intrinsic and induced metrics are bi-Lipschitz equivalent [3]) then the Hélder Semicomplex of the germ of Sing(X) at zo is completely determined by the Hélder Complex of X at 29 (Theorem 6.2). Some examples presented in the section 6 show that this result does not remain true if X is not normally embedded. Note that all results of the paper are hold for semianalytic set. As a consequence of this observation we obtain that every germ of a semian- alytic curve is bi-Lipschitz equivalent to a germ of some semialgebraic curve (Corollary 5.2). The authors are grateful to Maria Aparecida Ruas, Jean-Jacques Risler and the anonymous referee for very useful comments. 370 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIII, num. 2, 369-382LBV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF 2 Héider Semicomplexes Definition 1. A complete finite graph I’ with a rational valued function a: Ep + QM[1,00) defined on the set of edges Er of T is called a Holder Semicomples if a satifies the following: Isosceles Property. For every three vertices a1, a2, a3 € I’, we have: if (a1, a2) < a(a2,a3) < a(a1,a3) then a(a1,a2) = a(az,a3). (Note that, since T is a complete graph, any edge is completely determined by two vertices). Remark 1. A Holder Semicomplex can be defined in the following equivalent way. Let A be a finite set with a symmetric function a : Ax A — diag(Ax.A) + QN[1,00). If @ satisfies the isosceles property then the pair (A,@) can be identified with a Hélder Semicomplex. Remark 2. Consider a function d: Ax A > Q such that 1 d{a,as)=0 if a =a, and d(a1,ax) = > otherwise. Then (A,d) is an ultrametric space if and only if (A,q@) is a Hélder Semicomplex. Definition 2. Two Hélder Semicomplezes (P,a1) and (P2,e2) are called isomorphic (or combinatorially equivalent) if there exists a graph isomorphism h: T; — [2 such that, for each pair of vertices a,,a2 of Ty, we have: (h(a), h(a2)) = an (a1, a2). Definition 3. A morphism of (A,a1) to (A2,a2) is a map m: A, > Ap such that, for all ay,a2 € Ay, a2(m(a1),m(a2)) > ar(ai,a2) if may) A (az). Thus, we obtain that the set of all Holder Semicomplexes is a cat- egory, the isomorphism defined in Definition 2 is an isomorphism in this category and Remark 2 defines a functor from this category to the category of metric spaces. 371 REVISTA MATEMATICA COMPLUTENSE (2000) vol. X111, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF 3 Semialgebraic curves and Hélder Semicom- plex Let X C R” be a semialgebraic curve and rz) € X. By the standard conic structure theorem, we have the following statement: There exists a neighbourhood Uz, of zo in R" such that X A Uz, = k UX; where the sets X; have the following properties: m1 (1) X; is a semialgebraic subset of R” and homeomorphic to a semiseg- ment (1,0) by a homeomorphism A; : (1,0) > X; such that h,(0) = zo; (2) For every i Xj, X;0X; = {20}; (3) There exists a number ro such that, for every i and 0 0 and € > 0 such that, for r < ro, Xi B,(0) C Ce(vi) NB, (0); 373 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIII, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES Xj9B,(0) C C.(vj) N B,(0); Ce(vi) 1 Ce(v5) = {0}- Thus, 9:3 > dist(C.(vi) — B(0), Ce(vj) — By(0)) > er, for some c > 0. It means that Ajj = aiy = 1. Now suppose that X; and Xj have the same unit tangent vector vp at 0. Clearly, aj; < fi;. Suppose that ay < Bij. Observe that, for sufficiently small r, we can choose the points y;(r) € X; and y;(r) € X; such that gij(r) = |lyi(r) — y,(r)ll depends semialgebraically and continuously on r. Observe that either y;(r) = xi(r), or yj(r) = a,(r), for sufficiently small r (otherwise the function gj;(r) has to be locally constant what contradicts to semialgebraicity). Suppose that y;(r) = a;(r). Consider a triangle with vertices x;(r),y,(r) and 2;(r). Since Biz > a3, we have: lleelr) ~ vg(r)I] < [lxe(r) — 25(r)IL, for small r. Thus, the angle between z;(r) — 2,(r) and yj(r) — 2;(r) tends to zero as r tends to zero. But, since u(r) = 2j(r) x;(r)|] Tur > u, as rv, i we obtain that Z(2;(r) — 2;(r),v9) 3 0 as r > 0. On the other hand, X; and X; have the same tangent vector vp at 0. ‘Thus, for every sufficiently small € > 0, there exists ro > 0 such that, for every r < ro, we have x;(r) € Ce(vo)S;(v9) and 2;(r) € Ce(v9)MS-(v9). It means that Lailr) ~ 23(r),00) > F ~ He) where d(¢) — 0, as € -+ 0. This is a contradiction. 374 REVISTA MATEMATICA COMPLUTENSE, (2000) vol. X11, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF 4 H6lder Semicomplexes as bi-Lipschitz invari- ants Theorem 4.1. Germs of closed semialgebraic curves (X1, 11) and (Xp, 22) are bi-Lipschitz equivalent if and only if the corresponding Hélder Semi- complexes (A1,a) = sh(X1,21) and (Ap,) = sh(Xo,x2) are combi- natorially equivalent. Proof. (=)Let © : (X1,21) + (X2,2) be a given bi-Lipschitz map. Let {X}}f, and {X?}"2, be pancake decompositions of (X1,11) and (X2,x2). Since & is a homeomorphism, we have ky = ko = k and, for each i, ®(X}) = X? (we can choose another renumeration if necessary). Let (Ai,a1) and (A2,a2) be the Hélder Semicomplexes corresponding to (X1,21) and (X2,22). Let A, = {a}}£,, let Ap = {a?}*_, and put h(a}) = a}. Let us prove that a (a},a}) = a2(a?, a3), for each i, j. Since © is a bi-Lipschitz map, there exists K > 0 such that, for sufficiently small r, we have: (Xj — By(e1)) CX? — Bxr(22); (1) (Xj — By(a1)) CX} — Brr(22). (2) Let y}(r) € X} — B-(a1) and y}(r) € X} — B,(x1) be points such that lly (r) — yj (r)|| = dist(X} — B, (a1), X} - B,(21)) Thus, by (1),(2) we obtain: I®OHr)) — GE} ()IL = 9% (Kr) (here gj;,92; mean the function gj; defined for X and X» correspond- ingly, see section 3). Since ® is a bi-Lipschitz map, there exists a con- stant L > 0 such that G(r) = lve () — a) = LIP) — SEF) Thus, 937) > Lgij(Kr). 375 REVISTA MATEMATICA COMPLUTENSE (2000) vol. X111, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF ... Using the Newton-Puiseux Theorem and the Order Comparison Lemma, we obtain that a1(a},a}) < a2(a?,a3). Using the fact that ®—? is also a bi-Lipschitz map we obtain that ay(a!,a a}) > an(a?,a’ a3). The first part of the theorem is proved. (<=) Let (Ai,a1) and (42,02) be combinatorially equivalent. We suppose that the isomorphism preserves the enumeration. Let {X}}*, and {X?}*, be pancake decompositions of (X1,21) and (X2, 22). Let ro > 0 be a number such that x}(r) and x?(r) are well defined semi- algebraic functions, for every r < ro and for every i. Consider the semialgebraic function r : X; U Xz + R defined by r(a) -{ |la-a,||, for ce Xy; le —aall, for © Xp. Define & : X1 N Byo(t1) + X2M Byo(x2) by @(x) = 22(r(c)), for ze X}. Claim. There exists 6 > 0 such that ®|y,9,(2:) is a bi-Lipschitz map. Proof of the claim. Each X} and X? has a tangent vector at 2; and 2, respectively, hence, for some éy > 0, we have: Olying. (2) iS a bi-Lipschitz map, for all i. It is enough to prove that, for each pair (X},X}), the map ®|(x1u51)B,, (21) is bi-Lipschitz, for some 6; > 0. Let « € X} and y € X} be two points sufficiently close to 21. We can suppose that r(x) < r(y). Let z € X} such that r(z) = r(z). Consider the triangles (x,y,z) and (®(x), ®(y), O(z)). Since the curves X},X},X?, X} are semialgebraic sets, for sufficiently small r, they are close enough to their tangent vectors. Thus, the angle at the vertex z in the triangle (x,y,z) and the angle at the vertex &(z) in the triangle (B(x), (y), ®(z)) are bounded away from zero. Using this fact we obtain that there exist K,, K2 > 0 such that Kymax(r(y) — r(2), f5(r(@))) $ lly — all (3) and lly ~ || < Kemax(r(y) — r(x), fy (r(2))) (4) 2 (here fj, 2, mean the function fi; defined for X; and Xp correspond- ingly, see section 3). By the same way, we obtain that there exist 376 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XII1, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF ... 11, Lz > 0 such that Zymax(r(®(y)) — r(B(2)), £3 (r(P(@)))) < IPH) — B(@)|_— (5) and Py) ~ &(2)|| < Lamax(r(H(y)) — r(®(2)), FR(r(#(@))). 6) Since a (a},a}) = a2(a?,a?), there exist constants M,,M: > 0 such that Mifis(r) < £3 (r) < Mafi(r)- (7) Using the inequalities (3),(4),(5),(7) and the facts that r(y) = r(®(y)) and r(x) = r(®(<)) we obtain that is a bi-Lipschitz map. The claim is proved. And, thus, Theorem 4.1 is proved. 5 The Realization Theorem Theorem 5.1. Let (A,a) be a Hilder Semicomplez. Let #A =k. Then there exists a semialgebraic subset X C R?® with dimX = 1 satisfying the following conditions: (1) (A,q) is a Hilder Semicomplez corresponding to (X,0); (2) (X,0) has a pancake decomposition {X;}#_, such that X; is a graph of an algebraic function 1; : {0,€] > R with v;(0) =0. Remark 5. This theorem means that each Hélder Semicomplex can be realized as a plane semialgebraic curve. X is called a realization of (A,a). Proof of the theorem 5.1. We use induction on k. For k = 1, the statement is obvious. Suppose that the statement is proved, for some k, Let (A,a) be a Hilder Semicomplex such that #A = k +1. Let a = maxa(ai,aj). We can suppose that ao = a(ax,a,41). Put A=A- {ax41} and & = al, 7. Let X be a realization of (4,4). Let X; be a pancake corresponding to a, (i = 1,...,k). By the induction hypothesis, we have that X; is a graph of some algebraic function 4; : (0, €0] + R such that #;(0) = 0. Put 377 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XI, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF .. Periz) = sx + p(x). Since X contains a finite number of branches, there exist s and € < eo such that be41(x) # pi(2), for every i=1,...,k, if £0. k+1 Put di = Palla and let X; be the graph of #j. Let X = (J Xi. By i=l straightforward calculations we obtain that (A,q) is a Hélder Semicom- plex corresponding to (X, 0). Corollary 5.2. Every semianalytic subset X C R" such that dimX =1 is locally bi-Lipschitz equivalent to some semialgebraic subset X C R?. 6 Hdédlder Semicomplexes and Hélder Complexes Here we are going to compare the bi-Lipschitz invariants of germs of semialgebraic curves defined in this paper with bi-Lipschitz invariants of germs of semialgebraic surfaces (see [1]). A complete intrinsic bi- Lipschita invariant of two-dimensional semialgebraic sets is called Hélder Complex. Hilder Complex is a pair (I’, 8) where I’ is a finite graph and A is a rational valued function defined on the set of edges of . One can find all the definitions and results related to this subject in [1] and (6). Let (I, 8) be a Hélder Complex. A vertex a of I is called essential if, for every neighbourhood Us of a, we have: Uy is not a topological 1- manifold (in other words, a is neither artificial nor a loop vertex of I, see [1]). We are going to define a Hélder Semicomplex (A, &) corresponding to (I, A) in the following way: (1) The set Vz of vertices of A be the set of all essential vertices of i (2) Let a1, az be vertices of A. Let P be the set of all finite paths y= {91,---,9s} (here 91,...,9s € Er) connecting a, and aj. We define . 1, if a, a2 belong to different connected components of I; Gar, a2) = max min 8(9.). YEP oREY Blo) 378 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIII, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF . Proposition 6.1. (A,&) is a Hélder Semicomplez. Proof. Let us prove the isosceles property. Let a1, a2,a3 be three ver- tices of A such that G(a,,a2) < &(a2,a3) < &(a1,a3). We can suppose that a1,a2,a3 belong to the same connected component of I’. (Other- wise the proposition is trivial). Let {g1,...,g4} be a path in TP con- necting a; and a. Let {g},...,9,} be a path connecting a; and a3. Clearly, {ge,..-,91,9};---+94} is a path connecting az and a3. Thus, G(a2,a3) < min {&(a),a2),&(a1,a3)}. Since &(a;,a2) < &(ai,a3) we obtain &(a2, a3) = &(a1, a2). Let Y C R® be a two-dimensional semialgebraic set. Let Y = Y° UY! UY? be the canonical topological stratification of Y. This stratification can be obtained in the following way. Let Y? be the set of all points y € Y such that there exists a neighbourhood U, C R" such that Uy MY is a topological two-dimensional manifold. Let Y! be the set of points of Y — Y? such that, for each § € Y!, there exists a neigh- bourhood Uy such that Uj (Y — Y?) is a topological one-dimensional manifold. Set Y° = Y — ¥Y? — Y!. By [5], Y°,Y! and Y? are semialge- braic sets and, thus, Y° is finite (see also [9}). Let yo € Y° and yo € Cl(Y) (here Cl(Y') means a closure of Y!). Then the germ of Cl(Y!) at yo is a germ of one-dimensional semialge- braic set. Let (A, a) be a Hélder Semicomplex associated to (Cl(Y"), yo) (ie. (A,a) = sh(CU(Y"), yo). Let (I, 8) be a canonical Hélder Complex of Y at yo (see[1]). Let (A, @) be the Holder Semicomplex corresponding to (I, 8) defined above. Theorem 6.2. (1) There exists a morphism m : (A, &) — (A, a). (2) If is locally normally embedded at yo (see [3]) then (A, &) and (A,@) are isomorphic. Proof. (2) Let ¥;! and ¥} be two branches (pancakes) of Cl(¥?). Let a; and a; be the corresponding vertices in (I,8). Suppose that a; and aj belong to the same connected component of [. Let y = {g1,---,9s} be a “maximal” path in I connecting a; and a; (it means that (ai, aj) = min 6(gx)). By results of [1], the union of the curvilinear triangles corresponding to the edges gi,..., 95 is bi-Lipschitz equivalent 379 REVISTA MATEMATICA COMPLUTENSE, (2000) vol. XII1, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF ... to the standard &(aj, a;)-Hélder triangle. Let a;(r) € Yj! and a;(r) € Y} be points such that dina(ai(r), yo) = dina(aj(r); yo) = r- Using results of [1] or [10] we obtain that there exist two constants K, and Ko such that Kidi(ai(r), yo) dina(ai(r), a;(r)), for sufficiently small r. Since ¥ is normally embedded, we obtain that (A, &) is a Hélder Semicomplex associated to (Cl(¥!), go). Since dina(@i(r), a3(r)) = 18H) + ofr) and dinalai(r), a3(r)) = 72H) + ofr2()), we have that (ai, a5) < ee(aj, a5). (8) 380 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIII, num. 2, 369-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF ... Set m(a;) = aj, for all i. By (8), mis a morphism of Hélder Semicom- plexes, . Remark 6. If Y is not normally embedded (A,a) and (A,&) are not necessary isomorphic. To see it consider the following example. Let (P,8) be a Hélder Complex such that T has more than one essential vertices. Let a; and a2 be two essential vertices. Let I” be a new graph obtained from I’ by adding a new edge h connecting a; and a. Set Bi(h) > max B(g). ‘Then (I',6!) is a Holder Complex. Using the algo- . rithm from [6] we construct a set Y and a point yo such that (Y, yo) is a Geometric Hélder Complex corresponding to (I, 8). We can suppose that Y is normally embedded (otherwise we can obtain it using [3]). Consider now a semialgebraic set Y’ obtained from Y by cutting the triangle corresponding to the edge h. For this set the Hélder Semicom- plexes considered above are not isomorphic. Proposition 6.3. Let (Yi, yi) and (Yo, y2) be two germs of 2-dimensional semialgebraic sets such that (Yi,y1) and (Y2,y2) are bi-Lipschitz equiv- alent with respect to the induced metric. Then the corresponding Holder Semicomplexes (A,,@1) and (Ap,a2) are isomorphic. The proof is straightforward. References {1] Birbrair L. Local bi-Lipschitz. classification of 2-dimensional semialgebraic sets. - Houston Journal of Math., N3, vol.25, pp.453-472, 1999. (2] Brdker L., Kuppe M., Scheufler W. Inner metric properties of 2- dimensional semi-algebraic sets. Revista Matematica Complutense.vol.10, pp. 51-78, 1997. [3] Birbrair L., Mostowski T. Normal embedding of semialgebraic sets. - Michigan Math. Journal, vol. 47, pp. 125-132, 2000. [4] Bochnak J., Risler J-J. Sur les exposants de Lojasiewicz. Comment Math. Helvetici 50, 493-507, 1975. [5] Benedetti R., Risler J-J. Real algebraic and semialgebraic sets. Hermann, 1990. [6] Birbrair L., Sobolevsky M. Realization of Hélder Complexes. - Annales de la Faculte des Sciences de Toulouse, N1, vol. VIII, pp.35-44, 1998. 381 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIII, num. 2, 869-382LEV BIRBRAIR AND ALEXANDRE C.G. FERNANDES METRIC THEORY OF ... [7] Gibson C.G., Wirthmiiller K., du Plessis A.A., Looijenga E.J.N. Topologi- cal stability of smooth mappings. Lecture Notes in Mathematics, vol. 552, (8 9 (10) (11) Springer, 1976. Kurdyka K. On a subanalytic stratification satisfying Whitney property with exponent 1. Lecture Notes of Mathematics, 1524, pp. 316-322, 1992. King H. Topological invariance of intersection homology without sheaves. - Topology and its applications, N20, pp.149-160, 1985. Kurdyka K., Orro P. Distance geodesique sur un sous-analytique. Revista Matematica Complutence. vol 10. pp.173-182, 1997. Parusinski A. Lipschitz properties of semianalytic sets. Ann. Inst. Fourier (Grenoble) 38 no. 4, pp. 189-213, 1998. Lev Birbrair Departamento de Matematica Universidade Federal do Ceara Campus do Pici Bloco 914 CEP 60455-760- Fortaleza, Ce-Brasil E-mail: lev@mat.ufc.br Alexandre C.G. Fernandes Instituto de Matematica USP - Sao Carlos Caixa Postal 668, 13560-000 Sao Carlos, SP-Brasil E-mail: alex@icmc.sc.usp.br 382 Recibido: 24 de Mayo de 1999 Revisado: 6 de Marzo de 2000 REVISTA MATEMATICA COMPLUTENSE (2000) vol. XIIT, num. 2, 369-382