Table Of Content

A COMPLETE COMPLEX HYPERSURFACE IN THE BALL OF CN 4 Josip Globevnik 1 0 2 Abstract In 1977 P. Yang asked whether there exist complete immersed complex n submanifolds ϕ: Mk CN with bounded image. A positive answer is known for holomor- → a phic curves (k = 1) and partial answers are known for the case when k > 1. The principal J result of the present paper is a construction of a holomorphic function on the open unit 4 1 ball B of CN whose real part is unbounded on every path in B of finite length that N N ends on bB . A consequence is the existence of a complete, closed complex hypersurface ] N V in B . This gives a positive answer to Yang’s question in all dimensions k, N, 1 k < N, N ≤ C by providing properly embedded complete complex manifolds. . h t 1. Introduction and the main result a m Denote by ∆ the open unit disc in C and by B the open unit ball in CN, N 2. [ N ≥ In 1977 P. Yang asked whether there exist complete immersed complex submanifolds 1 v ϕ:Mk CN with bounded image [Y1, Y2]. The first answer was obtained by P. Jones → 5 [J] who constructed a bounded complete immersion ϕ:∆ C2 and a complete proper 3 → 1 holomorphic embedding ϕ:∆ B4. Since then there has been a series of results on → 3 bounded complete holomorphic curves (k = 1) immersed in C2 [MUY, AL1, AF] the most . 1 recent being that every bordered Riemann surface admits a complete proper holomorphic 0 immersion to B and a complete proper holomorphic embedding to B [AF]. The more 4 2 3 1 difficult complete embedding problem for k = 1 and N = 2 has been solved only very v: recently by A. Alarcón and F. J. López [AL2] who proved that every convex domain in C2 Xi contains a complete, properly embedded complex curve. r In the present paper we are interested primarilyin the higher dimensional case (k > 1) a where there are partial answers which are easy consequences of the results for complete curves. For instance, it is known that for any k IN there are complete bounded embedded ∈ complex k-dimensional submanifolds of C2k and it is an open question whether, in this case, N = 2k is the minimal possible dimension [AL2]. In the present paper we consider the case where ϕ is a proper holomorphic embedding. In this case ϕ(Mk) is a closed submanifold. We restate the definition of completeness for this case: DEFINITION 1.1 A closed complex submanifold M of B is complete if every path N p:[0,1) M such that p(t) 1 as t 1 has infinite length. → | | → → Note that this coincides with the standard definition of completeness since the paths p:[0,1) M such that p(t) 1 as t 1 are precisely the paths that leaveevery compact → | | → → subset of M as t 1 . → 1 Here is our main result: THEOREM 1.1 Let N 2. There is a holomorphic function f on B such that f is N ≥ ℜ unbounded on every path of finite length that ends on bB . N So our function f has the property that if p:[0,1] B is a path of finite length such N → that p(t) < 1 (0 t < 1) and p(1) = 1 then t f(p(t) is unbounded on [0,1). | | ≤ | | → ℜ The following corollary answers the question of Yang in all dimensions k and N by (cid:0) (cid:1) providing properly embedded complete complex manifolds. COROLLARY 1.2 For each k,N, 1 k < N, there is a complete, closed, k-dimensional ≤ complex submanifold of B . N Proof. We first prove the corollary for k = N 1 (that is, we first prove the existence − of the hypersurface, mentioned in the title). Let f be the function given by Theorem 1.1. By Sard’s theorem one can choose c C such that the level set M = z B :f(z) = c N ∈ { ∈ } is a closed submanifold of B . Let p:[0,1) M be a path such that p(t) bB as N N → → t 1. Assume that p has finite length. Then there is a point w on bB such that N → lim p(t) = w. By the properties of f, f is unbounded on p([0,1)). On the other hand, t→1 ℜ f((p(t)) = c (0 t < 1), a contradiction. So p must have infinite length. This proves that ≤ M is complete and so completes the proof of the corollary for k = N 1. Assume now that − 1 k N 2. By the first part of the proof there is a complete, closed, k dimensional ≤ ≤ − − complex submanifold M of B B . Clearly M is a complete, closed k dimensional k+1 N ⊂ − manifold of B . This completes the proof. N Remark If we want to have a connected, complete closed complex submanifold of B N then we simply take a connected component of M as above. Note also that the same function f gives many complete closed complex manifolds of B since, by Sard’s theorem, N one can use the same reasoning for almost every c in the range of f. 2. Outline of the proof of Theorem 1.1 Let M IN. For x IRM 0 and α IR, write ∈ ∈ \{ } ∈ H(x,α) = y IRM:< y x >= α , K(x,α) = y IRM:< y x > α . { ∈ | } { ∈ | ≤ } Assume that x IRM 0 (1 i n) and that i ∈ \{ } ≤ ≤ n P = K(x ,1) (2.1) i i=1 \ is a bounded set. Then P is a convex polytope, that is, the convex hull of a finite set. So P is a compact convex set that contains the origin in its interior. A convex subset F of P is called a face of P if any closed segment with endpoints in P whose relative interior meets F is contained in F. A k-face is a face F with dimF = k, that is, the affine hull of F is k-dimensional. A face of dimension M 1 is called a facet of P. Let P be a convex − polytope such that the representation (2.1) is irreducible, that is, n P = K(x ,1) for each k, 1 k n. i 6 ≤ ≤ i=1,i6=k \ 2 Then n bP = H(x ,1) P i ∩ i=1 [ and the sets F = H(x ,1) P, 1 i n, are precisely the facets of P. See [B] for the i i ∩ ≤ ≤ details. Given a convex set G, denote by ri(G) the relative interior of G in the affine hull of G. What remains of the boundary of a convex polytope P after we have removed relative interiors of all facets F , 1 i n, we call the skeleton of P (or more precisely, the i ≤ ≤ (M 2)-skeleton of P, the union of all (M 2)-dimensional faces of P) and denote by − − skel(P). Thus n skel(P) = F ri(F ) . i i \ i=1 [(cid:2) (cid:3) To prove Theorem 1.1 we first prove THEOREM 2.1 Let B be the open unit ball of IRM, M 3. There is a sequence of ≥ convex polytopes P , n IN, such that n ∈ ∞ P IntP P IntP B, P = B, 1 2 2 3 j ⊂ ⊂ ⊂ ⊂ ··· ⊂ j=1 [ such that if w skel(P ) (j IN) then j j ∈ ∈ ∞ w w = (2.2) j+1 j | − | ∞ j=1 X that is, the series in (2.2) diverges. In the proof of Theorem 1.1 we shall use the following COROLLARY 2.2 Let P , n IN be the sequence of convex polytopes from Theorem n ∈ ∞ 2.1. Let θ be a decreasing sequence of positive numbers such that θ < . For n n=1 n ∞ each n IN, let bP be the θ -neighbourhood of skel(P ) in bP , that is, = w n n n n n n ∈ U ⊂ P U { ∈ bP : dist(w,skel(P )) < θ . Let p: [0,1) B be a path such that p(t) 1 as t 1, n n n } → | | → → and such that for all sufficiently large n IN, p([0,1)) meets bP only at . Then p has n n ∈ U infinite length. Once we have proved Corollary 2.2 we prove Theorem 1.1 as follows. Let B be the N open unit ball of CN, N 2. Let P , n IN, be a sequence of convex polytopes as in n ≥ ∈ Theorem 2.1 with M = 2N, and let , n IN, be as in Corollary 2.2. Given ε > 0 and n n U ∈ L < we use an idea from [GS] to construct a function f , holomorphic on B , such n n N ∞ that f < ε on P and such that f > L on bP . By choosing L and ε n n n−1 n n n n n n | | ℜ ∞ \ U inductively in the right way we then see that f = f has all the required properties. n=1 n P 3. Beginning of the proof of Theorem 2.1 3 Let w be a sequence in Bsuch that w 1 as n . If w does not converge then n n n | | → → ∞ (2.2) holds so to prove Theorem 2.1 it is enough to consider only the convergent sequences w . n First, we try to explain the idea of the most important part of the proof. Suppose for a moment that we have a sequence P of convex polytopes with the desired properties and n that there is an increasing sequence R of positive numbers converging to 1 such that n bP R B R B (n IN). n n n−1 ⊂ \ ∈ Let W = U (1 ν,1 + ν) be a small open neighbourhood of z = (0,0, 0,1) in IRM × − ··· where U is a small open ball in IRM−1 centered at the origin and ν > 0 is small. Assume that U 1 ν R B. 0 ×{ − } ⊂ Let π be the orthogonal projection onto IRM−1, so π(x , ,x ) = (x , ,x ). 1 M 1 M−1 ··· ··· For each n, consider C , the part of bP W consisting of the facets of P contained in n n n ∩ W. The projection π is one to one on C and for each of these facets its image under π is n a convex polytope in U that is a cell of a partition of π(C ) into convex polytopes. Call n this partition and notice that as n , π(C ) tends to U. If we remove from each n n L → ∞ cell of its relative interior then we get what we call the skeleton of and denote by n n L L skel( ). Clearly π(skel(P ) C ) = skel( ). Since, by our assumption at the moment, n n n n L ∩ L every sequence w contained in W which meets skel(P ) for all sufficiently large n must n n satisfy (2.2), looking at z = π(w ) we conclude that every sequence z U such that n n n ∞ ∈ z skel( ) for all sufficiently large n must satisfy z z = . The idea now n ∈ Ln n=1| n+1− n| ∞ is to reverse the direction of reasoning. Let R be so close to 1 that U 1 ν R B. 0 0 P ×{ − } ⊂ In a typical induction step of constructing our polytopes the data will be a partition of L IRM−1 into convex polytopes and ρ and r, R < ρ < r < 1. Denote by the union of those 0 C cells of the partition that are contained in U and let be the set of their vertices. We L V will ”lift” to b(rB) by putting V = (π W b(rB))−1( ). We want V to be the set of V | ∩ V vertices of a convex polyhedral surface C such that π(C) = and such that π maps the C facets of C precisely onto the cells of . We will do this in such a way that C stays out C of ρB - for this, the cells of C, and consequently the cells of will have to be sufficiently C small, of size proportional to √r ρ. Then we will construct a convex polytope P such − that C will be a part of its boundary bP and such that ρB IntP P rB. ⊂ ⊂ ⊂ There is a potential problem already at the first step. Namely, the points of V need not be the vertices of a convex surface C. For this to happen we will need two things: will have to be a true Delaunay partition of IRM−1 and the ball U in the definition L of W will have to be sufficiently small so that the part of b(rB) contained in W will be sufficiently flat. 4. A Delaunay tessellation of IRM−1 Perturb the canonical orthonormal basis in IRM−1 a little to get an (M 1)-tuple of − vectors e ,e , ,e in general position so that the lattice 1 2 M−1 ··· M−1 Λ = n e : n Z, 1 i M 1 (4.1) i i i ∈ ≤ ≤ − i=1 (cid:8)X (cid:9) 4 will be generic, and, in particular, no more than M points of Λ will lie on the same sphere. For each point x Λ there is the Voronei cell V(x) consisting of those points of ∈ IRM−1 that are at least as close to x as to any other y Λ, so ∈ V(x)) = y IRM−1: dist(y,x) dist(y,z) for all z Λ . { ∈ ≤ ∈ } M−1 In our case it is easy to see how to get V(0). Consider the finite set E = n e : { j=1 i i 1 n 1,1 i M 1 and for each x E 0 , look at K(x, x 2/2), that is, at − ≤ i ≤ ≤ ≤ − } ∈ \{ } | | P the halfspace which contains the origin and is bounded by the hyperplane passing through x/2 which is perpendicular to x. Then V(0) = K(x, x 2/2). | | x∈E\{0} \ This is a convex polytope. It is known that the Voronei cells form a tessellation of RM−1 and in our case they are all congruent, of the form V(0)+x, x Λ [CS]. ∈ There is a Delaunay cell for each point that is a vertex of a Voronei cell. It is the convex polytope that is the convex hull of the points in Λ closest to that point - these points are all on a sphere centered at this point. In our case, when there are no more than M points of Λ on a sphere, Delaunay cells are (M 1)-simplices. Delaunay cells form a − tessellation of RM−1 [CS]. It is a true Delaunay tessellation , that is, for each cell, the circumsphere of each cell S contains no other points of Λ than the vertices of S. We shall denote by (Λ) the family of all simplices - cells of the Delaunay tessellation for the lattice D Λ. By periodicity there are finitely many simplices S , ,S such that every other sim- 1 ℓ ··· plex of (Λ) is of the form S +w where w Λ and 1 i ℓ. It isthen clear by periodicity i D ∈ ≤ ≤ that there is an η > 0 such that for every simplex S (Λ) in η-neighbourhood of the ∈ D closed ball bounded by the circumsphere of S there are no other points of Λ than the vertices of S. We shall typically replace the lattice Λ by the lattice Λ+q = x+q: x Λ where { ∈ } q IRM−1, or, more generally, by the lattice σ(Λ+ q) where σ > 0 is small. Again, we ∈ shall denote by (σ(Λ+q)) the family of all simplices - cells of the Delaunay tessellation D for σ(Λ+q). These are the simplices of the form σ(S +q) where S (Λ). Passing from ∈ D Λ to σ(Λ + q) everything in the reasoning will change proportionally. In particular, for every simplex S (σ(Λ+q)) in (ση)-neighbourhood of the closed ball bounded by the ∈ D circumsphere of S there will be no other points of σ(Λ+ q) than the vertices of S. We shall also need the notion of the skeleton of the Delaunay tessellation for σ(Λ+q). This is what remains after we remove the interiors of all S (σ(Λ+q)), hence ∈ D skel (σ(Λ+q)) = S IntS = IRM−1 IntS . D \ \ S∈D(σ(Λ+q)) S∈D(σ(Λ+q)) (cid:0) (cid:1) [ (cid:2) (cid:3) (cid:2) [ (cid:3) The author is gratefull to John M.Sullivan who suggested the use of a generic lattice for our purpose here. 5 5. Lifting the lattice from IRM−1 to the sphere Let z, W = U (1 ν,1+ν) and π be as in Section 3. Let Λ IRM−1 be as in (4.1). × − ⊂ Fix R , 0 < R < 1, so largethat U 1 ν R Band assume that R < ρ < r < 1. 0 0 0 0 ×{ − } ⊂ Thepart ofthe sphere b(rB)inW cannow be writtenasa graphofa real analyticfunction, call it ψ , so r b(rB) W = (x,ψ (x)): x U r ∩ { ∈ } where M−1 1/2 ψ (x) = ψ (x , ,x ) = r2 x2 . (5.1) r r 1 ··· M−1 − j (cid:18) j=1 (cid:19) X Note that (grad ψ )(0) = 0, R < r < 1. r 0 The map π maps W b(rB) in a one to one way onto U. We shall ”lift” (σΛ) from U ∩ to b(rB) W by the inverse of this map, that is, by the map x (x,ψ (x)). We want to r ∩ 7→ get a convex polyhedral surface C with vertices w = (v,ψ (v)) where v are the vertices of r those cells of the Delaunay tessellation for σΛ which are contained in U and we want that π maps the facets of the surface C precisely onto the Delaunay cells of σΛ contained in U. Let us describe the conditions for this to happen. Let S be a simplex of the Delaunay tessellation for σΛ. Let v , ,v be the vertices of S. We want that the simplex with 1 M ··· vertices w = (v ,ψ(v )), 1 j M, is a facet of a convex poyhedral surface. For this j j j ≤ ≤ to happen, all other points w = (v,ψ (v)), v σΛ U, v = v , ,v , must lie in the r 1 M ∈ ∩ 6 ··· open halfspace bounded by the hyperplane Π through w , 1 j M, which contains the j ≤ ≤ origin, that is, they must lie on b(rB) outside the ”small” sphere Γ = Π b(rB). Since ∩ π W b(rB) is one to one, this happens if and only if the points v σΛ which are the | ∩ ∈ vertices of the Delaunay cells of σΛ contained in U and are different from v , ,v , are 1 M ··· outside the projection π(Γ), an ellipsoid in IRM−1. Aswe shallsee, thiswillhappen for allsuch simplicesS iftheball U IRM−1 centered ⊂ at the origin will be small enough so that the the gradient of ψ and thus the Lipschitz r constant of ψ will be small enough on U. The choice of U will depend only on η from r Section 4 and the same reasoning will work for any σ > 0. LEMMA 5.1 Let π:IRM IRM−1 be the standard projection, π(x , ,x ) = (x , 1 M 1 → ··· ··· ,x ). Let Λ be the lattice in IRM−1 as in (4.1) and let η > 0. There is a constant M−1 ω > 0 such that for every σ > 0 the following holds. Let S IRM−1 be a simplex belonging ⊂ to (σΛ). Suppose that ψ is a Lipschitz function in a neighbourhood of S with Lipschitz D constant ω. Let v , ,v be the vertices of S and let w , ,w be the points in IRM 1 M 1 M ≤ ··· ··· given by w = (v ,ψ(v )), 1 j M. Let Π be the hyperplane in IRM containing the j j j ≤ ≤ points w , ,w and let Γ be the sphere in Π, containing these points (that is, let Γ be 1 M ··· the circumsphere of the (M 1)-simplex in Π with vertices w , ,w ). Then π(Γ) is 1 M − ··· contained in the (ση)-neighbourhood of the circumsphere of the simplex S. 6. Proof of Lemma 5.1 Let S (Λ) and let η > 0. If we replace ψ with ψ +c where c is a constant, Π will ∈ D change to Π+(0,c), Γ to Γ+(0,c) and consequently π(Γ) will not change. Thus, π(Γ) remains unchanged if we subtract ψ(v ) from each ψ(v ), 1 j M. Thus, π(Γ) will be M j ≤ ≤ 6 determined precisely once we know β = ψ(v ) ψ(v ), ,β = ψ(v ) ψ(v ). 1 1 M M−1 M−1 M − ··· − We shall show that π(Γ) changes continuously with (β , ,β ) near (0,0, ,0) if 1 M−1 ··· ··· w = (v ,β ), ,w = (v ,β ) and w = (v ,0). Note that when β = = 1 1 1 M−1 M−1 M−1 M M 1 ··· ··· β = 0, then Γ = π(Γ) is the circumsphere of S in IRM−1. Let w = (w , w ,1) M−1 0 01 0,M−1 ··· be the vector in IRM perpendicular to Π whose last component equals 1. So w must be 0 perpendicular to w w , 1 j M 1, so < w w w >= 0 (1 j M 1) j M j M 0 − ≤ ≤ − − | ≤ ≤ − which, if v = (v , ,v ), 1 j M 1, is the system of linear equations j j1 j,M−1 ··· ≤ ≤ − (v v )w + +(v v )w = β (1 j M 1). j1 M1 01 j,M−1 M,M−1 0,M−1 j − ··· − − ≤ ≤ − This is a system of M 1 linear equations for M 1 unknowns w , ,w whose 01 0,M−1 − − ··· matrix is nonsingular, since, S being a (M 1)-simplex, the vectors v v , 1 j j M − − ≤ ≤ M 1, are linearly independent. Its solution depends linearly on (β , ,β ). When 1 M−1 − ··· β = = β = 0 the solution is the zero vector. In this case w = (0, ,0,1). Let 1 M−1 0 ··· ··· z = (z , ,z ) be the center of the sphere in Π that contains w , ,w . Then z is in 1 M 1 M ··· ··· Π so < z w w >= 0. (6.1) M 0 − | Further, for each i, 1 i M 1, z isat equal distance from w and w , so z iscontained i M ≤ ≤ − in the hyperplane in IRM that passes through the midpoint of the segment joining w and i w , and is perpendicular to this segment, so z must satisfy M < [z (w +w )/2] [w w ] >= 0. i M i M − | − Thus, < z [w w ] >= (1/2) < [w +w ] [w w ] > (1 i M 1). i M i M i m | − | − ≤ ≤ − Together with (6.1) this becomes the following system of linear equations for z , ,z : 1 M ··· z (v v )+ +z (v v )+z β = ( w 2 w 2)/2 (1 i M 1) 1 i1 M1 M−1 i,M−1 M,M−1 M i i M − ··· − | | −| | ≤ ≤ − z w + +z w +z = v w + +v w . 1 01 M−1 0,M−1 M M1 01 M,M−1 0,M−1 ··· ··· Its matrix v v , ,v v ,β 11 M1 1,M−1 M,M−1 1 − ··· − ....   v v , ,v ,β M−1,1 M1 M−1,M−1 M−1  − ···  w , w , 1   01 ··· 0,M−1    is nonsingular for β = = β = 0 when w = = w = 0. The matrix de- 1 M−1 01 0,M−1 ··· ··· pends continuously on(β , β ) and so do the right sides (1/2)( v 2 v 2+β2), 1 1 ··· M−1 | i| −| M| i ≤ i, M 1, and, since w depends continuously on (β , ,β ), also v w + + 0 1 M−1 M1 01 ≤ − ··· ··· v w depends continuously on (β , ,β ). So the solution z = (z , ,z ), M,M−1 0,M−1 1 M−1 1 M ··· ··· the center of the sphere Γ, depends continuously on (β , ,β ) near (0,0, ,0) and 1 M−1 ··· ··· so does its radius z w = (z v )2+ +(z v )2+z2 )1/2. Recall that Π | − M| 1− 1 ··· M−1 − M−1 M (cid:0) 7 passes through w = (v ,0)anditsperpendicular directionw changes continuously with M M 0 (β , ,β ) so Π changes continuously with(β , ,β ). We have seen that the cen- 1 M−1 1 M−1 ··· ··· ter z of the sphere Γ in Π and its radius also change continuously with (β , ,β ) near 1 M−1 ··· (0,0, 0). Thus, π(Γ) changes continuously with (β , ,β ) near the origin where 1 M−1 ··· ··· π(Γ) = Γ is the circumsphere of S when β = β = β = 0. Thus, π(Γ) is contained 1 2 M−1 ··· in the η-neighbourhood of the circumsphere of of the simplex S in IRM−1 provided that ψ(v ) ψ(v ), ,ψ(v ) ψ(v ) are small enough. If ψ is a Lipschitz function with 1 M M−1 M − ··· − the Lipschitz constant ω then ψ(v ) ψ(v ) ω v v , 1 i M 1, so there i M i M | − | ≤ | − | ≤ ≤ − is an ω such that if ψ is a Lipschitz function with the Lipschitz constant not exceeding ω then π(Γ) is contained in the η-neighbourhood of the circumsphere of the simplex S. Recall that every simplex in (Λ) is of the form S +x, 1 i ℓ, x Λ. Repeating the i D ≤ ≤ ∈ reasoning above for each S , 1 i ℓ, we get the Lipschitz constant that works for every i ≤ ≤ simplex S in (Λ). This completes the proof for σ = 1. D Now, let σ > 0 be arbitrary and let S IRM−1 be a simplex in (σΛ). Let ψ be a ⊂ D Lipschitz function with Lipschitz constant not exceeding ω in a neighbourhood of S, so its graph is given by x = ψ(x , ,x ). Introduce new coordinates X , ,X in IRM M 1 M−1 1 M ··· ··· by x = σX , 1 j M. In new coordinates we have σX = ψ(σX , ,σX ) so j j M 1 M−1 ≤ ≤ ··· X = Ψ(X , X ) = (1/σ)ψ(σX , σX ). Bothψ andΨareLipschitzfunctions M 1 M−1 1 M−1 ··· ··· with the same Lipschitz constants, so in new coordinates Ψ is a Lipschitz function in a neighbourhood of S, which, in new coordinates, belongs to (Λ). Thus, applying the first D part of the proof we see that in new coordinates π(Γ) is contained in the η-neighbourhood of the circumsphere of S. In follows that in old coordinates π(Γ) is contained in the (ση)-neighbourhood of the circumsphere of S. This completes the proof. 7. Polyhedral convex surface contained in a spherical shell Let η > 0 be as in Section 4 and let ω be the one given by Lemma 5.1. Let again W = U (1 ν,1 + ν) where ν > 0 is small, U is a small open ball centered at the × − origin in IRM−1 and let R < 1 be so large that U 1 ν R B. For every r, R < 0 0 0 ×{ − } ⊂ r < 1, W b(rB) = (x,ψ (x): x U where the function ψ is as in (5.1). We have r r ∩ { ∈ } (grad(ψ ))(x) = (r2 x 2)−1/2x (x U) so we may, passing to a smaller U if necessary, r − −| | ∈ assume that (gradψ )(x) ω (x U, R < r < 1) so that for each r, R < r < 1, ψ is r 0 0 r | | ≤ ∈ a Lipschitz function on U with Lipschitz constant not exceeding ω. Let Λ be as in (4.1), let σ > 0 be small and let R < r < 1. Let ψ be as in (5.1). 0 r Then x Ψ = (x,ψ (x)) is a one to one map from U onto W b(rB). We now look at r r 7→ ∩ the points Ψ (x),x (σΛ) U and want to see them as vertices of a convex polyhedral r ∈ ∩ hypersurface in IRM. Consider a simplex S (σΛ) which is contained in U. Let v , ,v be its vertices. 1 M ∈ D ··· We can extend the restriction of the function ψ to this set of vertices to a function ϕ on r r all S by putting M M M ϕ α v = α ψ (v ) (0 α 1, 1 j M, α = 1) r j j j r j j j ≤ ≤ ≤ ≤ j=1 j=1 j=1 (cid:0)X (cid:1) X X to get an affine function ϕ on S so that x Φ (x) = (x,ϕ (x)) is an affine map mapping r r r 7→ S to Φ (S), the simplex with vertices Ψ (v ), ,Ψ (v ). We do this for every simplex r r 1 r M ··· 8 S (σΛ) that is contained in U. Thus, we get a piecewise linear function ϕ on the r ∈ D union of the simplices S (σΛ)contained in U and so the union C (σ) of all these Φ (S), r r ∈ D the graph of the function ϕ , is then a polyhedral surface in IRM. We shall show that the r function ϕ is convex so that C (σ) is a convex polyhedral surface. Later we shall show r r that the part of C (σ) contained in W = U (1 ν,1+ν) with U being a ball in RM−1 r 0 0 0 ∩ − centered at the origin, strictly smaller than U, is a part of the boundary bP of a suitable convex polytope P. Given S (σΛ), S U, let Π be the hyperplane in IRM that contains Φ (S). r ∈ D ⊂ Then Π b(rB) is the sphere in Π and which is the circumsphere of Φ (S) and which was r ∩ denoted by Γ in Section 5. By Lemma 5.1, π(Γ) is contained in the (ση)-neighbourhood of the circumsphere of S in IRM−1. We know that the ση-neighbourhood of the closed ball in IRM−1 bounded by the circumsphere of S contains no other points of σΛ than the vertices of S which implies that all points of Ψ (U (σΛ)) other than the vertices of Φ (S) r r ∩ lie outside of the small ”spherical cap” that Π cuts out of b(rB), that is, outside of the ”small” part of b(rB) bounded by Γ. This shows that all other vertices of the simplices in C (σ) that are not the vertices of Φ (S) are contained in the open halfspace of IRM r r bounded by Π that contains the origin. Thus, Φ (S) is a facet of C (σ). Since this holds r r for every S (σΛ),S U, it follows that the surface C (σ) is convex. r ∈ D ⊂ The simplices Φ (S) where S (σΛ), S U, have all their vertices on b(rB). We r ∈ D ⊂ want to estimate how far into rB they reach. To do this, we need the following PROPOSITION 7.1 Let 0 < r < 1, let a b(rB) and let A b(rB) be a set such ∈ ⊂ that x a γ for all x A where γ < r. Then the convex hull of A misses ρB where | − | ≤ ∈ ρ = r γ2/r. − Proof. A is contained in x b(rB): x a γ . With no loss of generality assume that { ∈ | − | ≤ } a = (r,0, ,0). Then A x b(rB): (x r)2+x2+ +x2 γ2 x b(rB): r2 ··· ⊂ { ∈ 1− 2 ··· M ≤ } ⊂ { ∈ − 2x r + r2 γ2 = x b(rB): 2r2 2x r < 2γ2 = x b(rB): x > (r2 γ2)/r 1 1 1 ≤ } { ∈ − } { ∈ − } x rB: x > (r2 γ2)/r . The last set is a convex set that contains A and misses ρB 1 ⊂ { ∈ − } which completes the proof. Denote by d the length of the longest edge of simplices in (Λ) so that σd is the length D of the longest edge of the simplices in (σΛ). Since ψ is a Lipschitz function with the r D Lipschitz constant not exceeding ω the length of the longest edge of the simplices Φ (S) r where S (σΛ), S U, does not exceed √1+ω2σd. Now, we use Proposition 7.1. If ∈ D ⊂ R < r < 1 then r γ2/r > r γ2/R . Thus, putting 0 0 − − (1+ω2)d2 λ = R 0 we get the following PROPOSITION 7.2 If R < r < 1 then the simplices Φ (S) where S (σΛ), S U, 0 r ⊂ D ⊂ miss ρB where ρ = r σ2λ. − 8. A convex polytope with a prescribed part of the boundary We keep the meaning of R ,U,d and λ. Recall that U is an open ball in IRM−1 0 9 centered at the origin. Let µ be its radius. Let 0 < µ < µ < µ < µ < µ and let 0 1 2 3 U = x IRM−1: x < µ , W = U (1 ν,1+ν), 0 i 3. i i i i { ∈ | | } × − ≤ ≤ Choose σ > 0 so small that 0 σ d < min µ µ ,µ µ ,µ µ ,µ µ . (8.1) 0 3 3 2 2 1 1 0 { − − − − } Then, since the maximal edge length of simplices in (σΛ) equals σd it follows that if D 0 < σ < σ then 0 - the simplices S (σΛ) that meet U are contained in U 0 1 ∈ D - the simplices S (σΛ) that are contained in U cover U . 2 ∈ D PROPOSITION 8.1 There is a κ > 0 such that whenever R R 1 and R < R′ < 0 ≤ ≤ R + κ then each hyperplane in IRM which meets W R′B RB and misses W RB 2 3 ∩ \ ∩ misses RB. (cid:0) (cid:1) Proof Suppose that there is no such κ > 0. Then there are a sequence R , R n 0 ≤ R 1 (n IN), and a sequence x W , such that x > R (n IN) and such that n n 2 n n ≤ ∈ ∈ | | ∈ x R 0 as n , and for each n a hyperplane H through x which misses n n n n | | − → → ∞ W R B and meets R B W . Since x IR 0 as n we may, passing to 3 n n 3 n n ∩ \ | | − → → ∞ subsequences if necessary, with no loss of generality assume that R converges to an R n and x converges to x b(RB) W . Since for each n, H misses W R B it follows n 2 n 3 n ∈ ∩ ∩ that H converges to H, the hyperplane through x tangent to b(RB) at x. In particular, n H (RB W ) is empty, so for sufficiently large n, H (R B W ) must be empty, a 3 n n 3 ∩ \ ∩ \ contradiction. This completes the proof. With no loss of generality, passing to a smaller σ if necessary, we may assume that 0 σ2λ < κ. Suppose now that 0 < σ < σ and let R ρ < r < 1 where ρ = r σ2λ. 0 0 0 ≤ − We know that the union C (σ) of the simplices Φ (S) where S (σΛ),S U, is a r r ∈ D ⊂ convex polyhedral surface which, by Proposition 7.2, is contained in rB ρB. Each of these \ simplices Φ (S) is contained in a hyperplane H. We want that these hyperplanes miss r ρB. Note that by (8.1) the simplices in (σΛ), contained in U cover U . So the function 3 D ϕ is well defined on U and its graph C (σ) W is contained in W (rB ρB). The r 3 r 3 3 ∩ ∩ \ function ϕ is piecewise linear and convex. Thus, if S (σΛ) meets U then, by (8.1), r 2 ∈ D S U and by the convexity of ϕ , the graph of ϕ U lies on one side of the hyperplane 3 r r 3 ⊂ | H that contains Φ (S) which, in particular, implies that H misses W ρB and thus, by r 3 ∩ Proposition 8.1, H misses ρB. This shows that the part of C (σ) contained in W can r 2 be described in terms of the hyperplanes that miss ρB. So we find x , ,x bB and 1 n ··· ∈ α , ,α , ρ < α r (1 i n), such that 1 n i ··· ≤ ≤ ≤ G = x B: < x x > α ,1 i n . 1 i i { ∈ | ≤ ≤ ≤ } is a convex set containing ρB in its interior, and is such that W bG = W C (σ). 2 1 2 r ∩ ∩ PROPOSITION 8.2 Let R < r < 1, let 0 < σ < σ , and let ρ = r σ2λ > R , There 0 0 0 − is a convex polytope P which contains ρB in its interior, such that bP rB ρB, and such ⊂ \ that every Φ (S) where S (σΛ), S U , is a facet of P. r 1 ∈ D ⊂ Proposition 8.2 implies in particular, that W skel(P) = Φ U skel( (σΛ)) 0 r 0 ∩ ∩ D (cid:0) (cid:1) 10