The separating semigroup of a real curve
Kummer, Mario;Shaw, Kristin
2017-07-25 00:00:00
MARIO KUMMER AND KRISTIN SHAW Abstract. We introduce the separating semigroup of a real algebraic curve of dividing type. The elements of this semigroup record the possible degrees of the covering maps obtained by restricting separating morphisms to the real part of the curve. We also introduce the hyperbolic semigroup which consists of elements of the separating semigroup arising from morphisms which are compositions of a linear projection with an embedding of the curve to some projective space. We completely determine both semigroups in the case of maximal curves. We also prove that any embedding of a real curve of dividing type to projective space of suciently high degree is hyperbolic. Using these semigroups we show that the hyperbolicity locus of an embedded curve is in general not connected. Resume. Nous introduisons le semi-groupe s eparant d'une courbe alg ebrique r eelle s eparante. Les el ements de ce semi-groupe gardent trace des degr es pos- sibles des rev^ etements du cercle obtenus par restriction a la partie r eelle de la courbe des morphismes s eparants. Nous introduisons aussi le semi-groupe hyperbolique, compos e des el ements du semi-groupe s eparant provenant des morphismes qui sont la composition d'une projection lin eaire et d'un plonge- ment de la courbe dans un espace projectif. Nous d eterminons les deux groupes dans le cas des courbes maximales. Nous d emontrons aussi que tout plongement d'une courbe r eelle s eparante de degr e susamment grand est hyperbolique. En utilisant ces semi-groupes, nous montrons que le lieu hyperbolique d'une courbe plong ee n'est en g en eral pas connexe. 1. Introduction Here a curve will always be a non-singular projective and geometrically irre- ducible algebraic curve over R. Furthermore, we always use P to denote the projective space de ned over R. For a variety V de ned over R we denote by V (R) and V (C) the real and complex points of V , respectively. A basic fact concerning the classi cation of real algebraic curves, or real Riemann surfaces, is the following dichotomy which goes back to Klein [Kle23, x23]: If X is a curve, then the set X (C)rX (R) has either one or two connected components. If the latter is the case, X is called of dividing type. Curves of dividing type are often called curves of type I or separating in the literature. If there exists a morphism 1 1 1 f : X ! P with the property f (P (R)) = X (R), then X (C) r X (R) can not be 1 1 connected. This is since P (C) r P (R) has two connected components and their preimages under f yield two connected components of X (C) r X (R). Therefore, such a morphism is also called separating since it certi es that X is of dividing type. It follows from the work of Ahlfors [Ahl50, x4.2], though proved in a dierent context, that conversely every separating curve admits a separating morphism. 2010 Mathematics Subject Classi cation. Primary: 14P99, 14H50; Secondary: 14H51. The second author's research was supported in part by a postdoctoral fellowship from the Alexander von Humboldt foundation. The project was completed while both authors were at the Max-Planck-Institute for Mathematics in the Sciences, Leipzig. We thank the institute for the excellent working conditions. arXiv:1707.08227v2 [math.AG] 7 Feb 2019 2 MARIO KUMMER AND KRISTIN SHAW Rokhlin [Rok78] used the existence of separating morphisms given by pencils as a certi cation of certain real plane curves being of dividing type. In Mikhalkin's [Mik00] study of extremal topology of real curves in (C ) , he showed that the logarithmic Gauss map of a simple Harnack curve is separating. Conversely, it was shown by Passare and Risler [PR11] that if a planar curve has separating logarithmic Gauss map, then it is a Harnack curve. The existence of separating morphisms and their properties have been considered by several authors [Hui01, Gab06, CH13, Cop13]. For example, Gabard [Gab06] showed that every separating curve X admits a separating morphism of degree at g+r+1 most where g is the genus and r the number of connected components of X (R). Later Coppens [Cop13] constructed, for every value of k between r and g+r+1 , a separating curve X of genus g with X (R) having r components such that k is the smallest possible degree of a separating morphism. In this work, we take a complementary approach. Namely, we x a curve X of dividing type of genus g and study the set of all separating morphisms X ! P . Let X (R) consist of r connected components X ; : : : ; X . Since X is of dividing 1 r type r + g must be odd. A separating morphism f is always unrami ed over X (R) [KS15, Theorem 2.19]. Therefore, the restriction of f to each X is a covering map of P (R). This implies that the degree of a separating morphism is at least r. Let N denote the positive integers, so 0 62 N. We denote by d (f ) 2 N the 1 r degree of the covering map X ! P and set d(f ) := (d (f ); : : : ; d (f ))2 N . For i 1 r d = (d ; : : : ; d ) 2 N we let jdj := d . Our rst main object of interest is the 1 r i i=1 set of all such degree partitions. De nition 1.1. The set r 1 Sep(X ) = fd(f ) 2 N j f : X ! P separatingg is called the separating semigroup. Since we assume that X is a separating curve, the set Sep(X ) is always non- empty. The term semigroup is justi ed by the fact that this set turns out to be closed under componentwise addition, see Proposition 2.1. In Corollary 3.4, we show that d + Z Sep(X ) for every d 2 Sep(X ) with jdj suciently large. Our main technique is making use of interlacing sections : Two sections s and s of a line bundle on X are called interlacing if they both have only simple and real zeros and if on each component X between each two consecutive zeros of s there is exactly one zero of s . This notion generalizes the notion of interlacing polynomials to sections of line bundles on algebraic curves. The concept of interlacing polynomials has attracted a lot of attention since Marcus, Spielman and Srivastava used it to solve the Kadison{Singer problem as well as to nd bipartite Ramanujan graphs of all degrees [MSS15a, MSS15b]. We will make 1 0 use of the fact that the morphism to P de ned by s and s is separating if and only if s and s are interlacing. This is proved in Lemma 2.10. We also study the subset Sep(X ) consisting of all degree partitions that are realized by a separating morphism which is actually a linear projection of some embedding of X in projective space from a linear space disjoint from X . This is motivated by the following de nition from [SV14]. n n De nition 1.2. Let X P be a curve and E P be a linear subspace of codimension two such that E \ X = ;. Then X is hyperbolic with respect to E if the linear projection : X ! P from E is separating. Following the terminology of [SV14] we will call such embedded curves hyperbolic. These curves are a generalization of planar curves de ned by hyperbolic polynomials in three homogeneous variables. In general hyperbolic polynomials have attracted THE SEPARATING SEMIGROUP OF A REAL CURVE 3 interest in dierent areas of mathematics like partial dierential equations [G ar51, H or05], optimization [Gul97, Ren06] and combinatorics [COSW04, Br a07]. Whereas at rst sight hyperbolicity might seem to be a rare phenomenon, it is actually quite ubiquitous in the case of curves as justi ed by the next theorem. It says that for a given separating curve X , every embedding of high enough degree turns out to be hyperbolic. We rst remark any divisor D on a curve X with degree k > 2g is very ample by [Har77, Corollary 3.2 b)], and therefore the map X ! P(L (D) ) is an embedding. Theorem 1.3. Let X be a curve of dividing type of genus g. There exists a k > 2g with the following property: For any divisor D of degree at least k the corresponding embedding of X to P(L (D) ) is hyperbolic. De nition 1.4. The hyperbolic semigroup Hyp(X ) is the set of all elements of Sep(X ) where the corresponding f can be chosen to be the composition of a linear projection with an embedding of X to some P , where the center of the projection is disjoint from X . n 3 Remark 1.5. Replacing P by P in De nition 1.4 results in an equivalent con- dition [KS15, x2]. Also, in the de nition of Hyp(X ), one could equivalently just require f to be separating and f O 1 (1) to be very ample. The set Hyp(X ) also turns out to be a semigroup, see Proposition 2.1. In Proposition 2.12, we give an equivalent criterion for a curve to be hyperbolic in terms of the linking numbers of its components with the linear subspace from which we project. For a planar curve X of dividing type, a pencil of curves is said to be totally real with respect to X if every curve in the pencil intersects X in only real points. In [Tou13], Fiedler-Le Touz e asks if for every planar curve of dividing type there exists a totally real pencil. Using our techniques we can answer this question in the armative even when the base points of the pencil are not contained in the curve. In the next theorem, we let V denote the subvariety of P de ned by a collection of homogeneous polynomials in R[x; y; z]. Theorem 1.6. If X P is a curve of dividing type, then there exists an integer k such that for any k k there are homogeneous polynomials f; g 2 R[x; y; z] of degree k for which V (f; g)\ X (R) = ; and such that V (f + g) intersects X only in real points for every ; 2 R not both zero. A curve X of genus g is called an M-curve if X (R) has r = g + 1 connected components. Every M -curve is of dividing type. In the case of M -curves we give a complete description of both the separating and hyperbolic semigroups. Theorem 1.7. Let X be an M-curve. a) If g = 0, then Hyp(X ) = Sep(X ) = N. 2 2 b) If g = 1, then Sep(X ) = N and Hyp(X ) = N rf(1; 1)g. g+1 g+1 g+1 c) If g > 1, then Sep(X ) = N and Hyp(X ) = fd 2 N : d g + 3g. i=1 Finally, in Section 5 we study the hyperbolicity locus of an embedded curve in an example. Given an embedded curve X in P , Shamovich and Vinnikov asked if the subset of the Grassmannian Gr(n 1; n + 1) corresponding to the linear spaces from which the projection of X is separating is connected [SV14]. In Example 5.1, using the hyperbolic semigroup we construct an example where the answer is negative. In fact, this example is a member of a family of curves constructed independently by Mikhalkin and Orevkov [MO18, Theorem 3]. Their construction immediately implies that there exists an M -curve in P of genus g such that its hyperbolicity locus consists of g + 1 connected components. 4 MARIO KUMMER AND KRISTIN SHAW 2. The separating and hyperbolic semigroups We begin by showing that the sets Sep(X ) and Hyp(X ) are indeed semigroups. Proposition 2.1. Let X be a curve of dividing type. Then both Sep(X ) and Hyp(X ) are semigroups. Proof. Let f ; f : X ! P be two separating morphisms. Let X be one of the 1 2 + connected components of X (C)rX (R). Without loss of generality we may assume that f (X ) = f (X ) = H is the upper half-plane. Identify P (C) with C[f1g 1 + 2 + + 1 1 and let : P (C) ! P (C) be a M obius transformation sending the circle jzj = 1 to 1 1 P (R). De ne the map g : X ! P (C) by 1 1 g(x) = (f (x)) (f (x)): 1 2 1 1 The preimage g (z) of any point z with jzj = 1 is contained in P (R). Moreover, this preimage consists of exactly d (f ) +d (f ) points on X for i = 1; : : : ; r. Then i 1 i 2 i the composition f = g(x) is a separating map which satis es d(f ) = d(f )+d(f ). 1 2 This proves that Sep(X ) is a semigroup. To show that Hyp(X ) is a semigroup, suppose that L = f O 1 (1) and L = 1 P 2 f O 1 (1) are both very ample. Then the line bundle f O 1 (1) = L L , where f P P 1 2 is de ned as above, is also very ample. Therefore Hyp(X ) is also a semigroup. 1 1 When X = P we have 1 2 Hyp(P ). Therefore, the next corollary is an imme- diate consequence of Proposition 2.1. 1 1 Corollary 2.2. We have Sep(P ) = Hyp(P ) = N. Example 2.3. In fact, we have Sep(X ) = N if any only if X = P , since a map of degree 1 is an isomorphism. Remark 2.4. For a dividing curve X of genus g, every d 2 Sep(X ) with jdj 2g +1 is also in Hyp(X ). This is because every line bundle L on X with deg(L ) 2g + 1 is very ample [Har77, Corollary 3.2 b)]. Example 2.5. Let X be a hyperelliptic curve of genus g = 2 given by x = p(x ; x ) 0 1 where p is a positive de nite form of degree six. We consider X to be a subvariety of the weighted projective space P (1; 1; 3), so that it is non-singular. Then the canonical map is separating. Since r + g must be odd and f is unrami ed over the real points, we conclude that r = 1. Thus, we have 2 2 Sep(X ). By [Ahl50, x4.2] we also have 3 2 Sep(X ) but we do not have 1 2 Sep(X ) because the curve is not rational. Thus, we have Sep(X ) = N . Remark 2.6. In general, the separating and hyperbolic semigroups do not only depend on r and g. For example, it is possible to construct a hyperelliptic curve of genus three whose canonical map is separating. In that case we have r = 2. But there are also separating curves of genus three with r = 2 that are not hyperelliptic and therefore do not admit a separating morphism of degree two. Example 2.7. Let g = 1 and suppose that r = 2. Then there is an automorphism of X that sends X to X . Thus, Sep(X ) is stable under the action of the symmetric 1 2 group S . From embeddings to P we obtain (2; 1); (1; 2) 2 Hyp(X ). One also has (1; 1) 2 Sep(X ) r Hyp(X ). Example 2.8. In this example we show that the separating semigroup of a planar hyperbolic curve X of degree k is not symmetric for k 4. This version of the argument was suggested by Erwan Brugall e following our original approach for k = 4. The number of connected components of X (R) is r = d e. A linear system of rank 2 on a curve X of genus g 3 is unique, [Nam79, x2.3] or [ACGH85, A.18]. So we can label the connected components X ; : : : ; X from the innermost oval X 1 r 1 THE SEPARATING SEMIGROUP OF A REAL CURVE 5 to the outermost oval X if k is even. If k is odd then X is the outermost r r 1 and X is the unique pseudoline. By our hyperbolicity assumption (2; 2; : : : ; 2) 2 Hyp(X ) Sep(X ) if k is even and (2; 2; : : : ; 2; 1) 2 Hyp(X ) Sep(X ) if k is odd. The gonality of a planar curve X of degree k is k 1 and moreover every map 1 2 1 f : X ! P of degree k 1 is induced by a projection P ! P whose center is a point on X , [Nam79, x2.3] or [ACGH85, A.18]. Therefore, we have (1; 2; : : : ; 2) 2 Sep(X ) if k is even and, if k is odd (1; 2; : : : ; 2; 1) 2 Sep(X ). However, no other permutation of these degree sequences is possible, since a projection whose center is not on the innermost oval of X would not be a separating morphism. Therefore, the semigroup Sep(X ) is in general not preserved under the action of the symmetric group. Moreover, it follows that the semigroup Sep(X ) is an invari- ant of the curve X which is capable of distinguishing which connected component of X (R) is the innermost oval of a hyperbolic embedding to P when g 3. 2.1. Separating morphisms and interlacing sections. The following de ni- tion generalizes the interlacing property for polynomials [Fis06] to sections of line bundles. De nition 2.9. Let L be a line bundle on X . Let s ; s 2 (X;L ) be two global 0 1 sections that both have only simple and real zeros. We say that s and s interlace 0 1 if each connected component of X (R) rfP j s (P ) = 0g contains exactly one zero of s and vice versa. Lemma 2.10. Let L be a line bundle on X and let s ; s 2 (X;L ) be global 0 1 sections that generate L and that both have only simple and real zeros. Then the morphism X ! P given by x 7! (s (x) : s (x)) is separating if and only if s and 0 1 0 s are interlacing. Proof. Assume that s and s interlace. If the map is not separating, then there is 0 1 a 2 R such that s + s has a double zero on X (R). Since s and s generate 0 1 0 1 L , the section s + s does not vanish on any zero of s for any . Thus, because 0 1 1 s has exactly one zero on each connected component of X (R) rfP j s (P ) = 0g, 0 1 a double root of s + s is impossible. Therefore, interlacing implies separating. 0 1 Conversely, assume that the morphism under consideration is separating. It follows immediately that all zeros of s and s are real. The other properties of 0 1 interlacing sections follow from the fact that the restriction of separating morphisms to the real part is unrami ed [KS15, Theorem 2.19]. To verify the interlacing property of a pair of sections we have the following sucient criterion which we will use later on. Proposition 2.11. Let L be a line bundle on X and let s ; s 2 (X;L ) be 0 1 global sections that generate L . Let s have only simple and real zeros. Let I be the set of indices i such that s has more than one zero on X . Assume that 0 i for each i 2 I there is exactly one zero of s on each connected component of X rfP j s (P ) = 0g. Then s and s are interlacing and the morphism X ! P i 0 0 1 given by x 7! (s (x) : s (x)) is separating. 0 1 Proof. As in the proof of the preceding lemma we show that there is no 2 R such that s + s has a double zero on X (R). Indeed, as shown in the proof of the 0 1 preceding lemma, two zeros on one of the X for i 2 I cannot come together as varies. Since there is only one zero of s on the other components, the claim is also true for those. 2.2. Conditions for hyperbolic morphisms. We rst point out that a curve being hyperbolic with respect to some linear space is a purely topological property. 6 MARIO KUMMER AND KRISTIN SHAW Figure 1. The twisted cubic drawn in green is hyperbolic with respect to the red line. We begin by recalling linking numbers of spheres embedded in the n-dimensional sphere S . For the general de nition of linking numbers and more detailed infor- mation we refer to [Pra07]. Suppose that X and Y are disjoint embedded oriented spheres in S of dimensions p and q respectively where n = p + q + 1. Consider the fundamental cycles [X ] and [Y ] as cycles in the integral homology of S . There exists a chain W whose boundary is [X ]. The linking number lk(X; Y ) is de ned to be the intersection number of W and [Y ]. n 1 n Now let K P (R) be the image of an embedding of S and L P (R) be n n a linear subspace of codimension 2. Let : S ! P (R) be an unrami ed 2 to 1 1 n 1 covering map. Notice that (L) is a sphere of dimension n 2 in S and (K ) is either an embedded circle or two embedded circles. De ne the linking number of n 1 1 n 1 K and L in P (R) to be the linking number of (K ) and (L) in S if (K ) is a single connected component and de ne it to be the sum of the linking numbers of K with L and K with L if K [ K = (K ). Now we are able to give a 1 2 1 2 topological characterization of hyperbolic curves in terms of linking numbers. n n Proposition 2.12. Let X P be a curve and E P be a linear subspace of dimension n 2 with X \ E = ;. Then X is hyperbolic with respect to E if and only if deg(X ) = j lk(X ; E(R))j. When X is hyperbolic with respect to E, i=1 then the tuple (j lk(X ; E(R))j; : : : ;j lk(X ; E(R))j) 1 r is the element in Hyp(X ) corresponding to the projection from E. Proof. The curve X is hyperbolic with respect to E if and only if every hyperplane H P that contains E intersects X in deg(X ) many distinct real points. Let n n : S ! P (R) be an unrami ed 2 to 1 covering map. For any choice of a hyper- n 1 plane H P that contains E, the preimage X = (E) is a sphere of dimension 1 1 n 2 inside (H ) which is a sphere of dimension n 1. Let W (H ) be a hemisphere whose boundary is (E). If X is hyperbolic with respect to E, then the absolute values of the linking numbers lk(X ; E(R)), which are the intersection numbers of the (X ) with W , i i sum up to deg(X ). Conversely, if the intersection number of W with the preimage of X (R) is deg(X ), then H has (at least) deg(X ) many real intersection points with X . The nal statement about the element of Hyp(X ) arising from the projection from E is immediate. THE SEPARATING SEMIGROUP OF A REAL CURVE 7 Example 2.13. Let Q P be the quadratic surface de ned by the equation 2 2 2 2 x + y = z + w . Its real part Q(R) is the hyperboloid. For a curve X contained in Q, we can describe a topological condition for X to be hyperbolic with respect to the line E given by x = y = 0. 1 1 The hyperboloid Q(R) is homeomorphic to the torus S S and taking a real line from each of the two rulings of Q gives a pair of generators of H (Q(R)) Z Z. We will assume that these lines are oriented in the upwards z direction in the ane chart w = 1. For each hyperplane H containing the line E we have [H \ Q(R)] = (1; 1) 2 H (Q(R)), up to switching the orientation of H \ Q(R). If X Q, then a connected component of X (R) realizes either the trivial class in H (Q(R)) or the class (p; q) for p; q coprime integers. Otherwise, the connected components of X (R) would have non-trivial intersections contradicting the fact that X is non-singular. In order for X to be hyperbolic with respect to E, no component of X (R) can realize the trivial class. This is because if a connected component X is trivial in homology, it is the boundary of a disc contained in Q which would not intersect E and so d = j lk(X ; E(R))j = 0: If furthermore deg(X ) = r (p + q), i i then X must be hyperbolic with respect to E. For example, we can construct a curve X of degree 2k in P which is hyperbolic with respect to E in the following way. Let X be the complete intersection of Q with a hypersurface S which is a small perturbation of the union of hyperplanes H with equations of the form z + a w for i = 1; : : : ; k. Then X (R) consists of i i k connected components, and each one realizes the class (1; 1) 2 H (Q(R)) up to changing the orientations of the connected components of X (R). By the above remark, the curve X is hyperbolic with respect to E and thus (2; : : : ; 2) 2 Hyp(X ). See the left hand side of Figure 2. Let E be any real line on Q. By our choice of generators of H (Q(R)), we can suppose that [E (R)] = (1; 0), up to a change in orientation. Therefore, the set of real points E (R) intersect X (R) in at least k points. The projection of X from E is a separating map of degree k with degree k 2 partition (1; : : : ; 1) 2 N . The curve X has genus (k 1) and X (R) has k connected components. In particular, if k 3, then X is not an M -curve, see Section 4. Example 2.14. We can carry out the construction from the previous example with 2 2 2 2 Q now being the quadratic surface de ned by x + y + z = w . Then Q(R) is the sphere. Again let S be the hypersurface which is a small perturbation of the union of hyperplanes H with equations of the form z + a w with 1 < a < 1 for i i i i = 1; : : : ; k. Let X be the complete intersection of S and Q. Then the resulting curve X is again hyperbolic with respect to the line E de ned by x = y = 0 and we have (2; : : : ; 2) 2 Hyp(X ). See the right hand side of Figure 2. As in Example 2.13, the curve X has genus (k 1) and X (R) has k connected components. But we will show that unlike in the preceding example, we have (1; : : : ; 1) 62 Sep(X ) if k > 2. In fact, there is no real morphism X ! P of degree k. Any complex line on Q intersects X in k points and the projection from this line gives a map to P of degree k. The gonality of the complexi ed curve is k and every morphism X ! P of degree k comes from the projection from a line by [Bas96] and [CL84]. However, since the surface Q does not contain any real lines, there are no real lines intersecting X (R) in more than two points proving the claim that there are no real morphisms of degree k to P . Example 2.15. Let X be a curve of genus g with X (R) having r connected compo- nents. Assume that there is a separating morphism f : X ! P with the property that f O 1 (1) is the canonical line bundle. The degree of a divisor of a non-zero real holomorphic dierential form on X restricted to any connected component of 8 MARIO KUMMER AND KRISTIN SHAW Figure 2. The two canonical curves of genus four from Examples 2.13 (left) and 2.14 (right) with real part having three connected components that are hyperbolic with respect to the red line. The curve on the left has (1; 1; 1) in its separating semigroup whereas the one on the right does not. X (R) must be even, see for example [GH81, Proposition 4.2]. Therefore, we have 0 0 r that d(f ) = 2d for some d 2 N . This implies that r g 1. Examples 2.5, 2.8, and 2.13 show that for g = 2; 3; 4 such a morphism exists for a curve with r = g 1. For planar curves X P of degree d 4 the canonical bundle is given by O (d 3). Furthermore, it was shown in [Tou13] that for some planar sextic curves with 9 ovals there exists a pencil of cubics without base points on X that gives rise to a separating morphism. Thus for g = 10 and r = 9 we can also nd such a morphism. For curves of genus dierent from 2; 3; 4 and 10 we do not know if this is the case. The next two lemmas exclude speci c elements from the hyperbolic semigroup. Lemma 2.16. Let X be a curve of dividing type. If d = (1; : : : ; 1) 2 Hyp(X ), then X is a rational curve, and hence d = (1). Proof. Assume that there is a hyperbolic embedding of X to P with each com- ponent having degree 1. The linear projection of X to the plane from any point not on X will send each X to a pseudoline. In P each two pseudolines intersect, which implies that the image of X will have at least r(r 1) simple nodes given that the center of projection was chosen generally enough. On the other hand, the degree is r. By the genus{degree formula this implies that 1 1 g (r 1)(r 2) r(r 1) = 1 r: 2 2 Therefore, if r = 1 then X is a rational curve and if r > 1 then X cannot be irreducible. This proves the lemma. Lemma 2.17. Let X be a curve of dividing type such that r > 2. Then no permu- tation of (2; 1; : : : ; 1) is in Hyp(X ). Proof. Assume that (2; 1; : : : ; 1) 2 Hyp(X ) and let X P be a realizing em- bedding. Let X be hyperbolic with respect to a line L and let H P be any hyperplane containing L. Let H \ X = fP ; P g and H \ X = fQ g for all 1 1 2 i i i = 2; : : : ; r. Note that X is not contained in any plane since r > 2. Thus, we can assume that there is a Q that is not on the line spanned by P and P . The set A i 1 2 of all real lines in H that contain neither P nor P has two connected components. 1 2 Let L H be a line through Q that is in the same connected component of A as L. Every hyperplane containing L intersects X in at least two real points because 1 THE SEPARATING SEMIGROUP OF A REAL CURVE 9 of Proposition 2.12 and j lk(X ; L(R))j = j lk(X ; L (R)j. Also it intersects every X 1 1 i 0 0 with i 2 in at least one real point. Now any hyperplane H spanned by L and a point Q 6= Q on X would intersect X in more than degree many points. i i 0 0 3. Some results for the general case In general it is not easy to determine the separating and the hyperbolic semi- group for a given curve. In this section we provide a method that allows us under some reasonable assumptions to construct from a separating morphism another sep- arating morphism of one degree higher. The main result of this section is that any embedding of a separating curve of high enough degree is hyperbolic. Therefore, hyperbolic embeddings are the rule rather than the exception. Lemma 3.1. Let L be a line bundle on X. Let s ; s 2 (X;L ) be two global 1 2 sections. Let (s ) = P +: : :+P with pairwise distinct P 2 X (R) and s (P ) 6= 0. 1 0 0 n j 2 0 Let U X (R) be an open neighbourhood of P for all j = 1; : : : ; n. There is an j j open neighbourhood U X (R) of P such that for every Q 2 U there are Q 2 U 0 0 0 0 j j for j = 1; : : : ; n and s 2 (X;L ) with (s ) = Q + : : : + Q . 3 3 0 0 n Proof. It suces to show the claim in the case where the U are pairwise disjoint and do not contain P for j = 1; : : : ; n. Let f be a rational function on X with the property that (f ) = (s ) (s ) . Since f has only simple and real zeros, there is 1 0 2 0 a c > 0 such that for all 2] c; c[ the rational function f + has a zero in each U for all j with P a zero of f . Now let U X (R) be an open neighbourhood j j 0 of P which is disjoint from each of the U and is contained in f (] c; c[). Then 0 j for every Q 2 U there is an 2] c; c[ such that Q is a zero of f + . Note 0 0 0 that f + has the same poles of the same orders as f . Thus, the eective divisor (f + ) + (s ) is of the form Q + : : : + Q where Q 2 U . 2 0 0 n j j Proposition 3.2. Let L be a line bundle on X. Let s ; s 2 (X;L ) be two global 0 1 sections that interlace. Let D = (s ) and let P 2 X (R) with s (P ) 6= 0 such that 0 0 0 0 0 0 `(D + P ) > `(D). Then there are interlacing sections s ; s 2 (X;L ) such that 0 1 0 0 (s ) = D + P where L is the line bundle corresponding to D + P . Proof. Without loss of generality, we can assume that s (P ) 6= 0. Let (s ) = D = 1 1 0 P ++P . Since `(D +P ) > `(D), there is a rational function g 2 L (D +P ) that 1 n 00 00 has a pole at P . Then D = D + P + (g) is an eective divisor with P 62 Supp D . 0 00 0 The eective divisors D+P , D +P , and D correspond to global sections s , f , and f of L , respectively. Let U be the connected component of X (R)r Supp(D +P ) 2 j that contains P for all j = 1; : : : ; n. Applying Lemma 3.1 to f and f shows that j 1 2 0 0 0 0 there is a global section s of L such that s and s interlace. 1 0 1 Remark 3.3. The condition `(D + P ) > `(D) is for example satis ed when D is non-special. Corollary 3.4. Let X be a curve of dividing type. Let d 2 Sep(X ) and l be the number of indices where d is odd. If jdj + l 2g 1, then d + Z Sep(X ). Proof. By [Hui03, Theorem 2.5] the divisor corresponding to a separating mor- phism realizing d is non-special. By Remark 3.3 the statement then follows from Proposition 3.2. Proof of Theorem 1.3. By Ahlfors' theorem [Ahl50, x4.2] there is a separating mor- phism f : X ! P . By Proposition 2.1 we can furthermore assume that the degree of f is more than 2g 2. Let L = f O 1 (1) and s ; s 2 (X;L ) be two global P 0 1 sections that interlace. Then D = (s ) is non-special. By [Sch00, Cor. 2.10, 0 0 0 Rem. 2.14], there is an integer n > 0 such that every real divisor on X of degree at 10 MARIO KUMMER AND KRISTIN SHAW least n is linearly equivalent to a sum of distinct points from X (R)r Supp D . We will show that the claim holds for k = max(2g + 1; n + deg D ). Indeed, let D be a divisor with deg D k. Then D D is linearly equivalent to a sum of distinct points from X (R)r Supp D since deg(D D ) n. An iterated application of the 0 0 previous proposition shows that the corresponding embedding of X to P(L (D) ) is hyperbolic. Proof of Theorem 1.6. The existence of a totally real pencil of curves of degree k satisfying our assumptions follows immediately by applying Theorem 1.3 to the line bundle O (k). Remark 3.5. If we allow base points on the curve, then the existence of a totally real pencil simply follows from Ahlfors' theorem [Ahl50, x4.2]. Indeed, let X P be a curve of dividing type and f : X ! P be a separating morphism de ned by two sections s ; s of a suitable line bundle. The rational function can be 0 1 expressed in the coordinates x; y; z of P as the fraction of two homogeneous forms f; g in x; y; z of the same degree. Then V (f + g) intersects X only in real points for every ; 2 R not both zero. However, the intersection V (f; g)\X is in general not empty, so the statement of Theorem 1.6 does not follow from this argument. Question 1. What are the smallest possible values for k in Theorem 1.3 and The- orem 1.6? Remark 3.6. Since our proof of Theorem 1.6 is not constructive, it further mo- tivates the desire for a bound on k in Question 1. In [Tou13], it was shown that for plane sextic curves of type h2 q 1h6ii and h6 q 1h2ii one can choose k = 3 in Theorem 1.6. Example 3.7. We consider a planar hyperbolic quartic, for example the Vinnikov quartic X P from [PSV11, Example 4.1] which is the planar curve de ned by 4 4 4 2 2 2 2 2 2 2x + y + z 3x y 3x z + y z = 0: We have g = 3 and r = 2. Moreover, the curve X is of dividing type since it is hyperbolic with respect to (0 : 0 : 1). Let X be the inner oval and X 1 2 be the outer oval. We have (1; 2); (2; 2) 2 Sep(X ), but Example 2.8 shows that (1; 1); (2; 1) 62 Sep(X ). We can also realize (3; 2) and (1; 3) with pencils of conics having 3 and 4 base points on X (R) respectively, see Figure 3. In coordinates, the p p (2z+ 2x 2y)( 2z+ 2x 4y) separating morphisms are given by the rational functions xy (2z+ 2x 2y)(z+x+y) and , respectively. By Corollary 3.4, we have that (1; 2) +Z xy Sep(X ). This determines the separating semigroup of the Vinnikov curve except for some cases when d = 1. In fact, we do not know whether or not (n; 1) is in Sep(X ) for n 2 N rf1; 2g. 4. M-curves Recall that an M -curve X has exactly g + 1 connected components in X (R) and that every M -curve is of dividing type. Here we prove Theorem 1.7, which completely determines the separating and hyperbolic semigroups of M -curves. Proof of Theorem 1.7. Part a) is just Corollary 2.2. By [Gab06, Proposition 4.1] or [Ahl50, x4.2] the all ones vector is in Sep(X ) for any genus. Moreover, any choice of divisor D = P + : : : + P where P 2 X for i = 1; : : : ; g + 1 is non-special 1 g+1 i i by [Hui01, Theorem 2.4]. Thus `(D) = 2 and any non-constant rational function f 2 L (D) gives rise to a separating morphism realizing (1; : : : ; 1) 2 Sep(X ). By THE SEPARATING SEMIGROUP OF A REAL CURVE 11 Figure 3. Realizing (3; 2) and (1; 3) with pencils of conics having 3 and 4 base points on X (R). iterated application of Proposition 3.2 and Remark 3.3 we nd that Sep(X ) = N . This proves the claim on Sep(X ) for X of any genus. By Halphen's theorem [Har77, Proof of Proposition 6.1] the space of divisors which are not very ample is of codimension at least one in the space of all divisors of degree at least g + 3. Therefore given an eective divisor D realizing d = (d ; : : : ; d ) 2 Sep(X ) with jdj g + 3 obtained from the above construction, we 1 r can assume that it is very ample. Thus d = (d ; : : : ; d ) 2 Hyp(X ) whenever 1 r jdj g + 3. For the hyperbolic semigroup in part b) note that (1; 1) 62 Hyp(X ) by Lemma 2.16 and that (1; 2); (2; 1) 2 Hyp(X ) by Example 2.7. Part c) follows from the above and Lemmas 2.16 and 2.17. 5. Domain of hyperbolicity Given a curve X P not contained in a hyperplane, we study the hyperbolicity locus H(X ) of X . De ne H(X ) := fL j X is hyperbolic with respect to Lg Gr(n 1; n + 1); where Gr(n 1; n + 1) is the space of codimension 2 subspaces of P . In [SV14, Question 3.13] the authors asked whether H(X ) is connected. This question is motivated by the case of planar curves where the answer is yes [HV07, Theorem 5.2]. Further evidence towards a positive answer was given by [SV14, Theorem 7.2] which states that H(X ) is the intersection of Gr(n 1; n + 1) with a convex N 1 cone in P (R) where N = . Here we consider Gr(n 1; n + 1) as a subset n 2 N 1 of P via the Plu cker embedding. However, the next example shows that the hyperbolicity locus is in general not connected. Example 5.1. Let X P be a planar elliptic curve with X (R) having two 0 0 connected components X and X . We can nd quadrics q ; q ; q ; q 2 O (2) 1 2 1 2 X 1 2 such that q and q interlace. Furthermore, we can assume that q intersects X in i i i exactly 4 points. Consider the image of X in P under the second Veronese embedding. The 0 0 5 0 quadrics q and q determine hyperplanes H and H in P . Because q and q i i i i i i interlace, the image of X is hyperbolic with respect to the 3-planes L = H \ i i H . The projection maps produce the elements (2; 4) and (4; 2) of the semigroup Hyp(X ). Therefore, the 3-planes L and L can not lie in the same connected 1 2 component of H(X ). Example 5.2. One can even obtain an example of a hyperbolic curve X P with H(X ) not connected. For this we proceed as is Example 5.1 to construct two 12 MARIO KUMMER AND KRISTIN SHAW Figure 4. The link diagram of the curve from Example 5.2 to- gether with two lines of hyperbolicity giving rise to dierent degree partitions. totally real pencils of conics, q + q and p + p , that give rise to dierent 0 1 0 1 degree partitions on a plane elliptic curve. Then X is obtained as the image the plane elliptic curve under the map x 7! (q (x) : q (x) : p (x) : p (x)): 0 1 0 1 3 2 For a precise example, take the image of the plane curve de ned by z + 2xz 3 2 2 2 x + y z = 0 under the above map with q = xy, q = x y , p = y(x 4z) and 0 1 0 p = (3x 4z y)(2x + y). Figure 4 shows the link diagram, up to isotopy, following a linear projection of this embedding back to P . The two components of the links are depicted in red and blue, and the image of the two lines of hyperbolicity are shown in black. The real projective plane is depicted as a disk with antipodal boundary points identi ed. Question 2. Is there an example of a non-singular curve X P where two connected components of H(X ) give rise to the same element in Hyp(X )? Acknowledgements. We would like to thank Erwan Brugall e, Ilia Itenberg and Claus Scheiderer for useful discussions. We are very grateful to Erwan Brugall e, Lionel Lang, Bernd Sturmfels, and to an anonymous referee for helpful comments on a preliminary version of this manuscript. References [ACGH85] Enrico Arbarello, Maurizio Cornalba, Phillip A. Griths, and Joe Harris, Geometry of algebraic curves. Vol. I, Grundlehren der Mathematischen Wissenschaften [Fun- damental Principles of Mathematical Sciences], vol. 267, Springer-Verlag, New York, 1985. MR 770932 [Ahl50] Lars L. Ahlfors, Open Riemann surfaces and extremal problems on compact subre- gions, Comment. Math. Helv. 24 (1950), 100{134. MR 0036318 (12,90b) [Bas96] B en edicte Basili, Indice de Cliord des intersections compl etes de l'espace, Bull. Soc. Math. France 124 (1996), no. 1, 61{95. MR 1395007 [Br a07] Petter Br and en, Polynomials with the half-plane property and matroid theory, Adv. Math. 216 (2007), no. 1, 302{320. MR 2353258 [CH13] Marc Coppens and Johannes Huisman, Pencils on real curves, Math. Nachr. 286 (2013), no. 8-9, 799{816. MR 3066402 [CL84] Ciro Ciliberto and Robert Lazarsfeld, On the uniqueness of certain linear series on some classes of curves, Complete intersections (Acireale, 1983), Lecture Notes in Math., vol. 1092, Springer, Berlin, 1984, pp. 198{213. MR 775883 [Cop13] Marc Coppens, The separating gonality of a separating real curve, Monatsh. Math. 170 (2013), no. 1, 1{10. MR 3032670 THE SEPARATING SEMIGROUP OF A REAL CURVE 13 [COSW04] Young-Bin Choe, James G. Oxley, Alan D. Sokal, and David G. Wagner, Homogeneous multivariate polynomials with the half-plane property, Adv. in Appl. Math. 32 (2004), no. 1-2, 88{187, Special issue on the Tutte polynomial. MR 2037144 [Fis06] Steve Fisk, Polynomials, roots, and interlacing, arXiv preprint math/0612833 (2006). [Gab06] Alexandre Gabard, Sur la repr esentation conforme des surfaces de Riemann a bord et une caract erisation des courbes s eparantes, Comment. Math. Helv. 81 (2006), no. 4, 945{964. MR 2271230 [G ar51] Lars G arding, Linear hyperbolic partial dierential equations with constant coe- cients, Acta Math. 85 (1951), 1{62. MR 0041336 [GH81] Benedict H. Gross and Joe Harris, Real algebraic curves, Ann. Sci. Ecole Norm. Sup. (4) 14 (1981), no. 2, 157{182. MR 631748 [Gul97] Osman Gul er, Hyperbolic polynomials and interior point methods for convex program- ming, Math. Oper. Res. 22 (1997), no. 2, 350{377. MR 1450796 [Har77] Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977, Graduate Texts in Mathematics, No. 52. MR 0463157 (57 #3116) [H or05] Lars H ormander, The analysis of linear partial dierential operators. II, Classics in Mathematics, Springer-Verlag, Berlin, 2005, Dierential operators with constant co- ecients, Reprint of the 1983 original. MR 2108588 [Hui01] Johannes Huisman, On the geometry of algebraic curves having many real components, Rev. Mat. Complut. 14 (2001), no. 1, 83{92. MR 1851723 [Hui03] , Non-special divisors on real algebraic curves and embeddings into real pro- jective spaces, Ann. Mat. Pura Appl. (4) 182 (2003), no. 1, 21{35. MR 1969741 [HV07] J. William Helton and Victor Vinnikov, Linear matrix inequality representation of sets, Comm. Pure Appl. Math. 60 (2007), no. 5, 654{674. MR 2292953 (2009a:93050) [Kle23] Felix Klein, Uber Riemanns Theorie der algebraischen Funktionen und ihrer Integrale, Gesammelte mathematische Abhandlungen, Springer, 1923, pp. 499{573. [KS15] Mario Kummer and Eli Shamovich, Real bered morphisms and Ulrich sheaves, arXiv preprint arXiv:1507.06760 (2015). [Mik00] Grigory Mikhalkin, Real algebraic curves, the moment map and amoebas, Ann. of Math. (2) 151 (2000), no. 1, 309{326. MR MR1745011 (2001c:14083) [MO18] Grigory Mikhalkin and Stepan Orevkov, Maximally writhed real algebraic links, arXiv preprint math/1710.00013 (2017). [MSS15a] Adam W. Marcus, Daniel A. Spielman, and Nikhil Srivastava, Interlacing families I: Bipartite Ramanujan graphs of all degrees, Ann. of Math. (2) 182 (2015), no. 1, 307{325. MR 3374962 [MSS15b] , Interlacing families II: Mixed characteristic polynomials and the Kadison- Singer problem, Ann. of Math. (2) 182 (2015), no. 1, 327{350. MR 3374963 [Nam79] Makoto Namba, Families of meromorphic functions on compact Riemann surfaces, Lecture Notes in Mathematics, vol. 767, Springer, Berlin, 1979. MR 555241 [PR11] Mikael Passare and Jean-Jacques Risler, On the curvature of the real amoeba, Pro- ceedings of the G okova Geometry-Topology Conference 2010, Int. Press, Somerville, MA, 2011, pp. 129{134. MR 2931884 [Pra07] Victor V. Prasolov, Elements of homology theory, Graduate Studies in Mathematics, vol. 81, American Mathematical Society, Providence, RI, 2007, Translated from the 2005 Russian original by Olga Sipacheva. MR 2313004 [PSV11] Daniel Plaumann, Bernd Sturmfels, and Cynthia Vinzant, Quartic curves and their bitangents, J. Symbolic Comput. 46 (2011), no. 6, 712{733. MR 2781949 [Ren06] James Renegar, Hyperbolic programs, and their derivative relaxations, Found. Com- put. Math. 6 (2006), no. 1, 59{79. MR 2198215 [Rok78] Vladimir A. Rokhlin, Complex topological characteristics of real algebraic curves, Us- pekhi Mat. Nauk 33 (1978), no. 5(203), 77{89, 237. MR 511882 [Sch00] Claus Scheiderer, Sums of squares of regular functions on real algebraic varieties, Trans. Amer. Math. Soc. 352 (2000), no. 3, 1039{1069. MR 1675230 [SV14] Eli Shamovich and Victor Vinnikov, Livsic-type determinantal representations and hyperbolicity, arXiv preprint arXiv:1410.2826 (2014). [Tou13] S everine Fiedler-Le Touz e, Totally real pencils of cubics with respect to sextics, arXiv preprint arXiv:1303.4341 (2013). Technische Universitat Berlin, Institut fur Mathematik, Strae des 17. Juni 136 10623 Berlin, Germany E-mail address : kummer@tu-berlin.de 14 MARIO KUMMER AND KRISTIN SHAW Department of Mathematics, University of Oslo, P.O. Box 1053, Blindern, 0316 Oslo, Norway E-mail address : krisshaw@math.uio.no
http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.pngMathematicsarXiv (Cornell University)http://www.deepdyve.com/lp/arxiv-cornell-university/the-separating-semigroup-of-a-real-curve-MMp0FVTTA5