Topics in Algebraic Geometry and Geometric Modeling, Contemp. Math., 334, 2003. Toric ideals, real toric varieties, and the algebraic moment map Frank Sottile Abstract. This is a tutorial on some aspects of toric varieties related to their potential use in geometric modeling. We discuss projective toric varieties and their ideals, as well as real toric varieties. In particular, we explain the relation between linear precision and a particular linear projection we call the algebraic moment map. Introduction We develop further aspects of toric varieties that may be useful in geometric modeling, building on Cox’s introduction to toric varieties, What is a toric vari- ety? [Cox03], which also appears in this volume. Notation and terminology follow that article, with a few small exceptions. This paper is organized into eight sections: 1. Projective Toric Varieties 2. Toric Ideals 3. Linear Projections 4. Rational Varieties 5. Implicit Degree of a Toric Variety 6. The Real Part of a Toric Variety 7. The Double Pillow 8. Linear Precision and the Algebraic Moment Map 1. Projective Toric Varieties arXiv:math/0212044v3 [math.AG] 18 Apr 2008 In this tutorial, we study toric varieties as subvarieties of projective space. This differs slightly from Cox’s [Cox03] presentation, where toric varieties are studied via the abstract toric variety XΣ of a fan Σ. The resulting loss of generality is compensated by the additional perspective this alternative view provides. Only in the last few sections do we discuss abstract toric varieties. 2000 Mathematics Subject Classification. 14M25, 14Q99, 13P10, 65D17, 68U05, 68U07. Key words and phrases. Toric Varieties, Moment Map, B´ezier Patches, Linear Precision. Research supported in part by NSF grant DMS-0134860. Corrected version of published article. 1998 ACM Computing Classification System: I.3.5 Computational Geometry and Object Modeling. c 2003 (American Mathematical Society) 1 2 FRANK SOTTILE A projective toric variety may be given as the closure of the image of a map (C∗)n −→ Pℓ , defined by Laurent monomials as in Section 13 of [Cox03]. There, the monomials had exponent vectors given by all the integer lattice points in a polytope. Here, we study maps given by any set of Laurent monomials. Our basic data structure will be a list of integer exponent vectors n A := {m0, m1,..., mℓ} ⊂ Z . Such a list gives rise to a map ϕA (written ϕ when A is understood), ∗ n ℓ ϕA : (C ) −→ P (1.1) m m m t 7−→ [t 0 , t 1 ,..., t ℓ ] . ∗ n We explain this notation. Given t = (t1,...,tn) ∈ (C ) and an exponent vector a a1 a2 an a = (a1,...,an), the monomial t is equal to t1 t2 · · · tn . In this way, the coor- dinates of Pℓ are naturally indexed by the exponent vectors lying in A. The toric ℓ variety YA is the closure in P of the image of the map ϕA. This map ϕA gives a parametrization of YA by the monomials whose exponents lie in A. We claim that YA is a toric variety as defined in Section 2 of [Cox03]. The map m m m (C∗)n ∋ t 7−→ (t 0 , t 1 ,..., t ℓ ) ∈ (C∗)1+ℓ is a homomorphism from the group (C∗)n to the group (C∗)1+ℓ of diagonal (1+ ℓ) by (1 + ℓ) matrices, which acts on Pℓ. Thus (C∗)n acts on Pℓ via this map. Since ∗ ℓ ∗ 1+ℓ ℓ scalar matrices (those in C I1+ℓ) act trivially on P , this action of (C ) on P ∗ 1+ℓ ∗ ∗ ℓ factors through the group (C ) /C I1+ℓ ≃ (C ) , which is the dense torus in the ℓ ∗ n toric variety P . Then YA is the closure of the image of (C ) in this torus, that image T acts on YA, and thus T is the dense torus of YA. n Suppose that A = ∆ ∩ Z , where ∆ is a lattice polytope. Then YA is the image of the abstract toric variety X∆ given by the normal fan of ∆ under the map of Section 13 in [Cox03]. When A has this form, we write Y∆ for YA. Example 1.2. Consider the three lattice polytopes [n], the line segment [0,n] ⊂ R , 2 △n, the triangle {(x, y) ∈ R | 0 ≤ x, y, x + y ≤ n} , and 2 m,n, the rectangle {(x, y) ∈ R | 0 ≤ x ≤ m, 0 ≤ y ≤ n} . The maps ϕ for these polytopes are t 7−→ [1,t,t2,...,tn] ∈ Pn , n n− n n+2 (s,t) 7−→ [1,s,t,s2,st,t2,...,s ,s 1t,...,t ] ∈ P( 2 ) , and (s,t) 7−→ [1,s,...,sm,t,st,...,smt,...,tn,stn,smtn] ∈ Pmn . and the resulting projective toric varieties are known (see [Har92]) as n Y[n] = the rational normal curve in P , n+2 2 ( 2 ) Y△n = the Veronese embedding of P in P , and 1 1 mn Ym,n = the Segre embedding of P × P in P of bidegree m,n. REAL TORIC VARIETIES, AND THE ALGEBRAIC MOMENT MAP 3 In geometric modeling these projective toric varieties give rise to, respectively, B´ezier curves, rational B´ezier triangles of degree n, and tensor product surfaces of bidegree (m,n). Example 1.3. Let n = 1 and A = {0, 2, 3}. Then the map (1.1) is t 7−→ [1,t2,t3] whose image YA is the cuspidal cubic 2 3 V(x0x2 − x1) = which is the non-normal toric variety of Example 3.2 in [Cox03]. Example 1.4. Let ∆ be the hexagon which is the convex hull of the six column 1 1 0 −1 −1 0 vectors 0 , 1 , 1 , 0 , −1 , −1 . We depict ∆ and its normal fan Σ∆. ∆ = and Σ∆ = Then ∆∩Z2 consists of these six vectors (the vertices of ∆) together with the origin 0 −1 −1 −1 −1 × 6 0 . Thus Y∆ = {[1,t,st,s,t ,s t ,s ] | s,t ∈ C }⊂ P . Remark 1.5. Suppose that the origin 0 is an element of A and that m0 = 0. Then the image of the map ϕ of (1.1) lies in the principal affine part of Pℓ ℓ ℓ U0 := {x ∈ P | x0 6=0} ≃ C , whose coordinates are [1, x1, x2,...,xℓ]. Thus U0 ∩ YA is an affine toric variety. In this case, the dimension of the projective toric variety YA is equal to the dimension of the linear span of the exponent vectors A. 2. Toric Ideals The toric ideal IA is the homogeneous ideal of polynomials whose vanishing ℓ defines the projective toric variety YA ⊂ P . Equivalently, IA is the ideal of all the ∗ n homogeneous polynomials vanishing on ϕA((C ) ). Our description of IA follows the presentation in Sturmfels’s book, Gr¨obner bases and convex polytopes [Stu96]. ℓ Let [x0, x1,...,xℓ] be homogeneous coordinates for P with xj corresponding to mj the monomial t in the map ϕA (1.1), where A = {m0, m1,..., mℓ}. A monomial xu in these coordinates has an exponent vector u ∈ N1+ℓ. Restricting the monomial u m0 m2 mℓ x to ϕA(t1,...,tn) = [t , t ,..., t ] yields the monomial m m m tu0 0+u1 1+···+uℓ ℓ . This exponent vector is Au, where we consider A to be the matrix whose columns are the exponent vectors in A. For the hexagon of Example 1.4, this is 01 1 0 −1 −1 0 0011 0 −1 −1 4 FRANK SOTTILE This discussion shows that a homogeneous binomial xu − xv with Au = Av ∗ n vanishes on ϕA((C ) ) and hence lies in the toric ideal IA. In fact, the toric ideal IA is the linear span of these binomials. Theorem 2.1. The toric ideal IA is the linear span of all homogeneous bino- mials xu − xv with Au = Av. We obtain a more elegant description of IA if the row space of the matrix A contains the vector (1,..., 1), for then the homogeneity of the binomial xu − xv follows from Au = Av. It is often useful to force this condition as follows. Given a list A of exponent vectors in Zn, the lift A+ of A to 1 × Zn is obtained be prepending a component of 1 to each vector in A. That is, A+ := {(1, m) | m ∈ A} The matrix A+ is obtained from the matrix A by adding a new top row of 1s. For the hexagon, this is 1111 1 1 1 A+ = 0 11 0 −1 −1 0 . 0011 0 −1 −1 Here is the lifted hexagon (shaded) −1 −1 −1 −1 × Then YA+ = {[r,rt,rst,rs,rt ,rs t ,rs ] | r,s,t ∈ C } = Y∆. This lifting does not alter the projective toric variety. Indeed, m0 m1 mℓ ϕA+ (t0,t1,...,tn) = [t0t ,t0t ,...,t0t ] m m m = [t 0 , t 1 ,..., t ℓ ] = ϕA(t1,...,tn) , and so YA = YA+ . The dimension of YA is one less than the dimension of the linear + span of A . Since IA = IA+ , we have u v Theorem 2.2. The toric ideal IA is the linear span of all binomials x − x with A+u = A+v. If u ∈ Z1+ℓ, then we may write u uniquely as u = u+ − u−, where u+, u− ∈ N1+ℓ but u+ and u− have no non-zero components in common. For example, if u = (1, 3, 2, −2, 2, −4), then u+ = (1, 3, 2, 0, 2, 0) and u− = (0, 0, 0, 2, 0, 4). We 1+ℓ describe a smaller set of binomials that generate IA. Let ker(A) ⊂ Z be the kernel of the matrix A. u+ u− + Corollary 2.3. IA = hx − x | u ∈ ker(A )i.