Arxiv:1806.05346V4 [Math.AG]

arXiv:1806.05346v4 [math.AG] 26 Mar 2020 onigtenme fsltoso ytm fplnmasv polynomials of systems of solutions of number the Counting o ayzeroes? many How iaiMondal Pinaki atI oi theory Toric I: Part agoer tinﬁnity at geometry ia

Preface

In this book we describe an approach through toric geometry to the following problem: “estimate the number (counted with appropriate multiplicity) of isolated solutions of n polynomial equations in n

variables over an algebraically closed field k.” The outcome of this approach is the number of solutions for “generic” systems in terms of their Newton polytopes, and an explicit characterization of what makes a system “generic.” The pioneering work in this field was done in the 1970s by Kushnirenko, Bernstein and Khovanskii, who completely solved the problem of counting solutions of generic systems on the “torus” n (k 0 ) . In the context of our problem, however, the natural domain of solutions is not the torus, but the \{ } n affine space k . There were a number of works on extension of Bernstein’s theorem to the case of affine space, and recently it has been completely resolved, the final steps having been carried out by the author. The aim of this book is to present these results in a coherent way. We start from the beginning, namely Bernstein’s beautiful theorem which expresses the number of solutions of generic systems in terms of the mixed volume of their Newton polytopes. We give complete proofs, over arbitrary algebraically closed fields, of Bernstein’s theorem and its recent extension to the affine space, and describe some open problems. We also apply the developed techniques to derive and generalize Kushnirenko’s results on Milnor numbers of hypersurface singularities which in 1970s served as a precursor to the development of toric geometry. Care was taken to make this book as elementary as possible. In particular, we develop all the necessary algebraic geometry (modulo some explicitly stated basic results). This book should be accessible to a second year graduate student. Contents

Chapter I. Introduction 1 1. The problemand the results 1 2. Prerequisites 7 3. Organization 8

Chapter II. A brief history of points of inﬁnity in geometry 10 1. Points at inﬁnity 10 2. Homogeneous coordinates 11 3. Projective space 12

Part 0. Preliminaries 14

Chapter III. Quasiprojective varieties over algebraically closed fields 15 1. Overview of classical results from algebra 15 2. Quasiprojective Varieties 16 3. Affine varieties 18 4. (Ir)reducibility 20 5. Rational functions and rational maps on irreducible varieties 21 6. Product spaces, Segre map, Veronese embedding 22 7. Completeness and compactification 23 8. Dimension 25 9. Constructible sets 28 10. Zariski closure of the image of a morphism 29 11. Tangent space, singularities, local ring at a point 31 12. Completion of the local ring at a point 35 13. Degree of a dominant morphism 37

Chapter IV. ∗Intersection multiplicity 40 1. Introduction 40 2. Closed subschemes of a variety 41 3. Possibly non-reduced curves 44 4. Intersection multiplicity at a nonsingular point of a variety 47 5. Intersection multiplicity of complete intersections 51

Chapter V. Convex polyhedra 53 1. Basic notions 53 2. Characterization of convex polyhedra 57 3. Basic properties of convex polyhedra 61 4. Normal fan of a convexpolytope 68 5. Rational polyhedra 70 6. ∗Volume of convex polytopes 72 7. ∗Volume of special classes of polytopes 75

Part 1. Toric Varieties and classical results 79

Chapter VI. Toric varieties over algebraically closed ﬁelds 80 CONTENTS

1. Algebraic torus 80 2. Toric varieties from ﬁnite subsets of Zn 81 3. Examples of toric varieties 82 4. Structureof XA 83 5. Toric varieties from polytopes 85 6. Nonsingularity in codimension one on XP 88 7. Extending closed subschemes of the torus to XP 89 8. Branches of curves on the torus 92 9. Points at inﬁnity on toric varieties 93 10. ∗Weighted projective spaces 95 11. ∗Weighted blow up 97

Chapter VII. Number of solutions on the torus: BKK bound 99 1. Introduction 99 2. Mixedvolume 99 3. Theorems of Kushnirenko and Bernstein 100 4. Proof of Bernstein-Kushnirenko non-degeneracy condition 102 5. ProofoftheBKK bound 106 6. Applications of Bernstein’s theorem to convex geometry 109 7. Some technical results 113 8. The problem of characterizing coefﬁcients which guarantee non-degeneracy 114 9. Notes 116

Chapter VIII. (Weighted) Bézout theorems 117 1. Weighted degree 117 n n 2. P (ω) as a compactification of k when the ωj are positive and ω0 =1 117 3. Weighted Bézout theorem 118 4. Products of weighted projective spaces 119 5. Weighted multi-homogeneous Bézout theorem 121 6. Notes 122

Part 2. Extension of Bernstein-Kushnirenko Theorem to the afﬁne space and other applications123

Chapter IX. Intersection multiplicity at the origin 124 1. Introduction 124 2. Generic intersection multiplicity 124 3. Characterization of minimal multiplicity systems 128 4. Proof of the non-degeneracy condition 130 5. Proofofthebound 133 6. The efﬁcient version of the non-degeneracycondition 136 7. Other formulae for generic intersection multiplicity 138 8. Monotonicity of generic intersection multiplicity 141 9. Notes 142

Chapter X. Number of solutions on (open subsets of) the affine space 143 1. Introduction 143 2. Thebound 143 3. Derivation of the formuale for the bound 146 4. Other formulaefor the bound 148 5. Examples motivating the non-degeneracy conditions 150 6. Non-degeneracy conditions 153 7. Proof of the non-degeneracy conditions 155 8. Weighted Bézout theorem: general version 161 9. Weighted multi-homogeneous Bézout theorem: general version 165 10. Open problems 166 CONTENTS

Chapter XI. Milnor number of a hypersurface at the origin 168 1. Introduction 168 2. Milnornumber 168 3. Generic Milnor number 169 4. Classical notions of non-degeneracy 171 5. Newton number: Kushnirenko’s formula for the generic Milnor number 173 6. Open problems 176 Chapter XII. Beyond this book 180 1. Toric varieties 180 2. Newton-Okounkov bodies 180 3. B´ezout problem 180 Appendix A. Notation 182 Appendix B. Miscellaneous commutative algebra 183 1. Fundamental theorem of ﬁnitely generated modules over a PID 183 2. Prime and Maximal ideals, Noetherian rings, Zero-divisors 183 3. Nakayama’s lemma 184 4. Localrings 184 5. Discrete valuation rings 185 6. Krull dimension 185 7. Primary Decomposition 185 8. Length of modules 187 9. Gradedrings 188 10. Macaulay’s Unmixedness Theorem 188 11. Properties of order at a point on a possibly non-reduced curve 191 12. The power series ring and monomial orders 195 13. Primitive elements of Zn 198 14. Symmetric multiadditive functions on a commutative semigroup 199 Appendix. Bibliography 203 CHAPTER I

Introduction

1. The problem and the results This book is about the problem of computing the number of solutions of systems of polynomials, or equivalently, the number of points of intersection of the sets of zeroes of polynomials. In this section we formulate the precise version of the problem we are going to study and give an informal description of the results. One natural observation that simpliﬁes the problem is that intersection multiplicity should be taken into account, e.g. even though a tangent line intersects a parabola at only one point, it should be counted with multiplicity two (see ﬁg. 1).

As secants approach the tangent at O more and more closely, both of the two points of intersection move arbitrarily close to O.

FIGURE 1. A tangent line intersects a parabola at a point with multiplicity two

The geometric intuition for intersection multiplicity is the “principle of continuity,” the principle that continuous perturbations of systems result in continuous changes of associated metrics or invariants1. Since the number of points of intersection is a discrete invariant of a system, it follows that it must not change under continuous perturbation. However, over real numbers points of intersection may disappear upon an infinitesimal deformation (see fig. 2). On the other hand, this problem disappears if one also counts “imaginary” solutions (this is why the intersection theory over complex numbers, or, more generally, an algebraically closed field, is easier than the intersection theory over real numbers). In this book we will consider polynomial systems defined over algebraically closed fields2.

y = x2

y = ǫ intersects the parabola at two real points with multiplicity one. (0, 0) y = 0 intersects the parabola at one real point with multiplicity two. y = −ǫ intersects the parabola at two imaginary points with multiplicity one.

FIGURE 2. Disappearance of real points of intersection

If there are inﬁnitely many solutions of a system of polynomials, then the solution set has positive dimensional components, and assigning multiplicity to these components is trickier; we bypass this problem

1“Consider an arbitrary figure in general position . . . Is it not obvious that if . . . one begins to change the initial figure by insensible steps, or applies to some parts of the figure an arbitrary continuous motion, then is it not obvious that the properties and relations established for the initial system remain applicable to subsequent states of this system provided that one is mindful of particular changes, when, say, certain magnitudes vanish, change direction or sign, and so on—changes which one can always anticipate a priori on the basis of reliable rules.” – J. V. Poncelet, the foremost exponent of the principle of continuity, in the introduction of Traitédes propriétés projectives des figures (1822), as cited in [Ros05]. 2. . . which Poncelet probably would not have approved of, given his attitude towards consideration of complex solutions; see [Gra11, Section 4.2] for a most interesting account of this history.

1 2 I. INTRODUCTION in this book and compute only isolated3 solutions. This implies in particular we do not consider “underdetermined systems,”4 since an underdetermined system over an algebraically closed field can only have either positive dimensional or empty sets of solutions. We also ignore “overdetermined systems”4 because of the relative difficulty in assigning multiplicities. The final form of the subject of this book is thus the following: Problem I.1 (Affine Bézout problem). Given n polynomials in n variables over an algebraically closed

field k, give a sharp estimate of the number of its isolated solutions counted with appropriate multiplicity, and determine the conditions under which it is exact. For n =1, the fundamental theorem of algebra gives a complete answer: a polynomial of degree d has precisely d zeroes counted with multiplicity. For n 2, there is a problem: points of intersection may run off to infinity (see fig. 3). ≥

y = x2 As secants approach the vertical line at O more and more closely, one of the points of intersection approaches O and the other goes to inﬁnity.

FIGURE 3. A vertical line intersects the parabola at one point with multiplicity one

Any reasonable approach to problem I.1 therefore must take into account “intersections at infinity.” A theorem named after E. Bézout (1730–1783) is the most basic result that does it satisfactorily. n THEOREM I.2 (Bézout’s theorem, affine version). The number of isolated solutions in k of n polynomials in n variables is at most the product of their degrees. Moreover, this bound is exact if and only if the only common solution of the leading forms5 of the polynomials is the origin. Example I.3. Consider the system in fig. 3 consisting of the parabola y x2 =0 and a line ax+by+c =0. The Bézout bound is 2 1=2, and the leading forms are x2 and ax−+ by. As long as b = 0, the only solution to x2 = ax +× by = 0 is (0, 0), so that the bound− is exact. However, if b = 0, i.e.6 the line is − vertical, then any point of the form (0, k), k k, is a common solution of the leading forms. Consequently the Bézout bound overestimates the number∈ of solutions in this case, as observed in fig. 3. From the perspective of projective geometry, the Bézout bound is the number of intersections of poly- n n nomial hypersurfaces in the projective space P , which is a compactification of the affine space k formed by adjoining a “hyperplane at infinity.” Therefore the Bézout bound is exact if and only if the hypersurfaces do not intersect at any point at infinity on Pn. However, as Gauss famously remarked,6 infinity is the limit of some process, and curves which approach arbitrarily close to each other in one process may n grow apart in another. A natural class of compactifications of k containing the projective space is that of weighted projective spaces. Given an n-tuple ω = (ω1,...,ωn) of positive integers, the correspond- ω n ing weighted rational curve C through a point a = (a ,...,a ) k is the curve parametrized by the a 1 n ∈ map t (a tω1 ,...,a tωn ). In the same way that in the projective space straight lines with different 7→ 1 n 3A point is isolated in a set S if it is open in S. 4A system is underdetermined or overdetermined depending on whether the number of equations is smaller or greater than the number of variables. 5The leading form of a polynomial is the sum of its monomial terms with the highest degree; e.g. if f = 2x3 + 7x2y − 9y2 + 7xy − x + 1, then its degree is 3 and the leading form is 2x3 + 7x2y. 6Discussing his friend H. Schumacher’s purported proof of the parallel postulate, Gauss wrote to him (as cited in [Wat79]), “I protest first of all against the use of an infinite quantity as a completed one, which is never permissible in mathematics. The infinite is only a façon de parler, where one is really speaking of limits to which certain ratios come as close as one likes while others are allowed to grow without restriction.” 1. THE PROBLEM AND THE RESULTS 3

n ω slopes are separated at infinity, in the weighted projective space P (1,ω) the curves Ca corresponding to ω distinct a are separated at infinity. See fig. 4 for an example with ω = (1, 2), in which case Ca a is the α α { } family of parabolas a 2y a x2 = 0 . The “weight” of a monomial x 1 x 2 xαn corresponding to { 1 − 2 } 1 2 ··· n ω is ω1α1 + + ωnαn. If f is a polynomial, then the corresponding weighted degree ω(f) of f is the maximum of··· the weights of all the monomials appearing in f. The leading weighted homogeneous form of f is the sum of all monomials (with respective coefficients) of f with the highest weight. Computing intersection numbers on Pn(1,ω) leads to the “weighted Bézout theorem,” of which the original theorem of Bézout (theorem I.2) is a special case (corresponding to ω = (1,..., 1)).

parabolas

parabolas lines parabolas lines lines

line at inﬁnity

2 2 2 (A) k (B) P (C) P (1, 1, 2)

FIGURE 4. P2 separates lines, but not parabolas, at inﬁnity, whereas P2(1, 1, 2) separates parabolas, but not lines, at inﬁnity

THEOREM I.4 (Weighted Bézout theorem for positive weights). Let ω be a weighted degree on the ring of polynomials with positive weights ωi for xi, i =1,...,n. Then the number of isolated solutions of n polynomials f1,...,fn on k is bounded above by ( j ω(fj ))/( j ωj). This bound is exact if and only if the leading weighted homogeneous forms of f1,...,fn have no common solution other than the origin. Q Q Example I.5. Let ω = (1, 2), f = y x2 and g = ax + c, a = 0. Then ω(f)=2, ω(g)=1, and the leading weighted homogeneous forms− of f and g are respectively6 y x2 and ax . The only solution to the leading weighted homogeneous forms of f and g with respect to ω−is (0, 0), so theorem I.4 implies that the number of solutions of f = g = 0 is precisely the weighted Bézout bound (ω(f)ω(g)/(ω(x)ω(y)) = (2 1)/(1 2)=1, as we saw in fig. 3. × ×

parabolas c(1,1),∞

ω = (1, 2)

lines ω = (1, 1) c(1,2),∞

(B) A coordinate chart near (A) P inﬁnity on XP

FIGURE 5. Parabolas and lines near curves at inﬁnity on XP

The main class of compactiﬁcations considered in this book are toric varieties associated to convex integral polytopes7. If is an n dimensional convex integral polytope in Rn, then the outer normal to each of its (n 1)-dimensionalP faces determines (up to a constant of proportionality) a weighted degree, and − in the corresponding toric variety XP , weighted rational curves corresponding to each of these weights are separated. See ﬁg. 5 for an example of a toric variety in which both parabolas and lines are separated

7 A convex integral polytope in Rn is the convex hull of ﬁnitely many points in Rn with integer coordinates. 4 I. INTRODUCTION

2 at inﬁnity. It has two curves at inﬁnity (with respect to k ) corresponding to the two edges of which P are not along the axes; we denote these curves by cω,∞, where ω is the corresponding weight. Each cω,∞ separates the family of weighted rational curves corresponding to ω. Computing intersection numbers of hypersurfaces on toric varieties yields a beautiful result of D. Bernstein, which we now describe.

y y

x x NP(f) NP(g) f = 1 − x + 3x3 + 4x2y − 7y2 g =2+ x − y + xy + x2y + xy2

FIGURE 6. Some Newton polytopes in dimension 2

The Newton polytope of a polynomial is the convex hull of all the exponents that appear in its expression, see fig. 6. V. I. Arnold noticed sometime in 1960s or 1970s that invariants of “generic” systems of polynomials tend not to depend on precise values of the coefficients of their monomials, but only on the combinatorial relations of the exponents of these monomials. The study of this phenomenon was a recur- ring topic at his seminars at Moscow University. While working on Arnold’s question on determination of the Milnor number8 at the origin of a generic polynomial, A. Kushnirenko discovered that if all polynomials have the same Newton polytope, then for generic systems the number of isolated solutions which do not belong to any coordinate hyperplane has a strikingly simple expression: it is simply n! times the volume of this polytope! D. Bernstein soon figured out how to remove the restriction on Newton polytopes (about 130 years before this F. Minding [Min41] discovered a special case of Bernstein’s theorem in dimension two9).

THEOREM I.6. Let N be the number (counted with appropriate multiplicities) of the isolated zeroes

∗ n n k of polynomials f ,...,f on (k ) := x =0 . 1 n \ i{ i } (1) Kushnirenko [Kou76]: If each f has the same Newton polytope , then N n! Vol( ). If j S Vol( ) is nonzero, then the bound is exact if and only if the followinPg condition≤ holds: P P for each nontrivial weighted degree ω, the corresponding lead- ( ) ∗ n ∗ ing forms of f1,...,fn do not have any common zero on (k ) . (2) Bernstein [Ber75]: In general N is bounded above by the mixed volume10 of the Newton polytopes of f . If the mixed volume is nonzero, then the bound is exact if and only if ( ) holds. j ∗ Example I.7. If the Newton polytope of each polynomial contains the origin, then theorem I.6 in fact gives n an upper bound on the number of isolated solutions on k and it is in general better than the bounds from theorems I.2 and I.4. For example, using the fact that mixedvolume of two planar bodies and is simply Area( + ) Area( ) Area( ) (example VII.3), we see that Bernstein’s boundP for theQ number of solutionsP ofQf =−g =0 isP the− area ofQ the region shaded in blue in fig. 7, which is equal to 8. Bézout bound, on the other hand is 3 3=9; it is not hard to show that the 9 is also the best possible weighed Bézout bound. ×

The natural domain of solutions of systems of polynomials over a ﬁeld k is however not the torus

∗ n n n

k k (k ) , but the afﬁne space . There are at least two different ways to extend Bernstein’s formula to . The approach motivated by the polynomial homotopy method for solving polynomial systems is as follows:

8The Milnor number is an invariant of a singularity, see section XI.2. 9A. Khovanskii gives a summary of Minding’s approach in [BZ88, Section 27.3]; an English translation of [Min41] by D. Cox and J. M. Rojas appears in [GK03]. 10The mixed volume is the canonical multilinear extension (as a functional on convex bodies) of the volume to n-tuples of convex bodies in Rn, see section VII.2 for a precise description. 1. THE PROBLEM AND THE RESULTS 5

+ =

NP(f) NP(g) NP(f)+NP(g)

FIGURE 7. Minkowski sum of Newton polytopes of f and g

given polynomials f1,...,fn, one starts with a deformed system f1 = c1,...,fn = cn with nonzero cj . For generic f1,...,fn all solutions of the deformed system are in fact on the torus, and their number is given by Bernstein’s theorem. Then one counts how many of these solutions approach isolated solutions of f1,...,fn as each cj 0. This approachis taken in [Kho78, HS95, LW96, RW96, Roj99]. In particular, B. Huber and B. Sturmfels→ [HS95] found the general formula through this approach; however they proved it in a special case, and only in characteristic zero. J. M. Rojas [Roj99] observed that Huber and Sturmfels’ formula works over all characteristics. The other approach is closer to Bernstein’s original proof of his theorem: here one computes the number of “branches” of the curve defined by f = = f = 0 and 2 ··· n then the sum of order of f1 along these branches. General formulae through this approach were obtained by A. Khovanskii [unpublished]11 and the author [Mon16]. This formula requires knowing the intersection multiplicity at the origin of generic systems of polynomials. As an illustration we now state the weighted Bézout formula for weighted degrees with possibly negative weights12. Let ω be a weighted degree with nonzero weights ω ,...,ω . If I := i : ω < 0 , then the number of isolated zeroes of f ,...,f is 1 n − { i } 1 n max ω(f ), 0 + ω deg (f ) j j i∈I− i xi j (1) N ( 1)|I−|−|I| { } | | ≤ − Q i Pωi IX⊆I− | | (theorem X.36). Note that this reduces to the weighted BézoutQ bound from theorem I.4 in the case that each ωi is positive, i.e. I− = . This bound is exact for generic f1,...,fn, provided ω(fj) is nonnegative for each j. In the general case,∅ define n ω(f) := α = (α1,...,αn) R : αi 0 for each i, ω, α ω(f), αk deg (f) for each j I− P { ∈ ≥ h i≤ ≤ xk ∈ }

3 3

2 2

1 1

0.5 0.5 1 1.5 1 1.5 2 2 2 2

(A) ω = (1, −1, 1), ω(f) = 1 (B) ω = (1, −1, −1), ω(f)= −1

FIGURE 8. (f) for f = x2 + x x2 + x x x Pω 2 2 3 1 2 3

(see ﬁg. 8). Given I 1,...,n , let RI be the I -dimensional coordinate subspace of Rn spanned by all x , i I. Then there⊆{ is a collection} T of subsets| | of 1,...,n such that for each I T , the number of i ∈ { } ∈

11Khovanskii described his result to the author in 2017. 12For simplicity here we do not allow zero weights; see theorem X.36 for the statement without this restriction. 6 I. INTRODUCTION

I distinct j such that ω(fj ) touches R is precisely I , and P | | I I ′ ′ ′ ′ (2) N MV j1 (ω) R ,..., j (ω R ) [πI ( j (ω)),...,πI ( j (ω))]0 ≤ P ∩ P |I| ∩ × P 1 P n−|I| I∈T X (see theorem X.37 for the precise statement), where j ,...,j (respectively, j′ ,...,j′ ) is the collection of indices j such that (ω) touches • 1 |I| 1 n−|I| Pj (respectively, does not touch) RI ; MV( ,..., ) is the mixed volume; • ′ · · I := 1,...,n I is the complement of I, and πI′ is the natural projection onto the coordinate • { n}\ ′ ′ subspace of R spanned by all xi′ , i I , and ∈ [ ,..., ]0 is the intersection multiplicity at the origin of systems of generic polynomials with • given· Newton· polytopes. The general formula for generic number of solutions on the afﬁne space is no more difﬁcult; it is of the same type as (2), i.e. it is a sum of products of mixed volumes with generic intersection multiplicity at the origin (see theorem X.4). However, to use it one needs to compute the generic intersection multiplicity at the origin. In the special case that each polynomial is “convenient,”13 a formula for generic intersection multiplicity was given by L. Ajzenberg and A. Yuzhakov [AY83]; a Bernstein-Kushnirenko type “non- degeneracy” condition, i.e. the condition for the bound being exact, was also known for convenient systems (see e.g. [Est12, Theorem 5]). In the general case Rojas [Roj99] gave a formula via Huber and Sturmfels’ polynomial homotopy method. The non-degeneracy condition for the general case was established by the author in [Mon16].

n As hinted above, the formula for the generic number of solutions on k is straightforward once one has the formula for generic intersection multiplicity at the origin. Sufﬁcient criteria under which the bound is exact can also be obtained easily by adapting the Bernstein-Kushnirenko non-degeneracy condition ( ); such criteria were given by several authors including Khovanskii [Kho78], Rojas [Roj99]. Precise non-∗ n degeneracy conditions, i.e. which are necessary and sufﬁcient for the bound to be exact on k , are however 3 more subtle than ( ); consider e.g. the problem of characterizing non-degenerate systems on k of the form ∗ f1 = a1 + b1x1x2 + c1x2x3 + d1x3x1

f2 = a2 + b2x1x2 + c2x2x3 + d2x3x1

f3 = x3(a3 + b3x1x2 + c3x2x3 + d3x3x1) ∗ where aj ,bj,cj , dj k (this system is discussed in example X.16). If all aj ,bj,cj , dj are generic, then it ∈ 3 is straightforward to check directly that all common zeroes of f1,f2,f3 on k are isolated and they appear ∗ 3 on (k ) . Consequently, Bernstein’s theorem implies that the number of solutions is the mixed volume of the Newton polytopes of the fj , which equals 2. Now if a1 = a2, b1 = b2, and the remaining coefﬁcients are generic, then ( ) continues to be true, so that Bernstein’s theorem applies and number of solutions on

∗ 3 ∗ 3 k (k ) is still 2; in particular, the system continues to be non-degenerate on . However, in this case the 3 set of common zeroes of f1,f2,f3 on k also has a positive dimensional component, namely the curve x3 = a1 + b1x1x2 =0 . This situation never arises in the case of Bernstein’s theorem; indeed, existence {of a positive dimensional} componentmakes a system violate ( ) and its straightforward adaptations. Unlike ∗ n the Bernstein-Kushnirenko non-degeneracy criterion, the correct non-degeneracy criterion for k needs to accommodate existence of positive dimensional components - this criterion has to recognize when such a component leads to a loss of isolated solutions, and when it does not; such a criterion was given by the author in [Mon16].

We mentioned above that the pioneering work of Kushnirenko on counting solutions of polynomial systems was motivated by his work on Milnor numbers of hypersurface singularities. In [Kou76] he gave a beautiful formula for a lower bound on the Milnor number, and showed that the bound is achieved by Newton non-degenerate singularities if either the characteristic is zero or if the polynomial is convenient. It was however clear from the beginning that Newton non-degeneracy is not necessary for the formula to

13 mj A polynomial or power series is convenient if for each j, there is mj ≥ 0 such that the coefficient of xj is nonzero. 2. PREREQUISITES 7 hold, and it also does not imply “finite determinacy.”14 C. T. C. Wall [Wal99] introduced another notion of non-degeneracy which implies finite determinacy and which also guarantees that the Milnor number can be computed by Kushnirenko’s formula. S. Brzostowski and G. Oleksik [BO16] found the combinatorial condition which under Newton non-degeneracy is equivalent to finite determinacy. The Milnor number of a hypersurface at the origin is same as the intersection multiplicity at the origin of the partial derivatives of the defining polynomial (or power series). The non-degeneracy condition for intersection multiplicity therefore gives a natural starting point to study Milnor numbers. This condition generalizes both New- ton non-degeneracy (for isolated singularities) and Wall’s non-degeneracy condition; the author showed in [Mon16] that in positive characteristic this condition is sufficient, and in zero characteristic it is both necessary and sufficient, for the Milnor number to be generic.

The purpose of this book is to give a unified exposition of the results described above. In addition to Bernstein’s theorem (over arbitrary algebraically closed fields), classical results proved in this book include weighted homogeneous and multi-homogeneous versions of Bézout’s theorem; complete proofs (or even, statements) of these results are otherwise hard to find. We followed Bernstein’s original proof for establishing the non-degeneracy conditions of his theorem; in particular we present his simple and ingenious trick to construct a curve of solutions that runs off to infinity in the case that the non-degeneracy condition ( ) is not satisfied15. This book is the first part of a series of works on a constructive approach to compactifications∗ of affine varieties started in the author’s PhD thesis [Mon10], for which the affine Bézout problem served as a motivation. Based on the results of this book, in the next part we give a solution to the general version of the affine Bézout problem, i.e. give a recipe to compute the precise number (counted with multiplicity) of solutions of any given system of n polynomials in n variables. The algorithm is inductive; it consists of finitely many steps, and at each step a non-degeneracy criterion determines if the correct number has been computed. The estimate and non-degeneracy criterion for number of solutions on n k from chapter X of this book serve as the initial step of that algorithm.

2. Prerequisites We tried to ensure that it is accessible to someone with the mathematical maturity and algebra back- ground of a second year mathematics graduate student. In the ideal case a reader would be familiar with the properties of algebraic varieties discussed in chapter III, so that (s)he could start with toric varieties in part 1 and only refer to results from part 0 if necessary. However, part 0 is mostly self contained (modulo the dependencies explicitly stated in sections III.1 and IV.3.1 and some commutative algebra results stated in appendix B) - it can potentially be used as the material for a first course in algebraic geometry. One possible strategy for such a course would be to cover the chapters on algebraic varieties (chapter III), toric varieties (chapter VI), Bernstein-Kushnirenko theorem (chapter VII) and (weighted) Bézout’s theorem (chapter VIII). The chapters on intersection multiplicity (chapter IV) and polytopes (chapter V) are included for completion - we believe in a first course the required results from these chapters should simply be explained, perhaps via examples and/or pictures, instead of working out the details of the proofs. In particular, the proofs (and exercises) given in chapter V are elementary and a student should not have much difficulty in following them. The most sophisticated part of chapter IV is the concept of “closed subschemes” of a variety and the fact that it can be locally defined by ideals determined by regular functions; the other results are basic facts about intersection multiplicity of n regular functions at a nonsingular point a of an n dimensional variety (e.g. that they can be defined via the “order” at a of one of the functions along the curve defined by the other functions) and relevant properties of the “order” function for a (possibly non-reduced) curve. While the proofs use somewhat complicated algebra, the statements are intuitive, at least if one has some familiarity with properties of (complex) analytic functions.

14i.e. it does not ensure that the singularity at the origin is isolated. 15The bound from Bernstein’s theorem and the sufficiency of (∗) for the bound can be established without much difficulty (and also very elegantly!) using the general machinery of intersection theory (see e.g. [Ful93, Section 5.4]). However, we do not know of any proof of the necessity of (∗) using this approach which does not involve an adaptation of Bernstein’s trick; in all probability it would be much more difficult otherwise, since establishing positivity of excess intersections is in general a hard problem. Bernstein’s trick is a nontrivial example of an elementary argument faring better than a formidable machinery. 8 I. INTRODUCTION

3. Organization Part 0 and the ﬁrst chapter of part 1 have been designed as parts of a textbook, with many exercises. The goal was to develop as efﬁciently (and in as elementary way) as possible the theory needed for the proof of the results in the subsequent chapters. These latter chapters are more like those of a monograph; there are no exercises, but they do contain a number of examples. We now give a short description of each chapter.

In chapter III we develop the required theory of algebraic varieties. We tried to stress the geometric point of view following [Mum95]. A considerable portion of the results have been developed through exercises; we tried to ensure that no exercise is very difficult. We also included some material usually omitted from first courses on algebraic varieties, e.g. the relation between completeness and compactness (exercises III.30 to III.32), and the relation between algebraic and topological degrees of dominant maps (theorem III.44). In chapter IV we describe basic properties of intersection multiplicity (of n regular functions at a nonsingular point of an n dimensional variety), in particular how it can be computed using curves. After giving simple examples to illustrate that a satisfactory treatment of intersection multiplicity would need to incorporate non-reduced rings, we give a short introduction to closed subschemes of a variety. Chapter V is a compilation (with complete proof) of the properties of convex polyhedra which, together with the results of chapter III, constitute the foundation on which we introduce toric varieties in chapter VI. In chapter VI we mainly discuss those properties of toric varieties which are required for the results in the subsequent chapters. In chapter VII we prove Bernstein’s theorem and present some of its basic applications to convex geometry. In chapter VIII we apply Bernstein’s theorem to prove the weighted homogeneous and multi-homogeneous versions of Bézout’s theorem. Chapter IX contains the results on the generic bound and non-degeneracy conditions for intersection multiplicity at the origin, which we use in chapter X to compute the generic bounds and non-degeneracy conditions for the number of solutions of

n n k polynomial systems on k . It turns out that onecan as easily replace by an arbitrary Zariski open subset n of k - the results of chapter X are derived in this greater generality. In chapter X we also use the main results to derive generalizations of weighted homogeneous and multi-homogeneous versions of B´ezout’s theorem applicable to weighted degrees with possibly zero or negative weights. In chapter XI we apply the results from chapter IX to the study of Milnor numbers; in particular, we derive and generalize classical results of Kushnirenko on Milnor numbers. Chapters VII, X and XI end with selections of open problems (mostly combinatorial in nature). 3. ORGANIZATION 9

Chapter Dependencies

Chapter V Polytopes

Chapter III Chapter VI Varieties Toric varieties

Chapter VII Chapter IV Chapter VIII Bernstein- Intersection (Weighted) Kushnirenko multiplicity B´ezout’s theorem theorem

Chapter IX Chapter XI Generic intersection Generic Milnor multiplicity number

Chapter X Generic number of n solutions on k CHAPTER II

A brief history of points of inﬁnity in geometry

In this chapter we give a brief historical overview of the concept of points at inﬁnity in geometry and the subsequent introduction of homogeneous coordinates on projective spaces.

1. Points at infinity Points at infinity seem to have first cropped up in Johannes Kepler’s work on conics in Ad Vitellionem paralipomena quibus astronomiae pars optica traditur1 (1604). It is in this text that Kepler introduces the term focus2 to denote each of the (unique) pair of points inside a conic such that the rays from any point on the conic make equal angles to the tangent at that point. For a circle the foci coincide at the center, and they separate as the circle deforms into an ellipse. As one continues to deform the ellipse so that in the end it turns into a parabola, Kepler concludes that “In the Parabola one focus ... is inside the conic section, the other to be imagined either inside or outside, lying on the axis at an infinite distance from the first, so that if we draw the straight line ... from this blind focus to any point ... on the conic section, the line will be parallel to the axis ...” [FG87, pp. 186-187], see fig. 1.

FIGURE 1. Foci of families of conics

Another inspiration, albeit indirect, of points at infinity is the theory of linear perspective. Application of perspectives were already present in early fourteenthcentury paintings from Italy [And07, Chapter I], the earliest surviving written account of geometric construction of perspective being Leon Battista Alberti’s De Pictura (1435). By the seventeenth century there were numerous treatises on perspective. In 1639 Girard Desargues, who had worked as a military engineer and written on perspective, circulated fifty copies of his Brouillon project d’une atteinte aux evenmens des rencontres du cone avec un plan (“Rough draft of an essay on the results of taking plane sections of a cone”). At the very beginning of Brouillon project Desargues introduced the notion that parallel lines intersect at a point at infinity and parallel planes intersect

1“Literally ‘Things omitted by’ (or ‘Supplements to’) ‘Witelo with which the optical part of astronomy is concerned’. [. . .] Witelo’s Perspectiva, probably written in the 1270s, appeared in several new editions in the sixteenth century, and seems to have been the standard textbook on Optics” [FG87, pp. 221-222]. 2Focus is the Latin word for hearth. “Since light was reﬂected to the focus, . . . the focus of the mirror was the position in which one would place the material one wished to burn” [FG87, p. 222].

10 2. HOMOGENEOUS COORDINATES 11 at a line at infinity; constructing essentially the projective plane P2(R) and the three dimensional projective space P3(R) over R. He made extensive use of the lines and planes at infinity to give a unified treatment of families of parallel lines and families of lines through a common point. The subject of projective geometry was born in Brouillon project.

2. Homogeneous coordinates The birth however, went practically unnoticed. Desargues’ manuscript was thought to have been lost and it did not inspire much new work (other than Blaise Pascal’s Essay pour les coniques (1640) which contains Pascal’s theorem on conics). Projective geometry was revived in the nineteenth century largely due to Jean-Victor Poncelet, who fought in Napoleon’s army in the battle of Krasnoi in November, 1812, and then was a prisoner of war in Saratov till Napoleon’s defeat in mid 1814. In the prison “he occupied himself summarising all he knew of the mathematical sciences in notebooks that he then distributed to his fellow prisoners who wanted to finish an education disrupted by the incessant military campaigns” [Gra11, p. 13]. In the process he discovered, and upon his return to France, championed, the unifying aspect of projective geometry (as opposed to the “analytic geometry” of RenéDescartes). A fundamental tool of this new geometry was the duality between points and lines on the plane. Initially applied by Charles Julien Brianchon and Poncelet to conics, the duality principle was extended to all planar curves by Joseph Diaz Gergonne3. All the details of the duality principle however were not clear, e.g. even though the principle suggests that dualising twice one should get back the original curve, it was soon discovered that dualising a curve of degree higher than two results in a curve of degree higher than that of the original curve. Poncelet had some ideas about resolving this paradox by taking into account the effects of cusps and double points on a curve, but his ideas were not very precise. The resolution came through the algebraic treatment of projective geometry by August Möbius in Der Barycentrische Calcul¨ (1827). Möbius observed that weights w0, w1 placed at the ends of a (weightless) rod uniquely determines a point P on the rod, namely their barycenter, i.e. the center of gravity; the ordered pair [w0 : w1] (we write it in this way to distinguish from the Cartesian coordinates of P ) are the barycentric coordinates of P . It is straightforward to work out the relation between the Cartesian and barycentric coordinates, e.g. if we identify the rod with the closed interval [a,b] on the real line, then the barycentric coordinates [w : w ] of x [a,b] satisfies: 0 1 ∈ aw + bw x = 0 1 w0 + w1 This formula can be readily extended to allow for w0 and w1 to be zero or negative. It follows that each point on the real line has barycentric coordinates, see fig. 2. It is also clear that the barycentric coordinates are homogeneous, i.e. [w0λ : w1λ] denote the same point as [w0 : w1] for every nonzero λ R. Finally, ∈ note that if w0 + w1 =0, then [w0 : w1] does not correspond to any point on the line; Möbius defined it as a point lying at infinity.

[2 : −1] [1 : 0] [0.5 : 0.5] [0 : 1] [−1 : 2]

−1 0 0.5 1 2

FIGURE 2. Barycentric coordinates in dimension one with respect to the interval [0, 1]

In dimension two one starts with a triangle ∆; assume for convenience that the vertices of ∆ are the points with Cartesian coordinates (0, 0), (1, 0) and (0, 1). Then the Cartesian coordinates (x, y) and the barycentric coordinates [w0 : w1 : w2] of a point P on the plane with respect to ∆ are related as follows: w w x = 1 , y = 2 w0 + w1 + w2 w0 + w1 + w2

3“. . . it is one thing to realise that dualising a ﬁgure is a good way to obtain new theorems, which is what Poncelet did, and quite another thing to claim that points and lines are interchangeable concepts which must logically be treated on a par. This was the view that Gergonne put forward in 1825. Interpreted in such generality, Gergonne’s principle of duality is one of the most profound and simple ideas to have enriched geometry since the time of the Greeks . . . ” [Gra11, p. 55]. 12 II. A BRIEF HISTORY OF POINTS OF INFINITY IN GEOMETRY

2 [0.5 : −1 : 1.5] [−1.5 : 1.5 : 1]

[0 : 0 : 1] 1

0 [1 : 0 : 0] [0 : 1 : 0]

[2.5 : −1 : −0.5] [0.5:1: −0.5] -1 -2 -1 0 1 2

FIGURE 3. Barycentric coordinates in dimension two with respect to ∆

As in the case of the real line, the barycentric coordinates of the points at infinity are [w0 : w1 : w2] with w0 + w1 + w2 =0. Möbius observed that many computations with barycentric coordinates become simpler upon a change of coordinates of the form [w : w : w ] [w + w + w : w : w ] 0 1 2 7→ 0 1 2 1 2 These new coordinates are nowadays usually denoted as homogeneous coordinates. In particular, the equation of the line ax+by+c =0 changes in the barycentric coordinates to aw1 +bw2 +c(w0 +w1 +w2)=0, and this in turn becomes aw1 +bw2 +cw0 =0 in homogeneous coordinates. And in homogeneous coordinates the pointsat infinity are described by w0 =0. Ashewasfinishing Der Barycentrische Calcul¨ , Möbius heard of the duality between points and lines studied by the French geometers, and noticed that the homogeneous coordinates gives a natural algebraic approach to duality, namely the line aw1 + bw2 + cw0 =0 corresponds simply to the point with homogeneous coordinates [a : b : c], and vice versa. This automatically ensured that concurrent lines go to collinear points under duality and that dualizing twice one gets back to the original curve. Julius Plücker, possibly independently of Möbius, gave an analogous theory of homogeneous coordinates in 1830, and later used it to completely resolve the duality paradox. The homogeneous coordinates were soon extended to higher dimensions, which opened the door to algebraic study of higher dimensional projective spaces.

3. Projective space

n+1 k Take an arbitrary ﬁeld k and ﬁx coordinates (x0,...,xn) on , n 0. The n-dimensional projec-

n n+1≥ k tive space P over k is theset of lines (with respectto (x0,...,xn)) in throughthe origin. Every point n+1 (a ,...,a ) k 0 determines a uniqueline through the origin which passes throughit; the homoge- 0 n ∈ \{ } neous coordinate of this line is [a0 : : an]. For each j =0,...,n, let Uj := [a0 : : an]: aj =0 . The map ··· { ··· 6 } (a ,...,a ) [a : : a : 1 : a : : a ] 1 n 7→ 1 ··· j−1 j ··· n

n k gives a one-to-one correspondence between k and Uj. In the case that = R or C, one can use this correspondence to induce a topology on Uj (by declaring a subset of Uj to be open if and only if its pre-image in Rn or Cn is open). It is straightforward to check that these topologies are compatible (i.e. they induce n n the same topology on their intersections), and accordingly turns P = j=0 Uj into a manifold. For a

n k general k, the usual topology put on is the Zariski topology, in which the closed subsets are zero-sets of n S n systems of polynomials, and the identiﬁcation of the Uj with k is used to give P the structure of an al-

gebraic variety over k. We review algebraic varieties and Zariski topology in chapter III. The complement n Hj of Uj in P is the set of lines (through the origin) which lie on the j-th coordinate hyperplane, so that

n n n k if we identify k with U0, then the set of points at inﬁnity, i.e. the complement of in P , is precisely n H0 := [0 : a1 : : an] , and the homogeneous coordinates on P are precisely those introduced by { ··· } n−1 M¨obius. Note that H0 is naturally isomorphic to P . 3. PROJECTIVE SPACE 13

n n Proposition II.1. The closure in P of each straight line on k has a unique point at inﬁnity. Two planar lines intersect at a common point at inﬁnity if and only if they are parallel.

n PROOF. Here we treat the case that k = C and the topology on P is that induced from the Euclidean n 2n n topology on C = R ; see exercise III.13 for the case of general k and Zariski topology on P . Let ∼ n L = (a1,...,an)+ t(b1 + ... + bn): t C be a line on C = U0. In homogeneous coordinates { ∈ } ∼ L = [1 : a1 + tb1 : : an + tbn]: t C { ··· ∈ } Therefore the set of all points on Cn+1 which correspond to points on L is ′ L = (s,sa1 + stb1,...,san + stbn): s,t C,s =0 { ∈ 6 } ′ The points on the closure of L which correspond to points on H0 are precisely those coming from limits of sequences (sk,tk) such that sk 0 and sktk λ C 0 , i.e. the set of points (0, λb1,...,λbn), → → ∈ \{ } n λ C 0 . Since all these points correspond to the single point [0 : b1 : : bn] on P , this proves both assertions∈ \{ of} the proposition. ··· Proposition II.1 shows that the projective space incorporates the intuition from the theory of perspectives that each straight line has a point at inﬁnity, and two parallel lines intersect at a point at inﬁnity. The

connection of projective spaces with the afﬁne B´ezout problem comes from the following property (which k you will prove in exercise III.12): if k is algebraically closed, then for each polynomial f [x1,...,xn], there is a one-to-one correspondence between the following sets: ∈

the set of points at infinity on the clo- n n n x k 0 : ld(f)(x)=0 sure in P of x k : f(x)=0 { ∈ \{ } } { ∈ } where ld(f) is the leading form of f. This correspondence provides the geometric explanation for the condition from Bézout’s theorem (theorem I.2) under which the Bézout estimate for the number of solutions is exact. Our proof of Bézout’s theorem will consist of showing that

given n polynomials f1,...,fn k[x1,...,xn], the number of isolated points (counted with • ∈ n appropriate multiplicity) on the intersection of the closures in P of fi = 0 is at most the { } product of the degrees of the fi, and

n k if k is algebraically closed, then this bound is attained with points inside if and only if there • is no “intersection at infinity” on Pn i.e. f =0 = f =0 { i } { i } i i \ \ Most of the mathematics of this book takes place on toric varieties (see chapter VI), a class of algebraic varieties of which the projective space is a special case. We will study natural analogues of Bèzout’s theorem on different classes of toric varieties, and Bézout’s theorem will fall out as a special case (see corollary VIII.3). Part 0

Preliminaries CHAPTER III

Quasiprojective varieties over algebraically closed ﬁelds

1. Overview of classical results from algebra In this chapter we give a quick overview of the properties of algebraic varieties that we use in this book. In this section we go overthe basic results that we are going to use without any proof. The rest of the chapter is essentially self-contained, provided one is willing to work through the exercises. As everywhere

else in this book, unless explicitly stated otherwise in this chapter k is an algebraically closed ﬁeld. The ﬁrst two results, which are due to D. Hilbert, are part of the foundation of modern algebraic geometry, and are used in this chapter implicitly or explicitly from the very beginning.

THEOREM III.1 (Hilbert’s Basis Theorem). The ring of polynomials in ﬁnitely many variables over a Noetherian ring is also Noetherian.

n k Given a subset X of k we write I(X) for the set of all polynomials in [x1,...,xn] which are zero at all points of X; it is straightforward to check that I(X) is an ideal of k[x1,...,xn]. Conversely, if q is

n k an ideal of k[x1,...,xn], we write V (q) for the set of points on on which each element of q is zero; it is straightforward to check that I(V (q)) q and V (I(X)) X. The following result describes the basic correspondence between I( ) and V ( ). ⊇ ⊇ · ·

THEOREM III.2 (Hilbert’s Nullstellensatz). Assume k is algebraically closed. Then for each ideal q k of k[x1,...,xn], I(V (q)) is the radical of q. In particular, the maximal ideals of [x1,...,xn] are of the n form I(a) for a k . ∈ The “principal ideal theorem” of W. Krull is a fundamental tool in algebraic treatments of dimension. We use the following special case frequently, starting from the proof that the dimension of the set of common zeroes of k polynomials in n-variables has dimension at least n k (theorem III.20). − THEOREM III.3 (A special case of Krull’s Principal Ideal Theorem). Let R be a finitely generated integral domain over a field k, f be a nonzero element of R and p be an isolated prime ideal of the principal ideal fR of R generated by f (i.e., p is minimal among the prime ideals containing fR). If f is not a unit, then tr. d. (R/p)=tr. d. (R) 1 (where tr. d. is the transcendence degree over k). k k − k We also use some of the basic properties of localization and completion of Noetherian rings. Short introductions to these notions are provided in appendix B.4 and section III.12. Theorems III.4 and III.5 are used in our proof that near a nonsingular point a variety is a “locally complete intersection” (corollary III.37) and theorem III.6 is used for the first time in the proof that the local ring at a nonsingular point of a variety is isomorphic to a power series ring (theorem III.42). We use theorems III.4 and III.6 only in the case that R is the ring of rational functions in n variables (for some n 1) over a field whose denominators have nonzero constant terms. ≥

THEOREM III.4 (Krull [AM69, Corollary 10.18]). Let R be a Noetherian local ring with maximal k ideal m. If q is an ideal of R, then q = k≥0(q + m ).

THEOREM III.5 (Basic properties ofT formal power series rings). Let R := k[[x1,...,xn]] be the ring of formal power series over a ﬁeld k. Let m be the ideal of R generated by x1,...,xn. (1) m is the unique maximal ideal of R. Every element in R m is invertible. \ (2) If f1,...,fn m are such that the linear parts of the fj are linearly independent (over k), then the map x ∈f , j =1,...,n, induces an isomorphism of R. j 7→ j

15 16 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

THEOREM III.6 (Exactness of completion [AM69, Proposition 10.12]). Let I J be ideals of a ⊂ Noetherian ring R and let R¯ := R/I. Let Rˆ (respectively (R¯ˆ) be the completion of R (respectively R¯) with ¯ ¯ˆ ˆ ˆ respect to J (respectively JR). Then R ∼= R/IR. Elsewhere in this book (but not in this chapter) we also use that formal power series rings (in finitely many variables over a field) are Noetherian; however, we provide a proof of this fact in theorem B.40. The next two results state fundamental properties of separable extensions of a field. These are used in exercise III.25 to show that every irreducible variety is birational to a hypersurface, which is in turn used in our proof that the dimension of a variety can be read off from the dimension of tangent spaces at points on it (proposition III.34). Theorem III.8 is not used anywhere else; the only other place we use theorem III.7 is in the proof of the defining properties of degree of a morphism (theorem III.44).

THEOREM III.7 (Primitive Element Theorem). If L is a finite separable extension field of a field K, then there is α L such that L = K(α) (i.e. every element of L can be expressed as a polynomial in α with coefficients∈ in K).

THEOREM III.8 (F. K. Schmidt [ZS75a, Chapter II, Theorem 31]). If k is a perfect field, then any finitely generated field extension K/k is separably generated, i.e. there are x ,...,x K which are 1 d ∈ algebraically independent over k, and K is a separable algebraic extension of k(x1,...,xn). 2. Quasiprojective Varieties

n+1 n+1 k Let (x0,...,xn) be a system of coordinates on k , n 0. Consider the relation on 0 , n+1 ≥ ∼ \{ } n 0, defined as follows: if a,b k 0 with coordinates (with respect to (x ,...,x )) respectively ≥ ∈ \{ } 0 n (a0,...,an) and (b0,...,bn), then a b if and only if there is λ k 0 such that aj = λbj for each j =0,...,n. It is straightforward to check∼ that is an equivalence∈ relation;\{ } the n-dimensional projective ∼ space Pn is the set of the equivalence classes of . We denote the equivalence class of containing a by ∼ ∼n [a0 : : an] and say that [a0 : : an] is the homogeneous coordinate of the point of P (k) determined by a.··· Recall that a polynomial··· is homogeneous if each of its monomials has the same degree. If f is a homogeneous polynomial in (x ,...,x ), then the set V (f) := [a : : a ]: f(a ,...,a )=0 of 0 n { 0 ··· n 0 n } zeroes of f is a well defined subset of Pn; we say that V (f) is the hypersurface of Pn determined by f. We leave it an exercise to check that there is a unique topology on Pn in which the basic closed subsets are intersections of hypersurfaces of Pn (exercise III.1); it is called the Zariski topology. A projective variety is a Zariski closed subset of the projective space (equipped with the topology induced from the Zariski topology of Pn)1. A quasiprojective variety is a Zariski open subset of a projective variety (also equipped with the Zariski topology induced from Pn). A subvariety of a quasiprojective variety X is a Zariski closed subset of X. The following theorem states the fundamental property that varieties can be “finitely presented” in terms of polynomials. THEOREM III.9. Let X be a subvariety of Pn. There are finitely many homogeneous polynomials f1,...,fk in (x0,...,xn) such that n X = V (f1,...,fk) := [x0 : : xn]: fj (x0,...,xn)=0 for each j =1,...,k P { ··· }⊆

n n+1 k PROOF. Let a = [a0 : ... : an] P and L(a) := (ta0,...,tan): t k be the “line in represented by a.” ∈ { ∈ }

Claim III.9.1. A polynomial in (x0,...,xn) vanishes on L(a) if and only if each of its homogeneous components vanishes on a. e

PROOF. Pick f k[x ,...,x ] such that f 0. Write f = f , where each f is ∈ 0 n |L(a) ≡ d=0 d d homogeneous of degree d. Then f(ta ,...,ta ) = e tdf (a) is a polynomial in t with infinitely 0 n d=0 d P many zeroes, so that it must be identically zero. Therefore fd(a)=0 for each d, as required. P n Now let X = i∈I V (gi) P , where each gi is a homogeneous polynomial in (x0,...,xn). Let ⊂ n+1 C(X) := L(a) be the cone over X in k . Hilbert’s basis theorem (theorem III.1), Nullstellensatz a∈X T 1TraditionallyS it was common to define a projective variety as an irreducible Zariski closed subset of the projective space (a set is “irreducible” if it cannot be expressed as a union of two closed proper subsets); however, our definition is also widely used. 2. QUASIPROJECTIVE VARIETIES 17

2 2 2 2 FIGURE 1. For k = C, the real points of X := V (x x x ) P can be pictured 0 − 1 − 2 ⊂ as a cross section of the real points of the cone C(X) over X in C3

(theorem III.2) and claim III.9.1 imply that the ideal I(C(X)) of polynomialsin (x0,...,xn) that vanish on C(X) is generated by ﬁnitely many homogeneous polynomials f ,...,f and that X = V (f ,...,f ) 1 k 1 k ⊆ Pn.

2.1. Regular functions, morphisms. If f,g are homogeneous polynomials in (x0,...,xn) of the same degree, then f/g is a well deﬁned function on the Zariski open subset Pn V (g) of Pn; we say \ that f/g is a rational function on Pn which is regular on Pn V (g). In general, a regular function on a \ quasiprojective variety X is a function φ : X k which can be “locally represented by rational functions,” i.e. there is an open cover U of X such that→ for each i, φ is the restriction to U of a rational function { i} |Ui i fi/gi such that gi does not vanish at any point of Ui; the set k[X] of regular functions on X has the natural

structure of a k-algebra. The following is straightforward to see from the deﬁnition of regular functions: Proposition III.10. If f is a regular function on a quasiprojective variety X, then V (f) := x X : f(x)=0 is Zariski closed in X. { ∈ }

Given J k[X], we denote by V (J) the “subvariety of X determined by J,” i.e. the set of points on X on which⊆ each f J vanishes; proposition III.10 implies that V (J) is indeed a subvariety of X. A morphism φ : X Y ∈of quasiprojective varieties is a continuous map with respect to Zariski topology such that for every→ Zariski open subset U of Y and every regular function h on U, the pullback h φ is a regular function on φ−1(U). An isomorphism is a bijective morphism whose inverse is also a morphism.◦ We say that a morphism φ : X Y is a closed embedding if φ(X) is a (closed) subvariety of Y and φ is an isomorphism between X and→φ(X). It is straightforward to check that a morphism φ : X Y induces

∗ →

k k via pullback a k-algebra homomorphism φ : [Y ] [X]. We say that φ is dominant or dominating if φ(X) is Zariski dense in Y . → Proposition III.11. If a morphism φ : X Y of quasiprojective varieties is dominant (in particular, if it is surjective), then φ∗ is injective. → PROOF. Given a nonzero regular function f on Y , the set Y V (f) is nonempty and, due to proposition III.10, Zariski open in Y . Since φ is dominant, φ(X) intersects\ Y V (f), so that U := φ−1(Y V (f)) is a nonempty Zariski open subset of X. Since φ∗(f) is nonzero on U\, the proposition follows. \ The converse of proposition III.11 is not true. Indeed, you will show in exercise III.6 that the only

n n n n k regular functions on P are constants, i.e. k[P ] = . It follows that if φ : P P is a constant → morphism that maps every point to a ﬁxed point on Pn, then φ∗ is injective even though φ is far from being dominant. 2.2. Exercises.

EXERCISE III.1. Show that the collection of Zariski closed sets satisfy the axioms of a topology, namely that it contains the empty set and the projective space itself, and it is closed under ﬁnite unions and arbitrary intersections. 18 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

n EXERCISE III.2. Given X P , let I(X) be the ideal of k[x0,...,xn] generated by all the homoge- ⊆ neous polynomials f in (x0,...,xn) such that f(x0,...,xn)=0 for each [x0 : : xn] X. Show that n ¯ ··· ∈ the closure of X in P under the Zariski topology is X := V (I(X)) = [x0 : : xn]: f(x0,...,xn)= { ··· 0 for each homogeneous f I(X) ; we say that X¯ is the Zariski closure of X in Pn. ∈ } EXERCISE III.3. This exercise illustrates a fundamental property of algebraically closed ﬁelds.

n k (1) Show that every non-constant polynomial over k in (x1,...,xn) vanishes at some point of .

[Hint: proceed by induction on n. Treat polynomials in (x1,...,xn) over k as polynomials in

xn over k[x1,...,xn−1], and use inductive hypothesis to reduce to the case of n =1.] (2) Deduce that for every homogeneous polynomial f in (x ,...,x ), n 1, the hypersurface 0 n ≥ V (f) Pn is nonempty. ⊂

(3) Show by examples that the preceding statements may be false if k is not algebraically closed.

1 EXERCISE III.4. Consider P with homogeneous coordinates [x0 : x1]. (1) Show that the map [x0 : x1] x1/x0 gives a one-to-one correspondence between U0 :=

1 7→ k P V (x0) and k. This correspondence is used to give the structure of a quasiprojective variety.\

(2) Show that the Zariski topology on k coincides with the coﬁnite topology (i.e. the topology in which open sets are precisely complements of ﬁnite subsets).

EXERCISE III.5. Prove that every regular function on a quasiprojective variety X is “Zariski contin-

uous,” i.e. continuous with respect to Zariski topology (recall from exercise III.4 that the Zariski topology k on k is the coﬁnite topology); in particular, morphisms from X are precisely regular functions on → X. Show that the only proper Zariski closed subsets of P1 are ﬁnite sets of points; conclude that every bijection of P1 is Zariski continuous. Show that there are Zariski continuous maps from P1 to P1 which are not morphisms.

3. Afﬁne varieties

3.1. Afﬁne space, basic open subsets of the projective space. For each j =0,...,n, Uj := [a0 : n n n { : an]: aj =0 = P V (xj ) is a Zariski open subset of P ; it is clear that P = j Uj. We say that ··· 6 } \ n n the Uj are basic open subsets of P . Each Uj has a one-to-one correspondence with k via the map S n

(3) k (x ,...,x ) [x : : x : 1 : x : : x ] U ∋ 1 n 7→ 1 ··· j j+1 ··· n ∈ j n The afﬁne space is k equipped with the Zariski topology induced by this correspondence. Unless stated n explicitly otherwise, in this book we always write k to denote the afﬁne space. Theorem III.13 below n shows that the Zariski topology on k does not depend on j from the correspondence in (3).

n n k Proposition III.12. k is irreducible, i.e. every pair of nonempty open subsets of has a nonempty n intersection. In particular, if two regular functions on k agree on some Zariski open subset, then they agree everywhere.

PROOF. Fix j, 0 j n. Since k is algebraically closed, the following can be proved by a straightforward induction on n≤(see≤ exercise III.3).

Claim III.12.1. Let H be a homogeneous polynomial in (x0,...,xn). Then V (H) Uj = if and only m ∩ ∅ if H = cx for some m 0 and c k 0 and m 0. j ≥ ∈ \{ } ≥ If F , F are homogeneous polynomials in (x ,...,x ) such that U V (F ) = for each i, then the 1 2 0 n j \ i 6 ∅ above claim implies that Uj V (F1F2) = , which proves the ﬁrst assertion of the proposition. The second assertion then follows from\ proposition6 III.10∅ .

n THEOREM III.13. The regular functions on k are precisely the polynomials in (x1,...,xn). Every n subvariety of k is an intersection of zero sets of ﬁnitely many polynomials in (x1,...,xn). n PROOF [Mum95, Proposition 1.11]. Denote the map k U from (3) by ψ . It is straightforward → j j to check that every polynomial in (x1,...,xn) corresponds to a regular function on Uj via ψj . We will now show that conversely every regular function on Uj comes from a polynomial in (x1,...,xn). Given a regular function H on Uj, let q be the ideal of k[x1,...,xn] consisting of all polynomials g such that 3. AFFINE VARIETIES 19

n n k gψ (H) is a polynomial on some nonempty open subset of k . For each a , there are homogeneous j ∗ ∈ polynomials F, G of same degree such that G(ψj (a)) = 0 and H = F/G on an open neighborhood U of ∗ 6 d ∗ d ψj (a). Let d := deg(F ) = deg(G). Write f = ψj (F/xj ) and g = ψj (G/xj ), i.e. f,g are constructed d d respectively from F/xj and G/xj by substituting xi for xi−1/xj for i = 1,...,j, and for xi+1/xj for i = j +1,...,n. Then g q and g(a) =0. It follows that V (q)= . The Nullstellensatz (theorem III.2) ∈ ∗ 6 ∅ n then implies that 1 q, i.e. ψj (H) agrees with a polynomial on some nonempty open subset of k . Since n ∈ ∗ k is irreducible (proposition III.12), it follows that ψj (H) is a polynomial, which completes the proof of the ﬁrst assertion of the theorem. The second assertion then follows from theorem III.9.

n 3.2. Affine varieties. An affine variety X is simply a subvariety of the affine space k . The ring

k[X] of regular functions on X is also denoted as the coordinate ring of X. The restriction of polynomials k in (x1,...,xn) to X induces an isomorphism k[X] ∼= [x1,...,xn]/I(X), where I(X) is the ideal of k[x1,...,xn] consisting of all polynomials which vanish on X (exercise III.8). Morphisms between two

afﬁne varieties are completely determined by the induced (k-algebra) homomorphisms of their coordinate rings. In particular, two afﬁne varieties are isomorphic if and only if their coordinate rings are isomorphic

as k-algebras (exercise III.9). Affine varieties are “building blocks” of the category of varieties: Proposition III.14. Every variety has a finite open cover by affine varieties.

n PROOF. Let V be a quasiprojective variety in P . Then V is covered by V Uj , j = 0,...,n, ∩ and V Uj are by definition open subsets of affine varieties. Therefore it suffices to prove that an open ∩ n subset U of a subvariety X of k in has a finite open cover by affine varieties. Theorem III.13 implies that

U = X V (g1,...,gl) for some polynomials gi. Since U = i(X V (gi)), the result follows from the fact that the\ complement of an hypersurface in an afﬁne variety is also\ an afﬁne variety (exercise III.11). S

n n n k

Identify k with the open subset U0 of P . Let X be a subvariety of defined by the ideal q of k k[x1,...,xn] vanishing on X. For each f [x1,...,xn], the homogenization of f in (x0,...,xn) is ˜ deg(f) ∈ the homogeneous polynomial f := x0 f(x1/x0,...,xn/x0). Given an ideal q of k[x1,...,xn], the Zariski closure of V (q) in Pn is the subvariety of Pn defined by ˜ (4) q˜ := f : f q k[x ,...,x ] { ∈ }⊆ 0 n (exercise III.12). A hypersurface of the affine (respectively, projective) space is a subvariety defined by a single (respectively, single homogeneous) polynomial, and a hyperplane is a hypersurface defined by a linear polynomial. Exercise III.12 implies that the Zariski closure of an affine hypersurface (respectively, hyperplane) is a projective hypersurface (respectively, hyperplane). 3.3. Exercises.

n k EXERCISE III.6. Show that k[P ] = [Hint: apply theorem III.13 to Ui and Uj for i = j; then

6 k compute k[Ui] [Uj ]]. Conclude that the only possible morphism from a projective space into an afﬁne space is the trivial∩ map which maps everything to a point.

∗ k EXERCISE III.7. Let φ : X Y be a morphism between afﬁne varieties such that φ : k[Y ] [X] is injective. Show that φ is dominant.→ [Hint: if φ is not dominant, then the closure of φ(X) is a→ proper subvariety of Y and there is a nontrivial regular function on Y whose pullback is zero on X.] Contrast this with proposition III.11.

n k

EXERCISE III.8. Let X be a subvariety of k , with coordinates (x1,...,xn). Show that [X] ∼= k k[x1,...,xn]/I(X), where I(X) is the ideal of [x1,...,xn] consisting of all polynomials which vanish on X [Hint: mimic the proof of theorem III.13].

EXERCISE III.9. Prove that two afﬁne varieties are isomorphic if and only if their coordinate rings are

isomorphic as k-algebras. Show that this is not true without the “afﬁneness” condition. 2

EXERCISE III.10. Let R be a finitely generated k-algebra which is reduced . Show that there is an affine variety X such that k[X] ∼= R. Exercise III.9 shows that X is unique up to an isomorphism, and therefore the categories of affine varieties and finitely generated reduced k-algebras are equivalent.

2A ring is reduced if it does not have any nonzero nilpotent elements. 20 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

EXERCISE III.11. Let f be a nonzero polynomial in (x1,...,xn).

n k (1) Show that k V (f) is an affine variety with coordinate ring [x1,...,xn, 1/f]. [Hint: consider \ n+1 (x1,...,xn,y) k : f(x)y =1 .] (2){ More generally, if∈X is an affine variety} and f is a regularfunction on X, then show that X V (f) \ is also an affine variety whose coordinate ring is the localization (see appendix B.4) k[X]f of

k[X] at f.

n n EXERCISE III.12 (Projective closure of afﬁne varieties). Identify k with the open subset U0 of P .

Given an ideal q of k[x1,...,xn], deﬁne q˜ as in (4). Prove that

n n n n k (1) V (q˜) is the Zariski closure in P of V (q) k . In particular, is Zariski dense in P . (2) V (q) is Zariski open in V (q˜). ⊆

n ¯ n ˜ k (3) If X = V (f) k for some f [x1,...,xn], then the Zariski closure X of X in P is V (f). ⊆ ∈

(4) If q is generated by homogeneous polynomials f ,...,f k[x ,...,x ], show that q˜ is also 1 k ∈ 1 k generated by f1,...,fk. In particular, f˜j = fj for each j, and X¯ = V (f˜1,..., f˜k) as a subvariety of Pn. [Hint: if q is generated by homogeneous polynomials, then show that f q if and only if each homogeneous component of f belongs to q.] ∈ ¯ ˜ ˜ (5) Show that in general, if X = V (f1,...,fk), then X $ V (f1,..., fk). [Hint: consider e.g. the 2 2 ˜ ˜ case that f1 = x1 and f2 = x2 x1. Show that V (f1,f2) k is a point, whereas V (f1, f2) consists of two points. Cf. example− I.3.] ⊂

n EXERCISE III.13. Let a = (a1,...,an),b = (b1,...,bn) k , b = 0, and L be the line a + bt :

n ∈ 6 { k t k . ∈ }⊂ (1) Show that ideal I(L) of polynomials vanishing on L is generated by linear polynomials of the form cixj ckxl (for appropriate ci,ck,j,l). Determine a set of generators of I(L) of this form. [Hint: at first− consider the case that a =0.] ¯ n n n Show that the closure L of L in P (where the inclusion k ֒ P is as in exercise III.12), is (2) → L 0 : b1 : : bn] . [Hint: use assertion (4) of exercise III.12.] ∪{ ··· } 2 2 (3) Conclude that two lines in k intersect at infinity in P if and only if they are parallel. EXERCISE III.14 (Converse to exercise III.12). Let V be a a subvariety of Pn defined by a homo-

n n k geneous ideal q of k[x0,...,xn]. The basic open subset U0 of P is isomorphic to with coordinates (x1/x0,...,xn/x0). Let q := f/xdeg(f) : f is a homogeneous polynomial in q 0 { 0 } Show that

(1) q0 is an ideal of k[x1/x0,...,xn/x0]. (2) V U is the subvariety of U deﬁned by q . ∩ 0 0 0 n EXERCISE III.15. Let S1,S2 be ﬁnite subsets of P such that S1 S2 = . ∩ ∅ (1) Show that there is a projective hypersurface X containing S1 but not containing any point of S2. [Hint: reduce to the case that S1 =1.] (2) If S

n n k this explains the statement of proposition III.12 that k is irreducible. A ﬁnite set of points in is irreducibleif and only if it consists of only onepoint. An irreducible component of X is a closed irreducible subset which is not properly contained in any other irreducible subset of X. Each point of a ﬁnite subset S

n n k of k is an irreducible component of S. If X = V (x1 xn) is the union of coordinate hyperplanes n ··· ⊂ of k , then the irreducible components of X are the coordinate hyperplanes V (xi), i = 1,...,n. In the preceding example note that that the ideal I(X) of X is x1 xn , which is precisely the intersection of the ideals of its irreducible components. This is in fact truh e··· for alli afﬁne varieties - since the ideal I(X) 5. RATIONAL FUNCTIONS AND RATIONAL MAPS ON IRREDUCIBLE VARIETIES 21

n of a subvariety X of k is radical, it can be represented as the intersection of ﬁnitely many prime ideals

(theorem B.16). There is a unique such representation I(X)= j pj which is minimal, and the irreducible components of X are precisely the zero sets V (pj ) of pj. In particular, X is irreducible if and only if I(X) is prime (exercise III.19). T

4.1. Exercises.

EXERCISE III.16. Show that a topologicalspace X is irreducibleif and only if any two of its nonempty n open subsets has a nonempty intersection. Deduce that k is irreducible for each n.

EXERCISE III.17. Let V1,...,Vk be closed subsets of a topological space X. If V is a closed irreducible subset of X such that V V , then show that V V for some j. ⊆ j j ⊆ j EXERCISE III.18. If U is a denseS subset of V , show that U is irreducible if and only if V is irreducible. Conclude that Pn is irreducible.

EXERCISE III.19. Let q be an ideal of k[x1,...,xn]. Show that V (q) is irreducible if and only if the radical of q is prime. Conclude that an afﬁne variety is irreducible if and only if its coordinate ring is an integral domain.

5. Rational functions and rational maps on irreducible varieties Let X be an irreducible quasiprojective variety. A rational function on X is a regular function on a nonempty Zariski open subset of X. Formally, a rational function is an equivalence class of pairs (f,U), where U is a nonempty Zariski open subset of X and f is a regular function on U, and the equivalence ′ ′ ′ ′ relation is deﬁned as follows: (f,U) (f ,U ) if and only if f and f agree on U U ; we write k(X) for the set of rational functions on X. The∼ irreducibility of X forces every pair of its∩ nonempty Zariski open

subsets to have a nonempty intersection, which in turn implies that k(X) is a field (exercise III.21). If X is k in addition affine, then k(X) can be identified with the field of fractions of [X].A rational map φ from a variety X to a variety Y is a morphism from a nonempty open subset U of X to Y ; usually a rational map is denoted by a broken arrow φ : X 99K Y It is called dominant if the image of U is dense in Y , and birational if there are open subsets U ′ of U and ′ V of Y such that φ U ′ : U V is an isomorphism. We say that X and Y are birational or X is birational to Y if there is a birational| map→ from X to Y . Fields of rational functions of birational irreducible varieties are isomorphic (exercise III.24). Every irreducible variety is birational to a hypersurface (exercise III.25).

5.1. Exercises.

n k EXERCISE III.20. Compute k(X) for the following cases: X = ; X = Ui Uj , for i = j, where n n ∩ 6 Uj are afﬁne open subsets of P from section III.3; X = P .

EXERCISE III.21. Let X be an irreducible variety. k (1) If U is a nonempty open subset of X, show that k(X)= (U).

(2) Show that k(X) is a field. k (3) If X is also affine, then show that k(X) can be identified with the field of fractions of [X]. Show that this may not hold if X is not affine.

(4) Which parts of the proof that “k(X) is a ﬁeld” go wrong when X is reducible? What is an

appropriate generalization of k(X) to the case of reducible varieties?

EXERCISE III.22. Let φ : X Y be a dominant rational map. Show that if X is irreducible, then so is Y . →

EXERCISE III.23. Let φ be a rational map from X to Y , both irreducible varieties. Show that

∗ k .(If φ is dominant, then φ induces an injection φ : k(Y ) ֒ (X (1)

→ k Conversely, every injection k(Y ) ֒ (X) is induced by a dominant rational map; contrast this (2) with proposition III.11. → 22 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

EXERCISE III.24. Let X and Y be irreducible varieties. Show that X is birational to Y if and only if k k(X) ∼= (Y ). EXERCISE III.25. Let X be an irreducible variety. In this exercise you will show that there is n 1 n ≥ and a polynomial f in (x ,...,x ) such that X is birational to V (f) k .

1 n ⊂ k (1) Use exercises III.8 and III.21 to show that k(X) is a ﬁnitely generated ﬁeld extension of .

(2) Use theorems III.7 and III.8 to show that there are x ,...,x k(X) such that 1 n ∈ (a) x1,...,xn−1 are algebraically independent over k,

(b) xn is algebraic over k(x1,...,xn−1),

n ∈ 1 n k (d) k(X)= (x1,...,xn). n (3) Compute the ﬁeld of rational functions of Y := V (f) k and use exercise III.24 to conclude that X is birational to Y . ⊂

6. Product spaces, Segre map, Veronese embedding 6.1. Product spaces, Segre map. The most basic topology on products of topological spaces is the “product topology,” whose open sets are unions of products of open subsets of each factor. The product topology on algebraic varieties however is very restrictive:

Example III.15. Since the only proper closed subsets of k are ﬁnite sets of points (exercise III.4), the

k k proper closed subsets of k = under the product topology are finite unions of sets of the form ∼ × S1 S2, where at least one of the Si is finite. Given a polynomial f in (x, y), it follows that the curve × 2 V (f) on k is closed in the product topology if and only if f can be expressed as g(x)h(y) for polynomials g,h in one variable. A more natural topology on algebraic varieties is constructed as follows: consider projective spaces m n P , P with homogeneous coordinates respectively [x0 : : xm] and [y0 : : yn]. A polynomial f in ··· ··· (x0,...,xm,y0,...,yn) is called bi-homogeneous of bi-degree(d, e) in the xi and yj if each monomial that appears in f has degree d in the x-variables and e in the y-variables. The set of zeroes of a bi-homogeneous polynomial on Pm Pn is well-defined, and the Zariski topology on Pm Pn is by definition the (unique) topology whose basic× closed subsets are intersections of zero sets of bi-homogeneous× polynomials. If V, W are quasiprojective varieties with their Zariski closures being subvarieties respectively of Pm, Pn, j = 1, 2, then the Zariski topology on V W is the topology induced from the Zariski topology on × Pm Pn. An equivalent formulation of the Zariski topology on product spaces can be given via the Segre × m n k map s : P P P , where k := (m + 1)(n + 1) 1. It is defined as follows: let zij , 0 i m, × → − ≤ ≤ 0 j n, be a system of homogeneous coordinates on Pk; then ≤ ≤ (5) s : ([x : : x ], [y : : y ]) the point z = x y 0 ··· m 0 ··· n 7→ ij i j The proof of the following result is left as an exercise (exercise III.27) Proposition III.16. The image of Pm Pn under the Segre map is the subvariety Z of Pk defined by × m n (homogeneous) quadrics of the form zij zkl zilzkj and s induces an homeomorphism between P P and its image. − × Proposition III.16 implies that the Zariski topology on a product space agrees with the topology induced by the Zariski topology on the projective space upon identification of the product space with its image under the Segre map. In particular, we view Pm Pn as a projective variety by identifying it with × the subvariety s(Pm Pn) of Pk. × n+1 6.2. Veronese embedding. Let d be a positive integer and d := α = (α0,...,αn) Z : V { ∈ ≥0 n α = d be the set of exponents of monomials of degree d in (x ,...,x ). The degree-d Veronese j=0 j } 0 n n |Vd|−1 map νd : P P is given by P → (6) ν :[x : : x ] [xα : α ] d 0 ··· n 7→ ∈Vd n |Vd|−1 Proposition III.17. νd is a closed embedding, i.e. νd(P ) is a subvariety of P , and νd is an isomor- n n phism between P and νd(P ). 7. COMPLETENESS AND COMPACTIFICATION 23

PROOF. It is straightforward to see that νd can be expressed as a composition δ s d π Pn Pn Pn d P(n+1) −1 99K P|Vd|−1 → ×···× → n where δ : x (x,...,x) is the diagonal map to the product of d copies of P , sd is the Segre map on the d 7→ factors, and π is a projection which omits “redundant” coordinates of sd δ (i.e. for each α d, π retains α ◦ ∈V only one of the coordinates of sd δ equalling x ). Since δ and sd are closed embeddings (exercises III.27 and III.29), so is s δ. It is then straightforward◦ to check that π s δ is a closed embedding as well. d ◦ ◦ d ◦ We use proposition III.17 to show that the complement of a projective hypersurface in a projective variety is affine. It is the projective analogue of the fact that the complement of a hypersurface in an affine variety is also affine (exercise III.11). For this result we will use the terminology of graded rings from

appendix B.9. The degree of polynomials induce the structure of a graded ring on S := k[x0,...,xn]

(example B.22). If f is a homogeneous polynomial, then the localization Sf = k[x0,...,xn, 1/f] is also a graded ring with a well-deﬁned degree given by: deg(g/f k) := deg(g) k deg(f) for each g S − ∈ (proposition B.25), and the subring of zero degree elements of Sf is denoted by S(f). If I is a homogeneous ideal of S, i.e. if I is generated by homogeneous polynomials, we write I(f) for the ideal of S(f) generated by I, i.e. I := g/f k : g is a homogeneous element of I, k 0, deg(g)= k deg(f) . (f) { ≥ } n Proposition III.18. Let X be a subvariety of P and f be a nonzero homogeneouspolynomialin (x0,...,xn).

Then X V (f) is an afﬁne variety with coordinate ring isomorphic to k[x0,...,xn](f)/I(X)(f), where I(X) is the\ ideal of homogeneous polynomials identically vanishing on X.

PROOF. Let d := deg(f). If d =0, then the result is trivial. If d =1, then after a linear isomorphism n (which preserves degree, and therefore induces an automorphism of P ) of k[x0,...,xn] we may assume that f = x . But then X V (f) is the basic open set U , which is isomorphic to the afﬁne space with ∼ 0 \ 0 coordinate ring k[x1/x0,...,xn/x0] (theorem III.13), as required. In the case that d > 1, let Y := |Vd|−1 νd(X) P , then νd maps X V (f) isomorphically onto Y V (h) for a linear polynomial h on νd, and the⊂ result follows from the d =1\ case [sort out the details carefully].\ 6.3. Exercises.

m n m k EXERCISE III.26. Show that k (with the topology induced from the Zariski topology of P n m+n × × P ) is homeomorphic to k (with the usual Zariski topology, i.e. the topology induced from the Zariski topology of Pm+n). EXERCISE III.27. Prove proposition III.16.

m n EXERCISE III.28. Show that the Zariski topology on k P can be described as follows: let × m (x1,...,xm) be a system of coordinates polynomial coordinates of k and [y0 : : yn] be homo- n m n ··· geneous coordinates on P . Then the closed sets on k P are (finite) intersections of zero sets of × polynomials f k[x ,...,x ,y ,...,y ] which are homogeneous in the y-variables, i.e. of the form ∈ 1 m 0 n f = (coefficient)xα1 xαm yβ0 yβn 1 ··· m 0 ··· n β0+···X+βn=d for some d 0 (in this case d is said to be the degree of f in (y ,...,y )). ≥ 0 n EXERCISE III.29. Let φ : X Y be a morphism of varieties. Show that its graph, i.e. the set (x, φ(x)) : x X is a subvariety→ (i.e. a Zariski closed subset) of X Y and it is isomorphic to X. {[Hint: reduce to∈ the case} that X and Y are affine. Then use exercise III.26×.]

7. Completeness and compactiﬁcation While studying manifolds, it is often necessary to compactify them in order that the intersection theory

of submanifolds is well behaved. The Noetherianness of ﬁnitely generated k-algebras (Hilbert’s basis theorem) implies that all algebraic varieties satisfy the usual deﬁnitions of compactness (exercise III.30). The property which plays in the case of varieties the role similar to that of compactness in the case of manifolds is completeness. A variety X is called complete if for every variety Y , the projection map X Y Y is closed, i.e. it maps closed sets on to closed sets. For many “natural” topological spaces (e.g. the× → 24 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

Euclidean topology), compactness is equivalent to completeness (exercise III.32). A compactiﬁcation3 of a variety X is a complete variety X¯ containing an open dense subset isomorphic to X. The projective space is complete (exercise III.35). Since every closed subset of a complete variety is complete (exercise III.33), every quasiprojective variety has a compactiﬁcation (namely its closure in an ambient projective space).

7.1. Exercises.

EXERCISE III.30. Recall that a topological space is called compact if each of its open covers has a finite subcover, and it is called sequentially compact if every infinite sequence of points has a convergent subsequence. Show that (1) Every quasiprojective variety is compact. [Hint: start with the affine varieties, use the Noetheri- anness of their coordinate rings.] (2) Every quasiprojective variety is sequentially compact. [Hint (the following strategy requires the notion of dimension covered in section III.8): start with a sequence of points on a quasiprojective variety X and take its closure Z in X. Reduce to the case that Z is irreducible and affine. Then consider two cases: (i) there is a nonzero regular function on Z that vanishes on an infinite subsequence, and (ii) every nonzero regular function vanishes on only finitely many points from that sequence.]

EXERCISE III.31. We say that a topological space X is first-countable if for every point x X, there is a sequence of open neighborhoods U of x in X such that every open neighborhood U ∈of x in X { j}j contains some Uj . Show that the following are equivalent: (1) X is first countable. (2) For every point x X, there is a sequence of open neighborhoods Uj j of x in X such that for every open neighborhood∈ U of x in X, there is N such that U contains{ }U for each j N. j ≥ EXERCISE III.32. Let X be a first countable topological space. Consider the following properties: (1) X is sequentially compact. (2) For every first countable topological space Y , the projection map X Y Y is closed with respect to the product topology on X Y . × → × 4 Show that (1) implies (2). If X is a first countable T1-space , show that (2) implies (1).

EXERCISE III.33. Show that (1) a complete subset of a variety is also (Zariski) closed, (2) every subvariety of a complete variety is complete, (3) the image of a complete variety under a morphism is also complete.

EXERCISE III.34. If f is a regular function on a complete variety X, show that f takes ﬁnitely many k values [Hint: consider the subvariety Z of X k given by tf = 1, where t is the coordinate on ].

Conclude that × k (1) Every regular function on a complete irreducible variety is constant, i.e. k[X]= . (2) If f is a morphism from an irreducible complete variety to the affine space, then the image of f is a point. (3) If X¯ is a compactification of an affine variety X and C is a complete subset of X¯ consisting of infinitely many points, then C (X¯ X) = . ∩ \ 6 ∅ EXERCISE III.35. In this exercise you will prove that Pn is complete (following the structure of [Mum95, Proof of Theorem 2.23]).

n m m k (1) Show that it suffices to prove that the projection map π : P k is closed for each m. [Hint: every variety has a finite open covering by closed sub×sets of→ the affine space (proposition III.14).]

3We would have liked to use “completion” instead of “compactiﬁcation”; however, it might suggest a (misleading) connection with the notion of completion of local rings or modules. 4 A topological space is T1 if for every pair of distinct points, each has an open neighborhood not containing the other. 8. DIMENSION 25

m (2) Let π be as in the preceding step. Choose afﬁne coordinates (Y1,...,Ym) on k and homoge- n n m neous coordinates [X0 : : Xn] on P . Let Z be a closed subset of P k . Exercise III.28 ··· × implies that Z = V (f1,...,fk) for polynomials fj (X, Y ) k[X0,...,Xn, Y1,...,Ym] which ∈ m are homogeneous of degree dj 0 in the X-variables. Show that for each y k , y π(Z) ≥ d ∈ 6∈ if and only if there is d 1 such that m is contained in the ideal of k[X ,...,X ] gener- ≥ 0 n ated by f1(X,y),...,fk(X,y), where m is the maximal ideal of k[X0,...,Xn] generated by X0,...,Xn. (3) For each integer e, let Ve be the vector space of homogeneous polynomials of degree e in

m k (X0,...,Xn) (with Ve = 0 if e < 0) and let me := dimk (Ve). For each y , and d 0, let T d(y) be the linear map V V V deﬁned by (g (X∈),...,g (X))≥ d−d1 ··· d−dk → d 1 k 7→ k f (X,y)g (X). Let [T d(y)] be the matrix of T d(y) with respect to ﬁxed bases of V j=1 j j L L j d−dj and V . Show that T d(y) is surjective if and only if there is an m m -minor of [T d(y)] with P d d d L nonzero determinant. × m (4) Conclude that k π(Z) is open, as required. \ n n (5) What goes wrong with the proof if P is replaced by k ?

EXERCISE III.36. Let φ : X Y be a dominant morphism. Show that φ can be extended to a surjective morphism φ′ : X′ Y→where X′ is a variety which contains (an isomorphic copy of) X as a dense subset. [Hint: let X→¯ be a compactiﬁcation of X. Take X′ to be the closure of the graph (exercise III.29) of φ in X Y .] × 8. Dimension

The dimension dim(X) of an irreducible quasiprojective variety X is the transcendence degree over k k of the ﬁeld (X) of rational functions on X. In general dim(X) is the maximum of the dimensions of all its irreducible components. The dimension of a variety is an analogue of the dimension of smooth manifolds. In particular, when k = C and X is irreducible, it turns out that there is an open Zariski dense subset of X (consisting of nonsingular points (see section III.11) of X) which is a smooth manifold of (complex) dimension dim(X). We say that X has pure dimension d if each of its irreducible components has dimension d. If a X and there is a Zariski open neighborhood U of a in X which has pure dimension d, we say that X has pure∈ dimension d near a.

8.1. Dimension of subvarieties. The following is a fundamental property of dimension:

THEOREM III.19. Every proper subvariety of an irreducible variety X has smaller dimension than that of X.

PROOF. Let Y be a proper subvariety of X. Since the ﬁeld of rational functions does not change after restricting to a (nonempty) open subset (exercise III.21), taking an open neighborhood of a point on Y if necessary we may assume without loss of generality that X, Y are afﬁne. Since X, Y are also irreducible, it follows that X = V (p) and Y = V (q) for prime ideals p ( q of k[x1,...,xn], n 1 (exercise III.19). ≥

Let d := dim(X). Without loss of generality we may assume that x1,...,xd are algebraically independent

k k

over k in (X). It sufﬁces to show [why?] that x1,...,xd are not algebraically independent over in

k k k(Y ). Pick a nonzero element q q. Since x ,...,x , q are algebraically dependent over in (X), ∈ 1 d there is an irreducible polynomial F (x ,...,x ,y) k[x ,...,x ,y] such that F (x ,...,x , q)=0 in

1 d ∈ 1 d 1 d k k[X]. Since q =0 in [X], it follows that F is not a multiple of y. Therefore F (x ,...,x , 0) is a nonzero

6 1 d

k polynomial in k[x1,...,xd] which represents an algebraic relation of x1,...,xd in [Y ], as required. We presented theorem III.19 above due to its simple proof (which we found in [Mum95, Proof of Proposition 1.14]). However, if one is willing to use more elaborate machinery, namely Krull’s principal ideal theorem (theorem III.3), then in a straightforward way one can obtain the following more precise version of theorem III.19; we leave its proof as an exercise (exercise III.39).

THEOREM III.20. Let f1,...,fk be regular functions on a variety X of pure dimension n. Show that

(1) V (f1) X is either empty or has pure dimension n 1. (2) The dimension⊆ of every irreducible component of V :=− V (f ,...,f ) X is at least n k. 1 k ⊆ − 26 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

(3) If dim(V ) = n k, then for all I [k], V (fi : i I) has pure dimension n I near every point of V . − ⊆ ∈ − | |

In general the dimension of V (f1,...,fk) may be greater than n k, e.g. V (x(x y),y(x y)) is 2 − − − the diagonal line x = y on k , and therefore has dimension 1 > 2 2. − 8.2. Curve selection lemma. A curve (respectively, surface) is a variety of pure dimension one (respectively, two). The following is our ﬁrst version of the algebraic “curve selection lemma.” A more reﬁned statement is given in proposition III.24. Corollary III.21 (Curve selection lemma I). Let U be a nonempty Zariski open subset of an irreducible variety X and a X. If dim(X) > 0, then there is an irreducible curve C on X containing a such that C U is nonempty.∈ ∩ PROOF. We prove this by induction on dim(X). If dim(X)=1, then it is trivially true, so assume

dim(X) > 1. Without loss of generality we may assume that X is afﬁne. Pick f k[X] such that f(a)=0 and f is not identically zero on X [why does such f exist?]. Let X′ be an irreducible∈ component of V (f) containing a. Theorem III.20 implies that dim(X′) = dim(X) 1, so that we are done by induction. − 8.3. Dimension of the image and ﬁbers of a morphism. Intuitively, the image of a map can not have bigger dimension than the source. The following result makes it precise: Proposition III.22. Let φ : X Y be a dominant morphism of varieties. Then dim(X) dim(Y ). → ≥

∗ k PROOF. This is immediate from the observation that φ : k(Y ) (X) is an injection (exercise III.23). →

2 2 k Now we study the dimension of the fibers of a morphism. Consider e.g. the map φ : k given 2 → by (x, y) (x, xy). The imageof φ is (k V (x)) (0, 0) . For most points on the image, in fact for all 7→ 2 −1 \ ∪{ } (u, v) φ(k ) (0, 0) , the fiber φ (u, v) over (u, v) is zero dimensional, and consists of a single point (u,v/u∈). On the\{ other hand,} φ−1(0, 0) is all of the y-axis, and hence has dimension one. In particular, the dimension of fibers is constant over a nonempty Zariski open subset of the image, and it is possible that some fibers have higher dimension. Theorem III.23 below shows that this is true in general.

THEOREM III.23. Let X be an irreducible variety and φ : X Y be a surjective morphism. Then for every y Y and every irreducible component V of φ−1(y), dim(→V ) dim(X) dim(Y ). Moreover, there is a nonempty∈ Zariski open subset U of Y such that φ−1(y) is of pure≥ dimension− dim(X) dim(Y ) for each y U. − ∈ PROOF [Sha94, Proof of Threorem I.7]. We may assume without loss of generality that Y is afﬁne. Let m := dim(Y ). Take y Y . The following can be proved via an induction on m, we leave its proof as an exercise (exercise III.40).∈

Claim III.23.1. There are regular functions h1,...,hm on Y such that V (h1,...,hm) U = y for some open neighborhood U of y in Y . ∩ { } −1 Pick h1,...,hm,U as in claim III.23.1. Since X is irreducible, it follows that dim(φ (U)) = −1 −1 dim(X) (exercise III.37). Since the fiber of φ (y) is the zero-set of hj φ, j = 1,...,m, on φ (U), the first assertion follows from theorem III.20. For the second assertion we◦ proceed by induction on n := dim(X). Clearly it is true for n =0. Let X′ be an arbitrary nonempty affine open subset of φ−1(Y ). Due to the first assertion it suffices [why?] to prove that there is a nonempty Zariski open subset Y ′ of Y such that (7) dim(φ−1(y) X′) n m for each y Y ′ φ(X′). ∩ ≤ − ∈ ∩ Since X′ is dense in X and φ is continuous (exercise III.5) and surjective, it follows that φ(X′) is dense in

′

k k Y . Therefore φ induces an injection k[Y ] [X ] (proposition III.11); we will consider [Y ] as a subring

′ → ′

k k k of k[X ]. Let g1,...,gM (respectively f1,...,fN ) be -algebra generators of [Y ] (respectively [X ]).

′ ′ k Since dim(X ) = dim(X) = n, the transcendence degree of k(X ) over (Y ) is n m. Therefore we − may assume without loss of generality that f1,...,fn−m are algebraically independent over k(Y ) and fj 8. DIMENSION 27

is algebraically dependent over k(Y )(f ,...,f ) for each j>n m. Fix j, n m +1 j N, and 1 n−m − − ≤ ≤ pick a nonzero polynomial Fj (u1,...,un−m, v) in variables u1,...,un−m, v with coefﬁcients in k[Y ] such that Fj (f1,...,fn−m,fj)=0. Write

dj i Fj (u1,...,un−m, v)= Fj,i(u1,...,un−m)v i=0 X where dj is the degree of Fj in v, and write F (u ,...,u )= g uα1 uαn−m j,dj 1 n−m j,α 1 ··· n−m α=(α1X,...,αn−m) j j j where each g k[Y ]. Fix α = (α ,...,α ) such that g j is a nonzero regular function on Y . Let j,α ∈ 1 n−m j,α Y˜ := Y V ( g j ). Theorem III.19 implies that Y˜ is a nonempty open subset of Y . \ j j,α Claim III.23.2.Q There is a nonempty open subset Y ′ of Y˜ such that for each y Y ′, no irreducible N ∈ component of φ−1(y) is contained in V := V (F ) X′. j=n−m+1 j,dj ∩ PROOF. If φ(V ) is not dense in Y , thenS we can simply take Y ′ to be the complement in Y˜ of the ′ closure of φ(V ). So assume φ(V ) is dense in Y . Then φ V can be extended to a surjective morphism φ : V ′ Y which contains V as a dense subset (exercise III.36| ). Since dim(V ′) = dim(V ) (exercise III.38), theorem→ III.19 implies that dim(V ′)

j j |W − ≤ j 1 n−m k polynomial in k[u1,...,un−m, v] constructed from Fj by evaluating each coefﬁcient from [Y ] at y. Since ¯ ¯ ¯ ¯ ¯ ¯ ¯ Fj (f1,..., fn−m, fj)=0, claim III.23.2 implies that fj is algebraically dependent on k(f1,..., fn−m)

¯ ¯

k k k [why?]. Since k[W ] = [f1,..., fN ], and since (W ) is the quotient ﬁeld of [W ] (exercise III.21), it follows that dim(W ) n m, as required. ≤ − 8.4. Exercises.

EXERCISE III.37. Show that a zero dimensional variety is the union of ﬁnitely many points. If X is an irreducible variety and U is a nonempty open subset of X, show that dim(X) = dim(U). Compute the n n dimension of k and P . EXERCISE III.38. Let X Y be irreducible quasiprojective varieties, and let X′ be the closure of X in X′. Show that dim(X′) = dim(⊆ X).

EXERCISE III.39. Prove theorem III.20 [Hint: use theorem III.3].

n EXERCISE III.40. Let X be an affine variety in k of dimension m 1 and a X. Fix d ≥ ∈ ≥ 1. Show that it is possible to find m polynomials h1,...,hm in (x1,...,xn) of degree d such that a n is an isolated point of V (h1,...,hm) X, i.e. there is a Zariski open subset U of a in k such that U X V (h ,...,h )= a . [Hint:∩ use induction on dimension as in the proof of corollary III.21.] ∩ ∩ 1 m { } EXERCISE III.41. Show that a positive dimensional variety contains infinitely many points.

EXERCISE III.42. Let C Pn be a projective curve, and φ : C P1 be a surjective morphism which ⊂ → is not constant on any of the irreducible components of C. Let T P1 be a finite set. In this exercise you will show that ⊂ (8) there is a finite set S P1 T such that φ−1(P1 S) is an affine curve. ⊂ \ \ (1) Pick a point on P1 T ; denote it by . Show that there is a hypersurface X of Pn containing φ−1( ) such that X\ φ−1(T )= and∞ X C < . [Hint: use exercise III.15.] (2) Show∞ that C′ := C X∩ is a subvariety∅ of|X′∩. [Hint:| ∞ use proposition III.18.] \ 28 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

1 ′ (3) Identifying P with k, show that φ C′ is induced by a regular function f on C . [Hint: use exercise III.5.]\{∞} | ′ ′ ′′ ′ (4) Let T := φ(X C) = φ(C C ) k. Show that C := C V ( t∈T ′ (f t)) is an afﬁne curve. [Hint:∩ use\ {∞}exercise III.11\.] ∩ \ − (5) Conclude that (8) holds with S := T ′ . Q ∪ {∞} 9. Constructible sets You have shown in exercise III.33 that the image of a morphism from the projective space (or more generally, a projective variety) to a variety X is closed in X. This is not true for morphisms from the

2 2 2

k k afﬁne space, e.g. the image of the map from k to given by (x, y) (x, xy) is ( V (x)) (0, 0)

2 7→ 2 \ ∪{ } k which is neither open nor closed in k . However, it is a constructible subset of : a constructible subset of a topological space is a finite union of open subsets of its closed subsets (i.e. constructible subsets of quasiprojective varieties are simply finite unions of quasiprojective varieties). The next result follows in a straightforward way from curve selection lemma I (corollary III.21) and the definition of constructible sets; we leave its proof as an exercise (exercise III.43). Proposition III.24 (Curve selection lemma II). Let W be a constructible subset of a variety X and W¯ be the Zariski closure of W in X. Pick an irreducible component Z of W¯ and a point a Z. If dim(Z) 1, then there is an irreducible curve C on Z containing a such that C W is nonempty and∈ (Zariski) open≥ in C. ∩ The main result of this section is the following result of C. Chevalley:

THEOREM III.25 (Chevalley’s theorem). The image φ(X) of a morphism φ : X Y of varieties is a constructible subset of Y . → We give a proof of theorem III.25 following [Mum95, Section 2C]. We leave it as an exercise (exercise III.47) to shows that it is equivalent to the following property of varieties (note that this property is precisely what you get from substituting “closed subsets” by “constructible subsets” in the deﬁnition of complete varieties).

THEOREM III.26. Let X, Y be varieties. Then the projection map X Y Y maps constructible sets to constructible sets. × →

n m k PROOF. It sufﬁces [why?] to consider the case that X = k and Y = . Then taking compositions

m m

k k we can further reduce it to the case of the projection π : k . It is straightforward to check, and we leave it as an exercise (exercise III.48) to show, that it sufﬁces× → to prove the following statement:

m m+1

k k If V is an irreducible subvariety of k = and U is a nonempty open subset (9) × ∼ m of V , then π(U) contains a nonempty open subset of the closure of π(V ) in k .

k k [ We now prove (9). Let W be the closure of π(V ) in k . Then π induces an injective map [W ] ֒ [V

→

k k

(proposition III.11), so that k[V ] ∼= [W ][x1]/p for some ideal p of the polynomial ring [W ][x1] in one k variable over k[W ]. At ﬁrst consider the case that p = 0. Then V = W . Let (a1,...,am+1) be any ′ × ′ point of U. Then W := a1 W is a subvariety of V and therefore U W is a nonempty open subset of W ′ whose projection is open{ }× in W . Now consider the remaining case that∩ p =0. Then theorem III.19 and ¯ 6 1 m ′ ¯ proposition III.22 imply that dim(V ) = dim(W ). Let V be the closure of V in P k and V := V U. 1 ¯ ′ × m \ ¯ Since P is complete (exercise III.35), it follows that π(V ) and π(V ) are closed in k . Since π(V ) is closed, it follows that π(V¯ ) W . On the other hand dim(V ′) < dim(V¯ ) = dim(V ) = dim(W ) (theorem III.19 and exercise III.38⊃ ), and therefore π(V ′) cannot contain W (proposition III.22). Since π(U) π(V¯ ) π(V ′) W π(V ′), the claim follows. ⊃ \ ⊃ \ Corollary III.27. Let φ : X Y be a morphism of varieties and k be a nonnegative integer. Then Y := y Y : dim(φ−1(y)) <→ k is a constructible subset of Y . k { ∈ } PROOF. We proceed by double induction on mφ := dim(φ(X)) and k. Due to theorems III.23 and III.25 the corollary is true whenever mφ = 0 or k dim(X) mφ. Now assume it is true for ≤′ − ′ all φ such that mφ < m. Pick φ with mφ = m and k such that the corollary holds for φ and k . ′ We will show that it holds for φ and k +1. By the inductive hypothesis Y Y ′ is constructible, and \ k 10.ZARISKICLOSUREOFTHEIMAGEOFAMORPHISM 29

0 ′0 therefore is a union of quasiprojective varieties. Let Yk′ be an irreducible component of Y Yk′ , and Xk′ −1 0 0 ′0 \ be an irreducible component of φ (Yk′ ). Note that both Yk′ and Xk′ are quasiprojective varieties, so ′0 ′0 that Yk′ := φ(Xk′ ) is constructible due to Chevalley’s theorem (theorem III.25). It sufﬁces to show that ′0 ′0 ′0 ′0 ′0 ′0 Yk′ Yk′+1 is constructible. Let φk′ be the restriction of φ to Xk′ . Then φk′ : Xk′ Yk′ is surjective ∩ ′0 −1 ′ ′0 ′ ′0 → ′0 and by construction dim((φk′ ) (y)) k for each y Yk′ . If k < dim(Xk′ ) dim(Yk′ ), then ′0≥ ∈ − theorem III.23 implies that Yk′+1 Yk′ = , which is trivially constructible. Otherwise theorem III.23 ′ ′0 ∩ ′0 ∅ ′0 ′0 implies that k = dim(Xk′ ) dim(Yk′ ) and there is a nonempty Zariski open subset Uk′ of Yk′ such ′0 −1 ′ − ′0 ′′0 ′0 ′0 ′′0 that dim((φk′ ) (y)) = k for each y Uk′ . Let Yk′ := Yk′ Uk′ and φk′ be the restriction of φ to −1 ′′0 ′′0 ∈ \ ′′0 φ (Yk′ ). Since dim(Yk′ ) < dim(Y ) = m, the corollary is true for φk′ (and all values of k), so that ′′0 ′0 ′0 ′′0 Y ′ Y ′ is constructible. Therefore Y ′ Y ′ = U ′ (Y ′ Y ′ ) is constructible as well. k +1 ∩ k k +1 ∩ k k ∪ k +1 ∩ k 9.1. Exercises.

EXERCISE III.43. Prove proposition III.24.

EXERCISE III.44. Let φ : X Y be a morphism of varieties and U be a constructible subset of Y . Show that the following are equivalent:→ (1) U contains a nonempty Zariski open subset of Y . (2) φ−1(U) contains a nonempty Zariski open subset of X.

EXERCISE III.45. Let φ : X Y be a morphism of varieties and U be a constructible subset of X. Show that for each constructible subset→ V of Y , φ−1(V ) U is a constructible subset of X. ∩ EXERCISE III.46. The dimension of a constructible subset U of a variety X is simply the dimension of the Zariski closure of U in X. Let φ : X Y be a dominant morphism of varieties. If X is irreducible, then show that the following are equivalent:→ (1) U contains a nonempty Zariski open subset of X. (2) there is a nonempty Zariski open subset Y ′ of Y such that for each y Y ′, dim(φ−1(y) U)= dim(X) dim(Y ). ∈ ∩ − EXERCISE III.47. Show that theorem III.25 is equivalent to theorem III.26.

m m

k k EXERCISE III.48. Show that it sufﬁces to prove(9) in order to provethat the projection k maps constructible sets to constructible sets. [Hint: use induction on dimension.] × →

10. Zariski closure of the image of a morphism In this section we describe the (Zariski) closures of images of morphisms to afﬁne and projective spaces.

Proposition III.28. Let f1,...,fN be regular functions on a (possibly reducible) afﬁne variety X. Let

N N k φ : X k be the morphism which maps x (f (x),...,f (x)), and Z be the closure of φ(X) in .

→ 7→ 1 N k Let R be the k-subalgebra of [X] generated by f1,...,fN .

(1) k[Z] = R.

∼ N M k (2) Let π : k be the natural projection onto the ﬁrst M coordinates, where M N, and → N ≤ let H be the coordinate subspace of k spanned by the ﬁrst M coordinates. Then Z H is contained in the closure of π φ(X). ∩

∗ N ∗◦

k k (3) If φ(X) (k ) (where := 0 ), then the coordinate ring of the closure of φ(X) in

∗ N ⊆ \{ } −1 −1 k (k ) is isomorphic to Rf1···fN := [f1,f1 ,...,fN ,fN ]. ∗ PROOF. Let φ : k[y ,...,y ] R be the ring homomorphism which maps each y f . Let I(Z) 1 N → i 7→ i be the ideal of k[y1,...,yN ] consisting of polynomials that vanish on Z. Then g I(Z) if and only if g ∈ ∗ vanishes on φ(X) if and only if g(f1,...,fN )(x) is zero for all x X if and only if φ (g)=0 in k[X].

∗ ∈ k Therefore I(Z) = ker φ , and R ∼= k[y1,...,yN ]/I(Z) ∼= [Z] (see exercise III.8 for the last isomorphism). The second assertion follows from a general property of retractions - see exercise III.49. Finally,

∗ N n ∗ N

k k since (k ) = V (y y ), the closure of φ(X) in ( ) is Z V (y y ) (exercise III.50), so \ 1 ··· N \ 1 ··· N that the coordinate ring of Z V (y1 yN ) is k[Z]y1···yN (exercise III.11), which by the ﬁrst assertion is \ ··· isomorphic to Rf1···fN , as required for the third assertion. 30 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

identify two ends

blow up map

FIGURE 2. Blowupofapointonadiskin R2

Corollary III.29. Let f0,...,fN be regular functions on an irreducible afﬁne variety X, and Z be the N N closure in P of the image of the map φ : X V (f0,...,fN ) P deﬁned by x [f0(x): : fN (x)]. \ N → N 7→ ··· Let [z0 : : zN ] be homogeneous coordinates on P and Ui := P V (zi), 0 i N. Then

··· \ ≤ ≤ k (1) k[Z U ] = [f /f ,...,f /f ],

∩ i ∼ 0 i N i k (2) k[Z Ui Uj ] ∼= [f0/fi,...,fN /fi,fi/fj]. ∩ ∩ n b bi (3) Assume there exists b Z>0 and b1,...,bN Z≥0 such that b = bi and f = f . ∈ ∈ i=1 0 i≥1 i Then Z U (Z U ). ∩ 0 ⊆ i≥1 ∩ i P Q (4) Let H = V (z ,...,z ), H′ := V (z ,...,z ) for some M N, and π : N H′ H M+1S N 0 M P be the natural projection onto the ﬁrst M coordinates. Then Z ≤H is contained in\ the closure→ of π φ(X V (f ,...,f )). ∩ ◦ \ 0 M N PROOF. Since each Ui ∼= k , assertion (1) follows from proposition III.28. Write Zi := Z Ui and Z := Z U U . Since Z is an afﬁne variety, f /f is a regularfunction on Z , and Z = Z V∩(f /f ),

ij ∩ i ∩ j i j i i ij i \ j i

k k it follows that k[Zij ]= [Zi]fj /fi = [f0/fi,...,fN /fi] (see exercise III.11 for the ﬁrst equality), which proves assertion (2). For assertion (3), note that zb zbi is identically zero on Z. It follows that 0 − i≥1 i Z U = Z V (zb)= Z V ( zbi ) (Z U ), as required. For the last assertion it sufﬁces to ∩ 0 \ 0 \ i≥1 i ⊆ i≥1 ∩ Qi show that Z H Ui is contained in the closure of π φ(X V (fi)) for each i =0,...,M. This follows from assertion∩ (2)∩ of propositionQIII.28. S ◦ \

10.1. Blow up. Let X be an irreducible afﬁne variety and q be an ideal of k[X]. Given generators N−1 f1,...,fN of q, the blow up Blq(X) of X at q is the closure in X P of the graph of the map N−1 × X V (q) P given by x [f1(x): : fN (x)]. You will prove in exercise III.53 below that \ → 7→ ··· Blq(X) does not depend on the choice of the fj. In exercise III.51 below you will show that the restriction N−1 to Blq(X) of the natural projection from X P to X induces a surjective map σ : Blq(X) X which is an isomorphism away from V (q) ×X; often this map is what one means by the “blow up→ of X at q.” If V is an irreducible subvariety of X⊂ not contained in V (q) X, then the strict transform of V −1 ⊂ on Blq(X) is the closure in Blq(X) of σ (V V (q)). In the case that q is the ideal of k[X] consisting \ of regular functions vanishing on a subvariety Y of X, we also write BlY (X) for Blq(X), and call it the 2 “blow up of Y in X.” The name “blow up” is probably motivated by the case that Y is a point in X = k , in which case the point is “blown up” to a line, see ﬁg. 2 and exercise III.52. 10.2. Exercises.

EXERCISE III.49. Let Y be a topological space, and r : Y H be a retract onto a subset H Y → ⊂ (which means r restricts to the identity map on H). Given W Y , show that W H r(W ). ⊂ ∩ ⊂ EXERCISE III.50. Let U be an open subset of a topological space Y and W U. Show that the closure of W in U is the intersection of U and the closure of W in Y . ⊆ EXERCISE III.51. Let q be an ideal of the coordinate ring of an irreducible afﬁne variety X and σ : Blq(X) X be th blow up of X at q. Show that → 11. TANGENT SPACE, SINGULARITIES, LOCAL RING AT A POINT 31

(1) σ is surjective, and −1 (2) σ induces an isomorphism Blq(X) σ (V (q)) ∼= X V (q). In particular, X and Blq(X) are birational. \ \ n EXERCISE III.52. Let X = k and Y be a point in X. Show that in BlY (X) the strict transforms of distinct lines through Y in X do not intersect, and the pre-image of Y is isomorphic to Pn−1. In the case 2 that k = R and X is a disk in R , show that BlY (X) is homeomorphic to a Mobius¨ strip (see ﬁg. 2).

EXERCISE III.53. Let X be an irreducible afﬁne variety and q be an ideal of k[X]. Let f1,...,fN and g1,...,gM be two sets of generators of q in k[X]. Write Z (respectively, Y ) for the closure in N−1 M−1 X P (respectively, X P ) of the graph of the map x [f1(x): : fN (x)] (respectively, × × 7→ ··· x [g (x): : g (x)]). In this exercise you will show that Y = Z. 7→ 1 ··· M ∼ (1) Show that it sufﬁces to consider the case that M = N +1 and gj = fj, j =1,...,N. (2) Consider the case of the preceding assertion. Assume gN+1 = f1h1 + + fN hN , where N−1 N ··· h1,...,hN k[X]. Let ψ : X P X P be the map that sends ∈ × → × (x, [z : : z ]) (x, [z : : z : h (x)z + + h (x)z ]) 1 ··· N 7→ 1 ··· N 1 1 ··· N N Show that ψ induces an isomorphism Z ∼= Y . 11. Tangent space, singularities, local ring at a point

11.1. The case of afﬁne varieties. Consider a straight line L = a + tv : t k through a point a =

N N { ∈ } k (a ,...,a ) k , where v = (v ,...,v ) determines the “direction” of L. Assume f(a)=0, 1 N ∈ 1 N ∈ where f is a polynomial in (x1,...,xN ). We say that L is tangent to V (f) at a if ordt(f(a + tv)) > 1, or equivalently, if N ∂f (10) v (a)=0 i ∂x i=1 i X More generally, L is tangent at a to an afﬁne variety X containing a if (10) holds for all f vanishing on X. The tangent space Ta(X) to X at a is the union of all tangent lines to X at a. It is straightforward to check (exercise III.54) that N ∂f (11) T (X)= V (x a ) (a): f I(X) a i − i ∂x ∈ i=1 i ! X where I(X) is the ideal in k[x1,...,xN ] of polynomials vanishing on X. It is clear from (11) that Ta(X) N is of the form V + a where V is a linear subspace (through the origin) of k ; the dimension of Ta(X) is simply the dimension of V (as a vector space over k). Let f1,...,fs be a set of generators of I(X), then exercise III.54 implies that

∂fi (12) dim Ta(X)= N Rank (a) − ∂xj 1≤i≤s 1≤j≤N

where Rank( ) is the rank over k of the corresponding matrix. · Proposition III.30. For each integer k, the set X := a X : dim T (X) k is Zariski closed in ≥k { ∈ a ≥ } X; in other words, the map X Z given by a dim Ta(X) is upper semicontinuous. 7→ 7→

PROOF [Mum95, Section 1A]. Let q be the ideal of k[x ,...,x ] generated by (N k +1) (N k 1 N − × − k + 1)-minors of the matrix ( ∂fi (a)). Identity (12) implies that X = V (I(X)+ q ). ∂xj ≥k k Let d := min dim T (X): a X . Assume X is irreducible. Then we say that a X is a singular { a ∈ } ∈ (respectively nonsingular) point if dim Ta(X) > d (respectively dim Ta(X) = d). Proposition III.30 implies that the set of nonsingular points of X is a nonempty Zariski open subset of X. 2 Example III.31. Every line or a quadric curve on k is everywhere nonsingular (exercise III.56). Both the curves C = x2 = y3 and C = x2 = y2 y3 are singular at the origin (exercise III.57); see ﬁg. 3. 1 { } 2 { − } The singularity of C1 is called a cusp, and that of C2 is called a node. 32 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

O 2 3 2 2 3 (A) x = y (B) x = y − y

FIGURE 3. Some curve singularities

11.2. Intrinsicness of the tangent space; definition of tangent space and singularities for arbitrary varieties. The definitions of tangent spaces and (non-)singular points given above applies only to affine varieties; moreover, they depend on the defining equations of the affine variety, and a priori it is not clear if they are preserved by isomorphisms. In this section we show that these notions depend only on the ring of regular functions on an (arbitrarily small) open neighborhood of the given point (in the affine variety in question), and extend these notions to arbitrary varieties. Given a point a of a subvariety X of

k k k , amap D : [X] is called a derivation centered at a if D satisﬁed the following properties: →

(a) D is k-linear,

(b) D(fg)= f(a)D(g)+ g(a)D(f) for all f,g k[X], ∈

k k Proposition III.32. Given a = (a1,...,aN ) k , the derivations D : [x1,...,xN ] centered at a N ∈ → are in one-to-one correspondence with k given by:

N ∂f

(13) λ = (λ ,...,λ ) D , where D (f)= λ (a), for all f k[x ,...,x ] 1 N 7→ λ λ i ∂x ∈ 1 N i=1 i X

The local ring of X at a is the localization of k[X] with respect to the multiplicative set of OX,a regular functions not vanishing at a; equivalently X,a is the ring of rational functions f/g : f,g O {′ ∈ k[x1,...,xN ], g(a) =0 modulo the ideal generated by polynomials vanishing on X. If X is an arbitrary 6 } ′ (i.e. not necessarily affine) variety and a X , then X′,a is by definition X,a for any open affine ′ ∈ O O neighborhood X of a in X ; exercise III.59 below shows that X′,a is well-defined, and exercise III.60 below shows that it is in fact a local ring. O

THEOREM III.33. The following spaces are isomorphic as vector spaces over k: (1) T (X) with the vector space structure on T (X) induced from that of T (X) a (in particular, a a a −

a is the origin of Ta(X)), k (2) the space of derivations k[X] centered at a, 2 ∗ → 2 (3) the space (ma/ma) of linear functions from ma/ma to k, where ma is the maximal ideal of generated by functions vanishing at a.

OX,a k PROOF [Mum95, Section 1A]. A derivation k[X] centered at a is the same as a derivation

→

k k k k[x1,...,xN ] centered at a which vanishes on I(X). Derivations [x1,...,xN ] centered at a → n → are of the form D from the map deﬁned in (13). Given λ k , identity (11) implies that D vanishes λ ∈ λ on all f I(X) if and only if λ + a Ta(X), which proves the isomorphisms between (1) and (2).

∈ ∈ 2 ∗ k Exercise III.61 below shows that every derivation k[X] centered at a deﬁnes an element of (m /m ) . → a a

Conversely, given a linear map m k such that 2 , let φ : a φ ma 0 λφ := (φ((x1 a1) X ),...,φ((xN N → | ≡ − | − aN ) X )) k , and deﬁne Dφ := Dλφ as in (13). Let f I(X). The Taylor series expansion of f | ∈ n ∂f ′ ∈′ shows that f = f(a)+ (xi ai) (a)+ f , where f is of order two or higher in the (xi ai). i=1 − ∂xi − ′ N ∂f Since f(a)=0 and φ(f X )=0= φ(f X ) [why?], it follows that φ((xi ai) X ) (a)=0, i.e.

|P | i=1 − | ∂xi k D (f)=0. It follows that D is a derivation k[X] . It is straightforward to check that the maps we φ φ P deﬁned between (2)and (3) are inverse to each other,→ and induce an isomorphism of vector spaces. 11. TANGENT SPACE, SINGULARITIES, LOCAL RING AT A POINT 33

′ ′ The tangent space to a variety X at a point a X is by deﬁnition Ta(X) for an afﬁne neighborhood ′ ∈ ′ 2 ∗ X of a in X [why does such X exist?]; equivalently, Ta(X ) = (ma/ma) , where ma is the maximal ideal of ′ generated by functions vanishing at a. OX ,a Proposition III.34. Assume X is an irreducible variety. Then min dim T (X): a X = dim(X). { a ∈ } PROOF. Let d := min dim Ta(X): a X and U := x X : dim Ta(X) = d . Proposi- tion III.30 implies that U is{ Zariski open in X∈. Since} X is birational{ ∈ to a hypersurface (exercise} III.25), and since dimension is invariant under birational maps (exercise III.24), without loss of generality we N may assume X = V (f) k for some polynomial f in (x1,...,xN ). If f = 0, then an easy com- ⊂ N putation shows that U = k (exercise III.55) and d = N, as required. Otherwise f is a nonzero ir-

reducible polynomial, and since k is algebraically closed, not all the partial derivatives ∂f/∂xj vanish identically on X (exercise III.62). It then follows the definition of the tangent space (or identity (12)) that d = N 1 = dim(X). − Now we can extend the notion of (non-)singular points to arbitrary (possibly reducible) varieties. Let X be a variety and a X. Define dima(X) to be the maximum of the dimensions of the irreducible components of X containing∈ a. We say that a is a singular (respectively nonsingular) point of X if dim Ta(X) > dima(X) (respectively dim Ta(X) = dima(X)). Propositions III.30 and III.34 imply that the subset of X consisting of nonsingular points is Zariski open and has a nonempty intersection with every irreducible component of X; in particular, a curve has at most finitely many singular points. A variety is called singular if it has a singular point; otherwise it is called nonsingular. The codimension of a subvariety Y of a variety X is the integer dim(X) dim(Y ). − 11.3. Equations near a nonsingular point. In this section we show that every nonsingular point on a variety of dimension n has an affine neighborhood which can be described by N n equations in N − k N (for some N n); we follow the approach of [Mum95, Proof of Theorem 1.16]. Let R := k ,0

ˆ ≥ O k and R := k[[x1,...,xN ]] be the ring of formal power series in (x1,...,xN ) over . We consider R as a ˆ N −1 j subring of R via the expansion (1 x g ) = 1+ (x g ) for any g ,...,g k[x ,...,x ]. − i=1 i i j≥1 i i 1 N ∈ 1 N Lemma III.35. If q is an ideal of RP, then qRˆ R = q.P ∩ PROOF. Given f qRˆ R, it sufﬁces to show that f q. Indeed, write f = φ f , where f are ∈ ∩ ∈ j j j j polynomials which generate q and φ are power series in (x ,...,x ). For each k > deg(f), if φ are j 1 n P j,k the polynomials consisting of all monomial terms of φ of order at most k, then g := f φ q, so j k j j j,k ∈ that f = g + f (φ φ ) q + mk, where m is the (unique) maximal ideal of Rˆ. It follows that k j j − j,k ∈ P f (q + mk)= q (the last equality follows from theorem III.4). ∈ k≥0 P

THEOREM III.36 ([Mum95, Theorem 1.16]). Let f1,...,fr be polynomials in (x1,...,xN ) with no constant term and linearly independent (over k) linear terms. Then the ideal p generated by f1,...,fr in N

k is prime. O ,0 PROOF. It follows from basic properties of formal power series that f1,...,fr generate a prime ideal in Rˆ (exercise III.63). The theorem then follows from lemma III.35.

Corollary III.37. Every nonsingular point on a variety of dimension n has an open neighborhood which N is irreducible and isomorphic to a subvariety of k (for some N n) whose ideal is generated by N n polynomials (in N-variables). ≥ −

PROOF. Let a be a nonsingular point of a variety X, with dim(X) = n. We may assume without

m m k loss of generality that X is a subvariety of k for some m n, and a is the origin in . Identity (12) implies that there are f ,...,f I(X) with no constant≥ term and linearly independent linear 1 m−n ∈ terms. Let q be the ideal of k[x1,...,xm] generated by f1,...,fm−n. Theorem III.36 implies that the ′

m m m k

k k ideal q k ,a generated by q in ,a is prime, so that q := (q ,a) [x1,...,xm] is also prime. Let ′ O ′ m O′ O ∩ X := V (q ) k . Note that X is an irreducible variety containing a. ⊆ Claim III.37.1. dim(X′)= n. There is a polynomial g such that g(a) =0 and X V (g)= X′ V (g). 6 \ \ 34 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

′ m ′ PROOF. Since X is deﬁned by m n equations in k , theorem III.20 implies that dim X n. − ′ ≥ On the other hand the assumptions of linear parts of fj and proposition III.34 implies that dim X n. ′ ′ ′ ′ ′ ≤ It follows that dim X = n. Let h1,...,hs be a set of generators of q . Then each fj can be expressed as h /g for some polynomials g ,h such that h q and g (a) = 0 [why?]. Let g := g . It is j j j j j ∈ j 6 j j straightforward to check that X V (g)= X′ V (g), as required. \ \ Q

′

k k Since k[X V (g)] ∼= [X ]g ∼= [x1,...,xm,y]/ f1,...,fm−n,yg 1 (see e.g. exercise III.11), the corollary holds\ with N := m +1. h − i 11.4. Exercises. N EXERCISE III.54. Prove identity (11). If f1,...,fs generate I(X), then show that i=1(xi ∂fj − ai) (a), j =1,...,s, generate the ideal of polynomials vanishing on Ta(X). ∂xi P

N N N

k k EXERCISE III.55. Show that k is nonsingular everywhere. Given a , compute T ( ). ∈ a 2 EXERCISE III.56. Let C = V (f) k , where f is an irreducible polynomial of degree 1 or 2. Show ⊂ that C is everywhere nonsingular. [Hint: recall that over an algebraically closed ﬁeld k every homogeneous 2 2 polynomial of degree 2 can be written as ax +by , a,b k, after an appropriate change of coordinates on 2 ∈ 2 2 k . If deg(f)=2, then use this fact to reduce to the following cases: (1) f = ax + by + c, a,b,c =0, and (2) f = ax2 + by, a,b =0.] 6 6 2 2 3 2 2 3 EXERCISE III.57. Let C = V (f) k . If either f = x y or f = x y + y , show that the origin is the only singular point of C. ⊂ − − EXERCISE III.58. Prove proposition III.32. EXERCISE III.59. Let X be an afﬁne variety and U be an open neighborhood of a X. ∈ (1) Show that = . OX,a ∼ OU,a (2) Let φ : X Y be an isomorphism. Show that = . → OX,a ∼ OY,φ(a)

(3) Assume a Z X, where Z is a subvariety of X. Let q be the ideal of k[X] consisting of ∈ ⊆ regular functions vanishing on Z. Then show that = /(q ). OZ,a ∼ OX,a OX,a EXERCISE III.60. Let a be a point on a variety X, and let m be the ideal of generated by the a OX,a regular functions which are deﬁned on a neighborhood of a on X and vanish at a. Show that ma is the unique maximal ideal of ; in other words, is a local ring.

OX,a OX,a k EXERCISE III.61. Let X be an afﬁne variety, a X, and D : k[X] be a derivation centered at a. If m is the maximal ideal of generated by polynomials∈ vanishing→ at a, show that a OX,a

(1) D extends to a linear map m k given by a →

D(f/g) := D(f)/g(a) for all f,g k[X] ∈ (2) D(h)=0 for all h m2. ∈ a EXERCISE III.62. Let f be an irreducible (nonzero) polynomial in (x1,...,xN ). Show that there is i

such that ∂f/∂xi does not identically vanish on V (f). [Hint: you have to use that f is irreducible. In the k case that char(k) is positive, you have to use that is algebraically closed.]

EXERCISE III.63. Let f1,...,fr be elements of k[[x1,...,xn]] with no constant term and linearly k independent (over k) linear terms. Then the ideal p generated by f1,...,fr in [[x1,...,xn]] is prime. [Hint: use theorem III.5.] EXERCISE III.64. Let a be a nonsingular point on a curve X. In this exercise you will show that X,a is a discrete valuation ring (see appendix B.5). Let ma be the (unique) maximal ideal of X,a. Pick O 2 2 O f m m , so that (the image of) f generates m /m over k (see assertion (3) of theorem III.33). ∈ a \ a a a (1) Show that ma is the principal ideal generated by f. [Hint: apply Nakayama’slemma (lemma B.5).] (2) Show that the radical of the ideal generated by a nonzero element in is either itself OX,a OX,a

or m . [Hint: Let C be an afﬁne neighborhood of a in X and R = k[C]. Every g 0 a ∈ OX,a \{ } can be expressed as g = g1/g2, where g1,g2 R, g2(a) =0. If g1(a) =0, then g is invertible in . Otherwise theorem III.20 and exercise∈ III.37 imply6 that V (g ) consists6 of ﬁnitely many OX,a 1 points including a. The Nullstellensatz then implies that m = g .] a h i p 12. COMPLETION OF THE LOCAL RING AT A POINT 35

k (3) For each g X,a 0 , show that there is k 0 such that g/f is a unit in X,a [Hint: let k := inf m∈ O0 : f m\{ }g . Show that k is a well≥ deﬁned integer.] O { ≥ ∈ h i} m (4) Let ν : X,a Z be the mapgivenby g inf m 0 : f g . Show that ν extends O → ∪{∞} 7→ { ≥ ∈ h i} to a discrete valuation on the ﬁeld of fractions of X,a [why is X,a an integral domain?], its valuation ring is , and f is a parameter of O. O OX,a OX,a EXERCISE III.65. Let a be a nonsingular point on a curve X and ν be the (unique)5 discrete valuation associated to X,a. If f is any parameter of X,a, show that each g X,a 0 can be expressed as m ′ O O ′ ′ ∈ O \{ } cf + g where c k 0 , m = ν(g), and g X,a with ν(g ) > m. [Hint: take c := u(a), where u := g/f m.] ∈ \{ } ∈ O

EXERCISE III.66. Let C be a curve and f : C PN be a morphism. If C¯ is a compactiﬁcation of C → such that every point of C¯ C is nonsingular, then show that f extends to a morphism C¯ PN . [Hint: \ → ﬁx a C¯ C and a parameter t of ¯ . Use exercise III.64 to show that there is an open neighborhood ∈ \ OC,a U of a in C¯ such that f can be expressed as [u tk0 : : u tkN ], where the u are invertible regular |U\a 0 ··· N j functions at a and kj Z. Now “clear the denominators.”] ∈ EXERCISE III.67. This exercise shows that the conclusion of exercise III.66 might fail if C¯ C has

2 \ k singular points with “more than one branch.” Assume char(k) = 2. Let C be the curve on defined 3 2 6 2 by the equation y = x(y 1). Let [x0 : x1 : x2] be homogeneous coordinates on P . The map −2 2 2 ¯ (x, y) [1 : x : y] identifies k with the basic open subset U0 := P V (x0) of P . Let C be the closure 7→ \ of C in P2. ¯ 3 2 2 2 ¯ (1) Show that C = V (x2 x1(x2 x0)) P [Hint: use exercise III.12] and C C = [0 : 1 : 1], [0 : 1 : 0] . − − ⊂ \ { (2) Show that the} projection from C to y-axis is one-to-one, and the inverse of this map extends to a morphism φ : P1 C¯ which is generically one-to-one, and φ−1([0 : 1 : 0]) consists of two points. → (3) Conclude that the conclusion of exercise III.66 fails with C, C¯ and f := y. (4) Show that O := [0 : 1 : 0] has an affine neighborhood in C¯ isomorphic to the plane curve v3 = (v w)(v + w). A drawing of this curve makes apparent the two “branches” (see section VI.8.1) at O− with “tangents” v w =0 and v + w =0 (see fig. 3b). −

12. Completion of the local ring at a point If I is an ideal of a ring R, the I-adic topology on R is the (unique) topology in which a subset S of R is open if and only if S f + Im for some f R and m 0. A Cauchy sequence in R is a sequence of elements (f ) of R such⊇ that for any open∈ neighborhood≥ U of 0, there is m 0 with the property j j≥0 ≥ that fi fj U for all i, j m. Two Cauchy sequences (fi)i and (gj )j are equivalent if the sequence − ∈ 6 ≥ (fi gi)i converges to 0 in R. The I-adic completion of R is the set of equivalence classes of Cauchy sequences.−

Example III.38. Let R := k[x1,...,xn] and I be the ideal of R generated by all polynomials vanishing n at some point a = (a ,...,a ) k . Let (f ) in R be a Cauchy sequence with respect to the I-adic 1 n ∈ j j topology. We treat each fj as a polynomialin yi := xi ai, i =1,...,n. For each k 0, there is Mk such that the degree (with respect to (y ,...,y )-coordinates)− of f f is greater than k≥for each i, j M . 1 n i − j ≥ k In particular, with respect to (y1,...,yn)-coordinates the homogeneous components fj,k of degree k of all f agree with each other whenever j M ; write F := f . It is straightforward to check that the map j ≥ k k Mk,k (f ) F j j 7→ k Xk

ˆ k induces a k-algebra isomorphism between R and [[x a ,...,x a ]]. 1 − 1 n − n

5A discrete valuation is necessarily unique - see proposition B.8. 6 Recall that on a topological space X, a sequence (xi)i converges to a point x if for every open neighborhood U of x in X there is an integer N such that xi ∈ U for all i ≥ N. 36 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

n k

Example III.39. We have seen (in the discussion preceding lemma III.35) that the local ring k of O ,a at a point a = (a1,...,an) can be treated as a subring of k[[x1 a1,...,xn an]]. We leave it as an ˆ − − n k exercise (exercise III.69) to check that k = [[x a ,...,x a ]]. O ,a ∼ 1 − 1 n − n Proposition III.40. Let Rˆ be the completion of a ring R with respect to an ideal I. Given f R, write ∈ fˆ for the equivalence class of the Cauchy sequence (f, f, . . .). Let φ : R Rˆ be the map which sends → f fˆ. 7→ m (1) ker φ = m≥1 I . (2) If R is a Noetherian local ring, and I is a proper ideal of R, then φ is injective. T PROOF. Since the zero element of Rˆ is the equivalence class of (0, 0,...), the ﬁrst assertion is immediate from the deﬁnition of completion. The second assertion then follows from theorem III.4.

If X is a variety and a X, then has a unique maximal ideal m , namely the ideal generated by ∈ OX,a a polynomials vanishing at a (exercise III.60). We write ˆ for the m -adic completion of . The ring OX,a a OX,a ˆ captures “very local” information about X at a. Example III.39 shows that if X is the afﬁne space, OX,a then ˆX,a is the ring of formal power series expansions centered at a; in general it is a quotient of a power seriesO ring:

N ˆ k Proposition III.41. Let a = (a ,...,a ) be a point of a subvariety X of k , and let R := [[x 1 N 1 − a ,...,x a ]]. Then ˆ = R/ˆ (I(X)Rˆ). 1 N − N OX,a ∼

ROOF N N k P . Since X,a ∼= k ,a/(I(X) ,a) (exercise III.59), the result follows from example III.39 and the exactness ofO completionO (theorem III.6O ).

THEOREM III.42. Let a be a nonsingular point of an irreducible variety X of dimension n. Then ˆ (1) = k[[x ,...,x ]]. OX,a ∼ 1 n

N k Assume X is a subvariety of k with coordinates (x1,...,xN ). Pick g1,...,gn [X] such that gi(a)= ∂gi ∂gi ∈ 0 for each i, and ( (a),..., (a)), i = 1,...,n, generate Ta(X) as a vector space over k (see ∂x1 ∂xN assertion (1) of theorem III.33). Then (2) it is possible to choose the isomorphism from assertion (1) such that gˆ x , i = 1,...,n i → i (where gˆi are deﬁned as in proposition III.40).

PROOF. Taking an appropriate open neighborhood of a we may assume that X is an irreducible N subvariety of k , and I(X) is generated by f1,...,fN−n (the second property can be ensured due to N corollary III.37). By a change of coordinates if necessary, we may assume a is the origin in k . Then ˆ ˆ ˆ ˆ = R/(I(X)R), where R := k[[x ,...,x ]] (theorem III.6 and exercise III.59). The nonsingular- OX,a ∼ 1 N ity of X at a implies that the linear parts of the fj are linearly independent over k, and we may choose ′ ′ g1,...,gn k[X] which satisfy the hypothesis of assertion (2) (namely, take polynomials g1,...,gn ∈ ′ ′ which vanish at a and are such that the linear parts of f1,...,fN−n,g1,...,gn are linearly indepen- ′ ˆ ˆ dent over k; then set g := g , i = 1,...,n). It then follows from theorem III.5 that R/(I(X)R) i i|X is isomorphic to the ring of power series in n-variables over k via an isomorphism that maps gˆi xi, i =1,...,n. →

Corollary III.43. Let a be a nonsingular point of an irreducible curve C, and x be a parameter of C,a ˆ O (see exercise III.64). Then = k[[x]]. OC,a ∼ PROOF. Pick g as in theorem III.42. By deﬁnition of a parameter, g = uf k for some k 0 and a 1 1 ≥ unit u of C,a. The propertiesof g1 implies that k =1. Let c := u(a) =0. Then g1 = uf = cf +(u c)f. Since (uO c)f m2 (where m is the maximal ideal of ), it follows6 that ∂((u c)f)/∂x =− 0 and − ∈ a a OC,a − i ∂g1/∂xi = c∂f/∂xi for each i =1,...,n, i.e. the linear parts of g1 and f are proportional. The corollary then follows from the arguments of the proof of theorem III.42. 13. DEGREE OF A DOMINANT MORPHISM 37

12.1. Exercises.

EXERCISE III.68. Show that open subsets in I-adic topology satisfy the axioms of open subsets of a topological space. n

k n

EXERCISE III.69. Let a = (a ,...,a ) , and m be the unique maximal ideal of k gener- 1 n ∈ a O ,a

ˆ n n k ated by polynomials vanishing at a. Show that the m -adic completion k of is isomorphic as a

a O ,a O ,a k k-algebra to [[x a ,...,x a ]]. 1 − 1 n − n EXERCISE III.70. Show that the R-adic completion of R is the zero ring; in particular, assertion (2) of proposition III.40 does not hold if I = R.

EXERCISE III.71. Let a X and q be an ideal of X,a such that X,a/q X,a is a ﬁnite dimen- ∈ Oˆ ˆ O O sional vector space over k. Show that X,a/q X,a ∼= X,a/q X,a. [Hint: if m is the maximal ideal of , then q mq for someOq > 0.O Then itO followsO from the deﬁnition of completion that OX,a ⊃ OX,a ( /mq)/(q /mq) = ( ˆ /mq ˆ )/(q ˆ /mq ˆ )] OX,a OX,a ∼ OX,a OX,a OX,a OX,a 13. Degree of a dominant morphism Let φ : X Y be a dominant morphism of irreducible varieties. Then φ induces an inclusion

→

k k k k(Y ) ֒ (X) (proposition III.11). The degree deg(φ) of φ is the degree [ (X): (Y )] of the induced extension→ of fields. Theorem III.23 implies that if dim(X) > dim(Y ), then deg(φ)= and φ−1(y) is infinite for each y Y ; • if dim(X) = dim(Y ), then deg(φ) is finite,∞ and| there| is a nonempty Zariski∈ open subset U of Y • such that φ−1(y) is finite for each y U. | | ∈

This suggests an analogy of the degree of morphisms with topological degree of continuous maps. This k analogy is the most evident in the case that char(k)=0 (or more generally, when (X) is separable over

k(Y )), since in that case there is a open dense subset U of Y such that for each y U, the number of elements in φ−1(y) is precisely deg(φ). The preceding statement is a special case of theorem∈ III.44 below. The statement of theorem III.44 is more complicated since it also covers the case of positive characteristics, p in which there are morphisms such as ψ : x x (where p := char(k)) for which the simpler version is

7→ k false. Note that the ﬁeld extension k(X)/ (Y ) induced by ψ in the preceding sentence is not separable. 7

We write degsep(φ) (respectively deginsep(φ)) for the separable degree (respectively inseparable degree) k of the extension k(X)/ (Y ). Note that

deg(φ) = degsep(φ)deginsep(φ) THEOREM III.44. Assume dim(X) = dim(Y ). There exists a nonempty Zariski open subset U of Y such that for each y U, ∈ (1) Y is nonsingular at y; (2) φ−1(y) = deg (φ); | | sep (3) for each x φ−1(y) (a) X is nonsingular∈ at x, ˆ ˆ (b) dimk ( X,x/my X,x) = deginsep(φ), where my is the maximal ideal of Y,y; (4) in particularO O O ˆ ˆ dim k ( X,x/my X,x) = deg(φ) −1 O O x∈φX(y)

PROOF. We may assume without loss of generality that X and Y are afﬁne and nonsingular [why?]. k Let L be the separable closure of k(Y ) in (X). Pick regular functions z1,...,zk on X such that L is the

ﬁeld of fractions of T := k[Y ][z1,...,zk]. There is (up to isomorphism) a unique afﬁne variety Z with k coordinate ring T (exercise III.10). The chain of inclusions k[Y ] ֒ T ֒ [X] induces a factorization of φ of the form: → →

φi φs N

X Z Y k −→ −→ ⊆ 7Given a ﬁeld extension E/F , the separable closure of F in E is the set F¯ of all elements in E which are separable over F . It can be shown that F¯ is a ﬁeld. The separable degree (respectively inseparable degree) of E/F is [F¯ : F ] (respectively [E : F¯]). 38 III. QUASIPROJECTIVE VARIETIES OVER ALGEBRAICALLY CLOSED FIELDS

By theorem III.7 there is g T which generates L over k(Y ). We can factor φ as: ∈ s ψ π

Z Y k Y −→ × −→ d where ψ maps z (φ (z),g(z)) and π is the projection onto Y . Let G(y,t) = s a (y)tds−i

7→ s i=0 i ∈ k k[Y ][t], where t is an indeterminate and ds := deg (φ), be the minimal polynomial of g over (Y ). The

sep P k separability of g over k(Y ) implies that (∂G/∂t) t=g is a non-zero element of [Z]. Let U0 be a nonempty Zariski open subset of Y (V (a ) φ (V ((∂G/∂t| ) ))) [why does such U exist?] and U ′ := (y,t): \ 0 ∪ s |t=g 0 0 {

ds ds−i ′ k i=0 ai(y)t = 0 U0 k. Then U0 is irreducible (since G is irreducible in (Y )[t]), and ψ } ⊂ × ′ induces a birational map from Z to U0 [why?]. Let Y0 U0 be a nonempty Zariski open subset such that P −1 ′ −1⊆ ′ ψ induces an isomorphism φ (Y0) = Y := (π U ′ ) (Y0). Let y0 Y0. Then Y contains (y0,t0) for s ∼ 0 0 0 −1 | ∈ all the roots t0 of G(y0,t). Let z0 := ψ (y0,t0). Since (∂G/∂t)(y0,t0) equals (∂G/∂t) t=g evaluated at z , it follows that (∂G/∂t)(y ,t ) = 0 for every root t of G(y ,t). Consequently,| π−1(y ) = 0 0 0 6 0 0 | 0 | deg(G(y0,t)) = ds = degsep(φ).

−1 ˆ ′ ˆ ′ Claim III.44.1. For each (y0,t0) π (y0), dimk ( Y ,(y ,t )/my Y ,(y ,t ))=1. ∈ O 0 0 0 0 O 0 0 0 −1 ˆ ROOF N+1 k P . Pick (y0,t0) π (y0). The image of G(y,t) in k ,(y0,t0) = [[y1 y0,1,...,yN ∈ NO − − y ,t t ]] (where (y ,...,y ) are coordinates of y in k ) is 0,N − 0 0,1 0,N 0 N ∂G ∂G G(y ,t )+ (y y ) (y ,t ) + (t t ) (y ,t )+ h.o.t. 0 0 j − 0,j ∂y 0 0 − 0 ∂t 0 0 j=1 j X where h.o.t. denotes terms with order (in (y y ,t t )) greater than one. Since G(y ,t )=0 and − 0 − 0 0 0 (∂G/∂t)(y0,t0) = 0, theorem III.5 implies that t is in the ideal of k[[y1 y0,1,...,yN y0,N ,t t0]] 6 ˆ − − − generated by G(y,t) and yj y0,j, j =1,...,N. Since Y ′,(y ,t ) is the quotient of k[[y1 y0,1,...,yN − O 0 0 0 − − y0,N ,t t0]] modulo the ideal generated by G(y,t) (proposition III.41), it follows that t is in the ideal of ˆ − Y ′,(y ,t ) generated by the yj y0,j , which implies the claim. O 0 0 0 − Due to the above claim and the sentence preceding it, in order to prove theorem III.44 it sufﬁces to show that ˜ ˜ −1 there is a nonempty open subset Y of Y0 such that for each y0 Y and z0 := φs (y0) Z, (14) −1 ˆ ˆ ∈ ∈

φ (z ) consists of a single x X and dimk ( /m ) = deg (φ).

i 0 ∈ OX,x yOX,x insep k Pick x1,...,xq k[X] such that [X] = T [x1,...,xq]. Set T0 := T and Tj := Tj−1[xj ] for j = ∈ ej 1,...,q. The minimal equation of x over the field of fractions of T is of the form a xp a =0, j j−1 j,0 j − j,1 where p := char(k). For each j = 0,...,q, let Xj be the unique affine variety with coordinate ring Tj. Fix j, 1 j q. The natural map φ : X X is one-to-one. Pick x X such that ≤ ≤ j j → j−1 ∈ j−1 (i) Xj−1 is nonsingular at x, ′ −1 (ii) Xj is nonsingular at x := φj (x), (iii) a (x) =0. j,0 6 Let us write n := dim(X). Then each Xi has dimension n. After choosing an affine neighborhood of m+n x if necessary, we may assume that x is the origin in k with coordinates (u1,...,um+n) and there are h1,...,hm k[u1,...,um+n] which cut out Xj−1 near x and the linear component of hj is uj, j =1,...,m. It∈ follows that ˆ (iv) k[[u ,...,u ]]. Xj−1 ,x ∼= m+1 m+n O e ′ p j Now, Xj is cut out in Xj−1 k near x by the irreducible polynomial h := xj aj,1/aj,0. Replacing × ′ − xj by xj + c, where c is an appropriate element in k, we may assume that x is the origin with respect to coordinates (u1,...,um+n, xj ). Let a¯j,1 be the image of aj,1 in k[[um+1,...,um+n]]. Since Xj is non- ′ ′ singular at x , it follows that the rank of the Jacobian matrix of (h1,...,hm,h) at x is m+1. Therefore the linear part of a¯j,1 contains at least one of um+1,...,um+n with a non-zero coefficient. After a reordering of (um+1,...,um+n) if necessary we may assume that the linear part of a¯j,1 is um+1. It then follows from pej theorem III.5 that u can be expressed in ˆ ′ as a power series in x ,u ,...,u , so that m+1 OXj ,x j m+2 m+n ˆ (v) ′ = k[[x ,u ,...,u ]]. OXj ,x ∼ j m+2 m+n 13. DEGREE OF A DOMINANT MORPHISM 39

ˆ ˆ k

Let I be any ideal of such that d := dimk ( /I) < . Pick a -basis (g ,...,g ) of OXj−1,x OXj−1 ,x ∞ 1 d ˆ /I. OXj−1 ,x k ej ˆ ˆ Claim III.44.2. x g , 0 k

∗Intersection multiplicity

1. Introduction 1In this chapter we define the intersection multiplicity of n hypersurfaces at a point on a nonsingular variety of dimension n, and prove some of its basic properties. As fig. 1 suggests, nontrivial considerations prop up even in the intersection of a parabola and a line. In this case, i.e the case of two planar polynomial 2 curves f = 0 and g = 0 intersecting at the origin on k the correct (algebraic) definition boils down to

the following: the intersection multiplicity at the origin is the dimension (as a vector space) over k of

the quotient of the power series ring k[[x, y]] by the ideal generated by f,g. For a wonderful axiomatic exposition of this definition see [Ful89, Section 3.3]. There is another more “geometric” way to compute the intersection multiplicity: find a “parametrization” φ(t) of the curve g = 0 such that φ(0) = 0, and compute the order of f(φ(t)) which measures how fast f is vanishing along the curve g =0. For example, as in fig. 1 let f = y x2 and g = y mx. Then φ(t) := (t,mt) parametrizes the line g = 0, and f(φ(t)) = mt t2. Consequently− ord −(f(φ(t))) = 1 if m = 0. If m = 0, i.e. g = 0 is a horizontal − t 6 line, then ordt(f(φ(t))) = 2, as expected. However, to use this approach in practice one needs to define order, parametrization etc. even in the case that the equations are “not reduced”; consider e.g. the case f = y x2 and g = (y mx)2. The algebraic definition gives the expected answer (which is 2 if m =0, and 4 if−m = 0), but what− is geometrically the object (y mx)2 = 0? The underlying space is still6 the − 2 same line y = mx, but the defining equation is different, and the “coordinate ring” k[x, y]/ (y mx) is “non-reduced,” (i.e. it has a nonzero nilpotent, namely the image of y mx). In particular,h in− order toi make the geometric approach more generally applicable one needs to − (1) build a theory of “non-reduced varieties,” and (2) define the notion of order at a point of a “non-reduced curve.” In section IV.2 we introduce the notion of “closed subschemes” which are the correct candidates for “non- reduced subvarieties,” and in section IV.3 we extend the notion of order to non-reduced curves. In sections IV.4 and IV.5 we apply these notions to the study of intersection multiplicity.

y − x2 = 0 As secants approach the tangent at O more and more closely, both of the two points of intersection move arbitrarily close to O.

FIGURE 1. A tangent line intersects a parabola at a point with multiplicity two

1The asterisk in the chapter name is to indicate that most of the material of this section might be skipped in the first reading and/or in a first course of algebraic geometry. The results of this section are not used in chapter VI. For the proof of Bernstein’s theorem and its applications one would mainly need lemma IV.17, theorems IV.12, IV.19 and IV.20, corollary IV.13, and propositions IV.15 and IV.16 - which might be explained without proof in a first course of algebraic geometry.

40 2. CLOSED SUBSCHEMES OF A VARIETY 41

2. Closed subschemes of a variety

A closed subscheme Z of an afﬁne variety X is a pair (Z′, R), where Z′ is a subvariety of X and k R ∼= k[X]/q for some (not necessarily radical) ideal q of [X] such that the subvariety of X determined by I is precisely Z′, i.e. I(Z′) = q in the notation of theorem III.2. We say that Z′ is the support of Z. ′ 2 One can picture Z as a “thickened” version of Z . For example, if X = k and q is the ideal generated 2 2 2 by x, y , then the image of g = a + bx + cy + dx + exy + fy + in k[x, y]/q is determined by ∂g ··· a = g(0, 0) and c = ∂y (0, 0) so that Z is the origin coupled with the vertical tangent line at the origin

(ﬁg. 2). For general q, one can picture k[X]/q as a union of thickened varieties corresponding to primary decompositions (appendix B.7) of q, and different primary decompositions may lead to different pictures - see [Eis95, Section 3.8] for an illuminating exposition.

FIGURE 2. q = x, y2 corresponds to the origin coupled with a vertical tangent h i

Defining subschemes on an arbitrary quasiprojective variety is a bit more complicated than the case of affine varieties, since the same set of equations can look different in different coordinate charts. We want the subscheme to be a notion which would keep track of equations. Giving a set of equations on a neighborhood of a point x is essentially same as giving an ideal of the local ring X,x of X at x. However, the equations at different points need to be compatible, i.e. the equations at all pointsO on a sufficiently small neighborhood must “come from the same set of equations.” This leads to the following definition: a sheaf

of ideals on a quasiprojective variety X is a product x∈X x, where each x is an ideal of X,x, such thatI I I O Q each x X has a nonempty open affine neighborhood U in X (17) ∈ ′ and an ideal I of k[U] such that ′ = I ′ for each x U. Ix OX,x ∈ For us the closed subscheme V ( ) of X determined by would be the product x∈X X,x/ x of quotient rings. (If you already are familiarI with schemes you willI note that we are identifyingO a closedI subscheme with its structure sheaf). An embedded affine chart of V ( ) is an affine open subsetQ U of X which satisfies condition (17). Let U be an open covering of X byI embedded affine charts of V ( ), i.e. for each j, { j} I condition (17) is satisfied with U = Uj and I = Ij for some ideal Ij of k[Uj ]. Then the union of the ′ ′ ′ subvarieties Zj of X determined by Ij is in fact a subvariety Z of X. We say that Z is the support of Z := V ( ), and write Z′ = Supp(Z). Given a point x Z′, we often abuse the notation and say I ∈ x Z. The local ring Z,x of Z at x is the quotient X,x/ x. We say that Z has (pure) dimension k if ∈ ′ O O I and only if Z has (pure) dimension k. If U is an open subset of X, then U := x∈U x is a sheaf of ideals on U; we denote the corresponding closed subscheme of U by the “scheme-theoreticI| I intersection” Z U, and say that it is an open subscheme of Z. There is also a scheme-theoreticQ intersection of two closed∩ subschemes: if Y = V ( ) is a closed subscheme of X corresponding to a sheaf of ideals on X, then the scheme-theoretic intersectionJ Y Z is the closed subscheme of X correspondingJ to the sheaf of ∩ ideals + := x∈X ( x + x). We identify the variety X with its closed subscheme defined by the sheaf ofI zeroJ ideals. ThisI in particularJ implies that the scheme-theoretic intersection Z X is simply Z, as expected. Q ∩

2.1. Rational functions. Usually the notion of rational functions is considered only for irreducible varieties. Exercise III.21 gives an hint that deﬁning a rational function on a reducible variety can get tricky due to the presence of nonzero regular functions which are zero-divisors, i.e. which vanish identically on 2 some irreducible component. Consider e.g. X = V (xy) k , i.e. X is the union of x and y-axes on 2 ⊂ k . In this case both x and y are zero-divisors (in fact xy = 0 on X), and it would be difﬁcult to give geometric interpretation of a ring containing 1/x and 1/y (what would be the “value” of 1/x +1/y at a point on X?). A standard solution is therefore not to allow zero-divisors in the denominator: let Z be a subscheme of X. For each x Z, let S be the localization of at the set of its non zero-divisors, i.e. ∈ x OZ,x 42 IV. ∗INTERSECTION MULTIPLICITY

′ Sx := f1/f2 : f1,f2 Z,x, f2 is not a zero-divisor in Z,x . Let Z := Supp(Z). A rational function { ∈ O ′ O } on Z is an element f = (f : x Z ) ′ S such that x ∈ ∈ x∈Z x each x Z′ has a nonemptyQ open affine neighborhood U in Z′ and (18) ∈ ′ f ,f k[U] such that f ′ = f /f S ′ for each x U. 1 2 ∈ x 1 2 ∈ x ∈ A rational function f = (f : x Z′) on Z is a regular function if each f , andit is an invertible x ∈ x ∈ OZ,x rational function if 1/f is also a rational function, i.e. if (18) holds with the additional condition that no fi ′ ′ is a zero-divisor in ′ for any x U Z . OZ,x ∈ ∩ 2.2. Completeness and compactification of schemes. Let φ : Y X be a morphism of varieties. If = is a sheaf of ideals on X, then φ∗ := φ∗( → ) is a sheaf of ideals on I x∈X Ix I y∈Y Iφ(y) OY,y Y . If φ is an isomorphism of varieties, then for each y Y , is naturally isomorphic to (φ∗ ) = Q Q φ(y) y φ∗( ) ; we say that φ∗ : V ( ) V (φ∗( )) is an∈ embeddedI isomorphism of closed subschemesI . Iφ(y) OY,y I → I A compactification of Z := V ( ) is a closed subscheme Z¯ of a compactification X¯ of X such that Z is embedded isomorphic to an openI subscheme of Z¯. A closed subscheme of a variety is complete if its support is complete; note that Supp(Z¯), being a closed subvariety of a complete variety, is complete (exercise III.33). A fundamental result of Nagata states that every closed subscheme Z of a given variety X can be compactified to a closed subscheme of a given compactification X¯ of X. In this chapter we will use Nagata’s result for the case that dim(Z)=1, which we now prove.

THEOREM IV.1. Let Z be a one dimensional closed subscheme of a quasiprojective variety X and X¯ be a projective compactiﬁcation of X. Then there is a closed subscheme Z¯ of X¯ such that ¯ (1) Z ∼= Z X (2) Supp(Z¯∩) is the closure in X¯ of Supp(Z). (3) every rational function on Z extends to a rational function on Z¯. (4) every invertible rational function on Z extends to an invertible rational function on Z¯.

PROOF. Since Z′ := Supp(Z) has dimension one, S := Z¯′ Z, where Z¯′ is the closure of Z′ in X¯, is finite. \ ¯ ¯ ¯ Claim IV.1.1. There is an affineopen subset W of X containing S and there is a non zero-divisor g k[W ] such that V (g) Z¯′ = S, W := W¯ V (g) X. In addition one can ensure that ∈ ∩ \ ⊂ (a) W is an embedded affine chart of Z, and

(b) the ideal I of k[W ] deﬁning Z W is unmixed, i.e. has no zero-dimensional associated prime ideals. ∩

PROOF. Choose a closed embedding X¯ ֒ PN . After composing φ with an appropriate Veronese embedding (see section III.6.2) if necessary we→ may ensure that (i) X¯ X V (x ), \ ⊆ 0 (ii) V (x0) does not contain neither any irreducible component of X¯ nor any irreducible component of Z¯′, (iii) S V (x )= . ∩ 1 ∅ N N Let Uj := P V (xj ), j = 0,...,N, be the basic open subsets of P . Property (iii) implies that Z′ U = , so\ that we can choose an afﬁne open subset W of X U such that ∩ 1 6 ∅ 1 ∩ 1 (iv) W1 is also an embedded afﬁne chart of Z, and

(v) the ideal I of k[W ] defining Z W is unmixed. 1 1 ∩ 1 Choose a regular function h on U1 such that ′ (vi) V (h) ((Z W1 V (x0)) ((X U1) W1)) (vii) h does⊇not vanish∩ at∩ any point∪ of S ∩U . \ ⊂ 1 Let W¯ := (X¯ U1) V (h) and W := W¯ V (x0). Then W¯ is an affine open subset of X¯ (exercise III.14). Property (vi) implies∩ \ that W = W V (\hx /x ) (note that x /x is a regular function on U W ). 1 \ 0 1 0 1 1 ⊃ 1 Since W1 is an embedded affine chart of Z, it is straightforward to check that so is W . Properties (ii), (v), (vi) and (vii) then imply that the claim holds with g = x0/x1. 2. CLOSED SUBSCHEMES OF A VARIETY 43

¯ Choose W,W,g as in claim IV.1.1. Let I be the ideal of k[W ] deﬁning Z W . Since g is a non ¯ ∩ zero-divisor in k[W ],

¯ ¯ k k (19) the natural map k[W ] [W ] = [W ] is injective. → g ¯ ¯ Deﬁne I := I k[W ]. It is straightforward to check (using basic properties of localizations found e.g. in

∩ ¯ ¯ ¯ k [AM69, Chapter 3]) that g is a non zero-divisor in k[W ]/I and I is generated by I in [W ]. If is the I sheaf of ideals on X defining Z, then it follows that I¯ ¯ is a sheaf of ideals on X¯ x∈X Ix × x∈S OX,x and the corresponding subscheme Z¯ of X¯ satisfies assertions (1) and(2). Now let f = (f : x Z′) be a Q Q x rational function on Z. ∈ Claim IV.1.2. There is an open subset U ′ of Z′ which intersects every irreducible component of Z′, and in addition (18) is satisfied with U = U ′. ′ ′ ′ ′ PROOF. For each irreducible component Zj of Z , choose an open subset Uj of Z such that (18) is ′ ′ ′ ′ satisfied with U = Uj , and Uj does not intersect any irreducible component of Z other than Zj. Then ′ ′ U := j Uj satisfies the claim. LetSU ′ be as in claim IV.1.2. Due to assertion (b) of claim IV.1.1 one can find a regular function q on

′ k W such that Z W V (q) U, and q is a non zero-divisor in both k[W ] and [W ]/I. Note that Z is ∗ ∩ \ ⊆ ∗ deﬁned in W := W V (q) by the ideal generated by I. Moreover, there is f1,f2 k[W ] such that f2

\ ∗ ∗ ∈ ∗ k is a non zero-divisor in k[W ]/I [W ], and fx = f1/f2 X,x/I X,x for each x W Z. Since

∗ ¯ ∈ O O ∈ ∩ ¯ ¯ k k k k[W ]= [W ]q = [W ]gq , it follows that f1/f2 can be representedin the total quotientring of [W ]/I as

q b ¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯ k g q¯ f1/f2 for some integers a,b and q,¯ f1, f2 k[W ] such that q¯ and f2 are non zero-divisors in [W ]/I.

∈ ∗ ∗ ¯ k This proves assertion (3). For assertion (4) note that if f1 is a non zero-divisor in k[W ]/I [W ], then f1 ¯ ¯ is also a non zero-divisor in k[W ]/I. 2.3. Irreducible components, local rings. Let Z be a closed subscheme of a variety X. The irreducible components of Z are simply the irreducible components of Supp(Z). Note that every irreducible component of a closed subscheme Z of X is a subvariety of X. Let Y be a closed irreducible subvariety of Supp(Z). The local ring Z,Y of Z at Y is the set of the equivalence classes of pairs (h,U), where U is an open subset of X suchO that U Y = and h is a regular function on the open subscheme Z U of Z, and the equivalence relation is∩ defined6 ∅ as follows: (h ,U ) (h ,U ) if and only if h = h∩ in ∼ 1 1 ∼ 2 2 1 2 OZ,x for each x U1 U2. It is straightforward to check that Z,Y is a k-algebra. A more explicit realization of is as∈ follows:∩ pick an embedded affine chart U ofOZ such that U Y = . Then U Z is the closed OZ,Y ∩ 6 ∅ ∩ subscheme of U defined by an ideal q of k[U]. On the other hand, since U Y is an irreducible subvariety ∩ of Supp(Z) U, it corresponds to a prime ideal p of k[U] containing q. Then can be identified with

∩ OZ,Y k the localization (k[U]/q)p of [U]/q at the ideal generated by p. This in particular implies that is a OZ,Y local ring. If Y = a is a point of Z, then it is straightforward to check that this definition of Z,Y agrees with the earlier definition of . O OZ,a Lemma IV.2. Let Z be a closed subscheme of an affine variety X, and Y be an irreducible subvariety of

Supp(Z). Given f k[X], if the image of f is invertible in , then it is also invertible in . ∈ OSupp(Z),Y OZ,Y PROOF. This immediately follows from the following observation: if φ : S T is a ring homomorphism such that ker(φ) is contained in the nilradical of S, and if u S is such that→φ(u) is invertible in T , then u is invertible in S. ∈ 2.4. Cartier divisors. A Cartier divisor is a closed subscheme generated locally by single non zero- divisors; it is the natural scheme-theoretic analogue of a “hypersurface.” More precisely, a closed subscheme Z = V ( ) of a variety X is called a Cartier divisor if each x X has a nonempty open afﬁne

I ∈ k neighborhood U in X and an element g k[U] which is not a zero-divisor in [U] such that x′ = g X,x′ for each x′ U. It is straightforward to∈ check that deﬁning a Cartier divisor on X is equivalentI toO pre- scribing a collection∈ (U ,g ) of pairs such that { i i }i U is an open afﬁne covering of X, and •{ i}

g k[U ] are such that • i ∈ i – gi is a non zero-divisor in k[Ui] for each i, and

– g /g is invertible in k[U U ] for each i, j. i j i ∩ j 44 IV. ∗INTERSECTION MULTIPLICITY

If X is of pure dimension n, it follows from theorem III.20 that the support of a Cartier divisor on X is either empty or has pure dimension n 1. − 3. Possibly non-reduced curves This section is devoted to task (2) outlined in section IV.1. A reduced curve is a variety of pure dimension one. By a possibly non-reduced curve we mean a pure dimension one closed subscheme Z of a variety. Unless explicitly stated otherwise, a curve will mean a reduced curve. Our convention of identifying a variety with its subscheme deﬁned by the zero ideal sheaf implies that a “curve” is indeed a special case of a “possibly non-reduced curve.” In section IV.3.1 we describe some properties of curves we are going to use without proof. In section IV.3.2 we deﬁne the notion of order at a point of a possibly non-reduced curve and describe some of its properties whose proofs are deferred to appendix B.11.

3.1. (Reduced) Curves. Curves are in a sense the simplest nontrivial algebraic varieties. Theo- rem IV.3 below states one of their basic properties, namely that they can be desingularized. In particular, the map π : C˜ C from theorem IV.3 is called the desingularization of C. It is the one-dimensional case of resolution of→ singularities, which is still an open problem for dimension greater than 3 in nonzero characteristic. Proofs of theorem IV.3 can be found in many introductory algebraic geometry texts; in particular [Ful89] gives an elementary (but long) proof, and [Kol07, Chapter 1] contains an illuminating exposition of many different proofs.

THEOREM IV.3 ([Sha94, Theorems II.5.6 and II.5.7]). Let C be an irreducible curve. Then there is a nonsingular irreducible curve C˜ and a surjective morphism π : C˜ C such that → (1) for each nonsingular point a C, π restricts to an isomorphism near π−1(a); ∈ (2) if φ : D C is any surjective morphism of curves, then there is a morphism φ˜ : D X such that the following→ diagram commutes. → D φ˜ φ

C˜ π C The curve C˜ is unique up to isomorphism. If C is projective, then so is C˜. Moreover, condition (2) is automatically satisfied if C˜ is projective and π satisfies condition (1). Let a be a nonsingular point on a curve C. Identify an affine neighborhood of a in C with a curve

n k in some k with coordinates (x1,...,xn) deﬁned by an ideal I of [x1,...,xn]. Identity (12) implies that there is f k[x1,...,xn] such that f(a)=0 and ( f)(a) := ((∂f/∂x1)(a),..., (∂f/∂xn)(a)) is not in the span∈ of ( g )(a),..., ( g )(a), where g ,...,g∇ are any set of generators of I. We say ∇ 1 ∇ k 1 k that f is a parameter of C at a. The local ring C,a of C at a is a discrete valuation ring with parameter f (theorem III.33 and exercise III.64); we denoteO the discrete valuation of by ord ( ), and given a OC,a a · rational function g on C, we say that orda(g) is the order of g at a. 2 2 2 Example IV.4. Let a be the origin and C be the parabola V (y x ) k . Since (y x ) a = (0, 1), it follows that x is a parameter of . Since y/x2 is invertible− on C,⊂ it follows that∇ord−(y |)=2. OC,a a |C THEOREM IV.5 ([Sha94, Corollary to Theorem III.2.1]). Let g be a nonzero rational function on a nonsingular curve C. Then there are only ﬁnitely many points a on C such that orda(g) = 0. If C is in addition projective, then 6

(20) orda(g)=0 aX∈C Note that identity (20) may not hold if C is not projective, consider e.g. any non-constant polynomial on the afﬁne line. The following result lists two basic properties of parameters.

n Proposition IV.6. Let C be a curve on k with coordinates (x1,...,xn). Show that

(1) Given a nonsingular point a = (a1,...,an) of C, there is j, 1 j n, such that xj aj is a parameter of C at a. ≤ ≤ − 3. POSSIBLY NON-REDUCED CURVES 45

(2) Given a polynomial f in (x1,...,xn), the property of f being a parameter at a nonsingular point of C is “Zariski open,” i.e. if f is a parameter of C at some nonsingular point of C, then it is a parameter of C at each nonsingular point on a nonempty Zariski open subset of C.

PROOF. The ﬁrst assertion follows immediately from the deﬁnition of parameters and identity (12). For the second assertion note that given a nonsingular point a of C, due to corollary III.37 one can assume,

after replacing C by an appropriate open neighborhood of a on C, that the ideal of C in k[x1,...,xn−1] deﬁned by n 1 polynomials g ,...,g . If f is a parameter of C at a, identity (12) then implies that − 1 n−1 the determinant of the n n-matrix of partial derivatives of g1,...,gn−1,f is nonzero at a; therefore it is nonzero on a nonempty Zariski× open subset U of C. Then f is a parameter of C at each point of U C, as required. ∩ The following is a standard result covered in most introductory books in algebraic geometry. We use it in the proof of the main result (theorem IV.12) of next section.

THEOREM IV.7 ([Sha94, Theorem II.5.8]). Every non-constant morphism φ : C D between two irreducible projective curves is finite, i.e. if U is any affine open subset of D, then φ−→1(U) is affine and

−1 k k[φ (U)] is a ﬁnite module over [U].

−1

k k k Consider the embedding of k 0 ֒ . The coordinate ring of 0 is [x, x ], which is \{ } → \{ } not a finite module over k[x]. This shows that the condition that C and D are projective is crucial in theorem IV.7. Also, if C is the union of the closures of x and y axes in P2, then the projection map from C to the x-axis is not finite; i.e. theorem IV.7 may fail to hold if C is reducible. 3.2. Order at a point on a possibly non-reduced curve. Let a be a point on a possibly non-reduced curve C and f . The order ord (f) of f at a is the dimension of /f as a vector space ∈ OC,a a OC,a OC,a over k. Note that ord (f)= if f vanishes on any irreducible component of Supp(C) containing a. a ∞ 2 2 2 Example IV.8. Let C be the parabola V (y x ) k and f = y. Since y x on C, it follows that − ⊂ ≡ C,0/y C,0 is a 2-dimensional vector space over k generated by 1 and x, so that ord0(y C )=2. Note thatO thisO agrees with the computation from example IV.4. More generally, part (4) of proposition| B.8 shows that when C is a nonsingular (reduced) curve, the two definitions of order agree. Proposition IV.9. Let a be a point on a possibly non-reduced curve C and f . ∈ OC,a (1) If f is a non zero-divisor in , then ord (f) < . OC,a a ∞ (2) orda(f)=0 if and only if f is invertible in C,a. (3) If f is a non zero-divisor in and g O , then ord (fg) = ord (f)+ord (g). OC,a ∈ OC,a a a a The proof of proposition IV.9 is given in appendix B.11. If h is a rational function on C, then h = f/g for some f,g such that g is a non zero-divisor in . We define ord (h) to be ord (f) ord (g). ∈ OC,a OC,a a a − a Proposition IV.9 shows that orda(h) does not depend on the choice of f or g. As example IV.8 suggests, it is straightforward to check using basic properties of discrete valuation rings that this definition of order agrees with the definition from section IV.3.1 when both are applicable, i.e. C is a nonsingular (reduced) curve. ′ 2 2 3 2 ′ Example IV.10. Let C = V (x y + y ) k . We saw in example III.31 that C is singular at the − ⊂ origin. It is straightforward to check that C′,0/y C′,0 is a 2-dimensional vector space over k generated O O ∗ ′ by 1 and x, so that ord0(y C′ )=2. Now we compute the order of π (y) for a desingularization π of C .

| ′

k k Consider the map π : k C defined as follows - given m 0 , the straight line (tu,u): u through the origin with slope→ 1/t intersects C′ at a single point∈ other\{ than} the origin - define{ π(t) to be∈ that} point (see fig. 3). It is straightforward to compute that π(t) = (t t3, 1 t2). It can be checked2 that π is a desingularization of C′. Note that π−1(0) consists of two points− t = −1. It is straightforward to check ∗ 2 −1 ±∗ that π (y ′ )=1 t is a parameter at each a˜ π (0), so that ord (π (y ′ ))=1. It follows that |C − ∈ a˜ |C ∗ ord0(y C′ )= orda˜(π (y C′ )) | −1 | a˜∈πX(0)

2 1 Checking condition (1) of theorem IV.3 is straightforward. To verify condition (2), note that π extends to a map from P to the 2 closure C¯′ of C′ in P , and use the last assertion of theorem IV.3. 46 IV. ∗INTERSECTION MULTIPLICITY

x = ty π(t)

FIGURE 3. Desingularization of the nodal cubic x2 = y2 y3 −

2 k Example IV.11. Let C be the closed subscheme of k corresponding to the ideal of [x, y] generated by (x2 y2 + y3)2. Note that C is non-reduced and Supp(C) is the singular curve C′ from example IV.10. − 2 3 It is not hard to see that C,0/y C,0 is a 4-dimensional vector space over k generated by 1, x, x , x , so that ord (y )=4. CombiningO O this with the observation from example IV.10 yields that 0 |C ∗ ord0(y C )=2 orda˜(π (y C′ )) | −1 | a˜∈πX(0) here the factor 2 on the right hand side is the multiplicity of C′ in C, which we now deﬁne. Let C be a closed subscheme of a variety and D be an irreducible component of Supp(C). Pick an

embedded affine chart U of C such that U D = . Recall that if q (respectively, p) is the ideal of k[U] defining C (respectively, D), then the local∩ ring6 ∅ of C at D can be identified with the localization

OC,D k (k[U]/q)p of [U]/q at the ideal generated by p. Since D is an irreducible component of C, it follows that p is a minimal prime ideal containing q. Therefore the length of C,D as a module over itself is ﬁnite O 2 (proposition B.21); wecall it the multiplicity of D in C and denoteit by µD(C). Inexample IV.11, U = k , 2 2 3 2 2 2 3 q (respectively, p) is the ideal of k[x, y] generated by (x y + y ) (respectively, x y + y ). It is − − straightforward to check directly that C,D ) p C,D ) 0 is a maximal chain of ideals of C,D, which O O O implies that µD(C)=2. The following result, whose proof is given in appendix B.11, is the key relation between orders at a point on a possibly non-reduced curve and the desingularizations of its irreducible components; it shows that our observation from example IV.11 holds in general.

THEOREM IV.12. Let a be a point on a possibly non-reduced curve C. Let C1,...,Cs be the irreducible components of Supp(C) containing a and πi : C˜i Ci be the desingularizations of Ci. If f is a non zero-divisor in , then → OC,a (21) ord (f)= µ (C)ord (f )= µ (C) ord (π∗(f )) a Ci a |Ci Ci a˜ i |Ci i i a˜∈π−1(a) X X Xi Corollary IV.13. Let C be a possibly non-reduced curve and h be an invertible rational function (see section IV.2.1) on C. If Supp(C) is a projective curve, then a∈C orda(h)=0.

PROOF. Let C ,...,C be the irreducible componentsP of C containing a and π : C˜ C be the 1 s i i → i desingularizations of Ci. Theorem IV.12 implies that ord (h)= µ (C) ord (π∗(h )) a Ci a˜ i |Ci i aX∈C X a˜X∈Ci

Corollary B.18 implies that each i, h Ci is a well-deﬁned rational function on Ci. It then follows from ∗ | theorem IV.5 that orda˜(π (h C ))=0 for each i, as required. a˜∈Ci i | i Corollary IV.14. PLet X be a projective variety and C be a projective (reduced) curve on X P1. Assume × no component of C is contained in X a for any a P1. Then each component of C intersects X a ×{ } ∈ ×{ } for every a P1. ∈ PROOF. We may assume without loss of generality that C is irreducible. Fix a point (x0,a0), where 1 1 x0 X and a0 P , on C. Choose an arbitrary point a = a0 P . We will show that C intersects ∈ ∈ 6 ∈ 4. INTERSECTION MULTIPLICITY AT A NONSINGULAR POINT OF A VARIETY 47

1 1 X a . Pick a point P a,a0 . Identify P with k, so that we can treat a,a0 as elements

×{ } ∞ ∈ \{ } \ {∞} k of k. Let t be a coordinate on . Corollary IV.13 implies that ord ((t a) )= ord ((t a ) )=0 (x,b) − |C (x,b) − 0 |C (x,bX)∈C (x,bX)∈C Since f := (t a)/(t a0) is regular and nonzero at all points of X [check that f(x, )=1 for each x X], proposition− − IV.9 implies that × {∞} ∞ ∈ ord ((t a) )= ord ((t a ) ) (x,b) − |C (x,b) − 0 |C

(x,b)∈CX∩(X×{∞}) (x,b)∈CX∩(X×{∞}) k Note that C (X )= C (X k), and for each (x, b) C (X ), \ × {∞} ∩ × ∈ ∩ × positive if b = a, ord(x,b)((t a) C )= − | (0 otherwise. It follows that ord ((t a) )= ord ((t a ) ) ord ((t a ) ) > 0 (x,a) − |C (x,a0) − 0 |C ≥ (x0,a0) − 0 |C (x,a)∈CX∩(X×{a}) (x,a0)∈CX∩(X×{a0}) This implies that C (X a ) = , as required. ∩ ×{ } 6 ∅ 4. Intersection multiplicity at a nonsingular point of a variety 4.1. Intersection multiplicity of power series. The intersection multiplicity at the origin of f ,...,f 1 n ∈

k[[x1,...,xn]] is k

(22) [ψ ,...,ψ ] := dim k ( [[x ,...,x ]]/ f ,...,f ) 1 n 0 1 n h 1 ni Every power series can be approximated up to arbitrarily high order by polynomials. The following result shows that the intersection multiplicity of power series can also be approximated up to arbitrarily high order by (sufﬁciently close) polynomial approximations. Recall that the order ord(f) of a power series f is the smallest m for which there is there is a monomial xα1 xαn with nonzero coefﬁcient in f such that 1 ··· n j αj = m.

PropositionP IV.15. Let f ,...,f k[[x ,...,x ]]. 1 n ∈ 1 n (1) If [f ,...,f ] < , then there exists m 0 such that [g ,...,g ] = [f ,...,f ] for all 1 n 0 ∞ ≥ 1 n 0 1 n 0 g1,...,gn k[x1,...,xn] such that ord(fj gj ) m. (2) If [f ,...,f∈ ] = , then for each N 0, there− exists≥ m 0 such that [g ,...,g ] N for 1 n 0 ∞ ≥ ≥ 1 n 0 ≥

all g ,...,g k[x ,...,x ] such that ord(f g ) m. 1 n ∈ 1 n j − j ≥ PROOF. It follows immediately from theorem B.42, by taking e.g. to be the graded lexicographic order (see example B.39). 4.2. Intersection multiplicity of regular functions. Let a be a nonsingular point of a variety X of dimension n. Recall that the completion ˆ of the local ring of X at a is isomorphic to OX,a OX,a k[[x1,...,xn]] (theorem III.42). If f1,...,fn are regular functions on a neighborhood of a in X, the intersection multiplicity at a of f1,...,fn is ˆ

(23) [f ,...,f ] := dimk ( / f ,...,f ) 1 n a OX,a h 1 ni where we identify each fj with its natural image in ˆX,a (see proposition III.40). Exercise III.71 implies that O

(24) [f ,...,f ] = dim k ( / f ,...,f ) 1 n a OX,a h 1 ni Now we deduce some basic properties of intersection multiplicity using results of the preceding sections. In addition, the “unmixedness theorem” of F. S. Macaulay (theorem B.29) is used in a fundamental way in all the upcoming results of this and the following section. Given a subset S of X and b S, we say that b is an isolated point of S if there is an open neighborhood U of b in X such that b is the only∈ point of S U. ∩ Proposition IV.16. Let V := V (f ,...,f ) X. 1 n ⊆ 48 IV. ∗INTERSECTION MULTIPLICITY

(1) [f ,...,f ] =0 if and only if a V . 1 n a 6∈ (2) 0 < [f1,...,fn]a < if and only if a is an isolated point of V . (3) If 0 < [f ,...,f ] ∞< , then there is a Zariski open neighborhood U of a in X such that 1 n a ∞ V (f2,...,fn) U has pure dimension one. (4) If there is a Zariski∩ open neighborhood U of a in X such that C := V (f ,...,f ) U is a pure 2 n ∩ dimension one subscheme of U, then [f1,...,fn]a = orda(f1 C ). ′ ′ | (5) [f1f1,f2,...,fn]a = [f1,...,fn]a + [f1,f2,...,fn]a. PROOF. Assertion (1) is clear. Assertion (2) follows from identity (24) and the fact that a is an isolated point of V if and only if the radical of the ideal generated by f ,...,f in is the maximal ideal 1 n OX,a of . The third assertion follows from assertion (2) and theorem III.20. Since / f ,...,f = OX,a OX,a h 1 ni ∼ C,a/f1 C,a, the fourth assertion follows from identity (24) and the definition of order. The fifth as- O O ′ sertion is obvious in the case that either [f1,...,fn]a or [f1,f2,...,fn]a is zero or infinite. Otherwise assertion (3) implies that there is C := V (f2,...,fn) is a possibly non-reduced curve near a, and then ′ assertion (2) and lemma IV.17 below imply that f1f1 is a non zero-divisor in C,a. Therefore assertion ′ ′ O ′ (4) and proposition IV.9 imply that [f1f1,f2,...,fn]a = orda((f1f1) C ) = orda(f1 C )+orda(f1 C )= ′ | | | [f1,...,fn]a + [f1,f2,...,fn]a, as required. Lemma IV.17. Let a be a nonsingular point of a variety X of dimension n. Let f ,...,f , m n, be 1 m ≤ regular functions on a neighborhood U of a on X such that a V (f1,...,fm) and V (f1,...,fm) U has pure dimension n m. Then for every j =1,...,m, f is a∈ non zero-divisor in / f ,...,f ∩ . − j OX,a h 1 j−1i ′ PROOF. If some fj is a zero-divisor in X,a/ f1,...,fj−1 , then there is an open neighborhood U O ′ h i of a in X such that fj is a zero-divisor in k[U ]/ f1,...,fj−1 . Due to proposition III.14 and corol- ′′ h ′ i ′′ n+r lary III.37 we can choose a neighborhood U of a in U such that U is an affine variety in k and the ′′ ′′ ideal I(U ) of U in k[x1,...,xn+r] is generated by m polynomials g1,...,gr. Applying Macaulay’s unmixedness theorem (theorem B.29) to g1,...,gr,f1,...,fm, we see that fj is a non zero-divisor in ′′ k[U ]/ f ,...,f , which is a contradiction. h 1 j−1i 4.3. Intersection multiplicity in a family. Let X be a nonsingular affine variety of dimension n. Let hi := j φi,j (t)gi,j , i = 1,...,n, where φi,j are rational functions in an indeterminate t and gi,j are regular functions on X. Let T be the set of all ǫ k such that each φi,j is defined at ǫ (in other words, T is P ∈ the complement in k of the poles of φ ). For each ǫ T , h := h are regular functions on X i,j i,j ∈ ǫ,i i|t=ǫ which can be thought of “deformations” of hǫ ,i for some fixed ǫ0 T . In this section we describe how, Q 0 ∈ as ǫ varies in T , the multiplicities of hǫ,i, i =1,...,n, change locally at a point of X, and globally on all of X.

Example IV.18. Let h(x, t) := (x + 2)(t 2)(xt 2) (see fig. 4) over a field k of characteristic different − − from 2 or 3, and let b0 := 2 k. We compute the multiplicity at x = b0 of h(x, t) for different values of t. If ǫ 1, 2 , the multiplicity− ∈ at x = 2 of h is the same as the multiplicity of x +2, which is 1. 6∈ {− } − |t=ǫ For ǫ =2, h t=ǫ is identically zero, and therefore has infinite multiplicity everywhere. On the other hand, for ǫ = 1, the| multiplicity at x = 2 is − − [h ] = [3(x + 2)2] = 2[x + 2] =2 |t=−1 x=−2 x=−2 x=−2 Note that the point (x, ǫ) = ( 2, 1) is on two distinct irreducible components of the curve h(x, t)=0, namely x = 2 and xt =2. − − − In example IV.18 the multiplicity of h(x, t) at x = b0 is the same for almost all values of t, and it jumps to higher values for t = ǫ only when either the point (b0,ǫ) is non-isolated in V (h(x, t) =0), or (b0,ǫ) is on an irreducible component of V (h(x, t)=0) different from the “vertical line” b0 k. In theorem IV.19 below we show that this is true in general, and then in theorem IV.20 we describe{ the}× “global” analogue of

theorem IV.19. In the statements and proof of these results, t will be a coordinate on k, and we will treat the h (x, t) as regular functions on X T . Let V := V (h ,...,h ) be the closed subscheme of X T j × 1 n × determined by h1,...,hn, and for each ǫ T , let Zǫ be the subvariety Supp(V ) (X ǫ ); in other words, Z is the set of zeroes of h ,...,h ∈,t ǫ. ∩ ×{ } ǫ 1 n −

THEOREM IV.19. Let b0 X, and m : T Z be the function given by ǫ [hǫ,1,...,hǫ,n]b0 . ∈ → ∪{∞} 7→ Let m∗ := min m(ǫ): m T and T˜ := ǫ T : either (b ,ǫ) Z or (b ,ǫ) is isolated in Z . Then { ∈ } { ∈ 0 6∈ ǫ 0 ǫ} 4. INTERSECTION MULTIPLICITY AT A NONSINGULAR POINT OF A VARIETY 49

t = 2

t = 0 t = −1

x = −2

FIGURE 4. Theset of zeroes of h(x, t) = (x + 2)(t 2)(xt 2) − −

(1) T˜ is a Zariski open subset of T . (2) m∗ < if and only if T˜ is nonempty. (3) m∗ =0∞if and only if there is ǫ T such that (b ,ǫ) Z . ∈ 0 6∈ ǫ (4) Assume m∗ < . Then ǫ T : m(ǫ)= m∗ is a nonempty Zariski open subset of T˜. (5) Assume m∗ < ∞. Then for{ ∈ each ǫ T , m(ǫ)}>m∗ if and only if one of the following is true: ∞ ∈ (a) ǫ T˜, i.e. (b ,ǫ) is a non-isolated zero of h ,...,h , or 6∈ 0 ǫ,1 ǫ,n

(b) there is an irreducible component of V (h ,...,h ) X k containing (b ,ǫ) other than 1 n ⊂ × 0

the “vertical line” b k. { 0}× PROOF. The ﬁrst assertion follows from the observation (due to theorem III.20) that (b0,ǫ) is isolated in Zǫ if and only if V has pure dimension one near (b0,ǫ). Assertions (2) and (3) are immediate consequences of the deﬁnition of intersection multiplicity. Now we prove the last two assertions. If there is ∗ ǫ T such that b0 is not a common zero of hǫ,1,...,hǫ,n, then m = 0, and both assertions (4) and (5) ∈ ∗ are true. So assume 0

Claim IV.19.1. is invertible in , where is the only point of k. For each , is f D¯ 0,(b0,∞) P j t ǫj O ∞ \ − invertible in ¯ for each ǫ k ǫ . OD0,(b0,ǫ) ∈ \{ j} Proposition IV.9 and claim IV.19.1 imply that

ord ((t ǫ ) ¯ ) = ord ((t ǫ ) ¯ ) (b0,∞) − 1 |D0 (b0,∞) − 2 |D0 Theorem IV.12, corollary IV.13, proposition IV.16, and claim IV.19.1 then imply that for each ǫ T , ∈ 0 [h ,...,h ] = [t ǫ,h ,...,h ] = ord ((t ǫ) ¯ ) ǫ,1 ǫ,n b0 − 1 n (b0,ǫ) (b0,ǫ) − |D0 = µ ¯ (D¯ )ord ((t ǫ) ¯ )= µ ¯ (D¯ ) C0 0 (b0,ǫ) − |C0 C0 0 To complete the proof it sufﬁces to show that [h ,...,h ] > µ ¯ (D¯ ) whenever ǫ T T . This ǫ,1 ǫ,n b0 C0 0 ∈ \ 0 inequality is clear if (b0,ǫ) is non-isolated in Zǫ, so assume (b0,ǫ) is an isolated point of Zǫ and there are irreducible components of V containing (b0,ǫ) other than C0; denote them by C1,...,Ck. Choose a Zariski open neighborhood W of (b ,ǫ) in X T such that C W,...,C W are the only irreducible 0 × 0 ∩ k ∩ 50 IV. ∗INTERSECTION MULTIPLICITY components of the open subscheme V W of V . Then ∩ [h ,...,h ] = [t ǫ,h ,...,h ] = ord ((t ǫ) ) ǫ,1 ǫ,n b0 − 1 n (b0,ǫ) (b0,ǫ) − |V ∩W k = µ (V W )ord ((t ǫ) ) >µ (V W ) Ci∩W ∩ (b0,ǫ) − |Ci∩W C0∩W ∩ i=0 X ¯ Since V ∩W,C0∩W = D¯ 0,C0 (see section IV.2.3), it follows that µC0∩W (V W ) = µC¯0 (D0), which completesO the proof. O ∩

Now we prove a global counterpart of theorem IV.19. To motivate the statement of this result, we compute for different values of ǫ the number (counted with multiplicities) of isolated solutions of h t=ǫ, for h(x, t) := (x + 2)(t 2)(xt 2) from example IV.18. It is straightforward to check (see fig. 4)| that − − if ǫ 0, 2 , then this number is 2; indeed, if ǫ 1, 0, 2 , then h t=ǫ has two solutions of multiplicity one:6∈( { 2,ǫ}), (2/ǫ,ǫ), and if ǫ = 1, then h 6∈has {− one solution} of| multiplicity two: ( 2, 1). On the − − |t=ǫ − − other hand, if ǫ =2, the polynomial h is identically zero on k, and therefore has zero isolated solutions. |t=ǫ Finally, for ǫ = 0 there is only one point of multiplicity one on h t=ǫ = 0, namely the point ( 2,ǫ); the other solution (2/ǫ,ǫ) “goes to infinity” at t = 0. In particular, the| total number of isolated solutions− of h t=ǫ is equal for almost all values of ǫ, and can only drop in exceptional cases when some of the solutions become| non-isolated or run to infinity. Theorem IV.20 below states that this is also the case in general; in particular, “minimum” in the local case (i.e. theorem IV.19) becomes “maximum” in the global case (i.e. theorem IV.20). The notation (in particular the meaning of V and Zǫ) below remains unchanged from theorem IV.19.

THEOREM IV.20. Let C X T be the union of all irreducible components of V containing at least one isolated point of Z for some⊂ ǫ× T . ǫ ∈ (1) Either C is empty, or it has pure dimension one. (2) C := C (X ǫ ) is ﬁnite for every ǫ T . ǫ ∩ ×{ } ∈ Now assume C is not empty. (3) Let T ∗ be the set ofall ǫ T such that all points on C are isolated in Z . Then T ∗ is a nonempty ∈ ǫ ǫ

Zariski open subset of k. ˜ (4) For each ǫ T , let Cǫ := (b,ǫ) X T : (b,ǫ) is isolated in Zǫ . The function M : T Z ∈ { ∈ × } → given by ǫ ˜ [hǫ,1,...,hǫ,n]b achieves the maximum on a nonempty Zariski open 7→ (b,ǫ)∈Cǫ subset of T ∗. P (5) If ǫ T , then M(ǫ) fails to attain the maximum if and only if at least one of the following is true: ∈ ∗ (a) ǫ T , i.e. there is a point on Cǫ which is a non-isolated zero of hǫ,1,...,hǫ,n, or (b) C6∈“has a point at infinity at t = ǫ”, i.e. if X¯ is a projective compactification of X and C¯ is the closure of C in X¯ P1, then C¯ ((X¯ X) ǫ ) = . × ∩ \ ×{ } 6 ∅ PROOF. If (b,ǫ) is an isolated point of Z , which is defined by n+1 regular functions h ,...,h ,t ǫ ǫ 1 n − on a variety of dimension n +1, theorem III.20 implies that V = V (h1,...,hn) has pure dimension one near (b,ǫ), which proves assertion (1). If assertion (2) does not hold, then assertion (1) implies that there is ǫ T such that X ǫ contains an irreducible component of C. But then no point on this component ∈ ×{ } is isolated in Zǫ, contradicting the definition of C. This proves assertion (2). For assertion (3), let Y be the union of irreducible components of V (f1,...,fn) X. Since C is nonempty, it follows that C (Y T ) ′ ′ ⊂ ∗ ′ ∩ × is a finite set. If (bj ,ǫj) j are the points in this intersection, then note that T = T ǫj j. It remains to prove the last two{ assertions.} Let \{ } C˜ := C˜ = (b,ǫ) C : (b,ǫ) is isolated in Z ǫ { ∈ ǫ} ǫ [ Then C˜ is a Zariski open subset of C, i.e. there is a Zariski open subset U of X T such that C˜ = U C. × ∩ Let D˜ := V U be the corresponding open subscheme of V . Now choose a projective compactification ¯ ∩ ¯ ¯ 1 ˜ X of X. Let C be the closure of C in X P . For every ǫ k and every (b,ǫ) C, lemma IV.17 × ∈ ∈ implies that t ǫ is a non zero-divisor in ˜ = . If ǫ ,ǫ are distinct elements of k, then − OD,(b,ǫ) OV,(b,ǫ) 1 2 theorem IV.1 implies that there is a closed subscheme D¯ of X¯ P1 containing D˜ as an open subscheme × 5.INTERSECTIONMULTIPLICITYOFCOMPLETEINTERSECTIONS 51 such that Supp(D¯)= C¯, and each of t ǫ1,t ǫ2,f := (t ǫ1)/(t ǫ2) extends to an invertible rational function on D¯. The following is an immediate− − consequence− of lemma−IV.2. 1 ¯ ¯ Claim IV.20.1. Denote the only point of P k by . For each b X such that (b, ) C, f is invertible \ ∞ ∈ ¯ ∞ ∈ in ¯ . For each j, t ǫ is invertible in ¯ for each (b,ǫ) C such that ǫ k ǫ . OD,(b,∞) − j OD,(b,ǫ) ∈ ∈ \{ j}

Proposition IV.9 and claim IV.20.1 imply that for each ǫ ,ǫ k, 1 2 ∈ ord(b,∞)((t ǫ1) D¯ )= ord(b,∞)((t ǫ2) D¯ ) ¯ − | ¯ − | b:(b,X∞)∈C b:(b,X∞)∈C

Theorem IV.12, corollary IV.13, proposition IV.16, and claim IV.20.1 then imply that for each ǫ k, ∈ ∗ M := ord ((t ǫ) ¯ )= µ (D¯)ord ((t ǫ) ) (b,ǫ) − |D Ci (b,ǫ) − |Ci ¯ ¯ i b:(b,ǫX)∈C b:(b,ǫX)∈C X is constant, where the Ci are irreducible components of C. On the other hand, theorem IV.12 and proposition IV.16 imply that for all ǫ T , ∈ M(ǫ)= [h ,...,h ] = [t ǫ,h ,...,h ] ǫ,1 ǫ,n b − 1 n (b,ǫ) ˜ ˜ b:(b,ǫX)∈Cǫ b:(b,ǫX)∈Cǫ = ord ((t ǫ) )= µ (D¯)ord ((t ǫ) ) (b,ǫ) − |D Ci (b,ǫ) − |Ci ˜ ˜ i b:(b,ǫX)∈Cǫ b:(b,ǫX)∈Cǫ X Since t ǫ is regular and has a zero at each point of C¯ (X ǫ ), it follows that M ∗ M(ǫ); moreover, ∗ − ˜ ¯ ¯ ∩ ×{ } ¯ ≥ M > M(ǫ) if and only if Cǫ $ C (X ǫ ). By construction of C, the latter condition is true if and only if at least one of the conditions∩ of assertion×{ } (5) holds. Since these conditions hold at at most finitely many points of C, assertion (4) also holds. 5. Intersection multiplicity of complete intersections Let f ,...,f be regular functions on a nonsingular variety X of dimension n k, and let Y be the 1 k ≥ closed subscheme of X defined by (the ideal generated by) f1,...,fk. If Z is an irreducible component of Y of dimension n k, then the intersection multiplicity [f ,...,f ] of f ,...,f along Z is the − 1 k Z 1 k multiplicity µZ (Y ) of Z in Y (recall that µZ (Y ) was defined in section IV.3.2 in the paragraph preceding theorem IV.12). If k = n, then Z is a singleton a , and = / f ,...,f , and corollary B.20 { } OY,Z ∼ OX,a h 1 ni implies that µZ (Y ) = dimk ( X,a/ f1,...,fn ), so that the definition of [f1,...,fn]Z from this section agrees with the definition fromO the precedingh section.i

Proposition IV.21. Let f1,...,fn−1 k[x1,...,xn]. Let Y := V (x1,...,xk). Assume ∈ n (1) Y is an irreducible component of V (f1,...,fk) k . (2) V (f ,...,f ) has a one dimensional irreducible⊂ component Z. k+1|Y n−1|Y (3) Z is not contained in any irreducible component of V (f1,...,fk) other than Y . For each ǫ = (0,..., 0,ǫ ,...,ǫ ) Y , and each j = 1,...,k, we write f for the polynomial in k+1 n ∈ j,ǫ (x1,...,xk) obtained by substituting ǫi for xi for i = k +1,...,n. Then [f ,...,f ] = [f ,...,f ] [f ,...,f ] 1 n−1 Z 1,ǫ k,ǫ 0 k+1|Y n−1|Y Z for generic ǫ Y . ∈ PROOF. We prove this by induction on n k. At ﬁrst consider the case that n k = 1. Then Z is − − the xn-axis, and for generic ǫ k, Z is the only irreducible component of V (f1,...,fn−1) containing a := (0,..., 0,ǫ). Since ord ∈((x ǫ) )=1, theorem IV.12 implies that ǫ aǫ n − |Z [f ,...,f ] = [x ǫ,f ,...,f ]

1 n−1 Z n − 1 n−1 aǫ k

= dim k ( [[x ,...,x , x ǫ]]/ x ǫ,f ,...,f )

1 n−1 n − h n − 1 n−1i k

= dim k ( [[x ,...,x ]]/ f ,...,f ) 1 n−1 h 1|xn=ǫ n−1|xn=ǫi = [f1,ǫ,...,fn−1,ǫ]0 as required. In the general case, pick a nonsingular point z = (0,..., 0,zk+1,...,zn) of Z. Then there is j, k +1 j n, such that (x z ) has order one at z (proposition IV.6, assertion (1)). Pick a generic ≤ ≤ j − j |Z 52 IV. ∗INTERSECTION MULTIPLICITY

ǫ k. Assertion (2) of proposition IV.6 implies that the set V (x ǫ ) Z is nonempty and contains j ∈ j − j ∩ a nonsingular point a of Z such that a is not in any other irreducible component of V (f1,...,fn−1), and orda((xj ǫj ) Z )=1. Since a is an isolated zero of xj ǫj ,f1,...,fn−1, theorem III.20 implies that − | − ′ ′ ′ xj ǫj ,f1,...,fn−2 deﬁnes a possibly non-reduced curve Z near a. Let Z1,...,Zs be the irreducible − ′ ˜ ′ components of Z and πi : Zi Zi be the desingularization. Theorem IV.12 and proposition IV.16 imply that →

[f ,...,f ] = [x ǫ ,f ,...,f ] = ord (f ′ ) 1 n−1 Z j − j 1 n−1 a a n−1|Z ′ ∗ = µZ′ (Z ) orda˜(π (fn−1 Z′ )) i i | i i a˜∈π−1(a) X Xi ∗ ′ ′ = [f1 xj =ǫj ,...,fn−2 xj =ǫj ]Z orda˜(π (fn−1 Z )) | | i i | i i a˜∈π−1(a) X Xi Let Y := Y V (x ǫ ). Then the inductive hypothesis implies that ǫj ∩ j − j ∗ ′ ′ [f1,...,fn−1]Z = [f1,ǫ,...,fk,ǫ]0 [fk+1 Yǫ ,...,fn−2 Yǫ ]Z orda˜(π (fn−1 Z )) | j | j i i | i i a˜∈π−1(a) X Xi = [f ,...,f ] ord (f ′′ ) 1,ǫ k,ǫ 0 a n−1|Z where Z′′ is the closed subscheme Y Z′ of Y . It follows that ∩ [f ,...,f ] = [f ,...,f ] [x ǫ ,f ,...,f ] 1 n−1 Z 1,ǫ k,ǫ 0 j − j k+1|Y n−1|Y a = [f ,...,f ] [f ,...,f ] 1,ǫ k,ǫ 0 k+1|Y n−1|Y Z as required.

Corollary IV.22. Let the assumptions be as in proposition IV.21. Let fn k[x1,...,xn] and a be a nonsingular point on Z. Then ∈ [f ,...,f ] = [f ,...,f ] [f ,...,f ] 1 n a 1,ǫ k,ǫ 0 k+1|Y n|Y a for generic ǫ Y . ∈ PROOF. Follows from propositions IV.16 and IV.21. CHAPTER V

Convex polyhedra

A “polytope” has two equivalent definitions: a convex hull of finitely many points, or a bounded intersection of finitely many “half spaces.” In sections V.1 and V.2 we prove the equivalence of these definitions after introducing the basic terminology. The rest of the chapter is devoted to different properties of polytopes which are implicitly or explicitly used in the forthcoming chapters.

1. Basic notions In this chapter we treat the spaces Rn, n 0, as vector spaces over R equipped with the Euclidean ≥ topology, and deal only with “affine maps” between these spaces. Recall that a map φ : Rn Rm is affine m n m → n if there is β R such that φ( ) = β + φ0( ) for some linear map φ0 : R R . Given S R ∈ · · → ⊂ and T Rm, and a map φ : S T , we say that φ is affine if it is the restriction of an affine map ⊂ → from Rn Rm; we say that φ is an affine isomorphism if φ is affine and bijective. An affine subspace → A of Rn is a subset of Rn which is the image of an affine map; in other words it is simply a translation of a linear subspace L of Rn. The dimension dim(A) of A is the dimension of L as a vector space over R. A hyperplane in Rn is an affine subspace of dimension n 1. The affine hull aff(S) of a set S of − Rn is the smallest affine subspace of Rn containing S; alternatively, if L is the linear subspace of Rn spanned by all elements of the form α β such that α, β S, then aff(S) = L + α for any α S. In − ∈ ∈ fig. 1, it is straightforward to check that aff(S) = R2, aff( C ) = C , and since A, B, E are collinear, aff( A, B, E ) is the (unique) line L through these points. { } { } { }

D L B

E ν A

FIGURE 1. S = A,B,C,D,E , In (S)= C , ld (S)= A,E,B { } ν { } ν { } It is straightforward to check that an affine map preserves affine subspaces, affine hulls, and if the map is injective, then also the dimension of affine hulls (exercise V.1). Given α Rn and ν (Rn)∗, we write ∈ ∈ ν, α for the “value of ν at α,” and write ν⊥ := α Rn : ν, α =0 . Define h i { ∈ h i } min(ν) := min ν, α : α S , provided the minimum exists. S {h i ∈ } max(ν) := max ν, α : α S , provided the maximum exists. S {h i ∈ } Inν (S) := α S : ν, α = min(ν) , provided min(ν) exists. { ∈ h i S } S ldν (S) := α S : ν, α = max(ν) , provided max(ν) exists. { ∈ h i S } S See fig. 1 for an illustration of these notions for a planar set. Note that in fig. 1 we depicted ν (R2)∗ ∈ on R2 by identifying it with an element (modulo “parallel translations”) in R2 via the “dot product.” A set is convex if it contains the line segment joining any two points in it. The convex hull conv(S) of S is

53 54 V. CONVEX POLYHEDRA the smallest convex set containing S. In fig. 1 the convex hull of the 5 points is the green triangle. Given n k α1,...,αk R , an expression of the form j=1 ǫj αj , where the ǫj are nonnegative numbers whose sum ∈ n is 1, is called a convex combination of the αj . The set of convex combinations of two elements in R is precisely the line segment joining them, andP from this observation it can be shown that the convex hull of a set consists of the convex combinations of its points (exercise V.2), i.e. k k (25) conv(S)= ǫ α : k 0, α S, ǫ 0 for each j, 1 j k, and ǫ =1 { j j ≥ j ∈ j ≥ ≤ ≤ j } j=1 j=1 X X Given a nonnegative real number r and S Rn, we define rS := rα : α S , i.e. rS is the “dilation of S by a factor of r.” We say that S is a cone⊂if it is “dilation invariant,”{ i.e.∈ if rS} S for each r 0. The convex cone generated by S is defined to be ⊆ ≥ k (26) cone(S) := ǫ α : k 0, α S and ǫ 0 for each j, 1 j k { j j ≥ j ∈ j ≥ ≤ ≤ } j=1 X We say that cone(S) is finitely generated if S is finite. A convex polyhedron is a subset of Rn defined by P finitely many linear inequalities, i.e. inequalities of the form a0 + a1x1 + + anxn 0. Geometrically, a polyhedron is a finite intersection of “half-spaces,” where a “half-space”··· is the set≥ of all points on one side of a hyperplane - see fig. 2. If is bounded, we call it a convex polytope, and if it is a cone, we call it a convex polyhedral cone. A convexP polyhedral cone is equivalently a set defined by finitely many linear inequalities with zero constant term (exercise V.6). In this book we only consider convex polyhedra, and therefore we will simply write “polyhedra,” “polytopes,” “cones,” “polyhedral cones” to mean “convex polyhedra,” “convex polytopes,” “convex cones,” “convex polyhedral cones” respectively. The dimension dim( ) of a polyhedron is the dimension of its affine hull. Figure 2 depicts a few two dimensional convexP polyhedra. P

y y y

(0, 0) (0, 0) x x x (0, 0)

Polytope Unbounded Polyhedral cone polyhedron

FIGURE 2. Some planar convex polyhedra; “half-planes” deﬁning the cone are also depicted.

A strongly convex cone is a convex cone which does not contain any line through the origin, equiva- C lently, for all α 0 , α . The Minkowski sum of two subsets , of Rn is + := α + β : α , β ∈C\{. It is straightforward} − 6∈ C to check that the Minkowski sumP ofQ convex setsP is alsoQ convex{ (ex- ∈P ∈ Q} ercise V.8); see fig. 3 for some examples in R2. We now show that every convex polyhedral cone or finitely generated cone1 can be represented as the Minkowski sum of a linear subspace and a strongly convex cone. Proposition V.1. Let be a convex cone in Rn. C (1) If is a polyhedralcone, then there is a strongly convex polyhedral cone ′ and a linear subspace C C L of Rn such that = ′ + L and aff( ′) L = 0 . C C C ∩ { } 1We will see in section V.2 that convex polyhedral cones and finitely generated cones are the same. However, we will use proposition V.1 in the proof of this equivalence. 1. BASIC NOTIONS 55

y y y y y y

x x x x x x

P Q P + Q P Q P + Q

y y y y y y

x x x x x x

P Q P + Q P Q P + Q

FIGURE 3. Minkowski sums of planar sets

(2) If is a finitely generated cone, then there is a strongly convex finitely generated cone ′ and a C C linear subspace L of Rn such that = ′ + L and aff( ′) L = 0 . C C C ∩ { } (3) Let L′ be a linear subspace of Rn such that L′ aff(C) = 0 . Then is a polyhedral cone (respectively, finitely generated cone) if and only∩ if + L′ is{ a} polyhedralC cone (respectively, finitely generated cone). C PROOF. Let L be the (unique) maximal linear subspace of Rn contained in (see exercise V.9). After a linear changeof coordinatesif necessary, we may assume that L is the coordinateC subspace spanned by the first k coordinates. At first assume is a polyhedral cone. Exercise V.6 implies that it is defined by finitely C many inequalities of the form ai,1x1 + + ai,nxn 0, 1 i m. For each j =1,...,n, write ej for n ··· ≥ ≤ ≤ the j-th standard unit vector in R . Since rej L for each j =1,...,k, and each r R, it follows ′ ∈ ⊆ C ∈ that ai,1 = = ai,k =0 for each i. Let be the polyhedral cone on the (n k)-dimensional coordinate ··· n C − subspace of R spanned by ek+1,...,en defined by the inequalities ai,k+1xk+1 + + ai,nxn = 0. Then assertion (1) holds with ′ and L. Now assume is the cone generated by finitely··· many elements n C C α1,...,αm R . Let αi := (αi,1,...,αi,n). Since α˜i := ( αi,1,..., αi,k, 0,..., 0) L , it ∈′ − − ∈ ⊆ C follows that αi := αi +α ˜i = (0,..., 0, αi,k+1,...,αi,n) for each i. Then assertion (2) holds with L and ′ := cone(α′ ,...,α′ ). The proof of assertion (3) follows∈C via the same arguments as in the proof of C 1 m the preceding assertions by choosing a system of coordinateson Rn such that L′ is the coordinate subspace spanned by the first k′ coordinates, where k′ := dim(L′). Remark V.2. If Rn and d is a positive integer, then d has two natural interpretations: one is the dilation (which isA how ⊂ we defined it), and the other is the MinkAowski sum of d copies of . If is convex, then these are equivalent (exercise V.10). A A 1.1. Exercises. EXERCISE V.1. Let φ : Rn Rm be an affine map and let S Rn. Show that → ⊆ (1) If A is an affine subspace of Rn, then φ(A) is an affine subspace of Rm, and dim(φ(A)) dim(A). ≤ (2) φ(aff(S)) = aff(φ(S)) and φ(conv(S)) = conv(φ(S)). (3) Assume φ is in addition a linear map, i.e. φ(0) = 0. Then φ(cone(S)) = cone(φ(S)). Now assume φ is in addition injective. Then show that (4) The inequality in assertion (1) holds with equality. (5) The converses of assertions (1),(2) and(3) are also true, i.e. (a) if A Rn is such that φ(A) is an affine subspace of Rm, then A is also an affine subspace ⊆ of Rn, (b) φ−1(aff(S)) = aff(S) and φ−1(conv(φ(S)) = conv(S), (c) if φ is a linear map, then φ−1(cone(φ(S)) = cone(S). EXERCISE V.2. Show that the line segment joining α, β Rn is precisely the set of their convex n ∗ ∈ n combinations. Use it to prove identity (25). If ν (R ) and S R are such that minS(ν) exists, then ∈ ⊂ use identity (25) to show that minconv(S)(ν) = minS(ν). 56 V. CONVEX POLYHEDRA

EXERCISE V.3. Check that a convex polyhedron as defined in section V.1 is actually convex. EXERCISE V.4. Let S Rn. ⊆ (1) Show that aff(S) is zero dimensional if and only if S is a “singleton” (i.e. it consists of a single point). Deduce that zero dimensional polyhedra are precisely singletons. (2) Assume S is convex and closed in Rn, and aff(S) is one dimensional. Show that there is an affine isomorphism φ : R aff(S) such that φ−1(S) is a (possibly unbounded) closed interval of R. → EXERCISE V.5. Show that every finitely generated two dimensional cone in R2 can be generated as a cone by two nonzero elements of R2 (these correspond to “edges” of S). If S is in addition strongly convex, show that the generators of S are linearly independent. EXERCISE V.6. Let Rn. P ⊆ (1) Assume is “dilation invariant,” i.e. for every α and every r 0, rα . Show that for P ∈ P ≥ ∈ P every ν (Rn)∗, inf ν, α : α is either 0 or . (2) Deduce∈ that is a convex{h i polyhedral∈ P} cone if and−∞ only if it can be defined by finitely many inequalities withP zero constant term, i.e. inequalities of the form a x + + a x 0. 1 1 ··· n n ≥ EXERCISE V.7. If S,T Rm and φ : Rm Rn is a linear map, show that φ(S +T )= φ(S)+φ(T ). ⊆ → EXERCISE V.8. Let , be subsets of Rn. P Q (1) If = conv(S) and = conv(T ), then show that + = conv(S + T ). [Hint: if δ = P Q P Q i i j ǫj =1, then i δiai + j ǫj bj = i,j δiǫj (ai + bj ).] (2) Deduce that if and are convex (respectively, convex hulls of finitely many points,P finitely Pgenerated cones)PP thenQ so is P+ . P P Q EXERCISE V.9. Let be a convex subset of Rn containing the origin. Show that contains a unique C C maximal linear subspace of Rn, i.e. there is a linear subspace L of Rn such that L , and if L′ is n ′ ⊆ C n any linear subspace of R contained in , then L L. [Hint: if L1,L2 are two linear subspaces of R contained in , then L + L .] C ⊆ C 1 2 ⊆C EXERCISE V.10. Let Rn and d be a positive integer. A⊂ (1) If is convex, then show that dα : α = α1 + + αd : αj for each j . (2) GiveA an example to show that the{ above∈ equality A} { may not··· hold if is∈ not A convex. } A EXERCISE V.11. Let S Rm and n := dim(aff(S)). ⊂ (1) Show that there is T Rn and an injective affine map φ : Rn Rn such that aff(T ) is Rn, and ⊆ → φ(T )= S. [Hint: aff(S) is a translation of an n-dimensional linear subspace of Rm.] (2) If S contains the origin (this is the case when e.g. S is a cone) then show that it is possible to ensure in assertion (1) that φ is in addition a linear map.

EXERCISE V.12. Show that each of the following properties is invariant under injective affine maps, n1 n2 i.e. if φ : R R is an injective affine map then each of the following properties holds with S = S1 → and n = n1 if and only if it holds with S = φ(S1) and n = n2. (1) S is an affine subspace of Rn, (2) S is a convex subset of Rn, (3) S is the convex hull of finitely many points in Rn, (4) S is convex polyhedron in Rn, (5) S is convex polytope in Rn. EXERCISE V.13. Show that each of the following properties is invariant under injective linear maps, n1 n2 i.e. if φ : R R is an injective linear map then each of the following properties holds with S = S1 → and n = n1 if and only if it holds with S = φ(S1) and n = n2. (1) S is a linear subspace of Rn, (2) S is a convex cone in Rn, (3) S is a strongly convex cone in Rn, (4) S is a finitely generated convex cone in Rn, (5) S is a convex polyhedral cone in Rn. 2. CHARACTERIZATION OF CONVEX POLYHEDRA 57

EXERCISE V.14. Let H be a linear subspace of Rn. (1) Show that every linear map ν : H R can be extended to a map Rn R. → → (2) Assume H η⊥ for some η (Rn)∗. Let S Rn be a ﬁnite set such that η, α > 0 for each α S,⊆ and let c be an arbitrary∈ real number.⊂ Show that in assertion (1) ith can bei ensured ∈ that ν′, α > c for each α S. [Hint: after a linear change of coordinates on Rn we may assumeh thati H is spanned by∈ the ﬁrst k-coordinates and η is the projection onto the (k + 1)-th coordinate.] EXERCISE V.15. Given S Rn, consider the following property: ⊂ (27) there is ν (Rn)∗ such that ν, α > 0 for each α S 0 . ∈ h i ∈ \{ } n1 n2 n1 Let φ : R R be an injective linear map and S1 R . Show that property (27) holds with S = S1 → ⊆ and n = n1 if and only if it holds with S = φ(S1) and n = n2. [Hint: use assertion (1) of exercise V.14.] EXERCISE V.16. Let S Rn 0 . Show that the origin is in the convex hull of S if and only if cone(S) contains a line through⊆ the\{ origin.} [Hint: 0 can be written as a convex combination of elements from S if and only if there is α cone(S), α =0, such that α is also in cone(S).] ∈ 6 − EXERCISE V.17. Let be a convex cone in Rn and π : Rn Rm be a linear map such that ker(π) C → ∩ = 0 . Show that is a strongly convex cone in Rn if and only if π( ) is a strongly convex cone in Rm. C[Hint:{ use} exercise V.16C .] C

2. Characterization of convex polyhedra In this section we describe a characterization of convex polytopes and polyhedral cones, and use it to show that every polyhedron is the Minkowski sum of a polytope and a polyhedral cone. The proofs we give are elementary and geometric, but not the most “efﬁcient”; see e.g. [Sch98, Section 7.2] for a quicker proof (which is somewhat less intuitive in the beginning), and [Zie95, Section 1.3] for a more algorithmic proof.

THEOREM V.3 (Farkas (1898, 1902), Minkowski (1896), Weyl (1935)). Let be a convex subset of P Rn. (1) is a polytope if and only if it is the convex hull of finitely many points. (2) P is a polyhedral cone if and only if it is a finitely generated cone. P (3) Assume is a polyhedral cone. Then it is strongly convex if and only if there is ν (Rn)∗ such that ν, αP > 0 for each α 0 . ∈ h i ∈P\{ } PROOF. We are going to prove assertions (2), (3) and ( ) implication of assertion (1). The ( ) direction of assertion (1) follows from these assertions; it is left⇐ as an exercise (exercise V.19). We start⇒ with the proof of assertion (3). For the ( ) direction note that if contains both α and α for some ⇐ P − nonzero α Rn, then for all ν (Rn)∗, either ν, α 0 or ν, α < 0. For the ( ) direction we proceed by∈ induction on dim( )∈. Due to exercisesh V.11i ≤, V.13 andh −V.15i we may assume⇒ without loss of P generality that dim( ) = n. Since the only strongly positive cones in R are R≥0 and R≤0, it holds for n = 1. In the generalP case is defined by finitely many inequalities of the form ν, x 0 for nonzero P h i ≥ ν (Rn)∗ (exercise V.6). Take one such ν. Since ν is a smaller dimensional strongly convex ∈ ⊥ ∩P polyhedral cone, it follows by induction that there is η (Rn)∗ which is positive on (ν⊥ ) 0 . ∈ ∩P \{ } Claim V.3.1. For sufficiently small ǫ> 0, ν + ǫη is positive on 0 . P\{ } PROOF. If the claim is false, then we can find a sequence of positive numbers ǫk 0 and αk n−1 n−1 n → ∈ S , where S is the unit sphere centered at the origin in R , such that ν + ǫkη, αk < 0. Since ∩P n−1 h n−1 i is closed and S is compact, we may assume that the αk converge to α S . Then ν, α = limP ν + ǫ η, α 0, so that ν, α = 0. It follows that α Sn−1 ∈ν⊥ P ∩ , so that η,h α >i 0. k→∞h k ki ≤ h i ∈ ∩ ∩P h i By continuity η is positive on an open neighborhood U of α, which means that ν + ǫkη, αk > 0 for sufficiently large k. This contradiction proves the claim. h i Claim V.3.1 finishes the proof of assertion (3). We now prove the ( ) direction of assertion (2) by induction on dim( ). Due to proposition V.1 and exercises V.11 and V.13⇒we may assume without loss of P generality that is an n-dimensional strongly convex polyhedral cone in Rn. As in the proof of assertion P 58 V. CONVEX POLYHEDRA

(3), the n =1 case follows directly from the observation that the only strongly convex polyhedral cones in n ∗ R are R≥0 and R≤0. Now consider the case n 2. Assertion (3) implies that there is ν (R ) such that ν, α > 0 for each α 0 . ≥ ∈ h i ∈P\{ } Claim V.3.2. ′ := α : ν, α =1 is bounded. P { ∈P h i } ′ PROOF. Indeed, otherwise take a sequence αk such that αk is unbounded, where is the n ∈P || || || · || Euclidean norm on R . Note that each αk/ αk is in the intersection of and the (n 1)-dimensional unit sphere Sn−1. Since Sn−1 is compact and|| ||is closed, we may assumeP without loss− of generality that n−1 P αk/ αk converge to α S . But then ν, α = limk→∞ ν, αk/ αk = 0, which contradicts the choice|| || of ν. ∈ P ∩ h i h || ||i

Exercise V.6 implies that is defined by inequalities of the form νj , x 0, j =1,...,N, for some n ∗ P ⊥ h i≥ ν1,...,νN (R ) . For each j, let j := νj . By the inductive hypothesis, each j is generated by ∈ n P P ∩ P finite sets Sj R . We claim that is generated by Sj . Indeed, take α 0 . There is r > 0 ⊂ P j ∈P\{ } such that rα ′. Take any straight line through rα on the hyperplane α : ν, α = 1 . Claim V.3.2 S implies that either∈ P rα for some j, or each end of the line intersects{ oneh of thei .} In any event, ∈ Pj Pj the inductive hypothesis implies that α is a nonnegative linear combination of elements from j Sj , as required to complete the proof of ( ) implication of assertion (2). Now we prove ( ) implications of S assertions (1) and (2) by induction⇒ on dim(aff( )). We will first show that it suffices⇐ to prove only the implication from assertion (2). P

α3 C2 α2

α1 C3

FIGURE 4. Convex hull of ﬁnitely many points is a ﬁnite intersection of translations of cones

Claim V.3.3. Let k 1. Assume the ( ) implication of assertion (2) holds whenever dim(aff( )) k. Then the ( ) implication≥ of assertion (⇐1) holds whenever dim(aff( )) k. P ≤ ⇐ P ≤ PROOF. Let be the convex hull of a finite set S, and m := dim(aff( )) k. Without loss of P m P ≤ generality we may assume that R (exercises V.11 and V.12). Let α1,...,αs be the elements of S. For each j, let be the coneP generated ⊂ by S α = α α : 1 i s . The hypothesis of Cj − j { i − j ≤ ≤ } the claim implies that each j is a polyhedral cone, which implies in turn that j + αj is a polyhedron C ′ C (exercise V.12). It follows that := j ( j + αj ) is also a polyhedron. Therefore it suffices to show that ′ P C = (fig. 4). By(25) every element α can be expressed as i ǫiαi where ǫi are nonnegative real P P T ∈P ′ numbers with i ǫi = 1. But then α = αj + i ǫi(αi αj ) j + αj for each j. Therefore . Now take α . To complete the proof of the claim it− suffices∈ C toP show that α ′. Since αP⊆P, it follows that cone(6∈P P α) is strongly convex (exerciseP V.16). By the hypothesis of6∈ the P claim cone(6∈ P α) P − P− is polyhedral, so that due to assertion (3) there is ν (Rn)∗ which is positive on α. Pick j such ∈ P − that ν, αj = min ν, αi : 1 i s . Then ν is nonnegative on j , whereas ν, α αj is negative. Thereforeh iα +{hα , andi consequently≤ ≤ } α ′, as required. C h − i 6∈ Cj j 6∈ P Now we start the proof of ( ) direction of assertion (2) by induction on dim(aff( )). Due to proposition V.1 and exercises V.11 and⇐V.13 we may assume without loss of generality that Pis a strongly convex n P cone generated by a finite set S R and n = dim(aff( )). For n = 1 the possibilities for are R≥0 ⊂ P P and R≤0, both of which are polyhedral. For general n we proceed by induction on S . The case S =1 is also covered by the case n =1. So assume S 2. | | | | | |≥ 2. CHARACTERIZATION OF CONVEX POLYHEDRA 59

Claim V.3.4. There is ν (Rn)∗ which is positive on S 0 . ∈ \{ } PROOF. Pick α S 0 . By the inductive hypothesis 1 := cone(S α ) is polyhedral, so that is defined by finitely∈ many\{ } inequalities of the form ν , x P 0, j =1,...,N\{ (exercise} V.6). We claim P1 h j i≥ that νj , α > 0 for some j. Indeed, otherwise νj , α 0 for each j, so that α 1. But then containsh thei line through the origin and α, contradictingh − thei ≥ strong convexity of .− So we∈ P can pick j suchP ⊥ P that νj , α > 0. Now let 2 := νj . Then dim(aff( 2))

PROOF. Follows immediately from theorem V.3 and exercise V.8. The next corollary of theorem V.3 shows that each convex polyhedron has a representation of the form = + , where is a polyhedral cone and is a polytope. In generalQ the decomposition is not unique,Q seeP ﬁg. 5C. However,C is uniquely determinedP from (exercise V.21). It is also possible to ﬁnd which is “minimal” (moduloC translations in the case that isQ not strongly convex), but we will not get intoP that. C

y y

y (0, 0) x x (0, 0)

P1 C (0, 0) x y y

(0, 0) x x Q (0, 0)

P2 C

FIGURE 5. = + for = 1 and = 2. The “minimal” possible choice for is the line segmentQ P .C P P P P P P2

Corollary V.5 (Motzkin (1936)). A subset of Rn is a convex polyhedron if and only if it is the Minkowski sum of a convex polyhedral cone and a convex polytope. PROOF. At first we prove the ( ) implication. Let be a polyhedron in Rn defined by inequalities a + a x + + a x ⇒ 0, q = 1,...,N.Q We will show that it is the sum of a polyhe- q,0 q,1 1 ··· q,n n ≥ dral cone and a polytope. Consider the polyhedral cone ′ in Rn+1 defined by the “homogenizations” Q ′ n aq,0x0 + aq,1x1 + + aq,nxn 0, q = 1,...,N, and x0 0. Note that = ( 1 R ). Theorem V.3 implies··· that ′ is a finitely≥ generated cone. We may≥ choose a set ofQ generatorsQ ∩ of{ }×′ of the Q n Q form (1, α1),..., (1, αk), (0,β1),..., (0,βl), where α1,...,αk,β1,...,βl R . Let be the convex ∈ P 60 V. CONVEX POLYHEDRA

n hull of α1,...,αk, and be the cone in R generated by β1,...,βl. Theorem V.3 implies that is a polytope and is a polyhedralC cone. P C Claim V.5.1. = + . Q P C PROOF. Let γ . Then (1,γ) ′, and therefore (1,γ)= a (1, α )+ b (0,β ) where the ∈Q ∈Q i i i j j j a ,b are nonnegative numbers. Then it follows that a = 1 and γ = α + β, where α = a α and i j i i P P i i i β = b β . It is clear that β . Since a =1, it follows that α is a convex combination of the α , j j j ∈C i i i so that α . Therefore + . Now we checkP the opposite inclusion. Let α and Pβ . Then P ∈P Q⊂P C P ∈P ′ ∈C α = i aiαi where the ai are nonnegative numbers such that i ai = 1. Since (1, αi) for each i, ′ ∈ Q it follows that (1, α) = i ai(1, αi) . On the other hand, it follows from the construction of that P ′ n ∈ Q ′ P C 0 = ( 0 R ), so that (0,β) . Since the sum of two elements of a convex cone is also {in}×C that cone,Q it follows∩ { }× thatP (1, α + β)=(1,∈Q α)+(0,β) ′. Therefore, α + β , as required. ∈Q ∈Q Claim V.5.1 finishes the proof of ( ) implication of the corollary. Now we prove the ( ) implication. ⇒ ⇐ Let be a convex polytope and be a convex polyhedral cone in Rn. We will show that + is a convex polyhedron.P By exercise V.12 weC may assume without loss of generality that contains theP origin.C Identify P Rn+1 with R Rn. Let ′ be the cone in Rn+1 generated by 1 , and ′ be the closure of the cone in × P { }×P C Rn+1 generated by 1 . Either theorem V.3 or exercise V.18 implies that ′ is a polyhedral cone, and exercise V.20 implies{ that}×C′ is a polyhedral cone. Corollary V.4 then impliesP that ′ + ′ is a polyhedral C P C cone, so that := ( ′ + ′) ( 1 Rn) is a polyhedron. Q P C ∩ { }× Claim V.5.2. = 1 ( + ). Q { }× P C PROOF. Let the inequalities defining be ai,0 + ai,1x1 + + ai,nxn 0, i = 1,...,k, and the inequalities defining be b x + + bP x 0, j =1,...,l···. Exercise V.18≥ shows that ′ is defined C j,1 1 ··· j,n n ≥ P by the inequalities ai,0x0 + ai,1x1 + + ai,nxn 0, i = 1,...,k, and x0 0. On the other hand ′ ··· ≥ ≥ exercise V.20 implies that is defined by the same inequalities as together with the inequality x0 0. If α and β , thenC it follows that (1, α) ′ and (0,β)C ′, so that (1, α + β) ′ +≥ ′, which∈ proves P that∈ C 1 ( + ). For the opposite∈ P inclusion,∈ pick C (1,γ) . We will∈ show P thatC γ + . WriteQ⊃{(1,γ) =}× (a, αP) +C (b,β), where (a, α) ′ and (b,β) ′,∈ and Qa,b are nonnegative numbers∈ P suchC that a + b = 1. An examination of the inequalities∈ P defining ∈C′ Cshows that β for each r 0. If a =0, then assertion (1) of exercise V.18 implies that α =0. Since the origin is in ∈, C it follows that≥ γ =0+ β + . On the other hand, if a = 0, then (1/a)α . Since 0 < a P1, and since 0 , it follows∈ by P convexityC of that α . Therefore6 γ = α + β ∈ P+ , as required.≤ ∈P P ∈P ∈P Q Claim V.5.2 finishes the proof of ( ) implication, and therefore the proof of corollary V.5. ⇐ Corollary V.6. Let be a polyhedron in Rn. P (1) has a decomposition of the form = + + L, where is a polytope, is strongly convex P P P0 C P0 C polyhedral cone, and L is a linear subspace of Rn such that L = 0 . Moreover, L and + L are uniquely determined by . C ∩ { } (2)C If is defined by inequalities a +P a x + + a x 0, i =1,...,N, then P i,0 i,1 1 ··· i,n n ≥ (a) L = x : ai,1x1 + + ai,nxn =0, i =1,...,N . (b) + L{ = x : a x···+ + a x 0, i =1,...,N} . C { i,1 1 ··· i,n n ≥ } (c) For each r 1, the set r := x : ai,1x1 + + ai,nxn = 0, i = 1,...,N, and r x r≫, j =1,...,nP is a polytope,{ and we··· can take = for any such r. − ≤ j ≤ } P0 Pr PROOF. Combine proposition V.1 and corollary V.5. Corollary V.7. The Minkowski sum of finitely many convex polyhedra is a convex polyhedron. PROOF. Follows directly from corollary V.4 and the arguments of the proof of corollary V.5. Corollary V.8. The image of a polyhedron under an affine map is also a polyhedron. PROOF. By exercise V.12 it suffices to consider the case of linear maps. Let be a polyhedron in n n m P R , and φ : R R be a linear map. Write = 0 + + L as in corollary V.6. Then φ( ) = φ( )+ φ( )+ φ→(L) (exercise V.7). Theorem V.3P andP exerciseC V.1 imply that φ( ) is a polytope,P and P0 C P φ( ) is a polyhedral cone. Since φ(L) is a linear subspace of Rm (due to linearity of φ), corollary V.7 impliesC that φ( ) is a polyhedron. P 3. BASIC PROPERTIES OF CONVEX POLYHEDRA 61

2.1. Exercises. n EXERCISE V.18. Let be a nonempty polytope in R defined by inequalities ai,0 + ai,1x1 + + P n ··· ai,nxn 0, i = 1,...,N. Let 0 := x R : ai,1x1 + + ai,nxn 0, 1 i N , and be the ≥ C { ∈ ··· ≥ ≤ ≤ } C cone generated by 1 in Rn+1 = R Rn. { }×P × (1) Show that 0 = 0 . [Hint: 0 is a cone, 0 + , and is bounded.] C { } C Cn+1 P⊂P P (2) Deduce that is a polyhedral cone in R defined by the inequalities x0 0 and ai,0x0 + C ≥ ai,1x1 + + ai,nxn 0, i =1,...,N. (3) Showthat··· theassumption“≥ is nonempty”is necessary in the precedingstatements. In particular, show by an example that Pmay be a nontrivial cone if = . C0 P ∅ EXERCISE V.19. Show that if the ( ) implication of assertion (2) of theorem V.3 holds for all n, then the ( ) direction of assertion (1) of theorem⇒ V.3 also holds for all n. [Hint: use exercise V.18.] ⇒ EXERCISE V.20. Let be a polyhedral cone in Rn. Exercise V.6 implies that is defined by finitely C n+1C many inequalities of the form a1x1 + +anxn 0. Show that the closure in R of the cone generated by 1 is a polyhedral cone defined··· by the≥ same inequalities as the ones defining coupled with x { 0}×C. C 0 ≥ EXERCISE V.21. Let Rn. P ⊂ (1) Assume = 1 + 1 = 2 + 2, where j are bounded and j are closed cones. Prove that = .P ShowP by examplesC P thatC it is possibleP to have = . C C1 C2 P1 6 P2 (2) Assume = 1 + L1 = 2 + L2 where j are strongly convex polyhedral cones and Lj are P C n C C linear subspaces of R such that j Lj = 0 . Prove that L1 = L2. Show by examples that it is possible to have = . C ∩ { } C1 6 C2 (3) Assume = 1 + 1 + L1 = 2 + 2 + L2 where j are polytopes, j are strongly convex P P C P C n P C polyhedral cones and Lj are linear subspaces of R such that j Lj = 0 . Prove that L1 = L2 and + L = + L . C ∩ { } C1 1 C2 2 [Hint: we may assume without loss of generality that contains origin. Then for the first assertion pick α such that rα for each r 0. Express α asP a sum of elements of 1 and , and then take ∈ P ∈ P ≥ r Pj Cj the limit as r to show that each j must be the largest cone contained in . For the second assertion → ∞ C n P n prove that each Lj is the largest linear subspace of R contained in as follows: pick α R such that P ∈ both α and α are in . Express both α and α as elements of j +Lj, add them up, and use the condition that L−= 0 toP show that α L .] − C Cj ∩ j { } ∈ j 3. Basic properties of convex polyhedra In this section we establish a few basic properties of convex polyhedra. Throughout this section P will denote a nonempty convex polyhedron in Rn. The first property we state follows directly from the definition of a polyhedron: n n ∗ Proposition V.9. For each α R , there is ν (R ) such that minP (ν) exists and ν, α < ∈ \ P ∈ h i minP (ν). Geometrically, proposition V.9 states that every point in the complement of is separated from by P n ∗ P a hyperplane - see fig. 6.A face of is a subset of the form Inν ( ) for some ν (R ) . P P ∈ Proposition V.10. Every face of is a convex polyhedron. If is a polytope (respectively polyhedral cone), then every face of is alsoP a polytope (respectively, polyhedralP cone). P PROOF. The case of a convex polyhedron and that of a polytope are direct consequences of the definitions of polyhedra, polytopes and faces. The case of a polyhedral cone follows from combining exercises V.6 and V.23. We also note the following property whose proof is left as an exercise: Proposition V.11. Let φ : be an affine isomorphism of convex polyhedra. Then φ induces a bijection of faces, i.e. Pis →Q a face of if and only if φ( ) is a face of . R⊆P P R Q PROOF. This is exercise V.24. 62 V. CONVEX POLYHEDRA

α C L

ν η

P B

FIGURE 6. In ( )= C , In ( )= BC. The line L separates from α. ν P { } η P P

Proposition V.10 implies that every face of is a polyhedron, and therefore has a well-defined dimension. A vertex (respectively, edge, facet) is aP face of dimension zero (respectively one, dim( ) 1). In fig. 6 the vertices of are A, B, C, and the edges are the three sides. Proposition V.12 belowP gives− a more precise descriptionP of vertices and edges. The part of the proposition about finiteness of numbers of vertices and edges is a special case of the more general result (corollary V.17) that a polyhedron can have only finitely many faces. We will also see later that every polytope is the convex hull of its vertices (as it is evident for from fig. 6). P Proposition V.12. Every vertex of a polyhedron is a singleton, and up to an affine isomorphism, every edge is a (possibly unbounded) closed interval of R. Every polytope has finitely many vertices. Every strongly convex polyhedral cone has only one vertex (namely, the origin) and finitely many edges.

PROOF. The first statement follows directly from exercise V.4. By theorem V.3 every polytope is the convex hull of a finite set. The first statement together with exercise V.2 then implies that has onlyP finitely many vertices. Now let be a strongly convex polyhedral cone. Theorem V.3 implies thatP the origin is a vertex of . Since everyfaceC of is also a cone (proposition V.10), and since the only zero-dimensional cone is the origin,C it follows that theC origin is the only vertex of . The finiteness of the numbers of edges of follows from exercise V.26 and finiteness of the numbers ofC vertices of polytopes. C Note in fig. 6 every pair of edges of intersects in a vertex. The following result shows that this corresponds to a general phenomenon: P Proposition V.13. The intersection of finitely many faces of is also a face of , provided the intersection is nonempty. P P

PROOF. Let := Inν ( ) and := Inη( ) be faces of with = . It is then straightforward to check that Q = In P( ). [WhereR is itP used that P = ?Q See ∩ R exercise 6 ∅ V.27.] Q ∩ R ν+η P Q ∩ R 6 ∅ Every polyhedron is a face of itself, since = In0( ). A proper face of is a face which is properly contained in P, and the relative interior relint(P ) of P is the complement inP of the union of its proper faces. In fig. 6 Pthe relative interior is the complementP inP of the union of the threeP edges. P Proposition V.14. Let , be faces of . Q R P (1) If is a proper face of , then dim( ) < dim( ). (2) If Q relint( ) = , thenP . Q P Q ∩ R 6 ∅ Q ⊃ R PROOF. The first assertion is an immediate consequence of the observation that if ν (Rn)∗ is non- ∈ constant on an affine subspace H of Rn, then for every r R, α H : ν, α = r is an affine subspace ∈ { ∈ h i } of Rn whose dimension is one less than that of H. For the second assertion, note that if = , then by proposition V.13 is a face of , and it can not be a proper face of if it containsQ ∩ aR point 6 ∅ from relint( ). Q ∩ R R R R Note that in fig. 6 the relative interior of is also its topological interior2 in R2. We will now see that this is a manifestation of a general property ofP polyhedra.

2 A point x ∈ S ⊆ Rn is in the topological interior of S if and only if S contains an open neighborhood of x in Rn. 3. BASIC PROPERTIES OF CONVEX POLYHEDRA 63

Proposition V.15. is the closure of its relative interior. In particular, relint( ) is nonempty, and it is precisely the topologicalP interior of in aff( ) (where aff( ) is equipped with theP relative topology from P P P Rn). In particular, relint( ) is a nonempty relatively open subset of aff( ). In the case that Rn is “full dimensional” (i.e. dim(P ) = n), and a + a x + + a x P 0, i = 1,...,N, areP ⊂ a set of P i,0 i,1 1 ··· i,n n ≥ “nontrivial inequalities”3 defining , then relint( ) is the nonempty open set of Rn defined by the strict inequalities a + a x + + aP x > 0, i =1P,...,N. i,0 i,1 1 ··· i,n n PROOF. Due to exercises V.11, V.12 and V.24 it suffices to consider the case that is full dimensional. 0 P n For each i, let Hi be the hyperplane ai,0 +ai,1x1 + +ai,nxn =0. Let := x R : ai,0 +ai,1x1 + ··· P 0 { ∈ + ai,nxn > 0, i = 1,...,N = i Hi. Exercise V.25 implies that is the topological interior ··· 0 } P\ P of . We claim that = . Indeed, for each i = 1,...,N, there is αi such that ai,0 + ai,1x1 + P P 6 ∅ S ∈ P + ai,nxn > 0, since otherwise would be contained in Hi, and dim( ) would be less than n. Since ··· N P P 0 is convex, α := (1/N) i=1 αi . It is straightforward then to check that α , as required. P 0 ∈ P ∈ P We now show that = relint( ). The full-dimensionality of also implies that Hi is a proper face of for each iP, so thatP 0 Prelint( ). On the other hand,P if α 0, then by openness∩P of 0, for P n ∗ P ⊃′ 0 P ′ ∈ P P every ν (R ) 0 , there is α such that ν, α < ν, α , so that α Inν ( ). This shows that 0 relint(∈ ), and\{ therefore} 0 =∈ relint( P ). Finally,h toi seeh thati is the closure6∈ ofP its relative interior, P ⊂ P P P P let β := (β1,...,βn) relint( ). Given γ = (γ1,...,γn) , let L := (1 ǫ)β + ǫγ : 0 ǫ β be the line segment from∈ β to γ.P We claim that L γ relint(∈ P ). Indeed,{ for− each i =1,...,N≤ ,≤ let } \{ }⊆ P ǫ := sup ǫ :0 ǫ 1, a + a ((1 ǫ)β + ǫγ )+ + a ((1 ǫ)β + ǫγ ) > 0 i { ≤ ≤ i,0 i,1 − 1 1 ··· i,n − n n } Note that each of the functions a + a ((1 ǫ)β + ǫγ )+ + a ((1 ǫ)β + ǫγ ) is linear in i,0 i,1 − 1 1 ··· i,n − n n ǫ, and it is strictly positive at ǫ = 0. Therefore if ǫi < 1, then it must be zero at ǫi and negative on the interval (ǫi, 1], contradicting the fact that L . Therefore ǫi = 1 for each i, and consequently, L γ 0 = relint( ). This implies that is⊂ the P closure of relint( ), as required. \{ }⊂P P P P n n ∗ Proposition V.16. Let 1,..., s be polyhedra in R , and ν (R ) . Then P P ∈ (1) min (ν) exists if and only if min (ν) exists for each j. Pj Pj Pj Now assume exists. Then minPj Pj (ν) (2) min (ν)= min (ν). Pj Pj j Pj (3) In ( )= In ( ). ν j Pj Pj ν Pj (4) In ( ) are the “unique maximal” subsets of whose sum is In ( ), i.e. if α are ν Pj P Pj ν j Pj j ∈ Pj such that α In ( ), then α In ( ) for each j. j j ∈ ν j Pj j ∈ ν Pj P (5) If each is a polytope, then In ( ) are in fact unique faces of whose sum is In ( ), PjP P ν Pj Pj ν j Pj i.e. if are faces of such that = In ( ), then = In ( ) for each j. Qj Pj j Qj ν j Pj Qj ν Pj P (6) Assume there is j such that P P (a) either j is a cone, or P n (b) j is a linear subspace of R . P ⊥ Then Inν ( j ) contains the origin and minPj (ν)=0. In addition in case (6b), j ν , and In ( )=P . P ⊂ ν Pj Pj

PROOF. It sufﬁces to treat the case s = 2. Assume at ﬁrst minP1 (ν) does not exist. Then there are α such that ν, α . If β is an arbitrary element in , then ν, α + β as well, so k ∈P1 h ki→−∞ P2 h k i→−∞ that minP P (ν) does not exist. On the other hand, if minPj (ν) exists for each j, then pick αj Inν ( j ) j j ∈ P for each j, and note that for all β , ν, β ν, α = ν, α . This simultaneously j ∈ Pj h j j i ≥ j h j i h j j i proves assertions (1) to (3). For assertion (4) note that if α In ( ), then ν, α > min (ν). It P P 1 6∈ ν P1 P h 1i P1 follows that for each α2 2, ν, α1 + α2 > minP1 (ν) + minP2 (ν), so that α1 + α2 Inν ( 1 + 2). ∈P h i n ∗ 6∈ P P For assertion (5) assume 1 ( Inν ( 1). Since j is a polytope, there is η (R ) such that minQ1 (η) > min (η). On the otherQ hand, InP ( ) is aQ polytope for each j, and therefore∈ by assertion (1), Inν (P1) ν Pj min (η) = min(η) + min (η) > min (η) + min (η) = min (η) Q1+Inν (P2) Q1 Inν (P2) Inν (P1) Inν (P2) Inν (P1)+Inν (P2)

3 “Nontrivial inequalities” means that we do not allows inequalities with ai,0 = ··· = ai,n = 0. 64 V. CONVEX POLYHEDRA

It follows that 1 + Inν ( 2) = Inν ( 1)(η)+Inν ( 2), as required. Finally, assertion (6) follows directly from exercise V.23Q . P 6 P P The example from ﬁg. 5 shows that assertion (5) of proposition V.16 may not be true if some of the j are not bounded [what goes wrong with the proof?]. On the other hand, if j are polytopes, then it is notP too hard to show that it remains true even if are allowed to be arbitraryP convex subsets of , i.e. Qj Pj Inν ( j ) are unique convex subsets of j whose sum is Inν ( j j ). Exercise V.28 asks you to show that convexityP is necessary for uniqueness.P P P Corollary V.17. Every polyhedron has ﬁnitely many distinct faces.

n ∗ n PROOF. Let ν (R ) . If 0 = conv(S) for some finite set S R , exercise V.29 implies that In ( ) = conv(In∈(S)); in particular,P every face of the polytope ⊂is the convex hull of a subset of ν P0 ν P0 S, and therefore 0 has finitely many distinct faces. Now consider the case that = cone(T ) for a finite n P C ⊥ subset T of R . If minC (ν) exists, then proposition V.16 implies that minC(ν)=0 and Inν ( )= ν = cone(ν⊥ T ). Therefore, the number of distinct faces of is bounded by the number ofC subsets∩C of T . By corollary∩ V.6 every polyhedron be can be expressed asC + + L, where is a polytope, is P P0 C P0 C a strongly convex polyhedral cone, and L is a linear subspace of Rn. Proposition V.16 then implies that ′ ′ ′ ′ every face of is of the form 0 + + L, where 0 (respectively, ) is a face of 0 (respectively, ). The result thenP follows from theP casesC of polytopes andP polyhedral cones.C P C

FIGURE 7. ABC is the minimal face containingedges AC and BC. The relative interior of the line segment from a point in relint(AC) to a point in relint(BC) is contained in relint(ABC).

As a corollary of the finiteness of the number of faces, we now prove a more technical result that we use in section V.4. It shows that given two faces , of a polyhedron , there is a “smallest” face of Q1 Q2 P Q that contains both j, and given two points in the relative interiors of the j , the relative interior of the lineP segment joiningQ these points is contained in the relative interior of , seeQ fig. 7. Q Corollary V.18. Let 1, 2 be faces of a convex polyhedron . Q Q P ′ ′ (1) There is a (unique) face of such that 1 2 , and for every face of such that ′.Q P Q ∪Q ⊂ Q Q⊂Q Q P Q1 ∪Q2 ⊂Q (2) Pick αi relint( i), i = 1, 2. Then for every ǫ (0, 1), the positive convex combination ǫα + (1∈ ǫ)α ofQα and α is in the relative interior∈ of . 1 − 2 1 2 Q PROOF. Let be the intersection of all faces of containing both j . Proposition V.13 and corollary V.17 imply thatQ is a face of , and it is clearly asP in the first assertion.Q The second assertion follows from exercise V.29 andQ the definitionP of relative interior. The following proposition shows that the property of being a face is transitive. Note that this is clear in figs. 6 and 7: every vertex of every edge of is also a vertex of . P P Proposition V.19. Every face of a face of is also a face of . P P 3. BASIC PROPERTIES OF CONVEX POLYHEDRA 65

n ∗ PROOF. Let := Inν ( ) and := Inη( ), ν, η (R ) . Let = 0 + + L be a decomposition as in corollary V.6Q. PropositionP V.16Rthen impliesQ that ∈ P P C

(i) = Inν ( 0)+Inν ( )+ L, where ν L 0 and minC(ν)=0; (ii) Q = In (InP ( )) +C In (In ( )) + L| , where≡ η 0, and min (η)=0. R η ν P0 η ν C |L ≡ Inν (C) Due to proposition V.11 after a translation of if necessary we may further assume that Inν ( 0) contains the origin, which implies that P P

(iii) minP0 (ν)=0, and (iv) Inν ( 0) Inν ( ) L. Q⊃ P ∪ C ∪ n By theorem V.3 there are finite sets S,T R such that 0 = conv(S) and = cone(T ). By observations (i) and (iii), ν is nonnegative on S T , and⊂ therefore weP may choose r 0Csuch that rν + η, α > 0 for each α (S T ) ν⊥. ∪ ≥ h i ∈ ∪ \ Claim V.19.1. = In ( ). R ν+rη P PROOF. This is left as exercise V.33. The proposition follows immediately from claim V.19.1. Proposition V.20. Let be a polytope and be the set of vertices of . P V P (1) = conv( ). (2) P is the uniqueV minimal set whose convex hull is . (3) Vrelint( ) is the set of positive convex combinationsP of its vertices, i.e. α relint( ) if and only P ∈ P if α = β∈V ǫββ, where ǫβ are positive real numbers such that β∈V ǫβ =1. PROOF. ForP the first assertion proceed by induction on dim( ). It is evidentP when dim( )=0. Now consider the case that dim( ) 1. Take α . If α relint(P ), then it is on a properP face of . Since dim( ) < dim( ), byP induction≥ α is in∈ the P convex6∈ hull ofP vertices of . Proposition V.19QimpliesP that every vertexQ of isP also a vertex of , so that α conv( ), as required.Q If α relint( ), then take a line L through α onQ aff( ). Since isP bounded, proposition∈ VV.15 implies that each∈ end ofPL intersects a proper face of . SinceP we alreadyP showed that every proper face of is in conv( ), it follows that α conv( ) andP completes the proof of the first assertion. The second assPertion and theP ( ) implication of∈ the thirdV assertion follow from the first assertion and exercise V.29. Now pick α relint(⇐ ). It remains to show that α is a positive convex combination of the vertices of . By proposition∈ V.12 weP may assume dim( ) 1. Since relint( ) is relatively open in aff( ) (propositionP V.15), α ǫ β relint( ) P ≥ P P − β∈V ∈ P for sufficiently small positive ǫ. By the first assertion then α ǫ β conv( ). It then follows that − β∈V ∈ VP α is a positive convex combination of elements from , as required. V P Corollary V.21. If is a strongly convex polyhedral cone, then C (3) is the cone generated by its edges. (4) relint(C ) is the set of positive linear combinations of nonzero elements of its edges, i.e. if is the set consistingC of one nonzero element from each of the edges of , then α relint( ) if andE only C ∈ C if α = β∈E rββ, where rβ are positive real numbers. PROOF. CombineP proposition V.20 and exercise V.26.

xk+2

η2 xk+1

η1

FIGURE 8. cone(π(S T )) is a two dimensional cone contained in the half-plane x 0, and intersects∪x -axis only at the origin. k+1 ≥ k+2 66 V. CONVEX POLYHEDRA

Proposition V.22. Every proper face of a polyhedron is the intersection of facets of containing . In particular, every proper face is containedQ in a facet. P P Q

PROOF. Let = 0 + + L be a decomposition of as in corollary V.6. Let = Inν ( ), n ∗ P P C P Q P ν (R ) , be a proper face of . Proposition V.16 implies that = Inν ( 0)+Inν ( )+ L, where ν ∈ 0 and min (ν)=0. DueP to proposition V.11 after an affineQ isomorphismP of ifC necessary we |L ≡ C P may further assume that dim( ) = n, and aff( ) is the coordinate subspace of Rn spanned by the first k-coordinates, where k := dim(P ) < n, andQν is the projection on to the (k + 1)-th coordinate. If n = k +1, then is a facet, andQ we are done. So assume n k 2. Choose finite sets S,T such Q − ≥ n that 0 = conv(S) and = cone(T ). Let π be the projection map from R onto the 2-dimensional coordinateP subspace spannedC by the (k + 1)-th and (k + 2)-th coordinate. By construction every nonzero element of π(S T ) has positive xk+1-coordinate. We claim that the cone generated by π(S T ) is two dimensional, i.e.∪ it is as in fig. 8. Indeed, by construction π(L)=0, so that exercises V.1 and∪V.7 imply that π( )= π( 0)+π( ) = conv(π(S))+cone(π(T )) cone(π(S T )). Therefore, if cone(π(S T )) is containedP in aP line L,C and would be contained in π−⊆1(L), contradicting∪ the full-dimensionality∪ of . It follows that dim(cone(π(SP T ))) = 2, and cone(π(S T )) has two edges (exercise V.5). As in fig.P8, ∪2 ∗ ∪ let η1, η2 be the elements in (R ) which attains their minima on the two edges of cone(π(S T )). It is straightforward to check that ∪

∗ for each j, j := Inπ (ηj )( ) is a face of such that j and dim( j ) > dim( ), • If n = k +2R, then each Pis a facet of ,P and = R ⊃Q. R Q • Rj P Q R1 ∩ R2 The proposition follows from these observations by a straightforward induction on dim( ) dim( ). P − Q Corollary V.23. Let be an n-dimensional polyhedron in Rn. Assume = Rn. Then is determined by the affine hyperplanesP corresponding to facets of . More precisely, P 6 P P (1) up to multiplications by nonzero real numbers, there is a unique minimal set of inequalities determining . P Let a + a x + + a x 0, i =1,...,N, be the minimal set of inequalities defining . Then i,0 i,1 1 ··· i,n n ≥ P (2) x : a + a x + + a x =0 is a facet of for each i; P∩{ i,0 i,1 1 ··· i,n n } P (3) for every face of , there is I 1,...,N such that = x : ai,0 + ai,1x1 + + a x =0 forQ eachPi I . ⊆ { } Q P∩{ ··· i,n n ∈ } PROOF. Let a + a x + + a x 0, i =1,...,N, be a minimal set of inequalities defining i,0 i,1 1 ··· i,n n ≥ . It suffices to show that i := x : ai,0 +ai,1x1 + +ai,nxn =0 , i =1,...,N, are the facets of P. Indeed, if is a facet ofQ , thenP∩{ proposition V.15 implies··· that } . Exercise V.31 and assertion P Q P Q⊂ i Qi (1) of proposition V.14 then imply that = i for some i. It remains to show that every i is a facet. Q Q S ′ Q Indeed, reorder the inequalities so that i are facets for i =1,...,M. Let be the polytope determined Q P′ by ai,0 + ai,1x1 + + ai,nxn 0, i =1,...,M. If M 0. • ∈ i,0 i,1 1 ··· i,n n L contains a point γ on the topological boundary4 of . • P But then proposition V.14 implies that γ is a point of a face of which is not on any facet of , contradicting proposition V.22. P P

The theory of toric varieties (to be discussed in chapter VI) exploits many similarities between polytopes and algebraic varieties. Here we note some of the more obvious analogues between a polyhedron and an irreducible variety X. P

4 Topological boundary of S ⊂ Rn is the complement in S of the topological interior of S. 3. BASIC PROPERTIES OF CONVEX POLYHEDRA 67

The dimension of every proper irre- The dimension of every proper ducible subvariety of X is less than face of is less than dim( ) dim(X) (theorem III.19). ↔ (propositionP V.14). P

If an irreducible subvariety of X is If a face of is contained in the contained in the union of ﬁnitely many union of ﬁnitelyP many faces of subvarieties of X, then it is contained ↔ , then it is contained in one of in one of them (exercise III.17). themP (exercise V.31). 3.1. Exercises.

EXERCISE V.22. Let be a face of . Show that = aﬀ( ). [Hint: if = Inν ( ), then ν, is constant on aﬀ( ).] Q P Q P ∩ Q Q P h ·i Q EXERCISE V.23. Let be a convex polyhedron and ν (Rn)∗. Assume inf ν, α : α > , and P ∈ {h i ∈ P} −∞ (1) either is a cone, or P (2) is a linear subspace of Rn. P Then In ( ) contains the origin and min (ν)=0. In addition, in case (2) ν⊥, and In ( ) = . ν P P P ⊂ ν P P [Hint: in either case contains the origin, so that minP (ν) 0. If there is α such that ν, α < 0, then ν,rα approachesP as r .] ≤ ∈ P h i h i −∞ → ∞ EXERCISE V.24. Prove proposition V.11. [Hint: use assertion (1) of exercise V.14.] EXERCISE V.25. Let S Rn and ν be a nonzero element in (Rn)∗ such that m := inf ν, α : α ⊂ {h i ∈ S > . Show that no point of the hyperplane x Rn : ν, α = m is in the topological interior of } −∞ { ∈ h i } S in Rn. EXERCISE V.26. Let be a positive dimensional strongly convex polyhedral cone. By theorem V.3 C there is ν (Rn)∗ such that ν is positive on 0 . Let := α : ν, α =1 . Claim V.3.2 implies that is a∈ polytope. C\{ } P { ∈P h i } P (1) If is a face of , show that cone( ) is a face of of dimension dim( )+1. (2) ShowQ that the correspondenceP Qcone( ) inducesC a bijection betweenQ faces of and positive dimensional faces of , Q 7→ Q P (3) Show that the above correspondenceC also preserves the relative interiors of the faces, i.e. If is a face of , then relint(cone( )) = cone(relint( )). Q P Q Q [Hint: change coordinates on such that η is the projection onto the last coordinate.]

EXERCISE V.27. Let be a polyhedron, and = Inν ( ) and = Inη( ) be faces of . If = , show by an exampleP that it may not be trueQ that In P ( )=R . P P Q ∩ R ∅ ν+η P Q ∩ R n EXERCISE V.28. Find examples of polytopes 1, 2 in R and faces j of j such that 1 + 2 is ′ P ′P ′ Q P Q Q a face of 1 + 2, and there are j ( j such that 1 + j = 1 + 2. [Hint: there are examples with n =1 withP P= for each j.]Q Q Q Q Q Q Qj Pj k n EXERCISE V.29. Let α = j=1 ǫjαj be a convex combination of α1,...,αk R . Assume each ǫj is positive. ∈ n ∗ P (1) If ν (R ) , then show that ν, α minj ν, αj , with equality if and only if ν, α1 = = ν, α∈ . h i≥ h i h i ··· h ki (2) Conclude that if the αj are points of a polyhedron , and is a face of , then contains α if and only if contains each α . P Q P Q Q j EXERCISE V.30. Let S be a ﬁnite subset of Rn. Let := conv(S) and be the cone generated by S. Show that P C (1) α relint( ) if and only if α = ǫ β, where ǫ are positive real numbers such that ∈ P β∈S β β ǫ =1. β∈S β P (2) α relint( ) if and only if α = r β, where r are positive real numbers. P∈ C β∈S β β P 68 V. CONVEX POLYHEDRA

[Hint: follow the arguments from the proof of assertion (3) of proposition V.20.]

EXERCISE V.31. Let 1,..., k be faces of a polyhedron . If is a convex subset (e.g. a face) of such that Q, then showQ that for some Pj. [Hint:Q otherwise for each j, there is P Q ⊆ j Qj Q⊆Qj α . Apply exercise V.29 to a positive convex combination of the α .] j ∈Q\Qj S j ′ EXERCISE V.32. Given distinct vertices α, α of a polytope , show that there are vertices β0 = ′ P α, β2,...,βk = α of such that there is an edge of connecting βj−1 to βj for each j = 1,...,k; in other words, the “edge-graph”P of is connected. [Hint:P reduce to the case that is full dimensional. Pick ′ n ∗ P ′ P ν,ν (R ) such that Inν ( )= α and Inν′ ( )= α . Consider ǫ := Inν+ǫν′ ( ) for ǫ 0. As ǫ ∈ P { } P′ { } P P ≥ goes from 0 to , ǫ transitions from α to α in finitely many steps. Apply induction on dimension to each .] ∞ P { } { } Pǫ 4. Normal fan of a convex polytope A fan5 in Rn is a collection Σ of convex polyhedral cones in Rn such that (1) Each face of a cone in Σ is also a cone in Σ. (2) The intersection of any two cones in Σ is a face of each of them. Any (finite dimensional) vector space V over real numbers can be identified with Rn after choosing a basis , and thus the notions of convex polyhedra, cones, polytopes, fans, etc., can be extended to subsets of V . B In this section we take V = (Rn)∗ and to be the basis dual to the standard basis of Rn. Exercise V.1 B implies that convex polyhedra (and therefore cones, polytopes, fans, etc) in (Rn)∗ which are defined in this way remain so after linear changes of coordinates on Rn. Let be a convex polytope in Rn. For each face of , define P Q P n ∗ n ∗ (28) σQ := ν (R ) : Inν ( ) (R ) { ∈ P ⊇Q}⊂ n ∗ It is straightforward to see that σQ is a convex polyhedral cone in (R ) (exercise V.35). Let ΣP := σQ : n ∗ { is a face of . We will show in corollary V.27 below that ΣP is a fan in (R ) ; this is called the normal Q P} 0 n ∗ fan of . Given a face of , let σQ := ν (R ) : Inν ( )= σQ. We show in corollary V.26 P 0 Q P { ∈ P Q} ⊂ below that σQ is the relative interior of σQ.

σ0 C AB

σ0 B 0 A P σB

0 σC A 0 0 σBC σCA

2 0 FIGURE 9. Normal fan of a triangle in R . σP = σ is the origin. P P

Proposition V.24. Let ,..., be the faces of containing . Then σ0 = σ σ . R1 Rk P Q Q Q \ j Rj PROOF. This follows directly from the definition of a face and the definition of σQS. n ⊥ Let LQ be the linear subspace of R spanned by all vectors of the form α β with α, β and LQ n ∗ n ∗ − ∈Q be the linear subspace of (R ) consisting of all ν (R ) such that ν LQ 0. ∈ | ≡ Proposition V.25. L⊥ = aff(σ ) =aff(σ0 ). In particular, dim(σ )= n dim( ). Q Q Q Q − Q 5Our definition of a fan differs from the definition in standard texts on toric varieties (e.g.[Ful93, CLS11]) in that we do not require the cones in a fan to be strongly convex. 4. NORMAL FAN OF A CONVEX POLYTOPE 69

′ n ∗ ′ ′ 0 PROOF. Since is a face of , there is ν (R ) such that = minP (ν ). It follows that ν σQ; 0 Q P ∈ Q ∈ in particular σQ is nonempty. Claim V.25.1. Let ν σ0 . Then for each µ L⊥, ν + ǫµ σ0 for all sufﬁciently small positive ǫ. ∈ Q ∈ Q ∈ Q

PROOF. Indeed, let m := minP (ν) and pick ǫ such that for all vertices α of not on , ǫ µ, α < ν, α m. P Q h− i h i− 0 ⊥ 0 Claim V.25.1 implies that aff(σQ) LQ. On the other hand, if ν σQ, then for each α, β , ⊥ ⊃ 0 ⊥ ∈ 0 ∈ Q ν, α β = 0, so that ν LQ. It follows that aff(σQ) = LQ. Since σQ is the union of the σR over all h − i ∈ ⊥ ⊥ ⊥ faces of containing , and since LR LQ whenever contains , it follows that aff(σQ)= LQ as well, asR required.P Q ⊂ R Q

Corollary V.26. Let be a face of . Q P ⊥ (1) If is a face of containing , then σR = LR σQ. In particular, σR is a face of σQ. 0R P Q ∩ (2) σQ is the relative interior of σQ. (3) Every face of σ is of the form σ for some face of containing . Q R R P Q PROOF. In this proof we will identify Rn with ((Rn)∗)∗ in the usual way by treating α Rn as the n ∗ ⊥ ∈ linear function on (R ) given by ν ν, α . Proposition V.25 implies that σR L σQ. Now pick 7→ h i ⊂ R ∩ ν L⊥ σ . Since ν L⊥ , it follows that ν is constant on . On the other hand, since ν σ , it ∈ R ∩ Q ∈ R R ∈ Q follows that minP (ν) is achieved on . Since , it follows that minP (ν) is achieved on all of , and ν σ . Therefore σ = L⊥ σQ. For assertionQ ⊂ R (1) it remains to show that σ is a face of σ . R ∈ R R R ∩ Q R Q

Claim V.26.1. Let α and β . Then β α induces a nonnegative linear function on σQ. In particular, min (β ∈α)=0 Q . ∈ R − σQ − PROOF. If ν,β α < 0 for some ν σ , then α In ( ), which contradicts the deﬁnition of h − i ∈ Q 6∈ ν P σQ.

Choose a basis of L of the form β α, j =1,..., dim(L ), where α , and each β . Then R j − R ∈Q j ∈ R claim V.26.1 implies that σ = L⊥ σ = (σ (β α)⊥)= (In (σ )) is an intersection R R ∩ Q j Q ∩ j − j βj −α Q of ﬁnitely many faces of σQ, and therefore is also a face of σQ. This completes the proof of assertion T 0 T (1), and due to proposition V.24 also implies that σQ, being the complement of a union of faces of σQ, contains relint(σQ). On the other hand, claim V.25.1 implies that for each ν σQ, one can ﬁt inside ∈ 0 σQ a dim(σQ)-dimensional ball Bν centered at ν. Proposition V.15 then implies that σQ relint(σQ). 0 ⊂ Consequently, σQ = relint(σQ), which proves assertion (2). By assertions (1) and(2), and corollary V.17 and exercise V.31, every proper face of σQ is a face of σR for some face of properly containing . Assertion (3) therefore follows from a straightforward induction on dim( R) dim(P ). Q P − Q n ∗ Corollary V.27. ΣP is a fan in (R ) .

PROOF. Corollary V.26 shows that ΣP satisﬁes property (1) of a fan. Let 1 and 2 be faces of , and let from corollary V.18 be the smallest face of containing both . Itis clearthatQ Qσ σ σ P. Q P Qj Q ⊂ Q1 ∩ Q2 On the other hand, since is contained in every face of containing both j, it follows that for every ν σ σ , In ( ) Q , so that ν σ . ThereforePσ σ = σ ,Q which is a face of each σ ∈ Q1 ∩ Q2 ν P ⊃ Q ∈ Q Q1 ∩ Q2 Q Qj (corollary V.26). This shows ΣP satisﬁes property (2) of a fan as well.

Recall that in our deﬁnition the cones in a fan do not have to be strictly convex. They do however turn out to be so in an important case.

n Proposition V.28. If is an n-dimensional convex polytope in R , then each cone of ΣP is strictly convex. P n ∗ PROOF. Indeed,if dim( )= n, then for every ν (R ) 0 , Inν ( ) is nonempty. If ν σQ for some face of , it thenP follows from (28) that In∈ ( ) \{=} P\, i.e. νP σ , as required. ∈ Q P −ν P ∩Q ∅ − 6∈ Q 70 V. CONVEX POLYHEDRA

4.1. Exercises. EXERCISE V.33. Prove claim V.19.1. [Hint: every α can be expressed as ǫ α+ r β+ ∈P α∈S α β∈T β γ, where γ L, rβ are nonnegativereal numbers, and ǫα are nonnegativereal numberssuch that α∈S ǫα = ∈′ ′′ ′ P ′′ P 1. Define α := ⊥ ǫαα, α := ⊥ ǫαα, β := ⊥ rββ and β := ⊥ rββ. α∈S∩ν α∈S\ν β∈T ∩ν β∈PT \ν Examine the value of rν + η on each of α′, α′′,β′,β′′. Show that rν + η, α min (η), with P P P InPν (P0) equality if and only if α′′ = β′′ =0, α′ In (In ( )) and β′ In h(In ( )).]i ≥ ∈ η ν P0 ∈ η ν C EXERCISE V.34. Let be an n-dimensional polyhedron in Rn, be a facet of , and α relint( ). Pn ∗ n Q P ∈ Q Let = Inν ( ), ν (R ) . If β is an arbitrary element of R such that ν,β 0, then show that rα +Qβ r forP all r∈ 1. [Hint: use corollary V.23 to show that α + ǫβ h if ǫiis ≥ a sufficiently small positive∈ number.]P ≫ ∈P EXERCISE V.35. Let be a face of a polytope . Show that σ defined in section V.4 is a convex Q P Q polyhedral cone in (Rn)∗. [Hint: express and as convex hulls of finite sets. Show that the condition P Q defining σQ can be expressed in terms of finitely many linear inequalities as in exercise V.6.] 5. Rational polyhedra We say that a convex polyhedron is rational if it can be defined in Rn by a finitely many inequalities P of the form a0 + a1x1 + + anxn 0 where each ai is a rational number (or equivalently, an integer). In ··· ≥ n particular, a rational affine hyperplane is the set ofzeroes in R of an equation a0 +a1x1 + +anxn =0, ··· where each ai Q, and (a1,...,an) = (0,..., 0). Proposition V.29 below gives a characterization of rational polyhedra.∈ In particular, it implies6 that a polytope is rational if and only if its vertices have rational coordinates, and a polyhedral cone is rational if and only if each of its edges have a nonzero element with rational coordinates. We say that a polytope in Rn is integral if all its vertices are in Zn. Proposition V.29. Let Rn. P ⊂ (1) is a rational polytope if and only if it is the convex hull of finitely many points with rational coordinates.P (2) is a rational polyhedral cone if and only if it is the cone generated by finitely many points with Prational coordinates. (3) is a rational polyhedron if and only if it is the Minkowski sum of a rational convex polyhedral coneP and a rational convex polytope. PROOF. At first assume = conv(S) + cone(T ) for S,T Qn. We claim that is a rational polyhedron. Due to exercise V.36Pthe claim reduces to the case that dim(aff(⊂ )) = n. By theP results from sec- P tion V.3, is an n-dimensional polyhedron in Rn, andif is a facet of , then = conv(S′)+ cone(T ′) for S′ PS and T ′ T . It is then straightforward to seeQ that aff( ) isP a rationalQ affine hyperplane. Since ⊂ ⊂ Q an n-dimensional polyhedron in Rn is determined by the hyperplanes containing its facets (corollary V.23), it follows that is a rational polyhedron, as required, and proves the ( ) implications of all three assertions of propositionP V.29. We now prove the ( ) implications of all three⇐ assertions simultaneously by induction on dim( ). Due to exercise V.36 we⇒ may assume without loss of generality that dim( ) = n. The cases of dim( P)=0 and dim( )=1 are then obvious. Now assume dim( ) 2. Since theP vertices and edges of areP also rational polyhedraP (corollary V.23), the inductive hypothesisP ≥ and proposition V.20 P and corollary V.21 imply that if is a polytope, then = conv(S) for a finite S Qn, and if is a P P ⊂ P strongly convex polyhedral cone, then = cone(T ) for a finite T Qn. In general, corollary V.6 implies that = + , where is a rationalP convex polytope and is⊂ a rational convex polyhedron. Due to P P0 C P0 C what we already proved, it suffices to show that = cone(T ) with finite T Qn, even if is not strongly convex. Indeed, if is not strongly convex, thenC proposition V.1 and corollary⊂ V.6 implyC that = ′ + L C C C where ′ is a cone and L is a positive dimensional rational linear subspace of Rn. Let k := dim(L). Then C we can choose a surjective linear map π : Rn Rn−k such that ker(π)= L, and π maps Qn onto Qn−k. Exercise V.37 implies that π( ) is a rational polyhedral→ cone, and therefore, by the inductive hypothesis C π( ) = cone(T ′) for some finite T ′ Qn−k. Another application of exercise V.37 then implies that C ⊆ = cone(T ) for a finite T Qn, as required. C ⊆ Corollary V.30. If is a convex rational polytope, then each cone in the normal fan of is also rational P P (with respect to the basis on (Rn)∗ which is dual to the standard basis of Rn). 5. RATIONAL POLYHEDRA 71

PROOF. Let be a face of . We will show that the cone σQ from section V.4 is rational. Corol- lary V.23 impliesQ that is also aP convex rational polytope. Therefore by proposition V.29 there are ﬁnite n Q n ∗ subsets S,T of Q such that each is = conv(S) and = conv(T ). Since σQ = ν (R ) : ′ ′ P P Q { ∈ ν,β β =0 for all β,β T , and ν, α β 0 for all α S, β T , it follows that σQ is a rational hpolyhedral− i cone, as required.∈ h − i≥ ∈ ∈ } n Recall that the Hausdorff distance of , R is sup infβ∈Q α β , where is the P Q ⊂ α∈P { || − ||} || · || Euclidean norm on Rn. Corollary V.31. Every polytope can be approximated arbitrarily closely (with respect to the Hausdorff distance) by rational polytopes.

PROOF. This follows from theorem V.3 and proposition V.29, since to approximate the convex hull of n α1,...,αN R , it sufﬁces to take the convex hull of β1,...,βN , where each βj has rational coordinates, ∈ and is sufﬁciently close to αj . Many results from the theory of rational polyhedra (including proposition V.29) are ultimately based on the following basic fact from linear algebra. Lemma V.32. Consider a linear system of equations of the form a + a x + + a x = 0, i,0 i,1 1 ··· i,n n i = 1,...,k. Assume each ai,j Q. Let HQ and HR be the set of common solutions of this system n n ∈ respectively in Q and R . Then dimQ(HQ) = dimR(HR). In particular, the system has a common solution over R if and only if it has a common solution over Q. PROOF. If the system has a solution, then it can be found by Gaussian elimination, which produces a solution in Q (since each ai,j Q). Moreover, in that case the dimension of the space of solutions is ∈ precisely the rank of the k n matrix (ai,j ), 1 i k, 1 j n. Since the rank over Q is the same as × ≤ ≤ ≤ ≤ the rank over R, this completes the proof. Corollary V.33 (Cf. proposition V.20). Let be a convex rational polyhedron and α be a point with rational coordinates in the relative interior ofP . P (1) If is a polytope with vertices α ,...,α , then there are positive rational numbers q ,...,q P 1 k 1 k such that j qj =1 and α = j qj αj . n (2) If is a cone generated by α1,...,αk Z , then there are positive rational numbers q1,...,qk P P P ∈ such that α = j qj αj .

k n k PROOF. Let φ : R P R be the map which sends (r1,...,rk) rj αj α. If is a cone, → 7→ j=1 − P then corollary V.21 implies that φ−1(0) k = . Lemma V.32 then implies that φ−1(0) k = , R>0 P Q>0 which together with corollary V.21 proves∩ assertion6 ∅ (2). Assertion (1) follows by the same arguments∩ from6 ∅ ′ k n+1 proposition V.20 and lemma V.32 by considering the map φ : R R given by (r1,...,rk) (φ(r ,...,r ), r + + r 1). → 7→ 1 k 1 ··· k − The following result is one of the foundations of the theory of toric varieties we will encounter in chapter VI. Lemma V.34 (Gordan’s lemma). If σ is a rational convex polyhedral cone in Rn, then σ Zn is a ﬁnitely generated semigroup. ∩ n PROOF. It is straightforward to check that σ Z is a semigroup. Pick α1,...,αs Sσ which s ∩ n ∈ n generate σ as a cone. Let K := i=1 tiαi : ti R, 0 ti 1 R . Since K is compact, K Z is n{ s∈ ≤ ≤ }⊂ s ∩ a ﬁnite set. Now pick α σ Z . Then α = i=1 riαi, where each ri 0. Write α = i=1 ri αi + β, ∈ ∩ P ≥ ⌊ ⌋ n where ri is the greatest integer less than or equal to ri for each i, and β := i(ri ri )αi K Z . ⌊ ⌋ n P −⌊m P⌋ ∈ ∩ It follows that σ Z is generated as a semigroup by all the αj together with K Z . ∩ P ∩ 5.1. Exercises.

EXERCISE V.36. Assume either is a rational polyhedron or = conv(S) + cone(T ), where P P S,T Qn. If dim(aff( )) < n, then show that there is a rational affine hyperplane in Rn containing ⊂ P . [Hint: if is a polyhedron in Rn with dim( ) < n, then the set of points for which each of the inequalitiesP definingP is strict must be empty, forP otherwise dim( ) would be n. In the other case, P P 72 V. CONVEX POLYHEDRA

n n = conv(S)+cone(T ), with S,T Q . Let LP be the linear subspace of R generated by all elements ofP the form α β, where α β ⊂. It is possible to choose a generator of L consisting of elements of − − ∈P P the form α1 α2 + β1 β2, where α1, α2 S and β1,β2 are scalar multiples of elements from T . Since − − n ∈ each such element is in Q , it follows that LP is contained in a rational hyperplane. is contained in a n P translation of LP by an element of Q .] EXERCISE V.37. Let π : Rn Rm be a linear map such that π maps Qn into Qm. Given Rm, show that → C ⊆ (1) = cone(T ) for some (ﬁnite) T Qm if and only if π−1( ) = cone(T ′) for some (ﬁnite) C ⊂ C T ′ Qn. ⊂ (2) is a rational polyhedral cone in Rm if and only if π−1( ) is a rational polyhedral cone in Rn. C[Hint: use exercise V.6.] C

6. ∗Volume of convex polytopes

6 n Given linearly independent elements α1,...,αd R , the parallelotope generated by the αj is the set ∈ := λ α + + λ α :0 λ 1, 1 j d P { 1 1 ··· d d ≤ j ≤ ≤ ≤ } n It is straightforward to check that is a d-dimensional polytope in R (exercise V.38). We denote by Voln P the usual n-dimensional volume (i.e. the Lebesgue measure) on Rn. Given an afﬁne subspace H of Rn, we write VolH for the measure induced on H by Voln. Here is a precise deﬁnition of VolH : let H0 be the n (unique) linear subspace of R which is a translate of H (i.e. H0 = H α for any α H), d := dim(H), − ∈ n and β1,...,βn−d be an orthonormal set (with respect to the dot product) of elements in R such that H0 n is precisely the set of elements in R whose dot product with each βj is zero. Let be the parallelotope Q generated by β1,...,βd. If is any subset of H, then is measurable with respect to VolH if and only if + is measurable with respectR to Vol , and Vol ( R) = Vol ( + ). Q R n H R n Q R

β1

β2 Q H

2 FIGURE 10. If is a segment of a line H R , then VolH ( ) is the length of , and equals the area ofQ the rectangle with base ⊂and height one. Q Q Q

n If R is such that the dim(aff( )) d, then we write Vold( ) for VolH ( ), where H is any d- P ⊂ n P ≤ P P dimensional affine subspace of R containing ; this is well defined since VolH ( ) does not depend on H (exercise V.39). The following properties of VolP follow from basic analysis; we useP these without proof.

THEOREM V.35. Let be a polytope in Rn. P (1) If dim( ) d n, then the map λ R≥0 Vold(λ ) is homogeneous of order d, i.e. P ≤ d ≤ ∈ 7→ P Vold(λ )= λ Vold( ) for all λ 0. P P ≥ n (2) As a real valued function from the set of the polytopes in R , Voln is continuous with respect to the Hausdorff distance.

Let be an n-dimensional polytope, and O be an arbitrary point of Rn. For each facet of , let P Q P Q,O be the convex hull of O , i.e. Q,O is the “cone with base and apex O.” Then it is intuitively Sclear that the volume of canQ∪{ be computed} S in terms of the volumes ofQ the , see ﬁg. 11. This leads to P SQ,O

6The asterisk in the section name is to indicate that the material of this section is not going to be used in chapter VI. It is only used in the proof of Bernstein’s theorem in chapter VII. 6. ∗VOLUME OF CONVEX POLYTOPES 73 our next result. To state it we introduce a notation: if is a facet of an n-dimensional polytope Rn, Qn ∗ P ⊂ then (up to a positive multiple) there is a unique ν (R ) 0 such that = Inν ( ). Define ∈ \{ } Q P 1 if ν, O > min (ν), sign ( , O)= h i P Q P 1 otherwise. (− Geometrically, sign ( , O) is 1 if and only if O and are on the same side of aff( ) - see fig. 11. Q P P Q

C C νCA O αCA νCA B αCA B O

A A

(A) Vol2(ABC) = Vol2(OAB) + (B) Vol2(ABC) = Vol2(OAB) + Vol2(OBC) + Vol2(OCA) Vol2(OBC) − Vol2(OCA)

2 2 FIGURE 11. If is a triangle in R and O R , then for each side of , Q,O is the P ∈ Q P S triangle formed by and O. In ﬁg. 11b signCA(ABC, O)= 1 since O and ABC are on different sides ofQ the line containing CA. −

THEOREM V.36. Let be an n-dimensional polytope in Rn and O Rn. For each facet of , let d(O, aff( )) denote the distanceP between O and the affine hull of . Then∈ Q P Q Q 1 (29) Vol ( )= sign ( , O)d(O, aff( )) Vol ( ) n P n Q P Q n−1 Q XQ where the sum is over all facets of . In particular, Q P 1 1 (30) Voln( )= max(ν) Voln−1(ldν ( )) = min(ν) Voln−1(Inν ( )) P n P P −n P P ν∈(Rn)∗ ν∈(Rn)∗ ||Xν||=1 ||Xν||=1 Remark V.37. The norm on (Rn)∗ in (30) is the Euclidean norm induced from Rn upon identification ||·|| of (Rn)∗ and Rn via the basis dual to the standard basis of Rn.

PROOF. Let + (respectively, −) be the collection of facets of such that signQ( , O)=1 (respectively, 1).I The results of sectionI V.3 imply that Q P P − (i) Q( , O) (exercise V.40); P ⊂ Q∈I+ S P (ii) if − then Q,O relint( )= and Q,O ′ Q′ ( , O) (exercise V.41); Q ∈S I S ∩ P ∅ S ⊂ Q ∈I+ S P (iii) if then ′ ′ ( , O) (exercise V.42); + Q,O Q ∈I− Q Q ∈ I ′ S \P ⊂ S P S (iv) if and are distinct facets of such that sign ( , O) = sign ′ ( , O), then dim( Q Q S P Q P Q P SQ,O ∩ ′ ) n 1 (exercise V.43). SQ ,O ≤ − These observations immediately imply that Vol ( )= sign ( , O) Vol ( ) n P Q P n SQ,O XQ Since every cross section of parallel to the hyperplane aff( ) is a dilation of , it follows that SQ,O Q Q d(O,aff(Q)) d(O,aff(Q)) 1 Vol ( )= Vol (r )dr = rn−1dr = d(O, aff( )) Vol ( ), n SQ,O n−1 Q n Q n−1 Q Zr=0 Zr=0 where the second equality follows from theorem V.35. This completes the proof of identity (29). Now for n ∗ each facet of , let νQ (R ) be the outward facing unit normal to , i.e. = ldν ( ) and ν =1. Q P ∈ Q Q P || || 74 V. CONVEX POLYHEDRA

We now apply identity (29) with O being the origin of Rn. If we identify (Rn)∗ with Rn via the basis n dual to the standard basis of R , then αQ := signQ( , O)d(O, aff( ))νQ is a point on aff( ) (see e.g. P ǫO,Q Q Q fig. 11). It follows that minP (νQ) = νQ, αQ = ( 1) dO,Q. The first equality of identity (30) now follows from identity (29). The secondh equalityi follows− from the first by replacing ν by ν. −

6.1. Exercises.

EXERCISE V.38. Let := λ1α1 + + λdαd : 0 λj 1 for each j = 1,...,d , where n P { ··· ≤ ≤ } α1,...,αd R . Show that ∈ (1) is the convex hull of the set consisting of the origin and all elements of the form α + +α , P i1 ··· ik where 1 i1 <

(1) is measurable with respect to VolH1 if and only if it is measurable with respect to VolH2 . (2)P If is measurable with respect to either of them, then Vol ( ) = Vol ( ). P H1 P H2 P EXERCISE V.40. Prove observation (i) from the proof of theorem V.36. [Hint: it suffices to prove that relint( ) Q( , O). If α relint( ) O , the line through α and O intersects the facets at P ⊂ Q∈I+ S P ∈ P \{ } two points. One of these is contained in facets from .] S I+ EXERCISE V.41. Prove observation (ii) from the proof of theorem V.36. [Hint: if = , then either I− 6 ∅ O or O is on a facet of . Pick α , where −. If = Inν ( ), for the first part show that ν,β6∈ P min (ν) for each βP on the line∈ Q segment fromQ ∈O I to α.Q For the secondP part it suffices to show h i ≤ P relint( Q,O) ′ . Every point of relint( Q,O) is on the line segment between O and a point on S ⊂ Q ∈I+ S relint( ). Extending this line segment hits another point of the topological boundary of which belongs S to a facetQ from .] P I+ EXERCISE V.42. Prove observation (iii) from the proof of theorem V.36. [Hint: it suffices to consider the case that O . If α relint( ) and , then the line segment L from O to α intersects 6∈ P ∈ Q Q ∈ I+ the boundary of at a point β “in between” O and α. Any facet of containing β is in − and L ( , OP).] R P I \P⊂SR P EXERCISE V.43. Prove observation (iv) from the proof of theorem V.36. [Hint: pick distinct facets

1, 2 of such that Qj ,O are full dimensional and relint( Q1 ,O) relint( Q2,O) = . It sufﬁces to showQ Q that signP ( , O)Shave different signs for j =1 and j =2S . Indeed,∩ if α isS a point in6 the∅ intersection, Qj P then the line through O and α intersects at β relint( ), j = 1, 2. Show that one of the β is “in P j ∈ Qj j between” O and the other βj .]

EXERCISE V.44. Given positive ω ,...,ω and nonnegative d, m ,...,m , where 0 p n, let 1 n 1 p ≤ ≤ (~ω, d, ~m) be the polytope in Rn determined by the following inequalities: Q x 0 for each i =1, . . . , n, i ≥ n ω x d, i i ≤ i=1 X x m , i =1, . . . , p. i ≤ i Let (~ω, d, ~m) be the collection of subsets I of 1,...,p such that ω m d. I { } i∈I i i ≤ (1) Showthat theverticesof (~ω, d, ~m) are precisely the elementsP αI = (αI,1,...,αI,n) and βI,j = nQ (βI,j,1,...,βI,j,k) R , indexed by I (~ω, d, ~m) and j 1,...,n I, and deﬁned as ∈ ∈ I ∈ { }\ 7. ∗VOLUME OF SPECIAL CLASSES OF POLYTOPES 75

follows:

mk if k I, αI,k := ∈ 0 if k I, ( 6∈ m if k I, k ∈ d−Pi∈I ωimi βI,j,k := if k = j,  ωj 0 if k I j . 6∈ ∪{ } (2) Let di,mi,1,...,mi,p, i =1, 2, be nonnegativereal numberssuch that (~ω, d1, ~m1)= (~ω, d2, ~m2)= (~ω, d +d , ~m +~m ). Thenshow that (~ω, d +d , ~m +~m )= (~ω,I d , ~m )+ (~ω,I d , ~m ). I 1 2 1 2 Q 1 2 1 2 Q 1 1 Q 2 2 (3) With ~ω = (1, 1, 1), and p =2, show that (~ω, 3, (1, 1)) + (~ω, 3, (3, 3)) ( (~ω, 6, (4, 4)), i.e. the conclusion of assertion (2) may not beQ true in the absenceQ of its assumption.Q

n EXERCISE V.45. (1) Givenpolytopes ∆1,..., ∆p R , show that ⊂ p p Vol ∆ = ( 1)q−1 Vol ∆ q − i q=1 ! q=1 ! [ X I⊆[pX], |I|=q i\∈I where [p] denotes the set 1,...,p (this is the so called “inclusion-exclusion principle,” it holds for all “measurable” sets with{ finite} volume, i.e. as long as the volume of each intersection on the right hand side is well defined and finite). (2) Given ~ω, d, ~m as in exercise V.44, define n ∆ := (x ,...,x ): x 0, i =1, . . . , n, ω x d 0 { 1 n i ≥ i i ≤ } i=1 X n ∆ := (x ,...,x ): x 0, i =1,...,n, x m , ω x d , q =1, . . . , p. q { 1 n i ≥ q ≥ q i i ≤ } i=1 X Show that the polytope (~ω, d, ~m) from exercise V.44 equals ∆ p ∆ . Conclude that Q 0 \ q=1 p p 1 S Vol( (~ω, d, ~m)) = ( 1)q (d ω m )n Q n!ω ω − − i i 1 n q=0 ··· X I⊆[pX], |I|=q Xi∈I Pi∈I ωimi

7. ∗Volume of special classes of polytopes 7In section V.7.1 we study the dependence of the volume of Minkowski sums of polytopes on its summands, and in section V.7.2 we give a formula of the volume of rational polytopes in terms of “lattice volumes” of its facets.

7.1. Minkowski sums. Theorem V.35 implies that volume interacts well with Minkowski addition, n in the sense that given compact convex subsets , of R , the function from R≥0 to R≥0 given by P Q λ Voln( + λ ) is continuous. However, it turns out that this function is much more than a continuous function,7→ itP is a polynomialQ . In this section we are going to prove this result for the case of polytopes. At ﬁrst we need the following result.

n s Lemma V.38. Let 1,..., s be subsets of R and λ := (λ1,...,λs) R . Then for different λ, the P P ∈ >0 afﬁne hull Aλ of λ1 1 + + λs s are translations of each other. In particular, dim(Aλ) is independent of λ. P ··· P

PROOF. Fix an arbitrary element αi of i, i = 1,...,n. Without loss of generality we may replace P s i by i αi and assume that each i contains the origin. For each λ R>0, it then follows that Aλ P P − P n ∈ contains each i, and therefore it is simply the linear subspace of R spanned by elements in i. P i P 7The asterisk in the section name is to indicate that the material of this section is not going to be used in chapterSVI. It is only used in the proof of Bernstein’s theorem in chapter VII. 76 V. CONVEX POLYHEDRA

n THEOREM V.39. Let 1,..., s be convex polytopes in R . Then there are nonnegative real numbers P P s vα( 1,..., s) for all α s := (α1,...,αs) Z≥0 : α1 + + αs = n such that for all λ = P P s ∈ E { ∈ ··· } (λ1,...,λs) R , ∈ ≥0 Vol (λ + + λ )= v ( ,..., )λα1 λαs n 1P1 ··· sPs α P1 Ps 1 ··· s αX∈Es where Voln is the n-dimensional volume.

PROOF. We proceed by induction on n. If n =1 each is of the form [a ,b ], so that Pi i i Vol (λ + + λ ) = Vol ([λ a + + λ a , λ b + + λ b ]) 1 1P1 ··· sPs 1 1 1 ··· s s 1 1 ··· s s = λ (b a )+ + λ (b a ) 1 1 − 1 ··· s s − s = λ Vol ( ) i 1 Pi i X n−1 n Now assume it is true for convex polytopes in R . Now pick convex polytopes 1,..., n in R . Since the volume is translation invariant, we may assume that P P

( ) the origin is in the relative interior of each j . ∗ P Let P := λ + + λ . If dim(P ) n 1, then the result is true with all v being zero. λ 1P1 ··· sPs λ ≤ − α So assume dim(Pλ) = n. Due to theorem V.35 it suffices to consider the case that each λi is positive. Then proposition V.16 and lemma V.38 imply that the number of facets of does not depend on λ, and Pλ moreover, if λ,1,..., λ,N are the facets of λ, then for each j, there are faces i,j of i, i =1,...,s, such that P P P P P = λ + + λ Pλ,j 1P1,j ··· sPs,j For each i, j, pick an arbitrary αi,j i,j . Let νj be the outward pointing unit normal to λ,j. Identity (30) implies that ∈ P Q 1 Voln( λ)= max(νj ) Voln−1( λ,j ) P n Pλ P j X 1 = ν , λ α + + λ α Vol (λ + + λ ) n h j 1 1,j ··· s s,j i n−1 1P1,j ··· sPs,j j X Condition ( ) implies that νj , αi,j is nonnegative for each i, j. Since for each j, all the i,j can be ∗ h i n−1 n−P1 identified (via a volume preserving affine map from aff( λ,j ) to R ) with polytopes in R , the result then follows from the inductive hypothesis. P

7.2. Rational polytopes. Let H be a d-dimensional rational afﬁne subspace of Rn. If β H Zn, n d ∈ ∩ lemma V.32 implies that GH := (H β) Z is isomorphic (as an abelian group) to Z . A fundamental lattice parallelotope on H is a polytope− of the∩ form +β, where isa(d-dimensional) parallelotope generated by d elements from G which generate G Pas an abelianP group. We write fund(H) := Vol ( ). H H H P Proposition V.40 below shows that fund(H) is well deﬁned. In this section we identify Rn with (Rn)∗ via the dot product, and given α, β Rn, write α, β for the dot product of α and β. Similarly we write ∈ h i β⊥ := γ Rn : β,γ =0 . { ∈ h i }

n Proposition V.40. Let H be a rational afﬁne subspace of R . If 1, 2 are two fundamental lattice parallelotopes of H, then Vol ( ) = Vol ( ). P P H P1 H P2 n PROOF. By translating H and the j if necessary we may assume H is a linear subspace of R , P n and each j is the parallelotope generated by αj,1,...,αj,d H Z , where d := dim(H). Pick an P n ⊥ ∈ ∩ orthonormal set β1,...,βn−d R such that H = i βi . For each j = 1, 2, let j be the basis of n ∈ B R consisting of β1,...,βn−d, αj,1,...,αj,d. By deﬁnition VolH ( j ) = Voln( j ), where j is the T n nP Q Q parallelotope generated by the elements of j. Let φ : R R be the linear map which changes B → n coordinates with respect to 1 to that of 2. Since αj,1,...,αj,d generate H Z (as an abelian group), it follows that the matrices ofB both φ andBφ−1 have only integer entries. This∩ means that the determinant 7. ∗VOLUME OF SPECIAL CLASSES OF POLYTOPES 77

B M

D A

′ ′ FIGURE 12. fund(L)= √2, fund(M)= √5. VolL(AB)=3, VolM (CD)=2.

of the matrix of φ is 1, and therefore φ preserves Voln. Since φ maps one of the j to the other, this completes the proof. ± Q

Let H′ be a rational affine subspace of Rn such that H′ H and dim(H′)= d +1. We now describe ⊃ the relation between fund(H) and fund(H′). Pick β H. Since (H β) Qn (H′ β) Qn is ∈ − ∩ ⊂ − ∩ an inclusion of vector spaces over Q, it follows from the elementary theory of vector spaces that there is η′ (H′ β) Qn such that H β = η′⊥ (H′ β). Pick r Q 0 such that η := rη′ is “primitive,” ∈ − ∩ − ∩ − ∈ \{ } i.e. there is no integer k> 1 such that η = kη′′ for some η′′ Zn. Let denote the Euclidean norm on ∈ || · || Rn. Proposition V.41. η fund(H)= || || fund(H′) min η, α : α ((H′ β) Zn) (H β) {|h i| ∈ − ∩ \ − } In particular, if d = n 1 and H′ = Rn, then fund(H)= η . − || || PROOF. Replacing H by H β if necessary we may assume that β = 0. Lemma B.43 implies that − n we may pick α1,...,αd+1,ud+1,...,un R such that n ∈ α1,...,αd generate H Z , • ∩ ′ n α1,...,αd+1 generate H Z , • u H′, H = u⊥ H∩′, u =1, • d+1 ∈ d+1 ∩ || d+1|| u ,...,u are orthonormal (with respect to dot product), and H′ = n u⊥. • d+2 n j=d+2 j ′ Let M be the matrix with column vectors α1,...,αd,ud+1,...,un and M beT the matrix with column vectors α ,...,α ,u ,...,u . Write u = c α + u′, where u′ H. Then 1 d+1 d+2 n d+1 d+1 d+1 ∈ fund(H)= det(M) = c det(M ′) = c fund(H′) | | | d+1 | | d+1| Note that ud+1 = η/ η . It follows that η,ud+1 = η . On the other hand, η,ud+1 = ′ ± || || h i ±||′ || n h i η,cd+1αd+1 + u = cd+1 η, αd+1 . Since for each α H Z , η, α is an integer multiple of hη, α , the resulti follows. h i ∈ ∩ h i h d+1i Definition V.42. Let H be a rational affine subspace of Rn of dimension d. The H-normalized volume is Vol′ ( ) := Vol ( )/ fund(H) H · d · see fig. 12 for some examples with n = 2. An integral element of Rn is an element with integral coordinates; an integral element of (Rn)∗ is one which has integral coordinates with respect to the basis which is dual to the standard basis of Rn. An integral element η of Rn or (Rn)∗ is primitive, if it is not of the ′ ′ n ∗ ′ form kη , where k > 1 and η is also integral. If ν is an integral element of (R ) , then we write Volν for ′ Volν⊥ . Corollary V.43. Let be a convex rational polytope in Rn. Then P 1 ′ 1 ′ (31) Voln( )= max(ν) Volν (ldν ( )) = min(ν) Volν (Inν ( )) P n P P −n P P ν ν X X where the sum is over all primitive integral ν (Rn)∗. ∈ 78 V. CONVEX POLYHEDRA

PROOF. Since is rational, every facet of is determined by integral elements of (Rn)∗. Therefore the result follows fromP combining eq. (30) and propositionP V.41. Remark V.44. If is a facet of an n-dimensional rational polytope in Rn, then the primitive inner Q nP ∗ (respectively outer) normal to is the unique primitive integral ν (R ) such that = Inν ( ) (respec- F ∈ Q P tively = ldν ( )). Note that the sum in (31) is practically ﬁnite: the only non-zero contributions come from thoseQ ν whichP are primitive outer normal to (n 1)-dimensional faces of . − P Part 1

Toric Varieties and classical results CHAPTER VI

Toric varieties over algebraically closed ﬁelds

This chapter introduces toric varieties, which are the setting of all the subsequent results of this book. Our treatment will be mostly based on the results from chapters III and V; only in section VI.7 we use the notion of closed subschemes discussed in section IV.2. Unless explicitly stated otherwise, from this chapter

∗

k k onward k denotes an algebraically closed ﬁeld (of arbitrary characteristic), and denotes 0 . \{ } 1. Algebraic torus

n ∗ n n ∗ n

k k k If (x1,...,xn) are coordinates on k , n 1, then ( ) = V (x1 xn). This implies that ( ) ≥ −1 \ ···−1 is an afﬁne variety, and its coordinate ring is the ring k[x1, x1 ,...,xn, xn ] of Laurent polynomials in

∗ n k x1,...,xn over k (exercise III.11). An (algebraic) torus is a variety X isomorphic to ( ) for some n 1. A system of coordinates on X is an ordered collection (x1,...,xn) of regular functions on

≥ −1 −1 k X such that the coordinate ring k[X] of X is [x1, x1 ,...,xn, xn ]. A basic property of a torus is that every morphism between two tori is a group homomorphism and also a monomial map with respect ∗ N to every set of coordinates. Indeed, let φ : X Y = (k ) be a morphism. Choose coordinates → ∼ (y1,...,yN ) on Y . If φ(x) = (φ1(x),...,φN (x)), then each φj must be a monomial in (x1,...,xn), for otherwise it will be zero at some point of X (exercise VI.1). This shows that φ is a monomial map. αj n ′ ′ ′ Write φj (x) = x , α1,...,αN Z . If x = (x1,...,xn), x = (x1,...,xn) X, then it follows ′ ′ ′∈ ′ α1 ′ αN ′ ∈ that φ(x x ) = φ(x1x1,...,xnxn) = ((x x ) ,..., (x x ) ) = φ(x) φ(x ), so that φ is indeed a homomorphism.· This implies in particular that· the multipli·cation with respect· to every set of (algebraic) coordinateson a torus induces the same groupstructure on it, and the imageof a morphismbetween two tori is a subgroup of the target. Proposition VI.1 below shows that it is in addition a Zariski closed subset of the target1. We use the following notation in proposition VI.1: givena monomialmap φ : x (xα1 ,...,xαN ) 7→ between tori with ﬁxed systems of coordinates, we write [φ] for the N n matrix whose rows are the αi. Some basic properties of [φ] are established in exercise VI.2. × n ¯ n Proposition VI.1. Let G be the subgroup of Z generated by the αi, and G := α Z : kα G for { ∈ ∈ some k 1 be the “saturation” of G in Zn. Let q be the index of G in G¯. Let r be the rank of [φ] as a ≥ } matrix over Q. Then (1) φ(X) is a torus and a closed subvariety of Y of dimension r. (2) ¯ r for positive integers such that . G/G ∼= j=1 Z/qj Z q1,...,qr q = j qj ¯ ∗ n−r (3) ker(φ) is an (n r)-dimensional subgroup of X isomorphic to (G/G) (k ) . In particular, Q Q if r = n, then the− degree of φ (as a map from X to φ(X)) is q. ×

n ∗ n−r ∗ n k (4) Pick a basis (β1,...,βn−r) of ker[φ] Z , and let η : (k ) ( ) be the morphism such ⊆ → that the column vectors of [η] are the βj . Then the irreducible component of ker(φ) containing (1,..., 1) is the image of η. PROOF. Let φ′ : Zn ZN be the map corresponding to multiplication by [φ]. With respect to → appropriate coordinates on Zn and ZN , the matrix of φ′ is of the form D 0 0 0 where D is a diagonal matrix with positive integers as diagonal entries (corollary B.44). This means that we can choose coordinates (x1,...,xn) on X and (y1,...,yN ) on Y with respect to which φ takes the

1 n n k Contrast this to the case of k : the additive group structures on with respect to different systems of algebraic coordinates

n n k are in general different, and the image of a morphism from k to is in general neither a subgroup nor a closed subvariety of the target (see section III.9).

80 2. TORIC VARIETIES FROM FINITE SUBSETS OF Zn 81

q1 qr form (x1,...,xn) (x1 ,...,xr , 1,..., 1) for positive integers q1,...,qr. All the assertions are now straightforward; their7→ proofs are left as exercise VI.3.

1.1. Exercises.

EXERCISE VI.1. Show that every polynomial in (x1,...,xn) which is not a monomial vanishes at ∗ n some points on (k ) . [Hint: use exercise III.3.]

EXERCISE VI.2. Let φ : X Y be a morphism between two tori. Let (x ,...,x ) (respectively, → 1 n (y1,...,yN )) be coordinates on X (respectively, Y ) and [φ] be the corresponding matrix of φ. Show that n β β[φ] (1) for each β = (β1,...,βN ) Z , (φ(x)) = x , where β[φ] is the product of β (regarded as a 1 N matrix) and [φ]; ∈ × (2) φ is an isomorphism if and only if N = n and [φ] is invertible over Z; (3) if Z is a torus and ψ : Y Z is a morphism, then [ψ φ] = [ψ][φ]. → ◦ EXERCISE VI.3. Complete the proof of proposition VI.1.

2. Toric varieties from finite subsets of Zn A toric variety is a variety X which contains an algebraic torus as a dense open subset such that the (multiplicative) action of the torus on itself extends to an action on all of X. Given a finite subset n ∗ n N = α0,...,αN of Z , we write φA : (k ) P for the map given by A { } → (32) x [xα0 : : xαN ] 7→ ··· 0 0 N We write XA for the image of φA and XA for the closure of XA in P . We will now show that XA 0 N is a toric variety with torus XA. Denote the homogeneous coordinates of P by [zα0 : : zαN ]. Let N N ··· Uα := P V (zα), α , be the basic open subsets of P . \ ∈ A Proposition VI.2. XA is a toric variety. More precisely, 0 0 (1) XA is a torus and XA = XA α∈A Uα. ∩ β (2) For each α A, X U is an affine variety with coordinate ring k[x : β S ], where S is ∈ A ∩ α T ∈ α α the subsemigroup of Zn generated by α := β α : β . A− 0{ − ∈ A} (3) The dimension of XA (and equivalently, of XA) equals the dimension (as a polytope) of the convex hull of in Rn. 0 A 0 N (4) XA acts on XA via the multiplicative action of XA on P given by: (33) [y : : y ] [z : : z ] := [y z : : y z ] α0 ··· αN · α0 ··· αN α0 α0 ··· αN αN 0 N for all [yα0 : : yαN ] X and [zα0 : : zαN ] P . ··· ∈ A ··· ∈ 0 ∗ N PROOF. Since X U := U , and since U (k ) via the map [z : : z ] A α∈A α ∼= α0 αN ⊂ 0 ∗ N α1−α0 ··· αn−α07→ (zα1 /zα0 ,...,zαN /zα0 ), it follows that XA is the image in (k ) of the map x (x ,...,x ). T0 7→ 0 Proposition VI.1 then implies that XA is a torus, and also implies assertion (3). It also says that XA is a 0 closed subset of U, which implies that XA = XA U, and proves assertion (1). Assertion (2) follows ∩ 0 N directly from corollary III.29. Finally, since for a fixed y XA, the action of y on P given by (33) is 0 ∈ an isomorphism (exercise VI.4) and since XA is closed under this action, it follows that XA is also closed under it (exercise VI.5), as required to prove assertion (4).

N Proposition VI.2 states in particular that XA is “equivariantly embedded” in P , i.e. the action of the N torus on XA extends to all of P . Conversely every equivariantly embedded projective toric variety is essentially of the form XA for some appropriate (see e.g. [GKZ94, Proposition 5.1.5]); we will not use this result. We now show that X0 and X dependA only on the afﬁne geometry of the set . A A A Proposition VI.3 ([GKZ94, Proposition 5.1.2]). Let Zn, Zm, and T : Zn Zm be an injective integer afﬁne transformation such that T ( )=A ⊂. Then XB0 ⊂= X0 and X = X →as subsets of A B A B A B PN , where N = 1. |A| − 82 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

PROOF. Let A = α0,...,αN , = β0,...,βN , where βj = T (αj), j =0,...,N. By deﬁnition { m } B { } n there is λ = (λ1,...,λm) Z and an n m matrix M such that for each γ = (γ1,...,γn) Z ,

∈ m × ∗ ∗ m ∗∈ n k T (γ)= λ + γM. Let µj Z be the j-th row vector of M, j = 1,...,n, and T : (k ) ( ) be ∈ → the map deﬁned by T ∗(y) = (yµ1 ,...,yµn ). Exercise VI.2 implies that T ∗(y)αj = yαj M = yβj −λ for each j = 0,...,N. Since the rank of M = [T ∗] is n, proposition VI.1 implies that T ∗ is surjective, and therefore

0 β0 βN ∗ m −λ β0 −λ βN ∗ m k X = [y : : y ]: y (k ) = [y y : : y y ]: y ( ) B { ··· ∈ } { ··· ∈ }

∗ α0 ∗ αN ∗ m α0 αN ∗ n 0 k = [T (y) : : T (y) ]: y (k ) = [x : : x ]: x ( ) = X { ··· ∈ } { ··· ∈ } A which completes the proof. 2.1. Exercises. EXERCISE VI.4. Show that for a ﬁxed y X0 , the action of y on PN given by (33) is an isomorphism ∈ A from PN to itself. EXERCISE VI.5. Let W be a subset of a topological space X and φ : X X be a continuous map. → Show that φ(W ) φ(W ) (where the “bar” indicates closure in X). Deduce that if φ(W ) W , then ⊆ ⊆ φ(W ) W . ⊆ EXERCISE VI.6. In this exercise you will show that XA is a binomial variety, i.e. XA is deﬁned in N N q N qj P by binomial equations. Given q = (q0,...,qN ) Z , write z := j=0 zαj . Let J be the ideal of ∈ q1 q2 n R := k[zα0 ,...,zαN ] generated by binomials of the form z z , where qi := (qi,0,...,qi,N ) Z , N N N − N Q ∈ i = 1, 2, are such that j=0 q1,j = j=0 q2,j and j=0 q1,jαj = j=0 q2,j αj . Let I be the ideal of R consisting of all homogeneous polynomials that vanish on XA. P P P P (1) Show that J I. ⊂ q α0 αN ∗ n (2) Let f = q∈ N cqz I. Use the fact that f(x ,...,x ) is identically zero on (k ) to Z≥0 ∈ n N N show thatP for each α Z , cq =0, where α := (q0,...,qN ) Z≥0 : qj αj = ∈ q∈Sα S { ∈ j=0 α . } n P q P (3) Given α Z , write fα := cqz . Use the preceding step to show that fα J. Deduce ∈ q∈Sα ∈ that I = J. P 3. Examples of toric varieties

Given a subset Y of a topological space X, in this section we write ClX (Y ) for the closure of Y in X. n We also write e1,...,en for the standard unit vectors of R , i.e. for each i, j, the j-th coordinate of ei is 0 if j = i, and 1 if j = i. 6 ∗ 1 Example VI.4. If = 0, 1 Z, then XA = ClP2 ( [1 : x]: x k = P . More generally, if A { } ⊂ { ∈ }∗ n n = 0,e1,...,en , then XA = Cl n ( [1 : x1 : : xn] : (x1,...,xn) (k ) )= P . A { } P { ··· ∈ } ∗ Example VI.5. If = (0, 0), (1, 0), (0, 1), (1, 1) , then XA = ClP3 ( [1 : x : y : xy]: x, y k ) = A3 { } { 3 ∈ 1} 1 V (z1z2 z0z3) P . In the literature XA is denoted as the quadric surface in P . Note that XA ∼= P P via the morphism− ⊂ × 1 1 P P ([x0 : x1], [y0 : y1]) [x0y0 : x0y1 : x1y0 : x1y1] XA × ∋ 7→ ∈ n 1 n More generally, if = 0, 1 , assertion (2) of proposition VI.6 below implies that XA = (P ) . A { } ∼ n Given a set R and a positive integer d, in this chapter we write d := α1 + + αd : αj A⊂n n A { ··· ∈n A for each j R . If , are ﬁnite subsets of Z , we write δA,B for the diagonal map from k to } ⊂ A B P|A|−1 P|B|−1 given by × (34) x (φ (x), φ (x)) 7→ A B where φA and φB are deﬁned as in (32).

ni Proposition VI.6. Let 1, 2 be ﬁnite subsets of Z and d1, d2 be positive integers. A A (1) X X . A1 ∼= d1A1 (2) X X = X . A1 × A2 ∼ d1A1×d2A2 4. STRUCTURE OF XA 83

∗ n (3) Assume n1 = n2 = n. Let Xd1,d2 be the closure in Xd1A1 Xd2A2 of δd1A1,d2A2 ((k ) ). Then X X X X . × A1+A2 ∼= d1A1+d2A2 ∼= d1,d2 ∼= 1,1

PROOF. We may assume without loss of generality that each i contains the origin. If νdi is the degree-d Veronese map (see section III.6.2), then it follows that φA = π ν φ , where π is a i diAi ◦ di ◦ Ai projection which omits “redundant” coordinates of νdi φAi (i.e. for each α di i, π retains only one α ◦ ∈ A of the coordinates of νdi φAi equalling x ). Assertion (1) then follows from proposition III.17. For ◦ ∗ n1+n2 assertion (2), let φA1 φA2 : (k ) XA1 XA2 be the morphism which maps (x1, x2) × ∗ ni → × |A1|−1 |A2|−1 |A1||A2|−1 7→ (φA1 (x1), φA2 (x2)), where xi (k ) , i = 1, 2. If s : P P P is the Segre ∈ × → map, then assertion (2) follows from assertion (1) and the observation that s (φA1 φA2 ) = φA1×A2 . The proof of assertion (3) is left as exercise VI.7. ◦ ×

Example VI.7. Let d be a positive integer. If = 0, 1, 2,...,d Z, then assertion (1) of proposi- A1 { } ⊂ tion VI.6 and example VI.4 imply that XA ∼= P . In the literature XA is denoted as the rational normal curve of degree d in Pd. 2 3 ∗ 2 3 2 Example VI.8. If = 0, 2, 3 Z, then XA = ClP2 ( [1 : x : x ]: x k )= V (z0z2 z1 ) P . A { }⊂ { ∈2 } 2 − ⊂ If (x, y) = (z1/z0,z2/z0) are coordinates on the basic open subset U0 = P V (z0) of P , then XA U0 is the curve deﬁned by y2 = x3. Note that the curve is singular at the origin (example\ III.31). ∩

∗ n A toric variety of dimension n by deﬁnition contains an open subset isomorphic to (k ) . For some n classes of , it is not hard to see that XA contains open subsets isomorphic to k , i.e. XA is a compactiﬁ- An cation of k .

Proposition VI.9. Let be a ﬁnite subset of Zn. An n

(1) Assume and . Show that k . Z≥0 0,e1,...,en U0 XA ∼= (2) More generally,A⊂ assumeA⊃{ there is a vertex α}of such that there∩ are precisely n edges ,..., A E1 En of conv( ) containing α and on each i, there is an element of of the form α + βi, where A n E n A β1,...,βn is a basis of Z . Then show that Uα XA = k . ∩ ∼ PROOF. This is exercise VI.8. 3.1. Exercises.

EXERCISE VI.7. Prove assertion (3) of proposition VI.6.

EXERCISE VI.8. Prove proposition VI.9.

4. Structure of XA We continue with the notation and the set up of section VI.2. In this section we study the complement 0 of XA in XA (i.e. the subvariety of XA “at infinity”) and catch a glimpse of its beautiful combinatorial structure. Let be the convex hull of . Then is a rational polytope (proposition V.29). A face of is by definitionP a set of the form A whereP is a face of . We say that is a facet (respectivelyA vertex, edge) of if B is a facet (respectivelyQ ∩ A vertex,Q edge) of ,P see fig. 1. B A Q P

Proposition VI.10. Let z X . Deﬁne := α : z U . Then is a face of . ∈ A Az { ∈ A ∈ α} Az A PROOF. Pick α , α . For each i, let be the (unique) face of which contains α in its 1 2 ∈ Az Qi P i interior. For every ǫ (0, 1), the convex combination αǫ := ǫα1 + (1 ǫ)α2 of α1 and α2 belongs to the relative interior of a face∈ of containing both (corollary V.18). It− sufﬁces to show that . Q P Qi Q∩A⊂Az Indeed, let ǫ be a rational number in (0, 1). Let β1,...,βk be the vertices of . There are positive integers k k Q k N1,...,Nk such that ( j=1 Nj )αǫ = j=1 Njβj (corollary V.33). Write N := j=1 Nj. Multiplying k Nj the Nj by some appropriate integer we may ensure that Nǫ is a positive integer. Then z = P P P j=1 βj zNǫzN−Nǫ on X . Since both α are in , it follows that z = 0 for each j. Since each β α1 α2 A i Az βj |z 6 Q ∈Q∩A is a convex rational linear combination of the βj , it follows by the same reasoning that zβ z = 0 for each β , as required. | 6 ∈Q∩A 84 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

C C′

D B

E D′ A A′ E′B′

(A) A = {A,B,C,D,E} (B) A′ = {A′, B′,C′, D′,E′} Facets: {A,E,B}, {B,C}, {C,A} Facets: {A′,E′, B′}, {B′,C′}, {C′, D′,A′} Vertices: A, B, C Vertices: A′, B′, C′

FIGURE 1. Faces of some planar sets

Given ′ , we write ′ (respectively, ′ ) to denote that ′ is a face (respectively, a proper face)B of⊆B⊆A. B B B B B B THEOREM VI.11. For each face of , deﬁne V := X U and O := V U . B A B A \ α6∈B α B B ∩ β∈B β (1) ′ . In particular, VB = B′B OB S T S k (35) X X0 = O = V = V A \ A B B Aj j=1 B[A B[A [ where ,..., are the facets of . A1 Ak A (2) There is a one-to-one correspondence between the collection of OB for and the set of 0 B 0 A orbits of XA on XA. In particular, each VB is invariant under the action of XA. (3) Each VB is a toric variety with torus OB. More precisely, the pair (VB, OB) is isomorphic to 0 |B|−1 (XB,X ). The isomorphism is given by the projection map πB : VB P which ‘forgets’ B → all the coordinates zα such that α ; in other words, πB([zα : α ]) = [zβ : β ]. 6∈ B 0 ∈ A ∈B 0 (4) The action of OB on VB is compatible with the action of XA. More precisely, assume yA XA 0 ∗ n α ∈ β

and correspond to the same k , i.e. and yB OB ∼= XB x ( ) yA = [x : α ] yB = [x : β ].∈ Then for all z V , ∈ ∈ A ∈B ∈ B y z = y z A ·A B ·B where we write (respectively, ) to denote the action of X0 (respectively, O ) on V . ·A ·B A B B PROOF. The ﬁrst statement of assertion (1) follows from proposition VI.10, and this in turn implies ′ the ﬁrst two equalities of (35). Since VB′ VB whenever , and since every proper face is contained in a facet (proposition V.22), the last equality⊆ of (35) follows.B B We now prove the remaining assertions. N Given , let HB := V (zα : α ) P be the coordinate subspace containing VB. Let πB : N B A 6∈ B ⊂ N 0 P V (zβ : β ) HB be the natural projection and ZB be the closure in P of πB(X ). \ ∈B → A Claim VI.11.1. VB = ZB.

PROOF. The inclusion V Z follows from a general property of morphisms (assertion (4) of B ⊆ B corollary III.29). For the opposite inclusion it suffices (due to the definitionof VB) to show that XA ZB. ′ ⊃ Pick β . We will show that XA Uβ ZB Uβ. We may assume β = α0. Write zi := zαi /zα0 , ∈ B N ∩′ ⊇ ′ ∩ ′ ′ ′ ′γ so that U k with coordinates (z ,...,z ). Let f(z ,...,z ) = N c z be a polynomial in β ∼= 1 N 1 N γ∈Z γ ′ ′ 0 (z1,...,zN ) which vanishes on XA Uβ. It suffices to show that f vanishes on ZB Uβ as well. Note that ′ ′ ∩ P ∩ 0 α1 αN ∗ n ′ ′ ′′ XA Uβ = (x ,...,x ): x (k ) , where αi := αi α0, i =1,...,N. Write f = f +f , where ∩ { ′ ∈ }′ − ′′ the monomials in f consist solely of the zi such that αi and each monomial in f contains at least one ′ ∈B zi such that αi . If B is the affine hull of αi α0 : αi , exercise VI.9 implies that the exponent of 6∈ B ′ α′ α′ { − ∈ B} ′′ α′ α′ each monomial in f (x 1 ,...,x N ) is on B, whereas the exponent of no monomial in f (x 1 ,...,x N ) ′ ′ is on B; in particular, the monomials that appear in f ′(xα1 ,...,xαN ) are distinct from those appearing in 5. TORIC VARIETIES FROM POLYTOPES 85

′ ′ ′ ′ ′ ′ ′′ α1 αN α1 αN ∗ n ′ α1 αN f (x ,...,x ). Since f(x ,...,x ) is identically zero on (k ) , it follows that f (x ,...,x ) ′ ′ ′ ′ ∗ n α1 αN ′ α1 αN is also identically zero on (k ) . This implies that f πB(x ,...,x )= f (x ,...,x )=0 for all ∗ n 0 ◦ x (k ) , so that f vanishes on π (X ), as required. ∈ B A

It is evident that ZB can be identified to XB by “forgetting” the coordinates zα for all α . Since 0 6∈ B VB = ZB, proposition VI.2 implies that this induces an identification of XB with OB, which proves assertion (3). Assertion (4) then follows from identity (33). Since assertion (4) in particular implies that 0 OB is an orbit of XA, this proves assertion (2) as well. Corollary VI.12 below is an immediate corollary of theorem VI.11. The second statement of corollary VI.12 in particular implies that the complement of the torus is locally defined by a single equation on X ; in the terminology of section IV.2.4, X X0 is the “support of a Cartier divisor” on X . P P \ P A Corollary VI.12. If is the set of vertices of , then X U . In particular, X X0 = V A A ⊂ α∈V α A \ A V ( zα) XA. α∈V ∩ S QPROOF. The proof of this is left as an exercise.

4.1. Exercises.

EXERCISE VI.9. Let be a face of a polyhedron Rn. Assume that the origin is in aﬀ( ). Let Q P ⊂ Q α = k r α , where r > 0 and α for each j. Show that α aﬀ( ) if and only if α for j=1 j j j j ∈ P ∈ Q j ∈ Q each j. [Hint: if = In ( ), consider the value of ν, on , and α.] P Q ν P h ·i Q P EXERCISE VI.10. Prove corollary VI.12. [Hint: use proposition V.19.]

5. Toric varieties from polytopes

′ If ′ are finite subsets of Zn, then the natural projection P|A |−1 99K P|A|−1 restricts to a A ⊂ A rational map πA,A′ : XA′ 99K XA. This map is in general not defined everywhereon XA′ . For example, if ′ 2 1 = (0, 0), (1, 0) and = (0, 0), (1, 0), (0, 1) , then example VI.4 implies that XA′ = P , XA = P , A { } A { } ∼ ∼ and πA,A′ maps [x0 : x1 : x2] [x0 : x1], which is not defined at the point [0 : 0 : 1]. However, if in ′ 7→ n addition and have the same convex hull in R , then corollary VI.12 implies that πA,A′ is well-defined A A n everywhere on XA′ (exercise VI.11). If is the convex hull of in R , this observation shows that there P A is a natural morphism X n X , and for every positive integer k there is a natural morphism P∩Z → A X n X n n (k+1)P∩Z → (kP∩Z )+(P∩Z ) (For subsets of Rn and a positive integer d, in chapter V we defined d as a “dilation,” whereas in section VI.3S we defined it as the sum of d-copies of . This does not leadS to any conflict for the S case of convex polytopes - see remark V.2.) Proposition VI.6 implies that X(kP∩Zn)+(P∩Zn) is isomorphic to a subset of X n X n , so that the projection onto the first factor induces a morphism kP∩Z × P∩Z X n n X n . Consequently, there is a sequence of morphisms (kP∩Z )+(P∩Z ) → kP∩Z π3,4 π2,3 π1,2 π1 (36) X n X n X n X ··· −−→ 3P∩Z −−→ 2P∩Z −−→ P∩Z −→ A Proposition VI.13. For k sufficiently large, π : X n X n is an isomorphism. k,k+1 (k+1)P∩Z → kP∩Z PROOF. Let α ,...,α be the vertices of , so that kα , j = 0,...,s, are the vertices of k . 0 s P j P Corollary VI.12 implies that X n is the union of affine open sets X n U , j = 0,...,s. kP∩Z kP∩Z ∩ kαj n For each j, proposition VI.2 implies that the coordinate ring of XkP∩Z Ukαj is the semigroup alge- n ∩ n bra k[Sj,k], where Sj,k is the subsemigroup of Z generated by α kαj : α k Z . Since { − ∈ P ∩ } α kαj = (α + αj ) (k + 1)αj , it follows that Sj,k Sj,k+1, and πk,k+1 is simply the morphism − − ⊂ n induced by this inclusion. Note that each Sj,k is a subsemigroup of αj Z , where αj is the rational n C ∩ C convex polyhedral cone in R generated by αi αj : i = 0,...,s . Gordan’s lemma (lemma V.34) n { − } implies that αj Z is finitely generated. Due to exercise VI.12 below we may choose an integer K such C ∩ that for each j, Sj,k contains each of these generators for each k>K. For each k>K and each j, it n n n follows that Sj,k = αj Z is independent of k; consequently X(k+1)P∩ = XkP∩ . C ∩ Z ∼ Z 86 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

CA B P CB

CC A

FIGURE 2. Cones of regular functions on basic open subsets of XP . The dots in each cone mark generators of the semigroup of integral points in the cone.

n Given a convex integral polytope R , we write XP for the toric variety isomorphic to XkP∩ n P ⊆ Z 0 0 n for sufficiently large k and XP for the torus XkP∩Zn of XP ; we also write φP for the morphism φkP∩Z : ∗ n 0 (k ) XP defined as in (32). The arguments from the proof of proposition VI.13 show that XP is the → ′ union of open affine subsets Uα := XP Uα corresponding to vertices α of , and the coordinate ring of ′ ∩ P each Uα is generated by the monomials whose exponents belong to the cone α generated by α. See fig. 2 for the cones corresponding to the vertices of a triangle. The following propositionC summarizesP − some basic properties of XP . n n Proposition VI.14. Let be a finite subset of Z and be the convex hull of in R . Let GP be the n A P A subgroup of Z generated by pairwise differences of integral elements in the affine hull of and GA be P the subgroup of GP generated by the pairwise differences of elements from . n A (1) If k is such that GP is generated by α β : α, β k Z , then XP = XkP∩ n . { − ∈ P ∩ } ∼ Z (2) The dimension of XP is the same as the dimension of . If dim( )= n, then φP is an isomor-

∗ n 0 P P∗ n k phism between (k ) and XP ; i.e. XP is a compactification of ( ) . (3) There is a natural finite-to-one morphism φP,A : XP XA defined as the composition of the following maps: →

πkP∩Zn,kA φkA,A (37) X = X n X X P ∼ kP∩Z −−−−−−−→ kA −−−−→ A where k is as in assertion (1), πkP∩Zn,kA is the projection which drops coordinates correspond- n ing to elements in (k Z ) k , and φkA,A is the inverse of the Veronese isomorphism of P ∩ \ A degree k (see proposition VI.6). The degree of φP,A is the index of GA in GP . In particular, if 0 0 GA = GP , then φP,A restricts to an isomorphism between XP and XA. ′ ′ (4) There is an open cover U n of X such that the coordinate ring of each U is the { α}α∈P∩Z P α semigroup algebra k[Sα], where Sα is the semigroup of integral points in the convex polyhedral ′ ′ cone α generated by α α : α . (5) ThereC is a natural one-to-one{ − correspondence∈ P} between the faces of and the orbits of X0 on P P XP . For each face of , let OQ be the corresponding orbit and VQ be the closure of OQ in Q P 0 XP . Then VQ is naturally isomorphic to XQ, and the isomorphism identiﬁes OQ with XQ. 0 (6) In particular, XP XP is the union of the VQ for the facets of . ′ \ n Q P (7) Let be a convex integral polytope in R . Then XP+P′ is isomorphic to the closure in XP P ∗ n × X ′ of the image of the diagonal map (k ) X X ′ which sends x (φ (x), φ ′ (x)). P → P × P 7→ P P Moreover, XkP+lP′ ∼= XP+P′ for every pair of positive integers k,l. PROOF. Due to proposition VI.3 we may assume dim( ) = n and contains the origin. If k is as in assertion (1), then after another application of propositionP VI.3 we mayP assume that each standard n unit vector is in k Z , which immediately implies that φP is an isomorphism and proves the ﬁrst two assertions. The lastP assertion ∩ is a straightforward corollary of propositions VI.6 and VI.13. The remaining statements follow from propositions VI.1 and VI.2 and theorem VI.11. We leave it as an exercise to complete the proof. Example VI.15. Let := 0, 2 Z. Since x 2x maps 0, 1 onto , proposition VI.3 and exam- A 1{ } ⊂ 7→ { } A ple VI.7 imply that XA = P . If = conv( ) R, then GP = Z and assertion (1) of proposition VI.14 ∼ P A ⊂ 5. TORIC VARIETIES FROM POLYTOPES 87

2 ∗ 2 is satisﬁed with k = 1, i.e. XP is the closure in P of the image of k under the map x [1 : x : x ]. 1 7→ Example VI.7 implies that XP is also isomorphic to P . The map φP,A : XP XA is the restriction to → XP of the projection [z0 : z1 : z2] [z0 : z2], and therefore on the level of the tori it is simply the map 2 → x x . Note that deg(φP,A)=2 is also the index of GA =2Z in Z = GP . 7→ Example VI.16. Let n be the n-dimensional simplex in Rn with vertices at the origin and at the elements S n n n n n n of the standard unit basis of R , and let := [0, 1] R . Then GSn = GKn = Z , and both and n satisfy assertion (1) of propositionK VI.14 with k⊂= 1. Therefore it follows from examples VI.4S K n 1 n and VI.5 that XSn ∼= P and XKn ∼= (P ) . Example VI.17. Consider ′ from ﬁg. 1b. The convex hull ′ of is a translation of 9 2, so that A P A 2 S propositions VI.6 and VI.14 and example VI.16 imply that XP′ = XS2 = P . Note that GP′ = GA′ = 2 ∼ ∼ Z , i.e. the map φP′,A′ : XP′ XA′ has degree one. However, it is not an isomorphism. Indeed,

→ 8 9 9 8 k proposition VI.2 implies that the coordinate ring of XA′ UA′ is k[x , x ,y] = [u,v,w]/ u v ∩ 3 ∼ 9 h 8− i (exercise VI.14). Therefore X ′ U ′ is isomorphic to the hypersurface in k deﬁned by u v = 0, A ∩ A − which is singular at all points of the w-axis, and φ ′ ′ : X ′ X ′ is a desingularization of X ′ . P ,A P → A A

Even though XP is by definition isomorphic to XkP∩Zn for k 1, in each of the preceding examples it suffices to take k =1. In general it suffices to take k dim( ) ≫1, see e.g. [CLS11, Theorem 2.2.12]. In particular, to find examples for which one needs to≥ take k >P 1−requires polytopes with dimensions at least 3.

5.1. Exercises.

EXERCISE VI.11. Let ′ be ﬁnite subsets of Zn such that conv( ) = conv( ′). Show that A ⊂ A ′ A A πA,A′ is well-deﬁned everywhere on XA′ . [Hint: and have the same set of vertices. Use corollary VI.12.] A A

EXERCISE VI.12. In the notation of the proof of proposition VI.13, show that for each β αj , there is K 0 such that β k kα for each k K. ∈C ≥ ∈ P− j ≥ EXERCISE VI.13. Complete the proof of proposition VI.14.

8 9 9 8 9 k EXERCISE VI.14. Show that k[x , x ,y] = [u,v,w]/ u v . Prove that the hypersurface V (u 8 3 ∼ h − i − v ) in k is singular at all points of the w-axis.

n EXERCISE VI.15. In the set up of ﬁg. 2, let SA (respectively, SB,SC) be the subsemigroup of Z consisting of integral elements in (respectively, , ). CA CB CC (1) Show that SA is generated as a semigroup by (1, 1), (2, 1); SB is generated as a semigroup by ( 1, 1), ( 1, 0), ( 2, 1); SC is generated as a semigroup by ( 1, 1), (0, 1), (1, 1).

− − − − 2 3 3 − − − −2 3

k k (2) Deduce that XP UA = k , XP UB = V (uv w ) , and XP UC = V (uv w ) .

∩ ∼ 2∩ ∼ − ⊂ −1 −1 ∩−2 −∼1 − ⊂

k k k k [Hint: k[XP UA] = [xy, x y], [XP UB] = [x y, x , x y ], [XP UC ] = −1 −1 −∩1 −1 ∩ ∩ k[x y ,y , xy ].] (3) Conclude that XP has precisely two singular points.

EXERCISE VI.16. Let be a convex integral polytope in Rn and α, α′ be vertices of . P P α′−α (1) Show that x is an invertible regular function on Uα Uα′ XP . ∩ ′ ∩ (2) Let be the smallest face of containing both α and α , and α be the cone generated by β Qα : β . Show that α′P α is in the relative interior of Q. { − ∈ Q} − Qα (3) Let H be the linear subspace of Rn spanned by β α : β . For each γ H Zn, show γ { − ∈ Q} ∈ ∩ n that x is an invertible regular function on Uα Uα′ XP . [Hint: Choose β1,...,βk Z such that β α, j = 1,...,k, generate ∩as a cone.∩ Use exercise V.30 and the∈ preceding Q ∩ j − Qα assertions to show that for each j, xβj −α is a regular function on U X which does not vanish α ∩ P at any point of Uα Uα′ XP .] ∩ ∩ 0 (4) Deduce that if = , then U U ′ X = X . Q P α ∩ α ∩ P P 88 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

6. Nonsingularity in codimension one on XP

Example VI.17 and exercise VI.15 above show that toric varieties XP from polytopes might be singular. However, we will see in this section that the XP is nonsingular outside a subvariety of dimension at most dim(X ) 2, i.e. X is “nonsingular in codimension one.” Note that this is in general not true P − P for varieties XA (see example VI.17 above). We continue to use the notation of section VI.5. Let be a facet of . We will show that X is nonsingular at all points of O . Due to proposition VI.3 weQ may P P Q assume without loss of generality that is full dimensional, i.e. is an n-dimensional polytope in Rn. P P n n Let ν be the primitive inner normal (see remark V.44) to . Let Zν≥0 := β Z : ν,β 0 and n n Q n { ∈ h i ≥ } Z ⊥ := β Z : ν,β = 0 . Choose an arbitrary element αν Z such that ν, αν = 1. We write ν { ∈ h i } ∈ h i Z≥0 αν := kαν : k Z, k 0 . The following is a straightforward implication of lemma B.43. h i { ∈ ≥ } n Lemma VI.18. There is a basis of Z of the form β1,...,βn, where βn = αν and β1,...,βn−1 constitute a basis of n . In particular, n n−1, and as a semigroup n is isomorphic to n + α Zν⊥ Zν⊥ ∼= Z Zν≥0 Zν⊥ Z≥0 ν ∼= n−1 h i Z Z≥0. × Proposition VI.19. Let U := X0 O . Q P ∪ Q (1) UQ is an open afﬁne neighborhood of OQ in XP .

β n n−1 ∗ n−1

k k k k (2) k . In particular, . [UQ]= [x : β Zν≥0] ∼= [Z Z≥0] UQ ∼= ( )

αν ∈ × αν n ×

k k (3) and k . In particular, the embedding OQ = V (x ) UQ [OQ] ∼= [UQ]/ x ∼= [Zν⊥ ]

⊂ ∗hn−1i ∗ n−1

k k . ( ) ֒ O ֒ U is isomorphic to the embedding (k ) 0 Q → Q ×{ } → × n PROOF. Let := k Z , where k is large enough so that XP ∼= XA and there is an integral element α in the relativeA interiorP ∩of k . Since is the smallest face of containing α (proposition V.14), 0 Q Q P 0 proposition VI.10 implies that UQ = XP Uα0 , which proves assertion (1). Let β1,...,βn = αν be a n ∩ n basis of Z as in lemma VI.18. Choosing a large enough k we can also ensure that α0 + βj k Z for each j = 1,...,n (exercise V.34). It is then straightforward to check that the semigroup gen∈ eratedP ∩ by n α0 is precisely Zν≥0. Assertions (2) and (3) then follow from proposition VI.14 and lemma VI.18 in A−a straightforward manner - we leave the proof as an exercise.

Corollary VI.20 (Nonsingularity of XP in codimension one). The set Sing(XP ) of singular points of XP ′ is contained in O ′ , where are faces of of dimension n 2. In particular, dim(Sing(X )) i Qi i P n 2. Q P ≤ − ≤ − S PROOF. XP OQ′ is the union of UQ over all facets of , which is nonsingular due to propo- \ i i Q P sition VI.19. Since dim( OQ′ )= n 2, the result follows. S i i − S C αν =(−1, 0) Q B NP(g) P NP(Tαν (g))

ν =(−1, −1)

αν = (0, −1) A

NP(Tαν (g))

FIGURE 3. Different choices for T (g) when g =2y2 3x2y3 +7x4y4 6x2y4 αν − −

β −1 −1 Let ν and αν be as in proposition VI.19. Given g = β cβx k[x1, x1 ,...,xn, xn ], the support n ∈ Supp(g) of g is the set of all β Z such that cβ = 0, and the Newton polytope NP(g) of g is the ∈ 6 P convex hull of Supp(g). Let m := minSupp(g)(ν) = minNP(g)(ν). Choose an arbitrary isomorphism n n−1 −mαν ′ ψν (β−mαν ) ψ : ⊥ = . Define T (g) := x g and In (g) := c x (the “T ” in ν Zν ∼ Z αν αν ,ψν hν,βi=m β T ( ) is supposed to imply “translation,” and “In” in In′ ( ) is to suggest “initial form”). See fig. 3 αν · αν ,ψν · P 7. EXTENDING CLOSED SUBSCHEMES OF THE TORUS TO XP 89 for an example with from fig. 2. The following result is an immediate corollary of proposition VI.19. Its proof is left as an exercise.P

Corollary VI.21. (1) Tαν (g) is a regular function on UQ for each Laurent polynomial g. ′ ∗ ∗ n−1 (2) The correspondence T (g) In (g) induces an isomorphism ψ : (k ) = O . αν OQ αν ,ψν ν ∼ Q ′ | n 7→ ′ m(α′ −α ) ′ ν ν (3) If α is another element in Z such that ν, α = 1, then Tαν (g)/Tα (g) = x is ν h ν i ν invertible on UQ.

A basic property of the varieties XP is that they are “normal,” and nonsingularity in codimension one follows from normality. In this book we do not treat this notion - see [Ful93, Section 2.1] or [CLS11, Section 2.4] for an exposition of this and other fundamental properties of XP including the following result n (which we do not use): assume R is full dimensional. Then XP is nonsingular if and only if both of the following are true for everyP vertex ⊂ α of : P α has precisely n edges, and • C n the primitive integral elements of the edges of α form a basis of Z . • C 6.1. Exercises.

EXERCISE VI.17. Complete the proof of proposition VI.19.

EXERCISE VI.18. Prove corollary VI.21.

7. Extending closed subschemes of the torus to XP Recall (from section IV.2) that a closed subscheme V of a variety X is essentially a Zariski closed subset V ′ of a variety together with a sheaf of ideals such that V ′ is precisely the set of zeroes of elements in . If X¯ is a variety containing X as a ZariskiI open subset, and V¯ is a closed subscheme of X¯, we say that VI¯ extends V if the scheme-theoretic intersection V¯ X is precisely V . In this section we will ∗ n ¯ ∩ study the case that X = (k ) and X is the toric variety XP corresponding to an n-dimensional convex integral polytope . We are specially interested in the case that P (a) both V and V¯ are Cartier divisors (see section IV.2.4), and (b) Supp(V¯ ) is the closure of Supp(V ).

−1 −1 ∗ n k Each Laurent polynomial g k[x1, x1 ,...,xn, xn ] defines a Cartier divisor V (g) on ( ) (in fact it is not hard to see, using the fact∈ that the ring of Laurent polynomials is a UFD, that every Cartier divisor on ∗ n (k ) is of the form V (g) for some Laurent polynomial g - but we will not use it). There are many ways to extend V (g) to X , e.g. if NP(g) , then exercise VI.19 below prescribes a way to extend V (g) to a P ⊂P Cartier divisor on XP which satisfies property (b) if and only if NP(g) intersects each facet of . However, ∗ n P we will shortly see that for a given , there are Cartier divisors on (k ) which can not be extended to P 1 1 XP in a way to satisfy property (b). Then we will define an open subset XP of XP such that XP XP ∗ n 1 \ has dimension n 2, and every Cartier divisor on (k ) does admit extensions to XP satisfying both properties (a) and≤ (b)−.

CA CA B NP(g) P CB CB

NP(h) A CC CC

FIGURE 4. Extending Cartier divisors in dimension two 90 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

Example VI.22. Consider and g from fig. 3. Theorem VI.11 and proposition VI.14 imply that XP U U U , and for eachP P A, B, C , the coordinate ring of U X consists of Laurent⊂ A ∪ B ∪ C ∈ { } P ∩ P polynomials supported at the cone P ; consider the (unique) translation of NP(g) which is contained in and touches both sides of (seeC the middle panel of fig. 4). More precisely, let CP CP 2 3x2y +7x4y2 6x2y2 if P = A, −−4 −2 2 − 4 2 2 2 gP := x y (2 3x y +7x y 6x y ) if P = B,  − − x−3y−3(2 3x2y +7x4y2 6x2y2) if P = C. − − Then it is straightforward to check that the pairs (UP XP ,gP ), P A, B, C , defines a Cartier divisor D ∩ ∈{ } ∗ 2 on XP such that D extends the Cartier divisor V (g) and Supp(D) is the closure in XP of V (g) (k ) (see exercise VI.20 for a more general result). On the other hand, if h is any Laurent polynomial⊂ with NP(h) as in the left panel of fig. 4, then the translation of NP(h) that is contained in and touches CB both sides of B is not integral - see the right panel of fig. 4. It follows that if E is any extension of ∗ 2C V (h) (k ) to XP , then Supp(E) must contain either VAB or VBC ; in particular, E does not satisfy property⊂ (b) of the extension. Supp(E) is greater than the closure in X of V (h) X (exercise VI.21). P ⊂ P

2 2

1 1 NP(g) 2 2 − 2 − 2 1 1 1 1 P 2 − 2 − − 2 − 2

(A) Crossections of P and NP(g) (B) P (C) NP(g)

FIGURE 5. Obstruction to extension of Cartier divisors in dimension 3 ≥

Example VI.23. Let be the polytope in R3 with vertices (2, 2, 0), ( 2, 2, 0), ( 2, 2, 0), (2, 2, 0), P − − − −

(0, 0, 2), and let g k[x,y,z] be a polynomial such that NP(g) is the “cuboid” with vertices (2, 1, 0), ( 2, 1, 0), ( 2, 1∈, 0), (2, 1, 0), (2, 1, 2), ( 2, 1, 2), ( 2, 1, 2), (2, 1, 2) (see ﬁg. 5). It is straight- − − − − − − − − forward to check that for each m 1, if a translation of m NP(g) is contained in P , then can not ≥ Q m ∗ 3 C Q touch all the facets of P , and therefore any extension to XP of V (g ) (k ) by a Cartier divisor fails property (b) (exercise VI.22C ). ⊂

∗ n In general, to find a Cartier divisor on X which extends V (g) (k ) and satisfies property (b) P ⊂ requires a “modification” of XP by finding a “common refinement” of the normal fans of and NP(g). This leads to beautiful combinatorial geometry, but we will not get into this. We will ratherP find an open 1 ∗ n 1 subset XP of XP so that every Cartier divisor of (k ) can be extended to a Cartier divisor on XP and the extension also satisfies property (b).

1 n 7.1. The subset XP . Let be an n-dimensional convex integral polytope in R . Proposition VI.14 ∗Pn 0 implies that we can identify (k ) with XP . Denote the facets of by 1,..., s, and the faces of of ′ ′ P Q Q P dimension n 2 by ,..., ′ . Deﬁne ≤ − Q1 Qs s s s′ s′ 1 0 ′ ′ XP := XP OQi = UQi = XP OQ = XP VQ ∪ \ i′ \ i′ i=1 i=1 ′ ′ [ [ i[=1 i[=1 where the equalities are implications of theorem VI.11 and propositions VI.14 and VI.19. In particular, it 1 follows that XP is Zariski open in XP , and it is the union of torus orbits in XP of “codimension smaller 1 than one” (this is the motivation for the “1” in the notation XP ). Let νi be the primitive inner normal to i, −1 −1 Q i =1,...,s. Pick α such that ν , α =1. Given g k[x , x ,...,x , x ], corollary VI.21 implies νi h i νi i ∈ 1 1 n n 7. EXTENDING CLOSED SUBSCHEMES OF THE TORUS TO XP 91

0 ∗ n that Tα (g) is a regular function on UQ for each i. Since UQ UQ = X = (k ) for i = j (propo- νi i i ∩ j P ∼ 6 sition VI.19) and since T (g)/T (g) is a monomial in x ,...,x , it follows that (U ,T (g)) ανi ανj 1 n Qi ανi i 1 1 1 { } deﬁnes a Cartier divisor VP (g) on XP . In exercise VI.24 you will check that VP (g) is an extension to 1 ∗ n XP of the Cartier divisor V (g) of (k ) , and it also satisﬁes property (b). This construction can be car-

∗ n −1 −1 k ried out for arbitrary closed subschemes of (k ) . Indeed, given g ,...,g [x , x ,...,x , x , 1 k ∈ 1 1 n n for each i, deﬁne I to be the ideal of k[U ] generated by T (g ), j = 1,...,k. Exercise VI.25 Qi Qi ανi j implies that the ideals IQi can be glued over their intersection to form a sheaf of ideals of XP . We 1 I O write VP (g1,...,gk) for the corresponding closed subscheme V ( ) of XP . Assertion (2) of exercise VI.25 1 I implies that VP (g1,...,gk) does not depend on the choice of the ανi . 1 Example VI.24. Exercise VI.24 implies that if k = 1, then VP (g1,...,gk) depends only on the ideal I −1 −1 of k[x1, x1 ,...,xn, xn ] generated by g1,...,gk. We now show that this is in general false if k > 1. 2 2 Let be the triangle in R with vertices (0, 0), (1, 0), (0, 1), so that XP = P (example VI.16), and with P ∼2 0 2 respect to corresponding homogeneous coordinates [z0,0 : z1,0 : z0,1] on P , XP = P V (z0,0z1,0z0,1) 1 \ and XP = XP [0 : 0 : 1], [0 : 1 : 0], [1 : 0 : 0] . Note that (x, y) := (z1,0/z0,0,z0,1/z0,0) is a \{ 0 } system of coordinates on XP . Let g1 := x 1 and g2 := y 1. It is straightforward to check that 1 1 0 − 0 1 − VP (g1,g2) (XP XP )= , and as a subscheme of XP , VP (g1,g2)= V (g1,g2). Now let h1 := x y ∩ \ ∅ 2 −1 −1 − and h2 := x + y 2 + (x y) , so that the ideal generated by h1,h2 in k[x, x ,y,y ] is the same as − − 1 1 0 the ideal generated by g1,g2. It can be checked that VP (h1,h2) contains the point [0 : 1 : 1] XP XP , so that V 1(h ,h ) = V 1(g ,g ). ∈ \ P 1 2 6 P 1 2 n Let be a facet of with primitive inner normal ν, and αν be an arbitrary element of Z such that Q P ν, αν = 1. Proposition VI.19 implies that the ideal of k[UQ] consisting of elements vanishing on OQ h i αν αν is generated by x . Therefore OQ is precisely the support of the closed subscheme Zαν := V (x ) of 1 UQ. Since OQ is Zariski closed in XP (exercise VI.23), it follows that Zαν is in fact a closed subscheme 1 of XP . We now determine the embedded isomorphism (see section IV.2.2) type of the “scheme-theoretic 1 ∗ ∗ n−1 intersection” V (g ,...,g ) Z . Let ψ : (k ) = O be as in corollary VI.21. P 1 k ∩ αν ν ∼ Q 1 Proposition VI.25. As a closed subscheme of OQ, the scheme-theoretic intersection VP (g1,...,gk) Zαν ∗ ′ ′ ∗ n−∩1 is embedded isomorphic via ψν to the closed subscheme V (Inβν ,ψν (g1),..., Inβν ,ψν (gk)) of (k ) .

1 αν PROOF. The ideal IQ of VP (g1,...,gk) Zαν is generated in k[UQ] by x ,Tαν (g1),...,Tαν (gk). ∩ αν Proposition VI.19 and corollary VI.21 imply that k[UQ]/ x ,Tαν (g1),...,Tαν (gk) is isomorphic via

∗ ∗ n−1 ′ ′ h i k ψ to k[( ) ]/ In (g ),..., In (g ) , which directly implies the result. ν h βν ,ψν 1 βν ,ψν k i 7.2. Exercises. EXERCISE VI.19. Let be an n-dimensional convex integral polytope in Rn and g be a Laurent polynomial with NP(g) P. ⊆P −α (1) Show that the collection (Uα XP , x g), where α varies over the vertices of , deﬁnes a ∩ ∗ n P Cartier divisor D on XP such that D extends the Cartier divisor V (g) of (k ) . [Hint: use exercise VI.16.] (2) Show that the following are equivalent: ∗ n (a) Supp(D) is the closure in XP of V (g) (k ) . (b) NP(g) = for each facet of . ⊂ ∩ Q 6 ∅ Q P (3) Assume in addition that := Zn satisﬁes the hypothesis of assertion (1) of proposition VI.9 n A P ∩ so that U0 XP = k . Show that the following are equivalent: ∩ ∼ n (a) Supp(D) is the closure in XP of V (g) k . (b) NP(g) = for each facet of which⊂ is not contained in any coordinate hyperplane ∩ Q 6 ∅ Q P of Rn. EXERCISE VI.20. Let be an n-dimensional convex integral polytope in Rn. For each vertex α of n P ′ ′ −1 −1 , let α be the cone in R generated by α α : α . Assume there is g k[x1, x1 ,...,xn, xn ] P C { − n ∈ P} ∈ such that for each vertex α of , there is βα Z such that βα + NP(g) α and βα + NP(g) touches every edge of . P ∈ ⊂ C Cα (1) Let α, α′ be vertices of , and be the smallest face of containing both α and α′. Let H be Pn Q P the linear subspace of R spanned by β α : β . Show that βα βα′ H. [Hint: use { − ∈ Q} − ∈ 92 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

exercise V.32 to reduce to the case that there is an edge of connecting α and α′. In that case P ′ show that βα βα′ is on the line through the origin and α α .] − βα − (2) Deduce that the collection of pairs (Uα XP , x g), where α varies over the vertices of , ∩ ∗ n P deﬁnes a Cartier divisor D on XP such that D extends the Cartier divisor V (g) of (k ) , and ∗ n Supp(D) is the closure in XP of V (g) (k ) . [Hint: use exercise VI.16.] (3) If is simplicial, i.e. each vertex of is⊂ connected to precisely n distinct edges, then show that

P −1 −P1 m ∗ n k for every g k[x1, x1 ,...,xn, xn ], there is m 1 such that V (g ) ( ) extends to a ∈ ≥ m⊂ ∗ n Cartier divisor D on X such that Supp(D) is the closure in X of V (g ) (k ) . P P ⊂ EXERCISE VI.21. Let and h be as in example VI.22. If D is any Cartier divisor on XP which m ∗ 2 P extends V (h ) (k ) , where m is a positive integer, then show that D satisﬁes property (b) if and only if m is a multiple⊂ of 3. EXERCISE VI.22. Prove the claims made in example VI.23. EXERCISE VI.23. Let be a facet of an n-dimensional convex integral polytope in Rn. Show that Q 1 P OQ is a closed subvariety of XP . −1 −1 EXERCISE VI.24. Let g k[x1, x1 ,...,xn, xn ] and be a convex integral polytope of dimension n. Show that ∈ P 1 1 (1) The Cartier divisor VP (g) of XP does not depend on the choice of the ανi . 1 1 (2) If g/h is a monomial in x1,...,xn, then VP (g)= VP (h). 1 1 ∗ n (3) Supp(V (g)) is the closure in X of V (g) (k ) . P P ⊂ −1 −1 EXERCISE VI.25. Let g1,...,gk k[x1, x1 ,...,xn, xn ] and be a convex integral polytope of dimension n. ∈ P n (1) Let 1, 2 be facets of , νi be the primitive inner normal to i, and ανi Z be such Q Q P Q ∈ that νi, ανi = 1. Show that for each x UQ1 UQ2 the ideal of XP ,x generated by T h(g ),...,Ti (g ) is the same as the ideal∈ generated∩ by T (g ),...,TO (g ). αν1 1 αν1 k αν2 1 αν2 k (2) If g /h is a monomial in x ,...,x for each j, then show that for each i and each x U , j j 1 n ∈ Qi the ideal of X ,x generated by Tα (g1),...,Tα (gk) is the same as the ideal generated by O P νi νi T (h ),...,T (h ). ανi 1 ανi k EXERCISE VI.26. Verify the claims made in example VI.24.

8. Branches of curves on the torus 8.1. Branch of a curve on a variety. Let C be a curve, i.e. a pure dimension one variety. Fix a desingularization π : C′ C of C and a nonsingular compactification C¯′ of C′. A branch of C is the germ of a point in C¯′.→ Equivalently, consider the equivalence relation on the collection of pairs (Z,z): z C¯′ and Z is an open neighborhood of z in C′ defined as follows:∼ (Z,z) (Z′,z′) if and only{ if z =∈z′. Then a branch of C is a equivalence class} of . Let X be a variety containing∼ C as a subvariety. Let X¯ be an arbitrary projective compactification of X∼ and C¯ be the closure of C in X¯. Then π extends to a map C¯′ C¯ (exercise III.66), which we also denote by π. If B := (Z,z) is a branchof C and y := π(z), we say that→y is the center of B on X¯, or equivalently, B is a branch of C at y. If y X, we say that (with respect to X) B is a branch at infinity, or that it is centered at infinity. Since C¯′ is6∈ nonsingular, is a discrete valuation ring (exercise III.64), and correspondsto a unique discrete valuation ord ( ) on OZ,z z · the field of fractions of Z,z (assertion 5 of proposition B.8). If f is a regular function on a neighborhood O ∗ of z on X, then we write ordz(f) for ordz(π (f)). ∗ n 8.2. Weights of a branch on the torus. Fix a system of coordinates (x1,...,xn) on (k ) . Let ∗ n B = (Z,z) be a branch of a curve on (k ) . The weight of xj corresponding to B is ordz(xj Z ). By νB n ∗ | we denote the element in (R ) with coordinates (ordz(x1),..., ordz(xn)) with respect to the basis dual to the standard basis of Rn. Lemma VI.26. (1) For each α Zn, xα restricts to a well-defined rational function on Z and α ∈ ordz(x )= νB, α . h i ∗ n (2) ν =0 if and only if B is centered at infinity with respect to (k ) . B 6 PROOF. This is left as an exercise. 9. POINTS AT INFINITY ON TORIC VARIETIES 93

8.3. Weighted order on Laurent polynomials. Let ν be an integral element ν of (Rn)∗. The weighted order corresponding to ν is an integer valued map, which by an abuse of notation we also denote α by ν, on the ring of Laurent polynomials defined as follows: given a Laurent polynomial f = α cαx , ν(f) := min ν, α : c =0 {h i α 6 } P In particular, if f is the zero polynomial, then ν(f) is defined to be . The initial form Inν (f) of f with α ∞ n respect to ν is the sum of all cαx such that ν, α = ν(f). Given a subset of R , we say that f is supported at if Supp(f) , and we writeh ( i) for the set of all LaurentS polynomials supported at S ⊆ S L S . In the case that Zn is a finite set, we equip ( ) with the Zariski topology by identifying it with S|S∩Zn| S ∩ L S k via the map α cαx (cα)α∈S∩Zn n 7→ α∈S∩XZ The result we will now prove ties these notions with those from section VI.8.2; we will encounter many of ∗ n its variants in the forthcoming chapters. Let B = (Z,z) be a branch of a curve on (k ) , and νB be as in section VI.8.2. By an abuse of notation, we denote by νB also the weighted order corresponding to νB. Proposition VI.27. Assume Zn has finitely many elements. There is a nonempty Zariski open subset U of ( ) such that ord (f)=S ∩ν (f) for each f U. More precisely, define L S z B ∈ ∗( ) := f ( ) : ord (f)= ν (f) L S { ∈ L S z B } Then ∗( ) is a constructible subset of ( ) of dimension . L S L S |S| PROOF. Pick a parameter ρ of Z,z. Then ordz(ρ)=1. For each i =1,...,n, if mi := νB(xi), then O mi ∗ there is a representation of the form xi = ciρ + hi where ci k , and hi Z,z such that ordz(hi) > n ∗ ∈ ∗ ∈ O mi (exercise III.65). For each Z , let B( ) := f ( ) : Supp(InνB (f)) InνB ( ) . It is straightforward to check thatA⊆S∩ L A { ∈ L A ⊂ A } n ∗ αi ( )= (c ) n : c =0 if α , (c ) c =0 LB A { α α∈S∩Z α 6∈ A i α 6 } α∈In (A) i=1 XνB Y which implies that ∗ ( ) is a nonempty open subset of a closed subset of ( ). Since ∗( ) is the union LB A L S L S of ∗ ( ) over all subsets of Zn, it follows that it is a constructible subset of ( ). The remaining LB A A S ∩ L S parts of the proposition follows from taking = Zn. A S ∩ 8.4. Exercises.

EXERCISE VI.27. Consider the singular curves C1, C2 from example III.31. Show that C1 has a single branch at its singular point, whereas C has two branches at its singular point. [Hint: t (t3,t2) induces 2 7→ a desingularization of C1. For C2 use exercise III.67.] EXERCISE VI.28. Prove lemma VI.26.

9. Points at inﬁnity on toric varieties

n 9.1. Centers of branches at inﬁnity on the torus. Let := α0,...,αN be a ﬁnite subset of Z

∗ n A {∗ n 0 } k and B = (Z,z) be a branch centered at infinity on (k ) . If φA : ( ) XA is the map from (32), then 0 → φA(B) is a branch centered at infinity on XA. We now determine the torus orbit of XA that contains the n ∗ center oB of φA(B) on XA; we will see that this orbitis completely determinedby the element νB (R ) defined in section VI.8.2. Let be the convex hull of and be a face of . Then the convex hull∈ of P A B A Q is a face of . As in section V.4 we write ΣP for the normal fan of , and denote the normal cone of B P 0 P Q by σQ and the relative interior of σQ by σQ. 0 Proposition VI.28. oB OB if and only if νB σQ. In particular, if is a facet of , then oB OB if and only if ν is a positive∈ multiple of the primitive∈ inner normal to . B A ∈ B Q α−β PROOF. Pick β . Theorem VI.11 implies that x = zα/zβ is a regular function on OB for ∈ B α−β each α . It also implies that oB OB if and only if the following holds: “x oB = 0 if and only if α ∈.” A Due to assertion (1) of lemma∈ VI.26 the latter condition is equivalent to| the6 condition that In (∈)= B , which is in turn equivalent to the condition that In ( )= , as required. νB A B νB P Q 94 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

σ0 C AB

B σ0 P 0 A σB

0 σC A 0 0 σBC σCA

FIGURE 6. Normal fan of P

We now describe the coordinates of oB. As in section VI.8.3, we write νB also for the weighted order −1 −1 on k[x , x ,...,x , x ] induced by ν . Fix an arbitrary element ρ such that ord (ρ )=1. 1 1 n n B B ∈ OZ,z z B|Z We say that ρB is a parameter of B. Deﬁne

xj ∗ InB(xj ) := k , (ρ )νB (xj ) ∈ (38) B z ∗ n In(B) := (InB(x1),..., InB(xn)) (k ) ∈ That InB(xj ) and In(B) are well deﬁned follows from exercise III.64. Note that In(B) depends on the choice of ρB. In all cases considered in this book, whenever a branch B of a curve is considered, a corresponding parameter ρB is assumed to be ﬁxed from the beginning. Let be the face of such that B ∗ n A 0 oB OB (proposition VI.28 shows that is uniquely determined by νB). Let φB : (k ) XB be the ∈ B ∗ n → map from (32). Theorem VI.11 implies that we may think of φB as a map from (k ) to OB XA, simply N ⊂ by adjoining a zero in place of each coordinate zα on P such that α . 6∈ B Proposition VI.29. oB = φB(In(B)). In particular, φB(In(B)) does not depend on the choice of ρB (even though In(B) does).

α−β α−β PROOF. Pick α, β . Proposition VI.28 implies that (zα/zβ) oB = x oB = (In(B)) . The result then follows immediately∈ B from theorem VI.11. | |

∗ n 9.2. Closure of subvarieties of the torus. Let W be a closed subvariety of (k ) deﬁned by Laurent

∗ n n ∗ n 0 k polynomials f1,...,fm in (k ) . Let be a ﬁnite subset of Z and φA : ( ) XA be the map from ¯ ′ A ′ 0 → (32). Write W for the closure in XA of W := φA(W ) XA. In this section we give a partial description of the points in W¯ ′. ⊂

Lemma VI.30. Let B be a branch of a curve contained in W . Then In(B) is a common zero of InνB (fi), i =1,...,m.

PROOF. Let B = (Z,z). Pick a parameter ρB of B. Exercises III.64 and III.65 on page 34 and on νB (xj ) page 35 imply that for each j, xj /ρB is a regular function on a neighborhood of z on Z, and it can be νB(fi) expressed as InB(xj )+ gj , where gj Z,z, ordz(gj ) > 0. Consequently, for each i, fi/ρB can be ∈ O νB (fi) expressed in Z,z as InνB (fi)(In(B)) + hi, where hi Z,z, ordz(hi) > 0. Since fi/ρB maps to the zero elementO in , it follows that In (f )(In(B))=0∈ O , as required. OZ,z νB i Let ν be an integral element of (Rn)∗; we write ν also for the corresponding weighted order on

−1 −1 ∗ n k k[x1, x1 ,...,xn, xn ] and deﬁne Wν (f1,...,fm) := V (Inν (f1),..., Inν (fm)) ( ) . Let be a

∗ n 0 ⊂ B ∗ n k face of . As in proposition VI.29, we regard the map φB : (k ) XB from (32) as a map from ( ) A → 0 to OB XA. As in proposition VI.28 we write for the convex hull of and σQ for the relative interior of the corresponding⊂ cone of the normal fan of theQ convex hull of . B A ¯ ′ Corollary VI.31. W OB ν∈σ0 φB(Wν (f1,...,fm)). ∩ ⊂ Q S 10. ∗WEIGHTED PROJECTIVE SPACES 95

¯ ′ ′ 0 PROOF. Let w W OB. If w W , then we must have that = . In that case 0 σQ. ∈ ∩ ∈ B ′A ∈ Since W0(f1,...,fm) = W , it follows that w φB(W0(f1,...,fm)) = W , as required. So assume w W¯ ′ W ′. Then there is an irreducible curve∈ C′ W ′ such that w is in the closure of C′ (propo- ∈ \ −1 ⊂ sition III.24). Pick a branch B = (Z,z) of φA (C) such that z w under the morphism induced by 0 7→ φA. Proposition VI.28 implies that νB σQ and proposition VI.29 implies that w = φB(In(B)). Since In(B) W (f ,...,f ) (lemma VI.30∈), the result follows. ∈ νB 1 m n ∗ For each i =1,...,m, there are only ﬁnitely many possibilities for Inν (fi) as ν varies over (R ) . It follows that the union in the statement of corollary VI.31 can be regarded as being over a ﬁnite collection of ν (Rn)∗. Exercise VI.29 shows that the containment of corollary VI.31 is in general proper. However, if W∈is a hypersurface (i.e. m =1), then exercise VI.30 shows that corollary VI.31 holds with “=” in place of . ⊂ 9.3. Exercises. 2 EXERCISE VI.29. Let := (0, 0), (1, 0), (0, 1) R and h1,h2 be as in example VI.24. Let A {∗ 2 } ⊂ W := V (h1,h2) = (1, 1) (k ) and := (1, 0), (0, 1) . Note that := conv( ) is an edge of := conv( ). { } ∈ B { } Q B P A ′ (1) Show that W¯ OB = . (2) Let ν be the primitive∩ inner∅ normal to . Show that W (h ,h ) = . Q ν 1 2 6 ∅

−1 −1 ∗ n k EXERCISE VI.30. Let f k[x1, x1 ,...,xn, xn ] and W := V (f) ( ) . Show that in the ¯ ′∈ ⊂ notation of corollary VI.31, W OB = ν∈σ0 φB(Wν (f)) for each face of . [Hint: use exercise VI.24 ∩ Q B A to prove it in the case that is a facet of . Then use induction on dim( ).] Q SP P 10. ∗Weighted projective spaces 2 Recall from section II.3 that the n-dimensional projective space is the space of straight lines through n+1 the origin in k . A weighted projective space is constructed in the same way, using weighted rational n+1 ∗ curves in place of straight lines. Let ω be an integral element of (R ) with coordinates (ω0,...,ωn) n+1 with respect to the basis dual to the standard basis of R . Assume each ωj is positive. The corresponding n n n+1 weighted projective space, which we denote by P (ω) or P (ω0,...,ωn), is the set of curves in k of

ω0 ωn n+1 k the form Ca := (a0t ,...,ant ): t k , where a := (a0,...,an) 0 . The weighted { ∈ } ∈ n\{ } homogeneous coordinates of Ca are [a0 : : an]. Note that the projective space P is the special case of ··· Pn(ω) for ω = (1,..., 1). Like Pn, each Pn(ω) can be given the structure of a complete algebraic variety. In this section we give two (equivalent) realizations of Pn(ω) as a toric variety.

νAB =(−1, 2) B P

νBC =(−1, −1) νCA = (1, −1)

A νBC + 2νAB + 3νBC = 0

2 FIGURE 7. XP ∼= P (1, 2, 3)

n n n ∗ 10.1. P (ω) via polytopes in R . Pick integral elements ν0,...,νn of (R ) such that ν0,...,νn n ∗ n n span (Z ) , and ωj νj = 0. Let be an n-dimensional integral simplex in R such that its inner j=0 P facet normals are ν0,...,νn; note that is uniquely determined by the νj up to translation and scaling - P n P see ﬁg. 7. We will show that P (ω) can be identiﬁed with XP .

2The asterisk in the section name is to indicate that the material of this section is not going to be used in the proof of Bernstein’s theorem. It is used for the ﬁrst time in chapter VIII. 96 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

′ P 3 1 2

2 4 6

′ 3 2 FIGURE 8. A polytope in R such that XP′ = P (1, 2, 3) P ∼

n n+1 ′ 10.2. P (ω) via polytopes in R . Let p := lcm(ω0,...,ωn) and be the n-dimensional simplex n+1 P in R with vertices βj := (p/ωj)ej , j = 0,...,n, where e0,...,en are the standard unit vectors in n+1 R (ﬁg. 8). We will show that XP′ = XP , where is from section VI.10.1. ∼ P 10.3. Equivalence of the constructions. Let be as in section VI.10.1. For each j = 0,...,n, let be the facet of with inner normal ν , andPα be the (unique) vertex of which is not on the Qj P j j P facet j . Pick j, 0 j n. Proposition VI.14 implies that the coordinate ring of Uαj XP is k[Sj ], Q ≤ ≤ n ∩ where Sj is the semigroup of integral points in the polyhedral cone αj R generated by αi αj , C ⊂ n − i = 0,...,n. Exercise VI.31 below implies that α Sj if and only if α Z and νi, α 0 for each n n+1∈ ∈ h i ≥ i =0,..., ˆj,...,n. Consider the map φ : Z Z given by α ( ν0, α ,..., νn, α ). 7→ 7→ h i h i ′ n+1 Claim VI.32. φ induces an isomorphism between Sj and S := β = (β0,...,βn) Z : ω,β = j { ∈ h i 0, β 0 for each i =0,..., ˆj,...,n . i ≥ } −1 ′ ′ PROOF. Exercise VI.31 implies that Sj = φ (Sj ). We now show that Sj = φ(Sj ). Indeed, let ′ n+1 ′ Hj and Hj be the subgroups of Z generated respectively by φ(Sj ) and Sj . It sufﬁces to show that ′ ′ n Hj = Hj . Since Hj Hj are subgroups of Z of the same rank n, we have to show that if kβ Hj for ⊂ n+1 ∈ some positive integer k and β = (β0,...,βn) Z , then β Hj . Indeed, if kβj = νj , α for each n ∗ ∈ ∈ n h i j = 0,...,n, then since the νj span (Z ) , it follows that α/k Z , and β = φ(α/k) Hj . Therefore ′ ∈ ∈ Hj = Hj , which proves the claim.

′ ′ As in section VI.10.2 let βj be the vertex of on the j-th axis. Exercise VI.31 implies that Sj is P ′ n+1 the semigroup of integral points in the polyhedral cone R generated by βi βj , i = 0,...,n. Cβj ⊂ − Therefore proposition VI.14 implies that φ induces an isomorphism XP ∼= XP′ .

n 10.4. Identification with P (ω). Let f k[x0,...,xn]. We say that f is weighted homogeneous with respect to ω (or in short, ω-homogeneous∈) if ω is constant on Supp(f). If f is ω-homogeneous, then n the set V (f) := [a0 : : an]: f(a0,...,an)=0 P (ω) of zeroes of f is a well defined subset n { ··· n } ⊂ n n of P (ω). As in the case of P , the basic open subsets of P (ω) are Uj := P (ω) V (xj ), j =0,...,n. \ k If f is ω-homogeneous with ω(f) a multiple of ωj , say ω(f) = kωj, k 0, then (f/xj ) Ca is constant k ≥ | for all Ca Uj and therefore f/xj is a well defined function on Uj. Exercise VI.32 shows that the k- ∈ k algebra Rj generated by all these f/xj , k 0, is finitely generated, and if h1,...,hs generate Rj as a ≥ ′ k-algebra, then they induce a bijection from Uj to an open affine subvariety of XP′ (where is as in n P n section VI.10.2) which extend to a bijection from P (ω) XP′ . We use this bijection to identify P (ω) → with XP′ = XP (where is as in section VI.10.1). In particular, it follows that each Uj is an open affine ∼ n P n subvariety of P (ω) with coordinate ring Rj . To completely describe the identification of P (ω) and XP′ n n it remains to explicitly identify points in P (ω) with those of XP′ . Given a := [a0 : : an] P (ω), ··· ∈ we now compute the ideal Ia of all ω-homogeneous polynomials in (x0,...,xn) that vanish at a. Let

Ja be the ideal of k[x0,...,xn] generated by all the xi such that ai = 0, and all binomials of the form α2 α1 α1 α2 n+1 a x a x where α1, α2 Z≥0 such that ω, α1 = ω, α2 . It is clear that Ja Ia. The following− proposition, which you will∈ prove in exerciseh VI.33i , showsh thati the converse is also true.⊂

Proposition VI.33. Ia = Ja. 11. ∗WEIGHTED BLOW UP 97

10.5. Exercises. EXERCISE VI.31. In the notation of section VI.10.3 show that n ˆ (1) αj = α R : νi, α 0, i =0,..., j,...,n . C′ { ∈ n+1h i≥ } (2) = β R : ω,β 0, βi 0, i =0,..., ˆj,...,n . Cβj { ∈ h i≥ ≥ } [Hint: use corollary V.23.] ′ EXERCISE VI.32. For each j = 0,...,n, let Rj be as in section VI.10.4 and Sj be as in section VI.10.3. ′ (1) Show that Rj = k[Sj ] for each j. ′ ′ ′ ′ (2) Let Uj := Uβj XP be the afﬁne open subset of XP (where is as in section VI.10.2)

∩ ′ P k with coordinate ring k[Sj ]. If h1,...,hs generate Rj as a -algebra, then show that the map n ′ x (h1(x),...,hs(x)) induces a bijection φj from Uj := P (ω) V (xj ) to Uj. 7→ \ ′ (3) Show that for distinct j, the maps φj are “compatible,” i.e. for each j, j , φj maps Uj Uj′ ′ ′ ∩ bijectively onto U U ′ . j ∩ j EXERCISE VI.33. Prove proposition VI.33. [Hint: use arguments analogous to those outlined in exercise VI.6.]

11. ∗Weighted blow up

3 −1 −1 Let ν be a weighted order on k[x1, x1 ,...,xn, xn ] with positive weights νj := ν(xj ), j = n ∗ 1,...,n. Identify ν with the integral element of (R ) with coordinates (ν1,...,νn) with respect to n n the basis dual to the standard basis of R . Fix a positive integer k. Let k := α Z≥0 : ν, α = kp , B { α∈ h i } where p := lcm(ν1,...,νn), and qk be the ideal of k[x1,...,xn] generated by x : α k . Recall that

n n n { ∈B }∗ n

k k k the blow up Blqk (k ) of at qk is the closure in XBk of the graph of the map φBk : ( ) XBk n × → defined as in (32). Let := α R≥0 : ν, α = p . Then is the (n 1)-dimensional simplex in n P { ∈ h i } P − n R with vertices (p/νj)ej , j = 1,...,n, where e1,...,en form the standard basis of R . Let be the n n Q n-dimensional simplex in R whose vertices are 0,e1,...,en, so that k can be naturally identified with n the basic open set U0 of XQ = P . If k is sufficiently large, then XBk = XP , so that assertion (7) of ∼ n ∼ proposition VI.14 implies that Blqk (k ) is isomorphic to the open subset of XP+Q which is the union of n basic open subsets Uej , j =1,...,n (note that each ej is a vertex of + ). In particular, Blqk (k ) are isomorphic for all sufficiently large k; we call the corresponding algebraicP Q variety the ν-weighted blow up

n n n n

k k k of k and denote it by Blν ( ). Let σ : Blν ( ) be the blow up map. The exceptional divisor

n −1 → n k on Blν (k ) is Eν := σ (0). Note that σ is an isomorphism on Blν ( ) Eν . If V is a subvariety of

n n n −1 \

k k k , the strict transform of V on Bl ( ) is the closure in Bl ( ) of σ (V 0 ). Note that is the ν ν \{ } P “lower” facet of + (see ﬁg. 9) and Eν is precisely the subvariety VP of XP+Q corresponding to the P Q n−1 facet of + . The construction in section VI.10.2 therefore shows that Eν ∼= P (ν). The following propositionP P makesQ this isomorphism more explicit.

2 1

2 1 0.5

3 1 4 3 2 0.5 2 1 1

2 0.5 4 5 6 1 Lower and upper faces of + P Q P Q 3 FIGURE 9. Construction of Bl(1,2,3)(k )

3The asterisk in the section name is to indicate that the material of this section is not going to be used in the proof of Bernstein’s theorem. It is used for the ﬁrst time in chapter IX. 98 VI.TORICVARIETIESOVERALGEBRAICALLYCLOSEDFIELDS

n ν1 νn n

k k Proposition VI.34. For each a = (a ,...,a ) k 0 , let C := (a t ,...,a t ): t . 1 n ∈ \{ } a { 1 n ∈ }⊂ The strict transform of each Ca intersects Eν at precisely one point which we denote by [a]. The map n−1 [a1 : : an] [a] yields an isomorphism between P (ν) and Eν . ··· → n ′ PROOF. Fix a = (a ,...,a ) k 0 and j such that a = 0. Let be the convex polyhe- 1 n ∈ \{ } j 6 Cj dral cone generated by (p/νi)ei (p/νj )ej , i = 1,..., ˆj,...,n, and j be the convex polyhedral cone ′ − ′ C ′ generated by j and ej . Let Sj and Sj be the semigroups of integral elements respectively in j and j.

C ′ ′ C C

k k k Proposition VI.14 implies that k[Uαj XP+Q] = [Sj ] and [Uαj VP ] = [Sj ]. Let Ca be the strict n ∩ ∩ transform of Ca in Blν (k ). It turns out that (exercise VI.34) α α ν ν (i) For every α Sj , either x identically vanishes on Ca or ordt(x (a1t 1 ,...,ant n )) is nonnega- ∈ ′ | tive. This implies that Ca Eν Uαj . ′ ∩ n ⊂ ˆ ′ (ii) Sj = α = (α1,...,αn) Z : ν, α =0, αi 0, i =1,..., j,...,n . If α Sj , it follows that { ∈ h i ≥ } ∈ α ′ (1) either αi > 0 for some i such that ai =0, in which case x identically vanishes on Ca, α α (2) or αi =0 if and only if ai =0, in which case x takes the constant (nonzero) value a on ′ 6 6 Ca. ′ These observations together with proposition VI.33 implies that Ca Eν corresponds precisely to the point n−1 ∩ [a1 : : an] P (ν), as required. ··· ∈ n−1 We write Oν for the torus OP of Eν = VP . The isomorphism between Eν and P (ν) from proposi- n−1 ∗ n−1 tion VI.34 induces an isomorphism Oν = P (ν) V (x1 xn) = (k ) . Proposition VI.19 implies

∼ n \ ··· ∼ I k that Oν is a nonsingular hypersurface of Blν (k ). Let I [n] := 1,...,n and := V (xj : j I) be

n ⊂ I { |I| } ′ 6∈

k k the corresponding coordinate subspace of k . We identify with . Let ν be a weighted order with ′ positive weights on k[xi : i I] such that (ν (xi): i I) is proportional to (νi : i I). It follows

∈ I ∈ ∈ I k from the deﬁnition of a weighted blow up that Blν′ (k ) can be identiﬁed with the strict transform of on

n I n

k k .( ) Bl (k ). The following proposition compiles some properties of the embedding Bl ′ ( ) ֒ Bl ν ν → ν Proposition VI.35. Assume gcd(νi : i I)=1. Let k := I . Then there is a Zariski open neighborhood n ∈ | | U of Oν′ in Blν (k ) and regular functions (z1,...,zn) on U such that

∗ k−1 n−k

k k (1) U ∼= k ( ) with coordinates (z1,...,zn), (2) z ,...,z× are monomials× in (x : i I), 1 k i ∈ (3) ν(z1)=1, ν(zi)=0, 2 i n, ′ ′ ≤ ≤ νi′ (4) for all i I, there is i such that zi′ = xi′ /z1 ,

6∈ ∗ k−1 n−k k (5) E U = V (z ) = (k ) , ν ∩ 1 ∼ × (6) Oν = (Eν U) V (zk+1 zn),

I ∩ \ ··· ∗ k−1

k k (7) Blν′ (k ) U = V (zk+1,...,zn) = ( ) , ∩ ∼ × ∗ k−1 (8) E ′ U = O ′ = V (z ,z ,...,z ) = (k ) . ν ∩ ν 1 k+1 n ∼ PROOF. This is left as an exercise. 11.1. Exercises.

EXERCISE VI.34. Verify observations (i) and (ii) from the proof of proposition VI.34.

EXERCISE VI.35. Prove proposition VI.35. CHAPTER VII

Number of solutions on the torus: BKK bound

1. Introduction In this chapter we derive Bernstein’s theorem for the number of isolated solutions of generic systems

∗ n k of n Laurent polynomials on the algebraic torus (k ) over an algebraically closed ﬁeld , apply it to derive some properties of mixed volume, and discuss some open problems related to Bernstein’s theorem. The term “BKK bound” in the title of this section refers to the bound from Bernstein’s theorem which is also known in the literature as “BKK theorem” after D. Bernstein, A. Kushnirenko and A. Khovanski.

2. Mixed volume The set of convex polytopes in Rn, n 1, is a commutative semigroup under Minkowski addition (see chapter V). The interaction between Minkowski≥ addition and volume gives rise to the theory of mixed volumes. The starting point of this theory is the following result proven in section V.7.1: n THEOREM V.39. Let 1,..., s be convexpolytopes in R . Then there are nonnegativereal numbers P P s vα( 1,..., s) for all α s := (α1,...,αs) Z≥0 : α1 + + αs = n such that for all λ = P P s ∈ E { ∈ ··· } (λ1,...,λs) R , ∈ ≥0 Vol (λ + + λ )= v ( ,..., )λα1 λαs n 1P1 ··· sPs α P1 Ps 1 ··· s αX∈Es where Voln is the n-dimensional Euclidean volume. n Deﬁnition VII.1. The mixed volume MV( 1,..., n) of convex polytopes 1,..., n in R is v(1,...,1)( 1,..., n). P P P P P P THEOREM VII.2. Let be any collection of convex polytopesin Rn which is invariant under Minkowski 1 K addition . Then MV : n R is the unique function such that K → (1) MV( ,..., )= n! Voln( ) for all , (2) MV isP symmetricP in its arguments,P andP ∈ K (3) MV is multiadditive, i.e. MV(k + k′ ′ , ,..., )= k MV( , ,..., )+ k′ MV( ′ , ,..., ) 1P1 1P1 P2 Pn 1 P1 P2 Pn 1 P1 P2 Pn ′ ′ for all k1, k Z≥0 and 1,..., n, . 1 ∈ P P P1 ∈ K Moreover, MV can be expressed in terms of the volume (we write [n] to denote 1,...,n ): { } (39) MV( ,..., )= ( 1)n−|I| Vol P1 Pn − n Pi I⊆[n] i∈I ! IX6=∅ X PROOF. This follows from combining theorem V.39, corollary B.48, and lemma B.45. Example VII.3. For n =1, a convex polytope is simply an interval and its mixed volume is its length. For n =2, if , are convex polygons in R2, then identity (39) implies (see ﬁg. 1) that P Q (40) MV( , ) = Area( + ) Area( ) Area( ) P Q P Q − P − Q Remark VII.4. Theorem V.39 implies that the mixed volume is nonnegative, and identity (39) implies that MV is invariant under volume preserving transformations of Rn. Moreover, theorem V.35 coupled with identity (39) implies that mixed volume is continuous with respect to the Hausdorff distance on polytopes. In section VII.6 we use Bernstein’s theorem to deduce some other basic properties of mixed volume.

1 E.g. K may be the set of all convex polytopes in Rn, or the set of convex integral polytopes in Rn.

99 100 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

+ =

P Q P + Q

FIGURE 1. MV( , ) is 8, which is the area of the blue part of + P Q P Q

3. Theorems of Kushnirenko and Bernstein

3.1. The bound. Given Laurent polynomials f1,...,fn in indeterminates x1,...,xn, the number of their zeroes counted with multiplicity is

∗ n [f1,...,fn](k ) := [f1,...,fn]a ∗ n a∈X(k ) where [f1,...,fn]a is the intersection multiplicity of f1,...,fn at a. If V := V (f1,...,fn) has non- iso ∗ n isolated points, then [f1,...,fn] k = (proposition IV.16). We write [f1,...,fn] ∗ n for the sum of ( ) ∞ (k ) intersection multiplicities of f1,...,fn at all the isolated points of V . Recall that the support Supp(f) of α n n a Laurent polynomial f = α∈A cαx is the set of all α Z such that cα =0. Given a subset of R , we say that f is supported at if Supp(f) , and we write∈ ( ) for the set6 of all Laurent polynomialsS PS ⊆ S L S n supported at . Given an ordered collection := ( 1,... m) of finite subsets of Z , we write ( ) for S A A A L A the set of all m-tuples (f1,...,fm) of Laurent polynomials such that fj is supported at j for each j. We say that some property holds for generic f supported at , j =1,...,m, if it holds forA all (f ,...,f ) j Aj 1 m m Pj |Aj | in a nonempty Zariski open subset of ( ) = ( ) = k . L A ∼ j=1 L Aj ∼ THEOREM VII.5 (Bernstein’s theorem: theQ bound). Let j be the convex hull of j , j =1,...,n. If Supp(f ) , j =1,...,n, then P A j ⊂ Aj iso (41) [f ,...,f ] ∗ n MV( ,..., ) 1 n (k ) ≤ P1 Pn Moreover, for generic f supported at , j =1,...,n, we have j Aj iso ∗ n n (42) [f ,...,f ] k = [f ,...,f ] ∗ = MV( ,..., ) 1 n ( ) 1 n (k ) P1 Pn The Newton polytope NP(f) of a Laurent polynomial f is the convex hull (in Rn) of the support of f. Bernstein’s theorem in particular states that the number of isolated solutions of a system of polynomials is bounded by the mixed volume of their Newton polytopes. A. Kushnirenko proved Bernstein’s theorem for the case that all the Newton polytopes are identical, in which case the mixed volume equals n! times the volume of any of these polytopes. D. Bernstein found theorem VII.5 while trying to understand and generalize Kushnirenko’s result. Kushnirenko however, not only gave the bound, but also gave a precise characterization of the collections of f1,...,fn for which the bound is achieved. There is a natural way to understand this characterization in the case that k = C; we describe it now. n 3.2. The non-degeneracy condition. Let 1,..., n be finite subsets of Z , and f1,...,fn be Lau- A A rent polynomials in x1,...,xn over C such that Supp(fj )= j , j =1,...,n. Assume there are Laurent A iso iso polynomials g1,...,gn such that Supp(gj ) j for each j, and [f1,...,fn](C∗)n < [g1,...,gn](C∗)n . ⊆ A ∗ n Write hj := (1 t)fj + tgj , j =1,...,n. Since the gj = hj t=1 have “more” common zeroes on (C ) − | ∗ n than the fj = hj t=0, intuitively we may expect that there is a curve C(t) on (C ) such that hj(C(t)) =0 | ∗ n and limt→0 C(t) is not on (C ) , i.e. as t approaches 0, either C(t) approaches one of the coordinate hyperplanes of Cn, or C(t) approaches infinity (see fig. 2). In any event, assuming our intuition is correct, | | there is a punctured neighborhood U of the origin on C and a parametrization U C(t) of the form ν1 νn ∗ n → γ : t (a1t + ,...,ant + ), where a = (a1,...,an) (C ) and for each j, νj is the order 7→ ··· ··· ∗ n ∈ (in t) of the j-th coordinate. Since limt→0 C(t) (C ) , it follows that not all the νj are zero. Let ν be 6∈ 3. THEOREMS OF KUSHNIRENKO AND BERNSTEIN 101

2 (f1, f2)=(6y −6x−1,y −(y −x) −1) 2 (g1,g2) =(6y −x−18,y −(y −x) −1)

t = 1 t = 0.6 t = 0.3 t = 0 NP(f1) = NP(g1) NP(f2) = NP(g2)

FIGURE 2. One of the common roots of (1 t)f + tg =0, j =1, 2, approaches inﬁnity as t 0 − j j →

n ∗ the element in (R ) with coordinates (ν1,...,νn) with respect to the basis dual to the standard basis of n n α R . Given a subset of R and a Laurent polynomial g = cαx supported at , we write S α S α Inν (g) Pif Supp(g) Inν ( ) = , (43) InS,ν (g) := cαx = ∩ S 6 ∅ (0 otherwise. α∈XInν (S) where Inν ( ) is defined as in section V.1. Let qj := minν ( j ). Since Supp(gj ) j = Supp(fj ), it follows thatS A ⊆ A h (γ(t)) = tqj In (f )(a)+ tqj +1( In (f )(a)+ tǫ In (g )(a)) + j Aj ,ν j − Aj ,ν j Aj ,ν j ··· where ǫ is nonnegative, and the orders in t of the omitted terms are higher than q . Since h (γ(t)) 0, it j j ≡ follows that InAj ,ν (fj)(a)=0 for each j =1,...,n. This leads to the following definition. n Definition VII.6. Let 1,..., m be finite subsets of Z and (f1,...,fm) ( ). We say that f ,...,f are ( ,...,A )-non-degenerateA if they satisfy the following condition:∈ L A 1 m A1 Am n ∗ ∗ for each ν (R ) 0 , there is no common (ND ) ∈ \{ } ∗ n root of InAj ,ν (fj ), j =1,...,m, on (k ) .

We say that f1,...,fm are BKK non-degenerate if they are (Supp(f1),..., Supp(fm))-non-degenerate.

The preceding argument suggests that for k = C, -non-degeneracy is sufficient for the maximality iso A of [f1,...,fn] ∗ n . It turns out that it is also necessary; Kushnirenko proved it in the case that the convex (k ) hulls of the are identical and Bernstein treated the general case. Both necessity and sufficiency remain Aj valid even if C is replaced by an arbitrary algebraically closed field: THEOREM VII.7 (Bernstein’s theorem: non-degeneracycondition). Let := conv( ), j =1,...,n. Pj Aj If the mixed volume of 1,..., n is nonzero, then the bound (41) is satisfied with an equality if and only if f ,...,f are ( ,...,P )-non-degenerate.P 1 n A1 An Remark VII.8. Recall from section VI.8.3 that integral elements in (Rn)∗ can be identified with weighted orders on the ring of Laurent polynomials. In the case that NP(fj ) = conv( j ) for each j = 1,...,m, condition (ND∗) is equivalent to the following condition: A

′∗ for each nontrivial weighted order ν, there is no (ND ) ∗ n common root of Inν (fj ), j =1,...,m, on (k ) . Remark VII.9. On the face of it condition (IX.24) consists of uncountably many conditions, one for each n ∗ element in (R ) . However, it is equivalent to ﬁnitely many conditions. Indeed, let j := conv( j ), P A j = 1,...,m, and := 1 + + m. For each face of , there are unique faces j of j such P P ··· P Q P α Q αP that = 1 + + m (proposition V.16). Given fj := cj,αx , let fj,Q := cj,αx be the Q Q ··· Q α α∈Qj “component” of f supported at . Then(ND∗) is equivalent to the following condition: j Qj P P ′′∗ for each face of dimension less than n of , there (ND ) Q P ∗ n is no common root of fj,Q, j =1,...,m, on (k ) . Kushnirenko’s theorem follows immediately by applying theorems VII.2, VII.5 and VII.7 to the case n that each fj is supported at the same (ﬁnite) subset of Z :

Corollary VII.10 (Kushnirenko [Kou76]). Let f1,...,fn be Laurent polynomials supported at a ﬁnite n iso subset of and := conv( ). Then [f1,...,fn] ∗ n n! Voln( ). If Voln( ) > 0, then the Z k A P A ( ) ≤ P P bound is satisﬁed with equality if and only if f ,...,f are ( ,..., )-non-degenerate. 1 n A A 102 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

n We say that convex polytopes 1,..., m in R are dependent if there is a nonempty subset I of [m] := 1,...,m such that dim(Q Q) < I . We use this notion in the following alternate for- { } i∈I Qi | | mulation of Bernstein’s theorem in which ( 1,..., n)-non-degeneracy is replaced by a combinatorial condition plus BKK non-degeneracy;weP willA prove itA in corollary VII.33. n THEOREM VII.11 (Bernstein’s theorem - alternate version). Let j be ﬁnite subsets of Z and fj be Laurent polynomials supported at , j =1,...,n. Then A Aj iso [f ,...,f ] ∗ n MV(conv( ),..., conv( )) 1 n (k ) ≤ A1 An If MV(conv( 1),..., conv( n)) > 0, then the bound is satisﬁed with an equality if and only if both of the followingA conditions hold:A

(1) for each nontrivial weighted order ν, the collection Inν (NP(fj )) : Inν ( j ) Supp(fj ) = of polytopes is dependent, and { A ∩ 6 ∅} ′∗ (2) f1,...,fn are BKK non-degenerate, i.e. they satisfy (ND ) with m = n. 4. Proof of Bernstein-Kushnirenko non-degeneracy condition

n m Let := ( 1,..., m), m 1, be a collection of ﬁnite subsets of Z . Let ( ) := ( j ) = A A A ≥ L A j=1 L A ∼ Pj |Aj | k and be the set of all -non-degenerate . In section VII.4.3 we prove ( ) (f1,...,fm) ( ) Q the followingN result:A A ∈ L A m THEOREM VII.12. ( ) is a Zariski open subset of ( ). If m min n, dim( conv( )) + N A L A ≥ { j=1 Aj 1 , then ( ) is nonempty. } N A P In the case that m = n, deﬁne iso iso

(44) [ ,..., ] ∗ n := max [f ,...,f ] ∗ n : Supp(f ) , j =1,...,n , k A1 An (k ) { 1 n ( ) j ⊆ Aj } iso iso

and let ( ) be the set of all (f ,...,f ) ( ) such that [f ,...,f ] ∗ n = [ ,..., ] ∗ n . In k M A 1 n ∈ L A 1 n (k ) A1 An ( ) sections VII.4.1 and VII.4.2 we prove the following result: THEOREM VII.13. Assume m = n. Then iso ( ) if [ 1,..., n] ∗ n =0, L A A A (k ) ( )= iso M A ( ) if [ 1,..., n] ∗ n > 0. (N A A A (k ) In particular, ( ) is a nonempty Zariski open subset of ( ). M A L A Theorem VII.12 is required (in section VII.4.2) for the proof of theorem VII.13. The reason to defer the proof of theorem VII.12 to section VII.4.3 is that it is somewhat more technical. 4.1. Sufficiency of Bernstein-Kushnirenko non-degeneracy condition. In this section we prove that ( ) ( ) in the case that m = n. In his proof of this part of his theorem, Bernstein used the theoryN ofA Puiseux⊆ M A series of curves, which only works in characteristic zero. For arbitrary characteristics, the analogous role is played by the notion of branches described in section VI.8.1. Let B = (Z,z) be a ∗ n n ∗ branch on (k ) . Let νB (R ) be the corresponding integral vector of weights, ρB be a parameter of ∗ n ∈ B and In(B) (k ) be the corresponding n-tuple of “initial coefficients” (see section VI.9.1). ∈ Lemma VII.14. Let B be a branch of a curve contained in the common zero set of Laurent polynomials ∗ n f1,...,fm on (k ) . Then In(B) is a common zero of InνB (fj ), j = 1,...,m. In particular, if B is a branch at infinity, and Supp(f ) for each j, then both (ND∗) and (ND′∗) are violated with ν = ν . j ⊆ Aj B PROOF. The first assertion is a direct corollary of lemma VI.30. If B is a branch at infinity, then νB is a nontrivial weighted order (lemma VI.26), so that the second assertion follows from the first assertion and the second identity from (43).

Corollary VII.15. Let fj be a Laurent polynomial supported at j , j = 1,...,m. If f1,...,fm are ∗ n A ( ,..., )-non-degenerate, then V (f ,...,f ) (k ) is ﬁnite. A1 Am 1 m ⊂ Proposition VII.16. Let fj be a Laurent polynomial supported at j , j = 1,...,n. If f1,...,fn are iso Aiso

( 1,..., n)-non-degenerate, then [f1,...,fn] ∗ n = [ 1,..., n] ∗ n . In particular, ( 1,..., n) k A A (k ) A A ( ) N A A ⊆ ( ,..., ). M A1 An 4. PROOF OF BERNSTEIN-KUSHNIRENKO NON-DEGENERACY CONDITION 103

PROOF. Assume to the contrary that there are Laurent polynomials gj supported at j such that iso iso A

[g1,...,gn] ∗ n > [f1,...,fn] ∗ n . It suffices to show that this leads to a contradiction. Define k (k ) ( ) ∗ n h := (1 t)f + tg , where t is a new indeterminate. Since the set of zeroes of f ,...,f on (k ) j − j j 1 n is finite (corollary VII.15), assertion (5) of theorem IV.20 implies that the set of zeroes of h1,...,hn in

∗ n ∗ n

k k k (k ) contains a curve which has a branch B = (Z,z) at inﬁnity (with respect to ( ) ) at × ∗ n × t = 0 and B (k ) 0 . Then t B 0 and B determines a well-deﬁned weighted order νB on

−1 6⊆ −1 ×{ } | 6≡ −1 −1 k k[x1, x1 ,...,xn, xn ,t] such that νB(t) > 0. Let ν be the restriction of νB to [x1, x1 ,...,xn, xn ]. Since ν (t) > 0 and Supp(g ) , it follows from (43) that B j ⊂ Aj

Inν (fj )=InνB (hj ) if Supp(fj ) Inν ( j ) = , InAj ,ν (fj )= ∩ A 6 ∅ (0 otherwise.

∗ n Therefore lemma VII.14 implies that InAj ,ν (fj ), j = 1,...,n, have a common zero on (k ) . Since B

∗ n k is centered at infinity with respect to (k ) and νB(t) > 0, it follows that ν is a nontrivial weighted −1 −1 × order on k[x , x ,...,x , x ]. This contradicts the -non-degeneracy of f ,...,f , as desired. 1 1 n n A 1 n 4.2. Necessity of Bernstein-Kushnirenko non-degeneracy condition. In this section we finish (us- iso ing theorem VII.12) the proof of theorem VII.13. Assume [ 1,..., n] ∗ n > 0 and (f1,...,fn) (k ) A A −1 −1∈ ( ) is a -degenerate system in , i.e. there is a nontrivial weighted order ν on k[x1, x1 ,...,xn, xn ] L A A L ∗ n such that InA1,ν (f1),..., InAn,ν (fn) have a common zero on (k ) . It suffices to show that

iso iso

[f ,...,f ] ∗ n < [ ,..., ] ∗ n k 1 n (k ) A1 An ( ) We show it in two ways.

∗ n Claim VII.17. Given any point a (k ) , there is (g ,...,g ) ( ) such that ∈ 1 n ∈ L A (1) InAj ,ν (gj )(a) =0 for each j =1,...,n, and 6 ∗ n (2) there is an isolated zero b of (g ,...,g ) on (k ) such that f (b) =0 for each j =1,...,n. 1 n j 6 PROOF. Due to theorem VII.12 we can pick -non-degenerate (h ,...,h ) ( ) which auto- A 1 n ∈ L A matically satisﬁes the ﬁrst property. For each j, let αj be an arbitrary element of j . For each ǫ := ∗ n αj A (ǫ1,...,ǫn) (k ) , let hǫj ,j := hj ǫjx . Since ( ) is Zariski open (theorem VII.12), it fol- ∈ − N A∗ n

k ∗ n lows that (hǫ1,1,...,hǫn,n) is in ( ) for generic ǫ ( ) , and therefore [hǫj ,1,...,hǫj ,n](k ) =

iso N A ∈ ∗ n ∗ n k [ 1,..., n] ∗ n > 0 (proposition VII.16); in particular the map Φ : (k ) ( ) deﬁned by x A A (k ) → 7→ −α1 −αn (x h1,...,x hn) is dominant. Therefore it sufﬁces to take gj := hǫj ,j for generic (ǫ1,...,ǫn) ∗ n ∈ (k ) Φ(V (f f )). \ 1 ··· n 4.2.1. Bernstein’s proof of necessity of non-degeneracy (Bernstein’s trick). After a linear change of n ∗ n coordinates of Z (which corresponds to a monomial change of coordinates of (k ) ) and translating n each j by an element in Z (which corresponds to multiplying each fj by a monomial) if necessary, A n−1 we may arrange that ν = (0,..., 0, 1) and Inν ( j ) = j (Z 0 ), j = 1,...,n. For each (g ,...,g ) ( ), it follows that A A ∩ ×{ } 1 n ∈ L A (45) In (g )= g Aj ,ν j j |xn=0 ∗ n ′ If (a1,...,an) is a common zero of the InAj ,ν (fj ) on (k ) , then it follows that a := (a1,...,an−1, 0) n is also a common zero of the InAj ,ν (fj ) on k . Let g1,...,gn and b be as in claim VII.17. Identity (45)

′ n ′ k implies that gj (a ) =0 for each j =1,...,n. Take any map c : k such that c(0) = a and c(1) = b 6 ′ → (e.g. we may take c(t)=(1 t)a + tb) and set hj (x, t) := gj (c(t))fj (x) fj (c(t))gj (x), j =1,...,n.

− ′ n+1 − k Then each hj vanishes on the curve C := (c(t),t): t k . Since (b, 1) is an isolated zero of

′ { ′ ∈ } ⊂ ∗ n k h1(x, 1),...,hn(x, 1), and since (a , 0) C is a “point at inﬁnity” with respect to (k ) , assertion ∈ iso iso×

(5) of theorem IV.20 implies that [h1(x, 0),...,hn(x, 0)] ∗ n < [h1(x, ǫ),...,hn(x, ǫ)] ∗ n for generic k (k ) ( ) iso iso

ǫ k. Since each hj (x, ǫ) is supported at j , it follows that [f1,...,fn] ∗ n < [ 1,..., n] ∗ n , as k ∈ A (k ) A A ( ) required. 104 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

4.2.2. A modified version of Bernstein’s trick. Here we give an alternative proof of the necessity of Bernstein-Kushnirenko non-degeneracy by adapting Bernstein’s trick to produce a curve C′ as in his orig- ∗ n inal proof without changing the coordinates on (k ) ; this will be useful later, e.g. in the proofs of the weighted Bézout theorem (section VIII.3) and the extension of the BKK bound to the affine space (sec- ∗ n tion X.7.3). Let a = (a1,...,an) be a common zero of the Inν (fj ) on (k ) . As in the original proof, let ∗ n ′ (g1,...,gn) ( ) and b = (b1,...,bn) (k ) be as in claim VII.17. Fix integers νj >νj := ν(xj ).

∈ L A n ∈ n

k k Let C be the rational curve on k parameterized by c(t) := (c1(t),...,cn(t)) : given by ′ → (46) c (t) := a tνj + (b a )tνj , j =1, . . . , n. j j j − j

∗ n ∗ n

k k Note that c(1) = b (k ) , so that c(t) ( ) for all but ﬁnitely many values of . It follows −mj ∈ −mj ∈ that t fj (c(t)) and t gj (c(t)) are well deﬁned rational functions in t. Let mj := minAj (ν), j = 1,...,n. The following claim, which follows from a straightforward computation, implies that t = 0 is −mj −mj not a pole of t fj (c(t)) or t gj (c(t)).

−mj Claim VII.18. Fix j, 1 j n. If p is a Laurent polynomial supported at then t p (c(t)) k[[t]]. ≤ ≤ j Aj j ∈

The following identity holds in k[[t]]:

−mj t pj(c(t)) = InAj ,ν (pj )(a)+ tqj (t) for some q (t) k[[t]]. j ∈ Deﬁne (47) h := t−mj f (c(t))g t−mj g (c(t))f j j j − j j −2mj Let T be the complement in k of all the poles of t fj (c(t))gj (c(t)). Then both 0 and 1 are in T . Moreover, for each j, Q hj (x, 1) = fj (b)gj (x) (since fj (b) =0= gj (b)), and • h (x, 0) = In (g )(a)f (x) (this6 follows from claim VII.18 since In (f )(a)=0 = • j − Aj ,ν j j Aj ,ν j 6 InAj ,ν (gj )(a)). In particular, hj (x, 0) and hj (x, 1) are (nonzero) constant multiples of respectively fj and gj . Each hj

′ n+1 ′ ∗ n k vanishes on the curve C := (c(t),t): t T k . Note that C (( ) 1 ) contains (b, 1) { ∈ } ⊂ ∩ ×{ } which is an isolated zero of h1(x, 1),...,hn(x, 1). On the other hand, since ν is nontrivial, it follows that ′ ∗ n C has a “point at inﬁnity” at t = 0 with respect to (k ) T . Therefore assertion (5) of theorem IV.20 iso × iso

implies that [h1(x, 0),...,hn(x, 0)] ∗ n < [h1(x, ǫ),...,hn(x, ǫ)] ∗ n for generic ǫ T . Since each k (k ) ( ) ∈ iso iso

such hj (x, ǫ) is supported at j , it follows that [f1,...,fn] ∗ n < [ 1,..., n] ∗ n , as required. k A (k ) A A ( ) 4.3. The set of non-degenerate systems. In this section we prove theorem VII.12. We start with some notation. Let J [m] := 1,...,m and J ( ) := j∈J ( j ). Write cj,α for the coefficient of α ⊆ { } L A L A x for each j J, so that (cj,α : j J, α j ) are the coordinates on J ( ). For f = (cj,α)j,α ∈ ∈ ∈ A Q L A α ∈ J ( ), we write fj for the corresponding element in ( j ), i.e. fj = cj,αx . Let J := ( j : L A L A α∈Aj A A j J) and ( ) be the set of all f ( ) which are -non-degenerate. We will at first show that ∈ NJ A ∈ LJ A AJ P J ( ) is Zariski open in J ( ). If = ( j : j J) is an ordered tuple of (finite) sets, we say that is Na faceAof , and write L A, if itB satisfiesB the following∈ property: B AJ B AJ n ∗ “there is ν (R ) such that j = Inν ( j ) for each j J.” ∈ B A ∈ For each and , we define α. Let = ( j : j J) J f = (cj,α)j,α J ( ) fj,Bj := α∈Bj cj,αx B B ∈ A ∈ L A ∗ n ( ) be the set of all f such that there is a common root of f , j J, on (k ) , so that DJ,B A ∈ LJ j,Bj ∈ P (48) ( ) ( )= ( ) LJ A \ NJ A DJ,B A B[AJ ′ ∗ n ¯′ n ′ ′ Let J ( ) := J ( ) (k ) and J ( ) := J ( ) P . Let J,B( ) J ( ) be the collection of L A L A × L A ∗ n L A × D A ⊂ L A ′ all (f,a), where f ( ) and a (k ) are such that f (a)=0 for each j J; let ¯ ( ) be ∈ DJ,B A ∈ j,Bj ∈ DJ,B A the closure of ′ ( ) in ¯′ ( ). Let π : ¯′ ( ) ( ) be the natural projection and ¯ ( ) := DJ,B A LJ A J LJ A → LJ A DJ,B A π ( ¯′ ( )). J DJ,B A

Claim VII.19. Let . Then ¯ ( ) ′ ′ ( ). B AJ DJ,B A ⊂ B B DJ,B A S 4. PROOF OF BERNSTEIN-KUSHNIRENKO NON-DEGENERACY CONDITION 105

0 0 ¯ 0 n 0 0 ¯′ PROOF. Let f = (c )j,α J,B( ) J,B( ). Pick a P such that (f ,a ) ( ). j,α ∈ D A \ D A ∈ ∈ DJ,B A We can ﬁnd an irreducible curve C on ¯′ ( ) such that (f 0,a0) C and C ′ ( ) is nonempty DJ,B A ∈ ∩ DJ,B A and open in C (proposition III.24). Let B = (Z,z) be a branch of C at (f 0,a0). For each j = 1,...,n, and each α j , we write x¯j , c¯j,α respectively for the restrictions of xj ,cj,α to C. Then for each j J, ∈ A α ′ ′ ∈ F¯j := c¯j,αx¯ is identically zero on C. Since C ( ) = , it follows that no x¯j is identically α∈Bj ∩ DJ,B A 6 ∅ zero on C. Let ν′ be the element in ( n)∗ with coordinates (ν (¯x ),...,ν (¯x )) with respect to the P B R B 1 B n ′ ′ ′ ′ basis dual to the standard basis. Fix j J. Let := Inν ( j ) and mj := minBJ (ν )= ν ,β for any ∈ Bj B B B h B i β ′ . Then for each α , since ord (¯c ) 0, it follows that ∈Bj ∈Bj z j,α ≥ ord (¯c x¯α) = ord (¯c )+ord (¯xα) ord (¯xα)= ν′ , α m z j,α z j,α z ≥ z h B i≥ j α ′ 0 Moreover, ordz(¯cj,αx¯ ) = mj if and only if ordz(¯cj,α)=0 and α j , i.e. if and only if cj,α = 0 and ′ 0 ∈ B ¯ 6 α j , i.e. if and only if α Supp(fj,B′ ). If ρB is a parameter at B, it follows that Fj can be expanded ∈B ∈ j in k((ρB )) as n c0 (a0)αi (ρ )mj + j,α i B ··· α∈B′ i=1 Xj Y 0 ¯ where ai = InB(¯xi), i =1,...,n, and the omitted terms have higher order in ρB. Since Fj is identically 0 0 0 α 0 zero on C, it follows that (a1,...,an) is a zero of α∈B′ cj,αx = fj,B′ for each j J. Therefore j j ∈ 0 0 f = (f : j J) is -degenerate and is an element of ′ ( ), as required. j ∈ AJ P DJ,B A

Corollary VII.20. For each , the set ′ ′ ( ) is Zariski closed in ( ). Consequently B AJ B B DJ,B A LJ A ( ) is Zariski open in ( ). NJ A LJ A S n ¯ ¯′ PROOF. Since P is complete (see section III.7), it follows that J,B( ) := πJ ( ( )) is Zariski D A DJ,B A closed in J ( ). Since the relation is transitive, claim VII.19.therefore implies the first assertion. The second assertionL A then follows from eq. (48) (and the transitivity of ). Let := conv( ), j = 1,...,n, and := . For each = ( : j J) , define Pj Aj PJ j∈J Pj B Bj ∈ AJ := conv( ). Note that is a face of . PJ,B j∈J Bj PJ,B PPJ ¯ Claim VII.21.P Let J be such that dim( J,B) < J . Then J,B( ) $ J ( ). B A P | | D A L A PROOF. We proceed by induction on J . If J = 1, then the assumption dim( J,B) < J is valid only if is a vertex of , and therefore| | is a| vertex| of for each j. In thatcasePf is a| monomial| PJ,B PJ BJ Aj j,Bj for each f J ( ), so that ¯J,B( ) = . Now assume J 2. Let d := dim( J,B). Since d < J , ′ ∈ L A ′D A ∅ | | ≥ P ′ | | ′ ′ there is J J such that J = J 1 and dim( J ,BJ′ ) = d, where J := ( j : j J ). Define ⊂ ′ | | | |− P B B ∈ ′ := ( : j J ) and BJ Bj ∈ J′,B( ) := J′ ( ) ¯J′,B′ ( ) N A L A \ ′ D A B [BJ′ dim(PJ′,B′ )

2 ∼ p It is straightforward to check that LJ′ (A) = k for some p ≥ 0. 106 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND j is the unique element of J J ′, then a generic f ( ) will satisfy the following: for each ′ , 0 \ j0 ∈ L Aj0 B B ′ ′ ′ fj0,B takes a nonzero value at each point of WB . It follows that f := (fj : j J) B′B J,B ( ). j0 ∈ 6∈ D A Claim VII.19 now implies that f ¯ ( ), as required. 6∈ DJ,B A S PROOF OF THEOREM VII.12. Apply corollary VII.20 with J = [m] := 1,...,m to see that ( ) is Zariski open in ( ). If dim( m )

−1 −1 Claim VII.19 implies that π˜ ( J,B( )) = π˜ ( ¯J,B( )), so that ˜ ( ) is Zariski open BAJ J D A BAJ J D A N A in ( ). Claim VII.21 implies that ˜ ( ) is nonempty. L A S N A S 5. Proof of the BKK bound

n n For each j = 1,...,n, let j be a ﬁnite subset of Z and j be the convex hull of j in R . In this section we show that A P A iso (49) [ ,..., ] ∗ n = MV( ,..., ) A1 An (k ) P1 Pn In section VII.5.1 we prove two relevant results from the theory of toric varieties, which we use in section VII.5.2 to prove (49). Throughout this section we follow the convention of section VI.8.3 to identify n ∗ weighted orders on the ring of Laurent polynomials in (x1,...,xn) with integral elements of (R ) .

5.1. Toric propositions. Let ν be a primitive integral element in (Rn)∗. For each Laurent polyno- ′ n mial g in (x1,...,xn), deﬁne Tαν (g) and Inαν ,ψν (g) as in corollary VI.21, where αν Z is such that n n−1 ∈ ν, αν =1 and ψν : Zν⊥ = Z is an isomorphism (of abelian groups). Let be an n-dimensional con- h i n∼ P vex integral polytope in R which has a facet with primitive inner normal ν. Let XP be the toric variety Q 1 corresponding to from section VI.5, and given Laurent polynomials g1,...,gk, let VP (g1,...,gk) be P ∗ n the extension from section VI.7.1 of the closed subscheme V (g1,...,gk) of (k ) to a closed subscheme 1 of the Zariski open subset XP of XP . Recall that a possibly non-reduced curve is a pure dimension one closed subscheme of a variety.

Proposition VII.23. Let f1,...,fn be Laurent polynomials in (x1,...,xn). ′ ′ ∗ n−1 (1) If Inαν ,ψν (f2),..., Inαν ,ψν (fn) have no common zero on (k ) , then there is a Zariski open subset U ′ of X containing O such that the support of V 1(f ,...,f ) U ′ is empty. P Q P 2 n ∩

3Unlike A-non-degeneracy, ν is allowed to be the trivial weighted order (0,..., 0). 5. PROOF OF THE BKK BOUND 107

′ ′ ∗ n−1 (2) If the number of common zeroes of Inαν ,ψν (f2),..., Inαν ,ψν (fn) on (k ) is nonzero and ﬁ- ′ ′ 1 nite, then there is a Zariski open subset U of XP containing OQ such that C := VP (f2,...,fn) ′ ′ ∗ n ∩ U is a possibly non-reduced curve and every irreducible component of C intersects (k ) . ′ (3) If in addition to the assumptions of assertion (2) Inαν ,ψν (f1) does not vanish at any of the com- ′ ′ ∗ n−1 mon zeroes of Inαν ,ψν (f2),..., Inαν ,ψν (fn) on (k ) , then f1 restricts to a nonzero rational ′ function on C which can be represented in C′,a as a quotient of non zero-divisors for every a C′, and O ∈ ′ ′ ′ ∗ n−1 orda(f1 C )= ν(f1)[Inαν ,ψν (f2),..., Inαν ,ψν (fn)](k ) ′ | a∈CX∩OQ ′ ′ ∗ n−1

ROOF k P . If Inαν ,ψν (f2),..., Inαν ,ψν (fn) have no common zero on ( ) , then proposition VI.25 1 ′ 1 1 implies that VP (f2,...,fn) OQ = ,so thatpart(1) holds with U := XP VP (f2,...,fn). Now assume ∩ ′ ∅ ′ ∗ n−1\ the number of common zeroes of Inαν ,ψν (f2),..., Inαν ,ψν (fn) on (k ) is nonzero and finite. Since by −ν(fj )αν αν definition Tαν (fj)= x fj , it follows (e.g.due to proposition VI.25) that Tαν (f2),...,Tαν (fn), x 0 are n regular functions on the n dimensional variety UQ := XP OQ such that their common zero set 1 ∪ Z is finite and nonempty. Theorem III.20 then implies that VP (f2,...,fn) has pure dimension one near 1 each point of Z. Since Z is precisely the set of points in VP (f2,...,fn) OQ, this implies assertion (2) is ′ ∩ αν satisfied with some U UQ. Since UQ is nonsingular, and the set of zeroes of x on UQ is precisely OQ ⊂ ′ (proposition VI.19), and since OQ does not contain any irreducible component of C , lemma IV.17 implies αν ′ ′ that x C′ is a non zero-divisor in C′,a for each a C . Now fix a C OQ. Since a corresponds | ′ O ′ ∈ ∗ n−1 ∈ ∩ to a common zero of Inαν ,ψν (f2),..., Inαν ,ψν (fn) on (k ) (proposition VI.25), under the assumption ′ of assertion (3), Tαν (f1) is a regular function near a which does not vanish at a. Therefore Tαν (f1) C is α ν(f ) | ′ ν 1 ′ a unit in C ,a, and f1 = Tαν (f1)(x ) is the quotient of two non zero-divisors in C ,a. Proposi- O α O α ′ ′ ν ′ ν ′ tion IV.9 then implies that orda(f1 C ) = orda(Tαν (f1) C )+ν(f1)orda(x C )= ν(f1)orda(x C ). On the other hand, assertion (4) of| proposition IV.16 implies| that | |

αν αν αν ′ ord (x ) = [x ,T (f ),...,T (f )] = dim k ( / x ,T (f ),...,T (f ) ) a |C αν 2 αν n a OXP ,a h αν 2 αν n i Proposition VI.25 then implies that

αν ′ ′

′ ∗ n−1 k

ord (x ) = dim ( k / In (f ),..., In (f ) ) a |C O( ) ,(ψν )∗(a) h αν ,ψν 2 αν ,ψν n i ′ ′ = [Inαν ,ψν (f2),..., Inαν ,ψν (fn)](ψν )∗(a)

∗ n−1 ∗ ∗ n−1 k where (ψν )∗ : OQ ∼= (k ) is the inverse of the isomorphism ψν : ( ) ∼= OQ from corollary VI.21. Assertion (3) now follows immediately.

Corollary VII.24. Let f ,...,f be Laurent polynomials in (x ,...,x ) and be the sum of Newton 1 n 1 n R polytopes of f2,...,fn. ′ ′ ∗ n−1 (1) Assume that Inαν ,ψν (f2),..., Inαν ,ψν (fn) have no common zero on (k ) for every primitive n ∗ ∗ n integral ν (R ) such that dim(Inν ( ))

∗ n n ∗ n−1 k (50) [f ,...,f ] = [f ,...,f ] ∗ = ν(f )[In (f ),..., In (f )] k 1 n ( ) 1 n (k ) − 1 αν ,ψν 2 αν ,ψν n ( ) ν X n ∗ where the sum is over all primitive integral ν (R ) such that dim(Inν ( )) = n 1. ∈ R − ′ ∗ n PROOF. Let C be the set of common zeroes of f2,...,fn on (k ) . If dim( ) < n 1, then the assumption of assertion (1) implies that C′ = and all three sides of (50) are zero.R Therefore− assume that ∅ dim is n 1 or n. Let be an n-dimensional convex integral polytope in Rn which satisfy the following property:R − P n ∗ (51) for every ν (R ) , if ν is an inner normal to a face of of dimension n∈ 1, then ν is also an inner normal to a facet ofR . − P 108 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

For example, if dim( )= n, then we can simply take = , andif dim( )= n 1,we cantake to be R P R R − P any (convex integral) polytope which has two facets parallel to . Let XP be the toric variety correspond- ∗ n 0 R ing to . Recall that we can identify (k ) with the torus XP of XP (assertion (2) of proposition VI.14). ¯′P ′ n ∗ Let C be the closure of C in XP . Fix a primitive integral element ν (R ) . Let := Inν ( ) and OQ ∈ Q P be the corresponding torus orbit on XP . Claim VII.24.1. Let ′ := In ( ). Q ν R ′ ′ ′ ∗ n−1 ¯′ (1) If dim( )

∗ n ∗ n k projective. Since C (k ) is the closed subscheme of ( ) deﬁned by f2,...,fn, proposition IV.16 implies that ∩

∗ n [f1,...,fn](k ) = orda(f1 C ) ∗ n | a∈CX∩(k ) Since Supp(C) is projective, corollary IV.13, proposition VII.23, and claim VII.24.1 then imply that ′ ′

∗ n ∗ n−1 k [f ,...,f ] = ord (f )= ν(f )[In (f ),..., In (f )] k 1 n ( ) − a 1|C − 1 αν ,ψν 2 αν ,ψν n ( ) ν a∈CX∩XP,∞ X as required. 5.2. Proof of identity (49). Deﬁne iso iso

(52) [ ,..., ] ∗ n := max [f ,...,f ] ∗ n : Supp(f ) , j =1,...,n k P1 Pn (k ) { 1 n ( ) j ⊆Pj } ′ n iso Let := j j , j = 1,...,n. It follows from the deﬁnition that [ 1,..., n] ∗ n = j Z (k ) ′ A ′ Piso ∩ ⊇ A iso P P

[ ,..., ] ∗ n [ 1,..., n] ∗ n . However, due to theorem VII.12 we may pick BKK non- k 1 n (k ) ( ) A A ≥ A A ′ degenerate Laurent polynomials f1,...,fn such that Supp(fj ) j j and NP(fj )= j for each j, and then proposition VII.16 implies that ⊆ A ⊆ A P iso iso ′ ′ iso iso

[ ,..., ] ∗ n = [f ,...,f ] ∗ n = [ ,..., ] ∗ n = [ ,..., ] ∗ n

k k k A1 An (k ) 1 n ( ) A1 An ( ) P1 Pn ( ) In order to prove (49) it therefore sufﬁces to show that iso (53) [ ,..., ] ∗ n = MV( ,..., ) P1 Pn (k ) P1 Pn Claim VII.25. Let be the set of convex integral polytopes in Rn regarded as a semigroup under K n iso Minkowski addition. The function given by ( 1,..., n) [ 1,..., n] ∗ n is symmet- R k K → Q Q 7→ Q Q ( ) ric and multiadditive. 6.APPLICATIONSOFBERNSTEIN’STHEOREMTOCONVEXGEOMETRY 109

′ PROOF. The symmetry is evident, so we prove the multiadditivity. Pick 1,..., n, 1 . The- Q′ Q Q ∈ K orem VII.12 implies that we may choose Laurent polynomials g1,...,gn,g1 such that NP(gj ) = j , ′ ′ ′ Q j = 1,...,n, NP(g1) = 1, and both g1,...,gn and g1,g2,...,gn are BKK non-degenerate. But then ′ Q ′ g1g1,g2,...,gn are also BKK non-degenerate; in particular, g1g1,g2,...,gn only have isolated zeroes on ∗ n (k ) (corollary VII.15). It follows that ′ iso ′ iso

[ + , ,..., ] ∗ n = [g g ,g ,...,g ] ∗ n (proposition VII.16) k Q1 Q1 Q2 Qn (k ) 1 1 2 n ( ) iso ′ iso

= [g ,...,g ] ∗ n + [g ,g ,...,g ] ∗ n (proposition IV.16, assertion (5)) k 1 n (k ) 1 2 n ( ) iso ′ iso

= [ ,..., ] ∗ n + [ , ,..., ] ∗ n (proposition VII.16) k Q1 Qn (k ) Q1 Q2 Qn ( ) as required.

iso Due to theorem VII.2 and claim VII.25, inorderto prove(53) it sufﬁces to show that [ ,..., ] ∗ n = (k ) n P P n! Voln( ) for each convex integral polytope in R . We proceed by induction on n. It is clearly true for P P n n =1, so assume it is true for n 1. Pick a convex integral polytope in R . Let f1,...,fn be properly -non-degenerate Laurent polynomials− (see section VII.4.4) such thatP the Newton polytope of each f A j is . In particular f1,...,fn are BKK non-degenerate (proposition VII.22). Therefore corollary VII.15 andP proposition VII.16 imply that iso n ∗ n [ ,..., ] ∗ = [f ,...,f ] k P P (k ) 1 n ( ) Since the fj are properly non-degenerate, they satisfy the hypothesis of corollary VII.24, and identity (50) implies that

′ ′

∗ n ∗ n−1 k [f ,...,f ] k = ν(f )[In (f ),..., In (f )] 1 n ( ) − 1 αν ,ψν 2 αν ,ψν n ( ) ν X where the sum is over all primitive integral ν (Rn)∗. Fix one such ν. The proper -non-degeneracy ′ ′ ∈ ∗ nA−1 of the fj implies that Inαν ,ψν (f2),..., Inαν ,ψν (fn) are BKK non-degenerate on (k ) . Moreover, the ′ n−1 Newton polytope of each Inαν ,ψν (fj ) is ν , which is the convex hull in R of ψν (β mν αν ): β n P { − ∈ Inν ( ) Z , where mν := minP (ν). The inductive hypothesis implies that P ∩ } ′ ′ ∗ n−1

[In (f ),..., In (f )] k = (n 1)! Vol ( ) αν ,ψν 2 αν ,ψν n ( ) − n−1 Pν ′ ′ It follows from the deﬁnition of Volν ( ) (see deﬁnition V.42) that Voln−1( ν ) = Volν (Inν ( )). It follows that · P P ′ ′ ∗ n [f1,...,fn](k ) = (n 1)! min(ν) Volν (Inν ( )) = (n 1)! max(ν) Volν (ldν ( )) = n! Voln( ) − − P P − P P P ν ν X X iso where the last equality uses corollary V.43. Therefore, [ ,..., ] ∗ n = n! Voln( ), as required. P P (k ) P 6. Applications of Bernstein’s theorem to convex geometry In this section we use Bernstein’s theorem to deducesome propertiesof mixed volume. In particular we characterize the conditions for mixed volume (of n convex polytopes in Rn) being zero (theorem VII.30) and the conditions under which it is “strictly monotonic” (corollary VII.31). As an application back to algebraic geometry, we prove the alternate version of Bernstein’s theorem (corollary VII.33). Throughout n this section 1,..., n denote convex polytopes in R , n 1. P P ≥ ′ n ′ Proposition VII.26 (Monotonicity of mixed volume). If are convex polytopes in R such that j Pj P ⊆Pj for each j, then MV( ,..., ) MV( ′ ,..., ′ ). P1 Pn ≤ P1 Pn PROOF. Theorem VII.5 implies that it holds for rational polytopes. The general case then follows from the observation that every polytope can be approximated arbitrarily closely (with respect to the Hausdorff distance) by rational polytopes (corollary V.31), and the mixed volume is continuous with respect to the Hausdorf distance (remark VII.4).

n ∗ n n Let ν be a nonzero integral element of (R ) and Rν⊥ := α R : ν, α = 0 . Choose an afﬁne n n { ∈ h i n } n n transformation ψν : R R such that ψν restricts to an automorphism of Z and maps R ⊥ Z → ν ∩ 110 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

n−1 n onto Z 0 . If 1,..., n−1 are rational polytopes in R such that each j is a translate of some ×{′ }n Q Q Q polytope R ⊥ , then we define Qj ⊂ ν (54) MV′ ( ,..., ) := MV(ψ ( ′ ),...,ψ ( ′ )) ν Q1 Qn−1 ν Q1 ν Qn−1 where the mixed volume on the right hand side is the (n 1)-dimensional mixed volume on Rn−1. Propo- ′ − sition V.40 and theorem VII.2 imply that MVν does not depend on the choice of ψν or the translations involved. Proposition VII.27. Assume ,..., are rational polytopes. Then P1 Pn (1) MV( ,..., ) = max (ν) MV′ (ld ( ),..., ld ( )), where the sum is over all P1 Pn ν P1 ν ν P2 ν Pn primitive integral ν (Rn)∗. ∈ P (2) Assume 1 is a line segment in the direction of ν, where ν is a primitive integral element in n P R . Let l( 1) be the “integer length” of 1 (i.e. the Euclidean length of 1 is l( 1) times the P P n ∗ P P length of ν1). Identify ν with an element of (R ) via the basis dual to the standard basis of n n n R . Let πν : R Rν⊥ be a “lattice projection in the direction normal to ν” (i.e. πν = −1 → n n−1 ψν π ψν , where π : R R 0 is the projection in the first (n 1)-coordinates). Then◦MV(◦ ,..., )= l( →) MV′ (π×{( }),...,π ( )). − P1 Pn P1 ν ν P2 ν Pn PROOF. Due to the multiadditivity of the mixed volume we may assume that each is integral. Pj Pick Laurent polynomials fj with Newton polytope j such that f1,...,fn are properly non-degenerate. ∗ n P Then they are BKK non-degenerate on (k ) , they satisfy the hypothesis of corollary VII.24, and for each n ∗ ′ ′ ∗ n−1 primitive integral ν (R ) , Inαν ,ψν (f2),..., Inαν ,ψν (fn) are BKK non-degenerate on (k ) . Then corollary VII.24 and∈ theorems VII.5 and VII.7 implies assertion (1). For assertion (2), change coordinates on Zn (using lemma B.43) so that ν = (0,..., 0, 1). Without loss of generality we may assume that 1 is the line segment bounded by the origin and (0,..., 0,l), where l := l( 1). Then for each a = P ∗ n P (a ,...,a ) (k ) , 1 n ∈ l [f1,...,fn]a = [xn a1,f2,...,fn]a = [f2 xn=ǫ,...,fn xn=ǫ](a2,...,an) − l | | ǫX=an Since πν is simply the projection in the first (n 1)-coordinates, for generic a1, the Newton polytope of f is π ( ) for each j. Assertion (2) therefore− follows from theorem VII.5. j |xn=ǫ ν Pj

2 2 2

1 1 1

0.6 0.8 1 2 3 1.5 2 0.2 0.4 1 0.5 1

0.5 1 2 1 2

P1 P2 P3

FIGURE 3. Newton polytopes of polynomials from example VII.28

2 3 2 3 2 Example VII.28. Let f1 = a1x + b1y + c1z + d1z , f2 = a2x + b2xz + c2y + d2yz , f3 = 2 2 2 2 a x + b xz + c y + d yz , where a ,b ,c , d ’s are generic elements of k, and let := NP(f ), 3 3 3 3 i i i i Pj j ∗ 3 j = 1, 2, 3 (see fig. 3). We compute [f1,f2,f3](k ) = MV( 1, 2, 3) using proposition VII.27. As- P P P ′ sertion (1) of proposition VII.27 implies that MV( 1, 2, 3)= ν maxP1 (ν) MVν (ldν ( 2), ldν ( 3)). ′ P P P P P Theorem VII.30 implies that MVν (ldν ( 2), ldν ( 3)) is nonzero only if ν is one of the six outer normals of facets of + (fig. 4). When ν P= ( 1, 0P, 0) or (0, 1, 0)P, then max (ν)=0, so it suffices to P2 P3 − − P1 consider the remaining four cases. The image of the leading faces of 2, 3 and 2 + 3 under (certain ′ P P P P choices of) ψν are given in fig. 5, and example VII.3 implies that MV (ldν ( 2), ldν ( 3)) is the area of the region shaded green inside ψ (ld ( + )). It then follows from fig. 5 thatP P ν ν P2 P3 6.APPLICATIONSOFBERNSTEIN’STHEOREMTOCONVEXGEOMETRY 111

MV( 1, 2, 3) = max(1, 1, 1) Area( )+max( 1, 1, 1) Area( ) P P P P1 · P1 − − − · + max(2, 2, 1) Area( )+max( 2, 2, 1) Area( ) P1 · P1 − − − · =2 2 1 4+2 3 1 1 · − · · − · =5

4 (2,2,1) (1,1,1)

3 (-1,0,0) 2 (0,-1,0)

1 (-1,-1,-1) (-2,-2,-1) 1 2 3 4 5 2 4 outer normals to facets 2 + 3 P P of 2 + 3 P P

FIGURE 4. + and the outer normals to its facets P2 P3

ν = (1, 1, 1) ν =(−1, −1, −1)

+ = + =

ldν (P2) ldν (P3) ldν (P2 + P3) ldν (P2) ldν (P3) ldν (P2 + P3)

ν = (2, 2, 1) ν =(−2, −2, −1)

+ = + =

ldν (P2) ldν (P3) ldν (P2 + P3) ldν (P2) ldν (P3) ldν (P2 + P3)

FIGURE 5. The image under ψ of leading faces of , and + ν P2 P3 P2 P3

n Definition VII.29. We say that convex polytopes 1,..., m in R are dependent if there is a nonempty subset I of [m] := 1,...,m such that dim( Q ) < QI ; otherwise we say that they are independent. { } i∈I Qi | | In particular if m 1 and j = for some j, then 1,..., m are dependent. ≥ Q ∅ P Q Q THEOREM VII.30 (Minkowski). MV( ,..., )=0 if and only if they are dependent. P1 Pn PROOF. Due to corollary V.31 and remark VII.4 and the multiadditivity of mixed volumes, it suffices to consider the case that each is integral. At first assume there is I [n] such that dim( ) < I . Pj ⊆ i∈I Pi | | A recursive application of assertion (1) of proposition VII.27 shows that MV( 1,..., n) can be ex- P ′ PP ′ pressed as a sum such that each summand has a multiplicative factor of the form MVν (ldν ( i): i I), ′ ′ P ∈ where MV denotes an I -dimensional mixed volume and i is a face of i for each i I. Now pick BKK non-degenerate| | Laurent polynomials f , i I, suchP that NP(f )P = for each∈ i. Since i ∈ i Pi dim( i∈I i) < I , it follows from the definition of BKK non-degeneracy that there is no common zero P | | ∗ n ′ ′ of ld (f ), i I, on (k ) . Theorem VII.5 then implies that MV (ld ( ): i I)=0. This in turn νP i ν ν i implies that MV(∈ ,..., )=0. Now assume that ,..., are independent.P ∈ We will show that P1 Pn P1 Pn 112 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

MV( 1,..., n) > 0. We proceed by induction on n. The case of n =1 is obvious. In the general case, P P n since dim( n) 1, after a change of coordinates on Z if necessary, we may assume that it has positive P ≥ n n−1 length along xn-axis. Let π : Z Z be the projection in the ﬁrst (n 1)-coordinates. We consider two cases: → −

Case 1: MV(π( ),...,π( )) = 0. Due to the inductive hypothesis, we may assume after a P1 Pn−1 reordering of the if necessary, that there is k, 1 k n 1, such that dim( k π( )) < k. Pj ≤ ≤ − j=1 Pj Since ,..., are independent, it follows that dim( k π( )) = k 1 and dim( k ) = k. P1 Pn j=1 Pj − P j=1 Pj After a translation of one of the j if necessary, we may assume that the affine hull aff( 1 + ... + k) of P P P Pn P 1 + + k passes through the origin. Due to lemma B.43 we can change the basis of Z to ensure that P ··· P n n k ′ n n−k the subgroup of Z generated by Z aff( 1 + ... + k) is Z (0,..., 0) . Let π : R R ∩ P P ×{′ } ′ → be the projection in the last (n k)-coordinates. We claim that π ( k+1),...,π ( n) are independent. ′ − P ′ P Indeed, if dim( j∈J π ( j )) < J for some J k +1,...,n , then setting J := 1,...,k J will P ′ | | ⊆{ } { } ∪ yield that dim( ′ ) < J , contradicting the independence of the . Now pick generic f ,...,f Pj∈J Pj | | Pj 1 n such that NP(fj ) = j for each j. Theorems VII.5 and VII.7 and the inductive hypothesis implies that P P ∗ k the number of solutions Zk of f1,...,fk on (k ) is nonzero, and for each a = (a1,...,ak) Zk, ∗ n−k ∈ the number of solutions of fk+1,...,fn on a (k ) is nonzero. Theorem VII.5 then implies that MV( ,..., ) > 0, as required. { }× P1 Pn ∗ n Case 2: MV(π( 1),...,π( n−1)) > 0. In this case, for each generic ǫ (k ) , and for generic fj with NP(f )= , jP=2,...,nP 1, ∈ j Pj −

∗ n ∗ n−1 k [x ǫ,f ,...,f ] = [f ,...,f ] k = MV(π( ),...,π( )) > 0 n − 1 n−1 ( ) 1|xn=ǫ n−1|xn=ǫ ( ) P1 Pn−1

∗ n Computing [xn ǫ,f1,...,fn−1](k ) using eq. (50) then impliesdue to theorem VII.5 thatthere is a primi- − n ∗ n ′ tive integral ν (R ) such that ν,en =0 (where en = (0,..., 0, 1) Z ) and MVν (ldν ( 1),..., ldν ( n−1)) > 0. After a translation∈ if necessary,h wei 6 may assume the origin is in∈ the relative interior ofP . ThenP Pn maxPn (ν) > 0, and therefore assertion (1) of proposition VII.27 implies that MV( 1,..., n) > 0, as required. P P

′ n Corollary VII.31 (Strict monotonicity of mixed volume). If j are convex polytopes in R such that ′ ′ ′ P j j for each j, then MV( 1,..., n) MV( 1,..., n). The bound is strict if and only if both of theP ⊆P following are true: P P ≤ P P ′ ′ (1) 1,..., n are independent, and P P n ∗ ′ (2) there is ν (R ) 0 such that the collection Inν ( j ): j Inν ( j ) = of polytopes is independent.∈ \{ } { P P ∩ P 6 ∅} Remark VII.32. Recall that an empty collection of polytopes is independent. Therefore condition (2) of n ∗ ′ corollary VII.31 holds if there is ν (R ) 0 such that j Inν ( )= for each j. ∈ \{ } P ∩ Pj ∅ PROOF OF COROLLARY VII.31. Due to proposition VII.26 and theorem VII.30 it sufﬁces to prove the following statement: if ′ , j =1,...,n, and MV( ′ ,..., ′ ) > 0, then MV( ′ ,..., ′ ) > Pj ⊂Pj P1 Pn P1 Pn MV( ,..., ) if and only if condition (2) of corollary VII.31 is true. P1 Pn ′ ′ So assume MV( 1,..., n) > 0. Due to corollary V.31 and remark VII.4 and the multiadditivity of mixed volumes, we mayP assumeP in addition that all the , ′ are integral. Then choose BKK non-degenerate Pj Pj Laurent polynomials f such that NP(f )= , j =1,...,n. Theorems VII.5 and VII.7 imply that j j Pj iso ′ ′ MV( ,..., ) = [f ,...,f ] ∗ n MV( ,..., ) P1 Pn 1 n (k ) ≤ P1 Pn and the inequality is strict if and only if there is ν (Rn)∗ 0 such that ∈ \{ } ∗ n In ′ (f ),..., In ′ (f ) have a common zero on (k ) . (55) Pj ,ν 1 Pn,ν n

Since fj are generic, then it follows from theorems VII.5 and VII.30 that condition (55) holds if and only if In ( ): In ( ′ ) = is an independent collection of polytopes. { ν Pj Pj ∩ ν Pj 6 ∅} 7. SOME TECHNICAL RESULTS 113

n Corollary VII.33 (Bernstein’s theorem - alternate version). Let j be ﬁnite subsets of Z and fj be Laurent polynomials supported at , j =1,...,n. Then A Aj iso [f ,...,f ] ∗ n MV(conv( ),..., conv( )) 1 n (k ) ≤ A1 An If MV(conv( 1),..., conv( n)) > 0, then the bound is satisﬁed with an equality if and only if both of the followingA conditions hold:A

PROOF. The bound of corollary VII.33 follows from Theorem VII.5. Theorem VII.7 implies that if MV(conv( ),..., conv( )) > 0, then the bound holds with an equality if and only if f ,...,f are A1 An 1 n BKK non-degenerate and MV(NP(f1),..., MV(NP(fn) = MV(conv( 1),..., conv( n)). Now the result follows from corollary VII.31. A A

7. Some technical results In this section we compile a few (technical) corollaries of Bernstein’s theorem that we use in latter chapters.

Proposition VII.34. Let the set up be as in theorem VII.5. Assume MV( 1,..., n) is nonzero. If the set ∗ n P P iso of common zeroes of f1,...,fn on (k ) has a positive dimensional component, then [f1,...,fn] ∗ n < (k ) MV( ,..., ). P1 Pn PROOF. Since it is possible to choose a curve on such a positive dimensional component (corol- ∗ n lary III.21) and since every such curve has a branch at infinity with respect to (k ) (exercises III.35 and III.41), the proposition follows from lemma VI.30 and theorem VII.7. ′ ′ Corollary VII.35. Let the notation be as in proposition VII.23. Assume V (Inαν ,ψν (f2),..., Inαν ,ψν (fn)) ∗ n ′ ′ ′ ′ ∩ (k ) = , so that C is a curve. Let Cj j be the irreducible components of C , and j,ν be the col- 6 ∅ ′ { } ∗ n B lection of all branches B of Cj at infinity (with respect to (k ) ) such that νB is proportional to ν. If ′ ′ ∗ n−1 V (In (f ),..., In (f )) (k ) = , then αν ,ψν 1 αν ,ψν n ∩ ∅ ord (f ′ )= ord (f ′ )[f ,...,f ] ′ a 1 C z 1 Cj 2 n Cj ′ | ′ | a∈C ∩OQ j (Z,z)∈B (56) X X X j,ν ′ ′ ∗ n−1 = ν(f1)[Inαν ,ψν (f2),..., Inαν ,ψν (fn)](k ) ′ ′ where [f ,...,f ] ′ are defined as in section IV.5. If in addition f ,...,f are BKK non-degenerate, 2 n Cj 2,ν n,ν then ′ (57) orda(f1 C′ ) = min (ν) MVν (Inν (NP(f2)),..., Inν (NP(fn))) NP(f ) ′ | 1 a∈CX∩OQ PROOF. Theorem IV.12 implies that

ord (f ′ )= [f ,...,f ] ′ ord (f ′ ) a 1 C 2 n Cj a 1 Cj ′ | ′ | a∈C ∩OQ j a∈C ∩O X X Xj Q ∗ ′ ′ = [f2,...,fn]Cj ordz(πj (f1 Cj )) ′ −1 | j a∈C ∩OQ z∈π (a) X Xj Xj where π : C˜ C′ are desingularizations of C′ . It follows from the definition of branchesin section VI.8 j j → j j that each z π−1(a), where a C′ O , correspondsto a branch B = (Z,z) at infinity of C′ . Moreover, ∈ j ∈ j ∩ Q j since is a facet of , proposition VI.28 implies that a branch B at infinity of C′ intersects O if and Q P j Q only if νB is proportional to ν. It follows that

′ ′ ′ orda(f1 C )= ordz(f1 Cj )[f2,...,fn]Cj ′ | ′ | a∈C ∩OQ j (Z,z)∈B X X X j,ν The result now follows from proposition VII.23 and theorems VII.5 and VII.7. 114 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

8. The problem of characterizing coefﬁcients which guarantee non-degeneracy

n Let = ( 1,..., n) be an n-tuple of finite subsets of Z , and as in section VII.4, let ( ) be the A A A L A collection of n-tuples (f1,...,fn) of Laurent polynomials such that each fj is supported at j . Given (f ,...,f ) ( ), theorem VII.5 implies that if the coefficients of the f are generic, then A 1 n ∈ L A j iso (58) [f ,...,f ] ∗ n = MV( ,..., ) 1 n (k ) P1 Pn where = conv( ), j = 1,...,n. On the other hand, it is straightforward to see that not all the Pj Aj coefficients of the fj have to be generic for the equality in (58). E.g. if j is the set of vertices of j and if the coefficient of xα in each f is fixed for each α ,(58) still holdsB provided the coefficientsP of xα j ∈ Aj \Bj in the fj are generic for all α j. J. M.Rojas [Roj99] posed the problem of identifying all ( 1,..., n) ∈B ∗ n B B which have this property. The precise version of Rojas’ problem for (k ) is as follows: let j j , ∗ n B ⊆ A j = 1,...,n and := ( 1,..., n). We say that guarantees (k ) -maximality on ( ) if for each B B α B B L A choice of coefficients in k of x in fj for each j and each α j j,(58) holds provided the coefficients α ∈ A \B4 of x in each fj are generic for all α j . Then Rojas’ problem is to classify all which guarantees ∗ n ∈ B B (k ) -maximality on ( ). L A ∗ n Proposition VII.36 (Solution of Rojas’ problem for (k ) ). Let j := conv( j), j =1,...,n. Then the following are equivalent: Q B ∗ n (1) guarantees (k ) -maximality on ( ); B L A (2) MV( 1,..., n) = MV( 1,..., n); (3) one ofQ the followingQ holds:P P (a) 1,..., n are dependent, or P P n ∗ (b) for each ν (R ) 0 , the collection Inν ( j ): j Inν ( j ) = of polytopes is dependent. ∈ \{ } { Q Q ∩ P 6 ∅}

∗ n PROOF. It is straightforward to check that guarantees (k ) -maximality on ( ) if and only if for each (f ,...,f ) ( ), there is (g ,...,g )B ( ) such that L A 1 n ∈ L A 1 n ∈ L B iso [f + g ,...,f + g ] ∗ n = MV( ,..., ) 1 1 n n (k ) P1 Pn The implication (1) (2) follows from taking (f1,...,fn) = (0,..., 0) and applying theorem VII.5 and proposition VII.26⇒. For the opposite implication (2) (1), assume MV( ,..., ) = MV( ,..., ). ⇒ Q1 Qn P1 Pn Pick -non-degenerate (g1,...,gn) ( ) such that set hj := (1 t)fj + tgj. Theorem VII.7 implies B iso ∈ L B iso −

that [g1,...,gn] ∗ n [h1 t=ǫ,...,hn t=ǫ] ∗ n for each ǫ k, and therefore theorem IV.20 implies k (k ) ( ) ≥ | | iso ∈iso

that for generic ǫ k, [h1 t=ǫ,...,hn t=ǫ] ∗ n = [g1,...,gn] ∗ n = MV( 1,..., n), which implies k ∈ | | (k ) ( ) P P condition (1). Finally, the equivalence of conditions (2) and(3) follows from corollary VII.31.

∗ n We now describe a natural variant of the notion of (k ) -maximality on ( ). Consider the case 2 L A that n = 2 and 1 = 2 = Z , where is the bigger triangle from fig. 6. Let 1 = 2 = 2 A A P ∩ P B B Z , where is the smaller triangle in fig. 6. If f1,f2 are polynomials in two variables with Newton Q ∩ Q ∗ polytope , it is straightforwardto check that (ND ) is satisfied if the coefficients of monomials in fj whose P ∗ 2 exponents are in are generic, and therefore the number of solutions of f ,f on (k ) is MV( , )= B 1 2 P1 P2 2 Area( ) = 48. Note however that MV( 1, 2) = 2 Area( ) = 12 < MV( 1, 2). This motivates problemPVII.37 below. In its statement weQ useQ the following notation:Q we write P0( P) for the collection L A of all (f1,...,fn) ( ) such that NP(fj ) = j for each j. Given = ( 1,..., n) with j j , j =1,...,n, we denote∈ L A by π : ( ) ( ) beA the natural projectionB whichB “forgets”B the coefficientsB ⊆ A B L A → L B corresponding to α j, j = 1,...,n, and we write := ( 1 1,..., n n). We say that ∗ n6∈ B 0 A\B A \B A \B guarantees (k ) -maximality on ( ) if for each (h1,...,hn) ( ), there is a nonempty B L A iso ∈ L A\B Zariski open subset of ( ) such that [f1,...,fn] ∗ n = MV( 1,..., n) for each (f1,...,fn) U L B (k ) P P ∈ 0( ) π−1 (h ,...,h ) π−1( ). L A ∩ A\B 1 n ∩ B U Problem VII.37. Classify all = ( 1,..., n) with j j , j = 1,...,n, such that guarantees ∗ n 0 B B B B ⊆ A B (k ) -maximality on ( ). L A 4 ∗ n In [Roj99] Rojas posed the problem in a more general context (instead of (k ) he allowed for a broader class of subsets of n k ) and presented a solution. 8. THE PROBLEM OF CHARACTERIZING COEFFICIENTS WHICH GUARANTEENON-DEGENERACY 115

∗ 2 FIGURE 6. The number of solutions on (k ) of polynomials with Newton polytope is maximal if the coefﬁcients of monomials from are generic P Q

In the remainder of this section we give some partial answers to problem VII.37. We start with a ∗ n 0 condition that guarantees (k ) -maximality on ( ). Given a weighted order ν on the ring of Laurent polynomials in (x ,...,x ) and a subset J of [nL], letAd := dim(In ( )) and e := j J : 1 n J,ν ν j∈J Pj J,ν |{ ∈ In ( ) = . ν Pj ∩Bj 6 ∅}| P Proposition VII.38. Assume for each nontrivial weighted order ν, one of the following holds:

(1) either Inν ( j ) is a vertex of j for some j, (2) or there existsP a nonempty subsetP J of [n] such that e d < J . J,ν ≥ J,ν | | ∗ n 0 Then guarantees (k ) -maximality on ( ). B L A PROOF. Fix (h ,...,h ) ( ). It sufﬁces to show that for generic (g ,...,g ) ( ), if 1 n ∈ L A\B 1 n ∈ L B (f ,...,f ) π−1 (h ,...,h ) π−1(g ,...,g ), then (ND∗) holds provided (f ,...,f ) 0( ). 1 n ∈ A\B 1 n ∩ B 1 n 1 n ∈ L A Let ν be a nontrivial weighted order on the ring of Laurent polynomials. If (1) holds for some j, then ∗ n In (f ) is a monomial whenever NP(f )= and therefore is nowhere zero on (k ) . So without loss of ν j j Pj generality we may assume (2) holds for some nonemptysubset J of [n]. Let Jν := j J : Inν ( j ) j = { ∈ P ∗ ∩Bn 6 , so that eJ,ν = Jν . If eJ,ν > dJ,ν , it is straightforward to check that V (Inν (fj): j Jν ) (k ) = . ∅} | | ′ ∈ ∩ ∅ On the other hand, if eJ,ν = dJ,ν , then there is j J Jν (since dJ,ν < J ), and it is straightforward ∈∗ n \ | | to check that V (Inν (fj ): j Jν ) V (fj′ ) (k ) = for generic (gj : j Jν ). This proves the proposition. ∈ ∩ ∩ ∅ ∈

Proposition VII.39 (Solution to problem VII.37 for n =2). Assume n =2. Then = ( 1, 2) guaran- ∗ 2 0 B B B tees (k ) -maximality on ( ) if and only if for each nontrivial weighted order ν, one of the following holds: L A

(1) either Inν ( j ) is a vertex of j for some j, (2) or there is jPsuch that In ( P) = . ν Pj ∩Bj 6 ∅ PROOF. The ( ) direction follows from proposition VII.38. For the opposite inclusion assume there is a nontrivial weighted⇐ order ν such that for each j, In ( ) is an edge of which does not intersect ν Pj Pj j . Then it is clear we can pick fj with NP(fj ) = j such that Inν (f1) Inν (fj ) have a common zero B ∗ 2 P ∩ on (k ) and for such (f1,f2), no choice of coefﬁcients of monomials from j would make that zero disappear. B

∗ 3 Proposition VII.40 (Solution to problem VII.37 for n = 3). Assume n = 3. Then guarantees (k ) - maximality on 0( ) if and only if for each nontrivial weighted order ν, one of the followingB holds: L A (1) Inν ( j ) is a vertex of j for some j, (2) or thereP are j = j suchP that In ( + ) has dimension one and In ( ) = , 1 6 2 ν Pj1 Pj2 ν Pj1 ∩Bj1 6 ∅ (3) or there are j1 = j2 such that Inν ( j1 + j2 ) has dimension two and Inν ( jk ) jk = for each k =1, 2, 6 P P P ∩B 6 ∅ (4) or there are j = j such that 1 6 2 116 VII.NUMBEROFSOLUTIONSONTHETORUS:BKKBOUND

5 (a) there is no positive dimensional polytope which is a Minkowski summand of Inν ( jk ) for each k =1, 2, P

(b) Inν ( jk ) jk = for each k =1, 2, and (c) In (P ) ∩B = ∅, where j is the single element of 1, 2, 3 j , j . ν Pj3 ∩Bj3 6 ∅ 3 { }\{ 1 2} PROOF. At ﬁrst we prove the ( ) implication. Fix (h1,...,hn) ( ), (g1,...,gn) ( ) −1 ⇐ −1 0 ∈ L A\B ∈ L B and (f1,...,fn) πA\B(h1,...,hn) πB (g1,...,gn) ( ). Pick a nontrivial weighted order ν. If ∈ ∩ ∩ L A ∗ n one of(1), (2)or(3) holds, then proposition VII.38 implies that V (In (f ),..., In (f )) (k ) is empty ν 1 ν n ∩ if g1,...,gn are generic. So assume (4) holds. Condition (4a) implies that Inν ( j1 ) and Inν ( j2 ) have −1 −1 P P no common non-invertible factor in k[x1, x1 ,...,xn, xn ]. This implies that V (Inν ( j1 ), Inν ( j2 )) is ∗ 3 P P either empty or have codimension 2 in (k ) . Therefore, by choosing generic gj3 , it is possible to ensure ∗ 3 that V (In ( ), In ( )) V (In (g )) (k ) = , which completes the proof of ( ) implication. ν Pj1 ν Pj2 ∩ ν j3 ∩ ∅ ⇐ For the opposite implication, assume there is a nontrivial weighted order ν such that Inν ( j ) is positive dimensional for each j, and one of the following holds: P

(i) either Inν ( j ) j = for each j, (ii) or there arePj =∩Bj such∅ that 1 6 2 (1) thereis a positive dimensionalpolytope which is a Minkowskisummandof each Inν ( jk ), k =1, 2, Q P (2) In ( ) = for each k =1, 2, and ν Pjk ∩Bjk ∅ (3) Inν ( j3 ) j3 = , where j3 is the single element of 1, 2, 3 j1, j2 . P ∩B 6 ∅ 0 { }\{ } ∗ n If (i) holds then one can choose (f ,f ,f ) ( ) such that In (f ) have a common zero on (k ) , 1 2 3 ∈ L A ν j and the zero would be unaffected by the coefﬁcients of monomials from the j, so that will not be able ∗ B B to ensure that (ND ) holds. On the other hand, if (ii) holds, then we can choose fj1 ,fj2 such that Inν (fj1 ) and Inν (fj2 ) have a common factor g with Newton polytope . Then for generic gj3 and generic fj3 −1 Q ∗ 3 ∈ π (g ), Bernstein’s theorem would imply that V (In (f , In (f , In (f ) (k ) V (g, In (f ) B3 j3 ν j1 ν j2 ν j3 ν j3 ∗ 3 ∗ ∩ ⊇ ∩ (k ) = , so that (f ,f ,f ) violates (ND ), as required. 6 ∅ 1 2 3 9. Notes All major results of this chapter are well known. A. Khovanskii in [BZ88, Section 27] gave a simple proof of Bernstein’s formula in zero characteristic. The main distinction between his proofand oursis in the handling of intersection multiplicity: he counts the numberof roots of systems which are nonsingular at the points of intersection, and then argues that every system can be deformed into such systems. We avoid this approach since in positive characteristics it would involve having as a technical overhead some versions of Bertini-type theorem every time we deal with intersection multiplicity. B. Huber and B. Sturmfels [HS95] gave a constructive proof of Bernstein’s theorem (in zero characteristic) which has had a deep impact on the “homotopy continuation” method to numerically compute the solutions of polynomial systems; the techniques of their proof also gives an efﬁcient way to compute mixed volumes. Our proof that the set of BKK non-degenerate polynomials is Zariski open follows the arguments from [Oka79], and our proof that it is nonempty comes from [Mon16]. J. M. Rojas [Roj94, Corollary 9] gave a characterization for strict monotonicity of mixed volume analogous to corollary VII.31.

5We say that a convex polytope P is a Minkowski summand of a convex polytope Q if there is a convex polytope R such that Q = P + R. CHAPTER VIII

(Weighted) Bezout´ theorems

In this chapter we use the results from chapter VII to prove B´ezout’s theorem (corollary VIII.3) and two of its classical generalizations: the weighted homogeneous version (theorem VIII.2) and the weighted multi-homogeneous version (theorem VIII.8). The weighted degrees considered in these results have the property that the weight of each variable is positive. In sections X.8 and X.9 we establish more general versions of these results involving arbitrary weighted degrees as special cases of the extension of Bernstein’s

n k theorem to k . We continue to assume that is an algebraically closed ﬁeld.

1. Weighted degree Let ω be an integral element of (Rn)∗. The weighted degree corresponding to ω, which by an abuse of notation we also denote by ω, is the map k[x1,...,xn] Z given by → ∪ {−∞} f max (ω) 7→ Supp(f) α α Given f = c x k[x ,...,x ], the leading form of f with respect to ω is ld (f) := c x . α α ∈ 1 n ω hω,αi=ω(f) α We say that f is weighted homogeneous with respect to ω (or in short, ω-homogeneous) if ω(xα) are equal P P for all α such that c =0, or equivalently, if Supp(ld (f)) = Supp(f). α 6 ω n n 2. P (ω) as a compactiﬁcation of k when the ωj are positive and ω0 =1 n+1 ∗ Assume ω is an integral element of (R ) with coordinates (ω0,...,ωn) with respect to the basis n+1 n dual to the standard basis of R such that each ωj is positive. The weighted projective space P (ω) corresponding to ω was constructed in section VI.10. In this section we treat the case that ω0 =1. For each n n j =0,...,n, let Uj := P (ω) V (xj ), be the “coordinate charts” of P (ω) considered in section VI.10.4. \ Since ω0 = 1, proposition VI.9 implies that the map (a1,...,an) [1 : a1 : : an] induces an

n n 7→ n ··· k isomorphism between k and U0, and therefore P (ω) is a compactiﬁcation of . The set of points at

n n n k inﬁnity (with respect to U0 = k ) on P (ω) is V (x0)= [0 : a1 : : an] : (a1,...,an) 0 = n−1 ∼ { ··· ∈ \{ }} ∼ P (ω1,...,ωn). Since ω0 = 1, for each polynomial f k[x1,...,xn], we can deﬁne its weighted ∈ ˜ ω(f) ω1 ωn homogenization with respect to ω as f := x0 f(x1/x0 ,...,xn/x0 ). It is straightforward to check that f˜ is ω-homogeneous and ldω(f˜) = ldω(f). n Proposition VIII.1. Identify k with U0.

˜ n n k (1) Let f k[x1,...,xn]. Then V (f) is the Zariski closure in P (ω) of V (f) . ∈ ⊂ (2) If f1,...,fk k[x1,...,xn], then the following are equivalent: k ∈˜ n (a) j=1 V (fj ) k = . \ ∅ n (b) there is no common zero of ld (f ),..., ld (f ) on k 0 . T ω 1 ω k \{ } PROOF. At ﬁrst we prove assertion (1). If f is a nonzero constant, then both V (f) and V (f˜) are empty. So assume f is a non-constant polynomial. If n = 1, then V (f) consists of ﬁnitely many points and it is ˜ 1 ˜ n straightforward to check that V (f)= V (f) P (ω0,ω1). So assume n 2. Since V (f) k = V (f), ⊂ ≥ ∩ theorem III.20 implies that each irreducible component of dim(V (f˜)) has dimension n 1. On the other hand, − ˜ n n ˜ (59) V (f) (P k )= V (f) V (x0)= V (ldω(f)) V (x0) ∩ \ ∩ ∩ n−1 Since ldω(f) is a nonzero polynomial and V (x0) ∼= P (ω1,...,ωn), theorem III.20 implies that each ˜ n irreducible component of V (f) k has dimension n 2, and therefore can not be an irreducible com- ˜ \ − ˜ n ponent of V (f). Consequently every irreducible component of V (f) intersects k and therefore contains

117 118 VIII. (WEIGHTED) BEZOUT´ THEOREMS an irreducible component of V (f). Since V (f˜) is Zariski closed in Pn(ω), this completes the proof of assertion (1). Assertion (2) follows from identity (59).

3. Weighted Bezout´ theorem We use the following notation throughout the rest of the book: for a nonnegative integer n we denote by [n] the set 1,...,n ; if I [n] and k is a ﬁeld (in most cases k will be either k or R), we write { } ⊆ (60) kI := (x ,...,x ) kn : x =0 if i I = k|I|, { 1 n ∈ i 6∈ } ∼ (61) (k∗)I := x =0 kI = (k∗)|I| { i 6 } ∩ ∼ Yi∈I where k∗ := k 0 . Note that k∅ = (k∗)∅ = 0 . \{ } { }

THEOREM VIII.2 (Weighted B´ezout theorem (theorem I.4)). Let ω be a weighted degreeon k[x1,...,xn] with positive weights ωj for xj , j = 1,...,n. Then the number of isolated solutions of polynomials n f1,...,fn on k is bounded above by ( j ω(fj ))/( j ωj ). This bound is exact if and only if the leading weighted homogeneous forms of f ,...,f have no common solution other than (0,..., 0). 1 Qn Q PROOF. Due to multiadditivity of intersection multiplicities (assertion (5) of proposition IV.16), we m may replace each fi by fi for an appropriate positive integer m and assume that mi,j := ω(fi)/ωj is an n integer for each i, j. Let i be the simplex in R with vertices at the origin and at mi,j ej , j = 1,...,n, P n iso where e ,...,e are the standard unit vectors in . Analogous to the deﬁnition of [ ,..., ] ∗ n in

1 n R 1 n k P P ( ) (52), deﬁne

iso iso

[ ,..., ] n := max [p ,...,p ] n : Supp(p ) , j =1,...,n k P1 Pn k { 1 n j ⊆Pj } It sufﬁces to show that iso (i) [ ,..., ] n = ( ω(f ))/( ω ), and 1 n k j j j j P P iso iso

(ii) [f ,...,f ] n = [ ,..., ] n if and only if thereis no commonzero of ld (f ),..., ld (f ) k 1 n k 1 n ω 1 ω n n PQ P Q on k 0 . \{ } Assertion (i) follows from Bernstein’s theorem (theorem VII.5) and the following claim.

Claim VIII.2.1. (1) MV( 1,..., n) = ( j ω(fj ))/( j ωj ). iso P Piso (2) [ ,..., ] n = [ ,..., ] ∗ n .

1 n k 1 n P P P P (k ) Q Q

ω(fj ) PROOF. Since j = 1 for each j = 1,...,n, the properties of mixed volume from theo- P ω(f1) P rem VII.2 imply that

ω(f ) ω(f ) n ω(f ) MV( ,..., ) = MV( , 2 ,..., n )= j MV( ,..., ) P1 Pn P1 ω(f )P1 ω(f ) P1 ω(f ) P1 P1 1 1 j=2 1 Y n n n n ω(f ) ω(f ) ω(f ) = j n! Vol ( )= j m = j ω(f ) n P1 ω(f ) 1,j ω j=2 1 j=2 1 j=1 j=1 j Y Y Y Y This proves the ﬁrst assertion of the claim. For the second assertion, let pj be an arbitrary polynomial iso iso supported at , j = 1,...,n. Since it is clear that [ ,..., ] n [ ,..., ] ∗ n , it sufﬁces to

j 1 n k 1 n (k ) P ′ P iso P ′ ≥ P′ iso P n show that there are p supported at such that [p ,...,p ] n [p ,...,p ] ∗ n . Write := ,

j 1 n k j j k Z j P ≤ 1 n ( ) A P ∩ j = 1,...,n. Pick ( ,..., )-non-degenerate (q ,...,q ) ( ,..., ), and set r (x, t) := A1 An 1 n ∈ L A1 An i

(1 t)p + tq . For each ǫ k, write r := r . Note that r = q and r = p . Pick a − i i ∈ ǫ,i i|t=ǫ 1,i i 0,i i generic ǫ k. Due to theorem VII.12 we may assume that (rǫ,1,...,rǫ,n) is also ( 1,..., n)-non- degenerate.∈ Since the intersection of an with a coordinatesubspace is a (nonempty)faceA of A, it is then Aj Aj straightforward to check (e.g. using remark VII.9) from the deﬁnition of ( 1,..., n)-non-degeneracythat

n ∗ n A A k rǫ,1,...,rǫ,n do not have any common zero on k ( ) . Assertion (4) of theorem IV.20 then implies iso iso \ iso

that [p ,...,p ] n [r ,...,r ] n = [r ,...,r ] ∗ n , as required. k 1 n k ǫ,1 ǫ,n ǫ,1 ǫ,n ≤ (k ) 4. PRODUCTS OF WEIGHTED PROJECTIVE SPACES 119

iso iso

Now we prove assertion (ii). At ﬁrst assume [f ,...,f ] n < [ ,..., ] n . We will show that k 1 n k 1 n P P n ′ the leading weighted homogeneous forms ldω(fi) of fi have a common zero on k 0 . Let ω := n n ′ \{ } (1,ω1,...,ωn). Embed k into P (ω ) via the map (x1,...,xn) [1 : x1 : : xn]. Set pi := fi, 7→ ··· i = 1,...,n. Let qi and ri := (1 t)pi + tqi be as in the proof of claim VIII.2.1. Let V be the (ﬁnite)

− n n+1 k set of common zeroes of q ,...,q on k and C be the union of irreducible components of 1 n ⊆ V (r1,...,rn) which intersect V 1 . Assertion (1) of theorem IV.20 implies that C is a curve. Since iso iso ×{ }

[f ,...,f ] n < [q ,...,q ] n , assertion (5) of theorem IV.20 implies that one of the following holds: k 1 n k 1 n n (iii) there is a positive dimensional component of V (f ,...,f ) k , or 1 n ⊂ (iv) C “has a point at inﬁnity at t = 0”, i.e. if C¯ is the closure of C in Pn(ω′) P1, then C¯ × ∩ ((Pn(ω′) X) 0 ) = . \ ×{ } 6 ∅ n ′ ˜ Denote the weighted homogeneous coordinates on P (ω ) by [x0 : : xn]. Let fi and q˜i be the weighted ′ ··· homogenization with respect to ω respectively of fi and qi. Proposition VIII.1 implies that the closures of n ′ ˜ V (fi) and V (qi) in P (ω ) are respectively V (fi) and V (˜qi). Since ω(fi) = ω(qi) for each i, it follows n ′ ˜ that the closure of V (ri) in P (ω ) k is V (˜ri), where r˜i := (1 t)fi + tq˜i. If (iii) holds, then the closure n ′ × ′ n ′ n − of V (f1,...,fn) in P (ω ) contains a point a P (ω ) k . The weighted homogeneous coordinates of ′ ∈ \ n ˜ ′ a are of the form [0 : a1 : : an] with a = (a1,...,an) k 0 . Then fi(a )=0, and therefore ··· ∈ \{ ′ } ¯ n ′ ldω(fi)(a)=0 for each i. On the other hand, if (iv) holds, then let (a , 0) C ((P (ω ) X) 0 ). ′ ∈ ∩ n \ ×{ } Since r˜ (a , 0)=0 for each i, this again yields a common zero of the ld (f ) on k 0 , as required. i ω i \{ } It remains to show the necessity of the non-degeneracy condition. Assume the leading weighted ho- n mogeneous forms ldω(fi) of fi have a common solution a k 0 . As in section VII.4.2, pick BKK n ∈ \{ } non-degenerate g1,...,gn with a commonzero b k such that for each i, NP(gi)= i and ldω(gi)(a) =

n ∈ P 6 n

k k 0 = fi(b). Deﬁne a rational curve C on k via the parametrization c(t) := (c1(t),...,cn(t)) : from6 (46) with ν := ω, i.e. → − c (t) := a t−ωj + (b a )t−ωj +kj , j =1, . . . , n. j j j − j where each k is a positive integer. Deﬁne h ,...,h as in (47) with m = ω(f )= ω(g ), i.e. j 1 n j − j − j h := tω(fj )f (c(t))g tω(gj )g (c(t))f j j j − j j ω(fj ) ω(gj ) Note that t fj (c(t)) and t gj (c(t)) are polynomials in t. The same arguments as in the alternate iso iso

proof of necessity of BKK non-degeneracyin section VII.4.2.2 show that [f ,...,f ] n < [g ,...,g ] n , k 1 n k 1 n so that the weighted homogeneous bound is not exact. Corollary VIII.3 (B´ezout’s theorem (theorem I.2)). The number of isolated solutions of polynomials n f1,...,fn on k is bounded above by j deg(fj ). This bound is exact if and only if the leading homogeneous forms of f ,...,f have no common solution other than (0,..., 0). 1 n Q PROOF. In theorem VIII.2 take ω to be the usual degree of polynomials. Combining the arguments of the proof of assertion (2) of claim VIII.2.1 with theorems III.20, VII.5 iso and VII.30 gives the following characterization of polytopes ,..., such that [ ,..., ] n =

1 n 1 n k iso n P H P P P [ 1,..., n] ∗ n . Given a coordinate subspace H of , we write T ( ) := j : j H = [n]. k R P P ( ) P { P ∩ 6 ∅} ⊂ n Lemma VIII.4. Let 1,..., n be convex integral polytopes in R . Then the following are equivalent: P P ≥0 iso iso

(1) [ ,..., ] n = [ ,..., ] ∗ n . k P1 Pn P1 Pn (k ) (2) For each proper coordinate subspace H of Rn, one of the following is true: ′ (a) either there is a coordinate subspace H′ of Rn such that H′ H and T H ( ) < dim(H′), or ⊃ | P | (b) T H ( ) is nonempty and H : j T H ( ) is a dependent collection of polytopes. P {Pj ∩ ∈ P } 4. Products of weighted projective spaces In this section we examine the closures of the hypersurfaces in a product of weighted projective spaces. This would be useful in the proof of the weighted multi-homogeneous B´ezout bound (theorem VIII.8). Let ¯ s nj X := j=1 P (ωj ), where each ωj is a weighted degree on Aj := k[xj,0,...,xj,nj ] such that the weight Q 120 VIII. (WEIGHTED) BEZOUT´ THEOREMS

ω of x is positive for each j, k. Let A := k[x :1 j s, 0 k n ], and for each j, let ω˜ be j,k j,k j,k ≤ ≤ ≤ ≤ j j the trivial extension of ωj to A, i.e.

ωk,l if k = j, ω˜j (xk,l)= (0 otherwise. We say that a polynomial h A is (weighted multi-) homogeneous with respect to Ω := ω ,...,ω , or ∈ { 1 s} in short, f is Ω-homogeneous, if it is ω˜j -homogeneous for each j =1,...,s. If h is Ω-homogeneous, then V (h) := a X¯ : h(a)=0 is a well deﬁned Zariski closed subset of X¯. { ∈ }

4.1. The case that ωj,0 =1 for each j =1,...,s. In this section we consider the case that ωj,0 = nj nj ¯ 1 for each j. In this case each P (ωj ) is a compactiﬁcation of k , and therefore X is a compactiﬁcation

n s n ¯ s k of k , where n := j=1 nj . More precisely, can be identiﬁed with X V ( j=1 xj,0), and we may

\ n k treat B := k[x : 1 j s, 1 k n ] A as the coordinate ring of . Let f B. Then j,k P ≤ ≤ ≤ ≤ j ⊂ Q ∈ Ω-homogenization f˜ of f is formed by substituting xj,k/xj,0 for xj,k in f for each j, k, and then clearing out the denominator. The following result is the analogue of assertion (1) of proposition VIII.1, and follows from the same arguments. ˜ ¯ Proposition VIII.5. Let f B. If k is algebraically closed, then V (f) is the Zariski closure in X of n ∈ V (f) k . ⊂ ¯ ¯ n s The set of points at infinity on X is X k = V ( x ) = Y , where the union is over all \ j=1 j,0 J J nonempty subsets J of [s] := 1,...,s , and Y are defined as follows: { } J Q S nj −1 ′ nj YJ = V (xj,0 : j J) V ( xj,0) = P (ω ) k ∈ \ ∼ j × jY6∈J jY∈J jY6∈J ′ where ωj are the restriction of ωj to k[xj,1,...,xj,nj ] for each j. Fix a nonempty subset J of [s]. Given ˜ α f B, we would like to compute the points at infinity on V (f) which belong to YJ . If f = α cαx , we ∈ α α write ldΩ,J (f) for be sum of cαx over all α such that ω˜j(x )=˜ωj(f) for each j J. In other words, ∈ P ld (f) is obtained from f˜ by substituting x =0 for each j J and x =1 for each j J. Ω,J j,0 ∈ j,0 6∈ Example VIII.6. Let s = 3, n1 = n2 = 1 and n3 = 2 (so that n = 4). Let ω1,ω2 be the usual degree in respectively x1,1 and x2,1 coordinates, and ω3 be the weighted degree in (x3,1, x3,2) coordinates 5 7 5 7 3 5 2 corresponding to weights 2 for x3,1 and 3 for x3,2. Let f = x1,1 + x2,1 + x1,1x2,1 + x3,1 + x1,1x3,2. Then ω1(f)=5, ω2(f)=7, ω3(f)=6, and

5 5 7 5 2 x1,1 + x1,1x2,1 + x1,1x3,2 if J = 1 , 7 5 7 { } x2,1 + x1,1x2,1 if J = 2 , 3 5 2 { } x3,1 + x1,1x3,2 if J = 3 , ldΩ,J (f)=  { } x5 x7 if J = 1, 2 ,  1,1 2,1  5 2 { } x1,1x3,2 if J = 1, 3 ,  { } 0 if J = 2, 3 or J = 1, 2, 3 .  { } { }  Note the following difference from the weighted homogeneous case: if J 2, then it might happen that | |≥ ldΩ,J (f)=0 even if f is a nonzero polynomial. Since ld (f) is ω˜ -homogeneous for each j J, it deﬁnes a well deﬁned Zariski closed subset of Ω,J j ∈ YJ . It is straightforward to check that this set is precisely the intersection of V (f˜) and YJ , which is the content of the next result. ˜ Proposition VIII.7. Assume k is algebraically closed. If f B, then V (f) YJ = V (ldΩ,J (f)) YJ . If f ,...,f B, then the following are equivalent ∈ ∩ ∩ 1 k ∈ k ˜ n (1) V (f ) k = . j=1 j \ ∅ (2) For each nonempty subset J of [s], there is no common zero of ldΩ,J (f1),..., ldΩ,J (fk) on

T nj nj k (k 0 ) . j∈J \{ } × j6∈J Q Q 5. WEIGHTED MULTI-HOMOGENEOUS BEZOUTTHEOREM´ 121

5. Weighted multi-homogeneous Bezout´ theorem

We now generalize the weighted B´ezout bound to the multi-projective setting. Let I := (I1,...,Is) be an ordered partition of [n] := 1,...,n , i.e. [n]= I and I = n. For each j =1,...,s, let { } j j j | j |

ωj be a weighted degree on k[xk : k Ij ] with positive weights ωj,k for xk, k Ij . Let f1,...,fn be n ∈ S P ∈ polynomials on k . Given di,j := ωj (fi), we wouldlike tocomputea (sharp)upperboundofthenumberof isolated solutions of f1,...,fn. Let nj := Ij and lj be the least common multiple of ωj,1,...,ωj,nj and nj | | s j be the simplex in R≥0 defined by α : ωj , α lj . Note that Supp(fi) i := j=1(di,j /lj) j S n { h i≤ } ⊆P S ⊂ R . By definition MV( 1,..., n) is the coefficient of λ1 λn in the polynomial P P ··· Q n n s s n s n s d d d Vol ( λ ) = Vol ( λ i,j ) = Vol ( ( λ i,j )) = ( λ i,j )nj Vol ( ) n iPi n i l Sj n i l Sj i l n Sj i=1 i=1 j=1 j j=1 i=1 j j=1 i=1 j j=1 X X Y Y X Y X Y s nj s n s n nj (lj /ωj,k) ( λidi,j ) j=1 k=1 nj j=1 i=1 = ( λidi,j ) = s (n !lnj ) s (n ! nj ω ) Q jQ=1 j j j=1 i=1 Qj=1 Pj k=1 j,k Y X s n The coefficient of λ Qλ in ( λ d )nj is the permanentQ of theQ following n n matrix: 1 ··· n j=1 i=1 i i,j × Q P n1 times ··· ns times d d d d 1,1 ··· 1,1 ······ 1,s ··· 1,s D(I , d~) := z. }| . { z }| { z. }| . {  . . . .  dn,1 dn,1 dn,s dn,s  ··· ······ ···  The preceding observations together with lemma VIII.4 imply that  I ~ iso iso perm(D( , d))

(62) n n k [f1,...,fn]k [ 1,..., n] = MV( 1,..., n)= ≤ P P P P ( j nj!)( j,k ωj,k)

Note that when s =1, then Ω consists of only one weighted degree ω and perm(Q D(QI , d~)) = n!ω(f1) ω(fn), n ··· so that the bound from (62) is precisely the weighted Bézout bound j=1(ω(fj )/ωj). Now we determine ′ the condition for the attainment of this bound. For each j =1,...,s, fix a new indeterminate uj . Let Bj be Q s nj ′ the ring of polynomials (over k) in uj and xk, k Ij . Note that each XPi is isomorphic to P (ω ), ∈ j=1 j where ω′ is the weighted degree on B′ such that ω′ (u )=1 and ω′ (x ) = ω , k I . For each j j j j j k j,k ∈Q j g k[x1,...,xn] and each nonempty subset J of [s], define ldΩ,J (g) as in section VIII.4.1. Finally, note

∈ n s Ij k I k that since is a partition of [n], we may identify with j=1 . THEOREM VIII.8 (Weighted multi-homogeneous B´ezoutQ theorem). The number of isolated solutions n of polynomials f1,...,fn on k is bounded by (62). This bound is exact if and only if the following holds: Ij for each nonempty subset J of [s], there is no common zero of ld (f ),..., ld (f ) on (k Ω,J 1 Ω,J n j∈J \

Ij n s Ij

k k 0 ) k (upon identiﬁcation of with ). { } × j6∈J j=1 Q

PROOFQ . We have already proved that (62) holds.Q Let f˜i be the Ω-homogenization of fi deﬁned as in section VIII.4.1. If the bound is not exact, then it follows by the same arguments as in the proof of s ˜ n theorem VIII.2 that V (f ) k = , and then proposition VIII.7 implies that the condition in the j=1 j \ 6 ∅ second assertion of theorem VIII.8 is violated. Now assume there is a nonempty subset J of [s] such

T Ij Ij k that ldΩ,J (f1),..., ldΩ,J (fn) have a common zero a = (a1,...,an) j∈J (k 0 ) j6∈J . m ∈ \{ } × Replacing each fi by f for some appropriate positive integer m, we may assume that the vertices of each i Q Q i have integer coordinates. Let I := i : ai = 0 [n]; in other words, I is the smallest subset of [n] P ∗ I { 6 } ⊆ such that a (k ) . Since J is nonempty, it follows that I is also nonempty; in fact Ij I = for each j J. For each∈ j J, let be the facet of determined by ω , α = l . Fix i [n]. Deﬁne∩ 6 ∅ ∈ ∈ Tj Sj h j i j ∈

I i := R (di,j /lj) j (di,j /lj) j Q ∩  T × S  jY∈J jY6∈J   Then i is a proper (nonempty) face of i. Let fi,Qi be the component of fi supported at i, i.e. if Q α, then c xPα. It is straightforward to check that f (a)=0 forQ each i. fi = α cαx fi,Qi = α∈Qi α i,Qi P P 122 VIII. (WEIGHTED) BEZOUT´ THEOREMS

n ∗ Now choose an integral element ν (R ) such that i = Inν ( i) for each i [you have to check that ∈ Q P n such ν exists!]. As in section VII.4.2, pick BKK non-degenerate g1,...,gn with a common zero b k ∈ n such that for each i, f (b) = 0, NP(g ) = and In (g )(a) = 0. Deﬁne a rational curve C on k i 6 i Pi ν i 6 via the parametrization from (46) and deﬁne h1,...,hn as in (47) with mi = ν(gi) for each i. Then −mj −mj t fj (c(t)) and t gj (c(t)) are polynomials in t and the same arguments as in section VII.4.2 show iso iso

that [f ,...,f ] n < [g ,...,g ] n , as required. k 1 n k 1 n 6. Notes A version of the weighted Bézout theorem appears in [Dam99]. I. Shafarevich gave the bound for the “multi-homogeneous” case (i.e. the case that all weights are 1) of the weighted multi-homogeneous Bézout theorem in the first edition of [Sha94] in the 1970s; see also [MS87]. We could not locate any past reference for the non-degeneracy condition or the estimate for the general weighted multi-homogeneous version. Part 2

Extension of Bernstein-Kushnirenko Theorem to the afﬁne space and other applications CHAPTER IX

Intersection multiplicity at the origin

1. Introduction In this chapter we consider the “local” version of the affine Bézout problem, i.e. the problem of esti- mating the intersection multiplicity of generic hypersurfaces at the origin. This computation is a crucial ingredient of the extension in chapter X of Bernstein’s theorem to the affine space. Recall that the support α of a power series f = α cαx k[[x1,...,xn]] is Supp(f) := α : cα =0 and we say that f is sup- n ∈ { 6 } n ported at Z if Supp(f) . Now let 1,..., n be (possibly infinite) subsets of Z≥0. In the case A⊂ P ⊂ A A A iso that are finite, we saw in chapter VII that within all f supported at , j =1,...,n, [f ,...,f ] ∗ n Aj j Aj 1 n (k ) takes the maximum value when f1,...,fn are generic. It is possible to talk about “generic” power series supported at even if is infinite, and it turns out that the intersection multiplicity [f ,...,f ] Aj Aj 1 n 0 of f1,...,fn at the origin takes the minimum value when fj are generic power series supported at j , j = 1,...,n (see theorem IX.8 for the precise statement); in this chapter we compute this minimumA and give a Bernstein-Kushnirenko type characterization of the systems which attain the minimum.

2. Generic intersection multiplicity

n Let j be a (possibly inﬁnite) subset of Z , j =1,...,n. Deﬁne A ≥0

(63) [ ,..., ] := min [f ,...,f ] : j, f k[[x ,...,x ]], Supp(f ) A1 An 0 { 1 n 0 ∀ j ∈ 1 n j ⊆ Aj } In this section we motivate, state and illustrate the formula for [ 1,..., n]0. Its proof is given in section IX.5. A A

A1 B1 B2

FIGURE 1. [ , ] =0 for any since a generic f supported at has a nonzero A1 A2 0 A2 1 A1 constant term; [ 1, 2]0 = since every gj supported at j , j = 1, 2, identically vanishes on the xB-axis.B ∞ B

2.1. Motivation of the formula. In this section we study informally the case k = C and try to motivate the formula for [ ,..., ] . It is not hard to understand precisely when [ ,..., ] is zero A1 An 0 A1 An 0 or infinity - see fig. 1 and corollary IX.13. So consider the case that 0 < [ 1,..., n]0 < , and pick f supported at , j =1,...,n, such that [f ,...,f ] = [ ,..., ] .A Then theA origin is∞ an isolated j Aj 1 n 0 A1 An 0 point of f1 = = fn = 0. Therefore theorem IV.12 and proposition IV.16 imply that near the origin f = = f ···=0 defines a curve C, and 2 ··· n (64) [f ,...,f ] = ord (f ) 1 n 0 0 1|B XB where the sum is over all “branches” B of C at the origin. We now try to compute this sum. So fix a branch B of C at the origin and an analytic parametrization γ = (γ1,...,γn): U B of B, where U is →n a neighborhood of the origin on C. Let HB be the smallest coordinate subspace of C containing B.

124 2. GENERIC INTERSECTION MULTIPLICITY 125

n 2.1.1. Case 1: HB = C . In this case no γi is identically zero, so that each γi can be expressed as γ = a tνi + i 1 ··· ∗ where ai C , t is an analytic coordinate on U, and νi is the order (in t) of γi. Note that each νi is ∈ positive, since the center of B is at the origin. Let νB be the weighted order on k[x1,...,xn] such that νB(xi)= νi, i =1,...,n. Then for each j, f (γ(t)) = In (f )(a ,...,a )tν(fj ) + j νB j 1 n ··· where In ( ) denotes the initial form with respect to ν , and the omitted terms have higher order in t. νB · B Since fj(γ(t)) 0 for j = 2,...,n, we have a system of (n 1) weighted homogeneous polynomial equations: ≡ −

(65) InνB (fj )(a1,...,an)=0, j =2, . . . , n. Theorem VII.5 implies that the number of solutions of (65) is the (n 1)-dimensional mixed volume of − the Newton polytopes of InνB (fj ), j =2,...,n. It is in fact “reasonable” to guess that in the generic case each solution of (65) corresponds to a distinct branch B of C and for each such B, the order of f at the 1|B origin is νB(f1). In other words, for each weighted order ν,

(66) ord0(f1 B)= ν(f1) MV(Inν (f2),..., Inν (fn)) n | HB =C νXB =ν n 2.1.2. Case 2: HB ( C . In this case we may assume without loss of generality that HB is the I coordinate subspace spanned by x1,...,xk, k

(67) In (f I )(a ,...,a )=0, j =2, . . . , k. νB j |C 1 k

where νB is the weighted order on k[x1,...,xk] corresponding to weights νi for xi, i = 1,...,k. Asin the preceding case, the number of solutions of (67) in the generic situation is the (k 1)-dimensional mixed − volume of the Newton polytopes of InνB (fj CI ), j =2,...,k, and each solution corresponds to a distinct branch B of C. However, each branch should| be counted with proper multiplicity, and therefore this mixed I volume should be multiplied by the “intersection multiplicity of fk+1,...,fn along C .” It turns out (see corollary IV.22) that for generic fk+1,...,fn, this is precisely the intersection multiplicity at the origin ∗ of fk+1 (x1,...,xk)=(ǫ1,...,ǫk),...,fn (x1,...,xk)=(ǫ1,...,ǫk), where ǫ1,...,ǫk are generic elements from C . If n | n−k | π : R R is the projection onto the last (n k)-coordinates, then the genericness of the ǫi imply that the→ support of f is precisely− π( ) for each j, and it is reasonable to guess k+j |(x1,...,xk)=(ǫ1,...,ǫk) Ak+j that if the fk+j and ǫi are generic, then [f ,...,f ] = [π( ),...,π( )] k+1|(x1,...,xk)=(ǫ1,...,ǫk) n|(x1,...,xk)=(ǫ1,...,ǫk) 0 Ak+1 An 0

It then follows as in the ﬁrst case that for each weighted order ν on k[x1,...,xk],

Therefore [ 1,..., n]0 should be the sum of the right hand side of identity (68) over all appropriate I and ν. TheoremA IX.1A states that this precisely the case, and the proof of theorem IX.1 in section IX.5.1 simply makes the preceding arguments rigorous.

A1 B1 B2

FIGURE 2. Newton diagrams (in red) of the sets from ﬁg. 1

2.2. Precise formulation. Let be a (possibly infinite) subset of Zn . The convexhull of +Rn in A ≥0 A ≥0 Rn is a convex polyhedron (corollary B.38); the Newton diagram ND( ) of is the union of the compact A A faces of this polyhedron (fig. 2). The Newton diagram of a power series f in (x1,...,xn), denoted ND(f), is the Newton diagram of Supp(f); it is the local analogue of the Newton polytope of a polynomial. Given n diagrams Γ1,..., Γn in R , define n n (69) [Γ1,..., Γn]0 := min [f1,...,fn]0 : j, fj k[[x1,...,xn]], ND(fj )+ R Γj + R { ∀ ∈ ≥0 ⊆ ≥0} We will see in theorem IX.1 below that [Γ1,..., Γn]0 can be expressed in terms of certain mixed volumes n of the faces of Γj , and if Γj are Newton diagrams of j Z , then [ 1,..., n]0 = [Γ1,..., Γn]0. A ⊆ ≥0 A A First we need to introduce some notation. We write [n] := 1,...,n . If I [n] and k is a field, I { } n ⊆ recall that k is the I -dimensional coordinate subspace (x1,...,xn) k : xi = 0 if i I , and ∗ I | | I { n ∈ I I 6∈ } (k ) = (x1,...,xn) k : xi = 0 if i I . For R , we write := R . We denote by π : kn { kI the projection∈ in the coordinates6 ∈ indexed} S by ⊂I, i.e. S S ∩ I → xj if j I (70) the j-th coordinate of πI (x1,...,xn) := ∈ 0 if j I. ( 6∈

Let ν be a weighted order on k[x1,...,xn] corresponding to weights νj for xj , j = 1,...,n. We n ∗ identify ν with the element in (R ) with coordinates (ν1,...,νn) with respect to the dual basis. We say that ν is centered at the origin if each νi is positive and that ν is primitive if it is nonzero and the greatest common divisor of ν1,...,νn is 1. If ν is centered at the origin, then it also extends to a weighted order on the ring of power series in (x1,...,xn). We write 0 for the set of weighted orders centered at the origin ′ V n and for the primitive elements in 0. Given polytopes Γ1,..., Γn in R , define V0 V ∗ ′ (71) [Γ1,..., Γn]0 := min(ν) MVν (Inν (Γ2),..., Inν (Γn)) Γ1 ν∈V′ X0 where MV′ ( ,..., ) is defined as in (54). ν · · n THEOREM IX.1 ([Mon16]). Let := ( 1,..., n) be a collection of subsets of Z≥0 and Γj be the A A A I I Newton diagram of j , j =1,...,n. For each I [n], let TA := j : j = be the set of all indices j A I ⊂ n { A 6 ∅} such that j touches the coordinate subspace R of R . Define A (72) T := I [n]: I = , T I = I , 1 T I A,1 { ⊆ 6 ∅ | A| | | ∈ A} Then (1) If 0 Γ and there is I [n] such that T I < I , then 6∈ j j ⊂ | A| | | S [ 1,..., n]0 = [Γ1,..., Γn]0 = A A ∞ (2) Otherwise I I I ∗ (73) [ 1,..., n]0 = [Γ1,..., Γn]0 = [Γ , Γ ,..., Γ ] [π (Γ ′ ),...,π (Γ ′ )]0 1 j2 j|I| 0 [n]\I j1 [n]\I jn−|I| A A T × I∈XA,1 2. GENERIC INTERSECTION MULTIPLICITY 127

where for each I T , j =1, j ,...,j are elements of T I , and j′ ,...,j′ are elements ∈ A,1 1 2 |I| A 1 n−|I| of [n] T I . \ A Remark IX.2. The productof 0 and , when/ifit occursin (73), isdefinedto be 0. Also empty intersection ∞ productsand mixed volumes are defined as 1. In particular, when n =1, theterm MV(Inν (Γ2),..., Inν (Γn)) from (71) is defined to be 1.

Remark IX.3 (Generic intersection multiplicity is monotonic). The formulae for [ 1,..., n]0 from the- n A A orem IX.1 do not change if the j are replaced by j + R≥0. This immediately implies that [ ,..., ]0 ′ A n A ′ ′ · · is monotonic, i.e. if j + R , j = 1,...,n, then [ ,..., ]0 [ 1,..., n]0. Precise Aj ⊆ A ≥0 A1 An ≥ A A characterization of the cases for which [ ′ ,..., ′ ] > [ ,..., ] is given in theorem IX.32. A1 An 0 A1 An 0 It is not obvious from the outset that the term computed by (73) is invariant under the permutations of the j . Some formulae which are invariant under permutations of the j are given in section IX.7. We nowA present an example to illustrate this invariance. A

1 2 (1,1,1) (1,1,1) 2 0.5 1 1 (2,2,1)

0.6 0.8 1 2 3 1.5 2 0.2 0.4 1 0.5 1

0.5 1 2 1 2

Γ1 Γ2 Γ3

FIGURE 3. Newton diagrams of polynomials from example IX.4 and inner normals to their facets

Example IX.4. Consider the polynomials f1,f2,f3 from example VII.28. If := (Supp(f1), Supp(f2), Supp(f3)), then it is straightforward to check (see ﬁg. 3) that T = 1, 2, 3 , 3 ,soA that (73) implies that A,1 {{ } { }} ∗ {3} ∗ [f1,f2,f3]0 = [Γ1, Γ2, Γ3]0 + [π{1,2}(Γ2), π{1,2}(Γ3)]0 [Γ1 ]0

The Newton diagrams of π{1,2}(Γ2) and π{1,2}(Γ3) are the same diagram consisting of the line segment from (1, 0) to (0, 1), which is the Newton diagram of linear polynomials with no constant terms. It follows that [π{1,2}(Γ2), π{1,2}(Γ3)]0 =1. The Newton diagram of Γ2 +Γ3 has two facets with inner normals in 3 (R>0) , and these inner normals are ν1 := (1, 1, 1) and ν2 := (2, 2, 1) (see ﬁg. 4). Then it follows from ﬁg. 5 and identity (40) that ′ ′ [f1,f2,f3]0 = min(ν1) MVν1 (Inν1 (Γ2), Inν1 (Γ3)) + min(ν2) MVν2 (Inν2 (Γ2), Inν2 (Γ3))+1 ordz(f1 x=y=0) Γ1 Γ1 · |

= min(1, 1, 1) Area( ) + min(2, 2, 1) Area( )+1 1=1 4+1 1+1=6. Γ1 · Γ1 · · · · ′ On the other hand with := (Supp(f ), Supp(f ), Supp(f )), one has T ′ = 1, 2, 3 . Since A 3 1 2 A ,1 {{ }} the Newton diagram of Γ1 +Γ2 has only one facet and that the primitive inner normal to that facet is ν1, we have from ﬁg. 5 and identity (40) that ∗ ′ [f1,f2,f3]0 = [Γ3, Γ1, Γ2]0 = min(ν1) MVν1 (Inν1 (Γ1), Inν1 (Γ2)) Γ3

= min(1, 1, 1) Area( )=2 3=6. Γ3 · · ′′ Similarly, with := (Supp(f ), Supp(f ), Supp(f )), one has T ′′ = 1, 2, 3 . The Newton A 2 3 1 A ,1 {{ }} diagram of Γ3 +Γ1 have two facets, with inner normals ν1 and ν2, we have from ﬁg. 5 and identity (40) 128 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

2 2 2

4 5 3 3 3 4 2 1 2 1 2 1

2 2 2 4 4

ND(Γ1 +Γ2) ND(Γ2 +Γ3) ND(Γ3 +Γ1)

FIGURE 4. Sum of the Newton diagrams of polynomials from example IX.4 that ∗ [f1,f2,f3]0 = [Γ2, Γ3, Γ1]0 ′ ′ = min(ν1) MVν1 (Inν1 (Γ3), Inν1 (Γ1)) + min(ν2) MVν2 (Inν2 (Γ3), Inν2 (Γ1)) Γ2 Γ2 = min(1, 1, 1) Area( ) + min(2, 2, 1) Area( )=3 2+4 0=6. Γ2 · Γ2 · ∅ · ·

ν1 = (1, 1, 1)

Inν1 (Γ1) Inν1 (Γ2) Inν1 (Γ3) Inν1 (Γ1 +Γ2) Inν1 (Γ2 +Γ3) Inν1 (Γ3 +Γ1)

ν2 = (2, 2, 1)

Inν2 (Γ1) Inν2 (Γ2) Inν2 (Γ3) Inν2 (Γ1 +Γ2) Inν2 (Γ2 +Γ3) Inν2 (Γ3 +Γ1)

FIGURE 5. Normalized faces of the diagrams of example IX.4

3. Characterization of minimal multiplicity systems

n Given a collection = ( 1,..., m) of (possibly inﬁnite) subsets of Z , we write 0( j ) for the A A A ≥0 L A space of all power series in (x ,...,x ) supported at , j =1,...,n, and ( ) := n ( ). For 1 n Aj L0 A j=1 L0 Aj the case that m = n, in this section we characterize the systems (f1,...,fn) 0( ), which achieve the minimum possible intersection multiplicity at the origin. The proofs of the results∈ L ofA thisQ section are given in sections IX.4, IX.6 and IX.8.

3.1. Non-degeneracy at the origin. As in the case of Bernstein’s theorem, we try to guess the correct non-degeneracy condition by considering the case k = C. Also assume for simplicity that each j is A

ﬁnite, so that every power series supported at is in fact a polynomial. Now pick f ,g k[x ,...,x ] Aj j j ∈ 1 n over C such that j = Supp(fj) Supp(gj) for each j, and [f1,...,fn]0 > [g1,...,gn]0. Write A ⊇ n hj := (1 t)fj + tgj , j = 1,...,n. Then it seems reasonable to expect that there is a curve C(t) on C such that −h (C(t)) =0 and lim C(t)=0 (see ﬁg. 6). Pick a parametrization U C(t), where U is a j t→0 → neighborhood of the origin on C the form γ : t (γ1(t),...,γn(t)), where each γi is a power series in t. 7→ Fix j =1,...,n. As in section VII.3.2, we examine the initial part of the expansion of hj (γ(t)). 3. CHARACTERIZATION OF MINIMAL MULTIPLICITY SYSTEMS 129

∗ n 3.1.1. Base case. At ﬁrst we consider the case that the image of γ intersects (C ) , i.e. no γi is νi ∗ identically zero. Let γi = ait + , where ai C , and νi := ordt(γi). Let ν be the weighted ··· ∈ order on k[x1,...,xn] corresponding to weights νi for xi, i = 1,...,n. Then it follows exactly as in section VII.3.2 that In (f )(a)=0 for each j =1,...,n, where In ( ) are deﬁned as in (43). Since Aj ,ν j Aj ,ν · limt→0 γ(t)=0, it follows in addition that each νj is positive, i.e. ν is centered at the origin. This leads to the following notion.

n Deﬁnition IX.5. Let := ( 1,..., m) be a collection of (possibly inﬁnite) subsets of Z and (f1,...,fm) A A A ≥0 ∈ 0( ). We say that f1,...,fm are ( , )-non-degenerate at the origin if they satisfy the following condi- Ltion:A A ∗

∗ for each weighted order ν centered at the origin, there is (ND0) ∗ n no common root of InAj ,ν (fj ), j =1,...,m, on (k ) .

We say that f1,...,fm are -non-degenerate at the origin if they are ( , )-non-degenerate at the origin with := (Supp(f ),..., Supp(∗ f )). B ∗ B 1 m

2 2 3 (f1, f2)=(y − x,y − x − x ) 2 2 3 (g1,g2)=(6y − x,y − x − x )

t = 1 t = 0.4 t = 0.2 t = 0 ND(f1) = ND(g1) ND(f2) = ND(g2)

FIGURE 6. One of the common roots of (1 t)f + tg =0, j =1, 2, approaches the origin as t 0 − j j →

3.1.2. General case. So far we ignored the possibility that some of the γj can be identically zero. This happens if γ(t) belongs to a proper coordinate subspace of Cn. Incorporating this possibility and running the same arguments as in the ﬁrst case leads to the following notion.

n Deﬁnition IX.6. Let := ( 1,..., m) be a collection of (possibly inﬁnite) subsets of Z and (f1,...,fm) A A A ≥0 ∈ I

0( ). For each I [n] and each j, write fj k for the power series obtained from fj by substituting 0 L A ⊆ I | I I I I for each xk such that k I, and write := ( 1,..., m) = ( 1 R ,..., m R ). We say that f ,...,f are -non-degenerate6∈ at theA origin ifA they satisfyA the followingA ∩ condition:A ∩ 1 m A I

I I k

k are -non-degenerate at (ND ) f1 ,...,fm ( , ) 0 the| origin for each| nonemptyA ∗ subset I of [n].

We say that f1,...,fm are non-degenerate at the origin if they are -non-degenerate at the origin with := (Supp(f ),..., Supp(f )). B B 1 m

The preceding discussion suggests that for k = C, -non-degeneracy at the origin is a sufﬁcient condition for minimum intersection multiplicity at the origin.A We will see that it is in fact necessary and

sufﬁcient for all (algebraically closed) k.

n 3.2. The results. Let := ( 1,..., n) be a collection of (possibly infinite) subsets of Z . The- A A A ≥0 orem IX.8 below states the necessary and sufficient condition for the minimality of [f1,...,fn]0 for (f1,...,fn) 0( ). It also states that [f1,...,fn]0 is minimal for “generic” (f1,...,fn) 0( ). We have to be∈ carefulL A about the notion of “genericness” though, since the spaces ( ) and∈ L (A ) L0 Aj L0 A are in general infinite dimensional vector spaces over k, and therefore they are not algebraic varieties. ′ ′ Let := ( 1 ND( 1),..., n ND( n)). Then 0( ) is an algebraic variety isomorphic to PjA|Aj ∩ND(Aj )|∩ A A ∩ ′A L A k . Let π : 0( ) 0( ) be the natural projection which “ignores” the coefficients L A → L A ′ corresponding to exponents not in ND( j ), j = 1,...,n. Write 0( ) (respectively, 0( )) for the A ′ M A M A set of all (f1,...,fn) in 0( ) (respectively, 0( )) with the minimum possible value for [f1,...,fn]0. L′ A L A ′ We will show that 0( ) is a nonempty Zariski open (and therefore Zariski dense) subset of 0( ), and ( )= π−1( M′ ). A L A M0 A M0 130 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

Remark IX.7. An ind-variety over a ﬁeld k is a set X along with a chain of subsets X0 X1 such that ⊂ ⊂ · · ·

(1) X = i Xi, (2) Each Xi is an algebraic variety over k, and .The inclusionsS X ֒ X are closed embeddings of algebraic varieties (3) i → i+1 It is not hard to see, taking arbitrary sequences of ﬁnite subsets such that = , Aj,0 ⊂ Aj,1 ⊂ · · · Aj i Aj,i that 0( ) is an ind-variety. The notion of Zariski topology has a natural extension to the case of ind- L A S varieties. Theorem IX.8 implies in particular that 0( ) is a nonempty dense open subset of 0( ) in this topology. M A L A

−1 ′ ′ THEOREM IX.8 ([Mon16]). 0( )= π ( 0) and 0( ) is a nonempty Zariski open subset of ( ′). If [ ,..., ] = , thenM A ( )= M( ). OtherwiseM A the following are equivalent: L0 A A1 An 0 ∞ M0 A L0 A (1) (f1,...,fn) 0( ) (2) f ,...,f are∈M-non-degenerateA at the origin. 1 n A Theorem IX.8 is proven in section IX.4. To check -non-degeneracy at the origin, one needs to check ( I , )-non-degeneracy for all (nonempty) subsets I ofA [n]. The following theorem, which we prove in sectionA ∗ IX.6, often signiﬁcantly limits the number of test cases. 1 THEOREM IX.9 ([Mon16]). Let := ( 1,..., m) be a collection of (possibly inﬁnite) subsets of n I A I A A I Z . For each I [n], let := j : = and IA := I [n]: I = , n I . Then for ≥0 ⊆ EA { Aj ∅} { ⊆ 6 ∅ |EA|≥ − | |} (f ,...,f ) ( ) the following are equivalent: 1 m ∈ L0 A (1) f1,...,fm are -non-degenerate at the origin. A I

I I I k (2) f k ,...,f are ( , )-non-degenerate at the origin for every I . 1| m| A ∗ ∈ A Remark IX.10. If m = n and 0 < [ ,..., ] < , then (due to corollary IX.13 below) A1 An 0 ∞ I = I [n]: I = , I = n I = I [n]: I = , T I = I A { ⊆ 6 ∅ |EA| − | |} { ⊆ 6 ∅ | A| | |} where T I := j : I = . A { Aj 6 ∅} n Now we go back to the m = n case, i.e. := ( 1,..., n) is a collection of subsets of Z . De- A A A ﬁne IA as in theorem IX.9. Similar to the characterization of strict monotonicity of mixed volume in corollary VII.31, we give (in theorem IX.32) a combinatorial characterization of strict monotonicity of [ 1,..., n]0. As a corollary in section IX.8 we prove the following result, which says that in the same A A ∗ n way as in the case of (k ) (theorem VII.11), -non-degeneracy of a system of power series at the origin equivalent to a combinatorial condition plus non-degeneraA cy at the origin with respect to their supports. Corollary IX.11. If [ ,..., ] < , then the following are equivalent for (f ,...,f ) ( ): A1 An 0 ∞ 1 n ∈ L0 A (1) f ,...,f are -non-degenerate at the origin. 1 n A (2) (a) for each nonempty subset I of [n], and each ν (Rn)∗ which is centered at the origin, the I I ∈ collection Inν (ND(fj ) R ):Inν ( ) Supp(fj ) = of polytopes is dependent, and { ∩ Aj ∩ 6 ∅} (b) f1,...,fn are non-degenerate at the origin. n ∗ (3) (a) for each I IA and each ν (R ) which is centered at the origin, the collection ∈ I I ∈ Inν (ND(fj) R ):Inν ( ) Supp(fj ) = of polytopes is dependent, and { ∩ Aj ∩ 6 ∅} (b) f1,...,fn are non-degenerate at the origin. 4. Proof of the non-degeneracy condition

In this section we prove theorem IX.8. Let := ( 1,..., m), m 1, be a collection of subsets of n ′ A A A I ≥ I Z≥0. Let 0( ), be as in section IX.3.2. Let I [n]; deﬁne TA := j : j = as in theorem IX.1. LI A AI ⊆ { A 6 ∅} Note that TA = TA′ . Lemma IX.12. Assume 0 . Then 6∈ j Aj I

k I I

k k (1) If TA < I , thenSdim ( [[xi : i I]]/ f1 k ,...,fm ) = for all (f1,...,fm) |( )|. | | ∈ h | | i ∞ ∈ L0 A 1Note that the number of subsets is m, which may be distinct from n. 4. PROOF OF THE NON-DEGENERACY CONDITION 131

I ∗ I ′ (2) If T I , then V (f ,...,f ) (k ) is isolated for generic f ,...,f ( ). | A| ≥ | | 1 m ∩ 1 m ∈ L0 A I ROOF I P . If T < I then for all (f ,...,f ) ( ), the number of f such that f k is nonzero | A| | | 1 m ∈ L0 A j j|

I k

is less than I . Since 0 for any j, each f k is in the maximal ideal of R := [[x : i | | 6∈ Aj j| I i ∈

I I k k I]]/ f1 k ,...,fm . Theorem III.20 implies that the transcendence degree of RI over is positive, so h | | i that dim k (R )= . The second assertion follows from Bernstein’s theorem. I ∞ Corollary IX.13 (cf. [Roj99, Lemma 2], [HJS13, Proposition 5]). Assume m = n. n (1) [ 1,..., n]0 =0 if and only if 0 i=1 i. A A ∈ n A I (2) [ 1,..., n]0 = if and only if 0 i and there is I [n] such that T < I . A A ∞ 6∈S i=1 A ⊆ | A| | | Let 0( ) be the set of all (f1,...,fm) S0( ) such that f1,...,fm are -non-degenerate at the origin. NoteN A that ( )= π−1( ( ′)), where∈ Lπ :A ( ) ( ′) is the naturalA projection. N0 A N0 A L0 A → L0 A ′ ′ m I Proposition IX.14. 0( ) is a Zariski open subset of 0( ). If either 0 i=1 i or TA I for N′ A L A ∈ A | | ≥ | | all I [n], then 0( ) is nonempty. ⊆ N A S n n ∗ PROOF. For each m-tuple = ( 1,..., m) of subsets of R and for each ν (R ) , we write B B ′ B n ∗ ∈ ′ Inν ( ) := (Inν ( 1),..., Inν ( m)). If = Inν ( ) for some ν (R ) , we say that is a face of and writeB that ′ B ; if in additionB ν isB centered atB the origin, we∈ write that ′ . B B B B B 0 B Claim IX.14.1. If ′ ′, then ′ ′. B B0 A B 0 A n ∗ ′ PROOF. By assumption there is ν (R ) centered at the origin such that = Inν ( ). Pick ′ n ∗ ′ ∈ B A ν (R ) such that = Inν′ ( ). If k is a sufficiently large positive integer, then each of the coordinates ∈ ′ B B n ′ ′ of kν + ν is positive with respect to the basis dual to the standard basis on R , and Inkν+ν′ ( )= , so that ′ ′. A B B 0 A ′ ′ ′ Given , and f = (f1,...,fm) 0( ), define fj,Bj as in section VII.4.3. Let B( ) be B A ′ ∈ L A ∗ n D A′ the set of all f 0( ) such that there is a common root of f1,B1 ,...,fm,Bm on (k ) . Let 0( ) := ′∈ L A ′ 0 ′ D A ′ ( ). Claim IX.14.1 implies that ( ) = ′ ′ ′ ( ), so that claim VII.19 B0A B 0 B0A B B B D A ′ D A ′ D A ′ ′I ′ implies that 0( ) is a Zariski closed subset of 0( ). Let I [n]. Replacing by := ( 1 SI ′ D AI ′I L A S ⊆S ′I A A A′ ∩ R ,..., m R ), it follows that 0( ) is a Zariski closed subset of 0( ). Let π¯0,I : 0( ) ′I A ∩ D A ′ ′ −L1 A ′I L A → 0( ) be the natural projection. Then 0( ) = 0( ) I⊆[n] π¯0,I ( 0( )) is Zariski open in L A′ N A L A \ D A 0( ). It now remains to prove the second assertion of proposition IX.14. If 0 i for some i, then L A ′ S ∈ A ′ any polynomial supported at i with a nonzero constant term would lead to an element in 0( ). On the other hand, if T I I forA every I [n], then claim VII.21 implies that ( ′I ) is aN properA Zariski | A| ≥ | | ⊆ D0 A closed subset of ( ′I ) for every I [n], so that ( ′) is nonempty, as required. L0 A ⊆ N0 A We now explore the relation between non-degeneracy at the origin and the intersection multiplicity at the origin. At first we need to extend the notion of weighted orders and “initial coefficients” corresponding

∗ n n k to branches on (k ) to the case of branches on .

n ∗ Deﬁnition IX.15. Let B := (Z,z) be a branch of a curve C k . Identify Z := Z z with its image on

IB ⊂ \ n k C and let IB := i : xi Z∗ 0 . Note that k is the smallest coordinate subspace of which contains ∗ { | 6≡ } −1 Z . We write νB for the weighted order on k[xi, xi : i IB ] corresponding to the weight ordz(xi Z ) for each i I . Fix a parameter ρ of B and define ∈ | ∈ B B 0 if j I 6∈ B InB(xj ) := xj (74)  νB (xj ) if j IB.  (ρB ) z ∈ ∗ IB In(B) := (In (x ),..., In (x )) (k )  B 1 B n ∈ ∗ n Compare this definition with the case of branches on (k ) defined in (38) in section VI.9.1. The following result is immediate from the definition. ∗ I Lemma IX.16. If the center of B is on (k ) , then I I . ⊂ B 132 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

Lemma IX.17. Let (f1,...,fm) 0( ) k[x1,...,xn] and B be a branch of a curve contained ∈ L A ∩ ∗ IB

in . Then I I k . If in addition is a k V (f1,...,fm) In(B) V (InνB (f1 k B ),..., InνB (fm B )) ( ) B ∈ | | ∗ ∩

branch at the origin, then I I violate condition (ND ) (from deﬁnition IX.5) with ;

k f k ,...,f ν = ν 1| B m| B 0 B in particular, f1,...,fm violate (ND0) (from deﬁnition IX.6) with I = IB.

PROOF. The ﬁrst assertion is a direct corollary of lemma VI.30. If B is a branch at the origin, then νB is centered at the origin, so that the second assertion follows from the ﬁrst one.

Corollary IX.18. If f1,...,fm 0( ) k[x1,...,xn] are -non-degenerate at the origin, then the ∈ L A ∩ n A origin can not be a non-isolated point of V (f ,...,f ) k . 1 m ⊂ n+1 4.1. Proof of theorem IX.8. Below sometimes we work with k with coordinates (x1,...,xn,t). n+1 In those cases we usually denote the coordinates of elements of k as pairs, with the last component of n+1 the pair denoting the t-coordinate. In particular, the origin of k is denoted as (0, 0). Take m = n and deﬁne ( ) as in section IX.3.2. M0 A Claim IX.19. ( ) ( ). M0 A ⊇ N0 A PROOF. Let (f1,...,fn) 0( ) 0( ). We will show that f1,...,fn are -degenerate at the origin. By our assumption there∈ is L(gA,...,g\M )A ( ) such that [g ,...,g ] < [fA,...,f ] . Due to 1 n ∈ L0 A 1 n 0 1 n 0 proposition IV.15 we may assume all gi and fj are polynomials in x1,...,xn. Let t be a new indeterminate. n An application of theorem IV.19 with hj := (1 t)fj + tgj , j =1,...,n, X = k , and (b0,ǫ0)=(0, 1) implies that −

(i) either the origin is a non-isolated zero of f1,...,fn,

n k (ii) or there is an irreducible component V of V (h ,...,h ) in k containing (0, 0) such that 1 n ×

V is different from 0 k. { }× In case (i) f1,...,fn are -degenerate at the origin (corollary IX.18), so consider that we are in case (ii). A n+1 Choose a branch B at the origin of a curve contained in V (h ,...,h ) k which is different from

1 n ⊂ k 0 k. Since B 0 , it follows that I := I [n] = . Let ν be the restriction of ν to { }× 6⊂ { }× B ∩ 6 ∅ B k[x : i I]. Then it follows as in the proof of proposition VII.16 that for each j =1,...,n, i ∈

I I I k k if Inν (fj )=InνB (hj k B ) Supp(fj ) Inν ( j ) = , I I (75) InA ,ν (fj k )= | | | ∩ A 6 ∅ j | (0 otherwise. ∗

Lemma IX.17 implies that I I violate (ND ) with , and therefore identity (75) k h1 k B ,...,hn B 0 ν = νB implies that f ,...,f are -degenerate| at the| origin, as desired. 1 m A Claim IX.20. Assume [ ,..., ] < . Then ( ) ( ). A1 An 0 ∞ M0 A ⊆ N0 A PROOF. If [ 1,..., n]0 = 0, then (f1,...,fm) 0( ) if and only if one of the fj has a nonzero constantA term, whichA immediately implies that ∈f ,...,f M A are -non-degenerate. So assume 1 n A 0 < [ 1,..., n]0 < . Pick (f1,...,fn) 0( ) 0( ). We will show that (f1,...,fn) 0( ) followingA (theA adapted∞ version of) Bernstein’s∈L trickA from\ N sectionA VII.4.2. Due to corollary IX.186∈ Mwe mayA assume that [f ,...,f ] < , and due to proposition IV.15 we may assume in addition that each f is 1 n 0 ∞ j a polynomial in (x1,...,xn). Since f1,...,fn violate (ND0), there is a nonempty subset I of [n] and a

k I I I I k weighted order ν centered at the origin on [xi : i I] such that In (f1 k ),..., In (fn ) have A1 ,ν An,ν ∗ I I∈ I | | a common zero a = (a ,...,a ) (k ) . Let T = j : = . Since [ ,..., ] < , corol- 1 n ∈ A { Aj 6 ∅} A1 An 0 ∞ lary IX.13 and propositions IV.15 and IX.14 imply that there is a system (g1,...,gn) 0( ) of polyno- I ∈ L A I I mials in (x1,...,xn) such that InA ,ν (gj k )(a) =0 for each j T . Deﬁne c(t) := (c1(t),...,cn(t)) : j | 6 ∈ A

I k k as follows: → νi ait if i I, ci(t) := ∈ (0 otherwise. I For each j T , let mj := minAI (ν). Deﬁne ∈ A j

−mj −mj I t gj (c(t))fj t fj(c(t))gj if j TA, hj := − ∈ (fj otherwise. 5. PROOF OF THE BOUND 133

I I I I k Note that each hj is a polynomial in (x1,...,xn,t). Since InA ,ν (gj k )(a) =0=InA ,ν (fj )(a) for j | 6 j | each j T I , it follows as in section VII.4.2.2 that h (x, 0) is a nonzero constant multiple of f for each ∈ A j j

j. By our assumption the origin is an isolated zero of f1,...,fn. Since h1,...,hn vanish on the curve k (c(t),t): t k , theorem IV.19 implies that [f1,...,fn]0 > [h1(x, ǫ),...,hn(x, ǫ)]0 for generic ǫ . {Since Supp(h∈ (x,} ǫ)) for each ǫ, it follows that [f ,...,f ] > [ ,..., ] , as required. ∈ j ⊂ Aj 1 n 0 A1 An 0 Theorem IX.8 now follows from corollary IX.13, proposition IX.14, and claims IX.19 and IX.20.

5. Proof of the bound In this section we prove theorem IX.1. The computation of intersection multiplicity becomes easier if a generic system satisﬁes a property which is stronger than (ND0); at ﬁrst we prove that such systems exist. The proof of theorem IX.1 is then given in section IX.5.1. We start with a notation: if g is a polynomial in n I (x1,...,xn), a = (a1,...,an) k and I [n], we write ga for the polynomial in (xi : i I) obtained ∈ ′ ⊆ ∈ from substituting ai′ for xi′ for each i not in I.

FIGURE 7. Illustration of lemma IX.21: f := x2 + (y 1)2 1=0 and g := z2 + x2 (y + z)2 = 0 intersect along the z-axis and an ellipse− which− intersect at (0, 0, 1). − 2 2 2 2 f z=1 = x 2y + y and g z=1 = x 2y y are degenerate at the origin, since with | − n ∗ | − − 2 ν = (1, 2) (R ) , Inν (f z=1)=Inν (g z=1)= x 2y. ∈ | | −

k I

Lemma IX.21. Let I [n] and f1,...,fk [x1,...,xn] such that fj k =0 for each j =1,...,k. As- ⊆ ∈ | I [n]\I [n]\I sume V (f1,...,fk) has an irreducible component V which is not contained in k . Then f1,a ,...,fk,a ∗ I are degenerate at the origin for each a (k ) V (see ﬁg. 7). ∈ ∩ ∗ I PROOF. Let a (k ) V and B be a branch centered at a of a curve contained in V such that

I ∈ ∩ n I

k k B k . Then IB % I (lemma IX.16) and πI (In(B)) = a, where πI : is the natural projection.

6⊂ ′ ′ ′ ′ → ∗ I k Let I := I I and ν be the restriction of ν to k[x ′ : i I ]. Since the center of B is on ( ) , it B \ B i ∈ follows that νB(xi)=0 for each i I. Using this it is straightforward to check that for each j =1,...,k, [n]\∈I ′ ′ ∗ I′ I ′ I′ , where ′ k . The result now follows

In (f k B )(In(B)) = In (f )(a ) a := π (In(B)) ( ) νB j ν j,a k I from lemma| IX.17. | ∈ Let f ,...,f be polynomials in (x ,...,x ) and I [n]. Deﬁne 1 m 1 n ⊆ I ∗ I [n]\I [n]\I

(76) D (f ,...,f ) := a (k ) : f ,...,f are degenerate at the origin 0 1 m { ∈ 1,a m,a } The following result is immediate from theorem III.20 and lemma IX.21. ′ ′ ′ ′ Corollary IX.22. Let J, J [m] and V be an irreducible component of V (fj′ : j J ) such that ′ I I ⊆ ′ ′ ′ ∈ V k . Assume dim(D0 (fj′ : j J ) V (fj : j J)) < I J . Then V does not contain any 6⊆ ∈ ∩ ∗ I ∈ | | − | | irreducible component of V (f : j J) (k ) . j ∈ ∩ 134 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

n Definition IX.23. Given a collection = ( 1,..., m), m 1, of finite subsets of Z≥0 and I [n], define A A A ≥ ⊆ I I T := j : j R = [m] A { A ∩ 6 ∅} ⊆ ′I := J = j ,...,j [m] T I : J = n I , [π ( ),...,π ( )] < TA { { 1 n−|I|}⊆ \ A | | − | | [n]\I Aj1 n\I Aj[n]−|I| 0 ∞} I I I := ( 1 R ,..., m R ) A A ∩ A ∩ n [n]\I where π[n]\I : R R is defined as in (70). Due to the finiteness of the j we can identify ( ) with the collection→ ( ) of polynomials (f ,...,f ) such that Supp(f ) A for each j. Given L0 A L A 1 m j ⊆ Aj (f1,...,fm) ( ), we say that f1,...,fm are strongly -non-degenerate if for all I [n], ∈ L A I A ⊆

I I k (a) f1 k ,...,fm are properly -non-degenerate (see section VII.4.4), | ′ | ′I [n]\I A′ ′ [n]\I (b) for all J A , fj′,(1,...,1), j J , are non-degenerate at the origin and NP(fj′ ,(1,...,1)) = ∈ T ′ ∈′ conv(π ( ′ )) for each j J , [n]\I Aj ∈ (c) for all J T I and J ′ ′I , ⊆ A ∈ TA I ′ ′ (77) dim(D (f ′ : j J ) V (f : j J)) < I J , 0 j ∈ ∩ j ∈ | | − | | where DI ( ) are as in (76). 0 · Note that property (b) with I = in particular implies that ∅ (d) NP(f ) = conv( ) for each j =1,...,m. j Aj We write ˇ ( ) be the collection of all (f ,...,f ) ( ) which are strongly -non-degenerate. Recall N A 1 m ∈ L A A that 0( ) stands for the collection of systems which are -non-degenerate at the origin. For proposi- tionsNIX.24A and IX.25 below we assume each is ﬁnite. A Aj Proposition IX.24. Assume either 0 m or T I I for all I [n]. Then ˇ ( ) ( ). In ∈ i=1 Ai | A| ≥ | | ⊆ N A ⊆ N0 A particular, if m = n and [ ,..., ] < , then ˇ ( ) ( ). A1 An 0 S ∞ N A ⊆ N0 A PROOF. Follows from proposition VII.22 and corollary IX.13.

Proposition IX.25. ˇ ( ) is constructible and it contains a nonempty Zariski open subset of ( ). N A L A PROOF. By propositions VII.22 and IX.14 the collection of systems that satisfy properties (a) and (b) of deﬁnition IX.23 is a nonempty Zariski open subset of ( ). Therefore we can concentrate only on L A I ′ ′I property (c). For a subset J [m], write J := ( j : j J). Let I [n], J TA and J A . Write I ⊆ A A ∈ ⊆ ⊆ ′ ∈ T′ ( J , J′ ) for the subset of ( J ) ( J′ ) consisting of all ((fj : j J), (fj′ : j J )) which NsatisfyA propertyA (77). Consider theL A set of× maps L A from ﬁg. 8, where ∈ ∈

πJ′ , πJ,J′ , πJ,I , πJ′,I are natural projections, • ′ ′ ∗ I [n]\I ′ σ is the “substitution map” which maps ((f ′ : j J ),a) ( ′ ) (k ) to (f ′ : j • j ∈ ∈ L AJ × j ,a ∈ J ′). 0( ) denotes the collection of systems which are degenerate at the origin, and • D · ∗ I := ((f : j J),a) ( ) (k ) : f (a)=0 for each j J • VJ,I { j ∈ ∈ L AJ × j ∈ }

∗ I ( ) ( ′ ) (k ) L AJ × L AJ × πJ′,I πJ,I πJ,J′

I ∗ I ∗ I k ( , ′ ) ( ) ( ′ ) ( ) (k ) ( ′ ) ( ) N AJ AJ L AJ × L AJ VJ,I L AJ × L AJ × πJ′ σ=substitution

( ′ ) (π ( ′ )) (π ( ′ )) L AJ D0 [n]\I AJ L [n]\I AJ

FIGURE 8. Maps from the proof of proposition IX.25 5. PROOF OF THE BOUND 135

∗ I ′ ′ Let be the subset of ( J ) ( J′ ) (k ) consisting of all ((fj : j J), (fj′ : j J ),a) Z L A × L A[n]\I× ′ ′ ∈ ∈ such that a V (fj : j J) and (fj′,a : j J ) are π[n]\I ( J′ )-degenerate at the origin. Then −1 ∈ ∈ −1 ∈ A = π ( ) (σ π ′ ) ( (π ( ′ ))). Since both and (π ( ′ )) are Zariski Z J,I VJ,I ∩ ◦ J ,I D0 [n]\I AJ VJ,I D0 [n]\I AJ closed (the closedness of 0( ) follows from proposition IX.14), it follows that is also Zariski closed. I ′ D · Z Since (J, J ) is the set of all elementsin ( ) ( ′ ) whose preimage under π ′ has dimension N L AJ ×L AJ J,J |Z less than I J , corollary III.27 implies that I (J, J ′) is constructible. We now show that it contains a | | − | | N ∗ I nonempty Zariski open subset of ( ) ( ′ ). Fix any a (k ) , and let σ be the composition L AJ × L AJ 0 ∈ 0 σ (( ′ ) a (π ( ′ ) ֒ ( ′ ) L AJ → L AJ ×{ 0} −→L [n]\I AJ ′ ′ ′ ′ ′I where the ﬁrst map simply takes (fj′ : j J ) ((fj′ : j; J ),a0). Since J A , proposition IX.14 ′ −1 ∈ 7→ ∈ ∈ T implies that := σ0 ( 0(π[n]\I ( J′ )) is a nonempty Zariski open subset of ( J′ ). Pick an arbitrary Y ′ ′ N ′ A′ L A system (f ′ : j J ) , and let σ be the composition j ∈ ∈ Y

∗ I ′ ′ ∗ I σ k (( ′ ) k ) ֒ (f ′ : j J ) ( ) (π) →{ j ∈ }× −→L [n]\I AJ ′ ′ ′ ′ ′ where the ﬁrst map simply takes a ((fj′ : j J ),a). Since σ (a0) = σ0(fj′ : j J ) 7→ I ∈ ′ ′ ′ −1 ∈ ∈ 0(π[n]\I ( J′ )), proposition IX.14 implies that D0(fj′ : j J ) = (σ ) ( 0(π[n]\I ( J′ )) is a proper N A ∗ I I ′∈ ′ D AI Zariski closed subset of (k ) ; in particular, dim(D0 (fj′ : j J )) < I . Since J TA, lemma IX.26 ∈ | | ⊂ ′ below implies that there is a nonempty open subset of ( J ) such that ((fj : j J), (fj′ : j ′ I ′ W L ′A ′ ∈ ∈′ J )) (J, J ) for each (fj : j J) . Since (fj′ : j J ) was an arbitrary element from Y , ∈ N I ∈′ ∈ W ∈ exercise III.46 implies that (J, J ) contains a nonempty Zariski open subset of ( J ) ( J′ ), as required. N L A × L A

∗ n Lemma IX.26. Let W be an irreducible subvariety (k ) and = ( 1,..., k) be a collection of ﬁnite n B B B nonempty subsets of Z . Let := (f1,...,fk) ( ) : dim(V (f1,...,fk) W ) dim(W ) k . Then is a constructible subsetW of {( ) and it contains∈ L B a nonempty Zariski open∩ subset≤ of ( ). − } W L B L B ′ PROOF. Let := ((f1,...,fk), (x1,...,xn)) ( ) W : fj(x1,...,xn)=0 for each j ′ W { ∈ L B × −1 } and πW : W be the natural projection. For each w W , πW (w) is a linear subspace of ( ) W → ∈ −1 L B deﬁned by k linearly independent linear equations, so that dim(πW (w)) = dim( ( ) k = j j k. ′ L B − |B |− Theorem III.23 then implies that dim( ) = j j k + dim(W ) = dim( ( )) + (dim(W ) k). W |B |− L B ′ P − Now the result follows from applying theorem III.23 and corollary III.27 to πB W′ : ( ), where ∗ n P | W → L B π : ( ) (k ) ( ) is the natural projection. B L B × → L B

5.1. Proof of theorem IX.1. Corollary IX.13 implies that theorem IX.1 holds when [ 1,..., n]0 = ′ A A 0 or . So assume 0 < [ 1,..., n]0 < . Let := ( 1 Γ1,..., n Γn). Pick strongly ′-non-degenerate∞ (f ,...,fA ) A( ′). Theorem∞ IX.8A and propositionA ∩ IX.24Aimply∩ that A 1 n ∈ L0 A [f ,...,f ] = [ ,..., ] 1 n 0 A1 An 0 Therefore it sufﬁces to show that

I I I ∗ (78) [f1,...,fn]0 = [Γ , Γ ,..., Γ ] [π (Γ ′ ),...,π (Γ ′ )]0 1 j2 j|I| 0 [n]\I j1 [n]\I jn−|I| T × I∈XA,1 where for each I T , j = 1, j ,...,j are elements of T I , and j′ ,...,j′ are elements of ∈ A,1 1 2 |I| A 1 n−|I| [n] T I . We proceed by induction on n. It is true for n =1 (see remark IX.2), so assume it is true for all \ A dimensions smaller than n. Since 0 < [f1,...,fn]0 < , proposition IV.16 implies that on a sufficiently n∞ small Zariski open neighborhood U of the origin in k , the subscheme of U defined by f2,...,fn is a I I curve C. For each I [n], let Cj j be the set of irreducible components of C such that k is the smallest ⊆ n { } I coordinate subspace of k containing each Cj . Claim IX.27. Let I [n], T be as in theorem IX.1 and ′I be as in definition IX.23. ⊆ A,1 TA (1) If CI is nonempty, then I T . { j }j ∈ A,1 (2) If I T , then [n] T I ′I . ∈ A,1 \ A ∈ TA 136 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

PROOF. For the ﬁrst assertion, pick I [n] such that CI is nonempty. Since 0 < [ ,..., ] < ⊆ { j }j A1 An 0 , corollary IX.13 implies that T I I . On the other hand, if T I 1 I , then the proper non- ∞ | A| ≥ | | | A \{ }| ≥ | |

I I k degeneracy of f1 k ,...,fn (property (a) of strong -non-degeneracy) and lemma IX.17 implies that I | | AI I CJ j is empty, which is a contradiction. Accordingly TA 1 = I 1 and TA = I , which imply that { T} | T\{ }| | |−′ | | I | | ′ I A,1, as required. For the second assertion, pick I A,1 and set J := [n] TA. Since J = n I , ∈ ∈ ′ \′ | | −| |′ we have to show that [π[n]\I ( j′ ),...,πn\I ( j′ )]0 < , where j ,...,j are elements of J . A 1 A [n]−|I| ∞ 1 n−|I| I′ ′ ′ Indeed, otherwise corollary IX.13 would imply that TA < I for some I ) I, which would in turn imply (since by assumption 0 ) that [ ,...,| ]| = | |, which is a contradiction. 6∈ j Aj A1 An 0 ∞ T I ′ ′ Pick I A,1. Let j1 = 1S, j2,...,j|I| be the elements of TA and j1,...,jn−|I| be the elements of ′ ∈I ′ ′I J := [n] TA. Since J A (claim IX.27), property (b) of strong -non-degeneracy and lemma IX.21 \ I ∈ T A imply that k is an irreducible component of V (f ′ ,...,f ′ ). On the other hand, applying prop- j1 jn−|I| erty (c) of strong -non-degeneracy with J = j2,...,j|I| , and then using corollary IX.22 shows that A { } I no irreducible component of V (f ′ ,...,f ′ ) other than k contains any irreducible component of j1 jn−|I| ∗ I V (fj2 ,...,fj|I| ) (k ) . Therefore claim IX.27, theorem IV.12, and propositions IV.16 and IV.21 imply that ∩ [n]\I [n]\I

I ∗ I ∗ I I k (79) [f1,...,fn]0 = [fj′ ,ǫ ,...,fj′ ,ǫ]0 ord0(f1 C )[fj2 (k ) ,...,fj|I| ( ) ]C 1 n−|I| | j | | j T j I∈XA,1 X ∗ n [n]\I ′ ′I where ǫ is a generic element of (k ) and f·,ǫ are as in lemma IX.21. Since J A , property (b) of strong -non-degeneracy and theorem IX.8 imply that ∈ T A [n]\I [n]\I [f ′ ,...,f ′ ] = [π ( ′ ),...,π ( ′ )] = [π (Γ ′ ),...,π (Γ ′ )] j ,ǫ j ,ǫ 0 [n]\I j1 [n]\I j 0 [n]\I j1 [n]\I j 0 1 n−|I| A A n−|I| n−|I| I It remains to compute the inner sum of the right hand side of (79). Let I [n]. Write R := k[xi : i I]. I I ⊆ ′I ∈ Let 0 be the set of weighted orders on R which are centered at the origin and 0 be the set of primitive V I ′I I V I elements in 0 . For each ν 0 , let 0,j,ν be the set of all branches at the origin of Cj such that νB is proportionalV to ν. Theorem IV.12∈ V impliesB that for each I, j,

ord0(f1 CI )= ordz(f1 CI ) | j | j ν∈V′I (Z,z)∈BI X0 X0,j,ν Therefore it sufﬁces to show that (80) ′ I I

I ∗ I ∗ I I k ordz(f1 C )[fj2 (k ) ,...,fj|I| ( ) ]C = min(ν) MVν (Inν (Γj2 ),..., Inν (Γj )) | j | | j ΓI |I| j (Z,z)∈BI 1 X X0,j,ν

∗ I ∗ I ∗ I

k k To see it, apply corollary VII.35 (with n = I ) to f1 (k ) ,fj2 ( ) ,...,fjI ( ) . Property (a) of strong -non-degeneracyimplies that all the assumptions| | of| proposition| VII.23 and corollary| VII.35 are satisﬁed. PartA 2 of proposition VII.23 implies that each irreducible component of the resulting curve C′ comes ∗ I ′

∗ I ∗ I k k from an irreducible component of V (fj2 (k ) ,...,fjI ( ) ) ( ) and therefore the collections j,ν | I | ⊂ B from corollary VII.35 are precisely the collections 0,j,ν. Corollary VII.35 then implies identity (80) and completes the proof of theorem IX.1. B

6. The efficient version of the non-degeneracy condition In this section we prove theorem IX.9. Given I [n], we write I for the set of weighted orders ˜ ⊆I I˜ V on k[x : i I]. Given I I, we say that ν and ν˜ are compatible if (ν(x ): i I) i ∈ ⊆ ∈ V ∈ V i ∈ and (˜ν(x ): i I) are proportional, with a positive constant of proportionality, and ν˜(x˜) > 0 for each i ∈ i ˜i I˜ I. Theorem IX.9 follows directly from lemma IX.28 below. ∈ \ n Lemma IX.28. Let := ( 1,..., m) be a collection of (possibly infinite) subsets of Z , I be a A A A ≥0 nonempty subset of [n], and ν I be such that ∈V n ∗ (81) Inν˜( j ) is finite for each ν˜ (R ) which is compatible with ν. A ∈

Let f ,...,f k[[x ,...,x ]] such that Supp(f ) for each j. Assume 1 m ∈ 1 n j ⊆ Aj 6.THEEFFICIENTVERSIONOFTHENON-DEGENERACYCONDITION 137

I I I (1) A

I I k (2) InA ,ν (fj k ), j =1,...,m, have a common zero u ( ) . j | ∈ ˜ Then there exists I˜ % I and ν˜ I such that ∈V (3) ν˜ is compatible with ν. ∗ n (4) In I˜ (fj I˜), j = 1,...,m, have a common zero u˜ (k ) such that πI (˜u) = πI (u), where A ,ν˜ k j | ∈

∗ n ∗ I k π : (k ) ( ) is defined as in (70). I → PROOF. Dueto(81) we may assume without any loss of generality that the support of each fj is finite, i.e. the fj are polynomials in (x1,...,xn). We may also assume that I = 1,...,k , 1 k n. Let ∗ I { } ≤ ≤ a := πI (u) (k ) and (a1,...,an) be the coordinates of a. At first consider the case that ν(xi)=0 for ∈ ∗ I each i I. Assumption (2) then says that a is a common zero of f1,...,fm on (k ) . Let yj := xj aj , ∈ n − j = 1,...,n, so that a is the origin of k with respect to (y1,...,yn) coordinates. Choose any integral

′ n ∗ n n k ν (R ) with positive coordinates with respect to the dual basis, and let π : Blν′ (k ) be the ′ ∈ n → ν -weighted blow up of k with respect to (y1,...,yn) coordinates (see section VI.11). Let Eν′ be the

′ ∗ I n I k exceptional divisor of π, and W be the strict transform of (k ) on Blν′ ( ). Since A

I ∗ n ∗ I |E | − | | k is an irreducible component V of V (fj : j A) (k ) properly containing ( ) . Then the strict trans- ′ ∈E ∩′ ′ ′ form V of (the closure of) V properly contains W . Pick a Eν′ W , and choose an irreducible curve ′ ′ ′ ′ ′ ′ ∈ ′ ′ ∩ ′ ′ C V such that a C W Eν′ , and a branch B = (Z ,z ) of C centered at a . Let I˜ := IB′ and

⊂ I˜ ∈ 6⊆ ∪ ∗ I ′ I k ν˜ := νB′ (deﬁnition IX.15). Since π is centered at a (k ) and since π(B ) , it follows that ˜ ∈ V ∈ ˜ 6⊂˜ I $ I, InB′ (xi)= ai and ν˜(xi)=0 for each i I, and ν˜(x˜i) is positive for each i I I. Fix j [m]. I ∈ ∈ \ ∈

˜ ˜ ˜ I I I k If j , it follows that min I (˜ν)=0 and In I (fj I ) = fj k = In (fj ). This implies that A A A ,ν˜ k A ,ν 6∈ E j j | | j | ′ I ′ ˜ ˜ I ˜ ˜

In I (f I )(In(B )) = f k (a)=0. On the other hand, if j , then In I (f I )(In(B )) = 0 k A ,ν˜ j k j A A ,ν˜ j j | | ∈ E j | due to lemma IX.17. The lemma is therefore true in the case that ν is the trivial weighted order.

Now assume ν is not the trivial weighted order. Identify ν with the element in (Rn)∗ with coordinates n (ν(x1),...,ν(xk)) with respect to the basis dual to the standard basis of R . Choose a basis α1,...,αk k α1 αk of Z such that ν, αj =0 for j =1,...,k 1, and ν, αk =1. Then (x ,...,x , xk+1,...,xn) are

h i ∗ k n−k − h i k coordinates on X := (k ) . Define × xαj aαj if 1 j k 1, α − ≤ ≤ − yj := x k if j = k, xj if k +1 j n. ≤ ≤ n ′ ′ Write Y for the affine space k with coordinates (y1,...,yn). Choose positive integers ν1,...,νn such that ν′ =1 and ν′ 1 for j = k +1,...,n. Let ν′ be the element in (Rn)∗ with coordinates (ν′ ,...,ν′ ) k j ≫ 1 n with respect to the basis dual to the standard basis, and π : Y ′ Y be the ν′-weighted blow up of Y with → ′ respect to (y1,...,yn) coordinates, E be the exceptional divisor of π, and W be the strict transform on ′ ′ Y of W := V (yk+1,...,yn) Y . Proposition VI.35 implies that there is an affine open subset U of Y such that ⊂

∗ k−1 n−k

k k (i) U = k ( ) with respect to coordinates (z1,...,zn) where z1,...,zk are mono- ∼ × × ′ νj ′ ′ mials in (y1,...,yk), zj = yj/z1 for j = k + 1,...,n, ν (z1)=1 and ν (zj)=0 for j =2,...,n,

∗ k−1 n−k k (ii) U E = V (z1) = (k ) , and

∩ ′ ∼ × ∗ k−1 k (iii) U W = V (z ,...,z ) = k ( ) . ∩ k+1 n ∼ × We treat X as an open subset of Y via the natural embedding. There is an irreducible component V of I ¯ ′ ¯ V (fj : j A) X such that its closure V in Y properly contains W . The strict transform V of V on Y ′ properly∈ E contains∩ W ′. Pick a′ U W ′ E. Choose an irreducible curve C′ V ′ such that a′ C′, and C′ E W ′, and C′ ∈ π−∩1(X) ∩= . Pick a branch B′ = (Z′,z′) of⊂C′ centered at a′.∈ Since π(B′) 6⊂X =∪ , we may treat∩ B′ as a branch6 ∅ (possibly at infinity) of a curve on X. Define ∩ 6 ∅ νB′ and IB′ as in definition IX.15. Since each of x1,...,xk is everywhere nonzero on X, it follows that ′ IB′ 1,...,k = I. On the other hand, since π(B ) W , it follows that there exists j > k such ⊃ { } 6⊂ ˜ that xj B′ 0. It follows that IB′ ) I. We show that properties (3) and (4) are true with I := IB′ and | 6≡ 138 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

′ ν˜ := νB′ . Indeed, since a E, for each j =1,...,n, either yj B′ 0, or ordz′ (yj B′ ) > 0. Therefore, for each j =1,...,k 1, ∈ | ≡ | α − α (iv) ν ′ (x j ) = ord ′ ((a j + y ) ′ )=0, since ord ′ (y ′ ) > 0. B z j |B z j |B αk Since ν ′ (x ) = ord ′ (y ′ ) > 0, it follows that ν ′ and ν are proportional on k[x : i I] with B z k|B B i ∈ a positive constant of proportionality. Pick j I ′ I. Then j > k, and ν ′ (x ) = ord ′ (y ′ ) > ∈ B \ B j z j |B 0. It follows that νB′ and ν are compatible. It remains to exhibit property (4). Since the center of B is on U E, properties (i) and (ii) of U imply that (ordz′ (y1 B′ ),..., ordz′ (yk B′ )) is proportional to ′ ∩′ ′ | | (ν1,...,νk). Since ν (z1) =1, it follows that the constant of proportionality is q := ordz′ (z1 B′ ). αk ′ αk | Therefore νB′ (x ) = ordz′ (yk B′ ) = νkq = q = qν(x ). Since νB′ is compatible with ν, it follows that |

(v) νB′ (xj )= qν(xj ) for j =1,...,k. ′ ′ On the other hand, since a U W E, properties (ii) and (iii) imply that ordz′ (zj B′ ) 1 if j > k. It follows that ∈ ∩ ∩ | ≥ ′ ′ (vi) for each j I ′ I, ν ′ (x ) = ord ′ (z ′ )+ ν ord ′ (z ′ ) > qν . ∈ B \ B j z j |B j z 1|B j Let u′ := In(B′) (k∗)IB′ . Observation (iv) implies that u′αj = aαj for j = 1,...,k 1. Proposi- ∈ ∗ ν(x1) ′ ν(xk) −′ tion VI.1 then implies that there is t k such that (a1,...,ak) = (t u1,...,t uk). Choose a ′ ∈ q-th root t of t in k and let u˜ = (˜u1,..., u˜n) be an element with coordinates

′νB′ (xj ) ′ t uj if j IB′ , u˜j := ∗ ∈ (arbitrary element in k otherwise. tν(xj )u′ = a if j I, j j ∈ ′νB′ (xj ) ′ = t uj if j IB′ I, ∗ ∈ \ arbitrary element in k otherwise. I Note that πI (˜u) = a = πI (u). Fix j [m]. If j A, then (v) and (vi) imply that choosing ′ ′ ∈ 6∈ E I I I I ν ,...,ν sufﬁciently large we can ensure that In ′ (fj ′ ) = In (fj k ), which would k+1 n B k B A ,ν Aj ,νB′ | j | I I I I I imply that In ′ (fj ′ )(˜u)=In (fj k )(a)=0. On the other hand, if j , then B k B Aj ,ν A Aj ,νB′ | | ∈E

′ min I (ν ′ ) B′ B Aj ′

In I ′ (fj I ′ )(˜u)= t In I ′ (fj I ′ )(In(B ))=0 k B k B B B Aj ,νB′ | Aj ,νB′ | due to lemma IX.17. This completes the proof of property (4).

7. Other formulae for generic intersection multiplicity 7.1. The formula of Huber-Sturmfels and Rojas. Let t be a new indeterminate. Fix positive integers

k ,...,k . Note that for each f ,...,f k[[x ,...,x ]], 1 n 1 n ∈ 1 n k1 kn [f1,...,fn]0 = [t,f1 + c1t ,...,fn + cnt ]0 n for any c1,...,cn k. It follows that, for each collection of subsets 1,..., n of Z , ∈ A A ≥0 [ ,..., ] = [ ˆ ,..., ˆ ] A1 An 0 A0 An 0 ˆ n+1 ˆ n+1 where 0 := (1, 0,..., 0) Z and j := (kj , 0,..., 0) ( 0 j ) Z for j =1,...,n. A { }⊂ ≥0 A { } ∪ { } × A ⊂ ≥0 ˆ ˆ ˆ T Let := ( 0,..., n). It follows from (72) that Aˆ,1 = [n + 1] and therefore, if [ 1,..., n]0 < , thenA theoremA IX.1 impliesA that { } A A ∞ (82) [ ,..., ] = [Γˆ ,..., Γˆ ]∗ = νˆ MV′ (In (Γˆ ),..., In (Γˆ )) A1 An 0 0 n 0 0 νˆ νˆ 1 νˆ n νˆ∈Vˆ′ X0 ˆ ˆ where Γj are the Newton diagramsof j , and νˆ ranges over the primitive weighted orders on k[t, x1,...,xn] A ′ ˆ ˆ which are centered at the origin, and νˆ0 :=ν ˆ(t). Note that MVνˆ(Inνˆ(Γ1),..., Inνˆ(Γn)) is positive only ′ if νˆ is the inner normal to a “lower” facet of Γˆ1 + + Γˆn (the designation “lower” comes from the fact that νˆ′ points “upward” along the t-coordinate). B.··· Huber and B. Sturmfels presented in [HS97] the idea of “lifting” subsets of Zn to one extra dimension and summing the mixed volumes of faces corresponding 7.OTHERFORMULAEFORGENERICINTERSECTIONMULTIPLICITY 139 to certain lower facets of the sum of the lifted bodies. J. M. Rojas [Roj99] observed that the expression in the right hand side of (82) gives the generic intersection multiplicity at the origin. Note that unlike the formula (73) from theorem IX.1, the expression in (82) is symmetric in 1,..., n (provided the kj are chosen to be equal). A A

A1 A2

FIGURE 9. 1 is convenient, whereas 2 is not. The subdiagram volume of j is the area of the regionA shaded in green. A A

n 7.2. Convenient Newton diagrams. We say that a subset of R≥0 is convenient if it contains a point − n on each coordinate axis. The subdiagram volume Vn ( ) of a subset of R≥0 is the n-dimensional volume of the “cone” whose base is the Newton diagramA of and apexA is at the origin; in other words, V −( ) is the n-dimensional volume of the union of all line segmentsA from the origin to ND( ) (ﬁg. 9). n A A n Proposition IX.29. Let 1,..., n be subsets of Z . Let Γj := ND( j ), j =1,...,n. A A ≥0 A (1) If Γ2,..., Γn are convenient, then ′ (83) [ 1,..., n]0 = min(ν) MVν (Inν (Γ2),..., Inν (Γn)) A A Γ1 ν∈V′ X0 (2) (Kushnirenko [AY83, Theorem 22.8]) If Γ is a convenient Newton diagram, and if Γj = Γ for each j, then (84) [ ,..., ] = n!V −(Γ) A1 An 0 n (3) (Ajzenberg and Yuzhakov [AY83, Theorem 22.10]) If Γ1,..., Γn are convenient, then (85) [ ,..., ] = ( 1)n−|I|V −( Γ ) A1 An 0 − n i I⊆[n] i∈I XI6=∅ X

PROOF. If Γ2,..., Γn are convenient, then TA,1 = [n] ,and(83) follows from (73). Now we prove { } n assertion (2). Let j j be the facets of Γ2 + +Γn with inner normals in Z . Then(83) implies that {Q } ··· >0 ′ ′ [ 1,..., n]0 = min(νj ) MVν (Inνj (Γ),..., Inνj (Γ)) = (n 1)! min(νj ) Volν (Inνj (Γ)) A A Γ j − Γ j j j X X where ν′ are the inner normals to and Vol′ are as in corollary V.43. Now ﬁx j, and let := j Qj νj Rj conv( j 0 ). Then j is a facet of j with outer primitive normal νj , and all other facets of j passesQ through∪{ } the origin.Q Since max R(ν ) = min (ν ), corollary V.43 implies that Vol ( )R = Rj j Γ j n Rj (1/n) Vol′ (In (Γ)). Since V −(Γ) = Vol ( ), identity (84) follows. Since [ ,..., ] is νj νj n j n Rj A1 An 0 multi-additive and symmetric in the j , assertion (3) then follows from corollary B.48. A P The following is a more precise version of assertion (2) of proposition IX.29.

n Proposition IX.30. Let 1,..., n be subsets of Z . Let Γj := ND( j ), j = 1,...,n, and Γ := A A ≥0 A ND( n ). For each I [n], let T I , where := ( ,..., ), be as in theorem IX.1. Then j=1 Aj ⊆ A A A1 An − S(1) (Kushnirenko [AY83, Theorem 22.8]) [ 1,..., n]0 n!Vn (Γ). − A A ≥ I (2) [ 1,..., n]0 = n!Vn (Γ) if and only if for each nonempty I [n], TA I and for each A A ⊆ | | ≥ | | I weighted order ν centered at the origin on k[xi : i I], the collection Inν (Γj R ): j I I ∈ { ∩ ∈ T , Γj Inν (Γ R ) = of polytopes is dependent. A ∩ ∩ 6 ∅} 140 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

PROOF. If [ ,..., ] = then both assertions of the proposition are satisfied (for the second A1 An 0 ∞ assertion one needs to use corollary IX.13). So assume [ 1,..., n]0 < . Then Γ is convenient, so that assertion (1) follows from assertion (2) of proposition IX.29A and theA definition∞ of generic intersection multiplicity. Regarding the second assertion, theorem IX.8 implies that [ 1,..., n]0 = [Γ,..., Γ]0 if and only A A n n if generic (f1,...,fn) 0( ) are -non-degenerateat the origin, where := ( j=1 j ,..., j=1 j ). The second assertion now∈ L followsA fromB theorem VII.30. B A A S S 7.3. Making convenient without changing [ ,..., ] . If [f ,...,f ] < , then the ideal Aj A1 An 0 1 n 0 ∞ generated by f1,...,fn in k[[x1,...,xn]] contains all sufficiently large powers of the maximal ideal of di,j k[[x1,...,xn]]. It follows that if we replace fj by fj + j ci,j xj , then [f1,...,fn]0 does not change for sufficiently large di,j . Since the Newton diagrams of the fj become convenient after these replacements, it nP follows that given any set of subsets 1,..., n of Z such that [ 1,..., n]0 < , we may use (83) A A ≥0 A A ∞ or (85) to compute [ 1,..., n]0 after adding to each Γj appropriate vertices on the coordinate axes. In this section we deriveA a “sharp”A explicit bound on the placement of these vertices which guarantees that the intersection multiplicity at the origin remains unchanged. A. Khovanskii told the author in 2017 that he also had obtained, but never published, such a bound.

m2(S,ν) H(S,ν) D ν A B m1(S,ν)

FIGURE 10. m ( ,ν) for = A,B,C,D i S S { }

Let be a compact subset of Rn and ν be an element of (Rn)∗ centered at the origin. Let H( ,ν) := nS S α R : ν, α = minS (ν) be the hyperplane perpendicular to ν which contains the “face” Inν ( ) of { ∈ h i } S corresponding to ν. We write mi( ,ν) for the i-th coordinate of the point of the intersection of H( ,ν) andS the i-th coordinate axis (see ﬁg.S10). Note that S

minS (ν) mi( ,ν)= S νi n where ν = (ν1,...,νn) with respect to the coordinate dual to the standard basis of R . Given a collection n T T = ( 1,..., n) of subsets of Z≥0, pick I A,1, where A,1 is as in (72). Let j1 = 1, j2,...,j|I| A A A I I ∈ I I ′I be the elements of T . For each j, let Γ be the Newton diagram of := j R . Let ( ) be the A j Aj A ∩ V0,1 A set of primitive weighted orders centered at the origin on k[xi : i I] such that the ( I 1)-dimensional mixed volume of In (ΓI ),..., In (ΓI ) is nonzero; recall that faces∈ with nonzero mixed| |− volume can be ν j2 ν j|I| n detected combinatorially (theorem VII.30). Let e1,...,en be the standard unit vectors in R . Deﬁne (86) I I ki if I = i , Γ1 = kiei , and fj k =0 for each j > 1 (case 1) I { ′}I { } | mi,1( ) := 1 else if 0,1( )= (case 2) A  I V A ∅ max ′I mi(Γ ,ν) otherwise (case 3)  ν∈V0,1(A) 1

 I See fig. 11 for an illustration of different cases in the definition of mi,1. Define I (87) mi,1( ) := max mi,1 A I∈TA,1 i∈I T [n] Since [n] A,1, it follows that mi,1( ), and therefore mi,1( ) is a well defined nonnegative rational ∈ ′ A′ ′ A ′ number for each i. Let 1 := 1 m1e1,...,mnen , where mi are arbitrary integers greater than or equal to m ( ). Note thatA ′ Ais convenient.∪{ Let ′ :=} ( ′ , ,..., ). i,1 A A1 A A1 A2 An 8. MONOTONICITY OF GENERIC INTERSECTION MULTIPLICITY 141

{2} {2} m2,1 (A) m2,1 (A) A1 A2 A1 A2 (Case 1) (Case 1) {1,2} m2,1 (A) (Case 2) {1,2} m1,1 (A) (Case 3)

(A) Cases 1 and 3 (B) Cases 1 and 2

FIGURE 11. Different cases of (86)

′ Proposition IX.31. Assume [ 1,..., n]0 < . Then [ 1, 2,..., n]0 = [ 1,..., n]0. If in addi- A A ∞ A′ A A A A ′′ tion [ 1,..., n]0 > 0, then the transformation 1 1 is sharp in the following sense: if 1 1 ′′ A A A′′ 7→ A ′′ ′′ A ⊃ A ∪ mi ei for any i and any nonnegative integer mi such that mi < mi,1( ), then [ 1 , 2,..., n]0 < {[ ,...,} ] . A A A A A1 An 0 PROOF. If [ ,..., ] =0, then 0 ′ , so that [ ′ , ,..., ] = 0 as A1 An 0 ∈ j Aj ⊂ A1 ∪ j≥2 Aj A1 A2 An 0 well. Now assume 0 < [ ,..., ] < . Then 0 , so that m (ΓI ,ν) > 0 for each i [n], I [n] A1 An 0 ∞ S 6∈ A1 S i 1 ∈ ⊆ and each nontrivial weighted order centered at the origin on k[xj : j I]. It follows that mi,1( ) > 0 for ′ ∈ T TA each i, and therefore corollary IX.13 implies that 0 < [ 1, 2,..., n]0 < and A,1 = A′,1. Now T A AI ′IA ∞ ′ I pick I A′,1. Let j1 = j, j2,...,j|I| be the elements of T ′ . If Γ is the Newton diagram of R , it ∈ A 1 A1 ∩ is straightforward to see using the deﬁnition of [ ,..., ]∗ that [Γ′I , ΓI ,..., ΓI ]∗ = [ΓI , ΓI ,..., ΓI ]∗, 0 1 j2 j|I| 0 1 j2 j|I| 0 ′ · · and therefore (73) implies that [ 1, 2,..., n]0 = [ 1,..., n]0. On the other hand, if for some ′I A A A A A ′′ ′′ I i I, Γ1 contains an element on the i-th axis with coordinates mi ei such that mi < mi,1(Γ1,ν) ∈ ′I ′I I I ∗ for some ν 0,1( ), then minΓ′I (ν) < minΓI (ν), and it would follow that [Γ1 , Γ ,..., Γ ]0 < ∈ V A 1 1 j2 j|I| [ΓI , ΓI ,..., ΓI ]∗, which implies the last assertion. 1 j2 j|I| 0

It is clear that given 1,..., n such that [ 1,..., n]0 < , repeating the above process n A ′ A ′ A A ∞ n ′ ′ times would yield a collection ,..., of convenient subsets of Z such that [ ,..., ]0 = A1 An ≥0 A1 An [ 1,..., n]0, as required. However, as ﬁg. 12 shows, the outcome of the process is in general not unique: differentA orderingA of the might result in different ′ ,..., ′ . Aj A1 An

Γ1 ′ Γ2 Γ2

′ ′ Γ2 Γ1 Γ2 ′ Γ1 Γ1

′ ′ (A) Γ1 =Γ2 = the line segment from (2, 0) to (0, 4) (B) Reversing the order of the Aj from ﬁg. 12a ′ ′ changes Γ1 (respectively Γ2) to the line segment from (2, 0) to (0, 2) (respectively, from (4, 0) to (0, 4)).

′ ′ ′ FIGURE 12. Dependenceof 1, 2 on the ordering of j : here Γj , Γi denote respectively the Newton diagram of{A , A′ .} A Aj Ai

8. Monotonicity of generic intersection multiplicity

n In remark IX.3 we saw that [ ,..., ]0 is “monotonic” as a function on n-tuples of subsets of Z≥0. In this section we characterize in theorem· IX.32· the conditions under which it is “strictly monotonic,” and as an application we prove the alternate formulation of non-degeneracy at the origin in corollary IX.11 (the counterparts of these results in the toric case are corollary VII.31 and theorem VII.11). We also state a curious implication (proposition IX.34) of theorem IX.8 that in the case the monotonicity is not strict, the intersection multiplicity is determined by the Newton diagram of the intersection. 142 IX. INTERSECTION MULTIPLICITY AT THE ORIGIN

n THEOREM IX.32. Let j conv( j )+ R , j = 1,...,n. Then [ 1,..., n]0 [ 1,..., n]0. B ⊆ A ≥0 A A ≤ B B If [ ,..., ] < , then the following are equivalent: A1 An 0 ∞ (1) [ ,..., ] = [ ,..., ] , A1 An 0 B1 Bn 0 (2) for each nonempty subset I of [n], and each ν (Rn)∗ which is centered at the origin, the I I ∈ collection Inν (ND( j )) : Inν (ND( j )) Supp( j ) = of polytopes is dependent, { B n ∗A ∩ B 6 ∅} I (3) foreach I IA and each ν (R ) which is centered at the origin, the collection Inν (ND( )) : ∈ ∈ { Bj In (ND( I )) Supp( ) = of polytopes is dependent. ν Aj ∩ Bj 6 ∅} PROOF. This follows exactly in the same way as the proof of corollary VII.31 by considering generic (f1,...,fn) 0( ), then observing that [f1,...,fn]0 = [ 1,..., n]0 if and only if f1,...,fn are - non-degenerate∈ L at theB origin, and ﬁnally applying theorems VII.5A andAVII.30 together with the genericnessA of f1,...,fn. Corollary IX.33. Corollary IX.11 holds.

PROOF. The equivalence of assertions (1) and (2) follows exactly as in the proof of corollary VII.33. Theorem IX.9 implies that assertions (2)and (3) are equivalent.

α In proposition IX.34 below we use the following notation: given f = α cαx k[[x1,...,xn]] and n α ∈ R , we write fS := cαx . S ⊆ α∈S P Proposition IX.34. Let (Pf1,...,fn) 0( ) be such that [f1,...,fn]0 = [ 1,..., n]0. Then ∈ L A A A [f ,...,f ] = [f ,...,f ] = [ ,..., ] 1,ND(A1) n,ND(An) 0 1 n 0 A1 An 0 n In particular, if j are subsets of j + R such that [ 1,..., n]0 = [ 1,..., n]0, then B A ≥0 B B A A [ ND( ),..., ND( )] = [ ,..., ] = [ ,..., ] B1 ∩ A1 Bn ∩ An 0 B1 Bn 0 A1 An 0 PROOF. Follows immediately from theorem IX.8, since f1,...,fn are non-degenerate at the origin if and only if f1,ND(A1),...,fn,ND(An) are non-degenerate at the origin. n The requirement that j j + R for each j is necessary for proposition IX.34. Indeed, if j , j, B ⊆ A ≥0 A B j =1, 2, are from ﬁg. 13, then [ , ] = [ , ] =9, but [ ND( ), ND( )] = . A1 A2 0 B1 B2 0 B1 ∩ A1 B2 ∩ A2 0 ∞

A1 A2 B1 B2

n FIGURE 13. Failure of proposition IX.34 when there is j such that j j + R B 6⊆ A ≥0

9. Notes For convenient Newton diagrams there is a formula for generic intersection multiplicity in terms of integer lattice points in the region bounded by the diagram and the coordinate hyperplanes (see e.g. [Est12, Theorem 5]). A. Khovanskii informed the author that he had obtained (but did not publish) a bound equivalent to (87) which reduces the computation of generic intersection multiplicity to the convenient case. [HJS17] lists some other formulae for generic intersection mulitplicity in the general case. CHAPTER X

Number of solutions on (open subsets of) the afﬁne space

1. Introduction

In this section we compute the number of solutions on k (or more generally, on any given Zariski n open subset of k ) of generic systems of polynomials with given supports, and give explicit BKK-type characterizations of genericness in terms of initial forms of the polynomials. As a special case we derive genralizations of weighted (multi-homogeneous)-B´ezout theorems involving arbitrary weighted degrees (i.e. weighted degrees with possibly negative or zero weights).

2. The bound

n 2.1. Khovanskii’s formula. For polynomials f1,...,fn, and any Zariski open subset U of k , asin iso section VII.3 let [f1,...,fn]U be the sum of intersection multiplicities of f1,...,fn at all the isolated n points of V (f1,...,fn) U. Given a collection := ( 1,..., n) of n ﬁnite subsets of Z , deﬁne ∩ A A A ≥0 [ ,..., ]iso := max [f ,...,f ]iso : Supp(f ) , j =1,...,n A1 An U { 1 n U j ⊂ Aj } In this section we give a formula for [ ,..., ]iso in terms of (mixed volumes of) convex hulls of . A1 An U Aj For I [n], let T I := j : I = as in theorem IX.1, and let ⊆ A { Aj 6 ∅} ˜ (88) E ( ) := I [n]: there is I˜ I such that T I < I˜ A { ⊆ ⊇ | A| | |} iso The following result, which follows immediately from theorem III.20, implies that [ 1,..., n]U = iso I A A [ 1,..., n] , where UA := U E k . A A UA \ i∈ (A)

n k Lemma X.1. Let f1,...,fm k[x1S,...,xn] and let V := V (f1,...,fm) . Given I [n], if there

˜ ∈ ˜ I ⊂ ⊆ k exists I I such that j : fj I˜ 0 < I , then no point of V is isolated in V . ⊇ |{ |k 6≡ }| | | ∩ Deﬁne I

(89) E (U) := I [n]: k U = , { ⊆ ∩ ∅} (90) T (U, ) := I [n]: I/ E (U) E ( ), T I = I . A { ⊆ ∈ ∪ A | A| | |} The following result is also a straightforward consequence of theorem III.20.

∗ I Lemma X.2. If I T (U, ), then (k ) V (f1,...,fn) U = for generic f1,...,fn such that Supp(f ) , j =16∈ ,...,nA. ∩ ∩ ∅ j ⊆ Aj Remark X.3. It is possible that (i.e. the empty set) is in T (U, ); this is the case if and only if the origin is in U and 0 < [ ,..., ] <∅ (see example X.5). A A1 An 0 ∞ 1 THEOREM X.4 (Khovanskii ). Let j be the convex hull of j , j = 1,...,n. For I [n], let I I n I P A ⊆ := j R , and let πI : R R be the natural projection (as in (70)). Then Pj P ∩ → iso I I (91) [ 1,..., n] = MV( ,..., ) [π ( ′ ),...,π ( ′ )]0 U j1 j|I| [n]\I j1 [n]\I jn−|I| A A T P P × P P I∈ X(U,A) where for each I T (U, ), j ,...,j are the elements of T I , and j′ ,...,j′ are the elements of ∈ A 1 |I| A 1 n−|I| [n] T I , and [ ,..., ] is deﬁned as in (63). \ A · · 0

1A. Khovanskii described this unpublished formula to the author during the Askoldfest, 2017

143 144 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

The interpretation of the right hand side of (91) is straightforward - for each I T (U, ), the cor- ∗ I ∈ A responding summand counts with multiplicity the number of solutions on (k ) U of generic systems ∩ supported at 1,..., n. In the next section we present another formula which sometimes is more efﬁ- cient, since itA involvesA summing over elements from a proper subset of T (U, ). A Example X.5. Let j be the support of fj from examples VII.28 and IX.4, and U be a nonempty Zariski 3 A open subset of k . Then E ( )= , and A ∅ 1, 2, 3 , 3 , if 0 U (Case 1), {{ } { } ∅} ∈ T (U, )= 1, 2, 3 , 3 if 0 U, but U contains a point on the z-axis (Case 2), A {{ } { }} 6∈  1, 2, 3 otherwise (Case 3). {{ }} In Case 3, identity (X.4) and example VII.28 imply that  [ , , ]iso = MV( , , )=5 A1 A2 A3 U P1 P2 P3 {3} Since the projections of 2 and 3 onto the (x, y)-plane have nontrivial linear part, and 1 has integer length 1 (see ﬁg. 1), identityP (X.4P) implies that in Case 2, P [ , , ]iso = 5 + MV( {3}) [π ( ), π ( )] =5+1 1=6 A1 A2 A3 U P1 × {1,2} P2 {1,2} P3 0 · Finally, in Case 1, identity (X.4) and the computation from example IX.4 imply that [ , , ]iso = 6 + [ , , ] =6+6=12 A1 A2 A3 U P1 P2 P3 0

2 {3} 2 2 P1 1 1 1

2 3 2 3 1.5 2 1 1 0.5 1

1 2 2 π{1,2}( 2) π{1,2}( 3) P 2 P

{3} FIGURE 1. Computing MV( ) [π ( ), π ( )] P1 × {1,2} P2 {1,2} P3 0 2.2. A formula in the same spirit as the formula for generic intersection multiplicity. If ν is a

weighted order on k[x1,...,xn], we say that ν is centered at inﬁnity if ν(xi) < 0 for some i [n]. Given I ∈′ I [n], we say that ν is centered at k if ν(xi) 0 for each i I and ν(xi′ ) > 0 for each i [n] I.

⊆ ≥ n ∈ n ∈ \ k Given a collection S of subsets of [n], we denoteby kS the complement in of the coordinate subspaces I k for all I S , i.e. ∈

n n I

k k (92) kS := \ S I[∈ I We write S for the union, over all I , of the sets of weighted orders centered at k , and for the V ∈ S V∞ set of weighted orders centered at inﬁnity; the collection of primitive elements in S and ∞ are denoted ′ ′ V V respectively as S and . V V∞

n n n n

k k k Example X.6. Taking S = gives kS = and S = . If we take S = , then S = 0 ∅ V ∅ {∅} \{ } and S is the set 0 of weighted orders centered at the origin (see section IX.2.2). If S is the set of all

V V n ∗ n k subsets of [n] consisting of n 1 elements, then kS = ( ) and S is the set of all nonzero weighted orders which are not centered− at infinity. V n n Let := ( 1,..., n) be a collection of subsets of Z and j be the convex hull in R of j , A A A ≥0 P A j =1,...,n. Given a collection S of subsets of [n], define ∗ ′ (93) [ 1,..., n]S := min(ν) MVν (Inν ( 2),..., Inν ( n)) P1 P P − ′ ′ P P ν∈VXS ∪V∞ where MV′ ( ,..., ) is defined as in (54). ν · · 2. THE BOUND 145

n THEOREM X.7 ([Mon16]). Let U be a Zariski open subset of k . We continue to use the notation of T T I E I theorem X.4. Deﬁne 1(U, ) := I (U, ):1 TA , and for each I [n], set (U, ) := J I : J E (U) E ( ) . ThenA { ∈ A ∈ } ⊆ A { ⊆ ∈ ∪ A } iso I I I ∗ (94) [ 1,..., n] = [ , ,..., ]E I [π ( ′ ),...,π ( ′ )]0 U 1 j2 j|I| (U,A) [n]\I j1 [n]\I jn−|I| A A T P P P × P P I∈ X1(U,A) where for each I T (U, ), j = 1, j ,...,j are the elements of T I , and j′ ,...,j′ are the ∈ 1 A 1 2 |I| A 1 n−|I| elements of [n] T I . \ A There is an obvious analogy between formula (94) and the formula (73) for intersection multiplicity at the origin. The interpretations of the terms on the right hand side of (94) are also analogous to the interpretations of the corresponding terms of (73) described in section IX.2.1. As in thecase of (73), and unlike (91), the symmetry of the right hand side of (94) with respect to permutations of the j is not at all obvious. We use (94) to derive the symmetric formula of Huber and Sturmfels [HS97] andP Rojas [Roj99] in section X.4.1.

4 (1,1,1) (2,2,1)

(0,1,0) 2 (1,0,0) (1,1,0)

(-2,-2,-1) 1 2 3 2 inner normals to facets 1 + 3 P P of 1 + 3 P P

FIGURE 2. + and the inner normals to its facets P1 P3

iso Example X.8. We continue with 1, 2, 3 from example X.5, and compute [ 1, 2, 3]U using theo- A A A 3 A A A rem X.7 for nonempty Zariski open subsets U of k . It is straightforward to check that 1, 2, 3 , 3 if U contains a point on the z-axis (Case 1 or 2 of example X.5), T1(U, )= {{ } { }} A 1, 2, 3 otherwise (Case 3 of example X.5). ({{ }} ′ On the other hand, if we change the order of the j , or equivalently, consider the collection := ′ A A ( 2, 1, 3), then one checks that T1(U, )= 1, 2, 3 for any (nonempty) U, and we apply identity (94A ) toA ′Ato reduce computation. In particular,A we{{ have }} A iso iso ∗ ′ [ 1, 2, 3]U = [ 2, 1 3]U = [ 2, 1, 3]E (U,A) = min(ν) MVν (Inν ( 1), Inν ( 3)) A A A A A A P P P − P2 P P ν∈V′ ∪V′ E (XU,A) ∞ ′ The inner normals to the facets of 1 + 2 are listed in ﬁg. 2. The only elementin ∞ is ( 2, 2, 1). If the origin is in U, then E (U)= , andP thereforeP it follows from example VII.3 andV ﬁg. 3 that− − − ∅ iso [ 1, 2, 3]U = min( 2, 2, 1) Area( )=6 2=12 A A A − P2 − − − · · If the origin is not in U, but U contains some other points of the z-axis, then for I = 1, 2, 3 , the set I ′ ′ { } E (U, ) contains the emptyset, but not 3 . It follows that I does not contain (1, 1, 0), but it A { } VE (U,A′) contains each (primitive) weighted order centered at the origin (see example X.6); in particular, it contains

(2, 2, 1) and (1, 1, 1). Since minP2 (ν)=0 when ν = (1, 0, 0) or (0, 1, 0), it does not matter if these two elements are in ′ . It then follows from ﬁg. 3 that VE I (U,A′) iso [ 1, 2, 3]U = 12 min(2, 2, 1) Area( ) min(1, 1, 1) Area( )=12 4 0 3 2=6 A A A − P2 · ∅ − P2 · − · − · 146 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

ν =(−2, −2, −1) ν = (2, 2, 1)

+ = + =

Inν (P1) Inν (P3) Inν (P1 + P3) Inν (P1) Inν (P3) Inν (P1 + P3)

ν = (1, 1, 1) ν = (1, 1, 0)

+ = + =

Inν (P1) Inν (P3) Inν (P1 + P3) Inν (P1) Inν (P3) Inν (P1 + P3)

FIGURE 3. The image under ψ of initial faces of , and + ν P1 P3 P1 P3

′ If U does not intersect the z-axis, then (1, 1, 0) is also an element of I , and it follows that VE (U,A′) iso [ 1, 2, 3]U =6 min(1, 1, 0) Area( )=6 1 1=5 A A A − P2 · − · The computations therefore agree with those from example X.5. Note that formulae (91) and (94) resolve the cases in the opposite order!

3. Derivation of the formuale for the bound

In this section we prove theorems X.4 and X.7. Throughout this section := ( 1,..., n) denotes n A n A A a collection of ﬁnite subsets of Z , and j denotes the convex hull of j in R , j = 1,...,n. Asin ≥0 P A the preceding chapters, we write ( ) for the space of n-tuples of polynomials supported at , and as in L A A deﬁnition IX.23, we write ˇ ( ) be the collection of all strongly -non-degenerate (f1,...,fn) ( ). N A n ˇ A ˇ ∈ L A Given a Zariski open subset U of k , let (U, ) for all (f1,...,fn) ( ) such that for all I E (U) E ( ), N A ∈ N A 6∈ ∪ A ∗ I

(95) (V (f ,...,f ) (k ) ) U = 1 n ∩ \ ∅ Proposition IX.25 and lemma IX.26 imply that ˇ (U, ) contains a nonempty Zariski open subset of ( ). Assertion (4) of theorem IV.20 therefore impliesN A that in order to prove theorems X.4 and X.7, it L A sufﬁces to show that [f1,...,fn]U equals the quantities from the right hand sides of (91) and (94) for all (f ,...,f ) ˇ (U, ). 1 n ∈ N A 3.1. Proof of theorem X.4. Theorem X.4 follows from the following result.

Proposition X.9. Let (f1,...,fn) ˇ (U, ) and I [n] (E (U) E ( )). ∈ N A ∗ I⊆ \ ∪ A (1) If I T (U, ), then V (f1,...,fn) (k ) = . (2) If I 6∈ T (U, A), then ∩ ∅ ∈ A I I (96) [f1,...,fn]a = MV( ,..., ) [π ( ′ ),...,π ( ′ )]0 j1 j|I| [n]\I j1 [n]\I jn−|I| ∗ I P P × P P a∈X(k ) where j ,...,j are elements of T I , and j′ ,...,j′ are elements of [n] T I . 1 |I| A 1 n−|I| \ A T E I PROOF. If I (U, ) ( ), then TA > I , and property (a) of strongly -non-degenerate 6∈ A ∪ A ∗ I | | | | A systems imply that V (f1,...,fn) (k ) = , which proves assertion (1). Now pick I T (U, ) and ∗ I ∩ ∅ ∈ A a V (f1,...,fn) (k ) . Since proper non-degeneracy implies BKK non-degeneracy when dimension of∈ the ambient afﬁne∩ space is equal to the number of polynomials, properties (a) and (d) of strongly -non- I I I A I degenerate systems imply that (fj k : j TA) are ( j : j TA)-non-degenerate, and corollary VII.15 ∗ |I ∈ A I ∈ implies that V (f1,...,fn) (k ) is ﬁnite. If j TA, theorem III.20 implies that C := V (fi : i = ∗ I ∩ ∈ 6 j) (k ) is purely one dimensional near a. Property (c) of strongly -non-degenerate systems and ∩ A 3.DERIVATIONOFTHEFORMUALEFORTHEBOUND 147

I ′ ′ I corollary IX.22 imply that k is an irreducible component of V := V (fj′ : j TA), and no irreducible ′ I 6∈ component of V other than k contains any irreducible component of C. Corollary IV.22 then implies that

I I ′ ′ k (97) [f1,...,fn]a = [fj k ,...,fj ]a [fj ,ǫ,...,fj ,ǫ]0 1 | |I| | × 1 n−|I| ∗ I for generic ǫ = (ǫ ,...,ǫ ) (k ) , where f ′ are formed from f ′ by substituting ǫ for x for all i I. 1 n jk,ǫ jk i i ∈ ∗ I ∈ Due to property (95), assertion (2) follows by summing (97) over all a (k ) due to theorems VII.7 I I ∈

I I k and IX.8 (after using ( ,..., )-non-degeneracy of fj k ,...,fj and non-degeneracy at the Aj1 Aj|I| 1 | |I| | origin of f ′ ,...,f ′ for generic ǫ). j1,ǫ jn−|I|,ǫ

3.2. Proofoftheorem X.7. Let (f1,...,fn) ˇ (U, ) and C be the subscheme deﬁned by f2,...,fn

′ I ∈ N ′A I I k on U := U I∈E (A) k . For each I [n] such that U = , let Cj j be the set of irreducible

\ I ⊆ ∩ 6 ∅ n { } I k components of C such that k is the smallest coordinate subspace of containing each C . S j ′ I ′I

T k Claim X.10. Let 1(U, ) be as in theorem X.7, I [n] such that U = , and A be as in definition IX.23. A ⊆ ∩ 6 ∅ T (1) If CI is nonempty, then I T (U, ). { j }j ∈ 1 A (2) If I T (U, ), then [n] T I ′I . ∈ 1 A \ A ∈ TA I E PROOF. For the first assertion, pick I [n] such that Cj j is nonempty. Since I ( ), it follows I ⊆I { } 6∈ A that TA I . On the other hand, if TA 1 I , then property (a) of strong -non-degeneracy | | ≥ | | |I \{ }| ≥ | | A and lemma IX.17 implies that each CJ j is a point, which contradicts theorem III.20. Accordingly I I { } T TA 1 = I 1 and TA = I , which imply that I 1(U, ), as required. The second assertion follows| \{ from}| the| |− definition| of E| ( |) and| corollary IX.13. ∈ A A

For each I T1(U, ), property (a) of strong -non-degeneracy and assertion (1) of corollary VII.24 ∈ I A I A imply that either Cj j is empty, or each Cj has dimension one. It then follows from claim X.10 that C is a curve, and therefore{ } lemma IV.17, proposition IV.16, and theorem IV.12 imply that

[f1,...,fn]U = orda(f1 C )= [f2,...,fn]CI orda(f1 CI ) | j | j aX∈C I,j,aX T where [f2,...,fn]CI are deﬁned as in section IV.5. Pick I 1(U, ). Let j1 = 1, j2,...,j|I| be the j ∈ A I ′ ′ I elements of TA, and j1,...,jn−|I| be the elements of [n] TA. Property (c) of strongly -non-degenerate I \ ′ A systems and corollary IX.22 imply that k is an irreducible component of V := V (f ′ ,...,f ′ ), andif j1 jn−|I| I ′ I Cj j is nonempty, then no irreducible component of V other than k contains any irreducible component of{ C}. Proposition IV.21 then implies that

I I I I ′ ′ k [f1,...,fn]U = orda(f1 C )[fj2 k ,...,fj|I| ]C [fj ,ǫ,...,fj ,ǫ]0 | j | | j × 1 n−|I| I,j,aX ∗ I for generic ǫ = (ǫ1,...,ǫn) (k ) , where fj′ ,ǫ are formed from fj′ by substituting ǫi for xi for all i I. ∈ k k ∈ Properties (b), (d) of strongly -non-degeneratesystems and theorem IX.8 imply that [fj′ ,ǫ,...,fj′ ,ǫ]0 = A 1 n−|I| I [π[n]\I ( j′ ),...,π[n]\I ( j′ )]0 for generic ǫ k . In order to prove theorem X.7 therefore it sufﬁces P 1 P n−|I| ∈ to show that for each I T (U, ), ∈ A I I I ∗

I I I I I k orda(f1 C )[fj2 k ,...,fj|I| ]C = [ 1 , j ,..., j ]E | j | | j P P 2 P |I| (U,A) j,a X

I k k Identify k with , where k := I . Proposition VII.23 and claim VII.24.1 imply that we can ﬁnd a k-dimensional convex rational polytope| | such that P I , k ֒ XP • → I

I I ¯ k V (fj2 k ,...,fjk ) extends to a complete curve C on XP , and • | | I I f1 restricts to a nonzero rational function on C¯ which is representable near every point of C¯ as • a quotient of non zero-divisors. 148 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

Corollary IV.13 implies that

I I I I I I I I

k k k orda(f1 C )[fj2 k ,...,fjk ]C = orda(f1 C )[fj2 ,...,fjk ]C | j | | j − | j | | j j a∈U ′ j a∈C¯I \U ′ X X X Xj ¯I I ¯I ′ where Cj are the closures of Cj in XP . Property(95)ofthe fj implies that a Cj U if and only if either I′ ′ ∈ I\

k E E a for some I (U) ( ), or (the germ at) a is a branch at inﬁnity of Cj . Corollary VII.35 then implies∈ that ∈ ∪ A

′ I I

I I I I k orda(f1 C )[fj2 k ,...,fjk ]C = min(ν) MVν (Inν ( j2 ),..., Inν ( jk )) | j | | j − PI P P j a∈U ′ ν∈V′I ∪V′I 1 X X E IX(U,A) ∞ ′I ′I where (respectively I ) is the set of primitive weighted orders on k[x : i I] which are V∞ VE (U,A) i ∈ I′ ′ I centered at inﬁnity (respectively at k for some I E (U, )). Since the right hand side of the preceding I I I ∗ ∈ A identity is precisely [ , ,..., ]E I , this completes the proof of theorem X.7. P1 Pj2 Pj|I| (U,A) 4. Other formulae for the bound

Throughout this section we continue to use to denote a collection ( 1,..., n) of ﬁnite subsets of n A A A Z and j to denote the convex hull of j , j =1,...,n. ≥0 P A 4.1. The formula of Huber-Sturmfels and Rojas. Let t be a new indeterminate. Fix positive integers

n k k ,...,k . Note that for each f ,...,f k[[x ,...,x ]], and each Zariski open subset U of , 1 n 1 n ∈ 1 n (98) iso k1 kn iso [f1,...,fn]U = [t,f1 + c1t ,...,fn + cnt ]U×k

for any c ,...,c k. It follows that 1 n ∈ [ ,..., ]iso = [ ˆ ,..., ˆ ]iso A1 An U A0 An Uˆ ˆ ˆ n+1 ˆ n+1 where U := U k, 0 := (1, 0,..., 0) Z and j := (kj , 0,..., 0) ( 0 j ) Z for × A { }⊂ ≥0 A { } ∪ { } × A ⊂ ≥0 j = 1,...,n. Let ˆ := ( ˆ0,..., ˆn). It is straightforward to check that T1(U,ˆ ˆ) = [n + 1] , so that theorem X.7 impliesA that A A A { } [ ,..., ]iso = [ ˆ ,..., ˆ ]∗ = νˆ MV′ (In ( ˆ ),..., In ( ˆ )) A1 An U P0 Pn E (Uˆ)∪E (Aˆ) − 0 νˆ νˆ P1 νˆ Pn νˆ∈Vˆ′ ∪Vˆ′ E (UˆX)∪E (Aˆ) ∞ where ˆ are the convex hulls of ˆ , and νˆ ranges over the collection ˆ′ ˆ′ of all primitive Pj Aj VE (Uˆ )∪E (Aˆ) ∪ V∞

I k weighted orders on k[t, x ,...,x ] which are either centered at inﬁnity or centered at for some I 1 n ∈ E (Uˆ) E ( ˆ), and νˆ :=ν ˆ(t). Now, since dim(Pˆ )=0, either theorem VII.5 or theorem VII.30 implies ∪ A 0 0 that MV( ˆ ,..., ˆ )=0. Therefore assertion (1) of proposition VII.27 implies that P0 Pn (99) [ ,..., ]iso = νˆ MV′ (In ( ˆ ),..., In ( ˆ )) A1 An U 0 νˆ νˆ P1 νˆ Pn νˆ6∈Vˆ′ ∪Vˆ′ E (UˆX)∪E (Aˆ) ∞ ˆ′ ˆ′ If ν is a primitive weighted order on k[t, x ,...,x ], then ν if and only if all the 1 n 6∈ VE (Uˆ )∪E (Aˆ) ∪ V∞ following hold: (i) ν is nonnegative, and (ii) for each I E (U) E ( ), there is i′ I such that ν(x )=0. ∈ ∪ A 6∈ i ˆ′ Let (U, ) be the set of all primitive weighted orders νˆ on k[t, x1,...,xn] which satisfy properties (i), (ii) andV inA addition satisfy the following:

(iii) νˆ0 :=ν ˆ(t) is positive.

Since a summand on the right hand side of (99) has a nonzero contribution only if νˆ0 is positive, it follows n that for any Zariski open subset U of k , (100) [ ,..., ]iso = νˆ MV′ (In ( ˆ ),..., In ( ˆ )) A1 An U 0 νˆ νˆ P1 νˆ Pn νˆ∈VˆX′(U,A) 4. OTHER FORMULAE FOR THE BOUND 149

n ∗ |J| n−|J|

k k k In the case that k = C, E ( ) = , and U = V ( xj ) = ( ) for some J [n], A ∅ \ j∈J ∼ × ⊆ formula (100) appeared in [HS97]. In this case ˆ′(U, ) consists of all primitive nonnegative weighted V AQ orders νˆ on k[t, x1,...,xn] such that νˆ0 is positive, and νˆ(xj )=0 for each j J. The sum on the right hand side of (100) in this case was termed in [HS97] as the I-stable mixed volume∈ (where I := [n] J) of \ 1,..., n. J. M.Rojas [Roj99] observed that the formula of [HS97] works over all algebraically closed Aﬁelds. A

4.2. Estimates in terms of single mixed volumes. If U is nonempty, identity (91) implies that iso ˜ [ 1,..., n]U MV( 1,..., n). On the other hand, since j j := j 0 , it trivially follows A A ≥iso P′ P ′ iso ′ ′ A ⊆ A′ A ∪{ } ′ that [ 1,..., n]U [ 1,..., n]U = MV( 1,..., n), where j are the convex hull of j , and the A A ≤ A A P P P n A last equality follows from (91). It follows that for nonempty Zariski open subsets U of k , (101) MV(conv( ),..., conv( )) [ ,..., ]iso MV(conv( 0 ) ..., conv( 0 )) A1 An ≤ A1 An U ≤ A1 ∪{ } An ∪{ } The upper bound in (101) is due to T. Y. Li and X. Wang [LW96]. We now examine when these bounds are exact. The lower bound is easier to handle; the following result follows directly from theorems VII.30 and X.4.

n Proposition X.11. Let U be a nonempty Zariski open subset of k . Then the following are equivalent: (1) The ﬁrst inequality in (101) holds with equality. T I I (2) For each I (U, ) [n] , j1 R ,..., j|I| R are dependent, where j1,...,j|I| are ∈ I A \{ } P ∩ P ∩ the elements of TA. Remark X.12. Since an empty collection of convex polytopes is by deﬁnition independent, condition (2) of proposition X.11 implies in particular that T (U, ), which in turn implies that either U does not contain the origin, or [ ,..., ] is zero or∅ 6∈(remarkAX.3). A1 An 0 ∞ Following A. Khovanskii [Kho78], We say that = ( 1,..., n) is regularly attached to the coor- A A A I dinate cross if for each proper subset I of [n], the set of nonempty elements of j R : j = 1,...,n {P ∩ } is dependent; in particular this implies (taking I = ) that the origin belongs to at least one of the j . The following is immediate from proposition X.11. ∅ A

Corollary X.13 (Khovanskii [Kho78]). If U is nonempty and is regularly attached to the coordinate cross, then [ ,..., ]iso = MV( ,..., ). A A1 An U P1 Pn Now let M := MV(conv( 1 0 ) ..., conv( n 0 )) be the upper bound from (101). Consider as in (98) the system A ∪{ } A ∪{ }

(102) fj + t =0, j =1, . . . , n, where fj are generic polynomials supported at j , with NP(fj ) = conv( j ) = j. For generic t = 0, A A P ∗ n 6 the corresponding system has precisely M isolated solutions, all of which are on (k ) . Therefore, the number of solutions of the system at t = 0 is also M if and only if there is no curve of solutions of ∗ n the system (102) on (k ) that escapes U or becomes non-isolated as t approaches 0. Theorems VII.5 and VII.30 and proposition VII.23 imply that such a curve exists if and only if there is a weighted order νˆ

on k[x1,...,xn,t] such that (i) νˆ(t) > 0,

I k (ii) the restriction of νˆ to k[x1,...,xn] is either centered at inﬁnity or at for some I E (U) E ( ), and ∈ ∪ Aˆ ˆ ˆ n+1 (iii) Inνˆ( 1),..., Inνˆ( n) are independent, where j := NP(fj + t) R . P P P ⊂ ≥0

Let ν be the restriction of νˆ to k[x ,...,x ]. Let m := max min (ν): j = 1,...,n . Since 1 n ν { Pj } dim(In ( ˆ )) dim(In ( ))+1, and since νˆ(t) > 0, it follows from the deﬁnition of depen- νˆ j Pj ≤ ν j Pj dence of polytopes that In ( ˆ ),..., In ( ˆ ) are independent if and only if P νˆ P1 P νˆ Pn (iv) mν > 0, and 150 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

(v) for all J [n], ⊆

J if minPj (ν)

ity [f ,...,f ] n = [ ,..., ] n , where := ( ,..., ) is a given collection of ﬁnite subsets of k 1 n k 1 n 1 n n A A A A nA Z , and (f1,...,fn) ( ). We consider the case U = k ; in a sense this is the most important case, ≥0 ∈ L A and it already captures the essence of the general case. We also assume for simplicity that E ( ) = , A ∅ which ensures in particular that for generic (f1,...,fn) ( ), all points in the set V (f1,...,fn) of n ∈ L A common zeroes of f1,...,fn in k are isolated. In this scenario, if we apply the intuitive reasoning from section VII.3.2 that motivated the non-degeneracy condition for Bernstein’s theorem, we are led to the following condition: I

(ND ) I I k ∞ f k ,...,f are -non-degenerate at infinity for each I [n]. 1| n| A ⊆ I I I I I where := ( 1,..., n) = ( 1 R ,..., n R ), and non-degeneracy at infinity is defined as A A A A ∩ A ∩ n follows: given a collection := ( 1,..., m) of finite subsets of Z and gj k[x1,...,xn], such B B B ≥0 ∈ that Supp(g ) , j = 1,...,m, we say that g ,...,g are - non-degenerate at infinity if for each j ⊆ Bj 1 m B weighted order ν centered at infinity (see section X.2.2), there is no common root of InBj ,ν (gj ), j = ∗ n 1,...,m, on (k ) . We now present a series of examples which illustrate how condition (ND∞) falls short of characterizing the correct non-degeneracy condition, and which also suggest the ways to amend it. In all these examples would denote the convex hull of , j =1,...,n. Pj Aj 3 Example X.16. Let 1 = 2 = (0, 0, 0), (1, 1, 0), (0, 1, 1), (1, 0, 1) Z≥0 and 3 = 1 + (0, 0, 1) . Then = is a tetrahedronA A and{ is a translation of , and therefore}⊂ A A { } P1 P2 P3 P1 MV( , , ) = 3! Vol( )=2 P1 P2 P3 P1 If f are polynomials such that NP(f )= , then j j Pj f1 = a1 + b1x1x2 + c1x2x3 + d1x3x1

f2 = a2 + b2x1x2 + c2x2x3 + d2x3x1

f3 = x3(a3 + b3x1x2 + c3x2x3 + d3x3x1)

∗ 3 ∗ k where aj ,bj,cj , dj k . We write V for the set of common zeroes of f1,...,f3 on , and V for ∗ 3 ∈ V (k ) . ∩ 2If dim(P)= n − 1, then both of the primitive normals to P are considered to be inner. 5.EXAMPLESMOTIVATINGTHENON-DEGENERACYCONDITIONS 151

1 2

0.8 1.8

0.6 1.6

0.4 1.4

0.2 1.2

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.5 0.5 1 1

= 3 = 1 + (0, 0, 1) P1 P2 P P { }

FIGURE 4. Newton polytopes of example X.16

∗ (a) Ifall aj,bj ,cj , dj are generic, then it is straightforward to check directly that V = V . Therefore iso iso

theorem VII.5 implies that [ 1, 2, 3] 3 = [f1,f2,f3] 3 = MV( 1, 2, 3)=2.

k k (b) If a = a , b = b , andA theA remainingA coefﬁcients are generic,P thenPV P= V ∗ C, where 1 2 1 2 ∪ C := x3 = a1 + b1x1x2 =0 is a positive dimensional component of V (f1,f2,f3). However, { ∗ } iso iso

f1,f2,f3 still satisfy (ND ), and theorem VII.7 implies that [f1,f2,f3] 3 = 2 = [ 1, 2, 3] 3 . k k A A A (c) If a1 = a2 = a3, b1 = b2 = b3, and the rest of the coefﬁcients are generic, then again V = V ∗ C. However, (ND∗) fails for the weighted order ν corresponding to weights ( 1, 1, 2) ∪ iso iso −

for (x,y,z), and theorem VII.7 implies that [f1,f2,f3] 3 < 2 = [ 1, 2, 3] 3 . (It is straight-

k k ∗ isoA A A forward to verify directly that in this case V = and [f ,f ,f ] 3 =0.) ∅ 1 2 3 k Part (b) of example X.16 shows that it is possible that V (f1,...,fn) has a positive dimensional n iso iso

component on k , but still [f ,...,f ] n = [ ,..., ] n (where = Supp(f )). (This does not k 1 n k 1 n j j ∗ n A A A happen in the case of (k ) , see proposition VII.34.) Moreover, since the intersection of the curve C with the “torus” of the (x1, x2)-plane is nonempty, in part (b) of example X.16, condition (ND∞) is violated for I = 1, 2 . However, note that the intersections of and with the (x , x )-plane (in { } P1 P2 1 2 Rn) are dependent in the terminology of deﬁnition VII.29. Moreover, in part (c) of example X.16, where iso iso

[f ,f ,f ] 3 < [ , , ] 3 , condition (ND ) is violated with I = 1, 2, 3 , and the corresponding

k 1 2 3 k 1 2 3 ∞ polytopes are independentA A A. This motivates the following definition. { } n Definition X.17. An ordered collection = ( 1,..., m), m 1, of collections of finite subsets of R I B B B ≥ I I is called R -dependent if there is a nonempty subset J of [m] such that := j R is nonempty for Bj B ∩ each j J, and the collection conv( I ): j J of convex polytopes is dependent (definition VII.29); ∈ { Bj ∈ } otherwise we say that is RI -independent. B Example X.16 suggests that I (i) Condition (ND∞) should be checked only for those I [n] such that is R -independent. ⊆ A This, however, is not enough, as the next example shows.

1 2 (-1,-1,1)

0.8 1.8

0.6 1.6 (1,1,0) (0,-1,0) 0.4 1.4

0.2 1.2 (0,1,-1) 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.5 0.5 1 1 outer normals to facets 1 = 2 3 P P P of 1 P

FIGURE 5. Newton polytopes of example X.18 152 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

Example X.18. Consider the following system of polynomials:

f1 = a1 + b1x1 + c1x2x3 + d1x3x1

f2 = a2 + b2x1 + c2x2x3 + d2x3x1

f3 = x3(a3 + b3x2 + c3x2x3 + d3x3x1) ∗ where aj ,bj,cj , dj k , and j := Supp(fj ), j =1, 2, 3. We continue to write V := V (f1,f2,f3) and ∗ ∗ 3 ∈ A ∗ V := V (k ) . When all the coefﬁcients are generic, then it is straightforward to check that V = V , so that theorem∩ VII.5 implies that iso iso

[f ,f ,f ] 3 = [ , , ] 3 = MV( , , ) k 1 2 3 k A1 A2 A3 P1 P2 P3 We compute MV( , , ) using proposition VII.27. If ν is the primitive outer normal to any of the P1 P2 P3 facets of 1 = 2, it is straightforwardto check that the image of the correspondingfacet under the map ψν P P ′ from the deﬁnition of MVν ( ,..., ) is (up to a translation) the triangle with vertices (0, 0), (0, 1), (1, 0), and therefore MV′ (ld ( )·, ld ( · )) = 2! Area( )=1. It follows thatT ν ν P1 ν P2 T iso [ , , ] 3 = MV( , , ) A1 A2 A3 k P1 P2 P3 = max(0, 1, 0)+ max(0, 1, 1)+ max(1, 1, 0)+ max( 1, 1, 1) P3 − P3 − P3 P3 − − =0+0+1+1 =2

If a1 = a2 and b1 = b2 and the other coefﬁcients are generic, then it is straightforward to check directly ∗ iso that V = and V is the curve x3 = a1 + b1x1 = 0 , so that [f1,f2,f3] 3 = 0. However, the ∅ { } k only I 1, 2, 3 such that is RI -independent is I = 1, 2, 3 , and it is straightforward to check ⊆ { } A { } I that fj k = fj are in fact -non-degenerate at inﬁnity. In particular, f1,f2,f3 satisfy condition (i), but | iso Aiso

[f1,f2,f3] 3 < [ 1, 2, 3] 3 . k k A A A

′ ∗ I′ k Given I I [n] and a weighted order ν on k[xi : i I], we say that ν is centered at ( ) if ⊆ ⊆ ′ ′ ∈ ν(xi)=0 for all i I and ν(xi) > 0 for all i I I . It is straightforward to check that in example X.18, ∈ ∈ \ ∗ 3 the only nonzero weighted orders ν such that Inν (fj ), j =1, 2, 3, have a common zero on (k ) are of the ∗ I′ ′ form (0, 0,ǫ) for ǫ> 0, i.e. they are centered at (k ) with I := 1, 2 . It turns out that is hereditarily I′ { } Aiso -dependent (see section X.6) and therefore example X.18 suggests that for [f ,...,f ] n to be equal R 1 n k iso to [ ,..., ] n , the following condition needs to be satisﬁed in addition to (i): A1 An k I I I (ii) For each I [n] such that is not hereditarily R -dependent, InA ,ν (fj k ) do not have any ⊆ A j |

∗ n ∗ I′

k k common zero on (k ) for all weighted orders ν on [xi : i I] which are centered at ( ) ′ ∈ for some I′ ( I such that is hereditarily RI -dependent. A Example X.19. Let f1,...,f4 be polynomials in (x1,...,x4) such that f1 and f2 are polynomials in (x1, x2) with nonzero constant terms, the (two dimensional) mixed volume of NP(f1) and NP(f2) is nonzero, and

f = x f (x , x )+ x f (x , x ) k[x , x , x , x ], j =3, 4, j 3 j,1 1 2 4 j,2 1 2 ∈ 1 2 3 4 where f3,1,f3,2,f4,1,f4,2 are nonzero polynomials in (x, y) such that NP(f3,1) and NP(f4,1) are positive dimensional. Let j be the support of fj and j be the Newton polytope of fj , j = 1,..., 4. The A P I only I 1, 2, 3, 4 such that the supports of fj are R -independent is I = 1, 2 . It follows that for ⊆ { } { } generic coefﬁcients, all the common zeroes of f1,...,f4 are isolated and contained in the (x1, x2)-plane. Moreover, theorem X.4 implies that

iso {1,2} {1,2} [ , , , ] 4 = MV( , ) [π ( ), π ( )] = MV( , ) A1 A2 A3 A4 k P1 P2 × {3,4} P3 {3,4} P4 0 P1 P2 ∗ 2 Now ﬁx BKK non-degenerate f1,f2, anda commonzero z = (z1,z2) of f1,f2 on (k ) . Take f3,1 and f4,1 such that f3,1(z)= f4,1(z)=0. Then (z1,z2, t, 0) : t k V (f1,f2,f3,f4), so that (z1,z2, 0, 0) is no { ∈ }⊆ {1,2} longer an isolated pointof V (f1,f2,f3,f4) (eventhoughit is an isolated point of V (f1,f2,f3,f4) k ). iso iso ∩

It follows that [f1,f2,f3,f4] 4 < MV( 1, 2) = [ 1, 2, 3, 4] 4 . However, f1,...,f4 satisfy both

k k conditions (i) and (ii). P P A A A A 6. NON-DEGENERACY CONDITIONS 153

Example X.19 leads us to another condition that needs to be satisﬁed in addition to (i) and (ii) for iso iso

[f ,...,f ] n to be equal to [ ,..., ] n : k 1 n k 1 n A A I I I (iii) For each I [n] such that is hereditarily R -dependent, InA ,ν (fj k ) do not have any com- ⊆ A j |

∗ n ∗ I′

k k mon zero on (k ) for all weighted orders ν on [xi : i I] which are centered at ( ) for ′ ∈ some I′ ( I such that is not hereditarily RI -dependent. A Note that condition (iii) is precisely the opposite of condition (ii).

6. Non-degeneracy conditions

n Let := ( 1,..., n) be a collection of finite subsets of Z and I [n]. We say that is A A A ≥0 ⊆ A hereditarily RI -dependent if is RI -dependent (see definition X.17) and there is I′ I such that ′ A ⊇ (a) is RI -dependent, A I′ ′ (b) TA = I , | ˜ | | | (c) T I > I˜ for each I˜ such that I I˜ ( I′. | A| | | ⊆ Remark X.20. If I = , then RI is the origin, and is RI -dependent if and only if the origin belongs to ∅ I A I some j . However, even if is R -dependent, it might not be hereditarily R -dependent, see fig. 6. A A

A2 A3 A1

I FIGURE 6. Both ( 1, 2) and ( 2, 3) are R -dependent for I = . However, only A A A A ∅ the latter pair is hereditarily RI -dependent (with I′ = 2 ). { } The relevance of “hereditary dependence” to affine Bézout problem is given by proposition X.21 be- I I low: it states that if is hereditarily R -dependent, then for all (f1,...,fn) ( ), either V := ∗ I A I ∈ L A V (f ,...,f ) (k ) is empty, or all pointsof V are non-isolated in (the possibly larger set) V (f ,...,f ) 1 n ∩ 1 n (however, points of V I might be isolated in V I itself!). This is not necessarily true if is simply RI - dependent, e.g. the system (x + y 1, 2x y 2) (over a field of characteristic not equalA to two) has an isolated zero on the coordinate subspace− y −=0.− I ∗ I Proposition X.21. If is hereditarily R -dependent, then V (f1,...,fn) has no isolated point on (k ) for each (f ,...,f ) A ( ). 1 n ∈ L A We prove proposition X.21 in section X.7.1. Now we introduce the correct non-degeneracy condition n for (arbitrary open subsets of) the affine space. Let U be a Zariski open subset of k . Let E ( ) and E (U) be as in section X.2.1. Define A (104) D(U, ) := I [n]: I E (U) E ( ), is hereditarily RI -dependent A { ⊆ 6∈ ∪ A A } (105) I (U, ) := I [n]: I E (U) E ( ), is not hereditarily RI -dependent A { ⊆ 6∈ ∪ A A } We say that polynomials f1,...,fn are (U, )-non-degenerate if the following conditions are true: A ∗ I

I I I k k (a) For each nonempty I (U, ), there is no common zero of f1 k ,...,fn on ( ) U

∈ A n | | n \ k (note that this condition is vacuously true when U = k , or more generally, when U = S (see (92)) for some collection S of subsets of [n]),

(b) For each nonempty I I (U, ) and for each weighted order ν on k[xi : i I] such that ν is ∈ ∗ I′A ′ ∈ centered at inﬁnity or at (k ) for some I E (U) E ( ) D(U, ), there is no common ∈ ∗ n ∪ A ∪ A

I I I I k k zero of InA ,ν (f1 k ),..., InA ,ν (fn ) on ( ) . 1 | n |

k I I I I I k centered at ( ) for some I (U, ), thereis nocommonzeroof In (f1 k ),..., In (fn ) A1 ,ν An,ν ∗ n ∈ A | | on (k ) . 154 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

We prove the following result in section X.7.3.

iso THEOREM X.22 ([Mon16]). Assume [ 1,..., n]U > 0. Then for each (f1,...,fn) ( ), the following are equivalent: A A ∈ L A iso iso (1) [f1,...,fn]U = [ 1,..., n]U , (2) f ,...,f are (U,A)-non-degenerate.A 1 n A The collection (U, ) of all (U, )-non-degenerate (f1,...,fn) ( ) is a nonempty Zariski open subset of ( ).N A A ∈ L A L A 3 In both examples X.16 and X.18, D(k , ) is the set , 1 , 2 , 1, 2 of all subsets of 1, 2 , 3 A {∅ { } { } { }} { } and I (k , ) is the set of remaining subsets of 1, 2, 3 . It is straightforward to check that in cases (a)

A 3 { } 3 k and (b) of example X.16, f1,f2,f3 are (k , )-non-degenerate, but in case (c), condition (b) of ( , )- non-degeneracy fails for I = 1, 2, 3 and νA= ( 1, 1, 2), which is centered at inﬁnity. In example X.18A , 3 { } − condition (b) of (k , )-non-degeneracy fails for I = 1, 2, 3 and ν = (0, 0, 1), which is centered at

∗ I′ ′ A { } 4 k (k ) , where I := 1, 2 . In the scenario of example X.19, I ( , ) is the collection of all subsets

4 { } A 4 k of 1, 2 , and D(k , ) is the set of remaining subsets of 1, 2, 3, 4 , and condition (c) of ( , )-non- { } A { } ∗ I′ ′ A degeneracy fails for I = 1, 2, 3, 4 and ν = (0, 0, 1, 2), which is centered at (k ) , where I := 1, 2 . { } { } 6.1. A more efﬁcient formulation of (U, )-non-degeneracy. In this section we describe a criterion equivalent to (U, )-non-degeneracy, but whichA involves checking fewer conditions. Recall that for I I A I ⊆ [n], we write T := j [n]: j R = . Deﬁne A { ∈ A ∩ 6 ∅} D ∗ D I (U, ) := I (U, ): I = , TA = I (106) A { ∈ A 6 ∅ | | | |} = I [n]: I = , T I = I , is RI -dependent { ⊆ 6 ∅ | A| | | A } I ∗ I I (U, ) := I (U, ): I = , TA = I (107) A { ∈ A 6 ∅ | | | |} = I [n]: I = , T I = I , is RI -independent { ⊆ 6 ∅ | A| | | A } Proposition X.23 ([Mon16]). For polynomials f1,...,fn in (x1,...,xn), the following are equivalent:

(1) fi’s are (U, )-non-degenerate. (2) (i) propertyA(a) of (U, )-non-degeneracy holds, (ii) property (b) of (U, A)-non-degeneracy holds with I (U, ) replaced by I ∗(U, ), (iii) property (c) of (U, A)-non-degeneracy holds with D(U, A) replaced by D ∗(U, A). A A A We prove proposition X.23 in section X.7.1. Example X.24 (Warning!). In property (b) of (U, )-non-degeneracy D(U, ) can not be replaced by D ∗(U, ), and in property (c) of (U, )-non-degeneracyA I (U, ) can not Abe replaced by I ∗(U, ). Indeed,A at ﬁrst consider the system A A A

f1 =1+ x1

f2 =1+ x1 + x2

f3 =1+ x1 +2x2 + ax3x4

f4 = x4(1 + x1 + bx2 + cx3 + dx4)

∗ 4 k for generic a,b,c,d k . Let U = and j = Supp(fj ), j = 1,..., 4. Then D(U, ) is the collection of all subsets∈ of 1, 2, 3 , I (U, ) isA the collection of all subsets of 1, 2, 3, 4 containingA 4, D ∗(U, ) = 1, 2, 3 , and{ I ∗(}U, ) =A 1, 2, 3, 4 . Condition (b) of (U, { )-non-degeneracy} fails A {{ }} A {{ }} ∗ I′ A ′ with I = 1, 2, 3, 4 and ν = (0, 1, 1, 1), so that the center of ν is (k ) , where I := 1 D(U, ). However, it{ is straightforward} to check that condition (b) would not have been violated had{ D} ∈(U, ) beenA replaced by D ∗(U, ). Now consider the system A A g1 =1+ x1 + x2 + x3

g2 =1+ x1 +2x2 +3x3 ′ ′ ′ g3 = x2(ax1 + bx2 + cx3)+ x3(a x1 + b x2 + c x3)

g4 = x4(1 + x1) 7. PROOF OF THE NON-DEGENERACY CONDITIONS 155

′ ′ ′ ∗ 4 k where a,b,c,a ,b ,c are generic elements in k . Let := (Supp(g1),..., Supp(g4)), and U := . Then D(U, ) is the collection of all subsets of 1, 2,B3, 4 containing 4, I (U, ) is the collection of all subsets ofB 1, 2, 3, 4 not containing 4, D ∗(U,{ ) = }1, 2, 3, 4 , and I ∗(U,B ) = 1, 2, 3 . It { } B {{ }} B {{ }} is straightforward to check that g1,g2,g3,g4 violate condition (c) of (U, )-non-degeneracy with I = ∗ I′ ′ B 1, 2, 3, 4 and ν = (0, 1, 1, 1) (so that the center of ν is (k ) , where I := 1 I (U, )). It is also straightforward{ } to check that condition (c) would not have been violated had {I }(U, ∈ ) beenB replaced by I ∗(U, ). B B iso The following combinatorial description of strict monotonicity of [ ,..., ]U follows from theorem X.22 and proposition X.23 exactly as in the proof of theorem IX.32. · · Corollary X.25. Let conv( ), j = 1,...,n. Then [ ,..., ]iso [ ,..., ]iso. If Bj ⊆ Aj A1 An U ≥ B1 Bn U [ ,..., ]iso > 0, then the following are equivalent: A1 An U (1) [ ,..., ]iso = [ ,..., ]iso, A1 An U B1 Bn U (2) (a) For each nonempty I I (U, ) and for each weighted order ν on k[xi : i I] such that ∈ A∗ I′ ′ ∈ ν is centered at infinity or at (k ) for some I E (U) E ( ) D(U, ), the collection In (ND( I )) : In (ND( I )) Supp( ) = ∈ of polytopes∪ A is∪ dependent,A and { ν Bj ν Aj ∩ Bj 6 ∅} (b) for each nonempty I D(U, ) and for each weighted order ν on k[xi : i I] such that ν ∗ I′ ∈ A′ I ∈ I is centered at (k ) for some I I (U, ), the collection In (ND( )) : In (ND( )) ∈ A { ν Bj ν Aj ∩ Supp( j ) = of polytopes is dependent. B 6 ∅} ∗ (3) (a) For each nonempty I I (U, ) and for each weighted order ν on k[xi : i I] such that ∈ ∗AI′ ′ ∈ ν is centered at infinity or at (k ) for some I E (U) E ( ) D(U, ), the collection I I ∈ ∪ A ∪ A Inν (ND( j )) : Inν (ND( j )) Supp( j ) = of polytopes is dependent, and { B ∗A ∩ B 6 ∅} (b) for each nonempty I D (U, ) and for each weighted order ν on k[xi : i I] such that ν ∗ I′ ∈ A′ I ∈ I is centered at (k ) for some I I (U, ), the collection In (ND( )) : In (ND( )) ∈ A { ν Bj ν Aj ∩ Supp( ) = of polytopes is dependent. Bj 6 ∅} 7. Proof of the non-degeneracy conditions 7.1. Reduction to the more efficient non-degeneracy criterion. In this section we prove propositions X.21 and X.23.

PROOF OF PROPOSITION X.21. Let I′ I be as in the deﬁnition of hereditary non-degeneracy. By I′ ⊇ ′ I restricting all fj ’s to k , we may assume that I = [n]. Let Z be the set of isolated points V (f1,...,fn) ∗ I I which are on (k ) . Assume to the contrary of the claim that Z = . Let (g1,...,gn) ( ) be a system such that 6 ∅ ∈ L A

(1) NP(gj ) = conv( j ), j =1,...,n, and A I˜

(2) g I˜,...,g I˜ are properly -non-degenerate (see section VII.4.4) for each I˜ [n], where

k 1 k n I˜| | I˜ An˜ ⊆ := ( 1 R ,..., n R ) (proposition VII.22 implies that the set of such systems is a AnonemptyA Zariski∩ openA subset∩ of ( )). L A I Let t be a new indeterminate, and hj := tgj +(1 t)fj k[x1,...,xn,t], j =1,...,n. Pick z Z . Then − n+1∈ ∈ (z, 0) is an isolated point of V (h1,...,hn,t) on k , and therefore theorem III.20 implies that there is a n+1 ′ Zariski open neighborhood U of (z, 0) in k such that C := V (h1,...,hn) U is a curve. Let C be an ¯′ ′ n∩ 1 ′ n irreducible component of C containing (z, 0), and C be the closure of C in P P . Then C k ǫ ¯′ n × ′ 6⊆¯′ ×{ } for any ǫ k, and corollary IV.14 implies that C intersects P 1 . Pick a branch B of C centered ∈ ′ ¯′ n ×{ } at a point (z , 1) C (P 1 ). Deﬁne IB [n + 1] and the weighted order νB on the coordinate

IB ∈ ∩ ×{ } ˜ ⊆ ˜ k ring of k as in deﬁnition IX.15. Let I := I [n] and ν˜ be the restriction of ν to [x : i I]. B ∩ B i ∈

Since NP(f ) NP(g ) for each j, it follows that ν˜(f I˜) ν˜(g I˜). Since ν (t 1) > 0, it follows k j ⊆ j j |k ≥ j | B − that , and therefore I I˜ for each . Lemma IX.17 then implies that νB(t)=0 Inν (hj k B ) = t Inν˜(gj ) j B | |k

∗ I˜ ¯′ ∗ I k Inν˜(gj I˜) have a common zero in (k ) . On the other hand, since C contains the point (z, 0) ( ) , |k ∈ it follows that I I˜, and therefore it follows from the deﬁnition of hereditary non-degeneracy (applied ′ ⊆ I˜ ˜ I˜ ˜ I with I = [n]) that either TA > I , or TA = I and is R -dependent. In any event, there is a I˜ | | | | | | | | A

nonempty subset J of T such that dim( NP(gj I˜)) < J . Since g1 I˜,...,gn I˜ are properly

I˜-non-degenerate, it then follows by definition of proper non-degeneracy that there is no common zero A ∗ I˜ of In (g I˜) on (k ) . This contradiction completes the proof. ν˜ j |k PROOF OF PROPOSITION X.23. Since D ∗(U, ) D(U, ) and I ∗(U, ) I (U, ), it suffices to show the following: A ⊆ A A ⊆ A (1) property (b) of (U, )-non-degeneracy holds if it holds with I (U, ) replaced by I ∗(U, ), (2) property (c) of (U, A)-non-degeneracy holds if it holds with D(U, A) replaced by D ∗(U, A). A A A It follows from the definition of hereditary non-degeneracy that for every I I (U, ) I ∗(U, ), ′ I ∗ ′ I′ ′ ′ ∈ A \′ ˜ A there exists I (U, ) such that I ( I and TA > I for each I such that I I ( I. The same statement∈ also holdsA if we replace I (U, ) and| I|∗(U,| |) respectively by D(U, )⊆and D ∗(U, ). I˜ A A A A Since restricting all fj to k yields a system with the same number of nonzero polynomials as the num- I ber of variables, it suffices to prove the following claim: “if there is I [n] such that TA > I and

∗ n ⊆ | | | | k

I I I I k k InA ,ν (f1 k ),..., InA ,ν (fn ) have a common zero on ( ) for some weighted order ν on [xi : 1 | n | ˜ ˜

k ˜ i I], then there is I ) I and a weighted order ν˜ on [x˜i : i I] such that ν˜ is compatible ∈ ∈ ∗ n

with ν, and In I˜ (f1 I˜),..., In I˜ (fn I˜) have a common zero in u˜ (k ) .” This follows from k A ,ν˜ k A ,ν˜ 1 | n | ∈ lemma IX.28. 7.2. Understanding condition (c) of (U, )-non-degeneracy. In this section we show that if condi- A ′ tion (c) of (U, )-non-degeneracy is violated for (f1,...,fn) ( ) with I, ν and I , then the common ∗ n A ∈ L A

k I I I I k zero on ( ) of In (f1 k ),..., In (f1 ) corresponds to a non-isolated point of V (f1,...,fn). A1 ,ν An,ν

| n I | k Recall the deﬁnition of π : k from (70). I →

∗ I k Lemma X.26. Let (f1,...,fn) ( ), and ν be a weighted order on k[x1,...,xn] centered at ( ) , I [n]. Assume ∈ L A ⊆ n (1) conv( 1),..., conv( n) are dependent polytopes in R , and A A ∗ n (2) InA1,ν (f1),..., InAn,ν (fn) have a common zero a (k ) . ∗ I ∈ Then πI (a) (k ) is a non-isolated point of the set V (f1,...,fn) of common zeroes of f1,...,fn on n ∈ k . Note that both conditions (1) and (2) are necessary for the conclusion of lemma X.26 to hold. For example, f1 = 1+ x1,f2 = 1+ x1 + x2 satisfy condition (2), but not (1), with I = 1 , ν = (0, 1), ∗ { } = Supp(f ), j = 1, 2, and a = ( 1,c) for some arbitrary c k . In this case π (a) = ( 1, 0) is Aj j − ∈ I − an isolated point of V (f1,f2). On the other hand, f1 =1+ x1,f2 = x2, and j = Supp(fj ), j = 1, 2, satisfy, with the same I, ν and a, condition (1), but not (2), and π (a) = ( 1, 0)Ais again an isolated point I − of V (f1,f2).

PROOF OF LEMMA X.26. It is immediate to check that πI (a) is in V (f1,...,fn); we only have to show that it is non-isolated in there. Since := conv( ), j = 1,...,n, are dependent, it follows that Pj Aj there is J [n] such that p := dim( j∈J j ) < J . Let Π be the (unique) p-dimensional linear subspace n ⊆ P | | of R such that j∈J j is contained in a translate of Π. Let νj := ν(xj ), j = 1,...,n. We identify ν nP∗ P with the element of (R ) with coordinates (ν1,...,νn) with respect to the basis dual to the standard basis n P n of R . Let Π0 := α Π: ν, α = 0 and r := dim(Π0). Choose a basis α1,...,αr of Π0 Z . Let αi { ∈ h i } ∗ n αi ∩ ¯ ci := a , i =1,...,n. Let Y be the subvariety of (k ) determined by x ci, i =1,...,r, and Y be

n ¯ − n k the closure of Y in k . Then Y , and therefore Y , is an irreducible variety of codimension r in . Claim X.26.1. π (a) Y¯ . Moreover, if g is any Laurent polynomial in (x ,...,x ) such that Supp(g) I ∈ 1 n ⊆ Π0, then g restricts to a constant function on Y with value g(a). PROOF. The second assertion is obvious, so we prove the ﬁrst assertion. If ν is the trivial weighted or- n der, then πI is the identity and therefore πI (a)= a Y . Otherwise let C be the curve on k parametrized ν1 νn ∗ ∈n ¯ by t c(t) := (a1t ,...,ant ). Then C (k ) Y , so that C Y . Since πI (a) = c(0) C, the claim7→ is proved. ∩ ⊆ ⊆ ∈

Note that r equals either p or p 1. If r = p, then Π0 = Π and for each j J, fj is ν-homogeneous βj − n ∈ and is of the form fj = x gj for some βj Z and Laurent polynomial gj such that Supp(gj ) Π0. Claim X.26.1 implies that f 0 for each∈j J, so that π (a) Y¯ V (f : j J). It follows⊆ that at j|Y ≡ ∈ I ∈ ⊆ j ∈ 7. PROOF OF THE NON-DEGENERACY CONDITIONS 157 least one of the irreducible components of V (fj : j J) containing πI (a) has codimension smaller than n ∈ J in k . The lemma then follows due to theorem III.20. It remains to consider the case that r = p 1. | | n − Lemma B.43 implies that α1,...,αr can be extended to a basis α1,...,αn of Z such that α1,...,αr+1 is a basis of Π, and ν, αj 0 for each j = 1,...,n; this in particular implies that ν, αr+1 > 0. Let αi h i ≥ ∗ n h i yi := x , i = 1,...,n. Then the yi form a system of coordinates on (k ) and the projection onto ∗ n−r n (yr+1,...,yn) restricts to an isomorphism Y = (k ) . Pick βj := (βj,1,...,βj,n) Z such that ∼ ∈

n βj,i ¯ ∗ n−r n k xj = i=1 yi , j = 1,...,n. Then Y is the closure of the image of the map ψ : (k ) given by → Q n n ′ β1,i ′ βn,i ψ(yr+1,...,yn) := (c1 yi ,...,cn yi ), i=r+1 i=r+1 Y Y

′ r βj,i ¯ ′ ′ ∗ n−r n k where cj := i=1 ci , j =1,...,n. Let Y be the closure of the image of the map ψ : (k ) given by → Q n n ′ β1,i βn,i ψ (yr+1,...,yn) := ( yi ,..., yi ) i=r+1 i=r+1 Y Y ¯ ′ ¯ ′ ′ ′ Then Y is isomorphic to Y via the map ρ : (x1,...,xn) (c1x1,...,cnxn). Let βj := (βj,r+1,...,βj,n), ′ ′ ′ ′ 7→ n−r j =1,...,n, and := β0,...,βn , where β0 is the origin in Z . Let XB′ be the corresponding toric B { } ′ variety. In the notation of theorem VI.11 Y¯ is isomorphic to the affine open subset X ′ U ′ of X ′ . Since B β0 B ¯ ′ ′ ′ n−r ∩ ′ dim(Y )= n r, it follows that the convex hull of in R has dimension n r. Let νj := ν, αj , − ′ n−r ∗ P B ′ − ′ h i j = r +1,...,n, and ν (R ) be the element with coordinates (νr+1,...,νn). Then for each ′ ′ n∈ n ′ j =1,...,n, ν ,βj = i=r+1 βj,i ν, αi = ν, i=1 βj,iαi = νj 0. It follows that minP′ (ν )=0 ′ ′ h i ′ h i h ′ i ′ ≥ ′ (since ν ,β0 = 0). Since νr+1 > 0, there is a facet of containing Inν′ ( ) such that the first coordinateh ofi the inner normalP with respect to theP dual basisQ isP positive (this followsP e.g. from corollar- ′ ′ ′ ies V.21 and V.26 and proposition V.28). Let := , and VC′ be the corresponding torus invariant C B ∩Q ′ ′ ′ codimension one subvariety of X ′ (see theorem VI.11). Let Z := Y¯ V ′ , and Z := ρ(Z ) Y¯ . B ∩ C ⊆ Claim X.26.2. π (a) Z V (f : j J). I ∈ ⊆ j ∈ n−r ∗ ′ PROOF. Let η (R ) be the primitive inner normal to , and (ηr+1,...,ηn) be the coordinates ∈ n−r ∗ Q ∗ n−r of η in the dual basis of (R ) . For each b = (br+1,...,bn) (k ) , consider the rational curve on Y¯ ′ parametrized by ∈ ′ ′ η η β′ η′ β′ η′ t c (t) := ψ (b t r+1 ,...,b t n ) = (b 1 t 1 ,...,b n t n ) 7→ b r+1 n ′ ′ ′ ′ where ηj := η,βj for each j =1,...,n. Since 0= β0 Inν′ ( ) , it follows that minP′ (η)=0, h′ i ′∈ ′ P ⊂Q ′ so that each ηj is nonnegative,and it is zero if and only if βj . In particular, cb(0) is well defined, and is

n ∈Q ′ ∗ n−r ¯ ′ k an element in k . Moreover,theorem VI.11 implies that the set cb(0) : b ( ) is precisely Y OC′ ; ′ { ∈ } ∩ in particular it is a dense Zariski open subset of Z . Therefore, in order to prove Z V (fj : j J), it ′ ⊆ ∈ suffices to show that fj (ρ(cb(0))) = 0 for each j J. Fix j J. Write fj as fj = fj,0 + fj,1 + , where f are ν-homogeneous polynomials with ν(∈f ) <ν(f∈ ) < . We will show that ··· j,k j,0 j,1 ··· ′ (108) fj,k(ρ(cb(0))) = 0 for each k. Let m := min ( ). Note that ν(f ) m for each k. At first consider the case that j ν Aj j,k ≥ j ν(fj,k) = mj . This implies that k = 0, Inν ( j ) Supp(fj ) is nonempty, and fj,k = InAj ,ν (fj ). Since A ∩ ′ In (f )(a)=0, in this case claim X.26.1 implies that f is identically zero on ρ(c (t)), t k; in Aj ,ν j j,k b ∈ particular, (108) holds. Now fix k such that ν(fj,k) >mj . It suffices to show that (109) ord (f ρ c′ (t)) > 0 t j,k ◦ ◦ b Pick α0 Inν ( j ) and α Supp(fj,k). Then α α0 Π, and ν, α α0 = ν(fj,k) mj > 0. It follows ∈ A r+1 ∈ − ∈ h − i − that α α = m α with integers m such that m > 0. Since xαj c for j =1,...,r, and − 0 j=1 j j j r+1 |Y ≡ j xαr+1 y , it follows that |Y ≡ r+1P|Y ord (xα ρ c′ (t)) = ord (xα0 ρ c′ (t)) + m η t ◦ ◦ b t ◦ ◦ b r+1 r+1 158 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

′ n Since ηr+1 > 0 (due to our choice of ) and α0 Z ,(109) follows. It remains to prove that πI (a) Z. Q ∈ ≥0 ∈ Consider the rational curve C′ on Y¯ ′ parametrized by n n ′ ′ β ,i βn,i t ψ′(c tνr+1 ,...,c tνn ) = (( c 1 )tν1 ,..., ( c )tνn ) 7→ r+1 n i i i=r+1 i=r+1 Y Y ′ ′ ′ Since Inν′ ( ) is a face of , proposition VI.28 implies that the center of the branch of C at t = 0 is contained inPZ′. Note that ρ(QC′) is precisely the curve C from claim X.26.1. It follows from claim X.26.1 that πI (a) is the center of the branch of C at t =0, and therefore it is on Z, as required. Since dim(Z)= n r 1 >n J , the lemma follows from claim X.26.2 and theorem III.20. − − − | |

Corollary X.27. Let I D(U, ) and ν be a weighted order on k[xi : i I] such that ν is centered

∗ I′ ′ ∈ A \{∅} ∈ ∗ n k

k I I I I k at ( ) for some I I, and In (f1 k ),..., In (fn ) have a common zero a ( ) . Then A1,ν An,ν ′ ⊆ | | ∈

∗ I n k π ′ (a) (k ) is a non-isolated point of the zero-set V (f ,...,f ) of f ,...,f on . I ∈ 1 n 1 n PROOF. It is straightforward to check that πI′ (a) V (f1,...,fn). There is J I such that is J J I˜ ∈ ⊇ A ˜ ˜ ˜ J R -dependent, TA = J , and TA > I for each I such that I I ( J. Replacing the fj by fj k and applying lemma|IX.28| reduces| | the| corollary| | | to the case that I = [n⊆]. Then it follows from lemma X.26| . 7.3. Proof of theorem X.22. We divide the proof of theorem X.22 in three parts: 7.3.1. Proof of the implication (2) (1) from theorem X.22. We proceed as in the proof of proposi- ⇒ iso iso tion VII.16. Pick (f1,...,fn), (g1,...,gn) ( ) such that [g1,...,gn]U > [f1,...,fn]U . We will show that f ,...,f are (U, )-degenerate. Deﬁne∈ L Ah := (1 t)f + tg , where t is a new indeterminate. 1 n A j − j j The set of isolated zeroes of g1,...,gn on U is nonempty. Theorem IV.20 implies that we can ﬁnd an

irreducible curve C contained in the set of zeroes of h ,...,h on U k such that 1 n × (i) C intersects Z 1 , ×{ } (ii) for generic ǫ k, the set Cǫ := C (U ǫ ) is nonempty, and each point of Cǫ is an isolated zero of h ∈,...,h on U; and∩ ×{ } 1|t=ǫ n|t=ǫ (iii) (1) either there is (z, 0) C such that z is a non-isolated zero of f1,...,fn on U, (2) or C has a “point at∈ infinity with respect to U” at t = 0, i.e. if C¯ is the closure of C in n ¯ P k, then there is (z, 0) C such that z U. × ∈ 6∈ ¯ n ˜ Claim X.28. Let B be a branch of C P k. Let I := IB [n] (where IB [n + 1] is defined as in ⊂ × ˜ ∩ ⊆ definition IX.15) and ν˜ be the restriction of ν to k[x : i I]. B i ∈

∗ I˜ k (1) C (k ) ǫ for generic ǫ . ǫ ⊆ ×{ } ∈ (2) I˜ I (U, ). ∈ A (3) Assume the center of B is (z, 0) where z Pn. Then ∈ ∗ n (a) In I˜ (fj I˜), j =1,...,n, have a common zero on (k ) . A ,ν˜ k j | n n ˜ (b) If z P k , then I = and ν˜ is centered at inﬁnity. In particular, f1,...,fn violate ∈ \ 6 ∅ condition (b) of (U, )-non-degeneracy with I = I˜ and ν =ν ˜. A ∗ J ˜ (c) Otherwise let J be the (unique) subset of [n] such that z (k ) . Then either I = J = , ˜ ∗ J ∈ ∅ or I = and ν˜ is centered at (k ) . In particular, if J E (U) E ( ) D(U, ), then 6 ∅ ∈ ∪ A ∪ A f ,...,f violate property (b) of (U, )-non-degeneracy with I = I˜, I′ = J and ν =ν ˜. 1 n A PROOF. Assertion (1) follows from the deﬁnition of I . Since (h ,...,h ) ( ) for all ǫ, B 1|t=ǫ n|t=ǫ ∈ L A assertion (1), property (ii) and lemma X.1 imply that I˜ E (U) E ( ), and then proposition X.21 implies 6∈ ∪ A that I˜ I (U, ). This proves assertion (2). Now assume we are in the situation of assertion (3). Fix j, ∈ A

. Since I , it follows that 1 j n ν (t k ) > 0 ≤ ≤ B | B I˜ I˜ I if Inν˜(fj )=Inν (hj k B ) Supp(fj ) Inν˜( ) = , k B j In I˜ (fj I˜)= | | ∩ A 6 ∅ A ,ν˜ k j | (0 otherwise. n n Assertion (3a) now follows from lemma IX.17. If z P k , then there is at least one j such that 1/xj ∈ \ is a regular function near z which vanishes at z. This j has to be in I˜ and ν˜(xj ) has to be negative, which proves the ﬁrst statement of part (3b). The second statement then follows from assertions (2)and(3a). The 7. PROOF OF THE NON-DEGENERACY CONDITIONS 159

ﬁrst statement of part (3c) is obvious. If J E (U) E ( ) D(U, ), then assertion (2) and the ﬁrst ∈ ∪ A ∪ A statement of part (3c) imply that I˜ = , and then the second statement follows from assertions (2) and (3a). 6 ∅

Now we resume the proof of the implication (2) (1) from theorem X.22. At first assume (iii.1) ⇒ holds. Pick a branch B of C centered at (z, 0) and let I,˜ J, ν˜ be as in part (3c) of claim X.28. Part (3c) of claim X.28 implies that f1,...,fn are (U, )-degenerate if J E (U) E ( ) D(U, ). So assume ′ A ′∈ ∪ A ∪ ′A J I (U, ). Pick an irreducible curve C on U such that z C V (f1,...,fn). Let J be the smallest ∈ A ′ J′ ′ ∈ ⊆ ′ subset of [n] such that C k . Since J J and J E (U) E ( ), it follows that J E (U) E ( ) ′ ⊆ ⊇ 6∈ ∪ A 6∈ ∪ A as well. If J I (U, ), then lemma IX.17 implies that f1,...,fn violate condition (b) of (U, )-non- ∈ A′ ′ A ′ degeneracy with I = J and some weighted order ν on k[xj : j J ] centered at infinity (take a branch B ′ ∈ ′ of C centered at infinity, and set ν = νB′ ). On the other hand, if J D(U, ), then f1,...,fn violate condition (c) of (U, )-non-degeneracy with I = J ′ and I′ = J (take∈ a branchAB′ of C′ centered at z and A take ν := νB′ ). This completes the proof of (2) (1) in the case that (iii.1) holds. Now assume we are in n ⇒¯ n n case (iii.2). Pick z P U such that (z, 0) C. If z P k , then part (3b) of claim X.28 implies ∈ \ ∈ n∈ \ that f1,...,fn are (U, )-degenerate. So assume z k U. Define J as in part (3c) of claim X.28. If J E (U) E ( ) DA(U, ), then part (3c) of claim∈X.28\implies that f ,...,f are (U, )-degenerate. ∈ ∪ A ∪ A 1 n A So assume J I (U, ). But then f1,...,fn violate condition (a) of (U, )-non-degeneracy. This completes the proof∈ of theA implication (2) (1). A ⇒ iso 7.3.2. Proof of the implication (1) (2) from theorem X.22. Assume [ 1,..., n]U > 0 and pick ⇒ A isoA iso an (U, )-degenerate system f1,...,fn ( ). We will show that [ 1,..., n]U > [f1,...,fn]U . RecallA the definition of I ∗(U, ) from (107∈ L). A A A A Claim X.29. There is I I ∗(U, ) such that one of the following holds: ∈ A

k I I

(1) Either I is nonempty, and there is a weighted order ν on [x : i I] such that In (f k ), i Aj ,ν j ∗ n ∈ | j =1,...,n, have a common zero a (k ) , and one of the following holds: (a) ν is centered at inﬁnity, ∈ ∗ I′ ′ (b) or ν is centered at (k ) for some I I, and πI′ (a) U, ∗ I′ ′ ⊆ 6∈ (c) or ν is centered at (k ) for some I I and πI′ (a) is a non-isolated zero of f1,...,fn. ⊆ I (2) Or there is an isolated point a of V (f1,...,fn) k U which is not isolated in V (f1,...,fn) n ∩ ∩ ⊂

k .

′ PROOF. If f1,...,fn violate property (a) of (U, )-non-degeneracy, then there is a common zero a ∗ I′ ′ A

of f I′ ,...,f I′ on (k ) U for some I I (U, ). Lemma IX.28 then implies that the claim k 1 k n holds| with case (1b| ) holds. If property\ (b) of (U, ∈)-non-degeneracyA fails with I I ∗(U, ), then either A ∈ A

k I I the claim holds with case (1a), or there is a weighted order ν on [xi : i I] such that InA ,ν (fj k ) ∈ j |

∗ n ∗ I′ ′ k have a common zero a (k ) , and ν is centered at ( ) for some I E (U) E ( ) D(U, ). It is ∈ ∈ ∪ A n∪ ′ A straightforward to check that πI′ (a) is in the set V of common zeroes of f1,...,fn on k . If I E (U),

∗ I′ ∗ I′ ′ ∈ k then we are in case (1b), since πI′ (a) (k ) and ( ) U = . If I E ( ) D(U, ), then πI′ (a) has to be a non-isolated point of V due∈ to lemma X.1 and∩ proposition∅ X.21∈ , whichA ∪ is caseA (1c). Due to proposition X.23 the only case left to consider is that of f1,...,fn violating property (c) of (U, )-non-

A ∗ J′ k degeneracy. Then there is J D(U, ) and a weighted order η on k[xj : j J] centered at ( ) ′ ∈ A ∈ ∗ n

I J J J J k k for some J (U, ) such that InA ,η(f1 k ),..., InA ,η(fn ) have a common zero b on ( ) . ∈ A 1 | n | Corollary X.27 implies that πJ′ (b) is a non-isolatedpointof V (f1,...,fn). Pick the smallest subset I of [n] ′ I ′ E E E E containing J such that TA = I . Since J (U) ( ), it follows that I (U) ( ), and since |′ | | | 6∈ ∪ A 6∈ ∪ A is not hereditarily RJ -dependent, it follows that is RI -independent; in particular, I I ∗(U, ). If A I A ′ ∈ A πJ′ (b) is an isolated point of V (f1,...,fn) k (which is e.g. the case if I = J = ), then case (2) holds ∩ ∅ I with a = πJ′ (b). Otherwise picking a branch at πJ′ (b) of a curve contained in V (f1,...,fn) k and applying lemmas IX.17 and IX.28 shows that case (1c) holds. ∩

Claim X.30. Assume case (2) of claim X.29 holds. Then there is (g1,...,gn) ( ) such that gj k = n ∈ L A |

I k f k for each j, and a is an isolated point of V (g ,...,g ) . j | 1 n ⊂ 160 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

˜ I E I˜ I ˜ PROOF. Fix I ) I. Since TA = I and I ( ), it follows that TA TA contains at least I I

| | | | 6∈ Ã ∗ I \ | | − | | k elements. For each weighted order ν on k[xi : i I] centered at ( ) , choosing generically coefficients I˜ I ∈ I˜ I ∗ n of fj , j TA TA, it can be ensured that InAI˜,ν (fj ), j TA TA, have no common zero a˜ on (k ) such ∈ \ j ∈ \ that πI (ã)= a. The claim now follows due to lemma IX.17.

At ﬁrst consider case (2) of claim X.29. Pick (g1,...,gn) as in claim X.30. Apply theorem IV.20 to

X = U and hj = (1 t)fj + tgj , j =1,...,n. It is straightforward to see that in this case a k is an irreducible component− of the curve C from theorem IV.20, so that { }× iso iso iso

n n

k [f ,...,f ] [f ,...,f ]k < [h ,...,h ] 1 n U ≤ 1 n 1|t=ǫ n|t=ǫ iso iso iso for generic ǫ k. It follows that [f ,...,f ] < [ ,..., ] n = [ ,..., ] , as required. Now 1 n U 1 n k 1 n U assume there∈ are I, ν and a as in case (1) of claim X.29A . WeA may assumeA withoutA loss of generality that I = T I = 1,...,k for some k, 1 k n. A { } ≤ ≤ Claim X.31. There is (g ,...,g ) ( ) such that 1 n ∈ L A I I (1) InA ,ν (gj k )(a) =0 for each j =1,...,k. j | 6 ∗ I (2) there is a common zero b of g1,...,gk on (k ) U such that n∩ (a) b is an isolated point of V (g1,...,gn) k , and (b) f (b) =0 for each j =1,...,k. ⊂ j 6 ′ PROOF. Let ( ) be the collection of all (g1,...,gn) ( ) such that L A J ∈ L A

J J k (i) g1 k ,...,gn are -non-degenerate (in the sense of deﬁnition VII.6) for all J I, and | | A ∗ I ⊃ (ii) there is no common zero of g ,...,g on (k ) U. 1 k \ Since I E (U) E ( ), theorem VII.12 and lemma IX.26 imply that ′( ) contains a nonempty Zariski 6∈ ∪ A ′ ′ L A I I open subset of ( ). It follows that ( ) := (g1,...,gn) ( ) : InA ,ν (gj k )(a) = 0 for each L A La A { ∈ L A j | 6 j = 1,...,k also contains a nonempty Zariski open subset of ( ). Since is RI -independent, it } I L A A follows that the (k-dimensional) mixed volume of conv( j ) R , j = 1,...,k, is nonzero. Due to (ii), A ∩ ′ the arguments of claim VII.17 then imply that we can ﬁnd (g1,...,gn) a( ) and a common zero b of ∗ I ∈ L A g ,...,g on (k ) U such that f (b) =0 for each j =1,...,k. Property (i) together with lemma IX.17 1 n ∩ j 6 then imply that b must be an isolated zero of g1,...,gn on U. ′ Now we follow the process from section VII.4.2.2. Fix integers νj > νj := ν(xj ). Let C be the

n n

k k rational curve on k parametrized by c(t) := (c (t),...,c (t)) : given by 1 n → ′ νj ν ajt + (bj aj )t j if 1 j k, cj (t) := − ≤ ≤ (0 otherwise.

Let mj := minAI (ν), 1 j k. Deﬁne j ≤ ≤

−mj −mj t fj(c(t))gj t gj (c(t))fj if 1 j k, hj := − ≤ ≤ (1 t)f + tg otherwise. ( − j j Assertion (1) of claim X.31 implies that each hj t=0 is a nonzero constant times fj, and assertion (2) of claim X.31 implies that each h is a nonzero| constant times g and c(1) = b is an isolated zero of j |t=1 j g1,...,gn on U. The assumptions of case (1) of claim X.29 implies that the center of C at t =0 is either ′ out of U, or it is a non-isolated zero of f1,...,fn. Since each hj vanishes on the curve C := (c(t),t):

n+1 iso { iso k t k , assertion (5) of theorem IV.20 implies that [f1,...,fn] < [h1 t=ǫ,...,hn t=ǫ] ∗ n U (k ) ∈ } ⊂ iso iso | | for generic ǫ k. It follows that [f ,...,f ] < [ ,..., ] , as required. ∈ 1 n U A1 An U 7.3.3. Proof that (U, ) is a nonempty Zariski open subset of ( ). As in section VII.4.3 we call N A nL∗ A = ( 1,..., n) a face of and write , if there is ν (R ) such that j = Inν ( j ) for each B B B A B A ∈ B A j; in that case we also write := Inν ( ) and we say that is centered at inﬁnity (respectively, centered

∗ I B A B ∗ I k at (k ) for some I [n]) if ν is centered at inﬁnity (respectively, centered at ( ) ). Moreover, if g ⊆ is a polynomial supported at j , we write gBj for InAj ,ν (g). Consider the systems of polynomials that violate either property (b) or propertyA (c) of (U, )-non-degeneracy: pick I [n] and a weighted order A ⊆ 8. WEIGHTED BEZOUTTHEOREM:GENERALVERSION´ 161

I I ν on k[xi : i I]. Let := Inν ( ) and B be the set of all (g1,...,gn) ( ) such that there is a ∈ B A D ∗ n ∈ L A I

I k common root of (gj k )Bj , j = 1,...,n, on ( ) . If (f1,...,fn) ( ) is in the closure of B, then | I ′ ∈ L A D claim VII.19 implies that (f ,...,f ) ′ for some . Note that 1 n ∈ DB B B (i) If is centered at inﬁnity, then ′ is also centered at inﬁnity.

B ∗ I′ B ′ ′ ∗ I′′ k (ii) If is centered at (k ) for some I E (U) E ( ), then is also centered at ( ) for someB I′′ E (U) E ( ). ∈ ∪ A B ∈ ∪ ∗ AI′ ′ (iii) If is centered at (k ) for some I [n], then for each j =1,...,n, B ⊆ I′ fj,Bj if j R = , f I′ = B ∩ 6 ∅ j k | (0 otherwise. It follows that ′ ′ ∗ I′′ ′′ (1) if I D(U, ) and is centered at (k ) for some I I (U, ), then (f1,...,fn) violate∈ propertyA (c) ofB(U, )-non-degeneracy with I, I′ replaced∈ respectivelyA by I′, I′′; ′ ′ A (2) if I I (U, ) and is centered at inﬁnity, then (f1,...,fn) violate property (b) of (U, ∈)-non-degeneracyA B with I, I′ replaced respectively by I′, I′′; ′A ′ ∗ I′′ ′′ (3) if I I (U, ) and is centered at (k ) for some I D(U, ), then (f1,...,fn) violate∈ propertyA (b) ofB(U, )-non-degeneracy with I, I′ replaced∈ respectivelyA by I′, I′′; A It follows from these observations that the set of systems which violate at least one of the properties (b) and (c) of (U, )-non-degeneracy is Zariski closed in ( ). Now we tackle property (a). Pick I I A I L A ∈ (U, ) and let U be the set of all (g1,...,gn) ( ) such that there is a common root of g1,...,gn ∗AI D ∈ L A I on (k ) U. If (f ,...,f ) ( ) is in the closure of , then the arguments from the proof of \ 1 n ∈ L A DU

k I I claim VII.19 imply that there is a weighted order ν on [x : i I] and a common zero a of In (f k ), i Aj ,ν j ∗ n ∈ | j =1,...,n, on (k ) , such that

(iv) either ν is centered at inﬁnity, in which case f1,...,fn violate property (b) of (U, )-non- degeneracy, A

∗ I′ ′ ∗ I′ k (v) or ν is centered at (k ) for some I I, and πI′ (a) ( ) U; in this case f1,...,fn violate property (a) (with I replaced by I′) of⊆(U, )-non-degeneracy∈ \ if I′ I (U, ), and property (b) of (U, )-non-degeneracy if I′ D(U, )A. ∈ A A ∈ A It follows that the set of (U, )-degenerate systems is Zariski closed in ( ), as required. A L A

8. Weighted Bezout´ theorem: general version

8.1. Weighted Bezout´ theorem II: all ω(fj ) ≥ 0. For weights ω = (ω1,...,ωn) to be applicable in weighted B´ezout theorem (theorem VIII.2), each ωi has to be positive. In this section we replace this condition by a weaker one - that ω(fj ) has to be nonnegativefor each j. This opens up a new possibility: if some ωi is nonpositive, then the set of polynomials f with ω(f) bounded above by a given integer will be

an inﬁnite dimensional vector space over k, and the number of (isolated) solutions of n such polynomials can be arbitrarily large. Therefore to estimate number of solutions one has to bound the degree in each xi such that ωi 0. It is natural then to consider, given an integer d, and a nonnegative integer mi for each i such that ω ≤ 0, the set of polynomials supported at the polytope i ≤ n (ω, d, ~m) := α R : ω, α d, αi 0, i =1,...,n, αi mi, i I0 I− P { ∈ h i≤ ≥ ≤ ∈ ∪ } where I := i : ω < 0 and I := i : ω =0 (see ﬁg. 7). The reason for our restriction to the case of − { i } 0 { i } nonnegative ω(fj) is the following observation:

Proposition X.32. Let d1,...,dk Z and mj,i Z≥0, j = 1,...,k, i I0 I−. If dj 0 for each j, then ∈ ∈ ∈ ∪ ≥ (110) (ω, d , ~m )= (ω, d , ~m ) P j j P j j j j j X X X PROOF. (ω, dj , ~mj) is the product of an (n I0 )-dimensional polytope of the same form with the I -dimensionalP box formed by the product over all−i | I| of the closed intervals from 0 to m on x -axis. 0 ∈ 0 j,i i 162 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

1 1 1

0.5 0.5 0.5

0.5 1 0.5 1 0.5 0.5 1 2 2 1 1

(A) ω = (1, 1, 1), (B) ω = (1, −1, 0), (C) ω = (1, −1, −1), d = 1 d = 1, m2 = m3 = 1 d = 1, m2 = m3 = 1

FIGURE 7. P (ω, d, ~m) for different ω, d, ~m

It is then straightforward to see that to prove (110) it sufﬁces to prove it under the additional condition that I = . So assume I = . Let m := max m and deﬁne 0 ∅ 0 ∅ i,j j,i x if i I x′ := i 6∈ − i m x if i I ( − i ∈ − ω′ := ω , i =1, . . . , n. i | i| ′ ′ It is straightforward to check that in (x1,...,xn)-coordinates, in the notation of exercise V.44, up to a reordering of the xi if necessary, (111) (ω, d , ~m )= (~ω′, d′ , ~m ) + (m′ ,...,m′ ) P j j Q j j j,1 j,n where d′ := d + ω m , and j j | i| i iX∈I− 0 if i I m′ := 6∈ − j,i m m if i I ( − j,i ∈ − It then follows from exercise V.44 that (ω, d , ~m )= (~ω′, d′ , ~m )+ (m′ ,...,m′ )= (ω, d , ~m ) P j j Q j j j,1 j,n P j j j j j j j j X X X X X X as required.

Remark X.33. (110) may fail to hold if dj < 0 for some j. Indeed, it follows from exercise V.44 and (111) that with ω = ( 1, 1, 1), m1,1 = m1,2 = 1 and m2,1 = m2,2 = 3, (ω, 1, ~m1)+ (ω, 3, ~m1) ( (ω, 2, ~m + −~m ).− P P − P − 1 2 We will now compute the mixed volume of (ω, dj, ~mj ), j = 1,...,n. Exercise V.45 and (111) imply that P

mi i∈I0 |I−|−|I| n−|I0| (112) Vol( (ω, d, ~m)) = ( 1) (d + ωi mi) P (n I0 )! i6∈I ωi − | | − |Q| 0 | | I⊆I− i∈I d+P X|ω |m >0 X Q i∈I i i In particular, if d 0, then ≥ mi i∈I0 |I−|−|I| n−|I0| Vol( (ω, d, ~m)) = ( 1) (d + ωi mi) P (n I0 )! i6∈I ωi − | | Q 0 I⊆I− i∈I − | | | | X X Therefore proposition X.32 implies that forQ each d ,...,d , λ ,...,λ 0, 1 n 1 n ≥ n n n ( λj mj,i) i∈I0 j=1 |I−|−|I| n−|I0| Vol( λj (ω, dj , ~mj)) = ( 1) ( λj (dj + ωi mj,i)) P (n I0 )! ωi − | | j=1 Q P i6∈I0 I⊆I j=1 i∈I X − | | | | X− X X Q 8. WEIGHTED BEZOUTTHEOREM:GENERALVERSION´ 163

The mixed volume of (ω, d1, ~m1),..., (ω, dn, ~mn) is the coefﬁcient of λ1 λn in the right hand side of the above expression.P For each JP [n] such that J = I , the coefﬁcient··· of λ in 0 0 0 j∈J0 j n ⊆ | | | | ( λj mj,i) is the permanent of the I0 I0 matrix i∈I0 j=1 | | × | | Q

Q P mj1,i1 mj1,ik . ··· . (113) D := . . I0,J0  . .  m m  jk ,i1 ··· jk,ik    where k := I0 = J0 and i1,...,ik (respectively, j1,...,jk) are elements of I0 (respectively, J0). (Note that perm(D| | )|does| not depend on the ordering of the elements of I or J . If I = J = , then I0,J0 0 0 0 0 ∅ DI0,J0 is the empty matrix, and its permanent is by convention 1.) On the other hand, the coefﬁcient of λ in ( n λ (d + ω m ))n−k is (n k)! (d + ω m ). Combining all j6∈J0 j j=1 j j i∈I i j,i j6∈J0 j i∈I i j,i these together yields: | | − | | Q P P Q P Proposition X.34. Let d1,...,dn Z and mj,i Z≥0, j = 1,...,k, i I0 I−. If dj 0 for each j, then the mixed volume of (ω, d , ~m∈ ),..., (ω,∈ d , ~m ) is ∈ ∪ ≥ P 1 1 P n n j∈[n]\J (dj + i∈I ωi mj,i) perm(D ) ( 1)|I−|−|I| 0 − | | I0,J0 ω − i∈[n]\I0 i J0⊆[n] I⊆I− Q P | | |J X|=|I | X 0 0 Q The following result describes the faces of (ω, d, ~m). Its proof is left as an exercise. Let η (Rn)∗ P n ∈ with coordinates (η1,...,ηn) with respect to the basis dual to the standard basis of R . Proposition X.35. Deﬁne M := sup η /ω : ω > 0 . Assume d 0. Then { i i i } ≥ (1) If M 0 (which is the case if in particular ωi 0 for each i), then ldη( (ω, d, ~m)) is the set of all (α≤,...,α ) such that ≤ P 1 n ∈P 0 if ηi < 0, αi = (mi if ηi > 0. (2) If M > 0, then ld ( (ω, d, ~m)) is the set of all (α ,...,α ) such that α ω = d and η P 1 n ∈P i i i ηi ηi P 0 if (ωi =0, ηi < 0) or (ωi > 0, M), ωi ωi αi =   ηi mi if (ωi =0, ηi > 0) or (ωi < 0,

n k any weighted degree ω on k[x1,...,xn], the number N of isolated solutions of f1,...,fn on satisﬁes: (114) max ω(f ), 0 + ω deg (f ) j∈[n]\J0 j i∈I− i xi j N perm(D ) ( 1)|I−|−|I| { } | | I0,J0 ω ≤ − Q i∈[n]\I0 Pi J0⊆[n] I⊆I− | | |J X|=|I | X 0 0 Q 164 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

where DI0,J0 is deﬁned as in (113) with mj,i := degxi (fj ). More generally, if d1,...,dn and mj,i are nonnegative integers such that each d ω(f ) and m deg (f ), then j ≥ j j,i ≥ xi j d + ω m j∈[n]\J0 j i∈I− i j,i (115) N perm(D ) ( 1)|I−|−|I| | | I0,J0 ω ≤ − Q i∈[n]\PI0 i J0⊆[n] I⊆I− | | |J X|=|I | X 0 0 Q If the right hand side of (115) is nonzero, i.e. (ω, d , ~m ),..., (ω, d , ~m ) are independent (which is P 1 1 P n n the case e.g. if each dj and each mj,i is positive), then (115) holds with an equality if and only if both of the following are true:

(1) For each nonempty subset I of I0 I−, ld ~mj ,I (fj ), j = 1,...,n, have no common zero on n ∪ k V (x ). \ i∈I i (2) For each (possibly empty) subset I of I0 I−, ldω,dj , ~mj ,I (fj ), j = 1,...,n, have no common S n ∪ zero on k V (x )) ( V (x )) (where I := i : ω > 0 ). \ ∪i∈I i ∩i∈I+ i + { i } PROOF. Inequalities (114) and (S115) follow directly from theorem X.4 and proposition X.34. To ﬁnd the non-degeneracy conditions we apply theorem X.22 and proposition X.23 with := ( (ω, d1, ~m1) n n n A P ∩ Z ,..., (ω, dn, ~mn) Z ) and U := k . It is straightforward to check that E ( ) = E (U) = , and if P(ω, d , ~m ), j ∩= 1,...,n, are independent then D(U, ) = and I ∗(U, A) = [n]. Therefore∅ P j j A ∅ A theorem X.22 and proposition X.23 imply that if (ω, dj , ~mj ), j =1,...,n, are independent (which due to theorem VII.30 and proposition X.34 is equivalentP to the right hand side of (115) being nonzero), then (115) holds with an equality if and only if the following holds:

for each weighted order ν centered at inﬁnity on k[x1,...,xn], there is (116) ∗ n no common zero of InA1,ν (f1),..., InAn,ν (fn) on (k ) .

Pick a weighted order ν centered at inﬁnity on k[x ,...,x ]. Now apply proposition X.35 to η := ν. If 1 n − M := sup ηi/ωi : ωi > 0 0, then I := i I0 I− : ηi > 0 is nonempty and proposition X.35 implies that{ } ≤ { ∈ ∪ }

In (f ) = ld (f ) ′ ′ Aj ,ν j ~mj ,I j |{xi′ =0:i ∈I } where I′ := i [n]: η < 0 . Therefore (116) is equivalent to the condition that ld (f ), j = { ∈ i } ~mj ,I j n ′ 1,...,n, have no common zero on k ′ V (x ) ′ ′ V (x ′ ). Since it is possible for I to \ i∈[n]\I i ∩ i ∈I i be any subset of [n] I, taking into account all such possibilities leads to condition (1). Now consider \ S T the case that M > 0. In this case deﬁne I := i I0 : ηi > 0 i I− : ηi/ωi < M and I′ := i I : η < 0 i I : η /ω >M { i∈ I : η /ω

THEOREM X.37 (Weighted B´ezout theorem III). Let dj Z and mj,i Z≥0 be such that dj ω(fj ) ∈ ∈ ≥ for each j and each mj,i deg (fj ). Given I I−, write (I) := j : mj,iωi > dj . Let ≥ xi ⊆ B { i∈I−\I } G := I I− : (I˜) I˜ for each I˜ I . Then the number N of isolated solutions of f1,...,fn on n { ⊆ |B | ≤ | | ⊆ } P k is I′ I′ (117) N MV ( ′ ,..., ′ ) [π ( ),...,π ( )] n−k j1 R jn−k R I j1 I jk 0 ≤ G P ∩ P ∩ × P P I∈ ,|BX(I)|=|I| where ′ := [n] I, • I are short\ for (ω, d , ~m ), • Pj P j j 9. WEIGHTED MULTI-HOMOGENEOUS BEZOUTTHEOREM:GENERALVERSION´ 165

j ,...,j (respectively, j′ ,...,j′ ) are elements of (I) (respectively, 1,...,n (I)), and • 1 k 1 n−k B { }\B MVn−k( ,..., ) is the (n k)-dimensional mixed volume. • · · − n The bound in (117) holds with an equality if and only if f1,...,fn are (k , )-non-degenerate, where n n A := ( 1 Z ,..., n Z ). A P ∩ P ∩ PROOF. Follows immediately from theorems X.4 and X.22 and proposition X.23.

9. Weighted multi-homogeneous Bezout´ theorem: general version In this section we generalize the weighted multi-homogeneous version of B´ezout’s theorem (theorem VIII.8) by replacing the weighted degrees with positive weights by weighted degrees from theorem X.36. As in the setting of theorem VIII.8, let I := (I1,...,Is) be an ordered partition of [n] :=

1,...,n , and for each j =1,...,s, let ωj be a weighted degree on k[xk : k Ij ] with weights ωj,k for {x , k I}. Let I (respectively, I ) be the set of all k I such that ω >∈0 (respectively, ω < 0). k ∈ j j,+ j,− ∈ j j,k j,k Given nonnegative integers di,j ωj (fi) and mi,j,k degxk (fi) for each k Ij,0 Ij,−, we consider the polytope ≥ ≥ ∈ ∪ s := (ω , d , ~m ) Pi P j i,j i,j j=1 Y Let nj := Ij , j = 1,...,s. The mixed volume of 1,..., n is the coefﬁcient of λ1 λn in the polynomial | | P P ··· n n s s n Vol ( λ ) = Vol ( λ (ω , d , ~m )) = Vol ( ( λ (ω , d , ~m ))) n iPi n i P j i,j i,j n iP j i,j i,j i=1 i=1 j=1 j=1 i=1 X X Y Y X s n n s n n = Vol ( (ω , λ d , λ ~m )) = Vol ( (ω , λ d , λ ~m )) n P j i i,j i i,j nj P j i i,j i i,j j=1 i=1 i=1 j=1 i=1 i=1 Y X X Y X X where the third equality follows from proposition X.32. After a reﬁnement of I if necessary, we may, and will, assume that ω =0 for each j, k, i.e. I = I I for each j. Then(112) implies that j,k 6 j j,+ ∪ j,− n s n 1 |Ij,−|−|I| nj Voln( λi i)= ( 1) ( λi(di,j + ωj,k mi,j,k)) P nj ! ωj,k − | |  i=1 j=1 k∈Ij i=1 X Y | | I⊆XIj,− X Xk∈I Let I := I . For each I QI , write  − j j,− ⊆ − (118) S d := d + ω m I,i,j i,j | j,k| i,j,k k∈XI∩Ij Let Ω := ω ,...,ω , and D(Ω, d,~ ~m, I) be the following n n matrix: { 1 s} × n1 times ··· ns times d d d d I,1,1 ··· I,1,1 ······ I,1,s ··· I,1,s D(Ω, d,~ ~m, I) := z . }| . { z }| { z . }| . {  . . . .  dI,n,1 dI,n,1 dI,n,s dI,n,s  ··· ······ ···  The preceding discussion together with theorem X.4 or lemma VIII.4 imply that  ~ iso iso |I−|−|I| perm(D(Ω, d, ~m, I))

(119) [f ,...,f ] n [ ,..., ] n = MV( ,..., )= ( 1) k 1 n k 1 n 1 n ≤ P P P P − ( j nj !)( j,k ωj,k) IX⊆I− It is straightforward to check that the bound (115) from weighted Bézout theoremQ II correspondsQ to the special case of (119) in which nj = 1 for all but possibly one j 1,...,s . We will now see that the criterion for attainment of this bound is an amalgam of the non-degeneracy∈ { criteria} of weighted multi- homogeneous Bézout theorem (theorem VIII.8) and weighted Bézout theorem II (theorem X.36). Given n I I−, J [s], and l [n], let ~ be the set of all α = (α1,...,αn) Z such that ⊆ ⊆ ∈ LΩ,d, ~m,I,J,i ∈ ≥0 α = m for each j [s] and k I I, and • k i,j,k ∈ ∈ j,− ∩ 166 X.NUMBEROFSOLUTIONSON(OPENSUBSETSOF)THEAFFINESPACE

ωj,kαk = di,j for each j J. • k∈Ij ∈ α Given g = c x k[x ,...,x ] supported at , define P α α ∈ 1 n Pi α P ldΩ,d,~ ~m,I,J,i(g) := cαx α∈L ΩX,d,~ ~m,I,J,i We are now ready to prove version II of the weighted multi-homogeneous Bézout theorem. But at first we recall the assumptions: (a) ω =0 for each j =1,...,s, and k I . j,k 6 ∈ j (b) di,j max ωj(fi), 0 for each i =1,...,n, and j =1,...,s. (c) m ≥ deg{ (f ) for} each i =1,...,n, and j =1,...,s, and k I . i,j,k ≥ xk i ∈ j,− THEOREM X.38 (Weighted multi-homogeneous Bézout theorem II). Under the above assumptions n the number of isolated solutions of polynomials f1,...,fn on k is bounded by (119). This bound is exact if and only if the following holds: for each pair I,J such that I I−, J [s], and at least ⊆ ⊆ n one of I and J is nonempty, there is no common zero of ld (f ),..., ld (f ) on k Ω,d,~ ~m,I,J,1 1 Ω,d,~ ~m,I,J,n n \ V (x )) ( V (x )) . ∪i∈I i ∪j∈J ∩k∈Ij,+ k PROOF. ThisS follows from theorems X.4 and X.22 and proposition X.23 via arguments similar to those in the proof of theorem X.36.

10. Open problems

10.1. Systems with isolated zeroes on a given coordinatesubspace. Given ﬁnite subsets 1,..., n

n I n A A k of Z≥0 and a coordinate subspace k of , it is straightforward to identify if for generic fj supported at I , there is any common zero of f ,...,f on k , and in case there are such points, if they are isolated Aj 1 n in V (f1,...,fn) or not. This is the content of the next result, which is straightforward to prove from theorem VII.30 and lemma X.1.

Proposition X.39 (Zeroes of generic systems). Let := ( 1,..., n). For all f = (f1,...,fn) ( ),

n ∗ I A A A ∗ I ∈ L A k write V (f) := V (f ,...,f ) k and (V ) (f) := V (f ,...,f ) ( ) . 1 n ⊂ 1 n ∩ (1) The following are equivalent: (a) (V ∗)I (f)= for generic f ( ). ∅ ∈ L A (b) is RI -dependent. (2) The followingA are equivalent: (a) (V ∗)I (f) = and all points of (V ∗)I (f) are isolated in V (f) for generic f ( ). 6 ∅ ∈ L A (b) is RI -independent and I E ( ). (3) The followingA are equivalent: 6∈ A (a) (V ∗)I (f) = and all points of (V ∗)I (f) are non-isolated in V (f) for generic f ( ). 6 ∅ ∈ L A (b) is RI -independent and I E ( ). A ∈ A Problem X.40 (Existence of systems with given support and isolated zeroes on given coordinate subspace). Characterize those I [n] for which there are f ,...,f such that Supp(f ) = , j = 1,...,n, ⊆ 1 n j Aj (V ∗)I (f ,...,f ) = and all points (or some points) of (V ∗)I (f ,...,f ) are isolated in V (f ,...,f ). 1 n 6 ∅ 1 n 1 n If I is as in problem X.40, then lemma X.1 implies that I E ( ); • theorems VII.7 and VII.306∈implyA that • – either I is RI -independent and T I = I (i.e. I I ∗(U, )), | A| | | ∈ A – or T I > I (in particular, is RI -dependent); | A| | | A proposition X.21 implies that is not hereditarily RI -dependent. • A In the context of these observations problem X.40 boils down to the following problem. ′ E I I Problem X.40 . If I ( ) and is not hereditarily R -dependent and TA > I , does there exist 6∈ A A ∗ I | | | | f1,...,fn such that Supp(fj) = j for each j, (V ) (f1,...fn) = and all points (or some points) of ∗ I A 6 ∅ (V ) (f1,...,fn) are isolated in V (f1,...,fn)? If not, then characterize those I [n] which satisfy the hypothesis of the preceding question but fail the conclusion. ⊆ 10. OPEN PROBLEMS 167

∗ n 10.2. Non-isolated zeroes and non-degeneracy. In contrast to the case of (k ) , example X.16 shows that for f = (f1,...,fn) ( ), the existence of non-isolated solutions of f1,...,fn does ∈ L A n not automatically mean that f1,...,fn are (k , )-non-degenerate. More precisely, part (b) of exam-

n A n k ple X.16 shows that if I D(k , ), then it is possible for f1,...,fn to be ( , )-non-degenerate even if (V ∗)I (f) has non-isolated∈ points.A On the other hand, condition (b) of (U, )-non-degeneracyA implies

n ∗ I A n k that if I I (k , ) and (V ) (f) has non-isolated points, then f1,...,fn are ( , )-degenerate. The question∈ is if it is sufﬁcient.A A

n Problem X.41. If I D(k , ), does there exist f = (f1,...,fn) such that Supp(fj )= j for each j, ∗ I ∈ A A and (V ) (f) is nonempty (and due to proposition X.21 necessarily positive dimensional), but f1,...,fn

n n k are (k , )-non-degenerate? If not, then characterize those I D( , ) for which there is no such f ( A). ∈ A ∈ L A 10.3. Simple criteria for equality of Li and Wang’s bound. Since the upper bound of Li and Wang from (101) is so simple, it would be interesting to find simple criteria under which it holds with equality. Proposition X.14 gives a characterization of all such scenarios, so the question is if it can be made “more explicit” in any sense, or if there are simpler criteria (e.g. as in corollary X.15 or assertion (1) of proposition X.14) which are sufficient. One possible criterion was proposed in [RW96] and [Roj99], but that turns out to be incorrect. Indeed, both [RW96, Theorem 1] and [Roj99, Affine Point Theorem II] imply the following: if E ( ) = and the intersection of each j with each of the n coordinate hyperplanes is nonempty, then Li andA Wang’s∅ bound holds with equality.A This is indeed the case for n 2, but as the following example shows, it is false in higher dimensions. ≤

2 ′ ′ ′ 2 k ′ ′ ′ Example X.42. Let f1 := ax+by+cx , f2 := a x+b y+c x , f3 := pz x+q, where a,b,c,a ,b ,c ,p,q

n k are generic elements in k and k 1. It is straightforward to check directly that on there are precisely ≥ k solutions for f1 = f2 = f3 = 0 and 2k solutions for f1 = f2 = f3 = t for generic t = 0, so that Li and Wang’s bound fails for := Supp(f ), j = 1, 2, 3. Note that both conditions (iv) and6 (v) from Aj j section X.4.2 hold with the weighted degree ν on k[x,y,z] corresponding to weights x k, y k, z 1. 7→ 7→ 7→ − 10.4. “Compact” formulae for general weighted and weighted multi-homogeneous versions of Bezout’s´ theorem. There are scenarios not covered in weighted Bézout theorem II (theorem X.36) for which very similar bound exists, e.g. in the case that I− = 1 (and no restriction that the dj have to be nonnegative). This motivates the question: is it possible| to| find a formula that is more explicit than the one from weighted Bézout theorem III (theorem X.37), and which is more general than version II? That would also lead to a more general version of weighted multi-homogeneous Bézout theorem II. CHAPTER XI

Milnor number of a hypersurface at the origin

1. Introduction The modern theory of applications of Newton polyhedra to affine Bézout problem started from A. Kushnirenko’s work aimed at answering V.I. Arnold’s questionon Milnor numbers of generic singularities. In [Kou76] Kushnirenko gave a beautiful formula for a lower bound of the Milnor number at the origin in terms of volumes of the region bounded by the Newton diagram, and showed that the bound is attained in the case that the singularity is Newton non-degenerate. In this chapter we reverse the arrow of applications and show that the notionof non-degeneracyat the origin introduced in chapter IX can be used to derive(and generalize) Kushnirenko’s result on Milnor numbers. In particular, based on non-degeneracy at the origin we introduce a non-degeneracycriterion which generalizes Newton non-degeneracy and inner Newton non- degeneracy1, the latter introduced by C. T. C. Wall [Wal99]. We show that in zero characteristic the new criterion is necessary and sufficient for the Milnor number to be the minimum, and the minimum Milnor number can be obtained by Kushnirenko’s bound. In positive characteristic this criterion is sufficient, but not necessary.

2. Milnor number

The Milnor number µ0(f) of a power series f in (x1,...,xn) is the dimension over k of the quotientof k[[x1,...,xn]] by the ideal generated by the partial derivativesof f, i.e. µ0(f) = [∂f/∂x1,...,∂f/∂xn]0. It is a fundamental measure of complexity of the singularity of V (f) at the origin (in the case that f is the

Taylor series of a rational function, or in the case that k = C and f is a analytic at the origin).

Proposition XI.1. Let f k[x ,...,x ] such that f(0) = 0. ∈ 1 n (1) µ0(f)=0 if and only the origin is a nonsingular point of V (f). (2) If V (f) has a non-isolated singularity at the origin, then µ (f) = . The converse holds if 0 ∞

char(k)=0.

PROOF. The ﬁrst assertion is clear, so we prove the second assertion. If V (f) has a non-isolated singularity at the origin, then the origin is a non-isolated point of V (∂f/∂x1,...,∂f/∂xn), so that µ (f) = (proposition IV.16). Now assume µ (f) = . Then the origin is a non-isolated point of 0 ∞ 0 ∞ V (∂f/∂x1,...,∂f/∂xn) (proposition IV.16) and therefore there is an irreducible curve C containing the origin such that C V (∂f/∂x ,...,∂f/∂x ) (corollary III.21). It sufﬁces to show that f 0 if

⊆ 1 n |C ≡

k k ˆ char( )=0. Pick a nonsingular point a C. Then there is an isomorphism φ : [[t]] ∼= C,a, and n ∈ ˆ O d(f φ)/dt = (∂f/∂x )(d(φ )/dt) 0 . Since char(k)=0, it follows that f is constant ◦ j=1 j j ≡ ∈ OC,a on C. Since 0 C and f(0) = 0, it follows that C V (f), as required. ∈P ⊆

Example XI.2. The converse to assertion (2) of proposition XI.1 may not be true if p := char(k) is positive; consider e.g. the case that n =1 and f(x) = xp, or n =2 and f(x, y)= xp + yq, where q 2 is relatively prime to p. The latter example in particular shows that Milnor number can be inﬁnite even≥ for isolated singular points.

In the case that k = C and the origin is an isolated singular point of V (f), Milnor originally deﬁned 2n−1 µ0(f) in [Mil68, Chapter 7] in the following way: let Sǫ be the sphere of radius ǫ centered at the origin of Cn = R2n and S2n−1 := S2n−1 be the unit sphere of Cn. Given a morphism g : Cn Cn such that ∼ 1 →

1This terminology is taken from [BGM12].

168 3. GENERIC MILNOR NUMBER 169 the origin is an isolated zero of g−1(0), the multiplicity of g at the origin is the degree of the mapping2 S2n−1 S2n−1 given by z g(z)/ g(z) (where is the Euclidean distance). Milnor defined ǫ 7→ 7→ || || || · || µ0(f) as the multiplicity at the origin of the map z (∂f/∂x1,...,∂f/∂xn). Milnor showed that for all 2n−1 1 7→ sufficiently small ǫ, if φ : Sǫ V (f) S is the map given by z f(z)/ f(z) , then each fiber of φ is a smooth (2n 2)-dimensional\ real manifold→ with homotopy type7→ of a “bouquet|| ”||Sn−1 Sn−1 − ∨···∨ of spheres, and µ0(f) is precisely the number of spheres in the bouquet. The fact that the multiplicity of n n a map g : C C at the origin equals [g1,...,gn]0 was left in [Mil68, Appendix B] as an exercise; a proof can be found→ in [AGZV85, Chapter I.5].

3. Generic Milnor number

n Let be a (possibly infinite) subset of Z . We write 0( ) for the set of all power series in A ≥0 L A (x ,...,x ) supported at . For each j =1,...,n, define 1 n A n (120) ∂ /∂xj := α ej : α , α ej Z , p does not divide αj A { − ∈ A − ∈ ≥0 } n where ej is the j-th standard unit vector in Z and p := char(k). Note that ∂ /∂xj is the support of A ∂g/∂x for generic3 g ( ). Define j ∈ L0 A µ ( ) := min µ (f): f ( ) 0 A { 0 ∈ L0 A } In theorem XI.3 below we estimate µ0( ) in terms of the intersection multiplicity [Γ1,..., Γn]0 at the origin of the Newton diagrams Γ of ∂A/∂x . Given f ( ), we say that f is partially -non- j A j ∈ L0 A A degenerate at the origin if the partial derivatives of f are (∂ /∂x1, . . . , ∂ /∂xn)-non-degenerate at the origin in the sense of definition IX.6; in the case that =A Supp(f), weA simply say that f is partially non-degenerate at the origin. In other words, f is partiallyA non-degenerate at the origin if it satisfies the following property: for each nonempty subset I of [n] and each weighted order ν centered at the origin on

(121) ∗ n k k I

[x : i I], there is no common zero of In (∂f/∂x k ), j =1,...,n, on ( ) . i ∈ ν j| Recall that one does not have to check this condition for all nonempty subsets of [n] - see theorem IX.9 and remark IX.10.

THEOREM XI.3 ([Mon16]). Let Γ := ND(∂ /∂x ), j =1,...,n. Assume 0 . Then j A j 6∈ A (122) µ ( ) [Γ ,..., Γ ] 0 A ≥ 1 n 0 Moreover, (1) The following are equivalent for all f ( ): ∈ L0 A (a) µ0(f) = [Γ1,..., Γn]0 < , (b) f is partially -non-degenerate∞ at the origin. ′ A (2) Let 0( ) be the set of all f 0( ) which are partially -non-degenerate at the origin. Let ′ M A ∈ L A ′ A ′ ′ be any ﬁnite subset of such that ∂ /∂xj Γj ∂ /∂xj, j =1,...,n. Then 0( ) is A A ′ A′ ⊇ −1∩ A′ ′ M A′ a Zariski open subset of 0( ), and 0( )= π ( 0( )), where π : 0( ) 0( ) is the natural projection. L A M A M A L A → L A ′ (3) If [Γ1,..., Γn]0 = , then 0( )= . If [Γ1,..., Γn]0 < , then µ0( ) = [Γ1,..., Γn]0 if ′ ∞ M A ∅ ∞ A and only if 0( ) is nonempty. M A ′ (4) If char(k)=0 and [Γ ,..., Γ ] < , then ( ) is nonempty and µ ( ) = [Γ ,..., Γ ] . 1 n 0 ∞ M0 A 0 A 1 n 0 Example XI.4. Assertion (4) of theorem XI.3 may not be true, i.e. the bound in (122) may be strict, in the 2 case that p := char(k) > 0. E.g. let := (p+1, 1), (1,p+1) Z≥0. Then ∂ /∂x1 = (p, 1), (0,p+ A { p }⊂p+1 A p+1 { p 1) and ∂ /∂x2 = (p +1, 0), (1,p) . If f1 = a1,1x1x2 + a1,2x2 and f2 = a2,1x1 + a2,2x1x2 are generic} polynomialsA { supported respectively} at ∂ /∂x and ∂ /x , then [f ,f ] = (p + 1)2. Therefore A 1 A 2 1 2 0 2The degree of a differentiable map φ : M → N between oriented differentiable manifolds of the same dimension, where M −1 is compact and N is connected, is the sum of sign of dfx over all x ∈ φ (y) for a generic y ∈ N, where dfx is the derivative map from the tangent space of M at x to the tangent space of N at y, and the sign of dfx is either 1 or −1 depending on whether dfx preserves or reverses orientation. 3 “Generic” refers to elements of a nonempty Zariski open (dense) subset of L0(A) in the Zariski topology mentioned in remark IX.7. 170 XI.MILNORNUMBEROFAHYPERSURFACEATTHEORIGIN

2 p+1 p+1 [Γ1, Γ2]0 = (p+1) . Ontheotherhand,if f = ax1 x2 +bx1x2 is a generic polynomialsupported at , p p p p p p pA then ∂f/∂x1 = x2(ax1 + bx2) and ∂f/∂x2 = x1(ax1 + bx2) so that µ0( ) = [x2(ax1 + bx2), x1(ax1 + p A bx2)]0 = > [Γ1, Γ2]0. It is straightforward to check that ∂f/∂x1,∂f/∂x2 are -degenerate at the origin, i.e. ∞ ′ ( )= . A M0 A ∅ PROOF OF THEOREM XI.3. Assertions (1),(2)and(3) follow from theorem IX.8 and corollary IX.18. ′ Therefore it sufﬁces to show that 0( ) is nonempty in the case that char(k)=0. We may assume without loss of generality that 0 <µM( A) < . Pick any ﬁnite subset ′ of satisfying the assumptions 0 A ∞ A A of assertion (2). Let I [n] and ν be a weighted order on k[xi : i I] centered at the origin. Denote by I the set of all g ⊆( ′) such that ∈ Lν ∈ L0 A Supp(g)= ′, and • A ∗ n

I I k k V (In (∂g/∂x k ),..., In (∂g/∂x )) ( ) = • ν 1| ν n| ∩ ∅ I ′ It sufﬁces to show that ν contains a nonempty Zariski open subset of 0( ). We may assume without loss of generality that I L= 1,...,k . Take g ( ′), and express it asL A { } ∈ L0 A n g = g (x ,...,x )+ x g (x ,...,x )+ 0 1 k i i 1 k ··· i=Xk+1 where the omitted terms have quadratic or higher order in (x ,...,x ). Since 0 < [Γ ,..., Γ ] < , k+1 n 1 n 0 ∞ I corollary IX.13 implies that ∂g/∂x k 0 for l k values of j; denote them by 1 j < < j n. j| 6≡ ≥ ≤ 1 ··· l ≤

I I k If g0 0, then each ji > k, and V (Inν (∂g/∂x1 k ),..., Inν (∂g/∂xn )) = V (Inν (gj1 ),..., Inν (gjl ). ≡ I | | Therefore g ν if gj1 ,...,gjl are BKK non-degenerate. Since l k, theorem VII.12 then implies I ∈ L ′ ≥ that ν contains a nonempty Zariski open subset of 0( ), as required. So assume g0 0. Let L ′ L A ∗ n ′ 6≡ h0 := Inν (g0) and V (h0) := V (∂h0/∂x1,...,∂h0/∂xk) (k ) . Let 0 := Supp(h0). Since ′ ∩ A

I I k V (Inν (∂g/∂x1 k ),..., Inν (∂g/∂xn )) V (h0), it sufﬁces to show that the set of all polynomi- ′| ′ | ⊆ ′ als h 0( 0) such that V (h) = contains a nonempty Zariski open subset of 0( 0). Let Z :=

∈ L A ∗ n ′ ∅ L A∗ n ′ k (x, h): x (k ) , h 0( 0), ∂h/∂x1(x) = = ∂h/∂xk(x) =0 ( ) 0( 0), { ∈ ∗ n ∈ L A ′ ··· } ⊂ × L A and let π1 : Z (k ) and π2 : Z 0( 0) be the natural projections. It suffices to show that → ′ → L A ′ dim(π2(Z)) < dim 0( 0). We prove this by a dimension count. Denote the elements in 0 by L A αi A′ αi = (αi,1,...,αi,n), i = 1,...,N, and the coefficients of x (in a polynomial supported at 0) by ∗ n −1 ′ A ai. Let x (k ) . Then π1 (x) is the subspace of 0( 0) defined by a system of linear equations of the form ∈ L A

α xα1−e1 α xαN −e1 a 1,1 ··· N,1 1 . . . (123)  . .   .  =0 α xα1−ek α xαN −ek a  1,k ··· N,k   N      k where e1,...,ek are the standard unit vectors in Z . Since char (k)=0 (note: this is the only place the

assumption of zero characteristic is used), the rank (as a matrix over k) of the left-most matrix in (123) is the same as the rank (as a matrix over Q) of α α 1,1 ··· N,1 . . B :=  . .  α α  1,k ··· N,k   Since ν has positive weights, and the αj belong to a level set (corresponding to a positive value) of ν, it follows that Rank(B) = 1 + dim(NP(h )) and therefore dim(π−1(x)) = N 1 dim(NP(h )). 0 1 − − 0 Since this is independent of x, it follows that dim(Z) = N + n 1 dim(NP(h0)). On the other ′ ′ − − hand, if x V (h) for some h 0( 0), then since the support of each of ∂h/∂xj is contained in ∈ ∈ L A β a translation of NP(h0), it is straightforward to check that ∂h/∂xj(xz )=0 for each j = 1,...,k, ∗ n n and each z (k ) and each β Z which is normal (with respect to the “dot product”) to NP(h0). ∈ ∈ Therefore the dimension of each ﬁber of π2 is at least n dim(NP(h0)). It follows that dim(π2(Z)) dim(Z) n + dim(NP(h )) = N 1, as required. − ≤ − 0 − 4. CLASSICAL NOTIONS OF NON-DEGENERACY 171

4. Classical notions of non-degeneracy

4.1. Newton non-degeneracy. For a power series f in (x1,...,xn), we write j(f) for the ideal generated by the partial derivatives of f. We say that f is Newton non-degenerate iff j(Inν (f)) has no zero

∗ n k on (k ) for each weighted order ν on [[x1,...,xn]] centered at the origin. Newton non-degeneracy is possibly the most studied non-degeneracy property of hypersurface germs: it is a Zariski open condition (in the same sense as partial non-degeneracy at the origin) and, in the characteristic zero case, also nonempty. In this section we discuss its relationship with partial non-degeneracy at the origin. In general Newton non-degeneracy does not imply “ﬁnite determinacy,” i.e. f can be Newton non-degenerate but still V (f) may have a non-isolated singularity at the origin (take e.g. f := x1 xn, n 2). However, if the Newton diagram of f is convenient then Newton non-degeneracy of f implies··· that the≥ origin is an isolated singularity of V (f) (corollary XI.7), which can be resolved by a “toric modiﬁcation.” As a result, the invariants of the singularity can be computed combinatoriallyin terms of the diagram(see e.g. [Oka97]). We show in proposition XI.6 that for isolated singularities, Newton non-degeneracy is a special case of partial non-degeneracy at the origin. However, the following example shows that even in the case of convenient diagrams partial non-degeneracy at the origin does not imply Newton non-degeneracy.

q Example XI.5. Let f := x1 + (x2 + x3) , where q 2. Then ND(f) is convenient and f is not Newton non-degenerate (take ν with weights (q +1, 1, 1) for≥(x,y,z)). However, f is partially non-degenerate at the origin with µ0(f)=0. We now show that Newton non-degeneracyimplies partial non-degeneracyat the origin. The following ′ notation is used in its proof: let I I′ [n], ν (RI )∗ and ν′ (RI )∗. We say that ν and ν′ are

⊆ ⊆ ∈ ∈ ′ ′ k compatible if the weighted order on k[xi : i I] induced by ν and the weighted order on [xi′ : i I ] induced by ν′ are compatible in the sense of section∈ IX.6. ∈

Proposition XI.6 ([Mon16]). Let f k[[x ,...,x ]] and Γ := ND(∂f/∂x ), j = 1,...,n. If f ∈ 1 n j j is Newton non-degenerate and [Γ1,..., Γn]0 < , then f is partially non-degenerate at the origin. In particular, if f is Newton non-degenerate and ND(∞f) is convenient, then f is partially non-degenerate at the origin.

PROOF. We start with a direct proof of the second assertion since it is easier to see. Assume Γ := ND(f) is convenient and f is Newton non-degenerate. Pick a nonempty subset I of [n] and a weighted I order ν on k[xi : i I] which is centered at the origin. Since Γ is convenient, Γ R = . Therefore we ∈ ′ ′ ∩ 6 ∅ I can ﬁnd a weighted order ν on k[x1,...,xn] such that ν is compatible with ν and Inν′ (Γ) R . Then ⊂ Inν′ (f) depends only on (xi : i I). Since f is Newton non-degenerate, it follows that ∂(Inν′ (f))/∂xi, ∈ ∗ n i I, do not have any common zero in (k ) . But if i I is such that ∂(In ′ (f))/∂x is not identically ∈ ∈ ν i

′ I I I

k k zero, then ∂(Inν (f))/∂xi = ∂(Inν (f k ))/∂xi = Inν (∂f/∂xi ). This implies that Inν (∂f/∂xi ), | ∗ I | | i I, do not have any common zero on (k ) , as required for partial non-degeneracy of f at the origin. ∈

Now we prove the ﬁrst assertion. If [Γ1,..., Γn]0 =0, then corollary IX.13 implies that f is partially non-degenerateat the origin. So assume 0 < [Γ1,..., Γn]0 < and f is partially degenerate at the origin. It sufﬁces to show that f is Newton degenerate. Pick I [n] and∞ a primitive weighted order ν centered at

⊆ ∗ n k k I the origin on [xi : i I] such that Inν (∂f/∂xi k ), i I, have a common zero (a1,...,an) ( ) . ∈ | ∈ ∈ ′ I

At ﬁrst consider the case that f k 0. Then as in the convenient case pick a weighted order ν on

′ | 6≡ k k[x ,...,x ] such that ν is compatible with ν and In ′ (f) [x : i I]. Then for each j =1,...,n, 1 n ν ∈ i ∈ ′ I

∂ Inν (f)/∂xj is either Inν (∂f/∂xj k ) or is identically zero. It follows that (a1,...,an) is a common ∗ n |

′ k I zero of j(In (f)) on ( ) , so that f is Newton degenerate, as required. Now assume that f k 0. ν | ≡ I We may assume that I = 1,...,k for some k, 1 k n, and ∂f/∂xj k 0 if and only if i = k +1,...,k + l. Then f can{ be expressed} as ≤ ≤ | 6≡

l (124) f = xk+j fj (x1,...,xk)+ xk+ixk+j fi,j (x1,...,xn) j=1 X i≥Xj≥1 Let h := In (f ), j =1,...,l, and B be the k l matrix with (i, j)-th entry (∂h /∂x )(a ,...,a ). We j ν j × j i 1 k claim that Rank(B) < k. Indeed, let νi := ν(xi), i = 1,...,k. For each j = 1,...,l, by assumption 172 XI.MILNORNUMBEROFAHYPERSURFACEATTHEORIGIN

ν ν I 1 k

(a ,...,a ) is a common zero of In (∂f/∂x k )= h (x ,...,x ), so that h (a t ,...,a t )=0 1 k ν k+j| j 1 k j 1 k for all t k. Note that ∈ k k d ∂hj ∂hj (h (a tν1 ,...,a tνk )) = (a tν1 ,...,a tνk )ν a tνi−1 = (a ,...,a )ν a tν(hj )−1 dt j 1 k ∂x 1 k i i ∂x 1 k i i i=1 i i=1 i X X ′ n ′ ′ If a := (ν1a1,...,νkak) k , then it follows that a B =0. Since ν is primitive, a =0, so that the map

k l ∈ 6 k k given by multiplication by B on the right is not injective. Therefore Rank(B) < k, as claimed. → Since 0 < [Γ1,..., Γn]0 < , corollary IX.13 implies that l k, so that there is b = (b1,...,bl) =0 l ∞ ≥ 6 ∈ k such that B times the transpose of b is zero. Let m be the number of nonzero coordinates of b. Without loss of generality we may assume that b =0 if and only if j =1,...,m. Then we have that j 6 m ∂h (125) b j (a ,...,a )=0 j ∂x 1 k j=1 i X ′ Fix positive integers q, qk+m+1,...,qn and let ν be the weighted order on k[x1,...,xn] such that

ν(xi) if i =1, . . . , k, ′ ν (xi)= q ν(f ) if i = k +1,...,k + m,  − i−k qi if i = k + m +1, . . . , n. m If and for , then identity (124) implies that ′ . q 1 qi q i = k + m +1 ,...,n Inν (f)= j=1 xk+j hj ′ ≫ ′ ≫ ∗ ′ ′ If b1,...,bn−k−m are arbitrary elements in k , it follows that (a1,...,ak,b1,...,bm,b1,...,bn−k−m) is ∗ n P a zero of j(Inν′ (f)) on (k ) . Therefore f is Newton degenerate, as required.

Corollary XI.7 (Brzostowski and Oleksik [BO16, Theorem 3.1]). Let Γj := ND(∂f/∂xj ), j =1,...,n. Assume f(0) = 0 and f is Newton non-degenerate. Then the following are equivalent: (1) µ (f) < . 0 ∞ (2) [Γ1,..., Γn]0 < . I ∞ (3) j :Γj R = I for each I [n]. |{ ∩ 6 ∅}| ≥ | | ⊆ PROOF. Combine corollary IX.13, theorem XI.3, and proposition XI.6. 4.2. Inner Newton non-degeneracy. A diagram Γ in Rn is the Newton diagram of some subset of n n R≥0, and a face ∆ of Γ is a compact face of Γ+ R≥0. ∆ is called an inner face if it is not contained in n α any proper coordinate subspace of R . If f = n cαx is a power series in (x1,...,xn) and ∆ is α∈Z≥0 a subset of n, we write α. We say that is inner Newton non-degenerate if there is a R f∆ := α∈∆ cαx P f convenient diagram Γ such that P (a) no point of Supp(f) “lies below” Γ, i.e. Supp(f) Γ+ Rn , and ⊆ ≥0 (b) for every inner face ∆ of Γ and for every nonempty subset I of [n], I ∗ I (126) ∆ R = V (j(f∆)) (k ) = ∩ 6 ∅ ⇒ ∩ ∅ The difference between Newton non-degeneracy and inner Newton non-degeneracy is most evident in the case of weighted homogeneous polynomials with isolated singularities, e.g. consider the polynomial q f := x1 + (x2 + x3) , where q 2, from example XI.5. The Newton diagram of f is convenient and has only one inner face, namely≥ the two dimensional face ∆ with vertices (1, 0, 0), (0, q, 0), (0, 0, q). In particular, f∆ = f and ∂f/∂x1 = 1, which implies that f is inner non-degenerate. However, as we saw in example XI.5, f is Newton degenerate. In fact C. T. C. Wall introduced inner Newton non- degeneracy in [Wal99] in order to ﬁnd a condition which (in the case of power series with convenient Newton diagrams) is weaker than Newton non-degeneracy, but still wide enough to include all weighted homogeneous polynomials with isolated singularities. We now show that inner Newton non-degeneracy n implies partial non-degeneracy at the origin. Given ∆ R , we write I∆ for the smallest subset of [n] ⊆ such that ∆ RI∆ . ⊆ Lemma XI.8. Let Γ be a convenient diagram, I [n] and ∆ be a face of Γ RI . Pick an integral element I ∗ ⊆ I ∩ ′ n ∗ ν (R ) centered at the origin such that ∆=Inν (Γ R ). Then there is an integral ν (R ) which ∈ ∩ ′ ∈ is centered at the origin and compatible with ν I such that ∆ := In ′ (Γ) is an inner face of Γ and |R ∆ ν ∆ = ∆′ RI∆ . ∩ 5. NEWTON NUMBER: KUSHNIRENKO’S FORMULA FOR THE GENERIC MILNORNUMBER 173

PROOF. We proceed by induction on ǫ := n I∆ . It is obvious if ǫ = 0. Now assume ǫ = 1. − | | I∆ ∗ Then without loss of generality we may assume I∆ = 1,...,n 1 . Pick ν (R ) such that I∆ { − } ∈ n ∗ ∆ = Inν (Γ R ). For each pair of positive integers q, r, let νq,r be the element on (R ) deﬁned as follows: ∩ ν , (α ,...,α ) := q ν, (α ,...,α ) + rα h q,r 1 n i h 1 n−1 i n If r q, then In (Γ) = ∆. On the other hand, since ν is centered at the origin and Γ is convenient, if ≫ νq,r q r, then Inνq,r (Γ) is a point on the xn-axis. Therefore we can ﬁnd q, r such that Inνq,r (Γ) contains ∆ ≫ ′ and also a point with positive n-th coordinate. The lemma then holds with ν := νq,r. Now assume ǫ 2. ′ [n]\{i′} ∗ ≥ Pick i [n] I∆. The induction hypothesis implies that there is ν˜ (R ) which is centered at the ∈ \ ˜ [n]\{i′}∈ [n]\{i′} origin and compatible with ν I∆ such that ∆ := Inν˜(Γ R ) is an inner face of Γ R and |R ∩ ∩ ∆ = ∆˜ RI∆ . The ǫ = 1 case then implies that there is ν′ (Rn)∗ which is centered at the origin and ∩ ′ ∈ ′ [n]\{i′} ˜ compatible with ν˜ such that ∆ := Inν′ (Γ) is an inner face of Γ and ∆ R = ∆. The lemma then holds with ν′, which completes the proof. ∩

Proposition XI.9 ([Mon16]). If f k[[x1,...,xn]] is inner Newton non-degenerate, then it is partially non-degenerate at the origin. ∈

PROOF. Assume f is inner Newton non-degenerate with respect to a convenient diagram Γ. Pick I I [n] and a weighted order ν on k[xi : i I]. Let ∆ := Inν (Γ R ). Lemma XI.8 implies that there is ′⊆ n ∗ ∈ ′ ∩ ′ I∆ ν (R ) compatible with ν RI∆ such that ∆ := Inν′ (Γ) is an inner face of Γ, and ∆ = ∆ R . Fix j, 1∈ j n. | ∩ ≤ ≤ Claim XI.9.1. One of the following holds:

I∆ (i) either ∂f∆′ /∂xj is identically zero on k , ′ I I

(ii) or k . ∂f /∂x k = In (∂f/∂x ) ∆ j | ∆ ν j| I∆ ′ ′ PROOF. Assume ∂f∆′ /∂xj is not identically zero on k . Let m := ν (f) = ν (f∆′ ). Then ′ ′ ′ ν (∂f∆′ /∂xj ) = m ν (xj ). There is a positive rational number λ such that ν RI∆ = λν RI∆ . Then − ′ ′ | | ′ I ′ I I

. Now pick k k ν(∂f∆ /∂xj k ∆ ) = ν (∂f∆ /∂xj ∆ )/λ = (m ν (xj ))/λ α Supp(∂f/∂xj ) | | −α ′ ∈ | \ ′ I . It sufﬁces to show that . Let be the unit vec- Supp(∂f∆ /∂xj k ∆ ) ν(x ) > (m ν (xj ))/λ ej | n ′ − ′ ′ tor along the (positive) -th axis on . Pick ′ I . Then and j R α Supp(∂f∆ /∂xj k ∆ ) α + ej ∆ ′ ′ ′ ′ ∈ ′ | ′ ′ ′ ∈ α + ej Γ ∆ . Therefore ν , α + ej > ν , α + ej , so that ν , α > ν , α = m ν (xj ). It follows that ν,∈ α =\ ν′, α /λ > (mh ν′(x i))/λh, as required.i h i h i − h i h i − j ∗ I

I I k k Claim XI.9.1 implies that V (Inν (∂f/∂x1 k ),..., Inν (∂f/∂xn )) ( ) is contained in the prod-

| ∗ I∆ ∗ I\I∆| ∩ k

uct of ′ I ′ I k with . The inner non-degeneracyof with k V (∂f∆ /∂x1 k ∆ ,...,∂f∆ /∂xn ∆ ) ( ) ( ) f | | ∩ ∗ I

I I k k respect to Γ then implies that V (In (∂f/∂x k ),..., In (∂f/∂x )) ( ) = , as required. ν 1| ν n| ∩ ∅ Corollary XI.10 (Wall [Wal99, Lemma 1.2]). Inner Newton non-degeneracy implies ﬁnite determinacy.

If p := char(k) is nonzero, partial Newton non-degeneracy at the origin is strictly weaker than inner Newton non-degeneracy, e.g. xp +yp +xp+1 +yp+1 is partially non-degenerateat the origin, but it is inner Newton degenerate. We do not know if this is true in zero characteristic (see section XI.6.2).

5. Newton number: Kushnirenko’s formula for the generic Milnor number Let Γ be a diagram in Rn. We write Γ¯ for the region bounded by the cone with base Γ and apex at the − ¯ origin, and Vk (Γ), 0 k n, for the sum of k-dimensional Euclidean volumes of the intersections of Γ ≤ ≤ n with the k-dimensional coordinate subspaces of R (in particular, V0(Γ) is deﬁned to be 1). The Newton number of Γ is n ( 1)n−kk!V −(Γ) if Γ is convenient, ν(Γ) := k=0 − k sup ν(Γ me ,...,me ): m 0 otherwise, (P { ∪{ 1 n} ≥ } n where ej are the unit vectors along the (positive direction of the) axes of R . Let f k[[x1,...,xn]] such that f(0) = 0. A. Kushnirenko proved in [Kou76] that µ (f) ν(ND(f)), and ∈µ (f) = ν(ND(f)) if 0 ≥ 0 f is Newton non-degenerate and if either char(k)=0 or ND(f) is convenient. C. T. C. Wall proved in 174 XI.MILNORNUMBEROFAHYPERSURFACEATTHEORIGIN

[Wal99] that if char(k)=0, then the equality µ0(f)= ν(ND(f)) continues to hold if f is inner Newton non-degenerate. In this section we present some generalizations of these results.

5.1. Preliminary results. Let be a subset of Zn not containing the origin. For each m 0, let A ≥0 ≥ := (α ,...,α ): n α m . For each j =1,...,n, let Am A∪{ 1 n j=1 j ≥ } ′ n P := α ej : α m, α ej Z ) Am,j { − ∈ A − ∈ ≥0} n ′ where ej are the unit vectors along the (positive direction of the) axes of R . Note that ∂ m/∂xj , A ⊆ Am,j and the inclusion is proper if char(k) is positive. In any event, theorem XI.3 and the monotonicity of intersection multiplicity (remark IX.3) imply that for each m 1, ≥ (127) µ ( ) [∂ /∂x , . . . , ∂ /∂x ] [∂ /∂x , . . . , ∂ /∂x ] [ ′ ,..., ′ ] 0 A ≥ A 1 A n 0 ≥ Am 1 Am n 0 ≥ Am,1 Am,n 0 Lemma XI.11. Let Zn 0 . Deﬁne A⊆ ≥0 \{ } ′ n := α ej : α , α ej Z Aj { − ∈ A − ∈ ≥0} n ′ where ej are the j-th standard unit vector in R , j =1,...,n. Assume each is convenient. Then is Aj A also convenient and [ ′ ,..., ′ ] = ν(ND( )). A1 An 0 A ′ Remark XI.12. Kushnirenko’stheorem implies that the assumption “each j is convenient” in lemma XI.11 is not needed - see corollary XI.16 below. A

′ ′ PROOF OF LEMMA XI.11. Note that [ 1,..., n]0 = [g1,...,gn]0 for all g1,...,gn k[[x1,...,xn]] such that A A ∈ (i) Supp(g )= ′ for each j, and j Aj (ii) g1,...,gn are non-degenerate at the origin.

We will show that there are g1,...,gn which satisfy both of the above properties and in addition satisfy [g1,...,gn]0 = ν(ND( )). Indeed, choose g1,...,gn which satisfy the above properties, and in addition satisfy A ′ I I (iii) for each pair of subsets I,J of [n] such that I = J , the restrictions gj k , j J are ( R : | | | | | ∈ Aj ∩ j J)-non-degenerate at the origin. ∈ Note that this is possible since each ′ is convenient and since “generic” systems are non-degenerate at Aj the origin (theorem IX.8). For each I,J [n] such that I + J = n, we write ⊆ | | | |

[(gi)i∈I , (xj )j∈J ]0 := [gi1 ,...,gik , xj1 ,...,xjn−k ]0 where I = i ,...,i and J = j ,...,j . Since ′ are convenient, corollary IX.13 and property { 1 k} { 1 n−k} Aj (iii) imply that [(g ) , (x ) ] is deﬁned for each I [n]. Consequently corollary B.50 implies that i i∈I j j∈[n]\I 0 ⊆ n−|I| n−|I|

I I k [g ,...,g ] = ( 1) [(x g ) , (x ) ] = ( 1) [x g k ,...,x g ] 1 n 0 − i i i∈I j j∈[n]\I 0 − i1 i1 | i|I| i|I| | 0 IX⊆[n] IX⊆[n] where i1,...,i|I| are elements of I for each I [n]. Fix I [n]. Since the Newton diagram of the union I ⊆ ⊆ I of the supports of xij gij k is ND( ) R (this is where the assumption 0 is used!), property (iii) and proposition IX.30 imply| that A ∩ 6∈ A

− I

I I k [xi1 gi1 k ,...,xi gi ]0 = V (ND( ) R ) | |I| |I| | |I| A ∩ The result now follows from the deﬁnition of ν( ). · Corollary XI.13. [ ′ ,..., ′ ] = ν(ND( )) for each m 1. Am,1 Am,n 0 Am ≥

Corollary XI.14 (Kushnirenko [Kou76, TheoremI, part (i)]). µ0(f) ν(ND(f)) for all f k[[x1,...,xn]]. In particular, if ν(ND(f)) = , then µ (f)= . ≥ ∈ ∞ 0 ∞ PROOF. Combine inequation (127) and corollary XI.13. 5. NEWTON NUMBER: KUSHNIRENKO’S FORMULA FOR THE GENERIC MILNORNUMBER 175

5.2. Characteristic zero case. Continue to assume that isa subsetof Zn not containing the origin. A ≥0

THEOREM XI.15. If char(k)=0, then µ ( )= ν(ND( )). 0 A A

PROOF. If char(k)=0, then theorem XI.3 implies that µ ( ) = [∂ /∂x , . . . , ∂ /∂x ] . At 0 A A 1 A n 0 ﬁrst consider the case that [∂ /∂x1, . . . , ∂ /∂xn]0 < . For m 1 then proposition IV.15 implies that [∂ /∂x , . . . , ∂ /∂xA ] = [∂ A/∂x , . . . , ∂ ∞/∂x ] . Since≫ in zero characteristic ′ = A 1 A n 0 Am 1 Am n 0 Am,j ∂ /∂x , inequation (127) and corollary XI.13 imply that µ ( ) = ν(ND( )). On the other hand, if Am j 0 A A [∂ /∂x1, . . . , ∂ /∂xn]0 = , then proposition IV.15 implies that supm[∂ m/∂x1, . . . , ∂ m/∂xn]0 = A, so that corollaryA XI.13 implies∞ that ν(ND( )) = , as required. A A ∞ A ∞ Corollary XI.16. Lemma XI.11 holds even without the assumption that each ′ is convenient. Aj PROOF. In zero characteristic µ ( ) = [ ′ ,..., ′ ] (theorem XI.3); now use theorem XI.15. 0 A A1 An 0

Corollary XI.17 (Cf. [Kou76, Characteristic zero case of Theorem I], [Wal99, Theorem 1.6], [BO16, k Corollaries 3.10 and 3.11]). Let f k[[x ,...,x ]] be such that f(0) = 0. Assume char( )=0. Then ∈ 1 n µ0(f) = ν(ND(f)) whenever f is partially non-degenerate at the origin. In particular, if f is either Newton non-degenerate or inner Newton non-degenerate, then µ0(f)= ν(ND(f)). PROOF. Combine theorems XI.3 and XI.15 and propositions XI.6 and XI.9. 5.3. The general case. We continue to use the notation of section XI.5.1. In particular, is a subset n ′ A of Z≥0 not containing the origin. Let 0( ) be as in theorem XI.3 the set of all f 0( ) which are partially -non-degenerate at the origin.M A Proposition IV.15 and theorem XI.3, inequation∈ L A (127) and corollary XI.13A imply the following result. Proposition XI.18. µ ( )= ν(ND( )) if and only if 0 A A (1) either ν(ND( )) = , or ′ A ∞ (2) (a) 0( ) is nonempty, and (b) M[∂ A/∂x , . . . , ∂ /∂x ] = ν(ND( )) for all m 1 not divisible by p. Am 1 Am n 0 Am ≫ Remark XI.19. Corollaries IX.13 and XI.16 imply that the following are equivalent: (1) ν(ND( )) = . (2) [ ′ ,...,A ′ ] ∞= . A1 An 0 ∞ (3) there is a nonempty subset I of [n] such that j : ′ RI = < I . |{ Aj ∩ 6 ∅}| | |

Let p := char(k). Example XI.4 shows that if p is positive, then condition (2a) of proposition XI.18 is nontrivial. The example below shows that condition (2b) is nontrivial as well.

pq/2 (p,p) pr/2

FIGURE 1. Subdivision of the area under the Newton diagram from example XI.20

r p p q Example XI.20. Assume p is positive. Let f = x1 + x1x2 + x2, where q, r are large positive integers not divisible by p. Let := Supp(f) and Γ := ND(f). It is straightforward to see from ﬁg. 1 that A r ν(Γ) = 2p(q + r)/2 (r + q)+1=(p 1)(q + r)+1. On the other hand, ∂f/∂x1 = rx1 and q−1 − − ′ ∂f/∂x2 = qx2 , so that f is partially non-degenerate at the origin. It follows that 0( ) = and r−1 q−1 M A 6 ∅ µ0( )= µ0(f) = [rx1 , qx2 ]0 = (r 1)(q 1). It follows that µ0( ) >ν(ND( )) for sufﬁciently largeAq, r. − − A A 176 XI.MILNORNUMBEROFAHYPERSURFACEATTHEORIGIN

We now state a condition which guarantees that condition (2b) of proposition XI.18 holds. Given a n subset of Z , let JB := j [n]: ∂ /∂xj = = j [n]: there is (β1,...,βn) such that p B ≥0 { ∈ B 6 ∅} { ∈ ∈ B does not divide β . The condition is the following: j } for each m 1 not divisible by p and for each face ∆ of ND( ), J = , (128) m Am∩∆ and the convex≫ hulls of ∂( ∆)/∂x , j J , are dependent.A 6 ∅ Am ∩ j ∈ Am∩∆ Proposition XI.21. Assume (128) holds. Then condition (2b) of proposition XI.18 also holds. In particular, µ ( )= ν(ND( )) if either ν(ND( )) = or ′ ( ) is nonempty. 0 A A A ∞ M0 A PROOF. Pick m 1 not divisible by p. The arguments from the proof of lemma XI.11 show that ≫ − it sufﬁces to prove that [x1g1,...,xngn]0 = n!Vn (ND( m)) for power series gj such that Supp(gj ) = ∂ /∂x and g ,...,g are non-degenerate at the origin.A Condition (128) ensures that the Newton dia- Am j 1 n gram of the union of the supports of xj gj is ND( m), and that the condition of assertion (2) of proposition IX.30 is also satisﬁed. Therefore the result followsA from proposition IX.30. Note that condition (128) is not necessary for (2b) of proposition XI.18 to hold - see section XI.6.3.

Corollary XI.22 ([Kou76, Positive characteristic case of Theorem I]). Let f k[[x ,...,x ]] such that ∈ 1 n f(0) = 0, ND(f) is convenient, and f is Newton non-degenerate. Then µ0(f)= ν(ND(f)). PROOF. Due to propositions XI.6 and XI.21 it suffices to show that := Supp(f) satisfies condition (128). Since is convenient, ND( ) = ND( ) for m 1. So pickA a face ∆ of ND( ). Define f A Am A ≫ A ∆ as in section XI.4.2. Let gj := ∂f∆/∂xj, j =1,...,n. Since f is Newton non-degenerate, it follows that ∗ n g1,...,gn are BKK non-degenerate. Since g1,...,gn have no common zero on (k ) , theorems VII.5, VII.7 and VII.30 imply that the Newton polytopes of gj are dependent, as required. 6. Open problems

n 6.1. Existence of non-degenerate polynomials. Let be a subset of Z≥0 not containing the origin. A ′ The estimate of µ0( ) from theorem XI.3 is exact if and only if the set 0( ) of power series which are partially -non-degenerateA at the origin is nonempty. Theorem XI.3 showsM A that in characteristic zero ′ A 0( ) is always nonempty, and example XI.4 shows that in positive characteristic there are such that M′ (A) is empty. This motivates the following problem. A M0 A ′ Problem XI.23. In the case that char(k) > 0, characterize those for which 0( ) is nonempty. Compute µ ( ) for those such that ′ ( ) is empty. A M A 0 A A M0 A ′0 Let 0 ( ) bethe set of allpower series f supported at such that ND(f) = ND(A), ND(∂f/∂xj)= N A A ′0 ND(∂ /∂xj ) for each j, and f is Newton non-degenerate. Proposition XI.6 implies that 0 ( ) ′ ( A) when µ ( ) < . The following is therefore a subproblem of problem XI.23 in that case.N A ⊆ M0 A 0 A ∞ ′0 Problem XI.24. In the case that char(k) > 0, characterize those for which ( ) is nonempty. A N0 A The proof of theorem XI.3 gives a sufﬁcient condition for existence of Newton non-degenerate polynomials: let be a ﬁnite subset of Zn and B be the n matrix whose columns are the elements of B.

B ≥0 × |B| k

Let Rankk ( ) denote the rank of a matrix over . The condition we are interested in is the following: ·

(129) Rankk (B) = dim(conv( ))+1 B Lemma XI.25. Assume (129) holds. Then the set of polynomials g supported at such that ∂g/∂xj, ∗ n B j =1,...,n, have no common zero on (k ) contains a nonempty Zariski open subset of the space of all polynomials supported at . B PROOF. This is the in fact the main content of the proof of theorem XI.3 (the only place where the zero characteristic played a role in that proof was to ensure that (129) holds). Corollary XI.26. If (129) holds with = ∆ for each face ∆ of ND( ), then ′0( ) = . B ∩ A A N0 A 6 ∅ Assertion (1) of theorem XI.3 implies that for to admit polynomials which are partially non- degenerate at the origin, it is necessary that A (130) [ ,..., ] < A1 An 0 ∞ 6. OPEN PROBLEMS 177

I Corollary IX.13 implies that (130) is equivalent to the condition that j : j R = I for each I [n]. The following is an immediate corollary of proposition XI.6 and|{ corollaryA ∩ XI.266 ∅}|. ≥ | | ⊆ Corollary XI.27. Assume (130) holds and (129) holds with = ∆ for each face ∆ of ND( ), then ′ ( ) = . B ∩ A A M0 A 6 ∅ Question XI.28. Is the condition from corollary XI.26 necessary for ′0( ) to be nonempty? N0 A 6.2. Relation among non-degeneracy conditions. Given a power series f, let us write (N), (I), (P ) to denote respectively the conditions that f is Newton non-degenerate, inner Newton non-degenerate

and partially non-degenerate at the origin. If p := char(k) is nonzero, then (P ) does not imply (I), e.g. xp + yp + xp+1 + yp+1 is partially non-degenerate at the origin, but it is inner Newton degenerate. This observation together with the discussion from section XI.4 implies the relations depicted in ﬁg. 2, where “(N )” denotes the condition that f is Newton non-degenerate and µ (f) < . µ0<∞ 0 ∞

(P ) (P )

/ / / ? ? ? (Nµ0<∞)/ (I) (Nµ0<∞)/ (I)

(A) zero characteristic (B) positive characteristic

FIGURE 2. Relation among non-degeneracy conditions

Problem XI.29. Determine if the question-marked implications from ﬁg. 2 are valid. We now show that in zero characteristic the implication (P ) (I) does hold in dimension 3.

⇒ ≤ k Proposition XI.30. Let f k[x1,...,xn]] be partially non-degenerate at the origin. If n 3 and char( ) is zero, then f is also inner∈ Newton non-degenerate. ≤

PROOF. Since all the ∂f/∂xj can not be identically zero on any axis, ND(f) satisfies the following property: (131) the distance from any axis to ND(f) can not be greater than 1. This leads to the possibilities of fig. 3 in the case that n = 2. If ND(f) is convenient, take Γ = ND(f), otherwise take Γ to be the union of ND(f) and edges (with appropriate slopes) from the end points of ND(f) to some points on the axes (e.g. the “dashed edges” in fig. 3). It is straightforward to check that f is inner Newton non-degenerate with respect to Γ.

FIGURE 3. Four possibilities for ND(f) in dimension two

Now assume that n = 3. If ND(f) does not touch (at least) one of the three coordinate hyperplanes, then f would be divisible by some xi. The partial non-degeneracy of f at the origin would then imply 3 that f = xig such that g(0) = 0, i.e. ND(f) = ei , where ei is the i-th standard unit vector in R . 6 { } In that case f would be inner Newton non-degenerate with any convenient diagram with a vertex at ei. Therefore we may assume that ND(f) touches every coordinate hyperplane. Let Γ be the Newton diagram of f + xN + xN + xN for some N 1. We claim that f is inner Newton non-degenerate with respect to 1 2 3 ≫ Γ. Indeed, let ∆ be an inner face of Γ and I 1, 2, 3 be such that ∆ RI = . We will check that (126) ⊆{ } ∩ 6 ∅ 178 XI.MILNORNUMBEROFAHYPERSURFACEATTHEORIGIN

′ I ′ ′ holds for ∆ and I. Let ∆ := ∆ R . If ∆ ND(f)= , then ∆ = Nei for some i. Assume without loss of generality that i =1. Since∩ ∆ ND(f∩) = (since∅ e.g. ∆ is an inner{ face),} it follows due to property (131) that there is an element on ∆ of∩ the form6 (∅k, 1, 0) or (k, 0, 1) for some k 0. Then xk j(f ), ≥ 1 ∈ ∆ and therefore V (j(f∆)) does not contain any point on x1-axis other than the origin, as required. It remains to consider the case that ∆′′ := ∆′ ND(f) = . If I = 1 or I = 3, it is straightforward to check that violation of (126) implies violation∩ of partial6 non-degeneracy∅ | | of| f|. So we may assume without loss of ′′ generality I = 1, 2 . Since ND(f) is not contained in the (x1, x2)-plane, it follows that dim(∆ ) 1. If ′ { } α1 α2 n ≤ dim(∆ )=0, then f∆ = cx1 x2 +x3g for some polynomial g, (α1, α2) Z≥0 0 and c =0. Itisthen ∈∗ I \{ } 6 ′′ clear that either ∂f∆/∂x1 or ∂f∆/∂x2 does not vanish at any point on (k ) . So assume dim(∆ )=1. It then follows that ∆′′ = ∆′, and if N is sufﬁciently large, then ∆ is in fact a (two dimensional) face of ND(f) containing ∆′, see ﬁg. 4a.

∂∆/∂x3

∂∆/∂x3 ∆ ′′ Λ Λ

ν ν ν Λ′ ∆′ (B) Case 1: ∂∆/∂x3 (C) Case 2: ∂∆/∂x3 does not (A) ∆ touches (x1,x2)-plane touch (x1,x2)-plane

FIGURE 4. Scenarios when dim(∆′′)=1

If ∂∆/∂x3 touches the (x1, x2)-plane, then applying the partial non-degeneracy condition(121) with ′ I ∗ I I = 1, 2 and ν equal to the “inner normal” to ∆ in R shows that V (j(f∆)) (k ) = . So assume that { } ∩ ∅ ∗ I ∂∆/∂x does not touch (x , x )-plane. In this case we claim that V (∂f ′ /∂x ,∂f ′ /∂x ) (k ) = . 3 1 2 ∆ 1 ∆ 2 ∩ ∅ Indeed, assume to the contrary that there is a common zero a = (a1,a2, 0) of ∂f∆′ /∂x1,∂f∆′ /∂x2 on ∗ I (k ) . The partial non-degeneracy implies that the intersection of ∂Γ/∂x3 with the (x1, x2)-plane is ′ nonempty, and if Λ is the face of the intersection determined by ν, then (∂f/∂x3)Λ′ does not vanish at ′ a. It is straightforward to check that there is a face Λ of ∂Γ/∂x3 which contains Λ and also intersects ′′ ∂∆/∂x3. Let Λ := Λ ∂∆/∂x3. If (∂f/∂x3)Λ′′ does not vanish after substituting x1 = a1 and x2 = a2, ∩ ∗ then one can ﬁnd a zero of (∂f/∂x ) of the form (a ,a ,b) for some b k , and this would contradict 3 Λ 1 2 ∈ partial non-degeneracy condition (121) with I = 1, 2, 3 and ν being an inner normal to Λ in R3. On { } ′′ the other hand, if (∂f/∂x3)Λ′′ vanishes after substituting x1 = a1 and x2 = a2, then Λ must be an edge parallel to ∆′ and the partial non-degeneracy condition (121) is violated with I = 1, 2, 3 and ν being an ′′ 3 ∗ I { } inner normal to Λ in R . This proves that V (∂f∆′ /∂x1,∂f∆′ /∂x2) (k ) = , as claimed. But then it ∗ I ∩ ∅ is clear that V (j(f )) (k ) is also empty, as required. ∆ ∩ 6.3. Conditions for validity of Kushnirenko’s formula in positive characteristic. Consider the

case that p := char(k) > 0. The characterization of the conditions under which µ0( ) = ν(ND( )) established in proposition XI.18, in particular condition (2b) of proposition XI.18, is notAexplicit. It wouldA be interesting to ﬁnd simple criteria under which this condition holds. The criterion from proposition XI.21 is sufﬁcient, but not necessary, e.g. if f = x + yp, then (128) fails (and f is not Newton non-degenerate as well), but µ0(f) = ν(ND(f)) = 0. Note that f is inner Newton non-degenerate. Y. Boubakri, G.-M. Greuel and T. Markwig claim in [BGM12, Theorem 3.5] that µ0(f) = ν(ND(f)) if f is inner Newton non-degenerate, i.e. Wall’s result [Wal99, Theorem 1.6] extends to positive characteristics.

Problem XI.31. Determine necessary and sufﬁcient conditions for µ0(f) = ν(ND(f)) to hold when

char(k) > 0. In particular, is inner Newton non-degeneracy necessary for this equality? 6. OPEN PROBLEMS 179

6.4. Monotonicity of Newton number. Theorems XI.3 and XI.15 and the monotonicity of intersection multiplicity (remark IX.3) implies that the Newton number ν(Γ) of a diagram Γ grows monotonically with Γ. Even though the arguments in this implication involve nontrivial algebraic geometry, the monotonicity of Newton numbers is a purely “elementary” convex geometric statement. It is a problem of V. I. Arnold [Arn04, Problem 1982-16] to find an elementary proof of this statement. S. K. Lando wrote in the commentary of that problem that he gave an elementary proof for n = 2. For n = 3 an elementary proof was given by S. Brzostowski, T. Krasiński and J. Walewska [BKW17]. Since the monotonicity of the formula (73) for intersection multiplicity is obvious (see remark IX.3), one possible strategy would be to find an elementary proof of the following identity (which is an immediate consequence of theorems XI.3 and XI.15): I I I ∗ ν(Γ) = [Γ , Γ ,..., Γ ] [π (Γ ′ ),...,π (Γ ′ )]0 1 j2 j|I| 0 [n]\I j1 [n]\I jn−|I| T × I∈XA,1 n n where the right hand side is as in theorem XI.15 with Γj := ND( α ej : α Γ+ R≥0, α ej Z≥0 ), j =1,...,n. { − ∈ − ∈ } CHAPTER XII

Beyond this book

In this ﬁnal chapter we outline some of the natural directions of pursuit for a reader of this book, and some further connections with topics of current research.

1. Toric varieties An obvious direction for further pursuit is to study toric varieties. Recent introductions to the theory of toric varieties include [Oda88, Ful93, CLS11]. Unlike this book, in these expositions toric varieties are deﬁned via fans, which reveals their combinatorial structure more cleanly. The combinatorics makes delightful appearances even in dimension two, e.g. continued fractions appear in resolutions of singularity [Ful93, Section 2.6] and Pick’s formula appears in intersection theory [Ful93, Section 5.3]. A main attrac- tion of toric varieties is their “computability,” due to which, in the words of Fulton, “toric varieties have provided a remarkably fertile testing ground for general theories” of algebraic geometry. We give a recent example. The (strong) factorization conjecture of Hironaka states that “Any birational map f : X 99K Y

of smooth complete varieties (over an algebraically closed ﬁeld k) can be factored into a succession of blow-ups at smooth centers followed by a succession of blow-downs at smooth centers.” The weak version of the factorization conjecture, formulated by K. Miyake andT.Oda[Oda78] states that any such f, which is an isomorphism over an open set U, can be factored as

f f fn−1 X = X 99K0 X 99K1 99K X = Y 0 1 ··· n where each Xi is a smooth complete variety and fi is a blow-up or blow-down at a smooth center which is an isomorphism over U. The toric version of the weak factorization conjecture was proven independently by R. Morelli [Mor96] and J. Włodarczyk[Wł97], and based on this toric case Włodarczyklater provedthe general case of the weak factorization conjecture in [Wł03]. The strong factorization conjecture remains unproven even in the toric case. S. Da Silva and K. Karu [DK11] proposed a combinatorial algorithm which they conjecture solves the toric case of the strong factorization conjecture.

2. Newton-Okounkov bodies The Newton-Okounkov body is a very fruitful recent generalization of the Newton polytope of a (Lau- rent) polynomial. It was originally introduced by A. Okounkov, who associated in [Oko96, Oko03] convex bodies to ample divisors on a smooth variety. R. Lazarsfeld and M. Mustata [LM09], and independently K. Kaveh and A. Khovanskii [KK12] made further generalizations of this construction. Since then there have been numerous articles on these structures. The construction of Kaveh and Khovanskii in particular is very elementary and leads to simple proofs of nontrivial results from intersection theory and also convex geometry. The series of articles by Kaveh and Khovanskii on Newton-Okounkov bodies, in particular [KK10, KK12], would be excellent reading material for a reader of this book.

3. Bezout´ problem Another natural extension of the material of this book would be a deeper study of the B´ezout problem of counting numbers of solutions of systems of polynomials. A. Khovanskii studied higher dimensional analogues of this problem. In particular, he computes in [Kho78] the Euler characteristics and in [Kho16] the number of irreducible components of generic complete intersections on the torus. P. Philippon and M. Sombra [PS08] studied a parametrized version of Bernstein’s theorem over a nonsingular curve, and gave an answer in terms of an associated “mixed integral,” which is a generalization of mixed volume to concave functions. In a sequel to this book we describe an inductive algorithm to compute the precise

180 3. BEZOUT´ PROBLEM 181 number (counted with multiplicity) of solutions of any given system of n polynomials in n variables starting from the estimate from chapter X. APPENDIX A

Notation

:= is deﬁned as Euclidean length || · || disjoint union ν, , ν (Rn)∗ the function induced by ν on Rn `h ·i ∈ n β, , β R dot product with β h ·i ∈ η⊥ α Rn : η, α =0 f, g, . . . , ideal{ ∈ generatedh byi f, g,} . . .

h i k

k[[x]] ring of formal power series in x with coefﬁcients in k

k((x)) field of Laurent series in x with coefficients in k k[X] ring of rational functions on the algebraic variety X defined over [n], n Z 1, 2,...,n

∈ ∗ n ∗ m { } k [T ],T : (k ) ( ) n m matrix of exponents of coordinates of T → α× xα, α = (α ,...,α ) x 1 xαn 1 n 1 ··· n 1n identity matrix of size n n aff( ) affine hull × cone(· S) cone generated by S conv( ) convex hull fund(H· ) volume of any fundamental lattice parallelotope of H gcd(a, b, ) greatest common positive divisor of the nonzero elements from a, b, ··· ··· Inν ( ) initial form of a (Laurent) polynomial or minimizing face of a polyhedron lcm(·a, b, ) lowest common positive multiple of the nonzero elements from a, b, ld ··· leading form of a (Laurent) polynomial or maximizing face of a polyhedron··· max (ν) max ν, α : α P {h i ∈ P} minP (ν) min ν, α : α MV mixed{h volumei ∈ P} ′ MVν normalized mixed volume n( ) nilradical ND(· ) Newton diagram NP(·) Newton polytope · local ring of the variety X at its subvariety Z OX,Z Q>0 q Q : q> 0 { ∈ } Q≥0 q Q : q 0 { ∈ ≥ } R>0 r R : r> 0 { ∈ } R≥0 r R : r 0 relint( ) relative{ ∈ interior≥ } Supp(f· ) support of f tr. d.k( ) transcendence degree over (a field) k V (f, g,· . . .) the set of zeroes of f, g, . . . ′ VolH normalized lattice vlume on an affine subspace H Vol′ Vol′ , where H := α Rn : ν, α =0 ν H { ∈ h i } Z>0 0, 1, 2,..., { } Z≥0 1, 2,..., { }

182 APPENDIX B

Miscellaneous commutative algebra

By a “ring” in this book we mean a commutative ring with identity, and an integral domain is a ring without zero-divisors. In this chapter we brieﬂy recall several notions related to rings.

1. Fundamental theorem of finitely generated modules over a PID A principal ideal domain (in short, PID) is an integral domain such that every ideal is principal. An irreducible element or prime of a PID is a nonzero element which is a non-unit1 and which can not be expressed as a product of two non-units. The following is usually called the “fundamental theorem” or the “structure theorem” of finitely generated modules over a PID. Its proof can be found in any standard introductory abstract algebra text, e.g. [DF04]. THEOREM B.1. Let M be a finitely generated module over a PID R. Then M = Rr R/ pα1 R/ pα2 R/ pαk ∼ h 1 i h 2 i · · · h k i for some r, α ,...,α 0 and primesMp ,...,p M. M 1 k ≥ 1 k 2. Prime and Maximal ideals, Noetherian rings, Zero-divisors

Given elements g1,...,gk ofa ring R, we write g1,...,gk to denote the ideal generated by R. Recall that a prime ideal of R is a proper ideal p such that hfg p impliesi either f or g is in p. A maximal ideal of R is a proper ideal m such that the only proper ideal of∈ R containing m is m itself. It is straightforward to check that every maximal ideal is prime. The following are some other basic properties of prime and maximal ideals - see e.g. [AM69, Chapter 1] for their proof. THEOREM B.2. Let R be a nonzero ring (remember that by “ring” we mean a commutative ring with identity). (1) Given any non-unit g R, there is a maximal ideal of R containing g. ∈ (2) If q is an ideal of R contained in the union of finitely many prime ideals p1,..., pk of R, then q p for some j. ⊆ j (3) If p is a prime ideal of R containing the intersection of finitely many ideals q1,..., qk, then p q for some j. If p = k q , then p = q for some j. ⊇ j j=1 j j (4) The intersection of all prime ideals of R is precisely its nilradical, i.e. the ideal consisting of all nilpotent2 elements of R . T Assertion (2) of theorem B.2 is sometimes referred to as “prime avoidance”; see [Eis95, Lemma 3.3] for a stronger variation and geometric interpretation. A ring R is called Notherian if every nonempty set of ideals in R has a maximal element with respect to , i.e. thereisan ideal q such that the only elementI of containing q is q itself. It is straightforward to⊆ check that the ring of integ∈ Iers, or more generally, any PIDI is Noetherian. Hilbert’s basis theorem (theorem III.1) implies that quotients of polynomial rings in finitely many variables over a field are also Noetherian. Most of the rings that appear in this book are Noetherian. The ring R[x , x , ] of polynomials in infinitely many variables x , j 1, over a ring R is 1 2 ··· j ≥ not Noetherian, since e.g. the sequence of ideals x1,...,xn , n 1, does not have any maximal element. A zero-divisor in a ring R is a nonzero element fh such that fgi =0≥ for some nonzero g. Given g R, one writes (0 : g) for the set of all f R such that fg =0; in other words (0 : g) is the annihilator of∈ g. It is clear that the set of zero-divisors∈ is precisely the union of the annihilators of the nonzero elements of R.

1A unit of a ring is an element which has a multiplicative inverse. 2g ∈ R is called nilpotent if there is n ≥ 1 such that gn = 0.

183 184 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

Proposition B.3. The set of zero-divisors in a Noetherian ring R is a union of prime ideals which are of the form (0 : g) for some g R. ∈ PROOF. Let be the set of all ideals of a Noetherian ring R of the form (0 : g), where g varies over nonzero elementsI of R. By Noetherianity of R, each q is contained in a maximal element p . We will show that p is prime. Indeed, pick g R such that∈ Ip = (0 : g). If p is not prime, then there∈ I is ∈ f1f2 R p such that f1f2 p, i.e. f1f2g =0. But then (0 : f2g) properly contains p. This contradiction (to the∈ maximality\ of p) completes∈ the proof.

In corollary B.17 below we prove a stronger version of proposition B.3.

3. Nakayama’s lemma In this section we present a brief discussion of Nakayama’s lemma following [AM69]. Let R be a ring, and j be its Jacobson ideal, i.e. the intersection of all the maximal ideals of R. Lemma B.4. 1+ f is a unit in R for each f j. ∈ PROOF OF LEMMA B.4. Pick f j. If 1+ f is not a unit, then there is a maximal ideal m of R that contains 1+ f (theorem B.2, assertion∈ (1)). Since m contains both f and 1+ f, it contains 1, which is a contradiction.

Lemma B.5 (Nakayama). Let M be a ﬁnitely generated R-module and q be an ideal of R contained in the Jacobson ideal of R. If qM = M, then M =0.

PROOF. Assume M = 0 and let u ,...,u be a minimal set of generators of M. Since u qM, 6 1 k k ∈ there is an equation of the form u = k f u , with each f q. It follows that (1 f )u = k j=1 j j j ∈ − j k k−1 u f . Since 1 f is a unit in R (lemma B.4), it follows that u is an R-linear combination of j=1 j j − j P k u1,...,uk−1. This means that M can be generated by u1,...,uk−1, a contradiction. P

Let R be a local ring with (the unique) maximal ideal m, and k = R/m be its residue field. Let M be a finitely generated R-module. Since M/mM is annihilated by m, it is a finitely generated R/m-module,

i.e. a ﬁnite dimensional vector space over k.

Corollary B.6. Let m ,...,m M be such that their images span M/mM as a vector space over k. 1 k ∈ Then m1,...,mk generate M as a module over R.

PROOF. Let N be the submodule of M generated by m1,...,mk. Since the image of N in M/mM is all of M/mM, it follows that M = N + mM. But then m(M/N) = (N + mM)/N = M/N. Lemma B.5 then implies that M = N, as required.

4. Local rings A local ring is a ring with a unique maximal ideal. If S is a multiplicative closed subset of a ring R, then the localization RS of R with respect to S is the equivalence class of “fractions” f/g : f R, g S under the equivalence relation that f/g f ′/g′ provided (fg′ f ′g)h = 0 for{ some h ∈ S. It∈ is } ∼ − ∈ easy to check that RS is a commutative ring with respect to the usual rules of addition and multiplication of fractions. Two cases of localizations are especially relevant for our purpose: k Case 1: S = f k for some f R. In this case we denote RS by Rf , and say that Rf is the • localization of{R at} f. ∈ Case 2: S = R p for some prime ideal p of R. In this case we denote R by Rp, and say that • \ S Rp is the localization of R at p.

It is straightforward to check that [the equivalence class of] every element of S is invertible in RS (unless it is the zero ring), and every ideal I of of R is generated by R I. It follows that if R is Noetherian, S ∩ then so is RS. We refer to [AM69, Chapter 3] for a lucid discussion of these and other basic properties of localizations of commutative rings. 7. PRIMARY DECOMPOSITION 185

5. Discrete valuation rings A discrete valuation on a ﬁeld K is a surjective map ν from K onto Z such that ν(0) = , ν(fg)= ν(f)+ ν(g) and ν(f + g) min ν(f),ν(g) for each f,g K.∪ It{∞} is straightforward to check∞ that the set of all f K such that ν(f≥) 0{is a subring} of K; it is called∈ the valuation ring of ν. ∈ ≥ Example B.7. A basic example of a discrete valuation is the order of rational functions in one variable. j Recall that the order of f = a t k[t], where t is an indeterminate, is ord(f) := min j : a = 0 , j j ∈ { j 6 } and the order can be extended to rational functions by deﬁning ord(f/g) := ord(f) ord(g). The P −

valuation ring of ord is precisely the localization of k[t] at the maximal ideal generated by t, i.e. the k subring f/g : f,g k[t], g(0) =0 of (t). { ∈ 6 } A discrete valuation ring is an integral domain which is the valuation ring of a discrete valuation on its field of fractions. We now record some properties of discrete valuation rings: their verification is straightforward, and is left as an exercise. Proposition B.8. Let ν be a discrete valuation on a field K, and R := f K : ν(f) 0 be the valuation ring of ν. { ∈ ≥ } (1) The units of R are precisely the elements f in K with ν(f)=0. (2) m := f K : ν(f) > 0 is the unique maximal ideal of R; in particular, R is a local ring. (3) If f is{ any∈ element in m such} that ν(f)=1, then m is the (principal) ideal of R generated by f. More generally, every proper ideal of R is a principal ideal generated by f k for some k 1. ≥

(4) If R contains a ﬁeld k which is isomorphic to R/m, then for each g R, R/gR is a vector space ∈

over k of dimension ν(g). (5) R uniquely determines ν, i.e. if ν′ is a discrete valuation on K such that R is also the valuation ring of ν′, then ν′ = ν. Proposition B.8 in particular implies that a discrete valuation ring R is a local ring whose maximal ideal is principal. A parameter of R is a generator of its maximal ideal.

6. Krull dimension

A chain of prime ideals of length n 0 in a ring R is a finite sequence p0 ( p1 ( ( pn of prime ideals of R. The Krull dimension of R is≥ the supremum of the lengths of all chains of prime··· ideals in R. Example B.9. Since the only maximal ideal of a nontrivial field F is the zero ideal, the Krull dimension of F is zero. Each of the following rings has Krull dimension one: the ring of integers, the ring of polynomials in one variable over a field, a discrete valuation ring.

7. Primary Decomposition A proper ideal q of a ring R is primary if fg q implies either f q or gk q for some k 1. ∈ ∈ ∈ ≥ Proposition B.10. Let q be an ideal of a ring R and √q be the radical3 of q. (1) If q is primary, then √q is prime. (2) If √q is maximal, then q is primary. (3) Write p := √q. Assume q and q′ are primary ideals of R such that √q′ = √q = p. Then q q′ ∩ is also primary and √q q′ = p. ∩ PROOF. The proof of the ﬁrst assertion is straightforward from the deﬁnition of a primary ideal - it is left as an exercise. For the second assertion, assume √q is maximal. Claim B.10.1. If g √q, then g is a unit in R/q. 6∈ PROOF. Indeed, there is f √q and u, v R such that gu + fv = 1. If k 1 is such that f k q, ∈ ∈ ≥ ∈ then the relation (gu + fv)k =1 reduces to gu′ =1 in R/q for an appropriate u′ R. ∈

3The radical of q is the ideal consisting of all g ∈ R such that gk ∈ q for some k ≥ 1. 186 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

Pick f,g R such that fg q. If g √q, then claim B.10.1 implies that f q. Therefore q is ∈ ∈ 6∈ ∈ primary. It remains to provethe third assertion. It is straightforward to check that √q q′ = √q √q′ = p. Now pick fg q q′ such that f q q′. Then f is not in either q or q′, and since∩ both are primary∩ with radical p, it follows∈ ∩ that g p = √6∈q ∩q′, so that q q′ is also primary. ∈ ∩ ∩ 2 Example B.11. The ideal q of k[x, y] generated by x and xy is not primary, since xy q, but x q and ∈ 6∈ y √q = x . In particular, assertion (2) of proposition B.10 does not hold if “maximal” is replaced by “prime.”6∈ h i

n Example B.12. The primary ideals of k[t] are the zero ideal and the ideals generated by f , n 1, for irreducible polynomials f. More generally, the nonzero primary ideals of a PID are precisely the≥ ideals generated by powers of irreducible elements.

2 Example B.13. The ideal q of k[x, y] generated by x and y is primary (proposition B.10, assertion (2)). The radical of q is the maximal ideal m generated by x, y. Note that m ) q ) m2, so that q is not a prime-power. A primary decomposition of an ideal q is an expression k (132) q = qj j=1 \ of q as a ﬁnite intersection of primary ideals qj . Given such a primary decomposition, excluding any redundant ideal from the right hand side and then grouping the ideals with the same radical, it can be ensured that q can not be expressed as an intersection of less than k of the q , and • j the radicals √qj of qj are distinct (due to assertion (3) of proposition B.10); • in that case we say that (132) is a minimal primary decomposition, and that the prime ideals √qj are associated with q (that √qj are prime follows from proposition B.10). Note that k √q = √qj j=1 \ is a (not necessarily minimal) primary decomposition of the radical √q of q; in particular, √q is a ﬁnite intersection of prime ideals associated with q. 2 Example B.14. Let q := x , xy k[x, y]. We saw in example B.11 that q is not primary. Each of the following is a minimal primaryh decompositioni ⊂ of q (since x is prime and the radical of the other ideal in the decomposition is maximal) h i q = x x2,y = x x2,xy,yk (k 1) h i ∩ h i h i ∩ h i ≥ In particular, minimal primary decompositions are in general not unique4. Proposition B.15. Let q be an ideal in R which has a primary decomposition. Then every prime ideal in R containing q contains one of the prime ideals associated with q. The minimal ideals among those associated with q are precisely the minimal elements in the set of all ideals in R containing q.

PROOF. Follows immediately from assertion (3) of theorem B.2. The following is a fundamental property of Noetherian rings; it is a combination of [AM69, Theorem 4.5, Proposition 7.17].

THEOREM B.16. Assume R is Noetherian. Then every proper ideal has a primary decomposition. In particular, every radical ideal is a ﬁnite intersection of prime ideals. The prime ideals associated with an ideal q are precisely those prime ideals of R which occur in the set of ideals (q : f) := g R : fg q as f varies over R, and hence are uniquely determined by q. { ∈ ∈ }

4However, the isolated primary ideals (i.e. primary ideals whose radicals are minimal among the radicals of all primary ideals appearing in the decomposition) in a primary decomposition are in fact unique (e.g. the ideal hxi will appear in every primary decomposition of hx2,xyi) - see [AM69, Corollary 4.11]. 8. LENGTH OF MODULES 187

Corollary B.17. The set of zero-divisors in a Noetherian ring R is the union of the prime ideals associated with the zero ideal. Every prime ideal of R contains a prime ideal associated with the zero ideal. The minimal ideals among the prime ideals associated with the zero ideal are precisely the minimal elements of the set of minimal5 prime ideals of R. In particular, every element of a minimal prime ideal of R is a zero-divisor.

PROOF. The ﬁrst assertion follows from combining proposition B.3 and theorem B.16. The remaining assertions then follow from proposition B.15.

Corollary B.18. Let f be a non zero-divisor in a ring R and p be a minimal prime ideal of R. If R is Noetherian, then (the image of) f remains a non zero-divisor in R/p.

PROOF. Corollary B.17 implies that f p. Therefore, if g p, then fg p. Therefore f is not a zero-divisor in R/p, as required. 6∈ 6∈ 6∈

8. Length of modules Let M be a module over a ring R.A composition series of M of length n 0 is a sequence ≥ (133) M = M0 ) M1 ) ) Mn =0 ··· of R-submodules which is “maximal,” or equivalently, each quotient Mi−1/Mi, 1 i n, is simple, (that is, has no non-zero proper submodule). Not every module has a composition series.≤ In≤ fact the following are equivalent ([AM69, Proposition 6.8]): (1) M has a composition series, (2) M satisfies both ascending and descending chain conditions6. If M has a composition series, then all composition series of M have the same length ([AM69, Proposition 6.7]). The length l(M) of M is defined to be infinite if it has no composition series; otherwise l(M) is the length of any composition series of M. If(133) is a composition series of M, then each Mi−1/Mi is isomorphic to R/m for some maximal ideal m of R

Proposition B.19. Let M be a module over a ring R with a composition-series M = M0 ) M1 ) ) M =0. Fix i, 1 i n. ··· n ≤ ≤ (1) Each Mi−1/Mi is isomorphic to R/m for some maximal ideal m of R. (2) Let m be as in assertion (1). If there is a subﬁeld k of R which maps isomorphically onto R/m, then Mi−1/Mi is a one dimensional vector space over k.

PROOF. Pick any nonzero element m¯ M /M . Since M /M is simple, the map R ∈ i−1 i i−1 i → Mi−1/Mi given by f fm is a surjective map and its kernel must be a maximal ideal of R. This implies the ﬁrst assertion.7→ The second assertion follows immediately from the ﬁrst.

Corollary B.20. Assume R is a local ring with maximal ideal m, and there is a ﬁeld k R which maps isomorphically onto R/m. Then for any R-module M, the length of M as an R-module⊆ is equal to the length of M as a k-module, which in turn is equal to the dimension of M as a k-vector space.

Proposition B.21. Let p be a minimal prime ideal of a Noetherian ring R. Then the localization Rp of R at p has a ﬁnite length as an Rp-module.

PROOF. Since p is a minimal prime ideal of R, the ideal generated by p is the unique prime ideal 7 of Rp, i.e. the Krull dimension of Rp is zero. Since Rp is also Noetherian, it follows that it is Artinian ([AM69, Theorem 8.5]) as well. This implies that the length Rp as a module over itself is ﬁnite ([AM69, Proposition 6.8]), as required.

5A minimal prime ideal p of a ring R is a prime ideal p such that the only prime ideal of R contained in p is p itself. 6 M satisﬁes ascending (respectively, descending) chain condition if for every chain M0 ⊆ M1 ⊆ ··· (respectively, M0 ⊇ M1 ⊇··· ) of submodules of M, there is k such that Mj = Mk for all j ≥ k. 7A ring R is Artinian if it satisﬁes the descending chain condition as a module over itself, i.e. if for every descending chain of ideals I0 ⊇ I1 ⊇··· , there is k such that Ij = Ik for all j ≥ k. 188 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

9. Graded rings By a graded ring in this book we mean a commutative ring S together with a decomposition S =

d∈Z Sd of S into a direct sum of abelian groups (with respect to the addition operator of S) such that SdSe Sd+e for each d, e Z. For each d Z, an element of Sd is called a homogeneous element of Ldegree⊆d. Every f S has a∈ unique homogeneous∈ decomposition f = f where f S for each ∈ d∈S d d ∈ d d; we say that the fd are the homogeneous components of f. An ideal of S is called homogeneous ideal if for every f I, every homogeneous component of f is also in I. P ∈ Example B.22. The most basic example of a graded ring is the polynomial ring, i.e. the ring of polynomials in a collection of indeterminates over a ﬁeld, where the grading is induced by the usual degree of polynomials, and the homogeneous elements are simply the homogeneous polynomials.

Proposition B.23. Let I be an ideal of a graded ring S = d∈Z Sd. (1) I is homogeneous if and only if it is generated by homogeneous elements. L (2) If I is homogeneous, then S/I is a graded ring with associated decomposition S/I = S /(I d∈Z d ∩ Sd). (3) If I is homogeneous, then the radical of I is also homogeneous. L

PROOF. The ﬁrst assertion is straightforward, and the second assertion follows immediately from the ﬁrst. For the third assertion assume to the contrary that √I is not homogeneous. Then there is nonzero f = f √I such that none of the nonzero homogeneous components of f is in √I. Pick k 1 such d d ∈ ≥ that k . If is the smallest integer such that is nonzero, then k is a homogeneous component f I d0 fd0 fd0 kP ∈ k of f . Since I is homogeneous, f I, so that fd √I, which gives the required contradiction. d0 ∈ 0 ∈ 2 3 Example B.24. I := x + x , x is not a homogeneousideal of k[x] (with respect to the grading given by h i degree), however √I = x is homogeneous. Therefore the converse of assertion (3) of proposition B.23 is not true in general. h i The following result is straightforward - its proof is left as an exercise.

Proposition B.25. Let f be a homogeneous element in a graded ring S = d∈Z Sd and Sf be the localization of S at f. L (1) Sf is also a graded ring. The homogeneous elements of degree e, where e is any integer, of Sf k are g/f : k Z, g Skd+e . { ∈ ∈ } (2) Assume in addition that S is a finitely generated algebra over a subring R of S0. Then Sf and S(f) are also finitely generated algebras over R. Given graded rings T,S, a graded ring homomorphism φ : T S is a ring homomorphism that → preserves degrees; if in addition φ is a k-algebra homomorphism for some field k T0, we say that it is a graded k-algebra homomorphism. If f is a homogeneous element of S, the subring⊆ of the localization Sf consisting of homogeneous elements of degree zero is denoted by S(f). The following result is also straightforward to verify using the definitions.

Proposition B.26. Let φ : T S be a homomorphism of graded k-algebras for some field k T0. For each homogeneous g T , φ induces→ a k-algebra homomorphism T S . ⊆ ∈ (g) → (φ(g)) 10. Macaulay’s Unmixedness Theorem In this section we give a proof, following [ZS75b, Section VII.8], of the unmixedness theorem of F. S. Macaulay for polynomial rings R := k[x1,...,xn] over an arbitrary field k. The statement of this theorem requires the notion of dimension8 dim(q) of an ideal q of R. If q is prime, then dim(q) is the transcendence degree (of the field of fractions) of R/q over k. In general dim(q) is the maximum of the dimensions of the prime ideals associated with q. Since the minimal elements of the set of prime ideals of R containing q are precisely the minimal elements of the set of ideals associated with q (proposition B.15), it follows that dim(q) is the maximum of the dimensions of the minimal prime ideals containing q. One

8It is in fact the same as the dimension of the algebraic variety determined by q as deﬁned in section III.8; here we give a purely algebraic treatment. 10. MACAULAY’S UNMIXEDNESS THEOREM 189 of the most fundamental properties of dimensions is given by Krull’s prime ideal theorem (theorem III.3)- the following result is its straightforward consequence.

Proposition B.27. Let k be an arbitrary ﬁeld, and q be a proper ideal of k[x1,...,xn] generated by f1,...,fm, m n. Then dim(q) n m. Ifin addition dim(q)= n m, then dim( f1,...,fj )= n j, for each j =1,...,m≤ . ≥ − − h i − The following is a special case of Macaulay’s unmixedness theorem (theorem B.29).

q1 qr Proposition B.28. Let g = g1 gr R := k[x1,...,xn], where gj are irreducible (non-constant) polynomials and q are positive integers.··· ∈ Then h R is a zero-divisor in R/ g if and only if h is divisible j ∈ h i by some gj .

PROOF. This immediately follows from the fact that R is a unique factorization domain.

THEOREM B.29 (Macaulay’s unmixedness9 theorem [ZS75b, Theorem VII.26]). Let k be an arbitrary ﬁeld, and f1,...,fm k[x1,...,xn], m n, be such that dim( f1,...,fm ) = n m. Then for each j =1,...,m, f is a non∈ zero-divisor in k≤[x ,...,x ]/ f ,...,fh . i − j 1 n h 1 j−1i PROOF. Write R := k[x1,...,xn]. We proceed by induction on m. The case of m = 1 is obvious, and the case of m = 2 follows from proposition B.28. So assume m 3. If fm is a zero-divisor in R/ f ,...,f , then it is contained in a prime ideal p associated with q≥:= f ,...,f . h 1 m−1i h 1 m−1i Claim B.29.1. dim(p)

Let d := dim(p). Then there are x1,...,xd which are algebraically independent over k in R/p. This ˜ means k[x ,...,x ] p = 0. Let R := k(x ,...,x )[x ,...,x ]. Given an ideal r of R, we write ˜r 1 d ∩ 1 d d+1 n for the ideal of R˜ generated by r. Claim B.29.2. Let r be a prime ideal of R contained in p and ˜r be the ideal of R˜ generated by r. (1) dim(˜r) = dim(r) d. (2) If r is associated with− q, then ˜r is associated to ˜q.

PROOF. The ﬁrst assertion follows immediately from choosing a transcendence basis (over k) of R/r containing x ,...,x . Now pick f R such that r = (q : f). Then it is straightforward to check that 1 d ∈ ˜r = (q˜ : f). Since ˜r is a prime ideal of R˜, the second assertion follows from theorem B.16.

In claim B.29.2 taking r = q′, where q′ is the ideal from the proof of claim B.29.1, implies that q˜ is a proper ideal of R˜ with dimension n d (m 1). Since ˜q is generated by m 1 elements, proposition B.27 implies that dim(q˜) is precisely≤ − n −d (m− 1). On the other hand, applying− claim B.29.2 − − − with r = p shows that p˜ is a zero dimensional ideal of R˜ which is also associated with ˜q. But then claim B.29.3 below gives a contradiction and proves theorem B.29. Claim B.29.3. Let b a be ideals in R := k[x ,...,x ] such that ⊂ 1 r (1) a is prime, (2) dim(a)=0, (3) b is generated by m 1 elements, and (4) dim(b)= r (m −1). − − If r m, then a is not associated with b. ≥

9An ideal is unmixed if all its associated prime ideals have the same dimension. Theorem B.29 is called the “unmixedness theorem” due to this equivalent formulation: “Let q be an ideal of k[x1,...,xn] generated by m polynomials. If dim(q) = n − m, then q is unmixed.” The equivalence of this statement with theorem B.29 follows in straightforward manner from corollary B.17 and Krull’s principal ideal theorem (theorem III.3). 190 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

To complete the proof of theorem B.29 it remains to prove claim B.29.3. Pick a set g1,...,gm−1 of generators of b. Let c be the ideal of R generated by g1,...,gm−2 (recall that m 3). Proposition B.27 implies that dim(c) = r (m 2). If h R is not in any of the minimal≥ prime ideal associated − − ∈ with c, the induction hypothesis then applies to g1,...,gm−2,h and implies that h is not a zero-divisor in R/c. Theorem B.16 and proposition B.27 then imply that all prime ideals associated to c have dimension r m +2; enumerate these ideals as p1,..., ps. Let ps+1,..., ps′ be the minimal (i.e. (r m + 1)- dimensional)− prime ideals associated to b. Since r m, each p has dimension greater than zero.− We will ≥ j use the following result to construct a special type of polynomial which does not belong to any pj .

LEMMA B.29.4. Given ﬁnitely many prime ideals p1,..., ps′ of positive dimensional prime ideals in R := k[x ,...,x ], there is e, 1 e s′, and a polynomial h(x ,...,x ) such that the image of 1 r ≤ ≤ 1 e−1 ye := xe + h(x1,...,xe−1) is transcendental over k in R/pj for each j.

PROOF. Pick the smallest integer i1 such that xi1 is transcendental over k in R/pj for some j. Reorder the pj in a way that xi1 is transcendental over k in R/p1,...,R/ps1 and algebraic over k in R/pj for ′ j>s1. If s1 i1 such that xi2 is transcendental over k in R/pj for ′ some j>s1. Then reorder ps1+1,..., ps in a way that xi2 is transcendental over k in R/ps1+1,...,R/ps2 and algebraic over k in R/pj for j>s2. Since each pj is positive dimensional, we can continue in this ′ way until there is an integer t such that st = s . In particular, ′ (a) xit is transcendental over k in R/pj for each j = st−1 +1,...,st = s . Now ﬁx j, s +1 j s . We claim that there is at most one integer q such that x + xq is t−2 t−1 it it−1 ≤ ≤ q1 q2 algebraic over k in R/pj. Indeed, otherwise there would be q1 = q2 such that x x is algebraic over 6 it−1 − it−1 k in R/pj; but this is impossible since by construction xit−1 is transcendental over k in R/pj. Therefore (b) there is a positive integer such that qt−1 is transcendental over in p for each qt−1 xit + xit−1 k R/ j j = st−2 +1,...,st−1. Continuing in this way one can choose integers q , q ,...,q such that for each t′ =1,...,t 1, t−1 t−2 1 − qt−1 qt−t′ (c) xi + x + + x is transcendental over k in R/pj for each j = st−t′−1 +1,...,st−t′ . t it−1 ··· it−t′ ′ ′ Since for each t = 1,...,t and each ist′−1, it follows qt−1 q1 that xi + x + + x satisﬁes the conclusion of the lemma with e := it. t it−1 ··· i1

We go back to the proof of claim B.29.3. Let ye be the polynomial from lemma B.29.4. Note that R = k[x1,...,xe−1,ye, xe+1,...,xr]. Since dim(a)=0, the image of ye in R/a is algebraic over k. Let f := φ(y ) k[y ] be the polynomial with the minimum degree in y which is zero in R/a. Note that e ∈ e e (i) f a, and ∈ (ii) f is not in any pj , since the image of ye is transcendental over k in R/pj. Claim B.29.5. g is not a zero-divisor in R/ c,f . m−1 h i PROOF. Note that R/ f ∼= k(¯ye)[x1,...,xe−1, xe+1,...,xr], where y¯e is the image of ye in R/a. Let b¯ be the ideal of R/ f hgeneratedi by b. Krull’s principal ideal theorem (theorem III.3) and observation (ii) above imply that dim(h ib¯) = dim(b) 1= r (m 1) 1= r 1 (m 1). Since b¯ is generated by m 1 elements (namely the images g¯− of g , 1− i −m −1) in the− polynomial− − ring R/ f over k(¯y ), − i i ≤ ≤ − h i e the induction hypothesis implies that g¯m−1 is not a zero-divisor in (R/ f )/ g¯1,..., g¯m−2 ∼= R/ f, c , as required. h i h i h i

Returning to the proof of claim B.29.3, assume to the contrary that a is associated with b. Then there is g R such that a (b : g) (theorem B.16); in particular gf b = c,gm−1 . Therefore there is h R such∈ that gf g ⊇h c, i.e. g h c,f . Claim B.29.5 then∈ impliesh thatih c,f , i.e. h af∈ c − m−1 ∈ m−1 ∈ h i ∈ h i − ∈ for some a R. It follows that gf gm−1af = (g gm−1a)f c. Since f is not in any prime ideal associated with∈ c (observation (ii)), f−is not a zero divisor− in R/c ∈(theorem B.16 and corollary B.17). It follows that g gm−1a c, so that g c,gm−1 = b. But then (b : g) = R a. This contradiction ﬁnishes the proof− of claim∈B.29.3 and theorem∈ h B.29i. 6⊆ 11. PROPERTIES OF ORDER AT A POINT ON A POSSIBLY NON-REDUCED CURVE 191

11. Properties of order at a point on a possibly non-reduced curve In section IV.3.2 we deﬁned the notion of “order at a point on a possibly non-reduced curve,” and stated some of its properties without proof. In this section we prove these results, namely proposition IV.9 and theorem IV.12, in respectively proposition B.30 and theorem B.34. The proof uses somewhat more involved commutative algebra than the rest of the book. Given a point a on a possibly non-reduced curve

C and f , the order ord (f) of f at a is the dimension of /f as a vector space over k. ∈ OC,a a OC,a OC,a Proposition B.30 (proposition IV.9). Let a be a point on a possibly non-reduced curve C and f C,a. ′ ∈ O Let C := Supp(C). Recall that ′ is a quotient of . OC ,a OC,a (1) If the image of f is a non zero-divisor in ′ , then ord (f) < . In particular, if f is a non OC ,a a ∞ zero-divisor in C,a, then orda(f) < . (2) If f is a non zero-divisorO in , then∞ord (f) < . OC,a a ∞ (3) orda(f)=0 if and only if f is invertible in C,a. (4) If f is a non zero-divisor in and g O , then ord (fg) = ord (f)+ord (g). OC,a ∈ OC,a a a a PROOF. Iftheimageof f is a non zero-divisor in ′ , then Krull’s principal ideal theorem [AM69, OC ,a Corollary 11.18] implies that the quotient C,a/f C,a of C,a by the ideal generated by f is a zero dimensional Noetherian local ring. ThereforeO it is alsoO ArtinianO [AM69, Theorem 8.5] and a ﬁnite dimen-

sional vector space over k [AM69, Exercise 8.3], which proves the ﬁrst assertion. The second assertion is straightforward. For the third assertion, without loss of generality we may assume that ord (g) < . Let a ∞

(f ,...,f ) be a basis of /f and (g ,...,g ) be a basis of /g over k. Let h . 1 l OC,a OC,a 1 m OC,a OC,a ∈ OC,a Write Kf and Kg for respectively the k-linear span of the fi and of the gj . Then C,a = Kf + f C,a = K + fK + fg , which implies that f ,...,f ,fg ,...,fg spans Omodulo the idealO gener-

f g OC,a 1 l 1 m OC,a k ated by fg. We claim that they are also linearly independent over k. Indeed, pick ci, dj such that c f + d fg = fgh, h . Then c f f , so that c = = c∈= 0. It fol- i i i j j j ∈ OC,a i i i ∈ OC,a 1 ··· l lows that f d g = fgh. Since f is a non zero-divisor, d g = gh g . It follows that P P j j j P j j j ∈ OC,a d1 = = dm =0 as well. ··· P P The proof of theorem IV.12 will be long. We start with a few auxiliary results. Lemma B.31. Let R be a ring containing a ﬁeld k, and t R be such that t is transcendental over k, and R contains the ring k[[t]] of power series in t over k. Assume∈ R/tR is a ﬁnite dimensional vector space over k generated by (the images in R/tR of) f1,...,fm R. Then f1,...,fm generate R as a module over k[[t]]. ∈

PROOF. Given g R, it can be successively expressed as g = j fjh0,j + tg1 = j fj (h0,j + 2 ∈ th )+ t g , and so on, resulting in an expression of the form g = f h with h k[[t]]. 1,j 2 jPj j j ∈ P Proposition B.32. Let a be a point on an possibly non-reduced affineP curve C. Assume C′ := Supp(C) is irreducible and nonsingular at a. Let t be a regular function on C such that t ′ is a parameter of ′ . |C OC ,a Assume t is not a zero-divisor in C,a. Then O ′ (1) orda(t) = µC′ (C), where µC′ (C) is the multiplicity of C in C (defined in section IV.3.2 in the paragraph preceding theorem IV.12), (2) ord (f) = ord (f ′ )ord (t) for each regular function f on C. a a |C a PROOF. We start with the first assertion. Without loss of generality we may assume that C is the

n+1 k closed subscheme of k determined by the ideal J in S := [x0,...,xn], and a corresponds to the ˆ ˆ ˆ point x0 = = xn = 0, and t is the restriction of x0. Theorem III.6 implies that C,a ∼= S/JS

ˆ ··· ˆ ˆ ˆ ˆ ˆ O k k where S := [[x ,...,x ]]. Let r := dimk (S/(x S + JS)) = dim ( /t ). Assertion (2) of 0 n 0 OC,a OC,a proposition B.30 and exercise III.71 imply that r = dim k ( C,a/t C,a) = orda(t). In particular, r < , ˆ ˆ O O ∞ so that lemma B.31 implies that S/JS is a ﬁnitely generated k[[x0]]-module. Since t is a non zero-divisor in C,a, theorem III.6 implies that x0 is a non zero-divisor in S/Jˆ Sˆ. The fundamental theorem of ﬁnitely

O ˆ ˆ s k generated modules over a PID (theorem B.1) then implies that S/JS ∼= k[[x0]] as a module over [[x0]] ˆ ˆ ˆ ′ ′ for some s 0. But then s = dimk (S/(x0S + JS)) = r. Since t C is a parameter of C ,a, for each i = 1,...,n≥, there is u and a positive integer m such that| x u xmi is nilpotentO in . i ∈ OC,a i i − i 0 OC,a Pick a representative ψ Sˆ of u and set y := x xmi ψ . Pick φ ,...,φ Sˆ such that S/Jˆ Sˆ = i ∈ i i i − 0 i 1 r ∈ 192 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

r k i=1 φi k[[x0]] as a [[x0]]-module. Since the linear parts of (x0,y1,...,yn) are linearly independent, ˆ ˆ theorem III.5 implies that S = k[[x ,y ,...,y ]]. Moreover, since each y is nilpotent modulo JS, L 0 1 n j each φi can be expressed as a k[[x0]]-linear combination of ﬁnitely many monomials in (y1,...,yn). Let yαj , j = 1,...,N, be a minimal collection of such monomials such that every other monomial in ˆ (y1,...,yn) is their k((x0))-linear combination modulo JS. After a Gauss-Jordan elimination process αi N αj and re-orderings of the αj if necessary we may assume that φi = y + j=r+1 φi,j (x0)y , i =1,...,r,

α1 ˆ k where φ (x ) k((x )). Now express y as a [[x ]]-linear combination of the φ modulo JS: i,j 0 ∈ 0 0 P i r r r N α1 αi αj y = ρi(x)φi = ρi(x)y + ρi(x)φi,j (x)y i=1 i=1 i=1 j=r+1 X X X X αi The minimality assumption on the y implies that ρ1(x)=1, and ρi(x)φi,j (x)=0 if either i > 1 or α1 αi j > 1. It follows that y = φ1. The same arguments inductively show that φi = y for each i =1,...,r. In particular, N = r.

Claim B.32.1. C,C′ (which was defined in section IV.2.3) can be identified with a k-subalgebra of [1/t]. O OC,a PROOF. Elements of C,C′ are of the form f/g where f,g C,a such that g C′ 0. Let m := O ∈ O | m6≡ ord (g ′ ). Lemma IV.2 implies that there is invertible u such that g = ut h where h a |C ∈ OC,a − is a nilpotent in . Pick k such that hk+1 = 0. Write f ′ := f/u and h′ := h/u. Then f/g = OC,a f ′/(tm h′) = t−m(k+1)f ′ k tmj h′k−j which is an element of [1/t]. Since t is a non zero- − j=0 OC,a divisor in , it follows that the map ′ [1/t] is injective. OC,a P OC,C → OC,a ˆ ˆ ˆ Claim B.32.1 and proposition III.40 imply that ′ is a k-subalgebra of [1/t] S /JS , C,C C,a ∼= x0 x0 ˆ ˆ O O where S := S[1/x ]= k((x ))[[y ,...,y ]]. Re-order α ,...,α if necessary to ensure that α α x0 0 0 1 n 1 r i − j ∈ n ˆ ˆ ˆ αi+1 αr Z only if i j. For each i =0,...,r, let Ji be the ideal of Sx0 /JSx0 generated by y ,...,y and ≥0 ≥ M := Jˆ ′ . Now fix i, 0 i r 1. i i ∩ OC,C ≤ ≤ − α α Claim B.32.2. y i+1 ,...,y r generate M as an ideal of ′ . i OC,C PROOF. Pick h M . Claim B.32.1 implies that there exists k 0 such that htk and hxk is ∈ i ≥ ∈ OC,a 0 αi+1 αr ˆ ˆ ′ a k[[x ]]-linear combination of the y ,...,y in S/JS. Write M for the ideal of generated by 0 i OC,a yαi+1 ,...,yαr and N ′ for the ideal of generated by M ′ and htk. Let L′ := N ′/M ′ be the quotient i OC,a i i i i of N ′/M ′ as a module over . Let Lˆ′ , Mˆ ′, Nˆ ′ be the completion (with respect to the maximal ideal of i i OC,a i i i ) of respectively L′ ,M ′,N ′. Theorem III.6 implies that Lˆ′ = Nˆ ′/Mˆ ′, and then [AM69, Propositions OC,a i i i i ∼ i i 10.13] implies that Lˆ′ ( ˆ N ′)/( ˆ M ′)=0. A theorem of Krull [AM69, Theorem i ∼= C,a OC,a i C,a OC,a i O′ ⊗ k O ′ ⊗ ′ 10.17] then implies that Li = 0, so that ht Mi . Since t is invertible in C,C′ , it follows that h Mi , as required. ∈ O ∈

α α Claim B.32.2 implies that M /M is generated by y i+1 as an ′ -module. Since y i+1 i i+1 OC,C ∈ Mi Mi+1, it follows that Mi/Mi+1 =0. On the other hand, if h is a nilpotent element in C,C′ , then as \ ˆ ˆˆ 6 O an element of Sx0 /JSx0 , h is in the ideal generated by y1,...,yn, and the ordering of the αj ensures that hM M . Combining these observations we see that as an ′ -module, M /M = ′ /n, i ⊆ i+1 OC,C i i+1 ∼ OC,C where n is the (maximal) ideal of nilpotent elements of ′ . It follows that ′ = M M OC,C OC,C 0 ⊃ 1 ⊃ Mr = 0 is a composition series of C,C′ , so that µC′ (C) = r, which proves the ﬁrst assertion of proposition··· ⊃ B.32. O

For the second assertion, we may assume without loss of generality that q := orda(f C′ ) < . Then q | ∞ there is u C,a such that u C′ is invertible and (t uf) C′ =0. Lemma IV.2 implies that u is invertible ∈ O q | − | 10 in C,a. Note that h := t uf is nilpotent in C,a. By replacing C by a smaller neighborhood of a in C ifO necessary we may assume− h is nilpotent in S/JO . Let z be a new indeterminatewhich we think of as the

~~n+1 n+2 n+2~~

k k k last coordinate of k = , and let D be the closed subscheme of determined by the ideal × q q q q K in T := k[x ,...,x ,z] generated by J and z h. In T/K we have uf = t z = (t ζ z), 0 n − − i=1 − i 10Since C is a possibly non-reduced curve, by a “neighborhood of a in C” we mean an open subscheme of CQcontaining a. 11. PROPERTIES OF ORDER AT A POINT ON A POSSIBLY NON-REDUCED CURVE 193

′ ′ where the ζi are the q-th roots of unity in k. Note that D := Supp(D) is isomorphic to C ; in particular a can be naturally identiﬁed with a point on D, which by an abuse of notation we also denote by a. Since z is nilpotent in T/K, it follows that (t ζ z) ′ = t ′ is a parameter of ′ for each i. Since − i |D |D OD ,a q−1 i q as a module over , it follows that T/K = i=0 z S/J ∼= (S/J) S/J (i) t is not a zero-divisor in , which implies that t ζ z is not a zero-divisor in for P OD,a − i OD,a any i (since it is easy to check that if g1 is a zero-divisor and g2 is nilpotent, then g1 + g2 is a zero-divisor), and (ii) ord (g )= q ord (g ) for each for each g S/J. a |D a |C ∈ These observationstogether with the ﬁrst assertion and proposition B.30 imply that q ord (f ) = ord ((uf) )= a |C a |D q−1 ord ((t ζ z) )= q ord (t )= q2 ord (t ), so that ord (f )= q ord (t ), as required.

~~i=0 a − i |D a |D a |C a |C a |C k~~

~~PLemma B.33. Let D be a possibly non-reduced afﬁne curve over k deﬁned by an ideal I of [x1,...,xn]. k~~

If f R := [x ,...,x ]/I is such that r := dimk (R/fR) < . Then r = ord (f). ∈ 1 n ∞ a∈D a PROOF. Since r < , there are ﬁnitely many zeroes of f on D. DenoteP them by a1,...,ak; let mj ∞ be the maximal ideal of a in R and ι : R Rm be the natural map. j j → j ∗ Claim B.33.1. fR = j ιj (fRmj ). PROOF (A` LA MUMFORDT [Mum95, Proposition 1.11]). We onlyneedto showthe “ ” inclusion. Let ∗ ⊃ h ι (fRm ) for each j. Then for each maximal ideal m of R, there exists u m such that uh fR. It ∈ j j 6∈ ∈ follows that the ideal (fR : h) := u R : uh fR of R is not contained in any maximal ideal of R. The Nullstellensatz then implies that{ 1∈ (fR : h∈), as required.} ∈ ′ ∗ ∗ Claim B.33.2. If j = j , then ι (fRm )+ ι ′ (fRm )= R. 6 j j j j′ ROOF P . Since Rmj /fRmj and Rmj′ /fRmj′ are Artinian local rings, their maximal ideals are nilpo- q q ′ tent [AM69, Proposition 8.6]. Therefore there exists q such that (mj ) Rmj fRmj and (mj ) Rmj′ q q ⊆ ⊆ fRm . Since 1 (m ) + (m ′ ) [why?], the claim follows. j′ ∈ j j Claims B.33.1 and B.33.2 and the Chinese remainder theorem imply that k

∗ k (134) r = dimk (R/fR)= dim (R/ιj (fRmj )) j=1 X Claim B.33.3. R/ι∗(fR ) R /fR for each j. j mj ∼= mj mj ∗ PROOF. It is straightforward to see that the natural map R/ιj (fRmj ) Rmj /fRmj is injective. q → For surjectivity, note that if h = g1/g2 is an element of (mj ) Rmj for some q 0, where g1 R and

q+1 −1 ≥ ∈ k g R m , then h cg (m ) R , where c k is the image of g in R /m R . It follows 2 j 1 j mj 2 ∼= mj j mj ∈ \ − ∈ ∈ ′ ′ q+1 by an induction on q that for each h Rmj and q 1, there exists h R such that h h (mj) Rmj . q+1 ∈ ≥ ∈ − ∈ Choosing q such that (mj) Rmj fRmj (which is possible due to the arguments in the proof of claim B.33.3) yields the required result.⊆ The result follows from (134) and claim B.33.3.

THEOREM B.34 (theorem IV.12). Let a be a point on a possibly non-reduced curve C. Let C1,...,Cs be the irreducible components of Supp(C) containing a and πi : C˜i Ci be the desingularizations of Ci. If f is a non zero-divisor in , then → OC,a (135) ord (f)= µ (C)ord (f )= µ (C) ord (π∗(f )) a Ci a |Ci Ci a˜ i |Ci i i a˜∈π−1(a) X X Xi PROOF. Write C′ := Supp(C). Without loss of generality we may assume that

~~(i) C is afﬁne, i.e. the closed subscheme determined by an ideal a of k[x1,...,xn]; in particular~~

~~′ k k[C ]= [x1,...,xn]/√a; (ii) C′ a is nonsingular; \{ } (iii) f is in the maximal ideal of C,a; (iv) f restricts to a regular functionO on C′, and 194 B. MISCELLANEOUS COMMUTATIVE ALGEBRA~~

(v) the restriction of f to every irreducible component of C′ is non-constant. Let C¯′ be the (unique) compactification of C′ such that C¯′ a is nonsingular (given a closed embedding of C′ into an affine space, C¯′ can be explicitly constructed\{ by} taking the closure of C′ in an projective completion of the affine space and then resolving the singularities at infinity of the closure of C′). The ′ ′ ¯′ 1 ′ restriction f := f C′ of f to C induces a morphism C P (exercise III.66); we use f to denote this | ′ 1 → morphism as well. Note that f (a)=0 k P (since f is in the maximal ideal of C,a). It follows that ∈ ⊂ O

~~′−1 11 k (vi) There is a ﬁnite set S of k 0 such that f ( S) is an afﬁne curve (exercise III.42). \{ } \~~

~~′ ′−1 ′ ′ ′ ′ ′ ′ ′ ′~~

~~k k Let D := f (k S) and p := s∈S(f s) [C ] [D ]. Then D C V (p ). Let q be~~

′ \ ′ − ′ ∈ ′ ′ ∩ ′ ⊇ ′\ ′ ′ ′ k an element in k[f ] such that q (a) = 0 and D V (p q ) is contained in C . Let g := p q [f ]. It is straightforward to check that gQ′6 is regular on\C′ D′, g′(a) = 0, D′ V (g′) = C′ V (∈g′), and

~~′ ′ ∪ 6 \ \ ′ k k[D ]g′ = [C ]g′ . In other words, if we write X for the open subscheme C V (g) of C, and X for the support of X, then there is a commutative diagram as in ﬁg. 1a. \~~

~~∼ XD= V (g) D X \ Supp Supp Supp Supp ∼ ∼ X′ = D′ V (g′) D′ X′ = D′ V (g′) D′ \ \ (A) (B)~~

~~FIGURE 1. Compactiﬁcation of C V (g) \~~

~~′ ′ Pick g k[f] such that g = g C′ . We now construct an afﬁne scheme D with support D such~~

∈ | ′ ′ k that the diagram in ﬁg. 1b commutes. Indeed, since g is invertible in k[X ] ∼= ( [x1,...,xn]/√a)g′ , it follows that g is invertible in A := (k[x1,...,xn]/a)g (to see this use the same arguments as in the proof of lemma IV.2). Pick k-algebra generators f1 =1/g,f2 = g,f3,...,fk of A such that

~~¯ ′ ′ k (vii) for each j = 2,...,k, the image fj in k[X ] of fj is in the image of the natural map [D ] ′ → k[X ], and~~

~~′ ¯ ¯ k (viii) k[D ]= [f2,..., fk].~~

~~k−1 k Let B := k[f2,...,fk] A, and D be the “closed subscheme of with coordinate ring B,” i.e. D is~~

⊂k−1 k the closed subscheme of k corresponding to the kernel of the surjective map [x ,...,x ] B which 2 k → maps each xj to fj . Then it is straightforward to check that the diagram in ﬁg. 1b commutes. For each i =1,...,s, we write Di for the irreducible component of D containing Ci. Let p := s∈S (f s) k[f], ′ ′ ′ − ∈ so that p = p C′ . Then p is invertible in k[D ], which implies that p is invertible in B. | Q

~~′ ′~~

~~k k Claim B.34.1. B (respectively k[D ]) is a ﬁnitely generated module over [f]p (respectively [f ]p′ ).~~

~~′ k PROOF. Theorem IV.7 implies that each k[Di] is a ﬁnitely generated [f ]p′ -module. Since the natural~~

~~′ ′ ′~~

~~k k k k map k[D ] i [Di] is injective, [D ] is isomorphic to a [f ]p′ -submodule of i [Di], and there-~~

→ ′ ′ ′ ′ k fore also a ﬁnitely generated module over k[f ]p′ . Pick a ﬁnite collection g1,...,gk of [f ]p′ -module Q ′ ′ Q generators of k[D ]. Pick gi B such that gi = gi D′ , i = 1,...,k. Let h1,...,hl be generators of the ideal n of nilpotent elements of∈ B. There is m 0 such| that nm+1 =0. We claim that B is generated as a α1 αm ≥ l k[f]p-module by gih1 hm , 1 i k, α Z≥0, α1 + + αl m. Indeed, let u B. Then there ··· ≤ ≤ ∈k ··· ≤ l ∈ are φ1,...,φk k[f]p such that u1 := u j=1 φj gj n. Then u1 = j=1 u1,j hj. Expressing the ∈ − ∈ u as k[f] -linear combinations of the g modulo n and continuing as above gives the claim. 1,j p j P P

~~11 ′−1 In fact f (k) is also afﬁne; however, all the proofs we know of use Riemann-Roch theorem, which we do not cover in this book. 12.THEPOWERSERIESRINGANDMONOMIALORDERS 195~~

~~The fundamental theorem of ﬁnitely generated modules over a PID (theorem B.1) implies that as a~~

~~k[f]p-module B has a decompositions of the form:~~

~~r k B = (k[f] ) [f] / φ (f) ∼ p  p h j i j M M   such that each φ (f) is a non-constant polynomial in f which does not vanish at f = c for any c S. For~~

j ∈ k each c k, it follows that f c is a zero-divisor in B if and only if t c divides some φ (t) in [t] (where ∈ − − j t is an indeterminate); in particular, there are only ﬁnitely many such c k. It is straightforward to see that

~~∈ k r = dim k (B/ f c for all c such that f c is a non zero-divisor in B. Lemma B.33 then implies that h − i ∈ −~~

~~r = ord (f c) b − b∈D f(Xb)=c~~

∗ k for all c k such that f c is a non zero-divisor in [D]. Now pick an arbitrary point a C a . Applying∈ the above construction− with a∗ and C∗ := C a respectively in place of a and C∈ yields\{ an} afﬁne possibly non-reduced curve D∗ such that Supp(D\{∗) containing} an open neighborhood of a∗, and the same arguments show that there is an integer r∗ such that

~~∗ r = ord ((f c) ∗ ) b − |D b∈D∗ f(Xb)=c~~

∗ ∗ for all c k such that f c is a non zero-divisor in the “coordinate ring” of D . Since D and D are birational,∈ and since our construction− guarantees that f is a non zero-divisor in coordinate rings of both D and D∗, it follows that r = r∗ and

~~(136) ord (f)= ord (f ∗ ) b b |D b∈D b∈D∗ f(Xb)=0 f(Xb)=0~~

−1 ∗ Let a1 = a,a2,...,ak be the points of f (0) on D. For each j > 1, the construction of D shows that ∗ ∗ ∗ −1 a D and a has a neighborhood on D which is isomorphic to D . Let S := (f ∗ ) (0) j ∈ j |Supp(D ) \ a ,...,,a D∗. Identity (136) implies that { 2 k}⊂

orda(f)= ordb(f D∗ ) ∗ | bX∈S It follows from the construction of D∗ that Supp(D∗) is nonsingular at every point on S. For each b S∗, let ∗ be the (unique) irreducible component of ∗ containing . Proposition B.32 then implies that∈ Dib D b

∗ orda(f)= µD∗ (D )ordb(f D∗ ) ib ib ∗ | bX∈S ∗ Since there Dib and Ci have isomorphic nonempty Zariski open subsets such that the corresponding open ∗ ∗ ∗ subschemes of D and C are isomorphic, it follows that µD (D ) = µCi (C), and since the desingular- ib b ˜ ∗ ∗ ∗ ization Ci of Ci is isomorphic to Di near b, it follows that ordb(f D ) = ordb(πi (f Ci )), which proves b | ib b | b the theorem.

12. The power series ring and monomial orders In this section we define “monomial order,” and use it to give simple proofs that the ring of power series (in finitely many variables over a field) is Noetherian (corollary B.41) and that the dimensions of the quotients of power series rings are “finitely determined” (theorem B.42). 196 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

12.1. Monomial order. A monomial order on Zn is a binary relation on Zn such that ≥0 ≥0 (a) is a total order12, n n (b) is compatible with the addition on Z≥0, i.e. if α β, then α + γ β + γ for all γ Z≥0, and n ∈ (c) 0 α for each α Z≥0. ∈ n We show below in corollary B.37 that every monomial order is also a well order on Z≥0, i.e. for every n ′ ′ nonempty subset S of Z , there is a unique α S such that α α for all α S. Let x1,...,xn ≥0 ∈ ∈ be indeterminates. Since a monomial order on Zn deﬁnes an ordering on the set of monomials in

≥0 k (x ,...,x ), we also say that is a monomial order on k[x ,...,x ] or on [[x ,...,x ]]. 1 n 1 n 1 n n n Example B.35. The lexicographical order lex on Z≥0 is defined as follows: if α, β Z≥0, then α lex β if either α = β, or α = β and the first non-zero coordinate (from the left) of α ∈ β is negative. It is 6 n − straightforward to check that lex is a monomial order and also a well order on Z . ≥0 A corner point of a subset S of Zn is an element α S such that there is no α′ S, α′ = α, such ≥0 ∈ ∈ 6 that α = α′ + β for some β Zn . ∈ ≥0 n Lemma B.36. Let S be a nonempty subset of Z≥0. The set S of corner points of S is finite and nonempty. n n C Moreover, S + Z = S + Z . ≥0 C ≥0 PROOF. It is immediate to check that the minimal element of S with respect to lex is a corner 0 point of S, which proves that S is nonempty. Now assume S has infinitely many elements. Let α = (α0,...,α0 ) be an arbitrary elementC of S. For each α = (α ,...,αC ) , there is j such that α < α0. 1 n 1 n ∈CS j j Fix such that there are infinitely many with 0 . Since there are finitely many choices for j1 α S αj1 < αj1 0 ∈C 1 1 1 αj , it follows that there is aj < α such that := α S : αj = aj is infinite. Now fix α . 1 1 j1 CS { ∈C 1 1 } ∈C Replacing 0 by 1 and by 1 and running the above procedure yields and 1 such that α α S S j2 = j1 aj2 < αj2 2 1 C C 6 n S := α S : αj2 = aj2 is infinite. Continuing this process we will end up with an infinite set S . But thisC is absurd,{ ∈C since n will} consist of a single element (a ,...,a ) by construction. This contradictionC CS 1 n implies that S was finite to begin with, which proves the first assertion. For the second assertion, pick C n (α1,...,αn) S and define S≤α := β S : α β Z≥0 . The first assertion implies that there is ∈ { ∈ − ∈ } n a corner point β of S≤α. It is clear that β is also a corner point of S and α β + Z≥0. This proves the second assertion. ∈

n n Corollary B.37. Every monomial order on Z≥0 is also a well order on Z≥0. n n PROOF. Let be a monomial order on Z and S be a nonempty subset of Z . Let S be the set of ≥0 ≥0 C corner points of S. Since S is finite and nonempty, it has a unique minimal element β0 with respect to . C n For every α S, lemma B.36 implies that there is β S such that α β Z , so that properties (b) ∈ ∈ C − ∈ ≥0 and (c) of monomial orders imply that β α. It follows that β0 is the minimal element of with respect to . S Corollary B.38. Let S Zn. Then the convex hull of S + Rn is a (convex) polyhedron. ⊂ ≥0 n n PROOF. By translating S if necessary, we may assume S Z≥0. Lemma B.36 implies that S+R≥0 = n ⊂ S + R≥0, where S is the finite set of corner points of . It is straightforward to see that the convex hull C C n S of the Minkowski addition of any finite set with R≥0 is a convex polyhedron.

n n n We say that a monomial order on Z has finite depth if for every α Z , the set [Z ]α := ≥0 ∈ ≥0 ≥0 β Zn : β α is finite. { ∈ ≥0 } Example B.39. The lexicographic order lex from example B.35 does not have finite depth. The graded n lexicographic order grlex on Z≥0 is defined as follows: if α = (α1,...,αn) and β = (β1,...,βn) n ∈ Z≥0, then α grlex β if either j αj < j βj, or if j αj = j βj and α lex β. It is straightforward n to check that grlex is a monomial order on Z of finite depth. P P ≥0 P P 12A total order on a set S is a binary relation on S which is reflexive (i.e. x x for all x ∈ S), transitive (i.e. if x y and y z then x z), antisymmetric (i.e. if x y and y x then x = y), and totally comparable (i.e. either x y or y x for each x,y ∈ S). 12.THEPOWERSERIESRINGANDMONOMIALORDERS 197

12.2. Rings of formal power series over a field. Let k be a field, x1,...,xn be indeterminates, and R := k[[x1,...,xn]] be the ring of formal power series in (x1,...,xn) over k. Lemma B.36 implies that every ideal of R which is generated by monomials is finitely generated. We now prove the more general n fact that every ideal of R is finitely generated. Fix a monomial order on Z≥0. For each nonempty subset n α S of Z≥0, we write In(S) for the minimal element of S with respect to . For f = α cαx R, the n ∈ support Supp(f) of f is the set α : cα =0 Z≥0; if f =0 and α := In(Supp(f)), we say that α is { 6 }⊆ α 6 P the initial exponent of f denoted by exp(f) and cαx is the initial form In(f) of f. For each subset Q n of R, we write exp(Q) := exp(f): f Q, f =0 Z≥0 for the set of initial exponents of non-zero elements in Q. { ∈ 6 }⊆ n THEOREM B.40. Let be a monomial order of finite depth on Z . Let I be an ideal and I = ≥0 C α1,...,αs be the set of corner points of exp(I). For each i = 1,...,s, pick fi I such that { }αi ∈ In(fi)= x . ′ ′ (1) Each g R can be expressed as g = i fihi + g for some h1,...,hs,g R such that either g′ =0,∈ or exp (g) exp (g′) and exp (g′) exp (I). ∈ P 6∈ (2) I is generated by f1,...,fs. n

~~(3) If g R 0 is such that Supp(g) Z≥0 exp(I), then g I. In particular, dim k (R/I)=~~

~~n ∈ \{ } ⊂ \ α n6∈ k Z exp (I) and if dim k (R/I) < , then x : α Z exp (I) form a -basis of | ≥0 \ | ∞ { ∈ ≥0 \ } R/I.~~

PROOF. At first we prove the first assertion. Pick g R. If α := exp(g) exp(I), there is ∈ 6∈ n nothing to do. Otherwise pick the smallest i1, 1 i1 s, such that α = αi1 + β1 for some β1 Z . ≤ ≤ ∈ ≥0 αi +β1 β1 Then In (g) = c x 1 for some c k. Write g := g and g := g c x f . Continuing with g 1 1 ∈ 0 1 − 1 i1 1 and repeating this procedure, yields a sequence of elements (gk)k≥0 such that exp(gk) exp(gk+1) for each k. Either this sequence is infinite, in which case the finite depth of ensures that f is a R-linear combination of the fj, or it stops at a stage k, in which case exp(gk) exp(I). This implies the first assertion. The second assertion follows from the first. The third assertion6∈ follows from the first assertion and finite depth of . Corollary B.41. R is Noetherian.

12.3. Finite determinacy of quotients of power series rings. Given formal power series f1,...,fs R := k[[x ,...,x ]], in this section we show that the dimension of R/ f ,...,f as a vector space over∈ 1 n h 1 si k can be finitely determined, i.e. it can be determined by polynomials g1,...,gs provided the gj are “sufficiently close” to to fj, where the “closeness” of elements of R will be measured by monomial orders of n α n finite depth. Let be a monomial order on Z . For each f = cαx R and each β Z write ≥0 α ∈ ∈ ≥0 [f] := c xα. β αβ α P THEOREMP B.42. Assume has finite depth. Let I be an ideal of R generated by f1,...,fs. For each n β Z≥0, let [I]β be the ideal of R generated by [fj ]β, j =1,...,s. ∈ n

~~′ k (1) If dimk (R/I) < , then there is β Z≥0 such that exp(I) = exp([I]β ) and dim (R/I)= ∞ ′ ∈ ′ dim k (R/[I]β ) for all β β. n~~

~~′ k (2) If dim k (R/I)= , then for each N 0, there is β Z such that dim (R/[I]β ) N for ∞ ≥ ∈ ≥0 ≥ all β′ β. ~~

PROOF. At first assume dimk (R/I) < . Due to theorem B.40 to prove assertion item 1 it suffices ′ ∞ to show that exp([I]β′ ) = exp(I) if β is sufficiently “high” with respect to . Theorem B.40 implies n n that Z≥0 exp(I) is finite. For every finite subset S of Z≥0, write ld(S) for the maximal element of S \ n ′ with respect to . Let β1 := ld(Z≥0 exp(I)). Then for each β β1 and each g1,...,gs R, we have \ ∈

g [f ] ′ = g ([f ] ′ f )+ g f j j β j j β − j j j j j j X X X Let h := j gj[fj ]β′ , h1 := j gj ([fj ]β′ fj) and h2 := j gj fj . Since exp(h2) exp(I) and n − ∈ exp (h1) ld(Z exp (I)), it is straightforward to see that if h = h1 + h2 = 0, then exp (h) P ≥0 \ P P 6 ∈ exp (I). It follows that exp ([I] ′ ) exp (I). Now let be the set of corner points of exp (I) and β ⊆ CI 198 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

β := ld (exp ( )). Fix α . There is g ,...,g R such that exp( g f )= α. Pick β′ β , 2 CI ∈CI 1 s ∈ j j j 2 and define h,h1,h2 as above. Then exp (h2) exp (h1), so that exp (h) = exp (h2)= α. Therefore P I exp([I]β′ ) and consequently lemma B.36 implies that exp(I) exp([I]β′ ). It follows that C ⊂ ′ ⊂ exp (I) = exp ([I] ′ ) whenever β β := ld β ,β , which proves the first assertion. For the β { 1 2} second assertion, fix N 0. Take an arbitrary finite subset S of Zn exp (I) such that S N and let ≥ ≥0 \ | |≥ β := ld (S). The same argument as in the beginning of the proof suggests that for each β′ β and each ′ ′ g1,...,gs R, exp( j gj[fj ]β ) S, so that theorem B.40 implies that dimk (R/[I]β ) S N, as required.∈ 6∈ ≥ | |≥ P

13. Primitive elements of Zn An element in Zn is primitive if it is nonzero and the greatest common divisor of its nonzero coordinates is one. Every member of a basis of Zn is primitive. The following lemma shows that the converse is also true. In this section we use , to denote the standard inner product on Zn, i.e. h· ·i (a ,...,a ), (b ,...,b ) := a b h 1 n 1 n i j j j X Lemma B.43. Let n be a positive integer and G be a subgroup of Zn. (1) If α is a primitive element in Zn, then there is a basis of Zn containing α. m n (2) G = Z for some m n, and there is a basis (α1,...,αn) of Z and positive integers ∼ ≤ k1,...,km such that (k1α1,...,kmαm) is a basis of G.

n ∗ PROOF. For the first assertion, let α1 := α. Since α1 is primitive, there is β1 (Z ) such that ⊥ n ∈ β1, α1 = 1. Let H1 := β1 := γ Z : β1,γ = 0 . If H1 = 0, pick a primitive element α2 H1 h i n ∗ { ∈ h i } ⊥6 ∈ and β2 (Z ) such that β2, α2 = 1, and set H2 := H1 β . Continue in this way up to the n-th ∈ h i ∩ 2 step. It is straightforward to see that β1,...,βn are linearly independent (over R), so that Hn = 0 and n (α1,...,αn) is a basis of Z , as required. For the second assertion, we may assume G = 0 . For each non-zero α G, let d be the greatest common divisor of the non-zero coordinates of6 α.{ Pick} α G ∈ α ∈ with the smallest possible dα. Due to the first assertion we may assume without loss of generality that α = dα(1, 0,..., 0). The minimality of dα implies that for all β = (β1,...,βn) G, the first coordinate ′ ′ ∈ n−1 β1 of β is divisible by dα. It follows that G = Zα + G , where G := G ( 0 Z ). The second assertion holds for G′ by induction on n, which in turn implies that it holds for∩ {G.}×

Corollary B.44. Let φ : Zn Zm be a homomorphism of abelian groups and r be the rank (over Q) of → the matrix of φ. Then the matrix of φ with respect to appropriate bases of Zn and Zm is of the form D 0 [φ]= 0 0 where D is an r r-diagonal matrix whose diagonal entries are positive integers. × m n m PROOF. Let ei and ej denote respectively the i-th standard unit element in Z and the j-th standard n unit element in Z . Let α1,...,αm be the rows of the matrix of φ with respect to the standard bases of n m n Z and Z . Let G be the subgroup of Z generated by α1,...,αm. Lemma B.43 implies that after a m n n change of basis of Z we may assume that G is generated by k1e1 ,...,krer , where each kj is a positive m n integer. Pick γ1 = (γ1,1,...,γ1,m) Z such that j γ1,j αj = k1e1 . Then γ1 must be a primitive m ∈ m m element of Z , so after a change of basis in Z , we may assume that γ1 = e1 , which in turn implies that n P n ′ ′ ′ α1 = k1e1 . Note that for each j =2,...,m, αj is of the form dj k1e1 + αj , with αj in the subgroup G of n n n m m m m Z generated by k2e2 ,...,krer . Therefore, after a change of basis of Z the form ej ej + dj k1e1 , j =2,...,m, the matrix of φ is of the form 7→

k1 0 0 M for some (n 1) (m 1) matrix M. Now apply induction (say, on n) to the homomorphism from − × − Zn−1 Zm−1 induced by M. → 14. SYMMETRIC MULTIADDITIVE FUNCTIONS ON A COMMUTATIVE SEMIGROUP 199

14. Symmetric multiadditive functions on a commutative semigroup Throughout this section ( , +) is a commutative semigroup and n is a positive integer. Let ρ bea map- n K ping from to R. We say that ρ is symmetric if ρ(f1,...,fn)= ρ(fσ1 ,...,fσn ) for each permutation σ of (1, 2,...,nK ). We say that ρ is multiadditive if

ρ(h1,...,hj−1,qf + rg,hj+1 ...,hn)= qρ(h1,...,hj−1,f,hj+1 ...,hn) (137) + rρ(h1,...,hj−1,g,hj+1 ...,hn) for each q, r Z≥0, j 1,...,n and f,g,h1,...,hj−1,hj+1,...,hn . Throughout this section ∈ ∈ { } ∈ K we use ρ to denote a symmetric multiadditive function from n to R. K

14.1. Existence from polynomial functions. For nonnegative integers j1,...,jk,n such that n = n n j1 + + jk, we write for the multinomial coefficient n!/(j1! jk!). Recall that is ··· j1,...,jk ··· j1,...,jk precisely the coefficient of xj1 xjk in (x + + x )n. 1 ··· k 1 ··· k Lemma B.45. Let ν : R be a function which satisfies the following property: for each s 1 and K → s ≥ f1,...,fs , there are να(f1,...,fs) R for all α s := (α1,...,αs) (Z : α1 + + αs = ∈ K ∈ ∈E { ∈ ≥0 ··· n such that for all λ1,...,λs Z≥0, } ∈ (138) ν(λ f + + λ f )= ν (f ,...,f )λα1 λαs 1 1 ··· s s α 1 s 1 ··· s αX∈Es 1 n Then ρ(f1,...,fn) := n! ν(1,...,1)(f1,...,fn) is a symmetric multiadditive function from to R such that ρ(f,...,f)= ν(f). K

n PROOF. Fix f . Applying (138) with s = 1 shows that ν(λf) = νn(f)λ for each λ Z≥0. ∈ K n ∈ Setting λ =1, we have νn(f)= ν(f), and therefore ν(λf)= λ ν(f) for all λ Z≥0. It follows that ∈ n ν(λ f + + λ f) = (λ + + λ )nν(f)= λα1 λαn ν(f) 1 ··· n 1 ··· n α ,...,α 1 ··· n α 1 n X so that 1 n ρ(f,...,f)= ν(f)= ν(f) n! 1,..., 1 It is clear that ρ is symmetric in its arguments. For multiadditivity, write elements of Zn+1 as (α, β, γ¯) := (α,β,γ1,...,γn−1) and note that

~~α β γ1 γn−1 ν(λf + µg + τ1h1 + + τn−1hn)= ν(α,β,γ¯)(f,g,h1,...,hn−1)λ µ τ1 τn−1 ··· ¯ ··· (α,β,Xγ)∈En+1 so that~~

~~α+β γ1 γn−1 ν(λ(f + g)+ τ1h1 + + τn−1hn)= ν(α,β,γ¯)(f,g,h1,...,hn−1)λ τ1 τn−1 ··· ¯ ··· (α,β,Xγ)∈En+1 It follows that~~

ν(δ,γ¯)(f + g,h1,...,hn−1)= ν(α,β,γ¯)(f,g,h1,...,hn−1) α+Xβ=δ In particular, 1 ρ(f + g,h ,...,h )= ν (f + g,h ,...,h ) 1 n−1 n! (1,...,1) 1 n−1 1 = (ν (f,g,h ,...,h )+ ν (f,g,h ,...,h )) n! (1,0,1,...,1) 1 n−1 (0,1,...,1) 1 n−1 = ρ(f,h1,...,hn−1)+ ρ(g,h1,...,hn−1) which implies that ρ is multiadditive, and completes the proof. 200 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

14.2. Identities from homogeneous polynomials. Let x1,...,xN be indeterminates and An be the abelian group spanned by homogeneouspolynomialsof degree n in x1,...,xN . Given f := (f1,...,fN ) N ∈ , ρ induces a map ρf : An R given by K →

j1 jk x x ρ(fi ,...,fi ,...,fi ,...,fi ), i1 ··· ik 7→ 1 1 k k where each fis is repeated js times. It is straightforward to check that ρf is a homomorphism of abelian groups. This implies the following principle:

~~Remark B.46. Any identity that is valid on An remains valid after replacing every term that appears in it by its image under ρf .~~

~~We now use this observation to show that a symmetric multiadditive function is uniquely determined by its diagonal part.~~

~~Lemma B.47. Let x ,...,x be indeterminates and I [n]. Let I := i ,...,i , where k = I . Write 1 n ⊆ { 1 k} | | s := (x + + x )n I i1 ··· ik~~

~~n j1 jk rI := xi1 xik j1,...,jk ··· j1+···+jk =n jl≥1,X l=1,...,k~~

~~Then~~

~~(139) r = s s + s + + ( 1)k−1 s I I − J J ··· − J J⊂I J⊂I J⊂I |J|X=k−1 |J|X=k−2 |JX|=1~~

~~In particular,~~

~~(140) n!x x = ( 1)n−|I|( x )n 1 ··· n − i I⊆[n] i∈I XI6=∅ X~~

~~PROOF. It sufﬁces to prove identity (139), since identity (140) follows from (139) by setting I = [n]. Straightforward algebra shows that~~

~~n j1 jk sI = xi1 xik = rI + rJ = rI + rJ + rJ j1,...,jk ··· j1+···+jk =n J(I J(I J(I X X |J|X=k−1 |J|≤Xk−2~~

= r + (s r ′ )+ r I J − J J J(I J′(J J(I |J|X=k−1 X |J|≤Xk−2 k−2 = r + s + (1 k + J )r = r + s + (1 k + l)r I J − | | J I J − J J(I J(I J(I l=1 J⊂I |J|X=k−1 |J|≤Xk−2 |J|X=k−1 X |XJ|=l k−3 = r + s (s r ′ )+ (1 k + l)r I J − J − J − J J(I J⊂I J′(J l=1 J⊂I |J|X=k−1 |J|X=k−2 X X |XJ|=l k−3 k l = r + s s + (1 (k l)+ − )r I J − J − − 2 J J(I J⊂I l=1 J⊂I |J|X=k−1 |J|X=k−2 X |XJ|=l 14. SYMMETRIC MULTIADDITIVE FUNCTIONS ON A COMMUTATIVE SEMIGROUP 201

In this way, at every step writing r = s ′ r ′ and rearranging terms, and observing that r = s J J − J (J J J J whenever J =1, we will have | | P s = r + s s + + ( 1)k−3 s I I J − J ··· − J J(I J⊂I J⊂I |J|X=k−1 |J|X=k−2 |JX|=2 k 1 k 1 + (1 (k 1) + − + + ( 1)k−2 − )s − − 2 ··· − k 2 J J⊂I − |JX|=1

Since k−2( 1)j k−1 = (1 1)k−1 ( 1)k−1 = ( 1)k−1, the lemma follows. j=0 − j − − − − − CorollaryP B.48 (See e.g. [hl]). ρ is uniquely determined by its diagonal part, i.e. the map from to R which sends f ρ(f,...,f). More precisely, K 7→ 1 (141) ρ(f ,...,f )= ( 1)n−|I|ρ( f ,..., f ) 1 n n! − i i I⊆[n] i∈I i∈I XI6=∅ X X

~~PROOF. Combine remark B.46 and identity (140).~~

~~Lemma B.49. Let x1,...,xk,y1,...,yk be indeterminates. Then (142) x x = ( 1)k−|I| (x + y ) y 1 ··· k − i i j IX⊆[k] Yi∈I j∈Y[k]\I~~

~~PROOF. We provethis by induction on k. For k =1 it boils down to the identity x1 = (x1 + y1) y1. In the general case, write −~~

k k j−1 k x x = (x + y ) ( x )y ( (x + y )) 1 ··· k i i − i j i i i=1 j=1 i=1 i=j+1 Y X Y Y j−1 Applying induction to the expression i=1 xi yields that k k Q k j−1−|I| x x = (x + y ) ( ( 1) (x + y ) y ′ )y ( (x + y )) 1 ··· k i i − − i i i j i i i=1 j=1 ′ i=j+1 Y X I⊆X[j−1] Yi∈I i ∈[Yj−1]\I Y k k k−|I|−1 = (x + y ) ( 1) (x + y ) y ′ i i − − i i i i=1 j=1 [k]⊇I⊇[k]\[j] i∈I i′∈[k]\I Y X XI6∋j Y Y k−|I| = ( 1) (xi + yi) yi′ − ′ IX⊆[k] Yi∈I i ∈Y[k]\I This completes the proof.

For the next result we assume ρ : n R is a symmetric multiadditive “rational” map, i.e. ρ may not be defined everywhere on n, but ifKρ(f→,...,f ) is defined, then ρ(f ,...,f ) are defined for all K 1 n σ1 σn permutations σ of (1,...,n), and all of them take the same value in R; and identity (137) holds whenever ρ is defined on at least two of the three elements of n that appear on (137). Given h ,...,h K 1 n ∈ K and I,J [n] such that I + J = n, we write ρ((hi)i∈I , (hj )j∈J ) for ρ(hi1 ,...,hik ,hj1 ,...,hjn−k ) (provided⊆ it is defined), where| | I| =| i ,...,i and J = j ,...,j . { 1 k} { 1 n−k} Corollary B.50. Let f1,...,fn,g1,...,gn be such that ρ((fi)i∈I , (gj )j∈[n]\I ) is defined for each I [n]. Then ∈ K ⊆ (143) ρ(f ,...,f )= ( 1)n−|I|ρ((f + g ) , (g ) ) 1 n − i i i∈I j j∈[n]\I IX⊆[n] 202 B. MISCELLANEOUS COMMUTATIVE ALGEBRA

PROOF. Let x1,...,xn,y1,...,yn be indeterminates and Bn be the abelian group generated by the monomials of degree n of the form x y for all I [n]. Then the map ρ which sends i∈I i j∈[n]\I j ⊆ f,g xi yj ρ((fi)i∈I , (gj )j∈[n]\I ) is a well deﬁned homomorphism from Bn to R. Since all i∈I j∈[n]\I 7→ Q Q the monomials that appear on (142) for k = n belong to B , identity (143) therefore follows from applying Q Q n ρf,g to the k = n case of (142). Bibliography

[AGZV85] V. I. Arnold, S. M. Gusein-Zade, and A. N. Varchenko. Singularities of Differentiable Maps, volume 1. Birkhäuser, 1985. [AM69] M. F. Atiyah and I. G. Macdonald. Introduction to commutative algebra. Addison-Wesley Publishing Co., Reading, Mass.- London-Don Mills, Ont., 1969. [And07] Kirsti Andersen. The geometry of an art. The history of the mathematical theory of perspective from Alberti to Monge. New York, NY: Springer, 2007. [Arn04] V. I. Arnold, editor. Arnold’s problems. Translated and revised edition of the 2000 Russian original. Berlin: Springer; Moscow: PHASIS, 2004. [AY83] L. A. Ajzenberg and A. P. Yuzhakov. Integral representations and residues in multidimensional complex analysis. Transl. from the Russian by H. H. McFaden, ed. by Lev J. Leifman. 1983. [Ber75] D. N. Bernstein. The number of roots of a system of equations. Funkcional. Anal. i Priloˇzen., 9(3):1–4, 1975. [BGM12] Yousra Boubakri, Gert-Martin Greuel, and Thomas Markwig. Invariants of hypersurface singularities in positive characteristic. Rev. Mat. Complut., 25(1):61–85, 2012. [BKW17] S. Brzostowski, T. Krasiński, and J. Walewska. Arnold’s problem on monotonicity of the Newton number for surface singularities. ArXiv e-prints, April 2017. [BO16] Szymon Brzostowski and Grzegorz Oleksik. On combinatorial criteria for non-degenerate singularities. Kodai Math. J., 39(2):455–468, 2016. [BZ88] Yu. D. Burago and V. A. Zalgaller. Geometric inequalities. Transl. from the Russian by A. B. Sossinsky. Transl. from the Russian by A.B. Sossinsky. Berlin etc.: Springer-Verlag, transl. from the russian by a.b. sossinsky edition, 1988. [CLS11] David A. Cox, John B. Little, and Henry K. Schenck. Toric varieties, volume 124 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2011. [Dam99] James Damon. A global weighted version of Bezout’s theorem. In The Arnoldfest (Toronto, ON, 1997), volume 24 of Fields Inst. Commun., pages 115–129. Amer. Math. Soc., Providence, RI, 1999. [DF04] David S. Dummit and Richard M. Foote. Abstract algebra. 3rd ed. Chichester: Wiley, 3rd ed. edition, 2004. [DK11] Sergio Da Silva and Kalle Karu. On Oda’s strong factorization conjecture. Tohoku Math. J. (2), 63(2):163–182, 2011. [Eis95] David Eisenbud. Commutative algebra, volume 150 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995. With a view toward algebraic geometry. [Est12] Alexander Esterov. Multiplicities of degenerations of matrices and mixed volumes of Cayley polyhedra. J. Singul., 6:27–36, 2012. [FG87] J. V. Field and J. J. Gray. The geometrical work of Girard Desargues. 1987. [Ful89] William Fulton. Algebraic curves. An introduction to algebraic geometry. Notes written with collab. of R. Weiss. new ed. Redwood City, CA etc.: Addison-Wesley Publishing Company, Inc., new ed. edition, 1989. [Ful93] William Fulton. Introduction to toric varieties, volume 131 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, 1993. The William H. Roever Lectures in Geometry. [GK03] Ron Goldman and Rimvydas Krasauskas, editors. Topics in algebraic geometry and geometric modeling. Proceedings of the workshop on algebraic geometry and geometric modeling, July 29–August 2, 2002, Vilnius, Lithuania. Providence, RI: American Mathematical Society (AMS), 2003. [GKZ94] I. M. Gel′fand, M. M. Kapranov, and A. V. Zelevinsky. Discriminants, resultants, and multidimensional determinants. Mathematics: Theory & Applications. Birkhäuser Boston Inc., Boston, MA, 1994. [Gra11] Jeremy Gray. Worlds out of nothing. A course in the history of geometry in the 19th century. 2nd corrected ed. London: Springer, 2nd corrected ed. edition, 2011. [HJS13] Mar´ıa Isabel Herrero, Gabriela Jeronimo, and Juan Sabia. Affine solution sets of sparse polynomial systems. J. Symb. Comput., 51:34–54, 2013. [HJS17] M. I. Herrero, G. Jeronimo, and J. Sabia. On the multiplicity of isolated roots of sparse polynomial systems. ArXiv e-prints, September 2017. [hl] Tom Leinster (https://mathoverflow.net/users/586/tom leinster). Do the elementary properties of mixed volume characterize it uniquely? MathOverflow. URL:https://mathoverflow.net/q/71980 (version: 2011-08-03). [HS95] Birkett Huber and Bernd Sturmfels. A polyhedral method for solving sparse polynomial systems. Math. Comp., 64(212):1541–1555, 1995. [HS97] Birkett Huber and Bernd Sturmfels. Bernstein’s theorem in affine space. Discrete Comput. Geom., 17(2):137–141, 1997. [Kho78] A. G. Khovanskii. Newton polyhedra, and the genus of complete intersections. Funktsional. Anal. i Priloˇzhen., 12(1):51–61, 1978. [Kho16] A. G. Khovanskii. Newton polytopes and irreducible components of complete intersections. Izv. Math., 80(1):263–284, 2016.

~~203 204 BIBLIOGRAPHY~~

[KK10] Kiumars Kaveh and A. G. Khovanskii. Mixed volume and an extension of intersection theory of divisors. Mosc. Math. J., 10(2):343–375, 479, 2010. [KK12] Kiumars Kaveh and A. G. Khovanskii. Newton-Okounkov bodies, semigroups of integral points, graded algebras and intersection theory. Ann. Math. (2), 176(2):925–978, 2012. [Kol07] János Kollár. Lectures on resolution of singularities., volume 166. Princeton, NJ: Princeton University Press, 2007. [Kou76] A. G. Kouchnirenko. Polyèdres de Newton et nombres de Milnor. Invent. Math., 32(1):1–31, 1976. [LM09] Robert Lazarsfeld and Mircea Mustat¸˘a. Convex bodies associated to linear series. Ann. Sci. Ec.´ Norm. Supér. (4), 42(5):783– 835, 2009. [LW96] T. Y. Li and Xiaoshen Wang. The BKK root count in Cn. Math. Comp., 65(216):1477–1484, 1996. [Mil68] John W. Milnor. Singular points of complex hypersurfaces. Annals of Mathematics Studies. No. 61. Princeton, N.J.: Prince- ton University Press and the University of Tokyo Press. 122 p. (1968)., 1968. [Min41] Ferdinand Minding. Ueber die bestimmung des grades einer durch elimination hervorgehenden gleichung. J. Reine Angew. Math., 1841(22):178–183, 1841. [Mon10] Pinaki Mondal. Towards a Bezout-type theory of affine varieties. http://hdl.handle.net/1807/24371, March 2010. PhD Thesis. [Mon16] Pinaki Mondal. Intersection multiplicity, Milnor number and Bernstein’s theorem. http://arxiv.org/abs/1607.04860, 2016. [Mor96] Robert Morelli. The birational geometry of toric varieties. J. Algebr. Geom., 5(4):751–782, 1996. [MS87] Alexander Morgan and Andrew Sommese. A homotopy for solving general polynomial systems that respects m- homogeneous structures. Appl. Math. Comput., 24:101–113, 1987. [Mum95] David Mumford. Algebraic geometry. I. Classics in Mathematics. Springer-Verlag, Berlin, 1995. Complex projective varieties, Reprint of the 1976 edition. [Oda78] Tadao Oda. Lectures on torus embeddings and applications. (Based on joint work with Katsuya Miyake.). Published for the Tata Institute of Fundamental Research, Bombay. Tata Institute of Fundamental Research Lectures on Mathematics and Physics. Mathematics. 57. Berlin, Heidelberg, New York: Springer-Verlag. XI, 175 p. DM 18.00; $ 8.00 (1978)., 1978. [Oda88] Tadao Oda. Convex bodies and algebraic geometry. An introduction to the theory of toric varieties., volume 15. Berlin etc.: Springer-Verlag, 1988. [Oka79] Mutsuo Oka. On the bifurcation of the multiplicity and topology of the Newton boundary. J. Math. Soc. Japan, 31:435–450, 1979. [Oka97] Mutsuo Oka. Non-degenerate complete intersection singularity. Paris: Hermann, 1997. [Oko96] Andrei Okounkov. Brunn-Minkowski inequality for multiplicities. Invent. Math., 125(3):405–411, 1996. [Oko03] Andrei Okounkov. Why would multiplicities be log-concave? In The orbit method in geometry and physics. In honor of A. A. Kirillov. Papers from the international conference, Marseille, France, December 4–8, 2000, pages 329–347. Boston, MA: Birkhäuser, 2003. [PS08] Patrice Philippon and Mart´ın Sombra. A refinement of the Bernˇstein-Kuˇsnirenko estimate. Adv. Math., 218(5):1370–1418, 2008. [Roj94] J. Maurice Rojas. A convex geometric approach to counting the roots of a polynomial system. Theoret. Comput. Sci., 133(1):105–140, 1994. Selected papers of the Workshop on Continuous Algorithms and Complexity (Barcelona, 1993). [Roj99] J. M. Rojas. Toric intersection theory for affine root counting. J. Pure Appl. Algebra, 136(1):67–100, 1999. [Ros05] B.A. Rosenfeld. The analytic principle of continuity. Am. Math. Mon., 112(8):743–748, 2005. [RW96] J. Maurice Rojas and Xiaoshen Wang. Counting affine roots of polynomial systems via pointed Newton polytopes. J. Com- plexity, 12(2):116–133, 1996. [Sch98] Alexander Schrijver. Theory of linear and integer programming. Repr. Chichester: Wiley, repr. edition, 1998. [Sha94] Igor R. Shafarevich. Basic algebraic geometry. 1. Springer-Verlag, Berlin, second edition, 1994. Varieties in projective space, Translated from the 1988 Russian edition and with notes by Miles Reid. [Wal99] C. T. C. Wall. Newton polytopes and non-degeneracy. J. Reine Angew. Math., 509:1–19, 1999. [Wat79] William C. Waterhouse. Gauss on infinity. Hist. Math., 6:430–436, 1979. [Wł97] Jarosław Włodarczyk. Decomposition of birational toric maps in blow-ups and blow-downs. Trans. Am. Math. Soc., 349(1):373–411, 1997. [Wł03] Jarosław Włodarczyk. Toroidal varieties and the weak factorization theorem. Invent. Math., 154(2):223–331, 2003. [Zie95] Günter M. Ziegler. Lectures on polytopes, volume 152 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1995. [ZS75a] Oscar Zariski and Pierre Samuel. Commutative Algebra. Vol. 1. With the cooperation of I. S. Cohen. 2nd ed., volume 28. Springer, New York, NY, 1975. [ZS75b] Oscar Zariski and Pierre Samuel. Commutative algebra. Vol. II. Springer-Verlag, New York, 1975. Reprint of the 1960 edition, Graduate Texts in Mathematics, Vol. 29.