AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS arxiv:5.597v1 [math.co] 7 May CHRISTOPHER J. BROOKS, ABRAHAM MARTÍN DEL CAMPO, AND FRANK SOTTILE Abstract. We show that the Galois group of any Schubert problem involving lines in projective space contains the alternating group. Using a criterion of Vakil and a special position argument due to Schubert, this follows from a particular inequality among Kostka numbers of two-rowed tableaux. In most cases, an easy combinatorial injection proves the inequality. For the remaining cases, we use that these Kostka numbers appear in tensor product decompositions of sl C-modules. Interpreting the tensor product as the action of certain commuting Toeplitz matrices and using a spectral analysis and Fourier series rewrites the inequality as the positivity of an integral. We establish the inequality by estimating this integral. Introduction The Schubert calculus of enumerative geometry[kl7] is a method to compute the number of solutions to Schubert problems, a class of geometric problems involving linear subspaces. One can reduce the enumeration to combinatorics; for example, the number of solutions to a Schubert problem involving lines is a Kostka number for a rectangular partition with two parts. A prototypical Schubert problem is the classical problem of four lines, which asks for the number of lines in space that meet four given lines. To answer this, note that three general lines l 1,l, and l lie on a unique doubly-ruled hyperboloid, shown in Figure 1. These three lines lie in one ruling, while the second ruling consists of the lines meeting the given three lines. The fourth line l 4 meets the hyperboloid in two points. Through each of these points there is a line in the second ruling, and these are the two lines m 1 and m meeting our four given lines. In terms of Kostka numbers, the problem of four lines reduces to counting the number of tableaux of shape λ = (,) with content (1,1,1,1). There are two such tableaux: 1 4 Galois groups of enumerative problems are subtle invariants about which very little is known. While they were introduced by Jordan in 187 [Jor7], the modern theory began with Harris in 1979, who showed that the algebraic Galois group is equal to a geometric 1 4 1991 Mathematics Subject Classification. 5E15, 14N15. Key words and phrases. Kostka numbers, Galois groups, Schubert calculus, Schubert varieties. Research supported in part by NSF grant DMS-91511 and the Institut Mittag-Leffler. 1
BROOKS, MARTÍN DEL CAMPO, AND SOTTILE m l l 4 l p m 1 l 1 Figure 1. The two lines meeting four lines in space. monodromy group [Har79]. In general, we expect the Galois group of an enumerative problem to be the full symmetric group and when it is not, the geometric problem possesses some intrinsic structure. Harris result gives one approach to studying the Galois group by directly computing monodromy. For instance, the Galois group of the problem of four lines is the group of permutations which are obtained by following the solutions over loops in the space of lines l 1,l,l,l 4. Rotating l 4 18 degrees about the point p (shown in Figure 1) gives a loop which interchanges the two solution lines m 1 and m, showing that the Galois group is the full symmetric group on two letters. Leykin and Sottile [LS9] used numerical homotopy continuation [SW5] to compute monodromy for many simple Schubert problems, showing that in each case the Galois group was the full symmetric group. (The problem of four lines is simple.) Billey and Vakil [BV8] gave an algebraic approach based on elimination theory to compute lower bounds for Galois groups. Vakil [Vak6b] gave a combinatorial criterion, based on group theory, which can be used to show that a Galois group contains the alternating group. He used this and his geometric Littlewood-Richardson rule [Vak6a] to show that the Galois group was at least alternatingforevery Schubert probleminvolving linesinprojectivespacep n forn 16. Brooks implemented Vakil s criterion and the geometric Littlewood-Richardson rule in python and used it to show that for n 4, every Schubert problem involving lines in projective space P n has at least alternating Galois group. Our main result is the following. Theorem 1. The Galois group of any Schubert problem involving lines in P n contains the alternating group. We prove this theorem by applying Vakil s criterion to a special position argument of Schubert, which reduces Theorem 1 to proving a certain inequality among Kostka numbers of
AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS two-rowed tableaux. For most problems, the inequality follows from a combinatorial injection of Young tableaux. For the remaining problems, we work in the representation ring of sl C, where these Kostka numbers also occur. We interpret the tensor product of irreducible sl C- modules in terms of commuting Toeplitz matrices. Using the eigenvector decomposition of the Toeplitz matrices, we express these Kostka numbers as certain trigonometric integrals. In this way, the inequalities of Kostka numbers become inequalities of integrals, which we establish by estimation. Note that the generalization of Theorem 1 to arbitrary Grassmannians is false. Derksen found Schubert problems in the Grassmannian of -planes in P 7 whose Galois groups are significantly smaller than the full symmetric group, and Vakil generalized this to problems in the Grassmannians of k 1 planes in P n 1 whose Galois groups are not the full symmetric group for every k and n k [Vak6b,.1]. 1. Preliminaries 1.1. Schubert problems of lines. Let G(1, n) be the Grassmannian of lines in n-dimensional projective space P n, which is an algebraic manifold of dimension n. A (special) Schubert subvariety is the set of lines X L that meet a linear subspace L P n ; that is, (1.1) X L := {l G(1,n) l L }. If diml = n 1 a, then X L has codimension a in G(1,n). A Schubert problem asks for the lines that meet fixed linear subspaces L 1,...,L m in general position, where diml i = n 1 a i for i = 1,...,m and a 1 + +a m = n. These are the points in the intersection (1.) X L1 X L X Lm. As the L i are in general position, the intersection (1.) is transverse and therefore zerodimensional. (Over fields of characteristic zero, transversality follows from Kleiman s Transversality Theorem [Kle74] while in positive characteristic, it is Theorem E in [Sot97].) We define the Schubert intersection number K(a 1,...,a m ) to be the number of points in the intersection (1.), which does not depend upon the choice of general L 1,...,L m. We call a := (a 1,...,a m ) the type of the Schubert problem (1.). Note that given positive positive integers a = (a 1,...,a m ) whose sum is even, K(a ) is a Schubert intersection number in G(1,n(a )), where n(a ) := 1 (a 1 + + a m + ). Henceforth, a Schubert problem will be a list a of positive integers with even sum. It is valid if a i n(a ) 1 (this is forced by diml i ). Theintersection number K(a )isakostkanumber, whichisthenumber ofyoungtableaux of shape (n(a ) 1,n(a ) 1) and content (a 1,...,a m ) [Ful97, p.5]. Let K(a ) be the set of such tableaux. These are two-rowed arrays of integers, each row of length n(a ) 1, such that the integers increase weakly across each row and strictly down each column, and there are a i occurrences of i for each i = 1,...,m. For example, here are the five Young tableaux in
4 BROOKS, MARTÍN DEL CAMPO, AND SOTTILE K(,,1,,), demonstrating that K(,,1,,) = 5. (1.) 1 1 4 4 5 5 5 1 1 4 4 5 5 5 1 1 4 5 5 5 1 1 4 4 5 5 5 1 1 4 4 5 5 5 1.. Vakil s Criterion for Galois groups of Schubert problems. In.4 of [Vak6b], Vakil explains how to associate a Galois group to a dominant map W X of equidimensional irreducible varieties and establishes his criterion for the Galois group to contain the alternating group. We discuss this for a Schubert problem a = (a 1,...,a m ). Define X := {(L 1,...,L m ) L i P n is a linear space of dimension n 1 a i }, where n := n(a ). Consider the incidence variety, W := {(l,l 1,...,L m ) (L 1,...,L m ) X and l L i, i = 1,...,m}. The projection map W G(1,n) realizes W as a fiber bundle over G(1,n) with irreducible fibers. As G(1, n) is irreducible, W is irreducible. Let : W X be the other projection; its fiber over a point (L 1,...,L m ) X is (1.4) 1 (L 1,L,...,L m ) = X L1 X L X Lm. Thus the map : W X contains all Schubert problems of type a. As the general Schubert problem is a transverse intersection containing K(a ) points, is a dominant map of degree K(a ). Under, the field K(X) of rational functions on X pulls back to a subfield of K(W), the field of rational functions on W, and the extension K(W)/K(X) has degree K(a ). Definition. The Galoisgroup of the Schubert problem of type a, G(a ), is the Galois group of the Galois closure of the field extension K(W)/K(X). This Galois group G(a ) is a subgroup of the symmetric group S K(a ) on K(a ) letters. We say that G(a ) is at least alternating if it contains the alternating group A K(a ). Vakil s Criterion is adapted to classical special position arguments in enumerative geometry. First, if Z X is a subvariety such that Y = 1 (Z) W is irreducible and the map Y Z has degree K(a ), then Y Z has a Galois group which is a subgroup of G(a ). This enables us to restrict the original Schubert problem to one derived from it through certain standard reductions. More interesting is when Z X is a subvariety such that Y = 1 (Z) decomposes into two smaller problems, Y = Y 1 Y, where Y i Z is a Schubert problem of type a (i) for i = 1,. In this situation, monodromy of Y Z gives a subgroup H of the product G(a (1) ) G(a () ) which projects onto each factor and includes into G(a ). Then purely group-theoretic arguments imply the following. Vakil s Criterion. If G(a (1) ) and G(a () ) are at least alternating, and either K(a (1) ) K(a () ) or K(a (1) ) = K(a () ) = 1; then G(a ) is at least alternating.
AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS 5. Inequalities A Schubert problem a = (a 1,...,a m ) is reduced if it is valid and if a i +a j n(a ) 1 for any i < j. Any Schubert problem is equivalent to a reduced one: If a is valid, but a m 1 +a m > n(a ) 1, then K(a 1,...,a m ) = K(a 1,...,a m, a m 1 1,a m 1), as the intersection (1.) for a is equal to an intersection for (a 1,...,a m,a m 1 1,a m 1). Iterating this procedure gives an equivalent reduced Schubert problem. Schubert [Sch86] observed that if the linear spaces are in a special position, then the Schubert problem decomposes into two smaller problems, which gives a (familiar) recursion for these Kostka numbers. Given a reduced Schubert problem a = (a 1,...,a m ), set n := n(a ). Let L 1,...,L m be linear subspaces which are in general position in P n, except that L m 1 and L m span a hyperplane Λ := L m 1,L m. If a line l meets both L m 1 and L m, then either it meets L m 1 L m or it lies in their linear span (while also meeting both L m 1 and L m ). This implies Schubert s recursion for Kostka numbers (.1) K(a 1,...,a m ) = K(a 1,...,a m, a m 1 +a m ) + K(a 1,...,a m, a m 1 1,a m 1). Observe that if a is reduced, then both smaller problems in (.1) are valid. An induction shows that if a is valid, then K(a ) >. For example, consider K(,,1,,). The first tableau in (1.) has both 4s in its second row (along with its 5s), while the remaining four tableaux have last column consisting of a 4 on top of a 5. If we replace the 5s by 4s in the first tableau and erase the last column in the remaining four tableaux, we obtain 1 1 4 4 4 4 4 1 1 4 5 5 1 1 5 5 1 1 4 5 5 1 1 4 5 5 which shows that K(,,1,,) = K(,,1,5)+K(,,1,1,). We state our key lemma. A rearrangement of a Schubert problem a 1,...,a m is simply a listing of the integers a 1,...,a m in some order. Lemma. Every reduced Schubert problem has a rearrangement (a 1,...,a m ) such that either (.) K(a 1,...,a m, a m 1 +a m ) K(a 1,...,a m, a m 1 1,a m 1), and both are nonzero, or else both are equal to 1. We use Lemma below to prove Theorem 1, then we devote the rest of the extended abstract to the proof of this Lemma. of Theorem 1. We use the notation of Subsection 1. and argue by induction on m and n(a ). Assume that a is reduced and let Z be the set of those (L 1,...,L m ) X such that L m 1 and L m span a hyperplane. Then the geometric arguments given before (.1) imply that the pullback 1 (Z) Z decomposes as the union of two Schubert problems, one for (a 1,...,a m, a m 1 +a m ) and the other for (a 1,...,a m, a m 1 1,a m 1). Therefore,
6 BROOKS, MARTÍN DEL CAMPO, AND SOTTILE Lemma and our induction hypothesis, together with Vakil s criterion, imply that G(a ) is at least alternating. While an induction shows that the only reduced Schubert problem where the two terms in (.) are both 1 is (1,1,1,1), the inequality of Lemma is not easy to prove. This is in part because there are no closed formulas for the numbers K(a ), except for the case a 1 = = a m 1 = 1 (in which case K(a ) is given by the hook-length formula)..1. Inequality of Lemma in most cases. We give an injection of sets of Young tableaux to establish Lemma when a i a j for some i,j. Lemma 4. Suppose that (b 1,...,b m,α,β,γ) is a reduced Schubert problem where α β γ with α < γ. Then (.) K(b 1,...,b m, α,β +γ) < K(b 1,...,b m, γ,β +α). To see that this implies Lemma in the case when a i a j, for some i,j, we apply Schubert s recursion to obtain two different expressions for K(b 1,...,b m,α,β,γ), K(b 1,...,b m, α,β+γ) + K(b 1,...,b m, α,β 1,γ 1) = K(b 1,...,b m, γ,β +α) + K(b 1,...,b m, γ,β 1,α 1). By the inequality (.), at least one of these expressions involves unequal terms. Since all four terms are from valid Schubert problems, none is zero, and this implies Lemma when not all a i are identical. of Lemma 4. We establish the inequality (.) via a combinatorial injection ι : K(b 1,...,b m, α,β +γ) K(b 1,...,b m, γ,β +α), which is not surjective. Let T be a tableau in K(b 1,...,b m, α,β+γ) and let A be its sub-tableau consisting of the entries 1,...,m. Then the skew tableau T \A has a bloc of (m+1) s of length a at the end of its first row, and its second row consists of a bloc of (m+1) s of length α a, followed by a bloc of (m+) s of length β+γ. Form the tableau ι(t) by changing the last row of T \A to a bloc of (m+1) s of length γ a followed by a bloc of (m+) s of length β+α. Since a α < γ, this map is well-defined. T = A α a a β+γ A γ a a β+α = ι(t). To see that ιisnot surjective, set b := (b 1,...,b m,γ α 1,β 1), which isavalidschubert problem. Hence K(b ) and K(b ). For any T K(b ), we may add α+1 columns to its end consisting of a m+1 above a m+ to obtain a tableau T K(b 1,...,b m, γ,β +α). As T has more than α (m+1)s in its first row, it cannot be in the image of the injection ι, which completes the proof of the lemma.
AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS 7. Kostka numbers as integrals Kostka numbers of two-rowed tableaux appear as the coefficients in the decomposition of the tensor products of irreducible sl C-modules. Let V a be the irreducible module of sl C with highest weight a. Given a Schubert problem a = (a 1,...,a m ), the Kostka number K(a ) is the multiplicity of the trivial sl C-module V in the tensor product V a1 V am. The representation ring R of sl C is the free abelian group on the isomorphism classes [V a ] ofirreducible modules, modulo the relations[v a ]+[V b ] [V a V b ]. Setting [V a ] [V b ] := [V a V b ] makes R into a ring. Writing e a := [V a ], multiplication by e a is a linear operator M a on R, (.1) M a (e b ) := e a e b = e b+a +e b+a + +e b a, by the Clebsch-Gordan formula. In the basis {e a }, the operator M a is represented by an infinite Toeplitz matrix with entries and 1 given by the formula (.1). For instance, we have M = 1 1 1 1 1 1 1 1 1 1 1 1...., M = 1 1 1 1 1 1 1 1 1 1 1 1 1 1 Since R is a commutative ring, the operators {M a a } commute. They have an easily described system of joint eigenvectors and eigenvalues, which may be verified using the identity sinα sinβ = cos(α β) cos(α+β), and noting that the resulting sums are telescoping. Proposition 5. For each θ and integer a, set v(θ) := (sinθ,sinθ,...,sin(j+1)θ,...) = j λ a (θ) := sin(a+1)θ. sinθ Then v(θ) is an eigenvector of M a with eigenvalue λ a (θ). These eigenvectors form a complete system of eigenvectors. Proposition 6. For any a =,1,,..., we have e j = It follows that for any a 1, we have M a (e ) = sin(j+1)θ v(θ)dθ. λ a (θ)sinθv(θ)dθ.. sin(j+1)θ e j, A consequence of Proposition 6 is an integral formula for the Kostka numbers.....
8 BROOKS, MARTÍN DEL CAMPO, AND SOTTILE Theorem 7. Let a = (a 1,...,a m ) be any valid Schubert problem. Then (.) K(a ) = ( m ) λ ai (θ) sin θdθ. i=1.1. Inequality of Lemma in the remaining case. We complete the proof of Theorem 1 by establishing the inequality in Lemma for those Schubert problems not covered in Lemma 4. For these, every condition is the same, so a = (a,a,...,a) =: a m. If a = 1, then we may use the hook-length formula. The Kostka number K(1 n,b), where n+b = c is even, is the number of Young tableaux of shape (c,c b), which is K(1 n,b) := n!(b+1) (c b)!(c+1)! When m = n is even, the inequality of Lemma is that K(1 n ) K(1 n,). We compute K(1 n ) = (n )!(1) n!(n+1)! and K(1 n,) = (n )!() (n )!(n+1)! and so K(1 n,)/k(1 n n!(n+1)! ) = (n )!(n+1)! = n 1 n+1 1, when n >, but when n = both Kostka numbers are 1, which proves the inequality of Lemma when each a i = 1. We now suppose that a = (a m+ ) where a > 1 and m a is even. Table 1 shows that when a = and m 16, the inequality of Lemma holds. However, the sign of K( m,4) K( m,1,1) changes at m = 14. In fact, we have the following lemma. Lemma 8. For all m 1, we have K( m,4) K( m,1,1). If m < 14 then K( m,4) < K( m,1,1) and if m 14, then K( m,4) > K( m,1,1). The remaining cases a have a more uniform behavior. Lemma 9. For a and for all m we have (.) K(a m, a) < K(a m,(a 1) ). WeomittheproofofLemma9fromthisextendedabstract, butincludeaproofoflemma8.
AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS 9 Table 1. The inequality (.) for the case a = ( m+ ) m K( m,4) K( m,1,1) Difference 1 1 1 1 1 1 1 4 4 6 9 5 15 1 6 6 4 51 11 7 15 7 8 8 4 9 75 85 85 1 5 188 16 11 55 5798 98 156 15511 485 1 46 4185 57 14 11841 1164 7 15 154 157 4848 16 877 85467 85.. Proof of Lemma 8. By the computations in Table 1, we only need to show that K( m,4) K( m,1,1) > for m 14. Using (.), we have K( m,4) K( m,1,1) = = = λ (θ) m( λ 4 (θ) λ 1 (θ) ) sin θ dθ λ (θ) m( sin5θ sinθ sin θ ) dθ λ (θ) m1 ( cos4θ cos6θ 1 ) dθ = 1 λ (θ) m( cos4θ cos6θ 1 ) dθ. The integrand f(θ) of the last integral is symmetric about θ = / in that f(θ) = f( θ). Thus, it suffices to prove that if m 14, then (.4) To simplify our notation, set λ (θ) m (cos4θ cos6θ 1)dθ >. F(θ) := cos4θ cos 6θ 1 and λ(θ) := λ (θ) = 1+cosθ.
1 BROOKS, MARTÍN DEL CAMPO, AND SOTTILE We display these functions and the integrand in (.4) for m = 8 in Figure. 1 1 4 1 1 1 4 F(θ) 1 λ(θ) 4 λ(θ) 8 F(θ) Figure. The functions F(θ), λ(θ), and λ(θ) 8 F(θ). In the interval [, ], the zeroes of F occur at, functions are positive on [, ], and so (.5) λ m (θ)f(θ)dθ λ m (θ)f(θ)dθ, and 5, and λ vanishes at. Both λ m (θ)f(θ) dθ. We show the positivity of (.4) by showing that the right hand side of (.5) is positive for m 14. This is equivalent to the following inequality, (.6) λ m (θ)f(θ)dθ > λ m (θ)f(θ) dθ + λ m (θ)f(θ) dθ. The function λ(θ) is monotone decreasing in the interval [, ], and it vanishes at, so the maximum of λ(θ) on this interval is λ( ) = 1. Also, F(θ) 4 for all θ [, ]. Thus we estimate the last integral in (.6), It is therefore enough to show that (.7) λ m (θ)f(θ) dθ λ m (θ)f(θ)dθ > 1 4dθ =. λ m (θ)f(θ) dθ +, for m 14. We establish (.7) by induction on m 14. This inequality holds for m = 14, as the left hand side is λ 14 (θ)f(θ)dθ = 69 4 + 674 167954168 + 1159.9 7 5555
AN INEQUALITY OF KOSTKA NUMBERS AND GALOIS GROUPS OF SCHUBERT PROBLEMS 11 whereas the right hand side is λ 14 (θ)f(θ) dθ+ = 65 61 1717 + 167954168 5555 87.1 Suppose now that the inequality (.7) holds for some m 14. As λ( ) = 1 + and λ is decreasing in [, ], we have λ(θ) 1 + for θ [, ]. Thus (.8) λ m+1 (θ)f(θ)dθ ( 1+ ) λ m (θ)f(θ)dθ. Similarly, when θ [, ] we have that λ(θ) 1+, as λ( ) = 1. Therefore, (.9) λ m+1 (θ)f(θ) ( dθ 1+ ) λ m (θ)f(θ) dθ. From the induction hypothesis and equations (.8), and (.9), we obtain (.1) λ m+1 (θ)f(θ)dθ > ( 1+ This completes the proof of Lemma 8. ( 1+ ) λ m (θ)f(θ) dθ ) λ m (θ)f(θ) dθ+(1+ ) References > λ m+1 (θ)f(θ) dθ+. [BV8] S. Billey and R. Vakil, Intersections of Schubert varieties and other permutation array schemes, Algorithms in algebraic geometry, IMA Vol. Math. Appl., vol. 146, Springer, New York, 8, pp. 1 54. [Ful97] Wm. Fulton, Young tableaux, London Mathematical Society Student Texts, vol. 5, Cambridge University Press, Cambridge, 1997. [Har79] J. Harris, Galois groups of enumerative problems, Duke Math. J. 46 (1979), 685 74. [Jor7] C. Jordan, Traité des substitutions, Gauthier-Villars, Paris, 187. [KL7] S. Kleiman and D. Laksov, Schubert calculus, Amer. Math. Monthly 79 (197), 161 18. [Kle74] S. Kleiman, The transversality of a general translate, Compositio Math. 8 (1974), 87 97. [LS9] A. Leykin and F. Sottile, Galois groups of Schubert problems via homotopy computation, Math. Comp. 78 (9), no. 67, 1749 1765. [Sch86] H. Schubert, Die n-dimensionalen Verallgemeinerungen der fundamentalen Anzahlen unseres Raume, Math. Ann. 6 (1886), 6 51, (dated 1884). [Sot97] F. Sottile, Enumerative geometry for the real Grassmannian of lines in projective space, Duke Math. J. 87 (1997), no. 1, 59 85. [SW5] A. Sommese and C. Wampler, The numerical solution of systems of polynomials, World Scientific Publishing Co. Pte. Ltd., Hackensack, NJ, 5.
BROOKS, MARTÍN DEL CAMPO, AND SOTTILE [Vak6a] R. Vakil, A geometric Littlewood-Richardson rule, Ann. of Math. () 164 (6), no., 71 41, Appendix A written with A. Knutson. [Vak6b], Schubert induction, Ann. of Math. () 164 (6), no., 489 5. Christopher J. Brooks, Department of Mathematics, Texas A&M University, College Station, Texas 7784, USA E-mail address: cbrooks9@neo.tamu.edu Abraham Martín del Campo, Department of Mathematics, Texas A&M University, College Station, Texas 7784, USA E-mail address: asanchez@math.tamu.edu URL: http://www.math.tamu.edu/~asanchez Frank Sottile, Department of Mathematics, Texas A&M University, College Station, Texas 7784, USA E-mail address: sottile@math.tamu.edu URL: http://www.math.tamu.edu/~sottile