Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
INEQUALITIES FOR HILBERT FUNCTIONS RALF FRÖBERG Let Sn = C[x1 , . . . , xn ], Tn = Sn /(x21 , x22 , . . . , x2n ), En = Chx1 , . . . , xn i/(xi xj + xi xj , x2i ). 1. Homogeneous ideals in Sn Let R = Sn /I, I homogeneous, let hi = dimC Ri . Given hd , we want an upper bound for hd+1 . d−1 + · · · + k11 with kd > kd−1 > · · · > k1 ≥ 0. There Let m, d > 0. Write m = kdd + kd−1 is a unique way to do this. Example d = 3, m = 28. 28 = 63 + 42 + 21 . 1 +1 d−1 +1 d−1 d +1 , so + kd−1+1 + · · · + k1+1 For m = kdd + kd−1 + · · · + k11 , let mhdi = kd+1 3 5 7 h3i 28 = 4 + 3 + 2 = 48. hdi Theorem 1 (Macaulay). hd+1 ≤ hd . Let M be a set of monomials of degree d. Order the monomials lexicographically with x1 < x2 < · · · < xn . M is called a lexsegment if v ∈ M, u > v =⇒ u ∈ M . It is easy to see that an ideal generated by a lexsegment in degree d generates a lexsegment in degree d + 1. It is natural that if we want an ideal generated by monomials of degree d which generates as little as possible, we should choose a lexsegment ideal. For a lexsegment ideal of degree d we have equality in the theorem. If M is a lexsegment of degree 3 in S5 such that h3 (S5 /(M )) = 28, then #M = 73 − 28 = 7, so M = {x35 , x25 x4 , x25 x3, x25 x2 , x25 x1 , x5 x24 , x5 x4 x3 }. The nonzero monomials of 6 4 degree 3 in C[x1 , x2 , x3 , x4 ] are 3 . The nonzero monomials in x5 C[x1 , x2 , x3 ] are 2 . The nonzero ideals in x5 x4 C[x1 , x2 ] are 21 . M generates the lexsegment {x45 , . . . , x5 x4 x3 x1 } in degree 4. The number of nonzero monomials of degree 4 in C[x1 , . . . , x4 ] are 74 , thoose in x5 C[x1 , x2 , x3 ] are 53 , thoose in x5 x4 C[x1 , x2 ] are 32 . In this way one can prove that there is equality for lexsegment ideals. The theoremk1is d−1 proved in general in [1]. They use a theorem by Green. For m = kdd + kd−1 +···+ 1 kd−1 −1 kd −1 k1 −1 let mhdi = d + d−1 + · · · + 1 . Theorem 2 (Green). Let R be a homogeneous C-algebra, and let d ≥ 1 be an integer. Then hd (R/hR) ≤ hd (R)hdi for a general linear form h. 1 2 RALF FRÖBERG Macaulay’s theorem is extended in Gotzmann’s persistence theorem. Theorem 3. If for R = Sn /I we have equality in Macaulay’s theorem between degrees d and d + 1, and if I has no generator of degree > d + 1, then we have equality for all higher degrees. 2. Homogeneous ideals in En Let ∆ be a simplicial complex, and let fi be the number of i-dimensional faces in ∆. If ∆ is the 2-dimensional simplicial complex with maximal faces {x1 , x2 , x3 } and {x3 , x4 }, then f−1 = 1, f0 = 4, f1 = 4, f2 = 1. Consider the ideal I∆ in En generated by all non-faces of ∆. In the example I∆ P = (x1 x4 , x2 x4 ). It is clear that the Hilbert series of En /I∆ (called the indicator algebra) is fi ti+1 . (This resembles the Stanley-Reisner algebra of ∆, which is defined as C[∆] = Sn /I∆ . If the Hilbert series of C[∆] is g(t) and the Hilbert series of the indicator algebra of ∆ is h(t), then k(t) = h(t/(1 − t)). kd−1 kd k1 d−1 + · · · + k11 let hdi m = d−1 + d−1−1 + · · · + 1−1 . For m = kdd + kd−1 kd−1 kd−1 +1 kd k1 kd +1 k1 +1 hdi For m = d + d−1 + · · · + 1 let m = d + d−1 + · · · + 1 . The following is proved in [2]. Theorem 4. (1) If (f−1 , f0 , . . . , fd−1 ) is the f -vector of a pure (all maximal faces have the same dimension) simplicial complex, then hii fi ≤ fi−1 for 1 ≤ i ≤ d − 1. (2) If hii fi = fi−1 for some i in a pure simplicial complex, then hji fj = fj−1 for all 1 ≤ j ≤ i. P (3) If 1 + n0 hi ti is the Hilbert series of a graded C-algebra En /I, then hi+1 ≤ hii hi for 0 < i ≤ n − 1. (4) If hi+1 = hii hi for some i, and I has no generator of degree > i + 1, then hj+1 = hji hj if j ≥ i. (1) was proved by Kruskal and Katona independently. 3. Generic forms in Tn and En Tn and En are isomorphic as graded vector spaces, but not as rings. (In En every odd element has square 0.) 3.1. One generic form in Tn . Theorem 5. The Hilbert series of Tn /(f ), f generic of degree d is [(1 − td )(1 + t)n ]. (Take only terms as long as they are positive.) Proof It suffices to get one example since we have an equality and the generic case is the worst. Take the sum of all squarefree monomials of degree d. Let the squarefree monomials {mi } of degree i − d denote the rows, and the squarefree monomials {ni } of degree i denote the columns in a matrix. The multiplication matrix then is an incidence matrix, in place (j, k) there is a 1 if mj |nk and 0 otherwise. That this matrix has full rank is well-known. INEQUALITIES FOR HILBERT FUNCTIONS 3 3.2. One generic form of even degree in En . It is proved in [3] that the same formula as for Tn is true. 3.3. One generic form of odd degree in En . We have that f· 0 −→ Ann(f )(−d) −→ En (−d) −→ En −→ En /(f ) −→ 0 is exact. It is clear that (f ) ⊆ Ann(f ). Sometimes the inclusion is strict, i.e. (f ) 6= Ann(f ). E.g. when d = 3 and n = 16 the differens in Hilbert series is 16t9 + 120t8 + 559t7 . If d > 3 (and odd) the differens is much smaller. If n − d = 0, 1, 3, 4, 5, the differens seems to be 0. If n − d = 2, the differens seems to be t. If n − d = 6, the differens is t3 if d = 5, 9, 13. If n − d = 11, d = 5 the differens is t6 . Conjectures If n − d = 0, 1, 3, 4, 5, the differens is 0. If n − d = 2, the differens is t. If n − d = 6, the differens is t3 if d = 1 mod 4. For fixed d, Ann(f ) = (f ) if n >> 0. 3.4. Several generic forms. Theorem 6. Consider theP exterior algebra E5 over P a vector space with basis {e1 , ..., e5 } (over any field). Let f1 = cij ei ∧ ej and f2 = dij ei ∧ ej be two forms in E5 . Then {ei ∧ fj } is linearly dependent. Proof It suffices to prove the theorem for generic forms (so we can suppose that the cij ’s and dij ’s are algebraically independent over the prime field of k). I calculated a relation. In [3] they calculate the Hilbert series for two quadratic forms in n ≤ 13 variables. It differs from the ”expected” series (1 − t2 )2 (1 + t)n with t3 for n = 5, with t4 for n = 7, 8, with t5 for n = 9, with 10t5 for n = 10, with t5 + t6 for n = 11, with 64t6 for n = 12, and with 13t6 + t7 for n = 13. Conjectures There is no difference for two forms of degree 2 from the ”expected” series [(1 − t2 )2 (1 + t)n ] in degrees < [(n + 1)/2]. For two forms of even degree d > 2, the series is the ”expected” [(1 − td )2 (1 + t)n ]. References [1] W. Bruns and J. Herzog, Cohen-Macaulay rings. [2] A. Aramova, J. Herzog, T. Hibi, Gotzmann Theorems for Exterior Algebras and Combinatorics, J. OF ALGEBRA 191, 174–211, 1997. [3] G. Moreno and J. Snellman, SOME CONJECTURES ABOUT THE HILBERT SERIES OF GENERIC IDEALS IN THE EXTERIOR ALGEBRA, Homology, Homotopy and Applications, vol.4(2), 2002, pp.409–426.