H A A}. ) < k, then there are constants c t such that c t α t = 0. j=1 H i j

Transcription

1 M ath. Res. Lett. 16 (2009), no. 1, c International Press 2009 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY Hal Schenck and Ştefan O. Tohǎneanu Abstract. The Orlik-Solomon algebra is the cohomology ring of the complement of a hyperplane arrangement A C n ; it is the quotient of an exterior algebra Λ(V ) on A generators. In [9], Orlik and Terao introduced a commutative analog Sym(V )/I of the Orlik-Solomon algebra to answer a question of Aomoto and showed the Hilbert series depends only on the intersection lattice L(A). In [6], Falk and Randell define the property of 2-formality; in this note we study the relation between 2-formality and the Orlik-Terao algebra. Our main result is a necessary and sufficient condition for 2- formality in terms of the quadratic component I 2 of the Orlik-Terao ideal I. The key is that 2-formality is determined by the tangent space T p(v (I 2 )) at a generic point p. 1. Introduction Let A = {H 1,..., H d } be an arrangement of complex hyperplanes in C n. In [7], Orlik and Solomon showed that the cohomology ring of the complement X = C n \ d i=1 H i is determined by the intersection lattice L(A) = { H A A}. H A The Orlik-Solomon algebra H (X, Z) is the quotient of the exterior algebra E = (Z d ) on generators e 1,..., e d in degree 1 by the ideal generated by all elements of the form e i1...i r := q ( 1) q 1 e i1 ê iq e ir, for which codim H i1 H ir < r. Throughout this paper, we work with an essential, central arrangement of d hyperplanes; this means we may always assume L(A) has rank n and d A = V (α i ) P n 1, i=1 where α i are distinct homogeneous linear forms such that H i = V (α i ). Write [d] for {1,..., d} and let Λ = {i 1,..., i k } [d]. If codim( k j=1 H i j ) < k, then there are constants c t such that c t α t = 0. t Λ In [9], Orlik and Terao introduced the following commutative analog of the Orlik- Solomon algebra in order to answer a question of Aomoto. Received by the editors April 9, Mathematics Subject Classification. 52C35, 13D02, 13D40. Key words and phrases. Hyperplane arrangement, Free resolution, Orlik-Terao algebra. Schenck supported by NSF , NSA

2 172 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU Definition 1.1. For each dependency Λ = {i 1,..., i k }, let r Λ = k j=1 c i j y ij R = K[y 1,..., y d ]. Define f Λ = (r Λ ) = k j=1 c i j (y i1 ŷ ij y ik ), and let I be the ideal generated by the f Λ. The Orlik-Terao algebra OT is the quotient of K[y 1,..., y d ] by I. The Artinian Orlik-Terao algebra is the quotient of OT by y1, 2..., yd 2. Orlik and Terao actually study the Artinian version, but for our purposes the OT algebra will turn out to be more interesting. The crucial difference between the Orlik-Solomon algebra and Orlik-Terao algebra(s) is not the difference between the exterior algebra and symmetric algebra, but rather the fact that the Orlik-Terao algebra actually records the weights of the dependencies among the hyperplanes. So in any investigation where actual dependencies come into play, the OT algebra is the natural candidate for study Formality. In [6], Falk and Randell introduced the concept of 2-formality: Definition 1.2. For an arrangement A, the relation space F (A) is the kernel of the evaluation map φ: d e Ke i α i i K[x1,..., x n ] 1. i=1 A is 2-formal if F (A) is spanned by relations involving only three hyperplanes. Example 1.3. Suppose we have an arrangement of 4 lines in P 2 given by the linear forms: α 1 = x 1, α 2 = x 2, α 3 = x 3, α 4 = x 1 + x 2 + x 3. Obviously any three of the forms are independent, so the only relation is α 1 + α 2 + α 3 α 4 = 0. Hence the OT algebra is K[y 1,..., y 4 ]/ y 2 y 3 y 4 + y 1 y 3 y 4 + y 1 y 2 y 4 y 1 y 2 y 3. This arrangement cannot be 2-formal, since there are no relations involving only three lines, whereas F (A) is nonzero. Many interesting classes of arrangements are 2-formal: in [6], Falk and Randell proved that K(π, 1) arrangements and arrangements with quadratic Orlik-Solomon ideal are 2-formal. In [3], Brandt and Terao generalized the notion of 2-formality to k formality, proving that every free arrangement is k formal. Formality of discriminantal arrangements is studied in [1], with surprising connections to fiber polytopes [2]. In [17], Yuzvinsky shows that free arrangements are 2-formal; and gives an example showing that 2-formality does not depend on L(A). Example 1.4. Consider the following two arrangements of lines in P 2 : A 1 =V (xyz(x+ y+ z)(2x+ y+ z)(2x+ 3y+ z)(2x+ 3y+ 4z)(3x+ 5z)(3x+ 4y+ 5z)) A 2 =V (xyz(x+ y+ z)(2x+ y+ z)(2x+ 3y+ z)(2x+ 3y+ 4z)(x+ 3z)(x+ 2y+ 3z)). For a graded R module M, the graded betti numbers of M are b ij = dim K Tor R i (M, K) i+j. We shall be interested in the case when M = R/I. The graded betti numbers of the Orlik-Terao and Artinian Orlik-Terao algebras of A 1 and A 2 are identical; for the

3 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY 173 Orlik-Terao algebras the graded betti numbers are: total The diagram entry in position (i, j) is simply b ij, so for example, dim K Tor R 2 (R/I, K) 4 = 120. Yuzvinsky shows that A 1 is 2-formal, and A 2 is not. The arrangements have the same intersection lattice and appear identical to the naked eye: Figure 1. The difference is that the six multiple points of A 1 lie on a smooth conic, while the six multiple points of A 2 do not. The quadratic OT algebra is K[y 1,..., y d ]/I 2, where I 2 consists of the quadratic generators of I. The graded betti numbers of the quadratic OT algebra of A 1 are: total while the quadratic OT algebra of A 2 has betti diagram: total This paper is motivated by the question raised by the previous example: Is 2- formality determined by the quadratic OT algebra? Our main result is: Theorem: Let A be an arrangement of d hyperplanes of rank n. A is 2-formal if and only if codim(i 2 ) = d n.

4 174 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU As noted, the class of 2-formal arrangements is quite large, and includes free arrangements, K(π, 1) arrangements, and arrangements with quadratic Orlik-Solomon ideal. The previous theorem gives necessary and sufficient conditions for 2-formality in terms of the Orlik-Terao algebra. We close the introduction with several examples. Definition 1.5. Let G be a simple graph on ν vertices, with edge-set E, and let A G = {z i z j = 0 (i, j) E} be the corresponding arrangement in C ν. For a graphic arrangement A G it is obvious that d = E, and easy to show that rank A G = ν 1. In [16], Tohaneanu showed that a graphic arrangement A G is 2- formal exactly when H 1 ( G ) = 0, where G is the clique complex of G a simplicial complex, whose i-faces correspond to induced complete subgraphs on i + 1 vertices. Example 1.6. The clique complex for G as in Figure 2 consists of the four triangles (and all vertices and edges). Clearly H i ( G ) = 0 for all i 1, so A G is 2-formal. The graded betti numbers of OT are Figure 2. total while the quadratic OT algebra of A G has betti diagram: total Since I 2 is clearly a complete intersection, codim(i 2 ) = 4 = 8 (5 1) = d n, giving another proof of 2-formality for this configuration. While the examples of 2-formal arrangements encountered thus far have I 2 a complete intersection, this is generally not the case.

5 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY 175 Example 1.7. The Non-Fano arrangement is the unique configuration of seven lines in P 2 having six triple points. It is free, hence 2-formal. The graded betti numbers of OT are total while the quadratic OT algebra has betti diagram: total So I 2 is not a complete intersection, though it is Gorenstein. In general, I 2 need not even be Cohen-Macaulay, and often has multiple components. For the Non-Fano arrangement the primary decomposition of I 2 is I 2 = I y 0, y 1, y 3, y 4 In particular, if I I 2, the primary decomposition is nontrivial. When A is 2-formal, it is possible to give a simple necessary and sufficient combinatorial criterion for I 2 to be a complete intersection. Corollary 1.8. Let A be a 2-formal arrangement of rank n, with A = d. Then I 2 is a complete intersection iff ( ) d b 2 = d + n 2 Proof. First, since b 2 is the dimension of the Orlik-Solomon algebra in degree two, b 2 = dim K Λ 2 (K d ) dim K J 2, where J denotes the Orlik-Solomon ideal. Since the Artinian Orlik-Terao algebra has the same Hilbert series as the Orlik-Solomon algebra, this implies Now, clearly b 2 = dim K (R/I + y 2 1,..., y 2 d ) 2 dim K (R/ y 2 1,..., y 2 d ) 2 = dim K Λ 2 (K d ). In the next section, it is proved that I is prime. It follows from this that no y 2 i I 2. Combining, we obtain that dim K I 2 = dim K J 2. Now, I 2 is a complete intersection if codim(i 2 ) = dim K (I 2 ). By the assumption of 2-formality, this concludes the proof.

6 176 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU 2. The quadratic Orlik-Terao algebra and 2-Formality We keep the setup of the previous section: A is an essential, central arrangement of d hyperplanes in P n 1, with relation space F (A). Since dim K F (A) = d n, A is 2-formal if and only iff dim K (span K ({3-relations})) = d n. Fix defining linear forms α i so that H i = V (α i ), let I R = K[y 1,..., y d ] denote the OT ideal, and I 2 the quadratic component of I. Proposition 2.1. I is prime, and V (I) is a rational variety of codimension d n. Proof. Consider the map R Φ K [ 1 α 1,..., ] 1 = C(A) α d given by y i 1 α i. An easy check shows that I ker Φ. Our assumption that A is essential implies that [ ] [ ] 1 K,..., K,...,, y 1 y n α 1 α d hence the field of fractions of R/ ker(φ) is K(y 1,..., y n ), giving rationality and the appropriate dimension (as an affine cone). In [15], Terao proved that the Hilbert series for C(A) is given by ( t ) HS(C(A), t) = P A,. 1 t where P (A, t) is the Poincaré polynomial of A. If H is a hyperplane of A, the deletion A is the subarrangement A \ H of A, and the restriction A is the arrangement {H H H A }, considered as an arrangement in the vector space H, and there is a relation P (A, t) = P (A, t) + tp (A, t). Thus the Hilbert series of C(A) satisfies the recursion HS(C(A), t) = HS(C(A ), t) + t 1 t HS(C(A ), t) If the quotient R/I satisfies the same recursion, then I = ker Φ will follow by induction. Let y 1 be a variable corresponding to H = H 1. In [10], Proudfoot and Speyer prove that the broken circuits are a universal Gröbner basis for I, hence in particular a lex basis. Let R = K[y 1,..., y n ] and R = K[y 2,..., y n ], and consider the short exact sequence 0 R( 1)/(in(I) : y 1 ) y1 R/in(I) R/(in(I), y 1 ) 0. The initial ideal of I has the form with f i, g j not divisible by y 1. Clearly in(i) = f 1,..., f k, y 1 g 1,..., y 1 g m, in(i) : y 1 = f 1,..., f k, g 1,..., g m

7 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY 177 In the rightmost module of the short exact sequence, quotienting by y 1 kills the generator and all relations involving y 1, hence R/(in(I), y 1 ) R / f 1,..., f k = R /in(i ), with I denoting the OT ideal of A. In the leftmost module, since no relation of in(i) : y 1 involves y 1, in(i) : y 1 R, so taking into account the degree shift and the fact that y 1 acts freely on the module, we see HS(R( 1)/(in(I) : y 1 ), t) = t 1 t HS(R / f 1,..., f k, g 1,..., g m, t). We now claim that HS(R / f 1,..., f k, g 1,..., g m, t) = HS(C(A ), t). To see this, note that the initial monomials f k express the fact that the dependencies among hyperplanes of A continue to hold in A. The relations g j represent the collapsing that occurs on restricting to H 1. For example, if there is a circuit (123) in A, this means that H 2 H1 = H 3 H1. In OT(A), the relation on (123) has initial term y 1 y 2, hence y 2 in(i) : y 1. This reflects the redundancy H 2 H1 = H 3 H1 in the restriction. Similar reasoning applies to the relations g i of degree greater than one, leading to: ( HS(R / f 1,..., f k, g 1,..., g m, t) = HS(C(A ), t) = P A t ),. 1 t Combining this with an induction and Terao s formula shows that I = ker Φ. Corollary 2.2. The variety V (I) is nondegenerate. Proof. Since I is prime, V (I) will be contained in a hyperplane V (L) iff L I. However, the assumption that the hyperplanes of A are distinct implies that there are no dependencies involving only two hyperplanes, so that I is generated in degree greater than two. In particular, I contains no linear forms. Theorem 2.3. Let A be an arrangement of d hyperplanes of rank n. A is 2-formal if and only if X = V (I 2 ) (C ) d 1 has codimension d n. Proof. The proof hinges on the fact that for p (C ) d 1 P d 1, the Jacobian ideal of I 2, evaluated at p, is (after a linear change of coordinates) exactly the matrix recording the dependencies among triples of hyperplanes. Let f = ay 2 y 3 + by 1 y 3 + cy 1 y 2 = (ay 1 + by 2 + cy 3 ), a, b, c 0, be a generator of I 2, let p = (p 1 :... : p d ) be a point on X, so mp p i = 0. Such points form a dense open subset of V (I) by Corollary 2.2. The Jacobian matrix of the generators of I 2 has a special form when evaluated at p V (I 2 ) (C ) d 1. Write f i for (f)/ (y i ). For f as above, if i 4 then f i = 0, and Evaluating the partials of f at p yields: f 1 = by 3 + cy 2 f 2 = ay 3 + cy 1 f 3 = ay 2 + by 1 f 1 (p) = p 2 p 3 (b 1 p 2 + c 1 p 3 ), f 2 (p) = p 1 p 3 (a 1 p 1 + c 1 p 3 ), f 3 (p) = p 1 p 2 (a 1 p 1 + b 1 p 2 ).

8 178 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU Since f(p) = 0, a 1 p 1 + b 1 p 2 + c 1 p 3 = 0, and simplifying using this relation, we obtain: f 1 (p) = p 2p 3 p 1 a, f 2 (p) = p 1p 3 p 2 b, f 3 (p) = p 1p 2 p 3 c. The p i are nonzero, so rescaling shows that the row of J p (I 2 ) corresponding to f is: [ a p 2, b 1 p 2, c ] 2 p 2, 0,..., 0 3 Multiplying J p (I 2 ) on the right by the diagonal invertible matrix with (i, i) entry p 2 i yields a matrix whose entries encode the dependencies among triples of the forms l i. Hence rank J p (I 2 ) is exactly the dimension of the space of three relations. Theorem 2.4. A is 2-formal if and only if codim(i 2 ) = d n. Proof. When I = I 2, the result follows from Theorem 2.3. There are two cases to consider. As in Example 1.7, it is possible that I I 2 but dim V (I) = dim V (I 2 ). In this case, the theorem holds, simply from the coincidence of dimensions. The second case is that dim V (I) < dim V (I 2 ). In this case, for a smooth point p V (I 2 ), the dimension of T p (V (I 2 )) will be greater than d n, hence rank J p (I 2 ) < d n. The dimension of the tangent space can only increase at singular points, so rank J p (I 2 ) < d n at all points of V (I 2 ), and thus the three relations cannot span the relation space. Proposition 2.5. If A is supersolvable, then I = I 2. Proof. Fix the reverse lexicographic order on R = K[y 1,..., y d ]. Suppose C is a circuit with C = p 4, and that (C) / I 2. From the set of circuits with (C) / I 2, C = p, select the circuit C = {H j1,..., H jp }, j 1 < < j p which has maximal lead term M = y j2 y jp. Since A is supersolvable, there exists j r, j s, 1 r < s p and u > j s such that D = {H jr, H js, H u } is a circuit. If u {j 1, j 2,..., j p } then C contains D, which would contradict the fact that C is a circuit with C 4. Hence u / {j 1, j 2,..., j p }. So D gives rise to a dependency: yielding an element D = b r α jr + b s α js + α u = 0, b r, b s 0, (D) = y jr y js + b r y js y u + b s y jr y u I 2. Note that C also gives a dependency a 1 α j1 + +a p α jp = 0, a k 0 and corresponding element (C) = a r y j1 ŷ jr y jp + a s y j1 ŷ js y jp + y jr y js P I, with P = a 1 y j2 ŷ jr ŷ js y jp + + a p y j1 ŷ jr ŷ js y jp 1. Consider the dependencies C r = a r D b r C and C s = a s D b s C. Writing y jr y js = (D) b r y js y u b s y jr y u and substituting into the expression for (C) yields (C) = (C r ) + (C s ) + (D)P.

9 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY 179 Now note that C r and C s are circuits of cardinality p with leading terms of (C r ) and (C s ) greater than M, since we replaced the variable y jr or y js by y u, with u > j s > j r. The choice of M and C now implies that (C r ) and (C s ) are both in I 2, a contradiction. 3. Combinatorial Syzygies Example 1.4 shows that the module of first syzygies on I 2 is not determined by combinatorial data. In this section we study linear first syzygies. First, we examine the syzygies which arise from an X L 2 (A) with µ(x) 3. If µ(x) = d 1, then d hyperplanes H 1,..., H d pass thru X. To simplify, we localize to the rank two flat, so that A consists of d points in P 1. Theorem 3.1. Suppose X L 2 (A) has µ(x) = d 1 3 and let I R = K[y 1,..., y d ] K[y 1,..., y n ] be the subideal of I 2 corresponding to A X. The ideal I has an Eagon-Northcott resolution S 2 (R 2 ) Λ 4 R( 1) d 1 (R 2 ) Λ 3 R( 1) d 1 Λ 2 R( 1) d 1 Λ 2 R 2 R/I 0. In particular, the only nonzero betti numbers are ( ) d 1 dim K T or i (R/I, K) i+1 = i. i + 1 Proof. Let X L 2 (A) with µ(x) = d 1. After a change of coordinates, X = V (x 1, x 2 ), and X H iff l H x 1, x 2. Localization is exact, so without loss of generality we may assume that A consists of d points in P 1. By Proposition 2.5, I is quadratic, and by Proposition 2.1, I has codimension d 2, so V (I) is an irreducible curve in P d 1. Since the irrelevant maximal ideal is not an associated prime, I has depth at least one, and by Corollary 2.2, X is not contained in any hyperplane, so y d is a nonzero divisor on R/I. This implies that deg X = deg V (I + y d ). Since A has rank two, any set of three hyperplanes is dependent and thus every triple {H i, H j, H k } yields an element of I. Therefore I, y d = J, y d, where J = {y i y j } 1 i<j d 1. It follows that the primary decomposition of I, y d is I, y d = y i1,..., y id 2, y d, 1 i 1< <i d 2 d 1 so deg V ( I, y d ) = d 1. Any smooth, irreducible nondegenerate curve of degree d 1 in P d 1 is a rational normal curve, and has an Eagon-Northcott resolution [4]. So we need only show that V (I) is smooth. Our main theorem implies A is 2-formal, and the proof of that result shows that dim T p (V (I)) = 1 for all p V (I). Hence V (I) is rational normal curve of degree d 1.

10 180 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU 3.1. Graphic arrangements. For a graphic arrangement, all weights of dependencies are ±1, so the Orlik-Terao ideal I G has a presentation that is essentially identical to that of the Orlik-Solomon ideal obtained in [12]. The proof of the next lemma is straightforward. Lemma 3.2. I G is minimally generated by the chordless cycles of G. Lemma 3.3. Every K 4 subgraph yields two minimal linear first syzygies on I G. Proof. Let K 4 be the complete graph on {1, 2, 3, 4}. There are four relations: y 12 + y 23 y 13, y 12 + y 24 y 14, y 13 + y 34 y 14, y 23 + y 34 y 24. Let y 1 = y 12, y 2 = y 13, y 3 = y 14, y 4 = y 23, y 5 = y 24, y 6 = y 34. I is generated by: f 4 = y 2 y 4 + y 1 y 2 y 1 y 4 f 3 = y 3 y 5 + y 1 y 3 y 1 y 5 f 2 = y 3 y 6 + y 2 y 3 y 2 y 6 f 1 = y 5 y 6 + y 4 y 5 y 4 y 6 A direct calculation yields the pair of (independent) linear syzygies: (y 1 y 2 )f 1 (y 1 + y 5 )f 2 + (y 2 + y 6 )f 3 (y 6 y 5 )f 4 = 0 (y 2 y 3 )f 1 (y 4 y 5 )f 2 + (y 4 y 2 )f 3 (y 5 y 3 )f 4 = 0 A Hilbert function computation as in Corollary 1.8 concludes the proof. Theorem 3.4. If κ i is the number of induced subgraphs of type K i+1 G, then κ 2 i = 1 T or i (R/I G ) i+1 = 2κ 3 i = 2 0 i 3. Proof. Combine a Hilbert function argument as in Corollary 1.8 with suitable modifications to the proofs of Corollary 6.6 and Lemma 6.9 of [12] The spaces R k (A). In [3], Brandt-Terao introduce higher relation spaces. Definition 3.5. For X L 2 (A), let F (A X ) be the subspace of F (A) generated by the relations associated to circuits of length 3 {H i, H j, H k }, with X H i, H j, H k. The inclusion map F (A X ) F (A) gives a map π : F (A X ) F (A). X L 2(A) The first relation space R 3 (A) = ker π. The space R 3 (A) captures the dependencies among the circuits of length 3 in A, and A is 2-formal iff π is surjective. It is clear that dim F (A X ) = µ(x) 1, where µ is the Möbius function. Proposition 3.6. Let X 1,..., X s L 2 (A) and let r 1 F (A X1 ),..., r s F (A Xs ) be nonzero relations. If L 1 (r 1 ) + + L s (r s ) = 0, L i 0 is a linear syzygy, then there exist a i K with a 1 r a s r s = 0.

11 THE ORLIK-TERAO ALGEBRA AND 2-FORMALITY 181 Proof. Suppose L 1 (r 1 ) + + L s (r s ) = 0, L i 0 is a linear syzygy, with L i = a i 1y 1 + +a i d y d. Define supp(l i ) to be the set of indices j for which a i j 0, and suppose supp(l i) supp(r i ) for some i. In the expression of L i (r i ) there exists a nonzero monomial of the form y u y 2 v with H u H v = X i. Since in the syzygy this monomial must be cancelled, there exists k i such that L k (r k ) contains a term y u y 2 v. Since y 2 v cannot occur in (r k ), we see that X k = H u H v, contradicting X k X i. Let Λ i = supp(r i ). Substituting ( 1 y 1,..., 1 y d ) in the syzygy yields: f 1 r f s r s = 0, y [d]\λ1 y Λ1 y [d]\λs y Λs f where i y [d]\λi = L i ( 1 y 1,..., 1 y d ) and f i K[y 1,..., y d ] are square-free non-zero polynomials, possibly divisible by a square-free monomial. Hence f 1 r f s r s = 0 and f 1 r 1 r 2,..., r s. If r 1 / r 2,..., r s, then f 1 r 2,..., r s and f 1 = P 2 r P s r s. Evaluating this expression at ( 1 y 1,..., 1 y d ) and clearing the denominators shows there exists a monomial m such that ml 1 (r 2 ),..., (r s ) I. By Proposition 2.1 and Corollary 2.2, I is nondegenerate prime ideal, so this is impossible, and r 1 r 2,..., r s. Acknowledgements Computations were performed using Macaulay2, by Grayson and Stillman, and available at: References [1] M. Bayer, K. Brandt, Discriminantal Arrangements, Fiber Polytopes and Formality, J. Alg. Comb., 6 (1997), [2] L. Billera, B. Sturmfels, Fiber polytopes, Ann. of Math, 135 (1992), [3] K. Brandt, H. Terao, Free Arrangements and Relation Spaces, Discrete Comput. Geom., 12 (1994), [4] D. Eisenbud, The geometry of syzygies, Graduate Texts in Mathematics, vol. 229, Springer- Verlag, Berlin-Heidelberg-New York, [5] M. Falk and R. Randell, On the homotopy theory of arrangements, Adv. Stud. Pure. Math. 8 (1986), [6] M. Falk and R. Randell, On the homotopy theory of arrangements II, Adv. Stud. Pure. Math. 27 (2000), [7] P. Orlik, L. Solomon, Combinatorics and topology of complements of hyperplanes, Invent. Math. 56 (1980), [8] P. Orlik, H. Terao, Arrangements of Hyperplanes, Grundlehren Math. Wiss., Bd. 300, Springer- Verlag, Berlin-Heidelberg-New York, [9] P. Orlik, H. Terao, Commutative algebras for arrangements, Nagoya Math. J., 134 (1994), [10] N. Proudfoot, D. Speyer, A broken circuit ring, Beiträge Algebra Geom., 47 (2006), [11] B. Shelton, S. Yuzvinsky, Koszul algebras from graphs and hyperplane arrangements, J. London Math. Soc. (2) 56 (1997), [12] H. Schenck, A. Suciu, Lower central series and free resolutions of hyperplane arrangements, Trans. Amer. Math. Soc. 354 (2002),

12 182 HAL SCHENCK AND ŞTEFAN O. TOHǍNEANU [13] H. Schenck, A. Suciu, Resonance, linear syzygies, Chen groups, and the Bernstein-Gelfand- Gelfand correspondence, Trans. Amer. Math. Soc. 358 (2006), [14] A. Suciu, Fundamental groups of line arrangements: Enumerative aspects, Contemporary Math. 276 (2001), AMS, Providence, RI, [15] H. Terao, Algebras generated by reciprocals of linear forms, J. Algebra, 250 (2002), [16] S. Tohaneanu, Topological criteria for k formal arrangements, Beiträge Algebra Geom., 48 (2007), [17] S. Yuzvinsky, First two obstructions to the freeness of arrangements, Trans. Amer. Math. Soc. 335 (1993), Schenck: Mathematics Department, University of Illinois, Urbana, IL 61801, USA address: schenck@math.uiuc.edu Tohaneanu: Math Department, University of Cincinnati, Cincinnati, OH 45221, USA address: stefan.tohaneanu@uc.edu