Hindawi Publishing Corporation

International Journal of Mathematics and Mathematical Sciences Volume 2012, Article ID 254791,22 pages doi:10.1155/2012/254791

Research Article

Subring Depth, Frobenius Extensions, and Towers

Lars Kadison

Departamento de Matematica, Faculdade de Ciencias, Universidade do Porto, Rua do Campo Alegre 687, 4169-007 Porto, Portugal

Correspondence should be addressed to Lars Kadison, lkadison@c2i.net Received 24 March 2012; Accepted 23 April 2012 Academic Editor: Tomasz Brzezinski

Copyright © 2012 Lars Kadison. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The minimum depth d(B,A) of a subring B C A introduced in the work of Boltje, Danz and Kulshammer (2011) is studied and compared with the tower depth of a Frobenius extension. We show that d(B,A) < to if A is a finite-dimensional algebra and Be has finite representation type. Some conditions in terms of depth and QF property are given that ensure that the modular function of a Hopf algebra restricts to the modular function of a Hopf subalgebra. If A D B is a QF extension, minimum left and right even subring depths are shown to coincide. If A D B is a Frobenius extension with surjective Frobenius, homomorphism, its subring depth is shown to coincide with its tower depth. Formulas for the ring, module, Frobenius and Temperley-Lieb structures are noted for the tower over a Frobenius extension in its realization as tensor powers. A depth 3 QF extension is embedded in a depth 2 QF extension; in turn certain depth n extensions embed in depth 3 extensions if they are Frobenius extensions or other special ring extensions with ring structures on their relative Hochschild bar resolution groups.

1. Introduction and Preliminaries

A basic lemma in representation theory states that if a subalgebra B of a finite-dimensional algebra A has ¡1 : A®BA ^ A, a ® a' ^ aa' a split epimorphism of A-A-bimodules, then A has finite representation type if B has. Weakening the condition on ¡1 to a split epimorphism of A-B-bimodules does not place any restriction on B C A, but the opposite hypothesis that a split monomorphism exists from A®gA into a multiple nA = A e ••• e A captures the notion of normality of a subalgebra in the context of group algebras [1], Hopf algebras [2], and semisimple algebras [3]. If A is a Frobenius extension of B, where AB is a progenerator module (but A and B may be infinite-dimensional algebras), the "depth two" condition as the opposite hypothesis is known as, implies that A is a Galois extension of B, where the bimodule endomorphism ring of the extension may be given the structure of a Hopf algebroid (which acts naturally on A with invariant subalgebra B) [4,5]. Such theorems first appeared

in [6, 7] for certain finite index subfactors of depth two. The left bialgebroid aspect of the definition of Hopf algebroid was influenced by a study of Lie groupoids in Poisson geometry [8]. The publication of [9] clarified the role played by Galois theory in depth two theory.

After the focus on depth two, the study of how to generalize depth three and more from subfactor theory to algebra occurred in three stages after [10]. At first the depth two condition was generalized from a subalgebra pair B c A to a tower of three rings C c B c A [11]. This was applied to the tower of iterated right endomorphism rings above a Frobenius extension B c A c A1 ^ A2 ^ •••, so that B c A has (tower) depth n if B ^ An-3 ^ An-2 has the generalized depth two property (called a depth 3 tower in [11]). This yields a compact matrix inequality condition

M[n+1] < qM[n-1]

(some q e N ) for when a subalgebra pair of semisimple complex algebras has depth n in terms of the inclusion matrix M, equivalently the incidence matrix of the Bratteli diagram of the inclusion B ^ A [3, 18]. Since M[2] = MM', M[3] = MMtM,..., already in this matrix condition the odd and even depth become distinguished from one another in terms of square and rectangular matrices. From [3], Boltje et al. [12] have extended the definition to a subring B c A, which has (right) depth 2n if the relative Hochschild n+1 bar resolution group Cn+1(A,B) maps as a split monomorphism into a multiple of a smaller group, qCm(A,B) as A-B-bimodules, and depth 2n +1 if this condition only holds as natural B-B-bimodules. Since subring B c A having depth m implies that it has depth m +1, the minimum depth d(B, A) is the more interesting positive integer.

The algebraic definition of depth of subring pairs of Artin algebras is closely related to induced and restricted modules or characters in the case of group algebras. The depths of several class subgroups are recently computed, both as induced complex representations [3] and as induced representations of group algebras over an arbitrary ground ring [12]. For example, the minimum depth of the permutation groups Sn c Sn+1 is 2n - 1 over any ground ring k and depends only on a combinatorial depth of a subgroup H < G defined in terms of G x H-sets and diagonal action in the same way as depth is defined for a subring [12]. The main theorem in [12] is that an extension k[G] 2 k[H] of finite group algebras over any ground ring k has finite depth, in fact bounded by twice the index [G : NG (H)] of the normalizer subgroup.

The notion of subring depth d(B,A) in [12] is defined in equivalent terms in (1.7). In case B and A are semisimple complex algebras, it is shown in an appendix of [12] how subring depth equals the notion of depth based on induction-restriction table, equivalently inclusion matrix M in [3] and given in (1.1). Such a pair A D B is a special case of a split, separable Frobenius extension; in Theorem 5.2 we show that subring depth is equal to the tower depth of Frobenius extensions [11] satisfying only a generator module condition. The authors of [12] define left and right even depth and show these are the same on group algebra extensions; Theorem 3.2 shows this equality holds for any quasi-Frobenius (QF) extension.

It is intriguing that the definition of subring depth makes use of the bar resolution groups of relative homological algebra, although in a fundamentally different way. The tower of iterated endomorphism rings above a ring extension becomes in the case of Frobenius extensions a tower of rings on the bar resolution groups Cn(A,B) (n = 0,1,2,...) with Frobenius and Temperley-Lieb structures explicitly calculated from their more usual iterative definition in Section 4.1. At the same time Frobenius extensions of depth more than 2 are

known to have depth 2 further out in the tower: we extend this observation in [11] with different proofs to include other ring extensions satisfying the hypotheses of Proposition 4.3. In Section 1 it is noted that a subalgebra B of a finite-dimensional algebra A has finite depth if its enveloping algebra Be has finite representation type.

1.1. H-Equivalent Modules

Let A be a ring. Two left A-modules, AN and AM, are said to be h-equivalent, denoted by

AM ~A N, if two conditions are met. First, for some positive integer r, N is isomorphic to a direct summand in the direct sum of r copies of M, denoted by AN e *~AMr ^

N | rM ^ 3fi e Hom(AM,AN), g e Hom(ANAM) : £f ◦ gt = id^ (1.2)

Second, symmetrically there is s e Z + such that M | sN. It is easy to extend this definition of h-equivalence (sometimes referred to as similarity) to h-equivalence of two objects in an

abelian category and to show that it is an equivalence relation.

If two modules are h-equivalent, AN ~A M, then they have Morita equivalent endomorphism rings, ¿N := EndAN and ¿M := EndAM, since a Morita context of bimodules is given by H(M,N) := Hom(AM,AN), which is an ¿N-¿M-bimodule via composition, and the bimodule ¿MH(N,M)En; these are progenerator modules, by applying to (1.2) or its reverse, M | sN, any of the four Hom-functors such as Hom (A - ,AM) from the category of left A-modules into the category of left EM-modules. Then, the explicit conditions on mappings for h-equivalence show that H(M,N)®EmH(N,M) — ¿N and the reverse mapping given by composition are surjections.

The theory of h-equivalent modules applies to bimodules tMb ~t NB by letting A = T®ZB°P, which sets up an equivalence of abelian categories between T-B-bimodules and left A-modules. Two additive functors F,G : C — D are h-equivalent if there are natural split epis F(X)n G(X) and G(X)m F(X) for all X in C. We leave the proof of the lemma below as an elementary exercise.

Lemma 1.1. Suppose two A-modules are h-equivalent, M ~ N, and two additive functors from

A-modules to an abelian category are h-equivalent, F ~ G. Then, F(M) ~ G(N).

For example, the following substitution in equations involving the ~-equivalence relation follows from the lemma:

aPt ~a Qt, tUb ~t Vb aP®tUb ~ aQ®tVb. (1.3)

Example 1.2. Suppose A is a finite-dimensional algebra with indecomposable A-modules {Pa | a e I} (representatives from each isomorphism class for some index set I). By Krull-Schmidt finitely generated modules MA and NA have a unique factorization into a direct sum of multiples of finitely many indecomposable module components. Denote the indecomposable constituents of MA by Indec (M) = {Pa | [Pa,M] = 0}, where [Pa,M] is

the number of factors in M isomorphic to Pa. Note that M | qN for some positive q if and only if Indec(M) c Indec(N). It follows that M ~ N if and only if Indec(M) = Indec(N).

Suppose Aa = n1P1 e ■■■ e nrPr is the decomposition of the regular module into its projective indecomposables. Let PA = P1 e ■■■ePr. Then, PA and AA are h-equivalent, so that A and End PA are Morita equivalent. The algebra End PA is the basic algebra of A.

1.2. Depth Two

A subring pair B c A is said to have left depth 2 (or be a left depth two extension [4])

if A®BA ~ A as natural B-A-bimodules. Right depth 2 is defined similarly in terms of h-equivalence of natural A-B-bimodules. In [4] it was noted that the left condition implies the right and conversely if A is a Frobenius extension of B. Also in [4] a Galois theory of Hopf algebroids was defined on the endomorphism ring H := End bAb as total ring and the centralizer R := AB as base ring. The antipode is the natural anti-isomorphism stemming from following the arrows:

EndAB A®BA (EndBA)op (1.4)

restricted to the intersection End BAB = End AB n End BA.

The Galois extension properties of a depth two extension A D B are as follows. If AB is faithfully flat, balanced or B equals its double centralizer in A, the natural action of H on A has invariant subalgebra AH satisfying the Galois property of AH = B. Also the well-known Galois property of the endomorphism ring as a cross-product holds: the right endomorphism ring End AB = A#H, where the latter has smash product ring structure on A®RH [4]. There is also a duality structure by going a step further along in the tower above B c A — End AB — End A®BAA, where the Hopf algebroid H' := (A®BA)B is the R-dual of H and acts naturally on EndAB in such a way that End(A®BA)A has a smash product ring structure [4] .

Conversely, Galois extensions have depth 2. For example, an H-comodule algebra A with invariant subalgebra B and finite-dimensional Hopf algebra H over a base field k, which has a Galois isomorphism from A®BA -— A®kH given by a' 0 a — a'a(0) 0 fl(1), satisfies (strongly) the depth two condition A®BA = AdimH as A-B-bimodules. The Hopf subalgebras within a finite-dimensional Hopf algebra, which have depth 2, are precisely the normal Hopf subalgebras; if normal, it has depth 2 by applying the observation about Hopf-Galois extension just made. The converse follows from an argument noted in Boltje-Kulshammer [2], which divides the normality notion into right and left (like the notion of depth 2), where left normal is invariance under the left adjoint action. In the context of an augmented algebra A their results extend to the following proposition. Let e : A — k be an algebra homomorphism into the ground field k. Let A+ denote ker e, and, for a subalgebra B c A, let B+ denote ker e n B.

Proposition 1.3. Suppose B c A is a subalgebra of an augmented algebra. If B c A has right depth 2, then AB+ c B+A.

The proof of this proposition is an exercise in tensoring both sides of A®BAe* ~ qA by the unit A-module k, then passing to the annihilator ideal of a module and a direct summand. The opposite inclusion is of course satisfied by a left depth 2 extension of augmented algebras.

Example 1.4. Let A = Tn(k) be the algebra of n by n upper triangular matrices where n > 1, and B = Dn(k) the subalgebra of diagonal matrices. Note that there are n augmentations si : A — k given by si(X) = Xii, and each of the B+ satisfies the inclusions above if left or right depth two. This is a clear contradiction, thus d(B,A) > 2. We will see below that d(B,A) = 3.

Also subalgebra pairs of semisimple complex algebras have depth 2 exactly when they are normal in a classical sense of Rieffel. The theorem in [3] is given below and one may prove the forward direction in the manner indicated for the previous proposition.

Theorem 1.5 ([3] Theorem 4.6). Suppose B c A is a subalgebra pair of semisimple complex algebras. Then, B c A has depth 2 if and only if, for every maximal ideal I in A, one has A(I n B) = (I n B)A.

For example, subalgebra pairs of semisimple complex algebras that satisfy this normality condition are then by our sketch above examples of weak Hopf-Galois extensions, since the centralizer R mentioned above is semisimple (see Kaplansky's Fields and Rings for a C*-theoretic reason), the extension is Frobenius [18], and weak Hopf algebras are equivalently Hopf algebroids over a separable base algebra [4].

1.3. Subring Depth

Throughout this paper, let A be a unital associative ring and B c A a subring where 1B = 1A. Note the natural bimodules BAB obtained by restriction of the natural A-A-bimodule (briefly A-bimodule) A, also to the natural bimodules BAA, AAB or BAB, which are referred to with no further ado.

Let Q(A,B) = B, and, for n > 1,

Cn(A,B) = A®b •■■<s>bA (n times A), (1.5)

For n > 1, Cn(A,B) has a natural A-bimodule structure, which restricts to B-A-, A-B-, and B-bimodule structures occurring in the next definition.

Definition 1.6. The subring B c A has depth 2n + 1 > 1 if as B-bimodules Cn(A,B) ~

Cn+1(A,B). The subring B c A has left (resp., right) depth 2n > 2 if Cn(A,B) ~ Cn+t(A,B) as B-A-bimodules (resp., A-B-bimodules).

It is clear that if B c A has either left or right depth 2n, it has depth 2n + 1by restricting the h-equivalence condition to B-bimodules. If it has depth 2n + 1, it has depth 2n + 2 by tensoring the h-equivalence by -®BA or A®B-. The minimum depth is denoted by d(B, A); if B c A has no finite depth, write d(B,A) = to.

Note that the minimum left and right minimum even depths may differ by 2 (in which case d(B, A) is the lesser of the two). In the next section we provide a general condition, which

includes a Hopf subalgebra pair B c A of symmetric (Frobenius) algebras, where the left and right minimum even depths coincide.

Also note that a subalgebra pair of Artin algebras B c A have depth 2n + 1 if and only if the indecomposable module constituents of Cn+m(A,B) remain the same for all m > 0 as those already found in Cn (A,B) (see Example 1.2). This corresponds well with the classical notion of finite depth in subfactor theory.

Example 1.7. Again let A = Tn(k) and B = Dn(k) = kn, where n > 1. Let ej denote the matrix units, ki the n simple B-modules, and kij for 1 < i < j < n the n(n + 1)/2 simple components of BAB. Note that A®BA as a B-bimodule has components keiS0BeSj = kij where i < s < j, so A0BA | nA as B-bimodules. Thus, d(B, A) < 3. But d(B, A) = 2 by the remark following Proposition 1.3; then d(B,A) = 3.

1.4. H-Depth

A subring B c A has H-depth 2n - 1 if Cn+1(A,B) ~ Cn(A,B) as A-A-bimodules (n = 1,2,3,...). Note that B has H-depth 2n - 1 in A implies that it has H-depth 2n + 1 (also that it has depth 2n). Thus, define the minimum H-depth dH(B,A) if it exists. Note that the definition of H-depth 2n - 1 is equivalent to the condition on a subring B c A that Cn+1(A,B) | qCn(A,B) for some q e N. This is clear for n = 2 since Cn(A,B) | Cn+1(A,B). For n = 1, the H-separability condition

aA^bAa e* ~ AAAq (1.6)

implies the separability condition AAA e* ~ AA®BAA as argued in the paper [13] by Hirata. The notion of H-depth is studied in [14] where it is noted that |dH(B, A) - d(B,A)| < 2ifone or the other minimum depth is finite. See Section 2 for which Hopf subalgebras satisfy the dH(B,A) = 1 condition in (1.6).

Remark 1.8. Suppose B is a subring of A. The minimum depth of the subring B c A as defined in Boltje-Danz-Kulshammer [12] coincides with d(B,A). In fact, for n > 0, the depth 2n + 1 condition in [12] is that for some q e Z+

Cn+1(A,B) | qCn(A,B) (1.7)

as B-bimodules. The left depth 2n condition in [12] is (1.7) more strongly as natural B-A-bimodules (and as A-B-bimodules for the right depth 2n condition). But (using a pair of classical face and degeneracy maps of homological algebra) we always have Cn(A,B) | Cn+1(A,B) as A-B-, B-A-, or B-bimodules, so that the depth 2n as well as 2n + 1 conditions coincide in the case of subring having depth 2n and 2n + 1 conditions above.

Note that depth 1 in this paper is equivalent to the subring depth 1 notion in, for example, [4, 12, 15] since A is h-equivalent to B as B-bimodules if and only if A is centrally projective over B (i.e., A | qB as B-bimodules). This follows from the lemma below.

Lemma 1.9. Suppose B is a subring of ring A such that BAB | mBBB for some integer m > 1. Then, bBb |bAb .

Proof. From the central projectivity condition on A, one obtains m maps hi e Hom(BAB,BBB) and m maps gi e Hom(BBB,BAB) — vi e AB such that ^m=1 vihi(a) = a for every a e A. It follows that A ~ B®Z(B) AB since hi(AB) c Z(B). Note that restricting the equation to the centralizer AB shows that AB is a finitely generated projective Z(B)-module. But Z(B) c AB is a commutative subring, whence AB is a generator Z(B)-module. From Z(B) e* = nAB for some positive integer n, it follows from the tensor algebra decomposition of A that BBB | nBAB. Whence there are n maps fi e Hom(BAB,BBB) and n elements ri e AB such that Xn=1 fi(ri) = 1A. Define a (condition expectation or) bimodule projection E(a) := ^n=1 fi(ari) of A onto B. □

Example 1.10. The paper [12] asks in its introduction about the depth d(B,A) of invariant subrings in classical invariant theory, where K is a field, A = K[X1,...,Xn], B = k[X1,.. .,Xn]G and G is a finite group in GLn(K) acting by linear substitution of the variables. In any case AB is finitely generated and B is a finitely generated affine K-algebra. We note here that if G is generated by pseudoreflections (such as G = Sn, the symmetric group) and the characteristic of K is coprime to |G|, B is itself an n-variable polynomial algebra and A is a free B-module; consequences of the Shephard-Todd Theorem [16, 17]. Since A is a commutative algebra, it follows that d(B, A) = 1.

Example 1.11. Let B c A be a subring pair of semisimple complex algebras. Then, the minimum depth d(B,A) may be computed from the inclusion matrix M, alternatively an r-by-s induction-restriction table of rB-simples induced to nonnegative integer linear combination of sA-simples along rows, and by Frobenius reciprocity, columns show restriction of A-simples in terms of B-simples. The procedure to obtain d(B,A) given in the paper [3] is to compute the bracketed powers of M given in Section 1, and check for which nth power of M satisfies the matrix inequality in (1.1): d(B,A) is the least such n by results in [12, appendix] (or Theorem 5.2 below combined with [3, 18]). One may note that d(B,A) < 2d - 1 where MM' has degree d minimal polynomial [3]. A GAP subprogram exists to compute d(B,A) for a complex group algebra extension by converting character tables to an induction-restriction table M, then counting the number of zero entries in the bracketed powers of M, which decreases nonstrictly with increasing even and odd powers of M, d(B, A) being the least point of no decrease.

In terms of the bipartite graph of the inclusion B c A, d(B,A) is the lesser of the minimum odd depth and the minimum even depth [3]. The matrix M is an incidence matrix of this bipartite graph if all entries greater than 1 are changed to 1, while zero entries are retained as 0: let the B-simples be represented by r black dots in a bottom row of the graph and A-simples by s white dots in a top row, connected by edges joining black and white dots (or not) according to the 0-1-matrix entries obtained from M. The minimum odd depth of the bipartite graph is 1 plus the diameter in edges of the row of black dots (indeed an odd number), while the minimum even depth is 2 plus the largest of the diameters of the bottom row where a subset of black dots under one white dot is identified with one another.

For example, let A = CS4, the complex group algebra of the permutation group on four letters, and B = CS3. The inclusion diagram pictured in Figure with the degrees of the irreducible representations is determined from the character tables of S3 and S4 or

the branching rule (for the Young diagrams labelled by the partitions of n and representing the irreducibles of Sn).

This graph has minimum odd depth 5 and minimum even depth 6, whence d(B,A) = 5.

Example 1.12. The induction-restriction table M of the inclusion of permutation groups Sn X Sm < Sn+m via

(c,t)-(i ••• n) n+* ••• n+M (i.8)

\c( 1 ) ••• c(n) n + t( 1 ) ••• n + t(m)J

may be computed combinatorially from the Littlewood-Richardson coefficients cYiv e N, where i is partition of n, v = (v 1,... ,vm) a partition of m, and X a partition of n + m. Briefly, the coefficient number c¡v is zero if y does not contain ¡i or is the number of Littlewood-Richardson fillings with content v of y with ¡i removed. A Littlewood-Richardson filling of a skew Young tableau is with integers i = 1 ,2,...,m occuring vi times in rows that are weakly increasing from left to right, columns are strictly increasing from top to bottom, and the entries when listed from right to left in rows, top to bottom row, form a lattice word [ 19].

For example, computing the matrix M for the subgroup S2 x S3 < S5 with respect to the ordered bases of irreducible characters of the subgroup X(12) x ¡(13), X(12) x ¡(2/1), X(12) x¡(3), X(2) x ¡(13), X(2) x ¡(2,1), X(2) x ¡(3) and of the group 7(155,^(2,13), ,1), 7(3,2), T(3,12), T(4,1), T(5) yields

/1 1 1 0 0 0 0\

0 11110 0 0 0 0 0 1 1 0 0 10 0 10 0 0 0 11110 \0 001011/

The bracketed powers of M satisfy a minimum depth 5 inequality (1.1) so that d(S2 x S3, S5) = 5. We mentioned before that d(Sn x S1, Sn+1) = 2n - 1 [3, 12]; however, a formula for d(Sn x Sm, Sn+m) is not known.

1.5. Finite Depth and Finite Representation Type

For the next proposition we adopt the notation Be for the (enveloping) algebra B®kE>op and recall that a finite-dimensional algebra has finite representation type if it only has finitely many isomorphism classes of indecomposable modules.

For example, a group algebra over a base field of characteristic p has finite representation type if and only if its Sylow p-subgroup is cyclic. Thus, B having finite representation type does not imply that Be has finite representation type.

Proposition 1.13. Suppose B c A is a subalgebra pair of finite-dimensional algebras where Be has in all r indecomposable Be-module isomorphism classes. Then, d(B,A) < 2r + 1.

Proof. This follows from the observation in Example 1.2 that since Cn(A,B) is the image of Cn+1(A,B) under an obvious split epimorphism of Be-modules (equivalently, B-bimodules), there is an increasing chain of subset inclusions

Indec(A) c Indec(A®BA) c Indec(A®BA®BA) c ... , (1.10)

which stops strictly increasing in at most r steps. When Indec(Cn(A, B)) = Indec(Cn+1 (A, B)), then Cn (A,B) ~ Cn+1 (A, B) as Be-modules, whence A D B has depth 2n + 1 < 2r + 1. □

Remarkably, the result in [12] is that all finite group algebra pairs have finite depth. The proposition says something about finite depth of interesting classes of finite-dimensional Hopf algebra pairs B c A, where research on which Hopf algebras have finite representation type is a current topic (although the paper [20] studies how tensor algebras seldom have finite representation type when the component algebras are not semisimple). (Note that Be is a Hopf algebra and semisimple if B is so.) For example, we have the following corollary.

Corollary 1.14. Suppose B is a semisimple Hopf subalgebra in a finite dimensional Hopf algebra A. Suppose that B has n nonisomorphic simple modules. Then, d(B, A) < 2n2 + 1.

2. When Frobenius Extensions of the Second Kind Are Ordinary

A (proper) ring extension A D B is a subring or more generally a monomorphism i: B — A, which is equivalent to a subring i(B) c A. Restricted modules such as Ai(B) and pullback modules AB are identified, and these are the type of modules we refer to below unless otherwise stated. (Almost all that we have to say holds for a ring homomorphism B — A and its pullback modules such as AB; however, certain conditions needed below such as AB is a generator imply that B — A is monic.)

A ring extension A D B is a left QF extension if the module BA is finitely generated projective and the natural A-B-bimodules satisfy A | qHom(BA,BB) for some positive integer q. A right QF extension is oppositely defined. A QF extension A D B is both a left and right QF extension and may be characterized by both AB and BA being finite projective, and two

hh h-equivalences of bimodules given by AAB ~a Hom(BA,BB)B and (BAA ~b Hom (AB,BB)a

[21]. For example, a Frobenius extension A D B is a QF extension since it is left and right

finite projective and satisfies the stronger conditions that A is isomorphic to its right B-dual

A* and its left B-dual *A as natural B-A-bimodules, respectively A-B-bimodules; the more

precise definition is given in the next section.

2.1. ¡5-Frobenius Extensions

In Hopf algebras and quantum algebras, examples of Frobenius extensions often occur with a twist foreseen by Nakayama and Tzuzuku, their so-called beta-Frobenius extension or Frobenius extensions of the second kind. Let ¡5 be an automorphism of the ring B and B c A a subring pair. Denote the pullback module of a module BM along ¡5 : B — B by ¡M,

the so-called twisted module. A ring extension A D B is a p-Frobenius extension if AB is finite projective and there is a bimodule isomorphism bAa = pHom(AB/BB). One shows that A D B is a Frobenius extension if and only if p is an inner automorphism. A subring pair of Frobenius algebras B c A is p-Frobenius extension so long as AB is finite projective and the Nakayama automorphism nA of A stabilizes B, in which case p = nB o nA [22]. For instance a finite-dimensional Hopf algebra A = H and B = K a Hopf subalgebra of H are a pair of Frobenius algebras satisfying the conditions just given: the formula for p reduces to the following given in terms of the modular functions of H and K and the antipode S [23, 7.8]: for x € K,

p(x) = £ OTh(x(1)) OTk(S(x(2))) x(3)/ (2.1)

Given the bimodule isomorphism above B AA—>pHom(AB, BB), apply it to 1A and let its value be E : A — B, which then is a cyclic generator of p Hom(AB,BB)A satisfying E(b1ab2) = p(b1)E(a)b2 for all b1,b2 € B,a € A. If xl,...,xm € A and € Hom(AB,BB) are

projective bases of AB, and Eyj := E(yj-) = fyj the equations

ZxiE(yja) = a' j=l

E(axj))yj = a

hold for all a € A. Call (E,xj,yj) a p-Frobenius coordinate system of A D B. Note that also BA is finite projective, that a p-Frobenius coordinate system is equivalent to the ring extension A | B being p-Frobenius and that p = idB if B is in the center of A. Additionally, one notes that there is an automorphism n of the centralizer subring AB such that E(ac) = E(n(c)a) for all a € A and c € AB. Also an isomorphism Ap®BA = EndAB is easily defined from the data and equations above, where xj ® yj — idA, so that if (E, zir wi) is another p-Frobenius coordinate system (sharing the same E : A — B), then ^i zi0Bwi = ^j xj0Byj in (Ap®bA)a.

When a p-Frobenius extension is a QF extension is addressed in the next proposition.

Proposition 2.1. A p-Frobenius extension A D B is a left QF extension if and only if there are ui,vi € A(i = l,...,n) such that sui = uip(s) and vis = p(s)vi for all i and s € B, and

p-l(s)^ uisvi. (2.3)

Proof. Suppose A D B is p-Frobenius extension with p-Frobenius system satisfying the equations above. Given the elements ui, vi € A satisfying the equations above, let Ei = E(ui-), which defines n mappings in (the untwisted) Hom(BAB,BBB). Also define n mappings fi € Hom(A(*A)B,AAB) by fi(g) = Y!m=l xjg(viyj) where it is not hard to show using the p-Frobenius coordinate equations that Xj xj®Bvyj € (A®BA)A for each i (a Casimir element). It follows that Xn=l fi(Ei) = lA and that A | n(*A) as natural A-B-bimodules, whence A is a left QF extension of B.

Conversely, assume the left QF condition BA*A | An, equivalent to AAB | n(*A) by applying the right B-dual functor and noting (*A)* = A as well *(A*) ~ A. Also assume the slightly rewritten ¡-Frobenius condition ¡-1AA ~ B(A*)A, which then implies ¡-1AA | nA. So there are n mappings gi e Hom (¡-1 AA,BAA) and n mappings fi e Hom(BAA,^-1 AA) such that Xn=1 fi 0 gi = idA. Equivalently, with ui := f (1A) and vi := g(1A), 2n=1 uivi = 1A, and the equations in the proposition are satisfied. □

The following corollary weakens one of the equivalent conditions in [24,25]. It implies that a finite dimensional Hopf algebra that is QF over a Hopf subalgebra is necessarily Frobenius over it. (Nontrivial examples of QF extensions occur for weak Hopf algebras over their separable base algebra [26].)

Corollary 2.2. Let H be a finite dimensional Hopf algebra and K a Hopf subalgebra. In the notation of (2.1) the following are equivalent.

(1) The automorphism ¡) = idK and H D K is a Frobenius extension.

(2) The algebra extension H D K is a QF extension.

(3) The modular functions mH (x) = mK (x) for all x e K.

Proof. (1 ^ 2) A Frobenius extension is a QF extension. (2 ^ 3) Set s = 1 in (2.3), and apply the counit e to see that e(£i uivi) = 1. Reapply e to (2.3) to obtain e 0 ¡3 = s. Apply e to (2.1), and use uniqueness of inverse in convolution algebra Hom (K, k), where mK 0 S = mK! and 1 = s, to show that mH = mK on K. (3 ^ 1) This follows from (2.1). □

The following observation for a normal Hopf subalgebra K c H has not been explicitly noted before in the literature.

Corollary 2.3. The modular function of a finite dimensional Hopf algebra H restricts to the modular function of a Hopf subalgebra K c H if K has depth d(K,H) < 2.

Proof. If the Hopf subalgebra K has depth 1 in H, it has depth 2. If it has depth 2, it is equivalently a normal Hopf subalgebra by the result of [2]. But a normal Hopf subalgebra K c H is an H-Galois extension: here H := H/HK+ denotes the quotient Hopf algebra, H — H, h — h denotes the quotient map, and the Galois isomorphism can : H0KH — H 0 H is given by can (h 0 h') = hh(1) 0 h'(2) [27]. In the same paper [27] it is shown that a Hopf-Galois extension of a finite dimensional Hopf algebra is a Frobenius extension. Then, ft = id in the corollary above, so mK = mH |K. □

The corollary extends to some extent to quasi-Hopf algebras [23] and Hopf algebras over commutative rings [28] , since the following identity may be established along the lines of [29] for the modular functions of subalgebra pairs of augmented Frobenius algebras B c A.

Lemma 2.4. Let (A, s) be an augmented Frobenius algebra with Nakayama automorphism nA, B a subalgebra and Frobenius algebra where nA(B) = B, and AB finitely generated projective. It follows that A D B is a ¡-Frobenius extension where ¡) = nB 0 nA, a relative Nakayama automorphism [22, Satz 7], [29, Paragraph 5.1]. Then the modular automorphisms of A and B satisfy

mA\B = mB o ß.

Proof. Let (fy, xi, yi, nA) be a Frobenius coordinate system for A, tA € A a right norm satisfying fytA = e, then tA is a right integral, satisfying tAx = tAe(x) for all x € A, spanning the one-dimensional space of integrals in A. Let mA be the augmentation on A defined by xtA = mA(x)tA for x € A. It follows that e = mA o nA by expressing tA in terms of dual bases, e and mA = tAfy (and note that (fy,yi,nA(xi)) are also dual bases) [29, Paragraph 3.2]. Similarly let (f, uj, vj,nB) be a Frobenius coordinate system for B and tB a right norm satisfying ftB = e|B, then tB is a right integral in B and xtB = mB (x) tB defines the fc-valued algebra homomorphism mB, which satisfies e|B = mB onB. It follows that mB op = mB onB onA- = eonA_1|B = mA|B. □

Note that (2.4) for Hopf subalgebras also follows from (2.l). Corollary 2.3 does not extend to depth 3 Hopf subalgebras by the next example.

Example 2.5. The Taft-Hopf algebra H over its cyclic group subalgebra K is a nontrivial p-Frobenius extension [23]. The algebra H is generated over C by a grouplike g of order n > 2, a nilpotent x of index n, and (g, l)-primitive element where xg = fgx for f € C a primitive nth root of unity. This is a Hopf algebra having right integral tH = xn-1 ^n-o gj with modular function mH(g) = f [23]. The Hopf subalgebra K is generated by g. Then the twist automorphism of K is given by p(gj) = fjgj. Of course, mH restricted to K is not equal to mK = e|K. The depth d(K,H) = 3 is computed in [30].

Finally we note that unimodular Hopf algebra extensions are trivial if the H-depth condition dH(B,A) = l is imposed.

Proposition 2.6. Suppose H is a finite-dimensional Hopf algebra and K is a Hopf subalgebra of H. If dH(K,H) = l, then K satisfies a double centralizer result; in particular, if H is unimodular, then K = H.

Proof. Since H is a finite-dimensional Hopf algebra, it is a free extension of the Hopf subalgebra K, therefore faithfully flat. If dH(K,H) = l, then the ring extension satisfies the generalized Azumaya condition H®KH = Hom Z(H (Ch(K), H) via x®Ky — Xx opy, left and right multiplication [23, 3l], where CH(K) denotes the centralizer subalgebra of K in H. If d € CH(Ch(K)), then it is obvious from this that d®KlH = lH®Kd, so that d € K: it follows that

K = Ch (Ch (K)). (2.5)

Since H is unimodular, it has a two-sided nonzero integral t. Note that t € Z(H) c CH(Ch(K)), whence t € K. Let X : H — k (where k is the arbitrary ground field) be the left integral in the dual Hopf algebra H* such that X ^ t = e. The bijective antipode S : H — H satisfies S(a) = 2(t) t(l)X(at(2)) since ^(a) a(l)S(a(2)) = lHe(a) and X ^ x = X(x)lH for all x,a € H. Since A(t) = 2(t) t(l) ® t(2) € K ® K, it follows that S(a) € K for all a € H. Thus H = K. □

3. Even Depth of QF Extensions

It is well known that for a Frobenius extension A D B, coinduction of a module, MB — Hom (AB,MB) is naturally isomorphic as functors to induction (MB — M®BA) (from the category of B-modules into the category of A-modules). Similarly, a QF extension has

h-equivalent coinduction and induction functors, which is seen from the naturality of the mappings in the next proof. Let T be an arbitrary third ring.

Proposition 3.1. Suppose T MB is a bimodule and A D B is a QF extension. Then, there is an h-equivalence ofbimodules,

tM0bAa ~t Hom(AB, Mb)a. (31)

Proof. Since AB is f.g. projective, it follows that there is a T-A-bimodule isomorphism

M0bHom(AB, Bb) ~ Hom(AB, Mb), (3.2)

given by m0B$ — m^(-) with inverse constructed from projective bases for AB. But the right B-dual of A is h-equivalent to BAA, so (3.1) holds by Lemma 1.1. □

The next theorem shows that minimum right and left even depth of a QF extension are equal (see Definition 1.6 where as before Cn(A, B) = A0B ■ ■ ■ 0BA, n times A).

Theorem 3.2. If A d B is QF extension, then A D B has left depth 2n if and only if A d B has right depth 2n.

Proof. The left depth 2n condition on A D B recall is Cn+1 (A, B) ~ Cn(A, B) as B-A-bimodules. To this apply the additive functor Hom(-A, AA) (into the category of A-B-bimodules), noting

thatHom(Cn(A,B)A,AA) = Hom(Cn-1(A,B)B,AB) via f — f (-0B----0B1A) for each integer

n > 1. It follows (from Lemma 1.1) that there is an A-B-bimodule h-equivalence,

Hom(Cn(A,B)b,Ab) ~ Hom(Cn-1(A,B)B,AB). (3.3)

(Then in the depth two case, the left depth two condition is equivalent to End AB ~ A as natural A-B-bimodules.)

Given bimodule AMB, we have AM0BAA ~a Hom(AB, MB )A by the previous lemma: apply this to Cn+1 (A, B) = Cn(A,B)0BA using the hom-tensor adjoint relation: there are h-equivalences and isomorphisms of A-bimodules,

Cn+1(A,B) ~ Hom(AB,Cn(A,B)B)

~ Hom(Ab, Hom (AB,Cn-1(A, B)b)b) (

= Hom(A0BAB,Cn-1(A,B)B) ••• ~ Hom(Cp(A,B)b,Cn-p+1 (A,B)b),

for each p = 1,2,...,n and n = 1,2,.... Compare (3.3) and (3.4) with p = n to get

ACn+1(A, B)B ~a Cn(A, B)B, which is the right depth 2n condition.

The converse is proven similarly from the symmetric conditions of the QF hypothesis.

The extent to which the theorem (and most of the results in the next section) extends to ¡-Frobenius or even twisted QF extensions presents technical problems and is unknown to the author.

4. Frobenius Extensions

As noted above a Frobenius extension A D B is characterized by any of the following four conditions [23]. First, AB is finite projective and BAA = Hom(AB,BB). Second, BA is finite projective and AAB ~ Hom(BA,BB). Third, coinduction and induction of right (or left) B-modules is naturally equivalent. Fourth, there is a Frobenius coordinate system (E : A — B; x1,..., xm ,y1, ...,ym € A), which satisfies

E € Hom(BAB,BBB), ^E(axi)yi = a = ^xiE(yia) (Va € A). (4.1)

i=1 i=1

These (dual bases) equations may be used to show the useful fact that ^i xi ® yi € (A®BA)A.

We continue this notation in the next lemma. Although most Frobenius extensions in the literature are generator extensions, by the lemma equivalent to having a surjective Frobenius homomorphism, Example 2.7 in [23] provides a somewhat pathological example of a matrix algebra Frobenius extension with a nonsurjective Frobenius homomorphism.

Lemma 4.1. The natural module AB is a generator A is a generator ^ there are elements {aj }n

j }j=i

and {cj j=1 such that £n=1 E(ajCj) = 1B ^ E is surjective.

Proof. The bimodule isomorphism bAa—BHom(AB,BB)A is realized by a — E(a-) (with inverse ty — X$(xùyù- If Ab is a generator, then there are elements {cj }"=1 of A and mappings {tyj }"=1 of A* such that Y!j=1 tyj(cj) = 1B• Let Eaj = tyj. Then, E(ajCj) = 1B.

Another bimodule isomorphism AAb—AHom(BA,BB)B is realized by a — E(-a) := aE. Then writing the last equation as ^j CjE(aj) = 1B exhibits bA as a generator.

The last of the equivalent conditions is implied by the previous condition and implies the first condition. Also note that any other Frobenius homomorphism is given by Ed for some invertible d e Ab . □

A Frobenius (or QF) extension A 2 B enjoys an endomorphism ring theorem [21, 32], which shows that E := End Ab 2 A is a Frobenius (resp., QF) extension, where the default ring homomorphism A ^ E is understood to be the left multiplication mapping X : a ^ Xa where Xa(x) = ax. It is worth noting that X is a left split A-monomorphism (by evaluation at 1A) so AE is a generator.

The tower of a Frobenius (resp., QF) extension is obtained by iteration of the endomorphism ring and X, obtaining a tower of Frobenius (resp. QF) extensions where occasionally we need the notation B := E-1r A = Eo and E = E1

B A E1 c—> E2 ^ ••• ^En^--- (4.2)

so E2 = End EA, and so forth. By transitivity of Frobenius extension or QF extension [21, 22], all subextensions Em — Em+n in the tower are also Frobenius (resp. QF) extensions.

The rings ¿n are h-equivalent to Cn+l(A,B) = A0B ■■■0BA as A-bimodules in case A D B is a QF extension. This follows from noting the

End AB ~ A®BHom(AB, BB) ~ A®BA (4.3)

also holding as natural ¿-A-bimodules, obtained by substitution of A* ~ A. This observation is then iterated followed by cancellations of the type A0AM = M.

4.1. Tower above Frobenius Extension

Specialize now to A D B a Frobenius extension with Frobenius coordinate system E and {xi}m=l, {yi}m=l. Then the h-equivalences above are replaced by isomorphisms, and ¿n ~ Cn+l(A,B) for each n > -l as ring isomorphisms with respect to a certain induced "E-multiplication." The E-multiplication on A0BA is induced from the endomorphism ring EndAB — A0BA given by f — ^i f (xi)0Byi with inverse a 0 a' — Xa o E o Xa. The outcome of E-multiplication on C2(A, B) is given by

(al®Ba2)(a3®Ba4) = al E(a2a3)0Ba4 (4.4)

with unity element ll = 2m xi0Byi. Note that the A-bimodule structure on ¿l induced by X : A — E corresponds to the natural A-bimodule A®BA. The E-multiplication is defined inductively on

¿n = ¿n-l®E„-2 ¿n-l (4.5)

using the Frobenius homomorphism En-l : ¿n-l — ¿n-2 obtained by iterating the following natural Frobenius coordinate system on ¿l ~ A®BA, given by El(a®Ba') = aa' and {xi®BlA}m=i, {lA®Byi}f=i [23] as one checks.

The iterative E-multiplication on Cn(A,B) clearly exists as an associative algebra, but it seems worthwhile (and not available in the literature) to compute it explicitly. The multiplication on C2n(A, B) is given by (® = ®B, n > l)

(al 0 ••• 0 a2n)(cl 0 ••• 0 c2n)

= ai 0 ••• 0 anE(an+iE(- ■ ■ E(a2n-iE(a2nCi)C2) ■ ■ ■ )cn-i)cn) 0 Cn+i 0 ••• 0 C2n. The identity on C2n(A, B) is in terms of the dual bases,

l2n-1 = X Xii ® Vin ®---®Vii.

¿1 = 1

The multiplication on C2n+l(A, B) is given by

(ai 0 ••• 0 a2n+i)(ci 0 ••• 0 C2n+i)

= ai 0 ••• 0 an+iE(an+2E(- ■ ■ E(a2nE(a2n+iCi)c2) ■ ■ ■ )Cn)Cn+i 0 ••• 0 C2n+i

with identity

l2n = X Xii 0 "' 0 Xin 0 1a 0 Уп 0 ••• 0 yii ■ (4-9)

Denote in brief notation the rings Cn(A,B) := An and distinguish them from the isomorphic rings ¿n-l (n = 0, l,...).

The inclusions An — An+l are given by a[n] — a[n] ln, which works out in the odd and even cases to

A2n-l — A2n,

ai 0 ••• 0 a2n-1 1—> y ai 0 ••• 0 anxi 0 yi 0 an+i 0 ••• 0 a2n-i,

i (4.10)

A2n — A2n+l,

ai 0 ••• 0 a2n 1—> ai 0 ••• 0 an 0 1a 0 an+i 0 ••• 0 a2n.

The bimodule structure on An over a subalgebra Am (with m < n via composition of left multiplication mappings X) is just given in terms of the multiplication in Am as follows:

(ri 0 ••• 0 rm)(ai 0 ••• 0 an) (4ll)

= [(rl 0 ••• 0 rm)(al 0 ••• 0 am)] 0 am+l 0 ••• 0 an

with a similar formula for the right module structure -

The formulas for the successive Frobenius homomorphisms E m • Am+1 ^ Am are given in even degrees by

E2n(ai 0 ••• 0 a2n+i) = ai 0 ••• 0 anE(an+i) 0 an+2 0 ••• 0 a2n+i (4 - 12)

for n > 0 - The formula in the odd case is

E2n+i(ai 0 ••• 0 a2n+2) = ai 0 ••• 0 an 0 an+ian+2 0 an+3 0 ••• 0 a2n+2 (4 - i3)

for n > 0

The dual bases of En denoted by xn and y^ are given by all-in-one formulas

Xn = Xi 0 in-i,

(4- i4)

yn = in-i 0 yi

for n > 0 (where i0 = iA)- Note that £i xn0Anyn = in+i-

With another choice of Frobenius coordinate system (F,Zj,Wj) for A D B, there is in fact an invertible element d in the centralizer subring AB of A such that F = E(d-) and '^ixi®Byi = XjZj®BdrlWj [22, 23]; whence an isomorphism of the E-multiplication onto the F-multiplication, both on A®BA, is given by r1 ® r2 — r1 ® d-1r2. If the tower with E-multiplication is denoted by AnE and the tower with F-multiplication is denoted by AnF, there is a sequence of ring isomorphisms

ae jl, Af A2n * A2n,

(4.15)

a1 ® ••• ® a2n i—> a1 ® ••• ® an ® d_1an+1 ® ••• ® d_1a2n,

aE = ■. Af

Aln+1 An+1 (4.16)

ai 0 ••• 0 a2n+1 -—> a1 0 ••• 0 an+1 0 d_1an+2 0 ••• 0 d_1a2n+1

which commute with the inclusions Aef — A^.

Theorem 4.2. The multiplication, module, and Frobenius structures for the tower An = A®b ■ ■ ■ ®B A (n times A) above a Frobenius extension A 2 B are given by formulas (4.4) to (4.16).

Proof. First define Temperley-Lieb generators iteratively by en = 1n-1 ®An-21n-1 e An+1 for n = 1,2,..., which results in the explicit formulas

e2n = X Xi1 0 0 Xin 0 Vinxin+1 0 Vin+1 0 yin-1 0 0 y«1 /

i1/.../in+1

e2n+1 = X Xi1 0 ••• 0 Xin 0 1a 0 1A 0 yin 0---0yix.

(4.17)

i1/.../in

These satisfy braid-like relations [4, page 106], namely,

eiej = ejeir \i - j| > 2, ei+1eiei+1 = ei+1, eiei+1ei = ei1i+1. (4.18)

(The generators above fail to be idempotents to the extent that E(1) differs from 1.) The proof that the formulas above are the correct outcomes of the inductive definitions may be given in terms of Temperley-Lieb generators, braid-like relations and important relations

enxen = enEn-1(x), Vx € An, yen = En yen en, Vy € An+1, (4.19)

xen = enx, Vx € An-1.

Reference [4, page 106] (for background see [33]) as well as the symmetric left-right relations. These relations and the Frobenius equations (4.1) may be checked to hold in terms of the equations above in a series of exercises left to the reader.

The formulas for the Frobenius bases follow from the iteratively apparent xin = xie1e2 ••• en and yn = en ••• e2e1yi and uniqueness of bases with respect to the same Frobenius

homomorphism. In fact en •••e2e\a = ln_i 0 a for any a e A,n = 1,2,... (a symmetrical formula holds as well) and 1n = ^i xie1 ■ ■ ■ en-1enen-1 ■ ■ ■ e1yi.

Since the inductive definitions of the ring and module structures on the An's also satisfy the relations listed above and agree on and below A2, the proof is finished with an induction argument based on expressing tensors as words in Temperley-Lieb generators and elements of A.

We note that

The formulas for multiplication (4.8), (4.6), and (4.ll) follow from induction and applying

For the next proposition the main point is that given a Frobenius extension there is a ring structure on the Cn(A,B)'s satisfying the hypotheses below (for one compares with (4.ll)). This is true as well if A is a ring with B in its center, since the ordinary tensor algebra on A0BA may be extended to an n-fold tensor product algebra A0B ■ ■ ■ 0B A.

Proposition 4.3. Let A d B be a ring extension. Suppose that there is a ring structure on each An := Cn(A,B) for each n > 0, a ring homomorphism An-l — An for each n > l, and that the composite B — An induces the natural bimodule given by b ■ (al 0 ••• 0 an) ■ b' = bal 0 a2 0 ■ ■ ■ 0 anb'.

Then, A D B has depth 2n + l if and only if An | B has depth 3.

Proof. If A D B has depth 2n + l, then An ~ An+l as B-bimodules. By tensoring repeatedly by

BA0B-, also An ~ A2n as B-bimodules. But A2n ~ An0BAn. Then, An D B has depth three.

Conversely, if An | B has depth 3, then A2n ~ An as B-bimodules. But An+l | A2n via the split B-bimodule epi al 0 ■ ■ ■ 0 a2n — al ■ ■ ■ an 0 an+l 0 ■ ■ ■ 0 a2n. Then, An+l | qAn for some q e Z+. It follows that A D B has depth 2n + l. □

One may in turn embed a depth three extension into a ring extension having depth two. The proof requires the QF condition. Retain the notation for the endomorphism ring introduced earlier in this section.

Theorem 4.4. Suppose A D B is a QF extension. If A D B has depth 3, then ¿ D B has depth 2. Conversely, if ¿D B has depth 2 and AB is a generator, then A d B has depth 3.

Proof. Since A is a QF extension of B, we have ¿ ~ A0BA as ¿-A-bimodules. Then, ¿0B¿ ~

A0BA0BA0BA as ¿-B-bimodules. Given the depth 3 condition, A0BA ~ A as B-bimodules,

hh it follows by two substitutions that ¿0B¿ ~ A0BA as ¿-B-bimodules. Consequently, ¿0B¿ ~

¿ as ¿-B-bimodules. Hence, ¿ D B has right depth 2, and since it is a QF extension by the

endomorphism ring theorem and transitivity, ¿D B also has left depth 2.

Conversely, we are given AB a progenerator, so that ¿ and B are Morita equivalent

rings, where BHom(AB,BB)£ and ¿AB are the context bimodules. If ¿ D B has depth

hh two, then ¿0B¿ ~ ¿ as ¿-B-bimodules. Then A0BA0BA0BA ~ A0BA as ¿-B-bimodules.

ai 0 ■ ■ ■ 0 a„+i = (ai 0 ■ ■ ■ 0 a„)(1„-i 0 a„+i)

= (ai 0 ■ ■ ■ 0 a„-i)(1„-2 0 a„)(e„■ ■ ■ eia„+i) = ■ ■ ■ = ai(eia2)(e2eia3) ■ ■ ■ (e„-i ■ ■ ■ eia„)(e„ ■ ■ ■ eia„+i).

(4.20)

the relations (4.18) through (4.20).

Since Hom(AB,BB)®EA = B as B-bimodules, a cancellation of the bimodules EAB follows,

so A®BA®BA ~ A as B-bimodules. Since A®BA | A®BA®BA, it follows that A®BA | qA for some q € Z +. Then A D B has depth 3. □

Example 4.5. To illustrate that the theorem does not extend to when A D B is not a QF extension, consider A = Tn(k), n > 2 (a hereditary algebra) and B = Dn(k) (a semisimple algebra), and left k be an algebraically closed field of characteristic zero. (Since B, A is, is not a QF-algebra it follows by transitivity that A D B is not a QF extension.) It was computed that d(B,A) = 3 in Example 1.7. Thinking of the columns of A as Aeii, it is quite easy to see that End AB ~ M1(k) x M2(k) x ••• x Mn(k) and that the inclusion of A — EndAB is given by

X (xn/ (X11 X12).....x). (4.21)

^X12 X22 Its restriction to B is given by

Diag(^1,...,^„) 1—> (^1,Diag(^1,^2),...,Diag(^1,...,^„)) (4.22)

with inclusion matrix M = ^< ej. Then, MM1 > 0, and from (1.1) we see that d(B, £) = 3.

5. When Tower Depth Equals Subring Depth

In this section we review tower depth from [11] and find a general case when it is the same as subring depth defined in (1.7) and in [12]. We first require a generalization of left and right depth 2 to a tower of three rings. We say that a tower A D B D T, where A D B and B D T are ring extensions, has generalized right depth 2 if A®BA hA as natural A-T-bimodules. (Note that if T = B, this is the definition of the ring extension A D B having right depth 2.)

Throughout the section below we suppose A D B is a Frobenius extension and Ei — Ei+1 is its tower above it, as defined in (4.2) and the ensuing discussion in Section 4. Following [11] (with a small change in vocabulary), we say that A D B has right tower depth n > 2 if the subtower of composite ring extensions B — £n-3 — ¿n_2 has generalized right depth 2; equivalently, as natural ¿n_2-B-bimodules,

¿n-2®£„-3En-2 ®* = qEn-2 (5.1)

for some positive integer q, since the reverse condition is always satisfied. Since ¿_1 = B and E0 = A, this recovers the right depth two condition on a subring B of A. To this definition we add that a Frobenius extension A D B has tower depth 1 if it is a centrally projective ring extension; that is, BAB | qB for some q € Z+. Left tower depth n is just defined using (5.1) but as natural B-£n_2-bimodules. By [11, Theorem 2.7] the left and right tower depth n conditions are equivalent on Frobenius extensions.

From the definition of tower depth and a comparison of (4.5) and Definition 1.6 we note that if A is a Frobenius extension of B of tower depth n > 1, then B c A has subring depth 2n _ 2; from (5.1) we obtain An | qAn-1 as A-B-bimodules, since An = En-1 = ¿n_2®E„_3¿n_2.

From [11, Lemma 8.3], it follows that if A D B has tower depth n, it has tower depth n + 1. Define dF(A, B) to be the minimum tower depth if A D B has tower depth n for some

integer n, dF(A,B) = to if the condition (5.l) is not satisfied for any n > 2 nor is it depth l. Notice that dp(A,B) = d(B, A) if d(B, A) < 2 or dp(A,B) < 2. This is extended to dp(A,B) = d(B, A) if d(B, A)ordF(A, B) < 3 in the next lemma.

Notice that tower depth n makes sense for a QF extension A D B: by elementary

considerations, it has right tower depth 3 if B — A <—¿ satisfies ¿0^ ~¿ as ¿-B-bimodules. It has been noted elsewhere that a QF extension has right tower depth 3 if and only if it has left tower depth 3 by an argument essentially identical to that in [ll, Theorem 2.8] but replacing Frobenius isomorphisms with quasi-Frobenius h-equivalences.

Lemma 5.1. A QF extension A D B such that AB is a generator has tower depth 3 if and only if B has depth 3 as a subring in A.

Proof. By the QF property, ¿ ~ A0BA as ¿-B-bimodules. By the tower depth 3 condition,

¿0^ ~ ¿ as ¿-B-bimodules. Then, A0BA0BA ~ A0BA as ¿-B-bimodules. Since AB is a

progenerator, we cancel bimodules ¿AB as in the proof of Theorem 4.4 to obtain A0BA ~ A as B-bimodules. Hence, B c A has depth 3.

Given BAB ~B A0BAB, by tensoring with ¿A0B- we get A0BA ~ A0BA0BA as

¿-B-bimodules. By the QF property, ¿0^ ~¿ as ¿-B-bimodules follows, whence A D B has tower depth 3. □

The theorem below proves that subring depth and tower depth coincide on Frobenius generator extensions, which are the most common Frobenius extensions, for example, including all group algebra extensions: the endomorphism ring extension of any Frobenius extension is a Frobenius generator extension. At a certain point in the proof, we use the following fundamental fact about the tower An above a Frobenius extension A D B: since the compositions of the Frobenius extensions remain Frobenius, the iterative construction of E-multiplication on tensor-squares isomorphic to endomorphism rings applies but gives isomorphic ring structures to those on the An. For example, the composite extension B — An is Frobenius with End(An)B ~ An0BAn = A2n, isomorphic in its EoEl o ■ ■ ■ oEn-l-multiplication or its E-multiplication given in (4.6) [10].

Theorem 5.2. Suppose A is a Frobenius extension of B and AB is a generator. Then, A d B has tower depth m for m = l, 2,... if and only if the subring B c A has depth m. Consequently, dF (A, B) = d(B, A).

Proof. The cases m = l, 2,3 have been dealt with above. We divide the rest of the proof into odd m and even m. The proof for odd m = 2n + l: if A D B has tower depth 2n + l, then A2n0Aln-l A2n | qA2n as A2n-B-bimodules. Continuing with A2n = A2n-l0A2n-2A2n-l, iterating and performing standard cancellations, we obtain

A2n+i | qA2n (5.2)

as End(An)B-B-bimodules. But the module (An)B is a generator for all n by Lemma 4.l, the endomorphism ring theorem for Frobenius generator extensions and transitivity of generator property for modules (if MA and AB are generators, then restricted module MB is clearly a generator). It follows that (An)B is a progenerator and cancellable as an End(An)B-B-bimodule (applying the Morita theorem as in the proof of Theorem 4.4).

Then, B(An+1)B|B(An)B after cancellation of An from (5.2), which is the depth 2n +1 condition in (1.7).

Suppose An+1 e* ~ An as B-bimodules. Apply to this the additive functor An®B_ from category of B-bimodules into the category of End(An)B-B-bimodules. We obtain (5.2), which is equivalent to the tower depth 2n + 1 condition of A D B.

The proof in the even case, m = 2n, does not need the generator condition (since even nongenerator Frobenius extensions have endomorphism ring extensions that are generators).

Given the tower depth 2n condition A2n_1<s>Aln_2A2n-1 ~ A2n is isomorphic as A2n-1-B-bimodules to a direct summand in qA2n-1 for some positive integer q, introduce a cancellable extra term in A2n = An®AAn+1 and in A2n-1 ~ An®AAn. Now note that A2n-1 ~ End(An)A, which is Morita equivalent to A. After cancellation of the End(An)A-A-bimodule An, we obtain An+1 | An as A-B-bimodules as required by (1.7).

(<H Given a (An+1) b A(An)B, we apply End(A„)AAn®A_ obtaining A2n | A2n_1 as A2n_1-B-bimodules, which is equivalent to the tower depth 2n condition. □

A depth 2 extension A D B may have easier equivalent conditions, for example, a normality condition, to fulfill than the B-A-bimodule condition A®BA | qA [2]. Thus, the next corollary (or one like it stated more generally for Frobenius extensions) presents a simplification in determining whether a special type of ring extension has finite depth. The corollary follows from the theorem above as well as [11, 8.6], Corollary 2.2, Proposition 4.3 and Theorem 4.4.

Corollary 5.3. Let K c H be a Hopf subalgebra pair of finite-dimensional unimodular Hopf algebras. Then, K has finite depth in H if and only if there is a tower algebra Hm such that K c Hm has depth 2.

Acknowledgments

The author thanks the referee for thoughtful comments. Research in this paper was funded by the European Regional Development Fund through the programme COMPETE and by the Portuguese Government through the FCT under the project PE-C/MAT/UI0144/2011.nts.

References

[1] L. Kadison and B. Kulshammer, "Depth two, normality, and a trace ideal condition for Frobenius extensions," Communications in Algebra, vol. 34, no. 9, pp. 3103-3122, 2006.

[2] R. Boltje and B. Kulshammer, "On the depth 2 condition for group algebra and Hopf algebra extensions," Journal of Algebra, vol. 323, no. 6, pp. 1783-1796, 2010.

[3] S. Burciu, L. Kadison, and B. Kulshammer, "On subgroup depth," International Electronic Journal of Algebra, vol. 9, pp. 133-166, 2011.

[4] L. Kadison and K. Szlachanyi, "Bialgebroid actions on depth two extensions and duality," Advances in Mathematics, vol. 179, no. 1, pp. 75-121, 2003.

[5] L. Kadison, "Infinite index subalgebras of depth two," Proceedings of the American Mathematical Society, vol. 136, no. 5, pp. 1523-1532, 2008.

[6] W. Szymaiiski, "Finite index subfactors and Hopf algebra crossed products," Proceedings of the American Mathematical Society, vol. 120, no. 2, pp. 519-528,1994.

[7] D. Nikshych and L. Vainerman, "A Characterization of depth 2 subfactors of II1 factors," Journal of Functional Analysis, vol. 171, no. 2, pp. 278-307, 2000.

[8] J.-H. Lu, "Hopf algebroids and quantum groupoids," International Journal of Mathematics, vol. 7, no. 1, pp. 47-70,1996.

T. Brzezinski and R. Wisbauer, Corings and Comodules, vol. 309, Cambridge University Press, Cambridge, UK, 2003.

L. Kadison and D. Nikshych, "Weak Hopf algebras and Frobenius extensions," Journal of Algebra, vol. 163, pp. 258-286, 2001.

L. Kadison, "Finite depth and Jacobson-Bourbaki correspondence," Journal of Pure and Applied Algebra, vol. 212, no. 7, pp. 1822-1839, 2008.

R. Boltje, S. Danz, and B. Kulshammer, "On the depth of subgroups and group algebra extensions," Journal of Algebra, vol. 335, pp. 258-281, 20ll.

K. Hirata, "Some types of separable extensions of rings," Nagoya Mathematical Journal, vol. 33, pp. 107-115,1968.

L. Kadison, "Odd H-depth and H-separable extensions," Central European Journal of Mathematics, vol. 10, no. 3, pp. 958-968, 2012.

R. Boltje and B. Kulshammer, "Group algebra extensions of depth one," Algebra & Number Theory, vol. 5, no. l, pp. 63-73, 20ll.

D. J. Benson, Polynomial Invariants of Finite Groups, vol. 190 of London Mathematical Society Lecture Note Series, Cambridge University Press, Cambridge, UK, 1993.

R. P. Stanley, "Invariants of finite groups and their applications to combinatorics," Bulletin of the American Mathematical Society, vol. l, no. 3, pp. 475-511,1979.

S. Burciu and L. Kadison, Subgroups of Depth Three, vol. 15 of Surveys in Differential Geometry, 20ll. W. Fulton, Young Tableaux, vol. 35 of London Mathematical Society Student Texts, Cambridge University Press, Cambridge, UK, 1997.

Z. Leszczyiiski, "On the representation type of tensor product algebras," Fundamenta Mathematicae, vol. 144, no. 2, pp. 143-161,1994.

B. Muller, "Quasi-Frobenius-Erweiterungen," Mathematische Zeitschrift, vol. 85, pp. 345-368,1964.

B. Pareigis, "Einige Bemerkungen uber Frobenius-Erweiterungen," Mathematische Annalen, vol. 153, pp. l-l3,1964.

L. Kadison, New Examples of Frobenius Extensions, vol. 14 of University Lecture Series, American Mathematical Society, Providence, RI, USA, 1999.

R. Farnsteiner, "On Frobenius extensions defined by Hopf algebras," Journal of Algebra, vol. 166, no. l, pp. 130-141,1994.

D. Fischman, S. Montgomery, and H.-J. Schneider, "Frobenius extensions of subalgebras of Hopf algebras," Transactions of the American Mathematical Society, vol. 349, no. 12, pp. 4857-4895,1997. M. C. Iovanov and L. Kadison, "When weak Hopf algebras are Frobenius," Proceedings of the American Mathematical Society, vol. 138, no. 3, pp. 837-845, 2010.

H. F. Kreimer and M. Takeuchi, "Hopf algebras and Galois extensions of an algebra," Indiana University Mathematics Journal, vol. 30, no. 5, pp. 675-692, l98l.

L. Kadison and A. A. Stolin, "An approach to Hopf algebras via Frobenius coordinates. II," Journal of Pure and Applied Algebra, vol. 176, no. 2-3, pp. 127-152, 2002.

L. Kadison and A. A. Stolin, "An approach to Hopf algebras via Frobenius coordinates," Beitrage zur Algebra und Geometrie, vol. 42, no. 2, pp. 359-384, 2001.

C. J. Young, Depth of Hopf algebras in smash products, dissertation [Ph.D. thesis], University of Porto.

P. A. A. B. Carvalho, "On the Azumaya locus of some crossed products," Communications in Algebra, vol. 33, no. l, pp. 51-72, 2005.

K. Morita, "The endomorphism ring theorem for Frobenius extensions," Mathematische Zeitschrift, vol. 102, pp. 385-404,1967.

F. M. Goodman, P. de la Harpe, and V. F. R. Jones, Coxeter Graphs and Towers of Algebras, vol. 14 of Mathematical Sciences Research Institute Publications, Springer, Heidelberg, Germany, 1989.

Copyright of International Journal of Mathematics & Mathematical Sciences is the property of Hindawi Publishing Corporation and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.