1 Introduction

Recall that an algebra \({\mathfrak {A}}\) with unity over a commutative ring R is called separable if \({\mathfrak {A}}\) is a projective left \({\mathfrak {A}}\otimes _R{\mathfrak {A}}^o\)-module under the action given by \((\sum _i x_i\otimes y_i)a=\sum _ix_iay_i\) for all \(a,x_i\in {\mathfrak {A}}, y_i\in {\mathfrak {A}}^o\), where \({\mathfrak {A}}^o\) is the opposite algebra of \({\mathfrak {A}}; {\mathfrak {A}}\) is called an Azumaya algebra if it is separable over its center \(Z({\mathfrak {A}})\). For separable algebra, the reader can be referred to the references [5, 16]. If R is a commutative ring with unity \(1_R\) and G a finite group of order n, then the group ring R[G] is separable if and only if \(n1_R\) is invertible in R. This is a well-known generalization of Maschke’s theorem. It is worthy to point out that if \(n1_R\) is invertible in R, then R[G] is an Azumaya algebra. So, R[G] is separable if and only if R[G] is Azumaya. On the other hand, Azumaya algebras form a class of very well-behaved PI-algebras that allow development of the Brauer group classification of commutative rings (see [6, 17]). They also prove to be very useful as an efficient tool in the study of general algebras satisfying polynomial identities. We refer to [2] for an important application of this type.

Z-separability of the integral semigroup ring Z[S] of an arbitrary finite semigroup S was first studied by Shapiro in [19]. Then Cheng [3] obtained a description of R-separable ring R[S] for an arbitrary commutative coefficient ring R. DeMeyer and Hardy independently considered this problem in [7]. An extension of these results to the class of so-called excellent extensions was given by Okniński in [12]. In [13], Okniński and Van Oystaeyen gave some necessary and sufficient conditions for a cancellative monoid algebra to be an Azumaya algebra. In a sequence of papers, Van Oystaeyen studied group-graded Azumaya algebras (see, [14, 15]).

A semigroup S is called idempotent-free if S has no idempotents except possibly identity and zero elements; S is called almost idempotent-free if it has exactly one nonzero \({\mathcal {J}}\)-class containing idempotents. Okniński pointed out that for a finitely generated semigroup S and a field KK[S] is an Azumaya algebra if and only if \(K[S]\cong \oplus _{i=1}^nK_0[S_i]\) and each \(K_0[S_i]\) is an Azumaya algebra, where \(S_i\) are almost idempotent-free semigroups (see [11, Corollary 17,  p. 325]). So, it is important to probe when semigroup algebras of almost idempotent-free semigroups are Azumaya algebras. Because of this view, Okniński raised the problem ([11, Problem 28, p. 332]):

Problem 1.1

Characterize Azumaya algebras K[S] of almost idempotent-free semigroups.

He conjectured that the equivalences of [11, Proposition 8,  p. 318] can be extended to idempotent-free semigroups; that is,

Conjecture 1.2

Let S be an idempotent-free monoid with the group of units U. Then the following statements are equivalent:

  1. (1)

    K[S] is an Azumaya algebra;

  2. (2)

    \(K[S]=Z(K[S])K[U]\) and K[U] is an Azumaya algebra;

  3. (3)

    \(C_{K[S]}(U)\) (the centralizer of U) \(=Z(K[S])\) and K[U] is an Azumaya algebra.

Furthermore, in his monograph, he raised [11, Problem 29, p. 332]:

Problem 1.3

Assume that S is a finite semigroup. Is an Azumaya algebra K[S] always isomorphic to a finite direct product of matrix algebras over semigroup algebras of idempotent-free monoids?

Our main aim is to give some partial answer to these problems.

We proceed as follows: after citing known results, we prove that for an almost idempotent-free semigroup S satisfying the regularity condition, if K[S] is Azumaya, then \(K[S]\cong M_n(K[G])\) where G is an idempotent-free monoid (Theorem 3.3). Based on this fact, we give a positive answer of Problem 1.3 when S satisfies the regularity condition (Corollary 3.6). In Sect. 4, we consider semigroup algebras of almost idempotent-free semigroups. We obtain a necessary and sufficient condition for the semigroup K[S] of an almost idempotent-free semigroup S satisfying the regularity condition and \(min_J\) to be Azumaya (Theorems 4.4 and 4.6). It is proved that Conjecture 1.2 is true for the case when the idempotent-free semigroup satisfying \(min_J\) (Theorem 4.10).

2 Preliminaries

Throughout this note, we use notations and terminologies from the monograph of Okniński [11] and the textbook of Clifford and Preston [4]. Other undefined terms can be found in the textbook [9].

2.1 Semigroups

Let S be a semigroup. Assume that I is an ideal of S. Form the set \(S/I=(S\backslash I)\cup \{0\}\) and define a multiplication \(\circ \) on S / I by the rule that

$$\begin{aligned} a\circ b=\left\{ \begin{array}{cl} ab &{} \quad \text { if } \,\, a,b,ab\in S\backslash I;\\ 0 &{} \quad \text { otherwise}, \end{array} \right. \end{aligned}$$

where ab is the product of a and b in the semigroup S. It is not difficult to check that \((S/I,\circ )\) is a semigroup. In what follows, we denote the above semigroup by S / I. In fact, S / I is the Rees factor semigroup of S modulo I (for detail, [4, p. 17]).

Also, S is called 0-simple if \(S^2\not =0\) and S has at most two ideals: 0 and S. In what follows, we always use \({\mathcal {J}}\) to stand for the usual Green’s relation: \((a,b)\in {\mathcal {J}}\) if and only if there exist \(x,y,u,v\in S^1\) such that \(a=xby\) and \(b=uav\). For \(a\in S\), we use J(a) to denote the smallest ideal \(S^1aS^1\) of S containing a, and \(J_a\) to denote the \({\mathcal {J}}\)-class of S containing a. We shall call the \({\mathcal {J}}\)-class containing regular elements the regular \({\mathcal {J}}\)-class of S. Define

$$\begin{aligned} J_a\le J_b \quad \text { if }\,\,\,J(a)\subseteq J(b). \end{aligned}$$

It is evident that \(\le \) is a partial order on the set \(S/\!{\mathcal {J}}\). When \(J_a\le J_b\) and \(J_a\not =J_b\), we shall denote \(J_a<J_b\). As in Howie [9], we say that S satisfies the condition \(min_J\) if the partially ordered set \(S/\!{\mathcal {J}}\) satisfies the minimal condition. Note that \(I(a)=\{x\in J(a): J_x< J_a\}\) is an ideal of S. The Rees factor semigroup \(J(a)/I(a)=J_a\cup \{0\}\) of J(a) modulo I(a) is either a 0-simple semigroup or a null semigroup. For convenience, we denote the zero element by \(0.\, J(a)/I(a)\) is called the principal factor of S determined by a. So, any principal factor is either a 0-simple semigroup or a null semigroup.

Let n be an integer. We say that S has the property \({\mathfrak {P}}_n\) if for any \(x_1,x_2,\ldots ,x_n\in S\), there exists a nontrivial permutation \(\sigma \) in the symmetric group \({\mathfrak {S}}_n\) such that \(x_1x_2\ldots x_n=x_{\sigma (1)}x_{\sigma (2)}\ldots x_{\sigma (n)}\). It is easy to see that any subsemigroup and any homomorphic image of a semigroup having the property \({\mathfrak {P}}_n\) have the property \({\mathfrak {P}}_n\). For simplicity, we say that S has the permutational property if some \({\mathfrak {P}}_n, n\ge 2\), is satisfied in S. In [8], Domanov proved the following result (also see [11, Theorem 17, p. 229]):

Lemma 2.1

If T is a 0-simple semigroup satisfying the permutational property, then T is a completely 0-simple semigroup.

2.2 Semigroup Algebras

We always assume that K is a field and S a semigroup. We denote by K[S] the semigroup algebra of S over K. In general, if I is a subset of S, then K[I] denotes the set of K-linear combinations of elements in I, that is, K[I] is a vector space with I as a basis. So each element of K[I] is a finite summation of the form \(\sum _{x\in I}r_xx, r_x\in K,x \in I\). In particular, if \(I_1\) and \(I_2\) are subsets of S, then \(K[I_1\cap I_2] =K[I_1]\cap K[I_2]\). If S is a semigroup with zero \(\theta \), then \(K[\theta ]\) is an ideal of K[S], and we define \(K_0[S]=K[S]/K[\theta ]\). This K-algebra \(K_0[S]\) is called the contracted semigroup algebra of S over K. If S has no zero, then we have \(K_0[S]=K[S]\). Clearly, an element a of \(K_0[S]\) is a finite linear combination \(a= \sum r_ss\) of elements \(s\in S\backslash \{\theta \}\). The support of \(a\in K_0[S]\), denoted by supp(a), is the set \(\{s \in S\backslash \{\theta \}\mid r_s\not =0\}\).

Lemma 2.2

[11, Proposition 1, p. 221] Assume that K[S] satisfies a polynomial identity of degree n. Then S has the property \({\mathfrak {P}}_n\).

We now recall some known results on Azumaya algebras (for example, see [11, Lemma 1, p. 313]).

Lemma 2.3

Assume that \({\mathfrak {A}}\) is an Azumaya algebra, and denote by \(Z({\mathfrak {A}})\) the center of \({\mathfrak {A}}\). Then

  1. (1)

    For every ideal \({\mathfrak {J}}\) of \({\mathfrak {A}}\) and \({\mathfrak {I}}\) of \(Z({\mathfrak {A}})\), we have \(({\mathfrak {J}}\cap Z({\mathfrak {A}})){\mathfrak {A}}={\mathfrak {J}}\) and \({\mathfrak {IA}}\cap Z({\mathfrak {A}})={\mathfrak {I}}\).

  2. (2)

    For every ideal \({\mathfrak {J}}\) of \({\mathfrak {A}}, Z({\mathfrak {A}}/{\mathfrak {J}})=(Z({\mathfrak {A}})+{\mathfrak {J}})/{\mathfrak {J}}\) and \({\mathfrak {A}}/{\mathfrak {J}}\) is an Azumaya algebra.

  3. (3)

    \({\mathfrak {A}}\) is a finitely generated projective module over its center \(Z({\mathfrak {A}})\), and \(Z({\mathfrak {A}})\) is a direct summand of this module.

In [1], Adjamagbo pointed out that

Lemma 2.4

[1, Recall 1.1 (8), p. 92] For any commutative algebra \({\mathfrak {A}}\) and commutative \({\mathfrak {A}}\)-algebra \({\mathfrak {C}}\), if \({\mathfrak {B}}\) is separable over \({\mathfrak {A}}\), then \(\mathfrak {B}\otimes _{{\mathfrak {A}}}{\mathfrak {C}}\) is separable over \({\mathfrak {C}}\). Conversely, if \(\mathfrak {B}\otimes _{{\mathfrak {A}}}{\mathfrak {C}}\) is separable over \({\mathfrak {C}}\) and if, in addition, \({\mathfrak {B}}\) is a finitely generated \({\mathfrak {A}}\)-module or \({\mathfrak {A}}\)-algebra and \({\mathfrak {C}}\) a faithfully flat \({\mathfrak {A}}\)-module, then \({\mathfrak {B}}\) is separable over \({\mathfrak {A}}\).

The following lemma follows immediately from [11, Proposition 13, p. 323].

Lemma 2.5

Let K[S] be an Azumaya algebra and \(t\in S\). If the principal factor of S determined by t is 0-simple, then K[StS] is a ring direct summand of K[S].

By the proof of [11, Proposition 13, p. 323], Lemma 2.5 is true for \(K_0[S]\).

The following observation is trivial.

Lemma 2.6

Let S be a semigroup. Then K[S] is an Azumaya algebra if and only if \(K_0[S]\) is an Azumaya algebra.

We need the following well-known result (see [18, Proposition 25B.8] or [6]).

Lemma 2.7

Let S be a monoid. Then \(M_n(K_0[S])\) is Azumaya if and only if \(K_0[S]\) is Azumaya.

3 The Regularity Condition

A semigroup is said to satisfy the regularity condition if all of its regular elements form a subsemigroup. Obviously, cancellative monoids satisfy the regularity condition. In this section, we study when semigroup algebras of almost idempotent-free semigroups satisfying the regularity are Azumaya algebras.

Lemma 3.1

Let S be an almost idempotent-free semigroup. If K[S] is an Azumaya algebra, then the following statements are true:

  1. (1)

    For any \(x\in S, x\) is not regular in S if and only if J(x) / I(x) is a null semigroup. Moreover, every element of the \({\mathcal {J}}\)-class of S containing a nonzero regular element is regular.

  2. (2)

    The nonzero regular \({\mathcal {J}}\)-class is the greatest element in the set \(S/\!{\mathcal {J}}\) with the above order \(\le \). Moreover, \(V:=\big (S\backslash \mathrm{Reg}(S)\big )\cup \{0\}\) is an ideal of S.

  3. (3)

    S / V is a completely 0-simple semigroup. Moreover, for any nonzero regular elements ab of \(S, a{\mathcal {D}}b\).

Proof

  1. (1)

    Assume that x is not regular in S. Then J(x) / I(x) is a null semigroup; if not, then J(x) / I(x) is 0-simple, but since K[S] is Azumaya, now K[S] is a PI-algebra, so by Lemmas 2.1 and 2.2, J(x) / I(x) is a completely 0-simple semigroup. So, the assertions follow.

  2. (2)

    By hypothesis, K[S] has an identity. Let e be the identity of K[S] and J a maximum element of the set \(A:=\{J_a:a\in \mathrm{supp}(e)\}\). If \(b\in J\cap \mathrm{supp}(e)\), then by \(be=b, bu=b\) for some \(u\in \mathrm{supp}(e)\). Since \(J=J_b=J_{bu}\le J_u\) and by the maximality of J, we have \(J=J_u\). That is, \(bu,u\in J\). This shows that \(J(b)/I(b)=J\cup \{0\}\) is 0-simple. Thus by Lemmas 2.1 and 2.2, J(b) / I(b) is a completely 0-simple semigroup, so b is regular in S, and therefore J is a nonzero regular \({\mathcal {J}}\)-class of S. For any \(w\in S\), since \(we=w\), there exists \(z\in \mathrm{supp}(e)\) such that \(w=wz\). It follows that \(J_w\le J_z\le J\) because J is the greatest element of A. Note that S has only a nonzero regular \({\mathcal {J}}\)-class. Therefore, the nonzero regular \({\mathcal {J}}\)-class is the greatest element in the set \(S/\!{\mathcal {J}}\). For any \(a,b\in S\), if \(ab\in \mathrm{Reg}(S)\backslash \{0\}\), then \(J_{ab}\le J_a,J_b\) and \(J_{ab}=J_a=J_b\) since the nonzero regular \({\mathcal {J}}\)-class is maximal, so \(a,b\in \mathrm{Reg}(S)\backslash \{0\}\), whence V is an ideal of S.

  3. (3)

    Because S is an almost idempotent-free semigroup and by \((2), S/V=J_a\cup \{0\}\) where a is regular. Since a is regular, we know that J(a) / I(a) is 0-simple, that is, S / V is a 0-simple semigroup. On the other hand, since K[S] is Azumaya, S / V satisfies the permutational property. Now by Lemma 2.1, S / V is a completely 0-simple semigroup. The rest is trivial. \(\square \)

Proposition 3.2

Let S be an almost idempotent-free semigroup. If K[S] is an Azumaya algebra, then S has finitely many idempotents.

Proof

By Lemma 3.1 (2), K[V] is an ideal of K[S], so that by Lemma 2.3 \((2), K_0[S/V]\cong K[S]/K[V]\) is Azumaya. It follows that \(K_0[S/V]\) has an identity. Now by [11, Proposition 2.5, p. 59], S / V has finitely many \({\mathcal {L}}\)-classes and \({\mathcal {R}}\)-classes. Therefore, as S / V is a completely 0-simple semigroup, S has finitely many idempotents. \(\square \)

Theorem 3.3

Let S be an almost idempotent-free semigroup. If S satisfies the regularity condition, then \(K_0[S]\) is an Azumaya algebra if and only if \(K_0[S]\cong M_n(K_0[G])\) where G is an idempotent-free monoid and \(K_0[G]\) is an Azumaya algebra.

Proof

Note that \(K_0[S]=K_0[S^0]\). So, we may assume that S has zero element. By Lemma 2.7, we need only to verify the necessity. Now, assume that \(K_0[S]\) is an Azumaya algebra. Then \(K_0[S]\) has an identity, say E. Let \(E=F+D\) with \(\mathrm{supp}(F)\subseteq \mathrm{Reg}(S)\) and \(\mathrm{supp}(D)\subseteq V\) where \(V=(S\backslash \mathrm{Reg}(S))\cup \{0\}\). For any \(a\in \mathrm{Reg}(S)\), we have \(a=aE=aF+aD\) and \(a=Ea=Fa+Da\). By Lemma 3.1(2), V is an ideal of S, and \(\mathrm{supp}(aD),\mathrm{supp}(Da)\subseteq V\). Thus \(a=Fa=aF\), in other words, F is an identity of \(K_0[\mathrm{Reg}(S)]\).

Compute

$$\begin{aligned} F=FE=F^2+FD=EF=F^2+DF. \end{aligned}$$

But \(DF,FD\in K[V]\) and \(F^2\in K[\mathrm{Reg}(S)]\), so \(F^2=F\) and \(FD=0=DF\). Since

$$\begin{aligned} F+D= & {} E=E^2=(F+D)^2=F^2+FD+DF+D^2\\= & {} F^2+D^2 \end{aligned}$$

and by \(D^2\in K[V]\) and \(F^2\in K[\mathrm{Reg}(S)]\), we observe that \(D=D^2\). Let \(x\in \mathrm{supp}(D)\) such that \(J_x\) is the maximum element of the set \(\big \{J_u:u\in \mathrm{supp}(D)\big \}\). The fact that \(D=D^2\) derives that \(x=yz\) for some \(y,z\in \mathrm{supp}(D)\). It follows that \(J_x\le J_y,J_z\). Thus \(J_x=J_y=J_z\) by the maximality of \(J_x\). Now, \(x=yz\) can show that J(x) / I(x) is 0-simple, and further by Lemma 3.1(1), x is a regular element, contradicting to the fact that \(\mathrm{supp}(D)\subseteq V\). Consequently, \(D=0\) and whence \(E=F\) and is the identity of \(K_0[S]\).

By Lemma 3.1 (3), S / V is isomorphic to some Rees matrix semigroup \(\mathcal {M}^0(H,I,\Lambda ;P)\) over a group H. Again by Lemma 3.1 \((2), S/V\cong \mathrm{Reg}(S)\), so that \(K_0[S/V]=K_0[\mathrm{Reg}(S)]\). For convenience, we identify \(\mathrm{Reg}(S)\) with \(\mathcal {M}^0(H,I,\Lambda ;P)\). By [11, Lemma 1, p. 48], the mapping \(\psi \) linearly spanned by the the mapping defined by

$$\begin{aligned} K_0\big [\mathrm{Reg}(S)\big ]\rightarrow {\mathfrak {M}}\big (K[H],I,\Lambda ;P\big ); (a,i,m)\mapsto (a_{jn}) \end{aligned}$$

is an algebra isomorphism, where \({\mathfrak {M}}\big (K[H],I,\Lambda ;P\big )\) is an algebra of matrix type and \((a_{jn})\) the \(I\times \Lambda \) matrix with entry \(a_{jn}=a\) if \(j=i,n=m\) and, otherwise, \(a_{jn}=0\). Since, by the forgoing proof, \(K_0\big [\mathrm{Reg}(S)\big ]\) has an identity, and by [11, Proposition 25, p. 59], this shows that \(|I|=|\Lambda |=n<+\infty \) and P is an invertible matrix in \(M_n\big (K[H]\big )\). It is not difficult to check that \(P^{-1}\) is the identity of \({\mathfrak {M}}\big (K[H],I,\Lambda ;P\big )\) and the mapping defined by \(\varphi : \;A\mapsto AP\) is an isomorphism of \({\mathfrak {M}}\big (K[H],I,\Lambda ;P\big )\) onto \(M_n(K[H])\).

Denote by \(\varepsilon _{i}\) the \(n\times n\) matrix with unity of K[H] at the row i, column i position and the zero 0 of K[H] in all other entries. So, \(F=\sum _{i=1}^nf_i\) where \(f_i=\psi ^{-1}\varphi ^{-1}\left( \varepsilon _{i}\right) \). Put

$$\begin{aligned} M_{im}=\big \{(a,i,m): a\in H\big \}. \end{aligned}$$

It is not difficult to know that each \(f_i\) is an idempotent, supp \((f_i)\subseteq M_{ii}\) and \(f_if_m\not =0\) only if \(i=m\).

Lemma 3.4

Let S satisfy the conditions of Theorem 3.3. Then \(K_0[S]=\bigoplus _{i=1}^ne_iK_0[S]\), where \(e_i=(p_{m_i,i}^{-1}, i, m_i) (m_i\in \Lambda )\) and \(e_iK_0[S]\cong e_jK_0[S]\) for any ij.

Proof

By the properties of Rees matrix semigroups, \(e_ix=x=xe_i\) for all \(x\in M_{ii}\). It follows that

$$\begin{aligned} f_i\cdot K_0[S]\subseteq \sum _{u\in \mathrm{supp}\; (f_i)}u\cdot K_0[S]\subseteq e_i\cdot K_0[S]. \end{aligned}$$

Because \(\mathrm{supp}\; (f_je_i)\subseteq M_{ji}\), we have

$$\begin{aligned} \mathrm{supp}\; \left( \sum _{j=1}^{i-1}f_je_i+\sum _{j=i+1}^nf_je_i\right) \; \bigcap M_{ii}=\emptyset \end{aligned}$$

and \(\mathrm{supp}\; (f_ie_i)\subseteq M_{ii}\). But

$$\begin{aligned} e_i=f_ie_i+\sum _{j=1}^{i-1}f_je_i+\sum _{j=i+1}^nf_je_i, \end{aligned}$$

now \(e_i=f_ie_i.\) Therefore, \(e_i\cdot K_0[S]\subseteq f_i\cdot K_0[S]\) and whence \(e_i\cdot K_0[S]= f_i\cdot K_0[S]\).

Note that \(F=\sum _{i=1}^nf_i\) is the identity of \(K_0[S]\) and \(f_i\)’s are orthogonal. We observe that \(K_0[S]=\bigoplus _{i=1}^nf_i\cdot K_0[S]=\bigoplus _{i=1}^ne_i\cdot K_0[S]\). In addition, \(e_i\)’s are all nonzero, further by Lemma 3.1 \((3), e_i\)s are related by \({\mathcal {D}}\). Therefore, by [10, Proposition (21.20), p. 315], \(e_iK_0[S]\cong e_jK_0[S]\) for any ij. \(\square \)

Now, we have

$$\begin{aligned} K_0[S]= & {} \bigoplus \nolimits _{i=1}^ne_i\cdot K_0[S]\cong End_{K_0[S]}(n e_1\cdot K_0[S])\cong M_n(e_1K_0[S]e_1)\\\cong & {} M_n(K_0[e_1Se_1]). \end{aligned}$$

On the other hand, since S is an almost idempotent-free semigroup, it is easy to see that \(E(e_1Se_1)=E(e_1\mathrm{Reg}(S)e_1)\). But \(S/V\cong \mathrm{Reg}(S)\) is a completely 0-simple semigroup, we have \(E(e_1\mathrm{Reg}(S)e_1)=\{e_1,0\}\). However, \(e_1Se_1\) is an idempotent-free monoid.

We have now proved that \(K_0[S]\cong M_n(K_0[e_1Se_1])\), so that \(M_n(Z(K_0[e_1Se_1]))\) is an Azumaya algebra. By Lemma 2.7, \(K_0[e_1Se_1]\) is an Azumaya algebra. The proof is finished. \(\square \)

Let us turn back to the proof of Theorem 3.3. If \(S\backslash V\) is a subsemigroup of S, then P has no zero entries. But P is invertible in \(M_n(K[H])\), so the cardinality of the sets I and \(\Lambda \) must be equal to 1. In other words, \(n=1\). Thus S is an idempotent-free monoid. Note that \(S\backslash V\) is a subsemigroup of S is equivalent to that the set of nonzero regular elements of S forms a subsemigroup of S, we have the following corollary.

Corollary 3.5

Let S be an almost idempotent-free semigroup whose set of nonzero regular elements forms a subsemigroup. If K[S] is Azumaya, then S is an idempotent-free monoid.

By applying the result [11, Corollary 17, p. 325], we have that for any finite semigroup \(S, K_0[S]\) is an Azumaya algebra if and only if \(K_0[S]\cong \oplus _{i=1}^mK_0[S_i]\) where each \(S_i\) is an almost idempotent-free semigroup and \(K_0[S_i]\) an Azumaya algebra. Furthermore, by Theorem 3.2, the following corollary is immediate. This corollary confirms Problem 1.3 when the semigroup satisfies the regularity condition.

Corollary 3.6

Let S be a finite semigroup. If S satisfies the regularity condition, then \(K_0[S]\) is an Azumaya algebra if and only if \(K_0[S]\cong \bigoplus _{i=1}^mM_{n_i}(K_0[G_i])\) where each \(G_i\) is an idempotent-free monoid and \(K_0[G_i]\) an Azumaya algebra.

4 The Condition \(min_J\)

The aim of this section is to determine when semigroup algebras of almost idempotent-free semigroups satisfying the condition \(min_J\) are Azumaya algebras.

Lemma 4.1

Let S be an almost idempotent-free semigroup satisfying the condition \(min_J\). Assume that K[S] is an Azumaya algebra. If I and J are \({\mathcal {J}}\)-classes of S that are not regular, then for any \(a\in I,b\in J\), we have \(J_{ab}<I,J\) and \(J_{ba}<I,J\).

Proof

For any \(a\in I,b\in J\), we have \(J_{ab}\le J_b=J\). If \(J_{ab}=J\), then there exist \(x,y\in S^1\) such that \(b=xaby\). Thus \(b=(xa)^nby^n\) for all positive integer n. Obviously, \((xa)^n\not =0\). Note that

$$\begin{aligned} \cdots \le J_{(xa)^3}\le J_{(xa)^2}\le J_{xa}. \end{aligned}$$

But S satisfies the condition \(min_J\), so there exists positive integer m such that

$$\begin{aligned} J_{(xa)^m}=J_{(xa)^{m+1}}=\cdots . \end{aligned}$$

It follows that \((xa)^m(xa)^m\in J_{(xa)^m}\), in other words, \(J_{(xa)^m}\) is 0-simple. Thus by Lemma 3.1 \((1), (xa)^m\) is regular, whence by Lemma 3.1(2), I is the unique nonzero regular \({\mathcal {J}}\)-class of S since \(J_{(xa)^m}\le J_a=I\), giving \(J_{(xa)^m}=I\). This is a contradiction. Consequently, \(J_{ab}<J\); similarly \(J_{ab}<I\). The rest can be similarly proved. \(\square \)

Lemma 4.2

Let S be an almost idempotent-free semigroup satisfying the condition \(min_J\). If K[S] is an Azumaya algebra, then for any \(a\in S\), either a is regular or \(a^n=0\) for some positive integer n if S has zero element 0.

Proof

Without loss of generality, we assume that S has zero element 0. Now, let a be not a regular element of S. Obviously,

$$\begin{aligned} J_a\ge J_{a^2}\ge \cdots \ge J_{a^n}\ge \cdots . \end{aligned}$$

But S satisfies the condition \(min_J\), so there exists positive integer m such that

$$\begin{aligned} J_{a^m}=J_{a^{m+1}}=\cdots . \end{aligned}$$

It follows that \(J_{a^m}=J_{a^{2m}}\), whence \(a^m\circ a^m\in J_{a^m}\). Thus \(a^m=0\) or \(J(a^m)/I(a^m)\) is 0-simple. If the second case holds, then by Lemma 3.1 \((1), a^m\) is regular. Note that \(J_{a^m}\le J_a\). By Lemma 3.1 \((2), J_{a^m}=J_a\) and whence a is regular, contrary to the hypothesis. Therefore, \(a^m=0\) and the result follows. \(\square \)

Recall that a semigroup S is called an ideal extension of nil semigroups by a completely 0-simple semigroup if S has a nil ideal I such that S / I is a completely 0-simple semigroup. The following corollary is immediate from Lemmas 3.1 and 4.2.

Corollary 4.3

Let S be an almost idempotent-free semigroup satisfying \(min_J\). If \(K_0[S]\) is Azumaya, then S is an ideal extension of nil semigroups by a completely 0-simple semigroup.

Based on Lemma 4.2, we have the following theorem.

Theorem 4.4

Let S be an almost idempotent-free semigroup satisfying the condition \(min_J\), and assume that S has no zero elements. If K[S] is an Azumaya algebra, then S is a group.

Proof

Assume that K[S] is an Azumaya algebra. By Lemma 4.2, all elements of S are regular, in other words, S is a regular semigroup. By Corollary 3.5, S is an idempotent-free monoid. Thus S is a group. \(\square \)

We shall simply denote

$$\begin{aligned} Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]=\left\{ \sum _{i=1}^mz_iy_i: z_i\in Z\big (K[S]\big ),y_i\in K\big [\mathrm{Reg}(S)\big ], m\in N\right\} , \end{aligned}$$

where N is the set of positive integers.

Lemma 4.5

Let S be an almost idempotent-free semigroup satisfying the condition \(min_J\). If K[S] is an Azumaya algebra, then \(K[S]=Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\).

Proof

Assume that K[S] is an Azumaya algebra. By Theorem 4.4, the theorem is clearly true for the case if S has no zero elements. So, we now let S have a zero element. Assume, on the contrary, that there exists \(\alpha \notin Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\). It follows that \(a_1\notin Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\) for some \(a_1\in \mathrm{supp}(\alpha )\). Obviously, \(a_1\in S\backslash \mathrm{Reg}(S)\). By Lemma 2.3(1), we have

$$\begin{aligned} \Big (K\big [J(a_1)\big ]\cap Z\big (K[S]\big )\Big )K[S]=K\big [J(a_1)\big ] \end{aligned}$$
(1)

and \(a_1=\alpha _1 (\beta _1+\beta '_1)=\alpha _1\beta _1+\alpha _1\beta '_1\) for some \(\alpha _1\in K[J(a_1)]\cap Z(K[S]), \beta _1\in K[\mathrm{Reg}(S)]\) and \(\beta '_1\in K[V]\). But \(a_1\notin Z(K[S])K[\mathrm{Reg}(S)]\), so \(\alpha _1\beta '_1\notin Z(K[S]) K[\mathrm{Reg}(S)]\) and \(\alpha _1\beta '_1\not =0\). Since \(J(a_1)=J_{a_1}\cup I(a_1)\), we know that \(J_{a_1}\) is the greatest element of the set \(\{J_x:x\in J(a_1)\}\), and whence \(J_{a_1}\) is bigger than the greatest element of the set

$$\begin{aligned} B:=\Big \{J_x: x\in \mathrm{supp}(u),u\in K\big [J(a_1)\big ]\cap Z\big (K[S]\big )\Big \}. \end{aligned}$$

In particular, \(J_{a_1}\ge J_x\) for any \(x\in \mathrm{supp}(\alpha _1)\). Now, by Lemma 4.1, \(J_{a_1}> J_y\) for any \(y\in \mathrm{supp}(\alpha _1\beta '_1)\). Because \(\alpha _1\beta '_1\notin Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\), we have \(a_2\in \mathrm{supp}(\alpha _1\beta '_1)\) such that \(a_2\notin Z(K[S])K[\mathrm{Reg}(S)]\). By the forgoing proof, \(J_{a_1}>J_{a_2}\).

By applying the same arguments to \(a_2\), we have \(a_3\) such that \(a_3\notin Z(K[S])K[\mathrm{Reg}(S)]\) and \(J_{a_2}>J_{a_3}\). Continuing this process, we may obtain the \({\mathcal {J}}\)-classes:

$$\begin{aligned} J_{a_1}>J_{a_2}>\cdots >J_{a_n}>\cdots , \end{aligned}$$

contrary to the hypothesis that S satisfies the condition \(min_J\). Consequently, \(K[S]=Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\). \(\square \)

We now arrive at the main result of this section.

Theorem 4.6

Let S be an almost idempotent-free semigroup satisfying \(min_J\). If, in addition, S satisfies the regularity condition, then K[S] is an Azumaya algebra if and only if \(K[S]=Z(K[S])K[\mathrm{Reg}(S)]\) and \(K\big [\mathrm{Reg}(S)\big ]\) is an Azumaya algebra.

Proof

Assume that K[S] is an Azumaya algebra. By Lemma 4.5, we need only to verify that \(K\big [\mathrm{Reg}(S)\big ]\) is an Azumaya algebra. In fact, by Lemma 3.1 (2), V is an ideal of S, so by Lemma 3.1 \((2), K_0[\mathrm{Reg}(S)]\cong K_0[S/V]\cong K[S]/K[V]\) and further is an Azumaya algebra by Lemma 2.3. However, by Lemma 2.6, \(K[\mathrm{Reg}(S)]\) is an Azumaya algebra.

Conversely, suppose that \(K[S]=Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\) and \(K\big [\mathrm{Reg}(S)\big ]\) is an Azumaya algebra. Since \(K[S]=Z\big (K[S]\big )K\big [\mathrm{Reg}(S)\big ]\), we know that the center \(Z\big (K[S]\big )\) of K[S] is equal to

$$\begin{aligned} \Big \{x\in K[S]: xu=ux \quad \text { for all }u\in K\big [\mathrm{Reg}(S)\big ]\Big \}. \end{aligned}$$

This means that the center \(Z\big (K[\mathrm{Reg}(S)]\big )\) of \(K\big [\mathrm{Reg}(S)\big ]\) is contained in \(Z\big (K[S]\big )\). By Lemma 2.4, we observe that \(Z\big (K[S]\big )\otimes _{Z\big (K\big [\mathrm{Reg}(S)\big ]\big )}K\big [\mathrm{Reg}(S)\big ]\) is an Azumaya algebra. It is easy to see that K[S] is a homomorphic image of \(Z(K[S])\otimes _{Z(K[\mathrm{Reg}(S)])}K[\mathrm{Reg}(S)]\) since \(K[S]=Z(K[S])K[\mathrm{Reg}(S)]\). Now by Lemma 2.3 (2), K[S] is an Azumaya algebra. \(\square \)

It is well known that finite semigroups satisfy the condition \(min_J\). Obviously, any infinite group satisfies the condition \(min_J\). So, not all semigroups satisfying the condition \(min_J\) are finite. The next example illustrates that there exist Azumaya semigroup algebras of almost idempotent-free semigroups satisfying the condition \(min_J\) but not finite.

Example 4.7

Let U be a null semigroup with zero 0, and G a group with identity 1 and such that K[G] is an Azumaya algebra. Assume that S is the disjoint union of U and G. Define a multiplication \(*\) by:

$$\begin{aligned} x*y=\left\{ \begin{array}{cl} x&{} \quad \text { if}\quad y\in G\quad \hbox {but}\quad x\in U\\ y&{} \quad \text { if}\quad x\in G\quad \hbox {but}\quad y\in U\\ xy&{} \quad \text { otherwise,} \end{array} \right. \end{aligned}$$

where xy is the product of x and y in the semigroup U or the group G. By computation, \((S,*)\) is a monoid with identity 1. It is easy to check that the Green’s relation \({\mathcal {J}}\) on S is equal to \((G\times G)\cup \Delta _U\) where \(\Delta _U\) is the identity relation on U. This shows that in the semigroup S, maximal chains of \({\mathcal {J}}\)-classes of S have the form: \(J_0<J_x<J_g,\) where \(x\in U\backslash \{0\}, g\in G\). So, S satisfies \(min_J\). On the other hand, it is easy to see that \(Z\big (K[S]\big )=K[U]+Z\big (K[G]\big )\) and whence \(K[S]=Z\big (K[S]\big )K[G]\) since \(1\in Z\big (K[S]\big )\). By Theorem 4.6, K[S] is an Azumaya algebra. When U is infinite, S is clear infinite.

The following example illustrates the condition that \(min_J\) is not necessary for a semigroup algebra K[S] to be an Azumaya algebra.

Example 4.8

Let G be a group such that K[G] is Azumaya. Let U be the \(\omega \)-chain \(\{e_0,e_1, e_2,\ldots \}\) with

$$\begin{aligned} e_0>e_1>e_2>\ldots \end{aligned}$$

(for detail, see [9, Example 4.6, p. 144]). The semigroup S constructed as in Example 4.7 has the \({\mathcal {J}}\)-class chain:

$$\begin{aligned} J_1>J_{e_0}>J_{e_1}>J_{e_2}>\cdots . \end{aligned}$$

This shows that S does not satisfy the condition \(min_J\). Note that U is in the center of S. Thus \(Z\big (K[S]\big )=Z\big (K[G]\big )+K[U]\), so that \(K[S]=Z\big (K[S]\big )K[G]\). It follows that K[S] is a homomorphic image of \(K[G]\otimes _{Z\big (K[G]\big )}Z\big (K[S]\big )\). But \(K[G]\otimes _{Z\big (K[G]\big )}Z\big (K[S]\big )\) is Azumaya (by Lemma 2.4). Therefore, K[S] is an Azumaya algebra.

The remainder of this section is devoted to semigroup algebras of idempotent-free semigroups.

Lemma 4.9

Let S be an idempotent-free semigroup. If K[S] is an Azumaya algebra, then S is a monoid satisfying the regularity condition and \(\mathrm{Reg}(S)\backslash \{0\}\) is a subgroup of S. Moreover, if S has no zero elements, then S is a group.

Proof

Because S is an idempotent-free semigroup and by Lemma 3.1 (3), S / V is a completely 0-simple semigroup without idempotents except possibly identity and zero element. It follows that S / V is a 0-group (that is, a group adjoining a zero). This shows that \(\mathrm{Reg}(S)\backslash \{0\}\) is a subgroup of S, and whence S satisfies the regularity condition. Now, by the proof of Theorem 3.3, the identity of K[S] coincides with the one of \(K[\mathrm{Reg}(S)]\), so that S is a monoid. The rest is trivial. \(\square \)

By Lemma 4.9 and Theorem 4.6, the following theorem is immediate. This theorem answers particularly Conjecture 1.2.

Theorem 4.10

Let S be an idempotent-free monoid with the group of units U. If S satisfies the condition \(min_J\), then K[S] is an Azumaya algebra if and only if \(K[S]=Z\big (K[S]\big )K[U]\) and K[U] is an Azumaya algebra.

It is clear to see that cancellative monoids are idempotent-free semigroups. By Theorem 4.4, we have the following corollary.

Corollary 4.11

Let S be a cancellative monoid satisfying the condition \(min_J\). If K[S] is an Azumaya algebra, then S is a group.

Proof

By Theorem 4.4, S is isomorphic to some Rees matrix semigroup over a group, and further a regular semigroup. But any regular cancellative monoid is a group, now S is a group. \(\square \)